Christodoulou 0805.3880

The Formation
of Black Holes
in General Relativity
arXiv:0805.3880v1 [gr-qc] 26 May 2008
Demetrios Christodoulou
May 18, 2008

Contents
Prologue
Chapter 1 : The Optical Structure Equations

1.1 The basic geometric setup
1.2 The optical structure equations
1.3 The Bianchi equations
1.4 Canonical coordinate systems
Chapter 2 : The Characteristic Initial Data

2.1 The characteristic initial data
2.2 Construction of the solution in an initial domain
Chapter 3 : L∞ Estimates for the Connection Coefficients

3.1 Introduction
3.2 L∞ estimates for χ′
3.3 L∞ estimates for χ′
3.4 L∞ estimates for η, η
3.5 L∞ estimates for ω, ω
3.6 The smallness requirement on δ
Chapter 4 : L4 (S) Estimates for the 1st Derivatives of the Connection

Coefficients
4.1 Introduction
4.2 L4 (S) estimates for ∇/χ
4.3 L4 (S) estimates for ∇/χ
4.4 L4 (S) estimates for ∇/η, ∇/η
4.5 L4 (S) estimates for /dω, /dω
4.6 L4 (S) estimates for Dω, Dω
Chapter 5 : The Uniformization Theorem

5.1 Introduction. An L2 (S) estimate for K − K
5.2 Sobolev inequalities on S. The isoperimetric constant
5.3 The uniformization theorem
5.4 Lp elliptic theory on S
Chapter 6 : L4 (S) Estimates for the 2nd Derivatives of the Connection

Coefficients
6.1 Introduction
6.2 L4 (S) estimates for ∇
/ 2 χ, K − K
4
6.3 L (S) estimates for ∇/ 2χ
2
6.4 L4 (S) estimates for ∇
/ 2 η, ∇
/ 2η
6.5 L4 (S) estimate for ∇
/ 2ω
6.6 L4 (S) estimate for ∇
/ 2ω
Chapter 7 : L2 Estimates for the 3rd Derivatives of the Connection

Coefficients
7.1 Introduction
7.2 L2 estimates for ∇
/ 2 η, ∇/ 2η
2
7.3 L elliptic theory for generalized Hodge systems on S
/ 3 χ′ , /dK
2
7.5 L estimates for ∇/ 3 χ′
/ 3 η, ∇/ 3η
2 3
7.7 L estimate for ∇/ ω
/ 2 ω and ∇ / 3ω
7.9 L estimates for /dDω, /dDω, D2 ω, D2 ω
2
Chapter 8 : The Multiplier Fields and the Commutation Fields

8.1 Introduction
8.2 L∞ estimates for the deformation tensors of L, K and S
8.3 Construction of the rotation vectorfields Oi
8.4 L∞ estimates for the Oi and ∇ /Oi
8.5 L∞ estimates for the deformation tensors of the Oi
Chapter 9 : Estimates for the Derivatives of the Deformation Tensors

of the Commutation Fields
9.1 L4 (S) estimates for the 1st derivatives of the deformation tensors of L, S
9.2 L4 (S) estimates for the 1st derivatives of the deformation tensors of the Oi
9.3 L2 estimates for the 2nd derivatives of the deformation tensors of L, S
9.4 L2 estimates for the 2nd derivatives of the deformation tensors of the Oi
Chapter 10 : The Sobolev Inequalities on the Cu and C u

10.1 Introduction
10.2 The Sobolev inequalties on the Cu
10.3 The Sobolev inequalities on the C u
Chapter 11 : The S-tangential Derivatives and the Rotational Lie

Derivatives
11.1 Introduction and preliminaries
11.2 The coercivity inequalities on the standard sphere
11.3 The coercivity inequalities on Su,u
Chapter 12 : Weyl Fields and Currents. The Existence Theorem

12.1 Weyl fields and Bianchi equations. Weyl currents
12.2 Null decompositions of Weyl fields and currents
3
12.3 The Bel-Robinson tensor. The energy-momentum density vectorfields
12.4 The divergence theorem in spacetime
12.5 The energies and fluxes. The quantity P2
12.6 The controlling quantity Q2 . Bootstrap assumptions and the comparison
lemma
12.7 Statement of the existence theorem. Outline of the continuity argument
Chapter 13 : The Multiplier Error Estimates

13.1 Preliminaries
13.2 The multiplier error estimates
Chapter 14 : The 1st Order Weyl Current Error Estimates

14.1 Introduction
14.2 The error estimates arising from J 1
Chapter 15 : The 2nd Order Weyl Current Error Estimates

15.1 The 2nd order estimates which are of the same form as the 1st order
estimates
15.2 The genuine 2nd order error estimates
Chapter 16 : The Energy-Flux Estimates. Completion of the Conti-

nuity Argument
16.1 The energy-flux estimates
16.2 Higher order bounds
16.3 Completion of the continuity argument
16.4 Restatement of the existence theorem
Chapter 17 : Trapped Surface Formation
Bibliography
4
Prologue
The story of the black hole begins with Schwarzschild’s discovery [Sc] of the
Schwarzschild solution in 1916, soon after Einstein’s foundation of the general
theory of relativity [Ei1] and his final formulation of the field equations of grav-
itation [Ei2], the Einstein equations, in 1915. The Schwarzschild solution is a
solution of the vacuum Einstein equations which is spherically symmetric and
depends on a positive parameter M , the mass. With r such that the area of
the spheres, which are the orbits of the rotation group, is 4πr2 , the solution
in the coordinate system in which it was originally discovered had a singular-
ity at r = 2M . For this reason only the part which corresponds to r > 2M
was originally thought to make sense. This part is static and represents the
gravitational field outside a static spherically symmetric body with surface area
corresponding to some r0 > 2M .
However, the understanding of Schwarzschild’s solution gradually changed.
First, in 1923 Birkoff [Bir] proved a theorem which shows that the Schwarzschild
solution is the only spherically symmetric solution of the vacuum Einstein equa-
tions. One does not therefore need to assume that the solution is static. Thus,
Schwarzschild’s solution represents the gravitational field outside any spheri-
cally symmetric body, evolving in any manner whatever, for example undergoing
gravitational collapse.
Eddington [Ed], in 1924, made a coordinate change which transformed the
Schwarzschild metric into a form which is not singular at r = 2M , however he
failed to take proper notice of this. Only in 1933, with Lemaı̂tre’s work [L],
it was realized that the singularity at r = 2M is not a true singularity but
rather a failure of the original coordinate system. Eddington’s transformation
was rediscovered by Finkelstein [Fi] in 1958, who realized that the hypersurface
r = 2M is an event horizon, the boundary of the region of spacetime which is
causally connected to infinity, and recognized the dynamic nature of the region
r < 2M . Now, Schwarzschild’s solution is symmetric under time reversal, and
one part of it, the one containing the future event horizon, the boundary of the
region of spacetime which can send signals to infinity, is covered by one type
of Eddington-Finkelstein coordinates, while the other part, the one containing
the past event horizon, the boundary of the region of spacetime which can re-
ceive signals from infinity, is covered by the other type of Eddington-Finkelstein
coordinates. Actually, only the first part is physically relevant, because only
future event horizons can form dynamically, in gravitational collapse. Systems
of coordinates that cover the complete analytic extension of the Schwarzschild
solution had been provided earlier (in 1950) by Synge [Sy], and a single most
covenient system that covers the complete analytic extension was discovered
independently by Kruskal [Kr] and Szekeres [Sz] in 1960.
Meanwhile in 1939, Oppenheimer and Snyder had studied the gravitational
collapse of a pressure-free fluid ball of uniform density, a uniform density “ball
5
of dust”. Even though this is a highly idealized model problem, their work was
very significant, being the first work on relativistic gravitational collapse. As
mentioned above, the spacetime geometry in the vacuum region outside the ball
is given by the Schwarzscild metric. Oppenheimer and Snyder analyzed the
causal structure of the solution. They considered in particular an observer on
the surface of the dust ball sending signals to a faraway stationary observer at
regularly spaced intervals as judged by his own clock. They discovered that the
spacing between the arrival times of these signals to the farway observer becomes
progressively longer, tending to infinity as the radius r0 corresponding to the
surface of the ball approaches 2M . This effect has since been called the infinite
redshift effect. The observer on the sufrace of the dust ball may keep sending
signals after r0 has become less than 2M , but these signals proceed to ever
smaller values of r until, within a finite afine parameter interval, they reach a
true singularity at r = 0. The observer on the surface of the ball reaches himself
this singular state within a finite time interval as judged by his own clock. The
concept of a future event horizon and hence of a region of spacetime bounded
by this horizon from which no signals can be sent which reach arbitrarily large
distances, was thus already implicit in Oppenheimer-Snyder work.
The 1964 work of Penrose [P1] introduced the concept of null infinity, which
made possible the precise general definition of a future event horizon as the
boundary of the causal past of future null infinity. A turning point was reached
in 1965 by the introduction by Penrose of the concept of a closed trapped surface
and his proof of the first singularity theorem, or, more precisely, incompleteness
theorem [P2]. Penrose defined a trapped surface as being a spacelike surface in
spacetime, such that an infinitesimal virtual displacement of the surface along
either family of future-directed null geodesic normals to the surface leads to a
pointwise decrease of the area element. On the basis of this concept, Penrose
proved the following theorem:
A spacetime (M, g) cannot be future null geodesically complete if:
1. Ric(N, N ) ≥ 0 for all null vectors N .
2. There is a non-compact Cauchy hypersurface H in M .
and:
3. There is a closed trapped surface S in M .
Here Ric is the Ricci curvature of g and condition 1 is always satisfied by
virtue of the Einstein equations and the physical positivity condition on the
energy-momentum-stress tensor of matter.
Once the notions of null infinity and of a closed trapped surface where intro-
duced, it did not take long to show that a spacetime with a complete future null
infinity which contains a closed trapped surface must contain a future event hori-
zon, the interior of which contains the trapped surface (see [H-E], Proposition
9.2.1). For the ideas and methods which go into Penrose’s theorem the reader
may consult, besides the monograph by Hawking and Ellis just mentioned, the
article of Penrose in [P3] as well as his monograph [P4]. Further singularity the-
orems, which also cover cosmological situations, where subsequently established
6
by Hawking and Penrose (see [H-E]), but it is the original sigularity theorem
quoted above which is of interest in the present context, as it corcerns gravita-
tional collapse. We should also mention that the term black hole for the interior
of the future event horizon was introduced by Wheeler in 1967 (see [Wh]).
Now, the 1952 work of Choquet-Bruhat [Cho1] (see also [Cho2] and [Cho3])
had shown that any initial data set (H, g, k), where H is a 3-dimensional mani-
fold g is a Riemannian metric on H and k is a symmetric 2-covariant tensorfield
on H such that the pair (g, k) satisfies the so called “constraint equations”, has
a future development (M, g), namely a 4-dimensional manifold M endowed with
a Lorentzian metric g satisfying the vacuum Einstein equations, such that H
is the past boundary of M , g and k are the first and second fudamental forms
of H relative to (M, g), and for each p ∈ M each past directed causal curve
initiating at p terminates at a point of H. The constraint equations are the
contracted Codazzi and twice contracted Gauss equations of the embedding of
H in M . The subsequent 1969 work of Choquet-Bruhat and Geroch [C-G] then
showed that each such an initial data set has a unique maximal future devel-
opment M ∗ , namely a future development, in the above sense, which extends
every other future development of the same initial data set. Geroch [Ge] subse-
quently showed that for any future development (M, g), M is diffeomorphic to
[0, ∞) × H. Moreover, the above theorems extend to the case where instead of
vacuum we have suitable matter, such as a perfect fluid, or an electromagnetic
field. In the light of the theorem of Choquet-Bruhat and Geroch condition 2 in
Penrose’s theorem may be replaced by the statement that (M, g) is the max-
imal future development of initial data on a complete non-compact spacelike
hypersurface.
In 1990 Rendall [R] solved in a very satisfactory manner the local character-
istic initial value problem for the vacuum Einstein equations (earlier work had
been done by Choquet-Bruhat [Cho4] and by Müller zum Hagen and Seifert
[M-S]). In this case we have, in the role of H, either two null hypersurfaces C
and C intersecting in a spacelike surface S, S being the past boundary of both
C and C, or a future null geodesic cone Co of a point o. The initial data on C
and C are the conformal intrinsic geometry of these null hypersurfaces, together
with the full intrinsic geometry of S, the initial rate of change of the area ele-
ment of S under displacement along C and C, and a certain 1-form on S (the
torsion). The initial data on Co are the conformal intrinsic geometry of Co and
certain regularity conditions at o. In contrast to the case where the initial data
are given on a spacelike hypersurface, there are no constraints, and the initial
data can be freely specified. The theorem of Rendall then shows that any such
characteristic initial data has S
a future development (M, g), bounded in the past
by a neighborhood of S in C C and of o in Co respectively. The theorem of
Choquet-Bruhat and Geroch, which applies to future developments, then shows
that there is a unique maximal future development (M ∗ , g) corresponding to
the given characteristic initial data.
Now the proof of the theorem of Penrose is by showing that if M where com-
plete, the boundary ∂J + (S) of the causal future J + (S) of the closed trapped
surface S would be compact. The integral curves of any timelike vectorfield on
7
M would define a continuous mapping of ∂J + (S) into H, M being a develop-
ment of H, and this mapping would have to be a homeomorphism onto its image,
∂J + (S) being compact. This leads to a contradiction with the assumption that
H is non-compact. We see that the proof makes no use of the strictly spacelike
nature of H other than through the assumption that M is a future development
of H. We may therefore replace H by a complete future null geodesic cone and
restate the theorem as follows. Here vacuum or suitable matter is assumed. We
do not state the first condition of Penrose’s because as we already mentioned,
it is automatically satisfied by virtue of the physical positivity condition that
the energy-momentum-stress tensor of matter satisfies.
Let us be given regular charactesistic initial data on a complete null geodesic
cone Co of a point o. Let (M ∗ , g) be the maximal future development of the
data on Co . Suppose that M ∗ contains a closed trapped surface. Then (M ∗ , g)
is future null geodesically incomplete.
An important remark at this point is that it is not a priori obvious that
closed trapped surfaces are evolutionary. That is, it is not obvious whether
closed trapped surfaces can form in evolution starting from initial conditions
in which no such surfaces are present. What is more important, the physically
interesting problem is the problem where the initial conditions are of arbitrarily
low compactness, that is, arbitrarily far from already containing closed trapped
surfaces, and we are asked to follow the long time evolution and show that,
under suitable circumstances, closed trapped surfaces eventually form. Only an
analysis of the dynamics of graviational collapse can achieve this aim.
Returning to our review of the historical development of the black hole con-
cept, a very significant development took place in 1963, shortly before the work
of Penrose. This was the discovery by Kerr [Ke] of a two parameter family of
axially symmetric solutions of the vacuum Einstein equations, with an event
horizon, the exterior of which is a regular asymptotically flat region. The two
parameters are the mass M , which is positive, and the angular momentum
L about the axis of symmetry, which is subject to the restriction |L| ≤ M 2 .
Kerr’s solution reduces in the spatial case of vanishing angular momentum to
Schwarzschild’s solution. The Kerr solution possesses an additional Killing field,
besides the generator of rotations about the axis, however this additional Killing
field, in contrast to the case of the Schwarzschild solution, is timelike not on
the entire exterior of the horizon, but only in the exterior of a non-spacelike
hypersurface containing the horizon. So only in this exterior region is the solu-
tion stationary in a strict sense. At every point of the region between the two
hypersurfaces, called ergosphere, the additional Killing field is spacelike, but the
plane which is the linear span of the two vectors at this point is timelike. On
the horizon itself the plane becomes null and tangent to the horizon, and the
null line generating this null plane defines the angular velocity of the horizon,
a constant associated to the horizon. Kerr’s solution is symmetric under time
reversal if also the sign of the angular momentum is reversed, hence it possesses,
besides the future even horizon, also an unphysical past event horizon, just like
8
the the Schwarzschild solution.
The fascinating properties of the Kerr solution where revealed in the decade
following its discovery. In particular Boyer and Lindquist [B-L] introduced a
more convenient coordinate system and obtained the maximal analytic exten-
sion. One of these fascinating properties which concerns us here, is that the
hypersurfaces of constant Boyer-Lindquist coordinate t are complete asymptot-
ically flat maximal spacelike hypersurfaces, their maximal future development
contains closed trapped surfaces and, in accordance with Penrose’s theorem,
is incomplete. Nevertheless the future boundary of the maximal development
is nowhere singular, the solution extending as an analytic solution across this
boundary. This future boundary is a regular null hypersurface, a Cauchy hori-
zon, illustrating the fact that incompleteness of the maximal future development
does not imply a singular future boundary.
Returning again to the question of whether closed trapped surfaces are evo-
lutionary, one may at first hand say that the question was already settled in
the affirmative by the Oppenheimer-Snyder analysis. This is because hyperbolic
systems of partial differential equations, such as the Einstein-Euler equations
describing a perfect fluid in general relativity, possess the property of continu-
ous dependence of the solution on the initial conditions. This holds at a given
non-singular solution, for a given finite time interval. Thus, since the initial
condition of a homogeneous dust ball leads to a trapped sphere within a finite
time interval, initial conditions which are sufficiently close to this special initial
condition will also lead to the formation of closed trapped surfaces within the
same time interval, the condition for a closed spacelike surface to be trapped
being an open condition. However, as we remarked above, the case that one is
really interested in is that for which the initial homogeneous dust ball is of low
compactness, far from already containing trapped spheres, and it is only by con-
tracting for sufficiently long time that a trapped sphere eventually forms. In this
case the closeness condition of the continuous dependence theorem may require
the initial conditions to be so unreasonably close to those of a homogeneous
dust ball that the result is devoid of physical significance.
With the above remarks in mind the author turned to the study of the
gravitational collapse of an inhomogeneous dust ball [Chr1]. In this case, the
initial state is still spherically symmetric, but the density is a function of the
distance from the center of the ball. The corresponding spherically symmetric
solution had already been obtained in closed form by Tolman in 1934 [T], in
comoving coordinates, but its causal structure had not been investigated. This
required integrating the equations for the radial null geodesics. A very different
picture from the one found by Oppenheimer and Snyder emerged from this
study. The initial density being assumed a decreasing function of the distance
from the center, so that the central density is higher than the mean density,
it was found that as long as the collapse proceeds from an initial state of low
compactness, the central density becomes infinite before a black hole has a
chance to form, thus invalidating the neglect of pressure and casting doubt on
the predictions of the model from this point on, in particular on the prediction
that a black hole eventually forms.
9
At this point the author turned to the spherically symmetric scalar field
model [Chr2]. This is the next simplest material model after the dust model.
The energy-momentum-stress tensor of matter is in this case that corresponding
to a scalar field φ:
1
Tµν = ∂µ φ∂ν φ + σgµν , σ = −(g −1 )µν ∂µ φ∂ν φ (1)
2
The integrability condition for Einstein’s equations, namely that Tµν is divergence-
free, is then equivalent to the wave equation for φ relative to the metric g. The
problem had been given to the author by his teacher, John Archibald Wheeler,
in 1968 (see [Chr3]), as a model problem through which insight into the dy-
namics of gravitational collapse would be gained. In the case of the dust model,
there is no force opposing the gravitational attraction , so there is no alternative
to collapse. This is not the case for the scalar field model and indeed in [Chr2]
it was shown that if the initial data are suitably small we obtain a complete
regular solution dispersing to infinity in the infinite future. So, for the scalar
field model there is a threshold for gravitational collapse. In this paper and
the papers on the scalar field that followed, the initial data where given on a
complete future null geodesic cone Co extending to infinity. The initial data on
Co consist of the function α0 = ∂(rφ)/∂s|Co , s being the affine parameter along
the generators of Co .
The next paper [Chr4] on the scalar field problem addressed the general case,
when the initial data where no longer restricted by a smallness condition. The
aim of this work was to prove the existence of a solution with a complete domain
of outer communications, that is, a development possessing a complete future
null infinity, the domain of outer communications being defined as the causal
past of future null infinity. This was tantamount to proving the weak cosmic
censorship conjecture of Penrose [P5] (called “asymptotic future predictability”
in [H-E]) in the context of the spherically symmetric scalar field model. The
aim was not reached in this paper. What was established instead was the
existence, for all regular asymptotically flat initial data, of a generalized solution
corresponding to a complete domain of outer communications. A generalized
solution had enough regularity to permit the study of the asymptotic behavior in
the domain of outer communications in the next paper, however no uniqueness
could be claimed for these generalized solutions, so the conjecture of Penrose
was left open.
In [Chr5] it was shown that when the final Bondi mass, that is, the infimum
of the Bondi mass at future null infinity, is different from zero, a black hole
forms of mass equal to the final Bondi mass, surrounded by vacuum. The rate
of growth of the redshift of light seen by faraway observers was determined and
the asymptotic wave behavior at future null infinity and along the event horizon,
was analyzed. However, the question of whether there exist initial conditions
which lead to a non-zero final Bondi mass was not addressed in this paper.
The next paper [Chr6] was a turning point in the study of the spherically
symmetric scalar field problem. Because of the fact that it has provided a
stepping stone for the present monograph, I quote its main theorem. Here Co
10
denotes the initial future null geodesic cone.
Consider on Co an annular region bounded by two spheres S1,0 and S2,0 with
S2,0 in the exterior of S1,0 . Let δ0 and η0 be the dimensionless size and the
dimensionless mass content of the region, defined by
r2,0 2(m2,0 − m1,0 )

δ0 = −1 η0 =
r1,0 r2,0
r1,0 , r2,0 and m1,0 , m2,0 being the area radii and mass contents of S1,0 , S2,0 re-
spectively. Let C 1 and C 2 be incoming null hypersurfaces through S1,0 and S2,0
and consider the spheres S1 and S2 at which C 1 and C 2 intersect future null
geodesic cones C with vertices on the central timelike geodesic Γ0 . There are
positive constants c0 and c1 such that if δ0 ≤ c0 and

1
η0 > c1 δ0 log
δ0
then S2 becomes trapped before S1 reduces to a point on Γ0 . There is a future

null geodesic cone C ∗ with vertex on Γ0 such that S2∗ is a maximal sphere in C ∗
while r1∗ > 0.
It was further shown that the region of trapped spheres, the trapped region,
terminates at a strictly spacelike singular boundary, and contains spheres whose
mass content is bounded from below by a positive constant depending only
on r1,0 , r2,0 , a fact which implies that the final Bondi mass is positive, thus
connecting with the previous work.
An important remark concerning the proof of the above theorem needs to
be made here. The proof does not consider at all the region interior to the
incoming null hypersurface C 1 . However, the implicit assumption is made that
no singularities form on Γ0 up to the vertex of C ∗ . If a smallness condition is
imposed on the restriction of the initial data to the interior of S1,0 , the argument
of [Chr2] shows that this assumption indeed holds. Also, by virtue of the way
in which the theorem was latter applied in [Chr9], the assumption in question
was a priori known to hold.
Since the spherical dust model had been disqualified by [Chr1] as establish-
ing the dynamical formation of trapped spheres, the work [Chr6] was the first
to establish the dynamical formation of closed trapped spheres in gravitational
collapse, although off course severely limited by the restriction to spherical sym-
metry and by the fact that it concerned an idealized matter model.
Solutions with initial data of bounded variation were considered in [Chr7]
and a sharp sufficent condition on the initial data was found for the avoidance
of singularities, namely that the total variation be sufficiently small, greatly
improving the result of [Ch2]. Moreover, a sharp extension criterion for solutions
was established, namely that if the ratio of the mass content to the radius of
spheres tends to zero as we approach a point on Γ0 from its causal past, then
the solution extends as a regular solution to include a full neighborhood of the
point. The structure of solutions of bounded variation was studied and it was
11
shown that at each point of Γ0 the solutions are locally scale invariant. Finally,
the behavior of the solutions at the singular boundary was analyzed.
In [Chr8] the author constructed examples of solutions corresponding to
regular asymptotically flat initial data which develop singularities which are not
preceeded by a trapped region but have future null geodesic cones expanding
to infinity. It was thus established for the first time that naked singularities do,
in fact, occur in the graviational collapse of a scalar field. Also, other examples
where constructed which contain singular future null geodesic cones which have
collapsed to lines and again are not preceeded by trapped regions.
The work on the spherically symmetric scalar field model culminated in
[Chr9]. Taking the space of initial data to be the space A of absolutely continous
functions on the non-negative real line, the theorem proved in [Chr9] was the
following.
Let us denote by R the subset of A consisting of those initial data which lead
to a complete maximal future development, and by S its complement in A. Let
also G ⊂ S be the subset consisting of those initial data which lead to a maximal
development possessing a complete future null infinity and a strictly spacelike
singular future boundary. Then E = S \ G has the following property. For each
initial data α0 ∈ E there is a function f ∈ A, depending on α0 , such that the
line Lα0 = {α0 + cf : c ∈ ℜ} in A is contained in G, except for α0 itself.
Moreover, the lines Lα0,1 , Lα0,2 corresponding to distinct α0,1 , α0,2 ∈ E do not
intersect.
The exceptional set E being, according this theorem, of codimension at least
1, the theorem established, within the spherically symmetric scalar field model,
the validity not only of the weak cosmic censorship conjecture of Penrose, but
also of his strong cosmic censorship conjecture, formulated in [P6]. This states,
roughly speaking, that generic asymptotically flat initial data have a maximal
development which is either complete or terminates in a totally singular future
boundary. The general notion of causal boundary of a spacetime manifold was
defined in [G-K-P]. The relationship between the two cosmic censorship conjec-
tures is discussed in [Chr10]. In the case of the spherically symmetric scalar
field model there is no system of local coordinates in which the metric extends
as a Lorentzian metric through any point of the singular future boundary.
The proof of the above theorem is along the following lines. It is first shown
that if Γ0 is complete, the maximal future development is also complete. Thus
one can assume that Γ0 has a singular end point e. We then consider C e , the
boundary of the causal past of e. This intersects the initial future null geodesic
cone Co in a sphere S0,e . Given then any sphere S0,1 exterior to S0,e on Co ,
but as close as we wish to S0,e , we consider the incoming null hypersurface
C 1 through S0,1 . Allowing a suitable modification of the initial data as in the
statement of the theorem, with f a function vanishing in the interior of S0,e
on Co , it is then shown that there exists a point p0 on Γ0 , earlier than e, such
that the annular region on the future null geodesic cone Cp0 , with vertex at p0 ,
bounded by the intersections with C e and C 1 satisfies the hypotheses of the
12
theorem of [Chr6]. It is in this part of the proof that the singular nature of
the point e is used. Application of the theorem of [Chr6] then shows that if
we consider future null geodesic cones Cp with vertices p on the segment of Γ0
between p0 and e, and the corresponding intersections with C e and C 1 , then
for some p∗ in this segment earlier than e, Cp∗ C 1 is a maximal sphere in
T
Cp∗ , and the part of C 1 to the future of this sphere lies in a trapped region.
We see therefore the essential role played by the formation of trapped spheres
theorem of [Chr6] in the proof of the cosmic censorship conjectures in [Chr9] in
the framework of the spherically symmetric scalar field model.
A model, closely related to the scalar field model but with surprizing new
features, was studied by Dafermos in [D1], [D2]. In this model we have in
addition to the scalar field an electromagnetic field. The two fields are only
indirectly coupled, through their interaction with the gravitational field, the
energy-momentum stress tensor of matter being the sum of 1 with the Maxwell
energy-momentum-stress tensor for the electromagnetic field. By the imposi-
tion of spherical symmetry, the electromagnetic field is simply the Coulomb
field corresponding to a constant charge Q. This is non-vanishing by virtue of
the fact that the topology of the manifold is ℜ2 × S 2 , like the manifold of the
Schwarzschild solution, so there are spheres which are not homologous to zero.
Dafermos showed that in this case part of the boundary of the maximal devel-
opment is a Cauchy horizon, through which the metric can be continued in a
C 0 manner, but at which, generically, the mass function blows up. As a conse-
quence, generically, there is no local coordinate system in any neighborhood of
any point on the Cauchy horizon in which the connection coefficients (Christof-
fel symbols) are square integrable. This means that the solution ceases to make
sense even as a weak solution of the Einstein-Maxwell-scalar field equations if we
attempt to include the boundary. The work of Dafermos illustrates how much
care is needed in formulating the strong cosmic censorship conjecture. In partic-
ular the formulation given in [Chr10] according to which C 0 extensions through
the boundary of the maximal development are generically excluded, turned out
to be incorrect. Only if the condition is added that there be no extension as
a solution, even in a weak sense, to include any part of the boundary, is the
counterexample avoided.
Before the work on the scalar field model was completed, the author intro-
duced and studied a model which was designed to capture some of the features
of actual stellar gravitational collapse while capitalizing to a maximum extent
on the knowledge gained in the study of the dust and scalar field models. This
was the two-phase model, introduced in [Chr11] and studied further in [Chr12]
and [Chr13]. Let us recall here that a perfect fluid model is in general defined
by specifying a function e(n, s), the energy per particle as a function of n, the
number of particles per unit volume, and s, the entropy per particle. This is
called the equation of state. Then the mass-energy density ρ, the pressure p and
the temperature θ are given by:
∂e ∂e
ρ = ne, p = n2 , θ= (2)
∂n ∂s
13
The mechanics of a perfect fluid are governed by the differential conservation
laws
∇ν T µν = 0, ∇µ I µ = 0 (3)
where T µν is the energy-momentum-stress tensor
T µν = ρuµ uν + p((g −1 )µν + uµ uν ), (4)
uµ being the fluid velocity, and
I µ = nuµ (5)
is the particle current. In the case of the two-phase model, p is a function of

ρ alone. For such fluids, called barotropic, p and ρ are functions of the single
variable
µ = nm(s) (6)
where m(s) a positive increasing function of s. In the two-phase model, if ρ is
less than a critical value, which by proper choice of units we may set equal to 1,
the matter is as soft as possible, the sound speed being equal to 0, while if ρ is
greater than 1, the matter is as hard as possible, the sound speed being equal
to 1, that is, to the speed of light in vacuum. Let us recall here that the sound
speed η is in general given by:

dp
η2 = (7)
dρ s
The pressure in the two-phase model is then given by:

0 : if ρ ≤ 1
p= (8)
ρ − 1 : if ρ > 1
The condition of spherical symmetry being imposed, the flow is irrotational.
The soft phase of the two-phase model coincides with the dust model while
the hard phase coincides with the scalar field model with the restriction that
−(g −1 )µν ∂ν φ be a future-directed timelike vectorfield. With
σ = −(g −1 )µν ∂µ φ∂ν φ, (9)
the density of mass-energy ρ and the fluid velocity uµ are given by:
1 (g −1 )µν ∂ν φ
ρ= (σ + 1), uµ = − √ (10)
2 σ
The energy-momentum-stress tensor in the hard phase is:
1
Tµν = ∂µ φ∂ν φ + (σ − 1)gµν (11)
2
so it differs from the standard one for a scalar field 1 by the term −(1/2)gµν , the
divergence-free condition on Tµν being again equivalent to the wave equation
for φ in the metric g.
14
Each of the two phases is by itself incomplete, the soft phase being limited
by the condition ρ ≤ 1 and the hard phase being limited by the condition σ ≥ 1.
The soft phase turns in contraction into the hard phase, while the hard phase
turns upon expansion into the soft phase. Only the two phases taken together
constitute a complete model. The hypersurface which forms the interface be-
tween the two phases has both spacelike and timelike components. Across a
spacelike component, the thermodynamic variables n, s or ρ, p and the fluid ve-
locity uµ are continuous, the final values of one phase providing the initial values
for the next phase. However, across a timelike component the thermodynamic
variables and the fluid velocity suffer discontinuities, determined by the integral
form of the conservation laws. These are of an irreversible character, each point
of a timelike component which is crossed by a flow line being a point of increase
of the entropy. A timelike component of the phase boundary is therefore a
shock, and the development of these shocks is a free boundary problem, which
was studied in [Chr12] and [Chr13]. With initial condition an inhomogeneous
dust ball at zero entropy these papers showed that the core of the ball turns con-
tinuously into the hard phase, however at a certain sphere a shock forms which
propagates outwards absorbing the exterior part of the original dust ball. Be-
hind this shock we have the hard phase at positive entropy, but the analysis was
not carried further to investigate under what initial conditions a black hole will
eventually form. We should also mention here that the two-phase model admits
a one parameter family of static solutions, balls of the hard phase, surrounded
by vacuum.
The goal of the effort in the field of relativistic gravitational collapse is the
study of the formation of black holes and singularities for general asymptotically
flat initial conditions, that is, when no symmetry conditions are imposed.
In this connection, an interesting theorem was established by Schoen and Yau
[S-Y1], as an outgrowth of their proof of the general case of the positive mass
theorem [S-Y2] (their earlier work [S-Y1] covered the case of a maximal spacelike
hypersurface of vanishing linear momentum; a different proof of the general
theorem was subsequently given by Witten [Wi]). In [S-Y1] it is shown that if
the energy density minus the magnitude of the momentum density of matter on a
spacelike hypersurface is everywhere bounded from below by a positive constant
b in a region which is large enough in a suitable sense which roughly corresponds
to linear dimensions of at least b−1/2 , then the spacelike hypersurface must
contain a closed trapped surface diffeomorphic to S 2 . Although this work does
not address the problem of evolution, the constraint equations alone entering
the proof, it is nevertheless relevant for the problem of evolution, in so far as it
reduces the problem of the dymamical formation of a closed trapped surface to
the problem of showing that, under suitable circumstances, the required material
energy concentration eventually occurs.
As far as the problem of evolution itself, let us first discuss the case where
the material model is a perfect fluid. Then, as we have seen in the spheri-
cally symmetric case, before closed trapped surfaces form, shock waves already
form. Now, the general problem of shock formation in a relativistic fluid, in the
physical case of 3 spatial dimensions has recently been studied by the author
15
in the monograph [Chr14]. This work is in the framework of special relativ-
ity. We should remark here that the only previous result in relation to shock
formation in 3 spatial dimensions was the result of Sideris [Si] which considers
the non-relativistic problem of a classical ideal gas with adiabatic index γ > 1.
Moreover, in that work it is only shown that the solutions cannot remain C 1
for all time, no information being given as to the nature of the breakdown. The
theorems proved in the monograph [Chr14] give, on the other hand, a detailed
picture of shock formation. In particular a detailed description is given of the
geometry of the boundary of the maximal development of the initial data and
of the behavior of the solution at this boundary. The notion of maximal devel-
opment in this context is not that relative to the background Minkowski metric
gµν , but rather the one relative to the acoustical metric
hµν = gµν + (1 − η 2 )uµ uν , uµ = gµν uν (12)
a Lorenzian metric, the null cones of which are the sound cones. It is not
appropriate to give here a complete summary of the results of [Chr14]. Instead,
the following short discussion should suffice to give the reader a feeling for the
present status of shock wave theory in the physical case of 3 spatial dimensions.
In [Chr14] it is shown that the boundary of the maximal development in the
above “acoustical” sense consists of a regular part and a singular part. Each
component of the regular part C is an incoming characteristic (relative to h)
hypersurface which has a singular past boundary. The singular part of the
boundary is the locus of points where the density of foliations by outgoing S
characteristic (relative to h) hypersurfaces blows up. It is the union ∂− B B,
where each component of ∂− B is a smooth embedded surface in Minkowski
spacetime, the tangent plane to which at each point is contained in the exterior
of the sound cone at that point. On the other hand, each component of B is a
smooth embedded hypersurface in Minkowski spacetime, the tangent hyperplane
to which at each point is contained in the exterior of the sound cone at that
point, with the exception of a single generator of the sound cone, which lies
on on the hyperplane itself. The past boundary of a component of B is the
corresponding component of ∂− B. The latter is at the same time the past
boundary of a component of C. This is the surface where a shock begins to
form. Now the maximal development in the acoustical sense, orS “maximal
classical solution”, is the physical solution of the problem up to C ∂− B, but
not up to B. In the last part of [Chr14] the problem of the physical continuation
of the solution is set up as the shock development problem. This a free boundary
problem associated to each component of ∂− B. In this problem one is required
to construct a hypersurface of discontinuity K, the shock, lying in the past of
the corresponding componentof B but having the same past boundary as the
latter, namely the given component of ∂− B, the tangent hyperplanes to K and
B coinciding along ∂− B. Moreover, one is required to construct a solution of the
differential conservation
S laws in the domain in Minkowski spacetime bounded
S in
the past by C K, agreeing with the maximal classical solution on C ∂− B,
while having jumps across K relative to the data induced on K by the maximal
classical solution, jumps satisfying the jump conditions which follow from the
16
integral form of the conservation laws. Finally, K is required to be spacelike
relative to the acoustical metic induced by the maximal classical solution, which
holds in the past of K, and timelike relative to the new solution, which holds
in the future of K.S The maximal classical solution thus provides the boundary
conditions on C ∂− B, as well as a barrier at B.
The shock development problem is only set up, not solved, in [Chr14]. The
author plans to address this problem in the near future. One final result of
[Chr14] needs to be mentioned here however. In the context of [Chr14], the
solution is irrotational up to K. At the end of that monograph a formula is
derived for the jump in vorticity across K of a solution of the shock development
problem. This formula shows that while the flow is irrotational ahead of the
shock, it acquires vorticity immediately behind.
This brings us to another problem. This is the problem of the long time
behavior of the vorticity along the fluid flow lines. By reason of the result just
quoted, this problem must be also be solved to achieve an understanding of the
dynamics, even when the initial conditions are restricted to be irrotational. Now,
even in the non-relativistic case, and even in the case that the compressibility
of the fluid is neglected, this is a very difficult problem. Indeed, the problem,
in the context of the incompressible Euler equations, of whether or not the
vorticity blows up in finite time along some flow lines, is one of the great unsolved
problems of mathematics (see [Co]).
In conclusion, it is clear that the above basic fluid mechanical problems must
be solved first, before any attempt is made to address the problem of the general
non-spherical gravitational collapse of a perfect fluid in general relativity.
However, once the restriction to spherical symmetry is removed, the dynam-
ical degrees of freedom of the gravitational field itself come into play, and the
thought strikes one that we may not need matter at all to form black holes.
Even in vacuum closed trapped surfaces could perhaps be formed by the focus-
ing of sufficiently strong incoming gravitational waves. It is in fact this problem
which John Wheeler related to the author back in 1968: the formation of black
holes in pure general relativity, by the focusing of incoming gravitational waves.
And it is this problem the complete solution of which is found in the present
monograph. Because of the absence of spherically symmetric solutions of the
vacuum Einstein equations other than the Schwarzchild solution, the problem
in question was far out of reach at that time, and for this reason John Wheeler
advised the author to consider instead the spherically symmetric scalar field
problem as a model problem, by solving which insights would be gained which
would prepare us to attack the original problem. Indeed there is some analogy
between scalar waves and gravitational waves, but whereas a scalar field is a
fiction introduced only for pedagogical reasons, gravitational waves are a fun-
damental aspect of physical reality. We should remember here the remarks of
Einstein in regard to the two sides of his equations. The right hand side, which
involves the energy-momentum-stress tensor of matter, he called “wood”, while
the left hand side, the Ricci curvature, he called “marble”, recalling, perhaps,
the simplicity of an ancient Greek temple.
We shall now state the simplest version of the theorem on the formation of
17
closed trapped surfaces in pure general relativity which this monograph estab-
lishes. This is the limiting version, where we have an asymptotic characteristic
initial value problem with initial data at past null infinity. Denoting by u the
“advanced time”, it is assumed that the initial data are trivial for u ≤ 0. Our
methods allow us to replace this assumption by a suitable falloff condition in
|u| for u ≤ 0, thereby extending the theorem. This would introduce no new
difficulties of principle, but would require more technical work, which would
have considerably lengthened the monograph, obscuring the main new ideas.
Let k, l be positive constants, k > 1, l < 1. Let us be given smooth asymp-
totic initial data at past null infinity which is trivial for advanced time u ≤ 0.
Suppose that the incoming energy per unit solid angle in each direction in the
advanced time interval [0, δ] is not less than k/8π. Then if δ is suitably small,
the maximal development of the data contains a closed trapped surface S which
is diffeomorphic to S 2 and has area
Area(S) ≥ 4πl2
The form of the smallness assumption on δ is specified in the precise form

of the theorem, stated in Chapter 17. We remark that by virtue of the scale
invariance of the vacuum Einstein equations, the theorem holds with k, l, and
δ, replaced by ak, al, and aδ, respectively, for any positive constant a.
The above theorem is obtained through a theorem in which the initial data is
given on a complete future null geodesic cone Co . The generators of the cone are
parametrized by an affine parameter s measured from the vertex o and defined
so that the corresponding null geodesic vectorfield has projection T at o along
a fixed unit future-directed timelike vector T at o. It is assumed that the initial
data are trivial for s ≤ r0 , for some r0 > 1. The boundary of this trivial region
is then a round sphere of radius r0 . The advanced time u is then defined along
Co by
u = r − r0 (13)
The formation of closed trapped surfaces theorem is similar in this case, the only
difference being that the “incoming energy per unit solid angle in each direction
in the advanced time interval [0, δ]”, a notion defined only at past null infinity,
is replaced by the integral
r02 δ
Z
edu (14)
8π 0
on the affine parameter segment [r0 , r0 +δ] of each generator of Co . The function
e is an invariant of the conformal intrinsic geometry of Co , given by:
1 2
e= |χ̂| (15)
2 /g
where g/ is the induced metric on the sections of Co corresponding to constant
values of the affine parameter, and χ̂ is the shear of these sections, the trace-free
part of their 2nd fundamental form relative to Co . The theorem for a cone Co is
18
established for any r0 > 1 and the smallness condition on δ is independent of r0 .
The domain of dependence, in the maximal development, of the trivial region
in Co is a domain in Minkowski spacetime bounded in the past by the trivial
part of Co and in the future by C e , the past null geodesic cone of a point e at
arc length 2r0 along the timelike geodesic Γ0 from o with tangent vector T at
o. Considering then the corresponding complete timelike geodesic in Minkowski
spacetime, fixing the origin on this geodesic to be the midpoint of the segment
between o and e, a segment of arc length 2r0 , the limiting form of the theorem
is obtained by letting r0 → ∞, keeping the origin fixed, so that o tends to the
infinite past along the timelike geodesic.
The theorem on the formation of closed trapped surfaces in this monograph
may be compared to the corresponding theorem in [Chr6] for the spherically
symmetric scalar field problem quoted above. In sharp contrast to that theorem
however, here almost all the work goes into establishing an existence theorem
for a development of the initial data which extends far enough into the future
so that trapped spheres have a evenlually chance to form within this develop-
ment. This theorem is first stated as Theorem 12.1 in the way in which it is
actually proved, and then restated as Theorem 16.1, in the way in which it can
most readily be applied, after the proof is completed. So all chapters of this
monograph, with the exception of the last (and shortest) chapter, are devoted
to the proof of the existence theorem. On the other hand, there is a wealth
of information in Theorem 16.1, which gives us full knowledge of the geome-
try of spacetime when closed trapped surfaces begin to form. The theorems
established in this monograph thus constitute the first foray into the long time
dynamics of general relativity in the large, that is, when the initial data are no
longer confined to a suitably small neighborhood of Minkowskian data. How-
ever, the existence theorem which we establish does not cover the whole of the
maximal development, and for this reason the question regarding the nature of
the future boundary of the maximal development is left unanswered.
We shall now give a brief discussion of the mathematical methods employed
in this monograph, for, as is generally acknowledged, the methods in a mathe-
matical work are often more important than the results. This monograph relies
on three methods, two of which stem from the author’s work with Klainerman
[C-K] on the stability of the Minkowski spacetime, and the third method is new.
We shall first summarize the first two methods.
The work [C-K] which established the global nonlinear stability of the Minkowski
spacetime of special relativity within the framework of general relativity, was
a work within pure general relativity, concerned, like the present one, with the
“marble side” of Einstein’s equations, the “wood” side having been set equal
to zero. Two where the main mathematical methods employed, and they were
both new at the time when the work was composed. The first method was pecu-
liar to Einstein’s equations, while the second had wider application, and could,
in principle, be extended to all Euler-Lagrange systems of partial differential
equations of hyperbolic type.
The first method was a way of looking at Einstein’s equations which allowed
estimates for the spacetime curvature to be obtained. A full exposition of this
19
method is given in Chapter 12, which is also self-contained, except for Proposi-
tions 12.1, 12.5 and 12.6, which are quoted directly from [C-K]. Only the barest
outline of the chief features will be given here. The method applies also in the
presence of matter, to obtain the required estimates for the spacetime curva-
ture. Its present form is dependent on the 4-dimensional nature of the spacetime
manifold, although a generalization to higher dimensions can be found.
Instead of considering the Einstein equations themselves, we considered the
Bianchi identities in the form which they assume by virtue of the Einstein
equations. We then introduced the general concept of a Weyl field W on a 4-
dimensional Lorentzian manifold (M, g) to be a 4-covariant tensorfield with the
algebraic properties of the Weyl or conformal curvature tensor. Given a Weyl
field W one can define a left dual ∗ W as well as a right dual W ∗ , but as a
consequence of the algebraic properties of a Weyl field the two duals coincide.
Moreover, ∗ W = W ∗ is also a Weyl field. A Weyl field is subject to equations
which are analogues of Maxwell’s equations for the electromagetic field. These
are linear equations, in general inhomogeneous, which we call Bianchi equations.
They are of the form:
∇α Wαβγδ = Jβγδ (16)
the right hand side J, or more generally any 3-covariant tensorfield with the
algebraic properties of the right hand side, we call a Weyl current. These equa-
tions seem at first sight to be the analogues of only half of Maxwell’s equations,
but it turns out that they are equivalent to the equations
1 µν
∇[α Wβγ]δǫ = ǫµαβγ J ∗µδǫ , ∗
Jβγδ = J ǫµνγδ (17)
2 β
which are analogues of the other half of Maxwell’s equations. Here ǫ is the
volume 4-form of (M, g). The fundamental Weyl field is the Riemann curvature
tensor of (M, g), (M, g) being a solution of the vacuum Einstein equations, and
in this case the corresponding Weyl current vanishes, the Bianchi equations
reducing to the Bianchi identities.
Given a vectorfield Y and a Weyl field W or Weyl current J there is a
“variation” of W and J with respect to Y , a modified Lie derivative L̃Y W ,
L̃Y J, which is also a Weyl field or Weyl current respectively. The modified Lie
derivative commutes with duality. The Bianchi equations have certain confor-
mal covariance properies which imply the following. If J is the Weyl current
associated to the Weyl field W according to the Bianchi equations, then the
Weyl current associated to L̃Y W is the sum of L̃Y J and a bilinear expression
which is on one hand linear in (Y ) π̃ and its first covariant derivative and other
the other hand in W and its first covariant derivative (see Proposition 12.1).
Here we denote by (Y ) π̃ the deformation tensor of Y , namely the trace-free part
of the Lie derivative of the metric g with respect to Y . This measures the rate
of change of the conformal geometry of (M, g) under the flow generated by Y .
From the fundamental Weyl field, the Riemann curvature tensor of (M, g), and
a set of vector fields Y1 , ..., Yn which we call commutation fields, derived Weyl
fields of up to any given order m are generated by the repeated application of
20
the operators L̃Yi : i = 1, ..., n. A basic requirement on the set of commutation
fields is that it spans the tangent space to M at each point. The Weyl currents
associated to these derived Weyl fields are then determined by the deformation
tensors of the commutation fields.
Given a Weyl field W there is a 4-covariant tensorfield Q(W ) associated to
W , which is symmetric and trace-free in any pair of indices. It is a quadratic
expression in W , analogous to the Maxwell energy-monentum-stress tensor for
the electromagnetic field. We call Q(W ) the Bel-Robinson tensor associated to
W , because it had been discovered by Bel and Robinson [Be] in the case of the
fundamental Weyl field, the Riemann curvature tensor of a solution of the vac-
uum Einstein equations. The Bel-Robinson tensor has a remarkable positivity
property: Q(W )(X1 , X2 , X3 , X4 ) is non-negative for any tetrad X1 , X2 , X3 , X4
of future directed non-spacelike vectors at a point. Moreover, the divergence of
Q(W ) is a bilinear expression which is linear in W and in the associated Weyl
current J (see Proposition 12.6). Given a Weyl field W and a triplet of future
directed non-spacelike vectorfields X1 , X2 , X3 , which we call multiplier fields we
define the energy-momentum density vectorfield P (W ; X1 , X2 , X3 ) associated to
W and to the triplet X1 , X2 , X3 by:
β γ δ
P (W ; X1 , X2 , X3 )α = −Q(W )α
βγδ X1 X2 X3 (18)
Then the divergence of P (W ; X1 , X2 , X3 ) is the sum of −(divQ(W ))(X1 , X2 , X3 )

and a bilinear expression which is linear in Q(W ) and in the deformation tensors
of X1 , X2 , X3 . The divergence theorem in spacetime, applied to a domain which
is a development of part of the initial hypersurface, then expresses the integral
of the 3-form dual to P (W ; X1 , X2 , X3 ) on the future boundary of this domain,
in terms of the integral of the same 3-form on the past boundary of the domain,
namely on the part of the initial hypersurface, and the spacetime integral of the
divergence. The boundaries being achronal (that is, no pair of points on each
boundary can be joined by a timelike curve) the integrals are integrals of non-
negative functions, by virtue of the positivity property of Q(W ). For the set of
Weyl fields of order up to m which are derived from the fundamental Weyl field,
the Riemann curvature tensor of (M, g), the divergences are determined by the
deformation tensors of the commutation fields and their derivatives up to order
m, and from the deformation tensors of the multiplier fields. And the integrals
on the future boundary give control of all the derivatives of the curvature up to
order m. This is how estimates for the spacetime curvature are obtained, once
a suitable set of multiplier fields and a suitable set of commutation fields have
been provided.
This is precisely where the second method comes in. This method constructs
the required sets of vectorfields by using the geometry of the two parameter
foliation of spacetime by the level sets of two functions. These two functions,
in the first realization of this method in [C-K], where the time function t, the
level sets of which are maximal spacelike hypersurfaces Ht of vanishing total
momentum, and the optical function u, which we may think of as “retarded
time”, the level sets of which are outgoing null hypersurfaces Cu . These are
21
chosen so that density of the foliation of each
T Ht by the traces of the Cu , that
is, by the surfaces of intersection St,u = Ht Cu , which are diffeomorphic to S 2 ,
tends to 1 as t → ∞. In other words, the St,u on each Ht become evenly spaced
in the limit t → ∞. It was already clear at the time of the composition of the
work [C-K] that the two functions did not enter the problem on equal footing.
The optical function u placed a much more important role. This is due to the
fact that the problem involved outgoing waves reaching future null infinity, and
it is the outgoing family of null hypersurfaces Cu which follows these waves.
The role of the family of maximal spacelike hypersurfaces Ht was to obtain a
suitable family of sections of each Cu , the family St,u corresponding to a given
u, to provide the future boundary, or part of the future boundary, of domains
where the divergence theorem is applied, and also to serve as a means by which,
in the proof of the existence theorem, the method of continuity can be applied.
The geometric entities describing the two parameter foliation of spacetime by
the St,u are estimated in terms of the spacetime curvature. This yields estimates
for the deformation tensors of the multiplier fields and the commutation fields
in terms of the spacetime curvature, thus connecting with the first method.
Another realization of this method is found in [Chr14]. There in the role of
the time function we have the Minkowskian time coordinate t which vanishes
on the initial hyperplane. The level sets of this function are then a family Ht
of parallel spacelike hyperplanes in Minkowski spacetime. In the role of the
optical function we have the acoustical function u, the level sets Cu of which are
outgoing characteristic hypersurfaces relative to the acoustical metric h. In this
case however, these are defined by their traces S0,u on the initial hyperplane
H0 , which are diffeomorphic to S 2 . The density of the foliation of each H Tt by
the traces of the Cu , that is, by the surfaces of intersection St,u = Ht Cu ,
in fact blows up in finite time t∗ (u, ϑ) for (u, ϑ) in an open subset of ℜ × S 2 ,
ϑ ∈ S 2 labeling the generators of each Cu , and this defines the singular boundary
B, whose past boundary ∂− B is the surface, not necessarily connected, from
which shocks begin to form. The relative roles of the two functions is even
clearer in this work, because the blow up of the density of foliations by outgoing
characteristic hypersurfaces is what characterizes shock formation.
Returning to general relativity, a variant of the method is obtained if we place
in the role of the time function t another optical function u, which we may think
of as “advanced time”, the level sets of which are incoming null hypersurfaces.
This approach had its origin in the author’s effort to understand the so-called
“memory effect” of gravitational waves [Chr15]. This effect is a manifestation of
the nonlinear nature of the asymptotic gravitational laws at future null infinity.
Now future null infinity is an ideal incoming null hypersurface at infinity, so the
analysis required the consideration of a family of incoming null hypersurfaces
the interiors of the traces of which on the initial spacelike hypersurface H0 give
an exhaustion of H0 . A two parameter family of surfaces diffeomorphic to S 2 ,
the “wave fronts”, was then obtained, namely the intersections of this incoming
family with the outgoing family of null hypersurfaces. A set of notes [Chr16]
was then written up where the basic structure equations of such a “double null”
foliation where derived, and a rough outline was given on how one can proceed
22
to estimate the geometric quantities associated with such a foliation in terms of
the spacetime curvature.
A double null foliation was subsequently employed by Klainerman and Nicolò
in [K-N] (where the aforementioned notes are gracefully acknowledged) to pro-
vide a simpler variant of the exterior part of the proof of the stability of
Minkowski spacetime, namely that part which considers the domain of depen-
dence of the exterior of a compact set in the initial asymtoptically flat spacelike
hypersurface. The developments stemming from the original work [C-K] include
the work of Zipser [Z], which extended the original theorem to the Einstein-
Maxwell equations, and most recently the work of Bieri [Bie], which extended
the theorem in vacuum by requiring a smallness condition only on up to the 1st
derivatives of the Ricci curvature of g0 , the induced metric on H0 , and up to
the 2nd derivatives of k0 , the 2nd fundamental form of H0 , instead of up to the
2nd and up to the 3rd derivatives respectively, as in the original theorem, and
moreover with weights depending on the distance from an origin on (H0 , g0 ),
which are reduced by one power of this distance, relative to the weights assumed
in [C-K].
In the present work, the roles of the two optical functions are reversed,
because we are considering incoming rather than outgoing waves, and it is the
incoming null hypersurfaces C u , the level sets of u, which follow these waves.
However, in the present work, taking the other function to be the conjugate
optical function u is not merely a matter of convenience, but it is essential for
what we wish to achieve. This is because the Cu , the level sets of u, are here,
like the initial hypersurface Co itself, future null geodesic cones with vertices
on the timelike geodesic
T Γ0 , and the trapped spheres which eventually form are
sections Su,u = C u Cu of “late” Cu , everywhere along which those Cu have
negative expansion.
We now come to the new method. This method is a method of treating the
focusing of incoming waves, and like the second method it is of wider application.
A suitable name for this method is short pulse method. Its point of departure
resembles that of the short wavelength or geometric optics approximation, in
so far as it depends on the presence of a small length, but thereafter the two
approaches diverge. The short pulse method is a method which, in the context
of Euler-Lagrange systems of partial differential equations of hyperbolic type,
allows us to establish an existence theorem for a development of the initial
data which is large enough so that interesting things have chance to occur
within this development, if a nonlinear system is involved. One may ask at this
point: what does it mean for a length to be small in the context of the vacuum
Einstein equations? For, the equations are scale invariant. Here small means by
comparison to the area radius of the trapped sphere to be formed.
With initial data on a complete future null geodesic cone Co , as explained
above, which are trivial for s ≤ r0 , we consider the restriction of the initial data
to s ≤ r0 + δ. In terms of the advanced time u, we restrict attention to the
interval [0, δ], the data being trivial for u ≤ 0. The retarded time u is set equal
to u0 = −r0 at o and therefore on Co , which is then also denoted Cu0 . Also,
u − u0 is defined along Γ0 to be one half the arc length from o. This determines
23
u everywhere. The development whose existence we want to establish is that
bounded in the future by the spacelike hypersurface H−1 where u + u = −1
and by the incoming null hypersurface C δ . We denote this development by
M−1 . We define L and L to be the future directed null vectorfields the integral
curves of which are the generators of the Cu and C u , parametrized by u and u
respectively, so that
Lu = Lu = 0, Lu = Lu = 1 (19)
The flow Φτ generated by L defines a diffeomorphism of Su,u onto Su+τ,u , while
the flow Φτ generated by L defines a diffeomorphism of Su,u onto Su,u+τ . The
positive function Ω defined by
g(L, L) = −2Ω2 (20)
may be thought of as the inverse density of the double null foliation. We denote
by L̂ and L̂ the corresponding normalized future directed null vectorfields
L̂ = Ω−1 L, L̂ = Ω−1 L, so that g(L̂, L̂) = −2 (21)
The first step is the analysis of the equations along the initial hypersurface
Cu0 . This analysis is performed in Chapter 2, and it is particularly clear and
simple because of the fact that Cu0 is a null hypersurface, so we are dealing with
the characteristic initial value problem and there is a way of formulating the
problem in terms of free data which are not subject to any constraints. The full
set of data which includes all the curvature components and their transversal
derivatives, up to any given order, along Cu0 , is then determined by integrating
ordinary differential equations along the generators of Cu0 . As we shall see in
Chapter 2, the free data may be described as a 2-covariant symmetric positive
definite tensor density m, of weight -1 and unit determinant, on S 2 , depending
on u. This is of the form:
m = exp ψ (22)
where ψ is a 2-dimensional symmetric trace-free matrix valued “function” on
S 2 , depending on u ∈ [0, δ], and transforming under change of charts on S 2
in such a way so as to make m a 2-covariant tensor density of weight -1. The
transformation rule is particularly simple if stereographic charts on S 2 are used.
Then there is a function O defined on the intersection of the domains of the north
and south polar stereographic charts on S 2 , with values in the 2-dimensional
symmetric orthogonal matrices of determinant -1 such that in going from the
north polar chart to the south polar chart or vise-versa, ψ 7→ ÕψO and m 7→
ÕmO. The crucial ansatz of the short pulse method is the following. We
consider an arbitrary smooth 2-dimensional symmetric trace-free matrix valued
“function” ψ0 on S 2 , depending on s ∈ [0, 1], which extends smoothly by 0 to
s ≤ 0, and we set:
δ 1/2 u
ψ(u, ϑ) = ψ0 , ϑ , (u, ϑ) ∈ [0, δ] × S 2 (23)
|u0 | δ
24
The analysis of the equations along Cu0 then gives, for the components of the
spacetime curvature along Cu0 :
sup |α| ≤ O2 (δ −3/2 |u0 |−1 )

Cu0
sup |β| ≤ O2 (δ −1/2 |u0 |−2 )

Cu0
sup |ρ|, sup |σ| ≤ O3 (|u0 |−3 )

Cu0 Cu0
sup |β| ≤ O4 (δ|u0 |−4 )

Cu0
sup |α| ≤ O5 (δ 3/2 |u0 |−5 ) (24)

Cu0
Here α, α are the trace-free symmetric 2-covariant tensorfields on each Su,u

defined by:
α(X, Y ) = R(X, L̂, Y, L̂), α(X, L̂, Y, L̂) (25)
for any pair of vectors X, Y tangent to Su,u at a point, β, β are the 1-forms on
each Su,u defined by:
1 1
β(X) = R(X, L̂, L̂, L̂), β(X) = R(X, L̂, L̂, L̂) (26)
2 2
and ρ, σ are the functions on each Su,u defined by:
1 1
ρ= R(L̂, L̂, L̂, L̂), R(X, Y, L̂, L̂) = σǫ/(X, Y ) (27)
4 2
for any pair of vectors X, Y tangent to Su,u at a point, /ǫ being the area form of
Su,u . The symbol Ok (δ p |u0 |r ) means the product of δ p |u0 |r with a non-negative
non-decreasing continuous function of the C k norm of ψ0 on [0, 1] × S 2 . The
pointwise magnitudes of tensors on Su,u are with respect to the induced metric
g/, which is positive definite, the surfaces being spacelike. The precise estimates
are given in Chapter 2. We should emphasize here that the role of the ansatz
23 is to obtain estimates of the form 24, that is with the same dependence on δ
and |u0 |, and analogous estimates for the L4 norms on the Su,u0 , u ∈ [0, δ], of
the 1st derivatives of the curvature components, and for the L2 norms on Cu0
of the 2nd derivatives of the curvature components (with the exception of the
2nd transversal derivative of α). If the quantities
supCu0 δ 3/2 |u0 ||α|

supCu0 δ 1/2 |u0 |2 |β|

supCu0 |u0 |3 |ρ| , supCu0 |u0 |3 |σ|

supCu0 δ −1 |u0 |4 |β|

supCu0 δ −3/2 |u0 |9/2 |α| ,

(28)
25
and analogous quantities for the 1st and 2nd derivatives, are assumed to have
bounds which are independent of |u0 | or δ, the ansatz 23 can be dispensed with,
and indeed the chapters following Chapter 2 make no reference to it, until, at the
end of Chapter 16, the existence theorem is restated, after it has been proven,
as Theorem 16.1. However the ansatz 23 is the simplest way to ensure that the
required bounds hold, and there is no loss of generality involved, ψ0 being an
arbitrary “function” on [0, 1] × S 2 with values in the 2-dimensional symmetric
trace-free matrices. Note here that, since |u0 | > 1, what is required of the last
of 28 is weaker than what is provided by the last of 24. A last remark before we
proceed to the main point is that the last three of the estimates 24 require more
than two derivatives of ψ0 , so there is an apparent loss of derivatives from what
would be expected of curvature components. This loss of derivatives is intrinsic
to the characteristic initial value problem and occurs even for the wave equation
in Minkowski spacetime (see [M]). It is due to the fact that one expresses the
full data, which includes transversal derivatives of any order, in terms of the free
data. No such loss of derivatives is present in our spacetime estimates, which
are sharp, and depend only on the L2 norm of up to the 2nd derivatives of the
curvature components on the initial hypersurface (with the exception of the 2nd
transversal derivative of α), precisely as in [C-K]. Nevertheless the initial data
are assumed to be C ∞ in this work, and the solutions which we construct are
also C ∞ .
To come to the main point, the reader should focus on the dependence on
δ of the right hand sides of 24. This displays what we may call the short pulse
hierarchy. And this hierarchy is nonlinear. For, if only the linearized form
of the equations was considered, a different hierarchy would be obtained: the
exponents of δ in the first two of 24 would be the same, but the exponents of δ
in the last three of 24 would instead be 1/2, 3/2, 5/2, respectively.
A question that immediately comes up when one ponders the ansatz 24,
is why is the “amplitude” of the pulse proportional to the square root of the
“length” of the pulse? (the factor |u0 |−1 is the standard decay factor in 3 spatial
dimensions, the square root of the area of the wave fronts). Where does this
relationship come from? Obviously, there is no such relationship in a linear
theory. The answer is that it comes from our desire to form trapped surfaces in
the development M−1 . If a problem involving the focusing of incoming waves
in a different context was the problem under study, for example the formation
of electromagnetic shocks by the focusing of incoming electromagnetic waves in
a nonlinear medium, the relationship between length and amplitude would be
different and it would be dictated by the desire to form such shocks within our
development.
Another remark concerning different applications of the short pulse method,
in particular applications to problems of shock formation, is that it is more
natural in these problems to use in the role of the retarded time u the time
function t whose level sets are parallel spacelike hyperplanes of the background
Minkowski metric, as in [Chr14]. However the analysis of the equations along
an outgoing characteristic hypersurface is indispensable as a crucial step of the
short pulse method, because, once the correct relationship between length and
26
amplitude has been guessed, it is this analysis which yields the short pulse
hierarchy.
The short pulse hierarchy is the key to the existence theorem as well as to
the trapped surface formation theorem. We must still outline however in what
way do we establish that the short pulse hierarchy is preserved in evolution.
This is off course the main step of the short pulse method. What we do is to
reconsider the first two methods previously outlined in the light of the short
pulse hierarchy.
Let us revisit the first method. We take as multiplier fields the vectorfields
L and K, where
K = u2 L (29)
In this monograph, as already metioned above, we take the initial data to be
trivial for u ≤ 0 and as a consequence the spacetime region corresponding to
u ≤ 0 is a domain in Minkowski spacetime. We may thus confine attention to
′
the nontrivial region u ≥ 0. We denote by M−1 this non-trivial region in M−1 .
To extend the theorem to the case where the data is non-trivial for u ≤ 0 but
satisfy a suitable falloff condition in |u|, in the region u ≤ 0 we replace L as a
multiplier field by
L + L = 2T (30)
and redefine K to be:
K = u2 L + u2 L (31)
Since, in any case, a smallness condition can be imposed on the part of the
data corresponding to u ≤ 0, we already know from the work on the stability of
Minkowski spacetime that in the associated domain of dependence, that is, in
the spacetime region u ≤ 0, the solution will satisfy a corresponding smallness
condition. In particular the said smallness condition will be satisfied along C 0 ,
and this suffices for us to proceed with our estimates in the region u ≥ 0 with
the multiplier fields L and K, with K as in 29. So all the difficulty lies in the
′
region M−1 where the pulse travels.
For each of the Weyl fields to be specified below, we define the energy-
momentum density vectorfields
(n)
P (W ) : n = 0, 1, 2, 3 (32)
where:
(0)
P (W ) = P (W ; L, L, L)
(1)
P (W ) = P (W ; K, L, L)
(2)
P (W ) = P (W ; K, K, L)
(3)
P (W ) = P (W ; K, K, K) (33)
As commutation fields we take L, S, defined by:
S = uL + uL, (34)
27
and the three rotation fields Oi : i = 1, 2, 3. The latter are defined according
to the second method as follows. In the Minkowskian region we introduce rect-
angular coordinates xµ : µ = 0, 1, 2, 3, taking the x0 axis to be the timelike
geodesic Γ0 . In the Minkowskian region, in particular on the sphere S0,u0 , the
Oi are the generators of rotations about the xi : i = 1, 2, 3 spatial coordidate
axes. The Oi are then first defined on Cu0 by conjugation with the flow of L
and then in spacetime by conjugation with the flow of L. The Weyl fields which
we consider are, besides the fundamental Weyl field R, the Riemann curvature
tensor, the following derived Weyl fields
1st order: L̃L R, L̃Oi R : i = 1, 2, 3, L̃S R

2nd order: L̃L L̃L R, L̃Oi L̃L R : i = 1, 2, 3, L̃Oj L̃Oi R : i, j = 1, 2, 3,
L̃Oi L̃S R : i = 1, 2, 3, L̃S L̃S R (35)
We assign to each Weyl field the index l according to the number of L̃L operators
in the definition of W in terms of R. We then define total 2nd order energy-
momentum densities
(n)
P2 : n = 0, 1, 2, 3 (36)
(n)
as the sum of δ 2l P (W ) over all the above Weyl fields in the case n = 3, all
the above Weyl fields except those whose definition involves the operator L̃S in
(n)
the cases n = 0, 1, 2. We then define the total 2nd order energies E 2 (u) as the
(n)
integrals on the Cu and the total 2nd order fluxes F 2 (u) as the integrals on
(n) (3)
the C u , of the 3-forms dual to the Of the fluxes only F 2 (u) plays a role
P 2.
in the problem. Finally, with the exponents qn : n = 0, 1, 2, 3 defined by:
1 3
q0 = 1, q1 = 0, q2 = − , q3 = − , (37)
2 2
according to the short pulse hierarchy, we define the quantities

(n) (n) (3) (3)
2qn 2q3
E2 = sup δ E2 (u) : n = 0, 1, 2, 3, F2 = sup δ F2 (u) (38)
u u
The objective then is to obtain bounds for these quantities in terms of the initial
data.
This requires properly estimating the deformation tensor of K, as well as
the deformations tensors of L, S and the Oi : i = 1, 2, 3 and their derivatives of
up to 2nd order. In doing this, the short pulse method meshes with the second
method previously described. This is the content of Chapters 3 - 9 and shall
be very briefly described in the outline of the contents of each chapter which
follows.
The estimates of the error integrals, namely the integrals of the absolute
(n)
values of the divergences of the P 2 , which is the content of Chapters 13 - 15,
28
then yield inequalities for the quantities 38. These inequalities contain, besides
the initial data terms
(n) (n)
2qn
D= δ E2 (u0 ) : n = 0, 1, 2, 3, (39)
terms of O(δ p ) for some p > 0, which are innocuous, as they can be made less
than or equal to 1 by subjecting δ to a suitable smallness condition, but they also
contain terms which are nonlinear in the quantities 38. From such a nonlinear
system of inequalities, no bounds can in general be deduced, because here, in
constrast with [C-K], the initial data quantities are allowed to be arbitrarily
large. However a miracle occurs: our system of inequalities is reductive. That
is, the inequalities, taken in proper sequence, reduce to a sequence of sublinear
inequalities, thus allowing us to obtain the sought for bounds.
We remark that although the first two methods on which the present work
is based stem from the work [C-K], it is only in the present work, in conjunction
with the new method, that the full power of the original methods is revealed.
In applying the short pulse method to problems in other areas of the field of
partial differential equations of hyperbolic type, an analogue of the first method
is needed. This is supplied in the context of Euler-Lagrange systems, that is,
systems of partial differential equations derivable from an action principle, by
the structures studied in [Chr17]. The analogue of the concept of a Weyl field is
the general concept of variation, or variation through solutions. The analogue
of the Bel-Robinson tensor is the canonical stress associated to such variations.
In the area of continuum mechanics or the electrodynamics of continuous media,
the fundamental variation is that with respect to a subgroup of the Poincare
group of the underlying Minkowski spacetime, while the higher order variations
are generated by the commutation fields, as in general relativity (see [Chr14]).
A particularly interesting problem that may be approached on the basis of the
methods which we have discussed, in conjunction with ideas from [Chr14], is the
formation of electromagnetic shocks by the focusing of incoming electromagnetic
waves in isotropic nonlinear media, that is, media with a nonlinear relationship
between the electromagnetic field and the electomagnetic displacement. In this
problem, unlike the problem of shock formation by outgoing compression waves
in fluid mechanics, there is a threshold for shock formation, as there is a threshold
for closed trapped sphere formation in the present monograph.
We shall now give a brief outline of the contents of the different chapters
of this monograph and of their logical connections. The basic geometric con-
struction, the structure equations of the double null foliation called the “optical
structure equations”, and the Bianchi identities, are presented in the introduc-
tory Chapter 1. The Einstein equations are contained in the optical structure
equations. The basic geometric entities associated to the double null foliation
are the inverse density function Ω, the metric g/ induced on the surfaces Su,u and
its Gauss curvature K, the second fundamental forms χ and χ of Su,u relative
to Cu and C u respectively, the torsion forms η and η of Su,u relative to Cu and
C u respectively, and the functions ω and ω, the derivatives of log Ω with respect
29
to L and L respectively. The torsion forms are given by:
η = ζ + /d log Ω, η = −ζ + /d log Ω (40)
where ζ may be caled the torsion. It is the obstruction to integrability of the

distribution of planes orthogonal to the tangent planes to the Su,u . In 40 /d
denotes the differential of the restriction of a function to any given Su,u . The
optical entities χ, χ, η, η, ω, ω are called connection coefficients in the succeeding
chapters, to emphasize their differential order, intermediate between the metric
entities Ω and g/, and the curvature entities K and the spacetime curvature
components. “Canonical coordinates” are defined in the last section of Chapter
1 and play a basic role in this monograph.
The subject of Chapter 2 is the characteristic initial data and the derivation
of the estimates for the full data in terms of the free data. This is where
the ansatz 23 is introduced and the short pulse hierarchy first appears. Thus
Chapter 2 is fundamental to the whole work.
Chapters 3 - 7 form a unity. The subject of these chapters is the derivation of
estimates for the connection coefficients in terms of certain quantities defined by
the spacetime curvature. These chapters are in logical sequence, which extends
to Chapters 8 and 9, however the place of the whole sequence of Chapters 3
- 9 in the logic of the proof of the existence theorem, Theorem 12.1, is after
Chapters 10 and 11. This is because the assumptions on which Chapters 3
- 9 rely, namely the boundedness of the quantities defined by the spacetime
curvature, is established, in the course of the proof of Theorem 12.1, through
the comparison lemmas, Lemmas 12.5 and 12.6, which make use of the results
of Chapters 10 and 11. Thus, Chapter 10 represents a new beginning. The
chapters following Chapter 12 are again in logical sequence.
Chapters 3 - 7 are divided by Chapter 5 into the two pairs of chapters, on one
hand Chapters 3 and 4, and on the other hand Chapters 6 and 7, each of these
two pairs forming a tighter unity. The first pair considers only the propagation
equations among the optical structure equations. These are ordinary differential
equations for the connection coefficients along the generators of the Cu and the
C u . The second pair considers coupled systems, ordinary differential equations
along the generators of the Cu or the C u coupled to elliptic systems on their
Su,u sections. This allows us to obtain estimates for the connection coefficients
which are of one order higher than those obtained through the propagation
equations, and are optimal from the point of view of differentiability. There is
however a loss of a factor of δ 1/2 in behavior with respect to δ, in comparison
to the estimates obtained through the propagation equations, in the case of the
entities η, η and ω. What is crucial, is that there is no such loss in the case
of the entities χ, χ, and ω, but the proof of this fact again uses the former
estimates.
In Chapter 3 the basic L∞ estimates for the connection coefficients are ob-
tained. The last section of Chapter 3 explains the nature of smallness conditions
on δ throughout the monograph. Chapter 4 derives L4 on the surfaces Su,u for
the 1st derivatives of the connection coefficients.
30
Chapter 5 is concerned with the isoperimetric and Sobolev inequalities on
the surfaces Su,u , and with Lp elliptic theory on these surfaces for 2 < p < ∞.
The main part of the chapter is concerned with the proof of the uniformization
theorem for a 2-dimensional Riemannian manifold (S, g/) with S diffeomorphic
to S 2 , when only an L2 bound on the Gauss curvature K is assumed. The
reason why this is required is that although the Gauss equation gives us L∞
control on K, the estimate is not suitable for our purposes because it involves
the loss of a factor of δ 1/2 in behavior with respect to δ. Thus one can only rely
on the estimate obtained by integrating a propagation equation, which although
optimal from the point of view of behavior with respect to δ, only gives us L2
control on K.
The Lp elliptic theory on the Su,u is applied in Chapter 6, in the case p = 4,
to the elliptic systems mentioned above, to obtain L4 estimates for the 2nd
derivatives of the connection coefficients on the surfaces Su,u . What makes
possible the gain of one degree of differentiability by considering systems of
ordinary differential equations along the gerenators of the Cu or the C u coupled
to elliptic systems on the Su,u sections, is the fact that the principal terms in
the propagation equations for certain optical entities vanish, by virtue of the
Einstein equations. In the case of the coupled system pertaining to χ and χ,
these entities are simply the traces trχ and trχ, and the Codazzi equations
constitute the elliptic systems for the trace-free parts χ̂ and χ̂ respectively. In
the case of the coupled systems pertaining to η and η, the entities are found at
one order of differentiation higher. They are the mass aspect functions µ and µ,
called by this name because of the fact, shown in [Chr15], that with r being the
area radius of the Su,u , the limits of the functions r3 µ/8π and r3 µ/8π at past
and future null infinity respectively, represent mass-energy per unit solid angle
in a given direction and at a given advanced or retarded time respectively. The
elliptic systems are Hodge systems constituted by one of the structure equations
and by the definition of µ and µ in terms of η and η respectively. Moreover, the
two sets of coupled systems, that for η on the Cu , and that for η on the C u , are
themselves coupled. (The propagation equations for η and η studied in Chapters
3 and 4 are similarly coupled). In the case of the coupled systems pertaining to
ω and ω, the entities which satify propagation equations in which the principal
terms vanish are found at one order of differentiation still higher. They are the
functions ω / and ω/ and the elliptic equations are simply the definitions of these
functions in terms of ω and ω respectively.
In the case of the χ system we have, besides what has already been de-
scribed, also a coupling with the propagation equation for the Gauss curvature
K, through the elliptic theory of Chapter 5 applied to the Codazzi elliptic sys-
tem for χ̂.
Chapter 7 applies L2 elliptic theory on the Su,u to the same coupled systems
to obtain L2 estimates on the Cu for the third derivatives of the connection
coefficients, the top order needed to obtain a closed system of inequalities in the
proof of the existence theorem.
One general remark concerning the contents of Chapters 3 -7 is that although
some of the general structure was already encountered in the work on the sta-
31
bility of Minkowski spacetime, the estimates and their derivation are here quite
different, and for two reasons. One is the obvious reason that some of the geo-
metric properties are here very different, in view of the fact that we are no longer
confined to a suitably small neighborhood of Minkowski spacetime and closed
trapped surfaces eventually form. The second is the fact, in conncetion with
the short pulse method, that behavior with respect to δ is here all-important.
In Chapter 8 the multiplier fields and the commutation fields are defined
and L∞ estimates for their deformation tensors are obtained. In Chapter 9 L4
estimates on the Su,u for the 1st derivatives of these deformation tensors and
L2 estimates on the Cu for their 2nd derivatives are obtained.
In Chapters 3 - 9 the symbol O(δ p |u|r ) denotes the product of δ p |u|r with
a non-negative, non-decreasing continuous function of certain initial data and
spacetime curvature quantities q1 , ..., qn . The set of quantities {q1 , ..., qn } is
gradually enlarged as we proceed through the sequence of chapters. The set
of quantities is replaced in the seventh section of Chapter 12 by a set which
includes only initial data quantities, and it is in this new sense that the symbol
O(δ p |u|r ) is meant throughout the proof of Theorem 12.1, which occupies the
four succeeding chapters.
As we mention above, Chapter 10 represents a new beginning. The point
is the following. Chapters 3 - 7 derive estimates for the connection coefficients
in terms of quantities involving the L∞ norms on the Su,u of the curvature
components, the L4 norms on the Su,u of the 1st derivatives of the curvature
components, and the L2 norms on the Cu of the 2nd derivatives of the curvature
components (with the exception of those involving α and the 2nd transversal
derivatives of β). The first two are to be estimated in terms of the last through
Sobolev inequalities on the Cu (except for the quantities involving α, which are
estimated in terms of the L2 norm on the C u of up to the 2nd derivatives of that
component through a Sobolev inequality on the C u ), but in establishing these
Sobolev inequalities one cannot rely on the results of the preceeding chapters,
otherwise the reasoning would be circular. So, the Sobolev inequalities on the Cu
and the C u are instead established on the basis of certain bootstrap assumptions.
The sharp form of the Sobolev inequality on the Cu given by Proposition 10.1
fits perfectly with the short pulse method and is essential to its success.
The subject of Chapter 11 is the coercivity properties of the operators L /Oi :
i = 1, 2, 3, the Lie derivatives of covariant tensorfields on the Su,u with respect
the rotation fields Oi : 1 = 1, 2, 3. These inequalities show that, for m = 1, 2
we can bound the sum of the squares of the L2 norms on Su,u of up to the
mth intrinsic to Su,u covariant derivatives of these tensorfields in terms of the
sum of the squares of the L2 norms on the Su,u of their Lie derivatives of
up to mth order with respect to the set of rotation fields. This is important
because only these rotational Lie derivatives of the curvature components (and
the Lie derivatives of the curvature components with respect to L and L), not
their covariant derivatives intrinsic to the Su,u , are directly controlled by the
energies and fluxes. To establish the coercivity inequalities additional bootstrap
assumptions are introduced.
Chapter 12 is the central chapter of the monograph. This chapter lays out
32
the first method and defines the energies and fluxes according to the short pulse
method as discussed above. These definitions are followed by the comparison
lemmas, Lemmas 12.5 and 12.6 which show that the quantity Q′2 , which bounds
all the curvature quantities which enter the estimates for the connection coeffi-
cients and the deformation tensors of Chapters 3 - 9, is itself bounded in terms
of the quantity:
(0) (1) (2) (3) (3)
P2 = max{ E 2 , E 2 , E 2 , E 2 ; F 2 } (41)
To establish the comparison lemmas additional bootstrap assumptions are intro-
duced. The last section of Chapter 12 gives the statement of Theorem 12.1 in the
way it is actually proved, and then gives an outline of the first and most impor-
tant part of the continuity argument, that which concludes with the derivation
(0) (1) (2) (3)
of the reductive system of inequalities for the quantities E 2 , E 2 , E 2 , E 2 and
(3)
F2 in the first section of Chapter 16.
Chapters 13 - 15 deal with the error estimates, namely the estimates for
the error integrals, the spacetime integrals of the absolute values of the diver-
(n)
gences of the energy-momenum density vectorfields P 2 . There are two kinds of
error integrals: the error integrals arising form the deformation tensors of the
multiplier fields and those arising from the Weyl currents generated by the com-
mutation fields. The first are treated in Chapter 13 and the second in Chapers
14 and 15. Because of the delicacy of the final estimates, all the error terms are
treated in a systematic fashion. All error integrals are estimated using Lemma
13.1. The concepts of integrability index and excess index are introduced. The
integrability index s being negative allows us to apply Lemma 13.1. The excess
index e then gives the exponent of δ contributed by the error term under con-
(0) (1) (2) (3)
sideration to the final system of inequalities for the quantities E 2 , E 2 , E 2 , E 2
(3)
and F 2 . All error terms turn out to have a negative integrability index and a
non-negative excess index. The terms with a positive excess index contribute
the innocuous terms O(δ e ) mentioned above. To the terms with zero excess
index are associated borderline error integrals. These contribute the nonlinear
terms to the final system of inequalities mentioned above.
Chapter 16 completes the proof of the existence theorem. The reductive
(0) (1) (2) (3) (3)
system of inequalities for the quantities E 2 , E 2 , E 2 , E 2 and F 2 is obtained in
the first section of Chapter 16, and the required bounds for these quantities are
deduced. The second section of Chapter 16 deduces the higher order bounds
for the spacetime curvature components and the connection coefficients. The
higher order estimates are of linear nature and are needed to show that the
solution extends as a smooth solution. Only the roughest bounds are needed.
The continuity argument is completed in the third section of Chapter 16. In this
section the work of Choquet-Bruhat ([Cho1], [Cho2], [Cho3]) and that of Rendall
[R] are used to obtain a smooth local extension of the solution in “harmonic”
(also called “wave”) coordinates. This is followed by an argument showing
that in a suitably small extension, contained in the extension just mentioned,
33
canonical null coordinates can be set up and the coordinate transformation from
harmonic coordinates to canonical null coordinates is a smooth transformation
with a smooth inverse, hence the metric extends smoothly also in canonical null
coordinates. The proof of the existence theorem is then concluded. In the last
section, the existence theorem is restated in the way in which it can most readily
be applied.
The last chapter, Chapter 17, establishes the theorem on the formation of
closed trapped surfaces, achieving the aim of this monograph.
The present monograph is off course a work in pure mathematics. However,
by virtue of the fact that Einstein’s theory is a physical theory, describing a
fundamental aspect of nature, this work is also of physical significance. For those
mathematicians who, by reading the present monograph, become interested in
the physical basis of general relativity, we recommend the excellent book [M-T-
W] where not only is the physical basis of the theory expained, but also a wealth
of information is given which illustrates how the theory is applied to describe
natural phenomena.
As this work was been completed the author learned that his old teacher,
John Wheeler, passed away. This monograph testifies to John Wheeler’s endur-
ing legacy in the scientific community.
34
Chapter 1
The Optical Structure

Equations
1.1 The Basic Geometric Setup

We consider a spacetime manifold (M, g), with boundary, smooth solution of
the vacuum Einstein equations, such that the past boundary of M is the future
null geodesic cone Co of a point o. (M, g) is to be a development of initial
data on Co . That is, for each point p ∈ M , each past directed causal curve
issuing from p is to terminate in the past at a point on Co . The generators
of Co are the future-directed null geodesics issuing from o. These are affinely
parametrized as follows. There is a future-directed unit timelike vector T at o
such that the tangent vector at o to each generator has projection T along T .
We denote by L′ the null geodesic vectorfield along Co , which is the tangent
field to each generator. The parameter of this vectorfield, measured from o, is
the affine parameter s. Each generator is to extend up to the value r0 + δ of the
parameter, where r0 and δ are constants, r0 > 1, 1 ≥ δ > 0. The null geodesic
cone Co is to contain no conjugate or cut points. For r ∈ (0, r0 + δ], let us
denote by Cor the truncated cone obtained by restricting each generator to the
interval [0, r] of the parameter. Let us also denote by Sor the boundary of Cor ,
the level surface s = r of the function s on Co .
The restrictions of the initial data to Cor0 are to be trivial, that is, they
are to coincide with the data corresponding to a truncated cone in Minkowski
spacetime. By the uniqueness theorem, the domain of dependence of Cor0 in
(M, g) coincides with a domain in Minkowski spacetime. The boundary of this
domain is a null hypersuface generated by the incoming null geodesic normals
from Sor0 . The solution extends as a domain in Minkowski spacetime up to the
past null geodesic cone C e of a point e at arc length 2r0 along the timelike
geodesic Γ0 from o with tangent vector T at o, the aforementioned incoming
null hypersurface from Sor0 coinciding with C e . Let us denote this Minkowskian
domain M0 . Thus the past boundary of M0 is Cor0 and the future boundary of
35
M0 is C e , o and e being respectively the past and future end points of Γ0 , and
the common boundary of Cor0 and C e being the surface Sor0 .
Let q ∈ Γ0 . We denote by Cq the boundary of the causal future of q in
M . Then Cq is an outgoing null hypersurface generated by the future directed
null geodesics issuing from q. Each of these geodesics has a future end point
which is either a point belonging to the future boundary of M or else it is a
point conjugate to q along the geodesic in question, or a cut point, a point at
which the given generator intersects another generator. We then denote Cq by
Cc , c ∈ [−r0 , 0], if q is at arc length 2(r0 + c) along Γ0 from o. In particular,
C−r0 = Co . Now for r ∈ (0, r0 ] the boundary of the domain of dependence of
Cor in M0 is C q , the past null geodesic cone of a point q ∈ Γ0 at arc length
2r from o, so that C q intersects Co in the surface Sor . We denote C q by C c ,
c = r − r0 ∈ (−r0 , 0]. In particular, C 0 = C e . For r ∈ (r0 , r0 + δ], we still
denote by C c , c = r − r0 , the boundary of the domain of dependence of Cor in
M . In this case c ∈ (0, δ] and C c is generated by the incoming null geodesic
normals to Sor , each generator having a future end point which is either a point
belonging to the future boundary of M or else it is a focal point of the incoming
null normal geodesic congruence to Sor along the given generator, or a cut point,
a point at which the given generator intersects another generator.
We now define on M the optical functions u and u by the requirements that
for each c the c-level set of u is the outgoing null hypersurface Cc and for each
c the c-level set of u is the incoming null hypersurface C c . In the following we
denote by Cu and C u an arbitrary level set of u and u respectively. In this sense
u and u denote not the optical functions themselves but values in the range of
the functions u and u respectively. We denote by Su,u the spacelike surfaces of
intersection: \
Su,u = C u Cu (1.1)
These are round spheres for u ∈ (u, 0], which corresponds to the Minkowskian
region M0 , degenerating to points for u = u, which corresponds to Γ0 . Also,
S−r0 +r,−r0 = Sor . The optical functions are increasing toward the future.
Let c∗ ∈ (u0 , −1], u0 = −r0 . For c ∈ (u0 + δ, c∗ ] we denote by Hc the
spacelike hypersurface:
Hc = {Su,u : u + u = c, u ∈ [u, δ)} (1.2)
Also, for c ∈ (2u0 , u0 + δ] we denote by Hc the spacelike hypersurface
Hc = {Su,u : u + u = c, u ∈ [u0 , u]} (1.3)
Note that for c ≤ u0 Hc is a spacelike hyperplane in the Minkowskian region

M0 . S
The future boundary of M is to be the union Hc∗ C δ . Thus, for each
c ∈ [u0 , c∗ − δ] an end point of a generator of Cc which belongs to the future
boundary of M , belongs to C δ , while for each c ∈ (c∗ − δ, c∗ ) an end point of
a generator of Cc which belongs to the future boundary of M , belongs to Hc∗ .
Note that for c ≥ c∗ the generators of Cc are contained in the Minkowskian
36
T
region M M0 hence end on Hc∗ . For each c ∈ (0, δ) a future end point of a
generator of C c which belongs to the future boundary of M , belongs to Hc∗T
. For
c ≤ 0 the generators of C c are contained in the Minkowskian region M M0
hence end on Hc∗ if c > c∗ /2, on Γ0 if c ≤ c∗ /2. We consider M as including
its past boundary but not its future boundary. Thus M corresponds to the
parameter domain:
[
D = {(u, u) : u ∈ [u, δ), u ∈ [u0 , c∗ − δ]}
{(u, u) : u ∈ [u, c∗ − u), u ∈ (c∗ − δ, c∗ /2)}
[
= {(u, u) : u ∈ [u0 , u], u ∈ [u0 , c∗ /2)}
{(u, u) : u ∈ [u0 , c∗ − u), u ∈ [c∗ /2, δ)} (1.4)
for c∗ ≥ u0 + δ and:
D = {(u, u) : u ∈ [u, c∗ − u), u ∈ [u0 , c∗ /2)}

[
= {(u, u) : u ∈ [u0 , u], u ∈ [u0 , c∗ /2)}
{(u, u) : u ∈ [u0 , c∗ − u), u ∈ [c∗ /2, c∗ − u0 )}
(1.5)
for c∗ < u0 + δ. The timelike geodesic Γ0 corresponds to:
A0 = {(u, u) : u = u, u ∈ [u0 , c∗ /2)} (1.6)
(see Figures 1.1, 1.2).

Our approach is a continuity argument in which the basic requirement to be
imposed on (M, g) is that the generators of the Cu and the C u have no end points
in M \ Γ0 . Thus, the Cu are to contain no conjugate or cut points in M and
the C u are to contain no focal or cut points in M \ Γ0 . Then for u ∈ [u0 , c∗ − δ]
the generators of Cu have their end points on C δ and for u ∈ (c∗ − δ, c∗ ) the
generators of Cu have their end points on Hc∗ . Also, for u ∈ [0, δ) the generators
of C u have their end points on Hc∗ . It follows that for (u, u) ∈ D \ A0 Su,u is a
spacelike surface embedded in M .
The optical functions satisfy the eikonal equation:
(g −1 )µν ∂µ u∂ν u = 0, (g −1 )µν ∂µ u∂ν u = 0 (1.7)
It follows that the vectorfields L′ and L′ defined by:
L′µ = −2(g −1 )µν ∂ν u, L′µ = −2(g −1 )µν ∂ν u (1.8)
are future directed null geodesic vectorfields:
∇L′ L′ = 0, ∇L′ L′ = 0 (1.9)
the integral curves of L′ being the generators of each Cu and the integral curves
of L′ being the generators of each C u , both affinely parametrized. For, we have,
37
u + u = c∗
u
u
=
=
δ
0
D′
A0 M0′
0
u
=
u
S
D = M0′ D′
Figure 1.1: Case c∗ ≥ u0 + δ

u
u
=
=
0
u + u = c∗
D′
A0 M0′
0
u
=
u
Figure 1.2: Case c∗ < u0 + δ
38
in arbitrary local coordinates,
gλµ L′ν ∇ν L′µ = −2L′ν ∇ν ∂λ u = −2L′ν ∇λ ∂ν u

= 4(g −1 )νκ ∂κ u∇λ ∂ν u = 2∂λ ((g −1 )νκ ∂ν u∂κ u) = 0
and similarly with u, L′ replaced by u, L′ respectively. Moreover, by virtue of

the condition that 2(r0 + u) along Γ0 coincides with arc length along Γ0 from
o, L′ along Γ0 has projection T along T , the unit future directed tangent field
to Γ0 . Also, L′ coincides along Cu0 with the vectorfield along Co previously
defined. The inner product g(L′ , L′ ) is then a negative function on M , hence
there is a positive function Ω on M defined by:
− g(L′ , L′ ) = 2Ω−2 (1.10)
We then define the normalized null vectorfields L̂ and L̂ by:
L̂ = ΩL′ , L̂ = ΩL′ (1.11)
These shall be used at each point in M \ Γ0 as a basis for the orthogonal

complement of the tangent plane to the surface Su,u through that point. We
have:
g(L̂, L̂) = −2 (1.12)
We proceed to define the null vectorfields L and L by:
L = ΩL̂ = Ω2 L′ , L = ΩL̂ = Ω2 L′ (1.13)
Then from 1.8 and 1.10 we have:
Lu = 0, Lu = 1
Lu = 1, Lu = 0 (1.14)
Thus the integral curves of L are the generators of each Cu parametrized by

u and the integral curves of L are the generators of each C u parametrized by
u. The flow Φt generated by L is defined on any given Cu as follows. Φ0 is
the identity mapping. The mapping Φt is defined at p ∈ Su,u by considering
the unique generator of Cu through p: Φt (p) ∈ Su+t,u is the point along this
generator at parameter value u + t. Φt is a diffeomorphism of Su,u onto Su+t,u .
Similarly, the flow Φt generated by L is defined on any given C u as follows. Φ0
is the identity mapping. The mapping Φt is defined at p ∈ Su,u by considering
the unique generator of C u through p: Φt (p) ∈ Su,u+t is the point along this
generator at parameter value u + t. Φt is a diffeomorphism of Su,u onto Su,u+t .
Note that:
g(L, L) = −2Ω2 (1.15)
39
1.2 The Optical Structure Equations
Let ξ be a 1-form on M \ Γ0 such that
ξ(L) = ξ(L) = 0 (1.16)
We then say that ξ is a S 1-form. Conversely, given for each (u, u) ∈ D, u > u,
a 1-form ξ intrinsic to the surface Su,u , then along each Su,u , ξ extends to the
tangent bundle of M over Su,u by the conditions 1.16. We thus obtain an S
1-form on M \Γ0 . Thus a S 1-form may be thought of as the specification of a 1-
form intrinsic to Su,u for each (u, u). Similar considerations apply to p-covariant
tensorfields. Thus, a p-covariant S tensorfield is a p-covariant tensorfield on
M \ Γ0 which vanishes if either L or L is inserted in one of its p entries, and can
be thought of as the specification of a p-covariant tensorfield intrinsic to Su,u for
each (u, u). Next, a S vectorfield is a vectorfield X defined on M \ Γ0 such that
X is at each point x ∈ M \ Γ0 tangential to the surface Su,u through x. Thus a
S vectorfield may be thought of as the specification of a vectorfield intrinsic to
Su,u for each (u, u). Similarly, a q-contravariant S tensorfield is a q-contravariant
tensorfield W defined on M \ Γ0 such that at each point x ∈ M \ Γ0 Wx belongs
to ⊗q Tx Su,u , and can be thought of as a q contravariant vectorfield intrinsic to
Su,u for each (u, u). Finally, a type Tpq S tensorfield θ is a type Tpq tensorfield
defined on M \ Γ0 such that at each x ∈ M \ Γ0 and each X1 , ..., Xp ∈ Tx M
we have θ(X1 , ..., Xp ) ∈ ⊗q Tx Su,u and θ(X1 , ..., Xp ) = 0 if one of X1 , ..., Xp is
either L or L. Thus a general type Tpq S tensorfield may be thought of as the
specification of a type Tpq tensorfield intrinsic to Su,u for each (u, u).
Let then ξ be a S 1-form. We define L /L ξ to be, for each (u, u) ∈ D, u > u,
the restriction to T Su,u of LL ξ, the Lie derivative of ξ with respect to L, a
notion intrinsic to each Cu . Then L /L ξ is an S 1-form as well. Note that in any
case we have (LL ξ)(L) = 0. We define L /L ξ to be, for each (u, u) ∈ D, u > u, the
restriction to T Su,u of LL ξ, the Lie derivative of ξ with respect to L, a notion
intrinsic to each C u . Then L /L ξ is an S 1-form as well. Note that in any case
we have (LL ξ)(L) = 0. For any p-covariant S tensorfield ξ, the derivatives L /L ξ
and L /L ξ are similarly defined.
Consider next the case of an S vectorfield Y .
Lemma 1.1 Let X be a vectorfield defined along a given Cu and tangential to
its Su,u sections. Then the vectorfield [L, X], defined along Cu is also tangential
to the Su,u sections. Also, let X be a vectorfield defined along a given C u and
tangential to its Su,u sections. Then the vectorfield [L, X], defined along C u is
also tangential to the Su,u sections.
Proof: To establish the first part of the lemma, let Ψs be the 1-parameter group
generated by X. Then Ψs maps each Su,u into itself. Since Φt maps Su,u into
Su+t,u , it follows that Ψ−s ◦ Φ−t ◦ Ψs ◦ Φt maps each Su,u into itself. This must
coincide to order ts as t, s → 0 with Ωts , where Ωr is the 1-parameter group
generated by [L, X]. It follows that [L, X] is tangential to the Su,u . The second
part of the lemma is established in a similar manner.
If then Y is a S vectorfield, according to the above lemma LL Y = [L, Y ] and
LL Y = [L, Y ] are also S vectorfields. We thus define the derivatives L /L Y and
40
L
/L Y to be simply LL Y and LL Y respectively. The case of any q-contravariant
S tensorfield W in the role of Y is formally identical to the case of a S vec-
torfield, such a tensorfield being expressible as the sum of tensor products of S
vectorfields. Thus the derivatives L /L W and L/L W are defined to be simply LL W
and LL W respectively, which are themselves q contravariant S tensorfields.
Consider finally the general case of an arbitrary type Tpq S tensorfield θ. We
may consider, in accordance with the above, θ as being, at each surface Su,u
and at each point x ∈ Su,u , a p-linear form in Tx Su,u with values in ⊗q Tx Su,u .
Then L /L θ is defined by considering θ on each Cu extended to T Cu according to
the condition that it vanishes if one of the entries is L and setting L /L θ equal
to the restriction to the T Su,u of the usual Lie derivative with respect to L of
this extension. Similarly, L /L θ is defined by considering θ on each C u extended
to T C u according to the condition that it vanishes if one of the entries is L and
setting L/L θ equal to the restriction to the T Su,u of the usual Lie derivative with
respect to L of this extension. Then L /L θ and L/L θ are themselves type Tpq S
tensorfields.
In the following we use the simplified notation:
L
/L θ = Dθ, L
/L θ = Dθ (1.17)
In particular, for a function f :
Df = Lf, Df = Lf (1.18)
For any function f defined on M \ Γ0 we denote by /df the S 1-form which

is the restriction to each surface Su,u of df , the differential of f .
Lemma 1.2 For any function f defined on M \ Γ0 we have:
Dd
/f = /dDf, /f = /dDf
Dd
Proof: Consider the first equality. If we evaluate each side on L or L then both
sides vanish by definition. On the other hand, if we evaluate on a S vectorfield
X the left hand side is:
(LL/df )(X) = L((d

/f )(X)) − (d
/f )([L, X]) = L(Xf ) − [L, X]f = X(Lf )
and the right hand side is:
(d
/Lf )(X) = X(Lf )
as well. This establishes the first equality. The second equality is established in
a similar manner.
Let X be a S vectorfield and ξ a S 1-form. Then L /X ξ is defined on each
Su,u as the usual Lie derivative of ξ as a 1-form on Su,u with respect to X as
a vectorfield on Su,u , a notion intrinsic to Su,u , which makes no reference to
the ambient spacetime. If ξ is any p covariant S tensorfield, L /X ξ is similarly
defined. Consider next the case of another S vectorfield Y . Then L /X Y is
defined to be simply LX Y = [X, Y ], which is itself a S vectorfield. The case of
41
any q contravariant S tensorfield W in the role of Y is formally identical to the
case of a S vectorfield, such a tensorfield being expressible as the sum of tensor
products of S vectorfields. Thus L /X W is defined to be simply LX W , which
is itself a q contravariant S tensorfield. Consider finally the general case of an
arbitrary type Tpq S tensorfield θ. Then L /X θ is again defined on each Su,u as
the usual Lie derivative of θ as a type Tpq tensorfield on Su,u with respect to X
as a vectorfield on Su,u , a notion intrinsic to Su,u , which makes no reference to
the ambient spacetime.
If η and θ are any two S tensorfields of any, not necessarily the same, type,
and we denote by η · θ an arbitrary contraction of the tensor product η ⊗ θ, then
the Leibniz rule:
L
/X (η · θ) = (L
/X θ) · θ + η · (L
/X θ) (1.19)
with X an arbitrary S vectorfield, holds, and so do the Leibniz rules:
D(η · θ) = (Dη) · θ + η · (Dθ)

D(η · θ) = (Dη) · θ + η · (Dθ) (1.20)
Let us recall that if X and Y are arbitrary vectorfields and θ an arbitrary

tensorfield on any manifold, we have:
LX LY θ − LY LX θ = L[X,Y ] θ
It then readily follows that if X and Y are S vectorfields and θ a S tensorfield

of any type, we have:
L
/X L
/Y θ − L
/Y L
/X θ = L
/[X,Y ] θ (1.21)
Moreover, we have the following lemma.
Lemma 1.3 Let X be a S vectorfield and θ a S tensorfield of any type. We

then have, noting that by Lemma 1.1 [L, X] and [L, X] are S vectorfields,
DL
/X θ − L
/X Dθ = L
/[L,X] θ
and
DL
/X θ − L
/X Dθ = L
/[L,X] θ
Proof: To establish the first part of the lemma we may confine ourselves to
a given Cu , considering the given Cu to be our manifold. Consider first the
case that θ is a S 1-form. Then if Y is any S vectorfield we have (L
/X θ)(Y ) =
(LX θ)(Y ) however (L/X θ)(L) = 0 while
(LX θ)(L) = −θ([X, L])
Since by 1.14 we have (du)(L) = 1, while (du)(Y ) = 0, the S 1-form L

/X θ is
given on the manifold Cu by:
L
/X θ = LX θ − f du where f = θ([L, X]) (1.22)
42
On the other hand, since (Dθ)(Y ) = (LL θ)(Y ) while (Dθ)(L) = 0 = (LL θ)(L),
the S 1-form Dθ is given on the manifold Cu simply by:
Dθ = LL θ (1.23)
Let then again Y be an arbitrary S vectorfield. From 1.22 we have:
(DL
/X θ)(Y ) = (LL L
/X θ)(Y ) = (LL (LX θ − f du))(Y ) = (LL LX θ)(Y ) (1.24)
the last step by virtue of the facts that (du)(Y ) = 0 and LL du = dLu = 0, by
1.14. Also, from 1.23,
(L
/X Dθ)(Y ) = (LX Dθ)(Y ) = (LX LL θ)(Y ) (1.25)
Subtracting 1.25 from 1.24 we then obtain:
(DL/X θ − L
/X Dθ)(Y ) = (LL LX θ − LX LL θ)(Y ) = (L[L,X] θ)(Y ) = (L /[L,X] θ)(Y )
(1.26)
This is the first part of the lemma in the case of a S 1-form. The case of a p
covariant tensorfield is similar. The case of a q contravariant S tensorfield is
immediate, in view of the fact that in this case the operators D and L /X reduce
to LL and LX respectively. Finally, the general case of a type Tpq S tensorfield
θ follows by taking arbitrary S vectorfields Y1 , ..., Yp , applying the lemma to the
q contravariant tensorfield θ(Y1 , ..., Yp ) and to each of the vectorfields Y1 , ..., Yp ,
and using the Leibniz rules 1.19 and the first of 1.20. (In fact this gives an
alternative proof for the case of a p covariant S tensorfield.)
The second part of the lemma is established in an analogous manner.
From 1.9, 1.11, 1.13 we obtain:

∇L̂ L̂ = ω̂L̂, ∇L̂ L̂ = ω̂ L̂ (1.27)
and:
∇L L = 2ωL, ∇L L = 2ωL (1.28)
where:
ω̂ = Ω−1 ω, ω̂ = Ω−1 ω (1.29)
and:
ω = D log Ω, ω = D log Ω (1.30)
The tangent hyperplane Tp Cu to a given null hypersurface Cu at a point
p ∈ Cu consists of all the vectors X at p which are orthogonal to L̂p :
Tp Cu = {X ∈ Tp M : g(X, L̂p ) = 0} (1.31)
We have L̂p ∈ Tp Cu , L̂p being orthogonal to itself. Similarly, the tangent

hyperplane Tp C u to a given null hypersurface C u at a point p ∈ C u consists of
all the vectors X at p which are orthogonal to L̂p :
Tp C u = {X ∈ Tp M : g(X, L̂p ) = 0} (1.32)
43
and we have L̂p ∈ Tp C u , L̂p being orthogonal to itself.
Thus the induced metrics on Cu and C u are degenerate. This is in fact the
definition of a null hypersurface in a spacetime manifold (M, g). On the other
hand, the induced metric g/ on each surface Su,u , a symmetric 2-covariant S ten-
sorfield, is positive definite. Any vector X ∈ Tp Cu can be uniquely decomposed
into a vector collinear to L̂p and a vector tangent to the Su,u section through p
which we denote by P X:
X = cL̂p + P X, P X ∈ Tp Su,u (1.33)
Thus we have at each p ∈ Cu a projection P of Tp Cu onto Tp Su,u . If X, Y is a

pair of vectors tangent to Cu at p, then:
g(X, Y ) = g/(P X, P Y ) (1.34)
Similarly, any vector X ∈ Tp C u can be uniquely decomposed into a vector

collinear to L̂p and a vector tangent to the Su,u section through p which we
denote by P X:
X = cL̂p + P X, P X ∈ Tp Su,u (1.35)
Thus we have at each p ∈ C u a projection P of Tp C u onto Tp Su,u . If X, Y is a
pair of vectors tangent to C u at p, then:
g(X, Y ) = g/(P X, P Y ) (1.36)
The 2nd fundamental form χ of a given null hypersurface Cu is a bilinear

form in Tp Cu at each p ∈ Cu defined as follows. Let X, Y ∈ Tp Cu . Then:
χ(X, Y ) = g(∇X L̂, Y ) (1.37)
The bilinear form χ is symmetric. For, if we extend X, Y to vectorfields along

Cu which are tangential to Cu , we have, in view of the fact that g(X, L̂) =
g(Y, L̂) = 0,
χ(X, Y ) − χ(Y, X) = −g(L̂, ∇X Y ) + g(L̂, ∇Y X) = −g(L̂, [X, Y ]) = 0
as [X, Y ] is also tangential to Cu hence orthogonal to L̂. We note that the 2nd
fundamental form of Cu is intrinsic to Cu , the vectorfield L̂ being tangential to
Cu . We have:
χ(X, Y ) = χ(P X, P Y ) (1.38)
Thus χ is a symmetric 2-covariant S tensorfield. We may consider χ to be the
2nd fundamental form of the sections Su,u of Cu relative to Cu .
The 2nd fundamental form χ of a given null hypersurface C u is a bilinear
form in Tp C u at each p ∈ C u defined as follows. Let X, Y ∈ Tp C u . Then:
χ(X, Y ) = g(∇X L̂, Y ) (1.39)
44
The bilinear form χ is symmetric. For, if we extend X, Y to vectorfields along
C u which are tangential to C u , we have, in view of the fact that g(X, L̂) =
g(Y, L̂) = 0,
χ(X, Y ) − χ(Y, X) = −g(L̂, ∇X Y ) + g(L̂, ∇Y X) = −g(L̂, [X, Y ]) = 0
as [X, Y ] is also tangential to C u hence orthogonal to L̂. We note that the 2nd
fundamental form of C u is intrinsic to C u , the vectorfield L̂ being tangential to
C u . We have:
χ(X, Y ) = χ(P X, P Y ) (1.40)
Thus χ is a symmetric 2-covariant S tensorfield. We may consider χ to be the
2nd fundamental form of the sections Su,u of C u relative to C u .
We have:
Dg/ = 2Ωχ, Dg/ = 2Ωχ (1.41)
To establish the first of 1.41, let X, Y be vectorfields along Cu which are tan-
gential to the Su,u sections. Then by Lemma 1.1:
(Dg/)(X, Y ) = (LL g/)(X, Y ) = L(g/(X, Y )) − g/([L, X], Y ) − g/(X, [L, Y ])

= L(g(X, Y )) − g([L, X], Y ) − g(X, [L, Y ]) = g(∇X L, Y ) + g(X, ∇Y L)
The second of 1.41 is established in a similar manner. We shall refer to 1.41 as

the first variational formulas.
Let now X, Y be vectors tangent to Su,u at a point p. We extend X, Y to
Jacobi fields along the generator of Cu through p by the conditions:
∇L X = ∇X L, ∇L Y = ∇Y L; that is: [L, X] = [L, Y ] = 0 (1.42)
Then X, Y are at each point tangential to the Su,u section through that point
and we have:
(Dχ)(X, Y ) = (LL χ)(X, Y ) = L(χ(X, Y )) = L(g(∇X L̂, Y ))

= Ωg(∇X L̂, ∇Y L̂) + g(∇L ∇X L̂, Y ) (1.43)
Now, by the definition of the spacetime curvature and 1.42:
∇L ∇X L̂ = ∇X ∇L L̂ + R(L, X)L̂
hence by 1.27:
g(∇L ∇X L̂, Y ) = ωg(∇X L̂, Y ) + g(R(L, X)L̂, Y ) = ωχ(X, Y ) − ΩR(X, L̂, Y, L̂)
(1.44)
In regard to the term g(∇X L̂, ∇Y L̂) we note that the vectors ∇X L̂, ∇Y L̂ are
tangential to Cu , for,
1 1
g(∇X L̂, L̂) = X(g(L̂, L̂)) = 0, g(∇Y L̂, L̂) = Y (g(L̂, L̂)) = 0
2 2
45
Thus 1.34 applies and we obtain:
g(∇X L̂, ∇Y L̂) = g/(P ∇X L̂, P ∇Y L̂) (1.45)
Now by the definition 1.37 we have:
P ∇X L̂ = χ♯ · X, P ∇Y L̂ = χ♯ · Y (1.46)
Here, if θ is any 2-covariant S tensorfield we denote by θ♯ the T11 -type S ten-

sorfield, which at each point p ∈ Su,u is the linear transformation of Tp Su,u
corresponding, through the inner product g/ at p, to the bilinear form θ at p.
That is, for any pair X, Y ∈ Tp Su,u we have:
g/(θ♯ · X, Y ) = θ(X, Y ) (1.47)
In terms of an arbitrary basis (eA : A = 1, 2) for Tp Su,u we have:

♯ C ♯ C
θ ♯ · e A = θA eC , θA = θAB (g/−1 )BC (1.48)
where θAB = θ(eA , eB ), g/AB = g/(eA , eB ) are the components of θ, g/ at p in this

basis. Thus:
g/(P ∇X L̂, P ∇Y L̂) = g/(χ♯ · X, χ♯ · Y ) = (χ × χ)(X, Y ) (1.49)
Here, if θ and θ′ are symmetric 2-covariant S tensorfields we denote by θ × θ′

the 2-covariant S tensorfield defined by:
(θ × θ′ )(X, Y ) = g/(θ♯ · X, θ′♯ · Y ) : for any pair X, Y ∈ Tp Su,u (1.50)
In terms of components in an arbitrary local frame field for Su,u ,

♯C ′♯D
(θ × θ′ )AB = g/CD θA θB = (g/−1 )CD θAC θBD
′
(1.51)
Note that for any symmetric 2-covariant S tensorfield θ we have:
g/ × θ = θ × g/ = θ (1.52)
Let α be the spacetime curvature component which is the symmetric 2-

covariant S tensorfield given by:
α(X, Y ) = R(X, L̂, Y, L̂) : for any pair X, Y ∈ Tp Su,u (1.53)
In view of 1.44, 1.45, 1.49 and 1.53, equation 1.43 holding for an arbitrary pair
X, Y ∈ Tp Su,u implies:
Dχ = ωχ + Ω(χ × χ − α) (1.54)
Let us define χ′ , the 2nd fundamental form of Cu with respect to the null
geodesic field L′ , by:
χ′ (X, Y ) = g(∇X L′ , Y ) : for any pair X, Y ∈ Tp Cu (1.55)
46
We then have:
χ′ = Ω−1 χ (1.56)
In terms of χ′ equation 1.54 takes the form:
Dχ′ = Ω2 χ′ × χ′ − α (1.57)
Let α be the spacetime curvature component which is the symmetric 2-

covariant S tensorfield given by:
α(X, Y ) = R(X, L̂, Y, L̂) : for any pair X, Y ∈ Tp Su,u (1.58)
Proceeding in an analogous manner, with C u , L, χ in the roles of Cu , L, χ re-

spectively we deduce the equation:
Dχ = ωχ + Ω(χ × χ − α) (1.59)
Defining χ′ , the 2nd fundamental form of C u with respect to the null geodesic
field L′ , by:
χ′ (X, Y ) = g(∇X L′ , Y ) : for any pair X, Y ∈ Tp C u (1.60)
we have:
χ′ = Ω−1 χ (1.61)
′
In terms of χ equation 1.59 takes the form:
Dχ′ = Ω2 χ′ × χ′ − α (1.62)
We shall refer to equations 1.54, 1.59, and their equivalents, equations 1.57, 1.62,
as the second variational formulas. The first of equations 1.41 and equation
1.54 or 1.57 are propagation equations along the generators of each Cu , while
the second of equations 1.41 and equation 1.59 or 1.62 are propagation equations
along the generators of each C u .
In the following we call conjugation the formal operation of exchanging Cu
and L with C u and L respectively. We call two entities conjugate if the definition
of one is obtained from the definition of the other by conjugation. Thus ω, χ
and α are the conjugates of ω, χ and α respectively. Also, the operator D is the
conjugate of the operator D. Similarly, we call two formulas conjugate if one is
obtained from the other by conjugation. Thus formulas 1.59 and 1.62 are the
conjugates of formulas 1.54 and 1.59 respectively. By virtue of the symmetry
of the geometric structure under investigation, if a formula holds its conjugate
must also hold.
The torsion of a given surface Su,u with respect to the null hypersurface Cu
is the 1-form ζ on Su,u defined at each p ∈ Su,u by:
1
ζ(X) = g(∇X L̂, L̂) : for any X ∈ Tp Su,u (1.63)
2
47
As this is defined for every (u, u), ζ is defined as a S 1-form on M \ Γ0 . The
entity conjugate to ζ is the torsion of Su,u with respect to C u , the 1-form ζ on
Su,u defined at each p ∈ Su,u by:
1
ζ(X) = g(∇X L̂, L̂) : for any X ∈ Tp Su,u (1.64)
2
However, by 1.12 we have, simply:
ζ = −ζ (1.65)
Let X be a vector tangent to Su,u at a point p. We extend X to a Jacobi

field along the generator of Cu through p by the condition (see 1.42):
∇L X = ∇X L, that is: [L, X] = 0 (1.66)
Then X is at each point tangential to the Su,u section through that point and
we have:
(Dζ)(X) = (LL ζ)(X) = L(ζ(X)) = 21 L(g(∇X L̂, L̂))

= 21 Ωg(∇X L̂, ∇L̂ L̂) + 21 g(∇L ∇X L̂, L̂) (1.67)
∇L ∇X L̂ = ∇X ∇L L̂ + R(L, X)L̂
hence by 1.27:
g(∇L ∇X L̂, L̂) = g(∇X ∇L L̂, L̂) + g(R(L, X)L̂, L̂)

= −2Xω + 2ωζ(X) − ΩR(X, L̂, L̂, L̂) (1.68)
In regard to the term g(∇X L̂, ∇L̂ L̂) we recall that the vector ∇X L̂ is tangential
to Cu hence 1.34 applies. By virtue of 1.46 and the definition 1.63 we then have:
∇X L̂ = −ζ(X)L̂ + χ♯ · X (1.69)
By 1.27 we have:
g(L̂, ∇L̂ L̂) = −g(∇L̂ L̂, L̂) = 2ω̂ (1.70)
To calculate g(Y, ∇L̂ L̂), where Y is a vector at a point q ∈ Cu which is tangential
to the Su,u section through that point, we first extend Y to a Jacobi field along
the generator of Cu through q by the condition:
∇L Y = ∇Y L, that is: [L, Y ] = 0
Then Y is at each point tangential to the Su,u section through that point and
we have:
g(Y, ∇L L̂) = −g(∇Y L, L̂) = 2Y Ω − 2Ωζ(Y ) (1.71)
48
by virtue of the definition 1.63. The equality of the first and last members of
1.71 holds in particular at q. Thus this equality holds for any vector at a point
of Cu which is tangential to the Su,u section through that point, in particular
the vector χ♯ · X. Taking also into account 1.70 we then conclude through 1.69
that:
Ωg(∇X L̂, ∇L̂ L̂) = −2ωζ(X) − 2Ωζ(χ♯ · X) + 2(χ♯ · X)Ω (1.72)
For any T11 type S tensor θ and any S 1-form ξ we define the S 1-form θ · ξ by:
(θ · ξ)(X) = ξ(θ · X) : ∀X ∈ T Su,u (1.73)
or, in terms of an arbitrary local frame field (eA : A = 1, 2) for Su,u ,

B B
(θ · ξ)(eA ) = ξB θA , θ(eA ) = θA eB , ξ(eA ) = ξA (1.74)
The 2nd term on the right in 1.72 is then −2Ω(χ♯ · ζ)(X) while the 3rd term is
2(χ♯ · /dΩ)(X).
Let β be the spacetime curvature component which is S 1-form given by:
1
β(X) = R(X, L̂, L̂, L̂) : for any X ∈ Tp Su,u (1.75)
2
In view of 1.68, 1.72 and the above remarks, equation 1.67 holding for an arbi-
trary vector X ∈ Tp Su,u implies:
/ω − χ♯ · (Ωζ − /dΩ) − Ωβ
Dζ = −d (1.76)
Let β be the spacetime curvature component which is the S 1-form given

by:
1
β(X) = R(X, L̂, L̂, L̂) : for any X ∈ Tp Su,u (1.77)
2
The conjugate of β is then −β. In view of 1.65 the conjugate of equation 1.76
reads:
− Dζ = −d /ω + χ♯ · (Ωζ + /dΩ) + Ωβ (1.78)
Also, the conjugate of equation 1.69 reads:
∇X L̂ = ζ(X)L̂ + χ♯ · X (1.79)
By virtue of Lemma 1.2 and the definitions 1.30 we have:
/dω = Dd
/ log Ω, /dω = Dd
/ log Ω (1.80)
Thus defining the S 1-form η and its conjugate η by:
η = ζ + /d log Ω, η = −ζ + /d log Ω (1.81)
equations 1.76 and 1.78 take the form:
Dη = Ω(χ♯ · η − β) (1.82)
49
Dη = Ω(χ♯ · η + β) (1.83)
Equation 1.82 is a propagation equation along the generators of each Cu , while
equation 1.83 is a propagation equation along the generators of each C u . The
S 1-form η may be considered to be the torsion of Su,u with respect to the null
geodesic field L′ , for,
Ω2
η(X) = g(∇X L′ , L′ ) : ∀X ∈ T Su,u (1.84)
2
Similarly, the S 1-form η may be considered to be the torsion of Su,u with
respect to the null geodesic field L′ , for,
Ω2
η(X) = g(∇X L′ , L′ ) : ∀X ∈ T Su,u (1.85)
2
From 1.70, 1.71 and the definitions 1.81 we obtain:
∇L̂ L̂ = −ω̂L̂ + 2η ♯ (1.86)
the conjugate of which reads:
∇L̂ L̂ = −ω̂L̂ + 2η ♯ (1.87)
Here η ♯ and η ♯ are the S vectorfields corresponding, through the metric g/ to the
S 1-forms η and η respectively. In general if ξ is an S 1-form we denote by ξ ♯
the S vectorfield defined by:
g/(ξ ♯ , X) = ξ(X) : ∀X ∈ T Su,u (1.88)
Thus we can write:

ξ ♯ = g/−1 · ξ (1.89)
Equations 1.86 and 1.87 imply:
∇L L = 2Ω2 η ♯ , ∇L L = 2Ω2 η ♯ (1.90)
hence:
[L, L] = 2Ω2 (η ♯ − η ♯ ) = 4Ω2 ζ ♯ (1.91)
This equation clarifies the geometrical significance of the torsion ζ. It is the
obstruction to integrability of the distribution of timelike planes which is orthog-
onal to the surfaces Su,u . This distribution is spanned by the vectorfields L,
L. As a consequence of the commutation formula 1.91 we have the following
lemma.
Lemma 1.4 Let θ be a S tensorfield of any type. We have:
DDθ − DDθ = L
/4Ω2 ζ ♯ θ
50
Proof: Consider first the case of a S 1-form. For an arbitrary S vectorfield
Y we have (Dθ)(Y ) = (LL θ)(Y ). Also, (Dθ)(L) = 0 = (LL θ)(L). However,
(Dθ)(L) = 0, while
(LL θ)(L) = −θ([L, L])
Since by 1.14 we have (du)(L) = 0, (du)(L) = 1, while (du)(Y ) = 0, the S
1-form Dθ is given by:
Dθ = LL θ − hdu where h = θ([L, L]) (1.92)
Similarly, the S 1-form Dθ is given by:
Dθ = LL θ + hdu (1.93)
Let again Y be an arbitrary S vectorfield. From 1.92 we have:
(DDθ)(Y ) = (LL Dθ)(Y ) = (LL (LL θ − hdu))(Y ) = (LL LL θ)(Y ) (1.94)
1.14. Also, from 1.93 we have:
(DDθ)(Y ) = (LL Dθ)(Y ) = (LL (LL θ + hdu))(Y ) = (LL LL θ)(Y ) (1.95)
1.14. Subtracting 1.95 from 1.94 we the obtain:
(DDθ − DDθ)(Y ) = (LL LL θ − LL LL θ)(Y ) = (L[L,L] θ)(Y )
/4Ω2 ζ ♯ θ)(Y )
= (L4Ω2 ζ ♯ θ)(Y ) = (L (1.96)
by the commutation formula 1.91. This is the lemma in the case of a S 1-form.
The case of a p covariant tensorfield is similar. The case of a q contravariant
S tensorfield is immediate, in view of the fact that in this case the operators
D and D reduce to LL and LL respectively. Finally, the general case of a type
Tpq S tensorfield θ follows by taking arbitrary S vectorfields Y1 , ..., Yp , applying
the lemma to the q contravariant tensorfield θ(Y1 , ..., Yp ) and to each of the
vectorfields Y1 , ..., Yp , and using the Leibniz rules 1.20. (In fact this gives an
alternative proof for the case of a p covariant S tensorfield.)
By 1.28:
1
ω = − Ω−2 g(∇L L, L) (1.97)
4
We have:
Dω = 21 Ω−2 ωg(∇L L, L) − 41 Ω−2 L(g(∇L L, L)) (1.98)
= −2ωω − 41 Ω−2 g(∇L ∇L L, L) − 41 Ω−2 g(∇L L, ∇L L)
By the definition of the spacetime curvature and 1.91 we have:

∇L ∇L L − ∇L ∇L L = R(L, L)L + ∇[L,L] L
= R(L, L)L − 4Ω2 ∇ζ ♯ L (1.99)
51
In view of the fact that by 1.90 g(∇L L, L) = 0, we have:
g(∇L ∇L L, L) = −g(∇L L, ∇L L) = −4Ω4 (η, η) (1.100)
where we have again appealed to 1.90. Here we denote by (ξ, ξ ′ ) the inner
product of the S 1-forms ξ and ξ ′ with respect to the induced metric g/:
(ξ, ξ ′ ) = g/−1 (ξ, ξ ′ ) = g/(ξ ♯ , ξ ′♯ ) (1.101)
Also, in the following, we denote by |ξ| the magnitude of the S 1-form ξ with
respect to g/:
|ξ| = (ξ, ξ)1/2 (1.102)
By 1.64, 1.65:
g(∇ζ ♯ L, L) = −2Ω2 ζ(ζ ♯ ) − 2Ωζ ♯ Ω

= −Ω2 (η − η, η) (1.103)
Defining the function ρ to be the spacetime curvature component:

1
ρ= R(L̂, L̂, L̂, L̂) (1.104)
4
we have:
g(R(L, L)L, L) = 4Ω4 ρ (1.105)
In view of 1.99, 1.100, 1.103, 1.105, we obtain:
1
− Ω−2 g(∇L ∇L L, L) = Ω2 (2(η, η) − |η|2 − ρ) (1.106)
4
Substituting in 1.98 and taking also into account the fact that by 1.28:
1
− Ω−2 g(∇L L, ∇L L) = 2ωω
4
we then obtain:
Dω = Ω2 (2(η, η) − |η|2 − ρ) (1.107)
This is a propagation equation along the generators of each Cu . The conjugate
equation
Dω = Ω2 (2(η, η) − |η|2 − ρ) (1.108)
is a propagation equation along the generators of each C u . Note that ρ is con-
jugate to itself.
/ be the covariant derivative intrinsic to Su,u associated to the induced
Let ∇
metric g/. The operator ∇/ satisfies the following. Let X, Y be S vectorfields.
Then:
∇/X Y = Π∇X Y (1.109)
Here Π is the projection operator to the surfaces Su,u , given by:
1 1
∀V ∈ Tp (M \ Γ0 ) : ΠV = V + g(V, L̂)L̂ + g(V, L̂)L̂ ∈ Tp Su,u (1.110)
2 2
52
Let R
/ be the intrinsic curvature of Su,u and let X, Y, Z, W be S vectorfields.
We have:
R
/(W, Z, X, Y ) = g/(W, R
/(X, Y )Z) = g(W, ∇
/X ∇
/Y Z −∇
/Y ∇
/X Z −∇
/[X,Y ] Z) (1.111)
Now by 1.109, 1.110,

1 1
∇
/Y Z = ∇Y Z + g(∇Y Z, L̂)L̂ + g(∇Y Z, L̂)L̂ (1.112)
2 2
and:
1 1
∇ /Y Z = Π∇X ∇Y Z + g(∇Y Z, L̂)Π∇X L̂ + g(∇Y Z, L̂)Π∇X L̂
/X ∇ (1.113)
2 2
Substituting 1.113 and an analogous formula with X and Y interchanged, as
well as 1.112 with Y replaced by [X, Y ], in 1.111, and recalling that
g(W, ∇X ∇Y Z − ∇Y ∇X Z − ∇[X,Y ] Z) = g(W, R(X, Y )Z) = R(W, Z, X, Y )
then yields:
/(W, Z, X, Y ) = R(W, Z, X, Y ) + 21 g(∇Y Z, L̂)g(W, ∇X L̂) + 21 g(∇Y Z, L̂)g(W, ∇X L̂)

R
− 21 g(∇X Z, L̂)g(W, ∇Y L̂) − 21 g(∇X Z, L̂)g(W, ∇Y L̂)
Noting that
g(∇Y Z, L̂) = −g(Z, ∇Y L̂) = −χ(Y, Z), g(∇Y Z, L̂) = −g(Z, ∇Y L̂) = −χ(Y, Z)
and similarly with Y replaced by X, we then obtain:
/(W, Z, X, Y ) + 21 χ(W, X)χ(Z, Y ) + 12 χ(W, X)χ(Z, Y )

R
− 12 χ(W, Y )χ(Z, X) − 21 χ(W, Y )χ(Z, X) = R(W, Z, X, Y ) (1.114)
Let (eA : A = 1, 2) be a local frame field for Su,u . Setting W = eA , Z = eB ,

X = eC , Y = eD , and denoting R
/(eA , eB , eC , eD ) = R
/ABCD , R(eA , eB , eC , eD ) =
RABCD , 1.114 takes the form:
/ABCD + 21 χAC χBD + 21 χAC χBD

R
− 21 χAD χBC − 12 χAD χBC = RABCD (1.115)
Now, since Su,u is 2-dimensional we have:
R
/ABCD = Kǫ/AB /ǫ CD = K(g/AC g/BD − g/AD g/BC ) (1.116)
where K is the Gauss curvature of Su,u and /ǫ AB = /ǫ (eA , eB ) are the components
of the area 2-form /ǫ of Su,u . Also, there is a function f such that:
RABCD = f /ǫ AB /ǫ CD = f (g/AC g/BD − g/AD g/BC ) (1.117)
53
Then the whole content of equations 1.115 is contained in the equation obtained
by contacting 1.115 with (1/2)(g/−1 )AC (g/−1 )BD , namely the equation:
1 1
K + trχtrχ − (χ, χ) = f (1.118)
2 2
Here if θ is a 2-covariant S tensorfield we denote by trθ its trace with repsect
to g/:
trθ = (g/−1 )AB θAB (1.119)
Also, if θ and θ′ are 2-covariant S tensorfields we denote by (θ, θ′ ) their inner
product with respect to g/:
(θ, θ′ ) = (g/−1 )AC (g/−1 )BD θAB θCD

′
(1.120)
and we denote by |θ| the magnitude of the 2-covariant S tensorfield θ with

respect to g/:
|θ| = (θ, θ)1/2 (1.121)
Note that:
(g/, θ) = trθ (1.122)
′
Also, note that if θ and θ are symmetric 2-covariant S tensorfields then:
tr(θ × θ′ ) = tr(θ′ × θ) = (θ, θ′ ) (1.123)
Now the vacuum Einstein equations read:
Ric = 0 (1.124)
where Ric is the Ricci curvature of (M, g), expressed in an arbitrary local frame
field for (M, g) by:
Ricµν = (g −1 )κλ Rκµλν (1.125)
The reciprocal metric g −1 is expressed in terms of the vectorfields L̂, L̂ and the
reciprocal induced metric g/−1 by:
1 1
g −1 = − L̂ ⊗ L̂ − L̂ ⊗ L̂ + g/−1 (1.126)
2 2
Complementing the frame field (eA : A = 1, 2) with the vectorfields e3 = L̂,

e4 = L̂, to obtain a frame field (eµ : µ = 1, 2, 3, 4) for M , we can express the
definition 1.125 in the form:
1 1
Ricµν = − R3µ4ν − R4µ3ν + (g/−1 )CD RCµDν (1.127)
2 2
In particular, equation RicAB = 0 reads:
1
(R3A4B + R4A3B ) = (g/−1 )CD RCADB = f g/AB (1.128)
2
54
by 1.117. On the other hand, equation Ric34 = 0 reads:
1
(g/−1 )AB RA3B4 = R4334 = −2ρ (1.129)
2
by the definition 1.104. Therefore:
ρ = − 21 (g/−1 )AB RA3B4 = − 41 (g/−1 )AB (RA3B4 + RB3A4 )

= − 41 (g/−1 )(R3A4B + R4A3B ) = −f (1.130)
Thus 1.117 becomes:
RABCD = −ρǫ/AB /ǫ CD = −ρ(g/AC g/BD − g/AD g/BC ) (1.131)
and 1.118 becomes:

1 1
K + trχtrχ − (χ, χ) = −ρ (1.132)
2 2
This is the Gauss equation of the embedding of the surfaces Su,u in the spacetime
manifold (M, g).
Let now ξ be a S 1-form and X, Y be S vectorfields. Then
(∇
/X ξ)(Y ) = X(ξ(Y )) − ξ(∇
/X Y ) = X(ξ(Y )) − ξ(∇X Y ) = (∇X ξ)(Y )
Thus ∇/X ξ is the restriction to T Su,u of ∇X ξ, and the same holds with ξ being
any p-covariant S tensorfield. Let then X, Y, Z be S vectorfields. We have:
(∇
/X χ)(Y, Z) − (∇
/Y χ)(X, Z) = (∇X χ)(Y, Z) − (∇Y χ)(X, Z) (1.133)
= X(χ(Y, Z)) − Y (χ(X, Z)) − χ([X, Y ], Z) − χ(Y, ∇X Z) + χ(X, ∇Y Z)
Now, by the definition 1.37:
X(χ(Y, Z)) − Y (χ(X, Z)) = X(g(∇Y L̂, Z)) − Y (g(∇X L̂, Z)) (1.134)
= g(∇X ∇Y L̂ − ∇Y ∇X L̂, Z) + g(∇Y L̂, ∇X Z) − g(∇X L̂, ∇Y Z)
= g(Z, R(X, Y )L̂) + g(∇[X,Y ] L̂, Z) + g(∇Y L̂, ∇X Z) − g(∇X L̂, ∇Y Z)
Substituting in 1.133 and recalling the definition 1.37 we then obtain:
(∇/X χ)(Y, Z)−(∇/Y χ)(X, Z) = R(Z, L̂, X, Y )+g(∇Y L̂, (∇X Z)⊥ )−g(∇X L̂, (∇Y Z)⊥ )
(1.135)
where for any vector V ∈ Tp (M \ Γ0 ) we denote by V ⊥ the part of V which
lies in the timelike plane which is the orthogonal complement of Tp Su,u . From
1.110:
1 1
V ⊥ = V − ΠV = − g(V, L̂)L̂ − g(V, L̂)L̂ (1.136)
2 2
From the definitions 1.37, 1.39 we have:
g(∇X Z, L̂) = −g(Z, ∇X L̂) = −χ(X, Z), g(∇X Z, L̂) = −g(Z, ∇X L̂) = −χ(X, Z)
55
and similarly with X replaced by Y . Hence from the definition 1.63:
g(∇Y L̂, (∇X Z)⊥ ) = χ(X, Z)ζ(Y )
and similarly with X and Y exchanged. Substituting in 1.135 we then obtain:
(∇
/X χ)(Y, Z) − (∇/Y χ)(X, Z) = R(Z, L̂, X, Y ) + χ(X, Z)ζ(Y ) − χ(Y, Z)ζ(X)
(1.137)
In terms of a local frame field (eA : A = 1, 2) for Su,u complemented by e3 = L̂,
e4 = L̂, setting X = eA , Y = eB , Z = eC , 1.137 takes the form:
∇
/A χBC − ∇
/B χAC = RC4AB + χAC ζB − χBC ζA (1.138)
Since Su,u is 2-dimensional we have:
RC4AB = ξC /ǫ AB (1.139)
where ξC are the components of a S 1-form ξ. The whole content of equations

1.138 is then contained in the equation obtained by contracting with (g/−1 )AC ,
namely the equation:
/A χ♯A
∇ ∗ ♯A
B − eB (trχ) = − ξB + trχζB − χB ζA (1.140)
Here ∗ ξB are the components of the S 1-form ∗

ξ, the left dual, relative to Su,u ,
of the S 1-form ξ:
∗
ξ = /ǫ ♯ · ξ (1.141)
where /ǫ ♯ is the T11 type S tensorfield corresponding through g/ to the area 2-form
/ǫ :
/ǫ ♯A C = /ǫ AB (g/−1 )BC (1.142)
Defining the intrinsic divergence of a symmetric 2-covariant S tensorfield θ to
be the S 1-form div
/ θ with components:
♯A ♯A
(div
/ θ)B = ∇
/ A θB , θB = θBC (g/−1 )CA (1.143)
equation 1.140 reads:
/ χ − /dtrχ = − ∗ ξ + trχζ − χ♯ · ζ
div (1.144)
Now from 1.127 we have:

1
− R344B + (g/−1 )CA RC4AB = Ric4B = 0 (1.145)
2
Thus, in view of the definition 1.75 we have:
∗
− ξB = (g/−1 )CA RC4AB = −βB (1.146)
hence:
ξ = − ∗ β, RC4AB = − ∗ βC /ǫ AB = −g/CA βB + g/CB βA (1.147)
56
and 1.144 becomes:
/ χ − /dtrχ + χ♯ · ζ − trχζ = −β
div (1.148)
The conjugate equation is:
/ χ − /dtrχ − χ♯ · ζ + trχζ = β
div (1.149)
Also, the conjugate of formula 1.147 reads:

∗
RC3AB = β C /ǫ AB = g/CA β B − g/CB β A (1.150)
Equation 1.148 and 1.149 are the Codazzi equations of the embedding of the
surfaces Su,u in the spacetime manifold (M, g). In terms of the forms χ′ , χ′ , η,
η, the Codazzi equations take the form:
/ χ′ − /dtrχ′ + χ′♯ · η − trχ′ η = −Ω−1 β

div (1.151)
/ χ′ − /dtrχ′ + χ′♯ · η − trχ′ η = Ω−1 β

div (1.152)
Let again X, Y be vectors tangent to Su,u at a point p. We extend X, Y to
Jacobi fields along the generator of Cu through p by the conditions 1.42. We
then have:
(D(Ωχ))(X, Y ) = (LL (Ωχ))(X, Y ) = L(Ωχ(X, Y )) = L(g(∇X L, Y ))

= g(∇L ∇X L, Y ) + g(∇X L, ∇Y L) (1.153)
∇L ∇X L = ∇X ∇L L + R(L, X)L
hence by 1.90:
g(∇L ∇X L, Y ) = 2g(∇X (Ω2 η ♯ ), Y ) + g(R(L, X)L, Y ) (1.154)

= 2Ω2 ((∇X η)(Y ) + 2(X log Ω)η(Y )) − Ω2 R(Y, L̂, X, L̂)
Also, by formulas 1.69, 1.79:
∇X L = η(X)L + Ωχ♯ · X, ∇Y L = η(Y )L + Ωχ♯ · Y (1.155)
hence:
g(∇X L, ∇Y L) = −2Ω2 η(X)η(Y ) + Ω2 (χ × χ)(X, Y ) (1.156)
Substituting 1.154 and 1.156 in 1.153 and taking into account the fact that
(∇X η)(Y ) = (∇
/X η)(Y ), and also that by the definitions 1.81 we have:
2d
/ log Ω = η + η (1.157)
57
we obtain:
(D(Ωχ))(X, Y ) = Ω2 {2(∇ /X η)(Y ) + 2η(X)η(Y ) + (χ × χ)(X, Y ) − R(Y, L̂, X, L̂)}

(1.158)
Now the restriction to T Su,u of the 2-form (1/2)R( , , L̂, L̂) is a S 2-form, hence
proportional to the area 2-form /ǫ . Therefore there is a function σ such that :
1
R(X, Y, L̂, L̂) = σǫ/(X, Y ) : for any pair X, Y ∈ Tp Su,u (1.159)
2
The left hand side of 1.159 is equal to
1
(R(X, L̂, Y, L̂) − R(Y, L̂, X, L̂))
2
by the cyclic property of the curvature tensor. On the other hand, from 1.129,
1.131 we have:
1
(R(X, L̂, Y, L̂) + R(Y, L̂, X, L̂) = −ρg/(X, Y ) (1.160)
2
We conclude that:
R(X, L̂, Y, L̂) = −ρg/(X, Y ) + σǫ/(X, Y ) : for any pair X, Y ∈ Tp Su,u (1.161)
Setting X = eA , Y = eB , the antisymmetric part of 1.158 reads, since the left

hand side is symmetric:
1
∇
/A η B − ∇
/B η A = (χ × χ − χ × χ)AB − σǫ/AB (1.162)
2
The whole content of 1.162 is contained in the equation obtained from 1.162 by
contracting with (1/2)ǫ/♯♯AB = (1/2)ǫ/CD (g/−1 )AC (g/−1 )BD , namely the equation:
1
curl
/ η= χ∧χ−σ (1.163)
2
Here we denote by curl
/ ξ the intrinsic curl of a S 1-form ξ:
1
/ ξ = /ǫ ♯♯AB (∇
curl /B ξA ) = /ǫ ♯♯AB ∇
/A ξB − ∇ /A ξB (1.164)
2
If ξ and ξ ′ are S 1-forms we denote:
ξ ∧ ξ ′ = /ǫ ♯♯AB ξA ξB
′
(1.165)
Also, if θ and θ′ are symmetric 2-covariant S tensorfields we denote:
θ ∧ θ′ = /ǫ ♯♯AB (g/−1 )CD θAC θBD

′
(1.166)
Note that:
θ × θ′ − θ′ × θ = θ ∧ θ′/ǫ (1.167)
58
The conjugate of σ being −σ, the conjugate of equation 1.163 is:
1
curl
/ η =− χ∧χ+σ (1.168)
2
Note that by the definitions 1.81:
curl
/ η = curl
/ ζ = −curl
/ η (1.169)
In view of 1.160 the symmetric part of 1.158 is the equation:
D(Ωχ) = Ω2 {∇ ˜ + 2η ⊗ η + (1/2)(χ × χ + χ × χ) + ρg/}

/η + ∇
/η (1.170)
The conjugate equation is:
D(Ωχ) = Ω2 {∇ ˜ + 2η ⊗ η + (1/2)(χ × χ + χ × χ) + ρg/}

/η + ∇
/η (1.171)
Now, by the definitions 1.81:
η = −η + 2d
/ log Ω, η = −η + 2d
/ log Ω
In view of Lemma 1.2 we then have:
Dη = −Dη + 2d
/ω, Dη = −Dη + 2d
/ω (1.172)
or, by 1.82, 1.83,

Dη = −Ω(χ♯ · η + β) + 2d
/ω (1.173)
♯
Dη = −Ω(χ · η − β) + 2d
/ω (1.174)
The first variational formulas 1.41, the second variational formulas 1.57, 1.62,
the propagation equations 1.82, 1.83, 1.107, 1.108, the Gauss equation 1.133,
the Codazzi equations 1.151, 1.152, equations 1.163, 1.169, and equations 1.170,
1.171, together with equations 1.173, 1.174, constitute the optical structure equa-
tions.
Let again (eA : A = 1, 2) be an arbitrary local frame field for the Su,u . We
complement (eA : A = 1, 2), as in the preceding, with the vectorfields e3 = L̂,
e4 = L̂, to obtain a frame field (eµ : µ = 1, 2, 3, 4) for M . In view of equations
1.27, 1.69, 1.80, 1.86, 1.87, and the relation 1.109, the connection coefficients of
the frame field (eµ : µ = 1, 2, 3, 4) are given by the following table:
/A eB + 21 χAB e3 + 12 χAB e4
∇A eB = ∇
∇3 eA = ∇
/3 eA + ηA e3 , / 4 eA + η A e4
∇4 eA = ∇
∇A e3 = χ♯B
A B
e + ζA e3 , ∇A e4 = χ♯B
A eB − ζA e4
∇3 e4 = 2η ♯A eA − ω̂e4 , ∇4 e3 = 2η ♯A eA − ω̂e3
∇3 e3 = ω̂e3 , ∇4 e4 = ω̂e4 (1.175)
Here:
η ♯A = (g/−1 )AB ηB , η ♯A = g/−1 )AB η B , χ♯B /−1 )BC χAC , χ♯B

A = (g A
= (g/−1 )BC χAC
59
Also, if X is a S vectorfield we denote:
∇
/3 X = Π∇L̂ X, ∇
/4 X = Π∇L̂ Y (1.176)
where Π is the projection operator to Su,u , defined by 1.110. If W is any q-

contravariant S tensorfield, ∇ /3 W and ∇ /4 W are defined in a similar manner,
extending the projection operator Π at each x ∈ M \ Γ0 to the tensor product
⊗q Tx Su,u . On the other hand if ξ is any p-covariant S tensorfield, we define ∇ /3 ξ
and ∇/4 ξ to be simply the restrictions to T Su,u of ∇L̂ ξ and ∇L̂ ξ respectively.
Finally, if θ is an arbitrary type Tpq S tensorfield, we define ∇ /3 θ and ∇/4 θ to be
the projections to Su,u of the restrictions to T Su,u of ∇L̂ θ and ∇L̂ θ respectively.
The components of the spacetime volume form ǫ are given by:
ǫAB34 = 2ǫ/AB (1.177)
From 1.53, 1.58, 1.75, 1.77, 1.104, 1.131, 1.147, 1.150, 1.159, 1.161, the
components of the curvature tensor of a metric which is a solution of the vacuum
Einstein equations 1.124 are given by:
RA3B3 = αAB RA4B4 = αAB

RA334 = 2β A RA434 = 2βA
R3434 = 4ρ RAB34 = 2σǫ/AB
∗
RA3BC = β A/ǫ BC = g/AB β C − g/AC β B
RA4BC = − ∗ βA/ǫ BC = −g/AB βC + g/AC βB
RA4B4 = −ρg/AB + σǫ/AB
RABCD = −ρǫ/AB /ǫ CD = −ρ(g/AC g/BD − g/AD g/BC ) (1.178)
Moreover we have:
trα = Ric33 = 0 trα = Ric44 = 0 (1.179)
Conversely, if the curvature tensor of a metric g is of the form 1.178 with α, α

verifying 1.179, then g is a solution of the vacuum Einstein equations.
1.3 The Bianchi Equations

For a solution of the vacuum Einstein equations, the Bianchi identities
∇[α Rβγ]δǫ := ∇α Rβγδǫ + ∇β Rγαδǫ + ∇γ Rαβδǫ = 0 (1.180)
are equivalent to the contracted Bianchi identities:
∇α Rαβγδ = 0 ∇α := (g −1 )αβ ∇β (1.181)
This is a special case of a more general fact to be discussed in the first section of
Chapter 12. We shall presently write down the components of these equations.
60
Proposition 1.1 For a solution of the vacuum Einstein equations the com-
ponents of ∇R, the covariant derivative of the curvature tensor, are given by:
∇4 RA4B4 = ∇
/4 αAB − 2ω̂αAB
∇3 RA3B3 = ∇
/3 αAB − 2ω̂αAB
∇3 RA4B4 ˆ AB
/3 αAB + 2ω̂αAB − 4(η ⊗β)
= ∇
∇4 RA3B3 = ∇ ˆ AB
/4 αAB + 2ω̂αAB + 4(η ⊗β)
∇A RB4C4 = ∇ ˆ ABC + 2ζA αBC
/A αBC − (χ⊗β)
∇A RB3C3 = ∇ ˆ ABC − 2ζA αBC
/A αBC + (χ⊗β)
∇4 RA434 /4 βA − 2ω̂βA − 2η ♯B αAB
= 2∇
∇3 RA334 /3 β A − 2ω̂β A + 2η ♯B αAB
= 2∇
∇3 RA434 /3 βA + 2ω̂βA − 6(ηA ρ + ∗ ηA σ)
= 2∇
∇4 RA334 /4 β A + 2ω̂β A + 6(η A ρ − ∗ η A σ)
= 2∇
∇A RB434 /A βB + 2ζA βB − χ♯C
= 2∇ α
A BC
− 3(χAB ρ + /ǫ ♯A C χCB σ)
∇A RB334 /A β B − 2ζA β B + χ♯C
= 2∇ ǫ ♯A C χCB σ)
A αBC + 3(χAB ρ − /
∇4 R3434 = 4e4 ρ − 8η ♯A βA
∇3 R3434 = 4e3 ρ + 8η ♯A β A
∇A R3434 /A ρ + 4(χ♯B
= 4d ♯B
A β B − χ A βB )
∇4 RAB34 = 2ǫ/AB e4 σ + 4(η A βB − η B βA )
∇3 RAB34 = 2ǫ/AB e3 σ + 4(ηA β B − ηB β A )
∇A RBC34 = 2ǫ/BC /dA σ + 2(χAB β C − χAC β B ) + 2(χAB βC − χAC βB )
∇4 RA4BC = g/AC ∇
/4 βB − g/AB ∇/4 βC − ω̂(g/AC βB − g/AB βC ) + ηB αAC − ηC αAB
∇3 RA3BC = −g/AC ∇ /3 β C + ω̂(g/AC β B − g/AB β C ) + η B αAC − η C αAB
/3 β B + g/AB ∇
∇3 RA4BC = g/AC ∇
/3 βB − g/AB ∇
/3 βC + ω̂(g/AC βB − g/AB βC )
+ 3(−g/AC ηB + g/AB ηC )ρ − (ǫ/AC ηB − /ǫ AB ηC + /ǫ BC ηA )σ
∇4 RA3BC /4 β B + g/AB ∇
= −g/AC ∇ /4 β C + ω̂(−g/AC β B + g/AB β C )
+ 3(−g/AC η B + g/AB η C )ρ + (ǫ/AC η B − /ǫ AB η C + /ǫ BC η A )σ
∇A RB4CD = g/BD ∇
/A βC − g/BC ∇ /A βD + ζA (g/BD βC − g/BC βD )
1
+ (χAC αBD − χAD αAC )
2
3
+ (χAD g/BC − χAC g/BD )ρ − (χAC /ǫ BD − χAD /ǫ BC + 2χAB /ǫ CD )σ
2
∇A RB3CD = −g/BD ∇/A β C + g/BC ∇ /A β D + ζA (g/BD β C − g/BC β D )
1
+ (χAC αBD − χAD αAC )
2
3
+ (χAD g/BC − χAC g/BD )ρ + (χAC /ǫ BD − χAD /ǫ BC + 2χAB /ǫ CD )σ
2
∇4 RA3B4 = −g/AB e4 ρ + /ǫ AB e4 σ + 2(η A βB − η B βA + η ♯C βC g/AB )
61
∇3 RA3B4 = −g/AB e3 ρ + /ǫ AB e3 σ + 2(ηA β B − ηB β A − η ♯C βC g/AB )
∇A RB3C4 = −g/BC /dA ρ + /ǫ BC /dA σ
+χAB βC − χAC βB + g/BC χ♯D β − χAB β C − χAC β B − g/BC χ♯D
A D A βD
∇4 RABCD = −(g/AC g/BD − g/AD g/BC )e4 ρ
+η A (βC g/DB − βD g/CB ) − η B (βC g/DA − βD g/CA )
+η C (βA g/BD − βB g/AD ) − η D (βA g/BC − βB g/AC )
∇3 RABCD = −(g/AC g/BD − g/AD g/BC )e3 ρ
−ηA (β C g/DB − β D g/CB ) + ηB (β C g/DA − β D g/CA )
−ηC (β A g/BD − β B g/AD ) + ηD (β A g/BC − β B g/AC )
∇A RBCDE = (g/BD g/CE − g/BE g/CD )d
/A ρ
1 1
+ χAB (−β D g/EC + β E g/DC ) − χAC (−β D g/EB + β E g/DB )
2 2
1 1
+ χAD (−β B g/CE + β C g/BE ) − χAE (−β B g/CD + β C g/BD )
2 2
1 1
+ χAB (βD g/EC − βE g/DC ) − χAC (βD g/EB − βE g/DB )
2 2
1 1
+ χAD (βB g/CE − βC g/BE ) − χAE (βB g/CD − βC g/BD )
2 2
ˆ ′ the symmetric trace-free 2-covariant
Here, for S 1-forms ξ, ξ ′ we denote by ξ ⊗ξ
S tensorfield:
ˆ ′ = ξ ⊗ ξ ′ + ξ ′ ⊗ ξ − (ξ, ξ ′ )g/
ξ ⊗ξ (1.182)
In components,
ˆ ′ )AB = ξA ξB
(ξ ⊗ξ ′ ′
+ ξB ξA − (ξ, ξ ′ )g/AB
Also, for a symmetric 2-covariant S tensorfield θ and a S 1-form ξ we denote
ˆ the 3-covariant S tensorfield, symmetric and trace-free in the last two
by θ⊗ξ
entries:
ˆ =θ⊗ξ+θ⊗
θ⊗ξ ˜ ξ − θ♯ · ξg/ (1.183)
In components,
♯D
ˆ ABC = θAB ξC + θAC ξB − θA
(θ⊗ξ) ξD g/BC
Proof: The proof is straightforward using tables 1.178, 1.175 and the definitions
of the operators ∇
/, ∇
/3 , ∇
/4 as applied to p-covariant S tensorfields, which imply
in particular that:
∇
/g/ = 0, ∇/ǫ/ = 0
and that:
∇
/3 g/ = 0, ∇
/3/ǫ = 0; ∇
/4 g/ = 0, ∇
/4/ǫ = 0 (1.184)
The fact that ∇
/3/ǫ = 0 is seen most easily by choosing the frame field (eA :
A = 1, 2) to be othonormal. Then g/AB = δAB , /ǫ AB is the antisymmetric 2-
dimensional symbol, and:
0 = e3 (g(eA , eB )) = g(∇3 eA , eB ) + g(eA , ∇3 eB ) = g(∇
/3 eA , eB ) + g(eA , ∇
/ 3 eB )
62
hence, setting:
/ 3 eA = M B
∇ A eB
we have:
MB A
A + MB = 0
that is, M B
A is an antisymmetric matrix. Thus:
MB
A = M/
ǫ AB
and we obtain:
(∇
/3/ǫ )AB = (∇3/ǫ )(eA , eB ) = e3 (ǫ/AB ) − /ǫ (∇
/3 eA , eB ) − /ǫ (eA , ∇
/ 3 eB )
= −ǫ/(∇ /3 eB ) = −M (ǫ/AC /ǫ CB + /ǫ BC /ǫ AC ) = 0
/3 eA , eB ) − /ǫ (eA , ∇
The fact that ∇ /4/ǫ = 0 is shown in a similar manner, while the fact that ∇
/3 g/ =
∇
/4 g/ = 0 follows trivially.
Using the expression 1.126 for g −1 we can write down the contracted Bianchi
identities 1.181 in the form:
1 1
− ∇3 R4βγδ − ∇4 R3βγδ + (g/−1 )AB ∇A RBβγδ = 0 (1.185)
2 2
In applying Proposition 1.1, we decompose χ and χ into their trace-free parts,
respectively χ̂ and χ̂, and their traces:
1 1
χ = χ̂ + g/trχ, χ = χ̂ + g/trχ (1.186)
2 2
We then make use of the following basic fact: for any pair of trace-free symmetric
2-covariant S tensorfields θ, θ′ we have:
θ × θ′ + θ′ × θ = (θ, θ′ )g/ (1.187)
In particular, for any trace-free symmetric 2-covariant S tensorfield θ we have:

1 2
θ×θ = |θ| g/ (1.188)
2
We also make use of the fact that if θ is a trace-free 2-covariant S tensorfield
then ∗ θ defined by:
∗
θAB = /ǫ ♯A C θCB (1.189)
is also a trace-free symmetric 2-covariant S tensorfield. We then conclude that
the components of the contracted Bianchi identities are the following system of
equations:
/3 α + 12 trχα + 2ω̂α − ∇
∇ ˆ − (4η + ζ)⊗β
/⊗β ˆ + 3χ̂ρ + 3 ∗ χ̂σ = 0
/4 α + 21 trχα + 2ω̂α + ∇
∇ ˆ + (4η − ζ)⊗β
/⊗β ˆ + 3χ̂ρ − 3 ∗ χ̂σ = 0
63
∇ / α − (η ♯ + 2ζ ♯ ) · α = 0
/4 β + 2trχβ − ω̂β − div
∇ / α + (η ♯ − 2ζ ♯ ) · α = 0
/3 β + 2trχβ − ω̂β + div
∗
∇
/3 β + trχβ + ω̂β − /dρ − /dσ − 3ηρ − 3 ∗ ησ − 2χ̂♯ · β = 0
/4 β + trχβ + ω̂β + /dρ −
∇ ∗
/dσ + 3ηρ − 3 ∗ ησ − 2χ̂♯ · β = 0
/ β − (2η + ζ, β) + 21 (χ̂, α) = 0
e4 ρ + 32 trχρ − div
e3 ρ + 23 trχρ + div
/ β + (2η − ζ, β) + 12 (χ̂, α) = 0
/ β + (2η + ζ) ∧ β − 21 χ̂ ∧ α = 0
e4 σ + 23 trχσ + curl
e3 σ + 23 trχσ + curl
/ β + (2η − ζ) ∧ β + 21 χ̂ ∧ α = 0 (1.190)
Here if ξ is an S 1-form we denote by ∇ˆ the symmetric trace-free 2-covariant

/⊗ξ
S tensorfield:
∇ ˆ =∇
/⊗ξ ˜ − g/div
/ξ + ∇
/ξ / ξ
We note that if ξ is a S 1-form we have:
(Dξ)(eA ) = (LL ξ)(eA ) = L(ξ(eA )) − ξ([L, eA ]) = L(ξ(eA )) − ξ(∇L eA ) + ξ(∇A L)

/4 ξ)(eA ) + χ♯B
= (∇L ξ)(eA ) + ξ(∇A L) = Ω((∇ A ξ(eB ))
or:
/4 ξ + χ♯ · ξ)
Dξ = Ω(∇ (1.191)
Similarly:
/3 ξ + χ♯ · ξ)
Dξ = Ω(∇ (1.192)
Also, if θ is a symmetric 2-covariant S tensorfield we have:
(Dθ)(eA , eB ) = (LL θ)(eA , eB ) = L(θ(eA , eB )) − θ([L, eA ], eB ) − θ(eA , [L, eB ])

= L(θ(eA , eB )) − θ(∇L eA , eB ) − θ(eA , ∇L eB ) + θ(∇A L, eB ) + θ(eA , ∇B L)
= (∇L θ)(eA , eB ) + θ(∇A L, eB ) + θ(eA , ∇B L)
/4 θ)(eA , eB ) + χ♯C
= Ω((∇ ♯C
A θ(eC , eB ) + χB θ(eA , eC ))
or:
Dθ = Ω(∇
/4 θ + χ × θ + θ × χ) (1.193)
Similarly:
Dθ = Ω(∇
/3 θ + χ × θ + θ × χ) (1.194)
In the case that θ is trace-free, decomposing χ and χ as in 1.186 and using
1.187, we obtain, denoting by D̂θ and D̂θ the trace-free parts of Dθ and Dθ
respectively, simply:
D̂θ = Ω(∇
/4 θ + trχθ), D̂θ = Ω(∇
/3 θ + trχθ) (1.195)
Taking into account 1.191, 1.192 and 1.195, we arrive at the form of the
Bianchi identities given by the following proposition.
64
Proposition 1.2 For a solution of the vaccuum Einstein equations the Bianchi
identities take the form of the following system of equations:
D̂α − 21 Ωtrχα + 2ωα + Ω{−∇ˆ − (4η + ζ)⊗β

/⊗β ˆ + 3χ̂ρ + 3 ∗ χ̂σ} = 0
D̂α − 21 Ωtrχα + 2ωα + Ω{∇ˆ + (4η − ζ)⊗β
/⊗β ˆ + 3χ̂ρ − 3 ∗ χ̂σ} = 0
Dβ + 23 Ωtrχβ − Ωχ̂♯ · β − ωβ − Ω{div
/ α + (η ♯ + 2ζ ♯ ) · α} = 0
Dβ + 23 Ωtrχβ − Ωχ̂♯ · β − ωβ + Ω{div
/ α + (η ♯ − 2ζ ♯ ) · α} = 0
Dβ + 21 Ωtrχβ − Ωχ̂ · β + ωβ − Ω{d
/ρ + ∗
/dσ + 3ηρ + 3 ∗ ησ + 2χ̂♯ · β} = 0
Dβ + 21 Ωtrχβ − Ωχ̂♯ · β + ωβ + Ω{d/ρ − ∗/dσ + 3ηρ − 3 ∗ ησ − 2χ̂♯ · β} = 0
Dρ + 32 Ωtrχρ − Ω div/ β + (2η + ζ, β) − 12 (χ̂, α) = 0

Dρ + 23 Ωtrχρ + Ω div/ β + (2η − ζ, β) + 12 (χ̂, α) = 0

Dσ + 23 Ωtrχσ + Ω curl/ β + (2η + ζ, ∗ β) − 21 χ̂ ∧ α = 0

Dσ + 23 Ωtrχσ + Ω curl/ β + (2η − ζ, ∗ β) + 21 χ̂ ∧ α = 0

1.4 Canonical coordinate systems

Given local coordinates (ϑA : A = 1, 2) on a domain U ⊂ S0,u0 , we can define
coordinates (u, u; ϑA : A = 1, 2) in the domain
[
MU = Φu−u0 (Φu (U )) ⊂ M \ Γ0 (1.196)
(u,u)∈D\A0
as follows. Given a point p ∈ MU , then p ∈ Su,u for a unique (u, u) ∈ D \

A0 . The coordinates (u, u) are then assigned to p accordingly. Moreover, p =
Φu−u0 (Φu (q)) for a unique q ∈ U ⊂ S0,u0 . Let (ϑ1 , ϑ2 ) be the coordinates of
q. The coordinates (ϑ1 , ϑ2 ) are then assigned to p accordingly. The coordinates
(u, u; ϑA : A = 1, 2) on MU ⊂ M \ Γ0 are thus canonically associated to the
coordinates (ϑA : A = 1, 2) on U ⊂ S0,u0 . Also, if {U1 , U2 } is a covering of S0,u0 ,
a sphere in Euclidean 3-dimensional space, then {MU1 , MU2 } is a covering of
M \ Γ0 . In these “canonical” coordinates the vectorfields L and L are given by:
∂ ∂ ∂
L= , L= + bA A (1.197)
∂u ∂u ∂ϑ
where:
bA (u, u0 , ϑ) = 0 (ϑ := (ϑ1 , ϑ2 )) (1.198)
According to the commutation relation 1.91 we have:
∂bA
= 4Ω2 ζ ♯A (1.199)
∂u
where:
ζ ♯A = (g/−1 )AB ζB (1.200)
65
and:
∂ ∂ ∂
g/AB = g/ A
, B , ζA = ζ (1.201)
∂ϑ ∂ϑ ∂ϑA
are the components of g/, ζ in the coordinate frame field (∂/∂ϑA : A = 1, 2) for
the Su,u . Taking also into account 1.15 and the fact that:

∂ ∂
g L, A = g L, A = 0 (1.202)
∂ϑ ∂ϑ
the vectorfield ∂/∂ϑA being tangential to the surfaces Su,u , we conclude that
the spacetime metric g is given on MU by:
g = −2Ω2 (du ⊗ du + du ⊗ du) + g/AB (dϑA − bA du) ⊗ (dϑB − bB du) (1.203)
The volume form ǫ = dµg of g is then given by:

p
ǫ = 2Ω2 detg/du ∧ du ∧ dϑ1 ∧ dϑ2 (1.204)
or:
dµg = 2Ω2 dµ/g ∧ du ∧ du (1.205)
where: p
dµ/g = /ǫ = detg/dϑ1 ∧ ϑ2 (1.206)
is the area form of Su,u .
In the Minkowskian region M0 we may introduce Cartesian coordinates
(x0 , x1 , x2 , x3 ) such that the x0 axis coincides with Γ0 and x0 = 0 at the
point e. The Cartesian coordinates are then unique up to an SO(3) rotation in
(x1 , x2 , x3 ). The surface S0,u on the boundary C 0 of M0 , is the sphere of radius
|u| in the hyperplane x0 = u:
p
|x| = |u|, where |x| = (x1 )2 + (x2 )2 + (x3 )2 (1.207)
The coordinates (x1 , x2 , x3 ) define stereographic coordinates (ϑ1 , ϑ2 ) on S0,u0 .

We have two stereographic charts, the north polar chart, induced by the projec-
tion from the south pole q2 = (0, 0, −|u0|) to the plane x3 = |u0 |, and the south
polar chart, induced by the projection from the north pole q1 = (0, 0, |u0 |) to
the plane x3 = −|u0 |. The domain of the north polar chart is U1 = S0,u0 \ q2
and the chart is the mapping of U1 onto ℜ2 by (x1 , x2 , x3 ) ∈ U1 7→ (ϑ1 , ϑ2 ) ∈ ℜ2
by:
2x1 2x2
ϑ1 = , ϑ2
= (1.208)
|x| + x3 |x| + x3
the north pole p1 being mapped to the origin in ℜ2 . The domain of the south
polar chart is U2 = S0,u0 \ q1 and the chart is the mapping of U2 onto ℜ2 by
(x1 , x2 , x3 ) ∈ U2 7→ (ϑ1 , ϑ2 ) ∈ ℜ2 by:
2x1 2x2
ϑ1 = , ϑ2 = (1.209)
|x| − x3 |x| − x3
66
2
the south pole p2 beingT mapped to the origin in ℜ2 . The image by either chart
of the intersection U1 U2 = S0,u0 \ {p1 , p2 } is ℜ \ (0, 0). The transformation
from north polar coordinates to south polar coordinates is the analytic mapping
f of ℜ2 onto itself given by:
4ϑ p
ϑ 7→ f (ϑ) = 2
, where ϑ = (ϑ1 , ϑ2 ), |ϑ| = (ϑ1 )2 + (ϑ2 )2 (1.210)
|ϑ|
The equator x3 = 0 of S0,u0 is mapped by either chart onto the circle of radius
2 in ℜ2 . Note that:
f = f −1 (1.211)
so the inverse transformation, from south polar coordinates to north polar co-
ordinates is identical in form. Let σ ∈ (0, 1). Then, restricting the domains U1
and U2 to:
U1,σ = {(x1 , x2 , x3 ) ∈ S0,u0 : x3 > −σ|u0 |},

U2,σ = {(x1 , x2 , x3 ) ∈ S0,u0 : x3 < σ|u0 |} (1.212)
{U1,σ , U2,σ } is a covering of S0,u0 . The image by the north polar chart of U1,σ
and the image by the south polar chart of U2,σ is D2ρ , the open disk of radius
2ρ in ℜ2 , where: r
1+σ
ρ= >1 (1.213)
1−σ
T
The image by either chart of U1,σ U2,σ is the open annulus:
2
Aρ = {ϑ ∈ ℜ2 : < |ϑ| < 2ρ} (1.214)
ρ
The mapping f maps Aρ onto itself and has bounded derivatives of all orders
on Aρ .
The stereographic coordinates on U1 , U2 induce canonical coordinates on
MU1 , MU2 which are unique up to the original SO(3) rotation. The induced
metric g/|S0,u on S0,u0 is given by:
0
◦
g/|S0,u = |u0 |2 g/ (1.215)
0
◦
where g/ is the standard metric on S 2 . In the canonical coordinates induced by
the stereographic coordinates 1.215 reads:
◦
g/AB (0, u0 , ϑ) = |u0 |2 g/AB (ϑ) (1.216)
◦
where g/AB (ϑ) is given in both charts by:
◦ δAB
g/AB (ϑ) = 2 (1.217)
1 + 14 |ϑ|2
67
Chapter 2
The Characteristic Initial

Data
2.1 The characteristic initial data

In the following we shall show how characteristic initial data is set up on Co . As
mentioned in Chapter 1, the restrictions of the initial data to Cor0 are trivial, that
is, they coincide with the data corresponding to a truncated cone in Minkowski
spacetime. We thus confine ourselves to
Co \ (Cor0 \ Sor0 ) (2.1)
where the data is non-trivial. From now on we denote by Cu0 , the non-trivial
part 2.1 of Co , rather than the whole of Co . Recall that C 0 is the outer boundary
of the Minkowskian region M0 . On C 0 we have the trivial initial data induced by
the Minkowski metric on M0 . The two null hypersurfaces Cu0 and C 0 intersect
at Sor0 = S0,u0 : \
C 0 Cu0 = S0,u0 (2.2)
a round sphere
T of radius r0 = |u0 | in Euclidean 3-dimensional space, the bound-
ary of Hu0 M0 , a ball of radius r0 in Euclidean 3-dimensional space.
The characteristic initial data on Cu0 is to be the specification of the con-
formal geometry of Cu0 , in the manner to be presently described. Let us denote
◦
again by g/ the standard metric on S 2 , given in stereographic coordinates by
1.217. The induced metric g/|Su,u : u ∈ [0, δ] may be expressed the form:
0
g/|Su,u = ( φ|Su,u )2 ĝ/ (2.3)

0 0 Su,u0
where the metric ĝ/ is subject to the requirement that Φ∗u ĝ/ , a metric
Su,u0 Su,u0
68
on S0,u0 , has the same area form as g/|S0,u :
0
dµ Φ∗ /ĝ = dµ /g|S (2.4)

u 0,u0
Su,u0
The positive function φ|Su,u is then defined by:

0
dµ /g|S = ( φ|Su,u )2 dµ /ĝ | (2.5)

u,u0 0 Su,u0
In terms of canonical coordinates on Cu0 induced by stereographic coordinates

on S0,u0 , the requirement 2.4 reads, in view of 1.215 - 1.217,
q
detĝ/(u, u0 , ϑ) = |u0 |2 w2 (ϑ) (2.6)
where
1
w(ϑ) = (2.7)
1 + 41 |ϑ|2
in both stereographic charts. Thus, ĝ/ is given by:
ĝ/AB (u, u0 , ϑ) = |u0 |2 w2 (ϑ)mAB (u, u0 , ϑ) (2.8)
where m satisfies:
detm = 1 (2.9)
and 2.5 becomes:
p q
detg/(u, u0 , ϑ) = φ2 (u, u0 , ϑ) detĝ/(u, u0 , ϑ) (2.10)
The characteristic initial data on Cu0 consists ofTthe specification of ĝ/ or

equivalently of m along Cu0 . Now, in changing in U1 U2 from the north polar
stereogeraphic chart to the south polar stereographic chart, or vice-versa, the
components of ĝ/ must transform as the components of a 2-covariant S tensofield.
Thus, with
ϑ′ = f (ϑ), ϑ = f (ϑ′ ) (2.11)
(see 1.210, 1.211), we must have:
′
ĝ/AB (u, u0 , ϑ) = TAC (ϑ)TBD (ϑ)ĝ/CD (u, u0 , ϑ′ ) (2.12)
where:
∂f C (ϑ) ϑC ϑA

4
TAC (ϑ) = A
= δCA − 2 (2.13)
∂ϑ |ϑ|2 |ϑ|2
In terms of matrix notation, with T̃ the transpose of T , the transformation rule
2.12 takes the form:
′
ĝ/(u, u0 , ϑ) = T̃ (ϑ)ĝ/ (u, u0 , ϑ′ )T (ϑ) (2.14)
69
◦
The components of g/, given by 1.217, similarly satisfy:
◦ ◦
g/ (ϑ) = T̃ (ϑ) g/′ (ϑ′ )T (ϑ) (2.15)
In view of the definition 2.7, we conclude that:
w2 (ϑ)
T̃ (ϑ)T (ϑ) = I (2.16)
w′2 (ϑ)
where:
w′ (ϑ) = w(ϑ′ ) : ϑ′ = f (ϑ) (2.17)
and I is the identity matrix. Note also from 2.13 that:
T̃ (ϑ) = T (ϑ) (2.18)
and:
w2
detT < 0, |detT | = (2.19)
w′2
(The last follows also from 2.16.) Substituting 2.8 in 2.14 and noting 2.17 we
conclude that the components of m must transform according to the rule:
w2 (ϑ)m(u, u0 , ϑ) = w′2 (ϑ)T̃ (ϑ)m′ (u, u0 , ϑ′ )T (ϑ) (2.20)
which in view of 2.19 we may write simply as:
m = |detT |−1 T̃ m′ T (2.21)
Taking determinants we obtain
detm′ = detm (2.22)
Thus the condition 2.9 on m implies the same condition on m′ and vice-versa.
The transformation rule 2.21 shows m to be a tensor density of weight -1.
Setting:
T
O= (2.23)
|detT |1/2
The matrix O is according to 2.13 given by:
C ϑC ϑA
OA (ϑ) = δCA − 2 (2.24)
|ϑ|2
By 2.16, 2.19 O is a symmetric orthogonal matrix of determinant −1:
ÕO = I, Õ = O, detO = −1 (2.25)
In terms of O the transformation rule 2.20 reads:
m(u, u0 , ϑ) = Õ(ϑ)m′ (u, u0 , ϑ′ )O(ϑ) (2.26)
70
which we may write simply as:
m = Õm′ O (2.27)
Now the matrix m at a given point on Cu0 is a 2-dimensional positive definite

symmetric unimodular matrix. Such a matrix has the form

Z +X Y
m= where Z 2 − X 2 − Y 2 = 1 and Z > 0 (2.28)
Y Z −X
Thus these matrices constitute the upper unit hyperboloid H1+ in ℜ3 . The
exponential mapping exp is an analytic mapping defined on the 4-dimensional
linear space of all 2-dimensional matrices A:
∞
X 1 n
exp A = I + A (2.29)
n=1
n!
The exponential map restricted to the 2-dimensional subspace Ŝ of all symmetric

trace-free 2-dimensional matrices is an analytic diffeomorphism of Ŝ onto H1+ .
A matrix A ∈ Ŝ is of the form:

a b
A= (2.30)
b −a
If O is an orthogonal transformation and A 7→ ÕAO, then:
exp(ÕAO) = Õ(exp A)O
thus if A is brought by a suitable orthogonal transformation to diagonal form:

λ 0
A=
0 −λ
then
eλ

0
exp A =
0 e−λ
In view of the above, we may express:
m = exp ψ (2.31)
where ψ belongs at each point to Ŝ.
Lemma 2.1 The transformation rule:
ψ = Õψ ′ O
for ψ, is equivalent to the transformation rule 2.27 for m.
71
Proof: Let ψ transform according to the above rule. Then by 2.25 we have:
ψ n = Õψ ′ (OÕψ ′ )n−1 O = Õψ ′n O
Therefore:
∞ ∞
!
X 1 n X 1 ′n
exp ψ = I + ψ = Õ I + ψ O = Õ(exp ψ ′ )O
n=1
n! n=1
n!
that is:
m = Õm′ O
as required. The converse then follows from the fact that exp is an analytic
diffeomorphism of Ŝ onto H1+ , so in particular it is one to one.
Let us denote the coordinates of the north polar chart by (ϑ1 , ϑ2 ) and those
of the south polar chart by (ϑ′1 , ϑ′2 ). The transformation rule of Lemma 2.1
takes the form:
ψ(u, u0 , ϑ) = Õ(ϑ)ψ ′ (u, u0 , ϑ′ )O(ϑ) : where ϑ′ = f (ϑ) (2.32)
Also, noting from 2.13 and 1.210 that:
O(ϑ′ ) = O(ϑ) : where ϑ′ = f (ϑ) (2.33)
the inverse transformation may be written in similar form (recall 1.211 and the
symmetry of O):
ψ ′ (u, u0 , ϑ′ ) = Õ(ϑ′ )ψ(u, u0 , ϑ)O(ϑ′ ) : where ϑ = f (ϑ′ ) (2.34)
We shall restrict the domain of the north polar chart to U 1,σ and that of the
south polar chart to U 2,σ (see 1.212) where σ ∈ (0, 1). The image by the north
polar chart of U 1,σ and the image by the south polar chart of U 2,σ is D2ρ , the
closed disk of radius 2ρ in ℜ2 , where ρ is given by 1.213:
r
1+σ
ρ= >1 (2.35)
1−σ
This closed disk is the range of the coordinates (ϑ1 , ϑ2 ) as well
T as the range of
the coordinates (ϑ′1 , ϑ′2 ). The image by either chart of U 1,σ U 2,σ is the closed
annulus:
2 2
Aρ = {ϑ ∈ ℜ2 : ≤ |ϑ| ≤ 2ρ} = {ϑ′ ∈ ℜ2 : ≤ |ϑ′ | ≤ 2ρ} (2.36)
ρ ρ
By 2.8, 2.31 and the transformation rule 2.32, or 2.34, the specification of a
smooth conformal metric ĝ/ on Cu0 is equivalent to the specification of a pair of
smooth mappings
ψu0 : [0, δ] × D2ρ → Ŝ, ψu′ 0 : [0, δ] × D2ρ → Ŝ (2.37)
72
such that with
ψ(u, u0 , ϑ) = ψu0 (u, ϑ), ψ ′ (u, u0 , ϑ′ ) = ψu′ 0 (u, ϑ′ ) (2.38)
the transformation rule 2.32, or 2.34, holds on [0, δ] × Aρ . Then with:
m(u, u0 , ϑ) = exp(ψ(u, u0 , ϑ)) : (u, ϑ) ∈ [0, δ] × D2ρ

m′ (u, u0 , ϑ′ ) = exp(ψ ′ (u, u0 , ϑ′ )) : (u, ϑ′ ) ∈ [0, δ] × D2ρ (2.39)
the conformal metric ĝ/ is given, in canonical coordinates, in the domain in Cu0
corresponding to U 1,σ by:
ĝ/AB (u, u0 , ϑ) = |u0 |2 w2 (ϑ)mAB (u, u0 , ϑ) (2.40)
and in the domain in Cu0 corresponding to U 2,σ by:

′
ĝ/AB (u, u0 , ϑ′ ) = |u0 |2 w2 (ϑ′ )m′AB (u, u0 , ϑ′ ) (2.41)
A crucial aspect of our setup is the following definition of the mappings 2.37.
We consider a pair of smooth mappings
ψ0 : [0, 1] × D2ρ → Ŝ, ψ0′ : [0, 1] × D2ρ → Ŝ (2.42)
such that the transformation rule 2.32, 2.34, that is:
ψ0 (s, ϑ) = Õ(ϑ)ψ0′ (s, ϑ′ )O(ϑ) : where ϑ′ = f (ϑ) (2.43)
or:
ψ0′ (s, ϑ′ ) = Õ(ϑ′ )ψ0 (s, ϑ)O(ϑ′ ) : where ϑ = f (ϑ′ ) (2.44)
holds for all (s, ϑ) ∈ [0, 1] × Aρ , (s, ϑ′ ) ∈ [0, 1] × Aρ . We require the mappings
ψ0 and ψ0′ to extend smoothly by zero to s < 0. In other words we require:
n n ′
∂ ψ0 ∂ ψ0
n
(0, ϑ) = 0 : ∀ϑ ∈ D 2ρ , (0, ϑ′ ) = 0 : ∀ϑ′ ∈ D2ρ (2.45)
∂s ∂sn
for all non-negative integers n. This is required so that the initial data on Cu0
are smooth extensions of the trivial data on Cor0 . We then set:
δ 1/2 u δ 1/2 ′ u ′
ψu0 (u, ϑ) = ψ0 ,ϑ , ψu′ 0 (u, ϑ′ ) = ψ ,ϑ (2.46)
|u0 | δ |u0 | 0 δ
The mappings ψu0 , ψu′ 0 so defined automatically satisfy the transformation rule
2.32, 2.34.
B ...B
In the following, for any multiplet of real numbers of the form ΞA11 ...Apq :
A1 , ..., Ap = 1, 2; B1 , ..., Bq = 1, 2 we denote by |Ξ| the “magnitude”:
s
B ...B
X
|Ξ| = (ΞA11 ...Apq )2 (2.47)
A1 ,...,Ap ;B1 ,...,Bq
73
Let kψ0 kC k ([0,1]×D2ρ ) and kψ0′ kC k ([0,1]×D2ρ ) be the C k norms on [0, 1] × D2ρ of
ψ0 and ψ0′ respectively:
∂m ∂n
kψ0 kC k ([0,1]×D2ρ ) = max sup ψ0 (s, ϑ)
m+n≤k
(s,ϑ)∈[0,1]×D2ρ ∂ϑm ∂sn
∂m ∂n ′
kψ0′ kC k ([0,1]×D2ρ ) = max sup ψ (s, ϑ) (2.48)
m+n≤k
(s,ϑ)∈[0,1]×D2ρ ∂ϑm ∂sn 0
Here, we denote by ∂ m /∂ϑm the multiplet of differential operators:
∂m
∂ϑA1 ...∂ϑAm
We then set:
Mk = max{kψ0 kC k ([0,1]×D2ρ ) , kψ0′ kC k ([0,1]×D2ρ ) } (2.49)
Let km − IkC k ([0,δ]×D2ρ ) and km′ − IkC k ([0,δ]×D2ρ ) be the following weighted
δ δ
C k norms on [0, δ] × D2ρ of m − I and m′ − I respectively:
∂m ∂n
km − IkC k ([0,δ]×D2ρ ) = max sup δn (m − I)(u, u0 , ϑ)
δ m+n≤k
(u,ϑ)∈[0,δ]×D2ρ
∂ϑm ∂un
∂m ∂n
km′ − IkC k ([0,δ]×D2ρ ) = max sup δn (m − I)(u, u0 , ϑ)
δ m+n≤k
(u,ϑ)∈[0,δ]×D2ρ
∂ϑm ∂un
(2.50)
The definitions 2.45, 2.37 and 2.38 imply that m and m′ are smooth functions
on [0, δ] × D2ρ and for each non-negative integer k we have:
max{km − IkC k ([0,δ]×D2ρ ) , km′ − IkC k ([0,δ]×D2ρ ) } ≤ δ 1/2 |u0 |−1 Fk (Mk ) (2.51)
δ δ
where Fk is a non-negative non-decreasing continuous function on the non-

negative real line.
Let ξ be a C k Tpq type S tensorfield on Cu0 , represented in the north polar
B ...B
stereographic chart by the C k component functions ξA11...Apq (u, u0 , ϑ) and in the
′B ...B
south polar stereographic chart by the C k component functions ξA11...Apq (u, u0 , ϑ).
We define the weighted C k norms of the component functions by:
∂m ∂n

kξkC k ([0,δ]×D2ρ ) = max sup δn ξ(u, u 0 , ϑ)
δ m+n≤k
(u,ϑ)∈[0,δ]×D2ρ
∂ϑm ∂un
∂m ∂n ′

kξ ′ kC k ([0,δ]×D2ρ ) = max sup δn ξ (u, u 0 , ϑ)
δ m+n≤k
(u,ϑ)∈[0,δ]×D2ρ ∂ϑm ∂un
(2.52)
74
We then define the weighted C k norm of ξ on Cu0 by:
kξkC k (Cu0 ) = max{kξkC k ([0,δ]×D2ρ ) , kξ ′ kC k ([0,δ]×D2ρ ) } (2.53)

δ δ δ
Moreover, we say that a smooth Tpq type S tensorfield ξ on Cu0 is Ml (δ r |u0 |s ),

for real numbers r, s and non-negative integer l,
ξ = Ml (δ r |u0 |s ) (2.54)
if for every non-negative integer k we have:

kξkC k (Cu0 ) ≤ δ r |u0 |s Fk (Mk+l ) (2.55)
δ

negative real line.
It follows from 2.51 and 2.8, in view of 2.7 and 1.217, that ĝ/ is a smooth
metric on Cu0 and for each non-negative integer k we have:
◦
k|u0 |−2 ĝ/− g/ kC k (Cu0 ) ≤ δ 1/2 |u0 |−1 Fk (Mk ) (2.56)
δ

negative real line. Thus, in accordance with the above definition we have:
◦
|u0 |−2 ĝ/− g/= M0 (δ 1/2 |u0 |−1 ) (2.57)
Let K̂ be the Gauss curvature of ĝ/. Then K̂ is a smooth function on Cu0 and
from 2.56 we deduce that for every non-negative integer k we have:
kK̂ − |u0 |−2 kC k (Cu0 ) ≤ δ 1/2 |u0 |−3 Fk (Mk+2 ) (2.58)

δ
where Fk is a (different) non-negative non-decreasing continuous function on

the non-negative real line. Thus, in accordance with the above definition we
have:
K̂ − |u0 |−2 = M2 (δ 1/2 |u0 |−3 ) (2.59)
Consider now equation 2.3. In canonical coordinates we have:
g/AB (u, u0 , ϑ) = φ2 (u, u0 , ϑ)ĝ/AB (u, u0 , ϑ) (2.60)
Differentiating with respect to u we obtain:
∂g/AB ∂ ĝ/ ∂f
= φ2 AB + 2φ ĝ/AB (2.61)
∂u ∂u ∂u
Now, along Cu0 we have:
Ω = 1 : along Cu0 (2.62)
u coinciding up to the additive constant r0 with the affine parameter s along
the generators of Cu0 , hence
1 = Lu = Ω2 L′ u = Ω2 L′ s = Ω2 : along Cu0
75
Thus, the first of equations 1.41 becomes along Cu0 :
Dg/ = 2χ (2.63)
or, in canonical coordinates,

∂
g/AB = 2χAB = 2χ̂AB + trχg/AB (2.64)
∂u
in terms of the decomposition 1.186. Now, from 2.8 we have:
∂ ĝ/ ∂mAB
(u, u0 , ϑ) = |u0 |2 w2 (ϑ) (u, u0 , ϑ) (2.65)
∂u ∂u
while by 2.9:
detĝ/(u, u0 , ϑ) = |u0 |4 w4 (ϑ) (2.66)
hence:
−1 AB ∂ ĝ
/AB ∂
(ĝ/ ) = log detĝ/ = 0 (2.67)
∂u ∂u
Comparing then 2.64 with 2.61 we conclude that:
1 2 ∂ ĝ/ 2 ∂φ
χ̂AB = φ trχ = (2.68)
2 ∂u φ ∂u
Denoting pointwise magnitudes of S tensorfields relative to g/ by | |/g to avoid

confusion with the notation 2.47, let us define on Cu0 the function:
1 2 1
e= |χ̂|/g = (g/−1 )AC (g/−1 )BD χ̂AB χ̂CD (2.69)
2 2
By 2.60 and the first of 2.68 we obtain:
1 −1 AC −1 BD ∂ ĝ/AB ∂ ĝ/CD
e= (ĝ/ ) (ĝ/ ) (2.70)
8 ∂u ∂u
This is a smooth non-negative function on Cu0 which depends only on ĝ/. From
2.56 we deduce that for every non-negative integer k we have:
kekC k (Cu0 ) ≤ δ −1 |u0 |−2 Fk (Mk+1 ) (2.71)

δ
where Fk is a (different) non-negative non-decreasing continuous function on

the non-negative real line. Thus, in accordance with the notation 2.54 we have:
e = M1 (δ −1 |u0 |−2 ) (2.72)
Equation 2.63 implies:
Dg/−1 = −2χ♯♯ : on Cu0 (2.73)
76
where χ♯♯ is the symmetric 2-contravariant S tensorfield with components:
χ♯♯AB = χCD (g/−1 )AC (g/−1 )BD (2.74)
It follows that for any 2-covariant S tensorfield θ on Cu0 we have:
Dtrθ = trDθ − 2(χ, θ) : on Cu0 (2.75)
Thus in view of 1.123, the second of 1.179, and 2.62, the trace of the second
variation equation 1.57 reads, along Cu0 :
1
Dtrχ = −|χ|/2g = − (trχ)2 − |χ̂|/2g : along Cu0 (2.76)
2
Substituting from 2.69 and the second of 2.68 this equation becomes the follow-
ing linear second order ordinary differential equation for the function φ along
the generators of Cu0 :
∂2φ
+ eφ = 0 (2.77)
∂u2
The initial conditions on S0,u0 for this equation are the following. First, by 2.5
and 2.4 at u = 0 we have:
φ|S0,u = 1 (2.78)
0
Then by the second of 2.68 at u = 0:

∂φ 1 1
= trχ|S0,u = (2.79)
∂u S0,u0 2 0 |u0 |
the last being the mean curvature of a round sphere of radius r0 = |u0 | in
Euclidean 3-dimensional space with respect to the outer normal.
The fact that e is non-negative, together with the fact that φ must be pos-
itive, implies that along each generator of Cu0 φ is a concave function of u.
Equation 2.77 together with the initial conditions 2.78, 2.79 has a unique smooth
solution on Cu0 . However, this solution may not be everywhere positive on Cu0 .
Consider a generator of Cu0 , corresponding to some ϑ ∈ S 2 . Then either φ is
positive everywhere on this generator, that is φ(u, u0 , ϑ) > 0 for all u ∈ [0, δ],
or there is a u∗ (ϑ) ∈ [0, δ] such that φ(u, u0 , ϑ) > 0 for all u ∈ [0, u∗ (ϑ)), but
φ(u∗ (ϑ), u0 , ϑ) = 0. (The point with coordinates (u∗ (ϑ), u0 , ϑ) may be a point
conjugate to o along the given generator, or a truly singular point.) If there
exists a generator, that is a ϑ ∈ S 2 , where the second alternative holds, then
the initial data are singular.
Lemma 2.2 The initial data are regular if δ|u0 |−2 is suitably small
depending on M1 .
Proof: For, φ(u, u0 , ϑ) is concave on [0, δ] if the first alternative holds, on
[0, u∗ (ϑ)] if the second alternative holds, hence, in view of the initial conditions
2.78, 2.79 we have:
u
φ(u, u0 , ϑ) ≤ 1 + (2.80)
|u0 |
77
for all u ∈ [0, δ] if the first alternative holds, for all u ∈ [0, u∗ (ϑ)] if the second
alternative holds. Integrating 2.77 once and using the initial condition 2.79 we
obtain: Z u
∂φ 1
(u, u0 , ϑ) = − φ(u′ , u0 , ϑ)e(u′ , u0 , ϑ)du′ (2.81)
∂u |u0 | 0
Substituting 2.80 and the fact that by 2.71 with k = 0:
sup e ≤ δ −1 |u0 |−2 F0 (M1 ) (2.82)

Cu0
we obtain:
∂φ 1 δ F0 (M1 )
(u, u0 , ϑ) ≥ − 1+ (2.83)
∂u |u0 | |u0 | |u0 |2
alternative holds. Integrating again and using the initial condition 2.78 we then
obtain:
u δ F0 (M1 )
φ(u, u0 , ϑ) ≥ 1 + 1− 1+ (2.84)
|u0 | |u0 | |u0 |
alternative holds. In particular, if the second alternative holds, then setting
u = u∗ (ϑ) we must have:
u∗

δ F0 (M1 )
0≥1+ 1− 1+ (2.85)
|u0 | |u0 | |u0 |
However, since |u0 | > 1, δ ≤ 1, u∗ ≤ δ, this is impossible if:
2δ|u0 |−2 F0 (M1 ) < 1 (2.86)
Therefore, under this smallness condition on δ, the second alternative is

ruled out and the initial data are regular.
In the following we shall in fact impose the stronger condition:
4δF0 (M1 ) ≤ 1 (2.87)
Since then 2.80 and 2.84 hold for all u ∈ [0, δ], we have:
u 1
sup φ − 1 − ≤ 2δ|u0 |−2 F0 (M1 ) ≤ (2.88)
Cu0 |u0 | 2
Substituting 2.88 and 2.82 in 2.77 then yields:

∂φ 1
sup − ≤ |u0 |−2 F0′ (M1 ) (2.89)
Cu0 ∂u |u0 |
where F0′ is a non-negative non-decreasing continuous function on the non-

negative real line.
78
We proceed to estimate ∂ m φ/∂ϑm . Here it is convenient to use multi-index
notation. If a = (a1 , a2 ) is a pair of non-negative integers, we denote:
a
∂ |a|

∂
= (2.90)
∂ϑ ∂(ϑ1 )a1 ∂(ϑ2 )a2
a differential operator of order |a|, where |a| is the “length”’ of the multi-index:
|a| = a1 + a2 (2.91)
If a and b are two multi-indices, we say that b ≤ a if b1 ≤ a1 and b2 ≤ a2 . Then

a − b is also a multi-index, of length |a| − |b|. We also denote:

a a!
a! = a1 !a2 ! = (2.92)
b b!(a − b)!
If f, g is a pair of smooth functions on [0, δ] × D2ρ , we have the Leibniz rule:

a X a ∂ b ∂ a−b
∂
= f g (2.93)
∂ϑ b ∂ϑ ∂ϑ
b≤a
We work in each of the two stereographic charts separately. Applying

(∂/∂ϑ)a to 2.77 we obtain:
a a
∂2 ∂ ∂
φ + e φ = fa (2.94)
∂u2 ∂ϑ ∂ϑ
where: b a−b
X ∂ ∂
fa = − e φ (2.95)
∂ϑ ∂ϑ
b≤a,|b|6=0
Applying (∂/∂ϑ)a for |a| = 6 0 to the initial conditions 2.78, 2.79 we obtain
simply:
a a
∂ ∂ ∂φ
φ(0, u0 , ϑ) = (0, u0 , ϑ) = 0 : for |a| =
6 0 (2.96)
∂ϑ ∂ϑ ∂u
Lemma 2.3 If condition 2.87 is satisfied, then for each non-zero multi-index
a we have:
a
∂
sup φ(u, u0 , ϑ ≤ δ|u0 |−2 F|a| (M|a|+1 )
(u,ϑ)∈[0,δ]×D2ρ
∂ϑ
and: a
∂ ∂φ
sup (u, u0 , ϑ) ≤ |u0 |−2 F|a|
′
(M|a|+1 )
(u,ϑ)∈[0,δ]×D2ρ ∂ϑ ∂u
79
where Fk , Fk′ are non-negative non-decreasing continuous functions on the non-
negative real line.
Proof: The proof is by induction on |a|. Let the lemma hold for all |a| =
1, ..., k − 1 (no assumption if k = 1). Then given any a with |a| = k, by
2.71, 2.88 and the inductive hypothesis, there is a non-negative non-decreasing
continous function G|a| such that:
sup |fa | ≤ δ −1 |u0 |−2 G|a| (M|a|+1 ) (2.97)

(u,ϑ)∈[0,δ]×D2ρ
We integrate 2.94 twice, using the initial conditions 2.96, to obtain:

a Z u Z u′ a
∂ ∂
φ(u, u0 , ϑ) = fa − e φ (u′′ , u0 , ϑ)du′′ du′ (2.98)
∂ϑ 0 0 ∂ϑ
It follows that:
a u u′
∂
Z Z
φ(u, u0 , ϑ) ≤ |fa (u′′ , u0 , ϑ)|du′′ du′
∂ϑ 0 0
u u′ a
∂
Z Z
+ e φ (u′′ , u0 , ϑ)du′′ du′ (2.99)
0 0 ∂ϑ
By 2.97 the first double integral on the right is bounded by:
1
δ|u0 |−2 G|a| (M|a|+1 ) (2.100)
2
Setting a
∂
xa (u) = sup φ(u′ , u0 , ϑ) (2.101)
u′ ∈[0,u] ∂ϑ
then by 2.82 the second double integral on the right in 2.99 is bounded by:
1
δ|u0 |−2 F0 (M1 )xa (u) (2.102)
2
xa (u) being a non-decreasing function of u. Hence we obtain:
a
∂ 1 1
φ(u, u0 , ϑ) ≤ δ|u0 |−2 F0 (M1 )xa (u) + δ|u0 |−2 G|a| (M|a|+1 ) (2.103)
∂ϑ 2 2
This inequality holds also with u replaced by u′ ∈ [0, u] on the left, the right
hand side being a non-decreasing function of u. Therefore taking on the left the
supremum with respect to u′ ∈ [0, u] yields:
1 1
xa (u) ≤ δ|u0 |−2 F0 (M1 )xa (u) + δ|u0 |−2 G|a| (M|a|+1 ) (2.104)
2 2
which by 2.87 implies:
4
xa (u) ≤ δ|u0 |−2 G|a| (M|a|+1 ) (2.105)
7
80
Setting u = δ this is the inequality of the lemma for φ with
4
Fk = Gk (2.106)
7
Integrating 2.94 once using the second of the initial conditions 2.96 we obtain:
a Z u a
∂ ∂φ ∂
(u, u0 , ϑ) = fa − e φ (u′ , u0 , ϑ)du′ (2.107)
∂ϑ ∂u 0 ∂ϑ
which implies:
a Z u
∂ ∂φ
(u, u0 , ϑ) ≤ |fa (u′ , u0 , ϑ)|du′
∂ϑ ∂u 0
Z u a
∂
+ e φ (u′ , u0 , ϑ)du′ (2.108)
0 ∂ϑ
By 2.97 the first integral on the right is bounded by:
|u0 |−2 G|a| (M|a|+1 ) (2.109)
while the second integral on the right is bounded by:
δ −1 |u0 |−2 F0 (M1 )xa (u) (2.110)
Substituting the bound 2.105 for the latter gives the inequality of the lemma
for ∂φ/∂u with
′ 4
Fk = 1 + F0 Gk (2.111)
7
(F0 (Mk+1 ) ≥ F0 (M1 ), F0 being a non-decreasing function). This completes the
inductive step and therefore the proof of the lemma.
Applying repeatedly ∂/∂u to equation 2.77 and using the estimates 2.88,
2.89 and 2.71 we deduce:
∂nφ
sup ≤ δ 1−n |u0 |−2 Fn,0 (Mn−1 ) : for n ≥ 2 (2.112)
Cu0 ∂un
where the Fn,0 are non-negative non-decreasing continuous functions on the

non-negative real line.
Similarly, applying repeatedly ∂/∂u to equation 2.94 and using Lemma 2.3
and the bound 2.82 we deduce:
a n
∂ ∂ φ
sup n
(u, u0 , ϑ) ≤ δ 1−n |u0 |−2 Fn,|a|
′′
(M|a|+n−1 )
(u,ϑ)∈[0,δ]×D2ρ ∂ϑ ∂u
: for n ≥ 2, |a| =
6 0 (2.113)
where the Fn,k are non-negative non-decreasing continuous functions on the real
line.
81
Putting together the results for the two stereographic charts we then obtain:
u
φ−1− ≤ δ|u0 |−2 F0,k (Mk+1 )
|u0 | Cδk (Cu0 )
∂φ 1
− ≤ |u0 |−2 F1,k (Mk+1 )
∂u |u0 | Cδk (Cu0 )
∂2φ
≤ δ −1 |u0 |−2 F2,k (Mk+1 ) (2.114)
∂u2 Cδk (Cu0 )
where the F0,k , F1,k , F2,k are new non-negative no-decreasing continuous func-
tions on the real line. Thus in accordance with the notation 2.54 we have:
u
φ−1− = M1 (δ|u0 |−2 )
|u0 |
∂φ 1
− = M1 (|u0 |−2 )
∂u |u0 |
∂2φ
= M1 (δ −1 |u0 |−2 ) (2.115)
∂u2
From 2.3, 2.57 and 2.115 we conclude that:
◦
|u0 |−2 g/− g/= M1 (δ 1/2 |u0 |−1 ) (2.116)
Also, from 2.68, 2.57 and 2.115 we conclude that:
2
χ̂ = M1 (δ −1/2 |u0 |) trχ − = M1 (|u0 |−2 ) (2.117)
|u0 |
◦
Now, since g/ is given in each stereographic chart by 1.217, that is:
◦
g/AB (ϑ) = w2 (ϑ)δAB (2.118)
where w is the function 2.6, if we denote by λ(u, u0 , ϑ) and Λ(u, u0 , ϑ) respec-
tively the smallest and largest eigenvalues of the matrix |u0 |−2 g/AB (u, u0 , ϑ),
then 2.116 implies that if δ is suitably small depending on M1 we have:
1 1
sup max{|λ(u, u0 , ϑ) − w2 (ϑ)|, |Λ(u, u0 , ϑ) − w2 (ϑ)|} ≤
(u,ϑ)∈[0,δ]×D2ρ 2 (1 + ρ2 )2
(2.119)
Since the function w satisfies:
1
≤ w(ϑ) ≤ 1 : ∀ϑD2ρ (2.120)
1 + ρ2
it follows that:
1 1
2 2
≤ λ(u, u0 , ϑ) ≤ Λ(u, u0 , ϑ) ≤ 1 +
2(1 + ρ ) 2(1 + ρ2 )2
: for all (u, ϑ) ∈ [0, δ] × D2ρ (2.121)
82
Let ξ be a Tpq type S tensorfield on Cu0 . Then by 2.121 there is a positive
constant Cp,q depending only on p, q and the initial choice of ρ such that:
−1
Cp,q |u0 |q−p |ξ| ≤
1/2
A ...A C ...C
|ξ|/g = g/A1 C1 ...g/Aq Cq (g/−1 )B1 D1 ...(g/−1 )Bp Dp ξB11...Bpq ξD11 ...Dqp
≤ Cp,q |u0 |q−p |ξ| (2.122)
Let then ξ be a C k Tpq type S tensorfield on Cu0 . We define the weighted

invariant C k norm of ξ by:
kξkCk (Cu0 ) = max sup δ n |u0 |m |∇

/m Dn ξ|/g

(2.123)
δ m+n≤k Cu
0
where ∇/ is the covariant derivative on the Su,u0 with respect to g/|Su,u . It follows
0
from 2.116 and 2.122 that there is a non-negative non-decreasing continuous
function Fk such that:
kξkCk (Cu0 ) ≤ |u0 |q−p Fk (Mk+1 )kξkC k (Cu0 ) (2.124)

δ δ
and:
kξkC k (Cu ) ≤ |u0 |p−q Fk (Mk+1 )kξkCk (Cu0 ) (2.125)
δ 0 δ
We say that a smooth Tpq type S tensorfield ξ on Cu0 is Ml (δ r |u0 |s ), for real
numbers r, s, and non-negative integer l,
ξ = Ml (δ r |u0 |s ) (2.126)
if for every non-negative integer k we have:
kξkCk (Cu0 ) ≤ δ r |u0 |s Fk (Mk+l ) (2.127)

δ
where Fk is a non-negative non-decreasing continuous function. Then 2.124,

2.125 imply the following lemma.
Lemma 2.4 Let ξ be a smooth Tpq type S tensorfield on Cu0 . Then for l ≥ 1
ξ = Ml (δ r |u0 |s ) implies ξ = Ml (δ r |u0 |s+q−p ) and conversely.
In particular 2.117 imply:

2
χ̂ = M1 (δ −1/2 |u0 |−1 ) trχ − = M1 (|u0 |−2 ) (2.128)
|u0 |
Also, 2.59 implies:
K̂ − |u0 |−2 = M2 (δ 1/2 |u0 |−3 ) (2.129)
Now K, the Gauss curvature of g/, is given in terms of K̂ and φ by:
K = φ−2 (K̂ − △
/ /ĝ log φ) (2.130)
83
while by the first of 2.115:
u
φ−1− = M1 (δ|u0 |−2 ) (2.131)
|u0 |
It then follows that:
K − |u0 |−2 = M3 (δ 1/2 |u0 |−3 ) (2.132)
We turn to consider the torsion ζ on Cu0 (see 1.63). By 2.62 and 1.81 we
have:
η = ζ, η = −ζ : along Cu0 (2.133)
Thus the propagation equation 1.82 becomes, along Cu0 :
Dζ + χ♯ · ζ = −β (2.134)
On the other hand, the Codazzi equation 1.148 reads:
/ χ − /dtrχ + χ♯ · ζ − trχ · ζ = −β
div (2.135)
Substituting for −β from 2.135 into 2.134 then yields the following propagation
equation for ζ along the generators of Cu0 :
Dζ + trχζ = ξ (2.136)
where ξ is the S 1-form:

ξ = div
/ χ − /dtrχ (2.137)
By 2.128:
ξ = M2 (δ −1/2 |u0 |−2 ) or, equivalently, ξ = M2 (δ −1/2 |u0 |−1 ) (2.138)
Equation 2.136 is a linear first order ordinary differential equation for ζ, which
in each stereographic chart reads:
∂ζA
+ trχζA = ξA (2.139)
∂u
The initial condition on S0,u is simply:
ζ|S0,u = 0 (2.140)
0
S0,u0 being a sphere lying on the spacelike hyperplane x0 = u0 with future

directed unit normal T̂ = ∂/∂x0 at the boundary of the Minkowskian region
M0 , the vectorfields L̂ and L̂ having projection T̂ along T̂ in M0 . Equation
2.139 together with the initial condition 2.140 has a unique smooth solution
ζ on Cu0 . This equation and the equations to follow as well as the derivative
equations, obtained by applying (∂/∂ϑ)a , for some multi-index a to the original
equations, are all equations of the form:
dx
+a·x=b (2.141)
dt
84
on the interval [0, δ], where the unkown x is a function of t with values in ℜn
for some n, a is a given smooth function of t with values in the n dimensional
matrices, and b is a given smooth function of t with values in ℜn . Denoting by
| |, as in 2.47, the Euclidean norm on ℜn , and by < , > the Euclidean inner
product on ℜn , equation 2.141 implies:
d 2
|x| = 2(− < x, a · x > + < x, b >) ≤ 2|x|(|a||x| + |b|) (2.142)
dt
2
where |a(t)| is the Euclidean norm of a(t) in ℜn , as in 2.47. It then follows
that (see Lemma 3.1 of the next chapter):
Z t
′
|x(t)| ≤ eA(t) |x(0)| + e−A(t ) |b(t′ )|dt′ (2.143)
0
where: Z t
A(t) = |a(t′ )|dt′ (2.144)
0
In this manner, using the second of 2.117 and the second of 2.138 we deduce, in
the present case:
ζ = M2 (δ 1/2 |u0 |−1 ) hence ζ = M2 (δ 1/2 |u0 |−2 ) (2.145)
We turn to trχ. Now, the first of 1.41 implies:
Dg/−1 = −2Ωχ♯♯ (2.146)
where χ♯♯ is the symmetric 2-contravariant S tensorfield with components:
χ♯♯AB = χCD (g/−1 )AC (g/−1 )BD (2.147)
It follows that for any 2-covariant S tensorfield θ we have:
Dtrθ = trDθ − 2Ω(χ, θ) (2.148)
Taking into account 2.148 and 1.123 the trace of equation 1.170, is the equation:
n o
D(Ωtrχ) = Ω2 2div/ η + 2|η|/2g − (χ, χ) + 2ρ (2.149)
Substituting for ρ from the Gauss equation 1.132 we then obtain the equation:
n o
D(Ωtrχ) = Ω2 2div/ η + 2|η|/2g − 2K − trχtrχ (2.150)
which along Cu0 becomes the following propagation equation for trχ along the
generators of Cu0 :
Dtrχ + trχtrχ = 2λ (2.151)
where λ is the function:
/ ζ + |ζ|/2g
λ = −K − div (2.152)
85
By 2.132 and 2.145:
λ + |u0 |−2 = M3 (δ 1/2 |u0 |−3 ) or, equivalently, λ + |u0 |−2 = M3 (δ 1/2 |u0 |−3 )
(2.153)
Equation 2.151 is a linear first order ordinary differential equation for trχ, which
∂trχ
+ trχtrχ = 2λ (2.154)
∂u
The initial condition on S0,u0 is:
2
trχ S0,u0
=− (2.155)
|u0 |
the last being the mean curvature of a round sphere of radius r0 = |u0 | in
Euclidean 3-dimensional space with respect to the inner normal. Equation 2.154
together with the initial condition 2.155 has a unique smooth solution trχ on
Cu0 . Setting:
2 2u
trχ = − + +ν (2.156)
|u0 | |u0 |2
and:
2
trχ = +ν (2.157)
|u0 |
equation 2.154 becomes the following linear first order ordinary differential equa-
tion for ν:
∂ν
+ trχν = 2λ′ (2.158)
∂u
where:
′ −2 2u u ν
λ = λ + |u0 | − 3
+ 1− (2.159)
|u0 | |u0 | |u0 |
and the initial condition 2.155 becomes simply:
ν|S0,u = 0 (2.160)
0
Now, according to the second of 2.117:
ν = M1 (|u0 |−2 ) (2.161)
which together with 2.153 yields:
λ′ = M3 (|u0 |−3 ) (2.162)
Using this and the second of 2.117 we readily deduce from 2.159, 2.160:
ν = M3 (δ|u0 |−3 ) hence also ν = M3 (δ|u0 |−3 ) (2.163)
86
We turn to χ̂. Equation 1.170 becomes along Cu0 :
Dχ = −∇ ˜ + 2ζ ⊗ ζ + 1 (χ × χ + χ × χ) + ρg/
/ζ − ∇
/ζ (2.164)
2
Decomposing χ and χ as in 1.186 and using the first of equations 1.41 we obtain:
1
Dχ = Dχ̂ + trχχ̂ + (Dtrχ + trχtrχ)g/ (2.165)
2
and by 2.148 applied to χ̂ we have:
trDχ̂ = 2(χ̂, χ̂) (2.166)
Substituting then 2.165 on the left in 2.164 and taking the trace free parts
of both sides, we obtain, taking into account the identity 1.187, the following
propagation equation for χ̂ along the generators of Cu0 :
1
Dχ̂ − (χ̂, χ̂)g/ − trχχ̂ = θ (2.167)
2
where θ is the trace-free symmetric 2-covariant S tensorfield:
1
θ = −∇ˆ + ζ ⊗ζ
/⊗ζ ˆ − trχχ̂ (2.168)
2
By 2.145, 2.156, 2.163, and the first of 2.128:
θ = M3 (δ −1/2 |u0 |−2 ) or, equivalently, θ = M3 (δ −1/2 ) (2.169)
Equation 2.169 is a linear first order ordinary differential equation for χ̂ which
∂ χ̂AB 1
− (χ̂, χ̂)g/AB − trχχ̂AB = θAB (2.170)
∂u 2
where:
(χ̂, χ̂) = (g/−1 )CE (g/−1 )DF χ̂CD χ̂EF
The initial condition on S0,u0 is simply:
χ̂ S0,u0
=0 (2.171)
S0,u0 being a round sphere in Euclidean 3-dimensional space, hence umbilical.

Equation 2.170 together with the initial condition 2.171 has a unique smooth
solution on Cu0 . Writing:
(χ̂, χ̂) = χ̂♯♯ · χ̂, (χ̂♯♯ )AB = (g/−1 )AC (g/−1 )BD χ̂CD
and noting that by 2.122:

|χ̂♯♯ | ≤ C|u0 |−4 |χ̂| (2.172)
87
√
while since |g/|/g = 2 we have:
|g/| ≤ C|u0 |2 , (2.173)
and using 2.117 and 2.169, we deduce:
χ̂ = M3 (δ 1/2 ) hence χ̂ = M3 (δ 1/2 |u0 |−2 ) (2.174)
We now consider the curvature components along Cu0 . Consider equation

1.57 along Cu0 :
Dχ = χ × χ − α (2.175)
Decomposing χ as in 1.186 and using the first of equations 1.41 we obtain:
1
Dχ = Dχ̂ + trχχ̂ + (Dtrχ + (trχ)2 )g/ (2.176)
2
and by 2.148 applied to χ̂ we have:
trDχ̂ = 2|χ̂|2 (2.177)
Substituting then 2.176 on the left in 2.175 and taking the trace-free parts of
both sides, we obtain, taking into account the identity 1.188,
Dχ̂ − |χ̂|/2g g/ = −α (2.178)
The first of 2.128 then yields:
α = M2 (δ −3/2 |u0 |−1 ) (2.179)
Next, β is given by 2.135. Using 2.128 and 2.145 we then obtain:
β = M2 (δ −1/2 |u0 |−2 ) (2.180)
Next, ρ is given by the Gauss equation 1.142 which we may write in the
form:
1 1
K + trχtrχ − (χ̂, χ̂) = −ρ (2.181)
4 2
Using 2.132, 2.128, 2.156, 2.163, and 2.174, we then obtain:
ρ = M3 (|u0 |−3 ) (2.182)
Also, σ is given by equation 1.163 (see 1.169) which reads:

1
curl
/ ζ + χ̂ ∧ χ̂ = σ (2.183)
2
Using 2.128, 2.145, and 2.174, we then obtain:
σ = M3 (|u0 |−3 ) (2.184)
88
To derive appropriate estimates for the remaining curvature components,
namely for β and α, we appeal to the Bianchi indentities, given by Proposition
1.2. The sixth Bianchi identity reads, along Cu0 :
1
Dβ + trχβ − χ̂♯ · β = κ (2.185)
2
where κ is the S 1-form:
κ = −d
/ρ + ∗
/dσ + 3ζρ − 3 ∗ ζσ + 2χ̂♯ · β (2.186)
By 2.182, 2.184, 2.145, 2.174, and 2.180:
κ = M4 (|u0 |−4 ) or, equivalently, κ = M4 (|u0 |−3 ) (2.187)
Equation 2.185 is a linear first order ordinary differential equation for β along
he generators of Cu0 , which in each stereographic chart reads:
∂β A 1
+ trχβ A − χ̂♯B
A β B = κA (2.188)
∂u 2
The initial condition on Cu0 is simply:
β S0,u0
=0 (2.189)
S0,u0 lying on the boundary of the Minkowskian region M0 . Equation 2.185

together with the initial condition 2.189 has a unique smooth solution β on
Cu0 . Using 2.117 and 2.187 we deduce:
β = M4 (δ|u0 |−3 ) hence β = M4 (δ|u0 |−4 ) (2.190)
In view of the fact that by 2.148 applied to α we have, along Cu0 :
trDα = 2(χ̂, α)g/ (2.191)
the second of the Bianchi identities of Proposition 1.2 reads, along Cu0 :
1
Dα − (χ̂, α)g/ − trχα = τ (2.192)
2
where τ is the trace-free symmetric 2-covariant S tensorfield:
τ = −∇
/⊗β ˆ − 3χ̂ρ + 3 ∗ χ̂σ
ˆ + 5ζ ⊗β (2.193)
By 2.190, 2.145, 2.174, 2.182, and 2.184:
τ = M5 (δ 1/2 |u0 |−5 ) or, equivalently, τ = M5 (δ 1/2 |u0 |−3 ) (2.194)
the principal term in τ being the term −∇ˆ but the leading terms in behavior
/⊗β,
with respect to δ being the terms −3χ̂ρ + ∗ χ̂σ. Equation 2.192 is a linear first
89
order ordinary differential equation for α along the generators of Cu0 , which in
each stereographic chart reads:
∂αAB 1
− (χ̂, α)g/AB − trχαAB = τAB (2.195)
∂u 2
where:
(χ̂, α) = (g/−1 )CE (g/−1 )DF χ̂CD αEF
α|S0,u = 0 (2.196)
0

together with the initial condition 2.196 has a unique smooth solution α on
Cu0 . Noting 2.173 and the fact that by 2.122:
|(χ̂, α)| ≤ |χ̂|/g |α|/g ≤ C|u0 |−4 |χ̂||α| (2.197)
and using 2.117 and 2.194, we deduce:
α = M5 (δ 3/2 |u0 |−3 ) hence α = M5 (δ 3/2 |u0 |−5 ) (2.198)
Of the connection coefficients only ω remains to be appropriately estimated

along Cu0 . This satisfies the propagation equation 1.107, which along Cu0 be-
comes:
Dω = h (2.199)
where h is the function:
h = −3|ζ|/2g − ρ (2.200)
By 2.145 and 2.182:
h = M3 (|u0 |−3 ) or, equivalently, h = M3 (|u0 |−3 ) (2.201)
Equation 2.199 reads in each stereographic chart, simply:

∂ω
=h (2.202)
∂u
ω|S0,u = 0 (2.203)
0
For, u is an affine parameter along the generators of C 0 , hence:
ω = 0 : along C 0 (2.204)
solution ω on Cu0 and by 2.201 we have:
ω = M3 (δ|u0 |−3 ) hence ω = M3 (δ|u0 |−3 ) (2.205)
90
Our construction requires that we also appropriately estimate Dω and Dα
along Cu0 . To estimate the first we consider again equation 1.107. Applying D
to this equation and using the commutation formula 1.91 we obtain:
DDω = −4Ω2 ζ ♯ · /dω + Ω2 2(η − η, Dη) + 2(η, Dη) − Dρ

+2Ω2 ω(2(η, η) − |η|/2g − ρ) + 2Ω3 (−2η · χ · η ♯ + η ♯ · χ · η ♯ )

(2.206)
Substituting for Dη from 1.173, for Dη from 1.83, and for Dρ from the eighth
Bianchi identity of Proposition 1.2, and evaluating the result along Cu0 , yields
the following propagation equation for Dω along Cu0 :
DDω = ḣ (2.207)
where ḣ is the function:
ḣ = −12(ζ, /dω) − 2ω(3|ζ|/2g + ρ) + div
/ β
3 1
+ trχρ + 7(ζ, β) + (χ̂, α) + 12ζ ♯ · χ · ζ ♯ (2.208)
2 2
By 2.205, 2.145, 2.174, 2.182, 2.190, and 2.198:
ḣ = M5 (|u0 |−4 or, equivalently, ḣ = M5 (|u0 |−5 ) (2.209)
Equation 2.209 reads in each stereographic chart, simply:
∂Dω
= ḣ (2.210)
∂u
In view of 2.204 the initial condition on S0,u0 is simply:
Dω|S0,u = 0 (2.211)
0
solution Dω on Cu0 and by 2.209 we have:
Dω = M5 (δ|u0 |−4 ) hence ω = M5 (δ|u0 |−4 ) (2.212)
Finally, to obtain an appropriate estimate for Dα along Cu0 we consider
again the second Bianchi identity of Proposition 1.2. Applying D to this identity
and taking into account Lemma 1.4 we obtain:
1
DDα − (Ωχ̂, Dα)g/ − ΩtrχDα + 2ωDα = −4L /Ω2 ζ ♯ α
2
1
+(D(Ωχ̂), α)g/ − D(Ωtrχ)α + 2Ωχ(Ωχ̂, α) − 4Ω2 (χ, χ̂ × α)g/
2
ˆ + (ζ − 4η)⊗β ˆ − 3χ̂ρ + 3 ∗ χ̂σ

−2(Dω)α + Ωω −∇ /⊗β
n
+Ω −∇ ˆ
/⊗Dβ + 2DΓˆ/ · β + 2Ωχ̂div
/ β
ˆ + (ζ − 4η)⊗Dβ
+(Dζ − 4Dη)⊗β ˆ − 2Ωχ̂(ζ − 4η, β)
−3(Dχ̂)ρ + 3(D ∗ χ̂)σ − 3χ̂Dρ + 3 ∗ χ̂Dσ (2.213)
91
Here DΓ is the Lie derivative with respect to L of the induced connection on
Su,u , given by (see Lemma 4.1):
/)C
(DΓ /−1 )CD (∇
AB = (g /A (Ωχ)BD + ∇
/B (Ωχ)AD − ∇
/D (Ωχ)AB ) (2.214)
ˆ/ is the trace-free part of DΓ
and DΓ / with respect to the lower indices:
ˆ/)C = (DΓ 1
(DΓ AB /)C
AB − g/AB (g/−1 )DE (DΓ
/)C
DE (2.215)
2
ˆ/ · β + 2Ωχ̂div
The terms 2DΓ / β represent the commutator:
−D∇ˆ +∇
/⊗β ˆ
/⊗Dβ
(see Lemma 4.1). We substitute for D(Ωχ) from 1.171, for Dω from 1.108,
for Dζ from 1.78, for Dη from 1.83, and for Dχ from 1.59, and evaluate the
results on Cu0 . Also, we substitute for Dβ from the fourth Bianchi identity of
Proposition 1.2, which reads, along Cu0 :
3
Dβ + trχβ − χ̂♯ · β − ωβ = −div
/ α+ζ (2.216)
2
and for Dρ and Dσ from the eighth and tenth Bianchi identities of Proposition
1.2, respectively, evaluated along Cu0 . Taking into account the fact that for any
trace-free symmetric 2-covariant S tensorfield θ we have:
∇ˆ / θ=△
/⊗div / θ − 2Kθ (2.217)
and the fact that by 2.62 we have:
ω = 0 : along Cu0 (2.218)
we deduce the following propagation equation for Dα along the generators of
Cu0 :
1
DDα − (χ̂, Dα)g/ − trχDα = τ̇ (2.219)
2
where τ̇ is the symmetric 2-covariant S tensorfield:
τ̇ = △
/ α − 2Kα − 4L
/ζ ♯ α + ∇ˆ
/⊗((3/2)trχβ − χ̂♯ · β − ωβ − ζ ♯ · α)
ˆ / α − ζ ♯ · α + (3/2)trχβ − χ̂♯ · β − ωβ)
−5ζ ⊗(div
+3χ̂(div
/ β + (3/2)trχρ + (ζ, β) + (1/2)(χ̂, α))
∗
/ β + (3/2)trχσ + (ζ, ∗ β) + (1/2)χ̂ ∧ α)
−3 χ̂(curl
+2χ̂div
/ β + ω(−∇ ˆ + 5ζ ⊗β
/⊗β ˆ − 3χ̂ρ + 3χ̂σ)
ˆ/ · β + (d
+2DΓ /ω − 5χ♯ · ζ − 5β)⊗β
ˆ
+((∇ˆ + ζ ⊗ζ
/⊗ζ ˆ + (1/2)trχχ̂ − (1/2)trχχ̂), α)g/
/ ζ + |ζ|/2g − (1/2)(χ, χ) + ρ)α + 2χ(χ̂, α) − 4(χ, χ̂ × α)g/
−(div
+2(3|ζ|2 + ρ)α − 10χ̂(ζ, β) − 3(ωχ̂ + |χ̂|/2g g/ − α)ρ + 3(ω ∗ χ̂ − ∗
α)σ
(2.220)
92
Using 2.128, 2.132, 2.145, 2.156, 2.163, 2.174, 2.182, 2.184, 2.190, 2.198, and
2.205, we find:
τ̇ = M7 (δ 1/2 |u0 |−6 ) or, equivalently, τ̇ = M7 (δ 1/2 |u0 |−4 ) (2.221)
the principal term in τ̇ being the term △ / α, but the leading terms in behavior
with respect to δ being the terms (9/2)trχ(χ̂ρ − ∗ χ̂σ). Equation 2.219 is a
linear first order ordinary differential equation for Dα along the generators of
Cu0 , which in each stereographic chart reads:
∂(Dα)AB 1
− (χ̂, Dα)g/AB − trχ(Dα)AB = τ̇AB (2.222)
∂u 2
where:
(χ̂, Dα) = (g/−1 )CE (g/−1 )DF χ̂CD (Dα)EF
Dα|S0,u = 0 (2.223)
0

together with the initial condition 2.223 has a unique smooth solution Dα on
Cu0 . Equation 2.219 is in fact identical in form to equation 2.195 but with τ̇ in
the role of τ . In view of 2.221, we deduce:
Dα = M7 (δ 3/2 |u0 |−4 ) hence Dα = M7 (δ 3/2 |u0 |−6 ) (2.224)
Finally it is clear that the procedure which we have discussed leads induc-
tively to the conclusion that Dn ω and Dn α are smooth on Cu0 for all non-
negative integers n and, together with the results already obtained, this im-
plies, through the optical structure equations and the Bianchi identities, that
the transversal derivatives of any order of the all the connection coefficients
and of all the curvature components are smooth on Cu0 . However, no further
estimates are needed.
2.2 Construction of the solution in an initial do-

main
Our setup, in summary, is the following. We have two null hypersurfaces, C 0
and Cu0 intersecting at a sphere S0,u0 . The freely specifiable initial data on each
of these hypersurfaces consists of the conformal metric ĝ/. In our case the data is
trivial on C 0 . Moreover, the full induced metric g/ is given on S0,u0 , in our case
the metric of a round sphere of radius |u0 | in Euclidean 3-dimensional space.
Also, the traces trχ and trχ of the 2nd fundamental forms of S0,u0 relative to
C 0 and Cu0 are given. In our case these are the mean curvatures of a round
sphere of radius |u0 | in Euclidean 3-dimensonal space relative to the inner and
outer normals, respectively. Finally, the torsion ζ of S0,u0 is given, which in our
93
case vanishes, being the torsion of a surface lying on a spacelike hyperplane in
Minkowski spacetime.
With initial data on two interesecting null hypersufaces set up in this way,
the theorem of Rendall [R] gives a spacetime neighborbood U of S0,u0 , bounded
in the past by the two null hypersurfaces, and a unique smooth solution of the
vacuumT Einstein
S equations in U , (U, g) being a development of the initial data
on U (C 0 Cu0 ). The solution is obtained in “harmonic” (also called “wave”)
coordinates, adapted to the two null hypersurfaces.
The construction to be presented in the third section of Chapter 16, then
produces a smooth solution in canonical coordinates, as defined in Chapter 1,
in a domain M ′ ⊂ U , of the form M ′ = M \ (M0 \ C 0 ), with M , as in Chapter
1, corresponding to some c∗ > u0 but sufficiently close to u0 .
94
Chapter 3
L∞ Estimates for the

Connection Coefficients
3.1 Introduction
As we discussed in Chapter 1, we are assuming that we have a spacetime man-
ifold (M, g) which is aTsmooth solution of the vacuum Einstein equations such
that the domain M0 M is isometric to a domain in Minkowski spacetime.
From now on we restrict attention to the non-trivial part of the spacetime man-
ifold (M, g): [
M ′ = M \ (M0 \ C 0 ) = Su,u (3.1)
(u,u)∈D′
′
where D is the parameter domain (see 1.4):
[
D′ = {(u, u) : u ∈ [0, δ), u ∈ [u0 , c∗ − δ]}
{(u, u) : u ∈ [0, c∗ − u), u ∈ (c∗ − δ, c∗ )}
= {(u, u) : u ∈ [u0 , c∗ − u), u ∈ [0, δ)} (3.2)
for c∗ ≥ u0 + δ, and:
D′ = {(u, u) : u ∈ [0, c∗ − u), u ∈ [u0 , c∗ )}
= {(u, u) : u ∈ [u0 , c∗ − u), u ∈ [0, c∗ − u0 )} (3.3)
for c∗ < u0 + δ (see Figures 1.1, 1.2).
The inner boundary of M ′ and outer boundary of the Minkowskian region
M0 is the incoming null hypersurface C 0 . We denote in the following by Cu the
part lying in M ′ of what was denoted in Chapter 1 by Cu .
We recall from Chapter 1 that according to the fundamental requirement on
(M ′ , g) of the continuity argument, the generators of the Cu and the C u have
no end points in M ′ . It follows that for each (u, u) ∈ D′ Su,u is a spacelike
surface embedded in M ′ .
95
The objective of the present chapter is to derive L∞ estimates for the connec-
tion coefficients in M ′ on the basis of L∞ bounds for the curvature components
in M ′ and the following basic bootstrap assumption:
1
A0: 2 ≤ Ω ≤ 2 : in M ′
The arguments of the present chapter, as well as those of the next chap-
ter, rely only on the propagation equations, equations 1.41, 1.57, 1.62, 1.82,
1.83, 1.107, 1.108 of Chapter 1. What is assumed in regard to the curvature
components in the present chapter is that the following quantities are finite:
R∞
0 (α) = sup(|u|δ 3/2 |α|)
M′
R∞
0 (β) = sup(|u|2 δ 1/2 |β|)
M′
R∞
0 (ρ) = sup(|u|3 |ρ|)
M′
R∞
0 (σ) = sup(|u|3 |σ|)
M′
R∞
0 (β) = sup(|u|4 δ −1 |β|)
M′
R∞
0 (α) = sup(|u|9/2 δ −3/2 |α|) (3.4)
M′
By the results of Chapter 2, the corresponding quantities on Cu0 , obtained by

replacing the supremum on M ′ by the supremum on Cu0 , are all bounded by
a non-negative non-decreasing continous function of M5 , the quantity requiring
Mk with the highest k being the one corresponding to α.
Throughout this monograph we shall denote by C various numerical con-
stants which are independent of u0 , u1 , c∗ or δ.
We shall first derive an L∞ estimate for χ′ , then an L∞ estimate for χ′ .
After that we shall derive L∞ estimates for η, η together, as the propagation
equations 1.82, 1.83 are coupled. Finally, we shall derive L∞ estimates for ω
and ω.
Now the first variational formulas 1.41 imply:
Dg/−1 = −2Ωχ♯♯ , Dg/−1 = −2Ωχ♯♯ (3.5)
where χ♯♯ , χ♯♯ , are the symmetric 2-contravariant S tensorfields with compo-
nents, in an arbitrary local frame field (eA : A = 1, 2) for Su,u ,
χ♯♯AB = χCD (g/−1 )AC (g/−1 )BD , χ♯♯AB = χCD (g/−1 )AC (g/−1 )BD (3.6)
It follows that for any 2-covariant S tensorfield θ we have:
Dtrθ = trDθ − 2Ω(χ, θ), Dtrθ = trDθ − 2Ω(χ, θ) (3.7)
Thus in view of 1.123 and equations 1.179, the traces of the second variation
equations 1.57 and 1.62 read:
1
Dtrχ′ = −Ω2 |χ′ |2 = − Ω2 (trχ′ )2 − Ω2 |χ̂′ |2 (3.8)
2
96
1
Dtrχ′ = −Ω2 |χ′ |2 = − Ω2 (trχ′ )2 − Ω2 |χ̂′ |2 (3.9)
2
Decomposing χ′ according to 1.186, we have, by the first of equations 1.41,
1
Dχ̂′ = Dχ′ − Ω2 trχ′ χ′ − g/Dtrχ′
2
′ 2 ′ ′ 1
Dtrχ′ + Ω2 (trχ′ )2 g/

= Dχ − Ω trχ χ̂ −
2
Substituting for Dχ′ from equation 1.57 and noting that by 1.52 and the identity
1.188:
1
χ′ × χ′ = χ̂′ × χ̂′ + trχ′ χ̂′ + g/(trχ′ )2
4
′ ′ 1 ′2 1
= trχ χ̂ + |χ̂ | + (trχ′ )2 g/
2 4
we conclude that we have simply:
D̂χ̂′ = −α (3.10)
Similarly we have:
D̂χ̂′ = −α (3.11)
In the following we shall make repeated use of the following elementary
lemma on ordinary differential inequalities.
Lemma 3.1 Let v be a non-negative function defined on an interval I with

initial point t0 and let v 2 be weakly differentiable in I and satisfy the ordinary
differential inequality
d 2
(v ) ≤ 2v(av + b)
dt
where a, b are integrable functions on I with b non-negative. We then have:
Z t
A(t) −A(t′ ) ′ ′
v(t) ≤ e v(t0 ) + e b(t )dt : ∀t ∈ I
t0
where Z t
A(t) = a(t′ )dt′
t0
Proof: Given a positive real number ε we set:

p
vε = ε2 + v 2 (3.12)
Then vε is weakly differentiable on I and satisfies:
dvε 1 d 2 v2 v
= (v ) ≤ a + b (3.13)
dt 2vε dt vε vε
97
Now, we have:
v2 ε2
vε − =√
vε ε2 + v 2
hence:
v2
0 ≤ vε − ≤ε (3.14)
vε
Therefore, writing:
v2 v2

a = avε − a vε − (3.15)
vε vε
we have:
v2

− a vε − ≤ ε|a| (3.16)
vε
Also, since
v
≤ 1,
vε
we have:
v
b ≤b (3.17)
vε
Substituting 3.15 - 3.17 in 3.13 we obtain:
dvε
≤ avε + b + ε|a| (3.18)
dt
Integrating from the initial point t0 , yields, for all t ∈ I:
Z t
p ′
A(t)
vε ≤ e 2 2
ε + v (t0 ) + e−A(t ) b(t′ )dt′
t0
Z t
′
−A(t ) ′ ′
+ε e |a|(t )dt (3.19)
t0
This holds for every ε > 0. Taking then the limit ε → 0 yields the lemma.
3.2 L∞ Estimates for χ′

Proposition 3.1 Under the bootstrap assumption A0 there are numerical
constants C such that if
1
δ≤
C(1 + R∞ 0 (α))
2
then the following inequalities hold on M ′ :

1 1 1 C
≥ trχ′ ≥ δ + (R∞ 2
: on M ′

− 0 (α))
|u| 2 |u| |u|2
CR∞ 0 (α)
|χ̂′ | ≤ : on M ′
|u|δ 1/2
98
Proof: Equations 3.8 and 3.10 together with the first of equations 1.41 constitute
a nonlinear system of ordinary differential equations along each generator of each
Cu for g/, trχ′ , χ̂′ , if we think of Ω and α as given. The initial conditions for
this system are given on C 0 , the outer boundary of the Minkowskian region M0 .
Along C 0 we have:
Ω = 1 : along C 0 (3.20)
u being an affine parameter along the generators of C 0 , and:
2
trχ′ = , χ̂′ = 0 : along C 0 (3.21)
|u|
S0,u being a Euclidean sphere of radius |u|.

Consider then a given generator of a given Cu . We set:
1 1
trχ′ = −x (3.22)
2 |u|
Integrating equation 3.8 from C 0 along the given generator using as an initial
condition the first of 3.21, which takes the form
x = 0 : on C 0 , (3.23)
yields:
u
1 1 ′2
Z
2
x= Ω (trχ ) + |χ̂ | du′
′ 2
(3.24)
0 4 2
Thus x ≥ 0.
For positive numerical constants a and k let P(t) be the property:
1
aδ + k(R∞ 2

P(t) : x ≤ 2 0 (α)) : for all u ∈ [0, t]
|u|
The constants a and k shall be chosen in the sequel. P(0) is true and it follows
by continuity that P(t) is true for sufficiently small positive t. Let t∗ be the
least upper bound of the set of values of t ∈ [0, min{δ, c∗ − u}) for which P(t) is
true. If t∗ = min{δ, c∗ − u} then P(t) is true for all t ∈ (0, min{δ, c∗ − u}), that
is, the estimate holds everywhere along the generator under consideration.
Suppose then that t∗ < min{δ, c∗ − u}. Then by continuity P(t∗ ) is true.
Consequently:
1
|trχ′ | ≤ |u|−1 max{1, |u|−1 aδ + k(R∞ 2 ∗

0 (α)) } : for all u ∈ [0, t ] (3.25)
2
along the generator in question. Since |u| ≥ |c∗ | ≥ 1, setting
y = 1 + k(R∞
0 (α))
2
(3.26)
then, provided that

aδ ≤ 1 (3.27)
99
we have
max{1, |u|−1 aδ + k(R∞ 2

0 (α)) } ≤ y
hence:
2y
|trχ′ | ≤ (3.28)
|u|
Now, by 3.5:
D|χ̂′ |2 = D(χ̂′ , χ̂′ ) = D((g/−1 )AC (g/−1 )BD χ̂′AB χ̂′CD )

= −4Ω2 χ′♯♯AC χ̂′AB χ̂′♯B
C + 2χ̂
♯♯AB
(Dχ̂′ )AB
= −4Ω (χ , χ̂ × χ̂ ) + 2(χ̂ , Dχ̂′ )
2 ′ ′ ′ ′
Since by the identity 1.188

1 ′2
χ̂′ × χ̂′ = |χ̂ | g/
2
and (χ′ , g/) = trχ′ , while (χ̂′ , Dχ̂′ ) = (χ̂′ , D̂χ̂′ ), we obtain, by equation 3.10:
D|χ̂′ |2 + 2Ω2 trχ′ |χ̂′ |2 = −2(χ̂′ , α) (3.29)
Since
|(χ̂′ , α)| ≤ |χ̂′ ||α|
3.29 implies:
D|χ̂′ |2 + 2Ω2 trχ′ |χ̂′ |2 ≤ 2|χ̂′ ||α| (3.30)
This is an ordinary differential inequality along the generator under considera-
tion. Applying Lemma 3.1 with initial point the point of intersection with C 0 ,
yields, in view of the second of 3.21:
Z u
′ −F
|χ̂ | ≤ e eF |α|du′ (3.31)
0
where: Z u
F = Ω2 trχ′ du′ (3.32)
0
By virtue of 3.28 and the bootstrap assumption A0 we have, recalling that
u ∈ [0, δ],
u u
8yδ
Z Z
′ 2 ′ ′′
|F (u) − F (u )| = Ω trχ du ≤ Ω2 |trχ′ |du′ ≤ (3.33)
u′ 0 |u|
Hence:
|F (u) − F (u′ )| ≤ log 2 (3.34)
provided that:
log 2
δ≤ (3.35)
8y
100
Substituting 3.34 and the first of the definitions 3.4 in 3.31 we obtain, recalling
again that u ∈ [0, δ],
2R∞ 0 (α)
|χ̂′ | ≤ (3.36)
|u|δ 1/2
Let us now evaluate 3.24 at u = t∗ . In view of the bounds 3.28 and 3.36 and
assumption A0 we obtain:
Z t∗
2(R∞ 2

4 0 (α)) 4
x(t∗ ) ≤ y 2
+ du ≤ (y 2 δ + 2(R∞ 2
0 (α)) ) (3.37)
0 |u|2 δ |u|2
since t∗ ∈ (0, δ]. Suppose then that:
(4y 2 − a)δ < 8(R∞
0 (α))
2
(3.38)
The inequality 3.37 would then imply that:
1
x(t∗ ) < aδ + 16(R∞ 2

0 (α)) (3.39)
|u|2
Therefore, setting
k = 16 (3.40)
∗
the inequality defining property P would not be saturated at u = t . Then, by
continuity, property P would be true for some t > t∗ contradicting the definition
of t∗ .
Let us set
a=8 (3.41)
Then 3.35 implies 3.27. Also, in view of the choice 3.40, definition 3.26 becomes:
y = 1 + 16(R∞
0 (α))
2
(3.42)
and condition 3.38 takes the form:
8(y 2 − 2)δ < y − 1 (3.43)
√
This is a non-trivial condition for y > 2. Comparing with 3.35, the minimum
value of the function
(y − 1)y √
2
on ( 2, ∞)
y −2
is √
1 + (3/4) 2
√ > log 2
1+ 2
Therefore 3.35 implies 3.38 as well. It follows that with the choices 3.40 and
3.41, t∗ = min{δ, c∗ − u}, P(t) is true for all t ∈ (0, min{δ, c∗ − u}), hence also
the estimate 3.36 holds for all u ∈ [0, min{δ, c∗ − u}), provided that δ satisfies
the smallness condition 3.35. Finally, the smallness condition
log 2
δ≤
128(1 + R∞0 (α))
2
implies condition 3.35. This completes the proof of the proposition.
101
3.3 L∞ Estimates for χ′
We proceed to derive an L∞ estimate for χ′ . Now along Cu0 we have:
Ω = 1 : along Cu0 (3.44)
u coinciding up to the additive constant r0 with the affine parameter s along

the generators of Cu0 .
Consider any (u1 , u1 ) ∈ D′ . Let D1 be the parameter subdomain:
D1 = [0, u1 ] × [u0 , u1 ] ⊂ D′ (3.45)
and let M1 be the corresponding subdomain of M ′ :

[
M1 = Su,u (3.46)
(u,u)∈D1
In the following lemmas we fix attention to M1 . We shall derive uniform esti-

mates in M1 which are independent of (u1 , u1 ) ∈ D′ . Uniform estimates in M ′
u
will then follow. We denote by Cu 1 the part of Cu which corresponds to u ≤ u1 .
u1
Thus for u ∈ [u0 , u1 ] Cu is the part of Cu lying in M1 .
Let s∗ be the least upper bound of the set of values of s ∈ [u0 , u1 ] such that:
1 u
|Ω − 1| ≤ |u|−2 : on Cu 1 , for all u ∈ [u0 , s] (3.47)
18
Then by continuity s∗ > u0 and we have:
1 −2 u
|Ω − 1| ≤ |u| : on Cu 1 , for all u ∈ [u0 , s∗ ] (3.48)
18
Let us denote:
D0∞ (χ̂) = sup(|u0 |2 δ −1/2 |χ̂|) (3.49)
Cu0
Here the supremum is meant to be taken on all of Cu0 , not only on the part
which lies in M ′ . This is the same for c∗ ≥ u0 + δ, but not for c∗ ∈ (u0 , u0 + δ).
∗
In the latter case the part of Cu0 lying in M ′ is the part Cuc 0 −u0 . By the results
of Chapter 2, D0∞ (χ̂) is bounded by a non-negative non-decreasing continuous
function of M3 .
∗
Also, let us denote by M1s the subset of M1 where u ≤ s∗ :
u
∗ [
M1s = Cu 1 (3.50)
u∈[u0 ,s∗ ]
Lemma 3.2 Suppose that on Cu0 trχ < 0 and:
2 1
+ |u0 | < : on Cu0
trχ 9
102
Then trχ′ < 0 on M ′ and there are numerical constants C such that if
C(D0∞ (χ̂) + R∞ 2
0 (α)) δ ≤ 1
∗
the following inequalities hold on M1s :
2 1 ∗
′
+ |u| ≤ : on M1s
trχ 3
δ 1/2 ∞ δ 3/2 ∞

|χ̂′ | ≤ C
∗
D (χ̂) + R (α) : on M1s
|u|2 0 |u|7/2 0
Proof : Equations 3.9 and 3.11 together with the second of equations 1.41
constitute a nonlinear system of ordinary differential equations along each gen-
erator of each C u for g/, trχ′ , χ̂′ , if we think of Ω and α as given. The initial
conditions for this system are given on Cu0 . Since by 3.9 Dtrχ′ ≤ 0 while by
the assumptions of the lemma trχ′ = trχ < 0 on Cu0 , it follows that trχ′ < 0
in M ′ .
Consider then a given generator of a given C u , u ∈ [0, u1 ]. We set:
2
x=− (3.51)
trχ′
Then by 3.9 x satisfies:
1
Dx = −Ω2 − Ω2 x2 |χ̂′ |2 (3.52)
2
Integrating this equation from Cu0 along the given generator yields:
Z u Z u
2 ′ 1 2 2 ′2 ′
x(u) = x(u0 ) − Ω du − Ω x |χ̂ | du (3.53)
u0 u0 2
Let P(t) be the property:

1
P(t) : |x − |u|| ≤ : for all u ∈ [u0 , t]
3
Then by the assumptions of the lemma P(0) is true and it follows by continuity
that P(t) is true for sufficiently small positive t. Let t∗ be the least upper bound
of the set of values of t ∈ [u0 , s∗ ] for which P(t) is true. Then by continuity
P(t∗ ) is true. It follows that:
1 1 1
≤ ≤ : for all u ∈ [u0 , t∗ ] (3.54)
|u| + (1/3) x |u| − (1/3)
or:
(1/3) 1 1 (1/3)
− ≤ − ≤ : for all u ∈ [u0 , t∗ ] (3.55)
|u|(|u| + (1/3)) x |u| |u|(|u| − (1/3))
103
which, since |u| ≥ 1, implies:
1 1 1
− ≤ : for all u ∈ [u0 , t∗ ] (3.56)
x |u| 2|u|2
that is:
1 1 1
trχ′ + ≤ : for all u ∈ [u0 , t∗ ] (3.57)
2 |u| 2|u|2
Also, we have:
2|u| 4|u|
≤ x(u) ≤ : for all u ∈ [u0 , t∗ ] (3.58)
3 3
Moreover, 3.57 and 3.58 together with 3.48 imply:
2 3
Ω2 trχ′ + ≤ : for all u ∈ [u0 , t∗ ] (3.59)
|u| 2|u|2
Now, following an argument similar to that leading from 3.10 to 3.30, we

derive from equation 3.11 the inequality:
D|χ̂′ |2 + 2Ω2 trχ′ |χ̂′ |2 ≤ 2|χ̂′ ||α| (3.60)
This is an ordinary differential inequality along the generator under considera-

tion. Applying Lemma 3.1 with initial point the point of intersection with Cu0 ,
yields: Z u
′
|χ̂′ (u)| ≤ e−F (u) |χ̂′ (u0 )| + eF (u ) |α|(u′ )du′ (3.61)
u0
where: Z u
F (u) = Ω2 trχ′ du′ (3.62)
u0
By virtue of 3.59 we have, recalling that |u| ≥ 1,

Z u Z u
′ 2 ′′ 2
F (u) − F (u ) + ′′
du = (Ω2 trχ′ )(u′′ ) + du′′
u′ |u | u′ |u′′ |
Z u Z u
2 2
≤ (Ω2 trχ′ )(u′′ ) + ′′ du′′ ≤ (Ω2 trχ′ )(u′ ) + du′
u′ |u | u0 |u′ |
Z u
3 3 3
≤ ′ 2
du′ ≤ ≤
u0 2|u | 2|u| 2
that is:
|u′ |2

′ 3
F (u) − F (u ) + log ≤ (3.63)
|u|2 2
In particular, setting u′ = u0 ,
|u0 |2

3
F (u) + log ≤ (3.64)
|u|2 2
104
Inequalities 3.63, 3.64 imply:
|u0 |2 ′ |u′ |2
e−F (u) ≤ e3/2 , e−F (u)+F (u ) ≤ e3/2 (3.65)
|u|2 |u|2
Substituting 3.65, the last of the definitions 3.4 and the definition 3.49, in 3.61,
noting that by 3.44 χ̂′ = χ̂ on Cu0 , we obtain:
δ 1/2

′ 2δ
|χ̂ (u)| ≤ e3/2 D0∞ (χ̂) + R∞
(α) : for all u ∈ [u0 , t∗ ] (3.66)
|u|2 3|u|3/2 0
Let us now evaluate 3.53 at u = t∗ :
Z t∗ t∗
1 2 2 ′2
Z
x(t∗ ) = x(u0 ) − Ω2 du − Ω x |χ̂ | du (3.67)
u0 u0 2
By virtue of 3.48 we have, recalling that t∗ ≤ −1,
Z t∗ Z t∗ Z t∗
2
Ω2 du ≤ 1 + (1/18)|u|−2 du ≤ 1 + (1/6)|u|−2 du

u0 u0 u0
∗ ∗ −1
≤ |u0 | − |t | + (1/6)|t | (3.68)
and:
Z t∗ Z t∗ 2
Z t∗
Ω2 du ≥ 1 − (1/18)|u|−2 1 − (1/9)|u|−2 du

du ≥
u0 u0 u0
≥ |u0 | − |t∗ | − (1/9)|t∗ |−1 (3.69)
Also, by virtue of 3.66, 3.58 and 3.48 we have:

Z t∗ Z t∗ 2
1 2 2 ′2
0≤ Ω x |χ̂ | du ≤ e3 δ D0∞ (χ̂) + (2/3)|u|−3/2 δR∞
0 (α) |u|−2 du
u0 2 u0
Z t∗
3
2 −2 2 ∗ −1
≤e δ D0∞ (χ̂) + δR∞
0 (α) |u| du ≤ e3 δ D0∞ (χ̂) + δR∞ 0 (α) |t |
u0
1 ∗ −1
≤ |t | (3.70)
18
the last step provided that:
2
18e3 D0∞ (χ̂) + R∞
0 (α) δ≤1 (3.71)
In view of 3.68, 3.69, 3.70, we conclude from 3.67 that:

2 1
x(u0 ) − |u0 | − |t∗ |−1 ≤ x(t∗ ) − |t∗ | ≤ x(u0 ) − |u0 | + |t∗ |−1 (3.72)
9 9
According to the assumptions of the lemma
1
|x(u0 ) − |u0 || < (3.73)
9
105
In view of the fact that |t∗ | ≥ 1, it follows that:
1
|x(t∗ ) − |t∗ || < (3.74)
3
Thus the inequality defing property P is not saturated at u = t∗ . Hence by
continuity, property P is true for some t > t∗ contradicting the definition of t∗ ,
unless t∗ = s∗ . This completes the proof of the lemma. Note from 3.57 that we
have shown that
1 1 1
trχ′ + ≤ : for all u ∈ [u0 , s∗ ] (3.75)
2 |u| 2|u|2
3.4 L∞ Estimates for η, η

We proceed to derive L∞ estimates for η, η. These satisfy the coupled system
1.84 and 1.85 of ordinary differential equations along the generators of each Cu
and each C u respectively. This is a linear system for η, η, if we think of Ω, g/,
χ, χ and β, β as given. The initial condition for 1.84 is on C 0 where we have:
η = 0 : along C 0 (3.76)
C 0 being the outer boundary of the Minkowskian region M0 . The initial condi-
tion for 1.85 is on Cu0 where, as we recall from Chapter 2, we have:
η = −η : along Cu0 (3.77)
Lemma 3.3 Under the assumptions of Proposition 3.1 and Lemma 3.2 the
∗
following estimates hold on M1s :
Cδ 1/2 ∞ Cδ 3/2 ∞
|η| ≤ R0 (β) + R (β)
|u|2 |u|3 0
Cδ 1/2 ∞ Cδ
|η| ≤ R (β) + 3 R∞ (β)
|u|2 0 |u| 0
Proof: By 3.5 we have:
D|η|2 = D(η, η) = D((g/−1 )AB ηA ηB )

= −2Ωχ♯♯AB ηA ηB + 2η ♯A (Dη)A
= −2Ωχ(η ♯ , η ♯ ) + 2(η, Dη) (3.78)
Since
|χ(η ♯ , η ♯ )| ≤ |χ||η|2 , |(η, Dη)| ≤ |η||Dη|,
106
and from equation 1.84:
|Dη| ≤ Ω(|χ||η| + |β|),
3.78 implies:
D|η|2 ≤ 2Ω|η|{|χ|(|η| + |η|) + |β|} (3.79)
Thinking of |η| as given, this is an ordinary differential inequality for |η| along
the generators of the Cu . Consider a given generator of a given Cu . We apply
Lemma 3.1 with initial point the point of intersection with C 0 . In view of the
initial condition 3.76 we obtain:
Z u Z u !
|η(u)| ≤ exp (Ω|χ|)(u )du (Ω(|χ||η| + |β|))(u′ )du′
′′ ′′
(3.80)
0 u′
Now, in view of the smallness condition on δ of Proposition 3.1, the estimates

of that proposition yield:
C(1 + R∞ 0 (α))
|χ| ≤ : on M ′ (3.81)
δ 1/2 |u|
Hence, the integral in the exponential in 3.80 is bounded by:
Z u
C(1 + R∞ 0 (α)) C(1 + R∞0 (α))
1/2
du′′ = 1/2
(u − u′ ) ≤ C(1 + R∞
0 (α))δ
1/2
u′ δ |u| δ |u|
The last is not greater than log 2 provided that δ satisfies a smallness condition
of the same form as that required in Proposition 3.1. Substituting then the
bound 3.81 and the second of the definitions 3.4 in 3.80 we obtain, along the
generator under consideration, the inequality:
Z u
C(1 + R∞ 0 (α)) ′ ′ Cδ 1/2 R∞0 (β)
|η(u)| ≤ |η(u )|du + 2
(3.82)
δ 1/2 |u| 0 |u|
Let us define:
y(u, u) = sup |η|, y(u, u) = sup |η| (3.83)
Su,u Su,u
Then 3.82 implies:

u
C(1 + R∞0 (α)) Cδ 1/2 R∞0 (β)
Z
|η(u)| ≤ 1/2
y(u′ , u)du′ + 2
(3.84)
δ |u| 0 |u|
Considering the generator of Cu through each point p ∈ Su,u and taking the
supremum over p ∈ Su,u we then obtain:
Z u
C(1 + R∞0 (α)) Cδ 1/2 R∞0 (β)
y(u, u) ≤ 1/2
y(u′ , u)du′ + 2
(3.85)
δ |u| 0 |u|
By 3.5 we have, in analogy with 3.78,
D|η|2 = −2Ωχ(η ♯ , η ♯ ) + 2(η, Dη) (3.86)
107
We now use the fact that
1 1
χ(η ♯ , η ♯ ) = trχ|η|2 + χ̂(η ♯ , η ♯ ) ≥ trχ|η|2 − |χ̂||η|2
2 2
together with the fact that:
|(η, Dη)| ≤ |η||Dη|
and equation 1.85 to deduce:

1
D|η|2 + 2Ω trχ − |χ̂| |η|2 ≤ 2Ω(|χ||η| + |β|)|η| (3.87)
2
Thinking of |η| as given, this is an ordinary differential inequality for |η| along
the generators of the C u . Consider a given generator of a given C u , u ∈ [0, u1 ].
We apply Lemma 3.1 with initial point the point of intersection with Cu0 to
obtain:
Z u
F̃ (u) −F̃ (u′ ) ′ ′
|η(u)| ≤ e |η(u0 )| + e (Ω(|χ||η| + |β|))(u )du (3.88)
u0
where: u
1
Z
F̃ (u) = Ω − trχ + |χ̂|) du′ (3.89)
u0 2
Let u ≤ s∗ . Then by Lemma 3.2 we have:
Z u Z u
1 1 1
F̃ (u) − F̃ (u′ ) − ′′ |
du ′′
= − (Ωtrχ)(u ′′
) − ′′ |
+ (Ω|χ̂|)(u ′′
) du′′
u′ |u u′ 2 |u
Z u Z u
1 1
≤ ′′
(Ωtrχ)(u ) + ′′ du′′ + (Ω|χ̂|)(u′′ )du′′
u′ 2 |u | u ′
1/2 ∞
+ δR∞
Z u Z u
3 ′
Cδ (D 0 (χ̂) 0 (α))
≤ ′ 2
du + ′ 2
du′
u0 4|u | u0 |u |
Z u
3 3 3
≤ ′ 2
du′ ≤ ≤
u0 2|u | 2|u| 2
the last line by virtue of a smallness condition on δ of the same form as that
required in Lemma 3.2. We have thus obtained:
′
′ |u | 3
F̃ (u) − F̃ (u ) − log ≤ (3.90)
|u| 2
In particular, setting u′ = u0 ,

|u0 | 3
F̃ (u) − log ≤ (3.91)
|u| 2
Also, by Lemma 3.2:
C ∗
|χ| ≤ : on M1s (3.92)
|u|
108
under the smallness condition on δ required in Lemma 3.2. Let us set:
D0∞ (η) = sup(|u0 |2 δ −1/2 |η|) (3.93)
u
Cu01
u
Here Cu01 , is the part of Cu0 corresponding to u ∈ [0, u1 ]. Substituting the
bounds 3.90, 3.91, 3.92, the next to last of the definitions 3.4, the first of defini-
tions 3.83 and definition 3.93 in 3.88 we deduce for u ∈ [u0 , s∗ ] the inequality:
|u | |η(u′ )| δR∞
Z u ′
0 (β)

|u0 |
|η(u)| ≤ C |η(u0 )| + C + du′
|u| u0 |u| |u′ | |u′ |4
Z u
Cδ 1/2 ∞ C Cδ
≤ D0 (η) + y(u, u′ )du′ + 3 R∞ (β) (3.94)
|u||u0 | |u| u0 |u| 0
Considering the generator of C u through each point p ∈ Su,u and taking the
supremum over p ∈ Su,u we then obtain, for each (u, u) ∈ [0, u1 ] × [u0 , s∗ ]:
Z u
Cδ 1/2 ∞ C Cδ
y(u, u) ≤ D (η) + y(u, u′ )du′ + 3 R∞ (β) (3.95)
|u||u0 | 0 |u| u0 |u| 0
This implies, for each (u, u) ∈ [0, u1 ] × [u0 , s∗ ]:
Z u
Cδ 3/2 ∞ Cδ 2 ∞
y(u′ , u)du′ ≤ D0 (η) + R (β)
0 |u||u0 | |u|3 0
Z u Z u
C
+ y(u′ , u′ )du′ du′ (3.96)
|u| u0 0
Hence, setting:
z(u) = sup y(u, u) = sup |η| (3.97)
u
u∈[0,u1 ] Cu 1
we obtain, for each (u, u) ∈ [0, u1 ] × [u0 , s∗ ]:

Z u
Cδ 3/2 ∞ Cδ 2 ∞
y(u′ , u)du′ ≤ D0 (η) + R (β)
0 |u||u0 | |u|3 0
Z u
Cδ
+ z(u′ )du′ (3.98)
|u| u0
Substituting in 3.85 we then obtain, for each (u, u) ∈ [0, u1 ] × [u0 , s∗ ]:
Cδ 1/2 ∞ Cδ 3/2 ∞
y(u, u) ≤ R (β) + (D0 (η) + R∞ 0 (β))
|u|2 0 |u|3
Z u
Cδ 1/2 (1 + R∞0 (α))
+ z(u′ )du′ (3.99)
|u|2 u0
Taking the supremum over u ∈ [0, u1 ] then yields the linear integral inequality:
Cδ 1/2 ∞ Cδ 3/2 ∞
z(u) ≤ R0 (β) + (D0 (η) + R∞
0 (β))
|u|2 |u|3
109
u
Cδ 1/2 (1 + R∞0 (α))
Z
+ z(u′ )du′
|u|2 u0
: ∀u ∈ [u0 , s∗ ] (3.100)
Setting Z u
Z(u) = z(u′ )du′ , we have Z(u0 ) = 0 (3.101)
u0
and 3.100 takes the form:
dZ
≤ aZ + b (3.102)
du
where
Cδ 1/2 (1 + R∞0 (α))
a(u) = 2
(3.103)
|u|
Cδ 1/2 ∞ Cδ 3/2 ∞
b(u) = 2
R0 (β) + (D0 (η) + R∞
0 (β)) (3.104)
|u| |u|3
Integrating 3.102 from u0 we obtain:
Z u Ru
a(u′′ )du′′
Z(u) ≤ e u′ b(u′ )du′ : ∀u ∈ [u0 , s∗ ] (3.105)
u0
We have:
u
Cδ 1/2 (1 + R∞
0 (α))
Z
a(u′′ )du′′ ≤ ≤ log 2 (3.106)
u′ |u|
the last step by virtue of a smallness condition on δ of the same form as that
required in Proposition 3.1. Therefore:
Z u
Z(u) ≤ 2 b(u′ )du′
u0
Cδ 1/2 ∞ Cδ 3/2 ∞
≤ R0 (β) + (D0 (η) + R∞
0 (β))
|u| |u|2
: ∀u ∈ [u0 , s∗ ] (3.107)
and substituting in 3.99 yields:
Cδ 1/2 ∞ Cδ 3/2 ∞
y(u, u) ≤ 2
R0 (β) + (D0 (η) + R∞
0 (β)) : ∀(u, u) ∈ [0, u1 ] × [u0 , s∗ ]
|u| |u|3
(3.108)
taking into account a smallness condition on δ of the same form as that required
in the proof of Proposition 3.1. Also, since
Z u
y(u, u′ )du′ ≤ Z(u),
u0
substituting 3.107 in 3.95 yields:

Cδ 1/2 |u| ∞

Cδ
y(u, u) ≤ D (η) + R0 (β) + 3 (D0∞ (η) + R∞
∞
0 (β)) (3.109)
|u|2 |u0 | 0 |u|
110
To complete the proof of the lemma we must derive an appropriate estimate
for D0∞ (η). To do this we consider 3.79 on Cu0 . Since along Cu0 condition 3.77
holds, 3.79 simplifies on Cu0 to:
D|η|2 ≤ 2Ω|η|{2|χ||η| + |β|} (3.110)
This is an ordinary differential inequality along the generators of Cu0 , to which

Lemma 3.1 applies. Taking into account 3.81 and a smallness condition on δ of
the same form as that required in the proof of Proposition 3.1, we deduce:
D0∞ (η) ≤ CR∞

0 (β) (3.111)
Substituting then 3.111 in 3.108 and 3.109, in view of the definitions 3.83, yields
the lemma.
3.5 L∞ Estimates for ω, ω

∗
We proceed to derive an L∞ estimate for ω on M1s . Using this estimate we
∗
shall derive an L∞ estimate for log Ω in M1s which improves the bound 3.48.
∗
This shall enable us to show that s = u1 so that the previous lemmas actually
hold on M1 and, since (u1 , u1 ) ∈ D′ is arbitrary, the estimates hold on all of M ′ .
We shall then derive an L∞ estimate for ω on M ′ . Finally, using the improved
estimate for log Ω we shall derive an estimate for trχ′ in M ′ which improves the
estimate given by Lemma 3.2.
Lemma 3.4 Under the assumptions of Proposition 3.1 and Lemma 3.2 the
∗
following estimate holds on M1s :
Cδ ∞ Cδ 2 ∞
|ω| ≤ R0 (ρ) + (R (β) + R∞
0 (β))
2
|u|3 |u|4 0
Proof: Equation 1.107 is a linear ordinary differential equation for ω along the
generators of the Cu . Substituting the bounds for η, η of Lemma 3.3 as well as
the third of the definitions 3.4 we obtain:
D|ω| ≤ |Dω| ≤ Ω(2|η|2 + |η|2 + |ρ|)

C ∞ Cδ
≤ R (ρ) + 4 (R∞ ∞
0 (β) + R0 (β))
2
(3.112)
|u|3 0 |u|
u
Integrating this inequality from C 0 along each generator of each Cu 1 , u ∈
[u0 , s∗ ], using the fact that by 3.20:
ω = 0 : along C 0 (3.113)
yields the lemma.
111
Lemma 3.5 Under the assumptions of Proposition 3.1 and Lemma 3.2 there
is a numerical constant C such that if δ satisfies the smallness condition:
C R∞ ∞ ∞ 2

0 (ρ) + (R0 (β) + R0 (β)) δ < 1
then we have s∗ = u1 and the estimate
Cδ ∞ Cδ 2 ∞
log Ω ≤ R (ρ) + (R (β) + R∞
0 (β))
2
|u|2 0 |u|3 0
holds on M1 .
Proof: According to the second of the definitions 1.30 we have:
D log Ω = ω (3.114)
This is a linear ordinary differential equation for log Ω along the generators of
the C u . Integrating this equation from Cu0 along a given generator of a given
C u , u ∈ [0, u1 ], using the initial condition 3.44 we obtain:
Z u
| log Ω(u)| ≤ |ω(u′ )|du′ (3.115)
u0
∗
Substituting the estimate for ω of Lemma 3.4 we conclude that on M1s we have:
Cδ ∞ Cδ 2 ∞
| log Ω| ≤ R0 (ρ) + (R (β) + R∞
0 (β))
2
(3.116)
|u|2 |u|3 0
∗
Since |Ω − 1| ≤ max{Ω, 1}| log Ω|, this implies that on M1s :
Cδ ∞
R0 (ρ) + δ(R∞ ∞ 2

|Ω − 1| ≤ 0 (β) + R0 (β)) (3.117)
|u|2
Thus, if we require δ to satisfy the smallness condition:

1
Cδ R∞ ∞ ∞ 2

0 (ρ) + (R0 (β) + R0 (β)) < (3.118)
18
we have:
1 −2 u
|Ω − 1| < |u| : on Cu 1 , for all u ∈ [u0 , s∗ ] (3.119)
18
hence by continuity 3.48 holds for some s > s∗ contradicting the definition of
s∗ , unless s∗ = u1 . This completes the proof of the lemma.
∗
Since by the above lemma M1s = M1 , the results of Lemmas 3.2, 3.3 and
3.4 hold on M1 . Moreover, since (u1 , u1 ) ∈ D′ is arbitrary we conclude:
Proposition 3.2 Under the assumptions of Proposition 3.1 and Lemmas 3.2
and 3.5, the estimates of Lemmas 3.2 - 3.5 hold on all of M ′ .
112
and 3.4 the following estimate holds on M ′ :
C ∞ Cδ
|ω| ≤ 2
R0 (ρ) + 3 (R∞ ∞
0 (β) + R0 (β))
2
|u| |u|
Proof: Equation 1.108 is a linear ordinary differential equation for ω along the
generators of the C u . Substituting the estimates for η, η of Lemma 3.3 as well
as the third of the definitions 3.4 we obtain:
D|ω| ≤ |Dω| ≤ Ω(2|η|2 + |η|2 + |ρ|)

C ∞ Cδ
≤ R (ρ) + 4 (R∞ ∞
0 (β) + R0 (β))
2
(3.120)
|u|3 0 |u|
Integrating this inequality from Cu0 along each generator of each C u using the
fact that by 3.44:
ω = 0 : along Cu0 (3.121)
yields the proposition.
Now, if the initial data along C0 where trivial we would have:

2
2 2 u δ
trχ = =− 1− +O : on Cu0
u−u |u0 | |u0 | |u0 |2
Motivated by the comparison with the trivial case we define:

∞ 3 −1 1 1 u
D0 (trχ) = sup |u0 | δ trχ + 1− (3.122)
Cu0 2 |u0 | |u0 |
Here, as in 3.49, the supremum is meant to be taken on all of Cu0 , not only
on the part which lies in M ′ . This is the same for c∗ ≥ u0 + δ, but not for
c∗ ∈ (u0 , u0 + δ). In the latter case the part of Cu0 lying in M ′ is the part
∗
Cuc 0 −u0 . By the results of Chapter 2 (see 2.156, 2.163) D0∞ (trχ) is bounded by
a non-negative non-decreasing continuous function of M3 .
and 3.5 the following estimate holds on M ′ :

1 ′ 1 u
≤ Cδ|u|−3 D0∞ (trχ) + (D0∞ (χ̂))2 + R∞

trχ + 1− 0 (ρ)
2 |u| |u|
+δ 2 |u|−3 (1 + D0∞ (trχ))2 + D0∞ (χ̂)R∞

0 (α)
+(R0 (β) + R0 (β) + R0 (α))2

∞ ∞ ∞

This improves the estimate for trχ′ given by Lemma 3.2.
113
Proof: We revisit equation 3.53 which we write in the form, subtracting |u| + u
from both sides,
Z u
x(u) − |u| − u = x(u0 ) − |u0 | − u − Ω2 du′ + |u0 | − |u|
u0
Z u
1 2 2 ′2 ′
− Ω x |χ̂ | du (3.123)
u0 2
Using the result of Lemma 3.5 we now estimate:
Z u Z u Z u
2 ′ 2 ′
Ω du − |u0 | + |u| = (Ω − 1)du ≤ |Ω2 − 1|du′
u0 u0 u0
u
Cδ ∞ Cδ 2 ∞
Z
≤C | log Ω|du′ ≤ R0 (ρ) + (R (β) + R∞
0 (β))
2
(3.124)
u0 |u| |u|2 0
Also, using the results of Lemma 3.2 we estimate, as in 3.70
Z u Z u 2
1 2 2 ′2 ′
0≤ Ω x |χ̂ | du ≤ Cδ D0∞ (χ̂) + |u′ |−3/2 δR∞0 (α) |u′ |−2 du′
u0 2 u0
Cδ ∞ 2
≤ D0 (χ̂) + |u|−3/2 δR∞ 0 (α) (3.125)
|u|
Moreover, setting
2 u
trχ|C =− 1− +ε (3.126)
u0 |u0 | |u0 |
we have, by definition 3.122,
|ε| ≤ δ|u0 |−2 D0∞ (trχ) (3.127)
Since
 

 1 u 
x(u0 ) − |u0 | − u = − 1+ − ε − ε |u0 |
 1− u +ε |u0 | 
|u0 |
2
u
|u0 | − ε |u0 |
= − ε|u0 |
u
1 − |u0 | + ε
2
u
= − ε x(u0 ) − ε|u0 |
|u0 |
we obtain, in view of 3.73,
2
u
|x(u0 ) − |u0 | − u| ≤ |ε| + C −ε |u0 |
|u0 |
≤ δ|u0 |−1 D0∞ (trχ) + Cδ 2 |u0 |−1 (1 + D0∞ (trχ))2
(3.128)
114
Substituting 3.124, 3.125, 3.128 in 3.123 yields:
|x − |u| − u| ≤ Cδ|u|−1 D0∞ (trχ) + (D0∞ (χ̂))2 + R∞

0 (ρ)
+δ 2 |u|−1 (1 + D0∞ (trχ))2 + D0∞ (χ̂)R∞

0 (α)
∞ ∞
+(R0 (β) + R0 (β) + R0 (α))2
∞

(3.129)
Since
1 1 x − |u| − u
trχ′ + =
2 |u| + u (|u| + u)x
and by Lemma 3.2
1 2|u|
x ≥ |u| − ≥ ,
3 3
the proposition follows.
3.6 The smallness requirement on δ

Let us now review the assumptions of Lemma 3.2, in the light of the definition
3.122. Consider first the assumption that trχ < 0 on Cu0 . Now, the condition
(1 + D0∞ (trχ))δ < 1 (3.130)
implies:
1 1
D0∞ (trχ) < |u0 |2 − (3.131)
δ |u0 |
The last implies:

1 1 u 1 δ
trχ + 1− < 1− (3.132)
2 |u0 | |u0 | |u0 | |u0 |
which in turn implies that trχ < 0 on Cu0 . Therefore the assumption that
trχ < 0 on Cu0 may be replaced by the smallness condition 3.130 on δ.
Consider next the assumption that
2 1
+ |u0 | < : on Cu0 (3.133)
trχ 9
Suppose that for some constant ε0 ∈ (0, 1) we have:
1 1 ε0
trχ + < : on Cu0 (3.134)
2 |u0 | |u0 |2
It follows in a straightforward manner that we then also have:
2 ε0
+ |u0 | < : on Cu0 (3.135)
trχ 1 − ε0
115
Thus, setting ε0 = 1/10, 3.133 follows if we assume that:
1 1 1 1
trχ + < : on Cu0 (3.136)
2 |u0 | 10 |u0 |2
On the other hand from the definition 3.122 we obtain:

1 1 δD0∞ (trχ) δ
trχ + ≤ + : on Cu0 (3.137)
2 |u0 | |u0 |3 |u0 |2
and the right hand side of 3.137 is less than (1/10)|u0|−2 if

1
(1 + D0∞ (trχ))δ < (3.138)
10
This being a stronger condition of the form 3.130 it impies both assumptions of
Lemma 3.2.
To summarize, the assumptions of the present chapter are, besides the basic
bootstrap assumption A0 and the finiteness of the quantities 3.4, the following
smallness conditions on δ:
C(1 + R∞ 2
0 (α)) δ ≤ 1 : from Proposition 3.1
C(1 + D0∞ (trχ)δ ≤ 1 : from Lemma 3.2, through 3.138
C(D0∞ (χ̂) + R∞ 2
0 (α)) δ ≤ 1 : from Lemma 3.2
C[R∞
0 (ρ) + (R∞
0 (β) + R∞ 2
0 (β)) ]δ ≤ 1 : from Lemma 3.5 (3.139)
Definining
D0∞ = max{D0∞ (trχ), D0∞ (χ̂)} (3.140)
R∞ ∞ ∞ ∞ ∞ ∞ ∞
0 = max{R0 (α), R0 (β), R0 (ρ), R0 (σ), R0 (β), R0 (α)} (3.141)
the above smallness conditions on delta may be simplified to:
C(1 + R∞ 2
0 ) δ ≤ 1
C(1 + D0∞ )δ ≤ 1
C(D0∞ + R∞ 2
0 ) δ ≤ 1
CR∞ ∞
0 (1 + R0 )]δ ≤ 1 (3.142)
A number n of smallness conditions on δ of the form:
F1 (D0∞ , R∞ ∞ ∞
0 )δ ≤ 1, . . . , Fn (D0 , R0 )δ ≤ 1 (3.143)
where F1 , ..., Fn are non-negative non-decreasing continuous functions defined

on the non-negative quadrant {(x, y) ∈ ℜ2 : x ≥ 0 and y ≥ 0}, may be replaced
by a single smallness condition of the same form:
F (D0∞ , R∞
0 )δ ≤ 1 (3.144)
116
where F is the upper envelope of the functions F1 , ..., Fn :
F (x, y) = max{F1 (x, y), ..., Fn (x, y)} : ∀{(x, y) ∈ ℜ2 : x ≥ 0 and y ≥ 0}

(3.145)
also a non-negative non-decreasing continuous function on the non-negative
quadarant. In conclusion, we may say that the results of the present chap-
ter hold if δ is suitably small depending on D0∞ and R∞ 0 , and the last phrase
means that there is a non-negative non-decreasing continous function F on the
non-negative quadarant such that 3.144 holds.
More generally the phrase if δ is suitably small depending on the non-negative
quantities q1 , ..., qn shall mean throughout this monograph the following: if there
is a non-negative non-decreasing continous function F on [0, ∞)n such that:
δF (q1 , ..., qn ) ≤ 1 (3.146)
Note that we call a function F defined on [0, ∞)n non-decreasing if yi ≥ xi

for all i = 1, ..., n implies F (y1 , ..., yn ) ≥ F (x1 , ..., xn ). Let then G1 , ..., Gn
be non-negative non-decreasing continous functions on [0, δ)m . These define a
continuous mapping:
G : [0, ∞)m → [0, ∞)n by: (r1 , ...rm ) 7→ (G1 (r1 , ..., rm ), ...Gn (r1 , ..., rm ))
(3.147)
The composition F ◦G is then a non-negative non-decreasing continuous function
on [0, ∞)m . In particular, taking
F (x1 , ..., xn ) = x1 + ... + xn , or

F( x1 , ..., xn ) = x1 ...xn , or
F( x1 , ..., xn ) = max{x1 , ..., xn } (3.148)
we conclude that the sum, product and upper envelope of non-negative non-
decreasing continous functions on [0, ∞)m is itself a non-negative non-decreasing
continous function on [0, ∞)m . Moreover, if the non-negative quantities q1 , ..., qn
satisfy bounds in terms of the non-negative quantities r1 , ..., rm of the form:
q1 ≤ G1 (r1 , ..., rm ), . . . , qn ≤ Gn (r1 , ..., rm ) (3.149)
and δ satisfies the smallness condition:
δ(F ◦ G)(r1 , ..., rm ) ≤ 1 (3.150)
then δ satisfies the smallness condition 3.146. We may therefore say that δ is
suitably small depending on q1 , ..., qn , provided that δ is suitably small depend-
ing on r1 , ..., rm .
117
Chapter 4
L4(S) Estimates for the 1st

Derivatives of the
4.1 Introduction
The objective of the present chapter is to derive L4 (S) estimates for the 1st
derivatives of the connection coefficients in M ′ on the basis of L4 (S) bounds
for the 1st derivatives of the curvature components in M ′ . In this and the
next three chapters the fundamental bootstrap assumption A0 of Chapter 3 is
understood to hold.
The arguments of the present chapter, like those of Chapter 3, rely only on
the propagation equations, equations 1.41, 1.57, 1.62, 1.82, 1.83, 1.107, 1.108
of Chapter 1. What is assumed in regard to the curvature components in the
present chapter is, besides the assumptions of Chapter 3 (see 3.4), that the
following quantities are finite:

R/41 (α) = sup |u|3/2 δ 3/2 k∇
/αkL4 (Su,u )
(u,u)∈D′

/41 (β) =
R sup |u|5/2 δ 1/2 k∇
/βkL4 (Su,u )
(u,u)∈D′

/41 (ρ) =
R sup |u|7/2 kd
/ρkL4 (Su,u )
(u,u)∈D′

/41 (σ) =
R sup |u|7/2 kd
/σkL4 (Su,u )
(u,u)∈D′

/41 (β) =
R sup |u|9/2 δ −1 k∇
/βkL4 (Su,u )
(u,u)∈D′

/41 (α) =
R sup |u|5 δ −3/2 k∇
/αkL4 (Su,u ) (4.1)
(u,u)∈D′
118
and:

R40 (Dρ) = sup |u|5/2 δkDρkL4 (Su,u )
(u,u)∈D′

R40 (Dσ) = sup |u|5/2 δkDσkL4 (Su,u ) (4.2)
(u,u)∈D′

replacing the supremum on D′ by the supremum on ([0, δ] × {u0 }) D′ , are
T
all bounded by a non-negative non-decreasing continuous function of M6 , the
quantity requiring Mk with the highest k being the one corresponding to ∇ /α.
We shall first derive an L4 (S) estimate for ∇/χ′ , then an L4 (S) estimate
for ∇/χ′ . After that we shall derive L4 (S) estimates for ∇/η, ∇
/η together, as
the propagation equations 1.82, 1.83 are coupled. After that we shall derive
L4 (S) estimates for /dω and /dω. Finally, we shall derive L4 (S) estimates for
Dω and Dω. The L4 (S) estimates for ∇ /η, ∇
/η imply through equations 1.170,
1.171 L4 (S) estimates for Dχ, Dχ, while the L4 (S) estimates for /dω, /dω imply
through equations 1.173, 1.174 L4 (S) estimates for Dη, Dη. In fact, taking into
account 3.7 and 1.123 the traces of equations 1.170, 1.171 are the equations:
D(Ωtrχ) = Ω2 2div / η + 2|η|2 − (χ, χ) + 2ρ

(4.3)
D(Ωtrχ) = Ω2 2div / η + 2|η|2 − (χ, χ) + 2ρ

(4.4)
while, in view of the fact that
ˆ − Ω2 trχχ̂,
D̂(Ωχ̂) = D(Ωχ) ˆ − Ω2 trχχ̂,
D̂(Ωχ̂) = D(Ωχ)
taking into account the identity 1.187 (see also 1.52), the trace-free parts of
equations 1.170, 1.171 are seen to reduce to the equations:

1 1
D̂(Ωχ̂) = Ω2 ∇ ˆ + η ⊗η
/⊗η ˆ + trχχ̂ − trχχ̂ (4.5)
2 2

1 1
D̂(Ωχ̂) = Ω2 ∇ ˆ + η ⊗η
/⊗η ˆ + trχχ̂ − trχχ̂ (4.6)
2 2
We begin with seven lemmas.
Lemma 4.1 Let θ be an arbitrary p-covariant S tensorfield on M ′ . We have,

with respect to an arbitrary local frame field for the Su,u ,
p
X
(D∇ /Dθ)AB1 ...Bp = −
/θ − ∇ /)C
(DΓ ABi θ C
B1 ...>Bi <...Bp
i=1
where DΓ / is the Lie derivative with respect to L of the induced connection on

the surfaces Su,u , a T21 -type S tensorfield, symmetric in the lower indices and
given by:
/)D
g/CD (DΓ AB = ∇
/A (Ωχ)BC + ∇
/B (Ωχ)AC − ∇
/C (Ωχ)AB
119
Similarly,
p
X
(D∇ /Dθ)AB1 ...Bp = −
/θ − ∇ /)C
(DΓ ABi θ C
B1 ...>Bi <...Bp
i=1
where DΓ / is the Lie derivative with respect to L of the induced connection on

the surfaces Su,u , a T21 -type S tensorfield, symmetric in the lower indices and
given by:
/)D
g/CD (DΓ AB = ∇
/A (Ωχ)BC + ∇
/B (Ωχ)AC − ∇
/C (Ωχ)AB
Here
C
> Bi <
signifies that the index Bi is missing and in its place we have the index C.
Proof: To derive the formula for D∇ /θ − ∇/Dθ we consider a given outgoing null
hypersurface Cu and work in a Jacobi field frame defined in a neighborhood V
of a generator γp of Cu , corresponding to a point p ∈ S0,u , of the form:
[
V= γq (4.7)
q∈U
where U is a neighborhood of the point p in S0,u and γq denotes the generator

of Cu corresponding to the point q ∈ S0,u . A Jacobi field frame (eA : A = 1, 2)
is characterized by the conditions:
[L, eA ] = 0 : A = 1, 2 (4.8)
It then follows by the Jacobi identity
[L, [eA , eB ]] + [eA , [eB , L]] + [eB , [L, eA ]] = 0
that also:
[L, [eA , eB ]] = 0 (4.9)
We have:
/θ)(eA , eB1 , ..., eBp ))
/θ)(eA , eB1 , ..., eBp ) = L((∇
(D∇ (4.10)
Also,
/θ)(eA , eB1 , ..., eBp ) =

(∇ /eA θ)(eB1 , ..., eBp )
(∇
X ∇
/eA eBi
= eA (θ(eB1 , ...eBp )) − θ(eB1 , ... > eBi < ..., eBp )
i
where
∇
/eA eBi
> eBi <
120
/eA eBi . Since
signifies that the ith entry eBi is replaced by ∇
/eA eB = Γ
∇ /C
AB eC (4.11)
where Γ/CAB are the coefficients of the induced connection on the surfaces Su,u
in the frame field (eA : A = 1, 2), we obtain:
p
X
/θ)(eA , eB1 , ..., eBp ) = eA (θB1 ...Bp ) −
(∇ /C
ΓABi θ C (4.12)
B1 ...>Bi <...Bp
i=1
Now, by 4.8,
L(eA (θB1 ...Bp )) = eA (L(θB1 ...Bp )) = eA ((Dθ)B1 ...Bp ) (4.13)
and, denoting
/C
L(Γ /)C
AB ) = (DΓ AB (4.14)
we have:
/C
L(ΓABi θ C ) = /C
(L(ΓABi ))θ C /C
+ΓABi L(θ C )
B1 ...>Bi <...Bp B1 ...>Bi <...Bp B1 ...>Bi <...Bp
= /)C
(DΓ ABi θ C /C
+ΓABi (Dθ) C
B1 ...>Bi <...Bp B1 ...>Bi <...Bp
(4.15)
Substituting 4.12 on the right hand side in 4.10 and taking into account 4.13
and 4.15, the right hand side of 4.10 becomes:
p
X
/Dθ)(eA , eB1 , ..., eBp ) −
(∇ /)C
(DΓ ABi θ C
B1 ...>Bi <...Bp
i=1
We conclude that:
p
X
(D∇ /Dθ)AB1 ...Bp = −
/θ − ∇ /)C
(DΓ ABi θ C (4.16)
B1 ...>Bi <...Bp
i=1
To obtain the formula for DΓ

/ we consider that:
eA (g/BC ) = eA (g/(eB , eC )) = g/(∇ /eA eC )

/eA eB , eC ) + g/(eB , ∇
/D
= g/(ΓAB eD , eC ) + g /D
/(eB , ΓAC eD ) = g /D
/DC ΓAB + g /D
/BD ΓAC
hence:
/D
L(eA (g/BC )) = (L(g/DC ))ΓAB + (L(g /D
/BD ))ΓAC + g /)D
/DC (DΓ AB + g /)D
/BD (DΓ AC
(4.17)
Now, according to 1.41:
L(g/AB ) = (Dg/)AB = 2ΩχAB (4.18)
121
hence by 4.8:
L(eA (g/BC )) = eA (L(g/BC )) = eA (2ΩχBC ) (4.19)
Thus 4.17 reads:
/D
eA (2ΩχBC ) = 2ΩχDC Γ /D
AB + 2ΩχBD ΓAC + g /)D
/DC (DΓ AB + g /)D
/BD (DΓ AC
which, since
/D
eA (ΩχAB ) − Γ /D
AB ΩχDC − ΓAC ΩχBD = ∇
/A (Ωχ)BC ,
is equivalent to:
1 1
∇
/A (Ωχ)BC = /)D
g/DC (DΓ AB + g /)D
/BD (DΓ AC (4.20)
2 2
Denoting this formula by {ABC}, we form the sum {ABC} + {BAC} − {CAB}
to obtain:
1
∇
/A (Ωχ)BC + ∇
/B (Ωχ)AC − ∇
/C (Ωχ)AB = /)D
g/CD ((DΓ /)D
AB + (DΓ BA )
2
1 1
/)D
+ g/AD ((DΓ BC /)D
− (DΓ CB ) + g /)D
/BD ((DΓ /)D
AC − (DΓ CA )
2 2
(4.21)
Now, we have:
/eB eA = (Γ
/eA eB − ∇
[eA , eB ] = ∇ /C /C
AB − ΓBA )eC (4.22)
Also, for arbitrary functions f A : A = 1, 2 we have, by 4.8,
[L, f A eA ] = (Lf A )eA (4.23)
It follows that:
/)C
((DΓ /)C
AB − (DΓ BA )eC = [L, [eA , eB ]] = 0
by virtue of 4.9. Thus:

/)C
(DΓ /)C
AB − (DΓ BA = 0 (4.24)
Taking this into account, 4.21 reduces to:
∇
/A (Ωχ)BC + ∇
/B (Ωχ)AC − ∇ /)D
/C (Ωχ)AB = g/CD (DΓ AB (4.25)
We have thus established the first part of the lemma. The second part then
follows by conjugation.
In the following we shall use a simplified notation when writing down com-
ponents of S tensorfields relative to an arbitrary frame field for the Su,u . That is
we shall use the components of g/−1 to raise indices. Thus, if θ is a 2-covariant S
122
tensorfield we denote the components of the corresponding T11 -type S tensorfield
θ♯ defined by 1.47 simply by θAB :
θAB = θAC (g/−1 )CB (4.26)
and the components of the corresponding 2-contravariant S tensorfield θ♯♯ sim-

ply by θAB :
θAB = (g/−1 )AC (g/−1 )BD θCD (4.27)
More generally, if θ is a p-covariant S tensorfield and {i1 , ..., ik } is an ordered
subset of {1, ..., p}, we denote
Bi . . . Bi
θA1 ...>Ai1 <...>Aik <...Ap = (g/−1 )Ai1 Bi1 ...(g/−1 )Aik Bik θA1 ...Ai1 ...Aik ...Ap (4.28)
1 k
Note that according to this notation we have g/AB = (g/−1 )AB . However to
avoid confusion we shall keep the notation (g/−1 )AB for the components of the
reciprocal metric.
Let θ, θ′ be a pair of arbitrary type Tpq S tensorfields. Their inner product
is defined by:
A ...A ′B ...B
(θ, θ′ ) = g/A1 B1 ...g/Aq Bq (g/−1 )C1 D1 ...(g/−1 )Cp Dp θC11...Cpq θD11...Dpq (4.29)
The magnitude of a Tpq S tensorfield θ is accordingly defined by:
|θ| = (θ, θ)1/2 (4.30)
Lemma 4.2 Let θ be an arbitrary p covariant S tensorfield on M ′ . We have:

p
X Bi
2 2
D(|θ| ) + pΩtrχ|θ| = 2(θ, Dθ) − 2 Ωχ̂A i A1 ...>Ai <...Ap
Bi θ θA1 ...Ap
i=1
and:
p
X Bi
D(|θ|2 ) + pΩtrχ|θ|2 = 2(θ, Dθ) − 2 Ωχ̂A

B
i A1 ...>Ai <...Ap
θ θA1 ...Ap
i
i=1
Proof: We have:
|θ|2 = (g/−1 )A1 B1 ...(g/−1 )Ap Bp θA1 ...Ap θB1 ...Bp
Thus, in view of the first of 2.3, which reads
(Dg/−1 )AB = −2ΩχAB ,
123
and the fact that the operator D satisfies the Leibniz rule, we obtain:
D(|θ|2 ) = (g/−1 )A1 B1 ...(g/−1 )Ap Bp {(DθA1 ...Ap )θB1 ...Bp + θA1 ...Ap (DθB1 ...Bp )}
p
X (−2ΩχAi Bi )
−1 A1 B1 −1 Ai Bi
+ (g/ ) ... > (g/ ) < ...(g/−1 )Ap Bp θA1 ...Ap θB1 ...Bp
i=1
p
A1 ...>Ai <...Ap
X
A1 ...Ap
= 2θ (Dθ)A1 ...Ap − 2 ΩχAi Bi θ Bi θA1 ...Ap (4.31)
i=1
Substituting the decomposition

1
χAi Bi = χ̂Ai Bi + trχ(g/−1 )Ai Bi
2
yields the first part of the lemma. The second part then follows by conjugation.
Let θ be an arbitrary p covariant S tensorfield on M ′ . Consider a given Cu .

The flow Φt generated by L on Cu (see Chapter 1) defines a diffeomorphism Φu
of S0,u onto Su,u . Then
θ(u) = Φ∗u θ|Su,u (4.32)
is a 1-parameter family of p covariant S tensorfields on S0,u and we have:
∂
θ(u) = θ̇(u) (4.33)
∂u
where
θ̇(u) = Φ∗u Dθ|Su,u (4.34)
Similarly, consider a given C u . The flow Φt generated by L on C u (see Chapter
1) defines a diffeomorphism Φu−u0 of Su,u0 onto Su,u . Then
θ(u) = Φ∗u−u0 θ|Su,u (4.35)
is a 1-parameter family of p covariant S tensorfields on Su,u0 and we have:
∂
θ(u) = θ̇(u) (4.36)
∂u
where
θ̇(u) = Φ∗u−u0 Dθ|Su,u (4.37)
Lemma 4.3 Consider a given Cu and let ρ be a non-negative function on Cu

and ρ(u) = Φ∗u ρ|Su,u = ρ|Su,u ◦ Φu be the corresponding 1-parameter family of
functions on S0,u . Then if δ is suitably small depending on R∞
0 (see 3.141) we
have, for each p ≥ 1:
2−1/p kρ(u)kLp (S0,u ) ≤ kρkLp(Su,u ) ≤ 21/p kρ(u)kLp (S0,u )
124
In particular, taking ρ = 1 we have, since Area(S0,u ) = 4π|u|2 ,
2π|u|2 ≤ Area(Su,u ) ≤ 8π|u|2
Also, consider a given C u and let ρ be a non-negative function on C u and

ρ(u) = Φ∗u−u0 ρ|Su,u = ρ|Su,u ◦ Φu−u0 be the corresponding 1-parameter family
of functions on Su,u0 . Then if δ is suitably small depending on D0∞ , R∞
0 (see
3.140, 3.141) we have, for each p ≥ 1:
1/p 1/p
|u|2 |u|2

−3/2p 3/2p
e kρ(u)kLp (Su,u0 ) ≤ kρkLp(Su,u ) ≤ e kρ(u)kLp(Su,u0 )
|u0 |2 |u0 |2
Proof: Consider first a non-negative function ρ on Cu . We have:

Z !1/p
p
kρkLp(Su,u ) = ρ dµ/g (4.38)
Su,u
and: !1/p
Z
p
kρ(u)kLp (S0,u ) = (ρ(u)) dµ/g (4.39)
S0,u
Now, Φu being a diffeomorphism of S0,u onto Su,u , we have:

Z Z
ρp dµ/g = (Φ∗u ρ)p Φ∗u dµ/g (4.40)
Su,u S0,u
Also,
Φ∗u dµ/g Su,u
= dµΦ∗u /g|S
u,u
and:
∂ ∗
Φ dµ/g = Φ∗u (Ddµ/g ) = Φ∗u (Ωtrχdµ/g )
∂u u Su,u Su,u Su,u
= ( (Ωtrχ)|Su,u ◦ Φu )Φ∗u dµ/g Su,u

(4.41)
Evaluating this at any given point q ∈ S0,u and integrating with respect to u,
noting that
Φ∗0 dµ/g S = dµ/g S
0,u 0,u
Φ0 being the identity mapping on S0,u , we obtain:

Z u
∗ ′
(Φu dµ/g S )(q) = exp (Ωtrχ)(Φu′ (q))du dµ/g S0,u
(q) : ∀q ∈ S0,u
u,u
0
(4.42)
Here
u′ 7→ Φu′ (q)
125
is the generator of Cu through q. Thus, the integral in the exponential is the
function F (u) defined by 3.32. If δ is suitably small depending on R∞
0 we have
(see 3.34):
|F (u)| ≤ log 2 (4.43)
It follows that:
2−1 dµ/g S0,u
≤ Φ∗u dµ/g Su,u
≤ 2 dµ/g S0,u
(4.44)
In view of 4.38 - 4.40 the first part of the lemma then follows.
Consider next a non-negative function ρ on C u . We have:
Z !1/p
p
kρkLp(Su,u ) = ρ dµ/g (4.45)
Su,u
and: !1/p
Z
p
kρ(u)kLp (Su,u0 ) = (ρ(u)) dµ/g (4.46)
Su,u0
Now, Φu−u0 being a diffeomorphism of Su,u0 onto Su,u , we have:

Z Z
ρp dµ/g = (Φ∗u−u0 ρ)p Φ∗u−u0 dµ/g (4.47)
Su,u Su,u0
Also,
Φ∗u−u0 dµ/g Su,u
= dµΦ∗u−u /
g|Su,u
0
and:
∂ ∗
Φ dµ/g Su,u = Φ∗u−u0 (Ddµ/g ) Su,u = Φ∗u−u0 (Ωtrχdµ/g )
∂u u−u0 Su,u
= ( (Ωtrχ) S ◦ Φu−u0 )Φ∗u−u0 dµ/g S (4.48)

u,u u,u
Evaluating this at any given point q ∈ Su,u0 and integrating with respect to u,
noting that
Φ∗0 dµ/g Su,u = dµ/g Su,u
0 0
Φ0 being the identity mapping on Su,u0 , we obtain:

Z u
(Φ∗u−u0 dµ/g S )(q) = exp (Ωtrχ)(Φu′ −u0 (q))du′ dµ/g Su,u0
(q) : ∀q ∈ Su,u0
u,u
u0
(4.49)
Here
u′ 7→ Φu′ −u0 (q)
is the generator of C u through q. Thus, the integral in the exponential is the
function F (u) defined by 3.62. If δ is suitably small depending on D0∞ , R∞0 we
have (see 3.64):
|u0 |2

3
F (u) + log ≤ (4.50)
|u|2 2
126
It follows that:
|u|2 |u|2
e−3/2 dµ/g ≤ Φ∗u−u0 dµ/g ≤ e3/2 dµ/g (4.51)
|u0 |2 Su,u0 Su,u |u0 |2 Su,u0
In view of 4.45 - 4.47 the second part of the lemma then follows.
Lemma 4.4 Let ψ be a non-negative function vanishing on C 0 such that ψ 2

is weakly differentiable along the generators of Cu and satisfies the inequality:
Dψ 2 + λΩtrχψ 2 ≤ 2lΩ|χ̂|ψ 2 + 2ψρ
where λ is a real number, ρ is a non-negative function and l a non-negative

constant. Then, if δ is suitably small depending on R∞ 0 , for each p ≥ 1 there is
a positive constant C depending only on p, λ, l such that:
Z u
kψkLp(Su,u ) ≤ C kρkLp(Su′ ,u ) du′
0
Proof: Consider the generator of Cu through an arbitrary point q ∈ S0,u :
u′ 7→ Φu′ (q)
We apply Lemma 3.1 to obtain:

!
Z u Z u
ψ(Φu (q)) ≤ exp fλ,l (Φu′′ (q))du′′ ρ(Φu′ (q))du′ : ∀q ∈ S0,u (4.52)
0 u′
where:
λ
fλ,l = − Ωtrχ + lΩ|χ̂| (4.53)
2
With the notations of Lemma 4.3, 4.53 reads:
!
Z Z u u
ψ(u) ≤ exp fλ,l (u′′ )du′′ ρ(u′ )du′ (4.54)
0 u′
an inequality for a 1-parameter family of non-negative functions on S0,u .

Now, by Proposition 3.1:
fλ,l ≤ C|u|−1 {lδ −1/2 R∞ ∞ 2

0 (α) + |λ|[1 + (R0 (α)) ]} (4.55)
hence:
Z u
fλ,l (u′′ )du′′ ≤ Cδ 1/2 |u|−1 {lR∞
0 (α)+|λ|δ
1/2
[1+(R∞ 2
0 (α)) ]} ≤ l+|λ| (4.56)
u′
the last step provided that δ is suitably small depending on R∞

0 (α). We conclude
that: Z u
ψ(u) ≤ C ρ(u′ )du′ (4.57)
0
127
It follows that:
Z u
kψ(u)kLp (S0,u ) ≤ C kρ(u′ )kLp (S0,u ) du′ (4.58)
0
Now by the first part of Lemma 4.3:
kψkLp(Su,u ) ≤ 21/p kψ(u)kLp (S0,u )
and:
kρ(u′ )kLp (S0,u ) ≤ 21/p kρkLp(Su′ ,u )
Therefore 4.58 implies:
Z u
kψkLp(Su,u ) ≤ C kρkLp(Su′ ,u ) du′ (4.59)
0
for a new constant C, depending on p, and the lemma is proved.
Lemma 4.5 Let ψ be a non-negative function such that ψ 2 is weakly differ-

entiable along the generators of C u and satisfies the inequality:
Dψ 2 + λΩtrχψ 2 ≤ 2lΩ|χ̂|ψ 2 + 2ψρ
where λ is a real number, ρ is a non-negative function and l a non-negative

constant. Then, if δ is suitably small depending on D0∞ , R∞ 0 , for each p ≥ 1
there is a positive constant C depending only on p, λ, l such that:
|u|λ−(2/p) kψkLp (Su,u ) ≤ C|u0 |λ−(2/p) kψkLp (Su,u0 )

Z u
+C |u′ |λ−(2/p) kρkLp (Su,u′ ) du′
u0
Proof: Consider the generator of C u through an arbitrary point q ∈ Su,u0 :
u′ 7→ Φu′ −u0 (q)
We apply Lemma 3.1 to obtain:

Z u
′
ψ(Φu−u0 (q)) ≤ exp f λ,l (Φu′ −u0 (q))du ψ(q) (4.60)
u0
Z u Z u
+ exp f λ,l (Φu′′ −u0 (q))du′′ ρ(Φu′ −u0 (q))du′
u0 u′
: ∀q ∈ Su,u0
where:
λ
f λ,l = − Ωtrχ + lΩ|χ̂| (4.61)
2
128
With the notations of Lemma 4.3, 4.60 reads:
Z u
′ ′
ψ(u) ≤ exp f λ,l (u )du ψ(u0 ) (4.62)
u
Z u 0 Z u
+ exp f λ,l (u )du ρ(u′ )du′
′′ ′′
u0 u′
an inequality for a 1-parameter family of non-negative functions on Su,u0 .

Now by Lemmas 3.2 and 3.5 and Proposition 3.4:

λ u
f λ,l − 1− ≤ |λ|δ|u|−3 F1 + lδ 1/2 |u|−2 F2 (4.63)
|u| |u|
Here F1 and F2 are non-negative non-decreasing continuous functions of the
quantities D0∞ , R∞ ∞
0 . It follows that if δ is suitably small depending on D0 and
∞
R0 it holds:
λ |λ| + l
f λ,l − ≤ (4.64)
|u| |u|2
Hence: λ
u
|u′ |
Z
′′ ′′
f λ,l (u )du ≤ C + log (4.65)
u′ |u|
In particular,
u λ
|u0 |
Z
f λ,l (u′ )du′ ≤ C + log (4.66)
u0 |u|
Substituting 4.65, 4.66 in 4.62 we conclude that:
Z u
|u|λ ψ(u) ≤ C|u0 |λ ψ(u0 ) + C |u′ |λ ρ(u′ )du′ (4.67)
u0
It follows that:
Z u
λ λ
|u| kψ(u)kLp (Su,u0 ) ≤ C|u0 | kψ(u0 )kLp (Su,u0 ) + C |u′ |λ kρ(u′ )kLp (Su,u0 ) du′
u0
(4.68)
Now by the second part of Lemma 4.3:
2/p
|u|
kψkLp (Su,u ) ≤ e3/2p kψ(u)kLp (Su,u0 )
|u0 |
and: 2/p
|u0 |
kρ(u′ )kLp (Su,u0 ) ≤ e3/2p kρkLp (Su,u′ )
|u′ |
|u|λ−(2/p) kψkLp (Su,u ) ≤ C|u0 |λ−(2/p) kψkLp (Su,u0 )

Z u
+C |u′ |λ−(2/p) kρkLp (Su,u′ ) du′ (4.69)
0
129
for a new constant C, depending on p, and the lemma is proved.
Lemma 4.6 Let θ be a r covariant S tensorfield on M ′ vanishing on C 0 and

satisfying along the generators of Cu the propagation equation:
ν
Dθ = Ωtrχθ + γ · θ + ξ
2
where ν is a real number, ξ is a r covariant S tensorfield on M ′ , and γ is a type
Trr S tensorfield on M ′ satisfying, pointwise,
|γ| ≤ mΩ|χ̂|
where m is a positive constant. Then, if δ is suitably small depending on R∞ 0 ,

for each p ≥ 1 there is a positive constant C depending only on p, r, ν, m such
that: Z u
kθkLp (Su,u ) ≤ C kξkLp (Su′ ,u ) du′
0
Proof: By virtue of the first part of Lemma 4.2 we have:
D|θ|2 + rΩtrχ|θ|2 ≤ 2(θ, Dθ) + 2rΩ|χ̂||θ|2 (4.70)
By the propagation equation,

ν
(θ, Dθ) = Ωtrχ|θ|2 + (θ, γ · θ) + (θ, ξ)
2
ν
≤ Ωtrχ|θ|2 + mΩ|χ̂||θ|2 + |θ||ξ| (4.71)
2
Substituting 4.71 in 4.70 we obtain the inequality:
D|θ|2 + (r − ν)Ωtrχ|θ|2 ≤ 2(m + r)Ω|χ̂||θ|2 + 2|θ||ξ| (4.72)
Setting then:
ψ = |θ|, ρ = |ξ|, λ = r − ν, l =m+r
and applying Lemma 4.4, yields the lemma.
Lemma 4.7 Let θ be a r covariant S tensorfield on M ′ satisfying along the

generators of C u the propagation equation:
ν
2
where ν is a real number, ξ is a r covariant S tensorfield on M ′ , and γ is a type
Trr S tensorfield on M ′ satisfying, pointwise,
|γ| ≤ mΩ|χ̂|
130
where m is a positive constant. Then, if δ is suitably small depending on D0∞ ,
R∞0 , for each p ≥ 1 there is a positive constant C depending only on p, r, ν, m
such that:
|u|r−ν−(2/p) kθkLp (Su,u ) ≤ C|u0 |r−ν−(2/p) kθkLp (Su,u0 )
Z u
+C |u′ |r−ν−(2/p) kξkLp (Su,u′ ) du′
u0
Proof: By virtue of the second part of Lemma 4.2 we have:

D|θ|2 + rΩtrχ|θ|2 ≤ 2(θ, Dθ) + 2rΩ|χ̂||θ|2 (4.73)
By the propagation equation,
ν
(θ, Dθ) = Ωtrχ|θ|2 + (θ, γ · θ) + (θ, ξ)
2
ν
≤ Ωtrχ|θ|2 + mΩ|χ̂||θ|2 + |θ||ξ| (4.74)
2
Substituting 4.74 in 4.73 we obtain the inequality:
D|θ|2 + (r − ν)Ωtrχ|θ|2 ≤ 2(m + r)Ω|χ̂||θ|2 + 2|θ||ξ| (4.75)
Setting then:
ψ = |θ|, ρ = |ξ|, λ = r − ν, l =m+r
and applying Lemma 4.5, yields the lemma.
According to the results of Propositions 3.1 - 3.4 we have, on M ′ :

δ −1 |u|2 log Ω ≤ CR∞0 (ρ) + O(δ|u|
−1
)
1 1
|u|2 trχ − ≤ C(R∞ 2
0 (α)) + O(δ)
2 |u|
δ 1/2 |u||χ̂| ≤ CR∞
0 (α)

1 1 u
δ −1 |u|3 trχ + 1− ≤ C{D0∞ (trχ) + (D0∞ (χ̂))2 + R∞
0 (ρ)} + O(δ)
2 |u| |u|
δ −1/2 |u|2 |χ̂| ≤ CD0∞ (χ̂) + O(δ|u|−3/2 )
δ −1/2 |u|2 |η| ≤ CR∞
0 (β) + O(δ|u|
−1
)
−1/2
δ 2
|u| |η| ≤ CR0 (β) + O(δ |u|−1 )
∞ 1/2
(4.76)
Here we denote by O(δ p |u|r ), for real numbers p, r, the product of δ p |u|r with
a non-negative non-decreasing continuous function of the quantities D0∞ , R∞ 0 .
4.2 L4(S) estimates for ∇

/χ
Proposition 4.1 We have:
kd
/trχkL4 (Su,u ) ≤ C|u|−5/2 R∞ /41 (α) + O(δ 1/2 |u|−5/2 )
0 (α)R
131
k∇
/χ̂kL4 (Su,u ) ≤ Cδ −1/2 |u|−3/2 R
/41 (α) + O(|u|−5/2 )
: ∀(u, u) ∈ D′
provided that δ is suitably small depending on D0∞ , R∞0 .

Proof: We consider equations 3.8 and 3.10. In view of the first of 3.5, equation
3.10 is equivalent to:
Dχ̂′ = −α + Ω2 |χ̂′ |2 g/ (4.77)
We apply Lemma 1.2 to equation 3.8 and Lemma 4.1 to equation 4.77 to obtain
the following linear system of propagation equations for /dtrχ′ , ∇
/χ̂′ along each
generator of each Cu :
/trχ′
Dd = f · /dtrχ′ + g · ∇
/χ̂′ + r
D∇/χ̂′ ′
= h · /dtrχ + i · ∇/χ̂′ + s (4.78)
Here f, g, r, h, i, s are respectively T11 , T13 , T10 , T31 , T33 , T30 type S tensorfields, the
components of which with respect to an arbitrary local frame field for the Su,u
are given by:
fAB = −Ω2 trχ′ δA

B
BCD
gA = −2Ω2 χ̂′CD δA B

1
rA = −2Ωd/A Ω (trχ′ )2 + |χ̂′ |2
2
1
hD 2 ′ D ′ D ′
ABC = − Ω (2χ̂BC δA + χ̂AC δB + χ̂AB δC − g
D
/AB χ̂′D
C −g /AC χ̂′D
B )
2
iDEF 2 D
ABC = 2Ω δA g /BC χ̂′EF
1 D E ′F
− Ω2 δ A (δB χ̂C + δBF ′E
χ̂C + δC E ′F
χ̂B + δC F ′E
χ̂B )
2
E D ′F D ′F F D ′E D ′E
+δA (δB χ̂C + δC χ̂B ) + δA (δB χ̂C + δC χ̂B )
′D E F F E ′D E F F E

−χ̂B (δA δC + δA δC ) − χ̂C (δA δB + δA δB )
/A αBC + 2Ω|χ̂′ |2/dA Ωg/BC
sABC = −∇
−2Ω(χ̂′D
B /dA Ω + χ̂′D
A /dB Ω − χ̂′AB /dD Ω)χ̂′DC
−2Ω(χ̂′D
C /dA Ω + χ̂′D dC Ω − χ̂′AC /dD Ω)χ̂′DB
A /
−Ωtrχ′ [2χ̂′BC /dA Ω + χ̂′AB /dC Ω + χ̂′AC /dB Ω
−(g/AB χ̂′DC + g/AC χ̂′DB )d
/D Ω

(4.79)
Thus 4.78 reads, in an arbitrary local frame field for the Su,u ,
/trχ′ )A = fAB (d
(Dd /trχ′ )B + gA
BCD
/χ̂′ )BCD + rA
(∇
/χ̂′ )ABC = hD
(D∇ ABC (d/trχ′ )D + iDEF
ABC (∇ /χ̂′ )DEF + sABC
We have, pointwise:
|g|, |h|, |i| ≤ CΩ|χ̂| (4.80)
132
Also, recalling that, from 1.81,
1
/d log Ω = (η + η) (4.81)
2
we have, by 4.76, the pointwise bounds:
|r| ≤ Cδ −1/2 |u|−4 {R∞ ∞ 2

0 (β)(R0 (α)) + O(δ
1/2
)}
−1/2
|s| ≤ |∇
/α| + Cδ |u| {R0 (β)(R0 (α)) + O(δ 1/2 )}
−4 ∞ ∞ 2
(4.82)
By Lemma 4.2 with the S 1-form /dtrχ′ in the role of θ,
/trχ′ |2 ) + Ωtrχ|d
D(|d /trχ′ |2 = 2(d
/trχ′ , Dd
/trχ′ ) − 2Ωχ̂AB (d
/trχ′ )A (d
/trχ′ )B
From the first of 4.78 and the first of 4.79:
/trχ′ , Dd
(d /trχ′ ) = /trχ′ |2 + (d
−Ωtrχ|d /trχ′ , r + g · ∇
/χ̂′ )
≤ /trχ′ |2 + |d
−Ωtrχ|d /trχ′ |(|r| + |g||∇
/χ̂′ |)
Since also |χ̂AB (d

/trχ′ )A (d
/trχ′ )B | ≤ |χ̂||d
/trχ′ |2 , we obtain:
/trχ′ |2 ) + 3Ωtrχ|d
D(|d /trχ′ |2 ≤ 2|d
/trχ′ |(|r| + |g||∇
/χ̂′ |) + 2Ω|χ̂||d
/trχ′ |2 (4.83)
/χ̂′ in the role of θ,

By Lemma 4.2 with the 3-covariant S tensorfield ∇
/χ̂′ |2 ) + 3Ωtrχ|∇
D(|∇ /χ̂′ |2 /χ̂′ , D∇
= 2(∇ /χ̂′ ) − 2Ωχ̂A /χ̂′ )DBC (∇
D (∇ /χ̂′ )ABC
−2Ωχ̂B /χ̂′ )ADC (∇
D (∇ /χ̂′ )ABC − 2Ωχ̂C /χ̂′ )ABD (∇
D (∇ /χ̂′ )ABC
From the second of 4.78:
/χ̂′ , D∇
(∇ /χ̂′ ) /χ̂′ , s + h · /dtrχ′ + i · ∇
= (∇ /χ̂′ )
/χ̂′ |(|s| + |h||d
≤ |∇ /trχ′ | + |i||∇
/χ̂′ |)
Since also each of χ̂A /χ̂′ )DBC (∇

D (∇ /χ̂′ )ABC , χ̂B /χ̂′ )ADC (∇
D (∇ /χ̂′ )ABC , χ̂C /χ̂′ )ABD (∇
D (∇ /χ̂′ )ABC
′ 2
is bounded in absolute value by |χ̂||∇ /χ̂ | (the last two are equal in view of the
symmetry of ∇ /χ̂′ in the last two indices) we obtain:
/χ̂′ |2 ) + 3Ωtrχ|∇
D(|∇ /χ̂′ |2 ≤ 2|∇
/χ̂′ |(|s| + |h||d
/trχ′ | + |i||∇
/χ̂′ |) + 6Ω|χ̂||∇
/χ̂′ |2 (4.84)
Let us define the non-negative function:

p
ψ = |d /trχ′ |2 + |∇
/χ̂′ |2 (4.85)
By 4.83, 4.84, and the pointwise bounds 4.80, ψ 2 satisfies the inequality:
D(ψ 2 ) + 3Ωtrχψ 2 ≤ 2CΩ|χ̂|ψ 2 + 2ψ(|r| + |s|) (4.86)
We then apply Lemma 4.4 with p = 4, λ = 3, l = C, and ρ = |r| + |s|, to obtain:

Z un o
kψkL4 (Su,u ) ≤ C krkL4 (Su′ ,u ) + kskL4 (Su′ ,u ) du′ (4.87)
0
133
Now for any function φ ∈ L∞ (Su,u ) and any p ≥ 1 we have, by Lemma 4.3,
kφkLp (Su,u ) ≤ (Area(Su,u ))1/p kφkL∞ (Su,u ) ≤ (8π|u|2 )1/p kφkL∞ (Su,u ) (4.88)
Consequently, the inequalities 4.82 imply, in view of the first of the definitions
4.1,
krkL4 (Su′ ,u ) ≤ Cδ −1/2 |u|−7/2 {R∞ ∞ 2

0 (β)(R0 (α)) + O(δ
1/2
)}
kskL4(Su′ ,u ) ≤ δ −3/2 |u|−3/2 R
/41 (α)
+Cδ −1/2 |u|−7/2 {R∞ ∞ 2
0 (β)(R0 (α)) + O(δ
1/2
)} (4.89)
Substituting in 4.87 then yields:
kψkL4 (Su,u ) ≤ Cδ −1/2 |u|−3/2 R

/41 (α) + Cδ 1/2 |u|−7/2 {R∞ ∞ 2
0 (β)(R0 (α)) + O(δ
1/2
)}
(4.90)
Going back to 4.83, that inequality implies:
D(|d /trχ′ |2 ≤ 2|d
/trχ′ |(|r| + CΩ|χ̂|ψ) (4.91)
Applying Lemma 4.4 with |d /trχ′ | in the role of ψ and |r| + CΩ|χ̂|ψ in the role
of ρ yields, taking p = 4:
Z u
′
/trχ kL (Su,u ) ≤ C
kd 4 krkL4 (Su′ ,u ) du′ (4.92)
0
Z u
+Cδ −1/2 |u|−1 R∞ 0 (α) kψkL4 (Su′ ,u ) du′
0
Substituting the estimate 4.90 as well as the first of the estimates 4.89 then
yields:
/trχ′ kL4 (Su,u ) ≤ C|u|−5/2 R∞

kd /41 (α)+Cδ 1/2 |u|−7/2 {R∞
0 (α)R
∞ 2
0 (β)(R0 (α)) +O(δ
1/2
)}
(4.93)
On the other hand, in view of the definition 4.85, the estimate 4.90 directly
implies:
/χ̂′ kL4 (Su,u ) ≤ Cδ −1/2 |u|−3/2 R

k∇ /41 (α)+Cδ 1/2 |u|−7/2 {R∞ ∞ 2
0 (β)(R0 (α)) +O(δ
1/2
)}
(4.94)
Since
/dtrχ = /d(Ωtrχ′ ) = Ω(d/trχ′ + (d / log Ω)trχ′ )
∇ /(Ωχ̂′ ) = Ω(∇
/χ̂ = ∇ /χ̂′ + (d
/ log Ω) ⊗ χ̂′ )
by 4.81 and the bounds 4.76 the estimates 4.93, 4.94 imply the estimates of the
proposition.
134
/χ
With the notations of Section 3 of Chapter 3, we consider any (u1 , u1 ) ∈ D′ and
fix attention in the following lemmas to the parameter subdomain D1 and the
corresponding subdomain M1 of M ′ (see 3.45, 3.46).
With a positive constant k to be appopriately chosen in the sequel, let s∗ be
the least upper bound of the set of values of s ∈ [u0 , u1 ] such that:
/ log ΩkL4 (Su,u ) ≤ kδ|u|−5/2

kd : for all (u, u) ∈ [0, u1 ] × [u0 , s] (4.95)
Then by continuity s∗ > u0 (recall that by 3.44 /d log Ω vanishes along Cu0 ) and
we have:
/ log ΩkL4 (Su,u ) ≤ kδ|u|−5/2

kd : for all (u, u) ∈ [0, u1 ] × [u0 , s∗ ] (4.96)
Let us denote:
/41 (trχ) = |u0 |7/2 δ −1 sup kd

D /trχkL4 (Su,u0 ) (4.97)
u∈[0,δ]
/41 (χ̂) = |u0 |5/2 δ −1/2 sup k∇

D /χ̂kL4 (Su,u0 ) (4.98)
u∈[0,δ]
Here, as in 3.49, 3.122, we are considering all of Cu0 , not only the part which
lies in M ′ . This is the same for c∗ ≥ u0 + δ, but not for c∗ ∈ (u0 , u0 + δ). In
∗
the latter case the part of Cu0 lying in M ′ is the part Cuc 0 −u0 , so u would be
restricted to the interval [0, c∗ − u0 ]. By the results of Chapter 2, D/41 (trχ) and
4
D
/1 (χ̂) are bounded by a non-negative non-decreasing continuous function of M4 .
∗
Also, let us denote by D1s the subset of D1 where u ≤ s∗ :
∗
D1s = [0, u1 ] × [u0 , s∗ ] (4.99)
∗
and by M1s the corresponding subset of M1 .
Lemma 4.8 We have:
Cδ|u|−7/2 k + (1 + D0∞ (χ̂) + O(δ))(D

/41 (trχ) + D
/41 (χ̂))

kd
/trχkL4 (Su,u ) ≤
+Cδ 3/2 |u|−9/2 (D0∞ (χ̂) + O(δ))2 (R∞
0 (β) + O(δ
1/2
))
+Cδ 2 |u|−5 (D0∞ (χ̂) + O(δ))R
/41 (α)
k∇
/χ̂kL4 (Su,u ) ≤ Cδ 1/2 |u|−5/2 (D
/41 (trχ) + D
/41 (χ̂))
+Cδ|u|−7/2 {k + (D0∞ (χ̂) + O(δ))(R∞
0 (β) + O(δ
1/2
))}
+Cδ 3/2 |u|−4 R
/41 (α)
∗
for all (u, u) ∈ D1s , provided that δ is suitably small depending on D0∞ , R∞
0 .
135
Proof: We consider equations 3.9 and 3.11. In view of the second of 3.5,
equation 3.11 is equivalent to:
Dχ̂′ = −α + Ω2 |χ̂′ |2 g/ (4.100)
We apply Lemma 1.2 to equation 3.9 and Lemma 4.1 to equation 4.100 to obtain
/χ̂′ along each
the following linear system of propagation equations for /dtrχ′ , ∇
generator of each C u :
/trχ′
Dd /χ̂′ + r
= f · /dtrχ′ + g · ∇
/χ̂′
D∇ /χ̂′ + s
= h · /dtrχ′ + i · ∇ (4.101)
Here f , g, r, h, i, s are respectively T11 , T13 , T10 , T31 , T33 , T30 type S tensorfields, the
are given by:
fB
A
= −Ω2 trχ′ δA
B
g BCD
A
= −2Ω2 χ̂′CD δA B

1 ′ 2 ′ 2
/AΩ (trχ ) + |χ̂ |
r A = −2Ωd
2
1 2
hABC = − Ω (2χ̂′BC δA
D D
+ χ̂′AC δB D
+ χ̂′AB δCD
− g/AB χ̂′DC
− g/AC χ̂′D
B
)
2
′EF
iDEF 2 D
ABC = 2Ω δA g /BC χ̂
1 2 h D E ′F F ′E E ′F F ′E
− Ω δA (δB χ̂C + δB χ̂C + δC χ̂B + δC χ̂B )
2
E D ′F D ′F F D ′E D ′E
+δA (δB χ̂C + δC χ̂B ) + δA (δB χ̂C + δC χ̂B )
i
−χ̂′D
B
(δ E F
A δ C + δ F E
A δ C ) − χ̂ ′D E F
C
(δ A δ B + δ F E
A δ B )
/A αBC + 2Ω|χ̂′ |2/dA Ωg/BC
sABC = −∇
−2Ω(χ̂′D
B A
′D
/d Ω + χ̂A /dB Ω − χ̂′AB /dD Ω)χ̂′DC
−2Ω(χ̂′D
C A
′D
/d Ω + χ̂A /dC Ω − χ̂′AC /dD Ω)χ̂′DB
h
−Ωtrχ′ 2χ̂′BC /dA Ω + χ̂′AB /dC Ω + χ̂′AC /dB Ω
i
−(g/AB χ̂′DC + g/AC χ̂′DB )d /D Ω (4.102)
Thus 4.101 reads, in an arbitrary local frame field for the Su,u ,
/trχ′ )A
(Dd = fB
A
/trχ′ )B + gBCD
(d A
/χ̂′ )BCD + r A
(∇
/χ̂′ )ABC
(D∇ = hD /trχ′ )D + iDEF
ABC (d /χ̂′ )DEF + sABC
ABC (∇
We have the pointwise bounds:
|g|, |h|, |i| ≤ CΩ|χ̂| (4.103)
136
Moreover, by virtue of 4.96 and the bounds 4.76:
∗
krkL4 (Su,u ) ≤ Ckδ|u|−9/2 : ∀(u, u) ∈ D1s (4.104)
provided that δ is suitably small depending on D0∞ , R∞

0 . Also, from the sixth
of the definitions 4.1 and the L∞ bound for /d log Ω which results from the
expression 4.81, we obtain:
kskL4 (Su,u ) ≤ δ 3/2 |u|−5 R

/41 (α) + Cδ|u|−9/2 (D0∞ (χ̂) + O(δ))(R∞
0 (β) + O(δ
1/2
))
(4.105)
provided that δ is suitably small depending on D0∞ , R∞ 0 .
Now, following an argument similar to that leading from equations 4.78 to
the inequalities 4.83 and 4.84, we deduce from equations 4.101, using Lemma
4.2, the inequalities:
D(|d /trχ′ |2 ≤ 2|d /χ̂′ |) + 2Ω|χ̂||d
/trχ′ |(|r| + |g||∇ /trχ′ |2 (4.106)
and:
/χ̂′ |2 )+3Ωtrχ|∇
D(|∇ /χ̂′ |2 ≤ 2|∇
/χ̂′ |(|s|+|h||d /χ̂′ |)+6Ω|χ̂||∇
/trχ′ |+|i||∇ /χ̂′ |2 (4.107)

q
ψ = |d /χ̂′ |2
/trχ′ |2 + |∇ (4.108)
By 4.106, 4.107, and the pointwise bounds 4.103, ψ 2 satisfies the inequality:
D(ψ 2 ) + 3Ωtrχψ 2 ≤ 2CΩ|χ̂|ψ 2 + 2ψ(|r| + |s|) (4.109)
We then apply Lemma 4.5 with p = 4, λ = 3, l = C, and ρ = |r| + |s|, to obtain:
|u|5/2 kψkL4 (Su,u ) ≤ C|u0 |5/2 kψkL4 (Su,u0 ) (4.110)

Z u n o
+C |u′ |5/2 krkL4 (Su,u′ ) + kskL4 (Su,u′ ) du′
u0
Substituting the bounds 4.104, 4.105 and noting that by the definitions 4.97,
4.98 and 4.108:
kψkL4 (Su,u0 ) ≤ δ 1/2 |u0 |−5/2 (D

/41 (trχ) + D
/41 (χ̂)) (4.111)
∗
we conclude that, for all (u, u) ∈ D1s :
kψkL4 (Su,u ) ≤ Cδ 1/2 |u|−5/2 (D /41 (χ̂))

/41 (trχ) + D
+Cδ|u|−7/2 {k + (D0∞ (χ̂) + O(δ))(R∞
0 (β) + O(δ
1/2
))}
+Cδ 3/2 |u|−4 R
/41 (α) (4.112)
Going back to 4.106, that inequality implies:
D(|d /trχ′ |2 ≤ 2|d
/trχ′ |(|r| + CΩ|χ̂|ψ) (4.113)
137
Applying Lemma 4.5 with |d /trχ′ | in the role of ψ and |r| + CΩ|χ̂|ψ in the role
of ρ yields, taking p = 4:
|u|5/2 kd
/trχ′ kL4 (Su,u ) /trχ′ kL4 (Su,u0 )
≤ C|u0 |5/2 kd (4.114)
Z u
+C |u′ |5/2 krkL4 (Su,u′ ) du′
u0
Z u
+Cδ 1/2 (D0∞ (χ̂) + O(δ)) |u′ |1/2 kψkL4 (Su,u′ ) du′
u0
Substituting the bounds 4.104 and 4.112 and recalling the definition 4.97, we
then obtain:
/41 (trχ) + D
/trχ′ kL4 (Su,u ) ≤ Cδ|u|−7/2 k + (1 + D0∞ (χ̂) + O(δ))(D /41 (χ̂))

kd
+Cδ 3/2 |u|−9/2 (D0∞ (χ̂) + O(δ))2 (R∞
0 (β) + O(δ
1/2
))
+Cδ 2 |u|−5 (D0∞ (χ̂) + O(δ))R
/41 (α) (4.115)
∗
0 .
On the other hand, in view of the definition 4.108, the estimate 4.112 directly
implies:
/χ̂′ kL4 (Su,u )

k∇ ≤ Cδ 1/2 |u|−5/2 (D
/41 (trχ) + D
/41 (χ̂))
+Cδ|u|−7/2 {k + (D0∞ (χ̂) + O(δ))(R∞
0 (β) + O(δ
1/2
))}
+Cδ 3/2 |u|−4 R
/41 (α) (4.116)
∗
for all (u, u) ∈ D1s . Since
/dtrχ = /d(Ωtrχ′ ) = Ω(d

/trχ′ + (d
/ log Ω)trχ′ )
∇ /(Ωχ̂′ ) = Ω(∇
/χ̂ = ∇ /χ̂′ + (d
/ log Ω) ⊗ χ̂′ )
by 4.96 the estimates 4.115 and 4.116 imply the estimates of the lemma.

/η, ∇
/η
∗
We proceed to derive L4 (S) estimates for ∇ /η in M1s . We consider equations
/η, ∇
1.84 and 1.85. Applying Lemma 4.1 to these equations we obtain the following
linear system of propagation equations for ∇/η, ∇
/η along the generators of each
Cu and each C u respectively:
D∇
/η = a·∇
/η + b
D∇
/η = a·∇
/η + b (4.117)
Here a, a are T22 type S tensorfields while b, b are T20 type S tensorfields, the
138
are given by:
aCD
AB = C D
ΩδA χB
aCD
AB = C D
ΩδA χB
bAB = / A χC
Ω{(∇ /A log Ω)χC
B )η C + (d B ηC }
/)C
−(DΓ AB ηC − Ω{∇
/A βB + (d
/A log Ω)βB }
bAB = / A χC
Ω{(∇ /A log Ω)χC
)η + (d
B C
η }
B C
/)C
−(DΓ AB η C + Ω{∇
/A β B + (d
/A log Ω)β B } (4.118)
Thus 4.117 read, in an arbitrary local frame field for the Su,u ,
(D∇
/η)AB = aCD /η)CD + bAB
AB (∇
/η)AB
(D∇ = aCD
AB (∇
/η)CD + bAB
We decompose:
1 C D
aCD
AB = C D
Ωδ δ trχ + ΩδA χ̂B
2 A B
1 C D C D
aCD
AB = Ωδ δ trχ + ΩδA χ̂B (4.119)
2 A B
Then the above equations take the form:
1
(D∇
/η)AB = /η)AB + Ωχ̂C
Ωtrχ(∇ B (∇
/η)AC + bAB
2
1
/η)AB
(D∇ = /η)AB + Ωχ̂C
Ωtrχ(∇ B
(∇
/η)AC + bAB (4.120)
2
To the first of 4.120 we apply Lemma 4.6 with ∇
/η in the role of θ and
1
/η + Ωχ̂ · ∇
Ωtrχ∇ /η + b
2
in the role of ξ. Then r = 2, ν = 0, γ = 0. Noting that

1
Ω |trχ| + |χ̂| ≤ ã (4.121)
2
where
ã = Cδ −1/2 |u|−1 (R∞
0 (α) + O(δ
1/2
)) (4.122)
we obtain, taking p = 4:
Z u
k∇
/ηkL4(Su,u ) ≤ Cã(u) /ηkL4 (Su′ ,u) du′
k∇
Z u0
+C kbkL4 (Su,u ) du′ (4.123)
0
139
To the second of 4.120 we apply Lemma 4.7 with ∇
/η in the role of θ and
1
Ωtrχ∇
/η + Ωχ̂ · ∇
/η + b
2
in the role of ξ. Then r = 2, ν = 0, γ = 0. Noting that

1
Ω |trχ| + |χ̂| ≤ ã (4.124)
2
where
ã = C|u|−1 (1 + O(δ 1/2 )) (4.125)
we obtain, taking p = 4:
|u|3/2 k∇
/ηkL4 (Su,u ) ≤ C|u0 |3/2 k∇
/ηkL4 (Su,u0 )
Z u
+C /ηkL4 (Su,u′ ) du′
|u′ |3/2 ã(u′ )k∇
u0
Z u
+C |u′ |3/2 kbkL4 (Su,u′ ) du′ (4.126)
u0
From 4.125 we obtain:

ã ≤ C|u|−1 (4.127)
(for a new numerical constant C) provided that δ is suitably small depending
on D0∞ , R∞
0 . Also, we define:
/41 (η) = |u0 |5/2 δ −1/2 sup k∇

D /ηkL4 (Su,u0 ) (4.128)
u∈[0,u1 ]
Substituting the above in 4.126 yields:

Z u
|u|3/2 k∇
/ηkL4 (Su,u ) ≤ C|u0 |−1 δ 1/2 D
/41 (η) + C |u′ |1/2 k∇
/ηkL4 (Su,u′ ) du′
u0
Z u
+C |u | kbkL4 (Su,u′ ) du′
′ 3/2
(4.129)
u0
∗
for all (u, u) ∈ D1s .
The inequalities 4.123, 4.129 constitute a linear system of integral inequali-
∗
/ηkL4 (Su,u ) in the domain D1s . Setting:
/ηkL4 (Su,u ) , k∇
ties for the quantities k∇
z(u) = sup k∇
/ηkL4 (Su,u ) (4.130)
u∈[0,u1 ]
and: Z u
B(u) = |u′ |3/2 sup kbkL4 (Su,u′ ) du′ (4.131)
u0 u∈[0,u1 ]
140
∗
4.129 implies, for all (u, u) ∈ D1s :
Z u
3/2
|u| k∇/ηkL4 (Su′ ,u ) du′ ≤ Cδ 3/2 |u0 |−1 D/41 (η) + CδB(u)
0
Z u
+Cδ |u′ |1/2 z(u′ )du′ (4.132)
u0
∗
Substituting in 4.123 we then obtain, for all (u, u) ∈ D1s :
k∇
/ηkL4 (Su,u ) ≤ C|u|−3/2 ã(u)(δ 3/2 |u0 |−1 D
/41 (η) + δB(u))
Z u Z u
′ −3/2
+C kbkL4(Su′ ,u ) du + C|u| ã(u)δ |u′ |1/2 z(u′ )du′
0 u0
Z u
z(u) ≤ λ(u) + ν(u) |u′ |1/2 z(u′ )du′ (4.133)
u0
where:
λ(u) = Cδ sup kbkL4(Su,u ) + Cδ|u|−3/2 ã(u)(δ 1/2 |u0 |−1 D

/41 (η) + B(u)) (4.134)
u∈[0,u1 ]
and:
ν(u) = Cδ|u|−3/2 ã(u) (4.135)
Setting Z u
Z(u) = |u′ |1/2 z(u′ )du′ , we have Z(u0 ) = 0 (4.136)
u0

dZ
≤ |u|1/2 (λ + νZ) (4.137)
du
Integrating from u0 we obtain:
Z u Z u
Z(u) ≤ exp |u | ν(u )du |u′ |1/2 λ(u′ )du′
′′ 1/2 ′′ ′′
: ∀u ∈ [u0 , s∗ ] (4.138)
u0 u′
From 4.135 and 4.122 we have:

Z u Z u
|u′′ |1/2 ν(u′′ )du′′ = Cδ |u′′ |−1 ã(u′′ )du′′
u′ u′
≤ C|u|−1 δ 1/2 (R∞
0 (α) + O(δ
1/2
))
≤ 1 (4.139)
provided that δ is suitably small depending on D0∞ , R∞ 0 . Therefore:

Z u
Z(u) ≤ C |u′ |1/2 λ(u′ )du′ : ∀u ∈ [u0 , s∗ ] (4.140)
u0
141
Z u
z(u) ≤ λ(u) + Cν(u) |u′ |1/2 λ(u′ )du′ : ∀u ∈ [u0 , s∗ ] (4.141)
u0
Moreover, since Z u
|u′ |1/2 k∇
/ηkL4 (Su,u′ ) du′
u0
setting, in analogy with 4.130,
z(u) = sup k∇
/ηkL4 (Su,u ) (4.142)
u∈[0,u1 ]
substituting 4.140 in 4.129 and taking the supremum over u ∈ [0, u1 ] yields:
Z u
|u|3/2 z(u) ≤ C|u0 |−1 δ 1/2 D
/41 (η) + CB(u) + C |u′ |1/2 λ(u′ )du′ : ∀u ∈ [u0 , s∗ ]
u0
(4.143)
/41 (η). To do this we recall
We shall now derive an appropriate estimate for D
that along Cu0 condition 3.77 holds:
η = −η : along Cu0 (4.144)
Thus, the first of 4.117 simplifies along Cu0 to:
D∇
/η = −a · ∇
/η + b : along Cu0 (4.145)
Equivalently, the first of 4.120 simplifies along Cu0 to:

1
/η)AB = −Ωχ̂C
/η)AB + Ωtrχ(∇
(D∇ B (∇
/η)AC + bAB (4.146)
2
We apply Lemma 4.6 with ∇ /η in the role of θ and b in the role of ξ. Here r = 2,
ν = −1, γ = −ΩI ⊗ χ̂♯ . Taking p = 4 we obtain:
Z u
k∇/ηkL4 (Su,u0 ) ≤ C kbkL4 (Su′ ,u ) du′ (4.147)
0
0
which, in view of 4.144, implies:
/41 (η) ≤ C|u0 |5/2 δ 1/2 sup kbkL4 (Su,u0 )

D (4.148)
u∈[0,u1 ]
We shall now derive bounds for b and b in L4 (Su,u ). From the expressions
for DΓ
/, DΓ/ of Lemma 4.1 we have, pointwise on Su,u ,
|DΓ
/| ≤ 3|∇
/(Ωχ)|, /| ≤ 3|∇
|DΓ /(Ωχ)|
hence:
/kL4(Su,u ) ≤ 3k∇
kDΓ /(Ωχ)kL4 (Su,u ) , kDΓ
/kL4(Su,u ) ≤ 3k∇
/(Ωχ)kL4 (Su,u ) (4.149)
142
Writing
∇
/(Ωχ) = Ω(∇
/χ + (d
/ log Ω) ⊗ χ)
from Proposition 4.1, 4.81 and the bounds 4.76 we obtain:
k∇
/(Ωχ)kL4 (Su,u ) ≤ Cδ −1/2 |u|−3/2 R
/41 (α) + O(|u|−5/2 )
: ∀(u, u) ∈ D′ (4.150)

0 . Also, writing
∇
/(Ωχ) = Ω(∇
/χ + (d
/ log Ω) ⊗ χ)
from Lemma 4.8 and 4.96 we obtain:
k∇
/(Ωχ)kL4 (Su,u ) ≤ Cδ 1/2 |u|−5/2 (D
/41 (trχ) + D
/41 (χ̂))
+Cδ|u|−7/2 k + O(δ|u|−7/2 ) + Cδ 3/2 |u|−4 R
/41 (α)
∗
: ∀(u, u) ∈ D1s (4.151)
Let us denote from now on by O(δ p |u|r ), for real numbers p, r, the product of
δ p |u|r with a non-negative non-decreasing continuous function of the quantities
D0∞ , R∞ /41 , R
0 , and D /41 , where:
/41 = max{D
D /41 (χ̂)}
/41 (trχ), D (4.152)
/41 = max{R
R /41 (α), R
/41 (β), R
/41 (ρ), R
/41 (σ), R
/41 (β), R
/41 (α)} (4.153)
With this notation 4.151 simplifies to:
k∇
/(Ωχ)kL4 (Su,u ) ≤ Cδ 1/2 |u|−5/2 (D /41 (χ̂))
/41 (trχ) + D
+Cδ|u|−7/2 k + O(δ|u|−7/2 )
∗
: ∀(u, u) ∈ D1s (4.154)
Note that the dependence on the constant k is, and must be, kept explicit.
Using 4.150, the first of 4.149, the second of the definitions 4.1, and 4.81
together with the bounds 4.76, we obtain from the third of 4.118:
kbkL4(Su,u ) ≤ Cδ −1/2 |u|−5/2 R

/41 (β) + O(|u|−7/2 )
: ∀(u, u) ∈ D′ (4.155)
Also, using 4.154, the second of 4.149, the fifth of the definitions 4.1, and 4.81
together with the bounds 4.76, we obtain from the fourth of 4.118:
n o
kbkL4 (Su,u ) ≤ Cδ|u|−9/2 R /41 (β) + (R∞
0 (β) + O(δ 1/2
))D
/ 4
1 (trχ) + D
/ 4
1 (χ̂))
+Cδ 3/2 |u|−11/2 (R∞
0 (β) + O(δ
1/2
))k + O(δ 3/2 |u|−11/2 )
∗
: ∀(u, u) ∈ D1s (4.156)
143
Hence (see 4.131):
n o
B(u) ≤ Cδ|u|−2 R/41 (β) + (R∞
0 (β) + O(δ
1/2
/41 (trχ) + D
))(D /41 (χ̂))
+Cδ 3/2 |u|−3 (R∞
0 (β) + O(δ
1/2
))k + O(δ 3/2 |u|−3 )
: ∀u ∈ [u0 , s∗ ] (4.157)
Substituting 4.148, 4.155, and 4.157, in 4.134 and recalling 4.122 we obtain:
λ(u) ≤ Cδ 1/2 |u|−5/2 R

/41 (β) + O(δ|u|−7/2 )
+Cδ 2 |u|−11/2 (R∞
0 (α) + O(δ
1/2
))(R∞
0 (β) + O(δ
1/2
))k
: ∀u ∈ [u0 , s∗ ] (4.158)
Substituting 4.158 in 4.141 and noting that from 4.135 and 4.122:
|u|3/2 ν(u) = Cδã(u) ≤ Cδ 1/2 |u|−1 (R∞

0 (α) + O(δ
1/2
)) ≤ 1 (4.159)

0 , yields:
/41 (β) + O(δ|u|−7/2 )

z(u) ≤ Cδ 1/2 |u|−5/2 R
+Cδ 2 |u|−11/2 (R∞
0 (α) + O(δ
1/2
))(R(β) + O(δ 1/2 ))k
: ∀u ∈ [u0 , s∗ ] (4.160)
Also, substituting 4.148, 4.155, and 4.157, 4.158, in 4.143, yields:
z(u) ≤ Cδ 1/2 |u|−5/2 R

/41 (β) + O(δ|u|−7/2 )
+Cδ 3/2 |u|−9/2 (R∞
0 (β) + O(δ
1/2
))k
: ∀u ∈ [u0 , s∗ ] (4.161)
In conclusion, recalling the definitions 4.130, 4.142 of z(u), z(u), we have estab-
lished the following lemma.
Lemma 4.9 We have:
k∇
/ηkL4 (Su,u ) ≤ Cδ 1/2 |u|−5/2 R
/41 (β) + O(δ|u|−7/2 )
+Cδ 2 |u|−11/2 (R∞
0 (α) + O(δ
1/2
))(R∞
0 (β) + O(δ
1/2
))k
k∇
/ηkL4 (Su,u ) ≤ Cδ 1/2 |u|−5/2 R
/41 (β) + O(δ|u|−7/2 )
+Cδ 3/2 |u|−9/2 (R∞
0 (β) + O(δ
1/2
))k
∗
0 .
144
4.5 L4(S) estimates for /dω, /dω
∗
We proceed to derive an L4 (S) estimate for /dω on M1s . Using this estimate we
∗
shall derive an L4 (S) estimate for /d log Ω in M1s which, with a suitable choice
of the constant k, improves the bound 4.96. This shall enable us to show that
s∗ = u1 so that the previous lemmas actually hold on the entire parameter
domain D1 , and, since (u1 , u1 ) ∈ D′ is arbitrary, the estimates hold on all of
D′ , that is, on all of M ′ . We shall then derive an L4 (S) estimate for /dω on M ′ .
Lemma 4.10 We have:
kd
/ωkL4 (Su,u ) ≤ Cδ|u|−7/2 R
/41 (ρ) + O(δ 3/2 |u|−9/2 )
+Cδ 3 |u|−13/2 (R∞
0 (β) + O(δ
1/2 2
)) k
∗
0 .
Proof: We apply /d to equation 1.107 to obtain, by virtue of Lemma 1.2, the
following propagation equation for /dω along the generators of each Cu :
/ ω = Ω2 l
Dd (4.162)
where:
l = −2(η − η) · ∇
/η + 2η · ∇
/η − /dρ
+(η + η)(−|η|2 + 2(η, η) − ρ) (4.163)
Here we have also used 4.81. To 4.162 we apply Lemma 4.6 with /dω in the role
of θ and Ω2 l in the role of ξ. Here r = 1, ν = 0, γ = 0. Taking p = 4 we obtain:
Z u
kd
/ωkL4 (Su,u ) ≤ C klkL4(Su′ ,u ) du′ : ∀(u, u) ∈ D′ (4.164)
0
Using the L4 (S) estimates for ∇ /η, ∇/η of Lemma 4.9 as well as the third of the
∗
definitions 4.1 we obtain, for all (u, u) ∈ D1s ,
klkL4(Su,u ) ≤ |u|−7/2 R
/41 (ρ) + O(δ 1/2 |u|−9/2 )
+Cδ 2 |u|−13/2 (R∞
0 (β) + O(δ
1/2 2
)) k (4.165)

0 . Substituting this in
4.164 yields the lemma.
Lemma 4.11 There is a numerical constant C such that with
/41 (ρ) + O(δ 1/2 )

k = CR
we have s∗ = u1 and the estimate:
/ log ΩkL4 (Su,u ) ≤ Cδ|u|−5/2 R

kd /41 (ρ) + O(δ 3/2 |u|−7/2 )
145
holds for all (u, u) ∈ D1 , provided that δ is suitably small depending on D0∞ ,
R∞0 .
Proof: By Lemma 1.2 and the second of the definitions 1.30:
Dd
/ log Ω = /dω (4.166)
To this we apply Lemma 4.7 with /d log Ω in the role of θ and /dω in the role of
ξ. Here r = 1, ν = 0, γ = 0. Taking p = 4 and taking into account the fact that
/d log Ω vanishes on Cu0 (see 3.44) we obtain:
Z u
1/2
|u| kd / log ΩkL4 (Su,u ) ≤ C |u′ |1/2 kd
/ωkL4 (Su,u ) du′ (4.167)
u0
Substituting the L4 (S) estimate for /dω of Lemma 4.10 we obtain:
kd
/ log ΩkL4 (Su,u ) ≤ Cδ|u|−5/2 R
/41 (ρ) + O(δ 3/2 |u|−7/2 )
+Cδ 3 |u|−11/2 (R∞
0 (β) + O(δ
1/2 2
)) k
∗
: ∀(u, u) ∈ D1s (4.168)
In reference to this estimate, let:
a /41 (ρ) + O(δ 1/2 )

= CR
b = C(R∞
0 (β) + O(δ
1/2 2
)) (4.169)
The estimate 4.168 implies:

∗
/ log ΩkL4 (Su,u ) ≤ δ(a + δ 2 bk)|u|−5/2
kd : ∀(u, u) ∈ D1s (4.170)
Choosing:
k = 2a (4.171)
we have:
a + δ 2 bk < 2a provided that 2bδ 2 < 1 (4.172)
The last is a smallness condition on δ depending on D0∞ , R∞
0 . The estimate
4.170 then implies:
∗
/ log ΩkL4 (Su,u ) < kδ|u|−5/2
kd : ∀(u, u) ∈ D1s = [0, u1 ] × [0, s∗ ] (4.173)
∗
Since by the above lemma D1s = D1 , with k as in the statement of Lemma
4.11, the results of Lemmas 4.8, 4.9 and 4.10 hold for all (u, u) ∈ D1 . Since
(u1 , u1 ) is arbitrary, these results hold for all (u, u) ∈ D′ . Moreover, by virtue
of the L4 (S) estimate for /d log Ω of Lemma 4.11, we can now estimate (see third
of 4.102):
krkL4 (Su,u ) ≤ Cδ|u|−9/2 R
/41 (ρ) + O(δ 3/2 |u|−11/2 ) (4.174)
146
Using this bound in place of the bound 4.104, we deduce, following the argument
leading to the estimate 4.115,
/trχ′ kL4 (Su,u ) ≤ Cδ|u|−7/2 R

4
kd /41 (trχ) + D
/1 (ρ) + (1 + D0∞ (χ̂) + O(δ))(D /41 (χ̂))
+O(δ 3/2 |u|−9/2 ) : ∀(u, u) ∈ D′ (4.175)
Also, applying Lemma 4.7, taking p = 4, with ∇ /χ̂′ in the role of θ, i in the role
′
of γ and h · /dχ + s in the role of ξ, to the second of 4.101, in which case r = 3,
ν = 0, and using the estimates 4.105 and 4.175, we deduce:
/χ̂′ kL4 (Su,u ) ≤ Cδ 1/2 |u|−5/2 D

k∇ /41 (χ̂) + O(δ|u|−7/2 ) (4.176)
We thus arrive at the following proposition.
Proposition 4.2 The following estimates hold for all (u, u) ∈ D′ :
Cδ|u|−7/2 R
4
kd
/trχkL4 (Su,u ) ≤ /1 (ρ) + (1 + D0∞ (χ̂) + O(δ))(D
/41 (trχ) + D
/41 (χ̂))
+O(δ 3/2 |u|−9/2 )
k∇
/χ̂kL4 (Su,u ) ≤ Cδ 1/2 |u|−5/2 D
/41 (χ̂) + O(δ|u|−7/2 )
k∇
/ηkL4 (Su,u ) ≤ Cδ 1/2 |u|−5/2 R
/41 (β) + O(δ|u|−7/2 )
k∇
/ηkL4 (Su,u ) ≤ Cδ 1/2 |u|−5/2 R
/41 (β) + O(δ|u|−7/2 )
kd
/ωkL4 (Su,u ) ≤ Cδ|u|−7/2 R
/41 (ρ) + O(δ 3/2 |u|−9/2 )

0 .
/ωkL4(Su,u ) ≤ C|u|−5/2 R
kd /41 (ρ) + O(δ 1/2 |u|−7/2 ) : ∀(u, u) ∈ D′

0 .
Proof: We apply /d to equation 1.108 to obtain, by virtue of Lemma 1.2, the
following propagation equation for /dω along the generators of each C u :
/ ω = Ω2 l
Dd (4.177)
where:
l = 2(η − η) · ∇
/η + 2η · ∇
/η − /dρ
+(η + η)(−|η|2 + 2(η, η) − ρ) (4.178)
(Equation 4.177 is the conjugate of equation 4.162.) To 4.177 we apply Lemma

4.7 with /dω in the role of θ and Ω2 l in the role of ξ. Here r = 1, ν = 0, γ = 0.
Recalling that ω vanishes on Cu0 , we obtain, taking p = 4,
Z u
|u|1/2 kd
/ωkL4(Su,u ) ≤ C |u′ |1/2 klkL4 (Su,u′ ) du′ (4.179)
u0
147
Substituting the L4 (S) estimates for ∇ /η, ∇
/η of Proposition 4.2 as well as the
third of the definitions 4.1 we obtain, for all (u, u) ∈ D′ :
klkL4 (Su,u ) ≤ |u|−7/2 R

/41 (ρ) + O(δ 1/2 |u|−9/2 ) (4.180)
Substituting this in 4.179 yields the proposition.
4.6 L4(S) estimates for Dω, Dω

kDωkL4 (Su,u ) ≤ Cδ −1 |u|−3/2 R40 (Dρ) + O(|u|−5/2 )

kDωkL4 (Su,u ) ≤ Cδ|u|−7/2 R∞ 2
0 (ρ) + O(δ |u|
−9/2
)

0 .
Proof: By the commutation formula 1.91 we have:
DDω = DDω + 2Ω2 (η − η)♯ · /dω (4.181)
and Dω is given by equation 1.107. Applying D to equation 1.107 and using

the first of equations 1.41 we obtain:
DDω = Ω2 {2((η − η), Dη) + 2(η, Dη)

−4Ω(η, χ♯ · η) + 2Ω(η, χ♯ · η)
+2ω(2(η, η) − |η|2 − ρ) − Dρ} (4.182)
Substituting for Dη, Dη from 1.82, 1.174, we obtain the following propagation
equation for Dω:
D(Dω) = Ω2 n (4.183)
where:
n = −Dρ + 6(η − η)♯ · /dω + 2Ω(η − 2η, β) (4.184)

♯ 2
−6Ω(η − η, χ · η) + 2ω(2(η, η) − |η| − ρ)
To 4.183 we apply Lemma 4.7 with Dω in the role of θ and Ω2 n in the role of
ξ. Here r = 0, ν = 0, γ = 0. Taking into account the fact that Dω vanishes on
Cu0 and setting p = 4 we obtain:
Z u
|u|−1/2 kDωkL4 (Su,u ) ≤ C |u′ |−1/2 knkL4 (Su,u′ ) du′ (4.185)
u0
Using the L4 (S) estimate for /dω of Proposition 4.3 as well as the first of the
definitions 4.2 we obtain, for all (u, u) ∈ D′ ,
knkL4 (Su,u ) ≤ Cδ −1 |u|−3/2 R40 (Dρ) + O(|u|−5/2 ) (4.186)
148
Substituting this in 4.185 yields the L4 (S) estimate for Dω of the proposition.
Next, we deduce the conjugate of equation 4.183, which is the following
propagation equation for Dω:
D(Dω) = Ω2 n (4.187)
where:
n = −Dρ − 6(η − η)♯ · /dω − 2Ω(η − 2η, β) (4.188)

♯ 2
+6Ω(η − η, χ · η) + 2ω(2(η, η) − |η| − ρ)
To 4.187 we apply Lemma 4.6 with Dω in the role of θ and Ω2 n in the role of
ξ. Here r = 0, ν = 0, γ = 0. Setting p = 4 we obtain:
Z u
kDωkL4 (Su,u ) ≤ C knkL4(Su′ ,u ) du′ (4.189)
0
4
Using the L (S) estimate for /dω of Proposition 4.2 as well as the third of the
definitions 4.2 we obtain, for all (u, u) ∈ D′ ,
knkL4(Su,u ) ≤ kDρkL4 (Su,u ) + O(δ|u|−9/2 ) (4.190)
Now, according to the eighth of the Bianchi identities of Proposition 1.2:

3 1
Dρ + Ωtrχρ = −Ω div / β + (2η − ζ, β) + (χ̂, α) (4.191)
2 2
Hence, by the results of Chapter 3 and the fifth of the definitions 4.1 we have:
kDρkL4 (Su,u ) ≤ C|u|−7/2 R∞

0 (ρ) + O(δ|u|
−9/2
) (4.192)
Substituting this in 4.190 and the result in 4.189 yields the L4 (S) estimate for
Dω of the proposition.
149
Chapter 5
The Uniformization
Theorem
5.1 Introduction. An L2 (S) estimate for K − K

The arguments of the next two chapters use elliptic theory on the surfaces
Su,u . In particular, the estimates of the next chapter use L4 elliptic theory
on these surfaces. To derive the necessary estimates we rely on the conformal
properties of the elliptic systems under consideration and make essential use
of the uniformization theorem to reduce the problem to one on the standard
sphere. The uniformization theorem states that the 2-dimensional Riemannian
manifold (Su,u , g/) is conformal to the standard sphere. Now, the uniformization
theorem has been established in [C-K] under the assumption that the Gauss
curvature K of g/ belongs to L∞ (Su,u ) and the bounds for the conformal factor
derived there depend on kK − KkL∞ (Su,u ) . Here, for an arbitrary function f
defined on M ′ we denote by f the function which on each Su,u is the mean value
of f on Su,u . By the Gauss Bonnet theorem:
Z
Kdµ/g = 4π (5.1)
Su,u
hence:
4π
K= (5.2)
Area(Su,u )
We may derive an L∞ bound for K on Su,u through the Gauss equation 1.132,
which we may write in the form:
1 1
K + trχtrχ − (χ̂, χ̂) = −ρ (5.3)
4 2
150
In fact, from the 2nd, 3rd, 4th and 5th of the bounds 4.76 and the 3rd of the
definitions 3.4 we have:
1
K− ≤ C|u|−3 {R∞ ∞ ∞ ∞
0 (α)(R0 (α) + D0 (χ̂)) + R0 (ρ)} + O(δ|u|
−3
) (5.4)
|u|2 |
This estimate is however not suitable for our present purposes, because in the
present situation the quantities D0∞ and R∞ 0 are allowed to be as large as we
wish. The appropriate bound for K is the L2 bound to be derived in the present
section. In the next section an appropriate version of the uniformization theorem
will be deduced, which relies only on such an L2 bound for K.
We begin with the following proposition. We define:
|u|2 δk∇
/ 2 αkL2 (Cu )

R/2 (α) = sup (5.5)
u∈[u0 ,c∗ )
where the L2 norm on Cu is with respect to the measure:
dµ/g du
Thus, for any S tensorfield θ we have:

 R R
Z  δ |θ|2
dµ /
g du : u ∈ [u0 , c∗ − δ]
0 S
kθk2L2 (Cu ) = |θ|2 = u,u
R c∗ −u R
2
Cu 
0
|θ| dµ/g du
Su,u
: u ∈ (c∗ − δ, c∗ )
(5.6)
for c∗ ≥ u0 + δ, and:
!
Z Z c∗ −u Z
kθk2L2 (Cu ) = 2
|θ| = 2
|θ |dµ/g : u ∈ [u0 , c∗ ) (5.7)
Cu 0 Su,u
for c∗ < u0 +δ (see 3.2, 3.3 and Figures 1.1, 1.2). By the results of Chapter 2, the
quantity corresponding to R /2 (α) on Cu0 , obtained by replacing the supremum
on [u0 , c∗ ) by the value at u0 , is bounded by a non-negative non-decreasing
continuous function of M4 .

q
/ 2 χ̂′ k2L2 (Su,u ) ≤ δ −1/2 |u|−2 R
/ 2 trχ′ k2L2 (Su,u ) + k∇
|∇ /2 (α) + O(|u|−3 )
for all (u, u) ∈ D′ , provided that δ is suitably small depending on D0∞ , R∞ 0 .

Proof: We apply ∇ / to the system 4.78. In view of Lemma 4.1 we obtain the
following linear system of propagation equations for ∇ / 2 trχ′ , ∇
/ 2 χ̂′ along the
generators of the Cu :
/ 2 trχ′
D∇ / 2 trχ′ + g · ∇
= f ·∇ / 2 χ̂′ + r′
/ 2 χ̂′
D∇ / 2 trχ′ + i · ∇
= h·∇ / 2 χ̂′ + s′ (5.8)
151
where:
r′ = ∇ /(Ωχ), /dtrχ′ ) + ∇
/r + (∇ /f · /dtrχ′ + ∇
/g · ∇/χ̂′
s′ = ∇
/s + (∇
/(Ωχ), ∇ ′
/χ̂ ) + ∇ ′
/h · /dtrχ + ∇ /χ̂′
/i · ∇ (5.9)
Here we denote by ( , ) a bilinear expression with coefficients which depend

only on g/.
/ 2 trχ′ in the
By Lemma 4.2 with the (symmetric) 2-covariant S tensorfield ∇
role of θ,
/ 2 trχ′ |2 ) + 2Ωtrχ|∇
D(|∇ / 2 trχ′ |2 = 2(∇
/ 2 trχ′ , D∇
/ 2 trχ′ ) + 2(Ωχ̂, ∇
/ 2 trχ′ , ∇
/ 2 trχ′ )
Here we denote by ( , , ) a trilinear expression with coefficients which depend

only on g/. From the first of 5.8 and the first of 4.79:
/ 2 trχ′ , D∇
(∇ / 2 trχ′ ) = / 2 trχ′ |2 + (∇
−Ωtrχ|∇ / 2 trχ′ , r′ + g · ∇
/ 2 χ̂′ )
≤ / 2 trχ′ |2 + |∇
−Ωtrχ|∇ / 2 trχ′ |(|r′ | + |g||∇
/ 2 χ̂′ |)
/ 2 trχ′ , ∇
Since also |(Ωχ̂, ∇ / 2 trχ′ )| ≤ CΩ|χ̂||∇
/ 2 trχ′ |2 , we obtain:
/ 2 trχ′ |2 ) + 4Ωtrχ|∇
D(|∇ / 2 trχ′ |2 ≤ 2|∇
/ 2 trχ′ |(|r′ | + |g||∇
/ 2 χ̂′ |) + 2CΩ|χ̂||∇
/ 2 trχ′ |2
(5.10)
By Lemma 4.2 with the 4-covariant S tensorfield ∇ / 2 χ̂′ in the role of θ,
/ 2 χ̂′ |2 ) + 4Ωtrχ|∇
D(|∇ / 2 χ̂′ |2 = 2(∇
/ 2 χ̂′ , D∇
/ 2 χ̂′ ) + 2(Ωχ̂, ∇
/ 2 χ̂′ , ∇
/ 2 χ̂′ )
From the second of 5.8:
/ 2 χ̂′ , D∇
(∇ / 2 χ̂′ ) / 2 χ̂′ , s′ + h · ∇
= (∇ / 2 trχ′ + i · ∇
/ 2 χ̂′ )
/ 2 χ̂′ |(|s′ | + |h||∇
≤ |∇ / 2 trχ′ | + |i||∇
/ 2 χ̂′ |)
/ 2 χ̂′ , ∇
Since also |(Ωχ̂, ∇ / 2 χ̂′ )| ≤ CΩ|χ̂||∇
/ 2 χ̂′ |2 , we obtain:
/ 2 χ̂′ |2 ) + 4Ωtrχ|∇
D(|∇ / 2 χ̂′ |2 / 2 χ̂′ |(|s′ | + |h||∇
≤ 2|∇ / 2 trχ′ | + |i||∇
/ 2 χ̂′ |)
/ 2 χ̂′ |2
+2CΩ|χ̂||∇ (5.11)

p
ψ ′ = |∇ / 2 trχ′ |2 + |∇
/ 2 χ̂′ |2 (5.12)
By 5.10, 5.11, and the pointwise bounds 4.80, ψ ′2 satisfies the inequality:
D(ψ ′2 ) + 4Ωtrχψ ′2 ≤ 2CΩ|χ̂|ψ ′2 + 2ψ ′ (|r′ | + |s′ |) (5.13)
We then apply Lemma 4.4 with p = 2, λ = 4, l = C, and ρ = |r′ | + |s′ |, to

obtain: Z un o
kψ ′ kL2 (Su,u ) ≤ C kr′ kL2 (Su′ ,u ) + ks′ kL2 (Su′ ,u ) du′ (5.14)
0
152
In reference to the first of 5.9 we have, for all (u, u) ∈ D′ :
/(Ωχ), /dtrχ′ )kL2 (Su,u ) ≤ Ck∇

k(∇ /trχ′ kL4 (Su,u ) ≤ O(δ −1/2 |u|−4 )
/(Ωχ)kL4 (Su,u ) kd
(5.15)
by Proposition 4.1, and:
/f · /dtrχ′ kL2 (Su,u ) ≤ Ck∇

k∇ /trχ′ kL4 (Su,u ) ≤ O(|u|−5 )
/(Ωtrχ)kL4 (Su,u ) kd
k∇ /χ̂′ kL2 (Su,u ) ≤ Ck∇
/g · ∇ /χ̂′ kL4 (Su,u ) ≤ O(δ −1 |u|−3 ) (5.16)
/(Ωχ̂)kL4 (Su,u ) k∇
by the first two of 4.79 and Proposition 4.1. Also, in view of the fact that from
4.81 we can express:
1
/ 2 log Ω = (∇
∇ /η + ∇/η) (5.17)
2
we obtain, by Proposition 4.1 and the 3rd and 4th of the estimates of Proposition
4.2:
/rkL2 (Su,u ) ≤ O(δ −1/2 |u|−4 )
k∇ (5.18)
Consequently:
kr′ kL2 (Su,u ) ≤ O(δ −1 |u|−3 ) (5.19)
In reference to the second of 5.9 we have, for all (u, u) ∈ D′ :
k(∇
/(Ωχ), ∇/χ̂′ )kL2 (Su,u ) ≤ Ck∇ /χ̂′ kL4 (Su,u ) ≤ O(δ −1 |u|−3 )
/(Ωχ)kL4 (Su,u ) k∇
(5.20)
by Proposition 4.1, and:
/h · /dtrχ′ kL2 (Su,u ) ≤ Ck∇

k∇ /trχ′ kL4 (Su,u ) ≤ O(δ −1/2 |u|−4 )
/(Ωχ̂)kL4 (Su,u ) kd
k∇ /χ̂′ kL2 (Su,u ) ≤ Ck∇
/i · ∇ /χ̂′ kL4 (Su,u ) ≤ O(δ −1 |u|−3 ) (5.21)
/(Ωχ̂)kL4 (Su,u ) k∇
by the third and fourth of 4.79 and Proposition 4.1. Also, in view of 5.17, we
obtain, by Proposition 4.1 and the 3rd and 4th of the estimates of Proposition
4.2:
k∇ / 2 αkL2 (Su,u ) + O(δ −1/2 |u|−4 )
/skL2(Su,u ) ≤ k∇ (5.22)
Consequently:
/ 2 αkL2 (Su,u ) + O(δ −1 |u|−3 )

ks′ kL2 (Su,u ) ≤ k∇ (5.23)
Substituting the estimates 5.19 and 5.23 in 5.14 and noting that:
Z u Z u 1/2
/ 2 αkL2 (Su′ ,u ) du′ ≤ u1/2
k∇ / 2 α|2L2 (Su′ ,u ) du′
k∇ ≤ δ 1/2 k∇
/ 2 αkL2 (Cu )
0 0
we obtain, in view of the definition 5.5,
kψ ′ kL2 (Su,u ) ≤ δ −1/2 |u|−2 R

/2 (α) + O(|u|−3 ) (5.24)
for all (u, u) ∈ D′ . Since
kψ ′ k2L2 (Su,u ) = k∇
/ 2 trχ′ k2L2 (Su,u ) + k∇
/ 2 χ̂′ k2L2 (Su,u )
153
5.24 is the estimate of the proposition.
Let now R /C
ADB be the components of the intrinsic curvature of Su,u with
respect to a Jacobi field frame defined in a neighborhood of a generator of Cu ,
as in the proof of the first part of Lemma 4.1. Then we have:
/)C
(DR /C
ADB = D(RADB ) = ∇ /C
/D DΓBA − ∇ /C
/B DΓDA (5.25)
hence, if Ric /C
/ AB = RACB are the components of the Ricci curvature of Su,u in
the same frame,
(DRic
/ )AB = D(Ric
/ AB ) = ∇ /C
/C DΓBA − ∇ /C
/B DΓCA (5.26)
Since Su,u is 2-dimensional,
/C
R C
ADB = K(δD g
C
/AB −δB / AB = Kg/AB , K = (g/−1 )AB Ric
g/AD ), Ric / AB (5.27)
where K is the Gauss curvature of Su,u . Thus, we have (by the first of 3.5):
1
DK = D((g/−1 )AB Ric
/ AB )
2
1
= −ΩχAB Ric/ AB + (g/−1 )AB (DRic
/ )AB
2
1
= −ΩtrχK + (g/−1 )AB (DRic/ )AB
2
Hence, from 5.26,
1 −1 AB 1 A B
DK + ΩtrχK = (g/ ) ∇ /C
/C DΓAB − ∇/ DΓ
/BA
2 2
Substituting the expression for DΓ of the first part of Lemma 4.1 we then obtain:
/B ∇
DK + ΩtrχK = ∇ /A (ΩχAB ) − △
/ (Ωtrχ)
or:
DK + ΩtrχK = div
/ div
/ (Ωχ) − △
/ (Ωtrχ)
1
= div
/ div
/ (Ωχ̂) − △/ (Ωtrχ) (5.28)
2
This is the propagation equation for the Gauss curvature along the generators
of the Cu .
By 5.2,
4π
DK = − DArea(Su,u )
(Area(Su,u ))2
and we have
Z
DArea(Su,u ) = Ωtrχdµ/g = ΩtrχArea(Su,u ) (5.29)
Su,u
154
hence:
DK = −Ωtrχ K (5.30)
From 5.28 and 5.30 we obtain the following propagation equation for K − K
along the generators of the Cu :
D(K − K) + Ωtrχ(K − K) = −(Ωtrχ − Ωtrχ)K

1
+div / (Ωχ̂) − △
/ div / (Ωtrχ) (5.31)
2
To this equation we apply Lemma 4.6, taking p = 2. Here r = 0, ν = −2, γ = 0.
We obtain:
Z u
kK − KkL2 (Su,u ) ≤ C k(Ωtrχ − Ωtrχ)KkL2 (Su′ ,u ) du′ (5.32)
0
Z u
1
+C kdiv / (Ωχ̂) − △
/ div / (Ωtrχ)kL2 (Su′ ,u ) du′
0 2
From the first two of the bounds 4.76:
2
Ωtrχ − ≤ C|u|−2 (R∞ 2
0 (α)) + O(δ|u|
−2
) (5.33)
|u|
hence also:
2
Ωtrχ − ≤ C|u|−2 (R∞ 2
0 (α)) + O(δ|u|
−2
)
|u|
It follows that:
|Ωtrχ − Ωtrχ| ≤ C|u|−2 (R∞ 2

0 (α)) + O(δ|u|
−2
) (5.34)
Together with 5.2 and the lower and upper bounds for Area(Su,u ) of Lemma
4.3 this implies that:
k(Ωtrχ − Ωtrχ)KkL2 (Su,u ) ≤ O(|u|−3 )
therefore the first integral on the right in 5.32 is bounded by O(δ|u|−3 ). Also,
by Proposition 5.1 together with expression 5.17 and Proposition 4.2:
1
kdiv / (Ωχ̂) − △
/ div / (Ωtrχ)kL2 (Su,u ) ≤ Cδ −1/2 |u|−2 R
/2 (α) + O(|u|−3 )
2
therefore the second integral on the right in 5.32 is bounded by Cδ 1/2 |u|−2 R
/2 (α)+
O(δ|u|−3 ). We thus arrive at the following proposition.
kK − KkL2 (Su,u ) ≤ Cδ 1/2 |u|−2 R

/2 (α) + O(δ|u|−3 )
for all (u, u) ∈ D′ , provided that δ is suitably small depending on D0∞ , R∞

0 .
155
5.2 Sobolev inequalities on S. The isoperimetric
constant
In the following we shall make use of the Sobolev inequalities of Su,u . These
inequalities are all derived from the isoperimetric Sobolev inequality (see [O])
of Su,u : If f is an arbitrary function which is integrable and with integrable
derivative on Su,u , then f is square-integrable on Su,u and we have:
Z Z !2
2
(f − f ) dµ/g ≤ I(Su,u ) |d
/f |dµ/g (5.35)
Su,u Su,u
Here I(Su,u ) is the isoperimetric constant of Su,u :
min{Area(U ), Area(U c )}
I(Su,u ) = sup (5.36)
U (Perimeter(∂U ))2
where the supremum is over all domains U with C 1 boundary ∂U in Su,u , and
U c = Su,u \ U denotes the complement of U in Su,u .
Lemma 5.1 Let (S, g/) be a compact 2-dimensional Riemannian manifold

and ξ a tensorfield on S, of arbitrary type, which is square integrable and with
square integrable first covariant derivative. Then for each 2 < p < ∞ ξ ∈ Lp (S)
and we have:
q
(Area(S)) −1/p
kξkL (S) ≤ Cp I′ (S)kξkW12 (S)
p
Here Cp is a numerical constant depending only on p,
I′ (S) = max{I(S), 1}
where I(S) is the isoperimetric constant of S, and we denote:
/ξkL2 (S) + (Area(S))−1/2 kξkL2 (S)

kξkW12 (S) = k∇
where ∇/ is the covariant derivative operator associated to the metric g/.

Proof: We first establish the lemma in the case of functions φ on S. In reference
to 5.35, which holds for an arbitrary compact 2-dimensional manifold (S, g/),
since
kf kL2 (S) = |f |(Area(S))1/2 , |f | ≤ (Area(S))−1 kf kL1(S) (5.37)
5.35 implies: q
kf kL2(S) ≤ I′ (S)kf kW11 (S) (5.38)
where we denote:
/f kL1(S) + (Area(S))−1/2 kf kL1(S)

kf kW11 (S) = kd (5.39)
156
/df being the differential of f on S.
We now rescale the metric g/ on S, setting:
g̃/ = (Area(S))−1 g/ so that dµ/g̃ = (Area(S))−1 dµ/g (5.40)
to a metric g̃/ of unit area. Taking into account the fact that relative to the new
metric, we have, in arbitrary local coordinates on S,
−1 AB
/φ|/2g̃ = (g̃/
|d ) ∂A φ∂B φ = Area(S)(g/−1 )AB ∂A φ∂B φ = Area(S)|d
/φ|/2g (5.41)
we obtain:
kφkW12 (S,g/) = kφkW 2 (S,/)
g̃ (5.42)
1
and we have:
kφkW 2 (S,/)
g̃ = kd
/φkL2 (S,/)
g̃ + kφkL2 (S,/)
g̃ (5.43)
1
Moreover, 5.38 and 5.39 read relative to g̃/:

q
kf kL2 (S,/)
g̃ ≤ I′ (S)kf kW 1 (S,/)
g̃ (5.44)
1
(the isoperimetric constant is invariant under rescalings) and:
kf kW 1 (S,/)
g̃
(5.45)
1
We now set:
1 |φ|
φ̃ = p ′ (5.46)
I (S) kφkW12 (S,/)
g̃
Then φ̃ ≥ 0 and:
1
kφ̃kW 2 (S,/)
g̃ =
p′ (5.47)
1
I (S)
Taking f = φ̃k , k > 1, in 5.44 we obtain:
q
kφ̃k kL2 (S,/)
g̃ ≤ I′ (S)kφ̃k kW 1 (S,/)
g̃ (5.48)
1
Since /d(φ̃k ) = k φ̃k−1/dφ̃ we have:
/(φ̃k )kL1 (S,/)

kd g̃ ≤ kkφ̃
k−1
kL2 (S,/)
g̃ kd
/φ̃kL2 (S,/)
g̃
and:
k−1 k−1
kφ̃kL1 (S,/)
g̃ = kφ̃ φ̃kL1 (S,/)
g̃ ≤ kφ̃ kL2 (S,/)
g̃ kφ̃kL2 (S,/)
g̃
hence, adding,
kφ̃k kW 1 (S,/)
g̃ ≤ kkφ̃
k−1
kL2 (S,/)
g̃ kφ̃kW 2 (S,/)
g̃ (5.49)
1 1
Substituting 5.47 in 5.49 and the result in 5.48 yields:
kφ̃k kL2 (S,/)

g̃ ≤ kkφ̃
k−1
kL2 (S,/)
g̃
157
which is equivalent to:
1/k 1−(1/k)
kφ̃kL2k (S,/)
g̃ ≤ k kφ̃k (5.50)
L2(k−1) (S,/)
g̃
We now set:
k = 2, 3, 4, ...
For k = 2 we have 2(k − 1) = 2 and taking f = φ̃ in 5.44 we obtain, by 5.46 and
5.47,
1 kφkL2 (S,/)
g̃
kφkW 1 (S,/)
g̃
1
kφ̃kL2 (S,/)
g̃ = √ ′ ≤ ≤1 (5.51)
I (S) kφk 2
W (S,/)
g̃ kφk 2
W (S,/)
g̃
1 1
From 5.50 we then obtain:

1/2
kφ̃kL4 (S,/)
g̃ ≤ 2
and in general, for any k = 2, 3, 4, ... the recursive inequality 5.50 implies:
kφ̃kL2k (S,/)
g̃ ≤ Ck (5.52)
where Ck is defined by the recursion:

Ck = k 1/k Ck−1 , C1 = 1
hence:
k
Y
Ck = j 1/j (5.53)
j=2
Moreover, since (S, g̃/) has unit area the norm kf kLp(S,/)
g̃ for a given function f
is a non-decreasing function of the exponent p. Therefore 5.52 implies that for
any 2 < p < ∞:
kφ̃kLp (S,/)
g̃
≤ Cp′ (5.54)
where Cp′ = Ck with k the smallest integer such that p ≤ 2k.
In view of the definition 5.46, the inequality 5.54 is equivalent to:
q
kφkLp (S,/)
g̃ ≤ Cp
′
I′ (S)kφkW 2 (S,/)
g̃ (5.55)
1
Morever, by 5.42 and the fact that:

−1/p
kφkLp (S,/)
g̃ = (Area(S)) kφkLp (S,g/) (5.56)
the inequality 5.55 is equivalent to:

q
(Area(S))−1/p kφkLp (S,g/) ≤ Cp′ I′ (S)kφkW12 (S,g/) (5.57)
This establishes the lemma in the case of functions φ. To deduce the lemma for
tensorfields ξ of arbitrary type, we set:
q
φ = ε2 + |ξ|/2g (5.58)
158
where ε is a positive constant. We then have:
(ξ, ∇
/ξ)/g
/dφ = q
ε2 + |ξ|/2g
hence:
|d
/φ|/g ≤ |∇
/ξ|/g (5.59)
The general form of the lemma then follows taking the limit ε → 0.
Lemma 5.2 Let (S, g/) be a compact 2-dimensional Riemannian manifold

and ξ a tensorfield on S, of arbitrary type, which belongs to Lp (S) and with
first covariant derivative which also belongs to Lp (S), for some p > 2. Then
ξ ∈ L∞ (S) and we have:
p
sup |ξ| ≤ Cp I ′ (S)(Area(S))(1/2)−(1/p) kξkW1p (S)
S
Here Cp is a numerical constant depending only on p, and we denote:
/ξkLp(S) + (Area(S))−1/2 kξkLp (S)

kξkW1p (S) = k∇
Proof: The proof is an adapation of an argument found in [G-T]. We first

establish the lemma in the case of functions φ on S. Following the proof of
Lemma 5.1 we now set:
1 |φ|
φ̃ = p ′ (5.60)
I (S) kφkW1p (S,/)
g̃
Then φ̃ ≥ 0 and:
1
kφ̃kW p (S,/)
g̃ =
p′ (5.61)
1
I (S)
Taking f = φ̃k , k > 1 in 5.44 we obtain:
q
kφ̃k kL2 (S,/)
g̃ ≤ I′ (S)kφ̃k kW 1 (S,/)
g̃ (5.62)
1
Since /d(φ̃k ) = k φ̃k−1/dφ̃, we now have, by Hölder’s inequality:
/(φ̃k )kL1 (S,/)

kd g̃ ≤ kkφ̃
k−1
kLp′ (S,/)
g̃ kd
/φ̃kLp (S,/)
g̃ (5.63)
and:
kφ̃k kL1 (S,/)

g̃ = kφ̃
k−1
φ̃kL1 (S,/)
g̃ ≤ kφ̃
k−1
kLp′ (S,/)
g̃ kφ̃kLp (S,/)
g̃ (5.64)
where p′ is the conjugate exponent:

1 1
+ =1 (5.65)
p p′
159
Hence, adding,
kφ̃k kW 1 (S,/)
g̃ ≤ kkφ̃
k−1
kLp′ (S,/)
g̃ kφ̃kW p (S,/)
g̃ (5.66)
1 1
Substituting 5.61 in 5.66 and the result in 5.62 then yields:
kφ̃k kL2 (S,/)

g̃ ≤ kkφ̃
k−1
kLp′ (S,/)
g̃
which is equivalent to:

1/k 1−(1/k)
kφ̃kL2k (S,/)
g̃ ≤ k kφ̃k (5.67)
Lp′ (k−1) (S,/)
g̃
This implies:
1/k 1−(1/k)
kφ̃kL2k (S,/)
g̃ ≤ k kφ̃k (5.68)
Lp′ k (S,/)
g̃
for, by virtue of the fact that S has unit area with respect to g̃/, the norm
kf kLp(S,/)
g̃ for a given function f is a non-decreasing function of the exponent
p. The ratio of the exponent on the left in 5.68 to the exponent on the right is
2/p′ > 1.
We now set: n
2
k= : n = 1, 2, 3, ... (5.69)
p′
For n = 1 the exponent on the right in 5.68 is 2, and taking f = φ in 5.44 we
obtain, by 5.61:
1 kφkL2 (S,/)
g̃
kφkW 1 (S,/)
g̃
1
kφ̃kL2 (S,/)
g̃ = p ′
≤ ≤1 (5.70)
I (S) kφk p
W1 (S,/)
g̃ kφk p
W1 (S,/)
g̃
It follows from 5.68 - 5.70, by induction on n, that for every n = 1, 2, 3, ...:

Pn m(2/p′ )−m
2 m=1
kφ̃kL2(2/p′ )n (S,/)
g̃ ≤ ′
(5.71)
p
Taking the limit n → ∞ we then obtain:
sup(φ̃) ≤ Cp (5.72)
S
where Cp is the constant:
P∞ (2/p′ )
m(2/p′ )−m
2 m=1 2 ((2/p′ )−1)2
Cp = = (5.73)
p′ p′
In view of the definition 5.60, 5.73 is equivalent to:

q
sup |φ| ≤ Cp I′ (S)kφkW p (S,/)
g̃ (5.74)
S 1
160
Now, by 5.40, 5.41, we have:
kφkW p (S,/)
g̃ = kd
/φkLp (S,/)
g̃ + kφkLp (S,/)
g̃
1
= (Area(S))(1/2)−(1/p) {kd
/φkLp(S,g/) + (Area(S))−1/2 kφkLp(S,g/) }
= (Area(S))(1/2)−(1/p) kφkW1p (S,g/) (5.75)
We thus conclude that:

q
sup |φ| ≤ Cp I′ (S)(Area(S))(1/2)−(1/p) kφkW1p (S,g/) (5.76)
S
This establishes the lemma in the case of functions φ. To deduce the lemma for
tensorfields ξ of arbitrary type, we define φ according to 5.58. Then in view of
5.59 the general form of the lemma follows taking the limit ε → 0.
We see from Lemmas 5.1 and 5.2 that the Sobolev inequalities of Su,u thus
involve the isoperimetric constant I(Su,u ) of Su,u . Consequently, to make use of
these inequalities we must first obtain an upper bound for I(Su,u ). Considering
a given Cu , the derivation of this upper bound shall be based on upper and
lower bounds for the eigenvalues of the metric g/(u) = Φ∗u g/ with respect to
Su,u
the metric g/(0) = g/|S0,u .
From 4.32 - 4.34 and the first of 1.41 we have:
∂
g/(u) = 2(Ωχ)(u) (5.77)
∂u
We consider the eigenvalues of g/(u) with respect to g/(0). Let λ(u) be the
smallest eigenvalue and Λ(u) be the largest eigenvalue.
Lemma 5.3 Under the assumptions of Proposition 3.1 we have:
λ(u) ≥ 1 − O(δ 1/2 |u|−1 ), Λ(u) ≤ 1 + O(δ 1/2 |u|−1 )
for all (u, u) ∈ D′ .

Proof: Let us define:
dµ/g(u) p
µ(u) = = λ(u)Λ(u) (5.78)
dµ/g(0)
and: s
1 Λ(u) Λ(u)
ν(u) = sup g/(u)(X, X) = = (5.79)
µ(u) |X|g/(0) =1 µ(u) λ(u)
We have:
∂ 1 ∂
µ(u) = tr/g(u) g/(u) µ(u) = (Ωtrχ)(u)µ(u) (5.80)
∂u 2 ∂u
161
Integrating with respect to u and noting that µ(0) = 1 yields:
Z u
′ ′
µ(u) = exp (Ωtrχ)(u )du (5.81)
0
In view of the second of the bounds 4.76 we then obtain:
|µ(u) − 1| ≤ O(δ|u|−1 ) (5.82)
To estimate ν(u) we set:

g/(u)
ĝ/(u) = (5.83)
µ(u)
Then according to the definition 5.79:
sup ĝ/(u)(X, X) = ν(u) (5.84)

|X|g
/(0) =1

∂ 1 ∂ g/(u) ∂
ĝ/(u) = g/(u) − µ(u)
∂u µ(u) ∂u (µ(u))2 ∂u
2 g/(u)
= (Ωχ)(u) − (Ωtrχ)(u)
µ(u) µ(u)
or, simply:
∂ 2
ĝ/(u) = (Ωχ̂)(u) (5.85)
∂u µ(u)
Consider any tangent vector X to S0,u such that |X|/g(0) = 1. Then:
∂ 2
ĝ/(u)(X, X) = (Ωχ̂)(u)(X, X) (5.86)
∂u µ(u)
Integrating with respect to u and noting that ĝ/(0)(X, X) = 1 yields:

Z u
(Ωχ̂)(u′ )(X, X) ′
ĝ/(u)(X, X) = 1 + 2 du (5.87)
0 µ(u′ )
hence: u
|(Ωχ̂)(u′ )(X, X)| ′
Z
ĝ/(u)(X, X) ≤ 1 + 2 du (5.88)
0 µ(u′ )
Now if ξ is an arbitrary type Tpq tensorfield on S0,u then, in components with
respect to an arbitrary local frame field on S0,u ,
A ...A B ...B
|ξ|/2g(u) = (g/(u))A1 B1 ...(g/(u))Aq Bq (g/−1 (u))C1 D1 ...(g/−1 (u))Cp Dp ξC11...Cpq ξD11 ...Dqp
(5.89)
It follows that:
(λ(u))q (Λ(u))−p |ξ|/2g(0) ≤ |ξ|/2g(u) ≤ (Λ(u))q (λ(u))−p |ξ|/2g(0) (5.90)
162
In particular, taking ξ = (Ωχ̂)(u) (q = 0, p = 2), we have:
|(Ωχ̂)(u)|/g(u) ≥ (Λ(u))−1 |(Ωχ̂)(u)|/g(0) (5.91)
hence:
|(Ωχ̂)(u)(X, X)| ≤ |(Ωχ̂)(u)|/g(0) |X|/2g(0) = |(Ωχ̂)(u)|/g(0) ≤ Λ(u)|(Ωχ̂)(u)|/g(u)

(5.92)
Substituting in 5.88, taking the supremum over X ∈ Tq S0,u such that |X|/g(0) = 1
at each q ∈ S0,u , and recalling 5.79 and 5.84 we obtain the following linear
integral inequality for ν(u):
Z u
ν(u) ≤ 1 + 2 |(Ωχ̂)(u′ )|/g(u′ ) ν(u′ )du′ (5.93)
0
which implies: Z u
′ ′
ν(u) ≤ exp 2 |(Ωχ̂)(u )|/g(u′ ) du (5.94)
0
In view of the third of the bounds 4.76 we then obtain (recall that by the
definition 5.79 ν(u) ≥ 1):
1 ≤ ν(u) ≤ 1 + O(δ 1/2 |u|−1 ) (5.95)
Since from the definitions 5.78, 5.79 we can express:
µ(u)
λ(u) = , Λ(u) = µ(u)ν(u) (5.96)
ν(u)
the lemma follows from the bounds 5.82 and 5.95.
Lemma 5.4 Under the assumptions of Proposition 3.1 we have:

1
I(Su,u ) ≤
π
for all (u, u) ∈ D′ , provided that δ is suitably small depending on R∞ 0 (α). In
particular, the constant I′ (Su,u ) appearing in the Sobolev inequalities, Lemmas
5.1 and 5.2 is equal to 1.
Proof: A domain Uu ⊂ Su,u with C 1 boundary ∂Uu is the image by Φu of a
domain U0 ⊂ S0,u with C 1 boundary ∂U0 .
We have:
Z Z Z
Perimeter(∂Uu ) = ds = ds(u), Perimeter(∂U0 ) = ds(0) (5.97)
∂Uu ∂U0 ∂U0
Here ds is the element of arc length of ∂Uu with respect to the metric g/|Su,u on
Su,u , while ds(u) is the element of arc length of ∂U0 with respect to the metric
163
g/(u) on S0,u and ds(0) the element of arc length of ∂U0 with respect to the
metric g/(0) = g/|S0,u on S0,u . Now:
p
ds(u) ≥ λ(u)ds(0) (5.98)
hence: p
Perimeter(∂Uu ) ≥ inf λ(u)Perimeter(∂U0 ) (5.99)
S0,u
Lemma 5.3 then implies:

Perimeter(∂Uu )
≥ 1 − O(δ 1/2 |u|−1 ) (5.100)
Perimeter(∂U0 )
provided that δ is suitably small depending on R∞ 0 (α) (recall the second and
third of the bounds 4.76 which come from Proposition 3.1).
On the other hand we have:
Z Z Z
Area(Uu ) = dµ/g = dµ/g(u) , Area(U0 ) = dµ/g(0) (5.101)
Uu U0 U0
where dµ/g is the area element of Su,u with respect to the induced metric g/|Su,u ,
while dµ/g(u) is the area element on S0,u induced by the metric g/(u) and dµ/g(0)
the area element on S0,u induced by the metric g/(0) = g/|S0,u . By 5.78:
dµ/g(u) = µ(u)dµ/g(0) (5.102)
hence:
Area(Uu ) ≤ sup µ(u)Area(U0 ) (5.103)
S0,u
The bound 5.82 then implies:

Area(Uu )
≤ 1 + O(δ|u|−1 ) (5.104)
Area(U0 )
The inequalities 5.100 and 5.104 together imply:
Area(Uu ) Area(U0 )
≤ (1 + O(δ 1/2 |u|−1 ) (5.105)
(Perimeter(∂Uu ))2 (Perimeter(∂U0 ))2
provided that δ is suitably small depending on R∞0 (α) (recall the second and
third of the bounds 4.76 which come from Proposition 3.1). Similarly we obtain:
Area(Uuc ) Area(U0c )
≤ (1 + O(δ 1/2 |u|−1 ) (5.106)
(Perimeter(∂Uu ))2 (Perimeter(∂U0 ))2
Therefore:
min{Area(Uu ), Area(Uuc )} min{Area(Uu ), Area(Uuc )}
≤ (1 + O(δ 1/2 |u|−1 ))
(Perimeter(∂Uu ))2 (Perimeter(∂Uu ))2
(5.107)
164
Taking the supremum over all domains Uu with C 1 boundary ∂Uu in Su,u we
conclude that:
I(Su,u ) ≤ (1 + O(δ 1/2 |u|−1 ))I(S0,u ) (5.108)
Now:
1
I(S0,u ) = (5.109)
2π
S0,u being a round sphere in Euclidean space, the supremum being achieved for
a hemisphere. The lemma then follows.
5.3 The uniformization theorem

We now turn to the uniformization problem. Given a 2-dimensional compact
Riemannian manifold (S, g/) with S diffeomorphic to S 2 the problem is to find
a suitable function φ such that the conformally related metric
◦
g = e−2φ g/ (5.110)
◦ ◦
has Gauss curvature K = 1, so that (S, g/) is isometric to the standard sphere
(the unit sphere in Euclidean space). In [C-K] a solution to this problem was
obtained in the form of the following proposition (Lemma 2.3.2 of [C-K]).
Proposition 5.3 Let (S, g/) be a 2-dimensional compact Riemannian manifold

with S diffeomorphic to S 2 . Then there exists a conformal transformation of
the metric
◦
g/= Ω2 g/
◦
such that K = 1. Moreover the conformal factor Ω can be chosen such that the
quantities Ω−1
m , ΩM , Ω1 have upper bounds which depend only on upper bounds
for
diam(S)
sup r2 |K|,
S r
Here, we denote: r
Area(S)
r=
4π
Ωm = inf rΩ, ΩM = sup rΩ
S S
−2
Ω1 = sup Ω |d
/Ω|
S
Also, for every 2 ≤ p < ∞ the quantity

Z 1/p
Ω2,p = Ω−3p+2 |∇
/ 2 Ω|p
S
165
has a bound which depends only on upper bounds for
diam(S)
sup r2 |K|,
S r
and on p. In addition, if with
km = inf r2 K, kM = sup r2 K
S S
we have km > 0 then the bounds on Ω−1 m , ΩM , Ω1 depend only on upper bounds
−1 −1
for km , kM , and the bound on Ω2,p depends only on upper bounds for km , kM
and on p.
As we have remarked in the introduction to the present chapter, we cannot

directly apply here the above proposition, because we do not possess the ap-
propriate L∞ bounds on K. We proceed instead as follows. We first define the
function ω to be the solution of
/ /g ω = K − K
△ such that ω = 0 (5.111)
Then the conformally related metric
g/′ = e2ω g/ (5.112)
has Gauss curvature K ′ given by:
K ′ = e−2ω (K − △
/ /g ω) = e−2ω K (5.113)
Since, according to 5.2,

1
K= (5.114)
r2
we shall obtain the required L∞ bounds for K ′ once we derive an appropriate
L∞ bound for ω.
Lemma 5.5 We have, for all (u, u) ∈ D′ :

1
sup |ω| ≤ log 2
Su,u 2
provided that δ is suitably small depending on D0∞ , R∞ /41 , R

0 , D /41 , and R
/2 (α).
Proof: We start from the following basic integral identity for the Laplacian △ / /g
acting on functions φ on the compact 2-dimensional manifold (S, g/) (see Chapter
2 of [C-K]): Z Z
/ 2 φ|2 + K|d
{|∇ /φ|2 }dµ/g = / φ|2 dµ/g
|△ (5.115)
S S
We apply this to the function ω obtaining, by virtue of equation 5.111,
Z Z
2 2 2
{|∇/ ω| + K|d /ω| }dµ/g = |K − K|2 dµ/g (5.116)
S S
166
Writing K = K + K − K we estimate:
Z
/ω|2 ≤ kK − KkL2 (S) kd
|K − K||d /ωk2L4 (S) (5.117)
S
Now, by Lemma 5.1 with p = 4 applied to /dω and Lemma 5.4:
kd / 2 ωk2L2 (S) + r−2 kd

/ωk2L4 (S) ≤ Cr{k∇ /ωk2L2(S) } (5.118)
In view of 5.114, 5.116 then implies:

2 2
1 − CrkK − KkL2 (S) k∇ / ωkL2 (S) + r−2 kd
/ωk2L2(S) ≤ kK − Kk2L2 (S)
(5.119)
Now, by Proposition 5.2:
r2 kK − KkL2 (S) ≤ Cδ 1/2 R

/2 (α) + O(δ) (5.120)
It follows that if δ is suitably small depending on D0∞ , R∞ /41 , R

0 , D /41 , and R
/2 (α),
then, in reference to the factror on the left in 5.119 we have:
1
CrkK − KkL2 (S) ≤
2
and 5.119 implies:
/ 2 ωk2L2 (S) + r−2 kd

k∇ /ωk2L2(S) ≤ 2kK − Kk2L2 (S) (5.121)
We now apply Lemma 5.1 with p = q to /dω. In view of Lemma 5.4 we obtain:
kd / 2 ωk2L2 (S) + r−2 kd

/ωk2Lq (S) ≤ Cq r4/q {k∇ /ωkL2(S) } (5.122)
Here q is any real number greater than 2. Substituting 5.121 then yields:
kd
/ωkLq (S) ≤ Cq r2/q kK − KkL2 (S) (5.123)
≤ Cq r(2/q)−2 {Cδ 1/2 R
/2 (α) + O(δ)}
by 5.120. Next, we apply Lemma 5.1 with p = q to ω to obtain:
/ωk2L2(S) + r−2 kωk2L2(S) }

kωk2Lq (S) ≤ Cq r4/q {kd (5.124)
Now, since ω = 0 the isoperimetric inequality 5.35 reduces in the case of ω to:
kωk2L2 (S) ≤ I(S)kd

/ωk2L1(S) (5.125)
Since
/ωk2L1(S) ≤ Area(S)kd
kd /ωk2L2(S)
5.125 implies:
r−2 kωk2L2 (S) ≤ 4πI(S)kd
/ωk2L2 (S) (5.126)
167
Substituting this in 5.124 then yields, in view of Lemma 5.4,
kωk2Lq (S) ≤ Cq r4/q kd

/ωk2L2(S) (5.127)
Hence, from 5.121:
kωkLq (S) ≤ Cq r(2/q)+1 kK − KkL2 (S) (5.128)

(2/q)−1 1/2
≤ Cq r {Cδ R
/2 (α) + O(δ)} (5.129)
Combining 5.123 with 5.129 we obtain:
kωkW1q (S) ≤ Cq r(2/q)−2 {Cδ 1/2 R

/2 (α) + O(δ)} (5.130)
This holds for any 2 < q < ∞. Taking then q = 4 and applying Lemma 5.2
with p = q we obtain, in view of Lemma 5.4,
sup |ω| ≤ Cr−1 {Cδ 1/2 R

/2 (α) + O(δ)} (5.131)
S
The lemma then follows.
According to the above lemma we have:

1 √
√ ≤ eω ≤ 2 (5.132)
2
It follows that with
r r
Area(S, g/) ′ Area(S, g/′ )
r= , r =
4π 4π
we have:
1 r′ √
√ ≤ ≤ 2 (5.133)
2 r
Since according to 5.113, 5.114
e−2ω
K′ = (5.134)
r2
the inequalities 5.132 and 5.133 imply:
1
≤ r′2 K ′ ≤ 4 (5.135)
4
Therefore the quantities
′
km = inf r′2 K ′ , ′
kM = sup r′2 K ′ (5.136)
S S
satisfy the bounds:

′ 1 ′
km ≥ , kM ≤4 (5.137)
4
168
We can then apply Proposition 5.3 to the Riemannian manifold (S, g/′ ) to con-
clude that there exists a conformal factor Ω′ such that the metric
◦
g/= Ω′2 g/′ (5.138)
◦
has Gauss curvature K = 1 and the quantities Ω′−1 ′ ′
m , ΩM , Ω1 , where
Ω′m = inf r′ Ω′ , Ω′M = sup r′ Ω′ , (5.139)

S S
Ω′1 = sup Ω′−2 |d

/Ω′ |/g′ , (5.140)
S
are bounded by numerical constants, while for every 2 ≤ p < ∞ the quantity
Z 1/p
Ω′2,p = Ω ′−3p+2
|∇
/ ′2
Ω′ |/pg′ dµ/g′ (5.141)
S
is bounded by a numerical constant depending only on p.

Setting ′ ′
Ω′ = e−φ = r′−1 e−ψ (5.142)
we have: ′ ′
Ω′m = inf e−ψ , Ω′M = sup e−ψ (5.143)
S S
therefore the bounds on Ω′m , Ω′M are equivalent to a bound for
sup |ψ ′ | (5.144)
S
Moreover, we have: ′
Ω′−2 |d
/Ω′ |/g′ = r′ eψ |d
/ψ ′ |/g′ (5.145)
therefore, modulo the bound for 5.144, the bound for Ω′1 is equivalent to a bound
for
sup r′ |d
/ψ ′ |/g′

(5.146)
S
Finally, by 5.142 we have:

Z 1/p
/ ′2 ψ ′ − /dψ ′ ⊗ /dψ ′ |/pg′ dµ/g′
′
Ω′2,p = r′2−(2/p) e(2p−2)ψ |∇ (5.147)
S
It follows that, modulo the bounds for 5.146 and 5.144, the bound for Ω′2,p is
equivalent to a bound for
r′2−(2/p) k∇
/ ′2 ψ ′ kLp (S,g/′ ) (5.148)
Combining the conformal transformations 5.112 and 5.138 we conclude that:

◦ ◦
g/ = e2φ g/= r2 e2ψ g/ (5.149)
169
where:
φ = φ′ − ω or ψ = log(r′ /r) + ψ ′ − ω (5.150)
Lemma 5.5 together with the inequalities 5.133 and the bound for 5.144 then
yield a bound for
sup |ψ| (5.151)
S
by a numerical constant.
Let us define:
k1,q = r1−(2/q) kd
/ψkLq (S,g/) (5.152)
Since by 5.150
/dψ = /dψ ′ − /dω, (5.153)
and by the bound for 5.146 and the inequalities 5.132 and 5.133
r1−(2/q) kd
/ψ ′ kLq (S,g/) ≤ C,
while by the estimate 5.123
r1−(2/q) kd
/ωkLq (S,g/) ≤ Cq r−1 {Cδ 1/2 R
/2 (α) + O(δ)}
≤ Cq′ , (5.154)
the last step provided that δ satisfies the smallness condition of Lemma 5.5, we
conclude that, under the assumptions of Lemma 5.5, for every 2 ≤ q < ∞ k1,q
is bounded by a numerical constant depending only on q. We have thus arrived
at the following proposition.
Proposition 5.4 Let the assumptions of Lemma 5.5 hold. Then for each
(u, u) ∈ D′ the induced metric g/ on Su,u can be expressed in the form:
◦
g/ = r2 e2ψ g/
◦ ◦ ◦
where the metric g/ has Gauss curvature K = 1, thus (Su,u , g/) is isometric to
the standard sphere. Moreover there is numerical constant C, and, for each
2 ≤ q < ∞ a numerical constant Cq depending only on q, such that for all
(u, u) ∈ D′ :
sup |ψ| ≤ C
Su,u
and:
k1,q := r1−(2/q) kd
/ψkLq (S,g/) ≤ Cq
where: r
Area(Su,u , g/)
r=
4π
We shall now proceed to derive a bound for the quantity

k1,∞ = r sup |d
/ψ|/g (5.155)
S
170
as well as for the quantity k2,4 , where, for any 1 ≤ p ≤ ∞:
k2,p = r2−(2/p) k∇
/ 2 ψkLp (S,g/) , (5.156)
under the assumption that K ∈ L4 (S).

Now we already have a bound for 5.146 by a numerical constant, and, in
view of the inequalities 5.132 and 5.133, this is equivalent to a bound for
/ψ ′ |/g
r sup |d (5.157)
S
by a numerical constant. Therefore, in view of 5.153, we shall obtain a bound

for k1,∞ once we derive a bound for
r sup |d
/ω|/g (5.158)
S
Also, for any 2 ≤ p < ∞ we already have a bound for 5.148 by a numerical
constant depending only on p, and, in view of the inequalities 5.132 and 5.133,
this is equivalent to a bound for
r2−(2/p) k∇
/ ′2 ψ ′ kLp (S,g/) (5.159)
for any 2 ≤ p < ∞ by a numerical constant depending only on p. Now, we have:
/ ′2 ψ ′ = ∇
∇ / 2 ψ ′ − △′ · /dψ ′ (5.160)
where △′ = Γ /′ − Γ
/ is the difference of the connections associated to g/′ and g/, a
type T21 tensorfield on S, symmetric in the lower entries, and expressed in local
coordinates on S by:
△′C C C
/−1 )CD g/AB ∂D ω
AB = δA ∂B ω + δB ∂B ω − (g (5.161)
We have, pointwise:
|△′ |/g ≤ C|d
/ω|/g (5.162)
hence:
/ 2 ψ ′ |/g ≤ |∇
|∇ / ′2 ψ ′ |/g + C|d /ψ ′ |/g
/ω|/g |d (5.163)
In view of the bound for 5.157 it follows that:
/ 2 ψ ′ |Lp (S,g/) ≤ k∇
k∇ / ′2 ψ ′ kLp (S,g/) + Cr−1 kd
/ωkLp(S,g/) (5.164)
Thus, in view of 5.154 with q = p, the bound for 5.159 implies a bound for
r2−(2/p) k∇
/ 2 ψ ′ kLp (S,g/) (5.165)
for every 2 ≤ p < ∞ by a numerical constant depending only on p. Therefore,

in view of 5.153 we shall obtain a bound for k2,4 once we derive a bound for:
r3/2 k∇
/ 2 ωkL4 (S,g/) (5.166)
171
To derive bounds for 5.158 and 5.166 in terms of kK −KkL4 (S,g/) , we consider
again equation 5.111. Now, the equation
△
/ /g ω = ρ (5.167)
◦
for a function ω on S is conformally covariant; that is, if g/ = e2φ g/ then ω
satisfies: ◦ ◦
/ ◦ ω =ρ where ρ= e2φ ρ
△ (5.168)
/
g
◦
Now, with φ = log r + ψ and ψ being as in Proposition 5.4, (S, g/) is isometric to
the standard sphere, therefore the standard Lp Calderon-Zygmund inequalities
◦
hold on (S, g/):
◦ ◦
/ 2 ωk
k∇ ◦ + kd
/ωk ◦ ≤ Cp k ρ k ◦ (5.169)
Lp (S,/
g) Lp (S,/
g) Lp (S,/
g)
for any 1 < p < ∞, the constant Cp depending only on p.

We have:
◦
| ρ |p dµ◦ = e(2p−2)φ |ρ|p dµ/g = r2p−2 e(2p−2)ψ |ρ|p dµ/g (5.170)
/
g
hence: ◦
kρk ◦ ≤ r2−(2/p) e(2−(2/p))ψM kρk ◦ (5.171)
Lp (S,/
g) Lp (S,/
g)
Here and in the following we denote:
ψM = sup ψ, ψm = inf ψ, oscψ = ψM − ψm (5.172)

S S
If ξ is a q-covariant tensorfield on S, then
|ξ|/pg dµ/g = e−(pq−2)φ |ξ|p◦ dµ◦ = r−(pq−2) e−(pq−2)ψ |ξ|p◦ dµ◦ (5.173)
/
g /
g /
g /
g
It follows that if pq ≥ 2 we have:
r−(q−(2/p)) e−(q−(2/p))ψM kξk ◦

Lp (S,/
g)
≤ kξkLp (S,g/) ≤ (5.174)
−(q−(2/p)) −(q−(2/p))ψm
r e kξk ◦
Lp (S,/
g)
◦
/ 2 ω (q = 2) we obtain, for
Applying this to ξ = /dω (q = 1) and to ξ =∇
2 ≤ p < ∞,
/ωkLp(S,g/) ≤ r−(1−(2/p)) e−(1−(2/p))ψm kd

kd /ωk ◦ (5.175)
Lp (S,/
g)
172
◦ ◦2
/ 2 ωkLp (S,g/) ≤ r−(2−(2/p)) e−(2−(2/p))ψm k ∇
k∇ / ωk ◦ (5.176)
Lp (S,/
g)
Now, we have:
◦ ◦
/ 2 ω =∇
∇ / 2 ω+ △ ·d
/ω (5.177)
◦ ◦ ◦
where △=Γ / −Γ / is the difference of the connections associated to g/ and g/, a
type T21 tensorfield on S, symmetric in the lower entries, and expressed in local
coordinates on S by:
◦
△C AB = C
−δA C
∂B φ − δB ∂B φ + (g/−1 )CD g/AB ∂D φ
C C
= −δA ∂B ψ − δB ∂B ψ + (g/−1 )CD g/AB ∂D ψ (5.178)
We have, pointwise:
◦
| △ |/g ≤ C|d
/ψ|/g (5.179)
It follows that:
◦
k △ ·d
/ωkLp(S,g/) ≤ Ckd
/ψkL2p(S,g/) kd
/ωkL2p(S,g/)
= Cr−1+(1/p) k1,2p kd
/ωkL2p(S,g/) (5.180)
From 5.177, 5.176, 5.169 and 5.171 we then obtain:
k∇/ 2 ωk p ≤ C e(2−(2/p))oscψ kρk
L (S,g
/) p Lp (S,g
/)
−1+(1/p)
+Cr k1,2p kd
/ωkL2p(S,g/) (5.181)
We apply this to equation 5.111 taking p = 4. Here ρ = K − K and from
Proposition 5.4 we have:
sup |ψ| ≤ C, k1,8 ≤ C (5.182)
S
while from 5.123 with q = 8:

kd
/ωkL8 (S,g/) ≤ Cr−7/4 {Cδ 1/2 R
/2 (α) + O(δ)}
′
≤ C (5.183)
0 , /41 ,
D /41 ,
R and R
/2 (α).
We then obtain:
/ 2 ωkL4 (S,g/) ≤ Cr3/2 kK − KkL4 (S,g/) + Cr−1
r3/2 k∇ (5.184)
Combining 5.184 with the bound for 5.165 for p = 4 yields:
k2,4 ≤ C{1 + r3/2 kK − KkL4 (S,g/) } (5.185)
Also, by Lemma 5.2 with p = 4 applied to /dω, Lemma 5.4, and the estimates
5.123 with q = 4 and 5.184 we obtain:
/ω| ≤ Cr3/2 kK − KkL4 (S,g/) + Cr−1
r sup |d (5.186)
S
hence, combining with the bound for 5.157 yields:

k1,∞ ≤ C{1 + r3/2 kK − KkL4 (S,g/) } (5.187)
173
5.4 Lp elliptic theory on S
Consider now the equation
div
/ /g θ = f (5.188)
for a trace-free symmetric 2-covariant tensorfield θ on S. Here f is a given
1-form on S. This is an elliptic equation for θ. In fact, the following integral
identity holds (see Chapter 2 of [C-K]):
Z Z
/θ|/g + 2K|θ|/g }dµ/g = 2 |f |/2g dµ/g
{|∇ 2 2
(5.189)
S S
◦
Equation 5.188 is conformally covariant; that is, if g/ = e2φ g/ then θ satisfies:
◦ ◦
div
/ ◦ θ =f where f = e2φ f (5.190)
/
g
◦
Recall that with φ = log r + ψ and ψ being as in Proposition 5.4, (S, g/) is iso-
metric to the standard sphere. The operator div/ ◦ from trace-free symmetric
/
g
2-covariant tensorfields to 1-forms on the standard sphere being a first order
elliptic operator with vanishing kernel, the following Lp Calderon-Zygmund in-
◦
equalities hold on (S, g/):
◦ ◦
k∇
/ θk ◦ + kθk ◦ ≤ Cp k f k ◦ (5.191)
Lp (S,/
g) Lp (S,/
g) Lp (S,/
g)
and:
◦ ◦◦ ◦
/ 2 θk
k∇ ◦ ≤ Cp {k ∇
/f k ◦ +kf k ◦ } (5.192)
Lp (S,/
g) Lp (S,/
g) Lp (S,/
g)
for any 1 < p < ∞, the constants Cp depending only on p.

◦ ◦
Applying 5.174 to ξ = θ (q = 2), to ξ =∇ / 2 θ (q = 4),
/ θ (q = 3), and to ξ =∇
we obtain:
kθkLp (S,g/) ≤ r−(2−(2/p)) e−(2−(2/p))ψm kθk ◦

Lp (S,/
g)
◦ ◦
k∇
/ θkLp (S,g/) ≤ r−(3−(2/p)) e−(3−(2/p))ψm k ∇
/ θk ◦
Lp (S,/
g)
◦ ◦
/ 2 θkLp (S,g/)
k∇ / 2 θk
≤ r−(4−(2/p)) e−(4−(2/p))ψm k ∇ ◦ (5.193)
Lp (S,/
g)
◦ ◦◦
Also, applying 5.174 to ξ =f (q = 1) and to ξ =∇
/f (q = 2) we obtain, for p ≥ 2,
◦ ◦
kf k ◦ ≤ r1−(2/p) e(1−(2/p))ψM k f kLp (S,g/)
Lp (S,/
g)
◦◦ ◦◦
k∇
/f k ◦ ≤ r2−(2/p) e(1−(2/p))ψM k ∇
/f kLp (S,g/) (5.194)
Lp (S,/
g)
174
◦
Moreover, since f = r2 e2ψ f , we have:
◦◦ ◦
/f = r2 e2ψ (∇
∇ / f + 2d
/ψ ⊗ f )
hence, recalling the definition 5.155:
◦
k f kLp (S,g/) ≤ r2 e2ψM kf kLp(S,g/)
◦◦ ◦
k∇
/f kLp (S,g/) / f kLp (S,g/) + 2r−1 k1,∞ kf kLp(S,g/) } (5.195)
≤ r2 e2ψM {k ∇
In view of 5.193 - 5.195 the inequalities 5.191 and 5.192 imply, for any 2 ≤ p <
∞:
kθkLp (S,g/) ≤ Cp re(2−(2/p))oscψ eψM kf kLp(S,g/) (5.196)
◦
/ θkLp (S,g/) ≤ Cp e(3−(2/p))oscψ kf kLp(S,g/)
k∇ (5.197)
and:
◦
/ 2 θkLp (S,g/) ≤
k∇
◦
Cp e(4−(2/p))oscψ k ∇
/ f kLp(S,g/) + r−1 (e−ψM + k1,∞ )kf kLp(S,g/)
(5.198)
We have, in arbitrary local coordinates on S:
◦ ◦ ◦
∇ /A θBC =△D AB θDC + △D AC θBD
/A θBC − ∇ (5.199)
◦ ◦
where △, the difference of connections associated to g/ and g/ is given by 5.178.
It follows that:
◦ ◦
k∇
/θkLp(S,g/) ≤ k ∇
/ θkLp (S,g/) + Ck △ kL∞ (S,g/) kθkLp (S,g/) (5.200)
and from 5.179 and the definition 5.155 we obtain:
◦
k △ kL∞ (S,g/) ≤ Cr−1 k1,∞ (5.201)
From 5.200, 5.196 and 5.197 we then obtain:
/θkLp (S) ≤ Cp e(3−(2/p))oscψ (1 + eψm k1,∞ )kf kLp(S,g/)
k∇ (5.202)
Next, we have:
◦ ◦
∇
/A ∇ /A ∇
/B θCD − ∇ /B θCD
◦ ◦ ◦ ◦ ◦
=∇ /B θCD + △E BC θED + △E BD θCE )− ∇
/A (∇ /A ∇
/B θCD
◦ ◦ ◦ ◦ ◦ ◦
/B θED + △E AD ∇
/E θCD + △E AC ∇
=△E AB ∇ /B θCE
◦ ◦ ◦ ◦
+ △E BC ∇
/A θED + △E BD ∇ /A △E BC +θCE ∇
/A θCE + θED ∇ /A △E BD
175
Substituting from 5.199 we then obtain:
◦ ◦ ◦ ◦ ◦
/ 2 θ− ∇
∇ / 2 θ = (△, ∇
/θ) + (∇
/ △, θ) + (△, △, θ) (5.203)
where:
◦ ◦ ◦ ◦
(△, ∇
/θ)ABCD = △E AB ∇
/E θCD + △E AC ∇
/B θED + △E AD ∇
/B θCE
◦ ◦
+ △E BC ∇
/A θED + △E BD ∇
/A θCE
◦ ◦ ◦
(∇
/ △, θ)ABCD = θED ∇ /A △E BD
/A △E BC +θCE ∇
◦ ◦ ◦ ◦ ◦ ◦
(△, △, θ)ABCD = −(△E AB △F EC + △E AC △F BE )θF D
◦ ◦ ◦ ◦
−(△E AB △F ED + △E AD △F BE )θCF
◦ ◦ ◦ ◦
−(△E AC △F BD + △E AD △F BC )θEF (5.204)
Now, from 5.201,

◦ ◦
k(△, ∇
/θ)kLp (S,g/) ≤ Ck △ kL∞ (S,g/) k∇
/θkLp (S,g/)
≤ C ′ r−1 k1,∞ k∇
/θkLp (S,g/) (5.205)
Moreover,
◦ ◦
k(∇
/ △, θ)kLp (S,g/) ≤ Ck∇
/ △ kLp (S,g/) kθkL∞ (S,g/)
and from 5.178 and the definition 5.156,
◦
/ △ kLp(S,g/) ≤ Cr−(2−(2/p)) k2,p
k∇ (5.206)
hence:
◦
/ △, θ)kLp (S,g/) ≤ Cr−(2−(2/p)) k2,p kθkL∞ (S,g/)
k(∇ (5.207)
Also,
◦ ◦ ◦
k(△, △, θ)kLp (S,g/) ≤ Ck △ k2L2p (S,g/) kθkL∞ (S,g/)
and from 5.178 and the definition 5.152,
◦
k △ kL2p (S,g/) ≤ Cr−(1−(1/p)) k1,2p (5.208)
hence:
◦ ◦
k(△, △, θ)kLp (S,g/) ≤ Cr−(2−(2/p)) (k1,2p )2 kθkL∞ (S,g/) (5.209)
In view of 5.205, 5.207, 5.209, we conclude from 5.203 that:

◦
/ 2 θ− ∇
k∇ / 2 θkLp (S,g/) ≤ Cr−1 k1,∞ k∇/θkLp(S,g/)
+Cr−(2−(2/p)) k2,p + (k1,2p )2 kθkL∞ (S,g/)(5.210)

176
We also have:
◦ ◦
∇ /A fB − △C AB fC
/ A fB = ∇ (5.211)
hence:
◦ ◦
k∇
/ f kLp (S,g/) ≤ k∇
/f kLp(S,g/) + k △ kL∞ (S,g/) kf kLp(S,g/)
≤ /f kLp(S,g/) + Cr−1 k1,∞ kf kLp(S,g/)
k∇ (5.212)
In view of 5.210 and 5.212 we conclude from 5.198 that:
/ 2 θkLp (S,g/) ≤
k∇
Cp e(4−(2/p))oscψ k∇/f kLp(S,g/) + r−1 (e−ψM + k1,∞ )kf kLp (S,g/)

n o
+C r−1 k1,∞ k∇ /θkLp (S,g/) + r−(2−(2/p)) k2,p + (k1,2p )2 kθkL∞ (S,g/)

(5.213)
We have thus established the following lemma.
Lemma 5.6 Let θ be a trace-free symmetric 2-covariant tensorfield on S

satisfying the equation
div
/ /g θ = f
where f is a given 1-form on S. Then with ψ being the function which appears
in Proposition 5.4 we have, for every 2 ≤ p < ∞ the estimates:
kθkLp (S,g/) ≤ Cp re(2−(2/p))oscψ eψM kf kLp(S,g/)
/θkLp (S) ≤ Cp e(3−(2/p))oscψ (1 + eψm k1,∞ )kf kLp(S,g/)

k∇
/ 2 θkLp (S,g/) ≤
k∇
Cp e(4−(2/p))oscψ k∇/f kLp(S,g/) + r−1 (e−ψM + k1,∞ )kf kLp (S,g/)

n o
+C r−1 k1,∞ k∇ /θkLp (S,g/) + r−(2−(2/p)) k2,p + (k1,2p )2 kθkL∞ (S,g/)

C and Cp being numerical constants, the last depending on p.
Setting now p = 4, we have the bounds 5.182, 5.185 and 5.187. Substituting
these bounds in Lemma 5.6 and taking into account the fact that from Lemma
4.3
1 r √
√ ≤ ≤ 2 (5.214)
2 |u|
we deduce the following lemma.
Lemma 5.7 Let θ be a trace-free 2-covariant S tensorfield in M ′ satisfying

the equation
div
/ θ=f
177
where f is a given S 1-form. Then, provided that δ is suitably small depending
on D0∞ , R∞ /41 , R
0 , D /41 , and R
/2 (α), θ satisfies the estimate:
/ 2 θkL4 (Su,u ) ≤ Ck∇

k∇ /f kL4 (Su,u )
+C{1 + |u|3/2 kK − KkL4 (Su,u ) } ·
n o
· |u|−1 kf kL4 (Su,u ) + k∇/θkL4 (Su,u ) + |u|−3/2 kθkL∞ (Su,u )
for every (u, u) ∈ D′ . Here C is a numerical constant.
Lemma 5.7 will be applied to the system consisting of the propagation equa-
tions for K − K, trχ′ and the Codazzi elliptic system for χ̂′ . For this application
it is crucial that the quantity kK − KkL4 (Su,u ) enters the estimate linearly. Af-
ter the appropriate estimate for this quantity has been established, a different
version of the lemma, the version given below as Lemma 5.9, will be applied.
From Lemma 5.6 we have, for every 2 ≤ p < ∞:
kθkW1p (S,g/) ≤ Cp e(3−(2/p))oscψ 1 + eψm (1 + k1,∞ ) kf kLp(S,g/)

(5.215)
and if p > 2 we have, by Lemma 5.2:

p
kθkL∞ (S,g/) ≤ Cp I ′ (S, g/)r1−(2/p) kθkW1p (S,g/) (5.216)
Hence, if p > 2 we have:
kθkL∞ (S,g/) ≤ Cp I ′ (S, g/)r1−(2/p) e(3−(2/p))oscψ 1 + eψm (1 + k1,∞ ) kf kLp(S,g/)

p
(5.217)
Substituting 5.215, 5.217 in the last estimate of Lemma 5.6 we obtain the fol-
lowing version of that lemma.
Lemma 5.8 Let θ be a trace-free symmetric 2-covariant tensorfield on S

satisfying the equation
div
/ /g θ = f
where f is a given 1-form on S. Then with ψ being the function which appears
in Proposition 5.4 we have, for every 2 < p < ∞ the estimate:
/ 2 θkLp (S,g/) + k∇
rk∇ /θkLp (S,g/) + r−1 kθkLp(S,g/)
≤ C e(3−(2/p))oscψ eoscψ rk∇

p /f k Lp (S,g
/) + l2,p kf kLp(S,g/)
where
l2,p = e−ψm + eoscψ k1,∞ + k1,∞ (1 + eψm k1,∞ )

n p o
+ 1 + eψm (1 + k1,∞ ) 1 + I ′ (S, g/) k2,p + (k1,2p )2

and Cp is a numerical constant depending only on p.
178
Setting p = 4, substituting the bounds 5.182, 5.185, 5.187, and taking into
account 5.214 as well as Lemma 5.4 we deduce the following.
Lemma 5.9 Let θ be a trace-free 2-covariant S tensorfield in M ′ satisfying

the equation
div
/ θ=f
where f is a given S 1-form. Then, provided that δ is suitably small depending
on D0∞ , R∞ /41 , R
0 , D /41 , and R
/ 2 θkL4 (Su,u ) + k∇
|u|k∇ /θkL4 (Su,u ) + |u|−1 kθkL4 (Su,u )
2
3/2
≤ C |u|k∇ /f kL (Su,u ) + 1 + |u| kK − KkL (Su,u ) kf kL (Su,u )
4 4 4
Consider now the system of equations:

div
/ /g θ = f
curl
/ /g θ = g (5.218)
for a 1-form θ on S. Here (f, g) is a given pair of functions on S. This is an
elliptic system for θ. In fact, the following integral identity holds (see Chapter
2 of [C-K]): Z Z
/θ|/2g + K|θk/2g }dµ/g = {f 2 + g 2 }dµ/g
{|∇ (5.219)
S S
◦
The system 5.218 is conformally covariant; that is, if g/ = e2φ g/ then θ satisfies:
◦
div
/ ◦θ = f
/
g
◦ ◦ ◦
curl
/ ◦θ = g where f = e2φ f, g= e2φ g (5.220)
/
g
◦
Recall again that with φ = log r + ψ and ψ being as in Proposition 5.4, (S, g/)
is isometric to the standard sphere. The operator (div
/ ◦ , curl
/ ◦ ) from 1-forms to
/
g /
g
pairs of functions on the standard sphere being a first order elliptic operator
with vanishing kernel, the following Lp Calderon-Zygmund inequalities hold on
◦
(S, g/):
◦ ◦ ◦
k∇
/ θk ◦ + kθk ◦ ≤ Cp {k f k ◦ +kg k ◦ } (5.221)
Lp (S,/
g) Lp (S,/
g) Lp (S,/
g) Lp (S,/
g)
and:
◦ ◦ ◦ ◦ ◦
/ 2 θk
k∇ ◦ ≤ Cp {kd
/f k ◦ /gk
+ kd ◦ +kf k ◦ +kg k ◦ }
Lp (S,/
g) Lp (S,/
g) Lp (S,/
g) Lp (S,/
g) Lp (S,/
g)
(5.222)
179
◦ ◦
Applying 5.174 to ξ = θ (q = 1), to ξ =∇ / 2 θ (q = 3)
/ θ (q = 2), and to ξ =∇
we obtain, for p ≥ 2,
kθkLp (S,g/) ≤ r−(1−(2/p)) e−(1−(2/p))ψm kθk ◦
Lp (S,/
g)
◦ ◦
k∇
/ θkLp (S,g/) ≤ r−(2−(2/p)) e−(2−(2/p))ψm k ∇
/ θk ◦
Lp (S,/
g)
◦ ◦
/ 2 θkLp (S,g/)
k∇ / 2 θk
≤ r−(3−(2/p)) e−(3−(2/p))ψm k ∇ ◦ (5.223)
Lp (S,/
g)
◦ ◦
Also, since f = r2 e2ψ f , g = r2 e2ψ g we have:
◦ ◦
kf k ◦ ≤ r2−(2/p) e(2−(2/p))ψM k f kLp (S,g/)
Lp (S,/
g)
◦ ◦
kgk ◦ ≤ r2−(2/p) e(2−(2/p))ψM k g kLp (S,g/) (5.224)
Lp (S,/
g)
◦ ◦
Also, applying 5.174 to ξ = /d f , /d g (q = 1) we obtain, for q ≥ 2,
◦ ◦
kd
/f k ◦ ≤ r1−(2/p) e(1−(2/p))ψM kd
/ f kLp (S,g/)
Lp (S,/
g)
◦ ◦
/gk
kd ◦ ≤ r1−(2/p) e(1−(2/p))ψM kd
/ g kLp (S,g/)
Lp (S,/
g)
which, since
◦ ◦
/d f = r2 e2ψ (d
/f + 2f /dψ), /d g = r2 e2ψ (d
/g + 2gd
/ψ),
yields, recalling the definition 5.155:
◦
kd
/f k ◦ /f kLp(S,g/) + 2r−1 k1,∞ kf kLp(S,g/) }
≤ r3−(2/p) e(3−(2/p))ψM {kd
Lp (S,/
g)
◦
/gk
kd ◦ /gkLp(S,g/) + 2r−1 k1,∞ kgkLp(S,g/) }
≤ r3−(2/p) e(3−(2/p))ψM {kd
Lp (S,/
g)
(5.225)
In view of 5.223 - 5.225 the inequalities 5.221 and 5.222 imply, for any 2 ≤ p <
∞:
kθkLp(S,g/) ≤ Cp re(1−(2/p))oscψ eψM {kf kLp(S,g/) + kgkLp(S,g/) } (5.226)
◦
/ θkLp (S,g/) ≤ Cp e(2−(2/p))oscψ {kf kLp(S,g/) + kgkLp(S,g/) }
k∇ (5.227)
and:
◦
/ 2 θkLp (S,g/) ≤ Cp e(3−(2/p))oscψ kd

k∇ /f kLp(S,g/) + kd
/gkLp(S,g/)
+r−1 (e−ψM + k1,∞ )(kf kLp (S,g/) + kgkLp(S,g/) )
(5.228)
180
We have, in arbitrary local coordinates on S:
◦ ◦
∇ /A θB =△C AB θC
/ A θB − ∇ (5.229)
Hence:
◦ ◦
k∇
/θkLp (S,g/) ≤ k ∇
/ θkLp (S,g/) + k △ kL∞ (S,g/) kθkLp (S,g/) (5.230)
From 5.226, 5.227 and 5.201 we then obtain:
/θkLp(S,g/) ≤ Cp e(2−(2/p))oscψ (1 + eψm k1,∞ ){kf kLp(S,g/) + kgkLp(S,g/) } (5.231)

k∇
Also, proceeding as in the case of a trace-free 2-covariant tensorfield on S,

we deduce in the present case of a 1-form on S, in analogy with 5.210,
◦
/ 2 θ− ∇
k∇ / 2 θkLp (S,g/) ≤ Cr−1 k1,∞ kθkLp(S,g/) (5.232)
−(2−(2/p)) 2

+Cr k2,p + (k1,2p ) kθkL∞ (S,g/)
which, together with 5.228 yields:
/ 2 θkLp (S,g/) ≤
k∇
Cp e(3−(2/p))oscψ kd

/f kLp(S,g/) + kd
/gkLp(S,g/)
+r−1 (e−ψM + k1,∞ ) kf kLp(S,g/) + kgkLp(S,g/)

n o
+C r−1 k1,∞ k∇
/θkLp (S,g/) + r−(2−(2/p)) k2,p + (k1,2p )2 kθkL∞ (S,g/)

(5.233)
Now, from 5.226 and 5.231 we have, for every 2 ≤ p < ∞:
kθkW1p (S,g/) ≤ Cp e(2−(2/p))oscψ 1 + eψm (1 + k1,∞ ) ·

·{kf kLp(S,g/) + kgkLp(S,g/) } (5.234)
and if p > 2 we have, by Lemma 5.2:

p
kθkL∞ (S,g/) ≤ Cp I ′ (S, g/)r1−(2/p) kθkW1p (S,g/) (5.235)
Hence, if p > 2 we have:
kθkL∞ (S,g/) ≤ Cp I ′ (S, g/)r1−(2/p) e(2−(2/p))oscψ 1 + eψm (1 + k1,∞ ) ·

p
·{kf kLp(S,g/) + kgkLp(S,g/) } (5.236)
Substituting this as well as 5.231 in 5.233 we obtain the following lemma.
Lemma 5.10 Let θ be a 1-form on S satisfying the system
div
/ /g θ = f
curl
/ /g θ = g
181
where f and g are given functions on S. Then with ψ being the function which
appears in Proposition 5.4 we have, for every 2 < p < ∞ the estimate:
/ 2 θkLp (S,g/) + k∇
rk∇ /θkLp(S,g/) + r−1 kθkLp (S,g/)
≤ C e(2−(2/p))oscψ eoscψ r kd

p /f k + kd
Lp (S,g
/) /gkLp(S,g/)

+l2,p kf kLp(S,g/) + kgkLp(S,g/)
where
l2,p = e−ψm + eoscψ k1,∞ + k1,∞ (1 + eψm k1,∞ )

n p o
+ 1 + eψm (1 + k1,∞ ) 1 + I ′ (S, g/) k2,p + (k1,2p )2

and Cp is a numerical constant depending only on p.
Setting p = 4, substituting the bounds 5.182, 5.185, 5.187, and taking into
account 5.214 as well as Lemma 5.4 we deduce the following.
Lemma 5.11 Let θ be a S 1-form in M ′ satisfying the system
div
/ θ = f
curl
/ θ = g
where f and g are given functions. Then, provided that δ is suitably small
depending on D0∞ , R∞ /41 , R
0 , D /41 , and R
/ 2 θkL4 (Su,u ) + k∇
|u|k∇ /θkL4 (Su,u ) + |u|−1 kθkL4 (Su,u )
n
≤ C |u| kd /f kL4 (Su,u ) + kd
/gkL4(Su,u )
2
3/2
+ 1 + |u| kK − KkL4 (Su,u ) kf kL4(Su,u ) + kgkL4 (Su,u )
182
Chapter 6
L4(S) Estimates for the 2nd

Derivatives of the
6.1 Introduction
The optical estimates of Chapters 3 and 4 relied only on the propagation equa-
tions among the optical structure equations. These are ordinary differential
equations along the generators of the null hypersurfaces Cu or C u . The esti-
mates loose one degree of differentiability, because of the presense of principal
terms on the right hand sides of the propagation equations. The present chapter
and the next, on the other hand, consider systems of optical structure equations
which consist of elliptic equations on the Su,u sections of the null hypersurfaces
Cu or C u coupled to ordinary differential equations along the generators of these
hypersurfaces. The ordinary differential equations considered here, in contrast
to the remaining propagation equations, do not contain principal terms on the
right hand side, by virtue of the Einstein equations. This approach, which al-
lows us to obtain optical estimates which are optimal from the point of view of
differentiability, was introduced in [C-K] and plays a basic role in the present
work as well. There is however in this approach, as we shall see, a loss of a factor
of δ 1/2 in behavior with respect to δ, in comparison to the estimates which rely
only on the propagation equations, in the case of η, η, and ω. What is crucial,
is that there is no such loss in the case of χ, χ, and ω, but the proof of this fact
uses again the former estimates.
The estimates on the present chapter use L4 elliptic theory on the Su,u
sections, the results derived in the previous chapter through the uniformization
theorem. The first step is the estimate for ∇ / 2 χ and, at the same time, the
4
estimate for K − K, both in L (S). This uses Lemma 5.7 applied to the Codazzi
equation 1.151. The statement of Lemma 5.7 shows that the two estimates must
183
be coupled.

/ 2 χ, K − K
We consider the propagation equation for /dtrχ′ , the first of 4.78, which reads
(see 4.79):
/trχ′ + Ωtrχd
Dd /trχ′ = −2Ω(χ̂, ∇ /χ̂′ ) + r (6.1)
where:
1
/ log Ω) (trχ)2 + |χ̂|2
r = −2(d (6.2)
2
and we denote:
/χ̂′ )A = χ̂BC ∇
(χ̂, ∇ /A χ̂′BC , (6.3)
in conjunction with the Codazzi equation 1.151, in the form:
1
/ χ̂′ = /dtrχ′ + i
div (6.4)
2
where:
1
i = Ω−1 −β + trχη − χ̂♯ · η , (6.5)
2
an elliptic equation for χ̂′ on each Su,u section.

kK − KkL4 (Su,u ) ≤ Cδ 1/2 |u|−5/2 (R
/41 (α) + R
/41 (β) + R∞ ∞
0 (α) + R0 (β))
+O(δ|u|−7/2 )
/ 2 trχkL4 (Su,u ) ≤
k∇
C|u|−7/2 (R
/41 (α) + R∞ /41 (α) + R
0 (α))(R /41 (β) + R∞ ∞
0 (α) + R0 (β))
+O(δ 1/2 |u|−7/2 )
/ 2 χ̂kL4 (Su,u ) + k∇
|u|k∇ /χ̂kL4 (Su,u ) + |u|−1 kχ̂kL4 (Su,u )
≤ Cδ −1/2 |u|−3/2 (R
/41 (β) + R∞
0 (β)) + O(|u|
−5/2
)
for all (u, u) ∈ D′ , provided that δ is suitably small depending on D0∞ , R∞ /41 ,
0 , D
4
R/1 , and R
/2 (α). In particular, under such a smallness condition on δ we have:
|u|3/2 kK − KkL4 (Su,u ) ≤ 1
Proof: We apply Lemma 5.7 to equation 6.4 to obtain:
/ 2 χ̂′ kL4 (Su,u ) ≤ C(k∇
k∇ / 2 trχ′ kL4 (Su,u ) + k∇
/ikL4(Su,u ) )
+C{1 + |u|3/2 kK − KkL4 (Su,u ) } ·
n
|u|−1 kd /χ̂′ kL4 (Su,u ) + kikL4 (Su,u )
/trχ′ kL4 (Su,u ) + k∇
o
+|u|−3/2 kχ̂′ kL∞ (Su,u ) (6.6)
184
Consider the second factor of the second term on the right in 6.6. From Propo-
sition 4.1, or, directly from the estimates 4.93 and 4.94 we have:
/trχ′ kL4 (Su,u )
kd ≤ C|u|−5/2 R∞ /41 (α) + O(δ 1/2 |u|−7/2 )
0 (α)R (6.7)
/χ̂′ kL4 (Su,u )
k∇ ≤ Cδ −1/2 |u|−3/2 R
/41 (α) + O(δ 1/2 |u|−7/2 ) (6.8)
and by the third of 4.76:
kχ̂kL∞ (Su,u ) ≤ Cδ −1/2 |u|−1 R∞
0 (α) (6.9)
Also, using the estimates of Chapter 3 we obtain:
kikL4 (Su,u ) ≤ C|u|1/2 kikL∞ (Su,u )
≤ C ′ δ −1/2 |u|−3/2 R∞
0 (β) + O(|u|
−5/2
) (6.10)
It follows from the above that:

|u|−1 kd /χ̂′ kL4 (Su,u ) + kikL4 (Su,u )
/trχ′ kL4 (Su,u ) + k∇
+|u|−3/2 kχ̂′ kL∞ (Su,u ) ≤ δ −1/2 |u|−5/2 l(u) (6.11)
where:
/41 (α) + R∞
l(u) = C(R ∞
0 (α) + R0 (β)) + O(δ
1/2
|u|−1 ) (6.12)
Using the results of Chapters 3 and 4 we deduce:
/ikL4(Su,u ) ≤ Cδ −1/2 |u|−5/2 R
k∇ /41 (β) + O(|u|−7/2 ) (6.13)
Substituting 6.11 and 6.13 in 6.6 we obtain:
/ 2 χ̂′ kL4 (Su,u ) ≤ Ck∇
k∇ / 2 trχ′ kL4 (Su,u ) + Cδ −1/2 |u|−1 l(u)kK − KkL4 (Su,u )
+Cδ −1/2 |u|−5/2 (R
/41 (α) + R
/41 (β) + R∞ ∞
0 (α) + R0 (β)) + O(|u|
−7/2
)
(6.14)
We now revisit the propagation equation for K − K, equation 5.31. We
apply Lemma 4.6 to this equation, taking p = 4. Here r = 0, ν = −2, γ = 0.
We obtain:
Z u
kK − KkL4 (Su,u ) ≤ C k(Ωtrχ − Ωtrχ)KkL4 (Su′ ,u ) du′ (6.15)
0
Z u
1
+C kdiv
/ div
/ (Ωχ̂) − △ / (Ωtrχ)kL4 (Su′ ,u ) du′
0 2
By 5.34 the integrant of the first integral on the right in 6.15 is bounded by
O(|u|−7/2 ). The integrant of the second integral on the right in 6.15 is bounded
by:
/ 2 trχ′ kL4 (S) + k∇
/ 2 χ̂′ kL4 (S)

C k∇
/trχ′ kL4 (S) + k∇
/χ̂′ kL4 (S)

+C kd / log ΩkL∞ (S) kd
/ 2 log ΩkL4 (S) (ktrχ′ kL∞ (S) + kχ̂′ kL∞ (S) )
+k∇
185
Using the estimates of Proposition 4.1, the expression 5.17 for ∇ / 2 log Ω and the
4
estimates of Proposition 4.2 for ∇
/η, ∇/η in L (S), as well as the estimates of
Chapter 3, the lower order terms are seen to be bounded by O(|u|−7/2 ). Thus
6.15 implies:
Z un o
kK − KkL4 (Su,u ) ≤ C k∇/ 2 trχ′ kL4 (Su′ ,u ) + k∇
/ 2 χ̂′ kL4 (Su′ ,u ) du′
0
+O(δ|u|−7/2 ) (6.16)
Substituting 6.16 in 6.14 yields a linear integral inequality, for fixed u, for
the quantity:
x(u) = k∇/ 2 χ̂′ kL4 (Su,u ) (6.17)
of the form: Z u
x(u) ≤ a x(u′ )du′ + b(u) (6.18)
0
Here a is the non-negative constant:
a = Cδ −1/2 |u|−1 l(u) (6.19)
and b(u) is the non-negative function:

Z u
b(u) / 2 trχ′ kL4 (Su,u ) + Cδ −1/2 |u|−1 l(u)
= Ck∇ / 2 trχ′ kL4 (Su′ ,u ) du′
k∇
0
+Cδ −1/2 |u|−5/2 (R
/41 (α) + R
/41 (β) + R∞ ∞
0 (α) + R0 (β)) + O(|u|
−7/2
)
(6.20)
The inequality 6.18 implies:

Z u
′
x(u) ≤ a ea(u−u ) b(u′ )du′ + b(u) (6.21)
0
Here
a(u − u′ ) ≤ aδ = Cδ 1/2 |u|−1 l(u) ≤ log 2 (6.22)
the last step provided that δ is suitably small depending on D0∞ , R∞
0 , /41 , R
D /41 .
Hence 6.21 implies: Z u
x(u) ≤ 2a b(u′ )du′ + b(u) (6.23)
0
From 6.20 and 6.22,
Z u Z u
′ ′
b(u )du ≤ C / 2 trχ′ kL4 (Su′ ,u ) du′
k∇ (6.24)
0 0
+Cδ 1/2 |u|−5/2 (R
/41 (α) + R
/41 (β) + R∞ ∞
0 (α) + R0 (β)) + O(δ|u|
−7/2
)
186
Therefore, substituting in 6.23 we obtain:
/ 2 χ̂′ kL4 (Su,u )

k∇ / 2 trχ′ kL4 (Su,u )
≤ Ck∇
Z u
+Cδ −1/2 |u|−1 l(u) / 2 trχ′ kL4 (Su′ ,u ) du′
k∇
0
+Cδ −1/2 |u|−5/2 (R
/41 (α) + R
/41 (β) + R∞ ∞
0 (α) + R0 (β))
+O(|u|−7/2 ) (6.25)
We now turn to the propagation equation 6.1. We apply Lemma 4.1 to this
/ 2 trχ′ :
equation to deduce the following propagation equation for ∇
/ 2 trχ′ + Ωtrχ∇
D∇ / 2 trχ′ = −2Ω(χ̂, ∇
/ 2 χ̂′ ) + e (6.26)
where:
/)C
eAB = −(DΓ dC trχ′ − /dA (Ωtrχ)d
AB / /B trχ′ − 2∇
/A (Ωχ̂CD )∇
/B χ̂′CD + ∇
/A rB (6.27)
and we denote:
/ 2 χ̂′ )AB = χ̂CD ∇
(χ̂, ∇ /B χ̂′CD
/A ∇ (6.28)
To estimate in L4 (S) the first three terms on the right in 6.27 we place the first
factors in L4 (S) and the second factors in L∞ (S). We then use Lemma 5.2 with
p = 4 to bound the L∞ (S) norms of the second factors in terms of their W14 (S)
norms. We have, in view of Lemma 5.4, in regard to the first term on the right
in 6.27:
/ · /dtrχ′ kL4 (Su,u ) ≤ kDΓ

kDΓ /trχ′ kL∞ (Su,u )
/kL4(Su,u ) kd
/kL4(Su,u ) {|u|1/2 k∇
≤ CkDΓ / 2 trχ′ kL4 (Su,u ) + |u|−1/2 kd
/trχ′ kL4 (Su,u ) }
(6.29)
Similarly, the L4 (S) norm of the second term on the right in 6.27 is bounded
by:
/(Ωtrχ)kL4 (Su,u ) {|u|1/2 k∇

Ckd / 2 trχ′ kL4 (Su,u ) + |u|−1/2 kd
/trχ′ kL4 (Su,u ) } (6.30)
and the L4 (S) norm of the third term on the right in 6.27 is bounded by:
/(Ωχ̂)kL4 (Su,u ) {|u|1/2 k∇

Ck∇ / 2 χ̂′ kL4 (Su,u ) + |u|−1/2 k∇
/χ̂′ kL4 (Su,u ) } (6.31)
Now, from Lemma 4.1 we have, pointwise:
|DΓ
/| ≤ C(|d
/(Ωtrχ)| + |∇
/(Ωχ̂)|) (6.32)
hence:
kDΓ
/kL4(Su,u ) ≤ C{kd
/(Ωtrχ)kL4 (Su,u ) + k∇
/(Ωχ̂)kL4 (Su,u ) }
≤ Cδ −1/2 |u|−3/2 R
/41 (α) + O(|u|−5/2 ) (6.33)
187
by Proposition 4.1. Substituting also the estimates 6.7, 6.8, we conclude that
the first three terms on the right in 6.27 are bounded in L4 (S) norm by:
n
Cδ −1/2 |u|−1 R
/41 (α) + O(|u|−2 ) k∇/ 2 trχ′ kL4 (Su,u ) + k∇/ 2 χ̂′ kL4 (Su,u )
o
+Cδ −1/2 |u|−5/2 R /41 (α) + O(|u|−7/2 )
(6.34)
2
Using the estimates of Proposition 4.1, the expression 5.17 for ∇
/ log Ω and the
estimates of Proposition 4.2 for ∇
/η, ∇/η in L4 (S), as well as the estimates of
Chapter 3, we obtain, in regard to the last term on the right in 6.27:
/rkL4 (Su,u ) ≤ O(δ −1/2 |u|−9/2 )
k∇ (6.35)
We then conclude that:
kekL4(Su,u ) ≤
n o
Cδ −1/2 |u|−1 R
/41 (α) + O(|u|−2 ) k∇ / 2 χ̂′ kL4 (Su,u )
/ 2 trχ′ kL4 (Su,u ) + k∇
+Cδ −1 |u|−7/2 (R
/41 (α))2 + O(δ −1/2 |u|−9/2 ) (6.36)
4
We also estimate in L (S) the first term on the right in 6.26:
/ 2 χ̂′ )kL4 (Su,u ) ≤ Cδ −1/2 |u|−1 R∞
kΩ(χ̂, ∇ / 2 χ̂′ kL4 (Su,u )
0 (α)k∇ (6.37)
by the third of 4.76. Substituting finally in 6.36 and 6.37 the estimate 6.25 we
conclude that:
/ 2 χ̂′ ) + ekL4 (Su,u ) ≤
k − 2Ω(χ̂, ∇
n
Cδ −1/2 |u|−1 m(u) k∇ / 2 trχ′ kL4 (Su,u )
Z u
k∇
0
+Cδ −1 |u|−7/2 (R
/41 (α) + R∞ /41 (α) + R
0 (α))(R /41 (β) + R∞ ∞
0 (α) + R0 (β))
+O(δ −1/2 |u|−9/2 ) (6.38)
where:
m(u) = C(R /41 (α) + R∞0 (α)) + O(δ
1/2
|u|−1 ) (6.39)
We now apply Lemma 4.6 with p = 4 to 6.26. Here r = 2, ν = −2, γ = 0.
We obtain:
Z u
2 ′
k∇/ trχ kL4 (Su,u ) ≤ C / 2 χ̂′ ) + ekL4 (Su′ ,u ) du′
k − 2Ω(χ̂, ∇ (6.40)
0
/ 2 trχ′ kL4 (S) .

Substituting the bound 6.38 then yields a linear integral inequality for k∇
Since for a non-negative function f (u) on [0, δ] we have:
Z u (Z u′ ) Z u Z u
f (u )du du′ ≤
′′ ′′
f (u′′ )du′′ du′
0 0 0 0
Z u Z u
′′ ′′
=u f (u )du ≤ δ f (u′ )du′ (6.41)
0 0
188
the contribution of the term in parenthesis in 6.38 involving the integral is
bounded by
Cδ 1/2 |u|−1 l(u) (6.42)
times the contribution of the first term in parenthesis in 6.38, and the factor
6.42 is less than or equal to 1 provided that δ is suitably small depending on
D0∞ , R∞ /41 , R
0 , D /41 . Therefore the resulting integral inequality simplifies to:
Z u
2 ′ −1/2 −1
k∇
/ trχ kL (Su,u ) ≤ Cδ
4 |u| m(u) k∇/ 2 trχ′ kL4 (Su′ ,u ) du′
0
+C|u|−7/2 (R
/41 (α) + R∞ /41 (α) + R
0 (α))(R /41 (β) + R∞ ∞
0 (α) + R0 (β))
+O(δ 1/2 |u|−9/2 ) (6.43)
This is a linear integral inequality of the form 6.18, but with b a positive con-
stant. Noting that
Cδ 1/2 |u|−1 m(u) ≤ 1 (6.44)
0 , D /41 , the integral
inequality 6.43 implies:
/ 2 trχ′ kL4 (Su,u ) ≤

k∇
C|u|−7/2 (R
/41 (α) + R∞ /41 (α) + R
0 (α))(R /41 (β) + R∞ ∞
0 (α) + R0 (β))
+O(δ 1/2 |u|−9/2 ) (6.45)
Substituting the estimate 6.45 in 6.25 we obtain:
/ 2 χ̂′ kL4 (Su,u )

k∇ ≤ Cδ −1/2 |u|−5/2 (R
/41 (α) + R
/41 (β) + R∞ ∞
0 (α) + R0 (β))
+O(|u|−7/2 ) (6.46)
Substituting then the estimates 6.45 and 6.46 in 6.16 yields:
kK − KkL4 (Su,u ) ≤ Cδ 1/2 |u|−5/2 (R

/41 (α) + R
/41 (β) + R∞ ∞
0 (α) + R0 (β))
+O(δ|u|−7/2 ) (6.47)
It follows that:
|u|3/2 kK − KkL4 (Su,u ) ≤ 1 (6.48)
provided that δ is suitably small depending on D0∞ , R∞ 0 , D/41 , R
/41 . We now
apply Lemma 5.9 to equation 6.4. In view of 6.48 the conclusion of that lemma
simplifies and we obtain:
/ 2 χ̂′ kL4 (Su,u ) + k∇

|u|k∇ /χ̂′ kL4 (Su,u ) + |u|−1 kχ̂′ kL4 (Su,u )
n
≤ C |u|(k∇ / 2 trχ′ kL4 (Su,u ) + k∇/ikL4(Su,u ) )
o
+kd /trχ′ kL4 (Su,u ) + kikL4(Su,u ) (6.49)
189
Substituting on the right the estimates 6.7, 6.45, 6.10, 6.13 then yields:
/ 2 χ̂′ kL4 (Su,u ) + k∇

≤ Cδ −1/2 |u|−3/2 (R
/41 (β) + R∞
0 (β)) + O(|u|
−5/2
) (6.50)
In view of the estimates 6.45, 6.47, 6.50, the proposition follows if we also take
into account the expression 5.17 for ∇/ 2 log Ω and the estimates of Proposition
4
4.2 for ∇
/η, ∇
/η in L (S).

/ 2χ
We now consider any (u1 , u1 ) ∈ D′ and fix attention in the following lemmas to
the parameter subdomain D1 and the corresponding subdomain M1 of M ′ (see
3.45, 3.46).
With a positive constant k to be appropriately chosen in the sequel, let s∗
be the least upper bound of the set of values of s ∈ [u0 , u1 ] such that:
/ 2 log ΩkL4 (Su,u ) ≤ kδ|u|−7/2

k∇ : for all (u, u) ∈ [0, u1 ] × [u0 , s] (6.51)
Then by continuity s∗ > u0 (recall that by 3.44 ∇

/ 2 log Ω vanishes along Cu0 )
and we have:
/ 2 log ΩkL4 (Su,u ) ≤ kδ|u|−7/2

k∇ : for all (u, u) ∈ [0, u1 ] × [u0 , s∗ ] (6.52)
In the estimates to follow the dependence on the constant k is made explicit.

Let us denote:
D / 2 trχkL4 (Su,u0 )
/42 (trχ) = |u0 |9/2 δ −1 sup k∇ (6.53)
u∈[0,δ]
Here, as in 3.49, 3.122, 4.97, 4.98, we are considering all of Cu0 , not only the
part which lies in M ′ . By the results of Chapter 2 this quantity is bounded by a
∗
non-negative non-decreasing continuous function of M5 . Let us denote by D1s
the subset of D1 where u ≤ s∗ :
∗
D1s = [0, u1 ] × [u0 , s∗ ] (6.54)
∗
We consider the propagation equation for /dtrχ′ , the first of 4.101, which
reads (see 4.102):
/trχ′ + Ωtrχd
Dd /χ̂′ ) + r
/trχ′ = −2Ω(χ̂, ∇ (6.55)
where:
1
r = −2(d
/ log Ω) (trχ)2 + |χ̂|2 (6.56)
2
190
and we denote:
/χ̂′ )A = χ̂BC ∇
(χ̂, ∇ /A χ̂′BC , (6.57)
in conjunction with the Codazzi equation 1.152, in the form:
1
/ χ̂′ = /dtrχ′ + i
div (6.58)
2
where:
1
i = Ω−1 β + trχη − χ̂♯ · η , (6.59)
2
an elliptic equation for χ̂′ on each Su,u section. Equations 6.55 and 6.58 are the
conjugates of equations 6.1 and 6.4 respectively.
Lemma 6.1 We have:
/ 2 trχkL4 (Su,u ) ≤ Cδ|u|−9/2 D

4
k∇ /41 (trχ) + D
/2 (trχ) + k + (D /41 (χ̂))2
+(D /41 (χ̂) + D0∞ (χ̂))(R
/41 (trχ) + D /41 (β) + R∞
0 (β))
+O(δ 3/2 |u|−11/2 )
/ 2 χ̂kL4 (Su,u ) + k∇
≤ Cδ 1/2 |u|−5/2 (R
/41 (β) + R∞
0 (β)) + Cδ|u|
−7/2
k + O(δ|u|−7/2 )
∗
0 ,
/41 , R
D /41 , and R
/2 (α).
Proof: We apply Lemma 5.9 to equation 6.58. In view of the last conclusion of
Proposition 6.1 the conclusion of Lemma 5.9 simplifies and we obtain:
/ 2 χ̂′ kL4 (Su,u ) + k∇

n
≤ C |u|(k∇ / 2 trχ′ kL4 (Su,u ) + k∇/ikL4 (Su,u ) )
o
+kd /trχ′ kL4 (Su,u ) + kikL4 (Su,u ) (6.60)
Using Proposition 4.2 and the results of Chapter 3 we deduce:
/ikL4 (Su,u ) ≤ Cδ 1/2 |u|−7/2 R

k∇ /41 (β) + O(δ|u|−9/2 ) (6.61)
/i in behavior with respect to δ coming from the

the leading contribution to ∇
term (1/2)Ω−1 trχ∇/η. Also, using the estimates of Chapter 3 we obtain:
kikL4 (Su,u ) ≤ C|u|1/2 kikL∞ (Su,u )

≤ C ′ δ 1/2 |u|−5/2 R∞
0 (β) (6.62)
191
the leading contribution to i in behavior with respect to δ coming from the term
(1/2)Ω−1 trχη. Substituting 6.61, 6.62 and the estimate 4.175 in 6.60 we obtain:
/ 2 χ̂′ kL4 (Su,u ) + k∇

/ 2 trχ′ kL4 (Su,u )
≤ C|u|k∇
+Cδ 1/2 |u|−5/2 (R
/41 (β) + R∞
0 (β)) + O(δ|u|
−7/2
) (6.63)
/ 2 trχ′ :
/ 2 trχ′ + Ωtrχ′ ∇
D∇ / 2 χ̂′ ) + e
/ 2 trχ′ = −2Ω(χ̂, ∇ (6.64)
where:
/)C
eAB = −(DΓ dC trχ′ − /dA (Ωtrχ)d
AB / /A (Ωχ̂CD )∇
/B trχ′ − 2∇ /B χ̂′CD + ∇
/A r B (6.65)
and we denote:
/ 2 χ̂′ )AB = χ̂CD ∇
(χ̂, ∇ /B χ̂′CD
/A ∇ (6.66)
To estimate in L4 (S) the first three terms on the right in 6.65 we place the first
factors in L4 (S) and the second factors in L∞ (S). We then use Lemma 5.2 with
p = 4 to bound the L∞ (S) norms of the second factors in terms of their W14 (S)
norms. We have, in view of Lemma 5.4, in regard to the first term on the right
in 6.65:
/ · /dtrχ′ kL4 (Su,u ) ≤ kDΓ

kDΓ /trχ′ kL∞ (Su,u )
/kL4 (Su,u ) kd
/kL4(Su,u ) {|u|1/2 k∇
≤ CkDΓ / 2 trχ′ kL4 (Su,u ) + |u|−1/2 kd
/trχ′ kL4 (Su,u ) }
(6.67)
Similarly, the L4 (S) norm of the second term on the right in 6.65 is bounded
by:
Ckd / 2 trχ′ kL4 (Su,u ) + |u|−1/2 kd

/(Ωtrχ)kL4 (Su,u ) {|u|1/2 k∇ /trχ′ kL4 (Su,u ) } (6.68)
and the L4 (S) norm of the third term on the right in 6.65 is bounded by:
/ 2 χ̂′ kL4 (Su,u ) + |u|−1/2 k∇

/(Ωχ̂)kL4 (Su,u ) {|u|1/2 k∇
Ck∇ /χ̂′ kL4 (Su,u ) } (6.69)
Now, from Lemma 4.1 we have, pointwise:
|DΓ
/| ≤ C(|d
/(Ωtrχ)| + |∇
/(Ωχ̂)|) (6.70)
hence:
kDΓ
/kL4 (Su,u ) ≤ C{kd
/(Ωtrχ)kL4 (Su,u ) + k∇
/(Ωχ̂)kL4 (Su,u ) }
≤ Cδ 1/2 |u|−5/2 (D
/41 (trχ) + D
/41 (χ̂)) + O(δ|u|−7/2 ) (6.71)
192
by Proposition 4.2 and Lemma 4.11. Using Proposition 4.2 and Lemma 4.11
we then conclude that the first three terms on the right in 6.65 are bounded in
L4 (S) norm by:

C δ 1/2 |u|−2 (D/41 (trχ) + D /41 (χ̂)) + O(δ|u|−3 ) ·
n
· k∇ / 2 trχ′ kL4 (Su,u ) + k∇ / 2 χ̂′ kL4 (Su,u )
o
+Cδ 1/2 |u|−7/2 (D /41 (trχ) + D/41 (χ̂)) + O(δ|u|−9/2 ) (6.72)
/r, contains the term:

The last term on the right in 6.65, ∇

1
− 2∇/ 2 log Ω (trχ)2 + |χ|2 (6.73)
2
We estimate this in L4 (S) using 6.52. We obtain a bound by:
Ckδ|u|−11/2
∗
for all (u, u) ∈ D1s . The remaining terms in ∇
/r are bounded in L4 (S) using
−15/2
Proposition 4.2 and Lemma 4.11 by O(δ|u| ). Thus:
∗
/rkL4 (Su,u ) ≤ Ckδ|u|−11/2 + O(δ 2 |u|−15/2 )
k∇ : ∀(u, u) ∈ D1s (6.74)
We then conclude that:
kekL4 (Su,u ) ≤

C δ 1/2 |u|−2 (D
/41 (trχ) + D /41 (χ̂)) + O(δ|u|−3 ) ·
n o
· k∇ / 2 trχ′ kL4 (Su,u ) + k∇ / 2 χ̂′ kL4 (Su,u )
+ Cδ|u|−11/2 k + (D /41 (trχ) + D/41 (χ̂))2 + O(δ 3/2 |u|−13/2 )

(6.75)
∗
for all (u, u) ∈ D1s . We also estimate in L4 (S) the first term on the right in
6.64:

kΩ(χ̂, ∇/ 2 χ̂′ )kL4 (Su,u ) ≤ C δ 1/2 |u|−2 (D0∞ (χ̂) + O(δ 3/2 |u|−7/2 ) k∇
/ 2 χ̂′ kL4 (Su,u )
(6.76)
by Proposition 3.2 (Lemma 3.2). Substituting finally in 6.75 and 6.76 the esti-
mate 6.63 we conclude that:
/ 2 χ̂′ ) + ekL4 (Su,u ) ≤

k − 2Ω(χ̂, ∇
Cδ 1/2 |u|−2 m(u)k∇ / 2 trχ′ kL4 (Su,u )
+Cδ|u|−11/2 k + (D /41 (trχ) + D/41 (χ̂))2

/41 (trχ) + D
+(D /41 (χ̂) + D0∞ (χ̂))(R
/41 (β) + R∞
0 (β))
+O(δ 3/2 |u|−13/2 ) (6.77)
193
∗
for all (u, u) ∈ D1s . Here:
/41 (trχ) + D
m(u) = C(D /41 (χ̂) + D0∞ (χ̂)) + O(δ 1/2 |u|−1 ) (6.78)

We obtain:
|u|7/2 k∇
/ 2 trχ′ kL4 (Su,u ) ≤ C|u0 |7/2 k∇
/ 2 trχ′ kL4 (Su,u0 ) (6.79)
Z u
+C / 2 χ̂′ ) + ekL4 (Su,u′ ) du′
|u′ |7/2 k − 2Ω(χ̂, ∇
u0
∗
for all (u, u) ∈ D1s . Substituting 6.77 and recalling the definition 6.53 yields a
linear integral inequality, for fixed u, for the quantity:
x(u) = |u|7/2 k∇
/ 2 trχ′ kL4 (Su,u ) (6.80)
of the form: Z u
x(u) ≤ a(u′ )x(u′ )du′ + b(u) (6.81)
u0
Here a is the non-negative function:
a(u) = Cδ 1/2 |u|−2 m(u) (6.82)
and b is the non-negative non-decreasing function:
b(u) = Cδ|u0 |−1 D

/42 (trχ)
Z u
Cδ|u|−2 k + (D /41 (χ̂))2
/41 (trχ) + D

+
u0
/41 (trχ) + D
/41 (χ̂) + D0∞ (χ̂))(R
/41 (β) + R∞

+(D 0 (β))
o
+O(δ 3/2 |u|−3 ) du′ (6.83)
Setting Z u
X(u) = a(u′ )x(u′ )du′ , we have X(u0 ) = 0 (6.84)
u0

dX
≤ a(X + b) (6.85)
du
It follows that:
Z u Ru
a(u′′ )du′′
X(u) ≤ e u′ a(u′ )b(u′ )du′ (6.86)
u0
( R u′ )
R u
R u u
db ′
Z
a(u′ )du′ − a(u′ )du′ − a(u′′ )du′′
= e u0
b(u0 ) − b(u)e u0
+ e u0
(u )du′
u0 du′
194
Hence:
x(u) ≤ X(u) + b(u)

Ru u
db ′
Z Ru
a(u′ )du′ a(u′′ )du′′
≤ e u0 b(u0 ) + e u′ (u )du′
u0 du′
(6.87)
Now, if δ is suitably small depending on D0∞ , R∞ /41 , R

0 , D /41 , we have (see 6.82
and 6.78): Z u
a(u′ )du′ ≤ log 2 (6.88)
u0
therefore 6.87 simplifies to:
u
db ′
Z
x(u) ≤ 2b(u0 ) + 2 du = 2b(u) (6.89)
u0 du′
In view of the definitions 6.80, 6.83 we conclude that:
/ 2 trχ′ kL4 (Su,u ) ≤ Cδ|u|−9/2 D

4
k∇ /41 (trχ) + D
/2 (trχ) + k + (D /41 (χ̂))2
/41 (trχ) + D
+(D /41 (χ̂) + D0∞ (χ̂))(R
/41 (β) + R∞
0 (β))
+O(δ 3/2 |u|−11/2 ) (6.90)

∗
for all (u, u) ∈ D1s .
Substituting the estimate 6.90 in 6.63 we obtain:
/ 2 χ̂′ kL4 (Su,u ) + k∇

≤ Cδ|u|−7/2 k + Cδ 1/2 |u|−5/2 (R
/41 (β) + R∞
0 (β)) + O(δ|u|
−7/2
)
(6.91)
∗
for all (u, u) ∈ D1s . In view of the estimates 6.90, 6.91, the lemma follows
taking also into account 6.52, the estimates of Proposition 4.2, as well as the
estimates of Chapter 3.

/ 2 η, ∇
/ 2η
The mass aspect function µ of the surface Su,u considered as a section of the
null hypersurface Cu is defined by:
1
µ = K + trχtrχ − div
/ η (6.92)
4
Similarly, the mass aspect function µ of the surface Su,u considered as a section
of the null hypersurface C u is defined by:
1
µ = K + trχtrχ − div
/ η (6.93)
4
195
Substituting for the Gauss curvature K from the Gauss equation 1.132, the
above definitions take the form:
1
µ = −ρ + (χ̂, χ̂) − div
/ η (6.94)
2
1
µ = −ρ + (χ̂, χ̂) − div
/ η (6.95)
2
If we consider µ as given, equation 6.94 together with equation 1.168 constitute
an elliptic system for η on each Su,u section of Cu :
1
div
/ η = −ρ + (χ̂, χ̂) − µ
2
1
curl
/ η = σ − χ̂ ∧ χ̂ (6.96)
2
Similarly, if we consider µ as given, equation 6.95 together with equation 1.163
constitute an elliptic system for η on each Su,u section of C u :
1
div
/ η = −ρ + (χ̂, χ̂) − µ
2
1
curl
/ η = −σ + χ̂ ∧ χ̂ (6.97)
2
(Note from 1.166 that if θ is any symmetric 2-covariant S tensorfield we have:
g/ ∧ θ = θ ∧ g/ = 0
Hence, if θ, θ′ is any pair of symmetric 2-covariant S tensorfields, then:
θ ∧ θ′ = θ̂ ∧ θ̂′
where θ̂ and θ̂′ are the trace-free parts of θ and θ′ respectively.) The elliptic
system 6.96 is to be considered in conjunction with a propagation equation for µ
along the generators of Cu . Similarly, the elliptic system 6.97 is to be considered
in conjunction with a propagation equation for µ along the generators of C u .
We shall presently derive these propagation equations.
We consider the expression 6.92 for µ. The Gauss curvature K satisfies along
the generators of Cu the propagation equation 5.28:
1
DK = −ΩtrχK + div / (Ωχ̂) − △
/ div / (Ωtrχ) (6.98)
2
Taking the trace of equation 1.170 and noting 1.123, the definition 6.95, and
the fact that by the first of 3.5 if θ is any 2-covariant S tensorfield we have:
Dtrθ = tr(Dθ) − 2Ω(χ, θ) (6.99)
we obtain the equation:

1 2 2
D(Ωtrχ) = Ω − trχtrχ − 2µ + 2|η| (6.100)
2
196
Writing trχtrχ = trχ′ (Ωtrχ) and using also the propagation equation 3.8 we
then deduce:

1 1 1 1 1
D(trχtrχ) = Ω − (trχ)2 trχ − trχ|χ̂|2 − trχµ + trχ|η|2 (6.101)
4 4 4 2 2
Next, we revisit the propagation equation 1.82 for η:

♯ 1
Dη = Ω χ̂ · η + trχη − β (6.102)
2
Now, from 6.4, 6.5 we can express β in the form:

1 1
−β = Ω div/ χ̂′ − /dtrχ′ − trχη + χ̂♯ · η
2 2
or:
1
− Ωβ = div
/ (Ωχ̂) − /d(Ωtrχ) (6.103)
2
1
+Ω −2χ̂♯ · /d log Ω + trχd
/ log Ω + χ̂♯ · η − trχη
2
Substituting in 6.102 and recalling that, from 1.81,
η + η = 2d
/ log Ω
yields:
1
/ (Ωχ̂) − /d(Ωtrχ) + Ωtrχη
Dη = div (6.104)
2
From Lemma 4.1 for any S 1-form ξ we have:
D∇
/ξ − ∇
/Dξ = −DΓ
/·ξ (6.105)
and:
tr(DΓ
/ · ξ) = 2(div
/ (Ωχ̂), ξ) (6.106)
Also, since div
/ ξ = tr∇
/ξ, we have, from 6.99,
Ddiv
/ ξ − tr(D∇
/ξ) = −2Ω(χ, ∇
/ξ)
Thus, taking the trace of 6.105 we obtain:
Ddiv
/ ξ − div / (Ωχ̂♯ · ξ) − Ωtrχdiv
/ Dξ = −2div / ξ (6.107)
Applying div
/ to 6.102 and substituting in 6.107 with η in the role of ξ we then
obtain:
1
− Ddiv
/ η = −div / (Ωχ̂) + △
/ div / (Ωtrχ)
2
+Ωtrχdiv
/ η + div
/ j (6.108)
197
where j is the S 1-form:
j = Ω(2χ̂♯ · η − trχη) (6.109)
Adding finally 6.98, 6.101 and 6.108 we obtain the desired propagation equation
for µ:
1
Dµ = −Ωtrχµ − Ωtrχµ
2
1 2 1 2
+Ω − trχ|χ̂| + trχ|η|
4 2
+div
/ j (6.110)
In a simmilar manner we derive the following propagation equation for µ:
1
Dµ = −Ωtrχµ − Ωtrχµ
2
1 1
+Ω − trχ|χ̂|2 + trχ|η|2
4 2
+div
/ j (6.111)
where j is the S 1-form:
j = Ω(2χ̂ · η − trχη) (6.112)
Equation 6.111 is simply the conjugate of equation 6.110. We note the crucial
fact that the right hand sides of these equations do not contain principal terms
(such terms would involve the first derivatives of the curvature or the second
derivatives of the connection coefficients).
In the following we denote by O(δ p |u|r ), for real numbers p, r, the product of
δ |u|r with a non-negative non-decreasing continuous function of the quantities
p
D0∞ , R∞ /41 , R
0 , D /41 , and D
/42 (trχ).
Lemma 6.2 We have:

/ 2 ηkL4 (Su,u ) + k∇
|u|k∇ /ηkL4 (Su,u ) + |u|−1 kηkL4 (Su,u )
≤ C|u|−5/2 Ã + O(δ 1/2 |u|−5/2 )
+ O(δ 5/2 |u|−11/2 )k
/ 2 ηkL4 (Su,u ) + k∇
≤ C|u|−5/2 Ã + O(δ 1/2 |u|−5/2 )
+ O(δ 3/2 |u|−9/2 )k
∗
for all (u, u) ∈ D1s , provided that δ is suitably small depending on D0∞ , R∞0 ,
/41 , R
D /41 , and R
/2 (α). Moreover, the coefficients of k depend only on D0∞ , R∞
0 .
Here:
/41 (ρ) + R
Ã = R /41 (σ) + R∞ ∞
0 (ρ) + R0 (σ) + R/41 (α)R∞ 0 (α)
4 ∞ ∞ 4
+(R /1 (α) + R0 (α))D0 (χ̂) + (D /1 (trχ) + D/1 (χ̂))R∞
4
0 (α)
198
Proof: We apply Lemma 5.11 to the systems 6.96 and 6.97. In view of the last
conclusion of Proposition 6.1 the conclusion of Lemma 5.11 simplifies and we
obtain:
/ 2 ηkL4 (Su,u ) + k∇
n
≤ C |u| kd /µkL4(Su,u ) + kd /ρkL4 (Su,u ) + kd
/σkL4 (Su,u )

+kd /(χ̂, χ̂)kL4 (Su,u ) + kd
/(χ̂ ∧ χ̂)kL4 (Su,u )
+kµkL4(Su,u ) + kρkL4 (Su,u ) + kσkL4 (Su,u )
o
+k(χ̂, χ̂)kL4 (Su,u ) + kχ̂ ∧ χ̂kL4 (Su,u ) (6.113)
and:
/ 2 ηkL4 (Su,u ) + k∇
n
≤ C |u| kd /µkL4 (Su,u ) + kd /ρkL4 (Su,u ) + kd
/σkL4 (Su,u )

+kd /(χ̂, χ̂)kL4 (Su,u ) + kd
/(χ̂ ∧ χ̂)kL4 (Su,u )
+kµkL4 (Su,u ) + kρkL4 (Su,u ) + kσkL4 (Su,u )
o
+k(χ̂, χ̂)kL4 (Su,u ) + kχ̂ ∧ χ̂kL4 (Su,u ) (6.114)
We have:
kd
/(χ̂, χ̂)kL4 (Su,u ) + kd
/(χ̂ ∧ χ̂)kL4 (Su,u )
n o
≤ C k∇ /χ̂kL4 (Su,u ) kχ̂kL∞ (Su,u ) + k∇
/χ̂kL4 (Su,u ) kχ̂kL∞ (Su,u )
≤ C|u|−7/2 [R /41 (trχ) + D
/41 (α)D0∞ (χ̂) + (D /41 (χ̂))R∞
0 (α)]
+ O(δ 1/2 |u|−9/2 ) (6.115)

by Propositions 4.1 and 4.2 and the results of Chapter 3. Also,
k(χ̂, χ̂)kL4 (Su,u ) + kχ̂ ∧ χ̂kL4 (Su,u )
≤ C|u|1/2 kχ̂kL∞ (Su,u ) kχ̂kL∞ (Su,u )
≤ C|u|−5/2 R∞ ∞
0 (α)D0 (χ̂) + O(δ|u|
−9/2
) (6.116)
by the results of Chapter 3. Moreover, by Proposition 4.2 and the results of
Chapter 3 we can estimate:
kµkL4 (Su,u ) , kµkL4 (Su,u ) ≤ C|u|−5/2 [R∞ ∞ ∞
0 (ρ) + R0 (α)D0 (χ̂)] + O(δ
1/2
|u|−5/2 )
(6.117)
Substituting the above in 6.113, 6.114 and taking also into account the third
and fourth of the definitions 4.1 we obtain:
/ 2 ηkL4 (Su,u ) + k∇
/µkL4 (Su,u ) + C|u|−5/2 A + O(δ 1/2 |u|−5/2 ) (6.118)
≤ C|u|kd
199
and:
/ 2 ηkL4 (Su,u ) + k∇
/µkL4 (Su,u ) + C|u|−5/2 A + O(δ 1/2 |u|−5/2 ) (6.119)
≤ C|u|kd
where:
/41 (ρ) + R
A = R /41 (σ) + R∞ ∞
0 (ρ) + R0 (σ)
/41 (α) + R∞
+(R ∞
/41 (trχ) + D
0 (α))D0 (χ̂) + (D /41 (χ̂))R∞
0 (α) (6.120)
We now turn to the propagation equations 6.110, 6.111. In view of Lemma

1.2 we deduce the following propagation equations for /dµ, /dµ:
1
Dd/µ = −Ωtrχd/µ − Ωtrχd/µ + h (6.121)
2
1
Dd/µ = −Ωtrχd/µ − Ωtrχd/µ + h (6.122)
2
where h, h are the S 1-forms:

1
h = /ddiv/ j − µ + µ /d(Ωtrχ)
2

1 1
/ Ω − trχ|χ̂|2 + trχ|η|2
+d (6.123)
4 2

1
h / j − µ + µ /d(Ωtrχ)
= /ddiv
2

1 1
/ Ω − trχ|χ̂|2 + trχ|η|2
+d (6.124)
4 2
The S 1-form /ddiv
/ j, the first term on the right in 6.123 is of the form:
/ddiv
/ j = / 2 η) + (Ωtrχ, ∇
(Ωχ̂, ∇ / 2 η)
+(∇/(Ωχ̂), ∇/η) + (d
/(Ωtrχ), ∇/η)
/ 2 (Ωχ̂), η) + (∇
+(∇ / 2 (Ωtrχ), η) (6.125)
The first two terms, placing the first factors in L∞ (S) and the second factors
in L4 (S), are bounded in L4 (S) by:
Cδ −1/2 |u|−1 R∞ / 2 ηkL4 (Su,u ) + C(|u|−1 + O(|u|−2 ))k∇
0 (α)k∇ / 2 ηkL4 (Su,u ) (6.126)
To estimate the second two terms in L4 (S), we place the first factors in L4 (S)
and the second factors in L∞ (S). We then use Lemma 5.2 with p = 4 to bound
the L∞ (S) norms of the second factors in terms of their W14 (S) norms. Using
also Propositions 4.1 and 4.2 we then obtain a bound by:
(Cδ −1/2 |u|−1 R
/41 (α) + O(|u|−2 ))k∇
/ 2 ηkL4 (Su,u ) + O(|u|−2 )k∇
/ 2 ηkL4 (Su,u )
+O(|u|−9/2 ) (6.127)
200
Finally, to estimate the last two terms in L4 (S), we place the first factors in
L4 (S) using Proposition 6.1 and the second factors in L∞ (S) using the results
of Chapter 3. We then obtain a bound by O(|u|−9/2 ). Combining the above
results yields:
kd
/div
/ jkL4 (Su,u ) ≤
Cδ −1/2 |u|−1 (R
/41 (α) + R∞
0 (α) + O(δ
1/2
|u|−1 ))k∇
/ 2 ηkL4 (Su,u )
+C|u|−1 (1 + O(|u|−1 )k∇
/ 2 ηkL4 (Su,u ) + O(|u|−9/2 ) (6.128)
/ j, the first term on the right in 6.124 is of the form:

The S 1-form /ddiv
/ddiv
/ j = / 2 η) + (Ωtrχ, ∇
(Ωχ̂, ∇ / 2 η)
+(∇/(Ωχ̂), ∇/η) + (d
/(Ωtrχ), ∇/η)
/ 2 (Ωχ̂), η) + (∇
+(∇ / 2 (Ωtrχ), η) (6.129)
The first two terms, placing the first factors in L∞ (S) and the second factors
in L4 (S), are bounded in L4 (S) by:
Cδ 1/2 |u|−2 (D0∞ (χ̂) + O(δ|u|−3/2 ))k∇
/ 2 ηkL4 (Su,u )
+C|u|−1 (1 + O(δ|u|−1 ))k∇
/ 2 ηkL4 (Su,u ) (6.130)
To estimate the second two terms in L4 (S), we place the first factors in L4 (S)
and the second factors in L∞ (S). We then use Lemma 5.2 with p = 4 to bound
the L∞ (S) norms of the second factors in terms of their W14 (S) norms. Using
also Proposition 4.2 we then obtain a bound by:
Cδ 1/2 |u|−2 (D
/41 (trχ) + D
/41 (χ̂) + O(δ 1/2 |u|−1 ))k∇
/ 2 ηkL4 (Su,u )
+O(δ|u|−3 )k∇
/ 2 ηkL4 (Su,u ) + O(δ|u|−11/2 ) (6.131)
Finally, to estimate the last two terms in L4 (S), we place the first factors in
L4 (S) and the second factors in L∞ (S) using the results of Chapter 3 to obtain
a bound by:
Cδ 1/2 |u|−2 (R∞
0 (β) + O(δ
1/2
|u|−1 ))k∇
/ 2 (Ωχ̂)kL4 (Su,u )
+Cδ 1/2 |u|−2 (R∞
0 (β) + O(δ|u|
−1
/ 2 (Ωtrχ)kL4 (Su,u )
))k∇
≤ O(δ 3/2 |u|−13/2 )k + O(δ|u|−11/2 ) (6.132)
∗
for all (u, u) ∈ D1s , by Lemma 6.1, taking also into account 6.52, the estimates
of Proposition 4.2 for ∇ /χ̂, as well as the estimates of Chapter 3. Note that the
coefficient of k depends only on D0∞ , R∞ 0 . Combining the above results yields:
kd / jkL4 (Su,u ) ≤
/div
Cδ 1/2 |u|−2 (D
/41 (trχ) + D
/41 (χ̂) + D0∞ (χ̂) + O(δ 1/2 |u|−1 ))k∇
/ 2 ηkL4 (Su,u)
+C|u|−1 (1 + O(δ|u|−2 ))k∇
/ 2 ηkL4 (Su,u )
+O(δ 3/2 |u|−13/2 )k + O(δ|u|−11/2 ) (6.133)
201
Consider next the second terms in each of 6.123, 6.124. Consider the expres-
sions 6.94, 6.95. By Lemma 5.2 with p = 4 and Proposition 4.2 we have:
k∇
/ηkL∞ (Su,u ) ≤ C|u|1/2 k∇
/ηkW14 (Su,u )
≤ C|u|1/2 k∇
/ 2 ηkL4 (Su,u ) + O(δ 1/2 |u|−3 ) (6.134)
k∇
/ηkL∞ (Su,u ) ≤ C|u|1/2 k∇
/ηkW14 (Su,u )
≤ C|u|1/2 k∇
/ 2 ηkL4 (Su,u ) + O(δ 1/2 |u|−3 ) (6.135)
Thus, taking also into account the results of Chapter 3 we obtain:
kµkL∞ (Su,u ) ≤ C|u|−3 (R∞ ∞ ∞

0 (ρ) + R0 (α)D0 (χ̂))
/ 2 ηkL4 (Su,u ) + O(δ 1/2 |u|−3 )

+C|u|1/2 k∇ (6.136)
kµkL∞ (Su,u ) ≤ C|u|−3 (R∞ ∞ ∞

0 (ρ) + R0 (α)D0 (χ̂))
+C|u|1/2 k∇
/ 2 ηkL4 (Su,u ) + O(δ 1/2 |u|−3 ) (6.137)
Placing then the first factors in the second terms in each of 6.123, 6.124 in
L∞ (S) and the second factors in L4 (S) using the results of Chapter 4 we obtain
that the second term in 6.123 is bounded in L4 (S) norm by:
C|u|−2 [R∞ /41 (α) + O(δ 1/2 )](k∇

0 (α)R / 2 ηkL4 (Su,u ) ) + O(|u|−11/2 )
/ 2 ηkL4 (Su,u ) + k∇
(6.138)
while the second term in 6.124 is bounded in L4 (S) norm by:
Cδ|u|−3 [R
/41 (ρ) + (1 + D0∞ (χ̂))(D
/41 (trχ) + D
/41 (χ̂)) + O(δ 1/2 )] ·
/ 2 ηkL4 (Su,u ) + k∇
·(k∇ / 2 ηkL4 (Su,u ) ) + O(δ|u|−13/2 )(6.139)
By the results of Chapters 3 and 4 the last term in 6.123 is bounded in L4 (S)
by:
Cδ −1 |u|−7/2 R∞ /41 (α) + O(δ −1/2 |u|−9/2 )
0 (α)R (6.140)
while the last term in 6.124 is bounded in L4 (S) by:
O(δ|u|−11/2 ) (6.141)
Combining the results 6.128, 6.138 and 6.140 yields:
khkL4(Su,u ) ≤ Cδ −1/2 |u|−1 (R

/41 (α) + R∞
0 (α) + O(δ
1/2
|u|−1 ))k∇
/ 2 ηkL4 (Su,u )
+C|u|−1 (1 + O(|u|−1 ))k∇
/ 2 ηkL4 (Su,u )
+Cδ −1 |u|−7/2 R∞ /41 (α) + O(δ −1/2 |u|−9/2 )
0 (α)R (6.142)
202
khkL4 (Su,u ) ≤ Cδ 1/2 |u|−2 (D /41 (χ̂) + D0∞ (χ̂)

/41 (trχ) + D
+O(δ 1/2 |u|−1 ))k∇
/ 2 ηkL4 (Su,u )
+C|u|−1 (1 + O(δ|u|−1 )k∇
/ 2 ηkL4 (Su,u )
+O(δ 3/2 |u|−13/2 )k + O(δ|u|−11/2 ) (6.143)
∗
for all (u, u) ∈ D1s . Substituting for k∇ / 2 ηkL4 (Su,u ) the estimates
/ 2 ηkL4 (Su,u ) , k∇
6.118, 6.119 we then obtain:
khkL4 (Su,u ) ≤ Cδ −1/2 |u|−1 (R

/41 (α) + R∞
0 (α) + O(δ
1/2
|u|−1 ))kd
/µkL4 (Su,u )
+C|u|−1 (1 + O(|u|−1 ))kd
/µkL4 (Su,u )
+Cδ −1 |u|−7/2 R∞ /41 (α) + O(δ −1/2 |u|−9/2 )
0 (α)R (6.144)
khkL4 (Su,u ) ≤ Cδ 1/2 |u|−2 (D

/41 (trχ) + D
/41 (χ̂) + D0∞ (χ̂)
+O(δ 1/2 |u|−1 ))kd
/µkL4 (Su,u )
+C|u|−1 (1 + O(δ|u|−1 ))kd
/µkL4 (Su,u ) + C|u|−9/2 A
+O(δ 3/2 |u|−13/2 )k + O(δ 1/2 |u|−9/2 )
(6.145)
where the coefficient of k depends only on D0∞ , R∞

0 .
We now turn to the propagation equations 6.121, 6.122. To 6.121 we apply
Lemma 4.6 with p = 4. Here r = 1, ν = −2,γ = 0, and we obtain:
Z u
1
kd/µkL4 (Su,u ) ≤ C − Ωtrχd/µ + h du′ (6.146)
0 2 4
L (Su′ ,u )
To 6.122 we apply Lemma 4.7 with p = 4. Here again r = 1, ν = −2, γ = 0,

and we obtain:
|u|5/2 kd
/µkL4 (Su,u ) ≤ C|u0 |5/2 kd
/µkL4 (Su,u0 )
Z u
1
+C |u′ |5/2 − Ωtrχd /µ + h du′ (6.147)
u0 2 L4 (Su,u′ )
Substituting in 6.146 and 6.147 the estimates 6.144 and 6.145 yields the following
system of linear integral inequalities for the quantities kd /µkL4 (Su,u ) , kd
/µkL4 (Su,u )
s∗
on the domain D1 :
Z u
/µkL4(Su,u ) ≤ a(u)
kd kd/µkL4 (Su′ ,u ) du′ (6.148)
0
Z u
+b(u) /µkL4 (Su′ ,u ) du′ + f (u)
kd
0
203
Z u
|u|5/2 kd
/µkL4 (Su,u ) ≤ |u′ |5/2 a(u′ )kd
/µkL4 (Su,u′ ) du′ (6.149)
u0
Z u
+ |u′ |5/2 b(u′ )kd
/µkL4(Su,u′ ) du′ + |u|5/2 f (u)
u0
Here:
a(u) = Cδ −1/2 |u|−1 (R

/41 (α) + R∞
0 (α) + O(δ
1/2
|u|−1 )) (6.150)
−1 −1
b(u) = C|u| (1 + O(|u| )) (6.151)
f (u) = C|u|−7/2 R∞
0 (α)R/41 (α) + O(δ 1/2 |u|−9/2 ) (6.152)
1/2 −2 4 4 ∞ 1/2 −1
a(u) = Cδ |u| (D /1 (trχ) + D /1 (χ̂) + D0 (χ̂) + O(δ |u| ))(6.153)
b(u) = C|u|−1 (1 + O(δ|u|−1 )) (6.154)
and:
f (u) = C|u|−5/2 |u0 |−1 D

/41 (µ) + C|u|−7/2 A
+O(δ 3/2 |u|−11/2 )k + O(δ 1/2 |u|−7/2 ) (6.155)
where we have defined:
/41 (µ) = |u0 |7/2 sup kd

D /µkL4 (Su,u0 ) (6.156)
u∈[0,u1 ]
and the coefficient of k depends only on D0∞ , R∞ 0 .

Consider first the integral inequality 6.148. At fixed u, considering kd
/µkL4 (Su,u )
as given, this inequality is of the form 6.18, with a(u) in the role of the constant
a and Z u
b(u) kd/µkL4 (Su′ ,u ) du′ + f (u)
0
in the role of the non-negative function b(u). If, as is the case here, the function
b(u) is non-decreasing, the result 6.21 implies:
x(u) ≤ eau b(u) (6.157)
Moreover, if
aδ ≤ log 2 (6.158)
6.157 in turn implies:
x(u) ≤ 2b(u) (6.159)
Condition 6.158 in the present case reads:
a(u)δ = Cδ 1/2 |u|−1 (R

/41 (α) + R∞
0 (α) + O(δ
1/2
|u|−1 )) ≤ log 2 (6.160)
and is indeed satisfied provided that δ is suitably small depending on D0∞ , R∞0 ,
/41 , R
D /41 . The result 6.159 then takes the form:
Z u
/µkL4 (Su,u ) ≤ 2b(u)
kd /µkL4 (Su′ ,u ) du′ + 2f (u)
kd (6.161)
0
204
Consider next the integral inequality 6.149. At fixed u, considering kd
/µkL4 (Su,u )
as given, this inequality is of the form 6.81 with a(u) in the role of the non-
negative function a and
Z u
|u′ |5/2 b(u′ )kd
/µkL4 (Su,u′ ) du′ + |u|5/2 f (u)
u0
in the role of the non-negative non-decreasing function b. Note in particular

from 6.155 that the function |u|5/2 f (u) is indeed non-decreasing. Recall that
the result 6.87 implies the result 6.89 provided that condition 6.88 holds. In
view of 6.153 this condition holds if
Cδ 1/2 (D /41 (χ̂) + D0∞ (χ̂) + O(δ 1/2 )) ≤ log 2

/41 (trχ) + D (6.162)
which is indeed satisfied if δ is suitably small depending on D0∞ , R∞ /41 , R

0 , D /41 .
The result 6.89 then takes the form:
Z u
|u|5/2 kd
/µkL4 (Su,u ) ≤ 2 |u′ |5/2 b(u′ )kd
/µkL4 (Su,u′ ) du′ + 2|u|5/2 f (u) (6.163)
u0
∗
We have thus reduced the system of integral inequalities 6.148, 6.149 on D1s to
the system 6.161, 6.163.
Let us set:
y(u) = sup kd
/µkL4 (Su,u ) (6.164)
u∈[0,u1 ]
y(u) = sup kd
/µkL4 (Su,u ) (6.165)
u∈[0,u1 ]
Since
Z u Z u
/µkL4 (Su′ ,u′ ) du′ du′
|u′ |5/2 b(u′ )kd
0 u0
Z u Z u
= /µkL4 (Su′ ,u′ ) du′ |u′ |5/2 b(u′ )du′
kd
u0 0
Z u
≤δ |u′ |5/2 b(u′ )y(u′ )du′
u0
replacing u by u′ in 6.163 and integrating with respect to u′ on [0, u] yields:

Z u Z u
/µkL4 (Su′ ,u ) du′ ≤ 2δ|u|−5/2
kd |u′ |5/2 b(u′ )y(u′ )du′ + 2δf (u) (6.166)
0 u0
∗
Substituting in 6.161 we then obtain, for all (u, u) ∈ D1s :
Z u
/µkL4 (Su,u ) ≤ 4δ|u|−5/2 b(u)
kd |u′ |5/2 b(u′ )y(u′ )du′
u0
+4δb(u)f(u) + 2f (u)
205
Z u
y(u) ≤ m(u) + n(u) |u′ |5/2 b(u′ )y(u′ )du′ (6.167)
u0
where:
m(u) = 4δb(u)f (u) + 2f (u) (6.168)
and:
n(u) = 4δ|u|−5/2 b(u) (6.169)
From 6.151, 6.152 and 6.155 we have:
m(u) ≤ C|u|−7/2 R∞ /41 (α) + O(δ 1/2 |u|−9/2 )

0 (α)R
+Cδ|u|−9/2 (1 + O(|u|−1 ))D
/41 (µ)
+O(δ 5/2 |u|−13/2 )k (6.170)
−7/2 −1
n(u) ≤ Cδ|u| (1 + O(|u| )) (6.171)
for new numerical constants C. Again, the coefficient of k depends only on D0∞ ,
R∞0 . Setting
Z u
Y (u) = |u′ |5/2 b(u′ )y(u′ )du′ , we have Y (u0 ) = 0 (6.172)
u0

dY
≤ |u|5/2 b(u)(m + nY ) (6.173)
du
Z u Z u
Y (u) ≤ exp |u | b(u )n(u )du |u′ |5/2 b(u′ )m(u′ )du′
′′ 5/2 ′′ ′′ ′′
(6.174)
u0 u′
for all u ∈ [u0 , s∗ ]. From 6.154 and 6.171 we have:

Z u
|u′′ |5/2 b(u′′ )n(u′′ )du′′ (6.175)
u′
Z u
≤ Cδ |u′′ |−2 (1 + O(|u′′ |−1 ))du′′ ≤ Cδ|u|−1 (1 + O(|u|−1 )) ≤ log 2
u′
the last step provided that δ is suitably small depending on D0∞ , R∞ /41 , R
0 , D /41 .
Therefore: Z u
Y (u) ≤ 2 |u′ |5/2 b(u′ )m(u′ )du′ (6.176)
u0

Z u
y(u) ≤ m(u) + 2n(u) |u′ |5/2 b(u′ )m(u′ )du′ (6.177)
u0
206
Moreover, since Z u
|u′ |5/2 b(u′ )kd
/µkL4 (Su,u′ ) du′ ≤ Y (u)
u0
substituting 6.176 in 6.163 and taking the supremum over u ∈ [0, u1 ] yields:
Z u
|u|5/2 y(u) ≤ 2|u|5/2 f (u) + 4 |u′ |5/2 b(u′ )m(u′ )du′ (6.178)
u0
/41 (µ). Recall that on Cu0

condition 3.77 holds:
η + η = 0 : on Cu0
Thus from the definitions 6.94, 6.95 we have:
µ + µ = −2ρ + (χ̂, χ̂) : on Cu0 (6.179)
Hence (see 6.115):
/µkL4 (Su,u0 ) ≤ kd
kd /µkL4(Su,u0 )
+C|u0 |−7/2 [R
/41 (ρ) + R
/41 (α)D0∞ (χ̂) + (D
/41 (trχ) + D
/41 (χ̂))R∞
0 (α)]
+O(δ 1/2 |u0 |−9/2 ) (6.180)

Substituting in 6.148 at u = u0 then yields the following linear integral inequal-
/µkL4 (Su,u0 ) :
ity for kd
Z u
kd/µkL4 (Su,u0 ) ≤ ã(u0 ) /µkL4 (Su′ ,u ) du′ + f˜(u0 )
kd (6.181)
0
0
where:
ã(u0 ) = a(u0 ) + b(u0 ) (6.182)
n
f˜(u0 ) = f (u0 ) + δb(u0 ) C|u0 |−7/2 [R
/41 (ρ) + R
/41 (α)D0∞ (χ̂)
o
/41 (trχ) + D
+(D /41 (χ̂))R∞
0 (α)] + O(δ
1/2
|u0 |−9/2 )
≤ C|u0 |−7/2 R∞ /41 (α) + O(δ 1/2 |u0 |−9/2 )
0 (α)R (6.183)
The integral inequality 6.181 implies:
/µkL4(Su,u0 ) ≤ eδã(u0 ) f˜(u0 )
kd (6.184)
Since (see 6.150, 6.151)

δã(u0 ) ≤ δ 1/2 |u0 |−1 (R
/41 (α) + R∞
0 (α) + O(δ
1/2
)) ≤ log 2 (6.185)
the last step provided that δ is suitably small depending on D0∞ , R∞ /41 , R
0 , D /41 ,
6.184 in turn implies:
kd
/µkL4 (Su,u0 ) ≤ 2f˜(u0 )
≤ C|u0 |−7/2 R∞ /41 (α) + O(δ 1/2 |u0 |−9/2 )
0 (α)R (6.186)
207
It then follows through 6.180 that:
/41 (µ) ≤ C R
4
D /1 (ρ)+
/41 (α)(D0∞ (χ̂) + R∞ /41 (trχ) + D
/41 (χ̂))R∞

R 0 (α)) + (D 0 (α)
+ O(δ 1/2 |u0 |−1 ) (6.187)
Substituting 6.187 in 6.170, the latter simplifies to:
m(u) ≤ C|u|−7/2 R∞ /41 (α) + O(δ 1/2 |u|−9/2 )

0 (α)R
+O(δ 5/2 |u|−13/2 )k (6.188)
Also, substituting 6.187 in 6.155 we obtain:
f (u) ≤ C|u|−7/2 Ã
+O(δ 3/2 |u|−11/2 )k + O(δ 1/2 |u|−7/2 ) (6.189)
where:
/41 (ρ) + R
Ã = R /41 (σ) + R∞ ∞
/41 (α)R∞
0 (ρ) + R0 (σ) + R 0 (α)
/41 (α) + R∞
+(R ∞
/41 (trχ) + D
0 (α))D0 (χ̂) + (D /41 (χ̂))R∞
0 (α) (6.190)
Substituting the definition 6.154 and the bounds 6.171 and 6.188 in 6.177 then
yields:
y(u) ≤ C|u|−7/2 R∞ /41 (α) + O(δ 1/2 |u|−9/2 ) + O(δ 5/2 |u|−13/2 )k
0 (α)R (6.191)

0 , D /41 . Also, substi-
tuting 6.189, the definition 6.154 and 6.188 in 6.178 yields:
y(u) ≤ C|u|−7/2 Ã + O(δ 3/2 |u|−11/2 )k + O(δ 1/2 |u|−7/2 ) (6.192)
provided again that δ is suitably small depending on D0∞ , R∞ /41 , R

0 , D /41 . More-
over, the coefficients of k in 6.191, 6.192 depend only on D0∞ , R∞0 .
In view of the definitions 6.164, 6.165, the estimates 6.191, 6.192 imply
through 6.118, 6.119 the estimates of the lemma.
6.5 L4(S) estimate for ∇

/ 2ω
∗
We proceed to derive an L4 (S) estimate for ∇ / 2 ω on M1s . Using this estimate
∗
4
we shall derive an L (S) estimate for ∇ / log Ω in M1s which, with a suitable
2
choice of the constant k, improves the bound 6.52. This shall enable us to show
that s∗ = u1 so that the previous lemmas actually hold on the entire parameter
domain D1 , and, since (u1 , u1 ) ∈ D′ is arbitrary, the estimates hold on all of
D′ , that is, on all of M ′ .
208
We introduce the function:
ω
/=△
/ ω − div
/ (Ωβ) (6.193)
We shall derive a propagation equation for ω

/ along the generators of the Cu .
Since
△ / 2ω
/ ω = tr∇
6.99 applies and we obtain:
D△ / 2 ω) − 2Ω(χ, ∇
/ ω = tr(D∇ / 2 ω) (6.194)
/ 2ω = ∇
On the other hand, since ∇ /(d
/ω), by Lemma 4.1 and Lemma 1.2 we have:
/ 2 ω)AB = (∇
(D∇ / 2 Dω)AB − (DΓ
/)C
AB (d
/ω)C (6.195)
and:
(g/−1 )AB (DΓ
/)C / (Ωχ̂))C
AB = 2(div (6.196)
We thus obtain:
D△
/ω = △
/ Dω − Ωtrχ△ / 2 ω) − 2(div
/ ω − 2Ω(χ̂, ∇ / (Ωχ̂), /dω) (6.197)
Moreover, from 1.107 we have:
△
/ Dω = / [Ω2 (2(η, η) − |η|2 )]
△
−Ω2 △ /(Ω2 ), /dρ) − ρ△
/ ρ − 2(d / (Ω2 ) (6.198)
The principal term on the right hand side is the term −Ω2 △
/ ρ.
Noting that
D(Ωβ) = Ω(Dβ + ωβ)
the sixth Bianchi identity of Proposition 1.2 reads:
1
D(Ωβ) + trχ(Ωβ) = Ω2 {−d
/ρ + ∗
/dσ − 3ηρ + 3 ∗ ησ + χ̂♯ · β + 2χ̂♯ · β} (6.199)
2
By 6.107 applied to the S 1-form Ωβ:
/ (Ωβ) = div
Ddiv / D(Ωβ) − Ωtrχdiv / (Ω2 χ̂♯ · β)
/ (Ωβ) − 2div (6.200)
Substituting from 6.199 then yields:

3 1
Ddiv
/ (Ωβ) = − Ωtrχdiv / (Ωβ) − (d /(Ωtrχ), Ωβ)
2 2
2 2
−Ω △ / ρ − (d /(Ω2 ), ∗/dσ)
/(Ω ), /dρ) + (d
/ [Ω2 (−χ̂♯ · β + 2χ̂♯ · β − 3ηρ + 3 ∗ ησ)]
+div (6.201)
/ ( ∗/dσ) = 0 identically, the Hessian of a function being symmetric.

Note that div
The principal term on the right hand side is the term −Ω2 △ / ρ and cancels the
209
principal term on the right hand side of 6.198 in considering the propagation
equation for ω/.
Combining 6.197, 6.198 and 6.201, we obtain the following propagation equa-
tion for the function ω
/ along the generators of the Cu :
Dω
/ + Ωtrχω / 2 ω) + m
/ = −2Ω(χ̂, ∇ (6.202)
where m is the function:

1
m / (Ωχ̂), /dω) + div
= −2(div / (Ω2 trχβ)
2
/(Ω2 ), /dρ) − (d
−(d /(Ω2 ), ∗/dσ) − ρ△
/ (Ω2 )
/ [Ω2 (2(η, η) − |η|2 )]
+△
/ [Ω2 (χ̂♯ · β − 2χ̂♯ · β + 3ηρ − 3 ∗ ησ)]
+div (6.203)
We note the crucial fact that the right hand side of equation 6.202 does not
contain principal terms (such terms would involve the second derivatives of the
curvature or the third derivatives of the connection coefficients). The propaga-
tion equation 6.202 is to be considered in conjunction with the elliptic equation
/ω = ω
△ / + div
/ (Ωβ) (6.204)
which is simply a re-writing of the definition 6.193.
Lemma 6.3 We have:
/ 2 ωkL4 (Su,u )
k∇ ≤ Cδ|u|−9/2 [R
/41 (β) + D0∞ (χ̂)R
/41 (β) + (D
/41 (trχ) + D
/41 (χ̂))R∞
0 (β)]
+O(δ 3/2 |u|−11/2 ) + O(δ 3 |u|−15/2 )k

∗
for all (u, u) ∈ D1s , provided that δ is suitably small depending on D0∞ , R∞ 0 ,
/41 , R
D /41 , and R
/2 (α). Moreover, the coefficient of k depends only on D0∞ , R∞0 .
∗
Proof: We shall first obtain a bound for kmkL4 (Su,u ) for all (u, u) ∈ D1s . The
first term on the right in 6.203 is bounded in L4 (S) by:
/(Ωχ̂)kL∞ (Su,u ) kd
Ck∇ /ωkL4 (Su,u )
and by Proposition 6.1 and Lemma 5.2 with p = 4:
/(Ωχ̂)kL∞ (Su,u ) ≤ Cδ −1/2 |u|−2 (R

k∇ /41 (β) + R∞
0 (β)) + O(|u|
−3
) (6.205)
Taking also into account the L4 (S) bound for /dω of Proposition 4.2 we then
conclude that the first term on the right in 6.203 is bounded in L4 (S) by:
O(δ 1/2 |u|−11/2 ) (6.206)
The second term on the right in 6.203 is bounded in L4 (S) by:
O(δ|u|−11/2 ) (6.207)
210
Writing (see 1.81)
1
/d log Ω =(η + η)
2
and using the results of Chapter 3, the third and fourth terms on the right in
6.203 are bounded in L4 (S) by:
O(δ 1/2 |u|−11/2 ) (6.208)
Also, writing
1
△
/ log Ω = (div
/ η + div
/ η)
2
and using the results of Chapter 4, the fifth term on the right in 6.203 is in
L4 (S) similarly bounded. The principal part of the sixth term on the right in
6.203 is:
2Ω2 ((η − η, △
/ η) + (η, △
/ η))
This part is bounded in L4 (S) using Lemma 6.2 and the results of Chapter 3
by:
O(δ 2 |u|−15/2 )k + O(δ 1/2 |u|−11/2 ) (6.209)
where the coefficient of k depends only on D0∞ , R∞0 . There is also a part of the
form:
2Ω2 (2(∇
/η, ∇ /η|2 )
/η) − |∇
This is estimated in L4 (S) by placing one factor of ∇
/η in L∞ (S) using Lemma
6.2 in conjunction with Lemma 5.2 with p = 4, which gives:
/ηkL∞ (Su,u ) ≤ C|u|−3 Ã + O(δ 1/2 |u|−3 ) + O(δ 5/2 |u|−6 )k

k∇ (6.210)
and the other factor of ∇ /η in L4 (S) using the results of Chapter 4. We

/η or ∇
4
then obtain an L (S) bound for this part by:
O(δ 3 |u|−17/2 )k + O(δ 1/2 |u|−11/2 ) (6.211)
/42 (trχ). The remainder of the sixth

where the coefficient of k is independent of D
term on the right in 6.203 is of lower order. Finally, we have the seventh term
on the right in 6.203. The part:
/ (Ω2 χ̂♯ · β)
−2div
gives the leading contribution in behavior with respect to δ. Using the results
of Chapters 3 and 4 this part is seen to be bounded in L4 (S):
C|u|−9/2 [D0∞ (χ̂)R

/41 (β) + (D
/41 (trχ) + D
/41 (χ̂))R∞
0 (β)] + O(δ
1/2
|u|−11/2 ) (6.212)
while the remainder of the seventh term is bounded in L4 (S) by:
O(δ 1/2 |u|−11/2 ) (6.213)
211
Collecting the above results (6.206 - 6.209, 6.211 - 6.213) we conclude that:
kmkL4 (Su,u ) ≤ C|u|−9/2 [D0∞ (χ̂)R
/41 (β) + (D
/41 (trχ) + D
/41 (χ̂))R∞
0 (β)]
+O(δ 2 |u|−15/2 )k + O(δ 1/2 |u|−11/2 ) (6.214)

0 , D /41 , the coefficient
4
of k being independent of D /2 (trχ).
Consider now equation 6.204. Setting θ = /dω, the S 1-form θ satisfies a
system of the form 5.218 with f = ω / + div
/ (Ωβ), g = 0. The estimate 5.231
holds. Substituting in this estimate the bound 5.187 for k1,∞ , we obtain, by
virtue of the last conclusion of Proposition 6.1, the estimate:
/θkLp(Su,u ) ≤ Cp {kf kLp(Su,u ) + kgkLp(Su,u ) }
k∇ (6.215)
In the present case taking p = 4 this yields:
/ 2 ωkL4 (Su,u ) ≤ Ckω
k∇ /41 (β) + O(δ 2 |u|−13/2 )
/kL4(Su,u ) + Cδ|u|−9/2 R (6.216)
It follows that in reference to the first term on the right in 6.202 we have:
/ 2 ω)kL4 (Su,u ) ≤ Cδ −1/2 |u|−1 R∞
kΩ(χ̂, ∇ 0 (α)kω/kL4 (Su,u ) + O(δ 1/2 |u|−11/2 )
(6.217)
We now apply Lemma 4.6 with p = 4 to the propagation equation 6.202. Here
r = 0, ν = −2, γ = 0 and we obtain:
Z u
kω
/kL4(Su,u ) ≤ C −2Ω(χ̂, ∇ / 2 ω) + m L4 (S ′ ) du′ (6.218)
u ,u
0
Substituting the estimates 6.218 and 6.214 we obtain the following linear integral
∗
inequality for the quantity kω /kL4(Su,u ) on the domain D1s :
Z u
kω/kL4(Su,u ) ≤ a(u) /kL4(Su′ ,u ) du′ + b(u)
kω (6.219)
0
Here:
a(u) = Cδ −1/2 |u|−1 R∞
0 (α) (6.220)
b(u) = Cδ|u|−9/2 [D0∞ (χ̂)R
/41 (β) + (D
/41 (trχ) + D
/41 (χ̂))R∞
0 (β)]
+O(δ 3/2 |u|−11/2 ) + O(δ 3 |u|−15/2 )k (6.221)
where the coefficient of k is independent of D /42 (trχ). At fixed u the inequality
6.219 is of the form 6.18, with a(u) in the role of the constant a and the constant
b(u) in the role of the non-negative function b(u). Then 6.157 holds, which, since
a(u)δ = Cδ 1/2 |u|−1 R∞
0 (α) ≤ log 2
provided that δ is suitably small depending on R∞

0 (α), implies 6.159, that is:
/kL4(Su,u )
kω ≤ /41 (β) + (D
Cδ|u|−9/2 [D0∞ (χ̂)R /41 (trχ) + D
/41 (χ̂))R∞
0 (β)]
+O(δ 3/2 |u|−11/2 ) + O(δ 3 |u|−15/2 )k (6.222)
212
/42 (trχ). Substituting in 6.216 then
where the coefficent of k is independent of D
yields the lemma.

k = C[D0∞ (χ̂)R
/41 (β) + (D
/41 (trχ) + D
/41 (χ̂))R∞
0 (β)] + O(δ
1/2
)
/ 2 log ΩkL4 (Su,u )
k∇ ≤ Cδ|u|−7/2 [R
/41 (β) + D0∞ (χ̂)R
/41 (β)
/41 (trχ) + D
+(D /41 (χ̂))R∞
0 (β)] + O(δ
3/2
|u|−9/2 )
holds for all (u, u) ∈ D1 , provided that δ is suitably small depending on D0∞ ,
R∞ /41 , R
0 , D /41 , and R
/2 (α).
Proof: By Lemma 4.1 and the second of the definitions 1.30:
/ 2 log Ω = ∇
D∇ / 2 ω − DΓ
/ · /d log Ω (6.223)
and we have:
/)C
(DΓ /A (Ωχ)BC + ∇
AB = ∇ /B (Ωχ)AC − ∇
/C (Ωχ)AB
By Lemma 6.1 and Lemma 5.2 with p = 4 (and the estimate of Proposition 4.2
for trχ):
/(Ωχ)kL∞ (Su,u ) ≤ Cδ 1/2 |u|−3 (R

k∇ /41 (β)+R∞
0 (β))+Cδ|u|
−4
k+O(δ|u|−4 ) (6.224)
hence also:
/kL∞ (Su,u ) ≤ Cδ 1/2 |u|−3 (R
kDΓ /41 (β) + R∞
0 (β)) + Cδ|u|
−4
k + O(δ|u|−4 ) (6.225)
∗
for all (u, u) ∈ D1s . Using then also Lemma 4.11 we can estimate:
/ · /d log ΩkL4 (Su,u )
kDΓ ≤ /kL∞ (Su,u ) kd
kDΓ / log ΩkL4 (Su,u )
≤ O(δ 3/2 |u|−11/2 ) + O(δ 2 |u|−13/2 )k (6.226)
where the coefficient of k is independent of D /42 (trχ). We now apply Lemma 4.7
with p = 4 to equation 6.223. Here r = 2, ν = 0, γ = 0. Taking into account
the fact that ∇/ 2 log Ω vanishes on Cu0 we obtain:
Z u
|u|3/2 k∇
/ 2 log ΩkL4 (Su,u ) ≤ |u′ |3/2 ∇
/ 2 ω − DΓ / · /d log Ω L4 (S ′ ) du′
u,u
u0
(6.227)
/ 2 ω in L4 (S) of Lemma
Substituting the estimate 6.226 and the estimate for ∇
6.3 we then obtain:
k∇ ≤ Cδ|u|−7/2 [R
/41 (β) + D0∞ (χ̂)R
/41 (β)
/41 (trχ) + D
+(D /41 (χ̂))R∞
0 (β)] + O(δ
3/2
|u|−9/2 )
+O(δ 2 |u|−11/2 )k (6.228)
213
/42 (trχ). In reference to this estimate,
where the coefficient of k is independent of D
let:
a = /41 (β) + D0∞ (χ̂)R

C[R /41 (β) + (D
/41 (trχ) + D
/41 (χ̂))R∞
0 (β)] + O(δ
1/2
)
b = O(1) (6.229)
so that the coefficient of k in 6.228 is δ 2 |u|−11/2 b, b being independent of D

/42 (trχ).
The estimate 6.228 then implies:
∗
/ 2 log ΩkL4 (Su,u ) ≤ δ(a + δbk)|u|−7/2 : ∀(u, u) ∈ D1s
k∇ (6.230)
Choosing:
k = 2a (6.231)
we have:
a + δbk < 2a provided that 2bδ < 1 (6.232)
The last is a smallness condition on δ depending on D0∞ , R∞ /41 , R
0 , D /41 . The
estimate 6.230 then implies:
∗
/ 2 log ΩkL4 (Su,u ) < kδ|u|−7/2
k∇ : ∀(u, u) ∈ D1s = [0, u1 ] × [0, s∗ ] (6.233)
∗
6.4, the results of Lemmas 6.1, 6.2 and 6.3 hold for all (u, u) ∈ D1 . Since (u1 , u1 )
is arbitrary, these results hold for all (u, u) ∈ D′ . Moreover, by virtue of the
L4 (S) estimate for ∇ / 2 log Ω of Lemma 6.4, we can now estimate (see 6.73):
k∇
/rkL4 (Su,u ) ≤
/41 (β) + (D
Cδ|u|−11/2 C[D0∞ (χ̂)R /41 (trχ) + D
/41 (χ̂))R∞
0 (β)]
+O(δ 3/2 |u|−13/2 ) (6.234)
Using this bound in place of the bound 6.74, we deduce, following the argument
leading to the estimate 6.90,
/ 2 trχ′ kL4 (Su,u ) ≤ Cδ|u|−9/2 B + O(δ 3/2 |u|−11/2 ) : ∀(u, u) ∈ D1

k∇ (6.235)
where:
B = /42 (trχ) + (D
D /41 (trχ) + D
/41 (χ̂))2 (6.236)
/41 (trχ) + D
+(D /41 (χ̂) + D0∞ (χ̂))(R
/41 (β) + R∞ /41 (β)
0 (β)) + R
214
/ 2 trχkL4 (Su,u ) ≤ Cδ|u|−9/2 B + O(δ 3/2 |u|−11/2 )

k∇
/ 2 χ̂kL4 (Su,u ) + k∇
≤ Cδ 1/2 |u|−5/2 (R
/41 (β) + R∞
0 (β)) + O(δ|u|
−7/2
)
2
/ ηkL4 (Su,u ) + k∇
≤ C|u|−5/2 Ã + O(δ 1/2 |u|−5/2 )
/ 2 ηkL4 (Su,u ) + k∇
≤ C|u|−5/2 Ã + O(δ 1/2 |u|−5/2 )
/ 2 ωkL4 (Su,u ) ≤ Cδ|u|−9/2 [R
k∇ /41 (β) + D0∞ (χ̂)R
/41 (β) + (D
/41 (trχ) + D
/41 (χ̂))R∞
0 (β)]
+O(δ 3/2 |u|−11/2 )

0 , D /41 , and R
/2 (α).
Here B and Ã are defined by 6.236 and 6.190 respectively.
6.6 L4(S) estimate for ∇

/ 2ω
Proposition 6.3 The following estimate holds for all (u, u) ∈ D′ :
/ 2 ωkL4 (Su,u ) ≤ Cδ −1/2 |u|−5/2 R

k∇ /41 (β) + O(|u|−7/2 )

0 , D /41 , and R
/2 (α).
Proof: We introduce the conjugate of the function ω / (see 6.193):
ω
/=△
/ ω + div
/ (Ωβ) (6.237)
Proceeding as in the derivation of the propagation equation 6.202 we derive the

following propagation equation for ω/, the conjugate of 6.202:
/ + Ωtrχω
Dω / 2 ω) + m
/ = −2Ω(χ̂, ∇ (6.238)
where m is the function:

1
m / (Ωχ̂), /dω) − div
= −2(div / (Ωtrχβ)
2
/(Ω2 ), /dρ) + (d
−(d /(Ω2 ), ∗/dσ) − ρ△
/ (Ω2 )
2 2
+△
/ [Ω (2(η, η) − |η| )]
/ [Ω2 (−χ̂♯ · β + 2χ̂♯ · β + 3ηρ + 3 ∗ ησ)]
+div (6.239)
The propagation equation 6.238 is to be considered in conjunction with the

elliptic equation
△/ω = ω/ − div
/ (Ωβ) (6.240)
215
We shall first obtain a bound for kmkL4 (Su,u ) . The first term on the right in
6.239 is bounded in L4 (S) by:
/(Ωχ̂)kL∞ (Su,u ) kd
Ck∇ /ωkL4(Su,u )
and by Proposition 6.2 and Lemma 5.2 with p = 4:
/(Ωχ̂)kL∞ (Su,u ) ≤ Cδ 1/2 |u|−3 (R

k∇ /41 (β) + R∞
0 (β)) + O(δ|u|
−4
) (6.241)
Taking also into account the L4 (S) bound for /dω of Proposition 4.3 we then
conclude that the first term on the right in 6.239 is bounded in L4 (S) by:
O(δ 1/2 |u|−11/2 ) (6.242)
The second term on the right in 6.239 is bounded in L4 (S) by:
Cδ −1/2 |u|−7/2 R
/41 (β) + O(δ 1/2 |u|−11/2 ) (6.243)
The third, fourth and fifth terms on the right in 6.239 are bounded in L4 (S) by:
O(δ|u|−13/2 ) (6.244)
by Lemmas 6.4 and 4.11, which in conjunction with Lemma 5.2 with p = 4
imply:
/ log ΩkL∞ (Su,u ) ≤ O(δ|u|−3 )
kd (6.245)
The sixth term on the right in 6.239 is bounded in L4 (S) by:
O(δ 1/2 |u|−11/2 ) (6.246)
using Lemma 6.4, Proposition 6.2, the estimate 6.245 and the results of Chapters
3 and 4. Finally, using the results of Chapters 3 and 4 we deduce that the seventh
term on the right in 6.239 is bounded in L4 (S) by:
O(|u|−9/2 ) (6.247)
Collecting the above results (6.242 - 6.244, 6.246) we conclude that:
kmkL4 (Su,u ) ≤ Cδ −1/2 |u|−7/2 R

/41 (β) + O(|u|−9/2 ) (6.248)

system of the form 5.218 with f = ω / − div
/ (Ωβ), g = 0. The estimate 6.215
holds. In the present case taking p = 4 yields:
k∇ /kL4 (Su,u ) + Cδ −1/2 |u|−5/2 R

/ 2 ωkL4 (Su,u ) ≤ Ckω /41 (β) + O(δ 1/2 |u|−9/2 ) (6.249)
/ 2 ω)kL4 (Su,u ) ≤ Cδ 1/2 |u|−2 (D0∞ (χ̂)+O(δ|u|−3/2 ))kω

kΩ(χ̂, ∇ /kL4(Su,u ) +O(|u|−9/2 )
(6.250)
216
|u|3/2 kω
/kL4(Su,u ) ≤ C|u0 |3/2 kω
/kL4(Su,u0 ) (6.251)
Z u
+C |u′ |3/2 −2Ω(χ̂, ∇
/ 2ω) + m L4 (Su,u′ )
du′
u0
Substituting the estimates 6.250 and 6.248 and noting that in view of the fact
that ω vanishes on Cu0 we have, from 6.237,
/kL4(Su,u0 ) ≤ Cδ −1/2 |u0 |−5/2 R

kω /41 (β) + O(δ 1/2 |u0 |−9/2 ), (6.252)
we obtain the following linear integral inequality for the quantity |u|3/2 kω
/kL4(Su,u ) :
Z u
|u|3/2 kω
/kL4(Su,u ) ≤ a(u′ )|u′ |3/2 kω
/kL4(Su,u′ ) du′ + b(u) (6.253)
u0
where:
a(u) = Cδ 1/2 |u|−2 (D0∞ (χ̂) + O(δ|u|−3/2 )) (6.254)

−1/2 −1
b(u) = Cδ |u| /41 (β)
R −2
+ O(|u| ) (6.255)
At fixed u the inequality 6.253 is of the form 6.81 with the quantity |u|3/2 kω
/kL4(Su,u )
in the role of x(u). Note that here the function b(u) is non-decreasing. Moreover,
we have Z u
a(u′ )du′ ≤ log 2
u0

0 . It follows that 6.89
holds, that is:
/kL4(Su,u ) ≤ Cδ −1/2 |u|−5/2 R

kω /41 (β) + O(|u|−7/2 ) (6.256)
Substituting finally this in 6.249 yields the lemma.
217
Chapter 7
L2 Estimates for the 3rd

Derivatives of the
7.1 Introduction
In the present chapter we shall assume that besides the quantity R
/2 (α) defined
by 5.5 also the following quantities are finite:
R
/2 (β) = sup (|u|3 k∇
/ 2 βkL2 (Cu ) )
u∈[u0 ,c∗ )
R
/2 (ρ) = sup (|u|4 δ −1/2 k∇
/ 2 ρkL2 (Cu ) )
u∈[u0 ,c∗ )
R
/2 (σ) = sup (|u|4 δ −1/2 k∇
/ 2 σkL2 (Cu ) )
u∈[u0 ,c∗ )
/2 (β) =
R sup (|u|5 δ −3/2 k∇
/ 2 βkL2 (Cu ) ) (7.1)
u∈[u0 ,c∗ )
and:
R
/1 (Dρ) = sup (|u|3 δ 1/2 kd
/DρkL2(Cu ) )
u∈[u0 ,c∗ )
R
/1 (Dσ) = sup (|u|3 δ 1/2 kd
/DσkL2 (Cu ) )
u∈[u0 ,c∗ )
R0 (D2 ρ) = sup (|u|2 δ 3/2 kD2 ρkL2 (Cu ) )

u∈[u0 ,c∗ )
R0 (D2 σ) = sup (|u|2 δ 3/2 kD2 σkL2 (Cu ) ) (7.2)

u∈[u0 ,c∗ )
and also:
R
/1 (Dβ) = sup (|u|5 δ −3/2 k∇
/DβkL2 (Cu ) ) (7.3)
u∈[u0 ,c∗ )
218
replacing the supremum on [u0 , c∗ ) by the value at u0 , are all bounded by a
non-negative non-decreasing continuous function of M7 , the quantity requiring
Mk with the highest k being the one corresponding to ∇/Dβ, which is expressed
in terms of ∇
/div
/ α by the fourth Bianchi identity of Proposition 1.2.

/ 2 η, ∇
/ 2η
As we have already remarked, the estimates of Chapter 6 for ∇ / 2 η, ∇
/ 2 η in L4 (S),
loose a factor of δ 1/2 in behavior with respect to δ in comparison with the L4 (S)
estimates for ∇ /η, ∇/η of Chapter 4. As a preliminary step we shall presently
derive L2 estimates for ∇ / 2 η, ∇
/ 2 η, using only the propagation equations, which
are similar to the estimates of Chapter 4 in behavior with respect to δ. We
consider any (u1 , u1 ) ∈ D′ and fix attention to the parameter subdomain D1
and the corresponding subdomain M1 of M ′ (see 3.45, 3.46).
We apply Lemma 4.1 to the propagation equations 4.120 to obtain:
1
/ 2 η)ABC
(D∇ = / 2 η)ABC + Ωχ̂D
Ωtrχ(∇ / 2 η)ABD + b′ABC
C (∇
2
1
/ 2 η)ABC
(D∇ = / 2 η)ABC + Ωχ̂D
Ωtrχ(∇ C
/ 2 η)ABD + b′ABC
(∇ (7.4)
2
where:
b′ABC = −(DΓ/)DAB ∇
/D ηC − (DΓ /)D
AC ∇
/B ηD
1
+ /dA (Ωtrχ)∇ /A (Ωχ̂D
/B η C + ∇ C )∇
/B η D
2
+∇
/A bBC
b′ABC = −(DΓ/)D /)D
/D η C − (DΓ
AB ∇ AC ∇
/B η D
1
+ /dA (Ωtrχ)∇ /A (Ωχ̂D
/B ηC + ∇ C
)∇
/B ηD
2
+∇
/A bBC (7.5)
/ 2 η in the role of θ and

To the first of 7.4 we apply Lemma 4.6 with ∇
1
/ 2 η + Ωχ̂ · ∇
Ωtrχ∇ / 2 η + b′
2
in the role of ξ. Then r = 3, ν = 0, γ = 0. In view of 4.121, 4.122 we obtain,
taking p = 2:
Z u
k∇/ 2 ηkL2 (Su,u ) ≤ Cã(u) k∇/ 2 ηkL2 (Su′ ,u ) du′
Z u0
+C kb′ kL2 (Su′ ,u ) du′ (7.6)
0
219
/ 2 η in the role of θ and
To the second of 7.4 we apply Lemma 4.7 with ∇
1
/ 2 η + Ωχ̂ · ∇
Ωtrχ∇ / 2η + b
2
in the role of ξ. Then r = 3, ν = 0, γ = 0. In view of 4.124, 4.125 we obtain,
taking p = 2:
Z u
|u|2 k∇
/ 2 ηkL2 (Su,u ) ≤ C|u0 |2 k∇
/ 2 ηkL2 (Su,u0 ) + C / 2 ηkL2 (Su,u′ ) du′
|u′ |2 ãk∇
u0
Z u
+C |u′ |2 kb′ kL2 (Su,u′ ) du′ (7.7)
u0
Defining:
/2 (η) = |u0 |3 δ −1/2 sup k∇
D / 2 ηkL2 (Su,u0 ) (7.8)
u∈[0,u1 ]
and taking into account 4.127, the inequality 7.7 reduces to:
Z u
|u|2 k∇
/ 2 ηkL2 (Su,u ) ≤ C|u0 |−1 δ 1/2 D
/2 (η) + C |u′ |k∇
/ 2 ηkL2 (Su,u′ ) du′
u0
Z u
′ 2 ′
+C |u | kb kL2 (Su,u′ ) du′ (7.9)
u0
The inequalities 7.6, 7.9 constitute a system of linear integral inequalities

/ 2 ηkL2 (Su,u ) , k∇
for the quantities k∇ / 2 ηkL2 (Su,u ) on the domain D1 . Setting:
z ′ (u) = / 2 ηkL2 (Su,u )

sup k∇ (7.10)
u∈[0,u1 ]
replacing u by u′ ∈ [0, u] in 7.9 and integrating with respect to u′ on [0, u] yields,

for all (u, u) ∈ D1 :
Z u Z u
2 2 ′ 3/2 −1
|u| / ηkL2 (Su′ ,u ) du ≤ Cδ |u0 | D
k∇ /2 (η) + Cδ |u′ |z ′ (u′ )du′
0 u0
Z u Z u1
+C ′ 2
|u | kb′ kL2 (Su′ ,u′ ) du′ du′ (7.11)
u0 0
Substituting 7.11 in 7.6 and taking the supremum over u ∈ [0, u1 ] then yields
the linear integral inequality:
Z u
z ′ (u) ≤ λ′ (u) + ν ′ (u) |u′ |z ′ (u′ )du′ (7.12)
u0
where:
Z u Z u1
′ −2 3/2 −1 ′ 2 ′ ′ ′
λ (u) = C|u| ã(u) δ |u0 | D /2 (η) + |u | kb kL2 (Su′ ,u′ ) du du
u0 0
Z u1
+C kb′ kL2 (Su′ ,u ) du′ (7.13)
0
220
and:
ν ′ (u) = Cδ|u|−2 ã(u) (7.14)
Setting Z u
Z ′ (u) = |u′ |z ′ (u′ )du′ , we have Z ′ (u0 ) = 0 (7.15)
u0

dZ ′
≤ |u|(λ′ + ν ′ Z ′ ) (7.16)
du
Z u Z u
′
Z (u) ≤ exp |u |ν (u )du |u′ |λ′ (u′ )du′ : ∀u ∈ [u0 , u1 ]
′′ ′ ′′ ′′
(7.17)
u0 u′

Z u Z u
|u′′ |ν ′ (u′′ )du′′ = Cδ |u′′ |−1 ã(u′′ )du′′ ≤ 1 (7.18)
u′ u′
provided that δ is suitably small depending on D0∞ , R∞ 0 . Therefore:

Z u
Z ′ (u) ≤ C |u′ |λ′ (u′ )du′ : ∀u ∈ [u0 , u1 ] (7.19)
u0
and 7.12 reads:

z ′ (u) ≤ λ′ (u) + ν ′ (u)Z ′ (u) (7.20)
2
Moreover, taking the L norm of 7.9 with respect to u on [0, u1 ], we obtain, in
view of the definitions 7.10 and 7.15,
|u|2 k∇
/ 2 ηkL2 (Cuu1 ) ≤ Cδ|u0 |−1 D
/2 (η) + Cδ 1/2 Z ′ (u)
Z u
+C |u′ |2 kb′ kL2 (C u1 ) du′ (7.21)
u0 u′
u
Here we denote by Cu 1 the part of Cu which corresponds to u ≤ u1 . This follows
since for any S tensorfield θ we have:
!
Z Z Z u1 Z u1
kθk2L2 (C u1 ) = u
|θ|2 = |θ|2 dµ/g du = kθk2L2 (Su,u ) du (7.22)
u
Cu 1 0 Su,u 0
(see 5.6, 5.7). In deriving 7.21 we have also used the fact that
1/2
kθkL2 (Cuu1 ) ≤ u1 sup kθkL2 (Su,u ) ≤ δ 1/2 sup kθkL2 (Su,u ) (7.23)
u∈[0,u1 ] u∈[0,u1 ]
to bound Z u
/ 2 ηkL2 (C u1 ) du′ ≤ δ 1/2 Z ′ (u)
|u′ |k∇
u0 u′
221
/2 (η). To do this we recall
that along Cu0 we have:
η = −η : along Cu0 (7.24)
Thus, the first of 7.4 simplifies along Cu0 to:
1
(D∇ / 2 η)ABC = −Ωχ̂D
/ 2 η)ABC + Ωtrχ(∇ / 2 η)ABD + b′ABC
C (∇ (7.25)
2
We apply to Lemma 4.6 with ∇ / 2 η in the role of θ and b′ in the role of ξ. Here
♯
r = 3, ν = −1, γ = −ΩI ⊗ χ̂ . Taking p = 2 we obtain:
Z u
2
/ ηkL2 (Su,u0 ) ≤ C
k∇ kb′ kL2 (Su′ ,u ) du′ (7.26)
0
0
which, in view of 7.24 implies:

Z u1
/2 (η) ≤ C|u0 |3 δ −1/2
D kb′ kL2 (Su′ ,u ) du′ (7.27)
0
0
Consider the integrals involving b′ , b′ in 7.13, 7.21 and 7.27. The principal
terms in b and b from the point of view of differentiability are the terms −Ω∇/β
and Ω∇/β respectively (see 4.118). Thus the principal parts of b′ and b′ are:
(P )
′
b / 2β
= −Ω∇
(P )
′
b / 2β
= Ω∇ (7.28)
and we have:
Z u (P ) Z u1
1
k b′ kL2 (Su′ ,u ) du′ ≤ C / 2 βkL2 (Su′ ,u ) du′
k∇
0 0
Z u1 1/2
1/2
≤ Cu1 / 2 βk2L2 (Su′ ,u ) du′
k∇
0
≤ Cδ 1/2 k∇
/ 2 βkL2 (Cu ) ≤ Cδ 1/2 |u|−3 R
/2 (β)(7.29)
and:
Z u1 (P ) Z u1
k b′ kL2 (Su′ ,u ) du′ ≤ C / 2 βkL2 (Su′ ,u ) du′
k∇
0 0
Z u1 1/2
1/2 2 ′
≤ Cu / βkL2 (Su′ ,u ) du
k∇
0
≤ / βkL2 (Cu ) ≤ Cδ 2 |u|−5 R
Cδ 1/2 k∇ 2
/2 (β) (7.30)
hence: Z u Z u1 (P )
|u′ |2 k b′ kL2 (Su′ ,u′ ) du′ du′ ≤ Cδ 2 |u|−2 R
/2 (β) (7.31)
u0 0
222
Also:
(P )
k b′ kL2 (C u1 ) ≤ Ck∇
/ 2 βkL2 (Cu ) ≤ Cδ 3/2 |u|−5 R
/2 (β) (7.32)
u
Writing
(P ) (N )
b′ = b + b′
′
(P ) (N )
b′ = b + b′
′
(7.33)
(N ) (N )
the non-principal parts b′ and b′ can be estimated in L2 (S) using the results
of Chapters 3, 4 and 6. We obtain:
(N )
k b′ kL2 (Su,u ) ≤ O(|u|−4 )
(N )
k b′ kL2 (Su,u ) ≤ O(δ|u|−5 ) (7.34)
Combining with 7.29 - 7.32 we then obtain:

Z u
1
kb′ kL2 (Su′ ,u ) du′ ≤ Cδ 1/2 |u|−3 R
/2 (β) + O(δ|u|−4 ) (7.35)
0
Z u1
kb′ kL2 (Su′ ,u ) du′ ≤ Cδ 2 |u|−5 R
/2 (β) + O(δ 2 |u|−5 ) (7.36)
0
Z u Z u1
′ 2
|u | kb′ kL2 (Su′ ,u′ ) du′ du′ ≤ Cδ 2 |u|−2 R
/2 (β) + O(δ 2 |u|−2 ) (7.37)
u0 0
kb′ kL2 (Cuu1 ) ≤ Cδ 3/2 |u|−5 R

/2 (β) + O(δ 3/2 |u|−5 ) (7.38)
In particular:
Z u1
kb′ kL2 (Su′ ,u ) du′ ≤ Cδ 1/2 |u0 |−3 R
/2 (β) + O(δ|u0 |−4 ) (7.39)
0
0
hence, from 7.27:

D /2 (β) + O(δ 1/2 |u0 |−1 )
/2 (η) ≤ CR (7.40)
Substituting the estimates 7.40, 7.37 and 7.35 in 7.13 (and recalling 4.122)
we obtain:
λ′ (u) ≤ Cδ 1/2 |u|−3 R
/2 (β) + O(δ|u|−4 ) (7.41)
Here and in the following we denote by O(δ p |u|r ), for real numbers p, r, the
product of δ p |u|r with a non-negative non-decreasing continuous function of the
quantities D0∞ , R∞ /41 , R
0 , D /41 , D
/42 (trχ), and R
/2 , where:
R
/2 = max{R
/2 (α), R
/2 (β), R
/2 (ρ), R
/2 (σ), R
/2 (β)} (7.42)
The bound 7.41 implies (see 7.19):
Z ′ (u) ≤ Cδ 1/2 |u|−1 R

/2 (β) + O(δ|u|−2 ) (7.43)
223
hence from 7.20 (recalling 7.14 and 4.122) we obtain:
z ′ (u) ≤ Cδ 1/2 |u|−3 R
/2 (β) + O(δ|u|−4 ) : ∀u ∈ [u0 , u1 ] (7.44)
Also, substituting the estimates 7.40, 7.43 and 7.38 in 7.21 we obtain:
/ 2 ηkL2 (Cuu1 ) ≤ Cδ|u|−3 R
k∇ /2 (β) + O(δ 3/2 |u|−4 ) : ∀u ∈ [u0 , u1 ] (7.45)
In view of the definition 7.10 and the fact that (u1 , u1 ) ∈ D′ is arbitrary the
estimates 7.44 and 7.45 yield the following Proposition:

/ ηkL2 (Su,u ) ≤ Cδ 1/2 |u|−3 R
k∇ 2
/2 (β) + O(δ|u|−4 ) : ∀(u, u) ∈ D′
/ 2 ηkL2 (Cu ) ≤ Cδ|u|−3 R

k∇ /2 (β) + O(δ 3/2 |u|−4 ) : ∀u ∈ [u0 , c∗ )
0 , D /41 , and R
/2 (α).
7.3 L2 elliptic theory for generalized Hodge sys-

tems on S
To estimate in L2 the third derivatives of the connection cefficients we shall
make use of the following two lemmas from Chapter 2 of [C-K] (Lemmas 2.2.2
and 2.2.3) for geneneralized Hodge systems on a 2-dimensional compact Rie-
mannian manifold (S, g/). For the sake of a self-contained presentation, we shall
also give the proofs of these lemmas. Let ξ be a totally symmetric r + 1 covari-
ant tensorfield on S. We define div/ ξ and curl
/ ξ to be the totally symmetric r
covariant tensorfields on S, given in a arbitrary local frame field by:
(div
/ ξ)A1 ...Ar = /B ξBA1 ...Ar
∇ (7.46)
(curl
/ ξ)A1 ...Ar = /ǫ BC ∇
/B ξCA1 ...Ar (7.47)
We also define trξ to be the totally symmetric r − 1 covariant tensorfield on S,
given in an arbitrary local frame field by:
(trξ)A1 ...Ar−1 = (g/−1 )BC ξBCA1 ...Ar−1 (7.48)
If r = 0, then by convention trξ = 0.
Lemma 7.1 Let ξ be a totally symmetric r + 1 covariant tensorfield on a

2-dimensional compact Riemannian manifold (S, g/), satisfying the generalized
Hodge system:
div
/ ξ = f
curl
/ ξ = g
trξ = h
224
where f, g are given totally symmetric r covariant tensorfields on S and h is a
given totally symmetric r − 1 covariant tensorfield on S. Then
Z Z
/ξ|2 + (r + 1)K|ξ|2 }dµ/g = {|f |2 + |g|2 + rK|h|2 }dµ/g
{|∇
S S
Proof: According to the definition of curl

/ ξ:
∇ /B ξCA1 ...Ar − /ǫ BC gA1 ...Ar

/C ξBA1 ...Ar = ∇
Differentiating and commuting covariant derivatives we obtain:
∇
/D ∇
/C ξBA1 ...Ar = ∇ /B ξCA1 ...Ar − /ǫ BC ∇
/D ∇ /D gA1 ...Ar
X r
M
= ∇ /D ξCA1 ...Ar −
/B ∇ R/Ai MBD ξCA 1 ...>Ai <...Ar
i=1
M
−R
/CMBD ξA 1 ...Ar
− /ǫ BC ∇
/D gA1 ...Ar (7.49)
where
R
/ABCD = (g/AC g/BD − g/AD g/BC )K
is the curvature tensor of (S, g/). Taking the trace of 7.49 by contracting with
(g/−1 )CD we deduce, using the first and third of the equations of the Hodge
system,
△
/ ξBA1 ...Ar = /C gA1 ...Ar
/B fA1 ...Ar − /ǫ BC ∇
∇
Xr
+(r + 1)KξBA1 ...Ar − g/Ai B KhA1 ...>Ai <...Ar (7.50)
i=1
BA1 ...Ar
Contracting then with ξ to form on the left hand side the inner product
(ξ, △
/ ξ), integrating on S, and integrating by parts on the left, yields the lemma.
Let ξ be a totally symmetric r + 1 covariant tensorfield on S. The sym-

metrized covariant derivative of ξ, denoted by (∇ /ξ)s , is the totally symmetric
r + 2 covariant tensorfield on S given by:
r+1
!
s 1 X
/ξ)BA1 ...Ar =
(∇ /B ξA1 ...Ar+1 +
∇ / Ai ξ
∇ B (7.51)
r+2 i=1
A1 ...>Ai <...Ar+1
Also, we denote by ( ∗ ξ)s the symmetrized dual of ξ, a totally symmetric r + 1

covariant tensorfield on S given by:
r+1
1 X B
( ∗ ξ)sA1 ...Ar+1 = /ǫ ξ (7.52)
r + 1 i=1 Ai A1 ...>Ai <...Ar+1
B
Lemma 7.2 Let ξ be a totally symmetric r + 1 covariant tensorfield on a 2-

dimensional Riemannian manifold (S, g/) satisfying the generalized Hodge system
225
of Lemma 7.1 for given f, g, h. Then ξ ′ = (∇
/ξ)s satisfies a similar system:
/ ξ′
div = f′
/ ξ′
curl = g′
trξ ′ = h′
where
1
f′ = /f )s −
(∇ ( ∗∇/g)s
r+2
2K
+(r + 1)Kξ − (g/ ⊗s h)
r+1
r+1
g′ = /g)s + (r + 1)K( ∗ ξ)s
(∇
r+2
2 r
h′ = f+ (∇/h)s
r+2 r+2
Here, we denote by ( ∗ ∇
/g)s is the totally symmetric r + 1 covariant tensorfield
given by:
r+1
1 X B
( ∗∇
/g)sA1 ...Ar+1 = /ǫ ∇ /B gA1 ...>Ai <...Ar+1
r + 1 i=1 Ai
Also, we denote by g/ ⊗s h the symmetrized tensor product of g/ and h, a totally
symmetric r + 1 covariant tensorfield given by:
X
(g/ ⊗s h)A1 ...Ar+1 = g/Ai Aj hA1 ...>Ai <...>Aj <...Ar+1
i<j=1,...,r+1
Proof: From 7.51 we have:

r+1
( )
′ 1 X
B
/ ξ )A1 ...Ar+1 =
(div / ξA1 ...Ar+1 +
△ ∇
/ ∇/ Ai ξ B (7.53)
r+2 i=1
A1 ...>Ai <...Ar+1
Consider first the sum on the right hand side. Commuting covariant derivatives
we deduce:
/B ∇
∇ / Ai ξ B = /BMBAi ξA
/Ai fA1 ...Ar+1 + R
∇ M
1 ...>Ai <...Ar+1
A1 ...>Ai <...Ar+1
X
MB
+ /Aj MBAi ξA
R 1 ...>Aj <...>Ai <...Ar+1
j=1,...,r+1
j6=i
= /Ai fA1 ...>Ai <...Ar+1 + (r + 1)KξA1 ...Ar+1 (7.54)
∇
X
− Kg/Aj Ai hA1 ...>Aj <...>Ai <...Ak+1
j=1,...,r+1
j6=i
It follows that:
r+1
X
/B ∇
∇ / Ai ξ B = /f )sA1 ...Ar+1
(r + 1)(∇ (7.55)
A1 ...>Ai <...Ar+1
i=1
+(r + 1)2 KξA1 ...Ar+1 − 2K(g/ ⊗s h)A1 ...Ar+1
226
We turn to the first term in parenthesis on the right in 7.53. This is given
by 7.50 with B replaced by Ar+1 . Now the left hand side of 7.50 is totally
symmetric. Therefore we may totally symmetrize the right hand side to obtain
a totally symmetric formula for △ / ξ. The formula which we obtain in this way
is:
2K
△ /f )s − ( ∗ ∇
/ ξ = (∇ /g)s + (r + 1)Kξ − (g/ ⊗s h) (7.56)
r+1
Combining 7.56 with 7.55 yields the formula for f ′ of the lemma.
From 7.51 we have:
r+1
( )
′ 1 BC
X
BC
/ ξ )A1 ...Ar+1 =
(curl /ǫ ∇ /C ξA1 ...Ar+1 +
/B ∇ /ǫ ∇ / Ai ξ
/B ∇ C
r+2 i=1
A1 ...>Ai <...Ar+1
(7.57)
In regard to the first term in parenthesis on the right we have:
1 BC
/ǫ BC ∇
/B ∇
/C ξA1 ...Ar+1 = /ǫ (∇ /C ξA1 ...Ar+1 − ∇
/B ∇ /C ∇/B ξA1 ...Ar+1 )
2
r+1
1 BC X M
= /ǫ R
/Ai MBC ξA 1 ...>Ai <...Ar+1
2 i=1
= (r + 1)K( ∗ ξ)sA1 ...Ar+1 (7.58)
Consider next the sum on the right in 7.57. Commuting covariant derivatives
we deduce:
/ǫ BC ∇
/B ∇/ Ai ξ C =
A1 ...>Ai <...Ar+1

/ǫ BC ∇ / Ai ∇/B ξ C
M
/CMBAi ξA
+R 1 ...>Ai <...Ar+1
A1 ...>Ai <...Ar+1



X
MC
+ /Aj MBAi ξA
R 1 ...>Aj <...>Ai <...Ar+1
j=1,...,r+1


j6=i
=∇/Ai gA1 ...>Ai <...Ar+1 (7.59)
 

 

X
− Kǫ/BC g/CAi ξ B − g
/ Aj B ξ Ai C
 A1 ...>Ai <...Ar+1 A1 ...>Aj <...>Ai <...Ar+1 
 j=1,...,r+1 
j6=i
It follows that:
r+1
X
/ǫ BC ∇ /g)s + (r + 1)2 K( ∗ ξ)s
/C ξA1 ...Ar+1 = (r + 1)(∇
/B ∇ (7.60)
i=1
Combining 7.60 with 7.58 we obtain the formula for g ′ of the lemma. Finally,
the formula for h′ is obtained in a straightforward manner.
227
Consider now, as in Lemma 5.6, a trace-free symmetric 2-covariant tensor-
field on S satisfying the equation
div
/ /g θ = f (7.61)
where f is a given 1-form on S. Then, noting the identity:
/ǫ AB /ǫ CD = g/AC g/CD − g/AD g/BC (7.62)
we have:
∗
/ θ)A = /ǫ AB (curl
(curl / θ)B = /ǫ AB /ǫ CD ∇
/C θDB
C −1 BD D −1 BC
= (δA (g/ ) − δA (g/ ) )∇ /C θDB = ∇ /B θAB = −(div
/A trθ − ∇ / θ)A
that is:
∗ ∗
curl
/ θ = −div
/ θ, curl
/ θ= div
/ θ (7.63)
Therefore θ satisfies the generalized Hodge system:
div
/ θ = f
∗
curl
/ θ = f
trθ = 0 (7.64)
Defining as in Lemma 7.2 θ′ to be the totally symmetric 3-covariant tensorfield

on S:
θ′ = (∇
/θ)s (7.65)
that is:
′ 1
θABC = (∇
/A θBC + ∇
/B θAC + ∇
/C θAB ) (7.66)
3
we have:
′ 1
∇
/A θBC = θABC + (2∇
/A θBC − ∇
/B θAC − ∇
/C θAB ) (7.67)
3
Since
∇
/A θBC − ∇ / θ)C = /ǫ AB ∗ fC
/B θAC = /ǫ AB (curl
/θ in terms of θ′ and f :
7.67 becomes the following expression for ∇
′ 1
∇
/A θBC = θABC + (ǫ/AB ∗ fC + /ǫ AC ∗ fB ) (7.68)
3
According to Lemma 7.2 θ′ satisfies the generalized Hodge system:
/ θ′
div = f′
/ θ′
curl = g′
trθ′ = h′ (7.69)
228
where:
1
f′ = /f )s − ( ∗ ∇
(∇ / ∗ f )s + 2Kθ
3
2 1
= (∇/f )s − g/div
/ f + 2Kθ
3 3
2
g′ = (∇/ ∗ f )s + 2K( ∗ θ)s
3
2
h′ = f (7.70)
3
Defining next as in Lemma 7.2 θ′′ to be the totally symmetric 4-covariant
tensorfield on S:
θ′′ = (∇
/θ′ )s (7.71)
that is:
′′ 1 ′ ′ ′ ′
θABCD = (∇
/A θBCD +∇
/B θACD +∇
/C θABD +∇
/D θABC ) (7.72)
4
we have:
′ ′′ 1 ′ ′ ′ ′
∇
/A θBCD = θABCD + (3∇
/A θBCD −∇
/B θACD −∇
/C θABD −∇
/D θABC ) (7.73)
4
Since
′ ′
∇
/A θBCD −∇
/B θACD / θ′ )CD = /ǫ AB gCD
= /ǫ AB (curl ′
/θ′ in terms of θ′′ and g ′ :

′ ′′ 1 ′ ′ ′
∇
/A θBCD = θABCD + (ǫ/AB gCD + /ǫ AC gBD + /ǫ AD gBC ) (7.74)
4
According to Lemma 7.2 θ′′ satisfies the generalized Hodge system:
/ θ′′
div = f ′′
/ θ′′
curl = g ′′
trθ′′ = h′′ (7.75)
where:
1
f ′′ /f ′ )s − ( ∗ ∇
= (∇ /g ′ )s
4
2K
+3Kθ′ − (g/ ⊗s h′ )
3
3
g ′′ = /g ′ )s + 3K( ∗ θ′ )s
(∇
4
1 ′ 1
h′′ = f + (∇ /h′ )s (7.76)
2 2
229
Substituting 7.70 in 7.76 we deduce the following pointwise bounds:
|f ′′ | ≤ C{|∇
/ 2 f | + |K||∇
/θ| + |d
/K||θ|}
′′
|g | ≤ C{|∇ / 2 f | + |K||∇
/θ| + |d
/K||θ|}
|h′′ | ≤ C{|∇
/f | + |K||θ|} (7.77)
According to Lemma 7.1 θ′′ satisfies the following integral identity on S:

Z Z
/θ′′ |2 + 4K|θ′′ |2 }dµ/g = {|f ′′ |2 + |g ′′ |2 + 3K|h′′ |2 }dµ/g
{|∇ (7.78)
S S
In regard to the second term in the integrant on the left hand side we write:
Z Z Z
K|θ′′ |2 dµ/g = K|θ′′ |2 dµ/g + (K − K)|θ′′ |2 dµ/g (7.79)
S S S
From this point on we take S to be diffeomorphic to S 2 . Then according to the

Gauss-Bonnet theorem: Z
Kdµ/g = 4π (7.80)
S
Denoting as in Chapter 5
r
Area(S) 1
r= so that K =
4π r2
we estimate the second integral on the right in 7.79 by:
kK − KkL2 (S) kθ′′ k2L4 (S) ≤ CI′ (S)rkK − KkL2 (S) {k∇
/θ′′ k2L2 (S) + r−2 kθ′′ k2L2 (S) }
by Lemma 5.1 with p = 4. Thus, the left hand side of 7.78 is greater than or
equal to:
h i
1 − 4CI′ (S)rkK − KkL2 (S) k∇/θ′′ k2L2 (S) + r−2 kθ′′ k2L2 (S)

(7.81)
Hence, if
1
4CI′ (S)rkK − KkL2 (S) ≤ (7.82)
2
the left hand side of 7.78 is greater than or equal to:
1h i
/θ′′ k2L2 (S) + r−2 kθ′′ k2L2 (S)
k∇ (7.83)
2
Substituting the pointwise bounds 7.77 on the right hand side of 7.78 we then
conclude that:
/θ′′ kL2 (S) ≤ C k∇

2
k∇ / f kL2 (S) + k|K||∇/θ|kL2 (S) + k|d /K||θ|kL2 (S)
Z 1/2
+C |K|{|∇ /f |2 + |K|2 |θ|2 }dµ/g (7.84)
S
230
We estimate:
k|K||∇
/θ|kL2 (S) ≤ kKkL4(S) k∇/θkL4 (S)
h i
−3/2
≤ Cr 1 + r3/2 kK − KkL4 (S) k∇
/θkL4 (S) (7.85)
noting that kKk4L4 (S) = 4πr−6 . Also:
k|d
/K||θ|kL2(S) ≤ kd
/KkL2(S) kθkL∞ (S)
q
≤ C I′ (S)r1/2 kd
/KkL2(S) (k∇/θkL4 (S) + r−1 kθkL4 (S) ) (7.86)
by Lemma 5.2 with p = 4. Moreover, the last term on the right in 7.84 is
bounded by:
1/2
C kKkL2(S) k∇ /f k2L4(S) + kKk3L3(S) kθk2L∞ (S) (7.87)

1/2 3/2
≤ C kKkL2 (S) k∇ /f kL4(S) + kKkL3 (S) kθkL∞ (S)
1/2
≤ C ′ r−1/2 1 + rkK − KkL2 (S)

k∇
/f kL4 (S)
q h i3/2
+ C ′ I′ (S)r−3/2 1 + r3/2 kK − KkL4 (S) (k∇/θkL4 (S) + r−1 kθkL4 (S) )
noting that kKk2L2 (S) = 4πr−2 , while kKkL3 (S) ≤ (4πr2 )1/12 kKkL4 (S) . We
substitute 7.85 - 7.87 in 7.84. Assuming that
r3/2 kK − KkL4 (S) ≤ 1 (7.88)
noting that rkK − KkL2 (S) ≤ (4π)1/4 r3/2 kK − KkL4 (S) and recalling from the
statement of Lemma 5.1 that by definition I′ (S) ≥ 1, we then obtain:
/θ′′ kL2 (S)

k∇ / 2 f kL2 (S) + r−1/2 k∇
≤ C{k∇ /f kL4(S) (7.89)
q
+ I′ (S)r−3/2 (1 + r2 kd
/KkL2(S) )(k∇/θkL4 (S) + r−1 kθkL4 (S) )}
Now, from 7.68, 7.74, we have, pointwise:
/ 3 θ|
|∇ / 2 θ′ | + C|∇
≤ |∇ / 2f |
/ 2 θ′ |
|∇ /θ′′ | + C|∇
≤ |∇ /g ′ | (7.90)
while from 7.70 we have, pointwise:
/g ′ | ≤ C{|∇
|∇ / 2 f | + |K||∇
/θ| + |d
/K||θ|} (7.91)
Therefore:
/ 3 θkL2 (S)
k∇ /θ′′ kL2 (S)
≤ k∇ (7.92)
2
+C k∇ / f kL2 (S) + k|K||∇
/θkL2 (S) + k|d
/K||θ|kL2 (S)
231
Substituting the estimates 7.85, 7.86 and noting the assumption 7.88 we then
obtain:
/ 3 θkL2 (S)
k∇ /θ′′ kL2 (S) + k∇
≤ C{k∇ / 2 f kL2 (S) (7.93)
q
+ I′ (S)r−3/2 (1 + r2 kd /θkL4 (S) + r−1 kθkL4 (S) )}
/KkL2(S) )(k∇
Combining 7.93 with 7.89 we establish the following lemma.
Lemma 7.3 Let θ be a trace-free symmetric 2-covariant tensorfield on a 2-

dimensional Riemannian manifold (S, g/), with S diffeomorphic to S 2 , satisfying
the equation
div
/ θ=f
where f is a given 1-form on S. Then there is a numerical constant C such that
if
1
4CI′ (S)rkK − KkL2 (S) ≤
2
while
r3/2 kK − KkL4 (S) ≤ 1
the following estimate holds:
/ 3 θkL2 (S)
k∇ / 2 f kL2 (S) + r−1/2 k∇
≤ C{k∇ /f kL4(S)
q
+ I′ (S)r−3/2 (1 + r2 kd
/KkL2(S) )(k∇/θkL4 (S) + r−1 kθkL4 (S) )}
Consider now , as in Lemma 5.9, a 1-form on S satisfying the Hodge system:
div
/ θ = f
curl
/ θ = g (7.94)
where (f, g) is a given pair of functions on S. Defining as in Lemma 7.2 θ′ to

be the symmetric 2-covariant tensorfield on S:
θ′ = (∇
/θ)s (7.95)
that is:
′ 1
θAB = (∇
/ A θB + ∇
/ B θA ) (7.96)
2
we have:
′ 1
∇
/A θB = θAB + (∇
/ A θB − ∇
/ B θA ) (7.97)
2
Since
∇
/ A θB − ∇
/B θA = /ǫ AB curl
/ θ = /ǫ AB g
/θ in terms of θ′ and g:
′ 1
∇
/A θB = θAB + /ǫ AB g (7.98)
2
232
According to Lemma 7.2 θ′ satisfies the generalized Hodge system:
/ θ′
div = f′
/ θ′
curl = g′
trθ′ = h′ (7.99)
where:
1 ∗
f′ = /df − /dg + Kθ
2
1
g′ = /dg + K ∗ θ
2
h′ = f (7.100)
We have the pointwise bounds:
|f ′ | ≤ C{|d
/f | + |d
/g| + |K||θ|}
|g ′ | ≤ C{|d
/g| + |K||θ|}
′
|h | = |f | (7.101)
′
According to Lemma 7.1 θ satisfies the following integral identity on S:
Z Z
/θ′ |2 + 2K|θ′ |2 }dµ/g = {|f ′ |2 + |g ′ |2 + K|h′ |2 }dµ/g
{|∇ (7.102)
S S
From this point on we take again S to be diffeomorphic to S 2 . Following the

argument leading to 7.81 we deduce that the left hand side of 7.102 is greater
than or equal to:
h i
[1 − 2CI′ (S)rkK − KkL2 (S) ] k∇
/θ′ k2L2 (S) + r−2 kθ′ k2L2 (S) (7.103)
Hence if
1
2CI′ (S)rkK − KkL2 (S) ≤ (7.104)
2
1h i
k∇/θ′ k2L2 (S) + r−2 kθ′ k2L2 (S) (7.105)
2
conclude that:
/θ′ kL2 (S)
k∇ ≤ C{kd
/f kL2(S) + kd /gkL2(S) + k|K||θ|kL2 (S) }
Z 1/2
+C |K||f |2 dµ/g (7.106)
S
We estimate (see 7.85):

k|K||θ|kL2 (S) ≤ kKkL4(S) kθkL4 (S)
h i
≤ Cr−3/2 1 + r3/2 kK − KkL4 (S) kθkL4 (S) (7.107)
233
Moreover, the last term on the right in 7.106 is bounded by:
1/2 1/2
CkKkL2 (S) kf kL4(S) ≤ C ′ r−1/2 1 + rkK − KkL2 (S)

kf kL4(S) (7.108)
(see 7.87). It follows that under the assumption 7.88 the inequality 7.106 implies:
/θ′ kL2 (S) ≤ C{kd

k∇ /gkL2(S) +r−3/2 |θkL4 (S) +r−1/2 kf kL4(S) } (7.109)
/f kL2(S) +kd
Now from 7.106 we have, pointwise:

/ 2 θ| ≤ |∇
|∇ /θ′ | + C|d
/g| (7.110)
therefore:
/ 2 θkL2 (S) ≤ k∇
k∇ /θ′ kL2 (S) + Ckd
/gkL2 (S) (7.111)
In view of 7.109, 7.111 we conclude that:
/ 2 θkL2 (S) ≤ C{kd

k∇ /gkL2(S) + r−1/2 kf kL4(S) + r−3/2 kθkL4 (S) }
/f kL2(S) + kd
(7.112)
Defining next as in Lemma 7.2 θ′′ to be the totally symmetric 3-covariant
tensorfield on S:
θ′′ = (∇
/θ′ )s (7.113)
that is:
′′ 1 ′ ′ ′
θABC = (∇
/A θBC +∇
/B θAC +∇
/C θAB ) (7.114)
3
we have:
′ ′′ 1 ′ ′ ′
∇
/A θBC = θABC + (2∇
/A θBC −∇
/B θAC −∇
/C θAB ) (7.115)
3
Since
′ ′
∇
/A θBC −∇
/B θAC / θ′ )C = /ǫ AB gC
= /ǫ AB (curl ′
/θ′ in terms of θ′′ and g ′ :

′ ′′ 1 ′ ′
∇
/A θBC = θABC + (ǫ/AB gC + /ǫ AC gB ) (7.116)
3
According to Lemma 7.2 θ′′ satisfies the Hodge system:
/ θ′′
div = f ′′
/ θ′′
curl = g ′′
trθ′′ = h′′ (7.117)
where:
1
f ′′ = /f ′ )s − ( ∗ ∇
(∇ /g ′ )s
3
+2Kθ′ − K(g/ ⊗s h′ )
2
g ′′ = /g ′ )s + 2K( ∗ θ′ )s
(∇
3
2 ′ 1
h′′ = f + (∇ /h′ )s (7.118)
3 3
234
Substituting 7.100 in 7.118 we deduce the following pointwise bounds:
|f ′′ | ≤ C{|∇
/ 2 f | + |∇
/ 2 g| + |K||∇
/θ| + |d
/K||θ|}
|g ′′ | ≤ C{|∇
/ 2 g| + |K||∇
/θ| + |d
/K||θ|}
|h′′ | ≤ C{|d
/f | + |d
/g| + |K||θ|} (7.119)
According to Lemma 7.1 θ′′ satisfies the following integral identity on S:

Z Z
/θ | + 3K|θ | }dµ/g = {|f ′′ |2 + |g ′′ |2 + 3K|h′′ |2 }dµ/g
{|∇ ′′ 2 ′′ 2
(7.120)
S S
Following the argument leading to 7.81 we deduce that the left hand side of
7.120 is greater than or equal to:
h i
1 − 3CI′ (S)rkK − KkL2 (S) k∇ /θ′′ k2L2 (S) + r−2 kθ′′ k2L2 (S)

(7.121)
Hence, if
1
3CI′ (S)rkK − KkL2 (S) ≤ (7.122)
2
1h i
k∇/θ′′ k2L2 (S) + r−2 kθ′′ k2L2 (S) (7.123)
2
conclude that:
/θ′′ kL2 (S)

k∇ / 2 f kL2 (S) + k∇
≤ C{k∇ / 2 gkL2 (S) }
+C{k|K||∇
/θ|kL2(S) + k|d/K||θ|kL2(S) } (7.124)
Z 1/2
+C |K|{|∇/f |2 + |∇
/g|2 + |K|2 |θ|2 }dµ/g
S
Proceeding as in the argument leading from 7.84 to 7.89 we deduce:
/θ′′ kL2 (S)

k∇ / 2 f kL2 (S) + k∇
≤ C{k∇ / 2 gkL2 (S) (7.125)
−1/2
+r (k∇/f kL4(S) + k∇/gkL4(S) )
q
/KkL2(S) )(k∇
under the assumption 7.88.

Now, from 7.98, 7.116 we have, pointwise:
/ 3 θ| ≤
|∇ / 2 θ′ | + C|∇
|∇ / 2 g|
/ 2 θ′ | ≤
|∇ /θ′′ | + C|∇
|∇ /g ′ | (7.126)
while from 7.100 we have, pointwise:
/g ′ | ≤ C{|∇
|∇ / 2 g| + |K||∇
/θ| + |d
/K||θ|} (7.127)
235
Therefore:
/ 3 θkL2 (S)
k∇ ≤ /θ′′ k
k∇ (7.128)
2
+C{k∇
/ gkL2 (S) + k|K||∇
/θ|kL2 (S) + k|d
/K||θ|kL2 (S) }
Substituting the analogues of the estimates 7.85, 7.86 and noting the assumption
7.88 we then obtain:
/ 3 θkL2 (S)
k∇ /θ′′ kL2 (S) + k∇
≤ C{k∇ / 2 gkL2 (S) (7.129)
q
/KkL2(S) )(k∇
Combining 7.129 with 7.125 and recalling 7.112 we conclude that the following
lemma has been established.
Lemma 7.4 Let θ be a 1-form on a 2-dimensional Riemannian manifold

(S, g/), with S diffeomorphic to S 2 , satisfying the Hodge system
div
/ θ = f
curl
/ θ = g
where (f, g) is a given pair of functions on S. Then there is a numerical constant
C such that if
1
3CI′ (S)rkK − KkL2 (S) ≤
2
while
r3/2 kK − KkL4 (S) ≤ 1
the following estimates hold:
/ 2 θkL2 (S)
k∇ ≤ C{kd /gkL2(S) + r−1/2 kf kL4 (S) + r−3/2 kθkL4 (S) }
/f kL2(S) + kd
/ 3 θkL2 (S)
k∇ / 2 f kL2 (S) + k∇
≤ C{k∇ / 2 gkL2 (S) + r−1/2 (k∇
/f kL4 (S) + k∇
/gkL4 (S) )
q
+ I′ (S)r−3/2 (1 + r2 kd/KkL2(S) )(k∇/θkL4 (S) + r−1 kθkL4 (S) )}

/ 3 χ′ , /dK
kd
/KkL2(Su,u ) ≤ Cδ 1/2 |u|−3 (R /41 (β) + R∞
/2 (β) + R 0 (β))
+O(δ|u|−4 )
/ 3 trχ′ kL2 (Su,u ) ≤

k∇
+C|u|−4 [R∞
0 (α)R /41 (α) + R∞
/2 (β) + (R /41 (β) + R∞
0 (α))(R 0 (β))]
1/2 −5
+O(δ |u| )
236
for all (u, u) ∈ D′ , and:
/ 3 χ̂′ kL2 (Cu )

k∇ ≤ C|u|−3 (R /41 (β) + R∞
/2 (β) + R 0 (β))
+O(δ 1/2 |u|−4 )
for all u ∈ [u0 , c∗ ), provided that δ is suitably small depending on D0∞ , R∞ /41 ,
0 , D
R/41 , and R
/2 . In particular, under such a smallness condition on δ we have:
|u|2 kd
/KkL2(Su,u ) ≤ 1
Proof: We apply Lemma 7.3 with χ̂′ in the role of θ to the Codazzi equation
6.4. The assumptions of Lemma 7.3 are satisfied by virtue of the first estimate
of Proposition 6.1 provided that δ is suitably small depending on D0∞ , R∞ /41 ,
0 , D
4
R
/1 and R/2 (α). Taking also into account Lemma 5.4 we obtain:
n
/ 3 χ̂′ kL2 (Su,u ) ≤ C k∇
k∇ / 3 trχ′ kL2 (Su,u ) + k∇
/ 2 ikL2 (Su,u ) (7.130)
o
+|u|−1/2 (k∇
/ 2 trχ′ kL4 (Su,u ) + k∇
/ikL4 (Su,u ) )
h in o
+C|u|−3/2 1 + |u|2 kd /KkL2(Su,u ) k∇ /χ̂′ kL4 (Su,u ) + |u|−1 kχ̂′ kL4 (Su,u )
Now, by the estimate 6.50,
/χ̂′ kL4 (Su,u ) + |u|−1 kχ̂′ kL4 (Su,u ) ≤ Cδ −1/2 |u|−3/2 (R

k∇ /41 (β) + R∞
0 (β)) + O(|u|
−5/2
)
(7.131)
Also, by the estimates 6.13 and 6.45,
|u|−1/2 (k∇ /ikL4(Su,u ) ) ≤ Cδ −1/2 |u|−3 R

/ 2 trχ′ kL4 (Su,u ) + k∇ /41 (β) + O(|u|−4 )
(7.132)
Moreover, using Propositions 5.1 and 7.1 as well as the results of Chapters 3
and 4 we deduce:
/ 2 ikL2 (Su,u ) ≤ Ck∇

k∇ / 2 βkL2 (Su,u ) + O(|u|−4 ) (7.133)

n o
/ 3 χ̂′ kL2 (Su,u ) ≤ C k∇
k∇ / 3 trχ′ kL2 (Su,u ) + k∇
/ 2 βkL2 (Su,u )
+Cδ −1/2 |u|−1 l(u)kd
/KkL2(Su,u ) (7.134)
+Cδ −1/2 |u|−3 (R
/41 (β) + R∞
0 (β)) + O(|u|
−4
)
where:
/41 (β) + R∞
l(u) = C(R 0 (β)) + O(δ
1/2
|u|−1 ) (7.135)
We now revisit the propagation equation for the Gauss curvature, equation
5.28. Applying /d to this equation we obtain, by virtue of Lemma 1.2, the
following propagation equation for /dK:
1
Dd
/K + Ωtrχd
/K = −Kd
/(Ωtrχ) + /ddiv
/ div
/ (Ωχ̂) − /d△
/ (Ωtrχ) (7.136)
2
237
We apply Lemma 4.6 to this equation, taking p = 2. Here r = 1, ν = −2, γ = 0.
We obtain:
Z u
kd
/KkL2(Su,u ) ≤ C kKd/(Ωtrχ)kL2 (Su′ ,u ) du′ (7.137)
0
Z u
1
+C kd
/div
/ div
/ (Ωχ̂) − /d△ / (Ωtrχ)kL2 (Su′ ,u ) du′
0 2
The integrant the first integral on the right in 7.137 is bounded by:
kKd
/(Ωtrχ)kL2 (S) ≤ Kkd
/(Ωtrχ)kL2 (S) + kK − KkL4 (S) kd
/(Ωtrχ)kL4 (S)
≤ O(|u|−4 ) (7.138)
by the first estimate of Proposition 6.1 and the results of Chapters 3 and 4. The
integrant of the second integral on the right in 7.137 is bounded by:
/ 3 trχ′ kL2 (S) + k∇

C(k∇ / 3 χ̂′ kL2 (S) )
+C{kd / 2 trχ′ kL2 (S) + k∇
/ log ΩkL∞ (S) (k∇ / 2 χ̂′ kL2 (S) )
/ 2 log ΩkL4 (S) (kd
+k∇ /trχ′ kL4 (S) + k∇
/χ̂′ kL4 (S) )
/ 3 log ΩkL2 (S) (ktrχ′ kL∞ (S) + kχ̂′ kL∞ (S) )}
+k∇
Using the estimates of Proposition 6.1 and Lemma 6.4 and writing (see 5.17):
1 2
/ 3 log Ω =
∇ (∇ / 2 η)
/ η+∇ (7.139)
2
the lower order terms are seen to be bounded by:
Cδ −1/2 |u|−1 (R∞

0 (α) + O(δ
1/2
/ 2 ηkL2 (S) + k∇
))(k∇ / 2 ηkL2 (S) ) + O(δ 1/2 |u|−5 )
Substituting the above in 7.137 yields:

Z un o
kd
/KkL2(Su,u ) ≤ C k∇/ 3 trχ′ kL2 (Su′ ,u ) + k∇
/ 3 χ̂′ kL2 (Su′ ,u )
0
+C|u|−1 (R∞
0 (α) + O(δ
1/2
/ 2 ηkL2 (Cu ) + k∇
))(k∇ / 2 ηkL2 (Cu ) )
+O(δ|u|−4 ) (7.140)
Substituting the estimates of Proposition 7.1 this simplifies to:

Z un o
kd
/KkL (Su,u ) ≤ C
2 / 3 trχ′ kL2 (Su′ ,u ) + k∇
k∇ / 3 χ̂′ kL2 (Su′ ,u )
0
+O(δ|u|−4 ) (7.141)
Substituting 7.141 in 7.134 yields a linear integral inequality, for fixed u, for
the quantity:
x(u) = k∇ / 3 χ̂′ kL2 (Su,u ) (7.142)
238
of the form: Z u
x(u) ≤ a x(u′ )du′ + b(u) (7.143)
0
Here a is the non-negative constant:
a = Cδ −1/2 |u|−1 l(u) (7.144)
and b(u) is the non-negative function:

Z u
b(u) = / 3 trχ′ kL2 (Su,u ) + Cδ −1/2 |u|−1 l(u)
Ck∇ / 3 trχ′ kL2 (Su′ ,u ) du′
k∇
0
/ 2 βkL2 (Su,u ) + Cδ −1/2 |u|−3 (R
+Ck∇ /41 (β) + R∞
0 (β)) + O(|u|
−4
)
(7.145)
Since
aδ = Cδ 1/2 |u|−1 l(u) ≤ log 2 (7.146)
0 , /41 ,
D /41 ,
R we deduce,
following the same argument as that leading from 6.18 to 6.23,
Z u
x(u) ≤ 2a b(u′ )du′ + b(u) (7.147)
0
From 7.145 and 7.146 (see 6.41),

Z u Z u
′ ′
b(u )du ≤ C k∇/ 3 trχ′ kL2 (Su′ ,u ) du′ (7.148)
0 0
+Cδ 1/2 |u|−3 (R /41 (β) + R∞
/2 (β) + R 0 (β)) + O(δ|u|
−4
)
recalling the first of the definitions 7.1. Therefore, substituting in 7.147 we

obtain:
/ 3 χ̂′ kL2 (Su,u )

k∇ / 3 trχ′ kL2 (Su,u )
≤ Ck∇
Z u
k∇
0
/ 2 βkL2 (Su,u ) + Cδ −1/2 |u|−3 (R
+Ck∇ /41 (β) + R∞
0 (β))
+O(|u|−4 ) (7.149)
/ 3 trχ′ :
D∇ / 3 trχ′ = −2Ω(χ̂, ∇
/ 3 χ̂′ ) + e′ (7.150)
where:
e′ABC = /)D
−(DΓ / 2 trχ′ )DC − (DΓ
AB (∇ /)D / 2 trχ′ )BD
AC (∇
−d / 2 trχ′ )BC − 2∇
/A (Ωtrχ)(∇ /A (Ωχ̂DE )∇ /C χ̂′DE
/B ∇
+∇
/A eBC (7.151)
239
and we denote:
/ 3 χ̂′ )ABC = χ̂DE ∇
(χ̂, ∇ /A ∇ /C χ̂′DE
/B ∇ (7.152)
2
To estimate in L (S) the first four terms on the right in 7.151 we place each of
the two factors of each term in L4 (S) using Propositions 4.1 and 6.1. This gives
a bound in L2 (S) for the sum of these terms by:
Cδ −1 |u|−4 R
/41 (α)(R
/41 (β) + R∞
0 (β)) + O(δ
−1/2
|u|−5 ) (7.153)
The contribution to k∇ /ekL2(S) of the first three terms in e as given by 6.27 is

/ 2 rkL2 (S) .
similarly bounded. Finally, the contribution of the last term in e is k∇
Now r is given by 6.2. Using the estimates of Proposition 6.1 and Lemma 6.4
as well as the expression 7.139 and the results of Chapters 3 and 4 we deduce:
/ 2 rkL2 (Su,u ) ≤
k∇
Cδ −1 |u|−2 [(R∞ 2
/ 2 ηkL2 (Su,u ) + k∇
0 (α)) + O(δ)](k∇ / 2 ηkL2 (Su,u ) )
+O(|u|−6 ) (7.154)
Combining the above results we obtain:
ke′ kL2 (Su,u ) ≤ Cδ −1 |u|−2 [(R∞ 2

/ 2 ηkL2 (Su,u ) + k∇
0 (α)) + O(δ)](k∇ / 2 ηkL2 (Su,u ) )
+Cδ −1 |u|−4 R
/41 (α)(R
/41 (β) + R∞
0 (β)) + O(δ
−1/2
|u|−5 ) (7.155)
We also estimate in L2 (S) the first term on the right in 7.150:
/ 3 χ̂′ )kL2 (Su,u ) ≤

kΩ(χ̂, ∇
n
Cδ −1/2 |u|−1 R∞
0 (α) / 3 trχ′ kL2 (Su,u )
k∇
Z u
+Cδ −1/2 |u|−1 l(u) k∇ / 3 trχ′ kL2 (Su′ ,u ) du′
0
−1/2 −1 ∞ 2
+Cδ |u| R0 (α)k∇
/ βkL2 (Su,u )
−1
+Cδ |u|−4 R∞ /41 (β) + R∞
0 (α)(R 0 (β)) + O(δ
−1/2
|u|−5 )(7.156)
by the estimate 7.149.

We obtain:
Z u
k∇/ 3 trχ′ kL2 (Su,u ) ≤ C / 3 χ̂′ ) + e′ kL2 (Su′ ,u ) du′
k − 2Ω(χ̂, ∇ (7.157)
0
Substituting the bounds 7.156, 7.155 then yields a linear integral inequality for
/ 3 trχ′ kL2 (S) . Recalling 6.41 the contribution of the term in parenthesis in
k∇
7.156 involving the integral is bounded by
Cδ 1/2 |u|−1 l(u) (7.158)
times the contribution of the first term in parenthesis in 7.156, and the factor
7.158 is less than or equal to 1 provided that δ is suitably small depending on
240
D0∞ , R∞ /41 , R
0 , D /41 . Taking also into account Proposition 7.1 and the first of the
definitions 7.1, noting that
Z u
/ 2 ηkL2 (Su′ ) du′ ≤ δ 1/2 k∇
k∇ / 2 ηkL2 (Cu ) ,
0
Z u
/ 2 βkL2 (Su′ ,u ) du′ ≤ δ 1/2 k∇
k∇ / 2 βkL2 (Cu ) ,
0
the resulting integral inequality simplifies to:
Z u
3 ′ −1/2 ∞
k∇/ trχ kL2 (Su,u ) ≤ Cδ R0 (α) / 3 trχ′ kL2 (Su′ ,u ) du′
k∇
0
+C|u|−4 [R∞
0 (α)R /41 (α) + R∞
/2 (β) + (R /41 (β) + R∞
0 (α))(R 0 (β))]
1/2 −5
+O(δ |u| ) (7.159)
This is a linear integral inequality of the form 6.15, but with b a positive con-
stant. Noting that
Cδ 1/2 |u|−1 R∞
0 (α) ≤ 1
provided that δ is suitably small depending on R∞

0 (α), the integral inequality
7.159 implies:
/ 3 trχ′ kL2 (Su,u ) ≤

k∇
+C|u|−4 [R∞
0 (α)R /41 (α) + R∞
/2 (β) + (R /41 (β) + R∞
0 (α))(R 0 (β))]
1/2 −5
+O(δ |u| ) (7.160)
This is the second conclusion of the proposition. Substituting this in 7.149 and
taking the L2 norm with respect to u on [0, δ) if u ∈ [u0 , c∗ − δ), on [0, c∗ − u) if
u ∈ (c∗ − δ, c∗ ), in the case c∗ ≥ u0 + δ, and on [0, c∗ − u), in the case c∗ < u0 + δ
(see 5.6, 5.7), the third conclusion follows. Finally, substituting the second and
third conclusions in 7.141 and noting that
Z u
k∇/ 3 χ̂′ kL2 (Su′ ,u ) du′ ≤ δ 1/2 k∇
/ 3 χ̂′ kL2 (Cu ) ,
0
yields the first conclusion of the proposition.

/ 3 χ′
We now consider any (u1 , u1 ) ∈ D′ and fix attention in the following lemmas to
the parameter subdomain D1 and the corresponding subdomain M1 of M ′ (see
3.45, 3.46).
With a positive constant k to be appropriately chosen in the sequel, let s∗
be the least upper bound of the set of values of s ∈ [u0 , u1 ] such that:
/ 3 log ΩkL2 (Cuu1 ) ≤ kδ 3/2 |u|−4

k∇ : for all u ∈ [u0 , s] (7.161)
241
Then by continuity s∗ > u0 (recall that by 3.44 ∇
/ 3 log Ω vanishes along Cu0 )
and we have:
/ 3 log ΩkL2 (Cuu1 ) ≤ kδ 3/2 |u|−4
k∇ : for all u ∈ [u0 , s∗ ] (7.162)
In the estimates to follow the dependence on the constant k is made explicit.

Let us denote:
/3 (trχ) = |u0 |5 δ −3/2 k∇
D / 3 trχkL2 (Cu0 ) (7.163)
Here, as in 3.49, 3.122, 4.97, 4.98, and 6.53, we are considering all of Cu0 , not
only the part which lies in M ′ . By the results of Chapter 2 this quantity is
bounded by a non-negative non-decreasing continous function of M6 . Let us
∗
denote by D1s the subset of D1 where u ≤ s∗ :
∗
D1s = [0, u1 ] × [u0 , s∗ ] (7.164)
∗
Lemma 7.5 We have:

/ 3 trχ′ kL2 (Cuu1 ) ≤ Cδ 3/2 |u|−5 D

k∇ /3 (trχ)
+(D /41 (χ̂) + D0∞ (χ̂))(R
/41 (trχ) + D /41 (β) + R∞ ∞
0 (β)) + D0 (χ̂)R
/2 (β) + k
+O(δ 2 |u|−6 )
/ 3 χ̂′ kL2 (C u1 ) ≤ Cδ|u|−4 (R

k∇ /41 (β) + R∞
/2 (β) + R 0 (β))
u
+Cδ 3/2 |u|−5 k + O(δ 3/2 |u|−5 )

for all u ∈ [u0 , s∗ ], provided that δ is suitably small depending on D0∞ , R∞ /41 ,
0 , D
4
R/1 , and R
/2 .
Proof: We apply Lemma 7.3 with χ̂′ in the role of θ to the Codazzi equation
6.58. In view of Lemma 5.4 and the last conclusion of Proposition 7.2 the
conclusion of Lemma 7.3 simplifies and we obtain:
n
k∇ / 3 χ̂′ kL2 (Su,u ) ≤ C k∇/ 3 trχ′ kL2 (Su,u ) + k∇ / 2 ikL2 (Su,u ) (7.165)
o
+|u|−1/2 (k∇ / 2 trχ′ kL4 (Su,u ) + k∇/ikL4 (Su,u ) )
n o
+C|u|−3/2 k∇ /χ̂′ kL4 (Su,u ) + |u|−1 kχ̂′ kL4 (Su,u )
Now, by the estimate 6.91 (and Lemma 6.4),

/χ̂′ kL4 (Su,u ) + |u|−1 kχ̂′ kL4 (Su,u ) ≤ Cδ 1/2 |u|−5/2 (R
k∇ /41 (β) + R∞
0 (β)) + O(δ|u|
−7/2
)
(7.166)
Also, by the estimates 6.61, 6.90 (and Lemma 6.4),
/ikL4 (Su,u ) ) ≤ Cδ 1/2 |u|−4 R
/ 2 trχ′ kL4 (Su,u ) + k∇
|u|−1/2 (k∇ /41 (β) + O(δ|u|−5 )
(7.167)
242
/ 2 trχ, ∇
Moreover, using the estimates for ∇ / 2 χ̂ of Proposition 6.2 as well as the
results of Chapters 3 and 4 we deduce:
/ 2 ikL2 (Su,u )
k∇ ≤ / 2 βkL2 (Su,u ) + C|u|−1 k∇
Ck∇ / 2 ηkL2 (Su,u )
+O(δ|u|−5 ) (7.168)

n
k∇/ 3 χ̂′ kL2 (Su,u ) ≤ C k∇/ 3 trχ′ kL2 (Su,u )
o
/ 2 βkL2 (Su,u ) + |u|−1 k∇
+k∇ / 2 ηkL2 (Su,u )
+Cδ 1/2 |u|−4 (R
/41 (β) + R∞
0 (β)) + O(δ|u|
−5
) (7.169)
/ 3 trχ′ :
D∇ / 3 χ̂′ ) + e′
/ 3 trχ′ = −2Ω(χ̂, ∇ (7.170)
where:
e′ABC = /)D
−(DΓ / 2 trχ′ )DC − (DΓ
AB (∇ /)D / 2 trχ′ )BD
AC (∇
−d
/A (Ωtrχ)(∇ /A (Ωχ̂DE )∇
/ 2 trχ′ )BC − 2∇ /C χ̂′DE
/B ∇
+∇
/A eBC (7.171)
and we denote:
/ 3 χ̂′ )ABC = χ̂DE ∇
(χ̂, ∇ /A ∇ /C χ̂′DE
/B ∇ (7.172)
To estimate in L2 (S) the first four terms on the right in 7.171 we place each of
the two factors of each term in L4 (S) using the estimates of Propositions 4.2
and 6.2 pertaining to χ. This gives a bound in L2 (S) for the sum of these terms
by:
Cδ|u|−6 (D
/41 (trχ) + D
/41 (χ̂))(R
/41 (β) + R∞
0 (β)) + O(δ
3/2
|u|−7 ) (7.173)
The contribution to k∇/ekL2 (S) of the first three terms in e as given by 6.65 is
/ 2 rkL2 (S) .
similarly bounded. Finally, the contribution of the last term in e is k∇
Now r is given by 6.56. Using the estimate of Lemma 6.4 and the estimates of
Proposition 6.2 pertaining to χ as well as the results of Chapters 3 and 4 we
deduce:
/ 2 rkL2 (Su,u ) ≤ C|u|−2 k∇

k∇ / 3 log ΩkL2 (Su,u ) + O(δ 2 |u|−8 ) (7.174)
Combining the above results we obtain:
ke′ kL2 (Su,u ) ≤ C|u|−2 k∇

+Cδ|u|−6 (D
/41 (trχ) + D
/41 (χ̂))(R
/41 (β) + R∞
0 (β))
+O(δ 3/2 |u|−7 ) (7.175)
243
We also estimate in L2 (S) the first term on the right in 7.170:
/ 3 χ̂′ )kL2 (Su,u ) ≤

kΩ(χ̂, ∇
n
Cδ 1/2 |u|−2 (D0∞ (χ̂) + O(δ|u|−3/2 )) k∇ / 3 trχ′ kL2 (Su,u )
o
+k∇/ 2 βkL2 (Su,u ) + |u|−1 k∇
/ 2 ηkL2 (Su,u )
/41 (β) + R∞
+Cδ|u|−6 D0∞ (χ̂)(R 0 (β)) + O(δ
3/2
|u|−7 ) (7.176)

We obtain:
/ 3 trχ′ kL2 (Su,u )

|u|4 k∇ ≤ C|u0 |4 k∇
/ 3 trχ′ kL2 (Su,u0 ) (7.177)
Z u
+C / 3 χ̂′ ) + e′ kL2 (Su,u′ ) du′
|u′ |4 k − 2Ω(χ̂, ∇
u0
Substituting the bounds 7.175, 7.176 yields a linear integral inequality, for fixed
u, for the quantity:
x(u, u) = |u|4 k∇
/ 3 trχ′ kL2 (Su,u ) (7.178)
of the form 6.81, that is:
Z u
x(u, u) ≤ a(u′ )x(u, u′ )du′ + b(u, u)
u0
with a the non-negative function:
a(u) = Cδ 1/2 |u|−2 (D0∞ (χ) + O(δ|u|−3/2 ) (7.179)
and b(u) the non-negative non-decreasing function, depending on u,
b(u, u) = C|u0 |4 k∇
/ 3 trχkL2 (Su,u0 )
+Cδ|u|−1 (D /41 (trχ) + D /41 (χ̂) + D0∞ (χ̂))(R/41 (β) + R∞ 0 (β))
Z u
+Cδ 1/2 |u′ |2 (D0∞ (χ̂) + O(δ|u′ |−3/2 )) ·
u0
n o
· k∇ / 2 βkL2 (Su,u′ ) + |u′ |−1 k∇/ 2 ηkL2 (Su,u′ ) du′
Z u
+C |u′ |2 k∇/ 3 log ΩkL2 (Su,u′ ) du′ + O(δ 3/2 |u|−2 ) (7.180)
u0
Since Z u
a(u′ )du′ ≤ log 2 (7.181)
u0

0 , we deduce, following
the same argument as that leading from 6.81 to 6.89,
x(u, u) ≤ 2b(u, u) (7.182)
244
We take the L2 norm with respect to u on [0, u1 ] of this inequality to obtain:
kx(·, u)kL2 ([0,u1 ]) ≤ 2kb(·, u)kL2 ([0,u1 ]) (7.183)
Since for any S tensorfield θ we have:
Z u 1/2
1
2
kkθkL2(S·,u ) kL2 ([0,u1 ]) = kθkL2 (Su,u ) du = kθkL2 (Cuu1 ) (7.184)
0
the inequality 7.183, implies, in view of the definitions 7.178, 7.180,

/ 3 trχ′ kL2 (Cuu1 ) ≤ C|u0 |4 k∇
|u|4 k∇ / 3 trχkL2 (Cuu1 )
0
+Cδ 3/2 |u|−1 (D /41 (trχ) + D/41 (χ̂) + D0∞ (χ̂))(R/41 (β) + R∞ 0 (β))
Z u
+Cδ 1/2 |u′ |2 (D0∞ (χ̂) + O(δ|u′ |−3/2 )) ·
u0
n o
· k∇ / 2 ηkL2 (C u1 ) du′
/ 2 βkL2 (C u1 ) + |u′ |−1 k∇
u′ u′
Z u
+C |u′ |2 k∇/ 3 log ΩkL2 (C u1 ) du′ + O(δ 2 |u|−2 ) (7.185)
u0 u′
By the second conclusion of Proposition 7.1:

k∇ / 2 ηkL2 (Cu ) ≤ Cδ|u|−3 R
/ 2 ηkL2 (C u1 ) ≤ k∇ /2 (β) + O(δ 3/2 |u|−4 ) (7.186)
u
By the fourth of the definitions 7.1:

/ 2 βkL2 (C u1 ) ≤ k∇
k∇ / 2 βkL2 (Cu ) ≤ δ 3/2 |u|−5 R
/2 (β) (7.187)
u
Moreover, by the definition 7.163:

k∇ / 3 trχkL2 (Cu0 ) ≤ δ 3/2 |u0 |−5 D
/ 3 trχkL2 (Cuu1 ) ≤ k∇ /3 (trχ) (7.188)
0
Substituting 7.186 - 7.188 in 7.185 and taking into account 7.162 we conclude
that:
/ 3 trχ′ kL2 (Cuu1 ) ≤ Cδ 3/2 |u0 |−1 D
|u|4 k∇ /3 (trχ)
3/2 −1
4
+Cδ |u| (D
/1 (trχ) + D/1 (χ̂) + D0∞ (χ̂))(R
4
/41 (β) + R∞
0 (β))
∞

+D0 (χ̂)R /2 (β) + k
+O(δ 2 |u|−2 ) (7.189)
for all u ∈ [u0 , s∗ ]. This yields the first conclusion of the lemma.
We take the L2 norm of the inequality 7.169 with respect to u on [0, u1 ] to
obtain:
n
/ 3 χ̂′ kL2 (Cuu1 ) ≤ C k∇
k∇ / 3 trχ′ kL2 (Cuu1 )
o
+k∇/2 βkL2 (Cuu1 ) + |u|−1 k∇
/ 2 ηkL2 (Cuu1 )
+Cδ|u|−4 (R
/41 (β) + R∞
0 (β)) + O(δ
3/2
|u|−5 ) (7.190)
245
Substituting 7.186, 7.187 and the estimate 7.189 then yields the second conclu-
sion of the lemma.

/ 3 η, ∇
/ 3η
In the following we denote by O(δ p |u|r ), for real numbers p, r, the product of
δ p |u|r with a non-negative non-decreasing continuous function of the quantities
D0∞ , R∞ /41 , R
0 , D /41 , D
/42 (trχ), R
/2 , and D
/3 (trχ).
Lemma 7.6 We have:
/ 3 ηkL2 (Cuu1 ) ≤ Cδ 1/2 |u|−4 M̃ + O(δ|u|−4 ) + O(δ 3 |u|−7 )k

k∇
/ 3 ηkL2 (C u1 ) ≤ Cδ 1/2 |u|−4 M̃ + O(δ|u|−4 ) + O(δ 2 |u|−6 )k
k∇
u
for all u ∈ [u0 , s∗ ], provided that δ is suitably small depending on D0∞ , R∞ /41 ,
0 , D
4 ∞ ∞
R/1 , and R
/2 . Moreover, the coefficients of k depend only on D0 , R0 . Here,
M̃ = R
/2 (ρ) + R
/2 (σ) + R/41 (ρ) + R
/41 (σ)
∞ 4 ∞
+(R0 (α) + R /1 (β) + R0 (β))(R /41 (β) + R∞
0 (β))
+(R/41 (α) + R∞ /41 (α)
0 (α))R
M̃ = M̃ + Ã
and Ã is given by 6.190.

Proof: We apply Lemma 7.4 with η and η in the role of θ to the Hodge systems
6.96 and 6.97 respectively. The assumptions of Lemma 7.4 are satisfied by
virtue of the first estimate of Proposition 6.1 provided that δ is suitably small
depending on D0∞ , R∞ /41 , R
0 , D /41 and R /2 (α). Moreover, in view of Lemma 5.4
and the last conclusion of Proposition 7.2 the conclusion of Lemma 7.4 simplifies
and we obtain:
n
k∇/ 3 ηkL2 (Su,u ) ≤ C k∇ / 2 ρkL2 (Su,u ) + k∇ / 2 σkL2 (Su,u ) + k∇/ 2 µkL2 (Su,u )
o
+k∇ / 2 (χ̂, χ̂)kL2 (Su,u ) + k∇/ 2 (χ̂ ∧ χ̂)kL2 (Su,u )
n
+C|u|−1/2 kd /ρkL4(Su,u ) + kd /σkL4 (Su,u ) + kd/µkL4(Su,u )
o
+kd /(χ̂, χ̂)kL4 (Su,u ) + kd /(χ̂ ∧ χ̂)kL4 (Su,u )
n o
+C|u|−3/2 k∇ /ηkL4 (Su,u ) + |u|−1 kηkL4 (Su,u ) (7.191)
and:
n
/ 3 ηkL2 (Su,u ) ≤ C k∇
k∇ / 2 ρkL2 (Su,u ) + k∇ / 2 σkL2 (Su,u ) + k∇
/ 2 µkL2 (Su,u )
o
+k∇/ 2 (χ̂, χ̂)kL2 (Su,u ) + k∇
/ 2 (χ̂ ∧ χ̂)kL2 (Su,u )
246
n
+C|u|−1/2 kd /ρkL4(Su,u ) + kd /σkL4 (Su,u ) + kd/µkL4 (Su,u )
o
+kd/(χ̂, χ̂)kL4 (Su,u ) + kd
/(χ̂ ∧ χ̂)kL4 (Su,u )
n o
+C|u|−3/2 k∇ /ηkL4 (Su,u ) + |u|−1 kηkL4 (Su,u ) (7.192)
Now, by the results of Chapters 3 and 4,
/ηkL4 (Su,u ) + |u|−1 kηkL4 (Su,u ) ≤ Cδ 1/2 |u|−5/2 (R

k∇ /41 (β) + R∞
0 (β)) + O(δ|u|
−7/2
)
(7.193)
and:
/ηkL4 (Su,u ) + |u|−1 kηkL4 (Su,u ) ≤ Cδ 1/2 |u|−5/2 (R

k∇ /41 (β) + R∞
0 (β)) + O(δ|u|
−7/2
)
(7.194)
Also, recalling the definitions 6.164, 6.165, by the estimates 6.191, 6.192 and
Lemma 6.4 we have:
/µkL4 (Su,u ) ≤ C|u|−7/2 R∞

kd /41 (α) + O(δ 1/2 |u|−9/2 )
0 (α)R (7.195)
and:
/µkL4 (Su,u ) ≤ C|u|−7/2 Ã + O(δ 1/2 |u|−7/2 )
kd (7.196)
where Ã is given by 6.190. Moreover, using the estimates of Propositions 6.1

and 6.2 pertaining to χ̂ and χ̂ we deduce:
/ 2 (χ̂, χ̂)kL2 (Su,u ) + k∇

k∇ / 2 (χ̂ ∧ χ̂)kL2 (Su,u )
≤ C|u|−4 (R
/41 (β) + R∞ 2
0 (β)) + O(δ
1/2
|u|−5 ) (7.197)
and (using also Lemma 5.2):
kd
/(χ̂, χ̂)kL4 (Su,u ) + kd
/(χ̂ ∧ χ̂)kL4 (Su,u )
≤ C|u|−7/2 (R
/41 (β) + R∞ 2
0 (β)) + O(δ
1/2
|u|−5 ) (7.198)
Substituting the above in 7.191, 7.192 we obtain:

n o
/ 3 ηkL2 (Su,u ) ≤ C k∇
k∇ / 2 µkL2 (Su,u ) + k∇
/ 2 ρkL2 (Su,u ) + k∇
/ 2 σkL2 (Su,u )
+C|u|−4 M + O(δ 1/2 |u|−4 ) (7.199)
and:
n o
/ 3 ηkL2 (Su,u ) ≤ C k∇
k∇ / 2 µkL2 (Su,u ) + k∇
/ 2 ρkL2 (Su,u ) + k∇
/ 2 σkL2 (Su,u )
+C|u|−4 M + O(δ 1/2 |u|−4 ) (7.200)
Here:
/41 (ρ) + R
M =R /41 (σ) + (R
/41 (β) + R∞ 2 ∞
/41 (α)
0 (β)) + R0 (α)R (7.201)
247
and:
/41 (ρ) + R
M =R /41 (σ) + (R
/41 (β) + R∞ 2
0 (β)) + Ã (7.202)
Next, we apply Lemma 4.1 to the propagation equations 6.121, 6.122 to
/ 2 µ, ∇
obtain the following propagation equations for ∇ / 2 µ:
1
D∇/ 2 µ = −Ωtrχ∇
/ 2 µ − Ωtrχ∇ / 2 µ + h′ (7.203)
2
1
D∇/ 2 µ = −Ωtrχ∇
/ 2 µ − Ωtrχ∇ / 2 µ + h′ (7.204)
2
where h′ ,h′ are the symmetric 2-covariant S tensorfields:

′ 1
h = −DΓ / · /dµ − µ + µ ∇ / 2 (Ωtrχ)
2

1 1
−d/(Ωtrχ) ⊗ /dµ + /dµ − /dµ + /dµ ⊗ /d(Ωtrχ)
2 2

1 1
+∇/ 2 Ω − trχ|χ̂|2 + trχ|η|2 +∇ / 2 div
/ j (7.205)
4 2

′ 1
h = −DΓ/ · /dµ − µ + µ ∇ / 2 (Ωtrχ)
2

1 1
/(Ωtrχ) ⊗ /dµ + /dµ − /dµ + /dµ ⊗ /d(Ωtrχ)
−d
2 2

1 1
/ 2 Ω − trχ|χ̂|2 + trχ|η|2
+∇ +∇/ 2 div
/ j (7.206)
4 2
We first estimate in L2 (S) the first four terms on the right in 7.205 and
7.206. We have:
/)C
(DΓ /A (Ωχ)BC + ∇
AB = ∇ /B (Ωχ)AC − ∇
/C (Ωχ)AB
By Proposition 6.1 and Lemma 5.2 with p = 4 (and the estimate of Proposition
4.1 for trχ):
/(Ωχ)kL∞ (Su,u ) ≤ Cδ −1/2 |u|−2 (R
k∇ /41 (β) + R∞
0 (β)) + O(|u|
−3
) (7.207)
hence also:
/kL∞ (Su,u ) ≤ Cδ −1/2 |u|−2 (R
kDΓ /41 (β) + R∞
0 (β)) + O(|u|
−3
) (7.208)
On the other hand, from 6.225 (and Lemma 6.4):

/kL∞ (Su,u ) ≤ Cδ 1/2 |u|−3 (R
kDΓ /41 (β) + R∞
0 (β)) + O(δ|u|
−4
) (7.209)
We can then estimate the first term on the right in each of 7.205, 7.206 by
placing the first factor in L∞ (S) using 7.208, 7.209 and the second factor in
L4 (S) using 7.195, 7.196. This gives:
/ · /dµkL2 (Su,u ) ≤ O(δ −1/2 |u|−5 ),
kDΓ / · /dµkL2 (Su,u ) ≤ O(δ 1/2 |u|−6 ) (7.210)
kDΓ
248
By 6.136, 6.137 and the estimates of Proposition 6.2 pertaining to η, η:
kµkL∞ (Su,u ) , kµkL∞ (Su,u ) ≤ C|u|−3 (R∞ ∞ ∞

0 (ρ)+ R0 (α)D0 (χ)+ Ã)+ O(δ
1/2
|u|−3 )
(7.211)
We can then estimate the second term on the right in each of 7.205, 7.206 by
placing the first factor in L∞ and the second factor in L4 (S) using Propositions
6.1 and 6.2. We obtain in this way bounds in L2 (S) for the second term on the
right in each of 7.205, 7.206 by O(|u|−6 ) and O(δ|u|−7 ) respectively. The third
and fourth terms on the right in each of 7.205, 7.206 are estimated in a similar
manner as the first terms. We obtain in this way bounds in L2 (S) for the third
and fourth terms on the right in each of 7.205, 7.206 by O(|u|−6 ) and O(δ|u|−7 )
respectively.
We proceed to the fifth terms on the right in each of 7.205, 7.206. Using
Proposition 6.1, the estimates of Proposition 6.2 for trχ and η, as well as the
results of Chapters 3 and 4, we can estimate the fifth term on the right in 7.205
in L2 (S) by:
Cδ −1 |u|−4 R∞ /41 (β) + R∞ /41 (α))2 + O(δ −1/2 |u|−5 ) (7.212)

0 (α)(R 0 (β)) + (R
Also, using the estimate of Proposition 6.1 for trχ, the estimates of Proposition
6.2 for trχ, χ̂ and η, as well as the results of Chapters 3 and 4, we can estimate
the fifth term on the right in 7.206 in L2 (S) by:
O(δ 1/2 |u|−6 ) (7.213)
We turn to the last terms on the right in 7.205 and 7.206 the symmetric 2-
/ 2 div
covariant S tensorfields ∇ / 2 div
/ j and ∇ / j. These are of the form (see 6.125,
6.129):
/ 2 div
∇ / j = / 3 η) + (Ωtrχ, ∇
(Ωχ̂, ∇ / 3 η)
+(∇ / 2 η) + (d
/(Ωχ̂), ∇ / 2 η)
/(Ωtrχ), ∇
/ 2 (Ωχ̂), ∇
+(∇ / 2 (Ωtrχ), ∇
/η) + (∇ /η)
/ 3 (Ωχ̂), η) + (∇
+(∇ / 3 (Ωtrχ), η) (7.214)
and:
/ 2 div
∇ / j = / 3 η) + (Ωtrχ, ∇
(Ωχ̂, ∇ / 3 η)
+(∇ / 2 η) + (d
/(Ωχ̂), ∇ / 2 η)
/(Ωtrχ), ∇
/ 2 (Ωχ̂), ∇
+(∇ / 2 (Ωtrχ), ∇
/η) + (∇ /η)
/ 3 (Ωχ̂), η) + (∇
+(∇ / 3 (Ωtrχ), η) (7.215)
The first pair of terms on the right in 7.214, placing the first factors in L∞ (S)
and the second factors in L2 (S), are bounded in L2 (S) by:
Cδ −1/2 |u|−1 R∞ / 3 ηkL2 (Su,u ) + C(|u|−1 + O(|u|−2 ))k∇

0 (α)k∇ / 3 ηkL2 (Su,u ) (7.216)
249
To estimate the second pair of terms on the right in 7.214 in L2 (S), we place
the first factors in L∞ (S) using 7.207 and the second factors in L2 (S) to obtain
a bound by:
Cδ −1/2 |u|−2 (R
/41 (β) + R∞0 (β) + O(δ
1/2
|u|−1 )) ·
/ 2 ηkL2 (Su,u ) + k∇
·(k∇ / 2 ηkL2 (Su,u ) )
≤ Cδ −1/2 |u|−2 (R
/41 (β) + R∞
0 (β) + O(δ
1/2
|u|−1 ))k∇
/ 2 ηkL2 (Su,u )
+O(|u|−5 ) (7.217)
taking into account the first estimate of Proposition 7.1. To estimate the third
pair of terms on the right in 7.214 in L2 (S), we place the first factors in L4 (S) us-
ing Proposition 6.1 and the second factors in L4 (S) using the results of Chapter
4 to obtain a bound by:
O(|u|−5 ) (7.218)
Finally, the last pair of terms on the right in 7.214, placing the first factors
in L2 (S) and the second factors in L∞ (S) using the results of Chapter 3, are
bounded in L2 (S) by:
Cδ 1/2 |u|−2 (R∞

0 (β) + O(δ
1/2
|u|−1 ))(k∇
/ 3 (Ωtrχ)kL2 (Su,u ) + k∇
/ 3 (Ωχ̂)kL2 (Su,u ) )
(7.219)
Moreover, expressing ∇ / 3 log Ω as in 7.139 and using the estimates of Proposition
6.1 and Lemma 6.4 and the fact that:
/ log ΩkL∞ (Su,u ) ≤ O(δ|u|−3 )

kd (7.220)
which follows from Lemmas 4.11 and 6.4 through Lemma 5.2, we can estimate:
/ 3 (Ωtrχ)kL2 (Su,u ) + k∇
k∇ / 3 (Ωχ̂)kL2 (Su,u )
n
≤ C k∇ / 3 trχ′ kL2 (Su,u ) + k∇
/ 3 χ̂′ kL2 (Su,u )
o
/ 3 log ΩkL2 (Su,u ) (ktrχ′ kL∞ (Su,u ) + kχ̂′ kL∞ (Su,u ) )
+k∇
+ O(δ 1/2 |u|−5 )
n o
≤ C k∇ / 3 χ̂′ kL2 (Su,u )
/ 3 trχ′ kL2 (Su,u ) + k∇
+ Cδ −1/2 |u|−1 (R∞
0 (α) + O(δ
1/2
/ 2 ηkL2 (Su,u ) + k∇
))(k∇ / 2 ηkL2 (Su,u ) )
+ O(δ 1/2 |u|−5 )
/ 3 χ̂′ kL2 (Su,u ) + Cδ −1/2 |u|−1 (R∞
≤ Ck∇ 0 (α) + O(δ
1/2
/ 2 ηkL2 (Su,u )
))k∇
+ O(|u|−4 ) (7.221)
the last step by virtue of the second estimate of Proposition 7.2 and the first
estimate of Proposition 7.1. Combining the above results 7.216 - 7.218 and
7.219, 7.221 we conclude that:
/ 2 div
k∇ / jkL2 (Su,u ) ≤ Cδ −1/2 |u|−1 R∞ / 3 ηkL2 (Su,u )
0 (α)k∇
250
+C|u|−1 (1 + O(|u|−1 ))k∇
/ 3 ηkL2 (Su,u )
+O(δ −1/2 |u|−2 )k∇
/ 2 ηkL2 (Su,u ) + O(δ 1/2 |u|−2 )k∇
/ 3 χ̂′ kL2 (Su,u )
+O(|u|−5 ) (7.222)
The first pair of terms on the right in 7.215, placing the first factors in L∞ (S)
and the second factors in L2 (S), are bounded in L2 (S) by:
Cδ 1/2 |u|−2 (D0∞ (χ̂) + O(δ|u|−3/2 ))k∇

/ 3 ηkL2 (Su,u )
+C|u|−1 (1 + O(δ|u|−1 ))k∇
/ 3 ηkL2 (Su,u ) (7.223)
To estimate the second pair of terms on the right in 7.215 in L2 (S), we place
the first factors in L∞ (S) using 6.224 (and Lemma 6.4) and the second factors
in L2 (S) to obtain a bound by:
Cδ 1/2 |u|−3 (R
/41 (β) + R∞
0 (β) + O(δ
1/2
|u|−1 )) ·
/ 2 ηkL2 (Su,u ) + k∇
·(k∇ / 2 ηkL2 (Su,u ) )
≤ Cδ 1/2 |u|−3 (R
/41 (β) + R∞
0 (β) + O(δ
1/2
|u|−1 ))k∇
/ 2 ηkL2 (Su,u )
+O(δ|u|−6 ) (7.224)
taking into account the first estimate of Proposition 7.1. To estimate the third
pair of terms on the right in 7.215 in L2 (S), we place the first factors in L4 (S) us-
ing Proposition 6.2 and the second factors in L4 (S) using the results of Chapter
4 to obtain a bound by:
O(δ|u|−6 ) (7.225)
Finally, the last pair of terms on the right in 7.215, placing the first factors
in L2 (S) and the second factors in L∞ (S) using the results of Chapter 3, are
Cδ 1/2 |u|−2 |(R∞

0 (β) + O(δ
1/2
|u|−1 ))(k∇
/ 3 (Ωtrχ)kL2 (Su,u ) + k∇
/ 3 (Ωχ̂)kL2 (Su,u ) )
(7.226)
Moreover, taking into account the estimates of Proposition 6.2, Lemma 6.4 and
7.220, we can estimate:
/ 3 (Ωtrχ)kL2 (Su,u ) + k∇
k∇ / 3 (Ωχ̂)kL2 (Su,u )
n
≤ C k∇ / 3 χ̂′ kL2 (Su,u )
/ 3 trχ′ kL2 (Su,u ) + k∇
o
/ 3 log ΩkL2 (Su,u ) (ktrχ′ kL∞ (Su,u ) + kχ̂′ kL∞ (Su,u ) )
+k∇
+ O(δ 3/2 |u|−6 )
n o
≤ C k∇ / 3 χ̂′ kL2 (Su,u )
/ 3 trχ′ kL2 (Su,u ) + k∇
+ C(|u|−1 + O(δ 1/2 |u|−2 ))k∇
+ O(δ 3/2 |u|−6 ) (7.227)
251
/ 3 log ΩkL2 (Su,u ) depends only on D0∞ , R∞
where the coefficient of k∇ 0 . Combin-
ing the above results 7.223 - 7.227 we conclude that:
k∇ / jkL2 (Su,u ) ≤ Cδ 1/2 |u|−2 (D0∞ (χ̂) + O(δ|u|−3/2 ))k∇

/ 2 div / 3 ηkL2 (Su,u )
+C|u|−1 (1 + O(δ|u|−1 ))k∇
/ 3 ηkL2 (Su,u )
+O(δ 1/2 |u|−3 )k∇
/ 2 ηkL2 (Su,u )
/ 3 χ̂′ kL2 (Su,u ) )
/ 3 trχ′ kL2 (Su,u ) + k∇
+O(δ 1/2 |u|−2 )(k∇
+O(δ 1/2 |u|−3 )k∇
/ 3 log ΩkL2 (Su,u ) + O(δ|u|−6 ) (7.228)
/ 3 log ΩkL2 (Su,u ) depends only on D0∞ , R∞

where the coefficient of k∇ 0 .
Combining the bound 7.130 with the previously obtained bounds on the first
five terms on the right in 7.205 we obtain:
kh′ kL2 (Su,u ) ≤ Cδ −1/2 |u|−1 R∞ / 3 ηkL2 (Su,u )

0 (α)k∇
+C|u|−1 (1 + O(|u|−1 ))k∇

/ 3 ηkL2 (Su,u )
+O(δ −1/2 |u|−2 )k∇
/ 2 ηkL2 (Su,u )
+O(δ 1/2 |u|−2 )k∇
/ 3 χ̂′ kL2 (Su,u )
+Cδ −1 |u|−4 R∞ /41 (β) + R∞ /41 (α))2

0 (α)(R 0 (β)) + (R
+O(δ −1/2 |u|−5 ) (7.229)
Combining the bound 7.228 with the previously obtained bounds on the first
five terms on the right in 7.206 we obtain:
kh′ kL2 (Su,u ) ≤ Cδ 1/2 |u|−2 (D0∞ (χ̂) + O(δ|u|−3/2 ))k∇

/ 3 ηkL2 (Su,u )
+C|u|−1 (1 + O(δ|u|−1 ))k∇
/ 3 ηkL2 (Su,u )
+O(δ 1/2 |u|−3 )k∇
/ 2 ηkL2 (Su,u )
+O(δ 1/2 |u|−2 )(k∇ / 3 χ̂′ kL2 (Su,u ) )
/ 3 trχ′ kL2 (Su,u ) + k∇
+O(δ 1/2 |u|−3 )k∇
/ 3 log ΩkL2 (Su,u ) + O(δ 1/2 |u|−6 ) (7.230)
where the coefficient of k∇/ 3 log ΩkL2 (Su,u ) depends only on D0∞ , R∞
0 . Substi-
tuting in 7.229, 7.230 the estimates 7.209, 7.210 yields:
kh′ kL2 (Su,u ) ≤ Cδ −1/2 |u|−1 R∞ / 2 µkL2 (Su,u )

0 (α)k∇
+C|u|−1 (1 + O(|u|−1 ))k∇

/ 2 µkL2 (Su,u )
+Cδ −1/2 |u|−1 (R∞
0 (α) + O(δ
1/2
/ 2 ρkL2 (Su,u ) + k∇
))(k∇ / 2 σkL2 (Su,u ) )
+O(δ −1/2 |u|−2 )k∇/ 2 ηkL2 (Su,u ) + O(δ 1/2 |u|−2 )k∇/ 3 χ̂′ kL2 (Su,u )
+Cδ −1 |u|−4 R∞ /41 (β) + R∞ /41 (α))2

0 (α)(R 0 (β)) + (R
+O(δ −1/2 |u|−5 ) (7.231)
252
and:
kh′ kL2 (Su,u ) ≤ Cδ 1/2 |u|−2 (D0∞ (χ̂) + O(δ|u|−3/2 ))k∇
/ 2 µkL2 (Su,u )
+C|u|−1 (1 + O(δ|u|−1 ))k∇
/ 2 µkL2 (Su,u )
+C|u|−1 (1 + O(δ 1/2 |u|−1 ))(k∇ / 2 σkL2 (Su,u ) )
/ 2 ρkL2 (Su,u ) + k∇
+O(δ 1/2 |u|−3 )k∇
/ 2 ηkL2 (Su,u )
/ 3 χ̂′ kL2 (Su,u ) )
/ 3 trχ′ kL2 (Su,u ) + k∇
+O(δ 1/2 |u|−2 (k∇
+O(δ 1/2 |u|−3 )k∇
+C|u|−5 M + O(δ 1/2 |u|−5 ) (7.232)
where the coefficient of k∇/ 3 log ΩkL2 (Su,u ) depends only on D0∞ , R∞
0 .
We now turn to the propagation equations 7.203, 7.204. To 7.203 we apply
Lemma 4.6 with p = 2. Here r = 2, ν = −2, γ = 0, and we obtain:
Z u
1
/ 2 µkL2 (Su,u ) ≤ C
k∇ − Ωtrχ∇ / 2 µ + h′ du′ (7.233)
0 2 2
L (S ′ ) u ,u
To 7.204 we apply Lemma 4.7 with p = 2. Here again r = 2, ν = −2, γ = 0,

and we obtain:
|u|3 k∇
/ 2 µkL2 (Su,u ) ≤ C|u0 |3 k∇
/ 2 µkL2 (Su,u0 ) (7.234)
Z u
1
+C |u′ |3 − Ωtrχ∇ / 2 µ + h′ du′
u0 2 L2 (Su,u′ )
Substituting in 7.233 and 7.234 the estimates 7.231 and 7.232 yields the fol-
lowing system of linear integral inequalities for the quantities k∇ / 2 µkL2 (Su,u ) ,
∗
k∇/ 2 µkL2 (Su,u ) on the domain D1s :
Z u
2
k∇
/ µkL2 (Su,u ) ≤ a(u) k∇/ 2 µkL2 (Su′ ,u ) du′
0
Z u
+b(u) k∇/ 2 µkL2 (Su′ ,u ) du′
0
+f (u) (7.235)
Z u
|u|3 k∇
/ 2 µkL2 (Su,u ) ≤ |u′ |3 a(u′ )k∇
/ 2 µkL2 (Su,u′ ) du′
u0
Z u
+ |u′ |3 b(u′ )k∇
/ 2 µkL2 (Su,u′ ) du′
u0
3
+|u| f (u, u) (7.236)
In 7.235,
a(u) = Cδ −1/2 |u|−1 R∞
0 (α) (7.237)
253
b(u) = C|u|−1 (1 + O(|u|−1 )) (7.238)
C|u|−4 R∞ /41 (β) + R∞ /41 (α))2

f (u) = 0 (α)(R 0 (β)) + (R
+O(δ 1/2 |u|−5 ) (7.239)
and we have taken into account the following three facts. First, that:
Z u
/ 2 ρkL2 (Su′ ,u ) + k∇
(k∇ / 2 σkL2 (Su′ ,u ) )du′
0
≤ δ 1/2 (k∇
/ 2 ρkL2 (Cu ) + k∇
/ 2 σkL2 (Cu ) )
≤ δ|u|−4 (R
/2 (ρ) + R
/2 (σ))
by the 2nd and 3rd of the definitions 7.1. Second, that:

Z u
k∇/ 2 ηkL2 (Su′ ,u ) du′
0
≤ δ 1/2 k∇
/ 2 ηkL2 (Cu )
≤ Cδ 3/2 |u|−3 R
/2 (β) + O(δ 2 |u|−4 )
by the 2nd estimate of Proposition 7.1. Third, that:

Z u
/ 3 χ̂′ kL2 (Su′ ,u ) du′
k∇
0
≤ Cδ 1/2 k∇
/ 3 χ̂′ kL2 (Cu )
≤ Cδ 1/2 |u|−3 (R /41 (β) + R∞
/2 (β) + R 0 (β)) + O(δ|u|
−4
)
by the 3rd estimate of Proposition 7.2.

In 7.236,
a(u) = Cδ 1/2 |u|−2 (D0∞ (χ̂) + O(δ|u|−3/2 )) (7.240)

−1 −1
b(u) = C|u| (1 + O(δ|u| )) (7.241)
Z un
|u|3 f (u, u) = C|u′ |2 (1 + O(δ 1/2 |u′ |−1 ))(k∇
/ 2 ρkL2 (Su,u′ ) + k∇
/ 2 σkL2 (Su,u′ ) )
u0
+O(δ 1/2 )k∇

/ 2 ηkL2 (Su,u′ )
+O(δ 1/2 |u′ |)(k∇ / 3 χ̂′ kL2 (Su,u′ ) )
/ 3 trχ′ kL2 (Su,u′ ) + k∇
o
+O(δ 1/2 )k∇ / 3 log ΩkL2 (Su,u′ ) du′
+C|u|−1 M + O(δ 1/2 |u|−1 ) + C|u0 |3 k∇
/ 2 µkL2 (Su,u0 ) (7.242)
where the coefficient of k∇ / 3 log ΩkL2 (Su,u ) depends only on D0∞ , R∞

0 .
Consider first the integral inequality 7.235. At fixed u, considering k∇/ 2 µkL2 (Su,u )
as given, this inequality is of the form 6.15, with a(u) in the role of the constant
a and Z u
b(u) / 2 µkL2 (Su′ ,u ) du′ + f (u)
k∇
0
254
in the role of the non-negative function b(u). If, as is the case here, the func-
tion b(u) is non-decreasing and 6.158 holds then the result 6.18 implies 6.159.
Condition 6.158 in the present case reads:
a(u)δ = Cδ 1/2 |u|−1 R∞

0 (α) ≤ log 2
and is indeed satisfied provided that δ is suitably small depending on R∞

0 . The
result 6.159 then takes the form:
Z u
/ 2 µkL2 (Su,u ) ≤ 2b(u)
k∇ / 2 µkL2 (Su′ ,u ) du′
k∇
0
+2f (u) (7.243)
Consider next the integral inequality 7.236. At fixed u, considering k∇/ 2 µkL2 (Su,u )
as given, this inequality is of the form 6.81 with a(u) in the role of the non-
negative function a and
Z u
|u′ |3 b(u′ )k∇
/ 2 µkL2 (Su,u′ ) du′ + |u|3 f (u, u)
u0
in the role of the non-negative non-decreasing function b. Note in particular

from 7.242 that the function |u|3 f (u, u) is indeed non-decreasing in u. Recall
that the result 6.87 implies the result 6.89 provided that the condition 6.88
holds. In view of 7.240 this condition holds if
Cδ 1/2 (D0∞ (χ̂) + O(δ)) ≤ log 2
which is indeed satisfied if δ is suitably small depending on D0∞ , R∞ 0 . The result

6.89 then takes the form:
Z u
|u|3 k∇
/ 2 µkL2 (Su,u ) ≤ 2 |u′ |3 b(u′ )k∇
/ 2 µkL2 (Su,u′ ) du′
u0
+2|u|3 f (u, u) (7.244)
∗
We have thus reduced the system of integral inequalities 7.235, 7.236 on D1s to
the system 7.243, 7.244.
Let us now define:
Z u 1/2
1
2 2
y(u) = k∇
/ µkL2 (Su,u ) du / 2 µkL2 (C u1 )
= k∇ (7.245)
u
0
Z u1 1/2
y(u) = / 2 µk2L2 (Su,u ) du
k∇ / 2 µkL2 (Cuu1 )
= k∇ (7.246)
0
Taking the L2 norm of 7.244 with respect to u on [0, u1 ] then yields:

Z u
3
|u| y(u) ≤ 2 |u′ |3 b(u′ )y(u′ )du′ + 2|u|3 g(u) (7.247)
u0
255
where:
g(u) = kf (·, u)kL2 ([0,u1 ]) (7.248)
From 7.242 we have:
Z un
3
|u| g(u) = C|u′ |2 (1 + O(δ 1/2 |u′ |−1 ))·
u0
/ 2 ρkL2 (C u1 ) + k∇
·(k∇ / 2 σkL2 (C u1 ) )
u′ u′
+O(δ 1/2 )k∇

/ 2 ηkL2 (C u1 )
u′
/ 3 χ̂′ kL2 (C u1 ) )
/ 3 trχ′ kL2 (C u1 ) + k∇
+O(δ 1/2 |u′ |)(k∇
u′ u′
o
1/2 3 ′
+O(δ )k∇ / log ΩkL2 (C u1 ) du
u′
1/2 −1 −1
+Cδ |u| M + O(δ|u| ) + C|u0 |−1 δ 1/2 D
/2 (µ) (7.249)
/ 3 log ΩkL2 (Cu1u′ ) in the integral depends only on D0∞ ,

where the coefficient of k∇
∞
R0 . By virtue of the 2nd and 3rd of the definitions 7.1, the 2nd estimate of
Proposition 7.1, Lemma 7.5, and 7.162, we then obtain:
|u|3 g(u) ≤ C|u|−1 δ 1/2 M ′ + O(δ|u|−1 ) + O(δ 2 |u|−3 )k

+C|u0 |−1 δ 1/2 D
/2 (µ) (7.250)
where the coefficient of k depends only on D0∞ , R∞ 0 . In the above we have

defined:
/2 (µ) = δ −1/2 |u0 |4 k∇
D / 2 µkL2 (C u1 ) (7.251)
u0
and:
M′ = M + R
/2 (ρ) + R
/2 (σ) (7.252)
Let us set:
z(u) = / 2 µkL2 (Su,u )
sup k∇ (7.253)
u∈[0,u1 ]
Going back to 7.243 and taking the supremum with respect to u on [0, u1 ] we
obtain: Z u1
z(u) ≤ 2b(u) / 2 µkL2 (Su,u ) du + 2f (u)
k∇ (7.254)
0
Now, Z u1
/ 2 µkL2 (Su,u ) du ≤ δ 1/2 y(u)
k∇ (7.255)
0
therefore substituting the bound 7.247 for y(u) we obtain:

Z u Z u
1
/ 2 µkL2 (Su,u ) du ≤ 2δ 1/2 |u|−3
k∇ |u′ |3 b(u′ )y(u′ )du′ + |u|3 g(u)
0 u0
(7.256)
Since
y(u) ≤ δ 1/2 z(u) (7.257)
256
7.256 implies:
Z u1 Z u
2 1/2 −3 1/2 ′ 3 ′ ′ ′ 3
k∇
/ µkL2 (Su,u ) du ≤ 2δ |u| δ |u | b(u )z(u )du + |u| g(u)
0 u0
(7.258)
Substituting the bound 7.258 in 7.254 then yields the following linear integral
inequality for z(u):
Z u
z(u) ≤ 4δ|u|−3 b(u) |u′ |3 b(u′ )z(u′ )du′
u0
+4δ 1/2 b(u)g(u) + 2f (u) (7.259)
Moreover, setting: Z u
Z(u) = |u′ |3 b(u′ )z(u′ )du′ (7.260)
u0
we have, replacing u by u′ in 7.257 multiplying by |u′ |3 b(u′ ) and integrating

with respect to u′ on [u0 , u],
Z u
|u′ |3 b(u′ )y(u′ )du′ ≤ δ 1/2 Z(u) (7.261)
u0
therefore substituting in 7.247 yields:
|u|3 y(u) ≤ 2δ 1/2 Z(u) + 2|u|3 g(u) (7.262)
Now, the inequality 7.259 takes in terms of the function Z the form:
dZ
≤ 4δbbZ + w (7.263)
du
where:
w(u) = 4δ 1/2 b(u)b(u)|u|3 g(u) + 2b(u)|u|3 f (u) (7.264)
Integrating 7.263 from u0 , noting that Z(u0 ) = 0, yields:
Z u Z u
Z(u) ≤ exp 4δ (bb)(u )du w(u′ )du′
′′ ′′
(7.265)
u0 u′
Now from the definitions 7.238, 7.241,
(bb)(u) = C|u|−2 (1 + O(|u|−1 ))
hence Z u
4δ (bb)(u′′ )du′′ ≤ log 2
u′
0 , and 7.265 simplifies
to: Z u
Z(u) ≤ 2 w(u′ )du′ (7.266)
u0
257
/2 (µ). Recall that on Cu0
6.179 holds. It follows that:
/ 2 µkL2 (Su,u0 )
k∇ / 2 µkL2 (Su,u0 ) + 2k∇
≤ k∇ / 2 ρkL2 (Su,u0 )
+C|u0 |−4 (R41 (β) + R∞ 2
0 (β)) + O(δ
1/2
|u0 |−5 )(7.267)
(see 7.197). Substituting in 7.235 at u = u0 then yields the following linear

/ 2 µkL2 (Su,u0 ) :
integral inequality for ∇
Z u
/ 2 µkL2 (Su,u0 ) ≤ ã(u0 )
k∇ / 2 µkL2 (Su′ ,u ) du′ + f˜(u0 )
k∇ (7.268)
0
0
where:
ã(u0 ) = a(u0 ) + b(u0 )

≤ Cδ −1/2 |u0 |−1 R∞ (α) + O(|u0 |−1 ) (7.269)
0
˜
f (u0 ) = f (u0 ) + b(u0 ) Cδ|u0 |−4 [R
/2 (ρ)
o
/41 (β) + R∞
+(R 2
0 (β)) ] + O(δ
3/2
|u0 |−5 )
≤ C|u0 |−4 [R∞ /41 (β) + R∞
0 (α)(R /41 (α))2 ] + O(δ 1/2 |u0 |−5 )
0 (β)) + (R
(7.270)
(see 7.247 - 7.249). The integral inequality 7.268 implies:
/ 2 µkL2 (Su,u0 ) ≤ eδã(u0 ) f˜(u0 )

k∇ (7.271)
Since
δã(u0 ) ≤ log 2
0 , 7.271 in turn implies:
/ 2 µkL2 (Su,u0 )
k∇ ≤ 2f˜(u0 )
≤ C|u0 |−4 [R∞ /41 (β) + R∞
0 (α)(R /41 (α))2 ]
0 (β)) + (R
+O(δ 1/2 |u0 |−5 ) (7.272)
It then follows through 7.267 that:
/41 (β) + R∞ 2

D
/2 (µ) ≤ C R /2 (ρ) + (R 0 (β))
+ R∞ /41 (β) + R∞ /41 (α))2

0 (α)(R 0 (β)) + (R
+O(δ 1/2 |u0 |−1 ) (7.273)
Substituting 7.273 in 7.250 we obtain:
|u|3 g(u) ≤ Cδ 1/2 |u|−1 M̃ + O(δ|u|−1 ) + O(δ 2 |u|−3 )k (7.274)
258
where (see 7.201):
M̃ = R
/2 (ρ) + R
/2 (σ) + R/41 (ρ) + R
/41 (σ)
∞ 4 ∞
+(R0 (α) + R /1 (β) + R0 (β))(R /41 (β) + R∞
0 (β))
/41 (α) + R∞
+(R /41 (α)
0 (α))R (7.275)
and the coefficient of k depends only on D0∞ , R∞ 0 . Substituting 7.274 and 7.239
in 7.264 we obtain:
w(u) ≤ C|u|−2 R∞ /41 (β) + R∞ /41 (α))2

0 (α)(R 0 (β)) + (R
+O(δ 1/2 |u|−3 ) + O(δ 5/2 |u|−5 )k (7.276)
hence, from 7.266:
C|u|−1 R∞ /41 (β) + R∞ /41 (α))2

Z(u) ≤ 0 (α)(R 0 (β)) + (R
+O(δ 1/2 |u|−2 ) + O(δ 5/2 |u|−4 )k (7.277)
where the coefficient of k depends only on D0∞ , R∞ 0 . Substituting 7.277 (see

7.260) as well as 7.274 and 7.239 in 7.259 then yields:
z(u) ≤ C|u|−4 R∞ /41 (β) + R∞ /41 (α))2

0 (α)(R 0 (β) + (R
+O(δ 1/2 |u|−5 ) + O(δ 5/2 |u|−7 )k (7.278)

0 . Also, substituting 7.277
and 7.274 in 7.262 yields:
y(u) ≤ Cδ 1/2 |u|−4 M̃ + O(δ|u|−4 ) + O(δ 2 |u|−6 )k (7.279)

0 .
Finally, taking the L2 norm of 7.199 with respect to u on [0, u1 ] we obtain,
in view of the definition 7.245 and the 2nd and 3rd of the definitions 7.1,
/ 3 ηkL2 (Cuu1 )
k∇ ≤ Cy(u) + Cδ 1/2 |u|−4 (R
/2 (ρ) + R
/2 (σ))
+Cδ 1/2 |u|−4 M + O(δ|u|−4 ) (7.280)
In view of 7.257, the bound 7.278 implies in conjunction with 7.280 the first
estimate of the lemma. Also, taking the L2 norm of 7.200 with respect to u
on [0, u1 ] we obtain, in view of the definition 7.246 and the 2nd and 3rd of the
definitions 7.1,
/ 3 ηkL2 (Cuu1 )
k∇ ≤ Cy(u) + Cδ 1/2 |u|−4 (R
/2 (ρ) + R
/2 (σ))
+Cδ 1/2 |u|−4 M + O(δ|u|−4 ) (7.281)
Substituting the bound 7.279 then yields the second estimate of the lemma.
259
7.7 L2 estimate for ∇
/ 3ω
u
We proceed to derive an estimate for ∇ / 3 ω in L2 (Cu 1 ) uniform in u for all
u ∈ [0, s∗ ]. Using this estimate we shall derive an estimate for ∇ / 3 log Ω in
u
L2 (Cu 1 ) uniform in u for all u ∈ [0, s∗ ], which, with a suitable choice of the
constant k, improves the bound 7.162. This shall enable us to show that s∗ = u1
so that the previous lemmas actually hold on the entire parameter domain D1 ,
and, since (u1 , u1 ) ∈ D′ is arbitrary, the estimates hold on all of D′ .
Lemma 7.7 We have:

/ 3 ωkL2 (C u1 ) ≤ Cδ 3/2 |u|−5 N + O(δ 2 |u|−6 ) + O(δ 7/2 |u|−8 )k
k∇
u
for all u ∈ [u0 , s ], provided that δ is suitably small depending on D0∞ , R∞

∗
/41 ,
0 , D
4 ∞ ∞
R/1 , and R
/2 . Moreover, the coefficient of k depends only on D0 , R0 . Here,
N = R /41 (β) + R
/2 (β) + R /41 (ρ) + D0∞ (χ̂)(R /41 (β))
/2 (β) + R
+(D /41 (χ̂))R∞
/41 (trχ) + D /41 (β) + R∞
0 (β) + (R 0 (β))
2
Proof: Consider the propagation equation for the function ω

/, equation 6.202.
Applying /d to this equation we obtain, in view of Lemma 1.2, the following
propagation equation for the S 1-form /dω
/:
Dd
/ω/ + Ωtrχd
/ω / 3 ω) + m′
/ = −2Ω(χ̂, ∇ (7.282)
where:
m′ = −d / − 2(∇
/(Ωtrχ)ω / 2 ω) + /dm
/(Ωχ̂), ∇ (7.283)
′ 2
We shall first estimate m in L (S). The first two terms on the right in 7.283
are estimated in L2 (S) by placing each of the two factors in L4 (S) using the
estimate of Proposition 6.2 pertaining to ω and the results of Chapter 4. We
obtain a bound for these terms in L2 (S) by:
O(δ 1/2 |u|−6 ) (7.284)
To estimate /dm in L2 (S), we consider the expression 6.203 for m. In regard to
the first term on the right in 6.203 we write:
/d(div
/ (Ωχ̂), /dω) = (d
/div
/ (Ωχ̂), /dω) + (div / 2 ω)
/ (Ωχ̂), ∇
Placing each of the two factors of the two terms on the right in L4 (S) using
Proposition 6.1, the estimate of Proposition 6.2 for ω, and the results of Chapter
4 we obtain an L2 (S) bound for the contribution of the first term on the right
in 6.203 to /dm by:
O(δ 1/2 |u|−6 ) (7.285)
In regard to the second term on the right in 6.203 we write:
/ (Ω2 trχβ) =
/ddiv Ω2 trχd
/div / 2 (Ωtrχ), β)
/ β + (∇
/(Ω2 trχ)div
+d /(Ω2 trχ), ∇
/ β + (d /β)
260
The first term on the right is bounded in L2 (S) by:
O(|u|−1 )k∇
/ 2 βkL2 (Su,u ) (7.286)
while the remaining terms are bounded in L2 (S) by:
O(δ|u|−7 ) (7.287)
using Proposition 6.1 and the results of Chapter 4. The contributions to /dm of
the third and fourth terms on the right in 6.203 are bounded in L2 (S) by:
O(δ|u|−3 )(k∇ / 2 σkL2 (Su,u ) ) + O(δ|u|−7 )

/ 2 ρkL2 (Su,u ) + k∇ (7.288)
by Lemmas 6.4 and 4.11, which in conjunction with Lemma 5.2 with p = 4
imply:
/ log ΩkL∞ (Su,u ) ≤ O(δ|u|−3 )
kd (7.289)
To estimate the contribution of the fifth term on the right in 6.203 we write:
1
△
/ log Ω = (div
/ η + div
/ η)
2
and use the first estimate of Proposition 7.1. We obtain in this way an L2 (S)
bound for the contribution in question by:
/ 2 ηkL2 (Su,u ) + O(δ 1/2 |u|−6 )

O(|u|−3 )k∇ (7.290)
The principal part of the contribution of the sixth term on the right in 6.203 to
/dm is:
2Ω2 ((η − η, /d△
/ η) + (η, /d△
/ η))
This part is bounded in L2 (S) by:
O(δ 1/2 |u|−2 )(k∇ / 3 ηkL2 (Su,u ) )

/ 3 ηkL2 (Su,u ) + k∇ (7.291)
while the remainder is bounded by:
O(δ 1/2 |u|−6 ) (7.292)
using the estimates for η, η of Proposition 6.2 and the results of Chapters 3 and
4. Finally, the contribution to /dm of the last term on the right in 6.203 is:
/ [Ω2 (χ̂♯ · β − 2χ̂♯ · β + 2ηρ − 3 ∗ ησ)]

/ddiv (7.293)
The part:
−2d / (Ω2 χ̂♯ · β)
/div
gives the leading contribution in behavior with respect to δ. Using the estimate
of Proposition 6.2 pertaining to χ̂, which in conjunction with Lemma 5.2 with
p = 4 implies:
/χ̂kL∞ (Su,u ) ≤ Cδ 1/2 |u|−3 (R

k∇ /41 (β) + R∞
0 (β)) + O(δ|u|
−4
) (7.294)
261
together with the results of Chapter 3, the contribution in question is bounded
in L2 (S) by:
Cδ 1/2 |u|−2 (D0∞ (χ̂) + O(δ|u|−3/2 ))k∇

/ 2 βkL2 (Su,u )
+C|u|−5 (R
/41 (β) + R∞ 2
0 (β)) + O(δ
1/2
|u|−6 ) (7.295)
The remainder of 7.293 is bounded in L2 (S) by:
O(δ −1/2 |u|−1 )k∇

/ 2 βkL2 (Su,u )
+O(δ 1/2 |u|−2 )(k∇
/ 2 ρkL2 (Su,u ) + k∇
/ 2 σkL2 (Su,u ) )
+O(|u|−3 )k∇
/ 2 ηkL2 (Su,u ) + O(δ 1/2 |u|−6 ) (7.296)
Collecting the above results (7.284 - 7.288, 7.290 - 7.292, 7.295 - 7.296) we
conclude that:
km′ kL2 (Su,u ) ≤ O(δ 1/2 |u|−2 )(k∇

/ 3 ηkL2 (Su,u ) + k∇
/ 3 ηkL2 (Su,u ) )
+O(|u|−3 )k∇
/ 2 ηkL2 (Su,u )
+Cδ 1/2 |u|−2 (D0∞ (χ̂) + O(δ|u|−3/2 ))k∇
/ 2 βkL2 (Su,u )
+O(δ −1/2 |u|−1 )k∇
/ 2 βkL2 (Su,u )
+O(δ 1/2 |u|−2 )(k∇ / 2 σkL2 (Su,u ) )
/ 2 ρkL2 (Su,u ) + k∇
+C|u|−5 (R
/41 (β) + R∞ 2
0 (β)) + O(δ
1/2
|u|−6 ) (7.297)

Hodge system of the type considered in Lemma 7.4 with f = ω / + div
/ (Ωβ),
g = 0. By the first estimate of Lemma 7.4 we then have:
n o
k∇/ 3 ωkL2 (Su,u ) ≤ C kd /ω / 2 βkL2 (Su,u )
/kL2(Su,u ) + k∇
+Cδ|u|−5 [R
/41 (β) + R
/41 (ρ) + D0∞ (χ̂)R
/41 (β)
/41 (trχ) + D
+(D /41 (χ̂))R∞
0 (β)]
+O(δ 3/2 |u|−6 ) (7.298)
using the estimate 6.222 (and Lemma 6.4) as well as the results of Chapter 4.
n o
kΩ(χ̂, ∇/ 3 ω)kL2 (Su,u ) ≤ Cδ −1/2 |u|−1 R∞0 (α) kd / 2 βkL2 (Su,u )
/kL2(Su,u ) + k∇
/ω
+O(δ 1/2 |u|−6 ) (7.299)
Z u
kd
/ω/kL2(Su,u ) ≤ C / 3 ω) + m′ L2 (S ′ ) du′
−2Ω(χ̂, ∇ (7.300)
u ,u
0
262
We substitute the estimates 7.299 and 7.297. For u ∈ [0, u1 ] we have:
Z u
/ 3 ηkL2 (Su′ ,u ) + k∇
(k∇ / 3 ηkL2 (Su′ ,u ) )du′
0
≤ δ 1/2 (k∇
/ 3 ηkL2 (Cuu1 ) + k∇
/ 3 ηkL2 (Cuu1 )
≤ O(δ|u|−4 ) + O(δ 5/2 |u|−6 )k (7.301)
by Lemma 7.6, the coefficient of k depending only on D0∞ , R∞ 0 ,
Z u
/ 2 ηkL2 (Su′ ,u ) du′ ≤ δ 1/2 k∇
k∇ / 2 ηkL2 (Cuu1 ) ≤ O(δ 3/2 |u|−3 ) (7.302)
0
by the 2nd estimate of Proposition 7.1, and:
Z u1
k∇ / 2 βkL2 (Cuu1 ) ≤ δ 1/2 |u|−3 R
/ 2 βkL2 (Su′ ,u ) du′ ≤ δ 1/2 k∇ /2 (β)
0
Z u1
/ 2 βkL2 (Su′ ,u ) du′ ≤ δ 1/2 k∇
k∇ / 2 βkL2 (Cuu1 ) ≤ δ 2 |u|−5 R
/2 (β)
0
Z u
1
/ 2 ρkL2 (Su′ ,u ) du′ ≤ δ 1/2 k∇
k∇ / 2 ρkL2 (C u1 ) ≤ δ|u|−4 R/2 (ρ)
u
0
Z u1
k∇ / 2 σkL2 (Cuu1 ) ≤ δ|u|−4 R
/ 2 σkL2 (Su′ ,u ) du′ ≤ δ 1/2 k∇ /2 (σ)
0
(7.303)
by the definitions 7.1. We then deduce the following linear integral inequality
∗
for the quantity kd
/ω /kL2(Su,u ) on the domain D1s :
Z u
−1/2 ∞
kd /kL2(Su,u ) ≤ Cδ
/ω R0 (α) kd
/ω /kL2(Su′ ,u ) du′
0
+Cδ|u|−5 D0∞ (χ̂)R /41 (β) + R∞ 2

/2 (β) + (R 0 (β))
+O(δ 3/2 |u|−6 ) + O(δ 3 |u|−8 )k (7.304)

the coefficient of k depending only on D0∞ , R∞ 0 . At fixed u this inequality
is of the form 6.18 with Cδ −1/2 R∞ 0 in the role of the constant a. Since here
condition 6.158 is verified if δ is suitably small depending on R∞ 0 (α), the result
6.159 follows, which here reads:
/kL2(Su,u ) ≤ Cδ|u|−5 D0∞ (χ̂)R /41 (β) + R∞ 2

kd
/ω /2 (β) + (R 0 (β))
+O(δ 3/2 |u|−6 ) + O(δ 3 |u|−8 )k (7.305)

∗
This holds for all (u, u) ∈ D1s . 2
Taking the L norm of 7.298 with respect to u
on [0, u1 ] we obtain, in view of the last of the definitions 7.1,
/ 3 ωkL2 (Cuu1 ) ≤ Ckd
k∇ /ω/kL2(Cuu1 )
+Cδ 3/2 |u|−5 [R /41 (β) + R
/2 (β) + R /41 (ρ) + D0∞ (χ̂)R
/41 (β)
+(D /41 (χ̂))R∞
/41 (trχ) + D 0 (β)]
+O(δ 2 |u|−6 ) (7.306)
263
Noting that
kd
/ω/kL2(Cuu1 ) ≤ δ 1/2 sup kd
/ω/kL2(Su,u )
u∈[0,u1 ]
and substituting the estimate 7.305 then yields the lemma.
k = CN + O(δ 1/2 )
/ 3 log ΩkL2 (C u1 ) ≤ Cδ 3/2 |u|−4 N + O(δ 2 |u|−5 )

k∇
u
holds for all u ∈ [u0 , u1 ], provided that δ is suitably small depending on D0∞ ,
R∞ /41 , R
0 , D /41 , and R
/2 .
Proof: Applying ∇ / to equation 6.223 we obtain, in view of Lemma 4.1,
/ 3 log Ω = ∇
D∇ / 3 ω − DΓ / 2 log Ω − ∇
/·∇ /DΓ
/ · /d log Ω (7.307)
Here, we denote:
(DΓ / 2 log Ω)ABC

/·∇ /)D
= (DΓ / 2 log Ω)AD
BC (∇ (7.308)
D 2 D 2
+(DΓ/)AC (∇
/ log Ω)BD + (DΓ
/)AB (∇
/ log Ω)CD
and:
(∇
/DΓ
/ · /d log Ω)ABC = ∇ /)D
/A (DΓ BC /
dD log Ω (7.309)
By Lemma 6.4 and the bound 7.209, the estimates of Proposition 6.2 pertaining
to χ and the bound 7.220, the second and third terms on the right in 7.307 are
O(δ 3/2 |u|−6 ) (7.310)
We apply Lemma 4.7 with p = 2 to equation 7.307. Here r = 3, ν = 0, γ = 0.
Taking into account the fact that ∇ / 3 log Ω vanishes on Cu0 we obtain:
Z u
|u|2 k∇
/ 3 log ΩkL2 (Su,u ) ≤ / 3 ωkL2 (Su,u′ ) du′
|u′ |2 k∇ (7.311)
u0
Z u
+ |u′ |2 kDΓ / 2 log Ω + ∇
/·∇ / · /dωkL2(Su,u′ ) du′
/DΓ
u0
By the bound 7.310 the second integral on the right is bounded by:
O(δ 3/2 |u|−3 ) (7.312)
We then take the L2 norm of 7.311 with respect to u on [0, u1 ] to obtain:

Z u
|u|2 k∇
/ 3 log ΩkL2 (C u1 ) ≤ |u′ |2 k∇
/ 3 ωkL2 (C u1 ) du′
u u
u0
+O(δ 2 |u|−3 ) (7.313)
264
/ 3 ω of Lemma 7.7 then yields:
Substituting the estimate for ∇
/ 3 log ΩkL2 (C u1 )
k∇ ≤ Cδ 3/2 |u|−4 N + O(δ 2 |u|−5 )
u
+O(δ 7/2 |u|−7 )k (7.314)

0 . In reference to this estimate
let:
a = CN + O(δ 1/2 )
b = O(1) (7.315)
so that the coefficient of k in 7.314 is δ 7/2 |u|−7 b, with b depending only on D0∞ ,
R∞0 . The estimate 7.314 then implies:
/ 3 log ΩkL2 (Cuu1 ) ≤ δ 3/2 (a + δ 2 bk)|u|−4 : ∀u ∈ [u0 , s∗ ]

k∇ (7.316)
Choosing
k = 2a (7.317)
we have
a + δ 2 bk < 2a provided that 2bδ 2 < 1 (7.318)
The last is a smallness condition on δ depending only on D0∞ , R∞
0 . The estimate
7.316 then implies
/ 3 log ΩkL2 (Cuu1 ) < kδ 3/2 |u|−4 : ∀u ∈ [u0 , s∗ ]
k∇ (7.319)
∗
u
7.8, the L2 (Cu 1 ) estimates of Lemmas 7.5, 7.6 and 7.7 hold for all u ∈ [u0 , u1 ],
in particular for u = u1 . Now, the right hand sides of these inequalities are
independent of u1 or u1 and (u1 , u1 ) ∈ D′ is arbitrary. Taking for any given
u1 ∈ [u0 , c∗ ) the limit u1 → min{δ, c∗ − u1 }, we conclude that the estimates
u
hold with Cu11 replaced by Cu1 , for all u1 ∈ [u0 , c∗ ).
u
Moreover, by virtue of the L2 (Cu 1 ) estimate for ∇ / 3 log Ω of Lemma 7.8, we
can now estimate (see 7.174):
/ 2 rkL2 (Cuu1 ) ≤ Cδ 3/2 |u|−6 N + O(δ 2 |u|−7 )
k∇ (7.320)
Using this bound we deduce, following the argument leading to the estimate
7.189,
/ 3 trχ′ kL2 (Cuu1 ) ≤ Cδ 3/2 |u|−5 B ′ + O(δ 2 |u|−6 ) : ∀u ∈ [u0 , u1 ]
k∇ (7.321)
where:
B′ = D
/3 (trχ) + R /41 (β) + R
/2 (β) + R /42 (ρ)
+(D /41 (χ̂) + D0∞ (χ̂))(R
/41 (trχ) + D /41 (β) + R∞
0 (β))
+D0∞ (χ̂)(R /41 (β)) + (R

/2 (β) + R /41 (β) + R∞
0 (β))
2
(7.322)
265
Proposition 7.3 The following estimates hold for all u ∈ [u0 , c∗ ):

/ 3 trχ′ kL2 (Cu ) ≤ Cδ 3/2 |u|−5 B ′ + O(δ 2 |u|−6 )
k∇
/ 3 χ̂′ kL2 (Cu ) ≤ Cδ|u|−4 (R
k∇ /41 (β) + R∞
/2 (β) + R 0 (β)) + O(δ
3/2
|u|−5 )
/ 3 ηkL2 (Cu ) ≤ Cδ 1/2 |u|−4 M̃ + O(δ|u|−4 )
k∇
/ 3 ηkL2 (Cu ) ≤ Cδ 1/2 |u|−4 M̃ + O(δ|u|−4 )
k∇
/ 3 ωkL2 (Cu ) ≤ Cδ 3/2 |u|−5 N + O(δ 2 |u|−6 )
k∇
0 , D /41 , and R
/2 . Here
′
B is defined by 7.322, M̃ and M̃ are defined in the statement of Lemma 7.6
and N is defined in the statement of Lemma 7.7.

/ 2 ω and ∇
/ 3ω
As we have already remarked, the estimate of Chapter 6 for ∇ / 2 ω in L4 (S) looses
1/2
a factor of δ in behavior with respect to δ in comparison with the L4 (S)
estimate for /dω of Chapter 4 (as well as a loss of a factor of |u|−1 in decay).
We shall presently derive an L2 estimate for ∇ / 2 ω using only the propagation
equation 4.177, in which no losses are present.

/ ωkL2 (Cu ) ≤ Cδ 1/2 |u|−3 R
k∇ 2
/2 (ρ) + O(δ|u|−4 )
0 , D
4
R/1 , and R
/2 .
Proof: We consider any (u1 , u1 ) ∈ D′ and fix attention to the parameter
subdomain D1 . We apply Lemma 4.1 to the propagation equation 4.177 to
obtain the following propagation equation for ∇ / 2 ω:
/ 2 ω = Ω2 l ′
D∇ (7.323)
where:
l′ = −Ω−2 DΓ / log Ω ⊗ l′ + ∇
/ · /dω + 2d /l (7.324)
To equation 7.323 we apply Lemma 4.7 with ∇ / ω in the role of θ and Ω2 l′ in
2
the role of ξ. Then r = 2, ν = 0, γ = 0. In view of the fact that ∇ / 2 ω vanishes

on Cu0 we obtain, taking p = 2:
Z u
/ 2 ωkL2 (Su,u ) ≤ C
|u|k∇ |u′ |kl′ kL2 (Su,u′ ) du′ (7.325)
u0
Now, by the bound 6.225 and Proposition 4.3 the first term on the right in 7.324
is bounded in L2 (S) by:
O(δ 1/2 |u|−5 ) (7.326)
266
By the bounds 7.289 and 4.180 the second term on the right in 7.324 is bounded
in L2 (S) by:
O(δ|u|−6 ) (7.327)
In regard to the last term on the right in 7.324, l is given by 4.178. Using the
estimates of Proposition 6.2 pertaining to η, η and the results of Chapters 3 and
4 we deduce:
k∇ / 2 ρkL2 (Su,u ) + O(δ 1/2 |u|−5 )

/lkL2 (Su,u ) ≤ k∇ (7.328)
Collecting the above results (7.326 - 7.328) we conclude that:
kl′ kL2 (Su,u ) ≤ k∇

/ 2 ρkL2 (Su,u ) + O(δ 1/2 |u|−5 ) (7.329)
Substituting this in 7.325 we obtain:

Z u
2
/ ωkL2 (Su,u ) ≤ C
|u|k∇ |u′ |k∇
/ 2 ρkL2 (Su,u′ ) du′ + O(δ 1/2 |u|−3 ) (7.330)
u0
Taking the L2 norm with respect to u on [0, u1 ] then yields:

Z u
/ 2 ωkL2 (Cuu1 ) ≤ C
|u|k∇ |u′ |k∇
/ 2 ρkL2 (C u1 ) du′ + O(δ|u|−3 ) (7.331)
u0 u′
Since by the second of the definitions 7.1
/ 2 ρkL2 (C u1 ) ≤ k∇
k∇ / 2 ρkL2 (Cu′ ) ≤ δ 1/2 |u′ |−4 R
/2 (ρ)
u′
it follows that:
/ 2 ωkL2 (C u1 ) ≤ Cδ 1/2 |u|−3 R

k∇ /2 (ρ) + O(δ|u|−4 ) (7.332)
u
This holds for any u ∈ [u0 , u1 ], in particular for u = u1 . Since (u1 , u1 ) ∈ D′

is arbitrary, taking for any given u1 ∈ [u0 , c∗ ) the limit u1 → min{δ, c∗ − u1 }
yields the lemma.
We now turn to estimate ∇ / 3 ω in L2 , using the propagation equation 6.238

for ω
/ coupled to the elliptic equation 6.240.
/ 3 ωkL2 (Cu ) ≤ C|u|−3 (R

k∇ /41 (β)) + O(δ 1/2 |u|−4 )
/2 (β) + R
0 , D
4
R/1 , and R
/2 .
Proof: We consider again any (u1 , u1 ) ∈ D′ and fix attention to the parameter
subdomain D1 . Applying /d to the propagation equation 6.238 we obtain, in
view of Lemma 1.2, the following propagation equation for the S 1-form /dω /:
/ω
Dd / + Ωtrχd
/ω / 3 ω) + m′
/ = −2Ω(χ̂, ∇ (7.333)
267
where:
m′ = −d
/(Ωtrχ)ω
/ − 2(∇ / 2 ω) + /dm
/(Ωχ̂), ∇ (7.334)
We shall first estimate m′ in L2 (S). The first two terms on the right in
7.334 are estimated in L2 (S) by placing each of the two factors in L4 (S) using
Proposition 6.3 (see 6.256) and the results of Chapter 4. We obtain a bound for
these terms in L2 (S) by:
O(|u|−5 ) (7.335)
To estimate /dm in L2 (S), we consider the expression 6.239 for m. In regard to
the first term on the right in 6.239 we write:
/d(div
/ (Ωχ̂), /dω) = (d
/div
/ (Ωχ̂), /dω) + (div / 2 ω)
/ (Ωχ̂), ∇
Placing each of the two factors of the two terms on the right in L4 (S) using
Propositions 6.2 and 6.3 and the results of Chapter 4 we obtain an L2 (S) bound
for the contribution of the first term on the right in 6.239 to /dm by:
O(|u|−5 ) (7.336)
In regard to the second term on the right in 6.239 we write:
/ (Ω2 trχβ) =
/ddiv Ω2 trχd
/div / 2 (Ωtrχ), β)
/ β + (∇
/(Ω2 trχ)div
+d /(Ω2 trχ), ∇
/ β + (d /β)
The first term on the right is bounded in L2 (S) by:
C(|u|−1 + O(δ|u|−2 ))k∇

/ 2 βkL2 (Su,u ) (7.337)
while the remaining terms are bounded in L2 (S) by:
O(δ 1/2 |u|−6 ) (7.338)
using Proposition 6.2 and the results of Chapter 4. The contributions to /dm
of the third, fourth, fifth, sixth and seventh terms on the right in 6.239 are
bounded in a similar manner as the contributions to /dm of the corresponding
terms on the right in 6.203 (see 7.288, 7.290 - 7.302, 7.295 - 7.296). We obtain
an L2 (S) bound for the contributions of these terms to /dm by:
O(δ 1/2 |u|−2 )(k∇ / 3 ηkL2 (Su,u ) )

/ 3 ηkL2 (Su,u ) + k∇
+O(|u|−3 )k∇
/ 2 ηkL2 (Su,u )
+O(δ 1/2 |u|−2 )k∇
/ 2 βkL2 (Su,u )
+O(δ −1/2 |u|−1 )k∇
/ 2 βkL2 (Su,u )
+O(δ 1/2 |u|−2 )(k∇
/ 2 ρkL2 (Su,u ) + k∇
/ 2 σkL2 (Su,u ) )
+O(|u|−5 ) (7.339)
268
Collecting the above results (7.335 - 7.339) we conclude that:
km′ kL2 (Su,u ) ≤ O(δ 1/2 |u|−2 )(k∇

/ 3 ηkL2 (Su,u ) + k∇
/ 3 ηkL2 (Su,u ) )
+O(|u|−3 )k∇
/ 2 ηkL2 (Su,u )
+C(|u|−1 + O(δ|u|−2 ))k∇
/ 2 βkL2 (Su,u )
+O(δ −1/2 |u|−1 )k∇
/ 2 βkL2 (Su,u )
+O(δ 1/2 |u|−2 )(k∇
/ 2 ρkL2 (Su,u ) + k∇
/ 2 σkL2 (Su,u ) )
+O(|u|−5 ) (7.340)

Hodge system of the type considered in Lemma 7.4 with f = ω / − div
/ (Ωβ),
g = 0. By the first estimate of Lemma 7.4 we then have:
n o
/ 3 ωkL2 (Su,u ) ≤ C kd
k∇ /ω / 2 βkL2 (Su,u )
/kL2(Su,u ) + k∇
+Cδ −1/2 |u|−3 R
/41 (β) + O(|u|−4 ) (7.341)
using the estimate 6.256 (and lemma 6.4) as well as the results of Chapter 4. It
follows that in refeference to the first term on the right in 7.282 we have:
/ 3 ω)kL2 (Su,u ) ≤
kΩ(χ̂, ∇
n o
Cδ 1/2 |u|−2 (D(χ̂) + O(δ|u|−3/2 )) kd
/ω / 2 βkL2 (Su,u )
/kL2(Su,u ) + k∇
+O(|u|−5 ) (7.342)
|u|2 kd
/ω/kL2(Su,u ) ≤ C|u0 |2 kω
/kL2(Su,u0 ) (7.343)
Z u
+C |u′ |2 −2Ω(χ̂, ∇
/ 3 ω) + m′ L2 (Su,u′ )
du′
u0
Substituting the estimates 7.342 and 7.340 and noting that in view of the fact
that ω vanishes on Cu0 we have, from 6.237,
kd
/ω / 2 βkL2 (Su,u0 ) + O(δ 1/2 |u0 |−5 )
/kL2(Su,u0 ) ≤ Ck∇ (7.344)
we obtain the following linear integral inequality for the quantity |u|2 kd
/ω/kL2(Sub,u ) :
Z u
|u|2 kd
/ω/kL2(Su,u ) ≤ a(u′ )|u′ |2 kd
/ω/kL2(Su,u′ ) du′ + b(u, u) (7.345)
u0
where:
a(u) = Cδ 1/2 |u|−2 (D0∞ (χ̂) + O(δ|u|−3/2 )) (7.346)
269
Z u n
b(u, u) = O(δ 1/2 )(k∇ / 3 ηkL2 (Su,u′ ) )
/ 3 ηkL2 (Su,u′ ) + k∇
u0
+ O(|u′ |−1 )k∇
/ 2 ηkL2 (Su,u′ )
+ C(|u′ | + O(δ 1/2 ))k∇
/ 2 βkL2 (Su,u′ )
+ O(δ 1/2 )(k∇
/ 2 ρkL2 (Su,u′ ) + k∇/ 2 σkL2 (Su,u′ ) )
o
+O(δ −1/2 |u′ |)k∇
/ 2 βkL2 (Su,u′ ) + O(|u′ |−3 ) du′
+C|u0 |2 k∇
/ 2 βkL2 (Su,u0 ) + O(δ 1/2 |u0 |−3 ) (7.347)
At fixed u the inequality 7.345 is of the form 6.81 with the quantity |u|2 kd
/ω/kL2(Su,u )
in the role of x(u). Note that the function b(u, u) is non-decreasing in u. More-
over, we have Z u
a(u′ )du′ ≤ log 2
u0
provided that δ is suitably small depending on D0∞ , R∞ 0 . It follows that 6.89
holds, that is:
|u|2 kd
/ω/kL2(Su,u ) ≤ 2b(u, u) (7.348)
taking the L2 norm of this inequality with respect to u on [0, u1 ] we then obtain:
|u|2 kd
/ω/kL2(Cuu1 ) ≤ 2kb(·, u)kL2 ([0,u1 ]) (7.349)
and from 7.347 we have:

Z u n
kb(·, u)kL2 ([0,u1 ]) = O(δ 1/2 )(k∇ / 3 ηkL2 (C u1 ) )
/ 3 ηkL2 (C u1 ) + k∇
u0 u′ u′
′ −1 2
+ O(|u | )k∇
/ ηkL2 (C u1 )
u′
+ C(|u′ | + O(δ 1/2 ))k∇

/ 2 βkL2 (C u1 )
u′
+ O(δ 1/2 )(k∇

/ 2 ρkL2 (C u1 ) + k∇
/ 2 σkL2 (C u1 ) )
u′ u′
o
−1/2 ′ 2 1/2
+O(δ / βkL2 (C u1 ) + O(δ
|u |)k∇ |u′ |−3 ) du′
u′
2
+C|u0 | k∇ 2
/ βkL2 (C u1 ) + O(δ|u0 |−3 ) (7.350)
u0
Substituting the second estimate of proposition 7.1, the estimates of Proposition

7.3 pertaining to η, η, yields, in view of the definitions 7.1,
kb(·, u)kL2 ([0,u1 ]) ≤ C|u|−1 R

/2 (β) + O(δ 1/2 |u|−2 ) (7.351)
therefore:
kd
/ω/kL2(Cuu1 ) ≤ C|u|−3 R
/2 (β) + O(δ 1/2 |u|−4 ) (7.352)
yields:
kd
/ω/kL2(Cu ) ≤ C|u|−3 R
/2 (β) + O(δ 1/2 |u|−4 ) (7.353)
270
for all u ∈ [u0 , c∗ ). Taking the L2 norm of 7.341 with respect to u on [0, δ), if
u ∈ [u0 , c∗ − δ], on [0, c∗ − u), if u ∈ (c∗ − δ, c∗ ), we obtain:
/ 3 ωkL2 (Cu )
k∇ ≤ Ckd
/ω/kL2(Cu ) (7.354)
+C|u|−3 (R /41 (β)) + O(δ 1/2 |u|−4 )
/2 (β) + R
Substituting then the estimate 7.353 yields the proposition.
7.9 L2 estimates for /dDω, /dDω, D2 ω, D 2 ω

/DωkL2 (Cu ) ≤ δ −1/2 |u|−2 R

kd /1 (Dρ) + O(δ 1/2 |u|−3 )
0 , D
4
R/1 , and R
/2 .
subdomain D1 . Applying /d to the propagation equation 4.183 we obtain, in
view of Lemma 1.2 the following propagation equation for /dDω:
/Dω) = Ω2 n′
D(d (7.355)
where:
n′ = 2(d
/ log Ω)n + /dn (7.356)
To equation 7.355 we apply Lemma 4.7 with /dDω in the role of θ and Ω2 n′ in
the role of ξ. Then r = 1, ν = 0, γ = 0. In view of the fact that /dDω vanishes
Z u
/DωkL2(Su,u ) ≤ C
kd kn′ kL2 (Su,u′ ) du′ (7.357)
u0
Taking the L2 norm of this inequality with respect to u on [0, u1 ] yields:

Z u
/DωkL2(Cuu1 ) ≤ C
kd kn′ kL2 (C u1 ) du′ (7.358)
u0 u′
Now, by the bounds 7.289 and 4.186 the first term on the right in 7.356 is
O(|u|−4 )
u
therefore in L2 (Cu 1 ) by:
O(δ 1/2 |u|−4 ) (7.359)
In regard to /dn, the second term on the right in 7.356, n is given by 4.184. By
the first of the definitions 7.2 the contribution to /dn of the first term on the
u
right in 4.184 is bounded in L2 (Cu 1 ) by:
δ −1/2 |u|−3 R
/1 (Dρ) (7.360)
271
By Proposition 7.4 and the results of Chapters 3 and 4 the contribution to /dn
u
of the second term on the right in 4.184 is bounded in L2 (Cu 1 ) by:
O(δ|u|−5 ) (7.361)
By the results of Chapters 3 and 4 the contribution to /dn of the remaining terms
on the right in 4.184 is bounded in L2 (S) by:
O(|u|−4 )
u
O(δ 1/2 |u|−4 ) (7.362)
Combining 7.360 - 7.362 we obtain:
/nkL2 (Cuu1 ) ≤ δ −1/2 |u|−3 R

kd /1 (Dρ) + O(δ 1/2 |u|−4 ) (7.363)
Combining the results 7.359 and 7.363 yields:
kn′ kL2 (C u1 ) ≤ δ −1/2 |u|−3 R

/1 (Dρ) + O(δ 1/2 |u|−4 ) (7.364)
u
We substitute this bound in 7.358 to obtain:
/DωkL2(Cuu1 ) ≤ δ −1/2 |u|−2 R

kd /1 (Dρ) + O(δ 1/2 |u|−3 ) (7.365)

/DωkL2 (Su,u ) ≤ Cδ|u|−4 R

kd /41 (ρ) + O(δ 2 |u|−5 )
0 , D
4
R/1 , and R
/2 .
Proof: Applying /d to the propagation equation 4.187, we obtain, in view of
Lemma 1.2, the following propagation equation for /dDω:
/Dω) = Ω2 n′
D(d (7.366)
where:
n′ = 2(d
/ log Ω)n + /dn (7.367)
To equation 7.366 we apply Lemma 4.6 with /dDω in the role of θ and Ω2 n′ in
the role of ξ. Here r = 1, ν = 0, γ = 0. Taking p = 2 we obtain:
Z u
kd
/DωkL2 (Su,u ) ≤ C kn′ kL2 (Su′ ,u ) du′
0
1/2
≤ Cδ kn′ kL2 (Cu ) (7.368)
272
Now, by the bounds 7.289 and 4.190 in which 4.192 is substituted, the first term
on the right in 7.367 is bounded in L2 (S) by:
O(δ|u|−6 )
therefore in L2 (Cu ) by:

O(δ 3/2 |u|−6 ) (7.369)
In regard to /dn, the second term on the right in 7.367, n is given by 4.188. The
contribution to /dn of the first term on the right in 4.188 is bounded in L2 (Cu )
by:
/DρkL2 (Cu ) ≤ Cδ 1/2 |u|−4 R
kd /41 (ρ) + O(δ 3/2 |u|−5 ) (7.370)
from 4.191 and the fourth of the definitions 7.1. By the estimate for ω of
Proposition 6.2 and the results of Chapters 3 and 4 the contribution to /dn of
the second term on the right in 4.188 is bounded in L2 (S) by:
O(δ 3/2 |u|−6 )

O(δ 2 |u|−6 ) (7.371)
By the results of Chapters 3 and 4 the contribution of the remaining terms on
the right in 4.188 is bounded in L2 (S) by:
O(δ|u|−5 )

O(δ 3/2 |u|−5 ) (7.372)
/nkL2(Cu ) ≤ Cδ 1/2 |u|−4 R

kd /41 (ρ) + O(δ 3/2 |u|−5 ) (7.373)
Combining the results 7.369 and 7.373 yields:
kn′ kL2 (Cu ) ≤ Cδ 1/2 |u|−4 R

/41 (ρ) + O(δ 3/2 |u|−5 ) (7.374)
Substituting this in 7.368 yields the proposition.
kD2 ωkL2 (Cu ) ≤ Cδ −3/2 |u|−1 R0 (D2 ρ) + O(δ −1/2 |u|−2 )
0 , D
4
R/1 , and R
/2 .
subdomain D1 . By the commutation formula 1.91 we have:
DD2 ω = DDDω + 2Ω2 (η − η)♯ · /dDω (7.375)
273
Applying D to equation 4.183 we then obtain the following propagation equation
for D2 ω:
D(D2 ω) = Ω2 ṅ (7.376)
where:
ṅ = 2(η − η)♯ · /dDω + 2ωn + Dn (7.377)
To equation 7.376 we apply Lemma 4.7 with D2 ω in the role of θ and Ω2 ṅ in
the role of ξ. Then r = 0, ν = 0, γ = 0. In view of the fact that D2 ω vanishes
Z u
|u|−1 kD2 ωkL2 (Su,u ) ≤ C |u′ |−1 kṅkL2 (Su,u′ ) du′ (7.378)
u0
Taking the L2 norm of this inequality with respect to u on [0, u1 ] yields:

Z u
|u|−1 kD2 ωkL2 (Cuu1 ) ≤ C |u′ |−1 kṅkL2 (C u1 ) du′ (7.379)
u0 u′
Now, by Proposition 7.6 and the results of Chapter 3 the first term on the right
u
in 7.377 is bounded in L2 (Cu 1 ) by:
O(|u|−4 ) (7.380)
Also, by the bound 4.186 and the results of Chapter 3 the second term on the
right in 7.377 is bounded in L2 (S) by:
O(δ −1 |u|−3 )
u
O(δ −1/2 |u|−3 ) (7.381)
By the third of the definitions 7.2 the contribution to Dn of the first term on
u
the right in 4.184 is bounded in L2 (Cu 1 ) by:
δ −3/2 |u|−2 R0 (D2 ρ) (7.382)
By Proposition 7.6, equations 1.82 and 1.174 and the results of Chapters 3 and
4 the contribution to Dn of the second term on the right in 4.184 is bounded in
u
L2 (Cu 1 ) by:
O(|u|−4 ) (7.383)
Now, the third of the Bianchi identities of Proposition 1.2 reads:
3
/ α + (η ♯ + 2ζ ♯ ) · α
Dβ + Ωtrχβ − Ωχ̂♯ · β − ωβ = Ω div

(7.384)
2
In view of the results of Chapter 3 it follows that:
kDβkL4 (Su,u ) ≤ Cδ −3/2 |u|−3/2 R

/41 (α) + O(δ −1 |u|−5/2 ) (7.385)
274
Also, the seventh of the Bianchi identities of Proposition 1.2 reads:

3 1
Dρ + ωtrχρ = ω div / β + (2η + ζ, β) − (χ̂, α) (7.386)
2 2
kDρkL4 (Su,u ) ≤ Cδ −1 |u|−5/2 D(χ̂)R∞

0 (α) + O(δ
−1/2
|u|−5/2 ) (7.387)
Using the estimates 7.385 and 7.387 together with equations 1.54, 1.82 and 1.174
and the results of Chapters 3 and 4 the contribution to Dn of the remaining
terms on the right in 4.184 is shown to be bounded in L2 (S) by:
O(δ −1 |u|−3 )
u
O(δ −1/2 |u|−3 ) (7.388)
Combining 7.382, 7.383 and 7.388 we obtain:
kDnkL2 (Cuu1 ) ≤ δ −3/2 |u|−2 R0 (D2 ρ) + O(δ −1/2 |u|−3 ) (7.389)
kṅkL2 (Cuu1 ) ≤ δ −3/2 |u|−2 R0 (D2 ρ) + O(δ −1/2 |u|−3 ) (7.390)
We substitute this bound in 7.358 to obtain:
kD2 ωkL2 (Cuu1 ) ≤ Cδ −3/2 |u|−1 R0 (D2 ρ) + O(δ −1/2 |u|−2 ) (7.391)

In the following proposition we assume, in addition to the previous assump-

tions on the curvature components, that also the quantity:

R40 (D̂α) = sup |u|5 δ −3/2 kD̂αkL4 (Su,u ) (7.392)
(u,u)∈D′
is finite. By the results of Chapter 2, the corresponding quantity on Cu0T, ob-

tained by replacing the supremum on D′ by the supremum on ([0, δ]×{u0 }) D′ ,
is bounded by a non-negative non-decreasing continuous function of M7 .
h i
kD2 ωkL2 (Su,u ) ≤ Cδ|u|−4 R∞ 2
0 (ρ) + Cδ |u|
−5
R/1 (Dβ) + R∞ 4
0 (α)R0 (D̂α)
+O(δ 2 |u|−5 )
0 , D
4
R/1 , and R
/2 .
275
DD2 ω = DDDω − 2Ω2 (η − η)♯ · /dDω (7.393)
Applying D to equation 4.187 we then obtain the following propagation equation

for D2 ω:
D(D2 ω) = Ω2 ṅ (7.394)
where:
ṅ = −2(η − η)♯ · /dDω + 2ωn + Dn (7.395)
2
To equation 7.394 we apply Lemma 4.6 with D ω in the role of θ and Ω2 ṅ in
the role of ξ. Here r = 0, ν = 0, γ = 0. Taking p = 2 we obtain:
Z u
kD2 ωkL2 (Su,u ) ≤ C kṅkL2 (Su′ ,u ) du′
0
≤ Cδ 1/2 kṅkL2 (Cu ) (7.396)
Now by Proposition 7.7 and the results of Chapter 3 the first term on the right
in 7.395 is bounded in L2 (S) by:
O(δ 3/2 |u|−6 )

O(δ 2 |u|−6 ) (7.397)
Also, by the bound 4.190 in which 4.192 is substituted and the results of Chapter
3, the second term on the right in 7.395 is bounded in L2 (S) by:
O(δ|u|−6 )

O(δ 3/2 |u|−6 ) (7.398)
Now, the fourth of the Bianchi identities of Proposition 1.2 reads:
3
Dβ + Ωtrχβ − Ωχ̂♯ · β − ωβ = −Ω div
/ α + (η ♯ − 2ζ ♯ ) · α

(7.399)
2
kDβkL4 (Su,u ) ≤ Cδ|u|−9/2 R∞

0 (β) + O(δ
3/2
|u|−5 ) (7.400)
The contribution to Dn of the first term on the right in 4.188 is −D2 ρ. This
can be expressed in terms of div
/ Dβ, Dα and Dρ by applying D to equation
4.191 and using equations 1.83, 1.171, 1.173, 3.9, and the conjugate of the
commutation formula 6.107, which reads, for an arbitrary S 1-form θ:
/ θ − div
Ddiv / (Ωχ̂♯ · θ) − Ωtrχdiv
/ Dθ = −2div / θ (7.401)
276
Using this expression and the bound 4.192 we deduce:
kD2 ρkL2 (Cu ) ≤ Cδ 1/2 |u|−4 R∞ 3/2

|u|−5 R/1 (Dβ) + R∞ 4

0 (ρ) + Cδ 0 (α)R0 (Dα)
+O(δ 3/2 |u|−5 ) (7.402)
By Proposition 7.7, equations 1.83 and 1.173 and the results of Chapters 3 and
4 the contribution to Dn of the second term on the right in 4.188 is bounded in
L2 (S) by:
O(δ 3/2 |u|−6 )
O(δ 2 |u|−6 ) (7.403)
Using the estimates 7.400 and 4.192 together with equations 1.59, 1.83 and 1.173
and the results of Chapters 3 and 4 the contribution to Dn of the remaining
terms on the right in 4.188 is shown to be bounded in L2 (S) by:
O(δ|u|−5 )

O(δ 3/2 |u|−5 ) (7.404)
h i
kDnkL2 (Cu ) ≤ Cδ 1/2 |u|−4 R∞
0 (ρ) + Cδ 3/2
|u|−5
R/ 1 (Dβ) + R∞
0 (α)R4
0 (D̂α)
+O(δ 3/2 |u|−5 ) (7.405)

h i
kṅkL2 (Cu ) ≤ Cδ 1/2 |u|−4 R∞
0 (ρ) + Cδ
3/2
|u|−5 R/1 (Dβ) + R∞ 4
0 (α)R0 (D̂α)
+O(δ 3/2 |u|−5 ) (7.406)
Finally, substituting this in 7.396 yields the proposition.
277
Chapter 8
The Multiplier Fields and

the Commutation Fields
8.1 Introduction
There are two kinds of vectorfields which play a fundamental role in our ap-
proach. The multiplier fields and the commutation fields. The two kinds of
vectorfields play distinct roles.
The multiplier vectorfields are used in conjunction with the Bel-Robinson
tensor associated to a Weyl field to construct positive quantities, the energies
and fluxes. This construction shall be presented in Chapter 12. It is the energies
and fluxes which are used to control the solution. Their positivity requires the
multiplier fields to be non-spacelike and future directed. We take as multiplier
fields the vectorfields L and K. We are already familiar with the vectorfield L
and its conjugate L. The vectorfield K is defined by:
K = u2 L (8.1)
and corresponds, in the present context, to the generator of inverted time trans-
lations in Minkowski spacetime.
The commutation fields are used to generate the higher order Weyl fields
from the fundamental Weyl field, the curvature of the Ricci-flat metric. These
higher order Weyl fields are variations of the fudamental Weyl field, modified
Lie derivatives of the curvature with respect to the commutation fields. The
modification is dictated by the conformal properties that a Weyl field is to
possess. This shall also be presented in Chapter 12. The basic requirement on
the set of commutation fields is that it spans the tangent space to the spacetime
manifold at each point. We take as commutation fields the vectorfields L, S
and the three rotation fields Oi : i = 1, 2, 3. The vectorfield S is defined by:
S = uL + uL (8.2)
278
and corresponds to the generator of scale transformations in Minkowski space-
time. It is spacelike in M ′ , the non-trivial part of the spacetime manifold,
timelike and past directed in the Minkowskian region M0 . The rotation fields,
which shall be constructed in the 3rd section of the present chapter, generate
an action of the 3-dimensional rotation group SO(3) on the spacetime mani-
fold. They are spacelike, being tangential to the surfaces Su,u . Note on the
other hand that the vectorfields L, S and K are orthogonal to the surfaces Su,u .
Note also that the vectorfield L plays a dual role, for, it plays the role of a
commutation field as well as that of a multiplier field.
The growth of the energies and fluxes shall be shown in Chapter 12 to
depend on the deformation tensors of the multiplier and commutation fields.
For any vectorfield X on spacetime, we denote by (X) π the Lie derivative of
the spacetime metric g with respect to X:
(X)
π = LX g (8.3)
The deformation tensor of X is then the trace-free part of (X) π, which we

denote by (X) π̃:
(X) 1
π̃ = (X) π − gtr (X) π (8.4)
4
The deformation tensor of a vectorfield X measures the variation of the confor-
mal geometry under the 1-parameter group of diffeomorphisms generated by X,
the trace part of (X) π giving rise only to variations of g within its conformal
class:
1
LX dµg = tr (X) πdµg (8.5)
2
dµg being the volume form of g. The reason why only variations of the confor-
mal geometry affect the growth of the energies and fluxes lies in the conformal
properties of the Bianchi equations satisfied by a Weyl field, which are simply
the Bianchi identities for a Ricci-flat metric in the case of the fundamental Weyl
field, the spacetime curvature.
For an arbitrary vectorfield X on spacetime, the associated tensorfield (X) π
decomposes into the symmetric 2-covariant S tensorfield (X) π /, the restriction
of (X) π to the Su,u , given on each Su,u by:
(X) (X)
π
/(Y, Z) = π(Y, Z) : ∀Y, Z ∈ Tp Su,u ∀p ∈ Su,u (8.6)
(X) (X)
the S 1-forms π
/3 , π
/4 , given by:
(X) (X) (X) (X)
π
/3 (Y ) = π(L̂, Y ), π
/4 (Y ) = π(L̂, Y ) : ∀Y ∈ Tp (Su,u ) ∀p ∈ Su,u
(8.7)
(X) (X)
and the functions π33 , π44 and (X) π34 , given by:
(X) (X) (X) (X) (X) (X)
π33 = π(L̂, L̂), π44 = π(L̂, L̂), π34 = π(L̂, L̂) (8.8)
The deformation tensor of X similarly decomposes into the symmetric 2-covariant

/, the S 1-forms (X) π̃
S tensorfield (X) π̃ /4 , and the functions (X) π̃33 ,
/3 , (X) π̃
279
(X) (X)
π̃44 and π̃34 . We have:
(X) 1
(X)
π̃
/= / − g/tr (X) π
π
4
(X)
/3 = (X) π
π̃ /3 , (X)
π̃
/4 = (X)
π
/4
(X) (X) (X) (X)
π̃33 = π33 , π̃44 = π44
(X) 1
π̃34 = (X) π34 + tr (X)
π (8.9)
2
Note that:
(X) (X) (X)
tr π = tr π
/− π34 (8.10)
In the following we denote:
(X) (X)
π̃
/= i
(X) (X) (X) (X)
π̃
/3 = m, π̃
/4 = m
(X) (X) (X) (X)
π̃33 = n, π̃44 = n
(X) (X)
π̃34 = j (8.11)
(X) X)
Also, we denote by î the trace-free part of i. Note that since
(X)
tr π̃ = 0
we have:
(X) (X)
tr i= j (8.12)
8.2 L∞ estimates for the deformation tensors of

L, K and S
If X, Y, Z are any three vectorfields on spacetime, we have:
(X)
π(Y, Z) = g(∇Y X, Z) + g(∇Z X, Y ) (8.13)
Consider the case X = L. Taking Y, Z to be S-tangential vectorfields, we

obtain, in view of the definition 1.37,
(L)
π
/ = 2Ωχ (8.14)
Taking Z = L̂ and Y to be an S tangential vectorfield we have:

(L)
π(L̂, Y ) = g(∇L̂ L, Y ) + g(∇Y L, L̂)
and from the second of 1.90
g(∇L̂ L, Y ) = 2Ωη(Y )
280
while from the second of 1.84 and 1.81
g(∇Y L, L̂) = g(∇Y (Ω2 L′ ), ΩL′ ) = 2Ω(η(Y ) − 2Y (log Ω)) = −2Ωη(Y )
Since from 1.81
η − η = 2ζ
we conclude that:
(L)
π
/3 = 4Ωζ (8.15)
Taking Z = L̂ and Y to be an S tangential vectorfield we have:
(L)
π(L̂, Y ) = g(∇L̂ L, Y ) + g(∇Y L, L̂) = 0
hence:
(L)
π
/4 = 0 (8.16)
Taking Y = Z = L̂ we have:
(L)
π(L̂, L̂) = 2g(∇L̂ L, L̂) = 0
by the second of 1.84. Hence:

(L)
π33 = 0 (8.17)
Taking Y = Z = L we have:
(L)
π(L̂, L̂) = 2g(∇L̂ L, L̂) = 0
hence:
(L)
π44 = 0 (8.18)
Finally, taking Y = L̂, Z = L̂ we have:
(L)
π(L̂, L̂) = g(∇L̂ L, L̂) + g(∇L̂ L, L̂) = g(∇L̂ L, L̂) = −4ω
by the second of 1.84 and the first of 1.28. Hence:

(L)
π34 = −4ω (8.19)
(L)
We collect the above results for the components of π in the following table:
(L)
π
/ = 2Ωχ
(L) (L)
π
/3 = 4Ωζ, π
/4 = 0
(L) (L)
π33 = 0, π44 = 0
(L)
π34 = −4ω (8.20)
(L)
The components of π̃ are then given by the following table:
(L)
î = 2Ωχ̂
(L)
j = Ωtrχ − 2ω
(L) (L)
m = 4Ωζ, m=0
(L) (L)
n= n=0 (8.21)
281
Consider next the case X = L. Proceeding along similar lines we derive the
conjugate of table 8.20, which gives the components of (L) π:
(L)
π
/ = 2Ωχ
(L) (L)
π
/3 = 0, π
/4 = −4Ωζ
(L) (L)
π33 = 0, π44 = 0
(L)
π34 = −4ω (8.22)
We turn to the vectorfield K. Now, if X is an arbitrary vectorfield and f an
arbitrary function on spacetime we have, in arbitrary local coordinates,
(f X) (X)
πµν = f πµν + ∂µ f Xν + ∂ν f Xµ (8.23)
where
Xµ = gµν X ν
In the case of the optical functions u, u, we have, from 1.8,
1 1
∂µ u = − L′µ , ∂µ u = − L′µ (8.24)
2 2
In view of 8.23, 8.24, the definition 8.1 implies:
(K)
πµν = u2 (L)
πµν − u(L̂µ L̂ν + L̂µ L̂ν ) (8.25)
Using the table 8.22 we then obtain the following table for the components of
(K)
π:
(K)
/ = 2u2 Ωχ
π
(K) (K)
π
/3 = 0, /4 = −4u2 Ωζ
π
(K) (K)
π33 = 0, π44 = 0
(K) 2
π34 = −4u ω − 4u (8.26)
(K)
The components of π̃ are then given by:
(K)
î = 2u2 Ωχ̂
(K)
j = u2 (Ωtrχ − 2ω) − 2u
(K) (K)
m = 0, m = −4u2 Ωζ
(K) (K)
n= n=0 (8.27)
Finally, we turn to the vectorfield S. In view of 8.23, 8.24, the definition 8.2
implies:
(S)
πµν = u (L) πµν + u (L) πµν − L̂µ L̂ν − L̂µ L̂ν (8.28)
Using tables 8.20 and 8.22 we then obtain the following table for the components
of (S) π:
(S)
π
/ = 2Ω(uχ + uχ)
(S) (S)
π
/3 = 4uΩζ, π
/4 = −4uΩζ
(S) (S)
π33 = 0, π44 = 0
(S)
π34 = −4(uω + uω + 1) (8.29)
282
(S)
the components of π̃ are then given by:
(S)
î = 2Ω(uχ̂ + uχ̂)
(S)
j = u(Ωtrχ − 2ω) + u(Ωtrχ − 2ω) − 2
(S) (S)
m = 4uΩζ, m = −4uΩζ
(S) (S)
n= n=0 (8.30)
We see from tables 8.21, 8.27, 8.30 that:
(X) (X)
n= n=0 (8.31)
for each of the vectorfields L, K, S. The results of Chapter 3 yield, through
these tables, the following L∞ estimates for the remaining components of the
deformation tensors of the vectorfields L, K and S.
(L)
k îkL∞ (Su,u ) ≤ Cδ −1/2 |u|−1 R∞
0 (α)
(L)
k jkL∞ (Su,u ) ≤ O(|u|−1 )
(L)
k mkL∞ (Su,u ) ≤ O(δ 1/2 |u|−2 )
(L)
m=0 (8.32)
(K)
k îkL∞ (Su,u ) ≤ Cδ 1/2 D0∞ (χ̂) + O(δ 3/2 |u|−3/2 )
(K)
k jkL∞ (Su,u ) ≤ O(δ)
(K)
m=0
(K)
k mkL∞ (Su,u ) ≤ O(δ 1/2 ) (8.33)
(S)
k îkL∞ (Su,u ) ≤ Cδ 1/2 |u|−1 (D0∞ (χ̂) + R∞
0 (α)) + O(δ
3/2
|u|−5/2 )
(S)
k jkL∞ (Su,u ) ≤ O(δ|u|−2 )
(S)
k mkL∞ (Su,u ) ≤ O(δ 3/2 |u|−2 )
(S)
k mkL∞ (Su,u ) ≤ O(δ 1/2 |u|−1 ) (8.34)
Here and in the estimates to follow we specify the leading term in behavior
with respect to δ in the case of the (X) î component of the deformation tensors.
We do this because this component enters the borderline error integrals as we
shall see in Chapters 13, 14 and 15. Since the above estimates depend only
on the results of Chapter 3 the symbol O(δ p |u|r ) may be taken here to have
the more restricted meaning of the product of δ p |u|r with a non-negative non-
decreasing continuous function of the quantities D0∞ and R∞ 0 alone.
283
8.3 Construction of the rotation vectorfields Oi
The action of the rotation group SO(3) on the spacetime manifold M is defined
as follows. First we have an action of SO(3) by isometries in the Minkowskian
region M0 , the central timelike geodesic Γ0 being the set of fixed points of the
action and the group orbits being the surfaces Su,u . This action in M0 commutes
with the flow of L as well as that of L in M0 . We extend the action to M ′ , the
non-trivial part of M , in the following manner.
First, we extend the action to Cu0 by conjugation with the flow of L on Cu0 :
(o ∈ SO(3), p ∈ Su,u0 ) 7→ op = Φu (oΦ−u p) ∈ Su,u0 (8.35)
Thus if q = Φ−u p ∈ S0,u0 is the point of intersection with S0,u0 of the generator
of Cu0 through p, then op = Φu (oq) is the point of intersection with Su,u0 of
the generator of Cu0 through oq.
The generating rotation fields Oi : i = 1, 2, 3 on Cu0 are then given by:
Oi (p) = dΦu (q) · Oi (q), q = Φ−u (p) ∈ S0,u0 : ∀p ∈ Su,u0 , ∀u ∈ [0, δ] (8.36)
The Oi are tangential to the surfaces Su,u0 and satisfy:
[L, Oi ] = 0, [Oi , Oj ] = ǫijk Ok : on Cu0 (8.37)
where ǫijk is the fully antisymmetric 3-dimensional symbol.

We then extend the action to M ′ by conjugation with the flow of L on each
C u in M ′ :
(o ∈ SO(3), p ∈ Su,u ) 7→ op = Φu−u0 (oΦu0 −u p) ∈ Su,u (8.38)
Thus if q = Φu0 −u p ∈ Su,u0 is the point of intersection with Su,u0 of the gen-
erator of C u through p, then op = Φu−u0 (oq) is the point of intersection with
Su,u of the generator of C u through oq.
The generating rotation fields Oi : i = 1, 2, 3 on M ′ are then given by:
Oi (p) = dΦu−u0 (q) · Oi (q), q = Φu0 −u (p) ∈ Su,u0 : ∀p ∈ Su,u , ∀(u, u) ∈ D′

(8.39)
The Oi are tangential to the surfaces Su,u and satisfy:
[L, Oi ] = 0, [Oi , Oj ] = ǫijk Ok : on M ′ (8.40)
According to the above definition, the Oi are given in a canonical coordinate

system simply by:
∂
Oi = OiA (ϑ) A (8.41)
∂ϑ
where the OiA (ϑ) are the components of Oi on S0,u0 . In the canonical coordinates
induced by the stereographic coordinates (see Chapter 1), the Oi are given in
284
the north polar chart on MU1 by:
( 2 ! )
1 1 1 2 ∂ 1 2 2 1 1 4 ∂
O1 = − ϑ ϑ + 1 + (ϑ ) − (ϑ )
1 + 41 |ϑ|2 ∂ϑ1 ∂ϑ2

2 4 16
( 2 ! )
1 1 2 1 ∂ 1 1 2 1 2 4 ∂
O2 = ϑ ϑ + 1 + (ϑ ) − (ϑ )
1 + 14 |ϑ|2 ∂ϑ2 ∂ϑ1

2 4 16
∂ ∂
O3 = ϑ1 2
− ϑ2 1 (8.42)
∂ϑ ∂ϑ
and in the south polar chart on MU2 by:
( 2 ! )
1 1 1 2 ∂ 1 2 2 1 1 4 ∂
O1 = ϑ ϑ + 1 + (ϑ ) − (ϑ )
1 + 14 |ϑ|2 ∂ϑ1 ∂ϑ2

2 4 16
( 2 ! )
1 1 2 1 ∂ 1 1 2 1 2 4 ∂
O2 = − ϑ ϑ + 1 + (ϑ ) − (ϑ )
1 + 41 |ϑ|2 ∂ϑ2 ∂ϑ1

2 4 16
∂ ∂
O3 = ϑ1 − ϑ2 1 (8.43)
∂ϑ2 ∂ϑ
From 8.13 we have, for any pair of vectorfields X, Y on spacetime,
(Oi )
π(X, Y ) = g(∇X Oi , Y ) + g(∇Y Oi , X) (8.44)
Taking Y = L and X to be an S tangential vectorfield we obtain:

(Oi )
π(L, X) = g(∇L Oi , X) + g(∇X Oi , L) = g(∇Oi L, X) + g(∇X Oi , L)
= −g(∇Oi X, L) + g(∇X Oi , L) = g([X, Oi ], L) = 0
by the first of 8.40 and the fact that [X, Oi ] is an S tangential vectorfield, both
X and Oi being S tangential vectorfields. We conclude that:
(Oi )
π
/3 = 0 (8.45)
Taking X = Y = L in 8.44 we obtain:

(Oi )
π(L, L) = 2g(∇L Oi , L) = g(∇Oi L, L) = 0
by the first of 8.40. We conclude that:

(Oi )
π33 = 0 (8.46)
Taking X = Y = L in 8.44 we obtain:

(Oi )
π(L, L) = 2g(∇L Oi , L) = −2g(Oi , ∇L L) = 0
hence:
(Oi )
π44 = 0 (8.47)
285
Taking X = L, Y = L in 8.44 we obtain:
(Oi )
π(L, L) = g(∇L Oi , L) + g(∇L Oi , L) = −g(Oi , ∇L L) − g(Oi , ∇L L)
= −2Ω2 (η(Oi ) + η(Oi )) = −4Ω2 Oi (log Ω)
by 1.84 and 1.81. We conclude that:

(Oi )
π34 = −4Oi (log Ω) (8.48)
The components (Oi ) π / and (Oi ) π

/4 of (Oi ) π remain to be considered. We
shall presently show that these components satisfy propagation equations along
the generators of the C u . We start from the identity:
LL LOi g − LOi LL g = L[L,Oi ] g (8.49)
Here LOi g = (Oi ) π and the right hand side vanishes by the first of 8.40. Thus
8.49 becomes:
LL (Oi ) π = LOi LL g (8.50)
Let X, Y be vectorfields defined on and tangential to Su,u0 . For each q ∈ Su,u0
we extend X(q), Y (q) to Jacobi fields along the generator of C u through q by:
X(Φu−u0 (q)) = dΦu−u0 · X(q), Y (Φu−u0 (q)) = dΦu−u0 · Y (q) (8.51)
Then X, Y are S tangential vectorfields satisfying:
[L, X] = [L, Y ] = 0 (8.52)
Let us evaluate 8.50 on X, Y . We have:

(Oi ) (Oi ) (Oi )
(LL π)(X, Y ) = L( π(X, Y )) = L( π
/(X, Y ))
(Oi ) (Oi )
= (LL π
/)(X, Y ) = (D π
/)(X, Y )
while:
(LOi LL g)(X, Y ) = Oi ((LL g)(X, Y )) − LL g([Oi , X], Y ) − LL g(X, [Oi , Y ])

= 2 Oi (Ωχ(X, Y )) − Ωχ([Oi , X], Y ) − Ωχ(X, [Oi , Y ])

=2 L /Oi (Ωχ) (X, Y )
We thus obtain:
(Oi )

(D π /Oi (Ωχ) (X, Y )
/)(X, Y ) = 2 L
and since the restrictions of X, Y to Su,u0 are arbitrary vectorfields tangential

to Su,u0 while Φu−u0 is a diffeomorphism of Su,u0 onto Su,u for each u, it follows
that:
D (Oi ) π /Oi (Ωχ)
/ = 2L (8.53)
286
This is the desired propagation equation for (Oi ) π
/. However, this equation must
be supplemented by an initial condition on Cu0 .
To determine (Oi ) π/ on Cu0 we derive a propagation equation for (Oi ) π
/ along
the generators of Cu0 . We start from the identity:
LL LOi g − LOi LL g = L[L,Oi ] g (8.54)
Now by the first of 8.37 the right hand side vanishes on Cu0 when evaluated on
a pair of vectorfields tangential to Cu0 . Let X, Y be vectorfields defined on and
tangential to S0,u0 . For each q ∈ S0,u0 we extend X(q), Y (q) to Jacobi fields
along the generator of Cu0 through q by:
X(Φu (q)) = dΦu · X(q), Y (Φu (q)) = dΦu · Y (q) (8.55)

Then X, Y are S tangential vectorfields along Cu0 satisfying:
[L, X] = [L, Y ] = 0 : along Cu0 (8.56)
Evaluating 8.54 along Cu0 on X, Y , we have:
(Oi )
(LL π)(X, Y ) = (LL LOi g)(X, Y ) = L( (Oi ) π(X, Y ))
= L( (Oi ) π
/(X, Y )) = (LL (Oi ) π
/)(X, Y ) = (D (Oi )
π
/)(X, Y )
and:
(LOi LL g)(X, Y ) = Oi ((LL g)(X, Y )) − LL g([Oi , X], Y ) − LL g(X, [Oi , Y ])
= 2 {Oi (Ωχ(X, Y )) − Ωχ([Oi , X], Y ) − Ωχ(X, [Oi , Y ])}
/Oi (Ωχ)) (X, Y )
= 2 (L
Since the right hand side of 8.54 vanishes on Cu0 when evaluated on X, Y , we
obtain:
(D (Oi ) π /Oi (Ωχ)) (X, Y )
/)(X, Y ) = 2 (L
and since the restrictions of X, Y to S0,u0 are arbitrary vectorfields tangential
to S0,u0 while Φu is a diffeomorphism of S0,u0 onto Su,u0 for each u, it follows
that:
D (Oi ) π /Oi (Ωχ) : on Cu0
/ = 2L (8.57)
(Oi )
This is the desired propagation equation for / along Cu0 . The initial con-
π
dition for this equation is simply:
(Oi )
π
/ = 0 : on S0,u0 (8.58)
for, the Oi generate isometries in the Minkowskian region M0 .

Note that if X, Y is a pair of S tangential vectorfields, we have:
(Oi )
/(X, Y ) = (Oi ) π(X, Y ) = (LOi g)(X, Y )
π
= Oi (g(X, Y )) − g([Oi , X], Y ) − g(X, [Oi , Y ])
= Oi (g/(X, Y )) − g/([Oi , X], Y ) − g/(X, [Oi , Y ])
/Oi g/) (X, Y )
= (L
287
therefore:
(Oi )
π /Oi g/
/=L (8.59)
Consider the propagation equation 8.53. Decomposing
1
χ = χ̂ + g/trχ
2
we have
1 1
/Oi (Ωχ̂) + g/Oi (Ωtrχ) + ΩtrχL
/Oi (Ωχ) = L
L /Oi g/
2 2
In view of 8.59 we then obtain:
1 1 (Oi )
L /Oi (Ωχ̂) + g/Oi (Ωtrχ) + Ωtrχ
/Oi (Ωχ) = L π
/ (8.60)
2 2
(Oi )
Substituting this in 8.53 brings the propagation equation for π
/ along the
C u to the following form:
(Oi ) (Oi )
D π
/ − Ωtrχ π /Oi (Ωχ̂) + g/Oi (Ωtrχ)
/ = 2L (8.61)
Similarly, we have:
1 1 (Oi )
L /Oi (Ωχ) + g/Oi (Ωtrχ) + Ωtrχ
/Oi (Ωχ) = L π
/ (8.62)
2 2
(Oi )
Substituting this in 8.57 brings the propagation equation for π
/ along Cu0
to the following form:
(Oi ) (Oi )
D π
/ − Ωtrχ π /Oi (Ωχ̂) + g/Oi (Ωtrχ)
/ = 2L (8.63)
We proceed to derive a propagation equation for (O)i π

/4 . We shall do this
wth the help of the following follow-up to Lemma 1.1.
Lemma 8.1 Let X be a vectorfield defined along a given Cu and tangential to

its Su,u sections. Then the vectorfield [L, X], defined along Cu is also tangential
to the Su,u sections and given by:
(X) ♯
[L, X] = Ω π
/4
Also, let X be a vectorfield defined along a given C u and tangential to its Su,u
sections. Then the vectorfield [L, X], defined along C u is also tangential to the
Su,u sections and given by:
(X) ♯
[L, X] = Ω π
/3
Proof: To establish the first part of the lemma, we know from Lemma 1.1 that
[L, X] is S-tangential. Let then Y be an arbitrary S-tangential vectorfield. We
have:
g([L, X], Y ) = g(∇L X, Y ) − g(∇X L, Y ) = g(∇L X, Y ) + g(L, ∇X Y )
288
and:
g(L, ∇X Y ) − g(L, ∇Y X) = g(L, [X, Y ]) = 0
[X, Y ] being S-tangential. Therefore:
(X) (X)
g([L, X], Y ) = g(∇L X, Y ) + g(∇Y X, L) = π(L, Y ) = Ω π
/4
The second part of the lemma is established in a similar manner.
From now on we denote:

Zi = [L, Oi ] (8.64)
According to the above lemma we have:
(Oi )
/4 = Ω−1 g/ · Zi
π (8.65)
(see 1.89). Consider the Jacobi identity:
[L, [L, Oi ]] + [L, [Oi , L]] + [Oi , [L, L]] = 0 (8.66)
Now, by the first of 8.40 the middle term vanishes while according to 1.91:
[L, L] = 4Ω2 ζ ♯
In view also of the definition 8.64, the identity 8.66 reads:
[L, Zi ] + 4[Oi , Ω2 ζ ♯ ] = 0 (8.67)
or:
/Oi (Ω2 ζ ♯ )
DZi = −4L (8.68)
This is the desired propagation equation for Zi . According to the first of 8.37
the initial condition on Cu0 is simply:
Zi = 0 : on Cu0 (8.69)
From the propagation equation 8.68 a propagation equation for (Oi ) π /4 can
readily be derived. However we shall work with the propagation equation 8.68,
as this is simpler, to derive estimates for Zi from which estimates for (Oi ) π
/4
shall thence be readily derived.
8.4 L∞ estimates for the Oi and ∇

/Oi
Proposition 8.1 Given any point p ∈ Su,u , let q = Φu0 −u (p) be the point
on Su,u0 correponding to p along the same generator of C u and let q0 = Φ−u (q)
the point on S0,u0 corresponding to q along the same generator of Cu0 . Then,
provided that δ is suitably small depending on D0∞ , R∞0 , we have:
1 |u| |u|
|Oi (q0 )| ≤ |Oi (p)| ≤ 2 |Oi (q0 )|
2 |u0 | |u0 |
289
for every point p ∈ M ′ .
Proof: We have, relative to an arbitary frame field for the Su,u ,
|Oi |2 = g/AB OiA OiB (8.70)
Consider first Oi along Cu0 . According to the first of 8.37 we have:
DOi = 0 : on Cu0 (8.71)
In view of the first of 1.41, that is
Dg/ = 2Ωχ,
D(|Oi |2 ) = (Dg/)(Oi , Oi )
= 2Ωχ(Oi , Oi )
= Ωtrχ|Oi |2 + 2Ωχ̂(Oi , Oi ) : on Cu0 (8.72)
Therefore:
2 1
D(|Oi | ) ≤ 2 Ωtrχ + Ω|χ̂| |Oi |2 (8.73)
2
and:
1
D(|Oi |2 ) ≥ 2 Ωtrχ − Ω|χ̂| |Oi |2 (8.74)
2
In reference to 8.73, setting:

1
v(t) = |Oi (Φt (q0 ))|, a(t) = Ωtrχ + Ω|χ̂| (Φt (q0 ))
2
we have:
d 2
(v ) ≤ 2av 2
dt
therefore Lemma 3.1 with b = 0 applies and we obtain:
Z u
1 ′
|Oi (q)| ≤ exp Ωtrχ + Ω|χ̂| (Φu′ (q0 ))du (8.75)
0 2
In reference to 8.74 setting:

1
v(t) = |Oi (Φ−t (q))|, a(t) = − Ωtrχ − Ω|χ̂| (Φ−t (q))
2
we again have:
d 2
(v ) ≤ 2av 2
dt
therefore Lemma 3.1 applies and we obtain:
Z u
1 ′
|Oi (q0 )| ≤ exp − Ωtrχ − Ω|χ̂| (Φ−u (q))du
′ (8.76)
0 2
290
Now by the results of Chapter 3 the integrals in the exponentials in 8.75 and
8.76 are bounded by O(δ 1/2 |u|−1 ). Therefore if δ is suitably small depending
on R∞0 these integrals do not exceed (1/2) log 2. It follows that:
1 √
√ |Oi (q0 )| ≤ |Oi (q)| ≤ 2|Oi (q0 )| (8.77)
2
Consider next Oi along C u . According to the first of 8.40 we have:
DOi = 0 : on M ′ (8.78)
In view of the second of 1.41, that is
Dg/ = 2Ωχ,
D(|Oi |2 ) = (Dg/)(Oi , Oi )
= 2Ωχ(Oi , Oi )
= Ωtrχ|Oi |2 + 2Ωχ̂(Oi , Oi ) : on M ′ (8.79)
Therefore:
1
D(|Oi |2 ) ≤ 2 Ωtrχ + Ω|χ̂| |Oi |2 (8.80)
2
and:
1
D(|Oi |2 ) ≥ 2 Ωtrχ − Ω|χ̂| |Oi |2 (8.81)
2
In reference to 8.80, setting:

1
v(t) = |Oi (Φt (q))|, a(t) = Ωtrχ + Ω|χ̂| (Φt (q))
2
we have:
d 2
(v ) ≤ 2av 2
dt
therefore Lemma 3.1 with b = 0 applies and we obtain:
Z u
1
|Oi (p)| ≤ exp Ωtrχ + Ω|χ̂| (Φu′ −u0 (q))du′ (8.82)
u0 2
In reference to 8.81 setting:

1
v(t) = |Oi (Φ−t (p))|, a(t) = − Ωtrχ − Ω|χ̂| (Φ−t (p))
2
we again have:
d 2
(v ) ≤ 2av 2
dt
291
therefore Lemma 3.1 applies and we obtain:
Z u
1
|Oi (q)| ≤ exp − Ωtrχ − Ω|χ̂| (Φu0 −u′ (p))du′ (8.83)
u0 2
Now by the results of Chapter 3 the integral in the exponential in 8.82 is bounded
by
|u|
log + O(δ 1/2 |u|−1 )
|u0 |
while the integral in the exponential in 8.83 is bounded by

|u|
− log + O(δ 1/2 |u|−1 )
|u0 |
If δ is suitably small depending on D0∞ , R∞

0 the terms O(δ
1/2
|u|−1 ) do not
exceed (1/2) log 2. It follows that:
1 |u| √ |u|
√ |Oi (q)| ≤ |Oi (p)| ≤ 2 |Oi (q)| (8.84)
2 |u0 |u0 |
Combining finally 8.84 with 8.77 yields the proposition.
We note that the three rotation fields Oi satisfy on S0,u0 the bound:
kOi kL∞ (S0,u0 ) ≤ |u0 | (8.85)
As S0,u0 lies on the boundary of the Minkowskian region M0 , this follows from
the expression:
∂
Oi = ǫijk xj k (8.86)
∂x
for the Oi in rectangular coordinates in M0 , with the x0 axis coinciding with
the central timelike geodesic Γ0 .

√
k∇
/Oi kL∞ (Su,u ) ≤ 3 2 + 1
0 , D
4
R/1 and R /2 .
Proof: Let X be an arbitrary S-tangential vectorfield. In analogy with Lemma
4.1 we have the commutation formulas:
D∇
/X − ∇
/DX = DΓ
/·X (8.87)
and:
D∇
/X − ∇
/DX = DΓ
/·X (8.88)
292
Here, with respect to an arbitrary local frame field for the Su,u ,
/ · X)B
(DΓ /)B
A = (DΓ
C
AC X , / · X)B
(DΓ /)B
A = (DΓ AC X
C
The formulas 8.87, 8.88 are established by the same argument as that of Lemma
4.1.
/Oi is a type T11 type S tensorfield we have, relative to an arbitrary
Since ∇
frame field for the Su,u ,
/Oi |2 = g/AB (g/−1 )CD ∇

|∇ /C OiA ∇
/D OiB (8.89)
It follows that:
/Oi |2 )
D(|∇ = 2ΩχAB (g/−1 )CD ∇
/C OiA ∇
/D OiB
−2ΩχCD g/AB ∇
/C OiA ∇
/D OiB
+2(∇
/Oi , D∇
/Oi )
and:
/Oi |2 )
D(|∇ = 2ΩχAB (g/−1 )CD ∇
/C OiA ∇
/D OiB
−2ΩχCD g/AB ∇
/C OiA ∇
/D OiB
+2(∇ /Oi )
/Oi , D∇
Substituting the decompositions

1 1
χ = χ̂ + g/trχ, χ = χ̂ + g/trχ
2 2
the trace terms cancel and we obtain:
/Oi |2 )
D(|∇ = 2Ωχ̂AB (g/−1 )CD ∇
/C OiA ∇
/D OiB
−2Ωχ̂CD g/AB ∇
/C OiA ∇
/D OiB
+2(∇
/Oi , D∇
/Oi ) (8.90)
and:
/Oi |2 )
D(|∇ = 2Ωχ̂AB (g/−1 )CD ∇
/C OiA ∇
/D OiB
−2Ωχ̂CD g/AB ∇
/C OiA ∇
/D OiB
+2(∇
/Oi , D∇
/Oi ) (8.91)
Consider first Oi along Cu0 . Since 8.71 holds, taking X = Oi in 8.87 we

obtain:
D∇/Oi = DΓ / · Oi : on Cu0 (8.92)
Substituting in 8.90 we deduce:
/Oi |2 ) ≤ 2|∇
D(|∇ /Oi |(CΩ|χ̂||∇
/Oi | + |DΓ
/||Oi |) (8.93)
293
Given any point q ∈ Su,u0 , let q0 = Φ−u (q) ∈ S0,u0 be the point corresponding
to q along the same generator of Cu0 . Applying Lemma 3.1 to 8.93 then yields:
Z u
′
|(∇
/Oi )(q)| ≤ exp CΩ|χ̂|(Φu′ (q0 ))du |(∇/Oi )(q0 )| (8.94)
0
( )
Z u Z u
+ exp CΩ|χ̂|(Φu′′ (q0 ))du′′ /||Oi |)(Φu′ (q0 ))du′
(|DΓ
0 u′
Now by the results of Chapter 3 the integrals in the exponentials in 8.94 are
bounded by O(δ 1/2 |u|−1 ). Therefore if δ is suitably small depending on R∞
0
these integrals do not exceed (1/2) log 2. Also, by the estimate 7.208,
/kL∞ (Su,u ) ≤ O(δ −1/2 |u|−2 )
kDΓ (8.95)
Taking also into account Proposition 8.1 and the bound 8.85, the second term on
the right hand side of 8.94 is bounded by O(δ 1/2 |u|−1 ) which does not exceed 1
if δ is suitably small depending on D0∞ , R∞0 , D/41 , R
/41 and R
/2 . Moreover, from the
expression 8.86 in the Minkowskian region M0 we deduce the following bound
on S0,u0 : √
/Oi kL∞ (S0,u0 ) ≤ 2
k∇ (8.96)
It follows that the first term on the right hand side of 8.94 does not exceed 2.
We conclude that:
|(∇
/Oi )(q)| ≤ 3
and, as this holds for all q ∈ Su,u0 ,
k∇
/Oi kL∞ (Su,u0 ) ≤ 3 (8.97)
Consider next Oi along C u . Since 8.78 holds, taking X = Oi in 8.87 we
obtain:
D∇/Oi = DΓ / · Oi : on M ′ (8.98)
Substituting in 8.91 we deduce:
/Oi |2 ) ≤ 2|∇
D(|∇ /Oi |(CΩ|χ̂||∇
/Oi | + |DΓ
/||Oi |) (8.99)
Given any point p ∈ Su,u , let q = Φu0 −u (p) ∈ Su,u0 be the point corresponding
to p along the same generator of C u . Applying Lemma 3.1 to 8.99 then yields:
Z u
|(∇
/Oi )(p)| ≤ exp CΩ|χ̂|(Φu′ −u0 (q))du′ |(∇ /Oi )(q)| (8.100)
u0
Z u Z u
+ exp CΩ|χ̂|(Φu′′ −u0 (q))du′′ (|DΓ /||Oi |)(Φu′ −u0 (q))du′
u0 u′
Now by the results of Chapter 3 the integrals in the exponentials in 8.100 are
bounded by O(δ 1/2 |u|−1 ). Therefore if δ is suitably small depending on D0∞ ,
R∞0 these integrals do not exceed (1/2) log 2. Also, by the estimate 7.209,
/kL∞ (Su,u ) ≤ O(δ 1/2 |u|−3 )

kDΓ (8.101)
294
Taking also into account Proposition 8.1 and the bound 8.85, the second term on
the right hand side of 8.100 is bounded by O(δ 1/2 |u|−1 ) which does not exceed
1 if δ is suitably small depending on D0∞ , R∞ /41 , R
0 , D /41 and R/2 . Moreover, in
view of the
√ bound 8.97 the first term on the right hand side of 8.100 does not
exceed 3 2. We conclude that:
√
|(∇/Oi )(p)| ≤ 3 2 + 1
and the proposition is proved.
8.5 L∞ estimates for the deformations tensors of

the Oi
(Oi )
Consider the propagation equations 8.63 and 8.61. Let us decompose π
/ into
its trace-free part (Oi ) π̂
/ and its trace:
(Oi ) (Oi ) 1 (Oi )

π
/= / + g/tr
π̂ π
/ (8.102)
2
In view of 3.5 we have:
(Oi ) (Oi ) (Oi )
tr(D π
/) = Dtr π
/ + 2Ω(χ, π
/)
(Oi ) (Oi ) (Oi )
= Dtr π
/ + Ωtrχtr π
/ + 2Ω(χ̂, π̂
/) (8.103)
and:
(Oi ) (Oi ) (Oi )
tr(D π
/) = Dtr π
/ + 2Ω(χ, π
/)
(Oi ) (Oi ) (Oi )
= Dtr π
/ + Ωtrχtr π
/ + 2Ω(χ̂, π̂
/) (8.104)
Also, since by 8.59

/Oi g/−1 = −
L (Oi ) ♯♯
π
/ , (8.105)
we have:
(Oi ) (Oi )

/Oi (Ωχ̂)) = (
tr (L π̂
/, Ωχ̂), /Oi (Ωχ̂) = (
tr L π̂
/, Ωχ̂) (8.106)
Taking the trace of equation 8.63 and using 8.103 and the first of 8.106 we obtain
the following propagation equation for tr (Oi ) π
/ along Cu0 :
(Oi )
Dtr π
/ = 2Oi (Ωtrχ) (8.107)
Also, taking the trace of equation 8.61 and using 8.104 and the second of 8.106
we obtain the following propagation equation for tr (Oi ) π
/ along the C u :
(Oi )
Dtr π
/ = 2Oi (Ωtrχ) (8.108)
295
Substituting in 8.63 the decomposition 8.102 and taking into account 8.107 we
obtain the following propagation equation for (Oi ) π̂
/ along Cu0 :
(Oi ) (Oi ) (Oi )
D π̂
/ − Ωtrχ π̂
/ = −Ωχ̂tr π /Oi (Ωχ̂)
/ + 2L (8.109)
Also, substituting in 8.61 the decomposition 8.102 and taking into account 8.108
we obtain the following propagation equation for (Oi ) π̂
/ along the C u :
(Oi ) (Oi ) (Oi )

D π̂
/ − Ωtrχ π̂
/ = −Ωχ̂tr π /Oi (Ωχ̂)
/ + 2L (8.110)

(Oi )
ktr /kL∞ (Su,u ) ≤ O(δ|u|−2 )
π
(Oi )
k /kL∞ (Su,u ) ≤ Cδ 1/2 |u|−1 (R
π̂ /41 (β) + R∞
0 (β)) + O(δ|u|
−2
)
provided that δ is suitably small depending on D0∞ , R∞ 0 , D/41 , R
/41 and R
/2 .
Proof: Consider first equation 8.107. Given any point q ∈ Su,u0 , let q0 =
Φ−u (q) ∈ S0,u0 be the point corresponding to q along the same generator of
Cu0 . Integrating 8.107 along this generator we obtain, in view of 8.58,
Z u
(Oi )
(tr π
/)(q) = 2(Oi (Ωtrχ))(Φu′ (q0 ))du′ (8.111)
0
Now by Proposition 6.1 and the results of Chapters 3 and 4 in conjunction with
Lemma 5.2, we have
/(Ωtrχ)kL∞ (Su,u ) ≤ O(|u|−3 )

kd (8.112)
which, together with Proposition 8.1 implies:
kOi (Ωtrχ)kL∞ (Su,u ) ≤ O(|u|−2 ) (8.113)
Substituting this estimate for u = u0 in 8.111 we deduce:

(Oi )
ktr /kL∞ (Su,u0 ) ≤ O(δ|u0 |−2 )
π (8.114)
Consider next equation 8.108. Given any point p ∈ Su,u , let q = Φu0 −u (p) ∈
Su,u0 be the point corresponding to p along the same generator of C u . Integrat-
ing 8.108 along this generator we obtain:
Z u
(tr (Oi ) π
/)(p) = (tr (Oi ) π
/)(q) + 2(Oi (Ωtrχ))(Φu′ −u0 (q))du′ (8.115)
u0
Now by Proposition 6.2 and the results of Chapters 3 and 4 in conjunction with
Lemma 5.2, we have
/(Ωtrχ)kL∞ (Su,u ) ≤ O(δ|u|−4 )

kd (8.116)
296
which, together with Proposition 8.1 implies:
kOi (Ωtrχ)kL∞ (Su,u ) ≤ O(δ|u|−3 ) (8.117)
Substituting this estimate as well as the estimate 8.114 in 8.115 we then deduce:
(Oi )
ktr /kL∞ (Su,u ) ≤ O(δ|u|−2 )
π (8.118)
We turn to equations 8.109, 8.110. By the first part of Lemma 4.2 equation
8.109 implies, along Cu0 ,

D(| (Oi ) π̂ /| CΩ|χ̂|| (Oi ) π̂
/|2 ) ≤ 2| (Oi ) π̂ /Oi (Ωχ̂) − Ωχ̂tr (Oi ) π
/| + 2L / (8.119)
Given again any point q ∈ Su,u0 , let q0 = Φ−u (q) ∈ S0,u0 be the point corre-
sponding to q along the same generator of Cu0 . Applying Lemma 3.1 to 8.119
then yields, in view of 8.58,
(Oi )
|( π̂
/)(q)| ≤ (8.120)
(Z )
Z u u
exp CΩ|χ̂|(Φu′′ (q0 ))du′′ 2L
/Oi (Ωχ̂) − Ωχ̂tr (Oi )
/ (Φu′ (q0 ))du′
π
0 u′
By the results of Chapter 3 the integral in the exponential is bounded by

O(δ 1/2 |u|−1 ). Therefore if δ is suitably small depending on R∞
0 this integral
does not exceed log 2. By Proposition 6.1 in conjunction with Lemma 5.2 we
have:
/(Ωχ̂)kL∞ (Su,u ) + kΩχ̂kL∞ (Su,u ) ≤ Cδ −1/2 |u|−1 (R
|u|k∇ /41 (β) + R∞
0 (β)) + O(|u|
−2
)
(8.121)
Now, for any S-tangential vectorfield X and any 2-covariant S tensorfield θ we
have, relative to an arbitrary local frame field for the Su,u ,
/X θ)AB = X C ∇
(L /A X C + θAC ∇
/C θAB + θCB ∇ /B X C (8.122)
Hence it holds, pointwise,
|L
/X θ| ≤ |X||∇
/θ| + 2|θ||∇
/X| (8.123)
Applying this taking X = Oi , θ = Ωχ̂ we obtain, in view of Propositions 8.1,
8.2 and the estimate 8.121,
/Oi (Ωχ̂)kL∞ (Su,u ) ≤ Cδ −1/2 |u|−1 (R
kL /41 (β) + R∞
0 (β)) + O(|u|
−2
) (8.124)
Substituting this estimate for u = u0 as well as the estimate 8.114 in 8.120 we
then deduce:
(Oi )
k /kL∞ (Su,u0 ) ≤ Cδ 1/2 |u0 |−1 (R
π̂ /41 (β) + R∞
0 (β)) + O(δ|u0 |
−2
) (8.125)
By the second part of Lemma 4.2 equation 8.110 implies, along the C u ,

/|2 ) ≤ 2| (Oi ) π̂
D(| (Oi ) π̂ /| CΩ|χ̂|| (Oi ) π̂ /Oi (Ωχ̂) − Ωχ̂tr (Oi ) π
/| + 2L / (8.126)
297
Given again any point p ∈ Su,u , let q = Φu0 −u (p) ∈ Su,u0 be the point corre-
sponding to p along the same generator of C u . Applying Lemma 3.1 to 8.126
then yields:
Z u
(Oi )
|( π̂
/)(p)| ≤ exp CΩ|χ̂|(Φu′ −u0 (q))du |( (Oi ) π̂
′
/)(p)| (8.127)
u0
Z u Z u
+ exp CΩ|χ̂|(Φu′′ −u0 (q))du′′ ·
u0 u′
(Oi )
/Oi (Ωχ̂) − Ωχ̂tr
· 2L / (Φu′ −u0 (q))du′
π

O(δ 1/2 |u|−1 ). Therefore if δ is suitably small depending on D0∞ , R∞
0 this inte-
gral does not exceed log 2. By Proposition 6.2 in conjunction with Lemma 5.2
we have:
|u|k∇/(Ωχ̂)kL∞ (Su,u ) + kΩχ̂kL∞ (Su,u ) ≤ Cδ 1/2 |u|−2 (R

/41 (β) + R∞
0 (β)) + O(|u|
−3
)
(8.128)
Applying then 8.123 taking X = Oi , θ = Ωχ̂ we obtain, in view of Propositions
8.1, 8.2 and the estimate 8.128,
/Oi (Ωχ̂)kL∞ (Su,u ) ≤ Cδ 1/2 |u|−2 (R

kL /41 (β) + R∞
0 (β)) + O(|u|
−3
) (8.129)
Substituting this estimate as well as the estimates 8.125 and 8.114 in 8.127 we
then deduce:
(Oi )
k /kL∞ (Su,u ) ≤ Cδ 1/2 |u|−1 (R
π̂ /41 (β) + R∞
0 (β)) + O(δ|u|
−2
) (8.130)
and the proof of the proposition is complete.
kZi kL∞ (Su,u ) ≤ O(|u|−1 )

0 , D /41 and R
/2 .
Proof: We consider the propagation equation 8.68 along the C u . We have
D(|Zi |2 ) = (Dg/)(Zi , Zi ) + 2(Zi , DZi )

= 2Ωχ(Zi , Zi ) − 8(Zi , L/Oi (Ω2 ζ ♯ )) (8.131)
hence:

2 1
D(|Zi | ) ≤ 2 /Oi (Ω2 ζ ♯ )|
Ωtrχ + Ω|χ̂| |Zi |2 + 8|Zi ||L (8.132)
2
Given any point p ∈ Su,u , let q = Φu0 −u (p) ∈ Su,u0 be the point corresponding
to p along the same generator of C u . In view of 8.69, applying Lemma 3.1 to
298
8.132 yields:
u Z u
1
Z
′′
|Zi (p)| ≤ exp Ωtrχ + Ω|χ̂| (Φu′′ −u0 (q))du ·
u0 u′ 2
/Oi (Ω2 ζ ♯ ) (Φu′ −u0 (q))du′(8.133)
·4 L

|u|
log + O(δ 1/2 |u|−1 ) (8.134)
|u′ |
and if δ is suitably small depending on D0∞ , R∞

0 the term O(δ
1/2
|u|−1 ) does
not exceed log 2. By Proposition 6.2 in conjunction with Lemma 5.2 we have,
recalling that (see 1.81)
2ζ = η − η,
/(Ω2 ζ)kL∞ (Su,u ) + kΩ2 ζkL∞ (Su,u ) ≤ O(|u|−2 )

|u|k∇ (8.135)
Now, for any pair of S-tangential vectorfields X, Y we have:
|L
/X Y | = |[X, Y ]| = |∇
/X Y − ∇
/Y X| ≤ |X||∇
/Y | + |Y ||∇
/X| (8.136)
Applying this taking X = Oi , Y = Ω2 ζ ♯ we obtain, in view of Propositions 8.1,

8.2 and the estimate 8.135,
/Oi (Ω2 ζ ♯ )kL∞ (Su,u ) ≤ O(|u|−2 )

kL (8.137)
Substituting this estimate as well as the result 8.134 above in 8.133 we then
deduce:
kZi kL∞ (Su,u ) ≤ O(|u|−1 ) (8.138)
and the proof of the proposition is complete.
According to 8.46, 8.47 we have:

(Oi ) (Oi )
n= n=0 (8.139)
while according to 8.45 we have:

(Oi )
m=0 (8.140)
The remaining components of the deformation tensor of the Oi are:

(Oi )
m = Ω−1 g/ · Zi (8.141)
(Oi ) 1 (Oi )
j= tr π
/ − 2Oi (log Ω) (8.142)
2
(see 8.48) and:
(Oi ) (Oi )
î = π̂
/ (8.143)
299
Propositions 8.3 and 8.4 (and the estimate 7.220) then yield the following esti-
mates for these components:
(Oi )
k îkL∞ (Su,u ) ≤ Cδ 1/2 |u|−1 (R
/41 (β) + R∞
0 (β)) + O(δ|u|
−2
)
(Oi )
k jkL∞ (Su,u ) ≤ O(δ|u|−2 )
(Oi )
k mkL∞ (Su,u ) ≤ O(|u|−1 ) (8.144)
300
Chapter 9
Estimates for the

Derivatives of the
Deformation Tensors of the
Commutation Fields
9.1 Estimates for the 1st derivatives of the de-

formation tensors of L, S
The components of the deformation tensor of the commutation field L are given
by table 8.21. From this table and the results of Chapters 3 and 4 we obtain:
/ (L) îkL4 (Su,u ) ≤ Cδ −1/2 |u|−3/2 R
k∇ /41 (α) + O(|u|−5/2 )
(L)
kd
/ jkL4 (Su,u ) ≤ O(|u|−5/2 )
/ (L) mkL4 (Su,u ) ≤ O(δ 1/2 |u|−5/2 )
k∇ (9.1)
Next, from table 8.21, equations 1.76, 3.8, 3.10 and the results of Chapters
3 and 4 we deduce:
(L)
kD̂ îkL4 (Su,u ) ≤ Cδ −3/2 |u|−1/2 R∞
0 (α) + O(δ
−1/2
|u|−5/2 )
(L)
kD jkL4 (Su,u ) ≤ O(δ −1 |u|−3/2 )
(L)
kD mkL4 (Su,u ) ≤ O(δ −1/2 |u|−3/2 ) (9.2)
Next, from table 8.21, equations 1.78, 1.108, 4.4, 4.6 and the results of
Chapters 3 and 4 we deduce:
(L)
kD̂ îkL4 (Su,u ) ≤ Cδ −1/2 |u|−3/2 R∞
0 (α) + O(δ
1/2
|u|−5/2 )
301
(L)
kD jkL4 (Su,u ) ≤ O(|u|−5/2 )
(L)
kD mkL4 (Su,u ) ≤ O(δ 1/2 |u|−5/2 ) (9.3)
The components of the deformation tensor of the commutation field S are
given by table 8.30. Let us define the function:
1
λ= Ω(utrχ + utrχ) − 1 (9.4)
2
(S)
In terms of the function λ the component j is expressed as:
(S)
j = 2(λ − uω − uω) (9.5)
By the results of Chapter 3 we have:
kλkL∞ (Su,u ) ≤ O(δ|u|−2 ) (9.6)
Moreover, by the results of Chapter 4 we have:
/λkL4 (Su,u ) ≤ O(δ|u|−5/2 )
kd (9.7)
Using equations 4.3 and 3.8 we deduce:
1
Dλ = − Ωtrχλ (9.8)
2
1
+ uΩ2 2ρ − (χ̂, χ̂) + 2div
/ η + 2|η|2

2
1
+uΩωtrχ − uΩ2 |χ̂|2
2
Also, using equations 4.4 and 3.9 we deduce:
1
Dλ = − Ωtrχλ (9.9)
2
1
+ uΩ2 2ρ − (χ̂, χ̂) + 2div
/ η + 2|η|2

2
1
+uΩωtrχ − uΩ2 |χ̂|2
2
The results of Chapters 3 and 4 together with 9.6 then yield:
kDλkL4 (Su,u ) ≤ O(|u|−3/2 ) (9.10)
kDλkL4 (Su,u ) ≤ O(δ|u|−5/2 ) (9.11)

From table 8.30, equation 9.5, the results of Chapters 3 and 4 and the
estimate 9.7 we obtain:
/ (S) îkL4 (Su,u ) ≤ Cδ 1/2 |u|−3/2 (D
k∇ /41 (χ̂) + R
/41 (α)) + O(δ|u|−5/2 )
(S)
kd
/ jkL4 (Su,u ) ≤ O(δ|u|−5/2 )
/ (S) mkL4 (Su,u ) ≤ O(δ 3/2 |u|−5/2 )
k∇
/ (S) mkL4 (Su,u ) ≤ O(δ 1/2 |u|−3/2 )
k∇ (9.12)
302
Next, from table 8.30, equation 9.5, together with equations 1.76, 3.8, 3.10,
4.3, 4.5, the results of Chapters 3 and 4 and the estimate 9.10 we deduce:
(S)
kD̂ îkL4 (Su,u ) ≤ Cδ −1/2 |u|−1/2 R∞
0 (α) + O(δ
1/2
|u|−3/2 )
(S)
kD jkL4 (Su,u ) ≤ O(|u|−3/2 )
(S)
kD mkL4 (Su,u ) ≤ O(δ 1/2 |u|−3/2 )
(S)
kD mkL4 (Su,u ) ≤ O(δ −1/2 |u|−1/2 ) (9.13)

Finally, from table 8.30, equation 9.5, together with equations 1.78, 3.9, 3.11,
4.4, 4.6, the results of Chapters 3 and 4 and the estimate 9.11 we deduce:
(S)
kD̂ îkL4 (Su,u ) ≤ Cδ 1/2 |u|−3/2 (D0∞ (χ̂) + R∞
0 (α)) + O(δ
3/2
|u|−5/2 )
(S)
kD jkL4 (Su,u ) ≤ O(δ|u|−5/2 )
(S)
kD mkL4 (Su,u ) ≤ O(δ 3/2 |u|−5/2 )
(S)
kD mkL4 (Su,u ) ≤ O(δ 1/2 |u|−3/2 ) (9.14)

Since the above estimates depend only on the results of Chapters 3 and 4
the symbol O(δ p |u|r ) may be taken here to have the more restricted meaning
of the product of δ p |u|r with a non-negative non-decreasing continuous function
of the quantities D0∞ , R∞ /41 , R
0 , D /41 .
9.2 Estimates for the 1st derivatives of the de-

formation tensors of the Oi
We begin with the following lemma.
Lemma 9.1 Let X be a S-tangential vectorfield and let θ be an arbitrary

type Tpq S tensorfield. We have, with respect to an arbitrary local frame field
for the Su,u ,
p
C ...C C ...C
X
1 q (X) D
(L
/X ∇
/θ − ∇
/L
/X θ)AB 1 ...Bp
= − /1,ABi θ 1 qD
π
i=1 B1 ...>Bi <...Bp
q D
(X) Cj C ...>C <...Cq
X
+ /1,AD θB11 ...Bp j
π
j=1
where (X) π
/1 is the Lie derivative with respect to X of the induced connection
/ on the Su,u , a type T21 S tensorfield, symmetric in the lower indices, given by:
Γ
(X) C 1 (X) C (X) C

π
/1,AB = (∇
/A π
/B +∇
/B π
/A /C
−∇ (X)
π
/AB )
2
303
Proof: We may restrict attention to a given Su,u . Then X is simply a vectorfield
and θ is a type Tpq tensorfield on Su,u . Let X generate the 1-parameter group
/
g
ft of diffeomorphisms of Su,u . Consider first the case q = 0. Then ∇
/θ =∇
/ θ is a
p + 1 covariant tensorfield on Su,u and we have:
/
g ft∗ /
g
ft∗ (∇
/ θ) / ft∗ θ
=∇ (9.15)
and:
d ∗
L
/X (∇
/θ) = f (∇
/θ) (9.16)
dt t t=0
ft∗ /
g
Let (ϑA : A = 1, 2) be local coordinates on Su,u and let Γ /C
AB be the connection
∗
coefficients of ft g/ (Christoffel symbols) in this coordinate system. In terms of
the coordinates (ϑA : A = 1, 2) we have:
p ∗
ft∗ /
g ∂(ft∗ θ)B1 ...Bp X fCt /g
/ ft∗ θ)AB1 ...Bp =
(∇ − /ABi (ft∗ θ)
Γ C (9.17)
∂ϑA i=1
B1 ...>Bi <...Bp
Hence:
∗
d ft /g ∗ ∂ d ∗
(∇
/ ft θ)AB1 ...Bp = (f θ)B1 ...Bp
dt ∂ϑA dt t t=0
t=0
p /
g
X d ∗
− /C
Γ ABi (ft θ) C
i=1
dt B1 ...>Bi <...Bp t=0
p f ∗/
g
X d Ct
− Γ
/ABi θ C (9.18)
i=1
dt B1 ...>Bi <...Bp
t=0
Since
d ∗
(f θ)B1 ...Bp = (L
/X θ)B1 ...Bp
dt t t=0
the first two terms on the right in 9.18 are simply
(∇
/(L
/X θ))AB1 ...Bp (9.19)
Also, since
ft∗ /
g
∂(ft∗ g/)BD ∂(ft∗ g/)AD ∂(ft∗ g/)AB

1 ∗ −1 CD
/C
ΓAB = ((f g/) ) + −
2 t ∂ϑA ∂ϑB ∂ϑD
we obtain:
f ∗/
g /
g /
g
d Ct 1 −1 CD d ∗ d ∗
Γ
/ = (g/ ) ∇
/A (f g/)BD +∇
/B (f g/)AD
dt AB 2 dt t t=0 dt t t=0
t=0
!
/
g
d ∗
−∇
/D (f g/)AB
dt t t=0
304
or, since
d ∗ (X)
(f g/) = π
/
dt t t=0
simply:
f ∗/
g
d Ct 1 −1 CD (X) (X) (X) (X) C
Γ
/ = (g/ ) (∇
/A π
/BD + ∇
/B π
/AD − ∇
/D π
/AB ) = π
/1,AB
dt AB 2
t=0
(9.20)
In view of 9.19, 9.20, and 9.16, the formula 9.18 becomes:
p
X
(L /θ)AB1 ...Bp = (∇
/X ∇ /X θ)AB1 ...Bp −
/L /C
π1,ABi θ C (9.21)
B1 ...>Bi <...Bp
i=1
which is the lemma in the case q = 0.

To obtain the lemma in the general case of a type Tpq tensorfield on Su,u ,
given such a tensorfield θ we apply what we have just proved to the p covariant
tensorfield
θ · (ξ1 , ..., ξq )
where ξ1 , ..., ξq are arbitrary 1-forms on Su,u . The general case then follows
by virtue of the Leibniz rule for Lie derivatives L /X as well as covariant for
derivatives ∇ /, if we also apply what we have just proved to the 1-forms ξ1 , ..., ξq .

(Oi )
kd
/tr /kL4 (Su,u ) ≤ O(δ|u|−5/2 )
π
k∇ /kL4 (Su,u ) ≤ Cδ 1/2 |u|−3/2 (R

/ (Oi ) π̂ /41 (β) + R∞
0 (β)) + O(δ|u|
−5/2
)
0 , D /41 and R
/2 .
Proof: Applying /d to the propagation equation 8.107 we obtain, in view of
Lemma 1.2, noting that for an arbitrary S-tangential vectorfield X and an
arbitrary function f it holds:
/dL
/X f = L
/X /df, (9.22)
(Oi )
the following propagation equation for /dtr / along Cu0 :
π
(Oi )
Dd
/tr π /Oi /d(Ωtrχ)
/ = 2L (9.23)
To this we apply Lemma 4.6 taking p = 4. Here r = 1, ν = 0, γ = 0 and we

obtain:
Z u
/tr (Oi ) π
kd /kL4 (Su,u0 ) ≤ C /Oi /d(Ωtrχ)kL4 (Su′ ,u ) du′
kL
0
0
≤ O(δ|u0 |−5/2 ) (9.24)
305
by Proposition 6.1, Propositions 8.1 and 8.2, and the results of Chapter 4.
Applying /d to the propagation equation 8.108 we obtain, in view of Lemma
1.2 and 9.22, the following propagation equation for /dtr (Oi ) π
/ along the C u :
(Oi )
/tr
Dd π /Oi /d(Ωtrχ)
/ = 2L (9.25)

obtain:
|u|1/2 kd
/tr (Oi )
π
/kL4 (Su,u ) ≤ C|u0 |1/2 kd
/tr (Oi ) π
/kL4 (Su,u0 ) (9.26)
Z u
+C /Oi /d(Ωtrχ)kL4 (Su,u′ ) du′
|u′ |1/2 kL
u0
Substituting the estimate 9.24 and using Proposition 6.2 together with Propo-
sitions 8.1 and 8.2 and the results of Chapter 4 we then deduce:
(Oi )
kd
/tr /kL4 (Su,u ) ≤ O(δ|u|−5/2 )
π (9.27)
Next, we apply ∇
/ to the propagation equation 8.109. Using Lemmas 4.1 and
/ (Oi ) π̂
9.1 we then obtain the following propagation equation for ∇ / along Cu0 :
/ (Oi ) π̂
D∇ /(Oi ) π̂
/ − Ωtrχ∇ / = −DΓ / + /d(Ωtrχ) ⊗ (Oi ) π̂
/ · (Oi ) π̂ /
(Oi )
/Oi ∇
+2L /(Ωχ̂) + 2 π
/1 · Ωχ̂ (9.28)
Here, we denote:
(Oi )
(DΓ
/· π̂ /)D
/)ABC = (DΓ AB
(Oi )
π̂ /)D
/DC + (DΓ AC
(Oi )
π̂
/BD (9.29)
(Oi ) (Oi ) D (Oi ) D
( π
/1 · Ωχ̂)ABC = π
/1,AB Ωχ̂DC + π
/1,AC Ωχ̂BD (9.30)
(Oi )
We decompose /1 into the T21 type S tensorfields, symmetric in the lower
π
indices,
(Oi ) (Oi ) (Oi )
π
/1 = π̂
/1 + tr π
/1 (9.31)
where:
(Oi )
C 1n C C o
π̂
/1,AB = /A (Oi ) π̂
∇ / B +∇ /B (Oi ) π̂
/ A−∇ /C (Oi ) π̂
/AB (9.32)
2
(Oi ) C 1 C
n o
tr π
/1,AB = δB /dA tr (Oi ) π C
/ + δA /dB tr (Oi ) π
/ − g/AB /dC tr (Oi ) π
/ (9.33)
2
/ (Oi ) π̂
Substituting in 9.28 we bring the propagation equation for ∇ / along Cu0 to
the form:
/ (Oi ) π̂
D∇ /(Oi ) π̂
/ − Ωtrχ∇ / (Oi ) π̂
/ = γ·∇ / + 2tr (Oi )
π /Oi ∇
/1 · Ωχ̂ + 2L /(Ωχ̂) + r (9.34)
where:
(Oi ) (Oi )
r = −DΓ
/· π̂
/ + /d(Ωtrχ) ⊗ π̂
/ (9.35)
306
In 9.34,
1 Dn o
/ (Oi ) π̂
(γ · ∇ /)ABC = /A (Oi ) π̂
Ωχ̂C ∇ /BD + ∇ /B (Oi ) π̂
/AD − ∇ /D (Oi ) π̂
/AB
2
1 n o
+ Ωχ̂D ∇
/ A
(Oi )
π̂
/ + ∇
/ C
(Oi )
π̂
/ − ∇
/ D
(Oi )
π̂
/
2 B CD AD AC
thus γ is the type T33 S tensorfield given by:
DEF 1 E D F D F
γABC = Ω χ̂C (δA δB + δB δA ) + χ̂F D E D E
C (δA δB + δB δA )
4
+ χ̂E D F D F F D E D E
B (δA δC + δC δA ) + χ̂B (δA δC + δC δA )
−χ̂D E F F E D E F F E
C (δA δB + δA δB ) − χ̂B (δA δC + δA δC ) (9.36)
To 9.34 we apply Lemma 4.6 taking p = 4. Here r = 3, ν = 2 and γ is given by
9.36. We obtain:
Z u
/ (Oi ) π̂
k∇ /kL4 (Su,u0 ) ≤ C 2tr (Oi ) π /Oi ∇
/1 · Ωχ̂ + 2L /(Ωχ̂) + r 4 du′
0 L (Su′ ,u0 )
(9.37)
Now, by the estimate 9.24:
(Oi )
ktr /1 · Ωχ̂kL4 (Su,u0 ) ≤ O(δ 1/2 |u0 |−7/2 )
π (9.38)
while by Proposition 6.1 together with Propositions 8.1 and 8.2:
/(Ωχ̂)kL4 (Su,u0 ) ≤ Cδ −1/2 |u0 |−3/2 (R
/Oi ∇
kL /41 (β) + R∞
0 (β)) + O(|u0 |
−5/2
) (9.39)
Also, by Proposition 8.3 and the estimates 8.95 and 8.112:
krkL4 (Su,u0 ) ≤ O(|u0 |−5/2 ) (9.40)
Substituting then 9.38 - 9.40 in 9.37 we conclude that:
k∇ /kL4 (Su,u0 ) ≤ Cδ 1/2 |u0 |−3/2 (R

/ (Oi ) π̂ /41 (β) + R∞
0 (β)) + O(δ|u0 |
−5/2
) (9.41)
Applying ∇ / to the propagation equation 8.110 we deduce, following steps

/ (Oi ) π̂
analogous to those leading to 9.34 the following propagation equation for ∇ /
along the C u :
/ (Oi ) π̂
D∇ /(Oi ) π̂
/ − Ωtrχ∇ / (Oi ) π̂
/ = γ·∇ / + 2tr (Oi )
π
/1 · Ωχ̂ + 2L
/Oi ∇
/(Ωχ̂) + r (9.42)
where:
(Oi ) (Oi )
r = −DΓ
/· π̂
/ + /d(Ωtrχ) ⊗ π̂
/ (9.43)
In 9.42 γ is the type T33 S tensorfield given by:
1 n E D F
γ DEF
ABC
= Ω χ̂C (δA δB + δB δA ) + χ̂F
D F
(δ D δ E + δB
C A B
D E
δA )
4
+ χ̂E (δ D δ F + δC
B A C
D F
δA ) + χ̂F (δ D δ E + δC
B A C
D E
δA )
o
−χ̂D (δ E δ F + δA
C A B
δB ) − χ̂D
F E
(δ E δ F + δA
B A C
F E
δC ) (9.44)
307
(this is the conjugate of γ). To 9.42 we apply Lemma 4.7 taking p = 4. Here
r = 3, ν = 2 and γ is given by 9.44. We obtain:
|u|1/2 k∇
/ (Oi ) π̂
/kL4 (Su,u ) ≤ C|u0 |1/2 k∇ / (Oi ) π̂
/kL4 (Su,u0 ) (9.45)
Z u
+C |u′ |1/2 2tr (Oi ) π
/1 · Ωχ̂ + 2L /Oi ∇
/(Ωχ̂) + r du′
u0 Su,u′ )
Now, by the estimate 9.24:

(Oi )
ktr /1 · Ωχ̂kL4 (Su,u ) ≤ O(δ 3/2 |u|−9/2 )
π (9.46)
while by Proposition 6.2 together with Propositions 8.1 and 8.2:
/Oi ∇
kL /(Ωχ̂)kL4 (Su,u ) ≤ Cδ 1/2 |u|−5/2 (R
/41 (β) + R∞
0 (β)) + O(δ|u|
−7/2
) (9.47)
Also, by Proposition 8.3 and the estimates 8.101 and 8.116:
krkL4 (Su,u0 ) ≤ O(δ|u|−7/2 ) (9.48)
Substituting then 9.46 - 9.48 and the estimate 9.41 in 9.45 we conclude that:
/ (Oi ) π̂
k∇ /kL4 (Su,u ) ≤ Cδ 1/2 |u|−3/2 (R
/41 (β) + R∞
0 (β)) + O(δ|u|
−5/2
) (9.49)
To derive an L4 (S) estimate for ∇/Zi , we must apply ∇ / to the propagation

equation 8.68. Here we must apply an analogue of the second part of Lemma 4.1
in the case where θ is an S-tangential vectorfield. In fact, we have the following
extension of Lemma 4.1 to the general case, where θ is an arbitrary type Tpq S
tensorfield.
Lemma 9.2 Let θ be an arbitrary type Tpq S tensorfield on M ′ . We have,

with respect to an arbitrary frame field for the Su,u ,
p
C ...C C1 ...Cq
X
(D∇
/θ − ∇ 1
/Dθ)AB q
1 ...Bp
= − /)D
(DΓ ABi θ D
i=1 B1 ...>Bi <...Bp
q D
C C ...>C <...Cq
X
j
+ (DΓ)AD θB11 ...Bp j
j=1
and:
p
C ...C C1 ...Cq
X
(D∇
/θ − ∇ 1
/Dθ)AB q
1 ...Bp
= − /)D
(DΓ ABi θ D
i=1 B1 ...>Bi <...Bp
q D
C C ...>C <...Cq
X
j
+ (DΓ)AD θB11 ...Bp j
j=1
308
Proof: In the case q = 0 the lemma reduces to Lemma 4.1. Consider now the
case p = 0, q = 1, namely the case that θ is a S-tangential vectorfield. We work,
as in the proof of Lemma 4.1, in a Jacobi field frame. We then have (see 4.8):
θ = θ B eB , Dθ = (LθB )eB (9.50)
and (see 4.11):
(∇ /eA Dθ = (eA (LθB ) + Γ

/Dθ)(eA ) = ∇ /B C
AC (Lθ ))eB (9.51)
On the other hand,

(D∇
/θ)(eA ) = [L, (∇
/θ)(eA )] (9.52)
while:
(∇ /eA θ = (eA (θB ) + Γ
/θ)(eA ) = ∇ /B C
AC θ )eB (9.53)
hence:
[L, (∇
/θ)(eA )] = (L(eA (θB ) + Γ
/B C
AC θ ))eB
= (eA (LθB ) + Γ
/B C
/)B
AC (Lθ ) + (DΓ
C
AC θ )eB (9.54)
Comparing with 9.51 we conclude that:
(D∇
/θ − ∇ /)B
/Dθ)(eA ) = (DΓ C
AC θ eB (9.55)
This is the first part lemma in the case p = 0, q = 1. The second part is
established in the same manner. The proof in the case p = 0, q arbitrary, is
simmilar. Finally, the general case is obtained by combining with the argument
of Lemma 4.1.
Applying ∇/ to equation 8.68 and using Lemmas 9.1 and 9.2 in the case p = 0,
q = 1, we obtain the following propagation equation for ∇
/Zi along the C u :
D∇ /(Ω2 ζ ♯ ) + s
/Oi ∇
/Zi = −4L (9.56)
where:
(Oi )
s = DΓ
/ · Zi + 4 / 1 · Ω2 ζ ♯
π (9.57)
Here:
/ · Zi )B
(DΓ /)B
A = (DΓ
C
AC Zi , ( (Oi )
/1 · Ω2 ζ ♯ )B
π A =
(Oi ) B
/1,AC Ω2 ζ C
π (9.58)
To equation 9.56 we must apply an analogue of Lemma 4.7 in the case where θ
is an T11 type S tensorfield. In fact, Lemma 4.7 can be extended to the general
case, where θ is an arbitrary type Trs S tensorfield. We first extend Lemma 4.2
to general case:
309
Lemma 9.3 Let θ be an arbitrary type Trs type S tensorfield on M ′ . We
have:
p Bi
i A1 ...>Ai <...Ap C1 ...Cq
X
2
D(|θ| ) + (p − q)Ωtrχ|θ| 2
= 2(θ, Dθ) − 2 Ωχ̂A
Bi θC1 ...Cq θA1 ...Ap
i=1
q
D A1 ...Ap C ...C
X
+2 Ωχ̂Cjj θ Dj
θA11 ...Aqp
j=1 C1 ...>Cj <...Cq
and:
p Bi
i A1 ...>Ai <...Ap C1 ...Cq
X
2
D(|θ| ) + (p − q)Ωtrχ|θ| 2
= 2(θ, Dθ) − 2 Ωχ̂A θ
Bi C1 ...Cq
θA1 ...Ap
i=1
q
A1 ...Ap C ...C
X
+2 Ωχ̂D
C
j
θ Dj
θA11 ...Aqp
j
j=1 C1 ...>Cj <...Cq
The proof is straightforward, along the lines of that of Lemma 4.2, starting from
the formula:
C ...C D ...D
|θ|2 = (g/−1 )A1 B1 ...(g/−1 )Ap Bp g/C1 D1 ...g/Cq Dq θA11 ...Aqp θB11...Bpq
Using the second part of Lemma 9.3 in place of the second part of Lemma 4.2
we establish, following the argument of Lemma 4.7, the following more general
lemma.
Lemma 9.4 Let θ be an arbitrary type Trs type S tensorfield on M ′ satisfying

along the generators of C u the propagation equation:
ν
2
where ν is a real number, ξ is a type Trs S tensorfield on M ′ , and γ is a type
r+s
Tr+s S tensorfield on M ′ satisfying, pointwise,
|γ| ≤ mΩ|χ̂|
where m is a positive constant. Then, if δ is suitably small depending on D0∞ ,
R∞0 , for each p ≥ 1 there is a positive constant C depending only on p, r, s, ν, m
such that:
|u|r−s−ν−(2/p) kθkLp (Su,u ) ≤ C|u0 |r−s−ν−(2/p) kθkLp (Su,u0 )
Z u
+C |u′ |r−s−ν−(2/p) kξkLp(Su,u′ ) du′
u0

/Zi kL4 (Su,u ) ≤ O(|u|−3/2 )
k∇
310
0 , D /41 and R
/2 .
Proof: We apply Lemma 9.4 to the propagation equation 9.56 taking p = 4.
Here r = s = 1, ν = 0, γ = 0. Taking also into accound the fact that ∇ /Zi
vanishes on Cu0 we obtain:
Z u
−1/2
|u| k∇
/Zi kL4 (Su,u ) ≤ C |u′ |−1/2 −4L /(Ω2 ζ ♯ ) + s L4 (S ′ ) du′ (9.59)
/Oi ∇
u,u
u0
Now, by Proposition 6.2 together with Propositions 8.1 and 8.2:
/Oi ∇
kL /(Ω2 ζ ♯ )kL4 (Su,u ) ≤ O(|u|−5/2 ) (9.60)
Also, by Proposition 8.4, the estimate 8.101 and Proposition 9.1:
kskL4 (Su,u ) ≤ O(δ 1/2 |u|−7/2 ) (9.61)
Substituting the above in 9.60 the proposition follows.

(Oi )
kDtr /kL4 (Su,u ) ≤ O(|u|−3/2 )
π
(Oi )
kD /kL4 (Su,u ) ≤ Cδ −1/2 |u|−1/2 (R
π̂ /41 (α) + R∞
0 (α)) + O(|u|
−3/2
)
0 , D /41 and R
/2 .
(Oi ) (Oi )
DDtr π
/ = DDtr / + 4Ω2 ζ ♯ · tr
π (Oi )
π
/ (9.62)
On the other hand, applying D to the propagation equation 8.108 we obtain:

Oi )
DDtr π
/ = 2D(Oi (Ωtrχ))
= 2Oi (D(Ωtrχ)) + 2Zi · /d(Ωtrχ) (9.63)
by virtue of 8.64. Substituting 9.63 in 9.62 yields the following propagation

equation for Dtr (Oi ) π
/ along the C u :
(Oi )
DDtr / = 2Oi (D(Ωtrχ)) + 4Ω2 ζ ♯ · /dtr
π (Oi )
π
/ + 2Zi · /d(Ωtrχ) (9.64)

obtain:
|u|−1/2 kDtr (Oi ) π

/kL4 (Su,u ) ≤ C|u0 |−1/2 kDtr (Oi ) π
/kL4 (Su,u0 ) (9.65)
Z u
+C |u′ |−1/2 Oi (D(Ωtrχ)) + 2Ω2 ζ ♯ · /dtr (Oi ) π
/ + Zi · /d(Ωtrχ) du′
u0 L4 (Su,u′ )
Now, by equation 8.107, Proposition 8.1 and the results of Chapters 3 and 4:
(Oi )
kDtr /kL4 (Su,u0 ) ≤ O(|u0 |−3/2 )
π (9.66)
311
By equation 4.3, Proposition 6.2, Proposition 8.1 and the results of Chapters 3
and 4:
kOi (D(Ωtrχ))kL4 (Su,u ) ≤ O(|u|−5/2 ) (9.67)
Also, by the estimate 8.116 and Propositions 8.4 and 9.1:
2Ω2 ζ ♯ · /dtr (Oi )

π
/ + Zi · /d(Ωtrχ) ≤ O(δ|u|−9/2 ) (9.68)
L4 (Su,u )
Substituting 9.66 - 9.68 in 9.65 we conclude that:

(Oi )
kDtr /kL4 (Su,u ) ≤ O(|u|−3/2 )
π (9.69)
By Lemma 1.4 we have:

(Oi ) (Oi ) (Oi )
DD π̂
/ = DD π̂
/+L/4Ω2 ζ ♯ π̂
/ (9.70)

(Oi ) (Oi ) (Oi ) (Oi )
DD π̂
/ − ΩtrχD π̂
/ = D(Ωtrχ) π̂
/ − D(Ωχ̂)tr π
/
(Oi )
−Ωχ̂Dtr π /Oi (Ωχ̂)
/ + 2DL (9.71)
and we have:
/Oi (Ωχ̂) = L
DL /Oi D(Ωχ̂) + L
/Zi (Ωχ̂) (9.72)
by virtue of 8.64. Substituting 9.72 in 9.71 and the result in 9.70 yields the
following propagation equation for D (Oi ) π̂
/ along the C u :
(Oi ) (Oi )
DD π̂
/ − ΩtrχD π̂ /Oi D(Ωχ̂) + v
/ = 2L (9.73)
where:
(Oi ) (Oi ) (Oi ) (Oi )
/Zi (Ωχ̂) + 4L
v = 2L /Ω2 ζ ♯ π̂
/ − Ωχ̂Dtr π
/π
/ + D(Ωtrχ) π̂
/ − D(Ωχ̂)tr
(9.74)
To 9.73 we apply Lemma 4.7 taking p = 4. Here r = 2, ν = 2, γ = 0. We
obtain:
|u|−1/2 kD (Oi )
π̂
/kL4 (Su,u ) ≤ C|u0 |−1/2 kD (Oi ) π̂
/kL4 (Su,u0 ) (9.75)
Z u
+C |u′ |−1/2 2L /Oi D(Ωχ̂) + v L4 (Su,u′ )
du′
u0
Now by equation 8.109, Propositions 8.1, 8.2 and 8.3 and the results of Chapters
3 and 4:
(Oi )
kD /kL4 (Su,u0 ) ≤ Cδ 1/2 |u0 |−1/2 (R
π̂ /41 (α) + R∞
0 (α)) + O(|u0 |
−3/2
) (9.76)
By equation 4.5, Propositions 8.1 and 8.2 and the results of Chapters 3 and 4:
/Oi D(Ωχ̂)kL4 (Su,u ) ≤ Cδ −1/2 |u|−3/2 (R

kL /41 (α) + R∞
0 (α)) + O(|u|
−5/2
) (9.77)
312
Using formula 8.122 with Zi in the role of X and Ωχ̂ in the role of θ, we obtain,
by Propositions 8.4 and 9.2,
/Zi (Ωχ̂)kL4 (Su,u ) ≤ O(δ 1/2 |u|−7/2 )

kL (9.78)
Also, using formula 8.122 with Ω2 ζ ♯ in the role of X and (Oi ) π̂

/ in the role of θ,
we obtain, by Propositions 8.3 and 9.1 and the results of Chapters 3 and 4,
(Oi )
kL
/Ω2 ζ ♯ /kL4 (Su,u ) ≤ O(δ|u|−7/2 )
π̂ (9.79)
The remaining three terms on the right in 9.74 are bounded in L4 (S) by O(δ 1/2 |u|−7/2 )
using the estimate 9.69, Proposition 8.3 and equations 4.3, 4.5. Hence we obtain:
kvkL4 (Su,u ) ≤ O(δ 1/2 |u|−7/2 ) (9.80)
Substituting 9.76, 9.77 and 9.80 in 9.75 we conclude that:

(Oi )
kD /kL4 (Su,u ) ≤ Cδ −1/2 |u|−1/2 (R
π̂ /41 (α) + R∞
0 (α)) + O(|u|
−3/2
) (9.81)
kDZi kL4 (Su,u ) ≤ O(δ −1/2 |u|−1/2 )

0 , D /41 and R
/2 .
Proof: By Lemma 1.4 we have:
DDZi = DDZi + 4[Ω2 ζ ♯ , Zi ] (9.82)
DDZi = /Oi (Ωζ ♯ )

−4DL
= /Oi D(Ω2 ζ ♯ ) − 4[Zi , Ω2 ζ ♯ ]
−4L (9.83)
by virtue of 8.64. Substituting 9.83 in 9.82 yields the following propagation

equation for DZi along the C u :
/Oi D(Ω2 ζ ♯ ) + 8[Ω2 ζ ♯ , Zi ]

DDZi = −4L (9.84)
To this we apply Lemma 9.4 taking p = 4. Here r = 0, s = 1, ν = 0, γ = 0.

Taking also into account the fact that DZi vanishes on Cu0 we obtain:
Z u
−3/2
|u| kDZi kL4 (Su,u ) ≤ C /Oi D(Ω2 ζ ♯ ) + 2[Ω2 ζ ♯ , Zi ] L4 (S ′ ) du′
|u′ |−3/2 −L
u,u
u0
(9.85)
Now, from equation 1.76, by Proposition 6.3 together with Propositions 8.1, 8.2
and the results of Chapters 3 and 4 we obtain:
/Oi D(Ω2 ζ ♯ )kL4 (Su,u ) ≤ O(δ −1/2 |u|−3/2 )

kL (9.86)
313
Also, by Propositions 8.4, 9.2 and the results of Chapters 3 and 4:
k[Ω2 ζ ♯ , Zi ]kL4 (Su,u ) ≤ O(δ 1/2 |u|−7/2 ) (9.87)
Substituting the above in 9.85 the proposition follows.
Estimates for D (Oi ) trπ

/, D (Oi ) π̂
/, and DZi , in L4 (S), follow directly from
equations 8.108, 8.110, and 8.68 respectively. We obtain:
(Oi )
kD /kL4 (Su,u ) ≤ O(δ|u|−5/2 )
trπ
(Oi )
kD /kL4 (Su,u ) ≤ Cδ 1/2 |u|−3/2 (R
π̂ /41 (β) + R∞
0 (β)) + O(δ|u|
−5/2
)
kDZi kL4 (Su,u ) ≤ O(δ 1/2 |u|−3/2 ) (9.88)
In view of the expressions 8.139 - 8.143 for the components of the deformation
tensor of the Oi , Propositions 9.1, 9.2, 9.3, 9.4 and the estimates 9.88 yield the
following estimates for the 1st derivatives of the non-vanishing components in
L4 (S):
/ (Oi ) îkL4 (Su,u ) ≤ Cδ 1/2 |u|−3/2 (R

k∇ /41 (β) + R∞
0 (β)) + O(δ|u|
−5/2
)
(Oi )
kd
/ jkL4 (Su,u ) ≤ O(δ|u|−5/2 )
/ (Oi ) mkL4 (Su,u ) ≤ O(|u|−3/2 )
k∇ (9.89)
(Oi )
kD̂ îkL4 (Su,u ) ≤ Cδ −1/2 |u|−1/2 (R
/41 (α) + R∞
0 (α)) + O(|u|
−3/2
)
(Oi )
kD jkL4 (Su,u ) ≤ O(|u|−3/2 )
(Oi )
kD mkL4 (Su,u ) ≤ O(δ −1/2 |u|−1/2 ) (9.90)
(Oi )
kD̂ îkL4 (Su,u ) ≤ Cδ 1/2 |u|−3/2 (R
/41 (β) + R∞
0 (β)) + O(δ|u|
−5/2
)
(Oi )
kD jkL4 (Su,u ) ≤ O(δ|u|−5/2 )
(Oi )
kD mkL4 (Su,u ) ≤ O(δ 1/2 |u|−3/2 ) (9.91)
In the second of 9.89, to estimate the contribution of the 2nd term on the right
in 8.142 we write:
/Oi /d log Ω
/dOi (log Ω) = L (9.92)
and use Lemma 6.4 together with Propositions 8.1 and 8.2. In the second of
9.90 to estimate the contribution of the 2nd term on the right in 8.142 we write:
DOi (log Ω) = Oi · /dω + Zi · /d log Ω (9.93)
and use Propositions 4.3 and 8.4. In the second of 9.91 to estimate the contri-
bution of the 2nd term on the right in 8.142 we write:
DOi (log Ω) = Oi · /dω (9.94)
314
and use the estimate for /dω in L4 (S) of Proposition 4.2. Finally, since
(Oi )
D m = g/ · DZi + 2Ωχ · Zi
the last of 9.91 requires the estimate
kZi kL4 (Su,u ) ≤ O(δ 1/2 |u|−1/2 ) (9.95)
which follows from the propagation equation 8.68 using the results of Chapter
4.
9.3 Estimates for the 2nd derivatives of the de-

formation tensors of L, S
We now assume, in addition to the previous assumptions on the curvature com-
ponents, that the following quantities are finite:

R40 (D̂α) = sup |u|1/2 δ 5/2 kD̂αkL4 (Su,u )
(u,u)∈D′

R40 (D̂α) = sup |u|5 δ −3/2 kD̂αkL4 (Su,u ) (9.96)
(u,u)∈D′
The second of these quantities has in fact already been introduced (see 7.392).
replacing the supremum on D′ by the supremum on ([0, δ] × {u0 }) D′ , are
T
bounded by a non-negative non-decreasing continuous function of M7 . In the
following we denote by O(δ p |u|r ), for real numbers p, r, the product of δ p |u|r
with a non-negative non-decreasing continuous function of the quantities D0∞ ,
R∞ /41 , R
0 , D /41 , D
/42 (trχ), R /3 (trχ), and R40 (D̂α), R40 (D̂α).
/2 , D
The components of the deformation tensor of the commutation field L are
given by table 8.21. Proposition 6.1 and Propositions 7.1 and 7.4 imply:
/2
k∇ (L)
îkL2 (Cu ) ≤ O(|u|−2 )
/2
k∇ (L)
jkL2 (Cu ) ≤ O(δ 1/2 |u|−3 )
/2
k∇ (L)
mkL2 (Cu ) ≤ O(δ|u|−3 ) (9.97)
for all u ∈ [u0 , c∗ ).

Next, from equations 1.76, 3.8, 3.10, in view of Propositions 7.4, 7.6 and the
results of Chapters 3 and 4, we deduce:
(L)
k∇
/D̂ îkL2 (Cu ) ≤ O(δ −1 |u|−1 )
(L)
kd
/D jkL2 (Cu ) ≤ O(δ −1/2 |u|−2 )
(L)
k∇
/D mkL2 (Cu ) ≤ O(|u|−2 ) (9.98)
for all u ∈ [u0 , c∗ ).
315
Next, from equations 1.78, 1.108, 4.4, 4.6, in view of Proposition 7.1, the
/ 2 ω of Proposition 6.2 and the results of Chapters 3 and 4, we
estimate for ∇
deduce:
(L)
k∇
/D̂ ikL2 (Cu ) ≤ O(|u|−2 )
(L)
kd
/D jkL2 (Cu ) ≤ O(δ 1/2 |u|−3 )
(L)
k∇
/D mkL2 (Cu ) ≤ O(δ|u|−3 ) (9.99)
for all u ∈ [u0 , c∗ ).

Next, applying D to equations 1.76, 3.8, 3.10, and using Propositions 7.6
and 7.8 and the results of Chapters 3 and 4, we deduce:
kD̂2 (L)
îkL2 (Cu ) ≤ O(δ −2 )
kD2 (L)
jkL2 (Cu ) ≤ O(δ −3/2 |u|−1 )
kD2 (L)
mkL2 (Cu ) ≤ O(δ −1 |u|−1 ) (9.100)
for all u ∈ [u0 , c∗ ).

Finally, applying D to equations 1.78, 1.108, 4.4, 4.6, and using equations
1.82, 1.107, 3.8, 3.10, 4.3, 4.5 and the results of Chapters 3 and 4, we deduce:
(L)
kD̂D̂ îkL2 (Cu ) ≤ O(δ −1 |u|−1 )
(L)
kDD jkL2 (Cu ) ≤ O(δ −1/2 |u|−2 )
(L)
kDD mkL2 (Cu ) ≤ O(|u|−2 ) (9.101)
for all u ∈ [u0 , c∗ ). Estimates for the 2nd Lie derivative with respect to L of the
deformation tensor of L shall not be needed.
The components of the deformation tensor of the commutation field S are
given by table 8.30. Moreover the component (S) j is expressed in terms of the
function λ by 9.5. From the definition 9.4, using Propositions 6.1 and 6.2 and
Lemma 6.4, we deduce:
/ 2 λkL2 (Cu ) ≤ O(δ 3/2 |u|−3 )

k∇ (9.102)
for all u ∈ [u0 , c∗ ). From equations 9.8 and 9.9, using Proposition 7.1, the
estimate 9.7 and the results of Chapters 3 and 4, we deduce:
/DλkL2 (Cu ) ≤ O(δ 1/2 |u|−2 )

kd
/DλkL2 (Cu ) ≤ O(δ 3/2 |u|−3 )
kd (9.103)
for all u ∈ [u0 , c∗ ). Moreover, applying D to equation 9.8 and using equations
1.174, 3.8, 3.10, 4.5, the estimate 9.10, Proposition 7.4, as well as the results of
Chapters 3 and 4, we deduce:
kD2 λkL2 (Cu ) ≤ O(δ −1/2 |u|−1 ) (9.104)
316
Also, applying D to equation 9.9 and using equations 1.173, 3.9, 3.11, 4.6, the
/ 2 ω of Proposition 6.2, as well as the results of
estimate 9.11, the estimate for ∇
kD2 λkL2 (Cu ) ≤ O(δ 3/2 |u|−3 ) (9.105)
for all u ∈ [u0 , c∗ ). Finally, applying D to equation 9.9 and using equations
1.82, 1.107, 3.10, 4.3, 4.5, and the results of Chapters 3 and 4, we deduce:
kDDλkL2 (Cu ) ≤ O(δ 1/2 |u|−2 ) (9.106)
for all u ∈ [u0 , c∗ ).

Propositions 6.1, 6.2, 7.4 and the estimate 9.102 imply:
/2
k∇ (S)
îkL2 (Cu ) ≤ O(δ|u|−2 )
/2
k∇ (S)
jkL2 (Cu ) ≤ O(δ 3/2 |u|−3 )
/2
k∇ (S)
mkL2 (Cu ) ≤ O(δ 2 |u|−3 )
/2
k∇ (S)
mkL2 (Cu ) ≤ O(δ|u|−2 ) (9.107)
for all u ∈ [u0 , c∗ ).

Next, from equations 1.76, 1.107, 3.10, 4.5, in view of Propositions 7.1, 7.4,
7.6, the first of the estimates 9.103, as well as the results of Chapters 3 and 4,
we deduce:
(S)
k∇
/D̂ îkL2 (Cu ) ≤ O(|u|−1 )
(S)
kd
/D jkL2 (Cu ) ≤ O(δ 1/2 |u|−2 )
(S)
k∇
/D mkL2 (Cu ) ≤ O(δ|u|−2 )
(S)
k∇
/D mkL2 (Cu ) ≤ O(|u|−1 ) (9.108)
for all u ∈ [u0 , c∗ ).

/ 2ω
Next, from equations 1.78, 1.108, 3.11, 4.6, in view of the estimate for ∇
of Proposition 6.2, Propositions 7.1, 7.7, the second of the estimates 9.103, as
well as the results of Chapters 3 and 4, we deduce:
(S)
k∇
/D̂ îkL2 (Cu ) ≤ O(δ|u|−2 )
(S)
kd
/D jkL2 (Cu ) ≤ O(δ 3/2 |u|−3 )
(S)
k∇
/D mkL2 (Cu ) ≤ O(δ 2 |u|−3 )
(S)
k∇
/D mkL2 (Cu ) ≤ O(δ|u|−2 ) (9.109)
for all u ∈ [u0 , c∗ ).

Next, applying D to equations 1.76, 1.107, 3.10, 4.5, and using equation
1.174, Propositions 7.4, 7.6, 7.8, the estimate 9.104, as well as the results of
kD̂2 (S)
îkL2 (Cu ) ≤ O(δ −1 )
317
kD2 (S)
jkL2 (Cu ) ≤ O(δ −1/2 |u|−1 )
kD2 (S)
mkL2 (Cu ) ≤ O(|u|−1 )
kD2 (S)
mkL2 (Cu ) ≤ O(δ −1 ) (9.110)
for all u ∈ [u0 , c∗ ).

Next, applying D to equations 1.78, 1.108, 3.11, 4.6, and using equation
/ 2 ω of Proposition 6.2, Propositions 7.7, 7.9, the esti-
1.173, the estimate for ∇
mate 9.105, as well as the results of Chapters 3 and 4, we deduce:
2 (S)
kD̂ îkL2 (Cu ) ≤ O(δ|u|−2 )
kD2 (S)
jkL2 (Cu ) ≤ O(δ 3/2 |u|−3 )
kD2 (S)
mkL2 (Cu ) ≤ O(δ 2 |u|−3 )
kD2 (S)
mkL2 (Cu ) ≤ O(δ|u|−2 ) (9.111)
for all u ∈ [u0 , c∗ ).

Finally, applying D to equations 1.78, 1.108, 3.11, 4.6,and using equations
1.82, 1.107, the estimate 9.106, and the results of Chapters 3 and 4 we deduce:
(S)
kD̂D̂ îkL2 (Cu ) ≤ O(|u|−1 )
(S)
kDD jkL2 (Cu ) ≤ O(δ 1/2 |u|−2 )
(S)
kDD mkL2 (Cu ) ≤ O(δ|u|−2 )
(S)
kDD mkL2 (Cu ) ≤ O(|u|−1 ) (9.112)
for all u ∈ [u0 , c∗ ).
9.4 Estimates for the 2nd derivatives of the de-

formation tensors of the Oi
We shall only require estimates for those 2nd derivatives of the deformation
tensors of the Oi for which the last derivative is tangential to the surfaces Su,u .
/ 2 tr
k∇ (Oi )
/kL2 (Cu ) ≤ O(δ 3/2 |u|−3 )
π
/2
k∇ (Oi )
/kL2 (Cu ) ≤ O(δ|u|−2 )
π̂
0 , D /41 and R/2 .
′
Proof: Consider any (u1 , u1 ) ∈ D and fix attention to the parameter subdomain
D1 and the corresponding subdomain M1 of M ′ (see 3.45, 3.46).
We first apply ∇
9.1 we then obtain the following propagation equation for ∇ / 2 tr (Oi ) π
/ along Cu0 :
/ 2 tr
D∇ (Oi )
π / 2 (Ωtrχ) + 2
/Oi ∇
/ = 2L (Oi )
π
/1 · /d(Ωtrχ) − DΓ
/ · /dtr (Oi )
π
/ (9.113)
318
obtain:
/ 2 tr (Oi ) π
k∇ /kL2 (Su,u0 ) ≤ (9.114)
Z u
C 2L/Oi ∇ / 2 (Ωtrχ) + 2 (Oi )
π
/1 · /d(Ωtrχ) − DΓ
/ · /dtr (Oi )
π
/ du′
0 L2 (Su′ ,u0 )
By Proposition 7.2, Lemma 7.8 together with Propositions 8.1 and 8.2:
Z u
/ 2 (Ωtrχ)kL2 (Su′ ,u ) du′ ≤ δ 1/2 kL
/Oi ∇
kL / 2 (Ωtrχ)kL2 (Cu0 ) ≤ O(δ|u0 |−3 )
/Oi ∇
0
0
(9.115)
Also, by Proposition 9.1 and the estimates 8.95 and 8.112,
Z u
2 (Oi ) π / · /dtr (Oi ) π
/1 · /d(Ωtrχ) − DΓ / 2 du′ ≤ O(δ 3/2 |u0 |−4 ) (9.116)
0 L (Su′ ,u0 )
Substituting in 9.114 we conclude that:
/ 2 tr
k∇ (Oi )
/kL2 (Su,u0 ) ≤ O(δ|u0 |−3 ) : ∀u ∈ [0, δ]
π (9.117)
We then apply ∇/ to the propagation equation 9.25. Using Lemmas 4.1 and
/ 2 tr (Oi ) π
9.1 we then obtain the following propagation equation for ∇ / along the
C u:
/ 2 tr
D∇ (Oi )
π / 2 (Ωtrχ) + 2
/Oi ∇
/ = 2L (Oi )
π
/1 · /d(Ωtrχ) − DΓ
/ · /dtr (Oi )
π
/ (9.118)

obtain:
/ 2 tr (Oi ) π
|u|k∇ / 2 trπ
/kL2 (Su,u ) ≤ C|u0 |k∇ /kL2 (Su,u0 ) (9.119)
Z u
+C |u′ | 2L / 2 (Ωtrχ) + 2 (Oi ) π
/Oi ∇ /1 · /d(Ωtrχ) − DΓ
/ · /dtr (Oi )
π
/ du′
u0 L2 (Su,u′ )
/ 2 tr (Oi ) π
|u|k∇ / 2 trπ
/kL2 (C u1 ) ≤ C|u0 |k∇ /kL2 (C u1 ) (9.120)
u u0
Z u
+C |u′ | 2L / 2 (Ωtrχ) + 2 (Oi ) π
/Oi ∇ /1 · /d(Ωtrχ) − DΓ
/ · /dtr (Oi )
π
/ u du′
u0 L2 (Cu′1 )
By Proposition 7.3 and Lemma 7.8 together with Propositions 8.1 and 8.2:
Z u
|u′ |kL / 2 (Ωtrχ)kL2 (C u1 ) du′ ≤ O(δ 3/2 |u|−2 )
/Oi ∇ (9.121)
u0 u′
Also, by Proposition 9.1 and the estimates 8.101 and 8.116,

Z u
|u′ | 2 (Oi ) π / · /dtr (Oi ) π
/1 · /d(Ωtrχ) − DΓ / u
1
du′ ≤ O(δ 2 |u|−3 ) (9.122)
u0 L2 (Cu′ )
319
Substituting in 9.120 and taking also into account the estimate 9.117 we con-
clude that:
/ 2 tr
k∇ (Oi )
/kL2 (Cuu1 ) ≤ O(δ 3/2 |u|−3 ) : ∀u ∈ [u0 , u1 ]
π (9.123)
Next we apply ∇/ to the propagation equation 9.34. Using Lemmas 4.1 and
/ 2 (Oi ) π̂
9.1 we then obtain the following propagation equation for ∇ / along Cu0 :
/2
D∇ (Oi )
π̂ /2
/ − Ωtrχ∇ (Oi )
π̂ /2
/ = γ ·∇ (Oi )
π̂
/ + 2∇
/tr (Oi )
π / 2 (Ωχ̂) + r′
/Oi ∇
/1 · Ωχ̂ + 2L
(9.124)
where:
r′ = −DΓ / (Oi ) π̂
/·∇ / (Oi ) π̂
/ + /d(Ωtrχ) ⊗ ∇ /+2 (Oi )
π
/1 · ∇
/(Ωχ̂)
(Oi ) (Oi )
+∇
/γ · ∇
/ π̂
/ + 2tr π
/1 · ∇
/(Ωχ̂) + ∇
/r (9.125)
To 9.124 we apply Lemma 4.6 taking p = 2. Here r = 4, ν = 2 and we obtain:

Z u
/ 2 (Oi ) π̂
k∇ /kL2 (Su,u0 ) ≤ C /tr (Oi ) π
2∇ / 2 (Ωχ̂) + r′ 2
/Oi ∇
/1 · Ωχ̂ + 2L du′
0 L (Su′ ,u0 )
(9.126)
Now from 9.33 and the estimate 9.117:
Z u
/tr (Oi ) π
k∇ /1 · Ωχ̂kL2 (Su′ ,u ) du′ ≤ O(δ 3/2 |u0 |−4 ) (9.127)
0
0
Z u
/ 2 (Ωχ̂)kL2 (Su′ ,u ) du′ ≤ δ 1/2 kL
/Oi ∇
kL / 2 (Ωχ̂)kL2 (Cu0 )
/Oi ∇
0
0
≤ O(δ 1/2 |u0 |−2 ) (9.128)
Also, using Propositions 6.1 and 9.1 we deduce:

Z u
kr′ kL2 (Su,u0 ) du′ ≤ O(δ|u0 |−3 ) (9.129)
0
/2
k∇ (Oi )
/kL2 (Su,u0 ) ≤ O(δ 1/2 |u0 |−2 )
π̂ (9.130)
We then apply ∇/ to the propagation equation 9.42. Using Lemmas 4.1 and
/ 2 (Oi ) π̂
9.1 we obtain the following propagation equation for ∇ / along the C u :
/2
D∇ (Oi )
π̂ /2
/ − Ωtrχ∇ (Oi )
π̂ /2
/ = γ ·∇ (Oi )
π̂
/ + 2∇
/tr (Oi )
π
/1 · Ωχ̂ + 2L / 2 (Ωχ̂) + r′
/Oi ∇
(9.131)
where:
r′ /·∇
= −DΓ / (Oi ) π̂ / (Oi ) π̂
/ + /d(Ωtrχ) ⊗ ∇ /+2 (Oi )
π
/1 · ∇
/(Ωχ̂)
+∇ / (Oi ) π̂
/γ · ∇ / + 2tr (Oi )
π
/1 · ∇
/(Ωχ̂) + ∇
/r (9.132)
320
To 9.131 we apply lemma 4.7 taking p = 2. Here r = 4, ν = 2 and we obtain:
/2
|u|k∇ (Oi )
π̂ /2
/kL2 (Su,u ) ≤ C|u0 |k∇ (Oi )
π̂
/kL2 (Su,u0 ) (9.133)
Z u
+C |u′ | 2∇
/tr (Oi )
π
/1 · Ωχ̂ + 2L / 2 (Ωχ̂) + r ′
/Oi ∇ du′
u0 L2 (Su,u′ )
/2
|u|k∇ (Oi )
π̂ /2
/kL2 (Cuu1 ) ≤ C|u0 |k∇ (Oi )
π̂
/kL2 (Cuu1 ) (9.134)
0
Z u
+ |u′ | 2∇
/tr (Oi )
π / 2 (Ωχ̂) + r ′
/Oi ∇
/1 · Ωχ̂ + 2L u du′
u0 L2 (Cu′1 )
Now from 9.33 and the estimate 9.123:

Z u
|u′ |k∇ /1 · Ωχ̂kL2 (C u1 ) du′ ≤ O(δ 2 |u|−4 )
/tr (Oi ) π (9.135)
u0 u′
Z u
|u′ |kL / 2 (Ωχ̂)kL2 (C u1 ) du′ ≤ O(δ|u|−1 )
/Oi ∇ (9.136)
u0 u′

Z u
kr′ kL2 (C u1 ) du′ ≤ O(δ 3/2 |u|−2 ) (9.137)
u0 u′
Substituting 9.135 - 9.137 in 9.134 and taking also into account the estimate
9.130 we conclude that:
/2
k∇ (Oi )
/kL2 (Cuu1 ) ≤ O(δ|u|−2 ) : ∀u ∈ [u0 , u1 ]
π̂ (9.138)
Since the right hand sides of the inequalities 9.123 and 9.138 are independent
u
of u1 or u1 and (u1 , u1 ) ∈ D′ is arbitrary the estimates hold with Cu 1 replaced
∗
by Cu , for all u ∈ [u0 , c ) and the proposition is established.
Proposition 9.6 The following estimate holds for all u ∈ [u0 , c∗ ):
/ 2 Zi kL2 (Cu ) ≤ O(δ 1/2 |u|−2 )

k∇

0 , D /41 and R
/2 .
′
Proof: Consider again any (u1 , u1 ) ∈ D and fix attention to the parameter
subdomain D1 and the corresponding subdomain M1 of M ′ . We apply ∇ / to the
propagation equation 9.56. Using Lemmas 9.1 and 9.2 in the case p = q = 1,
/ 2 Zi along the C u :
we obtain the following propagation equation for ∇
/ 2 Zi = −4L
D∇ / 2 (Ω2 ζ ♯ ) + s′
/Oi ∇ (9.139)
321
where s′ is the type T21 S tensorfield given by:
s′A
BC = −(DΓ/)DBC ∇/D ZiA + (DΓ/)A /C ZiD
BD ∇ (9.140)
(Oi ) D 2 ♯ A (Oi ) A 2 D A
−4 π
/1,BC ∇
/D (Ω ζ ) + 4 π
/1,BD ∇
/C (Ω ζ) + ∇
/ B sC
To 9.139 we apply Lemma 9.4 taking p = 2. Here r = 2, s = 1, ν = 0, γ = 0.

Taking also into account the fact that ∇/ 2 Zi vanishes on Cu0 we obtain:
Z u
/ 2 Zi kL2 (Su,u ) ≤ C
k∇ / 2 (Ω2 ζ ♯ ) + s′ L2 (S ′ ) du′
/Oi ∇
−4L (9.141)
u,u
u0

Z u
2
k∇/ Zi kL2 (Cu 1 ) ≤ C
u / 2 (Ω2 ζ ♯ ) + s′ L2 (C u1 ) du′
/Oi ∇
−4L (9.142)
u0 u′
Z u
/Oi ∇
kL / 2 (Ω2 ζ ♯ )kL2 (C u1 ) du′ ≤ O(δ 1/2 |u|−2 ) (9.143)
u0 u′
Also, using Propositions 6.2, 8.4, 9.1, 9.2 and 9.5 we deduce:
Z u
ks′ kL2 (C u1 ) du′ ≤ O(δ 3/2 |u|−3 ) (9.144)
u0 u′
/ 2 Zi kL2 (Cuu1 ) ≤ O(δ 1/2 |u|−2 ) : ∀u ∈ [u0 , u1 ]

k∇ (9.145)
Since the right hand side is independent of u1 or u1 and (u1 , u1 ) ∈ D′ is arbi-

u
trary the estimate holds with Cu 1 replaced by Cu , for all u ∈ [u0 , c∗ ) and the
proposition is established.

(Oi )
kd
/Dtr /kL2 (Cu ) ≤ O(δ 1/2 |u|−2 )
π
(Oi )
k∇
/D /kL2 (Cu ) ≤ O(|u|−1 )
π̂
0 , D /41 and R
/2 .
Proof: Consider again any (u1 , u1 ) ∈ D′ and fix attention to the parameter
subdomain D1 . We apply /d to equation 9.64 to obtain the following propagation
equation for /dDtr (Oi ) π
/ along the C u :
(Oi )
Dd
/Dtr π /Oi /dD(Ωtrχ) + /dh
/ = 2L (9.146)
where h is the function:
h = 4Ω2 ζ ♯ · /dtr (Oi )

π
/ + 2Zi · /d(Ωtrχ) (9.147)
322
To equation 9.146 we apply Lemma 4.7 taking p = 2. here r = 1, ν = 0, γ = 0
and we obtain:
(Oi )
kd
/Dtr π /Dtr (Oi ) π
/kL2 (Su,u ) ≤ Ckd /kL2 (Su,u0 ) (9.148)
Z u
+C /Oi /dD(Ωtrχ) + /dh L2 (S
2L du′
u,u′ )
u0

(Oi )
kd
/Dtr π /Dtr (Oi ) π
/kL2 (Cuu1 ) ≤ Ckd /kL2 (Cuu1 ) (9.149)
0
Z u
+C /Oi /dD(Ωtrχ) + /dh L2 (C u1 ) du′
2L
u0 u′
Now, by equation 8.107 we have on Cu0 :

(Oi )
/dDtr π /Oi /d(Ωtrχ)
/ = 2L (9.150)
Proposition 6.1 together with Propositions 8.1 and 8.2 then imply:
(Oi )
kd
/Dtr /kL2 (Cuu1 ) ≤ O(δ 1/2 |u0 |−2 )
π (9.151)
0
By equation 4.3, Proposition 7.3 together with Propositions 8.1 and 8.2 and the
results of Chapter 6:
Z u
/Oi /dD(Ωtrχ)kL2 (C u1 ) du′ ≤ O(δ 1/2 |u|−2 )
kL (9.152)
u0 u′

Z u
kd/hkL2 (C u1 ) du′ ≤ O(δ 3/2 |u|−4 ) (9.153)
u0 u′

(Oi )
kd
/Dtr /kL2 (Cuu1 ) ≤ O(δ 1/2 |u|−2 ) : ∀u ∈ [u0 , u1 ]
π (9.154)
Next we apply ∇
/D (Oi ) π̂
9.1 we then obtain the following propagation equation for ∇ / along the
C u:
(Oi ) (Oi )
D∇
/D π̂
/ − Ωtrχ∇
/D π̂ /D(Ωχ̂) + v ′
/Oi ∇
/ = 2L (9.155)
where:
v ′ = −DΓ
/·D (Oi )
π̂
/ + /d(Ωtrχ) ⊗ D (Oi )
π̂
/+2 (Oi )
π
/1 · D(Ωχ̂) + ∇
/v (9.156)
To 9.156 we apply Lemma 4.7 taking p = 2. Here r = 3, ν = 2, γ = 0 and we
obtain:
(Oi )
k∇
/D /kL2 (Su,u ) ≤ Ck∇
π̂ /D (Oi ) π̂
/kL2 (Su,u0 ) (9.157)
Z u
+C /D(Ωχ̂) + v ′ L2 (S
/Oi ∇
2L du′
u,u′ )
u0
323
(Oi )
k∇
/D /D (Oi ) π̂
/kL2 (C u1 ) ≤ Ck∇
π̂ /kL2 (C u1 ) (9.158)
u u0
Z u
+C /D(Ωχ̂) + v ′ L2 (C u1 ) du′
/Oi ∇
2L
u0 u′
Now, by equation 8.109 we have on Cu0 :

(Oi )
∇
/D π̂
/ = Ωtrχ∇ / + /d(Ωtrχ) ⊗ (Oi ) π̂
/ (Oi ) π̂ / (9.159)
(Oi ) (Oi ) (Oi )
−d
/tr π
/ ⊗ Ωχ̂ − tr π
/∇ /Oi ∇
/(Ωχ̂) + 2L /(Ωχ̂) + 2 π
/1 · Ωχ̂
Propositions 6.1, 9.1 together with Propositions 8.1 and 8.2 then imply:
(Oi )
k∇
/D /kL2 (Cuu1 ) ≤ O(|u0 |−1 )
π̂ (9.160)
0
By equation 4.5, Proposition 7.3 together with Propositions 8.1 and 8.2 and the
results of Chapter 6:
Z u
/Oi ∇
kL /D(Ωχ̂)kL2 (C u1 ) du′ ≤ O(|u|−1 ) (9.161)
u0 u′
Also, using Propositions 9.1, 9.2, 9.3. 9.5, 9.6 and the estimate 9.154 we deduce:
Z u
kv ′ kL2 (C u1 ) du′ ≤ O(δ 1/2 |u|−2 ) (9.162)
u0 u′

(Oi )
k∇
/D /kL2 (Cuu1 ) ≤ O(|u|−1 ) : ∀u ∈ [u0 , u1 ]
π̂ (9.163)
Since the right hand sides of the inequalities 9.154 and 9.163 are independent
u
of u1 or u1 and (u1 , u1 ) ∈ D′ is arbitrary the estimates hold with Cu 1 replaced
by Cu , for all u ∈ [u0 , c∗ ) and the proposition is established.
Proposition 9.8 The following estimate holds for all u ∈ [u0 , c∗ ):
/DZi kL2 (Cu ) ≤ O(|u|−1 )

k∇

0 , D /41 and R
/2 .
′
Proof: Consider again any (u1 , u1 ) ∈ D and fix attention to the parameter
subdomain D1 . We apply ∇ / to equation 9.84. Using then Lemmas 9.1 and 9.2
yields the following propagation equation for ∇/DZi along the C u :
D∇ /D(Ω2 ζ ♯ ) + w
/Oi ∇
/DZi = −4L (9.164)
where:
(Oi )
w = DΓ
/ · DZi + 4 /1 · D(Ω2 ζ ♯ ) + 8∇
π /[Ω2 ζ ♯ , Zi ] (9.165)
324
To 9.164 we apply Lemma 9.4 taking p = 2. Here r = s = 1, ν = 0, γ = 0.
Taking also into account the fact that ∇ /DZi vanishes on Cu0 we obtain:
Z u
|u|−1 k∇
/DZi kL2 (Su,u ) ≤ C |u′ |−1 −4L /D(Ω2 ζ ♯ ) + w L2 (S ′ ) du′
/Oi ∇
u,u
u0
(9.166)
Z u
|u|−1 k∇
/DZi kL2 (Cuu1 ) ≤ C |u′ |−1 −4L /D(Ω2 ζ ♯ ) + w L2 (C u1 ) du′ (9.167)
/Oi ∇
u0 u′
Now, from equation 1.76, by Proposition 7.5 together with Propositions 8.1, 8.2
and the results of Chapters 6 we obtain:
Z u
|u′ |−1 kL /D(Ω2 ζ ♯ )kL2 (C u1 ) du′ ≤ O(|u|−2 )
/Oi ∇ (9.168)
u0 u′
Also, using Propositions 9.1, 9.4 and 9.6 we deduce:

Z u
|u′ |−1 kwkL2 (C u1 ) du′ ≤ O(δ 1/2 |u|−3 ) (9.169)
u0 u′
/DZi kL2 (C u1 ) ≤ O(|u|−1 ) : ∀u ∈ [u0 , u1 ]

k∇ (9.170)
u
Since the right hand side is independent of u1 or u1 and (u1 , u1 ) ∈ D′ is arbi-

u
trary the estimate holds with Cu 1 replaced by Cu , for all u ∈ [u0 , c∗ ) and the
proposition is established.
Estimates for /dD (Oi ) trπ /D (Oi ) π̂

/, ∇ /DZi , in L2 (Cu ), follow directly
/, and ∇
from equations 8.108, 8.110, and 8.68 respectively. We obtain:
(Oi )
kd
/D /kL2 (Cu ) ≤ O(δ 3/2 |u|−3 )
trπ
(Oi )
k∇
/D /kL2 (Cu ) ≤ O(δ|u|−2 )
π̂
/DZi kL2 (Cu ) ≤ O(δ|u|−2 )
k∇ (9.171)
the last using Proposition 7.1.

In view of the expressions 8.139 - 8.143 for the components of the deformation
tensor of the Oi , Propositions 9.5, 9.6, 9.7, 9.8 and the estimates 9.171 yield
the following estimates in L2 (Cu ) for the 2nd derivatives of the non-vanishing
components of the deformation tensors of the Oi for which the last derivative is
tangential to the surfaces Su,u :
/2
k∇ (Oi )
îkL2 (Cu ) ≤ O(δ|u|−2 )
/2
k∇ (Oi )
jkL2 (Cu ) ≤ O(δ 3/2 |u|−3 )
/2
k∇ (Oi )
mkL2 (Cu ) ≤ O(δ 1/2 |u|−2 ) (9.172)
325
(Oi )
k∇
/D̂ îkL2 (Cu ) ≤ O(|u|−1 )
(Oi )
kd
/D jkL2 (Cu ) ≤ O(δ 1/2 |u|−2 )
(Oi )
k∇
/D mkL2 (Cu ) ≤ O(|u|−1 ) (9.173)
(Oi )
k∇
/D̂ îkL2 (Cu ) ≤ O(δ|u|−2 )
(Oi )
kd
/D jkL2 (Cu ) ≤ O(δ 3/2 |u|−3 )
(Oi )
k∇
/D mkL2 (Cu ) ≤ O(δ|u|−2 ) (9.174)
In the second of 9.172, in estimating the contribution of the 2nd term on the
right in 8.142 we bound (see 9.92)
/ 2 log Ω
/Oi ∇
L
in L2 (Cu ) using Lemma 7.8. In the second of 9.173 in estimating the contribu-
tion of the 2nd term on the right in 8.142 we bound (see 9.93)
/Oi /dω + /d(Zi · /d log Ω)

L
in L2 (Cu ) using Propositions 7.4 and 9.2. In the second of 9.174 to estimate
the contribution of the 2nd term on the right in 8.142 we bound (see 9.93)
/Oi /dω
L
in L2 (Cu ) using the estimate for ∇

/ 2 ω in L4 (S) of Proposition 6.2. Finally, since
(Oi )
∇
/D /DZi + 2∇
m = g/ · ∇ /(Ωχ) · Zi + 2Ωχ · ∇
/Zi
the last of 9.174 requires the estimate:
/Zi kL2 (Cu ) ≤ O(δ|u|−1 )

k∇ (9.175)
which follows from the propagation equation 9.56 using Proposition 7.1.
326
Chapter 10
The Sobolev Inequalities on

the Cu and the C u
10.1 Introduction
In this chapter we shall derive Sobolev inequalities on the Cu and the C u on
the basis of certain bootstrap assumptions. These Sobolev inequalities shall be
applied to obtain bounds for the spacetime curvature components in L∞ and
for their 1st derivatives in L4 (S) in terms of the L2 norms of up to their 2nd
derivatives on the Cu , in the case of the components α, β, ρ, σ, β, the L2 norms
of up to the 2nd derivatives on the C u , in the case of the component α.
In the present chapter, we shall not make use of the estimates on the con-
nection coefficients derived in the previous chapters, for these have relied on
the L∞ bounds for the spacetime curvature components and the L4 (S) bounds
for their 1st derivatives which we are presently to establish. We shall rely in-
stead on certain bootstrap assumptions the most basic of which control the ratio
Area(Su,u )/|u|2 and I(Su,u ), the isoperimetric constant of the surfaces Su,u .
These most basic bootstrap assumptions are:
1
A1.1: δΩ|trχ| ≤ 3 log 2 : in M ′
1
A1.2: δΩ|χ̂| ≤ 6 log 2 : in M ′
Let us denote by Ã(u, u) the ratio:

Area(Su,u )
Ã(u, u) = (10.1)
|u|2
Lemma 10.1 The assumption A1.1 implies:

Ã(u, u)
2−1/3 ≤ ≤ 21/3 : ∀(u, u) ∈ D′
4π
327
Also, the assumptions A1.1, A1.2 together imply:
1
I(Su,u ) ≤ : ∀(u, u) ∈ D′
π
that is, the conclusion of Lemma 5.4 holds.
Proof: We first revisit the proof of Lemma 5.3. Here u is fixed. From formula
5.81, assumption A1.1 implies:
2−1/3 ≤ µ(u) ≤ 21/3 (10.2)
Also, from the inequality 5.94, assumption A1.2 implies:
ν(u) ≤ 21/3 (10.3)
Hence, by the first of 5.96, assumptions A1.1, A1.2 imply:
λ(u) ≥ 2−2/3 (10.4)
Now (see 5.78, 5.101),

Z Z
Area(Su,u ) = dµ/g(u) = µ(u)dµ/g(0)
S0,u S0,u
while Z
Area(S0,u ) = dµ/g(0) = 4π|u|2
S0,u
Area(Su,u )
2−1/3 ≤ ≤ 21/3 (10.5)
4π|u|2
which is the first part of the lemma.

To obtain the second part of the lemma we revisit the proof of Lemma 5.4.
Consider again a domain Uu ⊂ Su,u with C 1 boundary ∂Uu , the image by Φu
of a domain U0 ⊂ S0,u with C 1 boundary ∂U0 . According to 5.99 we have:
Perimeter(∂Uu ) p
≥ inf λ(u) (10.6)
Perimeter(∂U0 ) S0,u
On the other hand, according to 5.103 we have:
Area(Uu )
≤ sup µ(u) (10.7)
Area(U0 ) S0,u
The inequalities 10.6 and 10.7 together imply:
Area(Uu ) supS0,u µ(u) Area(U0 )

2
≤ (10.8)
(Perimeter(∂Uu )) inf S0,u λ(u) (Perimeter(∂U0 ))2
328
Similarly, we obtain:
Area(Uuc ) supS0,u µ(u) Area(U0c )
≤ (10.9)
(Perimeter(∂Uu ))2 inf S0,u λ(u) (Perimeter(∂U0 ))2
Therefore:
min{Area(Uu ), Area(Uuc )} supS0,u µ(u) min{Area(U0 ), Area(U0c )}
≤ (10.10)
(Perimeter(∂Uu ))2 inf S0,u λ(u) (Perimeter(∂U0 ))2
Taking the supremum over all domains Uu with C 1 boundary ∂Uu in Su,u we
conclude that:
supS0,u µ(u)
I(Su,u ) ≤ I(S0,u ) (10.11)
inf S0,u λ(u)
Now, inequalities 10.2 and 10.4 imply:
supS0,u µ(u)
≤2 (10.12)
inf S0,u λ(u)
In view of 10.12 and 5.109 the second part of the lemma follows from 10.11.
Lemma 10.2 Let the basic bootstrap assumptions A1.1, A1.2 hold and let
ξ be an arbitrary p covariant S tensorfield. Suppose that ξ ∈ W12 (Su,u ) for some
(u, u) ∈ D′ . Then:
!( )
9 1
Z Z Z Z
6 4 2 2
|ξ| dµ/g ≤ |ξ| dµ/g |ξ| dµ/g + |∇
/ξ| dµ/g
Su,u π Su,u 18|u|2 Su,u Su,u
Proof: Recall that by Lemma 5.1 ξ ∈ Lq (Su,u ) for every q < ∞. The point
here is to establish the above inequality. Let f be the function:
f = |ξ|3 (10.13)
We apply to f the isoperimetric inequality 5.35 on Su,u :

Z Z Z !2
2 2
f dµ/g ≤ f dµ/g + I(Su,u ) |d
/f |dµ/g (10.14)
Su,u Su,u Su,u
Now, we have:
/df = 3|ξ|(ξ, ∇
/ξ) (10.15)
Thus,
/f | ≤ 3|ξ|2 |∇
|d /ξ| (10.16)
hence:
Z !2 Z ! Z !
4 2
|d
/f |dµ/g ≤9 |ξ| dµ/g |∇
/ξ| dµ/g (10.17)
Su,u Su,u Su,u
329
Also,
!2 !2
1 1
Z Z Z
2
f dµ/g = f dµ/g = |ξ|3 dµ/g
Su,u Area(S u,u ) Su,u Area(Su,u ) Su,u
! Z !
1
Z
4 2
≤ |ξ| dµ/g |ξ| dµ/g (10.18)
Area(Su,u ) Su,u Su,u
Substituting 10.17 and 10.18 in 10.14, recalling the definition 10.1, and noting
that by Lemma 10.1:

2 1
max I(Su,u ), ≤ (10.19)
Ã(u, u) π
yields the inequality of the lemma.
10.2 The Sobolev inequalities on the Cu

Proposition 10.1 Let the basic bootstrap assumptions A1.1, A1.2 hold.
Let ξ be an arbitrary p covariant S tensorfield vanishing on C 0 such that ξ, Dξ,
/ξ belong to L2 (Cu ) for some u ∈ [u0 , c∗ ). Then ξ ∈ L6 (Cu ) and there is a
∇
numerical constant Cp , depending only on p, such that:
n o1/3
1/3
|u|2/3 kξkL6 (Cu ) ≤ Cp kDξkL2 (Cu ) (1/18)kξk2L2(Cu ) + |u|2 k∇
/ξk2L2(Cu )
Also, ξ ∈ L4 (Su,u ) for each u and there is a numerical constant Cp′ , depending
only on p, such that:
n o1/4
1/2
sup |u|1/2 kξkL4 (Su,u ) ≤ Cp′ kDξkL2 (Cu ) (1/18)kξk2L2(Cu ) + |u|2 k∇
/ξk2L2 (Cu )
u
Proof: By the assumptions of the proposition ξ ∈ W12 (Su,u ) for almost all
u ∈ [0, δ) : if u ∈ [u0 , c∗ − δ], for almost all u ∈ [0, c∗ − u) : if u ∈ (c∗ − δ, c∗ ), in
the case c∗ ≥ u0 + δ, for almost all u ∈ [0, c∗ − u) in the case c∗ < u0 + δ (see
5.6, 5.7). Thus, Lemma 10.2 applies to ξ. We multiply the inequality of Lemma
10.2 by |u|4 and integrate with respect to u to obtain:
! Z
9 1
Z Z Z
4 6 2 4 2 2 2
|u| |ξ| ≤ sup |u| |ξ| |ξ| + |u| |∇
/ξ| (10.20)
Cu π u Su,u 18 Cu Cu
Consider now the surface integral (u is fixed):

Z
x(u) = |u|2 |ξ|4 dµ/g (10.21)
Su,u
We have:
dx
Z
|u|2 D(|ξ|4 ) + Ωtrχ|ξ|4 dµ/g

= (10.22)
du Su,u
330
By the first part of Lemma 4.2:
D(|ξ|4 ) + Ωtrχ|ξ|4 = 4|ξ|2 (ξ, Dξ) (10.23)

p
X Bi
−(2p − 1)Ωtrχ|ξ|4 − 4|ξ|2 Ωχ̂A

Bi ξ
i A1 ...>Ai <...Ap
ξA1 ...Ap
i=1
It follows that:
D(|ξ|4 ) + Ωtrχ|ξ|4 ≤ 4|ξ|3 |Dξ| + (|2p − 1|Ω|trχ| + 4pΩ|χ̂|)|ξ|4 (10.24)
Now, by assumptions A1.1, A1.2,

kp 1
|2p − 1|Ω|trχ| + 4pΩ|χ̂| ≤ , where: kp = log 2(|2p − 1| + 2p) (10.25)
δ 3
Substituting 10.25 in 10.24 and the result in 10.22 we obtain the following
ordinary differential inequality for x:
dx kp
≤ x+a (10.26)
du δ
where: Z
a(u) = 4 |u|2 |ξ|3 |Dξ|dµ/g (10.27)
Su,u
We have:
x(0) = 0 (10.28)
since ξ vanishes on C 0 . Integrating 10.26 then yields:
Z u
′
x(u) ≤ ekp (u−u )/δ a(u′ )du′
0
Z u
kp
≤ e a(u′ )du′ (10.29)
0
Since
Z u Z Z 1/2 Z 1/2
a(u′ )du′ ≤ 4 |u|2 |ξ|3 |Dξ| ≤ 4 |u|4 |ξ|6 |Dξ|2
0 Cu Cu Cu
(10.30)
10.29 implies:
Z ! Z 1/2 Z 1/2
2 4
sup |u| |ξ| ≤ 4ekp |u|4 |ξ|6 |Dξ|2 (10.31)
u Su,u Cu Cu
Substituting in 10.20 and cancelling one factor of

Z 1/2
|u|4 |ξ|6
Cu
331
from both sides we then obtain:
1/2 1/2 Z
36ekp
Z Z
1
Z
4 6 2 2 2 2
|u| |ξ| ≤ |Dξ| |ξ| + |u| |∇/ξ|
Cu π Cu 18 Cu Cu
(10.32)
which gives the first part of the proposition. Substituting 10.32 in turn in 10.31
yields:
!
144e2kp
Z Z
1
Z Z
2 4 2 2 2 2
sup |u| |ξ| ≤ |Dξ| |ξ| + |u| |∇/ξ|
u Su,u π Cu 18 Cu Cu
(10.33)
which gives the second part of the proposition.
We now apply the above proposition to the spacetime curvature componets

α, β, ρ, σ, β. We define:

R0 (α) = sup δkαkL2 (Cu )
u∈[u0 ,c∗ )

R0 (D̂α) = sup δ 2 kD̂αkL2 (Cu )
u∈[u0 ,c∗ )

R
/1 (α) = sup δ|u|k∇
/αkL2(Cu ) (10.34)
u∈[u0 ,c∗ )
and:
R[1] (α) = max{R0 (α), R0 (D̂α), R
/1 (α)} (10.35)
We define:

R0 (β) = sup |u|kβkL2 (Cu )
u∈[u0 ,c∗ )

R0 (Dβ) = sup δ|u|kDβkL2 (Cu )
u∈[u0 ,c∗ )
|u|2 k∇

R
/1 (β) = sup /βkL2 (Cu ) (10.36)
u∈[u0 ,c∗ )
and:
R[1] (β) = max{R0 (β), R0 (Dβ), R
/1 (β)} (10.37)
We define:

R0 (ρ) = sup δ −1/2 |u|2 kρkL2 (Cu )
u∈[u0 ,c∗ )

R0 (Dρ) = sup δ 1/2 |u|2 kDρkL2 (Cu )
u∈[u0 ,c∗ )

R
/1 (ρ) = sup δ −1/2 |u|3 kd
/ρkL2 (Cu ) (10.38)
u∈[u0 ,c∗ )
and:
R[1] (ρ) = max{R0 (ρ), R0 (Dρ), R
/1 (ρ)} (10.39)
332
Also:

R0 (σ) = sup δ −1/2 |u|2 kσkL2 (Cu )
u∈[u0 ,c∗ )

R0 (Dσ) = sup δ 1/2 |u|2 kDσkL2 (Cu )
u∈[u0 ,c∗ )

R
/1 (σ) = sup δ −1/2 |u|3 kd
/σkL2 (Cu ) (10.40)
u∈[u0 ,c∗ )
and:
R[1] (σ) = max{R0 (σ), R0 (Dσ), R
/1 (σ)} (10.41)
We define:

R0 (β) = sup δ −3/2 |u|3 kβkL2 (Cu )
u∈[u0 ,c∗ )

R0 (Dβ) = sup δ −1/2 |u|3 kDβkL2 (Cu )
u∈[u0 ,c∗ )

R
/1 (β) = sup δ −3/2 |u|4 k∇
/βkL2 (Cu ) (10.42)
u∈[u0 ,c∗ )
and:
R[1] (β) = max{R0 (β), R0 (Dβ), R
/1 (β)} (10.43)
We apply Proposition 10.1 to ξ = α. Since in regard to the first term on the
right in 10.23 we have, in this case,
(ξ, Dξ) = (ξ, D̂ξ) (10.44)
we may replace Dξ by D̂ξ on the right in 10.24 and 10.27, hence also in the
conclusions of Proposition 10.1. Defining:

R40 (α) = sup δ 3/2 |u|1/2 kαkL4 (Su,u ) (10.45)
(u,u)∈D′
we then conclude that there is a numerical constant C such that:
R40 (α) ≤ CR[1] (α) (10.46)
Next, we apply Proposition 10.1 to ξ = |u|β. Noting that Dξ = |u|Dβ, ∇

/ξ =
|u|∇
/β and defining:

R40 (β) = sup δ 1/2 |u|3/2 kβkL4 (Su,u ) (10.47)
(u,u)∈D′
we conclude that there is a numerical constant C such that:
R40 (β) ≤ CR[1] (β) (10.48)
333
Next, we apply Proposition 10.1 to ξ = |u|2 ρ, |u|2 σ. Noting that, accordingly,
Dξ = |u|2 Dρ, |u|2 Dσ and ∇/ξ = |u|2/dρ, |u|2/dσ, and defining:

R40 (ρ) = sup |u|5/2 kρkL4 (Su,u )
(u,u)∈D′

R40 (σ) = sup |u|5/2 kσkL4 (Su,u ) (10.49)
(u,u)∈D′
R40 (ρ) ≤ CR[1] (ρ)

R40 (σ) ≤ CR[1] (σ) (10.50)
Finally, we apply Proposition 10.1 to ξ = |u|3 β. Noting that Dξ = |u|3 Dβ,

/ξ = |u|3 ∇
∇ /β and defining:

R40 (β) = sup δ −1 |u|7/2 kβkL4 (Su,u ) (10.51)
(u,u)∈D′
R40 (β) ≤ CR[1] (β) (10.52)
We proceed to the 2nd order Sobolev inequalities on the Cu . If θ is an arbi-

trary p covariant S tensorfield vanishing on C 0 , to bound supu k∇ /θkL4 (Su,u )
we apply Proposition 10.1 to the p + 1 covariant S tensorfield ξ = ∇
/θ which also
vanishes on C 0 . In the resulting inequalities the quantity kD∇
/θkL2 (Cu ) appears
on the right hand side. By the first part of Lemma 4.1 we have, pointwise,
|D∇
/θ| ≤ |∇
/Dθ| + p|DΓ
/||θ| (10.53)
Taking the L2 norm of 10.53 on Su,u we obtain:
/θkL2 (Su,u ) ≤ k∇
kD∇ /DθkL2 (Su,u ) + pkDΓ
/kL4 (Su,u ) kθkL4 (Su,u ) (10.54)
Taking in turn the L2 norm of 10.54 with respect to u (on [0, δ) if u ∈ [u0 , c∗ −δ],
on [0, c∗ − u) if u ∈ (c∗ − δ, c∗ ), in the case c∗ ≥ u0 + δ, on [0, c∗ − u), in the
case c∗ < u0 + δ) we deduce:

kD∇ /DθkL2 (Cu ) + pδ 1/2 sup kDΓ
/θkL2 (Cu ) ≤ k∇ /kL4 (Su,u ) sup kθkL4 (Su,u )
u u
(10.55)
To control the factor supu kDΓ
/kL4 (Su,u ) we introduce the following bootstrap
assumptions:
/(Ωtrχ)kL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′

A2.1: δ|u|1/2 kd
334
1
A2.2: δ|u|1/2 k∇
/(Ωχ̂)kL4 (Su,u ) ≤ 2 : ∀(u, u) ∈ D′
From the formula for DΓ/ of Lemma 4.1, the assumptions A2.1, A2.2 imply:

sup δ|u|1/2 kDΓ
/kL4(Su,u ) ≤ 3 (10.56)
(u,u)∈D′
Substituting then 10.56 in 10.55 yields:

kD∇ /DθkL2 (Cu ) + 3pδ −1/2 |u|−1/2 sup kθkL4 (Su,u )
/θkL2 (Cu ) ≤ k∇ (10.57)
u

and the factor supu kθkL4 (Su,u ) is bounded by the second conclusion of Propo-
sition 10.1 with θ in the role of ξ.
Also, to bound supu kDθkL4 (Su,u ) , assuming that Dθ vanishes on C 0 , we
simply apply Proposition 10.1 with Dθ in the role of ξ.
We now apply the above proceedure to the spacetime curvature components
α, β, ρ, σ, β. We define:
δ|u|2 k∇
/ 2 αkL2 (Cu )

R
/2 (α) = sup
u∈[u0 ,c∗ )

R
/1 (D̂α) = sup δ 2 |u|k∇
/D̂αkL2 (Cu )
u∈[u0 ,c∗ )

R0 (D̂2 α) = sup δ 3 kD̂2 αkL2 (Cu ) (10.58)
u∈[u0 ,c∗ )
and:
R[2] (α) = max{R[1] (α), R /1 (D̂α), R0 (D̂2 α)}
/2 (α), R (10.59)
We define:
|u|3 k∇
/ 2 βkL2 (Cu )

R
/2 (β) = sup
u∈[u0 ,c∗ )
δ|u|2 k∇

R
/1 (Dβ) = sup /DβkL2 (Cu )
u∈[u0 ,c∗ )
2
δ 2 |u|kD2 βkL2 (Cu )

R0 (D β) = sup (10.60)
u∈[u0 ,c∗ )
and:
R[2] (β) = max{R[1] (β), R /1 (Dβ), R0 (D2 β)}
/2 (β), R (10.61)
We define:

R
/2 (ρ) = sup δ −1/2 |u|4 k∇
/ 2 ρkL2 (Cu )
u∈[u0 ,c∗ )

R
/1 (Dρ) = sup δ 1/2 |u|3 k∇
/DρkL2 (Cu )
u∈[u0 ,c∗ )

R0 (D2 ρ) = sup δ 3/2 |u|2 kD2 ρkL2 (Cu ) (10.62)
u∈[u0 ,c∗ )
335
and:
R[2] (ρ) = max{R[1] (ρ), R /1 (Dρ), R0 (D2 ρ)}
/2 (ρ), R (10.63)
Also:

R
/2 (σ) = sup δ −1/2 |u|4 k∇
/ 2 σkL2 (Cu )
u∈[u0 ,c∗ )

R
/1 (Dσ) = sup δ 1/2 |u|3 k∇
/DσkL2 (Cu )
u∈[u0 ,c∗ )

R0 (D2 σ) = sup δ 3/2 |u|2 kD2 σkL2 (Cu ) (10.64)
u∈[u0 ,c∗ )
and:
R[2] (σ) = max{R[1] (σ), R /1 (Dσ), R0 (D2 σ)}
/2 (σ), R (10.65)
Finally, we define:

R
/2 (β) = sup δ −3/2 |u|5 k∇
/ 2 βkL2 (Cu )
u∈[u0 ,c∗ )

R
/1 (Dβ) = sup δ −1/2 |u|4 k∇
/DβkL2 (Cu )
u∈[u0 ,c∗ )
o
R0 (D2 β) = sup δ 1/2 |u|3 kD2 βkL2 (Cu ) (10.66)
u∈[u0 ,c∗ )
and:
R[2] (β) = max{R[1] (β), R /1 (Dβ), R0 (D2 β)}
/2 (β), R (10.67)
The first of each of the definitions 10.58, 10.60, 10.62, 10.64, 10.66 has already
been introduced in the previous chapters, the first of 10.58 by 5.5, the first of
10.60, 10.62, 10.64, 10.66 by 7.1. Also, the second and third of each of the
definitions 10.62, 10.64 has already been introduced by 7.2.
We first apply the proceedure outlined above to bound supu k∇ /θkL4 (Su,u )
with α, |u|β, |u|2 ρ, |u|2 σ, |u|3 β in the role of θ. In terms of the first five of the
definitions 4.1, that is:

R/41 (α) = sup δ 3/2 |u|3/2 k∇
/αkL4 (Su,u )
(u,u)∈D′

/41 (β) =
R sup δ 1/2 |u|5/2 k∇
/βkL4 (Su,u )
(u,u)∈D′

/41 (ρ) =
R sup |u|7/2 kd
/ρkL4 (Su,u )
(u,u)∈D′

/41 (σ) =
R sup |u|7/2 kd
/σkL4 (Su,u )
(u,u)∈D′

/41 (β) =
R sup δ −1 |u|9/2 k∇
/βkL4 (Su,u ) (10.68)
(u,u)∈D′
336
we deduce that there are numerical constants C such that:
/41 (α) ≤ CR[2] (α)
R
/41 (β) ≤ CR[2] (β)
R
/41 (ρ) ≤ CR[2] (ρ)
R
/41 (σ) ≤ CR[2] (σ)
R
/41 (β) ≤ CR[2] (β)
R (10.69)
In deducing the first of the inequalities 10.69 we remark that with ∇ /α in the
role of ξ in Proposition 10.1, we may replace Dξ by D̂ξ, the trace-free, relative
to the last two entries, part of Dξ, and make use of the fact that:
1
[D̂, ∇
/]α = [D, ∇ /]α + (d/(trDα) − trD∇ /α) ⊗ g/
2
= [D, ∇ /]α + (α, ∇
/(Ωχ̂)) ⊗ g/ (10.70)
and the last term on the right is bounded in L2 (Cu ) by:

Cδ −1/2 |u|−1/2 sup kαkL2 (Su,u )
u
(compare with 10.57) using assumption A2.2.

We next apply the proceedure to bound supu kDθkL4 (Su,u ) with |u|β,
|u|2 ρ, |u|2 σ, |u|3 β in the role of θ. The same proceedure applies to bound

supu kD̂αkL4 (Su,u ) , remarking again that with D̂α in the role of ξ in Poposi-
tion 10.1, we may replace Dξ by D̂ξ. In terms of the definitions:

R40 (D̂α) = sup δ 5/2 |u|1/2 kD̂αkL4 (Su,u )
(u,u)∈D′

R40 (Dβ) = sup δ 3/2 |u|3/2 kDβkL4 (Su,u )
(u,u)∈D′

R40 (Dρ) = sup δ|u|5/2 kDρkL4 (Su,u )
(u,u)∈D′

R40 (Dσ) = sup δ|u|5/2 kDσkL4 (Su,u )
(u,u)∈D′

R40 (Dβ) = sup |u|7/2 kDβkL4 (Su,u ) (10.71)
(u,u)∈D′
(the third and fourth of which have already been introduced by 4.2) we deduce
that there are numerical constants C such that:
R40 (D̂α) ≤ CR[2] (α)
R40 (Dβ) ≤ CR[2] (β)
R40 (Dρ) ≤ CR[2] (ρ)
R40 (Dσ) ≤ CR[2] (σ)
R40 (Dβ) ≤ CR[2] (β) (10.72)
337
Consider finally the first of the definitions 10.68 in conjunction with defi-
nition 10.45, the second of the definitions 10.68 in conjunction with definition
10.47, the third and fourth of the definitions 10.68 in conjunction with ther first
and second of the definitions 10.49 respectively, and the firth of the definitions
10.68 in conjunction with definition 10.51. Then in view of Lemma 10.1, apply-
ing Lemma 5.2 taking p = 4 and recalling the first five of the definitions 3.4,
that is:
R∞
0 (α) = sup(δ
3/2
|u||α|)
M′
R∞
0 (β) = sup(δ 1/2 |u|2 |β|)
M′
R∞
0 (ρ) = sup(|u|3 |ρ|)
M′
R∞
0 (σ) = sup(|u|3 |σ)
M′
R∞
0 (β) = sup(δ −1 |u|4 |β|) (10.73)
M′
we deduce that there are numerical constants C such that:
R∞ /41 (α), R40 (α)}

0 (α) ≤ C max{R
R∞ /41 (β), R40 (β)}
0 (β) ≤ C max{R
R∞ /41 (ρ), R40 (ρ)}
0 (ρ) ≤ C max{R
R∞ /41 (σ), R40 (σ)}
0 (σ) ≤ C max{R
R∞ /41 (β), R40 (β)}
0 (β) ≤ C max{R (10.74)
Therefore, combining with the inequalities 10.69 and 10.46, 10.48, 10.50, 10.52,
we conclude that there are numerical constants C such that:
R∞
0 (α) ≤ CR[2] (α)
R∞
0 (β) ≤ CR[2] (β)
R∞
0 (ρ) ≤ CR[2] (ρ)
R∞
0 (σ) ≤ CR[2] (σ)
∞
R0 (β) ≤ CR[2] (β) (10.75)
10.3 The Sobolev inequalities on the C u

On C u we have the measure:
dµ/g du
Thus, for any S tensorfield θ and any p < ∞ we have:
!
Z Z c∗ −u Z
kθkpLp(C ) = |θ|p = |θ|p dµ/g du (10.76)
u
Cu u0 Su,u
338
We introduce the following bootstrap assumptions:
2 log 2 1
A3.1: Ωtrχ + |u| ≤ 3 |u|3/2 : in M ′
log 2 1
A3.2: Ω|χ̂| ≤ 6 |u|3/2 : in M ′
Proposition 10.2 Let the basic bootstrap assumptions A1.1, A1.2, as

well as the bootstrap assumptions A3.1, A3.2 hold. Let ξ be an arbitrary
p covariant S tensorfield such that for some u ∈ [0, δ) |u|q−1 ξ, |u|q Dξ, |u|q ∇
/ξ
belong to L2 (C u ) for some constant
1
q ≥p−
2
Moreover, let ξ ∈ L4 (Su,u0 ). Then |u|q ξ ∈ L6 (C u ) and there is a numerical
constant Cp , depending only on p such that:
k|u|q ξkL6 (C u ) ≤
n o1/6
Cp (1/18)k|u|q−1ξk2L2 (C ) + k|u|q ∇/ξk2L2 (C ) ·
u u
2/3
· |u0 |q kξkL4 (Su,u0 ) +
1/3 n o1/6
k|u|DξkL2 (C u ) (1/18)k|u|q−1 ξk2L2 (C q
) + k|u| ∇/ξk2L2 (C )
u u
Also, ξ ∈ L4 (Su,u ) for each u and there is a numerical constant Cp′ depending
only on p such that:

sup |u|q kξkL4 (Su,u )
u

1/2
≤ Cp′ |u0 |q kξkL4 (Su,u0 ) + k|u|q DξkL2 (C ) ·
u
n o1/4
· (1/18)k|u|q−1ξk2L2 (C ) + k|u|q ∇
/ξk2L2 (C )
u u
Proof: By the assumptions of the proposition ξ ∈ W12 (Su,u ) for almost all
u ∈ [u0 , c∗ − u). Thus Lemma 10.2 applies to ξ. We multiply the inequality of
Lemma 10.2 by |u|6q and integrate with respect to u to obtain:
!( Z )
9 1
Z Z Z
6q 6 4q 4 2q−2 2 2q 2
|u| |ξ| ≤ sup |u| |ξ| |u| |ξ| + |u| |∇/ξ|
Cu π u Su,u 18 C u Cu
(10.77)
Consider now the surface integral (u is fixed):
Z
x(u) = |u|4q |ξ|4 dµ/g (10.78)
Su,u
339
We have:

dx 4q
Z
4q 4 4
= |u| D(|ξ| ) + Ωtrχ − |ξ| dµ/g (10.79)
du Su,u |u|
By the second part of Lemma 4.2:
D(|ξ|4 ) + Ωtrχ|ξ|4 = 4|ξ|2 (ξ, Dξ) (10.80)

p
X Bi
−(2p − 1)Ωtrχ|ξ|4 − 4|ξ|2 Ωχ̂A

B
i
ξ A1 ...>Ai <...Ap
ξA1 ...Ap
i
i=1
It follows that:

4 4q
D(|ξ| ) + Ωtrχ − |ξ|4 ≤ 4|ξ|3 |Dξ| + ν|ξ|4 (10.81)
|u|
where:
4q
ν = −(2p − 1)Ωtrχ − + 4pΩ|χ̂| (10.82)
|u|
Now, by virtue of assumptions A3.1, A3.2 we have:
4(p − q) − 2 2
ν ≤ + |2p − 1| Ωtrχ + + 4pΩ|χ̂|
|u| |u|
kp
≤ (10.83)
|u|3/2
Substituting 10.83 in 10.81 and the result in 10.79 we obtain the following
ordinary differential inequality for x:
dx kp
≤ x+a (10.84)
du |u|3/2
where: Z
a=4 |u|4q |ξ|3 |Dξ|dµ/g (10.85)
Su,u
Integrating 10.84 from u0 we obtain:

Z u
kp ′
x(u) ≤ exp ′ 3/2
du x(u0 ) (10.86)
u0 |u |
Z u Z u
kp
+ exp ′′ 3/2
du a(u′ )du′
′′
u0 u′ |u |
which, since the integrals in the exponentials are bounded by 2kp , implies:
Z u
2kp ′ ′
x(u) ≤ e x(u0 ) + a(u )du (10.87)
u0
340
Since
Z u Z Z !1/2 Z !1/2
a(u′ )du′ ≤ 4 4q 3
|u| |ξ| |Dξ| ≤ 4 6q
|u| |ξ|6 2 2
|u| |Dξ|
u0 Cu Cu Cu
(10.88)
10.87 implies:
sup x(u) ≤ e2kp x(u0 ) + A1/2 y1/2 (10.89)
u
where: Z
y= |u|6q |ξ|6 (10.90)
Cu
and: Z
A = 16 |u|2q |Dξ|2 (10.91)
Cu
Substituting 10.89 in 10.77 we obtain, with:

1
Z Z
2(q−1) 2
B= |u| |ξ| + |u|2q |∇
/ξ|2 (10.92)
18 C u Cu
the inequality:
9e2kp
y ≤ mp x(u0 ) + A1/2 y1/2 B where: mp = (10.93)
π
This implies:
y ≤ mp B (2x(u0 ) + mp AB) (10.94)
which yields the first conclusion of the proposition. Moreover, 10.94 implies:
A1/2 y 1/2 ≤ (2mp ABx(u0 ))1/2 + mp AB
1
≤ x(u0 ) + 2mp AB (10.95)
2
Substituting then 10.95 in 10.89 we obtain:

3
sup x(u) ≤ e2kp x(u0 ) + 2mp AB (10.96)
u 2
which yields the second conclusion of the proposition.
We now apply the above proposition to the spacetime curvature component

α. We define:

R0 (α) = sup δ −3/2 k|u|3 αkL2 (C u )
u∈[0,δ)

R0 (D̂α) = sup δ −3/2 k|u|4 D̂αkL2 (C u )
u∈[0,δ)

/ (α) = sup
R δ −3/2 k|u|4 ∇
/αkL2 (C u ) (10.97)
1
u∈[0,δ)
341
and:
R[1] (α) = max{R0 (α), R0 (D̂α), R
/1 (α)} (10.98)
We also define:

D0′4 (α) = sup δ −3/2 |u0 |4 kαkL4 (Su,u0 ) (10.99)
u∈[0,δ)
Note that here, as in the case of the initial data quantities D0∞ , D /41 , D
/42 (trχ),
and D /3 (trχ) we are considering the whole of Cu0 , not only the part lying in M ′ .
For c∗ ≥ u0 + δ this is the same, but not for c∗ < u0 + δ. In the latter case the
part of Cu0 lying in M ′ corresponds to u < c∗ − u0 . By the results of Chapter 2
D0′4 (α) is bounded by a non-negative non-decreasing continuous function of M5 .
We apply Proposition 10.2 to ξ = α, taking q = 4. Since in regard to the
first term on the right in 10.78 we have, in this case,
(ξ, Dξ) = (ξ, D̂ξ)
we may replace Dξ by D̂ξ on the right in 10.81 and 10.85, hence also in the
conclusions of Proposition 10.2. Defining:

R40 (α) = sup δ −3/2 |u|4 kαkL4 (Su,u ) (10.100)
(u,u)∈D′
we then conclude that there is a numerical constant C such that:
R40 (α) ≤ C max{D0′4 (α), R[1] (α)} (10.101)
We proceed to the 2nd order Sobolev inequalities on the C u . If θ is an arbi-

trary p covariant S tensorfield defined on some C u , to bound supu |u|q+1 k∇ /θkL4 (Su,u )
we apply Proposition 10.2 to the p + 1 covariant tensorfield ξ = ∇ /θ, replacing q
by q + 1. In the resulting inequalities the quantity k|u|q+1 D∇ /θkL2 (C u ) appears
on the right hand side. By the second part of Lemma 4.1 we have, pointwise,
/θ| ≤ |∇
|D∇ /Dθ| + p|DΓ
/||θ| (10.102)
Taking the L2 norm of 10.102 on Su,u we obtain:
/θkL2 (Su,u ) ≤ k∇
kD∇ /DθkL2 (Su,u ) + pkDΓ
/kL4 (Su,u ) kθkL4 (Su,u ) (10.103)
Multiplying by |u|q+1 and taking the L2 norm with respect to |u| on [u0 , c∗ − u)
we then deduce:
k|u|q+1 D∇/θkL2 (C u ) ≤ k|u|q+1 ∇ /DθkL2 (C u ) (10.104)

Z c∗ −u !1/2

+p |u|2 kDΓ /k2L4(Su,u ) du sup |u|q kθkL4 (Su,u )
u0 u
342
To control the first factor in the second term on the right we introduce the
following bootstrap assumptions:
A4.1: |u|2 kd
/(Ωtrχ)kL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
1
A4.2: |u|2 k∇
/(Ωχ̂)kL4 (Su,u ) ≤ 2 : ∀(u, u) ∈ D′
From the formula for DΓ

/ of Lemma 4.1, the assumptions A4.2, A4.2 imply:
!1/2
Z c∗ −u
2
|u| /k2L4(Su,u ) du
kDΓ ≤3 (10.105)
u0
Substituting then 10.105 in 10.104 yields:

/DθkL2 (C u ) + 3p sup |u|q kθkL4 (Su,u )
/θkL2 (C u ) ≤ k|u|q+1 ∇
k|u|q+1 D∇ (10.106)
u

and the factor supu |u|q kθkL4 (Su,u ) is bounded by the second conclusion of
Proposition 10.2 with θ in the role of ξ.
Also, to bound supu |u|q+1 kDθkL4 (Su,u ) we simply apply Proposition 10.2
with Dθ in the role of ξ, replacing q by q + 1.
We now apply the above procedure to the spacetime curvature component
α. We define:

/ (α) = sup δ −3/2 k|u|5 ∇
R / 2 αkL2 (C u )
2
u∈[0,δ)

/1 (D̂α) = sup
R δ −3/2 k|u|5 ∇
/D̂αkL2 (C u )
u∈[0,δ)
2 2

R0 (D̂ α) = sup δ −3/2 k|u|5 D̂ αkL2 (C u ) (10.107)
u∈[0,δ)
and:
2
R[2] (α) = max{R[1] (α), R
/2 (α), R
/1 (D̂α), R0 (D̂ α)} (10.108)
We also define:

/1′4 (α) = sup
D δ −3/2 |u0 |5 k∇
/αkL4 (Su,u0 )
u∈[0,δ)

D0′4 (D̂α) = sup δ −3/2 |u0 |5 kD̂αkL4 (Su,u0 ) (10.109)
u∈[0,δ)
and:
′4
D[1] (α) = max{D0′4 (α), D
/1′4 (α), D0′4 (D̂α)} (10.110)
Note that here we are again considering the whole of Cu0 , not only the part
lying in M ′ . For c∗ ≥ u0 + δ this is the same, but not for c∗ < u0 + δ. In the
latter case the part of Cu0 lying in M ′ corresponds to u < c∗ −u0 . By the results
343
′4
of Chapter 2 D[1] (α) is bounded by a non-negative non-decreasing continuous
function of M7 .
We first apply the proceedure outlined above to bound supu |u|q+1 k∇ /θkL4 (Su,u )
with α in the role of θ, taking q = 4. In terms of the fifth of the definitions 4.1,
that is:
R/41 (α) = sup δ −3/2 |u|−5 k∇
/αkL4 (Su,u ) (10.111)
(u,u)∈D′
we deduce that there is a numerical constant C such that:

/41 (α) ≤ C max{D[1]
R ′4
(α), R[2] (α)} (10.112)
where In deducing 10.112 we remark that with ∇/α in the role of ξ in Proposition
10.2, we may replace Dξ by D̂ξ, the trace-free, relative to the last two entries,
part of Dξ, and make use of the fact that:
1
[D̂, ∇
/]α = [D, ∇
/]α + (d/(trDα) − trD∇ /α) ⊗ g/
2
= [D, ∇
/]α + (α, ∇
/(Ωχ̂)) ⊗ g/ (10.113)
and the last term on the right, multiplied by |u|5 , is bounded in L2 (C u ) by:

C sup |u|4 kαkL4 (Su,u )
u
(compare with 10.106 with q = 4) using assumption A4.2.

We
next apply
the proceedure outlined above to obtain an estimate for
5
supu |u| kD̂αk remarking again that with D̂α in the role of ξ in Proposition
10.2, we may replace Dξ by D̂ξ. In terms of the definition 7.392, that is:

R40 (D̂α) = sup δ −3/2 |u|5 kD̂αk (10.114)
(u,u)∈D′

′4
R40 (D̂α) ≤ C max{D[1] (α), R[2] (α)} (10.115)
Consider finally definition 10.111 in conjunction with definition 10.100. Then
in view of Lemma 10.1, applying Lemma 5.2 taking p = 4 and recalling the fifth
of the definitions 3.4, that is:
R∞
0 (α) = sup(δ
−3/2
|u|9/2 |α|) (10.116)
M′

R∞ /41 (α), R40 (α)}
0 (α) ≤ C max{R (10.117)
Therefore, combining with the inequalities 10.101 and 10.112, we conclude that
there is a numerical constant C such that:
R∞ ′4
0 (α) ≤ C max{D[1] (α), R[2] (α)} (10.118)
344
Chapter 11
The S-tangential
Derivatives and the
Rotational Lie Derivatives
11.1 Introduction and preliminaries

In the present chapter we shall establish certain coercivity inequalities for the
Lie derivatives of S tensorfields with respect to the rotation fields Oi :P
i = 1, 2, 3.
These inequalities show that for any covariant S tensorfield θ the sum i |L /Oi θ|2
2
bounds pointwise |∇ /θ| .
We define the rescaled induced metric g̃/ on the Su,u by:
g̃/ = |u|−2 g/ (11.1)

Since S0,u is a round sphere of radius |u| in Euclidean 3-dimensional space,
(S0,u , g̃/ ) is isometric to the unit sphere in Euclidean 3-dimensional space.
Su,u
Now, for each u ∈ [0, δ], Φu defines a diffeomorphism of S0,u0 onto Su,u0 and for
each (u, u) ∈ D′ , Φu−u0 defines a diffeomorphism of Su,u0 onto Su,u . Thus, for
each (u, u) ∈ D′ , Φu−u0 ◦ Φu defines a diffeomorphism of S0,u0 onto Su,u . We
consider the pullback:
g̃/(u, u) = (Φu−u0 ◦ Φu )∗ g̃/ = Φ∗u Φ∗u−u0 g̃/ (11.2)

Su,u Su,u
a metric on S0,u0 . We have:
g̃/(0, u0 ) = g̃/
S0,u0
◦
and (Su,u0 , g̃/ ) may be identified with (S 2 , g/), the unit sphere in Euclidean
S0,u0
3
ℜ . As a preliminary step we establish bounds on the eigenvalues of g̃/(u, u)
345
◦
with respect to g/ under the bootstrap assumptions A1.1, A1.2, A3.1, A3.2 of
Chapter 10.
Let θ be an arbitrary p covariant S tensorfield on M ′ and let θ(u, u) be the
pullback:
θ(u, u) = (Φu−u0 ◦ Φu )∗ θ|Su,u = Φ∗u Φ∗u−u0 θ|Su,u (11.3)
a p covariant tensorfield on S0,u0 . Then according to the discussion in the
paragraph which preceeds Lemma 4.3 we have:
∂θ
(u, u0 ) = Φ∗u Dθ|Su,u (11.4)
∂u 0
and:
∂θ
(u, u) = Φ∗u Φ∗u−u0 Dθ|Su,u (11.5)
∂u
From equations 1.41 we then obtain, in the case of the rescaled induced metric
g̃/:
∂ g̃/
(u, u0 ) = (2|u|−2 Ωχ)(u, u0 ) (11.6)
∂u
and:
∂ g̃/
(u, u) = (2|u|−2 Ωχ + 2|u|−1 g̃/)(u, u) (11.7)
∂u
Lemma 11.1 Let λ(u, u) and Λ(u, u) be respectively the smallest and largest
◦
eigenvalues of g̃/(u, u) relative to g̃(0, u0 ) =g/. Then, under the bootstrap as-
sumptions A1.1, A1.2, A3.1, A3.2 of Chapter 10 we have:
1
≤ λ(u, u) ≤ Λ(u, u) ≤ 4
4
Proof: The proof is along the lines of that of Lemma 5.3. We now define:
dµ/(u,u)
g̃ p
µ(u, u) = = λ(u, u)Λ(u, u) (11.8)
dµ/(0,u
g̃ 0)
and:
s
1 Λ(u, u) Λ(u, u)
ν(u, u) = sup g̃/(u, u)(X, X) = = (11.9)
µ(u, u) |X|g̃/(0,u ) =1 µ(u, u) λ(u, u)
0
By 11.6 and 11.7 we have:

!
∂µ 1 ∂ g̃/
(u, u0 ) = tr/(u,u (u, u0 ) µ(u, u0 ) = (Ωtrχ)(u, u0 )µ(u, u0 ) (11.10)
∂u 2 g̃ 0) ∂u
346
and:
!
∂µ 1 ∂ g̃/
(u, u) = trg̃/(u, u) (u, u) µ(u, u) = (Ωtrχ + 2|u|−1 )(u, u)µ(u, u)
∂u 2 ∂u
(11.11)
Integrating 11.10 with respect to u and noting that µ(0, u0 ) = 1 we obtain:
Z u
µ(u, u0 ) = exp (Ωtrχ)(u′ , u0 )du′ (11.12)
0
The bootstrap assumption A1.1 then implies:

2−1/3 ≤ µ(u, u0 ) ≤ 21/3 (11.13)
Integrating 11.11 with respect to u we obtain:
Z u
µ(u, u) = exp (Ωtrχ + 2|u|−1 )(u, u′ )du′ µ(u, u0 ) (11.14)
u0
The bootstrap assumption A3.1 then implies:

µ(u, u)
2−2/3 ≤ ≤ 22/3 (11.15)
µ(u, u0 )
Combining 11.13 and 11.15 we conclude that:
1
≤ µ(u, u) ≤ 2 (11.16)
2
To estimate ν(u, u) we set:
g̃/(u, u)
ĝ/(u, u) = (11.17)
µ(u, u)
Then according to the definition 11.9:
sup ĝ/(X, X) = ν(u, u) (11.18)

|X|g̃
/(0,u
=1
0)
From 11.6 and 11.10 we deduce:

∂ ĝ/
(u, u0 ) = (2|u|−2 µ−1 Ωχ̂)(u, u0 ) (11.19)
∂u
while from 11.7 and 11.11 we deduce:
∂ ĝ/
(u, u) = (2|u|−2 µ−1 Ωχ̂)(u, u) (11.20)
∂u
Consider any tangent vector X to S0,u0 such that |X|/(0,u
g̃ 0)
= 1. Then:
∂ ĝ/ 2
(u, u0 )(X, X) = (Ωχ̂)(u, u0 )(X, X) (11.21)
∂u |u0 |2 µ(u, u0 )
347
and:
∂ ĝ/ 2
(u, u)(X, X) = (Ωχ̂)(u, u)(X, X) (11.22)
∂u |u|2 µ(u, u)
Integrating 11.21 with respect to u and noting that ĝ/(0, u0 )(X, X) = 1 yields:
Z u
|(Ωχ̂)(u′ , u0 )(X, X)| ′
ĝ/(u, u0 )(X, X) ≤ 1 + 2 du (11.23)
0 |u0 |2 µ(u′ , u0 )
Also, integrating 11.22 with respect to u yields:
|(Ωχ̂)(u, u′ )(X, X)| ′

Z u
ĝ/(u, u)(X, X) ≤ ĝ/(u, u0 )(X, X) + 2 du (11.24)
u0 |u′ |2 µ(u, u′ )
Now if ξ is an arbitrary type Tpq tensorfield on S0,u0 then, in components with

respect to an arbitrary local frame field on S0,u0 ,
2 −1 −1
|ξ|/(u,u)
g̃
= (g̃/(u, u))A1 B1 ...(g̃/(u, u))Aq Bq (g̃/ (u, u))C1 D1 ...(g̃/ (u, u))Cp Dp
A ...A B ...B
ξC11..Cpq ξD11 ...Dqp (11.25)
It follows that:
(λ(u, u))q (Λ(u, u))−p |ξ|/(0,u

2
g̃
2
≤ |ξ|/(u,u) ≤ (Λ(u, u))q (λ(u, u))−p |ξ|/(0,u
2
0)g̃ g̃ 0)
(11.26)
In particular, taking ξ = (Ωχ̂)(u, u0 ) (q = 0, p = 2), we have:
|(Ωχ̂)(u, u0 )|/(u,u
g̃ 0)
≥ (Λ(u, u0 ))−1 |(Ωχ̂)(u, u0 )|/(0,u
g̃ 0)
(11.27)
hence:
2
|(Ωχ̂)(u, u0 )(X, X)| ≤ |(Ωχ̂)(u, u0 )|/(0,u
g̃ 0)
|X|/(0,u
g̃
(11.28)
0)
= |(Ωχ̂)(u, u0 )|/(0,u
g̃ 0)
≤ Λ(u, u0 )|(Ωχ̂)(u, u0 )|/(u,u
g̃ 0)
Also, taking ξ = (Ωχ̂)(u, u) (q = 0, p = 2), we have:
|(Ωχ̂)(u, u)|/(u,u)
g̃ ≥ (Λ(u, u))−1 |(Ωχ̂)(u, u)|/(0,u
g̃ 0)
(11.29)
hence:
2
|(Ωχ̂)(u, u)(X, X)| ≤ |(Ωχ̂)(u, u)k/(0,u
g̃ 0)
|X|/(0,u
g̃
(11.30)
0)
= |(Ωχ̂)(u, u)|/(0,u
g̃ 0)
≤ Λ(u, u)|(Ωχ̂)(u, u)|/(u,u)
g̃
Moreover, for an arbitrary type Tpq tensorfield ξ on S0,u0 we have:
|ξ|/(u,u)
g̃ = |u|p−q |ξ|/g(u,u) (11.31)
348
Thus 11.28 and 11.30 imply
|(Ωχ̂)(u, u0 )(X, X)| ≤ Λ(u, u0 )|u0 |2 |(Ωχ̂)(u, u0 )|/g(u,u0 ) (11.32)
and
|(Ωχ̂)(u, u)(X, X)| ≤ Λ(u, u)|u|2 |(Ωχ̂)(u, u)|/g(u,u) (11.33)
Substituting 11.32 in 11.23, taking the supremum over X ∈ Tq S0,u0 such that
|X|/(0,u
g̃ 0)
= 1 at each q ∈ S0,u0 , and recalling 11.9 and 11.18 we obtain the
following linear integral inequality for ν(u, u0 ):
Z u
ν(u, u0 ) ≤ 1 + 2 |(Ωχ̂)(u′ , u0 )|/g(u′ ,u0 ) ν(u′ , u0 )du′ (11.34)
0
This implies:
Z u
′
ν(u, u0 ) ≤ exp 2 |(Ωχ̂)(u , u0 )| g(u′ ,u0 )
/ du′ (11.35)
0
Also, substituting 11.33 in 11.24, taking the supremum over X ∈ Tq S0,u0 such
that |X|/(0,u
g̃ 0)
= 1 at each q ∈ S0,u0 , and recalling 11.9 and 11.18 we obtain the
following linear integral inequality for ν(u, u):
Z u
ν(u, u) ≤ ν(u, u0 ) + 2 |(Ωχ̂)(u, u′ )|/g(u,u′ ) du′ (11.36)
u0
This implies:
Z u
′ ′
ν(u, u) ≤ ν(u, u0 ) exp 2 |(Ωχ̂)(u, u )|/g(u,u′ ) du (11.37)
u0
Now, we have:
|(Ωχ̂)(u, u)|/g(u,u) = (|Ωχ̂|/g )(u, u), |(Ωχ̂)(u, u)|/g(u,u) = (|Ωχ̂|/g )(u, u) (11.38)
Thus assumption A1.2 yields through 11.35:
ν(u, u0 ) ≤ 21/3 (11.39)
while assumption A3.2 yields through 11.37:
ν(u, u) ≤ 22/3 ν(u, u0 ) (11.40)
Combining, we conclude that:
ν(u, u) ≤ 2 (11.41)
Since by the definitions 11.8 and 11.9:

µ(u, u)
λ(u, u) = , Λ(u, u) = µ(u, u)ν(u, u) (11.42)
ν(u, u)
349
11.16 together with 11.41 imply the lemma.
Let Γ
/(u, u) be the connection of g/(u, u) and Γ/(0, u0) the connection of g/(0, u0 )
(both are connections on T S0,u0 ). Then Γ /(u, u) − Γ/(0, u0 ) is a T21 -type S ten-
sorfield symmetric in the lower indices. By 11.4 and 11.5 we have:
∂Γ
/
(u, u0 ) = Φ∗u DΓ
/|Su,u = (DΓ
/)(u, u0 ) (11.43)
∂u 0
and:
∂Γ
/
(u, u0 ) = Φ∗u Φ∗u−u0 DΓ/|Su,u = (DΓ /)(u, u) (11.44)
∂u
Integrating 11.43 with respect to u we obtain:
Z u
/(u, u0 ) − Γ
Γ /(0, u0 ) = (DΓ/)(u′ , u0 )du′ (11.45)
0
Also, integrating 11.44 with respect to u we obtain:

Z u
Γ
/(u, u) − Γ
/(u, u0 ) = /)(u, u′ )du′
(DΓ (11.46)
u0
/(u, u) be the connection of g̃/(u, u). Since the metrics g̃/(u, u) and g/(u, u)
Let Γ̃
differ by a constant scale factor their connections coincide:
/(u, u) = Γ
Γ̃ /(u, u) (11.47)
Thus, the left hand sides of 11.45 and 11.46 may be replaced by Γ̃ /(u, u0 )−Γ̃
/(0, u0 )
and Γ̃
/(u, u) − Γ̃
/(u, u0 ) respectively. These equations then imply
Z u
/(u, u0 ) − Γ̃
|Γ̃ /(0, u0 )|/(0,u
g̃ 0)
≤ /)(u′ , u0 )|/(0,u
|(DΓ g̃ 0)
du′ (11.48)
0
and Z u
|Γ̃
/(u, u) − Γ̃
/(u, u0 )|/(0,u
g̃ 0)
≤ /)(u, u′ )|/(0,u
|(DΓ g̃ 0)
du′ (11.49)
u0
/)(u, u) and (DΓ
respectively. Now with (DΓ /)(u, u) in the role of ξ (q = 1, p = 2),
11.26 and 11.31 give:
Λ(u, u)
/)(u, u)|/(0,u
|(DΓ g̃ ≤ p /)(u, u)|/(u,u)
|(DΓ g̃ (11.50)
0
λ(u, u)
Λ(u, u) Λ(u, u)
= |u| p |(DΓ)(u, u)|/g(u,u) = |u| p /|/g )(u, u)
(|DΓ
λ(u, u) λ(u, u)
and:
Λ(u, u)
|(DΓ
/)(u, u)|/(0,u
g̃
≤ p |(DΓ
/)(u, u)|/(u,u)
g̃
(11.51)
0
λ(u, u)
Λ(u, u) Λ(u, u)
= |u| p |(DΓ)(u, u)|/g(u,u) = |u| p /|/g )(u, u)
(|DΓ
λ(u, u) λ(u, u)
350
Hence, by Lemma 11.1:
|(DΓ
/)(u, u)|/(0,u
g̃ 0)
≤ 8|u|(|DΓ
/|/g )(u, u) (11.52)
and:
/)(u, u)|/(0,u
|(DΓ g̃ 0)
≤ 8|u|(|DΓ
/|/g )(u, u) (11.53)
In view of 11.52, 11.53, the inequalities 11.48 and 11.49 imply:
Z u
/(u, u0 ) − Γ̃
|Γ̃ /(0, u0 )|/(0,u
g̃ 0)
≤ 8|u0 | /|/g )(u′ , u)du′
(|DΓ (11.54)
0
and: Z u
|Γ̃
/(u, u) − Γ̃
/(u, u0 )|/(0,u
g̃ 0)
≤8 |u′ |(|DΓ
/|/g )(u, u′ )du′ (11.55)
u0
respectively.
Now, the estimate 8.95 yields, through 11.54:
|Γ̃
/(u, u0 ) − Γ̃
/(0, u0 )|/(0,u
g̃ 0)
≤ O(δ 1/2 |u0 |−1 ) (11.56)
while the estimate 8.101 yields, through 11.55:
/(u, u) − Γ̃
|Γ̃ /(u, u0 )|/(0,u
g̃ 0)
≤ O(δ 1/2 |u|−1 ) (11.57)
Combining then yields the conclusion:
/(u, u) − Γ̃
|Γ̃ /(0, u0 )|/(0,u
g̃ 0)
≤ O(δ 1/2 |u|−1 ) (11.58)
Thus if δ is suitably small depending on D0∞ , R∞ /41 , R

0 , D /41 we have:
|Γ̃
/(u, u) − Γ̃
/(0, u0 )|/(0,u
g̃ 0)
≤1 (11.59)
However, the estimates 8.95, 8.101 depend on the results of Chapter 6, and
these in turn depend on the results of Chapters 3, 4 and 5, which rely on the
boundedness of the quantities R∞0 ,R/41 and R
/2 . The present chapter on the other
hand is to be antecedent to the establishment of bounds for these quantities in
the logic of the proof of the existence theorem. We thus introduce in present
chapter as a bootstrap assumption:
/(u, u) − Γ̃
B1: supS0,u0 |Γ̃ /(0, u0 )|/(0,u
g̃ 0)
≤ 1 : for all (u, u) ∈ D′
11.2 The coercivity inequalities on the standard

sphere
The coercivity inequalities on (Su,u , g/|Su,u ) shall be established on the basis
of the coercivity inequalities on (S0,u0 , g̃/ ), which, as we have seen, can
S0,u0
351
◦
be identified with (S 2 , g/), the unit sphere in Euclidean 3-dimensional space.
Consider the rotation fields Oi : i = 1, 2, 3 in Euclidean 3-dimensional space.
These are given in rectangular coordinates by:
∂
Oi = ǫijk xj (11.60)
∂xk
and are tangential to S 2 . In the following lemma we denote by ( , ) and | | the
◦ ◦
Euclidean inner product and norm. Then g/ and ∇
/ are the induced metric and
induced covariant derivative on
S 2 = {x ∈ ℜ3 : |x| = 1}
Also, we consider p covariant tensorfields on S 2 as p covariant tesorfields on

Euclidean 3-dimensional space defined along S 2 and vanishing if one of the
entries is normal to S 2 . Thus, if ξ is such a tensorfield its components ξa1 ...ap
in rectangular coordinates satisfy:
ξ b xb = 0, : i = 1, ..., p (11.61)
a1 ...>ai <...ap
Moreover, if ξ is any p covariant tensorfield on S 2 we have:
|ξ|◦ = |ξ| (11.62)

/
g
and if ξ, ζ are any two p covariant tensorfields on S 2 we have:
(ξ, ζ)◦ = (ξ, ζ) (11.63)

/
g
Repeated indices are as usual summed, nevertheless we shall explicitly denote

the summation over the three rotation fields Oi : i = 1, 2, 3.
Lemma 11.2 Let f be a function on S 2 . Then we have:

X
(Oi f )2 = |d
/f |2◦
/
g
i
Let ξ be a 1-form on S 2 . Then we have:

X ◦
/Oi ξ|2◦ = | ∇
|L / ξ|2◦ + |ξ|2◦
/
g /
g /
g
i
Let ξ be a p covariant tensorfield on S 2 with p ≥ 2. Then we have:

X ◦ X k∼l X i,j
|L / ξ|2◦ + p|ξ|2 +
/Oi ξ|2◦ = | ∇ (ξ, ξ )◦ − | tr ξ|2◦
/
g /
g /
g /
g
i k6=l k6=l
352
k∼l
Here for k 6= l we denote by ξ the transpose of ξ with respect to the kth and
lth index:
k∼l
ξ a1 ...ak ...al ...ap = ξa1 ...al ...ak ...ap
if k < l and similarly with the roles of k and l reversed if k > l. Also, for k 6= l
k,l
we denote by tr ξ the p − 2 covariant tensorfield on S 2 obtained by tracing the
kth and lth index:
k,l
tr ξa1 ...>ak <...>al <...ap = ξa1 ...b...b...ap
if k < l and similary with the roles of k and l reversed if k > l. Obviously,
k∼l l∼k k,l j,k
ξ = ξ and tr ξ = tr ξ
Proof : We have:
X X
(Oi )m (Oi )n = ǫijm xj ǫikn xk = (δjk δmn − δjn δkm )xj xk
i i
= δmn − xm xn = Πm
n (11.64)
where Πm n 2
n = Πm are the components of the orthogonal projection to S . The
first part of the lemma follows directly from this formula:
X X ∂f ∂f ∂f ∂f
(Oi f )2 = (Oi )m m
(Oi )n n = Πm
n /f |2 = |d
= |d /f |2◦
i i
∂x ∂x ∂xm ∂xn /
g
◦
To establish the rest of the lemma we note that ∇
/ Oi = Π·((Π·∇)Oi ), where
∇ is the covariant derivative in Euclidean space. By 11.60:
∂(Oi )n
(∇Oi )nj = = ǫijn (11.65)
∂xj
hence:
◦
/ Oi )m
(∇ j m
a = Πa Πn ǫijn (11.66)
We then have:
X ◦ X
(Oi )n (∇
/ Oi )m
a = ǫikn xk Πja Πm k j m
l ǫijl = (δkj δnl − δkl δnj )x Πa Πl
i i
= x Πja Πm
j l n m
n − x Πa Πl = 0 (11.67)
and:
X ◦ ◦ X
/ Oi )m
(∇ / Oi )nb =
a (∇ Πca Πm d n
k ǫick Πb Πl ǫidl
i i
= (δcd δkl − δcl δdk )Πca Πm d n
k Πb Πl
= Πca Πcb Πm n l k m n a m n m
k Πk − Πa Πb Πk Πl = Πb Πn − Πa Πb (11.68)
353
Let then ξ be a 1-form on S 2 . We have:
◦ ◦
/Oi ξ)a + ξm (∇
/Oi ξ)a = (∇
(L / Oi )m
a (11.69)
and:
◦ ◦
/Oi ξ)a = (Oi )m (∇
(∇ / ξ)ma (11.70)
Hence:
X X ◦ ◦
/Oi ξ|2◦ =
|L (Oi )m (∇
/ ξ)ma (Oi )n (∇
/ ξ)na (11.71)
/
g
i i
X ◦ ◦ X ◦ ◦
+2 (Oi )n (∇ / Oi )m
/ ξ)na ξm (∇ a + / Oi )m
ξm (∇ / Oi )na
a ξn (∇
i i
By 11.64 the first term on the right is:

◦ ◦ ◦
Πm
n (∇
/ ξ)ma (∇ / ξ|2
/ ξ)na = | ∇ (11.72)
By 11.67 the second term on the right in 11.71 vanishes, while by 11.68, which
implies:
X ◦ ◦
/ Oi )m
(∇ / Oi )na = Πm
a (∇ n (11.73)
i
the third term on the right in 11.71 is equal to:
Πm
n ξm ξn = |ξ|
2
(11.74)
We thus obtain:
X ◦
/Oi ξ|2 = | ∇
|L / ξ|2 + |ξ|2
i
which, in view of 11.62, is the second part of the lemma.
Let finally ξ be a p covariant tensorfield on S 2 , p ≥ 2. We have:
◦ p
X ◦
(L /Oi ξ)a1 ...ap +
/Oi ξ)a1 ...ap = (∇ ξ m / Oi )m
(∇ ak (11.75)
a1 ...>ak <...ap
k=1
and:
◦ ◦
/Oi ξ)a1 ...ap = (Oi )m (∇
(∇ / ξ)ma1 ...ap (11.76)
Hence:
X X ◦ ◦
/Oi ξ|2 =
|L / ξ)ma1 ...ap (Oi )n (∇
(Oi )m (∇ / ξ)na1 ...ap (11.77)
i i
p
XX ◦ ◦
+2 (Oi )n (∇
/ ξ)na1 ...ap ξ m / Oi )m
(∇ ak
a1 ...>ak <...ap
i k=1
p
X X ◦ ◦
+ ξ m / Oi )m
(∇ ak ξ n / Oi )nal
(∇
a1 ...>ak <...ap a1 ...>al <...ap
i k,l=1
354
By 11.64 the first term on the right is:
◦ ◦ ◦
Πm
n (∇ / ξ|2
/ ξ)na1 ...ap = | ∇
/ ξ)ma1 ...ap (∇ (11.78)
By 11.67 the second term on the right in 11.77 vanishes. The third term on the
right in 11.77 splits into:
p
XX ◦ ◦
ξ m / Oi )m
(∇ ak ξ n / Oi )nak
(∇
a1 ...>ak <...ap a1 ...>ak <...ap
i k=1
p
X X ◦ ◦
+ ξ m / Oi )m
(∇ ak ξ n / Oi )nal
(∇ (11.79)
a1 ...>ak <...ap a1 ...>al <...ap
i k6=l=1
By 11.73 the first sum is equal to:

p
X
Πm
nξ m ξ n = p|ξ|2 (11.80)
a1 ...>ak <...ap a1 ...>ak <...ap
k=1
while by 11.68 the second sum is equal to:

p
X
(Πaakl Πm m n
n − Πal Πak )ξ m ξ n
a1 ...>ak ...ap a1 ...>al ...ap
k6=l=1
Xp
= (δak al δmn − δmal δnak )ξ m ξ n
a1 ...>ak ...ap a1 ...>al ...ap
k6=l=1
p
X k∼l k,l
2
= (ξ, ξ ) − ( tr ξ) (11.81)
k6=l=1
We thus obtain:
X ◦ X k∼l X k,l
/Oi ξ|2 = | ∇
|L / ξ|2 + p|ξ|2 + (ξ, ξ ) − ( tr ξ)2
i k6=l k6=l
which, in view of 11.62, 11.63, is the third part of the lemma.
Given a p covariant tensorfield ξ on S 2 with p ≥ 2, for each k 6= l = 1, ..., p,

k,l
we may decompose ξ into a part θ which is symmetric in the kth and lth index
k,l
and a part ω which is antisymmetric in the kth and lth index:
k,l k,l k,l 1 k∼l k,l 1 k∼l
ξ =θ + ω, θ= (ξ+ ξ ), ω= (ξ− ξ ) (11.82)
2 2
k,l k,l
The symmetric part θ may be further decomposed into its trace-free part σ
and its trace part:
k,l k,l 1 ◦ k,l
θ = σ + g/ tr ξ (11.83)
2
355
For each k 6= l = 1, ..., p we have:
k,l k,l k,l k,l 1 k,l
|ξ|2 = | θ |2 + | ω |2 = | σ |2 + | ω |2 + | tr ξ|2 (11.84)
2
Hence:
X k,l k,l 1 k,l

p(p − 1)|ξ|2 = | σ |2 + | ω |2 + | tr ξ|2 (11.85)
2
k6=l
On the other hand we have, for each k 6= l = 1, ..., p,

k∼l k,l k,l k,l k,l 1 k,l
(ξ, ξ ) = | θ |2 − | ω |2 = | σ |2 − | ω |2 + | tr ξ|2 (11.86)
2
Hence:
X k∼l k,l
X
k,l k,l 1 k,l

(ξ, ξ ) − | tr ξ|2 = | σ |2 − | ω |2 − | tr ξ|2 (11.87)
2
k6=l k6=l
Comparing 11.87 with 11.85 we conclude that:

X k∼l k,l

(ξ, ξ ) − | tr ξ| ≥ −p(p − 1)|ξ|2
2
(11.88)
k6=l
Therefore the third conclusion of Lemma 11.2 implies that for any p covariant
tensorfield on S 2 with p ≥ 2 we have:
X ◦
|L / ξ|2◦ − p(p − 2)|ξ|2◦
/Oi ξ|2◦ ≥ | ∇ (11.89)
/
g /
g /
g
i
In view of the first two parts of the lemma this also holds in the cases p = 0, 1.
11.3 The coercivity inequalities on Su,u

We now derive on the basis of Lemmas 11.1 and 11.2 and the bootstrap assump-
tion B1, the coercivity inequalities on Su,u .
Proposition 11.1 Let the bootstrap assumptions A1.1, A1.2, A3.1, A3.2
and B1 hold. Let f be an arbitrary function defined on M ′ . Then for every
(u, u) ∈ D′ we have, pointwise on Su,u :
X
(Oi f )2 ≥ 4−1 |u|2 |d
/f |/2g
i
Let ξ be a p covariant S tensorfield defined on M ′ , p ≥ 1. Then for every

(u, u) ∈ D′ we have, pointwise on Su,u :
X /
g
/Oi ξ|/2g ≥ 2−1 4−1−2p |u|2 | ∇
|L / ξ|/2g − 2p(p − 1)|ξ|/2g
i
356
Proof : Given a function f on M ′ , we consider the function
f (u, u) = (Φu−u0 ◦Φu )∗ f |Su,u = Φ∗u Φ∗u−u0 f |Su,u = f |Su,u ◦Φu−u0 ◦Φu (11.90)
on S0,u0 . To this function we apply the first conclusion of Lemma 11.2 to obtain,
◦
in view of the identification of (S0,u0 , g̃/(0, u0 )) with (S 2 , g/),
X
(Oi (f (u, u)))2 = |d 2
/f (u, u)|/(0,u
g̃ )
(11.91)
0
i
Now, by 8.39 we have, for every p ∈ Su,u :
Oi (p)f = (dΦu−u0 (q) · Oi (q))f = Oi (q)(f ◦ Φu−u0 ), q = Φu0 −u (p) ∈ Su,u0

(11.92)
while by 8.36 we have, for every q ∈ Su,u0 :
Oi (q)(f ◦ Φu−u0 ) = (dΦu (q0 ) · Oi (q0 ))(f ◦ Φu−u0 )

= Oi (q0 )(f ◦ Φu−u0 ◦ Φu ), q0 = Φ−u (q) ∈ S0,u0 (11.93)
hence, combining:
Oi (p)f = Oi (q0 )f (u, u), q0 = Φ−u (Φu0 −u (p)) : ∀p ∈ Su,u (11.94)
In terms of the function (Oi f )(u, u) = (Φu−u0 ◦ Φu )∗ (Oi f )|Su,u , this reads:
(Oi f )(u, u) = Oi (f (u, u)) (11.95)
On the other hand, by Lemma 11.1 and inequality 11.26 with /df (u, u) in the
role of ξ (p = 1, q = 0) we have:
2 2
|d
/f (u, u)|/(0,u
g̃
≥ λ(u, u)|d
/f (u, u)|/(u,u) ≥ 4−1 |d 2
/f (u, u)|/(u,u) (11.96)
0) g̃ g̃
while by 11.31:
2
/f (u, u)|/(u,u)
|d g̃
/f (u, u)|/2g(u,u) = |u|2 (|d
= |u|2 |d /f |/2g )(u, u) (11.97)
hence, combining:
2
|d
/f (u, u)|/(0,u
g̃
≥ 4−1 |u|2 (|d
/f |/2g )(u, u) (11.98)
0)
In view of 11.95 and 11.98 we deduce from 11.91:

X
((Oi f )(u, u))2 ≥ 4−1 |u|2 (|d
/f |/2g )(u, u) (11.99)
i
which yields the first part of the proposition.

Next, given a p covariant S tensorfield ξ on M ′ , p ≥ 1, we consider the p
covariant tensorfield
ξ(u, u) = (Φu−u0 ◦ Φu )∗ ξ|Su,u = Φ∗u Φ∗u−u0 ξ|Su,u (11.100)
357
To this tensorfield we apply 11.89, a corollary of Lemma 11.2, to obtain, in view
◦
of the identification of (S0,u0 , g̃/(0, u0 )) with (S 2 , g/),
X /(0,u
g̃ 0)
2 2 2
/Oi (ξ(u, u))|/(0,u
|L g̃
≥| ∇
/ ξ(u, u)|/(0,u − p(p − 2)|ξ(u, u)|/(0,u
0) g̃ 0 ) g̃ 0)
i
(11.101)
Now, let Ψi,s be the 1-parameter rotation group generated by Oi . According
to the definition of the action of the rotation group of Chapter 8 we have, for
every (u, u) ∈ D′ :
Ψi,s ◦ Φu−u0 = Φu−u0 ◦ Ψi,s : on Su,u0 (11.102)
and:
Ψi,s ◦ Φu = Φu ◦ Ψi,s : on S0,u0 (11.103)
Composing 11.102 on the right by Φu acting on S0,u0 we obtain, by virtue of
11.103,
Ψi,s ◦ (Φu−u0 ◦ Φu ) = (Φu−u0 ◦ Φu ) ◦ Ψi,s : on S0,u0 (11.104)
This expresses the equality of two diffeomorphisms of S0,u0 onto Su,u . Therefore
the corresponding pullbacks, which map p covariant tensorfields on Su,u to p
covariant tensorfields on S0,u0 , coincide:
(Ψi,s ◦ (Φu−u0 ◦ Φu ))∗ = ((Φu−u0 ◦ Φu ) ◦ Ψi,s )∗
or:
(Φu−u0 ◦ Φu )∗ Ψ∗i,s = Ψ∗i,s (Φu−u0 ◦ Φu )∗ (11.105)
Applying this to ξ|Su,u we have:
(Φu−u0 ◦ Φu )∗ Ψ∗i,s ξ|Su,u = Ψ∗i,s (Φu−u0 ◦ Φu )∗ ξ|Su,u = Ψ∗i,s (ξ(u, u)) (11.106)
Taking the derivative of this with respect to s at s = 0, we have on the left hand
side the pullback by Φu−u0 ◦ Φu of:

d ∗
Ψ ξ| = (L/Oi ξ)|Su,u (11.107)
ds i,s Su,u s=0
and on the right hand side:

d ∗
Ψ (ξ(u, u)) /Oi (ξ(u, u))
=L (11.108)
ds i,s s=0
We thus obtain:
(Φu−u0 ◦ Φu )∗ (L
/Oi ξ)|Su,u = L
/Oi (ξ(u, u))
358
or:
/Oi ξ)(u, u) = L
(L /Oi (ξ(u, u)) (11.109)
In view of the equality 11.109, 11.101 is equivalent to:
X /(0,u
g̃ 0)
2 2 2
/Oi ξ)(u, u))|/(0,u
|(L g̃
≥| ∇
/ ξ(u, u)|/(0,u − p(p − 2)|ξ(u, u)|/(0,u
0) g̃ ) 0 g̃ 0)
i
(11.110)
/(u,u)
g̃
Consider next / ξ. We have, in an arbitrary local frame field on S0,u0 ¡
∇
/(u,u)
g̃ /(0,u
g̃ 0)
/ ξ(u, u))AB1 ...Bp = ( ∇
( ∇ / ξ(u, u))AB1 ...Bp
Xp
− (Γ̃ /(0, u0 ))C
/(u, u) − Γ̃ ABk ξ(u, u) C
B1 ...>Bk <...Bp
k=1
Hence:
/(u,u)
g̃ /(0,u
g̃ 0)
| / ξ(u, u)|/(0,u
∇ g̃ 0)
≤| ∇
/ ξ(u, u)|/(0,u
g̃ 0)
/(u, u) − Γ̃
+p|Γ̃ /(0, u0 )|/(0,u
g̃ 0)
|ξ(u, u)|/(0,u
g̃ 0)
/(0,u
g̃ 0)
≤| ∇
/ ξ(u, u)|/(0,u
g̃ 0)
+ p|ξ(u, u)|/(0,u
g̃ (11.111)
0)
by virtue of assumption B1. Thus, we have:

/(u,u)
g̃ /(0,u
g̃ 0)
2
| / ξ(u, u)|/(0,u
∇ g̃ 0)
≤ 2| ∇
/ ξ(u, u)|/(0,u
g̃ )
+ 2p2 |ξ(u, u)|/(0,u
2
g̃
0 0)
and combining with 11.110 we obtain:

X 1 /(u,u)
2
g̃
2 2
/Oi ξ)(u, u))|/(0,u
|(L |
g̃
≥
/ ξ(u, u)|/(0,u
∇ − 2p(p − 1)|ξ(u, u)|/(0,u
0) 2 g̃ 0) g̃ 0)
i
(11.112)
On the other hand, by Lemma 11.1 and inequality 11.26 applied to ξ(u, u) and
/Oi ξ)(u, u):
to (L
2
|ξ(u, u)|/(0,u
g̃
≤ (Λ(u, u))p |ξ(u, u)|/(u,u)
2 2
≤ 4p |ξ(u, u)|/(u,u) (11.113)
0) g̃ g̃
and:
2
/Oi ξ)(u, u)|/(0,u
|(L g̃
≤ (Λ(u, u))p |(L 2
/Oi ξ)(u, u)|/(u,u) ≤ 4p |(L 2
/Oi ξ)(u, u)|/(u,u)
0) g̃ g̃
(11.114)
/(u,u)
g̃
Also, by Lemma 11.1 and inequality 11.26 with ∇
/ ξ(u, u) in the role of ξ
(and (p + 1, 0) in the role of (p, q)):
/(u,u)
g̃ /(u,u)
g̃
2
| / ξ(u, u)|/(0,u
∇ g̃
≥ (λ(u, u))p+1 | / ξ(u, u)|/(u,u)
∇ g̃
0)
/(u,u)
g̃
≥ 4−p−1 | ∇
/ ξ(u, u)|/(u,u)
g̃ (11.115)
359
Moreover, by 11.31 we have:
2
|ξ(u, u)|/(u,u)
g̃
= |u|2p |ξ(u, u)|/g(u,u) , |(L 2
/Oi ξ)(u, u)|/(u,u)
g̃
/Oi ξ)(u, u)|/2g(u,u)
= |u|2p |(L
(11.116)
and, in view of 11.47:
/(u,u)
g̃ /
g(u,u)
| / ξ(u, u)|/(u,u)
∇ g̃
= |u|2p+2 | / ξ(u, u)|/g(u,u)
∇ (11.117)
Finally, for any diffeomorphism Ω of S0,u0 onto Su,u and any p covariant ten-
sorfield ξ on Su,u we have:
Ω∗ /
g /
g
/ (Ω∗ ξ) = Ω∗ (∇
∇ / ξ) (11.118)
In particular this holds in the case Ω = Φu−u0 ◦ Φu . Hence:
/
g(u,u) /
g
/ ξ(u, u) = (∇
∇ / ξ)(u, u) (11.119)
In view of 11.113 - 11.117, 11.119, the inequality 11.112 implies:

X 1 −2p−1 2 /g
/Oi ξ)(u, u)|/2g(u,u) ≥
|(L 4 / ξ)(u, u)|/2g(u,u) − 2p(p − 1)|ξ(u, u)|/2g(u,u)
|u| |(∇
i
2
(11.120)
Since
/Oi ξ)(u, u)|/g(u,u) = (|L

|(L /Oi ξ|/g )(u, u), |ξ(u, u)|/g(u,u) = (|ξ|/g )(u, u)
and
/
g /
g
|(∇
/ ξ)(u, u)|/g(u,u) = (| ∇
/ ξ|/g )(u, u)
the inequality 11.120 yields the second part of the proposition.
Let ξ be a p-covariant S tensorfield defined on M ′ with p ≥ 2. We denote by

tr/g ξ the p − 2 covariant S tensorfield obtained by tracing the last two indices.
In terms of components in an arbitrary local frame field for Su,u :
(tr/g ξ)A1 ...Ap−2 = (g/−1 )BC ξA1 ...Ap−2 BC (11.121)
Suppose now that ξ is a p covariant S tensorfield on M ′ , p ≥ 2, which is

symmetric and trace-free in the last two indices. Then we have:
(tr/g L /Oi g/−1 )BC ξA1 ...Ap−2 BC

/Oi tr/g ξ)A1 ...Ap−2 − (L
/Oi ξ)A1 ...Ap−2 = (L
(Oi ) BC
= π
/ ξA1 ...Ap−2 BC
or:
(Oi )
/Oi ξ = (
tr/g L π
/, ξ)/g (11.122)
360
Decomposing then L /Oi ξ into L̂ /Oi ξ, its trace-free part with respect to the last
two indices, and (1/2)tr/g ξ ⊗ g/, its trace part with respect to the last two indices,
1
/Oi ξ + tr/g ξ ⊗ g/
/Oi ξ = L̂
L (11.123)
2
we have:
X X 1

/Oi ξ|/2g
|L = 2
/Oi ξ|/g + |tr/g L
|L̂ /Oi ξ|/g 2
i i
2
X 1

= /Oi ξ|/2g + | (Oi ) π
|L̂ /|/2g |ξ|/2g (11.124)
i
2
therefore the bootstrap assumption:
(Oi )
B2: | /| ≤ 1 i=1,2,3 in M ′
π
implies:
X X 3
/Oi ξ|/2g ≥
|L̂ /Oi ξ|/2g − |ξ|/2g
|L (11.125)
i i
2
It follows that in the case of a p covariant S tensorfield which is symmetric and
trace-free in the last two indices Proposition 11.1 takes the form:
X /
g
/Oi ξ|/2g ≥ 2−1 4−1−2p |u|2 | ∇
|L̂ / ξ|/2g − Cp |ξ|/2g
i
: pointwise on Su,u , for every (u, u) ∈ D′ (11.126)
where:
3
Cp = 2p(p − 1) + (11.127)
2
We proceed to derive from Proposition 11.1 2nd order coercivity inequalities
on Su,u . Consider first the case of a function f , defined on M ′ . By the first
part of Proposition 11.1 applied to the function Oi f we have:
X
|u|2 |d
/Oi f |/2g ≤ 4 (Oj Oi f )2 (11.128)
j
hence, summing over i = 1, 2, 3,

X X
|u|2 /Oi f |/2g ≤ 4
|d (Oj Oi f )2 (11.129)
i i,j
Now, we have:
/Oi /df
/dOi f = L (11.130)
Thus, applying the second part of Proposition 11.1 in the case p = 1 to the S
1-form /df we obtain:
X
/ 2 f |/2g ≤ 2.43
|u|2 |∇ /Oi f |/2g
|d (11.131)
i
361
Combining 11.131 and 11.129 we conclude that:
X
|u|4 |∇
/ 2 f |/2g ≤ 2.44 (Oj Oi f )2 (11.132)
i,j
Consider next the case of a p covariant S tensorfield ξ defined on M ′ . In the

following, we shall denote by Cp various numerical constants depending only on
p. By the second part of Proposition 11.1 applied to the p covariant S tensorfield
/Oi ξ we have:
L
 
X 
|u|2 |∇/Oi ξ|/2g ≤ Cp
/L /Oj L
|L /Oi ξ|/2g
/Oi ξ|/2g + |L (11.133)
 
j

 
X X X 
|u|2 |∇/Oi ξ|/2g ≤ Cp
/L /Oi ξ|/2g +
/Oj L
|L /Oi ξ|/2g
|L (11.134)
 
i i,j i
We now express L /Oi ∇

/ξ in terms of ∇ /Oi ξ. By Lemma 9.1 we have, with respect
/L
to an arbitrary local frame field for the Su,u ,
p
X
(Oi ) C
/Oi ∇
(L /ξ − ∇/Oi ξ)AB1 ...Bp = −
/L π
/1,ABk ξ C (11.135)
B1 ...>Bk <...Bk
k=1
It follows that:
!
X X X
2 2 2 (Oi )
/Oi ∇
|L /ξ| ≤ 2 |∇/Oi ξ| + 2p
/L | /1 |/2g
π |ξ|/2g (11.136)
i i i
Applying the second part of Proposition 11.1 to the p + 1 covariant S tensorfield

∇
/ξ we obtain: ( )
X
2 2 2 2 2
/ ξ|/g ≤ Cp
|u| |∇ |L /ξ|/g + |∇
/Oi ∇ /ξ|/g (11.137)
i
Moreover, by Proposition 11.1 applied to ξ itself we have:

( )
X
2 2 2 2
|u| |∇
/ξ| ≤ Cp /Oi ξ|/g + |ξ|/g
|L (11.138)
i
hence, substituting in 11.137,

( )
X X
/ 2 ξ|/2g ≤ Cp |u|2
|u|4 |∇ /ξ|/2g +
/Oi ∇
|L /Oi ξ|/2g + |ξ|/2g
|L (11.139)
i i
362
Combining 11.139, 11.136 and 11.134 we conclude that:

X X
|u|4 |∇
/ 2 ξ|/2g ≤ Cp /Oi ξ|/2g +
/Oj L
|L /Oi ξ|/2g + |ξ|/2g
|L

i,j i
! )
X
+|u|2 /1 |/2g |ξ|/2g
| (Oi ) π (11.140)
i
We shall integrate this inequality on Su,u . The integral on Su,u of the last term
in parenthesis on the right is bounded by:

Z !2 1/2 Z !1/2
X
2
|u| | (Oi ) π
/1 |/2g  |ξ|/4g
Su,u i Su,u
X
≤ |u|2 k (Oi )
/1 k2L4 (Su,u ) kξk2L4 (Su,u )
π (11.141)
i
Now by Lemma 5.1 with p = 4 in that lemma and Lemma 10.1 there is a
numerical constant C such that:
n o
|u|kξk2L4 (Su,u ) ≤ C |u|2 k∇
/ξk2L2 (Su,u ) + kξk2L2 (Su,u )
Z ( )
X
2 2
≤ Cp /Oi ξ|/g + |ξ|/g dµ/g
|L (11.142)
Su,u i
the last step by 11.138 integrated on Su,u . At this point we introduce the
bootstrap assumption:
B3: |u|1/2 k∇
/ (Oi ) π
/kL4 (Su,u ) ≤ 1 : for all (u, u) ∈ D′
The assumption B3 implies that:

3
|u|1/2 k (Oi )
/1 kL4 (Su,u ) ≤
π : ∀(u, u) ∈ D′ (11.143)
2
Then in view of 11.142, the right hand side of 11.141 is bounded by:
Z ( )
X
2 2
Cp /Oi ξ|/g + |ξ|/g dµ/g
|L (11.144)
Su,u i
Integrating 11.140 on Su,u we then conclude that:

 
Z Z X X 
|u|4 |∇
/ 2 ξ|/2g dµ/g ≤ Cp /Oi ξ|/2g +
/Oj L
|L /Oi ξ|/2g + |ξ|/2g dµ/g
|L
Su,u Su,u  i,j

i
(11.145)
363
Consider finally the case of a p covariant S tensorfield ξ defined on M ′ , ≥ 2,
which is symmetric and trace-free in the last two indices. Then by 11.126 applied
to the p covariant S tensorfield L̂ /Oi ξ which is also symmetric and trace-free in
the last two indices we have:
 
X 
|u|2 |∇/Oi ξ|/2g ≤ Cp
/L̂ /Oj L̂
|L̂ /Oi ξ|/2g
/Oi ξ|/2g + |L̂ (11.146)
 
j

 
X X X 
|u|2 |∇/Oi ξ|/2g ≤ Cp
/L̂ /Oi |ξ|/2g +
/Oj L̂
|L̂ /Oi ξ|/2g
|L̂ (11.147)
 
i i,j i
/Oi ∇
We now express L̂ /ξ in terms of ∇/Oi ξ. From 11.122, 11.123 applied to ξ and
/L̂
to ∇
/ξ we deduce:
1
/Oi ∇
L̂ /ξ − ∇ /Oi ∇
/Oi ξ = L
/L̂ /ξ − ∇ / (Oi ) π
/Oi ξ + (∇
/L /, ξ)/g ⊗ g/ (11.148)
2
Together with the formula 11.135 this implies that:
!
X X X
/ξ|2 ≤ 2
/Oi ∇
|L̂ |∇/Oi ξ|2 + 2
/L̂ (2p2 | (Oi )
/1 |2 + |∇
π / (Oi ) π
/|2 ) |ξ|/2g
i i i
(11.149)
Applying 11.126 to the p + 1 covariant S tensorfield ∇ /ξ which is likewise sym-
metric and trace-free in the last two indices we obtain:
( )
X
2 2 2 2 2
/ ξ|/g ≤ Cp
|u| |∇ |L̂ /ξ|/g + |∇
/Oi ∇ /ξ|/g (11.150)
i
Moreover, by 11.126 applied to ξ itself we have:

( )
X
2 2 2 2
/ξ|/g ≤ Cp
|u| |∇ /Oi ξ|/g + |ξ|/g
|L̂ (11.151)
i
hence, substituting in 11.150,

( )
X X
/ 2 ξ|/2g ≤ Cp |u|2
|u|4 |∇ /ξ|/2g +
/Oi ∇
|L̂ /Oi ξ|/2g + |ξ|/2g
|L̂ (11.152)
i i
Combining 11.152, 11.149 and 11.147 we conclude that:


X X
|u|4 |∇
/ 2 ξ|/2g ≤ Cp /Oi ξ|/2g +
/Oj L̂
|L̂ /Oi ξ|/2g + |ξ|/2g
|L̂

i,j i
! )
X
+|u|2 (| (Oi ) π / (Oi ) π
/1 |/2g + |∇ /|/2g ) |ξ|/2g (11.153)
i
364
Now the integral on Su,u of the last term in parenthesis on the right is bounded
by:
X
|u|2 (k (Oi ) π / (Oi ) π
/1 k2L4 (Su,u ) + k∇ /|2L4 (Su,u ) )kξk2L4 (Su,u ) (11.154)
i
Again, by Lemma 5.1 with p = 4 in that lemma and Lemma 10.1 there is a
n o
|u|kξk2L4 (Su,u ) ≤ C |u|2 k∇
/ξk2L2 (Su,u ) + kξk2L2 (Su,u )
Z ( )
X
2 2
≤ Cp /Oi ξ|/g + |ξ|/g dµ/g
|L̂ (11.155)
Su,u i
the last step by 11.151 integrated on Su,u . Then by virtue of assumption B3

11.154 is bounded by:
Z ( )
X
2 2
Cp /Oi ξ|/g + |ξ|/g dµ/g
|L̂ (11.156)
Su,u i
Integrating 11.153 on Su,u we then conclude that in the case that ξ is a p

covariant S tensorfield on M ′ which is symmetric and trace-free in the last two
indices we have:
 
Z Z X X 
|u|4 |∇
/ 2 ξ|/2g dµ/g ≤ Cp /Oi ξ|/2g +
/Oj L̂
|L̂ /Oi ξ|/2g + |ξ|/2g dµ/g
|L̂
Su,u Su,u i,j i

(11.157)
365
Chapter 12
Weyl Fields and Currents.

The Existence Theorem
12.1 Weyl fields and Bianchi equations. Weyl

currents
We begin by recalling certain fundamental concepts from [C-K]. A Weyl field
on a general 4-dimensional spacetime manifold (M, g) is defined to be a tensor-
field on (M, g) with the same algebraic properties as the Weyl - or conformal
- curvature tensor, that is, in components relative to an arbitrary local frame
field, the antisymmetry in the first two as well as in the last two indices:
Wβαγδ = Wαβδγ = −Wαβγδ (12.1)
the cyclic condition:
Wα[βγδ] := Wαβγδ + Wαγδβ + Wαδβγ = 0 (12.2)
and the trace condition:
(g −1 )µν Wµανβ = 0 (12.3)
As is well known, the first two properties imply the symmetry under exchange
of the first and second pair of indices:
Wγδαβ = Wαβγδ (12.4)
∗
Given a Weyl field W we can define a right dual W as well as a left dual
∗
W . The right dual is defined as
1
∗
Wαβγδ = W µν ǫµνγδ (12.5)
2 αβ
by freezing the first pair of indices and considering W as a 2-form in the second
pair. The left dual is defined as
∗ 1 µν
Wαβγδ = W γδ ǫµναβ (12.6)
2
366
by freezing the second pair of indices and considering W as a 2-form in the first
pair. In 12.5, 12.6 ǫαβγδ are the components of the volume 4-form of (M, g)
and the indices are raised with respect to g. Now, by virtue of the algebraic
properties of W the two duals coincide:
∗
W = W∗ (12.7)
We shall thus only write ∗ W in the following. Moreover, ∗ W is also a Weyl
field. In fact, the cyclic condition for ∗ W is equivalent - modulo the other
conditions - to the trace condition for W , and vise-versa.
Given a Weyl field W and a vectorfield X, the Lie derivative with respect
to X of W , LX W , is not in general a Weyl field, because it does not in general
satisfy the trace condition 12.3. We can however define a modified Lie derivative,
L̃X W , which is a Weyl field:
(L̃X W )αβγδ = (LX W )αβγδ
1
− ( (X) π̃αµ Wµβγδ + (X) π̃βµ Wαµγδ + (X)
π̃γµ Wαβµδ + (X)
π̃δ µ Wαβγµ )
2
1
− tr (X) πWαβγδ (12.8)
8
Here (X) π̃ is the deformation tensor of X (see Chapter 8). The modified Lie
derivative commutes with duality:
L̃X ( ∗ W ) = ∗
(L̃X W ) (12.9)
The fundamental Weyl field W shall be the curvature tensor R of our solution
(M, g) of the vacuum Einstein equations. By virtue of the vanishing of the
Ricci curvature the curvature tensor reduces to Weyl tensor. The derived Weyl
fields shall be the iterated modified Lie derivatives of R with respect to the
commutation fields L, S and Oi : i = 1, 2, 3. In fact we shall consider the
following derived Weyl fields:
1st order: L̃L R, L̃S R, L̃Oi R : i = 1, 2, 3
2nd order: L̃L L̃L R, L̃S L̃S R, L̃Oj L̃Oi R : i, j = 1, 2, 3,
L̃Oi L̃L R : i = 1, 2, 3, L̃Oi L̃S R : i = 1, 2, 3 (12.10)
The homogeneous Bianchi equations for a Weyl field W are the equations:
∇[α Wβγ]δǫ = 0 (12.11)
where, as usual, we denote by [αβγ] the cyclic sum. We can write these equations
as:
DW = 0 (12.12)
to emphasize the analogy with the exterior derivative. Thus the homogeneous
Bianchi equations are seen to be analogues of the homogeneous Maxwell equa-
tions for the electromagnetic field F , a 2-form on (M, g):
dF = 0
367
However, D is not an exterior differential operator, so D2 6= 0. The equation
D2 W = 0,
a differential consequence of the homogenenous Bianchi equations, reduces in

fact to the following algebraic condition:
Rµαβγ ∗ Wναβγ − Rναβγ ∗ Wµαβγ = 0 (12.13)
Here R is the curvature tensor of the underlying metric g. One may inquire
about the analogues of the other Maxwell equations, which are in general inho-
mogeneous, but which in the absence of sources read:
d ∗F = 0
The remarkable fact here, which follows from the equality 12.7, is that the
equations:
D ∗W = 0 (12.14)
are equivalent to the equations DW = 0. In components, the equations
D ∗W = 0
read:
∇α Wαβγδ = 0 (12.15)
The fundamental Weyl field, the curvature tensor R of our solution of the
vacuum Einstein equations, satisfies the homogeneous Bianchi equations, for,
these become in this case simply the Bianchi identities for the curvature. How-
ever the derived Weyl fields do not satisfy the homogeneous Bianchi equations
but rather equations of the form:
∇α Wαβγδ = Jβγδ (12.16)
We call these equations inhomogeneous Bianchi equations and the right hand
side J we call Weyl current. In fact, given a Weyl field W equation 12.16 may
be thought of as simply the definition of the corresponding Weyl current J. The
algebraic properties of a Weyl current follow from those of a Weyl field and are
the following. The antisymmetry in the last two indices:
Jβδγ = −Jβγδ (12.17)
the cyclic condition:
J[βγδ] := Jβγδ + Jγδβ + Jδβγ = 0 (12.18)
and the trace condition:
(trJ)γ := (g −1 )βδ Jβγδ = 0 (12.19)
368
The inhomogeneous Bianchi equations 12.16 are equivalent to the equations:
∇α ∗ Wαβγδ = Jβγδ
∗
(12.20)
where the dual Weyl current J ∗ is defined by:
∗ 1 µν
Jβγδ = J ǫµνγδ (12.21)
2 β
Equations 12.20 are in turn equivalent to the equations:
∇[α Wβγ]δǫ = ǫµαβγ J ∗µδǫ (12.22)
while equations 12.16 are equivalent to the equations:
∇[α ∗ Wβγ]δǫ = −ǫµαβγ J µδǫ (12.23)
Thus, all four sets of equations are equivalent.

The dual J ∗ of a Weyl current J is also a Weyl current. In fact, the cyclic
condition for J ∗ is equivalent - modulo the remaining condition - equivalent to
the trace condition for J and vice-versa.
Given a Weyl current J and a vectorfield X, the Lie derivative with respect
to X of J, LX J, is not in general a Weyl current, because it does not in gen-
eral satisfy the trace condition 12.19. We can however define a modifed Lie
derivative, L̃X J, which is a Weyl current:
(L̃X J)βγδ = (LX J)βγδ

1
− ( (X) π̃βµ Jµγδ + (X) π̃γγ Jβµδ + (X)
π̃δ µ Jβγµ )
2
1
+ tr (X) πJβγδ (12.24)
8
Again, the modified Lie derivative commutes with duality:
L̃X (J ∗ ) = (L̃X J)∗ (12.25)
We consider next the conformal properties of the Bianchi equations. If the

Weyl field W is a solution of the Bianchi equations 12.16 on (M, g) with Weyl
current J, then, for any conformal factor Ω, setting:
g ′ = Ω−2 g, W ′ = Ω−1 W, J ′ = ΩJ (12.26)
the Weyl field W ′ satisfies the Bianchi equations on (M, g ′ ) with Weyl current
J ′:
∇′α Wαβγδ
′ ′
= Jβγδ (12.27)
It follows that if f is a conformal isometry of (M, g), that is, we have:
f ∗ g = Ω2 g (12.28)
369
for some conformal factor Ω, and if the Weyl field W is a solution of the Bianchi
equations on (M, g) with Weyl current J, then the Weyl field W ′ = Ω−1 f ∗ W is a
solution of the same equations on the same manifold (M, g) with Weyl current
J ′ = Ωf ∗ J. Suppose now that X is a vectorfield generating a 1-parameter
group ft of conformal isometries of (M, g), that is, a conformal Killing field.
Then if W is a solution of the Bianchi equations on (M, g) with Weyl current
J, then the Weyl field Ω−1 ∗
t ft W is a solution of the same equations on the same
manifold (M, g) with Weyl current Ωt f ∗ J. It follows, in view of the linearity of
the Bianchi equations that the Weyl field
d −1 ∗
Ω f W = L̃X W (12.29)
dt t t t=0
is likewise a solution of the same equations with Weyl current:

d
Ωt ft∗ J = L̃X J (12.30)
dt t=0
We see that the terms −(1/8)tr (X) πW in L̃X W and (1/8)tr (X) πJ in L̃X J
come from the conformal weights Ω−1 and Ω respectively.
Off course (M, g) does not in general admit a non-trivial conformal Killing
field. The following fundamental proposition from [C-K] shows the commutation
properties of the modified Lie derivative with respect to an arbitrary vectorfield
with the Bianchi equations.
Proposition 12.1 Let W be a Weyl field and J the corresponding Weyl

current, so the Bianchi equations
∇α Wαβγδ = Jβγδ
(X)
hold. Let also X be an arbitrary vectorfield with deformation tensor π̃.
Then the derived Weyl field L̃X W satisfies the Bianchi equations
∇α (L̃X W )αβγδ = (X)
J(W )βγδ
(X)
where the derived Weyl current J(W ) is given by:
(X) 1 (X) µν
J(W )βγδ = (L̃X J)βγδ + π̃ ∇ν Wµβγδ
2
1 1
+ (X)
pµ W µβγδ + ( (X)
qµβν W µνγδ + (X)
qµγν W µβ νδ + (X)
qµδν W µβγ ν )
2 2
Here:
(X)
pβ = ∇α (X)
π̃αβ
and:
(X) (X) (X) 1 (X) (X)
qαβγ = ∇β π̃γα − ∇γ π̃βα + ( pβ gαγ − pγ gαβ )
3
(X)
The 3-covariant tensorfield q has the algebraic properties of a Weyl current.
370
12.2 Null decompositions of Weyl fields and cur-
rents
Le (eA : A = 1, 2) be an arbitrary local frame field for the Su,u . We complement
(eA : A = 1, 2) with the vectorfields e3 = L̂ , e4 = L̂ to obtain a frame field (eµ :
µ = 1, 2, 3, 4) for M , as in Chapter 1. The algebraically independent components
of an arbitrary Weyl field W on M are, as in the case of the fundamental Weyl
field R, the curvature tensor of our solution (M, g) of the vacuum Einstein
equations, the trace-free symmetric 2-covariant S tensorfields α(W ) and α(W )
given by:
WA3B3 = αAB (W ), WA4B4 = αAB (W ) (12.31)
the S 1-forms β(W ) and β(W ) given by:
WA334 = 2β A (W ), WA434 = 2βA (W ) (12.32)
and the functions ρ(W ) and σ(W ) given by:
W3434 = 4ρ(W ), WAB34 = 2σ(W )ǫ/AB (12.33)
The remaining components of W are expressed in terms of these by:
WA3BC = ∗ β A (W )ǫ/BC = g/AB β C (W ) − g/AC β B (W )
WA4BC = − ∗ βA (W )ǫ/BC = −g/AB βC (W ) + g/AC βB (W )
WA3B4 = −ρ(W )g/AB + σ(W )ǫ/AB
WABCD = −ρ(W )ǫ/AB /ǫ CD = −ρ(W )(g/AC g/BD − g/AD g/BC ) (12.34)
(compare with 1.178). In the case of the fundamental Weyl field R, we shall, as
hitherto, omit the reference to W .
The components of ∗ W , the dual of W , are expressed in terms of the com-
ponents of W by:
α( ∗ W ) = ∗
α(W ), α( ∗ W ) = − ∗ α(W )
β( ∗ W ) = ∗
β(W ), β( ∗ W ) = − ∗ β(W )
ρ( ∗ W ) = σ(W ), σ( ∗ W ) = −ρ(W ) (12.35)
Here ∗ α, ∗ α, ∗ β, ∗ β denote the left duals relative to the Su,u of α, α, β, β
respectively. The left dual ∗ θ of a trace-free symmetric 2-covariant S tensorfield
θ is given by:
∗
θAB = /ǫ AC θCB (12.36)
We note that ∗ θ is also a trace-free symmetric 2-covariant S tensorfield. In
fact, the symmetry condition for ∗ θ is equivalent to the trace condition for θ
and vice-versa.
We shall now analyze the relationship, for a commutation field Y and Weyl
field W , between the null components of L̃Y W and L /Y applied to the null
components of W . We shall need the following lemma. Let us define:

(Y ) Y log Ω : for Y = L, Oi : i = 1, 2, 3
ν= (12.37)
1 + Y log Ω : for Y = S
371
Lemma 12.1 For all five commutation fields L, Oi : i = 1, 2, 3, S we have:
[Y, eA ] = Π[Y, eA ]
(Y )
[Y, e3 ] = − mA e A − (Y )
νe3
(Y ) A (Y )
[Y, e4 ] = − m eA − νe4
Proof: According to Lemma 1.1 and the commutation formula 1.91 we have:
[L, eA ] = Π[L, eA ]
[L, e3 ] = −4Ωζ A eA − (L log Ω)e3
[L, e4 ] = −(L log Ω)e4 (12.38)
and:
[L, eA ] = Π[L, eA ]
[L, e3 ] = −(LΩ)e3
[L, e4 ] = 4Ωζ A eA − (L log Ω)e4 (12.39)
In the case Y = L the lemma follows immediately from 12.38 in conjunction

with the table 8.21. In the case Y = S the lemma follows from 12.38, 12.39 and
the facts that
Lu = 1, Lu = eA u = 0; Lu = 1, Lu = eA u = 0
in conjunction with table 8.30. Finally, in the case Y = Oi we have, since both
Oi and eA are S-tangential vectorfields:
[Oi , eA ] = Π[Oi , eA ] (12.40)
Also, according to the first of 8.40:
[Oi , e3 ] = −(Oi log Ω)e3 (12.41)
and according to 8.64:
[Oi , e4 ] = −Ω−1 Zi − (Oi log Ω)e4 (12.42)
The lemma in the case Y = Oi then follows from 12.40 - 12.42 in conjunction
with 8.140, 8.141.
Proposition 12.2 Let Y be any of the five commutation fields L, Oi : i =

1, 2, 3, S and let W be an arbitrary Weyl field. Then the null components of
L̃Y W are given by:
(Y ) 1 (Y )
αAB (L̃Y W ) = (L̂
/Y α(W ))AB + ναAB (W ) − jαAB (W )
4
(Y ) ˆ
+( m⊗β(W ))AB
372
(Y ) 1 (Y )
αAB (L̃Y W ) = (L̂
/Y α(W ))AB + ναAB (W ) − jαAB (W )
4
(Y ) ˆ
−( m⊗β(W ))AB
(Y ) 1 (Y ) 1 (Y )
βA (L̃Y W ) = (L
/Y β(W ))A + 2 νβA (W ) + jβA (W ) − îAB βB (W )
4 2
1 (Y ) 3 3
+ mB αAB (W ) + (Y )
mA ρ(W ) + ∗(Y )
mA σ(W )
4 4 4
(Y ) 1 (Y ) 1 (Y )
/Y β(W ))A + 2
β A (L̃Y W ) = (L νβ A (W ) + jβ A (W ) − îAB β B (W )
4 2
1 (Y ) 3 (Y ) 3
− mB αAB (W ) − mA ρ(W ) + ∗(Y )
mA σ(W )
4 4 4
(Y ) 3 (Y )
ρ(L̃Y W ) = Y ρ(W ) + 3 νρ(W ) + jρ(W )
4
1 1
− (Y ) mA β A (W ) + (Y ) mA βA (W )
2 2
3
σ(L̃Y W ) = Y σ(W ) + 3 (Y ) νσ(W ) + (Y ) jσ(W )
4
1 ∗(Y ) A 1 ∗(Y ) A
+ m β A (W ) + m βA (W )
2 2
Proof: From the last of 8.20, 8.29 and from 8.48 we have:
(Y ) (Y )
π34 = −4 ν (12.43)
for all five commutation fields Y . It follows that:

(Y ) (Y ) (Y )
tr π=2 j+8 ν (12.44)
and:
(Y ) (Y ) (Y )
tr π
/=2 j+4 ν (12.45)
for all five commutation fields Y .
Using Lemma 12.1, the fact that
1 (Y )
L
/Y /ǫ = tr π
//ǫ (12.46)
2
and 12.45, we derive the following formulas:
(Y )
(LY W )A4B4 = (L
/Y α(W ))AB + 2 ναAB (W )
(Y ) (Y ) (Y )
+ mA βB (W ) + mB βA (W ) − 2 mC βC (W )g/AB
(LY W )A3B3 /Y α(W ))AB + 2 (Y ) ναAB (W )
= (L
(Y ) (Y ) (Y )
− mA β B (W ) − mB β A (W ) + 2 mC β C (W )g/AB
1
/Y β(W ))A + 3 (Y ) νβA (W )
(LY W )A434 = (L
2
1 1 3
+ (Y ) mB αAB (W ) + (Y ) mA ρ(W ) + ∗(Y )
mA σ(W )
2 2 2
373
1
(LY W )A334 = (L/Y β(W ))A + 3 (Y ) νβ A (W )
2
1 1 3
− (Y ) mB αAB (W ) − (Y ) mA ρ(W ) + ∗(Y )
mA σ(W )
2 2 2
1 (Y )
(LY W )3434 = Y ρ(W ) + 4 νρ(W )
4
+ (Y ) mA βA (W ) − (Y ) mA β A (W )
1 AB
/ǫ (LY W )AB34 = Y σ(W ) + 4 (Y ) νσ(W ) + (Y ) jσ(W )
4
1 1
+ ∗(Y ) mA βA (W ) + ∗(Y ) mA β A (W ) (12.47)
2 2
Consider next the expression
(Y )
[W ]αβγδ = (Y )
π̃αµ Wµβγδ + (Y )
π̃βµ Wαµγδ + (Y )
π̃γµ Wαβµδ + (Y )
π̃δ µ Wαβγµ
(12.48)
in formula 12.8. By a straightforward computation we find:
(Y ) (Y )
[W ]A4B4 = îAC αCB (W ) + (Y )
îBC αCA (W ) − 2 (Y )
mC βC (W )g/AB
(Y ) (Y )
[W ]A3B3 = îAC αCB (W ) + (Y )
îBC αCA (W ) + 2 (Y )
mC β C (W )g/AB
1 (Y ) (Y ) (Y )
[W ]A434 = − jβA (W ) + îAB βB (W )
2
1 (Y ) 1 3
+ mB αAB (W ) − (Y )
mA ρ(W ) + ∗(Y )
mB σ(W )
2 2 2
1 (Y ) (Y ) (Y )
[W ]A334 = − jβ A (W ) + îAB β B (W )
2
1 (Y ) 1 3
− mB αAB (W ) + (Y )
mA ρ(W ) + ∗(Y )
mA σ(W )
2 2 2
1 (Y )
[W ]3434 = −2 (Y ) jρ(W ) + (Y )
mA βA (W ) − (Y )
mA β A (W )
4
1 AB (Y )
/ǫ [W ]AB34 = 0 (12.49)
4
In view of formulas 12.47, 12.49, 12.43, the proposition follows through for-
mula 12.8, if we also take into account the fact that by virtue of the identity
1.187 for any trace-free symmatric 2-covariant S tensorfield θ we have:
1 (Y ) 1 1
(L
/Y θ)AB − îAC θCB − (Y )
îBC θCA = (L
/Y θ)AB − ( (Y )
î, θ)g/AB = (L̂
/Y θ)AB
2 2 2
(12.50)
The algebraically independent components of an arbitrary Weyl current J

on M , or more generally of a 3-covariant tensorfield on M with the algebraic
properties 12.17 - 12.19 of a Weyl current, are the S 1-forms Ξ(J) and Ξ(J)
given by:
1 1
J33A = ΞA (J), J44A = ΞA (J) (12.51)
2 2
374
the trace-free symmetric 2-covariant S tensorfields Θ(J) and Θ(J) to be defined
below, the functions Λ(J), K(J) and Λ(J), K(J) given by:
1 1
J343 = Λ(J), J434 = Λ(J) (12.52)
4 4
and:
1 AB 1 AB
/ǫ J3AB = K(J), /ǫ J4AB = K(J) (12.53)
4 4
and the S 1-forms I(J) and I(J) given by:
1 1
J43A = I A (J), J34A = IA (J) (12.54)
2 2
To define the components Θ(J) and Θ(J) we consider the 2-covariant S tensor-
fields J/3 and J/4 given by:
JA3B = (J/3 )AB , JA4B = (J/4 )AB (12.55)
Then Θ(J) and Θ(J) are the trace-free symmetric parts of J/3 and J/4 respec-
tively:
Θ(J)AB = 21 ((J/3 )AB + (J/3 )BA − g/AB trJ/3 )
Θ(J)AB = 21 ((J/4 )AB + (J/4 )BA − g/AB trJ/4 ) (12.56)
Note that by the trace condition 12.19 we have:
trJ/3 = −2Λ(J), trJ/4 = −2Λ(J) (12.57)
while by the cyclic condition 12.18 the antisymmetric parts of J/3 and J/4 are
given by:
1 1
((J/3 )AB − (J/3 )BA ) = K(J)ǫ/AB , ((J/4 )AB − (J/4 )BA ) = K(J)ǫ/AB (12.58)
2 2
Therefore the 2-covariant S tensorfields J/3 and J/4 are expressed as:
J/3 = Θ(J) − Λ(J)g/ + K(J)ǫ/
J/4 = Θ(J) − Λ(J)g/ + K(J)ǫ/ (12.59)
The remaining components of J are axpressed in terms of the above components
by:
JA34 = 2I A (J) − 2IA (J), JA43 = 2IA (J) − 2I A (J)
∗ ∗
JABC = ( IA (J) + I A (J))ǫ/BC (12.60)
∗
The components of J , the dual of J, are expressed in terms of the compo-
nents of J by:
Ξ(J ∗ ) = ∗
Ξ(J), Ξ(J ∗ ) = − ∗ Ξ(J)
Θ(J ∗ ) = ∗ Θ(J), Θ(J ∗ ) = − ∗ Θ(J)
Λ(J ∗ ) = −K(J), Λ(J ∗ ) = K(J)
K(J ∗ ) = Λ(J), K(J ∗ ) = −Λ(J)
I(J ∗ ) = ∗ I(J), I(J ∗ ) = − ∗ I(J) (12.61)
375
The following proposition specifies the relationship, for a commutation field
Y and Weyl curent J, between the null components of L̃Y J and L /Y applied to
the null components of J.
Proposition 12.3 Let Y be any of the five commutation fields L, Oi : i =

1, 2, 3, S and let J be an arbitrary Weyl current, or more generally an arbitrary
3-covariant tensorfield with the algebraic properties of a Weyl current. Then
the null components of L̃Y J are given by:
(Y ) 1 (Y ) 1 (Y )
ΞA (L̃Y J) = (L
/Y Ξ(J))A + 3 νΞA (J) + jΞA (J) − îAB ΞB (J)
2 2
1 (Y ) 3 3
+ mB ΘAB (J) − (Y )
mA Λ(J) − ∗(Y )
mA K(J)
4 4 4
(Y ) 1 (Y ) 1 (Y )
ΞA (L̃Y J) = (L
/Y Ξ(J))A + 3 νΞA (J) + jΞA (J) − îAB ΞB (J)
2 2
1 (Y ) B 3 3
+ m ΘAB (J) − (Y ) mA Λ(J) − ∗(Y ) mA K(J)
4 4 4
ΘAB (L̃Y J) = (L̂ /Y Θ(J))AB + 2 (Y ) νΘAB (J)
1 3 (Y )
+ ( (Y ) m⊗Ξ(J))
ˆ AB + ( ˆ
m⊗I(J)) AB
4 4
ΘAB (L̃Y J) = (L̂ /Y Θ(J))AB + 2 (Y ) νΘAB (J)
1 3 (Y )
+ ( (Y ) m⊗Ξ(J))
ˆ AB + ( ˆ
m⊗I(J)) AB
4 4
Λ(L̃Y J) = Y Λ(J) + 4 (Y ) νΛ(J) + (Y ) jΛ(J)
1 1 1
− ( (Y ) m, Ξ(J)) + ( (Y ) m, I(J)) − ( (Y ) m, I(J))
4 2 4
(Y ) (Y )
Λ(L̃Y J) = Y Λ(J) + 4 νΛ(J) + jΛ(J)
1 (Y ) 1 (Y ) 1
− ( m, Ξ(J)) + ( m, I(J)) − ( (Y ) m, I(J))
4 2 4
(Y ) (Y )
K(L̃Y J) = Y K(J) + 4 νK(J) + jK(J)
1 ∗(Y ) 1 ∗(Y ) 1
− ( m, Ξ(J)) − ( m, I(J)) − ( ∗(Y ) m, I(J))
4 2 4
(Y ) (Y )
K(L̃Y J) = Y K(J) + 4 νK(J) + jK(J)
1 ∗(Y ) 1 ∗(Y ) 1
− ( m, Ξ(J)) − ( m, I(J)) − ( ∗(Y ) m, I(J))
4 2 4
(Y ) 1 (Y ) 1
IA (L̃Y J) = (L/Y I(J))A + 3 νIA (J) + jIA (J) − (Y ) îAB IB (J)
2 2
1 1 1
+ (Y ) mB ΘAB (J) − (Y ) mA Λ(J) − ∗(Y ) mA K(J)
4 4 4
1 (Y ) 1 ∗(Y )
+ mA Λ(J) − mA K(J)
2 2
1 1
/Y I(J))A + 3 (Y ) νI A (J) + (Y ) jI A (J) − (Y ) îAB I B (J)
I A (L̃Y J) = (L
2 2
376
1 (Y ) 1 1
+ mB ΘAB (J) − (Y )
mA Λ(J) − ∗(Y )
mA K(J)
4 4 4
1 (Y ) 1 ∗(Y )
+ mA Λ(J) − mA K(J)
2 2
The proof is similar to that of Proposition 12.2.
Proposition 12.4 The following system of equations constitute the inhomo-

geneous Bianchi equations 12.16:
1
Ω−1 (D̂α(W ) + 2ωα(W )) − trχα(W ) − ∇ˆ
/⊗β(W ˆ
) − (4η + ζ)⊗β(W )
2
∗
+3χ̂ρ(W ) + 3 χ̂σ(W ) = −2Θ(J)
1
Ω−1 (D̂α(W ) + 2ωα(W )) − trχα(W ) + ∇ˆ
/⊗β(W ˆ
) + (4η − ζ)⊗β(W )
2
+3χ̂ρ(W ) − 3 ∗ χ̂σ(W ) = −2Θ(J)
Ω−1 (Dβ(W ) − ωβ(W )) − χ♯ · β(W ) + 2trχβ(W ) − div
/ α(W )
−(η ♯ + 2ζ ♯ ) · α(W ) = 2Ξ(J)
Ω−1 (Dβ(W ) − ωβ(W )) − χ♯ · β(W ) + 2trχβ(W ) + div
/ α(W )
+(η ♯ − 2ζ ♯ ) · α(W ) = −2Ξ(J)
Ω−1 (Dβ(W ) + ωβ(W )) − χ♯ · β(W ) + trχβ(W ) − /dρ(W ) − ∗/dσ(W )
−3ηρ(W ) − 3 ∗ ησ(W ) − 2χ̂♯ · β(W ) = −2I(J)
Ω−1 (Dβ(W ) + ωβ(W )) − χ♯ · β + trχβ(W ) + /dρ(W ) − ∗
/dσ(W )
+3ηρ(W ) − 3 ∗ ησ(W ) − 2χ̂♯ · β(W ) = 2I A (J)
3 1
Ω−1 Dρ(W ) + trχρ(W ) − div
/ β(W ) − (2η + ζ, β(W )) + (χ̂, α(W ))
2 2
= −2Λ(J)
3 1
Ω−1 Dρ(W ) + trχρ(W ) + div
/ β(W ) + (2η − ζ, β(W )) + (χ̂, α(W ))
2 2
= −2Λ(J)
3 1
Ω−1 Dσ(W ) + trχσ(W ) + curl
/ β(W ) + (2η + ζ) ∧ β(W ) − χ̂ ∧ α(W )
2 2
= −2K(W )
3 1
Ω−1 Dσ(W ) + trχσ(W ) + curl
/ β(W ) + (2η − ζ) ∧ β(W ) + χ̂ ∧ α(W )
2 2
= 2K(W )
The proof is similar to that of Proposition 1.2.
377
12.3 The Bel-Robinson tensor. The energy-momentum
density vectorfields
Given a Weyl field W , the Bel-Robinson tensor associated to W is the 4-
covariant tensorfield:
Q(W )αβγδ = Wαργσ Wβ ρδ σ + ∗

Wαργσ ∗ Wβ ρδ σ (12.62)
This is the analogue of the Maxwell energy-momentum stress tensor T (F ) as-

sociated to an electromagnetic field F :
T (F )αβ = Fαρ Fβ ρ + ∗
Fαρ ∗ Fβ ρ
We recall the following basic facts from [C-K].
Proposition 12.5 Given an arbitrary Weyl field W , the associated Bel-

Robinson tensor Q(W ) has the following algebraic properties:
1. Q(W ) is symmetric and trace-free in all pairs of indices.
2. Q(W )(X1 , X2 , X3 , X4 ) is non-negative for any tetrad X1 , X2 , X3 , X4 of
future-directed non-spacelike vectors at a point.
Proposition 12.6 Given a Weyl field W satisfying the Bianchi equations

12.16,
(divQ(W ))βγδ = ∇α Qαβγδ
the divergence of the associated Bel-Robinson tensor Q(W ), a 3-covariant ten-
sorfield which is symmetric and trace-free in all pairs of indices, is given by:
(divQ(W ))βγδ = Wβ µδ ν Jµγν + Wβ µγ ν Jµδν + ∗

Wβ µδ ν Jµγν
∗
+ ∗
Wβ µγ ν Jµδν
∗
We shall need expressions for those components of Q(W ) relative to the

frame field (eµ : µ = 1, 2, 3, 4) for which at least two of the indices are from the
set {3, 4}. These are given by the following lemma from [C-K].
Lemma 12.2 We have:
Q(W )3333 = 2|α(W )|2 , Q(W )4444 = 2|α(W )|2

Q(W )4333 = 4|β(W )|2 , Q(W )3444 = 4|β(W )|2
Q(W )3344 = 4(ρ(W )2 + σ(W )2 )
Also:
Q(W )A333 = −4αAB (W )β B (W )

Q(W )A444 = 4αAB (W )β B (W )
Q(W )A433 = −4ρ(W )β A (W ) − 4σ(W ) ∗ β A (W )
378
Q(W )A344 = 4ρ(W )βA (W ) − 4σ(W ) ∗ βA (W )
Q(W )AB33 = 2g/AB |β(W )|2 + 2ρ(W )αAB (W ) + 2σ(W ) ∗ αAB (W )
Q(W )AB44 = 2g/AB |β(W )|2 + 2ρ(W )αAB (W ) − 2σ(W ) ∗ αAB (W )
ˆ
Q(W )AB34 = −2(β(W )⊗β(W ))AB + 2g/AB (ρ(W )2 + σ(W )2 )
We shall also need expressions for those components of divQ(W ) for which
all the indices are from the set {3, 4}. These are given by the following lemma
from [C-K], which follows directly from Proposition 12.6.
Lemma 12.3 We have:
(divQ(W ))333 = 4(α(W ), Θ(J)) + 8(β(W ), Ξ(J))

(divQ(W ))444 = 4(α(W ), Θ(J)) − 8(β(W ), Ξ(J))
(div(Q(W ))433 = 8ρ(W )Λ(J) − 8σ(W )K(J) − 8(β(W ), I(J))
(div(Q(W ))344 = 8ρ(W )Λ(J) + 8σ(W )K(J) + 8(β(W ), I(J))
Given now a Weyl field W and three future directed non-spacelike vector-
fields X, Y, Z, we define the energy-momentum density vectorfield P (W ; X, Y, Z)
associated to W and to the triplet X, Y, Z by:
P (W ; X, Y, Z)α = −Q(W )αβγδ X β Y γ Z δ (12.63)
Then by the first statement of Proposition 12.5,
divP (W ; X, Y, Z) = ∇α P (W ; X, Y, Z)α ,
the divergence of the energy-momentum density vectorfield P (W ; X, Y, Z), is

given by:
divP (W ; X, Y, Z) = −(divQ(W ))(X, Y, Z) (12.64)

1
− Q(W )αβγδ ( (X) π̃ αβ Y γ Z δ + (Y ) αβ
π̃ XγZδ + (Z) αβ
π̃ XγY δ)
2
12.4 The divergence theorem in spacetime

Consider the equation
divP = τ (12.65)
for an arbitrary vectorfield P and function τ on a general spacetime manifold
(M, g). In arbitrary local coordinates this equation reads:
1 ∂ p
√ ( −detgP µ ) = τ (12.66)
−detg ∂xµ
379
Consider then equation 12.65 on our spacetime manifold (M ′ , g). We express
this equation in canonical coordinates (u, u; ϑA : A = 1, 2) (see Chapter 1,
Section 4). Expanding P in the associated coordinate frame field,

∂ ∂ ∂
, ; : A = 1, 2 ,
∂u ∂u ∂ϑA
∂ ∂ ∂
P = Pu + Pu + (P ϑ )A A (12.67)
∂u ∂u ∂ϑ
the equation takes the form (see 1.205, 1.206):

1 ∂ p ∂ p
p (2Ω2 detg/P u ) + (2Ω2 detg/P u ) + div
/ (2Ω2 M ) = 2Ω2 τ
detg/ ∂u ∂u
(12.68)
where M is the S-tangential vectorfield:
∂
M = (P ϑ )A (12.69)
∂ϑA
We integrate this equation on the Su,u with respect to the measure dµ/g to obtain
the equation:
! ! Z
∂ ∂
Z Z
2 u 2 u
2Ω P dµ/g + 2Ω P dµ/g = 2Ω2 τ dµ/g (12.70)
∂u Su,u ∂u Su,u Su,u
Consider any (u1 , u1 ) ∈ D′ . We integrate equation 12.70 with respect to (u, u)

on the parameter domain
D1 = [0, u1 ] × [u0 , u1 ] ⊂ D′
(see 3.45), to obtain, under the assumption that P u vanishes on C 0 ,

Z
E u1 (u1 ) − E u1 (u0 ) + F u1 (u1 ) = τ dµg (12.71)
M1
Here E u1 (u1 ) is the “energy”:

!
Z Z u1 Z
u1 2 u 2 u
E (u1 ) = u
2Ω P := 2Ω P dµ/g du (12.72)
Cu11 0 Su,u
and F u1 (u1 ) is the “flux”:

!
Z Z u1 Z
u1 2 u u
F (u1 ) = 2Ω P := 2ΩP dµ/g du (12.73)
u
C u1 u0 Su,u
1
Also, M1 is the subdomain of M ′ corresponding to D1 :

[
M1 = Su,u
(u,u)∈D1
380
(see 3.46), and we have (see 1.205):
Z Z Z Z !
2
τ dµg = 2Ω τ dµ/g dudu (12.74)
M1 D1 Su,u
Now, the vectorfield P can also be expanded in the frame field (eµ : µ =
1, 2, 3, 4) where
∂
eA = : A = 1, 2 (12.75)
∂ϑA
We thus have:
P = P µ eµ (12.76)
Now by 1.197 and 1.11:

∂ ∂ ∂
e3 = Ω−1 , e4 = Ω−1 + bA A (12.77)
∂u ∂u ∂ϑ
Substituting in 12.76 and comparing coefficients with the expansion 12.67 yields:
P u = Ω−1 P 3 , P u = Ω−1 P 4 (12.78)
and:
(P ϑ )A = P A + Ω−1 bA P 4 (12.79)
In conclusion, we have arrived at the following lemma.
Lemma 12.4 Let P be a vectorfield and τ a function defined on (M ′ , g) and

satisfying the equation:
divP = τ
Let also P 4 vanish on C 0 . Then, for every (u1 , u1 ) ∈ D′ we have:
Z
E u1 (u1 ) − E u1 (u0 ) + F u1 (u1 ) = τ dµg
M1
where E u1 (u1 ) is the “energy”:

!
Z Z u1 Z
u1 3 3
E (u1 ) = u
2ΩP := 2ΩP dµ/g du
Cu11 0 Su,u
and F u1 (u1 ) is the “flux”:

!
Z Z u1 Z
F u1 (u1 ) = 2ΩP 4 := 2ΩP 4 dµ/g du
u
C u1 u0 Su,u
1
381
12.5 The energies and fluxes. The quantity P2
We now define the energies and fluxes on which our entire approach is based.
We consider the energy-momentum density vectorfield P (W ; X, Y, Z), defined
by 12.63 with the multiplier fields L, K in the role of the future directed non-
spacelike vectorfields X, Y, Z. Taking in 12.63 the fundamental Weyl field W =
R we consider the energy-momentum density vectorfields:
P (R; L, L, L), P (R; K, L, L, ), P (R; K, K, L), P (R; K, K, K) (12.80)
Considering, for each u ∈ [u0 , c∗ ) the entire Cu in M ′ we define the correspond-

(n)
ing energies E 0 (u) : n = 0, 1, 2, 3:
(0)
Z
E 0 (u) = 2ΩP 3 (R; L, L, L)
Cu
(1)
Z
E0 (u) = 2ΩP 3 (R; K, L, L)
Cu
(2)
Z
E0 (u) = 2ΩP 3 (R; K, K, L)
Cu
(3)
Z
E0 (u) = 2ΩP 3 (R; K, K, K) (12.81)
Cu
Also, considering for each u ∈ [0, δ) the entire C u in M ′ we define the corre-
(n)
sponding fluxes F 0 (u) : n = 0, 1, 2, 3:
(0)
Z
F 0 (u) = 2ΩP 4 (R; L, L, L)
Cu
(1)
Z
F0 (u) = 2ΩP 4 (R; K, L, L)
Cu
(2)
Z
F0 (u) = 2ΩP 4 (R; K, K, L)
Cu
(3)
Z
F0 (u) = 2ΩP 4 (R; K, K, K) (12.82)
Cu
According to the definitions 12.63 and 8.1, and Lemma 12.2, we have:
(0)
Z
E 0 (u) = 2Ω4 |α|2
Cu
(1)
Z
E0 (u) = 4Ω4 |u|2 |β|2
Cu
(2)
Z
E0 (u) = 4Ω4 |u|4 (|ρ|2 + |σ|2 )
Cu
382
(3)
Z
E0 (u) = 4Ω4 |u|6 |β|2 (12.83)
Cu
Also, we have: Z
(3)
F0 (u) = 2Ω4 |u|6 |α|2 (12.84)
Cu
The above are the 0th order energies and fluxes.

Next, we consider the 1st order energies and fluxes. Taking in 12.63 the
Weyl field W = L̃L R we consider the energy-momentum density vectorfields:
P (L̃L R; L, L, L), P (L̃L R; K, L, L), P (L̃L R; K, K, L), P (L̃L R; K, K, K)
(12.85)
(n)
We then define the corresponding energies (L) E (u) : n = 0, 1, 2, 3:
(0)
Z
(L)
E (u) = 2ΩP 3 (L̃L R; L, L, L)
Cu
(1)
Z
(L)
E (u) = 2ΩP 3 (L̃L R; K, L, L)
Cu
(2)
Z
(L)
E (u) = 2ΩP 3 (L̃L R; K, K, L)
Cu
(3)
Z
(L)
E (u) = 2ΩP 3 (L̃L R; K, K, K) (12.86)
Cu
(n)
(L)
and fluxes F (u) : n = 0, 1, 2, 3:
(0)
Z
(L)
F (u) = 2ΩP 4 (L̃L R; L, L, L)
Cu
(1)
Z
(L)
F (u) = 2ΩP 4 (L̃L R; K, L, L)
Cu
(2)
Z
(L)
F (u) = 2ΩP 4 (L̃L R; K, K, L)
Cu
(3)
Z
(L)
F (u) = 2ΩP 4 (L̃L R; K, K, K) (12.87)
Cu
(0)
Z
(L)
E (u) = 2Ω4 |α(L̃L R)|2
Cu
(1)
Z
(L)
E (u) = 4Ω4 |u|2 |β(L̃L R)|2
Cu
(2)
Z
(L)
E (u) = 4Ω4 |u|4 (|ρ(L̃L R)|2 + |σ(L̃L R)|2 )
Cu
383
(3)
Z
(L)
E (u) = 4Ω4 |u|6 |β(L̃L R)|2 (12.88)
Cu
Also, we have: Z
(3)
(L)
F (u) = 2Ω4 |u|6 |α(L̃L R)|2 (12.89)
Cu
Taking in 12.63 the Weyl fields W = L̃Oi R : i = 1, 2, 3 we consider the

energy-momentum density vectorfields:
P (L̃Oi R; L, L, L), P (L̃Oi R; K, L, L), P (L̃Oi R; K, K, L), P (L̃Oi R; K, K, K)

: i = 1, 2, 3 (12.90)
(n)
We then define the corresponding energies (O) E (u) : n = 0, 1, 2, 3:
(0)
Z X
(O)
E (u) = 2Ω P 3 (L̃Oi R; L, L, L)
Cu i
(1)
Z X
(O)
E (u) = 2Ω P 3 (L̃Oi R; K, L, L)
Cu i
(2)
Z X
(O)
E (u) = 2Ω P 3 (L̃Oi R; K, K, L)
Cu i
(3)
Z X
(O)
E (u) = 2Ω P 3 (L̃Oi R; K, K, K) (12.91)
Cu i
(n)
(O)
and fluxes F (u) : n = 0, 1, 2, 3:
(0)
Z X
(O)
F (u) = 2Ω P 4 (L̃Oi R; L, L, L)
Cu i
(1)
Z X
(O)
F (u) = 2Ω P 4 (L̃Oi R; K, L, L)
Cu i
(2)
Z X
(O)
F (u) = 2Ω P 4 (L̃Oi R; K, K, L)
Cu i
(3)
Z X
(O)
F (u) = 2Ω P 4 (L̃Oi R; K, K, K) (12.92)
Cu i
(0)
Z X
(O)
E (u) = 2Ω4 |α(L̃Oi R)|2
Cu i
(1)
Z X
(O)
E (u) = 4Ω4 |u|2 |β(L̃Oi R)|2
Cu i
384
(2)
Z X
(O)
E (u) = 4Ω4 |u|4 (|ρ(L̃Oi R)|2 + |σ(L̃Oi R)|2 )
Cu i
(3)
Z X
(O)
E (u) = 4Ω4 |u|6 |β(L̃Oi R)|2 (12.93)
Cu i
Also, we have: Z
(3) X
(O)
F (u) = 2Ω4 |u|6 |α(L̃Oi R)|2 (12.94)
Cu i
Taking in 12.63 the Weyl field W = L̃S R we consider the energy-momentum

density vectorfield:
P (L̃S R; K, K, K) (12.95)
We then define the corresponding energy:
(3)
Z
(S)
E (u) = 2ΩP 3 (L̃S R; K, K, K) (12.96)
Cu
and flux: Z
(3)
(S)
F (u) = 2ΩP 4 (L̃S R; K, K, K) (12.97)
Cu
(3)
Z
(S)
E (u) = 4Ω4 |u|6 |β(L̃S R)|2 (12.98)
Cu
and: Z
(3)
(S)
F (u) = 2Ω4 |u|6 |α(L̃S R)|2 (12.99)
Cu
The above are the 1st order energies and fluxes.

Finally, we consider the 2nd order energies and fluxes. Taking in 12.63 the
Weyl field W = L̃L L̃L R we consider the energy-momentum density vectorfields:
P (L̃L L̃L R; L, L, L), P (L̃L L̃L R; K, L, L),
P (L̃L L̃L R; K, K, L), P (L̃L L̃L R; K, K, K) (12.100)
(n)
We then define the corresponding energies (LL) E : n = 0, 1, 2, 3:
(0)
Z
(LL)
E (u) = 2ΩP 3 (L̃L L̃L R; L, L, L)
Cu
(1)
Z
(LL)
E (u) = 2ΩP 3 (L̃L L̃L R; K, L, L)
Cu
(2)
Z
(LL)
E (u) = 2ΩP 3 (L̃L L̃L R; K, K, L)
Cu
(3)
Z
(LL)
E (u) = 2ΩP 3 (L̃L L̃L R; K, K, K) (12.101)
Cu
385
(n)
(LL)
and fluxes F : n = 0, 1, 2, 3:
(0)
Z
(LL)
F (u) = 2ΩP 4 (L̃L L̃L R; L, L, L)
Cu
(1)
Z
(LL)
F (u) = 2ΩP 4 (L̃L L̃L R; K, L, L)
Cu
(2)
Z
(LL)
F (u) = 2ΩP 4 (L̃L L̃L R; K, K, L)
Cu
(3)
Z
(LL)
F (u) = 2ΩP 4 (L̃L L̃L R; K, K, K) (12.102)
Cu
(0)
Z
(LL)
E (u) = 2Ω4 |α(L̃L L̃L R)|2
Cu
(1)
Z
(LL)
E (u) = 4Ω4 |u|2 |β(L̃L L̃L R)|2
Cu
(2)
Z
(LL)
E (u) = 4Ω4 |u|4 (|ρ(L̃L L̃L R)|2 + |σ(L̃L L̃L R)|2 )
Cu
(3)
Z
(LL)
E (u) = 4Ω4 |u|6 |β(L̃L L̃L R)|2 (12.103)
Cu
Also, we have: Z
(3)
(LL)
F (u) = 2Ω4 |u|6 |α(L̃L L̃L R)|2 (12.104)
Cu
Taking in 12.63 the Weyl fields W = L̃Oi L̃L R : i = 1, 2, 3 we consider the

energy-momentum density vectorfields:
P (L̃Oi L̃L R; L, L, L), P (L̃Oi L̃L R; K, L, L)

P (L̃Oi L̃L R; K, K, L), P (L̃Oi L̃L R; K, K, K)
: i = 1, 2, 3 (12.105)
(n)
We then define the corresponding energies (OL) E (u) : n = 0, 1, 2, 3:
(0)
Z X
(OL)
E (u) = 2Ω P 3 (L̃Oi L̃L R; L, L, L)
Cu i
(1)
Z X
(OL)
E (u) = 2Ω P 3 (L̃Oi L̃L R; K, L, L)
Cu i
(2)
Z X
(OL)
E (u) = 2Ω P 3 (L̃Oi L̃L R; K, K, L)
Cu i
386
(3)
Z X
(OL)
E (u) = 2Ω P 3 (L̃Oi L̃L R; K, K, K) (12.106)
Cu i
(n)
(OL)
and fluxes F (u) : n = 0, 1, 2, 3:
(0)
Z X
(OL)
F (u) = 2Ω P 4 (L̃Oi L̃L R; L, L, L)
Cu i
(1)
Z X
(OL)
F (u) = 2Ω P 4 (L̃Oi L̃L R; K, L, L)
Cu i
(2)
Z X
(OL)
F (u) = 2Ω P 4 (L̃Oi L̃L R; K, K, L)
Cu i
(3)
Z X
(OL)
F (u) = 2Ω P 4 (L̃Oi L̃L R; K, K, K) (12.107)
Cu i
(0)
Z X
(OL)
E (u) = 2Ω4 |α(L̃Oi L̃L R)|2
Cu i
(1)
Z X
(OL)
E (u) = 4Ω4 |u|2 |β(L̃Oi L̃L R)|2
Cu i
(2)
Z X
(OL)
E (u) = 4Ω4 |u|4 (|ρ(L̃Oi L̃L R)|2 + |σ(L̃Oi L̃L R)|2 )
Cu i
(3)
Z X
(OL)
E (u) = 4Ω4 |u|6 |β(L̃Oi L̃L R)|2 (12.108)
Cu i
Also, we have:
(3)
Z X
(OL)
F (u) = 2Ω4 |u|6 |α(L̃Oi L̃L R)|2 (12.109)
Cu i
Taking in 12.63 the Weyl fields W = L̃Oj L̃Oi R : i, j = 1, 2, 3 we consider

the energy-momentum density vectorfields:
P (L̃Oj L̃Oi R; L, L, L), P (L̃Oj L̃Oi R; K, L, L)

P (L̃Oj L̃Oi R; K, K, L), P (L̃Oj L̃Oi R; K, K, K)
: i, j = 1, 2, 3 (12.110)
(n)
We then define the corresponding energies (OO) E (u) : i, j = 1, 2, 3:
(0)
Z X
(OO)
E (u) = 2Ω P 3 (L̃Oj L̃Oi R; L, L, L)
Cu i,j
387
(1)
Z X
(OO)
E (u) = 2Ω P 3 (L̃Oj L̃Oi R; K, L, L)
Cu i,j
(2)
Z X
(OO)
E (u) = 2Ω P 3 (L̃Oj L̃Oi R; K, K, L)
Cu i,j
(3)
Z X
(OO)
E (u) = 2Ω P 3 (L̃Oj L̃Oi R; K, K, K) (12.111)
Cu i,j
(n)
(OO)
and fluxes F (u) : n = 0, 1, 2, 3:
(0)
Z X
(OO)
F (u) = P 4 (L̃Oj L̃Oi R; L, L, L)
C u i,j
(1)
Z X
(OO)
F (u) = P 4 (L̃Oj L̃Oi R; K, L, L)
C u i,j
(2)
Z X
(OO)
F (u) = P 4 (L̃Oj L̃Oi R; K, K, L)
C u i,j
(3)
Z X
(OO)
F (u) = P 4 (L̃Oj L̃Oi R; K, K, K) (12.112)
C u i,j
(0)
Z X
(OO)
E (u) = 2Ω4 |α(L̃Oj L̃Oi R)|2
Cu i,j
(1)
Z X
(OO)
E (u) = 4Ω4 |u|2 |β(L̃Oj L̃Oj R)|2
Cu i,j
(2)
Z X
(OO)
E (u) = 4Ω4 |u|4 (|ρ(L̃Oj L̃Oi R)|2 + |σ(L̃Oi L̃L R)|2 )
Cu i,j
(3)
Z X
(OO)
E (u) = 4Ω4 |u|6 |β(L̃Oj L̃Oi R)|2 (12.113)
Cu i,j
Also, we have:
(3)
Z X
(OO)
F (u) = 2Ω4 |u|6 |α(L̃Oj L̃Oi R)|2 (12.114)
Cu i,j
Taking in 12.63 the Weyl fields W = L̃Oi L̃S R we consider the energy-
momentum density vectorfields:
P (L̃Oi L̃S R; K, K, K) : i = 1, 2, 3 (12.115)
388
(3)
Z X
(OS)
E (u) = 2Ω P 3 (L̃Oi L̃S R; K, K, K) (12.116)
Cu i
and flux:
(3)
Z X
(OS)
F (u) = 2Ω P 4 (L̃Oi L̃S R; K, K, K) (12.117)
Cu i
(3)
Z X
(OS)
E (u) = 4Ω4 |u|6 |β(L̃Oi L̃S R)|2 (12.118)
Cu i
and: Z
(3) X
(OS)
F (u) = 2Ω4 |u|6 |α(L̃Oi L̃S R)|2 (12.119)
Cu i
Finally, taking in 12.63 the Weyl field W = L̃S L̃S R we consider the energy-
momentum density vectorfield:
P (L̃S L̃S R; K, K, K) (12.120)
(3)
Z
(SS)
E (u) = 2ΩP 3 (L̃S L̃S R; K, K, K) (12.121)
Cu
and flux: Z
(3)
(SS)
F (u) = 2ΩP 4 (L̃S L̃S R; K, K, K) (12.122)
Cu
(3)
Z
(SS)
E (u) = 4Ω4 |u|6 |β(L̃S L̃S R)|2 (12.123)
Cu
and: Z
(3)
(SS)
F (u) = 2Ω4 |u|6 |α(L̃S L̃S R)|2 (12.124)
Cu
The above are the 2nd order energies and fluxes.

We note that by the second statement of Pproposition 12.5 the integrants
of all the energies and fluxes are non-negative.
(n)
We now define the total 1st order energies E 1 (u) : n = 0, 1, 2, 3 by:
(0) (0) (0) (0)
E1 (u) = E 0 (u) + δ 2 (L)
E (u) +
(O)
E (u)
(1) (1) (1) (1)
E1 (u) =E 0 (u) + δ 2 (L)
E (u) +
(O)
E (u)
(2) (2) (2) (2)
E1 (u) =E 0 (u) + δ 2 (L)
E (u) +
(O)
E (u) (12.125)
389
and:
(3) (3) (3) (3) (3)
E1 (u) = E 0 (u) + δ 2 (L)
E (u) +
(O)
E (u) +
(S)
E (u) (12.126)
(n)
We also define the total 1st order fluxes F 1 (u) : n = 0, 1, 2, 3 by:
(0) (0) (0) (0)
F1 (u) = F 0 (u) + δ 2 (L)
F (u) +
(O)
F (u)
(1) (1) (1) (1)
F1 (u) =F 0 (u) + δ 2 (L)
F (u) +
(O)
F (u)
(2) (2) (2) (2)
F1 (u) =F 0 (u) + δ 2 (L)
F (u) +
(O)
F (u) (12.127)
and:
(3) (3) (3) (3) (3)
F1 (u) = F 0 (u) + δ 2 (L)
F (u) +
(O)
F (u) +
(S)
F (u) (12.128)
(n)
We then define the total 2nd order energies E 2 (u) : n = 0, 1, 2, 3 by:
(0) (0) (0) (0) (0)
E2 (u) = E 1 (u) + δ 4 (LL)
E (u) + δ
2 (OL)
E (u) +
(OO)
E (u)
(1) (1) (1) (1) (1)
E2 (u) =E 1 (u) + δ 4 (LL)
E (u) + δ
2 (OL)
E (u) +
(OO)
E (u)
(2) (2) (2) (2) (2)
E2 (u) =E 1 (u) + δ 4 (LL)
E (u) + δ
2 (OL)
E (u) +
(OO)
E (u)
(12.129)
and:
(3) (3) (3) (3) (3)
E2 (u) = E 1 (u) + δ 4 (LL)
E (u) + δ
2 (OL)
E (u) +
(OO)
E (u)
(3) (3)
(OS) (SS)
+ E (u) + E (u) (12.130)
(n)
We also define the total 2nd order fluxes F 2 (u) : n = 0, 1, 2, 3 by:
(0) (0) (0) (0) (0)
F2 (u) = F 1 (u) + δ 4 (LL)
F (u) + δ
2 (OL)
F (u) +
(OO)
F (u)
(1) (1) (1) (1) (1)
F2 (u) =F 1 (u) + δ 4 (LL)
F (u) + δ
2 (OL)
F (u) +
(OO)
F (u)
(2) (2) (2) (2) (2)
F2 (u) = F 1 (u) + δ 4 (LL)
F (u) + δ
2 (OL)
F (u) +
(OO)
F (u)
(12.131)
and:
(3) (3) (3) (3) (3)
F2 (u) = F 1 (u) + δ 4 (LL)
F (u) + δ
2 (OL)
F (u) +
(OO)
F (u)
(3) (3)
(OS) (SS)
+ F (u) + F (u) (12.132)
390
We now define the 0th order quantities:

(0) (0)
2
E 0 = sup δ E 0 (u)
u∈[u0 ,c∗ )

(1) (1)
E 0= sup E0 (u)
u∈[u0 ,c∗ )

(2) (2)
E 0= sup δ −1 E 0 (u)
u∈[u0 ,c∗ )

(3) (3)
−3
E 0= sup δ E 0 (u) (12.133)
u∈[u0 ,c∗ )
and:
(3) (3)
F 0= sup δ −3 F 0 (u) (12.134)
u∈[0,min{δ,c∗ −u0 })
and the quantity P0 by:

(0) (1) (2) (3) (3)
P0 = max{ E 0 , E 0 , E 0 , E 0 ; F 0 } (12.135)
Next, we define the 1st order quantities:

(0) (0)
E 1= sup δ 2 E 1 (u)
u∈[u0 ,c∗ )

(1) (1)
E 1= sup E1 (u)
u∈[u0 ,c∗ )

(2) (2)
E 1= sup δ −1 E 1 (u)
u∈[u0 ,c∗ )

(3) (3)
E 1= sup δ −3 E 1 (u) (12.136)
u∈[u0 ,c∗ )
and:
(3) (3)
F 1= sup δ −3 F 1 (u) (12.137)
u∈[0,min{δ,c∗ −u0 })

(0) (1) (2) (3) (3)
P1 = max{ E 1 , E 1 , E 1 , E 1 ; F 1 } (12.138)
Finally, we define the 2nd order quantities:

(0) (0)
2
E2 = sup δ E2 (u)
u∈[u0 ,c∗ )

(1) (1)
E 2= sup E2 (u)
u∈[u0 ,c∗ )
391

(2) (2)
E 2= sup δ −1 E 2 (u)
u∈[u0 ,c∗ )

(3) (3)
E 2= sup δ −3 E 2 (u) (12.139)
u∈[u0 ,c∗ )
and:
(3) (3)
F 2= sup δ −3 F 2 (u) (12.140)
u∈[0,min{δ,c∗ −u0 })

(0) (1) (2) (3) (3)
P2 = max{ E 2 , E 2 , E 2 , E 2 ; F 2 } (12.141)
12.6 The controlling quantity Q2. Bootstrap as-

sumptions and the comparison lemma
In reference to the quantities defined in Chapter 10, we set:
Q1 = max{R[1] (α), R[1] (β), R[1] (ρ), R[1] (σ), R[1] (β); R[1] (α), D0′4 (α)} (12.142)
and:
′4
Q2 = max{R[2] (α), R[2] (β), R[2] (ρ), R[2] (σ), R[2] (β); R[2] (α), D[1] (α)}
(12.143)
/1 (Dβ) defined by 7.3, we set:
Also, in reference to the quantity R
Q′2 = max{Q2 , R
/1 (Dβ)} (12.144)
Let us define:
R40 = max{R40 (α), R40 (β), R40 (ρ), R40 (σ), R40 (β), R40 (α)} (12.145)
By the results of Chapter 10, under the bootstrap assumptions A1.1, A1.2,
A3.1, A3.2 there is a numerical constant C such that:
R40 ≤ CQ1 (12.146)
Also, under the additional bootstrap assumptions A2.1, A2.2, A4.1 A4.2,
there are numerical constants C such that:
R∞
0 ≤ CQ2 (12.147)
and:
/41 ≤ CQ2
R (12.148)
Moreover, there are numerical constants C such that:
max{R40 (D̂α), R40 (Dβ), R40 (Dρ), R40 (Dσ), R40 (Dβ)} ≤ CQ2 (12.149)
392
and:
R40 (D̂α) ≤ CQ2 (12.150)
Reviewing Chapters 3 - 7, as well as Chapters 8 - 9, we see that the quantity
Q′2 bounds all the curvature norms which enter the estimates for the connection
coefficients and, through these, the estimates for the deformation tensors of the
multiplier fields and the commutation fields.
Our present purpose is to compare the quantity (Q′2 )2 to the quantity P2 .
We begin by comparing the quantity (Q1 )2 to the quantity P1 . To effect this
comparison we introduce the following bootstrap assumptions for the deforma-
tion tensors of the commutation fields.
C1.1: δ 3/4 k (L)

îkL∞ (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C1.2: δ 3/4 k (L)

jkL∞ (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C1.3: δ 3/4 k (L)

νkL∞ (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C1.4: δ −1/4 |u|k (L)

mkL∞ (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C2.1: δ −1/4 k (Oi )

îkL∞ (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C2.2: δ −1/4 k (Oi )

jkL∞ (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C2.3: δ −1/4 k (Oi )

νkL∞ (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C2.4: δ 1/4 |u|−1/2 k (Oi )

mkL∞ (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C3.1: δ −1/4 k (S)

îkL∞ (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C3.2: δ −1/4 k (S)

jkL∞ (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C3.3: δ −1/4 k (S)

ν − 1kL∞ (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C3.4: δ 1/4 |u|−1/2 k (S)

mkL∞ (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C3.5: δ −5/4 |u|k (S)

mkL∞ (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
(Recall from Chapter 8 that (L) m = (Oi ) m = 0.) Comparing with the
results of Chapter 8 we see that there is room of a factor of δ 1/4 at least in
the above bootstrap assumptions, a fact which will allow us to show that the
inequalities in these bootstrap assumptions are not saturated.
Lemma 12.5 Under the bootstrap assumptions A0, A1.1, A1.2, A3.1,
A3.2, B1, B2, and C1.1 - C1.4, C2.1 - C2.4, C3.1 - C3.5, there is a numer-
ical constant C such that:
(Q1 )2 ≤ C max{P1 , (D0′4 (α))2 }
393
provided that δ does not exceed a certain positive numerical constant. Moreover,
we have:
(0)
(R[1] (α))2 ≤ C E 1 +Cδ 3/2 max{P1 , (D0′4 (α))2 }
(1)
(R[1] (β))2 ≤ C E 1 +Cδ 1/2 max{P1 , (D0′4 (α))2 }
(2)
(R[1] (ρ))2 , (R[1] (σ))2 ≤ C E 1 +Cδ 3/2 max{P1 , (D0′4 (α))2 }
(3)
(R[1] (β))2 ≤ C E 1 +Cδ 1/2 max{P1 , (D0′4 (α))2 }
and:
(3)
(R[1] (α))2 ≤ C F 1 +Cδ 3/2 max{P1 , (D0′4 (α))2 }
Proof: Taking Y = L, Oi : i = 1, 2, 3 and W = R in Proposition 12.2 and
using the bootstrap assumptions C1.1 - C1.4, C2.1 -C2.4, as well as the basic
bootstrap assumption A0, we deduce, in view of 12.83, 12.86, 12.93, and 12.125,
12.126,
Z ( X
)
(0)
Z
2 2 2 2
|α| + δ |D̂α| + /Oi α|
|L̂ ≤ C E 1 (u) + Cδ −1/2 |u|2 |β|2
Cu i Cu
Z ( )
X
|u|2 |β|2 + δ 2 |Dβ|2 + /Oi β|2
|L
Cu i
(1)
Z Z
5/2 2 −1/2
≤C E1 (u) + Cδ |α| + Cδ |u|4 (|ρ|2 + |σ|2 )
Cu Cu
Z ( )
X
4
|u| |ρ|2 + |σ|2 + δ 2 (|Dρ|2 + |Dσ|2 ) + (|Oi ρ|2 + |Oi σ|2
Cu i
(2)
Z Z
≤C E1 (u) + Cδ 5/2 |u|2 |β|2 + Cδ −1/2 |u|6 |β|2
Cu Cu
Z ( )
X
|u|6 |β|2 + δ 2 |Dβ|2 + /Oi β|2
|L
Cu i
(3)
Z Z
≤C E1 (u) + Cδ 5/2 |u|4 (|ρ|2 + |σ|2 ) + Cδ −1/2 |u|7 |α|2
Cu Cu
(12.151)
Now,
Z
|α|2 ≤ δ −2 (R0 (α))2 ≤ δ −2 (Q1 )2
Cu
Z
|u|2 |β|2 ≤ (R0 (β))2 ≤ (Q1 )2
Cu
Z
|u|4 (|ρ|2 + |σ|2 ) ≤ δ((R0 (ρ))2 + (R0 (σ))2 ) ≤ 2δ(Q1 )2
Cu
394
Z
|u|6 |β|2 ≤ δ 3 (R0 (β))2 ≤ δ 3 (Q1 )2 (12.152)
Cu
and in view of the definition 10.100 and 12.146 we have:

Z
|u|7 |α|2 ≤ Cδ 4 (R40 (α))2 ≤ C ′ δ 4 (Q1 )2 (12.153)
Cu
Under the bootstrap assumptions B1, B2, we have, from the coercivity in-
equality 11.126, a corollary of Proposition 11.1, applied to α and α, and from
Proposition 11.1 itself applied to β, β, ρ, σ:
!
X
2 2 2 2
|u| |∇
/α| ≤ C /Oi α| + |α|
|L̂
i
!
X
2 2 2 2
/α| ≤ C
|u| |∇ /Oi α| + |α|
|L̂
i
X
|u|2 |∇
/β|2 ≤ C /Oi β|2
|L
i
X
2 2
|u| |∇
/β| ≤ C /Oi β|2
|L
i
X
2 2
|u| |d
/ρ| ≤ C |Oi ρ|2
i
X
2 2
|u| |d
/σ| ≤ C |Oi σ|2 (12.154)
i
Substituting 12.154 (except the second), 12.152 and 12.153, in 12.151 we then
obtain:
Z n o (0)
|α|2 + δ 2 |D̂α|2 + |u|2 |∇
/α|2 ≤ C E 1 (u) + Cδ −1/2 (Q1 )2
C
Z u
|u|2 |β|2 + δ 2 |Dβ|2 + |u|2 |∇/β|2

Cu
(1)
≤ C E 1 (u) + Cδ 1/2 (Q1 )2
Z
|u|4 |ρ|2 + |σ|2 + δ 2 (|Dρ|2 + |Dσ|2 ) + |u|2 (|d
/ρ|2 + |d
/σ|2

Cu
(2)
≤ C E 1 (u) + Cδ 5/2 (Q1 )2
Z
|u|6 |β|2 + δ 2 |Dβ|2 + |u|2 |∇
/β|2

Cu
(3)
≤ C E 1 (u) + Cδ 7/2 (Q1 )2 (12.155)
Multiplying the first of 12.155 by δ 2 , the second by 1, the third by δ −1 , and the
fourth by δ −3 , and taking the supremum over u ∈ [u0 , c∗ ), we obtain, in view
395
of the definitions 10.34 - 10.43 and 12.136,
(0)
(R[1] (α))2 ≤ C E 1 +Cδ 3/2 (Q1 )2
(1)
(R[1] (β))2 ≤ C E 1 +Cδ 1/2 (Q1 )2
(2)
(R[1] (ρ))2 , (R[1] (σ))2 ≤ C E 1 +Cδ 3/2 (Q1 )2
(3)
(R[1] (β))2 ≤ C E 1 +Cδ 1/2 (Q1 )2 (12.156)
Next, taking Y = L, Oi : i = 1, 2, 3, S and W = R in Proposition 12.2

and using the bootstrap assumptions C1.1 - C1.4, C2.1 -C2.4, C3.1 - C3.5,
as well as the basic bootstrap assumption A0, noting that for any trace-free
symmetric 2-covariant S tensorfield θ we have:
L̂
/S θ = uD̂θ + uD̂θ (12.157)
we deduce, in view of 12.84, the last of 12.87, 12.94, and 12.99, and 12.128,
Z ( )
X
6 2 2 2 2 2 2
|u| |α| + δ |D̂α| + |u| |D̂α| + /Oi α|
|L̂
Cu i
(3)
Z
≤C F1 (u) + Cδ 5/2 |u|4 |β|2 (12.158)
Cu
Now, in view of the definition 10.51 and 12.146 we have:

Z
|u|4 |β|2 ≤ Cδ 2 (R40 (β))2 ≤ C ′ δ 2 (Q1 )2 (12.159)
Cu
Substituting the second of 12.154 and 12.159 in 12.158 we obtain:
Z n o
|u|6 |α|2 + δ 2 |D̂α|2 + |u|2 |D̂α|2 + |u|2 |∇
/α|2
Cu
(3)
≤ C F 1 (u) + Cδ 9/2 (Q1 )2 (12.160)
Multiplying by δ −3 and taking the supremum over u yields, in view of the

definitions 10.97, 10.98 and 12.137,
(3)
(R[1] (α))2 ≤ C F 1 +δ 3/2 (Q1 )2 (12.161)
The inequalities 12.156 and 12.161 together imply, recalling the definitions
12.138 and 12.142,
(Q1 )2 ≤ C max{P1 , (D0′4 (α))2 } + Cδ 1/2 (Q1 )2 (12.162)
396
which, if δ is suitably small implies:
(Q1 )2 ≤ C ′ max{P1 , (D0′4 (α))2 } (12.163)
This is the first conclusion of the lemma. Substituting this bound in 12.156 and
12.161 then yields the remaining conlusions of the lemma.
We proceed to compare the quantity (Q′2 )2 to the quantity P2 . To effect this

comparison we introduce the following additional bootstrap assumptions for the
deformation tensors of the commutation fields.
C4.1: δ 7/4 |u|−1/2 kL̂

/L (L)
îkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C4.2: δ 7/4 |u|−1/2 kL (L)

jkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C4.3: δ 7/4 |u|−1/2 kL (L)

νkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C4.4: δ 3/4 |u|1/2 kL

/L (L)
mkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C4.5: δ 3/4 |u|−1/2 kL̂

/Oi (L)
îkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C4.6: δ 3/4 |u|−1/2 kOi (L)

jkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C4.7: δ 3/4 |u|−1/2 kOi (L)

νkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C4.8: δ −1/4 |u|1/2 kL

/Oi (L)
mkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C5.1: δ −1/4 |u|−1/2 kL̂

/Oj (Oi )
îkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C5.2: δ −1/4 |u|−1/2 kOj (Oi )

jkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C5.3: δ −1/4 |u|−1/2 kOj (Oi )

νkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C5.4: δ 1/4 |u|−1 kL

/Oj (Oi )
mkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C6.1: δ −1/4 |u|−1/2 kL̂

/Oi (S)
îkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C6.2: δ −1/4 |u|−1/2 kOi (S)

jkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C6.3: δ −1/4 |u|−1/2 kOi (S)

νkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C6.4: δ 1/4 |u|−1 kL

/Oi (S)
mkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C6.5: δ −5/4 |u|1/2 kL

/Oi (S)
mkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C6.6: δ −1/4 |u|−1/2 kL̂

/S (S)
îkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C6.7: δ −1/4 |u|−1/2 kS (S)

jkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
397
C6.8: δ −1/4 |u|−1/2 kS (S)
νkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C6.9: δ 1/4 |u|−1 kL

/S (S)
mkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
C6.10: δ −5/4 |u|1/2 kL

/S (S)
mkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
Comparing with the results of Chapter 9 we see that there is room of a factor
of δ 1/4 at least in the above bootstrap assumptions, a fact which will allow us
to show that the inequalities in these bootstrap assumptions are not saturated.
We shall also make use of the following supplementary bootstrap assump-
tions on the rotation fields.
D0: |u|−1 kOi kL∞ (Su,u ) ≤ 4 : ∀(u, u) ∈ D′

√
D1: k∇/Oi kL∞ (Su,u ) ≤ 2(3 2 + 1) : ∀(u, u) ∈ D′
Comparing with Proposition 8.1 together with the inequality 8.88, and Propo-
sition 8.2 we see that there is an extra factor of 2 in the above assumptions a fact
which will allow us to show that the inequalities of these bootstrap assumptions
are not saturated. We shall also need the following supplementary bootstrap
assumptions on the connection coefficients.
D2.1: δ 3/4 |u|1/2 ktrχkL∞ (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
D2.2: δ 3/4 kχ̂kL∞ (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
D3.1: |u|ktrχkL∞ (Su,u ) ≤ 4 : ∀(u, u) ∈ D′
D3.2: δ −1/4 |u|2 kχ̂kL∞ (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
D4.1: δ 1/4 |u|kηkL∞ (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
D4.2: δ 1/4 |u|kηkL∞ (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
D5: δ 1/4 |u|kωkL∞ (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
D6: δ −1/2 |u|2 kωkL∞ (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
D7: δ −1/2 |u|3/2 kd

/ΩkL4(Su,u ) ≤ 1 : ∀(u, u) ∈ D′
D8: δ 3/4 |u|kDtrχkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
/trχkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′

D9.1: |u|3/2 kd
D9.2: δ −1/4 |u|5/2 k∇

/χ̂kL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
D9.3: δ −3/4 |u|5/2 kD̂χ̂kL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
398
D10.1: δ 1/4 |u|3/2 k∇
/ηkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
D10.2: δ 1/4 |u|3/2 k∇

/ηkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
D10.3: δ 1/4 |u|3/2 kDηkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
D10.4: δ 1/4 |u|3/2 kDηkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
D11: δ 1/4 |u|3/2 kDωkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′
/ωkL4 (Su,u ) ≤ 1 : ∀(u, u) ∈ D′

D12: δ −1/2 |u|5/2 kd
Comparing with the results of Chapters 3 and 4 we see that there is room
of a factor of δ 1/4 at least in the above bootstrap assumptions, a fact which will
allow us to show that the inequalities in these bootstrap assumptions are not
saturated.
Lemma 12.6 Under the bootstrap assumptions A0, A1.1, A1.2, A2.1,
A2.2, A3.1, A3.2, A4.1, A4.2, B1, B2, B3, and C1.1 - C1.4, C2.1 - C2.4,
C3.1 - C3.5, C4.1 - C4.8, C5.1 - C5.4, C6.1 - C6.10, as well as D0, D1,
D2.1, D2.2, D3.1, D3.2, D4.1, D4.2, D5, D6, D7, D8, D9.1 - D9.3, D10.1
- D10.4, D11, D 12, there is a numerical constant C such that:
(Q′2 )2 ≤ C max{P2 , (D[1]
′4
(α))2 }
provided that δ does not exceed a certain positive numerical constant. Moreover,
we have:
(0)
(R[2] (α))2 ≤ C E 2 +Cδ 1/2 max{P1 , (D0′4 (α))2 }
(1)
(R[2] (β))2 ≤ C E 2 +Cδ 1/2 max{P1 , (D0′4 (α))2 }
(2)
(R[2] (ρ))2 , (R[1] (σ))2 ≤ C E 2 +Cδ 1/2 max{P1 , (D0′4 (α))2 }
(3)
(R[2] (β))2 ≤ C E 2 +Cδ 1/2 max{P2 , (D[1]
′4
(α))2 } + C max{P1 , (D0′4 (α))2 }
and:
(3)
′4
(R[2] (α))2 ≤ C F 2 +Cδ max{P2 , (D[1] (α))2 } + Cδ 1/2 max{P1 , (D0′4 (α))2 }
Proof: Taking Y = L, Oi : i = 1, 2, 3 and W = L̃L R and also Y = Oj : j =

1, 2, 3 and W = L̃Oi R : i = 1, 2, 3 in Proposition 12.2 and using the bootstrap
assumptions C1.1 - C1.4, C2.1 -C2.4, C4.1 - C4.8, C5.1 - C5.4, as well
as the basic bootstrap assumption A0, we deduce, in view of 12.103, 12.108,
12.113, and 12.129, 12.130, and 12.146, using also 12.155,
Z ( X
|α|2 + δ 2 |D̂α|2 + /Oi α|2
|L̂
Cu i
399

X X 
+δ 4 |D̂D̂α|2 + δ 2 /Oi D̂α|2 +
|L̂ /Oi α|2
/Oj L̂
|L̂

i i,j
Z ( )
(0) X
−1/2 2
≤C E2 (u) + Cδ |u| δ 2 |Dβ|2 + /Oi β|2
|L + δ −3/2 (Q1 )2
Cu i
Z (
X
|u|2 |β|2 + δ 2 |Dβ|2 + /Oi β|2
|L
Cu i

X X 
+δ 4 |DDβ|2 + δ 2 /Oi Dβ|2 +
|L /Oi β|2
/Oj L
|L

i i,j
Z ( )
(1) X
5/2 2 2 2
≤C E2 (u) + Cδ δ |D̂α| + /Oi α|
|L̂
Cu i
Z ( )
X
+Cδ −1/2 |u|4 δ 2 (|Dρ|2 + |Dσ|2 ) + (|Oi ρ|2 + |Oi σ|2 ) + Cδ 1/2 (Q1 )2
Cu i
Z (
X
|u|4 |ρ|2 + |σ|2 + δ 2 (|Dρ|2 + |Dσ|2 ) + (|Oi ρ|2 + |Oi σ|2 )
Cu i
X
4 2 2 2
+δ (|DDρ| + |DDσ| ) + δ (|Oi Dρ|2 + |Oi Dσ|2 )
i

X 
+ (|Oj Oi ρ|2 + |Oj Oi σ|2 )

i,j
Z ( )
(2) X
5/2 2 2 2 2
≤E 2 (u) + Cδ |u| δ |Dβ| + /Oi β|
|L
Cu i
Z ( )
X
−1/2 6 2 2 2
+Cδ |u| δ |Dβ| + /Oi β|
|L + Cδ 3/2 (Q1 )2
Cu i
Z (
X
|u|6 |β|2 + δ 2 |Dβ|2 + /Oi β|2
|L
Cu i

X X 
+δ 4 |DDβ|2 + δ 2 /Oi Dβ|2 +
|L /Oi β|2
/Oj L
|L

i i,j
Z ( )
(3) X
5/2 4 2 2 2 2 2
≤C E2 (u) + Cδ |u| δ (|Dρ| + |Dσ| ) + (|Oi ρ| + |Oi σ| )
Cu i
Z ( )
X
+Cδ −1/2 |u|7 δ 2 |D̂α|2 + /Oi α|2
|L̂ + Cδ 7/2 (Q1 )2 (12.164)
Cu i
400
We note here that by assumptions D0, D1:
X
/Oi α|2 ≤ C(|u|2 |∇
|L̂ /α|2 + |α|2 )
i
X
/Oi α|2 ≤ C(|u|2 |∇
|L̂ /α|2 + |α|2 )
i
X
/Oi β|2 ≤ C(|u|2 |∇
|L /β|2 + |β|2 )
i
X
/Oi β|2 ≤ C(|u|2 |∇
|L /β|2 + |β|2 )
i
X
(|Oi ρ|2 + |Oi σ|2 ) ≤ C|u|2 (|d
/ρ|2 + |d
/σ|2 ) (12.165)
i
Thus, in reference to the integrals on the right hand sides of the inequalities
12.164 we have:
Z ( X
)
2 2 2
δ |D̂α| + /Oi α|
|L̂ ≤ Cδ −2 (R[1] (α))2 ≤ Cδ −2 (Q1 )2
Cu i
Z ( )
X
2 2 2 2
|u| δ |Dβ| + /Oi β|
|L ≤ C(R[1] (β))2 ≤ C(Q1 )2
Cu i
Z ( )
X
4 2 2 2 2 2
|u| δ (|Dρ| + |Dσ| ) + (|Oi ρ| + |Oi σ| )
Cu i
≤ Cδ((R[1] (ρ))2 + (R[1] (σ))2 ) ≤ 2Cδ(Q1 )2
Z ( )
X
|u|6 δ 2 |Dβ|2 + /Oi β|2 ≤ Cδ 3 (R[1] (β))2 ≤ Cδ 3 (Q1 )2
|L
Cu i
(12.166)
Also, by 12.148 and 12.146:

Z X Z
7 2
|u|7 |u|2 |∇
/α|2 + |α|2

|u| /Oi α| ≤ C
|L̂
Cu i Cu
4
≤ Cδ /41 (α))2
((R + (R40 (α))2 ) ≤ C ′ δ 4 (Q2 )2 (12.167)
The integral Z
|u|7 |D̂α|2
Cu
remains to be appropriately bounded. This shall be done through a suitable

bound for kD̂αkL4 (Su,u ) . To derive this bound we consider the second of the
Bianchi identities, given by Proposition 1.2:
1
ˆ + 3χ̂ρ − 3 ∗ χ̂σ
ˆ + (4η − ζ)⊗β

D̂α − Ωtrχα + 2ωα = −Ω ∇/⊗β (12.168)
2
401
By 12.148,
k∇ˆ L4 (Su,u ) ≤ Cδ|u|−9/2 R

/⊗βk /41 (β) ≤ C ′ δ|u|−9/2 Q2 (12.169)
By assumption D3.2 and 12.146:

∗
kχ̂ρ − χ̂σkL4 (Su,u ) ≤ Cδ 1/4 |u|−2 kρkL4 (Su,u ) + kσkL4 (Su,u )
≤ Cδ 1/4 |u|−9/2 (R40 (ρ) + R40 (σ)) ≤ C ′ δ 1/4 |u|−9/2 Q1 (12.170)
Also, by assumptions D4.1, D4.2 and 12.146,
ˆ L4 (Su,u ) ≤ Cδ −1/4 |u|−1 kβkL4 (Su,u )

k(4η − ζ)⊗βk
≤ Cδ 3/4 |u|−9/2 R40 (β) ≤ C ′ δ 3/4 |u|−9/2 Q1 (12.171)
Moreover, by assumptions D2.1, D5 and 12.146,
ktrχαkL4 (Su,u ) ≤ δ −3/4 |u|−1/2 kαkL4 (Su,u )

≤ δ 3/4 |u|−9/2 R40 (α) ≤ Cδ 3/4 |u|−9/2 Q1 (12.172)
kωαkL4 (Su,u ) ≤ δ −1/4 |u|−1 kαkL4 (Su,u )

≤ δ 5/4 |u|−5 R40 (α) ≤ Cδ 5/4 |u|−5 Q1 (12.173)
In view of the bounds 12.169 - 12.173, we conclude from 12.168 that:
kD̂αkL4 (Su,u ) ≤ Cδ|u|−9/2 Q2 + Cδ 1/4 |u|−9/2 Q1 (12.174)

Z
|u|7 |D̂α|2 ≤ Cδ 3 (Q2 )2 + Cδ 3/2 (Q1 )2 (12.175)
Cu
Under the bootstrap assumptions B1, B2, B3, we have, from the coercivity
inequality 11.157 applied to α, α, the coercivity inequality 11.145 applied to β,
β and the coercivity inequality 11.132 applied to ρ, σ:
 
Z Z X X 
|u|4 |∇
/ 2 α|2 ≤ C /Oi α|2 +
/Oj L̂
|L̂ /Oi α|2 + |α|2
|L̂
Su,u Su,u  i,j i

 
Z Z X X 
|u|4 |∇
/ 2 α|2 ≤ C /Oi α|2 +
/Oj L̂
|L̂ /Oi α|2 + |α|2
|L̂
Su,u Su,u  i,j i

 
Z Z X X 
|u|4 |∇
/ 2 β|2 ≤ C /Oi β|2 +
/Oj L
|L /Oi β|2 + |β|2
|L
Su,u Su,u  i,j i

402
 
Z Z X X 
|u|4 |∇
/ 2 β|2 ≤ C /Oi β|2 +
/Oj L
|L /Oi β|2 + |β|2
|L
Su,u Su,u 
i,j i

X
|u|4 |∇
/ 2 ρ|2 ≤ C |Oj Oi ρ|2
i,j
X
4 2 2
|u| |∇
/ σ| ≤ C |Oj Oi σ|2 (12.176)
i,j
Also, under the bootstrap assumptions B1, B2, we have, from the coercivity
inequality 11.126, applied to D̂α, D̂α and D̂α and from Proposition 11.1 itself
applied to Dβ, Dβ and Dβ, ρ, σ:
!
X
2 2 2 2
|u| |∇
/D̂α| ≤ C /Oi D̂α| + |D̂α|
|L̂
i
!
X
2 2 2 2
|u| |∇
/D̂α| ≤ C /Oi D̂α| + |D̂α|
|L̂
i
!
X
|u|2 |∇
/D̂α|2 ≤ C /Oi D̂α|2 + |D̂α|2
|L̂
i
X
|u|2 |∇
/Dβ|2 ≤ C /Oi Dβ|2
|L
i
X
2 2
/Dβ| ≤ C
|u| |∇ /Oi Dβ|2
|L
i
X
2 2
|u| |∇
/Dβ| ≤ C /Oi Dβ|2
|L
i
X
2 2
|u| |d
/Dρ| ≤ C |Oi Dρ|2
i
X
2 2
|u| |d
/Dσ| ≤ C |Oi Dσ|2 (12.177)
i
Substituting 12.166, 12.167 and 12.175, as well as 12.154 (except the second)
and 12.176 (except the second), 12.177 (except the second, third and sixth) in
12.164 we obtain:
Z n
|α|2 + δ 2 |D̂α|2 + |u|2 |∇
/α|2
Cu
o
+δ 4 |D̂D̂α|2 + δ 2 |u|2 |∇
/D̂α|2 + |u|4 |∇
/ 2 α|2
(0)
≤ C E 2 (u) + δ −3/2 (Q1 )2
Z
|u|2 |β|2 + δ 2 |Dβ|2 + |u|2 |∇
/β|2

Cu
+δ 4 |DDβ|2 + δ 2 |u|2 |∇
/Dβ|2 + |u|4 |∇
/ 2 β|2
403
(1)
≤ C E 2 (u) + Cδ 1/2 (Q1 )2
Z
|u|4 |ρ|2 + |σ|2 + δ 2 (|Dρ|2 + |Dσ|2 ) + |u|2 (|d
/ρ|2 + |d
/σ|2 )

Cu
+δ 4 (|DDρ|2 + |DDσ|2 ) + δ 2 |u|2 (|d
/Dρ|2 + |d
/Dσ|2 )
+|u|4 (|∇
/ 2 ρ|2 + |∇
/ 2 σ|2 )
(2)
≤ E 2 (u) + Cδ 3/2 (Q1 )2
Z
|u|6 |β|2 + δ 2 |Dβ|2 + |u|2 |∇
/β|2

Cu
+δ 4 ||DDβ|2 + δ 2 |u|2 |∇
/Dβ|2 + |u|4 |∇
/ 2 β|2
(3)
≤ C E 2 (u) + Cδ 7/2 (Q2 )2 + Cδ 3 (Q1 )2 (12.178)
Multiplying the first of 12.178 by δ 2 , the second by 1, the third by δ −1 , and the
fourth by δ −3 , and taking the supremum over u ∈ [u0 , c∗ ), we obtain, in view
of the definitions 10.58 - 10.65 and 12.139,
(0)
(R[2] (α))2 ≤ C E 2 +Cδ 1/2 (Q1 )2
(1)
(R[2] (β))2 ≤ C E 2 +Cδ 1/2 (Q1 )2
(2)
(R[2] (ρ))2 , (R[2] (σ))2 ≤ C E 2 +Cδ 1/2 (Q1 )2
(3)
(R[2] (β))2 ≤ C E 2 +Cδ 1/2 (Q2 )2 + C(Q1 )2 (12.179)
Next, taking in Proposition 12.2 Y = L, Oi : i = 1, 2, 3 and W = L̃L R, and

Y = Oj : j = 1, 2, 3 and W = L̃Oi R : i = 1, 2, 3, and also W = L̃S R and
Y = Oi : i = 1, 2, 3, S, and using the bootstrap assumptions C1.1 - C1.4, C2.1
- C2.4, C3.1 - C3.5, C4.1 - C4.8, C5.1 - C5.4, C6.1 - C6.10, in view of
12.104, 12.109, 12.114, 12.132, and 12.146, using also 12.160, and noting that
by 12.157 for any trace-free symmetric 2-covariant tensorfield θ we have
/Oi L̂
L̂ /S θ = uL̂ /Oi D̂θ,
/Oi D̂θ + uL̂ (12.180)
we deduce:
Z (
X
|u| |α|2 + δ 2 |D̂α|2 + |u|2 |D̂α|2 +
6
/Oi α|2
|L̂
Cu i
X X
4 2 2
+δ |D̂D̂α| + δ /Oi D̂α| + |u|2
|L̂ 2
/Oi D̂α|2
|L̂
i i

X 
+ /Oi α|2 + |L̂
/Oj L̂
|L̂ /S α|2
/S L̂

i,j
Z ( )
(3) X
≤C F2 (u) + Cδ 5/2 |u|4 δ 2 |Dβ|2 + |L
/S β|2 + /Oi β|2
|L
Cu i
404
+Cδ 7/2 (Q1 )2 (12.181)
Now, by the fourth of 12.165 and 12.148, 12.146:

Z X Z
|u|4 /Oi β|2 ≤ C
|L |u|4 (|u|2 |∇
/β|2 + |β|2 )
Cu i Cu
2
≤ Cδ /41 (β))2
((R + (R40 (β))2 ) ≤ C ′ δ 2 (Q2 )2 (12.182)
Also, by 12.149:
Z
|u|4 |Dβ|2 ≤ C(R40 (Dβ))2 ≤ C ′ (Q2 )2 (12.183)
Cu
Since for any S 1-form ξ we have;
/S ξ = uDξ + uDξ
L (12.184)
to bound the integral Z

|u|4 |L
/S β|2
Cu
what remains to be bounded is the integral

Z
|u|6 |Dβ|2
Cu
For this purpose we consider the fourth of the Bianchi identities, given by Propo-
sition 1.2:
3
Dβ + Ωtrχβ − Ωχ̂♯ · β − ωβ = −Ω div / α + (η ♯ − 2ζ ♯ ) · α

(12.185)
2
By 12.148,
/ αkL4 (Su,u ) ≤ Cδ 3/2 |u|−5 R

kdiv /41 (α) ≤ C ′ δ 3/2 |u|−5 Q2 (12.186)
By assumptions D3.1, D3.2 and 12.146,
ktrχβkL4 (Su,u ) ≤ 4|u|−1 kβkL4 (Su,u ) ≤ 4δ|u|−9/2 R40 (β) ≤ Cδ|u|−9/2 Q1

kχ̂♯ · βkL4 (Su,u ) ≤ δ 1/4 |u|−3/2 kβkL4 (Su,u ) ≤ δ 5/4 |u|−5 R40 (β) ≤ Cδ 5/4 |u|−5 Q1
(12.187)
Also, by assumption D6 and 12.146,
kωβkL4 (Su,u ) ≤ δ 1/2 |u|−2 kβkL4 (Su,u ) ≤ δ 3/2 |u|−11/2 R40 (β) ≤ δ 3/2 |u|−11/2 Q1
(12.188)
Moreover, by assumptions D4.1, D4.2 and 12.146,
k(η ♯ − 2ζ ♯ ) · αkL4 (Su,u ) ≤ Cδ −1/4 |u|−1 kαkL4 (Su,u )

≤ Cδ 5/4 |u|−5 R40 (α) ≤ C ′ δ 5/4 |u|−5 Q1 (12.189)
405
In view of the bound 12.186 - 12.189, we conclude from 12.185 that:
kDβkL4 (Su,u ) ≤ Cδ 3/2 |u|−5 Q2 + Cδ|u|−9/2 Q1 (12.190)
Z
|u|6 |Dβ|2 ≤ Cδ 3 (Q2 )2 + Cδ 2 (Q1 )2 (12.191)
Cu
which together with 12.183 implies:

Z
|u|4 |L
/S β|2 ≤ Cδ 2 (Q2 )2 (12.192)
Cu
Substituting 12.182, 12.183, 12.192, as well as the second of 12.154 and

12.176 and the second and third of 12.177 in 12.181 we obtain:
Z n
|u|6 |α|2 + δ 2 |D̂α|2 + |u|2 |D̂α|2 + |u|2 |∇
/α|2
Cu
+δ 4 |D̂D̂α|2 + δ 2 |u|2 |∇
/D̂α|2 + |u|4 |∇
/D̂α|2
o
+|u|4 |∇/ 2 α|2 + |L̂ /S α|2
/S L̂
(3)
≤ C F 2 (u) + Cδ 9/2 (Q2 )2 + Cδ 7/2 (Q1 )2 (12.193)
We shall now derive, using the bound 12.193, an appropriate bound for the
integral Z
|u|10 |D̂D̂α|2 (12.194)
Cu
Using 12.157 and 1.14 we obtain:
L̂ /S α = u2 D̂D̂α + u2 D̂D̂α + uD̂α + uD̂α

/S L̂
+uu(D̂D̂α + D̂D̂α) (12.195)
Now, according to Lemma 1.4 for any S tensorfield θ we have:
DDθ − DDθ = 4L
/Ω2 ζ ♯ θ (12.196)
For any trace-free symmetric 2-covariant S tensorfield θ we have:
1
Dθ = D̂θ + g/(Ωχ̂, θ)
2
1
Dθ = D̂θ + g/(Ωχ̂, θ) (12.197)
2
It follows that in the case of a trace-free 2-covariant S tensorfield θ the following
commutation formula holds:
D̂D̂θ − D̂D̂θ = 4L̂
/Ω2 ζ ♯ θ (12.198)
1 1
− Ωχ̂(Ωχ̂, θ) + Ωχ̂(Ωχ̂, θ)
2 2
406
We consider this formula with α in the role of θ. Using assumptions D7 and
D10.1, D10.2 we deduce:
/Ω2 ζ ♯ αkL2 (Su,u ) ≤ Cδ −1/4 |u|−1 k∇

kL̂ /αkL2 (Su,u ) + Cδ −1/4 |u|−3/2 kαkL4 (Su,u )
(12.199)
Taking into account the fact that by 12.146
kαkL4 (Su,u ) ≤ δ 3/2 |u|−4 R40 (α) ≤ Cδ 3/2 |u|−4 Q1 (12.200)
it follows that: Z
|u|8 |L̂
/Ω2 ζ ♯ α|2 ≤ Cδ 5/2 (Q1 )2 (12.201)
Cu
Moreover, by assumptions D2.2, D3.2 we have:

Z Z
|u|8 |χ̂(χ̂, α)|2 , |u|8 |χ̂(χ̂, α)|2
Cu Cu
Z
≤ δ −1 |u|5 |α|2 ≤ δ 2 (Q1 )2 (12.202)
Cu
We then conclude from 12.198 with α in the role of θ that:

Z
|u|8 |D̂D̂α − D̂D̂α|2 ≤ Cδ 2 (Q1 )2 (12.203)
Cu
In view of the formula 12.195 and the bounds 12.193 and 12.202 the required
bound for the integral 12.194 will follow if we can appropriately estimate the
integral Z
|u|8 |D̂D̂α|2 (12.204)
Cu
For this purpose we apply D̂ to the Bianchi identity 12.168 to obtain an expres-
sion for D̂D̂α. In deriving this expression we make use of the following three
elementary facts. First, that for an arbitrary S 1-form ξ we have:
D̂∇ˆ =∇
/⊗ξ ˆ
/⊗Dξ ˆ/ · ξ − 2Ωχ̂div
− 2DΓ / ξ (12.205)
This formula follows in a straightforward manner from the second part of Lemma
4.1 together with the conjugate of formula 6.107. The second fact, which also
follows in a straightforward manner, is that for any pair of S 1-forms ξ, θ we
have:
ˆ = Dξ ⊗θ
D̂(ξ ⊗θ) ˆ + ξ ⊗Dθ
ˆ − 2Ωχ̂(ξ, θ) (12.206)
Finally, the third fact is that for any trace-free symmetric 2-covariant S tensor-
field θ we have:
D̂ ∗ θ = ∗ D̂θ (12.207)
407
(Note that an analogous formula does not hold for S 1-forms). To derive 12.207
we begin from the fact that:
Dǫ/ = Ωtrχǫ/ (12.208)
This implies that, with the notations of Chapter 1 (see 1.142),
Dǫ/♯ = −2Ω ∗ χ̂♯ (12.209)
Let then θ be an arbitrary trace-free symmetric 2-covariant S tensorfield. With

respect to an arbitrary local frame field for the Su,u we then have:
(D ∗ θ)AB = /ǫ ♯A C (Dθ)CB − 2Ω ∗ χ̂♯A C θCB (12.210)

This is symmetric, for,
/ǫ ♯♯AB (ǫ/♯A C (Dθ)CB − 2Ω ∗ χ̂♯A C θCB ) = (g/−1 )BC (Dθ)CB − 2Ωχ̂♯♯BC θCB = 0
Thus we can write:

1 ♯C
(D ∗ θ)AB = (ǫ/ (Dθ)CB + /ǫ ♯B C (Dθ)CA )
2 A
−Ω( ∗ χ̂♯A C θCB + ∗ χ̂♯B C θCA )
1
= (ǫ/♯A C (D̂θ)CB + /ǫ ♯B C (D̂θ)CA ) − Ωg/AB ( ∗ χ̂, θ)(12.211)
2
by the identity 1.187. Taking the trace free parts on both sides, and recalling
that the dual of a trace-free symmetric 2-covariant S tensorfield is also symmet-
ric and trace-free, then yields 12.207.
Applying D̂ to the Bianchi identity 12.168 and adding 32 Ωtrχ times the
Bianchi identity 12.168, and using the above three facts we deduce the following
expression for D̂D̂α + 23 ΩtrχD̂α:
3 1 1
D̂D̂α + ΩtrχD̂α = ΩtrχD̂α − 2ω D̂α + αD(Ωtrχ) − 2αDω
2 2 2
1 3
+Ω ωtrχ − 3ωtrχ + Ωtrχtrχ α
2 4
n
−Ω ∇ ˆ
/⊗Dβ ˆ/ · β − 2ω χ̂div
− 2DΓ / β
ˆ + (4η − ζ)⊗Dβ
+(4Dη − Dζ)⊗β ˆ − 2Ωχ̂(4η − ζ, β)
+3ρD̂χ̂ − 3σ ∗ D̂χ̂

3 3
+3χ̂ Dρ + Ωtrχρ − 3 ∗ χ̂ Dσ + Ωtrχσ
2 2

3
ˆ
ˆ + (4η − ζ)⊗β

−Ω Ωtrχ + ω ∇/⊗β
2
−3Ωω(χ̂ρ − ∗ χ̂σ) (12.212)
Consider first the principal term on the right. This is the term ∇ˆ
/⊗Dβ in
the second parenthesis. Now Dβ is given by the Bianchi identity 12.185. The
408
principal term in Dβ is the term −Ωdiv
/ α. The contibution of this term to the
integral Z
|u|8 |∇ˆ
/⊗Dβ|2
(12.213)
Cu
is bounded through 12.193 by:

(3)
C F 2 (u) + Cδ 9/2 (Q2 )2 + Cδ 7/2 (Q1 )2 (12.214)
Using assumptions D3.1, D3.2, D4.1, D4.2, D6, D7, D9.1, D9.2, D10.1,
D10.2, and D12, we deduce that the contributions of the remaining terms in
Dβ to ∇ ˆ
/⊗Dβ are bounded in L2 (Su,u ) by:
Cδ|u|−5 (R
/41 + R40 ) ≤ C ′ δ|u|−5 Q2 (12.215)
by 12.148 and 12.146. We then conclude that:

Z (3)
|u|8 |∇ˆ
/⊗Dβ|2
≤ C F 2 (u) + Cδ 2 (Q2 )2 (12.216)
Cu
Consider next the terms

3 3
3χ̂ Dρ + Ωtrχρ − 3 ∗ χ̂ Dσ + trχσ
2 2
in the second parenthesis on the right in 12.212. The eighth and tenth Bianchi
identities of Proposition 1.2 read:

3 1
Dρ + Ωtrχρ = −Ω div / β + (2η − ζ, β) + (χ̂, α)
2 2

3 ∗ 1
Dσ + Ωtrχσ = −Ω curl / β + (2η − ζ, β) + χ̂ ∧ α (12.217)
2 2
By assumptions D4.1, D4.2, D2.2 and 12.148, 12.146 the right hand sides of
equations 12.217 are bounded in L4 (Su,u ) by:
Cδ 3/4 |u|−4 (R
/41 + R40 ) ≤ C ′ δ 3/4 |u|−4 Q2 (12.218)
Taking also into account assumption D3.2 we then conclude that:

3 ∗ 3
χ̂ Dρ + Ωtrχρ − χ̂ Dσ + Ωtrχσ ≤ Cδ|u|−11/2 Q2
2 2 L2 (Su,u )
(12.219)
From the expression for DΓ/ of the second part of Lemma 4.1 and assumptions
D9.1, D9.2 and D3.1, D3.2 and D7 we deduce:
ˆ/k 4
kDΓ −3/2
L (Su,u ) ≤ C|u| (12.220)
409
Using assumptions D3.2, D5, D9.3, D10.3, D10.4 as well as the bounds
12.190 and 12.220 we deduce that the remaining terms in the second parenthesis
on the right in 12.212 are bounded in L2 (Su,u ) by:
Cδ|u|−11/2 Q2 + Cδ 3/4 |u|−5 Q1 (12.221)
Also, using assumptions D3.2, D4.1, D4.2, and D6 we deduce that the last
two terms in 12.212 are bounded in L2 (Su,u ) by:
Cδ|u|−5 Q2 + Cδ 3/4 |u|−5 Q1 (12.222)
The first five terms on the right in 12.212 remain to be considered. By

assumptions D2.1, D5, D6 and the bound 12.193 the contributions to the
integral 12.204 of the first and second terms in 12.212 are bounded by:
(3)
Cδ −3/2 F 2 (u) + Cδ 3 (Q2 )2 + Cδ 2 (Q1 )2 (12.223)
Also, by assumptions D8, D11 and 12.146 the third and fourth terms in 12.212
are bounded in L2 (Su,u ) by:
Cδ 3/4 |u|−5 Q1 (12.224)
Finally, by assumptions D2.1, D3.1, D5, D6 the fifth term in 12.212 is bounded
in L2 (Su,u ) by:
Cδ 3/4 |u|−5 Q1 (12.225)
Collecting the above results we conclude that:
2
3
Z (3)
|u|8 D̂ D̂α + ΩtrχD̂α ≤ Cδ −3/2 F 2 (u) + Cδ 2 (Q2 )2 + Cδ 3/2 (Q1 )2
Cu 2
(12.226)
Combining with the bound
2
3
Z (3)
|u|8 ΩtrχD̂α ≤ Cδ −2 F 2 (u) + Cδ 5/2 (Q2 )2 + Cδ 3/2 (Q1 )2 (12.227)
Cu 2
which follows from the estimate 12.193 and assumption D3.1, we then obtain:
Z (3)
|u|8 |D̂ D̂α|2 ≤ Cδ −2 F 2 (u) + Cδ 2 (Q2 )2 + Cδ 3/2 (Q1 )2 (12.228)
Cu
This estimate together with the estimate 12.203 implies:

Z (3)
|u|2 |u|8 |D̂D̂α + D̂D̂α|2 ≤ C F 2 (u) + Cδ 4 (Q2 )2 + Cδ 7/2 (Q1 )2 (12.229)
Cu
410
Using equation 12.195 to expresses u2 D̂D̂α in terms of L̂ /S α, uu(D̂D̂α +
/S L̂
2
D̂D̂α), u D̂D̂α, and uD̂α, uD̂α, we conclude from the estimates 12.193 and
12.229 that:
Z n
|u|6 |α|2 + δ 2 |D̂α|2 + |u|2 |D̂α|2 + |u|2 |∇
/α|2
Cu
+δ 4 |D̂D̂α|2 + δ 2 |u|2 |∇
/D̂α|2 + |u|4 |∇
/D̂α|2
o
+|u|4 |∇/ 2 α|2 + |u|4 |D̂D̂α|2
(3)
≤ C F 2 (u) + Cδ 4 (Q2 )2 + Cδ 7/2 (Q1 )2 (12.230)
Multiplying by δ −3 and taking the supremum over u yields, in view of the

definitions 10.105, 10.106 and 12.140,
(3)
(R[2] (α))2 ≤ C F 2 +Cδ(Q2 )2 + Cδ 1/2 (Q1 )2 (12.231)
The inequalities 12.178 and 12.231 together imply, recalling the definitions
12.141 and 12.142,
(Q2 )2 ≤ C max{P2 , (D[1]

′4
(α))2 } + Cδ 1/2 (Q2 )2 + C(Q1 )2 (12.232)
which, if δ is suitably small implies, taking also into account Lemma 12.5,
(Q2 )2 ≤ C ′ max{P2 , (D[1]

′4
(α))2 } (12.233)
Substituting this bound in 12.178 and 12.231 then yields the conclusions of the
lemma with the exception of the conclusion that there is a numerical constant
C such that:
′4
R/1 (Dβ) ≤ C max{P2 , (D[1] (α))2 } (12.234)
To obtain the remaining conclusion we take again Y = Oi : i = 1, 2, 3 and
W = L̃S R in Proposition 12.2 and use the bootstrap assumptions C2.1 - C2.4,
C3.1 - C3.5, C6.1 - C6.5, to deduce, in view of 12.118, 12.130, and the last
of 12.139:
Z Z ( )
X (3) X
|u|6 /S β|2 ≤ C E 2 (u) + Cδ 1/2
/Oi L
|L /S β|2 +
|u|6 |L |L/Oi β|2
Cu i Cu i
Z n o
−1/2
+Cδ |u|7 |L̂
/S α|2 + |L̂
/Oi α|2
Cu
Z X
+Cδ 5/2 |u|4 (|Oi ρ|2 + |Oi σ|2 )
Cu i
(12.235)
Now, from 12.184 together with 12.149 and 12.190 we deduce:

Z Z
/S β|2 =
|u|6 |L |u|6 |uDβ + uDβ|2 ≤ Cδ 3 (Q2 )2 (12.236)
Cu Cu
411
From 12.157 together with 12.150 and 12.174 we deduce:
Z Z
|u|7 |L̂
/S α|2 = |u|7 |uD̂α + uD̂α|2 ≤ Cδ 7/2 (Q2 )2 (12.237)
Cu Cu
Also, by the fourth of 12.165 together with 12.148 and 12.146:

Z X
|u|6 /Oi β|2 ≤ Cδ 3 (Q2 )2
|L (12.238)
Cu i
By the second of 12.165 together with 12.148 and 12.146:

Z X
|u|7 /Oi α|2 ≤ Cδ 4 (Q2 )2
|L̂ (12.239)
Cu i
and by the last of 12.165:

Z X
|u|4 (|Oi ρ|2 + |Oi σ|2 ) ≤ Cδ(Q1 )2 (12.240)
Cu i
Substituting the above in 12.235 we obtain:

Z X (3)
|u|6 /S β|2 ≤ C E 2 (u) + Cδ 3 (Q2 )2
/Oi L
|L (12.241)
Cu i
Now, by Proposition 11.1 applied to L /S β:

X
|u|2 |∇/S β|2 ≤ C
/L /S β|2
/Oi L
|L (12.242)
i
It follows that:
Z (3)
|u|8 |∇/S β|2 ≤ C E 2 (u) + Cδ 3 (Q2 )2
/L (12.243)
Cu
From 12.184,
∇
/L
/S β = u∇
/Dβ + u∇
/Dβ
and we have:
Z
|u|8 |u|2 |∇
/Dβ|2 ≤ δ 3 (R
/1 (Dβ))2 ≤ δ 3 (Q2 )2 (12.244)
Cu

Z (3)
/Dβ|2 ≤ C E 2 (u) + Cδ 3 (Q2 )2
|u|10 |∇ (12.245)
Cu
Multiplying this by δ −3 , taking the supremum over u ∈ [u0 , c∗ ), recalling the

definition 7.3, and taking into account the bound 12.233, then yields the desired
bound 12.234. This completes the proof of Lemma 12.6.
412
12.7 Statement of the existence theorem. Out-
line of the continuity argument
(n)
Let us denote by D : n = 0, 1, 2, 3 the following quantities, which depend only
on the initial data on Cu0 :
(0) (0)
2
D = δ E 2 (u0 )
(1) (1)
D = E 2 (u0 )
(2) (2)
−1
D= δ E2 (u0 )
(3) (3)
−3
D= δ E2 (u0 ) (12.246)
The spacetime manifold (M, g), solution of the vacuum Einstein equations,
which we have hitherto been considering, corresponds to the real number c∗ ∈
(u0 , −1]. Thus for the sake of clarity we denote it presently by (Mc∗ , g). We may
place any real number c ∈ (u0 , −1] in the role of c∗ and denote the corresponding
spacetime manifold by (Mc , g). We also denote by Mc′ the non-trivial part of
Mc and by Dc and Dc′ the parameter domains to which Mc and Mc′ respectively
correspond. The symbol c∗ shall from now on be reserved for the particular real
number in (u0 , −1] to be defined below.
(n)
We must first clarify the following point. The initial data quantities D : n =
/41 , D
0, 1, 2, 3, like the initial data quantities D0∞ , D /2 (trχ), D ′4
/3 (trχ), and D[1] (α)
are defined relative to the whole of Cu0 , even though we may be considering a
spacetime manifold Mc with c ∈ (u0 , u0 +δ), which contains only the part Cuc−u 0
0
′
of Cu0 . That is, given the pair of smooth mappings ψ0 , ψ0 of [0, 1] × D2ρ → Ŝ
(see 2.42) and the numbers δ ∈ (0, 1], u0 < −1, the procedure of Chapter 2
determines the full set of initial data along Cu0 , that is the metric g/, the 2nd
fundamental forms χ and χ, the torsion ζ, the function ω, its first transversal
derivative Dω, all curvature components α, β, ρ, σ, β, α, and the first transversal
derivative Dα of α. These suffice to determine the first transversal derivatives
of all connection coefficients along Cu0 directly through the optical structure
equations and the first transversal derivatives of all curvature componets directly
through the Bianchi equations. Moreover, the second transversal derivative D2 β
is also directly determined, in terms of ∇ /Dα, by applying D to the fourth Bianchi
identity of Proposition 1.2. To be sure, the Einstein equations are used to make
this determination, and if we are given a solution on Mc for some c ∈ (u0 , u0 +δ)
then these equations are satisfied by our solution only on the part Cuc−u 0
0
of Cu0 .
Nevertheless, our procedure determines entities on the whole of Cu0 , which must
coincide on Cuc−u 0
0
with the corresponding entities induced by our solution. Now
the components of the deformation tensors of the commutation fields L and S
depend only on the connection coefficients (see 8.21, 8.30) so these deformation
tensors as well as their transversal derivatives are determined in the above sense
on all of Cu0 . By 8.139 - 8.143 and 8.69 the only components of the deformation
413
tensors of the commutation fields Oi which do not vanish identically on Cu0 are:
(Oi ) 1 (Oi ) (Oi ) (Oi )

j= π
/ and î = π̂
/ (12.247)
2
◦
Since (Oi ) π
/=L /Oi g/= 0, the Oi being Killing fields of the standard
/Oi g/ and L
sphere, it follows from 2.116 that:
(Oi )
/ = M1 (δ 1/2 |u0 |−1 ) : on Cu0
π (12.248)
Moreovever, in canonical coordinates we have, along Cu0 :

p
(Oi ) 2 ∂(OiA detg/)
tr π
/= p (12.249)
detg/ ∂ϑA
and by 2.6, 2.10 (see also 2.118):

r
p ◦
detg/ = φ2 |u0 |2 det g/ (12.250)
hence: r
(Oi ) 2 ∂ ◦ 2 A ∂φ2
tr π
/= q (OiA φ2 det g/) = O
◦ ∂ϑA φ2 i ∂ϑA
φ2 det g/
that is:
Oi ) 4
tr π
/= Oi φ : on Cu0 (12.251)
φ
The first of 12.109 then gives:
(Oi )
tr / = M1 (δ|u0 |−2 )
π (12.252)
Thus, according to the construction of Chapter 2 we have on Cu0 :

(Oi )
j = M1 (δ|u0 |−2 ), (Oi )
î = M1 (δ 1/2 |u0 |−1 ) (12.253)
In view of Proposition 12.2, it follows that the energy densities are defined
everywhere on Cu0 , so the energy integrals 12.246 on the whole of Cu0 are well
defined. Moreover, the results of Chapter 2 imply that there is a non-negative
non-decreasing continuous function of M8 such that:
(0) (1) (2) (3)
max{ D , D , D , D } ≤ F (M8 ) (12.254)
the entity which requires Mk with the highest k being D2 β, which occurs in
(3)
(SS)
E (u0 ) and is expressed in terms of
/D̂α = M8 (δ 3/2 |u0 |−7 )

∇ (12.255)
414
′4
by 2.224. Also, by 2.198 and 2.224 the quantity D[1] (α) is bounded by a non-
negative non-decreasing continuous function of M7 (in fact by |u0 |−1 times such
a function), while by 2.156, 2.163 and 2.174, the quantities D0∞ , D/41 , D
/42 (trχ),
/3 (trχ) are bounded by non-negative non-decreasing continous functions of M3 ,
D
(n)
′4
M4 , M5 , M6 , respectively. Thus all of the quantities D : n = 0, 1, 2, 3, D[1] (α),
∞ 4 4
D0 , D
/1 , D
/2 (trχ), D
/3 (trχ) are bounded by a non-negative non-decreasing con-
tinuous function of M8 .
The aim of the next four chapters is to establish the following theorem.
Theorem 12.1 Let us be given smooth initial data on Cu0 as described in

Chapter 2. Let us define c∗ to be the supremum of the set A of real numbers
c in the interval (u0 , −1] such that to c there corresponds a spacetime mani-
fold (Mc , g), as defined in Chapter 1, smooth solution of the vacuum Einstein
equations, taking the given initial data along Cu0 , and such that:
1. The generators of the null hypersurfaces Cu and C u , as defined in Chapter

1, have no end points in Mc \ Γ0 . Thus, the Cu contain no conjugate or
cut points in Mc and the C u contain no focal or cut points in Mc \ Γ0 .
2. The bootstrap assumptions A0, A1.1, A1.2, A2.1, A2.2, A3.1, A3.2,
A4.1, A4.2, B1, B2, B3, and C1.1 - C1.4, C2.1 - C2.4, C3.1 - C3.5,
C4.1 - C4.8, C5.1 - C5.4, C6.1 - C6.10, as well as D0, D1, D2.1,
D2.2, D3.1, D3.2, D4.1, D4.2, D5, D6, D7, D8, D9.1 - D9.3, D10.1
- D10.4, D11, D 12, hold on Mc′ and Dc′ .
3. The quantity P2 corresponding to Mc′ satisfies the bound:
(0) (1) (2) (3)
P2 ≤ G( D , D , D , D )
where G is a positive continuous function, non-decreasing in each of its

arguments, which shall be specified in the sequel.
Then, if δ is suitably small depending on the initial data quantities:

(0) (1) (2) (3)
′4
D , D , D , D and D[1] (α)
as well as:
D0∞ , D
/41 and D
/42 (trχ), D
/3 (trχ)
we have:
c∗ = −1 ∈ A
We remark that condition 1 of the theorem may be restated as follows. With

Mc \ Γ0 = (Dc \ A0 ) × S 2 we have a smooth solution in canonical coordinates
on Mc \ Γ0 , the subdomain of Mc \ Γ where u ≤ 0 being isometric to the
corresponding domain in Minkowski spacetime.
415
We also remark that by the local existence theorem of Chapter 2 and the
argument to be presented in the 3rd section of Chapter 16, the set A is not
empty.
The proof of the theorem begins with the observation that c∗ ∈ A and
A = (u0 , c∗ ]. This is because c ∈ A implies (u0 , c] ⊂ A, hence A ⊃ (u0 , c∗ )
while Mc was defined in Chapter 1 to include its past boundary but not its
future boundary, hence: [
M c∗ = Mc (12.256)
c∈(u0 ,c∗ )
It follows that the 1st and 2nd of the above conditions for membership in the set
A hold for c∗ . Moreover, since (u0 , c∗ ) ⊂ A, for every c ∈ (u0 , c∗ ) the energies
(n) (3)
E2 : n = 0, 1, 2, 3 and flux F 2 corresponding to Mc satisfy:

(0) (0) (1) (2) (3)
sup δ 2 E 2 (u) ≤ G( D , D , D , D )
u∈[u0 ,c)

(1) (0) (1) (2) (3)
sup E2 (u) ≤ G( D , D , D , D )
u∈[u0 ,c)

(2) (0) (1) (2) (3)
−1
sup δ E 2 (u) ≤ G( D , D , D , D )
u∈[u0 ,c)

(3) (0) (1) (2) (3)
sup δ −3 E 2 (u) ≤ G( D , D , D , D )
u∈[u0 ,c)
and:
(3) (0) (1) (2) (3)
−3
sup δ F 2 (u) ≤ G( D , D , D , D )
u∈[0,min{δ,c−u0 })
(n) (3)
it follows that the energies E 2 : n = 0, 1, 2, 3 and flux F 2 corresponding to
Mc∗ likewise satisfy:

(0) (0) (1) (2) (3)
2
sup δ E 2 (u) ≤ G( D , D , D , D )
u∈[u0 ,c∗ )

(1) (0) (1) (2) (3)
sup E2 (u) ≤ G( D , D , D , D )
u∈[u0 ,c∗ )

(2) (0) (1) (2) (3)
sup δ −1 E 2 (u) ≤ G( D , D , D , D )
u∈[u0 ,c∗ )

(3) (0) (1) (2) (3)
−3
sup δ E 2 (u) ≤ G( D , D , D , D )
u∈[u0 ,c∗ )
and:
(3) (0) (1) (2) (3)
sup δ −3 F 2 (u) ≤ G( D , D , D , D )
u∈[0,min{δ,c∗ −u0 })
416
hence the quantity P2 corresponding to Mc∗ satisfies:
(0) (1) (2) (3)
P2 ≤ G( D , D , D , D ) (12.257)
Thus also the 3rd condition for membership in the set A holds for c∗ , conse-
quently c∗ ∈ A. Form this point and for the next four chapters our spacetime
manifold shall be (Mc∗ , g) with c∗ defined as in the statement of Theorem 12.1.
Since 2nd condition above holds with c∗ in role of c, the assumptions of
Lemmas 12.5 and 12.6 hold for Mc′∗ . Since also the 3rd condition holds with c∗
in the role of c, the conclusion of Lemma 12.6 yields:
(0) (1) (2) (3)
′4
(Q′2 )2 ≤ C max{G( D , D , D , D ), (D[1] (α))2 } (12.258)
More precisely, to obtain the results of Lemmas 12.5 and 12.6 for Mc′∗ , we first
apply Lemmas 12.5 and 12.6 with the subdomain Mc′∗ ,ε of Mc′∗ , defined by the
restrictions u ≤ δ − ε and u + u ≤ c∗ − ε, in the role of Mc′∗ , the L2 norms
on Cu and C u in the definitions of the quantities Q1 and Q′2 being replaced
by the corresponding L2 norms on Cu Mc′∗ ,ε and C u Mc′∗ ,ε . The quantities
T T
Q1 and Q′2 referring to Mc′∗ ,ε so defined, are finite, since Mc′∗ ,ε is a compact
subdomain of Mc′∗ and we have a smooth solution in canonical coordinates in
Mc′∗ . Lemmas 12.5 and 12.6 then give us bounds for these quantities in terms
of the quantities P1 and P2 referring to Mc′∗ , and these bounds are independent
of ε. Taking then the limit ε → 0 we obtain the results of Lemmas 12.5 and
12.6 for Mc′∗ itself.
As we have already remarked, the quantity Q′2 bounds all the curvature
norms which enter the estimates for the connection coefficients. These esti-
mates depend additionally on the initial data quantities D0∞ , D /41 and D/42 (trχ),
D
/3 (trχ). Thus if in a non-negative non-decreasing continuous function of the
quantities R∞0 , R/41 , R
/2 , and the quantities D0∞ , D
/41 , D
/42 (trχ), D
/3 (trχ), the first
four quantities are first replaced by their bounds in terms of Q′2 , (see 12.147,
12.148) and then Q′2 is replaced by its bound 12.258, the value of the function
will not be decreased and a non-negative non-decreasing continuous function of
(n)
′4
the quantities D : n = 0, 1, 2, 3; D[1] (α) and the quantities D0∞ , D
/41 , D
/42 (trχ),
D
/3 (trχ) will result (see last section of Chapter 3). Moreover, the same holds if
the original function depends additionally on R /2 (α), the quantities:
R40 (D̂α), R40 (Dβ), R40 (Dρ), R40 (Dσ), R40 (Dβ); R40 (D̂α),
the quantities:
R
/1 (D̂α), R
/1 (Dβ), R
/1 (Dρ), R
/1 (Dσ), R
/1 (Dβ); R
/1 (D̂α),
as well as the quantities:

2
R0 (D̂2 α), R0 (D2 β), R0 (D2 ρ), R0 (D2 σ), R(D2 β); R0 (D̂ α)
417
and R/1 (Dβ). We thus denote from now on by O(δ p |u|r ), for real numbers p, r,
the product of δ p |u|r with a non-negative non-decreasing continuous function of
(n)
′4
the quantities D : n = 0, 1, 2, 3; D[1] (α) and D0∞ , D
/41 , D
/42 (trχ), D
/3 (trχ). Also,
we denote simply by O(δ ), for a real number p, the product of δ p with a non-
p
(n)
negative non-decreasing continuous function of the quantities D : n = 0, 1, 2, 3;
′4
D[1] (α) and D0∞ , D
/41 , D
/42 (trχ), D
/3 (trχ). Then, in all the estimates for the con-
nection coefficients (Chapters 3 - 7) and all the estimates for the deformation
tensors of the multiplier fields and the commutation fields (Chapters 8 - 9),
the symbol O(δ p |u|r ) may be taken in the new sense. Moreover, the smallness
condition on δ required for these estimates to hold, a smallness condition de-
pending on the quantities R∞ /41 , R
0 , R /2 , and the quantities D0∞ , D
/41 , D
/42 (trχ),
D
/3 (trχ), may be replaced by a smallness condition depending on the quantities
(n)
′4
D : n = 0, 1, 2, 3; D[1] (α) and the quantities D0∞ , D
/41 , D
/42 (trχ), D
/3 (trχ). For,
the former smallness condition may be reduced to the form:
δF (R∞ /41 , R
0 ,R /2 , D0∞ , D
/41 , D
/42 (trχ), D
/3 (tr(χ)) ≤ 1
where F is a non-negative non-decreasing continuous function of its arguments

(see last section of Chapter 3).
Thus, under the assumptions of the theorem all the aforementioned estimates
hold on Mc′∗ . It follows that with Mc′∗ and Dc′ ∗ in the role of M ′ and D′
respectively, none of the inequalities in the bootstrap assumptions is saturated,
(n)
provided that δ is suitably small depending on D : n = 0, 1, 2, 3, D0∞ , D
/41 ,
4
D
/2 (trχ), D
/3 (trχ). In particular, in place of A0 we have:
2 3
≤Ω≤ : in Mc′∗ (12.259)
3 2
in place of D3.1 we have:
|u|ktrχkL∞ (Su,u ) ≤ 3 : ∀(u, u) ∈ Dc′ ∗ (12.260)
and the inequalities of all the other bootstrap assumptions hold on Mc′∗ and
Dc′ ∗ with their right hand sides halved. The aim of the next three chapters and
the first section of the fourth is to show that the quantity P2 corresponding to
Mc′∗ actually satisfies:
1 (0) (1) (2) (3)
P2 ≤ G( D , D , D , D ) (12.261)
2
so the inequality 12.257 is not saturated either. Then in the third section of
Chapter 16 we shall show that the solution extends to a larger domain Mc for
some c > c∗ and the three conditions in the statement of the theorem hold for
the extended solution as well, thereby contradicting the definition of c∗ , unless
c∗ = −1.
Going back to the proof of Lemma 12.5 and replacing assumptions C1.1 -
C1.4, C2.1 - C2.4, C3.1 - C3.5 by the corresponding estimates from Chapter
418
8, which improve the former by a factor of at least δ 1/4 , the secondary conclu-
sions of Lemma 12.5 may be replaced by:
(0)
(R[1] (α))2 ≤ C E 1 +O(δ 2 )
(1)
(R[1] (β))2 ≤ C E 1 +O(δ)
(2)
(R[1] (ρ))2 , (R[1] (σ))2 ≤ C E 1 +O(δ 2 )
(3)
(R[1] (β))2 ≤ C E 1 +O(δ)
(3)
(R[1] (α))2 ≤ C F 1 +O(δ 2 ) (12.262)
Also, going back to the proof of Lemma 12.6 and replacing assumptions C4.1 -
C4.8, C5.1 - C5.4, C6.1 - C6.10 by the corresponding estimates from Chapter
9, which improve the former by a factor of at least δ 1/4 , as well as assumptions
D2.1, D2.2, D3.1, D3.2, D4.1, D4.2, D5, D6, D7, D8, D9.1 - D9.3, D10.1
- D10.4, D11, D12 by the corresponding estimates from Chapters 3 and 4,
which improve the former by a factor of at least δ 1/4 , the secondary conclusions
of Lemma 12.6 may be replaced by:
(0)
(R[2] (α))2 ≤ C E 2 +O(δ)
(1)
(R[2] (β))2 ≤ C E 2 +O(δ)
(2)
(R[2 (ρ))2 , (R[2] (σ))2 ≤ C E 2 +O(δ)
(3)
(R[2] (β))2 ≤ C E 2 +O(δ)
(3)
(R[2] (α))2 ≤ C F 2 +O(δ) (12.263)
Let us now apply Lemma 12.4 to the energy-momentum density vectorfields

12.80, 12.85, 12.90, 12.95, 12.100, 12.105, 12.110, 12.115 and 12.120. Let us
denote the corresponding divergences by:
(0) (1)
τ = divP (R; L, L, L) τ = divP (R; K, L, L)
(2) (3)
τ = divP (R; K, K, L) τ = divP (R; K, K, K) (12.264)
(L) (0) (L) (1)

τ = divP (L̃L R; L, L, L) τ = divP (L̃L R; K, L, L)
(L) (2) (L) (3)
τ = divP (L̃L R; K, K, L) τ = divP (L̃L R; K, K, K)
(12.265)
(O) (0) (O) (1)

X X
τ= divP (L̃Oi R; L, L, L) τ= divP (L̃Oi R; K, L, L)
i i
419
(O) (2) (O) (3)
X X
τ= divP (L̃Oi R; K, K, L) τ= divP (L̃Oi R; K, K, K)
i i
(12.266)
(S) (3)
τ = divP (L̃S R; K, K, K) (12.267)
(LL) (0) (LL) (1)

τ = divP (L̃L L̃L R; L, L, L) τ = divP (L̃L L̃L R; K, L, L)
(LL) (2) (LL) (3)
τ = divP (L̃L L̃L R; K, K, L) τ = divP (L̃L L̃L R; K, K, K)
(12.268)
(OL) (0)
X
τ= divP (L̃Oi L̃L R; L, L, L)
i
(OL) (1)
X
τ= divP (L̃Oi L̃L R; K, L, L)
i
(OL) (2)
X
τ= divP (L̃Oi L̃L R; K, K, L)
i
(OL) (3)
X
τ= divP (L̃Oi L̃L R; K, K, K) (12.269)
i
(OO) (0)
X
τ= divP (L̃Oj L̃Oi R; L, L, L)
i,j
(OO) (1)
X
τ= divP (L̃Oj L̃Oi R; K, L, L)
i,j
(OO) (2)
X
τ= divP (L̃Oj L̃Oi R; K, K, L)
i,j
(OO) (3)
X
τ= divP (L̃Oj L̃Oi R; K, K, K) (12.270)
i,j
(OS) (3)
X
τ= divP (L̃Oi L̃S R; K, K, K) (12.271)
i
and:
(SS) (3)
τ = divP (L̃S L̃S R; K, K, K) (12.272)
(n)
We consider the total 1st order energy-momentum density vecctorfields P 1 :
n = 0, 1, 2, 3:
(0) X
P 1= P (R; L, L, L) + δ 2 P (L̃L R; L, L, L) + P (L̃Oi R; L, L, L)
i
420
(1) X
P 1= P (R; K, L, L) + δ 2 P (L̃L R; K, L, L) + P (L̃Oi R; K, L, L)
i
(2) X
P 1= P (R; K, K, L) + δ 2 P (L̃L R; K, K, L) + P (L̃Oi R; K, K, L)
i
(3) X
P 1= P (R; K, K, K) + δ 2 P (L̃L R; K, K, K) + P (L̃Oi R; K, K, K)
i
+P (L̃S R; K, K, K) (12.273)
(n)
We then consider the total 2nd order energy-momentum vectorfields P 2 : n=
0, 1, 2, 3:
(0) (0) X
P 2= P 1 +δ 4 P (L̃L L̃L R; L, L, L) + δ 2 P (L̃Oi L̃L R; L, L, L)
i
X
+ P (L̃Oj L̃Oi R; L, L, L)
i,j
(1) (1) X
P 2= P 1 +δ 4 P (L̃L L̃L R; K, L, L) + δ 2 P (L̃Oi L̃L R; K, L, L)
i
X
+ P (L̃Oj L̃Oi R; K, L, L)
i,j
(2) (2) X
P 2= P 1 +δ 4 P (L̃L L̃L R; K, K, L) + δ 2 P (L̃Oi L̃L R; K, K, L)
i
X
+ P (L̃Oj L̃Oi R; K, K, K)
i,j
(3) (3) X
P 2= P 1 +δ 4 P (L̃L L̃L R; K, K, K) + δ 2 P (L̃Oi L̃L R; K, K, K)
i
X
+ P (L̃Oj L̃Oi R; K, K, K)
i,j
X
+ P (L̃Oi L̃S R; K, K, K) + P (L̃S L̃S R; K, K, K) (12.274)
i
The total 1st order energy-momentum density vectorfields satisfy:

(n) (n)
div P 1 = τ 1 : n = 0, 1, 2, 3 (12.275)
(n)
where the τ 1 : n = 0, 1, 2, 3 are the total 1st order divergences:
(0) (0) (0) (O) (0)
τ 1 = τ +δ 2 (L) τ + τ
(1) (1) (1) (O) (1)
τ 1 = τ +δ 2 (L) τ + τ
(2) (2) (2) (O) (2)
τ 1 = τ +δ 2 (L) τ + τ
421
(3) (3) (3) (O) (3) (S) (3)
τ 1 = τ +δ 2 (L) τ + τ + τ (12.276)
The total 2nd order energy-momentum density vectorfields satisfy:

(n) (n)
div P 2 = τ 2 : n = 0, 1, 2, 3 (12.277)
(n)
where the τ 2 : n = 0, 1, 2, 3 are the total 2nd order divergences:
(0) (0) (LL) (0) (OL) (0) (OO) (0)
τ 2= τ 1 +δ 4 τ +δ 2 τ + τ
(1) (1) 4 (LL) (1) 2 (OL) (1) (OO) (1)
τ 2= τ 1 +δ τ +δ τ + τ
(2) (2) 4 (LL) (2) 2 (OL) (2) (OO) (2)
τ 2= τ 1 +δ τ +δ τ + τ
(3) (3) 4 (LL) (3) 2 (OL) (3) (OO) (3)
τ 2= τ 1 +δ τ +δ τ + τ
(OS) (3) (SS) (3)
+ τ + τ (12.278)
We apply Lemma 12.4 to 12.277 to obtain, in view of the definitions of the

energies and fluxes 12.81, 12.82, 12.86, 12.87, 12.91, 12.92, 12.96, 12.97, 12.101,
12.102, 12.106, 12.107, 12.111, 12.112, 12.116, 12.117, 12.121, 12.122, and 12.125
- 12.132, for any (u1 , u1 ) ∈ Dc′ ∗ :
(n) (n) (n) Z
(n)
E u1 2 (u1 )− E u1 2 (u0 )+ F u1 2 (u1 ) = τ2 dµg : n = 0, 1, 2, 3 (12.279)
M1
Since the energies and fluxes are integrals, on the Cu and C u respectively, of
non-negative functions this implies:
(n) (n)
Z
(n)
E u1 2 (u1 ) ≤ E 2 (u0 ) + | τ 2 |dµg : n = 0, 1, 2, 3 (12.280)
Mc′ ∗
and:
(n) (n)
Z
u1 (n)
F 2 (u1 ) ≤E 2 (u0 ) + | τ 2 |dµg : n = 0, 1, 2, 3 (12.281)
Mc′ ∗
for all (u1 , u1 ) ∈ Dc′ ∗ .

(n)
Consider 12.280 with u1 fixed. Since the energy E (u1 ) is the integral
over Cu1 of a non-negative function and the inequality 12.280 holds for any
u1 ∈ [0, δ) if u1 ∈ [u0 , c∗ − δ) and any u1 ∈ [0, c∗ − u1 ) if u1 ∈ (c∗ − δ, c∗ ), in the
case c∗ ≥ u0 + δ, any u1 ∈ [0, c∗ − u1 ), in the case c∗ < u0 + δ, it follows that:
(n) (n)
Z
(n)
E 2 (u1 ) ≤ E 2 (u0 ) + | τ 2 |dµg : n = 0, 1, 2, 3 (12.282)
Mc′ ∗
422
(n)
Consider 12.281 with u1 fixed. Since the flux F (u1 ) is the integral over C u1 of a
non-negative function and the inequality 12.281 holds for any u1 ∈ [u0 , c∗ − u1 ),
it follows that:
(n) (n)
Z
(n)
F2 1(u ) ≤ E2 0(u ) + | τ 2 |dµg : n = 0, 1, 2, 3 (12.283)
Mc′ ∗
Multiplying 12.282 by δ 2 in the case n = 0, by 1 in the case n = 1, by δ −1 in

the case n = 2, and by δ −3 in the case n = 3, and taking the supremum over
u1 ∈ [u0 , c∗ ) yields, recalling the definitions 12.139 and 12.246,
(0) (0)
Z
2 (0)
≤
E2 D +δ | τ 2 |dµg
Mc′ ∗
(1) (1)
Z
(1)
E 2≤ D + | τ 2 |dµg
Mc′ ∗
(2) (2)
Z
−1 (2)
E 2≤ D +δ | τ 2 |dµg
Mc′ ∗
(3) (3)
Z
−3 (3)
E 2≤ D +δ | τ 2 |dµg (12.284)
Mc′ ∗
Also, multiplying 12.283 in the case n = 3 by δ −3 and taking the supremum

over u1 ∈ [0, min{δ, c∗ − u0 }) yields, recalling the definition 12.140 and 12.246,
(3) (3)
Z
−3 (3)
F 2 ≤ D +δ | τ 2 |dµg (12.285)
Mc′ ∗
Let us define the exponents qn : n = 0, 1, 2, 3 by:

1 3
q0 = 1, q2 = − , q3 = −
q1 = 0, (12.286)
2 2
Then inequalities 12.284, 12.285 take the form:
(n) (n)
Z
2q (n)
E 2 ≤ D +δ n | τ 2 |dµg : n = 0, 1, 2, 3
Mc′ ∗
(3) (3)
Z
(3)
F 2≤ D +δ 2q3 | τ 2 |dµg (12.287)
Mc′ ∗
In view of the above, we shall obtain a closed system of inequalities for the
(0) (1) (2) (3) (3)
quantities E 2 , E 2 , E 2 , E 2 and F 2 , if we succeed in estimating appropriately
the error integrals:
Z
2qn (n)
δ | τ 2 |dµ/g : n = 0, 1, 2, 3 (12.288)
Mc′ ∗
(0) (1) (2) (3) (3)

in terms of the quantities E 2 , E 2 , E 2 , E 2 , F 2 themselves. This shall be our
aim in the next three chapters.
423
Chapter 13
The Multiplier Error

Estimates
13.1 Preliminaries
Let us introduce the following definitions.
Definition 13.1 Let ξ be a S tensorfield defined on Mc′∗ . We write:
ξ = O∞ (δ r |u|p )
for real numbers r, p, if:
kξkL∞ (Su,u ) ≤ O(δ r |u|p ) : ∀(u, u) ∈ Dc′ ∗
ξ = O4 (δ r |u|p )

1
kξkL4 (Su,u ) ≤ O(δ r |u|p+ 2 ) : ∀(u, u) ∈ Dc′ ∗
ξ = O(δ r |u|p )

1
kξkL2 (Cu ) ≤ O(δ r+ 2 |u|p+1 ) : ∀u ∈ [u0 , c∗ )
424
ξ = O(δ r |u|p )

3
k|u|−p− 2 ξkL2 (C u ) ≤ O(δ r ) : ∀u ∈ [0, δ)
In estimating the error integrals 12.288 we shall make repeated use of the
following lemma.
Lemma 13.1 Let {ξ1 , ξ2 , ξ3 } be a scalar trilinear expression in the S ten-

sorfields ξ1 , ξ2 , ξ3 , defined on Mc′∗ , with coefficients depending only on g/ and /ǫ .
Consider the following five cases.
Case 1: ξ1 = O∞ (δ r1 |u|p1 ), ξ2 = O(δ r2 |u|p2 ), ξ3 = O(δ r3 |u|p3 )
Case 4: ξ1 = O4 (δ r1 |u|p1 ), ξ2 = O4 (δ r2 |u|p2 ), ξ3 = O(δ r3 |u|p3 )
Case 5: ξ1 = O4 (δ r1 |u|p1 ), ξ2 = O4 (δ r2 |u|p2 ), ξ3 = O(δ r3 |u|p3 )
Then in each of the five cases
p1 + p2 + p3 + 3 < 0
implies: Z
|{ξ1 , ξ2 , ξ3 }|dµg ≤ O(δ r1 +r2 +r3 +1 )
Mc′
Proof: We begin by remarking that there is a numerical constant C such that

we have, pointwise,
|{ξ1 , ξ2 , ξ3 }| ≤ C|ξ1 ||ξ2 ||ξ3 | (13.1)
We shall show that in each of the five cases
p1 + p2 + p3 + 3 < 0
implies: Z
|ξ1 ||ξ2 ||ξ3 |dµg ≤ O(δ r1 +r2 +r3 +1 ) (13.2)
Mc′ ∗
Consider first Case 1. In this case we estimate:

Z Z c∗ Z
|ξ1 ||ξ2 ||ξ3 |dµg ≤ C |ξ1 ||ξ2 ||ξ3 | du
Mc′ ∗ u0 Cu
Z c∗ Z
r1 p1
≤ O(δ ) |u| |ξ2 ||ξ3 | du
u0 Cu
Z c∗
≤ O(δ r1 ) |u|p1 kξ2 kL2 (Cu ) kξ3 kL2 (Cu ) du
u0
425
Z c∗
1 1
r1
≤ O(δ ) |u|p1 O(δ r2 + 2 |u|p2 +1 )O(δ r3 + 2 |u|p3 +1 )du
u0
Z c∗
= O(δ r1 +r2 +r3 +1 ) |u|p1 +p2 +p3 +2 du
u0
≤ O(δ r1 +r2 +r3 +1 ) : if p1 + p2 + p3 + 3 < 0 (13.3)
Consider next Case 2. In this case we write:

Z
|ξ1 ||ξ2 ||ξ3 |dµg (13.4)
Mc′ ∗
Z !1/2 Z !1/2
2p3 +3 2 2 −2p3 −3 2
≤ |u| |ξ1 | |ξ2 | dµg |u| |ξ3 | dµg
Mc′ ∗ Mc′ ∗
and we have:
Z Z min{δ,c∗ −u0 }
3
|u|−2p3 −3 |ξ3 |2 dµg ≤ C k|u|−p3 − 2 ξ3 k2L2 (C ) du
u
Mc′ ∗ 0
≤ O(δ 2r3 +1 ) (13.5)
while:
Z Z c∗ Z
|u|2p3 +3 |ξ1 |2 |ξ2 |2 dµg ≤ C |u|2p3 +3 |ξ1 |2 |ξ2 |2 du
Mc′ ∗ u0 Cu
Z c∗
≤ O(δ 2r1 ) |u|2p1 +2p3 +3 kξ2 k2L2 (Cu ) du
u0
Z c∗
2r1 +2r2 +1
≤ O(δ ) |u|2(p1 +p2 +p3 )+5 du
u0
≤ O(δ 2r1 +2r2 +1 ) : if p1 + p2 + p3 + 3 < 0 (13.6)
Substituting 13.5 and 13.6 in 13.4 yields again 13.2.

Consider next Case 3. In this case we estimate:
Z
|ξ1 ||ξ2 ||ξ3 |dµ/g ≤ sup |u|p2 +p3 +3 |ξ1 |

(13.7)
Mc′ ∗ Mc′ ∗
Z !1/2 Z !1/2
−2p2 −3 2 −2p3 −3 2
· |u| |ξ2 | dµg |u| |ξ3 | dµg
Mc′ ∗ Mc′ ∗
and we have:
n o
sup |u|p2 +p3 +3 |ξ1 | = |u|p2 +p3 +3 kξ1 kL∞ (Su,u )

sup
Mc′ ∗ (u,u)∈Dc′ ∗
≤ O(δ r1 ) |u|p1 +p2 +p3 +3

sup
u∈[u0 ,c∗ )
≤ O(δ r1 ) : if p1 + p2 + p3 + 3 ≤ 0 (13.8)
426
while the two integrals on the right in 13.7 can be estimated in a similar manner
to 13.5 above. Substituting then in 13.7 yields again 13.2.
Consider next Case 4. In this case we write:
Z Z c∗ Z
|ξ1 ||ξ2 ||ξ3 | ≤ C |ξ1 ||ξ2 ||ξ3 | du
Mc′ ∗ u0 Cu
Z c∗
≤C k|ξ1 ||ξ2 |kL2 (Cu ) kξ3 kL2 (Cu ) du (13.9)
u0
Now,
Z (Z )
k|ξ1 ||ξ2 |k2L2 (Cu ) = |ξ1 |2 |ξ2 |2 du
Su,u
Z
≤ kξ1 k2L4 (Su,u ) kξ2 k2L4 (Su,u ) du
≤ δO(δ 2r1 |u|2p1 +1 )O(δ 2r2 |u|2p2 +1 )

= O(δ 2(r1 +r2 )+1 |u|2(p1 +p2 +1) ) (13.10)
and:
1
kξ3 kL2 (Cu ) ≤ O(δ r3 + 2 |u|p3 +1 ) (13.11)
Hence the right hand side of 13.9 is bounded by:
Z c∗
O(δ r1 +r2 +r3 +1 ) |u|p1 +p2 +p3 +2 du
u0
≤ O(δ r1 +r2 +r3 +1 ) : if p1 + p2 + p3 + 3 < 0 (13.12)
thus 13.2 results once again.

Consider finally Case 5. In this case we write:
Z
|ξ1 ||ξ2 ||ξ3 |dµg (13.13)
Mc′ ∗
Z !1/2 Z !1/2
2p3 +3 2 2 −2p3 −3 2
≤ |u| |ξ1 | |ξ2 | dµg |u| |ξ3 | dµg
Mc′ ∗ Mc′ ∗
We have:
Z Z c∗ Z
2p3 +3 2 2 2p3 +3 2 2
|u| |ξ1 | |ξ2 | dµg ≤ C |u| |ξ1 | |ξ2 | du
Mc′ ∗ u0 Cu
Z c∗
2(r1 +r2 )+1
≤ O(δ ) |u|2(p1 +p2 +p3 )+5 du
u0
≤ O(δ 2(r1 +r2 )+1 ) : if p1 + p2 + p3 + 3 < 0

(13.14)
427
where we have substituted for
Z
|ξ1 |2 |ξ2 |2 = k|ξ1 ||ξ2 |k2L2 (Cu )
Cu
the bound 13.10. Substituting 13.14 and 13.5 in 13.13 yields again 13.2.
Let us recall from Chapter 12 that the Weyl fields we are considering are:
0th order: W =R
1st order: W = L̃L R, W = L̃Oi R : i = 1, 2, 3, W = L̃S R,
2nd order: W = L̃L L̃L R, W = L̃Oi L̃L R : i = 1, 2, 3,
W = L̃Oj L̃Oi R : i, j = 1, 2, 3,
W = L̃Oi L̃S R : i = 1, 2, 3, W = L̃S L̃S R (13.15)
We assign to such a Weyl field W the index l which is the number of L̃L operators
in the definition of W in terms of R. Thus:
l = 0 for W = R, W = L̃Oi R : i = 1, 2, 3, W = L̃S R

and for W = L̃Oj L̃Oi R : i, j = 1, 2, 3,
W = L̃Oi L̃S R : i = 1, 2, 3, W = L̃S L̃S R
l = 1 for W = L̃L R, W = L̃Oi L̃L R : i = 1, 2, 3
l = 2 for W = L̃L L̃L R (13.16)
(n) (n) (n) (n)
Let us also recall from 12.276, 12.278 that in τ 1 , τ 2 , (X) τ and (Y X) τ
enter multiplied by the factor δ 2l where l is the number of L’s in (X) and in
(n)
(X) τ we have l = 1 for (X) = (L),
(Y X) respectively. Thus in the case of
(n)
l = 0 for (X) = (O) and (X) = (S), and in the case of (Y X) τ we have l = 2 for
(Y X) = (LL), l = 1 for (Y X) = (OL), l = 0 for (Y X) = (OO), (Y X) = (OS)
(n)
and (Y X) = (SS). In the case of τ itself we have l = 0. Bounds for the error
integrals:
Z
(n)
0th order: δ 2qn +2l | τ |dµg (13.17)
Mc′ ∗
Z
(X) (n)
1st order: δ 2qn +2l | τ |dµg
Mc′ ∗
: for (X) = (L), (O), (S)

Z
(Y X) (n)
2nd order: δ 2qn +2l | τ |dµg
Mc′ ∗
: for (Y X) = (LL), (OL), (OO), (OS), (SS)
then imply corresponding bounds for the error integrals 12.288.
428
From 12.83, 12.84, 12.88, 12.89, 12.93, 12.94, 12.98, 12.99, 12.103, 12.104,
12.108, 12.109, 12.113, 12.114, 12.118, 12.119, 12.123, 12.124, 12.125, 12.126,
12.128, 12.129, 12.130, 12.132, 12.139, 12.140, 12.141, in conjunction with the
inequality 12.257, we have:
1
α(W ) = O(δ − 2 −q0 −l |u|−1 )
1
β(W ) = O(δ − 2 −q1 −l |u|−2 )
1
(ρ, σ)(W ) = O(δ − 2 −q2 −l |u|−3 )
1
β(W ) = O(δ − 2 −q3 −l |u|−4 ) (13.18)
and:
α(W ) = O(δ −q3 −l |u|−9/2 ) (13.19)
for all Weyl fields W in 13.15 except for the Weyl fields
W = L̃S R, W = L̃Oi L̃S R : i = 1, 2, 3, W = L̃S L̃S R
and 13.19 and the last of 13.18 hold for these three Weyl fields as well.
13.2 The multiplier error estimates

Let us recall from 12.64 that τ (W ; X, Y, Z), the divergence of the energy-
momentum density vectorfield P (W ; X, Y, Z) associated to the Weyl field W
and the multiplier fields X, Y, Z is expressed as the sum:
τ (W ; X, Y, Z) = τc (W ; X, Y, Z) + τm (W ; X, Y, Z) (13.20)
where:
1 (X) αβ
τm (W ; X, Y, Z) = − Q(W )αβγδ ( π̃ Y γZδ + (Y ) αβ
π̃ XγZδ + (Z) αβ
π̃ XγY δ)
2
(13.21)
is generated by the deformation tensors of the multiplier fields, while:
τc (W ; X, Y, Z) = −(divQ(W ))(X, Y, Z) (13.22)
is generated (see Proposition 12.6) by the Weyl current J corresponding to W .
Thus the divergences
(n)
0th order: τ : n = 0, 1, 2, 3,
(X) (n)
1st order: τ : (X) = (L), (O), (S); n = 0, 1, 2, 3
(Y X) (n)
2nd order: τ : (Y X) = (LL), (OL), (OO), (OS), (SS); n = 0, 1, 2, 3
correspondingly split into the sums:
(n) (n) (n)
τ =τ c + τ
m
(X) (n) (n) (n)
τ = (X) τ c + (X) τ m
(Y X) (n) (n) (n)
τ = (Y X) τ c + (Y X) τ m (13.23)
429
In the present chapter we shall estimate the contributions of the multiplier error
(n) (n) (n)
terms τ m , (X) τ m , (Y X) τ m, to the error integrals 13.17. These error terms
are all of the form:
(0) 3 (L) µν
τm (W ) = − π̃ Qµνκλ (W )Lκ Lλ
2
(1) (L) µν 1
τm (W ) = − π̃ Qµνκλ (W )K κ Lλ − (K) µν
π̃ Qµνκλ (W )Lκ Lλ
2
(2) 1 (L) µν
τm (W ) = − π̃ Qµνκλ (W )K κ K λ − (K) µν
π̃ Qµνκλ (W )K κ Lλ
2
(3) 3 (K) µν
τm (W ) = − π̃ Qµνκλ (W )K κ K λ (13.24)
2
where W is each one of the Weyl fields 13.15. The integrals to be estimated,
the multiplier error integrals, are then:
Z
2qn +2l (n)
δ | τ m (W )|dµg (13.25)
Mc′ ∗
where l is the index associated to W .

Expanding the above expressions in the frame (eµ : µ = 1, 2, 3, 4) using
8.11, 8.12 and Lemma 12.2 we find:
(0) n
τ m (W ) = −3Ω2 2 (L) j|β(W )|2 − 2 (L) m♯ · α(W ) · β ♯ (W )
o
+ρ(W )( (L) î, α(W )) − σ(W ) (L) î ∧ α(W ) (13.26)
(1)
n
τm (W ) = −2Ω2 2|u|2 (L)
j((ρ(W ))2 + (σ(W ))2 )
−2|u|2 (L)
m♯ · (ρ(W )β(W ) − σ(W ) ∗ β(W ))
−|u|2 ( (L) î, β(W )⊗β(W
ˆ ))
(K)
+ j|β(W )|2 − (K)
m♯ · (ρ(W )β(W ) − σ(W ) ∗ β(W ))

1 (K) 1 (K)
+ ρ(W )( î, α(W )) − σ(W ) î ∧ α(W ) (13.27)
2 2
(2)
n
τm (W ) = −2Ω2 |u|4 (L)
j|β(W )|2
+|u|4 (L) m♯ · (ρ(W )β(W ) + σ(W ) ∗ β(W ))
1 1
+ |u|4 ρ(W )( (L) î, α(W )) + |u|4 σ(W ) (L) î ∧ α(W )
2 2
+2|u|2 (K) j((ρ(W ))2 + (σ(W ))2 )
+2|u|2 (K) m♯ · (ρ(W )β(W ) + σ(W ) ∗ β(W ))
o
−|u|2 ( (K) î, β(W )⊗β(W
ˆ ))
(13.28)
430
(3)
n
τm (W ) = −3Ω2 2|u|4 (K)
j|β(W )|2 + 2|u|4 m♯ · α(W ) · β ♯ (W )
(K)
o
+|u|4 ρ(W )( (K) î, α(W )) + |u|4 σ(W ) (K) î ∧ α(W )
(13.29)
In the above expressions each term is a trilinear expression {ξ1 , ξ2 , ξ3 } with co-
efficients depending only on g/ and /ǫ , where ξ is a component of the deformation
tensors of L or K, multiplied by Ω2 and the appropriate power of |u|2 , while ξ1
and ξ2 are components of W . Recalling from Chapter 8 that the components of
the deformation tensors of L and K satisfy the L∞ bounds 8.32 and 8.33, we
have:
(L)
î = O∞ (δ −1/2 |u|−1 )
(L)
j = O∞ (|u|−1 )
(L)
m = O∞ (δ 1/2 |u|−2 ) (13.30)
and:
(K)
î = O∞ (δ 1/2 )
(K)
j = O∞ (δ)
(K)
m = O∞ (δ 1/2 ) (13.31)
while the components of W satisfy 13.18, 13.19. Thus to each term on the
right in 13.26 - 13.29 we can apply one of the first two cases of Lemma 13.1.
The third case does not occur because there are no terms involving two α(W )
factors. Comparing with what is required to obtain a bound for 13.25 by O(1),
we define the excess index e of the contribution of a given term to 13.25 by:
e = 2qn + 2l + r1 + r2 + r3 + 1 (13.32)
Then the contribution of the given term to 13.25 is bounded by O(δ e ), provided
that the integrability index s of that term, defined by:
s = p1 + p2 + p3 + 3 (13.33)
is negative so that Lemma 13.1 applies. We obtain in this way the following
tables. The ordinals in these tables refer to the terms on the right hand side of
each of 13.26 - 13.29. We consider a pair of similar terms one involving ρ(W )
and the other σ(W ) as a single term. Thus, we consider the 2nd line of 13.26,
the 5th line of 13.27, the 3rd line of 13.28 and the 2nd line of 13.29, each as a
single term.
Case n = 0:
term e s
1st 2 −2
(13.34)
2nd 3/2 −2
3rd 1 −2
431
Case n = 1:
term e s
1st 1 −2
2nd 1 −2
3rd 1 −2 (13.35)
4th 1 −1
5th 1 −2
6th 0 −1
Case n = 2:
term e s
1st 2 −2
2nd 3/2 −2
3rd 1 −3/2 (13.36)
4th 1 −1
5th 3/2 −2
6th 1 −1
Case n = 3:
term e s
1st 1 −1
(13.37)
2nd 1 −3/2
3rd 0 −1/2
We see that all terms have negative integrability index so Lemma 13.1 indeed
applies. All terms have non-negative excess index. The terms with vanishing
excess index play a crucial role because they give rise to borderline error inte-
grals. These must be analyzed in more detail. They occur only in the cases
n = 1 and n = 3. The borderline terms are the 6th term in 13.27 and the 3rd
term in 13.29.
Consider the 6th term in 13.27. Replacing the first of 13.31 by the more
precise statement:
(K)
k îkL∞ (Su,u ) ≤ Cδ 1/2 (D0∞ (χ̂) + O(δ)) : ∀(u, u) ∈ Dc′ ∗ (13.38)
implied by the first of 8.33, replacing also the first and third of 13.18 by the
more precise statements:
(0)
kα(W )kL2 (Cu ) ≤ C( E 2 )1/2 δ −1−l : ∀u ∈ [u0 , c∗ )
(2)
k(ρ, σ)(W )kL2 (Cu ) ≤ C( E 2 )1/2 δ 1/2−l |u|−2 : ∀u ∈ [u0 , c∗ ) (13.39)
implied by the first and third of each of 12.83, 12.88, 12.93, 12.103, 12.108,
12.113, 12.125, 12.129, and 12.139, and following the proof of Case 1 of Lemma
432
13.1, we deduce that the contribution of the term in question to the error integral
13.25 with n = 1 is bounded by:
(0) (2)
CD0∞ (χ̂)( E 2 )1/2 ( E 2 )1/2 + O(δ) (13.40)
Consider the 3rd term in 13.29. Replacing again the first of 13.31 by 13.38,
the third of 13.18 by the second of 13.39, and 13.19 by the more precise state-
ment:
(3)
k|u|3 α(W )kL2 (C u ) ≤ C( F 2 )1/2 δ 3/2 : ∀u ∈ [0, δ) (13.41)
implied by 12.84, 12.89, 12.94, 12.99, 12.104, 12.109, 12.114, 12.119, 12.124,
12.128, 12.132, 12.140, and following the proof of Case 2 of Lemma 13.1, we
deduce that the contribution of the term in question to the error integral 13.25
with n = 3 is bounded by:
(2) (3)
CD0∞ (χ̂)( E 2 )1/2 ( F 2 )1/2 + O(δ) (13.42)
However, the above have not been shown to hold in the case of the three
Weyl fields which involve S:
W = L̃S R, W = L̃Oi L̃S R : i = 1, 2, 3, W = L̃S L̃S R
For these only the case n = 3 occurs so only 13.29 is to be considered. The
point here is that the second of 13.39 does not quite hold for these three Weyl
fields. The aim of the remainder of the present chapter is to show that for these
three Weyl fields it nevertheless holds:
(2)
k(ρ, σ)(W )kL2 (Cu ) ≤ Cδ 1/2 |u|−2 (( E 2 )1/2 + O(δ 1/2 )) (13.43)
(here l = 0) so that 13.42 is simply replaced by:

(2) (3)
CD0∞ (χ̂)( E 2 )1/2 ( F 2 )1/2 + O(δ 1/2 ) (13.44)
Consider first the case W = L̃S R. From Proposition 12.2 we have:
(S) 3 (S) 1 1
ρ(L̃S R) = Sρ + 3 νρ +jρ − (S) m♯ · β + (S) m♯ · β
4 2 2
3 1 1
σ(L̃S R) = Sσ + 3 (S) νσ + (S) jσ + ∗(S) m♯ · β + ∗(S) m♯ · β
4 2 2
(13.45)
In view of the estimates 8.34 together with the fact that by 12.37
(S)
ν − 1 = S log Ω = uω + uω (13.46)
and the estimates of Chapter 3 we have:

(S)
k ν − 1kL∞ (Su,u ) ≤ O(δ|u|−2 ) (13.47)
433
we deduce:
kρ(L̃S R)kL2 (Cu ) ≤ kSρkL2(Cu ) + 3kρkL2(Cu ) + O(δ 3/2 |u|−3 )

kσ(L̃S R)kL2 (Cu ) ≤ kSσkL2 (Cu ) + 3kσkL2 (Cu ) + O(δ 3/2 |u|−3 ) (13.48)
We have:
(Sρ, Sσ) = (uDρ + uDρ, uDσ + uDσ) (13.49)
and:
kuDρkL2 (Cu ) ≤ δkDρkL2 (Cu ) ≤ δ 1/2 |u|−2 R[1] (ρ)

(2)
1/2 −2 1/2
≤ δ |u| C( E 1 ) + O(δ)
kuDσkL2 (Cu ) ≤ δkDσkL2 (Cu ) ≤ δ 1/2 |u|−2 R[1] (σ)

(2)
≤ δ 1/2 |u|−2 C( E 1 )1/2 + O(δ) (13.50)
by 12.262. To obtain appropriate estimates for kuDρkL2 (Cu ) , kuDσkL2 (Cu ) , we

consider the eighth and tenth of the Bianchi identities, given by Proposition 1.2:
3 1
Dρ + Ωtrχρ = −Ω{div
/ β + (2η − ζ, β) + (χ̂, α)}
2 2
3 ∗ 1
Dσ + Ωtrχσ = −Ω{curl
/ β + (2η − ζ, β) + χ̂ ∧ α} (13.51)
2 2
By the results of Chapter 3 the right hand sides are bounded in L2 (Cu ) by
O(δ 3/2 |u|−4 ). Hence we obtain:
kuDρkL2 (Cu ) ≤ CkρkL2 (Cu ) + O(δ 3/2 |u|−3 )

kuDσkL2 (Cu ) ≤ CkσkL2 (Cu ) + O(δ 3/2 |u|−3 ) (13.52)
Combining 13.50 and 13.52 and taking into account the fact that:
(2)
kρkL2 (Cu ) , kσkL2 (Cu ) ≤ Cδ 1/2 |u|−2 ( E 0 )1/2 (13.53)
(see third of 12.83, 12.133) we conclude that:

(2)
kSρkL2(Cu ) , kSσkL2 (Cu ) ≤ Cδ 1/2 |u|−2 ( E 1 )1/2 + O(δ 3/2 |u|−2 ) (13.54)
Substituting 13.54 and 13.53 in 13.48 then yields:

(2)
k(ρ, σ)(L̃S R)kL2 (Cu ) ≤ Cδ 1/2 |u|−2 ( E 1 )1/2 + O(δ 3/2 |u|−2 ) (13.55)
Consider next the case W = L̃Oi L̃S R : i = 1, 2, 3. From Proposition 12.2 we

have:
(Oi ) 3 (Oi )
ρ(L̃Oi L̃S R) = Oi ρ(L̃S R) + 3 νρ(L̃S R) + jρ(L̃S R)
4
434
1 (Oi )
− m♯ · β(L̃S R)
2
(Oi ) 3 (Oi )
σ(L̃Oi L̃S R) = Oi σ(L̃S R) + 3 νσ(L̃S R) + jσ(L̃S R)
4
1 ∗(Oi )
+ m♯ · β(L̃S R) (13.56)
2
In view of the estimates 8.144 together with the fact that by 12.37
(Oi )
ν = Oi (log Ω) (13.57)
and the estimate 7.220 we have:

(Oi )
k νkL∞ (Su,u ) ≤ O(δ|u|−2 ) (13.58)
we deduce, taking also into account 13.55,
kρ(L̃Oi L̃S R)kL2 (Cu ) ≤ kOi ρ(L̃S R)kL2 (Cu ) + O(δ 3/2 |u|−4 )
kσ(L̃Oi L̃S R)kL2 (Cu ) ≤ kOi σ(L̃S R)kL2 (Cu ) + O(δ 3/2 |u|−4 ) (13.59)
Moreover, applying Oi to the expressions 13.45 and using the estimates 9.12 we
deduce:
kOi ρ(L̃S R)kL2 (Cu ) ≤ kOi SρkL2 (Cu ) + 3kOi ρkL2 (Cu ) + O(δ 3/2 |u|−3 )
kOi σ(L̃S R)kL2 (Cu ) ≤ kOi SσkL2 (Cu ) + 3kOi σkL2 (Cu ) + O(δ 3/2 |u|−3 ) (13.60)
We have:
(Oi Sρ, Oi Sσ) = (uOi Dρ + uOi Dρ, uOi Dσ + uOi Dσ) (13.61)
and:
kuOi DρkL2 (Cu ) ≤ δkOi DρkL2 (Cu ) ≤ δ 1/2 |u|−2 R[2] (ρ)

(2)
≤ δ 1/2 |u|−2 C( E 2 )1/2 + O(δ 1/2 )
kuOi DσkL2 (Cu ) ≤ δkDσkL2 (Cu ) ≤ δ 1/2 |u|−2 R[2] (σ)

(2)
1/2 −2 1/2 1/2
≤ δ |u| C( E 2 ) + O(δ ) (13.62)
by 12.263. To obtain appropriate estimates for kuOi DρkL2 (Cu ) , kuOi DσkL2 (Cu ) ,
we apply Oi to the Bianchi identities 13.51. Using the results of Chapters 3 and
4 we then deduce:
kuOi DρkL2 (Cu ) ≤ CkOi ρkL2 (Cu ) + O(δ 3/2 |u|−3 )

kuOi DσkL2 (Cu ) ≤ CkOi σkL2 (Cu ) + O(δ 3/2 |u|−3 ) (13.63)
435
Combining 13.62 and 13.63 and taking into account the fact that by 12.262:

(2)
kOi ρkL2 (Cu ) ≤ Cδ 1/2 |u|−2 R[1] (ρ) ≤ Cδ 1/2 |u|−2 ( E 1 )1/2 + O(δ)

(2)
kOi σkL2 (Cu ) ≤ Cδ 1/2 |u|−2 R[1] (σ) ≤ Cδ 1/2 |u|−2 ( E 1 )1/2 + O(δ) (13.64)
we conclude that:
(2)
kOi SρkL2 (Cu ) , kOi SσkL2 (Cu ) ≤ Cδ 1/2 |u|−2 ( E 2 )1/2 + O(δ|u|−2 ) (13.65)
Substituting 13.65 and 13.64 in 13.60 yields:

(2)
kOi ρ(L̃S R)kL2 (Cu ) ≤ Cδ 1/2 |u|−2 ( E 2 )1/2 + O(δ|u|−2 )
(2)
kOi σ(L̃S RkL2 (Cu ) ≤ Cδ 1/2 |u|−2 ( E 2 )1/2 + O(δ|u|−2 ) (13.66)
Substituting these in turn in 13.59 we conclude that:

(2)
k(ρ, σ)(L̃Oi L̃S R)kL2 (Cu ) ≤ Cδ 1/2 |u|−2 ( E 2 )1/2 + O(δ|u|−2 ) (13.67)
Consider finally the case W = L̃S L̃S R. From Proposition 12.2 we have:
(S) 3 (S)
ρ(L̃S L̃S R) = Sρ(L̃S R) + 3 νρ(L̃S R) + jρ(L̃S R)
4
1 (S) 1
− m♯ · β(L̃S R) + (S)
m♯ · β(L̃S R)
2 2
(S) 3 (S)
σ(L̃S L̃S R) = Sσ(L̃S R) + 3 νσ(L̃S R) + jσ(L̃S R)
4
1 ∗(S) 1
+ m♯ · β(L̃S R) + ∗(S)
m♯ · β(L̃S R) (13.68)
2 2
and:
(S) 1 (S) 1
β(L̃S R) = L
/S β + 2 νβ +jβ − (S) î♯ · β
4 2
1 (S) ♯ 3 (S) 3
+ m ·α+ mρ + ∗(S) mσ (13.69)
4 4 4
Taking into account the estimates 8.34 and 13.47 we obtain:
/S βkL2 (Cu ) + 2kβkL2 (Cu ) + O(δ 1/2 |u|−2 )

kβ(L̃S R)kL2 (Cu ) ≤ kL (13.70)
We have:
/S β = uDβ + uDβ
L (13.71)
and:
kuDβkL2 (Cu ) ≤ δkDβkL2 (Cu ) ≤ |u|−1 R[1] (β) ≤ O(|u|−1 ) (13.72)
436
To obtain an appropriate estimate for kuDβkL2 (Cu ) we consider the fifth of the
Bianchi identities of Proposition 1.2:
1
Dβ + Ωtrχβ + ωβ = Ω{χ̂♯ · β + /dρ + ∗ /dσ + 3ηρ + 3 ∗ ησ + 2χ̂♯ · β} (13.73)
2
Using the results of Chapter 3 we then deduce:
kuDβkL2 (Cu ) ≤ O(|u|−1 ) (13.74)
Combining with 13.72 yields:
/S βkL2 (Cu ) ≤ O(|u|−1 )

kL (13.75)
Substituting in 13.70 we then obtain:
kβ(L̃S R)kL2 (Cu ) ≤ O(|u|−1 ) (13.76)
Going back to 13.68 and using the estimates 8.34, 13.47 and 13.76 we deduce:
kρ(L̃S L̃S R)kL2 (Cu ) ≤ kSρ(L̃S R)kL2 (Cu ) + 3kρ(L̃S R)kL2 (Cu ) + O(δ 3/2 |u|−3 )
kσ(L̃S L̃S R)kL2 (Cu ) ≤ kSσ(L̃S R)kL2 (Cu ) + 3kσ(L̃S R)kL2 (Cu ) + O(δ 3/2 |u|−3 )
(13.77)
Applying S to the expressions 13.45 and using the estimates 12.236, 13.75
and the estimates 8.34 as well as the estimates:
(S)
kL̂
/S îkL4 (Su,u ) ≤ O(δ 1/2 |u|−1/2 )
(S)
kS jkL4 (Su,u ) ≤ O(δ|u|−3/2 )
(S)
kL
/S mkL4 (Su,u ) ≤ O(δ 3/2 |u|−3/2 )
(S)
kL
/S mkL4 (Su,u ) ≤ O(δ 1/2 |u|−1/2 ) (13.78)
which follow by combining the estimates 9.13 and 9.14, and in addition the
estimate:
kS (S) νkL4 (Su,u ) ≤ O(δ|u|−3/2 ) (13.79)
which follows from the estimates of Chapter 4 (see 13.46), we deduce:
kSρ(L̃S R)kL2 (Cu ) ≤ kS 2 ρkL2 (Cu ) + 3kSρkL2(Cu ) + O(δ 3/2 |u|−3 )

kSσ(L̃S R)kL2 (Cu ) ≤ kS 2 σkL2 (Cu ) + 3kSσkL2 (Cu ) + O(δ 3/2 |u|−3 )
(13.80)
Now, for any function f defined on Mc′∗ we have:
S 2 f = u2 D2 f + uu(DDf + DDf ) + u2 D2 f + Sf (13.81)
437
In particular, this holds for the functions ρ, σ. We have:
ku2 D2 ρkL2 (Cu ) ≤ δ 2 kD2 ρkL2 (Cu ) ≤ δ 1/2 |u|−2 R[2] (ρ)

(2)
≤ δ 1/2 |u|−2 C( E 2 )1/2 + O(δ 1/2 )
ku2 D2 σkL2 (Cu ) ≤ δ 2 kD2 σkL2 (Cu ) ≤ δ 1/2 |u|−2 R[2] (σ)

(2)
≤ δ 1/2 |u|−2 C( E 2 )1/2 + O(δ 1/2 ) (13.82)
To obtain appropriate estimates for
kuuDDρkL2 (Cu ) , kuuDDσkL2 (Cu ) , ku2 D2 ρkL2 (Cu ) , ku2 D 2 σkL2 (Cu )
we consider again the Bianchi identities 13.51. Applying D to these identities

we obtain:
3 3
DDρ + ΩtrχDρ + D(Ωtrχ)ρ
2 2
= −Ω div/ Dβ + (2Dη − Dζ, β) + (2η − ζ, Dβ)

1 1
+ (D̂ χ̂, α) + (χ̂, D̂α)
2 2
♯

+Ω 2div
/ (Ωχ̂ · β) + Ωtrχdiv/ β
o
+Ωtrχ(2η − ζ, β) + 2Ω(2η − ζ)♯ · χ̂ · β ♯ + Ωtrχ(χ̂, α)

1
−Ωω div / β + (2η − ζ, β) + (χ̂, α)
2
3 3
DDσ + ΩtrχDσ + D(Ωtrχ)σ
2 2
/ Dβ + (2Dη − Dζ, ∗ β) + (2η − ζ, ∗ Dβ)
= −Ω curl

1 1
+ D̂ χ̂ ∧ α + χ̂ ∧ D̂α
2 2
/ β + (2η − ζ, ∗ β) + χ̂ ∧ α

+Ω(Ωtrχ) curl

1
−Ωω curl / β + (2η − ζ, ∗ β) + χ̂ ∧ α (13.83)
2
To derive the above formulas we have made use of the following eight facts.
First, the commutation formula 6.107. Second, the fact that for any pair of
trace-free symmetric 2-covariant S tensorfields θ, θ′ we have:
D(θ, θ′ ) = (D̂θ, θ′ ) + (θ, D̂θ′ ) − 2Ωtrχ(θ, θ′ ) (13.84)
This is obtained using the identity 1.187 which implies that for any triplet of
trace-free symmetric 2-covariant S tensorfields θ, θ′ , θ′′ we have:
(θ, θ′ × θ′′ ) = 0 (13.85)
438
Third, the fact, which follows from 12.210, that for any S 1-form ξ we have:
D∗ ξ = ∗
Dξ − 2Ω ∗ χ̂♯ · ξ (13.86)
Fourth, the fact that if θ is a trace-free symmetric 2-covariant S tensorfield and

ξ an S 1-form, we have:
∗ ♯
θ · ξ + θ♯ · ∗ ξ = 0 (13.87)
This follows from the symmetry of ∗ θ. Remarking that if ξ is an arbitrary S
1-form we have:
curl / ∗ξ
/ ξ = div (13.88)
and using the formula 6.107 as well as the third and fourth facts just mentioned,
we deduce the commutation formula:
Dcurl
/ ξ − curl
/ Dξ = −Ωtrχcurl
/ ξ (13.89)
for an arbitrary S 1-form ξ. This is the fifth fact. Using the third and fourth
facts we deduce that for any pair of S 1-forms ξ, ξ ′ we have:
D(ξ, ξ ′ ) = (Dξ, ∗ ′
ξ ) + (ξ, ∗
Dξ ′ ) − Ωtrχ(ξ, ξ ′ ) (13.90)
This is the sixth fact. Finally, remarking that for any pair of trace-free sym-
metric 2-covariant S tensorfields θ, θ′ we have:
θ ∧ θ′ = (θ, ∗ ′
θ) (13.91)
and using the second of the above facts as well as the commutation formula
12.209, which we think of as the seventh fact, we deduce:
D(θ ∧ θ′ ) = D̂θ ∧ θ′ + θ ∧ D̂θ′ − 2Ωtrχθ ∧ θ′ (13.92)
This is the eighth fact.

We express in equations 13.83, D(Ωtrχ) by 4.3, Dη by 1.82, Dζ by 1.76,
that is:
Dζ = −d/ω + Ω(χ♯ · η − β) (13.93)
and D̂χ̂ by 3.10. We also remark that if we use, in place of the bootstrap
assumptions, the results of Chapters 3 and 4 (which improve the former by a
factor of at least δ 1/4 ) to derive from the Bianchi identity 12.168 an estimate
for kD̂αkL4 (Su,u ) , we obtain, in place of 12.174, the estimate:
kD̂αkL4 (Su,u ) ≤ O(δ 1/2 |u|−4 ) (13.94)
Then the right hand sides of 13.83 are seen to be bounded in L2 (Cu ) by
O(δ 1/2 |u|−4 and we obtain:
kuuDDρkL2 (Cu ) ≤ 3kuDρkL2 (Cu ) + O(δ 3/2 |u|−3 )

≤ 3δ 1/2 |u|−2 R[1] (ρ) + O(δ 3/2 |u|−3 )
439

(2)
≤ δ 1/2 |u|−2 C( E 1 )1/2 + O(δ)
kuuDDσkL2 (Cu ) ≤ 3kuDσkL2 (Cu ) + O(δ 3/2 |u|−3 )

≤ 3δ 1/2 |u|−2 R[1] (σ) + O(δ 3/2 |u|−3 )

(2)
1/2 −2 1/2
≤ δ |u| C( E 1 ) + O(δ) (13.95)
by 12.262.
Applying D to the Bianchi identities 13.51 and using the conjugates of the
eight facts mentioned above we obtain:
3 3
D2 ρ + ΩtrχDρ + D(Ωtrχ)ρ
2 2
/ Dβ + (2Dη − Dζ, β) + (2η − ζ, Dβ)
= −Ω div

1 1
+ (D̂ χ̂, α) + (χ̂, D̂α)
2 2
n
/ (Ωχ̂♯ · β) + Ωtrχdiv
+Ω 2div / β
o
+Ωtrχ(2η − ζ, β) + 2Ω(2η − ζ)♯ · χ̂ · β ♯ + Ωtrχ(χ̂, α)

1
−Ωω div/ β + (2η − ζ, β) + (χ̂, α)
2
3 3
D2 σ + ΩtrχDσ + D(Ωtrχ)σ
2 2
/ Dβ + (2Dη − Dζ, ∗ β) + (2η − ζ, ∗ Dβ)

= −Ω curl

1 1
+ D̂ χ̂ ∧ α + χ̂ ∧ D̂α
2 2
/ β + (2η − ζ, ∗ β) + χ̂ ∧ α

+Ω(Ωtrχ) curl

∗ 1
−Ωω curl/ β + (2η − ζ, β) + χ̂ ∧ α (13.96)
2
We express in these equations D(Ωtrχ) by 3.9, Dη by 1.173, Dζ by 1.78, that

is:
− Dζ = −d/ω + Ω(χ♯ · η + β) (13.97)
and D̂χ̂ by 4.6. Then, by the results of Chapters 3 and 4 and the estimate
12.190, the right hand sides of 13.96 are seen to be bounded in L2 (Cu ) by
O(δ 3/2 |u|−5 ) and we obtain:
ku2 D2 ρkL2 (Cu ) ≤ 3kuDρkL2 (Cu ) + 3kρkL2(Cu ) + O(δ 3/2 |u|−3 )

≤ Cδ 1/2 |u|−2 R[1] (ρ) + O(δ 3/2 |u|−3 )

(2)
≤ δ 1/2 |u|−2 C( E 1 )1/2 + O(δ)
440
ku2 D2 σkL2 (Cu ) ≤ 3kuDσkL2 (Cu ) + 3kσkL2(Cu ) + O(δ 3/2 |u|−3 )
≤ Cδ 1/2 |u|−2 R[1] (σ) + O(δ 3/2 |u|−3 )

(2)
≤ δ 1/2 |u|−2 C( E 1 )1/2 + O(δ) (13.98)
by 13.52.
Combining 13.82, 13.95, 13.98 and 13.54, and noting that by the commuta-
tion formula 1.91:
kuu(DD − DD)(ρ, σ)kL2 (Cu ) = k4uuΩ2 ζ ♯ · /d(ρ, σ)kL2 (Cu ) ≤ O(δ 2 |u|−4 ) (13.99)
we conclude through 13.81 that:

(2)
2 1/2 −21/2 1/2
kS ρkL2 (Cu ) ≤ δ |u| C( E 2 ) + O(δ )

(2)
2 1/2 −2 1/2 1/2
kS σkL2 (Cu ) ≤ δ |u| C E 2 ) + O(δ ) (13.100)
Combining this in turn with 13.80, 13.77, 13.55 and 13.55, we conclude that:
(2)
k(ρ, σ)(L̃S L̃S R)kL2 (Cu ) ≤ Cδ 1/2 |u|−2 ( E 2 )1/2 + O(δ|u|−2 ) (13.101)
We have thus established 13.43 for all three Weyl fields involving the commuta-
tion field S.
We summarize the results of this chapter in the following proposition.
Proposition 13.1 The multiplier error integrals 13.25 satisfy the estimates:
Z
(0)
δ 2q0 +2l | τ m |dµg ≤ O(δ)
Mc′ ∗
Z (0) (2)
(1)
δ 2q1 +2l | τ m |dµg ≤ CD0∞ (χ̂)( E 2 )1/2 ( E 2 )1/2 + O(δ)
Mc′ ∗
Z
(2)
δ 2q2 +2l | τ m |dµg ≤ O(δ)
Mc′ ∗
Z (2) (3)
(3)
δ 2q3 +2l | τ m |dµq ≤ CD0∞ (χ̂)( E 2 )1/2 ( F 2 )1/2 + O(δ 1/2 )
Mc′ ∗
441
Chapter 14
The 1st Order Weyl

Current Error Estimates
14.1 Introduction
In the present chapter and the next we shall estimate the contributions of the
Weyl current error terms:
(n)
τc : n = 0, 1, 2, 3 (14.1)
(n)
(X) τ
c : (X) = (L), (O), (S); n = 0, 1, 2, 3
(n)
(Y X) τ
c : (Y X) = (LL), (OL), (OO), (OS), (SS); n = 0, 1, 2, 3
the Weyl current error term τc (W ; X, Y, Z) associated to a Weyl field W and
the multiplier fields X, Y, Z being given by 13.22:
τc (W ; X, Y, Z) = −(divQ(W ))(X, Y, Z) (14.2)
This is generated by the Weyl current J corresponding to W , for, according to
Proposition 12.6 we have:
(divQ(W ))βγδ = Wβ µδ ν Jµγν +Wβ µγ ν Jµδν + ∗ Wβ µδ ν Jµγν
∗
+ ∗ Wβ µγ ν Jµδν
∗
(14.3)
Now the fundamental Weyl field W = R satisfies the homogeneous Bianchi
equations 12.15. Thus, the corresponding Weyl current vanishes, hence:
(n)
τ c= 0 : n = 0, 1, 2, 3 (14.4)
Only to the derived Weyl fields there correspond non-trivial Weyl currents.
These are given by Proposition 12.1. Given a Weyl field W and a commutation
field X let us define the Weyl currents (X) J 1 (W ), (X) J 2 (W ), (X) J 3 (W ) by:
(X) 1 1 (X) µν
J (W )βγδ = π̃ ∇ν Wµβγδ (14.5)
2
442
1 (X)
(X) 2
J (W )βγδ = pµ W µβγδ
2
1
(X) 3
J (W )βγδ = ( (X) qµβν W µνγδ + (X)
qµγν W µβδν + (X)
qµδν W µβγ ν )
2
Here:
(X)
pβ = ∇α (X)
π̃αβ (14.6)
and:
(X) (X) (X) 1 (X) (X)
qαβγ = ∇β π̃γα − ∇γ π̃βα + ( pβ gαγ − pγ gαβ ) (14.7)
3
the 3-covariant tensorfield (X) q having the algebraic properties of a Weyl cur-
rent.
Let us denote by (X) J the Weyl current corresponding to the 1st order
derived Weyl field L̃X R. Then according to Proposition 12.1 (X) J is given by:
(X) (X) 1 (X) 2 (X) 3
J= J (R) + J (R) + J (R) (14.8)
The 1st order Weyl current error terms are:
(L) (0)
τ c= −(divQ(L̃L R))(L, L, L)
(L) (1)
τ c= −(divQ(L̃L R))(K, L, L)
(L) (2)
τ c= −(divQ(L̃L R))(K, K, L)
(L) (3)
τ c= −(divQ(L̃L R))(K, K, K) (14.9)
(O) (0)
X
τ c= − (divQ(L̃Oi R))(L, L, L)
i
(O) (1)
X
τ c= − (divQ(L̃Oi R))(K, L, L)
i
(O) (2)
X
τ c= − (divQ(L̃Oi R))(K, K, L)
i
(O) (3)
X
τ c= − (divQ(L̃Oi R))(K, K, K) (14.10)
i
and:
(S) (3)
τ c= −(divQ(L̃S R))(K, K, K) (14.11)
By Lemma 12.3 the above are given by:
(L) (0)
n o
τ c= −4Ω3 (Θ( (L) J), α(L̃L R)) − 2(Ξ( (L) J), β(L̃L R))
(L) (1) n
τ c = −8Ω3 |u|2 Λ( (L) J)ρ(L̃L R) − K( (L) J)σ(L̃L R)
443
o
+(I( (L) J), β(L̃L R))
(2)
n
(L)
τ c = −8Ω3 |u|4 Λ( (L) J)ρ(L̃L R) − K( (L) J)σ(L̃L R)
o
−(I( (L) J), β(L̃L R))
(3)
n o
(L)
τ c = −4Ω3 |u|6 (Θ( (L) J), α(L̃L R)) + 2(Ξ( (L) J), β(L̃L R))
(14.12)
(O) (0)
X n o
τ c= −4 Ω3 (Θ( (Oi )
J), α(L̃Oi R)) − 2(Ξ( (Oi )
J), β(L̃Oi R))
i
(O) (1)
X n
τ c= −8 Ω3 |u|2 Λ( (Oi )
J)ρ(L̃Oi R) − K( (Oi )
J)σ(L̃Oi R)
i
o
+(I( (Oi ) J, β(L̃Oi R))
(2) X n
(O)
τ c = −8 Ω3 |u|4 Λ( (Oi ) J)ρ(L̃Oi R) − K( (Oi )
J)σ(L̃Oi R)
i
o
−(I( (Oi ) J), β(L̃Oi R))
(3) X n o
(O)
τ c = −4 Ω3 |u|6 (Θ( (Oi ) J), α(L̃Oi R)) + 2(Ξ( (Oi )
J), β(L̃Oi R))
i
(14.13)
and:
(S) (3)
n o
τ c= −4Ω3 |u|6 (Θ( (S)
J), α(L̃S R)) + 2(Ξ( (S)
J), β(L̃S R)) (14.14)
Let us denote by (Y X) J the Weyl current corresponding to the 2nd order

derived Weyl field L̃Y L̃X R. Then according to Proposition 12.1 (Y X) J is given
by:
(Y X) (X) (Y ) 1 (Y ) 2 (Y ) 3
J = L̃Y J+ J (L̃X R) + J (L̃X R) + J (L̃X R) (14.15)
The 2nd order Weyl current error terms are:
(LL) (0)
τ c= −(divQ(L̃L L̃L R))(L, L, L)
(LL) (1)
τ c= −(divQ(L̃L L̃L R))(K, L, L)
(LL) (2)
τ c= −(divQ(L̃L L̃L R))(K, K, L)
(LL) (3)
τ c= −(divQ(L̃L L̃L R))(K, K, K) (14.16)
(OL) (0)
X
τ c= − (divQ(L̃Oi L̃L R))(L, L, L)
i
444
(OL) (1)
X
τ c= − (divQ(L̃Oi L̃L R))(K, L, L)
i
(OL) (2)
X
τ c= − (divQ(L̃Oi L̃L R))(K, K, L)
i
(OL) (3)
X
τ c= − (divQ(L̃Oi L̃L R))(K, K, K) (14.17)
i
(OO) (0)
X
τ c= − (divQ(L̃Oj L̃Oi R))(L, L, L)
i,j
(OO) (1)
X
τ c= − (divQ(L̃Oj L̃Oi R))(K, L, L)
i,j
(OO) (2)
X
τ c= − (divQ(L̃Oj L̃Oi R))(K, K, L)
i,j
(OO) (3)
X
τ c= − (divQ(L̃Oj L̃Oi R))(K, K, K) (14.18)
i,j
and:
(OS) (3)
X
τ c= − (divQ(L̃Oi L̃S R))(K, K, K) (14.19)
i
(SS) (3)
τ c= −(divQ(L̃S L̃S R))(K, K, K) (14.20)
By Lemma 12.3 the above are given by:
(LL) (0)
n o
τ c= −4Ω3 (Θ( (LL) J), α(L̃L L̃L R)) − 2(Ξ( (LL) J), β(L̃L L̃L R))
(1)
n
(LL)
τ c = −8Ω3 |u|2 Λ( (LL) J)ρ(L̃L L̃L R) − K( (LL) J)σ(L̃L L̃L R)
o
+(I( (LL) J), β(L̃L L̃L R))
(2)
n
(LL)
τ c = −8Ω3 |u|4 Λ( (LL) J)ρ(L̃L L̃L R) − K( (LL) J)σ(L̃L L̃L R)
o
−(I( (LL) J), β(L̃L L̃L R))
(3)
n o
(LL)
τ c = −4Ω3 |u|6 (Θ( (LL) J), α(L̃L L̃L R)) + 2(Ξ( (LL) J), β(L̃L L̃L R))
(14.21)
(OL) (0)
X n
τ c= −4 Ω3 (Θ( (Oi L)
J), α(L̃Oi L̃L R))
i
o
(Oi L)
−2(Ξ( J), β(L̃Oi L̃L R))
445
(OL) (1)
X n
τ c= −8 Ω3 |u|2 Λ( (Oi L)
J)ρ(L̃Oi L̃L R) − K( (Oi L)
J)σ(L̃Oi L̃L R)
i
o
+(I( (Oi L) J, β(L̃Oi L̃L R))
(OL) (2) X n
τ c = −8 Ω3 |u|4 Λ( (Oi L) J)ρ(L̃Oi L̃L R) − K( (Oi L)
J)σ(L̃Oi L̃L R)
i
o
−(I( (Oi L) J), β(L̃Oi L̃L R))
(3) X n
(OL)
τ c = −4 Ω3 |u|6 (Θ( (Oi L) J), α(L̃Oi L̃L R))
i
o
(Oi L)
+2(Ξ( J), β(L̃Oi L̃L R)) (14.22)
(OO) (0)
X n
τ c= −4 Ω3 (Θ( (Oj Oi )
J), α(L̃Oj L̃Oi R))
i,j
o
−2(Ξ( (Oj Oi ) J), β(L̃Oj L̃Oi R))
(1) X n
(OO)
τ c = −8 Ω3 |u|2 Λ( (Oj Oi ) J)ρ(L̃Oj L̃Oi R) − K( (Oj Oi ) J)σ(L̃Oj L̃Oi R)
i,j
o
+(I( (Oj Oi ) J, β(L̃Oj L̃Oi R))
(2) X n
(OO)
τ c = −8 Ω3 |u|4 Λ( (Oj Oi ) J)ρ(L̃Oj L̃Oi R) − K( (Oj Oi )
J)σ(L̃Oj L̃Oi R)
i,j
o
−(I( (Oj Oi ) J), β(L̃Oj L̃Oi R))
(3) X n
(OO)
τ c = −4 Ω3 |u|6 (Θ( (Oj Oi ) J), α(L̃Oj L̃Oi R))
i,j
o
(Oj Oi )
+2(Ξ( J), β(L̃Oj L̃Oi R)) (14.23)
and:
(OS) (3)
X n
τ c= −4 Ω3 |u|6 (Θ( (Oi S)
J), α(L̃Oi L̃S R))
i
o
(Oi S)
+2(Ξ( J), β(L̃Oi L̃S R)) (14.24)
(SS) (3)
n o
τ c= −4Ω3 |u|6 (Θ( (SS)
J), α(L̃S L̃S R)) + 2(Ξ(s(SS) J), β(L̃S L̃S R))
(14.25)
The aim of the next two chapters is to obtain bounds for the Weyl current
error integrals:
Z
(n)
1st order: δ 2qn +2l | (X) τ c |dµg (14.26)
Mc′ ∗
446
: for (X) = (L), (O), (S)
Z
(Y X) (n)
2nd order: δ 2qn +2l | τc |dµg
Mc′ ∗
: for (Y X) = (LL), (OL), (OO), (OS), (SS)
In the present chapter we shall estimate the 1st order Weyl current error inte-
grals. The 2nd order Weyl current error integrals shall be estimated in the next
chapter.

We begin with the following proposition which is deduced in a straightforward
manner using the table 1.175 of connection coefficients of the frame (eµ : µ =
1, 2, 3, 4) and noting the basic identities 1.187 and 13.85.
Lemma 14.1 The components of the Weyl current (X) J 1 (W ) associated to
the commutation field X and to the Weyl field W are given by:
(X) 1 1 (X) −1
4ΞA ( J (W )) = − j Ω (Dβ(W ))A + (div / α(W ))A − χAB βB (W )
2
−trχβA (W ) − Ω−1 ωβA (W ) + (2ζ B − η B )αAB (W )
1 n
+ (X) mB 2∇ /B βA (W ) + Ω−1 (D̂α(W ))AB − χBC αAC (W )
2
−trχαAB (W ) + 2Ω−1 ωαAB (W ) + 2((ζ − 2η)⊗β(Wˆ ))AB
∗
−3(χAB ρ(W ) + χAB σ(W ))}
1 (X) B n −1
+ m Ω (D̂α(W ))AB − trχαAB (W )
2
−2Ω−1 ωαAB (W )
− (X) îBC ∇ ˆ

/C αAB (W ) − (χ⊗β(W ))CAB + 2ζC αAB (W )
1 n
4ΞA ( (X) J 1 (W )) = − (X) j −Ω−1 (Dβ(W ))A + (div / α(W ))A + χAB β B (W )
2 o
+trχβ A (W ) + Ω−1 ωβ A (W ) − (2ζ B + η B )αAB (W )
1 n
+ (X) mB −2∇ /B β A (W ) + Ω−1 (D̂α(W ))AB − χBC αAC (W )
2
−trχαAB (W ) + 2Ω−1 ωαAB (W ) + 2((ζ + 2η)⊗β(W ˆ ))AB
o
−3(χAB ρ(W ) − ∗ χAB σ(W ))
1 n
+ (X) mB Ω−1 (D̂α(W ))AB − trχαAB (W )
2
−2Ω−1 ωαAB (W )
− (X) îBC ∇ ˆ

/C αAB (W ) + (χ⊗β(W ))CAB − 2ζC αAB (W )

1 3
4ΘAB ( (X) J 1 (W )) = (X) j Ω−1 (D̂α(W ))AB + (∇ ˆ
/⊗β(W ))AB − trχα(W )
2 2
447
+2Ω−1 ωαAB (W ) + ((ζ − 4η)⊗β(W
ˆ ))AB
∗
−3(χ̂AB ρ(W ) + χ̂AB σ(W ))}
1n (X)
o
− m⊗ ˆ Ω−1 Dβ(W ) − χ♯ · β(W ) + Ω−1 ωβ(W )
2 AB
3 (X) 3 ∗(X)
+ m⊗ηˆ ρ(W ) + m⊗η ˆ σ(W )
2 AB 2 AB
1 (X)
n o
− m⊗ ˆ Ω−1 Dβ(W ) − χ̂♯ · β(W ) − Ω−1 ωβ(W )
2 AB
1 n (X) ♯ o 1 (X)
− η⊗ˆ m · α(W ) + m, η αAB (W )
2 AB 2
1
− (X) mC ∇ ˆ

/C αAB (W ) − (χ⊗β(W ))CAB + 2ζC αAB (W )
2
+ (X) îAD ∇/D βB (W ) + (X) îBD ∇/D βA (W ) − g/AB (X) îCD ∇ /D βC (W )
1 (X) CD
+ î χ̂CA αDB (W ) + χ̂CB αDA (W ) − g/AB χ̂CE αDE (W )
2
1 n (X) ♯ o
−( (X) î, χ̂)αAB (W ) + ( ˆ
î · ζ)⊗β(W )
2 AB
3 n
(X) ∗(X)
o
− trχ îAB ρ(W ) + îAB σ(W )
2
1 3
4ΘAB ( (X) J 1 (W )) = (X) j Ω−1 (D̂α(W ))AB − (∇ ˆ
/⊗β(W ))AB − trχα(W )
2 2
+2Ω−1 ωαAB (W ) + ((ζ + 4η)⊗β(W ˆ ))AB
o
−3(χ̂AB ρ(W ) − ∗ χ̂AB σ(W ))
1 n (X) o
− m⊗ ˆ −Ω−1 Dβ(W ) + χ♯ · β(W ) − Ω−1 ωβ(W )
2 AB
3 (X) 3 ∗(X)
+ m⊗ηˆ ρ(W ) − m⊗η ˆ σ(W )
2 AB 2 AB
1 (X)
n o
− m⊗ ˆ −Ω−1 Dβ(W ) + χ̂♯ · β(W ) + Ω−1 ωβ(W )
2 AB
1 n (X) ♯ o 1 (X)
− η⊗ˆ m · α(W ) + m, η αAB (W )
2 AB 2
1
− (X) mC ∇ ˆ

/C αAB (W ) + (χ⊗β(W ))CAB − 2ζC αAB (W )
2
− (X) îAD ∇/D β B (W ) − (X) îBD ∇/D β A (W ) + g/AB (X) îCD ∇ /D β C (W )
1 (X) CD
χ̂CA αDB (W ) + χ̂CB αDA (W ) − g/AB χ̂CE αDE (W )

+ î
2
1 n (X) ♯ o
−( (X) î, χ̂)αAB (W ) + ( ˆ
î · ζ)⊗β(W )
2 AB
3 n
(X) ∗(X)
o
− trχ îAB ρ(W ) − îAB σ(W )
2
1 3
4Λ( (X) J 1 (W )) = (X) j Ω−1 Dρ(W ) + div / β(W ) − trχρ(W )
2 2
448

1
+(ζ − 2η, β(W )) − (χ̂, α)
2
1 (X)

− m, Ω−1 Dβ(W ) − χ♯ · β(W ) + Ω−1 ωβ(W )
2
3 (X)
+ m, ηρ(W ) + ∗ ησ(W )
2
− (X) m, /dρ(W ) + χ♯ · β(W ) − χ♯ · β(W )

− (X) m, Ω−1 Dβ(W ) − χ♯ · β(W ) − Ω−1 ωβ(W )

+ (X) m, η ♯ · α(W )
1 (X)
+ î, ∇ ˆ
/⊗β(W ˆ
) + ζ ⊗β(W ) − 3(χ̂ρ(W ) + ∗ χ̂σ(W ))
2
1
− trχ( (X) î, α(W ))
4
(X) 1 1 (X) 3
4Λ( J (W )) = j Ω−1 Dρ(W ) − div / β(W ) − trχρ(W )
2 2

1
+(ζ + 2η, β(W )) − (χ̂, α)
2
1 (X)

− m, −Ω−1 Dβ(W ) + χ♯ · β(W ) − Ω−1 ωβ(W )
2
3 (X)
+ m, ηρ(W ) − ∗ ησ(W )
2
(X)
− m, /dρ(W ) − χ♯ · β(W ) + χ♯ · β(W )

− (X) m, −Ω−1 Dβ(W ) + χ♯ · β(W ) + Ω−1 ωβ(W )

+ (X) m, η ♯ · α(W )
1 (X)
+ î, −∇ ˆ
/⊗β(W ˆ
) + ζ ⊗β(W ) − 3(χ̂ρ(W ) − ∗ χ̂σ(W ))
2
1
− trχ( (X) î, α(W ))
4
1 3
4K( (X) J 1 (W )) = (X) j Ω−1 Dσ(W ) − curl / β(W ) − trχσ(W )
2 2

1
−(ζ − 2η) ∧ β(W ) + χ̂ ∧ α(W )
2
1 (X)
m ∧ Ω Dβ(W ) − χ♯ · β(W ) + Ω−1 ωβ(W )
−1
+
2
3
− (X) m ∧ {ηρ(W ) + ∗ ησ(W )}
2
− (X) m, /dσ(W ) + χ♯ · ∗ β(W ) + χ♯ · ∗ β(W )
+ (X) m ∧ Ω−1 Dβ(W ) − χ♯ · β(W ) − Ω−1 ωβ(W )

449

(X)
− m, η ♯ · ∗
α(W )
1 (X)

− î, ∇ˆ ∗ β(W ) + ζ ⊗
/⊗ ˆ ∗ β(W ) − 3( ∗ χ̂ρ(W ) − χ̂σ(W ))
2
1
+ trχ (X) î ∧ α(W )
4
1 3
4K( (X) J 1 (W )) = (X) j −Ω−1 Dσ(W ) + curl / β(W ) + trχσ(W )
2 2

1
−(ζ + 2η) ∧ β(W ) + χ̂ ∧ α(W )
2
1
+ (X) m ∧ −Ω−1 Dβ(W ) + χ♯ · β(W ) − Ω−1 ωβ(W )

2
3
− (X) m ∧ ηρ(W ) − ∗ ησ(W )

2
+ (X) m, /dσ(W ) + χ♯ · ∗ β(W ) + χ♯ · ∗ β(W )
+ (X) m ∧ −Ω−1 Dβ(W ) + χ♯ · β(W ) + Ω−1 ωβ(W )

− (X) m, η ♯ · ∗ α(W )
1 (X)
+ î, ∇
/⊗ˆ ∗ β(W ) − ζ ⊗
ˆ ∗ β(W ) + 3( ∗ χ̂ρ(W ) + χ̂σ(W ))
2
1
+ trχ (X) î ∧ α(W )
4
1
4IA ( (X) J 1 (W )) = (X) j Ω−1 (Dβ(W ))A + /dA ρ(W ) + ∗/dA σ((W )

2
−trχβA (W ) + Ω−1 ωβA (W ) − 3(ηA ρ(W ) + ∗ ηA σ(W ))
o
−χAB βB (W ) + 2χ̂AB β B (W )
1 (X)
mB g/AB Ω−1 Dρ(W ) + /ǫ AB Ω−1 Dσ(W )

−
2 o
+2(ηA β B (W ) − ηB β A (W ) + g/AB η C β C (W ))
1 (X)

−
2 o
−2(ηB βA (W ) − η A βB (W ) + g/AB η C βC (W )
1 n
− X) mB 2∇ /B βA (W ) + 2ζB βA (W ) − χBC αAC (W )
2
−3(χAB ρ(W ) + ∗ χAB σ(W ))}
(X) B
+ îA /dB ρ(W ) + ∗(X) îAB /dB σ(W )
n
+ (X) îBC χ̂CA βB (W ) − χ̂CB βA (W ) − g/AB χ̂CD βD (W )
o
+χ̂CA β B (W ) − χ̂CB β A (W ) + g/AB χ̂CD β D (W )
(X)
+2trχ îAB β B (W )
450
(X) 1 1 (X) −1
4I A ( J (W )) = j Ω (−Dβ(W ))A + /dA ρ(W ) − ∗/dA σ((W )
2
+trχβ A (W ) − Ω−1 ωβ A (W ) − 3(η A ρ(W ) − ∗ η A σ(W ))
o
+χAB β B (W ) − 2χ̂AB βB (W )
1 (X)
mB g/AB Ω−1 Dρ(W ) − /ǫ AB Ω−1 Dσ(W )

−
2 o
−2(ηA βB (W ) − η B βA (W ) + g/AB η C βC (W ))
1 (X)

−
2 o
+2(ηB β A (W ) − ηA β B (W ) + g/AB η C β C (W )
1 n
− (X) mB −2∇/B β A (W ) + 2ζB β A (W ) − χBC αAC (W )
2 o
−3(χAB ρ(W ) − ∗ χAB σ(W ))
(X)
+ îAB /dB ρ(W ) − ∗(X) îAB /dB σ(W )
n
+ (X) îBC −χ̂CA β B (W ) + χ̂CB β A (W ) + g/AB χ̂CD β D (W )
o
−χ̂CA βB (W ) + χ̂CB βA (W ) − g/AB χ̂CD βD (W )
(X)
−2trχ îAB βB (W )
(X) 1
In the expressions for the components of J (W ) given by the above
proposition we now substitute for:
D̂α(W ), Dβ(W ), Dρ(W ), Dσ(W ), Dβ(W ) and D̂α(W )
from the inhomogeneous Bianchi equations of Proposition 12.4. When consider-

ing the 1st order Weyl current error estimates, W is the fundamental Weyl field
R and these equations reduce to the corresponding homogeneous equations (the
Bianchi identities of Proposition 1.2). When considering, on the other hand,
the 2nd order Weyl current error estimates, W is one of the derived Weyl fields
L̃L R, L̃Oi R : i = 1, 2, 3, and L̃S R, and on the right hand sides of the inhome-
geneous Bianchi equations we have the components of the corresponding Weyl
currents (L) J, (Oi ) J : i = 1, 2, 3, and (S) J, respectively. Therefore the consid-
eration of the 2nd order Weyl current error estimates arising from J 1 , requires
that we first assess the 1st order Weyl currents. We thus confine ourselves in
this chapter to the 1st order Weyl current error estimates. The 2nd order Weyl
current error estimates shall be addressed in the next chapter.
The expression for each component of (X) J 1 (W ) given by Lemma 14.1 is a
sum of terms one factor of which is a component of (X) π̃. We consider the terms
with the same such factor as a single term. We thus consider the expression for
each component of (X) J 1 (W ) as consisting of four terms, proportional to (X) j,
451
(X)
m, (X) m, and (X) î, in the order in which these appear in the expression
given by Lemma 14.1.
Now, each component of (X) π̃ is O∞ (δ r1 |u|p1 ) for some r1 , p1 according to
the estimates of Chapter 8. In the case W = R, to which our attention is at
present confined, after the substitution for
D̂α, Dβ, Dρ, Dσ, Dβ and D̂α
from the homogeneous Bianchi equations, the other factor of each term in the
expression of a given component of (X) J 1 (R) becomes the sum of a principal
part, which is a sum of 1st derivatives of components of R, and a non-principal
part, which is a sum of terms consisting of two factors, one of which is a con-
nection coefficient and the other a component of R. Viewing these terms in a
way which would be valid also in the case of the 2nd order Weyl current error
estimates, we see each term as being either O(δ r |u|p ) in the case of the terms
involving the components α, β, ρ, σ, β and their first derivatives (which are ei-
ther ∇/ or D derivatives after the substitution), or O(δ r |u|p ) in the case of the
terms involving the component α and its first derivatives (which are either ∇ /
or D derivatives after the substitution). The values of r and p assigned are
those implied by the bound on the quantity Q1 , taking also into account the
L∞ estimates of Chapter 3 for the connection coefficients in the case of the
non-principal terms. We then define r2 to be the minimal r and p2 to be the
maximal p occuring in the terms of the other factor of a term involving a given
component of (X) π̃ in the expression of a given component of (X) J 1 (R). We
then set, for each term involving a given component of (X) π̃ in the expression
of a given component of (X) J 1 (R),
r′ = r1 + r2 , p′ = p1 + p2 (14.27)
We then assign, to each component of (X) J 1 (R), the pair r∗ , p∗ , where r∗ is

the minimal r′ and p∗ is the maximal p′ occuring in the four terms constituting
the expression of that component. We obtain in this way the following tables.
1. Case X = L
(L) 1
Ξ( J (R)) : r∗ = −2, p∗ = −3
term r′ p′
1st −3/2 −3
(14.28)
3rd −2 −3
4th −2 −3
(the 2nd term vanishes)
(L) 1
Ξ( J (R)) : r∗ = 1/2, p∗ = −6
452
term r′ p′
1st 1 −6
(14.29)
2nd 1/2 −6
4th 1/2 −6
(the 3rd term vanishes)
(L) 1
Θ( J (R)) : r∗ = −3/2, p∗ = −3
term r′ p′
1st −3/2 −3
(14.30)
3rd −1 −4
4th −3/2 −3
(L) 1
Θ( J (R)) : r∗ = −1/2, p∗ = −5
term r′ p′
1st 1/2 −6
(14.31)
2nd 1/2 −6
4th −1/2 −5
This gives rise, as we shall see, to a borderline error integral. The borderline
contribution comes from the last part of the 4th term namely from the terms:
3 n
(L) ∗(L)
o
− trχ îρ − îσ (14.32)
2
The contribution of these shall be estimated more precisely in the sequel. If
these terms are removed from the 4th term the values r′ = 1/2, p′ = −6 would
be assigned to the remainder, hence if we denote by Θ′ ( (L) J 1 (R)) what results
if we remove from Θ( (L) J 1 (R)) the terms 14.32 then to Θ′ ( (L) J 1 (R)) is to be
assigned r∗ = 1/2, p∗ = −6.
(L) 1
Λ( J (R)) : r∗ = −2, p∗ = −3
term r′ p′
1st −1 −4
(14.33)
3rd −1 −4
4th −2 −3
contribution comes from the last part of the 4th term namely from the term:
1 (L)
− trχ( î, α) (14.34)
4
453
The contribution of this shall be estimated more precisely in the sequel. If this
term is removed from the 4th term the values r′ = −1, p′ = −4 would be
assigned to the remainder, hence if we denote by Λ′ ( (L) J 1 (R)) what results
if we remove from Λ( (L) J 1 (R)) the term 14.34 then to Λ′ ( (L) J 1 (R)) is to be
assigned r∗ = −1, p∗ = −4.
(L) 1
Λ( J (R)) : r∗ = 0, p∗ = −5
term r′ p′
1st 0 −5
(14.35)
2nd 0 −5
4th 0 −6
(L) 1
K( J (R)) : r∗ = −2, p∗ = −3
term r′ p′
1st −1 −4
(14.36)
3rd −1 −4
4th −2 −3
1
trχ (L) î ∧ α (14.37)
4
The contribution of this shall be estimated more precisely in the sequel. If this
term is removed from the 4th term the values r′ = −1, p′ = −4 would be
assigned to the remainder, hence if we denote by K ′ ( (L) J 1 (R)) what results if
we remove from K( (L) J 1 (R)) the term 14.37 then to K ′ ( (L) J 1 (R)) is to be
assigned r∗ = −1, p∗ = −4.
(L) 1
K( J (R)) : r∗ = 0, p∗ = −5
term r′ p′
1st 0 −5
(14.38)
2nd 0 −5
4th 0 −6
(L) 1
I( J (R)) : r∗ = −1, p∗ = −4
term r′ p′
1st −1/2 −4
(14.39)
3rd −1 −4
4th −1/2 −5
454
(L) 1
I( J (R)) : r∗ = −1, p∗ = −4
term r′ p′
1st 0 −5
(14.40)
2nd −1/2 −5
4th −1 −4
2. Case X = Oi : i = 1, 2, 3
(Oi ) 1
Ξ( J (R)) : r∗ = −3/2, p∗ = −3
term r′ p′
1st −1/2 −4
(14.41)
2nd −3/2 −3
4th −1 −3
(Oi ) 1
Ξ( J (R)) : r∗ = 3/2, p∗ = −6
term r′ p′
1st 2 −7
(14.42)
3rd 3/2 −13/2
4th 3/2 −6
(Oi ) 1
Θ( J (R)) : r∗ = −1/2, p∗ = −4
term r′ p′
1st −1/2 −4
(14.43)
2nd −1/2 −4
4th −1/2 −4
(Oi ) 1
Θ( J (R)) : r∗ = 1/2, p∗ = −5
term r′ p′
1st 3/2 −7
(14.44)
3rd 1 −6
4th 1/2 −5
455
3 n
(Oi ) ∗(Oi )
o
− trχ îρ − îσ (14.45)
2
be assigned to the remainder, hence if we denote by Θ′ ( (Oi ) J 1 (R)) what results
if we remove from Θ( (Oi ) J 1 (R)) the terms 14.45 then to Θ′ ( (Oi ) J 1 (R)) is to
be assigned r∗ = 1, p∗ = −6.
(Oi ) 1
Λ( J (R)) : r∗ = −1, p∗ = −3
term r′ p′
1st 0 −5
(14.46)
2nd −1/2 −4
4th −1 −3
1 (Oi )
− trχ( î, α) (14.47)
4
The contribution of this shall be estimated more precisely in the sequel. If
this term is removed from the 4th term the values r′ = 0, p′ = −4 would be
assigned to the remainder, hence if we denote by Λ′ ( (Oi ) J 1 (R)) what results
if we remove from Λ( (Oi ) J 1 (R)) the term 14.47 then to Λ′ ( Oi ) J 1 (R)) is to be
assigned r∗ = −1/2, p∗ = −4.
(Oi ) 1
Λ( J (R)) : r∗ = 1, p∗ = −6
term r′ p′
1st 1 −6
(14.48)
3rd 1 −6
4th 3/2 −6
(Oi ) 1
K( J (R)) : r∗ = −1, p∗ = −3
term r′ p′
1st 0 −5
(14.49)
2nd −1/2 −4
4th −1 −3
456
1 (Oi )
trχ î ∧ α (14.50)
4
The contribution of this shall be estimated more precisely in the sequel. If
this term is removed from the 4th term the values r′ = 0, p′ = −4 would be
assigned to the remainder, hence if we denote by K ′ ( (Oi ) J 1 (R)) what results if
we remove from K( (Oi ) J 1 (R)) the term 14.50 then to K ′ ( (Oi ) J 1 (R)) is to be
assigned r∗ = −1/2, p∗ = −4.
(Oi ) 1
K( J (R)) : r∗ = 1, p∗ = −6
term r′ p′
1st 1 −6
(14.51)
3rd 1 −6
4th 1 −6
(Oi ) 1
I( J (R)) : r∗ = 0, p∗ = −5
term r′ p′
1st 1/2 −5
(14.52)
2nd 0 −5
4th 1/2 −5
(Oi ) 1
I( J (R)) : r∗ = 0, p∗ = −4
term r′ p′
1st 1 −6
(14.53)
3rd 0 −5
4th 0 −4
3. Case X = S
In this case only the case n = 3 occurs, therefore we only have to consider
(S) 1
Θ( J (R)) and Ξ( (S) J 1 (R)) (see 14.14).
(S) 1
Ξ( J (R)) : r∗ = 3/2, p∗ = −6
457
term r′ p′
1st 2 −7
2nd 3/2 −6 (14.54)
3rd 2 −13/2
4th 3/2 −6
(S) 1
Θ( J (R)) : r∗ = 1/2, p∗ = −5
term r′ p′
1st 3/2 −7
2nd 3/2 −6 (14.55)
3rd 3/2 −6
4th 1/2 −5
3 n o
− trχ (S) îρ − ∗(S) îσ (14.56)
2
be assigned to the remainder, hence if we denote by Θ′ ( (S) J 1 (R)) what results
if we remove from Θ( (S) J 1 (R)) the terms 14.56 then to Θ′ ( (S) J 1 (R)) is to be
assigned r∗ = 3/2, p∗ = −6.
We have completed the investigation of the components of the Weyl currents
(L) 1
J (R), (Oi ) J 1 (R) : i = 1, 2, 3, and (S) J 1 (R). These currents contribute to
(n) (n) (3)
the error terms (L) τ c : n = 0, 1, 2, 3, (O) τ c : n = 0, 1, 2, 3, and (S) τ c ,
respectively, according to 14.12, 14.13, and 14.14. We call these contributions
(n) (n) (3)
(L)
n = 0, 1, 2, 3, (O) τ c,1 : n = 0, 1, 2, 3, and (S) τ c,1 , respectively.
τ c,1 :
Each component of the Weyl currents (L) J 1 (R), (Oi ) J 1 (R) : i = 1, 2, 3, and
(S) 1
J (R) being written as a sum of terms in the manner discussed above, and
these expressions being substituted into 14.12, 14.13, and 14.14 respectively,
sums of trilinear terms result, two of the factors in each term being contributed
by the expression for a component of the Weyl current (L) J 1 (R) in the case of
14.12, (Oi ) J 1 (R) : i = 1, 2, 3 in the case of 14.13, and (S) J 1 (R) in the case of
14.14, and the other factor being a component of L̃L R, L̃Oi R : i = 1, 2, 3, and
L̃S R, respectively, multiplied by Ω3 and the appropriate power of |u|2 . Viewing
these third factors in a way which would be valid also in the case of the 2nd order
Weyl current error estimates, we see each third factor as being either O(δ r3 |u|p3 )
in the case of factors involving α(L̃X R), β(L̃X R), ρ(L̃X R), σ(L̃X R), β(L̃X R), or
O(δ r3 |u|p3 ) in the case of factors involving α(L̃X R). The values of r3 and p3
assigned are those implied by the bound on the quantity P1 . To each trilinear
term we can then apply accordingly one of the first three cases of Lemma 13.1.
Since the pair r′ , p′ corresponding to the first two factors of each trilinear term
458
may be replaced by the corresponding r∗ , p∗ , we then obtain a bound for the
contribution of all terms resulting from the product of a given component of
(X) 1
J with a given component of L̃X R to the corresponding error integral
Z
2qn +2l (n)
δ | (X) τ c,1 |dµg (14.57)
Mc′ ∗
by O(δ e ), where e is the excess index (see 13.32):
e = 2qn + 2l + r∗ + r3 + 1 (14.58)
provided that the integrability index s, defined by (see 13.33):
s = p∗ + p3 + 3 (14.59)
each of 14.12 - 14.14.
1. Case X = L : l = 1
(0)
(L) τ c,1
term e s
1st 1 −1 (14.60)
2nd 3/2 −2
(1)
(L) τ c,1
term e s
1st 0 −1
(14.61)
2nd 0 −1
3rd 1/2 −1
(2)
(L) τ c,1
term e s
1st 1 −1
(14.62)
2nd 1 −1
3rd 1 −1
(3)
(L) τ c,1
459
term e s
1st 0 −1/2 (14.63)
2nd 1/2 −1
2. Case X = O : l = 0
(0)
(O) τ c,1
term e s
1st 1 −2 (14.64)
2nd 1 −2
(1)
(O) τ c,1
term e s
1st 0 −1
(14.65)
2nd 0 −1
3rd 1/2 −2
(2)
(O)
τ c,1
term e s
1st 1 −2
(14.66)
2nd 1 −2
3rd 1 −1
(3)
(O) τ c,1
term e s
1st 0 −1/2 (14.67)
2nd 1/2 −1
3. Case X = S : l = 0
(3)
(S) τ c,1
term e s
1st 0 −1/2 (14.68)
2nd 1/2 −1
460
We see that all terms have negative integrability index so Lemma 13.1 indeed
applies. All terms have non-negative excess index. The terms with vanishing
excess index play a crucial role because they give rise to borderline error inte-
grals. These must be analyzed in more detail. They occur only in the cases
n = 1 and n = 3. The borderline terms are the 1st term in each of 14.63, 14.67,
and 14.68, and the 1st and 2nd terms in each of 14.61, 14.65.
Replacing Θ( (L) J 1 (R), Θ( (Oi ) J 1 (R)), and Θ( (S) J 1 (R) by Θ′ ( (L) J 1 (R)),
′ (Oi ) 1
Θ( J (R)) and Θ′ ( (S) J 1 (R)) in the 1st term of each of 14.63, 14.67, and
13.68, respectively, terms with excess indices of 1, 1/2, and 1, respectively,
(3) (3)
(L) τ c,1 , (O) τ c,1 ,
would result. Thus the actual borderline error terms in
(3)
(S) τ c,1 ,
and are the terms:
− 6Ω3 |u|6 trχ(( (L)

îρ − ∗(L)
îσ), α(L̃L R)) (14.69)
X
− 6Ω3 |u|6 trχ (( (Oi )
îρ − ∗(Oi )
îσ), α(L̃Oi R)) (14.70)
i
and:
− 6Ω3 |u|6 trχ(( (S)
îρ − ∗(S)
îσ), α(L̃S R)) (14.71)
contributed by 14.32, 14.45, and 14.56, respectively.
Using the precise bound:
(L)
k îkL∞ (Su,u ) ≤ Cδ −1/2 |u|−1 R∞ ′
0 (α) : ∀(u, u) ∈ Dc∗ (14.72)
(see first of 8.32), using also the facts that:

(2)
ktrχρkL2 (Cu ) ≤ Cδ 1/2 |u|−3 ( E 0 )1/2 : ∀u ∈ [u0 , c∗ )
(2)
ktrχσkL2 (Cu ) ≤ Cδ 1/2 |u|−3 ( E 0 )1/2 : ∀u ∈ [u0 , c∗ ) (14.73)
as well as the fact that:

(3)
kΩ3 |u|3 α(L̃L R)kL2 (C u ) ≤ Cδ 1/2 ( F 1 )1/2 : ∀u ∈ [0, δ) (14.74)
(see 12.89, 12.128 and 12.137), and following the proof of Case 2 of Lemma 13.1,
we deduce that the contribution of the term 14.69 to the error integral 14.57
with X = L (l = 1) and n = 3 is bounded by:
(2) (3)
CR∞
0 (α)( E 0 )
1/2
( F 1 )1/2 (14.75)
Using the more precise bound:

(Oi )
k îkL∞ (Su,u ) ≤ Cδ 1/2 |u|−1 (R
/41 (β) + R∞
0 (β) + O(δ
1/2
)) : ∀(u, u) ∈ Dc′ ∗
(14.76)
461
implied by the first of 8.144, using also 14.73 as well as the fact that:
(3)
kΩ3 |u|3 α((L̃Oi R)kL2 (C u ) ≤ Cδ 3/2 ( F 1 )1/2 : ∀u ∈ [0, δ) (14.77)
with X = O (l = 0) and n = 3 is bounded by:
(2) (3)
/41 (β) + R∞
C(R 0 (β))( E 0 )
1/2
( F 1 )1/2 + O(δ 1/2 ) (14.78)

(S)
k îkL∞ (Su,u ) ≤ Cδ 1/2 |u|−1 (D0∞ (χ̂)+R∞ ′
0 (α)+O(δ)) : ∀(u, u) ∈ Dc∗ (14.79)
implied by the first of 8.34, using also 14.73 as well as the fact that:
(3)
kΩ3 |u|3 α(L̃S R)kL2 (C u ) ≤ Cδ 3/2 ( F 1 )1/2 : ∀u ∈ [0, δ) (14.80)
with X = S (l = 0) and n = 3 is bounded by:
(2) (3)
C(D0∞ (χ̂) + R∞
0 (α))( E 0 )
1/2
( F 1 )1/2 + O(δ) (14.81)
Now, by the first of 10.73 and the first of 12.263:

(0)
R∞ ′
0 (α) ≤ CR[2] (α) ≤ C ( E 2 )
1/2
+ O(δ 1/2 ) (14.82)
It follows that 14.75 is in turn bounded by:

(0) (2) (3)
C( E 2 )1/2 ( E 0 )1/2 ( F 1 )1/2 + O(δ 1/2 ) (14.83)
By the second of 10.66, the second of 10.73 and the second of 12.263:
(1)
/41 (β) ≤ CR[2] (β) ≤ C ′ ( E 2 )1/2 + O(δ 1/2 )
R
(1)
R∞ ′
0 (β) ≤ CR[2] (β) ≤ C ( E 2 )
1/2
+ O(δ 1/2 ) (14.84)
It follows that 14.78 is in turn bounded by:

(1) (2) (3)
C( E 2 )1/2 ( E 0 )1/2 ( F 1 )1/2 + O(δ 1/2 ) (14.85)
Moreover, 14.81 is in turn bounded by, in view of 14.82,

(0) (2) (3)
∞ 1/2
C D0 (χ̂) + ( E 2 ) ( E 0 )1/2 ( F 1 )1/2 + O(δ 1/2 ) (14.86)
462
Replacing Λ( (L) J 1 (R), K( (L) J 1 (R)), and Λ( (Oi ) J 1 (R)), K( (Oi ) J 1 (R))
by Λ′ ( (L) J 1 (R)), K ′ ( (L) J 1 (R)) and Λ′ ( (Oi ) J 1 (R)), K ′ ( (Oi ) J 1 (R)) in the 1st
and 2nd terms of each of 14.61 and 14.65, respectively, terms with excess indices
of 1 and 1/2, respectively, would result. Thus the actual borderline error terms
(1) (1)
(L) τ c,1 (O) τ c,1 ,
in and are the terms:
n o
2Ω3 |u|2 trχ ( (L)
î, α)ρ(L̃L R) + (L)
î ∧ ασ(L̃L R) (14.87)
and:
Xn o
2Ω3 |u|2 trχ ( (Oi )
î, α)ρ(L̃Oi R) + (Oi )
î ∧ ασ(L̃Oi R) (14.88)
i
contributed by 14.34, 14.37 and 14.47, 14.50 respectively.

Using 14.72, the fact that:
(0)
ktrχαkL2 (Cu ) ≤ Cδ −1 |u|−1 ( E 0 )1/2 : ∀u ∈ [u0 , c∗ ) (14.89)
as well as the facts that:

(2)
kΩ3 |u|2 ρ(L̃L R)kL2 (Cu ) ≤ Cδ −1/2 ( E 1 )1/2 : ∀u ∈ [u0 , c∗ )
(2)
kΩ3 |u|2 σ(L̃L R)kL2 (Cu ) ≤ Cδ −1/2 ( E 1 )1/2 : ∀u ∈ [u0 , c∗ ) (14.90)
(see 12.88, 12.125 and 12.136), and following the proof of Case 1 of Lemma 13.1
with X = L (l = 1) and n = 1 is bounded by:
(0) (2)
CR∞
0 (α)( E 0 )
1/2
( E 1 )1/2 (14.91)
Using 14.76, 14.89, as well as the facts that:

(2)
kΩ3 |u|2 ρ(L̃Oi R)kL2 (Cu ) ≤ Cδ 1/2 ( E 1 )1/2 : ∀u ∈ [u0 , c∗ )
(2)
kΩ3 |u|2 σ(L̃Oi R)kL2 (Cu ) ≤ Cδ 1/2 ( E 1 )1/2 : ∀u ∈ [u0 , c∗ ) (14.92)
(see 12.93, 12.125 and 12.136), and following the proof of Case 1 of Lemma 13.1
with X = O (l = 0) and n = 1 is bounded by:
(0) (2)
/41 (β) + R∞
C(R 0 (β))( E 0 )
1/2
( E 1 )1/2 (14.93)
In view of 14.82 we conclude that 14.91 is in turn bounded by:

(0) (0) (2)
C( E 2 )1/2 ( E 0 )1/2 ( E 1 )1/2 + O(δ 1/2 ) (14.94)
463
Moreover, in view of 14.84 we conclude that 14.93 is in turn bounded by:
(1) (0) (2)
C( E 2 )1/2 ( E 0 )1/2 ( E 1 )1/2 + O(δ 1/2 ) (14.95)
We summarize the results of this section in the following proposition.
Proposition 14.1 The 1st order Weyl current error integrals arising from J 1 ,
14.57, satisfy the estimates:
Z
2q0 +2 (0)
δ | (L) τ c,1 |dµg ≤ O(δ)
Mc′ ∗
Z (0) (0) (2)
(L) (1)
δ 2q1 +2 | τ c,1 |dµg ≤ C( E 2 )1/2 ( E 0 )1/2 ( E 1 )1/2 + O(δ 1/2 )
Mc′ ∗
Z
(L) (2)
δ 2q2 +2 | τ c,1 |dµg ≤ O(δ)
Mc′ ∗
Z (0) (2) (3)
2q3 +2 (L) (3)
δ | τ c,1 |dµg ≤ C( E 2 )1/2 ( E 0 )1/2 ( F 1 )1/2 + O(δ 1/2 )
Mc′ ∗
Z
(O) (0)
δ 2q0 | τ c,1 |dµg ≤ O(δ)
Mc′ ∗
Z (1) (0) (2)
(O) (1)
δ 2q1 | τ c,1 |dµg ≤ C( E 2 )1/2 ( E 0 )1/2 ( E 1 )1/2 + O(δ 1/2 )
Mc′ ∗
Z
(O) (2)
δ 2q2 | τ c,1 |dµg ≤ O(δ)
Mc′ ∗
Z (1) (2) (3)
(O) (3)
δ 2q3 | τ c,1 |dµg ≤ C( E 2 )1/2 ( E 0 )1/2 ( F 1 )1/2 + O(δ 1/2 )
Mc′ ∗
and:
Z
(0) (2) (3)
(S) (3)
δ 2q3 | τ c,1 |dµg ≤ C D0∞ (χ̂) + ( E 2 )1/2 ( E 0 )1/2 ( F 1 )1/2 + O(δ 1/2 )
Mc′ ∗

For any commutation field X, the associated 1-form (X) p defined by 14.6 de-
composes into the S 1-form (X) /p, the restriction of (X) p to the Su,u , given on
each Su,u by:
(X) (X)
/p(Y ) = p(Y ) : ∀Y ∈ Tq Su,u ∀q ∈ Su,u (14.96)
464
(X) (X)
and the functions p3 , p4 , given by:
(X) (X) (X) (X)
p3 = p(L̂), p4 = p(L̂) (14.97)
The components of the Weyl current (X) J 2 (W ) associated to the commu-

tation field X and to the Weyl field W are then given by:
(X) 2 (X) (X) ♯
4Ξ( J (W )) = − p4 β(W ) − /p · α(W )
(X) 2 (X) (X) ♯
4Ξ( J (W )) = p3 β(W ) − /p · α(W )
(X) 2 (X) (X) ˆ
4Θ( J (W )) = − p3 α(W ) + /p⊗β(W )
(X) 2 (X) (X) ˆ
4Θ( J (W )) = − p4 α(W ) − /p⊗β(W )
(X) 2 (X) (X)
4Λ( J (W )) = − p4 ρ(W ) + ( /p, β(W ))
(X) 2 (X) (X)
4Λ( J (W )) = − p3 ρ(W ) − ( /p, β(W ))
(X) 2 (X) (X)
4K( J (W )) = − p4 σ(W ) − /p ∧ β(W )
(X) 2
4K( J (W )) = (X) p3 σ(W ) + (X) /p ∧ β(W )
(X) 2
4I( J (W )) = − (X) p3 β(W ) + (X) /pρ(W ) + ∗(X) /pσ(W )
(X) 2
4I( J (W )) = (X) p4 β(W ) + (X) /pρ(W ) − ∗(X) /pσ(W ) (14.98)
Using the table 1.175 of connection coefficients of the frame (eµ : µ =

1, 2, 3, 4) we obtain:
(X) 1
p4 = − Ω−1 D (X)
j + div
/ (X)
m
2
(X) 1 (X) (X)
−trχ j + ((η + 5η), m) − (χ̂, î) (14.99)
2
(X) 1
p3 = − Ω−1 D (X)
j + div
/ (X)
m
2
(X) 1 (X) (X)
−trχ j + ((η + 5η), m) − (χ̂, î) (14.100)
2
(X) 1 1
/p = − (D (X) m + D (X) m) + /d (X) j + div/ (X)
î
2 2
1 1 1 1
− Ω−1 ω (X) m − Ω−1 ω (X) m − trχ (X)
m − trχ (X)
m
2 2 2 2
+(η + η) (X) j + (η + η)♯ · (X) î (14.101)
Using the estimates 9.1 - 9.3 as well as 8.32 and the estimates of Chapter 3
we deduce, for X = L:
(L)
p4 = O4 (δ −1 |u|−2 )
(L)
p3 = O4 (|u|−2 )
(L)
/p = O4 (δ −1/2 |u|−2 ) (14.102)
465
Using the estimates 9.89 - 9.91 as well as 8.144 and the estimates of Chapter 3
we deduce, for X = Oi : i = 1, 2, 3:
(Oi )
p4 = O4 (|u|−2 )
(Oi )
p3 = O4 (δ|u|−3 )
(Oi )
/p = O4 (δ 1/2 |u|−2 ) (14.103)
Also, using the estimates 9.12 - 9.14 as well as 8.34 we deduce, for X = S:
(S)
p4 = O4 (|u|−2 )
(S)
p3 = O4 (δ|u|−3 )
(S)
/p = O4 (δ 1/2 |u|−2 ) (14.104)
The expression for each component of (X) J 2 (W ) given by 14.98 is a sum of

terms one factor of which is a component of (X) p and the other factor a compo-
nent of W . According to the above, each component of (X) p is O4 (δ r1 |u|p1 ) for
some r1 , p1 . Our attention is at present confined to the case W = R. Viewing
the components of W = R in a way which would be valid also in the case of
the 2nd order Weyl current error estimates we see each component as being
O4 (δ r2 |u|p2 ) according to:
α = O4 (δ −3/2 |u|−1 )
β = O4 (δ −1/2 |u|−2 )
ρ, σ = O4 (|u|−3 )
β = O4 (δ|u|−4 )
α = O4 (δ 3/2 |u|−9/2 ) (14.105)
implied by 12.146 and the bound on the quantity Q1 . We set, for each term in
the expression of a given component of (X) J 2 (R),
r′ = r1 + r2 , p′ = p1 + p2 (14.106)

the minimal r′ and p∗ is the maximal p′ occuring in the terms constituting the
expression of that component. We obtain in this way the following tables. The
ordinals in these tables refer to the terms in the expressions 14.98.
1. Case X = L
(L) 2
Ξ( J (R)) : r∗ = −2, p∗ = −3
term r′ p′
1st −3/2 −4 (14.107)
2nd −2 −3
466
(L) 2
Ξ( J (R)) : r∗ = 1, p∗ = −6
term r′ p′
1st 1 −6 (14.108)
2nd 1 −13/2
(L) 2
Θ( J (R)) : r∗ = −3/2, p∗ = −3
term r′ p′
1st −3/2 −3 (14.109)
2nd −1 −4
(L) 2
Θ( J (R)) : r∗ = 1/2, p∗ = −6
term r′ p′
1st 1/2 −13/2 (14.110)
2nd 1/2 −6
(L) 2
Λ( J (R)) : r∗ = −1, p∗ = −4
term r′ p′
1st −1 −5 (14.111)
2nd −1 −4
(L) 2
Λ( J (R)) : r∗ = 0, p∗ = −5
term r′ p′
1st 0 −5 (14.112)
2nd 1/2 −6
(L) 2
K( J (R)) : r∗ = −1, p∗ = −4
term r′ p′
1st −1 −5 (14.113)
2nd −1 −4
(L) 2
K( J (R)) : r∗ = 0, p∗ = −5
term r′ p′
1st 0 −5 (14.114)
2nd 1/2 −6
(L) 2
I( J (R)) : r∗ = −1/2, p∗ = −4
term r′ p′
1st −1/2 −4 (14.115)
2nd −1/2 −5
467
(L) 2
I( J (R)) : r∗ = −1/2, p∗ = −5
term r′ p′
1st 0 −6 (14.116)
2nd −1/2 −5
2. Case X = Oi : i = 1, 2, 3
(Oi ) 2
Ξ( J (R)) : r∗ = −1, p∗ = −3
term r′ p′
1st −1/2 −4 (14.117)
2nd −1 −3
(Oi ) 2
Ξ( J (R)) : r∗ = 2, p∗ = −13/2
term r′ p′
1st 2 −7 (14.118)
2nd 2 −13/2
(Oi ) 2
Θ( J (R)) : r∗ = −1/2, p∗ = −4
term r′ p′
1st −1/2 −4 (14.119)
2nd 0 −4
(Oi ) 2
Θ( J (R)) : r∗ = 3/2, p∗ = −6
term r′ p′
1st 3/2 −13/2 (14.120)
2nd 3/2 −6
(Oi ) 2
Λ( J (R)) : r∗ = 0, p∗ = −4
term r′ p′
1st 0 −5 (14.121)
2nd 0 −4
(Oi ) 2
Λ( J (R)) : r∗ = 1, p∗ = −6
term r′ p′
1st 1 −6 (14.122)
2nd 3/2 −6
(Oi ) 2
K( J (R)) : r∗ = 0, p∗ = −4
468
term r′ p′
1st 0 −5 (14.123)
2nd 0 −4
(Oi ) 2
K( J (R)) : r∗ = 1, p∗ = −6
term r′ p′
1st 1 −6 (14.124)
2nd 3/2 −6
(Oi ) 2
I( J (R)) : r∗ = 1/2, p∗ = −5
term r′ p′
1st 1/2 −5 (14.125)
2nd 1/2 −5
(Oi ) 2
I( J (R)) : r∗ = 1/2, p∗ = −5
term r′ p′
1st 1 −6 (14.126)
2nd 1/2 −5
3. Case X = S
(S) 2
Θ( J (R)) and Ξ( (S) J 2 (R)) (see 14.14).
(S) 2
Ξ( J (R)) : r∗ = 2, p∗ = −13/2
term r′ p′
1st 2 −7 (14.127)
2nd 2 −13/2
(S) 2
Θ( J (R)) : r∗ = 3/2, p∗ = −6
term r′ p′
1st 3/2 −13/2 (14.128)
2nd 3/2 −6
(L) 2
(n) (n) (3)
(n) (n) (3)
(L) τ c,2 : (O) τ c,2 : (S) τ c,2 ,
n = 0, 1, 2, 3, n = 0, 1, 2, 3, and respectively.
469
Each component of the Weyl currents (L) J 2 (R), (Oi ) J 2 (R) : i = 1, 2, 3, and
(S) 2
J (R) being expressed as a sum of terms by 14.98, and these expressions
being substituted into 14.12, 14.13, and 14.14, sums of trilinear terms result, two
of the factors in each term being contributed by the expression for a component
of the Weyl current (L) J 2 (R) in the case of 14.12, (Oi ) J 2 (R) : i = 1, 2, 3 in
the case of 14.13, and (S) J 2 (R) in the case of 14.14, and the other factor being
a component of L̃L R, L̃Oi R : i = 1, 2, 3, and L̃S R, respectively, multiplied
by Ω3 and the appropriate power of |u|2 . Viewing these third factors in a
way which would be valid also in the case of the 2nd order Weyl current error
estimates, we see each third factor as being either O(δ r3 |u|p3 ) in the case of
factors involving α(L̃X R), β(L̃X R), ρ(L̃X R), σ(L̃X R), β(L̃X R), or O(δ r3 |u|p3 )
in the case of factors involving α(L̃X R). The values of r3 and p3 assigned are
those implied by the bound on the quantity P1 . To each trilinear term we can
then apply accordingly one of the last two cases of Lemma 13.1. Since the pair
r′ , p′ corresponding to the first two factors of each trilinear term may be replaced
by the corresponding r∗ , p∗ , we then obtain a bound for the contribution of all
terms resulting from the product of a given component of (X) J 2 with a given
component of L̃X R to the corresponding error integral
Z
(n)
δ 2qn +2l | (X) τ c,2 |dµg (14.129)
Mc′ ∗
by O(δ e ), where e is the excess index (see 13.32):
e = 2qn + 2l + r∗ + r3 + 1 (14.130)
provided that the integrability index s, defined by (see 13.33):
s = p∗ + p3 + 3 (14.131)
each of 14.12 - 14.14.
1. Case X = L : l = 1
(0)
(L) τ c,2
term e s
1st 1 −1 (14.132)
2nd 3/2 −2
(1)
(L) τ c,2
470
term e s
1st 1 −2
(14.133)
2nd 1 −2
3rd 1 −1
(2)
(L) τ c,2
term e s
1st 1 −1
(14.134)
2nd 1 −1
3rd 3/2 −2
(3)
(L) τ c,2
term e s
1st 1 −3/2 (14.135)
2nd 1 −1
2. Case X = O : l = 0
(0)
(O) τ c,2
term e s
1st 1 −2 (14.136)
2nd 3/2 −2
(1)
(O) τ c,2
term e s
1st 1 −2
(14.137)
2nd 1 −2
3rd 1 −2
(2)
(O) τ c,2
term e s
1st 1 −2
(14.138)
2nd 1 −2
3rd 3/2 −2
(3)
(O) τ c,2
471
term e s
1st 1 −3/2 (14.139)
2nd 1 −3/2
3. Case X = S : l = 0
(3)
(S) τ c,2
term e s
1st 1 −3/2 (14.140)
2nd 1 −3/2
We see that all terms have negative integrability index so that Lemma 13.1
indeed applies. All terms have positive excess index, so there are no borderline
terms.
Z
(0)
δ 2q0 +2 | (L) τ c,2 |dµg ≤ O(δ)
Mc′ ∗
Z
(L) (1)
δ 2q1 +2 | τ c,2 |dµg ≤ O(δ)
Mc′ ∗
Z
(L) (2)
δ 2q2 +2 | τ c,2 |dµg ≤ O(δ)
Mc′ ∗
Z
2q3 +2 (L) (3)
δ | τ c,2 |dµg ≤ O(δ)
Mc′ ∗
Z
(O) (0)
δ 2q0 | τ c,2 |dµg ≤ O(δ)
Mc′ ∗
Z
2q1 (O) (1)
δ | τ c,2 |dµg ≤ O(δ)
Mc′ ∗
Z
(O) (2)
δ 2q2 | τ c,2 |dµg ≤ O(δ)
Mc′ ∗
Z
2q3 (O) (3)
δ | τ c,2 |dµg ≤ O(δ)
Mc′ ∗
472
and: Z
2q3 (S) (3)
δ | τ c,2 |dµg ≤ O(δ)
Mc′ ∗

As we remarked in Chapter 12, for any commutation field X the associated
3-covariant tensorfield (X) q defined by 14.7 has the algebraic properties of a
Weyl current. It therefore decomposes into the components Ξ( (X) q), Ξ( (X) q),
Θ( (X) q), Θ( (X) q), Λ( (X) q), Λ( (X) q), K( (X) q), K( (X) q), I( (X) q), I( (X) q).
The following lemma is established in a straightforward manner.
Lemma 14.2 The components of the Weyl current (X) J 3 associated to the
commutation field X and to the Weyl field W are given by:
4Ξ( J (W )) = −3I ♯ ( (X) q) · α(W ) + 3Λ( (X) q)β(W ) + 6K(

(X) 3 (X)
q) ∗ β(W )
+3Ξ( (X) q)ρ(W ) + 2 ∗ Ξ( (X) q)σ(W )
(X) 3
4Ξ( J (W )) = −3I ♯ ( (X)
q) · α(W ) − 3Λ( (X)
q)β(W ) − 6K( (X)
q) ∗ β(W )
(X)
+3Ξ( q)ρ(W ) − 2 ∗ Ξσ(W )
(X) 3 (X) (X)
4Θ( J (W )) = 3Λ( q)α(W ) − 3K( q) ∗ α(W ) − 6I( (X) ˆ
q)⊗β(W )
(X) ∗ (X)
+3Θ( q)ρ(W ) + 3 Θ( q)σ(W )
(X) 3 (X) (X) ∗ (X) ˆ
4Θ( J (W )) = 3Λ( q)α(W ) − 3K( q) α(W ) + 6I( q)⊗β(W )
+3Θ( (X) q)ρ(W ) − 3 ∗ Θ( (X) q)σ(W )
1
4Λ( (X) J 3 (W )) = (Θ( (X) q), α(W )) − 3Λ( (X) q)ρ(W ) + 3K( (X) q)σ(W )
2
+(Ξ( (X) q), β(W ))
1
4Λ( (X) J 3 (W )) = (Θ( (X) q), α(W )) − 3Λ( (X) q)ρ(W ) − 3K( (X) q)σ(W )
2
−(Ξ( (X) q), β(W ))
1
4K( (X) J 3 (W )) = − Θ( (X) q) ∧ α(W ) + 2I( (X) q) ∧ β(W )
2
−3Λ( (X) q)σ(W ) − 3K( (X) q)ρ(W ) + Ξ( (X) q) ∧ β(W )
1
4K( (X) J 3 (W )) = − Θ( (X) q) ∧ α(W ) − 2I( (X) q) ∧ β(W )
2
+3Λ( (X) q)σ(W ) − 3K( (X) q)ρ(W ) − Ξ( (X) q) ∧ β(W )
4I( J (W )) = −Ξ♯ ( (X) q) · α(W ) − 4Λ( (X) q)β(W )
(X) 3
−3I( (X)
q)ρ(W ) − 3 ∗ I( (X)
q)σ(W ) − 2Θ( (X)
q) · β ♯ (W )
(X) 3
4I( J (W )) = −Ξ♯ ( (X)
q) · α(W ) + 4Λ( (X)
q)β(W )
(X)
−3I( q)ρ(W ) + 3 ∗ I( (X)
q)σ(W ) + 2Θ( (X)
q) · β ♯ (W )
473
Using the table 1.175 of connection coefficients of the frame (eµ : µ =
1, 2, 3, 4) and taking into account the basic identity 1.187 we obtain:
(X)
2Ξ( q) = Ω−1 D (X)
m + χ♯ · m − Ω−1 ωm (14.141)
(X)
2Ξ( q) = Ω−1 D (X)
m + χ♯ · m − Ω−1 ωm (14.142)

(X) −1 1(X) (X) (X)
2Θ( q) = 2Ω D̂
î − Ωtrχ î − ∇ˆ
/⊗ m
2
+2χ̂ (X) j − (2η + ζ)⊗
ˆ (X) m (14.143)

(X) −1 1(X) (X) (X)
2Θ( q) = 2Ω D̂
î − Ωtrχ î − ∇ˆ
/⊗ m
2
+2χ̂ (X) j − (2η − ζ)⊗
ˆ (X) m (14.144)
(X)
2Λ( q) = −Ω−1 D (X)
j − trχ (X)
j + div
/ (X)
m
2
−(χ̂, (X) î) + (2η + ζ, (X) m) − (X)
p4 (14.145)
3
(X)
2Λ( q) = −Ω−1 D (X)
j − trχ (X)
j + div
/ (X)
m
(X) (X) 2 (X)
−(χ̂, î) + (2η − ζ, m) − p3 (14.146)
3
(X) (X) (X) (X)
2K( q) = curl
/ m + χ̂ ∧ î + ζ ∧ m (14.147)
(X) X) (X) (X)
2K( q) = curl
/ m + χ̂ ∧ î − ζ ∧ m (14.148)
(X)
2I( q) = Ω−1 D (X)
m − Ω−1 ω (X)
m − /d (X) j − 2η (X)
j
2
−2η♯ · (X) î + χ♯ · (X) m + (X) /p (14.149)
3
(X)
2I( q) = Ω−1 D X)
m − Ω−1 ω (X)
m − /d (X) j − 2η (X)
j
2
−2η ♯ · (X) î + χ♯ · (X) m + (X) /p (14.150)
3
We consider as a single term the first term on the right hand side of 14.144
because there is a crucial cancellation involved here as we shall presently see.
For the sake of symmetry we also consider as a single term the first term on the
right hand side of 14.143 although there is no cancellation involved.
Consider first the case X = L. Then by the first of 9.3:
(L)
D̂ î = O4 (δ −1/2 |u|−2 ) (14.151)
474
However, since according to the first of 8.22:
(L)
î = 2Ωχ̂, (14.152)
according to 4.6 we have:
(L) 1 (L)
D̂ î − Ωtrχ î = Ω2 {2∇ˆ + 2η ⊗η
/⊗η ˆ − trχχ̂} (14.153)
2
The estimates of Chapter 4 for the right hand side then give:
(L) 1 (L)
D̂ î − Ωtrχ î = O4 (δ 1/2 |u|−3 ) (14.154)
2
Consider next the case X = S. Then by the first of 9.14:
(S)
D̂ î = O4 (δ 1/2 |u|−2 ) (14.155)
However, according to the first of 8.30:

(S)
î = 2Ω(uχ̂ + uχ̂), (14.156)
hence:
(S) 1 (S)
D̂ î − Ωtrχ î = 2Ω2 uD̂(Ω−1 χ̂)
2
1
+2u D̂(Ωχ̂) − Ωtrχ(Ωχ̂)
2

1
+4Ωuωχ̂ + 2Ωχ̂ 1 − uΩtrχ (14.157)
2
Now, according to 3.11:
D̂(Ω−1 χ̂) = −α (14.158)
It follows that the first term on the right in 14.157 is O4 (δ 3/2 |u|−7/2 ). By 14.154
the second term is O4 (δ 3/2 |u|−3 ). Also, by the estimates of Chapter 3 the third
and fourth terms are O4 (δ 3/2 |u|−4 ) and O4 (δ 3/2 |u|−3 ) respectively. We conclude
that:
1
D̂ (S) î − Ωtrχ (S) î = O4 (δ 3/2 |u|−3 ) (14.159)
2
Consider finally the case X = Oi : i = 1, 2, 3. Then by the first of 9.91:
(Oi )
D̂ î = O4 (δ 1/2 |u|−2 ) (14.160)
(Oi ) (Oi )
However î = π̂
/ satisfies the propagation equation 8.110:
(Oi ) (Oi ) (Oi )
D̂ π̂
/ − Ωtrχ π̂
/ = −Ωχ̂tr π
/ + 2L̂
/Oi (Ωχ̂) (14.161)
hence we have:
(Oi ) 1 (Oi ) 1 (Oi ) (Oi )
D̂ î − Ωtrχ î = Ωtrχ π̂ /Oi (Ωχ̂) − Ωχ̂tr
/ + 2L̂ π
/ (14.162)
2 2
475
Let us define the symmetric 2-covariant S tensorfields:
(Oi )
ϑi = π /Oi (Ωχ̂)
/ + 2uL (14.163)
Let us also set:

1
ε=
uΩtrχ − 1 (14.164)
2
Then, taking into account the fact that:
/Oi (Ωχ̂) = g/−1 · L

trL /Oi (g/−1 ) = (Ωχ̂,
/Oi (Ωχ̂) = −Ωχ̂ · L (Oi )
π̂
/) (14.165)
14.162 takes the form:

1
u D̂ (Oi ) î − Ωtrχ (Oi )
î = ϑi + ε (Oi )
π̂
/ − ug/(Ωχ̂, (Oi )
π̂
/)
2
1 (Oi ) (Oi )
− g/tr π
/ − uΩχ̂tr π
/ (14.166)
2
Now, by the estimates of Chapter 3:
ε = O∞ (δ|u|−1 ) (14.167)
hence by Proposition 8.3 the second term on the right in 14.166 is O4 (δ 3/2 |u|−2 ).
Moreover, by the same proposition and the results of Chapter 3 the third,
fourth and fifth terms on the right in 14.166 are O4 (δ|u|−2 ), O4 (δ|u|−2 ) and
O4 (δ 3/2 |u|−3 ) respectively. It follows that we will have shown that:
(Oi ) 1 (Oi )
D̂ î − Ωtrχ î = O4 (δ|u|−3 ) (14.168)
2
once we establish the following lemma.
Lemma 14.3 We have:

ϑi = O4 (δ|u|−2 )
Proof: Consider first Dϑi along Cu0 . Since [L, Oi ] = 0 along Cu0 , we have:
/Oi (Ωχ̂) = L
DL /Oi D(Ωχ̂) : along Cu0 (14.169)
Now, taking into account the fact that
trD(Ωχ̂) = g/−1 · D(Ωχ̂) = −Ωχ̂ · D(g/−1 ) = 2(Ωχ̂, Ωχ̂) (14.170)
4.5 gives:

1 1
D(Ωχ̂) = Ω2 ∇ ˆ + η ⊗η
/⊗η ˆ + trχχ̂ − trχχ̂ + g/(χ̂, χ̂) (14.171)
2 2
Hence:
2 1
/Oi D(Ωχ̂) = Ω
L − trχL
/Oi χ̂ + κi (14.172)
2
476
where the κi are the symmetric 2-covariant S tensorfields:

1 1
κi = L/Oi (∇ˆ + η ⊗η)
/⊗η ˆ +L /Oi trχχ̂ + g/(χ̂, χ̂) − (Oi trχ)χ̂
2 2

1 1
+2(Oi log Ω) ∇ ˆ + η ⊗η
/⊗η ˆ + trχχ̂ − trχχ̂ + g/(χ̂, χ̂) (14.173)
2 2
By Proposition 6.2 and the estimates of Chapters 3 and 4 together with Propo-
sitions 8.1 and 8.2 the first term on the right in 14.173 is seen to be O4 (|u|−3 ).
Also, by the estimates of Chapters 3 and 4 and the estimate 7.220 together with
Propositions 8.1 and 8.2 the remaining terms on the right hand side of 14.173
are seen to be O4 (|u|−3 ) as well. We conclude that:
κi = O4 (|u|−3 ) (14.174)
Considering Dϑi along Cu0 and substituting for DL /Oi (Ωχ̂) along Cu0 from
14.169, 14.172 and for D (Oi ) π
/ along Cu0 from 8.63 (and recalling the definition
14.164 and the fact that Du = 0), we obtain the following propagation equation
for ϑi along Cu0 :
Dϑi − Ωtrχϑi = ϕi : along Cu0 (14.175)
where the ϕi are the symmetric 2-covariant S tensorfields:
/Oi (Ωχ̂) + 2(1 + ε)(Oi Ω)χ̂

ϕi = −2εL
−2uΩtrχL/Oi (Ωχ̂) + g/Oi (Ωtrχ) + 2uΩ2 κi (14.176)
By the results of Chapters 3 and 4 together with Propositions 8.1 and 8.2 the first
four terms on the right in 14.176 are seen to be O4 (δ 1/2 |u|−2 ), O4 (δ 1/2 |u|−3 ),
O4 (δ 1/2 |u|−3 ), O4 (|u|−2 ), respectively. The last term is O4 (|u|−2 ) by 14.174.
We conclude that:
ϕi = O4 (|u|−2 ) (14.177)
Noting that the ϑi vanish on C 0 , we apply Lemma 4.6 to the propagation
equation 14.175 along Cu0 , taking p = 4. Here r = 2, ν = 2 andγ = 0. We
obtain: Z u
kϑi kL4 (Su,u0 ) ≤ C kϕi kL4 (Su′ ,u ) du′ (14.178)
0
0
Substituting 14.177 then yields:
kϑi kL4 (Su,u0 ) ≤ O(δ|u0 |−3/2 ) (14.179)
Consider next Dϑi along each C u . Since [L, Oi ] = 0 on M ′ we have:
/Oi (Ωχ̂) = L
DL /Oi D(Ωχ̂) : on M ′ (14.180)
Now, taking into account the fact that:
trD(Ω−1 χ̂) = g/−1 · D(Ω−1 χ̂) = −Ω−1 χ̂ · D(g/−1 ) = 2|χ̂|2 (14.181)
477
14.158 gives:
D(Ωχ̂) = Ω2 −α + g/|χ̂|2 + 2Ωωχ̂

(14.182)
Considering Dϑi along each C u and substituting for DL /Oi (Ωχ̂) from 14.180 and
/Oi applied to 14.182 and for D (Oi ) π
L / along Cu0 from 8.61 (and recalling the
definition 14.164 and the fact that Du = 1), we obtain the following propagation
equation for ϑi along the C u :
Dϑi − Ωtrχϑi = ϕi (14.183)
where the ϕi are the symmetric 2-covariant S tensorfields:
/Oi (Ωχ̂) + g/Oi (Ωtrχ)

ϕi = −4εL
/Oi Ω2 −α + g/|χ̂|2 + 2Ωωχ̂

+2uL (14.184)
The results of Chapters 3 and 4 together with Propositions 8.1 and 8.2 imply:
ϕi = O4 (δ|u|−3 ) (14.185)
To the propagation equation 14.183 we apply Lemma 4.7 taking p = 4. Here

r = 2, ν = 2, γ = 0 and we obtain:
Z u
−1/2 −1/2
|u| kϑi k L4 (S u,u )
≤ C|u0 | kϑi k L4 (S u,u0 )
+C |u′ |−1/2 kϕi kL4 (Su,u′ ) du′
u0
(14.186)
Substituting 14.179 and 14.185 then yields:
kϑi kL4 (Su,u ) ≤ O(δ|u|−3/2 ) : ∀(u, u) ∈ Dc′ ∗ (14.187)
and the lemma is established.
Let us now go back to expressions 14.141 - 14.150. Consider first the case
X = L. Using the estimates 9.1 - 9.3 and 14.154 as well as 14.102 and the
estimates 8.32 together with the estimates of Chapter 3, we deduce:
(L)
Ξ( q) = 0
(L)
Ξ( q) = O4 (δ 1/2 |u|−3 )
(L)
Θ( q) = O4 (δ −3/2 |u|−1 )
(L)
Θ( q) = O4 (δ 1/2 |u|−3 )
(L)
Λ( q) = O4 (δ −1 |u|−2 )
(L)
Λ( q) = O4 (|u|−2 )
(L)
K( q) = 0
(L)
K( q) = O4 (|u|−3 )
(L)
I( q) = O4 (δ −1/2 |u|−2 )
(L)
I( q) = O4 (δ −1/2 |u|−2 ) (14.188)
478
Consider next the case X = Oi : i = 1, 2, 3. Using the estimates 9.89 - 9.91 and
14.168 as well as 14.103 and the estimates 8.144 together with the estimates of
Chapter 3, we deduce:
(Oi )
Ξ( q) = O4 (δ −1/2 |u|−1 )
(Oi )
Ξ( q) = 0
(Oi )
Θ( q) = O4 (δ −1/2 |u|−1 )
(Oi )
Θ( q) = O4 (δ|u|−3 )
(Oi )
Λ( q) = O4 (|u|−2 )
(Oi )
Λ( q) = O4 (δ|u|−3 )
(Oi )
K( q) = O4 (|u|−2 )
(Oi )
K( q) = O4 (δ|u|−3 )
(Oi )
I( q) = O4 (δ 1/2 |u|−2 )
(Oi )
I( q) = O4 (δ 1/2 |u|−2 ) (14.189)
Consider finally the case X = S. Using the estimates 9.12 - 9.14 and 14.159 as
well as 14.104 and the estimates 8.34 together with the estimates of Chapter 3,
we deduce:
(S)
Ξ( q) = O4 (δ −1/2 |u|−1 )
Ξ( (S) q) = O4 (δ 3/2 |u|−3 )
Θ( (S) q) = O4 (δ −1/2 |u|−1 )
(S)
Θ( q) = O4 (δ 3/2 |u|−3 )
(S)
Λ( q) = O4 (|u|−2 )
(S)
Λ( q) = O4 (δ|u|−3 )
(S)
K( q) = O4 (|u|−2 )
(S)
K( q) = O4 (δ|u|−3 )
(S)
I( q) = O4 (δ 1/2 |u|−2 )
(S)
I( q) = O4 (δ 1/2 |u|−2 ) (14.190)
The expression for each component of (X) J 3 (W ) given by Lemma 14.2 is

a sum of terms one factor of which is a component of (X) q and the other
factor a component of W . According to the above, each component of (X) q
is O4 (δ r1 |u|p1 ) for some r1 , p1 . Our attention is at present confined to the
case W = R. Viewing the components of W = R in a way which would be
valid also in the case of the 2nd order Weyl current error estimates we see each
component as being O4 (δ r2 |u|p2 ) according to 14.105. We set, for each term in
the expression of a given component of (X) J 3 (R),
r′ = r1 + r2 , p′ = p1 + p2 (14.191)

the minimal r′ and p∗ is the maximal p′ occuring in the terms constituting the
479
expression of that component. We obtain in this way the following tables. The
ordinals in these tables refer to the terms in the expressions of Lemma 14.2.
1. Case X = L
(L) 3
Ξ( J (R)) : r∗ = −2, p∗ = −3
term r′ p′
1st −2 −3 (14.192)
2nd −3/2 −4
(the remaining terms vanish)
(L) 3
Ξ( J (R)) : r∗ = 1/2, p∗ = −6
term r′ p′
1st 1 −13/2
2nd 1 −6
(14.193)
3rd 1 −7
4th 1/2 −6
5th 1/2 −6
(L) 3
Θ( J (R)) : r∗ = −3/2, p∗ = −3
term r′ p′
1st −3/2 −3
2nd −3/2 −4
(14.194)
3rd −1 −4
4th −3/2 −4
5th −3/2 −4
(L) 3
Θ( J (R)) : r∗ = 1/2, p∗ = −6
term r′ p′
1st 1/2 −13/2
3rd 1/2 −6 (14.195)
4th 1/2 −6
5th 1/2 −6
(L) 3
Λ( J (R)) : r∗ = −1, p∗ = −4
term r′ p′
1st −1 −4 (14.196)
2nd −1 −5
480
(L) 3
Λ( J (R)) : r∗ = 0, p∗ = −5
term r′ p′
1st 0 −11/2
2nd 0 −5 (14.197)
3rd 0 −6
4th 0 −5
(L) 3
K( J (R)) : r∗ = −1, p∗ = −4
term r′ p′
1st −1 −4
(14.198)
2nd −1 −4
3rd −1 −5
(L) 3
K( J (R)) : r∗ = 0, p∗ = −5
term r′ p′
1st 0 −11/2
2nd 1/2 −6
(14.199)
3rd 0 −5
4th 0 −6
5th 0 −5
(L) 3
I( J (R)) : r∗ = −1, p∗ = −4
term r′ p′
1st −1 −4
2nd −1/2 −4
(14.200)
3rd −1/2 −5
4th −1/2 −5
5th −1/2 −5
(L) 3
I( J (R)) : r∗ = −1/2, p∗ = −5
term r′ p′
2nd 0 −6
3rd −1/2 −5 (14.201)
4th −1/2 −5
5th 0 −5
(the 1st term vanishes)
481
2. Case X = Oi : i = 1, 2, 3
(Oi ) 3
Ξ( J (R)) : r∗ = −1, p∗ = −3
term r′ p′
1st −1 −3
2nd −1/2 −4
(14.202)
3rd −1/2 −4
4th −1/2 −4
5th −1/2 −4
(Oi ) 3
Ξ( J (R)) : r∗ = 2, p∗ = −13/2
term r′ p′
1st 2 −13/2
(14.203)
2nd 2 −7
3rd 2 −7
(Oi ) 3
Θ( J (R)) : r∗ = −1/2, p∗ = −4
term r′ p′
1st −1/2 −4
2nd −1/2 −4
(14.204)
3rd 0 −4
4th −1/2 −4
5th −1/2 −4
(Oi ) 3
Θ( J (R)) : r∗ = 1, p∗ = −6
term r′ p′
1st 3/2 −13/2
2nd 3/2 −13/2
(14.205)
3rd 3/2 −6
4th 1 −6
5th 1 −6
(Oi ) 3
Λ( J (R)) : r∗ = −1/2, p∗ = −4
term r′ p′
1st −1/2 −4
2nd 0 −5 (14.206)
3rd 0 −5
4th 1/2 −5
(Oi ) 3
Λ( J (R)) : r∗ = 1, p∗ = −11/2
482
term r′ p′
1st 1 −11/2
(14.207)
2nd 1 −6
3rd 1 −6
(the 4th term vanishes)
(Oi ) 3
K( J (R)) : r∗ = −1/2, p∗ = −4
term r′ p′
1st −1/2 −4
2nd 0 −4
(14.208)
3rd 0 −5
4th 0 −5
5th 1/2 −5
(Oi ) 3
K( J (R)) : r∗ = 1, p∗ = −11/2
term r′ p′
1st 1 −11/2
2nd 3/2 −6 (14.209)
3rd 1 −6
4th 1 −6
(the 5th term vanishes)
(Oi ) 3
I( J (R)) : r∗ = 1/2, p∗ = −5
term r′ p′
2nd 1/2 −5
3rd 1/2 −5 (14.210)
4th 1/2 −5
5th 1/2 −5
(the 1st term vanishes)
(Oi ) 3
I( J (R)) : r∗ = 1/2, p∗ = −5
term r′ p′
1st 1 −11/2
2nd 1 −6
(14.211)
3rd 1/2 −5
4th 1/2 −5
5th 1/2 −5
3. Case X = S
483
(S) 3
Θ( J (R)) and Ξ( (S) J 3 (R)) (see 14.14).
(S) 3
Ξ( J (R)) : r∗ = 3/2, p∗ = −6
term r′ p′
1st 2 −13/2
2nd 2 −7
(14.212)
3rd 2 −7
4th 3/2 −6
5th 3/2 −6
(S) 3
Θ( J (R)) : r∗ = 3/2, p∗ = −6
term r′ p′
1st 3/2 −13/2
2nd 3/2 −13/2
(14.213)
3rd 3/2 −6
4th 3/2 −6
5th 3/2 −6
(L) 3
(n) (n) (3)
(n) (n) (3)
(L)
n = 0, 1, 2, 3, (O) τ c,3 : n = 0, 1, 2, 3, and (S) τ c,3 , respectively.
τ c,3 :
Each component of the Weyl currents (L) J 3 (R), (Oi ) J 3 (R) : i = 1, 2, 3,
and (S) J 3 (R) being expressed as a sum of terms by Lemma 14.2, and these
expressions being substituted into 14.12, 14.13, and 14.14, sums of trilinear
terms result, two of the factors in each term being contributed by the expression
for a component of the Weyl current (L) J 3 (R) in the case of 14.12, (Oi ) J 3 (R) :
i = 1, 2, 3 in the case of 14.13, and (S) J 3 (R) in the case of 14.14, and the other
factor being a component of L̃L R, L̃Oi R : i = 1, 2, 3, and L̃S R, respectively,
multiplied by Ω3 and the appropriate power of |u|2 . Viewing these third factors
in a way which would be valid also in the case of the 2nd order Weyl current
error estimates, we see each third factor as being either O(δ r3 |u|p3 ) in the case of
factors involving α(L̃X R), β(L̃X R), ρ(L̃X R), σ(L̃X R), β(L̃X R), or O(δ r3 |u|p3 )
in the case of factors involving α(L̃X R). The values of r3 and p3 assigned are
those implied by the bound on the quantity P1 . To each trilinear term we can
then apply accordingly one of the last two cases of Lemma 13.1. Since the pair
r′ , p′ corresponding to the first two factors of each trilinear term may be replaced
by the corresponding r∗ , p∗ , we then obtain a bound for the contribution of all
terms resulting from the product of a given component of (X) J 3 with a given
484
component of L̃X R to the corresponding error integral
Z
2qn +2l (n)
δ | (X) τ c,3 |dµg (14.214)
Mc′ ∗
by O(δ e ), where e is the excess index, defined as in 14.130, provided that the
integrability index s, defined as in 14.131, is negative so that Lemma 13.1 ap-
plies. We obtain in this way the following tables. The ordinals in these tables
refer to the terms on the right hand side of each of 14.12 - 14.14.
1. Case X = L : l = 1
(0)
(L) τ c,3
term e s
1st 1 −1 (14.215)
2nd 3/2 −2
(1)
(L) τ c,3
term e s
1st 1 −2
(14.216)
2nd 1 −2
3rd 1/2 −1
(2)
(L) τ c,3
term e s
1st 1 −1
(14.217)
2nd 1 −1
3rd 3/2 −2
(3)
(L) τ c,3
term e s
1st 1 −3/2 (14.218)
2nd 1/2 −1
2. Case X = O : l = 0
(0)
(O)
τ c,3
485
term e s
1st 1 −2 (14.219)
2nd 3/2 −2
(1)
(O) τ c,3
term e s
1st 1/2 −2
(14.220)
2nd 1/2 −2
3rd 1 −2
(2)
(O) τ c,3
term e s
1st 1 −3/2
(14.221)
2nd 1 −3/2
3rd 3/2 −2
(3)
(O) τ c,3
term e s
1st 1/2 −3/2 (14.222)
2nd 1 −3/2
3. Case X = S : l = 0
(3)
(S) τ c,3
term e s
1st 1 −3/2 (14.223)
2nd 1/2 −1
We see that all terms have negative integrability index so that Lemma 13.1
indeed applies. All terms have positive excess index, so there are no borderline
terms.
486
Z
2q0 +2 (0)
δ | (L) τ c,3 |dµg ≤ O(δ)
Mc′ ∗
Z
(L) (1)
δ 2q1 +2 | τ c,3 |dµg ≤ O(δ 1/2 )
Mc′ ∗
Z
2q2 +2 (L) (2)
δ | τ c,3 |dµg ≤ O(δ)
Mc′ ∗
Z
(L) (3)
δ 2q3 +2 | τ c,3 |dµg ≤ O(δ 1/2 )
Mc′ ∗
Z
2q0 (O) (0)
δ | τ c,3 |dµg ≤ O(δ)
Mc′ ∗
Z
(O) (1)
δ 2q1 | τ c,3 |dµg ≤ O(δ 1/2 )
Mc′ ∗
Z
2q2 (O) (2)
δ | τ c,3 |dµg ≤ O(δ)
Mc′ ∗
Z
(O) (3)
δ 2q3 | τ c,3 |dµg ≤ O(δ 1/2 )
Mc′ ∗
and: Z
(S) (3)
δ 2q3 | τ c,3 |dµg ≤ O(δ 1/2 )
Mc′ ∗
We have now completed the 1st order Weyl current error estimates.
487
Chapter 15
The 2nd Order Weyl

Current Error Estimates
15.1 The 2nd order estimates which are of the

same form as the 1st order estimates
In the present section we shall estimate the contributions to the 2nd order Weyl
current error integrals of the terms
(Y ) 1 (Y ) 2 (Y ) 3
J (L̃X R) + J (L̃X R) + J (L̃X R)
in (Y X) J, (Y, X) = (L, L), (Oi , L) : i = 1, 2, 3, (Oj , Oi ) : i, j = 1, 2, 3, (Oi , S) :

i = 1, 2, 3, (S, S) (see 14.15). These contributions are identical in form to the
(n)
(LL) τ c
1st order Weyl current error integrals. Thus the contribution to is
(n)
(L) τ
identical in form to c but with the derived Weyl field L̃L R in the role of
(n)
(OL) τ c
the fundamental Weyl field R. The contribution to is identical in form
(n) (n)
(O) τ c (OO) τ c
to but with L̃L R in the role of R. The contribution to is
(n)
(O) τ c
identical in form to but with the derived Weyl fields L̃Oi R : i = 1, 2, 3 in
(3)
(OS) τc
the role of the fundamental Weyl field R. Finally, the contributions to
(3) (3) (3)
and to (SS) τ c are identical in form to (O) τ c and (S) τ c but with the
derived Weyl field L̃S R in the role of the fundamental Weyl field R. The genuine
2nd order Weyl current error integrals, namely the contributions to the 2nd order
Weyl current error integrals of the term
(X)
L̃Y J
in (Y X) J shall be estimated in the next section. We pay special attention to

the terms giving rise to borderline error integrals.
488
We first consider the contributions to the 2nd order Weyl current error in-
tegrals of the term
(Y ) 1
J (L̃X R)
i = 1, 2, 3, (S, S). As we have seen in the previous chapter, it is the Weyl current
J 1 which contributes borderline error integrals. We consider the expressions for
the components of (Y ) J 1 (L̃X R) given by Proposition 14.1 with the vectorfield
Y in the role of the vectorfield X and the derived Weyl field L̃X R in the role of
the Weyl field W . The expression for each component of (Y ) J 1 (L̃X R) is a sum
of terms one factor of which is a component of (Y ) π̃. We consider, as before,
the terms with the same such factor as a single term.
Consider first the cases X = L, Oi : i = 1, 2, 3. As we discussed in the
previous chapter, in these expressions we are to substitute for:
D̂α(L̃X R), Dβ(L̃X R), Dρ(L̃X R), Dσ(L̃X R), Dβ(L̃X R)
from the inhomogeneous Bianchi equations of Proposition 12.4 with L̃X R in the
role of the Weyl field W and (X) J in the role of the corresponding Weyl current
J. Then the above will involve:
(X) (X) (X) (X) (X)
Θ( J), I( J), Λ( J), K( J), Ξ( J)
respectively. After this substitution, the other factor of each term in the expres-
sion of a given component of (Y ) J 1 (L̃X R) becomes the sum of a principal part,
which is a sum of ∇/ and D derivatives of components of L̃X R, and D̂α(L̃X R), a
non-principal part, which is a sum of terms involving two factors, one of which is
a connection coefficient and the other a component of L̃X R, and terms involving
the above components of (X) J.
We estimate
∇
/α(L̃X R), ∇
/β(L̃X R), /dρ(L̃X R), /dσ(L̃X R), ∇
/β(L̃X R), ∇
/α(L̃X R)
through Proposition 11.1 (see also 11.126) in terms of
L̂
/Oj α(L̃X R), L /Oj β(L̃X R), L̂
/Oj β(L̃X R), Oj ρ(L̃X R), Oj σ(L̃X R), L /Oj α(L̃X R)
(j = 1, 2, 3) respectively, and the latter through Proposition 12.2, using also the
L∞ bounds for the components of (Oj ) π̃ to estimate the remainders, in terms
of
α(L̃Oj L̃X R), β(L̃Oj L̃X R), ρ(L̃Oj L̃X R), σ(L̃Oj L̃X R), β(L̃Oj L̃X R), α(L̃Oj L̃X R)
(j = 1, 2, 3) respectively. The last are bounded according to the bound on the

quantity P2 .
In the case X = L we estimate
D̂α(L̃L R), Dβ(L̃L R), Dρ(L̃L R), Dσ(L̃L R), Dβ(L̃L R), Dα(L̃L R)
489
through Proposition 12.2, using also the L∞ bounds for the components of (L)
π̃
to estimate the remainders, in terms of
α(L̃L L̃L R), β(L̃L L̃L R), ρ(L̃L L̃L R), σ(L̃L L̃L R), β(L̃L L̃L R), α(L̃L L̃L R)
respectively. The latter are bounded according to the bound on the quantity
P2 .
In the case X = Oi : i = 1, 2, 3, to estimate
D̂α(L̃Oi R), Dβ(L̃Oi R), Dρ(L̃Oi R), Dσ(L̃Oi R), Dβ(L̃Oi R), D̂α(L̃Oi R)
we first express these though Proposition 12.2 in terms of
/Oi β, DOi ρ, DOi σ, DL

/Oi α, DL
D̂L̂ /Oi α
/Oi β, D̂L̂
respectively. The remainders are bounded using the L∞ bounds for the compo-
nents of (Oi ) π̃ and the L4 (S) bounds for their D derivatives. We then express,
using Lemma 1.3 and the commutator 8.64,
(Oi ) (Oi )
D̂L̂ /Zi α +
/Oi D̂α + L̂
/Oi α = L̂ î(Ωχ̂, α) − Ωχ̂( î, α)
DL /Oi Dβ + L
/Oi β = L /Zi β
DOi ρ = Oi Dρ + Zi ρ
DOi σ = Oi Dσ + Zi σ
/Oi Dβ + L
/Oi β = L
DL /Zi β
(Oi ) (Oi )
/Oi α = L̂
D̂L̂ /Oi D̂α + L̂
/Zi α + î(Ωχ̂, α) − Ωχ̂( î, α) (15.1)
The second term on the right in each of the above is estimated using Propositions
8.4 and 9.2 and the bound on the quantity Q1 . The third term present on the
right in the first and last of the above is estimated using the L∞ bounds for
(Oi )
î and Ωχ̂. The first terms on the right in the first five of 15.1 are bounded
through the inequalities 12.164 after substituting on the right the inequalities
12.166, 12.167 and 12.175. In view of Lemma 12.6 we then obtain bounds in
terms of the quantity P2 . The first term in the last of 15.1 is bounded through
the inequality 12.181 after substituting on the right the inequalities 12.182,
12.83 and 12.192. In view of Lemma 12.6 we then obtain bounds in terms of
the quantity P2 .
Considering still the cases X = L, Oi : i = 1, 2, 3, we must also appropriately
estimate D̂α(L̃X R). In the case X = L we express D̂α(L̃L R) through Proposi-
tion 12.2 in terms of D̂D̂α. The remainder is bounded using the L∞ bounds for
the components of (L) π̃ and the L4 (S) estimates for their D derivatives. Then
D̂D̂α is bounded through the inequality 12.228. In view of Lemma 12.6 we then
obtain a bound in terms of the quantity P2 . In the case X = Oi : i = 1, 2, 3
we express D̂α(L̃Oi R) in terms of D̂α(L̃Oi R) and L̂ /S α(L̃Oi R). We already
discussed how the former is estimated. To estimate L̂ /S α(L̃Oi R) we express it
through Proposition 12.2 in terms of L̂ /Oi α. The remainder is bounded using
/S L̂
490
the L∞ bounds for the components of (Oi ) π̃ and the L4 (S) bounds for their L
/S
derivatives (which are expressed in terms of D and D derivatives). Using the
commutation relation
[S, Oi ] = uZi (15.2)
which follows from 8.64 and the first of 8.40, we then express, using Lemma 1.3,
(Oi ) (S) (S) (Oi )
L̂ /Oi α = L̂
/S L̂ /Zi α +
/S α + uL̂
/Oi L̂ î( î, α) − î( î, α) (15.3)
The second term on the right is estimated using Propositions 8.4 and 9.2 and
the bound on the quantity Q1 . The third term on the right is estimated using
the L∞ bounds for (Oi ) î and (S) î. Finally, in view of 12.180, the first term on
the right is estimated through the inequality 12.181 after substituting on the
right the inequalities 12.182, 12.83 and 12.192. In view of Lemma 12.6 we then
obtain a bound in terms of the quantity P2 .
Consider finally the case X = S. In this case only the case n = 3 occurs, so
we only have to consider the expressions for
(Y ) 1 (Y ) 1
Ξ( J (L̃S R)) and Θ( J (L̃S R))
given by Proposition 14.1 with Y = S, Oi : i = 1, 2, 3 in the role of X and L̃S R

in the role of W . These expressions contain
∇
/β(L̃S R), ∇
/α(L̃S R),
Dβ(L̃S R), D̂α(L̃S R), Dβ(L̃S R), D̂α(L̃S R)
We estimate
∇
/β(L̃S R), ∇
/α(L̃S R)
through Proposition 11.1 (see also 11.126) in terms of
L /Oi α(L̃S R) ; i = 1, 2, 3
/Oi β(L̃S R), L̂
respectively, and the latter through Proposition 12.2, using also the L∞ bounds
for the components of (Oi ) π̃ to estimate the remainders, in terms of
β(L̃Oi L̃S R), α(L̃Oi L̃S R) ; i = 1, 2, 3
respectively. The last are bounded according to the bound on the quantity P2 .
To estimate Dβ(L̃S R) we express it through the inhomogeneous Bianchi
equations (Proposition 12.4), with L̃S R in the role of W and (S) J in the role of
J, in terms of /dρ(L̃S R), /dσ(L̃S R), which are estimated through Proposition 11.1
in terms of Oi ρ(L̃S R), Oi σ(L̃S R); i = 1, 2, 3 respectively. The last are estimated
by 13.66. There is also a source term I( (S) J) in the inhomogeneous Bianchi
equation in question, which is estimated following the approach of Chapter 14.
To estimate D̂α(L̃S R) we express it through the inhomogeneous Bianchi
equations, with L̃S R in the role of W and (S) J in the role of J, in terms of
/β(L̃S R), which is estimated through Proposition 11.1 in terms of L
∇ /Oi β(L̃S R).
491
The latter is in turn estimated through Proposition 12.2 in terms of β(L̃Oi L̃S R).
The last is bounded according to the bound on the quantity P2 . There is also a
source term Θ( (S) J) in the inhomogeneous Bianchi equation in question, which
has already been estimated in Chapter 14.
To estimate D̂β(L̃S R) we express it through the inhomogeneous Bianchi
equations (Proposition 12.4) with L̃S R in the role of W and (S) J in the role of
J in terms of ∇ /α(L̃S R). We have already discussed how the latter is estimated.
There is also a source term Ξ( (S) J) in the inhomogeneous Bianchi equation in
question, which has already been estimated in Chapter 14.
Finally, to estimate D̂α(L̃S R) we express it in terms of D̂α(L̃S R) and
L̂
/S α(L̃S R). We have already discussed how the former is estimated. We es-
timate the latter through Proposition 12.2 in terms of α(L̃S L̃S R). The last is
bounded according to the bound on the quantity P2 .
According to the above discussion, the other factor in the expression of a
given component of (Y ) J 1 (L̃X R) is a sum of terms each of which may be viewed
as being either O(δ r |u|p ) or O(δ r |u|p ). We then define r2 to be the minimal r
and p2 to be the maximal p occuring in the terms of the other factor of a term
involving a given component of (Y ) π̃ in the expression of a given component of
(Y ) 1
J (L̃X R). We then set, for each term involving a given component of (Y ) π̃,
which is O∞ (δ r1 |u|p1 ), in the expression of a given component of (Y ) J 1 (L̃X R),
r′ = r1 + r2 , p′ = p1 + p2 (15.4)
We then assign, to each component of (Y ) J 1 (L̃X R), the pair r∗ , p∗ , where r∗ is

the minimal r′ and p∗ is the maximal p′ occuring in the four terms constituting
the expression of that component. We obtain in this way tables which are similar
to the tables 14.28 - 14.31, 14.33, 14.35, 14.36, 14.38 - 14.40, and 14.41 - 14.44,
14.46, 14.48, 14.49, 14.51 - 14.53, and 14.54, 14.55. In fact, the tables for the
case Y = X = L (l = 2) are identical to the tables for the case X = L (l = 1)
of Chapter 14, but with the values of r′ and r∗ decreased by 1. The tables for
the case Y = Oi : i = 1, 2, 3, X = L (l = 1) are identical to the tables for the
case X = Oi : i = 1, 2, 3 (l = 0) of Chapter 14 but with the values of r′ and r∗
decreased by 1. The tables for the case Y = Oj : j = 1, 2, 3, X = Oi : i = 1, 2, 3
(l = 0) are identical to the tables for the case X = Oj : j = 1, 2, 3 (l = 0) of
Chapter 14. The tables for the case Y = Oi : i = 1, 2, 3, X = S (l = 0) are
identical to the tables for the case X = Oi : i = 1, 2, 3 (l = 0) of Chapter 14.
Finally, the tables for the case Y = X = S (l = 0) are identical to the tables
for the case X = S (l = 0) of Chapter 14.
Each component of the Weyl currents (L) J 1 (L̃L R), (Oi ) J 1 (L̃L R) : i =
1, 2, 3, (Oj ) J 1 (L̃Oi R); i, j = 1, 2, 3 and (Oi ) J 1 (L̃S R) : i = 1, 2, 3, (S) J 1 (L̃S R)
being written as a sum of terms in the manner discussed above, and these ex-
pressions being substituted into 14.21, 14.22, 14.23 and 14.24, 14.25 respectively,
by the expression for a component of the Weyl current (L) J 1 (L̃L R) in the case
of 14.21, (Oi ) J 1 (L̃L R) : i = 1, 2, 3 in the case of 14.22, (Oj ) J 1 (L̃Oi R) : i, j =
1, 2, 3 in the case of 14.23 and (Oi ) J 1 (L̃S R) in the case of 14.24, (S) J 1 (L̃S R) in
492
the case of 14.25, and the other factor being a component of L̃L L̃L R, L̃Oi L̃L R :
i = 1, 2, 3, L̃Oj L̃Oi R : i, j = 1, 2, 3 and L̃Oi L̃S R : i = 1, 2, 3, L̃S L̃S R respec-
tively, multiplied by Ω3 and the appropriate power of |u|2 . Each third factor
is either O(δ r3 |u|p3 ) in the case of factors involving α(L̃Y L̃X R), β(L̃Y L̃X R),
ρ(L̃Y L̃X R), σ(L̃Y L̃X R), β(L̃Y L̃X R), or O(δ r3 |u|p3 ) in the case of factors in-
volving α(L̃Y L̃X R). The values of r3 and p3 assigned are those implied by the
bound on the quantity P2 . To each trilinear term we can then apply accordingly
one of the first three cases of Lemma 13.1. Since the pair r′ , p′ corresponding to
the first two factors of each trilinear term may be replaced by the corresponding
r∗ , p∗ , we then obtain a bound for the contribution of all terms resulting from
the product of a given component of (Y ) J 1 (L̃X R) with a given component of
L̃Y L̃X R to the corresponding error integral
Z
(n)
δ 2qn +2l | (Y X) τ c |dµg (15.5)
Mc′ ∗
by O(δ e ), where e is the excess index, defined as in 14.58, provided that the
integrability index s, defined as in 14.59, is negative so that Lemma 13.1 applies.
We obtain in this way tables which are similar to the tables 14.60 - 14.68. In
fact, the tables for the case (Y X) = (LL) (l = 2) are identical to the tables for
the case X = L (l = 1) of Chapter 14. The tables for the case (Y X) = (OL)
(l = 1) are identical to the tables for the case X = O (l = 0) of Chapter 14.
The tables for the case (Y X) = (OO) (l = 0) are identical to the tables for the
case X = O (l = 0) of Chapter 14. Finally, the table for the case (Y X) = (OS)
(l = 0) is identical to the table for the case X = O (l = 0) of Chapter 14 and
the table for the case (Y X) = (SS) (l = 0) is identical to the table for the case
X = S (l = 0) of Chapter 14.
We must now analyze in more detail the terms with vanishing excess index
as they give rise to borderline error integrals. They occur only in the cases
(3)
(LL) τc
n = 1 and n = 3. The terms which give borderline contributions to
(3) (3) (3) (3)
(OL) τ c , (OO) τ c (OS) τ c , (SS) τ c
, and are the terms:
− 6Ω3 |u|6 trχ(( (L) îρ(L̃L R) − ∗(L) îσ(L̃L R)), α(L̃L L̃L R)) (15.6)
X
− 6Ω3 |u|6 trχ (( (Oi ) îρ(L̃L R) − ∗(Oi ) îσ(L̃L R)), α(L̃Oi L̃L R) (15.7)
i
X
− 6Ω3 |u|6 trχ (( (Oj )
îρ(L̃Oi R) − ∗(Oj )
îσ(L̃Oi R)), α(L̃Oj L̃Oi R)) (15.8)
i,j
and:
X
− 6Ω3 |u|6 trχ (( (Oi )
îρ(L̃S R) − ∗(Oi )
îσ(L̃S R)), α(L̃Oi L̃S R)) (15.9)
i
− 6Ω3 |u|6 trχ(( (S)

îρ(L̃S R) − ∗(S)
îσ(L̃S R)), α(L̃S L̃S R)) (15.10)
respectively.
493
Using the precise bound 14.72, the facts that:
(2)
ktrχρ(L̃L R)kL2 (Cu ) ≤ Cδ −1/2 |u|−3 ( E 1 )1/2 : ∀u ∈ [u0 , c∗ )
(2)
ktrχσ(L̃L R)kL2 (Cu ) ≤ Cδ −1/2 |u|−3 ( E 1 )1/2 : ∀u ∈ [u0 , c∗ ) (15.11)
as well as the fact that:

(3)
kΩ3 |u|3 α(L̃L L̃L R)kL2 (C u ) ≤ Cδ −1/2 ( F 2 )1/2 : ∀u ∈ [0, δ) (15.12)
(see 12.104, 12.132 and 12.140), and following the proof of Case 2 of Lemma
13.1, we deduce that the contribution of the term 15.6 to the error integral 15.5
with (Y X) = (LL) (l = 2) and n = 3 is bounded by:
(2) (3)
CR∞
0 (α)( E 1 )
1/2
( F 2 )1/2 (15.13)
Using the precise bound 14.76, as well as 15.11 and

(2)
ktrχρ(L̃Oi R)kL2 (Cu ) ≤ Cδ 1/2 |u|−3 ( E 1 )1/2 : ∀u ∈ [u0 , c∗ )
(2)
ktrχσ(L̃Oi R)kL2 (Cu ) ≤ Cδ 1/2 |u|−3 ( E 1 )1/2 : ∀u ∈ [u0 , c∗ ) (15.14)
and:
(2)
ktrχρ(L̃S R)kL2 (Cu ) ≤ Cδ 1/2 |u|−3 (( E 1 )1/2 + O(δ)) : ∀u ∈ [u0 , c∗ )
(2)
ktrχσ(L̃S R)kL2 (Cu ) ≤ Cδ 1/2 |u|−3 (( E 1 )1/2 + O(δ)) : ∀u ∈ [u0 , c∗ )
(15.15)
which follow from 13.55, and the facts that:

(3)
kΩ3 |u|3 α(L̃Oi L̃L R)kL2 (C u ) ≤ Cδ 1/2 ( F 2 )1/2 : ∀u ∈ [0, δ) (15.16)
(3)
kΩ3 |u|3 α(L̃Oj L̃Oi RkL2 (C u ) ≤ Cδ 3/2 ( F 2 )1/2 : ∀u ∈ [0, δ) (15.17)
(3)
kΩ3 |u|3 α(L̃Oi L̃S R)kL2 (C u ) ≤ Cδ 3/2 ( F 2 )1/2 : ∀u ∈ [0, δ) (15.18)
(see 12.109, 12.114, 12.119, 12.132 and 12.140), and following the proof of Case
2 of Lemma 13.1, we deduce that the contribution of the term 15.7 to the error
integral 15.5 with (Y X) = (OL) (l = 1) and n = 3, the contribution of the
term 15.8 to the error integral 15.5 with (Y X) = (OO) (l = 0) and n = 3, and
the contribution of the term 15.9 to the error integral 15.5 with (Y X) = (OS)
(l = 0) and n = 3, are all three bounded by:
(2) (3)
/41 (β) + R∞
C(R 0 (β))( E 1 )
1/2
( F 2 )1/2 + O(δ 1/2 ) (15.19)
494
Using the precise bound 14.79, as well as 15.15 and the fact that:
(3)
kΩ3 |u|3 α(L̃S L̃S R)kL2 (C u ) ≤ Cδ 3/2 ( F 2 )1/2 : ∀u ∈ [0, δ) (15.20)
13.1, we deduce that the contribution of the term 15.10 to the error integral
15.5 with (Y X) = (SS) (l = 0) and n = 3 is bounded by:
(2) (3)
C(D0∞ (χ̂) + R∞
0 (α))( E 1 )
1/2
( F 2 )1/2 + O(δ) (15.21)
In view of 14.82, 14.84, the expressions 15.13, 15.19 and 15.21 are in turn
bounded by
(0) (2) (3)
C( E 2 )1/2 ( E 1 )1/2 ( F 2 )1/2 + O(δ 1/2 ) (15.22)
(1) (2) (3)
C( E 2 )1/2 ( E 1 )1/2 ( F 2 )1/2 + O(δ 1/2 ) (15.23)
and
(0) (2) (3)
C D0∞ (χ̂) + ( E 2 )1/2 ( E 1 )1/2 ( F 2 )1/2 + O(δ 1/2 ) (15.24)
respectively.
(1) (1) (1)
The terms which give borderline contributions to (LL) τ c , (OL) τ c , (OO) τ c
are the terms:
n o
2Ω3 |u|2 trχ ( (L) î, α(L̃L R))ρ(L̃L L̃L R) + (L) î ∧ α(L̃L R)σ(L̃L L̃L R) (15.25)
Xn o
2Ω3 |u|2 trχ ( (Oi )
î, α(L̃L R))ρ(L̃Oi L̃L R) + (Oi )
î ∧ α(L̃L R)σ(L̃Oi L̃L R)
i
(15.26)o
Xn
2Ω3 |u|2 trχ ( (Oj )
î, α(L̃Oi R))ρ(L̃Oj L̃Oi R) + (Oj )
î ∧ α(L̃Oi R)σ(L̃Oj L̃Oi R)
i,j
(15.27)
respectively.
(0)
ktrχα(L̃L R)kL2 (Cu ) ≤ Cδ −2 |u|−1 ( E 1 )1/2 : ∀u ∈ [u0 , c∗ ) (15.28)
as well as the facts that:
(2)
kΩ3 |u|2 ρ(L̃L RL̃L R)kL2 (Cu ) ≤ Cδ −3/2 ( E 2 )1/2 : ∀u ∈ [u0 , c∗ )
(2)
kΩ3 |u|2 σ(L̃L RL̃L R)kL2 (Cu ) ≤ Cδ −3/2 ( E 2 )1/2 : ∀u ∈ [u0 , c∗ )(15.29)
13.1 we deduce that the contribution of the term 15.25 to the error integral 15.5
(0) (2)
CR∞
0 (α)( E 1 )
1/2
( E 2 )1/2 (15.30)
495
Using 14.76, as well as 15.28 and
(0)
ktrχα(L̃Oi R)kL2 (Cu ) ≤ Cδ −1 |u|−1 ( E 1 )1/2 : ∀u ∈ [u0 , c∗ ) (15.31)
and the facts that:

(2)
kΩ3 |u|2 ρ(L̃Oi L̃L R)kL2 (Cu ) ≤ Cδ −1/2 ( E 2 )1/2 : ∀u ∈ [u0 , c∗ )
(2)
kΩ3 |u|2 σ(L̃Oi L̃L R)kL2 (Cu ) ≤ Cδ −1/2 ( E 2 )1/2 : ∀u ∈ [u0 , c∗ )(15.32)
(2)
kΩ3 |u|2 ρ(L̃Oj L̃Oi R)kL2 (Cu ) ≤ Cδ 1/2 ( E 2 )1/2 : ∀u ∈ [u0 , c∗ )
(2)
kΩ3 |u|2 σ(L̃Oj L̃Oi R)kL2 (Cu ) ≤ Cδ 1/2 ( E 2 )1/2 : ∀u ∈ [u0 , c∗ )(15.33)
(see 12.108, 12.113, 12.129 and 12.139), and following the proof of Case 1 of
Lemma 13.1, we deduce that the contribution of the term 15.26 to the error
integral 15.5 with (Y X) = (OL) (l = 1) and n = 1 and the contribution of the
term 15.27 to the error integral 15.5 with (Y X) = (OO) (l = 0) and n = 1, are
both bounded by:
(0) (2)
/41 (β) + R∞
C(R 0 (β))( E 1 )
1/2
( E 2 )1/2 + O(δ 1/2 ) (15.34)
In view of 14.82, 14.84 the expressions 15.30, 15.34 are in turn bounded by
(0) (0) (2)
C( E 2 )1/2 ( E 1 )1/2 ( E 2 )1/2 + O(δ 1/2 ) (15.35)
and
(1) (0) (2)
C( E 2 )1/2 ( E 1 )1/2 ( E 2 )1/2 + O(δ 1/2 ) (15.36)
respectively.
We turn to consider the contributions to the 2nd order Weyl current error
integrals of the terms
(Y ) 2 (Y ) 3
J (L̃X R) and J (L̃X R)

i = 1, 2, 3, (S, S). Now, the contributions of (L) J 2 (L̃L R) and (L) J 3 (L̃L R) to
(n) (n) (n)
(LL) τcare identical in form to (L) τ c,2 and (L) τ c,3 respectively, but with
L̃L R in the role of R. The contributions of (Oi ) J 2 (L̃L R) and (Oi ) J 3 (L̃L R) to
(n) (n) (n)
(OL) τcare indentical in form to (O) τ c,2 and (O) τ c,3 respectively, but with
L̃L R in the role of R. The contributions of (Oj ) J 2 (L̃Oi R) and (Oj ) J 3 (L̃Oi R) to
(n) (n) (n)
(OO)
are identical in form to (O) τ c,2 and (O) τ c,3 respectively, but with
τ c
L̃Oi R in the role of R. The contributions of (Oi ) J 2 (L̃S R) and (Oi ) J 3 (L̃S R) to
(3) (3) (3)
(OS) τc (O) τ c,2 (O) τ c,3
are identical in form to and respectively, but with
496
(S) 2 (S) 3
L̃S R in the role of R. Finally, the contributions of J (L̃S R) and J (L̃S R)
(3) (3) (3)
(SS) τ (S) τ (S) τ
to are identical in form to
c and c,2 respectively, but
c,3
with L̃S R in the role of R.
Now, by 12.149, 12.174 and the bounds on the quantity Q2 , we have:
D̂α = O4 (δ −5/2 |u|−1 )

Dβ = O4 (δ −3/2 |u|−2 )
Dρ = O4 (δ −1 |u|−3 )
Dσ = O4 (δ −1 |u|−3 )
Dβ = O4 (|u|−4 )
D̂α = O4 (δ 1/2 |u|−9/2 ) (15.37)
Taking also into account 14.105 and the L∞ bounds on the components of (L)
π̃,
we then deduce from Proposition 12.2 that:
α(L̃L R) = O4 (δ −5/2 |u|−1 )

β(L̃L R) = O4 (δ −3/2 |u|−2 )
ρ(L̃L R) = O4 (δ −1 |u|−3 )
σ(L̃L R) = O4 (δ −1 |u|−3 )
β(L̃L R) = O4 (|u|−4 )
α(L̃L R) = O4 (δ 1/2 |u|−9/2 ) (15.38)
These are similar to 14.105 but with the exponents of δ reduced by 1. It follows
that the tables for the components of (L) J 2 (L̃L R) and (L) J 3 (L̃L R) (l = 2) are
similar to the tables 14.107 - 14.116 and 14.192 - 14.201 for the components of
(L) 2
J (R) and (L) J 3 (R) (l = 1), respectively, but with the values of r′ and r∗
reduced by 1. It then follows that the tables for the contributions of (L) J 2 (L̃L R)
(n)
(L) 3 (LL) τc
and J (L̃L R) to are identical to the tables 14.132 - 14.135 and
(n) (n)
(L) τ c,2 and (L) τ c,3 respectively. Also,
14.215 - 14.218 for the tables for
(Oi ) 2
the components of J (L̃L R) and (Oi ) J 3 (L̃L R) (l = 1) are
similar to the
tables 14.117 - 14.126 and 14.202 - 14.211 for the components of (Oi ) J 2 (R) and
(Oi ) 3
J (R) (l = 0), respectively, but with the values of r′ and r∗ reduced by
1. Then the tables for the contributions of (Oi ) J 2 (L̃L R) and (Oi ) J 3 (L̃L R) to
(n)
(OL) τc are indentical to the tables 14.136 - 14.139 and 14.219 - 14.222 for
(n) (n)
(O) τ
c,2and (O) τ c,3 respectively.
By 12.148 and the bound on the quantity Q2 , together with 14.105 and
Propositions 8.1 and 8.2, we have:
/Oi α = O4 (δ −3/2 |u|−1 )

L̂
/Oi β = O4 (δ −1/2 |u|−2 )
L
Oi ρ = O4 (|u|−3 )
497
Oi σ = O4 (|u|−3 )
/Oi β = O4 (δ|u|−4 )
L
/Oi α = O4 (δ 3/2 |u|−9/2 )
L̂ (15.39)
Taking also into account 14.105 and the L∞ bounds on the components of (Oi )
π̃,
α(L̃Oi R) = O4 (δ −3/2 |u|−1 )
β(L̃Oi R) = O4 (δ −1/2 |u|−2 )
ρ(L̃Oi R) = O4 (|u|−3 )
σ(L̃Oi R) = O4 (|u|−3 )
β(L̃Oi R) = O4 (δ|u|−4 )
α(L̃Oi R) = O4 (δ 3/2 |u|−9/2 ) (15.40)
These are similar to 14.105. It follows that the tables for the components of
(Oj ) 2
J (L̃Oi R) and (Oj ) J 3 (L̃Oi R (l = 0) are similar to the tables 14.117 -
14.126 and 14.202 - 14.211 for the components of (Oj ) J 2 (R) and (Oj ) J 3 (R)
(l = 0), respectively. It then follows that the tables for the contributions of
(n)
(Oj )
J 2 (L̃Oi R) and (Oj ) 3
J (L̃Oi R) to (OO) τc are identical to the tables 14.136
(n) (n)
(O)
- 14.139 and 14.219 - 14.222 for and (O) τ
τ c,2 c,3 respectively.
Expressing
D̂α, Dβ, Dρ, Dσ, Dβ
by means of the homogeneous Bianchi equations and using 12.148 and the bound
on the quantity Q2 as well as 14.105 and the L∞ bounds of Chapter 3 for the
connection coefficients we deduce:
D̂α = O4 (δ −3/2 |u|−2 )
Dβ = O4 (δ −1/2 |u|−3 )
Dρ = O4 (|u|−4 )
Dσ = O4 (|u|−4 )
Dβ = O4 (δ|u|−5 ) (15.41)
Also, according to 12.150 and the bound on the quantity Q2 :
D̂α = O4 (δ 3/2 |u|−11/2 ) (15.42)
Combining 15.41 and 15.42 with 15.37 we conclude that:
/S α = O4 (δ −3/2 |u|−1 )
L̂
/S β = O4 (δ −1/2 |u|−2 )
L
Sρ = O4 (|u|−3 )
Sσ = O4 (|u|−3 )
/S β = O4 (δ|u|−4 )
L
/S α = O4 (δ 3/2 |u|−9/2 )
L̂ (15.43)
498
Taking also into account 14.105 and the L∞ bounds on the components of (S)
π̃,
α(L̃S R) = O4 (δ −3/2 |u|−1 )

β(L̃S R) = O4 (δ −1/2 |u|−2 )
ρ(L̃S R) = O4 (|u|−3 )
σ(L̃S R) = O4 (|u|−3 )
β(L̃S R) = O4 (δ|u|−4 )
α(L̃S R) = O4 (δ 3/2 |u|−9/2 ) (15.44)
These are similar to 14.105. It follows that the tables for the components of
(Oi ) 2
J (L̃S R) and (Oi ) J 3 (L̃S R) (l = 0) are similar to the tables 14.117 - 14.126
and 14.202 - 14.211 for the components of (Oi ) J 2 (R) and (Oi ) J 3 (R) (l = 0), re-
spectively. It then follows that the tables for the contributions of (Oi ) J 2 (L̃S R)
(3)
(Oi ) 3 (OS) τc
and J (L̃S R) to are identical to the tables 14.139 and 14.222
(S) (S)
(O) τ c,2 and (O) τ c,3 respectively. Also,
for the tables for the components
(S) 2
of J (L̃S R) and (S) J 3 (L̃S R) (l = 0) are
similar to the tables 14.127 -
14.128 and 14.212 - 14.213 for the components of (S) J 2 (R) and (S) J 3 (R)
(l = 0), respectively. Then the tables for the contributions of (S) J 2 (L̃S R)
(3)
(S) 3 (SS) τc
and J (L̃S R) to are identical to the tables 14.140 and 14.223 for
(3) (3)
(S) τ c,2 (S) τ c,3
and respectively.
We summarize the results of this section in the following Proposition.
Proposition 15.1 The contributions to the error integral

Z
(n)
δ 2qn +2l | (Y X)) τ c |dµg
Mc′ ∗
of the terms
(Y ) 1 (Y ) 2
J (L̃X R) + J (L̃X R) + J 3 (Y )
(L̃X R)
are bounded as follows.
1. For (Y X) = (LL).
In the case n = 0 by:
O(δ)
(0) (0) (2)
C( E 2 )1/2 ( E 1 )1/2 ( E 2 )1/2 + O(δ 1/2 )

O(δ)
499
(0) (2) (3)
C( E 2 )1/2 ( E 1 )1/2 ( F 2 )1/2 + O(δ 1/2 )
2. For (Y X) = (OL).
O(δ)
(1) (0) (2)
C( E 2 )1/2 ( E 1 )1/2 ( E 2 )1/2 + O(δ 1/2 )
O(δ)
(1) (2) (3)
C( E 2 )1/2 ( E 1 )1/2 ( F 2 )1/2 + O(δ 1/2 )
3. For (Y X) = (OO).
O(δ)
(1) (0) (2)
C( E 2 )1/2 ( E 1 )1/2 ( E 2 )1/2 + O(δ 1/2 )
O(δ)
(1) (2) (3)
C( E 2 )1/2 ( E 1 )1/2 ( F 2 )1/2 + O(δ 1/2 )
4. For (Y X) = (OS).
Here we only have the case n = 3, where we have a bound by:
(1) (2) (3)
C( E 2 )1/2 ( E 1 )1/2 ( F 2 )1/2 + O(δ 1/2 )
5. For (Y X) = (SS).

(0) (2) (3)
C D0∞ (χ̂) + ( E 2 )1/2 ( E 1 )1/2 ( F 2 )1/2 + O(δ 1/2 )
500
15.2 The genuine 2nd order estimates
In the present section we shall estimate the contribution to the 2nd order Weyl
current error integrals of the term
(X)
L̃Y J

i = 1, 2, 3, (S, S) (see 14.15). Again, we pay special attention to the terms giving
rise to borderline error integrals.
We first consider the contribution to the 2nd order Weyl current error inte-
grals of the term
L̃Y (X) J 1 (R)
Special care is needed here, because, as we have seen in the previous chapter, it
is the Weyl current J 1 which contributes borderline error integrals. We consider
the expressions for the components of (X) J 1 (R) given by Proposition 14.1 with
the fundamental Weyl field R in the role of the Weyl field W . The expression
for each component of (X) J 1 (R) is a sum of terms one factor of which is a
component of (X) π̃, and we consider the terms with the same such factor as a
single term. After the substitution for
D̂α, Dβ, Dρ, Dσ, Dβ and D̂α
from the homogeneous Bianchi equations, as discussed in the previous chap-

ter, the other factor of each term in the expression of a given component of
(X) 1
J (R) becomes a sum of a principal part, which is a sum of 1st derivatives
of components of R, and a non-principal part, which is a sum of terms consisting
of two factors, one of which is a connection coefficient and the other a compo-
nent of R. Moreover, only ∇ / or D derivatives of the components α, β, ρ, σ, β
and only ∇ / or D derivatives of the component α occur. Now, the components of
L̃Y (X) J 1 (R) are given by Proposition 12.3 in terms of the L/Y derivatives of the
components of (X) J 1 (R). The remainders are estimated using the L∞ bounds
for the components of (Y ) π̃ and the estimates for the components of (X) J 1 (R)
of Chapter 14. Now if L /Y is applied to a term in the expression of a given
component of (X) J 1 (R) a sum of two terms will result, one of which will have
/Y derivative of the corresponding component of (X) π̃ and
as a first factor the L
the same second factor as the original term, and the other term will have the
component of (X) π̃ as a first factor and L /Y applied to the second factor of the
original term, plus possibly a bilinear expression in (Y ) π/ and the second factor
of the original term, as the other factor. The bilinear expression results from
applying L /Y to the coefficients of the expression constituting the given original
term, which may involve g/ and /ǫ . The first of the two resulting terms is to be
estimated by placing the first factor in L4 (S) using L4 (S) estimates for the 1st
derivatives of the components of (X) π̃, and the second factor in L4 (S) using the
L4 (S) estimates for the 1st derivatives of the componets of R. The second of
the two resulting terms is to be estimated by placing the first factor in L∞ using
501
the L∞ estimates for the components of (X) π̃, the principal part of the second
factor in L2 (Cu ) or L2 (C u ) as shall be described in more detail below, and the
remainder of the second factor in L2 (Cu ). In fact, the estimates of Chapters 3
and 4 together with the L∞ estimates for the components of R and the L4 (S)
estimates for their 1st derivatives (and L∞ bounds for (Y ) π /) give an L4 (S)
2
estimate for the remainder, which implies a corresponding L (Cu ) estimate.
The principal part of the second factor of the second of the two terms re-
sulting from a given term in the expression of a given component of (X) J 1 (R)
by applying L /Y , is a term or a sum of terms of the form:
L̂
/Y ∇
/α, L
/Y ∇
/β, L
/Y /dρ, L
/Y /dσ, L /β and L̂
/Y ∇ /Y ∇
/α (15.45)
(the first and last of these being the trace-free parts relative to the last two
indices of L
/Y ∇
/α and L/Y ∇
/α respectively) and:
/Y D̂α, L
L̂ /Y Dβ and L̂
/Y Dβ, Y Dρ, Y Dσ, L /Y D̂α (15.46)
The first five of each of 15.45, 15.46 is to be estimated in L2 (Cu ), while the last
is to be estimated in L2 (C u ), in the manner to be presently discussed.
Consider first the case Y = L. Then by Lemma 4.1 (see also 10.68 and a
similar identity with α in the role of α) each of 15.45 is equal to
∇
/D̂α, ∇
/Dβ, /dDρ, /dDσ, ∇
/Dβ and ∇
/D̂α (15.47)
respectively, plus a remainder which can be estimated in L4 (S), therefore also

in L2 (Cu ) using the estimates of Chapter 4 together with the L∞ estimates for
the components of R. Also, each of 15.46 is equal to
D̂2 α, D2 β, D2 ρ, D2 σ, D2 β and D̂D̂α (15.48)
respectively. The first five of each of 15.47, 15.48 are estimated in L2 (Cu )
according to the bound on the quantity Q2 . The last of 15.47 is estimated in
L2 (C u ) through inequality 12.193 and the bounds on the quantities P2 and Q2 .
The last of 15.48 is estimated in L2 (C u ) through inequalities 12.203 and 12.228
and the bounds on the quantities P2 and Q2 . In this way we obtain:
L̂ /α = O(δ −5/2 |u|−2 )

/L ∇
L /β = O(δ −3/2 |u|−3 )
/L ∇
/L/dρ = O(δ −1 |u|−4 )
L
/L/dσ = O(δ −1 |u|−4 )
L
L /β = O(|u|−5 )
/L ∇
L̂ /α = O(δ 1/2 |u|−11/2 )
/L ∇ (15.49)
and:
/L D̂α = O(δ −7/2 |u|−1 )

L̂
502
/L Dβ = O(δ −5/2 |u|−2 )
L
LDρ = O(δ −2 |u|−3 )
LDσ = O(δ −2 |u|−3 )
/L Dβ = O(δ −1 |u|−4 )
L
/L D̂α = O(δ 1/2 |u|−11/2 )
L̂ (15.50)
Consider next the case Y = Oj : j = 1, 2, 3. Then each of 15.45 is equal to
/Oj ∇
L̂ /Oj ∇
/α, L /Oj /dσ, L
/Oj /dρ, L
/β, L /Oj ∇
/β and L̂
/Oj ∇ /α (15.51)
respectively, and each of 15.46 is equal to
L̂ /Oj Dβ, Oj Dρ, Oj Dσ, L

/Oj D̂α, L /Oj D̂α
/Oj Dβ and L̂ (15.52)
respectively. In view of Propositions 8.1, 8.2, the first five of each of 15.50, 15.51
are estimated in L2 (Cu ) and the last of each of 15.51, 15.52 are estimated in
L2 (C u ), according to the bound on the quantity Q2 . In this way we obtain:
/α = O(δ −3/2 |u|−2 )

/Oj ∇
L̂
/β = O(δ −1/2 |u|−3 )
/Oj ∇
L
/Oj /dρ = O(|u|−4 )
L
/Oj /dσ = O(|u|−4 )
L
/β = O(δ|u|−5 )
/Oj ∇
L
/Oj ∇
L̂ /α = O(δ 3/2 |u|−11/2 ) (15.53)
and:
/Oj D̂α = O(δ −5/2 |u|−1 )

L̂
/Oj Dβ = O(δ −3/2 |u|−2 )
L
Oj Dρ = O(δ −1 |u|−3 )
Oj Dσ = O(δ −1 |u|−3 )
/Oj Dβ = O(|u|−4 )
L
/Oj D̂α = O(δ 3/2 |u|−11/2 )
L̂ (15.54)
Consider finally the case Y = S. Then X = S, and since in this case

only the case n = 3 occurs, we only have to consider the terms resulting from
/S Ξ( (S) J 1 (R)) and L̂
L /S Θ( (S) J 1 (R)). From Proposition 14.1 we see that these
involve only
L/S ∇/β and L̂/S ∇
/α (15.55)
(the second being the trace-free part relative to the last two indices of L
/S ∇
/α)
and
/S Dβ and L̂
L /S D̂α (15.56)
503
By Lemma 4.1 the first and second of 15.55 are equal to
∇
/L
/S β and ∇
/L̂
/S α (15.57)
plus remainders which can be estimated in L4 (S), therefore also in L2 (Cu ) using
the estimates of Chapter 4 together with the L∞ estimates for the components
of R. The first of 15.57 is estimated through inequality 12.243 and the bounds on
the quantities P2 and Q2 . The second of 15.57 is estimated through inequality
12.193 and the bounds on the quantities P2 and Q2 . In this way we obtain:
L /β = O(δ|u|−5 )
/S ∇
L̂ /α = O(δ 3/2 |u|−11/2 )
/S ∇ (15.58)
The first of 15.56 is equal to:
/S Dβ = uDDβ + uD2 β
L (15.59)
The second term is estimated in L2 (Cu ) according to the bound on the quantity
Q2 . To estimate the first term we appeal to the sixth of the Bianchi identities,
given by Proposition 1.2:
1 n o
Dβ+ Ωtrχβ−Ωχ̂♯ ·β+ωβ = Ω −d /ρ + ∗/dσ − 3ηρ + 3 ∗ ησ + 2χ̂♯ · β (15.60)
2
Applying Dto this identity yields the following expression for DDβ:
1
DDβ = − ΩtrχDβ + Ωχ̂♯ · Dβ − ωDβ
2
1
− D(Ωtrχ)β + D̂(Ωχ̂) · β ♯ − (Dω)β
2 n o
+Ω2 (χ̂, χ̂)β − 2(χ̂ × χ̂) · β ♯
/Dρ + ∗/dDσ − 3ηDρ + 3 ∗ ηDσ

+Ω −d
−3(Dη)ρ + 3 ∗ (Dη)σ
o
+ 2χ̂♯ · Dβ + 2(D̂χ̂) · β ♯
−2Ω2 ∗ χ̂♯ · (d
/σ + 3ησ)
n o
+Ωω −d /ρ + ∗/dσ − 3ηρ + 3 ∗ ησ + 2χ̂♯ · β (15.61)
Consider first the principal terms on the right. These are the terms −d /Dρ +
∗
/dDσ in the second parenthesis. Now Dρ, Dσ are given by the Bianchi iden-
tities 13.51. The principal terms in Dρ and Dσ are the terms −Ωdiv / β and
−Ωcurl/ β respectively. Thus the principal part of the right hand side of 15.61
is:
Ω2 (d / β − ∗/dcurl
/div / β) = O(δ|u|−6 ) (15.62)
according to the bound on the quantity Q2 . Using 15.41, the L∞ bounds for the
components of R and the results of Chapters 3 and 4 we find that the remaining
504
terms on the right in 15.61 are O4 (|u|−5 ), therefore also O(|u|−5 ). We then
conclude that:
DDβ = O(|u|−5 ) (15.63)
/S Dβ = O(|u|−4 )
L (15.64)
The second of 15.56 is equal to:
2
/S D̂α = uD̂ α + uD̂D̂α
L̂ (15.65)
The first term is estimated in L2 (C u ) according to the bound on the quantity

Q2 , while the second term is estimated in L2 (C u ) through inequalities 12.203
and 12.228 and the bounds on the quantities P2 and Q2 . We then obtain:
/S D̂α = O(δ 3/2 |u|−11/2 )

L̂ (15.66)
According to the preceding discussion the first term resulting by applying L /Y

to a given term in the expression of a given component of (X) J 1 (R) consists of
two factors the first of which is O4 (δ r1,1 |u|p1,1 ) and the other is O4 (δ r2,1 |u|p2,1 ).
We set:
r1′ = r1,1 + r2,1 , p′1 = p1,1 + p2,1 (15.67)
The second term resulting by applying L /Y to a term in the expression of
a given component of (X) J 1 (R) consists of two factors the first of which is
O∞ (δ r1,2 |u|p1,2 ) and the second is a sum of terms each of which is either O(δ r |u|p )
or O(δ r |u|p ). We define r2,2 to be the minimal r and p2,2 to be the maximal p
occuring in the terms of the other factor. We set:
r2′ = r1,2 + r2,2 , p′2 = p1,2 + p2,2 (15.68)
We then set:
r′ = min{r1′ , r2′ }, p′ = max{p′1 , p′2 } (15.69)
′ ′
In this way a pair r , p is assigned to the pair of terms resulting by applying
/Y to a given term in the expression of a given component of (X) J 1 (R). Let
L
r1∗ be the minimal r′ and p∗1 be the maximal p′ occuring for a given component
of (X) J 1 (R). Now, each component of L̃Y (X) J 1 (R) is expressed by Proposi-
tion 12.3 as L /Y applied to the corresponding component of (X) J 1 (R) plus a
remainder which is a sum of bilinear terms, the first factor of each term being a
component of (Y ) π̃ and the second factor a component of (X) J 1 (R). For each
term the first factor is O∞ (δ r1,3 |u|p1,3 ) and the second factor is a sum of terms
which may be viewed as being either O(δ r |u|p ) or O(δ r |u|p ) according to the
analysis of Chapter 14. We define r2,3 to be the minimal r and p2,3 to be the
maximal p occuring in the terms of the second factor. The pair r2,3 , p2,3 is the
pair r∗ , p∗ assigned to that component of (X) J 1 (R) in Chapter 14. We set:
r3′ = r1,3 + r2,3 , p′3 = p1,3 + p2,3 (15.70)
505
We then define r2∗ to be the minimal r3′ and p∗2 to be the maximal p′3 occuring
in the terms of the remainder. Finally, we assign to the given component of
L̃Y (X) J 1 (R) the pair r∗ , p∗ where:
r∗ = min{r1∗ , r2∗ }, p∗ = max{p∗1 , p∗2 } (15.71)

It turns out that for all components:
r∗ = r1∗ , p∗ = p∗1 (15.72)
We obtain in this way tables which are similar to the tables 14.28 - 14.31, 14.33,
14.35, 14.36, 14.38 - 14.40, and 14.41 - 14.44, 14.46, 14.48, 14.49, 14.51 - 14.53,
and 14.54, 14.55. In fact, the tables for the case Y = X = L (l = 2) are identical
to the tables for the case X = L (l = 1) of Chapter 14, but with the values of
r′ and r∗ decreased by 1. The tables for the case Y = Oi : i = 1, 2, 3, X = L
(l = 1) are identical to the tables for the case X = L (l = 1) of Chapter 14. The
tables for the case Y = Oj : j = 1, 2, 3, X = Oi : i = 1, 2, 3 (l = 0) are identical
to the tables for the case X = Oi : i = 1, 2, 3 (l = 0) of Chapter 14. The tables
for the case Y = Oi : i = 1, 2, 3, X = S (l = 0) are identical to the tables for the
case X = S (l = 0) of Chapter 14. Finally, the tables for the case Y = X = S
(l = 0) are identical to the tables for the case X = S (l = 0) of Chapter 14.
Each component of the Weyl currents L̃L (L) J 1 (R), L̃Oi (L) J 1 (R) : i =
1, 2, 3, L̃Oj (Oi ) J 1 (R); i, j = 1, 2, 3 and L̃Oi (S) J 1 (R) : i = 1, 2, 3, L̃S (S) J 1 (R)
by the expression for a component of the Weyl current L̃L (L) J 1 (R) in the case
of 14.21, L̃Oi (L) J 1 (R) : i = 1, 2, 3 in the case of 14.22, L̃Oj (Oi ) J 1 (R) : i, j =
1, 2, 3 in the case of 14.23 and L̃Oi (S) J 1 (R) in the case of 14.24, L̃S (S) J 1 (R) in
volving α(L̃Y L̃X R). We then obtain a bound for the contribution of all terms
resulting from the product of a given component of L̃Y (X) J 1 (R) with a given
component of L̃Y L̃X R to the corresponding error integral 15.5 by O(δ e ), where
e is the excess index, defined as in 14.58, provided that the integrability index
s, defined as in 14.59, is negative so that Lemma 13.1 applies. We obtain in this
way tables which are similar to the tables 14.60 - 14.68. In fact, the tables for
the case (Y X) = (LL) (l = 2) are identical to the tables for the case X = L
(l = 1) of Chapter 14. The tables for the case (Y X) = (OL) (l = 1) are iden-
tical to the tables for the case X = L (l = 1) of Chapter 14. The tables for
the case (Y X) = (OO) (l = 0) are identical to the tables for the case X = O
(l = 0) of Chapter 14. Finally, the tables for the cases (Y X) = (OS) (l = 0)
and (Y X) = (SS) (l = 0) are identical to the table for the case X = S (l = 0)
of Chapter 14.
506
We must now analyze in more detail the terms with vanishing excess index
as they give rise to borderline error integrals. They occur only in the cases
(3)
(LL) τc
n = 1 and n = 3. The terms which give borderline contributions to
(3) (3) (3) (3)
(OL) τ c , (OO) τ c (OS) τ c , (SS) τ c
, and are the terms:
−6Ω3 |u|6 trχ(((D̂ (L)

î)ρ − ∗
(D̂ (L)
î)σ), α(L̃L L̃L R))
3 6 (L) ∗(L)
−6Ω |u| trχ(( îDρ − îDσ), α(L̃L L̃L R)) (15.73)
X
−6Ω3 |u|6 trχ (((L̂
/Oi (L)
î)ρ − ∗
(L̂
/Oi (L)
î)σ), α(L̃Oi L̃L R))
i
X
−6Ω3 |u|6 trχ (( (L)
îOi ρ − ∗(L)
îOi σ), α(L̃Oi L̃L R)) (15.74)
i
X
−6Ω3 |u|6 trχ (((L̂
/Oj (Oi )
î)ρ − ∗
(L̂
/Oj (Oi )
î)σ), α(L̃Oj L̃Oi R))
i,j
X
−6Ω3 |u|6 trχ (( (Oi )
îOj ρ − ∗(Oi )
îOj σ), α(L̃Oj L̃Oi R)) (15.75)
i,j
and:
X
−6Ω3 |u|6 trχ (((L̂
/Oi (S)
î)ρ − ∗
(L̂
/Oi (S)
î)σ), α(L̃Oi L̃S R))
i
X
−6Ω3 |u|6 trχ (( (S)
îOi ρ − ∗(S)
îOi σ), α(L̃Oi L̃S R)) (15.76)
i
−6Ω3 |u|6 trχ(((L̂

/S (S)
î)ρ − ∗
(L̂
/S (S)
î)σ), α(L̃S L̃S R))
−6Ω3 |u|6 trχ(( (S)
îSρ − ∗(S)
îSσ), α(L̃S L̃S R))
3 6 (S) ∗(S)
−6Ω |u| (Strχ + 2trχ)(( îρ − îσ), α(L̃S L̃S R)) (15.77)
respectively. The last results from the term 2 (S) νΘ(J) in the expression for
Θ(L̃S J) given by Proposition 12.3 with S in the role of Y (see 12.37). It is the
only borderline term contributed by the remainders in Proposition 12.3.
(L)
kD̂ îkL4 (Su,u ) ≤ Cδ −3/2 |u|−1/2 (R∞ ′
0 (α) + O(δ)) : ∀(u, u) ∈ Dc∗ (15.78)
implied by the first of 9.2, the facts that:
ktrχρkL4 (Su,u ) ≤ C|u|−7/2 R40 (ρ) : ∀(u, u) ∈ Dc′ ∗

ktrχσkL4 (Su,u ) ≤ C|u|−7/2 R40 (σ) : ∀(u, u) ∈ Dc′ ∗ (15.79)
507
as well as 15.12, and following the proof of Case 5 of Lemma 13.1, we deduce
that the contribution of the first of the terms 15.73 to the error integral 15.5
(3)
CR∞ 4 4
0 (α)(R0 (ρ) + R0 (σ))( F 2 )
1/2
+ O(δ) (15.80)
Also, using the precise bound 14.72, the facts that:
ktrχDρkL2 (Cu ) ≤ Cδ −1/2 |u|−3 R0 (Dρ) : ∀u ∈ [u0 , c∗ )

ktrχDσkL2 (Cu ) ≤ Cδ −1/2 |u|−3 R0 (Dσ) : ∀u ∈ [u0 , c∗ ) (15.81)
as well as 15.12, and following the proof of Case 2 of lemma 13.1, we deduce
that the contribution of the second of the terms 15.73 to the same error integral
is bounded by:
(3)
CR∞
0 (α)(R0 (Dρ) + R0 (Dσ))( F 2 )
1/2
(15.82)
By 14.82, and by 10.48 and the third of 12.262 which together imply:
(2)
R40 (ρ), R40 (σ) ≤ C( E 1 )1/2 + O(δ) (15.83)
15.79 is in turn bounded by:

(0) (2) (3)
C( E 2 )1/2 ( E 1 )1/2 ( F 2 )1/2 + O(δ 1/2 ) (15.84)
Also, by 14.82 and the third of 12.262, 15.81 is in turn bounded by 15.84 as
well.
The first of the bounds 9.1 together with the first of the bounds 8.32 and
Propositions 8.1 and 8.2 implies:
(L)
kL̂
/Oi îkL4 (Su,u ) ≤ δ −1/2 |u|−1/2 (R
/41 (α) + R∞
0 (α) + O(δ
1/2
)) : ∀(u, u) ∈ Dc′ ∗
(15.85)
Using this bound, 15.79, as well as 15.16, and following the proof of Case 5 of
Lemma 13.1, we deduce that the contribution of the first of the terms 15.74 to
the error integral 15.5 with (Y X) = (OL) (l = 1) and n = 3 is bounded by:
(3)
/41 (α) + R∞
C(R 4 4
0 (α))(R0 (ρ) + R0 (σ))( F 2 )
1/2
+ O(δ 1/2 ) (15.86)
Also, using the precise bound 14.72, the facts that by Proposition 8.1:
ktrχOi ρkL2 (Cu ) ≤ Cδ 1/2 |u|−3 R

/1 (ρ) : ∀u ∈ [u0 , c∗ )
ktrχOi σkL2 (Cu ) ≤ Cδ 1/2 |u|−3 R
/1 (σ) : ∀u ∈ [u0 , c∗ ) (15.87)
is bounded by:
(3)
CR∞
0 (α)(R /1 (σ))( F 2 )1/2
/1 (ρ) + R (15.88)
508
By the first of 10.67 and the first of 10.73 together with the first of 12.263:
(0)
/41 (α) + R∞
R 0 (α) ≤ C( E 2 )
1/2
+ O(δ 1/2 ) (15.89)
Taking also into account 15.83, we conclude that 15.86 is in turn bounded by:
(0) (2) (3)
C( E 2 )1/2 ( E 1 )1/2 ( F 2 )1/2 + O(δ 1/2 ) (15.90)
Also, by 14.82 and the third of 12.262, 15.88 is in turn bounded by 15.90 as
well.
(Oi )
kL̂
/Oj îkL4 (Su,u ) ≤ Cδ 1/2 |u|−1/2 (R
/41 (β) + R∞
0 (β) + O(δ
1/2
)) : ∀(u, u) ∈ Dc′ ∗
(15.91)
the error integral 15.5 with (Y X) = (OO) (l = 0) and n = 3 is bounded by:
(3)
/41 (β) + R∞
C(R 4 4
0 (β))(R0 (ρ) + R0 (σ))( F 2 )
1/2
+ O(δ 1/2 ) (15.92)
Also, using the more precice bound 14.76, 15.87 as well as 15.17, and following
the proof of Case 2 of lemma 13.1, we deduce that the contribution of the second
of the terms 15.75 to the same error integral is bounded by:
(3)
/41 (β) + R∞
C(R 0 (β))(R /1 (σ))( F 2 )1/2 + O(δ 1/2 )
/1 (ρ) + R (15.93)
By 14.84, 15.83 and the third of 12.262, 15.92 and 15.93 are both in turn bounded
by:
(1) (2) (3)
C( E 2 )1/2 ( E 1 )1/2 ( F 2 )1/2 + O(δ 1/2 ) (15.94)
(S)
kL̂
/Oi îkL4 (Su,u ) ≤ Cδ 1/2 (D
/41 (χ̂) + D0∞ (χ̂) + R
/41 (α) + R∞
0 (α) + O(δ
1/2
))
: ∀(u, u) ∈ Dc′ ∗ (15.95)
the error integral 15.5 with (Y X) = (OS) (l = 0) and n = 3 is bounded by:
(3)
/41 (χ̂)+ D0∞ (χ̂)+ R
C(D /41 (α)+ R∞ 4 4
0 (α))(R0 (ρ)+ R0 (σ))( F 2 )
1/2
+ O(δ 1/2 ) (15.96)
Also, using the more precice bound 14.79, 15.87 as well as 15.18, and following
the proof of Case 2 of lemma 13.1, we deduce that the contribution of the second
of the terms 15.76 to the same error integral is bounded by:
(3)
C(D0∞ (χ̂) + R∞
0 (α))(R /1 (σ))( F 2 )1/2 + O(δ)
/1 (ρ) + R (15.97)
509
By 15.89, 15.83 and the third of 12.262, 15.96 and 15.97 are in turn bounded
by:
(0) (2) (3)
/41 (χ̂) + D0∞ (χ̂) + ( E 2 )1/2 )( E 1 )1/2 ( F 2 )1/2 + O(δ 1/2 )
C(D (15.98)
and:
(0) (2) (3)
C(D0∞ (χ̂) + ( E 2 )1/2 )( E 1 )1/2 ( F 2 )1/2 + O(δ 1/2 ) (15.99)
respectively.
The first of the bounds 9.13, 9.14 together imply:
(S)
kL̂
/S îkL4 (Su,u ) ≤ Cδ 1/2 |u|−1/2 (D0∞ (χ̂) + R∞
0 (α) + O(δ)) (15.100)
the error integral 15.5 with (Y X) = (SS) (l = 0) and n = 3 is bounded by:
(3)
C(D0∞ (χ̂) + R∞ 4 4
0 (α))(R0 (ρ) + R0 (σ))( F 2 )
1/2
+ O(δ 1/2 ) (15.101)
Also, using the more precise bound 14.79, the bounds 13.54 which imply:
(2)
ktrχSρkL2 (Cu ) ≤ Cδ 1/2 |u|−3 (( E 1 )1/2 + O(δ)) : ∀u ∈ [u0 , c∗ )
(2)
ktrχSρkL2 (Cu ) ≤ Cδ 1/2 |u|−3 (( E 1 )1/2 + O(δ)) : ∀u ∈ [u0 , c∗ )
(15.102)
is bounded by:
(2) (3)
C(D0∞ (χ̂) + R∞
0 (α))( E 1 )
1/2
( F 2 )1/2 + O(δ) (15.103)
By 14.82 and 15.83, 15.101 and 15.103 are both in turn bounded by:
(0) (2) (3)
C(D0∞ (χ̂) + ( E 2 )1/2 )( E 1 )1/2 ( F 2 )1/2 + O(δ 1/2 ) (15.104)
Finally, we consider the contribution of the third of the terms 15.77. Using
equations 3.9 and 4.3 we deduce the following expression for Strχ + trχ:
Strχ + trχ = (uω − uω − ε)trχ

−uΩ|χ̂|2 + uΩ 2div
/ η + 2|η|2 − (χ̂, χ̂) + 2ρ (15.105)

where ε is the function defined by 14.164. Using the results of Chapters 3 and
4 we obtain:
Strχ + trχ = O4 (δ|u|−2 ) (15.106)
510
It follows that the contribution comming from the factor Strχ + trχ is not
borderline, being bounded by O(δ). We are left with the contribution of
− 6Ω3 |u|6 trχ(( (S)

îρ − ∗(S)
îσ), α(L̃S L̃S R)) (15.107)
which is similar to that of 14.71, but with α(L̃S L̃S R) in the role of α(L̃S R).
Using 5.20 in place of 14.80 we conclude that the contribution in question is
bounded by:

(0) (2) (3)
∞ 1/2
C D0 (χ̂) + ( E 2 ) ( E 0 )1/2 ( F 2 )1/2 + O(δ 1/2 ) (15.108)
(compare with 14.86).

(1) (1) (1)
The terms which give borderline contributions to (LL) τ c , (OL) τ c , (OO) τ c
are the terms:
n o
2Ω3 |u|2 trχ (D̂ (L) î, α)ρ(L̃L L̃L R) + D̂ (L) î ∧ ασ(L̃L L̃L R)
n o
+2Ω3 |u|2 trχ ( (L) î, D̂α)ρ(L̃L L̃L R) + (L) î ∧ D̂ασ(L̃L L̃L R)
(15.109)
Xn o
2Ω3 |u|2 trχ (L̂
/Oi (L)
î, α)ρ(L̃Oi L̃L R) + L̂
/Oi (L)
î ∧ ασ(L̃Oi L̃L R)
i
Xn o
3 2 (L) (L)
+2Ω |u| trχ /Oi α)ρ(L̃Oi L̃L R) +
î, L̂ /Oi ασ(L̃Oi L̃L R)
î ∧ L̂
i
(15.110)
Xn o
2Ω3 |u|2 trχ (L̂
/Oj (Oi )
î, α)ρ(L̃Oj L̃Oi R) + L̂
/Oj (Oi )
î ∧ ασ(L̃Oj L̃Oi R)
i,j
Xn o
3 2 Oi ) (Oi )
+2Ω |u| trχ ( /Oj α)ρ(L̃Oj L̃Oi R) +
î, L̂ /Oj ασ(L̃Oj L̃Oi R)
î ∧ L̂
i,j
(15.111)
respectively.
ktrχαkL4 (Su,u ) ≤ Cδ −3/2 |u|−3/2 R40 (α) : ∀(u, u) ∈ Dc′ ∗ (15.112)
as well as 15.29, and following the proof of Case 4 of Lemma 13.1, we deduce
that the contribution of the first of the terms 15.109 to the error integral 15.5
(2)
CR∞ 4
0 (α)R0 (α)( E 2 )
1/2
+ O(δ) (15.113)
511
Also, using the precise bound 14.72, the fact that:
ktrχDαkL2 (Cu ) ≤ Cδ −2 |u|−1 R0 (Dα) : ∀u ∈ [u0 , c∗ ) (15.114)
as well as 15.29, and following the proof of Case 1 of lemma 13.1, we deduce that
the contribution of the second of the terms 15.109 to the same error integral is
bounded by:
(2)
CR∞
0 (α)R0 (Dα)( E 2 )
1/2
(15.115)
By 14.82, and by 10.44 and the first of 12.262 which together imply:
(0)
R40 (α) ≤ C( E 1 )1/2 + O(δ) (15.116)
15.113 is in turn bounded by:

(0) (0) (2)
C( E 2 )1/2 ( E 1 )1/2 ( E 2 )1/2 + O(δ 1/2 ) (15.117)
Also, by 14.82 and the first of 12.262, 15.115 is in turn bounded by 15.117 as
well.
Using the bound 15.85, 15.112, as well as 15.32, and following the proof
of Case 4 of Lemma 13.1, we deduce that the contribution of the first of the
terms 15.110 to the error integral 15.5 with (Y X) = (OL) (l = 1) and n = 1 is
bounded by:
(2)
/41 (α) + R∞
C(R 4
0 (α))R0 (α)( E 2 )
1/2
+ O(δ 1/2 ) (15.118)
Also, using the bound 14.72, the facts that by Propositions 8.1 and 8.2:
/Oi αkL2 (Cu ) ≤ Cδ −1 |u|−1 max{R

ktrχL̂ /1 (α), R0 (α)} : ∀u ∈ [u0 , c∗ ) (15.119)
as well as 15.32, and following the proof of Case 1 of lemma 13.1, we deduce that
the contribution of the second of the terms 15.110 to the same error integral is
bounded by:
(2)
CR∞ /1 (α), R0 (α)}( E 2 )1/2
0 (α) max{R (15.120)
By 15.89 and 15.116, 15.118 is in turn bounded by:
(0) (0) (2)
C( E 2 )1/2 ( E 1 )1/2 ( E 2 )1/2 + O(δ 1/2 ) (15.121)
Also, by 14.82 and the first of 12.262, 15.120 is in turn bounded by 15.121 as
well.
Using 15.91, 15.112, as well as 15.33, and following the proof of Case 4 of
the error integral 15.5 with (Y X) = (OO) (l = 0) and n = 1 is bounded by:
(2)
/41 (β) + R∞
C(R 4
0 (β))R0 (α)( E 2 )
1/2
+ O(δ 1/2 ) (15.122)
512
Also, using the bound 14.76, 15.119 as well as 15.33, and following the proof
of Case 1 of lemma 13.1, we deduce that the contribution of the second of the
terms 15.111 to the same error integral is bounded by:
(2)
/41 (β) + R∞
C(R /1 (α), R0 (α)}( E 2 )1/2 + O(δ 1/2 )
0 (β)) max{R (15.123)
By 14.84, 15.116 and the first of 12.262, 15.122 and 15.123 are both in turn
bounded by:
(1) (0) (2)
C( E 2 )1/2 ( E 1 )1/2 ( E 2 )1/2 + O(δ 1/2 ) (15.124)
We now turn to consider the contribution to the 2nd order Weyl current
error integrals of the term
L̃Y (X) J 2 (R)
We consider the expressions for the components of (X) J 2 (R) given by 14.98 with
the fundamental Weyl field R in the role of the Weyl field W . The expression for
each component of (X) J 2 (R) is a sum of terms with two factors, one factor being
a component of (X) p and the other a component of R. Now, the components of
L̃Y (X) J 2 (R) are given by Proposition 12.3 in terms of the L
/Y derivatives of the
components of (X) J 2 (R). The remainders are estimated using the L∞ bounds
for the components of (Y ) π̃ and the estimates for the components of (X) J 2 (R)
of Chapter 14. Now if L /Y is applied to a term in the expression of a given
component of (X) J 2 (R) a sum of two terms will result, one of which will have
as a first factor the L/Y derivative of the corresponding component of (X) p and
the same second factor as the original term, and the other term will have the
component of (X) p as a first factor and L /Y applied to the second factor of the
original term, plus possibly a bilinear expression in (Y ) π
/ and the second factor
of the original term, as the other factor. The bilinear expression results from
applying L /Y to the coefficients of the expression constituting the given original
term, which may involve g/ and /ǫ . The first of the two resulting terms is to be
estimated by placing the first factor in L2 (Cu ) using the estimates to be given
below, and the second factor in L∞ using the L∞ bounds for the components
of R. The second of the two resulting terms is to be estimated by placing the
first factor in L4 (S) using the estimates 14.102 - 14.104, and the second factor
in L4 (S) using the estimates 15.37, 15.39, and 15.43.
The estimates 9.97 - 9.101, 9.107 - 9.112, 9.172 - 9.174, together with the
estimates 9.1 - 9.3, 9.12 - 9.14, 9.89 - 9.91 and 8.32, 8.34, 8.144, the results
of Chapters 3 and 4 and Propositions 8.1 and 8.2, imply through the formulas
14.99 - 14.101 the following L2 (Cu ) estimates for L/Y applied to the components
of (X) p:
(L)
L p4 = O(δ −2 |u|−2 )
(L)
L p3 = O(δ −1 |u|−2 )
(L)
L
/L /p = O(δ −3/2 |u|−2 ) (15.125)
(L)
Oi p4 = O(δ −1 |u|−2 )
513
(L)
Oi p3 = O(|u|−2 )
(L)
L
/Oi /p = O(δ −1/2 |u|−2 ) (15.126)
(Oi )
Oj p4 = O(|u|−2 )
(Oi )
Oj p3 = O(δ|u|−3 )
(Oi )
L
/Oj /p = O(δ 1/2 |u|−2 ) (15.127)
(S)
Oi p4 = O(|u|−2 )
(S)
Oi p3 = O(δ|u|−3 )
(S)
L
/Oi /p = O(δ 1/2 |u|−2 ) (15.128)
(S)
S p4 = O(|u|−2 )
(S)
S p3 = O(δ|u|−3 )
(S)
L
/S /p = O(δ 1/2 |u|−2 ) (15.129)
According to the preceding discussion the first term resulting by applying L /Y

to a given term in the expression of a given component of (X) J 2 (R) consists of
two factors the first of which is O(δ r1,1 |u|p1,1 ) and the other is O∞ (δ r2,1 |u|p2,1 ).
We set:
r1′ = r1,1 + r2,1 , p′1 = p1,1 + p2,1 (15.130)
O4 (δ r1,2 |u|p1,2 ) and the second is O4 (δ r2,2 |u|p2,2 ). We set:
r2′ = r1,2 + r2,2 , p′2 = p1,2 + p2,2 (15.131)
We then set:
r′ = min{r1′ , r2′ }, p′ = max{p′1 , p′2 } (15.132)
In this way a pair r′ , p′ is assigned to the pair of terms resulting by applying
/Y to a given term in the expression of a given component of (X) J 2 (R). Let r1∗
L
be the minimal r′ and p∗1 be the maximal p′ occuring for a given component of
(X) 2
J (R). Now, each component of L̃Y (X) J 2 (R) is expressed by Proposition
12.3 as L /Y applied to the corresponding component of (X) J 2 (R) plus a remain-
der which is a sum of bilinear terms, the first factor of each term being a com-
ponent of (Y ) π̃ and the second factor a component of (X) J 2 (R). For each term
the first factor is O∞ (δ r1,3 |u|p1,3 ) and the second factor is O(δ r2,3 |u|p2,3 ) accord-
ing to the analysis of Chapter 14. The pair r2,3 , p2,3 is the pair r∗ , p∗ assigned to
that component of (X) J 2 (R) in Chapter 14. For, if {ξ1 , ξ2 } is an S tensorfield
which is a bilinear expression in the S tensorfields ξ1 and ξ2 with coefficients
depending only on g/ and /ǫ and ξ1 = O4 (δ r1 |u|p1 ) while ξ2 = O4 (δ r2 |u|p2 ), then
{ξ1 , ξ2 } = O(δ r1 +r2 |u|p1 +p2 ). We set:
r3′ = r1,3 + r2,3 , p′3 = p1,3 + p2,3 (15.133)
514
r∗ = min{r1∗ , r2∗ }, p∗ = max{p∗1 , p∗2 } (15.134)
r∗ = r1∗ , p∗ = p∗1 (15.135)
We obtain in this way tables which are similar to the tables 14.107 - 14.116,
14.117 - 14.126, and 14.127 - 14.128. In fact, the tables for the case Y = X = L
(l = 2) are identical to the tables for the case X = L (l = 1) of Chapter 14, but
with the values of r′ and r∗ decreased by 1. The tables for the case Y = Oi :
i = 1, 2, 3, X = L (l = 1) are identical to the tables for the case X = L (l = 1)
of Chapter 14. The tables for the case Y = Oj : j = 1, 2, 3, X = Oi : i = 1, 2, 3
(l = 0) are identical to the tables for the case X = Oi : i = 1, 2, 3 (l = 0) of
identical to the tables for the case X = S (l = 0) of Chapter 14. Finally, the
tables for the case Y = X = S (l = 0) are identical to the tables for the case
s, defined as in 14.59, is negative so that Lemma 13.1 applies. We obtain in
this way tables which are similar to the tables 14.132 - 14.140. In fact, the
tables for the case (Y X) = (LL) (l = 2) are identical to the tables for the case
X = L (l = 1) of Chapter 14. The tables for the case (Y X) = (OL) (l = 1) are
identical to the tables for the case X = L (l = 1) of Chapter 14. The tables for
515
of Chapter 14. All terms have positive excess index so there are no borderline
terms.
We finally consider the contribution to the 2nd order Weyl current error
integrals of the term
L̃Y (X) J 3 (R)
We consider the expressions for the components of (X) J 3 (R) given by Lemma
14.2 with the fundamental Weyl field R in the role of the Weyl field W . The
expression for each component of (X) J 3 (R) is a sum of terms with two factors,
one factor being a component of (X) q, which has the algebraic properties of
a Weyl current, and the other a component of R. Now, the components of
L̃Y (X) J 3 (R) are given by Proposition 12.3 in terms of the L /Y derivatives of
the components of (X) J 3 (R). The remainders are estimated using the L∞
bounds for the components of (Y ) π̃ and the estimates for the components of
(X) 3
J (R) of Chapter 14. Now if L /Y is applied to a term in the expression of
a given component of (X) J 3 (R) a sum of two terms will result, one of which
will have as a first factor the L /Y derivative of the corresponding component
of (X) q and the same second factor as the original term, and the other term
will have the component of (X) q as a first factor and L /Y applied to the second
factor of the original term, plus possibly a bilinear expression in (Y ) π / and the
second factor of the original term, as the other factor. The bilinear expression
results from applying L /Y to the coefficients of the expression constituting the
given original term, which may involve g/ and /ǫ . The first of the two resulting
terms is to be estimated by placing the first factor in L2 (Cu ) using the estimates
to be given below, and the second factor in L∞ using the L∞ bounds for the
components of R. The second of the two resulting terms is to be estimated by
placing the first factor in L4 (S) using the estimates 14.102 - 14.104, and the
second factor in L4 (S) using the estimates 15.37, 15.39, and 15.43.
In deducing the L2 (Cu ) estimates for L /Y applied to the componets of (X) q
to be given below, we must take into account the crucial cancellation involved
in
1
D̂ (X) î − Ωtrχ (X) î (15.136)
2
the first term on the right in 14.144, as discussed in Chapter 14.
In the case X = L, 15.136 is given by equation 14.153. In view of equation
1.82 the results of Chapters 3 and 4 imply:

(L) 1 (L)
L̂
/L D̂ î − Ωtrχ î = O(δ −1/2 |u|−3 ) (15.137)
2
Also, Proposition 7.1 together with Propositions 8.1 and 8.2 and the results of
Chapters 3 and 4 implies:

(L) 1 (L)
/Oi D̂
L̂ î − Ωtrχ î = O(δ 1/2 |u|−3 ) (15.138)
2
In the case X = S, 15.136 is given by equation 14.154. Propositions 8.1 and
516
8.2 together with the results of Chapters 3 and 4 and 15.138 imply:

(S) 1 (S)
/Oi D̂
L̂ î − Ωtrχ î = O(δ 3/2 |u|−3 ) (15.139)
2
Taking into account the fact that by equation 1.173 and the L4 (S) estimate for
/ 2 ω of Proposition 6.2 we have:
∇
/Dη = O4 (δ 1/2 |u|−4 )

∇ (15.140)
Noting moreover that the function

1
ε= Ωutrχ − 1
2
(see 14.164) satifies:
1
Sε = (1 + ε)S log Ω + Ωu(Strχ + trχ) = O4 (δ|u|−2 ) (15.141)
2
by 15.104, and taking also into account the last of 15.43, we deduce:

1
/S D̂ (S) î − Ωtrχ (S) î = O(δ 3/2 |u|−3 )
L̂ (15.142)
2
In the case X = Oi , 15.136 is given by equation 14.166. In view of Proposi-

tion 9.1 and the fact that
Oj ε = O∞ (δ|u|−2 ) (15.143)
we will have shown that:

(Oi ) 1 (Oi )
/Oj D̂
L̂ î − Ωtrχ î = O(δ|u|−2 ) (15.144)
2
once we establish the following Lemma.
Lemma 15.1 We have:

/Oj ϑi = O(δ|u|−2 )
L
Proof: We first consider the propagation equation 14.175 along Cu0 . Since
/Oj to this equation we obtain, in view of
[L, Oj ] = 0 along Cu0 , applying L
Lemma 1.3,
/Oj ϑi − ΩtrχL
DL /Oj ϕi + Oj (Ωtrχ)ϑi : along Cu0
/Oj ϑi = L (15.145)
Now, ϕ is given by 14.176, with κi given by 14.75. In applying L /Oj to 14.176

and 14.173 terms of the form L /Oi ξ result, where ξ is a symmetric trace-free
/Oj L
2-covariant S tensorfield, one of
1
Ωχ̂, Ωχ̂, trχχ̂ + g/(χ̂, χ̂) and ∇ˆ + η ⊗η
/⊗η ˆ
2
517
/Oi ξ is to be estimated in L2 (Cu ). We have:
/Oj L
We shall presently show how L
/Oi ξ = Oi · ∇
L /ξ + ∇
/Oi · ξ + ξ · ∇
/Oi (15.146)
where, in terms of components in an arbitrary local frame field for Su,u0 ,
(∇ /A OiC ξCB ,
/Oi · ξ)AB = ∇ (ξ · ∇ /B OiC
/Oi ) = ξAC ∇
It follows that:
/Oj L
L /Oj Oi · ∇
/Oi ξ = L /Oj ∇
/ξ + Oi · L /Oj ∇
/ξ + (L /Oj ∇
/Oi ) · ξ + ξ · (L /Oi )
+∇ /Oj ξ · ∇
/Oj ξ + L
/Oi · L /Oi (15.147)
/Oj Oi = [Oj , Oi ] = ǫjik Ok the first term on the right is

In view of the fact that L
ǫjik Ok · ∇
/ξ. Also, by Lemma 9.1 in the case p = 0, q = 1:
(Oj ) (Oj )
/Oj ∇
L /Oi = ∇/Oj Oi +
/L π
/1 · Oi = ǫjik ∇
/Ok + π
/1 · Oi (15.148)
By Propositions 8.1 and 9.1 we have:

(Oj )
k /1 · Oi kL4 (Su,u ) ≤ O(δ 1/2 |u|−1/2 )
π (15.149)
Taking also into account Proposition 8.2, it follows that:
/Oi ξkL2 (Su,u ) ≤ C(|u|2 k∇

/Oj L
kL / 2 ξkL2 (Su,u ) + |u|k∇
/ξkL2(Su,u ) + kξkL2 (Su,u ) )
+O(δ 1/2 |u|−1/2 )kξkL4 (Su,u ) (15.150)
hence:
/Oi ξkL2 (Cu ) ≤ C(|u|2 k∇

/Oj L
kL / 2 ξkL2 (Cu ) + |u|k∇
/ξkL2(Cu ) + kξkL2 (Cu ) )
+O(δ|u|−1/2 ) sup kξkL4 (Su,u ) (15.151)
u
Using Propositions 7.3 and 6.1, 6.2 we then obtain:
/Oj κi = O(|u|−3 )
L (15.152)
and:
/Oj ϕ = O(|u|−2 )
L (15.153)
We now apply Lemma 4.6, taking p = 2, to the propagation equation 15.145
along Cu0 , noting that the L /Oj ϑi vanish along C 0 . Here r = 2, ν = 2 and γ = 0.
We obtain:
Z u
/Oj ϑi kL2 (Su,u0 ) ≤ C
kL /Oj ϕ + Oj (Ωtrχ)ϑi kL2 (Su′ ,u ) du′
kL
0
0
1/2
≤ Cδ /Oj ϕ + Oj (Ωtrχ)ϑi kL2 (Cu0 )
kL (15.154)
Substituting 15.153 and the estimate 14.179 yields:
/Oj ϑi kL2 (Su,u0 ) ≤ O(δ|u0 |−1 )

kL (15.155)
518
We then consider the propagation equation 14.183 along the C u . Since
[L, Oj ] = 0 on M ′ applying L/Oj to this equation we obtain, in view of Lemma
1.3,
DL /Oj ϑi − ΩtrχL/Oj ϑi = L/Oj ϕi + Oj (Ωtrχ)ϑi (15.156)
We apply Lemma 4.7 to this equation taking p = 2. Here r = 2, ν = 2, γ = 0
and we obtain:
|u|−1 kL
/Oj ϑi kL2 (Su,u ) ≤ C|u0 |−1 kL/Oj ϑi kL2 (Su,u0 ) (15.157)
Z u
+C /Oj ϕi + Oj (Ωtrχ)ϑi kL2 (Su,u′ ) du′
|u′ |−1 kL
u0
Now the symmetric 2covariant S tensorfields ϕi are given by 14.184. We esti-

mate the contribution of the term −2uΩ2 L /Oj L/Oi α in L/Oj ϕ to the integral on
the right in 15.157 by:
Z u Z u 1/2
′ ′ −6 ′
C /Oi αkL2 (Su,u′ ) du ≤ C
/Oj L
kL |u | du k|u|3 L /Oi αkL2 (C u )
/Oj L
u0 u0
3/2
≤ O(δ |u|−5/2 ) (15.158)
The contributions of the remaining terms are estimated using 15.150 and Propo-
sition 6.2. Substituting also the estimate 14.187 we obtain a bound for the in-
tegral on the right in 15.157 by O(δ|u|−2 ). In view also of the estimate 15.155,
we then conclude that:
/Oj ϑi kL2 (Su,u ) ≤ O(δ|u|−1 )

kL (15.159)
from which the lemma follows.
Let us now go back to expressions 14.141 - 14.150. Consider first the case
X = L. Using the estimates 9.97 - 9.101 and 15.137, 15.138 and 15.125, 15.126,
as well as the estimates 9.1 - 9.3, 14.154, 14.102, 8.32, Propositions 8.1, 8.2 and
the results of Chapters 3 and 4, we deduce:
(L)
L
/L Ξ( q = O(δ −1/2 |u|−3 )
(L)
L̂
/L Θ( q) = O(δ −5/2 |u|−1 )
(L)
L̂
/L Θ( q) = O(δ −1/2 |u|−3 )
(L)
LΛ( q) = O(δ −2 |u|−2 )
(L)
LΛ( q) = O(δ −1 |u|−2 )
(L)
LK( q) = O(δ −1 |u|−3 )
(L)
L
/L I( q) = O(δ −3/2 |u|−2 )
(L)
L
/L I( q) = O(δ −3/2 |u|−2 ) (15.160)
and:
(L)
/Oi Ξ(
L q) = O(δ 1/2 |u|−3 )
519
(L)
/Oi Θ(
L̂ q) = O(δ −3/2 |u|−1 )
(L)
/Oi Θ(
L̂ q) = O(δ 1/2 |u|−3 )
(L)
Oi Λ( q) = O(δ −1 |u|−2 )
Oi Λ( (L) q) = O(|u|−2 )
Oi K( (L) q) = O(|u|−3 )
(L)
/Oi I(
L q) = O(δ −1/2 |u|−2 )
(L)
/Oi I(
L q) = O(δ −1/2 |u|−2 ) (15.161)
(recall from 14.188 that Ξ( (L) q), K( (L) q) vanish).

Consider next the case X = Oi : i = 1, 2, 3. Using the estimates 9.172 - 9.174
and 15.144 and 15.127 as well as the estimates 9.89 - 9.91, 14.168, 14.103, 8.144,
Propositions 8.1, 8.2 and the results of Chapters 3 and 4, we deduce:
(Oi )
/Oj Ξ(
L q) = O(δ −1/2 |u|−1 )
(Oi )
L̂
/Oj Θ( q) = O(δ −1/2 |u|−1 )
(Oi )
/Oj Θ(
L̂ q) = O(δ|u|−3 )
(Oi )
Oj Λ( q) = O(|u|−2 )
(Oi )
Oj Λ( q) = O(δ|u|−3 )
(Oi )
Oj K( q) = O(|u|−2 )
(Oi )
Oj K( q) = O(δ|u|−3 )
(Oi )
/Oj I(
L q) = O(δ 1/2 |u|−2 )
(Oi )
/Oj I(
L q) = O(δ 1/2 |u|−2 ) (15.162)
(recall from 14.189 that Ξ( (Oi ) q) vanishes).

Consider finally the case X = S. Using the estimates 9.107 - 9.112 and 15.139,
15.142 and 15.128, 15.129 as well as the estimates 9.12 - 9.14, 14.159, 14.104,
8.34, Propositions 8.1, 8.2 and the results of Chapters 3 and 4, we deduce:
(S)
/Oi Ξ(
L q) = O(δ −1/2 |u|−1 )
(S)
/Oi Ξ(
L q) = O(δ 3/2 |u|−3 )
(S)
/Oi Θ(
L̂ q) = O(δ −1/2 |u|−1 )
(S)
/Oi Θ(
L̂ q) = O(δ 3/2 |u|−3 )
(S)
Oi Λ( q) = O(|u|−2 )
(S)
Oi Λ( q) = O(δ|u|−3 )
(S)
Oi K( q) = O(|u|−2 )
(S)
Oi K( q) = O(δ|u|−3 )
(S)
/Oi I(
L q) = O(δ 1/2 |u|−2 )
(S)
/Oi I(
L q) = O(δ 1/2 |u|−2 ) (15.163)
520
and:
(S)
L
/S Ξ( q) = O(δ −1/2 |u|−1 )
(S)
L
/S Ξ( q) = O(δ 3/2 |u|−3 )
(S)
L̂
/S Θ( q) = O(δ −1/2 |u|−1 )
(S)
L̂
/S Θ( q) = O(δ 3/2 |u|−3 )
(S)
SΛ( q) = O(|u|−2 )
SΛ( (S) q) = O(δ|u|−3 )
SK( (S) q) = O(|u|−2 )
SK( (S) q) = O(δ|u|−3 )
/S I( (S) q) = O(δ 1/2 |u|−2 )
L
(S)
L
/S I( q) = O(δ 1/2 |u|−2 ) (15.164)
According to the discussion of the paragraph preceding 15.136 the first term
resulting by applying L /Y to a given term in the expression of a given component
of (X) J 3 (R) consists of two factors the first of which is O(δ r1,1 |u|p1,1 ) and the
other is O∞ (δ r2,1 |u|p2,1 ). We set:
r1′ = r1,1 + r2,1 , p′1 = p1,1 + p2,1 (15.165)
O4 (δ r1,2 |u|p1,2 ) and the second is O4 (δ r2,2 |u|p2,2 ). We set:
r2′ = r1,2 + r2,2 , p′2 = p1,2 + p2,2 (15.166)
We then set:
r′ = min{r1′ , r2′ }, p′ = max{p′1 , p′2 } (15.167)
′ ′
In this way a pair r , p is assigned to the pair of terms resulting by applying
/Y to a given term in the expression of a given component of (X) J 3 (R). Let
L
r1∗ be the minimal r′ and p∗1 be the maximal p′ occuring for a given component
of (X) J 3 (R). Now, each component of L̃Y (X) J 3 (R) is expressed by Proposi-
tion 12.3 as L /Y applied to the corresponding component of (X) J 3 (R) plus a
remainder which is a sum of bilinear terms, the first factor of each term being a
component of (Y ) π̃ and the second factor a component of (X) J 3 (R). For each
term the first factor is O∞ (δ r1,3 |u|p1,3 ) and the second factor is O(δ r2,3 |u|p2,3 )
according to the analysis of Chapter 14. The pair r2,3 , p2,3 is the pair r∗ , p∗
assigned to that component of (X) J 3 (R) in Chapter 14. We set:
r3′ = r1,3 + r2,3 , p′3 = p1,3 + p2,3 (15.168)
r∗ = min{r1∗ , r2∗ }, p∗ = max{p∗1 , p∗2 } (15.169)
521
r∗ = r1∗ , p∗ = p∗1 (15.170)
We obtain in this way tables which are similar to the tables 14.192 - 14.201,
14.202 - 14.211, and 14.212 - 14.213. In fact, the tables for the case Y = X = L
(l = 2) are identical to the tables for the case X = L (l = 1) of Chapter 14, but
with the values of r′ and r∗ decreased by 1. The tables for the case Y = Oi :
i = 1, 2, 3, X = L (l = 1) are identical to the tables for the case X = L (l = 1)
of Chapter 14. The tables for the case Y = Oj : j = 1, 2, 3, X = Oi : i = 1, 2, 3
(l = 0) are identical to the tables for the case X = Oi : i = 1, 2, 3 (l = 0) of
identical to the tables for the case X = S (l = 0) of Chapter 14. Finally, the
tables for the case Y = X = S (l = 0) are identical to the tables for the case
s, defined as in 14.59, is negative so that Lemma 13.1 applies. We obtain in
this way tables which are similar to the tables 14.215 - 14.223. In fact, the
tables for the case (Y X) = (LL) (l = 2) are identical to the tables for the case
X = L (l = 1) of Chapter 14. The tables for the case (Y X) = (OL) (l = 1) are
identical to the tables for the case X = L (l = 1) of Chapter 14. The tables for
of Chapter 14. All terms have positive excess index so there are no borderline
terms.
522
Proposition 15.2 The contribution to the error integral
Z
2qn +2l (n)
δ | (Y X) τ c |dµg
Mc′ ∗
of the term
(X)
L̃Y J
is bounded as follows.
1. For (Y X) = (LL).
O(δ)
(0) (0) (2)
C( E 2 )1/2 ( E 1 )1/2 ( E 2 )1/2 + O(δ 1/2 )

O(δ)
(0) (2) (3)
C( E 2 )1/2 ( E 1 )1/2 ( F 2 )1/2 + O(δ 1/2 )
2. For (Y X) = (OL).
O(δ)
(0) (0) (2)
C( E 2 )1/2 ( E 1 )1/2 ( E 2 )1/2 + O(δ 1/2 )

O(δ)
(0) (2) (3)
C( E 2 )1/2 ( E 1 )1/2 ( F 2 )1/2 + O(δ 1/2 )
3. For (Y X) = (OO).
O(δ)
523
(1) (0) (2)
C( E 2 )1/2 ( E 1 )1/2 ( E 2 )1/2 + O(δ 1/2 )
O(δ)
(1) (2) (3)
C( E 2 )1/2 ( E 1 )1/2 ( F 2 )1/2 + O(δ 1/2 )
4. For (Y X) = (OS).

(0) (2) (3)
4 ∞ 1/2
C D /1 (χ̂) + D0 (χ̂) + ( E 2 ) ( E 1 )1/2 ( F 2 )1/2 + O(δ 1/2 )
5. For (Y X) = (SS).

(0) (2) (3)
∞ 1/2
C D0 (χ̂) + ( E 2 ) ( E 1 )1/2 ( F 2 )1/2 + O(δ 1/2 )
524
Chapter 16
The Energy-Flux Estimates.

Completion of the
Continuity Argument
16.1 The energy-flux estimates

We now consider the error integrals
Z
(n)
δ 2qn | τ 2 |dµg : n = 0, 1, 2, 3 (16.1)
Mc′ ∗
which appear on the right hand sides of the energy-flux inequalities 12.284 -
12.285 (see definitions 12.286). Let us recall the definitions 12.276, 12.278.
Combining the results of Propositions 13.1, 14.1, 14.2, 14.3, 15.1 and 15.2 we
arrive at the following bounds for the error integrals 16.1:
Z
2q0 (0)
δ | τ 2 |dµg ≤ O(δ)
Mc′ ∗
Z
(1) (0) (1) (0) (2)
δ 2q1 | τ 2 |dµg ≤ C D0∞ (χ̂) + ( E 2 )1/2 + ( E 2 )1/2 ( E 2 )1/2 ( E 2 )1/2
Mc′ ∗
+O(δ 1/2 )
Z
(2)
δ 2q2 | τ 2 |dµg ≤ O(δ)
Mc′ ∗
Z
(3)
δ 2q3 | τ 2 |dµg ≤
Mc′ ∗

(0) (1) (2) (3)
4 ∞ 1/2 1/2
C D/1 (χ̂) + D0 (χ̂) + ( E 2 ) + ( E 2 ) ( E 2 )1/2 ( F 2 )1/2
+O(δ 1/2 ) (16.2)
525
Now, the initial data quantities D0∞ (χ̂), D /41 (χ̂) come from the bound 13.38
for k (K) îkL∞ (Su,u ) , the bound 14.79 for k (S) îkL∞ (Su,u ) , the bound 15.95 for
/Oi (S) îkL4 (Su,u ) , and the bound 15.100 for kL̂
kL̂ /S (S) îkL4 (Su,u ) . The contribu-
tions of the initial data quantities in question to these bounds is through the
bounds of Chapters 3 and 4 for kχ̂kL∞ (Su,u ) and k∇ /χ̂kL4 (Su,u (see first of 8.27,
8.30). Evidently, the bounds of Chapters 3 and 4 hold equally well if in the
definitions 3.49, 3.122 Cu0 is replaced by its part lying in Mc′∗ and in the def-
initions 4.97, 4.98 the supremum over [0, δ] is replaced by the supremum over
[0, min{δ, c∗ − u0 }]. Thus, in the inequalities 16.2 we may replace D0∞ (χ̂) and
/41 (χ̂) by:
D

sup |u0 |2 δ −1/2 kχ̂kL∞ (Su,u0 ) (16.3)
u∈[0,min{δ,c∗ −u0 }]
and:
sup |u0 |5/2 δ −1/2 kχ̂kL4 (Su,u0 ) (16.4)
u∈[0,min{δ,c∗ −u0 }]
respectively.
Consider the estimate of Proposition 6.2 for χ̂:
/ 2 χ̂kL4 (Su,u ) + k∇
≤ Cδ 1/2 |u|−5/2 (R
/41 (β) + R∞
0 (β)) + O(δ|u|
−7/2
) (16.5)
in Dc′ ∗ . This estimate holds in particular at u = u0 . Through Lemma 5.2 with

p = 4, it implies (see definitions 3.49, 4.98):

sup /41 (β) + R∞
|u0 |2 δ −1/2 kχ̂kL∞ (Su,u0 ) ≤ C(R 0 (β)) + O(δ
1/2
)
u∈[0,min{δ,c∗ −u0 }]

sup |u0 |5/2 δ −1/2 kχ̂kL4 (Su,u0 ) ≤ C(R
/41 (β) + R∞
0 (β)) + O(δ
1/2
)
u∈[0,min{δ,c∗ −u0 }]
(16.6)
hence, by 14.84:
(1)
sup |u0 |2 δ −1/2 kχ̂kL∞ (Su,u0 ) ≤ C( E 2 )1/2 + O(δ 1/2 )
u∈[0,min{δ,c∗ −u0 }]
(1)
sup |u0 |5/2 δ −1/2 kχ̂kL4 (Su,u0 ) ≤ C( E 2 )1/2 + O(δ 1/2 )
u∈[0,min{δ,c∗ −u0 }]
(16.7)
In view of 16.7, the inequalities 16.2 simplify to:

Z
(0)
δ 2q0 | τ 2 |dµg ≤ O(δ)
Mc′ ∗
Z
(1) (0) (1) (0) (2)
2q1
δ | τ2 |dµg ≤ C ( E 2 )1/2 + ( E 2 )1/2 ( E 2 )1/2 ( E 2 )1/2 + O(δ 1/2 )
Mc′ ∗
526
Z
(2)
δ 2q2 | τ 2 |dµg ≤ O(δ)
Mc′ ∗
Z
(3) (0) (1) (2) (3)
2q3
δ | τ2 |dµg ≤ C ( E 2 )1/2 + ( E 2 )1/2 ( E 2 )1/2 ( F 2 )1/2 + O(δ 1/2 )
Mc′ ∗
(16.8)
The constant C in the second and fourth bound may be taken to be the same
constant.
Substituting the bounds 16.8 in the energy-flux inequalities 12.284, 12.285,
we obtain the inequalities:
(0) (0)
E 2≤ D +O(δ)

(1) (1) (0) (1) (0) (2)
1/2 1/2
E 2≤ D +C ( E 2 ) + ( E 2 ) ( E 2 )1/2 ( E 2 )1/2 + O(δ 1/2 )
(2) (2)
E 2≤ D +O(δ)

(3) (3) (0) (1) (2) (3)
E 2≤ D +C ( E 2 )1/2 + ( E 2 )1/2 ( E 2 )1/2 ( F 2 )1/2 + O(δ 1/2 )
(16.9)
and:

(3) (3) (0) (1) (2) (3)
F 2≤ D +C ( E 2 )1/2 + ( E 2 )1/2 ( E 2 )1/2 ( F 2 )1/2 + O(δ 1/2 ) (16.10)
(n)
′4
Choosing δ suitably small depending on the quantities D : n = 0, 1, 2, 3; D[1] (α)
and the quantities D0∞ , D
/41 , D /3 (trχ), we can make each of the O(δ 1/2 )
/42 (trχ), D
and O(δ) terms on the right hand sides of the inequalities 16.9 and 16.10 less
than or equal to 1. We then obtain the following closed system of inequalities
(0) (1) (2) (3) (3)
for the quantities E 2 , E 2 , E 2 , E 2 and F 2 :
(0) (0)
E 2≤ D +1

(1) (1) (0) (1) (0) (2)
E 2≤ D +C ( E 2 )1/2 + ( E 2 )1/2 ( E 2 )1/2 ( E 2 )1/2 + 1
(2) (2)
E 2≤ D +1

(3) (3) (0) (1) (2) (3)
E 2≤ D +C ( E 2 )1/2 + ( E 2 )1/2 ( E 2 )1/2 ( F 2 )1/2 + 1
(16.11)
and:

(3) (3) (0) (1) (2) (3)
1/2 1/2
F 2≤ D +C ( E 2 ) + ( E 2 ) ( E 2 )1/2 ( F 2 )1/2 + 1 (16.12)
527
Substituting the first and third of inequalities 16.11 in the second, we obtain:
(1) (1) (0) (2)
E 2≤ D +C( D +1)( D +1)1/2 + 1
(0) (2) (1)
+C( D +1)1/2 ( D +1)1/2 ( E 2 )1/2 (16.13)
Writing
(0) (2) (1) 1 2 (0) (2) 1 (1)
C( D +1)1/2 ( D +1)1/2 ( E 2 )1/2 ≤ C ( D +1)( D +1) + E 2
2 2
16.13 implies:
(1) (1)
E 2≤ D +A (16.14)
where:
(1) (0) (2) (0) (2)
A = 1 + ( D +1) + 2C( D +1)( D +1)1/2 + C 2 ( D +1)( D +1) (16.15)
(0) (1) (2)
is a positive continuous non-decreasing function of D , D , D . It follows that:
(0) (2) (0) (2) (1)
C( D +1)( D +1)1/2 + 1 + C( D +1)1/2 ( D +1)1/2 ( E 2 )1/2 ≤ A
(16.16)
which in turn through 16.13 implies 16.14.

Substituting the first and third of inequalities 16.11 and 16.14 in inequality
16.12 we obtain:
(3) (3) (3)
F 2≤ D +1 + B( F 2 )1/2 (16.17)
where:
(0) (1) (2)
1/2 1/2
B = C ( D +1) + ( D +A) ( D +1)1/2 (16.18)
(0) (1) (2)

is a positive continuous non-decreasing function of D , D , D . Writing
(3) 1 2 1 (3)
B( F 2 )1/2 ≤ B + F2
2 2
16.16 implies:
(3) (3)
F 2≤ 2( D +1) + B 2 (16.19)
It follows that:
(3) (3)
1 + B( F 2 )1/2 ≤ 1 + ( D +1) + B 2 (16.20)
which in turn through 16.17 implies 16.19. Substituting the first and third of
the inequalities 16.11 and 16.14 in the fourth of the inequalities 16.11 we obtain:
(3) (3) (3)
E 2≤ D +1 + B( F 2 )1/2 (16.21)
528
hence by 16.20, also:
(3) (3)
E 2≤ 2( D +1) + B 2 (16.22)
We conclude from the above that:
(0) (1) (0) (1) (2) (2) (3) (0) (1) (2)
P2 ≤ max{ D +1, D +A( D , D , D ), D +1, 2( D +1) + B 2 ( D , D , D )} (16.23)
Moreover, in view of 16.16 and 16.20 the error integrals 16.1 are bounded ac-
cording to:
Z
(0)
δ 2q0 | τ 2 |dµg ≤ 1
Mc′ ∗
Z
(1)
δ 2q1 | τ 2 |dµg ≤ A
Mc′ ∗
Z
(2)
δ 2q2 | τ 2 |dµg ≤ 1
Mc′ ∗
Z (3)
2q3 (3)
δ | τ 2 |dµg ≤ 1 + ( D +1) + B 2 (16.24)
Mc′ ∗
and the error integral bounds 16.23 imply through inequalities 12.284 - 12.285
the energy-flux bounds, namely the first and third of 16.11, 16.14, 16.19 and
(0) (1) (2) (3)
16.22. Let us define G( D , D , D , D ) by:
(0) (1) (2) (3)
G( D , D , D , D ) (16.25)
(0) (1) (0) (1) (2) (2) (3) (0) (1) (2)
= 2 max{ D +1, D +A( D , D , D ), D +1, 2( D +1) + B 2 ( D , D , D )}
(0) (1) (2) (3) (0) (1)
Then G( D , D , D , D ) is a positive non-decreasing continuous function of D , D
(2) (3)
, D , D and we have:
1 (0) (1) (2) (3)
P2 ≤ G( D , D , D , D ) (16.26)
2
(0) (1) (2) (3)
Thus, with the choice 16.25 for the function G( D , D , D , D ) we have established
inequality 12.261 of Chapter 12 which has been our aim.
16.2 Higher order bounds

In this section we shall derive uniform higher order bounds for the solution in
the domain Mc′∗ .
Consider the spacelike hypersurfaces Ht : t ∈ (u0 , c∗ ) corresponding to
u + u = t (see 1.7, 1.8). We denote by Ht′ the part of Ht which lies in the
non-trivial part of the spacetime manifold:
Ht′ = {p ∈ Ht : u(p) ≥ 0} (16.27)
529
Setting
λ=u−u (16.28)
the induced metric g on the Ht is given in terms of canonical coordinates (see
1.203) (λ, ϑ1 , ϑ2 ) by:
1 1
g = Ω2 dλ ⊗ dλ + g/AB (dϑA − bA dλ) ⊗ (dϑB − bB dλ) (16.29)
2 2
The volume form dµg of g is given by:
p
dµg = Ω detg/dλ ∧ dϑ1 ∧ dϑ2 = Ωdλ ∧ dµ/g (16.30)
The future directed timelike vectorfield

1
T = (L + L) (16.31)
2
generates a flow Fτ , which maps Ht onto Ht+τ . For,
Tt = 1 (16.32)
We have:
T = ΩT̂ (16.33)
where T̂ is the future directed timelike unit normal to the Ht .
We shall first derive L2 (Ht′ ) bounds, uniform in t ∈ (u0 , c∗ ), for the deriva-
tives of the curvature componets of all orders. This will allow us to show that
extending the manifold M Sc∗ by attaching the future boundary Hc∗ , the metric
extends smoothly to Mc∗ Hc∗ . Only rough bounds are needed here, the depen-
dence of the bounds on δ or u0 being of no consequence for our argument. Since
the derivation of these bounds is standard, we shall only outline the arguments
involved.
For c ∈ (u0 , c∗ ) the spacetime domain Mc′ is foliated by the Ht′ : t ∈ (u0 , c):
[
Mc′ \ S0,u0 = Ht′ (16.34)
t∈(u0 ,c)
′u
Given any u1 ∈ (0, δ), let us denote by Dc 1 the subdomain of the parameter do-
′u
main Dc′ corresponding to u ≤ u1 and let Mc 1 be the corresponding spacetime
domain:
′u
[
Mc 1 = Su,u (16.35)
′u1
(u,u)∈Dc
′u
a subdomain of the spacetime domain Mc′ . We also denote by Hc 1 , the part of
Hc′ where u ≤ u1 :
′u
Hc 1 = {p ∈ Hc′ : u(p) ≤ u1 } (16.36)
′u1 u ′u c−u
The past boundary of Mc is C c0 Cu01 and its future boundary is Hc 1 C u1 1 .
S S
We have the following form of the divergence theorem in spacetime (see Lemma
12.4).
530
Lemma 16.1 Let P be a vectorfield and τ a function defined on Mc′∗ and
satisfying the equation:
divP = τ
Let also P 4 vanish on C 0 . Then for c ∈ (u0 , u0 + u1 ] we have:
Z
E ′u1 (c) − E c−u0 (u0 ) = ′u
τ dµg
Mc 1
and for c ∈ (u0 + u1 , c∗ ) we have:

Z
E ′u1 (c) − E u1 (u0 ) + F c−u1 (u1 ) = ′u
τ dµg
Mc 1
′u
Here E ′u1 is the “energy”’ associated to Hc 1 :
Z Z
E ′u1 (c) = ′u
P T̂
dµg = ′u
(P 3 + P 4 )dµg
Hc 1 Hc 1
Proof: We integrate equation 12.70 with respect to (u, u) on the parameter

′u
domain Dc 1 . In view of the relations 12.78, the condition of vanishing of P 4
on C 0 , definition 12.72 with u0 in the role of u1 , definition 12.73, the above
definition of E ′u1 (c), and 16.30, the lemma follows.
Let us now consider the case where the components P 3 and P 4 of P are
non-negative functions, which is the case for the energy-momentum density
vectorfields. Then Lemma 16.1 implies, for all c ∈ (u0 , c∗ ):
Z
E ′u1 (c) ≤ E(u0 ) + |τ |dµg (16.37)
Mc′
and since this holds for every u1 ∈ (0, δ), taking the limit u1 → δ we obtain, for
all c ∈ (u0 , c∗ ): Z
E ′ (c) ≤ E(u0 ) + |τ |dµg (16.38)
Mc′
where E ′ (c) is the energy associated to Hc′ :

Z Z
E ′ (c) = P T̂ dµg = (P 3 + P 4 )dµg (16.39)
Hc′ Hc′
(n)
We denote by E ′ 0 (t) : n = 0, 1, 2, 3 the energies associated to the energy-
momentum density vectorfields 12.80 and to Ht′ . They are given by:
(0) Z
′
E 0 (t) = Ω3 (|α|2 + 2|β|2 )dµg
Ht
(1) Z
E ′ 0 (t) = Ω3 |u|2 (2|β|2 + 2|ρ|2 + 2|σ|2 )dµg
Ht
531
(2) Z
E ′ 0 (t) = Ω3 |u|4 (2|ρ|2 + 2|σ|2 + 2|β|2 )dµg
Ht
(3) Z
′
E 0 (t) = Ω3 |u|6 (2|β|2 + |α|2 )dµg (16.40)
Ht
(n)
We denote by (L) E ′ (t) : n = 0, 1, 2, 3 the energies associated to the energy-
(0) Z
(L) ′
E (t) = Ω3 (|α(L̃L R)|2 + 2|β(L̃L R)|2 )dµg
Ht′
(1) Z
(L) ′
E (t) = Ω3 |u|2 (2|β(L̃L R)|2 + 2|ρ(L̃L R)|2 + 2|σ(L̃L R)|2 )dµg
Ht′
(2) Z
(L)
E ′ (t) = Ω3 |u|4 (2|ρ(L̃L R)|2 + 2|σ(L̃L R)|2 + 2|β(L̃L R)|2 )dµg
Ht′
(3) Z
(L)
E ′ (t) = Ω3 |u|6 (2|β(L̃L R)|2 + |α(L̃L R)|2 )dµg (16.41)
Ht′
(n)
We denote by (O) E ′ (t) : n = 0, 1, 2, 3 the energies associated to the energy-
(0) Z X
(O) ′
E (t) = Ω3 (|α(L̃Oi R)|2 + 2|β(L̃Oi R)|2 )dµg
Ht′ i
(1) Z X
(O) ′
E (t) = Ω3 |u|2 (2|β(L̃Oi R)|2 + 2|ρ(L̃Oi R)|2 + 2|σ(L̃Oi R)|2 )dµg
Ht′ i
(2) Z X
(O) ′
E (t) = Ω3 |u|4 (2|ρ(L̃Oi R)|2 + 2|σ(L̃Oi R)|2 + 2|β(L̃Oi R)|2 )dµg
Ht′ i
(3) Z X
(O) ′
E (t) = Ω3 |u|6 (2|β(L̃Oi R)|2 + |α(L̃Oi R|2 )dµg (16.42)
Ht′ i
(3)
We denote by (S) E ′ (t) the energy associated to the energy-momentum density
vectorfield 12.95 and to Ht′ . It is given by:
(3) Z
(S) ′
E (t) = Ω3 |u|6 (2|β(L̃S R)|2 + |α(L̃S R)|2 )dµg (16.43)
Ht′
(n)
We denote by (LL) E ′ (t) : n = 0, 1, 2, 3 the energies associated to the energy-
(0) Z
(LL)
E ′ (t) = Ω3 (|α(L̃L L̃L R)|2 + 2|β(L̃L L̃L R)|2 )dµg
Ht′
532
(1) Z
(LL)
E ′ (t) = Ω3 |u|2 (2|β(L̃L L̃L R)|2
Ht′
+2|ρ(L̃L L̃L R)|2 + 2|σ(L̃L L̃L R)|2 )dµg

(2) Z
(LL)
E ′ (t) = Ω3 |u|4 (2|ρ(L̃L L̃L R)|2 + 2|σ(L̃L L̃L R)|2
Ht′
+2|β(L̃L L̃L R)|2 )dµg

(3) Z
(LL)
E ′ (t) = Ω3 |u|6 (2|β(L̃L L̃L R)|2 + |α(L̃L L̃L R)|2 )dµg (16.44)
Ht′
(n)
We denote by (OL) E ′ (t) : n = 0, 1, 2, 3 the energies associated to the energy-
(0) Z X
(OL) ′
E (t) = Ω3 (|α(L̃Oi L̃L R)|2 + 2|β(L̃Oi L̃L R)|2 )dµg
Ht′ i
(1) Z X
(OL) ′
E (t) = Ω3 |u|2 (2|β(L̃Oi L̃L R)|2
Ht′ i
+2|ρ(L̃Oi L̃L R)|2 + 2|σ(L̃Oi L̃L R)|2 )dµg

(2) Z X
(OL)
E ′ (t) = Ω3 |u|4 (2|ρ(L̃Oi L̃L R)|2 + 2|σ(L̃Oi L̃L R)|2
Ht′ i
+2|β(L̃Oi L̃L R)|2 )dµg

(3) Z X
(OL)
E ′ (t) = Ω3 |u|6 (2|β(L̃Oi L̃L R)|2 + |α(L̃Oi L̃L R)|2 )dµg
Ht′ i
(16.45)
(n)
We denote by (OO) E ′ (t) : n = 0, 1, 2, 3 the energies associated to the
energy-momentum density vectorfields 12.110 and to Ht′ . They are given by:
(0) Z X
(OO) ′
E (t) = Ω3 (|α(L̃Oj L̃Oi R)|2 + 2|β(L̃Oj L̃Oi R)|2 )dµg
Ht′ i,j
(1) Z X
(OO) ′
E (t) = Ω3 |u|2 (2|β(L̃Oj L̃Oi R)|2
Ht′ i,j
+2|ρ(L̃Oj L̃Oi R)|2 + 2|σ(L̃Oj L̃Oi R)|2 )dµg

(2) Z X
(OO)
E ′ (t) = Ω3 |u|4 (2|ρ(L̃Oj L̃Oi R)|2 + 2|σ(L̃Oj L̃Oi R)|2
Ht′ i,j
+2|β(L̃OJ L̃Oi R)|2 )dµg
533
(3) Z X
(OO)
E ′ (t) = Ω3 |u|6 (2|β(L̃Oj L̃Oi R)|2 + |α(L̃Oj L̃Oi R)|2 )dµg
Ht′ i,j
(16.46)
(3)
We denote by (OS) E ′ (t) the energy associated to the energy-momentum den-
sity vectorfields 12.115 and to Ht′ . It is given by:
(3) Z X
(OS)
E ′ (t) = Ω3 |u|6 (2|β(L̃Oi L̃S R)|2 + |α(L̃Oi L̃S R)|2 )dµg (16.47)
Ht′ i
(3)
Finally, we denote by (SS) E ′ (t) the energy associated to the energy-momentum
density vectorfield 12.120 and to Ht′ . It is given by:
(3) Z
(SS)
E ′ (t) = Ω3 |u|6 (2|β(L̃S L̃S R)|2 + |α(L̃S L̃S R)|2 )dµg (16.48)
Ht′
(n)
Next, we denote by E ′ 1 (t) : n = 0, 1, 2, 3 the total 1st order energies
associated to Ht′ . They are given by:
(0) (0) (0) (0)
E ′ 1 (t) =E ′ 0 (t) + δ 2 (L)
E ′ (t) + (O)
E ′ (t)
(1) (1) (1) (1)
E ′ 1 (t) =E ′ 0 (t) + δ 2 (L)
E ′ (t) + (O)
E ′ (t)
(2) (2) (2) (2)
E ′ 1 (t) =E ′ 0 (t) + δ 2 (L)
E ′ (t) + (O)
E ′ (t)
(3) (3) (3) (3) (3)
E ′ 1 (t) =E ′ 0 (t) + δ 2 (L)
E ′ (t) + (O)
E ′ (t) + (S)
E ′ (t) (16.49)
(n)
We denote by E ′ 2 (t) : n = 0, 1, 2, 3 the total 2nd order energies associated to
Ht′ . They are given by:
(0) (0) (0) (0) (0)
E ′ 2 (t) =E ′ 1 (t) + δ 4 (LL)
E ′ (t) + δ 2 (OL)
E ′ (t) + (OO)
E ′ (t)
(1) (1) (1) (1) (1)
E ′ 2 (t) =E ′ 1 (t) + δ 4 (LL)
E ′ (t) + δ 2 (OL)
E ′ (t) + (OO)
E ′ (t)
(2) (2) (2) (2) (2)
E ′ 2 (t) =E ′ 1 (t) + δ 4 (LL)
E ′ (t) + δ 2 (OL)
E ′ (t) + (OO)
E ′ (t)
(3) (3) (3) (3) (3)
E ′ 2 (t) =E ′ 1 (t) + δ 4 (LL)
E ′ (t) + δ 2 (OL)
E ′ (t) + (OO)
E ′ (t)
(3) (3)
(OS)
+ E ′ (t) + (SS)
E ′ (t) (16.50)
The 2nd order energies associated to Ht′ control the L2 norms on Ht′ of all 2nd
derivatives of all curvature components.
534
Finally, we define the quantities:
!
(n) (n)
′ 2qn ′
E 2= sup δ E 2 (t) : n = 0, 1, 2, 3 (16.51)
t∈(u0 ,c∗ )
(the exponents qn being defined by 12.286). Then in view of the inequality 16.38
we obtain:
(n) (n)
Z
(n)
′
E 2 ≤ D +δ 2qn | τ 2 |dµg : n = 0, 1, 2, 3 (16.52)
Mc′ ∗
The bounds 16.24 for the error integrals then imply the following sharp bounds
(n)
for the quantities E ′ 2 :
(0) (0)
′
E 2 ≤ D +1
(1) (1)
′
E 2 ≤ D +A
(2) (2)
′
E 2 ≤ D +1
(3) (3)
E 2 ≤ 2( D +1) + B 2
′
(16.53)
2
We now begin the outline of the derivation of the rough L bounds (Ht′ )
for the higher derivatives of the curvature components. Here we use only the
future directed timelike vectorfield T (see 16.30) as a multiplier field, and the
vectorfields L and L as commutation fields. We thus consider the derived Weyl
fields:
Wm,n = (L̃L )m (L̃L )n R (16.54)
where m and n are non-negative integers. The order of Wm,n is k = m + n. We
associate to Wm,n the energy-momentum density vectorfields:
Pm,n = P (Wm,n ; T, T, T ) (16.55)
′
(see 12.63). We then consider the energies Em,n (t) associated to Pm,n and to
′
Ht : Z
′ 3 4
Em,n = (Pm,n + Pm,n )dµg (16.56)
Ht′
(see 16.39). Now, by Lemma 12.2, the energy E ′ [W ](t) associated to P (W ; T, T, T )

and to Ht′ , that is:
Z
′
E [W ](t) = (P 3 (W ; T, T, T ) + P 4 (W ; T, T, T ))dµg (16.57)
Ht′
is given by:

1 3
Z
E ′ [W ](t) = |α(W )|2 + |β(W )|2 + (|ρ(W )|2 + |σ(W )|2 )
Ht′ 8 2

2 1 2
+|β(W )| + |α(W )| dµg (16.58)
8
535
Thus E ′ [W ](t) is equivalent to the sum of the squares of the L2 norms on Ht′
of all components of W . The total kth order energy Ek′ (t) associated to Ht′ is
then defined by: X
Ek′ (t) = ′
Em,n (t) (16.59)
m+n=k
Now, from 12.64 τm,n , the divergence of Pm,n , is given by:

3 (T )
τm,n = −(divQ(Wm,n )(T, T, T ) − Q(Wm,n )αβγδ π̃αβ T γ T δ (16.60)
2
the second term on the right being the “multiplier”’ part and the first term the
“Weyl current”’ part, given according to Proposition 12.6 by:
− 2((Wm,n )βµδν (Jm,n )µγν + ( ∗ Wm,n )βµδν (Jm,n

∗
)µγν )T β T γ T δ (16.61)
Here Jm,n is the Weyl current corresponding to the Weyl field Wm,n . By Propo-
sition 12.1 the Jm,n satisfy the following recursion relations:
1 1
(J0,n+1 )βγδ = (L̃L J0,n )βγδ + (L) µν
π̃ ∇ν (W0,n )µβγδ + (L)
pµ (W0,n )µβγδ
2 2
1
+ ( (L)
qµβν (W0,n )µνγδ + (L)
qµγν (W0,n )µβνδ + (L)
qµδν (W0,n )µβγν )
2
(16.62)
1 1
(Jm+1,n )βγδ = (L̃L Jm,n )βγδ + (L) µν
π̃ ∇ν (Wm,n )µβγδ + (L)
pµ (Wm,n )µβγδ
2 2
1
+ ( (L)
qµβν (Wm,n )µνγδ + (L)
qµγν (Wm,n )µβ νδ + (L)
qµδν (Wm,n )µβγν )
2
(16.63)
To determine the kth order currents, we first obtain J0,n for n = 1, ..., k from
the recursion relation 16.62 and the fact that J0,0 = 0. We then obtain Jm,n
for m = 1, ..., k − n, n = 0, ..., k from the recursion relation 16.63. The leading
terms in the expression for any component of Jm,n are of the following four
types. First, terms which are bilinear expressions, with coefficients depending
only on g/ and /ǫ , the first factor of which is Dm1 Di1 of a component of (L) π̃, and
the second factor is a (D, D or ∇ /) derivative of a component of Wm2 ,n−1+i2 −i ,
where m1 + m2 = m, i1 + i2 = i and i = 0, ..., n − 1. Second, terms which are
bilinear expressions, with coefficients depending only g/ and /ǫ , the first factor of
which is a (D, D or ∇ /) derivative of Dm1 Di1 of a component of (L) π̃, and the
second factor is a component of Wm2 ,n−1+i2 −i , where m1 + m2 = m, i1 + i2 = i
and i = 0, ..., n − 1. Third, terms which are bilinear expressions with coefficients
depending only on g/ and /ǫ , the first factor of which is Dj1 of a component
of (L) π̃, and the second factor is a (D, D or ∇ /) derivative of a component of
Wm−1+j2 −j,n , where j1 + j2 = j and j = 0, ..., m − 1. Fourth, terms which
are bilinear expressions, with coefficients depending only on g/ and /ǫ , the first
factor of which is a (D, D or ∇ /) derivative of Dj1 of a component of (L) π̃,
536
and the second factor is a component of Wm−1+j2 −j,n , where j1 + j2 = j and
j = 0, ..., m − 1. Since the components of (L) π, (L) π are expressed by 8.21,
8.22 in terms of the connection coefficients, we can simply say that the leading
terms in the expression of a component of Jm,n are terms which are bilinear
expressions, with coefficients depending only on g/ and /ǫ , the first factor of
which is a ith order derivative of a connection coefficient, and the second factor
a jth order derivative of a component of R, where i + j = m + n = k. The
lower order terms in the expression of a component of Jm,n are terms which are
are l-linear expressions, l ≥ 3, with coefficients depending only on g/ and /ǫ , the
first l − 1 factors of which are ip th order derivatives of connection coefficients,
p = 1, ..., l − 1 respectively, and the last factor is jth order derivative of a
component of R, where i1 + ... + il−1 = i − l + 2 and i + j = m + n = k. Note
that l ≤ i + 2.
As noted previously, we already have sharp L2 bounds on the Ht′ for the 2nd
derivatives of the curvature components. By the results of Chapter 6 we already
have sharp L4 (S) bounds for the 2nd derivatives of the connection coefficients,
except for /dDω, /dDω, D2 ω, D2 ω. For /dDω, D 2 ω we have sharp L2 (S) bounds
from Chapter 7. The L2 (S) and L4 (S) bounds imply L2 bounds on the Ht′ .
For /dDω, D2 ω we have sharp L2 bounds on the Cu from Chapter 7. We shall
presently show how sharp L2 bounds on the Ht′ may also be obtained for these.
Here and in the following we shall make use of the following elementary lemma.
We denote by λ∗ (t) the supremum of λ on Ht′ :
: for t ∈ (u0 + δ, c∗ )

∗ 2δ − t
λ (t) = sup λ = (16.64)
H′ t − 2u 0 : for t ∈ (u0 , u0 + δ]
t
The infimum of λ on Ht′ is −t.
Lemma 16.2 Let t ∈ (u0 , c∗ ) and let f (u, u) be a non-negative function which
is square integrable on Dt′ . Then we have:
 !2 1/2
1
λ∗ (t) 2 (t−λ)

Z 1
Z 
f (t + λ), u′ du′ dλ
 −t u0 2 
!1/2
t λ∗ (t′ )
1 ′ ′ 1 ′
Z Z
2 ′ ′
≤ f (t + λ ), (t − λ ) dλ dt′
u0 −t′ 2 2
and:

1
!2 1/2
λ∗ (t) 2 (t+λ)

Z 1
Z 
f u′ , (t − λ) du′ dλ
 −t 0 2 
!1/2
t λ∗ (t′ )
1 ′ ′ 1 ′
Z Z
2 ′ ′
≤ f (t + λ ), (t − λ ) dλ dt′
max{t−δ,u0 } −t′ 2 2
537
Proof: To establish the first inequality we define the function f˜(u, u) to be the
exension of f by zero to u < u0 (see Figure 16.1)
u + u = c∗
u+u=t
u
=
δ
u
=
0
u0
=
u
u + u = u0
Figure 16.1: The domain considered in regard to the first inequality
and write:
1
2 (t−λ)
Z t
1 1 1
Z
f ′ ′
(t + λ), u du = ˜
f (t + λ), t − (t + λ) dt′
′
(16.65)
u0 2 u0 2 2
Setting:
′ ˜ 1 ′ 1
g(t , λ) = f (t + λ), t − (t + λ) (16.66)
2 2
the left hand side of the inequality is:
(Z 2 )1/2
λ∗ (t) Z t Z t
′ ′
g(t , λ)dt dλ = g(t′ , ·)dt′ (16.67)
−t u0 u0 L2 (−t,λ∗ (t))
and we have:
Z t Z t
g(t′ , ·)dt′ ≤ kg(t′ , ·)kL2 (−t,λ∗ (t)) dt′ (16.68)
u0 L2 (−t,λ∗ (t)) u0
Now,
λ∗ (t)
1 1
Z
kg(t ′
, ·)k2L2 (−t,λ∗ (t)) = f˜2 ′
(t + λ), t − (t + λ) dλ (16.69)
−t 2 2
Setting λ′ = t − t′ + λ the last integral is:

t−t′ +λ∗ (t)
1 ′ ′ 1 ′
Z
f˜2 (t + λ ), (t − λ ) dλ′
′
(16.70)
−t′ 2 2
538
Now, for t′ > u0 + δ (hence also t > u0 + δ) we have, by 16.64:
t − t′ + λ∗ (t) = t − t′ + 2δ − t = 2δ − t′ = λ∗ (t′ )
hence, since f˜ coincides with f in Dt′ , 16.70 is:

λ∗ (t′ )
1 ′ ′ 1 ′
Z
2
f (t + λ ), (t − λ ) dλ′
′
(16.71)
−t′ 2 2
For t′ ≤ u0 + δ and t > u0 + δ we have, by 16.64:
t − t′ + λ∗ (t) = 2δ − t′ ≥ t′ − 2u0 = λ∗ (t′ )
hence, since f˜ coincides with f in Dt′ and vanishes for u < u0 , 16.70 is equal to
16.71. Finally, for t ≤ u0 + δ (hence also t′ ≤ u0 + δ) we have, by 16.64:
t − t′ + λ∗ (t) = 2t − t′ − 2u0 ≥ t′ − 2u0 = λ∗ (t′ )
hence again, since f˜ coincides with f in Dt′ and vanishes for u < u0 , 16.70 is
equal to 16.71. Thus, we have, in general,
!1/2
λ(t′ )
1 ′ ′ 1 ′
Z
′ 2 ′ ′
kg(t , ·)kL2 (−t,λ∗ (t)) = f (t + λ ), (t − λ ) dλ (16.72)
−t′ 2 2
and the first inequality of the lemma is established.

To establish the second inequality we define the function f˜ to be the exten-
sion of f by zero to u < 0 (see Figures 16.2, 16.3)
u + u = c∗
u
=
δ
u
=
0
u+u=t
0
u
=
u
u + u = u0
Figure 16.2: The domain considered in regard to the second inequality, case
t < u0 + δ
539
u + u = c∗
u+u=t
u
=
δ
δ
−
t
=
u
u+u = t−δ
u
=
0
0
u
=
u
Figure 16.3: The domain considered in regard to the second inequality, case
t ≥ u0 + δ
and, noting that by 16.64:

1 1
(t − λ) ≥ (t − λ∗ (t)) = max{t − δ, u0 }
2 2
we write:
Z 21 (t+λ) Z t
′ 1 ′ 1 1
f u , (t − λ) du = ′
f t − (t − λ), (t − λ) dt′
0 2 1
2 (t−λ) 2 2
Z t
˜ 1 1
= f t − (t − λ), (t − λ) dt′
′
max{t−δ,u0 } 2 2
(16.73)
Setting:
1 1
g(t′ , λ) = f˜ t′ − (t − λ), (t − λ) (16.74)
2 2
the left hand side of the inequality is:
 !2 1/2
Z λ∗ (t) Z t  Z t
g(t′ , λ)dt′ dλ = g(t′ , ·)dt′
 −t max{t−δ,u0 }  max{t−δ,u0 }
L2 (−t,λ∗ (t))
(16.75)
and we have:
Z t Z t
g(t′ , ·)dt′ ≤ kg(t′ , ·)kL2 (−t,λ∗ (t)) dt′
max{t−δ,u0 } max{t−δ,u0 }
L2 (−t,λ∗ (t))
(16.76)
540
Now,
λ∗ (t)
1 1
Z
kg(t′ , ·)k2L2 (−t,λ∗ (t)) = f˜2 t′ − (t − λ), (t − λ) dλ (16.77)
−t 2 2
Setting λ′ = t′ − t + λ the last integral is:

Z t′ −t+λ∗ (t)
1 ′ 1
f˜2 (t + λ′ ), (t′ − λ′ ) dλ′ (16.78)
t′ −2t 2 2
Since t′ − 2t ≤ −t′ and f˜ coincides with f on Dt′ and vanishes for u < 0, 16.78
is equal to:
Z t′ −t+λ∗ (t)
2 1 ′ ′ 1 ′
f (t + λ ), (t − λ ) dλ′
′
(16.79)
−t′ 2 2
Now, for t > u0 + δ we have, by 16.64:
t′ − t + λ∗ (t) = t′ − 2t + 2δ ≤ 2δ − t′ = λ∗ (t′ )
while for t ≤ u0 + δ (hence also t′ ≤ u0 + δ) we have, by 16.64:
t′ − t + λ∗ (t) = t′ − 2u0 = λ∗ (t′ )
In both cases 16.79 is less than or equal to:

Z λ∗ (t′ )
2 1 ′ ′ 1 ′
f (t + λ ), (t − λ ) dλ′
′
(16.80)
−t′ 2 2
Thus:
!1/2
λ∗ (t′ )
1 ′ ′ 1 ′
Z
′ 2 ′ ′
kg(t , ·)kL2 (−t,λ∗ (t)) ≤ f (t + λ ), (t − λ ) dλ (16.81)
−t′ 2 2
and the second inequality of the lemma is established as well.
Let now θ be an arbitrary S tensorfield. By 16.30 we have:

 
Z Z λ∗ (t) Z
kθk2L2 (H ′ ) := |θ|2 dµg =  |θ|2 Ωdµ/g  dλ (16.82)
t
Ht′ −t S 1 (t+λ), 1 (t−λ)
2 2
hence:
Z λ∗ (t) Z λ∗ (t)
−1
C kθk2L2 (S 1 ) dλ ≤ kθk2L2 (Ht′ ) ≤C kθk2L2 (S 1 ) dλ
(t+λ), 1 (t−λ) (t+λ), 1 (t−λ)
−t 2 2 −t 2 2
(16.83)
Aplying Lemma 16.2 to the function
f (u, u) = kθkL2 (Su,u )
541
we then obtain:
 !2 1/2
Z λ∗ (t) Z 1
2 (t−λ)
 Z t
kθkL2 (S 1 (t+λ),u′ ) du′ dλ ≤C kθkL2 (H ′′ ) dt′
2 t
 −t u0  u0
(16.84)
and:
 !2 1/2
1
Z λ∗ (t) Z 2 (t+λ)
 Z t
kθkL2 (Su′ , 1 (t−λ) ) du′ dλ ≤ kθkL2 (H ′′ ) dt′
2 t
 −t 0  max{t−δ,u0 }
(16.85)
We now revisit the proofs of Propositions 7.6 and 7.8. Setting u = 12 (t + λ),
u = 12 (t − λ) in 7.357, squaring, integrating with respect to λ on (−t, λ∗ (t)),
and taking the square root, we obtain, by 16.84 and 16.83:
Z t
kd/DωkL2(Ht′ ) ≤ C kn′ kL2 (H ′′ ) dt′ (16.86)
t
u0
Substituting for n′ we now obtain an estimate for /dDω in L2 (Ht′ ) in terms of

Z t
/DρkL2 (H ′′ ) dt′
kd (16.87)
t
u0
plus lower order terms. Now, /dDρ is bounded in L2 (Ht′ ) through the bounds
16.53. We obtain in this way the following sharp bound for /dDω in L2 (Ht′ ):
/DωkL2(Ht′ ) ≤ O(δ −1/2 |t|−2 ) : ∀t ∈ (u0 , c∗ )

kd (16.88)
Setting u = 12 (t + λ), u = 21 (t − λ) in 7.378, squaring, integrating with respect

to λ on (−t, λ∗ (t)), and taking the square root, we obtain, by 16.84 and 16.83:
Z t
−1 2
|t| kD ωkL (Ht ) ≤ C
2 ′ |t′ |−1 kṅkL2 (H ′′ ) dt′ (16.89)
t
u0
Substituting for ṅ we now obtain an estimate for D2 ω in L2 (Ht′ ) in terms of

Z t
kD2 ρkL2 (H ′′ ) dt′ (16.90)
t
u0
plus lower order terms. Now, D2 ρ is bounded in L2 (Ht′ ) through the bounds
16.53. We obtain in this way the following sharp bound for D2 ω in L2 (Ht′ ):
kD2 ωkL2 (Ht′ ) ≤ O(δ −3/2 |t|−1 ) : ∀t ∈ (u0 , c∗ ) (16.91)
Let us now consider the kth order energy Ek′ (t) associated to Ht′ (see 16.54 -
16.59). It seems at first sight that Ek′ (t) controls the L2 norm on Ht′ of only the
D, D derivatives of the curvature components of of order k and no ∇ / derivatives.
542
Nevertheless, we shall presently show that the El′ (t) for l = 3, ..., k together with
the sharp bounds which we have already obtained, actually control the L2 norms
on Ht′ of all D, D and ∇/ derivatives of the curvature components of order up to
k. This is by virtue of the Bianchi identities.
Consider in fact the Bianchi identities, given by Proposition 1.2. Denoting
by ≃ equality up to lower order terms (here products of connection coefficients
with curvature components), these equations read:
Ω−1 D̂α ≃ ∇ˆ
/⊗β
−1 ˆ
Ω D̂α ≃ −∇ /⊗β
Ω−1 Dβ ≃ div
/ α
−1
Ω Dβ ≃ −div/ α
Ω−1 Dβ ≃ /dρ + ∗
/dσ
−1 ∗
Ω Dβ ≃ −d
/ρ + /dσ
−1
Ω Dρ ≃ div
/ β
Ω−1 Dσ ≃ −curl
/ β
−1
Ω Dρ ≃ −div
/ β
Ω−1 Dσ ≃ −curl
/ β (16.92)
In the third and fourth of the above equations we have the elliptic operator div
/
from trace-free symmetric 2-covariant tensorfields to 1-forms on Su,u , analyzed
in Chapter 5 (and also in Chapter 7). The L2 -adjoint of this operator is the
operator
1
ξ 7→ − ∇ ˆ
/⊗ξ (16.93)
2
from 1-forms ξ to trace-free symmetric 2-covariant tensorfields on Su,u , and we
have:
∇ ˆ / θ=△
/⊗div / θ − 2Kθ (16.94)
for any trace-free symmetric 2-covariant tensorfield θ on Su,u . Therefore, ap-
plying D̂ to the first of equations 16.92 and substituting from the third we
obtain:
Ω−2 D̂D̂α ≃ △
/α (16.95)
Also, applying D̂ to the second of equations 16.92 and substituting from the
fourth we obtain:
Ω−2 D̂D̂α ≃ △/α (16.96)
In the last two pairs of equations 16.92 we have the elliptic operator (div
/ , curl
/ )
from 1-forms to pairs of functions on Su,u , analyzed in Chapter 5 (and also in
Chapter 7). The L2 -adjoint of this operator is the operator
∗
(f, g) 7→ −d
/f + /dg (16.97)
from pairs of functions (f, g) to 1-forms on Su,u , and we have:
∗
/ddiv
/ ξ− /dcurl
/ ξ=△
/ ξ − Kξ (16.98)
543
for any 1-form ξ on Su,u , and:
∗ ∗
(div
/ (d
/f − /dg), curl
/ (d/f − /dg)) = (△
/ f, △
/ g) (16.99)
for any pair of functions (f, g) on Su,u . Therefore applying D to the fifth of
equations 16.92 and substituting from the seventh and eighth we obtain:
Ω−1 DDβ ≃ △
/β (16.100)
and applying D to the sixth of equations 16.92 and substituting from the ninth
and tenth we obtain:
Ω−2 DDβ ≃ △ /β (16.101)
Also, applying D to the senth and eighth of equations 16.92 and substituting
from the fifth we obtain:
Ω−2 DDρ ≃ △
/ρ
Ω−2 DDσ ≃ △
/σ (16.102)
Applying D m Dn to equations 16.92 yields:

m+1
Ω−1 D̂ Dn α ≃ ∇ ˆ m Dn β)
/⊗(D
m
Ω−1 D̂ D̂n+1 α ≃ −∇ˆ m Dn β)
/⊗(D
m
Ω−1 Dm Dn+1 β ≃ div
/ (D̂ D̂n α)
m
Ω−1 Dm+1 Dn β ≃ −div
/ (D̂ D̂n α)
Ω−1 Dm+1 D̂n β ≃ /d(Dm Dn ρ) + ∗/d(Dm Dn σ)
Ω−1 Dm Dn+1 β ≃ −d /(Dm Dn ρ) + ∗/d(Dm Dn σ)
Ω−1 Dm Dn+1 ρ ≃ div
/ (Dm Dn β)
Ω−1 Dm Dn+1 σ ≃ −curl
/ (Dm Dn β)
Ω−1 Dm+1 Dn ρ ≃ −div
/ (Dm Dn β)
Ω−1 Dm+1 Dn σ ≃ −curl
/ (Dm Dn β) (16.103)
Here we take m + n = k − 1. In view of the identity 5.189, the third and fourth
m m
/D̂ D̂n αk2L2 (Su,u ) and k∇
of the above give us bounds for k∇ /D̂ D̂n αk2L2 (Su,u ) in
terms of kDm Dn+1 βk2L2 (Su,u ) and kDm+1 Dn βk2L2 (Su,u ) , respectively, plus lower
terms. In view of the identity 5.219, the seventh and eighth of 16.103 give us a
bound for k∇/Dm Dn βk2L2 (Su,u ) in terms of kDm Dn+1 ρk2L2 (Su,u ) , kDm Dn+1 σk2L2 (Su,u )
plus lower order terms, while the ninth and tenth of 16.103 give us a bound for
k∇/Dm Dn βk2L2 (Su,u ) in terms of kDm+1 Dn ρk2L2 (Su,u ) and kDm+1 Dn σk2L2 (Su,u )
plus lower order terms. Finally, in view of the identity:
Z Z
2 ∗ 2
/f − ∗/dg|2 dµ/g

/f | + | /dg| dµ/g =
|d |d (16.104)
Su,u Su,u
544
for any pair of functions f, g on Su,u , the fifth and sixth of 16.103 gives us bounds
/Dm Dn ρk2L2 (Su,u ) , kd
for kd /Dm Dn σk2L2 (Su,u ) in terms of kDm+1 Dn βk2L2 (Su,u ) and
kDm Dn+1 βk2L2 (Su,u ) respectively, plus lower order terms.
Applying Dm Dn to equations 16.95, 16.96, 16.100, 16.101, 6.99 yields:
m+1 m
Ω−2 D̂ D̂n+1 α ≃ △
/ (D̂ D̂n α)
m+1 m
Ω−2 D̂ D̂m+1 α ≃ △
/ (D̂ D̂n α)
Ω−2 Dm+1 Dn+1 β ≃ △
/ (Dm Dn β)
Ω−2 Dm+1 Dn+1 β ≃ △
/ (Dm Dn β)
Ω−2 Dm+1 Dn+1 ρ ≃ △
/ (Dm Dn ρ)
Ω−2 Dm+1 Dn+1 σ ≃ △
/ (D m Dn σ) (16.105)
Here we take m + n = k − l, l = 2, ..., k. Standard elliptic estimates for the

Laplacian △
/ on Su,u then give us bounds for
m
/ l D̂ D̂n αk2L2 (Su,u )
k∇
m
/ l D̂ D̂n αk2L2 (Su,u )
k∇
/ l Dm Dn βk2L2 (Su,u )
k∇
/ l Dm Dn βk2L2 (Su,u )
k∇
/ l Dm Dn ρk2L2 (Su,u )
k∇
/ l Dm Dn σk2L2 (Su,u )
k∇ (16.106)
in terms of
m+1
/ l−2 D̂
k∇ D̂n+1 αk2L2 (Su,u )
m+1
/ l−2 D̂
k∇ D̂n+1 αk2L2 (Su,u )
/ l−2 Dm+1 Dn+1 βk2L2 (Su,u )
k∇
/ l−2 Dm+1 Dn+1 βk2L2 (Su,u )
k∇
/ l−2 Dm+1 Dn+1 ρk2L2 (Su,u )
k∇
/ l−2 Dm+1 Dn+1 σk2L2 (Su,u )
k∇ (16.107)
respectively, plus lower order terms. We thus have a recursion which for l even
gives us bounds for 16.106 in terms of
m+(l/2)
kD̂ D̂n+(l/2) αk2L2 (Su,u )
m+(l/2)
kD̂ D̂n+(l/2) αk2L2 (Su,u )
kDm+(l/2) Dn+(l/2) βk2L2 (Su,u )
545
kDm+(l/2) Dn+(l/2) βk2L2 (Su,u )
kDm+(l/2) Dn+(l/2) ρk2L2 (Su,u )
kDm+(l/2) Dn+(l/2) σk2L2 (Su,u ) (16.108)
respectively, plus lower order terms, while for l odd gives us bounds for 16.106
in terms of
m+((l−1)/2)
k∇
/D̂ D̂n+((l−1)/2) αk2L2 (Su,u )
m+((l−1)/2)
k∇
/D̂ D̂n+((l−1)/2) αk2L2 (Su,u )
/Dm+((l−1)/2) Dn+((l−1)/2) βk2L2 (Su,u )
k∇
/Dm+((l−1)/2) Dn+((l−1)/2) βk2L2 (Su,u )
k∇
/Dm+((l−1)/2) Dn+((l−1)/2) ρk2L2 (Su,u )
k∇
/Dm+((l−1)/2) Dn+((l−1)/2) σk2L2 (Su,u )
k∇ (16.109)
respectively, plus lower order terms. According to the previous discussion the
latter are bounded in terms of
kDm+((l−1)/2) Dn+((l−1)/2)+1 βk2L2 (Su,u )

kDm+((l−1)/2)+1 Dn+((l−1)/2) βk2L2 (Su,u )
kDm+((l−1)/2) Dn+((l−1)/2)+1 ρk2L2 (Su,u )
+kDm+((l−1)/2) Dn+((l−1)/2)+1 σk2L2 (Su,u )
kDm+((l−1)/2)+1 Dn+((l−1)/2) ρk2L2 (Su,u )
+kDm+((l−1)/2)+1 Dn+((l−1)/2) σk2L2 (Su,u )
kDm+((l−1)/2)+1 Dn+((l−1)/2) βk2L2 (Su,u )
or kDm+((l−1))/2 Dn+((l−1)/2)+1 βk2L2 (Su,u ) (16.110)
respectively, plus lower order terms. Setting u = 21 (t + λ), u = 12 (t − λ) and

integrating the inequalities obtained in this way with respect to λ on (−t, λ∗ (t)),
we then obtain a bound for the sum of the squares of the L2 norms on Ht′ of
all derivatives of order k of all curvature componets in terms of the kth order
energy Ek′ (t) associated to Ht′ plus lower order terms. The lower order terms
are bounded in terms of the sum of the squares of the L2 norms on Ht′ of the
derivatives of the curvature components of order up to k − 1 and the sum of the
squares of the L2 norms on Ht′ of the derivatives of the connection coefficients
of order up to k − 1. As we shall presently show, any derivative of order k ′ of
any connection coefficient is bounded in L2 (Ht′ ) by the supremum for t′ ∈ (u0 , t]
of the sum of the L2 norms on Ht′′ of the derivatives of order k ′ of the curvature
components, plus lower order terms. We are thus able to establish inductively
546
that Ek′ (t) bounds the sum of the squares of the L2 norms on Ht′ of all derivatives
of order k of all curvature components, up to terms which are bounded in terms
of supt′ ∈(u0 ,t] Ek′ ′ (t′ ): for k ′ = 3, ..., k − 1 and the sharp bounds already obtained
for up to the second derivatives of the curvature components and the connection
coefficients. Here the Sobolev inequality on the Ht′ is used, which is similar to
the Sobolev inequality on the Cu , Proposition 10.1, but with the D derivative
replaced by the L /N derivative, where N is the vectorfield
1
N= (L − L) Nλ = 1 (16.111)
2
which is tangential to Ht′ and generates a flow Gσ on Ht′ which maps S 12 (t+λ), 12 (t−λ)
onto S 21 (t+λ+σ), 21 (t−λ−σ) . Note that on Ht′ we have: 1 − δ < |t|/|u| ≤ 1.
We shall now show that any derivative ofqorder k of any connection coefficient
′
is bounded in L2 (Ht′ ) by a constant times E k (t), where
′
E k (t) = sup Ek′ (t′ ) (16.112)
t′ ∈(u0 ,t]
plus a lower order term which is bounded in terms of the supremum for t′ ∈
(u0 , t] of the sum of the L2 norms on Ht′′ of the derivatives of the curvature
components of order up to k − 1 and the supremum for t′ ∈ (u0 , t] of the sum of
the L2 norms on Ht′′ of the derivatives of the connection coefficients of order up
to k − 1, and the sharp bounds for the derivatives of the curvature components
and the connection coefficients of up to second order which have already been
obtained. This shall be accomplished by appealing only to the propagation
equations.
It will then follow by induction that the sum of the L2 norms on Ht′ of the
derivatives
q of the connection coefficients of order k is bounded
q by a constant
′ ′
times E k (t) up to terms which are bounded in terms of E k′ (t): for k ′ =
3, ..., k−1 and the sharp bounds already obtained for up to the second derivatives
of the connection coefficients and the curvature components.
We first consider the propagation equations for ∇ / k trχ′ , ∇
/ k χ̂′ . These are
k−2
deduced by applying ∇ / to the equations 5.8 and using Lemma 4.1. We
obtain:
/ k trχ′ = f · ∇
D∇ / k trχ′ + g · ∇
/ k χ̂′ + rk
/ k χ̂′ = h · ∇
D∇ / k trχ′ + i · ∇
/ k χ̂′ + sk (16.113)
The coefficients f , g, h, i are given by 4.79, and we have:
rk ≃ 0
/ kα
sk ≃ −∇ (16.114)
Here, we denote by ≃ equality up to lower order terms involving the ∇

/ derivatives
of trχ′ , χ̂′ and /d log Ω = (1/2)(η + η) of order up to k − 1.
547
We apply Lemma 4.6 to the above propagation equations taking p = 2. In
the case of the first of 16.113 we have ∇ / k trχ′ in the role of θ, g · ∇ / k χ̂′ + rk in
the role of ξ, r = k, ν = −2 and γ = 0. In the case of the second of 16.113 we
/ k χ̂′ in the role of θ, h · ∇
have ∇ / k trχ′ + sk in the role of ξ, r = k + 2, ν = 0 and
γ = i. In view of the estimates of Chapter 3 we obtain:
Z u
k∇/ k trχ′ kL2 (Su,u ) ≤ C kg · ∇/ k χ̂′ + rk kL2 (Su′ ,u ) du′
0
Z u Z u
′ k ′ ′
≤C k∇/ χ̂ kL2 (Su′ ,u ) du + C krk kL2 (Su′ ,u ) du′
0 0
Z u
k∇/ k χ̂′ kL2 (Su,u ) ≤ C kh · ∇/ k trχ′ + sk kL2 (Su′ ,u )
Z 0u Z u
′
≤C k∇/ k trχ′ kL2 (Su′ ,u ) du′ + C ksk kL2 (Su′ ,u ) du′
0 0
(16.115)
Here and in the following we allow the constants to depend on δ, u0 , and the
(n)
′4
quantities D : n = 0, 1, 2, 3, D[1] (α), D0∞ , D /41 , D/42 (trχ), and D/3 (trχ), as only
rough bounds are required.
Setting u = 12 (t + λ), u = 21 (t − λ) in each of 16.115, squaring, integrating
with respect to λ on (−t, λ∗ (t)), and taking the square root, we obtain, by 16.85
and 16.83,
Z t Z t
/ k trχ′ kL2 (Ht′ ) ≤ C ′
k∇ / k χ̂′ kL2 (H ′′ ) dt′ + C
k∇ krk kL2 (H ′′ ) dt′
t t
u0 u0
Z t Zt
/ k χ̂′ kL2 (Ht′ ) ≤ C ′
k∇ / k trχ′ kL2 (H ′′ ) dt′ + C
k∇ ksk kL2 (H ′′ ) dt′
t t
u0 u0
(16.116)
Summing these two inequalities we obtain a linear integral inequality for the
quantity
/ k trχ′ kL2 (Ht′ ) + k∇
k∇ / k χ̂′ kL2 (Ht′ )
which implies:
Z t n o
/ k trχ′ kL2 (Ht′ ) + k∇
k∇ / k χ̂′ kL2 (Ht′ ) ≤ C ′′ krk kL2 (H ′′ ) + ksk kL2 (H ′′ ) dt′
t t
u0
(16.117)
Now by 16.114 the integral
Z t
krk kL2 (H ′′ ) dt′ (16.118)
t
u0
is bounded in terms of the supremum for t′ ∈ (u0 , t] of the sum of the L2 norms
of the derivatives of the connection coefficients of order up to k−1, and the sharp
548
bounds for the derivatives of up to the 2nd order of the connection coefficients
already obtained. Also, the integral
Z
ksk kL2 (H ′′ ) dt′ (16.119)
t
u0
is bounded by the integral

Z
/ k αkL2 (H ′′ ) dt′
k∇ (16.120)
t
u0
(u0 , t] of the sum of the L2 norms of the derivatives of the connection coefficients
of order up to k − 1, and the sharp bounds for the derivatives of up to the 2nd
order of the connection coefficients already obtained. Moreover, according to
what has previously
q been demonstrated, the integral 16.120 is bounded by a
′
constant times E k (t) plus a lower order term which is bounded in terms of
q
′
E k′ (t): for k ′ = 3, ..., k − 1, the supremum for t′ ∈ (u0 , t] of the sum of the L2
norms of the derivatives of the connection coefficients of order up to k − 1, and
the sharp bounds for the derivatives of up to the 2nd order of the connection
coefficients already obtained. We have thus shown what was required in the
case of ∇/ k trχ, ∇/ k χ̂.
We turn to ∇ / k χ̂′ . These statisfy the conjugates of the propagation
/ k trχ′ , ∇
equations 16.113:
/ k trχ′ = f · ∇
D∇ / k χ̂′ + rk
/ k trχ′ + g · ∇
/ k χ̂′ = h · ∇
D∇ / k χ̂′ + sk
/ k trχ′ + i · ∇ (16.121)
The coefficients f , g, h, i are given by 4.102, and we have:
rk ≃ 0
/ kα
sk ≃ −∇ (16.122)
Here, we denote by ≃ equality up to lower order terms involving the ∇ / derivatives
of trχ′ , χ̂′ and /d log Ω = (1/2)(η + η) of order up to k − 1.
We apply Lemma 4.7 to the above propagation equations taking p = 2. In
the case of the first of 16.121 we have ∇ / k χ̂′ + rk in
/ k trχ′ in the role of θ, g · ∇
the role of ξ, r = k, ν = −2 and γ = 0. In the case of the second of 16.121 we
have ∇/ k χ̂′ in the role of θ, h · ∇
/ k trχ′ + sk in the role of ξ, r = k + 2, ν = 0 and
γ = i. We obtain:
/ k trχ′ kL2 (Su,u ) ≤ C|u0 |k+1 k∇
|u|k+1 k∇ / k trχ′ kL2 (Su,u0 )
Z u
+C / k χ̂′ + rk kL2 (Su,u′ ) du′
|u′ |k+1 kg · ∇
u0
|u|k+1
/ χ̂ kL2 (Su,u ) ≤ C|u0 |k+1 k∇
k∇ k ′
/ k χ̂′ kL2 (Su,u0 )
Z u
+C / k trχ′ + sk kL2 (Su,u′ )(16.123)
|u′ |k+1 kh · ∇
u0
549
In view of the estimates of Chapter 3 these imply:
/ k trχ′ kL2 (Su,u ) ≤ C ′ k∇

k∇ / k trχ′ kL2 (Su,u0 )
Z u Z u
+C ′ k∇/ k χ̂′ kL2 (Su,u′ ) du′ + C ′ krk kL2 (Su,u′ ) du′
u0 u0
/ k χ̂′ kL2 (Su,u ) ≤ C ′ k∇
k∇ k ′
/ χ̂ kL2 (Su,u0 )
Z u Z u
+C ′ / k trχ′ kL2 (Su,u′ ) du′ + C ′
k∇ ksk kL2 (Su,u′ ) du′
u0 u0
(16.124)
Setting u = 12 (t + λ), u = 21 (t − λ) in each of 16.124, squaring, integrating

with respect to λ on (−t, λ∗ (t)), and taking the square root, we obtain, by 16.84
and 16.83,
/ k trχ′ kL2 (Ht′ ) ≤ C ′ k∇

k∇ / k trχ′ kL2 (Cu0 )
Z t Z t
′ k ′ ′ ′
+C / χ̂ kL2 (H ′′ ) dt + C
k∇ kr k kL2 (H ′′ ) dt′
t t
u0 u0
k ′ ′ k ′
/ χ̂ kL2 (Ht′ ) ≤ C k∇
k∇ / χ̂ kL2 (Cu0 )
Z t Z t
′ k ′ ′ ′
+C k∇
/ trχ kL (H ′ ) dt + C
2 ′ ksk kL2 (H ′′ ) dt′
t t
u0 u0
(16.125)
quantity
/ k χ̂′ kL2 (Ht′ )
/ k trχ′ kL2 (Ht′ ) + k∇
k∇
which implies:
n o
/ k χ̂′ kL2 (Ht′ ) ≤ C ′′ k∇
/ k trχ′ kL2 (Ht′ ) + k∇
k∇ / k trχ′ kL2 (Cu0 ) + k∇ / k χ̂′ kL2 (Cu0 )
Z tn o
+C ′′ kr k kL2 (H ′′ ) + ksk kL2 (H ′′ ) dt′
t t
u0
(16.126)
Now by 16.122 the integral

Z t
kr k kL2 (H ′′ ) dt′ (16.127)
t
u0
is bounded in terms of the supremum for t′ ∈ (u0 , t] of the sum of the L2 norms
of the derivatives of the connection coefficients of order up to k−1, and the sharp
already obtained. Also, the integral
Z
ksk kL2 (H ′′ ) dt′ (16.128)
t
u0
550
is bounded by the integral
Z
/ k αkL2 (H ′′ ) dt′
k∇ (16.129)
t
u0
of order up to k − 1, and the sharp bounds for the derivatives of up to the 2nd
order of the connection coefficients already obtained. Moreover, according to
what has previously
q been demonstrated, the integral 16.129 is bounded by a
′
constant times E k (t) plus a lower order term which is bounded in terms of
q
′
E k′ (t): for k ′ = 3, ..., k − 1, the supremum for t′ ∈ (u0 , t] of the sum of the L2
norms of the derivatives of the connection coefficients of order up to k − 1, and
the sharp bounds for the derivatives of up to the 2nd order of the connection
coefficients already obtained. We have thus shown what was required also in
the case of ∇ / k χ̂.
/ k trχ, ∇
k
We turn to ∇ / η,∇ / k η. These satisfy propagation equations deduced by ap-
plying ∇ / k−1 to the propagation equations 4.117 and using Lemma 4.1. We
obtain:
/ kη = a · ∇
D∇ / k η + bk
/ kη = a · ∇
D∇ / k η + bk (16.130)
The coefficients a, a are given by 4.118 and we have:
/ k β + Ω(∇
bk ≃ −Ω∇ / k−1 DΓ
/ k χ) · η − (∇ /) · η
bk ≃ Ω∇ / k χ) · η − (∇
/ k β + Ω(∇ / k−1 DΓ) · η (16.131)
Here we denote by ≃ equality up to lower order terms involving the ∇ / derivatives

of χ, χ, η, η of order up to k − 1 as well as the ∇ / derivatives of β, β of order up
to k − 1.
To the first of the propagation equations 16.130 we apply Lemma 4.6 taking
p = 2, with ∇/ k η in the role of θ and a · ∇
/ k η + bk in the role of ξ. Here r = k + 1,
ν = 0, γ = 0 and we obtain:
Z u
k
k∇/ ηkL2 (Su,u ) ≤ C / k η + bk kL2 (Su′ ,u ) du′
ka · ∇ (16.132)
0
To the second of the propagation equations 16.130 we apply Lemma 4.7 taking
/ k η in the role of θ and a · ∇
p = 2, with ∇ / k η + bk in the role of ξ. Here r = k + 1,
ν = 0, γ = 0 and we obtain:
|u|k k∇
/ k ηkL2 (Su,u ) ≤ C|u0 |k k∇/ k ηkL2 (Su,u0 )
Z u
+C / k η + bk kL2 (Su,u′ ) du′ (16.133)
|u′ |k ka · ∇
u0
551
In view of the estimates of Chapter 3, 16.132 and 16.133 imply:
Z u Z u
k ′ k ′ ′
k∇/ ηkL2 (Su,u ) ≤ C k∇
/ ηkL2 (Su′ ,u ) du + C kbk kL2 (Su′ ,u ) du′
0 0
/ k ηkL2 (Su,u ) ≤ C ′ k∇
k∇ / k ηkL2 (Su,u0 )
Z u Z u
+C ′ k∇/ k ηkL2 (Su,u′ ) du′ + C ′ kbk kL2 (Su,u′ ) du′
u0 u0
(16.134)
Setting u = 21 (t + λ), u = 12 (t − λ) in each of 16.124, squaring, integrating with

respect to λ on (−t, λ∗ (t)), and taking the square root, we obtain, by 16.85,
16.84 and 16.83,
Z t Z t
k ′ k ′ ′
k∇
/ ηkL2 (Ht′ ) ≤ C k∇
/ ηkL2 (H ′′ ) dt + C kbk kL2 (H ′′ ) dt′
t t
u0 u0
k ′ k
/ ηkL2 (Ht′ ) ≤ C k∇
k∇ / ηkL2 (Cu0 )
Z t Z t
+C ′ / k ηkL2 (H ′′ ) dt′ + C ′
k∇ kbk kL2 (H ′′ ) dt′
t t
u0 u0
(16.135)
quantity
/ k ηkL2 (Ht′ ) + k∇
k∇ / k ηkL2 (Ht′ )
which implies:
/ k ηkL2 (Ht′ ) + k∇
k∇ / k ηkL2 (Ht′ ) ≤ C ′′ k∇
/ k ηkL2 (Cu0 )
Z tn o
+C ′′ kbk kL2 (H ′′ ) + kbk kL2 (H ′′ ) dt′
t t
u0
(16.136)
Now, in view of 16.131, the integral

Z tn o
kbk kL2 (H ′′ ) + kbk kL2 (H ′′ ) dt′ (16.137)
t t
u0
is bounded by
Z t n o
C / k βkL2 (H ′′ ) + k∇
k∇ / k βkL2 (H ′′ ) dt′
t t
u0
Z t n o
+C ′ / k χkL2 (H ′′ ) + k∇
k∇ / k χkL2 (H ′′ ) dt′ (16.138)
t t
u0
552
and of the curvature components of order up to k − 1, and the sharp bounds
for the derivatives of up to the 2nd order of the connection coefficients and of
the curvature components already obtained. Moreover, according to what has
previously been demonstrated,
q the first of the integrals 16.138 is bounded by
′
a constant times E k (t) plus a lower order term which is bounded in terms
q
′
of E k′ (t): for k ′ = 3, ..., k − 1, the supremum for t′ ∈ (u0 , t] of the sum of
the L2 norms of the derivatives of the connection coefficients of order up to
k − 1, and the sharp bounds for the derivatives of up to the 2nd order of the
connection coefficients already obtained. Also, the second of the integrals 16.138
has already been appropriately estimated above. We have thus shown what was
required also in the case of ∇ / k η, ∇
/ k η.
l m
We turn to ∇/ D ω : l + m = k. Applying ∇ / k to the propagation equa-
tion 1.108 and using Lemmas 1.2 and 4.1 we obtain the following propagation
/ k ω:
equation for ∇
D∇/ k ω = nk,0 (16.139)
where
/ k η) + 2(η − η, ∇
nk,0 ≃ Ω2 2(η, ∇ / k η) − ∇
/ kρ

(16.140)
and we denote by ≃ equality up to lower order terms involving the ∇ / derivarives
of χ, η, η and ρ of order up to k−1. For n ≥ 1, we apply Dm−1 to the propagation
equation 4.184. In view of the commutation formula 1.91 we obtain:
DDm ω = n0,m (16.141)
where the functions n0,m satisfy the recursion relation:
n0,m = Dn0,m−1 + 2Ω2 (η − η)♯ · /dDm−1 ω (16.142)
and we have:
n0,0 = Ω2 {2(η, η) − |η|2 − ρ} (16.143)
Therefore the functions n0,m are given by:
m−1
X n o
′ ′
n0,m = Dm n0,0 + Dm 2Ω2 (η − η)♯ · /dDm−1−m ω (16.144)
m′ =0
It follows that:
n0,m ≃ Ω2 2(η, Dm η) + 2(η − η, Dm η) − Dm ρ + 2mΩ2 (η − η)♯ · /dDm−1 ω

(16.145)
where we denote by ≃ equality up to lower order terms involving the D and
∇
/ derivatives of the connection coefficients and of the curvature components of
order up to m−1. Moreover, applying Dm−1 to equations 1.82, 1.174 we obtain:
Dm η ≃ 0
/Dm−1 ω
Dm η ≃ 2d (16.146)
553
We conclude that:
n0,m ≃ 2(m + 2)Ω2 (η − η)♯ · /dDm−1 ω − Ω2 Dm ρ (16.147)
/ l , for l + m = k, to the propagation equation 16.141. Using

We now apply ∇
Lemmas 1.2 and 4.1 we then deduce the following propagation equation for
/ l Dm ω:
∇
D∇/ l Dm ω = nl,m (16.148)
where
/ l n0,m ≃ 2(m + 2)Ω2 (η − η)♯ · ∇

nl,m ≃ ∇ / l+1 Dm−1 ω − Ω2 ∇
/ l Dm ρ (16.149)
and we denote by ≃ equality up to lower order terms involving the D and ∇ /

derivatives of the connection coefficients and of the curvature components of
order up to k − 1.
To the propagation equation 16.139 we apply Lemma 4.7 taking p = 2. Here
r = k, ν = 0, γ = 0 and we obtain, in view of the fact that ∇ / k ω vanishes on
Cu0 ,
Z u
k−1 k
|u| / ωkL2 (Su,u ) ≤ C
k∇ |u′ |k−1 knk,0 kL2 (Su,u′ ) du′ (16.150)
u0
This implies: Z u
/ k ωkL2 (Su,u ) ≤ C ′
k∇ knk,0 kL2 (Su,u′ ) du′ (16.151)
u0
Setting u = 12 (t + λ), u = 12 (t − λ), squaring, integrating with respect to λ on

(−t, λ∗ (t)), and taking the square root, we obtain, by 16.84 and 16.83,
Z t
k ′
k∇
/ ωk L2 (Ht′ ) ≤C knk,0 kL2 (H ′′ ) dt′ (16.152)
t
u0
Now, in view of 16.140 the integral on the right in 16.152 is bounded by

Z t Z tn o
k ′ ′
C k∇/ ρkL2 (H ′ ) dt + C
′ / k ηkL2 (H ′′ ) + k∇
k∇ / k ηkL2 (H ′′ ) dt′ (16.153)
t t t
u0 u0
(u0 , t] of the sum of the L2 norms of the ∇/ derivatives of the connection coef-
ficients and of the curvature components of order up to k − 1, and the sharp
and of the curvature components already obtained. Moreover, according to what
has previously been q demonstrated, the first of the integrals 16.152 is bounded
′
by a constant times E k (t) plus a lower order term which is bounded in terms
q
′
of E k′ (t): for k ′ = 3, ..., k − 1, the supremum for t′ ∈ (u0 , t] of the sum of the
L2 norms of the derivatives of the connection coefficients of order up to k − 1,
554
and the sharp bounds for the derivatives of up to the 2nd order of the connec-
tion coefficients already obtained. Also, the second of the integrals 16.152 has
already been appropriately estimated above. We have thus shown what was
required in the case of ∇/ k ω.
We proceed to derive the appropriate estimates for ∇ / l Dm ω : l + m = k
for m = 1, ..., k by induction on m. Suppose then that ∇ / l+1 Dm−1 ω has been
appropriately estimated. We apply Lemma 4.7 to the propagation equation
16.148, taking p = 2. Here r = l, ν = 0, γ = 0 and we obtain, in view of the
fact that ∇/ l Dm ω vanishes on Cu0 ,
Z u
|u|l−1 k∇/ l Dm ωkL2 (Su,u ) ≤ C |u′ |l−1 knl,m kL2 (Su,u′ ) du′ (16.154)
u0
This implies:
Z u
/ l Dm ωkL2 (Su,u ) ≤ C ′
k∇ knl,m kL2 (Su,u′ ) du′ (16.155)
u0

Z t
k∇/ l Dm ωkL2 (Ht′ ) ≤ C ′ knl,m kL2 (H ′′ ) dt′ (16.156)
t
u0

Z t Z t
C / l Dm ρkL2 (H ′′ ) dt′ + C ′
k∇ / l+1 Dm−1 ωkL2 (H ′′ ) dt′
k∇ (16.157)
t t
u0 u0
(u0 , t] of the sum of the L2 norms of the ∇
/ and D derivatives of the connection
coefficients and of the curvature components of order up to k − 1, and the sharp
and of the curvature components already obtained. By the inductive hypothesis
the second of the integrals 16.157 has been appropriately estimated. Moreover,
according to what has previously beenqdemonstrated, the first of the integrals
′
16.157 is bounded by a constant times E k (t) plus a lower order term which is
q
′
bounded in terms of E k′ (t): for k ′ = 3, ..., k − 1, the supremum for t′ ∈ (u0 , t]
of the sum of the L2 norms of the derivatives of the connection coefficients of
order up to k − 1, and the sharp bounds for the derivatives of up to the 2nd
order of the connection coefficients already obtained. The inductive step is thus
complete and we have established the required estimates for ∇ / l Dm ω, for all
l + m = k.
The required estimates for ∇/ l Dm ω are deduced in a similar manner. Indeed
/ k ω satisfies the conjugate of the propagation equation 16.139:
∇
/ k ω = nk,0
D∇ (16.158)
555
where
nk,0 ≃ Ω2 2(η, ∇
/ k η) + 2(η − η, ∇
/ k η) − ∇
/ kρ

(16.159)
and we denote by ≃ equality up to lower order terms involving the ∇
/ derivarives
of χ, η, η and ρ of order up to k − 1. Also ∇/ l Dm ω : l + m = k, for m ≥ 1,
satisfies the conjugate of the propagation equation 16.148:
/ l Dm ω = nl,m
D∇ (16.160)
where
/ l+1 Dm−1 ω − Ω2 ∇
nl,m ≃ 2(m + 2)Ω2 (η − η)♯ · ∇ / l Dm ρ (16.161)
and we denote by ≃ equality up to lower order terms involving the D and ∇ /

derivatives of the connection coefficients and of the curvature components of
order up to k − 1.
Noting that ∇ / k ω vanishes on C 0 , we apply Lemma 4.6 to the propagation
equation 16.158, taking p = 2 . Here r = k, ν = 0, γ = 0 and we obtain:
Z u
k∇/ k ωkL2 (Su,u ) ≤ C knk,0 kL2 (Su′ ,u ) du′ (16.162)
0

Z t
/ k ωkL2 (Ht′ ) ≤ C ′
k∇ knk,0 kL2 (H ′′ ) dt′ (16.163)
t
u0

Z t Z tn o
C k∇/ k ρkL2 (H ′′ ) dt′ + C ′ k∇ / k ηkL2 (H ′′ ) dt′ (16.164)
/ k ηkL2 (H ′′ ) + k∇
t t t
u0 u0
which has already been appropriately estimated above. We have thus shown
what was required in the case of ∇ / k ω.
Finally, we derive the appropriate estimates for ∇ / l Dm ω : l + m = k
for m = 1, ..., k by induction on m. Suppose then that ∇ / l+1 D m−1 ω has been
l m
appropriately estimated. Noting that ∇ / D ω vanishes on C 0 , we apply Lemma
4.6 to the propagation equation 16.148, taking p = 2. Here r = l, ν = 0, γ = 0
and we obtain:
Z u
/ l D m ωkL2 (Su,u ) ≤ C
k∇ knl,m kL2 (Su′ ,u ) du′ (16.165)
0

Z t
/ l Dm ωkL2 (Ht′ ) ≤ C ′
k∇ knl,m kL2 (H ′′ ) dt′ (16.166)
t
u0
556
Z t Z t
C / l Dm ρkL2 (H ′′ ) dt′ + C ′
k∇ / l+1 Dm−1 ωkL2 (H ′′ ) dt′
k∇ (16.167)
t t
u0 u0
(u0 , t] of the sum of the L2 norms of the ∇
/ and D derivatives of the connection
coefficients and of the curvature components of order up to k − 1, and the sharp
and of the curvature components already obtained. By the inductive hypothesis
the second of the integrals 16.167 has been appropriately estimated. Moreover,
according to what has previously beenqdemonstrated, the first of the integrals
′
16.167 is bounded by a constant times E k (t) plus a lower order term which is
q
′
bounded in terms of E k′ (t): for k ′ = 3, ..., k − 1, the supremum for t′ ∈ (u0 , t]
of the sum of the L2 norms of the derivatives of the connection coefficients of
order up to k − 1, and the sharp bounds for the derivatives of up to the 2nd
order of the connection coefficients already obtained. The inductive step is thus
complete and we have established the required estimates for ∇ / l Dm ω, for all
l + m = k, as well.
Having established appropriate estimates for ∇ / k χ, ∇
/ k χ, ∇
/ k η, ∇
/ k η, and
l m l m
∇
/ D ω : l + m = k, ∇ / D ω : l + m = k, appropriate estimates for
all D, D and ∇ / derivatives of order k of all connection coefficients follow in
a straightforward manner from equations 1.54, 1.59, 1.82, 1.83, 1.107, 1.108,
1.170, 1.171, 1.173, 1.174.
We are now ready to derive bounds for the energies Ek′ (t) which are uniform
in t ∈ (u0 , c∗ ), for any k ≥ 3. The energy inequality 16.38 reads, in the case of
the energy-momentum vectorfield
X
Pk = Pm,n (16.168)
m+n=k
(see 16.55), Z
Ek′ (t) ≤ Ek (u0 ) + |τk |dµg (16.169)
Mt′
where X
τk = τm,n (16.170)
m+n=k
and τm,n is given by 16.60. In view of the fact that dµg , the volume form of
(Mc′∗ , g) is given in terms of dµg , the volume form of (Ht′ , g) by:
dµg = Ωdt ∧ dµg (16.171)
we have: Z Z t
|τk |dµg ≤ C kτk kL1 (H ′′ ) dt′ (16.172)
t
Mt′ u0
557
Starting from the sharp bounds for up to the 2nd derivatives of the connection
coefficients and of the curvature components on Mc′∗ which we have previously
obtained, we shall show by induction that the Ek′ (t) are uniformly bounded in
t ∈ (u0 , c∗ ) for any k ≥ 3. The inductive hypothesis for k ≥ 4 is that the
Ek′ ′ (t) are uniformly bounded in t ∈ (u0 , c∗ ) for k ′ = 3, ..., k − 1, and there is no
hypothesis for k = 3. In establishing the inductive step we allow the constants to
depend not only on δ, u0 and the ininial data quantities, but also on the uniform
bounds in t ∈ (u0 , c∗ ) for the Ek′ ′ (t) for k ′ = 3, ..., k − 1 provided by the inductive
hypothesis.
Now by the L∞ bounds on the components of (T ) π̃ which follow from the
∞
L bounds for the connection cefficients, the multiplier part of τm,n is bounded
in L1 (Ht′ ) by:
′
CEm,n (t) (16.173)
while the Weyl current part of τm,n , given by 16.61 is bounded in L1 (Ht′ ) by:
q
C Em,n′ (t)kJm,n kL2 (Ht′ ) (16.174)
where we denote by kJm,n kL2 (Ht′ ) the square root of the sum of the squares of
the L2 norms on Ht′ of the components of Jm,n (in the null decomposition).
According to the discussion following the recursion relations 16.62, 16.63, the
leading terms in the expression for any component of Jm,n are of four types. All
four types are bilinear expressions, with coefficients depending only on g/ and /ǫ ,
the first factor of which is a derivative of a connection coefficient of order k1 ,
and the second factor a derivative of a curvature component of order k2 , where
k1 + k2 = k. In the case k1 = 0, k2 = k, we place the first factor in L∞ (Ht′ )
using the results of Chapter 3, and the p second factor in L2 (Ht′ ) the resulting
quantity being bounded in terms of Ek′ (t). In the case k1 = k, k2 = 0 we
place the first factor in L2 (Ht′ ) using the estimates, just obtained, forq the kth
′
order derivatives of the connection coefficients in L2 (Ht′ ) in terms of E k (t),
and the second factor in L∞ (Ht′ ). In the case k1 = 1, k2 = k − 1, we place the
first factor in L4 (S 12 (t+λ), 12 (t−λ) ) using the results of Chapter 4, and the second
factor also in L4 (S 12 (t+λ), 12 (t−λ) ) the resulting quantity being bounded in terms
of Ek′ (t) by virtue of the Sobolev inequality on Ht′ (the analogue of the second
p
statement of Proposition 10.1). In the case k1 = k − 1, k2 = 1, we place the

first factor in L4 (S 12 (t+λ), 12 (t−λ) ) the resulting quantity being bounded in terms
of the L2 norm on Ht′ of the kth order derivatives of the connection coefficients
(by virtue ofqthe Sobolev inequality just mentioned), which is in turn bounded
′
in terms of E k (t), and the second factor also in L4 (S 12 (t+λ), 12 (t−λ) ). Finally,
in the remaining cases 2 ≤ k1 , k2 ≤ k − 2 (which exist only for k ≥ 4) we may
place both factors in L4 (S 12 (t+λ), 12 (t−λ) ), the resulting quantities being bounded
by constants in the above sense. The contributions of lower order terms in Jm,n
to kJm,n kL2 (Ht′ ) are similarly bounded by constants in the above sense. We thus
558
arrive at the following conclusion:
q
′
kJm,n kL2 (Ht′ ) ≤ C E k (t) + C ′ (16.175)
Substituting in 16.174 and combining with 16.173 we conclude that:

′
kτk kL1 (Ht′ ) ≤ CE k (t) + C ′ (16.176)
(for different constants C and C ′ ). Substituting in 16.172 and 16.169 then yields:
Z t
′
Ek′ (t) ≤ Ek (u0 ) + C E k (t′ )dt′ + C ′ : ∀t ∈ (u0 , c∗ ) (16.177)
u0
Since the right hand side is a nondecreasing function of t, this implies that also:
Z t
′
Ek′ (t′ ) ≤ Ek (u0 ) + C E (t′ )dt′ + C ′ : ∀t′ ∈ (u0 , t], ∀t ∈ (u0 , c∗ ) (16.178)
u0
Taking the supremum with respect to t′ ∈ (u0 , t] then yields the following linear
′
integral inequality for E k :
Z t
′ ′
E k (t) ≤ Ek (u0 ) + C E k (t′ )dt′ + C ′ (16.179)
u0
This implies:
′
E k (t) ≤ CEk (u0 ) + C ′ (16.180)
′
(for different constants C and C ), which shows that Ek′
is uniformly bounded
on (u0 , c∗ ) completing the inductive step.
Having establised that the Ek′ are uniformly bounded in (u0 , c∗ ) for all k,
it follows that the L2 norms on Ht′ of all the (D, D and ∇ /) derivatives of any
order of all the curvature components and of all the connection coefficients
are uniformly bounded in t ∈ (u0 , c∗ ). By the Sobolev inequality on the Ht′
(the analogue of the second statement of Proposition 10.1), it then follows that
the L4 norms on Su,u of all the derivatives of any order of all the curvature
components and of all the connection coefficients are uniformly bounded in
(u, u) ∈ Dc′ ∗ . Then by Lemma 5.2 with p = 4 all the derivatives of any order of
all the curvature components and of all the connection coefficients are uniformly
bounded in Mc′∗ .
16.3 Completion of the continuity argument

In this section we shall show that the solution extends as a smooth solution to
a larger domain Mc for some c > c∗ and the three conditions in the statement
of Theorem 12.1 hold for the extended solution as well, thereby contradicting
the definition of c∗ unless c∗ = −1. The proof of Theorem 12.1 will then be
complete.
559
It follows from what has just been established that the metric components
in a canonical coordinate system extend smoothly to the future boundary Hc∗
of Mc∗ as do the connection coefficients and curvature components. Thus Hc∗
is endowed with a smooth induced metric g (see 16.29) and a smooth 2nd fun-
damental form k. We denote
[ ′ [
M c∗ = M c∗ Hc ∗ , M c∗ = Mc′∗ Hc′ ∗ (16.181)
and:
2δ − c∗ : for c∗ ∈ (u0 + δ, −1)

∗ ∗
λ (c ) = sup λ = (16.182)
Hc′ ∗ c∗ − 2u0 : for c∗ ∈ (u0 , u0 + δ]
(see 16.27, 16.64). In the Minkowskian region M0 , which corresponds to u ≤ 0, g

coincides with the Euclidean metric e in the same (λ, ϑA : A = 1, 2) coordinates
and k vanishes. The 2nd fundamental form of the Ht , in particular of Hc∗ , is
given by:
k(X, Y ) = g(∇X T̂ , Y ) (16.183)
where X, Y is any pair of vectors tangent to Ht at a point. Let us supplement the
frame field (eA = ∂/∂ϑA : A = 1, 2) for the S 12 (t+λ), 12 (t−λ) with the vectorfield
N̂ = Ω−1 N (16.184)
the outward unit normal to the S 21 (t+λ), 12 (t−λ) , to a frame field for Ht . From
16.111 and 1.197 we have:

1 ∂ 1 ∂
N̂ = (e4 − e3 ) = Ω−1 + bA A (16.185)
2 ∂λ 2 ∂ϑ
From 16.31, 16.33 and the table 1.175 we then find, for the components of k in
the frame (N̂ , eA : A = 1, 2):
1 −1
k(N̂ , N̂ ) = Ω (ω + ω) = T̂ log Ω
2
k(eA , N̂ ) = −ζA
1
k(eA , eB ) = (χAB + χAB ) (16.186)
2
Therefore k is given in canonical coordinates (λ, ϑA : A = 1, 2) on Ht by:
1
k= Ω(ω + ω)dλ ⊗ dλ
2
1 1
−ΩζA (dϑA − bA dλ) ⊗ dλ − ΩζA dλ ⊗ (dϑA − bA dλ)
2 2
1 1 1
+ (χAB + χAB )(dϑA − bA dλ) ⊗ (dϑB − bB dλ) (16.187)
2 2 2
Now by virtue of the definition of a canonical coordinate system, the metric
g/AB (u, u)dϑA ⊗ dϑB is precisely the pullback metric (Φu−u0 ◦ Φu )∗ g/|Su,u on
560
◦ ◦
S0,u0 , while the metric g/|S0,u = |u0 |2 g/ where g/ is the standard metric on S 2 .
0
Then by Lemma 11.1, λ(u, u) and Λ(u, u), respectively the smallest and largest
◦
eigenvalues of |u|−2 g/AB (u, u) relative to g/AB , satisfy:
1
≤ λ(u, u) ≤ Λ(u, u) ≤ 4 (16.188)
4
′
for all (u, u) ∈ Dc′ ∗ , hence by continuity also for all (u, u) ∈ Dc∗ , where
′ [
Dc∗ = Dc′ ∗ {(u, u) : u + u = c∗ , u ∈ [0, u∗ (c∗ ))} (16.189)
where for any t ∈ (u0 , c∗ ] we denote:
: for t ∈ (u0 + δ, c∗ )

∗ δ
u (t) = sup u = (16.190)
Ht t − u0 : for t ∈ (u0 , u0 + δ]
The left inequality in 16.187 implies that the null hypersurfaces C u do not
contain focal points on Hc′ ∗ , the future boundary of Mc′∗ , either. Let now
λ′ (u, u) and Λ′ (u, u) be respectively the smallest and largest eigevalues of the
◦
pullback metric Φ∗u g/|Su,u on S0,u with respect to the metric g/|S0,u = |u|2 g/.
Then by Lemma 5.3, taking δ suitably small, we have:
1
λ′ (u, u) ≥ , Λ′ (u, u) ≤ 2 (16.191)
2
′
for all (u, u) ∈ Dc′ ∗ , hence by continuity also for all (u, u) ∈ Dc∗ . The left
inequality implies that the null hypersurfaces Cu do not contain conjugate points
on Hc′ ∗ either.
Now the induced metric g on Hc′ ∗ , given by 16.29, is equivalent to the Eu-
clidean metric e on Hc′ ∗ :
◦
e = dλ ⊗ dλ + λ2 g/AB dϑA ⊗ dϑB (16.192)
T
and, as mentioned above, g = e on Hc∗ M0 . It follows that the distance
relative to e between any pair of points on Hc∗ is bounded from above by a
constant multiple of the distance between the same points relative to g, and
conversely. Hence, if p′ and p′′ are points on Su,c∗ −u , u ∈ [0, u∗ (c∗ )), which
correspond to distinct points ϑ′ and ϑ′′ on S 2 , then their distance relative to
g is positive. As a consequence, the null hypersurfaces C u do not contain cut
points on Hc′ ∗ , the future boundary of Mc′∗ either. The generators of the Cu are
the integral curves of L, which is given in canonical coordinates by 1.197. Thus,
the generator of Cu initiating at a point q ∈ S0,u corresponding to the point
ϑ0 ∈ S 2 represented by the coordinates (ϑ10 , ϑ20 ), is represented in canonical
coordinates by:
u 7→ (u, u, ϑ(u; ϑ0 ))
561
where u 7→ ϑ(u; ϑ0 ) is the solution of:
dϑA
= bA (u, u, ϑ), ϑA (0; ϑ) = ϑA
0 (16.193)
du
a non-autonomous system of ordinary differential equations on S 2 . The vector-

field
∂
b = bA A (16.194)
∂ϑ
being smooth on M c∗ , it follows that ϑ′0 6= ϑ′′0 implies ϑ(u; ϑ′0 ) 6= ϑ(u; ϑ′′0 ) for
all u ∈ [0, c∗ − u], if u > c∗ − δ, for all u ∈ [0, δ), if u ≤ c∗ − δ, in the case
c∗ ≥ u0 + δ, for all u ∈ [0, c∗ − u], in the case c∗ < u0 + δ. Consequently, the
null hypersurfaces Cu do not contain cut points on Hc′ ∗ either. Moreover, Hc′ ∗
is smoothly foliated by the {S 12 (c∗ +λ), 12 (c∗ −λ) : λ ∈ (−c∗ , λ∗ (c∗ ))}, the lapse
function Ω (see 16.184) being bounded above and below by positive constants.
We now define “Cartesian” coordinates x1 , x2 , x3 on Hc∗ as follows. In the
Minkowskian region M0 these coincide with the Cartesian coordinates x1 , x2 , x3
previously defined. To extend these coordinates to Hc′ ∗ we simply stipulate that
they are given in terms of the “polar” coordinates λ, ϑ1 , ϑ2 by the same formulas
as in the Minkowskian region M0 . That is, the xi : i = 1, 2, 3 are given in the
north polar chart by:
λϑ1 λϑ2 1 − 41 |ϑ|2

1 2 3
x = , x = , x =λ (16.195)
1 + 41 |ϑ|2 1 + 14 |ϑ|2 1 + 41 |ϑ|2
and in the south polar chart by:
λϑ1 λϑ2 1 − 14 |ϑ|2

1 2 3
x = , x = , x = −λ (16.196)
1 + 14 |ϑ|2 1 + 41 |ϑ|2 1 + 14 |ϑ|2
the inverse of the transformation 16.195 being:
2x1 2x2
λ = |x|, ϑ1 = , ϑ2 = (16.197)
|x| + x3 |x| + x3
and the inverse of the transformation 16.196 being:
2x1 2x2
λ = |x|, ϑ1 = , ϑ2 = (16.198)
|x| − x3 |x| − x3
(see 1.208, 1.209) where

p
|x| = (x1 )2 + (x2 )2 + (x3 )2
In the north polar chart we have:
∂λ ϑA
= : A = 1, 2
∂xA 1 + 14 |ϑ|2
562
∂λ 1 − 14 |ϑ|2
=
∂x3 1 + 41 |ϑ|2
∂ϑA

1 1
λ B = 1 + |ϑ|2 δAB − ϑA ϑB : A, B = 1, 2
∂x 4 2
∂ϑA
λ = −ϑA : A = 1, 2 (16.199)
∂x3
and in the south polar chart we have:
∂λ ϑA
= : A = 1, 2
∂xA 1 + 14 |ϑ|2
∂λ 1 − 14 |ϑ|2
=−
∂x 3 1 + 41 |ϑ|2
∂ϑA

1 2 1
λ B = 1 + |ϑ| δAB − ϑA ϑB : A, B = 1, 2
∂x 4 2
∂ϑA
λ = ϑA : A = 1, 2 (16.200)
∂x3
Note that the right hand sides of 16.199, 16.200 are analytic functions of ϑ =
(ϑ1 , ϑ2 ) alone.
Let us write:
g =e+h (16.201)
The components eij of the Euclidean metric in Cartesian coordinates being
simply δij , we have:
∂λ ∂λ 2
◦ ∂ϑA ∂ϑB
eij = + λ g
/ AB = δij (16.202)
∂xi ∂xj ∂xi ∂xj
Therefore, the components gij of the metric g in Cartesian coordinates are given
by:
g ij = δij + hij (16.203)
Here hij are the components of the tensorfield h in Cartesian coordinates:

2 1 2 ∂λ ∂λ
hij = Ω − 1 + |b|
4 ∂xi ∂xj
∂λ ∂ϑA ∂ϑA ∂λ

1
− bA +
2 ∂xi ∂xj ∂xi ∂xj
◦ ∂ϑA ∂ϑB
+(g/AB − λ2 g/AB ) (16.204)
∂xi ∂xj
where:
bA = g/AB bB (16.205)
Letting again λ(u, u) and Λ(u, u) be respectively the smallest and largest
◦
eigenvalues of |u|−2 g/AB (u, u) relative to g/AB , following the proof of Lemma
563
11.1, and using the estimates of Chapter 3 in place of the bootstrap assumptions
A1.1, A1.2, A3.1, A3.2, we now obtain:
λ(u, u) ≥ 1 − O(δ 1/2 |u|−1 ), Λ(u, u) ≤ 1 + O(δ 1/2 |u|−1 ) (16.206)
It follows that:
◦ p
||u|−2 g/(u, u)− g/ |◦ = (λ(u, u) − 1)2 + (Λ(u, u) − 1)2
/
g
′
≤ O(δ 1/2 |u|−1 ) : ∀(u, u) ∈ Dc∗ (16.207)
Also, the L∞ bound on ζ from Chapter 3 yields through equation 1.199 the
following L∞ bound on b:
′
kbkL∞ (Su,u ) ≤ O(δ 1/2 )|u|−1 ) : ∀(u, u) ∈ Dc∗ (16.208)
The bounds 16.207, 16.208, together with the bound on log Ω for Chapter 3, yield
through 16.204, in view of the formulas 16.199, 16.200, the following bound for
the functions hij on Hc′ ∗ :
sup |hij | ≤ O(δ 1/2 |c∗ |−1 ) : i, j = 1, 2, 3 (16.209)

Hc′ ∗
The components of the tensorfield k in Cartesian coordinates are given by:

1 ∂λ ∂λ
kij = Ω(ω + ω) i j (16.210)
2 ∂x ∂x
A
∂ϑ 1 ∂λ ∂λ
−ΩζA − bA i
∂xi 2 ∂x ∂xj
∂λ ∂ϑA

1 A ∂λ
−ΩζA i − b
∂x ∂xj 2 ∂xj
A A
1 ∂ϑ 1 A ∂λ ∂ϑ 1 A ∂λ
+ (χAB + χAB ) − b − b
2 ∂xi 2 ∂xi ∂xj 2 ∂xj
The estimates of Chapter 3 together with the bound 16.208, imply, in view of
the formulas 16.199, 16.200, the following bound for the functions kij on Hc′ ∗ :
sup |kij | ≤ O(δ −1/2 |c∗ |−1 ) : i, j = 1, 2, 3 (16.211)

Hc′ ∗
The functions hij and kij , considered as functions of the Cartesian coordi-
nates x1 , x2 , x3 , are smooth functions on the open ball Bλ∗ (c∗ ) of radius λ∗ (c∗ )
in ℜ3 extending to smooth functions on the closed ball B λ∗ (c∗ ) , their derivatives
of all orders being bounded. We apply to these functions a radial total exten-
sion operator E for Bλ∗ (c∗ ) . This yields smooth functions Ehij and Ekij on the
closed ball B λ∗ (c∗ )+1 , and there is a numerical constant C such that:
sup |Ehij | ≤ C sup |hij |, sup |Ekij | ≤ C sup |kij | (16.212)

B λ∗ (c∗ )+1 Bλ∗ (c∗ ) B λ∗ (c∗ )+1 Bλ∗ (c∗ )
564
In particular, the bound 16.209 implies:
sup |Ehij | ≤ O(δ 1/2 ) (16.213)

B λ∗ (c∗ )+1
(n)
hence, if δ is suitably small (depending on the quantities D : n = 0, 1, 2, 3;
′4
D[1] (α) and the quantities D0∞ , D
/41 , D
/42 (trχ), D
/3 (trχ)) the tensorfield with com-
ponents:
δij + Ehij (16.214)
defines a positive-definite metric on B λ∗ (c∗ )+1 . From now on we denote this
metric simply by g ij and we denote Ekij simply by kij .
We shall now construct a local future development Vε′0 of the initial data
(g, k) on Hc∗ . This shall be done by the classical method of Choquet-Bruhat (see
[Cho1], [Cho2], [Cho3]), which uses “wave” coordinates (also called “harmonic”
coordinates) adapted to the initial spacelike hypersurface. We shall presently
briefly review this method. The wave coordinates, which we denote by xµ : µ =
0, 1, 2, 3 are functions φ on Vε′0 which are solutions of the wave equation
△φ = 0 (16.215)
on Vε′0 . Such a solution is determined by its initial data on Hc∗ which is φ

together with T̂ φ on Hc∗ . Here, we set:
x0 = c∗ : on Hc∗ , xi = xi : i = 1, 2, 3 : the given Cartesian coordinates on Hc∗

(16.216)
and:
T̂ x0 = 1 : on Hc∗ , T̂ xi = 0 : i = 1, 2, 3 : on Hc∗ (16.217)
These conditions state that the coordinate system is Gaussian normal along
Hc∗ . For, they state that:
∂
T̂ = : along Hc∗ (16.218)
∂x0
Since
∂
g(T̂ , T̂ ) = −1, g T̂ , i = 0
∂x
the vectorfields ∂/∂xi being by 16.216 tangential to Hc∗ , we have:
g00 = −1, g0i = 0 : i = 1, 2, 3 : along Hc∗ (16.219)
Also:
gij = g ij : i, j = 1, 2, 3 : along Hc∗ (16.220)
T
By the domain of dependence theorem, the domain of dependence of Hc∗ M0
in Vε′0 shall be the Minkowskian region Vε′0 M0 , the coordinates xµ : µ =
T
0, 1, 2, 3 coinciding with the original Cartesian coordinates in M0 .
565
The wave equation 16.215 expressed in an arbitrary coordinate system reads:
△φ = (g −1 )αβ ∂α ∂β φ − Γµ ∂µ φ = 0 where Γµ = (g −1 )αβ Γµαβ (16.221)
Thus the condition 16.215 imposed on the coordinate functions xµ is equivalent

to:
Γµ = 0 (16.222)
In view of 16.219, along Hc∗ we have:
∂0 gij = 2kij : i, j = 1, 2, 3 : along Hc∗ (16.223)
The ∂0 g0i : i = 1, 2, 3 and ∂0 g00 along Hc∗ are chosen so as to satisfy the
conditions:
Γµ = 0 : along Hc∗ (16.224)
A short calculation shows that:
∂0 g0i = Γi : i = 1, 2, 3 ∂0 g00 = 2trk : along Hc∗ (16.225)

j j j
where Γi = gij Γ and Γ = (g −1 )mn Γmn are the corresponding 3-dimensional
functions for the induced metric g ij . Also, trk = (g −1 )mn kmn . Let us denote
by Hc∗ ,1 the closed ball B λ∗ (c∗ )+1 on the coordinate hyperplane x0 = c∗ where
the extended functions gij and kij are defined. We then define gµν and ∂0 gµν ,
µ, ν = 0, 1, 2, 3, on Hc∗ ,1 , so as to satisfy 16.219, 16.220, 16.223, and 16.225,
relative to the extended functions g ij and kij .
The Ricci curvature components in an arbitrary coordinate system are given
by:
1
Ricµν = Hµν + Sµν (16.226)
2
Here:
1
Hµν = − (g −1 )αβ ∂α ∂β gµν + Bµν αβκλρσ
∂α gκλ ∂β gρσ (16.227)
2
where B is a homogeneous rational function of the metric components. Also:
Sµν = ∂µ Γν + ∂ν Γµ , Γµ = gµν Γν (16.228)
Thus in wave coordinates the vaccuum Einstein equations reduce to the quasi-
linear system of wave equations:
Hµν = 0 (16.229)
for the metric components, which are called reduced equations. Let us denote:
˜ µν = Ricµν − 1 gµν trRic

Ric (16.230)
2
trRic being the scalar curvature:
trRic = (g −1 )αβ Ricαβ (16.231)
566
Similarly, we denote:
1
S̃µν = Sµν − gµν trS, trS = (g −1 )αβ Sαβ (16.232)
2
The twice contracted Bianchi identities
˜ µν = 0
∇ν Ric (16.233)
imply that every solution of the reduced equations satisfies:
∇ν S̃µν = 0 (16.234)
The equations 16.234 constitute a homogeneous linear system of wave equations

for the Γµ :
(g −1 )αβ ∂α ∂β Γµ + Aαβ
µ ∂α Γβ = 0 (16.235)
where A is a linear form in the partial derivatives of the metric components, with
coefficients which are homogeneous rational functions of the metric components.
Moreover, the constraint equations:
˜ 0i = 0 : i = 1, 2, 3 Ric
Ric ˜ 00 = 0 : along Hc∗ (16.236)
imply that every solution of the reduced equations which satisfies 16.224 along
Hc∗ also satisfies:
∂0 Γµ = 0 : along Hc∗ (16.237)
Now, the constraint equations 16.236 are the contracted Codazzi and twice
contracted Gauss equations of the embedding of Hc∗ in spacetime:
j
∇ kij − ∂i trk = 0
trRic − |k|2 + (trk)2 = 0 (16.238)
Here ∇ is the covariant derivative operator on Hc∗ associated to g, Ric is the

Ricci curvature of g, and trRic = (g −1 )mn Ricmn the
q scalar curvature of g. Also,
|k| is the magnitude of k with respect to g: |k| = (g −1 )mi (g −1 )nj kmn kij . The
constraint equations 16.238 are satisfied along Hc∗ by virtue of the fact that
we have a smooth solution of the vacuum Einstein equations on Mc′∗ which
smoothly extends to its future boundary Hc∗ .
Let then (Vε0 , g) be a development of the extended initial data (gµν , ∂0 gµν )
on Hc∗ ,1 , that is, a domain Vε0 , with past boundary Hc∗ ,1 and a solution gµν of
the reduced equations 16.229 on Vε0 , taking the given initial data along Hc∗ ,1 ,
such that for each point x ∈ Vε0 each past directed causal curve (with respect
to gµν ) initiating at x terminates at a point of Hc∗ ,1 . Then equations 16.235
are satisfied on Vε0 and conditions 16.224 and 16.237 are satisfied along Hc∗ . It
follows that the Γµ vanish on Vε′0 , the domain of dependence of Hc∗ in (Vε0 , g).
Consequently, the restriction of gµν to Vε′0 is a solution of the vacuum Einstein
equations on Vε′0 .
567
Let f be a non-negative non-increasing C ∞ function on the real line such
that:
0 : r≥1
f (r) = (16.239)
1 : r ≤ 12
For any ε > 0 we define the domain Vε by:
Vε = {(x0 , x1 , x2 , x3 ) ∈ ℜ4 : 0 ≤ x0 − c∗ ≤ εf (|x| − λ∗ (c∗ )), |x| ≤ λ∗ (c∗ ) + 1}

(16.240)
and the local existence theorem for the reduced equations 16.229 states that
given any smooth initial data (gµν , ∂0 gµν ) on Hc∗ ,1 where gµν is a Lorentzian
metric defined along Hc∗ ,1 , in particular for any smooth initial data of the form
16.219, 16.220, 16.223, 16.225, there is an ε0 > 0 and a development (Vε0 , g),
taking the given initial data along Hc∗ ,1 . Moreover, for any ε1 ∈ (0, ε0 ] the
future boundary of Vε1 in Vε0 is a spacelike hypersurface.
The vectorfields L′ , L′ satisfy everywhere on Mc∗ Hc∗ , in particular along
S
Hc∗ the formulas 1.8:
L′µ = −2(g −1 )µν ∂ν u, L′µ = −2(g −1 )µν ∂ν u
Thus, in view of 16.218, 16.219, 16.220 and the facts that 2u = c∗ −λ, 2u = c∗ +λ
along Hc∗ , the components of L′ and L′ along Hc∗ in the wave coordinate system
are given by:
L′0 = 2∂0 u = 2T̂ u = Ω−1

xj
L′i = −2(g−1 )ij ∂j u = (g−1 )ij
|x|
L′0 = 2∂0 u = 2T̂ u = Ω−1
xj
L′i = −2(g−1 )ij ∂j u = −(g−1 )ij (16.241)
|x|
and the function Ω is determined along Hc∗ by the conditions:
gµν L′µ L′ν = gµν L′µ L′ν = 0
which yield:
xi xj
Ω−2 = (g −1 )ij (16.242)
|x| |x|
We extend L′ and L′ to Hc∗ ,1 by stipulating that 16.241 and 16.242 hold every-
where along Hc∗ ,1 . Then L′ and L′ are respectively outgoing and incoming null
normal fields to the surfaces of constant λ on Hc∗ ,1 .
Let us denote, for any η ∈ (0, 1], by Hc∗ ,η the closed ball B λ∗ (c∗ )+η on the
′
coordinate hyperplane x0 = c∗ . Let us also denote by H c∗ the closed annular
region B λ∗ (c∗ ) \B−c∗ on the coordinate hyperplane x0 = c∗ . This is Hc′ ∗ together
with its outer boundary in Hc∗ ,1 .
568
We now consider, for each point p ∈ Hc∗ ,1/4 \ B−c∗ −1/4 the null geodesic
initiating at p with initial tangent vector L′ (p). This is the first member of the
solution
xµ = xµ (s; p), L′µ = L′µ (s; p) (16.243)
of the first order system of ordinary differential equations:
dxµ
= L′µ
ds
dL′µ
= −Γµαβ (x(s; p))L′α L′β (16.244)
ds
corresponding to the initial conditions:
x0 (0; p) = c∗ , xi (0; p) = xip : i = 1, 2, 3 L′µ (0; p) = L′µ (p) (16.245)
the Cartesian coordinates of p being (c∗ , x1p , x2p , x3p ). We call this family of null
geodesics the “outgoing” family. We shall show in the following that there is a
ε1 ∈ (0, ε0 ] such that each null geodesic is defined as long as it remains in Vε1 .
Thus the outgoing null geodesic initiating at p is defined for all s ∈ [0, s∗ (p)]
where
x0 (s∗ (p); p) = c∗ + ε1 f (|x(s∗ (p); p)| − λ∗ (c∗ )) (16.246)
that is, x(s∗ (p); p) lies on the future boundary of Vε1 . The coefficients Γµαβ being
smooth functions on Vε0 , the mapping
(s, p) 7→ x(s; p) (16.247)
is a smooth mapping of the domain:
Kε1 = {(s, p) ∈ ℜ×(Hc∗ ,1/4 \B−c∗ −1/4 ) : s ∈ [0, s∗ (p)], p ∈ Hc∗ ,1/4 \B−c∗ −1/4 }
(16.248)
into Vε1 . Assigning polar coordinates (λ, ϑ) to p, and setting λ = c∗ − 2u, the
mapping 16.247 may be described as a smooth mapping
(s, u, ϑ) 7→ x(s; u, ϑ) (16.249)
of the domain:
Kε1 = {(s, u, ϑ) ∈ ℜ2 × S 2 : (16.250)

s ∈ [0, s (c − 2u, ϑ)], 2u ∈ [c − λ (c ) − 1/4), 2c + 1/4], ϑ ∈ S 2 }
∗ ∗ ∗ ∗ ∗ ∗
into Vε1 . The vectorfield L′ is thus extended to a null geodesic vectorfield

and s is the corresponding affine parameter function which vanishes on the
0 ∗
coordinate
T hyperplane x = c . An outgoing null geodesic initiating at a point
p ∈ Hc∗ M0 is described by:
xip
x0 (s; p) = c∗ +s, xi (s, p) = xip +s i = 1, 2, 3 L′µ (s; p) = L′µ (p) (16.251)
|xp |
569
as long as it remains in M0 .
′
Similarly, we consider, for each point p ∈ H c∗ the null geodesic initiating at
′
p with initial tangent vector L (p). This is the first member of the solution
xµ = xµ (s; p), L′µ = L′µ (s; p) (16.252)
of the first order system of ordinary differential equations:
dxµ
= L′µ
ds
dL′µ
= −Γµαβ (x(s; p))L′α L′β (16.253)
ds
corresponding to the initial conditions:
x0 (0; p) = c∗ , xi (0; p) = xip : i = 1, 2, 3 L′µ (0; p) = L′µ (p) (16.254)
the Cartesian coordinates of p being (c∗ , x1p , x2p , x3p ). We call this family of null
geodesics the “incoming”’ family. Again, we shall show in the following that
each null geodesic is defined as long as it remains in Vε1 . Thus the incoming
null geodesic initiating at p is defined for all s ∈ [0, s∗ (p)] where
x0 (s∗ (p); p) = c∗ + ε1 f (|x(s∗ (p); p)| − λ∗ (c∗ )) (16.255)
that is, x(s∗ (p); p) lies on the future boundary of Vε1 . The coefficients Γµαβ being
smooth functions on Vε0 , the mapping
(s, p) 7→ x(s; p) (16.256)
is a smooth mapping of the domain:
′
K ε1 = {(s, p) ∈ ℜ × H c∗ : s ∈ [0, s∗ (p)], p ∈ H c∗ } (16.257)
into Vε1 . Assigning polar coordinates (λ, ϑ) to p, and setting λ = 2u − c∗ , the
mapping 16.247 may be described as a smooth mapping
(s, u, ϑ) 7→ x(s; u, ϑ) (16.258)
of the domain:
K ε1 = {(s, u, ϑ) ∈ ℜ2 ×S 2 : s ∈ [0, s∗ (2u−c∗ , ϑ)], 2u ∈ [0, c∗ +λ∗ (c∗ )], ϑ ∈ S 2 }
(16.259)
into Vε1 . The vectorfield L′ is thus extended to a null geodesic vectorfield and s
is the corresponding affine parameter function which vanishes on the coordinate
hyperplane x0 = c∗ .
We shall now derive certain bounds in Vε1 . In the following, for any multiplet
µ ...µ
of real numbers of the form Aν11...νpq : µ1 , ..., µq ; ν1 , ..., νp = 0, 1, 2, 3 we denote
by |A| the “magnitude”:
s
µ ...µ
X
|A| = (Aν11..νp q )2 (16.260)
µ1 ,...,µq ;ν1 ,...,νp
570
Since the Γµαβ are smooth functions on Vε0 , there is a positive constant M such
that:
sup |Γ(x)| = M (16.261)
x∈Vε0
where, in accordance with the above notation,

sX
|Γ(x)| = (Γµαβ (x))2
µ,α,β
We shall presently show that if ε1 is chosen suitably small depending on M ,

then for the outgoing family of null geodesics we have:
1
|L′ (s; p)| ≤ 2, L′0 (s; p) ≥ : for all (s; p) ∈ Kε1 (16.262)
2
and for the incoming family of null geodesics we similarly have:
1
|L′ (s; p)| ≤ 2, L′0 (s; p) ≥ : for all (s; p) ∈ K ε1 (16.263)
2
Here, in accordance with the notation 16.260,
sX sX
′
′
|L (s; p)| = ′µ 2
(L ) , |L (s; p)| = (L′µ )2
µ µ
Consider first 16.262. By 16.241, 16.242 and 16.209 along Hc∗ ,1 we have:
√
|L′ | ≤ 2 + O(δ 1/2 |c∗ |−1 ), L0 ≥ 1 − O(δ 1/2 |c∗ |−1 ) (16.264)
therefore 16.262 hold at s = 0 provided that δ is suitably small. Consider now

a given p ∈ Hc∗ ,1/4 \ B−c∗ −1/4 . By continuity 16.262 hold at the given p and
for s suitably small. Let s be the largest value of s in the interval (0, s∗ (p)] for
which 16.262 holds at the given p and for all smaller values of s. Then either
s = s∗ (p), in which case what is required has been demonstrated, or one of the
inequalities must be saturated at s. By 16.261 and the second of 16.244 we
have:
dL′
≤ 4M : on [0, s] (16.265)
ds
hence:
|L′ (s; p)| − |L′ (0; p)| ≤ 4M s (16.266)
Moreover, again by 16.261 and the second of 16.244:
|L′0 (s; p) − L′0 (0; p)| ≤ 4M s (16.267)
Now by the first of 16.244 and the second of the inequalities 16.262 at p, which
holds for all s ∈ [0, s] we have:
dx0 1
≥ (16.268)
ds 2
571
hence:
s
x0 (s; p) − c∗ ≥ : ∀s ∈ [0, s] (16.269)
2
Since in Vε′1 we have x0 − c∗ ≤ ε1 , it follows that:
s ≤ 2ε1 (16.270)
Substituting this bound in 16.266 and 16.267 we obtain:
|L′ (s; p)| − |L′ (0; p)| ≤ 8M ε1 (16.271)
and:
|L′0 (s; p) − L′0 (0; p)| ≤ 8M ε1 (16.272)
Comparing with 16.264 we conclude that neither of the two inequalities 16.262
is saturated at s provided that:
1
ε1 ≤ ε1,0 := (16.273)
32M
(n)
′4
and δ is suitably small (depending on the quantities D : n = 0, 1, 2, 3; D[1] (α)
∞ 4 4 ∗
and the quantities D0 , D /1 , D
/2 (trχ), D
/3 (trχ)). Therefore s = s (p) and we have
established 16.262. The inequalities 16.263 are established in a similar manner.
It follows that the mapping 16.247 is indeed a smooth mapping of Kε1 into Vε1
and the mapping 16.256 is indeed a smooth mapping of K ε1 into Vε1 , and these
statements hold for any ε1 ∈ (0, ε1,0 ].
The first of the inequalities 16.262 and 16.263 together with the bounds (see
16.270):
′
s∗ (p) ≤ 2ε1 : ∀p ∈ Hc∗ ,1/4 \ B−c∗ −1/4 , s∗ (p) ≤ 2ε1 : ∀p ∈ H c∗ (16.274)
imply, integrating the first of equations 16.244 and 16.253,
|x(s; p) − xp | ≤ 4ε1 : ∀s ∈ [0, s∗ (p)], ∀p ∈ Hc∗ ,1/4 \ B−c∗ −1/4 (16.275)
and:
′
|x(s; p) − xp | ≤ 4ε1 : ∀s ∈ [0, s∗ (p)], ∀p ∈ H c∗ (16.276)
Here, as previously, and in contrast to the notation 16.260, we denote by |x| the
magnitude of the spatial part of x:
v
u 3
uX
|x| = t (xi )2
i=1
Let then (s1 , p1 ) and (s2 , p2 ) be two points of Kε1 which are mapped by the
mapping 16.247 to the same point of Vε1 . Then by 16.275 the Euclidean distance
of the points p1 and p2 is at most: 8ε1 Similarly, let (s1 , p1 ) and (s2 , p2 ) be two
points of K ε1 which are mapped by the mapping 16.256 to the same point of
572
Vε1 . Then by 16.276 the Euclidean distance of the points p1 and p2 is at most
8ε1 as well.
The inequality 16.275 also implies that:
1
|x(s∗ (p); p)| ≤ λ∗ (c∗ ) + + 4ε1 (16.277)
4
Therefore by 16.246 and the definition 16.239 of the function f :
x0 (s∗ (p); p) − c∗ = ε1 : ∀p ∈ Hc∗ ,1/4 \ B−c∗ −1/4 (16.278)
provided that:
1
ε1 ≤ (16.279)
16
Similarly, the inequality 16.276 also implies that:
|x(s∗ (p); p)| ≤ λ∗ (c∗ ) + 4ε1 (16.280)
therefore 16.279 a fortiori implies:

′
x0 (s∗ (p); p) − c∗ = ε1 : ∀p ∈ H c∗ (16.281)
Let us set:
ε2,0 = min{ε1,0 , 1/16} (16.282)
Then all the above hold in reference to Vε1 for any ε1 ∈ (0, ε2,0 ]. By the first of
16.244 and the first of the inequalities 16.262:
dx0
≤2 (16.283)
ds
hence integrating on [0, s∗ (p)] we obtain, by 16.278:
ε1
s∗ (p) ≥ : ∀p ∈ Hc∗ ,1/4 \ B−c∗ −1/4 (16.284)
2
Using 16.281 we obtain, in a similar manner:
ε1
s∗ (p) ≥ : ∀p ∈ H c∗ (16.285)
2
Consider now the Jacobian of the mapping 16.247 at s = 0. Let us assign
Cartesian coordinates (y 1 , y 2 , y 3 ) to p. Since s = 0 corresponds to Hc∗ ,1/4 \
B−c∗ −1/4 ⊂ Hc∗ ,1 , we have, by 16.245:
∂xµ
= L′µ |Hc∗ ,1 (16.286)
∂s s=0
and, for j = 1, 2, 3:
∂xµ ∂xµ

0 : for µ = 0
= = (16.287)
∂y j s=0 ∂y j Hc∗ ,1
δij : for µ = i = 1, 2, 3
573
Taking into account 16.241, we then conclude that the Jacobian of the mapping
16.247 at s = 0 is simply:
∂(x0 , x1 , x2 , x3 )
= Ω−1 (16.288)
∂(s, y 1 , y 2 , y 3 ) s=0
Hc∗ ,1
The Jacobian of the mapping 16.249 is then the product of the above by
∂λ/∂u|Hc∗ ,1 = −2 times the Jacobian of the transformation from polar to Carte-
sian coordinates on Hc∗ ,1 .
Consider next the Jacobian of the mapping 16.256 at s = 0. Let us again
assign Cartesian coordinates (y 1 , y 2 , y 3 ) to p. Since s = 0 corresponds to H c∗ ⊂
Hc∗ ,1 , we have, by 16.254:
∂xµ
= L′µ Hc∗ ,1
(16.289)
∂s s=0
and, for j = 1, 2, 3:
∂xµ ∂xµ

0 : for µ = 0
= = (16.290)
∂y j s=0 ∂y j Hc∗ ,1
δij : for µ = i = 1, 2, 3
Taking into account 16.241, we then conclude that the Jacobian of the mapping
16.256 at s = 0 is simply:
∂(x0 , x1 , x2 , x3 )
= Ω−1 (16.291)
∂(s, y 1 , y 2 , y 3 ) s=0
Hc∗ ,1
The Jacobian of the mapping 16.258 is then the product of the above by
∂λ/∂u|Hc∗ ,1 = 2 times the Jacobian of the transformation from polar to Carte-
sian coordinates on Hc∗ ,1 .
By 16.242 Ω−1 is bounded from below by a positive constant on Hc∗ ,1 . In
fact, by 16.213:
1
Ω−1 ≥ 1 − O(δ 1/2 ) ≥ on Hc∗ ,1 (16.292)
2
(n)
′4
if δ is suitably small depending on the quantities D : n = 0, 1, 2, 3; D[1] (α) and
∞ 4 4
the quantities D0 , D /2 (trχ), D
/1 , D /3 (trχ). We are then in a position to apply
the implicit function theorem to conclude that there is a ε3,0 ∈ (0, ε2,0 /2] and
a η0 > 0 such that for every p ∈ Hc∗ ,1/4 \ B−c∗ −1/4 the mapping 16.247, with
ε2,0 in the role of ε1 , restricted to [0, ε3,0 ] × B η0 (p), is a diffeomorphism onto
′
its image in Vε2,0 , and for every p ∈ H c∗ the mapping 16.256, with ε2,0 in the
role of ε1 , restricted to [0, ε3,0 ] × B η0 (p), is a diffeomorphism onto its image in
Vε2,0 . Here we denote by B η0 (p) the closed Euclidean ball on Hc∗ ,1 with center
′
at p, intersection Hc∗ ,1/4 \ B−c∗−1/4 and H c∗ respectively. We choose the upper
bound ε2,0 /2 for ε3,0 , so that by inequalities 16.284 and 16.285 the domain of
574
the restricted mappings is indeed contained in the domains Kε2,0 and K ε2,0
respectively. Consider now the mappings 16.247 and 16.256 with ε1 satisfying:
ε3,0
ε1 ≤ (16.293)
2
Then by virtue of the inequalities 16.274 we have:
[ [
[0, ε3,0 ] × B η0 (p) ⊃ Kε1 and [0, ε3,0 ] × B η0 (p) ⊃ K ε1
′
p∈Hc∗ ,1/4 \B−c∗ −1/4 p∈H c∗
(16.294)
respectively. Let then (s1 , p1 ) and (s2 , p2 ) be two points of Kε1 which are
mapped by the mapping 16.247 two the same point of Vε1 . Then as we have
shown earlier, the Euclidean distance of the points p1 and p2 is at most 8ε1 .
Therefore if ε1 satisfies:
η0
ε1 ≤ (16.295)
16
the points p1 and p2 are both contained in one of the balls B η0 . Moreover, by
the first of the inequalities 16.274, s1 , s2 ≤ 2ε1 , therefore if ε1 satisfies also:
ε3,0
ε1 ≤ (16.296)
2
the points (s1 , p1 ) and (s2 , p2 ) are both contained in one of the domains [0, ε3,0 ]×
B η0 (p). It then follows that (s1 , p1 ) = (s2 , p2 ). We have thus shown that setting:
ε1 = min{ε3,0 /2, η0 /16} (16.297)
the mapping 16.247 of Kε1 into Vε1 is one to one. Being locally a diffeomorphism
this mapping is then a diffeomorphism onto its image in Vε1 . A similar argument
shows that the mapping 16.256 of K ε1 into Vε1 is a diffeomorphism onto its image
in Vε1 . Moreover, it readily follows from the inequalities 16.275 and 16.276 that
the range of the mapping 16.247 includes the range of the mapping 16.256.
Now the mapping 16.249, which for the sake of clarity we denote from now
on by φ:
x = φ(s, u, ϑ) (16.298)
is the mapping 16.247 composed on the right with the mapping:
(s, u, ϑ) 7→ (s, p(c∗ − 2u, ϑ)) (16.299)
the mapping (λ, ϑ) 7→ p(λ, ϑ) being simply the transformation from polar to
Cartesian coordinates on Hc∗ ,1 . Since the last is a diffeomorphism of [−c∗ −
1/4, λ∗ (c∗ ) + 1/4] × S 2 onto Hc∗ ,1/4 \ B−c∗ −1/4 , it follows that the mapping
16.298 is a diffeomorphism of Kε1 , as defined by 16.250, onto its image in Vε1 .
Similarly, the mapping 16.258, which for the sake of clarity we denote from now
on by φ:
x = φ(s, u, ϑ) (16.300)
575
is the mapping 16.256 composed on the right with the mapping:
(s, u, ϑ) 7→ (s, p(2u − c∗ , ϑ)) (16.301)
The transformation from polar to Cartesian coordinates on Hc∗ ,1 being a diffeo-

′
morphism of [−c∗ , λ∗ (c∗ )] × S 2 onto H c∗ , it follows that the mapping 16.300 is
a diffeomorphism of K ε1 , as defined by 16.259, onto its image in Vε1 . Moreover,
the range Wε1 of the mapping 16.298 includes the range W ε1 of the mapping
16.300.
It follows from the above that u is a smooth function of x on Wε1 . Now,
by construction each level set Cu of u in Wε1 is generated by the congruence of
′
outgoing null geodesic normals to the surface Sc∗ −u,u in H c∗ . The hypersurface
Cu is smooth, being the image by the diffeomorphism φ of the “cylinder”
{(s, u, ϑ) : s ∈ [0, s∗ (c∗ − 2u, ϑ)], ϑ ∈ S 2 } ⊂ Kε1
and the generators all terminate on the part of the future boundary of Vε1 where
x0 = c∗ + ε1 . It follows that the vectorfield −2(g −1 )µν ∂ν u is collinear to L′µ ,
therefore it is null and u is a solution of the eikonal equation
(g −1 )µν ∂µ u∂ν u = 0 (16.302)
in Wε1 . It then follows that the vectorfield −2(g −1 )µν ∂ν u is a null geodesic
vectorfield, whose integral curves must coincide with those of L′µ . Since the
two vectorfields coincide along Hc∗ ,1/4 \ B−c∗ −1/4 it follows that:
− 2(g −1 )µν ∂ν u = L′µ (16.303)
in Wε1 . Similarly, by construction each level set C u of u in W ε1 is generated

by the congruence of incoming null geodesic normals to the surface Su,c∗ −u in
′
H c∗ . The hypersurface C u is smooth, being the image by the diffeomorphism
φ of the “cylinder”
{(s, u, ϑ) : s ∈ [0, s∗ (2u − c∗ , ϑ)], ϑ ∈ S 2 } ⊂ K ε1
and the future end points of the generators all lie on the part of the future
boundary of Vε1 where x0 = c∗ +ε1 . It follows that the vectorfield −2(g −1 )µν ∂ν u
is collinear to L′µ , therefore it is null and u is a solution of the eikonal equation
(g −1 )µν ∂µ u∂ν u = 0 (16.304)
in W ε1 . It then follows that the vectorfield −2(g −1 )µν ∂ν u is a null geodesic

vectorfield, whose integral curves must coincide with those of L′µ . Since the
′
two vectorfields coincide along H c∗ it follows that:
− 2(g −1 )µν ∂ν u = L′µ (16.305)
in W ε1 .
576
Now, since in particular C (c∗ +λ∗ (c∗ ))/2 , the outer boundary of W ε1 , is a
smooth null hypersurface generated by the congruence of incoming null geodesic
′
normals to S(c∗ +λ∗ (c∗ ))/2,(c∗ −λ∗ (c∗ ))/2 , the outer boundary of H c∗ , and the gener-
ators all terminate on the part of the future boundary of Vε1 where x0 = c∗ + ε1 ,
it follows that the closure of Vε′1 , the domain of dependence of Hc∗ in (Vε1 , g),
is: \ [
(M0 Vε1 ) W ε1 (16.306)
and its outer boundary is the outer boundary of W ε1 . We shall restrict attention
from now on to this domain, where we have aTsmooth solution gµν of the vacuum
Einstein equations. Moreover, since in M0 Vε1 gµν is simply the Minkowski
metric in Cartesian coordinates, we shall focus on the domain W ε1 where we
have a non-trivial solution.
Consider on W ε1 the function Ω defined by:
1
Ω−2 = − gµν L′µ L′ν (16.307)
2
This is a smooth positive function, the vectorfields L′ and L′ being future di-
rected null vectorfields which are nowhere collinear. By 16.241, the function Ω
′
coincides along H c∗ to the function previously defined there by 16.242. Now u
is a smooth function of x in W ε1 . For the sake of clarity we write:
u = f (x) (16.308)
Substituting x = φ(s, u, ϑ), that is, considering the composition:
h=f ◦φ (16.309)
a smooth function on K ε1 , we write:
u = h(s, u, ϑ) (16.310)
We then have: µ
∂h ∂φ
= (∂µ f ) ◦ φ (16.311)
∂s ∂s
which in view of the fact that:
∂φµ
= L′µ ◦ φ
∂s
becomes:
∂h
= (L′µ ∂µ u) ◦ φ = Ω−2 ◦ φ (16.312)
∂s
by 16.303 and 16.307.Therefore ∂u/∂s is a smooth positive function on K ε1 . It
then follows that the mapping π of K ε1 onto the domain:
Ñε1 = {(u, u, ϑ) ∈ ℜ2 × S 2 : (16.313)

u ∈ [c − u, h(s (2u − c , ϑ), u, ϑ)], 2u ∈ [0, c + λ (c )], ϑ ∈ S 2 }
∗ ∗ ∗ ∗ ∗ ∗
577
by:
(s, u, ϑ) 7→ (h(s, u, ϑ), u, ϑ) (16.314)
is a diffeomorphism. Then:
ψ = φ ◦ π −1 (16.315)
is a diffeomorphism of Ñε1 onto W ε1 . The coordinates (u, u, ϑ) are canonical

coordinates on W ε1 , which smoothly extend the canonical coordinates on Mc′∗ .
The mapping ψ being a diffeomorphism, we have smooth metric in canonical
coordinates on Ñε1 , a smooth extension of the metric in the same coordinates
on Mc′∗ . Let finally
Ω−2
m = inf Ω
−2
>0 (16.316)
Wε
1
Then along the generators of each C u , u ∈ [0, (c∗ + λ∗ (c∗ ))/2], we have:
∂h
≥ Ω−2
m (16.317)
∂s
Integrating we obtain:
1
h(s∗ (2u − c∗ , ϑ), u, ϑ) − (c∗ − u) ≥ Ω−2 ∗ ∗
m s (2u − c , ϑ) ≥ ε1 Ω−2
m (16.318)
2
by 16.285. Hence the domain Ñε1 includes the domain:
Nε2 = {(u, u, ϑ) : u + u ∈ [c∗ , c∗ + ε2 ], 2u ∈ [0, c∗ + λ∗ (c∗ )], ϑ ∈ S 2 } (16.319)
where:
1
ε2 = ε1 Ω−2
m (16.320)
2
Now, if c∗ ∈ [u0 + δ, −1), then by 16.182 λ∗ (c∗ ) = 2δ − c∗ and we have:
[
Mc′∗ Nε2 ⊃ Mc′∗ +ε2 (16.321)
If on the other hand c∗ ∈ (u0 , u0 + δ), then by 16.182 λ∗ (c∗ ) = c∗ − 2u0 and
we consider the characteristic initial value problem with initial hypersurfaces
one of which, the inner one, is the part of C c∗ −u0 which constitutes the outer
boundary of Nε2 , and the other, the outer one, is the part of Cu0 which lies
to the future of Sc∗ −u0 ,u0 . The intersection of these two null hypersurfaces is
the surface Sc∗ −u0 ,u0 . On the inner null hypersurface we have as characteristic
initial data the data induced from the solution on Nε2 , and on the outer null
hypersurace we have as characteristic initial data the original initial data on Cu0 .
We then apply the theorem of Rendall [R] which constructs a smooth solution
of this characteristic initial value problem for the vacuum Einstein equations in
a neighborhood of the surface Sc∗ −u0 ,u0 bounded in the past by the two null
hypersurfaces. The solution is again obtained in a wave coordinate system,
adapted in this case to the two intersecting null hypersurfaces. An entirely
578
analogous argument to the one presented above then demonstrates that there
is a
ε′2 ∈ (0, min{ε2 , u0 + δ − c∗ }) (16.322)
and a smooth solution in canonical coordinates on the domain:
Nε′ ′ = {(u, u, ϑ) : u + u ∈ [c∗ , c∗ + ε′2 ], u ≥ u0 , u ≥ c∗ − u0 , ϑ ∈ S 2 } (16.323)

2
Then [ [
Mc′∗ Nε2 Nε′ ′ ⊃ Mc′∗ +ε′ (16.324)
2 2
The same argument shows that there is a subdomain of the initial domain
U (see last section of Chapter 2) of the form Mu′ 0 +η0 , for some η0 > 0, where
we have a smooth metric in canonical coordinates.
Since we have a smooth metric S in canonical coordinates on Nε2 , in the case
c∗ ∈ [u0 + δ, −1], and on Nε2 Nε′ ′ , in the case c∗ ∈ (u0 , u0 + δ), all the
2
quantities which appear in the bootstrap assumptions A0, A1.1, A1.2, A2.1,
A2.2, A3.1, A3.2, A4.1, A4.2, B1, B2, B3, and C1.1 - C1.4, C2.1 - C2.4,
C3.1 - C3.5, C4.1 - C4.8, C5.1 - C5.4, C6.1 - C6.10, as well as D0, D1,
D2.1, D2.2, D3.1, D3.2, D4.1, D4.2, D5, D6, D7, S D8, D9.1 - D9.3, D10.1
- D10.4, D11, D 12, are continuous on Nε2 and Nε2 Nε′ ′ , respectively. Since
2
the inequalities involved are not saturated on Hc′ ∗ , it follows that there is a ε3 ,
ε3 ∈ (0, ε2 ] in the case c∗ ∈ [u0 + δ, −1], ε3 ∈ (0, ε′2 ] in the case c∗ ∈ (u0 , u0 + δ),
such that all the above bootstrap assumptions hold on Mc∗ +ε3 as well.
The same argument shows that there is a η1 ∈ (0, η0 ] such that the bootstrap
assumptions hold on Mu′ 0 +η1 ⊂ Mu′ 0 +η0 .
(n)
Finally, the error integrants δ 2qn τ 2 : n = 0, 1, 2, 3Sare integrable functions
on Nε2 , in the case c∗ ∈ [u0 + δ, −1], and on Nε2 Nε′ ′ , in the case c∗ ∈
2
(u0 , u0 + δ). It follows that there is a ε4 ∈ (0, ε3 ] such that:
Z
(n)
δ 2qn | τ 2 |dµg ≤ 1 : n = 0, 1, 2, 3 (16.325)
Mc′ ∗ +ε \Mc′ ∗
4
hence by the energy-flux inequalities 12.287 with Mc′∗ +ε4 in the role of Mc′∗ we
have:
(n) (n)
Z
2qn (n)
E 2 ≤ D +δ | τ 2 |dµg + 1 : n = 0, 1, 2, 3
Mc′ ∗
(3) (3)
Z
(3)
F 2≤ D +δ 2q3 | τ 2 |dµg + 1 (16.326)
Mc′ ∗
(n) (3)
where now the quantities E 2 n = 0, 1, 2, 3 and F 2 refer to Mc′∗ +ε4 . Combining
with the bounds 16.24 we then conclude that:
(0) (1) (2) (3)
P2 ≤ G( D , D , D , D ) (16.327)
579
where now the quantity P2 refers to Mc′∗ +ε4 .
Similarly, in reference to the domain Mu′ 0 +η1 , there is a η2 ∈ (0, η1 ] such
that: Z
2qn (n)
δ | τ 2 |dµg ≤ 1 : n = 0, 1, 2, 3 (16.328)
′
Mu
0 +η2
This implies that the quantity P2 corresponding to Mu′ 0 +η2 satisfies the bound
required in the 3rd condition of the statement of Theorem 12.1. We have thus
shown that u0 + η2 ∈ A, so the set A is indeed non-empty.
Returning now to the inequality 16.327, which refers to Mc′∗ +ε4 , we conclude
that all three conditions of Theorem 12.1 are satisfied with c∗ + ε4 in the role
of c. This contradictcs the definition of c∗ unless c∗ = −1. This completes the
proof of Theorem 12.1.
16.4 Restatement of the existence theorem

According to the discussion preceding the statement of Theorem 12.1 the quan-
(n)
′4
tities D : n = 0, 1, 2, 3, D[1] (α), D0∞ , D
/41 , D
/42 (trχ), D
/3 (trχ) are all bounded by a
non-negative non-decreasing continuous function of M8 . The theorem can now
be restated as follows.
Theorem 16.1 Let us be given smooth initial data on Cu0 as described in

Chapter 2. There is a non-negative non-decreasing continuous function F on
the non-negative real line such that if:
δF (M8 ) ≤ 1
the following hold:
1. With M−1 \ Γ0 = (D−1 \ A0 ) × S 2 there is a smooth solution g of the

vacuum Einstein equations in canonical coordinates on M−1 \ Γ0 , taking
the given initial data along Cu0 , the subdomain of M−1 \ Γ0 where u ≤ 0
being isometric to the corresponding domain in Minkowski spacetime. In
particular, the null hypersurfaces Cu and C u contain no focal or cut points
in M−1 \ Γ0 .
2. The assumptions A0, A1.1, A1.2, A2.1, A2.2, A3.1, A3.2, A4.1, A4.2
′
hold on M−1 .
3. There is a non-negative non-decreasing continuous function G on the non-
negative real line such that:
Q′2 ≤ G(M8 )
the quantity Q′2 referring to M−1

′
. It follows, by the results of Chapter 10,
that:
R∞ 0 ≤ CG(M8 ), R /41 ≤ CG(M8 )
580
max{R40 (D̂α), R40 (Dβ), R40 (Dσ), R40 (Dβ)} ≤ CG(M8 )
R40 (D̂α) ≤ CG(M8 )

′
all quantities referring to M−1 .
′
4. The results of Chapters 3 - 7, all hold on M−1 with the symbol O(δ p |u|r )
p r
re-interpreted to mean the product of δ |u| with a non-negative non-
decreasing continuous function of M8 .
581
Chapter 17
Trapped Surface Formation
We are now ready to reach the aim of this work, namely the analysis of the
formation of trapped surfaces.
By Theorem 16.1 and the results of Chapter 3 we have:
2
trχ + ≤ O(δ|u|−2 ) : on M−1
′
(17.1)
|u|
Therefore if δ is suitably small depending on M8 , trχ is everywhere negative on
′ ′
M−1 . Consequently, a surface Su,u contained in M−1 is a trapped sphere if and
only if everywhere on this surface trχ < 0, or equivalently trχ′ < 0.
Consider then equation 4.6:

1 1
D̂(Ωχ̂) = Ω2 ∇ ˆ + η ⊗η
/⊗η ˆ + trχχ̂ − trχχ̂ (17.2)
2 2
In view of the fact that D log Ω = ω, this equation takes the form:
1
D̂χ̂ − Ωtrχχ̂ = θ (17.3)
2
where θ is the trace-free symmetric 2-covariant S tensorfield:

1
θ=Ω ∇ ˆ + η ⊗η
/⊗η ˆ − trχχ̂ − ω χ̂ (17.4)
2
We apply the second part of Lemma 4.2 to obtain:
D(|χ̂|2 ) + 2Ωtrχ|χ̂|2 = 2(χ̂, D̂χ̂) − 4Ω(χ̂, χ̂ × χ̂)
In view of the identity 1.188 the last term vanishes and we obtain simply:
D(|χ̂|2 ) + 2Ωtrχ|χ̂|2 = 2(χ̂, D̂χ̂) (17.5)
Substituting for Dχ̂ from 17.3 yields:
D(|χ|2 ) + Ωtrχ|χ̂|2 = 2(χ̂, θ) (17.6)
582
We then consider the function:
f = |u|2 |χ̂|2 (17.7)
We have (recall that u < 0):
Df = |u|2 D(|χ̂|2 ) − 2|u||χ̂|2

2 2 2 2 2
|χ̂|2

= |u| D(|χ̂| ) + Ωtrχ|χ̂| − |u| Ωtrχ +
|u|
Substituting from 17.6 then yields the equation:
Df = g (17.8)
where g is the function:

2
g = |u|2 − Ωtrχ + |χ̂|2 + 2(χ̂, θ) (17.9)
|u|
Now, by Theorem 16.1 and the results of Chapter 3:
2
Ωtrχ + ≤ O(δ|u|−2 ) : on M−1
′
(17.10)
|u|
and:
|χ̂|2 ≤ O(δ −1 |u|−2 ) : on M−1
′
(17.11)
with the symbol O(δ p |u|r ) interpreted from this point on as in the 4th statement
of Theorem 16.1.
Consider next θ. From Theorem 16.1 and Proposition 6.2:
|u|k∇ /ηkL4 (Su,u ) ≤ O(|u|−5/2 )

/ 2 ηkL4 (Su,u ) + k∇ (17.12)
Applying Lemma 5.2 with p = 4 we then obtain, in view of Lemma 10.1,
/η| ≤ O(|u|−3 ) : on M−1

|∇ ′
(17.13)
Also, by Theorem 16.1 and the results of Chapter 3:
|η|2 ≤ O(δ|u|−4 ) : on M−1

′
|trχ||χ̂| ≤ O(δ 1/2 |u|−3 ) : on M−1

′
|ω||χ̂| ≤ O(δ 1/2 |u|−4 ) : on M−1

′
(17.14)

|θ| ≤ O(|u|−3 ) : on M−1
′
(17.15)
Since |(χ̂, θ)| ≤ |χ̂||θ|, the bounds 17.10, 17.11, 17.15, imply:
|g| ≤ O(δ −1/2 |u|−2 ) : on M−1

′
(17.16)
583
Let us work in a canonical coordinate system. Then equation 17.8 reads
simply:
∂f
=g (17.17)
∂u
Integrating with respect to u on [u0 , −1 − δ] we then obtain:
Z −1−δ
f (u, −1 − δ, ϑ) = f (u, u0 , ϑ) + g(u, u, ϑ)du (17.18)
u0
By 17.16 the integral on the right is bounded from below by:

Z −1−δ
− |g(u, u, ϑ|du ≥ −O(δ −1/2 ) (17.19)
u0
We conclude that:
f (u, −1 − δ, ϑ) ≥ f (u, u0 , ϑ) − O(δ −1/2 ) (17.20)
Now, according to equation 3.8 we have on each Cu , in particular on C−1−δ :
1
Dtrχ′ = − (trχ)2 − |χ̂|2 (17.21)
2
From the definition 17.7, on C−1−δ this implies:
Dtrχ′ ≤ −(1 + δ)−2 f (17.22)
Consider now each generator of C−1−δ , integral curve of L. If (0, −1−δ, ϑ0) is the
point where a generator intersects S0,−1−δ , the generator is given in canonical
coordinates by:
u 7→ (u, −1 − δ, ϑ(u; ϑ0 )) (17.23)
where ϑ(u; ϑ0 ) is the solution of the ordinary differential equation (see 1.197):
dϑA (u; ϑ0 )
= bA (u, −1 − δ, ϑ(u; ϑ0 )) (17.24)
du
corresponding to the initial condition:
ϑ(0, ϑ0 ) = ϑ0 (17.25)
Equation 17.24 represents a non-autonomous flow on S 2 , b at u = −1 − δ being
a vectorfield on S 2 depending on u. This is simply the flow Φu on C−1−δ .
Since
2
trχ′ |S0,−1−δ =
1+δ
integrating equation 17.22 along the generator of C−1−δ originating at (0, −1 −
δ, ϑ0 ) ∈ S0,u0 we obtain:
Z u
2 1
trχ′ (u, −1 − δ, ϑ(u; ϑ0 )) ≤ − f (u′ , −1 − δ, ϑ(u′ ; ϑ0 ))du′
1+δ (1 + δ)2 0
(17.26)
584
Suppose now that:
Z δ
f (u, −1 − δ, ϑ(u; ϑ0 ))du > 2(1 + δ) : for all ϑ0 ∈ S 2 (17.27)
0
Then there is a u∗ ∈ (0, δ) such that for all u ∈ (u∗ , δ) we have:
trχ′ (u, −1 − δ, ϑ(u; ϑ0 )) < 0 : for all ϑ0 ∈ S 2 (17.28)
therefore, since Φu is a diffeomorphism of S0,−1−δ onto Su,−1−δ ,
trχ′ < 0 : everywhere on Su,−1−δ
that is, Su,−1−δ is a trapped sphere.

By 17.20 the condition 17.27 is satisfied if:
Z δ
f (u, u0 , ϑ(u; ϑ0 ))du > 2 + O(δ 1/2 ) : for all ϑ0 ∈ S 2 (17.29)
0
(where we have absorbed δ into O(δ 1/2 ).) Recalling the definition 2.69 of the
function e on Cu0 we see that:
f (u, u0 , ϑ) = 2|u0 |2 e(u, u0 , ϑ) (17.30)
therefore condition 17.29 reads, in terms of the function e:

Z δ
2
|u0 | e(u, u0 , ϑ(u; ϑ0 ))du > 1 + O(δ 1/2 ) : for all ϑ0 ∈ S 2 (17.31)
0
Now condition 17.31 is not natural as it stands, because the integral is not
an integral along the generators of Cu0 , but rather along the images on Cu0 by
Φu0 +1+δ of the generators of C−1−δ . To turn it into a natural condition, we
must first estimate the deviation of the image on Cu0 by Φu0 +1+δ of a gener-
ator of C−1−δ from the corresponding generator of Cu0 , namely the generator
originating at the same point on S0,u0 . Analytically, the problem is to estimate:
◦
d (ϑ(u; ϑ0 ), ϑ0 )
◦ ◦
the distance d, relative to the standard metric g/ on S 2 , of the points ϑ(u; ϑ0 )
and ϑ0 . Now, the S tangential vectorfield b is the solution of 1.200:
Db = 4Ω2 ζ ♯ (17.32)
with the initial condition 1.199:
b = 0 : on Cu0 (17.33)
We have:
D(|b|2 ) = D(g/(b, b)) = 2Ωχ(b, b) + 2g/(b, Db) = Ωtrχ|b|2 + Ωχ̂(b, b) + 8Ω2 (b, ζ)
(17.34)
585
hence:
1
D(|b|2 ) ≤ 2|b| Ωtrχ + Ω|χ̂| |b| + 4Ω2 |ζ| (17.35)
2
Applying Lemma 3.1 we then obtain:
Z u Z u
|b(u, u, ϑ)| ≤ 4 exp ((1/2)Ωtrχ + Ω|χ̂|)(u, u′′ , ϑ)du′′
u0 u′
·(Ω2 |ζ|)(u, u′ , ϑ)du′(17.36)
By Theorem 16.1 and the results of Chapter 3 the integral in the exponential
does not exceed:
|u|
log + O(δ 1/2 |u|−1 ) (17.37)
|u′ |
Moreover, if δ is suitably small depending on M8 the O(δ 1/2 |u|−1 ) term does
not exceed log 2. It follows that 17.36 implies:
Z u
−1
|u| ||b(u, u, ϑ)| ≤ 8 |u′ |−1 (Ω2 |ζ|)(u, u′ , ϑ)du′ (17.38)
u0
Since by Theorem 16.1 and the results of Chapter 3 we have:
Ω2 |ζ| ≤ O(δ 1/2 |u|−2 ) : on M−1

′
(17.39)
it follows that:
|b| ≤ O(δ 1/2 |u|−1 ) : on M−1
′
(17.40)
In particular, this holds on C−1−δ and we obtain:
| b|Su,−1−δ | /g|S ≤ O(δ 1/2 ) (17.41)

u,−1−δ
Now, by Lemma 11.1 and the inequality 11.26 applied to b|Su,−1−δ there is a
|b(u, −1 − δ, ϑ)|◦ ≤ C| b|Su,−1−δ | /g|S (17.42)

/
g u,−1−δ
Hence also:
sup |b(u, −1 − δ, ϑ)|◦ ≤ O(δ 1/2 ) (17.43)
ϑ∈S 2 /
g
◦
This is a bound for the speed, relative to g/, of the flow defined by 17.24 on S 2 .
◦
It follows that the length, relative to g/ of the arc 17.23 (u ∈ [0, δ]) does not
exceed O(δ 3/2 ), therefore:
◦
3/2
d (ϑ(u; ϑ0 , ϑ0 )) ≤ O(δ ) (17.44)
Now, by 2.72:
/e|◦ ≤ O(δ −1 |u0 |−2 )
|d (17.45)
/
g
586
The bounds 17.44 and 17.45 together imply:
|u0 |2 |e(u, u0 , ϑ(u; ϑ0 )) − e(u, u0 , ϑ0 )| ≤ O(δ 1/2 ) (17.46)
Consequently, condition 17.31 is satisfied if:

Z δ
2
|u0 | e(u, u0 , ϑ0 )du > 1 + O(δ 1/2 ) : for all ϑ0 ∈ S 2 (17.47)
0
Since O(δ 1/2 ) is of the form δ 1/2 F (M8 ) where F is a non-negative non-decreasing
continuous function on the non-negative real line, given any constant k > 1,
Z δ
2
|u0 | e(u, u0 , ϑ0 )du ≥ k : for all ϑ0 ∈ S 2 (17.48)
0
and:
δ 1/2 F (M8 ) < k − 1 (17.49)
imply 17.47. The condition 17.48 is natural: it simply says that the integral of
|u0 |2 e along any generator of Cu0 , with respect to the affine parameter, is not
less than k.
Finally, from 2.40, 2.41, and 2.70, we have:
1 −1 AC −1 BD ∂mAB ∂mCD
e= (m ) (m ) (17.50)
8 ∂u ∂u
in the north polar chart, and:
1 ′−1 AC ′−1 BD ∂m′AB ∂m′CD
e= (m ) (m ) (17.51)
8 ∂u ∂u
in the south polar chart. Then by 2.38, 2.39, 2.46, and the fact that exp is an
analytic map with derivative at the origin equal to I, we have:
2
1 ∂ψ0 u
δ|u0 |2 e(u, u0 , ϑ) − ,ϑ ≤ O(δ 1/2 |u0 |−1 ) (17.52)
8 ∂s δ
in the north polar chart, and:

2
1 ∂ψ0′ u
δ|u0 |2 e(u, u0 , ϑ) − ,ϑ ≤ O(δ 1/2 |u0 |−1 ) (17.53)
8 ∂s δ
in the south polar chart. It follows that, given any constant k > 1,
2
1 1 ∂ψ0
Z
(s, ϑ0 ) ds ≥ k for all ϑ0 ∈ D2ρ (17.54)
8 0 ∂s
and:
1 2
1 ∂ψ0′
Z
(s, ϑ0 ) ds ≥ k for all ϑ0 ∈ D2ρ (17.55)
8 0 ∂s
587
(here | | denotes the Euclidean norm on ℜ2 , as in 2.47) then if also δ 1/2 |u0 |−1
is suitably small depending on M1 , the condition 17.48 is satisfied with 1 +
(1/2)(k − 1) > 1 in the role of k.
We have thus established the following theorem.
Theorem 17.1 Let 2 ≥ k > 1. With initial data on Cu0 as in Theorem 16.1
there is a non-negative non-decreasing continous function F on the non-negative
real line such that if:
δ 1/2 F (M8 ) < k − 1
and one of the following two conditions holds:
Z δ
1. |u0 |2 e(u, u0 , ϑ)du ≥ k : for all ϑ0 ∈ S 2
0
or:
1 2
1 ∂ψ0
Z
2. (s, ϑ) ds ≥ k for all ϑ ∈ D2ρ
8 0 ∂s
and:
1 2
1 ∂ψ0′
Z
(s, ϑ) ds ≥ k for all ϑ ∈ D2ρ
8 0 ∂s
then there is a u ∈ (0, δ) such that for all u ∈ (u∗ , δ) the surfaces Su,−1−δ ⊂
∗
′
C−1−δ ⊂ M−1 are trapped spheres. Moreover, their areas satisfy:
|Area(Su,−1−δ ) − 4π| ≤ O(δ)
Now the initial data on Cu0 can be extended to regular initial data on a
complete cone Co . This can be done, for example, in the following way. We recall
that the mappings ψ0 and ψ0′ extend smoothly by zero to s < 0. We may then
apply a total extension operator E for (−∞, 1] to ψ0 (·, ϑ), ψ0′ (·, ϑ) : ϑ ∈ D2ρ ,
obtaining smooth mappings Eψ0 , Eψ0′ of ℜ×D2ρ into Ŝ, with support in [0, 2]×
D2ρ , satisfying the transformation rule 2.43, 2.44. This gives asymptotically flat
data along Co . Let M ∗ be the unique maximal future development of the data
on Co . Then since the data on the part of Co which corresponds to u ≤ δ
coincides with the data we have been considering, the development M−1 is a
subdevelopment of M ∗ . Therefore M ∗ contains trapped spheres. We may then
apply the theorem of Penrose, with Co in the role of the non-compact Cauchy
hypersurface in that theorem (see Prologue), to conclude the following.
Theorem 17.2 Consider smooth regular initial data on a complete cone Co

which extends initial data satisfying the conditions of Theorem 17.1. Then the
maximal future development of the initial data on Co is future null geodesically
incomplete.
588
We conclude the present monograph with the following remarks. With a
fixed pair (ψ0 , ψ0′ ) of smooth mappings of [0, 1] × D2ρ into Ŝ as in Chapter
2, and fixed δ satisfying the smallness conditions of Theorems 16.1 and 17.1
relative to M8 , we may consider the sequence of characteristic initial value
problems corresponding to a decreasing sequence u0,n → −∞. The fact that we
have obtained bounds independent of u0 allows us to extract a subsequence such
that the corresponding sequence of solutions converges in the future of Cc for
any c < −1. We may construct in this way a solution such that the generators
of the C u for u ∈ [0, δ) are complete toward the past, and the interior of C 0
is the Minkowskian region which is the complete past of the point e (see first
section of Chapter 1). For this solution we have:
φ(u, u, ϑ) → 1 : as u → −∞ (17.56)
and:
w2 (ϑ)

∂(ψ0 )AB u
|u|−1 χ̂AB (u, u, ϑ) → , ϑ) : as u → −∞ (17.57)
2δ 1/2 ∂s δ
in the north polar chart, and similarly in the south polar chart, uniformly in
[0, δ] × D2ρ . Moreover, it can be shown, using the method of Friedrich [Fr],
that this solution is the unique solution of this type satisfying 17.57 and its
analogue, for a given pair (ψ0 , ψ0′ ). We thus have a unique solution of the
asymptotic characteristic initial value problem with initial data at past null
infinity determined by the pair (ψ0 , ψ0′ ). The limit 17.57 may be thought of as
the incoming radiative amplitude. It is a 2-covariant symmetric tensorfield on
S 2 (see 2.32), depending on u, which is trace-free relative to the standard metric
◦
g/.
Furthermore, if the data on Cu0,n are extended to regular asymptotically
flat data on a complete cone in the manner described in the paragraph preced-
ing Theorem 17.2, we can show that if |u0,n | is sufficiently large the maximal
development contains complete cones Cu for all u ∈ [u0,n , c], for some c < 0
with |c| suitably large. It can then be shown that the corresponding limiting
solution with initial data at past null infinity contains complete cones Cu for
all u ≤ c. The solution then contains a piece of future null infinity which is a
neighborhood of spacelike infinity.
As a consequence of 17.57 we have:
|u|2 e(u, u, ϑ) → e∞ (u, ϑ) as u → −∞ (17.58)
where:
2
1 ∂ψ0 u
e∞ (u, ϑ) = ,ϑ (17.59)
8δ ∂s δ
in the north polar chart, and similarly in the south polar chart, uniformly in
[0, δ] × D2ρ . Note here that since O is an orthogonal transformation (see 2.25),
the transformation rule 2.43 implies that for all (s, ϑ) ∈ [0, 1] × Aρ :
2 2
∂ψ0 ∂ψ0′
(s, ϑ) = (s, ϑ′ ) , ϑ′ = f (ϑ) (17.60)
∂s ∂s
589
hence if e∞ (u, ϑ) and e′∞ (u, ϑ′ ) are the functions corresponding to the north
polar and south polar charts respectively, then we have
e∞ (u, ϑ) = e′∞ (u, ϑ′ ), ϑ′ = f (ϑ) : ∀(s, ϑ) ∈ [0, δ] × Aρ (17.61)
as we should, e∞ and e′∞ being representations of the same function on [0, 1]×S 2.
The function e∞ , as a function on [0, δ] × S 2, has a simple physical meaning:
it is 8π times the incoming radiative power per unit solid angle, a function of
the advanced time u and the direction, the last being represented by a point on
S 2 . (This statement holds in units where Newton’s graviational constant is set
equal to 1). Thus the limiting form of both conditions of Theorem 17.1, namely:
Z δ
e∞ (u, ϑ)du ≥ k : for all ϑ ∈ S 2 (17.62)
0
means simply that the incoming energy per unit solid angle in each direction in
the advanced time interval [0, δ] is not less than k/8π. This corresponds to a
total incoming energy E in the advanced time interval p [0, δ] of at least k/2. The
“area radius” of Su,u being defined by r(Su,u ) = Area(Su,u )/4π, according
to Theorem 17.1 the area radius of the trapped spheres which form satisfies
|r − 1| ≤ O(δ). Therefore, in the limit k → 1, δ → 0, we recover the inequality
2E/r ≥ 1, familiar from the spherically symmetric case. However in the general
case the requirement for the formation of a trapped sphere is not on the total
incoming energy but rather on the incoming energy in each direction.
If the above condition is satisfied for some, but not all, ϑ ∈ S 2 , we can still
conclude that some generator of C−1−δ has a segment on which trχ′ < 0. It
then follows that if that generator is extended in the maximal development then
either there is a value s∗ of the affine parameter and a (non-zero) Jacobi field X
along the generator in question such that |X(s)| → 0 as s → s∗ , or the generator
does not extend to the interval [0, s∗ ) of the affine parameter, in which case the
maximal development is future null geodesically incomplete anyway. If the first
alternative holds, then either the generator extends to s∗ itself and the end
point is a point conjugate to o along the generator in question, or the generator
does not extend to s∗ , the end point being a singular point. The latter would
necessarily be the case if, for example, we can show that α(X(s), X(s)) does
not tend to zero as s → s∗ . But this cannot be shown without analyzing the
solution in the maximal development beyond M−1 . And off course the nature of
the future “boundary” of the maximal development, when incompletess holds,
remains an open question.
590
Bibliography
[Be] Bel, L. “Introduction d’un tenseur du quatrième ordre”, C. R. Acad. Sci.

Paris 248, 1297 - 1300 (1959).
[Bie] Bieri, L. An Extension of the Stability Theorem of the Minkowski Space
in General Relativity, Ph. D. thesis, Mathematics Department, ETH Zurich,
2007.
[Bir] Birkoff, G. D. Relativity and Modern Physics, Harvard University Press,
Cambridge Mass., 1923.
[B-L] Boyer, R. H. and Lindquist, R. W. “Maximal analytic extension of the
Kerr metric”, J. Math. Phys. 8, 265 - 281 (1967).
[Cho1] Choquet-Bruhat, Y. “Theorem d’existence pour certain systems d’equations
aux deriveés partielles nonlineaires”, Acta Mathematica 88, 141 - 225 (1952).
[Cho2] Choquet-Bruhat, Y. “The Cauchy problem”, in Gravitation: An Intro-
duction to Current Research, L. Witten (editor), Wiley, New York, 1962.
[Cho3] Choquet-Bruhat, Y. “Solutions C ∞ d’equations hyperboliques non
linéaires”, C. R. Acad. Sci. Paris 272, 386 - 388 (1968).
[Cho4] Choquet-Bruhat, Y. “Probleme des condition initiales sur un conoide
characteristique”, C. R. Acad. Sci. Paris, 3971 - 3973 (1963).
[C-G] Choquet-Bruhat, Y. and Geroch, R. P. “Global aspects of the Cauchy
problem in in general relativity”, Commun. Math. Phys. 14, 329 - 335 (1969).
[Chr1] Christodoulou, D. “Violation of cosmic censorship in the gravitational
collapse of a dust cloud”, Commun. Math. Phys. 93, 171 -195 (1984).
[Chr2] Christodoulou, D. “The problem of a self-gravitating scalar field”,
Commun. Math. Phys. 105, 337 - 361 (1986).
[Chr3] Christodoulou, D. Investigations in Gravitational Collapse and the
Physics of Black Holes, Ph. D. thesis, Physics Department, Princeton Univer-
sity, 1971.
[Chr4] Christodoulou, D. “Global existence of generalized solutions of the
spherically symmetric Einstein-scalar equations in the large”, Commun. Math.
Phys. 106, 587 - 621 (1986).
[Chr5] Christodoulou, D. “A mathematical theory of gravitational collapse”,
Commun. Math. Phys. 109, 613 - 647 (1987).
[Chr6] Christodoulou, D. “The formation of black holes and singularities in
spherically symmetric gravitational collapse”, Comm. Pure Appl. Math. 44,
339 -373 (1991).
[Chr7] Christodoulou, D. “Bounded variation solutions of the spherically sym-
metric Einstein-scalar field equations”, Comm. Pure Appl. Math. 46, 1131 -
1220 (1993).
[Chr8] Christodoulou, D. “Examples of naked singularity formation in the
gravitational collapse of a scalar field”, Ann. of Math. 140, 607 - 653 (1994).
[Chr9] Christodoulou, D. “The instability of naked singularities in the gravi-
tational collapse of a scalar field”, Ann. of Math. 149, 183 - 217 (1999).
591
[Chr10] Christodoulou, D. “On the global initial value problem and the issue
of singularities”, Class. Quantum Grav. 16 A, 23 - 35 (1999).
[Chr11] Christodoulou, D. “Self-gravitating relativistic fluids: a two phase
model”, Arch. Rational Mech. Anal. 130, 343 - 400 (1995).
[Chr12] Christodoulou, D. “Self-gravitating relativistic fluids: the continuation
and termination of a free phase boundary”, Arch. Rational Mech. Anal. 133,
333 - 398 (1996).
[Chr13] Christodoulou, D. “Self-gravitating relativistic fluids: the formation of
a free phase boundary in the phase transition from soft to hard”, Arch. Rational
Mech. Anal. 134, 97 - 154 (1996).
[Chr14] Christodoulou, D. The Formation of Shocks in 3-Dimensional Fluids,
EMS Monographs in Mathematics, EMS Publishing House, 2007.
[Chr15] Christodoulou, D. “Nonlinear nature of gravitation and gravitational
wave experiments”, Phys. Rev. Lett. 67, 1486 - 1489 (1991).
[Chr16] Christodoulou, D. “Notes on the geometry of null hypersurfaces”,
U.S. Copyright Office, Library of Congress, Registration Number TXu 832-728,
c 1991, Demetrios Christodoulou, All Rights Reserved.
[Chr17] Christodoulou, D. The Action Principle and Partial Differential Equa-
tions, Ann. Math. Stud. 146, Princeton University Press, 2000.
[C-K] Christodoulou, D. and Klainerman, S. The Global Nonlinear Stability of
the Minkowski Space, Princeton Mathematical Series 41, Princeton University
Press, 1993.
[Co] Constantin, P. “On the Euler equations of incompressible fluids”, Bull.
Amer. Math. Soc. 44, 603 - 621 (2007).
[D1] Dafermos, M. “Stability and instability of the Cauchy horizon for the
spherically symmetric Einstein-Maxwell-scalar field equations”, Ann. Math.
158, 875 - 928 (2003).
[D2] Dafermos, M. “The interior of charged black holes and the problem of
uniqueness in general relativity”, Comm. Pure Appl. Math. 58, 445 - 504
(2005).
[Ed] Eddington, A. S. “A comparison of Whitehead’s and Einstein’s formulas”,
Nature 113, 192 (1924).
[Ei1] Einstein, A. “Zür algemeinen Relativitätstheorie”, Preuss. Akad. Wiss.
Berlin, Sitzber. 44, 778 - 786; 799-801 (1915).
[Ei2] Einstein, A. “Die Feldgleichungen der Gravitation”,Preuss. Akad. Wiss.
Berlin, Sitzber. 50, 844 - 847 (1915).
[Fi] Finkelstein, D. “Past-future asymmetry of the gravitational field of a point
particle”, Phys. Rev. 110, 965 - 967 (1958).
[Fr] Friedrich, H. “The asymptotic characteristic initial value problem for
Einstein’s vacuum field equations as an initial value probelm for a first order
quasilinear symmetric hyperbolic system”, Proc. Roy. Soc. Lond. A 378, 401
- 421 (1981).
[Ge] Geroch, R. P. “The domain of dependence”, J. Math. Phys. 11, 437 -
439 (1970).
[G-K-P] R. P. Geroch, E. H. Kronheimer, and R. Penrose “Ideal points in
spacetime”, Proc. Roy. Soc. Lond. A 327, 545 - 567 (1972).
592
[G-T] Gilbarg, D. and Trudinger, N. S. Elliptic Partial Differential Equations
of Second Order, Springer-Verlag, 1977.
[H-E] Hawking, S. W. and Ellis, G. F. R. The Large Scale Structure of Space-
time, Cambridge Monographs on Mathematical Physics, Cambridge University
Press, Cambridge, 1973.
[Ke] Kerr, R. P. “Gravitational field of a spinning mass as an example of
algebraically special metrics”, Phys. Rev. Lett. 11, 237 - 238 (1963).
[Kr] Kruskal, M. D. “Maximal extension of the Schwarzschild metric”, Phys.
Rev. 119, 1743-1745 (1960).
[K-N] Klainerman, S. and Nicolò, F. The Evolution Problem in General Rela-
tivity, Progress in Math. Phys. 25, Birkhäuser, Boston, 2003.
[L] Lemaı̂tre, G. “L’univers en expansion”, Ann. Soc. Sci. Bruxelles I A53,
51 - 85 (1933).
[M] Müller zum Hagen, H. “Characteristic initial value problem for hyperbolic
systems of second order differential equations”, Ann. Inst. Henri Poincaré,
Phys. Theor. 53, 159 - 216 (1990).
[M-S] Müller zum Hagen, H. and Seifert, H. J. “On Characteristic initial value
and mixed problems”, Gen, Rel. Grav. 8, 259 - 301 (1977)
[M-T-W] Misner, C. W., Thorne, K.S., and Wheeler, J. A. Gravitation, W. H.
Freeman & Company, San Franscisco, 1973.
[O] Osserman, R. “The isoperimetric inequality”, Bull. Amer. Math. Soc. 84,
1182 - 1238 (1978).
[O-S] Oppenheimer, J. R. and Snyder, H. “On continued gravitational con-
traction”, Phys. Rev. 56, 455 - 459 (1939).
[P1] Penrose, R. “Conformal treatment of infinity”, in Relativity, Groups and
Topology, C. M. de Witt and B. de Witt (editors), Les Houches Summer School
1963, Gordon and Breach, New York, 1964.
[P2] Penrose, R. “Gravitational collapse and space-time singularities”, Phys.
Rev. Lett. 14, 57 - 59 (1965).
[P3] Penrose, R. “Structure of spacetime”, in Battelle Rencontres, C. M. de
Witt and J. A. Wheeler (editors), Benjamin, New York, 1968.
[P4] Penrose, R. Techniques of Differential Topology in Relativity, Regional
Conference Series in Applied mathematics, SIAM, 1972.
[P5] Penrose, R. “Gravitational collapse: the role of general relativity” Noovo
Cimento 1, 252 - 276 (1969).
[P6] Penrose, R. “Singularities and time-asymmetry”, in General Relativity, an
Einstein Centenary Survey, S. W. Hawking and W. Israel (editors), Cambridge
University Press, Cambridge, 1979.
[R] Rendall, A. D. “Reduction of the characteristic initial value problem to
the Cauchy problem and its applications to the Einstein equations”, Proc. Roy.
Soc. Lond. A 427, 221 - 239 (1990).
[Sc] Schwarzschild, K. “Über das Gravitationsfeld eines massenpunktes nack
der Einsteinschen Theorie”, Sitz. Deut. Akad. Wiss. Berlin, Kl. Math. - Phys.
Tech., 189 - 196 (1916).
[Sy] Synge, J. L. “The gravitational field of a particle”, Proc. R. Irish Acad.
A 53, 83 - 114 (1950).
593
[Sz] Szekeres, G. “On the singularities of a Riemannian manifold”, Publ. Mat.
Debrecen 7, 285 - 301 (1960).
[S-Y1] Schoen, R. and Yau, S. T. “The existence of a black hole due to
condensation of matter”, Commun. Math. Phys. 90, 575 - 579 (1983).
[S-Y2] Schoen, R. and Yau, S. T. “Proof of the positive mass theorem II”,
Commun. Math. Phys. 79, 231 - 260 (1981).
[S-Y3] Schoen, R. and Yau, S. T. “On the proof of the positive mass conjecture
in general relativity”, Commun. Math. Phys. 65, 45 - 76 (1979).
[T] Tolman, R. C. “Effect of inhomogeneity on cosmologicak models” Proc.
Nat. Acad. Sci. U. S. 20, 169 - 176 (1934).
[Wh] Wheeler, J. A. Geons, Black Holes & Quantum Foam, W. W. Norton &
Company, New York, London, 1998.
[Wi] Witten, E. “A new proof of the positive energy theorem” Commun. Math.
Phys. 80, 381 - 402 (1981).
[Z] Zipser, N. The Global Nonlinear Stability of the Trivial Solution of the
Einstein-Maxwell Equations, Ph. D. thesis, Harvard University, 2000.
594

Christodoulou 0805.3880

Uploaded by

Copyright:

Available Formats

Christodoulou 0805.3880

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Christodoulou 0805.3880

Uploaded by

Copyright:

Available Formats

The Formation

May 18, 2008

Chapter 1 : The Optical Structure Equations

Chapter 2 : The Characteristic Initial Data

Chapter 3 : L∞ Estimates for the Connection Coefficients

Chapter 4 : L4 (S) Estimates for the 1st Derivatives of the Connection

Chapter 5 : The Uniformization Theorem

Chapter 6 : L4 (S) Estimates for the 2nd Derivatives of the Connection

Chapter 7 : L2 Estimates for the 3rd Derivatives of the Connection

Chapter 8 : The Multiplier Fields and the Commutation Fields

Chapter 9 : Estimates for the Derivatives of the Deformation Tensors

Chapter 10 : The Sobolev Inequalities on the Cu and C u

Chapter 11 : The S-tangential Derivatives and the Rotational Lie

Chapter 12 : Weyl Fields and Currents. The Existence Theorem

Chapter 13 : The Multiplier Error Estimates

Chapter 14 : The 1st Order Weyl Current Error Estimates

Chapter 15 : The 2nd Order Weyl Current Error Estimates

Chapter 16 : The Energy-Flux Estimates. Completion of the Conti-

Chapter 17 : Trapped Surface Formation

r2,0 2(m2,0 − m1,0 )

then S2 becomes trapped before S1 reduces to a point on Γ0 . There is a future

T µν = ρuµ uν + p((g −1 )µν + uµ uν ), (4)

uµ being the fluid velocity, and

is the particle current. In the case of the two-phase model, p is a function of

σ = −(g −1 )µν ∂µ φ∂ν φ, (9)

The form of the smallness assumption on δ is specified in the precise form

Then the divergence of P (W ; X1 , X2 , X3 ) is the sum of −(divQ(W ))(X1 , X2 , X3 )

g(L, L) = −2Ω2 (20)

L̂ = Ω−1 L, L̂ = Ω−1 L, so that g(L̂, L̂) = −2 (21)

sup |α| ≤ O2 (δ −3/2 |u0 |−1 )

sup |β| ≤ O2 (δ −1/2 |u0 |−2 )

sup |ρ|, sup |σ| ≤ O3 (|u0 |−3 )

sup |β| ≤ O4 (δ|u0 |−4 )

sup |α| ≤ O5 (δ 3/2 |u0 |−5 ) (24)

Here α, α are the trace-free symmetric 2-covariant tensorfields on each Su,u

supCu0 δ 3/2 |u0 ||α|

supCu0 δ 1/2 |u0 |2 |β|

supCu0 |u0 |3 |ρ| , supCu0 |u0 |3 |σ|

supCu0 δ −1 |u0 |4 |β|

supCu0 δ −3/2 |u0 |9/2 |α| ,

1st order: L̃L R, L̃Oi R : i = 1, 2, 3, L̃S R

η = ζ + /d log Ω, η = −ζ + /d log Ω (40)

where ζ may be caled the torsion. It is the obstruction to integrability of the

The Optical Structure

1.1 The Basic Geometric Setup

Hc = {Su,u : u + u = c, u ∈ [u, δ)} (1.2)

Also, for c ∈ (2u0 , u0 + δ] we denote by Hc the spacelike hypersurface

Hc = {Su,u : u + u = c, u ∈ [u0 , u]} (1.3)

Note that for c ≤ u0 Hc is a spacelike hyperplane in the Minkowskian region

D = {(u, u) : u ∈ [u, c∗ − u), u ∈ [u0 , c∗ /2)}

for c∗ < u0 + δ. The timelike geodesic Γ0 corresponds to:

A0 = {(u, u) : u = u, u ∈ [u0 , c∗ /2)} (1.6)

(see Figures 1.1, 1.2).

(g −1 )µν ∂µ u∂ν u = 0, (g −1 )µν ∂µ u∂ν u = 0 (1.7)

It follows that the vectorfields L′ and L′ defined by:

L′µ = −2(g −1 )µν ∂ν u, L′µ = −2(g −1 )µν ∂ν u (1.8)

are future directed null geodesic vectorfields:

∇L′ L′ = 0, ∇L′ L′ = 0 (1.9)

Figure 1.1: Case c∗ ≥ u0 + δ