Gen Fun D

GENERALIZED FUNCTIONS LECTURES
Contents
0. Preface 1
1. The space of generalized functions on Rn 1
1.1. Motivation 1
1.2. Basic definitions 2
1.3. Remarks on operations on distributions 5
1.4. Translations of generalized functions 5
1.5. Derivatives of generalized functions 5
1.6. The support of generalized functions 6
1.7. Products and convolutions of generalized functions 7
1.8. Generalized functions on Rn 9
1.9. Generalized functions and differential operators 9
1.10. Regularization of generalized functions 10
2. Topological properties of Cc∞ (Rn ) 12
2.1. Normed spaces 12
2.2. Topological vector spaces 13
2.3. Defining completeness 14
2.4. Fréchet spaces 17
2.5. Sequence spaces 19
2.6. Direct limits of Fréchet spaces 20
2.7. Topologies on the space of distributions 21
3. Geometric properties of C −∞ (Rn ) 22
3.1. Sheaf of distributions 22
3.2. Filtration on spaces of distributions 24
3.3. Functions and distributions on a Cartesian product 26
4. p-adic numbers and `-spaces 27
4.1. Defining p-adic numbers 27
4.2. Misc. -not sure what to do with them (add to an appendix about
p-adic numbers?) 30
4.3. p-adic expansions 32
4.4. Inverse limits 32
4.5. Haar measure and local fields 33
4.6. Some basic properties of `-spaces. 34
Date: June 16, 2022.

1
GENERALIZED FUNCTIONS LECTURES 2
4.7. Distributions on `-spaces 36

4.8. Distributions supported on a subspace 36
5. Vector valued distributions 38
5.1. Smooth measures 38
5.2. Generalized functions versus distributions 39
5.3. Some linear algebra 40
5.4. Generalized functions supported on a subspace 42
6. Manifolds 43
6.1. Tangent space of a manifold 45
6.2. Types of maps between smooth manifolds. 46
6.3. Analytic manifolds and vector bundles 47
6.4. Sections of a bundle 49
6.5. Another description of vector bundles 50
7. Distributions on analytic manifolds and on smooth manifolds 51
7.1. Smooth sections of a vector bundle. 52
7.2. Distributions on manifolds 55
8. Operations on generalized functions 58
9. Fourier transform 62
10. Wave-front set 70
10.1. Definition of the wave-front set 71
10.2. Advanced properties of the wave-front set 72
10.3. Sketch of proof of the invariance to isomorphisms in the p-adic case 74
11. The Weil representation, the oscillator representation and an
application 76
12. Schwartz functions on Nash manifolds 79
12.1. Semi-algebraic sets and the Seidenberg-Tarski theorem 79
12.2. Nash manifolds 81
12.3. Schwartz functions on Nash manifolds 85
13. Invariant distributions 91
13.1. Proof for n = 1 91
13.2. Luna slice theorem, Frobenius reciprocity and Harish-Chandra
descent 92
13.3. Sketch of Proof for all n 92
References 92
0. Preface
??
1. The space of generalized functions on Rn
1.1. Motivation. One of the most basic and important examples of a generalized
function is the Dirac delta function. The Dirac delta function on R at point t is
usually denoted
 by δt , and while it is not a function, it can be intuitively described
∞ x = t ´∞
by δt (x) := , and by satisfying the equality δt (x)dx = 1. Notice
0 x 6= t −∞
that it also satisfies:

ˆ∞ ˆ∞
δt (x)f (x)dx = f (t) δt (x)dx = f (t).
−∞ −∞
The following are possible motivations for generalized functions:
• Many natural analytic operations, e.g. derivation and Fourier transform,

are not defined for arbitrary functions. In many cases, the reason is that
the result of the operation is in fact a generalized function.
• Every real function f : R → R can be constructed as an (ill-defined) sum
P
of continuum indicator functions f := f (t)δt .
t∈R
• In general, solutions to differential equations, and even just derivatives of
functions are not functions, but rather generalized function. Using the
language of generalized functions allows one to rigorize such notions.
• Generalized functions are useful in physics. For example, the density of a
point mass can be described by the Dirac delta function. This allows to
give a mathematical formulation to physical theories that unify both mass
distribution and point masses, by using mass distributions that are allowed
to be generalized functions.
• As a mathematical manifestation of the previous point, distributions can
be used in series theory, since sequences can be replaced by discrete com-
binations of δ-functions, and thus summation of series is integration of the
corresponding generalized function, and one can unify techniques of inte-
gration with techniques for series summation. For example, summation
by parts becomes a special case of integration by parts, and Fourier series
become a special case of Fourier Transform.
1.2. Basic definitions. In this book we will consider various spaces of functions
and functionals. Unless specified otherwise, all the functions and functionals will be
real-valued. All the statements below are also valid for complex-valued functions.
In order to define what is a generalized function we first need to introduce some
standard notation.
Definition 1.2.1 (Smooth functions of compact support).

(i) Denote by C ∞ (R) the space of smooth functions f : R → R, i.e. functions

that can be differentiated infinitely many times.
(ii) Define the support of a function f : R → R by
supp(f ) := {x ∈ R : f (x) 6= 0},
the closure of the set in which it does not vanish.

(iii) Denote by Cc∞ (R) ⊂ C ∞ (R) the space of smooth functions with compact sup-
port.
Definition 1.2.2 (Convergence in Cc∞ (R)). Given f ∈ Cc∞ (R) and a sequence
n=1 of smooth functions with compact support we say that {fn }n=1 converges
{fn }∞ ∞
to f in Cc (R) if:
∞
S
(i) There exists a compact set K ⊂ R for which supp(fn ) ⊆ K.
n∈N
(k) ∞
(ii) For every order k ∈ N0 , the derivatives (fn )n=1 converge uniformly to the
derivative f (k) .
Exercise 1.2.3. Fix a non-zero ρ ∈ Cc∞ (R). Show that

(i) the sequence x n
fn (x) := ρ(x) 1 +
n
converges in Cc∞ (R) to ρ(x)ex .
(ii) the sequence
ρ(x + n)
gn (x) :=
n
does not converge in Cc∞ (R), even though it satisfies condition (ii) of the
definition.
(iii) the sequence
sin(2n x)
hn (x) := ρ(x)
n
does not converge in Cc∞ (R), even though it satisfies condition (i) of the def-
inition, and converges uniformly to zero.
We can now define the notion of distributions (cf. [?, ??] and [Kan04, Section 2.3].
Definition 1.2.4 (Distributions). A linear functional ξ : Cc∞ (R) → R is con-

tinuous if for every convergent sequence {fm }∞ m=1 of functions fm ∈ Cc (R) we
∞
have
lim hξ, fm i = hξ, lim fm i.
m→∞ m→∞
We will usually use the notation hξ, f i instead of ξ(f ). We call a continuous linear
functional a distribution or a generalized function. The space of all generalized
functions on R is denoted by C −∞ (R) := (Cc∞ (R))∗ .
Remark 1.2.5. In §?? below we will define a natural topology on the space Cc∞ (R).
The convergence in this topology will be as in Definition 1.2.2, but this does not
define the topology uniquely since this topology is not first countable. We will show
that a linear functional on this topological space is continuous if and only if it
satisfies the condition in Definition 1.2.4.
Remark 1.2.6. For now the names generalized functions and distributions are
synonymous as there is no difference for R. We will discuss the difference in a
later part of the manuscript, when it will be relevant.
Example 1.2.7. For any a ∈ R, define δa ∈ C −∞ (R) by hδa , f i := f (a).
Recall that a function f is locally-L1 if the restriction to any compact subset in its
domain is an L1 function. We denote the space of such functions L1Loc . Given a
real-valued function f ∈ L1Loc (R) we define ξf : Cc∞ (R) → R to be the generalized
function
ˆ∞
hξf , φi := f (x) ∙ φ(x)dx.
−∞
Note that this integral converges as φ vanishes outside of some compact set K, and
(f φ)|K ∈ L1 (K). These are sometimes called regular generalized functions.
Exercise 1.2.8. For any f ∈ L1Loc (R), show that ξf is a well defined distribution.
Note that we have
C ∞ (R) ⊂ C(R) ⊂ L1Loc (R) ⊂ C −∞ (R),
where the last embedding is given by f 7→ ξf . This embedding motivates the name
generalized function.
Exercise 1.2.9. Prove that there exists a function f ∈ Cc∞ (R) which is not the
2
zero function. Hint: Use functions such as e−1/(1−x) .
n=1 weakly
Definition 1.2.10. We say that a sequence of generalized functions {ξn }∞
converges to ξ ∈ C −∞
(R) if for every f ∈ Cc (R) we have
∞
lim hξn , f i = hξ, f i.

n→∞
Note that in particular this definition applies to locally-L1 functions, since as we

have seen above they are contained in the space of generalized functions. Now
we can give an equivalent definition of the space of generalized functions - as the
sequential completion of Cc∞ (R) with respect to the weak convergence. For this we
need the notion of a weakly Cauchy sequence:
Definition 1.2.11. (i) A sequence {fn } in L1Loc (R) is called a weakly Cauchy
sequence if for every g ∈ Cc∞ (R) and > 0 there exists a number N ∈ N such
´∞
that for all m, n > N we have that (fn (x) − fm (x))g(x)dx < .
−∞
(ii) Two weakly Cauchy sequences are called equivalent if their difference weakly
converges to zero.
One can similarly define these notions for sequences of distributions.
Exercise 1.2.12. A sequence {fn } in C −∞ (R) is a weakly Cauchy sequence if and

only if for any g ∈ Cc∞ (R), the sequence hfn , gi converges.
However, weakly Cauchy sequences in Cc∞ (R) do not necessarily converge in Cc∞ (R).
Remark 1.2.13. One can define the space of generalized functions C −∞ (R) as the
space of equivalence classes of weakly Cauchy sequences in Cc∞ (R). As we will show
in ??, this definition is equivalent to Definition 1.2.4. It is important that we take
weakly Cauchy sequences rather than weakly Cauchy nets, since otherwise we would
get the full completion of Cc∞ (R), which is larger than C −∞ (R), as we will see in
??.
Exercise 1.2.14. Find a sequence of functions (fn )∞

n=1 in Cc (R) converging
∞
weakly to 0 that does not converge point-wise.
One can find a weakly Cauchy sequence that converges to the Dirac’s delta.
Definition 1.2.15. A sequence φn ∈ Cc (R) of continuous, non-negative, compactly

supported functions is said to be an approximation of identity if:
´∞
(1) φn satisfy φn (x) ∙ dx = 1 (that is have total mass 1), and
−∞
(2) for any fixed ε > 0, the functions φn are supported on [−ε, ε] for n suffi-
ciently large.
Exercise 1.2.16. An approximation of identity weakly converges to δ0 .
The reason for the name “approximation of identity” is that δ0 is the identity for
the convolution operation that we will define later.
Such sequences can be generated, for example, by starting with a non-negative,
continuous, compactly supported function φ1 of total integral 1, and by then setting
φn (x) = nφ1 (nx).
Exercise 1.2.17. Find an approximation of identity.
Note that given η ∈ C −∞ (R) of the form η = ξf , we can recover the value of f at
´∞
t via lim hξf , φn (x − t)i = lim f (x)φn (x − t)dx = f (t).
n→∞ n→∞ −∞
1.3. Remarks on operations on distributions. In general, many spaces of func-

tions can be defined as completions of Cc∞ with respect to various topologies. From
this point of view, in order to define an operation on functions such a space, it is
enough to define this operation for functions in Cc∞ and prove that it is continuous
with respect to the relevant topology.
For defining operations on distributions we will often use a different approach.

Suppose that we have an operation α on Cc∞ (R), and we would like to extend it to
generalized functions. We can try to do it in the following way. Given φ ∈ Cc∞ (R),
we can try to express hξα(f ) , ϕi in terms of the pairing of ξf with various functions in
Cc∞ (R). If we succeed, we can apply the same procedure to an arbitrary distribution
ξ in place of ξf . Let us now apply this approach to the notion of derivative.
1.4. Translations of generalized functions. For a function f ∈ C ∞ (R) and

t ∈ R denote by Rt f the translated function Rt f (x) = f (x − t). The motivation
behind the formula, including the minus sign, is that the graph of the translated
function is the translation of the graph by t.
This operation can be naturaly extended to generalized functions. For this purpose
observe that hξRt f , φi = hξf , R−t φi. This motivates the following definition.
Definition 1.4.1. Given a generalized function ξ, we define its translation Rt ξ by
hRt ξ, φi := hξ, R−t φi
Exercise 1.4.2. Define the notion of translation of a generalized function using

approximation by Cc∞ (R). In other words, prove that if {fn } is a weakly Cauchy
sequence in Cc∞ (R) then so is {Rt fn }.
1.5. Derivatives of generalized functions. Let f, φ ∈ Cc∞ (R). Since hξf 0 , φi =

´∞ 0
f (x) ∙ φ(x)dx we can use integration by parts to deduce that
−∞
ˆ∞
hξ , φi = f (x) ∙
f0 φ(x)|∞
−∞ − f (x) ∙ φ0 (x)dx.
−∞
−∞ = 0.
However, since φ and f have compact support, we know that f (x) ∙ φ(x)|∞
Thus, hξf 0 , φi = −hf, φ i. This motivates the following definition.
0
Definition 1.5.1. For any ξ ∈ C −∞ (R), define its derivative ξ 0 ∈ C −∞ (R) by
hξ 0 , φi := −hξ, φ0 i
for any φ ∈ Cc∞ (R).
For example, the derivative of δ0 can be intuitively described as



∞
 x → 0−
δ00 (x) := −∞ x → 0+



0 otherwise.
Example 1.5.2. We have

(n)
hδ0 , φi = (−1)n hδ0 , φ(n) i = (−1)n φ(n) (0).
Exercise 1.5.3. Find a function F ∈ L1Loc for which F 0 = δ0 as generalized func-

tions.
Exercise 1.5.4. Define the notion of derivative of a generalized function using

approximation by Cc∞ (R). In other words, prove that if {fn } is a weakly Cauchy
sequence in Cc∞ (R) then so is {fn0 }.
The following exercise enables to define the notion of derivative of a generalized

function using translations.
Exercise 1.5.5. Prove that ξ−R t

tξ
weakly converges to ξ 0 when t → 0. By this we
mean that for any test function φ ∈ Cc∞ (R) we have hξ 0 , φi = limt→0 h ξ−R
t
tξ
, φi.
1.6. The support of generalized functions. Let U ⊂ R be an open set and let
Cc∞ (U ) be the space of smooth functions f : U → R supported in some compact
subset of U . Given a compact subset K ⊂ R, we denote by CK ∞
(R) the space of
smooth functions f : R → R with supp(f ) ⊆ K. In particular CK ∞
(X) ⊆ Cc∞ (R)
for every compact K ⊆ R.
We cannot evaluate a generalized function at a point. Therefore, we cannot just
define its support as we did before for a function by supp(f ) := {x ∈ R | f (x) 6= 0}.
However, if for some neighborhood U ⊂ R we have for every f ∈ Cc∞ (U ) that
hξ, f i = 0, then it is natural to say that supp(ξ) ⊆ R r U . This leads us to the
following definition:
Definition 1.6.1. Let ξ ∈ C −∞ (R).

(i) For an open subset U ⊂ R we say that ξ vanishes on U if for any f ∈ Cc∞ (U )
we have hξ, f i = 0.
(ii) For ξ ∈ C −∞ (R) we define
[
supp(ξ) = R r {open U ⊂ R | ξ vanishes on U }
(iii) Denote by Cc−∞ (R) the space of distributions with compact support.
The following (difficult) exercise shows that the definition is meaningful.
Exercise 1.6.2 (*). Let ξ ∈ C −∞ (R).

(i) Let U1 , U2 be two open segments in R. Show that if ξ vanishes on U1 and U2
then ξ vanishes on U1 ∪ U2 . Hint: Use partition of unity.
(ii) Show that if I is a set of arbitrary cardinality and {Uα }α∈I is a collection of
open subsets of R with compact closures and ξ|Uα ≡ 0 for any α ∈ I then
ξ|Sα∈I Uα ≡ 0.
We will discuss this in more details and in larger generality in ?? below.
Remark 1.6.3. Note that supp(ξ) is always a closed set.
Example 1.6.4. The support of δ0 is {0}.

Remark 1.6.5. While the support of δ00 is also {0}, given some f ∈ C ∞ (R) for
which f (0) = 0 but f 0 (0) 6= 0, we get that hδ00 , f i = −hδ0 , f 0 i = −f 0 (0) 6= 0. In
other words, having f (0) = 0 is not enough to get hξ, f i = 0, for a distribution ξ
supported at {0}. However, if f vanishes at 0 with all its derivatives, it will imply
hξ, f i = 0 for any ξ supported at {0}, as follows from Exercise 1.6.7 below.
Exercise 1.6.6. Let ξ1 , ξ2 ∈ C −∞ (R) and a, b ∈ R. Show that:

(1) supp(aξ1 + bξ2 ) ⊆ supp(ξ1 ) ∪ supp(ξ2 ).
(2) supp(ξ) − supp(ξ)◦ ⊆ supp(ξ 0 ) ⊆ supp(ξ).
Exercise 1.6.7. Show that all the generalized functions ξ ∈ C −∞ (R) which are
P
n
supported on {0} are of the form ci δ (i) for some n ∈ N and ci ∈ R.
i=0
Hint: prove this in three steps.
(i) Show that there exists n such that ξ is bounded on the set {f | f (i) (x) < 1∀x ∈
R, ∀i < n}.
(ii) Show that there exists k ∈ N such that ξxk = 0, that is hξxk , f i = hξ, xk f i = 0
for every f ∈ Cc∞ (R).
P
k−1
(i)
(iii) From ξxk = 0 deduce that ξ = ci δ0 for some ci ∈ R.
i=0
1.7. Products and convolutions of generalized functions.
Definition 1.7.1. Let f ∈ C ∞ (R) and ξ ∈ C −∞ (R). We would like to have

´∞
(f ∙ ξ)(φ) = ξ(x) ∙ f (x) ∙ φ(x)dx. Thus, we define (f ∙ ξ)(φ) := ξ(f ∙ φ).
−∞
While we can multiply every smooth function f by any generalized function ξ, the
product of two generalized functions is not always defined. Notice that indeed the
product of two weakly Cauchy sequences is not always a weakly Cauchy sequence,
so we might not be able to approximate the product of two generalized functions
by the product of their approximations.
Recall that given two functions f, g, their convolution is defined by
ˆ∞
(f ∗ g)(x) := f (t) ∙ g(x − t)dt.
−∞
The convolution of two smooth functions is always smooth, if it exists. In addition,

if f and g have compact support, then so does f ∗ g:
Exercise 1.7.2.
(1) Show that supp(f ∗ g) ⊆ supp(f ) + supp(g), where supp(f ) + supp(g) is
the Minkowski sum of supp(f ) and supp(g). Thus f, g ∈ Cc∞ (R) implies
f ∗ g ∈ Cc∞ (R).
(2) Find an example in which the left hand side is strictly contained in the right
hand side.
Given f, g ∈ Cc∞ (R) we can write (f ∗ g)(x) = hξf , g̃x i, where g̃x (t) := g(x − t).
This motivates us to define the convolution ξ ∗ g as the function (ξ ∗ g)(x) = hξ, g˜x i.
Note that the convolution of a smooth function and a generalized function is always
a smooth function:
Exercise 1.7.3. Show that for φ ∈ Cc∞ (R) and ξ ∈ C −∞ (R) we get that ξ ∗ φ is a
smooth function.
Let us now define the convolution of two generalized functions. This will not be
defined for every pair of generalized functions, but for pairs such that at least one
of the generalized functions have compact support. Firstly, for f, g ∈ Cc∞ (R) we
would like to have ξf ∗ ξg = ξf ∗g . This means
ˆ∞ ˆ∞ ˆ∞
hξf ∗ ξg , φi = (f ∗ g)(x) ∙ φ(x)dx = f (t) ∙ g(x − t) ∙ φ(x)dtdx.
−∞ −∞ −∞
We would like like to express the right-hand side in terms of convolutions of dis-
tributions with functions. For this purpose, for a function h ∈ C ∞ (R) denote
h := h(−x). In these terms we have
ˆ∞ ˆ∞ ˆ∞ ˆ∞ ˆ∞
f (t)∙g(x−t)∙φ(x)dtdx = f (t) ˉ
g(x−t)∙φ(−x)dxdt = ˉ
f (t)∙(ξg ∗φ)(−t)dt ˉ
= ξf (ξg ∗ φ).
−∞ −∞ −∞ −∞ −∞
ˉ
Definition 1.7.4. We define hξf ∗ ξg , φi := hξf , ξg ∗ φi.
However, some formal justification is required. Given a compact K ⊂ R, we say ρ

is a cutoff function of K if ρ|K ≡ 1 and ρ|V ≡ 0, where R\V has compact closure.
Exercise 1.7.5. Let K and V be as above. Show that there exists a continuous
cutoff function. Hint: use Urysohn’s Lemma.
Thus, given some ξ ∈ Cc−∞ (R) with supp(ξ) ⊂ K we have that ξ(φ) = ξ(ρK ∙ φ).
This enables us to define ξ as a functional over all C ∞ (R) and not only on Cc∞ (R).
For every φ ∈ C ∞ (R) we define ξ(φ) = ξ(ρK ∙ φ) where K := supp(ξ) ⊂ R.
Exercise 1.7.6. Let ξ ∈ C −∞ (R). In an exercise above we showed: if φ ∈ Cc∞ (R)

then the convolution ξ ∗ φ is smooth. Show that if φ is smooth, and supp(ξ) is
compact, then ξ ∗ φ is still smooth.
To summarize, the convolution of two compactly supported distributions is well

defined and compactly supported, while the convolution of a compactly supported
distribution with an arbitrary distribution is well defined, but usually not compactly
supported.
Exercise 1.7.7. Show the following identities for any compactly supported distri-
butions ξ1 , ξ2 and ξ3 in C −∞ (R).
(1) δ0 ∗ ξ1 = ξ1 .
(2) δ00 ∗ ξ1 = ξ10 .
(3) ξ1 ∗ ξ2 = ξ 2 ∗ ξ 1 .
(4) ξ1 ∗ (ξ2 ∗ ξ3 ) = (ξ1 ∗ ξ2 ) ∗ ξ3 .
(5) (ξ1 ∗ ξ2 )0 = ξ1 ∗ ξ20 = ξ10 ∗ ξ2 .
Exercise 1.7.8. Let K ⊂ R be a compact set. Construct a function f ∈ Cc∞ (R)

such that f|K ≡ 1 and f|U ≡ 0 for some neighborhood K ⊂ U (Hint: convolve a
suitable approximation of identity with the indicator function of K).
1.8. Generalized functions on Rn . All the notions above make sense for func-
tions and generalized functions in several variables. The definitions and the state-
ments literally generalize to this case. For example, let us restate the definition of
convergence in Cc∞ (Rn ).
Definition 1.8.1 (Convergence in Cc∞ (Rn )). Given f ∈ Cc∞ (R) and a sequence
n=1 of smooth functions with compact support, we say that {fn }n=1 converges
{fn }∞ ∞
to f in Cc∞ (Rn ) if:

S
(1) There exists a compact set K ⊂ Rn for which supp(fn ) ⊆ K.
n∈N
(α)
(2) For every multi-index α, the partial derivatives (fn )∞
n=1 converge uni-
formly to the partial derivative f .
(α)
1.9. Generalized functions and differential operators. A differential equa-

tion is given by the equality Af = g, where A is a differential operator. Assume A
is a linear differential operator which is invariant under translations, i.e. we have
that ARt (f ) = Rt (Af ). Equivalently, we assume that A has constant coefficients,
e.g. Af := f 00 + 5f 0 + 6f .
Though classically one was interested in solutions in smooth functions, one can also
look for solutions in generalized functions. In this case, the right-hand side can also
be a generalized function. As we shall see, this approach is useful even if eventually
we are interested only in smooth functions.
An important special case is the equation Af = δ0 . Given a solution f = G, using
the invariance of A under translations, we get that AGx = δx , for Gx := Rx G.
Exercise 1.7.7 implies that A(f ∗ h) = (Af ) ∗ h for any two functions f, h and thus
A(G ∗ g) = AG ∗ g = δ0 ∗ g = g. Hence, we can find a general solution f for Af = g
by solving a single simpler equation AG = δ0 . The solution G is called Green’s
function of the operator.
Exercise 1.9.1. Let A be a differential operator with constant coefficients (i.e. as

above).
(1) Choose any solution for the equation AG = δ0 , and describe the conditions
G has to meet without using generalized functions.
(2) Without using generalized functions, explain the equation A(G ∗ g) = g we

got for the solution G.
∂2
(3) Solve the equation Δf = δ0 (where Δ = ∂x 2 is the Laplacian).
1.10. Regularization of generalized functions.
Definition 1.10.1. Let {ξλ }λ∈C be a family of generalized functions. We say

that the family is analytic if hξλ , f i is analytic as a function of λ ∈ C for every
f ∈ Cc∞ (R).
Example 1.10.2. We denote


xλ x>0
xλ+ := ,
0 x≤0
and define the family by ξλ := xλ+ for Re(λ) > −1. The behavior of the function
changes as λ changes: When Re(λ) > 0 we have a continuous function; if Re(λ) = 0
we get a step function and for Re(λ) ∈ (−1, 0), the function xλ+ will not be locally
bounded. However, for any λ with Re(λ) > −1, ξλ ∈ L1loc (R).
We would like to extend the definition analytically to the entire complex plane. To
do that we will have to leave the world of locally L1 functions.
Deriving xλ+ (both as a complex function or as we defined for generalized function)
we obtain ξλ0 = λ ∙ ξλ−1 . This is a functional equation which enables us to define
ξ0
ξλ := λ+1
λ+1
, and in this way extend ξλ to every λ ∈ C with Re(λ) > −2, except
λ = −1. We can use this procedure interatively to extend ξλ to the entire complex
plane. This extension is not analytic, but it is meromorphic: it has a pole in
λ = −1, and by the extension formula, in λ = −2, −3, . . ..
This is an example of a meromorphic family of generalized functions.
Definition 1.10.3. A family {ξλ }λ∈C of generalized functions is called meromor-

phic if there exists an entire function f on C such that the family f (λ)ξλ is analytic
(possibly after redefining it in a discrete set of points in C). In fact, the family ξλ
can be undefined in a discrete set of points, as in the case of meromorphic functions.
Exercise 1.10.4. Show that any meromorphic family of generalized functions can
be expanded into Laurent series in a punctured neighborhood of every pole, with
coefficients being generalized functions.
Exercise 1.10.5. Find the order and the leading coefficient of every pole of ξλ :=
xλ+ .
Example 1.10.6. For a given p ∈ C[x1 , . . . xn ], similarly to before set,


p(x , . . . x )λ x > 0
1 n
p+ (x1 , . . . xn )λ := .
0 x≤0
The problem of finding the meromorphic continuation of a general polynomial was

open for some time. It was solved by J. Bernstein (see [?]) by proving that there
exists a differential operator Dpλ+ := b(λ)∙pλ−1
+ , where b(λ) is a polynomial pointing
1
on the location of the poles.
Exercise 1.10.7. Solve the problem of finding an analytic continuation for p+ (x1 , . . . xn )λ
in the following cases:
(1) p(x, y, z) := x2 + y 2 + z 2 − a and a ∈ R.
(2) p(x, y, z) := x2 + y 2 − z 2 .
1There was a slightly earlier proof in [?, ?], using resolution of singularities.
2. Topological properties of Cc∞ (Rn )
We want to analyze the space of distributions C −∞ (Rn ). For this aim, we want to
introduce a topology on this space.
2.1. Normed spaces.
Definition 2.1.1. A normed space over R is a vector space V over R with a

function || ∙ || : V → R≥0 satisfying
(i) ||λv|| = |λ| ∙ ||v||
(ii) ||v + w|| ≤ ||v|| + ||w||
(iii) ||v|| = 0 ⇐⇒ v = 0
If we weaken (iii) to state only ||0|| = 0 we will get the definition of a semi-norm.
The norm defines a Hausdorff topology on V be letting the basis for open sets be
the open balls B(x, ε) := {v ∈ V | ||v − x|| < ε} for all x ∈ V , ε ∈ R>0 .
P
Example 2.1.2. (i) lp := {sequences xn in R | |xn |p < ∞}
(ii) Lp (R) := {measurable f : R → R | |f |p is integrable on R}.
(iii) C k (R) := functions with k continuous bounded derivatives,
k
X
||f || := sup |f (i) (x)|.
i=1 x∈R
There are several facts that we are used to in the Euclidean space Rn . For example,
the closed bounded sets are compact, linear subspaces are always closed, and linear
operators are always continuous. These properties still hold for finite-dimensional
normed spaces, but fail for infinite-dimensional ones.
Let us now discuss this in more detail. Fix a normed space V .
Exercise 2.1.3. If dim V is finite then the closed unit ball B := {v ∈ V | ||v|| ≤ 1}
is compact.
Corollary 2.1.4. Any finite-dimensional subspace of V is closed.
Proof. Let W ⊂ V be a finite-dimensional subspace. Let v ∈ W , and let K be the

ball in W with center at 0 and radius 2||v||. Then v lies in the closure of K. On
the other hand, K is compact and thus closed in V . Thus v ∈ K ⊂ W .
Exercise 2.1.5. Show that C 1 (R) ⊂ C 0 (R) is dense (and thus not closed).
Exercise 2.1.6. Find a non-continuous linear functional on C 0 (R). Hint: Show,

using Zorn’s lemma, that any linear functional on a subspace can be extended. Then,
use the embedding C 1 (R) ⊂ C 0 (R).
Exercise 2.1.7. Show that for any two normed spaces of the same finite dimension
there exists a linear homeomorphism between them.
Proposition 2.1.8. If the closed unit ball B ⊂ V is compact then dim V is finite.
S
Proof. Note that B can be covered by open balls of radius 1/2: B ⊂ x∈B B(x, 1/2).
If B is compact then this cover has a finite subcover. Denote the centers of the
subcover by {xi }ni=1 , and let W := Span({xi }ni=1 ). Then
n
[
B⊂ B(xi , 1/2) ⊂ W + 1/2B ⊂ W + 1/4B ⊂ . . .
i=1
Thus, for any k ∈ N, any v ∈ B can be presented as wk + zk where wk ∈ W and

||zk || < 2−k . Thus, v = lim wk . This shows that W is dense in V . Since W is
finite-dimensional, Corollary 2.1.4 implies that it is closed. Thus V = W and thus
dim V = n.
2.2. Topological vector spaces.
Definition 2.2.1. A topological vector space (or linear topological space) is a linear
space with a topology such that multiplication by scalar and vectors addition are
continuous. More precisely, the addition and the multiplication define continuous
maps:
(i) + : V × V → V ,
(ii) ∙ : R × V → V .
Remark 2.2.2. In this definition V is a vector space over R, but in the same way
one defines topological vector spaces over any topological field, e.g. over C or over
the field of p-adic numbers that we will define later.
Since addition is continuous, so is translation by a constant vector. This makes all

points of a topological vector space similar - the open neighborhoods of every point
x are, roughly speaking, the same as those of 0. Thus, in order to define a topology
on a vector space it is enough to give a basis for the topology at zero.
We assume the topological vectors spaces we consider are Hausdorff. Note that
given a non-Hausdorff space V , we can quotient V by the closure of {0} and get a
Hausdorff space.
Definition 2.2.3. Let V be a topological vector space over R.

(i) We say that a set A ⊆ V is convex if for every a, b ∈ A the linear combination
ta + (1 − t)b ∈ A for any t ∈ [0, 1].
(ii) We say that V is locally convex if it has a basis of its topology which consists
of convex sets.
(iii) We say that a set W ⊆ V is balanced if λW ⊆ W for all λ ∈ R satisfying
|λ| ≤ 1. Note that a convex set C is balanced ⇐⇒ it is symmetric (C = −C).
(iv) For every open convex balanced set C in V and x ∈ V we define a function
x
NC (x) = inf{α ∈ R>0 : ∈ C}.
α
Exercise 2.2.4. Let V be a topological vector space over R.

(i) Show that for every neighborhood U of 0 there exists an open balanced set W
such that 0 ∈ W ⊆ U .
(ii) (*)Find a topological vector space which is not locally convex (not necessarily
of finite dimension).
(iii) Prove that V is Hausdorff ⇐⇒ {0} is a closed set.
(iv) Show that if V is finite dimensional and Hausdorff then it is isomorphic to
Rn (as a topological vector space).
Remark 2.2.5. Note that the previous exercise shows that a linear topological space
satisfies the T1 separation axiom ⇐⇒ it satisfies T2 .
Exercise 2.2.6. Let C be an balanced open convex set in a topological vector space
V . Show that
(i) NC (x) < ∞ for all x ∈ V .
(ii) NC (x) is a semi-norm.
In a locally convex space we have a basis for the topology consisting of convex sets.
By Exercise 2.2.4(i) we can assume that all the sets in this basis are balanced.
Thus we obtain the next corollary.
Corollary 2.2.7 (Proof - exercise). The following are equivalent.

(i) V is locally convex.
(ii) There exists a collection of semi-norms nα on V such that the following sets
form a basis for the topology at zero:
Bα (r) := {x ∈ V | nα (x) < r}
(iii) There exists a collection of semi-norms nα on V such that the topology on V is

the weakest topology such that all the semi-norms nα are continuous functions
V → R≥0 .
Exercise 2.2.8. Find a locally convex (Hausdorff ) topological vector space V such
that V has no continuous norm on it. That is, every convex open set C contains a
line span{v}, so NC (v) = 0.
Note however that by Hausdorffness and Corollary 2.2.7, for every vector v ∈ V
there exists a semi-norm n on V such that n(v) > 0.
Generalizing the proof of Proposition 2.1.8, one can prove the following theorem.
Theorem 2.2.9 ([Rud06, Theorem 1.22]). Every locally compact topological vector
space has finite dimension.
2.3. Defining completeness. Recall that one can define the notion of Cauchy
sequence for a metric space, but not for a general topological space. However, one
can define it for any topological vector space V , in the following way.
Definition 2.3.1. A sequence (xn )∞

n=1 ⊂ V is called a Cauchy sequence, if for
every neighborhood U of 0 ∈ V there is n0 ∈ N such that m, n > n0 implies
xn − xm ∈ U .
Remark 2.3.2. More generally, if X has a uniform topology, then we can define a
notion of a Cauchy sequence. We will not give the definition of a uniform topology,
but we note that any topological group possesses a uniform topology, and indeed one
can define a notion of a left (resp. right) Cauchy sequence as follows: (xn )∞n=1 is
a Cauchy sequence if for every neighborhood U of e ∈ G there is an index n0 ∈ N
such that m, n > n0 implies x−1m xn ∈ U (resp. xn xm ∈ U ).
−1
Definition 2.3.3. Let V be a topological vector space. V is called sequentially

complete if every Cauchy sequence in it converges.
The notion of sequentially complete is well suited for topological vector spaces in
which the topology can be defined by sequences. Such are all the first-countable
(Hausdorff) topological spaces, but not all Hausdorff topological spaces, as we will
see from the following discussion.
Definition 2.3.4. A subset Y of a Hausdorff topological space X is called sequentially closed

n=1 in Y , the limit lies in Y .
if for every convergent sequence {yn }∞
Example 2.3.5. (i) Show that in a first-countable (Hausdorff ) topological space,

a set is sequentially closed if and only if it is closed.
(ii) Give an example of a first-countable (Hausdorff ) topological space, and a se-
quentially closed subset that is not closed. Hint. Take X be the real interval
[0, 1] with the co-countable topology on X, and Y := [0, 1).
Let us now define the general notion of a complete topological space.
Definition 2.3.6. Let V be a topological vector space.

(i) An embedding i : V ,→ W is called a strict embedding if i : V ,→ i(V ) is an
isomorphism of topological vector spaces.
(ii) V is called complete if for every strict embedding φ : V ,→ W , the image
φ(V ) is closed.
Exercise 2.3.7 (*). (i) Show that a complete topological vector space is always
sequentially complete.
(ii) Show that a first countable topological vector space V is complete if and only
if it is sequentially complete.
sec sec
??Solution: (i) Let V be the sequential completion of V . The map V → V is
a strict embedding ?? check how to define the topology on it.
(ii): Suppose that V is sequentially complete. Let κ be an ordered filtered set and
ϕ : κ → V s.t. for every neighborhood of zero U ⊂ V there exists α ∈ κ s.t. the
Minkowsky difference ϕ(κ≥α ) − ϕ(κ≥α ) lies in U . We want to prove that any such
?? converges. This means that there exists y ∈ V for any open neighborhood U 3 y
there exists β ∈ κ with ϕ(κ≥β ) ⊂ U . Indeed, choose a descending basis Ui of the
topology around zero. Let αi ∈ κ be such that ϕ(κ≥αi ) − ϕ(κ≥αi ) ⊂ h(Ui ), where
h(Ui ) is an open neighborhood of zero such that h(Ui ) + h(Ui ) ⊂ Ui . On the other
hand, ϕ(αi ) is a Cauchy sequence. Let y be its limit. Let us show that y satisfies
the condition. For that purpose, let U be a neighborhood of y. Let i be such that
ϕ(αi ) ∈ h(U ) and Ui ⊂ h(U ). We set β = αi . It is easy to see that it satisfies the
conditions above.
Now let ι : V → W be a strict embedding. We have to show that the image is
closed. Let y ∈ Imι. Let {Uα |α ∈ I} be a basis for topology at y. Let F (I) be
the set of all finite subsets of I, and for any f ∈ F (I) denote Vf := ∩α∈f Uα . For
any f ∈ F (I) choose xf ∈ f ∩ Imι. By the discussion above there exists z ∈ Imι
such that xf converges to z in the sense we discussed above. This implies (??) that
z = y.
Remark 2.3.8.
(i) Equivalently, we can define that a space V is complete if every Cauchy net is
convergent. From this definition it can be easily seen that any complete space
X is also sequentially complete.
(ii) In the category of first countable topological vector spaces, completeness is
equivalent to sequentially completeness, and indeed, there the notion of a
Cauchy net is equivalent to the notion of a Cauchy sequence, and a set Y ⊆ X
is closed ⇐⇒ it is sequentially complete.
Exercise 2.3.9. Find a sequentially complete space which is not complete. Hint:
see the above example.
Definition 2.3.10. Let V be a topological vector space. A space Vˉ is a completion

of V if Vˉ is complete and there is a strict embedding i : V → Vˉ and i(V ) is dense
in Vˉ .
Remark 2.3.11. We can also use a universal property in order to define the
completion of V . A strict (?) embedding i : V → Vˉ is a completion of V if:
(1) Vˉ is complete.
(2) For every map ψ : V → W where W is complete, there is a unique map
φW : Vˉ → W , such that ψ ≡ φW ◦ i.
Exercise 2.3.12 (*). Show that these two definitions of completeness are equiva-
lent.
It is often easier to show that a space is complete using the universal property. In
this way we avoid dealing with Cauchy nets or filters. However, in order to show
such completion exists one has to use these notions.
Exercise 2.3.13.
(1) (∗) Show that every Hausdorff topological vector space has a completion.
(2) Show that in the category of first countable topological vector spaces both
definitions of completion are equivalent to being sequentially complete.
2.4. Fréchet spaces. Reminder: A Banach space is a normed space, which is

complete with respect to its norm. A Hilbert space is an inner product space which
is complete with respect to its inner product.
Theorem 2.4.1. (Hahn-Banach) Let V be a normed topological vector space, W ⊆

V a linear subspace, and C ∈ R>0 . Let f : W → R be a linear functional such that
|f (x)| ≤ C ∙ kxk for all x ∈ W . Then there exists fe : V → R such that fe|W = f
and fe(x) ≤ C ∙ kxk for all x ∈ V .
Exercise 2.4.2. Let V be a locally convex topological vector space (i.e, not neces-
sarily normed), and let f : W → R be a continuous linear functional, where W ⊆ V
is a closed linear subspace of V . Show that f can be extended to V .
Definition 2.4.3. The space of all continuous functionals on a topological vector

space V is called the dual space and denoted by V ∗ .
Exercise 2.4.4. Let W ⊆ V be infinite-dimensional vector spaces. Show that any

linear functional on W can be extended to a linear functional on V . (There is no
topology in this exercise).
Definition 2.4.5. A topological space (X, τ ) is said to be metrizable if there exists

a metric which induces the topology τ on X.
Remark 2.4.6. Every normed space is Hausdorff and locally convex, since there is
a basis of its topology consisting of open balls, which are convex. We also know that
every normed space is metric. However, metrizability does not force local convexity
and vice versa.
Definition: A topological space X is called a Fréchet space if it is a locally convex,

complete space which is metrizable.
Exercise 2.4.7. Show that for a locally convex topological vector space V the fol-
lowing three conditions are equivalent.
(1) V is metrizable.
(2) V is first countable (that is it has a countable basis of its topology at every
point).
(3) There is a countable collection of semi-norms {ni }i∈N that defines the basis
of the topology of V , i.e, Ui, = {x ∈ V |ni (x) < } is a basis of the topology
at 0.
Hint: given a countable family of semi-norms define a metric by

∞
X ||x − y||k
d(x, y) :=
(1 + ||x − y||k )k 2
k=1
Exercise 2.4.8. Let V be a locally convex metrizable space. Prove that V is com-
plete (and consequentially is a Fréchet space) ⇐⇒ it is sequentially complete.
Recall that the completion of a normed space V with respect to its norm is the
quotient space Vˉ of all Cauchy sequences in X under the equivalence relation
(xn )∞
n=1 ∼ (yn )n=1
∞
⇐⇒ lim kxn − yn k = 0. In particular, Vˉ is a Banach
n→∞
space. Completing V with respect to a semi-norm N results in the elimination of
all elements {x ∈ V | n(x) = 0}. The quotient space equipped with the induced
norm on the quotient then yields a Banach space.
Example 2.4.9. Let V be the space of step functions on R, and consider the norm
kf k1 := R |f (x)|dx. The completion of V with respect to k ∙ k1 is isomorphic to the
´
Banach space L1 (R) (equipped with the norm on the quotient).
Let V be a Fréchet space, then we have a family of semi-norms {ni }i∈N on V . We

can form a new system of ascending semi-norms by replacing ni with max{nj }. Let
j≤i
Vi be the completion of V with respect to ni .
If ni and nj were norms (and not just semi-norms), which satisfy ∀x ∈ V, ni (x) ≥
nj (x), we would get a continuous inclusion Vi ,→ Vj . A sequence of ascending
norms n1 ≤ n2 ≤ . . . thus gives rise to a descending chain of completions V1 ←-
V2 ←- V3 ←- . . .. Our space V is then an inverse limit, V = limVi , which in this case
T ←
has a very nice description: it is an intersection V = Vi of the Banach spaces
i∈N
defined above.
If ni and nj are semi-norms, we get a continuous map Vi → Vj as every converging
sequence is mapped to a converging sequence which need not be injective. In this
case V will be the inverse limit limVi where the topology on V is generated by all
←
the sets of the form ϕ−1
i (Ui ) where Ui is an open set in Vi and ϕi : V = lim Vi → Vi
←
is the natural projection map which is part of the data of limVi .
←
Example 2.4.10. The following are examples of Fréchet spaces.

(1) V := C ∞ (S 1 ) is a Fréchet space. Define the norms {ni }i∈N by kf kni :=
max sup { f (j) (x) }. The completion with respect to nk is Vk = C k (S 1 ),
j≤i x∈S 1
the space of k-times differentiable functions. This family of norms satisfies
∀x ∈ V we have that nj (x) ≤ ni (x) if j ≤ i, so by the argument above we
T k 1
indeed have C ∞ (S 1 ) = C (S ).
k∈N
(2) V = C ∞ (R) is a Fréchet space. Define ni by kf kni := max sup { f (j) (x) }
j≤i x∈Ki
where Ki = [−i, i]. Notice that this gives an ascending chain of seminorms
so this defines a Fréchet space V = limVi . A similar argument shows that

←
C ∞ (Rn ) is a Fréchet space, as well as C ∞ (M ) for every smooth manifold
M . In these cases we take the supremum over all the possible directional
derivatives.
(3) Let K be a compact set and n ∈ N0 , then CK ∞
(Rn ) is a Fréchet space.
Let k ∈ N0 , CK (R ) is a Banach space and in particular a Fréchet space.
k n
Cc∞ (Rn ) is not Fréchet.
2.5. Sequence spaces. An important family of examples of Fréchet spaces are

sequence spaces.
Example 2.5.1. The space `p is the space of all sequences (xn )∞n=1 with values in
P
∞
R, such that |xn |p < ∞. It is a Banach space, and for p = 2 it is a Hilbert
n=1
space.
Let SW(N) be the space of all the sequences which decay to zero faster than any
polynomial, i.e. ∀n ∈ N, lim xi ∙ in = 0. A family of norms one can consider when
i→∞
analyzing these spaces is ||(xi )∞
n=1 ||n = sup{|xi ∙ i |}. It is not hard to see that with
n
i∈N
respect to these norms every Cauchy sequence converges. Define the topology on
SW(N) using by the family of norms || ∙ ||n , then SW(N) is a Fréchet space. This
is an example of a Fréchet space which is not a Banach space.
Remark 2.5.2. How can we see every Cauchy sequence converges? Why is not it
a Banach space?
The dual space SW(N)∗ is {(xi )∞i=1 | ∃n, c : |xi | < c ∙ i }. This is a union of Banach
n
spaces, as opposed to the intersection we had when defining the completion of a

Fréchet space (we will discuss the dual space more thoroughly next lecture). Note
that both SW(N) and SW(N)∗ contain the subspace of all sequences with compact
support (that is sequences with finitely many non-zero elements).
Example 2.5.3. Smooth functions on the unit circle, C ∞ (S 1 ), correspond to se-

quences (xi )∞ i=1 decaying faster than any polynomial. More precisely, we can view
f ∈ C (S ) as a periodic function in C ∞ (R) which can be written as f (x) =
∞ 1
P
∞
an ∙ eint (for functions of period 1). We thus attach to f the sequence (an )∞
n=1
n=−∞
where an decays faster then any polynomial.
Exercise 2.5.4.
(1) Show that the Fourier series map F : C ∞ (S 1 ) → SW(Z) via f 7→ an
is an isomorphism of Fréchet spaces. In other words, show that it is a
bijection and that for any semi-norm Pi of SW(Z) there exists a semi-
norm Sj of C ∞ (S 1 ) and C ∈ R such that for any f ∈ C ∞ (S 1 ) we have
that kF (f )kPi < C ∙ kf kSj (and vice-versa, or use Banach’s open mapping
theorem for Fréchet spaces).
(2) Define a Fréchet topology on
S(R) = {f ∈ C ∞ (R)| lim f (n) (x) ∙ xk → 0 for all k}.

x→±∞
2.6. Direct limits of Fréchet spaces.
Definition 2.6.1. The direct limit of an ascending sequence of vector spaces is the
S
space V∞ := Vn . This is not a Fréchet space, but a locally convex topological
n∈N T
vector space. A convex subset U ⊆ V∞ is open ⇐⇒ U Vn is open in Vn for all
n.
Every space of the form C ∞ (K) can be given the induced topology from C ∞ (R).
Taking the union of the ascending chain
C ∞ ([−1, 1]) ⊂ C ∞ ([−2, 2]) ⊂ . . .
gives all smooth functions with compact support Cc∞ (R) = lim C ∞ ([−n, n]) as a
n→∞
direct limit. However, this is not a Fréchet space - it is a direct limit and not an
inverse limit. A basis of the topology of Cc∞ (R) at 0 is given by the sets:
X
U(n ,kn ) := {f ∈ C ∞ (R) | supp(f ) ⊆ [−n, n] and |f (kn ) | < n },
n∈N
where n ∈ R0> , kn ∈ N0 and Σ denotes Minkowski sum, that is A+B := {a+b|a ∈

A, b ∈ B}.
Exercise 2.6.2. Show that a sequence (fn )∞ n=1 in Cc (R) converges to f ∈ Cc (R)
∞ ∞
with respect to the topology defined above if and only if it converges as was defined
in the first lecture (Definition 1.2.2), i.e.,
S
∞
(1) There exists a compact set K ⊆ R s.t. supp(fn ) ⊆ K.
n=1
(k)
(2) For every k ∈ N the derivatives fn (x) converge uniformly to f (k) (x).
Remark 2.6.3. Notice that the topology on Cc∞ (R) is complicated- it is a direct
limit of an inverse limit of Banach spaces!
Exercise 2.6.4. Show that taking a convex hull instead of a Minkowski sum (i.e.,
defining U(n ,kn ) := convn∈N {f ∈ C ∞ (R) | supp(f ) ⊆ [−n, n], f (kn ) < n }) will
result in the same topology. This shows that Cc∞ (R) is a locally convex topological
vector space (Note that this follows since this is a direct limit of Fréchet spaces).
Finally, Fréchet spaces have several more nice properties:

• Every surjective map φ : V1 → V2 between Fréchet spaces is an open map
(it is enough to demand that V2 is Fréchet and V1 is complete).
• In the previous item, defining K := ker φ, one can show that the quotient
V1 /K is a Fréchet space, and φ factors through V1 /K, that is φ : V1 →
V1 /K → V2 , where the map V1 /K → V2 is an isomorphism.
• Every closed map φ : V1 → V2 between Fréchet spaces can be similarly

decomposed. Firstly, by showing Im(φ) is a Fréchet space, and then by
writing V1 → Im(φ) → V2 .
2.7. Topologies on the space of distributions.
Remark 2.7.1. Let U ⊆ Rn be an open set, we define C −∞ (U ) to be the continuous

dual of Cc∞ (U ).
Definition 2.7.2.
(1) Let V be a topological vector space. A subset B ⊆ V is called bounded if for
every open U ⊆ V there exists λ ∈ R such that B ⊆ λ ∙ U .
(2) Denote V ∗ = {f : V → R : f is linear and continuous}. There are many
topologies one can define on V ∗ , we mention here two of those. Let > 0
and S ⊆ V , and set U,S = {f ∈ V ∗ : ∀x ∈ S, |f (x)| < }.
(a) The weak topology on V ∗ , denoted Vw∗ . The basis for the topology on
Vw∗ at 0 is given by:
Bw := {U,S : > 0, S is finite} .
(b) The strong topology on V ∗ , denoted VS∗ . The basis for the topology on
VS∗ at 0 is given by:
BS := {U,S : > 0, S is bounded} .
In particular, every open set in Vw∗ is open in Vs∗ .
By definition, a sequence {fn }∞n=1 in V converges to f ∈ V if and only if for every

∗ ∗
U,S ∈ B there exists N ∈ N s.t. (fn −f ) ∈ U,S for n > N . That is, ∀x ∈ S we have
that |fn (x) − f (x)| < . Therefore (fn )∞
n=1 converges to f w.r.t the weak topology
⇐⇒ it converges point-wise, and it converges to f w.r.t the strong topology ⇐⇒
it converges uniformly on every bounded set.
Remark 2.7.3. If the topology on V is given by a collection of semi-norms (such

as in Fréchet spaces), a set is bounded if and only if it is bounded with respect to
every semi-norm.
Theorem 2.7.4. (Banach-Steinhaus) Let V be a Fréchet space, W be a normed

vector space, and let F be a family of bounded linear operators Tα : V → W . If for
all v ∈ V we have that sup kT (v)kW < ∞ then there exists k such that
T ∈F
sup kT (v)kW < ∞.

T ∈F,kvkk =1
(we assume that ||v||k ≤ ||v||k+1 ∀k).
Example 2.7.5 (Fleeing bump function). Let V = R and let ψ be a bump function.
Notice that gn (x) = ψ(x + n) converges pointwise to 0 (and hence also in the weak
topology). Note that gn does not converge uniformly to 0, but it does converge
uniformly on bounded sets to 0, so it strongly converges to 0.
Assume V is a Fréchet space. Recall that we can define V as an inverse limit

T
∞
of Banach spaces V = Vi where Vi is the completion of V with respect to an
i=1
increasing sequence of semi-norms ni . If we dualize the system {Vi }∞
n=1 we get an
increasing sequence V1 ⊆ V2 ⊆ . . . ⊆ VS = limVi , and get that VS is a direct limit
∗ ∗ ∗ ∗ ∗
→
of Banach spaces (as a topological vector space).
Exercise 2.7.6. Let S ⊆ Cc∞ (R) be a bounded set, then there exists a compact
K ⊂ R such that S ⊆ CK
∞
(R).
Exercise 2.7.7. Consider the embedding Cc∞ (R) ,→ C −∞ (R), defined by f 7→ ξf .

Show that:
1) This embedding is dense with respect to the weak topology on C −∞ (R).
2) This embedding is dense with respect to the strong topology on C −∞ (R).
3) C −∞ (R) with the weak topology is sequentially complete but not complete.
4) C −∞ (R)w = Cc∞ (R)# where the latter is the full dual space (that is all function-
als, not necessarily continuous).
5) C −∞ (R)S is complete.
3. Geometric properties of C −∞ (Rn )
3.1. Sheaf of distributions.
Definition 3.1.1. Let U1 ⊆ U2 ⊆ Rn be open sets. Every function f ∈ Cc∞ (U1 )

can be extended to a function ext0 f ∈ Cc∞ (U2 ) by defining ext0 f |U2 \U1 ≡ 0, hence
we have an embedding Cc∞ (U1 ) ,→ Cc∞ (U2 ). This embedding defines a restriction
map C −∞ (U2 ) → C −∞ (U1 ), mapping ξ 7→ ξ |U1 , with hξ |U1 , f i = hξ, ext0 f i.
Remark 3.1.2.
(1) For an open U ⊂ Rn , the topology we defined on Cc∞ (U ) is generally not the
same as the induced topology when considering it as a subspace of Cc∞ (Rn ).
(2) For every compact K ⊂ U , we have CK ∞
(U ) ⊂ Cc∞ (U ). Here the topology
on CK∞
(U ) is indeed the induced topology from Cc∞ (U ).
Next we prove that with respect to the restriction operation for distributions defined
above, the space of distributions is equipped with a natural structure of a sheaf.
Lemma 3.1.3 (Locally finite partition of unity). Let I be an indexing set and
S
U = Ui be a union of open sets. Then there exist functions λi ∈ Cc∞ (U ) such
i∈I
that:
(i) supp(λi ) ⊂ Ui
(ii) For every x ∈ U , there exists an open neighborhood Ux of x in U and a finite
set Sx of indices such that λi |Ux ≡ 0 for all i ∈
/ S.
P
(iii) For every x ∈ U , i∈I λi (x) = 1.
Proof. Since Rn is paracompact, we can choose a locally finite refinement Vj of

Ui , i.e. a set J, a function α : J → I and an open cover {Vj }j∈J of U such
that Vj ∈ Uα(j) for any j ∈ J and any x ∈ U has an open neighborhood Ux that
intersects only finitely many Vj . Furthermore, we can assume that Vj are open
balls B(xj , rj ). Since the closures B(xj , rj ) are compact, there exist j such that
{B(xj , rj − j )}j∈J still cover U . For any j, let ρj be smooth non-negative bump
functions satisfying ρj |B(xj ,rj −j ) ≡ 1 and ρj |B(xj ,rj )c ≡ 0. For any i define

 P ρi (x) x ∈ Ui
j∈J ρj (x)
fi (x) = .
0 x∈ /U i
Note that the sum in the denominator is finite. Now, for every i define
X
λi (x) = fj (x).
j∈α−1 (i)
Theorem 3.1.4. With respect to the restriction map defined above, distributions
S
form a sheaf, that is given an open U ⊆ Rn , and open cover U = Ui , we have:
i∈I
(1) (Identity axiom) Let ξ ∈ C (U ). If for every i we have that ξ|Ui ≡ 0,

−∞
then ξ|U ≡ 0.
(2) (Glueability axiom) Given a collection of distributions {ξi }i∈I , where ξi ∈
C −∞ (Ui ), that agree on intersections, i.e. ∀i, j ∈ I we have that ξi |Ui ∩Uj ≡
ξj |Ui ∩Uj , there exists ξ ∈ C −∞ (U ) satisfying ξ|Ui ≡ ξi for any i.
P
Proof. Choose a locally finite partition of unity 1 = i∈I λi corresponding to the
cover Ui by Lemma 3.1.3.
(1) Given f ∈ Cc∞ (U ) we need to show hξ, f i = 0. Let fi := λi f . Then
Pn
f= fi , and
i=1
n
X n
X
hξ, f i = hξ, fi i = hξ, fi i = 0.
i=1 i=1
(2) Note that for any compact K ⊆ U we then have that λi |K ≡ 0 for all but
finitely many i. Now suppose we are given ξi ∈ C −∞ (Ui ) which agree on
pairwise intersections. For any f ∈ Cc∞ (U ) define
X
hξ, f i := hξi , λi f i.
i∈I
Since f is supported on some compact K this sum is finite. It is clear that

ξ is linear, we need to prove that it is continuous, and that ξ|Ui = ξi .
Let fn converge to f , where all functions lie in Cc∞ (U ). Then also

λi ∙ fn → λi ∙ f as the multiplication (f, g) 7→ f ∙ g is continuous. Since
S
∞
suppfn ⊆ K for some K ⊆ U , we have f λi ≡ 0 for all but finitely many
n=1
Pk Pk
indices i so we can write hξ, f i = i=1 hξi , λi f i and hξ, fn i = i=1 hξi , λi fn i
for any n. By continuity of ξi we get that hξi , λi ∙ fn i → hξi , λi ∙ f i and
therefore
X X
hξ, fn i = hξi , λi ∙ fn i → hξi , λi ∙ f i = hξ, f i,
i i
so ξ continuous. Now let f ∈ Cc∞ (Uj ),

then
X X X
hξ, f i = hξi , λi f i = hξj , λi f i = hξj , λi f i = hξj , f i,
i i i
where the second equality follows from the fact that λi f ∈ Cc∞ (Uj ∩ Ui )
and ξi |Ui ∩Uj ≡ ξj |Ui ∩Uj .
A second way to prove continuity of ξ is working with the open sets
in the topology of Cc∞ (U ). As ξi are continuous, they are bounded in
some convex open set 0 ∈ Bi , so |hξi , f i| < for every f ∈ Bi . Notice
S L
that conv( Bi ) is open in i∈I Cc∞ (Ui ), where each Bi is an open set in
i∈I L S
Cc∞ (Ui ) and hence a set in i∈I Cc∞ (Ui )) as conv( Bi ) ∩ Cc∞ (Ui ) = Bi .
L i∈I
Consider the map ϕ : i∈I Cc∞ (Ui ) → Cc∞ (U ) given by extension by zero
S
and summation. Note that B := ϕ(conv( Bi )) is open. Now let f ∈ B.
i∈I
P
n P
We can write f = ai fi where fi ∈ Bji and ai = 1. Therefore
P jP
i =1
ξ(f ) := ξi (ai fi ) < ai ∙ = and ξ is bounded on B.
3.2. Filtration on spaces of distributions.
Exercise 3.2.1. Let U ⊂ Rn , show that in Cc∞ (Rn ) we have
Cc∞ (U ) = {f ∈ Cc∞ (Rn ) : ∀x ∈

/ U, ∀ differential operator L, Lf (x) = 0} .
Consider U = Rn \Rk . We wish to describe the space of distributions supported on

Rk , which we denote CR−∞
k (R ). Notice that:
n
CR−∞
k (R ) = {ξ ∈ C
n −∞
(Rn )|∀f ∈ Cc∞ (Rn \Rk ) we have hξ, f i = 0}
and by continuity this is the same as:
{ξ ∈ C −∞ (Rn )|∀f ∈ Cc∞ (Rn \Rk ) we have hξ, f i = 0} = {ξ ∈ C −∞ (Rn )| ξ|V = 0},
where V = Cc∞ (U ) as described in Exercise 3.2.1. Notice that we can define a

natural descending filtration on V by:
∂if
V ⊆ Vm = {f ∈ Cc∞ (Rn )|∀i ∈ Nn−k
0 with |i| ≤ m we have | k
i R
= 0}.
(∂x)
We immediately see that f ∈ Vm implies f ∈ Vm−1 , hence this is a descending
chain. After dualizing, this defines an ascending filtration on CR−∞
k (R ) by:
n
Fm (CR−∞ n ⊥ −∞ n −∞
k (R )) = Vm = {ξ ∈ CRk (R ) : ξ|Vm = 0} ⊆ CRk (R ).
n
Exercise 3.2.2. Show the following.

T
(1) Vm = V = Cc∞ (Rn \Rk ).
m≥0
S
(2) Fm 6= CR−∞
k (R ).
n
m≥0
(3) Let U ⊆ Rn be open and U compact. Show that for every ξ ∈ CR−∞ k (R )
n
S∞ −∞
there exists ξ ∈ Fm such that ξ|U = ξ |U , thus i=0 Fi covers CRk (Rn )
0 0
locally.
(4) Show that Fn is stable under coordinate changes. More generally, let ϕ :
Rn → Rn be a smooth proper map that fixes Rk . Show that for every ξ ∈ Fi ,
ϕ∗ (ξ) ∈ Fi , where hϕ∗ (ξ), f i := hξ, f ◦ ϕi.
L ∂ i (C −∞ (Rk ))
Theorem 3.2.3. As vector spaces we have Fm ' ∂xi where i ∈ Nn−k
0
|i|≤m
∂ i (C −∞ (Rk )) ∂i
and ∂xi is the image of C −∞
(R ) under the differential operator
k
∂xi (note
that we only differentiate with respect to coordinates not lying in Rk ).
Proof. We prove here the statement for m = 0, and return to the case where m > 0
in section 5. Define a map res∗ : C −∞ (Rk ) → F0 by hres∗ ξ, f i = hξ, f |Rk i for every
ξ ∈ C −∞ (Rk ). Notice that res∗ ξ(f ) = 0 for any f ∈ F0 by definition so it is well
defined.
Furthermore, res∗ is injective since if hres∗ ξ, f i = hξ, f |Rk i = 0 for all f ∈ Cc∞ (Rn )
then ξ = 0 since the restriction res : Cc∞ (Rn ) → Cc∞ (Rk ) is surjective.
It is left to prove surjectivity. Define an extension map ext∗ : F0 → C −∞ (Rk )
by hext∗ η, f i = hη, ext(f )i where ext(f )|Rk = f for every f ∈ Cc∞ (Rk ). Note
that this is well defined since if we choose a different extension ext0 (f ) we get
that hη, ext0 (f ) − ext(f )i = 0 since (ext0 (f ) − ext(f ))|Rk ≡ 0 and thus ext∗ (η) =
ext0∗ (η). Also, we have that ext∗ η is a continuous functional, since we can choose
the extension ext(f ) in such a way that if (fn )∞ n=1 converges to f then (ext(fn ))n=1
∞
converges to ext(f ). Finally, note that since res ext (η) = η, we have that ext∗ is
∗ ∗
indeed surjective, and we are done.

L ∂ i (C −∞ (Rk ))
Remark 3.2.4. Note that if we now define Gm = ∂xi we get that
|i|≤m
Gm ' Fm /Fm−1 .
i −∞ k
Exercise 3.2.5. Show that Gm and G(i) = ∂ (C ∂xi(R )) , where i is some multi-
index, are not invariant under changes of coordinates, that is we might have that
ϕ(Gm ) 6= Gm and ϕ(G(i) ) 6= G(i) for a diffeomorphism ϕ : Rn → Rn .
3.3. Functions and distributions on a Cartesian product. Consider the nat-

ural map
ϕ : Cc∞ (Rn ) ⊗ Cc∞ (Rk ) → Cc∞ (Rn × Rk ), given by ϕ(f ⊗ g)(x, y) 7→ f (x)g(y).
Exercise 3.3.1. Show that this map is continuous and has a dense image.
Let us now define a natural map
Φ : C −∞ (Rn ) ⊗ C −∞ (Rk ) → C −∞ (Rn × Rk ) by

hΦ(ξ ⊗ η), F i := hη, f i, where f is given by f (y) := hξ, F |Rn ×{y} i.
Exercise 3.3.2. Show that this map is continuous and has a dense image.
Let us now denote by L(C −∞ (Rn ), C −∞ (Rk )) the space of all continuous linear
operators, and define a natural map
S : C −∞ (Rn × Rk ) → L(Cc∞ (Rn ), C −∞ (Rk )) by
h(S(ξ))(f ), gi := hξ, ϕ(f ⊗ g)i
Exercise 3.3.3. Show that the map S

(i) is continuous and has a dense image, where L(Cc∞ (Rn ), C −∞ (Rk )) is endowed
with the topology of bounded convergence.
(ii) maps C ∞ (Rn × Rk ) to L(Cc∞ (Rn ), C ∞ (Rk )) by the formula
ˆ
(S(f )(g))(y) = f (x, y)g(x)dx.
Rn
Remark 3.3.4. (i) Note that the map S is similar to the matrix multiplication.
(ii) The map S is in fact an isomorphism. This statement is the Schwartz kernel
theorem, see [Tre67, Theorem 51.7]
(iii) There are two natural topologies one can define on a tensor product: the injec-
tive one and the projective one. If the spaces are nuclear these two topologies
coincide. We will not define these notions in the present course, but all the
topological vector spaces we consider are nuclear, and thus our tensor products
possess natural topology. If we complete C −∞ (Rn ) ⊗ C −∞ (Rk ) with respect
to this topology, the map Φ will extend, and will become an isomorphism.
The analogous statement for the map ϕ does not hold, but it will hold if we
omit the compact support assumption. In other words, the extension of ϕ to
the completed tensor product C ∞ (Rn )⊗C ˆ ∞ (Rk ) by the same formula is an
isomorphism with C (R × R ), see [Tre67, Theorem 51.6].
∞ n k
4. p-adic numbers and `-spaces
One motivation to define the p-adic numbers comes from number theory. Assume
we are given a polynomial equation p(x) = 0 where p ∈ Z[x]. If it has an integral
solution x0 ∈ Z, then surely it satisfies the equation p(x) = 0 mod n for every
n ∈ N. Now, consider the converse question - if we know that it has a solution
modulo n for every n ∈ N, does it have an integral solution in characteristic zero?
In some cases, such as for quadratic forms, the answer, together with demanding
that there also exists a real solution, is yes (see the Hasse principle for more on
this). To know whether there exists a real solution, we can use simple methods
from analysis. The question of whether an equation has a solution mod n for every
n ∈ N can be simplified in two steps. Firstly, by the Chinese remainder theorem
it is enough to check whether the equation has a solution mod pn for every n ∈ N.
The second step is then by defining a ring Zp , called the ring of p-adic integers,
such that if there exists a solution x ∈ Zp it implies that there is a solution mod
pn for every n ∈ N. The field Q of p-adic numbers is then defined to be the field of
fractions of Zp .
A different motivation for introducing the p-adic numbers comes from a more an-
alytic point of view. One construction of the real numbers is via completing Q
with respect to its absolute value. An interesting question is whether this can be
generalized, that is what are the possible absolute value-like functions on Q and
their completions. It turns out that besides the standard and the trivial absolute
values, every absolute value (up to equivalence) is a p-adic absolute value (this is
essentially Theorem 4.1.8 below). The p-adic numbers are then obtained as the
completion of Q with respect to such an absolute value.
In this manuscript we take the second approach, starting with defining what prop-
erties we demand from an absolute value function.
4.1. Defining p-adic numbers.
Definition 4.1.1. A topological field is a field F , together with a topology, such

that addition, multiplication and the multiplicative and additive inverses are con-
tinuous operations with respect to this topology.
Definition 4.1.2. Given a field F , an absolute value is a function | ∙ | : F → R≥0

that satisfies:
(1) The triangle inequality : |x + y| ≤ |x| + |y|.
(2) |x||y| = |xy|.
(3) |x| = 0 ⇔ x = 0.
If furthermore |x+y| ≤ max{|x|, |y|}, we say that |∙| is a non-Archimedean absolute
value (and Archimedean otherwise).
For topological fields we demand the absolute value to be a continuous map. Notice
that every absolute value satisfies |1| = 1 (as |1| = |1| ∙ |1|, and |1| 6= 0).
Example 4.1.3. The following are absolute values:


0
x=0
(1) The trivial absolute value, defined by |x|0 :=
1 x 6= 0.

x x>0
(2) The standard absolute value on R: | ∙ |∞ =
−x x ≤ 0.
Definition 4.1.4. Let p be a prime number. We define the p-adic absolute value
of x ∈ Q by 
p−n , for x 6= 0,
|x|p =
0, for x = 0,
where x = pn ab and a, b ∈ Z, are coprime to p.
Exercise 4.1.5. Show that | ∙ |p is indeed an absolute value on Q, and that it is

non-Archimedean.
Definition 4.1.6. Two absolute values | ∙ | and | ∙ |0 on F are called equivalent, and
denoted | ∙ | ∼ | ∙ |0 if they induce the same topology on F .
Exercise 4.1.7. Let | ∙ | and | ∙ |0 be two absolute values on a field F . Show that
the following are equivalent:
(1) | ∙ | and | ∙ |0 are equivalent.
(2) There exists α ∈ R>0 such that | ∙ | = (| ∙ |0 )α .
(3) Every sequence which is Cauchy with respect to | ∙ | is Cauchy with respect
to | ∙ |0 .
Theorem 4.1.8. [Ostrowski’s Theorem] Every absolute value |∙| on Q is equivalent

to either | ∙ |p for a prime p, the standard absolute value | ∙ |∞ on Q induced from
R, or the trivial absolute value | ∙ |0 .
Proof. Let | ∙ | be an absolute value, we show it must be one of the above by

cases. Assume | ∙ | is non-Archimedean, i.e. |x + y| ≤ max{|x|, |y|}, and set a =
{x ∈ Z : |x| < 1}. This set is non empty as |0| = 0, and since we assume | ∙ | is
non-Archimedean it is an ideal of Z since,
(1) |x + . . . + x| ≤ |x|,
and thus if x ∈ a, meaning that |x| < 1, then xy = x + . . . + x ∈ a. Consider a

| {z }
y times
prime p. If |p| = 1 for every prime, we get that |x| = 1 for every 0 6= x ∈ Q,
as | p1 | = |p|−1 , and thus | ∙ | is the discrete absolute value. Thus we can assume
that there exists p ∈ a (note that for every integer |m| ≤ 1 by (1)), implying that
|p|
pZ ⊆ a ( Z and consequentially pZ = a. Now, if we put s = − log log p , we see that
|p| = p−s . Taking x = pn ab where a, b, n ∈ Z and a and b are coprime to p we get:
a a
|x| = |pn | = |pn | | | = |p|n = p−ns = |x|sp ,
b b
|{z}
=1
showing | ∙ | is equivalent to | ∙ |p .
Now, assume | ∙ | is an Archimedean absolute value. We must have that |n| ≥ 1 for
all non-zero integers n ∈ Z. Otherwise, let n be the smallest positive number such
that |n| < 1, and for every n < x ∈ N write it in base n:
(2) x = a0 + a1 n + a2 n2 + . . . + ar nr , for 0 ≤ ai ≤ n − 1, nr ≤ x.
We have |ai | ≤ ai ≤ n, thus

r
X r
X n(1 − |n|r+1 ) n
|x| ≤ |ai ni | ≤ n|n|i = ≤ .
i=0 i=0
1 − |n| 1 − |n|
Since n
1−|n| is independent of r, it bounds every x > n, and thus we must have that
|x| ≤ 1 for every x > n as otherwise |x|k > 1−|n| n
for k large enough. We get that
|x| ≤ 1 for x < n in the same way by considering n < xk for k large enough. But
this means that |x| ≤ 1 for all x ∈ Z, so by the previous step it is equivalent to a
p-adic absolute value, in contradiction to the fact that | ∙ | is Archimedean.
We can thus assume |n| ≥ 1 for all n ∈ N . Recall the number r defined in (2) and
log x
note that r ≤ log n . We now have:
Xr
log x log x
|x| ≤ |ai ||n| ≤ 1 +
i
n|n| log n .
i=0
log n
Using these bounds for xk :

log x k log x
|x|k ≤ (1 + k )n|n| log n ,
log n
implying s
r
X
1 log x 1
|x| log x ≤ |ai ||n|i ≤ k
(1 + k )n|n| log n .
i=0
log n
1 1
By taking k → ∞, we get |x| log x ≤ |n| log n . But by interchanging x and n we can
1 1
get that |n| log n ≤ |x| log x (note that if n < x we can repeat this process for x and
1 1 log |x|
nk for k large enough). Thus |x| log x = |n| log n = e log x is constant, implying that
|x| log |n|
s = log
log x = log n is constant. Now, note that |x| = x for every x and get that,
s
|x| = xs = |x|s∞ ,
finishing the proof.

Exercise 4.1.9. Show that given a field F and an absolute value | ∙ | on it the
topology it defines makes F a topological field, i.e. that addition, multiplication,
and the inverse operations are continuous.
Exercise 4.1.10. Non-Archimedean locally compact fields such as Qp have some

interesting properties. Prove the following two:
(1) For every open ball B(x, r) = {y ∈ Qnp : |xi − yi | < r} of radius r with
center x we have that B(x, r) = B(x, r) for every x0 ∈ B(x, r).
(2) Any two p-adic balls B(x, r) and B(x0 , r 0 ) in Qnp are either distinct or one
contains the other.
We can now define the p-adic numbers.
Definition 4.1.11. Let p be a prime number. We define the field of p-adic numbers
Qp to be the completion of Q with respect to the absolute value | ∙ |p .
Remark 4.1.12.
(1) The completion is defined just as we did in the case of the Archimedean
norm on Q; by equivalence classes of Cauchy sequences. Therefore, any
element a ∈ Qp is represented by a Cauchy sequence {an }∞ n=1 ⊂ Q with
respect to | ∙ |p .
(2) We get a space which is an uncountable field of characteristic 0, not alge-
braically closed, locally compact (every point has a compact neighborhood)
and totally disconnected, i.e. every connected component is a point.
Definition 4.1.13. We define the p-adic integers Zp to be the unit disc in Qp ,

explicitly Zp = {x ∈ Qp : |x|p ≤ 1}.
Exercise 4.1.14. Show that the p-adic absolute value extends from Q to Qp , that is
n=1 of elements in Q the limit lim |an |p
show that for every Cauchy sequence {an }∞
n→∞
exists.
Remark 4.1.15. Notice that just like R, this completion is not algebraically closed.
Try to find an equation in Qp for some p which does not have a solution Qp .
4.2. Misc. -not sure what to do with them (add to an appendix about
p-adic numbers?)
Theorem 4.2.1 (Taken from Koeblitz Theorem 2 page 11). Every equivalence
class a ∈ Qp for which |a|p ≤ 1 has exactly one representative Cauchy sequence of
the form {ai }∞
i=1 for which:
1) 0 ≤ ai < p for i = 1, 2, . . .
i
2) ai ≡ ai+1 (mod(pi )) for i = 1, 2, . . .
For the proof we will use the following lemma:

Lemma 4.2.2 (Taken from Koeblitz page 12). If x ∈ Q and kxkp ≤ 1, then for
any i there exists an integer α ∈ Z such that kα − xkp ≤ p−i . The integer α can be
chosen in the set {0, 1, 2, . . . pi − 1}.
Proof. Let x = a/b written in the form where (gcd(a, b) = 1). Since kxkp ≤ 1 it
follows that p does not divide b and therefore b and pi are relatively prime. Then
we can find m, n ∈ Z such that bm + npi = 1 . The intuition is that bm is close to
1 up to a small p-adic length so it is a good approximation to 1 so am is a good
approximation to a/b. So we pick α = am and get:
kα − xk = kam − a/bk = ka/bk ∙ kbm − 1k ≤ kbm − 1k = npi ≤ 1/pi

Note that we can add multiples of pi to α and still have
α − k ∙ pi − x ≤ max(1/pi , 1/pi ) ≤ 1/pi .
Therefore we can assume that α ∈ {0, . . . .pi − 1}.
Proof of Theorem 4.2.1. At first we prove the uniqueness: If {a0i } is a different

sequence satisfying (1) and (2) and if there exists i0 such that ai0 6= a0i0 then
ai 6= a0i (mod(pi0 )) for every i > i0 . Therefore kai − a0i k > 1/pi0 so {a0i }, {ai } are
not equivalent.
Now we prove existence: Suppose we have a Cauchy sequence {bi } ∈ Qp , we want
to find an equivalent sequence {ai } with the above property. Let Nj be the number
such that for every i, i0 > Nj we have kbi − bi0 k < p−j , and we can choose Nj to
be strictly increasing with j, and Nj > j. Observe that kbi k ≤ 1 if i > N1 . Indeed,
for all i0 > N1 we have that kbi − bi0 k < 1/p , kbi k ≤ max(kbi0 k , kbi − bi0 k) and for
i0 → ∞ we have that kbi0 k → kakp ≤ 1.
Now we use the lemma and get a sequence {aj } when 0 ≤ aj < pj such that
aj − bNj < p−j . We claim that{aj } is equivalent to {bi }, and satisfies the condi-
tions of the theorem. It indeed satisfies the conditions as:
kaj+1 − aj k = aj+1 − bNj+1 + bNj+1 − bNj − (aj − bNj )
≤ max( aj+1 − bNj+1 , bNj+1 − bNj , aj − bNj ) ≤ p−j

So aj+1 − aj has at least pj as a common divisor as required.
Furthermore, for any j and any i > Nj we have
kai − bi k = ai − aj + aj − bNj − (bi − bNj ) ≤ max(kai − aj k , aj − bNj , bi − bNj ) ≤ p−j .
So {ai } v {bi }.
Now, if we have some {a} ∈ Qp with kak ≥ 1 then there exists some m such that
ka ∙ pm k ≤ 1 and we have numbers with negative powers. Therefore we can present
the p-adic numbers as:

∞
X
Qp := { ai ∙ pi , where ai ∈ {0 . . . pi − 1}}.
i=−k
We define the ring of integers , denoted Zp as Zp := {x ∈ Qp | kxkp ≤ 1} or equiv-

P∞
alently Zp := { i=0 ai ∙ pi , where ai ∈ {0 . . . pi − 1}} or equivalently Zp := Zk kp -
the closure of Z with respect to the p-adic norm. Notice that Zp is indeed a ring
and that the only invertible elements are x ∈ Zp with kxkp = 1.
4.3. p-adic expansions. We want to write the p-adic expansions of elements q

in Q. If q ∈ N, that’s just writing its p-base expansion. For example, (126)5 =
” . . . 002001.” Let x := m
n be some rational number, with (n, m) = 1. It is enough
to describe the expansion when p - n (that is, when x ∈ Zp ∩ Q) as otherwise we
can multiply x by pk for some k, calculate the expansion, and move the point k
places to the left.
We can’t take remainder of x modulo p, as with integers. Instead, we can calculate
the fraction x = m n in Fpk for k ∈ N. Thus, the expansion of x in Qp is calculated
inductively:
• Write the digit x0 := [ m
n ] ∈ Fp .
• The nominator of the difference m n −x0 =
m−n∙x0
n is divisible by p. Redefine
our fraction to be x := p ∙ ( n − x0 ), and continue inductively.
1 m
Example 4.3.1. Calculate 12 ∈ Q7 . We start by solving the equation 2x0 =

1(mod7). The answer is x0 = 4. In the second step we calculate 17 ( 12 − 4) = x1 . So
2 ∙ (7x1 + 4) = 1(mod49). Therefore x1 = 3. We continue by induction and get the
required expansion.
Every ball in Qp is a disjoint union of p balls. For p = 2, the ball Z2 = Bc (0, 1) =

Bo (0, 2) consists of numbers with no digits to the right of the point. It’s a disjoint
union of two balls, B0 and B1 - where each Bi consists of all numbers ending with
S S
the digit 0 i0 . Similarly, B0 = B00 B01 , B1 = B10 B11 , where the elements in
Bij end with the digits 0 ij 0 . And so on.
This recursive structure implies p-adic integers are homeomorphic to the Cantor
set.
== Appendix material ends here ==
P
∞
Exercise 4.3.2. Show that an converges in Qp ⇐⇒ |an |p → 0.
n=0
Exercise 4.3.3. Show Zp is homeomorphic to the Cantor set as topological spaces,

where the Cantor set has its usual topology induced from the real numbers. In
particular this shows Zp is a compact set.
4.4. Inverse limits.

Definition 4.4.1. Let A1 ← A2 ← A3 ← . . . be a sequence of Abelian groups

{Ai } together with a system of homomorphisms {fij : Aj → Ai | j > i}, such that
fik = fij ◦ fjk , ∀i ≤ j ≤ k. An inverse limit of a sequence of Abelian groups is
defined by the collection of compatible sequences:
Y
lim
←− Ai = {a ∈ Ai : fij (aj ) = ai , ∀i ≤ j ∈ N}.
i∈N
Exercise 4.4.2. Prove the following:

(1) Let Ai := Z/pi Z, and let fij be the projection Z/pj Z → Z/pi Z. Show that
lim
←−Z/p Z ' Zp as a topological ring.
n
(2) Qp is the localization of Zp by p.

(3) Prove that Qp is homeomorphic to the Cantor set minus a point.
(4) Prove that Qnp and Qp are homeomorphic.
(5) Let U ⊂ Qnp be an open set. Show that either U is homeomorphic to the
Cantor set, or to Cantor set minus a point.
4.5. Haar measure and local fields. Let X be a topological space and let Cc (X)
be the space of continuous functions of compact support on X. Recall that the space
of continuous linear functionals Cc (X)∗ can be identified with the space of regular
Borel measures on X.
Theorem 4.5.1 (Haar’s theorem). Let G be a locally compact topological group.

Then:
(1) There exists a measure μ on G with values in C such that μ(U ) = μ(gU )
for any measurable set. Equivalently, there exists φ ∈ Cc (G)∗ such that for
any g ∈ G we have that φ(f ) = φ(fg ) where fg (x) = f (g −1 ∙ x).
(2) This measure is unique up to a scalar.
Exercise 4.5.2.
(1) Prove Haar’s theorem for (Qp , +).
(2) Given a Haar measure μ, we can define another invariant measure μa (B) =
μ(aB) for any a ∈ Qp . Show that μa = |a| ∙ μ.
Definition 4.5.3. A local field is a non-discrete topological field which is locally

compact.
Theorem 4.5.4. Any local field F is isomorphic as a topological field to one of the
following :
(1) R or C (if F is Archimedean).
(2) A finite extension of Qp for some prime p (if F is non-Archimedean of
characteristic 0).
P∞
(3) The field of formal Laurent series Fps ((t)) = { i=−k ai ti | ai ∈ Fps } for
some prime p and natural number s (if F is non-Archimedean of charac-
teristic p).
Proof. The main steps of the proof are as follows:

(1) Using Haar’s theorem define a measure on (F, +). Using this measure
we define an absolute value, up to scalar multiplication, that is, we set
|a| = α(a) where μa = α(a)μ.
(2) We show that for every local field the absolute value which was defined in
(1) defines its topology.
(3) We prove that every compact metric space is complete.
(4) Every local field of characteristic 0 contains Q and one of its completions.
This means that F contains R if it is archimedean, and Qp if it is non-
archimedean.
(5) We show that if F has characteristic 0, then F must be a finite extension
of R or Qp . Otherwise, if it is a non algebraic extension it must be non-
locally-compact.
(6) We show that any finite extension of Qp , R or Fq ((t)) is indeed a local field.
(7) For char(F ) 6= 0 we show that F contains a transcendental element, name
it t, and show that it contains Fq ((t)). We show that F must be a finite
extension of Fq ((t)).

4.6. Some basic properties of `-spaces.
Definition 4.6.1. An `-space X is a Hausdorff, locally compact and totally dis-

connected topological space.
Remark 4.6.2. We usually add the demand X is σ-compact, that is it is the union
of countably many compact spaces. Such a space is also sometimes called countable
at ∞.
Exercise 4.6.3. Find a compact `-space X and U ⊆ X such that U is not countable
at ∞.
Exercise 4.6.4. Show the following:

(1) Any non-archimedean local field is an `-space.
(2) Finite products, and open or closed subsets of an `-space are `-spaces. Note
that any subset of a totally disconnected topological space is totally discon-
nected.
Proposition 4.6.5. Let X be an `-space, then it has a basis of clopen (that is

closed and open) sets (i.e. it is zero-dimensional).
Proof. Taken from [AT08, 3.1.7]. Assume we have a point x ∈ W ⊆ K, with W

open and K = W compact and set Px = {U ⊆ K : U is clopen in K and x ∈ U }
T
and P = V . Note that K ∈ Px , thus P 6= ∅.
V ∈Px
Now, we claim that for every closed subset F of K such that F ∩ P = ∅ there
exists some W ∈ Px such that W ∩ F = ∅. Indeed, set η = {U ∩ F : U ∈ Px }. By
assumption, it is a family of non-empty closed subsets of F , and since F is compact if
T Tn T
n
V = ∅, then there is a finite collection of Vi such that Vi = Ui ∩ F = ∅
V ∈η i=0 i=0
(note that this is an equivalent characterization of compactness via closed sets).
Tn
Now set W := Ui ∈ Px . Since Px is closed under finite intersections, W ∈ Px .
i=0
Now we wish to show that P = {x}. Assume the contrary, i.e. P 6= {x}. P
is disconnected since X is totally disconnected, so there exists non-empty closed
x ∈ A and B such that A ∪ B = P and A ∩ B = ∅ which are open in K. Since K
is regular, (Hausdorff + locally compact implies regular), there exist open disjoint
sets U 3 U and V ⊇ B in K, where we have F = K\(U ∪ V ) closed in K and
P ∩ F = ∅. We showed that for such F we can find W ∈ Px such that F ∩ W = ∅.
Now, observe that the open set G = U ∩ W is also closed in K as,
G = U ∩ W ⊆ (K\V ) ∩ (K\F ) = K\(V ∪ F ) ⊆ U.
Therefore G ⊆ U ∩ W = G (W was closed). Since x ∈ G, we have G ∈ Px , but as

G ∩ B = ∅, we get that P = A ∪ B is not contained in G, which is a contradiction,
implying P = {x}.
Since for every open set x ∈ O in K the set K\O is compact and x ∈ / K\O, it
follows from the above claim that O contains some V ∈ Px .
Now, given an open set x ∈ A in X, we have that W ∩ A ⊂ K ∩ A, is open in K,
and thus contains a clopen U in the topology of K ∩ A from the above. Now, U is
closed in K and thus closed in X, and open in W ∩ A but contained in W ∩ A and
thus open in X. This finishes the proof.
Exercise 4.6.6. Show that every σ-compact, first countable `-space X is homeo-
morphic to one of the following:
(1) Countable (or finite) discrete space.
(2) Cantor set.
(3) Cantor set minus a point.
(4) Disjoint union of (2) or (3) with (1).
S
Definition 4.6.7. A refinement of an open cover Ui = X is an open cover
i∈I
{Vj }j∈J such that for any j, we have that Vj ⊆ Ui for some i.
Exercise 4.6.8.
(1) Let C ⊆ X be a compact subset of an `-space. Then any open cover has an
open compact disjoint refinement.
(2) Let X be a σ-compact `-space, then any open cover has an open compact
disjoint refinement.
4.7. Distributions on `-spaces.
Definition 4.7.1. Let X be an `-space. A function f : X → C is said to be smooth

if it is locally constant, that is for every point x ∈ X there is an open neighborhood
x ∈ U ⊆ X such that the restriction f |U is constant. Similarly to the archimedean
case, the space of smooth functions on X is denoted by C ∞ (X).
Proposition 4.7.2. Let X be an `-space. Show that smooth functions separate the
points in X. Assuming this, the Stone-Weierstrass theorem implies that C ∞ (X) is
dense in the space of all continuous functions C(X).
Proof. Let x, y ∈ X. As X is Hausdorff and has a basis of open compact, sets

there exists disjoint Ux and Uy which are compact and open. Set f|Ux = 1 and
f|X\Ux = 0. Then f is smooth and f (x) = 1, and f (y) = 0.
Definition 4.7.3. The space of smooth functions with compact support, Cc∞ (X) ⊂
C ∞ (X), are called Schwartz functions. We denote them by S(X). We also denote
Dist(X) = Cc∞ (X)∗ = S ∗ (X). We consider S(X) as a vector space, without any
topology.
Exercise 4.7.4. Let X be an `-space, show that S ∗ (X) is a sheaf.
Remark 4.7.5. In Rn , the Schwartz functions are the functions whose derivatives
decrease faster than any polynomial, and there is a strict containment Cc∞ (Rn ) ⊂
S(X) ⊂ C ∞ (Rn ). We will define them in the next lectures.
4.8. Distributions supported on a subspace. Recall that over R, the descrip-

tion of distributions on a space X that are supported on a closed subspace Z is
complicated (we did that using filtrations). Distributions on `-spaces behave much
better.
Definition 4.8.1. Let X be an `-space, we define the support of a distribution

T
ξ ∈ S ∗ (X) as we did for distributions on real spaces, by supp(ξ) = Dβ , where
ξ|Dc ≡0
β
Dβ ⊂ X are taken to be closed.
Proposition 4.8.2. (Exact sequence of an open subset). Let U ⊆ X be open and

set Z = X \ U . Then 0 → S(U ) → S(X) → S(Z) → 0 is exact.
Proof. It is clear that extension by zero S(U )→S(X) is injective, we show that
S(X)→S(Z) is onto. Let f ∈ S(Z). As f is locally constant and compactly
supported, we may assume that Z is compact and has a covering by a finite number
of open sets Ui (open in Z) with f |Ui = ci . Notice that each Ui is of the form
S
n
Ui = Wi ∩ Z, where Wi is open in X. Therefore, Z ⊆ Wi , and as Z is compact,
S i=1
m
we may refine {Wi }ni=1 and get that Z ⊆ j=1 Vj where Vj are open, compact
mutually disjoint and Vj ∩ Z ⊆ Wi ∩ Z = Ui for some i. We can thus extend f by

setting f (x) = ci for x ∈ Vj ⊆ Wi and zero otherwise.
It is left to prove exactness at S(X). Let f ∈ S(X) such that f |Z = 0. As f is
locally constant, there is an open set Z ⊆ V such that f |V = 0. This implies that
f is supported on Z c = U and therefore f |U ∈ S(U ).
Corollary 4.8.3. Let X be an l-space, and Z ⊂ X a closed subspace. Then:
(1) The inclusion i : S ∗ (Z) → SZ∗ (X) is an isomorphism.
(2) There is an exact sequence 0 → S ∗ (Z) → S ∗ (X) → S ∗ (X\Z) → 0.
Remark 4.8.4.
(1) Note that if we replace X by R, then the map i is not onto. For example,
(n)
for Z := {0} ⊂ R, the derivatives δ0 ∈ SZ∗ (Rn ) but they are not in the
image of i as in that case S ∗ (Z) ' C.
(2) This can be corrected by replacing S ∗ (X) by SZ∗ (X). Thus the following is
an exact sequence:
0 → SZ∗ (X) → S ∗ (X) → S ∗ (X\Z).

Exercise 4.8.5. Let V be a vector space (not necessarily finite dimensional) over
a field K, and L ⊂ V a linear subspace. Show that ∀f ∈ L∗ ∃g ∈ V ∗ such that
g|L ≡ f . Use Zorn’s lemma.
Proposition 4.8.6. Let X, Y be `-spaces. Given f1 ∈ S(X) and f2 ∈ S(Y ),
consider the bilinear map φ : S(X) ⊗ S(Y ) → S(X × Y ) via
(φ(f ⊗ g))(x, y) := f (x) ∙ g(y).
Then φ is an isomorphism of vector spaces.

Proof. It is easy to see that the image lies in the space of locally constant functions.
P
We first prove this map is surjective. Let f ∈ S(X × Y ), then f = cUi ×Vi and
by refining {Ui × Vi }i=1 we may assume that they are disjoint (note we are using
n
the fact that suppf is compact). Since each term cUi ×Vi ∈ Imφ we are done.
To show φ is injective, assume that
k
X k
X
φ( fi ⊗ gi ))(x, y) := fi (x) ∙ gi (y) = 0.
i=1 i=1
We can assume that {fi } are linearly independent and that {gi } are non zero and
that k is minimal with respect to these demands. If we take some y such that
Pk
g1 (y) 6= 0 we get that for any x ∈ X, we have i=1 fi (x) ∙ gi (y) = 0. This implies
that fi are linearly dependent. Contradiction. Hence gi ≡ 0 for all i, implying
fi ⊗ gi ≡ 0, contradicting the assumption that k is minimal.
Define a natural map
S : S ∗ (X × Y ) → HomF (S(X), S ∗ (Y )) by
hS(ξ)(f ), gi := hξ, φ(f ⊗ g)i.
Exercise 4.8.7.
(i) Show that the map S is a linear isomorphism.
(ii) Assume X = F n , Y = F m and let μ and ν be Haar measures on X and
Y . Embed C ∞ (X × Y ) ,→ S ∗ (X × Y ) and C ∞ (Y ) ,→ S ∗ (Y ) by multipli-
cation by the corresponding Haar measures. Then S maps C ∞ (X × Y ) to
HomF (S(X), C ∞ (Y )) by the formula
ˆ
(S(f )(g))(y) = f (x, y)g(x)μ.
X
Exercise 4.8.8. Consider the natural map
Φ : S ∗ (X) ⊗ S ∗ (Y ) → S ∗ (X × Y ) given by
hΦ(ξ ⊗ η), F i := hη, f i, where f is given by f (y) := hξ, F |X×{y} i.

(i) Show that Φ is not onto. Hint: take X = Y = Z.
(ii) Endow S(X × Y ) with the weak topology, i.e. ξn → ξ iff hξn , f i → hξ, f i for
every f ∈ S(X × Y ). Show that the map Φ has dense image.
5. Vector valued distributions
Definition 5.0.1. Let F be either R or C, let X be a locally compact space and

let V be a vector space over F . We define Cc∞ (X, V ) to be the space of smooth
functions on X with compact support with values in V . Here the smoothness of a
function is the usual coordinate-wise one.
Exercise 5.0.2. Let V be a topological vector space over F . Prove that Cc∞ (X, V ) ∼
=
Cc (X) ⊗F V as topological vector spaces, where the topology on Cc (X) ⊗F V is
∞ ∞
given by choosing a basis to identify V with F n and by then taking the product topol-
ogy on Cc∞ (X) ⊗F F n ∼ = (Cc∞ (X))n . In particular, this topology is independent of
a choice of a basis.
5.1. Smooth measures. Recall that a measure is a σ additive map from the σ-
algebra of Borel subsets of X into R. For us, the following characterization is
better:
Definition 5.1.1. Let X be a locally compact topological space. The space of signed
measures on X is Cc (X)∗ , i.e. all continuous functionals on Cc (X) (and all linear
functionals if X is an `-space). A signed measure is a measure if it is non-negative
on non-negative functions.
As the space Cc (X) is larger than Cc∞ (X), its dual is smaller. Explicitly, Cc (X)∗ ⊆
Cc∞ (X)∗ where the inclusion is the dual of the dense embedding Cc∞ (X) ,→ Cc (X).
If X is a group then in Cc (X)∗ there is a one-dimensional space of Haar measures,
which for X = Rn is just the space of multiples of the Lebesgue measure.
Remark 5.1.2. We usually consider the space of complex valued measures. As in

the real case, it can naturally be identified with Cc (X)∗ .
Definition 5.1.3. Let V be a locally compact vector space (note that it must be
finite dimensional as otherwise it is not locally compact). The space of Haar mea-
sures on V , denoted Haar(V ) ⊆ Cc (V )∗ , is the space of translation invariant mea-
sures (which exists by Haar’s theorem).
The fact that this space is one dimensional is non-trivial, but the intuition is as
follows: A Borel measure on V is determined by its value on cubes with rational
coordinates, as they form basis of the topology. It is not hard to see that if the
measure is translation invariant, the measures of these cubes are determined by the
measure of the unit cube.
Definition 5.1.4. Let V be a topological vector space. A measure μ on V is called

a smooth measure if μ ∈ C ∞ (V, Haar(V )), i.e. μ = f (x)h where f is smooth and h
is a Haar measure. We denote this space by μ∞ (V ), and the space of all compactly
c (V ).
supported measures inside it by μ∞
Exercise 5.1.5 (*). Let V be a vector space over a local field and let ξ ∈ Cc∞ (V )∗
be translation invariant. Prove that ξ is a Haar measure, i.e. show it is a mea-
sure (note that Cc∞ (V )∗ ! Cc (V )∗ so a-priori there might be translation invariant
distributions which are not measures).
Remark 5.1.6. Note that by definition μ∞ c (V ) ' Cc (V ) ⊗ Haar(V ) canonically.

∞
c (V ) ' Cc (V ) by choosing a Haar measure. This isomorphism

We also have that μ∞ ∞
is not canonical.
5.2. Generalized functions versus distributions. We are now in a position to

understand the difference between generalized functions and distributions.
A distribution on V is a continuous functional on the space of smooth functions
with compact support:
Dist(V ) := Cc∞ (V )∗ .
A generalized function is a continuous functional on the space of smooth measures
with compact support on V , i.e.
C −∞ (V ) := Cc∞ (V, Haar(V ))∗ .
As functions can be integrated against smooth measures of compact support, we

have a bilinear pairing
h∙,∙i
Cc∞ (V, Haar(V )) × Cc∞ (V ) → C.
Thus we have the following picture:

'
Cc−∞ (V ) ⇐⇒ Dist(V )
j↑ i↑
'
Cc (V )
∞
⇐⇒ μc (V )
∞
where the diagonals are dual to each other. Both inclusions i and j are obtained
via the pairing h∙, ∙i.
Exercise 5.2.1. Show that Haar(V ) ' Dist(V )V or equivalently that Dist(V )V is
one dimensional, for any finite dimensional vector space V over a local field F .
Definition 5.2.2. We can also define generalized functions with values in a vector
space by either:
1) C −∞ (V, E) := C −∞ (V ) ⊗ E
2) C −∞ (V, E) := Cc∞ (V, Haar(V ) ⊗ E ∗ )∗
and then C −∞ (V, Haar(V )) := C −∞ (V ) ⊗ Haar(V ) = Cc∞ (V )∗ = Dist(V ).
Exercise 5.2.3.
(1) Show that the two definitions of C −∞ (V, E) are equivalent.
(2) Describe an embedding Cc∞ (V, E) ,→ C −∞ (V, E).
5.3. Some linear algebra.
Definition 5.3.1. Let V be a finite dimensional vector space over a local field F .
(1) We define the exterior algebra as
∞
M
Λ(V ) = Λk (V ),
k=0
N
k
where Λ0 (V ) = F , and for k > 0 we have Λk (V ) = ( V )/Jk where Jk is
j=1
N
k
the vector space generated in V by the set
j=1
{v1 ⊗ . . . ⊗ vk : vi = vj for some i 6= j}.
(2) We define the symmetric algebra Sym(V ) as

∞
M
Sym(V ) = Symk (V ),
k=0
N
k
where Sym0 (V ) = F , and for k > 0 we have Symk (V ) = ( V )/Ik where
j=1
N
k
Ik is the vector space generated in V by the set
j=1
{v1 ⊗ . . . ⊗ vk − vσ(1) ⊗ . . . ⊗ vσ(k) : σ ∈ Sk }.

Note that this implies that the elements of the exterior algebra are anti-symmetric
(i.e. v ⊗ u = −u ⊗ v), and that Λk (V ) = 0 if k > dim V , since after choosing a
basis for V and decomposing an element in Λk (V ) to basic tensors, there must be
a basis element which appears at least twice.
Definition 5.3.2. Let V be an n-dimensional vector space over a local field F with
absolute value | ∙ |.
(1) We define the space of k-forms Ωk (V ) = Λk (V ∗ ).
(2) For a 1-dimensional space V we define a real vector space
|V | := {f : V ∗ → R : ∀α ∈ F, f (αv) = |α|f (v)}.
(3) We define the densities of V as
Dens(V ) := {f : V n → R : f (Av1 , . . . , Avn ) = | det(A)|f (v1 , . . . , vn )}.
Now, let Ωtop (V ) be the space of anti-symmetric n-forms on V . It is a one-

dimensional space, and Ωtop (V ) = Λn (V ∗ ).
Exercise 5.3.3. Let B be the space of bases of V .

(1) Show that Ωtop (V ) = {f : B → F : f (B1 ) = det(MBB12 )f (B2 )∀B1 , B2 ∈ B}
where MBB12 is the respective base changing matrix.
(2) Show that Ωtop (V ) = {f : V n →F : f (Av1 , . . . , Avn ) = det(A)f (v1 , . . . , vn )}.
Definition 5.3.4. For a finite dimensional real vector space V define the orienta-
tion line
Ori(V ) := {f : B → R : f (B1 ) = sign(det(MBB12 )) ∙ f (B2 )}.
Exercise 5.3.5. Using the tensor product of the natural maps Ωtop (V ) → Dens(V )
and Ωn (V ) → Ori(V ) show that Ωtop (V ) = Dens(V ) ⊗ Ori(V ).
Note that the orientation line is a linear space and not just two points as one is
used to think about orientations. However, we have two distinguished points in
Ori(V ), the two functions with absolute value 1. These are the usual orientations
we are used to thinking about.
Proposition 5.3.6. Show that there is a canonical isomorphism Dens(V ) ' Haar(V ).
Proof. A Haar measure can be viewed both as a functional on compactly supported,

continuous functions and as a function on a Borel algebra. The absolute value
of the determinant | det | : V n → R is an element of the one dimensional space
|Ωn (V )| (recall that for finite dimensional spaces V ∼ = V ∗∗ ). We have a canonical
isomorphism by choosing a basis {ei }i=1 for V , and bijecting between the element
n
ϕ ∈ |Ωn (V )| such that ϕ(e1 , . . . , en ) = 1 with the Haar measure normalized such
that it has the value 1 on the parallelogram spanned by the vectors {ei }ni=1 . This is
independent of choice of basis since given a different basis both elements would be
multiplied by the same factor of | det(M )|, where M is the change of basis matrix
with respect to these two bases.

(1) |L ⊗ M | = |L| ⊗ |M | for two one dimensional vector spaces L and M .
(2) |Ωtop (V )| ' Dens(V ).
(3) If W ⊆ V then Haar(W ) ⊗ Haar(V /W ) ∼ = Haar(V ).
(4) top ∼
If W ⊆ V then Ω (V ) = Ω (W ) ⊗ Ω (V /W ).
top top
(5) If F = R, then Ori(V ) ∼= Ori(W ) ⊗ Ori(V /W ).
5.4. Generalized functions supported on a subspace. Let W ⊆ V be lin-

ear spaces. We showed that over a non-archimedean local field F we have that
DistW (V ) = Dist(W ), and for F = R we described DistW (V ) for the case where
V = Rn and W = Rk . The goal now is to describe distributions on V supported
on W for any real linear spaces W ⊆ V . Recall we have defined a (non-exhausting)
filtration Vm (W ) on Cc∞ (V ) by
∂if
Vm (W ) = {f ∈ Cc∞ (V )|∀i ∈ Nn−k
0 where |i| ≤ m it holds that |
i W
= 0}.
(∂x)
where dim(V ) = n and dim(W ) = k. We then defined Fm (W ) ⊆ DistW (V ) by
∗
Fm (W ) = (Cc∞ (V )/Vm (W )) := {ξ ∈ Dist(V )|hξ, f i = 0 for any f ∈ FW
m
(V )}.
∗
Note that we have that Fm (W ) = Vm (W )⊥ where V2⊥ := (V1 /V2 ) for two vector
spaces V2 ⊆ V1 . We want to describe Fm (W )/Fm−1 (W ) in canonical terms, that is
such that the isomorphism will respect diffeomorphisms of V which preserve W .
Theorem 5.4.1. We have an isomorphism of vector spaces which commutes with

diffeomorphisms of V which preserve W :
Fm (W )/Fm−1 (W ) ∼
=can Cc∞ (W, Symm (W ⊥ ))∗ ' Dist(W ) ⊗ Symm (V /W ).
The proof of the theorem is based on the following lemma:
Lemma 5.4.2. Fm (W )/Fm−1 (W ) ∼

= (Vm−1 (W )/Vm (W ))∗ .
Proof. For any φ ∈ Fm (W ), the restriction φ|Vm−1 (W ) vanishes on Vm (W ), and

we send it to the induced functional on Vm−1 (W )/Vm (W ) which we denote by
e This is an injective morphism, since if φe = 0 then φ|V
φ. = 0 so φ ∈
m−1 (W )
Fm−1 (V )W . Surjectivity follows from the Hahn-Banach theorem in the following

∗
way: any ϕ ∈ (Vm−1 (W )/Vm (W ))∗ can be extended to ϕ e ∈ (Cc∞ (V )/Vm (W )) =
Fm (W ). Therefore [ϕ]
e + Fm−1 (W ) 7→ ϕ.
Hence, in order to prove the theorem it will be sufficient to prove that
Vm−1 (W )/Vm (W ) ∼
= Cc∞ (W, Symi (W ⊥ )).
For this we do the natural thing- we attach to f ∈ Vm−1 (W )/Vm (W ) its i-th
derivatives. Explicitly, we define:
Φ(f )(w)(v1 , . . . , vi ) = ∂v1 . . . ∂vi f (w).
It is well defined as f vanishes identically on W , so this form vanishes on all the

tangential derivatives. It is injective since if Φ(f ) = 0 then f vanishes with all of
its derivatives up to degree i, so it is in Vm (W ).
Exercise 5.4.3.
(1) Finish the proof of the lemma - show that Φ is onto, hence an isomorphism.
(2) Show that the isomorphism Fm (W )/Fm−1 (W ) ∼ =can Cc∞ (W, Symm (W ⊥ ))∗
is invariant with respect to diffeomorphism of (V, W ).
(3) Find ξ ∈ Dist(V \W ) such that there is no η ∈ Dist(V ) such that η|V \W =
ξ. That is, show that the natural map Dist(V ) → Dist(V \W ) is not onto.
To get a similar result for generalized functions, we twist by the one dimensional
space of Haar measures:
Fm (W )/Fm−1 (W ) ∼
= Cc∞ (W, Symm (W ⊥ ))∗ = C −∞ (W, Symm (W ⊥ ) ⊗ Haar(W )).
Take Gm (W ) = Fm (W ) ⊗ Haar(W )∗ ⊆ C −∞ (V ). We get by the compatibility of
tensor product and quotient the following:
Gm (W )/Gm−1 (W ) ∼
= C −∞ (W, Symm (W ⊥ ) ⊗ Haar(W )) ⊗ Haar(V )∗
∼
= C −∞ (W, Symm (W ⊥ ) ⊗ Haar(W ) ⊗ Haar(V )∗ ).
The next exercise shows that Haar(W ) ⊗ Haar(V )∗ can be presented in a simpler
manner:
Exercise 5.4.4. Let W ⊆ V .

(1) Show that Haar(W ) ⊗ Haar(V /W ) ∼ =can Haar(V ).
(2) Show that Haar(V ∗ ) ∼
=can Haar(V )∗ .
(3) Conclude that Haar(W ) ⊗ Haar(V )∗ ∼ =can Haar(W ⊥ ).
We arrive at the following corollary, yielding the desired description for generalized
functions.
Corollary 5.4.5. By the above argument it follows that:
Gm (W )/Gm−1 (W ) ∼
= C −∞ (W, Symm (W ⊥ ) ⊗ Haar(W ⊥ )).
6. Manifolds
After understanding generalized functions on vector spaces, we move to understand

generalized functions on spaces which locally look like vector spaces. For this we
define the notion of a manifold.
Definition 6.0.1. Let X be a topological space.

(1) A cover {Ui }i∈I is called locally finite, if for any x ∈ X there is a neigh-
borhood V such that V ∩ Ui 6= ∅ only for finitely many i ∈ I.
(2) X is called paracompact, if any open cover has a locally finite refinement.
(3) X is called a topological manifold if X is locally homeomorphic to Rn and
is Hausdorff and paracompact.
Exercise 6.0.2.
(1) Find a space X which is locally homeomorphic to Rn at every point and is
paracompact but is not Hausdorff.
(2) Find a space which is Hausdorff, locally isomorphic to Rn but is not para-
compact.
We now give a definition of a smooth manifold which is different than the usual
definition in differential topology and uses sheaves of functions.
Definition 6.0.3. A sheaf of (K-valued) functions F on a topological space X is

an assignment U 7→ F (U ) ⊆ {f : U → K| f is continuous} for every open U ⊂ X
such that:
(1) F (U ) is an algebra with unity.
(2) If f ∈ F (U ) and V ⊂ U then the restricted function f |V belongs to F (V ).
S
(3) (Gluing) For every open cover U = Ui , and every collection of functions
i∈I
{fi ∈ F (Ui )}i∈I such that fi |Ui ∩Uj ≡ fj |Ui ∩Uj for any i, j ∈ I, there exists
f ∈ F (U ) s.t. f |Ui ≡ fi for any i ∈ I.
Note that a sheaf of functions is a sheaf. The identity axiom is automatic.
Example 6.0.4. Continuous or smooth functions on a space X form a sheaf of

functions. So do locally constant functions.
Definition 6.0.5. A space with functions is a pair (X, F ), where X is a topological

space and F is a sheaf of functions on X. A morphism of spaces with functions
ϕ : (X, F ) → (Y, G) is a continuous map ϕ : X → Y such that for any open U ⊂ Y
and any function f ∈ G(U ), the composition f ◦ (ϕ|U ) lies in F (ϕ−1 (U ).
In the language of sheaves, the composition with ϕ defines ϕ# : G → ϕ∗ F
Definition 6.0.6. A smooth manifold is a space with a sheaf of functions (X, C ∞ (X))
such that X is a topological manifold and for every point x ∈ X there is an open
neighborhood U such that (U, C ∞ (X)|U ) ' (Rn , C ∞ (Rn )) as sheaves of functions.
Remark 6.0.7. The usual definition of manifolds adds an atlas to the structure
S
of X, that is an open cover X = Ui with diffeomorphisms φi : Ui → Rn . We
i∈I
also demand that φi ◦ φ−1
j is differentiable, so it seems like an additional demand
with respect to the definition above. Alas, further rumination shows that a pair
of isomorphisms ϕi : (Ui , C ∞ (Ui )) → (Rn , C ∞ (Rn )) and ϕj : (Uj , C ∞ (Uj )) →
(Rn , C ∞ (Rn )) implies that the following composition is an isomorphism:
#
ϕi ◦ ϕ−1
j |Ui ∩Uj : (Rn , C ∞ (Rn ))|ϕj (Ui ∩Uj ) → ϕi ◦ ϕ−1
j ∗
(Rn , C ∞ (Rn ))|ϕi (Ui ∩Uj ) .
In particular, by the next exercise we can deduce that ϕi ◦ ϕ−1

j |Ui ∩Uj is smooth
and is a diffeomorphism. Therefore Definition 6.0.6(2) is equivalent to the usual
definition of a smooth manifold.
Exercise 6.0.8.
(1) Show that C ∞ (Rn , Rk ) = {f : Rn → Rk : f ∗ (μ) ∈ C ∞ (Rn ) ∀μ ∈ C ∞ (Rk )}.
(2) Let M and N be smooth manifolds. Show that 6.0.6(1) is equivalent to the
usual definition of a morphism of smooth manifolds. That is, that a map
f : M → N is a smooth map of manifolds ⇐⇒ it is a morphism of ringed
spaces (where the sheaf is a sheaf of smooth functions).
Remark 6.0.9. Note that by a theorem of Whitney every n-dimensional manifold

can be embedded in R2n+1 . Thus we can always think about smooth manifolds sitting
in RN for N large enough.
6.1. Tangent space of a manifold. There are several equivalent ways to define
the tangent space to a smooth manifold M at a point x ∈ M . We first give a
categorical definition and then construct several objects which satisfy this definition.
Definition 6.1.1. We denote by ptMan the category of smooth pointed manifolds,

that is the objects are pairs consisting of a smooth manifold M and a point x ∈ M
and the morphisms are smooth maps of manifolds which preserve the distinguished
points.
Definition 6.1.2. A tangent space is a functor Tan : ptMan → Vect from pointed
smooth manifolds to vector spaces which satisfies the following conditions:
(1) The restriction of T an to the subcategory V ect ⊂ ptM an is the identity
functor.
(2) If f, g : (R, 0) → (C, 0) satisfy f 0 (0) = g 0 (0) then Tan(f ) = Tan(g).
(3) If ϕ : U ,→ M is an open embedding, then Tan(ϕ) is an isomorphism.
There are several structures that satisfy the above conditions:

(1) The space of all smooth paths Tx (M ) := {γ : ((−1, 1), 0) → (M, x)} modulo
the relation γ1 ∼ γ2 ⇐⇒ there exists a neighborhood U of x and an
isomorphism φ : U → Rn s.t. (φ ◦ γ1 )0 (x) = (φ ◦ γ2 )0 (x). It is easy to check
that this definition does not depend on the choice of (φ, U ).
(2) The space of derivations
Tx (M ) = {∂ : C ∞ (M ) → R : ∂ is linear, ∂(f ∙ g) = ∂f ∙ g(x) + f (x) ∙ ∂g}.

(3) Define mx := {f ∈ C ∞ (M ) : f (x) = 0}, and take Tx (M ) := (mx /m2x )∗ .
Exercise 6.1.3. Show the constructions of the tangent space given above are equiv-
alent.
Definition 6.1.4. Let φ : M → N be a smooth map. The differential of φ at

x ∈ M is the map dx φ : Tx (M ) → Tφ(x) (N ) defined by dx (φ)(γ) := φ ◦ γ for an
equivalence class of paths [γ] ∈ Tx (M ) .
Exercise 6.1.5.
(1) Show the differential is well defined, i.e. it does not depend on the repre-
sentative γ ∈ [γ].
(2) Show that given manifolds M, N, and K and maps φ : M → N and ψ :
N → K, the differentials satisfy dx (ψ ◦ φ) = dφ(x) (ψ) ◦ dx (φ).
6.2. Types of maps between smooth manifolds.
Definition 6.2.1. Let φ : M → N be a smooth map between smooth manifolds.

(1) φ is an immersion if dx φ is injective.
(2) φ is a submersion if dx φ is surjective.
(3) φ is a local isomorphism or étale if dx φ is an isomorphism.
(4) φ is an embedding if it is an immersion and defines a homeomorphism
M∼ = φ(M ).
(5) φ is a proper map if for every compact K ⊂ N , the preimage φ−1 (K) is
compact. In particular, in that case all the fibers of φ are compact in M .
(6) φ is a covering map if for every x ∈ N there exists a neighborhood U ⊆ N ,
such that φ|φ−1 (U ) : φ−1 (U ) → U is locally a diffeomorphism, and φ−1 (U ) ∼
=
U × D for some discrete set D.
Example 6.2.2.
(1) Let φ : [−1, 1] → R2 be a smooth path that slows to a stop at φ(0) = (0, 0),
but spends no time at (0, 0). That is, φ0 (0) = (0, 0), but φ(x) 6= 0 for all
x 6= 0 in some neighborhood [−ε, ε] of 0. Such a φ is locally injective at 0,
but since d0 φ = 0 it is not an immersion at 0.
(2) An immersion is not necessarily one-to-one. As an example, consider a
self-intersecting path φ : R → R2 with constant speed.
(3) Let L and D be finite dimensional linear spaces. The differential of a linear
map φ : L → V is φ itself. Thus, a one-to-one φ will be an immersion, an
onto φ will be a submersion, and an isomorphism of linear space will be an
étale map.
Exercise 6.2.3. Let M and N be smooth manifolds.

(1) Find a map φ : M → N which is an injective immersion, but is not an
embedding.
(2) Show that every proper map which is an injective immersion is a closed
embedding.
(3) Show that a proper map which is étale is a covering map, and that a covering
map with finite fibers is proper and étale.
Definition 6.2.4. Let X and Y be topological spaces. A fiber bundle is a map

p : X→Y , such that for every y ∈ Y there exists a neighborhood U ⊆ Y such that
p−1 (U ) ' U × Z for U ⊆ Y for some topological space Z.
Exercise 6.2.5. Show that a proper submersion is a fiber bundle.
6.3. Analytic manifolds and vector bundles. We would like to be able to talk
about manifolds for a general local field. In order to do so, for a non-archimedean
local field F we introduce the notion of an analytic F -manifold.
Definition 6.3.1. Let F be a non-archimedean local field. An analytic F -manifold

is a topological space M which is locally isomorphic to OFn together with the sheaf
of functions
X ~
An(U ) = {f : U → F : ∀x ∈ U, ∃r > 0 s.t. f|Br (x) (y) = a~k (x − y)k },
~
k∈Nn
~
where Br (x) is the ball of radius r around x, ~k is a multi-index, and (x − y)k =
Qn
(xi − yi )ki .
i=0
Remark 6.3.2. By Exercise 4.6.8 there is no need to use partition of unity for
F -analytic manifolds.
Example 6.3.3. There exist singular analytic manifolds, and any singular affine
algebraic variety is an example for such a manifold.
Definition 6.3.4. Let M be a smooth manifold or a p-adic analytic manifold.

A real vector bundle over M is a tuple (E, p) where E is a topological space and
p : E → M is a continuous surjection such that:
(1) For every x ∈ M we have a structure of a finite dimensional real vector
space on p−1 (x) = Vx .
(2) For every x ∈ M there exists an open x ∈ U and a local trivialization
ϕU : Vx × U → p−1 (U ) where ϕU is a homeomorphism (or diffeomorphism
if M is a real smooth manifold) and p ◦ ϕU (v, x) = x for all v ∈ Vx .
(3) The maps v 7→ ϕU (v, x) are linear isomorphisms.
If E ' V × M we say (E, p) is a trivial bundle over M .
If dim Vx = 1 for all x ∈ M we say (E, p) is a line bundle over M .
Exercise 6.3.5. It is known that the Mobius strip M is not homeomorphic to the
(finite) cylinder I × S 1 . By extending each segment I of M to R, we can define a
vector bundle E over the manifold S 1 . This way, points of E are such that the fiber
over θ ∈ S 1 is a line in R3 which intersects the z-axis with angle 0.5 ∙ θ. Define the
vector bundle above rigorously and show it is not diffeomorphic to the trivial bundle
S 1 × R.
Example 6.3.6. The tangent bundle of M = S 1 is T S 1 ' S 1 × R. The tangent

space at any point is one-dimensional, and changes smoothly as we move along the
circle. However, on M = S 2 the tangent bundle is not isomorphic to S 2 × R2 . This
happens since every vector field on S 2 vanishes at some point. (Hairy ball theorem).
Definition 6.3.7. Let (M, E) be a k-dimensional real vector bundle and π be its
projection. Given trivializing neighborhoods U and V , and trivializations ϕU : U ×
∼ ∼
Rk − → π −1 (U ) and ϕV : V × Rk − → π −1 (V ), one can consider ϕ−1 V ◦ ϕU : (U ∩
V ) × R → (U ∩ V ) × R . We can then write ϕ−1
k k
V ◦ ϕ U (x, v) = (x, gU,V (v)) where
gU,V ∈ GL(R ). The maps gU,V are called transition functions.
k
Notice that the set of transition functions gU,V , satisfy the cocycle conditions
gU,U (x) = Id and gU,V (x)gV,W (x) = gU,W (x).
Conversely, given a fiber bundle (E, X, π) of degree k with a transition map in

GL(Rk ) abiding the cocycle condition which acts in the standard way on the fiber
Rk , there is an associated a vector bundle. This is sometimes taken as the definition
of a vector bundle.
Proposition 6.3.8. Given a manifold M , vector bundles {(Ei , pi )}ni=1 each of

which with fiber of constant dimension mi over it, and a functor F : Vectn → Vect,
we can construct a vector bundle (F (E1 , . . . , En ), q) over M .
Proof. First, take a cover {Uα }α∈I of M which is a local trivialization of E (that
is, p−1 (Uα ) ' V × Uα ). Define the total space F (E) over each Uα by F (V ) × Uα ,
where the surjection q will be projecting to M , and glue every two pieces q −1 (Uα )
and q −1 (Uβ ) by setting (v, x) ∼ (gα,β (v), x) for every x ∈ Uα ∩ Uβ and v ∈ V ,
where gα,β = F (ϕ−1 Uβ ϕUα ). Finally, note that for any two elements of the cover
−1
gα,β = gβ,α , and that in order for our construction to be well defined we need to
show the cocycle condition, namely that gβ,γ gα,β = gα,γ when restricted to triple
intersections. This holds since
gβ,γ gα,β = F (ϕ−1 −1 −1

Uγ ϕUβ )F (ϕUβ ϕUα ) = F (ϕUγ ϕUα ) = gα,γ .
Note that if we want F (E) to have a smooth structure we need to demand that F
preserves smooth maps.
Example 6.3.9. Let E1 and E2 be two vector bundles over M . The direct sum
E1 ⊕ E2 is defined by applying our construction above to the direct sum functor
L
: Vect2 → Vect.
Exercise 6.3.10. Find two non-isomorphic bundles E and E 0 , such that E ⊕ F ∼

=
E ⊕ F for a bundle F (Hint: use vector bundles over S ).
0 2
Definition 6.3.11. Let M be a manifold.

(1) The tangent bundle of M is the disjoint union of its tangent spaces T M =
S
{x} × Tx M .
x∈M
(2) Given a submanifold N ⊆ M , and an embedding i : N → M , we define
the normal bundle to N in M to be NM N := i (T M )/T N , where i is the
∗ ∗
pullback of the bundle T M to N . Similarly, the conormal bundle to N in

M is CNM N := (NN ) .
M ∗
Example 6.3.12. For the sphere N = S 2 ⊂ R3 = M , the normal bundle at a

point is the normal line to it (i.e. the line passing at the point and at zero). It is
diffeomorphic to the trivial bundle on N .
Definition 6.3.13. For vector bundles E1 , E2 over a manifold (smooth or F -

analytic), we define the following:
(1) E1∗ .
(2) E1 ⊕ E2 .
(3) E 1 ⊗ E2 .
(4) For an embedding ϕ : E1 ,→ E2 , we define E2 /E1 .
Vk
(5) (E1 ), Symk (E1 ).
(6) We define the density bundle of E1 by Dens(E1 ).
Definition 6.3.14. Let M be either a smooth manifold or an F -analytic manifold,

we define its density bundle by Dens(M ) = |Ωtop (T M )|, that is the density bundle
of its tangent bundle.
6.4. Sections of a bundle. A set theoretic section of a function f : X → Y is a

function g : Y → X s.t. g ◦ f = idX .
Example 6.4.1. Let f : R2 → R be the projection f (x, y) = x. One example of a

section is g(x) := (x, sinx).
In many case sections of bundles give rise to important concepts:

• A section of the tangent bundle of a manifold is a vector field.
• A section of the k-th exterior power of the cotangent bundle of a manifold
is a differential form of degree k.
• A section of the density bundle is a density on the manifold.
• A section of the orientation bundle is a choice of an orientation on the
manifold.
Exercise 6.4.2.
(1) Show the every manifold has a Riemannian metric, i.e , an inner product
on tangent spaces
h∙, ∙ip : Tp M × Tp M → R
which varies smoothly.

(2) Let M be a smooth n-dimensional Riemannian manifold, that is a smooth
real manifold with a Riemannian metric. Construct explicitly a density over
M , that is a smooth section of the density bundle over M . The density
should respect coordinate changes, and be the standard density when M is
a linear space with the standard inner product.
Remark 6.4.3. We do not always have non-zero top differential forms on a man-
ifold M , and the Mobius strip is an example of a manifold with no non-zero top
differential form. However, we can always find a non-zero density on M . Since
with a density we can define a measure on the manifold, we can define integration
over manifolds.
6.5. Another description of vector bundles.
Definition 6.5.1. Let V be a finite dimensional vector space and X a topological

space.
(1) We define the constant sheaf V X to be the sheafification of the constant
presheaf, which assigns to every open set in X the vector space V .
(2) We say that a sheaf F over X is locally constant if for every x ∈ X there
exists an open x ∈ Ux and a finite dimensional vector space Vx such that
F|Ux ' Vx Ux .
Exercise 6.5.2. Let V be a finite dimensional vector space and X a topological

space.
(1) Show that V X (U ) consists of the locally constant functions from U to V .
(2) Show that if X is a σ-compact `-space then every locally constant sheaf F
such that Fx ' Fy for all x, y ∈ X is isomorphic to the constant sheaf.
Up to now we have used the Grothendieck definition of a sheaf. In some situations

the following definition is more useful.
Definition 6.5.3. A Leray sheaf on X is a pair (E, p) such that E is a topological

space and p : E → X is a homeomorphism locally in E, i.e. every point in E has
an open neighborhood U such that p(U ) is open and p|U defines a homeomorphism
U ' p(U ).
Theorem 6.5.4. The category of Leray sheaves is equivalent to the category of

Grothendieck sheaves.
Proof. Given a Leray sheaf (E, p) we define a Grothendieck sheaf by
F (U ) := {s : U → p−1 (U ) : f is continuous and p ◦ s = IdU }
with the obvious restriction maps.

F
For the other direction, given a Grothendieck sheaf F , we define E = x∈X Fx
with the natural projection map p : E → X. The basis of the topology of E is
given by Us,V = {(x, (s)x ) : x ∈ V } where V ⊆ X is open and s ∈ F (V ).
Exercise 6.5.5. Complete the proof by showing that this is indeed an equivalence
of categories.
Exercise 6.5.6.
(1) Show that covering spaces correspond to locally constant sheaves, and that
a covering space is trivial when it corresponds to a constant sheaf.
(2) Give an example of a locally constant sheaf arising from a covering space
which is not constant.
7. Distributions on analytic manifolds and on smooth manifolds
Definition 7.0.1. Let E be an F -analytic line bundle over an F -analytic manifold

X. Define a real vector bundle |E| as follows. As a set define |E| := {(x, v) :
x ∈ X, v ∈ |Ex |} and define a topology by giving C the discrete topology, so locally
E|U ' U × F and |E| |U ' U × |F | ' U × C. Hence, a base for the topology is
Vi,U,α = ϕi (U × {α}) where ϕi : U × C → |E| |U and α ∈ C.
Remark 7.0.2. Note that pe : |E| → X is a local homeomorphism as Vi,U,α ' U as

a topological space. Hence pe is a Leray sheaf. Its corresponding Grothendieck sheaf
is F (U ) := {continuous sections U → pe−1 (U )}. This is a locally constant sheaf
CX over X.
Definition 7.0.3. We can now define the density bundle over an F -analytic man-
ifold X in two ways:
Def 1 (Leray): Dens(X) := |Ωtop (X)|.
Def 2 (Grothendieck):
Dens(X)(U ) := {μ ∈ Measures(U )|∀ϕ ∈ OFn → U, there exists f ∈ C ∞ (OFn ) such that μ = ϕ∗ (f ∙Haar)}.
Lemma 7.0.4. Let ϕ : F n → F n be an analytic diffeomorphism, let f ∈ Cc (F n )

and let μ be a choice of a Haar measure on F n . Then
ˆ ˆ
hμ, f i =: f dx = (f ◦ ϕ) |det(Dx ϕ)| dx.
Exercise 7.0.5. Show that the above definitions are equivalent.

7.1. Smooth sections of a vector bundle. In this subsection we assume that

F = R and we are dealing with smooth manifolds.
Definition 7.1.1. We define smooth functions on a manifold M with compact
support and values in a vector bundle (E, π) by:
Cc∞ (M, E) := {f : M → E : π◦f = IdM and ∃K compact such that f |K C (m) = (m, 0)}.
Recall that Cc∞ (Rn , Rk ) = lim
−→CKm (R , R ) where Km is an increasing sequence of
∞ n k
S
∞
compact sets such that Km = Rn . We define a topology on Cc∞ (M, E) using
n=1
the topology on Cc∞ (Rn , Rk ):
Case 1- The trivial case: M ' Rn and E ' Rn × Rk with the projection to
the first component. Note that continuous sections from Rn to Rn × Rk are just
functions in Cc∞ (Rn , Rk ). Hence we give Cc∞ (M, E) the topology of Cc∞ (Rn , Rk ).
Exercise 7.1.2. Show that the above definition is well defined, i.e. it does not
depend on the isomorphism M ' Rn and E ' Rn × Rk → Rn . In other words,
show that:
(1) Given a diffeomorphism ϕ : Rn → Rn it induces a homeomorphism ϕ∗ :
Cc∞ (Rn , Rn × Rk ) → Cc∞ (Rn , Rn × Rk ) via precomposition.
(2) Given a smooth map ψ ∈ C ∞ (Rn , GLk (R)) we have that ψ∗ : Cc∞ (Rn , Rn ×
Rk ) → Cc∞ (Rn , Rn × Rk ) by ψ∗ (f ) = ψ ◦ f is a homeomorphism.
S
Case 2- General case: We can choose trivializing {Ui }i∈I such that M = Ui
i∈I
' '
where ϕi : Ui → Rn and ψi : E|Ui → Rn × Rk . We have a surjective map
M
ϕ: Cc∞ (Ui , E|Ui ) Cc∞ (M, E)
i∈I
by summation where surjectivity follows from partition of unity. We define the quo-
tient topology on Cc∞ (M, E) according to the map ϕ, that is, a set U ⊆ Cc∞ (M, E)
L ∞
is open if ϕ−1 (U ) is open in Cc (Ui , E|Ui ), where the latter is endowed with the
i∈I
direct sum topology.
Proposition 7.1.3. The topology on Cc∞ (M, E) is well defined. That is, the defi-
nition does not depend on the choice of the cover {Ui }i∈I of M .
Proof. We need to show that given a different cover {Vβ }β∈J of M which locally
trivializes M and E, we get the same topology.
Consider the cover {Wα,β } for Wα,β = Uα ∩ Vβ which refines both covers. We need
to show that for the addition map,
MM +
M
Cc∞ (Wα,β , E|Wα,β ) −
→ Cc∞ (Uα , E|Uα )
α∈I β∈J α∈I
a set in the range is open if and only if its preimage is open, where Wα,β ⊆ Uα ' Rn
and E|Wα,β ' E|Uα ' Rk . In order to show the above, it is enough to handle each
L +
case Cc∞ (Wα,β , Rk ) −
→ Cc∞ (Uα , Rk ) ' Cc∞ (Rn , Rk ) separately, since in the
β∈J L
direct sum topology a set is open if all the injections Di ,→ Di are continuous
(we furthermore assume our open sets are convex).
Given a basic open set U(Lm ,m ,Bm ) ⊆ Cc∞ (Uα , Rk ) where Lm are mixed differen-
S
∞
tiations, m ∈ R>0 and Bm are compact sets such that Bm = Rn , it is of the
P m=1
form U(Lm ,m ,Bm ) = VLm ,m ,Bm , where
m∈N

VLm ,m ,Bm = f ∈ C ∞ (Rn , Rk ) : supp(f ) ⊆ Bm , sup ||Lm (f )|| < m .
x∈Rn
P P P
l
Now, take a finite sum fβ ∈ +−1 (U{Lm ,m ,Bm } ) for fβ = f = fmi and
i=0
fmi ∈ VLmi ,mi ,Bmi . Let fβ = prβ (f ) be the projection of f into Cc∞ (Wα,β , Rk ),
mi −sup ||Lmi (fmi )||
and define N = #{β : prβ (f ) 6= 0} and 0mi = N and set 0m = Nm
if m 6= mi for all 0 ≤ i ≤ l. For Bm,β 0
⊆ Wα,β , compact sets which exhaust
Wα,β and such that Bm,β
0
⊆ Bm , the sets U(Lm ,0m ,Bm,β
0 ) are basic open sets in each
Cc (Wα,β , R ), and their direct sum is open in the direct sum. Now, we claim that,
∞ k
M
f∈ fβ + U(Lm ,0m ,Bm,β
0 ) ⊆+
−1
(U(Lm ,m ,Bm ) ).
β:fβ 6=0
P lβ
P
Given g = gβ where gβ ∈ U(Lm ,0m ,Bm,β
0 ) , then gβ = gβ,iβ where gβ,iβ ∈
β:fβ 6=0 iβ =1
V Ln i ,0n 0
,Bn .
β iβ iβ ,β
mi −sup ||Lmi (f )||
Thus if niβ = mi for some i, we have sup ||Lmi (gβ,mi )|| < 0mi = N
x∈Bmi
implying that,
X X X
sup Lmi (fmi ,β + gβ,mi ) ≤ sup Lmi (fmi ,β ) + sup ||Lmi (gβ,mi )||
x∈Bmi x∈Bmi x∈Bmi
β:fβ 6=0 β:fβ 6=0 β:fβ 6=0
X m − sup ||Lm (fm )||
< sup Lmi (fmi ) + i i i
x∈Bmi N
β:fβ 6=0
= mi .
Otherwise, if niβ 6= mi for all i, set n0 = niβ , and using the requirement sup ||Ln0 (gβ,n0 )|| <
x∈Bn0
n 0
N we note that:
X X n0
sup Ln0 (gβ,n0 ) ≤ sup ||Ln0 (gβ,n0 )|| < N = n0 .
x∈Bn0 x∈Bn0 N
β:fβ 6=0 β:fβ 6=0
P P l P P lβ
This allows us to conclude that f + g = fmi ,β + gβ,iβ lie in
β:fβ 6=0 i=1 β:fβ 6=0 iβ =1
P
U(Lm ,m ,Bm ) = VLm ,m ,Bm for all such functions g, implying that the ad-
m∈N
dition is continuous. For a less cumbersome approach, note that the embed-
L ∞
dings Cc (Wα,β , Rk ) →
− Cc∞ (Rn , Rk ) are continuous (a cookie for the person
β∈J
who finds a quick proof for this), so it is enough to show that the addition map
L ∞ n k + Lk
Cc (R , R ) − → Cc∞ (Rn , Rk ) ' Cc∞ (Rn ) is continuous. Since the domain
β∈J i=k
has the direct sum topology, it is enough to check this for a finite direct sum, which
follows by the continuity of addition in a topological vector space.
L ∞ +
To show the map is open, it is enough to consider CK,c (Wα,β , Rk ) − → CK ∞
(Rn , Rk ) '
β∈J
L
k
∞
CK (Rn ), for every compact K, and since the domain has the direct sum topol-
j=1
ogy and the basic open sets are finite sums of open sets in each coordinate, it is
L
m
+ L
k
enough to show it for a finite direct sum ∞
CK,c (Wi , Rk ) −
→ CK∞
(Rn ) where
i=1 j=1
S
m
K ⊂ Wi . Now, use partition of unity fi , with Ci = supp(fi ) ⊂ Wi where
i=1
P
m
fi |K ≡ 1 to get an onto map via the composition,
i=1
m
M m
M +
∞
CK∩C i
(Wi , Rk ) ,→ ∞
CK,c (Wi , Rk ) − ∞
→ CK (Rn ).
i=1 i=1
Since this is a continuous surjective map of Fréchet spaces, it must be open, implying
that the addition is open since the embedding is continuous.
We now give a different description of the topology of Cc∞ (Rn ). First observe that
f ∈ C(Rn ) is compactly supported if and only if f g is bounded for any g ∈ C(Rn ).
Now let D ∈ Diff(Rn ) be a differential operator on Cc∞ (Rn ). Define a seminorm
kf kD by sup |D(f )(x)|.
x∈Rn
Exercise 7.1.4. The topology on Cc∞ (Rn ) can be defined by the seminorms k kD .
Definition 7.1.5. Let M be a manifold and D : C ∞ (M ) → C ∞ (M ) be a map. We

S
say that D is a differential operator on M if for any trivializing cover Ui = M
i∈I
∼
and ϕi : Ui − → Rn we have ϕ−1i ◦ D ◦ ϕi ∈ Diff(Rn ). We denote the space of all
differential operators on M by Diff(M ).
We would like to define differential operators from C ∞ (M, E) to C ∞ (M, E 0 ), which

we denote by Diff(C ∞ (M, E), C ∞ (M, E 0 )). As before we divide the definition into
cases:
0
Case 1- the trivial case: Assume that E ' M × Rk and E 0 ' M × Rk . Then
0
Diff(C ∞ (M, E), C ∞ (M, E 0 )) ' Diff(C ∞ (M )k , C ∞ (M )k ) and the latter space is
isomorphic as a vector space to the space of k×k 0 matrices with values in Diff(C ∞ (M )).
Exercise 7.1.6. Show that the definition of the space of differential operators
Diff(C ∞ (M, E), C ∞ (M, E 0 )) does not depend on the isomorphisms E ' M × Rk
0
and E 0 ' M × Rk .
Case 2- the general case: Let A ∈ Hom(C ∞ (M, E), C ∞ (M, E 0 )). Then we say
that A ∈ Diff(C ∞ (M, E), C ∞ (M, E 0 )) if:
(1) For any f1 , f2 ∈ C ∞ (M, E) such that f1 |U = f2 |U , we have Af1 |U = Af2 |U .
(2) If E 0 |U is a trivialization then A|U ∈ Diff(U, E|U , E 0 |U ).
Definition 7.1.7 (Second definition to the topology on Cc∞ (M, E)). For D ∈
Diff(C ∞ (M, E), C ∞ (M, E)) define kf kD = sup |D(f )(x)|. Define the topology on
x∈M
Cc∞ (M, E) via
Cc∞ (M, E) = lim (Cc∞ (M, E)k∙kD ).
←D
Exercise 7.1.8. Given a manifold M and a vector bundle E over it show that the
two definitions of the topology on Cc∞ (M, E) are equivalent (one defined via taking
a cover of M and trivialization of E and the other through differential operators).
7.2. Distributions on manifolds.
Definition 7.2.1. Let M be a smooth or F -analytic manifold, and let E be a

smooth vector bundle over it (in the case of an analytic manifold it has the discrete
topology).
(1) The space of distributional E-sections is defined to be Dist(M, E) := Cc∞ (M, E)∗ .
(2) The space of generalized E-sections is defined to be C −∞ (M, E) = Dist(M, E ∗ ⊗
Dens(M )).
Although we do not have a natural injection from Cc∞ (M, E) to Cc∞ (M, E)∗ , we
have a natural injection
i : Cc∞ (M, E) ,→ C −∞ (M, E)
as follows: let μ ∈ Cc∞ (M, E ∗ ⊗ Dens(M )) and f ∈ Cc∞ (M, E). Note that
f ⊗ μ ∈ Cc∞ (M, E ∗ ⊗ E ⊗ Dens(M )),
that is, f ⊗ μ(m) = f (m) ⊗ μ(m). Note that we have a natural map
q : Cc∞ (M, E ∗ ⊗ E ⊗ Dens(M )) → Cc∞ (M, Dens(M ))
by pairing E with E ∗ and a natural map

ˆ
: Cc∞ (M, Dens(M )) → C
by integrating over M according to the measure defined by the section of the density
bundle. We define ˆ
hi(f ), μi := q(f ⊗ μ).
M
Therefore, the definition of generalized sections indeed generalizes smooth sections.
Proposition 7.2.2. Let X be either a smooth or an F -analytic manifold. Then

w
Cc∞ (X) = C −∞ (X).
Proof. Recall that C −∞ (X) = μ∞ c (X) . Given a topological vector space V , for
∗
W ⊆ V the space W is dense with respect to the weak topology if and only if
∗
W ⊥ = {v ∈ V : hw, vi = 0 ∀w ∈ W } = {0}. To see the relevant direction, if

W ⊥ = {0}, we will show that for every ξ ∈ V ∗ , finite set S ⊂ V and > 0
we can find w ∈ W such that ξ|S = w|S . Given such ξ ∈ V ∗ , S = {v1 , . . . , vn }
and > 0, assume S is a linearly independent set, and consider ρ : V ∗ → Rn
by ρ(η) = (hη, v1 i, . . . , hη, vn i). The map ρ|W is onto, since otherwise there exist
Pn
ci ∈ R such that ci hw, vi i = 0 for all w ∈ W (it must lie in some hyperplane,
i=1
P
n
and all hyperplanes are of this form), but this means that hw, ci vi i = 0 implying
i=1
P
n
ci vi ∈ W ⊥ = {0}. The surjectivity of ρ|W allows us to find the desired w ∈ W .
i=1
Thus it is enough to show that given η ∈ μ∞ c (X), if hf, ηi = 0 for all f ∈ Cc (X)
∞
then η = 0.
Assume X is a smooth manifold. Given a non-zero measure η, there exists some
Rn ' U ⊂ X such that η|U 6= 0, to see this either use the fact that distributions
form a sheaf, or view it as a positive function on Borel sets. Now, since U ' Rn
we must have that η|U = g ∙ μHaar where g ∈ C ∞ (Rn ). Taking some cutoff function
ψ ∈ Cc∞ (Rn ) such that ψ|B1 (0) ≡ 1 and ψ ≥ 0 implies the desired result as hgψ, ηi =
hgψ, g ∙ μHaar i = hg 2 ψ, μHaar i > 0 as this is an integral of a positive function.
For an F -analytic manifold we do the same procedure only this time ψ is the
indicator function of the open unit ball in F n .
Exercise 7.2.3. Let M, N be either smooth or F -analytic manifolds and let E and
I be complex vector bundles over M and N respectively.
(i) Show that the natural map Cc∞ (M, E) ⊗ Cc∞ (N, I) → Cc∞ (M × N, E I) is an
embedding with dense image, and is an isomorphism if M, N are F -analytic
manifolds.
(ii) Show that the natural map
Φ : Dist(M, E) ⊗ Dist(N, I) → Dist(M × N, E I)
given by
hΦ(ξ ⊗ η), F i := hη, f i, where f is given by f (y) := hξ, F |Rn ×{y} i

is an embedding with dense image.

(iii) Show that the natural map
Dist(M × N, E I) → L(Cc∞ (M, E), Dist(N, I))
is an embedding with dense image, and is an isomorphism if M, N are F -

analytic manifolds.
Definition 7.2.4. Let X be an `-space and F a sheaf over X. Define Fc (X) to

be the space of compactly supported global sections of F , that is all s ∈ F (X)
such that s|K C = 0 outside some compact K. Define Cc∞ (X, F ) := Fc (X) and
Dist(X, F ) = Cc∞ (X, F )∗ .
Theorem 7.2.5. Let i : Z ,→ X be `-spaces where Z is closed. Then:

(1) Dist(X, F )|Z ' Dist(Z, F|Z ) = i∗ (F ).
(2) We have the following short exact sequence:
0 → Dist(Z, F|Z ) → Dist(X, F ) → Dist(U, F|U ) → 0.
We now want to prove the following important theorem:
Theorem 7.2.6. Let N ⊆ M be a closed submanifold of a real manifold M , and

let E a bundle over M . Then there is a canonical filtration Fi ⊆ Dist(M, E) such
that:
(1) Every ξ ∈ Fi is supported on N .
S
∞
(2) Fi is locally exhaustive, i.e. Fi is locally DistN (M, E).
i=1
(3) Fi /Fi−1 ' Dist(N, E|N ⊗ Symi (CNM
N )).
In order to prove the theorem, we would like to define the notion of derivatives of
smooth sections f ∈ Cc∞ (M, E). Alas, the value of the derivative depends on the
chart defined on M , so it is not well defined. Fortunately, the notion of vanishing
of derivatives of certain order is well defined as the following exercise shows:
Exercise 7.2.7. Let f ∈ C ∞ (Rn ) such that f (α) (0) = 0 for every multi-index α
with |α| < k, and ϕ : Rn → Rn a diffeomorphism such that ϕ(0) = 0. Furthermore
let g ∈ C ∞ (Rn ) be a nowhere vanishing function, and set fe(x) = g(x)(f ◦ ϕ−1 (x)).
(1) Show that fe(α) (0) = 0 for every multi-index α with |α| < k.
(2) Show that:

∂k ∂k
fe (0) = f (0)g(0).
∂v1 . . . ∂vk ∂ ((dϕ)v1 ) . . . ∂ ((dϕ)vk )
(3) Find a counter example for part (1) if f (i) (0) 6= 0 for some |i| < k.
Remark 7.2.8. As a consequence of this exercise, given any f ∈ Cc∞ (M, E) whose
first k − 1 derivatives vanish we can define the k-th differential symbol of f denoted
dkx f : Tx M × . . . × Tx M → Ex by

∂k
dkx f (ξ1,i , . . . , ξk,i ) = (f ◦ ϕ−1
i ) (0),
∂ξ1,i . . . ∂ξk,i
0
where ϕi is a local chart and ξ1,i = (ϕi ◦ γ1 ) (0) are tangent vectors. If we choose
a different chart ϕj we get that

∂k ∂k
dkx f (ξ1,j , . . . , ξk,j ) = (f ◦ ϕ−1
j ) (0) = (f ◦ ϕ −1
i ◦ ϕ) (0)
∂ξ1,j . . . ∂ξk,j ∂ξ1,j . . . ∂ξk,j
where ϕ := ϕi ◦ ϕ−1 j . By the discussion above, we get that

∂k ∂k
(f ◦ ϕ−1
j ) (0) = (f ◦ ϕ −1
i ) (0),
∂ξ1,j . . . ∂ξk,j ∂(dϕ)ξ1,j . . . ∂(dϕ)ξk,j
but as
0
dx ϕ(ξ1,j ) = dx ϕ ∙ (ϕi ◦ γ1 ) (0) = (ϕ ◦ ϕi ◦ γ1 )0 (0) = ξ1,i
we have that dkx f (ξ1,j , . . . , ξk,j ) = dkx f (ξ1,i , . . . , ξk,i ) so this is well defined.
Proof of Theorem 7.2.6. Note that we can identify dkx f ∈ Symk (Tx∗ M ) ⊗ Ex . Let
N ⊆ M be a submanifold. Define:
FNi (Cc∞ (M, E)) = {f ∈ Cc∞ (M, E) : ∀x ∈ N, the first k − 1 derivatives of f vanish}.
i−1
Choose trivializations M |U ' Rn and N |U ∩N ' Rk . We showed that FW (V )/FW
i
(V ) ∼
=
Cc (W, Sym (W ) ⊗ Ex ) using the map f 7→ dx f . Hence we get that:
∞ i ⊥ k
FNi /FNi−1 ' Cc∞ (N, E|N ⊗C Symi (CNM

N )).
This gives a canonical filtration Fi ⊆ DistN (M, E) such that

∗
Fi /Fi−1 ' FNi /FNi−1 ' Cc∞ (N, E|N ⊗C Symi (CNM N )) = Dist(N, E|N ⊗C Sym (CNN )).
∗ i M
Corollary 7.2.9. We have
Gri (C −∞ (M, E)N ) = C −∞ (N, E|N ⊗ Dens(M )∗ |N ⊗ Symi (NM

N ) ⊗ Dens(N )).
Proof. We have
Gri (C −∞ (M, E)N ) = Gri (DistN (M, E ∗ ⊗ Dens(M )))
' Dist(N, E ∗ |N ⊗ Dens(M )|N ⊗ Symi (CNM

N )) =
= C −∞ (N, E|N ⊗ Dens(M )∗ |N ⊗ Symi (NM

N ) ⊗ Dens(N )).
8. Operations on generalized functions
In this section we assume X and Y are either `-spaces, analytic F -manifolds (with
or without complex bundles over them), or smooth manifolds.
Definition 8.0.1. Let ϕ : X → Y be a map. We define the pullback ϕ∗ :

C ∞ (Y ) → C ∞ (X) by ϕ∗ (f ) = f ◦ ϕ. It is easy to see that if ϕ is proper then
ϕ∗ : Cc∞ (Y ) → Cc∞ (X). By dualizing, we get an operation ϕ∗ : Dist(X) → Dist(Y )
on distributions, which we call pushforward, by
ϕ∗ (ξ)(f ) := ξ(ϕ∗ (f )) = ξ(f ◦ ϕ).
Note that if ϕ is not proper then we can define ϕ∗ : Dist(X)prop → Dist(Y ) where
Dist(X)prop := {ξ ∈ Dist(X)| ϕ|supp(ξ) is proper}. We would like to set hϕ∗ ξ, f i =
hξ, f ◦ϕi, but f ◦ϕ might not be compactly supported. Therefore we choose a cutoff
function ρ such that ρ|supp(ξ) = 1 and ρ|U C = 0 where U is a small neighborhood
of supp(ξ) and ϕ|U is proper (it is a hard task to find such a function). Hence we
can define
hϕ∗ ξ, f i := hξ, ρ ∙ (f ◦ ϕ)i.
Note that
supp(ρ ∙ (f ◦ ϕ)) ⊆ supp(ρ) ∩ ϕ−1 (supp(f )) ⊆ ϕ|−1

supp(ρ) (supp(f )).
Since ϕ|supp(ρ) is proper, and f is compactly supported, this is well defined. The
definition clearly does not depend on the choice of ρ.
Recall that for vector spaces we had that Dens(V ) ' Haar(V ) canonically. Hence
we can identify the space of smooth measures μ∞ c (X) with the space of smooth
sections of the density bundle Cc (X, Dens(X)). Note that we can define ϕ∗ :
∞
Cc∞ (X, Dens(X)) → Dist(X) by hϕ∗ (μ), f i = X f dμ.

´
Exercise 8.0.2. Let X and Y be either smooth or F -analytic manifolds and ϕ :

X → Y a map. Show that the pushforward of a compactly supported distribution is
compactly supported, that is ϕ∗ (Distc (X)) ⊆ Distc (Y ).
Proposition 8.0.3. Let X and Y be either smooth or F -analytic manifolds and

ϕ : X → Y be a submersion. Then:
(1) ϕ∗ (μ∞
c (X)) ⊆ μc (Y ).
∞
(2) ϕ∗ (f ∙ |ωX |) = g ∙ |ωY |, where |ωX | and |ωY | are non-vanishing densities on
X and Y respectively and
|ωX |
ˆ
g(y) = f ∗
ϕ−1 (y) |ϕ ωY |
|ωX |
where |ϕ∗ ωY | ⊗ |ωY | is the image of |ωX | under the natural isomorphism
Dens(X)x ' Dens(ϕ−1 (y))x ⊗ Dens(Y )ϕ(x) .
Proof.
(1) We prove the first statement in two steps. Case 1: X = F n , Y = F m
where n ≥ m and ϕ : F n → F m is the natural projection ϕ(x1 , . . . , xn ) =
x1 , . . . , xm . Recall that Haar(X) ' Haar(Y ) ⊗ Haar(X/Y ) or equivalently
that
Dens(X) ' Dens(Y ) ⊗ Dens(X/Y ).
Let φ ∈ Cc∞ (X, Dens(X))and note that φ = f ∙ dμX where f ∈ Cc∞ (X)
and μX is a normalized Haar measure (taking the value 1 on the unit ball
of X = F n ), so we can write μX = μY ⊗ μX/Y . By definition, for any
g ∈ Cc∞ (Y ) we have:
ˆ ˆ ˆ
hϕ∗ (φ), gi = hφ, g ◦ ϕi = f ∙ (g ◦ ϕ)dμX = f ∙ (g ◦ ϕ)dμY ⊗ μX/Y .
X Y X/Y
It is compactly supported, and since g ◦ ϕ(x1 , . . . , xn ) = g(x1 , . . . , xm )

depends only on Y we have
ˆ ˆ ! ˆ
hϕ∗ (φ), gi = f ∙ dμX/Y ∙ gdμY = fe ∙ gdμY
Y X/Y Y
where fe ∈ Cc∞ (Y ). Hence ϕ∗ (φ) is a smooth measure.

Case 2 - general case: Let ϕ : X → Y be a submersion. Take trivializing
S S
covers Y = Vj and X = Ui such that ϕ(Ui ) ⊆ Vj . For any i, j such
j∈J i∈I
that ϕ(Ui ) ⊆ Vj we can choose isomorphisms τi : Ui ' F n and ψj : Vj ' F m
(if X and Y are F -analytic we choose isomorphisms to some powers of OF
) such that ψj ◦ ϕ ◦ τi−1 is the natural projection F n → F m (respectively

OFn → OFm for F -analytic). Hence ψj ◦ ϕ ◦ τi−1 ∗ (μ∞ c (F )) ⊆ (μc (F )
n ∞ m
and ϕ∗ (Cc∞ (Ui , Dens(Ui ))) ⊆ Cc∞ (Vj , Dens(Vj )).

Now, let φ ∈ Cc∞ (X, Dens(X)). Using partition of unity, we can write
P
φ= fi μi where fi μi ∈ Cc∞ (Ui , Dens(Ui )). Note that this is a finite sum
i∈I
since φ is compactly supported and observe that:
!
X X X
ϕ∗ (φ) = ϕ∗ fi μi = ϕ∗ (fi μi ) = gi μ0i
i∈I i∈I i∈I
where gi ∈ Cc∞ (Vj , Dens(Vj )).

Each is a smooth compactly sup-
gi μ0i
P
ported measure, so the sum gi μi is a smooth section of the density
0
i∈I
bundle and we are done.
(2) Since ϕ is a submersion for any ϕ(x) = y ∈ Y the fiber ϕ−1 (y) is a sub-
manifold of X and the following sequence is exact:
0 → Tx ϕ−1 (y) → Tx (X) → Tϕ(x) (Y ) → 0.
Since this is an exact sequence of vector spaces it splits so Tx (X) = Tx ϕ−1 (y)⊕
Tϕ(x) (Y ) and by dualizing we get that
Tx∗ (X) = Tx∗ ϕ−1 (y) ⊕ Tϕ(x)

∗
(Y ).
This implies that Dens(X)x = Dens(ϕ−1 (y))x ⊗ (Dens(Y ))ϕ(x) .

We now reduce the problem to a small neighborhood. As before take

S S
trivializing covers Y = Vj and X = Ui such that ϕ(Ui ) ⊆ Vj , and
j∈J i∈I
choose appropriate isomorphisms τi and ψj for ϕ(Ui ) ⊆ Vj such that ψj ◦
ϕ ◦ τi−1 is the natural projection F n → F m (resp. OFn → OFm ).
We need to prove that for every h ∈ Cc∞ (Y ) we have
hϕ∗ (f |ωX |), hi = hf |ωX | , h ◦ ϕi = hg ∙ |ωY | , hi,
where g is as in the statement of the proposition. Construct a partition of

P
unity f = fi with respect to {Ui }i∈I . Then it is enough to prove the
i∈I
claim for fi |ωX | as then:
X Xˆ
ϕ∗ (f |ωX |)(h) = ϕ∗ ( fi |ωX |)(h) = gi h |ωY |
Y
P
where gi (y) = fi η since supp(fi η) ⊂ Ui . As g = gi we would
´
ϕ−1 (y)
i∈I
have that g(y) = ϕ−1 (y) f η as required.
´
In (1) we showed the case where ϕ is a projection. Using the fact that
for diffeomorphisms pushforward and pullback are inverse to one another
and using part (1) we get that:
ψj ◦ ϕ∗ (fi |ωX |) = ψj ◦ ϕ ◦ (τi−1 )∗ ((τi−1 )∗ (fi |ωX |))

= ψj ◦ ϕ ◦ (τi−1 )∗ (fi ◦ τi−1 ∙ (τi−1 )∗ ωX ) = gei (ψj−1 )∗ ωY
(τi−1 )∗ ωX
where gei (x) = fi ◦ τi−1 . Set gi := gei ◦ ψj , then
´
τi ◦ϕ−1 ◦ψj−1 (x) (ϕ◦τi−1 )∗ ωY
gei = gi ◦ ψj−1 .
Hence ϕ∗ (fi |ωX |) = gi |ωY | where,
(τi−1 )∗ ωX ωX
ˆ ˆ
gi (y) = gei (ψj (y)) = fi ◦ τi−1 −1 ∗ = fi ∗ .
τi ◦ϕ−1 (y) (ϕ ◦ τi ) ωY ϕ−1 (y) ϕ ωY

Definition 8.0.4. By the proposition, the map ϕ∗ : Cc∞ (X, Dens(X)) → Cc∞ (Y, Dens(Y ))
gives rise to a pullback ϕ∗ : C −∞ (Y ) → C −∞ (X).
Exercise 8.0.5. Let ϕ : X → Y be a submersion. We can define a pullback

ϕ∗ : C ∞ (Y ) → C ∞ (X) both by ϕ∗ (f ) = f ◦ ϕ and by the restriction of the map
ϕ∗ : C −∞ (Y ) → C −∞ (X) to the subspace C ∞ (Y ) ⊂ C −∞ (Y ). Show that these two
definitions coincide.
We can generalize the pushforward and pullback operations on functions and on

distributions to functions and distributions with values in a vector bundle:
Definition 8.0.6. Let ϕ : X → Y and let be π : E → Y a bundle.

(1) Define the pullback of the bundle (E, π) to X by
ϕ∗ (E) := {(x, e) ∈ X × E : ϕ(x) = π(e)}

with the natural projection to X.

(2) Using (1) define the the pullback of sections
ϕ∗ : C ∞ (Y, E) → C ∞ (X, ϕ∗ (E))
and by dualizing the pushforward of distributions
ϕ∗ : Dist(X, ϕ∗ (E))prop → Dist(Y, E).
(3) Define ϕ! (E) := ϕ∗ (E) ⊗ ϕ∗ (Dens(Y )∗ ) ⊗ Dens(X).
Proposition 8.0.7. Let ϕ : X → Y be a submersion. Then
ϕ∗ (Cc∞ (X, ϕ∗ (E) ⊗ Dens(X))) ⊆ C ∞ (Y, E ⊗ Dens(Y )).

In particular, this implies that ϕ∗ Cc∞ (X, ϕ! (E)) ⊆ C ∞ (Y, E).
Proof. As in the proof of the last proposition, we may reduce to the case where
ϕ : X → Y is the natural projection, and X = F n , Y = F m , and E ' F m × F k
is trivial (resp OFn ,OFn and OFm × OFk for F -analytic manifolds). As a consequence,
ϕ∗ (E) = F n × F k (resp. OFn × OFk ). Note reducing is possible since the notion of
smoothness of a distribution (that is, it is a smooth measure) is local.
Let φ = f μ ∈ Cc∞ (X, ϕ∗ (E) ⊗ Dens(X)). Then we have for any g ∈ C ∞ (Y, E),
ˆ ˆ ˆ !
hϕ∗ (φ), gi = hφ, g ◦ ϕi = f ∙ (g ◦ ϕ)μX = f ∙ μX/Y ∙ gμY
X Y X/Y
ˆ
= fe ∙ gμY = hf˜μY , gi,
Y
where f˜ = f ∙ μX/Y which is smooth, so ϕ∗ (φ) is smooth.

´
X/Y
9. Fourier transform
Definition 9.0.1. Let G be a locally compact Hausdorff abelian group. Define its
Pontryagin dual by,
G∨ = {χ : G → U1 (C) = S 1 ⊆ C| χ(g1 g2 ) = χ(g1 )χ(g2 ), χ is cts}.
The topology on G∨ is the compact open topology, i.e. a sub-basis of the topology is
comprised of sets M (K, V ) = {χ ∈ G∨ : χ(K) ⊆ V } where K ⊆ G is compact and
V ⊆ S 1 is open.
Theorem 9.0.2. Let G be a locally compact, Hausdorff abelian group, then G∨ is

a locally compact Hausdorff abelian group.
Proof. We see that characters form an abelian group. Since S 1 is a topological

group, the compact open topology on G∨ is equivalent to the topology of uniform
convergence on compact sets. Thus, in order to show that the multiplication and
inverse operations are continuous, it is enough to show that if fn → f and gn → g
uniformly on compact sets then fn ∙ gn−1 → f ∙ g −1 uniformly on compact sets. Now,

if K ⊂ G is compact, note that this follows from the following bound (∀x ∈ K):
|fn gn−1 − f g −1 | ≤ |fn (gn−1 − g −1 )| + |(fn − f )g −1 | = |gn − g| + |fn − f |.
Now to show it is locally compact, consider the space (S 1 )G of all functions f :

G → S 1 ' R/Z with the product topology (i.e. a basis is given by open sets in only
finitely many components). It is a compact space by Tychonoff’s theorem, and it
has the space
\
G̃ = {χ : G → S 1 : χ(g1 g2 ) = χ(g1 )(g2 )},
g1 ,g2 ∈G
as a closed subspace, implying that G̃ is compact. Furthermore, for every S ⊆ G

and > 0 the set A(S, ) = {χ ∈ (S 1 )G : χ(S) ⊆ [−, ]} is also closed and compact
in (S 1 )G as the complement is a union of sets of the form
{χ : G → S 1 : ∃s ∈ S s.t. χ(s) ∈ [−, ]c }
which are open.

In particular, taking an open neighborhood e ∈ U ⊂ G, the sets V (U, ) = A(U, ) ∩
G̃ are closed and compact in (S 1 )G . Take 0 < < 12 , we show that we have that
V (U, ) ⊆ G∨ . Start with an open e ∈ U0 = U ⊂ G, and choose a sequence of
neighborhoods (Un )∞ n=1 in G such that Un+1 ∙ Un+1 ⊂ Un for all n ∈ N and set
n = 2n . Taking χ ∈ V (Un , n ), we see that since for x ∈ Un+1 we have that

χ(x) ∈ [−n , n ] and x2 ∈ Un we get χ(x2 ) = χ(x)2 ∈ [−2n , 2n ] ⊆ [− 2n , 2n ],
implying that V (Un , n ) ⊆ V (Un+1 , n+1 ).
Now, take χ ∈ V (U, ) and a basic open set (−δ, δ) ⊂ S 1 for δ > 0. We have
that [−n , n ] ⊆ (−δ, δ) for n big enough, implying that e ∈ Un ⊆ χ−1 ((−δ, δ))
which means that χ is continuous at e. Since χ is a homomorphism, we can show
it is continuous everywhere; if χ(g) ∈ W ⊂ S 1 and W is open, we have that
(−δ, δ) ⊆ χ(g −1 )W for some δ > 0 and that,
χ−1 (χ(g −1 )W ) = {y ∈ G : χ(y) ∈ χ−1 (g)W } = {y ∈ G : χ(gy) ∈ W }

= g −1 {gy ∈ G : χ(gy) ∈ W } = g −1 χ−1 (W ).
Now, for some m ∈ N0 big enough, the following implies that g ∈ gUm ⊆ χ−1 (W ):
Um ⊂ χ−1 (χ(g −1 )W ) = g −1 χ−1 (W ).
We know that V (U, ) is compact in the product topology, and want to show it is
compact with respect to the compact open topology. For this, it is enough to show
that any net in V (U, ) has a converging subnet in the compact open topology.
Assume we are given some net (xα ) ∈ V (U, ), then it has a subnet (fβ ) → f
converging in the product topology with fβ , f ∈ V (U, ). Now, note that V (U, )
is uniformly equicontinuous, that is if g1 , g2 ∈ G and g1 g2−1 ∈ Un then for any
χ ∈ V (U, ),
|χ(g1 ) − χ(g2 )| = |χ(g1 )χ−1 (g2 ) − 1| = |χ(g1 g2−1 ) − 1| ≤ n .
Given a basic open neighborhood of the identity character 1G ∈ M (K, B0 (0)),
where K is compact, for every g ∈ K we have that g ∈ Un g (for n big enough).
Now, taking any g 0 ∈ Un g, we get that g 0 g −1 ∈ Un implying that for some big
enough β we have that |f (g) − fβ (g)| < n and that,
|f (g 0 ) − fβ (g 0 )| ≤ |f (g 0 ) − f (g)| + |fβ (g 0 ) − fβ (g)| + |f (g) − fβ (g)| < 3n .

0
Taking n > ng such that n < 3 , we see that fβ → f uniformly on Un g, but since
K is compact we can cover it with finitely many sets of the form Ung g, and take
n = max {ngi } and appropriate β.
0≤i≤k
To finish off the argument, note that by local compactness every g ∈ G has a
neighborhood g ∈ U with compact closure U ⊂ K, and we have that M (K, B (0)) ⊆
V (U, ) for an appropriate 12 > > 0.
Exercise 9.0.3. Let G be a locally compact, Hausdorff abelian group. Show that if
G is compact then G∨ is discrete, and that if G is discrete then G∨ is compact.
Theorem 9.0.4. For a locally compact abelian group G, we have that the natural
map ϕ : G → G∨∨ defined by g 7→ ϕg , where ϕg (χ) = χ(g), is an isomorphism
G∨∨ ' G.
Proof. This is complicated. Need a reference.
Exercise 9.0.5. Let G be a locally compact, Hausdorff abelian group, and H ≤ G

a closed subgroup. Show that:
(1) Pontryagin duality is a contravariant endofunctor in the category of locally
compact abelian groups.
(2) H ∨ ' G∨ /H ⊥ where H ⊥ = {χ ∈ G∨ : χ(h) = 1 ∀h ∈ H}, and that if
H and G are vector spaces then this is a homeomorphism (Hint: use an
appropriate version of the Hahn-Banach theorem).
Example 9.0.6.
(1) For any finite abelian group G we have that G ' G∨ .
(2) The dual of U1 (C) = S 1 is Z.
(3) We have that R∨ ' R.
Exercise 9.0.7. Let V be a topological vector space over a local field F . Then
V ∗ ⊗F F ∨ ' V ∨ .
Definition 9.0.8. Let G be a locally compact Hausdorff abelian group. The map
F : μc (G) → C(G∨ ) defined by F (μ)(χ) = χdμ is called the Fourier transform.
´

(1) F is continuous.
(2) Let G be a locally compact abelian group. For a character τ : G → S 1 define
shh (τ )(x) = τ (x + h). Show that for η ∈ μ∞c (G) and g ∈ G:
(a) F (shg (η))(χ) = χ(g)F (η)(χ) for all χ ∈ G∨ .
(b) F (χη) = shχ−1 (F (η)) for all χ ∈ G∨ .
Definition 9.0.10. Let X1 and X2 be locally compact topological vector spaces and
c (X1 ) and μ2 ∈ μc (X2 ). We define the external tensor product of such
let μ1 ∈ μ∞ ∞
measures μ1 μ2 ∈ μc (X1 × X2 ). In addition, If X1 = X2 = G, then we define

∞
the convolution of these measures by μ1 ∗ μ2 := m∗ (μ1 μ2 ) where m : G × G → G

is the multiplication map.
c (G) we have that F (α ∗β) = F (α)∙F (β).

Fact 9.0.11. For two measures α, β ∈ μ∞
Definition 9.0.12. Let V be a finite dimensional vector space over a local field F .
Define the space of Schwartz functions S(V ) on V by:
(1) If F is non-archimedean, then S(V ) = Cc∞ (V ), i.e. locally constant func-
tions on V .
(2) If F is archimedean, then
S(V ) = {f ∈ C ∞ (V )|∀i ∈ Nn , p ∈ F [V ], sup ∂ i f ∙ p(x) < ∞}.
In other words it is the space of rapidly decreasing smooth functions on V .
Proposition 9.0.13. The Fourier transform F : S(V, Haar(V )) → S(V ∨ ) is con-

tinuous for an Archimedean V and its image is indeed contained in S(V ∨ ) in both
cases.
Proof. Assume V is a real vector space of dimension n, and recall that the topology
β
on S(V ) is determined by the semi-norms kf kα,β = sup |Φα (x) ∂ ∂x
f (x)
β | where α, β ∈
x∈V
Q
n
α
Nn0 and Φα (x) = xj j . It is enough to show that for every f ∈ Cc∞ (V, Haar(V ))
j=1
and semi-norm k ∙ kα,β on S(V ∨ ) there exists a semi-norm k ∙ k0 on S(V, Haar(V ))
and a positive constant C such that kF (f )kα,β ≤ Ckf k0 . Now, recall that,
i∂F (f ) i∂ −iξ∙x
ˆ
= (e f (x))dx = F (xj f ),
∂ξj ∂ξj
Rn
where one can differentiate directly using the definition to verify the above proce-
dure. The other side of the coin is given by integration by parts,
−iξ∙x ∞ ξj −iξ∙x ∂f (x) −i∂(f )
ˆ ˆ
ξj F (f ) = ξj e −iξ∙x
f (x)dx = −e f (x) −∞ − e dx = F ( ).
−iξj ∂xj ∂xj
Rn Rn
Note that since the functions e converge weakly to zero as distributions as

−iξ∙x
|ξ| → ∞ we get that Schwartz measures are mapped into S(V ∨ ). We can now
bound F (f ) properly using the above relations:
(−i∂)β F (f )(x∨ ) (−i∂)α (Φβ (−x)f )

ˆ
kF (f )kα,β = sup Φα (x∨ ) = sup x∨ μ
x∨ ∈V ∨ ∂(x )
∨ β
x∨ ∈V ∨ ∂xα
V

∂ (Φβ (−x)f )
α
n+1 ∂ (Φβ (|x|)f )
α
ˆ
≤ sup x∨ μ ≤ C sup (1 + |x|) .
x∨ ∈V ∨ ∂xα x∈V ∂xα
V
where C = 1
Since the last expression is a linear combination of
´
(1+|x|)n+1 dμ(x).
V
norms of the form kf kα0 ,β 0 for |α0 | ≤ |α| + n + 1 and |β 0 | ≤ |β|, this implies that F
is continuous. Note that we can also use this to show that F (f ) is Schwartz, since
β
if all the norms k ∙ kα,β are bounded then the value of |Φα (x) ∂ ∂x f (x)
β | decays to 0 as
|x| → ∞ for every α and β.

The proof for vector spaces over non-Archimedean fields is analogous.
Definition 9.0.14. Let S(V ) be the space of Schwartz functions on V .

(1) We call ξ ∈ S ∗ (V ) the space of tempered distributions and G(V ) :=
S ∗ (V, Haar(V )) the space of tempered generalized functions.
(2) Finally, we define the Fourier transform on tempered distributions via du-
ality:
F ∗ : S ∗ (V ∨ ) → G(V ) := S ∗ (V, Haar(V )).
Taking V := V ∨ we get F ∗ : S ∗ (V ) → G(V ∨ ).
Theorem 9.0.15. The definition of Fourier transform of distributions is consistent

with the definition given for functions. In other words F ∗ |S(V,Haar(V )) = F .
Proof. Let f (x) ∙ dx ∈ S(V, Haar(V )) and g(χ) ∙ dχ ∈ S(V ∨ , Haar(V ∨ )). Then by
definition,
ˆ
hF ∗ (f (x) ∙ dx), g(χ) ∙ dχi := hf (x) ∙ dx, F (g(χ)dχ)i = f (x)F (g(χ) ∙ dχ)(x)dx
V
where F (g(χ) ∙ dχ)(x) := V ∨ χ(x)g(χ)dχ. Therefore we have:

´
ˆ ˆ ˆ ˆ ˆ
f (x)F (g(χ) ∙ dχ)(x)dx = f (x) χ(x)g(χ)dχdx = χ(x)f (x)dx g(χ)dχ
V V∨ V∨
ˆV V
= (F (f )(χ)) g(χ)dχ = hF (f (x) ∙ dx), g(χ) ∙ dχi.

V∨

Remark 9.0.16. We will usually omit the ∗ from the F ∗ notations, this should
cause no confusion.
In the following argument we would like to show the Fourier transform is a unitary
operator. For this we will first need to define a pairing between Haar(V ) and
Haar(V ∨ ). Given α ∈ Haar(V ) and β ∈ Haar(V ∨ ) we can define such a pairing
as follows. We choose f ∈ Cc∞ (V ∨ ) such that f (0) = 1 and then define hα, βi :=
hF (α), f ∙ βi.
Exercise 9.0.17.
(1) Show this is well defined. That is, given a different g ∈ Cc∞ (V ∨ ) such that
g(0) = 1, show that hF (α), (f − g) ∙ βi = 0.
(2) Show that Haar(V ∨ ) 'can Haar(V )∗ .
Exercise 9.0.18. Show that F (δa )(t) = exp(at).
Definition 9.0.19. We define a map Fn : S ∗ (V, Haar(V )⊗n ) → S ∗ (V ∨ , Haar(V ∨ )⊗(1−n) )

such that F0 is the Fourier transform. We use the following isomorphisms:
(1) The pairing Haar(V ∨ ) 'can Haar(V )∗ .
(2) The identification S ∗ (V, Haar(V )⊗n ) ' S ∗ (V ) ⊗ Haar(V )⊗−n .
(3) The identification S ∗ (V ∨ , Haar(V ∨ )⊗(1−n) ) ' S ∗ (V ∨ , Haar(V ∨ ))⊗Haar(V ∨ )⊗n .
The first item was shown in the previous exercise. The second identification is as
follows. Given ξ ⊗ β ∈ S ∗ (V ) ⊗ Haar(V ∨ )⊗n and f ∙ α ∈ S(V, Haar(V )⊗n ) we have
that hξ ⊗ β, f αi = hξ, f ihβ, αi. The third identification is similar.
Finally, we define the map by applying the Fourier transform on the first coordinate
of the right hand side of (2) and by applying the canonical map (1) on the second
coordinate.
Proposition 9.0.20. We have that F1 ◦F0 = flip where hflip(ξ), f (x)μi = hξ, f (−x)μi.
Proof. Note that span{δx }x∈V is a dense subspace of S ∗ (V ) with respect to the
weak topology. Hence it is enough to show that F1 ◦ F0 (δa ) = δ−a for all a ∈ V .
Note that hF0 (δ0 ), f βi := hδ0 , F0 (f β)i = V ∨ f dβ, this implies F0 (δ0 ) = 1.
´
As before, F1 : S ∗ (V ∨ , Haar(V ∨ )) → S ∗ (V ) is defined by identifying : S ∗ (V ∨ , Haar(V ∨ ))

with S ∗ (V ∨ ) ⊗ Haar(V ). Under this identification, 1 ∙ μ for a choice of a Haar mea-
sure μ on V ∨ is identified with 1 ⊗ η where η ∈ Haar(V ) and hμ, ηi = 1.
Given f ∈ S(V ), we have that (note we are using Theorem 9.0.15):
hF1 (1 ∙ μ), f i = hF ∗ (1) ⊗ η, f ∙ η ⊗ μi = hF (1), f ∙ ηihη, μi = hδ0 , f ∙ ηi ∙ 1 = f (0),
so F1 ◦ F0 (δ0 ) = δ0 .
Using Exercise 9.0.9, we now see that (here χ(a) is the function which substitutes
the value a in a given character χ):
F1 ◦ F0 (δa ) = F1 ◦ F0 (sha (δ0 )) = F1 (χ(a)F0 (δ0 )) = sh−a F1 ◦ F0 (δ0 ) = δ−a .
By continuity of F0 and F1 this implies that F1 ◦ F0 = flip.
Definition 9.0.21. Let F and K be local fields and χ : F × → K × a character.

For a 1-dimensional space V over F we define a functor by:
χ(V ) := {ϕ : V ∗ → K : ϕ(αv) = χ(α)ϕ(v)∀α ∈ F × , v ∈ V ∗ }.

Example 9.0.22. Let χ : F × → F × be the character x 7→ x2 and let V be a one

dimensional vector space of F . Then
χ(V ) := {ϕ : V ∗ → K : ϕ(αf ) = α2 ϕ(f )}.
Note that χ(V ) 'can V ⊗ V by v ⊗ w 7→ ϕv ∙ ϕw . Indeed, given ψ ∈ V ∗ , we have

ϕv ∙ ϕw (ψ) = ψ(v) ∙ ψ(w) and ϕv ∙ ϕw (aψ) = aψ(v) ∙ aψ(w) = a2 ϕv ∙ ϕw (ψ).
Definition 9.0.23. Let V be a 1-dimensional vector space over R.

(1) A positive structure on V is a non trivial subset P ⊆ V such that R≥0 ∙ P =
P.
(2) If V has a positive structure, we define
α α
V α := |V | = {ϕ : V ∗ → R : ϕ(βf ) = |β| ∙ ϕ(f )}.
Exercise 9.0.24. Let V be a real 1-dimensional vector space with a positive struc-
ture.
(1) Show that:
(a) V 'can |V |.
(b) V α+β 'can V α ⊗ V β where α, β ∈ Q× .
(2) Deduce that Haar(V )α ⊗ Haar(V )β ' Haar(V )α+β .
Definition 9.0.25. For α ∈ Q we define similarly to the procedure defined above,
Fα : S ∗ (V, Haar(V )) → S ∗ (V ∨ , Haar(V ∨ )1−α ).
In particular, choosing α = 1
2 we have:
1 1
F 12 : S ∗ (V, Haar(V ) 2 ) → S ∗ (V ∨ , Haar(V ∨ ) 2 ).
Theorem 9.0.26 (Functoriality of Fourier transform). Let W ⊂ V be vector spaces

over a local field, denote the inclusion of W in V by i, and set p : V ∨ → W ∨ for
the induced linear map on the duals, then the following diagrams commute:
i∗ i∗
S(V ) S(W ) S ∗ (V ) S ∗ (W )
F F F F
∗
p∗ p
S(V ∨ , Haar(V ∨ )) S(W ∨ , Haar(W ∨ )) G(V ∨ ) G(W ∨ ).
Note that this is possible since p is a submersion (linear and surjective) so pushing
Schwartz measures along it yields Schwartz measures.
Proof. We start by showing the right hand side diagram commutes. Since i∗ , the
Fourier transform and p∗ are continuous with respect to the weak topology, it is
enough to prove commutativity for a dense set in S ∗ (W ).
First take the delta function δ0 ∈ S ∗ (W ), it is a compactly supported measure, and
it holds that i∗ (δ0 ) = δ0 . Furthermore, since F : S ∗ (V ) → G(V ∨ ) is defined via
duality we have that F (δ0 ) = 1:

ˆ
hF (δ0 ), f μi = hδ0 , F (f μ)i = F (f μ)(0V ∨∨ ) = f dμ = h1, f μi,
V ∨
where the third equality is sensible since F (f μ) ∈ S(V ∨∨ ) and 0V ∨∨ (χ) = 1 for
all χ ∈ V ∨ . We can also show that p∗ (1) = 1. Consider G(W ∨ ) as a subspace of
C −∞ (W ∨ ), there the generalized Schwartz function 1 is a smooth function, and
note that the following diagram, where the horizontal arrows are the inclusions is
commutative:
G(V ∨ ) C −∞ (V ∨ ) C ∞ (V ∨ )
p∗ p∗ p∗
G(W ∨ ) C −∞ (W ∨ ) C ∞ (W ∨ ).
Now, note that every measure f μ ∈ μ∞ c (V ) can be treated either as a functional

∨
on smooth functions (since it has compact support as a distribution), or as the

parameter a generalized function takes values on. This is utilized in the second
equality bellow to yield the required result:
hp∗C −∞ (1), f μi = h1, p∗ (f μ)i = hp∗ (f μ), 1i = hf μ, p∗C ∞ (1)i = hf μ, 1i = h1, f μi.
Note that since p∗ is a submersion pushing forward a compactly supported smooth

measure along it yields a smooth measure.
Since δw for any w ∈ W is just a translation of δ0 by w, its Fourier transform is
F (δw )(χ) = χ(w), and i∗ and p∗ are invariant to translations, the diagram is com-
mutative for delta distributions. The space of delta distributions spanC {δw }w∈W
is dense w.r.t the weak topology since for every function f with f (x0 ) 6= 0 we can
take suitable c ∈ C such that |hξ − cδx , f i| is small as desired.
To see this implies the commutativity of the left diagram, it is enough to show
that if A∗ = 0 for A∗ : V2∗ → V1∗ where A∗ is the dual map to the linear map
A : V1 → V2 , then A = 0, and use this for F i∗ − p∗ F .
If A∗ = 0, we have for every ξ2 ∈ V2∗ and v1 ∈ V1 that 0 = hA∗ ξ2 , v1 i = hξ2 , Av1 i. If
there exists v1 ∈ V1 such that Av1 6= 0, then we can define a non-zero linear func-
tional ξ : spanC {Av1 } → C via hξ, Av1 i = 1, and extend it to a non-zero continuous
functional ξ2 ∈ V2∗ by the Hahn-Banach theorem. This yields a contradiction as
1 = hξ2 , Av1 i = hA∗ ξ2 , v1 i = h0, v1 i = 0.

10. Wave-front set
The wave-front set of a generalized function ξ is the collection of all points and
codirections in which ξ is not smooth. It is a very important invariant of the
generalized function. For example, there are some operations on functions, like
product or pull-back, that do not extend to arbitrary generalized functions but
do extend to generalized functions under some conditions on the wave-front set.
The term comes from physics. Every differential equation satisfied by a generalized
function will give a restriction on its wave-front set.
Example 10.0.1. Let ξ be the δ-function of the x-axis on R2 , i.e. the generalized
function given by hξ, f dxdyi := f (x)dxi, let L ⊂ R2 denote the x-axis and L⊥ ⊂
´
(R2 )∗ denote the subspace of functionals vanishing on L. Then WF(ξ) = L × L⊥ .
We will now define the wave-front set by characterizing properties.
Definition 10.0.2. The wave-front set is an assignment of a closed subset WF(ξ) ⊂

T ∗ M for any ξ ∈ C −∞ (M, E) such that
(1) For any isomorphism ν : (M, E) ' (M 0 , E 0 ),
ν̃(WF(ξ)) = WF(ν∗ (ξ)).
(2) WF(ξ) is conical in the cotangent directions, i.e.
(x, v ∈ WF(ξ) ⇒ (x, λv) ∈ WF(ξ) ∀λ ∈ F.
(3) pM (WF(ξ)) = WF(ξ) ∩ (M × {0}) = supp(ξ),

where pM : T ∗ M → M denotes the natural projection.
(4) ξ ∈ C ∞ (M, E) ⇔ WF(ξ) ⊂ M × {0}
(5) WF(f ξ + gη) ⊂ WF(ξ) ∪ WF(η)
(6) For another bundle E 0 over M , let η ∈ C −∞ (M, E) and consider ξ ⊕ η ∈
C −∞ (M, E ⊕ E 0 ). Then WF(ξ ⊕ η) = WF(ξ) ∪ WF(η).
(7) For any open subset U ⊂ M , WF(ξ|U ) = WF(ξ) ∩ p−1 M (U ).
(8) Let ν : N → M be a submersion. Then WF(ν ∗ ξ) = ν ∗ (WF(ξ)), where the
operation ν ∗ on subsets of cotangent bundles will be defined below.
(9) Let ν : M → N be a smooth map such that ν|supp(ξ) is proper. Then
WF(ν∗ (ξ)) ⊂ ν∗ (WF(ξ)), where the operation ν∗ on subsets of cotangent
bundles will be defined below.
Definition 10.0.3. (i) For sets A, B, and subsets X ⊂ A, Y ⊂ B, S ⊂ A × B

define subsets S∗ (X) ⊂ B and S ∗ (Y ) ⊂ A by
S∗ (X) = {y ∈ B | ∃x ∈ X s.t.(x, y) ∈ S} S ∗ (Y ) = {x ∈ A | ∃y ∈ Y s.t.(x, y) ∈ S}.
(ii) For a morphism smooth map ν : M → N , define Δν ⊂ T ∗ M × T ∗ N by
Δν := {((m, v), (n, w) ∈ T ∗ M × T ∗ N | ν(m) = n, d∗m ν(w) = v}.

For a subset X ⊂ T ∗ M define ν∗ (X) := (Δν )∗ (X) ⊂ T ∗ N and

for a subset Y ⊂ T ∗ N define ν ∗ (Y ) := (Δν )∗ (Y ) ⊂ T ∗ M.
Exercise 10.0.4 (*). Check whether the wave-front set is uniquely defined by the
above properties.
Hint. By (7), WF is local and thus it is enough to prove for ξ ∈ C −∞ (F n ). Let

(0, v) ∈/ WF(ξ), and let ρ ∈ Cc∞ (F n ) be constant 1 in a neighborhood of zero.
Consider v∗ (ρξ) ∈ C −∞ (F ). From (9) and (4), v∗ (ρξ) ∈ C ∞ (F ). Thus, if v∗ (ρξ) ∈ /
C (F ) for some ρ as above then (0, v) ∈ WF(ξ).
∞
Let us now find a necessary condition for (0, v) ∈ WF(ξ). We will use the Radon
transform, which maps f ∈ Cc∞ (V ) to the integrals of f on all affine hyperplanes,
i.e. hyperplanes not necessarily passing through the origin. For a vector space
W , denote W ˉ := P(W ⊕ F ). Note that this is a compact manifold and that the
manifold of all affine hyperplanes in V is V ∗ , and the manifold of all affine lines in
V is Vˉ . Define R := {(l, H) ∈ Vˉ × V ∗ | l ∈ H}, and let p1 : R → Vˉ and p2 : R → V ∗
be the projections. Then the Radon transform is (p2 )∗ ◦ p∗1 . It is known that this
transform is invertible, and thus any distribution is the Radon transform of another
one. Thus, the conditions (9) and (8) give an upper bound on WF. Hopefully, the
two bounds together determine WF.
10.1. Definition of the wave-front set. Let us now give a constructive definition
of WF, following Hörmander, and then prove the above properties. It will take
several steps.
Definition 10.1.1. For any local field F , let V be an F -vector space, v ∈ V and
f ∈ C ∞ (V ). We say that f vanishes asymptotically along v if there exists an
open neighborhood U of v and ρ ∈ Cc∞ (U ) such that p∗ (ρ)m∗ f ∈ S(U × F ), where
m : V × F → V is given by m(v, λ) := λv and p : V × F → F is the projection.
Exercise 10.1.2. TFAE:

(i) f vanishes asymptotically along v for any v 6= 0 ∈ V
(ii) f ∈ S(V ).
Exercise 10.1.3. Show that f vanishes asymptotically along 0 if and only if f = 0.
Exercise 10.1.4. For any Lie group G, we have F (Distc (G)) ⊂ C ∞ (Ǧ).
Definition 10.1.5. Let V be a vector space, and ξ ∈ Dist(V ).

(i) We say that ξ is smooth at (x, w) ∈ V × V ∗ if there exists ρ ∈ Cc∞ (V ) such
that ρ(x) = 1 and F (ρξ) vanishes asymptotically along w.
(ii) WF(ξ) = {(x, w) ∈ V × V ∗ | ξ is not smooth at (x, w)}.
(iii) For x ∈ V , let WFv (ξ) := WF(ξ) ∩ {x} × V ∗
Theorem 10.1.6 (The proof is complicated). Let ν : V → V be a diffeomorphism

s.t. ν(0) = 0 and d0 ν = Id. Then WF0 (ν ∗ (ξ)) = WF0 (ξ) for any ξ ∈ Dist(V ).
Exercise 10.1.7 (*). A distribution ξ on V is smooth at (x, w) ∈ V × V ∗ if and

only if for any ρ ∈ Cc∞ (V ) with ρ(x) = 1, the Fourier transform F (ρξ) vanishes
asymptotically along w.
Corollary 10.1.8. Let ν : V → V be a diffeomorphism. Then
WF(ν ∗ (ξ)) = ν ∗ (WF((ξ))).

Corollary 10.1.9. The definition of WF extends to generalized sections of vector
bundles on manifolds.
Exercise 10.1.10. Let L ⊂ V be linear spaces of dimensions 1 and 2. Fix a Haar
measure μ on L and define δL ∈ Dist(V ) by hδL , f i := L f μ. Compute WF(δL ).
´
Exercise 10.1.11. Properties (1)-(7) of WF hold.

Exercise 10.1.12. For any epimorphism of Lie groups p : G H, and any
ξ ∈ Distc (G), we have F (p∗ (ξ)) = F (ξ)|Ȟ .
Exercise 10.1.13. Property (8) holds.
Exercise 10.1.14. Property (9) holds.
Hint. Any map ν : M → N decomposes as a composition of the closed embedding
graph(ν) : M ,→ M × N and the projection M × N N . Thus it is enough
to prove for a closed embedding and for a projection. By the locality of WF and
invariance to isomorphisms it is enough to prove for a linear embedding and a linear
epimorphism.
10.2. Advanced properties of the wave-front set.

Definition 10.2.1. Let Γ ⊂ T ∗ M be a closed subset. Define
CΓ−∞ (M ) := {ξ ∈ C −∞ (M ) | WF(ξ) ⊂ Γ}.

Definition 10.2.2. (i) For a non-archimedean F , an F -analytic manifold M
and an F -vector space W define
S W (M ) := {f ∈ C ∞ (M × W ) | prM |supp(f ) is proper }.
(ii) For a smooth manifold M and an R-vector space W define
S W (M ) := {f ∈ C ∞ (M × W ) | ∀ compact K ⊂ M, ∀m, n ∈ Ndim W

∀D ∈ Dif f (M ). sup (n) m
||Dfx,w w || < ∞}.
(x,w)∈K×W
w
Definition 10.2.3. Define topology on CΓ−∞ (V ) by: ξi → ξ if ξ → ξ in C −∞ (V )
and for any v ∈ V there exists > 0 and ρ ∈ Cc∞ (B (v)) such that for any l ∈ V ∗
we have
m∗ F (ρξi )|B (l)×F → m∗ F (ρξi )|B (l)×F
in S F (B (l) × F ).
Definition 10.2.4. For a smooth map ν : M → N , define
Sν := {(ν(x), w) ∈ T ∗ N | dx ν ∗ (w) = 0}.
Note that Sν = ν∗ (M × {0}).
Theorem 10.2.5. Let ν : M → N and let Γ ⊂ T ∗ N be a closed subset such that
Γ ∩ Sν ⊂ N × {0}.
Then ν ∗ : C ∞ (N ) → C ∞ (M ) can be continuously extended to
ν ∗ : CΓ−∞ (N ) → Cν−∞
∗ (Γ) (M ).
Proof. For a submersion, one can pullback any distribution, CΓ−∞ (N ) = C ∞ (N ).

Also, the statement is local. Thus, enough to prove it for ν : V → W . Any such ν
decomposes to i : V → V × W given by i(v) := (v, 0), ν 0 : V × W → V × W given
by (v, w) 7→ (v, ν(v) + w) and the projection on W . Thus, it is enough to prove for
i. For i we have
Si = CN (V ) = V × W ∗ ⊂ (V ⊕ W ) × (V ∗ ⊕ W ∗ ) = T ∗ (V ⊕ W ).
Also, enough to prove for the case dim W = 1. This we do by twisting pushforward
by F .
More precisely, we use Theorem 9.0.26 on functoriality of Fourier transform. By this
theorem, for f ∈ S(V ⊗W ) we have F (i∗ f ) = p∗ (F (f )), where p : V ∗ ⊕W ∗ → V ∗ is
the projection. This formula extends to ξ ∈ CV−∞ ×W ∗ (V ⊕ W ), since F (ρξ) vanishes
asymptotically along W for any ρ ∈ Cc (V ⊕ W ). More precisely, we can choose a
∗ ∞
sequence ρn ∈ Cc∞ (V ⊕ W ) that on every compact becomes the constant 1 starting

from some index, and define ν ∗ ξ to be the limit of F −1 (p∗ (F (ρn ξ))).
Corollary 10.2.6. Let ξ, η ∈ C −∞ (M ) such that WF(ξ) ∩ WF(η) ⊂ M × {0}.

−∞
Then we can define the product ξ ∙ η ∈ CWF(ξ)+WF(η) (N ).
Proof. Define Δ : M → M × M by Δ(m) := (m, m). Then ξ ∙ η = Δ∗ (ξ ⊗ η).
Theorem 10.2.7. Let ξ ∈ C −∞ (V ) and let Z ⊂ V ∗ be a closed conical set such

that suppF (ξ) ⊂ Z. Then WF(ξ) ⊂ V × Z.
Intuitively, this theorem makes sense since the cotangent directions in the wave-
front set form the asymptotic support of the Fourier transform. Let us now give the
proof in the p-adic case, since the proof in the real case is similar, though longer.
Proof. Enough to show that for any ρ ∈ Cc∞ (V ), F (ρξ) is asymptotically supported
in Z. Note that F (ρξ) = F (ρ) ∗ F (ξ) and thus suppF (ρξ) ⊂ suppF (ρ) + Z. Since
suppF (ρ) is a compact set, F (ρξ) is eventually zero in any direction not in Z.
Since Z is closed, this implies that F (ρξ) is asymptotically supported in Z and
thus WF(ξ) ⊂ V × Z.
Theorem 10.2.8. Let M be a smooth manifold, D be a differential operator on

M , and ξ ∈ C −∞ (M ) such that Dξ = 0. Then Symb(D)(WF(xi)) = {0}.
Sketch of proof. Since the question is local, we can assume that M is a vector space
V . Fix x ∈ V and let p be the polynomial on V ∗ given by p(l) := Symb(D)(v, l).
Let l ∈ V ∗ be a cotangent direction such that p(l) 6= 0. We want to show that
p∈/ WFx (ξ). Define
ln |F (ξ)(λw)|
ν(ξ, l) := sup lim
w∈l+B (0) λ→∞ ln |λ|
This is the order of asymptotics as λ → ∞ of F (ξ) near the direction l. It is not
+∞ since F (ξ) is a tempered distribution. We want to show that ν(ξ, l) = −∞ for
small enough. Denote ψ(α) := exp(2πiα), and define fλ ∈ C ∞ (V ) by fλ (v) :=
ψ(λl(v)). We have
0 = hρDξ, fλ i = hξ, D(ρfλ )i.
Using the Leibnitz rule we get
degD
X degD
X
D(ρfλ )(v) = D(ρ(v)ψ(λl(v)) = λm ψ(λl(v))ρm = λm f ρ m
m=0 m=0
for some collection ρm ∈ Cc∞ (V

). Thus
X X X
0 = hξ, D(ρfλ )i = λm hξ, f ρm i = λm F (ρm ξ)(λl) = λm (F (ξ)∗F (ρm ))(λl).
Thus
F (ξ) ∗ F (ρdegD )(λl) = −λ−1 (F (ξ) ∗ F (ρdegD−1 ))(λl) + . . .
and thus ν(ξ, l) = ν(ξ, l) − 1. Thus ν(ξ, l) = −∞.
10.3. Sketch of proof of the invariance to isomorphisms in the p-adic case.

Let V := F n . Let ψ : F → U (1) ⊂ C × be a non-trivial unitary additive character
with ψ(B1 (0)) = {1}. Identify V̌ with V ∗ using ψ.
Notation 10.3.1. Let l ∈ V ∗ and r, ε > 0 ∈ R. Denote

−∞
Cl,r,ε (V ) := {ξ ∈ C −∞ (V ) | supp(ξ) ⊂ Br (0) and m∗ (F (ξ))|Bε (l)×F ∈ S F (Bε (l)×F )}
−∞
Cl,r,ε,α (V ) := {ξ ∈ C −∞ (V ) | supp(ξ) ⊂ Br (0) and supp(m∗ (F (ξ)))|Bε (l)×F ⊂ Bε ×Bα (F )}
−∞ −∞
Note that Cl,ε (V ) := ∪α∈R Cl,ε,α (V ).
−∞ −∞
Exercise 10.3.2. (i) If ξ ∈ Cl,r,ε,α (V ) with ||l|| = 1 then 1Bδ (l) ξ ∈ Cl,r,ε,α+δ −1 (V ).
−∞ −∞
(ii) If ξ ∈ Cl,r,ε (V ) and ρ ∈ Cc (V ) then ρξ ∈ Cl,r,ε (V ).
∞
Notation 10.3.3. For a diffeomorphism ν : V → V and x ∈ V denote by Af fx ν

the affine approximation to ν at x. Namely
Af fx ν(y) := ν(x) + dx ν(y − x).

Exercise 10.3.4. Let ν : V → V be a diffeomorphism s.t. ν(0) = 0 and d0 ν = Id.

Let r > 0. Show that ∃C ∀δ > 0 ∀x ∈ Br (0) we have
sup ||ν − Af fx ν|| < Cδ 2 .

Bδ (x)
Hint. Use Taylor series
Exercise 10.3.5. Let ξ ∈ Sc∗ (V ). Let ν1 , ν2 : V → V and let ε > 0 be s.t.
sup ||ν1 − ν2 || < ε.

suppξ
Then
F (ν1∗ ξ)|Bε−1 (0) = F (ν2∗ ξ)|Bε−1 (0) .
Exercise 10.3.6. Let r, ε, α > 0 with ε < 1 and let l ∈ V ∗ s.t. ||l|| = 1. Let
A : V → V be an affine transformation s.t. ||d0 A − Id|| < 1 and A(Br (0)) ⊂ Br (0).
Then
−∞
A∗ (Cl,r,ε,α (V )) ⊂ Cd−∞
∗ A(l),r,ε,α (V ).
0
This exercise follows from the rules for conjugation of linear transformations by
Fourier transform.
The theorem follows now from the following specialization.
Theorem 10.3.7. Let r, ε, α > 0 with ε < 1 and let l ∈ V ∗ s.t. ||l|| = 1. Let
ν : V → V be a diffeomorphism such that
(1) ν(0) = 0
(2) ||dx ν − Id|| < ε
(3) for any x ∈ Br (0) and any r0 ≤ r we have ν(Br0 (x)) = Br0 (ν(x)).
−∞
Let ξ ∈ Cl,r,ε,α (V ). Then
−∞
ν ∗ ξ ∈ Cl,r,ε (V ).
p
Proof. Let C be as in Exercise 10.3.4. Fix λ ∈ F with |λ| > α + C + C(C + 2α).
√
Let δ = 1/( 2λC). Thus α + δ −1 < |λ| < C −1 δ −2 . It is enough to show that for
any l0 ∈ Bε (l) we have F (ν ∗ ξ)(λl0 ) = 0. Present Br (0) as a disjoint union of balls
Ui = Bδ (xi ) of radius δ. We have
X
F (ν ∗ ξ)(λl0 ) = F (ν ∗ (1Ui ξ))(λl0 ).
i
Since |λ| < C −1 −2

δ , Exercises 10.3.4 and 10.3.5 imply
F (ν ∗ (1Ui ξ))(λl0 ) = F ((Af fxi ν)∗ (1Ui ξ))(λl0 ).

−∞
By Exercise 10.3.2 (i) and by the assumptions 1Ui ξ ∈ Cl,r,ε,α+δ −1 (V ). By Exercise
10.3.6 this implies
(Af fxi ν)∗ (1Ui ξ) ∈ Cd−∞

∗ ν(l),r,ε,α+δ −1 (V ).
xi
Since |λ| > α + δ −1 and l0 ∈ Bε (d∗xi ν(l)) we have
F ((Af fxi ν)∗ (1Ui ξ))(λl0 ) = 0 ∀i.
Summarizing, we have
X X
F (ν ∗ ξ)(λl0 ) = F (ν ∗ (1Ui ξ))(λl0 ) = F ((Af fxi ν)∗ (1Ui ξ))(λl0 ) = 0.
i i
Remark 10.3.8. (1) The method in the proof is called the stationary phase
method, which is a central method in microlocal analysis.
P
(2) In order to use the affine approximation we decomposed ξ = i 1Ui ξ. The
multiplication of ξ by 1Ui “damaged" ξ but this “damage" can be controlled
using 10.3.2(i), and is apparently moderate since the affine approximation
is good to the second order.
11. The Weil representation, the oscillator representation and an

application
In this section we show that the Fourier transform is not alone - it is part of an
infinite group of operators. This group is a representation of a double cover of
SL2 (F ). Thus this section requires some knowledge of representation theory. To
motivate the existence of this representation we first describe the Heisenberg group
and its representations.
Definition 11.0.1. Let V := F n and let ω be the standard symplectic form on

Wn := V ⊕ V ∗ . The Heisenberg group Hn is the algebraic group with underlying
algebraic variety Wn × F with the group law given by
(w1 , z1 )(w2 , z2 ) = (w1 + w2 , z1 + z2 + 1/2ω(w1 , w2 )).
Define a unitary character χ of R by χ(z) := exp(2πiz).
Definition 11.0.2. The oscillator representation of Hn is given on the space L2 (V )

by
(3) (σ(x, ϕ, z)f )(y) := χ(ϕ(y) + z))f (x + y).
Note that the center of Hn is 0 × F , and it acts on σ by the character χ, which can
be trivially extended to a character of V ∗ × F .
It is easy to see that σ is the unitary induction of (the extension of) the character
χ from V ∗ × R to Hn = (V ⊕ V ∗ ) × F .
Lemma 11.0.3. The space of smooth vectors in σ is S(V ), and the Lie algebra of
Hn acts on it by
(4) σ(v)f := ∂v f, σ(ϕ)f := ϕf, σ(z)f := 2πizf.

Proof. Formula (4) is obtained from (3) by derivation. Now, it is known that the
space of smooth vectors in a unitary induction consists of the smooth L2 functions
whose derivatives also lie in L2 .
Theorem 11.0.4 (Stone-von-Neumann). The oscillator representation σ is the

only irreducible unitary representation of Hn with central character χ.
Idea of the proof. Let me ignore all the analytic difficulties. Consider the normal
commutative subgroup A := V ×F . Conjugation in Hn defines an action of V on the
dual group of A. This action has only two orbits. The closed orbit is the singalton
{1} and the open orbit O is the complement to the closed one. The restriction σ|A
decomposes to a direct integral of characters in O, each “with multiplicity one".
The restriction of any non-zero subrepresentation ρ ⊂ σ to A will also include χ,
and thus the whole orbit O of χ. Thus ρ = σ and σ is irreducible.
Now let τ be any irreducible unitary representation of Hn with central character
χ. Then the restriction of τ to A will again include all the characters in O with
multiplicity one. Thus τ is the induction of an irreducible representation of the
stabilizer of χ in Hn . However, this stabilizer is A and thus τ ' σ.
Note that the symplectic group Sp(V ⊕ V ∗ ) acts on Hn by automorphisms, pre-

serving the center. Thus the theorem implies the following corollary.
Corollary 11.0.5. For every g ∈ Sp(V ⊕ V ∗ ) there exists a (unique up to a

scalar multiple) linear automorphism T of S(V ) such that For any h ∈ Hn we
have σ(hg ) = T σ(h)T −1 .
The uniqueness part of Corollary 11.0.5 follows from Schur’s lemmas. Corollary
11.0.5 defines a projective representation of Sp(V ⊕ V ∗ ) on S(V ), i.e. a map τ :
Sp(V ⊕ V ∗ ) → GL(S(V )) such that τ (gh) = λg,h τ (g)τ (h). It is not possible to
coordinate the scalars in order to obtain an honest representation of Sp(V ⊕ V ∗ ),
f ⊕ V ∗ ), called the
but it is possible to obtain a representation of a double cover Sp(V
metaplectic group. This was shown by A. Weil. We will not give the formulas for
f 2 (F )
the full Weil representation, but rather for its restriction to the subgroup SL
embedded by
E11 7→ IdV , E12 7→ T, E21 7→ T −1 , E22 7→ IdV ∗ ,
where by IdV we mean the operator that is Id on V and zero on V ∗ and by T the
operator that is zero on V ∗ and on V is given by ω.
Suppose that F is non-archimedean, and identify V with V ∗ using a non-degenerate
quadratic form q. Also, identify σ with S(V ). Let n := dim V . Then the Weil
representation is given by
!
1 u
(5) π( , ε)f (v) = εn ψ(uq(v))f (v)
0 1
!
t 0
(6) π( , ε)f (v) = εn |t|n/2 γ(q)γ(tq)−1 f (tv)
0 t−1
!
0 1
(7) π( , ε)f (v) = εn γ(q)F (f )(v)
−1 0
Here, γ(q) is a certain eights root of unity that depends on the quadratic form, and
ψ is the non-trivial unitary additive character that we use to identify V̌ with V ∗ .
Note that this representation defines a representation of SL2 (F ) if and only if n is
even. The existence of the Weil representation and the formulas above imply the
following corollary.
Corollary 11.0.6 (Rallis - Schiffmann). Let ξ ∈ S(V ) and let q be a non-degenerate

quadratic form on V . Let Z denote the zeros of q in V . Let ξ ∈ S ∗ (V ) such that
suppξ ⊂ V and suppF (ξ) ⊂ V . Then ξ = γ(q)F (ξ) and
hξ, f (tv)i = |t|−n/2 γ(q)−1 γ(tq)hξ, f (v)i.
Moreover, if ξ 6= 0 then n is even.

!
1 u
Proof. Since suppξ ⊂ Z, we have ψ(uq(v))ξ = ξ and thus π( , ε)ξ =
0 1
! ! ! !
0 1 1 u 0 1 0 1
εn ξ. Since = , and suppF (ξ) ⊂ Z,
−1 0 0 1 −1 0 −1 0
! !
1 0 1 ∗
(7) and (5) imply π( , ε)ξ = ε ξ. Since the subgroups
n
and
u 1 0 1
!
1 0 g2 (F ). We
generate SL2 (F ), we get that (g, ε)ξ = εn ξ for any (g, ε) ∈ SL
∗ 1
can assume ξ 6= 0. Then we get that n is even, and SL g2 (F ) acts trivially on ξ. The
lemma follows now from (6) and (7).
In the case F = R, one can prove a slightly weaker lemma. The problem in
generalizing the argument above to the archimedean case is that in this case the
condition suppF (ξ) ⊂ V does not imply ψ(uq(v))ξ = 0, but rather that there exists
k such that ψ(uq(v))k ξ = 0. On the other hand, in this case one can use the Lie
algebra sl2 .
For the Archimedean version of the corollary we will need the following definition.
Definition 11.0.7. For any t ∈ F × denote by ρ(t) the homotheties action on

Sc∗ (V ). Denote also δ(t) := γ(q)/γ(tq).
We say that a distribution ξ ∈ S ∗ (V ) is adapted to q if for any t ∈ F × we have
either (i) ρ(t)ξ = δ(t)|t|dim V /2 ξ or (ii) ρ(t)ξ = δ(t)t|t|dim V /2 ξ.
Theorem 11.0.8. Let L ⊂ SV∗ (Z(q)) be a non-zero subspace such that for all ξ ∈ L
we have Fq (ξ) ∈ L and q ∙ ξ ∈ L (here B is viewed as a quadratic function).
Then there exists a non-zero distribution ξ ∈ L which is adapted to q.
Using it one obtains the following result.
Proposition 11.0.9. Let V be a real vector space and let q be a non-degenerate

quadratic form on V . Let Z denote the zeros of q in V . Let ξ ∈ S ∗ (V ) such that
suppξ ⊂ V and suppF (ξ) ⊂ V . Then there exists a unitary character χ of R× and
m ∈ {0, 1} either such that for any t ∈ R× we have
hξ, f (tv)i = |t|−n/2−m χ(t)hξ, f (v)i.
12. Schwartz functions on Nash manifolds
12.1. Semi-algebraic sets and the Seidenberg-Tarski theorem. In this sec-

tion we follow [BCR].
Definition 12.1.1. A subset A ⊂ Rn is called a semi-algebraic set if it can

be presented as a finite union of sets defined by a finite number of polynomial
equalities and inequalities. In other words, if there exist finitely many polynomials
fij , gik ∈ R[x1 , ..., xn ] such that
r
[
A= {x ∈ Rn |fi1 (x) > 0, ..., fisi (x) > 0, gi1 (x) = 0, ..., giti (x) = 0}.
i=1
Lemma 12.1.2. The collection of semi-algebraic sets is closed with respect to finite
unions, finite intersections and complements.
Example 12.1.3. The semi-algebraic subsets of R are unions of finite number of

(finite or infinite) intervals.
In fact, a semi-algebraic subset is the same as a union of connected components of

an affine real algebraic variety.
Definition 12.1.4. Let A ⊂ Rn and B ⊂ Rm be semi-algebraic sets. A mapping

ν : A → B is called semi-algebraic iff its graph is a semi-algebraic subset of
Rm+n .
Proposition 12.1.5. Let ν be a bijective semi-algebraic mapping. Then the inverse

mapping ν −1 is also semi-algebraic.
Proof. The graph of ν is obtained from the graph of ν −1 by switching the coordi-
nates.
One of the main tools in the theory of semi-algebraic spaces is the Tarski-Seidenberg
principle of quantifier elimination. Here we will formulate and use a special case of
it. We start from the geometric formulation.
Theorem 12.1.6. Let A ⊂ Rn be a semi-algebraic subset and p : Rn → Rn−1 be

the standard projection. Then the image p(A) is a semi-algebraic subset of Rn−1 .
By induction and a standard graph argument we get the following corollary.
Corollary 12.1.7. An image of a semi-algebraic subset of Rn under a semi-

algebraic map is semi-algebraic.
Sometimes it is more convenient to use the logical formulation of the Tarski-

Seidenberg principle. Informally it says that any set that can be described in
semi-algebraic language is semi-algebraic. We will now give the logical formulation
and immediately after that define the logical notion used in it.
Theorem 12.1.8 (Tarski-Seidenberg principle, see e.g.[BCR, Proposition 2.2.4] ).

Let Φ be a formula of the language L(R) of ordered fields with parameters in R.
Then there exists a quantifier - free formula Ψ of L(R) with the same free variables
x1 , . . . , xn as Φ such that ∀x ∈ Rn , Φ(x) ⇔ Ψ(x).
Definition 12.1.9. A formula of the language of ordered fields with pa-

rameters in R is a formula written with a finite number of conjunctions, disjunc-
tions, negations and universal and existential quantifiers (∀ and ∃) on variables,
starting from atomic formulas which are formulas of the kind f (x1 , . . . , xn ) = 0
or g(x1 , . . . , xn ) > 0, where f and g are polynomials with coefficients in R. The
free variables of a formula are those variables of the polynomials which are not
quantified. We denote the language of such formulas by L(R).
Notation 12.1.10. Let Φ be a formula of L(R) with free variables x1 , . . . , xn . It

defines the set of all points (x1 , . . . , xn ) in Rn that satisfy Φ. We denote this set by
SΦ . In short,
SΦ := {x ∈ Rn |Φ(x)}.
Corollary 12.1.11. Let Φ be a formula of L(R). Then SΦ is a semi-algebraic set.
Proof. Let Ψ be a quantifier-free formula equivalent to Φ. The set SΨ is semi-

algebraic since it is a finite union of sets defined by polynomial equalities and
inequalities. Hence SΦ is also semi-algebraic since SΦ = SΨ .
Proposition 12.1.12. The logical formulation of the Seidenberg-Tarski principle

implies the geometric one.
Proof. Let A ⊂ Rn be a semi-algebraic subset, and pr : Rn → Rn−1 the standard

projection. Then there exists a formula Φ ∈ L(R) such that A = SΦ . Then
pr(A) = SΨ where
Ψ(y) = “∃x ∈ Rn (pr(x) = y ∧ Φ(x))”.
Since Ψ ∈ L(R), the proposition follows now from the previous corollary.
In fact, it is not difficult to deduce the logical formulation from the geometric one.
Let us now demonstrate how to use the logical formulation of the Seidenberg-Tarski
theorem.
Corollary 12.1.13. The closure of a semi-algebraic set is semi-algebraic.
Proof. Let A ⊂ Rn be a semi-algebraic subset, and let A be its closure. Then

A = SΨ where
Ψ(x) = “∀ε > 0 ∃y ∈ A |x − y|2 < ε”.
Clearly, Ψ ∈ L(R) and hence A is semi-algebraic.
Corollary 12.1.14. The derivative f 0 of any differentiable semi-algebraic function

f : R → R is semi-algebraic.
Proof. The graph of f 0 equals SΨ , where
Ψ(x, y) = “∀ε > 0 ∃δ > 0, s.t. ∀0 6= δ 0 ∈ (−δ, δ) we have

(f (x + δ 0 ) − f (x) − yδ 0 )2 < εδ 0 )2 ”.
Clearly, Ψ ∈ L(R) and hence f 0 is semi-algebraic.
Corollary 12.1.15.
(i) The composition of semi-algebraic mappings is semi-algebraic.
(ii) The R-valued semi-algebraic functions on a semi-algebraic set A form a ring,
and any nowhere vanishing semi-algebraic function is invertible in this ring.
(iii) Images and preimages of semi-algebraic sets under semi-algebraic mappings
are semi-algebraic.
Proposition 12.1.16. [BCR, Proposition 2.4.5] Any semi-algebraic set in Rn has

a finite number of connected components.
Remark 12.1.17. Over a non-archimedean local field F (e.g. F = Qp ) one con-

siders sets that are finite unions of finite intersections of sets of the form
{x ∈ F n s.t. p(x) is a k-th power}, or {x ∈ F n s.t. p(x) = 0.
An analog of the Seidenberg - Tarski theorem holds for such sets.
12.2. Nash manifolds. Let us now define the category of Nash manifolds, i.e.
smooth semi-algebraic manifolds. I like this category since the Nash manifolds
behave as tamely as algebraic varieties (e.g. posses some finiteness properties, and
admit an analog of Hironaka’s desingularization theorem), and in addition their
local structure is almost as easy as that of differentiable manifolds. In particular,

they are locally trivial, and analogs of the implicit function theorem and the tubular
neighborhood hold for them. The only thing we “loose" is the partition of unity.
Nash has shown that any compact smooth manifold has a unique structure of
an affine Nash manifold. It was later shown that it also has uncountably many
structures of non-affine Nash manifold. It is Artin-Mazur who first used the term
of Nash manifold. They gave a fundamental theorem which states that an affine
Nash manifold can be imbedded in a Euclidean space so that the image contains
no singular points of its Zariski closure.
In this section we follow [BCR, Shi].
Definition 12.2.1. A Nash map from an open semi-algebraic subset U of Rn to

an open semi-algebraic subset V ⊂ Rm is a smooth (i.e. infinitely differentiable)
semi-algebraic function. The ring of R-valued Nash functions on U is denoted by
N (U ). A Nash diffeomorphism is a Nash bijection whose inverse map is also
Nash.
As we are going to do semi-algebraic differential geometry, we will need a semi-

algebraic version of implicit function theorem.
Theorem 12.2.2 (Implicit Function Theorem for Nash manifolds, see e.g. [BCR,
Corollary 2.9.8]). Let (x0 , y 0 ) ∈ Rn+p , and let f1 , ..., fp be semi-algebraic smooth
functions on an open neighborhood of (x0 , y 0 ), such that fj (x0 , y 0 ) = 0 for j = 1, .., p
∂f
and the matrix [ ∂yji (x0 , y 0 )] is invertible. Then there exist open semi-algebraic
neighborhoods U (resp. V) of x0 (resp. y 0 ) in Rn (resp. Rp ) and a Nash mapping
φ, such that φ(x0 ) = y 0 and f1 (x, y) = ... = fp (x, y) = 0 ⇔ y = φ(x) for every
(x, y) ∈ U × V.
Definition 12.2.3. A Nash submanifold of Rn is a semi-algebraic subset of Rn

which is a smooth submanifold.
By the implicit function theorem it is easy to see that this definition is equivalent
to the following one, given in [BCR]:
Definition 12.2.4. A semi-algebraic subset M of Rn is said to be a Nash sub-

manifold of Rn of dimension d if, for every point x of M , there exists a Nash
diffeomorphism φ from an open semi-algebraic neighborhood Ω of the origin in Rn
onto an open semi-algebraic neighborhood Ω0 of x in Rn such that φ(0) = x and
φ(Rd × {0} ∩ Ω) = M ∩ Ω0 .
Definition 12.2.5. A Nash map from a Nash submanifold M of Rm to a Nash

submanifold N of Rn is a semi-algebraic smooth map.
Any open semi-algebraic subset of a Nash submanifold of Rn is also a Nash sub-

manifold of Rn .
Theorem 12.2.6 ([BCR, §2]). Let M ⊂ Rn be a Nash submanifold. Then it has

the same dimension as its Zarisky closure.
Unfortunately, open semi-algebraic sets in Rn do not form a topology, but only a

restricted topology. That is, the collection of open semi-algebraic sets is closed only
under finite intersections and unions but not under infinite unions. For this reason
we will consider only finite covers.
We will use this restricted topology to “glue” affine Nash manifolds and define
Nash manifolds exactly in the same way as algebraic varieties are glued from affine
algebraic varieties.
Definition 12.2.7. A R-space is a pair (M, OM ) where M is a restricted topo-

logical space and OM a sheaf of R-algebras over M which is a subsheaf of the sheaf
R[M ] of real-valued functions on M .
A morphism between R-spaces (M, OM ) and (N, ON ) is a continuous map f :
M → N , such that the induced morphism of sheaves f ∗ : f ∗ (R[N ]) → R[M ] maps
ON to OM .
◦
Example 12.2.8. Take for M a Nash submanifold of Rn , and for S(M ) the family
of all open subsets of M which are semi-algebraic in Rn . For any open (semi-
algebraic) subset U of M we take as OM (U ) the algebra N (U ) of Nash functions
U → R.
Definition 12.2.9. An affine Nash manifold is an R-space which is isomorphic

to an R-space of a closed Nash submanifold of Rn . A morphism between two affine
Nash manifolds is a morphism of R-spaces between them.
Example 12.2.10. Any real nonsingular affine algebraic variety has a natural
structure of an affine Nash manifold.
Remark 12.2.11. Let M ⊂ Rm and N ⊂ Rn be Nash submanifolds. Then a Nash

map between them is the same as a morphism of affine Nash manifolds between
them.
Let f : M → N be a Nash map. Since an inverse of a semi-algebraic map is semi-
algebraic, f is a diffeomorphism if and only if it is an isomorphism of affine Nash
manifolds. Therefore we will call such f a Nash diffeomorphism.
In [Shi] there is another but equivalent definition of affine Nash manifold.
Definition 12.2.12. An affine C ∞ Nash manifold is an R-space which is iso-

morphic to an R-space of a Nash submanifold of Rn .
The equivalence of the definitions follows from the following theorem, which imme-
diately follows from [BCR, Theorem 8.4.6] and Proposition 12.1.16.
Theorem 12.2.13. Any affine C ∞ Nash manifold is Nash diffeomorphic to a union

of finite number of connected components of a real nonsingular affine algebraic
variety.
The book [Shi] usually uses the notion of affine C ω Nash manifold instead of affine
C ∞ Nash manifold, that we called here just Nash manifold. The two notions are
equivalent by the theorem of Malgrange (see [Mal] or [Shi, Corollary I.5.7]) and
hence equivalent to what we call just affine Nash manifold. In other words, any
Nash manifold has a natural structure of a real analytic manifold and any Nash
map between Nash manifolds is analytic.
One also considers C r -Nash manifolds for any 0 ≤ r < ∞. These satisfy all the
properties listed here, and in addition partition of unity.
Definition 12.2.14. A Nash manifold is an R-space (M, NM ) which has a finite
cover (Mi ) by open sets Mi such that the R-spaces (Mi , NM |Mi ) are isomorphic to
R-spaces of affine Nash manifolds.
A morphism between Nash manifolds is a morphism of R-spaces between them.
Such morphisms are called Nash maps, and isomorphisms are called Nash diffeo-
morphisms.
By Proposition 12.1.16, any Nash manifold is a union of a finite number of connected
components. Any semi-algebraic set can be stratified by Nash manifolds.
Any Nash manifold has a natural structure of a smooth manifold. Any real non-
singular algebraic variety has a natural structure of a Nash manifold.
It is well-known that the real projective space RPn is affine, see e.g. [BCR, Theorem
3.4.4]. Since any number of polynomial equations over R have the same set of
solutions as a single equation (which is the sum of squares of the left hand sides),
we get that any quasi-projective Nash manifold is affine.
Remark 12.2.15. Note that the additive group of real numbers and and the mul-
tiplicative group of positive real numbers are isomorphic as Lie groups and as Nash
manifolds, but are not isomorphic as Nash groups. Recently, the structure theory
of (almost) linear Nash groups was developed in [Sun].
The following theorem is a version of Hironaka’s theorem for Nash manifolds.
Theorem 12.2.16 ([Shi, Corollary I.5.11]). Let M be an affine Nash manifold.
Then there exists a compact affine nonsingular algebraic variety N and a closed
algebraic subvariety Z of N , which is empty if M is compact, such that Z has
only normal crossings in N and M is Nash diffeomorphic to a union of connected
components of N − Z.
It implies that Nash manifolds are locally trivial.
Theorem 12.2.17 ([Shi, Theorem I.5.12]). Any Nash manifold has a finite cover
by open submanifolds Nash diffeomorphic to Rn .
Theorem 12.2.18 ([AG10, Theorem 2.4.3]). Let M and N be Nash manifolds and
ν : M → N be a surjective submersive Nash map. Then locally (in the restricted
S
k
topology) it has a Nash section, i.e. there exists a finite open cover N = Ui such
i=1
that ν has a Nash section on each Ui .
This implies that any etale Nash map is a local diffeomorphism.

In our work on Schwartz functions we frequently use the following
Theorem 12.2.19. (Nash Tubular Neighborhood). Let Z ⊂ M ⊂ Rn be closed

affine Nash submanifolds. Equip M with the Riemannian metric induced from Rn .
Then Z has a Nash tubular neighborhood, i.e. there exists a strictly positive Nash
Z ∈ N (Z) and a Nash diffeomorphism between between an open Nash
function ρM
neighborhood of Z in M and the open Nash neighborhood of the zero section of the
normal bundle given by
Z (z).
{(z, v) ∈ NZM s.t. ||v|| < ρM
12.3. Schwartz functions on Nash manifolds. The Fréchet space S(Rn ) of

Schwartz functions on Rn was defined by Laurant Schwartz to be the space of all
smooth functions such that they and all their derivatives decay faster than 1/|x|n
for all n. In other words, S(Rn ) is the space of all f ∈ C ∞ such that |df | is bounded
for every differential operator d with polynomial coefficients. This definition makes
sense verbatim on any smooth affine algebraic variety and was extended in [ dCl] to
affine Nash manifolds, and in [AG08] (using some ideas from unpublished notes of
Casselman) to arbitrary Nash manifolds. One can also define Schwartz sections of
Nash bundles.
Definition 12.3.1. Let M be a Nash manifold, and let E be a Nash bundle over
S
k
it. Let M = Ui be an affine Nash trivialization of E. Then we have a map
i=1
L
k
φ: S(Ui ) → C ∞ (M, E). We define the space S(M, E) of global Schwartz
n
i=1
sections of E by S(M, E) := Imφ. We define the topology on this space using the
Lk
isomorphism S(M, E) ∼
= S(Ui )n /Kerφ.
i=1
As Schwartz functions cannot be restricted to open subsets, but can be continued

by 0 from open subsets, they form a cosheaf rather than a sheaf.
Let M be a Nash manifold, E be a Nash bundle over M and let S(M, E) denote
the space of Schwartz sections of E.
The following two theorems summarize some results from [dCl, AG08, AG10, AG13].
Theorem 12.3.2. Let U ⊂ M be an open (Nash) submanifold and N ⊂ M be a

closed (Nash) submanifold. Then
(1) S(Rn ) = Classical Schwartz functions on Rn .
(2) Cc∞ (M, E) ⊂ S(M, E).

(3) The restriction maps S(M, E) onto S(Z, E|Z ).
(4) S(U, E) := S(U, E|U ) =
{ξ ∈ S(M, E)|ξ vanishes with all its derivatives on M − U }.
(5) Partition of unity: Let (Ui )ni=1 be a finite cover by open Nash submani-
folds. Then there exist smooth functions α1 , ..., αn such that supp(αi ) ⊂ Ui ,
Pn
αi = 1 and for any g ∈ S(M, E), αi g ∈ S(Ui , E).
i=1
(6) S(M, E) = S(M )S(M, E).
(7) S(M, E) is a nuclear Fréchet space.
ˆ
(8) For any Nash manifold M 0 we have S(M × M 0 ) = S(M )⊗S(M 0
).
Let us now sketch the proof of some parts of this theorem.

For part (1) we remark that Nash functions have polynomial growth.
Part (2) is obvious for affine M . For general M let f ∈ Cc∞ (M, E) and let M =
Sn P
i=1 Ui be an open affine cover of M . Choose a partition of unity 1 = i αi
corresponding to this cover. Then αi f ∈ Cc (Ui , E) ⊂ S(Ui , E) and thus f =
∞
P
αi f ∈ S(M, E).
Part (3) follows from the Nash tubular neighborhood theorem (Theorem 12.2.19).
Part (4) we will show later.
Part (5) is proven similarly to the classical partition of unity for smooth functions.
We first prove “tempered" partition of unity, i.e. the existence of functions αi
supported in Ui and with polynomial growth, and then use this partition for a
refined cover Vi ⊂ Ui . Then the temperedness of αi guarantees αi g ∈ S(M ) and
suppαi ⊂ Vi guarantees the vanishing of αi near the boundary of Ui , which in turn
implies αi f ⊂ S(Ui ).
Part (6) we will not prove here due to lack of time.
For part (7) we note that a quotient of a nuclear Fréchet space by a closed subspace
is nuclear Fréchet, and thus it is enough to show that S(Rn ) is nuclear Fréchet. This
follows from its definition. For this we should say a couple of words on what nuclear
means. It means that for any other Fréchet space, the projective and the injective
topologies on their tensor product coincide. These are two natural topologies, and
most other topologies are stronger than injective and weaker than projective. An
inverse limit of spaces with inclusion maps that are Hilbert-Schmidt are nuclear,
and that is why S(Rn ) is nuclear.
Part (8) we discussed already for compactly supported functions, and this discussion
implies that S(M )⊗S(M 0 ) naturally embeds into S(M ×M 0 ) with dense image. We
cannot prove the full statement at this point, since we have not discussed nuclear
spaces and completed tensor products.
We denote by S ∗ (M ) the dual space to S(M ) and call it the space of tempered
distributions. From property (4) we obtain
Corollary 12.3.3. Let M be a Nash manifold and U ⊂ M be an open Nash subset.

Let E be a Nash bundle over M . Then we have a short exact sequence
0 → SM
∗
\U (M ) → S (M ) → S (U ) → 0.
∗ ∗
Theorem 12.3.4. Let N ⊂ M be a closed submanifold. Denote
SN (M )i := {φ ∈ S(M ) s.t. φ is 0 on N with first i − 1 derivatives}.
Let CNM
N
denote the conormal bundle to N in M . Then
(8) S(M )i /S(M )i+1 ∼

= S(N, Symi+1 (CNM N
))
(9) S(M \ N ) ∼
= lim S(M )i /S(M )i+1 .
←
Corollary 12.3.5. Let M be a Nash manifold, and N ⊂ M be a closed Nash

submanifold. Then the space SN ∗
(M ) has a natural filtration Fi by the order of
S
transversal derivatives. This filtration satisfies SN
∗
(M ) = Fi and
Fi /Fi−1 ∼
= S ∗ (N, Symi+1 (CNM
N
)).
We frequently use this corollary when we have a natural stratification of M since it

allows to reduce the analysis of equivariant distributions from M to single strata.
Theorem 12.3.6 ([AG09], Theorem B.2.4). Let φ : M → N be a Nash submersion

of Nash manifolds. Fix Nash measures μ on M and ν on N . Then
(i) there exists a unique continuous linear map φ∗ : S(M ) → S(N ) such that for
any f ∈ S(N ) and g ∈ S(M ) we have
ˆ ˆ
f (x)φ∗ g(x)dν = (f (φ(x)))g(x)dμ.
x∈N x∈M
In particular, we mean that both integrals converge.

(ii) If φ is surjective then φ∗ is surjective.
In fact ˆ
φ∗ g(x) = g(z)dρ
z∈φ−1 (x)
for an appropriate measure ρ.
For the proof of part (4) we will need several lemmas.
Lemma 12.3.7. Let U ⊂ Rn be an open semi-algebraic subset. For any φ : U → R

denote by φe : M → R its extension by 0 outside U . Let φ ∈ S(U ). Then φe is
differentiable at least once and for any Nash differential operator D of order 1 on
^
Rn , Dφe = D| U φ.
Proof. We have to show that for any z ∈ Rn \ U , φe is differentiable at least once at

z and its derivative at z in any direction is 0. Denote Fz (x) := ||x − z||. Clearly,
1/Fz2 ∈ N (U ). Hence φ/Fz2 is bounded in U and therefore φ/F e 2 is bounded on
z
Rn \ {z}, which finishes the proof.
Lemma 12.3.8 ([BCR, Proposition 2.6.4] ). Let F : A → R be a semi-algebraic

function on a locally closed semi-algebraic set. Let Z(F ) := {x ∈ A|F (x) = 0} be
the set of zeros of F and let AF := A \ Z(F ) be its complement. Let G : AF → R
be a semi-algebraic function. Suppose that F and G are continuous. Then there
exists an integer N > 0 such that the function F N G, extended by 0 to Z(F ), is
continuous on A.
The following two lemmas are straightforward.
Lemma 12.3.9. Let U ⊂ Rn be open (semi-algebraic) subset. Then any Nash

Pk
differential operator D on U can be written as i=1 fi (Di |U ) where fi are Nash
functions on U and Di are Nash differential operators on Rn .
Lemma 12.3.10. Suppose α ∈ C ∞ (R) vanishes at 0 with all its derivatives. Then
for any natural number n, α(t) = (n!)−1 tn α(n) (θ) for some θ ∈ [0, t].
Proof of part (4). Denote Z := X \ U and
WZ := {ξ ∈ S(M, E)|ξ vanishes with all its derivatives on M − U }.
We have to show that the extension by zero defines a continuous isomorphism

between S(U ) and WZ .
Case 1 M = RN .
Lemma 12.3.7 implies by induction that the extension by zero continuously
maps S(U ) into WZ . Let us show that this map is onto.
Let φ ∈ WZ . For any point x ∈ RN define r(x) := dist(x, Z). Let
S := S(0, 1) ∈ RN be the unit sphere. Consider the function ψ on S ×Z ×R
defined by ψ(s, z, t) := φ(z + ts). From Lemma 12.3.10 we see that
∂n
ψ(s, z, t) = tn ψ(x, s, t)|t=θ
(∂t)n
n
for some θ ∈ [0, t]. As φ is Schwartz, it is easy to see that (∂t)
∂
n ψ(x, s, t) is
bounded on Z × S × R. Therefore |ψ(s, z, t)| ≤ C|t| for some constant C

n
and hence φ/r n is bounded on RN for any n.

Let h be a Nash function on U . By Lemma 12.3.8, rn h extends by 0
to a continuous semi-algebraic function on RN for n big enough. It can be
majorated by f ∈ N (RN ). Therefore
|φh| = |(φ/r n )rn h| ≤ |φf |/rn .
φf ∈ W, thus |φf |/rn is bounded and hence |φh| is bounded.

For any Nash differential operator D on RN , Dφ ∈ W . Hence hDφ is
bounded. By Lemma 12.3.9, every Nash differential operator on U is a sum
of differential operators of the form hD|U , where D is a Nash differential
operator on RN and h a Nash function on U . Hence φ|U ∈ S(U ).
Case 2 M is affine.
Follows from the previous case and property (3) (extension from a closed
Nash submanifold).
Case 3 General case.
Choose an affine cover of M . The theorem now follows from the previous
case and partition of unity.

Recall that this property implies, by the Hahn-Banach theorem, that the restriction
S ∗ (M ) → S ∗ (U ) is onto (Corollary 12.3.3). Let us demonstrate a classical corollary
of this fact.
Corollary 12.3.11. Any tempered generalized function on Rn has the form

Xk
∂ |αi |
f dx,
i=1
(∂x)αi
where αi are multi-indexes, dx is a Lebesgue measure and f is a continuous function
on Rn , with |f | bounded by a polynomial.
For example, δ0 ∈ S ∗ (R) is the second derivative of |x|.
Proof. Embed R into S 1 . This defines an embedding of Rn into T n = (S 1 )n . Let

ξ ∈ S ∗ (Rn ). By Corollary 12.3.3, ξ extends to η on T n . Let us use Fourier series.
Distributions correspond under these to sequences indexed by n numbers that are
bounded by some polynomial. Absolutely summable sequences define continuous
functions. Partial derivatives of functions correspond to multiplications of sequences
by polynomials in the indices. Thus, ξ corresponds to a sequence αi1 ,...,in bounded
by the product ij11 ∙ ∙ ∙ ijnn . Let q(i1 , . . . , in ) := i1j1 +2 ∙ ∙ ∙ injn +2 and let D be the
corresponding differential operator with constant coefficients. Arguing by induction
on n we can assume that if one of the numbers i1 , . . . , in vanishes then so does the
corresponding coefficient αi1 ,...,in . Let β := α/q and let h ∈ C(T n ) be the function
with the Fourier series β. Then η = D(hdt), where dt is the Haar probability
measure on T n . Thus ξ = D|Rn (h|Rn ). Now, h is bounded as a function on T n and
thus h|Rn is also bounded. When we take the stereographic projection into account
we get that D|Rn does not have constant coefficients. However, its coefficients are
rational functions with nowhere vanishing denominator. Similarly, dt|Rn = pdx,
where p is a rational function with nowhere vanishing denominator. Altogether,
we get that there exists a differential operator Δ with constant coefficients and a
continuous function f bounded by a polynomial such that
ξ = D|Rn (h|Rn dt|Rn ) = Δf dx.

Corollary 12.3.12. Let M be an affine Nash manifold and let ξ be a tempered

generalized function on M . Then there exists a Nash differential operator D on M
and a bounded continuous function f ∈ C(M ) such that ξ = Df .
Proof. For the affine case embed M in Rn and extend ξ to Rn . There find f and
D from the previous corollary and restrict them to M . The general case follows by
partition of unity.
Let us demonstrate how to use ?? on a simple example.
Lemma 12.3.13. For any homogeneity degree α, the space of α-homogeneous even
distributions on F is one-dimensional.
Proof. By applying Fourier transform, we can assume α ≥ 1. To prove the non-

vanishing, let ξ be |x|α−1 dx.
To prove that any distribution in this space has to be a multiple of |x|α−1 dx,
it is enough to prove that the restriction to F × is an embedding. Indeed, any
distribution supported at 0 is a combination of derivatives of the δ0 and thus has
negative homogeneity degree.
The same statement with the same proof holds for odd distributions. Here, even
distribution means invariant under the coordinate change x 7→ −x, and odd means
anti-invariant.
Finally, we would like to remark on a different, extrinsic, approach to Schwartz
functions, applied in [CHM] and [KS]. We can compactify our manifold and de-
fine Schwartz functions on it as smooth functions on the (smooth) compactification
that vanish to infinite order on the complement to M . If both M and the com-
pactification are Nash manifolds then this definition will be equivalent, by Theorem
12.3.2(2,4). This allows to define Schwartz functions on non-Nash (say, subanalytic)
manifolds, but this space will depend on the compactification.
Schwartz functions on non-smooth algebraic varieties and more generally on Nash
varieties (i.e. varieties that can be locally described as zeros of Nash functions) can
be locally defined by restriction from a smooth ambient space, see [ESh, Ela].
Another realm in which one can define Schwartz functions is the category of tem-
pered manifolds. Indeed, our only use of (semi-)algebraicity was to have a scale of
infinitesimals. Such a scale exists in the wider generality of tempered manifolds -
manifolds that can be covered by open subsets that are identified with open subsets
in Rn , and coordinate changes between these subsets are tempered functions. A
tempered manifold is said to be of finite type if this cover is finite. Many properties
of Schwartz functions listed above continue to hold for tempered manifolds of finite
type, see [Sha].
13. Invariant distributions
Let gln (F ) denote the vector space of all square matrices of order n with coefficients
in F and GLn (F ) ⊂ gln (F ) denote the group of all invertible matrices of!order n.
g 0
Consider also the embedding of GLn (F ) into GLn+1 (F ) by g 7→ .
0 1
For a distribution ξ ∈ S ∗ (GLn+1 (F )) denote by ξ t the distribution given by hξ t , f i :=
hξ, f t i, where f t (X) = f (X t ) and X t denotes the transposed matrix. We let
GLn (F ) act on GLn+1 (F ) by conjugation via the embedding above. This action
defines an action on Schwartz functions and thus also on tempered distributions.
Denote by S ∗ (GLn+1 (F ))GLn (F ) the subspace of distributions invariant under this
action. In this section we sketch the proof of the following theorem
Theorem 13.0.1. For any ξ ∈ S ∗ (GLn+1 (F ))GLn (F ) , we have ξ = ξ t .
13.1. Proof for n = 1. Note that GL1 (F ) = F × and denote by G := F × n {±1},

where −1 acts on F × by λ 7→ λ−1 . Extend the action of F × on GL2 (F ) to the
action of G by letting (1, −1) act by transposition. Let χ be the character of G
given by projection on the second coordinate. Let S ∗ (GL2 (F ))G,χ denote the space
of tempered distributions that change under the action of G by the character χ.
Then the theorem is equivalent to the statement S ∗ (GL2 (F ))G,χ = 0.
Proposition 13.1.1. If S ∗ (gl2 )G,χ = 0 then S ∗ (GL2 )G,χ = 0.
Proof. Let ξ ∈ S ∗ (GL2 (F ))G,χ . We have to prove ξ = 0. Assume the contrary.

Take p ∈ Supp(ξ). Let t = det(p). Let f ∈ S(F ) be such that f vanishes in a
neighborhood of zero and f (t) 6= 0. Consider the determinant map det : GL2 (F ) →
F . Consider ξ 0 := (f ◦ det) ∙ ξ. It is easy to check that ξ 0 ∈ S ∗ (GL2 (F ))G,χ and
p ∈ Supp(ξ 0 ). However, we can extend ξ 0 by zero to ξ 00 ∈ S ∗ (gl2 (F ))G,χ , which is
zero by the assumption. Hence ξ 0 is also zero. Contradiction.
Now, note that the action of G on gl2 (F ) is isomorphic to the action on F 2 × F 2 ,

where on the first copy the action is trivial and on the second copy it is given by
(λ, 1) ∙ (x, y) = (λx, λ−1 y), and (1, −1) ∙ (x, y) = (y, x).
We can thus consider the second copy only, and prove S ∗ (F 2 )G,χ = 0.
Let U := {(x, y) ∈ F 2 | xy 6= 0}. Then U is locally isomorphic as a G-manifold to
F × ×F × , where the action on the first F × is trivial, and the action on the second one
is given by (λ, ε)∙z = (λ2 z)ε . Clearly, with this action we have S ∗ (F × ×F × )G,χ = 0
and thus we obtain
(10) S ∗ (U )G,χ = 0.
Now, denote by ρ the action of F × on F 2 given by λ(x, y) = (λx, λy). It defines

an action on S(F 2 ) and S ∗ (F 2 ).
Lemma 13.1.2. Let ξ ∈ S ∗ (F 2 )G,χ and let σ : F × → C be a character. Suppose

that for any λ ∈ F × and f ∈ S(F 2 ) we have hξ, ρ(λ−1 )f i = σ(λ)hξ, f i. Then
|σ(λ)| = |λ|−n , where n ≥ 0.
Proof. Assume first that F is p-adic and consider the restriction of ξ to F × × F .

This restriction is supported on F × × {0} and thus is defined on F × × {0}. Since
it is F × -invariant, it is a multiple of the Haar measure. If it is a non-zero multiple,
then n = 0. Now suppose that this restriction is 0. Similarly we get that either
n = 0 or ξ is supported at the origin. In the latter case we again get n = 0.
Now, let F be R or C and consider the restriction of ξ to F × × F , again supported
on F × × {0}. By ?? the space S ∗ (F × × F )F × ×{0} has a filtration by the order of
transversal derivatives. The elements in the 0-th filtra are homogeneous as before.
Since the derivation operators have negative homogeneity degrees, for other filtras
we can have only negative degrees.
Now, note that the Fourier transform preserves the space S ∗ (F 2 )G,χ . By (10), for
any S ∗ (F 2 )G,χ ⊂ SZ∗ (F 2 ), where Z is the zero set of the quadratic form q(x, y) = xy.
Thus, by Corollary 11.0.6, Proposition 11.0.9 and 13.1.2 we have S ∗ (F 2 )G,χ = 0.
13.2. Luna slice theorem, Frobenius reciprocity and Harish-Chandra de-

scent. ??
13.3. Sketch of Proof for all n. ??
References
[AG08] A. Aizenbud, D. Gourevitch: Schwartz functions on Nash Manifolds, International
Mathematics Research Notices IMRN 2008, no. 5, Art. ID rnm 155, 37 pp. See also
arXiv:0704.2891v3 [math.AG].
[AG10] A. Aizenbud, D. Gourevitch:.: De-Rham theorem and Shapiro lemma for Schwartz func-
tions on Nash manifolds, Israel Journal of Mathematics 177 (2010), pp 155-188. See also
arXiv:0802.3305[math.AG].
[AG09] A. Aizenbud, D. Gourevitch:Generalized Harish-Chandra descent, Gelfand pairs and an
Archimedean analog of Jacquet-Rallis’ Theorem, Duke Mathematical Journal, 149, n. 3,
509-567 (2009). See also arXiv: 0812.5063[math.RT].
[AG13] A. Aizenbud, D. Gourevitch: Smooth Transfer of Kloostermann Integrals, American Jour-
nal of Mathematics 135, 143-182 (2013). See also arXiv:1001.2490[math.RT].
[AT08] A. Arhangel’skii and M. Tkachenko, Topological groups and related structures, Atlantis
Studies in Mathematics, Vol 1, Atlantis Press, Paris; World Scientific Publishing Co. Pte.
Ltd., Hackensack, NJ (2008).
[BCR] J. Bochnak; M. Coste, M-F. Roy: Real Algebraic Geometry Berlin: Springer, 1998.
[CHM] W. Casselman, H. Hecht, Henryk; D. Miličić: Bruhat filtrations and Whittaker vectors for
real groups. The mathematical legacy of Harish-Chandra (Baltimore, MD, 1998), 151-190,
Proc. Sympos. Pure Math., 68, Amer. Math. Soc., Providence, RI, (2000).
[dCl] F. du Cloux: Sur les representations differentiables des groupes de Lie algebriques. Annales
scientifiques de l E.N.S. 4e serie, tome 24, no 3, p. 257-318 (1991).
[Ela] B. Elazar: Schwartz functions on quasi-Nash varieties, preprint, arXiv:1711.05804.
[ESh] B. Elazar, A. Shaviv:Schwartz functions on real algebraic varieties, Canadian Journal of

Mathematics (to appear), DOI:10.4153/CJM-2017-042-6.
[Hör90] L. Hörmander, The analysis of linear partial differential operators. I. Distribution theory
and Fourier analysis. Grundlehren der Mathematischen Wissenschaften 256. Springer-Verlag,
Berlin, 1990.
[Kan04] Kanwal, Ram P., Generalized functions: Theory and Applications, third edition,
Birkhäuser Boston, Inc., Boston, MA, xviii+476 (2004).
[KS] M. Kashiwara, P. Schapira: Ind-Sheaves. Asterisques 271, (2001).
[Mal] B. Malgrange: Ideals of differentiable functions. Oxford: Oxford University Press, 1966.
[Rud06] Walter Rudin: Functional Analysis, McGraw-Hill (2006) 424 pages.
[Sha] A. Shaviv: Tempered manifolds and Schwartz functions on them, in preparation.
[Shi] M. Shiota: Nash Manifolds Lecture Notes in Mathematics 1269. New York: Springer, 1987.
[Sun] B. Sun: Almost linear Nash groups, arxiv:1310.8011.
[Tre67] F. Treves: Topological vector spaces, distributions and kernels, Academic Press, New
York-London xvi+624 pp (1967).

Gen Fun D

Uploaded by

Copyright:

Available Formats

Gen Fun D

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

Gen Fun D

Uploaded by

Copyright:

Available Formats

GENERALIZED FUNCTIONS LECTURES

Date: June 16, 2022.

4.7. Distributions on `-spaces 36

1. The space of generalized functions on Rn

that it also satisfies:

The following are possible motivations for generalized functions:

• Many natural analytic operations, e.g. derivation and Fourier transform,

Definition 1.2.1 (Smooth functions of compact support).

(i) Denote by C ∞ (R) the space of smooth functions f : R → R, i.e. functions

supp(f ) := {x ∈ R : f (x) 6= 0},

the closure of the set in which it does not vanish.

Exercise 1.2.3. Fix a non-zero ρ ∈ Cc∞ (R). Show that

Definition 1.2.4 (Distributions). A linear functional ξ : Cc∞ (R) → R is con-

Example 1.2.7. For any a ∈ R, define δa ∈ C −∞ (R) by hδa , f i := f (a).

Note that we have

C ∞ (R) ⊂ C(R) ⊂ L1Loc (R) ⊂ C −∞ (R),

lim hξn , f i = hξ, f i.

Note that in particular this definition applies to locally-L1 functions, since as we

One can similarly define these notions for sequences of distributions.

Exercise 1.2.12. A sequence {fn } in C −∞ (R) is a weakly Cauchy sequence if and

Exercise 1.2.14. Find a sequence of functions (fn )∞

weakly to 0 that does not converge point-wise.

Definition 1.2.15. A sequence φn ∈ Cc (R) of continuous, non-negative, compactly

Exercise 1.2.16. An approximation of identity weakly converges to δ0 .

Exercise 1.2.17. Find an approximation of identity.

1.3. Remarks on operations on distributions. In general, many spaces of func-

For defining operations on distributions we will often use a different approach.

1.4. Translations of generalized functions. For a function f ∈ C ∞ (R) and

Definition 1.4.1. Given a generalized function ξ, we define its translation Rt ξ by

hRt ξ, φi := hξ, R−t φi

Exercise 1.4.2. Define the notion of translation of a generalized function using

1.5. Derivatives of generalized functions. Let f, φ ∈ Cc∞ (R). Since hξf 0 , φi =

Definition 1.5.1. For any ξ ∈ C −∞ (R), define its derivative ξ 0 ∈ C −∞ (R) by

for any φ ∈ Cc∞ (R).

For example, the derivative of δ0 can be intuitively described as

Example 1.5.2. We have

Exercise 1.5.3. Find a function F ∈ L1Loc for which F 0 = δ0 as generalized func-

Exercise 1.5.4. Define the notion of derivative of a generalized function using

The following exercise enables to define the notion of derivative of a generalized

Exercise 1.5.5. Prove that ξ−R t

Definition 1.6.1. Let ξ ∈ C −∞ (R).

The following (difficult) exercise shows that the definition is meaningful.

Exercise 1.6.2 (*). Let ξ ∈ C −∞ (R).

We will discuss this in more details and in larger generality in ?? below.

Remark 1.6.3. Note that supp(ξ) is always a closed set.

Example 1.6.4. The support of δ0 is {0}.

Exercise 1.6.6. Let ξ1 , ξ2 ∈ C −∞ (R) and a, b ∈ R. Show that:

1.7. Products and convolutions of generalized functions.

Definition 1.7.1. Let f ∈ C ∞ (R) and ξ ∈ C −∞ (R). We would like to have

The convolution of two smooth functions is always smooth, if it exists. In addition,

However, some formal justification is required. Given a compact K ⊂ R, we say ρ

Exercise 1.7.6. Let ξ ∈ C −∞ (R). In an exercise above we showed: if φ ∈ Cc∞ (R)

To summarize, the convolution of two compactly supported distributions is well

Exercise 1.7.8. Let K ⊂ R be a compact set. Construct a function f ∈ Cc∞ (R)

to f in Cc∞ (Rn ) if:

1.9. Generalized functions and differential operators. A differential equa-

Exercise 1.9.1. Let A be a differential operator with constant coefficients (i.e. as

(2) Without using generalized functions, explain the equation A(G ∗ g) = g we

1.10. Regularization of generalized functions.

where n ∈ R0> , kn ∈ N0 and Σ denotes Minkowski sum, that is A+B := {a+b|a ∈

Bw := {U,S : > 0, S is finite} .

BS := {U,S : > 0, S is bounded} .