Twin Paradox and Space Topology

JeanPhilippe Uzan1,2 , JeanPierre Luminet3 , Roland Lehoucq4 and Patrick Peter3,5

(1) Laboratoire de Physique Theorique, CNRSUMR 8627, B
at. 210, Universite Paris XI, F-91405 Orsay cedex (France)
(2) Departement de Physique Theorique, Universite de Gen`eve, 24 quai E. Ansermet, CH-1211 Gen`
eve 4 (Switzerland)
(3) Departement dAstrophysique Relativiste et de Cosmologie, Observatoire de Paris, CNRSUMR 8629,
F92195 Meudon cedex (France)
(4) CESaclay, DSM/DAPNIA/Service dAstrophysique, F91191 Gif sur Yvette cedex (France).
(4) Institut dAstrophysique de Paris, 98 bis, boulevard Arago, 75005 Paris (France)
(June 20, 2000)
If space is compact, then a traveller twin can leave Earth, travel back home without changing
direction and find her sedentary twin older than herself. We show that the asymmetry between
their spacetime trajectories lies in a topological invariant of their spatial geodesics, namely the
homotopy class. This illustrates how the spacetime symmetry invariance group, although valid
locally, is broken down globally as soon as some points of space are identified. As a consequence, any
nontrivial space topology defines preferred inertial frames along which the proper time is longer
than along any other one.
Preprint numbers: LPTORSAY 00/45, UGVADPT 00/041080

tion, in III we investigate the case of a spacetime with

compact spatial sections and in IV, we show that the
root explanation of the twin paradox lies in the global
breakdown of the spacetime symmetry group by a non
trivial topology.

I. INTRODUCTION

The twin paradox is the best known thought experiment of special relativity, whose resolution provides interesting insights on the structure of spacetime and on the
applicability of the Lorentz transformations. In its seminal paper on special relativity, Einstein [1] pointed out
the problem of clocks synchronization between two inertial frames with relative velocity v. Later on, Langevin
[2] picturesquely formulated the problem by taking the
example of twins aging differently according to their respective worldlines. The keypoint for understanding the
paradox is the asymmetry between the spacetime trajectories of the sedentary twin and of the traveller twin.
The subject has been widely studied for pedagogical purpose [3,4], the role of acceleration was examined in details
[58] and a full general relativistic treatment was given
[9].
Although counter-intuitive, the twin paradox is clearly
not a logical contradiction, it merely illustrates the elasticity of time in relativistic mechanics. The experiment
was actually performed in 1971 with twin atomic clocks
initially synchronized, one of them being kept at rest on
Earth and the other one being embarked on a commercial flight: the time shifts perfectly agreed with the fully
relativistic calculations [10]
An interesting revisited paradox was formulated
[11,12] in the framework of a closed space (due to curvature or to topology). In such a case, the twins can
meet again without none of them being accelerated, yet
they aged differently. Both an algebraic and a geometric
solution were given [13].
Our present goal is to extend such explanations by
adding a topological characterization of reference frames,
which allows us to solve the twin paradox whatever the
global shape of space may be. We first briefly recall, in
II, the classical twin paradox and its standard resolu-

II. THE STANDARD TWIN PARADOX

Let observers 1 and 2 be attached to inertial frames

with relative velocity v. 1 is supposed to be at rest and to
experience no acceleration. At time t = 0, the observers
synchronise their clock (thus they can be called twins).
Then twin 2 travels away at velocity +v with respect to
1 and comes back with velocity v. According to special
relativity, the travel time, 2 , measured by 2 (its proper
time) is related to the proper time measured by 1, 1 ,
by
p
(1)
2 = 1 v 2 1 .
A paradox arises if one considers that the situation is perfectly symmetrical about 1 and 2, since 2 sees 1 travelling
away with velocity v and coming back with velocity +v.
If this was correct, one could reverse the reasoning to deduce that 1 should be younger than 2, with
p
(2)
1 = 1 v 2 2 ,
and an obvious contradiction would arise.
Indeed, the previous argument holds whenever 2 is not
accelerating. As first explained by Paul Langevin [2],
among all the worldlines that connect two spacetime
events (such as the departure and return of 2), the one
which has the longest proper time is the unaccelerated
one, i.e. the reference frame K of 1. The traveller twin 2
cannot avoid accelerating and decelerating to make its return journey; then he had to jump from an inertial frame
1

K0 moving relatively to K with velocity v to another inertial frame K00 moving with velocity v with respect
to K. Hence the situation is not symmetrical about the
twins: a kink (infinite acceleration) in the middle of the
path of twin 2 explains the difference, and there is no
contradiction in the fact that the sedentary twin 1 will
definitely be older than the traveller twin 2.
The same result holds in the framework of general relativity, dealing with a more realistic situation including
accelerations, gravitational fields and curved spacetime
(so that the kink is smoothed out): in order to achieve
its journey, the traveller 2 necessarily experiences a finite
and variable acceleration; thus her reference frame is not
equivalent to that of 1.
Such explanations, as rephrased by Bondi [14], are
equivalent to say that there is only one way of getting
from the first meeting point to the second without acceleration.
However, acceleration is not the only and essential
point of the twin paradox, as shown by the example of
nonaccelerated twins in a closed space, in which there
are several ways to go from the first meeting point to
the second one without accelerating [1113]. The key
explanation of the twin paradox is now some kind
of asymmetry between the spacetime paths joining two
events. We investigate below the nature of such an asymmetry when space topology is not trivial (i.e. simply
connected)

decomposed as the direct product

M=R

(3)

where the real axis R refers to the time direction and to

the three dimensional spatial sections. Now the topology
of spacetime amounts to the study of the various shapes
of the spatial sections .
The topology of a threedimensional Riemannian
space can be described in full generality by a fundamental polyhedron P and a holonomy group whose
elements g identify the faces of the polyhedron by pairs
(see e.g. [16] for a general discussion of the topological
properties of spacetimes). It follows that can be written as
= X/

(4)

where X is the universal covering space [16] (simply

connected and three dimensional) and the quotient refers
to the equivalence relation defined as
x, y X

x y (g

| x = gy) .

(5)

If reduces to the identity, space is simplyconnected,

in the sense that two points of space are connected
by only one geodesic. As soon as there are non
trivial holonomies which identify points, space is multi
connected and several geodesics connect two any distinct
points. In a cosmological context, multiconnected universe models lead to the existence of ghost images in
the observable universe when one topological length is
shorter than the horizon size. Many methods aimed to
detect the cosmic topology have been proposed [16,17],
so far with no definite answer coming from observational
data.
Returning to Minkowski space for the sake of clarity,
the holonomies of that preserve the flatness of the space
sections are the identity, the translations, the reflexions and the helicoidal displacements. The group classification leads to 18 Euclidean spaceforms with different topologies, all of them having as universal covering
space X the simplyconnected, infinite Euclidean space
R3 . Six of them are compact (i.e. of finite volume) and
orientable. Their fundamental polyhedron can be a parallelepiped or a hexagonal prism. For the sake of visualization, in the following we shall develop our reasoning in
flat spacetimes with 1+2 dimensions only, i.e. whose spatial sections are just surfaces. In such a case there are
5 Euclidean topologies: the cylinder, the Mobius strip,
the flat torus, the Klein bottle and the Euclidean plane
[16]. For a pedagogical purpose, we select the case where
space has a toruslike topology (see figure 1). However
we emphasize that our conclusions will remain valid in
(1+3) dimensions, whatever the topology and the (constant) space curvature may be.

III. TWINS IN A COMPACT SPACE

In a spacetime which has at least one compact space

dimension, one can actually start from one point, travel
along several straight geodesics and come back to the
same spatial position without accelerating nor decelerating. Einsteins relativity theory determines the local
properties of the spacetime M (its metrics), but gives
little information about its global properties (its topology) [15,16], so that special relativity (in the absence of
gravitational fields) or general relativity (involving gravitational fields in curved spacetimes) well describe the
local physics. For instance, the Minkowski spacetime
(M, ) used in special relativity is a manifold R4 with a
flat Lorentzian metric = diag(-1,1,1,1) and Euclidean
space sections . One can obain spacetimes locally identical to (M, ) but with different large scale properties
by identifying points in M under a group of transformations, called holonomies, which preserve the metric
(thus holonomies are isometries). For instance, identifying (x0 , x1 , x2 , x4 ) with (x0 + L, x1 , x2 , x3 ) changes the
topology from R4 to a cylinder S1 R3 and introduces
closed timelike lines. However, if causality is believed to
hold in the sense that no effect can preceed its cause, such
an identification is prohibited, and the study of spacetime topology is restricted so as to exclude closed timelike curves [15]. This is achieved if spacetime can be

Name

cylinder

FD and
identifications
a
b

a
b

Mbius
strip

torus

a
b

Shape

if (0) = (1) = x0 , where x0 is a point of . Two

loops at x0 , and , are said to be homotopic ( )
if they can be continuously deformed into one another,
i.e. if there exists a continuous map (called homotopy)
F : [0, 1[[0, 1[ X such that

Closed Orientable

NO

YES

NO

NO

u [0, 1[,

v [0, 1[,

c d
a
b

YES

YES

a
b

YES

NO

a
b
d

F (0, v) = F (1, v) = x0 .

The equivalence class of homotopic loops is denoted by

[]. We denote i the loop corresponding to the projection of the trajectory of twin i in .
In our example (see figure 2), the twins 1 and 2 have
homotopic trajectories: since 2 does not go around the
universe, the loop 2 can be continuously contracted into
the null loop 1 = {0}, so

c d

Klein
bottle

F (s, 0) = (u), F (u, 1) = (u)

FIG. 1. The four multiconnected topologies of the two dimensional Euclidean plane. They are constructed from a parallelogram or an infinite band (fundamental domain FD), by
identification of edges according to the allowable holonomies.
We indicate as well their compactness and orientability properties (from [14]).

1 2 {0}.
However, among these two homotopic loops, only one corresponds to an inertial observer going from O to P : that
of twin 1, which is thus older than twin 2, as expected in
the standard paradox.
Now, the twins 3 and 4 respectively go once around the
hole and around the handle of the torus. From a topological point of view, their paths can be characterized by
a socalled winding index. In a cylinder, the winding index is just an integer which counts the number of times a
loop rolls around the surface. In the case of a 2torus, the
winding index is a couple (m, n) of integers where m and
n respectively count the numbers of times the loop goes
around the hole and the handle. In our example, twins 1
and 2 have the same winding index (0,0), whereas twins
3 and 4 have winding indexes respectively equal to (1, 0)
and (0, 1). The winding index is a topological invariant
for each traveller: neither change of coordinates or reference frame (which ought to be continuous) can change
its value.
To summarize, we have the two situations:

Since a characteristics of multiconnected spaces is to

provide several geodesics between two any points, we
shall extend the concept of twins to more than two
and consider an ensemble of twins (strictly speaking,
initially synchronized clocks) labeled as 1,2,3 and 4 (see
figure 2 [left]). The twin 1 stays at home at point O and
her worldline can be identified with the time axis. The
twin 2 leaves O at t = 0, travels away and then turns back
to meet twin 1 in P . The twins 3 and 4 leave O in two
different directions along nonaccelerated worldlines and
travel away from 1; they respectively reach P 0 and P 00 ,
where they meet twin 1 . Due to warped space, twins
3 and 4 managed to come back without changing their
direction and inertial frame. Now, one wants to compare
the ages within the various pairs of twins when they meet
again. Whereas twin 2 undergone the standard paradox,
there seems to be a real paradox with twins 1, 3 and 4,
who all followed strictly inertial frames.
In [13] it was shown that, in the case of a cylinder, the
sedentary twin 1 is always older than any traveller because their states of motions, although nonaccelerated,
are not symmetrical. Which kind of asymmetry is to be
considered? As we emphasize below, acceleration is a
local concept, and the only explanation lies in a global
breakdown of symmetry due to a nontrivial topology.
Let us consider the projection of the worldlines onto a
constant time hypersurface , assumed to be a 2D flat
torus (see figure 2 [right]). Each projection is a loop
(u, x0 ) at x0 which can be parametrised by u [0, 1[

1. Two twins have the same winding index (twins 1

and 2 in our example), because their loops belong
to the same homotopy class. Nevertheless only one
(twin 1) can go from the first meeting point to the
second one without changing inertial frame. The
situation is not symmetrical about 1 and 2 due to
local acceleration, and 1 is older than 2.
2. Several twins (1, 3 and 4 in our example) can go
from the first meeting point to a second one at a
constant speed, but travel along paths with different winding indexes. Their situations are not
symmetrical because their loops belong to different
homotopy classes:
1 6 3 6 4 .

Under very special circumstances, twins 3 and 4 could also

meet altogether with twins 1 and 2 at the same point of
spacetime!

Twin 1 is older than twins 3 and 4 because her path

has a zero winding index.
3

so? In special relativity theory, two reference frames are

equivalent if there is a Lorentz transformation from one
frame of space-time coordinates to another system. The
set of all Lorentz transformations is called the Poincare
group a ten dimensional group which combines translations and homogeneous Lorentz transformations called
boosts.
The non equivalence between the inertial frames is
due to the fact that space topology breaks globally the
Poincare group. Indeed, cutting and pasting to compactify the spatial sections defines (i) particular directions, so
that space, even if locally isotropic, is no more invariant
under rotations, and (ii) a particular time: the one measured by an observer whose 4velocity is perpendicular
to (the twin 1 in our example). Let us call t the proper
time of such an observer; then the spacetime coordinates
p of any point P M can be decomposed in the inertial frame K as p = (t, x), and the choice of a topology
reduces to the choice of the identifications

P"
O"

time

O'

P'

1
P

3
2

space

spatial section

FIG. 2. Different twins in a (1+2) spacetime with toroidal

spatial sections. All of them leave 0 at the same time. While
1 remains at home, 2 [dash-dotted line] goes away and then
comes back to meet 1 in P (it corresponds to the standard
case), 3 [solid line] goes around the universe in a given direction from O to P 0 O0 and 4 [dashed line] goes around the
universe along another direction from O to P 00 O00 . On the
left plot, we used the fundamental domain representation (in
which the 2torus corresponds to a rectangle whose edges /
and // are respectively identified; see figure 1). On the right
plot, we depict the projection of their trajectory on a constant
time slice. The spacetime points O, O0 , O00 , P , P 0 and P 00 of
M are projected onto the same base point 1 in .

p = (t, x) (t, gx) = g(p),

g .

(6)

In the inertial frame K0 of the traveller, the coordinates

of P are given by a Lorentz transformation L(p) = (t0 , x0 )
with
t0 = (t v.x) ,

x0 = (x vt)

(7)

where (1 v 2 )1/2 . It is clear from (6) and (7) that

In order to solve exhaustively the twin paradox in a

multiconnected space, one would like not only to compare separately the ages of the travellers with the age of
the sedentary twin, but also to compare the ages of the
various travellers when they respectively meet each other.
It is clear that the only knowledge of the winding index
of their loops does not allow us, in general, to compare
their various proper time lapses. The only exception is
that of the cylinder, where a larger winding number always corresponds to a shorter proper time lapse. But
for a torus of unequal lengths, for instance when the diameter of the hole is much larger than the diameter of
the handle, a traveller may go straight around the handle
many times with a winding index (0, n), and yet be older
than the traveller which goes straight only once around
the hole with a winding index (1,0). The situation is still
more striking with a double torus, indeed a hyperbolic
surface instead of an Euclidean one (see e.g. [16]). The
winding indexes become quadruplets of integers and, as
for the simple torus, they cannot be compared to answer
the question on the ages of the travellers. As we shall
see below, this problem can be solved only by using an
additional metric information.

g 6 Id , p,

g L(p) 6= L g(p).

(8)

The identification (6) particularises a given foliation and

spatial sections, leading to the existence of an absolute
rest frame (the one of zero homotopy class). For any
other observers, these identifications are relations between events at different times (and thus in different spatial sections) and not a relation between points in a given
spatial section. As pointed in [12], the observers 3 and
4 will find that their constant time hypersurfaces do not
match in the universal covering space and that there are
points on these surfaces of simultaneity which are connected by timelike curves. Moreover, the only holonomy
g such that g L(p) = L g(p) for all p is the identity
g = Id , thus the holonomy group reduces to = {Id } and
reduces to X. We deduce that the only topology compatible with the Poincare group is the trivial topology.
In other words, the only flat spaceform invariant under
the full Poincare group is the original simplyconnected
Minkowski spacetime, and any additional discrete identification group is incompatible with the Lorentz transformations.
In conclusion, the oldest twin will always be the one
of homotopy class {0}, and between two twins of same
homotopy class, the oldest one will be the one who does
not undergo any acceleration. We can rephrase Bondis
formulation of the solution by saying that there is only
one way of a given homotopy class of getting from the
first meeting point to the second without acceleration.

IV. LANGEVIN AND POINCARE

We have found a topological invariant attached to each

twins worldline which accounts for the asymmetry between their various inertial reference frames. Why is it
4

the FriedmannLematre solutions are assumed to be

globally homogeneous and isotropic. From a geometrical
point of view, this means that spacelike slices have constant curvature and that space is spherically symmetric about each point. In the language of group theory,
the spacetime is invariant under a six-dimensional isometry group. The universal covering spaces of constant
curvature are R3 , S3 or H3 according to the zero, positive or negative value of the curvature. Any identification of points in these simply-connected spaces via a
holonomy group lowers the dimension of their isometry
group ; it preserves the threedimensional homogeneity
group (spacelike slices have still constant curvature), but
it breaks down globally the isotropy group (at a given
point there are a discrete set of preferred directions along
which the universe does not look the same).
Thus in FriedmannLematre universes, (i) the expansion of the universe and (ii) the existence of a nontrivial
topology for the constant time hypersurfaces both break
the Poincare invariance and single out the same privileged inertial observer who will age more quickly than
any other twin: the one comoving with the cosmic fluid
although aging more quickly than all her travelling sisters
may be not a real privilege!

This generalises the previous works [1113] by adding

topological considerations which are more general and
hold whatever the shape of space is. As concluded in
[12], it is not sufficient that [the] motion [of the twin]
is symmetrical in terms of acceleration felt and so on;
it must also be symmetrical in terms of the way that
their worldlines are embedded into the spacetime; this
latter symmetry is the one we have exhibited and which
is encoded in the homotopy class.
As mentioned in the previous section, the homotopy
classes only tell us which twin is aging the fastest: the
one who follows a straight loop homotopic to {0}. It does
not provide a classification of the ages (i.e. proper time
lengths) along all the straight loops. To do this, some
additional information is necessary, such as the various
identification lengths. Indeed there exists a simple criterion which works in all cases: a shorter spatial length
in the universal covering space will always correspond to
a longest proper time. To fully solve the question, it is
therefore sufficient to draw the universal covering space
as tessellated by the fundamental domains, and to measure the lengths of the various straight paths joining the
twin 1 position in the fundamental domain to its ghost
positions in the adjacent domains (see figure 3). As usual
in topology, all reasonings involving metrical measurements can be solved in the simplyconnected universal
covering space.

[1] A. Einstein, Zur Elektrodynamik bewegter K

orper,
Ann. Phys. (Leipzig) 17, 891 (1905).
[2] P. Langevin, Levolution de lespace et du temps, Scientia X, 31 (1911).
[3] R.H. Romer, Twin paradox in special relativity, Am.
J. Phys. 27, 131135 (1959).
[4] E. F. Taylor and J.A. Wheeler, Spacetime Physics, Second Edition, Freeman, 1992.
[5] W.G. Unruh, Parallax distance, time, and the twin
paradox, Am. J. Phys. 49 (6), 589592 (1981).
[6] H. Nikolic, Relativistic contraction of an accelerated
rod, Am. J. Phys. 67 (11), 10071012 (1999).
[7] R.H. Good, Uniformly accelerated reference frame and
twin paradox, Am. J. Phys. 50, 232238 (1982).
[8] S.P. Boughn, The case of the identically accelerated
twins, Am. J. Phys. 57 (9), 791793 (1989).
[9] R. Perrin, Twin paradox: a complete treatment from
the point of view of each twin, Am. J. Phys. 44, 317
319 (1970).
[10] J.C. Hafele and R.E. Keating, Around-the-world atomic
clocks: Observed relativistic time gains, Science, 177,
168-170 (1972).
[11] C.H. Brans and D.R. Stewart, Unaccelerated returning

O 2

FIG. 3. Straight paths in the universal covering space of a

(2+1)spacetime with flat, toruslike spatial sections. Paths
2, 3, 4, 5 are straight loops with respective winding indexes
(0,1), (1,0, (1,1), (1,2), allowing the traveller twins to leave
and meet again the sedentary twin O without accelerating.
The inertial worldlines are clearly not equivalent: the longest
the spatial length in the universal covering space, the shortest
the proper time travel in spacetime.

In the framework of general relativity, general solutions of Einsteins field equations are curved spacetimes
admitting no particular symmetry. However, all the exact known solutions admit symmetry groups (although
less rich than the Poincare group). For instance, the
usual big bang cosmological models described by

There is one exception: the projective space, obtained by

identifying the antipodal points of S3

[12]
[13]
[14]
[15]

[16]
[17]

twin paradox in flat spacetime, Phys. Rev. D8, 1662

1666 (1973).
R.J. Low, An acceleration free version of the clock paradox, Eur. J. Phys. 11, 2527 (1990).
T. Dray, The twin paradox revisited, Am. J. Phys. 58
(9), 822-825 (1990).
H. Bondi, Relativity and common sense, Doubleday and
Company Inc., N.Y. (1964).
R. Geroch, Spacetime structure from a global viewpoint, in General relativity and cosmology, proceedings
of course XLVII of the international course of physics
Enrico Fermi, Ed. R.K. Sachs, Academic Press (New
York, 1971).
M. Lachi`eze-Rey, J.-P. Luminet, Cosmic topology,
Phys. Rep. 254, 135 (1995).
W.P. Thurston and J. Weeks, The Mathematics of
Three-Dimensional Manifolds, Sci. Am., 251, 108113
(July 1984); J-P. Luminet, G.D. Starkman and J.R.
Weeks, Is space finite?, Sci. Am., 280, 9097 (april
1999); J-P. Uzan, R. Lehoucq and J-P. Luminet, New
developments in the search for the topology of the universe, Proc. of the XIXth Texas meeting, Paris 1418
december 1998, Eds. E. Aubourg, T. Montmerle, J. Paul
and P. Peter, article no 04/25.