Thermal time and the Tolman-Ehrenfest effect:

temperature as the speed of time

Carlo Rovelli, Matteo Smerlak1

arXiv:1005.2985v3 [gr-qc] 25 May 2010

Centre de Physique Theorique de Luminy , Case 907, F-13288 Marseille, EU

(Dated: May 26, 2010)

The thermal time hypothesis has been introduced as a possible basis for a fully general-relativistic
thermodynamics. Here we use the notion of thermal time to study thermal equilibrium on stationary
spacetimes. Notably, we show that the Tolman-Ehrenfest effect (the variation of temperature in

space so that T g00 remains constant) can be reappraised as a manifestation of this fact: at thermal
equilibrium, temperature is locally the rate of flow of thermal time with respect to proper time
pictorially, the speed of (thermal) time. Our derivation of the Tolman-Ehrenfest effect makes no
reference to the physical mechanisms underlying thermalization, thus illustrating the import of the
notion of thermal time.
I.

INTRODUCTION

The thermal time hypothesis (TTH) was proposed

by Connes and one of the present authors [1, 2] as a basis
for a fully general-relativistic (and quantum) thermodynamics a problem which is still open [25]. In a nutshell,
the TTH proposes that the most characteristic features of
the flow of time are essentially thermodynamical, and
can emerge statistically also in a quantum gravitational
context where other notions of time are meaningless [6].
To understand this hypothesis, it is useful to distinguish between two prevalent notions of time: mechanical,
and thermal time. Mechanical time is the time of dynamics, measured by standard clocks, and determined by the
spacetime metric. In non-relativistic physics it is represented by the Newtonian time t, in special-relativistic
physics by the Minkowski coordinate x0 and on a general
spacetime by the proper time s along timelike geodesics.
Thermal time, on the other hand, is the quantity characterizing the unfolding of irreversible phenomena, arising
as a byproduct of their incomplete description. It is e.g.
the variable in the heat equation T = T , where T
is the temperature, or in the second law of the thermodynamics S > 0, where S is the entropy.
The two concepts essentially match in Newtonian
physics,1 and we will later illustrate how this comes
about. But, significantly, this is not so in more general situations. In particular, the notion of thermal time
retains its validity also in a context where a classical
background metric is not available. The TTH, indeed,
is based on the observation that any statistical state on
an algebra of observables naturally induces a flow on it;

Unit
e mixte de recherche (UMR 6207) du CNRS et des Universit
es
de Provence (Aix-Marseille I), de la M
editerran
ee (Aix-Marseille
II) et du Sud (Toulon-Var); laboratoire affili
e`
a la FRUMAM (FR
2291).
1 Although Loschmidts paradox the difficulty to trace back irreversible processes to a time-symmetric dynamics may be seen
as already pointing to a distinction, and perhaps a tension, between two.

it postulates that the thermal time governing the thermodynamics of a macroscopic system described by a given
statistical state is this flow [1, 2].
So far the TTH has remained a rather abstract idea,
with few concrete physical applications [711]. The
doubt is legitimate whether its definition of thermal time
has any physical content at all, or it is just an empty
definition. In this note, we show that the TTH has substantial physical content. We do so by considering thermal equilibrium on stationary spacetimes an intermediate situation between non-relativistic mechanics and
full-fledged general relativity.
It was shown by Tolman and Ehrenfest in 1930 [12, 13]
that, in the presence of gravity, temperature is not constant in space at equilibrium. In a stationary spacetime,
using stationary coordinates (~x, t) such that t g = 0,
one has instead that

T g00 = const,

(1)

where T the temperature. In the Newtonian limit, this

corresponds to a temperature gradient
~g
T
= 2,
T
c

(2)

where ~g is the Galilean acceleration of gravity. In other

words: a vertical column of fluid at equilibrium is hotter
at the bottom. Of course, this is a slight 1/c2 relativistic
effect, negligible in most practical situations;2 from a theoretical persepective, however, it is very significant, for
it bridges the gap between thermodynamics and general
relativity.
The effect can be derived in a number of ways and
invoking different mechanisms underlying thermalization
[1222]. (We recall a simple derivation in Appendix A,
which uses the fact that energy falls). Here we derive it without any assumption about such dynamical
mechanisms, solely from the characterization of thermal

On the surface of the Earth,

T
T

= 1018 cm1 .

2
equilibrium in terms of thermal time. Besides shedding
some new light on the Tolman-Ehrenfest effect itself, we
believe this result illustrates the import and effectiveness
of this notion of thermal time.
This note is organized as follows. The notion of thermal time is presented in section II; its incarnation in the
context of stationary spacetimes, and the derivation of
the Tolman-Ehrenfest law, are discussed in section III;
concluding remarks are given in section IV.
II.

THE THERMAL TIME HYPOTHESIS

Let us start by recalling the mathematics of the TTH,

in the simplified setting of classical Hamiltonian mechanics. (The full quantum version of the TTH is recalled for
completeness in Appendix B.) A general relativistic statistical system can be described by a Poisson algebra A
of observables A on a phase space S. Statistical states
are normalized3 positive functions on S, interpreted as
probability densities on S.
Given a statistical state , we define the thermal time
flow : A A as the Poisson flow of ( ln ) in A.
That is
d (A)
= {A, ln }.
d

(3)

where the r.h.s. is the Poisson bracket. The TTH then

states heuristically that the physical time flow that governs thermodynamical processes in, or near, the state
described by is .
Let us apply this definition to the Bolzmann-Gibbs
state T describing the state in thermal equilibrium at
temperature T (in the canonical ensemble). If H is the
energy,
H

T = Z 1 e kT
where k is the Botzmann constant and Z =
The thermal time flow of T is

(4)
R

ds e kT .

1
1 dA
d T (A)
= {A, ln T } =
{A, H} =
d
kT
kT dt

(5)

It follows that the thermal time of an equilibrium state

at temperature T and Newtonian time t are related by
d
1 d
=
.
d
kT dt

(6)

From this, we observe that a thermal equilibrium state

has the two following properties:
d
is a symmetry of the (Galilean) space(i) The flow d
time. More precisely, it generates a one-parameter
group of timelike isometries of spacetime.

(ii) At every point in space, the ratio between the flow

of thermal time and the flow of mechanical time
t is the temperature kT .
In fact, these two properties can be taken as a characterization of thermal equilibrium. A thermal equilibrium
state is a state whose thermal time generates a timelike one-parameter group of symmetries of spacetime; its
temperature is the ratio between the flow of thermal time
and the flow of mechanical time. This characterization
is equivalent to the Boltzmann-Gibbs ansatz (4).
A crucial fact pointed out in [7, 11] is that while (i) is
essentially a global property, (ii) can be interpreted locally: at any given spacetime point, temperature is given
by the ratio between the two flows at that point. In the
next section, we use this observation to generalize this
characterization to stationary spacetimes and to derive
the Tolman-Ehrenfest law (1).
III.

Thermal equilibrium states are states towards which

irreversible processes converge. In standard statistical
mechanics, they can be characterized in a number of
ways. Stochastically, by condition of detailed balance of
microscopic probability fluxes; dynamically, by a condition of stability under small perturbations of the Hamiltonian; thermodynamically, by a condition of passivity4 ; information-theoretically, by the maximization of
entropy; quantum mechanically, by the periodicity of correlation functions in imaginary time, aka the KMS condition; and so on.
When moving to curved spacetimes, these characterizations tend to become problematic, because of the effect
of gravity. In the case of stationary spacetimes, several
generalizations of the condition of thermal equilibrium
have been studied. The first was proposed in the original work of Tolman and Ehrenfest [13], using the idea of
a radiation thermometer electromagnetic radiation
whose pressure p measures the local temperature T via
the Stefan-Boltzmann relation
p=

In the sense that

ds (s) = 1.

4 4
T ,
3c

(7)

with the Stefan-Boltzmann constant. Ebert and

Goebel later introduced the notion of relativistic Carnot
cycles, and characterized thermal equilibrium by the
condition of passivity, i.e. vanishing Carnot efficiency
[17]. Other arguments have been proposed as well [14
16, 18, 19, 21]. These various characterizations of thermal equilibrium all lead to the conclusion that temperature is not constant at equilibrium in general; rather, the

THE TTH IN STATIONARY SPACETIMES

Passivity refers to the impossibility to extract work from cyclic

processes in a system at thermal equilibrium Kelvins formulation of the second law.

3
Tolman-Ehrenfest law

IV.

T kk = constant

(8)

holds. Here is a timelike Killing vector. In stationary

coordinates, = t hence kk = g00 , and we have (1).

Let us now show how this result can be obtained using
the notion of thermal time.
Consider a macroscopic system, say a gas, in a stationary spacetime. The phase space S of such a system can
be thought as the set of solutions ~xn (t) of the equations
of motion of all the particules of the gas. Observables are
functions of these trajectories. Among them are the local
observbles Ax , which depend on the positions and momenta of the particles in a neighborhood of the spacetime
point x = (~x, t). For instance, the macroscopic density
n(x) can be defined by
Z
1 X
mn
d3 ~x (~xn (t), ~x),
(9)
n(~x, t) =
V n
(~
x)
where (~x) is a small region of volume V centered on x
and mn is the mass of the n-th particle.
Let be a statistical state and its thermal time
flow. Under which conditions is at equilibrium, and
what is its temperature? It is immediate to see that the
characterization of thermal equilibrium (i) and (ii) of the
previous section are still meaningful in this context. The
first reads
(i) The thermal time flow has a geometrical action
on the local observables
d (A)
(10)
= L Ax .
d
where the Lie derivative L acts on Ax seen as
a function on spacetime, and the vector field
generates a timelike symmetry of spacetime, that is,
is a timelike Killing field for the stationary metric.
The second, which gives the temperature, now reads
(ii) At every space point ~x, temperature is the ratio of
the thermal time flow to the mechanical time flow
is. By (10), the first is . At each ~x, namely
along each stationary timelike curve, mechanical
time as measured by stationary standard clock
is proper time s. Hence T is determined by
=

1 d
.
kT ds

CONCLUSION

(11)

Now, the key observation is that at each point a timelike Killing field is proportional to d/ds along a stationary
timelike curve, but in general the proportionality constant is not constant in space. By taking the norm of the
last equation, we have indeed
1
k k =
.
(12)
kT
which is the Tolman-Ehrenfest law (8). Thus, the influence of gravity on thermal equilibrium can be read out
straightforwardly from the characterization of the latter
in terms of thermal time.

We have shown that the non-relativistic characterization of thermal equilibrium in terms of thermal time, and
the observation that temperature is the ratio between the
mechanical and the thermal time flows, can be directly
generalized to stationary spacetimes; from this observation we have derived the Tolman-Ehrenfest effect.
Our derivation relies only on the notion of thermal
time, and makes reference to any special thermalization
mechanism (as we do for instance in Appendix A, where
we use E = mc2 and the fact that energy falls in a
gravitational field). This derivation confirms the idea
that Tolman-Ehrenfest law is a genuine thermodynamical
relation, independent on the dynamical processes underlying it. More importantly to us, it shows that thermal
time is not an empty definition: a genuine physical effect
can be derived from it.
The core of the TTH hypothesis comes into play in the
identification of the temperature as the ratio between the
local proper time and thermal time flows. This identification appears unimportant in flat space, were it amounts
to fixing an arbitrary unobservable scale for the thermal time, but it becomes heavily consequential in a more
general context, where the norm of the Killing field, and
hence the thermal time flow, varies from point to point.
An analogy is Newtons main law F = ma. At first
sight, it is just an empty definition of force, or worse,
force and mass. But by applying this law to, say, two
distinct masses mi each one subjected to two forces Fj
(say two springs), the law becomes predictive: from
Fj = mi aij , it follows a22 a12 = a21 a11 . Therefore we
can predict one acceleration from the other three. Similarly, we have shown that Tolman-Ehrenfest law, which
is predictive, follows from applying the defining relation
(11) at different points in space, and taking ratios. The
norm of the killing field is not determined by the metric,
but the ratio of its norms at different points is.
Finally, we remark the Tolman-Ehrenfest effect instantiates the idea that temperature can be thought as the
speed of time.5 More precisely: the speed of thermal time
with respect to proper time. From this perspective, indeed, theTolman-Ehrenfest law is simply the observation
that the stronger the gravitational potential, the slower
the proper time (with respect to the flow of the global
symmetry group), and hence the higher the temperature.
Whether this intuition temperature as the speed of time
has further heuristic power, or is peculiar to thermal
equilibrium in stationary spacetimes, we cannot tell yet.

It is amusing in this respect to note that, in biology, the expression thermal time is sometimes used to refer to the widely
observed linear relationship between development rate and temperature [23].

4
Appendix A: A simple derivation of the
Tolman-Ehrenfest effect

Appendix B: Thermal time hypothesis for general

relativistic quantum systems

Several intuitive arguments can be used to make physical sense of the Tolman-Ehrenfest effect. Here is one,
which makes use of E = mc2 and the equivalence of inertial and gravitational mass. Equilibrium between two
systems happens when the total entropy is maximized

A general covariant quantum system can be described

by an algebra A of observables A (which we take to be a
C algebra) and a space S of pure states s on A. For instance, in quantum gravity the states can be given by the
solutions of the Wheeler-DeWitt equation, and observables by self-adjoint operators on a Hilbert space defined
by these solutions (see for instance [6]); a state in this
Hilbert space defines a pure state s on A, as the positive
linear function s (A) = h|A|i.
A general covariant quantum statistical system is described by the algebra A of observables and a space S
of states which are general positive linear functionals on A. These are interpreted as statistical superpositions
P of pure states. That is, they can have the form
= n n sn where n are interpreted as probabilities.
Given a state , the thermal time flow is the Tomita
flow of the state in A [1]. It satisfies in particular

dS = dS1 + dS2 = 0.

(A1)

If a heat quantity dE1 leaves the first system, and the

same quantity dE2 = dE1 enters the second system, then
dS1 /dE1 dS2 /dE2 = 0.

(A2)

Since T := dS/dE, we obtain T2 = T1 .

However, if the two systems are at different gravitational potentials (in the Newtonian limit), the amount of
energy dE1 leaving, say, the upper one, is not the amount
of energy dE2 entering the lower one. Indeed, E = mc2
and the equality of inertial and gravitational mass imply that any form of energy has a gravitational mass,
and falls. Hence dE2 is dE1 increased by the potential
energy m, where is the gravitational potential:
dE2 = dE1 (1 + /c2 ),

(A3)

which yields immediately (2).

[1] A. Connes and C. Rovelli, Von Neumann algebra

automorphisms and time thermodynamics relation in
general covariant quantum theories, Class. Quant.
Grav. 11 (1994) 28992918,
[2] C. Rovelli, Statistical mechanics of gravity and the
thermodynamical origin of time, Class. Quant. Grav.
10 (1993) 15491566.
[3] M. Montesinos and C. Rovelli, Statistical mechanics of
generally covariant quantum theories: A Boltzmann-like
approach, Class. Quant. Grav. 18 (2001) 555569,
[4] C. Rovelli and F. Vidotto, Single particle in quantum
gravity and BGS entropy of a spin network, Phys. Rev.
D81 (2010) 044038,
[5] L. Smolin, On the intrinsic entropy of the gravitational
field, Gen. Rel. Grav. 17 (1985) 417.
[6] C. Rovelli, Quantum Gravity. Cambridge University
Press, Cambridge, UK, 2004.
[7] P. Martinetti and C. Rovelli, Diamondss temperature:
Unruh effect for bounded trajectories and thermal time
hypothesis, Class. Quant. Grav. 20 (2003) 49194932,
[8] P. Martinetti, A brief remark on Unruh effect and
causality, J. Phys. Conf. Ser. 68 (2007) 012027,
[9] C. Rovelli, The statistical state of the universe, Class.
Quant. Grav. 10 (1993) 1567.
[10] Y. Tian, De Sitter Thermodynamics from Diamondss
Temperature, JHEP 06 (2005) 045.
[11] P. Martinetti, Conformal mapping of Unruh
temperature, Mod. Phys. Lett. A24 (2009) 14731483,
[12] R. C. Tolman, On the Weight of Heat and Thermal

( A) = (A).

(B1)

This flow depends on the state, but the flows generated

by different states are equivalent up to inner automorphisms in A [24, 25]. Therefore we can further define an
outer thermal time flow , as the flow up to inner
automorphisms. Remarkably, this is state independent.

[13]

[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]

Equilibrium in General Relativity,

Phys. Rev. 35 (1930) 904924.
R. C. Tolman and P. Ehrenfest, Temperature
Equilibrium in a Static Gravitational Field, Phys. Rev.
36 (1930) no. 12, 17911798.
N. L. Balazs Astrophys. J. 128 (1958) 398.
N. Balazs and M. Dawson Physica 31 (1965) 222.
H. Buchdahl, The Concepts of Classical
Thermodynamics. Cambridge University Press, 1966.
R. Ebert and R. G
obel, Carnot cycles in general
relativity, Gen. Rel. and Grav. 4 (1973) 375386.
J. Ehlers ANt. Math.-Nat. Kl. Akad. Wiss. Mainz 11
(1961) 804.
L. Landau and E. M. Lifshitz, Statistical Physics.
Pergamon, Oxford, 1959.
J. Stachel, The Dynamical Equations of Black-Body
Radiation, Foundations of Physics 14 (1973) 1163.
G. E. Tauber and J. W. Weinberg Phys. Rev. 122
(1961) 1342.
R. C. Tolman, Relativity, Thermodynamics and
Cosmology. Oxford University Press, London, 1934.
R. Bonhomme, Bases and limits to using degree day
units, Eur. J. Agronomy 13 (2000) 110.
A. Connes, Une classification des facteurs de type III,
Ann. Sci. Ecole Normale Superieure 6 (1973) 133252.
A. Connes, Noncommutative geometry. Academic Press,
San Diego, CA, 1994.