a

r
X
i
v
:
a
s
t
r
o
-
p
h
/
9
2
0
6
0
0
5
v
1


2
4

J
u
n

1
9
9
2
BROWN-HET-859
June 1992
Entropy of a Classical Stochastic Field
and Cosmological Perturbations
R. Brandenberger
1), 3)
, V. Mukhanov
2),

and T. Prokopec
1), 3)
1) Physics Department, Brown University
Providence, RI 02912, USA
2) Institute of Theoretical Physics, ETH Honggerberg,
CH 8093 Z urich, Switzerland
3)Institute for Theoretical Physics, University of California
Santa Barbara, CA 93106, USA
PACS numbers: 04.20.CV, 05.20.Gg , 98.80.Dr
ABSTRACT
We propose a general denition of nonequilibrium entropy of a classical stochastic eld.
As an example of particular interest in cosmology we apply this denition to compute the
entropy of density perturbations in an inationary Universe. On the scales of structures in
the Universe, the entropy of density perturbations dominates over the statistical uctuations
of the entropy of cosmic microwave photons, indicating the relevance of the entropy of density
uctuations for structure formation.
on leave of absence from Institute for Nuclear Research, Academy of Sciences, 117312 Moscow, Russia
1. Introduction
The concept of entropy contains relevant information about a dynamical system.
[1]
In
systems with a nite number of degrees of freedom there is a natural way to dene entropy,
even if the system is out of thermal equilibrium. We are interested in systems with innitely
many degrees of freedom and which can be described by a stochastic Gaussian eld. In this
letter, we propose a general denition of nonequilibrium entropy of a stochastic Gaussian
eld.
An issue of considerable interest is to develop a consistent denition of entropy in general
relativity and cosmology. There have been some key results in this area. The observation that
all information about a particle crossing the Schwarzschild horizon is lost, led Bekenstein
[2]
and Hawking
[3]
to their famous formula for the entropy of a black hole. Penrose
[4]
suggested
that it may be possible to realize the second law of thermodynamics in cosmology by assigning
an entropy to the gravitational eld itself. He conjectured that the plausible denition of
entropy might be some integral of the Weyl tensor squared, and that the Universe starts in
a state of minimal gravitational entropy. In this picture structure formation and the second
law of thermodynamics are reconciled, since gravitational clustering leads to an increase in
the Weyl tensor, thus generating gravitational entropy. The connection between information
loss and entropy of the gravitational eld was explored in many papers.
[5]
In this letter we use our denition of nonequilibrium entropy of a stochastic Gaussian
eld to propose a new approach to the problem of gravitational entropy in cosmology. The
formalism is based on separating the entire system of gravitational plus matter elds into
background elds (chosen to have high space-time symmetries) and linearized uctuating
elds; the latter are the stochastic elds we apply our general denition of entropy to.
It has been demonstrated that the dynamics of perturbations can be reduced to the
dynamics of a single scalar eld (which comprises in a self-consistent manner both scalar
and/or tensor gravitational eld perturbations and matter eld uctuations) in the classical
space-time background (for a recent review see Ref. [6] ). The evolution of the background
eld is completely specied; this means it carries no entropy. On the other hand, the
uctuating eld carries signicant entropy. This statement needs justication.
In order to obtain growth of entropy, it is necessary to propose some kind of coarse grain-
ing in which some information is lost during the evolution. In this work, we consider a free
2
scalar eld in an expanding space-time background, in which there is abundant production
of perturbations by parametric amplication.
[7]
We assume that there is a mechanism which
generates stochasticity in the phases of the perturbations produced during the evolution.
This mechanism is eective for the eld modes within the horizon, and generates entropy in
the uctuating eld.
The letter is organized in the following way. The next section is devoted to the derivation
of the formula for the entropy of a stochastic Gaussian scalar eld. Using this formula, we
then calculate (in Sec. 3.) the entropy of cosmological density perturbations. In Sec. 4. we
hint to some additional possible applications of the formula in the context of cosmology.
2. Entropy of Classical Field
We wish to consider the entropy associated with a classical stochastic eld. For a given
real scalar eld and its canonical momentum , there is a probability distribution func-
tional P[, ] dened over an innite dimensional space spanned by functions {, }. The
probabilistic denition of entropy gives
S =
_
P[, ] ln P[, ] DD , (2.1)
where the probability functional P[, ] is normalized to unity. We assume a Gaussian
process, i.e. that the knowledge of two-point correlation functions suces to completely
specify the stochastic properties of the elds and . (Higher order correlation functions
can be then given in terms of two-point correlations.)
If the stochastic process is non-Gaussian, then the Gaussian approximation may still be
a good one, provided corrections due to higher order correlations are small. An additional
requirement is that correlations are of nite range. This is fullled in the cosmological set
up, because a natural cut-o for correlation is the horizon scale.
The Gaussian approximation breaks down when perturbations grow nonlinear, and ef-
fects of the nonlinearities (originating in the full theory) become signicant. In this case the
corrections arising in the higher order correlation functions become important and need to
be incorporated in the probability distribution.
3
The denition of entropy in Eq. (1) can be applied to cosmological perturbations in an
expanding universe (see Sec. 3.), when the evolution of perturbations can be well represented
by a Hamiltonian of second order in and . This means that the Gaussian character of
the probability distribution is preserved in the course of evolution.
We now sketch a derivation of the expression for entropy in terms of correlation functions.
Assume that at some time t the probability functional P[, ] has a general Gaussian form
P[, ] =
1
N
exp
_

_
1
2
_
(x)A(x, y)(y)
+(x)B(x, y)(y) + 2(x)C(x, y)(x, y)
_
d
3
xd
3
y
_
,
(2.2)
where N is a normalization constant, and A, B, and C are related to the two-point corre-
lation functions in a way yet to be determined. In a homogeneous space-time background,
A, B, and C are functions of (x y) only.
By a clever substitution, it is possible to bring Eq. (2.2) into a diagonal form, in which
and are replaced by new normal coordinates. It is then quite straightforward to evaluate
the normalization factor N of Eq. (2.2)
N =
_
det D(x y) , (2.3)
where D can be expressed in terms of correlation functions
D(x y) =
_
d
3
z
_
(x)(z) (z)(y) (x)(z)(z)(y)

. (2.4)
Using Eqs. (2.2) through (2.4), the expression (2.1) for the entropy gives
S = Tr (x y) + ln N . (2.5)
The rst term is an irrelevant constant. The relevant contribution comes from the second
term and can be rewritten as
S =
1
2
ln det D(x y) . (2.6)
The above equation is the main result of this section. Eqs. (2.4) and (2.6) can be used to
obtain the entropy of any stochastic classical scalar eld whose probability distribution can
be well approximated by the Gaussian probability distribution (2.2).
4
In order to calculate the determinant of D(x, y) , one needs to solve the eigenvalue prob-
lem associated with D. Under quite general conditions (assuming D is of nite support) and
using the -function regularization scheme, it is possible to show
[8]
that this determinant
can be expressed in terms of the spectral density D

k
, which is given by the Fourier transform
of D(x y)
D

k

_
d
3
z e
i

kz
D(z) = |

k
|
2
|

k
|
2
|

k
|
2
(2.7)
and it is positive denite. The entropy then reads
S = V
_
d
3
k
(2)
3
1
2
ln D

k
. (2.8)
The procedure to calculate the entropy of a classical Gaussian eld is now very simple.
Given two point correlation functions, one calculates D(xy) (Eq. (2.4)), Fourier transforms
it (Eq. (2.7)) and obtains the entropy according to Eq. (2.8). We now apply this prescription
to an example which is of interest in cosmology.
3. Entropy of Cosmological Perturbations
In this section we apply the method developed above to calculate the entropy of cos-
mological density perturbations. This is an example of relevance in cosmology, because it
is likely that the scalar density perturbations seed structures in the universe. We nd that
the entropy of scalar density perturbations on large scales in the universe is signicant when
compared to the statistical uctuations of the entropy of cosmic microwave photons on the
same scales.
Before we present any calculations, we give a short summary of the theory of density
perturbations. Density perturbations are scalar type metric perturbations which couple to
energy density and pressure. For matter which is in the form of a scalar eld, or an ideal gas,
it turns out that density perturbations can be described in a self-consistent manner in terms
of a single gauge invariant scalar eld , which is a linear combination of scalar eld matter
uctuations (ideal gas density uctuations) and longitudinal metric uctuations and whose
dynamics is given by a quadratic action. (For a comprehensive account of the gauge invariant
formalism of linear cosmological perturbation in Friedmann-Robertson-Walker backgrounds
see Ref. [6] , for pedagogical introduction see Ref. [9 ] .)
5
Hence, assuming that the gauge invariant eld is a stochastic Gaussian eld, we can
calculate the entropy associated with using Eqs. (2.4) and (2.6), or equivalently Eqs. (2.7)
and (2.8). In order to accomplish this, we need to know the two-point correlation functions
(x, t)(y, t), (x, t)(y, t), and (x, t)(y, t), where (x, t) =

t
(x, t).
As an example of a model in which the correlation functions exhibit nontrivial behavior,
we consider an expanding universe with initial quantum uctuations which evolve into clas-
sical ones as a result of evolution. In this case, particle pairs are produced via parametric
amplication, i.e., via coupling of matter elds to the nontrivial space-time background. Be-
cause of the generation of perturbations, the correlation functions become time dependent.
Here we consider the inationary universe scenario in which there is abundant production
of inhomogeneities.
The Hamiltonian governing the evolution of the single scalar eld and momentum
is quadratic in and , so that it is convenient to represent the evolution operator

U(t) in
a form in which the eects of free evolution (R) and interaction with the background (S)
are separated.

U(t) is the product of the rotation operator R and the two-mode squeeze
operator S

U(t) = R({

k
}) S({r

k
,

k
}) (3.1)
where
R({

k
}) =

k, kx>0
R(

k
)
S({r

k
,

k
}) =

k, kx>0
S(r

k
,

k
) .
(3.2)
The product

k, kx>0
is over half of the possible values of momenta

k (for deniteness,
say k
x
> 0). The rotation angles

k
=
_
t

k
(t

)dt

are given in terms of the frequency

k
of
mode

k; r

k
,

k
are the squeeze factor and phase, respectively, and can be expressed in terms
of parameters of the Hamiltonian (see Ref. [10] ). The two-mode squeeze operator S(r

k
,

k
)
acts on the vacuum |0
in
, creating pairs of particles with momenta

k and

k, so that the
total momentum of the pair is zero; hence S(r

k
,

k
) is a momentum conserving operator.
The operator S mathematically describes the process of parametric amplication.
Now we can express the two point correlation functions of quantum operators and in
terms of the parameters of the squeezed state which is obtained as a result of the evolution
6
of the initial vacuum state |0
in
of cosmological perturbations in an expanding Universe.
Simple, but rather lengthly calculation gives
[11]
0
in
| (x, t)(y, t) |0
in
=
_
d
3
k
(2)
3
e
i

k(xy)
1
2

k
(t)
_
(2 sinh
2
r

k
+ 1) sinh 2r

k
cos 2
_
_

k
dt

k
_
_
.
(3.3)
where the frequencies

k
and the squeeze factor r

k
depend on time because of the nontrivial
evolution of the background. Similar expressions are obtained for 0
in
| (x, t) (y, t) |0
in

and 0
in
| (x, t) (y, t) |0
in
.
Now we argue that the contribution due to the second (oscillating) term of Eq. (3.3)
can be neglected in the classical limit, when the number of produced particles per mode is
large. This can be justied if we are considering the transition from quantum uctuations
to classical ones and performing some coarse graining. The stochasticity which is induced in
the phases by coarse graining causes strong cancellation in the oscillating terms.
As a result we obtain for the spectral density of the operator D(x y)
D

k
= sinh
2
r

k
(1 + sinh
2
r

k
) . (3.4)
The entropy density per mode is then (Eq. (2.6))
s

k
=
S

k
V
=
1
2
ln D

k
(3.5)
and D

k
n
2

k
= sinh
2
r

k
(for n

k
1 ), where n

k
is the average number of particles in

k-mode,
whenever the notion of particles can be dened. Note that Eq. (3.5) can be applied even if
the notion of particle is not well dened. In particular this is the case for inhomogeneities in
a matter dominated Universe.
[6]
It is useful to dene the entropy density s

per logarithmic
wavelength 1/k interval
s

k
3
s

k

1

3
ln D

k
. (3.6)
To demonstrate how the technique developed above works, we now apply Eq. (3.5) to
estimate the entropy of cosmological perturbations produced during the inationary stage
7
of a model of chaotic ination.
[12]
The simplest potential for the inaton eld
I
is V (
I
) =
(1/2)m
2

2
I
, where m is the mass of the inaton, typically of the order 10
13
GeV. Considering
perturbations on scales which enter the horizon late in the radiation era, we obtain the
following result for the entropy density of perturbations on a typical scale
ph
a/k (see
Ref. [8 ] )
s

ph

1

3
ph
ln
_
(ml) ln

ph

ph
t
__

ph
l
_
_
, (3.7)
where l is the Planck length, t is the cosmological time and

is the typical wavelength of

the cosmic microwave background radiation. This formula is applicable for perturbations
which satisfy the condition t
ph

. The contribution to the total entropy of galactic

scale perturbations ( 1 100 Mpc) per corresponding galactic volume is S
gal

3
gal
s
gal

200 - 220 . The entropy of gravitational radiation can be estimated in the same manner (see
Ref. [8 ] ) .
Statistical uctuations of cosmic microwave photons are another potential source of
inhomogeneities. However, the entropy density of these uctuations scales as
3/2
compared
to the logarithmic dependence we found for the entropy density of density perturbations.
Hence, on scales of galaxies, the entropy of gravitational perturbations dominates over the
statistical uctuations of the entropy of cosmic microwave photons. The total entropy of
cosmological uctuations is, however, suppressed by a factor (H/m
pl
)
3/2
(where H is the
Hubble expansion rate at the end of ination) compared to that of the cosmic microwave
background. The dominance of the entropy of gravitational perturbations on large scales is
a sign of the relevance of this entropy for structure formation. A further application
[8]
of
this entropy is in the context of a collapsing Universe.
It is worth noting that because of the weak (logarithmic) dependence of entropy per
mode on the energy scale m of the model, our conclusion remains valid for a wide class of
inationary models.
8
4. Discussion
We derived a new formula for the entropy of a stochastic Gaussian scalar eld in terms
of two-point correlation functions. We then applied this result to the cosmologically relevant
example of density perturbations, using the formalism of linearized gauge invariant scalar
perturbations about a homogeneous classical space-time and matter background. We found
that the entropy of the system (scalar gravitational metric perturbations plus matter density
uctuations) grows as a logarithm of the number of particles created as the universe expands.
On the scales of large structures in the universe, the entropy of density perturbations in an
inationary Universe dominates over the entropy of statistical uctuations of the cosmic
microwave photons.
The formalism for calculation of entropy which we developed in this letter can be applied
to any (cosmological) problem, which can be reduced to the evolution of a classical stochastic
Gaussian eld.
The treatment of the full (nonlinear) gravitational eld is still an open problem, and
the corresponding formula for entropy is yet to be constructed. It would be very instructive
to demonstrate that the entropy of the gravitational eld continues to grow, even when the
perturbations become nonlinear.
Acknowledgements: For interesting discussions we are grateful to Andy Albrecht, Leonid
Grishchuk, Jim Hartle, Tony Houghton, Bei-Lok Hu and Henry Kandrup. Two of us (R. B.
and T. P. ) thank the ITP of the University of California in Santa Barbara for hospitality
during the completion of this work. V. M. thanks the Swiss National Science Foundation for
nancial support. At Brown, this work was supported by DOE grant DE-AC02-76ER03130
Task A and by an Alfred P. Sloan Foundation Fellowship to R. B. At the ITP in Santa
Barbara nancial support from NSF grant PHY89-04035 is acknowledged.
9
REFERENCES
1. See e.g. K. Huang, Statistical Mechanics (Wiley, New York, 1963);
R. Kubo, Statistical Mechanics (Elsevier, Amsterdam, 1965).
2. J. Bekenstein, Phys. Rev. D7, 2333 (1973).
3. S. Hawking, Comm. Math. Phys. 43, 199 (1975).
4. R. Penrose, in General Relativity : An Einstein Cetenary Survey , ed. by S. Hawking
and W. Israel (Cambridge University Press, Cambridge, 1979).
5. L. Smolin, Gen. Rel. Grav. 17, 417 (1985);
B. Hu and D. Pavon, Phys. Lett. 180 B, 329 (1986);
H. Kandrup, J. Math. Phys. 28, 1398 (1987);
B. Hu and H. Kandrup, Phys. Rev. D35 , 1776 (1987);
H. Kandrup, Phys. Rev. D37, 3505 (1988);
E. Calzetta and B. Hu, Phys. Rev. D37 , 2878 (1988).
6. V. Mukhanov, H. Feldman and R. Brandenberger, Theory of Cosmological Perturba-
tions , Phys. Rep. , in press (1992).
7. L. Parker, Phys. Rev. Lett. 21 , 562 (1968);
L. Parker, Phys. Rev. 183 , 1057 (1969);
L. Grishchuk, Zh. Eksp. Teor. Fiz. 67 , 825 (1974).
8. R. H. Brandenberger, V. Mukhanov and T. Prokopec, The Entropy of Gravitational
Perturbations, Brown University preprint, BROWN-HET-849 (1992).
9. R. Brandenberger, H. Feldman and V. Mukhanov, Gauge Invariant Cosmological
Perturbations , Brown preprint BROWN-HET-841 (1992) , to be published in the
Proceedings of ICGC-91 (Wiley Eastern Ltd. , New Delhi, 1992 ).
10. C. M. Caves and B. L. Schumaker, Phys. Rev. A31, 3068 and 3093 (1985).
11. T. Prokopec, Entropy of the Squeezed Vacuum, Brown preprint, BROWN-HET-861
(1992).
12. A. Linde, Phys. Lett. 129 B, 177 (1983) .
10