John Earman - The "Past Hypothesis" Not Even False

Studies in History and Philosophy of
Modern Physics 37 (2006) 399430

The Past Hypothesis: Not even false
John Earman
Department of History and Philosophy of Science, University of Pittsburgh, Pittsburgh, PA 15260, USA
Abstract
It has become something of a dogma in the philosophy of science that modern cosmology has
completed Boltzmanns program for explaining the statistical validity of the Second Law of
thermodynamics by providing the low entropy initial state needed to ground the asymmetry in
entropic behavior that underwrites our inference about the past. This dogma is challenged on several
grounds. In particular, it is argued that it is likely that the Boltzmann entropy of the initial state of
the universe is an ill-dened or severely hobbled concept. It is also argued that even if the entropy of
the initial state of the universe had a well-dened, low value, this would not sufce to explain why
thermodynamics works as well as it does for the kinds of systems we care about. Because the role of
Boltzmann entropy in our inferences to the past has been vastly overrated, the failure of the
Boltzmann program does not pose a serious problem for our knowledge of the past. But it does call a
different explanation of why thermodynamics works as well as it does. A suggestion is offered for a
different approach.
r 2006 Elsevier Ltd. All rights reserved.
Keywords: Entropy; Irreversibility; Cosmology; Boltzmann; Second Law of thermodynamics
1. Introduction
Ludwig Boltzmann bequeathed a set of concepts and techniques that lie at the core of
statistical mechanics, an essential part of modern physics. He also bequeathed a set of
foundational problems and puzzles that are the subjects of a debate among physicists and
philosophers that continues unabated after nearly a century following Boltzmanns death.
Despite the never-ending quality of the debate, a remarkable consensus has developed
around one of the key foundations issues: namely, the grounding of the relevant temporal
asymmetries in entropic behavior observed today is to be found in the fact (posit?) that the
ARTICLE IN PRESS
www.elsevier.com/locate/shpsb
1355-2198/$ - see front matter r 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.shpsb.2006.03.002
E-mail address: jearman+@pitt.edu.
early universe was in a low entropy state (the Past Hypothesis). To be sure, this idea
comes in a variety of forms and avors: for Albert (2000) the hypothesis of a low entropy
state for the early universe has something of a neo-Kantian status in that its truth is a
condition needed to make our knowledge of the past possible; for others (e.g. Penrose,
1979, 1989, 2004) the hypothesis has a (wannabe) law status; and for still others the
hypothesis rests on contingent facts that may or may not call for an explanation (see Price,
2004 vs. Callender, 2004).
1
Leaving these and other nuances aside, the agreement in the
philosophy of science community (where disagreement is the norm) is broad enough that
we have something that can be rightly dubbed a dogma.
This dogma, I contend, is ill-motivated and ill-dened, and its implementation consists
mainly in furious hand waving and wishful thinking. In short, it is (to borrow a phrase
from Pauli) not even false.
2
Such heresies are apt to get one burnt at the stake of academic
opinion, and if that is to be my fate, I want to make sure that I am burnt at the right stake.
So at the outset let me make clear that I am not denying what is a near truism; namely, if
some observed temporal asymmetry is not a consequence of laws of physics, then its origin
must be sought in initial/boundary conditions.
3
Nor am I denying that there is some sense
of entropy in which it is true that the early universe was in a low entropy state. What I
do deny is that, for the cosmologies described by classical general relativity theory, there is
any well-dened sense in which the Boltzmann entropy of the early universe has a very low
value. Furthermore, I claim that even if the Boltzmann entropy for the early universe were
a well-dened quantity, a low value would not sufce to explain the kinds of presently
observed entropic asymmetries we care about. And I question a long-standing tradition in
philosophy of science that accepts the idea that solving the conceptual puzzles Boltzmann
bequeathed us holds the key to understanding the fundamental ontological and
epistemological asymmetries between past and the future.
Roughly, the standard line runs that essential components of these asymmetries hinge on
the temporal asymmetry of records or traces, that the record/trace asymmetry is
grounded in entropic asymmetries, and that these entropic asymmetries are ultimately
traceable to the low entropy state of the early universe. None of the links in this chain, I
contend, can bear the weight that philosophers want to put on them when the said entropy
is Boltzmann entropy.
4
This radical sounding contention is in fact not radical at all. In
particular, it does not threaten skepticism about our knowledge of the past; indeed, once
we are freed of the obsessive insistence of seeing entropy as the key to ontological and
epistemological aspects of temporal asymmetries, it is evident that the ill-dened nature of
claims about entropy states of the early universe in no way compromises our knowledge of
the past. However, the heresy does entail the need for a new understanding of the
approximate validity of thermodynamics.
The plan of the paper is as follows. In Section 2, I sketch Boltzmanns explanation of the
statistical validity of the Second Law of thermodynamics and develop the apparatus that
will be needed for the later discussion of modern cosmology. Section 3 reviews some
ARTICLE IN PRESS
1
For other recent versions of the Past Hypothesis, see Bricmont (1996) and Goldstein (2001).
2
While I decry Paulis haughty dismissiveness, his phrase seems to me to exactly t the present case.
3
As Boltzmann (1904, pp. 170171) himself puts it, while afrming the antecedent: Since in the differential
equations of mechanics themselves there is absolutely nothing analogous to the second law of thermodynamics,
the latter can be mechanically represented only by means of assumptions regarding initial conditions.
4
Of course, these links can be maintained if one is willing to work with a sense of entropy that is sufciently
loose and elastic. But then the relevance of Boltzmanns apparatus becomes dubious.
J. Earman / Studies in History and Philosophy of Modern Physics 37 (2006) 399430 400
qualms about Boltzmanns explanation. These qualms are waived in order to concentrate
on two key problems with which Boltzmann struggled, the initial state problem and the
asymmetry problem. These problems are reviewed in Section 4, and Boltzmanns resort to
cosmology in order to resolve these problems is discussed in Section 5. His own version of
a cosmological solution is judged to be unsatisfying. Section 6 recounts how, according to
received opinion, modern general relativistic cosmology comes to Boltzmanns rescue. This
alleged rescue is questioned in Sections 7 and 8, rst by presenting competing intuition
pumps that yield conicting results about the improbability of a low entropy initial state
for the universe and then by presenting some precise model calculations indicating that in
the very cosmologies that were supposed to answer Boltzmanns prayers, Boltzmann
entropy is an ill-dened or severely hobbled concept. Section 9 adds the complaint that
even if the Boltzmann entropy of the initial state of the universe has a well-dened, low
value, this fact does not sufce to explain the temporally asymmetric behavior of the kinds
of thermodynamical systems of most interest to us. Section 10 argues that the dismal
prospects for using cosmology to complete Boltzmanns program does not threaten to
undermine our inferential practices. In the course of the argument I sketch an alternative
to Boltzmanns program for explaining the approximate validity of thermodynamics.
Section 11 reviews various interpretations of Boltzmanns speculation that the direction of
time is enslaved to the entropy gradient. Conclusions are presented in Section 12.
2. The logic of Boltzmanns explanation of the Second Law
I am not going to retell yet again the oft told story of Boltzmanns failed attempt to
provide a purely mechanical grounding for the strict validity of the Second Law of
thermodynamics, his (in)famous H-theorem, the reversibility objection of Loschmidt, the
recurrence objection of Zermelo, etc.
5
Sufce it to say that eventually Boltzmann was
forced to a conclusion which James Clerk Maxwell had reached many years earlier on the
basis of his thought experiment involving (what we now call) Maxwells Demon;
6
namely,
that the
Minimum Theorem [H-theorem], as well as the so-called Second Law of
Thermodynamics, are only theorems of probability. The Second Law can never be
proved by means of the equations of dynamics alone. (Boltzmann, 1895, p. 414).
That the Second Law, although not strictly true, retains a statistical validity is easy to say
but notoriously difcult to make precise. In what follows I will make no attempt to ag all
ARTICLE IN PRESS
5
For an authoritative version of the story, see Brush (1975). Foundational issues raised by the story have
received much attention in the philosophical literature; for overviews, see Callender (2001), Sklar (1993, 1995),
and van Lith (2001).
6
In a letter to Strutt dated 6 December 1870, Maxwell wrote: the 2nd law of thermodynamics has the same
degree of truth as the statement that if you throw a tumblerful of water into the sea, you cannot get the same
tumblerful out again (Strutt, 1968, p. 47). In a letter to Tait, Maxwell mocked Boltzmanns attempts to provide a
purely mechanical foundation for the Second Law: But it is rare sport to see these learned Germans contending
for the priority in the discovery that the Second Law of [thermodynamics] is the Hamiltonische Princip . . . . The
Hamiltonische Princip, the while, soars along in a region unvexed by statistical considerations while the German
Icari ap their waxen wings in nephelococcygia, amid those cloudy forms which the ignorance and nitude of
human science have invested with the incommunicable attributes of the invisible Queen of Heaven. Quoted in
Knott (1911, pp. 115116).
the niceties that are needed, and will concentrate on the ones that bear directly on the main
issues of concern here.
I will situate the discussion in a rather general and abstract mathematical setting that
will prove useful later when addressing the cosmological context. A deterministic, time
translationally invariant dynamical system is a quadruple (X; m; B; f
t
), where X is the state
space, m is a measure on X with m(X) = 1, B is the set of measurable sets of X, and for each
t c R, f
t
: X X is a oneone map. It is required that the f
t
are measure preserving, i.e.
for any A c B and for any t c R, m(f
t
(A)) = m(A), and that they have the group property
f
t
1
t
2
= f
t
2
f
t
1
with f
t=0
= id and f
t
= f
1
t
. The intended interpretation of f
t
is that if
at any given time t
i
, the state of the system is x c X, then f
t
(x) is the state at t
i
t.
7
The paradigm instantiation of these ideas is given by Hamiltonian dynamics. Abstractly,
a Hamiltonian system is given by a 2N-dim space X, a symplectic form o (i.e., a closed
non-degenerate 2-form), and a Hamiltonian function H : X R. The Hamiltonian ow
f
t
is dened by the integral curves of the vector eld V
H
on X determined by the condition
that V
H
o dH = 0,
8
and the measure m is the volume element associated with o, namely
o
N
= o . o . . o (N times). That f
t
preserves measure is the content of Liouvilles
theorem. More concretely, one can consider the Hamiltonian dynamics for a system of n
particles. The state space for this case is R
6n
. Coordinates (q; p), where q = (q
1
; . . . ; q
3n
)
records the particle positions and p = (p
1
; . . . ; p
3n
) records their momenta, can be chosen so
that o = dq
1
. . dq
3n
. dp
1
. . dp
3n
, and the associated measure is o
N
= o . o .
. o (N times). To assure that the measure normalizes, the region X R
6N
of the state
space available to the system has to be limited so that X has compact closure. This can be
guaranteed, for example, by conning the particles to a box, preventing the box from
exchanging energy with its environment, and requiring that the intra-particle interaction
potential is bounded from below. For such cases the relevant measure is m cut down to a
constant energy surface.
Next comes coarse graining in the form of the choice of a set {m
a
] of macrostates for
describing the outcomes of measurements that can be made on the system with
macroscopic instruments. It is assumed that the macrostates supervene on the microstates,
i.e. each m
a
corresponds to a measurable region M
a
_ X in the sense that at any time t, the
system is in macrostate m
a
just in case the microstate state x
t
at t belongs to M
a
. Introduce
the operations , v, and & where m, m v m
/
, and m&m
/
denote, respectively, the
macrostate that obtains if and only if m does not obtain, either m or m
/
obtains, and m and
m
/
both obtain; and assume that the micromacro correspondence satises the strictures
that (i) for any m c {m
a
], m corresponds to X M, and (ii) for any m; m
/
c {m
a
], m v m
/
and m&m
/
correspond, respectively, to M C M
/
and M M
/
. To make a probability space,
close {m
a
] under and under countable v-ing and &-ings. Then associated with the closure
{m
a
]
c
of the coarse graining is a probability measure Pr given by Pr(m) :=m(M) for
m c {m
a
]
c
.
Next comes Boltzmann entropy. The Boltzmann entropy S
B
(m) of a macrostate m is, by
denition, S
B
(m) :=k log(Pr(m)) = k log(m(M)). Assume that for each x c X there is a
ARTICLE IN PRESS
7
For a non-time translationally invariant dynamics, the time development of a state over a time interval D
would depend not only on D but also on the time at which the state obtains.
8
In canonical coordinates q
a
and p
a
, o = dq
a
. dp
a
and dH = (qH=qq
a
) dq
a
(qH=qp
a
) dp
a
. Thus, the
Hamiltonian vector eld is given by V
H
= (qH=qp
a
)(q=qq
a
) (qH=qq
a
)(q=qp
a
), and the ow f
t
is obtained by
solving Hamiltons equations _ q
a
= qH=qp
a
, _ p
a
= qH=qq
a
. The story is more complicated for constrained
Hamiltonian systems since o is degenerate; a relevant example is encountered in Section 8.
nest macrostate m
f
c {m
a
]
c
actualized by the microstate x, where nest means that if
m
/
c {m
a
]
c
is any other macrostate actualized by x, then M
f
_ M
/
. Then (relative to the
chosen coarse graining) the Boltzmann entropy S
B
(t) of the system at t is, by denition,
S
B
(t) :=S
B
(m
t
f
) = k log(Pr(m
t
f
)), where m
t
f
is the nest macrostate actualized by the
microstate at t. (From here on when I speak of the macrostate at time t, I will mean the
nest such state.)
We are now in a position to state the Boltzmannian version of the Second Law, or rather
a special case that will be the focus of the discussion that is to follow.
(B) Suppose that at t = 0 the Boltzmann entropy S
B
(0) of the system is low; then for
some appropriate t
1
40, it is highly probable that S
B
(t
1
)4S
B
(0).
The truth of (B) depends on features of the microdynamics. In particular, for (B) to be
true, f
t
must be such that for the overwhelming majority of microstates x in the region M
0
corresponding to the actual low entropy initial macrostate m
0
at t = 0, the macrostate m
1
at t = t
1
that results from the evolution x.f
t
1
(x) corresponds to a region M
1
such that
m(M
1
)bm(M
0
). The quasi-law status of (B) rests on the presumed fact that this feature of
the microdynamics does obtain for the sorts of systems we subject to thermodynamic
analysis and for the sorts of coarse graining relevant to explaining macroscopic
observations made on these systems over time periods of length comparable to the said t
1
.
It should be evident by now that Boltzmanns downgrading of the status of the Second
Law, signaled by the phrase so-called in the quotation at the beginning of this section, is
justied; and further justication will be supplied in the following section. Or, to be more
cautious, the downgrading is justied if the validity of the Second Lawwhatever form
that validity takesis supervenient on the microdynamics of the systems of interest. At the
end of the 19th century there were many physicists who held that the Second Law is not
just so-called but is one of the fundamental laws of nature; for them the failure of statistical
mechanics to underwrite this presumed fundamental lawlike status of the Second Law was
reason to reject Boltzmanns approach and to question the atomic hypothesis. By the end
of the rst decade of the 20th century, this attitude was conned to the fringes of physics.
Boltzmann had won; but digesting the fruits of his victory proved to be far from simple.
3. Qualms about Boltzmanns explanation
The difculties in implementing Boltzmanns explanation are well known, and I will
rehearse them only to the extent needed to set up the discussion in the following sections. A
second reason for the rehearsal is to counteract the impression given in some of the
philosophical literature that once the contraptions of Boltzmann statistical mechanics are
supplemented with some suitable Past Hypothesis, they function smoothly to underwrite
inferences about the past. In fact, the clanking sounds of these contraptions can be heard
from afar.
The special case (B) of Boltzmanns formulation of the Second Law speaks of a probable
increase in the entropy of the system. What is the justication for such talk? In the
preceding section I explained the truth conditions of (B) by assuming that probable is
judged by m-measure. But this denitional move leaves unexplained the connection
between the sense of probability so dened and the physical probabilityin a propensity
or a frequency sensefor the entropy to increase. Boltzmann realized that a connection
could be made between m and the limiting relative frequency sense of probability if the
ARTICLE IN PRESS
system is ergodic. This property can be dened in a number of equivalent ways. Here are
two:
Denition. A dynamical system (X; m; B; f
t
) is ergodic just in case (i) for any A c B such
that m(A)a0 and for almost any x c X there is a t40 such that f
t
(x) Aa+, or
equivalently, (ii) for any A c B, if f
t
(A) = A for all t, then m(A) = 0 or m(A) = 1.
9
For an ergodic dynamical system m can be interpreted in terms of the limiting relative
frequency of the time the state of the system spends in a region of X:
Lemma. If (X; f
t
; B; m) is ergodic, then for any A c B,
m(A) = lim
to
R
t
0
I
A
(f
t
(x)) dt
t
for almost any x c X, where I
A
(x) = 1 if x c A and 0 otherwise.
While some progress has been made, nagging issues remain. Are the physical systems of
interest ergodic? (See Earman & Re dei, 1996.) If not, will approximate ergodicity do? (See
Vranas, 1998.) How is limiting relative frequency of occupation time connected with
expectations about nite stretches of time? I propose to waive these naggings in order to
pass on to two further issues.
The rst is what can be termed the subjectivity challenge to Boltzmanns explanation.
Here is one rough attempt to get at the worry underlying this challenge. The fact to be
explainedthe tendency of the entropy of an adiabatically closed system to increaseis an
objective physical fact about the world. But Boltzmanns explanation (so the complaint
goes) has to do not with such objective physical tendencies but with our epistemology. The
coarse grained description in terms of macrostates is a way of codifying our ignorance of
the exact microstate of the system, and what (B) is telling us is that a certain measure of
this ignorance tends to grow with time. This may be interesting but (the complaint
concludes) it is not what was to be explained. I think this is not an unfair reading of
Poppers (1981) complaint in Quantum Theory and the Schism in Physics. And it is such
intuitions that, surely, underlie Alberts (1994) form of the complaint. Consider two
bodies, one hot and one cold which are brought into contact and whose temperatures
subsequently equilibrate. Then, writes Albert,
Nothing, surely, about what anybody may or may not know about these two bodies
. . . can have played any role in bringing it about (that is: in causing it to happen) that
the temperatures of those bodies subsequently approached each other! (Albert, 1994
p. 670)
This subjectivity challenge can be met, at least partially, by a more careful specication of
the goals of the inquiry. If the goal is to derive from statistical mechanics a fully objective,
observer independent arrow of time, then one is going to be disappointed, at least if the
underlying microdynamics of the system is time reversible.
10
But if the goal is to explain
the temporal asymmetries we observe in the entropic behavior of macroscopic systems,
then coarse graining is perfectly legitimate since what we observe corresponds to a coarse
ARTICLE IN PRESS
9
This is the modern formulation of ergodicity. For accounts of the history of this concept, see Brush (1975,
Section 10.10) and Gallavotti (1999, Section 1.9).
10
See Section 4 for a denition of time reversal invariance and a discussion of its implications.
grained description. And given an appropriate coarse graining, the increase in the
Boltzmann entropy is objective in the sense that it is grounded in the underlying
microdynamics of the system (assuming, of course, that the qualms raised above have been
adequately addressed). There would be a remaining gripe if the grounding were sensitive to
how the coarse graining is donee.g. if it worked when the energy surface is partitioned
into areas of
1
10
erg but not when it partitioned into areas of 1 erg. But in fact the grounding
appears to be robust under such changes.
11
Another worry focuses on the conditions of applicability of (B) to concrete physical
systems. One such condition is that for the system of interest, the microstate at t = 0 will be
typical of the microstates compatible with the observed macrostate at that time. The
idea can be made more precise in the following Statistical Postulate:
(SP) Let m
0
be the macrostate at t = 0 of the system of interest. Then the probability
at t = 0 that the microstate of this system lies in some measurable subset A _ M
0
of
the state space region M
0
_ X corresponding to m
0
is m(A)=m(M
0
).
It might seem that if (B) holds and if (SP) is true at t = 0, then (SP) cannot be true at the t
1
used in (B). From (B) it follows that for the overwhelming majority of microstates x in the
region M
0
, the macrostate m
1
at t = t
1
that results from the evolution x.f
t
1
(x)
corresponds to a region M
1
such that m(M
1
)bm(M
0
). By conservation of measure
m(M
0
) = m(f
t
1
(M
0
)). So, a fortiori, for the overwhelming majority of microstates in M
0
the
resulting macrostate m
1
at t = t
1
is such that m(M
1
)bm(f
t
1
(M
0
) M
1
), i.e. microstates in
M
1
that have evolved from M
0
are only a small m-fraction of the microstates of M
0
. Thus,
it seems that the microstate at t
1
cannot be typical of the microstates compatible with
the macrostate at t
1
. And now the Boltzmannian explanation threatens to unravel. For
t = 0 was not supposed to denote some special time but an arbitrary initial time. And if
(SP)s being true at some arbitrary initial time implies that it is false at a later time, how
can (SP) reasonably be assumed to apply to the time t = 0 referred to in the antecedent
of (B)?
One response is that although, of course, there is nothing special about the time t = 0
itself, there is something special about the initial state; namely, it is part of the applicability
of (B) that we have no information about the initial microstate other than that implied by
the observed fact that the system is in the macrostate m
0
. And given this state of
knowledge, a principle of indifference argument can be used to justify the assumption that
the initial microstate is equally likely to lie in equal m-volumes of M
0
. This response, while
not unreasonable, is prey to the subjectivity complaint of Popper and Albert.
The best response does not appeal to an ignorance interpretation of probability but
rather exploits a non-sequitur in the argument that was supposed to show that if (B) holds
and if (SP) is true at t = 0, then (SP) cannot be true at t
1
. The fact that the macrostate m
1
at t
1
is such that the microstates in M
1
that have evolved from the phase space region
corresponding to the initial macrostate at t = 0 are only a small m-fraction of the
microstates of M
1
does not entail that the microstate at t
1
cannot be typical of the
microstates compatible with the macrostate at t
1
; for the sense of typicality relevant to
ARTICLE IN PRESS
11
But (as emphasized to me by a referee) Boltzmanns approach cannot be completely absolved of the
subjectivity charge. For example, the approach depends on taking the state space to be the momentum-position
space rather than the velocity-position space; and the approach has to be limited to macroscopic variables that can
be represented by measurable phase functions.
(SP), it sufces that the microstates that have evolved from the initial macrostate are
sufciently spread over the measurable subsets of M
1
. And for the needed spreading one
can appeal to a cousin of the property of mixing, which is stronger than the property of
ergodicity:
Denition. A dynamical system (X; m; B; f
t
) is mixing just in case for any A; B c B,
lim
to
m(f
t
(A) B) = m(A)m(B).
Rather than the asymptotic property of the formal denition, what is needed in support
of the reapplication of (SP) at t
1
40 is for effective mixing to take place over the relevant
macrotime scale for those phase space regions corresponding to the macrostates at issue. In
particular, it would sufce if, for any macrostate m c {m
a
] and any measurable subset
F _ M
a
of any state space region M
a
corresponding to a macrostate m
a
, m(f
t
1
(M)
F) - m(M)m(F) = m(f
t
1
(M))m(F), where the approximate equality - is tight enough to
accord with what is macroscopically discriminable. This effective mixing guarantees that
whatever the initial macrostate m
0
at t = 0 and resulting macrostate m
1
at t
1
40, the phase
points that originate in M
0
appear on the macrolevel to be uniformly spread over the
region M
1
at t
1
40.
Doubts can be raised about whether effective mixing can be demonstrated for the
relevant physical systems. But the point I want to emphasize is independent of such doubts:
effective mixing succeeds in supporting (SP) only by undermining the Boltzmannian
apparatus as a means of making predictions about the macroscopic behavior of the system.
For effective mixing means that, insofar as our rational expectations are governed by the
measure m, the information value of knowing that at t = 0 the system is in a certain
macrostate m
0
is effectively lost over the relevant time scale since the phase points that
originate in M
0
are effectively uniformly spread over all the phase space regions
corresponding to macrostates at issue.
To allow for knowledge of macroscopic states to have predictive power while also
providing for (SP), one could hope for a Goldilocks form of effective mixing which can be
formulated as follows. Let m
j
, j = 0; 1; 2; . . . ; N, be any sequence of macrostates
compatible with the microdynamics, with m
0
being a low entropy state, with the times
of occurrence of successive states differing by an amount t
1
, and with Nt
1
being some
relevant length of observation time. Then Goldilocks mixing requires that for all j,
m(f
t
1
(M
j
) M
j1
)=m(M
j
) - 1; and moreover, for any measurable F
j1
_ M
j1
,
m(f
t
1
(M
j
) F
j1
) - m(M
j
)m(F
j1
). In words, the vast majority of phase points originating
in the region M
j
corresponding to the macrostate m
j
evolve over a time span t
1
to the
region M
j1
corresponding to m
j1
and also appear to be spread uniformly over M
j1
.
This Goldilocks form of effective mixing requires the dynamics to be balanced between
enough sensitive dependence on initial conditions to spread the phase points initially in M
j
over M
j1
vs. enough stability needed to avoid non-Goldilocks mixing that would
undermine the predictive and retrodictive power of macrostates. It remains to be seen
whether the systems to which thermodynamic reasoning is applied have a microdynamics
that performs this balancing act.
These qualms about the logic of Boltzmanns program will be waived because they will
seem minor when compared to the initial state and asymmetry problems to be discussed in
the following section.
ARTICLE IN PRESS
4. The initial state problem and the asymmetry problem
Boltzmanns (B) has the form of a conditional whose antecedent asserts that the
Boltzmann entropy is low at t = 0. Two questions immediately arise. First, what is the
origin of the initial low entropy state that instantiates the antecedent of (B)? Second, is it
reasonable to assume that the low entropy initial state satises the Statistical Postulate? I
will concentrate on the rst question.
In some instances the answer may simply be that that is the way a human agent arranged
things, e.g. at t = 0 an experimenter places a cold body in thermal contact with a hot body
and thermally isolates this two-body system; or the experimenter evacuates the gas in one
chamber of a box divided into half by a partition, and then at t = 0 removes the partition,
etc. But obviously such an answer does not extend to events that were not engineered as
part of an experiment. Before considering answers that cover events not subject to human
intervention it will be helpful to get a second issue on the table.
The second issue arises from the recognition that (B) does not capture in probabilistic
format the temporal asymmetry of the Second Law. That requires something like
(B
) Suppose that at t = 0 the Boltzmann entropy S

B
some appropriate t
1
40, it is highly probable that S
B
(t
1
)4S
B
(0) and it is highly
probable that S
B
(0)XS
B
(t
1
).
But (B
) is not a consequence of the Boltzmann apparatus. Indeed, given the presumed

time reversal invariance of the microdynamics, one would expect that if (B) is true, it
should also be true that
(B
n
) Suppose that at t = 0 the Boltzmann entropy S
B
the t
1
40 of (B) it is just as probable that S
B
(t
1
)4S
B
(0) as that S
B
(t
1
)4S
B
(0).
But (B
n
) contradicts (B
). And it also contravenes our (presumably) rational expecta-

tions.
12
If, for example, at t = 0 we observe a thermally isolated system consisting of an ice
cube oating in a glass of lukewarm water, we expect in accord with (B) that if there is no
intervention over (say) the next 5 min, then at the end of that time the ice cube will have
partially melted and the water will have cooled correspondingly. And further, in accord
with (B
) and contra-(B
n
), we infer that if there was no intervention over the preceding
5 min, then 5 min earlier the ice cube was less melted and the water was correspondingly
warmer. And more generally, to the extent that memories and records of the past depend
on entropic behavior, the truth of (B
n
) seems to undermine our knowledge of the past since
it seems to imply that what we take to be memories and records are imposters because they
are not lingering traces of past events but rather the results of uctuations from higher
entropy states.
13
Because of the importance of this issue it is worthwhile examining in some detail the
assumptions that go into generating (B
n
). Time reversal invariance for a dynamical system
(X; m; B; f
t
) is explicated as follows. First, there is reversal operation on states given
formally by a oneone mapping R : X X, denoted by
R
x, having the involutional
property that
R
(
R
x) = x. For the main application, Hamiltonian particle mechanics,
x = (q; p), where q denotes the instantaneous positions of the particles and p denotes their
ARTICLE IN PRESS
12
See, however, the discussion in Section 10.
13
See Albert (2000, Chapter 4) for an engaging discussion on this point.
instantaneous momenta, the reversal operation is just reversal of momenta, i.e.
R
(q; p) = (q; p).
14
Time reversal invariance for the dynamics means that if the system
evolves from an initial state x
i
to a nal state x
f
= f
D
(x
i
) in a time span D, then in the same
time span the reversed state
R
x
f
evolves to
R
x
i
= f
D
(
R
x
f
).
Suppose that the Boltzmann apparatus is used to calculate the conditional probability
that a system initially in macrostate m
i
at t = 0 will evolve to a macrostate m
f
at t = D40
as
Pr(m
f
(D)=m
i
(0)) = m(f
D
(M
i
) M
f
)=m(M
i
). (1)
Make the following assumptions:
(A1) the microdynamics is time reversal invariant,
(A2) for any Y c B, m(Y) = m(
R
Y), where
R
Y :={x c X:
R
x c Y],
(A3) for any macrostate m c {m
a
] in the coarse graining, if M _ X is the region
corresponding to m then the region
R
M corresponds to a macrostate
R
m c {m
a
].
Assumption (A3) allows us to say that a macrostate m is reversal invariant just in case
R
m = m. Then if the initial macrostate m
i
is reversal invariant, assumptions (A1)(A2)
yield
Pr (
R
m
f
(D)=m
i
(0)) = m(f
D
(M
i
)
R
M
f
)=m(M
i
)
= Pr(m
f
(D)=m(0)). (2)
The proof of the second equality in (2) simply involves mechanical manipulations, and is
left to the reader. Since by (A2) m(
R
M
f
) = m(M
f
), the macrostates m
f
and
R
m
f
have the
same Boltzmann entropies. Thus, it follows from (2) that if the initial macrostate m
i
of the
system has low Boltzmann entropy, the system is as equally likely to have come from a
higher entropy state as it is to be headed towards a higher entropy state.
Assumption (A2) can be motivated by the idea that the measure m should be adapted to
the symmetries of the dynamics. In any case, (A2) does hold for Hamiltonian mechanics.
(A2) would not need to be invoked in the proof of (2) if
R
m = m holds for all macrostates
and not just for the initial macrostate.
15
Whether or not this condition holds depends in
part on what is packed into the notion of a macrostate. If instantaneous rates of change of
macroparameters are admitted as part of the macrostate description, then the verdict is
negative. Consider, for example, the macrostate m of a gas in which the instantaneous
rate of change of volume is positive; here
R
mam since the microstates in
R
M are those of
a gas whose volume is contracting rather than expanding. On the other hand, if the
macrostate is dened solely by instantaneous values of macroparameters, like volume
and pressure, but not their instantaneous rates of change, then the verdict is probably
positive. It seems that nothing essential is lost by taking the latter route since the change in
the macroparameters can be captured by a succession of macrostates m(0); m
/
(); m
//
(2); . . .
for some appropriate small 40. The result (2) can be generalized to show, for example,
that if
R
m = m and
R
m
/
= m
/
, then Pr(m
//
(2)=m(0)&m
/
()) = Pr (
R
m
//
(2)=m(0)&m
/
()).
But note that it does not follow from time reversal invariance alone that
ARTICLE IN PRESS
14
In other applications the denition of the reversal operation can be a matter of contention; compare Albert
(2000, Chapter 1) to Earman (2002) and Malament (2004).
15
Albert (2000, Chapter 4) assumes that (A2) does hold for all macrostates.
Pr(m
//
(2)=m(0)&m
/
()) = Pr (
R
m
//
(2)=m(0)&m
/
()). Even though the apparatus is time
symmetric, if time asymmetric information is put in, time asymmetric conclusions can
come out.
5. Boltzmanns cosmological solutions to the initial state and asymmetry problems
Here is one possible solution to the initial state problem. (For future reference, call it
(S1).) Suppose that the dynamical system of interest is ergodic. Then the eventual
occurrence of low entropy states is virtually assured, for ergodicity guarantees that for
almost every dynamical trajectory and for any macrostate, even one with a very low
entropy, there will be a time at which the microstate actuates the said macrostate.
Furthermore, it follows from Poincare s recurrence theorem that the low entropy
macrostate will, in the fullness of time, occur again and again.
16
The fact that we happen to
be living during an epoch that began with a low entropy state requires explanation, but
perhaps it sufces to use a form of anthropic reasoning, viz. the fact in question should not
be regarded as surprising since a sufciently low entropy is required for the existence of
critters like ourselves.
17
To my knowledge, Boltzmann nowhere explicitly considers (S1),
but it is implicit in the entropytime graphs he draws.
His own earliest attempted solution to the initial state problem (call it (S2)) appealed not
to ergodicity and the fullness of time but to cosmology and the vastness of the universe.
The rst published version is found in a letter to Nature. The account begins with the
ansatz that the whole universe is, and rests forever, in thermal equilibrium.
The probability that one (only one) part of the universe is in a certain state, is the
smaller the further this state is from thermal equilibrium; but this probability is
greater, the greater is the universe itself. If we assume the universe great enough, we
can make the probability of one relatively small part being in any given state
(however far from the state of thermal equilibrium) as great as we please. We can also
make the probability great that, though the whole universe is in thermal equilibrium,
our world is in its present state. (Boltzmann, 1895, p. 415).
18
This ideathat the vast universe, which is in thermal equilibrium, is composed of many
worlds the size of our galaxy and that as a result of uctuations some of them nd
themselves in an initially improbable (i.e. low Boltzmann entropy) stateis repeated
several times (see Boltzmann, 18961898, Section 90, 1897a, Section 7, 1897b, Sections
45, 1904, pp. 171172). That our own world is one of these can, perhaps, be explained on
anthropic grounds.
In at least two places Boltzmann offers (S2) as one of two possibilities, the other (call it
(S3)) being that the entire universe began in a low entropy state, that ever since the increase
in entropy has been monotonic, and that the increase has been sufciently slow as to leave
the present value of entropy well below its maximum (see Boltzmann, 1897a, Section 7,
ARTICLE IN PRESS
16
Poincare s recurrence theorem applies to any dynamical system (X; m; B; f
t
) regardless of whether or not the
system is ergodic. The theorem shows that for almost any x c X and any A c B such that x c A and m(A)40, if
for some t it is the case that f
t
(x) A = +, then there is a t
/
4t such that f
t
(x) Aa+. Zermelos objection to
Boltzmanns H-theorem was based on Poincare s recurrence theorem. English translations of Zermelos papers
and Boltzmanns responses can be found in Brush (1966).
17
I will not broach the issue of whether anthropic explanations are genuine explanations or mere nostrums.
18
Boltzmann attributes this idea to his old assistant, Dr. Schuetz.
1897b, Section 45). Boltzmann obviously did not nd (S3) attractive, and it is not hard to
imagine why. Classical spacetime theories offer no provision for a beginning of time itself,
save for the articial one of deleting all spacetime points on or below some t = const time
slice. In a spacetime where time extends back to t = o, a beginning for the material
universe can be contemplated as the creation ex nihilo at some nite time in the past of the
matter that makes up the constituents of Boltzmanns worlds. Neither alternative is very
prepossessing. (Had he lived another two decades Boltzmann would have seen how general
relativistic cosmological models escape between the horns of these unattractive
alternativessee Section 6.) But leaving aside the issue of what a beginning of the
universe is supposed to mean in a classical spacetime setting, (S3) is obviously unattractive
in comparison with (S1) and (S2): since the Boltzmann apparatus suggests no mechanism
for grounding (S3), adopting (S3) has the appearance of a deus ex machina.
As far as I know Boltzmann never considered a fourth possible solution (call it (S4)),
which amounts to a denial of the presupposition of the initial state problem: namely, there
is no initial low entropy state for our world or for the universe; it is rather that in the
past direction there is monotonic decrease in entropy from its present value towards a
minimum at t = o.
Let us now examine Boltzmanns favored alternative (S2). There is no reason to doubt
that in a large universe in thermal equilibrium, relatively small parts will have low entropy,
at least in an intuitive sense; that is, some parts of the universe will display properties, such
as macroscopic temperature gradients and/or matter density gradients, that are indicative
of thermodynamical non-equilibrium, in which case one can say that these parts have low
entropy in some intuitive sense of that term. But assumptions are needed to connect this
intuitive sense of entropy with the Boltzmann sense. For Boltzmann entropy is a non-local
quantityit does not attach directly to spatial or spatiotemporal regions but rather to
macrostates of a dynamical system. Presumably, in suitable circumstances the Boltzmann
entropy of a dynamical system will lend itself to a description in terms of a local entropy
density and an entropy ux.
19
Grant that these circumstances obtain for the uctuation
that creates the initially low entropy for our world. Even so, there is an apparently fatal
objection to Boltzmanns (S2).
Feynman (1994) dismisses Boltzmanns uctuation story as ridiculous. Suppose for
sake of reductio that a uctuation out of an equilibrium state for the universe produces an
orderly patch in our neighborhood. Then (according to Feynman) the most likely
condition of the rest of the universe is that it is mixed up. Thus, he concludes, the
prediction would be that if we looked at a place where we had not looked before, it would
be disordered and a mess (p. 115). But this prediction is falsied by what we actually
observe. In response Boltzmann could appeal to a uctuation large enough to encompass
the portion of the universe visible to us. Such a uctuation, however, is much less probable
than one which encompasses only our immediate neighborhood. The logical extension of
this line of reasoning leads to what has become known as Boltzmanns brain paradox; viz, a
uctuation that encompasses our neighborhood and produces the orderly patterns we
perceive here is much less likely than a uctuation that simply creates Boltzmanns brain
ARTICLE IN PRESS
19
Alternatively the uctuation can be coupled with the creation of a branch system which is sufciently
isolated that, to good approximation, it can be treated as a closed dynamical system in its own right. Then the low
entropy initial state refers to the Boltzmann entropy of this branch system at the time of branching off. But
obviously such a scenario requires additional assumptions.
(or your brain or mine) complete with the faux memories of macroscopic objects
behaving in accord with the laws of thermodynamics.
20
The upshot is that the position best
supported by Boltzmanns uctuation story is solipsism of the present moment.
21
Boltzmanns solution to the initial state problem cannot be discussed independently of
the asymmetry problem because his solution to the former was supposed to do double duty
by also providing a solution to the latter. This is made clear in Boltzmanns Lectures on
Gas Theory:
That in nature the transition from a probable [high Boltzmann entropy] to an
improbable [low Boltzmann entropy] state does not take place as often as the
converse, can be explained by assuming a very improbable initial state of the entire
universe surrounding us, in consequence of which an arbitrary system of interacting
bodies will in general nd itself initially in an improbable state. (Boltzmann,
18961898, p. 447).
But the solution to the initial state problem provides a solution to the asymmetry problem
only if additional assumptions are satised. Suppose for sake of discussion that solipsism
of the present moment is rejected out of hand, and grant that our current perceptions and
memories of relevant temporal asymmetriese.g. that the temperatures of bodies in
thermal contact tend to equalize rather than to become more disparateare not illusions.
Then the initial conditions at some time in the past must be favorable, and if the
Boltzmann apparatus provides the correct explanatory machinery, then being favorable
means that the initial state had low Boltzmann entropy. That is not an hypothesis but a
deduction. What is an hypothesis by Boltzmanns lights is that the time of occurrence of
the initial state is sufciently far in the past that the inferences we make about the past
do not overreach it. We surely want to make inferences about events that predate our
earliest memories (which are stipulated to be veridical), and so the initial state must
predate them. There was nothing in the cosmology of Boltzmanns day to give any
theoretical backing to this hypothesis, and so for him it was simply a posit. One can try to
dignify this posit by calling it a presupposition of our knowledge of a past more distant
than the reach of our memories, but that begs the question of whether this alleged
knowledge deserves the name. Nor will anthropic reasoning ll the gap, for it is no skin off
the nose of any plausible anthropic principle that a low Boltzmann entropy state that
predates the reach of our memories is the product of a downward uctuation from a higher
entropy state rather than the result of evolution from a state of even lower entropy.
Furthermore, there is an obvious tension in Boltzmanns solution to the initial state and
asymmetry problems. To make sure that our inferences to the past do not overreach the
time of the initial low entropy state, the tendency is to push this state into the far past.
But the further into the past it is pushed, the less it is able to guarantee that the entropy of
the present state, though higher than that of the initial state, is nevertheless itself low.
That guarantee requires another posit. In his second reply to Zermelo, Boltzmann
admitted that the solution to the asymmetry problem requires additional posits over and
ARTICLE IN PRESS
20
This Boltzmanns brain paradox is discussed by a number of authors. It is often attributed to Barrow and
Tipler (1986), but the basic idea may have been due to Bronstein and Landau (1933, p. 72).
21
It is interesting to note that there is still a lively controversy in cosmology over how to calculate the most
likely outcome of a uctuation; compare, for example, Dyson, Kleban, and Susskind (2002) with Albrecht and
Sorbo (2004).
above those involved in his uctuation hypothesis. His own formulation of the needed
posit is labeled as Assumption A: that the universe, considered as a mechanical systemor
at least a very large part of it which surrounds usstarted from a very improbable state,
and still is in an improbable state (Boltzmann, 1897b, pp. 238239). In effect, Boltzmann
has been forced to move to something close to (S3). He characterizes Assumption A as a
comprehensible physical explanation of the peculiarity of the initial state, consistent with
the laws of mechanics; or better, it is a unied viewpoint corresponding to these laws,
which allows one to predict the type of peculiarity of the initial state in any special case; for
one can never expect that an explanatory principle must itself be explained (Boltzmann,
1897b, p. 239). This rhetorical ourish cannot disguise the avor of a solution gained by
too many posits and not enough honest toil.
22
This embarrassing predicament is presumably what led Boltzmann to entertain a
proposal that would obviate the need for so many posits. This proposal will be discussed in
Section 11. But rst I want to take up the contention that modern cosmology rescues
Boltzmann from the predicament by turning the posits into facts.
6. Modern cosmology to the rescue (?)
The standard big bang models of modern general relativistic cosmology imply a
beginning for time in the sense that there is an upper bound on the proper length of any
past-directed timelike curve as measured from any point of the spacetime. For the actual
universe, the bound is thought to be about 14 billion years for curves starting at the present
moment. Moreover, these models cannot be extended through the big bang singularity in
any physically meaningful way (e.g. Einsteins gravitational eld equations are not
meaningful, even in a distributional sense, at this singularity). Of course, in these models
there is no beginning for time in the sense of a rst instant. But this does not matter in
the present context, since for purposes of solving the initial state and asymmetry problems
the initial state can be taken as the state at some time or other in the very early universe.
The choice of the exact time does not matter as long as the chosen state has the requisite
low entropy and the Statistical Postulate (SP) applies. Then the measure on which to base
statistical reasoning about the thermodynamics of the universe would seem to be
m
I
():=m(=I), where m is the appropriate measure on the state space for the universe and I is
the region of the state space compatible with the low entropy initial macrostate state of the
relevant coarse graining. The hope is that this conditionalized measure will vindicate the
temporally asymmetric inferences we routinely make about thermodynamic systems.
Before going further it is well to note that applying of the Second Law of
thermodynamics to the universe is a problematic exercise. Standard cosmology describes
the large scale structure of the universe using the FriedmannRobertsonWalker (FRW)
models.
23
Current observational data are consistent with all three types of FRW models:
k = 0, 1, and 1, whose space sections have, respectively, zero curvature, constant
negative curvature, and constant positive curvature. In the rst two cases space is innite,
24
ARTICLE IN PRESS
22
Note also that in order to get the Boltzmann apparatus into gear, the initial state must satisfy the Statistical
Hypothesis (SP), as discussed in Section 3, requiring yet another posit.
23
The line element can be written as ds
2
= a(t)[dr
2
=(1 kr
2
) r
2
dy
2
r
2
sin
2
y dj
2
] dt
2
, where a(t) is the
scale factor and k = 0; 1; or 1.
24
Unless topological identications are made to make the space sections compact.
and as a result such thermodynamic notions as the energy of the system, the work done by
the system, etc., are problematic if the system is the entire universe.
25
To overcome this
difculty, one could try to apply the Second Law to a nite comoving patch lying within
the portion of the universe observable by us and argue that, although strictly speaking this
patch is not an adiabatically closed system, the observed homogeneity of the universe
indicates that there is little net entropy ux across the boundary of the patch (see Carroll &
Chen, 2004). But the universe is homogeneous only at large scalesabove 10 mega
parsecsand at the smaller scales at which we typically apply thermodynamic reasoning;
the inhomogeneities can be expected to be associated with entropy gradients. Additionally,
if the expansion of the universe is acceleratingas current evidence strongly indicates that
it is (see Carroll, 2004)then the application of thermodynamic reasoning to a volume
comoving with the expansion is complicated by the fact that the energy of the matter elds
in the comoving volume increases with time.
26
But as usual in these discussions, let us agree
to suspend doubts about such niceties.
Nevertheless, it would seem at rst blush that very early states of the standard hot big
bang models do not answer the call for a low entropy state; on the contrary, a thermalized
state (needed to explain the abundances of the light elements) with a very uniform
distribution of mass-energy would seem to have high entropy instead of the requisite low
entropy. The usual response is that our intuitions are misled by the failure to take into
account the entropy associated with the gravitational degrees of freedom. Since gravitation
tends to make matter clump, the ner the mesh of the coarse graining, the more special
or unlikely a macrostate in which matter-energy is uniformly distributed on the scale set
by the coarse graining. At this juncture it is standard in the literature to quote Penroses
(1989, Chapter 7) estimate that for a k = 1 (spatially closed) FRW model, the
unlikeliness of an appropriate smooth macrostate is 1 part in 10
10
123
.
It has to be emphasized that such numbers are the result of purely heuristic
considerations and that no connectionother than a purely verbal oneto the Boltzmann
apparatus has been made. In terms of Boltzmann entropy, the quoted number would have
to be taken to mean that the ratio of the state space volume corresponding to the initial
smooth macrostate to the total volume is 10
10
123
.
27
But the basis for establishing this
meaningthe denitions of the state space, the measure m, the dynamics, and the coarse
grainingare not given. This might seem like a quibble, and an unproductive one at that.
Theoretical physicists routinely give heuristic order of magnitude estimates, and it often
turns out, after a long and hard technical analysis, that the estimates of the rst-rate
physicists are roughly correct. Thus, it might be urged, we should not worry about whether
Penroses estimate is off by a factor of 10
10
or so, but rather should get on with digesting
the consequences of the extraordinary improbability of the initial state. My qualm is not
based on a quibble over a factor of 10
10
but rather on a worry about whether any value for
the Boltzmann entropy of the initial state is meaningful. This worry will be given concrete
form in Section 8. For now I want to enter into the spirit of the discussions which take
Penroses estimates at face value.
ARTICLE IN PRESS
25
If gravitational energy is taken into account, then in cases where space is compact one meets the difculty that
classical general relativity supplies no well-dened notion of the energy of the entire universe.
26
In a FRW model the source matter takes the form of a perfect uid characterized by an energy density m and a
pressure p. The energy of a volume element comoving with the expansion scales as a
3w
(t), where w:=m=p.
Accelerating expansion requires that wo
1
3
.
27
See, for example, Lebowitz (1993, p. 13; 1999, p. S350) who puts this gloss on Penroses estimate.
That spirit is captured by the many approving references to Penroses estimatefrom
experts in statistical mechanics (e.g. Lebowitz, 1993, 1999) and cosmology (e.g. Albrecht,
2004) as well as philosophers of science (e.g. Price, 1996, p. 83). A somewhat over-the-top
example from philosophy of science is Prices (2004, p. 228) remark that In my view, this
discovery about the cosmological origins of low entropy is the most important
achievement of late-twentieth-century physics .
28
Although other authors may not share
Prices importance ranking, they typically agree that a low entropy initial state is a
discovery of modern cosmology, and then turn to the issue of whether the specialness of
the initial state needs explaining and, if so, what form that explanation should take. The
following section briey reviews the controversy this issue has generated and argues that a
resolution of the controversy is not to be obtained by means of intuition pumps but rather
through precise model calculations. Such calculations threaten to undermine the notion
that modern cosmology completes Boltzmanns program.
7. Competing intuition pumps
The program of explaining initial conditions of the universe does not t with the
methodology of other branches of physics outside of cosmology; indeed, the notion that
various branches of physics are incomplete because they fail to offer theories of initial
conditions is apt to strike the practitioners in these branches as bizarre. Perhaps there is
something special about cosmology that calls for a theory of initial conditions; but if so, it
should be based on something more substantial than the notions that the big bang
represents a creation of the universe and that it is legitimate, if not obligatory, to inquire
why the creation went this way rather than another. The conceit of a blindfolded Creator
who actualizes a universe by randomly throwing a dart at the cosmic dartboard of initial
conditions for the universe might be useful for some purposes; if, for example, it is taken
seriously, it provides an explanation of why the Statistical Postulate (SP) is satised by the
initial state. But, arguably, this advantage is not worth the price of the metaphysical
baggage used to secure it. In sum, there is much to be said for the position that no
explanation of the initial conditions of the universe is called for.
29
Another possible positionwhich does resort to a theory of initial conditionsis that
the seeming specialness of the initial state of the universe is an illusion. This state is special
in the comparison class of logically possible states. But for physics the relevant comparison
class is the class of physically possible states, and this class (so the theory of initial
conditions goes) does not include the seemingly generic lumpy initial states, which are
ruled out by a lawlike constraint on initial conditions. Penrose (1979, 1989, Chapter 7,
2004, Chapter 28) has hypothesized that there is a time asymmetric (local) physical law
that only becomes important near spacetime singularities and that forces a smooth
spacetime geometry at the big bang but not at the big crunch (should there be one, as in the
k = 1 FRW model without cosmological constant). His tentative proposal for trying to
make this hypothesis precise is to take the time asymmetric law to require that the Weyl
curvature tends to zero as the initial singularity is approached. The details of Penroses
ARTICLE IN PRESS
28
See also Price (2002).
29
For a forceful argument for this position, see Callender (2004).
proposal do not matter for present purposes.
30
What does matter is that if his proposal is
on the right track, it seemingly undermines the discovery that modern cosmology
provides the answers to Boltzmanns prayers by providing an improbable or low entropy
initial state.
31
Inationary cosmology offers yet a different perspective on the issue of the initial state
of the universe by claiming to provide a mechanism that obviates the need for a special
beginning state. It allows that the universe began in a generic lumpy state and postulates
that subsequently a period of very rapid accelerated expansion (aka ination), which
initiated around 10
35
s after the big bang, effectively smoothed out the lumpiness.
32
Since
the inationary era is hypothesized to have lasted only a fraction of a second, the time of
the initial state can, without any awkwardness for the Boltzmann program, be taken to
coincide with the end of the inationary era.
33
Work in inationary cosmology was
initially stimulated by the promise of tying cosmology to particle physics by linking the
hypothesized ination eld, that is supposed to drive ination, to a specic mechanism
in elementary particle physics. To date this promise has not been fullled, and the
ination eld is just a name for a scalar eld that does what it has to do to produce the
right kind of ination at the required time. But the program has been successful in offering
a natural explanation of the nearly scale free spectrum of density perturbations revealed by
measurements of the cosmic microwave background radiation. Critics of inationary
cosmology respond that the fundamental mechanism used by inationary cosmologists in
their explanation of the spectrum of density perturbations may operate whether or not
ination occurs (see, for example, Hollands & Wald, 2002a). Such issues are outside the
scope of this paper. But what is relevant is that, as with Penroses solution to the cosmic
dartboard problem, the inationary solution threatens Boltzmanns program.
If ination is indeed an effective smoothing mechanism, then it might seem to be an anti-
thermodynamic machine that reduces Boltzmann entropy by taking a pre-ination lumpy
macrostate corresponding to a sizable volume of state space and evolving it into a post-
ination era smooth macrostate corresponding to a much smaller volume of phase space.
This in itself need cause no alarm for the Boltzmann program since, as noted above, the
initial state can be identied with a post-ination state. But what should cause alarm is
the recent discovery that the universe has entered a new era of ination ( = accelerating
expansion), albeit of a less intense form than in the very early universe. This is a genuine
discovery for which there are multiple independent lines of evidence based on different
observational techniques and background assumptions.
ARTICLE IN PRESS
30
How fast does the Weyl curvature go to zero as the initial singularity is approached, does the approach to zero
take place in a parallely propagated frame or some other frame, etc.?
31
How then can the entropy of the universe be expected to increase with time? Penroses answer is that since the
hypothesized lawlike constraint on the initial conditions is not macroscopically discernible at a time t sufciently
later than the time of the initial state, the entropy at t should be calculated not by using the phase volumes within
the manifold of states compatible with all the laws but rather by using the larger manifold of states for which the
initial constraint is not required to hold; see Penrose (1979, pp. 632633). But any entropy increase obtained by
reference to physically impossible states is vulnerable to the objection that the increase is merely epistemic.
32
There are precise theorems showing that for a homogeneous but non-isotropic universe, ination is an
effective mechanism in smoothing out anisotropies (see, for example Wald, 1983). There are also results that
attempt to show that ination also serves to smooth out inhomogeneities, but these results are based on much
more dubious and restrictive premises than the former results.
33
See, for example, Albrecht (2004, p. 364).
The discussion of these issues is typically carried out in terms of dueling heuristic
arguments or, in philosophers jargon, intuition pumps. Intuition pumps are notoriously
ckle and can often be reprimed to run in the opposite direction. As a case in point,
consider again a hot, relatively smooth state for the very early universe. Naively, such a
state seems to have a high entropy. But this impression, it was argued, is due to the neglect
of the gravitational degrees of freedom. Now consider a lumpy state, presumed by the
inationary scenario to be present at the beginning of the universe. The newly tutored
intuitions suggest that such a state has relatively high entropy. But, it could be argued, this
second impression is also mistaken, this time due to the neglect of the degrees of freedom
associated with the ination eld that drives ination in the early universe. And (the
argument continues) when these additional degrees of freedom are taken into account, the
pre-ination state will be seen to have low entropy. Thus, the threat that ination acts as
an anti-thermodynamic machine is undercut. But the price paid is that, by its own
standards, inationary cosmology fails to provide a satisfactory explanation of the post-
ination smooth state because it has to appeal to special pre-ination conditions for the
ination eld.
And in fact a number of critics of inationary cosmology have argued that the problem
of special initial conditions is not escaped because special conditions are needed for the
universe to enter an inationary era. Here is one version of the criticism that applies to a
universe that begins with a big bang and ends with a big crunch:
As the big crunch is approached, it seems overwhelmingly improbableand,
indeed, in apparent blatant contradiction with the second law of thermodynamics
that . . . a collapsing universe would undergo an era of deation just before the
big crunch. Thus, the region of nal data space that corresponds to a universe
that did not deate should have a much larger measure than a region corresponding
to a universe that did deate. But the time reverse of a collapsing universe that fails
to deate is, of course, an expanding universe that fails to inate. Thus, [since the
relevant laws are time reversal invariant] this argument strongly suggests that the
region of initial data space that fails to give rise to an era of ination has far larger
measure than the region that does not give rise to an inationary era, i.e., it is
overwhelmingly unlikely that ination did occur. (Hollands & Wald, 2002a, pp.
20452046).
And, one could add, if ination did take place, it was because of special pre-ination
conditions.
34
Needless to say, inationary cosmologists resist this conclusion, but in some
cases there is grudging recognition that work needs to be done to avoid it.
35
I do not pretend to be able to adjudicate such conicting claims. What I want to
emphasize at this juncture is that, despite the many condent pronouncements that
modern cosmology has delivered Boltzmanns sought-after low entropy state for the early
universe, the only extant grounds for this dictum come from intuition pumps and, as we
have seen, there are competing pumps that produce different verdicts. Since there are no a
priori reliable principles for identifying the correct pump, the way forward does not lie
in furiously cranking the handle of ones favorite pump or in producing ever more clever
pumps. The only responsible way to proceed is to study models where all the relevant terms
ARTICLE IN PRESS
34
For a similar argument, see Penrose (1986).
35
I interpret some of the remarks in Albrecht (2004) in this way.
can be precisely dened and calculations of the relevant quantities can be made. In the few
cases where this has been done, the results have a plague-on-all-houses character in that
what were taken to be clear questions about the likelihood of various states turn out to
have ill-dened answerswhich may help to explain why the intuition pumps do not yield
consistent results.
8. Model calculations
In order to study the issue of how probable ination is, Hawking and Page (1988)
investigated a FRW-F model in which the matter content is given by a minimally coupled
massive scalar eld F. Such a model is less than ideal since the FRW metric is
homogeneous and isotropic and, thus, does not provide a good test of inations ability to
smooth out lumpy states; but this fault is balanced by tractability. The state space X for
this system is four-dimensional and can be coordinatized using the scale factor a of the
FRW model, the eld F, and their respective conjugate momenta p
a
and p
F
. The equations
of motion form a constrained Hamiltonian system with the one and only constraint being
the vanishing of the Hamiltonian H= 0.
36
The three-dimensional subspace C X where
the constraint is satised is called the constraint surface. A reduced phase space, which is
free of gauge redundancy, can be formed by choosing a two-dimensional surface S that is
transverse to the dynamical trajectories on C. Then, as noted by Hawking and Page (1988),
the pullback of the (degenerate) symplectic form o = dp
a
. da dp
F
. dF denes a
volume measure
(2)
m on S that is invariant under dynamical evolution. Alternatively, as
noted by Hollands and Wald (2002b), an invariant volume measure
(3)
m can be dened on
C by the volume element
(3)
given by the condition dH.
(3)
=
(4)
, where
(4)
:=dp
a
.
da . dp
F
. dF is the Liouville volume element for X.
The upshot is a deterministic dynamical system. However, unlike the systems
encountered in classical mechanics, it cannot be deemed to be time translationally
invariant. Nor is there any reasonable sense in which the dynamics of this system can be
deemed ergodic, which undercuts one of the justications for taking the measure to be
interpreted as probability in the relative frequency sense. But these are minor worries
compared to the main one:
(2)
m(S) =
(3)
m(C) = o; i.e. neither of the natural invariant
measures normalizes. Since this result holds for the k = 1 FRW model, the non-
normalizability cannot be blamed on the innity of space implied by the k = 0 and 1
models. This non-normalizability obviously poses problems for assigning probabilities to
various macrostates.
But all is not lost. For any measurable M _ S, one of three possibilities holds with
respect to
(2)
m (see Hollands & Wald, 2002b): (i)
(2)
m(M)oo, (ii)
(2)
m(M) = o and
(2)
m(S M)oo, or (iii)
(2)
m(M) = o and
(2)
m(S M) = o. Now suppose that some
relevant coarse graining has been dened for S. With M being the region of S
corresponding to a macrostate m belonging to the coarse graining, set
(2)
Pr(m) = 0 in case
(i), set
(2)
Pr(m) = 1 in case (ii), and declare
(2)
Pr(m) undened in case (iii). Then it is easy to
show that
(2)
Pr is a (partial) probability measure on the coarse graining in the following
sense: 0 p
(2)
Pr(m) p1 if
(2)
Pr(m) is dened, and
(2)
Pr(m v m
/
) =
(2)
Pr(m)
(2)
Pr(m
/
) if
ARTICLE IN PRESS
36
The existence of this constraint indicates the presence of gauge freedom; see Earman (2003) for an
introductory account of these matters.
the region M M
/
corresponding to m&m
/
is null and if
(2)
Pr(m) and
(2)
Pr(m
/
) are
dened.
37
An exactly analogoussituation holds for
(3)
m and its associated
(3)
Pr dened on a
coarse graining of C.
In this setting the only thing that could be meant by the claim that a relevant initial
macrostate m
i
has low probability and correspondingly low Boltzmann entropy is that
(2)
Pr(m
i
) = 0 (or
(3)
Pr(m
i
) = 0). Although I do not know of any proof that this is not the
case, I conjecture that it is typically false because typically the third case obtains, i.e.
(2)
Pr(m
i
) and
(3)
Pr(m
i
) are not dened. Such ill-denedness is known to hold for the
probability of ination in the FRW-F model. In the best case scenario where the initial
macrostate turns out to have zero probability, the conditional probability assertions of the
form
(2)
Pr(=m
i
) = x or
(3)
Pr(=m
i
) = y are meaningless if conditional probability is given
its usual interpretation (viz. Pr(A=B) := Pr(A&B)= Pr(B)). Moreover, the only changes that
can take place in the Boltzmann entropy are maximal ip-ops corresponding to a change
from a zero-probability macrostate to a probability-one macrostate or vice versa.
Of course, it could be that the HawkingPage model is misleading and that the results
inimical to the Boltzmann program are artifacts of the idealizations of the model. But I
would expect that removing the idealizations will not remove the inimical results.
There is an historical irony here worth remarking. The opening salvo of Zermelos
attack on Boltzmann used Poincare s recurrence theorem. Part of Boltzmanns (1896, p.
218) response was to question the applicability of Poincare s recurrence theorem to the
universe; in particular, he deemed questionable the assumption that the space available
for motion, and the total energy [of the particles], are nite. If this assumption fails, then
so does the normalizability of the natural invariant measure, which is a necessary condition
of Poincare s theorem. By the same token, normalizability is also a necessary condition for
the applicability of a non-hamstrung version of Boltzmann entropy. As illustrated by the
model considered above, when the relevant dynamics is the Hamiltonian dynamics of eld
quantities, normalizability may fail even when the physical space available for motion is
nite.
In sum, those who advocate that the Past Hypothesis provides the answer to
Boltzmanns prayers proceed on the presumption that there is a well-dened thermo-
statistical mechanics for general relativity that incorporates the gravitational entropy of
the universe. In fact, no such theory exists, and the above discussion strongly suggests
that no such theory can exist, at least not for classical general relativity. It may be that a
quantum theory of gravity will come to the rescue, but at present that is only a pious hope.
9. Worse and worse . . .
For the sake of discussion, I want to assume for the moment that there is a well-dened
Boltzmann entropy of the universe, which includes the gravitational contribution to
entropy, and that the value of this quantity for the initial state of the universe is very
low. Does conditionalizing on this state solve Boltzmanns initial state and asymmetry
problems, as some proponents of the Past Hypothesis claim? The answer is negative for
several connected reasons.
First, there is no guarantee that there will be a monotonic increase in entropy up to the
present and that, nevertheless, the present value of the entropy is lowsuch a result
ARTICLE IN PRESS
37
However,
(2)
Pr need not be countably additive.
depends on the details of the initial state and the dynamics of the particular cosmological
model. The importance of the details of the initial state is emphasized in a second, related
point.
If the just-so story told by the advocates of the Past Hypothesis were correct, then it
would seem that the story would be even better if the entropy of the initial state of the
universe were even lower. The way to make it lower is to make the state more uniform. But
with a completely uniform initial state, the structure formation would not have taken
place, and there would be no subsystems in which statistical-thermodynamics of the
familiar form would apply. Keeping ones eye on what needs to be explainedi.e. why
statistical-thermodynamics of the familiar form works as well as it does for the systems of
interest to usleads to a third point.
To repeat once more, Boltzmann entropy is a global quantity that characterizes the
macrostate of the systemin this case, the system of the entire universe. That the value of
this quantity is low places very little constraint on the state of a small subsystem of the type
whose behavior we are interested in explaining.
38
In the present instance this is doubly so
since the Boltzmann entropy of the initial state of the universe is low primarily because of
the gravitational contribution, whereas that contribution is irrelevant for the kinds of
subsystems of interest to us. The length scale of typical thermodynamic systems of interest
to us is below the Jeans length, and as a result self-gravitation is unimportant.
39
For
example, for any normal box of gas considered in laboratory experiments, the pressure of
the gas is sufcient to prevent gravitational self-collapse of the gas molecules. Nor is it
plausible that the entropy increases we perceive in typical thermodynamic systems is the
result of an increase in gravitational entropy that is transferred to the non-gravitational
degrees of freedom; for on the time scales of typical thermodynamic observations and
inferences, the values of the variables that characterize the gravitational degrees of freedom
are effectively constant.
I am not denying the correct (if virtually tautologous) component of the Past
Hypothesis; namely, the explanation of the observed thermo-statistical asymmetries must
rely on initial/boundary conditions which, presumably, can be traced back to the early
universe. In broad strokes, the explanation would go as follows. Immediately after the big
bang (or, if inationary cosmology is correct, immediately after the inationary era), the
universe was in a thermalized state that was very homogeneous and isotropic. But not
completely homogeneous and isotropic. The small density perturbations were amplied by
the action of gravitation and grew into the structures that populate the present universe
stars, galaxies, galactic clusters, . . . as well as more exotic objects, such as black holes. In a
subset of these structures the conditions are such that (for the reasons given immediately
above) one does not have to worry about any gravitational contribution to the entropy,
and the ordinary concept of Boltzmann entropy applies. And a subsubset of these
structures will be in thermodynamic disequilibrium with low Boltzmann entropy. That our
world belongs to this subsubset can be justied on anthropic grounds. Nor does the need
to resort to anthropic considerations vitiate the explanation on offer; for the nature of the
ARTICLE IN PRESS
38
Winsberg (2004) has argued that the hypothesized low entropy state of the early universe places no effective
constraint on the microstate of a relatively small and isolated thermodynamic system, because the past
interactions of the system with the rest of the universe effectively randomizes this microstate.
39
For a box of gas, the Jeans length is dened by L
J
:=
pv
2
s
=Gr
p
, where v
s
is the speed of sound, r is the energy
density, and G is the gravitational constant.
initial state and the dynamics together guarantee that the subsubset in question will be
well-populated and that the thermo-statistical asymmetries of the member systems will be
concordant.
The contours of this explanation sketch partially trace the just-so story that is part of the
Past Hypothesis. But the departures from the story told by the advocates of the Past
Hypothesis should be noted; in particular, the new story does not invoke the just-so aspects
of the Past Hypothesis that appeal to the entropy of the universe, and it recognizes the
need for an anthropic element to the story.
Although the new story is an improvement on the Past Hypothesis story, it still contains
a serious lacuna. Most of the subsystems we actually encounter are never completely
isolated, e.g., a box of gas may be thermally isolated, but the thermal isolation does not
shield the gas molecules from the gravitational inuence of external bodies. This coupling
to outside systems is sometimes thought to be helpful to the Boltzmann program by
providing a kind of stirring that, intuitively at least, promotes effective mixing for the
system of interest (recall the discussion of Section 2). The trouble is, however, that the
Boltzmann apparatus developed above and all the resultsactual and desireddepend on
the assumption of a closed dynamical system. Nevertheless, for a subsystem of a dynamical
system (X; m; B; f
t
) one could dene a Boltzmann entropy as follows. Marginalize the
measure m to the degrees of freedom characterizing the subsystem to produce a measure m;
choose an appropriate coarse graining of macrostates { m
b
] for the subsystem; and take
K log( m(

M)) as the Boltzmann entropy of the macrostate m c { m
b
], where

M is the region
of the state space

X of the subsystem corresponding to m. Unless the degrees of freedom of
the subsystem do not interact with those of the rest of the system, f
t
projected down to

X
will generally not dene a deterministic ow on

X, and m will not be preserved under
dynamical evolution; and if not, results about the behavior of the Boltzmann entropy of
the subsystem cannot be derived by applying the Boltzmann apparatus to the subsystem.
Nor does an increase in the Boltzmann entropy for the total dynamical system (X; m; B; f
t
)
provide any guarantee of an increase in the entropy of the above dened Boltzmann
entropy of the subsystem. For one thing, the increase in the entropy for the total system
may be largely due to the behavior of a class of degrees of freedom that is irrelevant to the
entropy of the subsystem, and for another, the increase in the Boltzmann entropy for the
total system can be due to the establishment of correlations among the subsystems rather
than increases in the Boltzmann entropies of the subsystems.
In sum, the Past Hypothesis, even if true, does not explain why ordinary
thermodynamics works as well as it does for the types of systems of interest to us, and
the widespread celebration of this hypothesis has been counter-productive in obscuring the
hard work that still needs to be done to secure a satisfactory explanation.
10. But not to worry . . . but get to work
The combination of the worry of Section 8that the Boltzmann entropy of the early
universe is ill-denedand the worry of Section 9that even if the Boltzmann entropy of
the early universe is well-dened and has a low value, it does not sufce to ground the
temporally asymmetric behavior for the (sub)systems we care aboutshould create a
serious concern for those who believe that our knowledge of the past depends in essential
ways on thermodynamic reasoning and that Boltzmanns apparatus is needed to explain
the extent to which this reasoning is sound. I want to suggest that the importance of
ARTICLE IN PRESS
statistical mechanics in grounding our knowledge of the past should be downgraded,
especially when Boltzmann entropy is being relied upon to do the grounding, but more
generally as well. The suggestion can be eshed out by a series of remarks, the last of which
will offer an alternative to Boltzmanns program for explaining the approximate validity of
thermodynamics.
First, as a preliminary, I want to urge that philosophers have been too credulous in
buying into one crucial aspect of Boltzmanns attempt to resolve the asymmetry problem.
Begin by acknowledging the obvious; namely, that if time symmetric information
about the present state of affairs is plugged into time reversal invariant laws, then only
time symmetric predictions can emerge. But do not swallow the claim that the only
way to get out time asymmetric predictions is with the help of additional information or
posits about the distant past state of the system, whether in the form of the entropy value
of this state or some other. For we can, and do, get temporally asymmetric predictions
and retrodictions by plugging in time asymmetric information about the present and
near past states that comes, for example, from a short temporal sequence of observa-
tions (see Section 4). This procedure does presume that our short-term memories
are veridical. But without this presumption both common-sensical and scientic knowl-
edge would be impossible, and issues about inferences to the more remote past would be
moot.
Second, note that if the Boltzmann apparatus is essential in generating knowledge of the
part of the past that lies beyond the reach of our short-term memories and if the only way
to solve the asymmetry problem is with the help of a low entropy initial state, then on
pain of circularity the knowledge of this state can only be presuppositional. But the
knowledge of the early universe generated by modern cosmology is not presuppositional in
this sense, though it does presuppose a number of theories. More generally, while much of
our knowledge of the past is based on statistical reasoning broadly construed, much of it is
not in need of grounding or justication from the Boltzmann version of statistical
mechanics. For instance, we know with more precision and certainty the state of the
universe at the time of last scatteringover 13 billion years agothan we know the
state of the earth 100,000 years ago. This knowledge of the early universe comes from
observing it with optical and radio telescopes. Ellis, Nel, Maartens, Stoeger, and
Whitman (1985) have shown that the eld equations of general relativity theory plus data
that is in principle observable on a past null cone are together sufcient to determine the
spacetime geometry and the matter content on and in the interior of this cone. To be sure,
actual data is noisy and needs to be massaged by various statistical procedures and ltered
through a number of scientic theories, both low level and high level, in order to produce
useful conclusions; and, of course, the reliability of these inferences presupposes that we
are not being deceived by Cartesian demons or conspiratorial conditions that produce
systematically misleading impressions. The point is not that empirical knowledge does not
rely on substantial presuppositionsit most certainly doesbut that the validity of
Boltzmannian statistical mechanics and the Past Hypothesis do not gure prominently
among the presuppositions.
The third remark calls attention to the fact that many of our inferences to the past do
not concern closed systems to which Boltzmannian statistical mechanics might apply;
rather, the object is to make probable inferences about interventions on the system. The
hackneyed example of footprint-like indentations on a beach serves as a useful illustration.
The outline of the common sense inference that the indentations were made by a creature
ARTICLE IN PRESS
walking across the beach can be reconstructed as follows. First we ask, supposing that the
sand were an isolated system, how long would we have to wait on average for a
spontaneous uctuation to produce a pattern of indentations similar to the ones observed?
By analogy with simpler cases where the calculations have been carried out we are justied
in concluding that the waiting time is longer than the age of the universe. Thus, it would be
an extraordinary coincidence if we happened to be observing the beach shortly after such a
uctuation occurred. Next, we reect on the fact that the beach is not a closed system but is
acted on by the tide and the winds. But again, we estimate that the likelihood that these
forces would by chance conspire to produce the observed pattern is so small as to be
ignorable. Finally, we use plausibility considerations, some of a statistical variety and
others of a kind that subjective Bayesians would incorporate in prior probabilities, to
exclude various possibilities, such as a visitation by an alien spacecraft with foot-shaped
landing gear. And, in Sherlock Holmes fashion, after eliminating all of the other
contenders that come to mind, we arrive at the Robinson Crusoe solution. This crude
outline begs for more detail, but it seems wholly implausible that any useful precisication
will result from attempts to bring to bear Boltzmann entropy or from the investigation of
the initial conditions of the universe.
In sum, the notion that our knowledge of the past would crumble unless the Boltzmann
program can be made to work with the help of the Past Hypothesis is a kind of hysteria
that can only be generated by the febrile minds of philosophers who have become too
enamored with the Boltzmann apparatus. Nevertheless, it is certainly the case that some of
our reasoning about the past does need to be grounded on thermo-statistical physics. If the
explanation of why this physics works as well as it does is not to be based on Boltzmann
entropy and the Past Hypothesis, it is a fair challenge to ask how an alternative
explanation might go. Here I can offer only a sketchy suggestion.
Begin by observing that the temporally asymmetric behavior we want to explain can be
characterized independently of the concept of Boltzmann entropy. In particular, what we
want is an explanation of the approximate validity of the Poisson heat equationwhich
describes the phenomenology of heat diffusionand of the transport equationswhich
describe the phenomenology of mass diffusion, viscosity, and thermal conductivity. And
we want the explanation to apply to relatively small subsystems, such as a box of gas
or an ice cube in a glass of lukewarm water. The fact that such subsystems are typically
not dynamically isolated from the environment may play an important role in the sought
after explanation.
40
The coupling to the environment may help to drive a large basin
of states in the subsystem state space to follow trajectories that fulll to good
approximation the transport equations. Needless to say, some sort of Past Hypothesis
in the form of a restriction on initial conditions will have to be used. But it is to be
hoped the restriction can be justied without having to appeal to dubious and perhaps
ill-dened notions such as the Boltzmann entropy of the universe. In any case, this
program is best pursued in the Gibbsian rather than the Boltzmannian approach to
statistical mechanics.
41
ARTICLE IN PRESS
40
For some attempts to implement this idea, see Bergmann and Lebowitz (1955) and Blatt (1959).
41
In fairness to Boltzmann, it should be noted that he was one of the rst writers to emphasize the impact of the
environment on a thermodynamical system; for example, Boltzmann (1871) bases the ergodic hypothesis on the
multiplicity of outside forces. But Boltzmann did not seem to realize that this insight is not easily exploited within
his version of statistical mechanics.
11. From the ill-dened to the ill-considered and the ridiculous
In his second reply to Zermelo, Boltzmann considered a proposal for avoiding the posit-
your-way solution to the asymmetry problem. In reading the passage quoted below, recall
that Boltzmann is hypothesizing that the universe is in thermal equilibrium but that within
it are many worlds in non-equilibrium.
For the universe as a whole, the two directions of time are indistinguishable, just as in
space there is no up or down. However, just as at a particular place on the earths
surface we can call down the direction toward the center of the earth, so a living
being that nds itself in such a world at a certain period of time can dene the time
direction as going from the less probable [lower entropy] to the more probable
[higher entropy] states (the former will be the past and the latter will be the
future) and by virtue of this denition he will nd that this small region, isolated
from the rest of the universe, is initially always in an improbable [low entropy]
state. (Boltzmann, 1897b, p. 242)
It is not clear how seriously Boltzmann took this proposal for a dene-your-way
solution to the asymmetry problem. For example, in the paragraph immediately following
the quoted remarks he says that Whether one chooses to indulge such speculations is a
matter of taste (Boltzmann, 1897b, p. 243). Except for the fact that the proposal is
repeated in nearly identical language in the Lectures on Gas Theory (Boltzmann,
18961898, p. 447), one could ignore it as an expedient for escaping the embarrassment,
revealed in his exchanges with Zermelo, of having to posit what is needed for a solution to
the initial state and asymmetry problems.
42
Commentators tend to take Boltzmanns proposal seriously, and those who do are
deeply divided about its viability. Reichenbach (1971, p. 128) saw it as one of the keenest
insights into the problem of time, while Popper (1974, p. 128) found it staggering in its
boldness and beauty but ultimately untenable. My assessment lies more with Popper
than with Reichenbach, but instead of boldness and beauty, I nd muddles. I will
distinguish ve interpretations of Boltzmanns proposal and will argue that all of them
succumb to one of two ills: they are either tenable but uninteresting or else interesting but
untenable.
The rst interpretation (I1) would have it that there is an objective directionality for time
that is independent both of observers and of the entropic behavior of physical systems but
that the subjective perception of time order for critters such as us is enslaved to the entropy
gradient of their environment.
43
In more detail, consider a time translationally invariant,
deterministic dynamical system (X; m; B; f
t
) that describes not only the physics of gases
and liquids but also the physics of critters like us, and stipulate that the mental states of
these critters supervene on their physical states. Suppose that in some time interval
t
i
ptpt
f
of a dynamically possible history generated by f
t
(x), x c X, as t ranges from o
to o, the Boltzmann entropy is monotonically increasing. Then in the time interval
ARTICLE IN PRESS
42
In Curds (1982, p. 282) opinion, Boltzmanns proposal was introduced to appease Zermelos demand for a
physical explanation of Assumption A.
43
Since the directionality of time is used in a number of different ways in the literature, it is necessary to say
precisely what one means by this phrase. I will do that below. But for the moment the reader can take the existence
of an objective directionality for time to be equivalent to the existence of an objective, observer independent time
order of (relatively timelike) events. Nuances will be added at the appropriate place below.
t
f
ptp t
i
of the time reversed history generated by f
T
t
(x) the entropy is monotonically
decreasing. And using time translation invariance, we can shift events in f
T
t
(x) by an
amount (t
f
t
i
) so that the interval of monotonic entropy decrease coincides with
t
i
ptpt
f
. Boltzmanns proposal (on its present interpretation) implies that the critters in
the time reversed (and time translated) history have the subjective impression that entropy
is increasing as time increases from t
i
to t
f
and, hence, that they perceive the same series of
macrostates as the critters in the original history but in reverse temporal order.
44
So (the
story goes) critters such as us will not be able to distinguish between these two histories
since both will think that they are inhabiting the one in which entropy is increasing. I think
that Maudlins (2004) reaction to such stories is exactly right: we do not have to wait for
neuro-psychology to become a mature science to know that this just-so story is just as false
as it can bethe critters in the time reversed history will not have the subjective impression
of increasing entropy; rather they will not have any coherent temporal perceptions at all.
To arrive at the second and third interpretations, focus not on subjective temporal
perceptions but on objective temporal structures, and take Boltzmanns proposal to be that
objective temporal directionality is enslaved to the entropy gradient. Here two versions can
be distinguished, depending upon whether the directionality at issue is the directionality of
time itself or of physical processes within time. Consider rst the second reading, yielding
interpretation (I2). If the temporal directionality of physical processes is identied with the
monotonic increase in some genuine physical magnitude
45
assigned to these processes, then
in Boltzmanns day it would have been plausible to think that such behavior has to be
identical with, or parasitic upon, entropy increase. Modern cosmology provides an
alternative that is meaningful even when the concept of Boltzmann entropy of the universe
is not; namely, the expansion of the universe. Up to the present, this expansion has been
monotonic, and if the equation of state for the dark energy that is driving the current
acceleration of expansion does not change, the universe will continue to expandad
innitum in the cases where the dark energy is supplied by a positive cosmological constant
or by quintessence, and for a nite time in the case where the dark energy is supplied by
phantom matter until the density of this matter becomes innite and the universe ends in a
big splat.
46
In addition to the expansion of the universe, there are other non-
thermodynamic arrows of time, such as the electrodynamic arrow, making Boltzmanns
concentration on the entropic arrow seem parochial.
Now consider the alternative reading that would have it that the directionality of time
itself is enslaved to the entropy gradient, yielding (I3). I will offer a dilemma for (I3)
depending on two different ways to understand time reversal invariance: taking the time
reversal transformation as either an active or a passive transformation. In preparation, it is
necessary to be more precise about the presuppositions of the treatment of time reversal
invariance given in Section 4. Suppose that the state of a dynamical system is given by the
values of particle and eld variables and, possibly, the values of their time derivatives, on
some time slice of spacetime. Then consider what properties the spacetime must have to
make the denition of time reversal invariance offered in Section 4 meaningful. First, the
ARTICLE IN PRESS
44
This is assuming that if M is the region of state space corresponding to a macrostate m, then
R
M = M; see the
discussion in Section 4.
45
Some weasel words here are needed to rule out monotonic Cambridge change.
46
See Caldwell, Kamionkowski, and Winberg (2003). Before the big splat there will be a big rip in which
gravitationally bound matter will be torn apart.
spacetime must be time orientable.
47
Any such spacetime admits two possible time
orientations
48
; with one of these orientations singled out, a relation of temporal precedence
5is dened on spacetime points by the condition that p
1
5p
2
iff there is a future directed
timelike curve from p
1
to p
2
. Requiring that there are no closed timelike curves assures that
5 is not only transitive but asymmetric and, thus, is an order relation.
49
But still more is
required to guarantee that there is a global time function in the guise of a smooth function
t from the spacetime to R such that t(p
1
)ot(p
2
) iff p
1
op
2
.
50
With all of these conditions on
spacetime structure satised, the denition of time reversal invariance offered in Section 4
can be applied.
That denition implicitly assumes the active reading of time reversal: i.e. for any x c X
the dynamical trajectories generated by f
t
(x) and f
T
t
(x) as t ranges from o to o
correspond to physically distinct histories
51
; and what time reversal invariance of the laws
of motion means is that one of the histories satises the laws if and only if the other does.
Take a case where in f
t
(x) the Boltzmann entropy S
B
(t) increases monotonically with time.
Then in f
T
t
(x) the Boltzmann entropy S
B
(t) decreases monotonically with time. But
according to (I3) this latter history is not a live possibility, which is just to say that (I3)
must reject the active reading of time reversal invariance.
On the passive reading of time reversal invariance, the dynamical trajectories generated
by f
t
(x) and f
T
t
(x) as t ranges from o to o are merely different descriptions of the
same history. This gauge interpretation of the time reversal operation might seem to be
in blatant contradiction with the presuppositions of the denition of time reversal
invariance. However, if all of the fundamental laws are time reversal invariant, some of the
structures enumerated above can be regarded as ladders that can be kicked away once they
have served their purpose of getting us to the distinction between time reversal invariant
and non-time reversal invariant laws. But notice that this Wittgensteinian ladder trick
amounts to regarding the relation of temporal precedence 5as a mere auxiliary device and
to denying that events, stripped of their gauge-dependent dressing, are ordered as to earlier
and later.
52
But this removes all literal meaning from (I3).
ARTICLE IN PRESS
47
A spacetime, classical or relativistic, is time orientable just in case it admits a continuous, non-vanishing
timelike vector eld. For a classical spacetime a timelike vector is one that is oblique to the planes of absolute
simultaneity, while for a relativistic spacetime it is a vector that lies within the local null cone. Any spacetime that
has not been subjected to nasty Mo bius type identications admits such a vector eld; for relativistic spacetimes
the relevant result is that any simply connected spacetime admits the required vector eld.
48
For sake of deniteness, concentrate on relativistic spacetimes. In a time orientable spacetime, choose some
continuous non-vanishing, timelike vector eld. Call a timelike (tangent) vector at a spacetime point p future
pointing (respectively, past pointing) just in case it points into the same lobe of the local null cone at p as the vector
from the reference eld. It is easy to see that the freedom in the choice of the reference elds leads to exactly two
possible choices for the division of timelike vectors into past and future pointing. The choice of one of the two
constitutes a time orientation.
49
If there are closed timelike curves, then although there is a globally consistent directionality for time, there is
no globally consistent time order. If there are no closed timelike curves, then the existence of a time direction or
orientation is equivalent to the existence of a time order. I ignore this distinction since for present purposes it
makes no difference.
50
For a relativistic spacetime to admit a global time function it must be stably causalintuitively, there is a
widening the null cones by some nite amount that does not result in closed timelike curves.
51
Except in degenerate cases, e.g. a single particle which is always at rest.
52
Although, assuming that the spacetime admits a global time function, they are ordered by the three-place
relation of temporal betweenness Bet(p
1
; p
2
; p
3
) which, when the relation 5of temporal precedence is added to the
spacetime structure, is interpreted to mean that either p
1
5p
2
5p
3
or p
3
5p
2
5p
1
.
A fourth interpretation (I4) emerges from the passive reading of the time reversal
symmetry. Take Boltzmanns proposal to mean that we should view the time reversal
symmetry as a gauge symmetry and that we should work in the Boltzmann gauge in
which the relation of temporal precedence 5 is assigned in such a way that entropy
increases with increasing 5order. As such, (I4) is harmless but insignicant in just the way
the proposal to work in the Lorentz gauge in electromagnetism is harmless but
insignicant. (I4) can be made more signicant by combining it with the idea that the
Boltzmann gauge is to be preferred, since it corresponds to our subjective impressions of
time order. Here I have an objection similar to the one made against (I2); namely, the
augmented (I4) entails empirically false predictions. According to the augmented (I4), since
the dynamical trajectories generated by f
t
(x) and f
T
t
(x) as t ranges from o to o are
merely different descriptions of the same physics, the critters described by f
t
(x) and f
T
t
(x)
must have the same temporal experiences, if they have any at all. My conjecture is that this
prediction is false and, thus, that in order to obtain an explanation of the differences in
temporal experiences of critters in what (I4) regards as the physically the same history
under different gauge dressings, it is necessary to use the active interpretation of time
reversal invariance.
It is also worth noting that there is now strong evidence that the presupposition
of (I3) and (I4)namely, that all of the fundamental laws of physics are time reversal
invariantis false. This fact is often brushed aside on the grounds that the non-time
reversal invariant laws concern the weak interactions of elementary particles
53
and that the
exotic processes created in particle accelerators to demonstrate the failure of time reversal
invariance in these interactions have no connection with the temporal asymmetries of
ordinary thermodynamic processes or with our temporal perceptions. Even if this is
correctand I think the rush to judgment could prove prematurethe existence of
non-time reversal invariant laws does have an obvious and important consequence for
the issues at hand; namely, the time reversal operation cannot be interpreted as a
gauge symmetry. And in order to distinguish between which of two processes that are
time reverse images of each other is physically possible and which is not, there must be,
at least locally in spacetime, a time orientation; and if the physical laws governing
these process are the same in all regions of spacetime, then that orientation must be
globally dened.
54
A fth interpretation (I5) of Boltzmanns proposal has been offered by Curd (1982).
Without going into the details, the basic idea is that entropy increase in branch systems
satisfying appropriate boundary conditions serves as a criterion for the relation of
temporal precedence. If criterion simply means that entropy increase serves as a reliable
indicator of temporal precedence, then (I5) is surely acceptable to the extent that the
(temporally asymmetric) Second Law has statistical validity; indeed, (I5) is just a statement
of a consequence of this statistical validity. But there are many other criteria in the sense
of reliable indicators of temporal precedence, and no reason has been given for why
entropy increase should take pride of place among all these criteria, especially since some
of them (such as the expansion of the universe) are more reliable indicators. On the other
hand, if criterion means something stronger, e.g. that entropy increase is part of the
ARTICLE IN PRESS
53
See Sachs (1987, Chapter 9) for a summary of the evidence for the violation of time reversal invariance.
54
For the argument, see Earman (2002). One possible escape is to redene the time reversal operation by the
combination CPT, where T is the usual time reversal operation, and to appeal to CPT invariance.
meaning of temporal precedence, or that it is physically necessary for temporal precedence,
or that it holds the key to our perceptions of temporal order, then for all of the reasons
given above I would reject (I5).
Once philosophers get hold of a muddle, they are reluctant to let it go. But the
philosophy of time would be better off if the muddle started by Boltzmanns denitional
ploy were bid adieu.
12. Conclusion
The following line of reasoning, although a caricature, helps to explain the tenor of
recent discussions in the philosophical literature: many of the temporal asymmetries that
underlie the direction of time and undergird our knowledge of the past invoke entropy in
essential ways; this entropy is to be understood in Boltzmanns sense; hence, Boltzmanns
program for explaining the statistical validity of the Second Law must be made to work;
and to make it work requires a low entropy initial state for the universe; thus, it will be
found that cosmology ts the bill, if not by honest toil then by hand waving, intuition
pumps, or by ignoring merely technical considerations.
On the contrary, I claim, there is no must about it. Even before the initial state and
asymmetry problems are encountered, there are serious difculties with the logic of
Boltzmanns explanation of the Second Law. Moreover, there are good reasons for
thinking that the alleged solutions to these problems that invoke the Past Hypothesis is
badly awed. The initial state of the universe can have a low Boltzmann entropy only
because of the gravitational contribution to the entropy budget. But there does not exist at
present, and probably cannot exist, a Boltzmannian statistical mechanics for gravitating
systems as described by classical general relativity theory, and as a result, the
Boltzmannian entropy of the universe is an ill-dened concept. Moreover, even if the
entropy of the initial state of the universe had a well-dened, low value, this would not
sufce to explain why thermodynamics works as well as it does for the kinds of systems we
care about.
These negative verdicts on Boltzmanns program do not, I contend, threaten a disaster
for the inferential practices we use to generate conclusions about the past. But, if correct,
they do require an alternative to Boltzmanns program for understanding the statistical
validity of thermodynamics. I have given only the sketchiest indications of how this
alternative program might work. The difculties for orthodoxy seem to me sufciently
great that more investigation of heterodox approaches is in order. Extant heterodoxies
include the idea that thermodynamic asymmetries have to be grounded on quantum
considerationsperhaps using the state vector reduction scheme of Girardi, Rimini, and
Weber (as favored by Albert, 2000, Chapter 7) or by using the phenomenon of quantum
decoherence (as favored by Hemmo & Shenker, 2001). Since the world is at base quantum
mechanical, the explanation of thermodynamical asymmetries must, of course, be statable
in quantum mechanical terms. But I would be surprised if the asymmetries at issue were
inexplicable without the essential use of quantum concepts.
55
ARTICLE IN PRESS
55
I also have qualms about the mentioned mechanisms: I doubt that there is a satisfactory relativistically
invariant account of state vector reduction; and while I have no doubt that quantum decoherence is an important
part of the emergence of the classical world from the quantum mechanics, I do doubt that decoherence by itself
can explain the emergence.
Acknowledgments
I am very grateful to Jos Ufnk for number of helpful suggestions on a earlier draft of
this paper; needless to say, this does not imply that he shares the conclusions I advance.
The comments of two anonymous referees are also gratefully acknowledged.
References
Albert, D. (1994). The foundations of quantum mechanics and the approach to thermodynamic equilibrium.
British Journal for the Philosophy of Science, 45, 669677.
Albert, D. (2000). Time and chance. Cambridge, MA: Harvard University Press.
Albrecht, A. (2004). Cosmic ination and the arrow of time. In J. D. Barrow, P. C. W. Davies, & C. L. Harper
(Eds.), Science and ultimate reality: Quantum theory, cosmology, and complexity (pp. 363401). Cambridge,
MA: Cambridge University Press.
Albrecht, A., & Sorbo, L. (2004). Can the universe afford ination? Physical Review D, 70, 063528-1-10. hep-th/
0405270.
Barrow, J., & Tipler, F. (1986). The anthropic cosmological principle. Oxford: Oxford University Press.
Bergmann, P. G., & Lebowitz, J. L. (1955). New approach to nonequilibrium processes. Physical Review, 99,
578587.
Blatt, J. M. (1959). An alternative approach to the ergodic problem. Progress of Theoretical Physics, 22,
745756.
Boltzmann, L. (1871). Einige allgemeine Sa tze u ber Wa rmegleichgewicht. Sitzungsberichte der Kaiserlichen
Akademie der Wissenschaften. Mathematisch-Naturwissenschaftliche Klasse, Wien, 63, 679711. Reprinted in
F. Haseno hrl (Ed.), Ludwig Boltzmann: Wissenschaftliche Abhandlungen (Vol. 1) (pp. 259287). New York:
Chelsea Publishing Co., 1968.
Boltzmann, L. (1895). On certain questions of the theory of gases. Nature, 51, 413415.
Boltzmann, L. (1896). Entgegung auf die wa rmetheoretischen Betrachtungungen des Hrn. E. Zermelo. Annalen
der Physik, 57, 773784. (S. G. Brush, English Transl.) Kinetic theory (Vol. 2) (pp. 218228). New York:
Pergamon Press, 1966.
Boltzmann, L. (18961898). Lectures on gas theory. (S. G. Brush, English Transl.). Berkeley, CA: University of
California Press, 1964.
Boltzmann, L. (1897a). U
ber einige meiner weniger bekannten Abhandlungen u ber Gastheorie und deren
Verha ltnis zu derselben. Verhandlungen der Gesellschaft Deutscher Naturforscher und A
rzte, 69, 1926.

Reprinted in F. Haseno hrl (Ed.), Ludwig Boltzmann: Wissenschaftliche Abhandlungen (Vol. 3) (pp. 598698).
New York: Chelsea Publishing Co., 1968.
Boltzmann, L. (1897b). Zu Hr. Zermelos Abhandlung U
ber die mechanische Erkla rung irreversibler Vorga nge.

Annalen der Physik, 60, 392398. (S. G. Brush, English Transl.) Kinetic theory (Vol. 2) (pp. 238245). New
York: Pergamon Press, 1966.
Boltzmann, L. (1904). On statistical mechanics. Popula re Schriften, Essay 19. Reprinted in B. McGinness
(Ed.), Ludwig Boltzmann: Theoretical physics and philosophical problems (pp. 158172). Dordrecht: D. Reidel,
1974.
Bricmont, J. (1996). Science of chaos or chaos of science? In P. Gross, N. Levitt, & M. Lewis (Eds.), The ight
from science and reason (pp. 131175). New York: New York Academy of Sciences.
Bronstein, M., & Landau, L. (1933). On the second law of thermodynamics and the universe. Reprinted in English
translation in D. Ter Harr (Ed.), Collected papers of L. D. Landau (pp. 6972). New York: Gordon and
Breach, 1967.
Brush, S. G. (1966). Kinetic theory (Vol. 2). New York: Pergamon Press.
Brush, S. G. (1975). The kind of motion we call heat (Vols. 1, 2). Amsterdam: North-Holland.
Caldwell, R. R., Kamionkowski, M., & Winberg, N. N. (2003). Phantom energy and cosmic doomsday. Physical
Review Letters, 91, 071301 astro-ph/0302505.
Callender, C. (2001). Thermodynamic asymmetry in time. Stanford encyclopedia of philosophy. http:/
www.plato.stanford.edu/entries/time-thermo/).
Callender, C. (2004). There is no puzzle about the low-entropy past. In C. Hitchcock (Ed.), Contemporary debates
in philosophy of science (pp. 240255). London: Blackwell.
ARTICLE IN PRESS
Carroll, S. (2004). Why is the universe accelerating? In W. L. Friedman (Ed.), Measuring and modeling the
universe. Cambridge, MA: Cambridge University Press astro-ph/0310342.
Carroll, S., & Chen, J. (2004). Spontaneous ination and the origin of the arrow of time. hep-th/0410270.
Curd, M. (1982). Popper on the direction of time. In R. Sexl, & J. Blackmore (Eds.), Ludwig Boltzmann,
internationale Tagung: anla sslich des 75. Jahrestages seines Todes, 5.-8. September 1981: ausgewa hlte
Abhandlungen (pp. 263303). Graz: Akademische Druck-u. Verlagsanstalt.
Dyson, L., Kleban, M., & Susskind, L. (2002). Disturbing implications of a cosmological constant. Journal of
High Energy Physics, 0210, 011.
Earman, J. (2002). What time reversal invariance is and why it matters. International Studies in Philosophy of
Science, 16, 245264.
Earman, J. (2003). Tracking down gauge: An ode to the constrained Hamiltonian formalism. In K. Brading, & E.
Castellani (Eds.), Symmetries in physics: Philosophical reections (pp. 140162). Cambridge, MA: Cambridge
University Press.
Earman, J., & Re dei, M. (1996). Why ergodic theory does not explain the success of equilibrium statistical
mechanics. British Journal for the Philosophy of Science, 47, 6378.
Ellis, G. F. R., Nel, S. D., Maartens, R., Stoeger, W. R., & Whitman, A. P. (1985). Ideal observational
cosmology. Physics Reports, 124, 315417.
Feynman, R. P. (1994). The character of physical law. Cambridge, MA: MIT Press.
Gallavotti, G. (1999). Statistical mechanics: A short treatise. New York: Springer.
Goldstein, S. (2001). Boltzmanns approach to statistical mechanics. In J. Bricmont, et al. (Eds.), Chance in
physics: Foundations and perspectives, Lecture notes in physics (Vol. 574) (pp. 3954). New York: Springer.
Hawking, S. W., & Page, D. N. (1988). How probable is ination? Nuclear Physics B, 298, 789809.
Hemmo, M., & Shenker, O. (2001). Can we explain thermodynamics by quantum decoherence? Studies in History
and Philosophy of Modern Physics, 32, 555568.
Hollands, S., & Wald, R. M. (2002a). An alternative to ination. General Relativity and Gravitation, 34, 20432055
gr-qc/0205058.
Hollands, S., & Wald, R. M. (2002b). Comment on ination and alternative cosmology. hep-th/0210001.
Knott, C. G. (1911). Life and scientic work of Peter Guthrie Tait. Cambridge, MA: Cambridge University Press.
Lebowitz, J. L. (1993). Macroscopic laws, microscopic dynamics, times arrow and Boltzmanns entropy. Physica
A, 194, 127.
Lebowitz, J. L. (1999). Statistical mechanics: A selective review of two central issues. Reviews of Modern Physics,
71, S346S357.
Maudlin, T. (2004). On the passing of time, preprint.
Malament, D. (2004). On the time reversal invariance of classical electromagnetic theory. Studies in History and
Philosophy of Modern Physics, 35, 295315.
Penrose, R. (1979). Singularities and time asymmetry. In S. W. Hawking & W. Israel (Eds.), General relativity: An
Einstein centenary (pp. 581638). Cambridge, MA: Cambridge University Press.
Penrose, R. (1986). Review of G. W. Gibbons, S. W. Hawking, & S. T. C. Siklos (Eds.), The very early universe.
Cambridge, MA: Cambridge University Press. Observatory 106, 2021.
Penrose, R. (1989). The emperors new mind: Concerning computers, minds, and the laws of physics. Oxford: Oxford
University Press.
Penrose, R. (2004). The road to reality: A complete guide to the laws of the universe. London: Jonathan Cape.
Popper, K. (1974). Boltzmann and the arrow of time. In P. A. Schilpp (Ed.), The philosophy of Karl Popper
(pp. 124129). La Salle, IL: Open Court.
Popper, K. (1981). Quantum theory and the schism in physics. Totowa, NJ: Roman and Littleeld.
Price, H. (1996). Times arrow and Archimedes point. New York: Oxford University Press.
Price, H. (2002). Boltzmanns time bomb. British Journal for the Philosophy of Science, 53, 83119.
Price, H. (2004). On the origins of the arrow of time: Why there is still a puzzle about the low-entropy past. In C.
Hitchcock (Ed.), Contemporary debates in philosophy of science (pp. 219239). London: Blackwell.
Reichenbach, H. (1971). The direction of time. Los Angeles: University of California Press.
Sachs, R. G. (1987). The physics of time reversal. Chicago: University of Chicago Press.
Sklar, L. (1993). Physics and chance: Philosophical issues in the foundations of statistical mechanics. Cambridge,
MA: Cambridge University Press.
Sklar, L. (1995). The elusive object of desire: In pursuit of the kinetic equations and the Second Law. In S. F.
Savitt (Ed.), Times arrows today (pp. 191216). Cambridge, MA: Cambridge University Press.
Strutt, R. J. (1968). The life of John William Strutt. Madison, WI: University of Wisconsin Press.
ARTICLE IN PRESS
van Lith, J. (2001). Stir in stillness: A study of the foundations of equilibrium statistical mechanics. Ph.D.
dissertation, University of Utrecht.
Vranas, P. B. M. (1998). Epsilon-ergodicity and the success of equilibrium statistical mechanics. Philosophy of
Science, 65, 688708.
Wald, R. M. (1983). Asymptotic behavior of homogeneous cosmological models in the presence of a positive
cosmological constant. Physical Review D, 28, 21182120.
Winsberg, E. (2004). Can conditioning on the Past Hypothesis militate against the reversibility objection?
Philosophy of Science, 71, 489504.
ARTICLE IN PRESS

John Earman - The "Past Hypothesis" Not Even False

Transféré par

Informations du document

Description originale:

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

John Earman - The "Past Hypothesis" Not Even False

Transféré par

Droits d'auteur :

Formats disponibles

Studies in History and Philosophy of

Modern Physics 37 (2006) 399430

) Suppose that at t = 0 the Boltzmann entropy S

) is not a consequence of the Boltzmann apparatus. Indeed, given the presumed

). And it also contravenes our (presumably) rational expecta-

rzte, 69, 1926.

ber die mechanische Erkla rung irreversibler Vorga nge.

Vous aimerez peut-être aussi