You are on page 1of 10

Biophysical Chemistry 105 (2003) 211–220

Phase transitions, Kauzmann curves, and inverse melting

Frank H. Stillingera, Pablo G. Debenedettib,*
Department of Chemistry, Princeton University, Princeton, NJ 08544, USA
Department of Chemical Engineering, Princeton University, Princeton, NJ 08544, USA

Received 5 September 2002; accepted 30 September 2002


Walter Kauzmann’s classic 1948 review of liquid supercooling and glass formation drew attention to the temperatures
at which (by extrapolation) enthalpies and entropies of liquid and crystal phases would appear to become equal. In
the temperature–pressure (T, p) plane, the collection of such ‘Kauzmann temperatures’ generate characteristic curves.
The present study examines the connection of those Kauzmann loci to equilibrium inverse melting phenomena, i.e.
cases where isobaric heating causes freezing of the liquid. Such cases are associated with local minima or maxima in
the melting curve pm(T), and we point out the possible relevance of melting curve maxima to the thermodynamics
of protein folding. Both equal-enthalpy and equal-entropy Kauzmann curves must pass through melting curve extrema.
Three thermodynamic identities have been obtained to describe the vicinity of these points; they involve, respectively,
the slopes of the two Kauzmann curves, and the second temperature derivative of the melting pressure. The second
of these three equations is formally identical to the first Ehrenfest relation for second-order phase transitions, but
carries no phase-transition implication. For purposes of specific numerical illustration, the inverse-melting behavior
displayed by 3He at low temperature has been analyzed in detail.
䊚 2003 Elsevier Science B.V. All rights reserved.

Keywords: Kauzmann curves; Inverse melting; Kauzmann paradox

1. Introduction supercooled liquids, and the glasses that they form

as a result of supercooling. The historical source
It is always a pleasure to recognize and to honor of the underlying ideas with which we deal below
the lasting scientific contributions of an esteemed is a famous review article published by Walter
colleague and friend. In the present instance, it is Kauzmann in 1948 w1x, which for good reason
our good fortune to do just that for Professor continues to receive frequent literature citations
Walter Kauzmann. The focus of this paper arises even today. A vast amount of research in this
from the need to understand the properties of scientifically challenging field has occurred fol-
lowing appearance of that review, with a corre-
*Corresponding author. Tel.: q1-609-258-5480; fax: q1-
spondingly vast array of published papers. For
E-mail address: readers not intimately involved with supercooled
(P.G. Debenedetti). liquids and glasses, it may be helpful to point out

0301-4622/03/$ - see front matter 䊚 2003 Elsevier Science B.V. All rights reserved.
212 F.H. Stillinger, P.G. Debenedetti / Biophysical Chemistry 105 (2003) 211–220

that several recent comprehensive reviews also involves ambient pressure. However, a comprehen-
exist, documenting the experimental and theoreti- sive view of the properties of those metastable
cal advances that have been made subsequent to states includes the effects of varying pressure,
1948 w2–6x. extending even into the negative pressure (isotro-
Professor Kauzmann stressed the significance of pic tension) regime w13,14x. In the analysis pro-
the experimental fact that, for most liquids, upon vided below, we shall be interested in examining
supercooling the constant-pressure heat capacity behavior in the temperature–pressure (T,p) plane.
CpŽl.ŽT. exceeds that of the thermodynamically As examples, the two temperatures just mentioned
define curves TK(p) and TH(p) in that plane. These
stable crystal phase CpŽc.ŽT., with the discrepancy functions can be grouped with, but remain distin-
widening with increasing extent of supercooling. guished from, the familiar thermodynamic melting
Consequently, reduction in absolute temperature T temperature Tm(p), and the experimental conven-
causes the metastable liquid to lose both entropy tion-dependent glass-transition temperature Tg(p).
and enthalpy faster than does the stable crystal. Our primary objective in this paper is to elabo-
Although a glass transition or crystal nucleation rate upon an earlier suggestion w8x, specifically
event typically would intervene to frustrate contin- that comprehensive understanding of the many-
uation of this trend to very low temperature, body phenomena underlying the Kauzmann para-
smooth and ‘reasonable’ extrapolation of the heat dox should include analysis of ‘inverse melting’.
capacities indicates that the entropy of the super- This unusual and counter-intuitive phenomenon
cooled liquid would equal that of the crystal at a involves liquids at thermodynamic equilibrium that
positive ‘Kauzmann temperature’ TK. Further crystallize upon isobaric addition of heat, the
reduction in temperature toward absolute zero, reverse of the usual situation. The relevance to
following the same extrapolation, would then pro- supercooling and glass formation is that these cases
duce a non-crystalline state with entropy substan- permit straightforward identification of a Kauz-
tially below that of the stable crystal phase, in mann TK(p) curve, at least one point of which lies
clear violation of the third law of thermodynamics on the equilibrium melting curve Tm(p). This
w7x. This conundrum has traditionally been called avoids the usually encountered uncertainty associ-
the ‘Kauzmann paradox’ w8x. An analogous enthal- ated with heat capacity extrapolation below a glass
py crossing would occur (at least for ‘fragile’ glass transition temperature, one consequence of which
formers w3,4x) at another temperature, TH (where has been argument in the literature w15–17x over
0-TH-TK), implying that the corresponding prob- whether or not a thermodynamic ‘ideal glass tran-
lematical non-crystalline state at Ts0 would also sition’ in principle should exist close to TK(p).
be substantially lower in enthalpy than the crystal. Section 2 identifies some known real-substance
The notion that a supercooled liquid might run examples of the inverse melting phenomenon.
out of available molecular configurations at, or Interestingly, this listing cites an example drawn
near, a Kauzmann temperature TK, should have from protein physical chemistry, another subject
implications beyond thermodynamics, and so has toward which Walter Kauzmann has contributed
long influenced the interpretation of flow and fundamental insights w18x. Section 3 analyzes ther-
relaxation behavior. As an example, this has tended modynamic relations for the two phases exhibiting
to provide legitimacy to the presence of a the inverse-melting property, and deduces families
divergence temperature T0)0 in the of curves in the T,p plane that describe constant
Vogel–Tammann–Fulcher (‘VTF’) expression for entropy differences (generalized Ehrenfest rela-
the temperature dependence of shear viscosity tions), as well as curves of constant enthalpy
w9–11x. It also underlies the conceptual basis of differences, between the two phases of interest.
the Adam–Gibbs theory of relaxation phenomena Section 4 is devoted to the specific case of the
w12x. lighter helium isotope 3He and its inverse melting
For obvious reasons, the majority of the exper- behavior. Final remarks have been collected into a
imental studies of supercooled liquids and glasses concluding Section 5, where we speculate on the
F.H. Stillinger, P.G. Debenedetti / Biophysical Chemistry 105 (2003) 211–220 213

order along this entire curve, so at least one of

S (l)yS (c) and V (l)yV (c) is non-zero at every point
of the curve. It will further be assumed that only
a single crystal phase is involved, at least in the
region of the T,p plane depicted in Fig. 1.
The familiar ‘normal’ melting scenario involves
an increase in molar volume upon melting of the
crystal, which occurs as the system absorbs the
latent heat of transition. Consequently, in the right
member of the Clausius–Clapeyron Eq. (2.1), both
the numerator S (l)yS (c) as well as the denominator
V (l)yV (c) are positive, and so the slope of pm(T)
is positive. This circumstance corresponds in Fig.
1 to the portion of the full curve that lies between
points B and C.
Less common, though still familiar, is the case
in which the molar volume decreases upon melting
the crystal, while again heat is absorbed in the
melting process. This combination changes the
Fig. 1. Schematic melting curve in the T,p plane, to illustrate sign of the melting curve slope, and corresponds
various melting circumstances, including the inverse-melting to the portions of the curve shown in Fig. 1 that
phenomenon. lie between points C and D, and between E and
F. Examples of real substances exhibiting this
possibilities of synthesizing and studying new behavior (at modest pressures) are the elements
materials for laboratory study, and of devising new silicon, gallium, antimony, and bismuth w20x, and
models for theoretical analysis, that might display of course the familiar molecular substance water
the inverse melting phenomenon. w21x. The point C in Fig. 1 locates the position at
which the molar volumes of coexisting crystal and
2. Inverse-melting systems liquid are identical, which Eq. (2.1) requires to be
a point of infinite slope of the melting curve
An appropriate starting point for discussion of pm(T).
the inverse melting phenomenon is the Clau- The intervals between points A and B, and
sius–Clapeyron equation that describes the slope between points D and E, are ranges of inverse
of the melting curve in the T,p plane w19x: melting. Starting in the liquid phase just to the left
of either of these intervals, isobaric heating causes
dpmŽT.ydTsŽSŽl.ySŽc..yŽVŽl.yVŽc.. (2.1) crossing of the phase transition curve, i.e. causes
the system to freeze into the crystalline phase.
Here, pm(T) is the temperature-dependent melting That such a counter-intuitive inverse melting might
pressure, the function that is the inverse of Tm(p). occur in the real world was apparently first sug-
Molar entropies and volumes are denoted by S (v) gested by Tammann a century ago w22x. Inverse
and V (v), with superscripts vsl,c indicating liquid melting is characterized by the entropy inequality:
and crystal phases, respectively.
By way of illustrating the relevance of thermo- SŽl.ySŽc.-0 (2.2)
dynamic identity wEq. (2.1)x to the inverse melting
scenario, Fig. 1 offers a schematic melting curve The interval AB exhibits negative slope, while the
ABCDEF. This will help to classify the known interval DE exhibits positive slope, because V (l)y
examples of this phenomenon. It will be assumed V (c) is positive for the former, but negative for the
that the liquid-crystal phase transition remains first latter.
214 F.H. Stillinger, P.G. Debenedetti / Biophysical Chemistry 105 (2003) 211–220

The three points B, D and E are distinguished possesses a maximum of the type depicted in Fig.
by vanishing of the melting curve slope: 1 by point D and its neighborhood. The crystalline
phase in coexistence with the liquid has been
dpmŽT.ydTs0 ŽB,D,E. (2.3) reported to be tetragonal. This melting maximum
occurs at approximately 425 K and 6.5 kbar. The
These points thus are all associated with vanishing P4MP1 and helium isotope examples have been
entropy change: mentioned in a previous publication as relevant to
the Kauzmann paradox issue w8x.
SŽl.ySŽc.s0 ŽB,D,E. (2.4) Although smaller by many orders of magnitude
than macroscopic phases, individual protein mol-
because the difference in molar volumes can never ecules in solution possess, nevertheless, a large
be infinite. Consequently, these are ‘Kauzmann number of internal conformational degrees of free-
points’ occurring as natural loci in the equilibrium dom. The rather sharp transition between the struc-
phase diagram. Their existence does not hinge turally organized and biologically relevant native
upon the legitimacy of long heat capacity extrap- form, and the large group of biologically inactive
olations that are obliged to cross glass transitions. denatured forms, provides a strong analogy to the
Section 3 utilizes this identification to trace out conventional melting transition. It is therefore sat-
‘Kauzmann curves’ TK(p) as they pass through isfying to notice that at least one protein, ribonu-
these kinds of distinguished points. Notice also clease A, displays a ‘melting’ curve in the T,p
that B, D and E must lie on vanishing–enthalpy plane with a maximum of type D in Fig. 1. This
curves TH(p). This follows from the fact that at maximum is located approximately at 283 K and
any point along the melting curve, the chemical 4.0 kbar, with the native form lying below the
potentials of the coexisting phases are equal, and maximum, and the denatured form above the
hence the respective enthalpy and entropy differ- maximum w30x. Starting at a slightly lower tem-
ences satisfy DHsTDS. perature and pressure, an initially denatured mol-
With this schematic analysis as background, we ecule can be induced to renature at constant
can now classify examples of inverse melting that pressure by raising the temperature. By semantic
have been reported in the published literature. The analogy to the phrase ‘inverse melting’, this pro-
best-documented examples of inverse melting are cess might well be called ‘inverse denaturation’.
the two helium isotopes 3He w23x and 4He w24,25x Liquid crystals also supply a useful analogy to
at low temperature. Both isotopes display melting inverse melting. Basing his remarks on data report-
curves that qualitatively resemble the neighbor- ed by Cladis et al. w31x, Johari has pointed out
hood of point B in Fig. 1, but they differ signifi- that the first-order phase boundary between the
cantly from one another in quantitative detail. The smectic-A and nematic phases of 4-cyano-49-octyl-
He example has its zero-slope minimum at oxybiphenyl also exhibits a maximum in the T,p
approximately 0.315 K and 29.3 bar, with normal plane (type D in Fig. 1) w32x. This maximum
(i.e. non-superfluid) liquid in coexistence with occurs at approximately 350 K and 2100 bar. In
body-centered cubic crystal. By contrast, the 4He fact, the observed phase boundary is analogous to
minimum is very shallow, occurs at approximately a larger portion of the schematic curve in Fig. 1
0.8 K and 26.2 bar, and has superfluid liquid in that includes a point of equal molar volumes for
coexistence with hexagonal-close-packed crystal. the two coexisting phases (type C in Fig. 1).
The 3He case will be examined in numerical detail Starting at the appropriate nematic state (possess-
in Section 4. ing just molecular orientational order), isobaric
A very different inverse-melting material has heating can cause the material to reorder into the
recently come to attention. This is the polymeric smectic-A phase (possessing both orientational and
substance poly(4-methylpentene-1), denoted more partial translational order).
simply as ‘P4MP1’ w26–29x. Published reports The reader should be warned that the scientific
indicate that its melting curve in the T,p plane literature also uses the phrase ‘inverse melting’ in
F.H. Stillinger, P.G. Debenedetti / Biophysical Chemistry 105 (2003) 211–220 215

a circumstance that is not directly relevant to the the expansions for the separate phases will appear
present study. In particular, kinetically-controlled as follows:
irreversible transformations of supersaturated alloy
crystals to amorphous solids under heat treatment HŽv.ŽT, p. s HŽv.ŽX. q CpŽv.ŽX. dTqVŽv.ŽX.
have been described this way w33x.

y1 y TX aŽv.ŽX.z~dpq... |
3. Generalized Ehrenfest relations
Recalling that the zero-slope point X has the
In order to present a proper thermodynamic property that molar enthalpies are exactly equal
characterization of inverse melting phenomena, it there, we deduce a linear expression for the differ-
will be useful and convenient to use leading terms ence in molar enthalpies of the liquid and crystal
of Taylor expansions for properties of interest. For phases that is locally valid in the vicinity of point
the moment, let X represent any zero-slope point X:
of the melting curve (such as B, D and E in Fig.
1). We shall suppose that thermodynamic functions
are well defined for both phases in the vicinity of HŽl.ŽT, p.yHŽc.ŽT, p.(wyCpŽl.ŽX.yCpŽc.ŽX.z~dT
x |

X. The temperature and pressure at this point will qµVŽl.ŽX.yw1 y TXaŽl.ŽX.~zyVŽc.ŽX.

x |

be denoted by TX and pX, respectively. Near point

X, the molar volumes of the two phases formally x

y1 y TX aŽc.ŽX.~z ∂dp |

have the following series expansions (vsc,l): (3.5)

VŽv.ŽT,p.sVŽv.ŽX.qAŽv.dTqBŽv.dpq.... For any fixed value of the enthalpy difference

between the phases, dp is a linear function of dT:
dTsTyTX, dpspypX (3.1)
T CpŽl.yCpŽc. T
where subsequent terms involve quadratic and dpshyUT w Žl. X dT,
Žl. yVŽc. 1yTaŽc. z T
higher-order combinations of the deviations dT Vy
V Ž1
y Ta . Ž .~ YX

and dp. The linear coefficients A (v), B (v) are related

to the isobaric thermal expansions a(v) and iso- HŽl.yHŽc.
thermal compressibilities k(Tv) evaluated at position hs (3.6)
X: µVŽl.wy1yTaŽl.z~yVŽc.wy1yTaŽc.z~∂X
x | x |

AŽv.saŽv.ŽX.VŽv.ŽX. where the shorthand notation {«}X specifies eval-

uation at point X. An equivalent description is that
this expression generates a parallel family of
BŽv.sykTŽv.ŽX.VŽv.ŽX. (3.2) curves parameterized by h. The hs0 member of
the family of curves is the one passing precisely
Similar expansions express the behavior of the through point X; this is equivalent to the function
molar phase enthalpies in the neighborhood of TH(p) mentioned Section 1. Eq. (3.6) thus leads
point X. Starting with the thermodynamic identities to the following expression for the slope of the
equal-enthalpy Kauzmann curve at X:
B ≠H E
C F sCp
D ≠T Gp B dp E CpŽl.yCpŽc.
C F s (3.7)
D dT GDHs0 VŽl.Ž1yTaŽl..yVŽc.Ž1yTaŽc..
B ≠H E
C F sVŽ1yaT. (3.3)
D ≠p GT Consider specifically the case XsB in Fig. 1.
We know that at this point of the melting curve
216 F.H. Stillinger, P.G. Debenedetti / Biophysical Chemistry 105 (2003) 211–220

the liquid has a larger molar volume than the SŽl.ŽT,p.ySŽc.ŽT,p.(wyCpŽl.ŽX.


crystal. In addition, the fact that the slope of the

melting curve changes from negative through zero yCpŽc.ŽX.z~dTyTX

to positive upon passing through B allows us to

conclude from the Clausius–Clapeyron Eq. (2.1)
that at dps0, S (l)yS (c) and thus H (l)yH (c) are yaŽc.ŽX.VŽc.ŽX.∂dp (3.13)
increasing with temperature at least in linear order
in dT. From Eq. (3.5) we therefore conclude that: If the molar entropy difference is held fixed at
some constant value, then this last relation implies
that dp will have to be the following linear
CpŽl.ŽB.yCpŽc.ŽB.)0 (3.8) function of dT:

Formally, similar analyses apply to XsD and

T CpŽl.yCpŽc. WT
dpŽdT,s.sysqUT w Žl. Žl.
X dT,
Žc. Žc.z T
to XsE. For both of these cases, the molar volume V Tya V ya V ~ YX
x |

of the liquid is less than that of the crystal, but

the sign changes of the slope of the melting curve SŽl.ySŽc.
occur in opposite directions for these latter two ss (3.14)
cases. For D, the slope decreases from positive, Ža VŽl.yaŽc.VŽc..X

through zero, to negative as temperature increases. As in the case of constant enthalpy difference, this
By contrast, the slope of the melting curve as it expression represents a family of parallel curves,
passes through E changes from negative to zero to indexed now by s. On account of the differing
positive with increasing temperature. These fea- denominators in Eqs. (3.14) and (3.6), the curves
tures imply that: of fixed enthalpy difference and of fixed entropy
difference will generally display distinct slopes.
CpŽl.ŽD.yCpŽc.ŽD.)0 (3.9) The ss0 curve from the family wEq. (3.14)x
passes through point X, and corresponds to vanish-
ing entropy difference. For this special ‘Kauzmann
and that curve’ the differential form of Eq. (3.14) then may
be written:
CpŽl.ŽE.yCpŽc.ŽE.-0 (3.10)
B dp E CpŽl.yCpŽc.
C F s (3.15)
D dT GDSs0 TwyaŽl.VŽl.yaŽc.VŽc.z~
x |

Similar considerations can be brought to bear

on the entropy functions for the two phases.
In principle, this same relation could be integrated
Starting with the general thermodynamic relations:
to trace out an entire curve of vanishing entropy
difference in the T,p plane.
B ≠S E Cp It should not escape attention that Eq. (3.15) is
C Fs (3.11) formally identical to the first Ehrenfest relation for
D ≠T Gp T
a line of second-order phase transitions w34x. By
definition, that is a circumstance which also
and involves equality of molar entropies for the two
phases. However, the present application is differ-
B ≠S E B ≠V E ent in a fundamental way, because the curve
C F syC F syVa (3.12) generated by Eq. (3.15) is not a thermodynamic
D ≠p GT D ≠T Gp
coexistence curve, and in particular, crossing it
does not imply that an ideal glass transition has
we obtain the analog of the earlier Eq. (3.5) occurred.
F.H. Stillinger, P.G. Debenedetti / Biophysical Chemistry 105 (2003) 211–220 217

Fig. 2. Experimentally determined melting curve for 3He, in the neighborhood of its pressure minimum. Adapted from Dobbs w23x.

We close this Section 3 with a relationship This indicates that points B and E, exhibiting
obeyed by the melting curve itself, as it passes upward curvature in Fig. 1, must have a common
through one of the zero-slope points. When carried sign for the heat capacity difference, and the
to second order in temperature and pressure chang- volume difference, between the phases. By con-
es, the Taylor expansion of chemical potential is: trast, point D must entail opposite signs for these
mŽdT,dp.smŽX.ySŽX.dTqVŽX.dp Distinct positive or negative signs in principle
ŽX. E can occur for the various combinations of ther-
yC FŽdT.2qVŽX.aŽX.dTdp
modynamic quantities that appear as factors in the
basic enthalpy and entropy Eqs. (3.7) and (3.15).
ŽX.kTŽX. E Those signs jointly determine the signs of the
yC FŽdp.2q... (3.16) slopes of the enthalpy and entropy Kauzmann
D 2 G
curves as they pass through the zero-slope points
Along the melting curve, the chemical potentials of the melting curve. The slopes of those curves
of the two phases must be equal. Furthermore, in in principle can be either positive or negative, and
the immediate vicinity of the zero-slope points need not be the same for the vanishing-enthalpy
such as B, D and E, the pressure dp along the and vanishing-entropy loci at any given zero-slope
melting curve is proportional to (dT)2. By utilizing point of the melting curve.
these properties in connection with the Taylor
4. Melting curve for 3He
expansion wEq. (3.16)x, one obtains the following
expression for the second temperature derivative We now examine in quantitative detail the
of the melting curve at zero-slope points: inverse melting phenomenon for 3He. As men-
tioned earlier, a melting-curve minimum of type B
d2pm CpŽl.yCpŽc. in Fig. 1 is involved. This example has had a
s (3.17)
dT2 TŽVŽl.yVŽc.. notable technical application, specifically in the
‘Pomeranchuk refrigerator’ that attains millikelvin
218 F.H. Stillinger, P.G. Debenedetti / Biophysical Chemistry 105 (2003) 211–220

Table 1 (3.8). This leads in turn, to numerical evaluation

Experimentally measured values of thermodynamic properties of the slope of the equal-entropy Kauzmann curve,
for the coexisting liquid and crystal phases of 3He, at the zero-
slope melting minimuma Eq. (3.15), as it passes through the melting point
Property Measured
value B dp E
T (K) 0.315 C F (y437 bar Ky1 (4.2)
D dT GDSs0
p (bar) 29.31
d2pmydT 2 (bar Ky2) 70
V (c) (cm3 moly1) 24.80 In a similar fashion, the slope of the equal-enthalpy
V (l) (cm3 moly1) 26.04 curve at the same point can be evaluated from Eq.
a(c) (Ky1) 0.00070 (3.7):
a(l) (Ky1) y0.00696
All entries have been extracted from data presented in the B dp E
extensive review by Dobbs w23x. C F (y21.0 bar Ky1 (4.3)
D dT GDHs0

temperatures in the laboratory w35x. Fig. 2 shows

that part of the melting curve for this substance
5. Discussion
that is relevant to our inverse melting discussion,
specifically the vicinity of the zero-slope pressure
minimum. Measured values of various thermody- The point E included in the schematic Fig. 1
namic properties at that minimum have been col- involves liquid at higher pressure than the crystal,
lected in Table 1. and positive curvature to the melting curve at this
In contrast to the other inverse-melting examples zero-slope point. Currently, no known real-world
mentioned in Section 2 above, the helium isotopes example exists for this combination of attributes,
are distinctly quantum mechanical in their behav- but nothing in principle seems to exclude it.
ior. This is particularly so with the fermionic Although not included in Fig. 1 in the interest of
lighter isotope 3He, for which the deep melting simplicity, a phase-reversed version of point D can
curve minimum owes its existence to the involve- also be imagined, displaying liquid at lower pres-
ment of nuclear spin degrees of freedom. These sure than crystal, and (as with D itself) negative
nuclear spins are relatively free to reorient inde- second derivative for the melting curve at that
pendently in the crystal at the temperature range zero-slope point. Again, no known real-world
of concern here, thereby increasing entropy. By example with this characteristic currently exists.
contrast, particle exchange in the 3He liquid strong- All cases of inverse melting appear to involve,
ly couples nuclear spins in that phase, so the and probably require, a significant number of
entropy remains low in comparison to the crystal. microscopic degrees of freedom that are coupled
The bosonic heavier isotope 4He has no nuclear to the particle center-of-mass degrees of freedom.
spin, and its very shallow melting curve minimum It is localization of the latter that determine freez-
relies on the relative phonon densities of states for ing of a liquid into a spatially periodic solid.
the crystal and liquid phases w8x. Nuclear spins play this role of coupled degrees of
By inserting values from Table 1 into the pm freedom in 3He. Backbone and side-group confor-
curvature condition Eq. (3.17), one can evaluate mational degrees of freedom evidently play an
the heat capacity difference at the melting-point analogous part for the polymeric substance
minimum: P4MP1. The ribonuclease A and liquid–crystal
cases cited in Section 2 above also possess many
internal molecular degrees of freedom that are
CpŽl.yCpŽc.(2.73 J moly1 Ky1 (4.1) coupled respectively to native-state conformation,
and to orientational and positional order to produce
which is consistent with the inequality of Eq. inverse melting analogs. One might say that a
F.H. Stillinger, P.G. Debenedetti / Biophysical Chemistry 105 (2003) 211–220 219

preponderance of internal degrees of freedom, Acknowledgments

when properly coupled to molecular position, have
the capacity to drive inverse melting phenomena. PGD gratefully acknowledges financial support
These last considerations offer hope for the by the US Department of Energy, Division of
construction of theoretical many-body models that Chemical Sciences, Geosciences, and Biosciences,
upon analytical andyor simulational examination Office of Basic Energy Sciences (Grant DE-FG02-
will exhibit inverse melting. Within the regime of 87ER13714).
classical statistical mechanics, one such opportu-
nity may emerge from a suitably crafted polymer References
model. It was pointed out by Flory and Krigbaum
many years ago w36x that polymers suspended in w1x W. Kauzmann, The nature of the glassy state and the
suitable solvents experience repulsive effective behavior of liquids at low temperatures, Chem. Rev. 43
pair interactions that are essentially Gaussian func- (1948) 219–256.
w2x J. Jackle, Models of the glass transition, Rep. Prog.
tions of the centroid-to-centroid distance. This
Phys. 49 (1986) 171–231.
observation can be abstracted into the well-studied w3x M.D. Ediger, C.A. Angell, S.R. Nagel, Supercooled
Gaussian core model w37–39x, which is known to liquids and glasses, J. Phys. Chem. 100 (1996)
have a point of type C, Fig. 1, at which the melting 13200–13212.
volume changes sign. At present, a study is under- w4x P.G. Debenedetti, Metastable Liquids: Concepts and
way to insert ‘polymer’ internal degrees of free- Principles, Princeton University Press, Princeton, NJ,
dom into the Gaussian core model that implicitly w5x C.A. Angell, K.L. Ngai, G.B. McKenna, P.F. McMillan,
change its centroid-to-centroid effective interac- S.W. Martin, Relaxation in glassforming liquids, J. Appl.
tions as temperature changes. In fact, it has been Phys. 88 (2000) 3113–3157.
established that under proper circumstances, an w6x P.G. Debenedetti, F.H. Stillinger, Supercooled liquids
inverse melting phenomenon can arise w40x. and the glass transition, Nature 410 (2001) 259–267.
w7x H.B. Callen, Thermodynamics, John Wiley and Sons,
Finally, we return to remark about the Kauz- New York, 1985, p. 30.
mann curves, specifically those for entropy equal- w8x F.H. Stillinger, P.G. Debenedetti, T.M. Truskett, The
ity of liquid and crystal phases. The results Kauzmann paradox revisited, J. Phys. Chem. B 105
deduced in this paper add a new facet to the (2001) 11809–11816.
interpretation of the glass transition, and to the ¨
w9x H. Vogel, Das temperature-abhangigketsgesetz der vis-
¨ von flussigkeiten,
kositat ¨ Phys. Zeit. 22 (1921)
related ‘Kauzmann paradox’. We have shown that 645–646.
at least for those few systems which exhibit inverse ¨
w10x G. Tammann, W. Hesse, Die abhangigkeit ¨
der viskositat
melting, a Kauzmann curve of equal entropies is ¨
von der temperature bei unterkuhlten ¨
flussigkeiten, Z.
not itself a line of phase transitions in the T,p Anorg. Allg. Chem. 156 (1926) 245–257.
w11x G.S. Fulcher, Analysis of recent measurements of the
plane. It is conceivable that some inverse melting
viscosity of glasses, J. Am. Ceram. Soc. 8 (1925)
systems might also have glass transitions, and that 339–355.
continuation of the equal-entropy curve from the w12x G. Adam, J.H. Gibbs, On the temperature dependence
locale upon which we have focused to that glass of cooperative relaxation properties in glass-forming
transition locale might encounter the conventional liquids, J. Chem. Phys. 43 (1965) 139–146.
w13x P.G. Debenedetti, F.H. Stillinger, T.M. Truskett, C.J.
Kauzmann temperature. We therefore raise the
Roberts, The equation of state of an energy landscape,
question whether the Kauzmann temperature for J. Phys. Chem. B 103 (1999) 7390–7397.
supercooled liquids necessarily implies the exis- w14x M. Utz, P.G. Debenedetti, F.H. Stillinger, Isotropic
tence of an ideal glass transition (never directly tensile strength of molecular glasses, J. Chem. Phys.
observed, of course). It appears to be an equally 114 (2001) 10049–10057.
w15x J. Jackle, Models of the glass transition, Rep. Prog.
plausible proposition that Kauzmann curves for
Phys. 49 (1986) 171–231.
supercooled liquids are simply an interesting ther- w16x F.H. Stillinger, Supercooled liquids, glass transitions,
modynamic locus, without demanding an interpre- and the Kauzmann paradox, J. Chem. Phys. 88 (1988)
tation as an ideal glass transition. 7818–7825.
220 F.H. Stillinger, P.G. Debenedetti / Biophysical Chemistry 105 (2003) 211–220

w17x P.G. Debenedetti, F.H. Stillinger, C.P. Lewis, Model w29x A.L. Greer, Too hot to melt, Nature 404 (2000)
energy landscapes, unpublished manuscript. 134–135.
w18x W. Kauzmann, Some factors in the interpretation of w30x J. Zhang, X. Peng, A. Jonas, J. Jonas, NMR study of
protein denaturation, Adv. Protein Chem. 14 (1959) the cold, heat, and pressure unfolding of ribonuclease
1–63. A, Biochemistry 34 (1995) 8631–8641.
w19x H. Reiss, Methods of Thermodynamics, Dover, Mineola, w31x P.E. Cladis, D. Guillon, F.R. Bouchet, P.L. Finn, Reen-
NY, 1996, p. 137. trant nematic transitions in cyano-octyloxybiphenyl
w20x D.A. Young, Phase Diagrams of the Elements, Univer- (8OCB), Phys. Rev. A 23 (1981) 2594–2601.
sity of California Press, Berkeley, 1991. w32x G.P. Johari, The Tammann phase boundary, exothermic
w21x D. Eisenberg, W. Kauzmann, The Structure and Prop- disordering and the entropy contribution change on
erties of Water, Oxford University Press, New York, phase transformation, Phys. Chem. Chem. Phys. 3
1969. (2001) 2483–2487.
w22x G. Tammann, Kristallisieren und Schmelzen, Johann w33x W. Sinkler, C. Michaelson, R. Bormann, A transmission
Ambrosius Barth, Leipzig, 1903, pp. 26–46. electron microscopy investigation of inverse melting in
w23x E.R. Dobbs, Helium Three, Oxford University Press, Nb45Cr55, J. Mater. Res. 12 (1997) 1872–1884.
Oxford, 2002. w34x A.B. Pippard, The Elements of Classical Thermodynam-
w24x C. Le Pair, K.W. Taconis, R. De Bruyn Ouboter, P. Das, ics, Cambridge University Press, Cambridge, 1957, p.
A direct measurement of the minimum in the melting 140.
curve of 4He, Physica 29 (1963) 755–756. w35x J. Wilks, D.S. Betts, An Introduction to Liquid Helium,
w25x J. Wilks, The Properties of Liquid and Solid Helium, Clarendon Press, Oxford, 1987, pp. 79–80.
Clarendon Press, Oxford, 1967. w36x P.J. Flory, Principles of Polymer Chemistry, Cornell
w26x S. Rastogi, M. Newman, A. Keller, Pressure-induced University Press, Ithaca, 1953, p. 539.
amorphization and disordering on cooling in a crystal- w37x F.H. Stillinger, D.K. Stillinger, Negative thermal expan-
line polymer, Nature 353 (1991) 55–57. sion in the Gaussian core model, Physica A 244 (1997)
w27x S. Rastogi, M. Newman, A. Keller, Unusual pressure- 358–369.
induced phase behavior in crystalline poly-4-methyl- w38x A. Lang, C.N. Likos, M. Watzlawek, H. Lowen, ¨ Fluid
pentene-1, J. Polym. Sci. (Polym. Phys.) B 31 (1993) and solid phases of the Gaussian core model, J. Phys.
125–139. Condens. Matter 12 (2000) 5087–5108.
w28x S. Rastogi, G.W.H. Hohne, A. Keller, Unusual pressure- w39x A.A. Louis, P.G. Bolhuis, J.P. Hansen, Mean field fluid
induced phase behavior in crystalline poly(4-methylpen- behavior of the Gaussian core model, Phys. Rev. E 62
tene-10): calorimetric and spectroscopic results and (2000) 7961–7972.
further implications, Macromolecules 32 (1999) w40x M.R. Feeney, P.G. Debeneddetti, F.H. Stillinger, J.
8897–8909. Chem. Phys., submitted.