insight review articles

Supercooled liquids
and the glass transition
Pablo G. Debenedetti* & Frank H. Stillinger†‡
*Department of Chemical Engineering and ‡Princeton Materials Institute, Princeton University, Princeton, New Jersey 08544, USA
(e-mail: [email protected])
†Bell Laboratories, Lucent Technologies, Murray Hill, New Jersey 07974, USA

Glasses are disordered materials that lack the periodicity of crystals but behave mechanically like solids.
The most common way of making a glass is by cooling a viscous liquid fast enough to avoid crystallization.
Although this route to the vitreous state — supercooling — has been known for millennia, the molecular
processes by which liquids acquire amorphous rigidity upon cooling are not fully understood. Here we
discuss current theoretical knowledge of the manner in which intermolecular forces give rise to complex
behaviour in supercooled liquids and glasses. An intriguing aspect of this behaviour is the apparent
connection between dynamics and thermodynamics. The multidimensional potential energy surface as a
function of particle coordinates (the energy landscape) offers a convenient viewpoint for the analysis and
interpretation of supercooling and glass-formation phenomena. That much of this analysis is at present
largely qualitative reflects the fact that precise computations of how viscous liquids sample their landscape
have become possible only recently.

T
he glassy state is ubiquitous in nature and (for example, during annealing or ageing) and far below Tg,
technology1. It is crucial in the processing of is not discussed. The reader is referred to an excellent recent
foods2, the commercial stabilization of labile review8 for a thorough coverage of these and other topics.
biochemicals3, and the preservation of insect
life under extremes of cold or dehydration3. Phenomenology of supercooling and glass formation
Window glass, composed mostly of sand, lime and soda, is Figure 1 illustrates the temperature dependence of a liquid’s
the best-known example of an engineered amorphous volume (or enthalpy) at constant pressure4,9. Upon cooling
solid4. Optical fibres are made of very pure amorphous below the freezing point Tm, molecular motion slows down.
silica, occasionally carefully doped. Most engineering If the liquid is cooled sufficiently fast, crystallization can be
plastics are amorphous solids, as are some metallic glasses avoided10,11. Eventually molecules will rearrange so slowly
and alloys of interest because of their soft magnetism and that they cannot adequately sample configurations in the
corrosion resistance5. The silicon used in many available time allowed by the cooling rate. The liquid’s
photovoltaic cells is amorphous, and it is possible that structure therefore appears ‘frozen’ on the laboratory
most water in the Universe may be glassy6. Most of these timescale (for example, minutes). This falling out of
examples entail supercooling of a liquid to take advantage equilibrium occurs across a narrow transformation range
of viscous retardation of nucleation and crystallization. where the characteristic molecular relaxation time becomes
Understanding quantitatively the extraordinary viscous of the order of 100 seconds, and the rate of change of volume
slow-down that accompanies supercooling and glass or enthalpy with respect to temperature decreases abruptly
formation is a major scientific challenge7. (but continuously) to a value comparable to that of a
We begin by reviewing the phenomenology of vitrifica- crystalline solid. The resulting material is a glass. The inter-
tion and supercooling. A useful approach for relating this section of the liquid and vitreous portions of the volume
complex phenomenology to molecular-scale events is to versus temperature curve provides one definition of Tg,
focus attention on the liquid’s energy landscape, that is, the which usually occurs around 2Tm/3. The behaviour depict-
multidimensional surface generated by the system’s ed in Fig. 1 is not a true phase transition, as it does not
potential energy as a function of molecular coordinates. involve discontinuous changes in any physical property.
Accordingly, basic landscape concepts and a discussion of The slower a liquid is cooled, the longer the time
the important theoretical and computational progress available for configurational sampling at each temperature,
currently being made in this area are presented next. This is and hence the colder it can become before falling out of
followed by a discussion of alternative viewpoints, in which liquid-state equilibrium. Consequently, Tg increases with
narrowly avoided singularities are assumed to occur well cooling rate12,13. The properties of a glass, therefore, depend
above the glass-transition temperature. We then close with a on the process by which it is formed. In practice, the
summary of the important open questions. dependence of Tg on the cooling rate is weak (Tg changes by
It is impossible to do justice to the entire field of super- 3–5 7C when the cooling rate changes by an order of
cooled liquids and amorphous solids in an article of this magnitude14), and the transformation range is narrow, so
length. We have therefore limited the scope to the dynamics that Tg is an important material characteristic.
and thermodynamics of viscous liquids above and close to
the glass-transition temperature Tg — in other words, to the Slowing down
glass transition viewed ‘from the liquid’. The view ‘from the Another definition of Tg is the temperature at which the
solid’, including such topics as relaxation both relatively near shear viscosity reaches 1013 poise. Close to Tg the viscosity h
NATURE | VOL 410 | 8 MARCH 2001 | www.nature.com
© 2001 Macmillan Magazines Ltd 259
insight review articles
is extraordinarily sensitive to temperature. For silica this dependence along a ‘strong’ to ‘fragile’ scale. The viscosity and relaxation times
is reasonably well described by the Arrhenius functionality, (for example, dielectric relaxation) of the former behave in nearly
h = Aexp(E/kBT), where A and E are temperature-independent and Arrhenius fashion, whereas fragile liquids show marked deviations
kB is Boltzmann’s constant. Other liquids exhibit an even more from Arrhenius behaviour. Silica (SiO2) is the prototypical strong liq-
pronounced viscous slow-down close to the glass transition, which is uid, whereas o-terphenyl (OTP) is the canonical fragile glass-former.
reasonably well represented, over 2–4 orders of magnitude in viscosi- Strong liquids, such as the network oxides SiO2 and germanium
ty8, by the Vogel–Tammann–Fulcher (VTF) equation15–17 dioxide (GeO2), have tetrahedrally coordinated structures, whereas
the molecules of fragile liquids exert largely non-directional, disper-
h = Aexp[B/(T1To)] (1) sive forces on each other. Alternative scaling descriptions that
attempt to extract universal aspects of viscous slow-down have been
where A and B are temperature-independent constants. Understand- proposed22–24. Their relative merits are still being assessed8,25.
ing the origin of this extraordinary slow-down of relaxation Viscous liquids close to Tg exhibit non-exponential relaxation.
processes is one of the main challenges in the physics of glasses. The temporal behaviour of the response function F(t) (for example,
Figure 2 shows a Tg-scaled Arrhenius representation of liquid the polarization in response to an applied electric field, the strain
viscosities18–21. Angell has proposed a useful classification of liquids (deformation) resulting from an applied stress, or the stress in

Box 1
Entropy crises

Boltzmann’s entropy formula establishes the connection between the microscopic world of atoms and molecules and the bulk properties of
matter:

S(N, V, E )4kB ln V

In this equation, S is the entropy, kB is Boltzmann’s constant and V is the number of quantum states accessible to N particles with fixed energy
E in a volume V. Because V cannot be less than one, the entropy cannot be negative. When a crystal is cooled sufficiently slowly, it approaches
a unique state of lowest energy, and hence its entropy approaches zero as T→0. If the entropy of a supercooled liquid were to become smaller
than that of the stable crystal at the Kauzmann temperature, its entropy would eventually become negative upon further cooling. This
impossible scenario constitutes an entropy crisis46–48.
The Kauzmann temperature TK is given by9
Tm Dc
Dsm4E }} dT
p

TK T
where Dsm is the melting entropy (the difference between liquid and crystal entropies at the melting temperature), Tm is the melting temperature
at the given pressure, and Dcp is the temperature-dependent difference between the heat capacity of the liquid and the crystal at the given
pressure. The rate of change of entropy with temperature at constant pressure is given by
cp
1}!!}sT 2 4}T}
P

The entropy crisis arises because the heat capacity of a liquid is greater than that of the stable crystal. The entropy of fusion is therefore
consumed upon supercooling, and vanishes at TK. The entropy crisis entails no conflict with the second law of thermodynamics, as the
difference in chemical potential Dm between the supercooled liquid and the stable crystal at TK is a positive quantity. Because the chemical
potential is the Gibbs free energy per unit mass, this means that the system can reduce its Gibbs free energy by freezing, in accord with
experience. The chemical potential difference at TK is given by9

Dm(TK )4E Dcp }}11

Tm
1 2
Tm
dT
TK T
One way of avoiding the entropy crisis is for the liquid to form an ideal glass of unique configuration at TK. This is the thermodynamic view of the
glass transition, according to which the observable glass transition is a manifestation, masked by kinetics, of an underlying second-order phase
transition50 occurring at TK.
Because the glass transition intervenes before the entropy crisis occurs (Tg > TK ), estimates of the Kauzmann temperature involve an
extrapolation of liquid properties below Tg. The validity of such extrapolations, and hence of the very possibility of an entropy crisis, has been
questioned by Stillinger103, owing to the apparent necessity for configurational excitations out of the ideal glass state to require an unbounded
energy. Furthermore, recent computer simulations of polydisperse hard disks found no evidence of a thermodynamic basis underlying the glass
transition104. Although the notion of an ideal glass remains controversial103, this does not undermine the usefulness of TK as an empirical
classification parameter for glass-formers.
Experimentally, there are substances with known Kauzmann points. 4He is a liquid at 0 K and 1 bar (liquid He-II). Its equilibrium freezing
pressure at 0 K is 26 bar. At this point, the entropies of the liquid and the crystal are equal, and this is therefore a Kauzmann point, although not
an entropy crisis as both phases have zero entropy. The melting curves of 3He and 4He exhibit pressure minima: these occur at about 0.32 K
and 0.8 K, respectively105. These are also equal-entropy (Kauzmann) points. Experiments indicate that poly(4-methylpentene-1) exhibits a
pressure maximum along its melting curve106. Although the appearance of an additional phase complicates the interpretation of this
observation, the implication would be that the pressure maximum is a Kauzmann point, and the continuation of the melting curve to lower
temperatures and pressures beyond this point corresponds to endothermic freezing of a stable liquid into a crystal possessing higher
entropy107. How this liquid would avoid conflict with the third law is not understood, but may hinge on the vibrational anharmonicity of the two
phases with changing temperature.

260 © 2001 Macmillan Magazines Ltd NATURE | VOL 410 | 8 MARCH 2001 | www.nature.com
 insight review articles
response to an imposed deformation) can often be described by the fast that a modest extrapolation of experimental data predicts its
stretched exponential, or Kohlrausch–Williams–Watts (KWW) impending vanishing. In practice, the glass transition intervenes, and
function26,27 DS does not vanish. If the glass transition did not intervene, the liquid
entropy would equal the crystal’s entropy at a nonzero temperature
F(t)4exp[1(t/t)b] (b < 1) (2) TK (the Kauzmann temperature.) Because the entropy of the crystal
approaches zero as T tends to zero, the entropy of the liquid would
where F(t)4[s (t)1s (÷)]/[s (0)1s (÷)] and s is the measured eventually become negative upon cooling if this trend were to contin-
quantity (for example, the instantaneous stress following a step ue. Because entropy is an inherently non-negative quantity (Box 1),
change in deformation). t in equation (2) is a characteristic relax- the state of affairs to which liquids such as lactic acid are tending when
ation time, whose temperature dependence is often non-Arrhenius the glass transition intervenes is an entropy crisis46–48. The extrapola-
(exhibiting fragile behaviour). The slowing down of long-time tion needed to provoke conflict with the third law is quite modest for
relaxation embodied in equation (2) contrasts with the behaviour of many fragile liquids49, and the imminent crisis is thwarted by a
liquids above the melting point, which is characterized by simple kinetic phenomenon, the glass transition. This suggests a connection
exponential relaxation. Experimental and computational evidence between the kinetics and the thermodynamics of glasses47. The
indicates that this slow-down is related to the growth of distinct thermodynamic viewpoint that emerges from this analysis50
relaxing domains28–39 (spatial heterogeneity). Whether each of these considers the laboratory glass transition as a kinetically controlled
spatially heterogeneous domains relaxes exponentially or not is a manifestation of an underlying thermodynamic transition to an
matter of considerable current interest38,39. ideal glass with a unique configuration.
A formula of Adam and Gibbs51 provides a suggestive connection
Decouplings between kinetics and thermodynamics:
In supercooled liquids below approximately 1.2Tg there occurs a
decoupling between translational diffusion and viscosity, and t4Aexp(B/T sc) (3)
between rotational and translational diffusion30,39,40. At higher
temperatures, both the translational and the rotational diffusion In this equation, t is a relaxation time (or, equivalently, the viscosity)
coefficients are inversely proportional to the viscosity, in agreement and A and B are constants. sc, the configurational entropy, is related to
with the Stokes–Einstein and Debye equations, respectively. Below the number of minima of the system’s multidimensional potential
approximately 1.2Tg, the inverse relationship between translational energy surface (Box 2). According to the Adam–Gibbs picture, the
motion and viscosity breaks down, whereas that between rotational origin of viscous slow-down close to Tg is the decrease in the number
motion and viscosity does not. Near Tg, it is found that molecules of configurations that the system is able to sample. At the Kauzmann
translate faster than expected based on their viscosity, by as much as temperature the liquid would have attained a unique, non-crystalline
two orders of magnitude. This therefore means that, as the state of lowest energy, the ideal glass. Because there is no configura-
temperature is lowered, molecules on average translate progressively tional entropy associated with confinement in such a state, the
more for every rotation they execute. Yet another decoupling occurs Adam–Gibbs theory predicts structural arrest to occur at TK. In their
in the moderately supercooled range. At sufficiently high derivation of equation (3), Adam and Gibbs invoked the concept of a
temperature the liquid shows a single peak relaxation frequency cooperatively rearranging region (CRR)51. A weakness of their
(Fig. 3), indicative of one relaxation mechanism. In the moderately treatment is the fact that it provides no information on the size of
supercooled regime, however, the peak splits into slow (a) and fast such regions. The fact that the CRRs are indistinguishable from each
(b) relaxations41–43. The former exhibit non-Arrhenius behaviour other is also problematic, in light of the heterogeneity that is believed
and disappear at Tg; the latter continue below Tg and display to underlie stretched exponential behaviour8.
Arrhenius behaviour44.

Thermodynamics
The entropy of a liquid at its melting temperature is higher than that 14 m,o-Xylene
of the corresponding crystal. Because the heat capacity of a liquid is Strong
m,o-Fluorotoluene
higher than that of the crystal, this entropy difference decreases upon 12
Chlorobenzene
Log (viscosity in poise)

supercooling (Box 1). Figure 4 shows the temperature dependence of

10 Toluene
the entropy difference between several supercooled liquids and their
stable crystals45. For lactic acid this entropic surplus is consumed so 8 o-Terphenyl
K + Bi3+CI-
6
K+Ca2+NO -3
Figure 1 Temperature 4 GeO2
dependence of a Liquid
SiO2
liquid’s volume v or 2
Volume, Enthalpy

enthalpy h at constant
0
pressure. Tm is the
melting temperature. -2 Fragile
A slow cooling rate
produces a glass -4
transition at Tga; a Glass Tg/T
b
faster cooling rate a
leads to a glass tal Figure 2 Tg-scaled Arrhenius representation of liquid viscosities showing Angell’s
Crys
transition at Tgb. strong–fragile pattern. Strong liquids exhibit approximate linearity (Arrhenius
Tga Tgb Tm
The thermal Temperature behaviour), indicative of a temperature-independent activation energy
expansion coefficient E4dlnh/d(1/T ) ≈ const. Fragile liquids exhibit super-Arrhenius behaviour, their
ap4(!lnv /!T )p and effective activation energy increasing as temperature decreases. (Adapted from
the isobaric heat capacity cp4(!h /!T )p change abruptly but continuously at Tg. refs 9 and 11.)

NATURE | VOL 410 | 8 MARCH 2001 | www.nature.com © 2001 Macmillan Magazines Ltd 261
insight review articles
Nevertheless, equation (3) describes the relaxation behaviour of relation yields the VTF equation, which is mathematically
deeply supercooled liquids remarkably well. If the difference in heat equivalent to the Williams–Lendel–Ferry equation for the tempera-
capacities between a supercooled liquid and its stable crystalline ture dependence of viscosity in polymers53. This transformation is
form is inversely proportional to temperature52, the Adam–Gibbs predicated on the assumption that the vibrational entropies of the

Box 2
Statistics of landscapes

The complexity of many-body landscapes makes a statistical description inevitable. The quantity of interest is the number of minima of given
depth, which is given by108

dV
}}4Cexp[Ns (f )]
df
Here, dV denotes the number of potential energy minima with depth per particle (f4F/N) between f and f5df/2. C is an N-independent factor
with units of inverse energy, and s(f), also an N-independent quantity, is a so-called basin enumeration function. Taking the logarithm of the
above expression and comparing with Boltzmann’s entropy formula (Box 1), we see that s(f) is the entropy per particle arising from the existence
of multiple minima of depth f, or, in other words, the configurational entropy.
At low temperatures, it is possible to separate the configurational contribution to thermophysical properties, which arises from the exploration
of different basins, from the vibrational component, which arises from thermal motions confined to a given basin75,76. The Helmholtz free energy A
is then given by

A f
w av
}}4}}1s ( f
w )&}}
NkT kT kBT
where fw is the depth of the basins preferentially sampled at the given temperature, and av is the vibrational free energy per particle. Thus, the free
energy consists of an energetic component that reflects the depth of landscape basins sampled preferentially at the given temperature, an
entropic component that accounts for the number of existing basins of a given depth, and a vibrational component. The statistical description of a
landscape consists of the basin enumeration function s (f), from which the excitation profile f(T) is obtained through the free-energy minimization
condition

ds 1
}}4}}
df kBT
The above equation assumes that av depends on T, but not on f — that is, all basins have the same mean curvature at their respective minima.
The shape of a given system’s landscape is determined by the density (number of molecules per unit volume, N/V). Temperature governs the
manner in which the landscape is sampled. A different basin enumeration function and excitation profile corresponds to each density.
Temperature dictates the point along the enumeration curve and the excitation profile sampled by the system at fixed density (see figure below).
It is possible to construct the basin enumeration function and excitation profile of a system from experimental heat capacity data for the crystal
and the supercooled liquid109, and by computer simulation77,78. In the latter case, the calculations involve determining the probability distribution of
inherent structure energies sampled as a function of temperature. These calculations are at the limit of what is presently feasible with available
computational power. The enumeration function is often well represented by a parabola, indicative of a gaussian distribution of basins4,77,97.
At present it is not understood how the enumeration function deforms with density for a given system (but see ref. 96 for a recent example of
such a calculation), or how it depends on molecular architecture. Understanding such questions would provide a direct link between
landscape statistics and physical properties. The success of the Adam–Gibbs equation indicates that this link applies also to transport
properties such as diffusion and viscosity.

Box 2 Figure Schematic

representation of the basin
enumeration function (left)
and the excitation profile 1/kBT
(right). f is the potential
energy per particle in
mechanically stable
potential energy minima.
fcr is the corresponding
quantity in the stable
crystal. The number of
potential energy minima 1/kBTK
with depth between f and =0
f5df/2 is proportional to – cr T
exp[Ns(f)]. In the TK
thermodynamic limit (N of
the order of Avogadro’s number, 6.0221023), basins that possess larger (less negative) potential energies (shallow basins) are overwhelmingly
more numerous than deeper basins possessing very negative f-values. The slope of the enumeration function is inversely proportional to the
temperature. The excitation profile gives the depth of the inherent structures sampled preferentially at a given temperature. At the Kauzmann
temperature TK the system attains the state of a configurationally unique ideal glass (s40), corresponding to the deepest amorphous basin (see
Figs 5 and 8) and its inherent structure energy does not therefore change upon further cooling.

262 © 2001 Macmillan Magazines Ltd NATURE | VOL 410 | 8 MARCH 2001 | www.nature.com
 insight review articles
supercooled liquid and its stable crystal are equal9. For many fragile 1.0
glass-formers the VTF temperature of structural arrest, To, is very
close to TK obtained from calorimetric measurements (typically49
0.9 < TK/To < 1.1). This again indicates a connection between B2O3
dynamics and thermodynamics not present at higher temperatures. 0.8
Equally suggestive is the correspondence between kinetic fragilities
based on the temperature dependence of the viscosity (see Fig. 2) and
thermodynamic fragilities54, based on the temperature dependence
0.6
of the entropy surplus of the supercooled liquid with respect to its
∆S
stable crystal. ∆Sm Ethanol
Propanol

The energy landscape 0.4 Glycerol

A convenient framework for interpreting the complex phenomenol- Glucose
ogy just described is provided by the energy landscape44. This is the
Lactic acid
name generally given to the potential energy function of an N-body
0.2
system F(r1, …, rN), where the vectors ri comprise position,
orientation and vibration coordinates. In condensed phases,
whether liquid or solid, every molecule experiences simultaneous Tg
interactions with numerous neighbours. Under these conditions it is 0
convenient to consider the full N-body F-function. The landscape is 0 0.2 0.4 0.6 0.8 0.9
T/Tm
a multidimensional surface. For the simplest case of N structureless
particles possessing no internal orientational and vibrational degrees
of freedom, the landscape is a (3N&1)-dimensional object. Figure 5 Figure 4 Temperature dependence of the entropy difference between several
is a schematic illustration of an energy landscape. The quantities of supercooled liquids and their stable crystals at atmospheric pressure. DSm is the
interest are the number of potential energy minima (also called melting entropy and Tm is the melting temperature. The glass transition always
inherent structures) of a given depth (Box 2), and the nature of the intervenes before the vanishing of the entropy surplus. For fragile liquids such as
saddle points separating neighbouring minima. More than 30 years lactic acid, however, an entropy crisis is imminent. (Adapted from ref. 45.)
ago, Goldstein articulated a topographic viewpoint of condensed
phases55 that has come to be known as the energy landscape
paradigm. His seminal ideas have since been applied to protein function of temperature provides information on its dynamic
folding56–64, the mechanical properties of glasses65–67, shear-enhanced behaviour70. The way that a landscape deforms as a result of changes
diffusion68 and the dynamics of supercooled liquids69–71. in density provides information on the mechanical properties of a
material system72. Figure 6 shows the average inherent structure
Landscape sampling energy for a mixture of unequal-sized atoms, as a function of the
For an N-body material system in a volume V, the landscape is fixed. temperature of the equilibrated liquid70,73. In these calculations,
The manner in which a material system samples its landscape as a molecular dynamics simulations of the binary mixture were
performed to generate configurations. Periodically, the system’s
energy was minimized, yielding mechanically stable inherent
11
structures, the average energy of which is reported in the figure. At
10 high temperatures the inherent structure energy is virtually tempera-
9 ture-independent, and appears to have reached a plateau. When the
system has sufficient kinetic energy to sample its entire energy
8
landscape, the overwhelming number of minima that it samples are
7 shallow, reflecting the fact that deep minima are very rare (Box 2). But
6 as the reduced temperature decreases below about T = 1, the system is
unable to surmount the highest energy barriers, and is therefore
Log (fm/Hz)

5
forced to sample the much rarer deeper minima (Box 2). When this
4 happens, the kinetics of structural relaxation changes from exponen-
3
tial to stretched exponential, and the activation energy (and entropy)
associated with structural relaxation become super-Arrhenius, that
Tg = 131+_1 K
2 is to say they increase with decreasing temperature70.
1 These calculations established a connection between changes in
dynamics and the manner in which the static and thermodynamic
0
energy landscape is sampled as a function of temperature. Figure 6 also
-1 shows that at a low enough temperature the system becomes stuck in a
-2 single minimum, the depth of which increases as the cooling rate
decreases. This corresponds to the glass transition. Another important
-3
3.0 4.0 5.0 6.0 7.0 8.0 9.0
observation of this study was the existence of a temperature T ≈ 0.45,
103/T (K-1) below which the height of the barriers separating sampled inherent
structures increases abruptly. This temperature was found to
Figure 3 Temperature dependence of the peak dielectric relaxation frequency of the correspond closely to the crossover temperature predicted by mode-
glass-forming mixture chlorobenzene/cis-decalin (molar ratio 17.2/82.8%). At high coupling theory (MCT; see below) for this system. Here again, it is the
enough temperature there is a single relaxation mechanism. In the moderately manner in which the system samples its landscape, not the landscape
supercooled regime the peak splits into slow (a) and fast (b) relaxations, of which itself, that changes with temperature. (See ref. 74 for a recent, different
a-processes exhibit non-Arrhenius temperature dependence and vanish at Tg. interpretation of landscape sampling at this temperature.)
(Adapted from refs 9 and 41.) The landscape picture provides a natural separation of low-
temperature molecular motion into sampling distinct potential
NATURE | VOL 410 | 8 MARCH 2001 | www.nature.com © 2001 Macmillan Magazines Ltd 263
insight review articles
energy minima, and vibration within a minimum. It is possible to archetypal landscapes. Aside from multiplicity due to permutational
separate formally the corresponding configurational and vibrational symmetry, strong landscapes may consist of a single ‘megabasin’,
contributions to a liquid’s properties75,76. In two important computa- whereas fragile ones display a proliferation of well-separated
tional studies, the configurational entropy was calculated by probing ‘megabasins’ (Fig. 8).
systematically the statistics governing the sampling of potential Cooperative rearrangements such as those that must occur in
energy minima77,78 (Box 2). Using this technique, a remarkable OTP are unlikely to consist of elementary transitions between adja-
connection between configurational entropy and diffusion was cent basins. Rather, the likely scenario involves a complicated
identified in liquid water79. One of water’s distinguishing anomalies sequence of elementary transitions. At low temperatures, these
is the fact that, at sufficiently low temperature, its diffusivity increases rearrangements should be rare and long-lived on the molecular
upon compression80. As shown in Fig. 7, diffusivity maxima are timescale. Furthermore, the diversity of deep landscape traps and of
correlated strongly with configurational entropy maxima, the the pathways of configuration space that connect them should result
respective loci coinciding within numerical error. in a broad spectrum of relaxation times, as required for the stretched
The results shown in Fig. 7 and the success of the Adam–Gibbs exponential function in equation (2). This in turn suggests that
equation in describing experimental data on relaxation in a wide supercooled fragile liquids are dynamically heterogeneous, probably
variety of systems52 indicate that there exists a scaling relationship consisting at any instant of mostly non-diffusing molecules with a
between the depth distribution of basins and the height of the saddle few ‘hot spots’ of mobile molecules. This dynamic heterogeneity39
points along paths connecting neighbouring basins. Such scaling is has both experimental29,30,36 and computational31–35 support.
not a mathematical necessity, but arises from the nature of real The inverse relation between the self-diffusion coefficient and
molecular interactions. The topographic nature of this statistical viscosity embodied in the Stokes–Einstein equation is based on
scaling relationship between minima and saddle points is poorly macroscopic hydrodynamics that treats the liquid as a continuum.
understood (but see the recent computational investigation of saddle
points74). Its elucidation will explain the origin of the connection
Landscape
between the dynamics and thermodynamics of glass-forming liq- dominated
uids, and constitutes the principal theoretical challenge in this field. Landscape
Glass influenced Free diffusion
-6.85
Strong versus fragile behaviour
The extent to which the shear viscosity h deviates from Arrhenius
behaviour, h4hoexp(E/kBT), constitutes the basis of the classifica-
tion of liquids as either strong or fragile (Fig. 2). Molten SiO2, often -6.90
Potential energy per atom

considered as the prototypical strong glass-former, displays an

almost constant activation energy of 180 kcal mol11 (ref. 81). This
constancy indicates that the underlying mechanism, presumably
breaking and reformation of Si–O bonds, applies throughout the -6.95
Fast
entire landscape4. In contrast, the viscosity of OTP — the canonical cooling
fragile glass-former — deviates markedly from Arrhenius
behaviour82, showing an effective activation energy (dlnh/d1/T) that -7.00
increases 20-fold, from one-quarter of the heat of vaporization for
Slow
the liquid above its melting point to roughly five times the heat of cooling
vaporization near Tg. This means that OTP’s landscape is very hetero-
geneous. The basins sampled at high temperature allow relaxation by -7.05
surmounting low barriers involving the rearrangement of a small 0.0 0.5 1.0 1.5 2.0
Temperature
number of molecules. The very large activation energy at T ≈ Tg, on
the other hand, corresponds to the cooperative rearrangement of
many molecules. These differences between strong and fragile behav- Figure 6 Mean inherent structure energy per particle of a binary mixture of unequal-
iour imply a corresponding topographic distinction between the two sized Lennard–Jones atoms, as a function of the temperature of the equilibrated liquid
from which the inherent structures were generated by energy minimization. Molecular
dynamics simulations at constant energy and density were performed over a range of
Transition temperatures for 256 Lennard–Jones atoms, of which 20% are of type A and 80%
Basin states are of type B. The Lennard–Jones size and energy parameters are sAA41,
sBB40.88, sAB40.8, and ;AA41, ;BB40.5, ;AB41.5, respectively. Length,
temperature, energy and time are expressed in units of sAA, ;AA/kB, ;AA and
sAA(m/;AA )1/2, respectively, with m representing the mass of the particles. Simulations
were performed at a density of 1.2. The fast and slow cooling rates are 1.0821013
Potential energy

and 3.3321016. When T > 1, the system has sufficient kinetic energy to sample the
entire energy landscape, and the overwhelming number of sampled energy minima
are shallow. Under these conditions, the system exhibits a temperature-independent
activation energy for structural relaxation (calculations not shown). Between T41
and T ≈ 0.45, the activation energy increases upon cooling, the dynamics become
‘landscape-influenced’, and the mechanically stable configurations sampled are
Crystal strongly temperature-dependent. Below T ≈ 0.45, the height of the barriers
Ideal glass
separating sampled adjacent energy minima seems to increase abruptly
(calculations not shown). This is the ‘landscape-dominated’ regime. In it, particles
Coordinates execute rare jumps over distances roughly equal to interparticle separations. The
crossover between landscape-influenced and landscape-dominated behaviour
Figure 5 Schematic illustration of an energy landscape. The x-axis represents all corresponds closely with the mode-coupling transition temperature70,92. (Adapted
configurational coordinates. (Adapted from ref. 44.) from refs 70 and 72.)

264 © 2001 Macmillan Magazines Ltd NATURE | VOL 410 | 8 MARCH 2001 | www.nature.com
 insight review articles
This picture must clearly break down in supercooled fragile liquids,
which are dynamically heterogeneous. The failure of the Stokes– Figure 8 Schematic
Einstein equation, referred to above as one of the distinguishing representation of the
characteristics of fragile supercooled liquids, is therefore qualitatively energy landscapes of Strong
understandable. Plausible models for the low-temperature enhance- strong and fragile
ment of diffusive motion relative to hydrodynamic expectations based substances. The
on the viscosity have been proposed83–85, but an accurate predictive potential energy
theory is missing. The landscape viewpoint also provides a plausible increases vertically, and
interpretation for the a/b-relaxation decoupling shown in Fig. 3 — the horizontal direction
a-relaxations correspond to configurational sampling of neighbour- represents collective
ing megabasins (Fig. 8), whereas b-processes are thought to configurational Crystal
correspond to elementary relaxations between contiguous basins44. coordinates.
Fragile
Direct computational evidence of this interpretation is not available.

Avoided singularities
Alternative viewpoints to the landscape perspective have also con-
tributed to current understanding of some aspects of supercooling
and the glass transition. Two such interpretations invoke a narrowly
avoided singularity above Tg. Ideal glass
According to MCT86, structural arrest occurs as a result of the Crystal
following feedback mechanism: (i) shear-stress relaxation occurs
primarily through diffusive motion; (ii) diffusion and viscosity are
inversely related; and (iii) viscosity is proportional to shear-stress structural arrest occurs as a purely dynamic singularity, that is to say it
relaxation time. These facts lead to a viscosity feedback whereby is not accompanied by thermodynamic signatures such as a diverging
correlation length. What is now known as the idealized MCT87,88
predicts structural arrest to occur at a temperature Tx. Initially,
a therefore, it was thought that MCT was a useful theory for the
laboratory-generated glass transition. It is now widely understood
3
that this is not the case, as one finds that Tx > Tg, and the MCT-
predicted singularity does not occur. In subsequent modifications of
the theory89, additional relaxation mechanisms occur, often referred
to as ‘hopping’ or activated motions, which restore ergodicity (the
system’s ability to sample all configurations) below Tx, thereby avoid-
Scont/kB

2 ing a kinetic singularity. These additional relaxation modes arise as a

result of a coupling between fluctuations in density and momentum.
Although not a theory of the glass transition, MCT accurately
describes many important aspects of relaxation dynamics in liquids
above or moderately below their melting temperatures. In particular,
1 the theory makes detailed predictions about the behaviour of the
intermediate scattering function F, an experimentally observable
b 101 quantity that measures the decay of density fluctuations. After a fast
initial decay due to microscopic intermolecular collisions, MCT
predicts that the decay of F obeys the following sequence (Fig. 9):
100 (i) power-law decay towards a plateau, according to F4f&At1a; (ii) a
second power-law decay away from the plateau value F4f1Bt b; and
D (10-5cm2 s-1)

(iii) slow relaxation at longer times, which can be fitted by the KWW
10-1
function F4exp[1(t/t)b]. Here, f is the plateau value of the scatter-
ing function, which only appears at sufficiently low temperature; t is
10-2 time; A, B, a and b are constants; t is the characteristic, temperature-
dependent relaxation time; and b < 1 is the KWW stretch exponent.
The basic accuracy of these detailed predictions has been verified
10-3 experimentally and in computer simulations90–92.
Kivelson and co-workers have proposed a theory of supercooled
liquids that is based also on an avoided singularity24,93–95. According to
10-4 this viewpoint, the liquid has an energetically preferred local struc-
0.9 1.0 1.1 1.2 1.3 1.4
ture that differs from the structure in the actual crystalline phase. The
ρ (g cm-3)
system is prevented from crystallizing into a reference crystal with the
preferred local structure because of geometric frustration owing to
Figure 7 Relationship between diffusivity (D ) and configurational entropy (Sconf) of the fact that the latter does not tile space. An example of such energet-
supercooled water79 at six different temperatures. Filled and open symbols from top to ically favoured but non-space-tiling local structure is the icosahedral
bottom represent the following temperatures: 300 K, 260 K, 240 K, 230 K, 220 K and packing seen in computer simulations of the supercooled
210 K; r is density. The configurational entropy, which is related to the number of Lennard–Jones liquid73. At a temperature T* the system would, but
potential energy minima of a given depth (Box 2), was calculated by subtracting the for frustration, crystallize into the reference crystal. Instead, strain
vibrational contribution from the total entropy. The calculations involved performing build-up causes the system to break up into frustration-limited
molecular dynamics simulations of the extended simple point charge (SPC/E) model of domains, thereby avoiding a phase transition (singularity) at T*. The
water102 over a range of temperatures and densities. (Adapted from ref. 79.) avoided transition temperature T* acts as a critical point, below
which two length scales emerge, both of which are large compared to
NATURE | VOL 410 | 8 MARCH 2001 | www.nature.com © 2001 Macmillan Magazines Ltd 265
insight review articles
treatments of the vitreous state based on invoking analogies to
spin glasses99–101. Establishing a coherent theoretical perspective on
1
supercooled liquids and glasses is important. We believe that the
landscape formalism offers the natural technical tools for
0.8
accomplishing this task. ■
1. Angell, C. A. Formation of glasses from liquids and biopolymers. Science 267, 1924–1935 (1995).
2. Blanshard, J. M. V. & Lillford, P. (eds) The Glassy State in Foods (Nottingham Univ. Press,
T = 0.466 Nottingham, 1993).
0.6
Fs(q,t)

3. Crowe, J. H., Carpenter, J. F. & Crowe, L. M. The role of vitrification in anhydrobiosis. Annu. Rev.
Physiol. 60, 73–103 (1998).
4. Debenedetti, P. G., Stillinger, F. H., Truskett, T. M. & Lewis, C. P. Theory of supercooled liquids and
0.4 glasses: energy landscape and statistical geometry perspectives. Adv. Chem. Eng. (in the press).
5. Greer, A. L. Metallic glasses. Science 267, 1947–1953 (1995).
T = 5.0
6. Jenniskens, P. & Blake, D. F. Structural transitions in amorphous water ice and astrophysical
0.2 implications. Science 265, 753–756 (1994).
7. Anderson, P. W. Through a glass lightly. Science 267, 1615 (1995).
8. Angell, C. A., Ngai, K. L., McKenna, G. B., McMillan, P. F. & Martin, S. W. Relaxation in
glassforming liquids and amorphous solids. J. Appl. Phys. 88, 3113–3157 (2000).
0.0 9. Debenedetti, P. G. Metastable Liquids. Concepts and Principles (Princeton Univ. Press, Princeton, 1996).
10–2 10–1 100 101 102 103 104 105 10. Turnbull, D. Under what conditions can a glass be formed? Contemp. Phys. 10, 473–488 (1969).
t 11. Angell, C. A. Structural instability and relaxation in liquid and glassy phases near the fragile liquid
limit. J. Non-Cryst. Solids 102, 205–221 (1988).
12. Moynihan, C. T. et al. in The Glass Transition and the Nature of the Glassy State (eds Goldstein, M. &
Figure 9 Evolution of the self-intermediate scattering function for A-type atoms for
Simha, R.) Ann. NY Acad. Sci. 279, 15–36 (1976).
the same supercooled Lennard–Jones mixture as in Fig. 6, at qsAA47.251, 13. Brüning, R. & Samwer, K. Glass transition on long time scales. Phys. Rev. B 46, 318–322 (1992).
corresponding to the first peak of the static structure factor of species A (ref. 92). Here 14. Ediger, M. D., Angell, C. A. & Nagel, S. R. Supercooled liquids and glasses. J. Phys. Chem. 100,
q is the magnitude of the wave vector. Temperature and time are in units of ;AA/kB and 13200–13212 (1996).
15. Vogel, H. Das temperatur-abhängigkeitsgesetz der viskosität von flüssigkeiten. Phys. Zeit. 22,
sAA(m/48;AA)1/2, respectively. Temperatures from left to right are 5, 4, 3, 2, 1, 0.8,
645–646 (1921).
0.6, 0.55, 0.5, 0.475, and 0.466. The self-intermediate scattering function is the 16. Tammann, G. & Hesse, W. Die abhängigkeit der viskosität von der temperatur bei unterkühlten
space Fourier transform of the van Hove function Gs(r, t ), which is proportional to the flüssigkeiten. Z. Anorg. Allg. Chem. 156, 245–257 (1926).
probability of observing a particle at r5dr at time t given that the same particle was at 17. Fulcher, G. S. Analysis of recent measurements of the viscosity of glasses. J. Am. Ceram. Soc. 8, 339 (1925).
18. Laughlin, W. T. & Uhlmann, D. R. Viscous flow in simple organic liquids. J. Phys. Chem. 76,
the origin at t = 0. Note the two-step relaxation behaviour upon decreasing T. 2317–2325 (1972).
Molecular dynamics simulations of 1,000 atoms. (Adapted from refs 9 and 92.) 19. Angell, C. A. in Relaxations in Complex Systems (eds Ngai, K. & Wright, G. B.) 1 (Natl Technol.
Inform. Ser., US Dept. of Commerce, Springfield, VA, 1985).
20. Angell, C. A. Relaxation in liquids, polymers and plastic crystals—strong/fragile patterns and
problems. J. Non-Cryst. Solids 131–133, 13–31 (1991).
molecular dimensions. One is the critical correlation length, which 21. Green, J. L., Ito, K., Xu, K. & Angell, C. A. Fragility in liquids and polymers: new, simple
governs density fluctuations in the absence of frustration. The quantifications and interpretations. J. Phys. Chem. B 103, 3991–3996 (1999).
second is the frustration-limited domain size. From these 22. Novikov, V. N., Rössler, E., Malinovsky, V. K. & Surovstev, N. V. Strong and fragile liquids in
percolation approach to the glass transition. Europhys. Lett. 35, 289–294 (1996).
considerations there emerge predictions on the temperature 23. Fujimori, H. & Oguni, M. Correlation index (Tga1Tgb)/Tga and activation energy ratio D;aa/D;ab as
dependence of the viscosity. Experimental data analysed according to parameters characterizing the structure of liquid and glass. Solid State Commun. 94, 157–162 (1995).
the theory display universality24, but at the expense of introducing a 24. Kivelson, D., Tarjus, G., Zhao, X. & Kivelson, S. A. Fitting of viscosity: distinguishing the
number of fitting parameters. The improvement with respect to temperature dependencies predicted by various models of supercooled liquids. Phys. Rev. E 53,
751–758 (1996).
competing interpretations is a matter of controversy25. 25. Cummins, H. Z. Comment on “Fitting of viscosity: distinguishing the temperature dependencies
predicted by various models of supercooled liquids”. Phys. Rev. E 54, 5870–5872 (1996).
Challenges and open questions 26. Kohlrausch, R. Theorie des elektrischen rückstandes in der leidener flasche. Ann. Phys. Chem.
(Leipzig) 91, 179–214 (1874).
Important aspects of the complex behaviour of viscous liquids close
27. Williams, G. & Watts, D. C. Non-symmetrical dielectric relaxation behavior arising from a simple
to the glass transition can be explained qualitatively from the energy empirical decay function. Trans. Faraday Soc. 66, 80–85 (1970).
landscape perspective. Making this descriptive picture quantitative 28. Richert, R. & Blumen, A. in Disorder Effects on Relaxational Processes (eds Richert , R. & Blumen, A.)
and predictive is a major challenge. This will require investigating 1–7 (Springer, Berlin, 1994).
29. Cicerone, M. T. & Ediger, M. D. Relaxation of spatially heterogeneous dynamic domains in
how basic landscape features such as the basin enumeration function supercooled ortho-terphenyl. J. Chem. Phys. 103, 5684–5692 (1995).
depend on molecular architecture and, for a given substance or mix- 30. Cicerone, M. T. & Ediger, M. D. Enhanced translation of probe molecules in supercooled o-
ture, on density (see ref. 96 for an example of such a calculation). terphenyl: signature of spatially heterogeneous dynamics? J. Chem. Phys. 104, 7210–7218 (1996).
Equally important is the translation of qualitative pictures such as 31. Mel’cuk, A. I., Ramos, R. A., Gould, H., Klein, W. & Mountain, R. D. Long-lived structures in fragile
glass-forming liquids. Phys. Rev. Lett. 75, 2522–2525 (1995).
Fig. 8 into precise measures of strength and fragility based on the 32. Hurley, M. M. & Harrowell, P. Non-gaussian behavior and the dynamical complexity of particle
basin enumeration function. Uncovering the topographic nature of motion in a dense two-dimensional liquid. J. Chem. Phys. 105, 10521–10526 (1996).
the scaling relationship between basin minima and saddle points 33. Perera, D. N. & Harrowell, P. Measuring diffusion in supercooled liquids: the effect of kinetic
inhomogeneities. J. Chem. Phys. 104, 2369–2375 (1996).
holds the key to understanding the relationship between kinetics and
34. Perera, D. N. & Harrowell, P. Consequence of kinetic inhomogeneities in glasses. Phys. Rev. E 54,
thermodynamics in deeply supercooled liquids. All of these 1652–1662 (1996).
calculations are in principle straightforward, but computationally 35. Donati, C., Glotzer, S. C., Poole, P. H., Kob, W. & Plimpton, S. J. Spatial correlations of mobility and
at the very limit of what is currently feasible. The development of immobility in a glass-forming Lennard-Jones liquid. Phys. Rev. E 60, 3107–3119 (1999).
36. Böhmer, R., Hinze, G., Diezemann, G., Geil, B. & Sillescu, H. Dynamic heterogeneity on
theoretical models97 is therefore of paramount importance. supercooled ortho-terphenyl studied by multidimensional deuteron NMR. Europhys. Lett. 36,
MCT and the landscape perspective offer complementary view- 55–60 (1996).
points of the same phenomena. So far, however, not enough effort has 37. Wang, C.-Y. & Ediger, M. D. How long do regions of different dynamics persist in supercooled o-
been devoted to bridging the gap that separates these two approaches. terphenyl? J. Phys. Chem. B 103, 4177–4184 (1999).
38. Vidal Russell, E. & Israeloff, N. E. Direct observation of molecular cooperativity near the glass
Recent calculations70,74 offer the promise of establishing a clearer transition. Nature 408, 695–698 (2000).
connection between the static landscape viewpoint and the dynamic 39. Ediger, M. D. Spatially heterogeneous dynamics in supercooled liquids. Annu. Rev. Phys. Chem. 51,
perspective of MCT. At the least, what is required is a precise 99–128 (2000).
40. Fujara, F., Geil, B., Sillescu, H. H. & Fleischer, G. Translational and rotational diffusion in
landscape-based explanation of what ‘hopping’ and ‘activated
supercooled orthoterphenyl close to the glass transition. Z. Phys. B Cond. Matt. 88, 195–204 (1992).
processes’ really mean. Additional theoretical viewpoints of 41. Johari, G. P. Intrinsic mobility of molecular glasses. J. Chem. Phys. 58, 1766–1770 (1973).
supercooling and the glass transition include the instantaneous 42. Johari, G. P. & Goldstein, M. Viscous liquids and the glass transition. II. Secondary relaxations in
normal-mode perspective on liquid dynamics98 and thermodynamic glasses of rigid molecules. J. Chem. Phys. 53, 2372–2388 (1970).

266 © 2001 Macmillan Magazines Ltd NATURE | VOL 410 | 8 MARCH 2001 | www.nature.com
 insight review articles
43. Rössler, E., Warschewske, U., Eiermann, P., Sokolov, A. P. & Quitmann, D. Indications for a change isothermal compressibility. J. Chem. Phys. 109, 3983–3988 (1998).
of transport mechanism in supercooled liquids and the dynamics close and below Tg. J. Non-Cryst. 76. Stillinger, F. H. & Debenedetti, P. G. Distinguishing vibrational and structural equilibration
Solids 172–174, 113–125 (1994). contributions to thermal expansion. J. Phys. Chem. B 103, 4052–4059 (1999).
44. Stillinger, F. H. A topographic view of supercooled liquids and glass formation. Science 267, 77. Sciortino, F., Kob, W. & Tartaglia, P. Inherent structure entropy of supercooled liquids. Phys. Rev.
1935–1939 (1995). Lett. 83, 3214–3217 (1999).
45. Kauzmann, W. The nature of the glassy state and the behavior of liquids at low temperatures. Chem. 78. Büchner, S. & Heuer, A. Potential energy landscape of a model glass former: thermodynamics,
Rev. 43, 219–256 (1948). anharmonicities, and finite size effects. Phys. Rev. E 60, 6507–6518 (1999).
46. Simon, F. Über den zustand der unterkühlten flüssigkeiten und glässer. Z. Anorg. Allg. Chem. 203, 79. Scala, A., Starr, F. W., La Nave, E., Sciortino, F. & Stanley, H. E. Configurational entropy and
219–227 (1931). diffusivity in supercooled water. Nature 406, 166–169 (2000).
47. Wolynes, P. G. Aperiodic crystals: biology, chemistry and physics in a fugue with stretto. AIP Conf. 80. Prielmeier, F. X., Lang, E. W., Speedy, R. J. & Lüdemann, H.-D. Diffusion in supercooled water to
Proc. 180, 39–65 (1988). 300 Mpa. Phys. Rev. Lett. 59, 1128–1131 (1987).
48. Wolynes, P. G. Entropy crises in glasses and random heteropolymers. J. Res. Natl Inst. Standards 81. Mackenzie, J. D. Viscosity-temperature relationship for network liquids. J. Am. Ceram. Soc. 44,
Technol. 102, 187–194 (1997). 598–601 (1961).
49. Angell, C. A. Landscapes with megabasins: polyamorphism in liquids and biopolymers and the role 82. Greet, R. J. & Turnbull, D. Glass transition in o-terphenyl. J. Chem. Phys. 46, 1243–1251 (1967).
of nucleation in folding and folding diseases. Physica D 107, 122–142 (1997). 83. Stillinger, F. H. & Hodgdon, J. A. Translation-rotation paradox for diffusion in fragile glass-forming
50. Gibbs, J. H. & DiMarzio, E. A. Nature of the glass transition and the glassy state. J. Chem. Phys. 28, liquids. Phys. Rev. E 50, 2064–2068 (1994).
373–383 (1958). 84. Tarjus, G. & Kivelson, D. Breakdown of the Stokes-Einstein relation in supercooled liquids. J. Chem.
51. Adam, G. & Gibbs, J. H. On the temperature dependence of cooperative relaxation properties in Phys. 103, 3071–3073 (1995).
glass-forming liquids. J. Chem. Phys. 43, 139–146 (1965). 85. Liu, C. Z.-W. & Openheim, I. Enhanced diffusion upon approaching the kinetic glass transition.
52. Richert, R. & Angell, C. A. Dynamics of glass-forming liquids. V. On the link between molecular Phys. Rev. E 53, 799–802 (1996).
dynamics and configurational entropy. J. Chem. Phys. 108, 9016–9026 (1998). 86. Geszti, T. Pre-vitrification by viscosity feedback. J. Phys. C 16, 5805–5814 (1983).
53. Williams, M. L., Landel, R. F. & Ferry, J. D. The temperature dependence of the relaxation mechanisms 87. Bengtzelius, U., Götze, W. & Sjölander, A. Dynamics of supercooled liquids and the glass transition.
in amorphous polymers and other glass-forming liquids. J. Am. Chem. Soc. 77, 3701–3707 (1955). J. Phys. C 17, 5915–5934 (1984).
54. Ito, K., Moynihan, C. T. & Angell, C. A. Thermodynamic determination of fragility in liquids and a 88. Götze, W. & Sjögren, L. Relaxation processes in supercooled liquids. Rep. Prog. Phys. 55, 241–376 (1992).
fragile-to-strong liquid transition in water. Nature 398, 492–495 (1999). 89. Götze, W. & Sjögren, L. The mode coupling theory of structural relaxations. Transp. Theory Stat.
55. Goldstein, M. Viscous liquids and the glass transition: a potential energy barrier picture. J. Chem. Phys. 24, 801–853 (1995).
Phys. 51, 3728–3739 (1969). 90. Götze, W. Recent tests of the mode-coupling theory for glassy dynamics. J. Phys. Cond. Matt. 11,
56. Frauenfelder, H., Sligar, S. G. & Wolynes, P. G. The energy landscapes and motions of proteins. A1–A45 (1999).
Science 254, 1598–1603 (1991). 91. Kob, W. Computer simulations of supercooled liquids and glasses. J. Phys. Cond. Matt. 11,
57. Nienhaus, G. U., Müller, J. D., McMahon, B. H. & Frauenfelder, H. Exploring the conformational R85–R115 (1999).
energy landscape of proteins. Physica D 107, 297–311 (1997). 92. Kob, W. & Andersen, H. C. Testing mode-coupling theory for a supercooled binary Lennard-Jones
58. Abkevich, V. I., Gutin, A. M. & Shakhnovich, E. I. Free energy landscape for protein folding kinetics: mixture: the van Hove correlation function. Phys. Rev. E 51, 4626–4641 (1995).
intermediates, traps, and multiple pathways in theory and lattice model simulations. J. Chem. Phys. 93. Kivelson, D., Kivelson, S. A., Zhao, X., Nussinov, Z. & Tarjus, G. A thermodynamic theory of
101, 6052–6062 (1994). supercooled liquids. Physica A 219, 27–38 (1995).
59. Saven, J. G., Wang, J. & Wolynes, P. G. Kinetics of protein folding: the dynamics of globally 94. Kivelson, D. & Tarjus, G. SuperArrhenius character of supercooled glass-forming liquids. J. Non-
connected rough energy landscapes with biases. J. Chem. Phys. 101, 11037–11043 (1994). Cryst. Solids 235–237, 86–100 (1998).
60. Wang, J., Onuchic, J. & Wolynes, P. Statistics of kinetic pathways on biased rough energy landscapes 95. Kivelson, D. & Tarjus, G. The Kauzmann paradox interpreted via the theory of frustration-limited
with applications to protein folding. Phys. Rev. Lett. 76, 4861–4864 (1996). domains. J. Chem. Phys. 109, 5481–5486 (1998).
61. Plotkin, S. S., Wang, J. & Wolynes, P. G. Correlated energy landscape model for finite, random 96. Sastry, S. The relationship between fragility, configurational entropy and the potential energy
heteropolymers. Phys. Rev. E 53, 6271–6296 (1996). landscape of glass-forming liquids. Nature 409, 164–167 (2001).
62. Becker, O. M. & Karplus, M. The topology of multidimensional potential energy surfaces: theory 97. Speedy, R. J. Relations between a liquid and its glasses. J. Phys. Chem. B 103, 4060–4065 (1999).
and application to peptide structure and kinetics. J. Chem. Phys. 106, 1495–1517 (1997). 98. Keyes, T. Instantaneous normal mode approach to liquid state dynamics. J. Phys. Chem. A 101,
63. Dill, K. A. & Chan, H. S. From Levinthal to pathways and funnels. Nature Struct. Biol. 4, 10–19 (1997). 2921–2930 (1997).
64. Klepeis, J. L., Floudas, C. A., Morikis, D. & Lambris, J. D. Predicting peptide structure using NMR 99. Kirkpatrick, T. R. & Wolynes, P. G. Stable and metastable states in mean-field Potts and structural
data and deterministic global optimization. J. Comp. Chem. 20, 1354–1370 (1999). glasses. Phys. Rev. B 36, 8552–8564 (1987).
65. Lacks, D. J. Localized mechanical instabilities and structural transformations in silica glass under 100. Kirkpatrick, T. R., Thirumalai, D. & Wolynes, P. G. Scaling concepts for the dynamics of viscous
high pressure. Phys. Rev. Lett. 80, 5385–5388 (1998). liquids near an ideal glassy state. Phys. Rev. A 40, 1045–1054 (1989).
66. Malandro, D. L. & Lacks, D. J. Volume dependence of potential energy landscapes in glasses. J. 101. Mézard, M. & Parisi, G. Thermodynamics of glasses: a first principles computation. Phys. Rev. Lett.
Chem. Phys. 107, 5804–5810 (1997). 82, 747–750 (1999).
67. Malandro, D. L. & Lacks, D. J. Relationships of shear-induced changes in the potential energy 102. Berendsen, H. J., Grigera, J. R. & Stroatsma, T. P. The missing term in effective pair potentials. J.
landscape to the mechanical properties of ductile glasses. J. Chem. Phys. 110, 4593–4601 (1999). Phys. Chem. 91, 6269–6271 (1987).
68. Malandro, D. L. & Lacks, D. J. Molecular-level instabilities and enhanced self-diffusion in flowing 103. Stillinger, F. H. Supercooled liquids, glass transitions, and the Kauzmann paradox. J. Chem. Phys. 88,
liquids. Phys. Rev. Lett. 81, 5576–5579 (1998). 7818–7825 (1988).
69. Schulz, M. Energy landscape, minimum points, and non-Arrhenius behavior of supercooled 104. Santen, L. & Krauth, W. Absence of thermodynamic phase transition in a model glass former.
liquids. Phys. Rev. B 57, 11319–11333 (1998). Nature 405, 550–551 (2000).
70. Sastry, S., Debenedetti, P. G. & Stillinger, F. H. Signatures of distinct dynamical regimes in the energy 105. Wilks, J. The Properties of Liquid and Solid Helium (Clarendon, Oxford, 1967).
landscape of a glass-forming liquid. Nature 393, 554–557 (1998). 106. Rastogi, S., Höhne, G. W. H. & Keller, A. Unusual pressure-induced phase behavior in crystalline
71. Keyes, T. Dependence of supercooled liquid dynamics on elevation in the energy landscape. Phys. Poly(4-methylpentene-1): calorimetric and spectroscopic results and further implications.
Rev. E 59, 3207–3211 (1999). Macromolecules 32, 8897–8909 (1999).
72. Debenedetti, P. G., Stillinger, F. H., Truskett, T. M. & Roberts, C. J. The equation of state of an energy 107. Greer, A. L. Too hot to melt. Nature 404, 134–135 (2000).
landscape. J. Phys. Chem. B 103, 7390–7397 (1999). 108. Stillinger, F. H. Exponential multiplicity of inherent structures. Phys. Rev. E 59, 48–51 (1999).
73. Jonsson, H. & Andersen, H. C. Icosahedral ordering in the Lennard-Jones crystal and glass. Phys. 109. Stillinger, F. H. Enumeration of isobaric inherent structures for the fragile glass former o-terphenyl.
Rev. Lett. 60, 2295–2298 (1988). J. Phys. Chem. B 102, 2807–2810 (1998).
74. Angelani, L., Di Leonardo, R., Ruocco, G., Scala, A. & Sciortino, F. Saddles in the energy landscape
probed by supercooled liquids. Phys. Rev. Lett. 85, 5356–5359 (2000). Acknowledgements
75. Stillinger, F. H., Debenedetti, P. G. & Sastry, S. Resolving vibrational and structural contributions to P.G.D.’s work is supported by the US Department of Energy.

NATURE | VOL 410 | 8 MARCH 2001 | www.nature.com © 2001 Macmillan Magazines Ltd 267