Frustration-limited clusters in liquids Steven A. Kivelson Department of Physics, Universiq of California, Los Angeles, California 90024 Xiaolin Zhao, Daniel Kivelson, Thomas M. Fischer, and Charles M. Knobler Department of Chemistry and Biochemistry, University of California, Los Angeles, California 90024 (Received 10 January 1994; accepted 13 April 1994) We present a continuum theory of frustration-limited clusters in one-component glass-forming liquids that accounts, in part, for the recently reported [Fischer et al., J. Non-Cryst. Solids, 131133, 134 (1991)], and quite unexpected, presence in simple glass-forming liquids of stable clusters at low temperatures (T) and the even less expected persistence for very long times of these clusters at higher Ts. The model is based on the idea that there is a local structure that is energetically preferred over simple crystalline packing, which is strained (frustrated) over large distances; although in a curved space the preferred packing could lead to “ideal” crystallization at temperatures that are usually above the actual freezing temperature, in “flat” space this transition is narrowly avoided. We are led to a new ansatz for the T dependence of the viscosity, which permits us to collapse data for many liquids onto a universal curve. I. INTRODUCTION There is still no definitive understanding of glass formation. However, there is a body of thought that suggests that it is a consequence of geometric frustration.‘-5 In this view, a glass-forming substance has a preferred local structure into which it could crystallize in an appropriately curved space, but in three-dimensional space the strain associated with this “ideal structure” would diverge with the size of the system. This notion of geometric frustration has been useful for understanding the appearance of large-scale structures in a number of materials, such as Frank-Kasper phases in transition-metal alloys and the blue phases in liquid crystal~.~ However, no direct evidence for the role of geometric frustration has been reported for glass-forming materials. In this communication we show that this model can lead to “frustration-limited cluster” formation in the liquid: In equilibrium the concentration of clusters is low at high temperatures, T, and high at low T, but metastable high concentrations can be found even at high T’s. We further show that the model provides a simple and natural explanation for at least some of the observed6*7 light scattering anomalies in glassforming liquids that have been attributed by Fischer et aL6 to the presence of clusters; in particular, it provides a rationalization for the existence of finite clusters in equilibrium, and for the very long persistence of metastable cluster concentrations at temperatures well above the freezing point. Moreover, in turn, the success of this explanation provides support for the idea that frustration, and perhaps clusters, play major roles in glass formation. In their light scattering studies of the glass-feting liquid orthoterphenyl (OTP), Fischer et ~1.~ observed that the turbidity and Landau-Placzek ratio were greatly enhanced in the deeply supercooled liquid. The growth in the scattering is not observed immediately, but develops over hours. These observations suggest the formation of large “clusters,” a supposition that is further supported by experiment8 that show that the integrated intensity and the Landau-Placzek ratio dependupon scatteringangle,i.e., upon wave number q. Such behavior could be explained by a number of altema- tive models. However, Fischer et al. also observed that the enhanced scattering that develops at low temperatures persists for days after the system is heated well above the melting point. This latter observation was unexpected and remained unexplained. In our view, this metastability reflects the fact that even above the physical freezing point, the system can be well below the “ideal freezing temperature”, Tif, at which it would freeze into the preferred local structure in the appropriately curved space.8 Motivated by the results of Fischer ef ~1.~ on OTP, and initially somewhat sceptical of their findings (particularly the long persistence of clusters above the melting point), we have carried out detailed and comprehensive studies on a second good glass-forming neat fragile liquid, triphenyl phosphite (TPP);7 our results are in most ways similar to those of Fischer et al., but because of serious hysteresis effects and slow equilibration, we do not yet consider our work definitive. Nevertheless, we include here a description of some of our new experimental findings that seem particularly pertinent to the understanding of molecular glass-forming liquids. [A fragile liquid’ is defined to be one whose viscosity has an activation energy, E7( T), that increases greatly as the temperature is lowered from the melting point to the glass transition temperature.] II. THEORY We postulate that in a good molecular glass-forming liquid there exists a locally preferred structure, such as a closepacking configuration, that cannot be extended throughout space. This concept has been investigated by a series of workers, among them Bernal,” Boerdijk,” Frank and Kasper,12 Hoare,13 and Stillinger,5 who have recognized that the local-preferred structure of spherical particles is icosahedral, but that the extension of this structure is thwarted by frustration. Said differently, the inability to extend the locally preferred configuration throughout space is a kind of frustration. The formation of locally preferred clusters is analogous J. Chem. Phys. 101 (3), 1 August 1994 0021-9606/94/101(3)/2391/7/$6.00 Q 1994see American institute of Physics 2391 Downloaded 27 Jul 2001 to 141.14.233.190. Redistribution subject to AIP license or copyright, http://ojps.aip.org/jcpo/jcpcr.jsp Kivelson et a/.: Frustration-limited 2392 clusters in liquids to the formation of nuclei in the crystallization process, but whereas in crystallization the nuclei can grow and ultimately extend over the entire system, in our model the growth is limited by the frustration. To incorporate this physics we write the free energy F of a single cluster of characteristic size L as F=aL’-- +L3+.s(L)L3. 0) The first two terms are those commonly encountered in classical theories of nucleation,14 the first representing the surface free energy required to construct a cluster. The second term represents the bulk free energy difference between the liquid and an “ideal crystal” which can be formed in a reference system in which the locally preferred structure can be extended globally; it has been shown that for spheres the reference system can be taken as a properly curved space, and that frustration arises from the mapping of the crystal in curved space onto flat space.lT4 The third term represents the effect of the frustration; it can be thought of as the strain energy associated with forcing this structure into “flat” three-dimensional space; this term is absent if the solid phase is an ordinary crystal. Equation (1) is a continuum representation of a cluster free energy in which the physics that takes place on a molecular length scale is neglected and which, therefore, is not likely to be applicable to small clusters; the model is also restricted to dilute solutions of clusters where cluster-cluster interactions can be ignored. The coefficient 4, is a measure of the free energy density gained by ordering in the locally preferred structure; it is positive at low temperatures and vanishes as the temperature is raised towards a temperature Z”, the “ideal freezing” temperature in the absence of frustration. Typically we expect Tif to be greater than the observed freezing temperature because the ideal crystal has, by assumption, lower free energy than the ordinary crystal. The surface-energy coefficient (T is positive and should, we believe, be less temperature dependent than 4. The strain coefficient s(L)=s,L2 for L small compared to the radius of curvature of the ideal space because at short distances any space is locally flat; this leads to a superextensive dependence of the strain energy on L. At sufficiently large L the strain coefficient s(L) must be relieved by defects, which will result in a positive saturation value s(m); this may result in some complex structure, perhaps a crystal of defects.4 Because the preferred local and the ideal crystal structures are, by assumption, quite similar, the freezing transition in the absence of frustration (e.g., in the curved space) is likely to be either weakly first order or continuous. It is convenient to rewrite Eq. (1) in terms of dimensionless variables: f=E2-d3+P, (2) where f = (F/F,,) is the reduced free energy, 1= L/L,,, Lo= ( C7/sop3, Pa= mL& and r= ( &a)Lo . A related expression for the free energy of frustrated domains has been discussed by Stillinger.5 At I= 0, the reduced free energy f(O) = 0, which is a local minimum at all values of 3-,i.e., at all temperatures. In our model, l=O represents the liquid. Below a temperature T, , which, in turn, lies below Tif, a second minimum in f( I) I FIG. 1. f vs I at different T’S[See Eq. (Z)]. appears at 1>O, and, of course, there is a maximum between the two minima. We take the value of 1=I, at the second minimum to be the reduced size of the cluster, and f(l,) is the reduced local free energy of formation of the cluster. As T is lowered, f( 1,) decreases, and for temperatures less than T2 one finds f(1,) <f(O) (see Fig. 1). As T is lowered below T, , the reduced cluster size 1, increases. Although the free energy of the cluster relative to that of the liquid has been calculated only along a specific growth/decomposition coordinate (1), the model suggests that the free energy must indeed have a local minimum at I= 1,. The model thus implies the existence of finite-sized equilibrated clusters, in contrast to kinetically limited clusters. The model indicates that just below T, , equilibrated liquids should contain a small fraction of (presumably small) clusters. As T is lowered towards T, , the size and number of clusters should increase. However, if the activation energy for cluster formation is large at temperatures above T,, the equilibrium concentration of clusters may be difficult to achieve. Below T, the equilibrium concentrations of the now larger clusters should be high, and since the activation free energy for cluster formation is then relatively small, these high concentrations should be readily achieved. In other words, surprisingly, equilibrium cluster concentrations are more rapidly obtained at low temperature than at high temperature. Moreover, if now the low-temperature system containing clusters is heated to temperatures above T,, but below T1, one should be able to achieve large metastable cluster concentrations which, because of the high cluster-toliquid barrier, decay only slowly. All this is, in fact, observedFT7 The very fact that a molecular fluid is a good glassformer suggests that our assumption that Tif lies above the freezing point is reasonable; in our model the liquid super- J. Chem. Phys., Vol. 101, No. 3, 1 August 1994 Downloaded 27 Jul 2001 to 141.14.233.190. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp Kivelson et al.: Frustration-limited clusters in liquids I 1 0.14 I I I I ;I t I - -- transition from metaslable io stable transition from stable to metastable : I : 0.10 T WV 0.08 I : 0 .’ % iii. 0.06 - High T,. : : : : Low -I ,,” 2393 dynamics of the cluster-to-liquid process could be described as a first order decomposition, one would expect exponential decay governed by a decay rate with Arrhenius dependence on T. Since the rate is very slow, one would expect a large activation free energy; in fact, however, it appear8 that above the melting point the cluster concentration decays with time approximately as (tell) 1’3 over 2.5 orders of magnitude, with a rather mild decrease of to with increasing T. Later we speculate on a mechanism for this decay, keeping in mind that this particular power-law decay may be specific to the TPP system we have studied. IV. SUMMARY OF CLUSTER STUDY 0.04 - 0.02 - 0.00 1.6 1.7 1.0 1.9 2.0 2.1 2.2 2.3 2 FIG. 2. Reducedactivation energy. cools because the local-preferred structure is locally much more stable than the crystal structure: this suggests that the activation energy for transformation from finite clusters to the infinite crystal is very high. One might expect that the better the glass forming potential of the material, the greater the difference between Ti, and the physical freezing point. The same reasoning suggests that for a very good glass former, T, , and even T, , may lie above the freezing point. In many cases the locally preferred structure should be closepacked, in which case cluster formation should be accompanied by a density increase; this is the case, for example, for the frustrated icosahedral close-packing of spheres, and is what we observe experimentally for TPP.7 III. DYNAMICS The model represented by Eqs. (1) and (2) provides a reasonable picture for the thermodynamic properties of clusters in a one-component liquid, but it cannot yield complete information concerning the dynamics. It does yield free energies of activation for the liquid-to-cluster and cluster-toliquid processes (see Fig. 2), but these free energies are relevant only to a particular reaction coordinate, one in which the size (L) of dilute clusters changes according to the rules of preferred packing. The fact that this activation free energy for the liquid-to-cluster process can be large at Ts above T2 can account for the fact that clusters do not readily form at high temperatures. The fact that the activation free energy for the cluster-to-liquid process can also be quite high can account for the fact that clusters, once formed at low temperatures, can persist in high metastable concentrations at temperatures above the melting point. Attractive as this picture may be, one can only say that these high free energy barriers block easy passage ‘along one likely reaction coordinate, thereby encouraging the cluster-liquid process to proceed along alternate, presumably less direct paths. Actually, if the The concept of clusters of one kind or another in onecomponent liquids, particularly in supercooled liquids, is not new.5,9-13**5However, the experiments of Fischer et al. on OTP, and the further experiments on TPP in our laboratory, represent the most convincing and tangible evidence to date for their existence. A number of theories have postulated the existence of clusters, and the consequent inhomogeneity of the liquid, in order to explain the observed stretched exponential relaxation behavior and consequent inhomogeniety of the fluid, but our theory, we believe, is the first to provide a theoretical description for the observed clusters in a onecomponent liquid. The model is based on the idea that there is a local structure that is energetically preferred over simple crystalline packing but which is strained (frustrated) over large distances. The model is consonant with pictures of the glass built upon the concept of just such locally preferred but macroscopically strained structures.1-5 Our theory has the central feature that the finite size of the clusters is an equilibrium, not a kinetic property. The principal achievement of the theory, besides its simplicity and physical motivation, lies in the fact that it not only accounts for the formation of clusters at low temperatures and their existence at high T (apparently even at equilibrium7), but it also rationalizes the astonishingly long-lived (days) metastable cluster concentrations observed at temperatures well above the freezing temperature. The stability of clusters, even above the freezing temperature, can be understood by noting that it is the ideal freezing temperature, Tif, or the cluster-tbreshhold temperature, T, , and not the actual freezing temperature, that is relevant to cluster formation, and clusters can exist provide T is below T, . The theory also explains why an equilibrium distribution of clusters cannot be readily prepared in a liquid that has not been previously supercooled, and why such a distribution can apparently be prepared by allowing the clusters to form slowly at low T. If the liquid is cooled to temperatures below T2 and then reheated before large clusters have time to form, no high-T clusters are observed.7 Or if the liquid is deeply supercooled, allowed to crystallize, and then heated, no high-T metastable clusters are observed.7 The model explains these observations by specifying that the activation barrier for cluster formation is high at temperatures above T, and low below T,. As the clusters grow in a low-T, high-viscosity system, the observed density increases7 which suggests that indeed the stable clusters represent a locally close-packed configu- J. Chem. Phys., Vol. 101, No. 3, 1 August 1994 Downloaded 27 Jul 2001 to 141.14.233.190. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp Kivelson et a/.: Frustration-limited 2394 ration. The rate at which these clusters grow is observed to slow down as T is lowered. This can be understood by noting that according to the model the activation free energy for cluster growth should be small and the dynamics of cluster growth should then be controlled largely by the preexponential factor, which should be inversely proportional to the viscosity. The experimental data seem to confirm this.7 The simple theory presented here is based on a continuum model (not likely to be applicable to small clusters where molecular effects dominate); furthermore, it applies to dilute clusters in liquid, not to liquids with high concentrations of clusters such as one finds at low T and high 7,~ V. SPECULATION ON SUPERCLUSTERS (Phenomena above T2) Having described our model and the limits of its applicability, we now extend it to account for a number of observations not properly incorporated by the model alone. We speculate that the large clusters (several hundred nanometers) detected by light scattering may actually be agglomerates (superclusters) of smaller primary clusters described by the theory: we adduce this by noting that, as reported by Fischer et al.,6 the scattering correlation length of the clusters detected at high T (i.e., above T2 but below T,) decays at about the same rate as does the metastable concentration of clusters measured by the light scattering intensity. This can be understood if equilibration among clusters is rapid compared to the decay of the primary clusters. We propose a mechanism, based on the concept of rapidly equilibrating superclusters, in order to rationalize the slow and nearly temperature-independent nonexponential decay of the observed cluster concentrations at temperatures above T,. We assume that interactions between primary clusters (i.e., the presence of superclusters) reduce the surface energy of the primary clusters [i.e., reduce CTin Eq. (l)], thereby reducing the activation energy for cluster decomposition into liquid. This then suggests alternate paths to decomposition into liquid via bicluster, tricluster, tetracluster mechanisms, paths with successively lower activation energies. If one assumes rapid equilibration among dilute superclusters, these mechanisms then give rise, respectively, to 1.-1 ,t - 1127 t -1’3 asymptotic decays for the primary cluster concentration (and for the cluster-supercluster mean size) with successively reduced temperature dependences. And, indeed, our experiments on TPP seem to indicate power-law (approximately t - 1’3) decay over 2.5 orders of magnitude with only slight temperature dependence. In the proposed mechanism the slowness of the dynamics is associated with the low primary cluster concentrations. It is difficult to establish the uniqueness, or even the applicability of any reaction mechanism. That difficulty is operative here. But the proposed mechanism rationalizes the power-law decay in time of cluster concentrations, the weak temperature dependence along with the slowness of the rate, and the fact that the observed mean size of clusters decays at about the same rate as the number of clusters. To elevate this picture from rationalization to theory, it would be necessary to understand the dominance of a tetracluster mechanism. clusters in liquids VI. SPECULATION ON VISCOSITY We speculate that there exists a connection between the physics of geometric frustration that leads to cluster formation at high temperature and the dramatic low-temperature increase in viscosity that is characteristic of the “glass transition.” This speculation leads us to an intriguing fitting function for viscosity. That simple cluster formation at low temperature is not the direct cause of the high viscosity is suggested by the observations of Fischer et aL6 that the viscosity (measured by rotational relaxation) does not change markedly as the enhanced scattering develops slowly with time at low T, and that at high T the viscosity appears to be similar in liquids both with and without clusters.6 Although this high viscosity could be the consequence of a rapidly created high concentration of small primary clusters (too small to be observed), which then agglomerate slowly into observed superclusters that have little effect upon viscosity, this model would not explain why clusters are found at high temperature (above T,) only if the liquid is first supercooled and kept at low T until low-temperature clustering is observed. It is not yet clear whether the large, detected superclusters can be obtained in the absence of impurities; Fischer and co-workers6(b) suggested that supercooling and cluster formation were promoted by doping of the liquid, and we have found evidence that impurities may be needed to inhibit crystallization.7 Despite the fact that we have not established any direct correlation between clusters and viscosity, it is possible, nonetheless, that the scale length associated with frustration (e.g.,’that implicit in the projection of the ideal crystal into flat space) could govern the viscosity, though not necessarily through the presence of clusters.‘6 Thus we consider the possibility that the physics of supercooled liquids can be understood in terms of a narrowly avoided high temperature transition at Tif, rather than the approach to a low temperature critical point To. In this context, it is, therefore, not unreasonable to assume that the activation free energy, EJ T) = R T ln[ v( 7’)l v(m)] , for shear motion might also be a function of T,- T or T, - T. In fact, for the eleven simple liquids (both fragile and strong) and three polymeric liquids for which we have collected data from the literature, excellent fits to the viscosity can be obtained over the entire experimentally accessible temperature range (down to the glass temperature Tg) with the simple formula E,(T)-E,@J)=BT*~~ =o (820) (6<0), (3) where E,&m), B, and T* depend upon the material, and S= (T* - T)IT* (see Figs. 3 and 4 and Table I). T* might be taken as T, or Ti, or some related temperature; it need not necessarily be larger than the actual melting temperature, but except for cresol and salol, it is. (Note that the fractional change in T is usually small in the relevant temperature range.) The fits to Eq. (3) require adjustment of the temperutureindependent parameters rl(~), E,(m), T*, and B for each liquid. If one argues that the first two are obtained “indepen- J. Chem. Phys., Vol. 101, No. 3, 1 August 1994 Downloaded 27 Jul 2001 to 141.14.233.190. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp Kivelson et a/.: Frustration-limited 2395 pirically; 3 was taken as the integer exponent that gave the best results, although for some substances best fits yield exponents that are about 10% smaller. To elevate this fit from numerology to theory, one would need to understand the origin of the exponent; indeed, we have recently made progress on a scaling theory of the glass transition which, among other things, produces an exponent of about 813. Equation (3) suggests that it is neither the empirical Tg nor the Vogel-Fulcher critical temperature, To, that is fundamental to the description of supercooled liquids, but the high-temperature T*. The experimental realization of Eq. (3) presented in Fig. 4 suggests that viscosity data from all supercooled liquids, fragile and strong, can be scaled and superimposed on a single universal curve. Furthermore, in accord with the conclusions of Nagel and co-workers,20’2’Fig. 100 Fitted to E(m) + ET* ( 1 -T/T*) ..-,..-,Vogel-Fulcher TD/(T - TO) clusters in liquids 3 6 4 4 suggests that there is no turnover to Arrhenius-like behavior at temperatures just above Tg. However, to check this 250 300 350 400 Temperature 450 picture one would like to see some associated avoided critical behavior around T* in thermodynamic quantities. 500 [K] VII. DISCUSSION OF CLUSTER FORMATION FIG. 3. E(T) vs T for orthoterphenyl(Refs. 17 and 18) with T,,,=328 K, r,=243 K. and 6=(T*-T)IT*. E,(m)=3150 K and 7+)=1.O3X1O-3 cl? (a) (0 0 0) experimental E,,(T)=T ln[ all]; (b) (---) VogelFulcher .E,(T)=DT,T/(T-T,), with D=6.8, To=206 K, and TC,==308 K: (Ref. 19) (c) (-) cluster model fitted to Eq. (3) with B=488 and T*=361 K. dently” from the high-temperature asymptotic behavior of the viscosity, then the low-temperature behavior fitted by Eq. (3) requires only two adjustable parameters.” It is true, however, that the “universal” exponent 3 has also been set em- 0.12 0 crasol X dibutyl phthalate 0 glycerol I isopropyl benzene H n-butyl benzene X npropanol 0 naphthyl benzene W o-terphenyl A propylene carbonate i salols 0.08’ 2% 1 w m - 0.10, *& Throughout this article we have attributed enhanced angle-dependent light scattering to the presence of clusters. Such scattering could arise from the presence of microcrystals, but for a number of reasons we do not believe this to be the case: (1) At temperatures well below melting, crystallized and allegedly heavily clustered systems look very different, (2) the q dependence of the scattering is not characteristic of particles with sharp boundaries, and (3) most convincingly, the large enhanced scattering at temperatures above the melting point is achieved only from samples in which clusters 0.06 I b w 0.04, 0.02 -0.8 : 0.00 FIG. 4. [E,(T) - E,$m)]lBT* vs s’, with &O, where, presumably, -0.4 -0.2 0.0 0.2 0.4 1 -T/T* 0.02 0.04 0.06 (1 -T/T*)3 vs 6, and extended to 6~0, -0.6 0.10 0.12 “‘* for a number of glass-forming liquids (SeeTable I). In the inset the [E,(T) - EJm)]IBT* En(T) =E,(m). Note that TB falls at different values of 6 for different liquids. data are plotted J. Chem. Phys., Vol. 101, No. 3, 1 August 1994 Downloaded 27 Jul 2001 to 141.14.233.190. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp Kivelson et al.: Frustration-limited 2396 clusters in liquids TABLE I. Material parameters. B n-butyl benzene’ab) isopropyl benzene’ab) propylene carbonatelc) salol’d) dibutyl phthalate’a.b) o-terphenyl’.’ s-trinaphthyl benzene(‘) n-propanol’b’ n-phenyl-cresol”’ glycerol(s) boron oxide’h’ poly(p-chlorostyrene) yco:;’ poly (propylene-gl (i) poly(viny1 acetate)“) 331 210 338 478 169 488 334 61.4 543 98.9 68.1 566 741 342 T* W E,,(=J) (IO3 K) 204 216 241 317 298 361 531 200 301 335 1020 559 271 467 1.43 1.72 2.12 3.22 3.31 3.15 5.07 2.22 6.09 5.18 10.5 6.08 4.52 2.76 “Reference22. bReference23. ‘Reference24. dSeeRef. 20 (dielectric and heat capacity data). ‘Reference 17. have previously been allowed to grow at low temperature and not from samples that have previously been merely crystalized at low temperature.7 The enhanced scattering could also arise from large critical-like fluctuations that are not characterized by an appreciable activation barrier, but we do not believe this to be the case because one would then expect, contrary to observation, that this enhanced scattering would be observed above the melting point whether or not the system had been previously taken to low temperature and given time to form clusters. Fragile liquids are often associated with small values of the Vogel-Fulcher parameter, D, which is obtained from low-T fits of the viscosity.” Curiously, D correlates quite well with our high-T asymptotic viscosity activation energy E,(a). In any case, our model suggests that the degree of clustering should not be associated with D but with the ratio BT*lE,(~) (see Table I). To date, direct experimental observation of clustering has been reported for only two liquids, OTP and TPP;6,7 OTP has the largest BT*lE,(~) of all the liquids in Table I, and the clustering parameter of TPP has yet to be determined. BT*/(E,P)) T,,, (K) 47.1 26.5 38.5 47.2 5.2 56.0 35.9 5.53 26.9 6.40 6.63 52.0 44.4 57.8 185 174 218 318 238 331 472 147 327 293 723 D 2.6 4.5 5.3 5.4 6.4 6.8 8.7 15.7 15.7 17.5 6.3 4.7 6.9 To WI 112 112 136 189 127 206 278 48.3 160 130 35 338 176 260 ‘Reference18. gReference25. hReference26. ‘Reference27. Noting that clusters do actually seem to form, we have also postulated that the physics of cluster formation is related to the great increase in viscosity with decreasing temperature that is observed in supercooled liquids. Here we do not pursue this correlation in depth, but because the high viscosity is observed even before superclusters are formed, and also in liquids where superclusters have not been detected at all, we postulate the existence of smaller, primary clusters which do directly affect the viscosity. The large superclusters may or may not have a large effect on the viscosity (the final answer is not yet in), but in any case, according to the model, they are not central to the phenomenon of high viscosity in supercooled liquids. This approach is supported by the fact that it leads us to Eq. (3) which gives excellent fits for the viscosity over the entire accessible temperature range for both fragile and strong simple liquids and for polymeric liquids, as well. The fact that the viscosity (and so presumably the primary clusters) do not exhibit hysteresis, whereas the superclusters do, has been addressed, but it has not yet been satisfactorily resolved. ACKNOWLEDGMENTS VIII. COMMENTS We have, by means of Eq. (l), introduced a model of frustration-limited cluster formation in order to explain the existence and behavior of the clusters inferred from light scattering. Not only does the model give a rationale for the growth of clusters at low temperatures, but it explains why such clusters can persist for long periods when heated to temperatures above the melting point. Although previous cluster models5”0”3 have incorporated the concept of frustration by studying the stability of specific arrangements of spherical particles, we believe that our continuum or collective approach is the first to generalize these ideas in such a way so as to explain all the phenomena reported by Fischer et al. Our treatment of frustration has been based closely on that of Sethna.3*4 This work was supported in part by the National Science Foundation under Grants Nos. CHE 91-17192, DMR 9312606, and CHE 89-02354 at UCLA T.M. Fischer would like to thank the Humboldt Foundation for providing a fellowship. We would also like to thank Dr. Anders Karlhede and Professors Sidney Nagel and James Sethna for their valuable comments. ‘M. Kleman and J. F. Sadoc,J. Phys. (Paris), Lett. 40, L569 (1979); J. F. Sadoc and R. Mosseri, Philos. Mag. B 45, 467 (1982); M. Kleman, J. Phys. (Paris) 43, 1389 (1982). ‘D. R. Nelson, Phys. Rev. Lett. 50, 982 (1983); Phys. Rev. B 28, 5515 (1983). ‘J. P. Sethna, Phys. Rev. Lett. 51, 2198 (1983); Phys. Rev. B 31, 6278 (1985). 4For a review see,J. M. Carlson,S. A. Langer,and J. P. Sethna,Europhys. Lett. 5, 327 (1988). J. Chem. Phys., Vol. 101, No. 3, 1 August 1994 Downloaded 27 Jul 2001 to 141.14.233.190. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp Kivelson et a/.: Frustration-limited sF. H. Stillinger, J. Chem. Phys. 89, 6461 (1988). 6(a) E. W. Fischer, G. Meier, T. Rabenau,A. Patkowski, W. Steffen, and W. Thonnes, J. Non-Cry% Solids 131-133, 134 (1991); (b) B. Gerhartz, G. Meier, and E. W. Fischer, J. Chem. Phys. 92, 7110 (1990). ‘Data on TPP were obtained from X. Zhao, I. Cohen,A. Ha, and D. Kivelson (unpublished). ‘There. cannot be a true phasetransition nor an ideal freezing temperatureif the spacein which the locally preferred structure is unfrustratedis finite (as when it has negativecurvature). However, in this case we can take T, to be a crossover temperaturebetween liquidlike and ideal-solidlike behavior. ‘C. A. Angel], J. Non-Cryst. Solids 131-133, 13 (1991). “J . D . Bernal, in Liquids: Structure, Properties, Solid Interactions, edited by T. J. Hughel (Elsevier, Amsterdam, 1965). ” A. H. Boerdijk, Phillips Res. Rep. 7, 303 (1952). 12F.C. Frank and J. S. Kasper, Acta Crystallogr. 11, 184 (1958). 13M* R . Hoare, in Ann. NY Acad. Sci. 279, 186 (1976). This includes a bibliography and review of previous work. “See, for example, K. F. Kelton, Solid State Phys. 45, 75 (1991). “See review of glassesby J. JackJe,Rep. Prog. Phys. 49, 171 (1986). See also, E. McLaughlin and A. R. Ubbelohde,Trans. Faraday Sot. 54, 1804 (1958); G. Adam and J. H. Gibbs, J. Chem. Phys. 28, 373 (1965); M. F. Shearsand G. Wtlliams, J. Chem. Sot. Faraday2 69, 1050 (1973); M. H. Cohen and G. S. Grest, Phys. Rev. B 20, 1077 (1979); P. D. Hyde, T. E. Evert, and M. D. Ediger, J. Chem. Phys. 93, 2274 (1990); F. Fujara, B. Geil, H. Sillescu, and G. Fleischer,Z. Phys. B CondensedMatter 88, 195 (1992). “When at low T the concentrationof clusters gets large, the problem may not be one of stable clusters in a liquid, but rather of growing clusters interfering with eachother. Our model doesnot incorporatethesephenomena. clusters in liquids 2397 “W. T. Laughlin and D. R. Uhlmann, J. Phys. Chem. 76, 2317 (1972); M. Cukierman, J. W. Lane, and D. R. Uhlmann, J. Chem. Phys. 59, 3639 (1973). ‘*R. J. Greet and D. Turnbull, J. Chem. Phys. 46, 1243 (1967). “The much used Vogel-Fulcher expression, E?(T) = DTcT/( T- To) for ‘%Tcf, which introduces a critical divergence at 7’cm0.87’s, requires four adjustableparameters[D, Tc, 77(m),and a high temperaturefitting cut-off, Tcr] (seeFig. 3). If the temperaturerangeof the fitted data is very restricted,the results are not sensitiveto the choice of T,, . However,if, in addition, one wishes to fit the data at temperaturesabove Tcr, then one needs,at least, a fifth parameter,E?(m), where E,(T) = E(m) for T> Tcr. This analysis requiresone parametermore than does Eq. (3) and it yields poorer fits (seeFig. 3). Alternatively, the Vogel-Fulcher expressioncan be fitted to the high-temperaturedata with a low-T cut off; these fits work poorly at low temperatures. ‘OP. K. Dixon, Phys. Rev. B 42, 8179 (1990); P. K. Dixon, L. Wu, S. R. Nagel, B. D. Williams, and J. P. Carini, Phys. Rev. Lett. 65, 1108 (1990). *‘P K. Dixon, N. Menon, and S. Nagel, Phys. Rev. E (to be published). “A. J. Barlow, J. Lamb, and A. J. Matheson,Proc. R. Sot. (London) Ser.A 292, 322 (1960). 23A. Campbell Ling and J. E. Willard, J. Phys. Chem. 72, 1918 (1968). %L. Borjessonand W. S. Howells, J. Non-Cryst. Solids 131-133,53 (1991). =N. Menon, K. P. O’Brien, P. K. Dixon, L. Wu, S. R. Nagel, B. D. Williams, and J. P Csrini, J. Non-Cryst. Solids 141, 61 (1992) (dielectric data). “D. Sidebottom, R. Bergman, L. Bojesson, and L. M. Torell, Phys. Rev. Len. 71, 2260 (1993) (relaxation times). “A. Schonhals,F. Kremer, and E. Schlosser,Phys. Rev. Len. 67,999 (1991) (Note that polymer data does not extend into high-temperatureArrhenius region.) J. Chem. Phys., Vol. 101, No. 3, 1 August 1994 Downloaded 27 Jul 2001 to 141.14.233.190. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp
© Copyright 2026 Paperzz