Polyamorphism J.L. Yarger and C. A. Angell Department of Chemistry and Biochemistry, Arizona State University, Tempe, AZ 85287. Contents I. Introduction, Definitions and Classification (a) Concepts, terminology and definitions (b) Polyamorphic Classifications (c) Special role of Computer Simulations (d) Organization of the Review 1 3 5 6 7 II. Brief historical account of the subject (a) Developments before 1990. (b) Developments since 1990 8 III. Elementary level interpretation of polyamorphism IV. Laboratory and Simulation Studies (a) Equilibrium Liquid Polyamorphism (level 1) (i) Laboratory studies (ii) Computer simulations (b) Metastable Liquid Polyamorphism (level 2) (i) laboratory studies (ii) computer simulations 14 (c) Amorphous Solid Polyamorphism (level 3) (d) Amorphous-Amorphous Transitions (level 4) (e) Liquid State Critical Transitions (level 5) IV. Prognosis and Future Studies V. Concluding Remarks VI. Acknowledgments VII. References 10 10 25 25 33 3 3 9 12 I. Introduction, Definitions and Classification Polyamorphism, Vitreous Polymorphs and Amorphous Polymorphs are terms that have been used to describe the existence of two or more distinct amorphous phases of the same substance, phases that differ only in density. They may be elemental, stoichiometric compound or mixed compound in character, but the compositions of the two phases must 1 be identical to fall within the scope of this review. To date, these terms have been used to describe a range of structural changes in amorphous solids, glasses, supercooled liquids, and equilibrium liquids. In this article we will overview the existing experimental, computational, and theoretical studies of these phases, their transformations, and related phenomena within isochemical systems. As an initial part of this work, we will attempt to clear up some of the ambiguity in definitions and terms used to describe these phenomena in the recent and current literature. With clear definitions in place, we will then classify the existing literature results into several categories that pertain to the type of phase transition behavior exhibited, and to the equilibrium state of the system. It should be mentioned at the outset that the vast subject of liquid crystals is excluded from our consideration by the fact that these are not isotropic liquids. Over the past several years, several concentrated sources of information on our subject have appeared. There is a proceedings volume1 from the first conference on this subject area (“Unusual phase transitions...) and there have been both focused reviews and overview articles on the basic phenomenon of polyamorphism2-5 liquid-liquid6 transitions (LLT), and amorphous-amorphous transitions5,7 (AAT). However, to date, there has been no comprehensive review of the literature nor any systematic classification of the various types of systems that are said to show polyamorphic behavior. This article is intended to fill that need and to establish a sound framework for the future development of this growing field. I(a). Concepts, Terminology, Definitions Although it is only recently that the idea of polyamorphic forms has been given much attention (some 400 papers since 1990), the concept of distinct forms of the glassy state for specific substances is actually quite old. The term amorphous polymorphs was used by Kreidl and co-workers8 in 1966 to distinguish between the structure of normal SiO2 glass and the structure which SiO2 glass adopts under long exposure to radiation. Shortly afterwards (1970), a much stricter use of the equivalent term “vitreous polymorphs” was made by Angell and Sare.9 Although it was only expressed in a footnote to the discussion of the apparently different structures that could be adopted by water in the pure state and in the solution state (now considered to be closely related to the high 2 density amorphous HDA structure10), we reproduce their concept here because it corresponds closely with the definition we would like to recommend in this article. They wrote, “The question raised here is whether of not it is proper to think in terms of vitreous polymorphs of a substance, i.e. whether long-range disordered substances can generate sufficient differences in their short-range order, when prepared under different conditions, to have distinct and different thermodynamics properties, which are maintained above their glass transition temperatures (when the amorphous phase is in equilibrium) for a finite temperature interval”. This can serve as a definition for true (unambiguous) polyamorphism if we add the caveat that such distinct phases can co-exist in metastable, or stable equilibrium. Indeed, it turns out that some authority for this idea had already been given by Landau in his 1969 treatise11 where direct reference was made to the possibility of two liquid phases existing for the same compound. The term “polyamorphism” itself was first used by Palatnik12 in 1980 who wrote “by analogy with the phenomenon of polymorphism, well known for crystalline materials, we will refer to the phenomenon of the existence of several such varieties of the amorphous state of the same substance, as polyamorphism.” This term was later reintroduced by G.H. Wolf in 199213, and has now been used repeatedly to deal with the many new amorphous “transitions” discovered over the past 2 decades. Here we will define (true) polyamorphism as “that phenomenon in which two distinct phases of a substance, both isotropic, internally equilibrated, and of the same composition, equilibrium.” can co-exist in stable or metastable thermodynamic “Polyamorphs”, then, are the phases that participate in the polyamorphic equilibrium. If we were to restrict this article to cases in which the above definition applied faithfully, then the article could be short indeed. Most of the cases for which the term has been used in the past, are excluded by it. However, we believe it is important to have a thermodynamically unambiguous case as a reference for informed discussion of the remainder. The key ingredient in this definition is the possibility of co-existence in internal equilibrium, as is true for the familiar crystalline state phenomenon. By way of contrast, it is important to point out phenomena that are truly excluded from the discussion. For instance, when sufficient regions of P-T space are explored, it is 3 clear that all amorphous materials will exhibit great structural changes. As an extreme case, many gases undergo transitions between different structures, the most commonly cited (textbook) case being that of nitrogen dioxide and its dimer, dinitrogen tetroxide. However, this change in structure, which occurs continuously over a broad range of P-T conditions, is chemical equilibrium, not polyamorphism. On the other hand, if NO2 were to form a liquid phase that could coexist, at particular P, T combinations, with a different liquid whose structural units were N2O4 molecules, then this would indeed satisfy the definition. What has been confusing is the circumstance that most cases so far discussed are somewhere intermediate between these extremes. Liquids and glassformers can greatly change their structures in response to changes in pressure and temperature, but these changes are usually also continuous. It is so far only in a very few cases, mostly glasses, that the structural change can become abrupt like the first-order transition between crystalline polyamorphs. The more-or-less abrupt change observed in glass cases may be referred to as the amorphous-amorphous transition, AAT, a term popularized by Brazhkin7,14 to describe the continuous transition from different local or intermediate range structural topologies in amorphous solid materials. Solid polyamorphs13 (or "vitreous polymorphs"9), if defined so as to be consistent with the strict criteria applied to the liquid state phenomenon, present us with a problem. As with crystalline polymorphs, alternative polyamorphic solid forms should, strictly, be capable of co-existence in thermodynamic equilibrium. However, in the case of amorphous solid systems this is impossible because the glasses are themselves not in equilibrium, by definition. Nevertheless, transformations occur between alternative structures in the vitreous state, and they are receiving a lot of attention. Therefore we need to introduce a classification for the behavior that has been observed. We base it on how closely the behavior observed in a particular case approaches that of a thermodynamically ideal polyamorphic transition. Whether or not many cases of stable state isocompositional liquid-liquid equilibrium will ever be demonstrated, is for the future to determine. 4 I(b). Classification In the following listing, we provide a basis for separating the diverse phenomena, to date been ‘lumped together’ under the rubric of polyamorphism, into a hierarchy based on closeness of approach to the ideal polymorphic behavior. In descending order, we recognize the following levels in the subject. (The examples mentioned will be fully discussed in the appropriate parts of Section III). 1. Stable liquids in coexistence (examples: elemental phosphorus [laboratory] and model liquid systems [computer simulation]). 2. Metastable liquids in co-existence or distinguished by a clearly defined first-order phase transition (Example: Si, in computer simulations]. 3. Metastable liquids that undergo a transition from liquid to glass (only one phase in equilibrium. (Example: triphenyl phosphate [laboratory, no simulations available]. 4. Amorphous solid phases not formed from the liquid but separated by well-defined separating surfaces (Example: Amorphous H2O). 5. Amorphous solids that change structure in a more or less abrupt manner, and show hysteresis without, so far, showing any separating surface (Example: SiO2, GeO2 and BeF2). 6. Liquid state anomalies of the “smeared first-order type” that are well characterized and reversible, but are not polyamorphic (These highly cooperative processes are included because of the possibility that they could evolve into first order transitions by application of the right thermodynamic stress on the system. They have the special advantage of manifesting their cooperative transitions in the thermodynamically stable liquid state). I(c). The special role of computer simulation experiments in exploring polyamorphism. Unfortunately, the number of examples of laboratory materials that can be classified as exhibiting level 1 (equilibrium) polyamorphism is, at this time, exceedingly small, 5 namely, one, the case of phosphorus. Nature does not encourage this phenomenon, for reasons that can be understood and which will be discussed in more detail below. Level 2 (metastable liquid) polyamorphism is more common, but is fraught with problems stemming from the propensity of the low temperature polyamorphs to crystallize, seemingly a rather general phenomenon. For each of these “top levels” of polyamorphism, computer simulation studies have special advantages. At level 1, they are free from restriction to reality, which means that models of non-existent substances that exhibit equilibrium liquid-liquid coexistence can be created and characterized. In this way, computer simulations permit the physical reality of the phenomenon to be demonstrated, as we will review in section II1(a)(ii) below. At level 2 in the above classification, simulations of real substances can be conducted on time scales of nanoseconds, and the first-order character of the transition between liquid phases can be established unambiguously before any crystallization of the low temperature phase can occur. In larger systems, it should be possible to observe a true co-existence of the two liquid phases, though this has yet to be done. At the lower levels of the above hierarchy, the advantages swing increasingly in favor of the experimental studies because of the larger samples and longer timescales of the experiments involved. I(d). Organization of the review. After (i) a brief account of the way the subject has developed, empirically, since early times, and (ii) an elementary level account of how the phenomenon can be related back to deviations from ideal gas behavior in terms of specific attractive repulsive interactions between the systems’ particles, we will (iii) review the accumulated knowledge in the field according to the order of phenomenological rigor listed in the previous section, giving a separate section to each level. In each of these sections, we will review first the laboratory experiments that define the level in question, and will then describe the corresponding computer simulation studies that help to establish the appropriateness of the level in question. 6 II. Brief historical account. (a) Developments before 1990. To our knowledge, the possible existence of equilibrium between liquid phases of the same composition was first recognized in 1967 through the behavior of a simple theoretical model of the “two-species” liquid variety, proposed by Rapoport15. Two species models themselves go back to Roentgen’s early attempt to explain the properties of water16, but in Roentgen’s model and all others of its type17, the different “species” were assumed to mix together in an ideal (random) fashion. Rapoport15,18 introduced a non-ideal form of mixing into the model in his attempt to account for the existence of melting point maxima in certain substances. Rapoport’s contribution was to assume that the species mixed according to the regular solution model of binary solutions (which is well known to have a critical point below which two phases of different composition coexist). The model gave a quite reasonable account of the melting point behavior but its further implication that, at some extreme, an isocompositional liquid-liquid phase transition would occur, was regarded as an embarrassment, and discounted. While the “species” of Rapoport's model, like those of most other two-species models, have not been identified, the cooperative character of the model, and the phase transition predicted by its simple equations, have recently gained considerable attention.6 Different versions of it have appeared, usually without awareness of Rapoport’s contribution. In 1979, for instance, Aptekar19 published a paper predicting liquid-liquid equilibria in silicon and germanium, based on a model with the same mathematical form, in which the species were silicon (or germanium) atoms which somehow could have either metallic or semiconducting character and could nevertheless mix together, even if non-ideally, in different proportions above and below the phase transition. An equally unphysical "two liquids" version of this model, in which two distinct liquids are assumed to occupy the same physical space in different volume fractions depending on T and P, was later (in the 90’s) to be used by Ponyatovsky and co-workers20,21 to predict liquidliquid transitions in a variety of systems, again without recognition of any of the earlier versions. More physical versions of this type of model will be discussed later. Separately, 7 Mitus et al, in 198122,23, discussed liquid-liquid transitions on the basis of the properties of a Potts model. Probably the earliest demonstration in the laboratory of the solid state generation of an amorphous material quite different in properties from the known liquid phase, was made for the case of germanium. In 1970, DeNeufville and Turnbull24 observed spherical droplets of a non-crystalline semiconducting germanium phase to separate out from a reheated sample of a reduced form of germania glass in which the O:Si ratio was only 1.9:1. This was not a polyamorphic phase change because the new phase, pure Ge, was quite different in composition from the phase from which it formed. The point of importance to the present review is that the phase of germanium that formed was a tetrahedral semiconducting phase rather the higher coordination number, weak metal phase that is the liquid state of germanium.25 The temperature at which this semiconducting Ge amorph formed was just above the glass transition temperature of the reduced glass. The droplets were observed by scanning electron microscopy. A consequence of the glassy Ge precipitation was the disappearance of the glass temperature endotherm on second scanning.24 This is not too surprising since each of the newly formed amorphous phases (Ge and stoichiometric GeO2) is well below its glass temperature at the Tg of the original glass. GeO2 has a Tg of ???ºC, and amorphous Ge only recrystallizes at 450ºC. The amorphous Ge indeed crystallized….??… DeNeufville and Turnbull were aware of the confusing thermodynamics of germanium liquid and amorphous solid, and it seems they expected that a liquid-toamorphous transition would occur in the pure germanium phase if supercooling of the high temperature phase could occur without crystallization or, conversely, if the low temperature phase could be heated without crystallization. This transition was later formally predicted by Spaepen and Turnbull26, simultaneously with Bagley and Chen27 using a Gibbs energy construction. The temperature of the transition was predicted to be 969K, somewhat below the value subsequently assigned by experiment, but above the value obtained by simulation studies to be discussed. In fact, it was close to the value, 880K, predicted by Aptekar19. The phase change was thought by Turnbull et al.28 to be "melting" of the amorphous semiconducting solid, though it has recently been pointed 8 out29 that even the crystalline form of silicon has a viscous liquid-like value of the diffusivity at the putative phase transition temperature. In other words, the silicon transition would be seen to be a liquid-liquid transition (as Aptekar had suggested) if it could be observed on normal time scales.. More recently the efforts to observe the phase transition have been made from the liquid state by various supercooling techniques, particularly that of levitation melting.30,31 It has also become a subject for intense computer simulation efforts, which will be described further below. The field received a major impetus in 1984 when Mishima et al.32 demonstrated that ice collapsed to a dense amorphous phase under cold (196 K) compression. The amorphization pressure was at 1.6 GPa. Of special interest to our subject was the additional observation that, on decompression at an appropriate temperature, the dense phase suddenly expanded to a less dense but still amorphous phase – apparently a direct glass-glass transition. This started a wave of studies on amorphization, and much speculation on glass-glass transitions, the most notable effort being that of Mishima33 in which the water was cycled back and forth many times between the two states. (b) Key developments since 1990 A particular reason for the current level of excitement about the phenomenon derives from the paper by Poole et al, in 1992.34 Using molecular dynamics simulations of water in the Rahman-Stillinger ST2 potential, these authors demonstrated that, when the long range part of the interaction was computed using the "reaction field" method, that the glass-glass transition observed by Mishima et al probably had a liquid state equivalent, though it could not be observed directly by the simulation. What was observed was evidence of a van der Waals type P vs V relation with a loop in the supercooled liquid regime. Such a loop should become a liquid-liquid equilibrium domain in a larger system. The loop developed as the temperature fell below a critical temperature at moderately high pressures. The notion that there could be a second critical point in an isotropic liquid phase was novel because Aptekar's work on silicon was unknown. Furthermore, water is intrinsically a more interesting case. “Liquid-liquid” transitions were then directly observed in two laboratory systems during cooling of the high temperature liquid phase. The first case35 involved a mixed oxide system related to garnet, Y3Al5O12, and the second case concerned a pure 9 molecular liquid, tri-phenyl phosphite36-38 In each case there has been much controversy due to the presence of crystalline material in the low temperature phase and, later, of the realization that the low temperature phase was actually a glass (i.e. was below its Tg) as it was formed. The ambiguity created in the field by the latter observations has fortunately been dispelled by the observation of liquid-liquid equilibria in the stable fluid state (the case of elemental phosphorus39), and by the development of simple theoretical models in which the coexistence of two stable liquid phases can be demonstrated, and characterized in detail40,41. Details in each case will be given in the appropriate sections below. The rate at which interest in this field has increased can be gauged from Fig. 1, which is a histogram of papers addressing the phenomenon over the past two decades. Figure 1. Time evolution of interest in the polyamorphism field. III. Phase diagrams, and an elementary level interpretation of polyamorphism . Before we proceed to discuss the details of these papers according to the scheme given in section Ic, it is appropriate to give brief attention to the phase relations that can exist. In Fig. 1(a) and (b) we reproduce the first such phase diagram, given in the work of Aptekar19 and the generalized version used by Poole et al.42 In the case of silicon, the Clapeyron slope dT/dP for the liquid-liquid equilibrium is negative because of the open network character of the low temperature phase, while in the second case the slope is shown positive. Instances of each type have now been identified in laboratory studies, as will be discussed in section III below. ` 10 Figure 1. (a) Phase diagram for the element silicon showing …. (from ref. ) 11 (b) generalized phase diagram for the non-crystalline phases of a single component system, showing both a gas-liquid critical point and the second (liquid-liquid) critical point. (reproduced from ref. X, with permission of the AAAS.) Both cases shown in Fig.1 exhibit a critical point that terminates the liquid-liquid coexistence line. In the case of silicon it is at negative pressure and impossible to verify. In the more interesting case of water, which is expected to have the same general form, such a critical point has been suggested to exist at positive pressure on the basis of both simulation43 and laboratory44 studies. The positive Clapeyron slope, envisaged in Fig. 2, is the type of behavior shown by the model potential of Ref. X, see section III.a.(ii), and possibly manifested in practice by TPP (see section III.b. (i). The possible existence of such novel critical points has provided one of the major stimuli to the field, and is responsible for much of it’s present level of activity. To understand briefly how such behavior can arise in the condensed states of a system of simple particles, we may generalize the ideas behind the Poole model for polyamorphism in water. This is a modification of the van der Waals model for the gasto-liquid transition. Instead of invoking just the short range repulsion (on contact) and the long range attraction of the Van der Waals model, the Poole model introduces a second length scale for attractions which lowers the system energy maximally at a density lying between the close packed density and the gas density. In the case of water, this intermediate density arises because of hydrogen bonds between water molecules which cause water molecules to find their lowest energies when in four coordination, i.e. at a much lower density than the 10-12 coordination associated with close packing. Of course there are many other types of bonding that can lead to energy minima at densities lower than close packing, and any of these could provide a mechanism to drive a polyamorphic equilibrium. All of these open structure-promoting interactions can be overcome by the application of pressure so that eventually at increased pressures the system will be most stable in close packing. At intermediate pressures, the two types of packing may be able to coexist. Whether such coexistence will actually be manifested, i.e. whether a polyamorphic equilibrium will be seen in practice, depends on cooperative effects that prevent the simple mixing of low density and high 12 density structural elements. The interactions which cause the initial deviation from ideal gas behavior and lead to the two phase liquid solid coexistence are illustrated in Fig. 2 (a) and the manner in which the simple two phase coexistence range of the normal liquid is split into two separate coexistence domains by the presence of the low coordination number driving potential, is illustrated in Fig 2(b). The mathematical formulation of these concepts and the resulting phase diagrams for different parameterizations of the model, are given in ref. X (Poole et al). The case of water has the low density phase a homogeneous hydrogen-bonded network, which is somewhat unusual. A much more common way of arranging for an intermediate density phase would for it to be a molecular liquid, while the high density phase would be a, continuously bonded polymeric phase, obtained by rearranging the electrons around the constituent atoms. In the polymeric phase the free volume of the molecular liquid has been eliminated automatically increasing the density. This is essentially the condition met by the only case of polyamorphism to satisfy our strictest criterion, namely the elemental phosphorus case, in which both gas and low density liquid consist of P4 molecules, while the high density phase, generated by sufficient increase of pressure, is polymeric (of so-far undetermined structure). It might be expected that there should be many cases in which the rearrangement of electrons between atoms should permit this type of polyamorphism, and it is a little surprising that more cases have not yet been found. Some indications of where the problem lies may be seen in the failure, some years ago, of a conscious attempt to produce a high density phase by rebonding of the atoms in the molecular substance NSF. It is perhaps reassuring to find that NSF is extremely stable under pressure, withstanding in our case some 25 GPA without change in structure. Chemically, (as opposed to comically) it means that the PV work of compression to, say, 50% of the molar volume of NSF, a volume change of 20 ml, does not produce an energy change sufficient to compensate for the rearrangement of the bonding electrons in NSF to any other configuration. But there will surely be other cases in which it is sufficient, and then polyamorphism will follow unless crystallization intercedes 13 For further thermodynamic background material on conditions for, and the origin of, liquid-liquid phase equilibria, the reader is referred to refs. X,Y and Z. (a) (b) QuickTime™ and a TIFF (LZW) decompressor are needed to see this picture. Figure 2. Illustration of the relation between normal van der Waals interpretation of the coexistence of liquid and gas phases, and the modification necessary to understand the existence of a second critical point. (based on lectures of Peter Poole, private comm.) Jeff: I need to put into this Figure, in the T-V plots, a line for the glass transition temperature and a comment on the effect of pressure below the glass temperature in promoting a glass/glass conversion. III. Laboratory and Simulation Studies 14 In the body of this review, we will deal with the literature on polyamorphism in separate sections, according to the hierarchy in ambiguity described above in section Ib, and starting with the least ambiguous. In each section, we will first discuss the laboratory phenomena that have been reported and then discuss the parallel studies of the same types of system that have been carried out by computer simulation experiments. III(a). Equilibrium Liquid Polyamorphism (Level 1) In this level of the classification, there is but a single unambiguous laboratory example, (liquid crystals having been excluded from consideration because either one, or both, phases are anisotropic). However there are several rather detailed theoretical models to consider, along with their computer-simulated manifestations. Because of the desirability of tying this subject to real world phenomena, we will consider the laboratory case first, and then will describe the more complete accounts of isocompositional liquids in equilibrium that have recently been provided by model system computer simulation studies. (i) Laboratory Systems. Phosphorus (ii) Computer Simulations and Theoretical Modeling In this section we consider (a) simulations of the laboratory case of phosphorus, which can only be executed using ab initio computational methods and which in consequence are extremely expensive, in computation time, to simulate. As a result, information is only available on small samples studied for short times. (b) simulations of model systems with simple potentials which are easy to study in large samples and which therefore are now well elaborated. The systems so far investigated in detail are systems with soft repulsions which effectively provide the second length scale necessary for the liquid-liquid co-existence to be manifested. Because of the greater detail obtainable in the latter cases, we find it most fruitful to discuss these first. 15 Model systems Systems with step, multistep and ramp potentials Reflecting on the “structure-modified van der Waals model” introduced in section 1, we need to have some mechanism by which the volume change on passing from the gas to close-packed system can be interrupted at a structurally stable intermediate density. The simplest model approach to obtaining this result is to have repulsion component of the interaction potential contain a step, such that a second shorter distance repulsion replaces the initial repulsion above a given pressure. This “softened core” idea was used in theoretical studies by Stell some time ago, though not with polyamorphism in mind. Jagla reintroduced it and showed that it generated water-like features in the behavior. Then it was realized that a two step repulsion would yield the same result and so, by adding more steps, the idea of a repulsive ramp modle arose. This simple idea is all that is . .. . which can only be studied in very small samples, with limited output have been executed by several groups (XYZ) but demand ab initio treatments also those of model systems which are simple to simulate, and which therefore have yielded Phosphorus Several first-principles molecular dynamics simulations on liquid phosphorus at high temperature and pressure have been carried out in order to investigate the observed liquid-liquid phase transition52-55. These computer simulations all show that the transition is caused by the breakup of P4 tetrahedral molecules to form a polymeric liquid. The major electronic rearrangements in phosphorus involved in passing from the tetrahedral molecular fluid phase to the rebounded polymeric phase, requires this system be simulated by ab initio methods. Currently, these simulations are very computer-timeintensive and no more than 100 atoms can be simulated for physically useful times. This size of simulation is typically considered too small to reveal accurate phase transition behavior clearly. However, it has been used very successfully to show the different molecular and network motifs adopted be the various liquid phosphorus phases. 16