Density controls the kinetic stability of ultrastable glasses

We use a swap Monte Carlo algorithm to numerically prepare bulk glasses with kinetic stability comparable to that of glass films produced experimentally by physical vapor deposition. By melting these systems into the liquid state, we show that some of our glasses retain their amorphous structures longer than 10 times the equilibrium structural relaxation time. This “exceptional” kinetic stability cannot be achieved for bulk glasses produced by slow cooling. We perform simulations at both constant volume and constant pressure to demonstrate that the density mismatch between the ultrastable glass and the equilibrium liquid accounts for a major part of the observed kinetic stability. Copyright c © EPLA, 2017 Introduction. – Physical vapor deposition is an efficient way to prepare amorphous thin films with tunable physical properties. Molecules are slowly deposited onto a substrate held at constant temperature, and a glassy film is constructed layer by layer [1]. For well-chosen substrate temperatures, the resulting glass may exhibit “exceptional” [2] physical properties. It can have higher density [3,4], lower enthalpy [5,6] and lower heat capacity [7,8] than glasses conventionally prepared by slow cooling. These vapor-deposited glasses have been classified as “ultrastable”, and have now been prepared from a wide range of molecules [9–13]. Although produced in an unusual way, these glasses are thought to be equivalent to glasses that have been aged for unacheivably long times. The kinetic stability of vapor-deposited glasses can be estimated in two ways, both of which involve melting the glass. A glass can be heated slowly and the “onset temperature” at which it starts to melt back to the liquid state measured [2]. The higher the onset temperature, the more stable the glass. The second measure is through a “stability ratio” which allows direct comparison of glasses formed from different materials [14]. The material is rapidly heated above the glass transition, and the ratio between the time it takes the glass to melt and the equilibrium relaxation time at the melting temperature is measured. This is the stability ratio, S. For vapordeposited ultrastable glasses, the stability ratio is found in the range S = 10 (for materials with low stability [13]) to 10 (for the majority of ultrastable glasses), up to 10 for trisnaphthylbenzene [14] and 10 for o-terphenyl [15], which seems to set the experimental record. Ultrastable glasses represent a new class of amorphous materials with interesting applications [1], but their properties are not well understood yet. For instance, it is not known how to quantitatively relate the degree of equilibration of ultrastable glasses to their measured kinetic stability, despite recent progress in this direction [16–18]. Computer simulations provide a valuable tool for achieving this understanding, as complete knowledge of microscopic information provides direct insight into the properties of stable glasses. However, computational work in this area is challenging, as materials this stable have effective preparation times that are extremely large. Several efforts have been made to simulate stable glasses using very slow cooling [19,20], random pinning [21], nonequilibrium sampling [22], or by directly simulating the deposition process [23–26], but the largest reported stability ratio to date remains a modest S ∼ 10 [19]. In this article, we report stability ratios that can be as large as S ≈ 10 for a simulated bulk glass-former, comparing favourably with the largest values reported in experiments for ultrastable glassy films. We achieve this record value by preparing glasses using swap Monte Carlo [27,28]. By considering how these glasses melt in different numerical ensembles (isochoric or isobaric), we demonstrate that a major part of their large kinetic stability stems from the density mismatch between the ultrastable glass and the equilibrium fluid, because the dense glass needs to expand to accomodate the invading fluid during melting.