Reduced probability densities of long-lived metastable states as those of distributed thermal systems: possible experimental implications for supercooled fluids

Abstract

When liquids are cooled sufficiently rapidly below their melting temperature, they may bypass crystalization and, instead, enter a long-lived metastable supercooled state that has long been the focus of intense research. Although they exhibit strikingly different properties, both the (i) long-lived supercooled liquid state and (ii) truly equilibrated (i.e., conventional equilibrium fluid or crystalline) phases of the same material share an identical Hamiltonian. This suggests a mapping between dynamical and other observables in these two different arenas. We formalize these notions via a simple theorem and illustrate that given a Hamiltonian defining the dynamics: (1) the reduced probability densities of all possible stationary states are linear combinations of reduced probability densities associated with thermal equilibria at different temperatures, chemical potentials, etc. (2) Excusing special cases, amongst all of these stationary states, a clustering of correlations is only consistent with conventional thermal equilibrium states (associated with a sharp distribution of the above state variables). (3) Other stationary states may be modified so as to have local correlations. These deformations concomitantly lead to metastable (yet possibly very long-lived) states. Since the lifetime of the supercooled state is exceptionally long relative to the natural microscopic time scales, their reduced probability densities may be close to those that we find for exact stationary states (i.e., a weighted average of equilibrium probability densities at different state variables). This form suggests several new predictions such as the existence of dynamical heterogeneity stronger than probed for thus far and a relation between the specific heat peak and viscosities. Our theorem may further place constraints on the putative "ideal glass" phase.