Learning the Short- and Medium-Range Enthalpic Structural Manifolds of Disordered Materials

Enthalpy minimization pulls the local atomic configurations in materials onto a manifold; by parameterizing that enthalpic structural manifold for a particular material, we can obtain a minimal set of structural descriptors that describe the important degrees of freedom in the local atomic configurations in that material. We sampled local atomic configurations from a variety of experimental and simulated glassy materials, compared those configurations using the Gaussian Integral Inner Product distance, and applied hierarchical clustering and diffusion maps to extract low-dimensional discrete and continuous parameterizations of the materials' enthalpic structural manifolds. We used nonlinear regression to interpret these learned parameters in terms of physically meaningful quantities. This presentation will focus on our structure parameterization methodology and physical interpretation of that parameterization, and will also include physical insights gleaned in the process. SNL is managed and operated by NTESS under DOE NNSA contract DE-NA0003525 (SAND2022-8094 A).