Sparse expansions of multicomponent oxide configuration energy using coherency and redundancy
Luis Barroso-Luque, Julia H. Yang, Gerbrand Ceder
Abstract
The computational study of configuration thermodynamics of crystalline materials relies heavily on accurate representations of the internal energy in terms of configurational variables. Considering the growth of configuration space dimensionality and the relatively constant number of training points, compressed sensing (CS) has been used as an effective paradigm to fit accurate expansions. However, when using a basis, CS only provides guarantees on accurate coefficient recovery under strict sampling requirements. Here, the authors demonstrate how replacing bases with frames, and thus obtaining redundant representations, allows obtaining accurate and highly sparse expansions with less strict sampling requirements.