(73a) Multiscale Electronic Coarse-Graining of Soft Materials Via Supervised Machine Learning

Authors: 
de Pablo, J., University of Chicago
The computational design of soft materials with advanced electronic, ionic, or optical functionalities depends on the accurate treatment of (i) classical morphologies and (ii) quantum-mechanical electronic structure. This effort is challenging due to the multiscale nature of soft materials in which the strong coupling of disparate length and time scales mediates material performance. Traditionally, a hybrid classical and quantum-mechanical approach is employed, for which the sampling of conformational space can be demanding. Coarse-grained (CG) models can produce representative morphologies, but CG configurations must be back-mapped into atomistic representations to perform quantum-mechanical calculations, which is computationally intensive and inconsistent with the spatial resolution of the CG models.

Here, a supervised machine learning methodology, denoted as artificial neural network electronic coarse graining (ANN-ECG), is described in which the conformationally dependent electronic structure of a system is directly mapped to CG degrees of freedom. By averaging over decimated degrees of freedom, ANN-ECG accelerates simulations by orders of magnitude via eliminating backmapping and repeated quantum-chemical calculations. The approach is accurate, consistent with the CG spatial resolution, and can be used to derive physical intuition regarding collective variables that mediate electronically functional soft materials.