(588h) Bypassing Backmapping By Learning the Noise of Electronically Coarse-Grained Models

Coarse-graining methods for molecular and polymeric materials facilitate the “bottom-up” development of multiscale structural prediction models in soft materials and biological applications. Recently, we introduced Electronic Coarse-Graining (ECG) as a novel coarse-graining strategy for developing electronic structure prediction models utilizing coarse-grained representations of soft materials. Here we extend the ECG methodology within a formalism of Gaussian Processes to learn the all-atom electronic noise distributions associated with the coarse-graining decimation procedure. The ability to learn all-atom noise distributions at the coarse-grained resolution within ECG provides an avenue for stochastically sampling the coarse-grained model’s noise distribution for electronic predictions, further enabling the bypassing of traditional multiscale backmapping procedures.