Abstract: Coarse-graining of nuclear degrees of freedom constitutes an integral part of the multiscale simulation toolset for soft materials. Here, I introduce electronic coarse-graining as a technique for mapping the configurationally dependent electronic structure of an underlying all-atom system directly to coarse-grained representations via supervised machine learning. By eliminating the atomistic backmapping procedure and ad nauseum quantum-chemical calculations, electronic coarse-graining has the potential to dramatically accelerate multiscale simulation workflows for predicting configurationally dependent electronic structure in soft materials. The methodology is introduced in the context of predicting HOMO and LUMO band energy levels, charge density distributions, optical spectra, and intermolecular electronic couplings directly from the coarse-grained representations of thiophene oligomers. It is further demonstrated that electronic coarse-graining can be employed to identify coarse-grained mappings that optimally capture the configurational dependence of a molecule’s electronic structure variations. Finally, it is shown how the methodology can be integrated with phenomenological quantum-mechanical Hamiltonians to enable molecular weight transferability of electronic predictions in polymeric materials entirely at a coarse-grained resolution.
AI & HPC Seminar