Abstract
<jats:p>Coarse-grained (CG) simulations efficiently sample the mesoscale morphologies that govern charge, ion, and exciton transport in soft electronic materials, but they integrate out the atomistic degrees of freedom needed to evaluate electronic properties. Recovering these properties conventionally requires expensive backmapping to atomistic structures followed by quantum chemistry. Electronic coarse-graining (ECG) circumvents this by directly learning the conditional distribution of an electronic observable given each CG configuration. Here we introduce conditional flow matching (CFM) as a scalable and distributionally flexible framework for ECG. Unlike Gaussian-process approaches, CFM imposes no functional form on the conditional distribution and trains at a cost that scales linearly with dataset size. Using the HOMO energy of a substituted oligothiophene (ST3M) and the intermolecular HOMO–HOMO coupling of condensed-phase BTBT dimers as benchmarks, we show that CFM accurately captures heteroscedastic, skewed, and heavy-tailed conditional distributions and reconstructs the ensemble electronic distribution upon marginalizing over the equilibrium CG ensemble. We further show that a temperature-conditioned CFMmodel learns a smoothly evolving family of transport maps, enabling accurate interpolation and extrapolation of electronic distributions to unseen temperatures from only a few reference temperatures. Together, these results establish CFM as a scalable, temperature-transferable, and assumption-free framework for recovering electronic property distributions directly from CG simulations.</jats:p>