(747i) Developing Theory and Data-Driven Benchmarks for General Coarse-Grained Mapping Operators

Coare-graining (CG) is a dimensionality reduction technique used to more efficiently simulate multiscale systems. CG methods rely on grouping similar particles together (mapping) and then redefining the potential energy in terms of the reduced particle number. There is a mature theory about generating potentials for CG models. There is not a rigorous theory for generating mappings from the fine-grain (FG) system to the CG system. This missing component is essential because past CG work shows that many mappings lead to homogeneous, weakly interacting, gas-like CG models. Identifying a mapping operator that preserves relevant physics is usually the most heuristic but essential step in developing CG models. The number of possible mappings is exponential with respect to the number of atoms, which is especially a problem since CG mapping are desired when the number of atoms is large.

In this talk I will outline a new graphical rigorous theory for CG mappings with techniques from the field of computer vision. Due to the ill-defined problem of evaluating specific mapping operators, the talk will emphasize data-driven benchmarks for general mapping strategies. Resulting mappings for solute-solvent and liquids will be explored.