The role of many-body correlations in determining potentials for coarse-grained models of equilibrium structure

Coarse-grained (CG) models often employ pair potentials that are parametrized to reproduce radial distribution functions (rdf's) determined for an atomistic model. This implies that the CG model must reproduce the corresponding atomistic mean forces. These mean forces include not only a direct contribution from the corresponding interaction but also correlated contributions from the surrounding environment. The many-body correlations that influence this second contribution present significant challenges for accurately reproducing atomistic distribution functions. This work presents a detailed investigation of these many-body correlations and their significance for determining CG potentials while using liquid heptane as a model system. We employ a transparent geometric framework for directly determining CG potentials that has been previously developed within the context of the multiscale coarse-graining and generalized Yvon-Born-Green methods. In this framework, a metric tensor quantifies the relevant many-body correlations and precisely decomposes atomistic mean forces into contributions from specific interactions, which then determine the CG force field. Numerical investigations reveal that this metric tensor reflects both the CG representation and also subtle correlations between molecular geometry and intermolecular packing, but can be largely interpreted in terms of generic considerations. Our calculations demonstrate that contributions from correlated interactions can significantly impact the pair mean force and, thus, also the CG force field. Finally, an eigenvector analysis investigates the importance of these interactions for reproducing atomistic distribution functions.