Coarse-grained modeling of simple molecules at different resolutions in the absence of good sampling

Many problems of interest in modern science originate from the complex network of interactions of different molecular structures, each possessing its own typical length and time scale of relevance. In such materials, nontrivial properties emerge from the different length and time scales involved that could not be predicted from the properties of each individual subunit taken alone. A solution to the formidable theoretical and computational issues raised by these systems involves coarse-graining, a procedure in which multiple atoms are grouped into a few interaction sites. The coarse-grained approach aims at constructing an effective Hamiltonian from available information about the system and then using this Hamiltonian to investigate the behavior of the system on the length and time scales of interest. In this paper, we aim at determining how far we can coarse-grain a system using only the commonly used pairwise, spherically symmetric potentials, as well as assessing the impact of poor initial sampling on the quality of the resulting coarse-grained model. Coarse-graining is performed following the multiscale coarse-graining (MS-CG) methodology, and we use as a model system the N-methylacetamide (NMA) molecule, a simple representation of a peptide bond, which can adopt two conformations, cis and trans. Our simulations reveal that as the coarse-graining becomes more aggressive multibody effects start to emerge and that the initial sampling of conformations can adversely bias the model in the case of heavy coarse-graining.