Improving accuracy of systematic coarse-grained simulations

Billion atom simulations are just now becoming possible in molecular simulation for nanoseconds. We've also crossed the millisecond barrier for simulating biomacromolecules. What's left? Unfortunately, a typical cell contains 100 trillion atoms. Even simulating something like a polymer nanoparticle (~100 million atoms) has timescales of interest far beyond nanoseconds. One way around the length-scale limitation is coarse-grained simulation. Coarse-graining requires two ingredients: (i) a mapping that determines how to group atoms into coarse beads and (ii) a force field that describes these interactions. In this talk, I will describe our recent progress on determining mapping operators, which has previously been an arcane topic with little rigor. We've developed novel theory, shown what role symmetry plays, and developed ML models that find mappings for arbitrary molecular systems. Finding the force field of a coarse-grained model is a rich field with a long history. Typically, it is broken into two types: top-down, where we choose the force field to reproduce an observed phonemenon in experiment; and bottom-up, where we draw upon the observed forces in a molecular simulation. I will describe our recent work on combining these approaches to create hybrid top-down/bottom-up models via the principle of maximum entropy.