(189w) GPU Accelerated Experiment Directed Metadynamics for Scattering Profile Biasing

Molecular dynamics (MD) simulations can reveal key information about the behavior of materials at the atomic scale, but mismatches between experimental conditions and inaccuracies in force-fields can make comparison and iterative improvement from experimental data difficult. We can mitigate this issue by minimally bias collective variables to match ensemble averages through maximum entropy biasing techniques. With Experiment Directed Simulation, we can even match entire PMFs to experimental data. Examples of when a reference PMF is available include radial distribution function matching in coarse-graining or structure factors derived from small-angle x-ray scattering data. A major challenge is that these techniques require accurate evaluation of the structure factor, and its derivative, at each timestep. In this work, I will show our GPU-accelerated implementation of structure factor calculations, including treating background subtraction from reference simulated systems. This new approach enables both rapid evaluation of Monte Carlo searches of crystal structures and MD of solutions.