Robust Statistical Mechanics with Fast Particle Toolkit

Many of the analysis formulas in statistical mechanics are just as numerically demanding as the original simulation producing classical molecular dynamics trajectory data. Example methods include coordinate transformations, accumulating spatial functions, computing energy differences, particle insertion, and time-integals of work. This talk presents a type-based mathematical framework implemented in performance-portable SYCL C++ for carrying out simulation and analysis tasks. Its design creates a domain-specific language for declaring 1-body (particle-cell) and 2-body (particle-particle) operations using C++ code. Its utility is demonstrated by showing a fully self-contained, DSL-based, computation of the pressure-volume curve for steam. Both simulation and analysis make use of full hardware-acceleration on CPUs and GPUs.