Numerical Methods for Monte Carlo Event Generators
Computational and numerical-methods research around Monte Carlo event generation: adaptive integration, rejection sampling, and scientific computing for NLO/NNLO-matched generators.
Monte Carlo event generators such as POWHEG rely on adaptive integration and rejection sampling to turn differential cross sections into unweighted events. The efficiency of rejection sampling depends directly on how tight an upper bound can be constructed for the event-generation probability; poor bounds translate into wasted computing time. I am interested in how modern numerical and machine-learning methods — from tree-based models to simple surrogate models — could improve upper-bound construction and sampling efficiency in this setting, as well as in GPU acceleration and parallel evaluation of matrix elements more broadly.
This is an active area of methodological exploration rather than a line of published results. Concrete findings will be added here, with code and benchmarks, once they are reproducible and public.