Extending the Reach of Quantum Monte Carlo Methods via Machine Learning

Explore a conference talk on extending quantum Monte Carlo methods through machine learning. Discover how active learning with AMPTorch predicts QMC-quality forces for molecular geometry relaxation and dynamics simulations. Learn about using Gaussian Process Regression to accurately predict solid energies in the thermodynamic limit. Examine case studies on carbon dimers, water, and H2O molecules, and understand the challenges of short bond lengths. Investigate the effects of statistical error bars and the generalization to multiple degrees of freedom. Gain insights into coupling machine learning methods with quantum Monte Carlo techniques to enable and accelerate complex calculations.

Intro
A QMC DREAM Heterogeneous Catalysis
PREDICTION WORKFLOW
COMPARISONS WITH BENCHMARKS
EXTRAPOLATION COMPARISONS
COMPARISONS WITH SUBTRACTION TRIC • Comparison with the
AB INITIO MOLECULAR DYNAMICS AND RELAXATION
LEARNING FORCE FIELDS
MOLECULAR CASE STUDIES Carbon Dimer, Water, H, 0
MACHINE LEARNING WORKFLOW
C, ENERGY AND FORCE PREDICTIONS Challenges at Short Bond Lengths
C, MOLECULAR DYNAMICS Does Averaging Help? NVE Bond Distance vs. Time
EFFECTS OF STATISTICAL ERROR BARS HO Modeled via AMPTorch-DMC
CH CI: A MORE SOPHISTICATED EXAMPLE Generalization to 9 Degrees of Freedom
CONCLUSIONS AND OUTLOOK Machine Learning Methods Can Be Coupled with Quantum Monte Carlo Methods to Enable and Accelerate Calculations Difficult to Perform Using QMC Alone.