Algorithmic Development for Quantum Monte Carlo

One of the most important weapons in our arsenal for the attack on the quantum many-body problem is the quantum Monte Carlo (QMC) algorithm. The ability to exactly simulate bosons in thermal equilibrium to within statistical uncertainties has fostered a massive increase in our grasp of quantum matter and exotic phases. A major aspect of my research has been through core contributions to the Del Maestro group continuous space QMC code. Society is currently coasting on the fumes of Moore’s law which has made abundantly clear that the most productive way to achieve more measurements, larger system sizes, and better understanding of the quantum many-body problem through the lens of QMC simulations is algorithmic improvements. One tantalizing new prospect is applying machine learning to the spatial continuum. Open source libraries offer the ability to use deep learning for phase discrimination and perhaps learn more efficient traversal paths through the Markov chain. Possible high impact outcomes include a sign-problem free QMC algorithm. Another course for enhancement is to use evolutionary computation to optimize heuristically chosen weights through optimization of number of measurements, autocorrelation time, and CPU hours. I plan to continue development on QMC algorithms with a focus on improving computational performance. Immediate gains can be realized through leveraging parallel processing when calculating long range interactions by adding CUDA support or porting the code to Julia. A code port could harness built-in multithreading, LAPACK, and BLAS capabilities as well as have the additional benefit of a lower barrier to entry for code development by the scientific community.