Title: Predictive Simulations of Correlated Materials with Quantum Monte Carlo
I will present an overview of the recent efforts in my group to extend the applicability of the Auxiliary-Field quantum Monte Carlo method to real materials, with the ultimate goal of developing a highly predictive, parameter-free, robust and flexible computational tool for the study of correlated solids.
I will discuss recent developments in the method which have helped reduce the severity of many of the challenges presented by the extension of the method to realistic solids. Some of these developments include: i) the use of efficient representations of the second-quantized Hamiltonian to reduce memory requirements, ii) efficient GPU implementations, iii) optimized basis sets for solids, and iv) systematically improvable trial wave-functions. With the rapidly evolving landscape in scientific high performance computing, driven by the continuous introduction of new and disruptive hardware, it is more important than ever to develop carefully designed scientific codes that can be easily adapted to significantly different hardware architectures. I will discuss our efforts to develop a performance portable implementation of AFQMC, capable of running efficiently on emerging GPU architectures without requiring significant code reimplementation. As I will show, the AFQMC method is very well suited for current GPUs, obtaining large performance gains compared to CPU implementations. This represents a dramatic extension in recent years in the applicability and utility of the method. By harnessing the power of upcoming exascale HPC systems and combining it with the extreme scalability and predictive capability of quantum Monte Carlo approaches, I believe we can achieve results on an unprecedented scale in the study of strongly correlated materials in the near future.