|Name:||Taking a Quantum Leap in Time to Solution for Materials Simulations|
|Time:||Monday, June 17, 2013
1:20 PM - 1:40 PM
CCL - Congress Center Leipzig
|Speakers:||Thomas Schulthess, CSCS|
|Abstract:||Modern supercomputers have become very complex machines, built with thousands or tens of thousands of complex nodes consisting of multiple CPU cores or, most recently, a combination of CPU and GPU processors. Efficient simulations on such high-end computing systems require tailored algorithms that optimally map numerical methods to particular architectures. The biggest opportunities for improving time to solution for materials simulations, however, come from methodological improvements that can beat the exponentially scaling complexity of the underlying many-body Schrödinger equation. In this presentation I will discuss recent advances in simulations methods for strongly correlated electron systems that allow us to massively reduce time to solution and thus open new avenues in microscopic modeling of high-Tc superconducting materials. The advances break up in two categories: (1) improvement in quantum Monte Carlo (QMC) algorithms as well as their implementation on modern multi-core and hybrid GPU-GPU systems; (2) improvements to quantum cluster methods that allow us to delay the fermionic sign problem of QMC algorithms and solve the simulations in polynomial rather than exponential time. With scalable implementations on the largest supercomputers available today, we are able to converge the computation of observable quantities, such as the superconducting transition temperature in models of the high-Tc superconductors.
The present work would not have been possible without continued access to resources at the Oak Ridge Leadership Computing Facility (OLCF) and the Swiss National Supercomputing Center (CSCS).