Transient Response of Strongly Correlated materials to Large Electric Fields: Utilizing the Large Memory Capacity of ARSC's Midnight Machine in a Capability Applications Project

Strongly correlated materials can be tuned to pass through a metal-insulator transition as a function of doping, pressure, or temperature. This high tunability makes them strong candidates for use in so-called "smart materials" that exhibit tunability of their properties. In this capabilities application project (CAP) phase ii project we used the Arctic Region Supercomputing Center (ARSC) Opteron computer (Midnight) to exactly solve for the nonlinear response of these materials to a strong electric field. Our computational algorithm breaks into two portions, one serial and one parallel; the serial part does not scale while the parallel part scales perfectly with the number of compute nodes. The code is transportable having been run on a wide variety of machines with high efficiency and requiring only FORTRAN, LAPACK and BLAS. The main bottlenecks that determine how large a system can be simulated stem from the memory per node on the machine, and the total computational time available. We found that the performance on Midnight was excellent for large jobs (where the executable size was larger than 4 GB) due to the fact that the job could be run with shared memory on one board and fewer processors. Doing so required simply using the proper calling protocol within message passing interface (MPI) to exclusively share memory within the same motherboard. Midnight also showed improved performance over other machines in the 2 GB-4 GB range for the executable because each processor can use up to 4 Gb memory without any sharing, and many other machines require sharing of memory when larger than 2 GB.