Density Functional Tight Binding (DFTB) is a semi-empirical electronic structure method typically used to examine large chemical systems. Linear algebra operations, such as matrix-matrix multiplication, Cholesky factorizations, and diagonalizations, are the most computationally expensive components of the DFTB calculations. Many of these operations are performed using pre-compiled Linear Algebra PACKage (LAPACK) function calls.  There are certain libraries that allow LAPACK calls to run with multiple threads, however the linear algebra routines are inherently serial codes.  Versions of DFTB using LAPACK work extremely well for evaluating small chemical systems, but become impractical as the system size increases. Replacing LAPACK function calls with their parallelized Scalable LAPACK (ScaLAPACK) counterparts will make evaluating larger systems more feasible by running calculations across multiple cores.  In preliminary benchmarking tests, ScaLAPACK function calls were shown to be up to three times faster than LAPACK.  Current work consists of modifying DFTB with the parallel ScaLAPACK routines and then conducting further investigations on how scaling affects the speed up of the parallelized code.

Density functional tight binding with ScaLAPACK for efficient electronic structure calculations

251st National Meeting (2016)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.