Get 10 materials for the price of 1: computationally efficient DFT
ORAL
Abstract
We have developed a numerical algorithm that will allow us to run density functional theory (DFT) computations ten times faster. Our improvements in efficiency result from accelerating the convergence rate of the band energy calculation by employing local polynomial interpolation and adaptive mesh refinement. Tests of the algorithm on 2D toy pseudopotentials show an order of magnitude improvement over the rectangular method, the method currently implemented in DFT programs. Our tests demonstrate two counterintuitive results: 1) accurately approximating the Fermi surface is critical to accurately calculate the band energy, and 2) band crossings are inconsequential. Implementing our algorithm in DFT programs will accelerate both the growth of theoretical materials databases and the discovery of materials.
*Authors acknowledge funding from ONR (MURI N00014-13-1-0635).
–
Presenters
-
Jeremy Jorgensen
- Brigham Young University