TIME: 3:30 PM

LOCATION: GMCS 314

SPEAKER: Mehmet Dogan, San Diego State University (Physics)

ABSTRACT: Using density functional theory (DFT) and pseudopotentials, we can accurately describe the electrons in most materials. In practice, though, solving the Kohn–Sham equations (the working equations of DFT) is still very expensive on a computer. This usually limits us to systems containing at most a few thousand atoms. In this work, I present new numerical methods that allow us to study systems with more than 100,000 atoms. Our approach is based on using a real-space basis with finite-difference techniques. This introduces several key advantages, such as avoiding heavy global communication and simple control of accuracy via the grid density. To efficiently solve the large linear-algebra problems that arise from real-space DFT, we use a technique called Chebyshev-filtered subspace iteration (CheFSI). Using CheFSI in a real-space finite-difference framework, we have developed a method that can handle confined systems with over 100,000 atoms (about 400,000 electrons). Our tests show that this approach significantly reduces communication overhead and makes better use of the vector-processing capabilities available on modern parallel supercomputers.

BIO: Mehmet Dogan is a new faculty member in SDSU Department of Physics. Dr. Dogan studies technologically relevant materials at the sub-nanometer scale using fundamental physical equations as implemented by software. He received a Ph.D. in Physics from Yale University and has worked as a researcher at UC Berkeley and UT Austin. His current interests are high-pressure hydride superconductors, novel 2D semiconductor alloys, defect-based quantum emitters in 2D materials, and real-space methods.

HOST: CSRC