Quantum-Chemical Methods for Large Systems: Low-, Linear-, and Sublinear-Scaling Methods

TH Department Online Seminar
Date: Aug 19, 2022
Time: 11:00 AM (Local Time Germany)
Speaker: Prof. Christian Ochsenfeld
Chair of Theoretical Chemistry, University of Munich (LMU), Germany
Location: https://zoom.us/j/96434291324?pwd=SmFDT3lZUnZHcU5KVmloWkN0L0VjUT09
Room: Meeting ID: 964 3429 1324 | Passcode: 487475
Host: TH Department

An overview of our recently developed low-, linear-, and sublinear-scaling methods ranging from HF, DFT, MP2 to RPA is given. These methods allow — also in combination with graphics processing units (GPUs) — for the efficient description of large systems at QM and QM/MM levels, where QM spheres with typically 500-1000 atoms are necessary for reliable studies.

In particular, we recently introduced a novel semi-numerical integration scheme that allows for linear-scaling exchange-type contractions on both CPUs and GPUs leading to two to three orders of magnitude speedups [1,2] as compared to conventional integration. This not only accelerates static calculations, but also ab-initio molecular dynamics simulations and the necessary sampling for complex systems.
Furthermore,linear-scaling reformulations for computing electron-correlation effects via RPA (random-phase approximation) and beyond RPA methods are described that employ atomic orbitals and Cholesky-decomposed densities (e.g., Ref. [3]). Finally, reformulations for computing molecular response properties such as g-tensors and NMR chemical shieldings are discussed and linear- and sublinear-scaling reformulations open up new possibilities for bridging theory and experiment.

[1] H. Laqua, T. H. Thompson, J. Kussmann, C. Ochsenfeld; JCTC 16, 1456 (2020).

[2] H. Laqua, J. Kussmann, C. Ochsenfeld; JCP 154, 214116 (2021).

[3] V. Drontschenko, D. Graf, H. Laqua, C. Ochsenfeld; JCTC 17, 5623 (2021).