Stochastic methods for electronic structure and correlation in large molecular systems
EMSL Project ID
48414
Abstract
Post Hartree-Fock (HF), fifth rung density functional theory (DFT), and many-body perturbation theory (MBPT) calculations are of importance for the estimation of observables dependent on correlation energy in large systems. Examples include the calculation of cohesion energies and molecular geometries, phonon/vibrational properties in large molecules, electron dynamics, and quasiparticle energies and gaps in nanocrystals. Application of such methods in a straightforward way to large systems of experimental relevance is often hampered by the steep scaling of the computational effort (both CPU and memory) with system size. We therefore propose to develop a novel strategy, facilitating the application of rigorous first-principles electronic structure methods to large systems of unprecedented size. The main goal of the proposal is to develop a new reformulation of existing electron correlation theories, which enables a fast, easily parallelizable, linear scaling (for large systems) and sub-linear scaling (for huge systems) computation effort with very modest memory requirements. The crucial aspect of this approach is the novel combination of operator methods and stochastic sampling techniques, which facilitates a tremendous reduction in the complexity of the calculation. The stochastic aspect bypasses the need to compute and store the molecular orbitals or the DM by expressing all desired variable as stochastic traces over appropriate operators or expressions. We have already achieved sub-linear scaling in calculating MP2 and RPA energies for very large systems, and envision applying our method to systems of an even larger size.
Project Details
Project type
Large-Scale EMSL Research
Start Date
2014-10-01
End Date
2016-09-30
Status
Closed
Released Data Link
Team
Principal Investigator
Team Members