Efficient simulation of large materials clusters using the jaguar quantum chemistry program: Parallelization and wavefunction initialization

Abstract

An outline of the improvements to the pseudospectral electronic structure program Jaguar is presented, showing efficient and robust performance of hybrid‐DFT calculations for large systems with thousands of basis functions, focusing on materials applications. The improvements include re‐engineered parallelization, the design of a fragment‐based initial guess generation method, and the validation of small eigenvalue cutoff values. An OpenMP/MPI hybrid parallelization has been implemented for the pseudospectral algorithm, which extends Jaguar's scalability to up to 256 cores in tests of TiO2 clusters with 1295‐4961 basis functions. In the largest test case, the code delivers 84.4× speedup for 128 cores in total calculation time. In addition, a fragment‐based initial guess method has been constructed for large systems containing many transition metals, where the conventional (atomic) approach often fails. Overall, Jaguar is now capable of efficiently and robustly performing hybrid‐DFT geometrical optimizations for large systems with more than 600 atoms in reasonable runtimes. © 2015 Wiley Periodicals, Inc.

Highly Efficient, Linear-Scaling Seminumerical Exact-Exchange Method for Graphic Processing Units

“…During the last 15 years, graphic processing units (GPUs) have gained increasing interest within the quantum chemistry community, focusing, in particular, on the evaluation of 4-center-2-electron (4c-2e) integrals, which represent the major bottleneck in most Hartree–Fock and Kohn–Sham calculations. − Since, for Hartree–Fock (HF) and hybrid density functional theory (DFT) calculations, the mandatory computation of exact (Fock-like) exchange matrices is particularly expensive, efficient and linear-scaling implementations have been developed since the late 1990s − including recent developments. , However, for larger molecules and particularly in combination with larger basis sets (i.e., triple-ζ or larger), resolution-of-the-identity (RI) , or seminumerical methods, that is, grid-based methods employing 3-center-1-electron (3c-1e) integrals, − are possibly more efficient due to their superior O ( N bas 2 ) formal scaling compared to the formal O ( N bas 4 ) scaling of the conventional 4c-2e integral-based methods.…”

Accelerating the Convergence of Self-Consistent Field Calculations Using the Many-Body Expansion

“…Nevertheless, there is surprisingly little interest in the literature devoted to this subject. − The most popular choice for the initial guess is the superposition of atomic densities (SAD) guess which is used as the default guess in most popular quantum chemistry packages. However, this conventional initial guess method via a block-diagonal density matrix (DM) arising from a set of atomic densities is not sufficient for large systems, especially ones with delocalized electron densities, and therefore is far from the converged value.…”

Assessment of Initial Guesses for Self-Consistent Field Calculations. Superposition of Atomic Potentials: Simple yet Efficient

“…Such a procedure has recently been advocated for real-space calculations; the use of confinement potentials has also been found to help SCF convergence in the case of extended real-space basis sets . In some cases it is also possible to decompose the system into either single molecules or chemically meaningful molecular fragments, ,, and “glue” the orbitals together to form a good guess density for the original calculation. However, the problem of the proper choice of the guess orbitals is not solved by either of these approaches, but rather just moved to the additional calculation(s) in the smaller basis set, or delegated to the isolated molecular fragments.…”