2025 Theses Doctoral

Applications of Phaseless Auxiliary Field Quantum Monte Carlo to Transition Metal Containing Systems

Vuong, Hung Tien

This thesis presents progress towards the generation benchmark quality reference data for transition metal complexes from first principle quantum chemistry with the phaseless Auxiliary Field Quantum Monte Carlo (ph-AFQMC). Chapter 1 presents an algorithmic improvement to the ph-AFQMC method, reducing both the asymptotic scaling and memory requirement of ph-AFQMC using multi-Slater determinant trial wavefunctions by exploiting the low-rank structure of the Coulomb integrals in localized orbitals basis. Implementation of this algorithm on GPUs has been significantly optimized, leading to an order of magnitude reduction in wall time for several test systems.

With the new machinery developed, the next three chapters present the application of ph-AFQMC using multi-Slater determinants trials for transition metal containing systems. In chapter 2, vertical ionization potentials of 28 transition metal complexes with varying degree of static correlation is computed with both ph-AFQMC and canonical CCSD(T) in the small double-𝛇 basis. Careful assessment of multireference diagnostics has been carried out, and we delineate the regime where the use of CCSD(T) is appropriate, based on the agreement between the two methods. For multireference molecules, we demonstrate that it is possible to generate reference-quality data with ph-AFQMC, by converging away the phaseless bias.

Chapter 3 presents an attempt to develop a cheaper complementary approach to generate multi-Slater determinants for TM complexes than the ones employed in Chapter 2. Three different approaches are evaluated on the dataset in Chapter 2, where we find that a combination of both orbital optimization and large active space is necessary. We take the best-performed protocol to compute the vertical ionization of a set of metallocenes, extrapolating to complete basis set (CBS) limit and compare with experiments.The performance of AFQMC using our newly developed protocols has reasonable errors from experimental values, and by using a very accurate CISD trial wavefunction, it is possible converge to the experimental values within 3 kcal/mol.

In chapter 4, the performance of both ph-AFQMC and DLPNO-CCSD(T) on three datasets of reaction energy and barrier heights for large organometallic complexes is assessed. Agreement between ph-AFQMC and DLPNO-CCSD(T) is found to be excellent at the triple-𝛇 level for 177/189 data points, representing the majority of the combined datasets.While the outstanding outlier cases need to be investigated further, the current results (once an appropriate CBS extrapolation scheme is employed), already present a substantial amount of data to assess existing DFT functionals, or train new ones, to scale up calculations on TM systems containing hundreds of atoms.