Open-shell Configuration State Functions as trial states for phaseless auxiliary-field quantum Monte Carlo
Don Danilov, Hugh G. A. Burton, James Shee
Abstract
The accurate description of open-shell states in molecular systems poses challenges for approximate electronic structure methods. One emerging approach is to variationally optimize the orbitals of a single user-selected Configuration State Function (CSF), i.e., a linear combination of Slater determinants that is constructed as an eigenstate of the total spin operator. In this work we explore the use of these orbital-optimized single CSFs as trial wavefunctions in phaseless auxiliary-field quantum Monte Carlo (ph-AFQMC), a stochastic method that recovers electron correlation by evolving an initial state in imaginary time, subject to a constraint imposed by the trial state. We investigate the symmetric stretch of H-He-H, a series of small p-block compounds with diradicaloid singlet states, twisted ethene conformers, and a bimetallic iron-sulfur complex. We find that accurate energetics are obtained for states dominated by open-shell character, and that for certain atoms and diatomics the accuracy can be significantly improved when the trial is made to also preserve the relevant angular momentum symmetries. Furthermore, our study reveals that the open-shell CSF trial is suboptimal when the target electronic state reflects a significant mixture of ionic and covalent character, e.g., at shorter bond lengths in H-He-H and for the superexchange-stabilized ground state of [2Fe-2S]2+. Our findings suggest that ph-AFQMC using CSF trial states is a promising method for many (but not all) strongly correlated molecular states, especially given its relatively low computational cost, and that it provides a reasonable non-perturbative way to recover some of the dynamic correlation that is otherwise missing in the CSF.
Create a lesson
Related papers
When optimized geometries are not optimal: Implications for inverted singlet-triplet gap predictions
Atreyee Majumdar, Elamathi Selvan Samynathan, Raghunathan Ramakrishnan
Machine Learning the Complete Active-Space Wavefunction Reconstruction
Aniruddha Seal, Bhavnesh Jangid, Helen S. Clifford et al.
Brueckner Orbitals in Local Pair Natural Orbital Coupled-Cluster Theory: DLPNO-BCCD(T) and DLPNO-BCCD(T)Lambda2
Andy Jiang, Tibor Gyori, Elizabeth C. Toth et al.
Exact periodic electrostatics for latent-charge machine-learning potentials
Eslam Ibrahim
Exploiting floating point emulation in tensor contractions in relativistic coupled cluster theory
Maxime Le, Ryan Benazzouk, Jan Brandejs et al.
Effective Harmonic and Anharmonic Models for Vibronic Spectra
Vincenzo Barone