Growing multiconfigurational potential energy surfaces with applications to X+H2 (X=C,N,O) reactions

Abstract

A previously developed method, based on a Shepard interpolation procedure to automatically construct a quantum mechanical potential energy surface (PES), is extended to the construction of multiple potential energy surfaces using multiconfigurational wave functions. These calculations are accomplished with the interface of the PES-building program, GROW, and the GAMESS suite of electronic structure programs. The efficient computation of multiconfigurational self-consistent field surfaces is illustrated with the C+H2, N+H2, and O+H2reactions. KeywordsSurface states, Hydrogen reactions, Interpolation, Potential energy surfaces, Exchange reactions Disciplines Chemistry CommentsThe following article appeared in Journal of Chemical Phsyics 124, 154104, and may be found at doi:10.1063/ 1.2185641. Growing multiconfigurational potential energy surfaces with applications to X + H 2 "X = C , N , O… reactions A previously developed method, based on a Shepard interpolation procedure to automatically construct a quantum mechanical potential energy surface ͑PES͒, is extended to the construction of multiple potential energy surfaces using multiconfigurational wave functions. These calculations are accomplished with the interface of the PES-building program, GROW, and the GAMESS suite of electronic structure programs. The efficient computation of multiconfigurational self-consistent field surfaces is illustrated with the C + H 2 , N+H 2 , and O + H 2 reactions.

Machine learning enables long time scale molecular photodynamics simulations

“…Recently, increasing effort has been devoted to ML potentials,17,18 where an accurate representation of the ground state PES including bond breaking19 and formation is promised 16,20–32. Similarly, modified Shepard interpolation is used to construct PESs in low-dimensional systems and adapt them in out-of-confidence regions 33,34. However, the problem of obtaining accurate full-dimensional PESs for excited states in order to simulate long time photodynamics has not been solved yet.…”

Efficient and accurate evaluation of potential energy matrix elements for quantum dynamics using Gaussian process regression

“…6 demonstrates that a scheme in which a database of GPR reference points is constructed during the course of the quantum dynamics simulations would be expected to further improve the accuracy of the GPR approximation. In passing, we note that such a scheme, comparable to the "grow" methodology employed previously using Shepard interpolation, 76,77 is already under development and will be reported shortly.…”

“…Furthermore, within the Shepard interpolation scheme it is possible to "grow" a PES by continuously adding new data points as new regions of configuration space are sampled. 76,77 However, the Shepard interpolation method embodied in Eqs. (20) and (21) 53 ); as noted already, this calculation can be computationally expensive if using ab initio electronic structure calculations to describe the PES.…”

“…In this way, one can "grow" a GPR interpolation during the course of a quantum dynamics simulation, a strategy that has been previously explored in the context of Shepard interpolation. 76,77 Furthermore, we note the statistical basis of GPR allows evaluation of an error estimate for the interpolated PES at any new configuration; 65 as a result, as GWPs evolve according to either variational or non-variational equations-of-motion, one can assess how accurate the GPR interpolation is at a point q i (t) and use this information to assess whether or not additional ab initio calculations are required. As a second refinement of the basic GPR strategy outlined here, we note that the determination of the GPR weights (Eq.…”

Mark S. Gordon

“…Each of these topics has to do with unpairing of electrons, with immediate implications for chemical applications such as: elucidation of methane's inversion [50], proton transfers in excited states [51], treatment of species such as Ti 8 C 12 metcars [52] which are intrinsically not single reference, explanation of observed vibrational structure in spin-orbit coupled levels of diatomics [53], and any number of chemical reactions such as the ring opening mechanism of silacyclobutane [54]. Of course, methodological improvements to the MCSCF program in GAMESS enabled these chemical applications, including a new orbital converger [55], adoption of determinant CI codes from the group of Professor Ruedenberg, scalable MCSCF analytic Hessians [56], and connection of GAMESS' MCSCF code to the surface growing program of Mick Collins [57].…”