AIMSWISS—Ab initio multiple spawning with informed stochastic selections

Yorick Lassmann,Basile F E Curchod

doi:10.1063/5.0052118

Abstract

Ab initio multiple spawning (AIMS) offers a reliable strategy to describe the excited-state dynamics and nonadiabatic processes of molecular systems. AIMS represents nuclear wavefunctions as linear combinations of traveling, coupled Gaussians called trajectory basis functions (TBFs) and uses a spawning algorithm to increase as needed the size of this basis set during nonadiabatic transitions. While the success of AIMS resides in this spawning algorithm, the dramatic increase in TBFs generated by multiple crossings between electronic states can rapidly lead to intractable dynamics. In this Communication, we introduce a new flavor of AIMS, coined ab initio multiple spawning with informed stochastic selections (AIMSWISS), which proposes a parameter-free strategy to beat the growing number of TBFs in an AIMS dynamics while preserving its accurate description of nonadiabatic transitions. The performance of AIMSWISS is validated against the photodynamics of ethylene, cyclopropanone, and fulvene. This technique, built upon the recently developed stochastic-selection AIMS, is intended to serve as a computationally affordable starting point for multiple spawning simulations.

Highlights

Simulating the excited-state dynamics of molecules is a highly complex computational task that necessitates letting go of an approximation at the heart of almost all of ground-state chemistry—the Born–Oppenheimer approximation (BOA).1,2 After being electronically excited, a molecule will relax and reach regions of configuration space where two or more electronic states come close in energy, at which point a coupling between nuclear and electronic motions—the so-called nonadiabatic couplings—will mark the breakdown of the BOA
We shall compare the performance of AIMSWISS to that of Ab initio multiple spawning (AIMS) and the original formulation of stochasticselection ab initio multiple spawning (SSAIMS) (ESSAIMS) on the three previously introduced molecular systems—ethylene, fulvene, and cyclopropanone
ESSAIMS (ε = 10−5 a.u.) is less aggressive in its stochasticselection process, leading to a number of electronic structure (ES) calls that rises to 300 at around 250 atu and oscillates near this value until the basis set is fully reduced to a single trajectory basis functions (TBFs) per initial condition at around 800 atu. In their converged form, become less computational expensive than AIMS at around 700 atu, even if AIMSWISS arrives at this point slightly earlier than ESSAIMS (ε = 10−5 a.u.). In this Communication, we introduced the AIMSWISS approach, which offers a simple strategy to reduce the number of TBFs produced in a multiple spawning dynamics while preserving the accuracy of employing coupled TBFs during nonadiabatic events

Summary

Introduction

Simulating the excited-state dynamics of molecules is a highly complex computational task that necessitates letting go of an approximation at the heart of almost all of ground-state chemistry—the Born–Oppenheimer approximation (BOA). After being electronically excited, a molecule will relax and reach regions of configuration space where two or more electronic states come close in energy, at which point a coupling between nuclear and electronic motions—the so-called nonadiabatic couplings—will mark the breakdown of the BOA. In most cases, moving beyond this approximation means employing the Born–Huang representation of the molecular wavefunction This representation engenders the common picture of photochemistry in which nuclear wavepackets move on distinct electronic potential energy surfaces (PESs) and can transfer amplitude between each other due to nonadiabatic couplings.. In FMS/AIMS, the phase space center of a given TBF moves on a single adiabatic surface along a classical trajectory, and nonadiabatic effects are included by allowing the number of TBFs to increase when needed. Examples are the variational multiconfigurational Gaussian (vMCG) method, which utilizes the time-dependent variational principle to propagate the TBFs, and multiconfigurational Ehrenfest (MCE), where TBFs are evolved classically on a mean-field potential

Methods

Results

Conclusion