We describe a method for performing nuclear quantum dynamics calculations using standard, grid-based algorithms, including the multiconfiguration time-dependent Hartree (MCTDH) method, where the potential energy surface (PES) is calculated “on-the-fly”. The method of Gaussian process regression (GPR) is used to construct a global representation of the PES using values of the energy at points distributed in molecular configuration space during the course of the wavepacket propagation. We demonstrate this direct dynamics approach for both an analytical PES function describing 3-dimensional proton transfer dynamics in malonaldehyde and for 2- and 6-dimensional quantum dynamics simulations of proton transfer in salicylaldimine. In the case of salicylaldimine we also perform calculations in which the PES is constructed using Hartree–Fock calculations through an interface to an ab initio electronic structure code. In all cases, the results of the quantum dynamics simulations are in excellent agreement with previous simulations of both systems yet do not require prior fitting of a PES at any stage. Our approach (implemented in a development version of the Quantics package) opens a route to performing accurate quantum dynamics simulations via wave function propagation of many-dimensional molecular systems in a direct and efficient manner.

Methods for solving the time-dependent Schrödinger equation via basis set expansion of the wave function can generally be categorized as having either static (time-independent) or dynamic (time-dependent) basis functions. We have recently introduced an alternative simulation approach which represents a middle road between these two extremes, employing dynamic (classical-like) trajectories to create a static basis set of Gaussian wavepackets in regions of phase-space relevant to future propagation of the wave function [J. Chem. Theory Comput., 11, 8 (2015)]. Here, we propose and test a modification of our methodology which aims to reduce the size of basis sets generated in our original scheme. In particular, we employ short-time classical trajectories to continuously generate new basis functions for short-time quantum propagation of the wave function; to avoid the continued growth of the basis set describing the time-dependent wave function, we employ Matching Pursuit to periodically minimize the number of basis functions required to accurately describe the wave function. Overall, this approach generates a basis set which is adapted to evolution of the wave function while also being as small as possible. In applications to challenging benchmark problems, namely a 4-dimensional model of photoexcited pyrazine and three different double-well tunnelling problems, we find that our new scheme enables accurate wave function propagation with basis sets which are around an order-of-magnitude smaller than our original trajectory-guided basis set methodology, highlighting the benefits of adaptive strategies for wave function propagation.

•A new direct scheme for quantum dynamics simulations of photoexcited states.•Automatic generation of potential energy surfaces with machine learning.•Propagation of diabatic states enables direct grid-based quantum dynamics.•Requires fewer electronic structure evaluations than generation of fitted diabatic surfaces.We present a method for performing non-adiabatic, grid-based nuclear quantum dynamics calculations using diabatic potential energy surfaces (PESs) generated “on-the-fly”. Gaussian process regression is used to interpolate PESs by using electronic structure energies, calculated at points in configuration space determined by the nuclear dynamics, and diabatising the results using the propagation diabatisation method reported recently (Richings and Worth, 2015). Our new method is successfully demonstrated using a grid-based approach to model the non-adiabatic dynamics of the butatriene cation. Overall, our scheme offers a route towards accurate quantum dynamics on diabatic PESs learnt on-the-fly.Download high-res image (90KB)Download full-size image

In a recent article [ J. Chem. Phys. 2015, 143, 094106], we introduced a novel graph-based sampling scheme which can be used to generate chemical reaction paths in many-atom systems in an efficient and highly automated manner. The main goal of this work is to demonstrate how this approach, when combined with direct kinetic modeling, can be used to determine the mechanism and phenomenological rate law of a complex catalytic cycle, namely cobalt-catalyzed hydroformylation of ethene. Our graph-based sampling scheme generates 31 unique chemical products and 32 unique chemical reaction pathways; these sampled structures and reaction paths enable automated construction of a kinetic network model of the catalytic system when combined with density functional theory (DFT) calculations of free energies and resultant transition-state theory rate constants. Direct simulations of this kinetic network across a range of initial reactant concentrations enables determination of both the reaction mechanism and the associated rate law in an automated fashion, without the need for either presupposing a mechanism or making steady-state approximations in kinetic analysis. Most importantly, we find that the reaction mechanism which emerges from these simulations is exactly that originally proposed by Heck and Breslow; furthermore, the simulated rate law is also consistent with previous experimental and computational studies, exhibiting a complex dependence on carbon monoxide pressure. While the inherent errors of using DFT simulations to model chemical reactivity limit the quantitative accuracy of our calculated rates, this work confirms that our automated simulation strategy enables direct analysis of catalytic mechanisms from first principles.

Methods for solving the time-dependent Schrödinger equation generally employ either a global static basis set, which is fixed at the outset, or a dynamic basis set, which evolves according to classical-like or variational equations of motion; the former approach results in the well-known exponential scaling with system size, while the latter can suffer from challenging numerical problems, such as singular matrices, as well as violation of energy conservation. Here, we suggest a middle road: building a basis set using trajectories to place time-independent basis functions in the regions of phase space relevant to wave function propagation. This simple approach, which potentially circumvents many of the problems traditionally associated with global or dynamic basis sets, is successfully demonstrated for two challenging benchmark problems in quantum dynamics, namely, relaxation dynamics following photoexcitation in pyrazine, and the spin Boson model.