Accurate Global Potential Energy Surfaces for the H + CH₃OH Reaction by Neural Network Fitting with Permutation Invariance

Abstract

The H + CH 3 OH reaction, which plays an important role in combustion and the interstellar medium, presents a prototypical system with multi channels and tight transition states. However, no globally reliable potential energy surface (PES) has been available to date. Here we develop global analytical PESs for this system using the permutation-invariant polynomial neural network (PIP-NN) and the high-dimensional neural network (HD-NN) methods based on a large number of data points calculated at the level of the explicitly correlated unrestricted coupled cluster single, double, and perturbative triple level with the augmented correlation corrected valence triple-ζ basis set (UCCSD(T)-F12a/AVTZ). We demonstrate that both machine learning PESs are able to accurately describe all dynamically relevant reaction channels. At a collision energy of 20 kcal/mol, quasi-classical trajectory calculations reveal that the dominant channel is the hydrogen abstraction from the methyl site, yielding H 2 + CH 2 OH. The reaction of this major channel takes place mainly via the direct rebound mechanism. Both the vibrational and rotational states of the H 2 product are relatively cold, and large portions of the available energy are converted into the product translational motion.

Full-dimensional potential energy surface for acetylacetone and tunneling splittings

“…Several comparative studies of these methods have recently appeared. [10][11][12][13] These methods all have in common that they do not rely on a model for representing potentials, e.g., Lennard-Jones, LEPS, exp/6, force-fields, etc. In this sense they are all non-parametric in the language of machine-learning (ML).…”

“…6,8 For example, the recent PIP-NN and atom-based HD-NN PESs for H+CH 3 OH were fit to 75,000 electronic energies. 13 The PIP approach developed in our group spans both these limits, i.e., it can easily fit ca. 10 5 energies (and/or gradients) and it also provides a faithful fit with very few configurations.…”

“…The above ML methods have been employed over the past 10 or so years to develop high-dimensional PESs for reactive systems. 8,9,12,13,[20][21][22][23] Some time ago Fu et al reported a PIP PES for the 7-atom O( 3 P)+C 2 H 4 reaction (which included spin-orbit coupling to the singlet PES). This PES was used in quasiclassical trajectory calculations which yielded excellent agreement with experiment for the branching ratio of numerous products.…”

“…23 Recently, Li et al reported PIP-NN and HD-NN PESs for several chemical reactions. 12,13 The most recent example is the 7-atom H+CH 3 OH reaction. Although these are both neural network methods, they are different in the way the potential is represented.…”

Breaking the Coupled Cluster Barrier for Machine-Learned Potentials of Large Molecules: The Case of 15-Atom Acetylacetone

“…Examples of potentials for six- and seven-atom chemical reactions, which are fits to tens of thousands or even 100 000 CCSD(T) energies, have also been reported. 1 , 3 , 4 , 7 , 8 …”