Search

We calculated reaction rate constants including atom tunneling for the hydrogen abstraction reaction CH3OH+H -> CH2OH+H2 with the instanton method. The potential energy was fitted by a neural network, that was trained to UCCSD(T)-F12/VTZ-F12 data. Bimolecular gas-phase rate constants were calculated using microcanonic instanton theory. All H/D isotope patterns on the CH3 group and the incoming H atom are studied. Unimolecular reaction rate constants, representing the reaction on a surface, down to 30 K, are presented for all isotope patterns. At 30 K they range from 4100 for the replacement of the abstracted H by D to ~ 8 for the replacement of the abstracting H to about 2--6 for secondary KIEs. The $^\text{12}$C/$^\text{13}$C kinetic isotope effect is 1.08 at 30 K, while the $^\text{16}$O/$^\text{18}$O kinetic isotope effect is vanishingly small. A simple kinetic surface model using these data predicts high abundances of the deuterated forms of methanol.

Artificial neural networks are used to fit a potential energy surface. We demonstrate the benefits of using not only energies, but also their first and second derivatives as training data for the neural network. This ensures smooth and accurate Hessian surfaces, which are required for rate constant calculations using instanton theory. Our aim was a local, accurate fit rather than a global PES, because instanton theory requires information on the potential only in the close vicinity of the main tunneling path. Elongations along vibrational normal modes at the transition state are used as coordinates for the neural network. The method is applied to the hydrogen abstraction reaction from methanol, calculated on a coupled-cluster level of theory. The reaction is essential in astrochemistry to explain the deuteration of methanol in the interstellar medium.

Artificial neural network (ANN) potentials enable the efficient large-scale atomistic modeling of complex materials with near first-principles accuracy. For molecular dynamics simulations, accurate energies and interatomic forces are a prerequisite, but training ANN potentials simultaneously on energies and forces from electronic structure calculations is computationally demanding. Here, we introduce an efficient alternative method for the training of ANN potentials on energy and force information based on an extrapolation of the total energy via a Taylor expansion. By translating the force information to approximate energies, the quadratic scaling with the number of atoms exhibited by conventional force-training methods can be avoided, which enables the training on reference data sets containing complex atomic structures. We demonstrate for different materials systems, clusters of water molecules, bulk liquid water, and a lithium transition-metal oxide, that the proposed force-training approach provides substantial improvements over schemes that train on energies only. Including force information for training reduces the size of the reference data sets required for ANN potential construction, increases the transferability of the potential, and generally improves the force prediction accuracy. For a set of water clusters, the Taylor-expansion approach achieves around 50% of the force error improvement compared to the explicit training on all force components, at a much smaller computational cost. The alternative force training approach thus simplifies the construction of general ANN potentials for the prediction of accurate energies and interatomic forces for diverse types of materials, as demonstrated here for water and a transition-metal oxide., Comment: 19 pages, 13 figures