Neural network potential energy surface and dynamical isotope effects for the N⁺(³P) + H₂ → NH⁺ + H reaction

Abstract

Dynamical isotope effects are calculated for the N+(3P) + H2 → NH+ + H reaction on a new neural network potential energy surface.

Neural Network Potential Energy Surfaces for Small Molecules and Reactions

“…The three-body term was fitted with about 10 200 MRCI points to an rmse of 0.095 kcal/mol (33 cm –1 ) and used in quasiclassical trajectory studies. The three-body term of the PES for the N + ( 3 P) + H 2 → NH + + H reaction was fitted with the PIP-NN method in ref and time-dependent wave packet calculations were performed of the rovibrational state distributions and state resolved angle distributions of the product and of the isotopic effect on them. The PES was fitted to about 30 000 MRCI points to an rmse of 2.7 meV (22 cm –1 ).…”

“…Even if they converge to the desired accuracy, they may require astronomic and unnecessary numbers of epochs; NN fits with tens of thousands of epochs with the Adam optimizer have been reported. 324 W e 6 9 , 1 0 1 , 1 1 5 , 1 1 7 , 1 1 8 , 1 4 2 , 1 6 5 , 1 6 6 , 1 6 8 , 1 8 0 , 2 1 9 a n d others 143,145,147,148,160,161,170,171,189,197,206,224,261,266,271,325 found that it is necessary to use the Levenberg−Marquardt (LM) algorithm. 196 While the cost per epoch is higher with Hessian-based algorithms, the convergence is much better and the overall cost lower.…”

Data Quality, Data Sampling and Data Fitting: A Tutorial Guide for Constructing Full-dimensional Accurate Potential Energy Surfaces (PESs) of Small Molecular Systems

“…48,146 Practically, the required PIPs under the order of M can be obtained by the software, MAGMA 147 and SINGULAR. 148 Full-dimensional accurate PESs of the following systems have been fitted by the PIP-NN method: F + HCl, 149 H + Li2 → Li + LiH, 150 Si + H2 → SiH + H, 151 N + O2 (doublet, quartet, and sextet states), 152 H + MgH → Mg+H2, 153 O + + H2, 154 S + + H2 → SH + + H, 155 K + H2 → H + KH, 156 H + HBr → H2 + Br, 157 N + + H2 → NH + + H, 158 Au + + H2 → H + AuH + , 159 OH + SO → H + SO2, 102,103 C + H2O → CO + H2/2H, 160 F + H2O → HF + OH, 94,126,161 Cl + H2O → HCl + OH, 84 HCl + OH → Cl + H2O, 162 OH + CO → H + CO2, 163 HOCO -, 164 CO + CO, 132 O + H2O, 126,165 H + HO2, 166 C2H2, 167 HF + HF, 168 KRb + KRb → K2 + Rb2, 169 H + H2S → H2 + SH, 170 NH + + H2 → N + H3 + /NH2 + + H, 171 OH + O2 → H + O3/ HO2 + O, 88 OH + + H2 → H + H2O + , 172 H2 + HF, 173 NH3, 174 H2 + HCl, 124 N2 + HF, 175 N2 + HOC + → N2H + + CO/N2 + HCO, 176 CH2OO, 114 dioxirane, 117 O + C2H2 → H + HCCO/CO + CH2, 177 CO + H2O, 123 C2H2 + Ne, 178,179 OH + H2O → H2O + OH, ...…”

Representing globally accurate reactive potential energy surfaces with complex topography by combining Gaussian process regression and neural networks

“…For example, it usually takes tens of thousands of points to build complex-forming reactive PESs dominated by a well even for the simplest triatomic systems. [10][11][12] Therefore, it is a key issue to efficiently sample data points in the nuclear configuration space to make the construction of polyatomic or complex structured PESs truly possible.…”