Machine Learning Quantum Reaction Rate Constants.

Area of Science:

• Computational Chemistry
• Chemical Kinetics
• Machine Learning in Chemistry

Background:

• Calculating exact quantum reaction rate constants is computationally expensive.
• This high cost hinders rapid kinetic analysis for complex reaction systems.

Purpose of the Study:

• To develop a deep neural network (DNN) for predicting quantum reaction rate constants.
• To overcome the computational limitations of traditional ab initio methods.

Main Methods:

• A DNN was trained on approximately 1.5 million quantum reaction rate constants.
• The dataset included various one-dimensional potentials, reactant masses, and temperatures.
• The DNN predicted the logarithm of rate constants multiplied by reactant partition functions.

Main Results:

• The DNN achieved a 1.1% relative error in predicting the logarithm of the rate product.
• At temperatures below 300 K, DNN predictions showed a 31% relative error compared to classical transition state theory.
• Accurate predictions were observed for diverse systems like H + H2 and hydrogen diffusion on Ni(100).

Conclusions:

• DNNs provide a computationally efficient method for estimating quantum reaction rates.
• This approach enables faster exploration of chemical reactivity in the quantum regime.
• The DNN shows promise for accelerating the design of chemical kinetics for complex reactions.