Automated Calculation of Thermal Rate Coefficients using Ring Polymer Molecular Dynamics and Machine-Learning Interatomic Potentials with Active Learning

Ivan S. Novikov,Yury V. Suleimanov,Alexander V. Shapeev

DOI: https://doi.org/10.1039/C8CP06037A

2018-11-15

Abstract:We propose a methodology for fully automated calculation of thermal rate coefficients of gas phase chemical reactions, which is based on combining the ring polymer molecular dynamics (RPMD) with the machine-learning interatomic potentials actively learning on-the-fly. Based on the original computational procedure implemented in the RPMDrate code, our methodology gradually and automatically constructs the potential energy surfaces (PESs) from scratch with the data set points being selected and accumulated during the RPMDrate simulation. Such an approach ensures that our final machine-learning model provides reliable description of the PES which avoids artifacts during exploration of the phase space by RPMD trajectories. We tested our methodology on two representative thermally activated chemical reactions studied recently by RPMDrate at temperatures within the interval of 300--1000~K. The corresponding PESs were generated by fitting to only a few thousands automatically generated structures (less than 5000) while the RPMD rate coefficients retained the deviation from the reference values within the typical convergence error of RPMDrate. In future, we plan to apply our methodology to chemical reactions which proceed via complex-formation thus providing a completely general tool for calculating RPMD thermal rate coefficients for any polyatomic gas phase chemical reaction.

Chemical Physics,Computational Physics

What problem does this paper attempt to address?

This paper aims to solve the problem of accurate and efficient calculation of thermal chemical reaction rate coefficients. Specifically, the author proposes a fully - automatic method for calculating the thermal rate coefficients of gas - phase chemical reactions, which combines ring - polymer molecular dynamics (RPMD) with machine - learning interatomic potential surfaces (IP) based on active learning (AL). The main innovation of this method lies in its ability to automatically construct potential energy surfaces (PESs) from scratch and gradually accumulate data sets during the RPMD simulation process, ensuring that the final machine - learning model can provide a reliable PES description and avoid human - made errors during the phase - space exploration process.

### Core issues of the paper
1. **Accurate calculation of thermal chemical reaction rate coefficients**: Traditional quantum - dynamics calculations are limited in multi - atomic systems due to the exponential growth of computational complexity, while classical descriptions can be used for practical simulations of multi - atomic systems, but quantum effects such as zero - point energy and tunneling effects become very important at low temperatures. Therefore, a computational method that takes into account quantum effects and can efficiently handle multi - atomic systems is required.

2. **Automated construction of potential energy surfaces**: The existing RPMDrate code can only be applied to a limited number of chemical reactions because the PESs of these reactions must be provided in an analytical form in advance. In order to make RPMDrate a general - purpose tool, it is necessary to efficiently combine RPMD with electronic - structure evaluation and automatically construct PESs.

### Solutions
- **Combination of RPMD and machine learning**: By combining RPMD and machine - learning potential energy surfaces, especially by using an active - learning strategy, the training set can be adaptively selected during the simulation process, thereby gradually constructing a reliable PES.
- **Active - learning strategy**: Use the D - optimality criterion and the maxvol algorithm to select the configurations in the training set, ensuring that the training set can cover the key regions of the phase - space and avoid human - made errors.

### Application and testing
- **Test systems**: The author selected two representative thermally - activated chemical reactions for testing, namely OH + H₂ → H + H₂O and CH₄ + CN → CH₃ + HCN. The PESs of these reactions have been constructed by high - precision quantum - mechanical models and are used as references.
- **Result verification**: The accuracy of the AL - MTP method was verified by comparison with the reference values. The results show that the deviation between the RPMD rate coefficients calculated using the AL - MTP method and the reference values is within the typical convergence error range.

### Future prospects
- **Extension to complex - formation reactions**: There are plans to extend this method to chemical reactions through complex formation, in order to provide a general - purpose tool applicable to the RPMD thermal rate coefficient calculation of any multi - atomic gas - phase chemical reaction.
- **Calculation of other dynamic properties**: This method is not limited to the calculation of rate coefficients, but can also be used for the calculation of other dynamic properties, such as RPMD diffusion coefficients.

Through these innovations, this paper provides a new, efficient and reliable method for the calculation of thermal chemical reaction rate coefficients.