Self-learning path integral hybrid Monte Carlo with mixed ab initio and machine learning potentials for modeling nuclear quantum effects in water

Bo Thomsen,Yuki Nagai,Keita Kobayashi,Ikutaro Hamada,Motoyuki Shiga

2024-11-15

Abstract:The introduction of machine learned potentials (MLPs) has greatly expanded the space available for studying Nuclear Quantum Effects computationally with ab initio path integral (PI) accuracy, with the MLPs' promise of an accuracy comparable to that of ab initio at a fraction of the cost. One of the challenges in development of MLPs is the need for a large and diverse training set calculated by ab initio methods. This data set should ideally cover the entire phase space, while not searching this space using ab initio methods, as this would be counterproductive and generally intractable with respect to computational <a class="link-external link-http" href="http://time.In" rel="external noopener nofollow">this http URL</a> this paper, we present the self-learning PI hybrid Monte Carlo Method using a mixed ab initio and ML potential (SL-PIHMC-MIX), where the mixed potential allows for the study of larger systems and the extension of the original SL-HMC method [Nagai et al., Phys. Rev. B 102, 041124 (2020)] to PI methods and larger systems. While the MLPs generated by this method can be directly applied to run long-time ML-PIMD simulations, we demonstrate that using PIHMC-MIX with the trained MLPs allows for an exact reproduction of the structure obtained from ab initio PIMD. Specifically, we find that the PIHMC-MIX simulations require only 5,000 evaluations of the 32-bead structure, compared to the 100,000 evaluations needed for the ab initio PIMD result.

Chemical Physics,Materials Science

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is to develop a new self - learning path integral hybrid Monte Carlo method (SL - PIHMC - MIX) by combining machine - learning potentials (MLPs) and ab initio methods, in order to efficiently simulate nuclear quantum effects (NQEs) in water. Specifically, this method aims to:

1. **Reduce computational cost**: Use machine - learning potentials to significantly reduce computational cost while maintaining an accuracy comparable to that of the ab initio path - integral (PI) method. This makes it possible to study larger systems, as traditional ab initio methods are computationally intractable for large - scale systems.

2. **Expand the scope of application**: By using a mixture of ab initio potentials and machine - learning potentials, this method can cover a wider phase space, thereby better simulating nuclear quantum effects. This solves the problem that the training data set of traditional machine - learning potentials needs to cover the entire phase space but cannot rely on ab initio methods for a large amount of computation.

3. **Improve sampling efficiency**: Introduce a hybrid potential function, allowing for a larger acceptance rate and faster phase - space sampling. This is crucial for improving the efficiency and accuracy of the self - learning process.

4. **Accurately reproduce the structure**: By using the PIHMC - MIX method, the structure of water obtained by ab initio path - integral molecular dynamics (PIMD) can be accurately reproduced. Specifically, the paper mentions that using the PIHMC - MIX method requires only 5,000 evaluations of 32 - bead structures, while the traditional ab initio PIMD method requires 100,000 evaluations.

In summary, through the development of the SL - PIHMC - MIX method, this paper aims to provide an efficient and accurate method for simulating nuclear quantum effects in water, especially when dealing with large - scale systems, significantly reducing computational cost and improving sampling efficiency.