Abstract:The rise of machine learning has greatly influenced the field of computational chemistry, and that of atomistic molecular dynamics simulations in particular. One of its most exciting prospects is the development of accurate, full-dimensional potential energy surfaces (PESs) for molecules and clusters, which, however, often require thousands to tens of thousands of ab initio data points restricting the community to medium sized molecules and/or lower levels of theory (e.g. DFT). Transfer learning, which improves a global PES from a lower to a higher level of theory, offers a data efficient alternative requiring only a fraction of the high level data (on the order of 100 are found to be sufficient for malonaldehyde). The present work demonstrates that even with Hartree-Fock theory and a double-zeta basis set as the lower level model, transfer learning yields CCSD(T)-level quality for H-transfer barrier energies, harmonic frequencies and H-transfer tunneling splittings. Most importantly, finite-temperature molecular dynamics simulations on the sub-microsecond time scale in the gas phase are possible and the infrared spectra determined from the transfer learned PESs are in good agreement with experiment. It is concluded that routine, long-time atomistic simulations on PESs fulfilling CCSD(T)-standards become possible.

What problem does this paper attempt to address?

The paper primarily aims to address the following issues: ### Research Background and Objectives - **Application of Machine Learning in Chemistry**: In recent years, machine learning methods have had a significant impact on computational chemistry, especially in atomic-scale molecular dynamics simulations. One important direction is the development of accurate and full-dimensional potential energy surfaces (PESs) for molecules and clusters. - **Limitations of Traditional Methods**: Constructing multidimensional potential energy surfaces suitable for long-time molecular dynamics simulations remains challenging. Traditional methods rely on pre-calculating reference energies and forces from electronic structure methods at an appropriate theoretical level, and these data are used to fit parameterized forms to represent the PES. The difficulty of this approach lies in the need to "guess" sufficiently flexible parameterized forms to capture both local and global features of the PES. ### Solution - **Application of Transfer Learning**: The paper proposes a method based on transfer learning (TL) to improve potential energy surfaces from a lower theoretical level (LL) to a higher theoretical level (HL). This method requires only a small amount of high-level data to achieve efficient data utilization. - **Research Object**: The paper uses malonaldehyde (MA) as the research object because it is a molecular system that has been extensively studied experimentally and can serve as a suitable benchmark system for evaluation. ### Main Results - **Energy Barriers, Frequencies, and Tunneling Splitting**: The PES obtained through transfer learning can accurately predict the energy barriers for hydrogen transfer, harmonic frequencies, and tunneling splitting of hydrogen transfer. - **Molecular Dynamics Simulations**: The paper demonstrates that even using Hartree-Fock theory and a double-zeta basis set as the lower theoretical level model, transfer learning can produce results close to the quality of the Coupled Cluster Singles, Doubles, and Perturbative Triples (CCSD(T)) level. This makes gas-phase molecular dynamics simulations on the microsecond timescale possible. - **Consistency of Infrared Spectra**: The infrared spectra obtained from dynamics simulations on the PES derived from transfer learning are in good agreement with experimental results, validating the effectiveness and practicality of the method. In summary, the paper aims to improve the potential energy surfaces at a lower theoretical level using a transfer learning approach with a small amount of high-level theoretical data, thereby enhancing the accuracy and efficiency of molecular dynamics simulations.

Transfer-Learned Potential Energy Surfaces: Towards Microsecond-Scale Molecular Dynamics Simulations in the Gas Phase at CCSD(T) Quality

Combining state-of-the-art quantum chemistry and machine learning make gold standard potential energy surfaces accessible for medium-sized molecules

PhysNet Meets CHARMM: A Framework for Routine Machine Learning / Molecular Mechanics Simulations

Towards Exact Molecular Dynamics Simulations with Machine-Learned Force Fields

Learning from the Density to Correct Total Energy and Forces in First Principle Simulations

High-Fidelity Potential Energy Surfaces for Gas-Phase and Gas- Surface Scattering Processes from Machine Learning

Challenges for Machine Learning Force Fields in Reproducing Potential Energy Surfaces of Flexible Molecules

Quantum Hardware-Enabled Molecular Dynamics via Transfer Learning

Molecular Dynamics with Neural-Network Potentials

A transferable active-learning strategy for reactive molecular force fields

Training Transferable Interatomic Neural Network Potentials for Reactive Chemistry: Improved Chemical Space Sampling

An automated framework for exploring and learning potential-energy surfaces

Advanced Potential Energy Surfaces for Molecular Simulation

Incorporating Electronic Information into Machine Learning Potential Energy Surfaces via Approaching the Ground-State Electronic Energy as a Function of Atom-Based Electronic Populations

Training Neural Nets To Learn Reactive Potential Energy Surfaces Using Interactive Quantum Chemistry in Virtual Reality

Large-Scale Atomic Simulation via Machine Learning Potentials Constructed by Global Potential Energy Surface Exploration

Machine Learning in QM/MM Molecular Dynamics Simulations of Condensed-Phase Systems

FINETUNA: Fine-tuning Accelerated Molecular Simulations

Machine-learning-accelerated simulations to enable automatic surface reconstruction

Subsystem Density Functional Theory Augmented by a Delta Learning Approach to Achieve Kohn–Sham Accuracy

Beyond potential energy surface benchmarking: a complete application of machine learning to chemical reactivity