Machine Learning-Enhanced Structure-Based Gaussian Expansion for Efficient Wavepacket Calculations.

Area of Science:

• Computational Chemistry
• Quantum Mechanics
• Molecular Dynamics

Background:

• Theoretical treatment of molecular wavepackets is computationally intensive, limiting applications to complex systems.
• The structure-based Gaussian (SBG) expansion method was previously developed to address this challenge by focusing basis functions on reaction pathways.

Purpose of the Study:

• To develop a highly efficient and versatile approach for wavepacket calculations by integrating machine learning into the SBG expansion method.
• To reduce the computational cost and improve the practicality of theoretical treatments for molecular wavepackets in complex chemical systems.

Main Methods:

• Incorporation of principal component analysis (PCA) for systematic construction of the SBG basis set.
• Integration of Gaussian process regression (GPR) for accurate interpolation of potential energy surfaces.
• Application of the enhanced SBG method to construct full-dimensional nuclear wave functions.

Main Results:

• Successfully reproduced experimental vibrational energies for H3O+ umbrella inversion tunneling up to overtone excited states using only 19 quantum chemical calculations.
• Demonstrated the method's efficiency with 33 SBG bases, significantly outperforming previous computational demands.
• Confirmed the feasibility of the approach for larger, complex systems, including intramolecular hydrogen transfer in 9-hydroxyphenalenone and its deuterated analogs.

Conclusions:

• The machine learning-enhanced SBG expansion method offers a significant advancement in computational efficiency and versatility for molecular wavepacket calculations.
• This approach enables accurate theoretical treatment of complex chemical systems previously considered computationally prohibitive.
• The method holds promise for broader applications in quantum chemistry and molecular dynamics simulations.