Gaussian process model of 51-dimensional potential energy surface for protonated imidazole dimer

Area of Science:

• Computational Chemistry
• Materials Science
• Quantum Mechanics

Background:

• Accurate potential energy surfaces (PESs) are crucial for understanding molecular behavior.
• High-dimensional molecular systems pose significant computational challenges for PES calculations.
• Traditional methods require extensive ab initio computations, limiting applicability.

Purpose of the Study:

• To develop a system-agnostic method for accurate PES generation using limited ab initio data.
• To leverage Gaussian processes (GPs) for efficient modeling of high-dimensional PESs.
• To demonstrate the extrapolation capabilities of GP models for PES prediction.

Main Methods:

• Probabilistic modeling using Gaussian processes (GPs).
• Utilizing composite kernels to enhance Bayesian information content.
• Representing the global PES as a sum of full-dimensional and fragment-based GP models.
• Application to a 51-dimensional PES and the protonated imidazole dimer (19 atoms).

Main Results:

• Achieved global accuracy of <0.2 kcal/mol for a 51-dimensional PES using 5000 ab initio calculations.
• Successfully constructed the global PES for the protonated imidazole dimer.
• Demonstrated accurate extrapolation of PES from low (<10,000 cm⁻¹) to high (>20,000 cm⁻¹) energies.

Conclusions:

• Gaussian processes provide an efficient and accurate approach for high-dimensional PES modeling.
• The method significantly reduces the number of required ab initio calculations.
• GP extrapolation capabilities open new avenues for studying phase transitions and accelerating Bayesian optimization in complex systems.