Efficient Generation of Torsional Energy Profiles by Multifidelity Gaussian Processes for Hindered Rotor Corrections

Area of Science:

• Computational Chemistry
• Chemical Physics
• Theoretical Chemistry

Background:

• Accurate thermochemistry requires precise treatment of torsional modes.
• The one-dimensional hindered rotor model offers computational efficiency for potential energy surface calculations.
• Multifidelity approaches can balance computational cost and accuracy.

Purpose of the Study:

• To demonstrate the utility of multifidelity modeling for hindered rotor calculations.
• To improve the accuracy and efficiency of thermochemical predictions.
• To develop strategies for optimal data selection in computational chemistry.

Main Methods:

• Utilizing a multifidelity approach combining low- and high-fidelity potential energy calculations.
• Applying smooth interpolation with uncertainty estimates for low-fidelity data.
• Employing Bayesian prediction to guide the selection of next computational points.
• Developing acquisition functions tailored for statistical thermodynamics.

Main Results:

• Successful smooth interpolation of low-fidelity data with uncertainty quantification.
• Identification of high-fidelity data crucial for reordering conformational energies.
• Demonstrated efficiency of Bayesian prediction in extending coarse grids for rotor scans.
• Significant computational time savings for one- and multidimensional hindered rotors.

Conclusions:

• Multifidelity modeling is a powerful tool for accurate and efficient thermochemistry.
• The proposed methods enhance the prediction of molecular conformations and transition states.
• Bayesian optimization effectively reduces computational cost in complex molecular systems.