FCHL revisited: Faster and more accurate quantum machine learning.

Area of Science:

• Computational Chemistry
• Machine Learning in Chemistry
• Materials Science

Background:

• Accurate prediction of molecular energies and forces is crucial for chemical simulations.
• Existing atomic representations can be computationally expensive for large systems.

Purpose of the Study:

• Introduce the revised FCHL19 representation for atomic environments.
• Develop fast and accurate machine learning models for chemical predictions.

Main Methods:

• Discretization and Monte Carlo optimization of FCHL19 features.
• Integration with Gaussian kernel functions and elemental screening.
• Application of Gaussian process regression and operator quantum machine learning.

Main Results:

• Achieved chemical accuracy for energy learning on QM7b and QM9 datasets.
• Demonstrated low mean absolute error for binding energies of water clusters.
• Attained state-of-the-art accuracy for force learning on the MD17 dataset.
• Enabled predictions in milliseconds per atom using the operator quantum machine learning regressor.

Conclusions:

• The revised FCHL19 representation enables rapid and accurate predictions of atomic forces and energies.
• The model is efficient for both general chemistry problems and molecular dynamics simulations.