Machine Learning for Absorption Cross Sections.

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Machine Learning Applications

Background:

• The nuclear ensemble approach (NEA) is crucial for computing absorption cross sections.
• Traditional NEA methods face challenges with statistical sampling errors.
• Accelerating these computations is vital for broader applicability.

Purpose of the Study:

• To present a novel machine learning (ML) method to accelerate the NEA.
• To reduce computational cost and improve statistical accuracy in cross-section calculations.
• To enable accurate predictions for large ensembles of nuclear geometries.

Main Methods:

• Developed ML-NEA, integrating machine learning with the nuclear ensemble approach.
• Utilized the KREG model (kernel-ridge-regression with RE descriptor) within MLatom.
• Calculated electronic properties (excitation energies, oscillator strengths) for a subset of geometries using a reference method.
• Employed ML to predict properties for the remaining geometries, significantly reducing computational expense.

Main Results:

• ML-NEA successfully computed statistically converged absorption cross sections.
• Demonstrated effectiveness on challenging systems like benzene and a dicyanomethylene acridine derivative.
• Achieved accurate results even with a limited number of training points (hundreds).
• Significantly reduced the computational cost associated with large ensemble calculations.

Conclusions:

• ML-NEA offers a powerful and efficient strategy for calculating absorption cross sections.
• The method overcomes limitations of traditional NEA by enhancing statistical sampling.
• This approach holds promise for accurate and cost-effective computational spectroscopy.