Identifying Coarse-Grained Representations for Electronic Predictions.

Area of Science:

• Computational Chemistry
• Materials Science
• Multiscale Modeling

Background:

• Coarse-grained (CG) simulations are vital in chemistry and materials science.
• Bottom-up CG models aim to capture electronic structure variations but face challenges in selecting effective reduced representations.
• Preserving crucial electronic structure information in simplified models is a key limitation.

Purpose of the Study:

• To develop methods for identifying essential atomic degrees of freedom for electronic structure.
• To create a scoring system for evaluating coarse-grained representations in electronic predictions.
• To bridge the gap between optimized CG representations and the development of simplified model Hamiltonians.

Main Methods:

• A physics-based approach incorporating nuclear vibrations and quantum chemical calculations.
• A machine learning technique using equivariant graph neural networks to assess the contribution of atomic degrees of freedom.
• Integration of both methods to identify critical atomic coordinates and score CG representations.

Main Results:

• Successfully identified important electronically coupled atomic degrees of freedom.
• Developed a robust method for scoring the efficacy of CG representations for electronic predictions.
• Established a link between optimized CG representations and the potential for simplified Hamiltonian development.

Conclusions:

• The proposed methods enhance the accuracy and applicability of coarse-grained simulations.
• This work facilitates the creation of more predictive and efficient multiscale models.
• It paves the way for incorporating complex vibrational modes into simplified computational models.