Developing a User-Friendly Code for the Fast Estimation of Well-Behaved Real-Space Partial Charges.

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Machine Learning Applications

Background:

• The Quantum Theory of Atoms in Molecules (QTAIM) offers a robust method for determining atomic partial charges based on electron density topology.
• Previous machine learning (ML) models could compute QTAIM charges efficiently but lacked charge conservation, limiting their practical use.

Purpose of the Study:

• To develop a user-friendly code (NNAIMGUI) that integrates ML with an equilibration strategy for accurate and conserved partial atomic charges.
• To address the limitations of atomistic charge predictions by ensuring charge conservation within molecular systems.

Main Methods:

• Development of NNAIMGUI, a code combining ML inference with an equilibration strategy for partial charge calculation.
• Testing the approach across diverse scenarios, including interpolation, extrapolation (chemical reactions), and large molecular systems.

Main Results:

• NNAIMGUI successfully generates partial charges that are chemically accurate and conserve molecular charge.
• The equilibrated charges maintain the high accuracy previously achieved by ML models.
• The code demonstrates reliable performance in interpolation and extrapolation regimes.

Conclusions:

• NNAIMGUI provides a significant advancement in calculating physically meaningful partial atomic charges.
• The flexible architecture allows for training custom models for various atomic properties.
• The GUI-interfaced code enhances the accessibility and intuition of QTAIM descriptor computation, extending its reach beyond theoretical chemistry.