QMLMaterial─A Quantum Machine Learning Software for Material Design and Discovery.

Area of Science:

• Computational Chemistry
• Materials Science
• Artificial Intelligence

Background:

• Experimental structural elucidation is complex.
• Theoretical chemistry aids in understanding material properties but faces search space limitations.
• Global search algorithms are crucial for identifying optimal structures.

Purpose of the Study:

• To present QMLMaterial, an AI-powered software for automated in silico structural determination.
• To enable efficient discovery of optimal structures across diverse chemical systems.
• To reduce the computational cost of materials design and discovery.

Main Methods:

• Utilizes an active learning approach with machine learning regression algorithms.
• Employs uncertainty quantification (Bayesian statistics, K-fold cross-validation, bootstrap resampling) for informed structure selection.
• Integrates with quantum chemistry codes and atomic descriptors (e.g., many-body tensor representation).

Main Results:

• Demonstrates QMLMaterial's capability in determining structures for atomic clusters, doped systems, adsorbed molecules, and encapsulated clusters.
• Successfully applied to systems like Na20, Mo6C3 (including spin multiplicity), H2O@CeNi3O5, Mg8@graphene, and Na3Mg3@CNT.
• The active learning strategy enhances the probability of finding global minima with fewer calculations.

Conclusions:

• QMLMaterial offers a powerful and efficient platform for accelerating materials design and discovery.
• The AI-driven approach overcomes limitations of traditional computational methods.
• Facilitates the exploration of complex chemical systems for novel material identification.