Machine-learning accelerated geometry optimization in molecular simulation.

Area of Science:

• Computational Materials Science
• Surface Science
• Quantum Chemistry

Background:

• Geometry optimization is crucial for determining ground state structures and reaction pathways in materials and surface science.
• Current methods using quantum chemical codes like density functional theory (DFT) are computationally expensive, limiting optimization speed.
• Accelerating geometry optimization allows for more efficient research and analysis of thermodynamic and kinetic properties.

Purpose of the Study:

• To develop and present a novel method for accelerating local geometry optimization.
• To enable simultaneous optimization of multiple atomic configurations.
• To reduce the computational cost associated with geometry optimization tasks.

Main Methods:

• Implementation of a neural network (NN) ensemble based active learning approach.
• Simultaneous local geometry optimization for multiple configurations.
• Utilizing the Atomic Simulation Environment (ASE)-optimizer Python package for practical application.

Main Results:

• Demonstrated acceleration of geometry optimization across various case studies, including metal surfaces and reactions.
• The NN ensemble method consistently required fewer DFT calculations compared to standard optimization techniques.
• Successful application to bare metal surfaces, surfaces with adsorbates, and nudged elastic band calculations for reactions.

Conclusions:

• The proposed NN ensemble active learning method effectively accelerates geometry optimization.
• This acceleration leads to a significant reduction in computational resources (DFT calculations).
• The ASE-optimizer package facilitates the adoption and use of this advanced technique in scientific research.