LASP to the Future of Atomic Simulation: Intelligence and Automation

Area of Science:

• Computational Materials Science
• Chemical Physics
• Artificial Intelligence in Chemistry

Background:

• Accurate potential energy surfaces and efficient rare event capture are crucial for atomic simulations.
• LASP software (large-scale atomic simulation with a Neural Network Potential) was developed to address these challenges.

Purpose of the Study:

• To review recent developments in LASP software, focusing on increased intelligence and automation.
• To highlight LASP's capabilities in solving complex material and reaction problems.

Main Methods:

• Utilizing global many-body function corrected neural network (G-MBNN) for improved potential energy surface accuracy and linear scaling efficiency.
• Incorporating ASOP and ML-interface methods for surface/interface structure prediction.
• Employing ML-TS and MMLPS methods for identifying lowest energy reaction pathways.

Main Results:

• LASP 3.7 achieves higher accuracy in potential energy surface description at low cost.
• The software demonstrates linear scaling efficiency for large-scale atomic simulations.
• New functionalities enable automated prediction of complex structures and reaction pathways.

Conclusions:

• LASP serves as an intelligent data generator for creating computational databases.
• The software facilitates database construction for applications like zeolite design and catalyst development.