Model selection in atomistic simulation.

Area of Science:

• Computational chemistry and materials science.
• Statistical modeling and machine learning.

Background:

• Atomistic simulation methods vary significantly in computational cost, accuracy, transferability, and parameterization.
• Fair comparison of these diverse methods is challenging but crucial for scientific advancement.

Purpose of the Study:

• To introduce a statistical model selection framework for fair comparison of atomistic simulation methods.
• To demonstrate the utility of this framework in developing new computational models.
• To apply the framework to create a semiempirical model for hydrogen clusters.

Main Methods:

• Utilized statistical model selection techniques to establish a fair comparison metric for different simulation methods.
• Developed a semiempirical model for hydrogen clusters as a case study.

Main Results:

• The statistical model selection approach provides a robust method for comparing diverse simulation techniques.
• The developed semiempirical model for hydrogen clusters demonstrates the practical application of the framework.
• Identified trade-offs between computational cost and accuracy for different simulation methods.

Conclusions:

• Statistical model selection offers a powerful and objective approach to evaluate and develop atomistic simulation methods.
• This methodology facilitates the creation of more efficient and accurate computational tools for scientific research.
• The study highlights the potential for developing novel methods that optimize the balance between accuracy and computational expense.