Replacing Quantum Chemistry With Machine-Learned Interatomic Potentials: Revolution or Evolution?

Area of Science:

• Computational Chemistry
• Materials Science
• Quantum Chemistry

Background:

• Developing general-purpose interatomic force fields is a long-standing goal.
• Current methods like density functional theory (DFT) offer high accuracy but are computationally expensive.
• High computational cost limits the scope of explorable physical problems.

Purpose of the Study:

• To examine the implications and limitations of machine-learned interatomic potentials (MLIPs).
• To outline a research agenda for leveraging MLIPs in computational chemistry and materials science.

Main Methods:

• Review of current computational chemistry and materials science methodologies.
• Analysis of the impact of machine learning advancements on interatomic potential calculations.
• Exploration of the potential acceleration offered by MLIPs compared to traditional methods like DFT.

Main Results:

• MLIPs are poised to accelerate DFT-level accuracy calculations by up to a million times.
• This acceleration will significantly transform the field of computational chemistry.
• MLIPs offer a pathway to explore complex physical problems previously intractable due to computational constraints.

Conclusions:

• MLIPs represent a paradigm shift in computational chemistry and materials science.
• Further research is needed to fully realize the potential of MLIPs.
• Addressing the limitations of MLIPs is crucial for their widespread adoption and impact.