Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A New Paradigm for Computational Chemistry

Raphael T. Husistein, Markus Reiher

Abstract

Computational chemistry has become an indispensable tool for generating data and insights, pervading all branches of experimental chemistry. Its most central concept is the potential energy hypersurface, key to all chemistry and materials science, as it assigns an energy to a molecular structure, the necessary ingredient for reaction mechanism elucidation and reaction rate calculation. Density functional theory (DFT) has been the most important method in practice for obtaining such energies, which is mirrored in the use of high-performance computing hardware. In the last two decades, a new class of surrogate potential energy functions has been evolving with remarkable properties: quantum accuracy combined with force-field speed. Until very recently, their application was hampered by the fact that they needed to be trained on truly large system-specific data sets, generated before a computational chemistry study could be started (in sharp contrast to DFT, which, as a first-principles method, works out of the box, but at a far higher price of computational cost). Very recently, this roadblock has been overcome by so-called foundation machine learning interatomic potentials, which are poised to completely change the way we do computational chemistry, likely prompting us to abandon DFT as the prime method of choice for this purpose in less than a decade.

A New Paradigm for Computational Chemistry

Abstract

Computational chemistry has become an indispensable tool for generating data and insights, pervading all branches of experimental chemistry. Its most central concept is the potential energy hypersurface, key to all chemistry and materials science, as it assigns an energy to a molecular structure, the necessary ingredient for reaction mechanism elucidation and reaction rate calculation. Density functional theory (DFT) has been the most important method in practice for obtaining such energies, which is mirrored in the use of high-performance computing hardware. In the last two decades, a new class of surrogate potential energy functions has been evolving with remarkable properties: quantum accuracy combined with force-field speed. Until very recently, their application was hampered by the fact that they needed to be trained on truly large system-specific data sets, generated before a computational chemistry study could be started (in sharp contrast to DFT, which, as a first-principles method, works out of the box, but at a far higher price of computational cost). Very recently, this roadblock has been overcome by so-called foundation machine learning interatomic potentials, which are poised to completely change the way we do computational chemistry, likely prompting us to abandon DFT as the prime method of choice for this purpose in less than a decade.

Paper Structure

This paper contains 15 sections, 4 equations, 1 figure, 1 table.

Table of Contents

  1. Introduction
  2. Machine Learning Interatomic Potentials
  3. The Breakthrough: Foundation MLIPs
  4. Static Generalist Models vs. (Repeated) System-Focused Fine Tuning of Foundation MLIPs
  5. Accuracy
  6. Catastrophic Forgetting
  7. Data and Compute Cost
  8. Uncertainty Quantification
  9. Performance of MLIPs
  10. Challenges pointing to Future Developments
  11. Physical Fidelity
  12. Scalability and Efficiency
  13. Data and Evaluation
  14. Conclusion and Outlook
  15. Acknowledgment

Figures (1)

  • Figure 1: Illustration of the message passing scheme applied to a molecule. (1) A local environment is defined for each atom using a cutoff radius. (2) The resulting graph is constructed, and each atom is assigned an initial feature vector. (3) Each atom passes its feature vector to its neighbors. An atom then uses its own feature vector and those gathered from its neighbors to update its representation. For clarity, only the update of the carbon atom is shown.