Solving the Schrödinger Equation in the Configuration Space with Generative Machine Learning.

Area of Science:

• Quantum chemistry
• Computational materials science
• Machine learning applications

Background:

• Configuration interaction (CI) is a powerful method for solving the Schrödinger equation for molecules and materials.
• The practical application of CI is severely limited by its unfavorable computational scaling.
• Efficiently selecting significant configurations is crucial for improving CI methods.

Purpose of the Study:

• To develop a machine learning approach for efficient configuration selection in electronic structure calculations.
• To accelerate the convergence to chemical accuracy in quantum chemistry simulations.
• To enable broader applications of generative models in solving the electronic structure problem.

Main Methods:

• A machine learning approach using a generative model is proposed.
• The generative model is iteratively trained to preferentially generate important configurations.
• The method was tested on molecular applications.

Main Results:

• The machine learning approach significantly accelerates convergence compared to random sampling or Monte Carlo CI.
• Chemical accuracy can be achieved much more rapidly.
• Demonstrates the potential of generative models in quantum chemistry.

Conclusions:

• The proposed machine learning method offers a more efficient route to solving the electronic structure problem.
• This work paves the way for wider adoption of generative models in computational chemistry and materials science.
• Accelerated convergence to chemical accuracy is a key benefit for practical applications.