Electronic excited states in deep variational Monte Carlo

Area of Science:

• Computational Chemistry
• Quantum Mechanics
• Machine Learning in Science

Background:

• Accurate computation of electronic ground and excited states is vital for understanding chemical reactions and material properties.
• Variational quantum Monte Carlo (QMC) is an ab initio method that scales favorably for large electronic systems.
• Deep QMC, utilizing neural network ansatzes, has shown promise for highly accurate ground-state calculations.

Purpose of the Study:

• To extend the deep QMC approach, specifically the PauliNet ansatz, for the accurate calculation of electronic excited states.
• To demonstrate the method's capability on small atomic and molecular systems.
• To showcase the potential for larger, more complex systems like benzene and ethylene.

Main Methods:

• Extension of the PauliNet neural network ansatz within the deep QMC framework.
• Application of the extended method to compute low-lying electronic excited states.
• Validation against established high-level computational methods for accuracy comparison.

Main Results:

• Consistent high accuracy achieved for low-lying excited states of small atoms and molecules.
• Successful computation of the first excited state of benzene, a larger system.
• Accurate determination of the conical intersection of ethylene, matching results from more computationally expensive methods.

Conclusions:

• The extended PauliNet ansatz provides a highly accurate and scalable method for computing electronic excited states.
• This approach overcomes limitations of traditional methods for large and complex molecular systems.
• The method holds significant potential for advancing research in various fields requiring precise electronic structure calculations.