Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Open-Source Fermionic Neural Networks with Ionic Charge Initialization

Shai Pranesh, Shang Zhu, Venkat Viswanathan, Bharath Ramsundar

TL;DR

This work tackles accurate electronic structure calculations for molecules and ions using variational Monte Carlo with neural-network wavefunctions by integrating the FermiNet into the open-source DeepChem platform. It introduces a Mulliken-based ionic charge initialization to correctly assign electron counts in ions, and employs HF baselines to stabilize and accelerate training. Empirical results on H2 and LiH+ show FermiNet energies closely tracking CCSD benchmarks and outperforming HF, illustrating practical gains in accuracy and convergence. The open-source integration aims to enable rapid, differentiable-physics–driven experimentation in quantum chemistry, with future plans to incorporate performance optimizations and physics priors from related models such as PauliNet.

Abstract

Finding accurate solutions to the electronic Schrödinger equation plays an important role in discovering important molecular and material energies and characteristics. Consequently, solving systems with large numbers of electrons has become increasingly important. Variational Monte Carlo (VMC) methods, especially those approximated through deep neural networks, are promising in this regard. In this paper, we aim to integrate one such model called the FermiNet, a post-Hartree-Fock (HF) Deep Neural Network (DNN) model, into a standard and widely used open source library, DeepChem. We also propose novel initialization techniques to overcome the difficulties associated with the assignment of excess or lack of electrons for ions.

Open-Source Fermionic Neural Networks with Ionic Charge Initialization

TL;DR

This work tackles accurate electronic structure calculations for molecules and ions using variational Monte Carlo with neural-network wavefunctions by integrating the FermiNet into the open-source DeepChem platform. It introduces a Mulliken-based ionic charge initialization to correctly assign electron counts in ions, and employs HF baselines to stabilize and accelerate training. Empirical results on H2 and LiH+ show FermiNet energies closely tracking CCSD benchmarks and outperforming HF, illustrating practical gains in accuracy and convergence. The open-source integration aims to enable rapid, differentiable-physics–driven experimentation in quantum chemistry, with future plans to incorporate performance optimizations and physics priors from related models such as PauliNet.

Abstract

Finding accurate solutions to the electronic Schrödinger equation plays an important role in discovering important molecular and material energies and characteristics. Consequently, solving systems with large numbers of electrons has become increasingly important. Variational Monte Carlo (VMC) methods, especially those approximated through deep neural networks, are promising in this regard. In this paper, we aim to integrate one such model called the FermiNet, a post-Hartree-Fock (HF) Deep Neural Network (DNN) model, into a standard and widely used open source library, DeepChem. We also propose novel initialization techniques to overcome the difficulties associated with the assignment of excess or lack of electrons for ions.
Paper Structure (10 sections, 6 equations, 1 figure, 1 table, 1 algorithm)

This paper contains 10 sections, 6 equations, 1 figure, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. DeepChem and Differentiable Physics
  3. Implementation
  4. Electron Sampler
  5. Hartree-Fock Baselines
  6. Ionic Charge Initialization
  7. FermiNet Model
  8. FermiNet Training
  9. Results
  10. Conclusion

Figures (1)

  • Figure 1: H2 ground state energy - FermiNet vs CCSD. Plotted using 21 different data points. The values obtained from FermiNet can be seen to closely match that of the values calculated via CCSD. Running the model with more MCMC steps and iterations can give more reliable results.