High Accuracy Wavefunctions Using Deep-Learning-Based Variational Monte Carlo

Explore high-accuracy wavefunctions using deep-learning-based variational Monte Carlo in this conference talk by Michael Scherbela from the University of Vienna. Delve into a novel architecture for fermionic wavefunctions that achieves superior accuracy at reduced computational costs compared to previous approaches. Discover how this method calculates the most accurate variational ground state energies for atoms and small molecules to date. Examine the impact of physical prior knowledge on accuracy and learn about accelerating VMC-optimization when calculating solutions for multiple molecular geometries. Gain insights into embedding, invariants, envelope orbitals, and backflow factors, as well as their applications in studying potential energy surfaces, with a focus on ethylene as an example.

Introduction
Deep learningbased variational Monte Carlo
Deep neural networks
Embedding
Invariants
Envelope orbitals
Spin
Results
Comparison
Relative energies
Backflow factor
Pretraining
Potential energy surface
Architecture
Example ethylene
Conclusion