Mode Connectivity in Neural Quantum States

Vinicius Hernandes, Thomas Spriggs, Eliska Greplova

Keywords: Mode connectivity, Neural Quantum States, Physics, Variational Monte Carlo

Abstract: We investigate the phenomenon of mode connectivity in Neural Quantum States, a neural network-based approximation to quantum wavefunctions. By analyzing the energy landscape of networks trained to fit the ground-state of the transverse-field Ising model, we find that minima corresponding to the solutions in different physical phases exhibit distinct connectivity properties. Notably, we observe an asymmetry across the quantum phase transition: models trained in the disordered phase are linearly connected to a wider range of models than those trained in the ordered phase, suggesting a connection between the energy landscape geometry and underlying physical phenomena.

Serve As Reviewer: ~Vinicius_Hernandes1

Submission Number: 28