Characterize optimization dynamics for finite-parameter neural quantum states

Investigate and resolve the outstanding questions about the optimization landscape and training dynamics of neural-network quantum state models with a finite number of parameters, going beyond the infinite-width neural tangent kernel regime to understand practical implementations.

Background

The paper highlights that our understanding of the optimization landscape for neural quantum states is incomplete. Although recent work has analyzed the infinite-width limit using neural tangent kernel techniques, practical models have finite parameter counts.

The authors explicitly state that open questions persist in the finite-parameter regime, emphasizing the need for theoretical and empirical understanding that can guide real-world optimization strategies.