Infinite Neural Network Quantum States: Entanglement and Training Dynamics

Di Luo,James Halverson

DOI: https://doi.org/10.1088/2632-2153/ace02f

2023-09-28

Abstract:We study infinite limits of neural network quantum states ($\infty$-NNQS), which exhibit representation power through ensemble statistics, and also tractable gradient descent dynamics. Ensemble averages of Renyi entropies are expressed in terms of neural network correlators, and architectures that exhibit volume-law entanglement are presented. A general framework is developed for studying the gradient descent dynamics of neural network quantum states (NNQS), using a quantum state neural tangent kernel (QS-NTK). For $\infty$-NNQS the training dynamics is simplified, since the QS-NTK becomes deterministic and constant. An analytic solution is derived for quantum state supervised learning, which allows an $\infty$-NNQS to recover any target wavefunction. Numerical experiments on finite and infinite NNQS in the transverse field Ising model and Fermi Hubbard model demonstrate excellent agreement with theory. $\infty$-NNQS opens up new opportunities for studying entanglement and training dynamics in other physics applications, such as in finding ground states.

Quantum Physics,Disordered Systems and Neural Networks,Machine Learning,High Energy Physics - Theory

What problem does this paper attempt to address?

The paper aims to address the following key issues:

1. **Entanglement properties of Neural Network Quantum States (NNQS)**: Investigate the quantum state neural networks (∞-NNQS) in the infinite limit, exploring their statistical properties and traceable gradient descent dynamics. Specifically, the paper demonstrates how to express the ensemble average of entanglement entropy through neural network correlation functions and proposes architectures that exhibit volume law entanglement.

2. **Analytical understanding of gradient descent dynamics**: Develop a general framework to study the gradient descent dynamics of Neural Network Quantum States (NNQS) using the Quantum State Neural Tangent Kernel (QS-NTK). For infinitely wide NNQS, in certain limits, the training dynamics become simplified and analytically solvable.

3. **Target wave function recovery in supervised learning**: Prove that in quantum state supervised learning, training infinitely wide NNQS with a positive definite QS-NTK can accurately recover any target wave function. This has important applications for initializing quantum states for ground state and real-time simulations and helps in understanding the representational capacity of neural network architectures.

By combining theoretical analysis with numerical experiments, the paper demonstrates that these techniques can accurately predict the training dynamics of finite-width NNQS ensembles, and as the network width increases, the performance approaches the infinite width limit. This provides theoretical insights and practical guidance for understanding and optimizing the application of neural network quantum states in other physical contexts.