Building Neural Quantum States for Frustrated Spin Systems

The intersection of many-body physics and deep learning has opened a new frontier: Neural Quantum States (NQS). Traditional methods struggle with high-dimensional frustrated systems, while the global attention mechanism of Transformers provides a powerful tool for capturing complex quantum correlations.

In this guide, we implement a research-grade Variational Monte Carlo (VMC) pipeline using NetKet and JAX to solve the frustrated J1–J2 Heisenberg spin chain. We will build a custom Transformer-based NQS architecture, optimize the wavefunction using Stochastic Reconfiguration, and benchmark our results against exact diagonalization.

By the end of this guide, you will have a scalable, physically grounded simulation framework capable of exploring quantum magnetism beyond the reach of classical exact methods. We install all required libraries and configure JAX for stable high-precision computation, defining the J1–J2 frustrated Heisenberg Hamiltonian using a custom colored graph representation.

We implement a Transformer-based neural quantum state using Flax. We encode spin configurations into embeddings, apply multi-layer self-attention blocks, and aggregate global information through pooling. The output is a complex log-amplitude, allowing our model to represent highly expressive many-body wavefunctions.

We sweep across multiple J2 values to explore the frustrated phase diagram. We train a separate variational state for each coupling strength and record the final energy, computing the structure factor peak for each point to detect possible ordering transitions.