One More Time: Revisiting Neural Quantum States from a Reinforcement Learning Perspective

2026-07-02 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Quantum Machine Learning · Depth: Expert, quick

Summary

Proximal Wavefunction Optimization (PWO) is introduced as a principled trust-region algorithm designed to enhance the optimization of autoregressive Neural Quantum States (NQS). This method addresses the shortcomings of existing techniques like Adam, which overlooks function space geometry, and stochastic reconfiguration, known for its high cost and numerical fragility. PWO reframes variational energy minimization as an advantage policy-gradient problem, clipping probability-ratio changes in the amplitude channel and phase increments in the phase channel. It avoids explicit matrix inversion and reuses samples, combining scalability with theoretical guarantees. Benchmarked against Adam, minSR, and SPRING, PWO demonstrated improved stability and wall-clock convergence across Ising and frustrated J1-J2 one- and two-dimensional spin systems. Notably, the research fine-tuned a 1.5B-parameter RWKV-7 model, achieving NQS optimization at a scale over three orders of magnitude beyond previous efforts.

Key takeaway

For Research Scientists and Machine Learning Engineers developing Neural Quantum States for quantum simulations, you should consider adopting Proximal Wavefunction Optimization (PWO). This algorithm provides a principled, scalable trust-region approach that significantly improves stability and convergence over traditional methods like Adam and stochastic reconfiguration. By enabling optimization of models with billions of parameters, PWO allows you to tackle previously intractable quantum many-body problems, accelerating your research into complex spin systems.

Key insights

Variational energy minimization for NQS can be optimized via a trust-region policy-gradient approach, improving stability and scale.

Principles

• NQS optimization benefits from policy-gradient framing.
• Trust-region methods stabilize complex wavefunction updates.

Method

PWO clips probability-ratio changes in the amplitude channel and phase increments in the phase channel, reusing samples across updates.

In practice

• Optimize autoregressive NQS models with PWO.
• Scale NQS training to billions of parameters.

Topics

• Neural Quantum States
• Reinforcement Learning
• Trust-Region Optimization
• Autoregressive Models
• Quantum Many-Body Physics
• PWO Algorithm

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Finding Stationary Points by Comparisons — A novel algorithm finds non-convex function stationary points using only value comparisons, with a quantum speedup.
• Quantum-inspired tensor networks in machine learning models — Tensor networks from quantum physics offer machine learning benefits in efficiency, explainability, and privacy.
• cuNNQS-SCI: A Fully GPU-Accelerated Framework for High-Performance Configuration Interaction Selection with Neural Network Quantum States — Fully GPU-accelerated quantum chemistry framework cuNNQS-SCI significantly boosts scalability and performance by eliminating CPU bottlenecks.
• Subject to: Laureano Escudero — Laureano Escudero's career exemplifies dedication to mathematical optimization and its real-world applications.
• Reusability report: Optimizing T count in general quantum circuits with AlphaTensor-Quantum — General agents trained on diverse quantum circuits can optimize T-count without retraining, outperforming specialized agents.
• QuantFPFlow: Quantum Amplitude Estimation for Fokker--Planck Policy Optimisation in Continuous Reinforcement Learning — Quantum amplitude estimation can quadratically speed up Fokker-Planck partition function computation in continuous reinforcement learning.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.