Quantum-informed machine learning for predicting spatiotemporal chaos with practical quantum advantage.

Area of Science:

• Computational physics
• Quantum machine learning
• Fluid dynamics

Background:

• Modeling high-dimensional chaotic systems is computationally challenging.
• Classical methods struggle with long-term prediction accuracy and fine-scale dynamics.
• Integrating quantum computing principles can enhance machine learning models.

Purpose of the Study:

• Introduce a quantum-informed machine learning (QIML) framework.
• Improve long-term modeling of high-dimensional chaotic systems.
• Leverage quantum generative models for enhanced predictive capabilities.

Main Methods:

• Developed a QIML framework combining a quantum generative model and a classical autoregressive predictor.
• The quantum model learns a quantum prior (Q-Prior) to guide small-scale interactions.
• Evaluated QIML on Kuramoto-Sivashinsky, 2D Kolmogorov flow, and 3D turbulent channel flow.

Main Results:

• QIML improved predictive distribution accuracy by up to 17.25% and full-spectrum fidelity by up to 29.36% over classical baselines.
• For turbulent channel inflow, QIML produced stable, physically consistent forecasts outperforming PDE solvers.
• The Q-Prior, trained on a quantum processor, was essential for prediction stability and accuracy.

Conclusions:

• QIML framework effectively models long-term behavior of chaotic systems.
• Quantum priors enhance predictive accuracy and stability in complex simulations.
• QIML offers significant memory compression and scalable integration of quantum resources for scientific modeling.