Toward quantum-aware machine learning: Improved prediction of quantum dissipative dynamics via complex valued neural...

Area of Science:

• Quantum Physics
• Computational Science
• Machine Learning

Background:

• Modeling quantum dissipative dynamics is complex due to environmental factors and memory effects.
• Real-valued neural networks (RVNNs) used in machine learning often fail to capture the complex-valued nature of quantum mechanics, leading to inconsistencies.
• Existing methods struggle with preserving amplitude-phase correlations crucial for quantum state fidelity.

Purpose of the Study:

• Introduce complex-valued neural networks (CVNNs) as a novel, physics-consistent framework for learning quantum dissipative dynamics.
• Address the limitations of RVNNs in accurately representing quantum mechanical principles.
• Develop a scalable and robust classical simulation method for open quantum systems.

Main Methods:

• Developed and implemented complex-valued neural networks (CVNNs) that operate directly on complex-valued quantum data.
• Preserved the inherent algebraic structure and coherence of quantum states within the neural network framework.
• Benchmarked CVNNs against RVNNs using the spin-boson model and Fenna-Matthews-Olson complex.

Main Results:

• CVNNs demonstrated superior convergence speed and training stability compared to RVNNs.
• Achieved significantly improved physical fidelity, including better trace conservation and Hermiticity.
• Performance advantages of CVNNs increased with system size and complexity of quantum coherence.

Conclusions:

• CVNNs provide a robust, scalable, and quantum-aware classical approach for simulating open quantum systems.
• This method is particularly valuable in the pre-fault-tolerant quantum era.
• CVNNs offer a promising direction for advancing quantum dynamics simulations.