Driving enhanced exciton transfer by automatic differentiation

Emanuele Ballarin, Diana A. Chisholm, A. Smirne, M. Paternostro, Fabio Anselmi, S. DonadiMach. Learn. Sci. Technol 2025 · DOI

TL;DR

This paper introduces a novel method for optimizing exciton transfer in complex quantum networks by using automatic differentiation to tune external driving fields. The approach significantly enhances energy transfer efficiency, especially under off-resonant conditions, offering a practical and computationally agile pathway for designing efficient quantum systems.

What Problem Does This Paper Address?

Efficiently transferring energy, like excitons in photosynthetic complexes or excitations in quantum networks, is crucial for many natural and technological processes. While quantum mechanics offers potential for highly efficient transfer, controlling these complex systems to achieve optimal performance is a significant challenge. Traditional methods for optimizing quantum systems often involve complex control strategies that are difficult to implement experimentally or scale poorly with system size. This paper tackles the problem of enhancing exciton transfer efficiency in various quantum network architectures by exploring a more accessible and computationally efficient control mechanism: external driving fields.

Methods

The authors model exciton transfer using a simplified system comprising a radiation field, an antenna (a qubit), a network of interconnected sites (also qubits), and a sink. They employ a Lindblad Master Equation to describe the system's evolution. The core of their methodology lies in using automatic differentiation, specifically through the PyTorch library, coupled with the Adam optimizer. This allows them to efficiently compute gradients and optimize parameters. They explore three optimization strategies: tuning network energies, adjusting couplings between components, and, most importantly, optimizing external driving fields applied to the antenna and the sink. The effectiveness of these strategies is evaluated by maximizing the time-integrated probability of the excitation reaching the sink.

Datasets

Not discussed in the paper. The paper focuses on a theoretical model and simulations rather than training on empirical datasets.

Key Insights and Ideas

  • Driving Fields as a Powerful Control Knob: The most significant finding is that optimizing external driving fields is remarkably effective in enhancing exciton transfer across various network types and sizes. This strategy consistently outperforms optimizing network energies or couplings, especially in off-resonant scenarios.
  • Off-Resonance Advantage: The benefits of optimization, particularly through driving fields, become dramatically more pronounced when the incident radiation is off-resonant with the system. This suggests that driving fields can "bridge the gap" and facilitate efficient energy transfer even when direct resonance is absent.
  • Computational Agility: Automatic differentiation provides a computationally efficient way to optimize complex quantum systems. This approach scales better than many traditional methods, making it a valuable tool for future quantum engineering.
  • Simplicity of Driving: Surprisingly, a small number of driving terms (a low value of 'R' in their model) is often sufficient to achieve near-optimal results, indicating that complex driving patterns are not always necessary.
  • Noise-Assisted Transport Phenomenon: The paper observes an interesting phenomenon where increasing noise levels in an unoptimized system can sometimes lead to higher integrated sink probability, hinting at noise-assisted transport.

Key Results

  • Order-of-Magnitude Improvement: In off-resonant cases, optimizing driving fields led to an increase in excitation transfer efficiency by more than an order of magnitude compared to the undriven case.
  • Consistent Effectiveness: The driving optimization strategy proved effective across different network architectures, including nearest-neighbor, star, and the more complex Fenna-Matthews-Olson (FMO) complex.
  • Coupling Optimization Limitations: While optimizing couplings was effective for simpler networks (NN, SN), it was less so for the FMO complex, suggesting that driving fields offer a more universally applicable control mechanism for complex systems.
  • Energy Optimization Ineffectiveness: Optimizing network energies generally showed the least improvement across all tested scenarios.

Limitations and Future Work

  • Understanding Coupling Optimization: The paper acknowledges that a deeper understanding of why learning couplings is less efficient than learning optimal drivings is needed and will be a focus of future research.
  • Experimental Feasibility: While driving local energies might be more experimentally feasible than manipulating couplings, the paper suggests that further investigation into the practical implementation of these strategies is warranted.
  • Model Simplification: The model used is a simplification of real-world systems. Future work could explore more complex network structures, environmental interactions, and non-linear effects.
  • Generalizability: While the results are promising across various network types, further testing on even more complex and realistic quantum systems would be beneficial.

Summary updated

BibTeX
@article{ballarin2025driving,
  author = {Emanuele Ballarin and Diana A. Chisholm and A. Smirne and M. Paternostro and Fabio Anselmi and S. Donadi},
  title = {{Driving enhanced exciton transfer by automatic differentiation}},
  journal = {Mach. Learn. Sci. Technol.},
  year = {2025},
  doi = {10.1088/2632-2153/add23b},
}
Figure