Efficient Convex Optimization for Bosonic State Tomography

Shengyong Li, Youling Yue, Ying Hu, Rui-Yang Gong, Qianchuan Zhao, Zhihui Peng, Pengtao Song, Ze-Liang Xiang, Jing Zhang, Zhang, JingarXiv 2025 · DOI · arXiv

Abstract

Quantum states encoded in electromagnetic fields, also known as bosonic states, have been widely applied in quantum sensing, quantum communication, and quantum error correction. Accurate characterization is therefore essential yet difficult when states cannot be reconstructed with sparse Pauli measurements. Tomography must work with dense measurement bases, high-dimensional Hilbert spaces, and often sample-based data. However, existing convex optimization-based techniques are not efficient enough and scale poorly when extended to large and multi-mode systems. In this work, we explore convex optimization as an effective framework to address problems in bosonic state tomography, introducing three techniques to enhance efficiency and scalability: efficient displacement operator computation, Hilbert space truncation, and stochastic convex optimization, which mitigate common limitations of existing approaches. Then we propose a sample-based, convex maximum-likelihood estimation (MLE) method specifically designed for flying mode tomography. Numerical simulations of flying four-mode and nine-mode problems demonstrate the accuracy and practicality of our methods. This method provides practical tools for reliable bosonic mode quantum state reconstruction in high-dimensional and multi-mode systems.

Introduction

Bosonic states, encoded in the continuous variables of electromagnetic fields, are fundamental to modern quantum technologies. Reconstructing these states through Quantum State Tomography (QST) is critical for verifying quantum operations. However, bosonic QST faces the "curse of dimensionality." Unlike discrete qubits, bosonic systems inhabit infinite-dimensional Hilbert spaces. As the number of modes increases, the computational complexity of traditional reconstruction methods—often relying on discretized histograms of phase-space data—becomes intractable. For instance, a 4-mode system would require a grid with $10^{12}$ cells, far exceeding typical memory capacities.

Methodology

To overcome these bottlenecks, this work introduces a framework centered on efficient convex optimization.

Efficient Displacement Operator Computation (EDOC)

A core operation in bosonic QST is the calculation of the displacement operator $D(\alpha)$. Standard approaches using matrix exponentials (e.g., Padé approximation) are computationally expensive. The proposed EDOC method leverages the tridiagonal structure of the operators in a specific basis. By transforming the problem into a vector-based calculation using shared eigenvector properties, the computational cost is significantly reduced.

Time cost and acceleration ratio of the EDOC method compared to standard matrix exponential calculations.

Hilbert Space Truncation and Stochastic Optimization

To handle high-dimensional matrices, a systematic Hilbert Space Truncation (HST) strategy is employed. This involves a two-step projection that eliminates redundant dimensions during matrix operations, optimizing trace calculations. Furthermore, to manage large datasets without memory overflow, the framework utilizes Stochastic Projected Gradient Descent (SPGD) and its accelerated variant, ASSG-r. These algorithms process data in "mini-batches," ensuring convergence to a globally optimal, physically valid density matrix while maintaining a low memory footprint.

Sample-based Maximum Likelihood Estimation

Rather than relying on discretized histograms, the authors propose a sample-based Maximum Likelihood Estimation (MLE) method. This approach treats each measurement as a discrete point in phase space, bypassing the exponential scaling of grid-based methods and enabling the analysis of multi-mode "flying" states.

Results and Discussion

Numerical simulations were conducted on various benchmark states, including single-mode cat states, two-mode binomial Bell states, and multi-mode W states.

Speed and Scalability

The EDOC method achieved an average acceleration ratio of approximately 20 compared to traditional methods. For a single-mode system with dimension $N=100$, reconstruction was completed in under 10 seconds. Most notably, the framework successfully reconstructed a 9-mode flying W state with a fidelity exceeding 98%, demonstrating unprecedented scalability for convex-based bosonic tomography.

Performance benchmarks for multi-mode bosonic state reconstruction, showing high fidelity across increasing dimensions and modes, as well as robustness to noise.

Robustness to Noise

The method's resilience was tested against additive Gaussian and thermal noise. Even with noise levels at 25% of the maximum signal amplitude, the reconstruction fidelity remained above 98% for complex multi-mode configurations. The stochastic approach also proved memory-efficient, reconstructing 2-mode systems using approximately 10 GB of RAM.

Conclusion

This work provides a robust and efficient framework for bosonic state tomography in high-dimensional and multi-mode systems. By integrating efficient operator computation with stochastic convex optimization and sample-based MLE, the researchers have bypassed traditional scaling limits. Future work will focus on validating these methods with experimental data from superconducting circuits and exploring more complex non-Gaussian noise models.

Summary updated

Abstract

Quantum states encoded in electromagnetic fields, also known as bosonic states, have been widely applied in quantum sensing, quantum communication, and quantum error correction. Accurate characterization is therefore essential yet difficult when states cannot be reconstructed with sparse Pauli measurements. Tomography must work with dense measurement bases, high-dimensional Hilbert spaces, and often sample-based data. However, existing convex optimization-based techniques are not efficient enough and scale poorly when extended to large and multi-mode systems. In this work, we explore convex optimization as an effective framework to address problems in bosonic state tomography, introducing three techniques to enhance efficiency and scalability: efficient displacement operator computation, Hilbert space truncation, and stochastic convex optimization, which mitigate common limitations of existing approaches. Then we propose a sample-based, convex maximum-likelihood estimation (MLE) method specifically designed for flying mode tomography. Numerical simulations of flying four-mode and nine-mode problems demonstrate the accuracy and practicality of our methods. This method provides practical tools for reliable bosonic mode quantum state reconstruction in high-dimensional and multi-mode systems.
BibTeX
@misc{li2025efficient,
  author = {Shengyong Li and Youling Yue and Ying Hu and Rui-Yang Gong and Qianchuan Zhao and Zhihui Peng and Pengtao Song and Ze-Liang Xiang and Jing Zhang and Zhang, Jing},
  title = {{Efficient Convex Optimization for Bosonic State Tomography}},
  year = {2025},
  eprint = {2509.06305},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2509.06305},
  howpublished = {arXiv},
}
Figure