arXiv:2610.05035
Process tensor simulations of non-Markovian open quantum systems with large uncoupled subspaces
Thomas K. Bracht, Moritz Cygorek
Abstract
Process tensors in matrix product operator representation (PT-MPOs) are a powerful tool to calculate the temporal evolution and multi-time correlations of open quantum systems. While much progress has been made in making the calculation of process tensors for larger system dimensions possible, the numerical cost of the subsequent propagation of the system typically scales as $\mathcal{O}(d^6)$, which renders PT-MPO methods prohibitively expensive for system dimensions above several dozens. Here, we show how PT-MPO methods can be used efficiently when only some of the system states couple to the environment, and how combination with a Taylor expansion method allows for a reduction of the numerical complexity down to $\mathcal{O}(d^3)$. We show that these methods enable efficient simulations of systems with several hundreds of states. This is particularly relevant for solid-state cavity-QED systems with multiple modes, or when calculating frequency-filtered correlation functions by coupling sensor systems to the system of interest.
AI Summary
Process-tensor matrix-product operators (PT-MPOs) can encode an environment efficiently when it couples directly only to a small subsystem, but propagating the remaining system can still be expensive. This paper addresses that bottleneck for systems with large indirectly coupled subsystems and additional Lindblad dissipation. The method adapts an existing Taylor-series integrator for Lindblad equations to PT-MPO propagation. Instead of constructing and multiplying by a Liouville-space exponential, it recursively applies the system Liouvillian through Hilbert-space matrix multiplications, independently for each process-tensor bond index. For fixed Taylor order $N_T$, number of Lindblad channels $N_L$, and bond dimension $\chi$, system propagation costs $\mathcal{O}(N_T(N_L+1)\chi d^3)$, avoiding dense exponentiation costing $\mathcal{O}(d^6)$ and storage of a $d^4$-element propagator. Benchmarks use a phonon-coupled two-level system interacting with a truncated cavity mode. They exhibit cubic runtime scaling up to dimensions around $d=1000$, with speedups over direct exponentiation beginning near $d=6$. An accuracy test at $d=6$ finds errors approaching machine precision at Taylor order six. A frequency-filtered photon-correlation calculation with two sensor systems takes approximately two hours per data point at $d=24$; restricting to at most two excitations reduces this to twelve minutes at $d=12$. The approach assumes a precomputed PT-MPO and gains efficiency when the directly environment-coupled subsystem remains small. Taylor truncation introduces an additional convergence parameter, alongside existing time-discretization and PT-MPO compression parameters; the reported accuracy comparison concerns the small benchmark system.
Error found: The large-dimension benchmarks establish runtime but not accuracy: cavity loss makes the generator increasingly stiff as the cutoff grows, and the fixed Taylor orders used are not uniformly stable over the reported dimensions, while convergence is tested only at d=6.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 7 pages, 3 figures