arXiv:2609.23429
Liouvillian Response for Temporal Information in Quantum Reservoir Computing
Jiande Cao, Rui-Yang Gong, Zhongjin Lin, Yexiong Zeng, Ze-Liang Xiang
Abstract
Open quantum systems offer a physical substrate for temporal information processing in quantum reservoir computing, yet the microscopic mechanisms linking their dynamics to computational performance remain unclear. Here we establish a microscopic response-to-performance framework that connects Liouvillian dynamics directly to task performance. We decompose observable Volterra weights as $W^{(q)}=BH^{(q)}$, separating internal input-history pathways from their readout visibility, and show that a dynamical response contributes to a task only if it is input-generated, readout-visible, target-correlated, and resolvable above the regularization scale. This framework reveals how Liouvillian modes govern the retention and propagation of task-relevant information, while symmetries determine which components remain visible to the readout. It further establishes how distinct covariance modes encode different temporal information structures and how system parameters and processing protocols can reshape their contributions to improve QRC performance. Thermalization drives the reservoir from processing that retains input history to a response dominated by the most recent input before regularization suppresses the residual information. We finally show that weak measurement can activate task-relevant covariance modes by lifting symmetry-imposed channel equivalence, whereas delayed feedback creates controlled return pathways for historical information. These results establish a microscopic connection between open-system dynamics and computational performance, providing principles for engineering task-relevant information flow in quantum reservoirs.
AI Summary
The paper asks how microscopic open-system dynamics determine task performance in quantum reservoir computing, beyond memory lifetime or prediction error alone. It develops a Liouvillian–Volterra description in which observable response weights factor as $W^{(q)}=BH^{(q)}$: $H^{(q)}$ describes input injection, mode conversion and historical propagation, while $B$ projects onto measured observables. Contracting these weights with input–target statistics gives feature covariances, target correlations and ridge-regression performance. Analytical derivations and density-matrix simulations use four-spin interacting reservoirs and a five-spin multipoint-waveguide reservoir, primarily with Mackey–Glass prediction tasks. Under assumed response-sector decay, fixed-order weights obey $|W_\mu^{(q)}(\mathbf m)|\leq A_q(1+m_q)^{q-1}e^{-\alpha_0m_q\Delta t}$. The analysis distinguishes observable initial-state persistence from readout mismatch; parity symmetry eliminates all odd-order responses in the specified $\sigma_z$ readout. Reconstructions through third order reproduce prediction errors with approximately $0.67\%$ mean relative discrepancy across an interaction scan and at most $1.09\%$ across a dephasing scan. Thermalization progressively removes historical response while leaving a latest-input predictor supported by signal correlations; subsequent degradation reflects covariance suppression relative to regularization. Input-removal diagnostics separate these contributions. Spatially inhomogeneous weak measurement activates additional covariance directions by breaking channel equivalence, whereas delayed feedback introduces recursively returning response pathways. The stated scope is conditional: history convergence is established at each fixed Volterra order under decay assumptions, not for the full infinite-order series. Kernel reuse requires inputs within the same response regime; performance formulas assume stationary training and evaluation statistics. Numerical validation concerns the small reservoirs and parameter regimes examined.
Error found: The claimed implication from contraction of the driven observable propagator to exponential decay of zero-input Volterra kernels is unsupported; the appendix instead assumes the required kernel bound separately. The exact all-order response-to-performance construction requires convergence beyond the fixed-order history convergence proved in the appendix, while numerical validation establishes accurate third-order reconstruction only in the tested regimes. The covariance-based NRMSE and memory formulas give population ridge performance, not generally the exact independent-test performance of the finite-sample trained estimator.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 43 pages, 11 figures, comments welcome