arXiv:2609.27072
Programmable Coherent Memory Kernels for Quantum Reservoir Computing in Waveguide QED
Hong Jiang, Yu Wu, Yue Ban, Xi Chen, Juan José García-Ripoll
Abstract
Reservoir computing requires not only long memory, but also memory distributed over the timescales relevant to a given task. In most quantum reservoir computing architectures, however, the memory profile emerges from intrinsic dynamics and cannot be directly engineered. Here we introduce a waveguide-QED quantum reservoir in which coherent memory return is controlled by the coupling geometry. The reservoir consists of a driven Kerr-nonlinear resonator coupled to a one-dimensional waveguide at spatially separated points. An unmeasured propagating field carries information about past inputs and coherently returns it to the resonator after a finite propagation time, forming a non-Markovian feedback channel without intermediate measurement or classical reinjection. Delay-resolved memory benchmarks reveal two complementary functions: waveguide propagation determines when past information returns, whereas Kerr dynamics convert the returned field into nonlinear temporal features accessible through a linear readout. NARMA-$n$ prediction shows that the optimal feedback delay shifts systematically with task order and approximately tracks the explicit input-product lag. These results establish coherent delayed feedback as a mechanism for engineering task-relevant memory in quantum reservoir computing.
AI Summary
The paper asks whether coherent delayed feedback can shape a quantum reservoir’s memory at task-relevant lags, rather than merely extend its duration. It studies a driven Kerr resonator coupled to a one-dimensional waveguide at spatially separated points. Eliminating the waveguide produces a memory kernel with geometry-set delayed terms; an unmeasured propagating field returns past-input information to the resonator. Numerical tests focus on a two-point, single-delay geometry, comparing a Kerr reservoir with $U=30\Gamma$ against a linear reservoir with the same feedback. Quantum dynamics are simulated using time-bin matrix-product states; quadrature expectations, temporally multiplexed within each input interval, feed a trained linear readout. Short-term-memory revivals occur near $d=\tau/\Delta t$ in both reservoirs. Delayed-product reconstruction develops a corresponding revival in the Kerr reservoir but is negligible in the linear reference, separating memory-return timing from nonlinear processing. For NARMA orders $n=5$–$10$, the optimal delay approximately follows $\tau_{\mathrm{opt}}/\Delta t=n-1$, shifting from four to nine input steps. Mean-field nonlinear controls give weaker delayed-product responses and higher prediction errors in the tested configurations. The study is numerical and concentrates on single-delay feedback, local Kerr nonlinearity, and first-order-moment readout. Multiple coupling points produce several linear-memory revivals but no clearly resolved delayed-product revivals in the tested regime. Mackey–Glass forecasting exhibits broader, horizon-dependent delay optima rather than a unique matching lag. Loss, dephasing, and imperfect feedback remain implementation considerations, and comparisons with classical reservoirs are restricted to the tested parameter and resource regimes.
Error found: The two-point classical reference equations use half the delayed-feedback coefficient implied by the stated general memory kernel, leaving the claimed same-kernel quantum–classical comparisons internally inconsistent. The reported linear-reservoir initial-state washout diagnostic is incompatible with the specified first-order readout because initial states |0⟩ and |1⟩ produce identical first moments under the same linear driven dynamics.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.