The Role of Entanglement in Quantum Reservoir Computing with Coupled Kerr Nonlinear Oscillators

Quantum Reservoir Computing (QRC) uses quantum dynamics to efficiently process temporal data. In this work, we investigate a QRC framework based on two coupled Kerr nonlinear oscillators, a system well-suited for time-series prediction tasks due to its complex nonlinear interactions and potentially high-dimensional state space. We explore how its performance in forecasting both linear and nonlinear time-series depends on key physical parameters: input drive strength, Kerr nonlinearity, and oscillator coupling, and analyze the role of entanglement in improving the reservoir’s computational performance, focusing on its effect on predicting non-trivial time series. Using logarithmic negativity to quantify entanglement and normalized root mean square error (NRMSE) to evaluate predictive accuracy, our results suggest that entanglement provides a computational advantage on average – up to a threshold in the input frequency – that persists under some levels of dissipation and dephasing. In particular, we find that higher dissipation rates can enhance performance. While the entanglement advantage manifests as improvements in both average and worst-case performance, it does not lead to improvements in the best-case error. These findings contribute to the broader understanding of quantum reservoirs for high performance, efficient quantum machine learning and time-series forecasting.

💡 Research Summary

In this paper the authors investigate a quantum reservoir computing (QRC) architecture built from two coupled Kerr‑nonlinear oscillators and assess how genuine quantum entanglement influences the system’s ability to forecast both linear and nonlinear time‑series. The physical model consists of two bosonic modes a and b described by a Hamiltonian H(t)=H_nl+H_int+H_drive. The nonlinear part H_nl contains Kerr terms K_a N_a^2 and K_b N_b^2, the interaction term H_int = g(a b†+a† b) provides coherent hopping, and the drive term injects the external signal with amplitude ε (the same for both modes) and dissipation rates κ_a, κ_b. Open‑system dynamics are captured by a Lindblad master equation that includes photon loss (collapse operators √2κ a, √2κ b) and pure dephasing (√κ_φ N_a, √κ_φ N_b). The Hilbert space is truncated at N_cut=3, ensuring that mean occupations stay below one photon to avoid truncation artefacts.

Input data are constructed as a sum of 20 sinusoidal components with frequencies evenly spaced between f/5000 and f/50, where f is a controllable frequency scale. The normalized signal (range 0–1) is fed into the reservoir by setting the drive amplitudes ε_a=ε_b=ε_0 X_i at each time step. The system is initialized in the vacuum, driven for a time step δt=100, and then measured in the Fock basis. Time‑multiplexing creates m=10 virtual nodes per physical step, yielding a high‑dimensional feature vector x_k composed of occupation probabilities. A linear ridge regression (with Tikhonov regularisation) maps x_k to a target y_k that is the original signal delayed by Δ=10 steps. Training uses the first half of the sequence; testing uses the second half, both after a wash‑out period to reach steady state. Performance is quantified by the normalized root‑mean‑square error (NRMSE), while bipartite entanglement between the two modes is quantified by logarithmic negativity E_N = log₂‖ρ^{Γ_A}‖₁.

The authors perform extensive parameter sweeps. Varying the input drive strength ε (30 random values in