Connection between memory performance and optical absorption in quantum reservoir computing

Technical Summary: Connection between Memory Performance and Optical Absorption in Quantum Reservoir Computing

Problem Statement
Quantum reservoir computing (QRC) has emerged as a promising paradigm for machine learning on noisy intermediate-scale quantum (NISQ) devices, leveraging quantum properties such as entanglement and exponentially scaling phase spaces. However, the evaluation of QRC performance has historically relied on information-theoretic benchmarks, such as Short-Term Memory Capacity (STMC). While effective for quantifying performance, these metrics fail to provide insights into the underlying physical mechanisms governing quantum neural networks. Consequently, there is a lack of systematic understanding regarding how concrete physical properties—specifically dissipation and absorption—govern the fundamental capabilities and practical optimization of these systems. This gap hinders the targeted engineering of quantum reservoirs for specific tasks.

Methodology
The authors investigate a fully-connected transverse-field Ising model (TFIM) as the physical platform for computation. The system consists of $N$ qubits with random coupling matrix elements $J_{ij}$ scaled to a spectral radius $J_0 = 0.5$.

• Input Encoding: Unlike previous approaches that utilize "erase-and-write" maps (which introduce fixed dissipation), this study employs a coherent light source coupled resonantly to the qubits. The input signal $s(t)$ is encoded via an amplitude-modulated coherent drive, modeled effectively by a time-dependent $\sigma_y$ term in the Hamiltonian. This allows for complete control over dissipation without inherent injection losses.
• Dissipation Modeling: The system dynamics are governed by the Gorini–Kossakowski–Sudarshan–Lindblad (GKSL) equation, incorporating unitary evolution and qubit decay (rate $\gamma$). The steady state is calculated by reformulating the dynamics in Fock-Liouville space to solve for the null space of the Liouvillian superoperator.
• Performance Metrics: The system's memory performance is quantified using Short-Term Memory Capacity (STMC). This involves training a linear readout layer to reproduce Legendre polynomials of past inputs. The capacity $C_i$ is defined as the sum of squared Pearson correlation coefficients between the reservoir output and target sequences over a range of delays, truncated at a noise threshold determined by random target correlations.
• Physical Property Analysis: The authors calculate the optical absorption spectrum $\alpha_{s,\gamma}(\omega)$ using linear response theory. Specifically, they analyze the resonant absorption $\alpha_{s,\gamma}(0)$, derived from the Fourier spectrum of the dipole correlation function, averaged over various signal strengths $s$.

Key Results

1. Dissipation Regimes: The study identifies three distinct regimes based on the qubit decay rate $\gamma$:
   • Low Dissipation ($\gamma \lesssim 10^{-2}$): The system evolves nearly unitarily, failing to fade memory, resulting in negligible STMC.
   • High Dissipation ($\gamma \gtrsim 10^{2}$): The system relaxes too quickly to the steady state, erasing input information immediately, also resulting in zero STMC.
   • Intermediate Dissipation: A "sweet spot" exists (approximately $10^{-1} < \gamma < 10^{1}$) where the system reaches a new steady state distinct from the zero-input state without erasing information too rapidly, yielding optimal memory capacity.
2. Correlation with Optical Absorption: The authors establish a quantitative link between the STMC and the resonant optical absorption of the quantum reservoir. The curve of average optical absorption $\bar{\alpha}_\gamma$ as a function of decay rate $\gamma$ exhibits a striking qualitative similarity to the STMC curve. Both metrics peak at the same intermediate dissipation level and vanish at the extremes of very weak and very strong dissipation.
3. Robustness: This correlation holds across different coupling strengths and system topologies, as verified by additional simulations of a 4-qubit reservoir.

Key Contributions

• Physical Explanation of the "Sweet Spot": The paper provides a physical explanation for the previously observed "sweet-spot" behavior in QRC performance as a function of dissipation. It demonstrates that optimal memory performance aligns directly with maximal optical absorption.
• Bridging Domains: The work bridges the gap between quantum information theory (STMC) and experimentally accessible physical properties (optical absorption). It moves beyond abstract metrics to link computational capability with tangible, measurable system responses.
• Engineering Guidance: By identifying optical absorption as a proxy for memory performance, the authors suggest that experimentalists can optimize QRC systems by tuning physical parameters (such as dissipation rates or coupling strengths) to maximize absorption, rather than relying solely on iterative information-theoretic benchmarking.

Significance
The paper claims that understanding physical properties, rather than just abstract information measures, is crucial for guiding quantum technology development. The established connection offers a pathway for the targeted engineering of quantum reservoir computers optimized for specific tasks. The authors note that these findings are relevant to various physical systems proposed for QRC, including photonic systems, nonlinear oscillators, and specifically arrays of Rydberg atoms, which naturally implement the TFIM and have demonstrated remarkable scaling in recent experiments. The results aim to foster the development of new performance metrics grounded in both physics and information theory.