Nikolaos Petropoulos and Elena Blokhina from University College Dublin and Equal1 Labs report that a spin-qubit reservoir computing system unexpectedly increases prediction accuracy at a coherence time of 1 μs, even as its overall memory capacity diminishes. Their simulations of a six-qubit spin chain reveal that combining single- and two-qubit measurements outperforms simpler readout methods, exceeding the performance of a system without decoherence at coherence times of 1 μs and 5-10 μs. The work demonstrates that the impact of coherence on this quantum reservoir depends heavily on the specific forecasting task, with prediction horizons shortening across ten simulations for chaotic forecasting at the 1 μs mark.
Spin-Qubit Reservoir: Six-Qubit Chain and Parameters
Researchers at Equal1 Labs and University College Dublin investigated how dephasing affects learning within this simulated system, designed to mimic electron-spin qubits realized in silicon-silicon germanium heterostructures. The simulated reservoir, a chain of six qubits coupled by Heisenberg exchange, was subjected to Markovian dissipation and exact expectation-value readout to model realistic operating conditions. The study focused on understanding how coherence, the duration for which quantum information is preserved, impacts the reservoir’s ability to learn and predict complex signals.
Impact of Decoherence on NARMA Prediction Accuracy
Open-loop NARMA prediction accuracy diminishes as coherence time decreases, with the metric MC10 falling from 0.49 to between 0.26 and 0.33 at a T2 value of 5-10 μs. This suggests the observed gains in prediction are not simply due to improved recall of input signals. The ability to accurately recall the input u(k-n), which enters the recurrence for yk, does not improve linearly with coherence, indicating a more complex relationship between decoherence and predictive performance.
This finding challenges the intuitive expectation that memory loss would hinder accuracy and points to a detailed role for dephasing in reservoir computing. Combining single- and two-qubit observables gives better predictions than a population-only readout at comparable polynomial feature counts.
Coherence Times and Memory Capacity Relationship
Measurements of memory capacity reveal that dephasing does not simply erase information, but rather redistributes it within the spin-qubit reservoir. At coherence times of 5 to 10 μs, total memory capacity remains remarkably high, between 0.95 and 0.99 of the value achieved in a system without decoherence, and extends to longer delays.
This suggests that gains in predictive accuracy at these coherence levels are not solely due to preserving more information, but to how that information is organized and accessed. The researchers measured this capacity using a standard construction, fitting each delay and recording a metric, MCd = 1-NMSEd, to quantify memory retention.
Observable Selection: Single vs. Two-Qubit Readouts
Researchers contrasted a full observable set generating 1710 features against a Z-only readout of degree 7, which produced 1715 features, ensuring a comparable regressor count for a fair comparison. The simulations consistently demonstrated that the full set outperformed the population-only approach across all 27 cells of a tested grid on two independent reservoir instantiations. This improved performance occurs despite a drop in total memory capacity at the 1 microsecond coherence time, a counterintuitive result given the expectation that diminished memory would hinder predictive ability.
Further analysis revealed a minimal difference between degree 6 and degree 7 Z-only readouts, with a median absolute normalized mean squared error (NMSE) difference of 8 x 10-5 across 54 matched cells, indicating the robustness of the findings. A quadratic readout on the full observable set at 10 microseconds consistently outperformed the Z-only approach across these different NARMA orders. A quadratic readout on this full set contains products of two-body expectation values, offering a richer representation of the quantum state than the Z-only measurement.
Limit-Cycle Persistence Independent of Reservoir Coherence
Prediction horizons for chaotic signals shorten consistently as coherence decreases to 1 μs within the simulated spin-qubit reservoir, a finding observed across ten independent model realizations. For autonomous generation, the readout was trained with scheduled sampling and iterative dataset aggregation to expose it to states reached under its own feedback. The team did not find a systematic dependence of limit-cycle persistence on coherence across the reservoir ensemble.
Chaotic Forecasting Performance with Reduced Coherence
The capacity of a simulated spin-qubit reservoir to predict chaotic signals diminishes consistently across ten independent realizations when coherence drops to 1 μs, a finding that challenges expectations about the role of quantum memory. While total memory capacity decreases at this timescale, the model still retains a degree of functionality, prompting investigation into how coherence impacts different predictive tasks.
This suggests that the impact of decoherence is not uniform, but instead alters how information is stored and retrieved within the reservoir. The observed redistribution of memory, where capacity remains relatively constant but is shifted across delays, is a key finding.
Noise-Driven Strange Attractor Response to Coherence
This outcome, detailed in recent work, demonstrates a specific limitation of the reservoir’s predictive capacity when confronted with this type of dynamical system, even with varying degrees of quantum coherence. This indicates that, for chaotic systems, reduced coherence demonstrably hinders the reservoir’s ability to extend its predictive reach, a finding that contrasts with some observations regarding non-chaotic signals. This detailed relationship suggests that optimizing quantum reservoir computing systems requires careful consideration of the specific application and the inherent characteristics of the data being processed.
👉 More information
🗞 To predict or to persist: how decoherence shapes learning in a spin-qubit quantum reservoir
✍️ Nikolaos Petropoulos and Elena Blokhina
🧠 ArXiv: https://arxiv.org/abs/2611.26215v1




See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.
