The Signal
Ghosh Team Finds Squeezed Sensors Lose Ground in Negative Noise
A study of correlated phase noise finds spin-squeezed states offer nearly optimal precision for positive correlations, while tensor-network-optimized strategies can offer an advantage for negative correlations. The researchers compared thes
Quantum.world Team · Written with AI · fact-checked ·
Tensor-network searches compared quantum sensing strategies under correlated phase noise, with calculations extending to about 30 channel uses. Spin-squeezed probes nearly reach the available precision for positively correlated fluctuations, while optimized protocols can outperform them when successive fluctuations are negatively correlated. The study by Srijon Ghosh, Arkadiusz Kobus, Stanisław Kurdziałek and Rafał Demkowicz-Dobrzański appeared in Quantum on October 28, 2026.
Noise correlation changes the sensor comparison
The paper models phase estimation across a sequence of quantum channels, each affected by a fluctuating phase. A correlation parameter distinguishes noise that tends to move successive phases in the same direction from noise that tends to move them in opposite directions; zero corresponds to independent fluctuations. The authors use a Gaussian model whose correlation strength can vary continuously, rather than treating noise only as a small set of discrete phase shifts.
That distinction matters because a sensing strategy suited to one noise pattern need not be best for another. The team compares one-axis-twisting spin-squeezed states, measured through collective spin observables, with parallel and adaptive protocols optimized using tensor-network numerical methods. They assess those strategies against upper bounds on the quantum Fisher information, a measure of how precisely a parameter can be estimated.
“We find that the spin-squeezed states offer practically optimal performance in the regime where phase fluctuations are positively correlated but can be outperformed by tensor-network-optimized strategies for negatively correlated fluctuations.”
The authors of the paper write
Negative correlations can favor optimized protocols
For positive correlations, the spin-squeezed protocol sits close to the study’s bounds, leaving limited room for another strategy to improve precision. The authors report that matrix-product-state probes outperform spin-squeezed states for both positive and negative correlations, though that advantage is expected to fade asymptotically for positive correlations. For negative correlations, they say, the advantage is substantial: at about 30 channel uses, the ratio of quantum Fisher information to channel number exceeds the asymptotic ratio achievable with the spin-squeezed strategy.
The paper also examines whether adding higher-order measurements to the spin-squeezed strategy would provide a comparable improvement. It reports that this measurement advantage appears only in strong-noise and strongly negative-correlation regimes, not at the moderate noise strength discussed earlier in the analysis. For Gaussian noise with small enough variances, the authors say higher-order moments do not improve the estimation scheme because the first-order-moment estimator reaches the classical Cramér–Rao bound asymptotically.
The result is a qualified case for more elaborate sensing protocols, not a blanket replacement for spin squeezing. The advantage depends on the noise statistics: a sensor facing positively correlated phase shifts may get nearly all the available benefit from the simpler squeezed-state strategy, while a sensor facing negative correlations may have more to gain from optimization. The paper discusses time-varying magnetic fields as one physical motivation for estimating a frequency in the presence of fluctuations.
Bounds and finite numerical models set the limits
To benchmark the strategies, the researchers derive a tighter classical-simulation bound and compare it with bounds from the quantum-comb extension method. They use tensor-network calculations to search protocols for systems with up to about 30 channel uses. The calculations approximate the continuous noise model with a discretized one using the Rouwenhorst method. The paper says this approximation is sufficient for moderate correlations, but does not reproduce properties of the continuous Gaussian model when correlations are strong.
The findings are theoretical comparisons, not results from a sensing experiment. Their practical message therefore depends on whether a real sensor’s phase-noise correlations can be characterized well enough to choose a matching protocol, and whether the predicted advantage holds beyond finite numerical calculations. The paper’s reported gains from optimized strategies are most pronounced for negative correlations; its analysis of higher-order measurements specifies that their advantage occurs in the strong-noise regime.