Two-qutrit steering detection turns to correlation-enhanced quantum reservoirs: unseen Werner-state accuracy reaches 96%, and new steerable partially entangled states are flagged
Synopsis
Deciding whether a mixed two-qutrit state is steerable has relied on semidefinite programming over a finite measurement set, and that does not necessarily carry over to a different state family. Correlation-enhanced quantum reservoir processing (CEQRP) adds two-body spin-correlation readouts from a spin-1 reservoir to the feature vector, letting logistic regression reach high accuracy on unseen Werner and isotropic states. With training augmented by samples certified by two steering criteria, classifiers flagged partially entangled states whose steerability semidefinite programming then independently confirmed using seven projective measurement settings.
Interpretation
Two-qutrit steering detection now has a feature representation that transfers across state families, not just a classifier that works inside its training distribution. Classifiers trained on the full-information representation or the steering-ellipsoid-like representation previously generalized only to a limited degree to unseen Werner states, while performing well on isotropic states. At measurement-setting counts m=4 and m=5, logistic regression on full-information CEQRP features reached 96.00% and 96.03% on unseen Werner states and 99.35% and 97.08% on isotropic states, respectively.
The features come from the dynamical response of a fixed spin-1 quantum reservoir, with the input state encoded into its first two sites. The reservoir uses a transverse-field Ising model with single-ion anisotropy, and the readout adds longitudinal Ising correlations and transverse spin correlations between site pairs to the local spin observables. Expectation values of these observables at several evolution times are stacked into a feature vector and passed to classical readouts such as logistic regression and AdaBoost; training and testing use semidefinite-programming-labeled two-qutrit datasets, with Werner and isotropic states excluded from training.
Perspective
The demonstration is confined to two-qutrit systems: classifiers train on semidefinite-programming-labeled random states and are tested on held-out Werner and isotropic families; the newly found steerable partially entangled states are confirmed at fixed parameters by semidefinite programming with seven projective measurement settings on Alice's side.
Which two-body correlation readouts actually drive the cross-family gain, and whether any newly found partially entangled state can be shown steerable from Alice to Bob while unsteerable from Bob to Alice under all projective measurements, remain open questions raised in the text.