A Finite-Window Recovery Hierarchy for Local Quantum Memory

AI summary

Overview

Research area: Quantum physics — open quantum systems, quantum information, and many-body dynamics. Specifically, the paper sits at the intersection of many-body localization physics, quantum channel recovery theory, and variational/learned quantum control.

Technical level: Advanced. The paper assumes fluency with CPTP maps, Choi states, coherent information, semidefinite programming, Floquet dynamics, and matrix-product-state tensor networks.

Scope (one sentence): The paper defines and numerically validates a locality-constrained benchmark — finite-window recoverability, implemented by a protocol called local variational recovery (LVR) — that asks whether quantum information which has left a target qubit can still be recovered by bounded-depth control acting only on a small surrounding window.

What This Paper Is About

When quantum information stored in a single qubit seems to vanish, two very different things may have happened: the information may have been destroyed, or it may have leaked into nearby degrees of freedom in a form that shallow local control cannot easily reach. Existing diagnostics — spectral statistics, target-site observables, echo protocols, full channel tomography — each answer a different question and none of them asks what a bounded-depth decoder acting on a fixed finite window can actually return to the target qubit. The authors close this gap by defining recovery relative to an explicit, deliberately narrow access model (local state preparation, many-body evolution, depth-limited control on a radius-r window, and target-qubit Pauli readout), then benchmarking three recovery scores against each other on a disordered kicked-Ising Floquet chain.

Key Contributions

1. A three-level recovery hierarchy. The paper formalizes three channel-level quantities on the same physical setup: the target-only optimum Q₀^opt (what is recoverable from the target site alone), the shallow score Q_r^shallow (what a bounded-depth decoder actually recovers from the radius-r window), and the same-window optimum Q_r^opt (what is recoverable in principle from that identical window). The ordering Q₀^opt < Q_r^shallow < Q_r^opt separates existence of off-site recoverable memory from its accessibility to shallow control.

2. Local variational recovery (LVR) and two derived certification quantities. LVR is the operational protocol: prepare the target qubit in one of six cardinal probe states against a fixed product-state background, evolve, apply a layered local decoder on the window, and read out target-qubit Paulis. From this the authors define the certified gain Δ_r^cert = Q_r^shallow − Q₀^opt and the shallow-accessibility fraction η_r = (Q_r^shallow − Q₀^opt)/(Q_r^opt − Q₀^opt).

3. A no-go observation that makes the gain meaningful. Observation 2 proves that if the window channel factors through the target channel via any CPTP post-processing (ℰ_r = 𝒯_r ∘ ℰ₀), then Q_r^opt = Q₀^opt. Consequently, any observed Q_r^shallow > Q₀^opt is a genuine certificate that recoverable information exists outside the target site but inside the chosen window — not merely a post-processing artifact.

4. A hierarchy-guided carrier-deletion repair test. Using the disorder window the hierarchy selects, the authors delete and reset the original target register after the dynamics and ask whether the surrounding halo can repair the input. A depth-8 decoder on a radius-3 halo succeeds, and outperforms unrestricted CPTP decoders restricted to every one-, two-, and three-site halo subwindow — turning the diagnostic into a local repair stress test.

Main Findings

• The five-site hierarchy holds in the crossover regime. With a depth-6 decoder on a radius-2 (five-site) window, the ordering Q₀^opt < Q₂^shallow < Q₂^opt is realized across the crossover regime of the N = 12 kicked-Ising chain, with positive certified gain for most disorder realizations.

• Certified gain peaks where on-site memory has already weakened. The median certified gain Δ₂^cert is positive across the full disorder scan and largest in the intermediate/crossover regime — the opposite of a monotonic "more disorder means more memory" trend.

• Shallow accessibility rises strongly with disorder. The median shallow-accessibility fraction η₂ rises from about 0.50 at W ≈ 2.36 to about 0.94 by W ≈ 5.09, and is already about 0.83 near the certified-gain peak at W ≈ 3.73. These fractions are computed on the full 100-realization exact benchmark.

• The signal is not target-site persistence. Target-site magnetization retention ⟨Zi(D)⟩{|0⟩} grows almost monotonically with disorder, whereas Δ₂^cert turns over in the intermediate regime. The two profiles, normalized by their own maxima, have visibly different shapes.

• The signal is not the tomographic channel-information increment. The reconstructed-channel coherent-information increment ΔI_c,2 counts channel content retained anywhere in the window, while Δ₂^cert asks what beats the optimal on-site baseline under bounded local decoding. Appendix H adds an operator shell-weight argument: local support inside Λ₂ is necessary but not sufficient for shallow target recovery, and shell weight can keep increasing after Δ₂^cert has already peaked.

• Carrier deletion is repairable. At the frozen held-out working point W = 3.3, D = 15, r = 3, L = 8, evaluated on 80 held-out disorder realizations (IDs 20–99), the full-halo shallow decoder reaches median Q = 0.516, i.e. F_avg = 0.758. The 95% bootstrap interval Q ∈ (0.478, 0.567) lies above the single-qubit classical benchmark Q = 1/3 (F_avg = 2/3), and 95% of held-out realizations exceed that benchmark.

• Small halo subwindows cannot explain the repair. Best one-, two-, and three-site halo-subwindow CPTP medians are Q = 0.233, 0.336, and 0.452 respectively, against 0.516 for the full-halo shallow decoder. Paired gains are Σ₁ = 0.260, Σ₂ = 0.136, and Σ₃ = 0.035, all with 95% bootstrap intervals strictly positive. The three-site separation is described as modest but statistically clean.

• Positive radius-2 gain survives longer-chain embedding. With an independent open-boundary MPS/TEBD backend at N = 24, 32, 40 (32 disorder realizations, bond dimension χ = 128, singular-value cutoff 10⁻¹⁰), the intermediate-disorder signal persists: for N = 24 near W ≈ 3.2–3.6 at D = 20–40, for N = 32 near W ≈ 4.4 at D = 20 and W ≈ 3.6 at D = 30, and for N = 40 the peak remains in the crossover-to-strong-disorder window. On the subset where the same-window SDP was solved, typical η₂ is about 0.65–0.85 for N = 32 and 0.55–0.80 for N = 40.

• What is not reported: the paper presents no experimental data and no hardware demonstration; all results are numerical. It also explicitly does not claim algorithmic scalability for windows that grow with system size — the tensor-network section tests persistence of the witness, not scaling.

Methodology in Plain English

The authors fix a target spin in a chain and define a "recovery window" as the target plus its r nearest neighbors on each side (radius 2 gives five sites). They prepare the target qubit in six probe states — |0⟩, |1⟩, |±⟩, |±i⟩ — while everything else starts in a fixed Néel product-state background, let the system evolve under the kicked-Ising Floquet dynamics, and then try to undo the damage.

Three recovery attempts are compared on exactly the same data. The first is the best possible decoder that sees only the target qubit — a rigorous ceiling computed by semidefinite programming over all CPTP maps. The third is the best possible decoder that sees the whole window — another SDP ceiling, but with more information available. The second, and the one that could actually be run in a lab, is a shallow layered circuit of on-site Euler rotations and controlled-Z gates restricted to the window, with a fixed depth (L = 6 for the main scan) and parameters trained by Adam with learning rate 0.03 for 300 steps. Because the six probe states form a qubit 2-design, the average fidelity over those six states equals the Haar-averaged fidelity F_avg, and the score Q = 2F_avg − 1 is computed from six target-qubit Pauli expectation values without any tomography of the window.

For the repair test, they go further: after the evolution they trace out the target site and replace it with a fresh |0⟩ ancilla, so any input-dependent information must reside entirely in the surrounding halo. They pick the working point using a shallow-only scan on realization IDs 0–19, freeze it, and report results only on IDs 20–99. Statistics use one-sided Wilcoxon signed-rank tests (Pratt zero handling) and percentile bootstrap intervals from 10⁴ paired resamples. To check that none of this is an artifact of the N = 12 exact-diagonalization setting, they re-run the same radius-2 task with an MPS/TEBD backend on open chains of 24, 32, and 40 sites, tracing down to the same five-site window before computing the identical scores.

Why This Matters

Impact on research. The paper draws a sharp operational line between three questions that are routinely conflated: does memory survive on the site, does it survive anywhere in a neighborhood, and can a realistic bounded-depth controller actually retrieve it. Proving that Q_r^opt = Q₀^opt whenever the window channel factors through the target channel gives the certified gain a clean information-theoretic interpretation, so a positive Δ_r^cert is not a numerical artifact of ansatz choice. This gives many-body-localization and Floquet-thermalization studies a channel-level diagnostic that is distinct from spectral statistics and from local-imbalance observables, and it gives quantum error-correction research a way to quantify how much of the recoverable content a shallow decoder of a given depth can reach.

Potential application areas (these are domains the framework speaks to; the paper itself reports only numerical benchmarks, not deployed applications):

• Quantum memory and storage validation. Distinguishing "the qubit lost its state" from "the state is in the neighborhoood but inaccessible" changes how a memory device's failure is diagnosed and whether it is worth attempting retrieval.
• Quantum error-correction code and decoder design. The shallow-accessibility fraction is a direct, hardware-agnostic measure of how much of the theoretically correctable information a depth-limited decoder actually captures, which is the practical bottleneck for real-time decoders.
• Benchmarking quantum simulators. The Floquet-chain protocol is a concrete, locally executable task that could be run on an analog or digital simulator to characterize its effective memory retention, using only local preparation and single-qubit readout.
• Quantum networking and repeater nodes. Where local information must be recovered after propagation and storage with only bounded local operations, the carrier-deletion test is a natural stress test for whether a node's surroundings can repair a reset register.

Industry relevance. For quantum-computing companies, the benchmark is attractive because it is cheap in the operational direction: LVR needs only local state preparation, a depth-L circuit on a small window, and Pauli readout — no tomography of the window. The exact comparators Q₀^opt and Q_r^opt require SDP post-processing and are used only as numerical reference points on tractable instances. This makes the shallow score a plausible device-level metric, and the paper's explicit protocol bookkeeping (frozen held-out points, disjoint selection and test realization sets, paired statistics) models the kind of methodology a hardware benchmarking effort would need.

Future Directions

• Larger recovery windows and deeper decoders. The paper explicitly declines to claim scalability for windows that grow with system size; the natural next step is to ask how the hierarchy behaves as radius and decoder depth increase, and whether the depth-6, radius-2 point sits near a genuine accessibility threshold.

• From numerics to hardware. All reported results are numerical. Running the same LVR protocol — six probe states, local preparation, Pauli readout — on a physical disordered Floquet simulator would test whether the certified gain and the carrier-deletion repair survive real noise and control error.

• Microscopic theory of the accessibility gap. The operator shell-weight analysis in Appendix H establishes that local support inside Λ₂ is necessary but not sufficient for shallow recovery, and that shell weight can keep growing after Δ₂^cert has peaked. Understanding which structural features of the spreading operator determine whether that support is refocusable by bounded-depth control remains open.

• Relation to approximate quantum error correction. The paper positions itself in the spirit of recovery-map and approximate-correction formulations; connecting the measured Δ_r^cert and η_r to explicit code-theoretic quantities — such as whether the window channel is correctable or only approximately so — would turn the benchmark into a design tool rather than a diagnostic.

Target Audience

This paper is written for researchers already working in quantum information and many-body physics: theorists studying many-body localization, Floquet dynamics, and scrambling; quantum-information scientists working on channel recovery, approximate quantum error correction, and decoders; and experimentalists who benchmark quantum simulators and memory devices and who want a locally executable task rather than a tomography-heavy one. Readers need comfort with CPTP maps, Choi states, coherent information, and semidefinite programming. Those without that background can still extract the central qualitative message — that losing local recoverability is not the same as losing the information, and that a shallow decoder on a small window can certify and partially repair the difference — but the quantitative content requires an advanced footing.

Authors’ abstract

When quantum information initially stored in a local qubit disappears, it need not be lost: it may have moved into nearby degrees of freedom or become inaccessible to shallow local control. We introduce finite-window recoverability as an operational channel benchmark that separates these possibilities. It compares optimal recovery from the target site, recovery by a bounded-depth decoder on a finite window, and the unrestricted optimum for that window. Its operational component, local variational recovery, uses local state preparation, window-local control, and target-qubit Pauli readout to certify recoverable memory beyond the target and quantify how much of the same-window advantage is accessible to shallow control. In a disordered kicked-Ising Floquet chain, a depth-6 decoder on a five-site window realizes $Q^{\mathrm{opt}}_0&lt;Q^{\mathrm{shallow}}_2&lt;Q^{\mathrm{opt}}_2$ across the crossover regime, with positive certified gain for most disorder realizations and substantial shallow-accessibility fractions. The signal differs from target-site persistence and reconstructed coherent-information increments. Positive radius-2 gain also persists when the task is embedded in longer open chains using an independent tensor-network backend. Guided by this hierarchy, we test a carrier-deletion task in which the original target register is reset after the dynamics. A depth-8 decoder repairs the input from a radius-3 surrounding halo with held-out median $F_{\mathrm{avg}}=0.758$, above the single-qubit classical benchmark $2/3$, and outperforms optimal one-, two-, and three-site halo-subwindow counterfactuals. These results establish finite-window recovery as a local-control benchmark for off-site quantum memory, diagnosing both where local quantum information remains and whether bounded-depth control can refocus it.