Quantum non-Markovian response spectra

AI summary

Overview

Research area: Quantum physics — open quantum systems, quantum stochastic processes, and the process-tensor framework for non-Markovian memory.

Technical level: Advanced. The paper assumes comfort with Choi operators, instruments, the generalized Born rule, singular-value decomposition, and semidefinite programming, though its central idea (a measured matrix whose spectrum reveals memory) is stated in fairly intuitive terms.

Scope (one sentence): The paper introduces a "response matrix" built from selected causal-break experiments that reveals the structure, spectrum, and quantum character of environmental memory in a quantum process without requiring full process tensor tomography.

What This Paper Is About

Quantum devices are never perfectly isolated: hidden environmental degrees of freedom can store information from earlier operations and release it later, so a system's future behaviour depends on more than its present state. Detecting that memory is comparatively easy (do a causal break and see whether the future still depends on the past), but learning how that memory is structured, where it lives in time, and whether it is genuinely quantum normally requires reconstructing the entire multi-time process tensor, whose cost grows exponentially with the number of intervention times. The paper asks whether a small, well-chosen set of controlled "break, prepare, probe" experiments can expose that temporal structure directly.

Key Contributions

1. The response matrix. A construction that arranges measured causal-break data with future records as rows and conditioned histories as columns, so that each entry is a single independent black-box query. Because no query depends on another, the matrix can be assembled in parallel in simulation and built identically from experimental data or from learned open-system models.

2. An exact relation to the underlying process. A tester-frame factorization, (V_c = F,\Upsilon_{F\leftarrow H},H^{\mathsf T}), linking the sampled response matrix to the process Choi operator regrouped across the history–future cut, together with a prescription for recovering the branch probability of each history from a "deterministic" future tester. The authors state conditions under which the rank and normalized singular spectrum of (V_c) match those of the underlying process.

3. Spectral diagnostics for memory. Proof that rank greater than one certifies history dependence at the cut (Proposition 1: rank one holds if and only if the normalized future response is the same for every tested history with nonzero probability), plus mode weights (p_m = \sigma_m^2/\lVert V_c\rVert_F^2), a response entropy (S_V = -\sum_m p_m \log p_m), and an effective mode number (R = \exp(S_V)).

4. A witness search formulated as a semidefinite program. Building on temporal-Choi-entanglement witnesses, the authors formulate the search for a quantum-memory witness inside the span of the selected instrument probes as an SDP that enforces non-negativity on every process separable across the history–future cut. A negative value excludes the adopted classical feed-forward model without reconstructing the Choi operator; failure to find one is explicitly stated to be inconclusive.

Main Findings

• Rank is an exact memoryless test within the probes. A response matrix of rank one means the normalized future response is history-independent; rank greater than one certifies that memory is visible. Rank one implies genuine process factorization only when the frames are informationally complete, and global Markovianity additionally requires the condition at every relevant cut and for all admissible instruments.

• The spectrum resolves how memory is organized, not just whether it exists. The singular values partition the squared Frobenius norm of the response, response modes pair a combination of histories with its associated future response, and retaining the (k) largest modes gives the optimal rank-(k) approximation with error (\sum_{m>k}\sigma_m^2). (S_V) ranges from (0) to (\log r) and (R) from (1) to (r); a rank-one response has (S_V = 0), (R = 1), and equal mode weights give (R = r).

• Spectral diagnostics are probe-relative, as illustrated by a classical-bit bath. Appendix A describes a bath storing a classical bit across a system reset: when the probes resolve both bit values, (V_c) has two nonzero modes, with (S_V = \ln 2) for equally likely values and small entropy for a strongly biased bit despite perfect recall. If the history outcome is ignored, only one response column remains and (S_V = 0) even though the bath dynamics are unchanged.

• Ising benchmark: qualitative tracking without exact recovery. In an interacting Ising process, the authors report that the response entropy and spectrum track the qualitative changes in the corresponding process tensor MPO bond entropy and spectrum, even when the exact-recovery conditions are not imposed. The study examines how the response spectrum changes with coupling, temporal cut, and intervening resets.

• Two-qubit benchmark: memory outlives classical reproducibility. In a storage-and-retrieval benchmark on two qubits, memory remains visible after it becomes classically reproducible, and an analytically supplied witness (evaluated using the response data) distinguishes the quantum and classical regimes. Because the paper text provided here is truncated, the specific numerical witness values and tomographic sample counts for these benchmarks are not reported in the available content.

• No full tomography is required. Useful information about non-Markovian spectra is obtained without process tensor tomography or an informationally complete probe set; outside the derived recovery conditions the diagnostics remain probe-dependent by design.

Methodology in Plain English

1. Break the memory link deliberately. At a chosen time, measure the accessible system, record the outcome, throw the system away, and prepare a fresh known state. Anything that still influences the future must have been carried by degrees of freedom that were not reset.

2. Vary the past and the future separately. Run this protocol for many different pre-break histories (including the break outcome) and many different post-break preparations, later controls, and final readouts. Each combination is one independent run.

3. Tabulate. Put every measured coefficient at the intersection of one conditioned history (column) and one future probe (row). This table is the response matrix.

4. Take a singular-value decomposition. The singular values form the "response spectrum"; the paired directions are "response modes." Rank above one flags memory, the spread of the singular values (via entropy and effective mode number) says whether the response is concentrated in one pattern or spread over several.

5. Repeat at different cuts and with intervening resets to see how long the memory patterns persist.

6. Test for quantumness separately. Using the same tabulated data, run a semidefinite program to look for a witness with negative expectation value inside the accessible probe span; a negative result rules out the chosen classical model.

Why This Matters

Impact on research. The work fills a practical gap between a binary "is there memory?" test and full process tensor tomography, whose resources grow exponentially in the number of intervention times. It connects to predictive-state and Hankel-realization approaches for inferring hidden dynamics from observable responses, and it gives an operational, probe-tailored alternative to reconstructing a process tensor matrix product operator (PT-MPO).

Real-world applications (as motivated in the paper):

• Diagnosing correlated noise in quantum processors, where spectators, leakage levels, material modes, or slowly drifting control parameters can retain information — including the documented case where one quantum job leaves an environmental trace affecting the next even after qubits are reset.
• Validating or falsifying open-system and Markovian noise models used in device characterization and benchmarking, since memory can distort benchmarking signals.
• Memory-aware control and error-mitigation design, where strategies that ignore system–environment correlations are known to underperform.
• Protecting quantum error correction, which the paper notes can be compromised by environmental memory.

Industry relevance. Hardware vendors, compiler and calibration teams, and cloud quantum-computing providers all need cheap diagnostics of job-to-job memory effects; because the queries are independent, the matrix construction parallelizes naturally on the same hardware it diagnoses.

Future Directions

• Experimental deployment on hardware. All benchmarks here are numerical; whether the same response spectra can be measured reliably against shot noise on real devices is not reported in the available content.
• Statistical error control and sampling. The paper states that increasing the number of sampled history and future probes tests convergence of (S_V) and (R), that repeated measurements improve precision, and that singular values above a numerical or statistical resolution threshold are retained — but the truncated text does not give the resulting error bars or the thresholds used, leaving practical sample-complexity guidance open.
• Choosing probe sets deliberately. Since the diagnostics are probe-dependent outside the recovery conditions, methods for selecting or adaptively refining instruments — or certifying what a given restricted probe set can and cannot resolve — remain an open design question.
• Extending and combining the witness search. The SDP finds a certificate only for the accessible space and chosen positive map; connecting it to larger probe families and to compressed representations such as PT-MPOs is a natural next step.

Target Audience

Researchers in open quantum systems and quantum stochastic processes; quantum-computing experimentalists and calibration engineers interested in characterizing correlated noise and job-order effects; and theorists working on process tensors, quantum memory witnesses, predictive-state and realization methods, and tensor-network descriptions of non-Markovian dynamics. Readers should have graduate-level familiarity with quantum channels and Choi representations to follow the derivations, though the conceptual narrative of the response matrix and its spectrum is accessible to a broader physics audience.

Authors’ abstract

Quantum systems are rarely isolated from their surroundings. These hidden environments can retain information and return it later, producing correlated errors in quantum computers and complicating the interpretation of experiments. On a quantum processor, for example, one job can therefore leave an environmental trace that affects the next, even after the qubits themselves have been reset. Learning how this hidden influence persists and affects later behavior is therefore important, but the environment is often impossible to measure directly, and fully reconstructing its effects through tomography rapidly becomes impractical. Here, we introduce a response matrix that uses controlled interventions and observations of the system alone to reveal the temporal structure of environmental memory. The matrix records how future observations change across different controlled histories. Its spectrum characterizes how memory is structured and how it persists under specified system resets, while suitable linear combinations of its entries can rule out specified classical models of that memory. Because its history-future queries can be evaluated independently, numerical response matrices can be assembled in parallel. We establish the framework theoretically, characterize response spectra in an interacting spin process, and certify quantum memory in a two-qubit benchmark. This approach turns system observations into a probe of inaccessible environments and could support the diagnosis of correlated noise, tests of open system models, and memory-aware control design.