Memory dimension detection in open quantum dynamics via pseudo-control

AI summary

Overview

Research area: Quantum information and open quantum systems — specifically, characterising the environmental memory (non-Markovianity) of open quantum dynamics using a circuit-level protocol.

Technical level: Advanced. The paper assumes familiarity with open quantum systems, process tensors/quantum combs, tensor-network diagrams, and interference/coherent-control circuits. The motivating question and headline result are stated in relatively accessible terms, but the proof machinery (appendices C and E) is specialist material.

Scope: The paper proposes and numerically tests a "pseudo-control" interferometric protocol that certifies a lower bound on the memory dimension of a two-interaction open quantum process from experimentally accessible output measurements, applied to five model many-body environments.

What This Paper Is About

Open quantum systems retain information about past interactions in environmental degrees of freedom that are usually inaccessible, and the size of the smallest effective memory needed to reproduce that influence is called the memory dimension. Existing approaches either measure the magnitude of non-Markovianity or reconstruct the entire multi-step process tensor, which is experimentally demanding. This paper asks whether measurements of accessible outputs alone — without touching the environment and without full process-tensor tomography — can certify a lower bound on that memory dimension.

Key Contributions

1. A pseudo-control protocol. A control qubit in |+⟩ coherently routes a system through one of two system–environment interactions separated by a waiting interval, so that the correlation between the two interaction times appears as a dependence between rows and columns of an interference matrix M_RS (the off-diagonal control block ω₀₁^RS).

2. A certified lower bound (Theorem 1). For a pure reference-system input, rank(M_RS) ≤ d_m², hence d_m ≥ ⌈√(rank(M_RS))⌉. The proof is given via tensor-network rank cuts (Appendix C) with an equivalent quantum-comb derivation (Appendix E). The bound is explicitly one-sided: high rank certifies large memory, but low rank does not certify small memory.

3. An operational reconstruction scheme. M_RS is not a direct observable; the paper shows it is recovered exactly from X- and Y-basis control measurements combined with conditional tomography on the retained reference–system output: M_RS = ½[Ω₊ − Ω₋ − i(Ω₊ᵢ − Ω₋ᵢ)]. Rank is taken from the singular-value spectrum because M_RS need not be Hermitian.

4. A numerical survey of five many-body environments (models A–E) examining how the resolved rank depends on interaction time, interval time, coupling strength, internal environmental dynamics and coupling geometry, including a threshold-based version r_ε = #{k : s_k(M_RS) > ε} for imperfect switch-off.

Main Findings

• Reference size strengthens the certificate. For a pure input of Schmidt rank s = rank(ρ_S), rank(M_RS) ≤ s·d_S; raising s from 1 to d_S lifts the rank ceiling from d_S to d_S². The simulations therefore use the maximally entangled input |Φ_{d_S}⟩_RS with full Schmidt rank s = d_S.

• Interaction time has competing effects. At τ = 0 no information is exchanged and M_RS stays at its rank-one baseline. As τ increases, r_ε initially increases for all five models; at longer τ it generally decreases for the dissipative models A–D, while the nondissipative model E retains a larger rank. Models C and E share the same coherent Hamiltonian, so their difference isolates the effect of depolarisation.

• Memory decays with the interval, with local revivals. With S and E both taken as three-qubit chains, fixing τ = τ₀ and λ = 1 and varying T, r_ε generally decreases with T for models A–D as singular values fall below the threshold ε. The broader (τ, T) landscape shows local increases superimposed on this decrease, attributed to coherent redistribution of information inside E during the interval.

• Coupling strength produces a rank ceiling that depends on internal dynamics. At λ = 0 the system and environment do not interact, so r_ε = 1. With boundary contact and no environmental propagation (model A, H_E = 0), only E₁ can carry memory and the rank saturates at r_ε = 4, giving d_m ≥ 2. In the commuting longitudinal-field Ising chain (model B) E₂ can also contribute through the E₁–E₂ interaction, reaching r_ε = 8 and d_m ≥ 3.

• Geometry controls the strong-coupling behaviour. For boundary contact, r_ε decreases again at large λ. The same decrease appears in model E without depolarisation, so it is not attributable to dissipation; the authors explain it as the strongly coupled S_n–E₁ pair becoming an access bottleneck to the rest of the environment. Replacing boundary contact with a ladder contact (each S_j coupled to E_j, with a 1/n factor giving equal spectral width) removes the decrease and lifts the rank ceilings found in models A and B.

• The bound survives imperfect decoupling. When the system–environment interaction cannot be fully switched off during the interval, the resolved rank r_ε still yields d_m ≥ ⌈√(r_ε)⌉ for a threshold ε whose choice depends on the form of the residual interaction. Appendix D also shows the protocol detects classical as well as quantum memory.

Methodology in Plain English

The authors build an interferometer out of a single control qubit. The control is put in a superposition, and depending on its value the system is sent through either the first or the second system–environment interaction, with the other interaction receiving a fixed auxiliary input instead. Because both routes are traversed coherently, the two probing histories interfere, and the interference term — the off-diagonal block of the control's final state — records how much the first interaction influenced the second. Measuring the control in the X and Y bases while performing tomography on the retained system-plus-reference output lets the experimenter reconstruct that block exactly. The rank of this matrix is then bounded by the square of the memory dimension, which converts a measured rank into a certified lower bound.

To test the idea, the authors simulate the full joint system–environment dynamics with an explicit master equation: interaction time τ under a Hamiltonian with coupling strength λ plus a depolarising dissipator on the environment, then an interval of length T with the coupling switched off and the environment evolving under its own Hamiltonian and dissipator. They sweep τ, T and λ across five environment models deliberately chosen to separate different physical effects: a non-propagating baseline (A), a commuting Ising chain (B), an XXZ-type chain (C), a nonintegrable transverse-plus-longitudinal-field Ising chain (D), and a dissipation-free version of C (E). They also compare two coupling geometries — a single boundary link versus a ladder of parallel links — to test whether a strong-coupling drop in rank is about access to the environment rather than about memory itself.

Why This Matters

Impact on research. The protocol gives an operational, experimentally grounded alternative to process-tensor tomography for a question that previously required either full multi-time reconstruction or indirect proxies such as measures of non-Markovianity. Because the certified bound responds to coupling geometry and environmental dynamics, it turns a resource-theoretic quantity (memory dimension) into a diagnostic of system–environment structure.

Potential applications (as motivated in the paper):

• Characterising temporally correlated noise in quantum hardware, including gate-dependent non-Markovian noise models of the kind studied with randomized benchmarking.
• Assessing memory requirements for simulating general open quantum dynamics and quantum many-body dynamics.
• Informing performance analysis for quantum communication over memory-bearing channels.
• Informing resource requirements for fault-tolerant quantum computation, where environmental memory affects overheads.

Industry relevance. Quantum computing and communication platforms must characterise non-Markovian noise to bound error-correction overheads and validate hardware. A protocol that certifies a memory dimension without requiring environment access or full tomography is directly relevant to benchmarking and noise-model validation workflows.

Future Directions

• Simpler circuit implementations. The authors explicitly flag reducing experimental overhead as future work, since the current protocol requires conditional tomography on the retained reference–system output.
• Connecting propagation to required memory. The paper proposes clarifying how information propagation in many-body environments determines the memory dimension needed to describe the resulting open-system dynamics.
• Closing the one-sided gap. The bound certifies only that memory is at least ⌈√rank⌉; conditions under which a low observed rank does reflect genuinely small memory remain unaddressed.
• Overlap with related work. The acknowledgements note forthcoming related work by Lisa Kolesnyk and Ingo Roth on probing the effective environment dimension of non-Markovian, gate-dependent noise models via randomized benchmarking, raising the question of how the two approaches compare.

Target Audience

Researchers in quantum information and open quantum systems working on non-Markovian dynamics, process tensors and quantum combs; experimentalists in quantum hardware characterisation and noise benchmarking; and theorists interested in tensor-network rank arguments and coherent-control/interferometric circuit techniques. Readers without a background in process tensors will find the conceptual motivation accessible but the proofs demanding.

Authors’ abstract

Characterising open quantum dynamics requires identifying the environmental degrees of freedom that shape the system's evolution. The memory dimension, defined as the size of the smallest effective memory needed to reproduce this influence, quantifies the dynamically relevant environmental resources and provides a common basis for comparing different many-body environments. Here we introduce a pseudo-control protocol that encodes information about the memory dimension in an experimentally accessible interference matrix, whose rank certifies a lower bound on this dimension without direct access to the environment or full process-tensor tomography. Applied to quantum many-body environments, the protocol allows flexible adjustment of the interaction and interval times, revealing how these parameters, environmental dynamics, and system-environment interactions shape detectable memory. Our approach provides a new operational technique for characterising open quantum dynamics and probing the structure of system-environment interactions through their memory signatures.