{"id":"a7815b6c-c085-4e3a-a29f-0fd6b30506ff","arxiv_id":"2506.23267","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Pseudo-density-matrix measures detect eternal CP-indivisibility that temporal steering, trace distance, and entropic measures miss, and a new continuous-time causality measure generalizes the Rivas-Huelga-Plenio witness.","lead":"Two quantum-correlation witnesses of non-Markovianity, based on pseudo-density matrices and temporal steering, are benchmarked against trace-distance and entropic measures on three known channels. The paper finds that pseudo-density-based measures catch eternal non-Markovianity that the steering-based and distance-based measures miss, and proposes a continuous-time witness for CP-indivisible memory.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The CCM witness is ill-defined as written: Eq. (13) with Eq. (12) does not have ||R||1→1 as ε→0 (identity intermediate gives ||R||1=2), so μ diverges; no normalization or monotonicity proof ensures CP intermediate maps give μ≤0.","rationale":"The reader's weakest assumption is precisely the load-bearing point: the positivity of μ(t) for CP intermediate maps is assumed but not proven, and the normalization that would make the identity intermediate map have unit trace norm is missing. I agree, and I would sharpen the concern: Eq. (13) as written is not merely unproven but undefined, because for the identity intermediate map the numerator ||R_{t+ε,t}||1−1 tends to 1, not 0, so μ(t) diverges. Equation (14) implicitly uses a base operator I4, which is not what Eq. (12) produces for V=I. The claim that CCM faithfully captures eternal non-Markovianity therefore depends on an unstated normalization and on a monotonicity theorem that the paper does not supply. The example-based evidence (three channels) is too narrow, especially since a single CP-divisible counterexample would invalidate the measure. I do not find a reason to change the reader's CONDITIONAL verdict: the LCM measure based on the total PDM's log negativity is essentially a Choi-negativity witness and is likely salvageable, but the paper should either provide the explicit normalized definition and a proof of the monotonicity property, or downgrade the CCM claims to numerical observation. The internal inconsistency in Sec. IVB ('stronger' where 'less sensitive' is meant) is minor by comparison.","tokens_in":13561,"tokens_out":16657,"duration_ms":148318,"concrete_test":"Recompute μ(t) from Eq. (12)–(13) for the CP-divisible pure dephasing channel with γ_z(t)=1 for all t and for a CP semigroup with a nonzero Hamiltonian term, using the explicit normalization that makes the identity intermediate map have ||R||1=1. If either computation yields μ(t)>0 for any input state ρ, the CCM witness flags a Markovian process. Also, check Eq. (14) by substitution: with ρ=11_2/2 and L=0, Eq. (14) gives ||I4||1=1 but Eq. (12) gives ||R||1=2, proving the two definitions are inconsistent unless an unstated normalization changes the base operator.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central new step is the continuous causality measure N_CCM built from μ(t)=lim_{ε→0+}(||R_{t+ε,t}||1−1)/ε in Eq. (13), with R_{t+ε,t} defined in Eq. (12). For the identity intermediate map, Eq. (12) gives R_{t,t}={ρ⊗11_2/2,S}; for ρ=11_2/2 this is 1/4∑_i σ_i⊗σ_i = SWAP/2, whose trace norm is 2. Thus the numerator in Eq. (13) tends to 1 rather than 0, so μ(t) diverges for every process; the sentence 'requires to be normalized appropriately' is the only acknowledgment, and the actual normalization is never specified. Even after a normalization is imposed, the sign of μ depends on the chosen base operator, and no theorem is given that the trace norm of a partial-transposed Choi operator is non-increasing under CP intermediate maps (this is not the usual LOCC monotonicity of the Choi state's negativity). Without such a theorem, a CP-divisible dephasing or unitary segment could give μ>0 for some ρ, and the claim that CCM 'faithfully capture[s] eternal non-Markovianity' is not established.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper compares two temporal-correlation witnesses of non-Markovianity—pseudo-density-matrix (PDM) causality measures and temporal steerable weight (TSW)—against trace-distance and entropic measures. It introduces a continuous PDM-based witness μ(t), defines a normalized measure N_CCM, and claims that PDM-based measures faithfully capture CP-indivisibility, including eternal non-Markovianity, while TSW captures only P-indivisibility. The paper also presents numerical examples for a non-unital generalized amplitude damping channel, a phase-covariant channel, and an eternally non-Markovian unital channel, and interprets the results in terms of weak versus strong forms of quantum direct cause.","tokens_in":13860,"tokens_out":16440,"duration_ms":157181,"significance":"If the proposed continuous causality measure were properly defined and accompanied by a proof of its monotonicity under CP intermediate maps, the paper would provide an optimization-free witness of CP-indivisibility that catches eternal non-Markovianity where trace distance, entropic measures, and temporal steering fail. The comparison between TSW and Choi-matrix-based measures is a useful contribution to the hierarchy of temporal quantum correlations. The relation between PDM negativity and the Choi matrix is also a valuable observation. However, the central new measure is not well defined as written, so the main claim currently rests on an unstated normalization and unproved monotonicity.","major_comments":[{"comment":"The continuous causality measure is not defined as written. For the intermediate map V=I and input ρ=11/2, Eq. (12) gives R_{t,t} = {11/4, S} = SWAP/2, whose trace norm is 2. Hence the numerator in Eq. (13) tends to 1, and μ(t) diverges for every process, regardless of whether the process is CP-divisible. The sentence after Eq. (14) saying that the witness 'requires to be normalized appropriately' is the only acknowledgment, but the normalization is never given; Figure 1 therefore cannot be reproduced from the formulas in the text. This is load-bearing because the CCM is the new measure claimed to 'faithfully capture eternal non-Markovianity'.","section":"II A, Eqs. (12)–(13)"},{"comment":"No theorem establishes the sign of μ(t) for general CP intermediate maps. The positivity of μ is asserted to indicate CP-indivisibility, but the only evidence is a few channel examples. For an arbitrary input ρ, R_{t+ε,t} is an input-dependent partial transpose of the Choi operator of V; it is not a positive operator, and no monotonicity result for its trace norm under CP maps is supplied. Since the trace norm of a partial-transposed Choi matrix can exceed 1 for perfectly CP maps (e.g., dephasing with ρ=11/2 gives ||R||1=2−2εγ), even a normalized version would require a separate proof that the derivative is non-positive in the CP case. Without that, the witness could flag CP-divisible dynamics as non-Markovian.","section":"II A, Eq. (13) and Definition 1"},{"comment":"The claimed reduction to the Rivas-Huelga-Plenio (RHP) measure is not correct as stated. For ρ=11/2, R_{t,t} is proportional to the partial transpose of the Choi matrix, not to the Choi matrix itself; for a CP dephasing intermediate map the trace norm of this partial-transposed object is greater than 1, whereas the RHP witness is based on the trace norm of the Choi state (which equals 1 for CP maps). The relation between μ(t) and decay-rate negativity therefore needs to be stated through the correct normalized object, or the claim of equivalence to RHP should be withdrawn.","section":"II A, after Eq. (15)"},{"comment":"The definition of N_CCM is incomplete: the integrals in Eq. (15) have an upper limit t that is also the time variable in the integrand, and the denominator ∫χ(t) divides by the total duration of positive-μ intervals, but the limit t→∞ is never specified. Moreover, with μ divergent as written, tanh(μ)=1 and the measure saturates at 1 for any process, which makes the numerical results in Figs. 3–5 impossible to interpret. The 0/0=0 convention and the ε-regularization of χ need to be made explicit and shown to be independent of the regularization parameter.","section":"II A, Eq. (15)"}],"minor_comments":[{"comment":"The statement that the logarithmic form's failure to detect the phase-covariant channel 'implies it is a stronger measure than the continuous case' is unclear and appears to contradict the usual meaning of stronger as more sensitive; please rephrase.","section":"IV B"},{"comment":"The caption gives p(t)=sin 2(5t), while the text defines p(t)=sin 2t; the two should be reconciled.","section":"Figure 2 caption"},{"comment":"The notation I4 and I2 is used without defining the relevant Hilbert-space dimensions, and the statement that a first-order Taylor expansion suffices for the limit needs justification, since higher-order terms can affect the limit when the trace norm is not analytic.","section":"Eq. (14)"},{"comment":"There are several typographical and grammatical issues, including 'temoprality', 'vis-versa', and the incomplete sentence 'see e.g,' in the Introduction; the manuscript should be carefully proofread.","section":"Introduction and throughout"},{"comment":"The claim that 'PDM-based LCM is found to be stronger than TSW-based measure' is not supported by Fig. 3, where TSW detects the phase-covariant non-Markovianity while LCM does not; please clarify which notion of strength is intended.","section":"Conclusions"}],"recommendation":"major_revision","confidential_remarks":"The central issue is the undefined normalization in Eq. (13). This is fixable in a revision: the authors should specify the normalization, prove the relevant monotonicity, and reconcile the claimed RHP reduction with the partial-transpose relation. The conceptual contribution—that strong direct cause (PDM) may witness CP-indivisibility while weak direct cause (TSW) witnesses only P-indivisibility—is interesting, and the numerical comparisons are potentially useful. I recommend major revision rather than rejection, provided the definitional gap is closed and the figures are regenerated from the stated formulas."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear Colleague,\n\nThe new thing here is a continuous-time PDM-based measure of CP-indivisibility, plus a comparison showing temporal steerable weight misses eternal non-Markovianity. The comparison is genuinely useful and the conceptual point—weak direct cause tracks P-indivisibility, strong direct cause tracks CP-indivisibility—is worth taking seriously. The paper is not a waste of time; it benchmarks three channels and the qualitative story is plausible.\n\nThe problem is that the central new measure, N_CCM, is not defined as written. Equation (13) defines μ(t) as the limit of (||R_{t+ε,t}||_1 − 1)/ε. For an identity intermediate map and ρ = 11/2, R_{t,t} = (1/2)SWAP, whose trace norm is 2. The numerator tends to 1, not 0, so μ(t) diverges for every process. The paper says the witness \"requires to be normalized appropriately\" but never supplies the normalization. This is not a minor typo; the measure as stated is ill-posed.\n\nBeyond that, even after a normalization is imposed, there is no proof that the trace norm of the intermediate PDM is non-increasing when the intermediate map is CP. Without such a monotonicity result, the claim that μ(t) is a faithful witness of CP-indivisibility is unsupported. The paper shows examples where it works, but that is not the same.\n\nI also want to flag that for ρ = 11/2 the measure reduces to the Rivas-Huelga-Plenio measure, so the agreement with decay-rate negativity in that case is a restatement of a known result, not a new finding. The genuinely new part is the input dependence, and that is precisely the part that lacks proof.\n\nThe numerics are not reproducible because no code or data is provided, and the input state for the PDM plots is not specified. There are also a few typos and an incomplete sentence in the introduction.\n\nWhat is good: the observation that TSW misses eternal non-Markovianity, while Choi-based measures catch it, is important and likely correct. The framing in terms of quantum direct cause is a nice way to organize the hierarchy of witnesses. The LCM part is standard.\n\nOverall: the paper should not be accepted as is. The main new measure needs a careful redefinition and a proper monotonicity proof. If that is done, the paper could be a solid contribution to the non-Markovianity literature. I'd send it to peer review—the question is worth referee time—but the referees should be instructed to scrutinize Eq. (13) and ask for the normalization and a theorem.\n\nBest.","headline":"The comparison of witnesses is useful, but the paper's central new measure is ill-defined as written; a serious revision is needed before the main claims can be accepted.","tokens_in":14429,"tokens_out":5430,"would_cite":false,"duration_ms":49674,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Pseudo-density-matrix measures catch non-Markovianity other witnesses miss","keywords":["non-Markovianity","pseudo-density matrix","temporal steering","CP-indivisibility","information backflow","quantum direct cause","Choi matrix","divisibility"],"falsifier":"Evaluate $\\mu(t)$ from Eq. (14) for a channel known to be CP-divisible, such as pure dephasing with a non-negative decay rate, using an input state other than the maximally mixed one. If any such CP-divisible step yields $\\mu(t)>0$, the measure reports non-Markovianity for a Markovian process and the central claim is refuted.","tokens_in":13292,"feed_emoji":"⏳","tokens_out":10059,"duration_ms":105212,"temperature":0.7,"pith_summary":"The paper claims that two witnesses built from the pseudo-density matrix—a two-time object assembled from Pauli correlation measurements—faithfully detect and quantify non-Markovian memory in any indivisible quantum process. Using counterexamples from the literature, it shows these PDM-based measures register non-Markovianity where trace distance, quantum Jensen-Shannon divergence, and temporal steering do not, including eternally non-Markovian channels that are P-divisible but not completely-positive-divisible. If the claim is right, the measures fill a practical gap: they are optimization-free witnesses of CP-indivisibility and sharpen the distinction between weak quantum direct cause (temporal steering, tied to information backflow) and strong quantum direct cause (temporal non-separability, tied to total memory).","feed_headline":"PDM-based measures catch non-Markovianity other witnesses miss","feed_subtitle":"New temporal-correlation witnesses detect eternal memory effects that trace distance and steering cannot.","key_machinery":"The central object is the two-point pseudo-density matrix $R_{t+\\epsilon,t} = (I \\otimes V_{t+\\epsilon,t})[\\{\\rho \\otimes \\mathbb{1}_2/2, S\\}]$ built from the intermediate dynamical map $V_{t+\\epsilon,t}$, the input state $\\rho$, and $S = \\frac{1}{2}\\sum_{i=0}^3 \\sigma_i\\otimes \\sigma_i$. Its trace norm $\\|R\\|_1$ gives the logarithmic causality measure, and its right-derivative defines the continuous causality measure $\\mu(t)$. Because the PDM is the partial transpose of the corresponding Choi matrix, $\\mu(t)$ diagnoses whether the intermediate map is completely positive, and it does so without optimizing over initial states or measurement bases.","core_discovery":"The central claim is that the logarithmic causality measure and the new continuous causality measure capture complete-positivity indivisibility and quantify total quantum memory, because a pseudo-density matrix is the partial transpose (up to normalization) of the corresponding channel's Choi matrix. The continuous witness is the instantaneous growth rate $\\mu(t) = \\lim_{\\epsilon\\to 0^+} (\\|R_{t+\\epsilon,t}\\|_1 - 1)/\\epsilon$ of the trace norm of the intermediate PDM; a positive $\\mu(t)$ signals a non-completely-positive intermediate map. On the standard eternal non-Markovian channel, trace distance, quantum Jensen-Shannon divergence, and temporal steerable weight all stay flat while the PDM measures detect the always-negative decay rate. The paper reads this as evidence that temporal steerable correlations express a weaker form of quantum direct cause than PDM correlations: the former track P-indivisibility (information backflow), the latter track CP-indivisibility (total memory).","pith_inferences":["The continuous measure could be turned into an experimental calibration tool on current platforms, since it needs only two-time Pauli measurements rather than an entangled ancilla.","Once the normalization of the continuous witness is fixed sharply, the PDM-to-Choi relation could yield direct temporal-correlation estimators for other channel properties, not just non-Markovianity.","The weak/strong direct-cause split suggests a natural grading of non-Markovianity measures: weak witnesses certify information backflow, strong witnesses certify CP-indivisibility, and no single distance measure can do both in full generality."],"forward_implications":["The PDM-based measures provide optimization-free witnesses of CP-indivisibility, unlike temporal steerable weight, which requires solving a semidefinite program.","Eternal non-Markovianity, which is invisible to trace distance, quantum Jensen-Shannon divergence, and temporal steering, becomes detectable and quantifiable.","For the maximally mixed input state, the continuous causality measure reduces to the standard ancilla-based divisibility measure, giving a known benchmark within the new formalism.","The weak-versus-strong direct-cause distinction gives a principled ordering: temporal steering witnesses information backflow (P-indivisibility), while PDM witnesses total memory (CP-indivisibility), so a process can be weakly but not strongly non-Markovian."],"supporting_citations":[{"why":"Supplies the eternally non-Markovian channel that is P-divisible but CP-indivisible, the key test the PDM measures pass.","marker":"[15]"},{"why":"Supplies the purely non-unital channel where trace distance fails and the PDM and steering measures succeed.","marker":"[16]"},{"why":"Defines CP-divisibility and the ancilla-based measure to which the continuous PDM measure reduces for the maximally mixed input.","marker":"[8]"},{"why":"Introduces PDM negativity as a causality measure, the basis of the logarithmic causality measure.","marker":"[23]"},{"why":"Defines the logarithmic causality measure as log trace norm of the PDM.","marker":"[24]"},{"why":"Establishes the PDM as the partial transpose of the Choi matrix, the link that grounds the CP-indivisibility reading.","marker":"[29]"},{"why":"Introduces the temporal steerable weight measure that the paper shows fails on eternal non-Markovianity.","marker":"[35]"},{"why":"Introduces the earlier PDM-based non-Markovianity measure that this work extends to the continuous-time regime.","marker":"[36]"},{"why":"Provides the entropic distinguishability measures (including QJSD) used as benchmarks that miss certain non-Markovian dynamics.","marker":"[57]"}],"fun_headline_variants":["PDM measures catch eternal memory trace distance misses","New witness spots non-Markovianity others can't","Quantum direct cause reveals hidden memory effects","Steering blind to eternal non-Markovianity, PDM sees"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that, under the normalization used in Eq. (13), a completely positive intermediate map always leaves the trace norm of the intermediate pseudo-density matrix non-increasing, so any positive growth rate $\\mu(t)$ is a sure sign of CP-indivisibility.","fun_headline_variants_meta":{"raw":{"variants":["PDM measures catch eternal memory trace distance misses","New witness spots non-Markovianity others can't","Quantum direct cause reveals hidden memory effects","Steering blind to eternal non-Markovianity, PDM sees"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00015,"raw_usage":{"total_tokens":1181,"prompt_tokens":911,"completion_tokens":270,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":527,"completion_tokens_details":{"reasoning_tokens":204}},"tokens_in":527,"tokens_out":270,"duration_ms":3499,"temperature":1.0,"reasoning_tokens":204,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T21:49:59.026609+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Evaluate $\\mu(t)$ from Eq. (14) for a channel known to be CP-divisible, such as pure dephasing with a non-negative decay rate, using an input state other than the maximally mixed one. If any such CP-divisible step yields $\\mu(t)>0$, the measure reports non-Markovianity for a Markovian process and the central claim is refuted.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the eternally non-Markovian channel that is P-divisible but CP-indivisible, the key test the PDM measures pass."},{"cited_title":"Nonuni- tal non-markovianity of quantum dynamics.Phys","cited_arxiv_id":null,"evidence_quote":"Supplies the purely non-unital channel where trace distance fails and the PDM and steering measures succeed."},{"cited_title":"Entanglement and non-markovianity of quantum evolu- tions","cited_arxiv_id":null,"evidence_quote":"Defines CP-divisibility and the ancilla-based measure to which the continuous PDM measure reduces for the maximally mixed input."},{"cited_title":"Quantum correlations which imply causation.Sci- entific reports, 5:18281, 2015","cited_arxiv_id":null,"evidence_quote":"Introduces PDM negativity as a causality measure, the basis of the logarithmic causality measure."},{"cited_title":"Hierarchy in temporal quan- tum correlations.Physical Review A, 98(2):022104, 2018","cited_arxiv_id":null,"evidence_quote":"Defines the logarithmic causality measure as log trace norm of the PDM."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the PDM as the partial transpose of the Choi matrix, the link that grounds the CP-indivisibility reading."},{"cited_title":"Quantify- ing non-markovianity with temporal steering","cited_arxiv_id":null,"evidence_quote":"Introduces the temporal steerable weight measure that the paper shows fails on eternal non-Markovianity."},{"cited_title":"Ob- servation of a full hierarchy of temporal quantum corre- lations with a superconducting qubit","cited_arxiv_id":null,"evidence_quote":"Introduces the earlier PDM-based non-Markovianity measure that this work extends to the continuous-time regime."},{"cited_title":"En- tropic bounds on information backflow.Physical Review Letters, 127(3):030401, 2021","cited_arxiv_id":null,"evidence_quote":"Provides the entropic distinguishability measures (including QJSD) used as benchmarks that miss certain non-Markovian dynamics."}],"review_version":1}