{"id":"ad0eb2d0-73dd-4589-aac8-0639214a2f3a","arxiv_id":"2505.02286","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For Bell-diagonal states with a<=1/3, successful recurrence EPP guarantees fidelity at least the average input; optimal purification time is earliest or latest depending on noise and figure of merit.","lead":"This paper proves when the standard entanglement purification protocol is guaranteed to improve noisy entangled pairs, and when to run it in time to maximize final fidelity under memory noise. It shows the best purification time can flip from earliest to latest depending on the type of decoherence.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Prop. III.4's strict inequalities fail at F1=F2=1/2; the appendix proofs support non-strict versions, so the proposition needs correction but the central threshold result survives.","rationale":"I focused on the central guaranteed-improvement proposition rather than the memory-noise model identified by the reader. The most concrete correctness issue is the strict-inequality wording in Prop. III.4: at the boundary F1=F2=1/2 the successful-output fidelity equals the AIS fidelity for every value of a, directly contradicting statement 1's 'greater than' and statement 3's 'there does not exist ... >='. The appendix proves non-negative / non-positive differences, so the technical machinery is correct and the intended threshold at a=1/3 remains intact. This is a real but localized error in the statement of a central proposition; it requires a wording correction rather than a change to the scientific conclusions. The optimal-time results, the a=1/3 transition, and the scoped non-strict guaranteed-improvement claim are unaffected. I therefore do not regard this as a verdict-changing objection, but it should be fixed before publication.","tokens_in":38219,"tokens_out":27440,"duration_ms":321613,"concrete_test":"Evaluate F_avg_incr_BDS from Eq. (B20) at F1=F2=1/2 for a=0, a=1/3, and a=0.75. Direct substitution gives numerator 1/4 + a^2/4 and denominator [(1+a)^2 + (1-a)^2]/4 = (1+a^2)/2, so FBDS = 1/2 and F_avg_incr_BDS = 0 in every case. This settles that Prop. III.4(1)'s 'greater than' and Prop. III.4(3)'s 'does not exist ... >=' are false as written, and that changing them to 'no less than' and 'no greater than', respectively, matches the appendix proofs.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In Sec. III.C, Prop. III.4(1) states that the successful-output fidelity FBDS is always greater than the average-input fidelity (F1+F2)/2 for a<=1/3, and Prop. III.4(3) states that for a>1/2 there does not exist (F1,F2) in [1/2,1]^2 with FBDS >= (F1+F2)/2. Both statements are false as printed. At F1=F2=1/2, Eq. (1a) with the common error parameter a gives FBDS = (1/4 + a^2/4) / [(1+a)^2/4 + (1-a)^2/4] = 1/2 for every a. Thus for a<=1/3 the output equals the AIS fidelity rather than exceeding it, and for a>1/2 the pair (1/2,1/2) is an example with FBDS = (F1+F2)/2, contradicting the claimed nonexistence. The appendix proof (Eq. B20 and the surrounding arguments in Appendix B.5) actually establishes F_avg_incr_BDS >= 0 for a<=1/3 and F_avg_incr_BDS <= 0 for a>=1/2, i.e. the non-strict versions of the inequalities. The intended content — that the AIS fidelity is the correct crossover at a=1/3 and that no strict improvement is possible for a>=1/2 — is sound; the printed strict-inequality wording is the error.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper analyzes the BBPSSW/DEJMPS recurrence entanglement purification protocol for Bell-diagonal states in quantum network settings. It asks (i) what improvement is guaranteed conditioned on success when the two input states are non-identical, and (ii) what is the optimal time to run the EPP when the input pairs are generated at different times and stored in decohering quantum memories. The main results are: rank-2 bit-flip inputs always improve over the better input; Werner-state inputs are guaranteed to improve over the average-input-state fidelity; general BDS inputs with a common error parameter a are guaranteed to improve over the AIS fidelity for a≤1/3 and never for a>1/2; for bit-flip memory noise the optimal purification time is the latest possible time (or the earliest for normalized fidelity and concurrence), while for depolarizing noise it is a computable intermediate time; and a numerical/analytical study of the transition between early and late optimal times for general Pauli channels. Detailed proofs are provided in Appendices A–D, with robustness and transition-border analysis in Appendices E–F.","tokens_in":38513,"tokens_out":14542,"duration_ms":160637,"significance":"If the results are made fully rigorous, the paper is a useful contribution to the quantum-network EPP scheduling literature. The clean thresholds a=1/3 and a=1/2, the parameter-independent optimal-time statements, and the explicit transition-border approximation give practically actionable guidance and are derived without fitted parameters. The paper is also careful to distinguish its guaranteed-improvement claims from the universality no-go results of arXiv:2407.21760. The main value is analytical: it identifies exactly when postselection on the recurrence protocol guarantees improvement and when delaying purification is beneficial, which is significant for repeater and memory-buffer design. However, the current manuscript contains several statement-level errors and a domain error in one of the optimal-time theorems that must be corrected before the results can be accepted as stated.","major_comments":[{"comment":"Statements 1 and 3 of Prop. III.4 are false as printed. At F1=F2=1/2, Eq. (1a) gives FBDS = (1/4 + a^2/4) / ((1+a)^2/4 + (1-a)^2/4) = 1/2 for every a. Hence for a≤1/3 the output fidelity equals the AIS fidelity rather than being strictly greater, and for a>1/2 the pair (1/2,1/2) satisfies FBDS=(F1+F2)/2, contradicting the claimed nonexistence of (F1,F2) with FBDS≥(F1+F2)/2. The appendix proof (Eq. B20 and Sec. B.5) establishes the non-strict forms F_avg_incr_BDS≥0 for a≤1/3 and F_avg_incr_BDS≤0 for a≥1/2. Please correct the proposition to non-strict inequalities and specify the equality cases.","section":"Sec. III.C, Prop. III.4"},{"comment":"The branch \"if t1<t1^*, perform EPP at t*\" is incomplete because t* is independent of t2 and may lie outside the feasible interval [t1,t2]. The derivative calculation in Eq. (D20) shows that the end-time fidelity increases on [t1,t*] and decreases after t*. Therefore, if t2<t*, the constrained maximum is at t2, not at t*. The proposition needs an additional condition, such as \"and t2≥t*\", plus a branch for t2<t* (perform at the latest possible time t2). Relatedly, the proposition and Eq. (9) should state that the threshold t1^* is positive only for F0>(3√3−2)/4≈0.799; for smaller F0 the earliest-time branch applies for all t1.","section":"Sec. IV.B, Prop. IV.4 and Appendix D.4"},{"comment":"The proof of the fidelity threshold Fth≈0.939 depends on numerically locating the root F1,root≈0.879 of E_low-up_D(F1,1) (Eq. B16). As stated, Prop. III.3 is a guarantee for all F1,F2≥Fth, but a purely numerical root location does not by itself provide a rigorous threshold. Please provide a rigorous enclosure of the root (for example, interval arithmetic or a Sturm-sequence argument) and a complete justification of the \"simple geometric argument\" that reduces the rectangle to the two boundaries examined, or explicitly present the threshold as a numerically supported conjecture rather than a proven proposition.","section":"Sec. III.B, Prop. III.3 and Appendix B.4"}],"minor_comments":[{"comment":"The sentence \"which is obviously negative for a<0\" should read \"for a>0\"; additionally, the example uses F1=1, where the ratio a=λ2/(1−F1) is not defined, so it should be flagged as a limiting case.","section":"Sec. III.C"},{"comment":"The phrase \"guaranteed to be higher than at most the average input fidelity\" is misleading; the proven statement is that the output fidelity is at least the AIS fidelity.","section":"Sec. III.B"},{"comment":"There are several typos that should be corrected: \"twriling\" in Prop. III.3, \"lastest\" in Props. IV.1 and IV.3, and \"red/w A\" in the first-page affiliation line.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is worth pursuing. The most serious issue is the missing upper-bound constraint in Prop. IV.4; I verified the flaw by taking F0=0.95, κ=1, t1=0.01, t2=0.1, for which t*>t2 and the claimed optimal point is infeasible. The authors should also fix the strict/non-strict boundary in Prop. III.4. With these corrections, I expect the paper to be acceptable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a solid analytical paper on the CNOT-based recurrence EPP, and the main theorems are new and essentially correct. It deserves a serious referee. But there is a small bug in Prop. III.4 that needs fixing before publication: at F1=F2=1/2 the output fidelity exactly equals the average input fidelity for every a, so the strict inequalities in statements 1 and 3 are false as printed. The proofs in Appendix B actually give the non-strict versions, so the transition at a=1/3 and the no-improvement region a>1/2 survive intact.\n\nWhat is genuinely new: the guaranteed-improvement results for non-identical inputs with a common error model, including the clean a=1/3 crossover and the a>1/2 no-improvement region, and the optimal-time theorems for bit-flip and depolarizing memory noise. The parameter-independent \"purify at the latest (or earliest) possible time\" results are nice, and the numerical transition phenomenon with an approximate analytic border is a good addition. The appendices are explicit and mostly self-contained, with parameter-free derivations from the standard recurrence equations; the single self-citation to the no-go paper is used to position the result, not as an assumption, which is fine.\n\nSoft spots, in order of severity. (1) The strict-inequality bug in Prop. III.4: at F1=F2=1/2 the output fidelity equals (F1+F2)/2 for all a, so the printed \"greater than\" and \"does not exist ... with >=\" are both wrong. The appendix proves non-strict versions, and the intended qualitative claims are unchanged, but the proposition and the surrounding text need rewording. (2) Prop. III.3 uses a numerically located root (F1,root≈0.879) to set the threshold Fth≈0.939; that is honest and easily checkable, but it is not fully analytic. (3) In Prop. IV.4, t1^* is negative for F0 below about 0.8, making the \"if t1 ≥ t1^*\" condition vacuous there; the statement is still true, but this should be stated explicitly. (4) Memory decoherence is modeled as independent identical Pauli channels; the paper says so clearly, and the transition phenomenon may be sensitive to that assumption, which is a scope limitation rather than a flaw.\n\nWho this is for: anyone working on quantum repeater scheduling, buffer-time optimization, or practical EPP policy. It is not a framework paper, but it gives useful, citable rules of thumb. My recommendation: send it to peer review. The central claims hold up, the appendix proofs are careful, and the strict-inequality bug is a correctable wording issue, not a load-bearing flaw.","headline":"Solid, mostly-correct analytical results on guaranteed EPP improvement and optimal scheduling; one proposition has a strict-inequality bug at the F=1/2 boundary that is easy to fix.","tokens_in":39026,"tokens_out":3441,"would_cite":true,"duration_ms":42272,"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":"This paper proves when the CNOT-based recurrence EPP improves entanglement and fixes the best time to run it under memory decoherence.","keywords":["entanglement purification","quantum networks","Bell-diagonal states","quantum memory decoherence","optimal scheduling","Pauli channels","fidelity improvement","entanglement measures"],"falsifier":"Set up the bit-flip scenario of Proposition IV.1 with $F_0=0.95$, $t_1=0.01/\\kappa$, $t_2=0.1/\\kappa$, and measure the fidelity at $t_2$ after successful purification at several times $t$; if the latest $t$ does not give the highest fidelity, the claimed parameter-independent optimal time fails.","tokens_in":38008,"feed_emoji":"⚛️","tokens_out":6953,"duration_ms":80790,"temperature":0.7,"pith_summary":"This paper asks when a two-to-one entanglement purification step is worth doing and when it should be scheduled in a quantum network where one pair arrives before another and both memories decay. It proves that the standard CNOT-based recurrence protocol, on two Bell-diagonal states of the same error shape, always improves output fidelity over the average input state as long as the noise parameter $a$ is at most $1/3$, and that for $a$ above $1/2$ such improvement never happens. For memory noise that is a single detectable Pauli channel (bit-flip), it proves the best time to purify is the latest possible time before the pair is used, independent of arrival times, raw fidelity, or decoherence rate. For depolarizing memory noise the optimal time depends on parameters and is given explicitly, and for general Pauli memory noise the strategy switches sharply between purify-immediately and purify-last, with an approximate border formula. The results give network operators simple timing rules and clarify what the protocol can guarantee when inputs are non-identical.","feed_headline":"Purify late: bit-flip memory noise favors the last moment","feed_subtitle":"A proof fixes the best schedule for CNOT recurrence purification and maps where the strategy flips by noise type.","key_machinery":"The load-bearing object is the recurrence EPP on Bell-diagonal states, where two noisy pairs are combined with a bilateral CNOT and a parity measurement, succeeding when the two measurement outcomes agree. For such inputs the protocol reduces to a four-line map on the Bell probabilities $\\lambda_1,\\dots,\\lambda_4$: on success, $\\lambda_1$ becomes $(\\lambda_1\\lambda_1'+\\lambda_2\\lambda_2')/p$, and analogous products replace the other components, with $p$ the success probability. The proof machinery combines that map with a one-parameter family of input states $\\lambda_1=F$, $\\lambda_2=a(1-F)$, $\\lambda_3+\\lambda_4=(1-a)(1-F)$, and with a continuous-time Pauli channel model for memory decoherence whose action on Bell-diagonal states is a $4\\times 4$ matrix. The parameter $a$ and the monotonicity of entanglement measures in fidelity carry the guaranteed-improvement results; the Pauli channel matrix carries the optimal-time results, which are reduced to signs of derivatives of the fidelity at $t_2$.","core_discovery":"The central claim is that the CNOT-based recurrence EPP has a precise, provable improvement boundary and a robust optimal schedule. For Bell-diagonal inputs with identical error pattern, letting $a$ be the fraction of the non-fidelity weight in the $\\Phi^-$ component, the successful output fidelity is always at least the average input fidelity when $a\\le 1/3$; for $a>1/3$ there exist inputs where purification lowers fidelity relative to that average, and for $a>1/2$ purification never beats it. Under bit-flip memory decoherence, the fidelity at the utilization time conditioned on successful purification is monotonically increasing in the purification time, so the EPP should be run as late as possible, with no dependence on $t_1$, $t_2$, $F_0$, or $\\kappa$; if success probability is included in a normalized figure of merit, the optimum flips to as early as possible for fidelity, concurrence and negativity, while normalized distillable entanglement again favors the latest time. Under depolarizing memory decoherence a parameter-dependent intermediate optimal time exists with an explicit formula. For general continuous-time Pauli memory channels the optimal time undergoes a transition between earliest and latest, and the paper gives an approximate $t_1,t_2$-independent expression for the transition border as a function of raw fidelity.","pith_inferences":["If the transition border formula is as general as the numerics suggest, small errors in characterizing the X/Y/Z balance of a quantum memory can flip a repeater from purify-at-end to purify-immediately; this sharp sensitivity could be tested in simulation without changing the noise model.","The guaranteed-improvement threshold $a\\le 1/3$ suggests a simple protocol-level diagnostic: estimate $a$ by twirling and measuring the $\\Phi^-$ component; below $1/3$ the average-input baseline is safe, while above it the protocol should be combined with a different error-correction step or a different EPP.","Extending the derivative analysis beyond Pauli channels, for example to amplitude-damping noise, may remove the parameter-independence; the paper also leaves classical communication delay out, so including a fixed two-way communication time should shift the optimal schedule by that delay, which is a direct testable extension."],"forward_implications":["A network operator who knows its memory decoherence is bit-flip-dominated can set the purification schedule without tracking $t_1$, $t_2$, $F_0$, or $\\kappa$: run the EPP as late as possible before use.","If the metric includes the success probability (normalized fidelity, concurrence, negativity), the same operator should purify as early as possible; normalized distillable entanglement reverses this again because of convexity.","For depolarizing memories, there is a unique intermediate time $t^*$ given by an explicit formula, and whether to wait depends on how long after the first pair the second pair arrives.","For a general Pauli memory channel, the optimal strategy is extreme, either immediately or at the end, with a sharp transition border approximated by $B\\approx (8F_0^2-4F_0+5)/(20F_0^2-4F_0+2)$; misidentifying the noise pattern can select the wrong extreme strategy and even make EPP worse than discarding the old pair."],"supporting_citations":[{"why":"Defines the recurrence protocol whose CNOT-and-parity circuit and success condition are the object under study.","marker":"[23]"},{"why":"Defines the DEJMPS recurrence variant and its fidelity transformations.","marker":"[24]"},{"why":"Supplies the standard derivation of the success-branch transformation for Bell-diagonal states used in Eqs. (1)-(2).","marker":"[25]"},{"why":"No-go theorems for universal entanglement purification; the paper's baselines are chosen to avoid violating these universality limits.","marker":"[58]"},{"why":"Provides evidence that the CNOT-based recurrence protocol is optimal for identical inputs, motivating the choice of protocol.","marker":"[59]"},{"why":"Pauli twirling justifies restricting inputs to Bell-diagonal states without changing their fidelity.","marker":"[64]"},{"why":"Coherent information is used as the lower bound on distillable entanglement in the threshold proofs.","marker":"[72]"},{"why":"Rains bound is used as the upper bound on distillable entanglement in the Werner-state threshold proof.","marker":"[73]"}],"fun_headline_variants":["Purify late: bit-flip memory noise favors the final step","Optimal purification time flips with error type","Guaranteed improvement: purification has a provable cutoff","Bit-flip memory noise: purify at the very end","Purification schedule: early or late depends on noise"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that memory decoherence is a pair of independent, identical continuous-time Pauli channels, one on each node's quantum memory; if the noise is non-Pauli, correlated, or has memory, the parameter-independent optimal times and the transition-border formula need not hold.","fun_headline_variants_meta":{"raw":{"variants":["Purify late: bit-flip memory noise favors the final step","Optimal purification time flips with error type","Guaranteed improvement: purification has a provable cutoff","Bit-flip memory noise: purify at the very end","Purification schedule: early or late depends on noise"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000756,"raw_usage":{"total_tokens":3426,"prompt_tokens":1073,"completion_tokens":2353,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":689,"completion_tokens_details":{"reasoning_tokens":2273}},"tokens_in":689,"tokens_out":2353,"duration_ms":25648,"temperature":1.0,"reasoning_tokens":2273,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T00:55:55.862630+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Set up the bit-flip scenario of Proposition IV.1 with $F_0=0.95$, $t_1=0.01/\\kappa$, $t_2=0.1/\\kappa$, and measure the fidelity at $t_2$ after successful purification at several times $t$; if the latest $t$ does not give the highest fidelity, the claimed parameter-independent optimal time fails.","supporting_citations":[{"cited_title":"Rozpędek, T","cited_arxiv_id":null,"evidence_quote":"Provides evidence that the CNOT-based recurrence protocol is optimal for identical inputs, motivating the choice of protocol."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Pauli twirling justifies restricting inputs to Bell-diagonal states without changing their fidelity."},{"cited_title":"Schumacher and M","cited_arxiv_id":null,"evidence_quote":"Coherent information is used as the lower bound on distillable entanglement in the threshold proofs."}],"review_version":1}