{"id":"20f02273-47ae-4544-b1e7-90adafb63d95","arxiv_id":"2608.03698","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"A fidelity-threshold Lipschitz certificate and recentering algorithm certify a finite perturbation radius for scalar Hamiltonian uncertainty in piecewise-constant quantum control.","lead":"This paper proves certified finite-time robustness margins for quantum gate control when one physical parameter is perturbed. The margins are structure-specific and can rank controllers differently than nominal fidelity or local sensitivity alone.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified: the uniform-bound premise is proven by Lemma 3; the only caveat is numerical certification rigor.","rationale":"The reader's weakest assumption centers on the uniform bound f(μ) ≤ C_hatH. On close reading, this bound is not merely asserted: Lemma 3 plus the affine structure (4) proves it for all μ in the domain, because the derivative of the Hamiltonian with respect to μ is a constant matrix α^(k)R_hatH. The centering argument is also valid, since the trace term only multiplies the propagator by a global phase and therefore does not affect the fidelity or its derivative. Thus the mathematical certificate in Theorem 2 and Algorithm 1 is sound in exact arithmetic. The remaining substantive issue is that the paper's numerical implementation claims 'certified' margins while relying on floating-point fidelity evaluations without error control. That is a legitimate caveat for the reported numbers, but it is not a flaw in the central theorem. The reader's conditions on making the C_hatH derivation explicit and improving artifact accessibility were reasonable; the derivation is correct, and the artifact remains an external dependency. Since no load-bearing mathematical concern survives, the reader's CONDITIONAL verdict can stand unchanged, with the numerical rigor point as an additional minor condition.","tokens_in":9870,"tokens_out":26904,"duration_ms":277102,"concrete_test":"Re-run Algorithm 1 for the three structures H0, H1, H2 using interval-arithmetic or rigorously error-bounded fidelity evaluations (e.g., verified matrix exponentials) and compare the resulting certified M to the reported floating-point M. If any margin differs by more than the paper's own post-hoc bracket (M_upper − M)/M ≤ 5.5e−4, or if a reported floating-point M exceeds the interval-certified value, then the floating-point implementation is not a rigorous certificate; if the values agree to that tolerance, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The reader's weakest assumption—that the uniform bound f(μ) ≤ C_hatH on the safe interval is asserted but unproven—is actually resolved in the text. Lemma 3 gives ||∂U^(k)/∂μ||_F ≤ Δ ||∂H^(k)/∂μ||_F for every μ, and for the affine structure (4) the derivative of H^(k) with respect to μ is α^(k) R_hatH, independent of μ. Hence f(μ) ≤ Δ Σ |α^(k)| ||R_hatH||_F = C_hatH globally, not merely on I. The centering step is also sound: subtracting the trace adds only a parameter-dependent global phase to the propagator, which leaves the trace-amplitude fidelity and its derivative unchanged, while the centered norm is no larger. The proof of Theorem 2 and the Algorithm 1 update then follow as written in exact arithmetic. The one genuine gap is practical rather than mathematical: Algorithm 1's certified steps require the fidelity values F_ν used in (F_ν − F_T)/L_hatH to be exact. The paper reports floating-point evaluations without rigorous error bounds; if a computed F_ν is slightly too high, the step can exceed the true safe radius, and the bisection safeguard only reacts when the numerical evaluation reports F < F_T. This does not invalidate the theorem, but it tempers the meaning of 'certified' for the reported numerical margins.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper considers finite-dimensional closed quantum systems under piecewise-constant coherent control and develops a scalar-parameter robustness margin for trace-amplitude gate fidelity. It derives a differential sensitivity bound (Theorem 1) from the skew-Hermitian nature of propagated perturbations and a Hermitian/skew-Hermitian decomposition, then converts the bound into a threshold-dependent Lipschitz constant on the connected safe component (Lemmas 3–4). The central result, Theorem 2, gives a certified radius r_nu = (F_nu - F_T)/L_hatH around any recentering point nu, and Algorithm 1 iterates this step to move toward the first threshold crossing. A three-qubit gate-control study with 61 controllers compares margins for drift and two control-Hamiltonian uncertainty structures. The main mathematical steps are checkable and the exact-arithmetic certificate is genuine; the principal weakness is the gap between the exact-arithmetic certificate and the floating-point numerical implementation used for the reported margins.","tokens_in":10180,"tokens_out":12005,"duration_ms":141105,"significance":"If the result is taken at face value, the paper provides a useful post-design certification tool: the Lipschitz constant is derived from the declared perturbation structure and the fidelity threshold, with no fitted parameters, and Theorem 2 gives an explicit, constructive safe-step certificate. The proof of the uniform bound f(mu) <= C_hatH, which a reader might initially treat as an assumption, is indeed supplied by Lemma 3 together with the affine structure of Eq. (4). The paper also ships reproducible code and data (Zenodo release), which raises confidence in the numerical experiments. The main mathematical contribution, a structure-specific fidelity-threshold margin with iterative recentering, is sound. The numerical certification gap, discussed below, tempers but does not invalidate the central theorem.","major_comments":[{"comment":"The word 'certified' is load-bearing for the numerical margins reported in Section VI, but Algorithm 1 is certified only in exact arithmetic. The update nu+ = nu +/- (F_nu - F_T)/L_hatH uses Theorem 2 with the current fidelity F_nu, and the theorem requires F_nu to be exact. In the implementation, F_nu is a floating-point evaluation, and the bisection branch (Algorithm 1, lines 8–11) is only a numerical safeguard: it cannot detect or correct a case where a rounded-up F_nu causes the certified step to overshoot the true safe radius. The paper is partially transparent about this ('floating-point evaluations of analytically certified lower bounds'), but the abstract and conclusion continue to describe the margins and steps as 'certified' without this qualification. Please either implement the algorithm with interval/validated arithmetic for the reported numbers, or consistently state that t","section":"Algorithm 1 and Section VI"}],"minor_comments":[{"comment":"After centering, the notation for the perturbation structure is inconsistent: the text defines hat H_mu = hat H_mu - N^{-1}(Tr hat H_mu)I, and later writes 'Writing H_m = Hm - N^{-1}(Tr H_m)I' without the hat. This should be made uniform to avoid confusion between the original and centered structures.","section":"Section V(b)"},{"comment":"The phrase 'the lower half of Lemma 4' is unclear: Lemma 4 states a two-sided Lipschitz inequality. Presumably the intended use is the lower bound |F_b - F_a| <= L |b-a|, which in the proof yields F_gamma(lambda) >= F_nu - L lambda |mu_1-nu|. Please rephrase.","section":"Lemma 4 proof"},{"comment":"The definition of B_T is written as B_T = sqrt((1-F_T^2)/N), but the derivation in Theorem 1 gives B_T = sqrt(1-F_T^2)/N. The square root covers only 1-F_T^2, not the division by N. Please check the typography in this displayed equation; the subsequent algebra appears correct.","section":"Eq. (12)"},{"comment":"The threshold status 'UNSET' is used but never explained. If it is just an internal flag, that is fine; still, a one-line clarification would help reproducibility.","section":"Algorithm 1"},{"comment":"The correlations are reported descriptively without confidence intervals or significance tests. Since the sample size is only 61, the reader cannot assess sampling variability. Adding a bootstrap interval or at least an explicit disclaimer that the entries are descriptive would be appropriate.","section":"Table I"}],"recommendation":"major_revision","confidential_remarks":"The mathematical core of the paper appears sound, and I found no circularity or hidden parameter fitting in Theorem 2. My main reservation is the numerical-certification gap: the abstract and conclusions call the reported finite margins 'certified' while the implementation relies on floating-point fidelity evaluations. This is fixable either by quantifying the rounding error or by systematically softening the certified language for the numerical results. If the authors choose the second route, the paper might be suitable after minor revision; I recommend major_revision because the current wording makes the numerical claim a central part of the contribution."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"For your money: this is a legitimate methodological contribution, not a rehash. The genuinely new piece is the threshold-dependent Lipschitz certificate in Theorem 2 plus the recentering iteration in Algorithm 1. The sensitivity bound extends their earlier static-control lemma to piecewise-constant control and vanishes at perfect fidelity. The uniform bound C_hatH that the reader flagged as an unproven premise is actually proven: Lemma 3 plus the affine structure gives f(mu) <= C_hatH globally, not just on the safe interval. So that concern does not land. The centering argument is also fine: it only changes the propagator by a global phase and the bound tightens. The proofs of Lemmas 1-4 and Theorem 2 are standard and checkable: skew-Hermitian/Hermitian orthogonality, Cauchy-Schwarz, and the continuity-based safe-component argument all go through. The paper also does a good job of stating what the algorithm does not certify: the status flags are honest, the DOMAIN-TRUNC case is clearly labeled as only a lower bound, and no claim is made about disconnected safe sets. The numerical validation uses a versioned, frozen toolbox with data and scripts, which is more than most papers do. The factor-of-three ranking observation is interesting but descriptive: it shows that nominal sensitivity alone doesn't rank controllers for control-structure perturbations, at least for these 61 controllers on one spin-chain family. That's a modest empirical claim, and the authors don't oversell it. The real soft spot is practical rather than mathematical. Algorithm 1's certified steps require exact fidelity values, but the implementation uses floating-point evaluation without rigorous error bounds. If a computed F_nu is slightly too high, the step can overshoot the true safe radius; the bisection safeguard only reacts after the numerical evaluation reports F < F_T. That tempers the word 'certified' for the reported margins. The post hoc upper witness brackets them to about 5e-4, which suggests the error is small in practice, but it's still a gap between the exact-arithmetic theorem and the floating-point numbers in Table I. A short discussion of interval-arithmetic or a posteriori error bounds would close it. Minor: the single numerical example limits generality, and the toolbox not being independently run means I can't fully verify the numbers, but the frozen artifact is a good start. Who this is for: quantum control people doing post-design robustness certification, and anyone building on norm-based fidelity bounds. It deserves a serious referee. I'd send it out, with a request that the referee check the numerical certification caveat and see whether the recentering iteration can be made rigorous with validated numerics. In my own work I'd cite it for the threshold-dependent Lipschitz certificate.","headline":"The certificate math is sound and the recentered continuation is a real extension; the only soft spot is that the reported numerical margins are not rigorous floating-point certificates, though the authors are mostly upfront about that.","tokens_in":10665,"tokens_out":1374,"would_cite":true,"duration_ms":18229,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper certifies a finite perturbation radius for quantum gate controls: the fidelity gap divided by a Lipschitz bound is a safe step, and repeating it gives a proven lower bound on the distance to the first threshold crossing.","keywords":["quantum control","robustness margin","structured perturbations","gate fidelity","Lipschitz bound","piecewise-constant control","trace-amplitude fidelity","fidelity threshold"],"falsifier":"Pick one of the 61 controllers, compute the certified margin M for the H2 structure, and evaluate the actual fidelity at parameter offset M - 10^-6. If the fidelity falls below F_T, the certificate is violated. Equivalently, numerically maximize f(mu) = sum_k ||partial-tilde-U(k)/partial-mu||_F over the safe interval; any value exceeding C_hatH refutes the uniform-bound premise.","tokens_in":9796,"feed_emoji":"⚛️","tokens_out":5845,"duration_ms":60482,"temperature":0.7,"pith_summary":"The paper aims to answer a practical question: for a quantum gate implemented by a piecewise-constant pulse sequence, how far can a scalar physical parameter (a drift error or a control amplitude error) move before the gate fidelity drops below an acceptable threshold? It proves that the answer can be certified using only the nominal controller, the threshold, and a structure-dependent constant: at any point the fidelity gap above threshold divided by a Lipschitz constant is a radius inside which the fidelity is guaranteed to stay above threshold, and recentering and repeating this step walks toward the true boundary. The bound is derived from a trace-amplitude fidelity sensitivity inequality that vanishes at perfect fidelity and is then made global by a uniform bound on the perturbed propagators. A three-qubit gate-control study shows the resulting margins depend on which Hamiltonian is perturbed and can disagree with rankings from nominal fidelity or local sensitivity by up to a factor of three. If correct, this gives a cheap, structure-aware certification tool for post-design gate robustness.","feed_headline":"Certified radius keeps quantum gates above fidelity threshold","feed_subtitle":"A Lipschitz bound turns one-parameter quantum control errors into provable safe steps toward the first fidelity failure.","key_machinery":"The central object is the threshold-dependent Lipschitz constant L_hatH = B_T C_hatH, where B_T = sqrt((1-F_T^2)/N) and C_hatH bounds the supremum over the safe interval of the sum of perturbed propagator Frobenius norms. The bound is obtained by decomposing the trace-amplitude fidelity derivative into Hermitian and skew-Hermitian parts of the propagator factors, cancelling the Hermitian contribution, and bounding the remainder with Lemma 2; this gives a local sensitivity inequality that is converted into a global Lipschitz bound on the connected safe component. The certified step r_nu = (F_nu - F_T)/L_hatH is the mechanism of Algorithm 1: recentering at each new point makes the certificate","core_discovery":"Given a nominal controller with fidelity F greater than the threshold F_T, the paper proves that for any parameter value nu in the connected safe interval, the step r_nu = (F_nu - F_T)/L_hatH is a certified safe radius: every parameter value within r_nu of nu has fidelity at least F_T, and strictly inside it stays within the safe component. The Lipschitz constant L_hatH = B_T C_hatH comes from bounding the derivative of the trace-amplitude gate fidelity by sqrt((1-F_T^2)/N) times a structure-dependent constant C_hatH that bounds the sum of the perturbed propagator Frobenius norms. C_hatH is explicit for affine scalar structures: t_f||H_0||_F for drift uncertainty and Delta||f_m||_1||H_m||_F","pith_inferences":["Since each step in the case study is certified and the post-hoc bracket between the reported margin and an upper witness is tight, the iteration likely reaches near the true boundary for many controllers; a testable extension is to use the directional statuses to produce two-sided brackets without extra bisection.","The same recentering argument could be adapted to simultaneous scalar parameters by cycling coordinates or replacing the interval Lipschitz constant with a gradient-norm bound over a rectangle, but the paper explicitly leaves multiparameter extensions open.","The structure-dependent ranking suggests margin-aware synthesis: optimizing the certified margin as part of the controller objective might select controllers that nominal fidelity alone would rank differently; that goes beyond what the paper demonstrates.","The reported anti-correlation between nominal drift sensitivity magnitude and finite drift margin implies local differential sensitivity can be misleading as a robustness proxy; checking this on other spin-chain lengths or gate targets would be a cheap falsifiable follow-up."],"forward_implications":["For any controller meeting F_T < F_mu0, the certificate gives a proven safe interval around the nominal parameter without exhaustive search over perturbation directions for a single scalar structure.","Iterating the recentered step yields a certified lower bound on the distance to the first threshold crossing in each direction, usable for post-design certification of gates.","The derivative bound vanishes at unit fidelity, so near-perfect gates inherit an O(epsilon^-1/2) scaling of the log-sensitivity of fidelity error, matching static-field observations; the margin itself remains finite and threshold-dependent.","Because L_hatH depends only on the threshold and the perturbation structure, not on the current point, the algorithm can be run with only local fidelity evaluations.","Margins for drift versus control Hamiltonians rank controllers differently: the drift margin anticorrelates with nominal sensitivity while control-structure margins do not, so finite margins carry off-nominal information not present in nominal fidelity or local sensitivity.",""],"supporting_citations":[{"why":"Supplies the earlier conservative norm-based sensitivity bounds and iterative local worst-case searches that the certified continuation step improves on.","marker":"[6]"},{"why":"Gives the derivative formula for quantum control sensitivities and the continuity argument on which the safe-set structure depends.","marker":"[7]"},{"why":"Provides the static-control lemma, vanishing at unit fidelity, that Corollary 1 extends to piecewise-constant controls.","marker":"[15]"},{"why":"Documents the divergence of log-sensitivity of fidelity error for static fields, with which the O(epsilon^-1/2) scaling check is consistent.","marker":"[16]"},{"why":"Defines the three-spin Heisenberg-chain gate-control model used for the case study.","marker":"[17]"},{"why":"Freezes the toolbox, controller data, and reproduction scripts that produce the figures and Table I; load-bearing for the numerical claims.","marker":"[18]"}],"fun_headline_variants":["Lipschitz bound certifies safe quantum gate steps","Certified radius for quantum control fidelity","Provable safe radius for quantum control","Fidelity threshold radius for finite-time quantum control"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The whole certificate rests on the constant C_hatH being a true upper bound on the sum of perturbed propagator norms everywhere in the connected safe interval; if that uniform bound fails at even one interior point, the certified radius can be too large.","fun_headline_variants_meta":{"raw":{"variants":["Lipschitz bound certifies safe quantum gate steps","Certified radius for quantum control fidelity","Provable safe radius for quantum control","Fidelity threshold radius for finite-time quantum control"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000695,"raw_usage":{"total_tokens":2948,"prompt_tokens":681,"completion_tokens":2267,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":425,"completion_tokens_details":{"reasoning_tokens":2210}},"tokens_in":425,"tokens_out":2267,"duration_ms":17505,"temperature":1.0,"reasoning_tokens":2210,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T14:31:10.241141+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Pick one of the 61 controllers, compute the certified margin M for the H2 structure, and evaluate the actual fidelity at parameter offset M - 10^-6. If the fidelity falls below F_T, the certificate is violated. Equivalently, numerically maximize f(mu) = sum_k ||partial-tilde-U(k)/partial-mu||_F over the safe interval; any value exceeding C_hatH refutes the uniform-bound premise.","supporting_citations":[{"cited_title":"Robustness of dynamic quantum control: Differential sensitivity bounds,","cited_arxiv_id":null,"evidence_quote":"Supplies the earlier conservative norm-based sensitivity bounds and iterative local worst-case searches that the certified continuation step improves on."},{"cited_title":"Sensitivity bounds for quantum control and time-domain performance guarantees,","cited_arxiv_id":null,"evidence_quote":"Gives the derivative formula for quantum control sensitivities and the continuity argument on which the safe-set structure depends."},{"cited_title":"Geometric interpre- tation of sensitivity to structured uncertainties in spintronic networks,","cited_arxiv_id":null,"evidence_quote":"Provides the static-control lemma, vanishing at unit fidelity, that Corollary 1 extends to piecewise-constant controls."},{"cited_title":"Time-domain sensitivity of the tracking error,","cited_arxiv_id":null,"evidence_quote":"Documents the divergence of log-sensitivity of fidelity error for static fields, with which the O(epsilon^-1/2) scaling check is consistent."},{"cited_title":"Robust quantum gates for open systems via optimal control: Markovian versus non- Markovian dynamics,","cited_arxiv_id":null,"evidence_quote":"Defines the three-spin Heisenberg-chain gate-control model used for the case study."},{"cited_title":"Fidelity-based quantum robustness margins,","cited_arxiv_id":null,"evidence_quote":"Freezes the toolbox, controller data, and reproduction scripts that produce the figures and Table I; load-bearing for the numerical claims."}],"review_version":1}