{"id":"ecf99034-ece9-4973-b0c8-effd7e444428","arxiv_id":"2412.12762","paper_version":5,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Eigenvalue perturbation under bit- and phase-flips is studied for matrices written as linear combinations of permutations, with exact invariance of the dominant eigenvalue for positive coefficients.","lead":"This paper analyzes how bit-flip and phase-flip errors affect the eigenvalues of quantum operators built as sums of permutation matrices, and reports that positive-coefficient sums are resilient. It proposes design guidelines for error-resilient block-encoded quantum algorithms, but the error model uses the average operator rather than single-run errors.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Dominant-eigenvalue invariance is correct, but it does not imply application-level resilience: the expected-operator model of Eq. (5) can equal A exactly while actual bit-flip runs make the success probability drop by p(1-p).","rationale":"The reader's weakest-assumption diagnosis is correct, and the counterexample above makes it quantitative. I agree the dominant-eigenvalue theorem is valid; in fact, every fixed error realization has the same lambda_max. The failure is the leap from spectral data to application correctness. Because f_overlap in Eq. (14) uses the first-moment operator B = A(p,b), it can miss all errors when they cancel in expectation, as in the I+X example. The paper's numerics use 30 random trials without error bars, so they cannot reveal this structural issue. The Appendix Gershgorin bound is loose, but the main concern is the overinterpretation of the spectral result. The correct fix is to scope the claim as a spectral statement about the expected operator, or to replace Eq. (14) with a proper average over random error realizations of the relevant success probability. I would keep the reader's CONDITIONAL verdict because the mathematical core is salvageable and the code is public, but the revision must remove or reframe the application-resilience claim.","tokens_in":11686,"tokens_out":14521,"duration_ms":150652,"concrete_test":"Run the paper's public simulator on the counterexample: n=1, K=2, alpha = [1,1], Pi_1 = I, Pi_2 = X, and for p in [0,0.5] compute (i) the eigenvalue errors in Eqs. (12)-(13) and f_overlap in Eq. (14) from A(p,b), and (ii) the exact average success probability sum_e Prob(e) |<+|B_e|0>|^2 / ||B_e|0>||^2 over the four error patterns. If (i) is identically zero/one while (ii) equals 1 - p + p^2, the expected-operator analysis cannot support the application-resilience conclusion.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Sec. 2.2.1's spectral fact is correct: for positive alpha_i, every bit-flipped term X_b Pi_i is still a permutation, so every error realization A_e has column sums sum_i alpha_i and hence lambda_max(A_e) = sum_i alpha_i. The unsupported step is the inference from this spectral invariance to the statement that 'applications ... are resilient to quantum bit-flip errors.' Eq. (5) is only the first moment of the random error operators, while measurement probabilities are second moments, so the expected operator cannot determine whether an algorithm succeeds. A concrete counterexample within the paper's own model: n=1, K=2, alpha_1 = alpha_2 = 1, Pi_1 = I, Pi_2 = X, with an independent bit-flip of probability p on each term. The expected operator A(p,b) is exactly A = I + X for every p, so Eqs. (12)-(14) report zero error. In an actual run, with probability 2p(1-p) the applied operator is 2I or 2X; on input |0> and measurement in the |+>/|-> basis, those runs succeed with probability 1/2 instead of 1, giving average success probability 1 - p + p^2. Thus the central application-resilience claim is not established; what is established is only invariance of the dominant eigenvalue of the expectation operator.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper considers matrices expressed as linear combinations of permutation matrices, A = sum_i alpha_i Pi_i, and studies how bit-flip and phase-flip errors perturb the eigenvalues. The author introduces a probabilistic error model in which each permutation term is replaced by a mixture of the identity and an X or Z operation, defines the expected operator A(p,b), and proves that for positive coefficients the dominant eigenvalue of this expected operator is invariant under bit-flips because every error realization preserves column sums. The paper also gives Gershgorin-type bounds, reports numerical experiments on eigenvalue errors and output-state fidelities, and discusses implications for block-encoding implementations and for algorithms whose operators have this form.","tokens_in":11906,"tokens_out":7254,"duration_ms":68415,"significance":"The core spectral observation is correct and easy to verify: for positive coefficients, every bit-flipped permutation term is still a permutation, so the all-ones vector remains an eigenvector with eigenvalue sum_i alpha_i. The paper is also commendably concrete: the model is explicit, the numerical experiments are documented, and the code is made available. However, the paper's central applied claim, that such operators are resilient to quantum bit-flip errors, is not established by the given analysis. The problem is that the model in Eq. (5) is the expectation of the random error operators, whereas a single run applies one error realization; measurement probabilities are second moments, so the expected operator does not determine algorithm success. The same gap propagates into the fidelity analysis of Sec. 3. Because the main advertised contribution depends on this inference, the manuscript needs substantial revision before it can be considered sound.","major_comments":[{"comment":"The proof that the dominant eigenvalue of the averaged operator in Eq. (5) equals sum_i alpha_i under bit-flips is correct, but the subsequent claim that 'applications ... are resilient to quantum bit-flip errors' does not follow. In a single run of the circuit each term applies either Pi_i or X_{b_i}Pi_i; Eq. (5) is only the first moment of the error distribution, whereas success probabilities are second moments. Concretely, for n=1, K=2, alpha_1=alpha_2=1, Pi_1=I, Pi_2=X, and independent bit-flip probability p on each term, A(p,b)=I+X for every p, so Eqs. (12)-(14) and Fig. 1 report zero error. In an actual run, with probability 2p(1-p) the applied operator is 2I or 2X, and on input |0> with measurement in the |+>/|-> basis the average success probability is 1-p+p^2, which is less than 1 for p>0.","section":"Sec. 2.2.1 and Abstract"},{"comment":"The statement that a bit flip on the control register swaps the coefficients of the permutations while leaving the eigenvalue spectrum unchanged is false in general. For example, with Pi_1=I, Pi_2=X, alpha_1=1, alpha_2=2, swapping the coefficients changes A=I+2X into A'=2I+X; the eigenvalue sets are {3,-1} and {3,1}, respectively. The claim should be deleted or restricted to special cases where the operator is actually invariant, such as when the swapped coefficients are equal.","section":"Sec. 2.2.2"},{"comment":"The fidelity quantities f_overlap and f_re are computed against the output of the expected operator A(p,q,b,phi), not against the distribution of actual error operators applied in individual runs. Consequently, the numerical fidelity curves in Figs. 4-7 inherit the same gap identified in Sec. 2.2.1: they can remain at 1 for erroneous realizations such as 2I and 2X in the counterexample above, while the probability of obtaining the correct measurement outcome decreases. The paper should replace or supplement these quantities with the average fidelity over the random per-term bit-flip and phase-flip distribution, e.g., E_B[|<psi|A^dagger B|psi>|^2 / (||B|psi>||^2 ||A|psi>||^2)] for B drawn from the error model.","section":"Sec. 3, Eqs. (14)-(15)"},{"comment":"The claim that the Perron-Frobenius dominant eigenvalue is 'isolated' with a gap to the second largest eigenvalue is not guaranteed for general nonnegative matrices; a reducible nonnegative matrix can have multiple eigenvalues on the spectral circle. The correct and sufficient statement for the paper's purpose is that the spectral radius is sum_i alpha_i because the all-ones vector is a right eigenvector with that eigenvalue and the column sums are constant. The simplicity claim should be removed or proved under an extra primitivity assumption.","section":"Sec. 2.2.1"}],"minor_comments":[{"comment":"The description of phase-flips as 'potentially reversing eigenvalue contributions (e.g., alpha_i -> -alpha_i)' does not match Eq. (6), where a Z gate flips signs of selected basis amplitudes inside each permutation term rather than changing the coefficients alpha_i globally; the wording should be aligned with the model.","section":"Sec. 2.3"},{"comment":"The statement that any square matrix can be expressed as a linear combination of permutations multiplied by two diagonal scaling matrices 'via Sinkhorn's theorem' overstates the classical result: Sinkhorn's theorem concerns positive matrices, and arbitrary complex or indefinite matrices require additional conditions or an extended decomposition. The sentence should be qualified.","section":"Sec. 1.1 and Sec. 4"},{"comment":"The Gershgorin derivation in Eq. (32) assumes that the two changed positions are off-diagonal; if a flipped permutation places a 1 on the diagonal, the change to the Gershgorin radius differs because diagonal entries are excluded from the radius. The bound should be derived with explicit bookkeeping of diagonal versus off-diagonal positions.","section":"Appendix A, Eq. (32)"},{"comment":"The text says 'K = 256 (i.e. K = 2n)' for n=8; this should be K = 2^n, while K=16 corresponds to 2n. The same typo appears in Sec. 2.2.3.","section":"Sec. 2.2.1 and Sec. 2.2.3"},{"comment":"The sentence 'we see the same error resilient observed in Fig. 1' should read 'the same error resilience observed in Fig. 1'.","section":"Sec. 3.2"},{"comment":"Reference [24] contains the typo 'Quantum omputation and quantum information', and the data-availability URL in Sec. 5 is broken across a line as 'sum-of-p ermutations.git'; the URL should be typeset as a single token.","section":"References and data availability"}],"recommendation":"major_revision","confidential_remarks":"The paper's title and abstract promise more than the technical content currently delivers. The dominant-eigenvalue invariance for the averaged operator is correct, but it does not by itself imply application-level resilience to bit-flip errors, and the control-register claim in Sec. 2.2.2 is wrong as stated. If the author reworks the analysis to treat average fidelity over actual error realizations and trims the overstated claims, the paper could become a useful technical contribution about eigenvalue stability of linear-combination-of-permutations operators. In its present form, however, it is likely to mislead readers who cite it for bit-flip resilience."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"I'll keep this short. The core observation is correct: for positive coefficients, any linear combination of permutation matrices has the all-ones vector as an eigenvector with eigenvalue equal to the coefficient sum, and since bit-flipping a permutation just gives another permutation, every error realization preserves that dominant eigenvalue exactly. That's a clean fact and the paper states it clearly.\n\nWhat doesn't hold is the inference from that spectral stability to the claim that 'applications ... are resilient to quantum bit-flip errors.' Equation (5) defines an expected operator, a first-moment object. The paper's error metrics (Eqs. 12–14) compare the eigenvalue of that expectation to the original, not the behavior of actual runs. The stress-test counterexample is decisive: with A = I + X, equal coefficients, and independent bit-flip probability p on each term, the expected operator equals A exactly, so the paper reports zero error. The real average success probability is 1 - p + p^2. Eigenvalues of the first moment don't determine measurement statistics, which are second moments. The paper never addresses this gap.\n\nSecondary issues: the Gershgorin appendix claims an equality for ΔR_j that assumes the perturbed entries were initially zero; in a sum over K terms that's generally false, so the bound is a worst-case inequality, not the derivation shown. And the Sinkhorn scaling generalization loses the all-ones eigenvector property, so the invariance doesn't automatically transfer to arbitrary matrices; the paper doesn't analyze that case.\n\nWhat's genuinely useful is the heuristic that positive-coefficient block encodings keep the dominant eigenvalue stable under bit-flips. That could inform error mitigation design. The numerics are modest (n=8, no error bars) but reproducible.\n\nMy recommendation: send it to a serious referee. The kernel is correct and the fix is conceptual rather than mathematical—either scope the claim to the dominant eigenvalue of the expected operator or redo the fidelity analysis as an average over random error realizations. As written, the abstract overstates the resilience.\n\nBest.","headline":"Correct spectral fact about bit-flip invariance of the dominant eigenvalue, but the application-resilience claim is not supported.","tokens_in":12501,"tokens_out":5708,"would_cite":false,"duration_ms":50413,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["15A18","15B51","81P68"],"pacs":["03.67.Lx","03.67.Pp"],"model":"deepseek-v4-flash","headline":"For quantum operators written as positive linear combinations of permutation matrices, bit-flip errors leave the dominant eigenvalue exactly unchanged, and the paper bounds how phase-flips and mixed-sign coefficients perturb the spectrum.","keywords":["linear combination of permutations","eigenvalue perturbation","bit-flip resilience","phase-flip errors","Perron-Frobenius theorem","Gershgorin circle theorem","block encoding","quantum error analysis"],"falsifier":"Prepare the dominant eigenvector of a positive-coefficient linear combination of permutations, apply the circuit with a known bit-flip on one term, and measure the fidelity of the output against the no-error output; if the fidelity drops while the dominant eigenvalue remains exactly $\\sum_i\\alpha_i$, the eigenvalue invariance is real but the claimed application-level resilience is not. A second check: compare the measured average output fidelity under random bit-flips with the fidelity predicted from the averaged operator $A(p,b)$; any systematic mismatch would falsify the error model itself.","tokens_in":11402,"feed_emoji":"⚛️","tokens_out":15817,"duration_ms":131920,"temperature":0.7,"pith_summary":"This paper analyzes quantum operators written as $A=\\sum_i \\alpha_i \\Pi_i$, where each $\\Pi_i$ is a permutation matrix (a matrix that reorders the entries of a vector), and asks how bit-flip and phase-flip errors disturb the eigenvalues. Its central claim is that when all coefficients $\\alpha_i$ are positive, the dominant eigenvalue of $A$ is exactly $\\sum_i \\alpha_i$, and bit-flips on any qubits in any term leave that value unchanged, because every flipped term is still a permutation and the column sums stay fixed. The paper argues that applications relying on this dominant eigenvalue are therefore resilient to quantum bit-flip errors. For mixed-sign coefficients and for phase-flips, the exact invariance is lost, but a Gershgorin-circle bound plus numerical simulations show the eigenspectrum and output-state fidelity are only slightly perturbed at small error rates. Because any square matrix can be written as a scaled linear combination of permutations through Sinkhorn's theorem, the analysis offers a general framework for error analysis in block-encoded quantum numerical linear algebra.","feed_headline":"Bit-flips keep the top eigenvalue of positive permutation sums fixed","feed_subtitle":"A dominant eigenvalue equal to the coefficient sum survives every bit-flip, anchoring error-tolerant quantum linear algebra.","key_machinery":"The load-bearing object is the decomposition $A=\\sum_i \\alpha_i \\Pi_i$ into permutation matrices with scalar coefficients. The argument runs on two mechanisms. First, a bit-flip is a Pauli-$X$ gate (a NOT gate that flips a qubit), and composing it with a permutation gives another permutation, so a flipped term is still a permutation term; when the $\\alpha_i$ are positive the whole matrix stays non-negative and keeps constant column sums, which pins the dominant eigenvalue at $\\sum_i\\alpha_i$ by the Perron-Frobenius theorem. Second, the Gershgorin circle theorem bounds how far any eigenvalue can move: each bit-flipped term changes a row's Gershgorin radius by at most $2p_i|\\alpha_i|$, so the full perturbation obeys $\\Delta R_j\\le 2\\sum_i p_i|\\alpha_i|$, and a similar bound holds for phase-flips. A block-encoded implementation adds a control register whose bit-flips swap the coefficients attached to the permutation terms, which the paper argues preserves the eigenvalue spectrum for positive coefficients.","core_discovery":"The paper's core discovery is an exact spectral invariance for positive linear combinations of permutations. For $A=\\sum_i \\alpha_i\\Pi_i$ with $\\alpha_i>0$, $A$ is a non-negative matrix whose columns all sum to $S=\\sum_i \\alpha_i$, so the Perron-Frobenius theorem makes $S$ the isolated dominant eigenvalue. Under the proposed bit-flip error model $A(p,b)=\\sum_i \\alpha_i\\bigl((1-p_i)I+p_i X_{b_i}\\bigr)\\Pi_i$, each term $\\alpha_i X_{b_i}\\Pi_i$ is again a positive permutation term, so $A(p,b)$ has the same column sums and therefore the same dominant eigenvalue $S$; the paper concludes that the dominant eigenvalue is not affected by bit-flip errors and that applications built on such matrices are resilient to those errors. Non-dominant eigenvalues are not protected, because their eigenvectors have mixed signs and the Gershgorin radius changes by at most $2\\sum_i p_i|\\alpha_i|$. For mixed-sign $\\alpha_i$ and for phase-flips, exact invariance is lost, yet the same Gershgorin machinery bounds the perturbation and the numerics show small spectral and fidelity changes at low error rates; the paper also shows that in a block-encoded implementation, bit-flips on the control register only permute coefficients and leave the spectrum unchanged.","pith_inferences":["One extension left implicit in the paper: encode the quantity an algorithm needs as the dominant eigenvalue of a positive linear combination of permutations, and system-register bit-flip noise becomes structurally irrelevant to that eigenvalue without any error correction.","A strengthening not stated in the paper: every single error realization is still a positive linear combination of permutations with the same column sums, so the dominant-eigenvalue invariance holds realization by realization; the residual vulnerability sits in the eigenvectors and the non-dominant spectrum.","A testable extension not explored in the paper: compare bit-flip errors on the control register versus the system register of the same block encoding; the model predicts control-register flips only permute coefficients and preserve the spectrum, so error budgets for the two registers could be set very differently.","The framework also points toward permutation-based error mitigation: because averaging over bit-flips preserves the dominant eigenvalue, deliberately randomizing over permutations could recover the eigenvalue exactly; this mitigation idea is not developed in the paper."],"forward_implications":["For any positive-coefficient linear combination of permutations, the dominant eigenvalue is exactly $\\sum_i\\alpha_i$, and bit-flips on any qubits in any term do not change it.","A positive-coefficient operator embedded by block encoding with a control register keeps the same eigenvalue spectrum under control-register bit-flips, since those only swap coefficients among terms.","For mixed-sign coefficients, the Gershgorin bound $\\Delta R_j\\le 2\\sum_i p_i|\\alpha_i|$ shows that small bit-flip probabilities keep the spectral perturbation small, although eigenvalues with small magnitude can suffer large relative changes.","Output-state fidelity is governed by the relative operator error: the bound $\\sigma_{\\max}(A-B)/\\sigma_{\\min}(A)$ shows that ill-conditioned operators amplify the effect of any bit-flip or phase-flip perturbation.","Phase-flips act like sign reversals on the matrix entries and produce larger deviations than bit-flips for mixed-sign matrices, but at low flip rates both the eigenvalue error and the fidelity loss remain small in the simulations."],"supporting_citations":[{"why":"Supplies the Perron-Frobenius theorem that makes the dominant eigenvalue of a non-negative matrix equal to its constant column sum, and the norm inequalities used in the fidelity bounds.","marker":"[5]"},{"why":"Supplies the Gershgorin circle theorem that bounds eigenvalue movement under bit-flip and phase-flip perturbations.","marker":"[26]"},{"why":"Sinkhorn's theorem is the basis for the claim that any positive square matrix can be scaled into the analyzed linear-combination-of-permutations form.","marker":"[17]"},{"why":"Birkhoff's theorem motivates the representation by writing doubly stochastic matrices as convex combinations of permutations.","marker":"[16]"},{"why":"Provides the block-encoding framework that embeds the analyzed operator as a block in a larger unitary, the application context for the error analysis.","marker":"[15]"},{"why":"Supplies the standard independent bit-flip and phase-flip error model on which the probabilistic operators A(p,b) and A(q,phi) are built.","marker":"[24]"},{"why":"Cited as the class of doubly-stochastic-matrix applications that are claimed to inherit resilience from the unchanged dominant eigenvalue.","marker":"[25]"}],"fun_headline_variants":["Positive permutation sums dodge bit-flip eigenvalue shifts","Bit-flips can't move the dominant eigenvalue of positive sums","Error analysis: bit-flips leave top eigenvalue fixed for positive coefficients","Spectral invariance under bit-flips for positive permutation combinations","Block-encoded bit-flips only permute coefficients, spectrum unchanged"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The model replaces each error-prone permutation term with a weighted average of the unflipped and flipped operator, so the claims describe the averaged operator $\\bigl((1-p_i)I+p_i X_{b_i}\\bigr)\\Pi_i$ rather than the behavior of any single run of the circuit; the conclusion that applications are resilient may not transfer if single-run behavior is what matters.","fun_headline_variants_meta":{"raw":{"variants":["Positive permutation sums dodge bit-flip eigenvalue shifts","Bit-flips can't move the dominant eigenvalue of positive sums","Error analysis: bit-flips leave top eigenvalue fixed for positive coefficients","Spectral invariance under bit-flips for positive permutation combinations","Block-encoded bit-flips only permute coefficients, spectrum unchanged"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000219,"raw_usage":{"total_tokens":1482,"prompt_tokens":1022,"completion_tokens":460,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":638,"completion_tokens_details":{"reasoning_tokens":375}},"tokens_in":638,"tokens_out":460,"duration_ms":4425,"temperature":1.0,"reasoning_tokens":375,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T13:45:37.119481+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Prepare the dominant eigenvector of a positive-coefficient linear combination of permutations, apply the circuit with a known bit-flip on one term, and measure the fidelity of the output against the no-error output; if the fidelity drops while the dominant eigenvalue remains exactly $\\sum_i\\alpha_i$, the eigenvalue invariance is real but the claimed application-level resilience is not. A second check: compare the measured average output fidelity under random bit-flips with the fidelity predicted from the averaged operator $A(p,b)$; any systematic mismatch would falsify the error model itself.","supporting_citations":[{"cited_title":"JHU press, 2013","cited_arxiv_id":null,"evidence_quote":"Supplies the Perron-Frobenius theorem that makes the dominant eigenvalue of a non-negative matrix equal to its constant column sum, and the norm inequalities used in the fidelity bounds."},{"cited_title":"Tres observaciones sobre el algebra lineal","cited_arxiv_id":null,"evidence_quote":"Birkhoff's theorem motivates the representation by writing doubly stochastic matrices as convex combinations of permutations."},{"cited_title":"Cambridge Univer- sity Press, 2010","cited_arxiv_id":null,"evidence_quote":"Supplies the standard independent bit-flip and phase-flip error model on which the probabilistic operators A(p,b) and A(q,phi) are built."},{"cited_title":"Some applications of doubly stochastic matrices","cited_arxiv_id":null,"evidence_quote":"Cited as the class of doubly-stochastic-matrix applications that are claimed to inherit resilience from the unchanged dominant eigenvalue."}],"review_version":1}