{"id":"e7310996-5058-4c54-9d21-d3198be95e40","arxiv_id":"2504.13168","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"If the noise operators commute with the signal Hamiltonian and a specific linear equation admits a solution, an ancilla-free autonomous QEC scheme restores Heisenberg scaling with error O(kappa T / R^c).","lead":"This paper identifies conditions under which autonomous quantum error correction, a passive method, can restore the Heisenberg scaling limit in quantum phase estimation. The result could make high-precision quantum sensors simpler by removing the need for active feedback and noiseless ancillas.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The QFI continuity bound in SM Step 3 is not merely unproven; it gives the wrong time scaling, so the claimed epsilon = O(kappa T/R^c) does not follow from the AutoQEC state-distance bound.","rationale":"The reader's weakest-assumption analysis correctly targets SM Step 3. Our stress-test sharpens this: the QFI continuity bound is not merely missing a proof; it fails quantitatively. For the logical qubit, F = 4 t^2 (h0-h1)^2 |c|^2, so a state-distance of order kappa t/R^c forces a QFI error of order t^2 * (kappa t/R^c) = kappa t^3/R^c. Thus the claimed epsilon = O(kappa T/R^c) cannot be derived from (S26). The power-series justification is also invalid for rank-deficient ideal states; a small component in a different H eigenspace can produce an O(1) change in QFI. This does not destroy the qualitative HS-restoration claim, because for any fixed T and epsilon one can still pick R large enough (with epsilon = O(kappa T^3/R^c)). The theorem's existence statement is likely true, but the advertised error scaling and its proof need revision. Therefore the conditional verdict is appropriate; the condition should require a corrected QFI bound with the proper t-dependence.","tokens_in":22162,"tokens_out":30316,"duration_ms":273480,"concrete_test":"Simulate the AutoQEC dynamics for the 3-qubit example (Eqs. 11-13) over t in [0,T] for several R. At each time, compute the corrected logical state sigma_t = tilde P_E[rho(t)] and the coherence c(t) = <mu0|sigma_t|mu1>. Test whether the coherence deficit delta(t) = 1/2 - |c(t)| scales as O(kappa t/R) (linear in t) or O(kappa/(t R)). If delta(t) grows linearly with t, then the QFI loss 4 t^2 (h0-h1)^2 delta(t) scales as O(kappa t^3/R), contradicting the theorem's epsilon = O(kappa T/R). Additionally, directly compute the ratio |F[sigma_t] - F[rho_id(t)]| / ||sigma_t - rho_id(t)|| over the same data; if this ratio grows like t^2, the Step 3 O(x) bound with a t-independent M0 is false.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The proof of Theorem 1 hinges on SM S1, Step 3, which asserts that |F[rho] - F[sigma]| <= M0 ||rho - sigma|| for a t-independent constant M0. For the two-dimensional code space used in the theorem, the QFI of any logical state sigma is exactly F[sigma] = 4 t^2 (h0-h1)^2 |<mu0|sigma|mu1>|^2. The AutoQEC bound (S26) provides ||tilde P_E[rho(t)] - rho_id(t)|| <= M kappa t / R^c, which bounds the coherence deficit by | |c| - 1/2 | = O(kappa t/R^c). Inserting this into the QFI formula gives a QFI error of order kappa t^3/R^c, not kappa t/R^c. The power-series expansion of F in the state difference is also unjustified because QFI can be discontinuous when rank support changes (e.g., adding an O(eps) component of |+> to |0> changes the QFI of a Z-encoding qubit by O(1)). Hence the claimed error scaling epsilon = O(kappa T/R^c) is not established and is inconsistent with the paper's own state-distance bound. The existence part of Theorem 1 (finite R for any epsilon) survives, but the advertised quantitative scaling is unsupported.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a sufficient condition for autonomous quantum error correction (AutoQEC) to restore Heisenberg scaling in noisy phase estimation. It proves (Theorem 1) that if all Lindblad operators commute with the signal Hamiltonian (T1) and a constrained linear system A_i^{[~c]} p_i = A_j^{[~c]} p_j admits a probability-vector solution (T2), then there exist ancilla-free codewords built from degenerate eigenspaces of H that satisfy the Knill-Laflamme condition for the error set E^{[~c]}. Using the engineered dissipation of Ref. [31], the AutoQEC state is shown to be close in operator norm to the ideal state, and the paper claims that the QFI approaches the ideal value (h_i-h_j)^2 t^2 with additive error epsilon = O(kappa T / R^c). The paper also provides a linear-programming method for checking (T2), discusses violations of the sufficient condition through properties (P1) and (P2), and presents numerical simulations for correlated and local dephasing noise.","tokens_in":22502,"tokens_out":12196,"duration_ms":134070,"significance":"If the claimed scaling were established, the paper would give the first general ancilla-free AutoQEC construction for metrology, with a checkable sufficient condition and a quantitative benefit from higher AutoQEC order c. The code construction itself is clean: T1 forces the Knill-Laflamme cross-terms to vanish, T2 forces the diagonal terms to match, and the reduction to the prior AutoQEC lemma of Ref. [31] is plausible. The paper also usefully distinguishes the HNLS condition from the AutoQEC-specific conditions and provides numerical evidence. However, the advertised error scaling is load-bearing and is not established by the provided proof; the main contribution as stated therefore needs substantial revision.","major_comments":[{"comment":"The bound |F[rho]-F[sigma]| <= M0 ||rho-sigma|| with a time-independent M0 is not proven and is in fact contradicted by the explicit QFI formula for the two-dimensional code used in Theorem 1. For any sigma in the code space, F[sigma] = 4 t^2 (h0-h1)^2 |<mu0|sigma|mu1>|^2. With sigma = tilde P_E[rho(t)] and rho_id(t) the ideal balanced pure state, the AutoQEC bound (S26) gives |<mu0|sigma|mu1> - 1/2| <= M kappa t / R^c up to constants, so |F[sigma]-F[rho_id]| is of order kappa t^3 / R^c, not kappa t / R^c. Consequently the theorem's stated epsilon = O(kappa T / R^c) and the abstract's emphasis on this scaling are unsupported. The existential part of Theorem 1 (finite R for any fixed epsilon) may survive with a modified bound epsilon = O(kappa T^3 / R^c), but that is not what the paper proves or claims.","section":"SM S1, Step 3, Eqs. (S27)-(S31)"},{"comment":"The power-series expansion rho - sigma = sum_k M_k x^k and F[rho] - F[sigma] = sum_k F_k x^k with t-independent coefficients is not justified. For two states with different rank support, and for perturbations that change the support of the state, the QFI is not generally analytic in the operator-norm perturbation; adding an O(epsilon) component in an orthogonal direction can change the QFI by O(1). Moreover, even where an expansion exists, the coefficients F_k depend on t through the unitary evolution of both states, so the step from (S30) to (S31) requires a proof that the Lipschitz constant M0 is uniformly bounded in t over the interval [0,T]. No such proof is given, and the explicit two-level formula above shows that the natural dependence is t^2.","section":"SM S1, Step 3, Eqs. (S28)-(S30)"}],"minor_comments":[{"comment":"There is a typo in the abstract: 'rate,)' should read 'rate).' Also, the paper uses 'HS' both for exact Heisenberg scaling and for the approximate preservation up to additive error; please clarify this in the introductory paragraphs.","section":"Abstract and main text"},{"comment":"The infinite-R limit in Theorem 2 is analyzed with the ansatz rho(t) = sum_k R^{-k} rho^{(k)}(t), but the paper does not explicitly justify that the limit R -> infinity commutes with the time evolution. Please make this limit argument precise.","section":"SM S3.A, Eq. (S52)"},{"comment":"The caption does not state which residual-state choices |Phi_q> are used for each curve; for c=2 there is no residual space, while for c=1 the choice |Phi_q> = (|mu0>+|mu1>)/sqrt(2) is made. Please make the parameter choice for each curve explicit.","section":"Fig. 3 caption"},{"comment":"The inequality F[rho(t)] <= F[rho_id(t)] is asserted without proof. The final inequality (S34) is trivially true when F[rho(t)] > F[rho_id(t)], but the assumption should still be stated as an assumption or justified, since it is used to frame the main bound.","section":"SM S1, Eq. (S33)"}],"recommendation":"major_revision","confidential_remarks":"The construction and the sufficient condition are interesting and likely salvageable, but the central quantitative claim is not established. The flaw is not merely presentational: the claimed epsilon = O(kappa T / R^c) is contradicted by the explicit QFI formula for the code used in the theorem, which yields a natural t^3 dependence. I recommend asking the authors to supply a correct QFI continuity analysis or to revise the theorem and all dependent statements to a scaling they can actually prove, and then to re-check the numerical discussion in light of the corrected scaling. There is no circularity problem in the paper's logic; the conditions are stated assumptions and Lemma 1 is used as an external tool."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper does something new: it gives a sufficient condition under which an ancilla-free autonomous QEC code can restore Heisenberg scaling, with a concrete construction and a linear-programming check. That's a real step beyond the prior AutoQEC metrology attempts, and the code construction is elegant. The distinction from Ref. [42] is honest, and the numerical evidence is consistent with the existence claim.\n\nThe soft spot is in the proof of the error scaling. SM S1 Step 3 asserts that |F[rho]-F[sigma]| <= M0 ||rho-sigma|| with M0 independent of t. For the two-dimensional code used in Theorem 1, the QFI of any logical state sigma is exactly 4 t^2 (h0-h1)^2 |<mu0|sigma|mu1>|^2. The state-distance bound (S26) gives ||tilde P_E[rho(t)] - rho_id(t)|| <= M kappa t/R^c, so the coherence deficit is O(kappa t/R^c). Plugging that into the QFI formula gives a QFI error of order kappa t^3/R^c, not kappa t/R^c. So the claimed epsilon = O(kappa T/R^c) is not established. The power-series expansion in Step 3 is also unjustified—QFI is not Lipschitz in the operator norm when rank changes—but even ignoring that, the t^2 prefactor in the exact QFI formula shows a t-independent M0 cannot exist.\n\nThe existence part of Theorem 1 survives: for any T and epsilon you can choose R large enough to suppress the kappa T^3/R^c term. But the advertised quantitative scaling law, which is a central selling point, is wrong as stated.\n\nMinor issue: no code or data for the numerics is provided, so the plots are not independently checkable.\n\nThis paper deserves serious peer review. The core idea is important and the gap is concrete and likely fixable—either by proving a refined QFI continuity bound with explicit t^2 dependence, or by changing the error criterion. I would send it to a competent referee but would not accept the current version at face value.","headline":"A genuinely useful sufficient condition and code construction for ancilla-free AutoQEC metrology, but the advertised error scaling is not proven: the QFI continuity bound in SM Step 3 fails, and the paper's own distance bound actually yields O(kappa T^3/R^c) rather than O(kappa T/R^c).","tokens_in":23008,"tokens_out":13683,"would_cite":true,"duration_ms":125238,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["03.67.-a","03.67.Lx"],"model":"deepseek-v4-flash","headline":"Autonomous quantum error correction can restore Heisenberg scaling in phase estimation under Markovian noise, without noiseless ancilla, whenever the noise operators commute with the signal Hamiltonian and a constrained linear equation…","keywords":["Heisenberg scaling","autonomous quantum error correction","quantum metrology","quantum Fisher information","Markovian noise","dephasing","Knill-Laflamme condition","engineered dissipation"],"falsifier":"The linear bound $|F[\\hat{\\rho}] - F[\\hat{\\sigma}]| = O(\\|\\hat{\\rho} - \\hat{\\sigma}\\|_{\\rm op})$ used in the supplemental proof fails for rank-changing states: for $\\hat{\\sigma} = |0\\rangle\\langle 0|$, $\\hat{\\rho}_\\varepsilon = (1-\\varepsilon)|0\\rangle\\langle 0| + \\varepsilon|+\\rangle\\langle +|$, and $\\hat{H} = |0\\rangle\\langle 1| + |1\\rangle\\langle 0|$, the operator-norm distance is $\\varepsilon/\\sqrt{2}$ yet $F[\\hat{\\rho}_\\varepsilon] - F[\\hat{\\sigma}] \\to -2$ as $\\varepsilon \\to 0$. A numerical check of the paper's own 3-qubit example---whether the QFI deficit at fixed $t$ decays as $1/R$ for $R = 10^3, 10^4, 10^5$---would settle whether this pathology invalidates Theorem 1 in the intended setting.","tokens_in":21993,"feed_emoji":"🎯","tokens_out":27911,"duration_ms":227617,"temperature":0.7,"pith_summary":"This paper aims to show that autonomous quantum error correction (AutoQEC), which uses only engineered dissipation and needs no continuous measurement or feed-forward, can restore Heisenberg scaling in quantum phase estimation despite Markovian noise. The authors prove a sufficient condition: if every Lindblad operator of the noise commutes with the fixed signal Hamiltonian, and a constrained linear equation over probability vectors admits a solution, then a simple ancilla-free two-word code exists whose quantum Fisher information is at least $(h_i - h_j)^2 t^2 - \\epsilon$ for any sensing time up to $T$, with $\\epsilon = O(\\kappa T / R^c)$. Here $R$ is the ratio of engineered to natural dissipation and $c$ is the AutoQEC order, so higher order needs less dissipation to reach a given precision. The result matters because it identifies exactly when the experimentally easier always-on dissipation route can match the metrological performance of active error correction, and it comes with a linear-programming procedure to find the code. Numerical simulations under correlated and local dephasing confirm the predicted recovery of Heisenberg scaling.","feed_headline":"Autonomous error correction restores Heisenberg-limited sensing","feed_subtitle":"No active feedforward or noiseless ancilla needed; engineered dissipation alone keeps precision near Heisenberg limit.","key_machinery":"The load-bearing object is a family of matrices $A_i^{[\\sim c]}$ with entries $[A_i^{[\\sim c]}]_{kl} = \\langle h_i^{(l)} | \\hat{K}_k | h_i^{(l)} \\rangle$, where $\\hat{K}_k$ runs over the operators $\\hat{E}_a^\\dagger \\hat{E}_b$ formed from the error set $E^{[\\sim c]}$ and $|h_i^{(l)}\\rangle$ are the eigenvectors of the signal Hamiltonian with eigenvalue $h_i$. The code consists of two codewords $|\\mu_0\\rangle, |\\mu_1\\rangle$ built by taking square roots of probability vectors $p_i, p_j$ as amplitudes on the two eigenspaces; condition (T2) is exactly the statement that these vectors can be chosen so the diagonal Knill-Laflamme checks match across codewords, while (T1) makes the off-diagonal checks vanish automatically. With the Knill-Laflamme condition satisfied for $E^{[\\sim c]}$, the engineered dissipation from the standard AutoQEC construction (Lemma 1 of the paper) suppresses the noise to order $1/R^c$, and the quantum Fisher information data-processing inequality for the CPTP projector $\\tilde{P}_E$ converts that suppression into the Heisenberg-scaling lower bound. Two structural properties that follow from the conditions do the real work: (P1) the signal Hamiltonian commutes with all correctable error operators, and (P2) it commutes with the projectors onto the $n$th-order correctable error spaces, so the signal never leaks amplitude into uncontrolled sectors.","core_discovery":"The central claim is Theorem 1: given a signal Hamiltonian $\\hat{H}$ and Markovian noise with Lindblad operators $\\hat{L}_{n,a}$, if (T1) $[\\hat{H}, \\hat{L}_{n,a}] = 0$ for all $a$, and (T2) there exist two distinct eigenvalues $h_i, h_j$ of $\\hat{H}$ and probability vectors $p_i, p_j$ with $A_i^{[\\sim c]} p_i = A_j^{[\\sim c]} p_j$ (where $A_i^{[\\sim c]}$ records the diagonal expectations of the error-correlation operators $\\hat{K}$ in the eigenspace of $h_i$), then for any $T$ and any $\\epsilon > 0$ one can build an ancilla-free AutoQEC scheme with finite $R$ such that the quantum Fisher information of the evolving probe satisfies $F[\\hat{\\rho}(t)] \\ge (h_i - h_j)^2 t^2 - \\epsilon$ for all $0 \\le t \\le T$, with $\\epsilon = O(\\kappa T / R^c)$. The code words are superpositions of Hamiltonian eigenstates from the two chosen eigenspaces with amplitudes taken from $p_i, p_j$; (T1) makes the signal and all correctable errors diagonal in the same basis, eliminating cross-codeword error terms, and (T2) enforces the Knill-Laflamme condition so the standard AutoQEC dissipation of order $c$ suppresses errors, and the quantum Fisher information data-processing inequality converts the resulting trace-norm closeness into the Fisher information bound.","pith_inferences":["The commuting condition (T1) rules out many natural noise models such as amplitude damping for typical transverse-field sensors, so Theorem 1 is likely most useful in dephasing-dominated settings; mapping which realistic noise-signal pairs admit commuting Lindblad operators is a natural follow-up.","Because the QFI bound passes through trace-norm closeness, the constant in the error $\\epsilon = O(\\kappa T / R^c)$ is state-dependent; an experiment comparing measured Fisher information against this bound at several $R$ values would reveal how conservative the estimate is in practice.","The linear-programming formulation suggests an automated code-search loop: given a signal and a noise correlation matrix, scan eigenvalue pairs in descending order of $|h_i - h_j|$ to find the code that maximizes the Heisenberg slope, turning the sufficiency condition into a design tool.","Combined with the paper's infinite-$R$ result (Theorem 2), the practical prescription is to use the Theorem 1 code when it exists and to reserve active feedback or very large $R$ for the HNLS-only regime; quantifying that crossover could guide resource allocation between engineered dissipation and measurement-based correction."],"forward_implications":["Any phase-estimation setup satisfying (T1) and (T2) inherits a concrete, ancilla-free AutoQEC code whose quantum Fisher information stays within $O(\\kappa T / R^c)$ of $(h_i - h_j)^2 t^2$ over any prescribed sensing window.","Higher AutoQEC order $c$ reduces the engineered-dissipation ratio $R$ required for a fixed error, so the scheme becomes easier to implement as $c$ grows, up to the dimension constraints of the code.","When the sufficient condition holds, the Hamiltonian-not-in-Lindblad-span (HNLS) condition is automatically satisfied, so the scheme also meets the necessary condition for any QEC-based restoration of Heisenberg scaling.","If the condition fails, either the signal fails to commute with the errors (violating P1) or it couples distinct correctable error spaces (violating P2); the numerical examples show degraded Fisher information, worst when P2 is violated, because the signal then generates errors faster than natural dissipation.","Finding codewords that satisfy (T2) reduces to a linear program, so for a given noise model one can efficiently check feasibility and, when feasible, obtain the codewords."],"supporting_citations":[{"why":"Supplies the definition of AutoQEC up to order c and Lemma 1, the engineered dissipation and Hamiltonian that achieve it once the Knill-Laflamme condition holds; Theorem 1 builds directly on this lemma.","marker":"[31]"},{"why":"Introduces ancilla-free QEC codes for metrology and the argument that signal-noise commutation makes cross-codeword Knill-Laflamme checks vanish without an ancilla; the proof's Step 1 adapts this technique.","marker":"[22]"},{"why":"Establishes the HNLS condition as the key necessary condition for QEC to restore Heisenberg scaling; the paper uses it as benchmark and as the contrast case in the infinite-R Theorem 2.","marker":"[24]"},{"why":"Defines the CPTP projector tilde-P_E = lim e^{L_E u} that stabilizes the code space, used in Definition 1 and in converting AutoQEC closeness to a Fisher-information statement.","marker":"[44]"},{"why":"Provides the data-processing inequality for quantum Fisher information, which the proof uses to turn trace-norm closeness of the error-corrected state into the lower bound on F[rho(t)].","marker":"[48]"}],"fun_headline_variants":["AutoQEC recovers Heisenberg scaling in metrology","No ancilla needed: AutoQEC retains Heisenberg limit","Dissipation alone restores Heisenberg-limited sensing","Ancilla-free error correction restores quantum precision","Autonomous QEC regains optimal metrology scaling"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof relies on the unproved assertion that small changes in a quantum state cause only proportionally small changes in the quantum Fisher information, even when the two states have different numbers of nonzero components (different rank).","fun_headline_variants_meta":{"raw":{"variants":["AutoQEC recovers Heisenberg scaling in metrology","No ancilla needed: AutoQEC retains Heisenberg limit","Dissipation alone restores Heisenberg-limited sensing","Ancilla-free error correction restores quantum precision","Autonomous QEC regains optimal metrology scaling"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000663,"raw_usage":{"total_tokens":3099,"prompt_tokens":1087,"completion_tokens":2012,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":703,"completion_tokens_details":{"reasoning_tokens":1944}},"tokens_in":703,"tokens_out":2012,"duration_ms":14999,"temperature":1.0,"reasoning_tokens":1944,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T12:14:41.245899+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"The linear bound $|F[\\hat{\\rho}] - F[\\hat{\\sigma}]| = O(\\|\\hat{\\rho} - \\hat{\\sigma}\\|_{\\rm op})$ used in the supplemental proof fails for rank-changing states: for $\\hat{\\sigma} = |0\\rangle\\langle 0|$, $\\hat{\\rho}_\\varepsilon = (1-\\varepsilon)|0\\rangle\\langle 0| + \\varepsilon|+\\rangle\\langle +|$, and $\\hat{H} = |0\\rangle\\langle 1| + |1\\rangle\\langle 0|$, the operator-norm distance is $\\varepsilon/\\sqrt{2}$ yet $F[\\hat{\\rho}_\\varepsilon] - F[\\hat{\\sigma}] \\to -2$ as $\\varepsilon \\to 0$. A numerical check of the paper's own 3-qubit example---whether the QFI deficit at fixed $t$ decays as $1/R$ for $R = 10^3, 10^4, 10^5$---would settle whether this pathology invalidates Theorem 1 in the intended setting.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the definition of AutoQEC up to order c and Lemma 1, the engineered dissipation and Hamiltonian that achieve it once the Knill-Laflamme condition holds; Theorem 1 builds directly on this lemma."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces ancilla-free QEC codes for metrology and the argument that signal-noise commutation makes cross-codeword Knill-Laflamme checks vanish without an ancilla; the proof's Step 1 adapts this technique."},{"cited_title":"Demkowicz-Dobrzański, J","cited_arxiv_id":null,"evidence_quote":"Establishes the HNLS condition as the key necessary condition for QEC to restore Heisenberg scaling; the paper uses it as benchmark and as the contrast case in the infinite-R Theorem 2."},{"cited_title":"Reiter, A","cited_arxiv_id":null,"evidence_quote":"Defines the CPTP projector tilde-P_E = lim e^{L_E u} that stabilizes the code space, used in Definition 1 and in converting AutoQEC closeness to a Fisher-information statement."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the data-processing inequality for quantum Fisher information, which the proof uses to turn trace-norm closeness of the error-corrected state into the lower bound on F[rho(t)]."}],"review_version":1}