{"id":"bd29d589-8aa1-4b53-b27d-806840492e92","arxiv_id":"2506.21455","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"An iterative polar-decomposition method for finding a unitary U with sigma approx U rho U* from state pairs, with a claim of convergence to a local minimum and a limited-data reconstruction scheme.","lead":"Researchers propose an iterative algorithm using polar decomposition to recover an unknown unitary quantum channel from limited input-output state pairs, proving convergence to a local minimum of a least-squares objective. The paper also outlines a reconstruction procedure using roughly n^2 + 3n measurements, but the derivations contain errors.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The reconstruction protocol in Section 5 uses auxiliary 'quantum states' that are not positive semidefinite, so the claimed n^2+3n-observable recovery is not physically realizable as stated.","rationale":"The paper's central claim has two parts: a convergent polar-decomposition iteration and a limited-data reconstruction protocol. I checked the convergence proof because the reader flagged Theorem 5. The assertion that I-(U^s)^*U^{s+1} is positive semidefinite is indeed false: for a unitary A this matrix need not be Hermitian, and its eigenvalues are 1-e^{iθ}. However, the monotonicity inequality that Theorem 5 needs does not actually depend on that assertion. Lemma 4.1 gives Re⟨σU^sρ, U^{s+1}⟩ ≥ Re⟨σU^sρ, U^s⟩, and with Δ=U^{s+1}-U^s one obtains g(U^s)-g(U^{s+1}) = 2Re⟨σ,U^sρΔ*⟩ + Re⟨σ,ΔρΔ*⟩ ≥ 0, because the first term is the Lemma 4.1 difference and the second is the trace of σ with ΔρΔ*, both positive semidefinite. Thus the convergence theorem may be repairable, and I do not rest the rejection on that gap. The reconstruction protocol, by contrast, is built on objects that are not physical states. Section 5 defines (ρ_{p,q,r})_+ and (ρ_{p,q,r})_- as v_r v_r^* plus an off-diagonal term between v_p and v_q. These matrices have trace one and are Hermitian, but on the two-dimensional subspace spanned by v_p and v_q the off-diagonal block [[0,1/2],[1/2,0]] has eigenvalues ±1/2. Hence the full matrix has a negative eigenvalue (for n=3, {1,1/2,-1/2}), so it is not a density operator. A quantum channel is a physical operation on states; feeding it a non-positive matrix is not an experimentally meaningful step, and it also violates the paper's own standing assumption that (σ,ρ) are Hermitian positive definite. The recovery of D in Steps 3-4 uses exactly these matrices as inputs, so the central claim of recovering U from n^2+3n observables is unsupported. There is also an indexing/count problem in the same step, but the non-positivity alone is decisive. Because this concern is about the physical validity of the advertised reconstruction, the reader's REJECT verdict is appropriate, although the specific weakest assumption identified by the reader is not the one on which I would rest the rejection.","tokens_in":1223,"tokens_out":1482,"duration_ms":293208,"concrete_test":"Compute the spectrum of the 3x3 Hermitian matrix (ρ_{2,3,1})_+ = v_1v_1^* + 1/2(v_2v_3^*+v_3v_2^*) in an orthonormal basis; it equals {1, 1/2, -1/2}. This single calculation establishes that the auxiliary 'quantum states' in Section 5, Step 3 are not positive semidefinite, so the recovery procedure cannot be implemented as a quantum-state protocol.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing problem is in the reconstruction protocol, not in the convergence proof. The states (ρ_{p,q,r})_± introduced in Section 5, Step 3 are Hermitian and have trace one, but they are not density matrices. Restrict (ρ_{p,q,r})_+ to span{v_p,v_q}: it becomes [[0,1/2],[1/2,0]], whose eigenvalues are ±1/2; the full matrix therefore has a negative eigenvalue (for n=3, the spectrum is {1,1/2,-1/2}). The same holds for (ρ_{p,q,r})_-. Hence these matrices cannot be prepared as inputs to a unitary quantum channel and do not satisfy the 'Hermitian positive definite' condition stated for Problem 1 and Algorithm 1. Steps 3-4 of the recovery procedure rely on passing exactly these matrices through the channel, so the advertised recovery of U with n^2+3n observables is not a valid quantum-state protocol. Separately, the text says to compute u_1^*Φ(...)u_1; with U0=D and u_1=e_1 this evaluates to 1 regardless of D, so the claimed recovery of d_pp d_qq^* is also not supported by the written computation. The Theorem 5 semidefinite assertion flagged by the reader is a genuine proof defect, but the descent inequality can be obtained directly from Lemma 4.1, so the convergence claim may be repairable; the invalid auxiliary states are more decisive.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes an iterative algorithm based on polar decomposition to identify a unitary quantum channel Φ(ρ)=UρU* from input-output pairs. It proves that the solutions of the one-pair problem form an equivalence class modulo a diagonal unitary (Theorem 1) and that channel representations are unique up to a global phase (Theorem 2). Algorithm 1 uses a polar-decomposition update; Theorems 5-6 claim monotone decrease of the objective and convergence to a critical point. Section 5 gives a reconstruction procedure to recover U using n^2+3n tomography steps. Numerical experiments on random 10×10 channels and a synthetic 8×8 quantum circuit are reported.","tokens_in":12571,"tokens_out":23125,"duration_ms":235819,"significance":"If the results were correct, the paper would offer a simple first-order polar-decomposition algorithm for unitary channel estimation with a convergence guarantee, plus a measurement-efficient reconstruction whose n^2+3n count improves on standard process tomography. The equivalence-class result (Theorem 1) and global-phase uniqueness (Theorem 2) are clean and appear correct. The numerical experiments, while not reproducible from the text, are consistent with convergence and accurate recovery. However, the advertised reconstruction protocol is not physically valid as written, and the convergence proofs contain incorrect algebraic assertions. The central idea is promising, but the manuscript in its current form does not rigorously support its headline claims.","major_comments":[{"comment":"The assertion in the proof of Theorem 5 that 'the matrix I − (U(s))* U(s+1) is positive semidefinite' is false for unitary matrices: for any eigenvalue e^{iθ} of U(s)*U(s+1), the corresponding eigenvalue of I − U(s)*U(s+1) is 1 − e^{iθ}, which is not real nonnegative unless θ=0. The proof of monotone decrease therefore does not hold as written. The claim itself appears repairable: from Lemma 4.1 one obtains Re Tr(σU(s)ρU(s+1)*) ≥ Re Tr(σU(s)ρU(s)*), and the remaining cross term Re Tr(σΔUρΔU*) is a nonnegative squared Frobenius norm. Please replace the false positive-semidefinite argument with this direct derivation.","section":"Section 4, Theorem 5"},{"comment":"The diagonalization step contains algebraic errors. The displayed equality Re(Tr(QρΛρQρ* ΔU QσΛσQσ* ΔU*)) = Re(Tr((ΛσQσ*ΔU*Qρ)(ΛρQρ*ΔU*Qσ))) is incorrect: with the stated R = Qσ*ΔU*Qρ, the correct second factor should be ΛρR* = ΛρQρ*ΔUQσ, not ΛρQρ*ΔU*Qσ. The proof also invokes the identity 'tr(AB*) = tr(A*B)', which is false in general; only the real parts coincide for the Hermitian matrices considered. The lemma's conclusion may still be true, but the proof as written is not sound.","section":"Section 4, Lemma 4.2"},{"comment":"The auxiliary matrices (ρ_{p,q,r})_± introduced in Step 3 are not density matrices. For example, in the subspace spanned by v_p and v_q, the matrix 1/2(v_p v_q* + v_q v_p*) has eigenvalues ±1/2, so the full matrix (which also contains v_r v_r*) has a negative eigenvalue for n≥3. Since Problem 1 and Algorithm 1 restrict to Hermitian positive definite inputs, these matrices cannot be prepared as quantum states or passed through the channel in the proposed protocol. The claimed reconstruction using n^2+3n quantum observables is therefore not a valid quantum-state procedure as stated.","section":"Section 5, Step 3"},{"comment":"The expression for the channel has the wrong placement of D: from U0 = U V D V* one gets U = U0 V D* V*, so Φ(ρ) = U0(V D* V*)ρ(V D V*)U0*, whereas the text writes U0(V D V*)ρ(V D* V*)U0*. Moreover, with the displayed formula and u1 equal to the first column of U0, the computation u1*Φ((ρ_{p,q,1})_±)u1 evaluates to 1 independently of D (take V=I and U0=D), so the claim that this yields d_pp d_qq* is not supported. The authors need to specify the correct measurement, e.g., cross terms u_p*Φ(ρ)u_q, and the correct ordering of D and D*.","section":"Section 5, Steps 2-4"},{"comment":"The abstract states that the algorithm identifies 'a critical point, which is also a local minimum,' but Theorem 6 only establishes convergence to a critical point; no second-order analysis is provided. In addition, the proof that the number of critical points is finite uses Lemma 4.5, but the first-order condition together with the unitary constraint is not a square polynomial system of the type covered by that lemma, and Lemma 4.5 only counts nonsingular solutions. Please either prove the local-minimum claim or revise the abstract, and provide a more careful finiteness argument.","section":"Abstract and Theorem 6"}],"minor_comments":[{"comment":"The inner product ⟨X,Y⟩_R is defined with conjugation in equation (7) but as Σ x_ij y_ij (no conjugation) in the proof of Lemma 3.1; this inconsistency makes the derivative computation hard to verify.","section":"Section 2 and Lemma 3.1"},{"comment":"The notation switches between Cm×m and Cn×n in the tangent-space formulas; please unify the dimensions.","section":"Section 2, equations (5)-(6)"},{"comment":"The comment that eigenvalues of ρ and σ are 'positive and strictly less than one' is not an assumption of Problem 1, which only requires positive definiteness; the inequality used is valid for all positive eigenvalues, so the condition is unnecessary.","section":"Section 4, Lemma 4.2"},{"comment":"There are several typos: 'p, q≠ i' should read 'p,q≠ r'; 'consided' should be 'considered'; in Example 2, 'result result' is duplicated.","section":"Section 5 and Section 6"},{"comment":"The text states that the algorithm terminates after 1000 iterations, but the x-axes of the figures extend to 10^5; please clarify. Also, no random seed or code is provided, so the numerical experiments are not reproducible.","section":"Section 6, Figures 1-2"},{"comment":"The novelty statement ('the use of matrix optimization methods to solve the quantum channel identification problem has not yet been explored') is too strong; there is existing work on optimizing over unitary matrices for quantum process learning. Please temper this claim and add references.","section":"Introduction"}],"recommendation":"major_revision","confidential_remarks":"This is a borderline case. The reconstruction protocol's use of non-positive-semidefinite 'states' and the wrong D ordering/measurement formula are genuine errors in a core contribution. They appear fixable (e.g., by choosing PSD probe states and measuring cross matrix elements), so I recommend major revision rather than reject. The convergence-proof errors are also fixable. The authors should be asked to rewrite Section 5, correct the proofs of Theorem 5 and Lemma 4.2, and align the abstract's 'local minimum' claim with the theorems."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Punchline: the polar-decomposition iteration is a sensible idea and the convergence theorem is plausibly repairable, but the reconstruction protocol in Section 5 is not physically valid, so the paper's main practical claim does not hold as written.\n\nWhat is new: the update by polar decomposition of 2σUρ is a natural first-order method on the unitary group, and I do not know a prior reference for exactly this update applied to unitary channel identification. The convergence analysis via Lemma 4.1 (closest unitary to a given matrix) is a reasonable strategy, and Theorem 2 (global phase uniqueness) is clean. The one-shot equivalence class theorem is standard but carefully stated. The numerics show the objective decreasing to machine precision, so there is a working algorithm underneath.\n\nSoft spots: the stress test on Section 5 is right. The matrices (ρ_p,q,r)_± defined in Step 3 are Hermitian and trace one but not positive semidefinite; restricted to span{v_p,v_q}, (ρ_p,q,r)_+ is [[0,1/2],[1/2,0]] with eigenvalues ±1/2, so it is not a valid quantum input. The same holds for the minus sign. Passing a non-state through a channel is not a well-defined operation. Also, the text claims computing u_1* Φ(...) u_1 with U0 = D and u_1 = e_1 recovers d_pp d_qq*, but the expression is identically 1. So Steps 3–4 do not support the claimed n^2+3n-observable recovery.\n\nThe convergence proof also has a genuine flaw: Theorem 5 asserts I − (U^(s))*U^(s+1) is positive semidefinite, which is false for general unitaries. The descent inequality might still follow from Lemma 4.1 directly, so this is likely repairable. Lemma 4.2's diagonalization step contains algebra mistakes and assumes eigenvalues of ρ and σ are less than 1 without that being assumed anywhere. These are presentation-level defects compared with the Section 5 issue, but they are real.\n\nWho this is for: someone working on manifold optimization for quantum process identification, or a numerical linear algebra audience interested in polar decomposition iterations. It is not a quantum information breakthrough; standard process tomography solves the identification problem with more data, and the paper does not compare against it.\n\nRecommendation: send it out. The core algorithm and convergence analysis deserve a serious referee, and the Section 5 problem is concrete enough that a revision could fix it. If the reconstruction part cannot be repaired, the paper shrinks to a convergence analysis of a known-type iteration, which may still be publishable if honest about scope.","headline":"The polar-decomposition iteration is plausible, but the Section 5 recovery protocol uses non-physical quantum states and a sign error, so the paper's central practical claim fails as written.","tokens_in":13040,"tokens_out":2322,"would_cite":false,"duration_ms":26680,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65F25","15A23","81P45"],"pacs":[],"model":"deepseek-v4-flash","headline":"A polar-decomposition iteration recovers an unknown unitary quantum channel from limited input-output pairs.","keywords":["unitary quantum channel","polar decomposition","Stiefel manifold","Frobenius norm","quantum state tomography","channel reconstruction","local minimum","convergence analysis"],"falsifier":"Run Algorithm 1 from $U^{(0)}=I$ with $\\sigma=\\operatorname{diag}(2,1)$ and $\\rho=\\operatorname{diag}(1,2)$: at the first step the matrix $I-(U^{(0)})^*U^{(1)}$ is generically not Hermitian positive semidefinite, so checking whether $g(U^{(1)})>g(U^{(0)})$ directly tests the claimed monotone-decrease guarantee.","tokens_in":12099,"feed_emoji":"⚛️","tokens_out":8693,"duration_ms":88790,"temperature":0.7,"pith_summary":"This paper claims that an unknown unitary quantum channel—the reversible evolution that sends a state $\\rho$ to $\\sigma = U\\rho U^*$—can be recovered from very few input-output pairs by solving a least-squares problem on the manifold of unitary matrices. The proposed iteration alternates a matrix multiplication with polar decomposition, $U^{(s+1)} = \\mathrm{polar}(2\\sigma U^{(s)}\\rho)$, and the authors prove that the squared Frobenius misfit decreases and the iterates converge to a critical point, hence a local minimum, for almost all Hermitian positive definite pairs $(\\sigma,\\rho)$. They also prove that with one non-degenerate input state the candidate unitaries form an equivalence class differing by a diagonal unitary in the eigenbasis of $\\rho$, and that two unitaries implementing the same channel differ only by a global phase. The reconstruction procedure uses $n^2+3n$ observables, which is less than the baseline $4n^2-2n-4$ Hermitian matrices needed for general unitary process tomography.","feed_headline":"Polar decomposition finds the unitary behind a quantum channel","feed_subtitle":"Starting from one input-output pair, it needs only n²+3n observables to recover the channel.","key_machinery":"The key object is the polar decomposition of the gradient-scaled product $2\\sigma U^{(s)}\\rho$, whose unitary factor is the nearest unitary matrix to that product in Frobenius norm. This makes the polar factor a natural projected-gradient step on the unitary group: the update is the unitary closest to the negative Euclidean gradient, so the fixed-point equation $U P = -\\nabla g(U)$ exactly characterizes critical points of the constrained problem. The analysis is carried on the Stiefel manifold using its tangent-space geometry, expressed through skew-Hermitian coordinates, to compute projections, and the search-space reduction uses an equivalence relation $\\#_V$ in which two solutions differ by a diagonal unitary in the eigenbasis $V$ of the non-degenerate input state $\\rho$.","core_discovery":"The central discovery is an iterative, first-order matrix algorithm that turns quantum channel identification into an optimization problem on the Stiefel manifold, the set of $n\\times n$ unitary matrices. Starting from any unitary $U^{(0)}$, the update $U^{(s+1)} = \\mathrm{polar}(2\\sigma U^{(s)}\\rho)$ is designed so that every fixed point satisfies the first-order optimality condition $U P = -\\nabla g(U)$, and the paper argues that the objective $g(U)=\\tfrac12\\|\\sigma - U\\rho U^*\\|_F^2$ never increases along the sequence and converges to a critical point for almost all data pairs. The reconstruction side shows that a single well-chosen non-degenerate state reduces the search to a diagonal phase matrix in the eigenbasis of the input state; probing the channel on $2n$ additional special states fixes those phases, yielding the full unitary up to a global phase with $n^2+3n$ tomography steps.","pith_inferences":["If the monotonicity argument is completed, the same update rule extends directly to multiple input-output pairs by minimizing a sum of squared Frobenius losses, since the polar-decomposition step is unchanged.","The equivalence-class reduction implies that after one non-degenerate pair only $n$ phase parameters remain to be measured, so a parallel hardware implementation of the $2n$ phase-probing steps could cut wall-clock time while keeping the $n^2+3n$ sample count.","The polar-factor step is a natural projected-gradient template that could be adapted to other matrix manifolds where the nearest feasible point is available in closed form, such as orthogonal or symplectic channel estimation.","The global-phase ambiguity proved for equivalent channel descriptions means downstream phase-sensitive algorithms must treat the overall phase as a gauge, while phase-insensitive uses can ignore it."],"forward_implications":["A single non-degenerate input-output pair already determines the channel up to a diagonal unitary in the eigenbasis of the input, so the effective search dimension drops from $n^2$ to $n$.","Two unitaries describing the same channel on all states differ only by a complex scalar of modulus 1; the global phase is the only remaining ambiguity.","Reconstructing the full unitary requires $n^2+3n$ observables, beating the $4n^2-2n-4$ Hermitian matrices required by generic unitary process tomography.","The iteration is implementable with standard matrix operations and converges numerically on random $10\\times10$ channels and on an 8-dimensional composite gate made of Hadamard and CNOT gates.","Because the update uses only matrix products and a polar decomposition, the method is a first-order, dataset-light alternative to machine-learning-based channel identification."],"supporting_citations":[{"why":"Supplies the polar-decomposition optimality result, that the unitary polar factor is the nearest unitary matrix, used to derive monotone decrease in Theorem 5.","marker":"[2]"},{"why":"Provides the Stiefel manifold tangent-space and projection formulas used to characterize critical points and justify the polar-decomposition update.","marker":"[4]"},{"why":"Gives the theorem that the number of nonsingular solutions to a polynomial system is finite, used to bound the number of critical points for almost all pairs (σ, ρ).","marker":"[1]"},{"why":"Supplies the real-analysis lemma that a bounded scalar sequence with finitely many limit points and vanishing consecutive gaps converges, used in Theorem 6.","marker":"[5]"},{"why":"Supplies the iterative-map convergence framework for a continuous map on a compact set with finitely many accumulation points, used to exclude cyclic behavior.","marker":"[19]"},{"why":"Establishes the baseline that 4n²−2n−4 Hermitian matrices are needed for generic unitary channel tomography, against which the paper's n²+3n count is compared.","marker":"[20]"}],"fun_headline_variants":["Iterative polar decomposition recovers quantum unitary from few pairs","Quantum channel search: one pair plus n²+3n observables","New algorithm: unitary channel from input-output pairs","Polar decomposition trick identifies unitary quantum channel","Recover a unitary channel with n²+3n measurements"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that $I-(U^{(s)})^*U^{(s+1)}$ is positive semidefinite at every iteration, but for two unitary matrices this matrix is generically not even Hermitian, so the monotone decrease and convergence to a critical point are not actually established.","fun_headline_variants_meta":{"raw":{"variants":["Iterative polar decomposition recovers quantum unitary from few pairs","Quantum channel search: one pair plus n²+3n observables","New algorithm: unitary channel from input-output pairs","Polar decomposition trick identifies unitary quantum channel","Recover a unitary channel with n²+3n measurements"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000632,"raw_usage":{"total_tokens":2858,"prompt_tokens":826,"completion_tokens":2032,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":442,"completion_tokens_details":{"reasoning_tokens":1952}},"tokens_in":442,"tokens_out":2032,"duration_ms":13308,"temperature":1.0,"reasoning_tokens":1952,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T22:26:42.172731+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run Algorithm 1 from $U^{(0)}=I$ with $\\sigma=\\operatorname{diag}(2,1)$ and $\\rho=\\operatorname{diag}(1,2)$: at the first step the matrix $I-(U^{(0)})^*U^{(1)}$ is generically not Hermitian positive semidefinite, so checking whether $g(U^{(1)})>g(U^{(0)})$ directly tests the claimed monotone-decrease guarantee.","supporting_citations":[{"cited_title":"Higham, Computing the polar decomposition—with applications , SIAM Journal on Scientific and Statistical Computing, 7(4):1160–1174, 1986","cited_arxiv_id":null,"evidence_quote":"Supplies the polar-decomposition optimality result, that the unitary polar factor is the nearest unitary matrix, used to derive monotone decrease in Theorem 5."},{"cited_title":"Lin, Quantify- ing measurement-induced disturbance to distinguish correlations as classical or quantum, Journal of Scientific Computing, 99(1):Paper No","cited_arxiv_id":null,"evidence_quote":"Provides the Stiefel manifold tangent-space and projection formulas used to characterize critical points and justify the polar-decomposition update."},{"cited_title":"Sommese and Charles W","cited_arxiv_id":null,"evidence_quote":"Gives the theorem that the number of nonsingular solutions to a polynomial system is finite, used to bound the number of critical points for almost all pairs (σ, ρ)."},{"cited_title":"Chu and J","cited_arxiv_id":null,"evidence_quote":"Supplies the real-analysis lemma that a bounded scalar sequence with finitely many limit points and vanishing consecutive gaps converges, used in Theorem 6."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the iterative-map convergence framework for a continuous map on a compact set with finitely many accumulation points, used to exclude cyclic behavior."},{"cited_title":"https: //doi.org/10.1063/1.4867625 20","cited_arxiv_id":null,"evidence_quote":"Establishes the baseline that 4n²−2n−4 Hermitian matrices are needed for generic unitary channel tomography, against which the paper's n²+3n count is compared."}],"review_version":1}