{"id":"472a376f-f33f-489d-9fde-9047ed9e4aa3","arxiv_id":"2501.10118","paper_version":2,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Almost every quantum channel and almost every Lindblad semigroup turns the time series of one non-trivial expectation value into a complete tomographic probe, except unitary evolution with simple depolarizing noise, which remains insufficient beyond qubits.","lead":"The paper proves that for almost any noisy quantum evolution, the time series of a single expectation value contains enough information to reconstruct the full quantum state. This means one binary measurement made repeatedly at different times can, in principle, replace the many different measurements normally required.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection to central claim; Remark 3's all-H_d injectivity claim is false for traceless H0.","rationale":"I read the central claim as the injectivity of the finite time-series map on density matrices for almost every channel and almost every Lindblad semigroup element, with the evolution and times known. The proof in Section 5 is careful and the main steps check out: after removing null sets such as the boundary of C_d, degenerate spectra, and non-invertible channels, the map on traceless perturbations is represented by a Vandermonde matrix times a diagonal matrix whose diagonal entries are tr[X_j^* H_Q]. The determinant is a polynomial in the channel entries; it is not identically zero because one can construct a channel arbitrarily close to the completely depolarizing channel with the required spectral projectors. For the Lindblad case, since L_d has nonempty interior in T_d, the same nonconstant polynomial cannot vanish on all of L_d, so the bad set is null there as well. Thus the central theorem survives scrutiny. The genuine flaw I found is confined to Remark 3: for traceless H0 and generic T, the identity component of the state is invisible in the time series, so injectivity on all Hermitian matrices cannot hold. The reader's weakest assumption concerned the known-channel premise, which is a stated limitation rather than a gap in the theorem. Because the false remark is peripheral and does not affect the density-matrix tomography result, I would keep the reader's ACCEPT verdict unchanged while recommending that Remark 3 be revised or deleted.","tokens_in":17806,"tokens_out":38523,"duration_ms":398606,"concrete_test":"For d=2 take H0 = sigma_z and a Pauli channel T(X) = (1/2) tr[X] 1 + (lambda_1/2) tr[sigma_x X] sigma_x + (lambda_2/2) tr[sigma_y X] sigma_y + (lambda_3/2) tr[sigma_z X] sigma_z with lambda_1, lambda_2, lambda_3 distinct and not all zero. Compute alpha(c1) and alpha(0) for the four-point series i=0..3; both are identically zero because T^i(sigma_z) = lambda_3^i sigma_z, so alpha is not injective on H_d, confirming that Remark 3 overstates the result.","verdict_should_be":"UNCHANGED","load_bearing_attack":"No load-bearing concern for the central claim of Theorem 4 and Corollary 4 on density matrices. The generic-channel argument via Vandermonde determinants and null-set reduction is sound: for generic T with distinct eigenvalues, the time-series map factors through an invertible Vandermonde matrix times a diagonal matrix, and the bad set is a proper real-algebraic hypersurface. The Lindblad variant follows because L_d has nonempty interior. The concrete error I find is in Remark 3 after Theorem 4: it claims that adding one data point makes alpha injective on all of H_d. This is false when H0 is traceless. For generic T with distinct eigenvalues, T^i(H0) has zero trace for every i, because H0 has no component along the fixed point 1 and the left eigenvector 1 is orthogonal to all other eigenspaces. Hence rho and rho + c1 produce identical time series. This does not affect the density-matrix tomography claim, where tr[rho] = 1 is fixed, but Remark 3 should be corrected.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies whether the homogeneous time evolution of a single expectation value can provide enough information for full quantum state or observable tomography in finite dimensions. The main positive results (Thm. 4 and Cor. 4) assert that for every nontrivial observable, almost every quantum channel, and almost every Lindblad semigroup element, the time series of length d^2-1 (for states) or d^2 (for observables) is injective on the relevant parameter set, and that a Takens-like dimension-dependent version holds (Thm. 5). The paper also proves series-extension theorems (Thms. 1-3) that lead to no-go results for unitary and simply depolarizing evolutions beyond qubits (Sec. 4), gives finite-statistics estimation bounds (Sec. 6), and provides a detailed qubit analysis (Sec. 7).","tokens_in":17967,"tokens_out":9794,"duration_ms":69269,"significance":"If correct, the results are significant: they rigorously identify non-trivial noise as a resource for tomography, provide a linear analogue of Takens' embedding theorem adapted to quantum channels, and give quantitative finite-statistics bounds. The proofs are detailed and combine null-set arguments, Sard's theorem, Vandermonde determinantal identities, and known generic-embedding results in a parameter-free way. The paper also gives explicit no-go counterparts, showing that unitary evolution and simply depolarizing noise are insufficient beyond the qubit case, and that jointly recovering an unknown state and an unknown observable is generically impossible. The main claims are clearly scoped: the evolution map is known and fixed, and the time series is generated from an identically prepared ensemble.","major_comments":[{"comment":"The claim that adding one more data point makes alpha injective on all of H_d is false when H0 is traceless. For a generic T with distinct eigenvalues, each T^i(H0) has zero trace, because the left eigenvector 1 is orthogonal to all other eigenspaces and H0 has no component along 1. Hence rho and rho + c1 produce identical time series for every c, so alpha cannot be injective on all of H_d. The statement should be restricted to trace-one states, or one should require H0 to have nonzero trace. This error does not affect the main density-matrix tomography results in Theorem 4 and Corollary 4, but the remark should be corrected.","section":"§5, Remark 3 after Theorem 4"}],"minor_comments":[{"comment":"The phrase 'evolved by any quantum channel, except for a null set' could be misread as 'for any fixed channel, all but a null set of measurements'; the precise quantifier order in Theorem 4 is that for each fixed H0 the set of failing channels is null and vice versa. A short clarification would avoid ambiguity.","section":"Abstract"},{"comment":"The least-squares estimator rho_hat = rho0 + alpha^{-1}(f - alpha(rho0)) is not constrained to be positive semidefinite and may not be a physical density matrix. The authors should note explicitly that this is a linear (possibly unphysical) estimator and that the bound in Thm. 6 is on the Hilbert-Schmidt distance to the true state, not on an estimator constrained to density-matrix space.","section":"§6, estimator in Eq. (25)"},{"comment":"The reference [JAM00] is missing the author names; only the title is listed.","section":"References"},{"comment":"In the proof of Corollary 2 the symbol delta is used both for the Kronecker delta delta_{lambda,0} and for the degree of the minimal polynomial delta(T_lambda); while the context is clear, a different symbol for one of the two would improve readability.","section":"§4, Corollary 2 proof"}],"recommendation":"minor_revision","confidential_remarks":"The false claim in Remark 3 is the only technical error I found; it is localized and does not affect Theorem 4, Corollary 4, or the paper's central conclusions. The use of external results ([HMW13], [Rob10]) as black boxes is appropriate and the generic-channel arguments are sound. The paper should be accepted after the remark is corrected and the minor presentation issues are addressed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The headline result is the positive one, and it is real: for finite-dimensional systems, for any H0 not proportional to identity, almost every quantum channel (and almost every Lindblad semigroup element) makes the finite time series rho -> tr[rho T^i(H0)] injective on density matrices. Same for observables with known state. That settles the qualitative folklore intuition — observability is generic in the quantum linear setting — and the paper does it with clean proofs: Vandermonde determinants, Sard's theorem, null-set arguments. The Takens-inspired Theorem 5 with prior information (Minkowski dimension) is a nice addition, and Theorem 6 gives honest finite-statistics bounds without overselling efficiency. The no-go results for unitary and simply depolarized evolutions are also solid, recovering and extending [MRFD10]. Credit where due: this is the first place I know that states the generic-channel claim at this level of precision, and the proofs are self-contained enough to check.\n\nSoft spots, in proportion. The one concrete error is in Remark 3 after Theorem 4: the claim that one extra data point makes alpha injective on all of Hd is false when H0 is traceless. Because T^i(H0) is traceless for every i (since T* is unital as well), rho and rho + c1 produce identical time series. That does not touch the density-matrix claims, where trace is fixed, but the remark needs correcting. Also, the abstract's wording about \"simply depolarizing noise\" being insufficient for states is a bit loose: Corollary 2 gives insufficiency for rank-r states for a certain range of r, not for all states in dimension 3. Minor presentational point. The framework assumes the channel and measurement times are known exactly; Corollary 3 makes clear that joint recovery of state and observable is impossible, so this is not a hidden flaw — but readers should not expect anything about channel uncertainty.\n\nBottom line: the central theorem is correct as far as I can see, the proofs are honest, and the overclaim is a localized remark, not a load-bearing beam. I would send this to a serious referee and expect it to be accepted after a small revision. I will cite it and would happily discuss it in a reading group.","headline":"A genuinely new genericity result: almost every noisy quantum channel turns a single expectation-value time series into full tomography, with unitary no-go theorems that are clean; one small overclaim in Remark 3 should be fixed.","tokens_in":18500,"tokens_out":1636,"would_cite":true,"duration_ms":18138,"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":"Almost every noisy quantum evolution makes a single measured expectation value enough to reconstruct any quantum state, while unitary-only evolution falls short.","keywords":["quantum state tomography","single expectation value","observability","quantum channels","Lindblad semigroups","Takens embedding theorem","Minkowski dimension","finite statistics"],"falsifier":"Fix $d=3$, choose $H_0=\\operatorname{diag}(1,0,-1)$, and pick distinct times $t_1,\\dots,t_9$. For an open ball of Lindblad generators $L$ parameterized as in Eq. (2), form the $9\\times 8$ matrix with entries $\\operatorname{tr}[X_j^* e^{t_k L}H_0]$, where $X_j$ runs over a traceless Hermitian basis. Corollary 4 says this matrix has rank 8 for all $L$ outside a null set; if its rank drops below 8 on an open set of generators, the paper's continuous-time claim is false.","tokens_in":17591,"feed_emoji":"📈","tokens_out":24992,"duration_ms":228664,"temperature":0.7,"pith_summary":"This paper asks how much of a quantum state can be learned from watching a single expectation value evolve. Its central answer is that, in any finite dimension, almost every noisy quantum evolution makes one such time series tomographically complete: once $d^2-1$ consecutive values are recorded, the state is uniquely determined, and the same is true for an unknown observable when the state is known. The clear exception is unitary evolution, which even with added simply depolarizing noise cannot achieve this beyond the qubit case, so non-trivial noise is a resource rather than an obstacle. The paper also proves a Takens-style embedding theorem for quantum channels, showing that prior information about the state reduces the number of required time points, and it supplies finite-statistics bounds for a least-squares estimator.","feed_headline":"One evolving measurement can identify any quantum state","feed_subtitle":"Almost any noisy quantum channel turns one measured expectation value into full state tomography. Unitary-only cannot.","key_machinery":"The central object is the time-series map $\\alpha(\\rho)=(\\operatorname{tr}[\\rho T^i(H_0)])_{i=t_0}^{t_1}$, restricted to density matrices, and its traceless reduction $\\alpha_0(\\sigma)_i=\\operatorname{tr}[\\sigma T_Q^i(H_Q)]$, where $T_Q=QTQ$ acts on the $(d^2-1)$-dimensional subspace of traceless Hermitians. The argument rides on the factorization $\\alpha_0=\\Lambda M$: $\\Lambda$ is a Vandermonde matrix—powers of the channel's eigenvalues—invertible whenever $T$ has $d^2$ distinct eigenvalues (a generic condition), and $M$ is diagonal with entries $\\operatorname{tr}[X_j^* Q(H_0)]$, the overlaps of the observable with the channel's left eigenmodes. Whenever one of these overlaps vanishes, injectivity fails; otherwise the state is uniquely recoverable. For the Takens-style theorem, the machinery is the map $\\tau(T)=(T(H_0),\\dots,T^m(H_0))$ from channels to $m$-tuples of observables, which has non-zero-measure image and pulls null sets back to null sets, so a generic embedding theorem can be imported.","core_discovery":"On the paper's own terms, almost every quantum channel—and almost every element of a Lindblad semigroup—turns a single non-trivial expectation value into a complete tomographic record: with $d^2-1$ consecutive time steps (or $d^2$ samples in continuous time) the map from density matrices to time series is injective, and the same is true for observables when the state is known. The exceptions are the ones that look noise-free: unitary evolutions, even with simply depolarizing noise, cannot do this beyond qubits, which the paper reads as evidence that non-trivial noise is necessary. The proofs identify a precise algebraic condition for injectivity: after removing the identity component, the time-series map factors as a Vandermonde matrix in the eigenvalues of $T$ times a diagonal matrix of overlaps between the observable and the channel's eigenmodes, so injectivity is lost only on a null set of channels or observables. A Takens-style theorem extends the construction to prior information, and a finite-statistics bound quantifies how many shots are needed.","pith_inferences":["Beyond the paper: the factorization in the proof gives a pre-certification test—compute the overlaps $\\operatorname{tr}[X_j^* Q(H_0)]$ from the channel's eigen-decomposition and check that none vanish—so an experimentalist can verify a channel's tomographic power before collecting data.","Beyond the paper: Corollary 3 suggests that evolution-based tomography is best used one-sided: fix and calibrate either the observable or the state independently, then use a single time series to reconstruct the other; attempting to reconstruct both from the same record is generically hopeless.","Beyond the paper: the results suggest replacing random measurement design with random evolution design—a single generic fixed channel produces an informationally complete record, so the randomness enters through the channel rather than through many measurement settings."],"forward_implications":["For almost every quantum channel (and almost every Lindblad semigroup element), $d^2-1$ consecutive values of a single expectation value determine the density matrix uniquely, so no family of different measurements is required once the evolution is generic.","With the state known, $d^2$ time points determine an unknown observable for almost every channel; in continuous time, $d^2$ samples at arbitrary distinct times suffice.","Unitary evolutions—even with simply depolarizing noise—are insufficient beyond qubits, so non-trivial noise is a genuine resource for evolution-based tomography.","Prior information about the state reduces the data requirement: if states are confined to a set $S$, then $m>D(S-S)$ time points suffice generically, and the reconstruction is Hölder-stable.","Under finite statistics, a least-squares estimator built from the time series obeys $\\mathbb{E}\\|\\hat{\\rho}-\\rho\\|_2^2 \\le \\|\\alpha^{-1}\\|^2\\Delta^2(d^2-1)/(4n)$, so the smallest singular value of the time-series map controls the sample complexity."],"supporting_citations":[{"why":"Supplies the delay-embedding theorem that the paper adapts to quantum channels; the positive results are framed as a linear counterpart.","marker":"[Tak81]"},{"why":"Provides the generic m-dimensional embedding theorem and the Minkowski-dimension properties used in Theorem 5 and Corollary 1.","marker":"[Rob10]"},{"why":"Transfers the generic-measurement theorem to quantum tomography under prior information, giving the m > D(S-S) criterion used in Theorem 5.","marker":"[HMW13]"},{"why":"Supplies the matrix-analysis backbone: minimal-polynomial degree, Jordan decomposition, and Vandermonde determinants used in Theorems 1-4 and Lemma 3.","marker":"[HJ13]"},{"why":"Gives the null-set property of zero sets of real analytic functions, used to show degenerate spectra and non-injective channels form null sets.","marker":"[Mit20]"},{"why":"Prior result that repeated unitary evolution cannot reconstruct states for d >= 3, which Corollary 2 recovers and extends to observables.","marker":"[MRFD10]"},{"why":"Characterizes Lindblad generators as conditionally completely positive maps, used in Lemma 1 to show semigroup elements form a non-null set.","marker":"[Eva77]"},{"why":"Computes the Minkowski dimension of rank-r density matrices, used in Corollary 2 to compare state-set dimension with minimal-polynomial degree.","marker":"[GKM05]"}],"fun_headline_variants":["One expectation value in time pins down any quantum state","Full tomography from a single evolving expectation value","One measurement over time reveals any quantum state","Noise makes a single expectation enough for full tomography"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the time-evolution map and the single observable are known exactly and held fixed while the time series is gathered—so that the theorems cover recovery of the state (or observable) alone, not joint recovery of unknown state, observable, or channel, which Corollary 3 shows is generally impossible.","fun_headline_variants_meta":{"raw":{"variants":["One expectation value in time pins down any quantum state","Full tomography from a single evolving expectation value","One measurement over time reveals any quantum state","Noise makes a single expectation enough for full tomography"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000536,"raw_usage":{"total_tokens":2539,"prompt_tokens":874,"completion_tokens":1665,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":490,"completion_tokens_details":{"reasoning_tokens":1606}},"tokens_in":490,"tokens_out":1665,"duration_ms":13064,"temperature":1.0,"reasoning_tokens":1606,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T19:26:50.019569+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Fix $d=3$, choose $H_0=\\operatorname{diag}(1,0,-1)$, and pick distinct times $t_1,\\dots,t_9$. For an open ball of Lindblad generators $L$ parameterized as in Eq. (2), form the $9\\times 8$ matrix with entries $\\operatorname{tr}[X_j^* e^{t_k L}H_0]$, where $X_j$ runs over a traceless Hermitian basis. Corollary 4 says this matrix has rank 8 for all $L$ outside a null set; if its rank drops below 8 on an open set of generators, the paper's continuous-time claim is false.","supporting_citations":[],"review_version":1}