{"id":"b36a95f8-b479-4d30-990a-3597e760049b","arxiv_id":"2506.10483","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Post-processing orthogonalization of time-delayed quadrature correlation measurements reconstructs the covariance matrix of highly multimode Gaussian light states.","lead":"This paper proposes a tomography method that reconstructs highly multimode Gaussian states of light from time-delayed quadrature correlation measurements, without needing to know the optical mode structure in advance. If it works in practice, it would let researchers characterize ultrafast quantum light pulses in the THz to mid-infrared range, where standard homodyne detectors struggle.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. (7) mixes bases in the vacuum-compensation term; for rank-deficient Z_LO it does not recover the projected covariance.","rationale":"The reader's weakest_assumption focuses on completeness of the measured subspace. That is a valid practical limitation, but the most load-bearing issue is more fundamental: the central reconstruction equation, Eq. (7), is basis-inconsistent in the rank-deficient case. Deriving from Eq. (6), with Z_LO = U Σ V^T and Q = V Σ^+, one obtains Q^T corr Q = P U^T cov_ρ U P - 1/2 P, so the vacuum compensation in the rotated basis must be 1/2 P, not 1/2 U P U^T. The paper's printed form mixes the rotated-basis left side with an original-basis projector on the right. In the full-rank case P=I the two coincide, which may explain why the error escaped notice; but the entire method relies on thresholding singular values, so the rank-deficient case is the intended operating regime. The correct formula is readily recoverable, so the intended algorithm is likely sound, but the paper as written does not specify it correctly. This strengthens the case for a CONDITIONAL verdict requiring correction of Eq. (7) and a numerical demonstration of the reconstruction on a rank-deficient example. The completeness concern also remains relevant and should be addressed, for instance by a convergence check increasing N and bandwidth. Overall, the reader's CONDITIONAL verdict is unchanged.","tokens_in":39011,"tokens_out":23469,"duration_ms":267895,"concrete_test":"Simulate a two-mode Gaussian state with a known 4x4 covariance C (imax=2, so 4 phase-space dimensions). Use N=1 time delay, producing Z_LO as a 4x2 matrix with rank 2, and compute corr = Z_LO^T C Z_LO - 1/2 Z_LO^T Z_LO. Apply Eq. (7) literally and compare with U P U^T C U P U^T, the exact projection of C onto the measured subspace. Then apply the corrected formula P U^T C U P = (V Σ+)^T corr V Σ+ + 1/2 P and conjugate by U. If the literal result differs outside the measured subspace and the corrected version matches exactly, Eq. (7) as printed is inconsistent and must be revised.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In Sec. II, Eq. (7) states P cov_{ρ,U} P = (V Σ+)^T corr V Σ+ + 1/2 U P U^T. The left-hand side and the first right-hand term are expressed in the U-rotated basis (because cov_{ρ,U} = U^T cov_ρ U), while the vacuum term 1/2 U P U^T is expressed in the original basis. When Z_LO is rank-deficient, which is the generic situation after the 10^-3 singular-value cutoff, P is not the identity and U P U^T does not equal P. Adding U P U^T directly to a rotated-basis matrix couples measured and unmeasured coordinates, so Eq. (7) does not return the projection of the state covariance onto the measured subspace. The correct identity is P U^T cov_ρ U P = (V Σ+)^T corr V Σ+ + 1/2 P, or, in the original basis, U P U^T cov_ρ U P U^T = U (V Σ+)^T corr V Σ+ U^T + 1/2 U P U^T. The basis mismatch is a concrete error in the central reconstruction equation, whereas the completeness issue raised by the reader is a separate limitation: the method can only recover the projection onto the span of the time-delayed local oscillators, and no self-check certifies that this span covers all significant modes.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes 'correlation tomography,' a Gaussian-state reconstruction scheme based on time-domain quadrature correlation measurements with two independently time-delayed local-oscillator pulses. The central idea is to assemble the measurement correlation matrix from many delay settings and then orthogonalize the local-oscillator states in post-processing via SVD, recovering the covariance matrix of a highly multimode Gaussian state projected onto the spanned temporal-mode subspace. The scheme is developed for both homodyne detection and electro-optic sampling, the latter including a nonperturbative treatment of the nonlinear detection interaction and optimization of the probe amplitude. The paper further analyzes thermalization and quantum correlations in the detected state, and derives the joint measurement statistics for non-Gaussian states, showing how correlation measurements can extract spectral information about pulsed Fock states.","tokens_in":39327,"tokens_out":6234,"duration_ms":79058,"significance":"If the central reconstruction formula is correct, the paper offers a useful conceptual advance: it avoids the need for a priori knowledge of the mode basis, which is a known bottleneck in multimode continuous-variable tomography. The detailed symplectic derivations, the Magnus-expansion treatment of electro-optic sampling, and the closed-form Fock-state joint statistics are valuable and go beyond previous work. The claimed scaling of the number of reconstructable modes with the number of time delays and LO bandwidth is falsifiable and testable. However, the core reconstruction equation contains a basis/sign error in the rank-deficient case, and the numerical section does not actually demonstrate reconstruction of a known multimode state; the significance is therefore conditional on these points being fixed.","major_comments":[{"comment":"Equation (7) is not correct as written when Z_LO is rank deficient. With Z_LO = U\\Sigma V^T and P = \\Sigma\\Sigma^+, the left-hand side P cov_{\\rho,U} P is expressed in the U-rotated basis, while the vacuum term (1/2) U P U^T is expressed in the original basis. Since U P U^T \\neq P when P \\neq I, this term couples measured and unmeasured coordinates. The correct identity is P cov_{\\rho,U} P = (V\\Sigma^+)^T corr V\\Sigma^+ + (1/2) P if corr is the raw measurement correlation matrix, or with no additive term if corr already has the vacuum contribution subtracted as in Eq. (6). Please correct this equation and re-derive the reconstruction step; it is the central formula of the paper.","section":"Sec. II, Eq. (7)"},{"comment":"The numerical demonstration does not include an end-to-end reconstruction test. The paper shows the rank of the projector P and the reconstructed mode functions, but never generates a known multimode squeezed state, applies the protocol, and compares the reconstructed covariance matrix with the true one (e.g., via fidelity or trace distance). Without such a test, the claim that the scheme 'reconstructs' highly multimode Gaussian states is not fully supported, especially because the singular-value cutoff in step 3 projects the state onto a subspace. Please add a numerical reconstruction experiment with a known input state and report the reconstruction error.","section":"Sec. II, Fig. 2"},{"comment":"The paper correctly notes that only the projection onto the span of the time-delayed local-oscillator vectors is recovered, and that singular values below 10^-3 times the maximum are discarded. For an unknown state, however, there is no self-check that the chosen delay grid and LO bandwidth cover all significant modes. The claim of 'full' tomography is therefore stronger than what the algorithm guarantees. Please either provide a convergence criterion (for example, increasing the number of delays and checking stabilization of the reconstructed covariance) or explicitly state that the scheme performs partial tomography of the spanned subspace; this limitation should be acknowledged in the abstract and conclusions.","section":"Sec. II, step 3, and Fig. 2a"}],"minor_comments":[{"comment":"Please clarify that Z_LO is assembled from column vectors and state its dimensions explicitly; the notation (\\zeta_LO(\\Gamma_i) | i \\leq 2N) is ambiguous.","section":"Eq. (5)"},{"comment":"The symbol cov_{\\rho,U} is introduced only in this equation. Please define it explicitly in the text and, after correcting Eq. (7), state whether the reconstructed matrix is in the U-rotated basis or in the original basis.","section":"Eq. (7)"},{"comment":"The symbols \\sigma_x and \\sigma_p are used both for singular values of the Schur complement and for quadrature variances. Please use distinct notation to avoid confusion.","section":"Sec. IV and Methods C"},{"comment":"The two panels in Fig. 2a (homodyne and electro-optic sampling) are visually crowded; separate color bars and clearer panel labels would improve readability.","section":"Fig. 2"},{"comment":"The related work in Ref. [84] is mentioned only in a note. Since it appears close in content, please briefly discuss its relation to the present scheme in the introduction or conclusion.","section":"Note added"}],"recommendation":"major_revision","confidential_remarks":"The error in Eq. (7) appears to be a correctable algebraic mistake rather than a fundamental flaw, but because it is the core reconstruction formula it must be fixed before publication. The absence of an end-to-end numerical reconstruction strengthens the need for revision. I would be willing to review a revised version."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The core idea is worth taking seriously, but the central reconstruction equation has a genuine basis bug, and the paper never shows an end-to-end reconstruction. The stress-test note is right, and it lands harder than the reader's report suggests.\n\nWhat is new and good: the SVD-based orthogonalization of time-delayed local oscillator pulses is a real step beyond the same group's earlier time-domain tomography [49]. The electro-optic sampling extension with the non-perturbative treatment of the nonlinear interaction is substantial, and the analysis of thermalisation during detection is thoughtful. The joint statistics for Fock states is a useful opening toward non-Gaussian states. The supplement is detailed and technically rich.\n\nThe soft spot is in Sec. II, Eq. (7). The paper defines cov_{ρ,U} = U^T cov_ρ U and P = ΣΣ+, then writes P cov_{ρ,U} P = (VΣ+)^T corr VΣ+ + 1/2 U P U^T. The first term on the right is in the rotated U-basis; the vacuum term 1/2 U P U^T is in the original basis. When Z_LO is rank-deficient, which is exactly the generic situation after their 10^-3 singular-value cutoff, U P U^T is not equal to P, so the equation does not return the projected covariance. The correct vacuum term is 1/2 P, with the whole equation consistently in the rotated basis (or equivalently rotate the entire equation back). This is not cosmetic: the cutoff is an integral part of the algorithm, and as written the reconstruction step is mathematically inconsistent. The paper needs a corrected equation and a numerical demonstration that the reconstructed projected covariance actually matches the true one.\n\nThe reader's completeness concern is real but separate: there is no self-check that a chosen delay grid and LO bandwidth cover all significant modes, so the method can only recover the projection onto the measured subspace, and the user is given no way to certify completeness. Also, despite the phrase \"theoretically demonstrate,\" there is no end-to-end reconstruction of a multimode Gaussian state in the paper. Fig. 2 shows mode ranks and shapes, not covariance reconstruction fidelity. No uncertainty estimates are given either. These are fixable, but currently missing.\n\nWho is this for: people working on ultrafast quantum optics, electro-optic sampling, and multimode CV tomography. The idea has real potential. Send it to peer review, but a serious referee should require a corrected Eq. (7), an end-to-end simulation of the reconstruction, and an explicit discussion of the completeness limitation.","headline":"A promising tomography scheme whose central reconstruction equation has a basis error that must be fixed before the method can be trusted.","tokens_in":39807,"tokens_out":9989,"would_cite":false,"duration_ms":107853,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P50","81V80"],"pacs":["03.65.Wj","42.50.Xa","42.65.Ky"],"model":"deepseek-v4-flash","headline":"Time-delay scans can reconstruct the unknown modes of multimode quantum light.","keywords":["quantum state tomography","quadrature correlation measurements","multimode Gaussian states","electro-optic sampling","time-domain measurement","singular value decomposition","subcycle resolution","temporal modes"],"falsifier":"Prepare a multimode squeezed vacuum whose third principal mode is centered outside the local-oscillator bandwidth and scan only delays that cover the first two modes. If the algorithm's recovered covariance matrix is claimed to be complete, it will instead return rank two, and the measured correlation matrix will contain residuals above the shot-noise floor that cannot be explained by the reconstructed state; demonstrating such residuals would falsify the completeness of the reconstruction.","tokens_in":1764,"feed_emoji":"🔬","tokens_out":1588,"duration_ms":118571,"temperature":0.7,"pith_summary":"This paper claims that the complete second-moment information of a highly multimode Gaussian state of light can be reconstructed from time-domain quadrature correlation measurements, without knowing the optical mode basis in advance. The key move is to use two short local-oscillator pulses that are independently time-delayed against the quantum pulse, so that each measurement picks out a different, overlapping temporal window. A singular value decomposition performed on the measured correlation matrix orthogonalizes those overlapping windows in post-processing and yields the covariance matrix projected onto the accessible mode subspace. The number of reconstructable modes is set by the number of time delays and the bandwidth of the local oscillator. If correct, this supplies a full tomography route for multimode Gaussian light and, via electro-optic sampling, extends subcycle-resolution measurements into the THz-to-mid-infrared range.","feed_headline":"Delay scans reconstruct the modes of quantum light","feed_subtitle":"A new tomography scheme recovers Gaussian states without knowing the mode basis, reaching subcycle THz dynamics.","key_machinery":"The central object is the singular value decomposition of the matrix Z_LO of time-delayed local-oscillator phase-space vectors. Because the two local-oscillator pulses overlap in time, the measured correlators mix the underlying orthogonal modes; the SVD orthogonalizes these overlapping vectors, the projector P = Sigma Sigma^+ selects the numerically well-conditioned directions, and singular values below $10^{-3}$ times the maximum are set to zero to avoid instability. The same formalism covers both homodyne detection and electro-optic sampling, with the electro-optic case replacing the local-oscillator vector by its image under the symplectic transformation M_NL($\\alpha$) generated by the nonlinear crystal.","core_discovery":"The central claim is that for a multimode Gaussian state, whose Wigner function is entirely fixed by its covariance matrix, the covariance matrix can be recovered from correlation measurements of two simultaneously detected quadratures at many time delays. Writing the measured correlation matrix as corr = Z_LO^T cov_rho Z_LO - Z_LO^T cov_vac Z_LO, where Z_LO collects the time-delayed local-oscillator phase-space vectors, the singular value decomposition Z_LO = U Sigma V^T and the Moore-Penrose pseudoinverse give P cov_rho,U P = (V Sigma^+)^T corr V Sigma^+ + (1/2) U P U^T. This recovers the covariance matrix projected onto the subspace spanned by the local-oscillator states, with the rank of the projector equal to the number of resolved temporal modes, and it does not require prior knowledge of the mode basis of the quantum state.","pith_inferences":["A natural testable extension is to feed the algorithm a known state engineered with modes outside the chosen delay grid; the recovered covariance should match the true projection, and the residual unmeasured correlations should reveal exactly which modes were missed.","Because the SVD cutoff effectively imposes a low-rank approximation, the scheme could be combined with compressive or adaptive delay-grid strategies to minimize the number of measurements when only a few modes carry significant variance.","The joint-statistics formalism developed for Fock states suggests that the same correlation setup could be used to reconstruct wavefunctions of non-Gaussian THz sources, not just their spectral information.","The mode-independent entanglement discussion in the methods hints that correlation data could certify entanglement in a way that does not require choosing a mode basis in advance."],"forward_implications":["A highly multimode Gaussian state can be tomographed from time-domain data alone, with no prior mode basis, by scanning N time delays and reading out the resulting rank(P) modes.","The number of reconstructable modes grows linearly with the number of time delays until a plateau set by the local-oscillator bandwidth and the SVD cutoff, so shorter local oscillators resolve more modes.","In the electro-optic-sampling implementation, the reconstructed modes are shifted toward lower frequencies, reaching the THz-to-mid-infrared range and potentially resolving subcycle dynamics.","Time-local measurements are insufficient for strongly squeezed or thermal states because detection thermalises the state through entanglement breaking; two-time correlation measurements are necessary, and they reveal quantum correlations such as discord even when entanglement vanishes.","The derived full joint statistics also allow spectral information about pulsed Fock states to be extracted, since the compatibility of the two quadrature measurements oscillates with twice the carrier frequency of the quantum pulse."],"supporting_citations":[{"why":"Supplies the time-domain quantum state tomography model, including the thermalisation due to entanglement breaking that motivates the correlation measurement.","marker":"[49]"},{"why":"Shows why time-local subcycle sampling cannot recover the principal modes of a multimode squeezed state, the gap this paper addresses.","marker":"[50]"},{"why":"Provides the first direct electro-optic sampling of vacuum fluctuations, the experimental basis for the electro-optic implementation.","marker":"[54]"},{"why":"Demonstrates electric-field correlation measurements on the vacuum state, the starting point extended here to arbitrary Gaussian states.","marker":"[56]"},{"why":"Gives the paraxial theory of direct electro-optic sampling of the quantum vacuum used for the nonlinear interaction model.","marker":"[57]"},{"why":"Supplies the Gaussian-state formalism and the covariance-matrix description of multimode Wigner functions.","marker":"[3]"},{"why":"Provides the symplectic transformation tools and Gaussian-state background used in the reconstruction and correlation analysis.","marker":"[32]"},{"why":"Defines the Moore-Penrose pseudoinverse used in the SVD orthogonalization step.","marker":"[82]"},{"why":"Defines the generalized inverse used in the reconstruction formula for the covariance matrix.","marker":"[83]"},{"why":"Related independent work on electro-optic sampling of Gaussian light in the perturbative regime, identified by the authors as overlapping their approach.","marker":"[84]"}],"fun_headline_variants":["Time-delayed correlations unveil quantum modes","Quantum state tomography without mode basis","Delay scans reconstruct multimode quantum light","Subcycle THz quantum light tomography via correlations","Correlation delays recover unknown quantum modes"],"cache_read_input_tokens":41984,"weakest_assumption_plain":"The reconstruction assumes that the measured set of time-delayed local-oscillator states spans the subspace containing every mode that contributes significantly to the quantum state; if the delay grid or the local-oscillator bandwidth misses a significant mode, the SVD projector silently discards it and the recovered covariance is incomplete.","fun_headline_variants_meta":{"raw":{"variants":["Time-delayed correlations unveil quantum modes","Quantum state tomography without mode basis","Delay scans reconstruct multimode quantum light","Subcycle THz quantum light tomography via correlations","Correlation delays recover unknown quantum modes"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000719,"raw_usage":{"total_tokens":3273,"prompt_tokens":1034,"completion_tokens":2239,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":650,"completion_tokens_details":{"reasoning_tokens":2176}},"tokens_in":650,"tokens_out":2239,"duration_ms":19888,"temperature":1.0,"reasoning_tokens":2176,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T04:25:52.298385+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Prepare a multimode squeezed vacuum whose third principal mode is centered outside the local-oscillator bandwidth and scan only delays that cover the first two modes. If the algorithm's recovered covariance matrix is claimed to be complete, it will instead return rank two, and the measured correlation matrix will contain residuals above the shot-noise floor that cannot be explained by the reconstructed state; demonstrating such residuals would falsify the completeness of the reconstruction.","supporting_citations":[{"cited_title":"Leonhardt and H","cited_arxiv_id":null,"evidence_quote":"Supplies the time-domain quantum state tomography model, including the thermalisation due to entanglement breaking that motivates the correlation measurement."},{"cited_title":"Wallentowitz and W","cited_arxiv_id":null,"evidence_quote":"Shows why time-local subcycle sampling cannot recover the principal modes of a multimode squeezed state, the gap this paper addresses."},{"cited_title":"Knyazev, K","cited_arxiv_id":null,"evidence_quote":"Provides the first direct electro-optic sampling of vacuum fluctuations, the experimental basis for the electro-optic implementation."},{"cited_title":"Hubenschmid, T","cited_arxiv_id":null,"evidence_quote":"Demonstrates electric-field correlation measurements on the vacuum state, the starting point extended here to arbitrary Gaussian states."},{"cited_title":"Gil-Lopez, Y","cited_arxiv_id":null,"evidence_quote":"Provides the symplectic transformation tools and Gaussian-state background used in the reconstruction and correlation analysis."},{"cited_title":"Lindel, A","cited_arxiv_id":null,"evidence_quote":"Defines the Moore-Penrose pseudoinverse used in the SVD orthogonalization step."},{"cited_title":"Schubert, M","cited_arxiv_id":null,"evidence_quote":"Defines the generalized inverse used in the reconstruction formula for the covariance matrix."},{"cited_title":"Langer, M","cited_arxiv_id":null,"evidence_quote":"Related independent work on electro-optic sampling of Gaussian light in the perturbative regime, identified by the authors as overlapping their approach."}],"review_version":1}