{"id":"f8a8f774-add4-41ab-89ec-cd39b246ecb4","arxiv_id":"2504.17873","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A semidefinite program over quadratic phase-space observables computes the Holevo Cramér-Rao bound for Gaussian states with parameters encoded in both the first moments and the covariance matrix.","lead":"This paper introduces a semidefinite program that computes the Holevo Cramér-Rao bound, the fundamental precision limit for estimating several parameters at once, for any Gaussian quantum state. It makes a previously hard infinite-dimensional optimization tractable using only the state's covariance matrix and first moments.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Unproven epsilon-regularization limit for pure normal modes leaves the 'general Gaussian states' HCRB claim conditional; a direct comparison with known pure-state HCRBs would settle it.","rationale":"I read the paper as attempting to prove that the HCRB for any finite-mode Gaussian state can be evaluated exactly by the finite-dimensional SDP in Eq. (36), based on restricting the search for unbiased observables to quadratic polynomials of the canonical operators. For full-symplectic-rank states, the proof is explicit and reasonably solid: App. D reduces D-invariance to the linear equation Re(S_theta) Zbar = -Im(S_theta) Xbar, Eq. D8, which is solvable when Re(S_theta) is invertible, and the resulting SDP then follows from the RLD inner product formula in Eq. (43). The weak point is precisely the extension to states with pure normal modes, where Re(S_theta) is not invertible. The paper's own Sec. IIB acknowledges this and proposes a covariance-matrix regularization, but it does not prove that the epsilon -> 0 limit of the regularized SDP equals the true HCRB, nor that the D-invariance property survives the limit. Because the abstract and the main-text summary claim applicability to 'general Gaussian states', this unproven limiting step is load-bearing. The case studies include pure-mode states (e.g., coherent-state outputs after loss) and compare against analytic formulas, so a targeted numerical or analytic check could either close the gap or reveal a genuine counterexample. I am not raising a claim of incorrectness: the regularization heuristic may well be valid, and the finite-dimensional analogue suggests it should be. But as written, the paper does not supply the missing argument, and the reader's CONDITIONAL verdict is the appropriate response. The only additional observation I would make is that the cited code repository in Ref. [84] is not accessible from the preprint, which hampers reproducibility but is secondary to the mathematical gap. Therefore I do not recommend changing the reader's verdict.","tokens_in":31714,"tokens_out":34875,"duration_ms":369387,"concrete_test":"For a pure-mode model with a known analytic HCRB, e.g. single-mode displaced squeezed vacuum estimating (Re alpha, Im alpha, r) from Ref. [61] or the coherent-state phase-loss model with C_H = 2/|alpha|^2 from Sec. IIC, evaluate Eq. (36) directly at epsilon = 0 using the PSD square root R_theta = sqrt(S_theta) (allowing rank-deficient R), and separately with sigma_epsilon = (1-epsilon)sigma + epsilon I for epsilon = 10^-6, 10^-8, 10^-10, extrapolating to zero. If the direct epsilon = 0 value matches the analytic HCRB and the extrapolated values converge to the same number within solver tolerance, the regularization route is supported; if they disagree, the pure-mode limit is not exact. Cross-check with a truncated-Fock HCRB SDP at increasing cutoff for a pure two-mode Gaussian example.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim requires that for states with pure normal modes the HCRB is exactly the epsilon-to-zero limit of the regularized SDP (Sec. IIB, 'Pathological states with pure normal modes'; App. D). The authors prove D-invariance of the quadratic-observable subspace only when Re(S_theta) is invertible (App. D, Eq. D8), then assert that regularizing sigma -> (1-epsilon)sigma + epsilon I and taking epsilon -> 0 'is equivalent' to the quotient construction, without proof. No argument is given that (i) the HCRB of the regularized states converges to the HCRB of the original pure-mode state, or that (ii) the limit commutes with the minimization in Eq. (36). Since S_theta is singular for pure modes, the SDP data (S_epsilon, D_epsilon) undergo a rank drop at epsilon = 0; optimal values of SDPs are not automatically continuous under such a drop, and kernel directions of S_theta (for example, the squeezed-mode number operator N_b) could in principle serve as zero-cost unbiased observables if their overlap with derivative directions did not vanish. The paper supplies no compatibility check of this kind. This gap is load-bearing because the advertised scope is 'general Gaussian states', which includes the coherent and pure squeezed probes used in the case studies of Sec. IIC.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a phase-space method for evaluating multiparameter quantum Cramér-Rao bounds for Gaussian states. The central object is a finite block-diagonal matrix Sθ, Eq. (27), which represents the RLD inner product between zero-mean observables that are at most quadratic in the canonical operators. Using this inner product, the SLD and RLD QFI matrices are expressed in terms of first-moment and covariance-matrix derivatives, and the Holevo Cramér-Rao bound is reformulated as the semidefinite program in Eq. (36), with analogous SDP formulations for the SLD and RLD scalar bounds. The method is applied to joint estimation of phase and loss and to joint estimation of displacement and squeezing for single- and two-mode Gaussian states, with comparisons against analytic limits and against each other. For states with full symplectic rank the derivation is coherent. For states with pure normal modes, the paper invokes a covariance-matrix regularization, σ → (1−ε)σ + εI, and asserts that the ε→0 limit reproduces the true HCRB without proving the limit; this is the main unresolved point in the claim that the method applies to general Gaussian states.","tokens_in":31999,"tokens_out":19084,"duration_ms":198716,"significance":"If the result holds in the stated generality, it is a substantial contribution to continuous-variable multiparameter quantum metrology: it supplies the first general numerical SDP evaluation of the Holevo bound for Gaussian states, unifies the SLD, RLD, and Holevo bounds in one phase-space framework, and is accompanied by reproducible code and analytic consistency checks. The derivation is parameter-free and the case studies reproduce known limits, such as C^H = 2/|α|² for coherent-state phase-loss estimation. The main weakness is the unproved regularization limit for pure normal modes, which prevents the 'general Gaussian states' claim from being fully established even though the gap appears local and fixable.","major_comments":[{"comment":"The advertised scope is 'general Gaussian states', but the proof of D-invariance of the quadratic-observable subspace is completed only when Re(Sθ) is invertible, i.e. for full symplectic rank, as stated after Eq. (D8). For states with pure normal modes, the paper asserts that σ → (1−ε)σ + εI followed by ε→0 'is equivalent' to the quotient construction, but gives no proof. This is load-bearing: the optimal value of an SDP is not automatically continuous under a rank drop of Sθ at ε=0, and kernel directions of Sθ could in principle interact with the derivative vectors D̄ and change the optimum. I ask the authors to either prove that the ε→0 limit of the regularized SDP equals the HCRB of the original state, or provide a direct derivation in the quotient space; a numerical check for the n=0 displaced-squeezed states of Ref. [61] would be useful but would not replace the proof.","section":"§IIB, 'Pathological states with pure normal modes' and App. D"},{"comment":"The argument that Dρ-invariance follows because the SLD operators of Gaussian states are at most quadratic cites Refs. [63,66], but the explicit phase-space formulas used in those references require invertibility of Re(Sθ). For pure or partially pure states, the SLD is not unique and the cited formulas need qualification. As written, the proof of the central reduction to quadratic observables is therefore incomplete for exactly the states covered by the regularization claim. Please state explicitly the domain of validity of the cited SLD results, or prove invariance of the quadratic subspace directly on the quotient space defined by the kernel of Sθ.","section":"§IIB, 'Optimality of quadratic observables'"},{"comment":"For an input coherent state (r=0) at η=1/2 the output state has σ=I, so Sθ has a kernel and the formulas (32) and (36) are not directly defined without regularization. The reported RLD and HCRB values for this case, e.g. C^R = C^H = 2/|α|², must therefore come from the same ε-regularization, but the text does not say so explicitly when presenting these results. This should be stated and justified together with the regularization proof requested above; otherwise the analytic pure-state results in Sec. IIC and App. E inherit the unproven limit.","section":"§IIC and App. E, coherent-state phase-loss results"}],"minor_comments":[{"comment":"The constraint in Eq. (72) is written 'barX⊤ D̄ = Ip'; the bar over X is missing in the typeset constraint and should be \t5X\t⊤.","section":"Eq. (72)"},{"comment":"The text says 'still by fixing η=1/2 and for a squeezed vacuum state, that is for r=0'; the phrase 'that is for r=0' should read 'that is for α=0', since a squeezed vacuum state has r≠0 in general.","section":"§IIC, after Eq. (88)"},{"comment":"The dimension z=2m(1+2m) counts the full 4m² quadratic coefficients even though A(2) is restricted to symmetric matrices; a sentence explaining that the redundant components are harmless because they do not enter D̄ and cannot reduce the optimal value would help the reader.","section":"Eqs. (27), (36)"},{"comment":"The final portion of the appendix contains garbled glyphs ('/uni00000013...') that appear to be a rendering artifact; if present in the source, the caption and surrounding text should be fixed.","section":"App. E, Figure 3 caption"},{"comment":"The companion notebook is cited as available in Ref. [84], but no URL is given; a permanent link would make the reproducibility claim actionable.","section":"Ref. [84]"}],"recommendation":"major_revision","confidential_remarks":"The paper is within the journal's scope and is likely to be influential if the pure-state regularization gap is closed. I would not insist on a fully general proof if the authors choose to restrict the main theorem to full symplectic rank and present pure-state cases as limits with explicit numerical validation, but the current wording overclaims. Adding a direct comparison with the exact pure-state HCRBs of Ref. [61] at n=0 would be a strong and inexpensive check."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a useful paper. The main contribution is the SDP in Eq. (36) for the HCRB of Gaussian states when parameters enter both first and second moments. Earlier SDPs (Refs. [20,62]) only covered displacement encoding. The authors also put SLD, RLD, and HCRB in one common phase-space inner-product framework, which is a nice unification and makes the difference between the bounds transparent. The formula (43) for the RLD inner product of quadratic observables is a workhorse that will likely be reused.\n\nThe derivation is careful and mostly convincing for full-symplectic-rank states. The block-diagonal S_theta is derived from characteristic functions, the Schur complement step is standard, and the case studies reproduce known analytic limits (e.g., C^H = 2/|alpha|^2 for coherent-state phase-loss). No parameter is fit and no target result is presupposed. That is real evidence.\n\nWhere the paper gets soft is the treatment of states with pure normal modes. In Sec. IIB ('Pathological states...') the authors say that the quotient construction used in the finite-dimensional case is 'equivalent' to regularizing sigma -> (1-epsilon)sigma + epsilon I and taking epsilon -> 0. That equivalence is asserted, not proved. The matrix S_theta has a kernel at epsilon=0, so the SDP data undergo a rank drop; the optimal value of an SDP is not automatically continuous under such changes, and you would need to show that no zero-cost unbiased observable appears in the limit. This is a genuine gap, and it matters because the advertised scope is 'general Gaussian states' and the pure case is used in the examples. I suspect the claim is true and the gap is fixable, but as it stands the theorem is conditional: fully proved for full symplectic rank, asserted for the rest.\n\nA smaller issue: the code is cited as Ref. [84] but no link or repository is provided in the preprint, so the numerics are not independently checkable. That should be fixed.\n\nBottom line: for readers working on multiparameter estimation with Gaussian states, this is a valuable tool. It deserves a serious referee; I would send it to peer review with a request to (i) prove or properly replace the epsilon-limit statement, and (ii) make the code available. I would cite it once the regularization point is settled.","headline":"New SDP for the Holevo bound with Gaussian states is real and mostly sound, but the 'general Gaussian states' claim rests on an unproven regularization limit for pure normal modes.","tokens_in":32518,"tokens_out":2245,"would_cite":true,"duration_ms":21112,"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":"An SDP built from covariance data evaluates the Holevo Cramér-Rao bound for any Gaussian state.","keywords":["multiparameter quantum estimation","Holevo Cramér-Rao bound","Gaussian states","semidefinite programming","quantum Fisher information","continuous-variable systems","phase and loss estimation","squeezing estimation"],"falsifier":"Take a pure single-mode displaced squeezed vacuum state, for which an analytic HCRB is known, solve the SDP with $\\sigma_\\epsilon=(1-\\epsilon)\\sigma+\\epsilon I$ for $\\epsilon=10^{-3},10^{-6},10^{-9}$, and check whether the optima converge to the analytic value; any systematic discrepancy as $\\epsilon\\to0$ would show that the regularised SDP does not reproduce the HCRB for pure Gaussian states.","tokens_in":31501,"feed_emoji":"🎯","tokens_out":10353,"duration_ms":93878,"temperature":0.7,"pith_summary":"The paper claims that the Holevo Cramér-Rao bound (HCRB) for any multimode Gaussian state can be evaluated exactly as a finite-dimensional semidefinite program built only from the covariance matrix, the first-moment vector, and their derivatives with respect to the estimated parameters. Previously the HCRB had an SDP treatment for finite-dimensional systems and for Gaussian displacement estimation, but not for parameters encoded in the covariance matrix. The same construction yields SDP evaluations of the symmetric and right logarithmic derivative bounds, so all three precision limits live in one framework. Two worked examples, joint phase and loss estimation and joint displacement and squeezing estimation, show that the HCRB is tighter than both scalar bounds and that the gap can be substantial. The practical consequence is that the most fundamental multiparameter precision limit for infinite-dimensional Gaussian metrology platforms becomes a routine convex optimisation.","feed_headline":"One SDP computes the Holevo bound for any Gaussian state","feed_subtitle":"Only covariance data is needed, so infinite-dimensional bounds become convex optimisation.","key_machinery":"The central object is the block-diagonal RLD inner-product matrix $S_\\theta=\\tfrac12\\mathrm{diag}(\\sigma_\\theta-i\\Omega,\\,(\\sigma_\\theta-i\\Omega)\\otimes(\\sigma_\\theta-i\\Omega))$, which encodes $\\mathrm{Tr}[\\hat{B}^\\dagger\\rho_\\theta\\hat{A}]=\\bar{B}^\\dagger S_\\theta\\bar{A}$ for zero-mean observables at most quadratic in the canonical operators, with $\\bar{A}=(A^{(1)};\\mathrm{vec}[A^{(2)}])$; its real part $\\mathrm{Re}(S_\\theta)$ is the SLD inner product on the same space. The mechanism is that every ingredient of the HCRB, namely the Gram matrix $Z(X)$, the local unbiasedness constraints, and the Schur-complement form of $V\\ge Z(X)$, is expressed through $S_\\theta$ and the derivative matrix $\\bar{D}$, converting an optimisation over infinite-dimensional operators into a finite-dimensional SDP. The restriction to quadratic observables is justified through the commutation superoperator $\\mathcal{D}_\\rho$, defined by $\\{\\mathcal{D}_\\rho(\\hat{X}),\\rho\\}=i[\\hat{X},\\rho]$, whose action on a quadratic operator for a Gaussian state is the same calculation as an SLD for a unitarily encoded parameter and therefore stays quadratic. Inverting $\\mathrm{Re}(S_\\theta)$ is the step that requires all symplectic eigenvalues to exceed one, which is why pure normal modes need regularisation.","core_discovery":"On the paper's own terms, the central claim is that for an $m$-mode Gaussian state with first moment $d_\\theta$ and covariance matrix $\\sigma_\\theta$, the HCRB equals the optimum of the semidefinite program $$\\mathrm{minimize}\\ \\mathrm{tr}[WV]\\quad \\text{subject to}\\quad \\begin{pmatrix} V & \\bar{X}^\\top R_\\$\\theta$^\\dagger \\\\ R_\\$\\theta$ \\bar{X} & I_r \\end{pmatrix} \\ge 0,\\qquad \\bar{X}^\\top \\bar{D}=I_p,$$ where $S_\\theta=\\tfrac12\\mathrm{diag}(\\sigma_\\theta-i\\Omega,\\,(\\sigma_\\theta-i\\Omega)\\otimes(\\sigma_\\theta-i\\Omega))$ is the RLD inner-product matrix on zero-mean observables at most quadratic in the canonical operators, $S_\\theta=R_\\theta^\\dagger R_\\theta$, and $\\bar{D}$ collects $\\partial d_\\theta/\\partial\\theta_j$ together with $\\tfrac12\\mathrm{vec}[\\partial\\sigma_\\theta/\\partial\\theta_j]$. The constraint $\\bar{X}^\\top\\bar{D}=I_p$ encodes local unbiasedness, and the search over observables can be restricted to quadratic ones because the commutation superoperator of a Gaussian state maps the quadratic subspace into itself. The scalar SLD and RLD bounds are the same optimisation with, respectively, the real part of the inner product or complex-valued $\\bar{X}$, so the three bounds are unified. For states with pure normal modes the matrix $S_\\theta$ is not invertible; the paper's recipe is to regularise $\\sigma\\to(1-\\epsilon)\\sigma+\\epsilon I$ and take $\\epsilon\\to0$, asserting, without a proof, that this limit gives the true HCRB.","pith_inferences":["A natural next step is to read off the optimal $\\bar{X}$ from the SDP solution and construct explicit quadratic measurements that attain the HCRB, something the paper does not do.","If the $\\epsilon\\to0$ regularisation is exact, the method covers pure Gaussian probes; a comparison against the known analytic pure-state HCRB would be a cheap check of that remaining assumption.","The phase-space inner-product formulation should transfer to fermionic Gaussian systems, where a similar quadratic-operator subspace and symplectic structure exists.","Because the paper notes the techniques likely apply to the Bayesian HCRB, one plausible extension is efficient evaluation of global, non-local multiparameter bounds for Gaussian states."],"forward_implications":["The HCRB for Gaussian statistical models with parameters in both first moments and covariance matrix can be evaluated numerically with a global-optimality guarantee, without truncating the infinite-dimensional Hilbert space.","The SLD-CRB and RLD-CRB are obtained from the same SDP by relaxing constraints, making comparisons among all three bounds routine for concrete metrology problems.","In the phase-loss example the HCRB is tighter than both scalar bounds in generic regimes; it coincides with the RLD bound for coherent probes and at specific squeezed-vacuum parameters, while elsewhere it can be substantially above it.","For displacement and squeezing estimation with single- and two-mode displaced squeezed thermal states, the HCRB matches the analytical upper bound in the studied regimes, and the RLD bound approaches it as the thermal photon number grows.","The same SDP structure extends to estimating $q\\le p$ smooth functions of the parameters and to singular statistical models, provided unbiased observables for those functions exist."],"supporting_citations":[{"why":"Supplies the finite-dimensional SDP formulation of the HCRB that this paper adapts to Gaussian states.","marker":"[56]"},{"why":"Provides the existing displacement-only Gaussian SDP that the new construction generalises to parameters in the covariance matrix.","marker":"[20]"},{"why":"Defines the HCRB and the commutation-superoperator argument that allows restricting the search to invariant subspaces.","marker":"[11]"},{"why":"Establishes the asymptotic attainability of the HCRB by collective measurements and the hierarchy relating it to the SLD and RLD bounds.","marker":"[17]"},{"why":"Gives the phase-space derivation that SLD operators for Gaussian states are at most quadratic, a premise of the invariance proof.","marker":"[63]"},{"why":"Provides the covariance-regularisation trick used to handle pure normal modes.","marker":"[67]"},{"why":"Supplies the analytic HCRB for pure single- and two-mode Gaussian states used as a comparison in the displacement-squeezing example.","marker":"[61]"}],"fun_headline_variants":["One SDP computes all three Gaussian quantum bounds","Efficient SDP evaluates Holevo bound for any Gaussian state","Unified SDP for HCRB, RLD, SLD on Gaussian states","Gaussian estimation bounds solved via a single SDP"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Everything rests on the claim that optimal HCRB observables can be chosen among operators at most quadratic in the canonical variables; for states with pure normal modes this claim is supported only by the unproven assumption that regularising the covariance matrix as $\\sigma\\to(1-\\epsilon)\\sigma+\\epsilon I$ and taking $\\epsilon\\to0$ gives the exact HCRB.","fun_headline_variants_meta":{"raw":{"variants":["One SDP computes all three Gaussian quantum bounds","Efficient SDP evaluates Holevo bound for any Gaussian state","Unified SDP for HCRB, RLD, SLD on Gaussian states","Gaussian estimation bounds solved via a single SDP"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000565,"raw_usage":{"total_tokens":2803,"prompt_tokens":1193,"completion_tokens":1610,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":809,"completion_tokens_details":{"reasoning_tokens":1539}},"tokens_in":809,"tokens_out":1610,"duration_ms":10713,"temperature":1.0,"reasoning_tokens":1539,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T10:32:26.843863+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a pure single-mode displaced squeezed vacuum state, for which an analytic HCRB is known, solve the SDP with $\\sigma_\\epsilon=(1-\\epsilon)\\sigma+\\epsilon I$ for $\\epsilon=10^{-3},10^{-6},10^{-9}$, and check whether the optima converge to the analytic value; any systematic discrepancy as $\\epsilon\\to0$ would show that the regularised SDP does not reproduce the HCRB for pure Gaussian states.","supporting_citations":[{"cited_title":"Optimal estimation of joint parameters in phase space","cited_arxiv_id":"1206.4867","evidence_quote":"Provides the existing displacement-only Gaussian SDP that the new construction generalises to parameters in the covariance matrix."},{"cited_title":"A tight Cram\\'er-Rao bound for joint parameter estimation with a pure two-mode squeezed probe","cited_arxiv_id":"1702.02271","evidence_quote":"Gives the phase-space derivation that SLD operators for Gaussian states are at most quadratic, a premise of the invariance proof."}],"review_version":1}