{"id":"fe722639-90ca-455e-b3fe-8c3718baf073","arxiv_id":"2412.01074","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"One nonclassical state plus one coherent state is both necessary and sufficient for Heisenberg-limited estimation of arbitrary weighted combinations of distributed phase shifts.","lead":"This paper proves that one nonclassical light source plus one ordinary laser is the minimum needed to measure arbitrary combinations of many sensor phases at the ultimate quantum precision limit. It also gives a general formula for the best achievable error in such distributed networks.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Universal lower bound Eq. (4) is not fully proven: S2 neglects off-diagonal N terms and assumes u⊥v remains optimal when N is present, so 'no network can beat the bound' is not established.","rationale":"The reader's weakest assumption and the present stress-test converge on the same point: the explicit construction appears correct, but the 'universal' status of the lower bound in Eq. (4) depends on an approximate treatment of the standard-quantum-limit matrix N in Supplemental S2. Because the authors themselves list the two limitations, it would be unfair to call this a hidden flaw; rather, it is an acknowledged proof gap that is directly load-bearing. If the numerical test shows no counterexample, the gap is benign and the paper strengthens to ACCEPT. If a counterexample appears, the universal lower bound must be weakened, though the constructive Heisenberg-limited scheme could survive. Therefore the existing CONDITIONAL verdict is the right calibration: do not reject a paper whose central construction is explicit and reproducible, but do not certify the universal claim until the S2 approximation is either removed or validated. The concern is not about the authors' integrity or about disagreement with consensus; it is about a specific mathematical step, namely the neglected off-diagonal cross terms e_j^T N n_j in the inverse of the QFIM.","tokens_in":19046,"tokens_out":26098,"duration_ms":234216,"concrete_test":"Perform a numerical search over 4 × 4 unitaries U for a concrete case: d = 2, weights w = (0.4, -0.4, 0.1, -0.1), squeezed-vacuum and coherent inputs with n1 = n2 = 10, and cs = 0. For each sampled U, build the exact QFIM F = cu uu^T + cv vv^T + 4 Diag(n1 |U_j1|^2 + n2 |U_j2|^2), with u_j = |U_j1|^2 and v_j = U_j1 U_j2^*, and compute Δ² = 4 w^T F^{-1} w by exact matrix inversion. If any unitary gives Δ² < 1/(N + 2 n2 W), the universal bound of Eq. (4) is violated; if the minimum is attained at the chosen aligned network and never falls below the bound, the S2 approximation is benign and the proof can be completed. Repeat with a state with cs ≠ 0, since off-diagonal N cross terms are the suspected loophole.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The constructive part of the paper is on solid ground: for a target weight vector w, the choice U_j1 = sqrt(|w_j|), U_j2 = w_j/sqrt(|w_j|) gives v = w, u ⊥ v, and N = 4N Diag(|w_j|). Under this alignment, Eq. (S.14) is an exact inverse and Eq. (S.15) follows, so the proposed network really attains the stated sensitivity. The load-bearing gap is the preceding claim that this is a universal lower bound over all linear networks. In Supplemental S2, the inverse of F is approximated by dropping e_j^T N n_j cross terms in the diagonalized basis (S.12), and the optimum condition w = v, u ⊥ v is derived from that approximate inverse. The authors explicitly acknowledge both limitations: off-diagonal terms of N are neglected, and u ⊥ v is only guaranteed optimal when N is ignored. If an off-axis configuration with u · v ≠ 0, or with w not exactly parallel to v, can use the N cross terms to decrease w^T F^{-1}w below the value in Eq. (S.15), then the universal part of Eq. (4) fails, even though the explicit construction remains a valid Heisenberg-limited scheme. The exact bound for the aligned network is necessary for the paper's main result, but it is not sufficient to prove that no other network can do better.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies continuous-variable distributed quantum metrology in a linear optical network with two non-vacuum inputs: one single-mode nonclassical state (pure or mixed) and one coherent state, with 2d modes arranged as d phase-reference pairs. The central object is the quantum Fisher information matrix, which the authors decompose as F = c_u uu^T + c_v vv^T + c_s(uv^T+vu^T) + N, Eq. (3), where u and v encode the network action and c_v = 8 n_2 W with W the metrological power. On this basis the paper claims: (i) two inputs, only one nonclassical, are necessary and sufficient for Heisenberg-limited estimation of an arbitrary linear combination of the d phases; (ii) a universal lower bound of the form Δq ≥ 1/√(N + 2 n_2 W), Eq. (4), tight when c_s = 0; (iii) local photon-number detection saturates this bound for a class of nonclassical states; (iv) arbitrary analytic functions of the phases can be estimated with the same scaling; and (v) the network has two operating regimes, one at the Heisenberg limit and one where a weak nonclassical state multiplicatively enhances classical sensitivity. The supplemental material contains the inverse-QFIM calculations, the mixed-state generalization, a (d+1)-mode reduced scheme, a four-mode example, and the photon-counting optimality derivation.","tokens_in":1661,"tokens_out":1903,"duration_ms":80154,"significance":"If the universal bound is valid, this is a substantial and useful unification: it reduces to known two-mode Mach-Zehnder results at d = 1, identifies the metrological power W as the resource controlling Heisenberg scaling, and provides a concrete four-mode network that realizes arbitrary two-parameter weights. The constructive part is carefully executed and convincing: for the choice U_j1 = √|w_j| and U_j2 = w_j/√|w_j| one has v = w, u ⟂ v, and N = 4N diag(|w_j|), so that the inverse in Eq. (S.14) is exact and Eq. (S.15) gives the advertised sensitivity; the d=1 limit checks out; and the extension to mixed input states through generalized coefficients in Supplemental S4 is well organized. The gap is in the converse direction: the claim that no other linear network can do better than the constructed one is not established, because the optimization over the network in Supplemental S2 is performed with an approximate inverse. The paper is therefore a strong candidate for publication once the universality claim is either proven exactly or properly restricted.","major_comments":[{"comment":"The claimed universal lower bound is not proven as stated. The derivation of the optimal alignment uses Eq. (S.12), in which the inverse of F is approximated by dropping the off-diagonal terms e_j^T N n_j, and the condition u ⟂ v is carried over from the N = 0 analysis. The authors explicitly acknowledge in S2 that \"we neglected the off-diagonal terms when we expanded N\" and that \"v⊥u may not always hold\" when N is included. Since Eq. (4) is used to conclude that no network can beat the constructed one, this is a load-bearing step. I ask the authors either to provide an exact proof that for every unitary U the quantity w^T F^{-1}w is bounded below by 1/(4N + c_v - c_s^2/(4N + c_u)), or to restate Eq. (4) as the sensitivity of the explicitly constructed aligned network and soften the \"universal\" and \"minimum\" claims accordingly.","section":"Eq. (4) and Supplemental S2"},{"comment":"The necessity of two inputs is argued from the N = 0 form F ≈ c_u uu^T: since u has positive entries, a signed weight vector cannot be parallel to u, so no Heisenberg scaling is obtained. This is a heuristic statement: once N is included, the optimal weighting vector is not exactly parallel to u, and the proof in S2 that the N-induced component contributes only at order 1/O(N) rests on the same approximate inverse flagged in the previous comment. To make the claim that two inputs are the minimum required, the single-input case needs a direct bound showing that for every network, w^T(c_u uu^T + N)^{-1}w cannot scale as 1/N^2 when w has both positive and negative entries.","section":"Single-mode input paragraph"}],"minor_comments":[{"comment":"The phrase \"universal and tight upper bound on the sensitivity\" is misleading: Eq. (4) is a lower bound on the estimation error Δq. Please reword as a lower bound on uncertainty or equivalently an upper bound on sensitivity only if sensitivity is defined as the inverse error.","section":"Abstract and Introduction"},{"comment":"There are several typographical errors: \"discucssed\" should be \"discussed\", and \"su fficient\" should be \"sufficient\". Please proofread the LaTeX source.","section":"Model section"},{"comment":"The notation 1/O(N) in Eq. (S.13) is informal: it should say that the term is of order 1/N with the constant depending on the omitted cross terms, rather than using the symbol O(N) as a scalar denominator.","section":"Supplemental S2, Eq. (S.13)"},{"comment":"The saturation result for local photon-number detection matches Eq. (4) only when α1 = β1 = 0 and either the weights are uniform, (||w||_3)^3 = |w|^4, or n1, n2 ≫ 1. The main text should state these conditions explicitly when claiming that photon-number detection achieves the maximum sensitivity, rather than leaving them only in the supplement.","section":"Supplemental S6, Eq. (S.26)"},{"comment":"The lower panel legend would benefit from stating explicitly that the solid curves are from Eq. (4) and that the dashed lines mark the SQL and HL scaling; currently the distinction between solid and dashed curves is not described in the caption.","section":"Figure 2"}],"recommendation":"major_revision","confidential_remarks":"The constructive part of the paper is solid and the QFIM decomposition is valuable. The main risk is overclaiming the universality of Eq. (4): the S2 inverse is approximate in a way that touches the central \"no network can beat this\" assertion. If the authors can supply an exact optimization or honestly restrict the claim, the paper would be publishable. I would not recommend rejection because the gap is localized and likely fixable within the manuscript's scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Take a look at this one if you care about CV distributed sensing. The genuinely new thing is the minimality theorem: two non-vacuum inputs, one coherent and one nonclassical, are necessary and sufficient for Heisenberg-limited estimation of arbitrary linear combinations of d phases, and the construction is explicit (set U_j1 = sqrt(|w_j|), U_j2 = w_j/sqrt(|w_j|)). The bound in terms of metrological power W, Δq ≥ 1/√(N + 2n2W), is clean and reproduces known MZI results when d = 1. They also characterize which nonclassical states give a quantum advantage and show that local photon-number detection suffices for some inputs. That is real progress.\n\nThe soft spot is the proof of the universal bound. In Supplemental S2 the inverse of the QFIM is computed after dropping off-diagonal terms of N, and the orthogonality condition u⊥v is only shown to be optimal when N is ignored. The authors admit both limitations in the note at the end of S2. The explicit network does achieve the stated sensitivity, and that part is exact, but the claim that no linear network can do better is not established by the given derivation. The stress-test note puts this correctly: the bound is necessary for the main result but not sufficient to prove universality. If you want to cite the bound as a theorem, wait for a rigorous proof or a corrected version.\n\nAlso worth noting: the note added discloses overlap with Pezzè and Smerzi's concurrent work on local-measurement schemes, so the local-measurement part is not fully novel. The two-step function estimation is a nice add-on, though it inherits the same bound. The citation pattern and math are otherwise careful; W is imported from their earlier work but it is an independent quantity with known properties, so I see no circularity. The paper is honest about its limitations and the literature.\n\nThis deserves a serious referee. The referee should push on the universality claim and ask for either a rigorous treatment of N or a restatement of the bound as a property of the constructed network rather than a universal optimum.","headline":"Solid minimality theorem and explicit construction for CV distributed sensing, but the claimed 'universal' lower bound rests on an approximate inverse that the authors themselves flag.","tokens_in":19838,"tokens_out":2558,"would_cite":true,"duration_ms":23414,"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":"Two light inputs, one quantum and one classical, are necessary and sufficient for Heisenberg-limited estimation of arbitrary weighted sums of distributed phases, with a tight bound set by the nonclassical state's metrological power.","keywords":["distributed quantum metrology","continuous variables","Heisenberg limit","quantum Fisher information","metrological power","squeezed vacuum","cat states","linear optical networks"],"falsifier":"Numerically optimize the full, unapproximated quantum Fisher information matrix for a small network (for instance $d=2$) over passive unitaries $U$ and over the photon-number split between a squeezed-vacuum input and a coherent input at fixed total $N$, computing the exact inverse including the $N$ term; any configuration with $\\Delta q < 1/\\sqrt{N + 2 n_2 W}$ would disprove the universal bound. A complementary check is to test whether a single non-vacuum input with a weight vector containing both signs can beat the SQL, which the paper says is impossible.","tokens_in":18780,"feed_emoji":"🎯","tokens_out":10375,"duration_ms":94913,"temperature":0.7,"pith_summary":"Distributed quantum metrology asks how many resources a network needs to measure many unknown phases with the best precision quantum mechanics allows. This paper shows that two non-vacuum inputs — one nonclassical state (for example squeezed vacuum or a cat state) and one ordinary coherent laser — are both necessary and sufficient to estimate any balanced weighted sum of $d$ distributed phases at the Heisenberg limit, where the error scales as $1/N$ with $N$ the total photon number. The central result is a universal error bound $\\Delta q \\ge 1/\\sqrt{N + 2 n_2 W}$, in which $n_2$ is the coherent input's mean photon number and $W$ is the metrological power of the nonclassical state; the bound is tight for squeezed vacuum, cat, squeezed-thermal, and Fock states. A single non-vacuum input cannot do the job for weights of both signs, because it produces only positive-weight sensitivity. The paper also shows that the same network estimates arbitrary analytic functions of the phases, and that local photon-number detection saturates the bound for a useful class of states.","feed_headline":"Two inputs suffice for Heisenberg-limited distributed phase sensing","feed_subtitle":"One nonclassical beam and one ordinary laser can measure any weighted sum of many phases at the quantum limit.","key_machinery":"The machinery is the rank-plus-diagonal decomposition of the QFIM, $F = c_u uu^T + c_v vv^T + c_s(uv^T + vu^T) + N$, with $u_j = |U_{j1}|^2$ and $v_j = U_{j1} U_{j2}^*$ determined by the first two columns of the network unitary. The scalar $W$ is the metrological power of the nonclassical state, its maximal quantum advantage for quadrature sensing, given for pure states by $W = n_1 - |\\alpha_1|^2 + |\\xi_1 - \\alpha_1^2|$; this quantity controls the quantum-enhanced term $c_v = 8 n_2 W$. The argument proceeds by forcing the estimated weight vector $w$ to be parallel to $v$, which is the only zero-sum network vector, and then inverting the QFIM to obtain the bound. The same decomposition also identifies which input states help: only the first three moments of the annihilation operator enter $W$ and $c_s$, so a wide class of nonclassical states, not just squeezed light, can supply the quantum advantage.","core_discovery":"The paper's central discovery is the complete structure of the quantum Fisher information matrix (QFIM) for any passive linear network fed by one nonclassical state plus one coherent state: $F = c_u u u^T + c_v v v^T + c_s (u v^T + v u^T) + N$, where $u$ and $v$ are determined by the network's first two columns, $c_v = 8 n_2 W$ is set by the coherent state's energy times the nonclassical state's metrological power, and $N$ is the standard-quantum-limit diagonal term. Because the phase-reference condition forces every valid weight vector $w$ to have zero sum, and because $v$ is the only network vector with zero sum, the high-energy part of the estimation must come from $v$; the explicit choice $U_{j1}=\\sqrt{|w_j|}$, $U_{j2}=w_j/\\sqrt{|w_j|}$ makes $v=w$ and gives $\\Delta q \\ge 1/\\sqrt{N + 2 n_2 W}$. The bound is saturated whenever the cross term $c_s$ vanishes, which includes squeezed vacuum, cat, squeezed-thermal, and Fock states. This establishes necessity and sufficiency of two inputs: with only one non-vacuum input the QFIM reduces to $c_u u u^T + N$, and $u$ has strictly positive entries, so no Heisenberg-limited estimation of a linear combination containing both positive and negative weights is possible.","pith_inferences":["Read as a recipe, the paper implies that the hard resource to supply is metrological power $W$ rather than high photon number in the nonclassical state; a modest squeezed source backed by a strong coherent beam is the practical configuration.","Because $W$ is defined for quadrature sensing and is independent of the network, the same two-input minimality likely transfers to distributed displacement sensing, though the paper only analyzes phase shifts.","A natural next test is whether a compensating output unitary $V$ can make local photon-number detection saturate the bound for states with nonzero $c_s$; the paper's saturation proof assumes $c_s = 0$.","The QFIM decomposition suggests a resource-theoretic reading: among all nonclassical states at fixed energy, the one maximizing $W$ (squeezed vacuum) should also be optimal for arbitrary distributed weights, generalizing the two-mode result."],"forward_implications":["Two input ports are minimal: no single non-vacuum input can provide Heisenberg-limited estimation of an arbitrary linear combination with both positive and negative weights, while one nonclassical plus one coherent input can.","The universal bound is tight for every nonclassical state with vanishing cross term $c_s$, so squeezed vacuum, cat, squeezed-thermal, and Fock states saturate it with a suitable passive network.","In the weak-nonclassical regime ($1 < W \\ll n_2$), the network still multiplies the classical sensitivity by about $\\sqrt{1 + 2W}$, so modest squeezing can enhance sensing with strong lasers.","The same network estimates arbitrary analytic functions $f(\\theta)$ with variance $\\|\\nabla f\\|_1^2/(4N + 8 n_2 W)$, reaching Heisenberg scaling under the optimal two-stage resource split.","For input states satisfying the moment conditions $\\alpha_1 = \\beta_1 = 0$ plus a reality condition, photon-number-resolving detection at each node saturates the quantum bound."],"supporting_citations":[{"why":"Establishes the distributed quantum metrology QFIM framework and the SQL/HL classification that Eq. (3) extends to two non-vacuum inputs.","marker":"[9]"},{"why":"The first CV DQM protocol for arbitrary linear combinations at the Heisenberg limit; this paper proves that scheme's two-input structure is minimal.","marker":"[43]"},{"why":"Introduces metrological power $W$ and its properties, the quantity that enters the universal bound and the quantum-enhanced coefficient $c_v$.","marker":"[45]"},{"why":"Defines the weak-nonclassical amplification regime and the multiplicative sensitivity factor used for the second operating regime.","marker":"[46]"},{"why":"Supplies the multi-parameter quantum Cramér–Rao bound and QFIM definition underlying Eq. (1) and the inversion argument.","marker":"[53]"},{"why":"Provides the two-mode Mach–Zehnder result that coherent-plus-nonclassical inputs and photon-number detection are optimal, which the $d$-mode bound reduces to and generalizes.","marker":"[57]"},{"why":"Gives the two-step protocol for estimating arbitrary functions of parameters, which the paper adapts to obtain Heisenberg-limited function estimation.","marker":"[16]"},{"why":"Supplies the exact matrix-inversion identity used in the supplement to compute the inverse QFIM under the orthogonality condition.","marker":"[68]"}],"fun_headline_variants":["Two inputs hit Heisenberg limit for any weighted phase sum","Heisenberg-limited distributed metrology with two inputs","Two beams, one classical: quantum limit for any weight","Arbitrary weighted sums at quantum limit: two inputs suffice","Squeezed vacuum plus laser: Heisenberg-limited distributed sensing"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The universal part of the bound depends on the supplement's approximate inversion of the Fisher matrix, where the standard-quantum-limit diagonal term is expanded with off-diagonal pieces neglected and the network condition $u \\perp v$ is known to be optimal only when that term is absent; if some network exploits the neglected term to do better, the claimed universality could fail, although the explicit construction still reaches the Heisenberg limit.","fun_headline_variants_meta":{"raw":{"variants":["Two inputs hit Heisenberg limit for any weighted phase sum","Heisenberg-limited distributed metrology with two inputs","Two beams, one classical: quantum limit for any weight","Arbitrary weighted sums at quantum limit: two inputs suffice","Squeezed vacuum plus laser: Heisenberg-limited distributed sensing"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001314,"raw_usage":{"total_tokens":5431,"prompt_tokens":1097,"completion_tokens":4334,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":713,"completion_tokens_details":{"reasoning_tokens":4263}},"tokens_in":713,"tokens_out":4334,"duration_ms":30174,"temperature":1.0,"reasoning_tokens":4263,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T04:43:40.447572+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Numerically optimize the full, unapproximated quantum Fisher information matrix for a small network (for instance $d=2$) over passive unitaries $U$ and over the photon-number split between a squeezed-vacuum input and a coherent input at fixed total $N$, computing the exact inverse including the $N$ term; any configuration with $\\Delta q < 1/\\sqrt{N + 2 n_2 W}$ would disprove the universal bound. A complementary check is to test whether a single non-vacuum input with a weight vector containing both signs can beat the SQL, which the paper says is impossible.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the distributed quantum metrology QFIM framework and the SQL/HL classification that Eq. (3) extends to two non-vacuum inputs."},{"cited_title":"Malitesta, A","cited_arxiv_id":null,"evidence_quote":"The first CV DQM protocol for arbitrary linear combinations at the Heisenberg limit; this paper proves that scheme's two-input structure is minimal."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces metrological power $W$ and its properties, the quantity that enters the universal bound and the quantum-enhanced coefficient $c_v$."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the weak-nonclassical amplification regime and the multiplicative sensitivity factor used for the second operating regime."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the two-step protocol for estimating arbitrary functions of parameters, which the paper adapts to obtain Heisenberg-limited function estimation."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the exact matrix-inversion identity used in the supplement to compute the inverse QFIM under the orthogonality condition."}],"review_version":1}