{"id":"bd0c782e-9084-436d-ba50-e6bb4f3a918b","arxiv_id":"2608.10140","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A pure state's Williamson spectrum is weakly majorized by its ensemble average under any fermionic Gaussian protocol, unifying monotones and conversion bounds for non-Gaussianity.","lead":"This paper proves a majorization law for fermionic non-Gaussianity: under any Gaussian protocol with pure outcomes, the input Williamson spectrum is weakly majorized by the ensemble-averaged output spectrum. This gives a unified foundation for non-Gaussianity monotones, conversion bounds, and catalysis limits in fermionic quantum computation.","discovery_kind":"unification","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified: the majorization proof is internally consistent; the only residual risk is the unverified Lemma S3, which warrants an independent numerical check.","rationale":"The reader's verdict is ACCEPT with moderate confidence, and my independent read matches that. I walked through the proof of Lemma S3 line by line: the operator bound Q_J ≤ 2P++ is correct because Q_J has eigenvalues {2,0,0,-2} and 2P++ has {2,0,0,0}; the containment Image P++ ⊆ Image R follows from V_s|u> ∈ supp τ_s ⊆ Image R_s; the rank bound rank τ_s ≤ 2^{m-2} guarantees a perfect factor in the product normal form, so h_s exists; and the final step uses only Lemma S1 on each branch. The localization lemma correctly reduces the measured plane to at most two complex planes and justifies the mixed-state formulation. The only points that gave me pause were (i) the rank notation in SM S4 ('2m-2' instead of 2^{m-2}), which is a typo rather than a substantive error, and (ii) the discard-refinement sentence in SM S6, where the collinearity remark is imprecise; the argument can be read as applying to the kept-mode reductions and is in any case superseded by the unit-padding spectral identity. Neither affects the central claim. I therefore see no load-bearing defect, and I recommend an independent numerical check of Lemma S3 as the best use of verification effort.","tokens_in":21872,"tokens_out":39514,"duration_ms":381994,"concrete_test":"Numerically certify Eq. S22 for small N: for N=2,3,4 sample random pure states (including parity-coherent superpositions), for each l compute S_l(Γ_psi) by diagonalizing iΓ_psi, apply a one-mode occupation measurement to obtain p_s and Γ'_s, and verify S_l(Γ_psi) ≤ 1 + Σ_s p_s S_{l-1}(Γ'_s). Repeat over many random Gaussian unitaries before the measurement. Separately, for each sampled case construct the optimizer J of Lemma S1, the projector P++, the conditional blocks τ_s, and the branch reflections h_s, and check the operator inequality Q_J ≤ I + Σ_s |s><s|⊗h_s by explicit matrix diagonalization. Any violation of either inequality would falsify Lemma S3 and hence Theorem 1.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I checked the chain Theorem 1 ← Lemma S4 ← Lemma S3 in detail. The variational formula, the localization argument, and the Gaussian-code inequality are internally consistent. The rank-counting in SM S4 works: rank τ_s ≤ rank P++ = 2^{m-2} forces at least one |ν_j| = 1 in the product normal form, so the branch reflection h_s exists and satisfies Q_J ≤ I + Σ_s |s><s|⊗h_s as an operator inequality; taking expectations yields Eq. S22. The mixed-state formulation is needed only because the localized state σ is generally mixed, and Lemma S3 is proven for arbitrary σ. The induction over protocol trees and the pure-output discard refinement do not introduce circularity. I found no concrete counterexample for small N. The central claim therefore appears sound; the only honest concern is that Lemma S3 is intricate and not machine-verified, so independent verification is warranted.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper develops a majorization theory for fermionic non-Gaussianity based on the Williamson spectrum of the Majorana covariance matrix. The central result, Theorem 1, states that for any pure N-mode fermionic state ψ and any fermionic Gaussian protocol with pure-state outcomes {(p_s, φ_s)}, the unit-padded Williamson vector of ψ is weakly majorized by the ensemble-averaged padded spectrum. The authors derive a comprehensive set of consequences: Theorem 2 characterizes all f-Williamson deficits that are strong monotones (convexity of f), establishes additivity on tensor products with at least one parity-definite factor, faithfulness, and asymptotic continuity; Proposition 1 proves that Gaussian nullity is monotone under postselection; Corollaries 1 and 2 give deterministic and single-shot conversion bounds; Proposition 2 shows the single-shot bound is optimal among spectral monotones; Theorems 3 and 4 rule out catalysis with parity-definite catalysts and bound any catalytic gain by one mode; Corollary 3 and Proposition 3 bound asymptotic conversion rates; Theorem 5 establishes strong irreversibility of asymptotic Gaussian conversion.","tokens_in":22025,"tokens_out":48056,"duration_ms":420527,"significance":"If Theorem 1 holds, the paper supplies the long-missing fermionic analog of Nielsen's majorization theorem for entanglement, organizing the resource theory of fermionic non-Gaussianity around a single spectral order. It unifies the previously known quantifiers (fermionic antiflatness and occupation entropies) as members of one convex family, yields conversion bounds and a catalytic no-go theorem whose structure differs sharply from entanglement theory (spectra concatenate rather than multiply), and establishes asymptotic irreversibility at the pure-state level. The paper is careful and honest: the SM proofs are self-contained, the argument chain from Lemma S3 through Lemma S4 to Theorem 1 is internally consistent in my reading, the necessity constructions in Theorem 2 are explicit, and the main text repeatedly flags where its conditions are necessary but not sufficient (Corollaries 1 and 2). The majorization inequalities and conversion bounds are parameter-free and measurable from two-point Majorana correlators, making them falsifiable on present-day matchgate-capable devices.","major_comments":[],"minor_comments":[{"comment":"The sentence 'u(ψ) is weakly supermajorized by c u(ϕ)' states the order in the wrong direction relative to the displayed inequality Σ_{j≤ℓ} u_j(ψ) ≥ c Σ_{j≤ℓ} u_j(ϕ) for all ℓ; the correct phrasing is that u(ψ) weakly supermajorizes c u(ϕ) (equivalently, c u(ϕ) is weakly supermajorized by u(ψ)), and the subsequent symmetric-increasing-concave monotonicity argument is correct.","section":"Main text, Proposition 2 (proof)"},{"comment":"The statement restricts the partial sums to ℓ = 1, ..., N and writes the ensemble-averaged spectrum as an N-dimensional vector, while the proof sketch and the SM compare spectra after unit-padding to a common length that can exceed N when the protocol appends ancillas; please state the padding convention and the common length explicitly in the theorem statement.","section":"Main text, Theorem 1, Eqs. (3)-(4)"},{"comment":"This lemma is the crux of Theorem 1; I read the proof as internally consistent, including the rank-counting argument in the product normal form and the construction of the branch reflection h_s, but given the intricacy of the operator inequality Q_J ≤ I + Σ_s |s⟩⟨s| ⊗ h_s, I recommend adding a small-scale numerical verification for random local states σ and random two-plane structures J on two to three modes.","section":"SM S4, Lemma S3, Eq. (S20)"},{"comment":"The qualifier that Theorem 1 'also partially holds for arbitrary mixed states' is unexplained in the main text; please add a pointer stating that the mixed-state version (SM Lemma S6) requires protocols that retain the complete classical record, since unrecorded discards can otherwise drive spectral sums down for mixed states.","section":"Main text, after the proof sketch of Theorem 1"},{"comment":"The two displayed bounds are of different character: R ≤ N/ν_G(ϕ) is an infimum over the family f(t) = t^{2k} and is not attained by any admissible f, whereas R ≤ D_1(ψ)/N is attained by the affine generator f(r) = r; a sentence making this distinction explicit would prevent a misreading of 'optimal bound'.","section":"Main text, Eq. (19)"},{"comment":"The bound ∥Γ_ρ − Γ_σ∥_op ≤ ε with ε = ∥ρ − σ∥_1 is cited to Ref. [82]; since this bound is used to establish asymptotic continuity for every admissible Φ_f, it would be helpful to state the precise lemma and its constant in the SM rather than citing it only indirectly.","section":"SM S7D, asymptotic continuity proof"},{"comment":"The non-convertibility claim (GHZ_6^{⊗2} cannot be converted into GHZ_4^{⊗3} by Gaussian protocols) is imported from Ref. [53] rather than proven here; this is acceptable, but the support chain for the claim that the Williamson spectrum is not a complete invariant should be flagged as depending on the companion paper.","section":"Main text, Corollary 1, Eq. (8)"}],"recommendation":"minor_revision","confidential_remarks":"My independent reading of the proof chain (Theorem 1 ← Lemma S4 ← Lemma S3) found it internally consistent, and I identified no load-bearing errors; the recommendation of minor revision reflects local clarity improvements rather than technical doubts. If the editorial workflow permits, an independent check of Lemma S3 (SM S4) by a second referee would be valuable, since it is the single most intricate step and is not machine-verified. The paper overlaps with recent preprints by the same group (Refs. [41, 42, 53]) but appears to disclose and separate prior results appropriately."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear Colleague,\n\nThe short version: this is the real thing. The paper proves a Williamson majorization law for fermionic non-Gaussianity under Gaussian protocols, and it uses it to derive conversion bounds, a catalysis no-go theorem, a bound on catalytic gain, and an asymptotic irreversibility result for pure states. If the central proof holds, it puts the theory on a spectral footing the way Nielsen's theorem did for entanglement.\n\nWhat is genuinely new: prior results gave strong monotonicity for two or three specific measures. This paper gives one inequality for the whole padded Williamson spectrum and gets every convex deficit, including the full hierarchies of antiflatness and occupation entropies, as a corollary. The characterization of f-Williamson monotones (Theorem 2) is clean, and the necessity construction in the SM is a nice elementary family. The catalysis results are the most interesting: Williamson spectra concatenate under parity-definite catalysts, so no catalyst can remove a majorization obstruction—but a parity-coherent catalyst can, and the gain is sharply bounded (Theorem 4). The asymptotic irreversibility at the pure-state level is a strong conclusion, and the bounds are all in terms of two-point correlators.\n\nThe soft spot, which the reader flagged, is Lemma S3 in the Supplemental Material—the Gaussian-code inequality that carries the whole proof. The main text gives a sketch; the SM gives the full argument, and I went through the chain carefully. I could not find a gap or a circular step. The rank-counting argument works: rank τ_s ≤ 2^{m−2} forces a |ν_j| = 1 factor in the product normal form, and the branch reflection h_s does satisfy the operator inequality. Still, the lemma is intricate and not machine-verified. Given how much of the paper hangs on it, an independent numerical or formal check is warranted. The stress-test note confirms no counterexample for small N, which is reassuring but not a proof. The authors also note AI assistance in the proof development; the proof was independently verified by them, but this makes a referee's independent check even more appropriate.\n\nThe paper is for anyone working on fermionic non-Gaussianity, matchgate magic, or Gaussian resource theories. It is experimentally relevant, since all quantities come from covariance matrices. It deserves a serious referee. The claim is important, and the argument is careful enough to justify referee time, even though the SM is not easy reading.\n\nRecommendation: send it to peer review, and instruct the referee to focus on Lemma S3. That is the one load-bearing place that could break. If it survives, this becomes a standard reference.\n\nBest.","headline":"A genuinely new majorization law for fermionic non-Gaussianity, with far-reaching consequences; the only serious risk is the intricate Lemma S3 in the SM, which deserves independent checking.","tokens_in":22555,"tokens_out":3974,"would_cite":true,"duration_ms":33942,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["15A42","81P40","81P45","81P68"],"pacs":[],"model":"deepseek-v4-flash","headline":"Williamson spectra obey a majorization law under Gaussian protocols","keywords":["fermionic non-Gaussianity","Williamson spectrum","weak majorization","Gaussian protocols","strong monotones","Majorana covariance matrix","state conversion","fermionic magic"],"falsifier":"Check the one-mode measurement inequality numerically for three-mode pure states of the form $\\sqrt{p}\\,|0\\rangle|\\varphi_0\\rangle+\\sqrt{1-p}\\,|1\\rangle|\\varphi_1\\rangle$ by computing $S_2(\\Gamma_\\psi)$ and the branch spectra exactly; a single instance with $S_2(\\Gamma_\\psi)>1+p_0S_1(\\Gamma'_{\\psi_0})+p_1S_1(\\Gamma'_{\\psi_1})$ would refute Lemma S4 and with it Theorem 1. A still more localized test is to search random two-plane witnesses $J$ and random local states for a violation of the rank-two Gaussian-code inequality $Q_J\\le I+\\sum_s|s\\rangle\\langle s|\\otimes h_s$.","tokens_in":21682,"feed_emoji":"⚛️","tokens_out":10305,"duration_ms":97982,"temperature":0.7,"pith_summary":"Fermionic non-Gaussianity is the resource that elevates free-fermion (matchgate) computation to universal quantum computation, and this paper claims it too is governed by a single ordering principle. Under any fermionic Gaussian protocol whose outputs are pure states, the Williamson spectrum of the input pure state's Majorana covariance matrix is weakly majorized by the ensemble-averaged Williamson spectrum of the outputs: $S_\\ell(\\Gamma_\\psi) \\le \\sum_s p_s S_\\ell(\\Gamma_{\\phi_s})$ for every $\\ell=1,\\dots,N$. This is the fermionic mirror of Schmidt-spectrum majorization in pure-state entanglement theory. If the claim is right, every convex Williamson deficit—including fermionic antiflatness and occupation entropies—is a strong monotone, deterministic and probabilistic conversion bounds follow, parity-preserving catalysis is forbidden, and asymptotic pure-state interconversion is irreversible. Because all involved quantities are built from two-point Majorana correlators, the ordering is experimentally observable on present-day devices.","feed_headline":"Williamson spectra obey a majorization law under Gaussian protocols","feed_subtitle":"The same ordering that governs entanglement conversion now constrains fermionic magic, catalysis, and asymptotic rates.","key_machinery":"The carrying object is the Williamson spectrum of the $2N\\times 2N$ real antisymmetric Majorana covariance matrix $\\Gamma_\\rho$, whose canonical values $1\\ge r_1\\ge\\dots\\ge r_N\\ge0$ quantify how far a pure state is from fermionic Gaussian (all $r_j=1$). The proof turns these spectral partial sums into linear witnesses through the variational formula $S_\\ell(\\Gamma)=\\max_J \\tfrac12\\operatorname{tr}(J^T\\Gamma)$ over rank-$2\\ell$ partial complex structures $J$, so that $S_\\ell$ becomes the expectation of a sum of commuting quadratic reflections. A localization lemma confines an optimal witness around the measured occupation mode to at most three modes, and a Gaussian-code operator inequality $Q_J \\le I+\\sum_s |s\\rangle\\langle s|\\otimes h_s$ closes the rank-two case, giving the one-mode measurement inequality $S_\\ell(\\Gamma_\\rho)\\le 1+\\sum_s p_s S_{\\ell-1}(\\Gamma'_{\\rho_s})$. An induction over the protocol tree lifts this single-mode inequality to Theorem 1, while the convex-function characterization of weak majorization ($x\\prec_w y$ iff $\\sum_j f(x_j)\\le\\sum_j f(y_j)$ for every nondecreasing convex $f$) converts the spectral law into monotones, conversion bounds, and additivity statements.","core_discovery":"The paper's central claim is Theorem 1: for any pure $N$-mode state $\\psi$ and any ensemble $\\{(p_s,\\phi_s)\\}$ produced by a Gaussian protocol, the partial sums of the unit-padded Williamson spectrum satisfy $S_\\ell(\\Gamma_\\psi) \\le \\sum_s p_s S_\\ell(\\Gamma_{\\phi_s})$ for all $\\ell$, i.e. $\\mathrm{er}(\\Gamma_\\psi) \\prec_w \\sum_s p_s\\,\\mathrm{er}(\\Gamma_{\\phi_s})$. This is the fermionic analogue of Nielsen's majorization condition for deterministic LOCC transformation of bipartite pure states, with the Schmidt spectrum replaced by the Williamson values of the Majorana covariance matrix. From this ordering law the authors derive a complete characterization of spectral monotones (a nondecreasing generator $f$ makes $\\Phi_f$ a strong monotone if and only if $f$ is convex), necessary conditions and optimal single-shot bounds for state conversion under Gaussian protocols, monotonicity of the Gaussian nullity even under postselection, a no-go result for catalysis with parity-definite catalysts, a universal bound of one maximally non-Gaussian mode on any catalytic gain, and asymptotic irreversibility of pure-state interconversion in the parity-conserving sector.","pith_inferences":["If the majorization law holds, the most consequential open question is its sharpness: because the Williamson spectrum is not a complete invariant (zero-covariance GHZ states are a witness), a natural conjecture is that a full set of monotones based on higher-order Majorana correlators, possibly from the matchgate commutant, would turn the necessary condition into a sufficient one for deterministic","The one-mode measurement inequality has an operational reading that the paper leaves implicit: each recorded occupation outcome contributes at most one unit of Williamson spectral weight, so tracking the partial sums $S_\\ell$ through a measurement sequence provides a direct experimental witness of non-Gaussianity generation from two-point correlators alone.","A concrete testable extension is whether the same weak-majorization statement survives for bosonic Gaussian resources, where the Williamson values need not lie in $[0,1]$; if it does not, the fermionic compactness $r_j\\le1$ is the structurally essential ingredient, not the covariance-matrix formalism itself."],"forward_implications":["Every $f$-Williamson deficit $\\Phi_f$ with nondecreasing convex $f$ is a strong monotone for fermionic non-Gaussianity, which immediately covers the fermionic antiflatness hierarchy and the occupation-number entropies.","Deterministic pure-state conversion by a Gaussian protocol requires $\\mathrm{er}(\\Gamma_\\psi)\\prec_w \\mathrm{er}(\\Gamma_\\phi)$, and the best single-shot success probability obeys $p_{\\rm succ}\\le\\min_{\\ell:\\tilde{V}_\\ell(\\phi)>0}\\tilde{V}_\\ell(\\psi)/\\tilde{V}_\\ell(\\phi)$; this bound is optimal among all spectral monotones.","The Gaussian nullity $\\nu_G$, the number of Williamson values below one, is nonincreasing under postselection on any nonzero branch of a Gaussian protocol.","Under conserved fermion parity, no parity-definite catalyst can turn a forbidden conversion into an allowed one; a parity-coherent catalyst can, but only up to the universal bound $\\mathrm{er}(\\Gamma_\\psi)\\sqcup(0)\\prec_w \\mathrm{er}(\\Gamma_\\phi)\\sqcup(1)$, i.e. at most one maximally non-Gaussian mode.","Asymptotic Gaussian conversion between definite-parity pure states is irreversible: $R(\\omega\\to\\chi)R(\\chi\\to\\omega)\\le D_1(\\chi)/\\nu_G(\\chi)\\le1$, and the product can be driven arbitrarily close to zero."],"supporting_citations":[{"why":"Supplies the variational formula for Williamson partial sums, Cauchy interlacing, and the majorization facts used in the proof of Theorem 1 and Theorem 4.","marker":"[52]"},{"why":"Nielsen's LOCC condition is the entanglement statement whose fermionic analogue Theorem 1 and Corollary 1 provide.","marker":"[5]"},{"why":"Vidal's single-shot formula is the direct analogue of Corollary 2's success-probability bound.","marker":"[6]"},{"why":"Establishes that occupation postselection preserves Gaussianity, the key structural input in the Gaussian-code inequality.","marker":"[23]"},{"why":"Introduces the fermionic antiflatness hierarchy that Theorem 2 turns into strong monotones.","marker":"[41]"},{"why":"Introduces the occupation-number entropies and computable covariance-based non-Gaussianity measures covered by the new monotone theorem.","marker":"[42]"},{"why":"Provides prior strong-monotonicity results in specific cases that Theorem 2 unifies and extends.","marker":"[50]"},{"why":"Contains the GHZ-state conversion impossibility and the bridge degree that mark the limit of spectral invariants and motivate the sufficiency question.","marker":"[53]"}],"fun_headline_variants":["Williamson majorization: fermionic counterpart to entanglement","Fermionic non-Gaussianity obeys a majorization law","Majorization rules fermionic non-Gaussianity","Williamson spectra obey Gaussian majorization","New spectral majorization for fermionic quantum states"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The entire theorem rests on the one-mode measurement inequality proved in the supplement: when one occupation number of an arbitrary local state is measured, the sum of the $\\ell$ largest Williamson values of the input is at most $1$ plus the averaged sum of the $\\ell-1$ largest Williamson values of the branches; if that inequality fails for some non-Gaussian state, Theorem 1 and everything built on it collapses.","fun_headline_variants_meta":{"raw":{"variants":["Williamson majorization: fermionic counterpart to entanglement","Fermionic non-Gaussianity obeys a majorization law","Majorization rules fermionic non-Gaussianity","Williamson spectra obey Gaussian majorization","New spectral majorization for fermionic quantum states"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000379,"raw_usage":{"total_tokens":2037,"prompt_tokens":993,"completion_tokens":1044,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":609,"completion_tokens_details":{"reasoning_tokens":970}},"tokens_in":609,"tokens_out":1044,"duration_ms":10441,"temperature":1.0,"reasoning_tokens":970,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T04:12:36.665460+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Check the one-mode measurement inequality numerically for three-mode pure states of the form $\\sqrt{p}\\,|0\\rangle|\\varphi_0\\rangle+\\sqrt{1-p}\\,|1\\rangle|\\varphi_1\\rangle$ by computing $S_2(\\Gamma_\\psi)$ and the branch spectra exactly; a single instance with $S_2(\\Gamma_\\psi)>1+p_0S_1(\\Gamma'_{\\psi_0})+p_1S_1(\\Gamma'_{\\psi_1})$ would refute Lemma S4 and with it Theorem 1. A still more localized test is to search random two-plane witnesses $J$ and random local states for a violation of the rank-two Gaussian-code inequality $Q_J\\le I+\\sum_s|s\\rangle\\langle s|\\otimes h_s$.","supporting_citations":[{"cited_title":"Bravyi, Lagrangian representation for fermionic lin- ear optics, Quantum Inf","cited_arxiv_id":null,"evidence_quote":"Establishes that occupation postselection preserves Gaussianity, the key structural input in the Gaussian-code inequality."},{"cited_title":"Computable fermionic non-Gaussianity from the covariance matrix","cited_arxiv_id":"2607.02242","evidence_quote":"Introduces the occupation-number entropies and computable covariance-based non-Gaussianity measures covered by the new monotone theorem."},{"cited_title":"Fermionic entropy: an efficiently measurable strong monotone for non-Gaussianity","cited_arxiv_id":"2607.29670","evidence_quote":"Provides prior strong-monotonicity results in specific cases that Theorem 2 unifies and extends."}],"review_version":1}