{"id":"596b2575-8f02-4aea-9951-68d78bc34f9a","arxiv_id":"2505.24212","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper constructs optimal-depth, optimal-gate-count quantum circuits that sample Haar-random active and passive fermionic linear optics unitaries directly from angle distributions, plus an optimal Clifford FLO sampler.","lead":"This paper gives recipes for building random fermionic linear optics (matchgate) circuits with the fewest possible gates and shortest possible depth, by specifying exactly how to sample each rotation angle. The result matters because random FLO circuits are a basic building block in quantum benchmarking, fermionic shadow tomography, and quantum advantage experiments.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The proof that the turnover compression produces the exponent pattern f_n in Eq. (6) is the load-bearing step; it is presented as a diagrammatic sketch, and a hidden swap-ordering or boundary error would invalidate Theorem 1.","rationale":"I read the paper in good faith. The main advertised claim is a sampling algorithm for Haar random active and passive FLO with optimal asymptotic depth and gate count. The individual ingredients are credible: Hurwitz's decomposition is standard, the Lie-algebra isomorphism is explicit, and Lemma 4's Jacobian computation is internally consistent. The n=4 numerical frame potentials provide genuine evidence for low moments. The load-bearing point is the global measure transformation under the turnover compression. The manuscript gives the final exponents in Eq. (A48) and asserts the swap order works, but does not supply a rigorous induction or a machine-checkable verification. Since the entire sampling distribution is defined by that exponent pattern, an error there would not change the circuit architecture but would silently destroy Haar randomness. This is exactly the reader's weakest-assumption identification, so I agree with the CONDITIONAL verdict: the result is plausible and well-supported at low n, but the compression proof needs to be made formal or verified symbolically before the theorem can be accepted as fully proven. No ad hominem or theatrical language is intended; this is a technical gap in presentation and verification, not an accusation of error.","tokens_in":35757,"tokens_out":25065,"duration_ms":290404,"concrete_test":"Use a computer-algebra system to symbolically implement the SI A.4 compression for n=5: start from the triangular circuit of Lemma 3 with sine powers k-j, apply the turnover map of Eq. (A39) in the exact order specified before Eq. (A48), and track the Jacobian and the sine-power of every gate. Verify factor-by-factor that the final density is exactly Eq. (A3)-(A6) with f_5 from Eq. (6); any mismatch in any gate exponent would show that Theorem 1's measure is incorrect.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that sampling gate parameters from Eqs. (4)-(6) yields Haar random active FLO. The proof route is: Hurwitz's triangular decomposition (Lemma 3) plus a sequence of turnover transformations (Lemma 4) to reach the brick-wall architecture. Lemma 4 is a correct local three-gate identity, but it only preserves the measure when the middle sine exponent equals r+r'+1. The global compression in SI A.4 asserts that the ordering of swaps captured by Sigma_k in Eq. (A47) guarantees this condition at every step, and then states the final anti-diagonal vectors in Eq. (A48) without a formal induction or a check of boundary diagonals d_1 and d_{2n-1}. If the order of the Sigma_k operations or the boundary handling is wrong, the sampled distribution would differ from Haar even though the circuit architecture and the n=4 frame-potential numerics might look plausible. The passive-FLO result (Theorem 2) is explicitly delegated as 'completely analogous', so it inherits this fragility. This is the weakest load-bearing assumption in the paper.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents circuit architectures for sampling Haar-random active and passive fermionic linear optics (FLO) unitaries directly from gate parameters, avoiding the usual O(n^3) classical matrix-sampling and compilation step. The central results are Theorem 1 and Theorem 2, which give brick-wall decompositions with n(2n-1) and n^2 parameters, respectively, and explicit product-of-sine probability densities (Eqs. (4)-(6) and (11)-(13)). The proofs start from Hurwitz's SO(2n)/U(n) decompositions, map them to matchgate circuits via the Majorana representation, and then use turnover gate identities to compress triangular circuits into brick-wall form. The paper also gives an algorithm for sampling Clifford FLO circuits with O(n^2) average two-qubit gate count, proves that Clifford FLO do not form a 4-design over FLO, and reports numerical frame-potential checks for n=4 that support the low-order moment structure.","tokens_in":35970,"tokens_out":8385,"duration_ms":87072,"significance":"If the central claims are valid, this is a useful advance for quantum information practice: it removes the O(n^3) classical compilation step of previous direct sampling methods and provides circuits with simultaneous Theta(n) depth and Theta(n^2) gate count, which is relevant for randomized benchmarking, fermionic classical shadows, and fermionic random circuit sampling. The explicit formulas are concrete and falsifiable, the local lemmas (Lemmas 1-4 and 6) are derived in detail, and the numerical frame-potential comparisons give nontrivial evidence for the low moments. The main gap is the global compression proof, which is presented as a diagrammatic sketch rather than a formal induction; this gap also affects the passive-FLO theorem, whose proof is delegated as 'completely analogous'.","major_comments":[{"comment":"The global compression step that converts the triangular circuit into the brick-wall architecture is the load-bearing step for Theorem 1, but it is asserted rather than proven. The text states that the ordering captured by Sigma_k and the initial assignment 'ensures' the turnover condition, and then lists the final anti-diagonal vectors in Eq. (A48) without a formal induction. Please provide a complete proof, ideally a rigorous induction, that (i) the sequence of turnovers is executable in the stated order, (ii) the middle-gate exponent condition r+r'+1 is satisfied at every turnover, and (iii) the boundary anti-diagonals d_1 and d_{2n-1}, including the k'=n-1 case, yield exactly Eq. (A48) and hence Eq. (6). A hidden swap-ordering or boundary error here would change the sampled distribution and invalidate the main theorem.","section":"SI A.4, Eqs. (A47)-(A49)"},{"comment":"The passive-FLO proof is delegated with the statements 'this optimization is completely identical' and 'completely analogous' to the active case. This is not sufficient: the passive case has n anti-diagonals rather than 2n-1, a different turnover lemma (Lemma 6), and additional phase parameters. A boundary or ordering error in the passive compression would break Theorem 2 just as in the active case. The SI should present the passive analogue of the Sigma_k argument explicitly, including the modified exponent vector definitions and boundary handling, rather than relying on analogy.","section":"SI B.4"},{"comment":"The sentence 'This is optimal as one cannot further compress the n(2n-1) gates' is a gate-count lower bound, not a depth lower bound. To justify the advertised 'optimal down-to-the-constant-factor Theta(n) depth', the paper needs an explicit depth lower bound for any nearest-neighbor matchgate circuit sampling the full FLO group, together with a matching analysis of the depth of the 2n layers in Eq. (3), including the intra-layer scheduling of non-commuting XX gates on adjacent edges.","section":"II A, depth claim"}],"minor_comments":[{"comment":"There are typos in the intermediate factors of the Lemma 6 proof: 'sin(t1)' should be 'sin(t3)', and '(sin theta3 sin theta3)' should be '(sin theta2 sin theta3)'. The final formula is correct, but these errors make verification harder.","section":"SI B.4, Eq. (B26)"},{"comment":"The stated range 'for 1<k<n' conflicts with the layers L_1 and L_{2n} appearing in Eq. (2). The range should be stated consistently, e.g., 1<=k<=n, with the angle domains specified accordingly.","section":"Eq. (3)"},{"comment":"The text states that the sampled frame potentials match the exact commutant dimensions 'despite the statistical uncertainty' but does not quantify that uncertainty. Adding error bars or a table of the numerical values used for n=4 and t=1,...,5 would make the verification reproducible.","section":"Fig. 8"},{"comment":"The notation [k] is used without a definition; the proof assumes l is chosen from {1,...,k}. Please define the interval notation explicitly in the main text and use it consistently in the algorithm and its proof.","section":"Algorithm 1 and SI F"}],"recommendation":"major_revision","confidential_remarks":"The core construction appears likely to be correct, and the missing formal induction is a completeness gap rather than a demonstrated error. However, because the central claim depends on the global compression step and the passive case is delegated by analogy, I cannot recommend acceptance without the authors supplying the full proof. I saw no issues with novelty or attribution; the manuscript is within the scope of the journal."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The useful part of this paper is real: it gives explicit angle distributions for brick-wall FLO circuits that sample the Haar measure, with n(2n−1) gates and depth 3n, plus a Clifford FLO sampler with n²/2 two-qubit gates and a proof that Clifford FLO is not a 4-design. The main technical development is Lemma 4, which tracks how the measure transforms under gate turnovers, and the anti-diagonal bookkeeping in the SI. This is the first construction I know of that avoids O(n³) classical compilation while keeping linear depth and optimal gate count.\n\nTwo things bother me. The “optimal depth” claim is not actually proven. They argue 3n is optimal because n(2n−1) gates require 2n layers, but a depth lower bound is a different statement; the claim “one cannot further compress” is an assertion. Given the literature’s 6n−7 triangular depth, the improvement is real, but the optimality should be softened or proven.\n\nThe bigger issue is the stress-test concern: the global compression proof in SI A.4 is a sketch. Lemma 4 is proven locally, but the claim that the Sigma_k ordering produces the f_n exponents in Eq. (6) is just stated with a diagram and a formula for the resulting anti-diagonals. There is no formal induction, and the boundary diagonals d_1 and d_{2n−1} are not explicitly checked. If the swap ordering has a hidden bug, Theorem 1’s distribution would be wrong. The n=4 frame-potential numerics are reassuring but not a substitute, and Theorem 2 is explicitly delegated as “completely analogous”, so it inherits the same fragility. This is a fixable presentation issue, but it is load-bearing.\n\nWho this is for: people doing randomized benchmarking, fermionic shadows, or anything needing Haar random FLO on hardware. The construction is likely correct; I’d use it after the proof is tightened. It deserves a serious referee, though my recommendation would be major revision rather than acceptance as-is.","headline":"A genuinely useful construction for Haar random FLO with explicit angle distributions and optimal gate count, but the load-bearing compression proof is a sketch and the depth-optimality claim is unproven.","tokens_in":36535,"tokens_out":1848,"would_cite":true,"duration_ms":21205,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P68","60B20"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that Haar-random fermionic linear optics can be sampled directly on a fixed brick-wall matchgate circuit with linear depth, quadratic gate count, and quadratic classical overhead.","keywords":["fermionic linear optics","Haar measure","matchgate circuits","random quantum circuits","brick-wall architecture","Clifford group","unitary designs"],"falsifier":"Take $n=4$, sample $10^6$ circuits using the distributions in Theorems 1 and 2, compute the empirical frame potentials $\\operatorname{Tr}[\\tau_{\\Lambda}^{(t)}]$ for $t=2$ and $t=3$, and compare them with the exact Haar values, namely the $t$-fold commutant dimensions computable by the SI G method. Since the second moment already distinguishes the Haar measure from any other independent-parameter product measure, a deviation beyond sampling error would show the exponent pattern is wrong; agreement for all $t$ up to at least 3 would corroborate the claim.","tokens_in":35567,"feed_emoji":"🎲","tokens_out":12698,"duration_ms":118605,"temperature":0.7,"pith_summary":"This paper aims to make Haar-random sampling of fermionic linear optics (FLO) circuits as simple as running a fixed circuit and drawing numbers. The authors prove that a nearest-neighbor brick-wall matchgate architecture on $n$ qubits, with $2n$ layers and $n(2n-1)$ gates, produces Haar-random active (non-particle-preserving) FLO when every gate angle is sampled independently from an explicitly given product-of-sines distribution; a parallel construction with $n$ layers and $n^2$ parameters produces Haar-random passive (particle-preserving) FLO. The circuit depth is linear in $n$, the gate count matches the dimension of the relevant Lie algebra, and the classical overhead is only the cost of sampling $\\Theta(n^2)$ parameters, avoiding the $\\mathcal{O}(n^3)$ matrix compilation used before. The paper also gives a direct algorithm for uniformly sampling Clifford FLO circuits with an optimal average of $n^2/2$ two-qubit gates, and shows Clifford FLO is not a 4-design over FLO. These results matter because FLO sampling underpins randomized benchmarking of continuous gate sets, fermionic classical shadows, and fermion-sampling experiments.","feed_headline":"Optimal depth and gate count for Haar-random fermionic circuits","feed_subtitle":"Fixed brick-wall matchgate circuits with sine-power parameter sampling replace cubic-cost classical compilation.","key_machinery":"The load-bearing mechanism is the exact pullback of the Haar measure through a sequence of gate rearrangements. The starting point is a classical angle-coordinate factorization of $\\mathrm{SO}(2n)$ and $\\mathrm{U}(n)$, whose Haar measures are products of powers of sines. A Lie-algebra isomorphism sends the infinitesimal plane rotations $L_{jk}$ to fermionic bilinears $c_jc_k/2$, which in the qubit representation become the matchgate generators $iZ_q$ and $iX_qX_{q+1}$, so the factorization becomes a triangular matchgate circuit. The paper then uses the turnover identity $U_{R_j}(\\alpha)U_{R_{j+1}}(\\beta)U_{R_j}(\\gamma) = U_{R_{j+1}}(a)U_{R_j}(b)U_{R_{j+1}}(c)$ to reorder gates into a brick wall, and proves (Lemma 4) that under this reordering the unnormalized density $\\sin^r(\\alpha)\\sin^{r+r'+1}(\\beta)\\sin^{r'}(\\gamma)$ maps to $\\sin^{r'}(a)\\sin^{r+r'+1}(b)\\sin^r(c)$: the middle gate's exponent is inherited, while the two outer exponents are exchanged. Bookkeeping with antidiagonals and swap operators yields the final exponent functions $f_n$ and $g_n$, so that the product measure is literally the Haar measure, not an approximation.","core_discovery":"The central claim is stated as Theorem 1: every FLO circuit can be written, up to a global sign, as $U = L_{2n} \\cdots L_1$, where odd layers are products of nearest-neighbor $e^{i\\alpha_{jk} X_j X_{j+1}}$ rotations and even layers are products of single-qubit $e^{i\\beta_{jk} Z_j}$ rotations. With the angles in their stated ranges, the normalized Haar measure on the adjoint representation is exactly the product measure $d\\mu(U) = N \\prod_{j,k} \\sin(\\alpha_{jk})^{f_n(2j,2k-1)} \\sin(\\beta_{jk})^{f_n(2j-1,2k)}\\, d\\alpha\\, d\\beta$, with integer exponents $f_n(u,v)$ given by the min-formula in Eq. (6). Sampling each angle independently from its sine-power distribution therefore yields Haar-random active FLO on a fixed, depth-optimal circuit, with no classical compilation. Theorem 2 is the analogous statement for passive FLO, with $n$ layers and exponent functions $g_n$; Theorem 3 says the Clifford FLO subgroup is not a 4-design over FLO, while earlier results had established it as a 3-design. Algorithm 1 samples Clifford FLO uniformly by choosing angles from $\\{0,\\pi/2,\\pi,3\\pi/2\\}$ in a triangular matchgate layout, achieving the optimal average two-qubit gate count of $n^2/2$.","pith_inferences":["One can test the same turnover calculus on other matchgate connectivities, such as periodic boundary conditions, two-dimensional grids, or long-range couplings, by re-deriving the exponent functions $f_n$ and $g_n$ and certifying the result with a frame-potential test; the paper does not do this, but nothing in the method is restricted to open chains.","Because most angles concentrate near $\\pi/2$ as $n$ grows, truncating the peaked sine distributions or rounding angles to Clifford values should produce approximate, rather than exact, $t$-designs for small $t$; the paper gives the exact distributions but does not quantify the error of such truncations.","The exact commutant-dimension computation used to verify the frame potentials could serve as a general certificate that any proposed random-circuit ensemble is Haar-like; the paper presents it as a verification tool rather than as a standalone sampling method."],"forward_implications":["Randomized benchmarking of continuous matchgate gate sets can run on a fixed hardware-native brick-wall circuit with linear depth, making the protocol practical on near-term devices.","Fermionic classical shadow protocols can sample their random unitaries directly from gate parameters, removing the cubic compilation bottleneck that limited previous implementations.","Since most sampled angles approach $\\pi/2$ for large $n$, Haar-random FLO circuits are mostly Clifford-like, which can be exploited in implementations that prefer Clifford gates.","Algorithm 1 provides uniform Clifford FLO sampling with the provably optimal average number $n^2/2$ of two-qubit gates, giving a cheaper way to form a 3-design over FLO for applications where the third moment suffices.","The negative result that Clifford FLO is not a 4-design means fourth-moment quantities, such as certain shadow variances, require true Haar FLO or a different design."],"supporting_citations":[{"why":"Supplies the classical angle-coordinate factorization of SO(d) and U(d) together with the invariant product-of-sines Haar measures that the constructions start from.","marker":"[37]"},{"why":"Provides the gate-turnover identity used to reorder matchgates while preserving the product of unitaries.","marker":"[27]"},{"why":"Gives the algebraic quantum-circuit compression viewpoint used to justify compressing sampled triangular Clifford circuits and to motivate the turnover calculus.","marker":"[41]"},{"why":"Establishes the fermionic-bilinear to qubit-Pauli correspondence that turns plane rotations into nearest-neighbor matchgate rotations.","marker":"[17]"},{"why":"Supplies the dynamical Lie algebra of matchgate circuits, isomorphic to so(2n), through which the classical factorization is translated into quantum circuits.","marker":"[23]"},{"why":"Provides a previous triangular matchgate-benchmarking sampling construction; its suboptimal depth is the baseline the brick-wall circuits improve on.","marker":"[21]"},{"why":"Provides the previous triangular matchgate classical-shadows sampling framework that the paper's construction supersedes and to which gate counts are compared.","marker":"[35]"},{"why":"Describes the standard Gaussian and QR method for sampling Haar orthogonal matrices with cubic classical cost, the baseline the new parameter-sampling method avoids.","marker":"[36]"},{"why":"Uses compiled Haar FLO for fermion sampling and is one of the previous approaches whose compilation cost the new method removes.","marker":"[29]"}],"fun_headline_variants":["Optimal-depth Haar-random matchgate circuits without classical compilation","Haar-random fermionic circuits with linear depth and quadratic gate count","Matchgate Haar sampling without cubic-cost classical compilation","Optimal circuits for Haar-random fermionic linear optics sampling"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The result rests on the assumption that the antidiagonal swap bookkeeping in Eq. (A48) exactly describes how the sine-power exponents change at every gate turnover, including the boundary cases, and that the passive-FLO proof really is completely analogous to the active one; if the tracking is wrong at any gate, the sampled circuits will not be Haar random.","fun_headline_variants_meta":{"raw":{"variants":["Optimal-depth Haar-random matchgate circuits without classical compilation","Haar-random fermionic circuits with linear depth and quadratic gate count","Matchgate Haar sampling without cubic-cost classical compilation","Optimal circuits for Haar-random fermionic linear optics sampling"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000896,"raw_usage":{"total_tokens":3903,"prompt_tokens":1028,"completion_tokens":2875,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":644,"completion_tokens_details":{"reasoning_tokens":2808}},"tokens_in":644,"tokens_out":2875,"duration_ms":20920,"temperature":1.0,"reasoning_tokens":2808,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T12:31:57.988519+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take $n=4$, sample $10^6$ circuits using the distributions in Theorems 1 and 2, compute the empirical frame potentials $\\operatorname{Tr}[\\tau_{\\Lambda}^{(t)}]$ for $t=2$ and $t=3$, and compare them with the exact Haar values, namely the $t$-fold commutant dimensions computable by the SI G method. Since the second moment already distinguishes the Haar measure from any other independent-parameter product measure, a deviation beyond sampling error would show the exponent pattern is wrong; agreement for all $t$ up to at least 3 would corroborate the claim.","supporting_citations":[{"cited_title":"Oszmaniec and Z","cited_arxiv_id":null,"evidence_quote":"Supplies the classical angle-coordinate factorization of SO(d) and U(d) together with the invariant product-of-sines Haar measures that the constructions start from."},{"cited_title":"Diaconis and P","cited_arxiv_id":null,"evidence_quote":"Gives the algebraic quantum-circuit compression viewpoint used to justify compressing sampled triangular Clifford circuits and to motivate the turnover calculus."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the dynamical Lie algebra of matchgate circuits, isomorphic to so(2n), through which the classical factorization is translated into quantum circuits."}],"review_version":1}