{"id":"f2892970-0c33-433a-a1d2-e0f5441d59a0","arxiv_id":"2607.29551","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"iSWAP gates maximise the second-moment spectral gap of random two-local quantum circuits on every connected graph with at least three qubits, for Hermitian gate ensembles.","lead":"The paper proves that a particular two-qubit operation called iSWAP is the fastest way to make random quantum circuits look random at the level of two-copy averages, on any connected network of three or more qubits. The result settles a conjecture about which gates give the best convergence rate for random quantum circuits.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The proof's final step relies on an imported and unverified localisation result (Prop 7.1) that equates the full iSWAP spectral gap with the compressed gap; without it, the argument only bounds competitors by γ0, not by the true iSWAP gap.","rationale":"The reader's weakest_assumption identifies Proposition 7.1, the imported localization result, as the bridge between the compressed comparison and the full iSWAP gap. My stress test reaches the same point: the proof genuinely needs ∆(T^iSWAP) = γ0, not merely γ0 as an upper bound. Lemma 2.2 supplies the wrong inequality for this purpose—it shows the full gap is at most γ0, whereas the theorem requires equality. The manuscript does not reproduce or verify the cited lemma, and it is the only external ingredient on which the final inequality depends. All the paper's own contributions—the affine envelope, the boundary reduction, the Z-matrix/Perron–Frobenius certificate, and the cone invariance—are internally coherent as far as I can see, and the concrete numerical test proposed would settle the external assumption cleanly. Since the existing CONDITIONAL verdict already flags this dependency, my read does not change the verdict. The scope issues noted by the reader (uniform edge selection and the Hermitian qualification) are secondary exposition concerns, not additional load-bearing risks.","tokens_in":17249,"tokens_out":22907,"duration_ms":234471,"concrete_test":"For n=3, on both the path graph and the triangle graph, construct the full iSWAP second-moment operator T using (2.7) with the ironed iSWAP gadget on every edge (dimension 16^3 = 4096). Numerically diagonalize T and compute its spectral gap ∆_full = 1−max{λ⋆, |λ_min|}. Separately build the 6×6 transient block M0 at α0 = 10/9 and compute γ0 = min Spec(M0). Check that ∆_full equals γ0 to numerical precision and that the eigenvector attaining λ⋆ is contained in W. Any discrepancy would falsify Proposition 7.1 in the regime used by Theorem 1.1.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central comparison is completed in two stages. First, Theorem 6.4 and Lemma 2.2 show that for any admissible Hermitian ensemble T^E, λ⋆(T^E) ≥ 1−γ0, hence ∆(T^E) ≤ 1−λ⋆(T^E) ≤ γ0. Second, Proposition 7.1 is invoked to identify the full iSWAP moment operator's spectral gap with 1−λ⋆(T^iSWAP|_W) = γ0. This second step is not proved here; it is imported from [17, Lemma 4.13 and Corollary 4.14]. The direction of Lemma 2.2 only gives λ⋆(full) ≥ λ⋆(compressed), so without localization the full iSWAP gap could be strictly smaller than γ0—either because a nontrivial eigenvalue outside W exceeds the compressed one, or because |λ_min| dominates. In that case the proven inequality ∆(T^E) ≤ γ0 would not imply ∆(T^E) ≤ ∆(T^iSWAP), and Theorem 1.1 would be unsupported exactly where it needs to be sharp. The condition n ≥ 3 is supplied by the import and is not independently verified in the manuscript, even though Remark 7.2 stresses that n = 2 is exceptional. Since the entire theorem is conditional on this external localization result, it is the most load-bearing assumption in the paper.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves Theorem 1.1: on any connected graph with n≥3, for any edge-dependent two-local unitary ensemble whose second-moment operator is Hermitian, the spectral gap is at most that of the iSWAP circuit. The proof proceeds by reducing every ironed gadget to a two-parameter family (α,β), bounding the allowed parameters by an affine envelope, replacing the Hamiltonian by a boundary family, applying a ground-state similarity transform to obtain a Z-matrix block, and then proving a cone-invariance result for asymmetric four-point inequalities. The Perron–Frobenius eigenvector of the iSWAP block is placed in this cone, yielding a componentwise comparison certificate. Finally, a local Haar compression step and an imported localization result from Kong–Li–Liu identify the compressed iSWAP eigenvalue with the full spectral gap, completing the proof of the conjecture.","tokens_in":17635,"tokens_out":5475,"duration_ms":55660,"significance":"If correct, the paper resolves a conjecture of Kong, Li, and Liu and provides a strong, parameter-free universality result: iSWAP is the optimal Hermitian two-local gate ensemble for second-moment convergence on every connected graph with at least three vertices. The main technical novelty—the invariant polyhedral cone of four-point inequalities and the Perron–Frobenius certificate—is original and likely to be useful for other spectral comparison problems. The proof is detailed and structurally coherent, and the comparison is genuinely parameter-free. The central caveat is that the final identification of the full iSWAP spectral gap with the compressed eigenvalue is imported from [17] rather than proved here, making the main theorem conditional on that external result.","major_comments":[{"comment":"The proof of Theorem 1.1 relies on the imported localization result that the largest non-trivial eigenvalue of the full iSWAP moment operator is attained in the compressed space W and that the full spectral gap equals γ0. Without this statement, the argument only yields ∆(T^E) ≤ γ0, and since ∆(T^iSWAP) is not independently established, the claimed inequality (1.1) would not follow. The condition n ≥ 2/(1+λ_min) and the value λ_min(T^iSWAP_2) = -1/3 are asserted but not verified in this manuscript. Please either provide a self-contained proof of Proposition 7.1 (or a precise quotation of [17, Lemma 4.13 and Corollary 4.14] with all hypotheses checked), or explicitly state that Theorem 1.1 is conditional on that result. As it stands, the main theorem's sharpness rests entirely on an external, unproved-in-text statement.","section":"Section 7, Proposition 7.1"}],"minor_comments":[{"comment":"The displayed expansion contains an arithmetic error: the coefficient of f(A_j) should be -7, not -10, in both occurrences. The final identity is correct, but the intermediate expression is wrong.","section":"Lemma 5.2, proof of identity (5.10)"},{"comment":"Reference [19] (the authors' companion paper) is listed in the bibliography but never cited in the text. It should be cited where relevant or removed from the reference list.","section":"References"},{"comment":"The computation of λ_min(T^iSWAP_2) = -1/3 is stated without proof or reference. Please provide a citation to the known result or a short derivation.","section":"Section 7, Proposition 7.1"},{"comment":"The paper would be easier to evaluate if the imported localization result were summarized more explicitly, including the definition of the full moment operator and the sense in which the compressed eigenvalue determines the gap. Currently the reader must consult [17] to verify the key step.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The core of the paper—the cone-invariance argument and the compressed spectral comparison—appears sound and is a genuine advance. The main risk is the dependence on an external localization result (Proposition 7.1) that is not proved here. If the editors accept reliance on a preprint for a load-bearing step, the paper is close to acceptable; if self-containedness is required, this is a blocker. The uncited self-reference [19] should be fixed. I recommend major revision rather than rejection, because the issue is addressable by either importing the result more rigorously or proving it in an appendix."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my take on Liang-Zhu. The main result is exactly what it claims: for any connected graph with n≥3 and any Hermitian two-local ensemble with uniform edge selection, the iSWAP gate maximizes the second-moment spectral gap. This resolves Kong-Li-Liu's conjecture and generalizes their complete-graph result. The paper does something real: it builds a four-point cone, shows it's invariant under the transpose semigroup, and uses the Perron–Frobenius eigenvector as a componentwise certificate. That's a new technique, and the proof is carefully written. I checked the linear algebra in Section 6 and the cone identities in Appendix B; they hold up. The reduction to the boundary family is clean, and the Collatz–Wielandt step is legitimate.\n\nThe soft spot is the bridge at the very end. Proposition 7.1—that for n≥3 the iSWAP full spectral gap equals the compressed gap γ0—is imported from [17] without proof. The paper states it as a specialization and gives the condition, but the actual verification is in the earlier paper. That's not fatal if [17] is correct, and the authors are transparent about it. But it means the headline theorem is only as solid as that external result. A referee should check it separately, and the authors should either reproduce the argument or point to exactly where in [17] it is proved. Also, the abstract says 'all two-local unitary circuit ensembles' but the theorem requires uniform edge selection; that's an overstatement in the abstract, not in the math.\n\nThere's a smaller issue: [19] is listed in the references but never cited in the body. Either cite it or remove it. The reader flagged this; I agree.\n\nThe n≥3 restriction is genuinely structural (see Remark 7.2), so that's not a flaw. And the edge-dependent ensemble statement is a real strengthening, not a restatement.\n\nOverall: this is a serious paper that resolves a conjecture with a new method. I'd send it to a referee, and the referee should spend most time on Proposition 7.1 and on the edge-dependent compression to make sure no eigenvalue outside W breaks the gap. If those confirm, it's a strong result. I'd take it to reading group.","headline":"Resolves a real conjecture with a genuinely new invariant-cone argument; the core comparison is solid, but the final step leans on an imported localization result that a referee must check.","tokens_in":18083,"tokens_out":2453,"would_cite":true,"duration_ms":22534,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P45","05C50","15A18","60J10"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that the iSWAP gate maximises the second-moment spectral gap among all Hermitian two-local circuit ensembles on every connected graph with at least three vertices, making it the fastest such circuit for forming unitary 2-d","keywords":["random quantum circuits","iSWAP gate","spectral gap","second-moment operator","unitary 2-designs","Perron-Frobenius","four-point inequalities","KAK parameters"],"falsifier":"Compute, for a small connected graph such as a triangle or a four-vertex path, the full second-moment spectrum of an edge-dependent Hermitian ensemble with parameters on the boundary β=2−3α/5. If any choice of αₑ in [0,10/9] gave min Spec(M_α) > γ₀, the certified comparison in Theorem 6.3 would fail; more directly, if any Hermitian two-local ensemble had ∆(Tᴱ) > ∆(T^iSWAP), Theorem 1.1 would be false. The explicit matrix formulas make this a finite numerical search.","tokens_in":17169,"feed_emoji":"🎲","tokens_out":7733,"duration_ms":77249,"temperature":0.7,"pith_summary":"Random quantum circuits are a practical way to generate approximate unitary 2-designs, but how quickly they converge depends on which two-qubit gate is applied. This paper proves that the iSWAP gate is the best possible choice in a precise sense: on every connected graph with three or more qubits, and among all two-local circuit ensembles whose second-moment operator is Hermitian, the iSWAP circuit has the largest spectral gap. A larger spectral gap means faster convergence toward the uniform Haar distribution over unitaries. The proof resolves a conjecture from the recent gate-design literature by reducing the infinite family of gates to a two-parameter affine boundary family and certifying the comparison through a Perron-Frobenius eigenvector and a cone of four-point inequalities. The result holds for mixtures of gates and for edge-dependent gate choices, not just for a single fixed gate.","feed_headline":"iSWAP gates are the fastest mixers in random quantum circuits","feed_subtitle":"On any connected graph with three or more qubits, no Hermitian two-local gate ensemble beats iSWAP's 2-design speed.","key_machinery":"The local second-moment gadget of any two-qubit unitary reduces to h(a,c) = αP₊ + βP₋, where P₊ and P₋ are two fixed orthogonal rank-one projections and α,β are functions of the KAK coordinates, constrained to the affine polytope 0≤α≤10/9 and 0≤β≤2−3α/5. The proof works on the upper boundary β=2−3α/5, applies the similarity transform that maps the two product zero modes to the consensus states, and obtains a block M_α indexed by non-empty proper subsets of V; this block is a Z-matrix, meaning all off-diagonal entries are non-positive. The named device carrying the argument is the cone C of asymmetric four-point inequalities 4f(A∪{i}) + f(A∪{j}) ≥ 2f(A) + 2f(A∪{i,j}), plus the swapped version","core_discovery":"The central claim is Theorem 1.1: for any connected graph G with n≥3, and any prescribed two-qubit probability distribution on each edge of G, if the resulting second-moment operator is Hermitian, then its spectral gap is at most that of the homogeneous iSWAP ensemble. Since the spectral gap controls the asymptotic convergence rate of the random circuit to a unitary 2-design, this identifies iSWAP as the universal maximiser of second-moment mixing speed among Hermitian two-local circuits. The proof has three stages. First, every ironed two-qubit gadget is shown, through its KAK coordinates, to lie in a convex envelope whose upper boundary is a one-parameter family h_α. Second, a ground-space","pith_inferences":["A natural testable extension is to non-uniform edge selection: the theorem assumes each edge is chosen uniformly at each step, and the edge-dependent comparison uses that uniform weight; reweighting edges would require a modified derivative certificate and is not covered by the paper.","The proof is specific to the second moment; a similar question for third- or higher-order moments would need a new local compression, since the two-projector reduction and the four-point cone rely on the t=2 structure, and nothing in the paper suggests the same gate would automatically be optimal at higher order.","Because the four-point cone is finite and explicit, one could independently verify the spectral comparison on any small fixed graph by solving the cone inequalities numerically—an easy check for n=3 or n=4 that would exercise the proof without full circuit simulation.","If the imported localisation step for iSWAP were shown to hold for other gate families with a less negative minimum eigenvalue, the same compression argument could yield optimality statements for broader families of gates; this is an editorial speculation, not a claim of the paper."],"forward_implications":["Any Hermitian two-local circuit ensemble on a connected graph with n≥3 converges to a unitary 2-design no faster than the iSWAP circuit; iSWAP sets a universal speed limit for second-moment mixing.","The optimality survives mixtures and edge-dependent gate distributions, so practical circuit implementations may vary gates from edge to edge without exceeding the iSWAP bound.","The theorem covers Hermitian ensembles that may have negative eigenvalues; even allowing such negative spectrum does not improve the spectral gap beyond iSWAP.","The proof reduces the infinite gate comparison to a finite algebraic certificate per graph, so the claimed optimality can in principle be verified mechanically on any fixed graph."],"fun_headline_variants":["iSWAP sets the speed limit for random quantum circuits","iSWAP maximizes mixing speed in random quantum circuits","iSWAP proves fastest for two-local quantum mixing","iSWAP is the optimal two-local gate for 2-design speed"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The proof leans on a previously established localisation result that for iSWAP circuits on n≥3 qubits the slowest mode of the full second-moment operator lies in the single-qubit Haar-invariant subspace, and that the full spectral gap equals the compressed gap; if that localisation failed, the comparison would only constrain a compressed operator and would not transfer to the true circuit gap.","fun_headline_variants_meta":{"raw":{"variants":["iSWAP sets the speed limit for random quantum circuits","iSWAP maximizes mixing speed in random quantum circuits","iSWAP proves fastest for two-local quantum mixing","iSWAP is the optimal two-local gate for 2-design speed"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000863,"raw_usage":{"total_tokens":3519,"prompt_tokens":623,"completion_tokens":2896,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":367,"completion_tokens_details":{"reasoning_tokens":2828}},"tokens_in":367,"tokens_out":2896,"duration_ms":18622,"temperature":1.0,"reasoning_tokens":2828,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T04:47:08.588063+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute, for a small connected graph such as a triangle or a four-vertex path, the full second-moment spectrum of an edge-dependent Hermitian ensemble with parameters on the boundary β=2−3α/5. If any choice of αₑ in [0,10/9] gave min Spec(M_α) > γ₀, the certified comparison in Theorem 6.3 would fail; more directly, if any Hermitian two-local ensemble had ∆(Tᴱ) > ∆(T^iSWAP), Theorem 1.1 would be false. The explicit matrix formulas make this a finite numerical search.","supporting_citations":[],"review_version":1}