{"id":"e9719347-7355-4339-b98e-c16e69900e0b","arxiv_id":"2607.18824","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For any irreducible representation of a compact Lie group, a Kac–Moody unitarising measure at level κ yields a non-trivial positive-matrix-valued random measure on the circle in the L²-phase.","lead":"A new kind of random measure—'non-Abelian multiplicative chaos'—produces matrix-valued measures on a circle from Lie-group representations and Kac-Moody symmetry, extending Gaussian multiplicative chaos beyond scalars. The construction works in the L²-phase and reproduces conformal weights from WZW theory, though one part of the convergence proof is only sketched.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"L^2 proof only controls scalar trace, not H_V-valued measure; Theorem 1.1's matrix convergence is not established.","rationale":"The reader's verdict already flags the scalar-trace issue in the rationale, so I partially agree. However, the reader's 'weakest_assumption' field emphasizes the dependence on the unverified Kac–Moody measure νκ, whereas I see the most load-bearing concern as an internal gap: the proof's L^2 estimate does not match the claimed H_V-valued convergence. This gap is concrete and directly affects the central theorem, independent of whether νκ exists. The νκ dependency is a serious external concern, but it is acknowledged and would be settled by inspecting the companion paper; the L^2 gap requires new estimates within this paper. I recommend no change to the conditional verdict because the gap is likely fixable via additional two-point estimates, but the paper as written does not prove the stated convergence.","tokens_in":12532,"tokens_out":6290,"duration_ms":54595,"concrete_test":"Recompute the L^2 distance in H_V directly: show that E[||M^ε(A)-M^δ(A)||^2_HS] → 0 as ε,δ→0, where ||·||_HS is the Hilbert–Schmidt norm. This requires an estimate on the two-point function of the form E[Tr(ϱ(H(z1)) ϱ(H(z2)))] or E[Tr(ϱ(H(z1)) b) Tr(ϱ(H(z2)) b')] for arbitrary b,b' in the Lie algebra. If such an estimate cannot be derived from the existing Proposition 3.4, or if a numerical or explicit counterexample shows the distance does not vanish, then the proof of Theorem 1.1 is incomplete. A simpler partial check: replace (3.12) with the Hilbert–Schmidt analogue and see whether the same G_{κ,ρ} bound suffices; if extra divergent terms appear, the proof fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 3.3 claims that (M^ε(A))_ε is Cauchy in L^2(νκ;H_V), but the displayed computation (3.12) only controls the scalar trace difference: it computes E[(Tr(M^ε(A)-M^δ(A)))^2], not E[||M^ε(A)-M^δ(A)||^2_{H_V}]. The space H_V is a cone of positive Hermitian operators; two positive operators can have arbitrarily close traces while their Hilbert–Schmidt or operator distance is large. Therefore the computed convergence of Tr(M^ε(A)) does not imply convergence of the matrix coefficients or of the H_V-valued measure. The two-point function G_{κ,ρ} in Proposition 3.4 is defined via products of traces, and the proof only provides estimates for such trace correlations. No estimate is given for correlations with insertions such as Tr(ϱ(H(z1)) b) Tr(ϱ(H(z2)) b') or Tr(ϱ(H(z1)) ϱ(H(z2))), which would be needed to control the matrix-valued second moment. Hence Theorem 1.1's conclusion that M^ε converges in probability as an H_V-valued measure does not follow from the argument presented. This is an internal gap in the proof, independent of the status of the Kac–Moody measure νκ.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a non-Abelian, non-Gaussian generalisation of multiplicative chaos on the unit circle. For an irreducible unitary representation ρ of a compact connected Lie group G, and for κ below the L² threshold, it defines a renormalised family of random H_V-valued measures M^ε_{κ,ρ} using the 'Kac–Moody unitarising measure' ν_κ of [Bav26]. Theorem 1.1 asserts weak convergence in probability to a non-trivial H_V-valued measure. The proof relies on an exact one-point function (Prop 3.3) and a two-point bound (Prop 3.4), derived from 'one-dimensional Knizhnik–Zamolodchikov equations' via an integration-by-parts formula. The final step in §3.3 aims to show the measures are Cauchy in L².","tokens_in":12855,"tokens_out":15718,"duration_ms":127552,"significance":"If correct, this would be a substantial contribution: it constructs a matrix-valued multiplicative chaos in a non-Gaussian setting, with explicit exponents matching WZW conformal weights, and no fitted parameters. The one-point computation is clean and the overall strategy is attractive. However, the paper currently leaves a load-bearing gap between the scalar trace estimates and the claimed H_V-valued convergence, and the two-point inequality (3.7) appears to be false as stated. The construction is also entirely conditional on the unpublished preprint [Bav26]. For these reasons the theorem cannot yet be regarded as established.","major_comments":[{"comment":"Equation (3.12) controls only the L² norm of the scalar trace difference Tr_V(M^ε(A)−M^δ(A)). Convergence of traces does not imply convergence in H_V: two positive Hermitian operators can have arbitrarily close traces while their Hilbert–Schmidt or operator distance is large. The sentence 'This proves that the sequence (M^ε(A)) is Cauchy in L²(νκ;H_V)' is therefore not justified. To prove Theorem 1.1 one needs second-moment estimates for the matrix coefficients, for example for ⟨u,ϱ(H(z))v⟩⟨u',ϱ(H(w))v'⟩ or at least for Tr(ϱ(H(z))b)Tr(ϱ(H(w))b'). As written, the argument only establishes convergence of the trace measures Tr_V(M^ε), not of the H_V-valued measures.","section":"§3.3, Eq. (3.12)"},{"comment":"The derivation of (3.7) from (3.11) is invalid as a two-sided inequality for all z1,z2∈D. The prefactor 4Re(ζ|z|^2/(ζ|z|^2−1)) = 2∂_r log|1−r^2ζ| changes sign as r varies for generic ζ. Hence the weighted average of the spectrum of S=Σ_b ρ(b)⊗ρ(b) cannot be sandwiched between λ_− and λ_+ uniformly. In fact, since λ_+≥λ_− and κ+ˇh<0, the claimed lower and upper bounds cross at |1−z1\\bar z2|=1; for e.g. z1=0.9, z2=0.9i the stated lower bound exceeds the stated upper bound. The L¹ domination argument on S¹×S¹ relies on this inequality, so the convergence statement in Proposition 3.4 is not established. The authors should correct the bound (e.g., with a constant and a restricted regime) or provide a separate argument away from the diagonal.","section":"Prop. 3.4, Eq. (3.7)"},{"comment":"The proof depends on unproved properties of ν_κ from [Bav26]: the integration-by-parts formula (2.4), G-invariance of the lifted measure, and the conjugation symmetry law(g)=law(z↦g^*(z̄)) used in (3.10). The latter is not proved here, and the 'straightforward adaptations' leading to (3.9) and the antisymmetry argument after (3.10) are omitted. Since [Bav26] is an unpublished preprint by the same author, the paper should either state the precise theorem from [Bav26] that supplies these properties or prove them. As written, Theorem 1.1 is conditional on external facts.","section":"Prop. 3.4 and §2.3"}],"minor_comments":[{"comment":"In the last term of the integrand, the second argument should be e^{-δ+iθ_2}, not e^{-δ+iθ_1}.","section":"Eq. (3.12)"},{"comment":"The symbol 'tr' is used both for the normalized trace on g_C and for Tr_V(ϱ(·)); this should be disambiguated, especially in (3.9)–(3.11).","section":"Throughout"},{"comment":"The construction relies crucially on [Bav26], which is an unpublished preprint. The authors should mark it clearly as such and state which precise results are being imported.","section":"References"},{"comment":"The 'weak topology' on M(S¹;H_V) is defined by convergence on all Borel sets, which is stronger than the usual weak topology on measures. This choice should be clarified, as it affects the notion of 'weak convergence in probability' used in Theorem 1.1.","section":"Section 2.4"}],"recommendation":"major_revision","confidential_remarks":"The dependence on the author's own unpublished preprint [Bav26] is a significant evaluation issue. I recommend asking the author to either include the necessary statements from [Bav26] or make the preprint available in final form. The paper is potentially interesting but not yet self-contained, and the trace-versus-matrix gap in §3.3 is a serious technical hurdle that will require new estimates."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague —\n\nThe one thing to know: the construction is genuinely new and the algebra is beautiful, but the proof of the main theorem has a load-bearing gap. The conclusion that the renormalised measures converge as H_V-valued measures does not follow from the estimates provided.\n\nWhat's actually new: this is the first non-Gaussian, non-Abelian matrix-valued multiplicative chaos on the circle. The one-point function is computed exactly and reproduces WZW conformal weights. The use of 1D Knizhnik–Zamolodchikov equations to derive correlation bounds is a nice idea and, as far as I can tell, correct in structure. The paper is also honest about the L^2-phase limitation and about depending on the author's previous unpublished Kac–Moody unitarising measure.\n\nThe soft spot: Section 3.3, equation (3.12). The author claims that convergence of Tr(M^epsilon(A) - M^delta(A)) in L^2 makes the sequence Cauchy in L^2(nu; H_V). That's a non sequitur. Trace convergence of positive operators does not control operator or Hilbert–Schmidt distance — two positive matrices can have arbitrarily close traces while being far apart. To get convergence of the matrix coefficients you need second moments with insertions like Tr(rho(H) b) Tr(rho(H') b'), not just products of full traces. Those are not computed. So Theorem 1.1's statement that M^epsilon converges weakly in probability as an H_V-valued measure is not proven by the argument given. This is an internal gap, independent of the status of [Bav26].\n\nThe dependence on [Bav26] is a dependency, not circularity, but it's a heavy one: existence, G-invariance, integration by parts, and a conjugation symmetry are all imported from an unpublished preprint. For a paper built on that, there should be a verification or a clear statement of what is assumed.\n\nThe two-point bound itself is plausible but has several 'straightforward' adaptations in the proof of Proposition 3.4 that I did not check line by line. That part may need cleaning, but I don't see an obvious error beyond the missing details.\n\nIf the trace-level convergence can be upgraded to full matrix-valued second moment estimates, the result would be solid and significant for the multiplicative chaos / CFT community. As it stands, the main theorem overstates what the proof establishes.\n\nMy recommendation: send it to peer review — the idea is worth refereeing — but the referee should ask for the matrix-valued moment computation or a corrected convergence argument before acceptance. It's a paper to engage with, not to dismiss.\n\nBest,\n\n[You]","headline":"New and interesting object, but the convergence proof for the matrix-valued measure only controls the scalar trace; Theorem 1.1 as written is not established.","tokens_in":13333,"tokens_out":3182,"would_cite":true,"duration_ms":30680,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60G57","81R10"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper constructs a non-Gaussian, non-Abelian matrix-valued multiplicative chaos measure on the circle, for every irreducible representation of a compact connected Lie group, in the L2 range.","keywords":["multiplicative chaos","non-Gaussian","non-Abelian","Kac-Moody unitarising measure","Knizhnik-Zamolodchikov equations","Hermitian Yang-Mills metrics","random matrix-valued measures","conformal field theory"],"falsifier":"Check the construction of ν_κ in the earlier preprint: if the integration-by-parts formula (2.4) or the conjugation symmetry law(g)=law(z↦g*(z̄)) fails for any admissible u, the derivation of Lemmas 3.1–3.2 and hence the differential equation for the one-point function breaks. Concretely, computing E[Tr ϱ(H(z))] from the definition of ν_κ and checking whether it equals (1-|z|²)^{Ω_ρ/(κ+h)} for a single non-trivial representation would settle the central claim.","tokens_in":12423,"feed_emoji":"🌀","tokens_out":5294,"duration_ms":47297,"temperature":0.7,"pith_summary":"This paper aims to construct a multiplicative chaos measure that is simultaneously non-Gaussian and non-Abelian: for every irreducible unitary representation of a compact connected Lie group, a family of renormalised random Hermitian positive definite matrix-valued measures on the circle is shown to converge weakly in probability to a non-trivial limit. The construction feeds a Kac–Moody unitarising measure on Hermitian Yang–Mills metrics in the disc into the representation, renormalises by a power of the Poincaré metric, and takes the boundary limit. The proof is a second-moment argument, valid in the L2-range of the parameter κ; at the critical boundary value κ = -h - 2λ+ρ the method stops, not because of a true phase transition but because second moments fail. A reader should care because this extends multiplicative chaos beyond Gaussian fields and scalar values, and because the correlation functions are obtained explicitly by solving a one-dimensional Knizhnik–Zamolodchikov equation, linking random geometry to conformal field theory.","feed_headline":"Non-Abelian multiplicative chaos built on the circle","feed_subtitle":"Renormalised matrix-valued measures converge for every irreducible representation of a compact Lie group.","key_machinery":"The engine of the proof is the integration-by-parts formula (2.4) for the Kac–Moody unitarising measure, which shifts the effective level from κ to κ + h, the dual Coxeter number. Two variational lemmas (Lemmas 3.1–3.2) convert this formula into differential equations for the one- and two-point functions of Tr ϱ(H(z)); the resulting equations are a one-dimensional version of the Knizhnik–Zamolodchikov equations. The solution for the one-point function is the Poincaré-metric power (1-|z|²)^{Ω_ρ/(κ+h)}, and the two-point bounds use the extremal eigenvalues λ±_ρ of ∑_b ρ(b)⊗ρ(b) on the tensor-product decomposition. These estimates control the collision singularity and give the L1 convergence ne","core_discovery":"The central claim, Theorem 1.1, is that the renormalised measure dMε_{κ,ρ}(e^{iθ}) = (1-e^{-2ε})^{-Ω_ρ/(κ+h)} ϱ(H(e^{-ε+iθ})) dθ, with H drawn from the Kac–Moody unitarising measure, converges weakly in probability as ε→0 to a non-trivial H_V-valued measure M_{κ,ρ}, for every irreducible unitary representation ϱ and for κ < min(-2h, -h - 2λ+_ρ). The normalisation is chosen so that the one-point function is exactly the identity: E[ϱ(H(z))] = (1-|z|²)^{Ω_ρ/(κ+h)} Id_V. The two-point function is sandwiched between powers of |1-z1 ar z2|, with exponents controlled by the spectrum of the Casimir operator on V⊗V. These explicit expressions are derived, rather than estimated abstractly, from the in","pith_inferences":["If the imported Kac–Moody measure exists with the assumed properties, the same one-dimensional KZ machinery could produce explicit higher-order correlation functions, giving access to joint laws of the limit measure at several boundary points.","The suggested rational multifractal spectrum could be tested by simulating the finite-dimensional truncations of the Kac–Moody measure and measuring local scaling of Tr ϱ(H(e^{-ε}z)) on the circle.","The absence of a unitary limit for an imaginary counterpart suggests an analytic-continuation route in κ, mirroring imaginary Gaussian chaos; whether such a distribution-valued object exists is open.","The method may adapt to higher-dimensional domains or to non-trivial holomorphic bundles, where the role of the Hermitian Yang–Mills equation and the KZ equation would need a higher-dimensional generalisation."],"forward_implications":["For every irreducible representation ϱ in the stated range, the limiting object is a genuine random measure with values in positive definite Hermitian matrices; its total mass is finite and its one-point density is normalised to the identity.","The Kac–Moody unitarising measure, whose samples do not converge on the boundary, is shown to have a boundary limit after renormalisation, so the construction supplies an intrinsic circle sample space.","The correlation exponents coincide with the conformal weights of the WZW model, making the measure a probabilistic realisation of a boundary conformal-field-theory object.","The theorem is confined to the L2 regime: at κ = -h - 2λ+_ρ the proof breaks down, and the paper explicitly notes that no matrix-valued multiplicative chaos in the full L1 phase is currently available.","A heuristic Mellin computation suggests that, beyond the L2 phase, the multifractal spectrum would be a rational function rather than the polynomial spectrum of Gaussian multiplicative chaos."],"fun_headline_variants":["Circle multiplicative chaos goes non-Abelian","Exact correlations for non-Gaussian chaos on a circle","KZ equations pin down non-Abelian chaos","Matrix-valued chaos measures from Lie groups","Non-Abelian twist on multiplicative chaos"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that the Kac–Moody unitarising measure ν_κ exists on the space of Hermitian Yang–Mills metrics with the integration-by-parts formula (2.4), G-invariance, and the conjugation symmetry law(g)=law(z↦g*(z̄)) used in (3.10); these properties are imported from an earlier unpublished preprint and are not proved or independently verified in this paper, so Theorem 1.1 collapses if any of them fails.","fun_headline_variants_meta":{"raw":{"variants":["Circle multiplicative chaos goes non-Abelian","Exact correlations for non-Gaussian chaos on a circle","KZ equations pin down non-Abelian chaos","Matrix-valued chaos measures from Lie groups","Non-Abelian twist on multiplicative chaos"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001614,"raw_usage":{"total_tokens":6264,"prompt_tokens":747,"completion_tokens":5517,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":491,"completion_tokens_details":{"reasoning_tokens":5447}},"tokens_in":491,"tokens_out":5517,"duration_ms":38662,"temperature":1.0,"reasoning_tokens":5447,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T14:11:52.387948+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Check the construction of ν_κ in the earlier preprint: if the integration-by-parts formula (2.4) or the conjugation symmetry law(g)=law(z↦g*(z̄)) fails for any admissible u, the derivation of Lemmas 3.1–3.2 and hence the differential equation for the one-point function breaks. Concretely, computing E[Tr ϱ(H(z))] from the definition of ν_κ and checking whether it equals (1-|z|²)^{Ω_ρ/(κ+h)} for a single non-trivial representation would settle the central claim.","supporting_citations":[],"review_version":1}