{"id":"ec449ff6-3ae4-415b-940b-67cbf2a7ff3e","arxiv_id":"2412.14813","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For rotationally symmetric interactions on high-dimensional spheres, the paper characterizes bifurcation branches and proves a sufficient condition for a discontinuous phase transition.","lead":"This paper finds conditions under which the stationary McKean-Vlasov equation on a sphere or another curved space has multiple solutions and sharp phase transitions. It applies these results to models inspired by transformer neural networks, liquid crystals, and opinion dynamics.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The proof of Theorem 5.11 is not internally consistent: the energy expansion in Lemma 5.12 mixes normalized and unnormalized spherical measures, so the cancellation that gives Fγ♯(ρ♯)<Fγ♯(¯ρ) is not demonstrated. Until this is fixed, γ_c<γ♯ is unproven.","rationale":"The strongest claim is the pair of results in Theorem 4.9 (bifurcation branches) and Theorem 5.11 (discontinuous transition below γ♯). The proof of Theorem 5.11 is the most fragile part of the paper: it depends on Lemma 5.12, whose Taylor expansion must be computed with one fixed measure. The manuscript as written mixes the normalized volume measure m of Section 2 with the unnormalized spherical measure σ of Section 3, and the displayed cancellation in Lemma 5.12 is sensitive to that choice. This is a concrete, load-bearing gap, not a question of taste or consensus. It is also distinct from the reader's weakest_assumption about the genericity of Assumption 5.8: even granting Assumption 5.8, the proof of the strict energy inequality is not yet rigorous as written. The defect is repairable—one expects the same competitor to work after a consistent rescaling in the normalized Hilbert space—so the appropriate verdict remains CONDITIONAL rather than REJECT. The bifurcation part is on much firmer ground: the Crandall–Rabinowitz framework is applied to an invariant subspace Ls2, the linearized operator is diagonal in the zonal harmonics, and the Fredholm/simplicity checks are standard, modulo minor typos in the stated regularity lemmas. I do not see a reason to change the reader's conditional verdict, but the normalization issue should be fixed before the phase-transition theorem is taken as established.","tokens_in":43694,"tokens_out":39399,"duration_ms":294646,"concrete_test":"Rewrite Lemma 5.12 entirely in normalized measure m=σ/ω_n: let the uniform density be 1, set ρ♯=1+εξu with u=Σ c_lY_{l,0} orthonormal in L²(m) and ∫u dm=0, and denote by W*_m u = Σ \\hat W_l c_l Y_l its spectral multiplier. Expand E_m and I_m consistently and recompute Fγ♯(ρ♯)−Fγ♯(1). If the second-order terms cancel exactly at γ♯ and the ε³ term controls the remainder under Assumption 5.8, the theorem stands; if an extra factor ω_n² or a missing ¯ρ appears, Lemma 5.12 and Theorem 5.11 require revision. As a numerical cross-check, evaluate the corrected expression for the Onsager kernel at n=3 using γ♯ from Proposition 6.4 and verify the energy difference is negative.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The discontinuous-transition claim (Theorem 5.11) rests entirely on Lemma 5.12, which must produce a state ρ♯ with Fγ♯(ρ♯)<Fγ♯(¯ρ). In the proof, ρ♯=¯ρ(1+εξu) is treated sometimes as a density w.r.t. the unnormalized sphere measure σ and sometimes as a density w.r.t. the normalized volume m. Section 2 defines E(µ)=∫ρ logρ dm with m(M)=1, while Section 3.1 uses σ, ω_n=σ(S^{n-1}), ¯ρ=1/ω_n, and the scalar product ⟨f,g⟩=ω_n^{-1}∫fg dσ. With that scalar product, an orthonormal Y_{l,0} satisfies ∫Y_{l,0}² dσ=ω_n, and Equation (22) gives ∫u(W*u)dσ = ω_n² Σ_l \\hat W_l \\hat u_l², not the expression used in Lemma 3.15 and Lemma 5.12. The Taylor expansion in Lemma 5.12 cancels the second-order entropy term (1/(2γ♯))ε²∥u∥² against a term (1/(2γ♯))ε²¯ρ Σ \\hat W_l c_l²∥Y_{l,0}∥²; this cancellation can hold at γ♯ only for one particular normalization, which is never stated or derived. Therefore the sign of Fγ♯(ρ♯)−Fγ♯(¯ρ) is not established as written. Since this strict inequality is the only argument putting the transition point strictly below γ♯, the discontinuity conclusion is unproven pending a consistent recalculation. The issue appears repairable—the same competitor written as density 1+εξu in L²(m) likely works—but the current manuscript does not contain that calculation.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies stationary solutions of the McKean-Vlasov equation on compact Riemannian manifolds, with emphasis on the high-dimensional sphere. It claims three main bodies of results: (i) equivalence between weak solutions, zeros of the Gibbs map, and critical points of a free-energy functional, together with existence, uniqueness and small-noise limits on general compact manifolds; (ii) for rotationally symmetric kernels on S^{n-1}, bifurcation branches from the uniform state at values γ_k = -1/\\hat W_k determined by spherical-harmonic coefficients, with an explicit description of the bifurcating states in terms of the harmonics Y_{k,0}; and (iii) a sufficient condition, called the relaxed resonance condition, for the existence of a discontinuous phase transition point γ_c strictly below the linear-stability threshold γ_♯. The theoretical results are illustrated on the noisy transformer model, the Onsager model, a spherical opinion-dynamics model, and a localized spherical Gaussian kernel.","tokens_in":44147,"tokens_out":18515,"duration_ms":153223,"significance":"If the results are correct, the paper gives a substantial manifold analogue of the torus analysis of Carrillo–Gvalani–Pavliotis–Schlichting, with explicit spectral conditions and concrete applications to models of current interest. Strengths include the self-contained derivation, the explicit assumptions in Theorems 4.9 and 5.11, and the concrete Gegenbauer/Bessel computations in the examples; no parameters are fitted and the claims are falsifiable. However, the proof of the discontinuous-transition theorem currently rests on an energy comparison whose normalization is inconsistent as written, and the proof of the central equivalence in Proposition 2.5 uses a false regularity assertion. These issues are load-bearing for the paper's main new claims, so the results cannot be accepted in their present form.","major_comments":[{"comment":"The proof asserts that every H^1 function on a compact Riemannian manifold is almost everywhere continuous, which is false in dimension n ≥ 2. The subsequent argument evaluating γ^{-1}log ρ + W*ρ at a 'point of continuity' x_0 is therefore not justified, and the implication that every critical point of F_γ is a zero of the Gibbs map is not proven as written. A Lebesgue-point argument should replace the continuity claim; this is likely repairable, but the repair is not present in the manuscript.","section":"§2.1, Proposition 2.5, proof of (3)→(2)"},{"comment":"The cancellation of the second-order terms in the energy difference F_{γ_♯}(ρ_♯) - F_{γ_♯}(ρ̄) mixes normalizations. With the scalar product ⟨f,g⟩ = ω_n^{-1} ∫ f g dσ introduced in Section 3.1, an orthonormal spherical harmonic satisfies ∫ Y_{l,0}^2 dσ = ω_n, and Equation (22) yields ∫ u (W*u) dσ = ω_n^2 Σ_l \\hat W_l \\hat u_l^2, not the expression used in Lemma 3.15 and Lemma 5.12. The entropy expansion is naturally written with respect to the normalized volume m, while the interaction expansion is written with respect to σ; the second-order cancellation at γ_♯ works only under one particular normalization of ∥u∥^2, which is never stated or derived. Consequently the strict inequality F_{γ_♯}(ρ_♯) < F_{γ_♯}(ρ̄) is not established as written. Since Theorem 5.11 derives γ_c < γ_♯ solely from this lemma, the discontinuous-phase-transition conclusion is currently unproven. The calculation appears repairable by working with the density 1 + εξu in L^2(m), but that calculation must be supplied.","section":"§5.3, Lemma 5.12"},{"comment":"Theorem 2.8 proves the existence of an L^2 minimizer of F_γ, but it does not prove that this minimizer is a critical point of F_γ or a zero of the Gibbs map; Proposition 2.5 requires ρ ∈ H^1(M) ∩ P^+_ac(M). The introduction's Theorem 1.1 states that for every γ there exists a solution ρ_γ ∈ H^1(M) ∩ P_ac(M) of (1), but the proof chain from the minimizer to the weak solution is missing. This is a further load-bearing gap in the existence claim, even though the final statement is plausible and probably standard.","section":"§2.2, Theorem 2.8 and Theorem 1.1"}],"minor_comments":[{"comment":"The normalizing constant is written as c^{n-2/2} in Definition 3.6 and later as c^λ; Definition 1.2 uses α_k without defining it. Please align these notations.","section":"§3.1, Definition 3.6"},{"comment":"The summation index set K_{α,δ} is not defined; it should presumably be K_{♯,δ} from the preceding display.","section":"§5.3, Assumption 5.8"},{"comment":"The displayed formula for γ_k appears to be missing a fraction or an explicit denominator; please check it against the computed spherical-harmonic coefficient \\hat W_{β,k}.","section":"§6.1, Proposition 6.1"},{"comment":"The phrase 'rational interaction kernel' is confusing for the kernel W_p(x,y) = -(1+⟨x,y⟩)^p; it should presumably be 'rotationally symmetric' or 'polynomial-type interaction kernel'.","section":"§6.3"},{"comment":"There is a typo in the displayed estimate: '(\\|u\\|2 + \\|v\\|)2' should read '(\\|u\\|^2 + \\|v\\|^2)'.","section":"§2.4, Theorem 2.19 proof"}],"recommendation":"major_revision","confidential_remarks":"The paper is a well-motivated extension of the torus analysis to manifolds, and the bifurcation and example computations are valuable. The normalization issue in Lemma 5.12 is likely repairable, but as it stands the strict inequality γ_c < γ_♯ is not proven, so Theorem 5.11 cannot be accepted. I see no reason to doubt the novelty or the relevance; the requested revision is local in nature but essential."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nBottom line: this is a serious paper that deserves a referee, but the headline discontinuity claim is not proven as written. The normalization in Lemma 5.12 mixes the normalized volume measure m and the unnormalized spherical measure σ; the cancellation that yields Fγ♯(ρ♯)<Fγ♯(ρ̄) at the linear stability threshold does not actually go through. The stress-test note is right: with the scalar product ⟨f,g⟩=ω_n^{-1}∫fg dσ, an orthonormal Y_{l,0} has ∫Y_{l,0}²dσ=ω_n, and equation (22) gives ∫u(W*u)dσ = ω_n² Σ \\hat W_l \\hat u_l², not the expression used in Lemma 3.15 and Lemma 5.12. So γ_c<γ♯ is currently unproven. This is repairable—a competitor written as density 1+εξu in L²(m) likely works—but the calculation isn't in the manuscript.\n\nWhat is genuinely new and good: the paper gives the first systematic treatment of stationary McKean-Vlasov solutions on arbitrary-dimensional spheres, including the free-energy/Gibbs-map equivalence on compact manifolds, a geodesic convexity criterion, explicit bifurcation branches under a unique most-negative spherical harmonic (Theorem 4.9), and a verifiable resonance condition for discontinuous transitions. The single-mode self-resonance for k=2,4 is a nice sphere-specific observation with no torus analogue. The examples (Onsager, opinion dynamics, localized heat kernel, noisy transformer) are concrete and useful, and the computations are mostly checkable.\n\nSoft spots beyond the Lemma 5.12 issue: Proposition 2.5 claims H^1 functions are almost everywhere continuous, which is false for n≥2; the proof needs a Lebesgue-point or smoothing argument. Assumption 5.8 is under-specified about what \"linear combination\" and the normalization of u mean; that should be tightened. Lemma 4.10's appendix proof also has a questionable step, but it looks fixable.\n\nWho this is for: anyone working on McKean–Vlasov equations, mean-field limits with phase transitions, or interacting particle systems on manifolds. The bifurcation half of the paper is solid and citable; the phase-transition half needs a revision. I'd bring it to reading group and would cite the bifurcation results, but I'd avoid citing Theorem 5.11 until the normalization is fixed.\n\nRecommendation: accept for peer review, with a request that the referee check Lemma 5.12 and the regularity claim carefully. With those fixed, it's a strong paper.","headline":"Solid bifurcation theory on the sphere, but the discontinuous phase transition proof has a normalization gap that currently leaves the main discontinuity claim unproven.","tokens_in":44644,"tokens_out":5404,"would_cite":true,"duration_ms":37085,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35B32","35Q83","58J35","82C22"],"pacs":[],"model":"deepseek-v4-flash","headline":"For rotationally symmetric interaction kernels on a sphere, the paper proves that a unique negative spherical-harmonic coefficient produces bifurcation branches from the uniform state, and that a relaxed resonance condition forces a…","keywords":["McKean-Vlasov equation","stationary solutions","bifurcation","phase transition","spherical harmonics","spherical convolution","free energy functional","Riemannian manifolds"],"falsifier":"Run a numerical continuation of the stationary McKean-Vlasov equation on $S^2$ with the Onsager kernel $W(x,y)=\\sqrt{1-\\langle x,y\\rangle^2}$: the paper predicts that the uniform state stops being a global minimizer at some $\\gamma_c$ strictly below $\\gamma_\\sharp = -1/\\hat W_2$, and if the first non-uniform minimizer instead appears exactly at $\\gamma_\\sharp$, the strict inequality in Theorem 5.11 would be contradicted. A second check is to compute $\\int_{S^{n-1}} Y_{1,0}^3\\,d\\sigma$ for the noisy transformer kernel; it vanishes because $Y_{1,0}$ is odd, confirming that the relaxed resonance condition is objectively absent there.","tokens_in":43473,"feed_emoji":"🌐","tokens_out":7878,"duration_ms":59056,"temperature":0.7,"pith_summary":"The paper studies stationary states of the McKean-Vlasov equation, the mean-field PDE for interacting diffusions, placed on a compact Riemannian manifold and in particular on the unit sphere $S^{n-1}$. It establishes that stationary states coincide exactly with fixed points of the Gibbs map and critical points of a free energy functional, and that minimizers exist for every inverse temperature $\\gamma$ while being unique for small $\\gamma$ under a curvature condition. On the sphere, the paper shows that a rotationally symmetric interaction kernel is diagonalized by spherical harmonics, so each simple negative coefficient $\\hat W_k$ produces a bifurcation branch of stationary states emerging from the uniform state at $\\gamma_k = -1/\\hat W_k$. The main new result is a sufficient spectral condition: if the most-negative modes can be combined into a function whose cube has nonzero spherical average, the system has a discontinuous phase transition at some $\\gamma_c$ strictly below the linear-stability threshold $\\gamma_\\sharp$. The condition is verified for the Onsager model, a spherical opinion-dynamics kernel, and a localized Gaussian heat kernel, while it fails for the noisy transformer kernel, leaving that transition type open.","feed_headline":"Discontinuous phase transition pinned below stability threshold","feed_subtitle":"On spheres, a resonant cubic mode makes the uniform state lose global optimality before it loses linear stability","key_machinery":"The load-bearing object is the spherical harmonics decomposition $\\hat W_k$ of the rotationally symmetric kernel, defined through Gegenbauer polynomials, together with the spherical convolution theorem: for any function $u$, the projection of $W * u$ onto the $k$-th harmonic subspace equals $\\omega_n \\hat W_k$ times the projection of $u$. This diagonalizes the linearized Gibbs map and reduces the bifurcation problem to a simple characteristic value of a compact operator. The phase-transition argument additionally uses the cubic moment $\\int_{S^{n-1}} u^3\\,d\\sigma$ of a competitor built from the minimal modes: the third-order term in the Taylor expansion of the free energy is what makes the uniform state lose global minimality strictly before linear instability. The relaxed resonance condition quantifies how small the bandwidth around the minimal coefficient must be for this cubic term to dominate.","core_discovery":"The central claim is that on $S^{n-1}$, for an interaction kernel of the form $W(x,y)=W(\\langle x,y\\rangle)$, the stationary McKean-Vlasov equation can be solved spectrally. Using the spherical convolution theorem, the linearization of the Gibbs map at the uniform state has eigenvalues $\\lambda_l = -l(n+l-2)(1+\\gamma \\hat W_l)$, so the uniform state becomes linearly unstable exactly when $\\gamma$ exceeds $\\gamma_\\sharp = -1/\\min_l \\hat W_l$. The paper proves (Theorem 4.9) that if a single coefficient $\\hat W_k<0$ is unique among all coefficients, a nontrivial branch $\\rho_\\gamma(t) = \\bar\\rho + f(t)Y_{k,0} + o(f(t))$ bifurcates from the uniform state at $\\gamma_k = -1/\\hat W_k$. It then proves (Theorem 5.11) that under the relaxed resonance condition, the uniform state is already not a global minimizer at $\\gamma_\\sharp$, so the phase transition occurs at some $\\gamma_c \\in (0,\\gamma_\\sharp)$ and is discontinuous. This transfers the resonance mechanism for discontinuous transitions from the torus to the sphere, where a single self-resonant mode such as $k=2$ or $k=4$ can suffice.","pith_inferences":["Beyond the paper, the strength of the cubic moment $U_3 = \\int u^3\\,d\\sigma$ could be computed numerically for any rotationally symmetric kernel, turning the relaxed resonance condition into a practical spectral test for discontinuous phase transitions on spheres.","Because the relaxed resonance condition fails for the noisy transformer kernel, where $k_{\\min}=1$ is odd and $\\int Y_{1,0}^3\\,d\\sigma=0$, the transition type for that model likely depends on higher-order terms in the free-energy expansion, which the paper leaves as an open quantitative question.","The strict gap $\\gamma_c < \\gamma_\\sharp$ predicts hysteresis in particle simulations: slow cooling from high noise should jump to the non-uniform state at $\\gamma_c$, while slow heating from an ordered state could persist until $\\gamma_\\sharp$.","The paper's manifold-level equivalence between stationary solutions, Gibbs fixed points, and critical points suggests that numerical free-energy minimization on the sphere could be used to locate bifurcating branches without solving the PDE directly."],"forward_implications":["For any compact Riemannian manifold with bounded curvature and a sufficiently regular kernel, stationary McKean-Vlasov states are exactly Gibbs fixed points and critical points of the free energy; minimizers exist for all $\\gamma$ and are unique for small $\\gamma$ under a geodesic-convexity condition.","On $S^{n-1}$, every simple negative spherical-harmonic coefficient $\\hat W_k$ yields a bifurcation branch from the uniform state of the form $\\bar\\rho + f(t)Y_{k,0}+o(f(t))$ at $\\gamma_k = -1/\\hat W_k$.","If the relaxed resonance condition holds, the phase transition is discontinuous and occurs at $\\gamma_c < \\gamma_\\sharp$; in particular the uniform state is globally suboptimal while still linearly stable on $(\\gamma_c, \\gamma_\\sharp)$.","For the Onsager kernel and the opinion-dynamics kernel with $p=n+2$, the minimal mode is $k=2$ and is self-resonant, so a discontinuous transition is guaranteed on spheres of arbitrary dimension.","For the localized hyperspherical heat kernel with sufficiently small $\\varepsilon$, the bandwidth condition of the relaxed resonance condition is satisfied, giving a discontinuous transition for small $\\varepsilon$."],"supporting_citations":[{"why":"Supplies the torus Fourier framework for bifurcation branches and resonance conditions that the sphere analysis adapts to spherical harmonics.","marker":"[CGPS20]"},{"why":"Provides the spherical-harmonics orthonormal basis and the convolution theorem used to diagonalize rotationally symmetric kernels.","marker":"[Dai13]"},{"why":"Defines continuous and discontinuous transition points and supplies the lemma that a continuous transition occurs exactly at the linear-stability threshold.","marker":"[CP10]"},{"why":"Gives the bifurcation-from-a-simple-eigenvalue theorem used to construct nontrivial branches.","marker":"[Dei13]"},{"why":"Establishes instability of the kernel as the classical criterion for existence of a phase transition, used in Proposition 5.3.","marker":"[GP70]"},{"why":"Supplies the geodesic-convexity machinery for the entropy term used to prove existence and small-$\\gamma$ uniqueness on manifolds.","marker":"[Stu05]"},{"why":"Introduces the exponential self-attention kernel studied here as the noisy transformer example.","marker":"[GLPR24]"},{"why":"Provides the prior bifurcation analysis of the Onsager model on $S^2$ that the arbitrary-dimension result extends.","marker":"[Vol17]"},{"why":"Gives the explicit hyperspherical heat-kernel representation used for the localized Gaussian example.","marker":"[ZS18]"}],"fun_headline_variants":["Resonant mode forces earlier phase jump on sphere","Sphere: discontinuous transition before instability threshold","McKean-Vlasov on sphere: resonance shifts transition point","One self-resonant mode suffices for early jump on sphere"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The paper's discontinuous-transition result rests on the relaxed resonance condition: the most-negative spherical-harmonic modes must combine into a bounded function whose cube has a nonzero average over the sphere, with the spread of near-minimal modes kept small enough.","fun_headline_variants_meta":{"raw":{"variants":["Resonant mode forces earlier phase jump on sphere","Sphere: discontinuous transition before instability threshold","McKean-Vlasov on sphere: resonance shifts transition point","One self-resonant mode suffices for early jump on sphere"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00023,"raw_usage":{"total_tokens":1475,"prompt_tokens":929,"completion_tokens":546,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":545,"completion_tokens_details":{"reasoning_tokens":480}},"tokens_in":545,"tokens_out":546,"duration_ms":4594,"temperature":1.0,"reasoning_tokens":480,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T11:53:18.771163+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a numerical continuation of the stationary McKean-Vlasov equation on $S^2$ with the Onsager kernel $W(x,y)=\\sqrt{1-\\langle x,y\\rangle^2}$: the paper predicts that the uniform state stops being a global minimizer at some $\\gamma_c$ strictly below $\\gamma_\\sharp = -1/\\hat W_2$, and if the first non-uniform minimizer instead appears exactly at $\\gamma_\\sharp$, the strict inequality in Theorem 5.11 would be contradicted. A second check is to compute $\\int_{S^{n-1}} Y_{1,0}^3\\,d\\sigma$ for the noisy transformer kernel; it vanishes because $Y_{1,0}$ is odd, confirming that the relaxed resonance condition is objectively absent there.","supporting_citations":[],"review_version":1}