{"id":"ad9df4d7-b872-4efd-a614-53ff054e0b1f","arxiv_id":"2607.28660","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A framework using finitely additive normalized means extends Souriau geometric thermodynamics to infinite-dimensional Hamiltonian systems, with conditional applications to fluid and field equations.","lead":"This paper replaces ordinary probability measures with 'normalized means' to build equilibrium thermodynamics on infinite-dimensional Hamiltonian systems. It proves a general existence and uniqueness theorem and applies it conditionally to fluid, wave, and field-theory examples.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"In the Gaussian infinite-dimensional examples, the unrenormalized partition function vanishes for β>0, so the central admissibility hypothesis of Theorem 2.10 is not satisfied; only a renormalized theory—not developed—could apply.","rationale":"The reader correctly identified that translation compatibility is unverified in the infinite-dimensional examples. However, the more pressing issue is that the unrenormalized partition function vanishes in those examples, so the central theorem's admissibility domain D is empty. The paper's own appendix demonstrates this divergence and pivots to relative normalization, but the abstract framework is never extended to cover renormalized partition functions. This does not contradict the conditional correctness of Theorem 2.10, but it means the intended application to infinite-dimensional Hamiltonian systems is not demonstrated. The reader's verdict of CONDITIONAL remains appropriate: the paper is a conditional contribution whose hypotheses need verification in a concrete infinite-dimensional model. My concern is more fundamental than translation compatibility, but it supports the same verdict, so no change is needed.","tokens_in":30219,"tokens_out":7616,"duration_ms":69084,"concrete_test":"Compute m0(e^{-βH}) for the Example 4.2 current-group Gaussian reference using the normalized cut-off mean of Example 1.5: m0(f)=lim_{N→∞} (∫ f dμ_N)/(∫ dμ_N). For F=−βH, the cut-off partition functions are Z_N(β,0)=(1+β)^{-(2N+1)/2}. If the limit is 0 for β>0, then F∉E(m0) by Definition 1.10, and D is empty. To be conclusive, also test whether any admissible β>0 exists after including a nonzero linear constraint λΦ; the appendix's relative formula log Z_rel(β,λ)= (2πλ)^2/[2(1+β)] suggests the unrenormalized mean of e^{-βH−λΦ} is still 0 for β>0, so no point in D.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Theorem 2.10 requires (β,λ)∈D, where D is defined in Definition 2.3 by 0<m0(e^{Fβ,λ})<∞. In the only explicit infinite-dimensional Gaussian computations (Appendix A, Example 4.2), the cut-off partition function at λ=0 is Z_N(β,0)=(1+β)^{-(2N+1)/2}. As N→∞, this tends to 0 for every β>0. Therefore the normalized mean defined by the cut-off limit gives m0(e^{-βH})=0, so F=−βH is not admissible. Consequently D is empty for β>0, and Theorem 2.10 has no instance in these examples. The appendix acknowledges this: 'Only the resulting relative or renormalized free energy and its derivatives enter the corresponding renormalized variational principle.' But no renormalized variational principle is stated or proved. Thus the central claim—existence and uniqueness of equilibrium means under the stated assumptions—is not instantiated in any genuinely infinite-dimensional, non-σ-additive example. This is more fundamental than the unverified translation compatibility highlighted in the reader's verdict: even if translation compatibility held, the strict positivity of the partition function already fails.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a geometric-analytic framework for equilibrium statistical mechanics on infinite-dimensional Hamiltonian systems, replacing σ-additive probability measures with normalized means (positive normalized linear functionals on a vector lattice algebra L). It defines the relative entropy H(n∥m0) via a Donsker–Varadhan-type supremum over exponentially admissible potentials, introduces exponential tilts, and, under admissibility, translation compatibility, and separation hypotheses, proves a Gibbs variational principle with explicit exponential-family minimizer (Theorem 2.10). It also proves equivariance properties (Theorem 1.17), stationarity under Hamiltonian flows preserving the reference mean and constraints (Theorem 2.12), a classical Poisson–KMS identity (Theorem 2.17), smoothness/convexity of the log-partition functional under local exponential regularity (Proposition 1.20), and a Legendre–Fenchel dual entropy (Propositions 1.27–1.28). Applications are sketched for H^s geodesic equations on current groups and diffeomorphism groups, including Camassa–Holm/EPDiff and 2D Euler relations, with the repeated caveat that the abstract hypotheses are assumed to hold.","tokens_in":30572,"tokens_out":6186,"duration_ms":58222,"significance":"If the framework were instantiated in a genuinely infinite-dimensional, non-σ-additive setting, it would extend Souriau's Gibbs formalism beyond Gaussian or σ-additive reference measures and provide a common language for equilibrium states of Hamiltonian PDEs. The algebraic core of the paper is clean: the change-of-reference identity, the covariance Hessian computation, and the Poisson–KMS identity are proved by explicit finite-order estimates, with no uncontrolled interchange of limits. The paper is also honest in flagging its conditional assumptions. The obstacle is that no such infinite-dimensional instantiation is actually provided: the only explicit Gaussian computations, in Appendix A, show that the unrenormalized partition function for the energy tilt vanishes in the infinite-dimensional limit, so the central admissibility hypothesis 0<m0(e^{-βH})<∞ fails. Consequently the main theorem is not known to have any non-σ-additive infinite-dimensional instance, and the claimed existence/uniqueness result is at present a conditional algebraic statement.","major_comments":[{"comment":"The only explicit infinite-dimensional Gaussian calculation violates the central admissibility condition. The cut-off log-partition for the energy tilt is log Z_N(β,0)=-(2N+1)/2 log(1+β), so Z_N(β,0)→0 for every β>0; hence m0(e^{-βH})=0 and F_{β,0}∉E(m0). Thus D in Definition 2.3 is empty for the energy tilt in this model, and Theorem 2.10 has no instance there. The appendix's statement that 'only the resulting relative or renormalized free energy... enter the corresponding renormalized variational principle' acknowledges the difficulty, but no renormalized variational principle is stated or proved. A revision must either supply a genuinely admissible example or develop a renormalized version of Theorem 2.10.","section":"Appendix A (Camassa–Holm/EPDiff Gaussian computation)"},{"comment":"Translation compatibility E(m_F)+F=E(m_0) is never verified in any example. The applications repeatedly say 'whenever F belongs to E(m0) and satisfies the translation-compatibility and separation hypotheses' without checking them. Since this condition is used in the change-of-reference identity (Theorem 2.7) to conclude that the exponential tilt is the global minimizer, the main theorem remains conditional in every concrete case. At least one nontrivial class of means and potentials satisfying translation compatibility is needed, or the theorem should be presented purely as an implication with no claimed instances.","section":"Definition 2.6 / Theorem 2.10 / Sections 3–4"},{"comment":"The separation and local-perturbation hypotheses are ad hoc and unverified. The proof requires a subspace T separating normalized means and differentiability of t↦log m0(e^{t h}) at t=0 with derivative m0(h). For finitely additive means this differentiation is not automatic, and the manuscript does not give a criterion by which to verify it. Since uniqueness in Proposition 2.9 and Theorem 2.10 relies on H(n∥m*)=0 ⇒ n=m*, the uniqueness assertion is also conditional on an unverified analytic assumption.","section":"Proposition 1.13 / Theorem 2.10"},{"comment":"The admissible domain D is defined by requiring 0<m0(e^{F_{β,λ}})<∞, so part of the 'existence' conclusion is assumed rather than derived. The theorem constructs the tilt from an assumed finite positive partition function; it does not prove that such a partition function exists for any given system. The paper is often careful about this, but the abstract and the statement of Theorem 2.10 say 'existence and uniqueness' without sufficiently emphasizing that existence is contingent on admissibility. The central claim should be restated as a conditional: if F_{β,λ}∈E(m0) and translation compatibility holds, then the tilt is the unique minimizer.","section":"Definition 2.3 / Theorem 2.10 / Abstract"}],"minor_comments":[{"comment":"Several LaTeX artifacts remain: 'F r´echet', 'smallers', 'ordersof', and missing spaces in the references. A careful proofreading pass is needed.","section":"General typography"},{"comment":"Definition 1.19 refers to the parameter domain D before D is formally defined in Definition 2.3; add a forward pointer or move the definition earlier.","section":"Definition 1.19 vs Definition 2.3"},{"comment":"The proof says 'Since the admissible potentials separate normalized means on L', but the separation assumption is stated only for a subspace T⊂L. Clarify that T⊂E(m0), or extend the separation assumption explicitly to all admissible potentials.","section":"Proof of Proposition 1.13"},{"comment":"The sentence 'all β>0 are admissible at that level' could mislead, since the infinite-dimensional limit requires relative normalization. Please distinguish finite-cut-off admissibility from infinite-dimensional admissibility in D.","section":"Appendix D, normalizability summary"},{"comment":"Reference [19] appears incomplete ('A classification of coadjoint orbits carrying Gibbs ensembles' lacks a period/journal) and [14] is a self-citation to a 2026 Annals of Physics article; verify bibliographic data.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The main risk is the empty-instance problem: the only explicit infinite-dimensional Gaussian computations show the partition function vanishes, so the main theorem is not instantiated in a non-σ-additive infinite-dimensional setting. A revision should either produce a concrete admissible example (possibly non-Gaussian) or explicitly reframe the paper as an abstract conditional framework plus a renormalization program, with the main theorem and abstract adjusted accordingly. The algebraic identities are correct and worth publishing if the gap is addressed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague—\n\nThis paper is a coherent and honestly written attempt to extend Souriau's geometric thermodynamics to infinite-dimensional Hamiltonian systems using normalized means instead of σ-additive measures. What's genuinely new is the packaging: normalized means on vector lattice algebras, a Donsker–Varadhan style relative entropy, exponential families, and a classical Poisson–KMS identity, all on weak symplectic Fréchet manifolds. The writing is careful, the definitions are precise, and the conditional proofs (Theorems 1.17, 2.10, 2.17) check out under the stated assumptions. I believe the author is not hiding anything: the paper repeatedly says \"whenever\" and the appendix openly notes that renormalization is required.\n\nBut the soft spot is load-bearing and bigger than the reader's summary suggests. The reader flags translation compatibility as the fragile assumption—correctly—but there is an even more basic failure in the only explicit infinite-dimensional examples. The admissibility domain D requires 0 < m0(e^{Fβ,λ}) < ∞. In the Gaussian cut-off computations of Appendix A, the finite-dimensional partition function is (1+β)^-(2N+1)/2 up to a linear factor, which tends to 0 for every β>0. So the normalized mean limit gives m0(e^{-βH})=0, Fβ,λ is not admissible, and D is empty for β>0. The appendix acknowledges this and proposes a \"renormalized variational principle\" based on relative normalization, but no such principle is stated or proved anywhere. Thus Theorem 2.10, the paper's central existence/uniqueness claim, has no verified infinite-dimensional instance. The examples in Sections 3–4 are all conditional on hypotheses that are never verified, and in the Gaussian case those hypotheses are actually false.\n\nThe finite-dimensional Euler top example works, but it reduces to standard Souriau theory. So what survives is a formal framework: if one ever finds a non-σ-additive reference mean and a potential satisfying translation compatibility, separation, and strict positivity of the partition function, then the Gibbs formula gives the unique minimizer. That is more or less true by construction—the minimizer is defined to be the exponential tilt, and the variational identity is essentially the change-of-reference formula. It is a useful organizing language, not a theorem about any concrete infinite-dimensional system.\n\nIs it worth refereeing? Yes. The framework is novel enough and the exposition good enough that a serious referee could push the author to either produce a non-empty infinite-dimensional example with all hypotheses verified or to recast the paper as a formal algebraic-geometric theory. The appendix's renorm hint should be developed into a theorem. The citation pattern is fine—self-citations to earlier work on normalized means are legitimate.\n\nFor your own work, I wouldn't cite it as a proof source, but it may be worth reading for the Poisson–KMS formulation.","headline":"A careful conditional framework for Souriau thermodynamics via normalized means, but its central theorem has no verified infinite-dimensional instance—the Gaussian examples make the admissibility domain empty for β>0.","tokens_in":30946,"tokens_out":5307,"would_cite":false,"duration_ms":47646,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["37K05","53D20","82B10","46L55","58D05","37L50","82C31","46A03"],"pacs":[],"model":"deepseek-v4-flash","headline":"Replacing probability measures by normalized means, the paper extends equilibrium statistical mechanics to infinite-dimensional Hamiltonian systems and proves existence and uniqueness of exponential-family equilibrium states.","keywords":["normalized means","equilibrium statistical mechanics","infinite-dimensional Hamiltonian systems","weak symplectic Fréchet manifolds","exponential families","relative entropy","entropy geometry","hydrodynamic models"],"falsifier":"Exhibit an infinite-dimensional Hamiltonian system where, after tilting the reference mean by the Hamiltonian, the admissible potentials for the tilted mean are not exactly the original admissible potentials shifted by that Hamiltonian (translation compatibility fails), or where the finite-dimensional cut-off partition functions converge to zero; either would invalidate the uniqueness theorem.","tokens_in":30108,"feed_emoji":"⚖️","tokens_out":10002,"duration_ms":71497,"temperature":0.7,"pith_summary":"The paper's central claim is that equilibrium statistical mechanics can be carried over to infinite-dimensional Hamiltonian systems even when no probability measure exists to play the role of a reference state. The device is the normalized mean, a positive normalized linear functional on an algebra of observables that generalizes probability measures and invariant means. Given such a reference, the paper defines relative entropy and free energy, and shows that the free energy has a unique minimizer given by an explicit exponential tilt of the reference. Under a local exponential regularity condition, the log-partition function is smooth and convex, its Hessian is the covariance of the observables, and its Legendre–Fenchel dual is a concave entropy. If correct, this gives a rigorous equilibrium formalism for infinite-dimensional Hamiltonian PDEs and hydrodynamic models.","feed_headline":"Equilibrium states exist in infinite dimension, no probability measures","feed_subtitle":"Replacing probability measures by normalized means makes equilibrium states well-defined for Hamiltonian PDEs.","key_machinery":"The central mechanism is the exponential tilt of a normalized mean: for an admissible potential F, the tilted mean m_F(f) = m0(f e^F)/m0(e^F). The argument rests on a change-of-reference identity H(n∥m_F) = H(n∥m0) − n(F) + log m0(e^F), which is valid precisely when the admissible classes satisfy translation compatibility, i.e. E(m_F)+F = E(m_0). This identity converts the variational problem into a relative-entropy minimization and yields the unique minimizer; a local exponential regularity condition then justifies differentiating the partition function and identifies the Hessian with the covariance form.","core_discovery":"On a weak symplectic Fréchet manifold equipped with a normalized mean m0, the paper defines the free energy G_{β,λ}(n) = H(n∥m0) + β n(H) + Σ λ_a n(Φ_a) and proves (Theorem 2.10) that, under translation compatibility and separation hypotheses, it has the unique minimizer n*_{β,λ}(f) = m0(f e^{−βH−Σλ_aΦ_a}) / m0(e^{−βH−Σλ_aΦ_a}). The logarithmic partition function ψ = log m0(e^{−βH−Σλ_aΦ_a}) is then smooth, with Hessian equal to the covariance form of the Hamiltonian and constraints; the Legendre–Fenchel dual of ψ is a concave entropy on extensive variables. The construction is presented as the infinite-dimensional analogue of the classical geometric thermodynamics of coadjoint orbits, with n","pith_inferences":["If translation compatibility can be verified for concrete Hamiltonian PDEs, the normalized-mean equilibrium states would supply rigorous stationary statistical states for systems where no invariant probability measure is currently known.","The covariance Hessian defines a metric on the parameter space, suggesting an information-geometric notion of distance between equilibria that could be used to study thermodynamic processes in infinite dimension.","The classical Poisson–KMS identity could serve as a selection criterion identifying equilibrium states among all stationary normalized means, paralleling the role of KMS states in algebraic quantum statistical mechanics.","The appendix's renormalized determinants suggest a concrete recipe for constructing reference means: relative normalization or second-order determinant renormalization of cut-off partition functions may yield admissible normalized means even when the covariance perturbation is only Hilbert–Schmidt."],"forward_implications":["Equilibrium states exist and are unique for infinite-dimensional Hamiltonian systems without assuming a σ-additive reference measure, whenever the admissibility and translation-compatibility conditions hold.","The log-partition function is smooth and convex, and its Hessian is the covariance form; this yields strict convexity modulo thermodynamically null directions and a metric structure on the reduced parameter space.","The Legendre–Fenchel dual of the log-partition is a concave entropy that agrees with the constrained variational entropy on the equilibrium image, so intensive/extensive duality survives in infinite dimension.","Equilibrium means are stationary under Hamiltonian flows that preserve the reference mean and the constraints, and satisfy a classical Poisson–KMS identity when the reference mean is Poisson invariant.","The framework covers geodesic equations on current groups and diffeomorphism groups, including hydrodynamic and field-theoretic examples, with exponential-family equilibria given by explicit tilts."],"fun_headline_variants":["Infinite dimension, no probability measure: equilibrium exists","Replace probability measures with normalized means for equilibrium","Hamiltonian systems get equilibrium via normalized means, not measures","Infinite-dimensional equilibrium: skip the measure, use a mean","Normalized means replace measures for entropy in Hamiltonian systems"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The construction collapses if the class of admissible exponential weights does not shift exactly by the tilt potential after reweighting (translation compatibility); the paper assumes this in every infinite-dimensional example without verifying it.","fun_headline_variants_meta":{"raw":{"variants":["Infinite dimension, no probability measure: equilibrium exists","Replace probability measures with normalized means for equilibrium","Hamiltonian systems get equilibrium via normalized means, not measures","Infinite-dimensional equilibrium: skip the measure, use a mean","Normalized means replace measures for entropy in Hamiltonian systems"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000533,"raw_usage":{"total_tokens":2415,"prompt_tokens":775,"completion_tokens":1640,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":519,"completion_tokens_details":{"reasoning_tokens":1577}},"tokens_in":519,"tokens_out":1640,"duration_ms":12386,"temperature":1.0,"reasoning_tokens":1577,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T00:44:48.708797+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Exhibit an infinite-dimensional Hamiltonian system where, after tilting the reference mean by the Hamiltonian, the admissible potentials for the tilted mean are not exactly the original admissible potentials shifted by that Hamiltonian (translation compatibility fails), or where the finite-dimensional cut-off partition functions converge to zero; either would invalidate the uniqueness theorem.","supporting_citations":[],"review_version":1}