{"id":"010a74c7-fad3-4e17-97ac-505c56685182","arxiv_id":"2505.08852","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A class of continuous-time autoregressive moving-average processes taking values in cones of measures is constructed via analytically weak solutions of linear state-space equations driven by Lévy subordinators.","lead":"Researchers extend CARMA time-series models so the modeled quantity is a measure over space rather than a single number, driven by jump processes. The paper proves these processes exist, stay positive, and gives moment and stationarity formulas, aimed at energy-market and weather-derivative modeling.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The proposed parameter examples cannot jointly meet the cone-invariance and stationarity hypotheses for p≥2; the advertised family needs explicit feasible parameters.","rationale":"The reader's weakest_assumption correctly identifies the operator-level conditions: quasi-positivity of Ap and complete monotonicity of the transfer function. I agree with that diagnosis but make it sharper: the examples offered are not merely unverified; they contradict one of the two requirements for p≥2. In the convolution construction all blocks are positive and no stabilizing negative term appears, so a stationary kernel cannot decay; in the scalar construction the negative coefficients needed for stability are exactly the blocks that violate positivity of e^{tAp} on the product cone. Since §3.4.2 advertises recovering classical CARMA(p,q) and §3.4.3 builds multi-parameter fields from these processes, the claimed family of measure-valued CARMA processes is not supported beyond the trivial p=1 OU-type case. Definition 2.2's quasi-monotonicity condition is additionally vacuous as written, so the proof of cone-invariance cannot currently be checked from the stated assumptions. These issues do not invalidate the analytic weak solution framework or the Laplace-transform computations if feasible parameters are found, but they must be corrected before the stationarity and cone-invariance claims are accepted. Hence the reader's CONDITIONAL verdict remains appropriate; my stress-test does not move it.","tokens_in":30888,"tokens_out":22718,"duration_ms":254952,"concrete_test":"Take p=2, B=L1(E) with the cone L1_+(E)^2, and scalar coefficients as in §3.4.2: A1=-a1 I, A2=-a2 I, a1,a2>0, Ep injecting into the last coordinate, Cq=(c0 I,c1 I). Compute the semigroup action on the boundary element (α,0): the Taylor expansion e^{tA_p}(α,0) = (α, -t a2 α) + O(t^2). Since -t a2 α∉L1_+, the semigroup is not positive; simultaneously, the characteristic polynomial λ^2+a1λ+a2 is Hurwitz precisely when a1,a2>0, so the stationarity side is satisfied while the cone-invariance side fails. Repeating for a2≤0 to make the (2,1) block nonnegative shows e^{tA_p} can be positive but the polynomial is not Hurwitz, so no stationary positive solution. This verifies that the claimed parameter family is empty for p=2; extending to p>2 gives the same sign conflict for the entries -a_p,...,-a_2.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Proposition 2.5 makes stationarity depend on complete monotonicity of Q(λ)P(λ)^{-1}, and Definition 3.2 makes cone-invariance depend on quasi-positivity of Ap. The manuscript never exhibits a non-degenerate tuple satisfying both. Worse, its own examples are incompatible. In §3.3.1 the convolution blocks Ai are positive operators; then Ap is a nonnegative companion matrix with no stabilizing negative diagonal, its spectral bound is not negative, so the stationary representation (2.28) cannot exist. In §3.4.2 the authors take Ai = -ai I with ai > 0. For p=2, the (2,1) block of Ap is -a2 I. Applying e^{tAp} to (α,0) with α∈L1_+\\{0} gives second component -t a2 α for small t, outside L1_+; hence e^{tAp} is not positive on the product cone. The Hurwitz condition for stationarity requires a2>0, so making the block nonnegative by a2≤0 destroys stationarity. Thus there are no stationary cone-valued CARMA(2,q) examples with these coefficients, and the p≥2 generalization advertised in §3.4.2–3.4.3 lacks a realized parameter set. This is not a missing computation: the positivity and stationarity constraints are mutually exclusive for this family. Cone-invariance is also not rescued by Definition 2.2, since as written the quasi-monotonicity condition is vacuous: ⟨f,x⟩=⟨f,y⟩ for all f∈K* forces x=y, so the inequality is only checked on equal elements.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a Banach-space framework for Lévy-driven CARMA processes with values in convex cones, with primary focus on measure-valued processes. It defines the state process as an analytically weak solution of a linear state-space model driven by a Lévy subordinator, proves existence and cone-invariance under quasi-positivity conditions, derives Laplace transforms and moment formulas, and gives stationarity conditions via complete monotonicity of the operator-valued transfer function Q(λ)P(λ)^{-1}. The authors then specialize to the space L1(E) of absolutely continuous finite signed measures and to M+(E) with finite-dimensional noise, calling the output a pure-jump measure-valued CARMA(p,q) process, and discuss applications to renewable energy and flow forwards.","tokens_in":31134,"tokens_out":19383,"duration_ms":217780,"significance":"If the technical gaps were repaired, the framework would be a useful contribution: it provides explicit variation-of-constants representations, Laplace transforms, moment formulas, and a clean link between measure-valued CARMA processes and ambit fields. The analytical-weak-solution approach for cone-valued state processes driven by subordinators is a natural and potentially generalizable idea, and the explicit formulas in Propositions 2.4, 3.1, and 4.1 would be valuable for applications. However, the impact is currently limited by several load-bearing issues: the quasi-monotonicity definition as stated is vacuous, the duality statement for M(E) is reversed, and the paper does not exhibit a feasible stationary cone-valued parameter set for p≥2. These issues must be resolved before the advertised class of stationary positive measure-valued CARMA processes can be considered established.","major_comments":[{"comment":"As written, the quasi-monotonicity condition is vacuous for the cones considered in the paper. The definition requires that for all x,y in the domain, if x≤K y and ⟨f,x⟩=⟨f,y⟩ for all f∈K*, then ⟨f,Ax⟩≤⟨f,Ay⟩ for all f∈K*. For K=L1_+(E), the premises force x=y, so the implication carries no information. Consequently, the citation to [32, Theorem 1] in support of St(K^p)⊆K^p is not justified, and the cone-invariance asserted in Proposition 2.1, Proposition 2.3(i), and Definition 3.2 is not established. The standard formulation, in which one requires f(Ax)≤f(Ay) for individual f∈K* satisfying f(x)=f(y), should replace the current statement.","section":"Section 2.2, Definition 2.2"},{"comment":"The statement “C0(E) is the dual of M(E)” reverses the Riesz–Markov–Kakutani theorem: for a locally compact Polish space E, C0(E)* is isometrically isomorphic to M(E), while the dual of M(E) under the total variation norm is not C0(E). This error matters because Proposition 3.2 computes Laplace transforms and moments only for g∈C0(E), whereas the appropriate dual cone in the absolutely continuous case is L∞_+(E). The weak-solution and Laplace-transform claims are therefore established only for a proper subspace of admissible test functions, not for the full dual cone needed in Definition 3.1 and Proposition 3.1.","section":"Section 3.1"},{"comment":"The manuscript never exhibits a non-degenerate parameter tuple satisfying both cone-invariance (Definition 3.2) and stationarity (Proposition 2.5) for p≥2, and the two concrete families are mutually incompatible. In Section 3.3.1 the convolution blocks Ai and Cj are positive, so Ap has a nonnegative spectral bound; the stationary representation (2.28) cannot converge and the transfer Q(λ)P(λ)^{-1} cannot be completely monotone in the required sense. In Section 3.4.2 the choice Ai=−aiI with ai>0 gives, for p=2, the block (Ap)_{2,1}=−a2I; applying the semigroup to (α,0) with α∈L1_+\\{0} yields second component −t a2 α+O(t^2) for small t, which is outside the positive cone. Setting a2≤0 to restore quasi-positivity destroys Hurwitz stability. The paper must either provide feasible parameters for stationary cone-valued CARMA(p,q) with p≥2 or explicitly separate the stationarity result for output positivity from the state-cone-invariance requirement in Definition 3.2.","section":"Sections 3.3.1 and 3.4.2; Proposition 2.5"},{"comment":"The proof of the stationary representation cites exponential bounds ‖K(±u)‖≤η e^{−w(S)|u|} from [23], but the standing assumption (2.27) only excludes singularities on the imaginary axis; it does not imply that the spectral bound of Ap is negative, nor that the kernel is exponentially integrable. Thus the claimed “if and only if” between complete monotonicity of Q(λ)P(λ)^{-1} and the existence of a stationary K-valued solution is not fully demonstrated. In particular, the proof should show that complete monotonicity, together with the rational decay of the transfer function, forces the poles into the open left half-plane and yields the required integrability of the kernel.","section":"Proposition 2.5"},{"comment":"The covariance formulas in Proposition 2.6 are inconsistent with the driving noise used in the paper. Theorem 2.1 and Proposition 2.1 concern pure-jump subordinators with Q=0, whereas Proposition 2.6 assumes a square-integrable Lévy process with covariance operator Q and refers to Proposition 2.1 for its existence. For a jump-driven process the conditional covariance of Yt given Fs is not simply CqΣ_{t,s}Cq* with Σ built from a Gaussian covariance operator; the jump part contributes an additional integral of squares with respect to the Lévy measure, as Proposition 2.4 itself shows. Moreover, in a general Banach space such as L1(E), the existence of a covariance operator in this form is not automatic. The proposition should be restated under assumptions that are compatible with the pure-jump setting, or its formulas should be corrected.","section":"Proposition 2.6"}],"minor_comments":[{"comment":"The sentence defining the measure-valued CARMA process contains a duplicated article: “a a pure-jump measure-valued CARMA(p,q) process.”","section":"Definition 3.2"},{"comment":"The displayed Laplace transform in Proposition 2.2(i) has unbalanced parentheses and a misplaced “du” after the second exponential; the plus sign before “exp” also appears inconsistent with the product formula derived in the preceding lines.","section":"Equation (2.12)"},{"comment":"The variance formula is missing the time integral: as printed, Var[⟨g,Y_t⟩] is time-independent and does not match Proposition 2.4, which contains ∫0^t∫K ... ds.","section":"Proposition 3.2(iii)"},{"comment":"In the Laplace transform for the L1_+(Rd)-valued subordinator, the displayed expression contains e^{⟨g,uφ(·−y)⟩} where the exponent should be −⟨g,uφ(·−y)⟩ for a Laplace transform with g∈L∞_+; as printed the expression is not a Laplace transform.","section":"Example 3.3.2"},{"comment":"The notation A:D(A)⊂L1_+(E)^p→L1_+(E)^p is problematic for a linear operator: a linear generator has a domain that is a subspace of L1(E)^p, not a subset of the positive cone, unless the operator is trivial.","section":"Definition 3.1"}],"recommendation":"major_revision","confidential_remarks":"The paper contains a plausible central construction and several useful explicit formulas, but the revision must resolve the feasibility gap for stationary cone-valued p≥2 examples and correct the foundational errors in Definition 2.2 and Section 3.1. If no stationary positive parameter sets exist for p≥2, the authors should substantially revise the claims in the abstract and Section 3.4 rather than leaving the current contradictory examples."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThe core of this paper is a real extension: measure-valued CARMA processes built from analytically weak solutions of linear state-space equations in L1 or in finite measures, with explicit Laplace transforms, moment formulas, a stationarity criterion in terms of complete monotonicity of an operator-valued transfer function, and an Esscher transform. That is worth taking seriously. The L1-based theory is mostly coherent, and the authors are honest about the non-separability of M(E) and about using finite-dimensional noise there.\n\nThe soft spots are concentrated in the examples and in a few definitions. Section 3.1 states that C0(E) is the dual of M(E); the duality runs the other way, C0(E)* = M(E), and the dual of M(E) under total variation is strictly larger. Since the paper then moves to L1, this is not fatal, but it should be corrected. Definition 2.2 is vacuous as written: requiring equality under all f in K* forces x=y, so the quasi-monotonicity condition is never tested. The intended condition is presumably the standard one with a single f and f(x-y)=0; as it stands, Proposition 2.3(i) does not actually establish cone invariance.\n\nThe bigger problem is the parameter sets. In §3.3.1 all Ai are positive convolution operators; then Ap is a nonnegative companion matrix with no negative feedback, its spectral bound is not negative, and the stationary representation (2.28) cannot converge. In §3.4.2 the authors take Ai = -ai I with ai > 0. For p=2 the (2,1) block is -a2 I, so applying e^{tAp} to (α,0) produces a negative second component for small t; the semigroup is not positive on the product cone. Demanding a2 ≤ 0 for positivity makes the spectral bound nonnegative, so stationarity is lost. Thus the paper contains no non-degenerate p≥2 tuple satisfying both cone-invariance and stationarity, and the advertised CARMA(2,q) examples do not fall under the theory as stated. This is a genuine gap, not a missing computation. There are also typos in (2.12), Proposition 3.2(iii), and Proposition 4.2, and the covariance operator Q in Proposition 2.6 is ambiguous for a pure-jump subordinator.\n\nWho should read it: people building infinite-dimensional CARMA or measure-valued affine models, and energy/weather modelers. The ideas are useful, but the current examples overstate what is established. It deserves a serious referee, with the expectation of substantial revision: fix the definitions, correct the dual-space claim, and either exhibit feasible p≥2 parameters or state clearly that the method currently yields only p=1-type examples. I would not desk-reject it.","headline":"A genuinely new measure-valued CARMA framework with a coherent L1 core, but the paper's own parameter examples fail the cone-invariance and stationarity assumptions for p≥2 and the current draft needs major revision.","tokens_in":31721,"tokens_out":8982,"would_cite":false,"duration_ms":97293,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60G51","60G57","60G10","60H15","47D06"],"pacs":[],"model":"deepseek-v4-flash","headline":"Measure-valued CARMA processes can be defined as positive, jump-driven, stationary measure flows with explicit Laplace transforms.","keywords":["measure-valued CARMA","Lévy subordinator","linear state-space model","cone invariance","stationarity","Pettis integral","operator-valued transfer function","spatio-temporal random fields"],"falsifier":"Take Section 3.3.1's convolution example with p=2, q=0, C0=I, pick two nonnegative convolution kernels, and compute the kernel K(t) by inverting the Fourier integral in (2.29); if K(t) is not a positive operator for some t while all the paper's stated positivity hypotheses hold, Proposition 2.5's sufficiency claim is false.","tokens_in":30630,"feed_emoji":"⚡","tokens_out":10984,"duration_ms":105518,"temperature":0.7,"pith_summary":"CARMA processes are continuous-time analogues of ARMA time series, and this paper moves them from real or vector values to measures: the state is a density or measure on a spatial domain, and the dynamics are a linear state-space equation driven by a Lévy subordinator. The paper shows that, when the companion operator and output operators preserve positivity, the solution exists as an analytically weak process, stays in the cone of positive measures, and has closed-form Laplace transforms, moments, and autocovariance. It gives a stationarity condition in terms of complete monotonicity of an operator-valued transfer function, and exhibits concrete parameter sets including convolution operators and Poisson-type noise. The point is a tractable jump-driven, cone-valued model for spatio-temporal aggregates such as regional renewable production, flow-forward prices, and power purchase agreements.","feed_headline":"Measure-valued CARMA processes proven to exist and stay positive","feed_subtitle":"Jump-driven measure flows with explicit Laplace transforms for spatio-temporal aggregates like renewable production.","key_machinery":"The central object is the companion block operator matrix A_p, built from the operator polynomial P(λ)=Iλ^p−$A_1λ^{{p−1}}$−...−A_p, together with the output polynomial Q(λ)=C_0+C_1λ+...+C_qλ^q. The transfer function λ ↦ Q(λ)P(λ)^{-1} carries the argument: its complete monotonicity with respect to the cone π(K) of positive operators is exactly the condition that makes the stationary kernel K(t) positive, so that the convolution against a Lévy subordinator stays inside the cone. Existence is handled through Pettis stochastic integrals, a weak notion of stochastic integration valid in general Banach spaces, and through quasi-positive semigroups that preserve the cone of positive measures.","core_discovery":"The paper's central claim is that a pure-jump measure-valued CARMA(p,q) process can be defined as the analytically weak solution of a linear state-space model in a Banach space of measures: dX_t = A_p X_t dt + E_p dL_t, Y_t = C_q X_t, with A_p the companion block operator matrix of P(λ)=Iλ^p−$A_1λ^{{p−1}}$−...−A_p, E_p injecting noise into the last coordinate, and C_q the output operator built from C_0,...,C_q. The process (t,A) ↦ ∫_A Y_t(x)λ(dx) is measure-valued, and under quasi-positivity of A_p, positivity of C_j and E, and a Lévy subordinator L, it exists and remains in the cone of nonnegative measures. The paper derives the Laplace transform of the transition semigroup, explicit first and second moments, conditional covariance and autocovariance, and proves that a unique stationary cone-valued version exists exactly when the operator-valued transfer function λ ↦ Q(λ)P(λ)^{-1} is completely monotone with respect to the positive-operator cone, in which case Y_t = ∫_{−∞}^{∞} K(t−s) dL_s with K given by a Fourier integral. It also connects the construction to classical CARMA, Hilbert-space CARMA, CARMA random fields, and ambit fields.","pith_inferences":["A natural extension the paper leaves implicit is that the Laplace-transform formalism would support characteristic-function estimation of measure-valued CARMA parameters from aggregate spatial observations, without observing the full measure field.","The complete-monotonicity condition is probably restrictive; even natural positive convolution kernels can be checked case by case, and failures would rule out stationarity (not positivity) for those parameter sets.","Because the existence proof needs a separable space, arbitrary finite signed measures are not covered; an integration theory for non-separable measure-valued Lévy processes would be needed to reach the full space of finite signed measures advertised in the title."],"forward_implications":["Under the positivity assumptions, every measure-valued CARMA(p,q) process remains in the cone of positive densities for all times, so it can model non-negative quantities like capacity factors and forward prices.","The Laplace transform of any spatial average ∫_A Y_t(x)λ(dx) has a closed form, so aggregate quantities inherit explicit exponential-affine dynamics.","First and second moments, conditional covariance, and autocovariance are given by explicit operator integrals, enabling moment-based statistical inference.","When the transfer function is completely monotone, the process has a unique stationary version Y_t=∫_{−∞}^{∞}K(t−s)dL_s, so all temporal dependence is captured by the kernel K.","Evaluating the process on indicator functions yields multi-parameter CARMA random fields, and evaluating it on spatial sets recovers classical real-valued CARMA processes."],"supporting_citations":[{"why":"It supplies the Pettis-integration and stochastic-integration theorems used to prove existence of the analytically weak solution and the variation-of-constants formula.","marker":"[40]"},{"why":"It supplies the cone-valued multivariate CARMA framework and the quasi-monotonicity arguments adapted in Proposition 2.3.","marker":"[5]"},{"why":"It gives the characterization of Lévy subordinators on cones used to define the driving noise.","marker":"[41]"},{"why":"It defines complete monotonicity for operator-valued functions and provides the Bernstein-type theorem used in the stationarity criterion.","marker":"[2]"},{"why":"It provides the semigroup spectral representation and growth bounds used to construct the stationary kernel K(t).","marker":"[23]"},{"why":"It establishes existence of Ornstein-Uhlenbeck-type stochastic integrals on Hilbert space, used for the infinite-horizon integrals in the stationary representation.","marker":"[17]"},{"why":"It defines the classical real-valued Lévy-driven CARMA process that the measure-valued construction generalizes.","marker":"[15]"},{"why":"It introduces CARMA processes in Hilbert space, the infinite-dimensional formulation this paper extends to Banach cones.","marker":"[10]"}],"fun_headline_variants":["Measure-valued CARMA processes exist and stay in positive cone","New proof: jump-driven measure flows keep positivity","Pure-jump CARMA on Banach spaces: existence and stationarity","Measure-valued CARMA: weak solutions with explicit moments","CARMA processes in measure spaces: cone-invariant dynamics"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the companion operator A_p generates a positivity-preserving semigroup and that the operator-valued transfer function λ↦Q(λ)P(λ)^{-1} is completely monotone in the positive-operator order, a condition the paper assumes abstractly rather than verifying for its concrete convolution examples.","fun_headline_variants_meta":{"raw":{"variants":["Measure-valued CARMA processes exist and stay in positive cone","New proof: jump-driven measure flows keep positivity","Pure-jump CARMA on Banach spaces: existence and stationarity","Measure-valued CARMA: weak solutions with explicit moments","CARMA processes in measure spaces: cone-invariant dynamics"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000605,"raw_usage":{"total_tokens":2838,"prompt_tokens":979,"completion_tokens":1859,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":595,"completion_tokens_details":{"reasoning_tokens":1778}},"tokens_in":595,"tokens_out":1859,"duration_ms":12030,"temperature":1.0,"reasoning_tokens":1778,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T21:47:51.862739+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take Section 3.3.1's convolution example with p=2, q=0, C0=I, pick two nonnegative convolution kernels, and compute the kernel K(t) by inverting the Fourier integral in (2.29); if K(t) is not a positive operator for some t while all the paper's stated positivity hypotheses hold, Proposition 2.5's sufficiency claim is false.","supporting_citations":[{"cited_title":"Stochastic integration for Lévy processes with values in Banach spaces","cited_arxiv_id":null,"evidence_quote":"It supplies the Pettis-integration and stochastic-integration theorems used to prove existence of the analytically weak solution and the variation-of-constants formula."},{"cited_title":"E., and Karbach, S","cited_arxiv_id":null,"evidence_quote":"It supplies the cone-valued multivariate CARMA framework and the quasi-monotonicity arguments adapted in Proposition 2.3."},{"cited_title":"Subordinators in a class of Banach spaces","cited_arxiv_id":null,"evidence_quote":"It gives the characterization of Lévy subordinators on cones used to define the driving noise."},{"cited_title":"Generators of positive semigroups and resolvent positive o perators","cited_arxiv_id":null,"evidence_quote":"It defines complete monotonicity for operator-valued functions and provides the Bernstein-type theorem used in the stationarity criterion."},{"cited_title":"J., and Nagel, R","cited_arxiv_id":null,"evidence_quote":"It provides the semigroup spectral representation and growth bounds used to construct the stationary kernel K(t)."},{"cited_title":"On processes of Ornstein-Uhlenbeck type in Hilbert space","cited_arxiv_id":null,"evidence_quote":"It establishes existence of Ornstein-Uhlenbeck-type stochastic integrals on Hilbert space, used for the infinite-horizon integrals in the stationary representation."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It defines the classical real-valued Lévy-driven CARMA process that the measure-valued construction generalizes."},{"cited_title":"E., and Süss, A","cited_arxiv_id":null,"evidence_quote":"It introduces CARMA processes in Hilbert space, the infinite-dimensional formulation this paper extends to Banach cones."}],"review_version":1}