{"id":"802a74bd-c6a2-4dab-b356-55e9897f438a","arxiv_id":"2505.11389","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Products of m Poisson multiple integrals are square-integrable exactly when iterated add-one cost expectations lie in L2, and their chaos kernels are explicit partition sums.","lead":"This paper gives a complete answer to when the product of several random integrals driven by a Poisson point process has finite variance, and it writes the product as a sum of simpler random integrals. The result completes a line of research in stochastic analysis that began in 1984 and affects limit theorems for random geometric structures.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Proposition 4.11's diagram identity has an edge case: when a word exhausts all kernel variables, the inner sum is over Π≥2(0,...,0), which the paper defines as empty, so (4.12) omits the deterministic terms needed for h_q=(1.9).","rationale":"The reader's concern about Condition A-(loc) is legitimate but is a stated limitation, not a flaw. My stress-test found a more concrete issue: Proposition 4.11, the only proof of Part II's identification h_q=(1.9), is not just compressed; under the paper's explicit definition of Π≥2 for the all-zero vector it fails for words in which every variable of every kernel is fixed by the z's. This is exactly the m=2 example f_1(z)f_2(z), and analogous deterministic contributions arise for m≥3 whenever a word uses every variable of every kernel. The fix is a one-line convention, namely letting the empty partition contribute the constant product, and with that convention the identity is likely correct since the same convention is implicit in Theorem 4.5. Therefore this does not warrant rejection; it does warrant conditional acceptance with the convention and the direct-inspection step made explicit. No independent verification of Proposition 4.11 was provided in the paper, and the illustrative examples do not cover this edge case. Part I of Theorem 1.6 and Theorem 5.1 are supported by a coherent argument using Last–Penrose and the new p-Poincaré inequalities, so the necessary-and-sufficient criterion appears sound. I agree with the reader that Part II is the weak point, but for a more specific reason than the one stated.","tokens_in":20845,"tokens_out":17375,"duration_ms":168195,"concrete_test":"Specialize (4.12) and (1.9) to m=2, k_1=k_2=1, q=1, with symmetric f_1,f_2∈L^2(μ), and compute both sides using the paper's stated definition Π≥2(0,0)=∅. The left side of (4.12) evaluates to 0 while h_1=f_1f_2, demonstrating the missing empty-partition term. Repeat for m=3, k=(1,1,1), q=2, word W=({1,2},{3}): (4.12) again gives 0 for that word unless Π≥2(0,0,0) is taken to be {∅}. If the authors confirm the empty-partition convention, recompute the proof of Proposition 4.11 with that convention and verify (4.13) for these cases; if equality then holds, the theorem survives with a clarified convention.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Part II of Theorem 1.6 rests on Proposition 4.11, specifically the assertion that the word sum in (4.12) equals the Surgailis partition sum h_q in (1.9). The proof is only by 'direct inspection', and under the paper's own definitions the displayed identity is false for words W=(A_1,...,A_q) with d_i=k_i for every i. Section 4.2.1 defines Π≥2(k_1,...,k_m)=Π(k_1,...,k_m)=∅ when all k_i=0. Take m=2, k_1=k_2=1, q=1, W=({1,2}). Since D^W_z(I_1(f_1),I_1(f_2))=f_1(z)f_2(z), the left side of (4.11) contains f_1(z)f_2(z). In (4.12) the inner sum for this W is over Π≥2(0,0)=∅, hence contributes 0; the other words {1} and {2} also contribute 0. So (4.12) gives 0, while h_1(z)=f_1(z)f_2(z) from (1.9). The same omission arises for m=3, k=(1,1,1), q=2, whenever all three indices are assigned to the two blocks, e.g. W=({1,2},{3}). The intended repair is to adopt the convention, already used silently in Theorem 4.5's 'by definition' clause, that Π≥2(0,...,0) contains the empty partition and that the integral over Z^0 equals the constant product. As written, however, the bridge proving (1.9) has a concrete gap, so Part II is not fully proved on the stated definitions. Part I and the p-Poincaré inequalities are not affected by this edge case.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper establishes necessary and sufficient conditions for the square-integrability of products of multiple Wiener-Itô integrals with respect to a general Poisson random measure, and gives an explicit chaos decomposition under a local integrability condition. The central result, Theorem 1.6, has two parts: Part I characterizes $\\Phi \\in L^2(P)$ via integrability of iterated add-one cost operators, with the top-order chaos kernel identified as the symmetrized tensor product; Part II, under Condition A-(loc), identifies the lower chaos kernels with the partition sums (1.9) originally written by Surgailis for the integrable case. The proof combines a new family of $p$-Poincaré inequalities for almost surely finite variables (Theorem 3.2), a propagation result (Proposition 3.4), a word calculus for iterated add-one costs (Section 4), and the Last-Penrose formula. The paper is addressed to open problems in Surgailis (1984) and completes the $m=2$ work of Döbler-Peccati for $m \\geq 3$.","tokens_in":21189,"tokens_out":21907,"duration_ms":223722,"significance":"If the main results are correct, this is a substantial contribution to Poisson stochastic analysis. The paper gives a genuine necessary-and-sufficient criterion for square-integrability of products of Poisson multiple integrals, answers a long-standing question of Surgailis, and provides explicit diagram formulae under natural local integrability. The newly introduced $p$-Poincaré inequalities for non-integrable variables are of independent interest. The paper is also commendable for being largely self-contained, for importing only standard external results, and for providing directly checkable formulas in the single-integral cases of Example 1.8; these formulas serve as concrete falsifiable predictions. The main theorem is not obtained by fitting parameters, and the $m \\geq 3$ result does not reduce to the previously known $m=2$ case by construction.","major_comments":[{"comment":"The identity (4.13) is false on the stated definitions when a word W has d_i = k_i for every i. Indeed, Section 4.2.1 defines Π≥2(0,...,0) = ∅, so the inner sum in (4.12) contributes 0 for such a word, while the left-hand side contains a nonzero deterministic product. For example, take m=2, k1=k2=1, q=1, and W=({1,2}); then D^W_z(I1(f1),I1(f2)) = f1(z)f2(z), so the left-hand side of (4.12) equals f1(z)f2(z), but the inner sum over Π≥2(0,0) is empty and the right-hand side is 0. The same omission occurs for m=3, k=(1,1,1), q=2 with W=({1,2},{3}). The proof of Proposition 4.11 introduces the auxiliary set Π≥2(B(T)) and for empty B(T) implicitly uses the empty partition, which is inconsistent with the convention adopted for Π≥2(0,...,0). Since Part II of Theorem 1.6 and the explicit kernel formula (1.9) are proved through Proposition 4.11, Part II is not fully proved as written. The repair is local: adopt the convention that Π≥2(0,...,0) contains the empty partition and that the integral over Z^0 of the empty product equals the product of the residual constants; this convention is already used silently in Theorem 4.5. Part I and the p-Poincaré inequalities are not affected by this edge case.","section":"Section 4.2.1 and Proposition 4.11, Eqs. (4.12)-(4.13)"}],"minor_comments":[{"comment":"The sentence 'Iterating M times (3.1)' is too terse, because Theorem 3.1 is stated for integrable G and F itself is not assumed integrable in Proposition 3.4. The argument becomes complete if the iteration is applied to D^{(q)}F for q ≥ 1 and the zero-order base case is handled through Theorem 3.2, but this should be spelled out.","section":"Section 3, proof of Proposition 3.4"},{"comment":"Condition (ii) in Theorem 1.6 is stated for q=1,...,K-1, whereas Theorem 5.1's condition (ii-2) includes q=K. The text should explicitly note that the q=K condition is automatic when f_i ∈ L^2(µ^{k_i}) for all i, because h_K = sym(f1 ⊗ ... ⊗ fm) is automatically in L^2(µ^K); this would make the 'direct consequence' claim fully transparent.","section":"Theorem 1.6, Part I versus Theorem 5.1"},{"comment":"The asserted equivalence between Condition A-(loc) and the conditions (1.12)-(1.13) is described only as a 'standard exercise'. Since Part II of Theorem 1.6 is stated in terms of (1.12)-(1.13), a short proof or a precise reference would improve the paper's self-containedness.","section":"Definition 4.8 and Remark 4.9"},{"comment":"In item 3 the notation 'sym{...}' is used without explicitly indicating the variables (z1,z2) in the displayed formula; the meaning is clear from context, but the formula would be easier to parse if the argument were written out as in (1.1).","section":"Example 1.8"}],"recommendation":"major_revision","confidential_remarks":"The Part II gap identified above is localized and easily repairable by adopting the empty-partition convention that the authors already use implicitly elsewhere. The core ideas and Part I appear sound, and I would be willing to accept after a revision that fixes the statement and proof of Proposition 4.11 and adds the small clarifications listed as minor comments."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The key thing to know: this paper actually moves the Surgailis questions forward. For m≥3 it gives necessary and sufficient L² conditions for products of multiple Poisson integrals, and an explicit chaos formula under a local integrability condition. The m=2 case was Döbler–Peccati; the m≥3 cases were open. That is a real result, not an incremental replay.\n\nThe paper does several things well. Theorem 3.2 extends Trauthwein's p-Poincaré inequality to almost surely finite (not necessarily integrable) variables, and the truncation-plus-Fatou proof is sound. Theorem 5.1 is a clean general criterion for products of finite-chaos functionals, and Corollary 5.3 settles h_K = sym(f1⊗...⊗fm) under square-integrability. Part I of Theorem 1.6 is directly proved from the Last–Penrose formula and the Poincaré tool; as far as I can tell it holds. The paper also has no fitted parameters, no circularity, and the citation practice is fair: the m=2 result is credited, Trauthwein is external, and the remaining imports are standard.\n\nThe main soft spot is in Proposition 4.11, which is the bridge to Part II. The stress-test note is right. When a word W has d_i = k_i for every i, the residual vector (k_1-d_1,...,k_m-d_m) is all zeros, and Section 4.2.1 defines Π≥2(0,...,0) to be empty. Then the inner sum in (4.12) contributes zero, even though the true expectation of that word's term is a nonzero product of constants (for example, m=2, k=(1,1), q=1, W=({1,2}) should give f1(z)f2(z), and (4.12) gives 0). The intended repair is straightforward: use the same convention as Theorem 4.5, where the all-zero case is defined to be the product of the constants. But as written the displayed identity is false, and Part II is therefore not fully proved on the stated definitions. This is an exposition/verification gap, not a fatal one; Part I and the Poincaré inequalities are unaffected. A referee should still ask for the argument to be written carefully.\n\nTwo smaller concerns. The induction in Proposition 3.4 is compressed; iterating (3.1) requires care because (3.1) assumes integrability, and the truncation argument from Theorem 3.2 needs to be spelled out to justify each step. Also, Condition A-(loc) is genuinely strong, and the paper honestly says it cannot be easily removed for m≥3; Part II is conditional on that hypothesis, which limits but does not destroy the contribution.\n\nWho this is for: anyone working on Poisson stochastic analysis, multiple integrals, or product formulae. It deserves a serious referee. I would not desk-reject. My recommendation: send it to review, with a request to fix the Π≥2(0,...,0) convention and expand the proof of Proposition 3.4.","headline":"Genuine advance on long-open Poisson multiplication questions, but Part II's combinatorial bridge has a small, repairable gap when a word exhausts all kernel variables.","tokens_in":21819,"tokens_out":3717,"would_cite":true,"duration_ms":36798,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60H07","60H05","60E15","60G55","60G57"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves a necessary and sufficient condition for the square-integrability of products of Poisson multiple Wiener-Itô integrals, expressed through iterated add-one cost operators, and supplies the explicit chaos multiplication…","keywords":["Poisson functionals","multiple Wiener-Itô integrals","square-integrability","product formula","add-one cost operator","p-Poincaré inequalities","partition diagrams","Wiener chaos"],"falsifier":"One concrete test: take a triple of $L^2$ kernels with $K=3$, for instance the setting of Remark 4.10 with $Z=(1,\\infty)$, $\\mu(dz)=z^{-5/2}$, and $f_1=f_2=f_3(z)=z^{1/2}$, and evaluate both sides of (4.11)-(4.13) numerically on finite-volume approximations. A mismatch between the partition sum (1.9) and the empirical chaos coefficient would show where the local integrability condition is doing real work; a match would confirm the formula extends beyond the global integrability requirement.","tokens_in":20570,"feed_emoji":"🎲","tokens_out":12948,"duration_ms":127438,"temperature":0.7,"pith_summary":"Multiplying two or more Poisson multiple integrals can destroy square-integrability, because Poisson chaos is not hypercontractive. This paper proves that, for a product $\\Phi=\\prod_{i=1}^m I_{k_i}(f_i)$, the property $\\Phi\\in L^2(P)$ is equivalent to a finite list of add-one cost checks: for every $q=1,\\dots,K-1$ (with $K=k_1+\\cdots+k_m$), the $q$-th iterated add-one cost of $\\Phi$ is integrable and its expectation lies in $L^2(\\mu^q)$. When this holds, $\\Phi$ belongs to the finite chaos sum $\\oplus_{q=0}^K C_q$ and its top kernel is the symmetrized tensor product $\\mathrm{sym}(f_1\\otimes\\cdots\\otimes f_m)$. Under a local integrability condition on the kernels, the paper also gives the full chaos expansion through a partition-and-diagram formula, completing a program opened in [59] and extending the $m=2$ case of [18] to arbitrary $m$.","feed_headline":"Iterated add-one costs decide square-integrability of Poisson products","feed_subtitle":"For any number of Poisson integrals, square-integrability comes down to iterated add-one costs.","key_machinery":"The argument is carried by two tools. The first is the add-one cost operator $D^+_z F=F(\\eta+\\delta_z)-F(\\eta)$ and its iterates $D^{(q)}$, connected to Wiener chaos by the identity $h_q=(1/q!)E[D^{(q)}\\Phi]$; this reduces square-integrability of products to moment estimates on costs. The second is a new family of $p$-Poincar\\'e inequalities for almost surely finite Poisson functionals, $E|F-F'|^p\\le 2^{3-p}\\int E|D^+_zF|^p\\,\\mu(dz)$ for $p\\in[1,2]$, which lets the proof work without assuming $F$ is integrable, an essential feature because for $m>2$ the product may have infinite first moment. The explicit multiplication formula uses word-and-partition combinatorics: words $W=(A_1,\\dots,A_q)$ record which factors receive an add-one cost, and partitions $\\sigma\\in\\Pi(k_1,\\dots,k_m)$ organize which variables are identified and which are integrated, producing kernels $H(\\sigma,A;f_1,\\dots,f_m)$ whose sum over $|A|+|\\sigma_1|=q$ is $h_q$.","core_discovery":"On Poisson space, the product of finitely many random variables each lying in a finite Wiener chaos is square-integrable exactly when the intermediate iterated add-one costs do not blow up in expectation. Formally, for $\\Phi=\\prod_{i=1}^m I_{k_i}(f_i)$, Theorem 1.6 Part I shows that $\\Phi\\in L^2(P)$ if and only if, for every $q=1,\\dots,K-1$, $D^{(q)}_{z_1,\\dots,z_q}\\Phi\\in L^1(P)$ for $\\mu^q$-almost every $(z_1,\\dots,z_q)$ and the function $(z_1,\\dots,z_q)\\mapsto E[D^{(q)}_{z_1,\\dots,z_q}\\Phi]$ belongs to $L^2(\\mu^q)$. If this holds, $\\Phi$ lies in $\\oplus_{q=0}^K C_q$ and its projection onto the $K$-th chaos is $I_K(\\mathrm{sym}(f_1\\otimes\\cdots\\otimes f_m))$, the Wick-product property conjectured in [59]. Under Condition A-(loc), a local $L^1$ requirement on the fixed-variable slices of the kernels, these conditions are equivalent to the coefficients $h_q$ of (1.9) being square-integrable, and the partition formula then gives the actual chaos expansion.","pith_inferences":["The equivalence in Part I suggests a numerical certification route: on finite configuration spaces one can approximate the add-one cost expectations and estimate their $L^2$ norms; because the conditions are necessary and sufficient, this could give empirical confirmation of square-integrability before computing any partition sum.","A natural companion question is an $L^1$ criterion: the paper notes that no necessary and sufficient condition for $E|\\Phi|<\\infty$ is currently known, and the same cost-operator technology may yield one if the $p$-Poincar\\'e inequalities are iterated in a different way.","Because the proof is built on the add-one cost stochastic calculus, the same square-integrability criterion may transfer to marked Poisson processes or finite point processes with the same independence structure, potentially becoming a general tool for geometric functionals.","The explicit formula (1.9) is not claimed to hold when Condition A-(loc) fails; in such cases square-integrability could still hold, so the partition formula may admit a renormalized extension governed by the add-one cost expectations rather than by finite integrals of absolute values."],"forward_implications":["Whenever the product is square-integrable, it has nonzero chaos terms only of orders $0,\\dots,K$, and the top-order term is $I_K(\\mathrm{sym}(f_1\\otimes\\cdots\\otimes f_m))$.","For products of $m$ single integrals, the criterion becomes explicit: for $m=3$, $\\Phi\\in L^2(P)$ if and only if $f_1f_2f_3\\in L^2(\\mu)$ and $\\mathrm{sym}\\{(f_1f_2)\\otimes f_3+(f_1f_3)\\otimes f_2+(f_2f_3)\\otimes f_1\\}\\in L^2(\\mu^2)$, with closed formulas for the first three chaos coefficients.","Under Condition A-(loc), formula (1.9) computes each chaos kernel $h_q$ directly from partitions, so no separate verification of the intermediate add-one cost conditions is needed.","The new $p$-Poincar\\'e inequalities yield a conditional $L^p$ criterion for $p<2$: if the expected iterated costs lie in $L^p(\\mu^q)$, then the product lies in $L^p(P)$; the converse is false for $p<2$, making $p=2$ the sharp threshold.","For $m=2$, the theorem reproduces the previously known characterization, so the result is a strict extension to all $m\\ge 2$ rather than a new proof of a special case."],"supporting_citations":[{"why":"supplies the classical sufficient diagram formula and the open questions (s1)-(s3) that Theorem 1.6 answers.","marker":"[59]"},{"why":"settles the $m=2$ case with a contraction-based product formula, the base case this paper extends to arbitrary $m$.","marker":"[18]"},{"why":"establishes the $p$-Poincar\\'e inequalities for integrable variables that Theorem 3.2 generalizes to almost surely finite variables.","marker":"[62]"},{"why":"connects chaos kernels to expectations of iterated add-one costs, the identity behind Part I.","marker":"[38]"},{"why":"provides the partition-and-diagram calculus, the notation $\\Pi(k_1,\\dots,k_m)$, and the expectation formula used in (1.9).","marker":"[39]"},{"why":"is the source of the expectation theorem that turns iterated costs into partition sums under Condition A-(loc).","marker":"[41]"},{"why":"is cited for the fact that in the $m=2$ case the integrability assumptions behind the explicit formula can be removed.","marker":"[19]"}],"fun_headline_variants":["Poisson product integrability: iterated add-one costs are decisive","Wick product holds for Poisson chaos products under minimal conditions","Iterated add-one costs settle Poisson multiplication formula","Poisson products: square-integrability exactly when costs don't blow up","New Poincaré inequalities yield Poisson multiplication formula"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The explicit chaos coefficients of Part II are guaranteed only when the kernels' fixed-variable slices satisfy a local integrability condition (Condition A-(loc)); for $m\\ge 3$ this may fail even when the product itself is square-integrable, so the fully explicit multiplication formula rests on that condition.","fun_headline_variants_meta":{"raw":{"variants":["Poisson product integrability: iterated add-one costs are decisive","Wick product holds for Poisson chaos products under minimal conditions","Iterated add-one costs settle Poisson multiplication formula","Poisson products: square-integrability exactly when costs don't blow up","New Poincaré inequalities yield Poisson multiplication formula"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000803,"raw_usage":{"total_tokens":3538,"prompt_tokens":962,"completion_tokens":2576,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":578,"completion_tokens_details":{"reasoning_tokens":2494}},"tokens_in":578,"tokens_out":2576,"duration_ms":18359,"temperature":1.0,"reasoning_tokens":2494,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T20:54:39.150263+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"One concrete test: take a triple of $L^2$ kernels with $K=3$, for instance the setting of Remark 4.10 with $Z=(1,\\infty)$, $\\mu(dz)=z^{-5/2}$, and $f_1=f_2=f_3(z)=z^{1/2}$, and evaluate both sides of (4.11)-(4.13) numerically on finite-volume approximations. A mismatch between the partition sum (1.9) and the empirical chaos coefficient would show where the local integrability condition is doing real work; a match would confirm the formula extends beyond the global integrability requirement.","supporting_citations":[{"cited_title":"Surgailis","cited_arxiv_id":null,"evidence_quote":"supplies the classical sufficient diagram formula and the open questions (s1)-(s3) that Theorem 1.6 answers."},{"cited_title":"FourthmomenttheoremsonthePoissonspace:analytic statements via product formulae.Electron","cited_arxiv_id":null,"evidence_quote":"settles the $m=2$ case with a contraction-based product formula, the base case this paper extends to arbitrary $m$."},{"cited_title":"Trauthwein","cited_arxiv_id":null,"evidence_quote":"establishes the $p$-Poincar\\'e inequalities for integrable variables that Theorem 3.2 generalizes to almost surely finite variables."},{"cited_title":"Last and M","cited_arxiv_id":null,"evidence_quote":"connects chaos kernels to expectations of iterated add-one costs, the identity behind Part I."},{"cited_title":"Last and M","cited_arxiv_id":null,"evidence_quote":"provides the partition-and-diagram calculus, the notation $\\Pi(k_1,\\dots,k_m)$, and the expectation formula used in (1.9)."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"is the source of the expectation theorem that turns iterated costs into partition sums under Condition A-(loc)."},{"cited_title":"Döbler and G","cited_arxiv_id":null,"evidence_quote":"is cited for the fact that in the $m=2$ case the integrability assumptions behind the explicit formula can be removed."}],"review_version":1}