Product Formula of Multiple Integrals of Levy Process

Neha Sharma; Nishant Agrawal; Yaozhong Hu

REVIEW 2 major objections 4 minor 10 references

Product Formula of Multiple Integrals of Levy Process

T0 review · 2 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The product of any finite number of multiple stochastic integrals of a pure-jump Lévy process is a finite sum of single multiple integrals of symmetrically contracted kernels.

desk verdict Elegant formal computation of a general product formula for Lévy multiple integrals, but the stated L2 domain is indefensible; the theorem needs extra integrability conditions before it can be used. read the letter →

arxiv 1908.01225 v2 pith:ZDVYYF2P submitted 2019-08-03 math.PR

Nishant Agrawal , Yaozhong Hu , Neha Sharma This is my paper · ORCID

classification math.PR MSC 60H0560G5160H30

keywords Lévyprocessnonlinearfunctionalofmultipleintegralschaosexpansionproductformulaexponentialvectorpolarizationtechnique

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper gives a single closed-form formula for the product of any finite number of multiple integrals with respect to the compensated Poisson random measure of a purely discontinuous Lévy process. The formula expands the product into a finite sum of multiple integrals whose kernels are obtained by symmetrized contractions of the original kernels, with explicit factorial coefficients. This generalizes the classical Brownian product formula and unifies earlier two-integral results. The proof is short: it compares the chaos expansions of a product of exponential vectors in two ways and reads off coefficients.

What carries the argument

The exponential vector $\mathcal{E}(\rho)=\exp\{\int_{\mathbb{T}}\rho\,d\tilde N - \int_{\mathbb{T}}(e^\rho-1-\rho)\,d\nu dt\}$, whose chaos expansion has kernels $(e^\rho-1)^{\hat\otimes n}$. The proof expands the product of $m$ such exponential vectors in two ways—directly as a product of independent chaos expansions, and as a single exponential vector with an extra deterministic correction factor—and then polarizes by taking $\rho_k=\log(1+u_k p_k)$. Comparing coefficients of $u_1^{q_1}\cdots u_m^{q_m}$ yields the contraction formula. The combinatorial core is the pair of operators $\hat\otimes_i^\mu$ and $V_j^\nu$ that glue or integrate shared time–jump variables.

What would settle it

Take a Lévy process with Lévy measure $\nu(dz)=\mathbf{1}_{(0,1)}(z)\,dz$, set $m=3$, and let $f_1=f_2=f_3=h$ with $h(t,z)=z^{-1/4}$ on $(0,1)^2$. Then $h\in L^2(dt\,d\nu)$ but $h^3\notin L^2(dt\,d\nu)$. The right-hand side of (2.12) contains the term $I_1(h^3)$ (the triple contraction with one shared variable), which is not defined, while the left-hand side $I_1(h)^3$ is a well-defined random variable. Hence the theorem is false as stated for arbitrary $L^2$ kernels; checking just this term would settle the claimed domain.

Watch

Extended reading notes

Core claim

Theorem 2.2 asserts that for $m\ge2$ and symmetric kernels $f_k \in (L^2([0,T]\times \mathbb{R}_0, dt\otimes \nu))^{\hat\otimes q_k}$, the product $\prod_{k=1}^m I_{q_k}(f_k)$ equals a finite sum, indexed by multi-indices $\vec l,\vec n$, of single multiple integrals $I_{|q|+|\vec n|-|\chi(\vec l,\vec n)|}(\hat\otimes_{\vec i}^{\vec l}\hat\otimes V_{\vec j}^{\vec n}(f_1,\dots,f_m))$, with weights $\prod_k q_k! /(\prod_\alpha l_{i_\alpha}! \prod_\beta \mu_{j_\beta}! \prod_k (q_k-\chi(k,\vec l,\vec n))!)$. In words, the product of several multiple jump integrals is a linear combination of one multiple jump integral of a symmetrized kernel in which groups of the original variables are glued together or integrated out. For $m=2$ the formula reduces to a two-integral expansion; in the Brownian limiting case it reduces to the classical contraction formula.

Load-bearing premise

The proof shows the identity for kernels that are finite sums of tensor products of one-variable functions, then asserts it extends to all square-integrable kernels by a routine limiting argument; that extension requires the contracted kernels on the right-hand side to remain square-integrable, which need not happen when a term glues together three or more kernels.

Editorial extensions

If this is right

For any $m\ge2$, the product of $m$ multiple integrals with respect to the jump measure is a finite sum of multiple integrals; no infinite series appears beyond the finite indexing by subsets.
The $m=2$ case recovers the known formula for products of two Lévy integrals and, in the Brownian limit, the classical contraction rule.
Because square-integrable functionals of the Lévy process have a unique chaos expansion, the formula allows products of arbitrary chaos expansions to be re-expanded into a single chaos series.
The compact form of the coefficients makes the formula suitable for moment computations and limit theorems for statistics built from jump processes.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The stated domain over all $L^2$ kernels is probably too broad: the contraction operator $V_j^\nu$ for $|j|\ge3$ does not preserve $L^2$, so the theorem likely needs an added integrability condition on the kernels or on the Lévy measure.
The exponential-vector derivation suggests an analogous product formula for multiple integrals with respect to other random measures that admit a Wick-type exponential and a chaos decomposition, as long as the deterministic correction factor is computed correctly.
A concrete check of the formula for $m=3$ and a Poisson process with unit jumps, comparing $I_1(1)^3$ against the direct polynomial $(N(T)-T)^3$, would pin down the contraction coefficients and expose any missing terms.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Dear colleague,

You should know that this paper has a nice proof idea and a genuinely new-looking formula for products of m≥3 multiple integrals with respect to the compensated Poisson random measure. But Theorem 2.2 as stated is false, not merely unproved: the RHS can contain terms that are not defined for arbitrary square-integrable kernels. The stress-test example is right. For m=2, n=m=1, the formula gives I(f)^2 = I_2(f⊗f) + I_1(f^2) + ∫ f^2 ν dt. If ν=δ_1, T=1, f(t)=(1-t)^{-1/4}, then f∈L^2 but f^2∉L^2, so I_1(f^2) is undefined (and the fourth moment diverges). So the 'routine limiting argument' at the end of Section 3 cannot work.

What is genuinely good: the exponential-vector/chaos-expansion derivation is short and elegant. It reduces correctly to Shigekawa's Brownian formula and to the m=2 result of Lee-Shih when the kernels are nice. For m≥3, the general formula appears new, and the bookkeeping with multi-indices is clear enough to follow. The proof works verbatim for kernels that are finite linear combinations of tensor products of bounded functions, say, or whenever all contracted kernels remain in L^2.

The soft spots are exactly where the reader's report points. The contraction operators V^ν_j and ⊗̂^μ_i do not preserve L^2 when |j|≥2 or |i|≥2, and no attempt is made to check integrability of the output. Also, the exponential functional (3.1) requires something like e^ρ-1 ∈ L^2; for ρ=log(1+up) with p∈L^2, B contains ∫ p_i p_j ν dt which may be infinite unless p_i p_j ∈ L^1. That is the same root issue.

This is a load-bearing flaw, but a fixable one. The paper should state the theorem for kernels satisfying explicit sufficient conditions (e.g., all needed contractions in L^2, or work with bounded functions and compact support and then close under a norm that controls the contractions). The m=2 case with the extra condition fg∈L^2 is known and fine; the general m version, with such conditions, would be a useful contribution.

Who should read it: researchers in stochastic analysis who want product formulas for Lévy multiple integrals. It deserves a serious referee rather than a desk reject, because the idea is sound and the gap is a standard domain issue, not an error of principle. But my own verdict: as written, the theorem is not true on its stated domain.

Referee Report

2 major / 4 minor

Summary. The paper proposes a general product formula for the product of m multiple stochastic integrals with respect to the compensated Poisson random measure of a pure-jump Lévy process. The formula expresses the product as a finite sum of multiple integrals whose kernels are obtained by a family of contraction and multiplication operations. The proof is based on exponential vectors: the authors expand an exponential functional in two ways, compare coefficients, and then extend by polarization and a claimed routine limiting argument. The m=2 case is stated as Theorem 2.3 and is shown to reduce to the classical Brownian product formula. The central claim is Theorem 2.2, equation (2.12).

Significance. The product formula for multiple integrals is a standard tool in stochastic analysis; a compact general formula for m factors in the Lévy setting would be useful. The exponential-vector method is elegant, the coefficient comparison is explicit, and the paper's formal derivation is a strength. If the formula were valid on the stated L2 domain, it would unify and extend known results (e.g., Shigekawa and Lee–Shih). However, the stated domain is not correct: as detailed below, the right-hand side can contain terms that are not well-defined multiple integrals for admissible L2 kernels, and the proof's final density argument is invalid. Thus the main result as stated cannot be accepted; the useful part of the computation would require a corrected statement with additional integrability assumptions.

major comments (2)

[§2, Theorem 2.3 (Eq. (2.13))] The formula is not well-defined for all f in L2. In the case m=2, q1=q2=1, Eq. (2.13) reads I_1(f)^2 = I_2(f⊗f) + I_1(f^2) + ∫_T f^2 dλdν. For f∈L2(T,dλ×dν) there is no guarantee that f^2∈L2(T,dλ×dν), so the term I_1(f^2) is undefined. For example, take T=[0,1], ν=δ_1, and f(t)=(1−t)^{−1/4}. Then f∈L2, but ∫ f^4 dλ = ∞, so f^2∉L2; consequently the right-hand side is not defined, and in fact E[I_1(f)^4]=∞, so the left-hand side does not even belong to L2. Hence Theorem 2.2 cannot hold for the stated domain.
[§3, last paragraph ('routine limiting argument')] The proof only establishes the identity for kernels that are finite linear combinations of symmetric tensor products p_1⊗...⊗p_q with p_i∈L2. The extension to all f_k∈L2 requires the contraction maps in (2.9)–(2.10) to be continuous on L2 and to preserve square-integrability. They do not: for a multi-index j with |j|≥3, V_j^ν(f_1,...,f_m) contains a product of several L2 functions evaluated at the same variables (t_1,z_1),...,(t_ν,z_ν), and such a product need not be integrable; even in the two-factor case the map f↦f^2 shows the problem. Therefore the 'routine limiting argument' is not a technical gap but an impossibility. A correct statement would need to impose extra integrability conditions on the kernels or on the Lévy measure and then prove the identity for that restricted domain; as written, the argument cannot be repaired by a denseness argument.

minor comments (4)

[§3, Eq. (3.8)] There is a typographical error in the second exponential factor: an unmatched closing parenthesis appears after 'm−1'. The intended expression is exp{∫_T ( e^{Σρ_k} − Σ e^{ρ_k} + m − 1 ) ν(dz)ds }.
[§2, Theorem 2.3 (2.13)–(2.14)] The notation In(fn)Im(gm) is confusing; in Theorem 2.3 the symbols f and g denote kernels of order n and m, not functions named f_n and g_m. Also, the kernel definition in (2.14) loses some arguments in the displayed formula; it should show the dependence on (s_1,z_1),...,(s_{n+m−k−2l}, z_{n+m−k−2l}) explicitly.
[§2, after Eq. (2.2)] The symbol T is used both for the time horizon and for the domain [0,T]×R0. This overload is confusing; a different letter for the product domain would improve readability.
[§3, Eq. (3.12)] There is a typo in the exponent of the last term: 'u^{lliκm}_{iκm}' should read 'u^{liκm}_{iκm}'.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the product formula is derived from Itô's formula, chaos expansion of exponential functionals, and coefficient comparison, with no fitted parameter or self-citation forcing the conclusion.

full rationale

The derivation chain is self-contained and non-circular. The paper establishes the chaos expansion of the exponential functional E(ρ) directly from Itô's formula (Section 3, equations (3.1)-(3.2)), then computes the product of m exponential functionals in two ways: as a power series in u_k whose coefficients are products of multiple integrals (3.7), and via the factorization A·B into an exponential functional factor and a deterministic factor (3.8), each expanded independently using (3.2)-(3.3). Comparing the coefficients of u_1^{q_1}...u_m^{q_m} yields the product formula (2.12) for tensor-product kernels, and polarization extends it to finite linear combinations of tensor products. The identity is not assumed anywhere; it is the output of algebraic coefficient comparison. The cited results, including Hu's textbook [2] used for the polarization technique and Shigekawa's formula [9] mentioned as a Brownian special case, are external standard tools and are not used to assume Theorem 2.2. The only substantive weakness is the closing 'routine limiting argument': the contraction operators in (2.12) do not appear to preserve L^2 for general f_k in the stated symmetric L^2 spaces, so the stated domain may be too broad and the extension step is underproved. That is a correctness/domain concern, not circularity, because the formula is still being derived rather than assumed. Accordingly, the circularity score is 0.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The proof relies on standard chaos expansion and polarization but introduces two hidden domain assumptions: the contracted kernels must be square-integrable, and the limiting argument must be valid. These are not free parameters, but they restrict the domain of validity of the central theorem.

assumptions (4)

standard math Wiener-Ito chaos expansion for square-integrable functionals of a pure jump Levy process (Theorem 2.1).
Invoked at the start of Section 3 and cited to standard references; not proved in this paper.
domain assumption The exponential functional E(rho) admits the chaos expansion (3.2)-(3.3) with convergence in L2 for the functions rho used.
Requires (e^rho - 1) in L2; for sums of rho_k of the form log(1+u_k p_k), the kernel prod(1+u_k p_k)-1 may not be in L2, so the series is only formal unless extra conditions hold.
ad hoc to paper The identity (2.12) extends via a routine limiting argument from tensor-product kernels to all f_k in L2.
This is the load-bearing gap: the RHS operators are not continuous on L2, and the contracted kernels may not be square-integrable, so the stated limiting argument is not routine.
ad hoc to paper All functions appearing as integrands in the multiple integrals on the RHS of (2.12) are in the appropriate L2 spaces.
Implicit in Theorem 2.2 but not stated; it fails for arbitrary L2 f_k, for example the k-contraction terms in the m=2 formula can be non-L2.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Product Formula of Multiple Integrals of Levy Process." pith.science (2026). https://pith.science/paper/ZDVYYF2P

@misc{pith2026190801225,
  author       = {Pith},
  title        = {Pith review of: Product Formula of Multiple Integrals of Levy Process},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZDVYYF2P}},
  note         = {Machine review of arXiv:1908.01225}
}

read the original abstract

We derive a product formula for the multiple stochastic integrals with respect to Levy process. The idea is to use exponential vectors and the polarization technique which greatly simplify the argument.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

10 extracted references · 9 canonical work pages

[1]

L\'evy processes and stochastic calculus

Applebaum, D. L\'evy processes and stochastic calculus. Second edition. Cambridge Studies in Advanced Mathematics, 116. Cambridge University Press, Cambridge, 2009

work page 2009
[2]

and ksendal, B

Di Nunno, G. and ksendal, B. and Proske, F

work page
[3]

Analysis on Gaussian spaces

Hu, Y. Analysis on Gaussian spaces. World Scientific Publishing Co. Pte. Ltd., Hackensack, NJ, 2017

2017
[4]

and ; Shih, H.-H

Lee, Y.-J. and ; Shih, H.-H. The product formula of multiple L\'evy-It\^o integrals. Bull. Inst. Math. Acad. Sinica 32 (2004), no. 2, 71-95

work page 2004
[5]

Multiple Wiener-It\^o integrals

Major, P. Multiple Wiener-It\^o integrals. With applications to limit theorems. Second edition. Lecture Notes in Mathematics, 849. Springer, Cham, 2014

work page 2014
[6]

Meyer, P. A. Quantum probability for probabilists. Lecture Notes in Mathematics, 1538. Springer-Verlag, Berlin, 1993

work page 1993
[7]

The Malliavin calculus and related topics

Nualart, D. The Malliavin calculus and related topics. Second edition. Probability and its Applications (New York). Springer-Verlag, Berlin, 2006

work page 2006
[8]

Protter, P. E. Stochastic integration and differential equations. Second edition. Version 2.1. Corrected third printing. Stochastic Modelling and Applied Probability, 21. Springer-Verlag, Berlin, 2005

work page 2005

Show all 10 references

[9]

L\'evy processes and infinitely divisible distributions

Sato, K. L\'evy processes and infinitely divisible distributions. Translated from the 1990 Japanese original. Revised edition of the 1999 English translation. Cambridge Studies in Advanced Mathematics, 68. Cambridge University Press, Cambridge, 2013

1990
[10]

Derivatives of Wiener functionals and absolute continuity of induced measures

Shigekawa, I. Derivatives of Wiener functionals and absolute continuity of induced measures. J. Math. Kyoto Univ. 20 (1980), no. 2, 263-289

1980

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [1]

L\'evy processes and stochastic calculus

Applebaum, D. L\'evy processes and stochastic calculus. Second edition. Cambridge Studies in Advanced Mathematics, 116. Cambridge University Press, Cambridge, 2009

work page 2009

[2] [2]

and ksendal, B

Di Nunno, G. and ksendal, B. and Proske, F

work page

[3] [3]

Analysis on Gaussian spaces

Hu, Y. Analysis on Gaussian spaces. World Scientific Publishing Co. Pte. Ltd., Hackensack, NJ, 2017

2017

[4] [4]

and ; Shih, H.-H

Lee, Y.-J. and ; Shih, H.-H. The product formula of multiple L\'evy-It\^o integrals. Bull. Inst. Math. Acad. Sinica 32 (2004), no. 2, 71-95

work page 2004

[5] [5]

Multiple Wiener-It\^o integrals

Major, P. Multiple Wiener-It\^o integrals. With applications to limit theorems. Second edition. Lecture Notes in Mathematics, 849. Springer, Cham, 2014

work page 2014

[6] [6]

Meyer, P. A. Quantum probability for probabilists. Lecture Notes in Mathematics, 1538. Springer-Verlag, Berlin, 1993

work page 1993

[7] [7]

The Malliavin calculus and related topics

Nualart, D. The Malliavin calculus and related topics. Second edition. Probability and its Applications (New York). Springer-Verlag, Berlin, 2006

work page 2006

[8] [8]

Protter, P. E. Stochastic integration and differential equations. Second edition. Version 2.1. Corrected third printing. Stochastic Modelling and Applied Probability, 21. Springer-Verlag, Berlin, 2005

work page 2005