Poissonian Malliavin Calculus

• Poissonian Malliavin Calculus is a stochastic analysis framework that extends classical Malliavin calculus to Poisson random measures and pure-jump processes.
• It employs discrete difference operators, chaos decompositions, and dual operators to derive quantitative limit theorems and normal approximation bounds.
• The calculus is crucial for sensitivity analysis in SDEs/SPDEs with jumps, financial Greeks computation, and establishing absolute continuity of Poisson functionals.

Poissonian Malliavin calculus is the stochastic differential calculus of variations formulated for Poisson random measures, Lévy processes with jumps, and general pure-jump processes. Its central objects are discrete difference operators and their duals, which contrast fundamentally with the derivations and structure underlying classical Malliavin calculus on Wiener (Gaussian) space. The resulting calculus is fundamental for quantitative limit theorems, stochastic analysis of SDEs with jumps, sensitivity formulas (“Greeks”) in finance, stochastic partial differential equations driven by Lévy noise, absolute continuity results, and the Malliavin–Stein program for Gaussian approximations.

1. Construction of the Poisson Space and Chaos Decomposition

Given a probability space $(\Omega, \mathcal{F}, \mathbb{P})$ carrying a Poisson random measure $N$ on $[0,\infty)\times \mathbb{R}$ with Lévy measure $\nu$, the compensator is $\tilde N(dt,dx) = N(dt,dx) - dt\,\nu(dx)$ (R, 2017). Every $F \in L^2(\Omega)$ admits a unique Wiener–Itô chaos expansion: $F = \sum_{q=0}^\infty I_q(f_q),$ where $I_q(f_q)$ are multiple compensated Poisson integrals with symmetric, square-integrable kernels $f_q$ over $(\mathbb{R}_+\times \mathbb{R})^q$ and orthogonal decomposition (R, 2017, Andersson et al., 2017, Laukkarinen, 2016). Isometries and product/contracting formulas for these multiple integrals provide a Fock-space structure analogous to the Gaussian case (Bourguin et al., 2021).

2. Fundamental Operators: Difference, Divergence, and OU Generator

Gradient (Difference/Malliavin) Operator $N$0

For measurable $N$1, the discrete gradient is

$N$2

acting as a closed, unbounded operator from $N$3 into $N$4, with domain those $N$5 whose chaos kernels $N$6 satisfy $N$7. On the chaos expansion, $N$8 (R, 2017, Minh, 2011, Andersson et al., 2017).

Difference Rules

Unlike the Gaussian case, the product and chain rules incorporate jump terms: for $N$9,

$[0,\infty)\times \mathbb{R}$0

and for a smooth $[0,\infty)\times \mathbb{R}$1,

$[0,\infty)\times \mathbb{R}$2

Divergence (Skorohod) Operator $[0,\infty)\times \mathbb{R}$3

$[0,\infty)\times \mathbb{R}$4 is the formal adjoint of $[0,\infty)\times \mathbb{R}$5. For a process $[0,\infty)\times \mathbb{R}$6, if $[0,\infty)\times \mathbb{R}$7,

$[0,\infty)\times \mathbb{R}$8

and if $[0,\infty)\times \mathbb{R}$9 with $\nu$0, then $\nu$1 (R, 2017, Andersson et al., 2017).

Ornstein–Uhlenbeck Generator $\nu$2 and Pseudo-Inverse $\nu$3

$\nu$4 acts on chaos expansions via

$\nu$5

with domain determined by $\nu$6. The fundamental commutation relation $\nu$7 holds on a suitable domain, and the pseudo-inverse is given by $\nu$8, so $\nu$9 (R, 2017, Andersson et al., 2017).

3. Advanced Calculus: Chain Rule, Energy Bracket, Gamma Calculus

The Poissonian chain rule is substantially more complex than in Gaussian analysis. For a multivariate smooth function $\tilde N(dt,dx) = N(dt,dx) - dt\,\nu(dx)$0 and points $\tilde N(dt,dx) = N(dt,dx) - dt\,\nu(dx)$1,

$\tilde N(dt,dx) = N(dt,dx) - dt\,\nu(dx)$2

for some $\tilde N(dt,dx) = N(dt,dx) - dt\,\nu(dx)$3, with the last term reflecting jump compensation absent in Wiener analysis (R, 2017).

The carré du champ and energy bracket are extended via the Poisson version of the Gamma calculus. For $\tilde N(dt,dx) = N(dt,dx) - dt\,\nu(dx)$4,

$\tilde N(dt,dx) = N(dt,dx) - dt\,\nu(dx)$5

and

$\tilde N(dt,dx) = N(dt,dx) - dt\,\nu(dx)$6

provide higher-order variance/covariance control in limiting theorems (Herry, 2019, Bourguin et al., 2021).

4. Limit Theorems, Quantitative Approximation, and Stein–Malliavin Bounds

The “Nourdin–Peccati (NP) bound” for the normal approximation in Poisson space states that for centered, unit variance $\tilde N(dt,dx) = N(dt,dx) - dt\,\nu(dx)$7, the Wasserstein distance satisfies

$\tilde N(dt,dx) = N(dt,dx) - dt\,\nu(dx)$8

with explicit expressions for the remainder involving higher-order derivatives and the jump terms (R, 2017). A second-order Poincaré inequality further refines this to

$\tilde N(dt,dx) = N(dt,dx) - dt\,\nu(dx)$9

providing the explicit asymptotic rate of convergence, often $F \in L^2(\Omega)$0 or $F \in L^2(\Omega)$1 (R, 2017, Minh, 2011, 1940.10181).

In the Hilbert space–valued setting, functional limit theorems and fourth-moment bounds admit clean formulations involving kernel contractions and Gamma operators, extending classical “fourth-moment phenomena” to the Poisson framework (Bourguin et al., 2021).

5. Applications: SDEs with Jumps, SPDEs, Sensitivity, and Hawkes Processes

SDEs and SPDEs with Poisson Noise

The Poisson Malliavin calculus supports rigorous sensitivity analysis and error bounds for both finite-dimensional SDEs with jumps and infinite-dimensional SPDEs (particularly with α-stable drivers) (Andersson et al., 2017, Sojudi et al., 2 Feb 2025, Maurer et al., 6 Oct 2025). The integration by parts and commutation relations

$F \in L^2(\Omega)$2

enable sharp error estimates in weak and strong approximations. The Poisson–Alekseev–Gröbner formula provides an explicit stochastic error representation between solutions of SDEs differing in coefficients, under minimal regularity (Maurer et al., 6 Oct 2025).

Sensitivity (Greeks) in Financial Mathematics

Poissonian Malliavin calculus, via Skorokhod integration and explicit chaos expansions, yields efficient representations for Greeks in mean-field jump-diffusion models and path-dependent discontinuous payoff functions, paralleling the Bismut–Elworthy–Li formula but for pure jump processes (Sojudi et al., 2 Feb 2025). Formulae for Delta involve the computation of predictable Skorohod weights solving integration-by-parts equations, with theoretical and computational advantages over naive finite-difference estimators.

Poisson–Malliavin–Stein Program for Gaussian Approximation

Key results connect the Malliavin covariance $F \in L^2(\Omega)$3 to optimal bounds in normal approximation for functionals of Poisson process, Poisson U-statistics, and functionals of Hawkes processes (Minh, 2011, Hillairet et al., 2021, Khabou, 2021, Bourguin et al., 2021). The multidimensional and functional functional CLTs employ explicit contraction/kernels–based bounds and are central in stochastic geometry, random graphs, and non-diffusive stochastic models.

6. Absolute Continuity, Malliavin-Sobolev Spaces, and Regularity

Sharp conditions for absolute continuity of the law of Poisson functionals arise naturally via Malliavin calculus. The criterion

$F \in L^2(\Omega)$4

on an event ensures absolute continuity of the law of $F \in L^2(\Omega)$5 restricted to that event (León et al., 2012). Appropriate weighted Lebesgue space characterizations of differentiability and fractional differentiability also follow, with the key result that for $F \in L^2(\Omega)$6-measurable $F \in L^2(\Omega)$7,

$F \in L^2(\Omega)$8

and for fractional spaces via real interpolation,

$F \in L^2(\Omega)$9

demonstrating the explicit relationship between jump-count and regularity on Poisson spaces (Laukkarinen, 2016).

7. Comparison: Poisson, Wiener, and Mixed Wiener–Poisson Calculi

The Poissonian Malliavin calculus is part of a general unified stochastic calculus of variations, admitting Gaussian (Wiener), Poissonian (jump), and mixed Wiener–Poisson frameworks (R, 2017, Sojudi et al., 2 Feb 2025). In the Wiener case, the chain rule is global and the chaos is Hermitian. In the Poissonian setting, extra compensation terms and different commutation/measurability properties arise, and the semigroup structure is more intricate. Mixed spaces combine both differentials, support joint limit theorems, and exhibit error rates inherited from both the Gaussian and Poissonian cases.

---

References:

(R, 2017) Normal Convergence Using Malliavin Calculus With Applications and Examples (Andersson et al., 2017) Poisson Malliavin calculus in Hilbert space with an application to SPDE (Laukkarinen, 2016) A note on Malliavin smoothness on the Lévy space (León et al., 2012) Local Malliavin Calculus for Lévy Processes and Applications (Minh, 2011) Malliavin-Stein method for multi-dimensional U-statistics of Poisson point processes (Bourguin et al., 2021) Functional Gaussian approximations on Hilbert-Poisson spaces (Herry, 2019) Stable limit theorems on the Poisson space (Sojudi et al., 2 Feb 2025) Sensitivity Analysis for Mean-Field SDEs With Jump By Malliavin Calculus: Chaos Expansion Approach (Maurer et al., 6 Oct 2025) A Poisson-Alekseev-Gröbner formula through Malliavin calculus for Poisson random integrals (Hillairet et al., 2021) The Malliavin-Stein method for Hawkes functionals (Khabou, 2021) Malliavin-Stein method for the multivariate compound Hawkes process