L1 Formula in Poisson Analysis

• L1 Formula is a Clark–Ocone-type representation for integrable Poisson functionals that uses pathwise add‐one differences and truncation to yield a predictable integrand.
• It differs from the classical Clark–Ocone formula by employing the Poisson measure (not the compensated measure), working in L1 instead of L2, and using a direct pathwise operator.
• The formula provides a unique, explicit decomposition of any integrable Poisson functional, with significant implications for Poisson analysis and Malliavin calculus.

Searching arXiv for the specified paper and closely related Clark–Ocone/Poisson work. arXiv search query: (Hillairet et al., 2024) Poisson imbedding meets the Clark-Ocone formula The $L^{1}$ formula in the setting of Poisson analysis is a Clark–Ocone-type representation for integrable Poisson functionals on a filtered probability space carrying a Poisson random measure. In the formulation developed in "Poisson imbedding meets the Clark-Ocone formula" (Hillairet et al., 2024), the representation extends the Poisson imbedding for point processes and differs from the classical Clark–Ocone formula in three explicit ways: it is written with respect to the Poisson measure rather than the compensated measure, it holds in $L^{1}$ rather than $L^{2}$, and its integrand is defined as a pathwise operator rather than as an $L^{2}$-limiting object. The resulting integrand is characterized as a predictable integrable process (Hillairet et al., 2024).

1. Probabilistic setting and statement of the formula

The representation is formulated on a filtered probability space carrying a Poisson random measure $N$ on $$([0,T]\times\mathbb X,\mathcal B([0,T])\otimes\mathcal X)$$ with intensity measure $\mu=dt\otimes\pi$, where $\pi$ is $\sigma$-finite and atomless. The predictable $\sigma$-field on $L^{1}$0 is denoted by $L^{1}$1 (Hillairet et al., 2024).

In the finite-mass case, namely under the assumption

$L^{1}$2

the main $L^{1}$3-representation theorem applies to any $L^{1}$4. The theorem introduces the empty-configuration value

$L^{1}$5

and then asserts that the random variable $L^{1}$6 admits the almost sure decomposition

$L^{1}$7

Equivalently,

$L^{1}$8

The theorem further states that $L^{1}$9 belongs to $L^{2}$0, meaning that it is predictable and integrable against $L^{2}$1, and that the decomposition is unique: if

$L^{2}$2

for some $L^{2}$3 and $L^{2}$4, then necessarily $L^{2}$5 and $L^{2}$6 almost everywhere (Hillairet et al., 2024).

In this context, the label $L^{2}$7 refers to the integrability class of both the functional $L^{2}$8 and the predictable integrand, rather than to an $L^{2}$9-norm construction or to an $L^{2}$0-type time discretization.

2. Pathwise operators and the predictable integrand

The formula is built from two pathwise operations. The first is the usual add-one difference, or Malliavin operator,

$L^{2}$1

The second is truncation at time $L^{2}$2, defined by

$L^{2}$3

that is, all atoms at times $L^{2}$4 are removed. The integrand is then defined by composition: $L^{2}$5 Because $L^{2}$6 is measurable with respect to the left-limit $L^{2}$7-field $L^{2}$8, the process $L^{2}$9 is $N$0-predictable by construction (Hillairet et al., 2024).

The ambient integrability space is

$N$1

so that $N$2 is well defined in $N$3 and almost surely (Hillairet et al., 2024).

This construction is technically significant because the integrand is not obtained through a projection or limiting procedure in a Hilbert space. Instead, it is a direct pathwise operator $N$4. A plausible implication is that the representation emphasizes the causal structure of the Poisson configuration: the value at $N$5 is computed by adding an atom to the truncated past, rather than by conditioning a derivative taken on the full configuration.

3. Relation to the classical Clark–Ocone formula

The representation differs from the classical Clark–Ocone formula on three stated accounts. First, it is expressed with respect to the Poisson measure $N$6 itself, not the compensated measure. Second, it holds in $N$7 and not in $N$8. Third, the integrand is a pathwise operator and not an $N$9-limiting object (Hillairet et al., 2024).

These differences are structural rather than cosmetic. The classical Clark–Ocone framework is ordinarily associated with martingale decompositions and compensated stochastic integrals, whereas the present formula is tailored to uncompensated Poisson integration. The paper explicitly identifies the result as a Pseudo-Clark–Ocone representation formula, and the proof uses Malliavin calculus together with a pseudo-chaotic decomposition with uncompensated iterated integrals (Hillairet et al., 2024).

The theorem therefore occupies a distinct position between two traditions. On one side lies the classical Clark–Ocone formula, which is centered on compensated noise and $$([0,T]\times\mathbb X,\mathcal B([0,T])\otimes\mathcal X)$$0-methods; on the other lies Poisson imbedding for point processes, which is inherently uncompensated and pathwise. The new formula connects these by showing that an $$([0,T]\times\mathbb X,\mathcal B([0,T])\otimes\mathcal X)$$1, uncompensated, predictable representation is available for any integrable Poisson functional.

4. Proof architecture and pseudo-chaotic expansion

The proof proceeds through a pseudo-chaotic expansion. For any square-integrable $$([0,T]\times\mathbb X,\mathcal B([0,T])\otimes\mathcal X)$$2, Theorem 3.1 gives

$$([0,T]\times\mathbb X,\mathcal B([0,T])\otimes\mathcal X)$$3

almost surely, where $$([0,T]\times\mathbb X,\mathcal B([0,T])\otimes\mathcal X)$$4 is the $$([0,T]\times\mathbb X,\mathcal B([0,T])\otimes\mathcal X)$$5-th pathwise difference at the empty path and $$([0,T]\times\mathbb X,\mathcal B([0,T])\otimes\mathcal X)$$6 denotes the $$([0,T]\times\mathbb X,\mathcal B([0,T])\otimes\mathcal X)$$7-fold integral against the factorial measure $$([0,T]\times\mathbb X,\mathcal B([0,T])\otimes\mathcal X)$$8 (Hillairet et al., 2024).

The argument then introduces finite-chaos approximants by truncating the series at level $$([0,T]\times\mathbb X,\mathcal B([0,T])\otimes\mathcal X)$$9. For each such approximation,

$\mu=dt\otimes\pi$0

and the integrand $\mu=dt\otimes\pi$1 is computed explicitly term by term. At this stage it is manifestly predictable and belongs to $\mu=dt\otimes\pi$2 (Hillairet et al., 2024).

The passage from finite chaos to general $\mu=dt\otimes\pi$3-functionals uses an $\mu=dt\otimes\pi$4-approximation argument. By Lemma 4.1, the operators $\mu=dt\otimes\pi$5 are continuous in $\mu=dt\otimes\pi$6 for $\mu=dt\otimes\pi$7, so $\mu=dt\otimes\pi$8 in $\mu=dt\otimes\pi$9. The stochastic integrals then converge in $\pi$0, yielding the desired decomposition for $\pi$1. Uniqueness follows by pathwise difference arguments: any other predictable integrand must agree with the pathwise quantity $\pi$2 (Hillairet et al., 2024).

The proof strategy is notable because it avoids compensators and avoids identifying the integrand as an orthogonal projection. This suggests a reformulation of Poisson representation theory in which uncompensated iterated integrals and pathwise truncation replace the more familiar $\pi$3-martingale machinery.

5. Poisson imbedding as an illustrative example

The paper’s main example concerns a point process $\pi$4 on $\pi$5 with predictable intensity $\pi$6. By standard thinning,

$\pi$7

Direct computation yields

$\pi$8

and after truncation these higher-order terms disappear: $\pi$9 Accordingly, the Pseudo-Clark–Ocone formula reduces exactly to

$\sigma$0

which is the classical Poisson-imbedding formula (Hillairet et al., 2024).

This example clarifies the operational meaning of the integrand. In the present framework, the correct predictable integrand is obtained by the simple pathwise difference at time $\sigma$1. The paper contrasts this with the standard Clark–Ocone integrand, for which one would have to take conditional expectations of the full future term and obtain a more cumbersome expression (Hillairet et al., 2024). The example therefore serves not merely as an application but as an explanation of why the pathwise operator $\sigma$2 is the natural integrand in the uncompensated $\sigma$3 setting.

6. Scope, terminology, and conceptual significance

The theorem identifies a representation for any integrable Poisson functional in the finite-mass case. Its output is a predictable integrable process, and its normalization term is the empty-configuration value $\sigma$4 (Hillairet et al., 2024). These two ingredients replace the more familiar pair consisting of expectation and compensated stochastic integral.

A frequent source of ambiguity is the phrase $\sigma$5 formula. Here it does not denote an $\sigma$6-penalized variational principle, an entrywise $\sigma$7-norm, or the classical L1 finite-difference formula for Caputo derivatives. Rather, it denotes a representation theorem valid under $\sigma$8, with the integrand belonging to $\sigma$9. The distinction from the $\sigma$0-based Clark–Ocone theory is explicit in the statement of the result (Hillairet et al., 2024).

The conceptual contribution can be summarized in three linked features. The representation is uncompensated, pathwise, and uniquely determined by truncation plus add-one difference. This suggests that the theorem is not only an extension of Poisson imbedding, but also a reorganization of Poisson Malliavin calculus around predictable pathwise operators. Within that perspective, the formula

$\sigma$1

is the central identity: it expresses an arbitrary integrable Poisson functional as its value on the empty configuration plus the accumulated predictable effect of adding atoms to the past.