Operator Factorization Beyond Hilbert Spaces

3 Banach Energy Space Framework

This chapter formalizes the abstract operator factorization theory from [1]. The derivative \(D = \delta ^*\) is constructed (not axiomatized) via the Banach dual pairing \(\langle DF, u \rangle = \mathbb {E}[F \cdot \delta (u)]\), and its properties—linearity, vanishing on constants, kernel characterization—are all proved from this single identity. The main results are: Theorem A (fluctuation factorization on \(\mathrm{im}(\delta )\)), the representability obstruction (centered even functionals cannot lie in \(\mathrm{im}(\delta _L)\), proved by the even-annihilation/self-pairing argument), Theorem B (the product rule with jump defect \(\Gamma \)), and Theorem C (chaos kernel characterization of \(\ker (D_L)\)). The chapter concludes by instantiating the abstract framework for symmetric \(\gamma \)-stable processes and verifying backward compatibility with the Hilbert formalization via the Hilbert bridge.

3.1 Abstract Framework

The Banach energy space axiomatizes the minimal structure—divergence, duality pairings, and expectation—needed to develop the operator theory without assuming a Hilbert space identification.

Definition 44 Banach energy space
✓
#

A Banach integrand space \(H_X\) with dual \(H_X^*\), \(L^p\)–\(L^q\) function spaces, divergence \(\delta : H_X \to L^p\), duality pairings, and expectation. Unlike the Hilbert framework, no Riesz identification \(H_X^* \cong H_X\) is assumed, which is precisely the generalization required by the singular Lévy measure.

Definition 45 Centering
✓
#

\(\mathrm{center}(G_0) = G_0 - \mathrm{const}(\mathbb {E}[G_0])\): removes the mean, preserving variance and symmetry. Centering is the first step in preparing the obstruction witness.

Theorem 46 Center is centered
✓
#

\(\mathbb {E}[\mathrm{center}(G_0)] = 0\), by linearity of expectation. One of the three hypotheses of the representability obstruction, ensuring the witness has zero mean.

Proof ▶

Expand \(\mathbb {E}[G_0 - \mathrm{const}(\mathbb {E}[G_0])]\) using linearity (expect_sub) and \(\mathbb {E}[\mathrm{const}(c)] = c\) (expect_const), giving \(\mathbb {E}[G_0] - \mathbb {E}[G_0] = 0\).

3.2 The Derivative \(D\)

The operator-covariant derivative \(D\) is constructed via the Banach dual pairing, and its fundamental properties—linearity, vanishing on constants, and kernel characterization—are derived from a single defining identity.

Definition 47 Operator-covariant derivative
✓
#

Constructed as \(D(F) := \mathrm{mk\_ dual}(u \mapsto \mathbb {E}[F \cdot \delta (u)])\). Bounded linearity on \(H_X\) follows from Hölder, so \(D(F) \in H_X^*\). This operator generalizes the Malliavin derivative [3] to the non-Hilbert setting.

Theorem 48 \(D\) defining identity
✓
#

\(\langle DF, u \rangle = \mathbb {E}[F \cdot \delta (u)]\). All downstream results follow from this single identity, which serves as the axiomatic interface for \(D\).

Proof ▶

Unfold \(D\) and apply mk_dual_spec.

Theorem 49 \(D\) linear
✓
#

\(D(\alpha F + G) = \alpha DF + DG\). Linearity of \(D\) is inherited from linearity of expectation and the duality pairing.

Proof ▶

Evaluate on test \(u\) via the defining identity. Linearity of \(\mathbb {E}\) and Hölder give equality; nondegeneracy of the pairing concludes.

Theorem 50 \(D\) on constants
✓
#

\(D(\mathrm{const}(c)) = 0\), since \(\mathbb {E}[c \cdot \delta (u)] = c \cdot 0 = 0\) by (H1). Constants carry no stochastic information, so their derivative vanishes.

Proof ▶

Apply the defining identity, factor out \(c\), use holder_const_left and centering of \(\delta \).

Theorem 51 \(\ker (D)\) characterization
✓

\(DF = 0\) iff \(\mathbb {E}[F \cdot \delta (u)] = 0\) for all \(u\): the kernel is the annihilator of \(\mathrm{im}(\delta )\). This characterization reduces kernel membership to a testable orthogonality condition.

Proof ▶

Forward: \(DF = 0\) means \(\langle DF, u \rangle = 0\) for all \(u\); apply defining identity. Reverse: the pairing is zero for all \(u\); apply pair_nondegenerate.

3.3 Fluctuation Factorization (Theorem A)

Every centered functional is shown to be representable as a divergence, providing the factorization that the derivative \(D\) inverts.

Theorem 52 Fluctuation factorization
✓

Theorem A. Under (H3): \(F - \mathbb {E}[F] = \delta (\delta ^{-1}(F - \mathbb {E}[F]))\). Every centered functional is representable as a divergence, providing the factorization that \(D\) inverts.

Proof ▶

By hypothesis \(F - \mathbb {E}[F] \in \mathrm{im}(\delta )\), so the recovery map gives \(v\) with \(\delta (v) = F - \mathbb {E}[F]\). Apply recovery_map_spec.

Theorem 53 Duality recovery
✓

\(\langle DF, u \rangle = \mathbb {E}[\delta (v) \cdot \delta (u)]\) where \(v = \delta ^{-1}(F - \mathbb {E}[F])\). This expresses the derivative purely in terms of the divergence operator, closing the duality loop.

Proof ▶

Apply the defining identity \(\langle DF, u \rangle = \mathbb {E}[F \cdot \delta (u)]\).

3.4 Representability Obstruction

The even-annihilation argument shows that centered, even, nonzero functionals cannot lie in \(\mathrm{im}(\delta _L)\), providing the abstract engine for the main non-representability theorem.

Definition 54 Distributional symmetry
✓
#

\(L \stackrel{d}{=} -L\): the negation map preserves even functionals while flipping \(\delta _L(u)\). This symmetry is the structural reason that even functionals are invisible to the first chaos.

Theorem 55 Even annihilates \(\mathrm{im}(\delta )\)
✓

If \(G\) is even, then \(\mathbb {E}[G \cdot \delta (u)] = 0\) for all \(u\). This vanishing is the mechanism by which distributional symmetry produces the obstruction.

Proof ▶

Under \(L \mapsto -L\): \(G\) is invariant, \(\delta (u)\) flips sign. So \(\mathbb {E}[G \cdot \delta (u)] = \mathbb {E}[G \cdot (-\delta (u))] = -\mathbb {E}[G \cdot \delta (u)]\), forcing the value to \(0\). Uses holder_neg and sym.neg_delta.

Theorem 56 Representability obstruction
✓

Proposition 5.1. Centered, even, nonzero \(G \Rightarrow G \notin \mathrm{im}(\delta _L)\). This is the abstract engine of the main theorem [1, Proposition 5.1]: three checkable properties yield non-representability.

Proof ▶

By contradiction: assume \(G = \delta (u)\). Then \(\| G\| _2^2 = \mathbb {E}[G \cdot G] = \mathbb {E}[G \cdot \delta (u)] = 0\) by even annihilation. So \(G = 0\), contradicting \(G \neq 0\).

3.5 Leibniz Defect (Theorem B)

The Leibniz defect \(\Gamma (F,G)\) measures the failure of the product rule for \(D\), which is shown to be entirely determined by the jump structure of the driving process.

Definition 57 Leibniz defect
✓
#

\(\Gamma (F,G) := D(FG) - F \cdot DG - G \cdot DF \in H_X^*\): measures failure of the product rule. \(\Gamma = 0\) iff \(D\) is a derivation, which holds precisely when the driving process has no jumps.

Theorem 58 Product rule with jump defect
✓

Theorem B. \(\langle \Gamma (F,G), u \rangle = \mathbb {E}[\sum _s \Delta _s(f,g) \cdot \delta (u)]\). The defect is entirely determined by the jump structure of the process [1, Theorem B].

Proof ▶

Apply the Itô formula to \(fg\), multiply by \(\delta (u)\), take expectations. The Leibniz part cancels, leaving the jump residual \(\sum _s \Delta _s(f,g)\). Uses product_ito_paired_derived.

Theorem 59 Defect vanishes without jumps
✓

No jumps \(\Rightarrow \) \(\Gamma = 0\): the classical product rule for continuous processes. This recovers the Hilbert setting as a special case, confirming consistency with the companion formalization.

Proof ▶

If \(\sum _s \Delta _s(f,g) = 0\), the product rule identity reduces to \(\langle \Gamma , u \rangle = 0\) for all \(u\), giving \(\Gamma = 0\) by nondegeneracy.

3.6 Chaos Characterization (Theorem C)

The kernel of \(D_L\) is characterized as the space of functionals whose chaos projections vanish, showing that even kernels and higher chaoses make \(\ker (D_L)\) infinite-dimensional.

Theorem 60 Chaos kernel characterization
✓

Theorem C. \(D_L F = 0\) on deterministic integrands iff \(\Phi _M \to 0\) as \(M \to \infty \). Even kernels and higher chaoses satisfy the vanishing condition, making \(\ker (D_L)\) infinite-dimensional. This characterization shows that the kernel of \(D_L\) captures precisely the chaos-orthogonal functionals [1, Theorem C]; cf. [9, 10] for the Poisson chaos expansion.

Proof ▶

The \(L^2\) isometry reduces \(\mathbb {E}[F \cdot \delta _L^M(u)] = \int u \cdot \Phi _M\, ds\). Truncation convergence passes \(M \to \infty \). The conclusion follows from nondegeneracy of the \(L^q\)-\(L^p\) pairing.

3.7 Stable Lévy Instantiation

The abstract Banach framework is instantiated for symmetric \(\gamma \)-stable processes, assembling the three witness properties to obtain the central non-representability result.

Definition 61 Stable noise data
✓
#

Packages \(\gamma \in (1,2)\), distributional symmetry, jump count \(G_0\), rate \(\lambda {\gt} 0\), and terminal time \(T {\gt} 0\). This bundle collects all parameters needed to instantiate the abstract obstruction for the stable case.

Theorem 62 Variance positive
✓
#

\(\mathrm{Var}(G_0) = \lambda T {\gt} 0\). Positive variance is the quantitative input that prevents the centered witness from being trivially zero.

Proof ▶

Apply mul_pos to \(\lambda {\gt} 0\) (from jump_rate) and \(T {\gt} 0\) (from hT_pos).

Theorem 63 Centered witness even
✓
#

\(\mathrm{center}(G_0)\) is even: \(G_0\) depends on \(|\Delta L_s|\) (invariant under \(L \mapsto -L\)), and centering preserves evenness. This evenness is inherited from the symmetry of the Lévy measure established in Theorem 2.

Proof ▶

\(G_0\) is even by jump_count_factors_through_abs and const_even. Centering preserves evenness: \(\mathbb {E}[G_0]\) is constant (hence even), and the difference of even functions is even (by neg_sub).

Theorem 64 Centered witness nonzero
✓
#

\(\mathrm{center}(G_0) \neq 0\): \(\| \mathrm{center}(G_0)\| _2^2 = \lambda T {\gt} 0\). The positive variance from the Poisson distribution ensures the witness is genuinely nontrivial.

Proof ▶

\(\lambda T {\gt} 0\) implies \(\mathrm{Var}(G_0) {\gt} 0\), so the \(L^2\) norm is nonzero, hence \(\mathrm{center}(G_0) \neq 0\).

Theorem 65 Stable obstruction
✓

\(G_0 - \mathbb {E}[G_0] \notin \mathrm{im}(\delta _L)\): the paper’s central result. The three prerequisites—centered, even, nonzero—are verified for the stable Lévy jump count, completing the argument.

Proof ▶

Assemble the three properties—centered (center_centered), even (jump_count_center_even), nonzero (jump_count_center_nonzero)—and apply representability_obstruction.

3.8 Hilbert Bridge

The Banach framework is shown to specialize to the Hilbert case when no jumps are present, confirming backward compatibility with the companion formalization.

Theorem 66 Leibniz defect vanishes
✓

In the Hilbert case (no jumps), \(\Gamma = 0\): the Banach framework specializes to \(p = q = 2\), \(H_X^* \cong H_X\) via Riesz, recovering the companion Hilbert formalization. This confirms that the Banach generalization is a strict extension of the original theory.

Proof ▶

Apply defect_vanishes_without_jumps with the no_jumps hypothesis from HilbertEnergyData.