1 Volterra Kernel and Combined Energy Space
This layer defines the causal kernel \(K\), the combined energy space \(\mathcal{H}_{VL} = H_W \times \tilde{H}_J\), the combined divergence \(\delta _{VL}\), and proves the splitting \(D_{VL} = (D_W, \tilde{D}_J)\). The splitting is the key structural fact: \(D\) inherits its decomposition from the energy space, not from any Gaussian assumption [1].
1.1 The Volterra Kernel
VolterraKernel A causal kernel \(K : [0,T]^2 \to \mathbb {R}\) with \(K(t,s) = 0\) for \(s {\gt} t\), terminal time \(T {\gt} 0\), and \(L^2\)-integrability: \(\int _0^t K(t,s)^2\, ds {\lt} \infty \) for all \(t \in [0,T]\). The Volterra–Lévy process is \(V_t = \int _0^t K(t,s)\, dL_s\), where \(L\) is a Lévy driver with both continuous and jump components [1, Definition 3.1].
CombinedEnergySpace.fractionalKernel \(K_H(t,s) = (t-s)^{H-1/2}\) for \(s {\lt} t\), zero otherwise. The fractional kernel with Hurst parameter \(H\) governs rough volatility: \(H {\lt} 1/2\) gives rougher-than-Brownian paths, \(H = 1/2\) recovers the Markov (classical) case [1, Example 2.2].
CombinedEnergySpace.fractionalKernel_causal def:fractionalKernel \(K_H(t,s) = 0\) for \(s {\gt} t\). Causality: the kernel looks backward only.
By definition: the if branch is false when \(s \geq t\).
CombinedEnergySpace.fractionalKernel_nonneg def:fractionalKernel \(K_H(t,s) \geq 0\) for \(H \geq 1/2\) and \(s {\lt} t\).
From Real.rpow_nonneg applied to \(t - s {\gt} 0\).
1.2 The Combined Energy Space
CombinedEnergySpace The product space \(\mathcal{H}_{VL} = H_W \times \tilde{H}_J\) where \(H_W = L^2([0,T] \times \Omega , \mathcal{P}, \sigma ^2 ds \otimes d\mathbb {P})\) is Hilbert (continuous integrands, with divergence \(\delta _W(u) = \sigma \int u_t\, dW_t\)) and \(\tilde{H}_J = L^p([0,T] \times \mathbb {R} \times \Omega , dt \otimes \nu (dz) \otimes d\mathbb {P})\) is Banach (jump integrands, with divergence \(\tilde{\delta }_J(h) = \int \! \! \int h(t,z)\, \tilde{N}(dt,dz)\)). Carries centering conditions, expectation linearity, Hölder and duality pairings [1, Definition 4.1].
CombinedEnergySpace.δ_VL def:CombinedEnergySpace \(\delta _{VL}(u,h) := \delta _W(u) + \tilde{\delta }_J(h)\). The combined stochastic integral acts componentwise on \(H_W \times \tilde{H}_J\). This is a definition, not an axiom.
CombinedEnergySpace.δ_VL_centered def:delta_VL \(\mathbb {E}[\delta _{VL}(u,h)] = 0\). Both stochastic integrals are centered (\(\mathbb {E}[\delta _W(u)] = 0\), \(\mathbb {E}[\tilde{\delta }_J(h)] = 0\)), and expectation distributes over addition. ALGEBRAIC: from expect_add, centered_W, centered_J.
simp [δ_VL, expect_add, centered_W, centered_J, add_zero].
CombinedEnergySpace.D_W def:CombinedEnergySpace \(D_W F\) is defined by \(\langle D_W F, u \rangle := \mathbb {E}[F \cdot \delta _W(u)]\), the \(D = \delta ^*\) construction restricted to the Hilbert component \(H_W\) [1, Definition 4.2]. This is the continuous Malliavin-type derivative.
CombinedEnergySpace.D_J def:CombinedEnergySpace \(\tilde{D}_J F\) is defined by \(\langle \tilde{D}_J F, h \rangle := \mathbb {E}[F \cdot \tilde{\delta }_J(h)]\), the \(D = \delta ^*\) construction restricted to the Banach component \(\tilde{H}_J\) [1, Definition 4.2]. This is the jump operator derivative (a Banach-space analogue of the Picard difference operator).
CombinedEnergySpace.D_VL_split def:D_W, def:D_J, def:delta_VL \(\langle D_W F, u \rangle + \langle \tilde{D}_J F, h \rangle = \mathbb {E}[F \cdot \delta _{VL}(u,h)]\). The combined operator derivative splits automatically into continuous and jump components, inherited from the energy space decomposition. This is a theorem, not an assumption—the splitting is forced by the linearity of the Hölder pairing [1, Proposition 3.4]. ALGEBRAIC: from mk_dual specs and holder_add_right.
Unfold \(D_W\), \(D_J\), \(\delta _{VL}\) via simp only, then apply holder_add_right.