1 Poisson Infrastructure
This chapter develops the measure-theoretic foundations for symmetric \(\gamma \)-stable Lévy processes. We define the stable Lévy density \(\nu _\gamma \), prove its symmetry and moment properties, and establish the Poisson identities (mean and variance) that underpin the obstruction argument. The chapter culminates in the canonical Poisson construction: an explicit probability space \((\mathbb {N}, \mathrm{poissonMeasure}(\lambda T))\) on which the large-jump count \(G_0\) is realized as a concrete random variable with a proved (not axiomatized) Poisson distribution.
1.1 Stable Lévy Measure
The stable density \(\nu _\gamma \) and its basic properties (symmetry, nonnegativity) are established here.
The symmetric \(\gamma \)-stable Lévy measure density is \(\nu _\gamma (dz) = c_\gamma |z|^{-1-\gamma }\, dz\) on \(\mathbb {R} \setminus \{ 0\} \), where \(c_\gamma {\gt} 0\) and \(\gamma \in (1,2)\). The singularity at the origin is strong enough that \(\int |z|^2\, \nu _\gamma (dz) = \infty \), forcing the energy space to be Banach rather than Hilbert. This density is the starting point for the entire Lévy–Itô construction and all moment estimates that follow; see [4, 5] for background on stable processes.
\(\nu _\gamma (-z) = \nu _\gamma (z)\) for all \(z \neq 0\), from the absolute value in \(c_\gamma |z|^{-1-\gamma }\). Symmetry of \(\nu _\gamma \) implies \(L \stackrel{d}{=} -L\), the key ingredient in the even-annihilation argument. Without this property, the obstruction witness could not be shown to lie outside \(\mathrm{im}(\delta _L)\).
Unfold the definition and use \(|-z| = |z|\). Closed by simp [abs_neg].
\(c_\gamma |z|^{-1-\gamma } {\gt} 0\) for \(z \neq 0\) when \(c_\gamma {\gt} 0\), ensuring \(\nu _\gamma \) is a valid Lévy measure. Nonnegativity is required for the measure-theoretic construction of compound Poisson integrals.
Product of nonneg factors: \(c_\gamma \geq 0\) by hypothesis and \(|z|^{-1-\gamma } \geq 0\) by rpow_nonneg.
1.2 Moment Computations
We compute the large-jump rate and the small-jump moment integrals that determine whether the energy space is Hilbert or Banach.
\(\lambda = \nu _\gamma (\{ |z| {\gt} 1\} ) = 2c_\gamma /\gamma {\gt} 0\). Positivity ensures the Poisson parameter \(\lambda T {\gt} 0\), giving positive variance for the large-jump count witness. This rate controls the intensity of the compound Poisson component in the Lévy–Itô decomposition.
Apply div_pos to \(2c_\gamma {\gt} 0\) (from mul_pos two_pos hc) and \(\gamma {\gt} 0\) (from \(\gamma {\gt} 1\)).
For \(\alpha {\gt} \gamma \): \(\int _{|z|\leq 1} |z|^\alpha \, \nu _\gamma (dz) = 2c_\gamma /(\alpha - \gamma ) {\lt} \infty \). In particular, the second moment (\(\alpha = 2 {\gt} \gamma \)) is finite, enabling the \(L^2\) isometry for truncated compensated integrals. This finiteness is what makes the compensated integral well-defined in \(L^2\).
Apply div_pos to \(2c_\gamma {\gt} 0\) and \(\alpha - \gamma {\gt} 0\) (from linarith).
For \(\alpha {\lt} \gamma \): \(\int _{|z|\leq 1} |z|^\alpha \, \nu _\gamma (dz) = \infty \). This divergence forces the integrand space \(H_L = L^p\) with \(p {\lt} \gamma {\lt} 2\) into the Banach setting, ruling out the standard Hilbert-space energy framework.
The exponent \(\alpha - \gamma {\lt} 0\) by linarith.
1.3 Poisson Identities
This section derives the Poisson mean, second factorial moment, and variance from a single recurrence relation.
\((k+1) \cdot P_r(k+1) = r \cdot P_r(k)\) for \(P_r(k) = e^{-r} r^k / k!\). This recurrence allows reindexing \(\sum k P_r(k)\) into \(r \sum P_r(k) = r\), and is the key tool for proving both the mean and second factorial moment.
\(\sum _{n=0}^{\infty } n \cdot P_r(n) = r\). Used in the variance computation and centering of the obstruction witness. This classical identity is derived here from the recurrence to keep the development self-contained.
Use the recurrence to show \(\texttt{HasSum}(\lambda k.\; (k+1)P_r(k+1), r)\) via simp_rw [poissonPMFReal_succ_mul]. Shift the index to get \(\texttt{HasSum}(\lambda n.\; n \cdot P_r(n), r)\) via comp_nat_add, split off the \(n=0\) term (which is zero), and conclude with tsum_eq.
\(\sum n(n-1) P_r(n) = r^2\), proved by applying the recurrence twice. Combined with the mean, this yields \(\mathrm{Var} = r^2 + r - r^2 = r\), completing the variance calculation.
Apply the double recurrence \((k+2)(k+1)P_r(k+2) = r^2 P_r(k)\) to get \(\texttt{HasSum}(\lambda k.\; r^2 P_r(k), r^2)\), then shift by 2 and note that the \(n=0\) and \(n=1\) terms vanish.
\(\mathrm{Var}(\mathrm{Poisson}(r)) = r\). Positive variance \(\lambda T {\gt} 0\) implies the centered witness is nonzero, which is essential for the representability obstruction.
Decompose \((n-r)^2 P_r(n) = n(n-1)P_r(n) + (1-2r)\cdot nP_r(n) + r^2 P_r(n)\) pointwise (by ring). Combine three HasSum instances via HasSum.add: the values sum to \(r^2 + (1-2r)r + r^2 = r\) (by ring). Use convert with ext n; ring for the summand and ring for the value.
1.4 Quadratic Defect
The exact quadratic Taylor remainder is computed, showing that the jump defect for \(f(x) = x^2\) is sharp and cannot be improved.
\((V + Kz)^2 - V^2 - 2V(Kz) = (Kz)^2\): the Taylor remainder for \(f(x) = x^2\) is exact. Serves as the prototype for the jump defect that drives the Leibniz failure in the Banach setting.
Expand and cancel. Closed by ring.
\(|g_f(V,K,z)| = (Kz)^2\) for \(f(x) = x^2\): the Taylor bound is achieved with equality. This sharpness shows that the defect estimate cannot be improved for quadratic nonlinearities.
Rewrite via quadratic_defect_exact, then apply abs_of_nonneg (sq_nonneg _).
1.5 Canonical Poisson Construction
An explicit probability space \((\mathbb {N}, \mathrm{poissonMeasure}(\lambda T))\) is constructed to realize the large-jump count as a concrete random variable with a proved Poisson distribution.
Constructs the canonical Poisson space \((\mathbb {N}, \mathrm{poissonMeasure}(\lambda T))\) with \(G_0(n) = n\) as the large-jump count. Provides an explicit probability space for the obstruction witness, bridging abstract existence to concrete computation.
\(\mathrm{Measure.map}\; G_0\; \mu = \mathrm{poissonMeasure}(\lambda T)\), trivially by \(\mathrm{Measure.map\_ id}\) since \(G_0 = \mathrm{id}\). This confirms that the concrete construction realizes the desired Poisson distribution.
Unfold \(G_0 = \mathrm{id}\) and \(P = \mathrm{poissonMeasure}(\lambda T)\); apply Measure.map_id.