OctoInformBeta

OpenAI preprint, September 2026 · §5. Homogeneous increments and signed Poisson sampling

§5 (setting): the power-law intensity and signed sums: Homogeneous integration and logarithmic moments accompany convolution powers and compound Poisson measures

Power intensities and signed jump sums.

Hover over a blue or green term in the text, or a node in the graph, to find it in the other. Select it to read its declaration both in Lean and in words.

The Setting

Signed sampling for real LpL_p spaces involves a measure on positive jump magnitudes and measures of sums of real vector-valued jumps. Let EE be the real additive space in which the jumps take values, and write *\ast for convolution of measures on EE.

The Claim

Definition 1. For p∈ℝp\in\mathbb R, the power intensity νp\nu_p is the Borel measure on ℝ\mathbb R determined by (1) for every Borel set AA, where u+=max⁡{u,0}u_+=\max\{u,0\}: νp(A)=∫A∩(0,∞)(pr−p−1)+dr.\nu_p(A)=\int_{A\cap(0,\infty)}\bigl(p r^{-p-1}\bigr)_+\,dr.(1)✦

The symmetrization μsym\mu^{\mathrm{sym}} of a measure μ\mu on ℝ\mathbb R is the measure in (2), where ι(x)=−x\iota(x)=-x and ι#μ\iota_\#\mu denotes the pushforward of μ\mu by ι\iota: μsym=12μ+12ι#μ.\mu^{\mathrm{sym}}=\frac12\mu+\frac12\iota_\#\mu.(2)✦

The convolution powers μ*n\mu^{*n} of a measure μ\mu on EE are the measures specified by (3), with δ0\delta_0 the unit point mass at zero: μ*0=δ0,μ*(n+1)=μ*n*μ(n≥0).\mu^{*0}=\delta_0,\qquad \mu^{*(n+1)}=\mu^{*n}\ast\mu\quad(n\ge0).(3)✦

The compound Poisson measure Ct(μ)C_t(\mu) on EE, with rate t≥0t\ge0 and jump measure μ\mu, is the measure defined by (4): Ct(μ)=∑n=0∞e−ttnn!μ*n.C_t(\mu)=\sum_{n=0}^{\infty}e^{-t}\frac{t^n}{n!}\,\mu^{*n}.(4)✦

Theorem 1. Let p>0p>0. For any function g:ℝ→Fg\colon\mathbb R\to F, where FF is a real normed vector space, integration against νp\nu_p is given by (5): ∫ℝg(r)dνp(r)=∫(0,∞)pr−p−1g(r)dr.\int_{\mathbb R}g(r)\,d\nu_p(r) =\int_{(0,\infty)}p r^{-p-1}g(r)\,dr.(5)✦ For a>0a>0, the tail is given by (6): νp((a,∞))=a−p.\nu_p\bigl((a,\infty)\bigr)=a^{-p}.(6)✦ For c∈ℝc\in\mathbb R and an even function f:ℝ→ℝf\colon\mathbb R\to\mathbb R with f(0)=0f(0)=0, the scaling identity is (7): ∫ℝf(cr)dνp(r)=|c|p∫ℝf(r)dνp(r).\int_{\mathbb R}f(cr)\,d\nu_p(r) =|c|^p\int_{\mathbb R}f(r)\,d\nu_p(r).(7)✦ For 0<a≤b0<a\le b, the critical truncated moment is (8): ∫(a,b]rpdνp(r)=plog⁡ba.\int_{(a,b]}r^p\,d\nu_p(r)=p\log\frac ba.(8)✦

What Is Checked

The measure constructions are stated over arbitrary jump measures and nonnegative rates; the integral identities concern positive exponents and real normed target spaces, with even scalar functions vanishing at zero in the scaling identity.