OctoInformBeta

OpenAI preprint, September 2026 · §5. Output-dependent truncations and energy jumps

Lemma 5.1: Finite Vitali selection; weak L1L^1 and 18218\sqrt{2} L3/2L^{3/2} bounds for nonnegative functions on ℤ\mathbb{Z} with 2e1/22e^{1/2} Gaussian domination for σ≥h2>0\sigma \ge h^2 > 0

Discrete maximal estimates and Gaussian domination on finite sets.

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

Given a finite family of nonempty sets with real weights, one seeks a pairwise disjoint subcollection that still intersects every member of the family at no loss in weight. For a function on ℤ\mathbb{Z} nonnegative on a finite subset D⊂ℤD \subset \mathbb{Z}, the companion questions are how the level sets and L3/2L^{3/2} sums of symmetric discrete interval averages and maximal operators compare with those of the function itself, and how a Gaussian sum sampled on a grid of step h>0h > 0 with variance σ≥h2\sigma \ge h^2 is bounded pointwise by those maximal operators.

Discrete Maximal Operators and Gaussian Densities

For an integer n∈ℤn \in \mathbb{Z} and a nonnegative integer r∈ℤ≥0r \in \mathbb{Z}_{\ge 0}, the discrete interval I(n,r)⊂ℤI(n, r) \subset \mathbb{Z} of radius rr centered at nn is the finite set of integers (1): I(n,r)={m∈ℤ:n−r≤m≤n+r}.I(n, r) = \{m \in \mathbb{Z} : n - r \le m \le n + r\}.(1)✦ Given a finite set D⊂ℤD \subset \mathbb{Z} and a function f:ℤ→ℝf : \mathbb{Z} \to \mathbb{R}, the discrete interval average ADf(n,r)∈ℝA_D f(n, r) \in \mathbb{R} at center n∈ℤn \in \mathbb{Z} and radius r∈ℤ≥0r \in \mathbb{Z}_{\ge 0} is the real number (2): ADf(n,r)=12r+1∑m∈D∩I(n,r)f(m).A_D f(n, r) = \frac{1}{2r + 1} \sum_{m \in D \cap I(n, r)} f(m).(2)✦

The support radius RD(n)∈ℤ≥0R_D(n) \in \mathbb{Z}_{\ge 0} of a finite set D⊂ℤD \subset \mathbb{Z} around an integer n∈ℤn \in \mathbb{Z} is the nonnegative integer (3): RD(n)=supm∈D|m−n|,with R∅(n)=0.R_D(n) = \sup_{m \in D} |m - n|, \qquad \text{with } R_\emptyset(n) = 0.(3)✦ With D⊂ℤD \subset \mathbb{Z} finite, the discrete maximal operator MDM_D maps each function f:ℤ→ℝf : \mathbb{Z} \to \mathbb{R} to the function MDf:ℤ→ℝM_D f : \mathbb{Z} \to \mathbb{R} whose value at n∈ℤn \in \mathbb{Z} is (4): MDf(n)=supr∈ℤ≥0r≤RD(n)ADf(n,r).M_D f(n) = \sup_{\substack{r \in \mathbb{Z}_{\ge 0} \\ r \le R_D(n)}} A_D f(n, r).(4)✦

When D⊂ℤD \subset \mathbb{Z} is finite and f:D→ℝf : D \to \mathbb{R} is a real-valued function on DD, the extension by zero f̃:ℤ→ℝ\tilde{f} : \mathbb{Z} \to \mathbb{R} is the function defined by (5): f̃(n)={f(n),n∈D,0,n∉D.\tilde{f}(n) = \begin{cases} f(n), & n \in D, \\ 0, & n \notin D. \end{cases}(5)✦ The line maximal operator ℳD\mathcal{M}_D associated with a finite set D⊂ℤD \subset \mathbb{Z} sends each function f:D→ℝf : D \to \mathbb{R} to the function ℳDf:ℤ→ℝ\mathcal{M}_D f : \mathbb{Z} \to \mathbb{R} given by (6): ℳDf(n)=MDf̃(n)(n∈ℤ).\mathcal{M}_D f(n) = M_D \tilde{f}(n) \qquad (n \in \mathbb{Z}).(6)✦

For a real variance σ∈ℝ\sigma \in \mathbb{R}, the variance-σ\sigma Gaussian density gσ:ℝ→ℝg_\sigma : \mathbb{R} \to \mathbb{R} is the function defined by (7): gσ(t)=(2πσ)−1/2exp⁡(−t2/(2σ))(t∈ℝ).g_\sigma(t) = (2\pi\sigma)^{-1/2} \exp(-t^2/(2\sigma)) \qquad (t \in \mathbb{R}).(7)✦ The mesh mass constant CmeshC_{\mathrm{mesh}} is the real number 2e1/22 e^{1/2}.

The Claim

Theorem 1.

  1. Given a set XX, a finite index set ss, a family {Ba}a∈s\{B_a\}_{a \in s} of nonempty subsets of XX, and a function δ:s→ℝ\delta : s \to \mathbb{R}, there exists a subset u⊆su \subseteq s satisfying (8): Ba∩Bb=∅for all distinct a,b∈u,and for every a∈s there exists b∈u with Ba∩Bb≠∅ and δ(a)≤δ(b).\begin{aligned} & B_a \cap B_b = \emptyset \quad \text{for all distinct } a, b \in u, \\ & \text{and for every } a \in s \text{ there exists } b \in u \text{ with } B_a \cap B_b \neq \emptyset \text{ and } \delta(a) \le \delta(b). \end{aligned}(8)✦

  2. Let D,s⊂ℤD, s \subset \mathbb{Z} be finite sets, f:ℤ→ℝf : \mathbb{Z} \to \mathbb{R} be nonnegative on DD, r:ℤ→ℤ≥0r : \mathbb{Z} \to \mathbb{Z}_{\ge 0} be a radius function, and ρ≥0\rho \ge 0 satisfy ADf(n,r(n))≥ρA_D f(n, r(n)) \ge \rho for all n∈sn \in s; then the cardinality |s||s| obeys (9): ρ|s|≤3∑m∈Df(m).\rho |s| \le 3 \sum_{m \in D} f(m).(9)✦

  3. If D,s⊂ℤD, s \subset \mathbb{Z} are finite sets, f,g:ℤ→ℝf, g : \mathbb{Z} \to \mathbb{R} satisfy f≥0f \ge 0 on DD and g≥0g \ge 0 on ss, and r:ℤ→ℤ≥0r : \mathbb{Z} \to \mathbb{Z}_{\ge 0} satisfies ADf(n,r(n))≥g(n)A_D f(n, r(n)) \ge g(n) for all n∈sn \in s, then (10) holds: ∑n∈sg(n)3/2≤182∑m∈Df(m)3/2.\sum_{n \in s} g(n)^{3/2} \le 18\sqrt{2} \sum_{m \in D} f(m)^{3/2}.(10)✦

  4. For finite sets D,s⊂ℤD, s \subset \mathbb{Z} and any function f:ℤ→ℝf : \mathbb{Z} \to \mathbb{R} nonnegative on DD, the discrete maximal function satisfies (11): ∑n∈s(MDf(n))3/2≤182∑m∈Df(m)3/2.\sum_{n \in s} \bigl(M_D f(n)\bigr)^{3/2} \le 18\sqrt{2} \sum_{m \in D} f(m)^{3/2}.(11)✦

  5. Whenever D⊂ℤD \subset \mathbb{Z} is finite, f:D→ℝf : D \to \mathbb{R} is nonnegative, κ\kappa is a finite index set, and u:κ→ℤu : \kappa \to \mathbb{Z} is injective, the line maximal function admits the bound (12): ∑k∈κ(ℳDf(u(k)))3/2≤182∑m∈Df(m)3/2.\sum_{k \in \kappa} \bigl(\mathcal{M}_D f(u(k))\bigr)^{3/2} \le 18\sqrt{2} \sum_{m \in D} f(m)^{3/2}.(12)✦

  6. Take a finite set D⊂ℤD \subset \mathbb{Z}, a function f:ℤ→ℝf : \mathbb{Z} \to \mathbb{R} nonnegative on DD, real parameters h>0h > 0 and σ≥h2\sigma \ge h^2, and an integer n∈ℤn \in \mathbb{Z}; then the Gaussian sum obeys (13): ∑m∈Df(m)hgσ((m−n)h)≤CmeshMDf(n).\sum_{m \in D} f(m) h g_\sigma\bigl((m - n)h\bigr) \le C_{\mathrm{mesh}} M_D f(n).(13)✦

  7. Given a finite set D⊂ℤD \subset \mathbb{Z}, a nonnegative function f:D→ℝf : D \to \mathbb{R}, real numbers h>0h > 0 and σ≥h2\sigma \ge h^2, and an integer n∈ℤn \in \mathbb{Z}, the Gaussian sum satisfies (14): ∑m∈Df(m)hgσ((m−n)h)≤CmeshℳDf(n).\sum_{m \in D} f(m) h g_\sigma\bigl((m - n)h\bigr) \le C_{\mathrm{mesh}} \mathcal{M}_D f(n).(14)✦

In the paper, the result reads: writing MM for the centered maximal operator on either ℝ\mathbb{R} or ℤ\mathbb{Z} (denoted MℤM_{\mathbb{Z}} on ℤ\mathbb{Z}) and μ\mu for the underlying measure, μ{Mf>ρ}≤(3/ρ)∥f∥1\mu\{Mf > \rho\} \le (3/\rho)\|f\|_1 for every f∈L1f \in L^1 and ρ>0\rho > 0; ∥Mf∥qq≤(6q2q−1/(q−1))∥f∥qq\|Mf\|_q^q \le \bigl(6q 2^{q-1}/(q - 1)\bigr)\|f\|_q^q for every 1<q<∞1 < q < \infty and f∈Lqf \in L^q; and there is an absolute constant CgC_g such that ∑m∈ℤhgσ((m−n)h)|f(m)|≤CgMℤf(n)\sum_{m \in \mathbb{Z}} h g_\sigma((m - n)h) |f(m)| \le C_g M_{\mathbb{Z}} f(n) whenever h>0h > 0, σ≥h2\sigma \ge h^2, and n∈ℤn \in \mathbb{Z}.

What Is Checked

The claims (8) through (14) are stated over a finite family of nonempty subsets of a set XX with real weights δ\delta, finite subsets D,s⊂ℤD, s \subset \mathbb{Z}, functions on ℤ\mathbb{Z} nonnegative on DD or ss (with interval averages on ss bounded below by ρ\rho or gg in (9) and (10)), nonnegative functions on DD, radius functions r:ℤ→ℤ≥0r : \mathbb{Z} \to \mathbb{Z}_{\ge 0} and levels ρ≥0\rho \ge 0, discrete intervals I(n,r)I(n, r), averages ADf(n,r)A_D f(n, r), support radii RD(n)R_D(n), extensions by zero f̃\tilde{f}, discrete and line maximal operators MDM_D and ℳD\mathcal{M}_D, injective maps u:κ→ℤu : \kappa \to \mathbb{Z} on finite sets κ\kappa, and grid parameters h>0h > 0 and σ≥h2\sigma \ge h^2 with the Gaussian density gσg_\sigma and mesh mass constant CmeshC_{\mathrm{mesh}}.