Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

§4.3, p. 477 — ‖Σ α_iΦ(x_i)‖² = Σ α_iα_j k(x_i, x_j), hence F ⊆ {x ↦ ⟨w, Φ(x)⟩ : ‖w‖ ≤ B}

Proved
RadGauss.Kernel.feature_map_inclusion

by mikedeng1 · Oct 5, 2026 · Mathlib 0df444a (Lean v4.33.1)

kernel-methodslearning-theoryp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1

Let kkk be a kernel on a compact space X\mathcal XX, and let Φ:X→H\Phi : \mathcal X \to \mathcal HΦ:X→H be a feature map of kkk: H\mathcal HH is a real Hilbert space with inner product ⟨⋅,⋅⟩\langle \cdot, \cdot \rangle⟨⋅,⋅⟩ and norm ∥⋅∥\|\cdot\|∥⋅∥, and k(x1,x2)=⟨Φ(x1),Φ(x2)⟩k(x_1, x_2) = \langle \Phi(x_1), \Phi(x_2) \ranglek(x1​,x2​)=⟨Φ(x1​),Φ(x2​)⟩ for all x1,x2∈Xx_1, x_2 \in \mathcal Xx1​,x2​∈X. Let B≥0B \ge 0B≥0 and let FFF be the class of kernel expansions x↦∑i=1mαik(x,xi)x \mapsto \sum_{i=1}^m \alpha_i k(x, x_i)x↦∑i=1m​αi​k(x,xi​) with ∑i,jαiαjk(xi,xj)≤B2\sum_{i,j}\alpha_i\alpha_j k(x_i, x_j) \le B^2∑i,j​αi​αj​k(xi​,xj​)≤B2. Then

  1. for every mmm, all x1,…,xm∈Xx_1, \dots, x_m \in \mathcal Xx1​,…,xm​∈X and all α1,…,αm∈R\alpha_1, \dots, \alpha_m \in \mathbb Rα1​,…,αm​∈R,
∥∑i=1mαiΦ(xi)∥2=∑i,jαiαj k(xi,xj);\Big\|\sum_{i=1}^m \alpha_i \Phi(x_i)\Big\|^2 = \sum_{i,j} \alpha_i \alpha_j\, k(x_i, x_j);​i=1∑m​αi​Φ(xi​)​2=i,j∑​αi​αj​k(xi​,xj​);
  1. consequently
F⊆{ x↦⟨w,Φ(x)⟩  :  w∈H, ∥w∥≤B }.F \subseteq \{\, x \mapsto \langle w, \Phi(x) \rangle \;:\; w \in \mathcal H,\ \|w\| \le B \,\}.F⊆{x↦⟨w,Φ(x)⟩:w∈H, ∥w∥≤B}.

The inclusion places the kernel class inside a ball of linear functionals of the feature map, which is how its complexity is bounded in Lemma 22.

Formalization Note The feature map is a hypothesis (any real Hilbert space H\mathcal HH and any Φ\PhiΦ with k=⟨Φ(⋅),Φ(⋅)⟩k = \langle\Phi(\cdot), \Phi(\cdot)\ranglek=⟨Φ(⋅),Φ(⋅)⟩); its existence for every kernel is the cited RKHS theorem, available on the platform as FoundationsML.Kernels.RKHS_exists. The paper takes B>0B > 0B>0; here B≥0B \ge 0B≥0 (for B<0B < 0B<0 the right-hand set is empty while FFF contains the zero function, so a sign condition is needed).

Preamble
import Mathlib
import Definitions.Def_RadGauss_Kernel_KernelClass
Formal statement
namespace RadGauss.Kernel

/-- §4.3, p. 477 (Bartlett–Mendelson 2002), the feature-map displays. Let `k` be a kernel on `𝒳`
with feature map `Φ : 𝒳 → ℋ` into a real Hilbert space, `k(x₁, x₂) = ⟨Φ(x₁), Φ(x₂)⟩`. Then
`‖Σ_i α_i Φ(x_i)‖² = Σ_{i,j} α_i α_j k(x_i, x_j)` for every finite family, and hence, for `B ≥ 0`,
`F ⊆ {x ↦ ⟨w, Φ(x)⟩ : ‖w‖ ≤ B}`. The paper takes `B > 0`; `0 ≤ B` is assumed here. -/
theorem feature_map_inclusion {X H : Type*} [TopologicalSpace X]
    [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H]
    (k : X → X → ℝ) (hk : IsKernel k) (Φ : X → H)
    (hΦ : ∀ x₁ x₂, k x₁ x₂ = inner ℝ (Φ x₁) (Φ x₂)) (B : ℝ) (hB : 0 ≤ B) :
    (∀ (m : ℕ) (x : Fin m → X) (α : Fin m → ℝ),
        ‖∑ i, α i • Φ (x i)‖ ^ 2 = ∑ i, ∑ j, α i * α j * k (x i) (x j)) ∧
      kernelClass k B ⊆ {f | ∃ w : H, ‖w‖ ≤ B ∧ f = fun x => inner ℝ w (Φ x)} := by sorry

end RadGauss.Kernel
Source
Bartlett, Mendelson, Rademacher and Gaussian Complexities: Risk Bounds and Structural Results, J. Mach. Learn. Res. 3 (2002), p. 477 (PDF p. 15), §4.3, first and second displays
Human review
  • Endorsed by Shuze Chen · Oct 5, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 5, 2026

    Confirmed by the mission captain (proposal self-audit).

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me