Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 3.28 — Gagliardo–Nirenberg–Sobolev inequality with an explicit constant

Proved
HunterPDE.Sobolev.gagliardo_nirenberg_sobolev

by mikedeng1 · Sep 28, 2026 · Mathlib 0df444a (Lean v4.33.1)

p2o-batch-books5p2o-gran-per-chapterp2o-plan-bookp2o-v1pdesobolev-inequalitysobolev-spaces

Let n≥2n \ge 2n≥2, 1≤p<n1 \le p < n1≤p<n, and let p∗=np/(n−p)p^* = np/(n-p)p∗=np/(n−p) be the Sobolev conjugate of ppp (Definition 3.25). Then for every f∈Cc∞(Rn)f \in C_c^\infty(\mathbb{R}^n)f∈Cc∞​(Rn)

∥f∥p∗≤p (n−1)2 (n−p) ∥Df∥p,\|f\|_{p^*} \le \frac{p\,(n-1)}{2\,(n-p)}\, \|Df\|_p ,∥f∥p∗​≤2(n−p)p(n−1)​∥Df∥p​,

where ∥Df∥p=(∫Rn∣Df∣p dx)1/p\|Df\|_p = \big(\int_{\mathbb{R}^n} |Df|^p\, dx\big)^{1/p}∥Df∥p​=(∫Rn​∣Df∣pdx)1/p and ∣Df∣=(∑i(∂if)2)1/2|Df| = \big(\sum_i (\partial_i f)^2\big)^{1/2}∣Df∣=(∑i​(∂i​f)2)1/2 is the Euclidean length of the gradient.

The inequality says that a compactly supported smooth function whose gradient is in LpL^pLp is itself in Lp∗L^{p^*}Lp∗, with p∗>pp^* > pp∗>p: one derivative in LpL^pLp buys integrability p∗p^*p∗. It underlies every Sobolev embedding for p<np < np<n, e.g. W1,p(Rn)↪Lq(Rn)W^{1,p}(\mathbb{R}^n) \hookrightarrow L^q(\mathbb{R}^n)W1,p(Rn)↪Lq(Rn) for p≤q≤p∗p \le q \le p^*p≤q≤p∗ (Theorem 3.31).

Formalization Note. The constant is nnn times the constant C(n,p)=p2n⋅n−1n−pC(n,p) = \frac{p}{2n}\cdot\frac{n-1}{n-p}C(n,p)=2np​⋅n−pn−1​ printed in (3.11). With the Euclidean ∣Df∣|Df|∣Df∣ the printed constant is false: at p=1p = 1p=1 it is 12n\frac1{2n}2n1​, below the sharp constant 1nαn1/n\frac{1}{n\alpha_n^{1/n}}nαn1/n​1​ the notes themselves quote on p. 65–66 (αn\alpha_nαn​ the volume of the unit ball), and it is also below Talenti's sharp constant for some 1<p<n1 < p < n1<p<n (e.g. n=4n = 4n=4, p=1.2p = 1.2p=1.2). The error is in the last step of the proof, which replaces (∏i∥∂if∥p)1/n(\prod_i \|\partial_i f\|_p)^{1/n}(∏i​∥∂i​f∥p​)1/n by 1n(∑i∥∂if∥pp)1/p\frac1n (\sum_i \|\partial_i f\|_p^p)^{1/p}n1​(∑i​∥∂i​f∥pp​)1/p. The proof's display ∥f∥p∗≤s2(∏i∥∂if∥p)1/n\|f\|_{p^*} \le \frac{s}{2} (\prod_{i} \|\partial_i f\|_p)^{1/n}∥f∥p∗​≤2s​(∏i​∥∂i​f∥p​)1/n with s=p(n−1)n−ps = \frac{p(n-1)}{n-p}s=n−pp(n−1)​, which is correct, together with ∣∂if∣≤∣Df∣|\partial_i f| \le |Df|∣∂i​f∣≤∣Df∣, gives the constant s2\frac s22s​ stated here. Rn\mathbb{R}^nRn is EuclideanSpace ℝ (Fin n); ∣Df∣|Df|∣Df∣ is ‖gradient f x‖, which equals the operator norm of fderiv ℝ f x; norms are eLpNorm in [0,∞][0,\infty][0,∞] with exponents ENNReal.ofReal p and ENNReal.ofReal p*; Cc∞C_c^\inftyCc∞​ is ContDiff ℝ ((⊤ : ℕ∞) : WithTop ℕ∞) f ∧ HasCompactSupport f.

Preamble
import Mathlib
import Definitions.Def_HunterPDE_Sobolev_SobolevConjugate

open MeasureTheory
Formal statement
namespace HunterPDE.Sobolev

/-- Theorem 3.28 of Hunter, *Notes on PDEs* (revised 6/18/2014), p. 63 (Gagliardo–Nirenberg–Sobolev
inequality), with the constant its proof establishes. Let `n ≥ 2`, `1 ≤ p < n` and
`p* = np/(n − p)` (Definition 3.25). Then for every `f ∈ C_c^∞(ℝⁿ)`
`‖f‖_{p*} ≤ C ‖Df‖_p` with `C = p(n − 1) / (2(n − p))`, where `|Df|` is the Euclidean norm of the
gradient. The printed constant (3.11), `p(n − 1)/(2n(n − p))`, is `1/n` times this and is false
with the Euclidean `|Df|` (it is below the book's own sharp constant `1/(n αₙ^{1/n})` at `p = 1`);
the proof's display `‖f‖_{p*} ≤ (s/2) (∏ᵢ ‖∂ᵢf‖_p)^{1/n}`, `s = p(n−1)/(n−p)`, together with
`|∂ᵢf| ≤ |Df|`, gives the constant stated here. -/
theorem gagliardo_nirenberg_sobolev {n : ℕ} (hn : 2 ≤ n) {p : ℝ} (hp : 1 ≤ p) (hpn : p < n)
    (f : EuclideanSpace ℝ (Fin n) → ℝ) (hf : ContDiff ℝ ((⊤ : ℕ∞) : WithTop ℕ∞) f)
    (hfc : HasCompactSupport f) :
    eLpNorm f (ENNReal.ofReal (sobolevConjugate n p)) volume ≤
      ENNReal.ofReal (p * ((n : ℝ) - 1) / (2 * ((n : ℝ) - p))) *
        eLpNorm (fun x => ‖gradient f x‖) (ENNReal.ofReal p) volume := by sorry

end HunterPDE.Sobolev
Source
Hunter, Notes on Partial Differential Equations (revised 6/18/2014), p. 63, Theorem 3.28 (constant corrected; see the Formalization Note)
Read-back

What the Lean code literally says, in plain math · claude-opus-5-5

The statement concerns a natural number nnn, a real number ppp and a function f:Rn→Rf : \mathbb{R}^n \to \mathbb{R}f:Rn→R, where Rn\mathbb{R}^nRn is the Euclidean space of real nnn-tuples with the Euclidean (ℓ2\ell^2ℓ2) norm and inner product. The hypotheses are:

  • n≥2n \ge 2n≥2;
  • 1≤p1 \le p1≤p and p<np < np<n, with nnn viewed as a real number;
  • fff is infinitely differentiable (C∞C^\inftyC∞) on all of Rn\mathbb{R}^nRn;
  • fff has compact support, meaning the closure of {x:f(x)≠0}\{x : f(x) \neq 0\}{x:f(x)=0} is compact.

What it asserts. Write p∗p^\astp∗ for the real number sobolevConjugate(n,p)\mathrm{sobolevConjugate}(n, p)sobolevConjugate(n,p). The conclusion is

∥f∥Lp∗(Rn)  ≤  p (n−1)2 (n−p)⋅∥ ∣∇f∣ ∥Lp(Rn).\|f\|_{L^{p^\ast}(\mathbb{R}^n)} \;\le\; \frac{p\,(n-1)}{2\,(n-p)} \cdot \big\|\, |\nabla f| \,\big\|_{L^{p}(\mathbb{R}^n)} .∥f∥Lp∗(Rn)​≤2(n−p)p(n−1)​⋅​∣∇f∣​Lp(Rn)​.

The pieces of this inequality are as follows.

  • The norms. Both are extended-valued LqL^qLq norms with respect to Lebesgue measure on Rn\mathbb{R}^nRn. For 0<q<∞0 < q < \infty0<q<∞ this is ∥g∥Lq=(∫Rn∣g(x)∣q dx)1/q\|g\|_{L^q} = \left(\int_{\mathbb{R}^n} |g(x)|^q\,dx\right)^{1/q}∥g∥Lq​=(∫Rn​∣g(x)∣qdx)1/q, and it may equal +∞+\infty+∞.
  • The exponents. Each exponent is passed through the map t↦max⁡(t,0)t \mapsto \max(t, 0)t↦max(t,0) into [0,∞)[0, \infty)[0,∞) before use.
  • The gradient term. ∇f(x)∈Rn\nabla f(x) \in \mathbb{R}^n∇f(x)∈Rn is the gradient of fff at xxx, and ∣∇f(x)∣|\nabla f(x)|∣∇f(x)∣ is its Euclidean length. The right-hand norm is therefore that of the scalar function x↦∣∇f(x)∣x \mapsto |\nabla f(x)|x↦∣∇f(x)∣.
  • The constant. p(n−1)2(n−p)\frac{p(n-1)}{2(n-p)}2(n−p)p(n−1)​ is likewise passed through max⁡(⋅,0)\max(\cdot, 0)max(⋅,0). Under the hypotheses it is strictly positive, so that map does not change it.
  • The arithmetic. The product and the inequality are taken in the extended non-negative reals [0,+∞][0, +\infty][0,+∞].

The exponent p∗p^\astp∗. Its definition lives in an imported module that was not provided to me, so I cannot unfold it. The statement says nothing more about p∗p^\astp∗ than that it is the value of that imported function at (n,p)(n, p)(n,p). In particular, the conclusion is only as strong as that definition makes it.

Degenerate cases.

  • Dimensions. n=0n = 0n=0 and n=1n = 1n=1 are excluded by n≥2n \ge 2n≥2. Since n≥2n \ge 2n≥2, the range 1≤p<n1 \le p < n1≤p<n is non-empty; for example p=1p = 1p=1 always satisfies it. The hypotheses are therefore satisfiable.
  • The zero function. f≡0f \equiv 0f≡0 is allowed. Both sides are then 000, and the claim holds trivially.
  • Non-positive p∗p^\astp∗. If the imported definition gives p∗≤0p^\ast \le 0p∗≤0 for some admissible (n,p)(n, p)(n,p), the left-hand exponent becomes 000. By the library's convention an L0L^0L0 "norm" is 000, so the inequality would then hold trivially.
  • Infinite exponent. The exponent +∞+\infty+∞ cannot arise on either side, because each exponent is a real number mapped to [0,∞)[0, \infty)[0,∞).
  • The right-hand exponent. Since p≥1p \ge 1p≥1, the right-hand exponent is exactly ppp.
  • The gradient's default value. The gradient's default value 000, used where fff is not differentiable, never applies, because fff is differentiable everywhere.
  • Finiteness of the right side. Because fff is smooth and compactly supported, ∣∇f∣|\nabla f|∣∇f∣ is continuous and compactly supported. Its LpL^pLp norm is therefore finite, so the right side is a finite number.
  • Infinite left side. The statement does not separately assert that the left side is finite. That follows from the inequality itself only because the right side is finite.
  • The product 0⋅∞0 \cdot \infty0⋅∞. No 0⋅∞0 \cdot \infty0⋅∞ convention is triggered, since the constant is strictly positive and the right-hand norm is finite.
Human review
  • Endorsed by Shuze Chen · Oct 2, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 2, 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