Documentation

TauCeti.Analysis.PDE.Caccioppoli.Logarithmic

The logarithmic Caccioppoli inequality for positive supersolutions #

Let a be measurable and uniformly elliptic on Ω ⊆ ℝⁿ with constants 0 < λ ≤ Λ, and let u ∈ H¹(Ω) be a weak supersolution of the divergence-form equation

-∂ᵢ(aⁱʲ ∂ⱼu) ≥ 0 in Ω,

meaning a(u, v) ≥ 0 for every nonnegative v ∈ H¹₀(Ω). If u is bounded below by a positive constant on the support of a smooth cutoff ψ compactly supported in Ω, then

∫_Ω ψ² ‖∇u‖² / u² ≤ (2Λ/λ)² ∫_Ω ‖∇ψ‖².

This is the logarithmic Caccioppoli inequality: ‖∇u‖ / u = ‖∇ log u‖, so the gradient of log u is controlled by the cutoff alone, independently of the size of u. Combined with the Poincaré–Wirtinger inequality on balls it shows that log u has bounded mean oscillation: on every ball B(x₀, R) with B(x₀, 2R) ⊆ Ω, if u ≥ m almost everywhere on the doubled ball B(x₀, 2R) for some constant m > 0, then

∫_{B(x₀, R)} (log u - (log u)_{B(x₀, R)})² ≤ C (Λ/λ)² |B(x₀, R)|,

with C depending only on the dimension. This estimate is the bridge between the bounds for subsolutions and for supersolutions in Moser's proof of the Harnack inequality: through the John–Nirenberg inequality, or the Bombieri–Giusti lemma, it is what allows the averages of a small positive power of u and of its reciprocal to be compared.

The proof tests the supersolution inequality against ψ²/u, which lies in H¹₀(Ω), with weak gradient 2ψ ∇ψ / u - ψ² ∇u / u². Since u is only bounded below on the support of ψ, the reciprocal is realised as a composition F ∘ u with a C¹ function F that has bounded derivative, vanishes near 0 and agrees with t ↦ 1/t for t ≥ m/2; the logarithm is handled in the same way for the mean-oscillation bound.

Main declarations #

References #

theorem TauCeti.PDE.UniformlyEllipticOn.setIntegral_sq_mul_norm_gradient_sq_div_sq_le {ι : Type u_1} [Fintype ι] {mu : MeasureTheory.Measure (EuclideanSpace ℝ ι)} [mu.IsAddHaarMeasure] {Omega : TopologicalSpace.Opens (EuclideanSpace ℝ ι)} {a : EuclideanSpace ℝ ι → Matrix ι ι ℝ} {lam Lam : ℝ} [DecidableEq ι] (h : UniformlyEllipticOn (↑Omega) a lam Lam) (ha : MeasureTheory.AEStronglyMeasurable a (mu.restrict ↑Omega)) {u : ↥(W1p mu Omega 2)} (hu : ∀ (v : ↥(W1p0 mu Omega 2)), (∀ᵐ (x : EuclideanSpace ℝ ι) ∂mu.restrict ↑Omega, 0 ≤ ↑↑(W1p.value ↑v) x) → 0 ≤ energyFormH1 a 0 0 u ↑v) {ψ : EuclideanSpace ℝ ι → ℝ} (hψ : ContDiff ℝ (↑⊤) ψ) (hcpt : HasCompactSupport ψ) (hts : tsupport ψ ⊆ ↑Omega) {m : ℝ} (hm : 0 < m) (hum : ∀ᵐ (x : EuclideanSpace ℝ ι) ∂mu.restrict (tsupport ψ), m ≤ ↑↑(W1p.value u) x) :
∫ (x : EuclideanSpace ℝ ι) in ↑Omega, ψ x ^ 2 * ‖↑↑(W1p.gradient u) x‖ ^ 2 / ↑↑(W1p.value u) x ^ 2 ∂mu ≤ (2 * Lam / lam) ^ 2 * ∫ (x : EuclideanSpace ℝ ι) in ↑Omega, ‖gradient ψ x‖ ^ 2 ∂mu

The logarithmic Caccioppoli inequality. Let a be measurable and uniformly elliptic on Ω with constants 0 < λ ≤ Λ, and let u ∈ H¹(Ω) be a weak supersolution of -∂ᵢ(aⁱʲ ∂ⱼu) ≥ 0, that is a(u, v) ≥ 0 for every nonnegative v ∈ H¹₀(Ω). Let ψ be smooth and compactly supported in Ω, and suppose that u ≥ m almost everywhere on the support of ψ for some m > 0. Then

∫_Ω ψ² ‖∇u‖² / u² ≤ (2Λ/λ)² ∫_Ω ‖∇ψ‖².

The left side is ∫_Ω ψ² ‖∇ log u‖²; the bound does not depend on m, on the size of u, or on any regularity of the coefficients beyond measurability.

theorem TauCeti.PDE.exists_setIntegral_ball_norm_gradient_sq_div_sq_le {ι : Type u_1} [Fintype ι] [DecidableEq ι] :
∃ (c : ℝ), 0 < c ∧ ∀ {mu : MeasureTheory.Measure (EuclideanSpace ℝ ι)} [inst : mu.IsAddHaarMeasure] {Omega : TopologicalSpace.Opens (EuclideanSpace ℝ ι)} {a : EuclideanSpace ℝ ι → Matrix ι ι ℝ} {lam Lam : ℝ} {u : ↥(W1p mu Omega 2)} {x₀ : EuclideanSpace ℝ ι} {r R m : ℝ}, UniformlyEllipticOn (↑Omega) a lam Lam → MeasureTheory.AEStronglyMeasurable a (mu.restrict ↑Omega) → (∀ (v : ↥(W1p0 mu Omega 2)), (∀ᵐ (x : EuclideanSpace ℝ ι) ∂mu.restrict ↑Omega, 0 ≤ ↑↑(W1p.value ↑v) x) → 0 ≤ energyFormH1 a 0 0 u ↑v) → 0 < r → r < R → Metric.ball x₀ R ⊆ ↑Omega → 0 < m → (∀ᵐ (x : EuclideanSpace ℝ ι) ∂mu.restrict (Metric.ball x₀ R), m ≤ ↑↑(W1p.value u) x) → ∫ (x : EuclideanSpace ℝ ι) in Metric.ball x₀ r, ‖↑↑(W1p.gradient u) x‖ ^ 2 / ↑↑(W1p.value u) x ^ 2 ∂mu ≤ (2 * Lam / lam) ^ 2 * (c / (R - r)) ^ 2 * mu.real (Metric.ball x₀ R)

The logarithmic Caccioppoli inequality on concentric balls. There is a constant c > 0, depending only on the dimension, such that the following holds for every additive Haar measure μ. Let a be measurable and uniformly elliptic on Ω with constants 0 < λ ≤ Λ, and let u ∈ H¹(Ω) be a weak supersolution of -∂ᵢ(aⁱʲ ∂ⱼu) ≥ 0. If B(x₀, r) ⊆ B(x₀, R) ⊆ Ω with 0 < r < R, and u ≥ m almost everywhere on B(x₀, R) for some m > 0, then

∫_{B(x₀, r)} ‖∇u‖² / u² ≤ (2Λ/λ)² (c / (R - r))² |B(x₀, R)|.

theorem TauCeti.PDE.exists_setIntegral_ball_log_sub_setAverage_sq_le {ι : Type u_1} [Fintype ι] [DecidableEq ι] :
∃ (C : ℝ), 0 < C ∧ ∀ {mu : MeasureTheory.Measure (EuclideanSpace ℝ ι)} [inst : mu.IsAddHaarMeasure] {Omega : TopologicalSpace.Opens (EuclideanSpace ℝ ι)} {a : EuclideanSpace ℝ ι → Matrix ι ι ℝ} {lam Lam : ℝ} {u : ↥(W1p mu Omega 2)} {x₀ : EuclideanSpace ℝ ι} {R m : ℝ}, UniformlyEllipticOn (↑Omega) a lam Lam → MeasureTheory.AEStronglyMeasurable a (mu.restrict ↑Omega) → (∀ (v : ↥(W1p0 mu Omega 2)), (∀ᵐ (x : EuclideanSpace ℝ ι) ∂mu.restrict ↑Omega, 0 ≤ ↑↑(W1p.value ↑v) x) → 0 ≤ energyFormH1 a 0 0 u ↑v) → 0 < R → Metric.ball x₀ (2 * R) ⊆ ↑Omega → 0 < m → (∀ᵐ (x : EuclideanSpace ℝ ι) ∂mu.restrict (Metric.ball x₀ (2 * R)), m ≤ ↑↑(W1p.value u) x) → ∫ (x : EuclideanSpace ℝ ι) in Metric.ball x₀ R, (Real.log (↑↑(W1p.value u) x) - ⨍ (y : EuclideanSpace ℝ ι) in Metric.ball x₀ R, Real.log (↑↑(W1p.value u) y) ∂mu) ^ 2 ∂mu ≤ C * (Lam / lam) ^ 2 * mu.real (Metric.ball x₀ R)

Bounded mean oscillation of the logarithm of a positive supersolution (Moser). There is a constant C > 0, depending only on the dimension, such that the following holds for every additive Haar measure μ. Let a be measurable and uniformly elliptic on Ω with constants 0 < λ ≤ Λ, and let u ∈ H¹(Ω) be a weak supersolution of -∂ᵢ(aⁱʲ ∂ⱼu) ≥ 0. If B(x₀, 2R) ⊆ Ω and u ≥ m almost everywhere on B(x₀, 2R) for some m > 0, then

∫_{B(x₀, R)} (log u - (log u)_{B(x₀, R)})² ≤ C (Λ/λ)² |B(x₀, R)|,

where (log u)_{B(x₀, R)} is the mean of log u over the ball. The bound does not depend on m, on R, or on the size of u.