Total variation of a convolution #
Convolving with an integrable kernel ρ averages the translates u (· - t) of a function with
weights ρ t. Each translate has total variation at most |Du|(Ω) on an open set Ω' with
Ω' - tsupport ρ ⊆ Ω, so the convolution satisfies the convolution estimate
|D(ρ ⋆ u)|(Ω') ≤ ‖ρ‖₁ |Du|(Ω).
For a normalized smooth bump the factor ‖ρ‖₁ is one. On the whole space the estimate combines
with lower semicontinuity under L¹_loc convergence: the variations of the mollifications of a
locally integrable function converge to its variation. Since a mollification is smooth, its
variation is the integral of the norm of its gradient. Thus every locally integrable function u
on the whole space satisfies |Du|(E) = lim ∫ ‖∇(ρ_ε ⋆ u)‖, where |Du|(E) is the variational
supremum TauCeti.totalVariationOn, finite or infinite. For integrable u the mollifications
also converge to u in L¹ (TauCeti.tendsto_eLpNorm_normed_convolution_sub), so they are a
strict approximation of u by smooth functions on the whole space.
The convolution estimate holds for every s-finite right-invariant measure. The limit statements use an additive Haar measure, against which the normalized bumps have integral one.
Main results #
TauCeti.totalVariationOn_convolution_le: the convolution estimate|D(ρ ⋆ u)|(Ω') ≤ ‖ρ‖₁ |Du|(Ω)whenΩ' - tsupport ρ ⊆ Ω.ContDiffBump.totalVariationOn_normed_convolution_le: mollification by a normalized bump of outer radiusrdoes not increase variation fromΩto any open set whose closedr-neighbourhood lies inΩ.TauCeti.exists_mollifier_seq_of_isCompact_closure: local smooth approximation with an eventual variation bound on an open set whose closure is a compact subset of the domain.TauCeti.tendsto_totalVariationOn_normed_convolution: on the whole space, the variations of mollifications with shrinking radii converge to the variation of the function.TauCeti.tendsto_lintegral_enorm_fderiv_normed_convolution: the same limit written as the integrals∫ ‖∇(ρ_ε ⋆ u)‖of the smooth mollifications.
References #
- L. Ambrosio, N. Fusco, D. Pallara, Functions of Bounded Variation and Free Discontinuity Problems, §3.1.
- L. C. Evans, R. F. Gariepy, Measure Theory and Fine Properties of Functions, §5.2.2.
The convolution estimate for total variation. If ρ is integrable with compact support
and every translate Ω' - t with t ∈ tsupport ρ lies in Ω, then convolving a function locally
integrable on Ω with ρ gives |D(ρ ⋆ u)|(Ω') ≤ ‖ρ‖₁ |Du|(Ω).
Mollification does not increase variation. For a normalized smooth bump φ of outer
radius r, if the closed ball of radius r about every point of Ω' lies in Ω, then
|D(φ ⋆ u)|(Ω') ≤ |Du|(Ω) for every u locally integrable on Ω.
On an open set with compact closure inside Ω, a locally integrable function admits
C¹ approximations converging in L¹ on each compact subset of that closure, with variation
eventually bounded by its variation on Ω. The construction localizes the function to a compact
neighbourhood before mollifying it, as in the scalar coarea proof.
Mollification approximates the variation on the whole space. For a locally integrable
function u and normalized smooth bumps φ i whose outer radii tend to zero, the variations of
the mollifications φ i ⋆ u on the whole space converge to the variation of u, finite or
infinite.
The variation as a limit of gradient integrals. For a locally integrable function u on
the whole space and normalized smooth bumps φ i whose outer radii tend to zero, the integrals
∫ ‖∇(φ i ⋆ u)‖ of the smooth mollifications converge to the variation |Du|(E), finite or
infinite.