Documentation

TauCeti.Analysis.BoundedVariation.TotalVariation

The total variation of a function on an open set #

Let Ω be an open set of a finite-dimensional real inner product space E. The total variation of a function u : E → ℝ on Ω is the extended supremum

|Du|(Ω) = ⨆ { ∫ u div φ | φ ∈ 𝓓(Ω, E), ‖φ x‖ ≤ 1 for all x }

over smooth vector fields compactly supported in Ω with pointwise norm at most one (TauCeti.totalVariationOn). The functions of bounded variation on Ω are the u ∈ L¹(Ω) for which it is finite, and the perimeter of a set is the total variation of its indicator. When the distributional derivative Du is a vector-valued Radon measure, this supremum is the total variation measure |Du| evaluated on Ω; that comparison is not proved here. The definition by duality is meaningful for every locally integrable function, with no differentiability assumption, which is what lower semicontinuity and compactness arguments need.

The measure μ is a parameter, as in TauCeti.HasWeakFDerivOn. The pairing ∫ u div φ ∂μ is the distributional pairing for an additive Haar measure μ, and that hypothesis is carried by the results that need it: a constant function has total variation zero only against a translation-invariant measure.

For u not locally integrable on Ω the integrals in the supremum may be Mathlib's junk value 0 for non-integrable functions, so the results that split integrals assume local integrability.

Main declarations #

References #

The total variation of u : E → ℝ on the open set Ω, relative to the measure μ: the extended supremum of ∫ div φ * u ∂μ over test vector fields φ ∈ 𝓓(Ω, E) with ‖φ x‖ ≤ 1 everywhere. For μ an additive Haar measure this is the variation |Du|(Ω) of the distributional derivative of u on Ω, defined by duality.

Instances For
    theorem TauCeti.totalVariationOn_def {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] [MeasurableSpace E] {μ : MeasureTheory.Measure E} {Ω : TopologicalSpace.Opens E} {u : E → ℝ} :
    totalVariationOn μ Ω u = ⨆ (φ : TestFunction Ω E ⊤), ⨆ (_ : ∀ (x : E), ‖φ x‖ ≤ 1), ENNReal.ofReal (∫ (x : E), (TestFunction.divergenceCLM φ) x * u x ∂μ)

    The defining supremum of TauCeti.totalVariationOn.

    An admissible test vector field bounds the total variation from below.

    theorem TauCeti.totalVariationOn_le_iff {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] [MeasurableSpace E] {μ : MeasureTheory.Measure E} {Ω : TopologicalSpace.Opens E} {u : E → ℝ} {c : ENNReal} :
    totalVariationOn μ Ω u ≤ c ↔ ∀ (φ : TestFunction Ω E ⊤), (∀ (x : E), ‖φ x‖ ≤ 1) → ENNReal.ofReal (∫ (x : E), (TestFunction.divergenceCLM φ) x * u x ∂μ) ≤ c

    Upper bounds for the total variation are tested on admissible test vector fields.

    Every admissible pairing is bounded in absolute value by the total variation.

    @[simp]

    The total variation of the zero function vanishes.

    Locality. The total variation on Ω only depends on u almost everywhere on Ω.

    An admissible test vector field on Ω whose support lies in another open set Ω' is admissible on Ω', so its pairing is bounded by the total variation on Ω'.

    Monotonicity in the open set. A test vector field on a smaller open set is one on the larger set, so the total variation grows with the open set.

    theorem TauCeti.totalVariationOn_iSup {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] [MeasurableSpace E] {μ : MeasureTheory.Measure E} {u : E → ℝ} {ι : Type u_2} {Ω : ι → TopologicalSpace.Opens E} (hΩ : Directed (fun (x1 x2 : TopologicalSpace.Opens E) => x1 ≤ x2) Ω) :
    totalVariationOn μ (⨆ (i : ι), Ω i) u = ⨆ (i : ι), totalVariationOn μ (Ω i) u

    Inner continuity in the open set. The total variation on the union of a directed family of open sets is the supremum of the total variations on its members.

    theorem TauCeti.totalVariationOn_iSup_le_tsum {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] [MeasurableSpace E] {μ : MeasureTheory.Measure E} {u : E → ℝ} [OpensMeasurableSpace E] {ι : Type u_2} {Ω : ι → TopologicalSpace.Opens E} (hu : MeasureTheory.LocallyIntegrableOn u (↑(⨆ (i : ι), Ω i)) μ) :
    totalVariationOn μ (⨆ (i : ι), Ω i) u ≤ ∑' (i : ι), totalVariationOn μ (Ω i) u

    Countable subadditivity in the open set. For u locally integrable on ⋃ Ωᵢ, |Du|(⋃ Ωᵢ) ≤ ∑ |Du|(Ωᵢ).

    Subadditivity in the open set. For u locally integrable on Ω ∪ Ω', |Du|(Ω ∪ Ω') ≤ |Du|(Ω) + |Du|(Ω').

    Additivity on disjoint open sets. For u locally integrable on Ω ∪ Ω' with Ω and Ω' disjoint, |Du|(Ω ∪ Ω') = |Du|(Ω) + |Du|(Ω').

    @[simp]

    The total variation of -u is that of u.

    The total variation is positively homogeneous: |D(c u)|(Ω) = c |Du|(Ω) for c ≥ 0.

    The total variation is absolutely homogeneous: |D(c u)|(Ω) = |c| |Du|(Ω).

    A test vector field bounded by c pairs to at most c |Du|(Ω).

    Subadditivity. For u locally integrable on Ω, |D(u + v)|(Ω) ≤ |Du|(Ω) + |Dv|(Ω). No integrability of v is needed: when div φ * v is not integrable, neither is div φ * (u + v), and its pairing is 0.

    @[simp]

    Against an additive Haar measure, adding a constant does not change the total variation. No integrability of u is needed.

    @[simp]

    Against an additive Haar measure, c - u has the same total variation as u. No integrability of u is needed.

    @[simp]

    Against an additive Haar measure, a constant function has total variation zero.

    Lower semicontinuity #

    theorem TauCeti.totalVariationOn_le_liminf_of_tendsto_integral {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] [MeasurableSpace E] {μ : MeasureTheory.Measure E} {Ω : TopologicalSpace.Opens E} {ι : Type u_2} {l : Filter ι} {u : ι → E → ℝ} {u₀ : E → ℝ} (h : ∀ (ψ : TestFunction Ω ℝ ⊤), Filter.Tendsto (fun (i : ι) => ∫ (x : E), ψ x * u i x ∂μ) l (nhds (∫ (x : E), ψ x * u₀ x ∂μ))) :
    totalVariationOn μ Ω u₀ ≤ Filter.liminf (fun (i : ι) => totalVariationOn μ Ω (u i)) l

    Lower semicontinuity under distributional convergence. If u i → u₀ as distributions on Ω, that is ∫ ψ * u i → ∫ ψ * u₀ for every scalar test function ψ on Ω, then |Du₀|(Ω) ≤ liminf |Du i|(Ω).

    theorem TauCeti.totalVariationOn_le_liminf {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] [MeasurableSpace E] {μ : MeasureTheory.Measure E} {Ω : TopologicalSpace.Opens E} [OpensMeasurableSpace E] {ι : Type u_2} {l : Filter ι} {u : ι → E → ℝ} {u₀ : E → ℝ} (hu : ∀ᶠ (i : ι) in l, MeasureTheory.LocallyIntegrableOn (u i) (↑Ω) μ) (hu₀ : MeasureTheory.LocallyIntegrableOn u₀ (↑Ω) μ) (h : ∀ (K : Set E), IsCompact K → K ⊆ ↑Ω → Filter.Tendsto (fun (i : ι) => ∫⁻ (x : E) in K, ‖u i x - u₀ x‖ₑ ∂μ) l (nhds 0)) :
    totalVariationOn μ Ω u₀ ≤ Filter.liminf (fun (i : ι) => totalVariationOn μ Ω (u i)) l

    Lower semicontinuity under L¹_loc convergence. If u i → u₀ in L¹_loc(Ω), that is ∫_K |u i - u₀| → 0 for every compact K ⊆ Ω, then |Du₀|(Ω) ≤ liminf |Du i|(Ω).

    theorem TauCeti.tendsto_totalVariationOn_of_le_add {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] [MeasurableSpace E] {μ : MeasureTheory.Measure E} {Ω : TopologicalSpace.Opens E} {u : E → ℝ} [OpensMeasurableSpace E] {ι : Type u_2} {l : Filter ι} {w : ι → E → ℝ} (hw : ∀ (i : ι), MeasureTheory.LocallyIntegrableOn (w i) (↑Ω) μ) (hu : MeasureTheory.LocallyIntegrableOn u (↑Ω) μ) (hL : ∀ (K : Set E), IsCompact K → K ⊆ ↑Ω → Filter.Tendsto (fun (i : ι) => ∫⁻ (x : E) in K, ‖w i x - u x‖ₑ ∂μ) l (nhds 0)) {b : ι → ENNReal} (hb : Filter.Tendsto b l (nhds 0)) (hwb : ∀ (i : ι), totalVariationOn μ Ω (w i) ≤ totalVariationOn μ Ω u + b i) :
    Filter.Tendsto (fun (i : ι) => totalVariationOn μ Ω (w i)) l (nhds (totalVariationOn μ Ω u))

    Convergence of variations from an upper bound. If w i → u in L¹_loc(Ω) and the variations of the w i exceed that of u by amounts tending to zero, then the variations converge. Lower semicontinuity (totalVariationOn_le_liminf) gives the lower bound.

    theorem TauCeti.totalVariationOn_le_of_tendsto_L1 {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] [MeasurableSpace E] {μ : MeasureTheory.Measure E} {Ω : TopologicalSpace.Opens E} [OpensMeasurableSpace E] {ι : Type u_2} {l : Filter ι} {u : ι → ↥(MeasureTheory.Lp ℝ 1 (μ.restrict ↑Ω))} {v : ↥(MeasureTheory.Lp ℝ 1 (μ.restrict ↑Ω))} {C : ENNReal} (h : Filter.Tendsto u l (nhds v)) (hC : ∃ᶠ (i : ι) in l, totalVariationOn μ Ω ↑↑(u i) ≤ C) :
    totalVariationOn μ Ω ↑↑v ≤ C

    A bound on the total variation holding frequently along l is preserved under convergence in L¹(Ω) along l.

    Weakly differentiable functions #

    A weakly differentiable function has variation at most ∫_Ω ‖Du‖. If u has weak derivative U on Ω, then |Du|(Ω) ≤ ∫_Ω ‖U‖. In particular a function in W^{1,1}(Ω) has bounded variation on Ω.

    theorem TauCeti.HasWeakFDerivOn.exists_totalVariationOn_ne_top {E : Type u_1} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E] [MeasurableSpace E] {μ : MeasureTheory.Measure E} {Ω : TopologicalSpace.Opens E} {u : E → ℝ} [OpensMeasurableSpace E] {U : E → E →L[ℝ] ℝ} (h : HasWeakFDerivOn μ Ω u U) {K : Set E} (hK : IsCompact K) (hKΩ : K ⊆ ↑Ω) :
    ∃ (V : TopologicalSpace.Opens E), K ⊆ ↑V ∧ totalVariationOn μ V u ≠ ⊤

    A weakly differentiable function has locally finite variation. If u has a weak derivative on Ω, every compact K ⊆ Ω has an open neighbourhood on which the variation of u is finite.

    The total variation of a weakly differentiable function. If u has weak derivative U on Ω, then against an additive Haar measure |Du|(Ω) = ∫_Ω ‖U‖. In particular a function in W^{1,1}(Ω) has bounded variation on Ω, and its variation is the L¹(Ω) norm of its weak derivative.

    A weakly differentiable function has finite total variation on Ω, against an additive Haar measure, exactly when its weak derivative is integrable on Ω.

    The total variation of a locally Lipschitz function, in particular of a C¹ function, is the integral of the norm of its classical derivative: |Du|(Ω) = ∫_Ω ‖Du‖, against an additive Haar measure.

    A compactly supported Lipschitz function, in particular a compactly supported C¹ function, has finite total variation on every open set, against an additive Haar measure.