TheoremBase

The Lipschitz Truncation of a Convex Function: a Global Lipschitz Convex Minorant Agreeing with It Where the Slope is Small

Infimal convolution of a convex function on an open convex set with a multiple of the norm produces a convex function on the whole space, Lipschitz with that multiple as constant, lying below the original; at a point whose subgradients are bounded by the constant the two functions agree, and a unique subgradient there is inherited by the truncation.

Statement

We work in the setting of Product Measures and Lebesgue Measure on ℝⁿ: Standing Notation and Background, whose notation is fixed for every dimension and is used here with a natural number nn satisfying 1≤n1\le n: the real numbers and sequences, and the Euclidean norm ∥ ⋅ ∥\lVert\,\cdot\,\rVert, dot product, distance dEd_{E} and topology, are as fixed there. The space Rn\mathbb{R}^{n} is open in itself by claim 1 of Euclidean Space Is Open in Itself and is convex, every convex combination of two of its points being again one of its points.

Let G⊆RnG\subseteq\mathbb{R}^{n} be open and convex and nonempty, let ϕ:G→R\phi:G\to\mathbb{R} be convex on GG, with subdifferential ∂Gϕ\partial_{G}\phi, and let LL be a real number for which there are x0∈Gx_{0}\in G and q0∈∂Gϕ(x0)q_{0}\in\partial_{G}\phi(x_{0}) with ∥q0∥≤L\lVert q_{0}\rVert\le L; such an LL exists, since ∂Gϕ(x0)\partial_{G}\phi(x_{0}) is nonempty for every x0∈Gx_{0}\in G by The Subdifferential of a Convex Function on an Open Convex Set is Nonempty. The Lipschitz truncation of ϕ\phi at level LL is the function ϕL:Rn→R\phi^{L}:\mathbb{R}^{n}\to\mathbb{R} given by

ϕL(x)=inf⁡{ϕ(z)+L∥x−z∥ : z∈G}(x∈Rn),\phi^{L}(x)=\inf\bigl\{\phi(z)+L\lVert x-z\rVert\ :\ z\in G\bigr\}\qquad(x\in\mathbb{R}^{n}),

the infimum existing in R\mathbb{R} by claim 1 below.

1. (A Lipschitz convex minorant) For every x∈Rnx\in\mathbb{R}^{n} the set {ϕ(z)+L∥x−z∥:z∈G}\{\phi(z)+L\lVert x-z\rVert:z\in G\} is nonempty and bounded below, so that ϕL(x)\phi^{L}(x) is a real number. The function ϕL\phi^{L} is convex on Rn\mathbb{R}^{n} and is Lipschitz with constant LL from (Rn,dE)(\mathbb{R}^{n},d_{E}) to R\mathbb{R} with the absolute-value metric, and ϕL(x)≤ϕ(x)\phi^{L}(x)\le\phi(x) for every x∈Gx\in G.

2. (Agreement and subgradients where the slope is small) Let x∈Gx\in G and let p∈∂Gϕ(x)p\in\partial_{G}\phi(x) satisfy ∥p∥≤L\lVert p\rVert\le L. Then

ϕL(x)=ϕ(x),p∈∂RnϕL(x).\phi^{L}(x)=\phi(x),\qquad p\in\partial_{\mathbb{R}^{n}}\phi^{L}(x).

3. (A unique subgradient is inherited) Let x∈Gx\in G and suppose that ∂Gϕ(x)={p}\partial_{G}\phi(x)=\{p\} with ∥p∥≤L\lVert p\rVert\le L. Then ∂RnϕL(x)={p}\partial_{\mathbb{R}^{n}}\phi^{L}(x)=\{p\}.

Proofs

Log in to submit a proof.

Loading...

Citations

Loading…

Dependencies

Loading…

Related

0 relations

Curated associations between results. These are editable and subjective — they do not replace the dependency graph, which is derived from the references in the text.

No relations recorded yet.

Comments

Log in to comment.

Loading…