[Paper Review] Metric Spaces with Linear Extensions Preserving Lipschitz Condition

Alexander Brudnyi, Yu. A. Brudnyi|ArXiv.org|Apr 17, 2004

Advanced Harmonic Analysis Research9 references4 citations

TL;DR

This paper introduces a new bi-Lipschitz invariant, λ(M), measuring the worst-case distortion in linearly extending Lipschitz functions from subsets of a metric space M to M itself. It proves λ(M) is finite for key classes including metric trees, groups of polynomial growth, Gromov-hyperbolic groups, and certain Riemannian manifolds of bounded geometry, while constructing a counterexample with λ(M) = ∞ for a two-dimensional Riemannian manifold of bounded geometry.

ABSTRACT

We study a new bi-Lipschitz invariant λ(M) of a metric space M; its finiteness means that Lipschitz functions on an arbitrary subset of M can be linearly extended to functions on M whose Lipschitz constants are enlarged by a factor controlled by λ(M). We prove that λ(M) is finite for several important classes of metric spaces. These include metric trees of arbitrary cardinality, groups of polynomial growth, Gromov-hyperbolic groups, certain classes of Riemannian manifolds of bounded geometry and finite direct sums of arbitrary combinations of these objects. On the other hand we construct an example of a two-dimensional Riemannian manifold M of bounded geometry for which λ(M)=\infty.

Motivation & Objective

• To define and analyze a new bi-Lipschitz invariant λ(M) that quantifies the distortion in linearly extending Lipschitz functions from subsets of a metric space M to M.
• To determine for which classes of metric spaces the invariant λ(M) is finite, indicating the existence of uniformly bounded linear extension operators.
• To establish sharp or asymptotic bounds on λ(M) for specific spaces, including metric trees, Carnot groups, and direct sums of such spaces.
• To construct a counterexample of a two-dimensional Riemannian manifold of bounded geometry for which no such linear extension operator exists (λ(M) = ∞).

Proposed method

• Define λ(M) as the supremum over all subsets S ⊂ M of the infimum operator norm of linear extension operators from Lip(S) to Lip(M).
• Use the duality between Lipschitz functions vanishing at a basepoint and the dual of a Banach space constructed from evaluation functionals.
• Represent the space of 1-Lipschitz functions vanishing at a fixed point via an isometric embedding into l∞(B), where B is the unit ball of Lipschitz functions.
• Construct a Banach space V as the closed linear span of the image of M under this embedding, and identify its dual V* with the space of 1-Lipschitz functions vanishing at a basepoint.
• Apply the Hahn-Banach theorem and Gelfand transform to represent functionals on V and control their operator norms via regular Borel measures.
• Use oscillation estimates and approximation by finite convex combinations of evaluation functionals to bound the Lipschitz norm of the extension.

Experimental results

Research questions

• RQ1 For which metric spaces M is the linear extension constant λ(M) finite?
• RQ2 What are sharp or asymptotic bounds for λ(M) in specific classes such as metric trees, Carnot groups, or direct sums of such spaces?
• RQ3 Can the existence of a linear extension operator with controlled norm be characterized in terms of the geometry of the metric space?
• RQ4 Are there metric spaces of bounded geometry for which no such linear extension operator exists, i.e., λ(M) = ∞?
• RQ5 How does the finiteness of λ(M) relate to the local doubling property and other geometric invariants of M?

Key findings

• λ(M) is finite for all metric trees of arbitrary cardinality, with λ(⊕_{i=1}^n T_i) bounded between C₁√n and C₂n for absolute constants C₁, C₂ > 0.
• λ(Z^n) satisfies c√n ≤ λ(Z^n) ≤ 24n for some 0 < c < 1, where Z^n is equipped with the word metric.
• The same upper bound λ(G) ≤ 24n holds for any Carnot group of homogeneous dimension n.
• λ(M) is finite for groups of polynomial growth and Gromov-hyperbolic groups.
• λ(M) is finite for Riemannian manifolds of bounded geometry that are locally doubling, including finite direct sums of such spaces.
• There exists a two-dimensional Riemannian manifold Σ of bounded geometry such that Ext(S,Σ) = ∅ for some subspace S, implying λ(Σ) = ∞.