Markov type of Alexandrov spaces of nonnegative curvature

We prove that Alexandrov spaces $X$ of nonnegative curvature have Markov type 2 in the sense of Ball. As a corollary, any Lipschitz continuous map from a subset of $X$ into a 2-uniformly convex Banach space is extended as a Lipschitz continuous map on the entire space $X$.

💡 Research Summary

The paper establishes that every Alexandrov space with non‑negative curvature possesses Markov type 2 in the sense introduced by Ball. The authors begin by recalling the classical notions of Rademacher type and cotype for Banach spaces and their nonlinear extensions: Enflo type and Ball’s Markov type and cotype. Markov type 2 is defined via reversible stationary Markov chains; it requires that the expected squared distance after one step of the chain does not exceed a universal constant times the expected squared distance after a “smoothing” operation defined by the matrix C = (1‑α)(I‑αA)⁻¹. This property is stronger than Enflo type 2 and is known to hold for Hilbert spaces, their bi‑Lipschitz equivalents, 2‑uniformly smooth Banach spaces, trees, hyperbolic groups, and certain negatively curved manifolds.

The core of the paper is the proof that Alexandrov spaces of non‑negative curvature satisfy the Markov type 2 inequality with a universal constant K = 1 + √2, independent of dimension. The key tool is a characterization of non‑negative curvature due to Sturm: for any finite collection of points {x_i} and any probability weights {a_i} summing to one, the quadratic form Σ_{i,j} a_i a_j (d(x_i,x_j)² – d(x_i,y)² – d(x_j,y)²) is non‑positive for every y. This inequality mirrors the parallelogram law in Hilbert spaces and provides a convexity‑type estimate for squared distances.

Using this inequality, the authors define for a given reversible Markov chain the quantity E(l) = ∑{i,j} π_i a^{(l)}{ij} d(x_i,x_j)², where a^{(l)} denotes the l‑step transition matrix. Lemma 4.1 shows that E(2l) ≤ 2E(l). By induction they prove the stronger bound E(l) ≤ (1 + √2)^{2l} E(1) for all integers l. Substituting this estimate into the equivalent formulation of Markov type (inequality (2.8)) yields the desired constant K = 1 + √2, establishing that every non‑negatively curved Alexandrov space has Markov type 2 with M₂(X) ≤ 1 + √2.

An immediate corollary follows from Ball’s extension theorem: if V is a reflexive Banach space with Markov cotype 2 (in particular any 2‑uniformly convex Banach space), then any Lipschitz map f defined on a subset Z of an Alexandrov space X can be extended to a Lipschitz map ˜f : X → V with Lipschitz constant at most 6·M₂(X)·C₂(V)·Lip(f). Since M₂(X) is bounded by a universal constant, the extension constant depends only on the geometry of V and not on the dimension of X. This yields dimension‑free Lipschitz extension results for maps from non‑negatively curved Alexandrov spaces into a broad class of Banach spaces, including all L^p spaces with 1 < p ≤ 2.

The paper concludes by emphasizing that Sturm’s curvature inequality serves as a bridge between curvature bounds and probabilistic Markov estimates, opening a new avenue for studying metric‑geometric properties of spaces with lower curvature bounds via Markov type techniques. The results enrich the catalog of spaces known to have Markov type 2 and suggest further investigations into other curvature regimes and their interaction with nonlinear functional‑analytic invariants.