A solution to the extreme point problem and other applications of Choquet theory to Lipschitz-free spaces

We prove that every element of a Lipschitz-free space admits an expression as a convex series of elements with compact support. As a consequence, we conclude that all extreme points of the unit ball of Lipschitz-free spaces are elementary molecules, solving a long-standing problem. We also deduce that all elements of a Lipschitz-free space with the Radon-Nikodým property can be expressed as convex integrals of molecules. Our results are based on a recent theory of integral representation for functionals on Lipschitz spaces which draws on classical Choquet theory, due to the third named author.

💡 Research Summary

The paper investigates the structure of Lipschitz‑free Banach spaces F(M) over a complete metric space M, using a recent Choquet‑type theory developed by the third author. The authors first recall the classical De Leeuw representation: every functional m in the dual of the Lipschitz space Lip₀(M) can be written as Φ*μ for some signed Radon measure μ on the Stone‑Čech compactification β f M of the set f M = {(x,y)∈M×M : x≠y}. While optimal (norm‑minimal) measures are known to exist, they are far from unique and may involve points of the compactification that are not in M.

To overcome this, the authors introduce a quasi‑order ≼ on the cone of positive measures M(β f M)⁺. The order is defined by testing against a convex cone G of continuous functions that contains all incremental quotients Φf for f∈Lip₀(M). One writes μ≼ν iff ∫g dμ ≤ ∫g dν for every g∈G. This order captures the idea of “efficiency”: a measure that concentrates its mass on fewer points (or avoids the diagonal where the distance is zero) is considered smaller. Minimal elements with respect to ≼ are then studied; they are abundant by Zorn’s lemma and inherit optimality when the original measure is optimal.

The central technical result (Theorem 1.1, originally due to Smith) states that for any m∈F(M) there exists a measure μ that is simultaneously optimal, minimal, and concentrated on p⁻¹(M×M), where p:β f M→β M×β M is the natural projection. In other words, the representation of a free‑space element can be achieved using only genuine points of M, never the “extra” points of the compactification. Moreover, the support of μ may be taken inside A×A with A = supp m ∪ {0}.

Using this refined representation, the authors prove a compact‑decomposition principle (Theorem 2.1): every m∈F(M) can be written as a convex series m=∑ₙ mₙ with ∥m∥=∑ₙ∥mₙ∥ and each mₙ having compact support (indeed, supp mₙ⊂supp m). The construction proceeds by repeatedly splitting the minimal optimal measure μ into pieces supported on Borel subsets whose shadows are compact; the corresponding functionals give the desired mₙ. This result is essentially equivalent to Theorem 1.1 but is presented in a more “user‑friendly” form. It can be viewed as a weak inner‑regularity property for elements of F(M) and as an isometric analogue of the compact‑reduction principle known for other Banach spaces.

The compact‑decomposition is then leveraged to solve the long‑standing extreme‑point problem for Lipschitz‑free spaces. The problem, raised by Weaver in the 1990s, asks whether every extreme point of the unit ball B_{F(M)} must be an elementary molecule m_{x,y} = (δ(x)−δ(y))/d(x,y). The answer was known for compact M (Albiac–Kalton, etc.) but remained open for general metric spaces. By applying Theorem 2.1, any extreme point can be approximated by a convex combination of compact‑supported elements, each of which, by the known compact case, must be an elementary molecule. The convexity forces the original extreme point itself to be a single molecule. Thus Theorem 3.1 establishes that all extreme points of B_{F(M)} are elementary molecules, settling the problem completely.

Next, the paper addresses the situation where F(M) has the Radon‑Nikodým property (RNP). In Banach spaces with the RNP, every element can be expressed as a Bochner integral of extreme points of the unit ball. Using the minimal‑optimal representation together with the compact‑decomposition, the authors prove Theorem 4.1: if F(M) has the RNP, then every m∈F(M) admits a representation as a convex integral of elementary molecules, i.e. m=∫_{M×M} (δ(x)−δ(y))/d(x,y) dν(x,y) for some probability measure ν concentrated on M×M. As a corollary (4.5) they obtain new information about Lipschitz functions that attain their Lipschitz constant: such functions must have a “supporting pair” (x,y) where the incremental quotient reaches the norm.

Finally, Theorem 5.4 provides a finer decomposition: any m∈F(M) can be split into a part that is a convex integral of molecules and a remainder whose representing measures are concentrated on the diagonal set d⁻¹(0) (i.e., pairs with zero distance). This separates the “good” part of the functional from a pathological component that lives on the “long diagonal”.

Overall, the paper demonstrates that the Choquet‑type quasi‑order and the associated minimal‑optimal representations give a powerful new toolkit for Lipschitz‑free spaces. They allow one to reduce many problems to the compact case, resolve the extreme‑point conjecture, and obtain integral representations under the RNP. The results deepen the connection between metric geometry, Banach space theory, and classical Choquet theory, and they are likely to influence future work on the geometry of free spaces, optimal transport, and nonlinear functional analysis.