Markov type characterizes superreflexivity

Proves that a Banach space with nontrivial Markov type is superreflexive, answering a question of Naor.

1 paper September 23, 2026 Explained by AI

Markov type measures how far a random walk can drift when its positions are mapped into a space of vectors. Superreflexivity means the space has an equivalent norm with round, uniformly curved balls. It was known that superreflexive spaces have nontrivial Markov type. This paper proves the converse for real Banach spaces, and Lean checks the result. So nontrivial Markov type and superreflexivity are the same property.

First step

2012

Assaf Naor, who asked the question in his survey of the Ribe program

Significance

Major

It completes a metric characterization of superreflexivity that the Ribe program had left open, with a Lean check.

Trust

Formal proof listed

Field

Functional analysis

1 paper

The explanation

Flip a fair coin and step one unit left or right on the number line. After 100 steps, the average squared distance from the start is exactly 100. So a typical walker is only about 10 units away, not 100. Random walks in straight-line spaces drift like the square root of time.

Markov type turns this into a test for a space of vectors. A Banach space is a space of vectors with a norm, a rule for length, in which limits behave well. Take a Markov chain, a random walk on finitely many states that moves by fixed chances. Call it stationary and reversible if it looks the same run forward or backward. Map its states to vectors by any function f. Look at the p-th power of the distance moved, averaged over the walk. The space has Markov type p with constant K if, after t steps, this average is at most K^p times t times the one-step average. Keith Ball introduced this idea to study how maps extend between spacesSource 1.

A Banach space is superreflexive if it has an equivalent norm that is uniformly convex. Uniformly convex means the midpoint of two far-apart unit vectors is clearly shorter than one. Bourgain showed earlier that such spaces are exactly those that cannot hold large binary trees with small distortionSource 2. Naor, Peres, Schramm and Sheffield showed that suitably smooth norms give Markov typeSource 3. So every superreflexive space has nontrivial Markov type, meaning Markov type for some p above 1. Naor asked whether the converse holdsSource 4. One obstacle is that a weaker test fails. Rademacher type, which controls sums with random signs, does not force superreflexivitySource 5.

The paper proves the converse: every real Banach space with Markov type p for some p above 1 is superreflexiveSource 6. The proof argues by contradictionSource 7. If the space is not superreflexive, it builds a special norm on finite lists of numbers. In that norm, adding neighbouring entries together never makes a list longerSource 8. For each time n, it then builds a reversible walk whose steps all have length 1. With chance at least 3/20, the sum of its first n steps has length at least n/24Source 9. The 3/20 comes from 1/4 minus 1/10. The walk reaches a reduced length of at least n/4 with chance at least 1/4. It fails a technical condition with chance at most 1/10Source 10. Then dividing n/4 by 6 gives n/24.

Here is why that breaks Markov type 2. The average squared length after n steps is at least 3/20 times (n/24) squared. For n = 384,000, n/24 is 16,000. Then 3/20 times 16,000 squared is 38,400,000, which is 100 times n. Markov type 2 says, roughly, that this average stays below K squared times n. Here K squared would have to be at least n divided by 3,840. As n grows, no fixed K works.

How sure is this? Lean, a computer proof checker, has a formal statement of the full result. It says nontrivial Markov type holds exactly when the space has an equivalent uniformly convex norm. So the paper proves this, as far as the Lean statement matches the paper. No specialist review is published.

What it doesn't show. This does not cover general metric spaces. The theorem is stated for real Banach spaces only.

Where it came from

This result comes from an unreleased OpenAI model, published in the openai/math release on 6 October 2026.

Hype check

Major and Lean-checked. It settles Naor's question for real Banach spaces, not for general metric spaces. No specialist review is published.

Words used

Banach space: A space of vectors with a length rule, called a norm, in which limits of sequences behave well.
Norm: A rule that gives each vector a length, with the usual triangle rule.
Markov chain: A random walk on a set of states where each move depends only on the current state.
Stationary and reversible: A chain whose statistics stay the same over time and look the same run backward.
Markov type: A bound saying a walk mapped into the space drifts no faster than a fixed power law in time.
Uniformly convex: A norm where the midpoint of two far-apart unit vectors is shorter than 1 by a fixed margin.
Superreflexive: A Banach space that has an equivalent uniformly convex norm.
Rademacher type: A bound on the length of sums of vectors with independent random plus or minus signs.

The original paper

Nontrivial Markov Type Forces Superreflexivity (PDF, opens in a new tab)
September 23, 2026 Formal proof listed

Lean notes

All quoted lines (10)

Ball introduced Markov type to study Lipschitz extension from metric spaces.
Nontrivial Markov Type Forces Superreflexivity Back to text
Bourgain characterized it by the distortion of finite complete binary trees; finite diamonds provide another metric test.
Nontrivial Markov Type Forces Superreflexivity Back to text
Naor, Peres, Schramm, and Sheffield showed that uniform smoothness of power type $p$ implies Markov type $p$ for $1<p\leq2$ .
Nontrivial Markov Type Forces Superreflexivity Back to text
Pisier's power-type renorming theorem therefore gives nontrivial Markov type to every superreflexive space; Markov type is invariant under equivalent norms. Naor asked whether a Banach space could have nontrivial Markov type without an equivalent uniformly smooth norm.
Nontrivial Markov Type Forces Superreflexivity Back to text
Ordinary Rademacher type, which controls sums with independent random signs, does not suffice for this conclusion: James constructed nonreflexive spaces of Rademacher type $2$ .
Nontrivial Markov Type Forces Superreflexivity Back to text
Every real Banach space with Markov type $p$ for at least one $p>1$ is superreflexive.
Nontrivial Markov Type Forces Superreflexivity Back to text
We argue by contradiction.
Nontrivial Markov Type Forces Superreflexivity Back to text
From it we construct a spreading norm $N$ on finite sequences for which summing consecutive entries is contractive.
Nontrivial Markov Type Forces Superreflexivity Back to text
For each time $n$ , the central task is a finite reversible walk with unit edge labels, changing sign under reversal, whose sum has norm at least $n/24$ with probability at least $3/20$ . Before normalization, the labels are sums of signed differences of standard basis vectors in finite coefficient sequences.
Nontrivial Markov Type Forces Superreflexivity Back to text
With probability at least $1/4-1/10=3/20$ , all steps are good and $\ell\geq n/4$ .
Nontrivial Markov Type Forces Superreflexivity Back to text

Other sources

openai/math, Lean scope note for family 327 The formalized result proves that a real Banach space has nontrivial Markov type exactly when it admits an equivalent uniformly convex norm, hence exactly when it is superreflexive.
openai/math release notes (2026): model provenance and review status manuscripts and supporting proof artifacts produced by an internal OpenAI model

332Metric Markov cotype of ℓ 1 and Hilbert-space Lipschitz extensionFunctional analysis
211The geometric phase diagram, diffusion, and spectra of random planar mapsProbability and statistical mechanics
331Reflexive midpoint convexity and diamond distortionFunctional analysis
227Critical SK autocorrelation processes and dynamics across the temperature transitionProbability and statistical mechanics
326The cotype–cotype conjecture under the approximation propertyFunctional analysis

Browse Functional analysis Open since the 2010s Lean-checked results