Question

At the 2010 Clay Research Conference, Gromov explained that we know of only 7 different methods for constructing smooth manifolds. Working from memory, and hence not necessarily respecting the order he used:

• Algebraic geometry (affine and projective varieties, ...)
• Lie groups (homogeneous spaces, ...)
• General position arguments (Morse theory, Pontryagin-Thom construction, ...)
• Solutions to PDE (Moduli spaces in gauge theory, Floer theory, ...)
• Surgery (Cut and paste techniques, ...)
• Markov processes

I realise that I only gave 6 constructions; this was the number of separate items listed on his slides, and since he failed to discuss this part, I am left to guess that he either listed two different constructions on one line, which I interpreted to be variants of the same construction, or that failed to include one altogether.

Question How does one construct a smooth manifold from Markov processes?

I asked Gromov after the talk for explanation, but due to the rudimentary nature of my Gromovian, I was unable to understand the answer. The only word I managed to parse is "hyperbolic," though I wouldn't put too much stock in that.

Answer 1

Unfortunately I missed the talk, but on the other hand Gromov have just produced a new paper called

Manifolds : Where do we come from ? What are we ? Where are we going ?

It can be found on his web page. From the title I guess there could be some intersection with the talk. In particular in section 11 called Crystals, Liposomes and Drosophila Gromov is speaking about "Markov quotients". This sounds like a way to produce "spaces" (generalisation of manifolds, I guess).

http://www.ihes.fr/~gromov/PDF/manifolds-Poincare.pdf

Answer 2

I suspect (but am far from certain) that Gromov may be referring to the correspondence between symbolic and hyperbolic dynamics.

The idea is basically that the 0-1 matrix corresponding to the sparsity pattern of a stochastic matrix encodes a subshift of finite type or topological Markov chain. Usually, however, one goes from the hyperbolic dynamics to the Markov description via a Markov partition or section.

I am not aware of a way to go in the other direction in general, although placing certain conditions on the Markov process would facilitate the construction of a Markov partition (which can then be made as small as one likes), for which covering sets would constitute an atlas.

---

Update: So I did a little digging and came across a paper by Coornaert and Papadopoulos called "Symbolic coding for the geodesic flow associated to a word hyperbolic group" (Manuscripta Math. 109, 465–492 (2002), DOI 10.1007/s00229-002-0321-9, PDF available here). In it the authors discuss an idea of Gromov whereby a to each "word hyperbolic group" $\Gamma$ a space with a flow defined up to orbit equivalence is given: this flow is called the geodesic flow associated to $\Gamma$. I quote:

In the case where $\Gamma$ is the fundamental group of a compact Riemannian manifold $M$ of negative curvature, then $\Gamma$ is word hyperbolic and [the geodesic flow associated to $\Gamma$] is, up to orbit equivalence, the geodesic flow on the tangent bundle of $M$.

Nowhere, however, is it indicated that the space so constructed is generically a manifold. Still, this construction is quite closely associated with the ideas mentioned earlier, as the introduction to this paper points out.