Differentiability and Poincaré-type inequalities in metric measure spaces (Q1649353)

In his seminal paper [Geom.\ Funct.\ Anal. 9, No. 3, 428--517 (1999; Zbl 0942.58018)], \textit{J. Cheeger} showed that in certain metric spaces, Lipschitz functions are differentiable almost everywhere. An important part of this statement is making sense of the notion of differentiability. The metric spaces considered by Cheeger support Poincaré inequalities, and this initiated a search in how this condition can be relaxed. Here, I would like to mention \textit{S. Keith}'s work [Adv.\ Math. 183, No. 2, 271--315 (2004; Zbl 1077.46027)], in which the author showed that there is a condition weaker than requiring a Poincaré inequality that still implies Cheeger's conclusion. Another direction of generalization of Cheeger's result has been pursued by Cheeger together with Bruce Kleiner. They showed that Cheeger's conclusion also holds for RNP-targets, that is for vector spaces that satisfy the Radon-Nikodym property. There are many equivalent definitions of this property, but for our purposes the most convenient is that a vector space has the Radon-Nikodym property if all Lipschitz maps from the real line to the space are differentiable almost everywhere. Having the domain and the target generalized separately, the question is whether one can generalize both at the same time. This is where the paper under review comes in. The authors show that a metric measure space is a RNP-LDS (a Lipschitz differentiability space with the RNP) if and only if it satisfies an asymptotic nonhomogeneous Poincaré inequality (a term coined by the authors) and all porous sets have measure zero. \textit{S. Eriksson-Bique} [``Classifying Poincaré Inequalities and the local geometry of RNP-differentiability spaces'', Preprint, \url{arXiv:1607.07428}] showed that spaces with an asymptotic nonhomogeneous Poincaré inequality are countable unions of PI spaces (doubling metric spaces satisfying a Poincaré inequality) up to a set of measure zero. The reader might also want to have a look at [\textit{A. Schioppa}, ``An example of a differentiability space which is PI-unrectifiable'', Preprint, \url{arXiv:1611.01615}], which shows that LDS and RNP-LDS can be different. After the introduction, the authors lie down the preliminaries. There I would like to point out that the authors show that a RNP-LDS has a nice decomposition into sets with Alberti representations. The next section shows a variant of quasiconvexity for RNP-LDS. The authors continue with proving that RNP-LDS satisfy an inequality reminiscent of the Poincaré inequality. They turn it into a definition: asymptotic non-homogeneous Poincaré inequality. The main point of the rest of the section is to show that such Poincaré inequalities imply that the space is a RNP-LDS. For a function \(f\) from a metric space \(X\) to a Banach space \((V,\|\cdot\|)\), we recall the following definition \[ \text{Lip} f(x)=\limsup_{r\to 0+}\sup_{y\in B(x,r)} \frac{\| f(y)-f(x)\|}{r}. \] The next section's objective is to prove a result alluded earlier in the paper roughly stating that if \(\text{Lip}\) is small almost everywhere in a set \(S\), then it has to be small everywhere. The proof involves constructing an RNP-valued Lipschitz function that is not differentiable in certain points; these points then need to form a set of measure zero. Next, the authors look at tangent spaces and show that the tangent spaces at almost all points of an RNP-LDS are quasiconvex RNP-LDS as well. The end of the paper is devoted to a second characterization of RNP-LDS, but this time in terms of Alberti representations.