Carnot rectifiability and Alberti representations (Q7034628)

The paper characterizes Carnot-rectifiability in terms of a Pansu-type differentiability condition and a strict dimensional condition on the measure. The Euclidean case was solved by \textit{D. Bate} and \textit{S. Li} [Ann. Sci. Éc. Norm. Supér. (4) 50, No. 1, 1--37 (2017; Zbl 1369.28002)]. In that case one considers \(n\)-rectifiability and differentiability in the sense of \textit{J. Cheeger} [Geom. Funct. Anal. 9, No. 3, 428--517 (1999; Zbl 0942.58018)]. The present paper adapts the techniques of Bate and Li and proves a wide-ranging and intriguing generalization.\N\NPansu differentiability spaces are spaces \(X\) that can be covered by charts \((U,\phi)\), where \(\phi:X\to \mathbb{G}\) is a Lipschitz map, and such that every Lipschitz map \(f:X\to \mathbb{H}\) admits a Pansu differential almost everywhere in \(U\), i.e., a group homomorphism \(df:\mathbb{G}\to \mathbb{H}\) such that\N\[\N\lim_{y\to x} \frac{d_{\mathbb{H}}(f(y),f(x)df(\phi(y)^{-1}\phi(x)))}{d(y,x)}=0.\N\]\NThe restricted charts \(\phi|_U\) need not be bi-Lispchitz or even homeomorphisms. On the other hand, \(\mathbb{G}\)-rectifiability corresponds to having such bi-Lipschitz maps. Such bi-Lipschitz maps, by Pansu's theorem [\textit{P. Pansu}, Ann. Math. (2) 129, No. 1, 1--60 (1989; Zbl 0678.53042)], yield Pansu differentiability. In short: rectifiability implies differentiability.\N\NWhat about the other direction? In general nothing holds, even in the case where \(\mathbb{G}=\mathbb{R}^n\), and the simplest examples of this are Laakso-spaces [\textit{T. J. Laakso}, Geom. Funct. Anal. 10, No. 1, 111--123 (2000; Zbl 0962.30006)]. Such counterexamples are in a sense ``bigger'' dimensional than their targets. However, if we additionally assume that \(X\) and \(\mathbb{G}\) have the same dimension, then strong restrictions for \(\phi\) ensue. More precisely, one needs to assume that the measure on \(X\) has an upper and lower \(Q\)-density bound, where \(Q\) is the homogeneous dimension of \(\mathbb{G}\). An argument then shows that charts \(\phi\) have big pieces of bi-Lipschitz maps, which yields rectifiability.\N\NPerhaps the main technical idea is hidden in how to construct \(df\). One can first construct a horizontal derivative \(df_H\) by differentiating \(f\) along horizontal curves (more precisely: curve fragments). Indeed, any Lipschitz function has a directional derivative almost everywhere along a curve. \(\mathbb{G}\)-differentiability implies \(\mathbb{R}\)-differentiability, and a result of \textit{D. Bate} [J. Am. Math. Soc. 28, No. 2, 421--482 (2015; Zbl 1307.30097)] gives a rich family of curves to differentiate on, i.e., Alberti representations. By a delicate measurability argument, one can select a basis of curves and define the horizontal part of \(df\) to agree with the directional derivatives that come from these curves.\N\NA second delicate part of the argument is extending \(df\) to all of \(\mathbb{G}\), since not every linear map from the horizontal bundle extends to a group homomorphism. However, curves coming from Alberti representations can be concatenated and this together with bracket generation and \(\mathbb{G}\)-differentiability guarantees a unique way of extending \(df\) to a group homomorphism. The concatenations do not connect every pair of points, but this is not a problem, since it is enough to find curve fragments \(\gamma\) such that \(\phi\circ \gamma\) connects \(\phi(x)\) to \(\phi(y)\). To pass back to the space \(X\) from the image, one uses completeness of charts. This argument is quite similar to the basic proof of calculus: a multi-variable function whose partial derivatives exist and are continuous is differentiable. One proves this by decomposing a diagonal movement into its horizontal components. Here, one decomposes a general movement to movements along curves coming from Alberti representations. \N\NThe paper also gives a generalization of the resolution of Cheeger's conjecture for Pansu differentiability spaces, and poses an intriguing conjecture on characterizing Carnot rectifiability in Conjecture 7.1. A reader interested in understanding the details of the paper might find it easier to first read the paper of Li and Bate [loc. cit.], since that paper introduces already all of the main arguments. The present paper adds the complications that come from the non-commutativity in Carnot groups. There are also many changes and new ideas needed to execute this.\N\NThe main arguments are contained in Sections 2, 3 and 6, and at first reading one might de-emphasize the somewhat delicate and tedious measurability issues. Measurability is important, but here it obscures some of the beautiful and conceptually simple main ideas.