BiLipschitz decomposition of Lipschitz maps between Carnot groups (Q496239)

The main result of this paper is that a Lipschitz map between two Carnot groups \(G\), \(H\), endowed with their Carnot-Carathéodory metrics, is ``almost'' bi-Lipschitz. Starting with a ball of \(G\), one can remove a set whose image has small Hausdorff content, and decompose what remains into a fixed finite number of closed sets, such that the restriction onto each such set is bi-Lipschitz. To be precise, we quote here the paper's main Theorem 1.1 (with slight notational changes): {Theorem.} Let \(G\) and \(H\) be Carnot groups endowed with their Carnot-Carathéodory metrics \(d_{G,cc}, d_{H,cc}\), and let \(N\) be the Hausdorff dimension of \(G\). There exists some \(c > 0\), depending only on \(G\) and \(H\), with the following property: For each \(\delta > 0\), there exists some \(M = M(\delta) > 1\) so that if \(x \in G\), \(R > 0\), and \(f : B(x,R) \to H\) is a \(1\)-Lipschitz function, then there exist closed sets \(\{F_i\}_{i=1}^M\) of \(B(x,R)\) so that \[ \mathcal{H}_\infty^N\left(f\left(B(x,R) \setminus \bigcup_{i=1}^M F_i \right) \right) < c \delta R^N, \] and \[ d_{H,cc}(f(x), f(y)) \geq \delta d_{G,cc} (x,y), \qquad \forall x,y \in F_i, \forall i \in \{1,\dots,M\}. \] Here \(\mathcal{H}^N_\infty\) denotes the \(N\)-dimensional Hausdorff content in \(H\). Note that in the paper, the last line is written as \(| f(x) - f(y)| \geq \delta | x-y| \); but the author has confirmed to the reviewer in personal communication that this is meant to refer to the Carnot-Carathéodory metrics (or any equivalent metrics), not the Euclidean metrics. A key step in the proof is to equip \(H\) with a carefully constructed metric \(d_H\), different from the Carnot-Carathéodory metric \(d_{H,cc}\) yet bi-Lipschitz equivalent to it, for which the desired properties are easier to prove. The metric \(d_H\) is actually induced by a certain homogeneous norm on \(H\). This paper's main result was previously obtained in [\textit{W. Meyerson}, Pac. J. Math. 263, No. 1, 143--170 (2013; Zbl 1295.43009)] for groups \(G\) which are \textit{discretizable}; meaning, roughly speaking, that there exists a discrete subgroup generated by horizontal elements whose linear span is \(G\). (Recall that in a Carnot group \(G\), the exponential map is a diffeomorphism, and therefore \(G\) inherits a natural vector space structure from its Lie algebra.) Thus, the novelty of the present paper is to remove the assumption of discretizability. Proposition 2.2 demonstrates that this is a real improvement by giving an explicit example of a Carnot group of step 6 which is not discretizable and therefore not covered by Meyerson's result.