1: \begin{abstract} %{{{

2: 

3: Hierarchical clustering is a popular method for analyzing data which associates

4: a tree to a dataset. Hartigan consistency has been used extensively as a

5: framework to analyze such clustering algorithms from a statistical point of

6: view. Still, as we show in the paper, a tree which is Hartigan consistent with a

7: given density can look very different than the correct limit tree. Specifically,

8: Hartigan consistency permits two types of undesirable configurations which we

9: term \emph{over-segmentation} and \emph{improper nesting}.  Moreover, Hartigan

10: consistency is a limit property and does not directly quantify difference

11: between trees.

12: 

13: In this paper we identify two limit properties, \emph{separation} and

14: \emph{minimality}, which address both over-segmentation and improper nesting and

15: together imply (but are not implied by) Hartigan consistency. We proceed to

16: introduce a \emph{merge distortion metric} between hierarchical clusterings and

17: show that convergence in our distance implies both separation and minimality. We

18: also prove that uniform separation and minimality imply convergence in the merge

19: distortion metric.  Furthermore, we show that our merge distortion metric is

20: stable under perturbations of the density.

21: 

22: Finally, we demonstrate applicability of these concepts by proving convergence

23: results for two clustering algorithms.   First, we show convergence (and hence

24: separation and minimality) of the recent robust single linkage algorithm of

25: \cite{chaudhuri_2010}. Second, we provide convergence results on

26: manifolds for  topological  split tree clustering.

27: \end{abstract}