1: \begin{abstract}

2: In this paper we examine the use of topological methods for multivariate

3: statistics.  Using persistent homology from computational

4: algebraic topology, a random sample is used to construct estimators of

5: persistent homology.  This estimation procedure can then be

6: evaluated using the bottleneck distance between the estimated

7: persistent homology and the true persistent homology.  The connection

8: to statistics comes from the fact that when viewed as a nonparametric

9: regression problem, the bottleneck distance is bounded by the sup-norm

10: loss.  Consequently,

11: a sharp asymptotic minimax bound is determined under the sup--norm risk

12: over H\"{o}lder classes of functions for the nonparametric regression

13: problem on manifolds.  This provides good convergence properties for

14: the persistent homology estimator in terms of the expected bottleneck

15: distance.

16: \end{abstract}

17: