1: \begin{abstract}

2: Fr\'echet means of samples from a probability measure~$\mu$ on any

3: smoothly stratified metric space~$\MM$ with curvature bounded above

4: are shown to satisfy a central limit theorem (CLT).  The methods and

5: results proceed by introducing and proving analytic properties of the

6: \emph{escape vector} of any finitely supported measure~$\delta$

7: in~$\MM$, which records infinitesimal variation of the Fr\'echet

8: mean~$\bmu$ of~$\mu$ in response to perturbation of~$\mu$ by adding

9: the mass $t\delta$ for $t \to 0$.  The CLT limiting distribution~$\cN$

10: on the tangent cone~$\Tmu$ at the Fr\'echet mean~$\bmu$ is

11: characterized in four ways.  The first uses tangential collapse~$\LL$

12: to compare $\Tmu$ with a linear space and then applies a distortion

13: map to the usual linear CLT to transfer back to $\Tmu$.  Distortion

14: % generalizes the inverse Hessian of the Fr\'echet function~$F_\mu$,

15: % since the Hessian is singular in this non-smooth setting, and

16: is defined by applying escape after taking preimages under~$\LL$.  The

17: second characterization constructs singular analogues of Gaussian

18: measures on smoothly stratified spaces and expresses~$\cN$ as the

19: escape vector of any such \emph{Gaussian mass}.  The third

20: characterization expresses~$\cN$ as the directional derivative, in the

21: space of measures on~$\MM$, of the barycenter map at~$\mu$ in the

22: (random) direction given by any Gaussian mass.  The final

23: characterization expresses~$\cN$ as the directional derivative, in the

24: space~$\cC$ of continuous real-valued functions on~$\Tmu$, of a

25: minimizer map, with the derivative taken at the Fr\'echet function $F

26: \in \cC$ along the (random) direction given by the negative of the

27: Gaussian tangent field induced by~$\mu$.  Precise mild hypotheses on

28: the measure~$\mu$ guarantee these CLTs, whose convergence is proved

29: via the second characterization of~$\cN$ by formulating a duality

30: between Gaussian masses and Gaussian tangent fields.

31: \end{abstract}

32: