1: \begin{abstract}  

2: We consider data from the Grassmann manifold $G(m,r)$ of all vector subspaces of dimension $r$ of $\R^m$, and  focus on the  Grassmannian statistical model.

3: Canonical Grassmannian distributions $\mathbb{G}_\sig$ on $G(m,r)$

4:  are indexed by parameters $\Sigma$ from the manifold 

5: $\mathcal{M}= \pas$ of positive definite symmetric matrices of determinant $1$. Robust  M-estimates of scatter (GE) for general probability measures $\shp$ on $G(m,r)$ are studied. Such estimators 

6: are defined to be the maximizers

7: of

8: the Grassmannian log-likelihood 

9: $-\ell_\shp(\sig)$ as  function of $\Sigma$. One of the novel features of this work is a strong use of the fact that $\mathcal{M}$ is a CAT(0) space with known visual boundary at infinity $\partial \mathcal{M}$, allowing us to study existence and unicity of GEs; along the way it is proved that 

10: $\ell_\shp(\gamma(t))$ is convex (and under further conditions even strictly convex) when evaluated on geodesics $\gamma(t)$ of $\mathcal{M}$. We also recall that the sample space $G(m,r)$ is a part of  $\partial \mathcal{M}$, show the distributions $\G_\sig$ are $\SL(m, \RR)$--quasi-invariant, and

11: that $\ell_\shp(\sig)$ is a weighted Busemann function. Let $\shp_n =(\delta_{U_1}+\cdots+\delta_{U_n})/n$ be the empirical probability measure for $n$-samples of random i.i.d. subspaces $U_i\in G(m,r)$ 

12: of common distribution $\shp$ with support $G(m,r)$. For $\Sigma_n$ and $\Sigma_\shp$ the \gmle s\ of $\shp_n$ and $\shp$, we show the almost sure convergence of $\Sigma_n$ toward $\Sigma$ as $n\to\infty$ using methods from geometry, and provide a central limit theorem for the rescaled process 

13: $C_n = \frac{m}{\tr(\Sigma_{\shp}^{-1} \Sigma_n)}g^{-1} \Sigma_n g^{-1}$,  where $\Sigma =g g$ with $g \in \SL(m, \RR)$ the unique symmetric positive-definite square root of $\Sigma$.

14: 

15: 

16: \end{abstract}

17: