1: \begin{abstract}

2: This study aims at contributing to lower bounds for 

3: empirical compatibility constants or empirical restricted

4: eigenvalues. This is of importance in compressed sensing and

5: theory for $\ell_1$-regularized estimators. Let $X$ be an $n \times p$ data matrix

6: with rows being independent copies of a $p$-dimensional random variable.

7: Let $\hat \Sigma := X^T X / n$ be the inner product matrix.

8: We show that the quadratic forms $u^T \hat \Sigma u$ are lower bounded by a value converging to one,

9: uniformly  over the set

10: of vectors $u$ with $u^T \Sigma_0 u $ equal to one and $\ell_1$-norm at most $M$.

11: Here $\Sigma_0 := \EE \hat \Sigma$ is the theoretical inner product matrix which we assume to exist. 

12: The constant $M$ is required to be of small order $\sqrt {n / \log p }$.

13: We assume moreover $m$-th order isotropy for some $m >2$ and sub-exponential

14: tails or moments up to order $\log p$  for the entries in $X$. 

15: As a consequence we obtain convergence of the empirical

16: compatibility constant to its theoretical counterpart, and

17: similarly for the empirical restricted eigenvalue. If the data matrix $X$

18: is first normalized so that its columns all have equal length we obtain lower bounds assuming only isotropy and

19: no further moment conditions on its entries.

20: The isotropy condition is shown to hold for

21: certain martingale situations. 

22: \end{abstract}

23: