1: \begin{abstract}

2: We introduce and study the Group Square-Root Lasso (GSRL)  method for

3: estimation in high dimensional sparse regression models with group

4: structure. The new estimator minimizes the square root of the residual

5: sum of squares plus a penalty term proportional to the sum of the

6: Euclidean norms of groups of the regression parameter vector. The net

7: advantage of the method over the existing Group Lasso (GL)-type

8: procedures consists in the form of the proportionality factor used in

9: the penalty term, which for GSRL is independent of the variance of the

10: error terms. This is of crucial importance in models with more

11: parameters than the sample size, when estimating the variance of the

12: noise becomes as difficult as the original problem.  We show that the

13: GSRL estimator adapts to the unknown sparsity of the regression

14: vector, and has the same optimal estimation and prediction accuracy as

15: the GL estimators, under the same minimal conditions on the

16: model. This extends the results recently established for the

17: Square-Root Lasso, for sparse regression without group

18: structure. Moreover, as a new type of result for Square-Root Lasso

19: methods, with or without groups, we study correct pattern recovery,

20: and show that it can be achieved under conditions similar to those

21: needed by the Lasso or Group-Lasso-type methods, but with a simplified

22: tuning strategy.  We implement our method via a new algorithm, with

23: proved convergence properties, which, unlike existing methods, scales

24: well with the dimension of the problem.  Our simulation studies

25: support strongly our theoretical findings.

26: \end{abstract}

27: