1: \begin{abstract}

2: 

3: We present a renormalization group (RG) approach to explain

4: universal features of extreme statistics, applied here

5: to independent, identically distributed

6: variables. The outlines of the theory have been described in

7: a previous Letter, the main result being that finite-size

8: shape corrections to the limit distribution can be obtained

9: from a linearization of the RG transformation near a

10: fixed point, leading to the computation of stable perturbations as

11: eigenfunctions.  Here we show details of the RG theory which exhibit

12: remarkable similarities to the RG known in statistical physics.

13: Besides the fixed points explaining universality, and the least

14: stable eigendirections accounting for convergence rates and

15: shape corrections, the similarities include

16: marginally stable perturbations

17: which turn out to be generic for the Fisher-Tippett-Gumbel class.

18: Distribution functions containing unstable perturbations are

19: also considered. We find that, after a transitory divergence, 

20: they return to the universal fixed line at the same or at a different 

21: point depending on the type of perturbation. 

22: \end{abstract}

23: