1: \begin{abstract}

2: In proofs of $L_2$-differentiability, 

3: Lebesgue densities of a central distribution are often assumed right from 

4: the beginning. Generalizing  \citet[Theorem~4.2]{Hu:81}, we show that in the class of smooth

5: parametric group models these densities are in fact consequences of

6: a finite Fisher information of the model, provided a suitable representation

7:   of the latter is used. The proof uses  the notions of absolute continuity

8: in $k$ dimensions and weak differentiability. \\

9: As examples to which this theorem applies, we spell out a number of models

10: including a correlation model and the general multivariate location and

11: scale model.\\

12: %

13: As a consequence of this approach, we show that in  the (multivariate)

14: location scale model, finiteness of Fisher information as defined here

15: is in fact equivalent to $L_2$-differentiability and

16:  to a log-likelihood expansion giving local asymptotic normality of the model.\\

17: %

18: Paralleling Huber's proofs for  existence and uniqueness of a minimizer of

19: Fisher information to our situation, we get existence of a minimizer in

20: any weakly closed set ${\cal F}$ of central distributions $F$.

21: If, additionally to analogue assumptions to those of \citet{Hu:81},

22:  a certain identifiability condition for the transformation

23:  holds, we obtain uniqueness of the minimizer.

24:  This identifiability condition is satisfied in the multivariate location

25:  scale model.

26: \end{abstract}