1: \begin{abstract}

2: Under the assumption that the true density is decreasing, it is well

3: known that the

4: Grenander estimator converges at rate $n^{1/3}$ if the true density is curved

5: [\textit{Sankhy\=a Ser. A} \textbf{31} (1969) 23--36]

6: and at rate $n^{1/2}$ if the density is flat

7: [\textit{Ann. Probab.} \textbf{11} (1983) 328--345;

8: \textit{Canad. J. Statist.} \textbf{27} (1999) 557--566].

9: In the case that the true density is misspecified, the results of

10: Patilea [\textit{Ann. Statist.} \textbf{29} (2001) 94--123]

11: tell us that the

12: global convergence rate is of order $n^{1/3}$ in Hellinger distance.

13: Here, we show that the local convergence rate is~$n^{1/2}$ at a point

14: where the density is misspecified. This is not in contradiction with the results

15: of Patilea [\textit{Ann. Statist.} \textbf{29} (2001) 94--123]:

16: the global convergence rate simply comes from locally curved \emph{well-specified} regions.

17: Furthermore, we study global convergence under misspecification by

18: considering linear functionals.

19: The rate of convergence is $n^{1/2}$ and we show that the limit is made

20: up of two independent terms:

21: a mean-zero Gaussian term and a second term (with nonzero mean) which

22: is present only if the

23: density has well-specified locally flat regions.

24: \end{abstract}