1: \begin{abstract}

2: 

3: The $k$-monotone classes of densities defined on $(0, \infty)$ have been known 

4: in the mathematical literature but were for the first time considered from a statistical point of view by 

5: \cite{MR2382657} and \cite{MR2830965}. 

6: In these works, the authors generalized the results established for 

7: monotone ($k=1$)  and convex ($k=2$) densities by giving a 

8: characterization of the Maximum Likelihood and Least Square 

9: estimators (MLE and LSE) and deriving minimax bounds for rates 

10: of convergence.  For $k \ge 3$, the pointwise asymptotic behavior of the 

11: MLE and LSE 

12: studied by

13: %given by 

14: \cite{MR2382657} %\cite{bw10} 

15: would show that the MLE and LSE attain the minimax lower 

16: bounds in a local pointwise sense. 

17: %are minimax efficient

18: However, the theory assumes that a certain conjecture about the 

19: approximation error of a Hermite spline holds true. 

20: The main goal of the present  note is to 

21: show why such a conjecture cannot be true. 

22: We also suggest how to bypass the conjecture and rebuilt the key proofs in the limit theory of the estimators.

23: \end{abstract}

24: