1: \begin{abstract}

2: The adaptive regularization algorithm for unconstrained nonconvex

3: optimization was shown in \cite{NestPoly06,CartGoulToin11d} to

4: require, under standard assumptions, at most

5: $\mathcal{O}(\epsilon^{3/(3-q)})$ evaluations of

6: the objective function and its derivatives of degrees one and two to

7: produce an $\epsilon$-approximate critical point of order

8: $q\in\{1,2\}$. This bound was shown to be sharp in

9: \cite{CartGoulToin10a,CarmDuchHindSidf20} for $q=1$ and in

10: \cite{CartGoulToin22} for arbitrary $q \in\{1,2\}$. This note revisits

11: these results and shows that the example for which slow convergence is

12: exhibited is not isolated, but that this behaviour occurs for a subset

13: of univariate functions of nonzero measure.

14: \end{abstract}

15: