Failure of the linear $A_2$ bound for the strong maximal operator

We show that the strong maximal operator on ${\mathbb{R}}^2$ does not satisfy a linear weighted $L^2$ estimate in terms of the rectangular $A_2$ characteristic. More precisely, we construct weights with arbitrarily large characteristic for which the operator norm is bounded below by $[w{]}_{A_2^{\mathrm{s}\mathrm{t}\mathrm{r}}}\sqrt{\log [w{]}_{A_2^{\mathrm{s}\mathrm{t}\mathrm{r}}}}$. The proof is elementary. It uses an explicit two-parameter mass recurrence on a product of geometric interval partitions, together with a direct comparison of arbitrary rectangular averages with anchored averages.