Translation averages of dyadic weights are not always good weights. (Q1809939)

The translation average \(\varphi(x)=\int_0^1 \varphi_t(x+t)\, dx\) of a collection \(\varphi_t\) of functions tends to improve certain properties of these functions, and it is known that translation averaging of dyadic BMO functions produces true BMO functions. In this paper it is proved that this is not the case for a number of dyadic properties; an appropriate construction, similar for the three cases, is used to show that improvement fails for dyadic doubling measures, for dyadic reverse Hölder weights, and for dyadic Muckenhoupt weights.