Recall that $n\in\mathbb{N}$ is $k$-full if, for a prime number $p$ : $p\mid n$ implies $p^{k}\mid n$. In the paper I am currently reading it is said that there are $O(B^{1/k})$ $k$-full positive integers in the dyadic interval $(B/2,B]$ for any $B\geq 1$. How to obtain this bound?
Obviously, any $k^{\text{th}}$ power is a $k$-full number, so there are at least $\approx B^{1/k}$ $k$-full numbers in $(0,B]$, and there are $O(B^{1/k})$ powers in $(B/2,B]$. Note that if number $n$ is $k$-full, then $n = x^{k}y,\;y\mid x$, so $B/2<x^{k}y\leq B$ and $y\leq x/2$, so $B<x^{k+1}$, so $x>B^{\frac{1}{k+1}}$ and clearly $x\leq B^{\frac{1}{k}}$. Thus, we are left with $B^{1/k} - B^{1/(k+1)}$ choices for $x$, which is still bad, because it is hard to find bound for choices of $y$. Trivial bound would be something like $O(B^{1/k+1})$, where $+1$ comes from the bound for choices of $y$. Maybe on average there are constant number of choices for $y$, in which case we get the desired $O(B^{1/k})$, but I don't know how to make the estimate. Any help is appreciated.