In his piece on $\pi^{\pi^{\pi^\pi}}$, Matt Parker plots a graph showing that if $\pi$ is approximated to $n$ digits, then whenever $m \leq 2 n$ (roughly), $\pi^m$ is correct to the nearest whole number.
Is there any theoretical underpinning for this?
Suppose $p>0$ is an approximation of $\pi$ with error $\epsilon$ (say $\epsilon=|p-\pi|$). The first digit where $p$ and $\pi$ disagree is in the $10^{\lceil\log_{10}\epsilon\rceil}$'s column.
Now we can trivially bound the error between $\pi^n$ and $p^n$ since $$|p^n-\pi^n|=\epsilon \cdot \bigg(\sum_{k=0}^{n-1}p^{k}\pi^{n-k-1} \bigg) < \epsilon \cdot n \cdot \max(\pi,p)^{n-1}.$$ Thus they definitely agree up to the $10^{(n-1)\lceil\log_{10}(\epsilon n\max(\pi,p))\rceil}$'s column.