The following is a problem in our exam which I couldn't get really far with.
Let $F$ be a field and $E$ a non-trivial extension of $F$ such that for all $x \in E$ there exists $n_x \in \mathbb{N}$ such that $x^{n_x} \in F$, prove that $Char(F) \neq 0$
My progress:
Let $x \in E/F$ and $x^p \in F$, there exists $n \in \mathbb{N}$ such that $(x+1)^n,(x+2)^n,\dots,(x+p)^n \in F$, this means the $p$ fold iteration of the discrete derivative on $f(x)=x^n$ gives $0$.
This was as far as I went, Hints and solutions are appreciated.
I’m pretty sure that this argument isn’t optimal, but it gives a complete solution. I think you can also extend it to show more generally that in general, if $F$ is infinite, then $E/F$ is purely inseparable.
Assume that $F$ has characteristic zero and $E \neq F$. Note that $E/F$ is algebraic.
Let $x \in E \backslash F$. Then, there is a $F$-embedding $\sigma: E \rightarrow \Omega$ (where $\Omega$ is algebraically closed) such that $\sigma(x) \neq x$.
By the hypothesis, we see that $\sigma(x)=\omega x$ for some root of unity $\omega \neq 1$.
For the same reason, we know that $\omega x+1=\sigma(x+1)=\omega’ (x+1)$ for some root of unity $\omega’ \neq 1$. It follows that $\omega \neq \omega’$ and thus $x=\frac{\omega’-1}{\omega-\omega’}$.
In particular, since $E \neq F$, $E$ is contained in the field generated by the roots of unity over $\mathbb{Q}$. Hence $E/F$ is automatically abelian.
Next, pick some $y \in E \backslash F$ (which is algebraic over $\mathbb{Q}$) and let $E’=\mathbb{Q}[y]$, $F’=\mathbb{Q}[y] \cap F$. It’s easy to see that the extension $E’/F’$ still works (and thus is still abelian).
The interest of this construction is that $E’$ has finitely many roots of unity.
Since $y \notin F’$, there exists $\sigma \in \mathrm{Gal}(E’/F’)$ such that $\sigma(y) \neq y$.
Then, for every integer $n$, $u_n=\frac{\sigma(y+n)}{y+n}$ is a root of unity contained in $E$ and distinct from $1$.
Choose two distinct integers $n,m$ such that $u_n=u_m$. Then $(\sigma(y)+n)(y+m)=(\sigma(y)+m)(y+n)$, hence $ny+m\sigma(y)=my+n\sigma(y)$, which implies $y=\sigma(y)$, a contradiction.