# Expansions by cube roots of 1 (mod n)

Fix some integer $n$. I would like to know when it is possible to find a cube root of 1 mod $n$, say $r$, so that it is possible to write

$m = a(r^{2} - 1) + b(r - 1) + c$ mod n

for all $0 \leq m \leq n-1$, with coefficients $a$, $b$ and $c$ that are `small'. I am not so picky about what small should mean, but am broadly looking for theorems like 'for $n$ in infinite family $A$, it is possible to find such $r$ with all coefficients $a,b,c = O(n^{0.6})$'. I'm also happy for theorems stating that there are no such $r$.

Thanks for any help.

EDIT in response to Kevin Buzzard: Thanks for the response, and I'm sorry for being unclear. I agree that that is one example of an infinite family of choices of $n$ and $r$ with good properties. I am trying to understand what types of behaviour are possible, and maybe it would be especially nice to know about an infinite family that was fairly dense. Maybe there is even a literature that would help one figure out this type of question - certainly I'm having a lot of trouble trying to figure out plausible guesses, or what values of $n$ are likely to be in families together, and so on.

The small number of other families I know about are:

Good:

1) Small modifications of your example (e.g. $A$ is parameterized by odd r, $n = \frac{r^{3} - 1}{2}$)

Isn't this just trivial: can't I let $r$ be any positive integer, set $n=r^3-1$ and then I can get $a,b,c=O(r)=O(n^{1/3})$? –  Kevin Buzzard Mar 24 '12 at 17:55
If $n$ is prime and $\equiv 1 \mod 3$, you don't need $r^2$ since $r^2+r+1=0$. So you can take $a=0$ and you can have $b,c = O(n^{1/2})$, since if two such numbers are congruent mod $p$, then their difference is zero and taking the norm of the corresponding Eisenstein integer, gives you a contradiction. So there is your $A$.