Let $\sigma_r(n)=\sum_{d|n}d^r$ where the sum is over all the integers $d=1,\dots,n$ which divide $n$. I am conjecturing
$$\sum_{m=1}^{p-1}m^2 \sigma_3(m)\sigma_3(p-m)\not\equiv 0\pmod{p^2}$$ for any prime $p>3$.
I have checked the truth of this statement up to $p\sim 30'000$ with some C++ program I wrote, but going above that would require to use some arbitrary precision library (which I tried, but failed).
Can anyone test the conjecture for some higher bound? Also, any ideas on how to attack the problem/any semi-obvious reason why this shouldn't be true?
Without going too much into details, this formula comes from fiddling with the Fourier coefficients of the Eisenstein series for $SL_2(\mathbb{Z})$. I have no reason to assume the conjecture is true (or false) but I have absolutely no idea on how to start working on it, mostly because it mixes "additive" (sum) and "multiplicative" ($\sigma$) number theory, so I am not actually expecting it to be easy.