let A be an abelian variety of dimension g over an algebraically closed field k of characteristic p ≥ 0. I'm trying to prove that the subgroup A' which is the union of all torsion points a ∈ A(k) of order prime to p is Zariski dense in A.
The statement would follow if the Zariski closure B (which by construction is a group variety) of A' in A would again be an abelian variety of dimension d, because assuming d<g, the \ell-primary part of B would still be (\Q\ell/\Z_\ell)2g, while it SHOULD be of rank 2d<2g, contradiction.
However, I fail to see why B should be irreducible. Does anyone see a way to salvage the argument, or a different, (simpler) argument?