ahem, yes very simple I guess, given all read come only from this gene, that's what you want right?

let r be the read length, g be the gene length, r<=g then each read contributes coverage c = r/g. So you need N/(r/g) reads. E.g. if r=g and aimed coverage is 1 then you need 1 read. Let alone those reads that do not fully overlap the gene, they contribute ofc only partially. If you want to include these into the calculation and want the exact coverage, you can only do it after you placed each read randomly and then recompute the coverage.

Or did I miss something completely??