It's just a rough (but certain for all n) upper bound. The precise probability for your case seems to be slightly greater than 0.01. The point is that you can r

It's just a rough (but certain for all n) upper bound. The precise probability for your case seems to be slightly greater than 0.01. The point is that you can replace a precise, expensive equality test with a second hash with greater range.