Abstract
Using dispersion relations and the convergence properties of t-channel isospin amplitudes, we show that (a) ΔI=2 electromagnetic mass difference should be correctly obtained by summing the self-energy contributions of a few low-lying states; (b) ΔI=1 mass differences cannot be obtained in this way, and a subtraction term is always necessary; (c) the subtraction term has the correct sign for explaining the proton-neutron mass difference.