A good example would be an expression morph designed to work with a custom face morph. If you create the expression over top the custom face and apply it as a morph target and dial it to 1, the custom face is baked into the expression morph and applied again. Telescoping. The custom face is doubled and surely won't look right when you dial in the expression. But if you subtract the custom face from the expression morph, which was built in relation to that custom face, and apply the result you are left with just the expression. Works properly. It's the difference derived through subtraction.

PML and GoZ can do this. You can also do it manually by dialing the baked in morph's original dial to -1, the new morph (that was created with the original baked in) to 1, export the result as a morph target, apply that morph target and the baked in morph will be subtracted from the resulting morph. So when you activate that morph it will behave accordingly in relation to the morph that was initially present without applying that reference morph again.