Well this site isn't supposed to be us doing your homework but it's a simple question so let's have a crack at it:
We'll assume the points you make originally are valid (about the scoring).
Suppose the contrary, that there exists some local alignment which is less than G. If this were true, then it means your best local alignment (meaning you started somewhere away from G's starting or end points) is actually less efficient than your global alignment. But we know this can't be the case because the local alignment is a subset of your global alignment (worst case scenario, your local alignment IS your global alignment).
Therefore we prove that there are no counterexamples so this statement must hold.
Hope that makes sense!