This question popped up as I was thinking about intersection theory on fibered surfaces, which is slightly different from the situation I'm about to describe, and as a result I may have gotten some facts wrong. Anyway, here goes:
Let $X$ be a regular projective surface. If $D$ be an ample effective divisor on $X$, corresponding to an ample invertible sheaf $\mathcal{O}_X(D)$ on $X$. Then, for every closed curve $C\subset X$, $\mathcal{O}_X(D)|_C$ is ample on $C$, and hence has positive degree as an invertible sheaf on $C$. In terms of intersection theory, this means that $D\cdot C > 0$ for every closed curve $C\subset X$.
But...isn't the intersection pairing negative semidefinite? Hence, $D\cdot D\le 0$. At first, this seems like it would break the ampleness of $D$, and hence regular projective surfaces $X$ can't have any ample divisors, but this is plainly false since you can just take $X = \mathbb{P}^2$, and $D$ to be any prime divisor.
Where is the problem(s) with my logic??