UMD 808J: Algebraic Surfaces

3.5 Properties of the intersection pairing related to fibers

Here \(X\) is a smooth, projective surface. The goal is to study the intersection pairing on the components of the fiber of a morphism \(f:X\to Y\) in the case when the fiber is a divisor. Curiously, we don’t really use the Hodge Index Theorem.

Lemma 3.33

Suppose \(V\) is a \(\mathbb {Q}\)-vector space with a symmetric bilinear form \((x,y)\mapsto x\cdot y\), and suppose \(\{ e_i\} _{i\in I}\) generated \(V\) and with \(e_i\cdot e_j\geq 0\) for all \(i\neq j\). Suppose further that there exists a \(z = \sum a_i e_i\) with \(a_i {\gt} 0\) such that \(z\cdot e_i = 0\) for all \(i\).

Then \(x\cdot x \leq 0\) for all \(x\in V\) and \(W := \{ x\in V: x\cdot x = 0\} \) is a liner subspace. Moreover, if the \(\{ e_i\} _{i\in I}\) form a basis, then the dimension of \(W\) is the number of connected components of the graph whose vertices are the elements of \(I\) and whose edge \(\{ i,j\} \) correspond to pairs \(\{ i,j\} \) with \(e_i\cdot e_j {\gt} 0\).

Proof

We can replace \(e_i\) with \(a_i e_i\) and then assume that \(z = \sum e_i\). Then pick \(x\in V\). We have \(x = \sum c_i e_i\) with \(c_i\in \mathbb {Q}\). Then we have

\begin{align*} x\cdot x & = \sum _i c_i^2 e_i\cdot e_i + 2 \sum _{i{\lt}j} c_i c_j e_i\cdot e_j\\ & \leq \sum _i c_i^2 e_i\cdot e_i + \sum _{i{\lt}j} (c_i^2 + c_j^2) e_i\cdot e_j\\ & = \sum _{i,j} c_i^2 (e_i\cdot e_j) = \sum _i c_i^2 z\cdot e_i = 0. \end{align*}

Now, suppose the \(e_i\) are linearly independent. Then the first inequality is strict if and only if \(2 c_i c_j = c_i^2 + c_j^2\) for all \(i\neq j\), and this holds if and only if \(e_i\cdot e_j\neq 0 \Rightarrow c_i = c_j\). That implies the last two statements.

Corollary 3.34

Suppose \(f: X\to C\) is a morphism from \(X\) to a nonsingular curve \(C\), and let \(D = [f^{-1}(y)]\) be the fiber of \(D\). Write \(D = \sum a_i [D_i]\) where the \(D_i\) are prime divisors and with the \(a_i{\gt}0\). Then, for every divisor of the form \(D' = \sum a_i D_i\) with \(a_i\in \mathbb {Z}\), we have \((D')^2 \leq 0\). Moreover, if \(f\) has connected fibers then \((D')^2 = 0\) if and only if there exists \(a\in \mathbb {Q}\) with \(D' = aD\).

Proof

Since \(C\) a curve, \(f\) is actually flat. So \(f^{!} = f^*\) and \(f^*[y] = D\). If we pick another point \(y^{\dagger }\in C\) and set \(D^{\dagger } = f^{*}y^{\dagger }\), then, since \([y]\) and \([y^{\dagger }]\) are numerically equivalent, so are \(D\) and \(D^{\dagger }\). So, \([D][D_i] = [D^{\dagger }][D_i] = 0\) for all \(i\). On the other hand, \(D_i\cdot D_j \geq 0\) for all \(i,j\) because the \(D_i\) are distinct. The result then follows from the Lemma 3.33. If we pick another point \(y^{\dagger }\in C\) and set \(D^{\dagger } = f^{*}[y^{\dagger }]\), then, since \(y\) and \(y^{\dagger }\) are numerically equivalent, so are \(D\) and \(D^{\dagger }\). Therefore, that \(D_i\cdot D = D_i\cdot D^{\dagger } = 0\) for all \(i\).

So, in Lemma 3.33, take \(V\) to be the free vector space generated by the divisors \([D_i]\) and take \(z = D\).

kk

Remark 3.35

If \(f:X\to C\) is a morphism as in Corollary 3.34, then, by Stein factorization, we can factor \(f\) as \(X\xrightarrow {g} C' \xrightarrow {h} C\) where \(g\) has connected fibers and \(h\) is finite.

Corollary 3.36

Suppose \(f:X\to Y\) is a birational morphism from a smooth surface \(X\) to a normal surface \(Y\). Let \(y\in Y\) be a closed point and suppose that \([f^{-1} y] = \sum _{i=1}^r E_i\), where the \(E_i\) are pairwise distinct prime divisors on \(X\). Then the intersection matrix \(E_i\cdot E_j\) is negative definite.

Proof

Pick a very ample divisor \(H\) on \(X\). Using Bertini  [ 4 , Theorem II.8.18 and III.7.9.1 ] , we can find an integral curve \(C\in |H|\), which is distinct from all the divisors \(E_i\). Let \(\bar C = f(C)\). Then \(\bar C\) is necessarily a curve in \(Y\) containing the point \(y\). Let \(U\) be a Zariski open neighborhood of \(y\in Y\). Then we can find a nonzero \(g\in \Gamma (U,\mathcal{O}_Y)\) such that \(\bar C\cap U = V(g)\). Regarding \(g\) as an element of the function field \(K(Y)\), we see that \(\operatorname{\mathrm{div}}f^*(g) = \sum _{i=1}^r m_i E_i + D\) with \(C\cap f^{-1} U = D\cap f^{-1} U\). Since \(D|_{f^{-1}(U)}\) is effective and \(E_i\subseteq f^{-1}(U)\), we get that \(E_i\cdot D = E_i\cdot C = E_i\cdot H\neq 0\) for all \(i\). But then

\begin{align} E_j \cdot (\sum m_i E_i + D) & = E_j\cdot \operatorname{\mathrm{div}}f^*(g) = 0\text{ for all $j$},\\ D\cdot (\sum m_i E_i + D) & = 0. \end{align}

Now, apply Lemma 3.33 to the free vector space on the \(E_i\) and \(D\), with \(z = \sum m_i E_i + D\). We get the result.