In this section we prove some lemmas on representing Cartier divisors by suitable effective Cartier divisors on blowups. These lemmas can be found in [Section 2.4, F]. We have adapted the formulation so they also work in the non-finite type setting. It may happen that the morphism b of Lemma 115.23.11 is a composition of infinitely many blowups, but over any given quasi-compact open W \subset X one needs only finitely many blowups (and this is the result of loc. cit.).
Proof.
Note that the quasi-coherent ideal sheaf \mathcal{I} = \mathcal{I}_{D_1} + \mathcal{I}_{D_2} defines the scheme theoretic intersection D_1 \cap D_2 \subset X. Since Z is a union of connected components of D_1 \cap D_2 we see that for every z \in Z the kernel of \mathcal{O}_{X, z} \to \mathcal{O}_{Z, z} is equal to \mathcal{I}_ z. Let b : X' \to X be the blowup of X in Z. (So Zariski locally around Z it is the blowup of X in \mathcal{I}.) Denote E = b^{-1}(Z) the corresponding effective Cartier divisor, see Divisors, Lemma 31.32.4. Since Z \subset D_1 we have E \subset f^{-1}(D_1) and hence D_1 = D_1' + E for some effective Cartier divisor D'_1 \subset X', see Divisors, Lemma 31.13.8. Similarly D_2 = D_2' + E. This takes care of assertions (1) – (5).
Note that if W' is as in (7) (a) or (7) (b), then the image W of W' is contained in D_1 \cap D_2. If W is not contained in Z, then b is an isomorphism at the generic point of W and we see that \dim _\delta (W) = \dim _\delta (W') = n - 1 which contradicts the assumption that \dim _\delta (D_1 \cap D_2 \setminus Z) \leq n - 2. Hence W \subset Z. This means that to prove (6) and (7) we may work locally around Z on X.
Thus we may assume that X = \mathop{\mathrm{Spec}}(A) with A a Noetherian domain, and D_1 = \mathop{\mathrm{Spec}}(A/a), D_2 = \mathop{\mathrm{Spec}}(A/b) and Z = D_1 \cap D_2. Set I = (a, b). Since A is a domain and a, b \not= 0 we can cover the blowup by two patches, namely U = \mathop{\mathrm{Spec}}(A[s]/(as - b)) and V = \mathop{\mathrm{Spec}}(A[t]/(bt -a)). These patches are glued using the isomorphism A[s, s^{-1}]/(as - b) \cong A[t, t^{-1}]/(bt - a) which maps s to t^{-1}. The effective Cartier divisor E is described by \mathop{\mathrm{Spec}}(A[s]/(as - b, a)) \subset U and \mathop{\mathrm{Spec}}(A[t]/(bt - a, b)) \subset V. The closed subscheme D'_1 corresponds to \mathop{\mathrm{Spec}}(A[t]/(bt - a, t)) \subset U. The closed subscheme D'_2 corresponds to \mathop{\mathrm{Spec}}(A[s]/(as -b, s)) \subset V. Since “ts = 1” we see that D'_1 \cap D'_2 = \emptyset .
Suppose we have a prime \mathfrak q \subset A[s]/(as - b) of height one with s, a \in \mathfrak q. Let \mathfrak p \subset A be the corresponding prime of A. Observe that a, b \in \mathfrak p. By the dimension formula we see that \dim (A_{\mathfrak p}) = 1 as well. The final assertion to be shown is that
\text{ord}_{A_{\mathfrak p}}(a) \text{ord}_{A_{\mathfrak p}}(b) > \text{ord}_{B_{\mathfrak q}}(a) \text{ord}_{B_{\mathfrak q}}(s)
where B = A[s]/(as - b). By Algebra, Lemma 10.124.1 we have \text{ord}_{A_{\mathfrak p}}(x) \geq \text{ord}_{B_{\mathfrak q}}(x) for x = a, b. Since \text{ord}_{B_{\mathfrak q}}(s) > 0 we win by additivity of the \text{ord} function and the fact that as = b.
\square
Proof.
Let us first prove this in the quasi-compact case, since it is perhaps the most interesting case. In this case we produce inductively a sequence of blowups
X = X_0 \xleftarrow {b_0} X_1 \xleftarrow {b_1} X_2 \leftarrow \ldots
and finite sets of effective Cartier divisors \{ D_{n, i}\} _{i \in I_ n}. At each stage these will have the property that any triple intersection D_{n, i} \cap D_{n, j} \cap D_{n, k} is empty. Moreover, for each n \geq 0 we will have I_{n + 1} = I_ n \amalg P(I_ n) where P(I_ n) denotes the set of pairs of elements of I_ n. Finally, we will have
b_ n^{-1}(D_{n, i}) = D_{n + 1, i} + \sum \nolimits _{i' \in I_ n, i' \not= i} D_{n + 1, \{ i, i'\} }
We conclude that for each n \geq 0 we have (b_0 \circ \ldots \circ b_ n)^{-1}(D_ i) is a nonnegative integer combination of the divisors D_{n + 1, j}, j \in I_{n + 1}.
To start the induction we set X_0 = X and I_0 = I and D_{0, i} = D_ i.
Given (X_ n, \{ D_{n, i}\} _{i \in I_ n}) let X_{n + 1} be the blowup of X_ n in the closed subscheme Z_ n = \bigcup _{\{ i, i'\} \in P(I_ n)} D_{n, i} \cap D_{n, i'}. Note that the closed subschemes D_{n, i} \cap D_{n, i'} are pairwise disjoint by our assumption on triple intersections. In other words we may write Z_ n = \coprod _{\{ i, i'\} \in P(I_ n)} D_{n, i} \cap D_{n, i'}. Moreover, in a Zariski neighbourhood of D_{n, i} \cap D_{n, i'} the morphism b_ n is equal to the blowup of the scheme X_ n in the closed subscheme D_{n, i} \cap D_{n, i'}, and the results of Lemma 115.23.8 apply. Hence setting D_{n + 1, \{ i, i'\} } = b_ n^{-1}(D_ i \cap D_{i'}) we get an effective Cartier divisor. The Cartier divisors D_{n + 1, \{ i, i'\} } are pairwise disjoint. Clearly we have b_ n^{-1}(D_{n, i}) \supset D_{n + 1, \{ i, i'\} } for every i' \in I_ n, i' \not= i. Hence, applying Divisors, Lemma 31.13.8 we see that indeed b^{-1}(D_{n, i}) = D_{n + 1, i} + \sum \nolimits _{i' \in I_ n, i' \not= i} D_{n + 1, \{ i, i'\} } for some effective Cartier divisor D_{n + 1, i} on X_{n + 1}. In a neighbourhood of D_{n + 1, \{ i, i'\} } these divisors D_{n + 1, i} play the role of the primed divisors of Lemma 115.23.8. In particular we conclude that D_{n + 1, i} \cap D_{n + 1, i'} = \emptyset if i \not= i', i, i' \in I_ n by part (6) of Lemma 115.23.8. This already implies that triple intersections of the divisors D_{n + 1, i} are zero.
OK, and at this point we can use the quasi-compactness of X to conclude that the invariant
115.23.11.1
\begin{equation} \label{obsolete-equation-invariant} \epsilon (X, \{ D_ i\} _{i \in I}) = \max \{ \epsilon _ Z(D_ i, D_{i'}) \mid Z \subset X, \dim _\delta (Z) = d - 1, \{ i, i'\} \in P(I)\} \end{equation}
is finite, since after all each D_ i has at most finitely many irreducible components. We claim that for some n the invariant \epsilon (X_ n, \{ D_{n, i}\} _{i \in I_ n}) is zero. Namely, if not then by Lemma 115.23.8 we have a strictly decreasing sequence
\epsilon (X, \{ D_ i\} _{i \in I}) = \epsilon (X_0, \{ D_{0, i}\} _{i \in I_0}) > \epsilon (X_1, \{ D_{1, i}\} _{i \in I_1}) > \ldots
of positive integers which is a contradiction. Take n with invariant \epsilon (X_ n, \{ D_{n, i}\} _{i \in I_ n}) equal to zero. This means that there is no integral closed subscheme Z \subset X_ n and no pair of indices i, i' \in I_ n such that \epsilon _ Z(D_{n, i}, D_{n, i'}) > 0. In other words, \dim _\delta (D_{n, i}, D_{n, i'}) \leq d - 2 for all pairs \{ i, i'\} \in P(I_ n) as desired.
Next, we come to the general case where we no longer assume that the scheme X is quasi-compact. The problem with the idea from the first part of the proof is that we may get and infinite sequence of blowups with centers dominating a fixed point of X. In order to avoid this we cut out suitable closed subsets of codimension \geq 3 at each stage. Namely, we will construct by induction a sequence of morphisms having the following shape
\xymatrix{ X = X_0 \\ U_0 \ar[u]^{j_0} & X_1 \ar[l]_{b_0} \\ & U_1 \ar[u]^{j_1} & X_2 \ar[l]_{b_1} \\ & & U_2 \ar[u]^{j_2} & X_3 \ar[l]_{b_2} }
Each of the morphisms j_ n : U_ n \to X_ n will be an open immersion. Each of the morphisms b_ n : X_{n + 1} \to U_ n will be a proper birational morphism of integral schemes. As in the quasi-compact case we will have effective Cartier divisors \{ D_{n, i}\} _{i \in I_ n} on X_ n. At each stage these will have the property that any triple intersection D_{n, i} \cap D_{n, j} \cap D_{n, k} is empty. Moreover, for each n \geq 0 we will have I_{n + 1} = I_ n \amalg P(I_ n) where P(I_ n) denotes the set of pairs of elements of I_ n. Finally, we will arrange it so that
b_ n^{-1}(D_{n, i}|_{U_ n}) = D_{n + 1, i} + \sum \nolimits _{i' \in I_ n, i' \not= i} D_{n + 1, \{ i, i'\} }
We start the induction by setting X_0 = X, I_0 = I and D_{0, i} = D_ i.
Given (X_ n, \{ D_{n, i}\} ) we construct the open subscheme U_ n as follows. For each pair \{ i, i'\} \in P(I_ n) consider the closed subscheme D_{n, i} \cap D_{n, i'}. This has “good” irreducible components which have \delta -dimension d - 2 and “bad” irreducible components which have \delta -dimension d - 1. Let us set
\text{Bad}(i, i') = \bigcup \nolimits _{W \subset D_{n, i} \cap D_{n, i'} \text{ irred.\ comp. with }\dim _\delta (W) = d - 1} W
and similarly
\text{Good}(i, i') = \bigcup \nolimits _{W \subset D_{n, i} \cap D_{n, i'} \text{ irred.\ comp. with }\dim _\delta (W) = d - 2} W.
Then D_{n, i} \cap D_{n, i'} = \text{Bad}(i, i') \cup \text{Good}(i, i') and moreover we have \dim _\delta (\text{Bad}(i, i') \cap \text{Good}(i, i')) \leq d - 3. Here is our choice of U_ n:
U_ n = X_ n \setminus \bigcup \nolimits _{\{ i, i'\} \in P(I_ n)} \text{Bad}(i, i') \cap \text{Good}(i, i').
By our condition on triple intersections of the divisors D_{n, i} we see that the union is actually a disjoint union. Moreover, we see that (as a scheme)
D_{n, i}|_{U_ n} \cap D_{n, i'}|_{U_ n} = Z_{n, i, i'} \amalg G_{n, i, i'}
where Z_{n, i, i'} is \delta -equidimensional of dimension d - 1 and G_{n, i, i'} is \delta -equidimensional of dimension d - 2. (So topologically Z_{n, i, i'} is the union of the bad components but throw out intersections with good components.) Finally we set
Z_ n = \bigcup \nolimits _{\{ i, i'\} \in P(I_ n)} Z_{n, i, i'} = \coprod \nolimits _{\{ i, i'\} \in P(I_ n)} Z_{n, i, i'},
and we let b_ n : X_{n + 1} \to X_ n be the blowup in Z_ n. Note that Lemma 115.23.8 applies to the morphism b_ n : X_{n + 1} \to X_ n locally around each of the loci D_{n, i}|_{U_ n} \cap D_{n, i'}|_{U_ n}. Hence, exactly as in the first part of the proof we obtain effective Cartier divisors D_{n + 1, \{ i, i'\} } for \{ i, i'\} \in P(I_ n) and effective Cartier divisors D_{n + 1, i} for i \in I_ n such that b_ n^{-1}(D_{n, i}|_{U_ n}) = D_{n + 1, i} + \sum \nolimits _{i' \in I_ n, i' \not= i} D_{n + 1, \{ i, i'\} }. For each n denote \pi _ n : X_ n \to X the morphism obtained as the composition j_0 \circ \ldots \circ j_{n - 1} \circ b_{n - 1}.
Claim: given any quasi-compact open V \subset X for all sufficiently large n the maps
\pi _ n^{-1}(V) \leftarrow \pi _{n + 1}^{-1}(V) \leftarrow \ldots
are all isomorphisms. Namely, if the map \pi _ n^{-1}(V) \leftarrow \pi _{n + 1}^{-1}(V) is not an isomorphism, then Z_{n, i, i'} \cap \pi _ n^{-1}(V) \not= \emptyset for some \{ i, i'\} \in P(I_ n). Hence there exists an irreducible component W \subset D_{n, i} \cap D_{n, i'} with \dim _\delta (W) = d - 1. In particular we see that \epsilon _ W(D_{n, i}, D_{n, i'}) > 0. Applying Lemma 115.23.8 repeatedly we see that
\epsilon _ W(D_{n, i}, D_{n, i'}) < \epsilon (V, \{ D_ i|_ V\} ) - n
with \epsilon (V, \{ D_ i|_ V\} ) as in (115.23.11.1). Since V is quasi-compact, we have \epsilon (V, \{ D_ i|_ V\} ) < \infty and taking n > \epsilon (V, \{ D_ i|_ V\} ) we see the result.
Note that by construction the difference X_ n \setminus U_ n has \dim _\delta (X_ n \setminus U_ n) \leq d - 3. Let T_ n = \pi _ n(X_ n \setminus U_ n) be its image in X. Traversing in the diagram of maps above using each b_ n is closed it follows that T_0 \cup \ldots \cup T_ n is a closed subset of X for each n. Any t \in T_ n satisfies \delta (t) \leq d - 3 by construction. Hence \overline{T_ n} \subset X is a closed subset with \dim _\delta (T_ n) \leq d - 3. By the claim above we see that for any quasi-compact open V \subset X we have T_ n \cap V \not= \emptyset for at most finitely many n. Hence \{ \overline{T_ n}\} _{n \geq 0} is a locally finite collection of closed subsets, and we may set U = X \setminus \bigcup \overline{T_ n}. This will be U as in the lemma.
Note that U_ n \cap \pi _ n^{-1}(U) = \pi _ n^{-1}(U) by construction of U. Hence all the morphisms
b_ n : \pi _{n + 1}^{-1}(U) \longrightarrow \pi _ n^{-1}(U)
are proper. Moreover, by the claim they eventually become isomorphisms over each quasi-compact open of X. Hence we can define
U' = \mathop{\mathrm{lim}}\nolimits _ n \pi _ n^{-1}(U).
The induced morphism b : U' \to U is proper since this is local on U, and over each compact open the limit stabilizes. Similarly we set J = \bigcup _{n \geq 0} I_ n using the inclusions I_ n \to I_{n + 1} from the construction. For j \in J choose an n_0 such that j corresponds to i \in I_{n_0} and define D'_ j = \mathop{\mathrm{lim}}\nolimits _{n \geq n_0} D_{n, i}. Again this makes sense as locally over X the morphisms stabilize. The other claims of the lemma are verified as in the case of a quasi-compact X.
\square
Comments (0)