Definition 10.125.1. Suppose that R \to S is of finite type, and let \mathfrak q \subset S be a prime lying over a prime \mathfrak p of R. We define the relative dimension of S/R at \mathfrak q, denoted \dim _{\mathfrak q}(S/R), to be the dimension of \mathop{\mathrm{Spec}}(S \otimes _ R \kappa (\mathfrak p)) at the point corresponding to \mathfrak q. We let \dim (S/R) be the supremum of \dim _{\mathfrak q}(S/R) over all \mathfrak q. This is called the relative dimension of S/R.
10.125 Dimension of fibres
We study the behaviour of dimensions of fibres, using Zariski's main theorem. Recall that we defined the dimension \dim _ x(X) of a topological space X at a point x in Topology, Definition 5.10.1.
In particular, R \to S is quasi-finite at \mathfrak q if and only if \dim _{\mathfrak q}(S/R) = 0. The following lemma is more or less a reformulation of Zariski's Main Theorem.
Lemma 10.125.2. Let R \to S be a finite type ring map. Let \mathfrak q \subset S be a prime. Suppose that \dim _{\mathfrak q}(S/R) = n. There exists a g \in S, g \not\in \mathfrak q such that S_ g is quasi-finite over a polynomial algebra R[t_1, \ldots , t_ n].
Proof. The ring \overline{S} = S \otimes _ R \kappa (\mathfrak p) is of finite type over \kappa (\mathfrak p). Let \overline{\mathfrak q} be the prime of \overline{S} corresponding to \mathfrak q. By definition of the dimension of a topological space at a point there exists an open U \subset \mathop{\mathrm{Spec}}(\overline{S}) with \overline{q} \in U and \dim (U) = n. Since the topology on \mathop{\mathrm{Spec}}(\overline{S}) is induced from the topology on \mathop{\mathrm{Spec}}(S) (see Remark 10.18.5), we can find a g \in S, g \not\in \mathfrak q with image \overline{g} \in \overline{S} such that D(\overline{g}) \subset U. Thus after replacing S by S_ g we see that \dim (\overline{S}) = n.
Next, choose generators x_1, \ldots , x_ N for S as an R-algebra. By Lemma 10.115.4 there exist elements y_1, \ldots , y_ n in the \mathbf{Z}-subalgebra of S generated by x_1, \ldots , x_ N such that the map R[t_1, \ldots , t_ n] \to S, t_ i \mapsto y_ i has the property that \kappa (\mathfrak p)[t_1\ldots , t_ n] \to \overline{S} is finite. In particular, S is quasi-finite over R[t_1, \ldots , t_ n] at \mathfrak q. Hence, by Lemma 10.123.13 we may replace S by S_ g for some g\in S, g \not\in \mathfrak q such that R[t_1, \ldots , t_ n] \to S is quasi-finite. \square
Lemma 10.125.3. Let R \to S be a ring map. Let \mathfrak q \subset S be a prime lying over the prime \mathfrak p of R. Assume
R \to S is of finite type,
\dim _{\mathfrak q}(S/R) = n, and
\text{trdeg}_{\kappa (\mathfrak p)}\kappa (\mathfrak q) = r.
Then there exist f \in R, f \not\in \mathfrak p, g \in S, g \not\in \mathfrak q and a quasi-finite ring map
such that \varphi ^{-1}(\mathfrak qS_ g) = (\mathfrak p, x_{r + 1}, \ldots , x_ n)R_ f[x_{r + 1}, \ldots , x_ n]
Proof. After replacing S by a principal localization we may assume there exists a quasi-finite ring map \varphi : R[t_1, \ldots , t_ n] \to S, see Lemma 10.125.2. Set \mathfrak q' = \varphi ^{-1}(\mathfrak q). Let \overline{\mathfrak q}' \subset \kappa (\mathfrak p)[t_1, \ldots , t_ n] be the prime corresponding to \mathfrak q'. By Lemma 10.115.6 there exists a finite ring map \kappa (\mathfrak p)[x_1, \ldots , x_ n] \to \kappa (\mathfrak p)[t_1, \ldots , t_ n] such that the inverse image of \overline{\mathfrak q}' is (x_{r + 1}, \ldots , x_ n). Let \overline{h}_ i \in \kappa (\mathfrak p)[t_1, \ldots , t_ n] be the image of x_ i. We can find an element f \in R, f \not\in \mathfrak p and h_ i \in R_ f[t_1, \ldots , t_ n] which map to \overline{h}_ i in \kappa (\mathfrak p)[t_1, \ldots , t_ n]. Then the ring map
becomes finite after tensoring with \kappa (\mathfrak p). In particular, R_ f[t_1, \ldots , t_ n] is quasi-finite over R_ f[x_1, \ldots , x_ n] at the prime \mathfrak q'R_ f[t_1, \ldots , t_ n]. Hence, by Lemma 10.123.13 there exists a g \in R_ f[t_1, \ldots , t_ n], g \not\in \mathfrak q'R_ f[t_1, \ldots , t_ n] such that R_ f[x_1, \ldots , x_ n] \to R_ f[t_1, \ldots , t_ n, 1/g] is quasi-finite. Thus we see that the composition
is quasi-finite and we win. \square
Lemma 10.125.4. Let R \to S be a finite type ring map. Let \mathfrak q \subset S be a prime lying over \mathfrak p \subset R. If R \to S is quasi-finite at \mathfrak q, then \dim (S_{\mathfrak q}) \leq \dim (R_{\mathfrak p}).
Proof. If R_{\mathfrak p} is Noetherian (and hence S_{\mathfrak q} Noetherian since it is essentially of finite type over R_{\mathfrak p}) then this follows immediately from Lemma 10.112.6 and the definitions. In the general case, let S' be the integral closure of R_\mathfrak p in S_\mathfrak p. By Zariski's Main Theorem 10.123.12 we have S_{\mathfrak q} = S'_{\mathfrak q'} for some \mathfrak q' \subset S' lying over \mathfrak q. By Lemma 10.112.3 we have \dim (S') \leq \dim (R_\mathfrak p) and hence a fortiori \dim (S_\mathfrak q) = \dim (S'_{\mathfrak q'}) \leq \dim (R_\mathfrak p). \square
Lemma 10.125.5.slogan Let k be a field. Let S be a finite type k-algebra. Suppose there is a quasi-finite k-algebra map k[t_1, \ldots , t_ n] \subset S. Then \dim (S) \leq n.
Proof. By Lemma 10.114.1 the dimension of any local ring of k[t_1, \ldots , t_ n] is at most n. Thus the result follows from Lemma 10.125.4. \square
Lemma 10.125.6. Let R \to S be a finite type ring map. Let \mathfrak q \subset S be a prime. Suppose that \dim _{\mathfrak q}(S/R) = n. There exists an open neighbourhood V of \mathfrak q in \mathop{\mathrm{Spec}}(S) such that \dim _{\mathfrak q'}(S/R) \leq n for all \mathfrak q' \in V.
Proof. By Lemma 10.125.2 we see that we may assume that S is quasi-finite over a polynomial algebra R[t_1, \ldots , t_ n]. Considering the fibres, we reduce to Lemma 10.125.5. \square
In other words, the lemma says that the set of points where the fibre has dimension \leq n is open in \mathop{\mathrm{Spec}}(S). The next lemma says that formation of this open commutes with base change. If the ring map is of finite presentation then this set is quasi-compact open (see below).
Lemma 10.125.7. Let R \to S be a finite type ring map. Let R \to R' be any ring map. Set S' = R' \otimes _ R S and denote f : \mathop{\mathrm{Spec}}(S') \to \mathop{\mathrm{Spec}}(S) the associated map on spectra. Let n \geq 0. The inverse image f^{-1}(\{ \mathfrak q \in \mathop{\mathrm{Spec}}(S) \mid \dim _{\mathfrak q}(S/R) \leq n\} ) is equal to \{ \mathfrak q' \in \mathop{\mathrm{Spec}}(S') \mid \dim _{\mathfrak q'}(S'/R') \leq n\} .
Proof. The condition is formulated in terms of dimensions of fibre rings which are of finite type over a field. Combined with Lemma 10.116.6 this yields the lemma. \square
Lemma 10.125.8. Let R \to S be a ring homomorphism of finite presentation. Let n \geq 0. The set
is a quasi-compact open subset of \mathop{\mathrm{Spec}}(S).
Proof. It is open by Lemma 10.125.6. Let S = R[x_1, \ldots , x_ n]/(f_1, \ldots , f_ m) be a presentation of S. Let R_0 be the \mathbf{Z}-subalgebra of R generated by the coefficients of the polynomials f_ i. Let S_0 = R_0[x_1, \ldots , x_ n]/(f_1, \ldots , f_ m). Then S = R \otimes _{R_0} S_0. By Lemma 10.125.7 V_ n is the inverse image of an open V_{0, n} under the quasi-compact continuous map \mathop{\mathrm{Spec}}(S) \to \mathop{\mathrm{Spec}}(S_0). Since S_0 is Noetherian we see that V_{0, n} is quasi-compact. \square
Lemma 10.125.9. Let R be a valuation ring with residue field k and field of fractions K. Let S be a domain containing R such that S is of finite type over R. If S \otimes _ R k is not the zero ring then
In fact, \mathop{\mathrm{Spec}}(S \otimes _ R k) is equidimensional.
Proof. It suffices to show that \dim _{\mathfrak q}(S/k) is equal to \dim (S \otimes _ R K) for every prime \mathfrak q of S containing \mathfrak m_ RS. Pick such a prime. By Lemma 10.125.6 the inequality \dim _{\mathfrak q}(S/k) \geq \dim (S \otimes _ R K) holds. Set n = \dim _{\mathfrak q}(S/k). By Lemma 10.125.2 after replacing S by S_ g for some g \in S, g \not\in \mathfrak q there exists a quasi-finite ring map R[t_1, \ldots , t_ n] \to S. If \dim (S \otimes _ R K) < n, then K[t_1, \ldots , t_ n] \to S \otimes _ R K has a nonzero kernel. Say f = \sum a_ I t_1^{i_1}\ldots t_ n^{i_ n}. After dividing f by a nonzero coefficient of f with minimal valuation, we may assume f\in R[t_1, \ldots , t_ n] and some a_ I does not map to zero in k. Hence the ring map k[t_1, \ldots , t_ n] \to S \otimes _ R k has a nonzero kernel which implies that \dim (S \otimes _ R k) < n. Contradiction. \square
Comments (0)