The Stacks project

Proof. In this paragraph we reduce to the case where $R \to S$ is of finite presentation. Namely, write $S = R[A, B, x_1, \ldots , x_ n]/J$ for some ideal $J \subset R[x_1, \ldots , x_ n]$ where $A$ and $B$ map to $a$ and $b$ in $S$. Then $J$ is the union of its finitely generated ideals $J_\lambda \subset J$. Set $S_\lambda = R[A, B, x_1, \ldots , x_ n]/J_\lambda $ and denote $a_\lambda , b_\lambda \in S_\lambda $ the images of $A$ and $B$. Then for some $\lambda $ the fibres of

have dimension $\leq d$, see Limits, Lemma 32.18.1. Fix such a $\lambda $. If we can find $n_0$ which works for $R \to S_\lambda $, $a_\lambda $, $b_\lambda $, then $n_0$ works for $R \to S$. Namely, the fibres of $f_{a_\lambda ^ n + b_\lambda } : \mathop{\mathrm{Spec}}(S_\lambda ) \to \mathbf{A}^1_ R$ contain the fibres of $f_{a^ n + b} : \mathop{\mathrm{Spec}}(S) \to \mathbf{A}^1_ R$. This reduces us to the case discussed in the next paragraph.

Assume $R \to S$ is of finite presentation. In this paragraph we reduce to the case where $R$ is of finite type over $\mathbf{Z}$. By Algebra, Lemma 10.127.18 we can find a directed set $\Lambda $ and a system of ring maps $R_\lambda \to S_\lambda $ over $\Lambda $ whose colimit is $R \to S$ such that $S_\mu = S_\lambda \otimes _{R_\lambda } R_\mu $ for $\mu \geq \lambda $ and such that each $R_\lambda $ and $S_\lambda $ is of finite type over $\mathbf{Z}$. Choose $\lambda _0 \in \Lambda $ and elements $a_{\lambda _0}, b_{\lambda _0} \in S_{\lambda _0}$ mapping to $a, b \in S$. For $\lambda \geq \lambda _0$ denote $a_\lambda , b_\lambda \in S_\lambda $ the image of $a_{\lambda _0}, b_{\lambda _0}$. Then for $\lambda \geq \lambda _0$ large enough the fibres of

have dimension $\leq d$, see Limits, Lemma 32.18.4. Fix such a $\lambda $. If we can find $n_0$ which works for $R_\lambda \to S_\lambda $, $a_\lambda $, $b_\lambda $, then $n_0$ works for $R \to S$. Namely, any fibre of $f_{a^ n + b} : \mathop{\mathrm{Spec}}(S) \to \mathbf{A}^1_ R$ has the same dimension as a fibre of $f_{a_\lambda ^ n + b_\lambda } : \mathop{\mathrm{Spec}}(S_\lambda ) \to \mathbf{A}^1_{R_\lambda }$ by Morphisms, Lemma 29.28.3. This reduces us the the case discussed in the next paragraph.

Assume $R$ and $S$ are of finite type over $\mathbf{Z}$. In particular the dimension of $R$ is finite, and we may use induction on $\dim (R)$. Thus we may assume the result holds for all situations with $R' \to S'$, $a$, $b$ as in the lemma with $R'$ and $S'$ of finite type over $\mathbf{Z}$ but with $\dim (R') < \dim (R)$.

Since the statement is about the topology of the spectrum of $S$ we may assume $S$ is reduced. Let $S^\nu $ be the normalization of $S$. Then $S \subset S^\nu $ is a finite extension as $S$ is excellent, see Algebra, Proposition 10.162.16 and Morphisms, Lemma 29.54.10. Thus $\mathop{\mathrm{Spec}}(S^\nu ) \to \mathop{\mathrm{Spec}}(S)$ is surjective and finite (Algebra, Lemma 10.36.17). It follows that if the result holds for $R \to S^\nu $ and the images of $a$, $b$ in $S^\nu $, then the result holds for $R \to S$, $a$, $b$. (Small detail omitted.) This reduces us to the case discussed in the next paragraph.

Assume $R$ and $S$ are of finite type over $\mathbf{Z}$ and $S$ normal. Then $S = S_1 \times \ldots \times S_ r$ for some normal domains $S_ i$. If the result holds for each $R \to S_ i$ and the images of $a$, $b$ in $S_ i$, then the result holds for $R \to S$, $a$, $b$. (Small detail omitted.) This reduces us to the case discussed in the next paragraph.

Assume $R$ and $S$ are of finite type over $\mathbf{Z}$ and $S$ a normal domain. We may replace $R$ by the image of $R$ in $S$ (this does not increase the dimension of $R$). This reduces us to the case discussed in the next paragraph.

Assume $R \subset S$ are of finite type over $\mathbf{Z}$ and $S$ a normal domain. Consider the morphism

The assumption tells us that $f_ a$ has fibres of dimension $\leq d$. Hence the fibres of $f : \mathop{\mathrm{Spec}}(S) \to \mathop{\mathrm{Spec}}(R)$ have dimension $\leq d + 1$ (Morphisms, Lemma 29.28.2). Consider the morphism of integral schemes

corresponding to the $R$-algebra map $R[x, y] \to S$ sending $x$ to $a$ and $y$ to $b$. There are two cases to consider

1. $\phi $ is dominant, and

2. $\phi $ is not dominant.

We claim that in both cases there exists an integer $n_0$ and a nonempty open $V \subset \mathop{\mathrm{Spec}}(R)$ such that for $n \geq n_0$ the fibres of $f_{a^ n + b}$ at points $q \in \mathbf{A}^1_ V$ have dimension $\leq d$.

Proof of the claim in case (1). We have $f_{a^ n + b} = \pi _ n \circ \phi $ where

is the flat morphism corresponding to the $R$-algebra map $R[x] \to R[x, y]$ sending $x$ to $x^ n + y$. Since $\phi $ is dominant there is a dense open $U \subset \mathop{\mathrm{Spec}}(S)$ such that $\phi |_ U : U \to \mathbf{A}^2_ R$ is flat (this follows for example from generic flatness, see Morphisms, Proposition 29.27.1). Then the composition

is flat as well. Hence the fibres of this morphism have at least codimension $1$ in the fibres of $f|_ U : U \to \mathop{\mathrm{Spec}}(R)$ by Morphisms, Lemma 29.28.2. In other words, the fibres of $f_{a^ n + b}|_ U$ have dimension $\leq d$. On the other hand, since $U$ is dense in $\mathop{\mathrm{Spec}}(S)$, we can find a nonempty open $V \subset \mathop{\mathrm{Spec}}(R)$ such that $U \cap f^{-1}(p) \subset f^{-1}(p)$ is dense for all $p \in V$ (see for example Lemma 37.24.3). Thus $\dim (f^{-1}(p) \setminus U \cap f^{-1}(p)) \leq d$ and we conclude that our claim is true (as any fibres of $f_{a^ n + b} : \mathop{\mathrm{Spec}}(S) \to \mathbf{A}^1_ R$ is contained in a fibre of $f$).

Case (2). In this case we can find a nonzero $g = \sum c_{ij} x^ i y^ j$ in $R[x, y]$ such that $\mathop{\mathrm{Im}}(\phi ) \subset V(g)$. In fact, we may assume $g$ is irreducible over $\text{Frac}(R)$. If $g \in R[x]$, say with leading coefficient $c$, then over $V = D(c) \subset \mathop{\mathrm{Spec}}(R)$ the fibres of $f$ already have dimension $\leq d$ (because the image of $f_ a$ is contained in $V(g) \subset \mathbf{A}^1_ R$ which has finite fibres over $V$). Hence we may assume $g$ is not contained in $R[x]$. Let $s \geq 1$ be the degree of $g$ as a polynomial in $y$ and let $t$ be the degree of $\sum c_{is} x^ i$ as a polynomial in $x$. Then $c_{ts}$ is nonzero and

provided that $n$ is bigger than the degree of $g$ as a polynomial in $x$ (small detail omitted). For such $n$ the polynomial $g(x, -x^ n)$ is a nonzero polynomial in $x$ and maps to a nonzero polynomial in $\kappa (\mathfrak p)[x]$ for all $\mathfrak p \subset R$, $c_{st} \not\in \mathfrak p$. We conclude that our claim is true for $V$ equal to the principal open $D(c_{ts})$ of $\mathop{\mathrm{Spec}}(R)$.

OK, and now we can use induction on $\dim (R)$. Namely, let $I \subset R$ be an ideal such that $V(I) = \mathop{\mathrm{Spec}}(R) \setminus V$. Observe that $\dim (R/I) < \dim (R)$ as $R$ is a domain. Let $n'_0$ be the integer we have by induction on $\dim (R)$ for $R/I \to S/IS$ and the images of $a$ and $b$ in $S/IS$. Then $\max (n_0, n'_0)$ works. $\square$