## 37.18 Flat modules and relative assassins

In this section we will prove that the support of a flat module is (in some sense) equidimensional over the base in geometric situations. For the Noetherian case we refer the reader to [IV Proposition 12.1.1.5, EGA]. First, we prove two helper lemmas.

Lemma 37.18.1. Let $A$ be a valuation ring. Let $A \to B$ is a local homomorphism of local rings which is essentially of finite type. Let $u : N \to M$ be a map of finite $B$-modules. Assume $M$ is flat over $A$ and $\overline{u} : N/\mathfrak m_ A N \to M/\mathfrak m_ A M$ is injective. Then $u$ is injective and $M/u(N)$ is flat over $A$.

Proof. We will deduce this lemma from Algebra, Lemma 10.128.4 (please note that we exchanged the roles of $M$ and $N$). To do the reduction we will use More on Algebra, Lemma 15.25.7 to reduce to the finitely presented case.

By assumption we can write $B$ as a quotient of the localization of a polynomial algebra $P = A[x_1, \ldots , x_ n]$ at a prime ideal $\mathfrak q$. Then we can think of $u : N \to M$ as a map of finite $P_\mathfrak q$-modules. Hence we may and do assume that $B$ is essentially of finite presentation over $A$.

Next, the $B$-module $N$ is finite but perhaps not of finite presentation. Write $N = \mathop{\mathrm{colim}}\nolimits N_\lambda$ as a filtered colimit of finitely presented $B$-modules with surjective transition maps. For example choose a presentation $0 \to K \to B^{\oplus r} \to N \to 0$, write $K$ as the union of its finite submodules $K_\lambda$, and set $N_\lambda = \mathop{\mathrm{Coker}}(K_\lambda \to B^{\oplus r})$. The module $N/\mathfrak m_ A N$ is of finite presentation over the Noetherian ring $B/\mathfrak m_ A B$. Hence for $\lambda$ large enough we have $N_\lambda /\mathfrak m_ A N_\lambda = N/\mathfrak m_ A N$. Now, if we can show the lemma for the composition $u_\lambda : N_\lambda \to M$, then we conclude that $N_\lambda = N$ and the result holds for $u$. Hence we may and do assume $N$ is of finite presentation over $B$.

By More on Algebra, Lemma 15.25.7 the module $M$ is of finite presentation over $B$. Thus all the assumptions of Algebra, Lemma 10.128.4 hold and we conclude. $\square$

Lemma 37.18.2. Let $f : X \to S$ be a morphism of schemes. Let $y \in X$ be a point with image $t \in S$. Denote $Y \subset X$ the closure of $\{ y\}$ viewed as an integral closed subscheme of $X$. Let $s \in S$ and let $x \in Y_ s$ be a generic point of an irreducible component of $Y_ s$. There exists a cartesian diagram

$\xymatrix{ X' \ar[r]_{g'} \ar[d]_{f'} & X \ar[d]^ f \\ S' \ar[r]^ g & S }$

with the following properties:

1. $S'$ is the spectrum of a valuation ring with generic point $t'$ and closed point $s'$,

2. $g(t') = t$ and $g(s') = s$,

3. there exists a point $y' \in X'_{t'}$ which is a generic point of an irreducible component of $(S' \times _ S Y)_{t'} = Y_ t \times _ t t'$ and satisfies $g'(y') = y$,

4. denoting $Y' \subset X'$ the closure of $\{ y'\}$ viewed as an integral closed subscheme of $X'$ there exists a point $x' \in Y'_{s'}$ which is a generic point of an irreducible component of $Y'_{s'}$ with $g'(x') = x$.

Proof. We choose a valuation ring $R$, we set $S' = \mathop{\mathrm{Spec}}(R)$ with generic point $t'$ and closed point $s'$, and we choose a morphism $h : S' \to X$ with $h(t') = y$ and $h(s') = x$. See Schemes, Lemma 26.20.4. Set $g = f \circ h$ so that $g(t') = t$ and $g(s') = s$. Consider the base change

$\xymatrix{ X' \ar[r]_{g'} \ar[d] & X \ar[d] \\ S' \ar@/^1em/[u]^\sigma \ar[r]^-g & S }$

We obtain a section $\sigma$ of the base change such that $h = g' \circ \sigma$.

Of course $\sigma$ factors through the base change $S' \times _ S Y$ of $Y$ as $h$ factors through $Y$. Let $y' \in X'_{t'} \subset X'$ be the generic point of an irreducible component of the fibre

$(S' \times _ S Y)_{t'} = Y_ t \times _ t t'$

containing the point $\sigma (t')$, i.e., such that $y' \leadsto \sigma (t')$. Since $g'(y') \in Y_ t$ and $g(y') \leadsto g(\sigma (t')) = y$ we find that $g'(y') = y$ because $y$ is the generic point of the fibre $Y_ t$. Denote $Y' \subset X'$ the closure of $\{ y'\}$ in $X'$ viewed as an integral closed subscheme. Then $\sigma$ factors through $Y'$ as $\sigma (t') \in Y'$. Choose a generic point $x' \in Y'_{s'}$ of an irreducible component of $Y'_{s'}$ which contains $\sigma (s')$, i.e., we get $x' \leadsto \sigma (s')$ and hence $g'(x') \leadsto g'(\sigma (s')) = x$. Again as $x$ is a generic point of an irreducible component of $Y_ s$ by assumption and as $g'(Y') \subset Y$ we conclude that $g'(x') = x$. $\square$

Lemma 37.18.3. Let $f : X \to S$ be a morphism of schemes which is locally of finite type. Let $\mathcal{F}$ be a quasi-coherent finite type $\mathcal{O}_ X$-module. Let $y \in \text{Ass}_{X/S}(\mathcal{F})$ with image $t \in S$. Denote $Y \subset X$ the closure of $\{ y\}$ in $X$ viewed as an integral closed subscheme. Let $s \in S$ and let $x \in Y_ s$ be a generic point of an irreducible component of $Y_ s$. If $\mathcal{F}$ is flat over $S$ at $x$, then $x \in \text{Ass}_{X/S}(\mathcal{F})$ and $\dim _ x(Y_ s) = \dim (Y_ t)$.

Proof. Choose a diagram as in Lemma 37.18.2. Set $\mathcal{F}' = (g')^*\mathcal{F}$. Divisors, Lemma 31.7.3 implies that $y' \in \text{Ass}_{X'/S'}(\mathcal{F}')$. By our choice of $y'$ we also see that $\dim (Y'_{t'}) = \dim (Y_ t)$, see for example Algebra, Lemma 10.116.7. By Algebra, Lemma 10.125.9 we see that $Y'_{s'}$ is equidimensional of dimension equal to $\dim (Y_ t)$. Since $\mathcal{F}$ is flat at $x$ over $S$ we see that $\mathcal{F}'$ is flat at $x'$ over $S'$, see Morphisms, Lemma 29.25.7.

Suppose that we can show $x' \in \text{Ass}_{X'/S}(\mathcal{F}')$. Then Divisors, Lemma 31.7.3 implies that $x \in \text{Ass}_{X/S}(\mathcal{F})$ and that the irreducible component $C'$ of $Y'_{s'}$ containing $x'$ is an irreducible component of $C \times _ s s'$ where $C \subset Y_ s$ is the irreducible component containing $x$. Whence $\dim (C) = \dim (C') = \dim (Y_ t)$ (see above) and the proof is complete. This reduces us to the case discussed in the next paragraph.

Assume $S = \mathop{\mathrm{Spec}}(A)$ where $A$ is a valuation ring and $t$ and $s$ are the generic and closed points of $S$. We will assume $x \not\in \text{Ass}_{X/S}(\mathcal{F})$ in order to get a contradiction. In other words, we assume $x \not\in \text{Ass}_{X_ s}(\mathcal{F}_ s)$ where $\mathcal{F}_ s$ is the pullback of $\mathcal{F}$ to $X_ s$. Consider the ring map

$A \longrightarrow \mathcal{O}_{X, x} = B$

and the module $N = \mathcal{F}_ x$ over $B = \mathcal{O}_{X, x}$. Then $B/\mathfrak m_ A B = \mathcal{O}_{X_ s, x}$ and $N/\mathfrak m_ A N$ is the stalk of $\mathcal{F}_ s$ at the point $x$. Denote $\mathfrak q \subset B$ the prime ideal corresponding to the point $y$, see Schemes, Lemma 26.13.2. Since $x$ is a generic point of $Y_ s$ we see that the radical of $\mathfrak q + \mathfrak m_ A B$ is $\mathfrak m_ B$. Then $\text{Ass}_{B/\mathfrak m_ A B}(N/\mathfrak m_ A N)$ is a finite set of prime ideals (Algebra, Lemma 10.63.5) which doesn't contain the maximal ideal of $B/\mathfrak m_ A B$ since $x \not\in \text{Ass}_{X/S}(\mathcal{F})$. Thus the image of of $\mathfrak q$ in $B/\mathfrak m_ A B$ is not contained in any of those prime ideals. Hence by prime avoidance (Algebra, Lemma 10.15.2) we can find an element $g \in \mathfrak q$ whose image in $B/\mathfrak m_ A B$ is a nonzerodivisor on $N/\mathfrak m_ A N$ (this uses the description of zerodivisors in Algebra, Lemma 10.63.9). Since $N = \mathcal{F}_ x$ is $A$-flat by Lemma 37.18.1 we see that

$g : N \longrightarrow N$

is injective. In particular, if $K = \text{Frac}(A)$ is the fraction field of $A$, then we see that

$g : N \otimes _ A K \longrightarrow N \otimes _ A K$

is injective. Observe that $\mathfrak q$ corresponds to a prime ideal of $B \otimes _ A K$. Denote $\mathcal{F}_ t$ the restriction of $\mathcal{F}$ to the generic fibre $X_ t$. We have $(B \otimes _ A K)_{\mathfrak q} = \mathcal{O}_{X_ t, y}$ and $(N \otimes _ A K)_\mathfrak q$ is the stalk at $y$ of $\mathcal{F}_ t$. Hence we find that $g \in \mathfrak m_ y \subset \mathcal{O}_{X_ t, y}$ is a nonzerodivisor on the stalk $(\mathcal{F}_ t)_ y$ which contradicts our assumption that $y \in \text{Ass}_{X/S}(\mathcal{F})$. $\square$

Lemma 37.18.4. Let $f : X \to S$ be a morphism of schemes which is locally of finite type. Let $\mathcal{F}$ be a quasi-coherent $\mathcal{O}_ X$-module of finite type. Let $y \in \text{Ass}_{X/S}(\mathcal{F})$. Denote $Y \subset X$ the closure of $\{ y\}$ in $X$ viewed as an integral closed subscheme. Denote $T \subset S$ the closure of $\{ f(y)\}$ viewed as an integral closed subscheme. We obtain a commutative diagram

$\xymatrix{ Y \ar[r] \ar[d] & X \ar[d] \\ T \ar[r] & S }$

where $Y \to T$ is dominant. Assume $\mathcal{F}$ is flat over $S$ at all generic points of irreducible components of fibres of $Y \to T$ (for example if $\mathcal{F}$ is flat over $S$). Then

1. if $s \in S$ and $x \in Y_ s$ is the generic point of an irreducible component of $Y_ s$, then $x \in \text{Ass}_{X/S}(\mathcal{F})$, and

2. there is an integer $d \geq 0$ such that $Y \to T$ is of relative dimension $d$, see Morphisms, Definition 29.29.1.

Proof. This follows immediately from the pointwise version Lemma 37.18.3. Note that to compute the dimension of the locally algebraic schemes $Y_ s$ it suffices to look near the generic points, see Varieties, Section 33.20. $\square$

Remark 37.18.5. Here are some cases where the material above, especially Lemma 37.18.4, allows one to conclude that a morphism $f : X \to S$ of schemes has relative dimension $d$ as defined in Morphisms, Definition 29.29.1. For example, this is true if

1. $X$ is integral with generic point $\xi$,

2. the transcendence degree of $\kappa (\xi )$ over $\kappa (f(\xi ))$ is $d$,

3. $f$ is locally of finite type, and

4. there exists a quasi-coherent $\mathcal{O}_ X$-module $\mathcal{F}$ of finite type which is flat over $S$ with $\text{Supp}(\mathcal{F}) = X$.

Another set of hypotheses that work are the following:

1. $S$ is irreducible with generic point $\eta$,

2. $X_\eta$ is dense in $X$,

3. every irreducible component of $X_\eta$ has dimension $d$,

4. $f$ is locally of finite type, and

5. there exists a quasi-coherent $\mathcal{O}_ X$-module $\mathcal{F}$ of finite type which is flat over $S$ with $\text{Supp}(\mathcal{F}) = X$.

Of course, we can relax the flatness condition on $\mathcal{F}$ and require only that $\mathcal{F}$ is flat over $S$ in codimension $0$, i.e., that $\mathcal{F}$ is flat over $S$ at every generic point of every fibre. If we ever need these results, we will carefully state and prove them here.

In your comment you can use Markdown and LaTeX style mathematics (enclose it like $\pi$). A preview option is available if you wish to see how it works out (just click on the eye in the toolbar).