6

I am looking for help in writing down the following formally/mathematically:


Let $(A, \frak p)$ be a local ring with $I \subset \frak p$ and $J \subset \frak p$.

The Zariski cotangent space of $A/I$ at $\frak p$ can be identified with $\frac{\frak p/\frak p^2}{(I+\frak p^2)/\frak p^2}$ which tells us that the tangent space of $A/I$ at $\frak p$ is a subspace of the tangent space of $A$ at $\frak p$. Similarly for the closed subscheme cut out by $J$.

The cotangent space of $A/(I+J)$ at $\frak p$ can be identified with $\frac{\frak p/\frak p^2}{(I+J+\frak p^2)/\frak p^2}$.

Informally

  • the tangent space of $A/I$ at $\frak p$ is cut out by the ideal $(I+\frak p^2)/\frak p^2$,
  • the tangent space of $A/J$ at $\frak p$ is cut out by the ideal $(J+\frak p^2)/\frak p^2$,
  • the tangent space of $A/(I+J)$ at $\frak p$ is cut out by the ideal $(I+J+\frak p^2)/\frak p^2$.

From this I gather that the intersection of the tangent spaces of $A/I$ and $A/J$ at $\frak p$ will be cut out by the ideal $(I+J+\frak p^2)/\frak p^2$? Not sure how to really write this out.

Therefore, the tangent space of the intersection $\operatorname{Spec}A/I \cap \operatorname{Spec} A/J=\operatorname{Spec}A/(I+J)$ at $\frak p$ is the intersection of the tangent spaces of $\operatorname{Spec}A/I$ and $\operatorname{Spec}A/J$ at $\frak p$.

user5826
  • 11,982

2 Answers2

4

Call these $X = \mathrm{Spec}(A)$ and $Y = \mathrm{Spec}(A/I)$ and $Z = \mathrm{Spec}(A/J)$. Denote the point $\mathfrak{p} \in \mathrm{Spec}(A)$ by $x \in X$.

Dualizing with respect to the field $\kappa(x)$, your quotients gives inclusions $T_x Y \to T_x X$ and $T_x Y \to T_xX$. By the same argument we get a diagram of inclusions, $\require{AMScd}$ \begin{CD} T_x (Y \cap Z) @>>> T_x Y\\ @V V V @VV V\\ T_x Z @>>> T_x X \end{CD} and we want this to be an intersection. This is the same as asking for this diagram to be a pullback in vectorspaces but we won't really need this. Going back to the undualed versions (which I write as $T^*_x X = \mathfrak{m}_x / \mathfrak{m}_x^2$ etc to save space), the above diagram is an intersection if and only if the diagram, $\require{AMScd}$ \begin{CD} T^*_x (Y \cap Z) @<<< T^*_x Y\\ @AAA @AAA\\ T^*_x Z @<<< T^*_x X \end{CD} is a ``gluing diagram'' or pushout of vector spaces. Explicitly, this means it identifies $T^*_x(Y \cap Z)$ with $T_x^* Y \oplus T^*_x Z$ modulo the subspace generated by $(v,-v)$ for $v \in T_x^* X$.

See if you can show this explicitly (or verify the universal property if you prefer) in terms of ideals.

Ben C
  • 1,423
  • I saw this answer here as well: https://stacks.math.columbia.edu/tag/0BEB – user5826 Dec 08 '21 at 17:03
  • But I got this exercise from Vakil's book. Is it a typo then? – user5826 Dec 08 '21 at 17:04
  • @hm2020 you certainly can but it's mostly a memory trick. The canonical algebraic gadget is the cotangent space so you need to turn "intersection" into its dual notion which swaps injections for surjections. Turning pullbacks into pushouts is one way of remembering what the corresponding operation to "intersection" is on the dual side. – Ben C Dec 09 '21 at 15:49
  • @hm2020 the fiber product is the intersection. – user5826 Dec 09 '21 at 19:18
  • @hm2020 if $i : X \to S$ is an open immersion then it is unramfied and thus $T_{X/S} = 0$. – Ben C Dec 09 '21 at 21:29
  • @hm2020 This is also true when $Y,Z$ are closed subschemes, as they are here. Check Vakil 9.2.C. – user5826 Dec 09 '21 at 23:41
  • @user46372819 the product is $X\times_k Y$, the intersection of $X$ and $Y$ as subschemes of some common scheme $Z$ is $X\times_Z Y$. Pay special attention to the product over $k$ versus the product over $Z$ - this is what's causing your confusion. This exercise is correct as stated, and this solution is correct and insightful. +1 and a good job to Ben C. – KReiser Dec 10 '21 at 03:55
  • @KReiser It looks like hm2020 deleted their comment, but when he wrote the fiber product over $k$, I was assuming it was a typo since we are not looking at the fiber product over $k$. We are looking at the intersection of two closed subschemes of $\operatorname{Spec}A$ which is given by the fiber product over $A$. – user5826 Dec 10 '21 at 21:38
  • @user46372819 right, which makes the stacks project link you were asking about not applicable. The way things are set up there just doesn't give the scenario you're curious about in this question. Ben C's answer above is exactly what you want. – KReiser Dec 10 '21 at 21:53
  • @KReiser With respect to the stacksproject link, why can't we take $S=\operatorname{Spec}A$ as the base scheme, and $X=\operatorname{Spec}A/I, Y=\operatorname{Spec}A/J$? Then that result says that the tangent space of the intersection is the direct sum of the tangent spaces? – user5826 Dec 11 '21 at 01:12
  • @user46372819 as I indicated above, the stacks project is referring to relative tangent spaces. For an unramified map (i.e. an embedding) these are always identically zero. – Ben C Dec 11 '21 at 02:02
  • @KReiser Got it. Thank you. So, is the definition of tangent space found in the following stacks project link (tag0B2C) different from the tangent space of a scheme at a point? https://stacks.math.columbia.edu/tag/0B2C – user5826 Dec 11 '21 at 16:03
  • @user46372819 the stacks project is defining the relative tangent space to a morphism, which may or may not be equal to the absolute tangent space that you're considering in your question. – KReiser Dec 11 '21 at 16:42
0

Question: "How do I formally show that the Zariski tangent space of the intersection of two closed subschemes is the intersection of the tangent spaces?"

Answer: For any ideals $I \subseteq I' \subseteq p \subseteq A$ there is a commutative diagram

$\require{AMScd}$ \begin{CD} 0 @>>> I @>>> p @>>> p_I:=p/I @>>> 0 \\ @VVV @VV V @VVV @VVV \\ 0 @>>> I' @>>> p @>>> p_{I'} @>>> 0 \end{CD}

This induce a canonical surjection

$$p_I/p_I^2 \rightarrow p_{I'}/p_{I'}^2$$

If $S:=Spec(A)$ and $U:=V(I), V:=V(J)$ with $\mathfrak{p} \in U \cap V$ it follows there are canonical surjections

$$A_{\mathfrak{p}} \rightarrow (A/I)_{\mathfrak{p}_I}$$

inducing surjections

$$ \mathfrak{p}/\mathfrak{p}^2 \rightarrow \mathfrak{p}_I/\mathfrak{p}_I^2 \rightarrow \mathfrak{p}_{I+J}/\mathfrak{p}_{I+J}^2$$

and

$$\mathfrak{p}/\mathfrak{p}^2 \rightarrow \mathfrak{p}_J/\mathfrak{p}_J^2 \rightarrow \mathfrak{p}_{I+J}/\mathfrak{p}_{I+J}^2.$$

When you dualize you get injections

$$(*)\text{ }T_\mathfrak{p}(U\cap V) \subseteq T_\mathfrak{p}(U) \subseteq T_\mathfrak{p}(S)$$

and

$$(**)\text{ }T_\mathfrak{p}(U\cap V) \subseteq T_\mathfrak{p}(V) \subseteq T_\mathfrak{p}(S)$$

and it follows you may intersect the vector spaces

$$T_\mathfrak{p}(U)\cap T_\mathfrak{p}(V) \text{ inside }T_\mathfrak{p}(S).$$

To intersect two vector spaces $V_1,V_1$ these vector spaces must be embedded into one common ambient vector space. The ambient vector space above is $T_{\mathfrak{p}}(S)$.

Hence you get a canonical inclusion of vector spaces

$$(***)\text{ }T_\mathfrak{p}(U\cap V) \subseteq T_\mathfrak{p}(U)\cap T_\mathfrak{p}(V) .$$

Claim: "Therefore, the tangent space of the intersection $\operatorname{Spec}A/I \cap \operatorname{Spec} A/J=\operatorname{Spec}A/(I+J)$ at $\frak p$ is the intersection of the tangent spaces of $\operatorname{Spec}A/I$ and $\operatorname{Spec}A/J$ at $\frak p$."

Example 1: Let $k$ be the complex numbers and let $A:=k[x,y],f:=y^2-x^3, g:=y$ and $B:=A/(f), C:=A/(g)$ with $S:=Spec(A), C_1:=Spec(V), C_1:=Spec(C)$.

It follows $C_1\cap C_2:=Spec(k[x,y]/(y,x^3))\cong Spec(k[x]/(x^3)):=Spec(R)$.

The ring $R$ has a unique maximal ideal $\mathfrak{m}:=(x)$ and $R_{\mathfrak{m}} \cong R$ and $\mathfrak{m}/\mathfrak{m}^2 \cong k\overline{x}$ is one dimensional. Hence $Spec(R)$ is zero dimensional with a 1-dimensional tangent space at the point $0$ corresponding to $\mathfrak{m}$.

We get an equality

$$T_0(C_1)\cap T_0(C_2)=T_0(C_1\cap C_2):=T_0(Spec(R)) \subseteq T_0(\mathbb{A}^2).$$

Example 2: If you let $k$ have characteristic $p>0$ and let $I,J\subseteq k[x_1^p,..,x_1^p]$ be two ideals with $C_1:=V(I), C_2:=V(J)$ where $C_i$ are curves with zero dimensional intersection and $p\in C_1\cap C_2(k)$, it follows $dim_k(T_p(C_1))=dim_k(T_p(C_2))=n$, hence $T_p(C_1)\cap T_p(C_2)$ has dimension $n$. The subscheme $C_1\cap C_2$ is zero dimensional and is defined by the ideal $I+J \subseteq k[x_1^p,..,x_n^p]$. For this reason it follows $dim_k(T_p(C_1 \cap C_2))=n$ hence there is an equality

$$T_p(C_1\cap C_2) = T_p(C_1) \cap T_p(C_2).$$

Note: When $A,B$ are $k$-algebras there is an isomorphism

$$P1.\text{ }\Omega^1_{A\otimes_k B/k} \cong \Omega^1_{A/k}\otimes_k B\oplus A\otimes_k \Omega^1_{B/k}.$$

Moreover if $k \rightarrow A \rightarrow A_i$ for $i=1,2$ are commutative unital rings, there is a surjection

$$0 \rightarrow I_{1,2} \rightarrow A_1\otimes_k A_2 \rightarrow A_1\otimes_A A_2 \rightarrow 0.$$

Note: When you take the intersection, you tensor over $A$, when you take the fiber product you tensor over $k$. It follows there is a surjection

$$ A_1\otimes_A A_2\otimes_{A_1\otimes_k A_2} \otimes \Omega^1_{A_1\otimes_k A_2/k} \rightarrow \Omega^1_{A_1\otimes_A A_2/k} \rightarrow 0.$$

Hence there is a quotient map

$$\Omega^1_{A_1\otimes_A A_2/k} \cong \Omega^1_{A_1\otimes_k A_2/k}/I_{1,2}(\Omega^1_{A_1\otimes_k A_2/k}).$$

Example: Let $X:=Spec(A), Y:=Spec(A_1):=Spec(A/I), Z:=Spec(A_2):=Spec(A/J)$. It follows

$$Y\cap Z:=Y\times_X Z:=Spec(A_1\otimes_A A_2) \cong Spec(A/(I+J)).$$

When you dualize you get a sequence of injections

$$T(Y\cap Z/k)\cong T(Y\times_X Z/k) \subseteq T(Y\times_k Z/k) \cong p^*T(Y/k)\oplus q^*T(Z/k) \subseteq T(X/k).$$

where $p,q:Y\times_k \rightarrow Y,Z$ are the projection maps.

The fiber of this injection at $p$ is the inclusion of tangent spaces coming from $(*)$ and $(**)$ above.

In this link

Tangent space of a product of algebraic group.

you find a proof that for any point $(y,z)\in Y\times_k Z$ it follows

$$T_{(y,z)}(Y\times_k Z/k)\cong T_y(Y/k) \oplus T_z(Z/k).$$

Example: There are two interesting exercises in Hartshorne on this: Ex.I.5.3+5.4 on multiplicity, intersection multiplicity and tangent directions.

If $C_1,C_2 \subseteq S:=\mathbb{A}^2_k$ are two plane curves and assume $C_1 \cap C_2$ is a scheme supported at a point $P\in C_1 \cap C_2$. The result in the "exercise 12.1.C" implies the following: If $C_i$ have the same tangent direction at $P$, this should imply that $T_P(C_1\cap C_2/k)\cong ke$ is one dimensional. If $C_1$ and $C_2$ do not have the same tangent direction at $P$ this should imply that $T_P(C_1 \cap C_2)=(0)$. Hence the exercise is related to how the curves $C_i$ intersect at $P$ and Ex HH.I.5.3, 5.4 above.

Example: Let $f:=x,g:=y,C_1:=V(f), C_2:=V(g)$. It follows the tangents of $C_i$ are not parallell at $p:=(0,0)$, and moreover $C_1 \cap C_2:=Spec(k[x,y]/(x,y))\cong Spec(k)$ which has zero tangent space. If $f:=y-x^2, g:=y$, it follows $C_1 \cap C_2:=Spec(k[x,y]/(y,x^2)) \cong Spec(k[x]/(x^2))$ which has 1-dimensional tangent space. The curves have parallell tangents at $p$.

hm2020
  • 1
  • I see. I got this exercise from Vakil’s book, exercise 12.1.C. So it is a typo? – user5826 Dec 08 '21 at 16:41
  • The exercise states: "Suppose $Y$ and $Z$ are closed subschemes of $X$, both containing the point $p \in X$. Show that $T_{Y \cap Z,p} = T_{Y,p} \cap T_{Z,p}$." Vakil does not mention what kind of scheme $X$ is, so I was assuming it is a general scheme. – user5826 Dec 08 '21 at 16:43
  • According to your edit, you say the exercise is not a typo. But haven't you provided a counterexample to the exercise with your Example 2? – user5826 Dec 08 '21 at 16:58
  • Also: When studying such intersection multiplicities and intersections you loose information when passing to the tangent bundle. – hm2020 Dec 11 '21 at 09:54