Sorry, your browser cannot access this site
This page requires browser support (enable) JavaScript
Learn more >

Presentation 定理二号, Krull PIT 是维数理论中十分重要的一个定理. 用它可以刻画出Noetherian scheme 上的tangent space 的维数大于等于局部维数的性质, 从而定义那些nonsingular 的点. Hartshorne 第一章就有这个定理, 关于tangent space的讨论则被藏在了习题5.10 里面. 值得注意的是Hartshorne 定理1.8A是一个非常不平凡的结果, 这个定理对于finitely generated kk algebra成立, 然而对于一般的catenary ring, 即任何prime ideal chain 长度相等的ring 却不一定成立, 也挺有意思的.

Krull Principal Ideal Theorem

In general idea of Zariski tangent space, we have dimTp,XdimRp\dim \mathbb{T}_{\mathfrak{p},X}\geq \dim R_{\mathfrak{p}}, where X=Spec(R)X=\mathrm{Spec}(R) and RR Noetherian. We will prove this in algebraic sense by the Krull’s principal ideal theorem.

Recall the Nakayama lemma first:

Lemma 1 (Nakayama). For RR be a ring and R\mathfrak{R} be its Jacobson radical, a\mathfrak{a} is an ideal contained in R\mathfrak{R}. Then

  1. For MM be a finite generated RR-module, M/aM=0M /\mathfrak{a}M=0 implies M=0M=0.

  2. For MM be a RR-module, NN' is a finite generated submodule of MM and M=aN+NM=\mathfrak{a}N'+N. Then M=NM=N.

  3. For {x1,,xn}M\{x_{1},\dots,x_{n}\}\subset M satisfies {x1ˉ,,xnˉ}\{\bar{x_{1}},\dots,\bar{x_{n}}\} generates M/aMM /\mathfrak{a}M, then {x1,,xn}\{x_{1},\dots,x_{n}\} generates MM.

Proof

Proof.

  1. Omitted. See the textbook.

  2. M=aN+NM=\mathfrak{a}N'+N implies M/N=a(N/N)M /N=\mathfrak{a}(N' /N), so M/NM /N is finitely generated. Then M/N=a(N/N)a(M/N)M/NM /N=\mathfrak{a}(N' /N)\subset \mathfrak{a}(M /N)\subset M /N, M/N=a(M/N)M /N=\mathfrak{a}(M /N). By (1), M/N=0M /N=0, i.e. M=NM=N.

  3. Consider a map f:RnMf:R^{n}\to M with f:(r1,,rn)r1x1++rnxnf:(r_{1},\dots,r_{n})\mapsto r_{1}x_{1}+\cdots+r_{n}x_{n}, we want to show ff is surjective, i.e. M/imf=0M /\mathrm{im}f=0. Let π:MM/aM\pi:M\to M /\mathfrak{a}M, then aRnker(πf)\mathfrak{a}R^{n}\subset \mathrm{ker}(\pi\circ f) so πf\pi\circ f factor through Rn/aRnR^{n}/ \mathfrak{a}R^{n}. Let fˉ:Rn/aRnM/aM\bar{f}:R^{n}/\mathfrak{a}R^{n}\to M /\mathfrak{a}M. fˉ\bar{f} maps (r1ˉ,,rnˉ)(\bar{r_{1}},\dots,\bar{r_{n}}) to r1ˉx1ˉ++rnˉxnˉ\bar{r_{1}}\bar{x_{1}}+\cdots+\bar{r_{n}}\bar{x_{n}} is surjective. Notice that we can write M=imf+aMM=\mathrm{im}f+\mathfrak{a}M, by (2)(2), we have M=imfM =\mathrm{im}f. ◻

We give the definition of height of an prime ideal.

Definition 1. Let p\mathfrak{p} be an prime ideal of RR, we call the dimension of local ring RpR_{\mathfrak{p}} the height of p\mathfrak{p}, i.e. ht(p)=dimRp\mathrm{ht}(\mathfrak{p})=\dim R_{\mathfrak{p}}.

Theorem 1 (Principal Ideal Theorem). Let RR be a Noetherian ring, for xRx\in R, then the minimal prime ideal p\mathfrak{p} containing (x)(x) has ht(p)1\mathrm{ht}(\mathfrak{p})\leq 1.

Proof

Proof. Localize at p\mathfrak{p} we can deduce to the case of Noetherian local ring R=RpR'=R_{\mathfrak{p}} and regard p\mathfrak{p} to be the maximal ideal of RR' and (x)(x) be the image in RR' of the principal ideal. Take q\mathfrak{q} be a prime ideal of RR' and qp\mathfrak{q}\subsetneq \mathfrak{p}, then we want to show q\mathfrak{q} to be the minimal prime ideal of RR', i.e. dimRq=0\dim R'_{\mathfrak{q}}=0, RqR'_{\mathfrak{q}} is Artinian.

Let q\mathfrak{q}' be the maximal ideal of RqR'_{\mathfrak{q}}, so q=qRq\mathfrak{q}'=\mathfrak{q}R'_{\mathfrak{q}}. If we can show (q)n=0(\mathfrak{q}')^{n}=0 for some nn, then for any prime ideal aRq\mathfrak{a}\subset R'_{\mathfrak{q}}, (q)na(\mathfrak{q}')^{n}\subset \mathfrak{a}, which implies qa\mathfrak{q}'\subset \mathfrak{a}. Then we have a=q\mathfrak{a}=\mathfrak{q}' and RqR'_{\mathfrak{q}} is Artinian.

Now we show (q)n=0(\mathfrak{q}')^{n}=0 for some nn. Let qn=(qn)c={rRrs(q)n for some sq}\mathfrak{q}_{n}=(\mathfrak{q}^{n})^{c}=\{r\in R'|rs\in (\mathfrak{q})^{n} \text{ for some }s\notin \mathfrak{q}\}. Notice that qn\mathfrak{q}_{n} is saturated and therefore qnRq=(q)n\mathfrak{q}_{n}R'_{\mathfrak{q}}=(\mathfrak{q}')^{n}. It’s easy to see we have q1q2\mathfrak{q}_{1}\supset \mathfrak{q}_{2}\supset\cdots. p\mathfrak{p} is minimal over (x)(x), we have dimR/(x)=0\dim R' /(x)=0, R/(x)R' /(x) is Artinian. The descending chain q1/(x)q2/(x)\mathfrak{q}_{1} /(x)\supset \mathfrak{q}_{2} /(x)\supset\cdots is stable. Assume for nNn\geq N, qn+1/(x)=qn/(x)\mathfrak{q}_{n+1}/(x)=\mathfrak{q}_{n}/(x), then aqn\forall a\in \mathfrak{q}_{n}, a+rx=bqn+1a+rx=b\in \mathfrak{q}_{n+1} for some rRr\in R and bqn+1b\in\mathfrak{q}_{n+1}. Then ba=rxqnb-a=rx\in \mathfrak{q}_{n}. Since p\mathfrak{p} is minimal over (x)(x), xqx\notin \mathfrak{q}, we have rqr\in \mathfrak{q}. Hence qn=qn+1+qn(x)\mathfrak{q}_{n}=\mathfrak{q}_{n+1}+\mathfrak{q}_{n}(x). By Nakayama lemma, qn=qn+1\mathfrak{q}_{n}=\mathfrak{q}_{n+1}, therefore qnRq=qn+1Rq\mathfrak{q}_{n}R'_{\mathfrak{q}}=\mathfrak{q}_{n+1}R'_{\mathfrak{q}}. (q)n=(q)n+1(\mathfrak{q}')^{n}=(\mathfrak{q}')^{n+1}, apply Nakayama lemma again, (q)n=0(\mathfrak{q}')^{n}=0. ◻

For general case, we have:

Theorem 2. Let RR be a Noetherian ring, p\mathfrak{p} is a minimal prime ideal containing (x1,,xn)(x_{1},\dots,x_{n}), then ht(p)n\mathrm{ht}(\mathfrak{p})\leq n.

Proof

Proof. W.l.o.g., assume RR is a Noetherian local ring. Let qp\mathfrak{q}\subset \mathfrak{p} with no prime ideal between p\mathfrak{p} and q\mathfrak{q}. We want to show q\mathfrak{q} is minimal over some (y1,,yn1)(y_{1},\dots,y_{n-1}). Since p\mathfrak{p} is minimal over (x1,,xn)(x_{1},\dots,x_{n}), (x1,,xn)⊄q(x_{1},\dots,x_{n})\not\subset\mathfrak{q}. Assume xnqx_{n}\notin \mathfrak{q}, then p\mathfrak{p} is minimal over (q,xn)(\mathfrak{q},x_{n}). Therefore dimR/(q,xn)=0\dim R /(\mathfrak{q},x_{n})=0 and R/(q,xn)R /(\mathfrak{q},x_{n}) is Artinian, the chain (p/(q,xn))(p/(q,xn))2(\mathfrak{p}/(\mathfrak{q},x_{n}))\supset (\mathfrak{p} /(\mathfrak{q},x_{n}))^{2}\supset\cdots is stable. By Nakayama lemma (p/(q,xn))m=0(\mathfrak{p} /(\mathfrak{q},x_{n}))^{m}=0 for some mm, i.e. pm(q,xn)\mathfrak{p}^{m}\subset (\mathfrak{q},x_{n}). Then xim=aixn+yix_{i}^{m}=a_{i}x_{n}+y_{i} for i=1,,n1i=1,\dots,n-1, for yiqy_{i}\in \mathfrak{q}.

Now we show q\mathfrak{q} is minimal over (y1,,yn1)(y_{1},\dots,y_{n-1}). First, p\mathfrak{p} is minimal over (y1,,yn1,xn)(y_{1},\dots,y_{n-1},x_{n}), otherwise (y1,,yn1,xn)pp(y_{1},\dots,y_{n-1},x_{n})\subset\mathfrak{p}'\subsetneq \mathfrak{p}. Since R/(x1,,xn)R /(x_{1},\dots,x_{n}) is Artinian, pk(x1,,xn)\mathfrak{p}^{k}\subset (x_{1},\dots,x_{n}) for some kk, then pkm(y1,,yn1,xn)ppp\mathfrak{p}^{km}\subset(y_{1},\dots,y_{n-1},x_{n})\subset \mathfrak{p}'\Rightarrow \mathfrak{p}\subset \mathfrak{p}', this shows p=p\mathfrak{p}'=\mathfrak{p}. Pass to R/(y1,,yn1)R /(y_{1},\dots,y_{n-1}), p/(y1,,yn1)\mathfrak{p} /(y_{1},\dots,y_{n-1}) is minimal over (xnˉ)(\bar{x_{n}}), the residue of (xn)(x_{n}) in R/(y1,,yn1)R /(y_{1},\dots,y_{n-1}). Then ht(p/(y1,,yn1))1\mathrm{ht}(\mathfrak{p} /(y_{1},\dots,y_{n-1}))\leq 1 and ht(q/(y1,,yn1))=0\mathrm{ht}(\mathfrak{q} /(y_{1},\dots,y_{n-1}))=0. So q\mathfrak{q} is minimal over (y1,,yn1)(y_{1},\dots,y_{n-1}). ◻

We also have the converse of Krull’s principal ideal theorem.

Theorem 3. Let RR be a Noetherian ring, p\mathfrak{p} is prime ideal and ht(p)=n\mathrm{ht}(\mathfrak{p})=n, then p\mathfrak{p} is minimal over and ideal a\mathfrak{a} which is generated by nn elements.

Proof

Proof. We use induction to prove the result.

First n=0n=0, obvious.

Assume for n1n-1. For qp\mathfrak{q}\subset \mathfrak{p} with no prime ideal between them, then ht(q)=n1\mathrm{ht}(\mathfrak{q})=n-1. By induction hypothesis, there are x1,,xn1x_{1},\dots,x_{n-1} s.t. q\mathfrak{q} is minimal over (x1,,xn1)(x_{1},\dots,x_{n-1}). Since RR is Noetherian, there are finitely many minimal primes over (x1,,xn1)(x_{1},\dots,x_{n-1}), denote them as q1,,qm\mathfrak{q}_{1},\dots,\mathfrak{q}_{m}. By prime avoidence, pi=1mqi\mathfrak{p}\neq\bigcup\limits_{i=1}^{m}\mathfrak{q}_{i} otherwise ppi\mathfrak{p}\subset \mathfrak{p}_{i} for some ii. Then there is xnpx_{n}\in \mathfrak{p} but xnqix_{n}\notin \mathfrak{q}_{i} for any ii. So the minimal prime ideal over (x1,,xn)(x_{1},\dots,x_{n}) has height at least nn, hence it’s p\mathfrak{p}. ◻

Application of Krull’s PIT

We will have the following geometric implication of Krull’s principal ideal theorem:

Theorem 4. For any pX\mathfrak{p}\in X, XX is a affine variety, let m\mathfrak{m} be the maximal ideal of the local ring RpR_{\mathfrak{p}}, then $$\dim \mathbb{T}_{\mathfrak{p},X}\geq \dim X$$ The equality holds if and only if RpR_{\mathfrak{p}} is a regular local ring, i.e. XX is nonsingular at p\mathfrak{p}.

Proof

Proof. Notice that dimTp,X=dimRp/mm/m2\dim \mathbb{T}_{\mathfrak{p},X}=\dim_{R_{\mathfrak{p}}/\mathfrak{m}} \mathfrak{m}/\mathfrak{m}^{2}. For {x1,,xn}m\{x_{1},\dots,x_{n}\}\subset \mathfrak{m} s.t. {x1ˉ,,xnˉ}\{\bar{x_{1}},\dots,\bar{x_{n}}\} generates m/m2\mathfrak{m}/\mathfrak{m}^{2} as Rp/mR_{\mathfrak{p}}/\mathfrak{m}-vector space, {x1ˉ,,xnˉ}\{\bar{x_{1}},\dots,\bar{x_{n}}\} also generates m/m2\mathfrak{m}/\mathfrak{m}^{2} as RpR_{\mathfrak{p}}-module. By Nakayama lemma, {x1,,xn}\{x_{1},\dots,x_{n}\} generates m\mathfrak{m}. From Krull’s PIT, ht(p)=dimRpn\mathrm{ht}(\mathfrak{p})=\dim R_{\mathfrak{p}}\leq n. ◻

For example, k[x1,,xn](x1,,xn)k[x_{1},\dots,x_{n}]_{(x_{1},\dots,x_{n})} is a regular local ring.

Theorem 5. Prime ideals in Noetherian ring RR also satisfies the descending chain condition.

Proof

Proof. For any prime p\mathfrak{p}, p\mathfrak{p} is finitely generated by nn elements, then ht(p)n\mathrm{ht}(\mathfrak{p})\leq n. ◻

Theorem 6. A Noetherian integral domain RR is UFD if and only if every prime ideal of height 1 is principal.

Proof

Proof. (\Leftarrow): Let pp be an irreducible element, q\mathfrak{q} is minimal over (p)(p), then ht(q)1\mathrm{ht}(\mathfrak{q})\leq 1. Since RR is domain, ht(q)=1\mathrm{ht}(\mathfrak{q})=1. Therefore q\mathfrak{q} is principal, pp is irreducible so (p)=q(p)=\mathfrak{q}.

(\Rightarrow): Let p\mathfrak{p} be an prime ideal s.t. ht(p)=1ht(\mathfrak{p})=1. By the converse of Krull’s PIT, p\mathfrak{p} is minimal over some (x)(x). Let x=cp1pnx=cp_{1}\cdots p_{n} with p1,,pnp_{1},\dots,p_{n} prime and cc unit. (p1)(pn)p(pi)p(p_{1})\cdots(p_{n})\subset \mathfrak{p}\Rightarrow (p_{i})\subset \mathfrak{p} for some ii. ht(p)=1\mathrm{ht}(\mathfrak{p})=1 implies (pi)=p(p_{i})=\mathfrak{p}. ◻

Lemma 2 (Generalized Noether Normalization Lemma). Let R=k[x1,,xn]R=k[x_{1},\dots,x_{n}] be a polynomial ring, a\mathfrak{a} be an ideal of RR. Let r=inf{ht(p)ap}r=\inf\{\mathrm{ht}(\mathfrak{p})|\mathfrak{a}\subset \mathfrak{p}\}. Then we can choose y1,,ynRy_{1},\dots,y_{n}\in R such that RR is integral over k[y1,,yn]k[y_{1},\dots,y_{n}] and ak[y1,,yn]=(y1,,yr)\mathfrak{a}\cap k[y_{1},\dots,y_{n}]=(y_{1},\dots,y_{r}).

We can derive the Noether normalization lemma from above lemma. From the lemma above we can show the following result:

Lemma 3. Let R=k[x1,,xn]R=k[x_{1},\dots,x_{n}] be a polynomial ring, p\mathfrak{p} be a prime ideal of RR. Then dimR/p=dimRht(p)\dim R /\mathfrak{p}=\dim R-\mathrm{ht}(\mathfrak{p}).

Proof

Proof. Let ht(p)=m\mathrm{ht}(\mathfrak{p})=m, then we can find y1,,ynRy_{1},\dots,y_{n}\in R such that RR integral over k[y1,,yn]k[y_{1},\dots,y_{n}] and pk[y1,,yn]=(y1,,ym)\mathfrak{p}\cap k[y_{1},\dots,y_{n}]=(y_{1},\dots,y_{m}). Then R/pR /\mathfrak{p} is integral over k[y1,,yn]/(pk[y1,,yn])k[y_{1},\dots,y_{n}]/(\mathfrak{p}\cap k[y_1,\dots,y_{n}]) and therefore $$\dim R /\mathfrak{p}=\dim k[y_{1},\dots,y_{n}]/(\mathfrak{p}\cap k[y_1,\dots,y_{n}])=\dim k[y_{m+1},\dots,y_{n}]=n-m$$ 

Theorem 7. Let kk be a field. An affine variety XAnX\subset\mathbb{A}^{n} has dimension n1n-1 if and only if it is the zero locus of a single nonconstant irreducible polynomial in k[x1,,xn]k[x_{1},\dots,x_{n}].

Proof

Proof. (\Leftarrow): For fk[x1,,xn]f\in k[x_{1},\dots,x_{n}] irreducible, (f)(f) is prime since k[x1,,xn]k[x_{1},\dots,x_{n}] is UFD, then Z(f)\mathcal{Z}(f) is a variety. By Krull’s PIT, we have ht(f)=1\mathrm{ht}(f)=1. Therefore dimZ(f)=dimk[x1,,xn]/(f)=dimk[x1,,xn]ht(f)=n1\dim\mathcal{Z}(f)=\dim k[x_{1},\dots,x_{n}] /(f)=\dim k[x_{1},\dots,x_{n}]-\mathrm{ht}(f)=n-1.

(\Rightarrow): Let X=Z(p)X=\mathcal{Z}(\mathfrak{p}) for some p\mathfrak{p} prime. We have dimk[x1,,xn]/p=n1\dim k[x_{1},\dots,x_{n}] /\mathfrak{p}=n-1. Notice that k[x1,,xn]k[x_{1},\dots,x_{n}] is UFD and ht(p)=1\mathfrak{ht}(\mathfrak{p})=1, by lemma above, we have p\mathfrak{p} is principal ideal. Thus X=Z(f)X=\mathcal{Z}(f) for some ff. ◻

We have a counterexample: A prime ideal p\mathfrak{p} of height 2 in k[x1,,xn]k[x_{1},\dots,x_{n}] may not be principal. Let XX be the affine algbraic set in A3\mathbb{A}^{3} (chark=0\mathrm{char} k=0) defined by {(t3,t4,t5)tk}\{(t^{3},t^{4},t^{5})|t\in k\}. Then use Grobner basis we can calculate I(X)=(xzy2,x3yz,z2x2y)\mathcal{I}(X)=(xz-y^{2},x^{3}-yz,z^{2}-x^{2}y) and ht(I(X))=2\mathrm{ht}(\mathcal{I}(X))=2.

评论