Patrick Lei

All notes

Commutative algebra

Lectures by Eric Urban, Columbia, Fall 2020.

Commutative Algebra
Fall 2020
Notes by Patrick Lei
Lectures by Eric Urban

Columbia University

Disclaimer

These notes were taken during lecture using the vimtex package of the editor neovim. Any errors are mine and not the instructor’s. In addition, my notes are picture-free (but will include commutative diagrams) and are a mix of my mathematical style and that of the instructor. If you find any errors, please contact me at plei@math.columbia.edu.

Any similarties to Matsumura’s Commutative Algebra are due to the instructor.

Acknowledgements

I would like to thank Caleb Ji, Nicolás Vilches, and Alex Xu for pointing out typos in these notes.

1 Basic Notions

The references we will use in this course are Matsumura’s Commutative Algebra and Serre’s Algèbre Locale, Multiplicités. There is an English translation of Serre. We will begin with general results on rings and modules. We will assume all rings are commutative and unital. Recall that and ideal I of a ring A is prime if and only if A/I is a domain, and I is maximal if and only if A/I is a field.

1.1 Basics of Ideals

Definition 1.1.1.

Let I⊂A be an ideal. Then the radical I of I is the set

Definition 1.1.2.

An ideal I⊂A is primary if I≠A and the zero divisors in A/I are nilpotent. Thus if x⁢y∈I and x∉I, then yn∈I for some n.

Proposition 1.1.3.

If Q⊂A is primary, then Q is a prime ideal.

Proof.

If x⁢y∈Q, then xn⁢yn∈Q. If xn∉Q, then y∈q because (yn)a∈Q. □

Remark 1.1.4.

The converse to Proposition 1.1.3 is false in general.

Definition 1.1.5.

Let A be a ring. Then the spectrum Spec⁡A of A as a set is the set of prime ideals of A. We may place the Zariski topology on this set, where the basis of open sets is given by Df=Spec⁡A∖Vf, where Vf is the set of prime ideals containing f.

If φ:A→B is a morphism of rings, the morphism φ∗:Spec⁡B→Spec⁡A is continuous in the Zariski topology.

Exercise 1.1.6.

In particular, if π:A→A/I, then π∗ is an embedding.

Exercise 1.1.7.

Let I⊂A be an ideal. Then let P1,…,Pr be ideals of A that are all prime except possibly two of them. Show that if I⊄Pi for all i, then I⊄⋃iPi.

Exercise 1.1.8.

Let 𝔞1,…,𝔞r be ideals of A such that 𝔞i+𝔞j=A. Then

  1. 1.

    ⋂i𝔞i=𝔞1⁢⋯⁢𝔞r

  2. 2.

    There is an isomorphism of rings A/⋂i𝔞i≅∏iA/𝔞i.

1.2 Localization

Let S⊂A be a multiplicative subset. The main examples are Sf={1,f,f2,…} and S𝔭=A∖𝔭 for a prime ideal 𝔭. Then if 0∉S, there is at least one ideal 𝔞 such that 𝔞∩S=∅. Denote the set of such 𝔞 by ℳS. Then any maximal element of ℳS is a prime ideal in A. Existence of a maximal element is seen using Zorn’s lemma.

To see that maximal elements of ℳS are prime ideals, note that (x)+P is not in ℳS, so if x,y∉P, there exist a,b∈A and s,s′∈S such that a⁢x≡smodP and b⁢y≡s′modP. Therefore a⁢b⁢x⁢y∉P, so x⁢y is not in P

Lemma 1.2.1.

Let nil⁡A be the set of all nilpotent elements. Then

Proof.

One direction is easy, so let x be contained in all prime ideals. Then consider the set Sx. If 0∉Sx, then ℳSx is nonempty, so it has a maximal element. This is a prime ideal, which implies x is not contained in some prime. □

Corollary 1.2.2.

Let Q be an ideal of A. Then Q is the intersection of all prime ideals containing Q.

Now fix a multiplicative subset S. Then we will define an equivalence relation on A×S. We write

if there exists t∈S such that t⁡(a⁢r−b⁢s)=0. If A is a domain, then this says that as=br. Now we will define the localization S−1⁢A to be the set of equivalence classes for this relation. Note there is a natural morphism A→S−1⁢A that sends a↦a1.

Note that the localization has a universal property: If φ:A→B is a morphism such that φ⁡(S)⊂B×, then φ factors uniquely through S−1⁢A.

Localization gives a map Spec⁡S−1⁢A→Spec⁡A, and in particular, if S={1,f,f2,…}, we recover the set Df=Spec⁡Af.

1.3 Modules

Let A be a ring. Then an A-module M is an abelian group with an action of A. If M is an A-module and S⊂A is a multiplicative set, then S−1⁢M is the set of equivalence classes for (m,s)∼(m′,s′) if there exists t∈S such that t⁡(s′⁢m−s⁢m′)=0. This is an S−1⁢A-module.

Lemma 1.3.1.

Let M be an A-module. Then the map

is injective.

Proof.

Let x∈M be nonzero. Then the annihilator of x is a proper ideal of A, so it is contained in a maximal ideal. This implies that xp∈Mp is nonzero. □

Corollary 1.3.2.

Let A be a domain. Then A=⋂𝔭A𝔭, where this intersection makes sense inside the fraction field of A.

Proof.

Apply the previous lemma to M=K/A. □

Definition 1.3.3.

Let M be an A-module and x∈A. Then x is M-regular if the morphism m↦x⁢m is injective. Additionally, if x is A-regular, then it is called regular.

The set S0 of all regular elements in A is multiplicative, and the ring S0−1⁢A is called the total ring of fractions. If A is a domain, then S0=A∖{0}, and S0−1⁢A is the field of fractions.

Definition 1.3.4.

A ring A is a local ring if A has only one maximal ideal. In this case, all elements not in the maximal ideal are units.

Remark 1.3.5.

If I⊂A is an ideal such that A∖I=A×, then A is a local ring and I its maximal ideal.

Example 1.3.6.

Now let A be a general ring and 𝔭∈Spec⁡A. Then A𝔭 is a local ring with maximal ideal 𝔭⁢A𝔭.

Definition 1.3.7.

Now suppose A,B are local rings. Then a morphism φ:A→B of rings is local if φ⁡(𝔪A)⊂𝔪B. This means we have a commutative diagram

where kA=A/𝔪A is the residue field of A.

Recall that the nilradical is the set of all nilpotent elements, or equivalently the intersection of all prime ideals. Then the Jacobson radical rad⁡A is defined to be the intersection of all maximal ideals.

Proposition 1.3.8.

Let x∈A. Then x∈rad⁡A if and only if 1+x⁢a is a unit for any a∈A.

Proof.

If (1+x)⁢A≠A, then 1+x is contained in some maximal ideal m⁢f⁢m, which implies 1∈𝔪. In the other direction, suppose there exists some maximal ideal 𝔪 such that x∉𝔪. Then x is nonzero in A/𝔐. Thus there exists b such that 1−x⁢b∈m, which contradicts the assumption that 1+x⁢a is a unit for any a. □

Lemma 1.3.9 (Nakayama’s Lemma).

Let M be a finitely generated A-module. Then let I be an ideal such that I⁢M=M. Then there exists x∈I such that (1+x)⁢M=0. In particular, if I⊆rad⁡A, then M=0.

Proof.

We will induct on the number of generators. If M=A.m, then m=x⁢m for some x∈I, and thus (1−x)⁢m=0. Now suppose M=A⁢m1+⋯+A⁢mr. Let M′=M/A⁢mr. By the inductive hypothesis, (1+x)⁢M′=0 for some x∈I. Therefore (1+x)⁢M⊂A⁢mr, so (1+x)⁢I⁢M=(1+x)⁢M⊂I⁢mr. Therefore (1+x)⁢mr=y⁢mr for some y∈I, and thus (1+x−y)⁢mr=0. Thus (1+x)⁢(1+x−y)⁢M⊂(1+x−y)⁢A⁢mr=0. □

Corollary 1.3.10.

Let N,N′⊂M and I⊂A such that M=N+I⁢N′. Then if either

  1. 1.

    I is nilpotent;

  2. 2.

    I⊂rad⁡A and N′ is finitely generated,

then M=N.

Proof.
  1. 1.

    Suppose I is nilpotent. Then

    because I is nilpotent.

  2. 2.

    Let I⊆rad⁡A and N′ be finitely generated. Then set M0=M/N=I⁢N0′, where N0′ is the image of N′ inside M0. Because N0′ is finitely generated, so is M0. Therefore M0=I⁢M0=0, so M=N.

□

Remark 1.3.11.

Most of the time, we apply this result when A is local and I is the maximal ideal of A. In this case, M/𝔪⁢M is a finite-dimensional vector space over A/𝔪.

1.4 Artinian and Noetherian Rings

Definition 1.4.1.

We say that an A-module M satisfies the ascending chain condition if any ascending chain of submodules of M becomes stationary. Similarly, M satisfies the descending chain condition if any descending chain of submodules becomes stationary. If M satisfies the ascending chain condition, it is called Noetherian, and if M satisfies the descending chain condition, it is Artinian.

Proposition 1.4.2.

Assume we have a short exact sequence of A modules

Then M is Noetherian (resp. Artinian) if and only if N and P are.

Proof.

Proving that if M is Noetherian, then N and P are is left to the reader. Now consider a chain

Then let Pi be the image of the Mi in P and Ni=N∩Mi. Then we have an exact sequence

Because (Ni) and (Pi) stabilize, so must Mi from the exact sequence. □

Corollary 1.4.3.

If A is Noetherian (resp. Artinian), then any finitely generated A-module is Noetherian (resp. Artinian).

Corollary 1.4.4.

Assume A is Noetherian. Then any finitely generated A-module M has a projective resolution by finite free A-modules. In other worse, there exists an exact sequence

such that each Fi=Ami.

Proof.

Suppose M is finitely generated. Then M=A⁢m1+⋯+A⁢mr, so we have a sequence

Then ker⁡φ0=N0 and F0=Ar. Then we repeat this process with N0 taking the role of M. □

Proposition 1.4.5.

An A-module M is noetherian if and only if any submodule of M is finitely generated.

Proof.

Let N⊆M. Then choose n1∈N. Then if A⁢n1≠N, choose n2∈N∖A⁢n1. This process will stop because M is Noetherian, so N is finitely generated.

Now suppose any submodule is finitely generated. Given a chain

set N=⋃iMi. This is finitely generated and is also equal to the first Mi that contains all of the generators. □

This means that a ring A is noetherian if and only if all ideals of A are finitely generated. In particular, fields and principal ideal domains are Noetherian.

Proposition 1.4.6.

Let M be Noetherian and suppose S is a multiplicative subset of A. Then S−1⁢M is Noetherian.

Proof.

Consider the morphism M→S−1⁢M. Then let Ni be a chain of S−1⁢A-modules in S−1⁢M. Their preimages Mi form a chain, and they are stationary, so Ni is also stationary. □

Theorem 1.4.7.

Let A be a Noetherian ring. Then A⁡[X] is Noetherian.

Proof.

Let I⊂A⁡[X]. Then 𝔄n⊂A be generated by the dominant coefficients of polynomials in I of degree at most n. Then we can write a∈𝔄n as a=∑αi⁢βi where αi∈A and βi a dominant coefficient of a polynomial of degree at most n in I. Thus the 𝔄n form a chain of ideals of A that stabilizer for n≥N. Then 𝔄N=(β1,…,βr). Set Qi=βi⁢XN+⋯∈I. If P∈I, then there exists S such that P=Q⁢S+R such that Q∈A⁢Q1+⋯+A⁢Qr and deg⁡R<N.

Therefore P∈(Q1,…,Qr)+A⁢[X]N−1∩I, so I⊂(Q1,…,Qr)+A⁢[X]N−1∩I and is thus finitely generated. □

Corollary 1.4.8.

Let B be a finitely-generated A-algebra. Then if A is Noetherian, B is also Noetherian.

Corollary 1.4.9.

Any finitely generated algebra over a field is Noetherian.

Remark 1.4.10.

Suppose A is Noetherian and M an A-module. If M is finitely generated, then M is Noetherian, but submodules are not necessarily Noetherian. However, they are finitely generated.

Suppose A⊂B is an inclusion of rings. Then we say that x∈B is integral over A if there exists a monic polynomial Q∈A⁡[t] such that Q⁡(x)=0.

Proposition 1.4.11.

The following are equivalent:

  1. 1.

    x∈B is integral over A;

  2. 2.

    A⁡[x] is a finitely-generated A-module;

  3. 3.

    There exists A⁡[x]⊂C⊂B such that C is a finitely-generated A-module.

  4. 4.

    There exists a faithful A⁡[x]-module M which is finitely generated over A.

Proof.
1 implies 2

Note that A⁡[x] is generated by 1,x,x2,…,xm, where Q has degree m.

2 implies 3

Set C=A⁡[x].

3 implies 4

Choose M=C.

4 implies 1

Write M=A⁢m1+⋯+A⁢mr. M is an A⁡[x]-module, so we can consider x.M⊂M. Then for all i, we have x⁢mi=∑ai⁢j⁢mj, so if we write consider the matrix T=(ai⁢j), then this matrix represents the map given by multiplication by x. Therefore we have

so set Q=det⁡(T−x⁢Ir). This is monic and Q⁡(x).mi=0 for all i, and therefore Q⁡(x) acts by 0 on M. Because M is a faithful A⁡[x]-module, we have Q⁡(x)=0.

□

Exercise 1.4.12.

Let x,y∈B⊃A. Show that if x and y are integral over A then so are x+y,x⁢y.

Proposition 1.4.13.

Let A⊂B⊂C. Assume that A is Noetherian and that C is a finitely-generated A-algebra. If C is a finitely-generated B-module, then B is a finitely-generated A-algebra.

Proof.

Write C=B⁢c1+⋯+B⁢cr. Also, we can write C=A⁡[x1,…,xm] for some xi∈C. Then we can write xi=∑bi⁢j⁢cj and ci⁢cj=∑bi⁢j⁢k⁢ck for bi⁢j,bi⁢j⁢k∈B. Then B0=A⁡[bi⁢j,bi⁢j⁢k] is a finitely-generated A-algebra. Any element of C is a polynomial in the xi with coefficients in A, so C is a finitely-generated B0-module. In particular, B0 is Noetherian. Because B⊂C, this implies that B is a finitely generated B0-module, so it is a finitely-generated A-algebra. □

Corollary 1.4.14.

Let k be a field and E a finitely-generated k-algebra. If E is a field, then E is a finite extension of k.

Proof.

Let E be a finitely-generated k-algebra. Then there exist x1,…,xr∈E that are algebraically independent over k. Then E is algebraic over k⁡(x1,…,xr), which is the field of fractions of k⁡[x1,…,xr]. However, this gives an inclusion k⊂F⊂E, where E is a finitely-generated k-algebra and E is algebraic over F.

By the proposition, F is a finitely-generated k-algebra. Therefore, we can write F=h⁡[y1,…,ys], where yi=figi. Because k⁡[x1,…,xn] is a UFD, then we can write

h is repatively prime to all of the gi, so 1h∉k[y!,…,ys]. This gives a contradiction, so E must be algebraic over k. □

1.4.1 Primary Decomposition in Noetherian Rings

Definition 1.4.15.

An ideal 𝔞⊂A is irreducible if for any decomposition 𝔞=𝔟∩𝔠, then either 𝔞=𝔟 or 𝔞=c.

Example 1.4.16.

If 𝔞 is a prime ideal, then 𝔞 is irreducible. To see this, if 𝔞∣𝔟⁢𝔠, then 𝔞 contains one of 𝔟,𝔠, and so either 𝔞=𝔟 or 𝔞=𝔠.

Remark 1.4.17.

Suppose 𝔪⊂A is a maximal ideal. Then any power 𝔪n of 𝔪 is primary.

Proof.

We want to prove that the zero divisors of A/𝔪n are nilpotent. Because 𝔪 is maximal, then A/𝔪n is a local ring with maximal ideal 𝔪/mn. But then A/𝔪n∖𝔪/𝔪n are all units, so everything in 𝔪 is nilpotent. □

Lemma 1.4.18.

If A is Noetherian, then every irreducible ideal is primary.

Proof.

Let 𝔞⊂A be irreducible. Then we can pass to the quotient, so we may asusme 𝔞=0. Let x,y be nonzero with x⁢y=0. We want to show that x is nilpotent.

Because A is Noetherian, then there exists n such that Ann⁡xn=Ann⁡xn+1. We want to show that (xn)∩(y)=0, so choose z=a⁢xn=b⁢y. Then z⁢x=a⁢xn+1=b⁢y⁢x=0, so a∈Ann⁡xn+1=Ann⁡xn. However, this means z=0. Because 0 is irreducible, then (xn)=0, so xn=0. □

Corollary 1.4.19.

If A is Noetherian, then every ideal of A has a primary decomposition. In other words, we can write I=𝔞1∩⋯∩𝔞r, where each 𝔞𝔦 is primary.

Proof.

Let S be the set of ideals with no primary decomposition. If S is nonempty, then S has a maximal element I. To see this, we can use the fact that A is Noetherian, so any chain of ideals in S eventually stabilizes. We know that I is not irreducible, we can write I=𝔞∩𝔟 such that I≠𝔞,𝔟. In addition, 𝔞,𝔟∉S, so they have a primary decomposition. This implies that 𝔞∩𝔟=I has a primary decomposition. □

Remark 1.4.20.

This decomposition is not unique. For example, consider I=⟨x2,x⁢y⟩⊂k⁡[x,y]. Then I=⟨x⟩∩⟨x2,x⁢y,yn⟩ for all n>0.

1.4.2 Artinian Rings

Proposition 1.4.21.

Assume that A is Artinian.

  1. 1.

    Every prime ideal of A is maximal.

  2. 2.

    A has finitely many maximal ideals.

  3. 3.

    The Jacobson radical of A is nilpotent.

Proof.
  1. 1.

    Fix a prime ideal 𝔭 and consider the domain B=A/𝔭. Choose B∋x≠0 and consider the decreasing chain (xn) of ideals. This stabilizes, so there exists (xn)=(xn+1), so we can write xn=xn+1⁢y for some y∈B, and therefore 1=x⁢y because B is a domain. Therefore x has an inverse, so B is a field. Thus 𝔭 is maximal.

  2. 2.

    Suppose we have infinitely many maximal ideals 𝔭1,…,𝔭n,… that are pairwise distinct. Then we form a chain

    which becomes stationary. Therefore 𝔭1⁢⋯⁢𝔭n⊂𝔭n+1, so 𝔭n+1 contains some 𝔭i. Because these ideals are maximal, this is a contradiction.

  3. 3.

    Consider I=rad⁡(A)=𝔪1∩⋯∩𝔪n. Then the chain I⊃I2⊃⋯ stabilizes, so In=In+1 for some n. Let J=((0):In).1 We will show that J=A. If not, let J′⊋J such that J′ is minimal for this property. Such a J′ exists because A is Artinian.

    Let x∈J′∖J and consider the ideal A⁢x+J By minimality of J′, we see that I⁢x+J⊊J′ (otherwise J=J′ by Nakayama’s lemma). Therefore I⁢x+J=J, so I⁢x⊂J and thus x∈(J:I). Therefore, In+1⁢x⊂In⁢J=(0). This implies In⁢x=0, so x∈J and thus J′=J. □

Definition 1.4.22.

An A-module M is called irreducible if 0 and M are the only submodules of M.

Definition 1.4.23.

An A-module M is said to be of finite length if there exists a (finite) decreasing sequence of submodules

such that Mi/Mi+1 is irreducible for i=0,…,n. In this case, n is actually unique and depends only on M. We will call n the length of M.

Proposition 1.4.24.

Let A be a ring. Then A is Artinian if and only if A is of finite length as an A-module.

Proof.

If A is of finite length, then we have a sequence A=M0⊋⋯⊋Mn+1=0 where Mi/Mi+1 is irreducible. If 𝔞1⊃𝔞2⊃⋯ is a decreasing chain of ideals, so 𝔞i∩Mn is a decreasing chain of ideals. However, each is either Mn or 0, so this chain stabilizes. Similarly, the chain (Mj∩𝔞i)/(Mj+1∩𝔞i) also stabilizes for all j. Therefore, there exists N such that for all i>N, Mj∩𝔞i/(Mj+1∩𝔞i) is constant for all j, so 𝔞i is constant for all i>N.

Now suppose that A is Artinian. Choose I=rad⁡(A)=𝔪1∩⋯∩𝔪m, where the 𝔪i are the maximal ideals of A. Then I is nilpotent, so there exists n>0 such that

Then A=A/In=∏A/𝔪jn by the Chinese remainder theorem, so A/𝔪jn is clearly a local ring and is of finite length as an A-module. Note that the A/𝔪j-vector space 𝔪ji/𝔪ji+1 is finite-dimensional because A is Artinian. Therefore 𝔪ji/𝔪ji+1 is of finite length. □

Exercise 1.4.25.

If there is an exact sequence 0→N→M→P→0 of A-modules, then M is of finite length if and only if N and P are of finite length. Moreover, ℓ⁡(M)=ℓ⁡(P)+ℓ⁡(N).

Theorem 1.4.26.

A is Artinian if and only if A is Noetherian and dimA=0.

Proof.

If A is Artinian, we have already proved that dimA=0. By the previous proposition, because A is of finite length, A is Noetherian. To see this, for a chain 𝔞1⊂𝔞2⊂⋯, note that 𝔞m∩Mi/𝔞m∩Mi+1 stabilizes. We can do this for each i, so any increasing chain stabilizes.

Now assume A is Noetherian and has dimension 0. We know that (0) has a primary decomposition, so we can write (0)=𝔮1∩⋯∩𝔮r, where each 𝔮i is primary. Then 𝔪i=𝔮i is a prime ideal, so it is maximal because dimA=0. Because A is Noetherian and for all x∈𝔪i, xn∈𝔮i for n≫0, so there exists N such that 𝔪iN⊂𝔮i for each i. Therefore

so 𝔪1N⁢⋯⁢𝔪1N=0. Therefore, A≅A/𝔪1N×⋯×A/𝔪rN. Each A/𝔪iN is of finite length (because each 𝔪ij/𝔪ij+1 is a finite-dimensional vector space), so A is of finite length. □

Proposition 1.4.27.

Let A be a Noetherian local ring with maximal ideal 𝔪. Then one of the following holds:

  1. (a)

    Either 𝔪n⊋𝔪n+1 for all n, or;

  2. (b)

    𝔪n=0 for n≫0 and in this case, A is Artininian.

Proof.

If 𝔪n=𝔪n+1, then 𝔪n=0 by Nakayama’s lemma. This implies that A=A/𝔪n is of finite length. Then if 𝔭 is prime, then 𝔪n=(0)⊂𝔭, so 𝔪⊂𝔭. Because 𝔪 is maximal, 𝔪=𝔭, so dimA=0. □

Theorem 1.4.28 (Structure Theorem for Artinian Rings).

An Artinian ring is uniquely up to isomorphism a finite product of Artinian local rings.

Proof.

Previously, we proved that A=∏A/𝔪iN. Each of these is a local Artinian ring. □

2 Linear Algebra of Modules

Proposition 2.0.1.

Assume M,N,P are A-modules.

  1. 1.

    The sequence N→M→P→0 is exact if and only if for all A-modules Q, the sequence

    is exact.

  2. 2.

    The sequence 0→N→M→P is exact if and only if for all A-modules Q, the sequence

    is exact.

Proof.

This is left as an exercise. □

Remark 2.0.2.

In general, if 0→M→N→P→0 is exact, then

are exact but the last morphism is not necessarily surjective.

Definition 2.0.3.

A module Q is projective if the functor Hom⁡(Q,−) is exact. Here, exact means that short exact sequences are preserved. Similarly, a module I is injective if the functor Hom⁡(−,I) is exact.

Proposition 2.0.4.

A module Q is projective if and only if Q is a direct factor of a free module. In other words, there exists a free module F and A-module Q′ such that F=Q⊕Q′.

Proof.

Suppose Q is projective. Then there is a surjection π:A(S)→Q→0. Because Q is projective, there exists a map θ such that π∘θ=id. Therefore A(S)≅Q⊕Q′, where Q′ is the kernel of π.

On the other hand, if A(S)=Q⊕Q′, then for any diagram of the form

we can embed Q in A(S) and then use projectivity of free modules (because Hom⁡(A,M)=M). □

Remark 2.0.5.

If M is projective and finitely generated, then it is a direct factor of a finite free module.

Definition 2.0.6.

A projective resolution of an A-module M is a right bounded complex

of projective modules such that there exists P0→M such that

is exact.

Exercise 2.0.7.

Show that any module has a projective resolution (Hint: construct a free resolution). In addition, any two projective resolutions are homotopic.

Definition 2.0.8.

kq:Mq→Nq+1 such that ϕq=dN∘kq+kq−1∘dM.

2.1 Tor and Ext Functors

Note that for a complex, we can compute the homology Hq⁢(M∙)≔ker⁡dq/Imdq+1. This measures the defect of the complex from being exact. For functors that are not exact, we can construct derived functors that measure the defect of exactness. Let F:A⁢-⁢𝖬𝗈𝖽→A⁢-⁢𝖬𝗈𝖽 be right exact. Then for any M, we can consider a projective resolution P∙→M→0. Applying F to P., then the left derived functor L∙⁢F⁢(M) is defined by L∙⁢F⁢(M)=H∙⁢(F⁡(P∙)).

Proposition 2.1.1.

If 0→M→N→P→0 is exact, then we have a long exact sequence

Recall that the tensor product M⊗N of two modules M,N is an A-module with a bilinear map M×N→M⊗N such that all bilinear maps M×N→P factor through M⊗N.

Proposition 2.1.2.

The functors −⊗N,Hom(N,−) are an adjoint pair.

Corollary 2.1.3.

If N→M→P→0 is exact, then

is exact.

Definition 2.1.4.

A module Q is flat if −⊗Q is exact.

We can defined the left derived functors Torq⁡(Q,M) of the tensor product.

Proposition 2.1.5.

Any projective module is flat.

Proof.

Clearly free modules are flat, so write Q⊕Q′=A(S) and then note that the tensor product distributes over the sum. □

Proposition 2.1.6.

The following are equivalent:

  1. 1.

    M is flat over A.

  2. 2.

    If N′↪N, then M⊗N′↪M⊗N.

  3. 3.

    For all finitely generated ideals I⊂A, I⊗M↪M.

  4. 4.

    For any finitely generated ideals I⊂A, Tor1⁡(M,A/I)=0.

  5. 5.

    For any finitely generated module N, we have Tor⁡(M,N)=0.

  6. 6.

    For all ai∈A and xi∈M such that ∑ai⁢xi=0 there exist y1,…,ys∈M and bi⁢j such that xi=∑bi⁢j⁢yj.

Proof.

It is clear that 1 is equivalent to 2 implies 3 implies 4 implies 5. The directions 3 implies 2 and 4 implies 3 are left to the reader.

1 implies 6

Choose ai∈A,xi∈M such that ∑i=1rai⁢xi=0. Then define a map Ar→𝑓A by

and define K=ker⁡f. Because M is A-flat, we have an exact sequence

Then (x1,…,xr)∈ker⁡f⊗idM. Therefore there exists b1,…,bs∈K and y1,…,ys∈M such that

Writing bj=(b1⁢j,…,br⁢j), we obtain the identity

and thus xi=∑bj⁢i⁢yj.

6 implies 3

Choose an ideal I⊂A. Consider the map 0→I⊗M→M. Then for any element in the kernel, we can write

Then we can write xi=∑bi⁢j⁢yj and so

and thus I⊗M→M is injective. □

Let ϕ:A→B be a map of rings and let M be a B-module. Define ϕ to be flat if B is flat as an A-module.

Proposition 2.1.7.

If ϕ:A→B is flat and M is a flat B-module, then M is also flat as an A-module.

Proof.

Let S be an A-module. Then S⊗AM=S⊗A(B⊗BM)=(S⊗AB)⊗BM. If 0→N1→N2 is an exact sequence of A-module, by flatness of B as an A-module, then

is exact. Because M is flat over B, we see that

is exact, as desired. □

Now let M be an A-module. Then for any map A→ϕB, we can consider the B-module M(B)≔M⊗AB.

Proposition 2.1.8.

If M is A-flat, then M(B) is B-flat.

Proof.

For a B-module S, write

Thus if 0→S1→S2 is exact, then 0→S1⊗AM→S2⊗AM is exact because M is A-flat, as desired. □

Proposition 2.1.9.

Let S⊂A be a multiplicative subset of A. Then the morphism of rings A→S−1⁢A is flat.

The proof is left to the reader. This can be reformulated as M⊗AS−1⁢A≅S−1⁢M.

Now we will give some remarks about the Ext functors. For any left exact functor, we may define the right derived functors R∙⁢F by

where M→I∙ is an injective resolution. Then we will define the right derived functors of HomA⁡(N,−) by ExtAi⁡(N,−).

Proposition 2.1.10.

If M is injective, then ExtAi⁡(N,M)=0 for all i>0. Similarly, if N is projective, then ExtAi⁡(N,M)=0 for all I>0.

Remark 2.1.11.

We can compute E⁢x⁢ti⁢(N,M) using a projective resolution of N.

Proposition 2.1.12.

Let A→B be a morphism of rings and let M,N be A-modules. Then let M(B),N(B) be their base changes to B. Then we have

and

if B is A-flat.

Proof.

This follows from the definition of Ext,Tor using projective resolutions using the following facts.

  1. 1.

    If M is A-projective, then M(B) is B-projective.

  2. 2.

    Since B is A-flat, for any complex X∙ of A-modules, then H∙⁢(X(B)∙)=H∙⁢(X∙)(B).

□

2.2 Flatness

Proposition 2.2.1.

Let A be a local ring. Then any finitely generated flat A-module is free. In particular, free, projective, and flat are equivalent for A-modules.

Proof.

We know that free implies projective implies flat. Therefore we will show that if M is flat, then it is free. Assume that M is finitely generated and A-flat. Let k=A/𝔪 be the residue field of A. Define M¯=M⊗Ak, which is a vector space of finite dimension over k. Then there exists x1,…,xr∈M that descend to a basis of 𝔐.

Then the map Ar→M,(ai)↦∑ai⁢xi is surjective by Nakayama’s lemma. We will prove that this map is injective by induction. If r=1, then suppose a⁢x1=0. Then there exist y1,…,ys,b11,…,b1⁢s such that

where a⁢bi⁢j=0 for all j=1,…,s. Because x¯1≠0, there exists j such that b¯1⁢j≠0 so b1⁢j is invertible in A. Thus a=0.

Now suppose a1⁢x1+⋯+ar⁢xr=0. Then there exist y1,…,ys and bi⁢j such that

and

Because x¯r≠0, we see that b¯r⁢j≠0 for some j and thus bi⁢j is a unit. Then a1⁢b1⁢j+⋯+ar⁢br⁢j=0, so we can write

We know that x¯1−c1⁢x¯r,…,x¯r−1−cr−1⁢x¯r are linearly independent over k, so from the induction a1=⋯=ar=0 and thus ar=0. □

Remark 2.2.2.

If M is not finitely generated, the proposition is false. An example is given by taking the field of fractions of a local domain.

When proving the proposition, we in fact proved that

Lemma 2.2.3.

If x1,…,xr∈M with M a flat A-module for A a local ring and x¯1,…,𝔵r are linearly independent in M⊗Ak, then x1,…,xr are linearly independent in M.

Proposition 2.2.4.

Suppose that A→B is flat and I1,I2 are ideals of A. Then

  1. 1.

    (I1∩I2)⁢B=I1⁢B∩I2⁢B;

  2. 2.

    If I2 is finitely generated, then (I1:I2)⁢B=(I1⁢B:I2⁢B).

Proof.

The proof is a formal consequence of flatness.

  1. 1.

    Consider the exact sequence 0→I1∩I2→A→A/I1×A/I2. Tensoring with B, we obtain an exact sequence

    But then (I1∩I2)⊗B=(I1∩I2)⁢B, but the kernel of the last map is clearly I1⁢B∩I2⁢B.

  2. 2.

    Set I2=(x1,…,xr). Then because

    it suffices to prove the result for I2 a principal ideal. We have an exact sequence

    Tensoring by B, we obtain

    and by analysing the kernel, we see that (I1:x⁢A)⁢B=(I1⁢B:x⁢B). By repeated application of the previous part, the desired result follows. □

Example 2.2.5.

We will give an example where the previous proposition is not true in general. Let A=k⁡[x,y] and B=A/x⁢A=k⁡[y]. Then choose I1=(x+y),I2=(y), so I1∩I2=I1⁢I2=((x+y)⁢y). But then we have (I1∩I2)⁢B=y2⁢B, but I1⁢B∩I2⁢B=y⁢B.

Another example is A=k⁡[x,y],B=k⁡[x,y,z]/(x⁢z−y)≅k⁡[x,z],I1=x⁢A,I2=y⁢A. Here we can check that (I1∩I2)=(x⁢y), that (I1∩I2)⁢B=x2⁢z⁢B, but I1⁢B∩I2⁢B=x⁢z⁢B. Viewing this geometrically as Spec⁡B→Spec⁡A, we can check the fiber over (0,0) and see that the map is not flat.

Proposition 2.2.6.

Let A→𝜑B be a ring homomorphism. The following are equivalent:

  1. 1.

    B is flat over A;

  2. 2.

    B𝔓 is flat over A𝔭 for all 𝔓∈Spec⁡B and 𝔭=φ−1⁢(𝔓).

  3. 3.

    B𝔓 is flat over A𝔭 for any 𝔓 maximal.

Proof.
1 implies 2:

We know that B𝔭 is flat over A𝔭. But then B𝔓 is flat over B𝔭 because it is a localization. By transitivity of flatness, B𝔓 is flat over A𝔭.

2 implies 3:

This is obvious.

3 implies 1:

Note that for all 𝔓 maximal, ToriA⁡(B,N)𝔓=0 for i>0. This implies that Tori!(B,N)=0, and thus B is flat over A. To get that the first Tor is zero we need to use the lemma belos. □

Lemma 2.2.7.

Let φ:A→B be a morphism of rings and choose 𝔓∈Spec⁡B. Then let 𝔭=φ−1⁢(𝔓) and N an A-module. Then ToriA⁡(B,N) is a B-module and ToriA⁡(B,N)𝔓=ToriA𝔭⁡(B𝔓,N𝔭).

Proof.

Let X∙→N be a projective resolution. Then Tor is computed by the homology of the complex B⊗AX∙. When we localize, we localize the homology at the B term. However, B𝔭⊗AX∙=B𝔓⊗A𝔭(X∙)𝔭, so because Xi is A-projective, then X𝔭 is A𝔭-projective, and thus (X∙)𝔭 is a projective resolution of N𝔭. Thus the complex B𝔓⊗A𝔭(X∙)𝔭 computes the Tor as desired. □

Definition 2.2.8.

An A-module N is said to be faithfully flat if

  1. 1.

    N is A-flat;

  2. 2.

    For any sequence P→Q→R of A-modules, if P⊗N→Q⊗N→R⊗N is exact, then P→Q→R is exact.

Theorem 2.2.9.

Let M be an A-module. Then the following are equivalent:

  1. 1.

    M is faithfully flat over A;

  2. 2.

    M is flat and for any nonzero N, M⊗N≠0;

  3. 3.

    M is flat and for all maximal ideals 𝔪⊂A, 𝔪.M≠M.

Proof.
1 implies 2:

Choose the sequence 0→N→0. Then tensor with M. If N⊗M=0, the sequence is now exact, then the original sequence is exact, and thus N=0.

2 implies 3:

Consider N=A/𝔪. Then N⊗M=M/𝔪⁢M≠0, so M≠𝔪⁢M.

3 implies 2:

Choose 0≠x∈N and set I=Ann⁡(x)⊂𝔪 for some maximal ideal 𝔪. Then A⁢x=A/I, so A⁢x⊗M≅M/I⁢M≠0. Because A⁢x injects into N, A⁢x⊗M injects into N⊗M, which must be nonzero.

2 implies 1:

Consider the sequence P→𝑓Q→𝑔R. Then because M is flat, ker⁡(g⊗idM)=ker⁡(g)⊗M and Im(f⊗idM)=Im(f)⊗M. If g∘f=0, then Im(g∘f)=0, which happens iff Im(g∘f)⊗M=0. Then Im((g⊗idM)∘(f⊗idM))=0.

If P⊗M→Q⊗M→R⊗M is exact, then P→Q→R is a complex. Finally, we need that ker⁡g=Imf. By flatness of M, we can tensor to find that ker⁡g/Imf⊗M=0 and then we see that ker⁡g/Imf=0. □

Corollary 2.2.10.

Let A→B be a local homomorphism and let M be a finitely-generated B=module. Then M if flat over A if and only if M is faithfully flat over A.

Proof.

Clearly faithfully flat implies flat. Then we need to show that M≠𝔪A.M. By Nakayama’s lemma, we know that M⊗kB≠0, so 𝔪𝔅⁢M≠M. In particular, 𝔪A⁢M≠M. In particular, this implies that item 3 of the previous theorem holds. Thus M is faithfully flat over A. □

Remark 2.2.11.

This also shows that flat and faithfully flat are equivalent over local rings. Alternatively, we can use the equivalence of flat and free.

Remark 2.2.12.

Faithful flatness is transitive. In addition, if A→B is a morphism of rings, and M is faithfully flat over A, then M⊗AB is faithfully flat over B.

Proposition 2.2.13.

Let M be a faithfully flat B-module which is faithfully flat over A. Then B is faithfully flat over A.

Proof.

Let N be an A-module. Then (B⊗AN)⊗BM=M⊗AN≠0 if N≠0. This implies that B⊗AN is nonzero. Now it suffices to show that B is flat over A.

Let (S) be an exact sequence of A-modules. Then if we consider ((S)⊗AM)=(S)⊗AM, this is exact by flatness of M over A. By faithful flatness of M over B, this implies that (S)⊗AB is exact. □

Proposition 2.2.14.

Let A→B be faithfully flat. Then

  1. 1.

    For any A-module N, the map N→N⊗AB is injective;

  2. 2.

    If I⊂A is an ideal, then I⁢B∩A=I;

  3. 3.

    The map Spec⁡B→Spec⁡A is surjective.

Proof.
  1. 1.

    Let 0≠x∈N. Then A⁢x⊗B↪N⊗B. Because B is faithfully flat, A⁢x⊗B≠0.

  2. 2.

    Recall that B/I⁢B=B⊗A/I. Then the map A/I→B/I⁢B is injective. Therefore we have a map A→B/I⁢B which has kernel I=I⁢B∩A.

  3. 3.

    Choose 𝔭∈Spec⁡A. Then B𝔭=B⊗AA𝔭 and A𝔭→B𝔭 is faithfully flat by base change. This means that 𝔭⁢B𝔭⊊B𝔭. Thus if we choose 𝔪 to be a maximal ideal of B𝔭 containing 𝔭⁢B𝔭, we see that 𝔪∩A𝔭⊃𝔭⁢A𝔭 and thus they are equal by maximality of 𝔭⁢A𝔭. Then if we choose 𝔓=𝔪∩B, we see that

    Thus the image of 𝔓 is 𝔭. □

Theorem 2.2.15.

Let φ:A→B be a map of rings. The following are equivalent:

  1. 1.

    The map φ is faithfully flat.

  2. 2.

    The map φ is flat and Spec⁡B→Spec⁡A is surjective.

  3. 3.

    The map φ is flat and for all maximal ideals 𝔪 of A, there exists some maximal ideal 𝔪′ of B such that φ−1⁢(𝔪′)=𝔪.

Proof.
1 implies 2:

This is the previous proposition.

2 implies 3:

Choose a maximal ideal 𝔪⊂A. Then there exists 𝔓∈Spec⁡B such that φ−1⁢(𝔓)=𝔪. But then if 𝔪′ is any maximal ideal containing 𝔓, we see that φ−1⁢(𝔪′)=φ−1⁢(𝔓)=𝔪 by maximality of 𝔪.

3 implies 1:

We want to prove that B≠𝔪⁢B for any maximal ideal 𝔪 of A. Then there exists 𝔪′ such that φ−1⁢(𝔪′)=𝔪. But then B⊋𝔪′⊃𝔪⁢B. □

Proposition 2.2.16 (Descent).

Let A→B be faithfully flat and M be an A-module. Then

  1. 1.

    M is flat (resp. faithfully flat) if and only if M⊗AB is B-flat (resp. B-faithfully flat).

  2. 2.

    Assume A is a local ring and M is finitely-generated. Then M is free if and only if M⊗AB is B-free.

Proof.
  1. 1.

    Let (S) be an exact sequence. Then (S)⊗AB is exact, so S⊗AB⊗M⊗AB=(S⊗AM)⊗AB is exact. By faithful flatness of B, (S)⊗AM is exact. Now if N≠0 is another A-module, we know that M(B)⊗N(B)≠0, but this is the same as (M⊗AN)(B), so M⊗AN is nonzero.

  2. 2.

    Assume that A is local. Then suppose M⊗AB is free. Therefore M⊗AB is faithfully flat. But then, M is faithfully flat over A, which means that M is free because M is finitely generated. □

Exercise 2.2.17.

Let A⊂B be integral domains. Assume that A and B have the same field of fractions. Prove that A↪B is faithfully flat if and only if A=B.

2.3 More on Integral Dependence

Recall Proposition 1.4.11.

Corollary 2.3.1.

Let x1,…,xn∈B. If each xi is integral over A, then A⁡[x1,…,xn] is a finitely-generated A-module.

Corollary 2.3.2.

Let C⊂B be the set of integral elements over A. Then C is a subring of B.

Proof.

Note that x+y,x⁢y∈A⁡[x]⁢[y]∈A⁡[x,y], which is a finitely-generated A module. Therefore they are integral over A. □

Remark 2.3.3.

The ring C is not necessarily finitely-generated over A. For an example, choose ℤ⁢ℤ¯⊂ℚ¯.

Definition 2.3.4.

Let A⊂B. Then we say that B is integral over A if all elements of B are integral over A.

Corollary 2.3.5.

Let A⊂B⊂C be extensions of rings. Then if B is integral over A and C is integral over B, then C is integral over A.

Proof of this is left to the reader.

Definition 2.3.6.

Let A be an integral domain. We say that A is integrally closed if for all x∈K=Frac⁡A, then x is integral over A if and only if x∈A.

Definition 2.3.7.

Assume that A⊂B is an inclusion of rings. Then the integral closure of A inside B is the set of all elements of B that are integral over A.

Example 2.3.8.

A typical example of this situation is when A is a domain, K is its fraction field, and L/K is a field extension. Then we can consider the integral closure B of A inside L. In number theory, if K is a number field, we define its ring of integers 𝒪K to be the integral closure of ℤ in K.

Exercise 2.3.9.

If K=ℚ⁡(ζp), prove that 𝒪K=ℤ⁡[ζp].

Exercise 2.3.10.

Let B be an integral domain and A⊂B. Prove that the integral closure of A inside B is integrally closed.

Lemma 2.3.11.

Let B be a domain that is integral over A. Then A is a field if and only if B is a field.

Proof.

Assume that A is a field. Now choose 0≠x∈B. But then we know that

for some ai∈A and a0≠0. But then we have

and so x−1∈B.

Now assume that B is a field. Choose 0≠x⁢i⁢n⁢A. Then x−1∈B. This means that x−1 is integral over A, which means that

for some ai∈A,a0≠0. Then if we multiply by an−1, we obtain

which means x−1∈A. □

Corollary 2.3.12.

Let A⊂B and B be integral over A. Let 𝔓∈Spec⁡B and 𝔭=A∩𝔓. Then 𝔭 is maximal if and only if 𝔓 is maximal.

Proof.

Note that B/𝔓 is integral over A/𝔭. Then we apply the lemma to A/𝔭⊂B/𝔓. □

We can refine this into going-up and going-down. Let ϕ:A→B be a morphism of rings and let ϕ∗:Spec⁡B→Spec⁡A be the induced map of spaces.

Definition 2.3.13 (Going-up).

A ring homomorphism ϕ satisfies the Going-up property if the following holds:

Let 𝔭⊂𝔭′ be prime ideals of A and suppose that ϕ∗⁢(𝔓)=𝔭. Then there exists 𝔓′⊃𝔓 such that ϕ∗⁢(𝔓′)=𝔭′.

Definition 2.3.14 (Going-down).

A ring homomorphism ϕ satisfies the Going-down property if the following holds:

Let 𝔭⊂𝔭′ be prime ideals of A. Then let 𝔓′∈Spec⁡B satisfy ϕ∗⁢(𝔓′)=𝔭′. Then there exists 𝔓⊂𝔓′ with ϕ∗⁢(𝔓)=𝔭.

Lemma 2.3.15.

The going-down property is equivalent to the following:

For all 𝔭∈Spec⁡A and 𝔓 a minimal prime ideal of B containing 𝔭⁢B, we have 𝔓∩A=𝔭.

Proof.

First suppose that going-down holds. Then choose 𝔓 be a minimal prime containing 𝔭⁢B. Then 𝔭′=ϕ−1⁢(𝔓)⊃𝔭. If 𝔭′≠𝔭, then there exists 𝔓0⊂𝔓 such that φ−1⁢(𝔓0)=𝔓, which contradicts minimality.

Now suppose the other condition holds. Suppose 𝔓′ goes to 𝔭′⊃𝔭. Then we know that 𝔭⁢B⊂𝔭′⁢B⊂𝔓′. If we fix 𝔓0 to be the minimal prime containing 𝔭⁢B, then we see that 𝔓0∩A=𝔭. □

Theorem 2.3.16.

If ϕ:A→B is flat, then going-down holds.

Proof.

Fix 𝔭⊂𝔭′ and let 𝔓′ lie over 𝔭′. Then we know that B𝔓′ is flat over 𝔄𝔭′. Because A𝔭′ is local and the map A𝔭′→𝔅𝔓′ is local, it is faithfully flat. This implies that the map Spec⁡B𝔓′→Spec⁡A𝔭′ is surjective, so there exists 𝔓1∈Spec⁡B𝔓′ such that ϕ−1⁢(𝔓1)=𝔭⁢A𝔭′.

Now set 𝔓≔𝔓1∩B. Then we see that

We will see consequences of this result in algebraic geometry.

We will now consider integral ring extensions A⊂B.

Theorem 2.3.17 (Cohen-Seidenberg).

Suppose A⊂B is an integral extension. Then the following hold:

  1. 1.

    The map Spec⁡B→Spec⁡A is surjective.

  2. 2.

    There are no inclusion relations between the prime ideals of B which are above a fixed prime ideal of A.

  3. 3.

    Going-up holds for A⊂B.

  4. 4.

    If A is local with maximal ideal 𝔪, then the prime ideals of B lying over 𝔪 are precisely the maximal ideals of B.

  5. 5.

    Assume further that A and B are integral domains and that A is integrally closed. Then going-down holds for A⊂B.

  6. 6.

    If B is the integral closure of A in a normal extension of field L of K≔Frac⁡A, then any two prime ideals of B lying over the same prime ideal of A are conjugate by an element of Aut⁡(L/K).

Proof.

We prove 4, then 1, 2, and 3, then 6, and finally 5.

  1. 1.

    Let 𝔭∈Spec⁡A. Then B𝔭=B⊗AA𝔭 is integral over A𝔭. Applying 4, we obtain the desired result.

  2. 2.

    Consider B𝔭 again. By 4, because ideals lying over 𝔭 are maximal, there cannot be inclusion relations between them.

  3. 3.

    Let 𝔭⊂𝔭′ and 𝔓∈Spec⁡B lying over 𝔭. Then B/𝔓 is integral over A/𝔭. By 1, we know that Spec⁡B/𝔓→Spec⁡A/𝔭 is surjective. Thus there exists 𝔓¯′∈Spec⁡B/𝔓 lying over 𝔭¯′=𝔭′/𝔭. Then we know that 𝔓¯′=𝔓′/𝔓 for some prime ideal 𝔓′ of B, and this is the ideal we are looking for.

  4. 4.

    This is a consequence of Lemma 2.3.15.

  5. 5.

    Write L=Frac⁡B⊃K=Frac⁡A. Then let L1 be the normal closure of L/K. Then let 𝔭⊂𝔭′ in A and 𝔓′ in B lie over 𝔭′. Then let 𝔓1⊂𝔓1′ in B1 the integral closure of A in L1. These exist thanks to 1 and 3.

    Let 𝔓1′′ in B1 such that 𝔓1′′∩B=𝔓′. Then there exists σ such that 𝔓1′′=σ⁡(𝔓1′) because both ideals are above 𝔭′. Then we can choose

    We need to show that 𝔓∩A=𝔭. But this is simply

  6. 6.

    We know that A is integrally closed in K. Then let L/K be a finite Galois (we can always reduce to this case) extension and B the integral closure of A in L. Then let 𝔓,𝔓′∈Spec⁡B lie above 𝔭∈Spec⁡A. We will show there exists σ∈Gal⁡(L/K) such that σ⁡(𝔓)=𝔓′.

    Suppose that no such σ exists. Then for all σ∈Gal⁡(L/K), 𝔓′≠σ⁡(𝔓). In particular, 𝔓′⊄σ⁡(𝔓). Then there exists x∈𝔓′ which is not in any σ⁡(𝔓) then we see that

    is integral over A, so y∈A. Also, y∉𝔓 because x∉σ⁡(𝔓), so x∈𝔓′ and thus y∈𝔓′, so y∈𝔭⊂𝔓. This gives a contradiction. □

Corollary 2.3.18.

Assume that B is integral over A.

  1. 1.

    If 𝔓0⊊𝔓1⊊⋯⊊𝔓r is a chain of prime ideals of B, the the 𝔭i≔𝔓i∩A for a chain of prime ideals of A.

  2. 2.

    If p0⊊⋯⊊𝔭r is a chain of prime ideals of A, then there exists a chain 𝔓0⊊⋯⊊𝔓r of prime ideals of B above it.

  3. 3.

    If A is integrally closed and B is a domain, then for any chain 𝔭0⊊⋯⊊𝔭r and 𝔓r∈Spec⁡B above 𝔭r, there exists a chain 𝔓0⊊⋯⊊𝔓r above the chain in A.

Proof.

The proof is clear and left to the reader. □

Definition 2.3.19.

Let 𝔭∈Spec⁡A. Then define the height of 𝔭 by

Then define the dimension of A by

Corollary 2.3.20.

Let A⊂B be an integral extension. Then

  1. 1.

    Suppose 𝔓∈Spec⁡B lies above 𝔭∈Spec⁡A. Then ht⁡(𝔓)≤ht⁡(𝔭).

  2. 2.

    dimA=dimB.

  3. 3.

    If A is integrally closed and B is a domain, then we have ht⁡(𝔓)=ht⁡(𝔭).

Proof.

This is an immediate consequence of the previous corollary. □

2.4 Associated Primes

Let M be an A-module and 𝔭∈Spec⁡A.

Definition 2.4.1.

We say that 𝔭 is an associated prime of M if one of the two following equivalent conditions hold.

  1. 1.

    There exists x∈M such that AnnA⁢(x)=𝔭;

  2. 2.

    There is an injection A/𝔭↪M.

We will denote the set of associated primes using the unfortunate notation AssA⁢(M). Then the set of primes 𝔭 such that M𝔭≠0 will be denoted SuppA⁢(M).

Proposition 2.4.2.

Let 𝔭 be a maximal element of {Ann⁡(x)∣x∈M,x≠0}. Then 𝔭∈AssA⁡(M).

Proof.

We will show that such a maximal element is actually a prime ideal. Suppose a⁢b∈𝔭. Then 𝔭=Ann⁡(x) for some nonzero x, so b.x≠0. Then Ann⁡(x)⊂Ann⁡(b⁢x)≠A. By maximality, Ann⁡(x)=Ann⁡(b⁢x). Because a⁢b⁢x=0, then a∈Ann⁡(b⁢x)=𝔭. □

Corollary 2.4.3.

Let A be Noetherian.

  1. 1.

    M is nonzero if and only if AssA⁡(M) is nonempty.

  2. 2.

    The set of zero divisors for M is the union of the associated primes of M.

Proof.
  1. 1.

    If there is some associated prime, then clearly M≠0. In the other direction, the set of annihilators has a maximal element because A is Noetherian, so there must be an associated prime.

  2. 2.

    Let a∈Ann⁡(x) for some nonzero x∈M. Then Ann⁡(x)⊂𝔭 is contained in some associated prime (because it is contained in some maximal element), and thus every zero divisor is contained in an associated prime. The other direction is obvious. □

Lemma 2.4.4.

Let S⊂A be a multiplicative set and M an A-module. Then

Proof.

Let 𝔭∈AssA⁡(S−1⁢M). Then 𝔭=AnnA⁡x1 for some x∈M, so 𝔭∩S must be empty. Next, we see that the set {AnnA⁡(s⁢x)∣s∈S} contains some maximal element 𝔪 because A is Noetherian. But then 𝔪=AnnA⁡(s0⋅x)=𝔭.

On the other hand, if a∈𝔭, then a⁢x1=0, which means a⁢s⁢x=0 for some s∈S. Then a∈AnnA⁡(s⁢x)⊂Ann⁡(s0⁢s⁢x)=Ann⁡(s0⁢x). Thus 𝔭⊂Ann⁡(s0⁢x). Thus we have shown that

The other inclusion is clear. □

Theorem 2.4.5.

Let A be Noetherian and M and A-module. Then AssA⁡(M)⊂SuppA⁡(M) and any minimal element of SuppA⁡(M) is inside AssA⁡(M).

Proof.

Let 𝔭∈AssA⁡(M). Then A/𝔭 injects in M, so we have an injection A𝔭/𝔭⁢A𝔭↪M𝔭. Thus 𝔭∈SuppA⁡(M).

Now choose a minimal 𝔭∈SuppA⁡(M). Thus M𝔭 is nontrivial, so there exists a prime ideal 𝔮⊂𝔭 such that 𝔮⁢A𝔭∈AssA𝔭⁡(M𝔭). Thus M𝔮=(M𝔭)𝔮⁢A𝔭 is nonzero, so 𝔮∈Supp⁡(M). By minimality, 𝔮=𝔭 and thus 𝔭⁢A𝔭∈AssA𝔭⁡(M𝔭). Therefore 𝔭∈AssA⁡(M). □

Definition 2.4.6.

If 𝔭∈AssA⁡(M), then 𝔭 is not necessarily minimal in the support of M. Then such a prime is called an embedded prime.

Proposition 2.4.7.

Let A be Noetherian and M a finitely-generated A-module. Then

  1. 1.

    There exists a chain

    such that Mi/Mi−1≅A/𝔭i for i=1,…,n and 𝔭i∈SuppA⁡(M).

  2. 2.

    Given such a sequence, we have AssA⁡(M)⊂{𝔭1,…,𝔭n}. In particular this set is finite.

Proof.
  1. 1.

    Suppose M≠0 Then choose 𝔭1∈AssA⁡(M) and let M1 be the image of A/𝔭1 in M. Then if M/M1 is nonzero, choose 𝔭1∈AssA⁡(M/M1) and M2 defined analogously to M1. This gives a sequence of submodules of M such that A/𝔭i≅Mi/Mi−1. Because M is Noetherian, this sequence becomes stationary. Thus there exists n such that Mn=M.

  2. 2.

    This is a consequence of the next lemma.

□

Remark 2.4.8.

In general the support of a module is not finite.

Lemma 2.4.9.

Assume we have an exact sequence of modules 0→M′→M→M′′. Then Ass⁡(M)⊂Ass⁡(M′)∪Ass⁡(M′′).

Proof.

If 𝔭∈Ass⁡(M), there exists N⊂M such that N≅A/𝔭. Then if N∩M′=0, N↪M′′ and 𝔭∈Ass⁡(M′′). If the intersection is nonzero, then there exists some nonzero x∈N∩M′ such that AnnA⁡(x)=p because A/𝔭 is a domain. Thus 𝔭∈Ass⁡(M′). □

Definition 2.4.10.

We say that M is coprimary if AssA⁡(M)={𝔭}.

Definition 2.4.11.

Let N⊂M. Then we say that N is 𝔭-primary if AssA⁡(M/N)={𝔭}. Alternatively, we say that N belongs to 𝔭.

Lemma 2.4.12.

A module M is coprimary if and only if M is nonzero and any zero divisor for M is locally nilpotent (for all x∈M, there exists n>0 such that an.x=0).

Proof.

Suppose that M is coprimary. Now suppose that a∈𝔭 and x∈M. Then Ass⁡(A⁢x)={𝔭}, so 𝔭 is minimal in the support of Ax, which is V⁡(Ann⁡(x)). Therefore, 𝔭=Ann⁡(x). Thus for a∈𝔭, an∈Ann⁡(x).

In the other direction, let 𝔭 be the set of locally nilpotent elements with respect to M. This is clearly an ideal of A. Then let 𝔮∈AssA⁡(M). Then x∈M, so 𝔮=Ann⁡(x). Therefore 𝔭⊂𝔮 because 𝔮 is a prime ideal. However, 𝔮 is contained in the set of zero divisors, which is precisely 𝔭. □

Remark 2.4.13.

Let I⊂A be an ideal. Then AssA⁡(A/I)={𝔭} if and only if the zero divisors of A/I are locally nilpotent. This is equivalent to I being primary.

Lemma 2.4.14.
  1. 1.

    Let Q1,Q2⊂M be 𝔭-primary submodules. Then Q1∩Q2 is 𝔭-primary.

  2. 2.

    Let N=Q1∩⋯∩Qr be an irredundant decomposition (i.e. Qi is 𝔭i-primary) for distinct 𝔭i. Then AssA⁡(M/N)={𝔭1,…,𝔭r}.

Proof.
  1. 1.

    Note that M/Q1∩Q2 injects in M/Q1⊕M/Q2. The desired result follows from the previous lemma.

  2. 2.

    First, note that M/N↪⨁M/Qi. Then suppose Ass⁡(M/N)={𝔭1,…,𝔭r}. Then we have an injection

    and thus Ass⁡((Q1∩⋯∩Qr)/N) is contained in Ass⁡(M/N). By the exact sequence

    we see that Ass⁡(Q2,∩⋯∩Qr/N)={𝔭1}. □

Theorem 2.4.15.

Let M be a module over a Noetherian ring A. Then for all 𝔭∈Ass⁡(M), there exists a 𝔭-primary submodule Q⁡(𝔭)⊂M such that

Proof.

Fix 𝔭∈Ass⁡(M). Consider the set

This set is nonempty because 0∈𝒮𝔭. Next, if Nλ∈𝒮𝔭 is a chain, then the module N=⋃Nλ is a submodule of M. In addition, Ass⁡(N)⊂⋃Ass⁡(Nλ). This implies that 𝒮𝔭 contains a maximal element by Zorn’s lemma. Choose such a maximal element Q⁡(𝔭).

We will show that M/Q⁡(𝔭) is coprimary. By the exact sequence

if 𝔭′∈Ass⁡(M/Q⁡(𝔭)), then 𝔭′=𝔭 because otherwise A/𝔭′ would inject in M/Q⁡(𝔭) as Q′/Q⁡(𝔭). Then Ass⁡(Q′)⊂Ass⁡(Q⁡(𝔭))∪Ass⁡(Q′/Q⁡(𝔭)), so Q′⊋Q⁡(𝔭), contradicting minimality. Thus Ass⁡(M/Q⁡(𝔭))={𝔭}.

The second part of the claim follows immediately from the fact that Ass⁡(⋂Q⁡(𝔭))=⋂Ass⁡(Q⁡(𝔭))=∅. □

Corollary 2.4.16.

Let M be an A-module of finite type. Then any N⊂M has a primary decomposition

such that

  1. 1.

    The Qi are 𝔭i-primary;

  2. 2.

    No Qi can be omitted;

  3. 3.

    This decomposition is irredundant: Ass⁡(M/N)={𝔭1,…,𝔭r}.

Proof.

Apply the previous theorem to M/N. Because M/N is of finite type, Ass⁡(M/N) is finite. Then use the previous lemma. □

Exercise 2.4.17.

Let A→𝜑B be a morphism of rings and let M be a B-module. Then prove that

where φ∗:Spec⁡B→Spec⁡A is the induced map of spaces.

3 Dimension Theory

3.1 Graded Rings and Modules

Let A=⨁n≥0An be a graded ring. This means that An⋅Am⊂An+m. Then an A-module M is a graded module if

such that An⋅Mm⊂Mn+m. We will call Mm the homogeneous elements of degree m on M.

Now let N⊂M be a submodule. We say that N is a graded submodule if N=⨁N∩Mm. N is also called homogeneous. A homogeneous element of M is an element of some Mm. Being a graded submodule is the same as every element being a sum of homogeneous elements.

Lemma 3.1.1.

The following are equivalent:

  1. 1.

    N is a homogeneous submodule.

  2. 2.

    N is generated by homogeneous elements.

  3. 3.

    If x=xr+⋯+xn∈N for xi∈Mi, then for all i, xi∈N.

Moreover, if N⊂M is homogeneous, then so is M/N, and

Proof.

The proof is left as an exercise to the reader. □

Example 3.1.2.

Let k be any ring. Then the ring A=k⁡[x1,…,xr] is a graded ring where the grading is by the degree of each monomial. In particular, A0=k. Then an ideal I⊂A is graded if I=⨁nIn where In=I∩An. In addition, A/I is a graded ring.

Proposition 3.1.3.

Let A be a Noetherian graded ring and M a graded A-module. Then

  1. 1.

    If 𝔭∈Ass⁡(M), then 𝔭 is a graded ideal of A and there exists a homogeneous x∈M such that 𝔭=Ann⁡(x).

  2. 2.

    One can choose a 𝔭-primary graded submodule Q⁡(𝔭) such that 0=⋂𝔭∈Ass⁡(M)Q⁡(𝔭).

Proof.

Let x∈M and set 𝔭=Ann⁡(M). Then write x=xe+xe−1+⋯+x0. Then for f∈𝔭, write f=fr+⋯+fs. If f⁢x=0, then we can write

and deduce that 0=fr⁢xe=fr2⁢xe−1=⋯. Then fre∈𝔭, so fr∈𝔭. By induction, all fi∈𝔬, so 𝔭 is graded.

The proof of the second part is simply the following lemma. □

Lemma 3.1.4.

Let 𝔭 be a graded prime ideal and Q⊂M such that Q is 𝔭-primary. Let Q′⊂Q be the submodule of Q generated by the homogeneous elements of Q. Then Q′ is 𝔭-primary.

Proof.

This will be proved later. □

We will now discuss filtrations of rings. A filtration is a sequence of subgroups

such that Jn⋅Jm=Jn+m. If we set

then A′ is a graded ring.

The basic example is Jm=Im for some fixed ideal I⊂A. in this case, the filtration is called the I-adic filtration.

Lemma 3.1.5.

Let A be a Noetherian ring and set I⊂A. Then

is a Noetherian graded ring.

Proof.

Because I is finitely-generated, then I/I2 is a finitely-generated A/I-module. Thus grI⁡(A) is a finitely-generated A/I-algebra. If x1,…,xr is a set of generators of I, then

is surjective, so because A/I is Noetherian, so is A/I⁡[x1,…,xr] and thus so is grI⁡A. □

Let A be an Artinian ring and B=A⁡[x1,…,xr] be a graded ring. Then let M be a finitely-generated graded B-module. Each graded piece Mn is an A-module, so write FM⁢(n)=ℓA⁢(Mn). Because M is finitely generated, we have a map

Here, B⁡(di)=B as a B-module with the gradation B⁢(d)n=Bn−d. Thus M is generated by homogeneous elements xdi of degree di. This gives us the map

and thus ℓA⁢(Mm)≤∑i=1rℓA⁢(Bm−di). But then Bm is a free A-module, and thus

Theorem 3.1.6.

Let A,B,M be as above. Then there is a polynomial fM⁢(x)∈ℚ⁡[x] such that

for n≫0. This is called the Hilbert-Samuel polynomial for M. The degree of this polynomial will give the first definition for the dimension of M.

Proof.

Say that M satisfies the property P⁡(M) if there exists f∈ℚ⁡(x) such that ℓ⁡(Mn)=f⁡(n) for n≫0.

  1. 1.

    First, we will show that if N1,N2⊂M and P⁡(M/N1),P⁡(M/N)⁢2 hold, then P⁡(M/N1∩N2) holds.

  2. 2.

    Second, if N is irreducible, then P⁡(M/N) holds.

If we prove these two things, then we simply use a primary decomposition of 0⊂M. Then the second statement implies that P⁡(M/Ni) holds and then we simply repeatedly apply the first to see that P⁡(M) holds.

  1. 1.

    Suppose N=N1∩N2 with N,N1,N2 graded. We then have an exact sequence

    and we know that N1/(N1∩N2)=(N1+N2)/N1 is graded. Then we see that FM/(N1∩N2)=FM/N+F(N1+N2)/N1, so we only need to prove that FN1+N2/N1 exists. But then FN1+N2/N1=FM/N2−FM/N1+N2 and because P⁡(M/N1+N2) holds, so does P⁡(M/N1∩N2).

  2. 2.

    Let N be irreducible. We know that M′=M/N is coprimary, so N is 𝔭-primary for some prime ideal 𝔭⊂B. Write I=(x1,…,xm). If I⊂𝔭, then Mm′=0 for n≫0. Indeed, if d is the maximal degree of a system of generators of M′, then Mn+d′=In⋅Md′. On the other hand, because M′ is 𝔭-primary, then elements in 𝔭 are locally nilpotent. Thus there exists k≫0 such that pk⋅Md′=0 and thus Mn+d′=0 for n≥k. Thus FM/N exists and is identically zero.

    In the second case, I⊄𝔭. Then suppose that x1∉𝔭. Thus x1 is not a zero divisor for M′. Thus, we have an exact sequence

    which then gives

    when restricting to a single graded piece. Thus N⊊N+x1⁢M. This implies that fM/N+x1⁢M exists because above, we proved that if P⁡(M/N′) holds for any N′⊋N implies that P⁡(M/N) holds. Then for n≥n0, we have ℓ⁡((M/N)m)−ℓ⁡((M/N)n−1)=fM/N+x1⁢M⁢(n). This implies that

    Then f⁡(n)+⋯+f⁡(n0)=g⁡(n) for some polynomial g of degree deg⁡f+1 and then fM/N=g+ℓ⁡((M/N)n0). □

Now let A⊃I and M be an A-module with filtration

We say that the filtration is

  1. 1.

    I-admissible if I⁢Mn⊂Mn+1 for all n≫0;

  2. 2.

    I-acic if I⁢Mn=Mn+1 for all n≥0;

  3. 3.

    essentially I-adic if I⁢Mn=Mn+1 for n≫0.

Remark 3.1.7.

A filtration on M defines a topology on M so that M is a topological group. Here, a system of neighborhoods of 0 is (Mn)n≫0. If ⋂Mn=0, then the topology is Hausdorff. If the filtration is essentially I-adic, then the topology is called the I-adic topology.

Lemma 3.1.8.

Let A and I be as before. Let M be an A-module with an admissible filtration. Let A′=⨁n=1∞In⁢xn⊂A⁡[x] and

  1. 1.

    M′ is a A′-module.

  2. 2.

    The filtration is essentially I-adic if and only if M′ is a finitely-generate A′-module.

Proof.
  1. 1.

    This is trivial.

  2. 2.

    Note that M′ is a graded A′-module. If M′ is finitely generated, then write M′=A′⁢m1+⋯+A′⁢mr. Then we see that Mn′=I⁢x⁢Mn−1′ for n>max⁡{deg⁡mi}. Thus Mn is essentially I-adic.

    Conversely, if Mn=In−n0⁢Mn0 for n≥n0, then, then it is clear that M′ is generated by Mn0⁢xn0+⋯+M1⁢x+M0 and is thus finitely generated. □

Theorem 3.1.9 (Artin-Rees).

Let A be a Noetherian ring and I⊂A. Then let M be a finitely-generated A-module and N⊂M be a submodule. Then there exists r>0 such that

for all n≥r.

Proof.

Let Mn=In⁢M be the I-adic filtration. Then Nn=In⁢M∩N is I-admissible. Then both N′⊂M′ are both A′-modules. We know that A′ is Noetherian, so because Mn, M′ is finitely generated. Thus N′ is also Noetherian, so it is also finitely-generated. This implies that Nm is essentially I-adic, as desired. □

Remark 3.1.10.

This theorem is saying that the filtration In⁢M∩N is essentially I-adic.

Corollary 3.1.11 (Krull Intersection Theorem).

Let A,I,M as above.

  1. 1.

    If N=⋂n=0∞In⁢M, then I⁢N=N.

  2. 2.

    If I⊂rad⁡(A), then ⋂n=0∞In⁢M=0.

Proof.
  1. 1.

    Note that N⊂M. Then apply the Artin-Rees theorem to N=In⁢M∩N.

  2. 2.

    Apply Nakayama’s lemma. □

Corollary 3.1.12.

Let A be a Noetherian domain and let I⊂A be a proper ideal. Then ⋂nIn=0.

Proof.

Let N=⋂n=0∞In. By the previous corollary, I⁢N=N. Then N is finitely generated because A is Noetherian. Thus there exists x∈I such that (1+x)⁢N=0, which implies that N=0 because A is a domain. □

Exercise 3.1.13.

Let A be Noetherian and M a finitely-generated A-module. Then let I,J be generated by M-regular elements. Then there exists r>0 such that (In⁢M:J)=In−r⁢(Ir⁢M:J). Here, (N:J)={m∈M∣J⁢m⊂N}.

3.2 Other Notions of Dimension

Let A be a ring. Then we define the Krull dimension

Then for any ideal I⊂A, define the height of I to be

Proposition 3.2.1.

For any ideal I, we have dim(A/I)+ht⁡(I)≤dimA.

Proof.

Consider a chain

where d′=dim(A/I). Then we see that dimA≥d′+ht⁡(𝔭0)≥d′+ht⁡(I), as desired. □

Then if M is an A-module, define dimM=dim(A/AnnA⁡M).

Proposition 3.2.2.

Assume that A is Noetherian and M is a finite A-module. Then the following are equivalent:

  1. 1.

    M is of finite length.

  2. 2.

    A/AnnA⁡M is Artinian.

  3. 3.

    dimM=0.

Proof.

Clearly conditions 2 and 3 are equivalent. Then M is a quotient of (A/AnnA⁡M)r, so 2 implies 1. Thus we need to prove that 1 implies 3.

Assume that ℓ⁡(M)<∞. If we write A′=A/AnnA⁡(M), then M is a finite A′-module. If dimA′>0, then there exists 𝔭⊂A′ that is minimal but not maximal. Then because AnnA′⁡(M)=0, we have 𝔭∈V⁡((0))=Supp⁡(M). But then 𝔭∈AssA′⁡(M) and thus we have an embedding A′/𝔭↪M. But then dimM′/𝔭>0, so ℓ⁡(A′/𝔭)=∞ and thus ℓ⁡(M)=∞. □

Now let A be a semilocal ring. Let 𝔪=Rad⁡A. Then an ideal I⊂A is called an ideal of definition of A if there exists s>0 such that 𝔪s⊂I⊂𝔪.

Remark 3.2.3.

I is an ideal of definition if and only if A/I is Artinian.

Let A∗=grI⁢(A)=⨁n=0∞In/In+1 be the graded ring with respect to the I-adic filtration and let M∗=grI⁢(M) be the corresponding graded A∗-module. Then if I=(x1,…,xi)⊂A, define

Then we have a map B↠A∗, so M∗ is a B-module. Now define χ⁡(M,I,n)≔ℓ⁡(M/In⁢M). If M is a finite A-module, then M/In⁢M is of finite length (because A/In is Artinian) and thus

Then if ℓ⁡(Is⁢M/Is+1⁢M) is a polynomial in s of degree at most r−1 for s≫0, then ℓ⁡(M/In⁢M) is a polynomial of degree at most r for n≫0.

Now if J is another ideal of definition, then there exists s such that Js⊂I and thus χ⁡(M,J,n⁢s)≥χ⁡(M,I,n). Therefore

and so the degree of χ⁡(M,I,n) is independent of I. Denote this degree by d⁡(M). We know that d⁡(M)≥r, which is the number of generators of I.

Lemma 3.2.4.

Assume we have an exact sequence of finite A-modules

Then d⁡(M)=max⁡{d⁡(M′),d⁡(M′′)} and χ⁡(M,I,n)−χ⁡(M′,I,n)−χ⁡(M′′,I,n) is a polynomial of degree strictly less that d⁡(M′′).

Proof.

For each n we have an exact sequence

Then Δ≔χ⁡(M,I,n)−χ⁡(M′′,I,n)=ℓ⁡(M/In⁢M)−ℓ⁡(M′′/In⁢M)=ℓ⁡(M′/In⁢M∩M′). By Artin-Rees, there exists r such that M′∩In⁢M=In−r⋅(M′∩Ir⁢M). But then

because χ⁡(M′,I,n)−χ′⁢(M′,I,n−r) has degree strictly less than d⁡(M′), and the desired result follows. □

Lemma 3.2.5.

Let A be a Noetherian semilocal ring. Then d⁡(A)≥dimA. In particular, dimA<∞.

Proof.

We will induct on d⁡(A). If d⁡(A)=0, then 𝔪n=𝔪n+1 for n≫0. By Nakayama, we see that 𝔪n=0, so ℓ⁡(A)<∞ and thus A is Artinian.

Assume that d⁡(A)>0 and dimA>0. Let

be a chain of prime ideals of length ℓ>0. Choose x∈𝔭ℓ−1∖𝔭ℓ. Then dim(A/𝔭+x⁢A)≥ℓ−1. Because we have the exact sequence

we have d⁡(A/𝔭)=max⁡(d⁡(A/𝔭),d⁡(A/𝔭+x⁢A)) and that χ⁡(A/𝔭+x⁢A,I,n) has degree less than d⁡(A/𝔭). Therefore

By induction, dim(A/𝔭+x⁢A)≤d⁡(A/𝔭+x⁢A) and thus ℓ−1≤d⁡(A/𝔭+x⁢A)≤d⁡(A)−1. This holds for any chain of ideals, so dimA≤d⁡(A). □

Corollary 3.2.6.

If A is Noetherian and 𝔭∈Spec⁡A, then 𝔥⁢𝔱⁢(𝔭)<∞.

Proof.

ht⁡(𝔭)=dimA𝔭<∞ by the previous lemma. □

Lemma 3.2.7.

Let A be a Noetherian semilocal ring and M a finite A-module. Choose x∈Rad⁡(A). Then d⁡(M)≥d⁡(M/x⁢M)≥d⁡(M)−1.

Proof.

Let I⊂A be an ideal of definition such that x∈I. Then

Then because x∈I, we see that In−1⁢M⊂(In⁢M:x), so

This implies that χ⁡(M/x⁢M,I,n)≥χ⁡(M,I,m)−χ⁡(M,I,n−1), so d⁡(M/x⁢M)≥d⁡(M)−1. □

Lemma 3.2.8.

Let A and M be as before. Let r=dimM>0. Then there exists x1,…,xr∈Rad⁡(A) such that ℓ⁡(M/x1⁢M+⋯+xr⁢M)<∞.

Proof.

Let 𝔭1,…,𝔭t be the minimal prime ideals containing AnnA⁡(M) such that dim(A/𝔭i)=r. Because r>0, then the 𝔭i are not maximal and therefore Rad⁡(A)⊄𝔭i. In particular, it is not contained in ⋃𝔭i. Choose x1∈Rad⁡(A)∖⋃𝔭i.

If 𝔮⊃Ann⁡(M/x1⁢X)⊃Ann⁡(M)+x1⁢A is prime and minimal, then 𝔮∉𝔭i because x1∉𝔭i. This implies dimA/𝔮≤r−1. By induction, we can then find x2,…,xr such that

where M¯=M/x1⁢M. □

Theorem 3.2.9.

Let A be semilocal and M a finite A-module. Then d⁡(M)=dimM is the smallest integer r such that there exists x1,…,xr∈Rad⁡(A) such that ℓ⁡(M/x1⁢M+⋯+xr⁢M)<∞.

Proof.

Choose x1,…,xr∈𝔪=Rad⁡(A). If ℓ⁡(M/x1⁢M+⋯+xr⁢M)<∞, then we know that d⁡(M/x1⁢M+⋯|xr⁢M)≥d⁡(M)−r. Then because M/x1⁢M+⋯+xr⁢M has finite length, its dimsnion is zero and thus r≥d⁡(M). Then let r0 be the smallest such integer. By the previous lemma, we deduce that dimM≥r0≥d⁡(M).

We will show that d⁡(M)≥dimM. Consider a sequence

such that Mi/Mi+1≅A/𝔭i for some prime ideals 𝔭i. Then Ass⁡(M)⊂{𝔭0,…,𝔭n}⊂Supp⁡M are the minimal primes containing AnnA⁡(M), so we see that

Remark 3.2.10.

If M=A, then d⁡(M) is the smallest integer r such that there exists x1,…,xr∈Rad⁡(A) such that (x1,…,xr) is an ideal of definition.

Corollary 3.2.11.

If A is Noetherian and I=(x1,…,xr)⊂A, then any minimal prime ideal 𝔭 containing I has height at most r. In particular, ht⁡(I)≤r.

Proof.

First, note that A𝔭/I⁢A𝔭 is Artinian because the image of 𝔭⁢A𝔭 is both maximal and minimal. Then ℓ⁡(A𝔭/x1⁢A𝔭+⋯+xr⁢A𝔭)<∞ and thus ht⁡(𝔭)=dim(A𝔭)≤r. □

Now let M be a finitely generated A-module. Define M̂=lim←M/In⁢M for an ideal of definition I. We call M̂ the I-adic completion of M.

Corollary 3.2.12.

dimM̂=dimM.

Proof.

We know that M̂/In⁢M̂=M/In⁢M. Thus the two modules have the same Hilbert-Samuel polynomial. □

Corollary 3.2.13.

Let A be Noetherian with 𝔭∈Spec⁡A. Let n be an integer. The following are equivalent:

  1. 1.

    ht⁡(𝔭)≤n.

  2. 2.

    There exits I generated by n elements such that 𝔭 is minimal in V⁡(I).

Proof.
1 implies 2

Suppose that ht⁡(𝔭)≤n. Then there exists an ideal of definition J of A𝔭 generated by n elements. If J=(x1s,…,xns), then I=(x1,…,xn)⊂𝔭 and 𝔭 is minimal containing I.

2 implies 1

Let I=(x1,…,xn) such that 𝔭⊃I is minimal. Therefore A𝔭/I⁢A𝔭 is Artinian, so it has finite length. Thus dimA𝔭≤n, as desired. □

Definition 3.2.14.

A system of parameters for M is a set of elements x1,…,xs∈Rad⁡(A) such that

  • •

    ℓ⁡(M/x1⁢M+⋯+xs⁢M)<∞;

  • •

    s=dimM.

Proposition 3.2.15.

Let x1,…,xr∈Rad⁡(A). Then dim(M/(x1,…,xi)⁢M)≥dimM−r and we have equality if and only if x1,…,xr belong to a system of parameters for M.

Proof.

By induction d⁡(M/x⁢M)≥d⁡(M)−1 for any x∈Rad⁡(A). Then we know that

Then assume that we have equality. Let y1,…,yp be a system of parameters for M/(x1,…,xr)⁢M. Then dim(M/(x1,…,xr)⁢M)=p=dimM−r. However, if M¯=M/(x1,…,xr), then

and thus y1,…,yp,x1,…,xr is a system of parameters for M.

Conversely suppose that x1,…,xr,y1,…,yp is a system of parameters for M. Then

but we have equality because

and so p≥d⁡(M/(x1,…,xr)⁢M). □

Now we turn to the case of local Noetherian rings A with maximal ideal 𝔪 and residue field k. Then if d=dimA, any ideal of definition has at least d generators. Then let x1,…,xd∈𝔪 such that ℓ⁡(A/(x1,…,xd))<∞. Thus I=(x1,…,xd) is an ideal of definition and (x1,…,xd) is a system of parameters of A.

Definition 3.2.16.

A local ring A is a regular local ring if there is a system of parameters generating the maximal ideal of A. Such a system is called a regular system of parameters.

Note that dimA≤dimk𝔪/𝔪2 and that A is regular if and only if dimA=dimk𝔪/𝔪2.

Proposition 3.2.17.

Let (A,𝔪,k) be a local Noetherian ring. Let (x1,…,xd) be a system of parameters of A. Then dim(A/(x1,…,xi)⁢A)=d−i and the image of (xi+1,…,xd) in A/(x1,…,xi)⁢A is a system of parameters of this quotient.

3.3 Dimension in the Relative Setting

Consider a morphism A→𝜑B. We have the pullback φ∗:Spec⁡B→Spec⁡A. For 𝔭∈Spec⁡A, we have a bijection between (φ∗)−1⁢(𝔭) and Spec⁡(B𝔭/𝔭⁢B𝔭). This latter ring is isomorphic to B⊗k⁡(𝔭), where k⁡(𝔭) is the residue field of A𝔭.

Theorem 3.3.1.

Let 𝔓∈Spec⁡B lie over 𝔭. Then

  1. 1.

    ht⁡(𝔓)≤ht⁡(𝔭)+ht⁡(𝔓/𝔭⁢𝔓). Equivalently, dimB𝔓≤dimA𝔭+dim(B𝔓/𝔭⁢B𝔓).

  2. 2.

    Equality holds is equivalent to the going-down property for φ and in particular if φ is flat.

  3. 3.

    If φ∗ is surjective and the going-down property holds, then dimB≥dimA and ht⁡(I)=ht⁡(I⁢B) for I⊂A.

Proof.
  1. 1.

    Set A=A𝔭,B=B𝔓. We need to prove that dimB≤dimA+dimB/𝔭⁢B, where 𝔭 is the maximal ideal of A. Write r=dimA and let x1,…,xr be a system of parameters for A. Then I=(x1,…,xr) is an ideal of definition, so 𝔭n⊂I⊂𝔭 for some n. Thus 𝔭n⁢B⊂I⁢B⊂𝔭⁢B and all of these ideals have the same nilradical. Therefore

    for some integer s. If y1,…,ys is a system of paramaters for B/I⁢B, then x1,…,xr,y1,…,ys generate an ideal of definition for B, so r+s≥dimB.

  2. 2.

    Let 𝔓=𝔓0⊋𝔓1⊋⋯⊋𝔓s be a chain of ideals of B/𝔭⁢B of length s=dimB/𝔭⁢B. Then for i=0,…,s we know 𝔭⊂φ∗⁢(𝔓i) and thus φ∗⁢(𝔓i)=𝔭 for all i. Now by the Going Down property we can find

    such that 𝔭i=φ−1⁢(𝔓s+i). Thus we have

    where r=dimA. This gives us the chain

    and thus dimB≥r+s.

  3. 3.

    The first inequality follows from 2. Note that dimB=dimA+dim(B/𝔭⁢B)≥dimA.

    To prove the equality, let 𝔓∈V⁡(I⁢B) be minimal such that ht⁡(𝔓)=ht⁡(I⁢B). Let 𝔭=φ∗⁢(𝔓). Then 𝔭⊃I and 𝔓/𝔭⁢B is minimal, so ht⁡(𝔓/𝔭⁢B)=0. This tells us that dimB𝔓=dimA𝔭 and thus ht⁡(𝔓)=ht⁡(𝔭). Thus ht⁡(𝔓)≥ht⁡(I).

    Conversely, let 𝔭⊃I be minimal with ht⁡(𝔭)=ht⁡(I). Let 𝔓∈Spec⁡B such that φ∗⁢(𝔓)=𝔭. Then 𝔓⊃𝔭⁢B⊃I⁢B and so we may suppose it is minimal for this property. Then we see that

    as desired. □

Corollary 3.3.2.

Let B⊃A be Noetherian rings such that B is integral over A.

  1. 1.

    dimA=dimB;

  2. 2.

    For all 𝔓∈Spec⁡B, ht⁡(𝔓)≤ht⁡(𝔓∩A).

  3. 3.

    If the going-down property holds, then for any ideal J⊂B, have ht⁡(J)=ht⁡(J∩A).

Proof.

The proof of this is left as an exercise to the reader. □

Exercise 3.3.3.

Let A→ϕB be a morphism of rings and assume that going-down holds for ϕ. Let 𝔭⊃𝔮 be prime ideals of A. Prove that dim(B⊗k⁡(𝔭))≥dim(B⊗k⁡(𝔮)).

Now we will consider finitely generated extensions of rings. Here B will be a finitely-generated A-algebra.

Theorem 3.3.4.

Let A be Noetherian. Then dimA⁡[X]=dimA+1.

Proof.

Let 𝔭∈Spec⁡A and let 𝔓∈Spec⁡B such that 𝔓∩A=𝔭. Choose 𝔓 to be maximal for this property. We will show that ht⁡(𝔓/𝔭⁢B)=1. After localization at 𝔭, we may assume that 𝔭 is maximal and A is local. Then B/𝔭⁢B=A/𝔭⁡[X], and A/𝔭 is a field. Thus B/𝔭⁢B is a PID, so 𝔓/𝔭⁢B is a nonzero principal ideal, so it must have height exactly equal to 1.

Previous we have seen that because B is flat, dimB𝔓=dimA𝔭+1, and thus ht⁢𝔓=ht⁡(𝔭)+1, and we obtain the desired result. □

Corollary 3.3.5.
  1. 1.

    dimA⁡[x1,…,xm]=dimA+m.

  2. 2.

    If k is a field, then dimk⁡[x1,…,xm]=m. Moreover, ht⁡((x1,…,xi))=i.

Proof.

We only need to prove the part about the height of (x1,…,xi). Then we have

Then clearly ht⁡((x1,…,xi))≥i and the inequality cannot be strict because otherwise (x1,…,xn) has height strictly larger than n. □

Exercise 3.3.6.

Let A be Noetherian, I⊂A, and I′⊂A⁡[X]. Suppose I′=I⁡[X]. Show that ht⁡(I′)=ht⁡(I).

Theorem 3.3.7 (Noether Normalization).

Let A be a finitely generated k-algebra over a field k. Let 𝔞1⊊𝔞2⊊⋯⊊𝔞p be a chain of prime ideals of A. Then there exist elements x1,…,xm∈A algebraically independent such that:

  1. 1.

    A is integral over B=k⁡[x1,…,xm];

  2. 2.

    For all i=1,…,p, there exists an integer h⁡(i)≥0 such that 𝔞i∩B=(x1,x2,…,xh⁡(i)). In particular, ht⁡(𝔞i)=h⁡(i).

Remark 3.3.8.

Recall that x1,…,xm are algebraically independent means that the map k⁡[X1,…,Xm]→A sending Xi↦xi is injective.

Proof of Theorem.

We will treat the case where A=k⁡[y1,…,ym] because any finitely generated algebra over k which is an integral domain is integral over such a ring. We will prove the result by induction on m and p. The case when m=1 is clear because k⁡[y1] is a PID.

We will assume the result is true for m−1. Now we will form an induction on p. In the case p=1, first assume that 𝔞1=(x1), where x1∉k. Then x1=f⁡(y1,…,ym)∈k⁡[y1,…,ym]. For i=2,…,m we will introduce xi=yi−y1ri for some integer ri. We want to choose the ri such that y1 is integral over k⁡[x1,…,xm]. Writing

we see that f⁡(p¯)=p1+r2⁢p2+⋯⁢rm⁢pm is the maximal degree of y1 in this expression. Then it is possible to choose r2,…,rm such that f⁡(p¯) are all distinct, for example ri=ki for k>max⁡{pi}, where the max is taken over all pi that occur in the polynomial.

Then choosing the p¯ for which f⁡(p¯) is maximal, we can write

Thus y1 is integral over k⁡[x1,…,xm], so yi=xi+y1ri is integral over k⁡[x1,…,xm]. Therefore A is integral over k⁡[x1,…,xm]. Finally, x1,…,xm are algebraically independent because otherwise the transcendance degree of Frac⁡(A) is smaller than m. We now show that 𝔞1∩B=(x1). If y∈𝔞1∩B, then write y=b′⁢x1 for some b′∈A. But then b′∈A∩Frac⁡(B), so because B is integrally closed, b′∈B and thus y∈B⁢x1.

For the general case, suppose 𝔞1 is generated by more than one element. Choose x1∈𝔞1∖k and choose t2,…,tm such that A is integral over C=k⁡[x1,t2,…,tm] and x1⁢A∩C=C⁢x1. By the induction hypothesis on m, there exist x2,…,xm such that k⁡[t2,…,tm] is integral over k⁡[x1,…,xm] and 𝔞1∩k⁡[t1,…,tm]∩k⁡[x2,…,xm]=(x2,…,xh). To see this, choose z∈𝔞1∩k⁡[x1,…,xh]. Then there exist hj∈k⁡[x2,…,xm] such that

bceause x1∈𝔞1∩k⁡[x1,…,xm]. Thus h0∈𝔞1∩k⁡[x1,…,xm]=(x2,…,xh) and thus z∈(x1,…,xh). This finishes the case p=1.

Now we complete the induction on p. Suppose we have a chain of prime ideals 𝔞1⊊⋯⊊𝔞p in A. Then we choose t1,…,tm such that

  • •

    A is integral over k⁡[t1,…,tm]

  • •

    𝔞i∩k⁡[t1,…,tm]=(t1,…,th⁡(i)) for i≤p−1.

Now we apply the case p=1 to the ideal 𝔞p∩k⁡[tr+1,…,tm] where r=h⁡(p−1). Thus there exist xr+1,…,xm such that k⁡[tr+1,…,tm] is integral over k⁡[x1,…,xm] and 𝔞p∩k⁡[xr+1⁣…,xm]=(xr+1,…,xh).

First, it is clear that A is integral over k⁡[t1,…,tr,xr+1,…,xm]. If we set xi=ti for i≤i, then we will show that

One direction is obvious from the inductive hypothesis. In the other direction, if we write y=∑ah¯⁢x¯h¯, then because x1,…,xr∈𝔞p−1∈𝔞p−1∩k⁡[x1,…,xm], we see that

Thus y∈(x1,…,xh), as desired. □

Corollary 3.3.9.

Let A be an integral domain of finite type over a field k. Then dimA equals the transcendence degree of the fraction field of A.

Proof.

There exist x1,…,xm such that A is integral over k⁡[x1,…,xm]. Then Frac⁡(A) is algebraic over k⁡(x1,…,xm). On the other hand, we know that dimA=m, which is the transcendance degree of k⁡(x1,…,xm). □

Corollary 3.3.10 (Nullstellensatz).

Let A be an algebra of finite type over a field k. Then for any maximal ideal 𝔪⊂A, A/𝔪 is algebraic over k.

Proof.

Note that dimA/𝔪=0, but this is also the transcendence degree over k by the previous corollary. □

Proposition 3.3.11.

Let A be an integral domain of finite type over a field k. Then for any prime ideal 𝔭∈Spec⁡A, we have ht⁡(𝔭)+dimA/𝔭=dimA.

Proof.

Let 𝔭∈Spec⁡A and hbe its height. Then by Noether normalization, there exist x1,…,xm, where m=dimA, such that A is integral over A′=k⁡[x1,…,xm] and 𝔭′=𝔭∩A′=(x1,…,xh). Then we know that A′/𝔭′≅k⁡[xh+1,…,xn]. Because A/𝔭 is integral over A′/𝔭′, we see that dimA/𝔭=dimA′/𝔭′=n−h. But then ht⁡(𝔭)=ht⁡(𝔭′)=h because A′ is integrally closed and A is integral over A′. Therefore,

Remark 3.3.12.

Let A be a finitely generated k-algebra. Then for any maximal ideal 𝔪⊂A, we know that A/𝔪 is an algebraic extension of k. Therefore we have a correspondence

In addition, any prime ideal 𝔭 defines a subvariety Spec⁡A/𝔭=V⁡(𝔭) of Spec⁡A.

Proposition 3.3.13.

Let A,A′ be two finitely-generated k-algebras that are domains. Then for any minimal prime ideal 𝔭⊂A⊗kA′, we have dimA⊗kA′/𝔭=dimA+dimA′.

Proof.

Choose B,B′ polynomials over k such that A (resp A′) is integral over B (resp B′). Then write d,d′=dimA,dimA′. Then A⊗A′ is torsion free over B⊗B′. Then because 𝔭⊂Spec⁡(A⊗A′) is minimal, we see that B⊗B′∩𝔭=0. Therefore A⊗A′/𝔭 is integral over B⊗B′ and thus the desired result follows using integrality. □

Remark 3.3.14.

We can think of Spec⁡A⊗A′ as the product Spec⁡A×Spec⁡A′. The proposition says that irreducible components of the product variety have the expected dimension.

Proposition 3.3.15 (Hilbert’s Nullstellensatz).

Let k be a field, A be a finitely-generated k-algebra and I⊊A be a proper ideal. Then I=⋂𝔪⊃I𝔪⁢ maximal𝔪.

Proof.

One direction is obvious because I is the intersection of all prime ideals containing I. Let a∈J=⋂𝔪∖I. Then S={1,a,a2,…}∩I=∅ and thus S−1⁢I⊂S−1⁢A is a proper ideal. Thus there exists a maximal ideal of S−1⁢A such that S−1⁢I⊂𝔪0. Because S−1⁢A is a finitely generated k-algebra, we have

Then writing 𝔪=𝔪0∩A, we see that k⊂A/𝔪⊂S−1⁢A/𝔪0 and thus dimA/𝔪=0 and therefore 𝔪⊃I is maximal. However, a∉𝔪 by hypothesis, which gives us a contradiction. □

3.4 Rings of Dimension 1

Definition 3.4.1.

A local ring A is called a discrete valuation ring if it is a principal ideal domain and has a nonzero prime ideal.

This prime ideal is naturally maximal becuase if 𝔭⊂𝔪⊂A given by (a)⊂(b), then we know a=b⁢s for some s∈A, but then s∈𝔭, so s=a⁢s′ and thus a=b⁢a⁢s′, so b⁢s′=1. In particular, dimA=1.

Definition 3.4.2.

A discrete valuation on A is a surjective function v:A∗→ℤ such that

  • •

    v⁡(x⁢y)=v⁡(x)+v⁡(y);

  • •

    v⁡(x+y)≥min⁡{v⁡(x),v⁡(y)}.

We define v⁡(0)=∞.

If A is a DVR, then choose x≠0. We define v⁡(x)=sup{n≥0∣x∈(πn)}, where π generates the maximal ideal of A. Then v⁡(x) is well-defined because ⋂0(πn)=0 by the Krull intersection theorem. We can extend this valuation to K=Frac⁡A by v⁡(x/y)=v⁡(x)−v⁡(y).

Proposition 3.4.3.

If K is a field and v:K∗→ℤ is a valuation, then A={x∈K∣v⁡(x)≥0}∪{0} is a discrete valuation ring.

Proof of this is left as an exercise.

Example 3.4.4.

Consider the ring ℤ(p) where p is a prime number. Then for any xy with y coprime to p, define vp⁢(xy) to be the maximal power of p dividing x.

Now let k be a field and let k⁡[[T]] be the ring of formal power series in T. Then any series a0+a1⁢T+⋯ is invertible iff a0≠0, and thus any element is a product of Tn and a unit for some n. Thus v⁡(F) is the degree of the first monomial with nonzero coefficient.

Proposition 3.4.5.

Let A be a ring. Then the following are equivalent:

  1. 1.

    A is a DVR.

  2. 2.

    A is a local noetherian ring and 𝔪A is generated by an element π which is not nilpotent.

Proof.

One direction is clear. If A is a DVR, then it is a domain, and thus the generator of the prime ideal is not nilpotent.

In the other direction, suppose 𝔪=(π). Then by the Krull intersection theorem, ⋂n(πn)=0. Then for 0≠x∈A, there exists a maximal n such that x∈(πn), so x=πn⁢u for some u∈A×. But then for y∈A,y≠0, write y=πm⁢v for v∈A×. Thus x⁢y=πm+n⁢u⁢v≠0 because π is not nilpotent. Therefore A is a domain and thus is a DVR. □

Proposition 3.4.6.

Let A be a local Noetherian ring. Then A is a DVR if and only if

  1. 1.

    A is integrally closed;

  2. 2.

    A has a unique nonzero prime ideal.

Proof.

One direction is clear. In the other direction, assume A is integrally closed and has a unique nonzero prime. First, we note that A must be local and 𝔪≠0. Then if we write

this is an A-module. If we write y∈𝔪, then y⁢𝔪′⊂A and thus 𝔪′⊂A⁢y−1. This implies that 𝔪′ is finitely generated. Then we have 𝔪⊂𝔪⁢𝔪′⊂A. We will show that this cannot equal 𝔪 by contradiction.

Let x∈𝔪′. Then x⁢𝔪⊂𝔪 and thus x is integral over A. Because A is integrally closed, x∈A. Thus 𝔪′=A. Now we set S={1,x,x2,…}. Then S−1⁢A=K because it has no nonzero prime ideals. If we choose z∈A∖0, then we can write 1z=yxn for some n≥0. This tells us that xn∈(z). Because 𝔪 is finitely generated, we see that 𝔪N⊂(z). Then let N0 be the smallestt integer such that 𝔪N0⊂(z)⊂𝔪 and let y∈𝔪N0−1∖(z). Then we have 𝔪⁢y⊂𝔪N0⊂(z) and thus yz∈𝔪′. This implies that 𝔪′⊋A and thus 𝔪⁢𝔪′=A. Therefore we can write

where xi∈𝔪,yi∈𝔪′. Therefore there exists i such that xi⁢yi−1∉𝔪 and thus yi−1⁢𝔪=A. Therefore 𝔪=(yi) is a principal ideal, so A is a DVR. □

Proposition 3.4.7.

Let A be a Noetherian domain. The following are equivalent:

  1. 1.

    For all 0≠𝔭∈Spec⁡A, the localization A𝔭 is a discrete valuation ring.

  2. 2.

    A is integrally closed and of dimension 1.

Proof.
1 implies 2:

Let x∈K be integral over A. But then x∈A𝔭 for all 𝔭≠0 and thus x∈⋂𝔭≠0A𝔭=A. Being of dimension 1 is easy. If 0≠𝔭⊂𝔪⊂A, we localize at 𝔪 and see that 𝔭⁢A𝔪=𝔪⁢A𝔪, so 𝔭=𝔪.

2 implies 1:

For all 𝔭, we know that A𝔭 is integrally closed. Because it has dimension 1, it must be a DVR.

□

Definition 3.4.8.

A ring A is called a Dedekind domain if it is a domain satisfying the properties of the previous proposition.

Example 3.4.9.

ℤ is a Dedekind domain. It is clearly a domain, having dimension 1 follows from being a PID, and being integrally closed is obvious. More generally, any principal ideal domain is a Dedekind domain.

Example 3.4.10.

Let A be a Dedekind domain and K=Frac⁡(A). Then let L/K be a finite extension and B be the integral closure of A in L. Then B is a Dedekind domain. In particular, if K is a number field, then its ring of integers 𝒪K is a Dedekind domain.

Here is another example of this. Let k be a field and A=k⁡[x],k=k⁡(X). Then if L/K is a field extension and B is the integral closure of A in L, then Spec⁡B→Spec⁡A is a smooth affine curve with a map to 𝔸1.

Remark 3.4.11.

All smooth curves can be obtained in this way (of taking the integral closure of some ring). Also, normalization resolves all singluarities of curves.

Definition 3.4.12.

A fractional ideal 𝔞⊂K=Frac⁡(A) is an A-submodule of finite type.

Note that for 𝔞,𝔟 nonzero fractional ideals, then 𝔞⋅𝔟 is a fractional ideal.

Proposition 3.4.13.

If A is a Dedekind domain, then the set of nonzero fractional ideals form an abelian group.

Proof.

It is easy to see that the multiplication is associative and commutative. Now we need to show that inverses exist. For 𝔞⊂K a fractional ideal, we need to find another fractional ideal 𝔟 such that 𝔞⁢𝔟=A. Set

Clearly we know that for 0≠x∈𝔞, we have 𝔞=1⊂x−1⁢A and thus 𝔞−1 is of finite type. Then for any prime ideal 𝔭 we know that 𝔞𝔭⋅𝔟𝔭=(𝔞⋅𝔟)𝔭, and thus 𝔞𝔭−1=(𝔞−1)𝔭. Therefore we have

and thus 𝔞⋅𝔞−1=A. □

Now if A is a Dedekind domain and 0≠𝔭≠𝔭′⊂A, then 𝔭+𝔭′=A, so 𝔭⁢𝔭′=𝔭∩𝔭′. This implies that for any 0≠𝔞⊂A that 𝔭⋅𝔞⊊𝔞 because otherwise 𝔭=A.

Lemma 3.4.14.

If x∈A is nonzero, then there are only finitely many maximal ideals 𝔭 such that x∈𝔭.

Proof.

Let x∈𝔭. Then 𝔞−1⊂x−1⁢A, so if x∈𝔭1,𝔭2,… for infinitely many maximal ideals, then

is a strictly decreasing infinite chain of ideals containing x. Thus we have

but this is impossible because x−1⁢A is Noetherian. □

Remark 3.4.15.

For any nonzero ideal I⊂A, there are only finitely many prime ideals 𝔭 such that I⊂𝔭.

Definition 3.4.16.

Let 𝔭 be a maximal ideal of A. Then let v𝔭⁢(I) be the n𝔭 such that I⁢A𝔭=ω𝔭n𝔭⁢A𝔭. Thus I⊂𝔭n𝔭 but I⊈𝔭n𝔭+1.

Corollary 3.4.17.

Let I be a fractional ideal of A, where A is a Dede,ind domain. Then

Proof.

If we denote the product by J, then I𝔭=J𝔭 for all 𝔭 and thus I=J. □

Remark 3.4.18.

This corollary gives the primary decomposition of an ideal in a Dedekind domain. In number theory, this replaces the prime factorization of an integer.

3.5 Depth

Let A be a ring and M be an A-module. Then for a1,…,ar∈A, write a¯=(a1,…,ar)⊂A.

Definition 3.5.1.

The sequence a1,…,ar is an M-regular sequence if it satisfies the following:

  1. 1.

    ai is not a zero divisor of M/(a1,…,ai−1)⁢M for i=1,…,r.

  2. 2.

    a¯⋅M⊊M. Therefore we have

Lemma 3.5.2.

Assume that a¯=(a1,…,ar) is M-regular and let m1,…,mr∈M such that ∑i=1rai⁢mi=0. Then mi∈a¯⋅M for all i.

Proof.

We will induct on r. If r=1, then a1⁢m1=0 implies m1=0. Now assume a1⁢m1+⋯+ar⁢mr=0 implies that ar⁢m¯r=0 in M/(a1,…,ar−1)⁢M. Then there exists n1,…,nr−1 such that mr=a1⁢n1+⋯+ar−1⁢nr−1, and thus

Thus mi+ar⁢ni∈(a1,…,ar−1)⁢M and thus mi∈(a1,…,ar)⁢M. □

Theorem 3.5.3.

Assume that (a1,…,ar) is an M-regular sequence. Then for any integers n1,…,nr, the sequence a1n1,…,arnr is M-regular.

Proof.

It is sufficient to prove that a1n,a2,…,ar is an M-regular sequence. We will induct on n. Assume that a1n−1,…,a2,ar is M-regular. First, multiplication by a1n is clearly injective.

Now if a1n,a2,…,aj−1 is M-regular, then let m∈M such that

By induction on n, we can write

Multiplying this by aj and combining the two equations, we obtain

By the previous lemma, we see that a1⁢m1−aj⁢m1′∈(a1n−1,…,aj−1)⁢M. Therefore, aj⁢m1′∈(a1,…,aj−1)⁢M, so m1′∈(a1,…,aj−1)⁢M. This implies that m∈(a1n,…,aj−1)⁢M. □

Definition 3.5.4.

The sequence (a1,…,ar) is said to be M-quasi-regular if one of the following equivalent conditions holds:

  • •

    For all F⁡(x1,…,xr)∈M⁡[x1,…,xr]=A⁡[x1,…,xr]⊗AM homogeneous of degree n such that F⁡(a1,…,ar)∈In+1⁢M, this implies that F⁡(x1,…,xr)∈I⁢M⁢[x1,…,xr], where I=(a1,…,ar).

  • •

    If F⁡(x1,…,xr)∈M⁡[x1,…,xr] is homogeneous and such that F⁡(a1,…,ar)=0, then F∈I⁢M⁢[x1,…,xr]

  • •

    The map

    is an isomorphism.

Lemma 3.5.5.

Assume that (a1,…,ar) is M-quasi-regular and x∈A. Then if (I⁢M:x)=I⁢M, then (In⁢M:x)=In⁢M for all n≥1.

Proof.

We will induct on n. Suppose that m∈(In⁢M:x). Then x⁢m∈In⁢M⊂In−1 and thus m∈In−1⁢M. Therefore there exists g⁡(X1,…,Xr) homogeneous of degree n−1 such that m=g⁡(a1,…,ar). This implies that x⁢g⁢(X1,…,Xr)∈M⁡[X¯], so

and then quasi-regularity gives us that x⁢g⁢(X1,…,Xr)∈I⁢M⁢[X¯]. This implies that

as desired. □

Proposition 3.5.6.

Using the same notation, if (a1,…,ar) is M-regular, then it is M-quasi-regular. Conversely, if (a1,…,ar) is M-quasi-regular and M,M/a1⁢M,…,M/(a1,…,ar)⁢M are Hausdorff in the I-adic topology, then (a1,…,ar) is M-regular.

Proof.

First we prove that regular implies quasi-regular by induction on r. Clearly r=1 is obvious. Suppose g⁡(x1)∈M⁡[x] is homogeneous of degree n. Then if g⁡(a1)∈a1n+1⁢M, we have a1n+1⁢m′=a1n⁢m and thus a1n⁢(m−a1⁢m′)=0. By regularity, we have m=a1⁢m′ and thus g⁡(x1)∈a1⁢M⁢[x1].

For the inductive step, suppose that (a1,…,ar) is regular. Then we know that (a1,…,ar−1) is M-quasi-regular, so now choose F⁡(x1,…,xr)∈M⁡[x1,…,xr] homogeneous of degree q and such that F⁡(a1,…,ar)=0. Then we can write

where H is homogeneous of degree q−1. Then G⁡(a1,…,ar−1)∈I0q⁢M, where I0=(a1,…,ar−1). This implies that ar⁢H⁢(a1,…,ar−1)∈I0q⁢M, which implies that H⁡(a1,…,ar−1)∈(I0q⁢M:ar). Because a1,…,ar−1 is quasi-regular and a1,…,ar is regular, we have (I0⁢M:ar)=I0⁢M. This implies that (I0q⁢M:ar)=I0q⁢M and thus H⁡(a1,…,ar)∈I0q⁢M. Then H⁡(a1,…,ar)=h⁡(a1,…,ar−1) where h is homogeneous of degree q. Now let

Because g⁡(a1,…,ar−1)=F⁡(a1,…,ar)=0, we see that g⁡(x1,…,xr−1)∈I0⁢M⁢[x1,…,xr−1] by induction. We conclude that G∈I⁢M⁢[x1,…,xr−1]. Because H∈I0⁢M⁢[x1,…,xr], we have F∈I⁢M⁢[x1,…,xr].

Now in the other direction, we will induct on r. If r=1, assume that a1 is M-quasi-regular. We need to show that m↦a1⁢m is injective. Suppose that a1⁢m=0. Then if we consider the polynomial g0⁢(x)=m, we see that a1⁢g0⁢(a1)=0 and thus x1⁢g0⁢(x1)∈a1⁢M⁢[x1]. This means that g0⁢(x1)∈a1⁢M⁢[x1] and thus m∈a1⁢M. Then there exists g1 homogeneous of degree 1 such that m=g1⁢(a1). Then we see that x1⁢g1⁢(x1)∈I⁢M⁢[x1] and thus g1⁢(x1)∈I⁢M⁢[x1], so m=g1⁢(a1)∈I2⁢M. Then there exists g2∈M⁡[x1] homogeneous of degree 2 such that m=g2⁢(a1). Then a1⁢m=a1⁢g2⁢(a1)=0, so we deduce that x1⁢g2⁢(x1)∈I⁢M and thus g2⁢(x1)∈I⁢M. Evaluating at a1, we see that m∈I3⁢M. In particular, we see that m∈⋂n≥1In⁢M=0, where the last equality uses the Hausdorff condition, and thus m=0.

Now for the induction, we know that a1 is M-regular. We need to show that (a2,…,ar) is M/a1⁢M-regular. This follows from the inductive hypothesis if we check that (a2,…,ar is M/a1⁢M≕M¯)-quasi-regular. Choose F⁡(x2,…,xr)∈M⁡[x2,…,xr] homogeneous of degree n such that F⁡(a2,…,ar)∈a1⁢M. Then we can write F⁡(a1,…,ar)=a1⁢m, so let i be such that m∈Ii⁢M. Then let G∈M⁡[x1,…,xr] be homogeneous of degree i and satisfy m=G⁡(a1,…,ar). Then the polynomial

vanishes at (a1,…,ar). If i<n−1, then a1⁢G⁢(a1,…,ar)=F⁡(a2,…,ar)∈In⁢M⊂Ii+2⁢M. But then x1⁢G⁢(x1,…,xr) is homogeneous of degree i+1 and thus x1⁢G⁢(x1,…,xr)∈I⁢M⁢[x1,…,xr] by quasi-regularity. This implies that

We can repeat this until m∈In−1⁢M and G is of degree n−1. Then

is homogeneous of degree n and g⁡(a1,…,ar)=0. This implies that g⁡(x1,…,xr)∈I⁢M⁢[x1,…,xr] and thus F⁡(x2,…,xr)∈I⁢M⁢[x2,…,xr]. Thus

Then (a2,…,ar) is M¯-quasi-regular, so they are M¯-regular by induction. □

Remark 3.5.7.

For A Noetherian and M of finite type, regular and quasi-regular are equivalent.

Definition 3.5.8.

Let I⊂A be an ideal and M and A-module. Then the I-depth of M is the (possibly infinite) length of the longest M-regular sequence in I.

Before we continue, recall that Ext∙ are the right derived functors of Hom and are computed by taking projective resolutions of the first argument or injective resolution of the second argument. In particular, if 0→A→B→C→0 is an exact sequence, we have an exact sequence

and similarly,

Theorem 3.5.9.

Assume that A is Noetherian, M a finite A-module, and I⊂A an ideal such that I⁢M≠M. Let m∈ℤ>0. Then the following are equivalent:

  1. 1.

    Extn⁡(N,M)=0 for all i<n and any finitely A-module N such that supp⁡(N)⊂V⁡(I).

  2. 2.

    Exti⁡(A/I,M)=0 for all i<n.

  3. 3.

    Exti⁡(N,M)=0 for all i<n for some finite A-module N such that supp⁡(N)=V⁡(I).

  4. 4.

    There exists a M-regular sequence (a1,…,an) of length n inside I.

Proof.

Clearly 1 implies 2 implies 3. Now we prove 3 implies 4 implies 1.

3 implies 4:

Assume that Hom⁡(N,M)=0. If I does not contain any M-regular element, then I⊂⋃𝔭∈Ass⁡M𝔭 and thus I⊂𝔭 for some 𝔭∈Ass⁡(M). Then A/𝔭↪M, which is equivalent to A𝔭/p⁢A𝔭↪M𝔭. On the other hand, we know 𝔭∈V⁡(I)=supp⁡(N), so N𝔭≠0. By Nakayama, we see that N𝔭⊗k⁡(𝔭)≠0, and thus HomA𝔭⁡(N𝔭,M𝔭)≠0, but this is simply HomA⁡(N,M)𝔭. Thus Hom⁡(N,M)≠0 and thus there exists a1∈I that is M-regular. Then we have an exact sequence

Writing M1=M/a1⁢M, we have proved the case n=1 and now proceed by induction on n. Applying Extn⁡(N,−) to the above exact sequence, we have the exact sequence

and deduce that Exti⁡(N,M1)=0 for i<n−1. Applying the case n−1 to M1, we obtain an M1-regular sequence a2,…,an∈I and thus (a1,…,an) is M-regular.

4 implies 1:

We will induct on n. We have a sequence a1,…,an∈I that is M-regular. Then set M1=M/a1⁢M, so we have an exact sequence

Then choose N such that supp⁡(N)⊂V⁡(I). Then we have an exact sequence

If i<n, then i−1<n−1. By the inductive hypothesis, this implies that Exti−1⁢(N,M1)=0. Thus the map Exti⁡(N,M)→×a1Exti⁡(N,M) is injective, so because supp⁡(N)⊂V⁡(I), we have I⊂Ann⁡(N). Thus there exists m such that a1m∈Ann⁡(N). From the exact sequence, we know Exti⁡(N,M)→×a1mExti⁡(N,M) is injective, which means that Exti⁡(N,M)=0. □

Definition 3.5.10.

If A is a local Noetherian ring and M is an A-module, then we define the depth of M to be

This is the same as the maximal length of an M-regular sequence in 𝔪.

Corollary 3.5.11.

Let A be local Noetherian and M be a finitely-generated A-module. Then depth⁡(M)=n if and only if there exists a M-regular sequence a1,…,an such that Exti⁡(k,M)=0 for all i<n and Extn⁡(k,M)=Hom⁡(k,M¯), where M¯=M/(a1,…,am)⁢M and k=A/𝔪.

Proof.

We know that the equivalence of 2 and 4 from the theorem implies the corollary except for Extn⁡(k,M)=Hom⁡(k,M¯), but this fact be proved by induction on n using the fact that

is exact because this implies that Extn−1⁡(k,M/a1⁢M)=Extn⁡(k,M). This gives the desired result. □

Lemma 3.5.12.

Let A be a local Noetherian and M,N finite A-modules. Then if k=depth⁡(M) and r=dim(N), then Exti⁡(N,M)=0 for all i<k−r.

Proof.

We use induction on r. For the case r=0, we know that supp⁡(N)={𝔪} and this follows by the previous theorem. Now for the inductive hypothesis, we may assume that N=A/𝔭 and dimA/𝔭=r. This is possible because we can consider a filtration on N with successive quotients of the form A/𝔭i with dimA/𝔭i≤r. We know that 𝔪≠𝔭, so there exists x∈𝔪∖𝔭, and this x is N-regular. Then we have the exact sequence

where N′=N/x⁢N. Because x¯ is not in any minimal prime ideal of A/𝔭, we know that r′≔dimN′<dimA/𝔭. By induction, we know that Extj⁡(N′,M)=0 for k<k−r′. Now if i<k−r, we know that i+1<k−r′ and thus Exti+1⁡(N′,M)=0. Now considering the exact sequence of Ext groups, we have

and thus Exti⁡(N,M)/x⁢Exti⁡(N,M)=0 and thus Exti⁡(N,M)=0 by Nakayama’s lemma. □

Remark 3.5.13.

If N,M are finitely-generated, then Exti⁡(N,M) is also finitely-generated.

Theorem 3.5.14.

Let A be local Noetherian and M be a finitely-generated A-module. Then for all 𝔭∈Ass⁡(M), we have depth⁡M≤dimA/𝔭.

Proof.

We know A/𝔭↪M, so Hom⁡(A/𝔭,M)≠0. Thus 0≥depth⁡M−dimA/𝔭 by the lemma. □

Corollary 3.5.15.

Let A be local Noetherian. Then depth⁡A≤dimA.

In general, this inequality is strict, so we will later study the rings for which this is an equality.

Lemma 3.5.16.

Let A be a local Noetherian ring. Then let M be a finitely-generated A-module and (a1,…,ar) be an M-regular sequence. Then dimM/(a1,…,ar)<=dimM−r.

Proof.

We prove this by induction on M. It suffices to do this for r=1, so choose x∈A an M-regular element. We know that dimM/x⁢M≥dimM−1 for any x∈A, so we need to prove this is an equality. Then we know supp⁡(M/x⁢M)=supp⁡(M)∩V⁡(x), and thus x is not contained in any minimal prime ideal in supp⁡(M) by regularity. Therefore supp⁡(M/x⁢M) does not contain any minimal ideal of V⁡(Ann⁡(M))=supp⁡(M). In particular, this means that dimM/x⁢M<dimM, as desired. □

Lemma 3.5.17.

Let M,N be finitely-generated A-modules. Then supp⁡(M⊗N)=supp⁡(M)∩Supp⁡(N).

Proof.

Let 𝔭∈Supp⁡(M)∩Supp⁡(N). Then M𝔭,N𝔭≠0. By Nakayama, we have M𝔭⊗k⁡(𝔭)≠0 and N𝔭⊗k⁡(𝔭)≠0 where k⁡(𝔭)=A𝔭/𝔭⁢A𝔭. This implies that

and therefore M𝔭⊗k𝔭⁢N𝔭=(M⊗AN)𝔭≠0. Thus supp⁡(M)∩supp⁡(N)⊂supp⁡(M⊗N). Note that for N=A/x⁢A, we have supp⁡(M/x⁢M)=supp⁡(M)∩V⁡(x) because M⊗N=M/x⁢M. □

4 Geometric Aspects of Commutative Algebra

4.1 Cohen-Macaulay Rings and Modules

Let A be a local Noetherian ring and M be a finitely-generated A-module.

Definition 4.1.1.

Recall that dimM≥depth⁡M. Then M is Cohen-Macaulay if dimM=depth⁡M.

Theorem 4.1.2.

Let A be local Noetherian and M be finitely generated.

  1. 1.

    If M is Cohen-Macaulay, then for any 𝔭∈Ass⁡(M), depth⁡M=dimA/𝔭.

  2. 2.

    If f∈A is M-regular and M′=M/f⁢M, then M is Cohen-Macaulay if and only if M′ is.

  3. 3.

    If M is Cohen-Macaulay, then for all 𝔭∈Spec⁡A, M𝔭 is Cohen-Macaulay and depth𝔭⁡M=depthA𝔭⁡M𝔭.

Proof.
  1. 1.

    Let M≠0 and dimM=depth⁡M. Then let 𝔭∈Ass⁡(M)⊂supp⁡(M). This implies that dimA/𝔭≤dimM, but also dimA/𝔭≥depth⁡(M)=dimM.

  2. 2.

    Let f be M-regular. Then we know depth⁡M/f⁢M=depth⁡M−1 (this follows from the Theorem 3.5.9, applying Ext⁡(k,−) to the exact sequence 0→M→M→M/f⁢M→0). But then we proved earlier that dimM/f⁢M=dimM−1, and thus M is C-M if and only if M/f⁢M is C-M.

  3. 3.

    Let 𝔭∈supp⁡(M). Then 𝔭⊃Ann⁡(M) and M𝔭≠0. But then if x1,…,xr∈𝔭 is M-regular where r=depth𝔭⁡M, then x1,…,xr∈𝔭⁢A𝔭 is M𝔭-regular by exactness of localization. This implies that depth⁡M𝔭≥r=depth𝔭⁡M. We know that dimM𝔭≥depth⁡M𝔭, so we need to prove that r=dimM𝔭. We do this by induction on depth𝔭⁡M. If r=0, then we know Hom⁡(A/𝔭,M)≠0. Thus there exists 𝔭′⊃𝔭 such that A/𝔭′↪M, so 𝔭′∈Ass⁡M. By minimality of associated primes, we have 𝔭′=𝔭. Now dimM𝔭′=0 because 𝔭 is maximal in A𝔭 and minimal in supp⁡(M).

    Now in general, assume depth𝔭⁡(M)>0. Let a∈𝔭 be M-regular. Thus a is M𝔭-regular, so set M1=M/a⁢M. We know that dim(M1)𝔭=depth𝔭⁡M1 by the inductive hypothesis. This implies that dimM𝔭=depth𝔭⁡M because dim(M1)𝔭=dimM𝔭−1 and depth𝔭⁡M1=depth𝔭⁡M−1. □

Theorem 4.1.3.

Let A be local Noetherian and Cohen-Macaulay. Then

  1. 1.

    For all proper ideals I⊊A, we have ht⁡(I)+dimA/I=dimA.

  2. 2.

    For all P⊃Q prime ideals, then ht⁡(P)−ht⁡(Q)=ht⁡(P/Q), where the last height is taken in A/Q.

  3. 3.

    Let x1,…,xr∈𝔪. Then the following are equivalent:

    1. (a)

      x1,…,xr are A-regular.

    2. (b)

      ht⁡(x1,…,xi)=i for all i≤r.

    3. (c)

      ht⁡(x1,…,xr)=r.

    4. (d)

      There exists xr+1,…,xn∈𝔪 with n=dimA such that x1,…,xn is a system of parameters for A.

Proof.

We will first prove 3 and then use this to prove 1 and 2.

a implies b:

We already know that ht⁡(x1,…,xi)≤i. First, we know that x1 does not belong to any minimal prime ideal of A because x1 is A-regular. Thus ht⁡(x1)=1. Then in A/x1, the sequence x¯1,…,x¯r is A/x1-regular, and thus ht⁡(x¯2,…,x¯i)=i−1. Thus we can find

such that 𝔭¯i⊋⋯⊋𝔭¯1⊋0. This implies that ht⁡(𝔭i)≥1 and thus ht⁡(x1,…,xr)≥i.

b implies c:

This is trivial.

c implies d:

Assume r<dimA. Then there exists xr+1∈𝔪 that does not belong to any minimal prime ideal containing (x1,…,xr). Therefore ht⁡(x1,…,xr+1)=r+1 and then we can find xr+2,…,xn such that ht⁡(x1,…,xn)=n. But this implies that I=(x1,…,xn) is an ideal of definition for A and so (x1,…,xn) is a system of parameters.

d implies a:

Note that so far we have not used the fact that A is Cohen-Macaulay. We show that a system of parameters is A-regular. Let 𝔭∈Ass⁡(A). Then dimA/𝔭=n. This implies that 𝔭 is minimal and therefore x¯1≠0 in A/𝔭 because otherwise x¯1,…,x¯n would be a system of parameters in A/𝔭. But then this means x1∉𝔭 for all 𝔭∈Ass⁡(A). This implies that x1 is A-regular, and thus A/x1⁢A is Cohen-Macaulay of dimension n−1. On the other hand, x¯2,…,x¯n is a system of parameters for A/x1⁢A and the desired result follows by induction.

Now we will prove 1 and 2.

  1. 1.

    Let I⊊A be an ideal with height ht⁡(I)=r. Then we can choose a1⁢…,ar∈I such that ht⁡(a1,…,ar)=r. By 3a, this means that a1,…,ar is A-regular and thus r≤depthI⁡(A). On the other hand, if we have b1,…,bs∈I an A-regular sequence, we see that s≤ht⁡(I). This tells us that ht⁡(I)=depthI⁡A. But then recall that

    and that

    This means we can assume that I=p is a prime ideal with 𝔭=r. In this case, we know that A𝔭 is Cohen-Macaulay and

    so there exists x1,…,xr∈I an A-regular sequence such that A/(x1,…,xr) is Cohen-Macaulay of dimension n−r. Then 𝔭¯ is minimal in A/(x1,…,xr) because otherwise we would be able to fine xr+1∈𝔭 which is not a zero divisor in A/(x1,…,xr) and then depth⁡(𝔭)≥r+1. This implies

    and thus dimA/𝔭+ht⁡(𝔭)=n−r+r=n=dimA.

  2. 2.

    Suppose P⊃Q are prime ideals. We know AP is Cohen-Macaulay. Then we know that dimAP=ht⁡(Q⁢AP)+dim(AP/Q⁢AP) by 1 for AP. But then we have the desired result because dimAP=ht⁡(P), ht⁡(Q⁢AP)=ht⁡(Q), and dim(AP/Q⁢AP)=ht⁡(P/Q). □

4.2 Normal and Regular Rings

Let A be a domain and K=Frac⁡(A). If x∈K, we say that it is almost integral if there exists a∈A such that a⁢xm∈A for all n≥1.

Remark 4.2.1.

Clearly if x is integral, then x is almost integral. Also, if x is almost integral and A is Noetherian, x is integral.

To prove the first part, write

Then choose a such that a⁢xi∈A for i≤N−1 and then by induction we have a⁢xi∈A for i≥n.

To prove the second part, note that if a⁢xi∈A for all a≥1, then A⁡[x]⊂a−1⁢A. Because A is noetherian, then A⁡[x] is a finitely-generated A-module and then use the characterization of integral elements over a ring.

Definition 4.2.2.

We say that A is a normal domain (rep. completely normal domain) if every integral element (resp. almost integral element) of K over A belongs to A.

Remark 4.2.3.

Thering A is normal if and only if A𝔭 is normal for all prime ideals 𝔭. Additionally, for any normal ring A, all localizations S−1⁢A are normal.

We also note that if A is normal (resp. completely normal) then so is A⁡[x] when A is Noetherian. The proof is easy and uses the fact that K⁡[x] is a UFD and is thus completely normal. Then for f∈K⁡(x) is almost integral, it belongs to K⁡[x] and then we write

where ai∈K. Then there exists Q∈A⁡[x] such that Q⁢fn∈A⁡[x] for all n≥1 and thus αd⁢arn∈A for all n≥1, where Q=αd⁢xd+⋯+α0. This means that ar is almost integral and therefore ar∈A. Then f−ar⁢xr is almost integral and finally we proceed by induction to see that f∈A⁡[x].

Now we consider the normal rings of dimension 1.

Proposition 4.2.4.

Let A be a local Noetherian ring. Then the following are equivalent:

  1. 1.

    A is a discrete valuation ring.

  2. 2.

    A is normal and dimA=1.

  3. 3.

    A is normal and there exists 0≠a∈𝔪 such that (a) is 𝔪-primary.

  4. 4.

    The maximal ideal 𝔪 of A is principal.

Proof.

We will proceed quickly.

1 implies 2:

This was done previously.

2 implies 3:

Note that 𝔪 is the only prime ideal containing (a) because A has dimension 1 and is a local domain. This means that (a) is 𝔪-primary.

3 implies 4:

We know that 𝔪∈Ass⁡(A/(a)) and thus 𝔪=AnnA⁡(x¯) for some 0≠x¯∈A/a⁢A. If x∈A is a lift of x¯, then x⁢a−1∉A. However, we know that 𝔪⋅x⊆(a), so 𝔪⁢x⁢a−1⊂A. This tells us that 𝔪⊊𝔪⁢x⁢a−1 is a strict inclusion because otherwise x⁢a−1 is integral over A and thus x⁢a−1∈A. Therefore 𝔪⁢x⁢a−1=A, so there exists t∈𝔪 such that t⁢x⁢a−1=1 and thus t⁢A⊂𝔪∋y and thus y=t⁢x⁢a−1⁢y=t⁡(x⁢a−1⁢y), where x⁢a−1⁢y∈A, and thus y∈t⁢A, so 𝔪=t⁢A is principal.

4 implies 1:

If 𝔪=(t), then we know that

and so for all x≠0 there exists m such that x=u⁢tm where u∈A×, and thus A is a DVR. □

Now let A be a ring and I⊂A such that ⋂nIn=(0) and A is I-separated. Then for all 0≠a∈A, there exists n such that a∈In∖In+1. Then define a∗ to be the class of a in In/In+1 and

Theorem 4.2.5.

Let A,I be as before. Then

  1. 1.

    If A∗ is a domain, then so is A.

  2. 2.

    If A is Noetherian and I⊂rad⁡(A), then if A∗ is a normal domain, so is A.

Proof.
  1. 1.

    Choose a,b∈A nonzero. Then a∗,b∗≠0 and so a∗⁢b∗≠0 because A∗ is a domain. This means that if a∈In∖In+1,b∈Iq∖Iq+1, we know that a⁢b∈In+q∖In+q+1 and thus a⁢b≠0.

  2. 2.

    Let x∈K be integral over A. Write x=ab where a,b∈A. Then A/b⁢A is Noetherian and therefore by the Krull intersection theorem this is the same thing as

    To show that x∈A, we will prove that a∈b⁢A or that a∈In+b⁢A for all n≥0. If a∈In−1+b⁢A, then we know that a=a′+b⁢s, where a′∈In−1 and s∈A. Then we know that because ab is integral, so is a′b. Then we may assume that a∈In−1 and ab is almost integral. Then there exists c∈A such that c⁢am∈bm⁢A for all m≥1 and thus because a∗≠0, we know that (a∗)m=(am)∗≠0 and thus c∗⁢(a∗)m∈(b∗)m⁢A. Because A∗ is Noetherian and normal, it is completely normal, and thus a∗b∗∈A∗ and thus a∗=b∗⁢d∗ for some 0≠d∈A. This means that a−b⁢d∈IN+1 for N such that a∈IN∖IN+1. But we know this for N=n−1, and thus a−b⁢d∈In, so a∈In+b⁢A. □

Recall that if A is local Noetherian, we say that A is regular if there exists a system of parameters that generates the maximal idea 𝔪 of A.

Theorem 4.2.6.

Let A be local Noetherian, 𝔪 be its maximal ideal, and k be the residue field. Then A is regular if and only if

where d=dimA.

Proof.

Assume that A is regular. Then 𝔪=(x1,…,xd) is a system of parameters because dimk𝔪/𝔪2=dimA=d. Therefore we have a map

This map is clearly surjective, so we need to show that it is injective. If Q≠0 is in the kernel, then let q=deg⁡Q. Then (Q)⊂ker. We know that

and thus dimA=d⁡(A)≤d−1, a contradiction.

Now assume that gr𝔪⁢(A)≅k⁡[x1,…,xd]. By definition, this means that d⁡(A)=d. Also dim𝔪/𝔪2 is the dimension of the space of homogeneous polynomials of degree 1, so dim𝔪/𝔪2=d. Thus A is regular. □

Corollary 4.2.7.

Let A be a local Noetherian ring. Then the following are equivalent:

  1. 1.

    A is a discrete valuation ring.

  2. 2.

    A is a normal ring of dimension 1.

  3. 3.

    A is regular of dimension 1.

Proof.

We have seen before that 1 is equivalent to 2 which implies 3. To prove that 3 implies 1, we need to show that A is a domain. But then gr𝔪⁢(A)≅k⁡[X] is a domain and thus A is a domain. □

Corollary 4.2.8.

If A is regular, then A is a normal domain.

Theorem 4.2.9.

Let A be a regular local Noetherian ring and (x1,…,xd) be a regular system of parameters. Then

  1. 1.

    A is a normal domain.

  2. 2.

    x1,…,xd is an A-regular sequence and A is Cohen-Macaulay.

  3. 3.

    The ideal 𝔭i=(x1,…,xi) is a prime ideal of height i and A/𝔭i is regular of dimension d−i.

  4. 4.

    If 𝔭∈Spec⁡(A) and A/𝔭 is regular, then there exists a system of parameters y1,…,yd such that 𝔭=(y1,…,yi) and dimA/𝔭=d−i.

Proof.

We prove each part.

  1. 1.

    We already proved that the graded ring gr𝔪⁢(A) is a normal domain.

  2. 2.

    Use the previous theorem. Recall that k⁡[x1,…,xd]≃gr𝔪⁢(A). This means that x1,…,xd is an A-quasi-regular sequence. Because A is Noetherian, then x1,…,xd is an A-regular sequence. This implies that

    and thus A is Cohen-Macaulay.

  3. 3.

    The previous implies that dimA/(x1,…,xi)=d−i. Then the ideal 𝔪¯=𝔪/𝔭i is generated by the d−i elements x¯i+1,…,x¯d and these form a system of parameters. By definition, A/𝔭i is regular.

  4. 4.

    Write 𝔪¯=𝔪/𝔭. Then let i≤d such that dim𝔪¯/𝔪¯2=d−i. This means that i=dim(𝔭+𝔪2)/𝔪2, so we can choose y1,…,yi∈𝔭 such that y¯1,…,y¯i is a basis of (𝔭+𝔪2/𝔪2). We then complete this to a basis of 𝔪/𝔪2. Thus y1,…,yd is a regular system of parameters for A, and is thus is regular sequence. Then 𝔭′=(y1,…,yi)⊂𝔭 is a prime ideal and dimA/𝔭′=d−i. But then dimA/𝔭=p−i because A/𝔭 is regular by assumption, so we must have 𝔭=𝔭′. □

We now return to some definitions.

Definition 4.2.10.

Let M be an A-module. Then 𝔭∈Ass⁡(M) is called an embedded prime if it is not minimal in the support.

Recall that if M is Cohen-Macaulay, then for any 𝔭∈Ass⁡(M), then dimA/𝔭=dimM. Therefore M has no embedded primes.

Definition 4.2.11.

Let I be an ideal of A. We say that I is unmixed if for all 𝔭∈Ass⁡(A/I) we have ht⁡(𝔭)=ht⁡(I). Note this implies that A/I has no embedded primes.

Definition 4.2.12.

We say that the unmixed theorem holds for A if for all I=(a1,…,ar) with r=ht⁡(I), then I is unmixed.

Remark 4.2.13.

This is equivalent to saying that for all I=(x1,…,xr) of height ht⁡(I)=r, A/I has no embedded primes.

To see this, note that if A/I has no embedded primes, choose a minimal Q∈Ass⁡(A/I). We know that ht⁡(Q)≤r because Q is minimal containing I. However, this is at least

Theorem 4.2.14.

Let A be Noetherian. Then A is Cohen-Macaulay if and only if the unmixed theorem holds for A.

Proof.

Suppose that the unmixed theorem holds for A. Let 𝔭 be a prime ideal. Then we know r=ht⁡(𝔭)=dimAp, so we can choose a1,…,ar∈𝔭 such that ht⁡((a1,…,ai))=i. However, if A satisfies the unmixed theorem, then ai+1 is not in any prime ideal of Ass⁡(A/(a1,…,ai)). Here, we use the fact that A/(a1,…,ai) has no embedded primes because (a1,…,ai) is unmixed. Thus a1,…,ar is an A-regular sequence and is also A𝔭-regular. This implies that r≤depth⁡(A𝔭)≤dimAp=ht⁡(p)=r and thus A𝔭 is Cohen-Macaulay.

Now assume that A is Cohen-Macaulay. Then if I=(a1,…,ar) has height r, then a1,…,ar is an A-regular sequence. Thus A/(a1,…,ar) is Cohen-Macaulay and thus has no embedded primes, so (a1,…,ar) is unmixed. □

Returning to normal domains, we have the following result:

Theorem 4.2.15.

Let A be a Noetherian normal domain. Then any nonzero principal ideal is unmixed and

In particular, A is Cohen-Macaulay if dimA≤2.

Proof.

Let 0≠a∈A. To show that (a) is unmixed, we need to show that ht⁡((a))=1. Let 𝔭∈Ass⁡(A/(a)). Then 𝔭=AnnA⁡(b¯) for some b≠0. This means that

This means that

and thus we can replace A by A𝔭. This implies that 𝔭−1⁢A𝔭≠A𝔭, so

Then one of these two inclusions is an equality, so if the first inclusion is an equality, we know that elements of 𝔭−1⁢A𝔭 are integral over A𝔭 and thus 𝔭−1⁢A𝔭⊂A𝔭 because A is a normal domain. This is impossible, so 𝔭−1⁢A𝔭⋅𝔭⁢A𝔭=A𝔭 and thus 𝔭⁢A𝔭 is a principal ideal. Therefore A𝔭 is a DVR, so ht⁡(𝔭)=1. Thus (a) is unmixed.

Now we will prove that A is the intersection of the localizations at the height 1 prime ideals. One direction is obvious, so now let ba∈⋂ht⁡(𝔭)=1A𝔭. Then if ba∉A, we know that A/(a)≠0 and thus there exists 𝔭 such that Ass⁡(A/(a))≠∅. We know that ht⁡(𝔭)=1, so A𝔭≠𝔭=Ann⁡(b), a contradiction with the fact that ba∈A𝔭. □

4.3 Homological Theory

Lemma 4.3.1.

Let A be a ring and M be an A-module. Then

  1. 1.

    M is projective if and only if ExtA1⁡(M,N)=0 for all A-modules N.

  2. 2.

    M is injective if and only if ExtA1⁡(A/I,M)=0 for all ideals I⊂A.

Proof.

We only need to prove the second part. One direction follows from the direction, so now suppose we have 0→N→N′ injective. We need to show that Hom⁡(N′,M)→Hom⁡(N,M) is surjective. Consider the set

We will show that N′∈𝒮. First, 𝒮 contains a maximal element N1 by an application of Zorn’s lemma. If Nα is a chain, then ⋃αNα∈𝒮. Now we show that N1=N′. Otherwise, there exists x∈N′∖N1 and set N2=N1+A⁢x⊂N′. This gives an exact sequence

Now we apply the functor Hom⁡(−,M) and we see that Hom⁡(N2,M)↠Hom⁡(N1,M) because Ext1⁡(A/I,M)=0. But this implies that N2∈𝒮 and this contradicts maximality of N1. □

Definition 4.3.2.

Let A be a ring and M an A-module. Then we define the projective dimension proj.dim⁡M to be the length of the shortest projective resolution of M. Equivalently, we define the injective dimension inj.dim⁡M to be the length of the shortest injective resolution of M.

Lemma 4.3.3.

Let n be a positive integer. Then the following are equivalent:

  1. 1.

    All A-modules M have projective dimension at most n;

  2. 2.

    All A-modules M of finite type have projective dimension at most n;

  3. 3.

    All A-modules M have injective dimension at most n;

  4. 4.

    ExtAn+1⁡(M,N)=0 for all A-modules M,N.

Proof.

Clearly 1 implies 2.

2 implies 3:

Consider

such that U0,…,Un−1 are injective. Now we have an exact sequence

where V0 is the kernel of U1→U2. For any ideal I, we know that Extn+1⁡(A/I,M)=Extn⁡(A/I,V0) because U0 is injective. Now we obtain that Extn⁡(A/I,V0)=Extn−1⁡(A/I,V1) where V1=ker⁡(U2→U3). This implies that

By 2, we know that Extn+1⁡(A/I,M)=0 and therefore C is injective. Thus there is an injective resolution of M of length n.

3 implies 4:

This is trivial.

4 implies 1:

The proof is similar to the proof that 2 implies 3. □

Corollary 4.3.4.

The two numbers

are equal and are called the global homological dimension of A.

Lemma 4.3.5.

Suppose A is Noetherian and M is an A-module of finite type. Then M is projective if and only if Ext1⁡(M,N)=0 for all N of finite type.

Proof.

Because M is finite type, we can write

where K is finite type. Then if we apply ExtA⁡(−,K), we see that Ext1⁡(M,K)=0 and thus Hom⁡(An,K)→Hom⁡(K,K) is surjective. Thus there exists ϕ:An→K such that ϕ∘ψ=idK and thus the exact sequence is split. Thus M is a direct factor of a free module and is thus projective. □

Lemma 4.3.6.

Let A be a local Noetherian ring and M be a finite A-module. Then proj.dim⁡M≤n if and only if Torn+1A⁡(M,k)=0.

Proof.

Let P∙→M→0 be a projective resolution of length at most n. Then using this resolution, we compute Torn+1⁡(M,k)=0.

In the other direction, first assume that n=0. If Tor1A⁡(M,k)=0, we want to show that M is projective. Because M is finite type, then there exists Ar↠M such that Ar⊗k→M⊗k is an isomorphism. Applying the −⊗k functor to

we obtain the exact sequence

and this implies N⊗k=0. By Nakayama’s lemma, we see that N=0 and thus M is free. The general case is similar. We construct

where each Fi is projective, and then we see that Tor0⁡(C,k)=⋯=Torn+1⁡(M,k)=0. This implies C⊗k=0, so C=0 and thus we have a projective resolution of length n. □

Corollary 4.3.7.

Let A be Noetherian and M be a finite A-module. Then

  1. 1.

    The projective dimension of M is the supremum of the projective dimensions of M𝔪 for all maximal ideals 𝔪 of A.

  2. 2.

    proj.dim⁡M≤n if and only if Torn+1⁡(M,A/𝔪)=0 for all maximal ideals 𝔪 of A.

Proof.
  1. 1.

    This follows from the fact that ExtAi⁡(M,N)𝔪=ExtA𝔪i⁡(M𝔪,N𝔪) and the fact that for an A-module M we have M=0 if and only if M𝔪=0 for all 𝔪 maximal.

  2. 2.

    This follows from 1, the previous lemma, and the fact that Tor commutes with localization. □

Lemma 4.3.8.

Let A be noetherian. The following are equivalent:

  1. 1.

    gl.dim⁡A≤n.

  2. 2.

    proj.dim⁡M≤n for all finite A-modules M.

  3. 3.

    inj.dim⁡M≤n for all finite A-modules M.

  4. 4.

    ExtAn+1⁡(M,N)=0 for all finite A-modules M,N.

  5. 5.

    Torn+1A⁡(M,N)=0 for all finite A-modules M,N.

Proof.

We know that 1 is equivalent to 2, which implies 3. We know that 3 implies 4 and 2 implies 5 are trivial. The previous corollary gives us 5 implies 2, so we need to prove that 4 implies 2. Let

such that Fi are free of finite type. Then we know that

This implies that C is projective and thus proj.dim⁡M≤n. □

Corollary 4.3.9.

We have the local-global identity

Theorem 4.3.10.

Let A be local Noetherian with residue field k. Then gl.dim⁡A≤n if and only if Torn+1A⁡(k,k)=0. In particular, gl.dim⁡A=proj.dimA⁡k.

Proof.

Suppose that Torn+1⁡(k,k)=0. Thus proj.dim⁡k≤n, and thus Torn+1⁡(M,k)=0 for all M of finite type. This implies that proj.dim⁡M≤n for all finite A-modules M. Thus gl.dim⁡A≤n. □

Lemma 4.3.11.

Let (A,𝔪,k) be local Noetherian and M a finite A-module. Let x∈A be an M-regular element. Then if proj.dim⁡M≤∞, so is proj.dim⁡M/x⁢M and proj.dim⁡M/x⁢M=proj.dim⁡M+1.

Proof.

Let r=proj.dim⁡M<∞. Then we have the long exact sequence

so after applying the functor −⊗k, we know that Tori⁡(M,k)=0 if i>r. Thus if i>r+1, then Tori⁡(M/x⁢M,k)=0. For i=r+1, we have an exact sequence

but x∈M annihlates k and thus Torr+1⁡(M/x⁢M,k)≃Torr⁡(M,k)≠0, so proj.dim⁡(M/x⁢M)=r+1. □

Theorem 4.3.12.

Let (A,𝔪,k) be a regular local ring of dimension n. Then gl.dim⁡A=n.

Proof.

Let x1,…,xn be a regular system of parameters. Then we know k=A/(x1,…,xn). This implies that

and thus gl.dim⁡A=n. □

Corollary 4.3.13.

Let k be a field. Then gl.dim⁡k⁡[x1,…,xn]=n.

Proof.

Let A=k⁡[x1,…,xn]. Then A𝔪≅k′⁢[T1,…,Tn] for some other field k′ and is thus regular of dimension n. Thus gl.dim⁡A𝔪=n for all maximal ideals 𝔪, and thus gl.dim⁡A=n. □

4.4 Koszul Complex

Our goal is to prove the following result:

Theorem 4.4.1 (Serre).

A local Noetherian ring is regular if and only if its global dimension is finite.

To do this, we will introduce the Koszul complex. First, recall that if L∙,M∙ are two complexes, (L∙⊗M∙)n=⨁i+j=nLi⊗Mj and the differential satisfies the graded Leibniz rule

For x∈A, define K⁡(x)=→0→A→×xA→0→⋯. We can treat any A-module M as a complex by putting it in degree 0. Now for x1,…,xr∈A and M an A-module, define

Then for an integer p, we have

This satisfies

Now consider the exact sequence

Now if C is a complex, write C∙⁢(x)=C∙⊗K⁡(x), so we have a short exact sequence

This gives us a long exact sequence of modules

By definition of K⁡(x) it is easy to see that δp=(−1)p⁢x. As a consequence, if Hp⁢(C∙) is trivial, then so is Hp⁢(C∙⁢(x)) for p>1. In addition, we have

Thus if x is H0⁢(C)-regular, then H1⁢(C⁡(x))=0 and H0⁢(C⁡(x))=H0⁢(C)/x⁢H0⁢(C). If we iterate this, we have now proven the following result:

Theorem 4.4.2.

Let M be an A-module and x1,…,xr an M-regular sequence. Then

for all p>0 and H0⁢(x¯,M)=M/(x1,…,xr)⁢M.

Corollary 4.4.3.

If x1,…,xr is an A-regular sequence, then K⁡(x1,…,xr,A) is a finite free resolution of A/(x1,…,xr)⁢A.

Now we will discuss minimal resolutions. Suppose (A,𝓂,k) is a local ring. Then recall from the homework that u:L→M is called minimal when u⊗1k is an isomorphism. When M is finite, this is equivalent to u surjective and ker⁡(u)⊂𝔪⁢L. We now say that a resolution L∙→M is minimal if Li⊗k≃ker⁡(di)⊗k, where Li→diLi−1. In particular, writing L¯i=Li⊗k, then the complex

satisfies d=0, so Tori⁡(M,k)=Hi⁢(L¯∙)=L¯i. Therefore if L∙ is a minimal resolution of M, then rk⁡(Li)=dimkTori⁡(M,k).

Proposition 4.4.4.

Now assume A is local Noetherian and M is a finite A-module. Then there exists a minimal free resolution of M and any two such resolutions are isomorphic.

Proof.

To prove existence, begin with M¯. Choose a basis e1,…,en0∈M of M⊗k=M¯. This gives us a morphism 0→K0→An0→M→0 that gives an isomorphism kn≃M¯. Now K0 is a finite A-module, so we can construct a minimal An1→K0 with kernel K1. Iterating this procedure, we have a minimal resolution.

To prove uniqueness, if L∙,L∙′ are minimal free resolutions, we will construct a map

that makes this diagram commute. In degree zero, we have f0:L0→L0′ where f¯0 is an isomorphism, so L0≅L0′=An. Thus det⁡f0∈A×, so f0 is an isomorphism. The rest follows by iterating this procedure. □

Lemma 4.4.5.

Let F∙→M be a complex such that

  1. 1.

    Fp is a finite free A-module for all p;

  2. 2.

    F¯0→M¯ is surjective.

  3. 3.

    d(Fi)⊂𝔪⁢Fi−1 for all i>0 and induces a surjection F¯i↪𝔪/𝔪2⊗F¯i−1=𝔪⊗F¯i−1.

Then rkA⁢(Fi)≤dimkTori⁡(M,k).

Proof.

Consider a minimal free resolution L∙→M of M. Because free implies projective, we can find a map

Then the map F¯0→L¯0 must be injective because F¯0↪M¯ and L¯0≃M¯. Therefore, we can split L0≅F0⊕F0′. In particular, this means that rkA⁢(F0)≤rkA⁢(L0).

Now we have a diagram

where we have 𝔪⁢L0=𝔪⁢F0⊕𝔪⁢F0′, so downstairs we now have

and therefore F¯1→L¯1 is surjective. This gives us a splitting L1=F1⊕F1′ and thus iterating this, we obtain the desired result. □

Theorem 4.4.6.

Let (A,𝔪,k) be a local Noetherian ring and let s=dim𝔪/𝔪2. Then

In particular, Tori⁡(k,k)≠0.

Proof.

Write Fi≔Ki⁢(x1,…,xs;A) where x1,…,xs∈𝔪 is a basis for 𝔪/𝔪2. Thus we have F0=A→H0⁢(x¯,A)=A/(x1,…,xs)=k. Thus the F∙ satisfies the first two hypotheses of the previous lemma, so we check the third one. Here, we have

Reducing modulo 𝔪, because x¯1,…,x¯s is a basis of 𝔪=𝔪/𝔪/𝔪2, we see that this differential is injective. Now we can apply the previous lemma to obtain the desired result. □

We are now ready to prove the theorem of Serre.

Proof of Theorem 4.4.1.

We proved one direction last time. Now set s=dim𝔪/𝔪2. We know that Tors⁡(k,k)≠0 and thus s≤gl.dim⁢A=proj.dimA⁢(k). On the other hand, by the Auslander-Buchsbaum formula, we have

and because depth⁡k=0, we see that dimA≥depth⁡A≥s. However, we know that s≥dimA, and thus s=dimA=depth⁡A, so A is regular. □

Corollary 4.4.7.

Let A be a regular local ring and choose 𝔭∈Spec⁡A. Then A𝔭 is regular.

Proof.

This requires showing that A𝔭 has finite global dimension. If M is an A𝔭-module, then M is an A-module. Then if F∙→M is a projective resolution of M of length at most dimA, exactness of localization tells us that (F∙)A𝔭→M is an A𝔭-projective resolution of M. Thus proj.dimA𝔭⁢(M)<dimA, so A𝔭 has finite global dimension. □

4.5 Unique Factorization

Recall that a unique factorization domain, or a UFD, is a domain A such that every element has a unique factorization into irreducibles (up to multiplication of irreducibles by units). Here, π≠0 is irreducible if π=a⁢b implies a or b is a unit. For example, when A=ℤ, the irreducibles are precisely the prime numbers.

Exercise 4.5.1.

If A is Noetherian, then any element can be written as a product of irreducibles.

We say that two irreducibles π,π′ are equivalent if (π)=(π′). Then for each equivalence class, we can choose one representative, and the unique factorization means that if

then u=u′ and ni=ni′.

Lemma 4.5.2.

A Noetherian domain is a UFD if and only if (π) is a prime ideal for all irreducibles π.

Proof.

Suppose a⁢b∈(π). Then if a=u⁢∏iπini,b=v⁢∏iπini where π=π1, we know that n1+m1≥1, so at least one of them is at least 1, and thus either a∈(π) or b∈(π).

In the other direction, we will prove that Gauss’s lemma is satisfied. This says that if π∣a⁢b and π∤a, then π∣b. This implies that our domain is a UFD, and is equivalent to (π) being a prime ideal for all units. □

Lemma 4.5.3.

Let A be a Noetherian domain. Then A is a UFD if and only if every prime ideal of height 1 is principal.

Proof.

Suppose A is a unique factorization domain. Then let 𝔭 be a prime ideal of height 1. Because p≠0, there exists a=u⁢π1n1⁢⋯⁢πrnr. Thus there exists an irreducible π such that π∈𝔭, and thus because 0⊊(π)⊂𝔭 and ht⁡(𝔭)=1, we see that (π)=𝔭.

In the other direction, we will show that for any irreducible π∈A, the ideal (π) is prime. Let 𝔭 be a minimal prime ideal containing (π). We know that ht⁡(𝔭)≤1, and thus ht⁡(𝔭)=1. But then we know that 𝔭=(α), so π=α⁢β for some β∈A. But because π is irreducible, this means that α or β is a unit, and we know α is not a unit, so β is a unit and thus (π)=(α)=𝔭. □

Lemma 4.5.4.

Let A be Noetherian and x≠0 such that (x) is a prime ideal. Then Ax=A⁡[x−1] is a UFD if and only if A is.

Proof.

Suppose A is a UFD. Clearly Ax us a UFD. In the other direction, Then if

we must have ux=u⁢xn,ux′=u′⁢xn′, then we obtain u=u′,n=n′. □

Theorem 4.5.5.

Let A be a local Noetherian regular ring. Then A is a UFD.

Proof.

We induct on dimA. If dimA=0, then A is a field. If dimA=1, then A is a DVR and the result is clear.

For the inductive step, choose x∈𝔪∖𝔪2. Then (x) is a prime ideal, and we want to show that Ax is a UFD. Choose a prime ideal 𝔭′ of height 1 in Ax. We will show that 𝔭′ is principal. Write 𝔭=𝔭′∩A, so 𝔭′=𝔭⁢Ax. Then because A is regular, it has finite global dimension, and thus 𝔭 has a finite projective resolution

where Fi=Ami. Now for any prime ideal 𝔭¯⊂Ax, we have (Ax)𝔭¯=A𝔭¯∩A. This implies that if x∉𝔭¯∩A, then 𝔪≠𝔭¯∩A and thus ht⁡(𝔭¯∩A)<dimA. Then we know that A𝔭¯∩A is regular, so we can apply the inductive hypothesis and it is thus a UFD. Then we know that

and so because 𝔭𝔭¯′ is principal, it is free of rank 1 and thus has projective dimension 0. Thus 𝔭′ is a projective Ax-module, and so we have an exact sequence

where Fi′=Axmi. Then we decompose this into short exact sequences

Because 𝔭′ is projective, the exact sequences split and all Ki′ are projective. Therefore

Writing the first direct sum as F and the second as G, we have F=G⊕𝔭′. Thus F,G are free Ax-modules, and rank⁡F=rank⁡G+1. Write rank⁡G=r. Now we have ⋀r+1F=⋀rG⊗𝔭′ because ⋀i𝔭′=0 if i>1 (after localization, we obtain a free module of rank 1). In particular, we have Ax≅⋀r+1F=𝔭′ and thus 𝔭′ is free of rank 1 as an Ax-module, so it is a principal ideal. □

Definition 4.5.6.

Let A be a Noetherian ring. We say that A is regular if for all 𝔭∈Spec⁡A, A𝔭 is a local Noetherian regular ring.

4.6 Criterion for Normality

Let A be a ring, k≥0 an integer, and consider the conditions

(Rk):

For all 𝔭∈Spec⁡A such that ht⁡(𝔭)≤k, then A𝔭 is regular.

(Sk):

For all 𝔭∈Spec⁡A, depth⁡(A𝔭)≥inf(k,ht⁡(𝔭)).

Theorem 4.6.1.

Let A be a Noetherian ring. Then A is normal if and only if (S2) and (R1) hold.

Before we prove this, we will prove several lemmas.

Lemma 4.6.2.

The ideal (0) is unmixed if and only if (S1) holds. In other words, for all 𝔭∈Spec⁡A, we have depth⁡(A𝔭)≥inf(1,ht⁡(𝔭)).

Proof.

Assume that (0) is unmixed and let 𝔭∈Spec⁡A with ht⁡(𝔭)≥1. Then (0) is also unmixed in A𝔭, so Ass⁡(A𝔭)≅{𝔮⊊𝔭∣𝔮⁢ minimal}. Therefore there exists x∈𝔭⁢A𝔭∖⋃𝔮𝔮⁢A𝔭. Thus x is not a zero divisor of A𝔭, so it must be A𝔭-regular, and thus depth⁡(A𝔭)≥1.

Conversely, suppose (0) is not unmixed. Let 𝔭∈Ass⁡(A) with ht⁡(𝔭)≥1. But then 𝔭⁢A𝔭∈Ass⁡(A𝔭), and thus 𝔭⁢A𝔭 is contained in the set of zero-divisors. Thus depth⁡(A𝔭)=0. □

Corollary 4.6.3.

The property (Sk) holds if and only if for all i<k, any A-regular sequence (a1,…,ai) is unmixed. For i=0, this is saying that (0) is unmixed.

Proof.

Use the previous lemma and an induction argument. Also use the fact that depth⁡(A𝔭/a⁢A𝔭)=depth⁡(A𝔭)−1 and the same for dimension. □

Recall that a domain A is normal if it is integrally closed. Also, A=⋂𝔪A𝔪, so A is a normal domain if and only if all localizations at maximal ideals are normal domains.

Definition 4.6.4.

A ring A is called normal if and only if for all 𝔭∈Spec⁡A, A𝔭 is a normal domain.

It is not hard to see that it suffices to check this condition for maximal ideals.

Corollary 4.6.5.

Assume that A is reduced and that (S2) holds. Then

Here, the intersection is well-defined because the intersection of all minimal prime ideals is {0}.

Proof.

Assume that (S2) holds. Then consider αβ∈⋂𝔥⁢𝔱⁢(𝔭)=1A𝔭. We will consider the ideal I={x∈A∣x⁢αβ∈A}=Ann⁡(α¯∈A/β⁢A). Therefore we have an embedding A/I↪A/β⁢A given by x↦x⁢α¯. Assume I≠A, so A/I≠0. Then if Q∈Ass⁡(A/I), Q must be of height 1. Also, because I⊂Q, then IQ⊂Q⁢AQ. However, we know that αβ∈AQ. But then this implies that IQ=AQ, which is a contradiction. □

Lemma 4.6.6.

Assumes that A satisfies (S1) and (R0). Then A is reduced.

Proof.

Choose x∈A be nilpotent, so xn=0 for some n≥1. We will show that x=0. If 𝔭∈Spec⁡A with ht⁡(𝔭)=0, then by (R0) we know that A𝔭 is a field. This implies that x𝔭=0, so there exists s∉𝔭 such that s⁢x=0. This implies that Ann⁡(x)⊈𝔭 for any minimal prime 𝔭.

Now assume that x≠0 and let 𝔮 be a maximal element of the set of ideals {Ann⁡(a⁢x)∣a∈A}. Then we know that 𝔮 is a non-minimal prime ideal, so by (S1), we have depth⁡(A𝔮)≥1. Also, 𝔮=Ann⁡(y) for some y=a⁢x. Therefore A/𝔮↪A. But this implies that Ext∙⁡(A𝔮/𝔮⁢A𝔮,A𝔮)≠0 and thus depth⁡(A𝔮)=0, which is a contradiction. □

Proof of Theorem 4.6.1.

First assume that A is normal and choose 𝔭∈Spec⁡A. Then if ht⁡(𝔭)=0, we know A𝔭 is a field. If ht⁡(𝔭)=1, then A𝔭 has dimension 1 and is integrally closed, so it must be a DVRaand is thus regular. Thus (R1) is satisfied. Now if ht⁡(𝔭)=2, then A𝔭 is normal of dimension 2 and is thus Cohen-Macaulay. Therefore, depth⁡(A𝔭)=2. If ht⁡(𝔭)>2, we can choose 𝔮⊂𝔭 of height 2 and then depth⁡(A𝔭)≥depth⁡(A𝔮)=2, so (S2) is satisfied.

Now assume that (R1) and (S2) hold. By the previous lemma, we know that A is reduced. This embeds A↪Tot⁡(A)=∏iKi, where Ki=Frac⁡(A/𝔭i) with 𝔭1,…,𝔭r the minimal primes of A. It suffices to prove that A is integrally closed in Tot⁡(A). This is because if εi=(0,…,0,1,0,…,0), then εi2=εi and thus εi=A and thus A=∏i=1rA/𝔭i and then each A/𝔭i is a normal domain.

Let x∈Tot⁡(A) be integral over A. Then x𝔭∈Tot⁢(A)𝔭=Tot⁡(A). Then we know x𝔭 is integral over A𝔭, so if ht⁡(𝔭)=1, then A𝔭 is regular, so it is a DVR and is thus integrally closed. Therefore x𝔭∈A𝔭 for all 𝔭 of height 1. This implies that