PMATH 646: INTRODUCTION TO COMMUTATIVE ALGEBRA NOTES

HEESUNG YANG

1. January 06
Throughout this course, we will assume that R is a commutative ring with unity 1.
Proposition 1.1 (Zorn’s lemma). (S , ≤) is a partially ordered set, and suppose that for
(sα )α∈J , we have α <J β ⇒ sα ≤S sβ . If it exists s ∈ S such that s ≥ sα for all α ∈ J then
S has a maximal element.
Remark 1.1. In this course, we will use Zorn’s lemma mostly with S = collection of ideals
in a ring R, with ≤ given by ⊆. For instance, if R is a non-trivial ring (i.e., 0 6= 1, then R
has a maximal ideal. In fact, if I is a proper ideal of R then there exists a maximal ideal M
such that I ⊆ M .
Proof of the above remark. Let S = {J ⊆ R : J ideal, I ⊆ J ( R}. If I is a proper ideal
of R, then I ∈ S so S 6= ∅. Notice that ifSX is a totally ordered set and {Iα }α∈X isa
chain in S (i.e., α <X β ⇒ Iα ⊆ Iβ ), then Iα =: J ⊇ I. If a, b ∈ J, then there exists
α∈X
α, β ∈ X such that a ∈ Iα and b ∈ Iβ with either Iα ⊆ Iβ ⇒ a, b ∈ Iβ ⇒ a + b ∈ Iβ ⊆ J or
β ≤X α ⇒ Iβ ⊆ Iα ⇒ b, a ∈ Iα ⇒ b + a ∈ Iα ⊆ J.
If α ∈ J, r ∈ R then there exists α ∈ X such that a ∈ Iα ⇒ ra ∈ Ia ⊆ J. So J is an
ideal and J ⊇ I. To show that J ∈ S we must show that J ( R. But this follows from the
fact that if J = R then 1 ∈ J hence 1 ∈ Iα for some α, which is a contradiction. Hence we
can conclude that every chain in S has an upper bound, so by Zorn’s lemma there exists
a maximal ideal M ∈ S . Thus M ⊇ I. If there is another maximal ideal M 0 such that
M ( M 0 ( R then M ∈ S and M 0 ) M , contradicting the maximality. Thus M must be
the only maximal ideal, as required. 
Example 1.2. Take R to be an abelian group that does not have a maximal proper subgroup.
m
For instance, let R = {α ∈ C∗ : there exists m ≥ 1 such that α2 = 1}. Observe that any
proper subgroup of R is finite, so we can always make any proper subgroup bigger. Thus
R has no maximal proper subgroups. To make R into a ring without unity, we will define
addition (⊕) and multiplication ( ) as r s = 1 (here, 1 is actually the additive identity 0)
and r ⊕ s = r · s. This R has no maximal ideals. Thus, for our above argument to work, we
need R to have unity.
Theorem 1.3 (Chinese remainder theorem for rings). Let R be a commutative ring with 1
and I1 , . . . , Ik ⊆ R ideals satisfying
k
\
Ii = (0) and Ii + Ij = R whenever i 6= j
i=1

Date: 6 April 2015.

1
(in other words, the Ii are pairwise comaximal). Then we have
k
R∼
Y
= R/Ii .
i=1

Definition 1.4. Let R be a ring. Then an R-module M is just an abelian group (M, +)
endowed with a map R×M → M defined as (r, m) 7→ rm satisfying, for r ∈ R and m, n ∈ M :
(1) r · (s · m) = (rs) · m
(2) r(m + n) = rm + rn
(3) (r + s)m = rm + sm
(4) 1R m = m.
Example 1.5. If R = F a field and V is an F-module, then V is an F-vector space. If R = Z,
then M is an abelian group.
Example 1.6. If R = R[x] and a M = C, define p(x) · λ = p(i) · λ.
Definition 1.7. Let R be a ring and M an R-module. Then the annihilator of M denoted
by AnnR (M ) is defined to be
AnnR (M ) = {r ∈ R : rm = 0 for all m ∈ M }.
Remark 1.2. AnnR (M ) is an ideal of R. If r, s ∈ AnnR (M ) implies that (r + s) · m =
r · m + s · m = 0M + 0M = 0M for all m ∈ M . If a ∈ AnnR (M ) and r ∈ R, then
(ra)m = r(am) = r · 0M = 0M . It is trivially true that 0R ∈ AnnR (M ).
Remark 1.3. If S = R/ AnnR (M ) then M has the structure of an S-module. Indeed, if
s ∈ S and s = r + AnnR (M ) and we define s · m := r · m then the given multiplication is
well-defined by the annihilator’s property.

2. January 08
Recall that if I = Ann(M ), then M inherits a structure as an R/I-module. If r ≡ s
(mod I), then r − s ∈ I, or (r − s) · m = 0 for all m ∈ M . Therefore r · m = s · m. Therefore,
M gets an R/I-module structure via the rule (r + I) · m := r · m ∈ M .
Definition 2.1. An R-module M is faithful if AnnR (M ) = (0).
Remark 2.1. If I = AnnR (M ), then M is a faithful R/I-module.
Definition 2.2. If N ⊆ M and M is an R-module and N is an R-module, then we will call
N an R-submodule of M if:
(1) n1 , n2 ∈ N implies n1 + n2 ∈ N ;
(2) r ∈ R, n ∈ N implies r · n ∈ N ; and
(3) 0M ∈ N .
Remark 2.2. Some remarks on modules and submodules:
(1) R is an R-module.
(2) If I is an ideal of R, then I is an R-submodule of R.
2
 (3) If M is an R-module and I is an ideal of R, then we can construct an R-submodule
IM of M , where
( n )
X
IM = xi mi : xi ∈ I, mi ∈ M .
i=1

(4) If N ⊆ M is an R-submodule of M , we can form a quotient module M/N = {m+N :

m ∈ M }, where
(m1 + N ) + (m2 + N ) = (m1 + m2 ) + N and r(m + N ) = rm + N.
Definition 2.3. If R is a ring and M, N are two R-modules, then we say that a map
f : M → N is an R-module homomorphism if:
(1) f (m1 + m2 ) = f (m1 ) + f (m2 ) for all m1 , m2 ∈ M
(2) f (rm) = rf (m) for all r ∈ R and m ∈ M .
Remark 2.3. More facts about modules:
(1) ker(f ) = {m ∈ M : f (m) = 0} ⊆ M is a submodule.
(2) im(f ) = {f (m) : m ∈ M } ⊆ N is a submodule.
(3) ker(f ) = (0) iff f is injective
(4) im(f ) = N iff f is surjective
(5) If f : M → M 0 is a surjective R-module homomorphism, then M/ ker(f ) ∼
= M0
(“First isomorphism theorem”).
Definition 2.4. For two R-modules M and N , define
HomR (M, N ) := {f : M → N | f is an R-module homomorphism}.
If f : M → M , then f is said to be an endomorphism, and we write
HomR (M, M ) = EndR (M, M ).
Remark 2.4. Notice that HomR (M, N ) is itself an R-module, where (f + g)(m) := f (m) +
g(m) and (rf )(m) := rf (m) = f (rm). Then it is a straightforward verification. If M = N ,
then HomR (M, M ) = EndR (M, M ) is a ring. Define the composition map ◦ as multiplication.
Example 2.5. If R = C and M = C2×1 and N = C3×1 , then HomC (C2 , C3 ) = M3×2 (C).
Similarly, if R = C and M = C2 , then EndC (C2 ) ∼
= M2 (C).
Definition 2.6. A module M is simple if (0) and M are its only R-submodules.
Definition 2.7. We call D a division ring if every non-zero element of D has a multiplicative
inverse. Note that a division ring need not be commutative. Thus, a commutative division
ring is a field.
Lemma 2.8 (Schur’s lemma). If M is a simple R-module then EndR (M ) is a division ring.
2.1. Direct sum and direct product.
Definition 2.9. Let X be an index set and {Mα }α∈X is a collection of R-modules. Then
the direct sum of Mα is defined as
M
Mα = {(mα )α∈X : mα ∈ Mα and {α ∈ X : mα 6= 0} is finite}
α∈X
3
The direct product of Mα is
Y
Mα = {(mα )α∈X : mα ∈ Mα }.
α∈X
L Q
Clearly, we have Mα ⊆ Mα .
Example 2.10. Suppose R = Z. Then is M1 ∼
= M2 as modules, where
∞
M ∞
Y
M1 := Z, M2 := Z?
i=1 i=1

The answer is no, since M1 is countable while M2 is uncountable. Therefore M1 and M2

have different cardinality.
Definition 2.11. An R-module M is called free if there exists a set X such that

M∼
M
= R.
i∈X

More intuitively, M is free if there exists a subset B = {mx : x ∈ X} ⊆ M (“basis”) such

that every element of M has a unique expression
X
m= rx mx where rx = 0 for all but finitely many x ∈ X.
x∈X
L
Remark 2.5. To see the equivalence, let ey ∈ R and y ∈ X be the sequence with a 1 in
x∈X
the y-th coordinate and zeroes everywhere else. Then
M X
f: R → M, f ((rx )x∈X ) = rx mx
x∈X x∈X

is bijective.
∞
Q
Question (Hard question). Is Z a free Z-module? (Answer: No. You will prove this in
i=1
Assignment #1.)
Remark 2.6. If R is a field, then every R-module is free (Zorn’s lemma exercise!). But if
R = Z and M = Z/2Z, then M is not free. Note that, since 2 · m = 0 · m for all m ∈ M ,
there can be no unique representation of m. That is, we can get a (non-empty) spanning
set, but it is not linearly independent.
Definition 2.12. We will write that RX :=
L
R and if |X| = n < ∞, then we write
x∈X
n
n X
L
R := R. If R = R , then we call |X| the rank of the free module M .
i=1

Remark 2.7. We cannot answer this yet, but it is indeed true that the rank is well-defined,
i.e., Rn ∼
= Rm implies n = m. Note that the rank is well-defined only when R is commutative.
If R is non-commutative, one can construct an example where R ∼ = R2 .
4
 3. January 09
3.1. Exact sequences.
Definition 3.1. Suppose that M, M 0 , M 00 are R-modules, and let f : M 00 → M and g :
f g
M → M 0 . Then the sequence M 00 → M → M 0 is said to be exact at M if im(f ) = ker(g).
More generally, if
f1 f2 f3 fn
M1 → M2 → M3 → · · · → Mn+1 ,
then the sequence is exact if im(fi ) = ker(fi+1 ) for all i = 1, 2, . . . , n − 1. More specifically,
an exact sequence of the form
f g
0 → M 00 → M → M 0 → 0
is called a short exact sequence.
Remark 3.1. If the sequence
f g
0 → M 00 → M → M 0 → 0 (1)
is short exact, then:
f
(a) 0 → M 00 → M is exact, so f is injective, and im(0 → M 00 ) is )). So we have
(0) = ker(f ), so f is indeed injective.
f g
(b) M 00 → M → M 0 is exact, so im(f ) = ker(g).
g
(c) M → M 0 → 0 is exact, so im(g) = ker(M 0 → 0) = M 0 . Thus g is surjective.
If (1) is short-exact, then M/f (M 00 ) ∼
= M 0 as R-modules, by the first isomorphism theorem.
Indeed, note that g : M → M 0 is surjective, M/ ker(g) ∼= im(g) = M 0 . But ker(g) = im(f ) =
00 00 ∼ 0
f ((M ) so M/f (M ) = M .
Example 3.2. If M and N are R-modules, then the mappings
i : M ,→ M ⊕ N, π2 : M ⊕ N → N
defined as i(m) = (m, 0) and π2 (m, n) = n give a short exact sequence
i π
0 → M → M ⊕ N →2 N → 0.
Example 3.3. Consider the short exact sequence
f g
0 → Z → Z → Z/2Z → 0,
where f (n) = 2n and g(n) = n + 2Z. But note that we have Z ⊕ Z/2Z ∼
6= Z.
3.2. Splitting.
Definition 3.4. A short-exact sequence
g
0 / M 00
f
/ Mo / M0 / 0
τ

is said to split if there exists an R-module homomorphism τ : M → M such that g ◦τ = idM .

Lemma 3.5 (Splitting lemma). If
f g
0 → M 00 → M → M 0 → 0
is a short exact sequence, then the following are equivalent:
5
 (1) there exists an R-module isomorphism θ : M → M 0 ⊕ M 00 such that
(a) θ ◦ f (m00 ) = (0, m00 ) for all m00 ∈ M 00 and
(b) π1 ◦ θ(m) = g(m) for all m ∈ M .
(2) there exists τ : M 0 → M such that g ◦ τ = idM 0 .
(3) there exists σ : M → M 00 such that σ ◦ f = idM 00 .
MO
θ / M 9 0 ⊕ M 00
i
f π1
g % 
M 00 M0
Proof. ((1) ⇒ (2)) Suppose we have the map θ : M → M 0 ⊕ M 00 . We want to construct a
?
map τ : M 0 → M . Do this via τ (m0 ) = σ −1 (m0 , 0). Then g ◦ τ (m0 ) = g(θ−1 (m0 , 0)) = m0 .
But then π1 ◦ θ(m) = g(m), so we can plug in m = θ−1 (m0 , 0). Thus π1 (θ(θ−1 (m0 , 0)) =
g(θ−1 (m0 , 0)) = π1 (m0 , 0) = m0 .
((1) ⇒ (3)) We have θ and want to construct σM → M 00 . Define σ(m) := π2 (θ(m)). Now
need to verify: σ ◦ f = id00M , i.e., we must show that σ(f (m00 )) = m00 , and π2 (θ(f (m0 ))) =
π2 ◦ i(m00 ) = π2 (0, m00 ) = m00 .
((3) ⇒ (1)) We want to define θ : M → M 0 ⊕ M 00 . Define θ(m) = (g(m), σ(m)), where
f
0 / M 00 o / M
g
/ M0 / 0
σ

?
We need π1 (g(m), σ(m)) = π1 (θ(m)) = g(m) and σ◦f (m00 ) = (0, m00 ). But then by exactness,
(g(f (m00 )), σ ◦ f (m00 ) = (0, m00 ).
((2) ⇒ (1)) This time define instead ψ : M 0 ⊕ M 00 → M by ψ(m0 , m00 ) = τ (m0 ) + f (m).
g
0 / M 00
f
/ Mo / M0 / 0
τ

We claim that ψ is an isomorphism. For injectivity, consider ker(ψ). Notice that if τ (m0 ) +
f (m) = 0 and we apply g, then g ◦ τ (m0 ) + g ◦ f (m00 ) = 0. But by exactness, we have
g ◦ f (m00 ) = 0 and g ◦ τ (m0 ) = m0 . Hence m0 = 0 (by injectivity of f ), as required. Now
we need to show surjectivity. Pick m ∈ M . We must show that there exists m00 ∈ M 00
and m0 ∈ M 0 such that m = τ (m0 ) + f (m00 ). Apply g to get g(m) = m0 . Notice that
m − τ (m0 ) = m − τ (g(m)) = f (m00 ) for some m, since m − τ (g(m)) ∈ ker g = im f . So there
exist m0 , m00 such that ψ(m, m00 ) = m.
Now define θ(m) = ψ −1 (m). Check if π1 ◦ ψ −1 (m) = g(m) and ψ −1 ◦ f (m00 ) = (0, m00 ).
This is a straightforward verification. 

4. January 13
Note that if M ∼
= M 00 ⊕ M 0 and
f g
0 → M 00 → M → M 0 → 0
is a short exact sequence, we do not necessarily have a section for g or f .
Example 4.1. If R = Z and M 0 = (Z/2Z)ω , M 00 = Z, M = Z ⊕ (Z/2Z)ω , then clearly
M ∼= M00 ⊕M 0 . Let f : M 00 → M, g : M → M 0 such that f (n) = (2n, 0, 0, 0, 0, . . . ) and
g(n, ε1 , ε2 , ε3 , . . . ) = (n + 2Z, ε1 , ε2 , ε3 , . . . ). Then the given sequence is exact. But note that
6
g does not have a section, i.e., there cannot exist τ : M 0 → M such that g ◦ τ = idM 0 , since
every element in (Z/2Z)ω has order 2, but M is torsion-free (Z cannot have any torsion
element). Notice that g ◦ τ (1, 0, 0, . . . ) = (0, ∗, ∗, ∗, . . . ), so g ◦ τ cannot possibly be idM 0 .
4.1. Structure theorem for modules over a PID.
Definition 4.2. An R-module M is called cyclic if there exists m ∈ M such that M = Rm.
Example 4.3. If R = Z, then N = Z/5Z = h1 + 5Zi is cyclic, but M = Z ⊕ Z is not.
Remark 4.1. If M is cyclic with M = Rm, we have an R-module homomorphism ϕ : R → M
is surjective, and is given by ϕ(r) = rm. Then I := ker ϕ is an ideal of R, and by the fist
isomorphism theorem, R/I ∼ = M as modules. Note that I = AnnR (M ).
Definition 4.4. Let R be a ring and let M be an R-module. Recall that M is a finitely-
generated R-module if there exist d ≥ 1 and m1 , . . . , md ∈ M such that M = Rm1 + Rm2 +
· · · + Rmd .
Theorem 4.5 (Structure theorem for finitely-generated modules over a PID). Let R be a
PID and let M be a f.g. R-module. Then there exists unique d ≥ 0 and some prime elements
π1 , π2 , . . . , πs ∈ R (s ≥ 0 and not necessarily distinct but is unique up to ordering) such that
M∼ = Rd ⊕ R/(π1i1 ) ⊕ · · · ⊕ R/(πsis ).
Proof. We prove it by induction on the number of generators d. Let d = 1 (base case). This
case is immediate since this means M is cyclic. Thus M ∼ = R/I for I = AnnR (M ). We have
∼ 1
two cases. If I = (0), then M = R , the free R-module of rank 1. If I 6= (0), then I = (a)
for some non-zero a ∈ R. Since every PID is a UFD, a has a unique factorization uπ1i1 · · · πsis
where u is a unit and πi ’s are distinct prime elements. So I = (π1i1 · · · πsis ). For k = 1, . . . , s,
let Jk = (πkik ). We claim that Jk ’s are comaximal (i.e., k 6= l implies Jk + Jl = R) and
s
Jk = I. Indeed, if k 6= l then Jk + Jl = (πkik , πlil ) = (b) since R is a PID. Therefore
T
k=1
(b) ⊇ (πkik ) and (b) ⊇ (πlil ). Hence b | πkik and b | πlil . Since πk and πl are distinct, it follows
that b = 1, as required. For the second part of the claim, one direction is easy: note that
s s
Jk , then b ∈ (πkik ) for all k.
T T
since I is contained in Jk for all k we have I ⊆ Jk . If b ∈
k=1 k=1
Hence πkik | b for all k, hence π1i1 · · · πsis | b. Hence b ∈ (π1i1 · · · πsis ) = I.
So by the Chinese remainder theorem,
s s
R/I ∼
Y Y
= R/Ji = R/(πkik )
i=1 k=1
as rings. But this is a stronger condition that being isomorphic as modules. Thus we can
make the analogous statement as modules. This completes the base case.
Now suppose the results holds whenever M is generated by fewer than d elements. Consider
M = hm1 , . . . , md i = Rm1 + Rm2 + · · · + Rmd .
Case 1. r1 m1 + · · · + rd md = 0 implies r1 = r2 = · · · = rd = 0.
In this case, it is immediate that M ∼ = Rd .
Case 2. There exists non-zero (r1 , r2 , . . . , rd ) such that r1 m1 + r2 m2 + · · · + rd md = 0.
Consider the set S consisting of all d-tuples (n1 , . . . , nd ) ∈ M d such that M = Rn1 +
Rn2 + · · · + Rnd . Given (n1 , . . . , nd ) ∈ S , let J(n1 ,...,nd ) := {r ∈ R : rn1 ∈ Rn2 + · · · + Rnd } =
7
Ann(M/(Rn2 + · · · + Rnd )). Notice that if r1 n1 + · · · + rd nd = 0 then r1 n1 = −(r2 n2 +
· · · + rd nd ) ∈ Rn2 + · · · + Rnd . It is not hard to see that J(n1 ,...,nd ) is an ideal. Here comes
the key trick: pick (n1 , n2 , . . . , nd ) ∈ S such that J(n1 ,...,nd ) is maximal in the collection of
ideals {J(n1 ,...,nd ) : (n1 , . . . , nd ) ∈ M d }. Here, the following fact comes in handy: if T is a
non-empty collection of ideals in a PID then there exists J ∈ T that is maximal in T with
respect to ⊇. We will finish the proof in the next lecture. 
5. January 15
Lemma 5.1. If T is a non-empty collection of ideals in a PID then there exists J ∈ T that
is maximal in T with respect to ⊇.
Proof. Let (ai ) ∈ T , and if (ai ) is maximal, we are done. If not, choose (a2 ) so that
(a2 ) ) (a1 ) and if (a2 ) maximal, stop. We see that either we will eventually produce a
maximal element of T or we will produce an infinite ascending chain
(a1 ) ⊆ (a2 ) ⊆ (a3 ) ⊆ · · ·
of ideals in T . Let ∞
[
J := (ai ) = (b)
i=1
(since R is a PID). Then b ∈ J, so there exists j such that b ∈ (aj ), or (b) = (aj ). This is a
contradiction since this implies (b) ⊇ (aj+1 ) ) (aj ) ⊇ (b). 
Proof of Theorem 4.5 (continued). Now that we proved Lemma 5.1, we can finish off the
proof. Pick (n1 , n2 , . . . , nd ) ∈ S if J(n1 ,...,nd ) is maximal in the set {J(v1 ,...,vd ) : (v1 , . . . , vd ) ∈
S }. Note that there exists r ∈ R such that J(n1 ,...,nd ) = (r) since R is a PID. We claim (to
be proved later) that if rn1 + · · · + rd nd = 0, then r | r2 , . . . , r | rd . So write ri = rai for all
2 ≤ i ≤ d and let n01 = n1 + a2 n2 + · · · + ad nd and n0i = ni for all 2 ≤ i ≤ d. Now if we let
N1 := Rn1 ⊆ M and N2 := Rn2 + Rnd ⊆ M , then by induction hypothesis, N1 and N2 both
have a decomposition of the desired form; hence, so does N1 ⊕ N2 .
We clearly have N1 + N2 = M . Thus we only need to check that N1 ∩ N2 = (0) to ensure
that M ∼ = N1 ⊕ N2 . So if a ∈ N1 ∩ N2 , then a ∈ N1 , i.e., a = un01 , and since a ∈ N2 ,
a must be of the form u2 n02 + · · · + ud n0d . Notice that rn01 = r(n1 + a2 n2 + · · · + ad nd ) =
rn1 + ra2 n2 + · · · + rad nd = rn1 + r2 n2 + · · · + rd nd = 0. Therefore, if a ∈ N1 ∩ N2 and a 6= 0,
then r - u. Hence (u, r) ) (r) if a 6= 0. But un01 = u2 n02 + · · · + ud n0d ∈ Rn02 + Rn03 + · · · + Rn0d
and rn01 = 0 ∈ Rn02 + · · · + Rn0d , form which it follows that
J(n01 ,...,n0d ) ⊇ (u, r) ) (r) = J(n1 ,...,nd ) .
But this contradicts the fact that our choice of J(n1 ,...,nd ) is maximal. Now it remains to
prove the claim we initially assumed. 
Lemma 5.2. If rn1 + · · · + rd nd = 0, then r | r2 , . . . , r | rd .
Proof. Suppose that rn1 + r2 n2 + · · · + rd nd = 0 and there exists i > 1 such that r - ri .
Without loss of generality, let i = 2. Let s = gcd(r1 , r2 ) and (s) ) (r). Write r = sa, r2 = sb
with gcd(a, b) = 1. There fore there exist c, d ∈ R such that ca + db = 1. Take the
relation: rn1 + r2 n2 + · · · + rd nd = s(an1 + bn2 ) + r3 n3 + · · · + rd nd = 0. Make a new
spanning set for M : n01 = an1 + bn2 , n02 = −dn1 + cn2 , n03 = n03 , . . . , nd = n0d . From this
we have cn01 − bn02 = (ac + bd)n1 + (bc − bc)n2 = n1 while we have dn01 + an02 = n2 . Now,
8
note that s(an1 + bn2 ) + r3 n3 + · · · + rd nd = 0, hence sn01 + r3 n03 + · · · + rd n0d = 0. Thus
J(n01 ,n02 ,...,n0d ) ⊇ (s) ) (r), contradicting the maximality assumption. 
6. January 15: Introduction to tensor products
Definition 6.1. Let R be a ring and let M, N be two R-modules. A module M ⊗R N is
called the tensor product of M and N over R.
So how do we build this tensor product?
(1) Start by building a free module F with a basis {e(m,n) : (m, n) ∈ M × N }. We will
take a submodule G ⊆ F . G will be the R-submoudle of F spanned by all elements of
the following forms: e(m1 +m2 ,n) − e(m1 ,n) − e(m2 ,n) , e(m,n1 +n2 ) − e(m,n1 ) − e(m,n2 ) , e(rm,n) −
re(m,n) , e(m,rn) − re(m,n) .
(2) Define M ⊗R N := F/G, and define e(m,n) + G =: m ⊗ n.
Remark 6.1. It is important to know that not every element can be expressed as m ⊗ n.
Question. What is (Z/2Z) ⊗Z (Z/3Z)?
Solution: In this case, the free module we need is F = Ze(0,0) ⊕ Ze(0,1) ⊕ Ze(0,2) ⊕ Ze(1,0) ⊕
Ze(1,1) ⊕ Ze(1,2) . We claim in fact that F = G. For any i ∈ Z/2Z and j ∈ Z/3Z, we have
e(i,j) = e(i·1,j·1) = ie(1,j·1) = ije(1,1) ≡ 0 (mod G),
since e(1,1) = e(3,1) = 3e(1,1) = e(1,3) = e(1,0)=0 by bilinearity. Thus F/G = (0).
7. January 16
We defined tensor product last time. Now we talk about the most useful property of tensor
products: universal property. Consider a bilinear map
φ : M × N → M ⊗R N
defined as (m, n) 7→ m ⊗ n = e(m,n) + G ∈ F/G. Note, however, that φ is generally not
surjective.
Proposition 7.1 (Universal property of tensor products). Let M, N, P be R-modules and
suppose that f : M × N → P is R-bilinear. Then
f
M ×N / P
:
φ
 ∃!f˜
M ⊗R N
Then these exists a unique f˜ ∈ Hom(M ⊗R N, P ) such that f˜ ◦ φ = f .
Proof. Suppose F is a free R-module with basis {eα } and if P is an R-module, then any map
φ : {eα } → P extends uniquely to an element φb ∈ Hom(F, P ). So what do we do? For any
R-module homomorphism f : M × N → P , we have a unique homomorphism ψ : F → P
such that e(m,n) 7→ f (m, n). This implies that ψ|G = 0 since
ψ(e(rm,n) − re(m,n) ) = ψ(e(rm,n) ) − rψ(e(m,n) ) = f (rm, n) − rf (m, n) = 0,
with the last equality following from the fact that f is bilinear. This means that we can
define an R-module homomorphism f˜ : F/G → P such that f˜(x + G) = ψ(x). This is
9
well-defined since x + G = y + G ⇔ x − y ∈ G ⇔ ψ(x) = ψ(y). So f˜ ◦ φ(m, n) = f˜(m ⊗ n) =
f˜(e(m,n) + G) = ψ(e(m,n) ) = f (m, n). 
Remark 7.1. Note that M ⊗R N is spanned as an R-module by {m ⊗ n : m ∈ M, n ∈ N }.
Write any x + G ∈ F/G as follows:
X X
x+G= ri e(mi ,ni ) + G → ri mi ⊗ ni ∈ M ⊗R N.
Example 7.2. Compute (Z/2Z) ⊗Z (Z/2Z). In this case F = Ze(0,0) + Ze(0,1) + Ze(1,0) + Ze(1,1) .
As we did last time, e(i,j) ≡ ie(1,j) ≡ ije(1,1) (mod G). So F/G = Z(e(1,1) + G) = Z(1 ⊗ 1).
Since 2(1 ⊗ 1) = 2 ⊗ 1 = 0 ⊗ 1 = 0(1 ⊗ 1) = 0, so F/G is either Z/2Z or (0). We claim that
(Z/2Z) ⊗Z (Z/2Z) = Z/2Z. Let P = Z/2Z. We say that if the tensor product were zero,
then f˜ will have to send everything to zero, but f (1, 1) = 1 6= 0. So the claim follows.
Theorem 7.3 (Uniqueness of tensor product). There exists a unique R-module M ⊗R N
with bilinear φ : M × N → M ⊗R N with respect to having the universal property for all
R-module P and bilinear f : M × N → P .
Proof. Suppose that we have R-modules A and B with bilinear map φ : M × N → A and
ψ : M × N → B with the universal property. So by the universal property, there exists a
unique ψ̃ such that ψ̃ ◦ φ = ψ.
ψ φ
M ×N / B and M × N / A
; ;
φ ψ
 ∃!ψ̃  ∃!φ̃
A B
Similarly, there exists a unique φ̃ such that φ̃ ◦ ψ = φ.
We claim that φ̃ ◦ ψ̃ = idA and ψ̃ ◦ φ̃ = idB . We just need to do one of them since we can
apply the symmetric argument for the other one. We know that the image of φ must span
A. So φ̃ ◦ ψ̃(φ(m, n)) = φ(m, n) for all m, n. Another possible method is to consider the fact
that the following diagram must commute:
φ
M ×N / A
;
φ
 ∃!φ̃◦ψ̃
A
Since φ̃ ◦ ψ̃ = idA works and that must be unique, the claim follows. 
Example 7.4. If m, n ≥ 2, what is Z/mZ ⊗Z Z/nZ? We claim that Z/mZ ⊗Z Z/nZ ∼
= Z/dZ
where d = gcd(m, n). Consider the commutative diagram
ψ
Z/mZ × Z/nZ / Z/dZ
7
φ

Z/mZ ⊗Z Z/nZ
ψ : (a, b) 7→ (ab) is bilinear and onto. The tensor product is spanned by i ⊗ j = ij(1 ⊗ 1).
But then d(1 ⊗ 1) = (am + bn)(1 ⊗ 1) = am ⊗ 1 + bn ⊗ 1 = 0. Hence the claim follows.
10
Proposition 7.5 (Properties of tensor product). The following hold:
(1) (commutativity) M ⊗R N ∼ = N ⊗R M
(2) (associativity) (M ⊗R N ) ⊗R P ∼
= M ⊗R (N ⊗R P ) ∼ = M ⊗R N ⊗R P

8. January 20
Proposition 8.1. M ⊗R N ∼
= N ⊗R M .
Proof. Consider the following commutative diagram:
f
M ×N
τ / N ×M / P
:
ψ
φ &  ∃![
f ◦τ
M ⊗R N

where τ : (m, n) 7→ (n, m) and ψ : (n, m) 7→ m ⊗ n. Note f[

◦ τ ◦ ψ = f.
So M ⊗R N satisfies the universal property for N ×M with ψ : N ×M → M ⊗R N bilinear.
Therefore M ⊗R N ∼ = N ⊗R M . 

8.1. Tensor product of maps.

Definition 8.2. If f : M → P and g : N → Q are homomorphisms, then we can make a
bilinear map ψ : M × N → P ⊗R Q such that (m, n) 7→ f (m) ⊗ g(n). This map f ⊗ g is said
to be the tensor product of f and g.
This map is bilinear, since
ψ
(rm1 + m2 , n) 7→ f (rm1 + m2 ) ⊗ g(n)
= (rf (m1 ) + f (m2 )) ⊗ g(n)
= rf (m1 ) ⊗ g(n) + f (m2 ) ⊗ g(n) = rψ(m1 , n) + ψ(m2 , n).

So if ψb : M ⊗R N → P ⊗R Q is defined as ψ(m ⊗ n) = f (m) ⊗ g(n), then ψb ◦ φ = ψ. In

b ⊗ n) = f (m) ⊗ g(n).
particular, ψ(m, n) = ψ(m
It is customary to let f ⊗ g denote ψ. b So in summary, if f ∈ Hom(M, P ) and g ∈
Hom(N, Q) then there exists a unique f ⊗g ∈ Hom(M ⊗N, P ⊗Q) such that (f ⊗g)(m⊗n) =
f (m) ⊗ g(n).
One important special case is the following: if f : M → N , then f ⊗ idC : M ⊗ C → N ⊗ C
is (f ⊗ idC )(m ⊗ c) = f (m) ⊗ c.
f g
Theorem 8.3. If M → N → P → 0 is exact and C is an R-module, then the sequence
f ⊗id g⊗id
M ⊗R C →C N ⊗R C →C P ⊗R C → 0
is exact also.
One can check that g ⊗ idC is onto. Note that P ⊗R C is generated by p ⊗ c ↔ g(n) ⊗ c ↔
(g ⊗ idC )(n ⊗ c). In general, however 0 → M → N being exact does not imply that
0 → M ⊗ C → N ⊗ C is exact.
11
Example 8.4. Let M = N = Z and C = Z/2Z, and R = Z. If f : M → N such that
f (m) = 2m, then indeed f is injective. However,
f ⊗id
0 → Z ⊗ Z/2Z →C Z ⊗ Z/2Z
is not exact: f ⊗ idC is the zero map, since (f ⊗ idC )(n ⊗ ε) = 2n ⊗ ε = n ⊗ 2ε = n ⊗ 0 = 0.
Definition 8.5. An R-module is flat if 0 → A ⊗ M → B ⊗ M is exact whenever 0 → A → B
is exact. M is faithfully flat if M is flat, and A ⊗R M 6= (0) whenever A 6= (0).
Example 8.6. Q is faithful and flat as a Z-module but is not faithfully flat. Recall that
Q ⊗Z Z/2Z = (0).
Proposition 8.7. Let R be a ring and M an R-module. Then R ⊗R M ∼ = M.
Proof. Let φ : R × M → M given by (r, m) 7→ rm. It is easy to verify that this is bilinear.
Suppose that R × M → P is bilinear.
ψ
R×M / P
;
φ
 ∃!ψ
e
M
Our goal is to find a unique homomorphism ψe such that ψ◦φ
e = ψ. Since ψ is bilinear, we have
?
ψ(r, m) = ψ(r · 1, m) = rψ(1, m). We want ψ(m)
e = ψe ◦ φ(1, m) = ψ(1, m). This (uniquely
determined) ψe is a homomorphism, since ψ is bilinear: ψ(rm
e 1 + m2 ) = ψ(1, rm1 + m2 ) =
rψ(1, m1 ) + ψ(1, m2 ) = rψ(m1 ) + ψ(m2 ).
e e 
8.2. Direct sums.
Theorem 8.8. If {Mα }α∈I is a collection of R-modules and C is an R-module, then
!
Mα ⊗R C ∼
M M
= (Mα ⊗R C).
α∈I α∈I

Proof. We will prove the two direct sum case (M ⊕N )⊗C ∼= (M ⊗C)⊕(N ⊗C), since the same
argument can be extended for the general case. Let π1 : M ⊕ N → M and π2 : M ⊕ N → N
be the projection maps. We will make linear maps π1 ⊗ id : (M ⊕ N ) ⊗ C → M ⊗ C
and π2 ⊗ id : (M ⊕ N ) ⊗ C → N ⊗ C and construct (π1 ⊗ id, π2 ⊗ id) : (M ⊕ N ) ⊗ C →
(M ⊗ C) ⊕ (N ⊗ C). Let i1 : M ,→ M ⊕ N and i2 : N ,→ M ⊕ N be inclusion maps
i1 (m) = (m, 0) and i2 (n) = (0, n). The inclusion maps give maps i1 ⊗ id : M ⊗ C →
(M ⊕ N ) ⊗ C and i2 ⊗ id : N ⊗ C → (M ⊕ N ) ⊗ C. The two maps extend to a map
(M ⊗ C) ⊕ (N ⊗ C) → (M ⊕ N ) ⊗ C such that (i1 ⊗ id)(m ⊗ c, n ⊗ c0 ) = i1 (m) ⊗ c. So we
get a map h : (M ⊗ C) ⊕ (N ⊗ C) → (M ⊕ N ) ⊗ C where h = i1 ⊗ id +i2 ⊗ id. Notice
(π1 ⊗ id, π2 ⊗ id) ◦ h(m ⊗ c, n ⊗ c0 ) = (π1 ⊗ id, π2 ⊗ id)[(m, 0) ⊗ c + (0, n) ⊗ c0 ]
= (m ⊗ c + 0 ⊗ c0 , 0 ⊗ c + n ⊗ c0 ) = (m ⊗ c, n ⊗ c0 ).
Hence (π1 ⊗ id, π2 ⊗ id) ◦ h = id(M ⊗C)⊕(N ⊗C) . For the other way,
h ◦ (π1 ⊗ id, π2 ⊗ id)((m, n) ⊗ c) = h(m ⊗ c, n ⊗ c)
= (m, 0) ⊗ c + (0, n) ⊗ c = (m, n) ⊗ c.
12
Therefore h ◦ (π1 ⊗ id, π2 ⊗ id) = id(M ⊕N )⊗C . Notice that this is enough to prove that it is
identity since we proved the identity over all the generating elements, which implies that it
is the identity over all elements. 

9. January 22
Corollary 9.1. If RX ∼
= RY (X, Y index sets, R a commutative ring), then |X| = |Y |.
Proof. Let P be a maximal ideal of R. Let F = R/P . Then RX ∼ = RY so RX ⊗R F ∼ =
X X X ∼ Y X ∼ Y
(R/P ) = F . Therefore R = R ⇒ F = F as R-modules. Recall the following fact:
Claim. If M and N are isomorphic R-modules then Ann(M ) = Ann(N ) = I is an ideal of
R. Then M and N are isomorphic as R/I-modules.
Proof of the claim. Let φ : M → N be an R-module isomorphism, and let r = r + I ∈ R/I.
Create an R/I-module isomorphism φ(m) = φ(m) and φ(r · m) = φ(rm) = rφ(m) = rφ(m).
One can check that φ is well-defined and bijective. 
Now we consider the annihilators of F X . Note that F X ∼ = (R/P )X , Since Ann(F X ) =
Ann(F Y ) = P and since F X ∼= F Y as R-modules, ist follows that F X ∼
= F Y as F -modules.
Therefore |X| = |Y |. 
9.1. Algebras, base change, and extension of scalars.
Definition 9.2. Let R be a ring. Then an R-algebra S is just a ring equipped with a ring
homomorphism α : R → S satisfying α(1R ) = 1S .
Example 9.3. Every ring is a Z-algebra, since α : Z → R given by α(n) = n · 1R is a ring
homomorphism.
Example 9.4. C[x, y] is a C-algebra, with α : C → C[x, y] being defined to be α(c) = c, where
c is a constant polynomial.
Example 9.5. C is an R-algebra, with α : R → C given by α(c) = c.
Notice that if S is an R-algebra then S inherits an R-module structure as well. Given
r ∈ R, s ∈ S, define r · s := α(r)s ∈ S, where α : R → S is a ring homomorphism such that
α(1R ) = 1S . Notice also then that if S is an R-algebra and M is an R-module then we can
form the tensor product S ⊗R M which is another R-module.
But note also that we can endow S ⊗R M with an S-module structure, via the rule
s1 · (s2 ⊗ m) := s1 s2 ⊗ m,
and extend linearly, i.e., (s1 + s2 )(s0 ⊗ m) = s1 (s0 ⊗ m) + s2 (s0 ⊗ m).
So we started with an R-module M for some ring R; and then for some R-algebra S, we
created an S-module S ⊗R M , hence “extending” scalars. This process is therefore called
either base change or extension of scalars. So why is this interesting?
Remark 9.1. As a motivating example, imagine that V is a Q-vector space. And it is easy
to see that C is a Q-algebra. Let VC := V ⊗Q C is a C-vector space. If V ∼= Qn , then
n
V ⊗Q C ∼ ⊕ · · · ⊕ Q) ⊗Q C ∼
= (Q ⊗Q C) ∼
= Cn .
L
= (Q| {z } i−1
n times
13
Remark 9.2 (A useful construction). If A and B are R-algebras then we can form A ⊗R B,
which is an R-module. But it actually has the structure of an R-algebra. For instance, if
we let (a1 ⊗ b1 ) · (a2 ⊗ b2 ) = a1 a2 ⊗ b1 b2 , and α defined to be α(r) ⊗ 1 = 1 ⊗ β(r), define
γ : R → A ⊗R B to be γ(r) = α(r) ⊗ 1 = 1 ⊗ β(r). This gives us the R-algebra structure.
9.2. Noetherian rings.
Definition 9.6. Let R be a ring. We say R is Noetherian if every ascending chain of ideals
of R terminates. That is, if I1 ⊆ I2 ⊆ · · · is an ascending chain, then there exists n such
that In = In+1 = In+2 = · · · .
Example 9.7. Every PID is Noetherian. Also, a field is Noetherian, since it only has two
ideals: itself and the zero ideal. Alternately, you can observe that a field is a PID, hence
Noetherian.
Example 9.8. Let R = C[x1 , x2 , . . . ]. Then the following chain
(x1 ) ( (x1 , x2 ) ( (x1 , x2 , x3 ) ( · · ·
is an ascending chain of infinite length. Hence R is not Noetherian.
Definition 9.9. Let R be a ring and let M be an R-module. THen M is Noetherian if every
ascending chain of submodules of M terminates.
Remark 9.3. R is Noetherian as a ring if and only if R is Noetherian as an R-module.
Example 9.10. If R = Z and M = Q, then Q is not Noetherian as a Z-module, since the
following ascending chain does not terminate: Z · 1 ( Z · 12 ( Z · 14 ( Z · 18 ( · · · .
Proposition 9.11. Let R be a ring. And the the following statements are equivalent:
(1) R is Noetherian
(2) Every ideal I is finitely generated
(3) Every non-empty collection of S of ideals has a maximal element with respect to ⊆.
Remark 9.4. One can prove analogous statements for modules in a similar manner.
Proof. ((2) ⇒ (3)) Let S be a non-empty collection of ideals, and if S does not have S a
maximal element, then we can produce a chain I1 ( I2 ( I3 ( · · · . Now let J = Ii .
By (2), we have J = (a1 , a2 , . . . , as ). For all i ∈ {1, 2, . . . , s}, there exists ni such that
ai ∈ Ini . Let N = max(n1 , n2 , . . . , ns ), so J = (a1 , . . . , as ) ∈ In ( J, or In = J. But this is a
contradiction. S
((2) ⇒ (1)) Same idea: start with I1 ( I2 ( · · · then let J = Ii cannot be finitely-
generated by the same reason.
(¬(2) ⇒ ¬(3)) Let J = (a1 , a2 , a3 , . . . ) be an ideal that is not finitely generated and
ai+1 ∈∈ / (a1 , . . . , ai ) for all i. Let S = {(a1 , . . . , ai ) : i ≥ 1}. Then S has no maximal
element, since (a1 , . . . , an ) ( (a1 , . . . , an+1 ). 

14
 10. January 23
Proposition 10.1. Let R be a ring and let M be an R-module. Let N be a submodule of
M . Then M is Noetherian if and only if N and M/N are Noetherian.
Proof. Let π : M → M/N be the canonical surjection. Recall that by the correspondence
theorem, the map P ⊂ M 7→ π(P ) ⊂ M/N gives an inclusion-preserving bijection between
submodules of M/N and submodules of M that contain N .
(⇒) If N1 ⊆ N2 ⊆ · · · is a chain in N , then it is also a chain in M . Therefore it must
terminate. Similarly, if A1 ⊆ A2 ⊆ · · · is a chain in M/N then π −1 (A1 ) ⊆ π −1 (A2 ) ⊆
π −1 (A3 ) · · · is a chain in M . Therefore it must terminate also. The claim now follows.
(⇐) Suppose that N and M/N are Noetherian, and let M1 ⊆ M2 ⊆ · · · are Noetherian.
Then M1 ∩ N ⊆ M2 ∩ N ⊆ · · · is a chain in N . So there exists m such that Mm ∩ N =
Mm+1 ∩ N = · · · . Also, π(M1 ) ⊆ π(M2 ) ⊆ · · · is a chain in M/N so there exists some
p such that π(Mp ) = π(Mp+1 ) = · · · . Let n = max(m, p). We claim that this implies
Mn = Mn+1 = Mn+2 = · · · .
Let x ∈ Mn+i . We know that Mn+i ⊇ Mn . It is enough to show that x ∈ Mn . Since
π(Mn+i ) = π(Mn ), there exists y ∈ Mn such that π(x) = π(y). Then π(x − y) ≡ 0, so
x − y ∈ N ∩ Mn+i = N ∩ Mn . So x − y = z ∈ Mn so x = y + x ∈ Mn . 
Corollary 10.2. If M and N are Noetherian, then M ⊕ N is Noetherian.
Proof. Note that (M ⊕ N )/M ∼
= N , since M ∼
= M ⊕ (0). 
Example 10.3. Note that the infinite
L direct sums of Noetherian modules need not be Noe-
therian. Let R = Z and M = Z. Then M is not finitely generated as a Z-module, so M
cannot be Noetherian.
Corollary 10.4. If R is Noetherian and M is a finitely-generated R-module, then M is
Noetherian.
Proof. Prove by induction in the number of generators d. If d = 0 then the claim is evident.
Let d = 1. Then there exists m ∈ M such that M = Rm. Then r 7→ rm is a surjective
map, and by the first isomorphism theorem M ∼ = R/I. Then M is Noetherian since R is
Noetherian as an R-module and R/I is a quotient.
So assume that this is true for all d < n. Let M = hm1 , . . . , mn i. Then N = hm1 , . . . , mn−1 i.
By the induction hypothesis, N is Noetherian and M/N = R(m+N ), so M/N is Noetherian.
Therefore M is Noetherian as well. 
10.1. Maximality principle.
Meta-theorem. If R is a Noetherian ring and one chooses an ideal I in R that is maximal
with respect to some “nice” property, then I is a prime ideal.
Definition 10.5. Recall that we define the product of the two ideals I and J to be
( d )
X
IJ := it jt : d ≥, i1 , . . . , ik ∈ I, j1 , . . . , jk ∈ J ⊆ R.
i=1

15
Theorem 10.6 (Noether). Let R be a Noetherian ring and I be a proper ideal of R.
Then there exist m ≥ 1 and prime ideals P1 , . . . , Pm (not necessarily distinct) such that
P1 P2 P3 · · · Pm ⊆ I.
Proof. Suppose the statement is not true. Let S be the collection of proper ideals that do
not contain a finite product of prime ideals. By assumption, S 6= ∅. Pick I ∈ S maximal.
We will show that I must be prime. And then we will let m = 1 and P1 = I hence Pi ⊆ I,
which is a contradiction.
Suppose I is not prime. Then there must exist a, b ∈ R \ I such that ab ∈ I. Now let J1 :=
I + Ra and J2 := I + Rb. Notice that I ( J1 and I ( J2 . Also note that J1 J2 = (I + Ra)(I +
Rb) ⊂ I + Rab ⊆ I. By maximality, J1 , J2 ∈ / S . So there exist P1 , . . . , Pm , Q1 , . . . , Qn prime
ideals such that P1 P2 · · · Pm ⊆ J1 and Q1 Q2 · · · Qn ⊆ J2 . Therefore P1 P2 . . . Pm Q1 . . . Qn ⊆
J1 J2 ⊆ I. 
Remark 10.1. Without loss of generality, we may choose the Pi so that they contain I. If I
is a proper ideal of R and R is Noetherian and P1 P2 · · · Pm ⊆ I and P1 , P2 , . . . Pm ⊇ I, then
if Q is a prime ideal and Q ⊇ I then Q ⊇ Pi .
Assume not. Then for i = 1, 2, . . . , m,there exist ai ∈ Pi \ Q so a1 a2 · · · am ∈ P1 · · · Pm ⊆
I ⊆ Q. This is a contradiction: note that a1 · · · an = 0 in R/Q so there exists j such that
aj = 0 hence aj ∈ Q.
So there are only finitely many primes in R that are minimal with respect to containing
I.
Definition 10.7. Let R be a ring and I an ideal of R. Then the radical of I is
√ \
I := P.
P ⊃I
P prime
p
Example 10.8. If R = Z/4Z, then what is (0)? Note that out ofp three ideals (0), 2(Z/4Z),
4(Z/4Z), we see that 2(Z/4Z) is the only prime ideal. Therefore (0) = 2(Z/4Z).

11. January 27
√
Note that if R is Notherian, then I = P1 ∩ · · · ∩ Ps , i.e., there are only finitely many
ideals containing I.
p p
Theorem 11.1. Let R be a ring. Then (0) is nil ideal, i.e., if x ∈ (0) then there exists
n = n(x) ≥ 1 such that xn = 0.
p
Proof. Suppose not. THen there exists x ∈ I := (0) such that x is not nilpotent. Let
/ T . Let S = {J : J ideal, J ∩ T = ∅}. Then S 6= ∅ because
T = {1, x, x2 , x3 , · · · }. Then 0 ∈
(0) ∩ T = ∅, meaning (0) ∈ S .
If R is Noetherian, we just let J be a maximal element of S . If not, then we use Zorn’s
lemma to produce a maximal element. Namely, let {Jα } be a chain of S and look at Jα .
T
Think about why this is in S . Therefore Zorn’s lemma gives a maximal element of S .
Let J be a maximal element of S . We claim that J is prime. To see this, if we assume that
J is not prime, then there must exist a, b ∈ R \ J such that ab ∈ J. Write J1 := J + Ra and
J2 := J + Rb. Since J is maximal in S , and J ( J1 , J2 , we see that J1 , J2 ∈ / S . Thus there
16
must exist n1 , n2 ≥ 1 such that xn1 ∈ J1 and xn2 ∈ J2 . Then xn1 +n2 ∈ J1 J2 ⊆ J + Rab ⊆ J,
but this contradicts the fact that J ∩ T = ∅.
Therefore J ∈ S is prime. But this is a contradiction! Recall that
\ p
J⊇ P = (0),
P ⊇(0)
P prime

since J is a prime; therefore x ∈ J. The claim follows. 

√
Corollary 11.2. If I is an ideal of R and x ∈ I, then there exists n = n(x) ≥ 1 such that
xn ∈ I.
p
Proof.√Let S := R/I and
√ apply the correspondence theorem to see that (0) ∈ S is matched
with I in R. If x ∈ I then (x + I)n = 0 in S, so xn ∈ I. 
Proposition 11.3. If R is a ring, then a0 + a1 x + · · · + an xn ∈ R[x] is a unit in R[x] if and
only if a0 is a unit in R and a1 , . . . , an are nilpotent.
Proposition 11.4. Let R be a Noetherian ring and I a nil ideal of R. Then I is nilpotent,
i.e., I n = (0) for some n ≥ 1.
Proof. Suppose otherwise. Let S = {J : J ideal of R, π(I) := (I + J)/J is not nilpotent in
R/J, π : R → R/J canonical projection}. Since (0) ∈ S , then S 6= ∅. Let J ∈ S be a
maximal element. We claim that J is prime. Otherwise, then we can find a, b ∈ R \ J such
that ab ∈ J. Let J1 = J + Ra, J2 = J + Rb. By maximality of J, neither J1 nor J2 can be
in S . Hence π(I) in R/Ji is nilpotent for i = 1, 2.
(Aside: What does it meant to say that π(I) is in R/L, where π : R → R/L? Note that
π(I) is nilpotent if and only if π(I)n = (0) if and only if (I + L)n /L = L/L if and only
if I n ⊆ L. Therefore there exist n1 , n2 ≥ 1 such that I n1 ⊆ J1 and I n2 ⊆ J2 , from which
I n1 +n2 ⊆ J1 J2 ⊆ J follows.)
Therefore we produced J ∈ S that is also a prime. But then we see that π(I) is a nil
ideal but not a nilpotent ideal in R/J (integral domain). But the only nilpotent element in
R/J is 0, meaning π(I) ⊆ (0) and this is a contradiction. The result follows. 
Theorem 11.5 (Hilbert’s basis theorem). If R is Noetherian, then R[x] is Noetherian also.
Remark 11.1. Subring of a Noetherian ring need not be Noetherian! Note that R =
C(x1 , x2 , . . . ) is Noetherian since R is a field, but within R lies a polynomial ring C[x1 , x2 , . . . ].
The polynomial ring is not Noetherian.
Lemma 11.6. If S is Noetherian, then so is S/I (correspondence).
Corollary 11.7. The converse of Hilbert’s basis theorem holds also. Just let S := R[x] and
I := (x).
Corollary 11.8. If R is a Noetherian, then R[x1 , x2 , . . . , xs ] is Noetherian also (by induc-
tion). Therefore R[x1 , . . . , xs ]/I is Noetherian.
Definition 11.9. A ring of the form R[x1 , . . . , xs ]/I is called a finitely generated R-algebra.
Particularly, if k is a field, then a finitely generated k-algebra is Noetherian.
Notation. Given p(x) = p0 + p1 x + · · · + pn xn ∈ R[x] with pn 6= 0, define in(p(x)) = pn ∈ R.
17
Proof. Let I be an ideal of R[x]. We will show that I is finitely generated as an ideal.
Pick f1 (x) ∈ I \ {0} of smallest degree. Then let d1 = deg f1 (x); a1 = in(f1 (x)); and let
J1 = a1 R. Let I1 = f1 (x)R(x) ⊆ I. If I = I1 , we are done. Otherwise, choose f2 (x) ∈ I \ I1
with minimal degree. Let d2 = deg f2 , and let a2 = in(f2 ). Let J1 ⊆ J2 := a1 R + a2 R
and I2 = f1 (x)R[x] + f2 (x)R[x] ⊆ I. Note that d2 ≥ d1 because we picked f1 to be the
polynomial of the smallest degree. The idea is that this process should stop at some point,
and we will use that fact to show that Ji must terminate also. 

12. January 29
Theorem 12.1 (Hilbert basis theorem). If R is Noetherian, then R[x] is Noetherian also.
Proof. Let I be an ideal of R[x]. We will show that I is finitely generated as an ideal.
Without loss of generality let I be a non-zero ideal.
Step 1. Pick f1 (x) = a1 xd1 + lower degree terms ∈ I \ {0} of minimal degree d1 . Note
a1 = in(f1 ) ∈ R; then let I1 = f1 (x)R[x] ⊆ I, and define J1 = a1 R.
Step 2. If I = I1 , stop. We are done by letting I = (f1 (x)). Otherwise, pick f2 (x) =
a2 xd2 + lower degree terms ∈ I \ I1 of minimal degree. Let I2 = f1 (x)R[x] + f2 (x)R[x], and
let J2 = a1 R + a2 R. Recall that d2 ≥ d1 . In general, we keep going in order to produce
f1 (x), . . . , fn−1 (x) ∈ R[x], a1 , . . . , an−1 ∈ R, d1 , . . . , dn−1 ∈ N; and let In−1 := f1 (x)R[x] +
· · · + fn−1 (x)R[x] and Jn−1 = a1 R + a2 R + · · · + an−1 R, where I1 ⊆ I2 ⊆ · · · ⊆ In−1 ⊆ I and
J1 ⊆ J2 ⊆ · · · ⊆ Jn−1 ⊆ R.
If I = In−1 , we can stop: I is finitely generated. Otherwise, pick fn (x) = an xdn +· · ·+a0 ∈
I \ In−1 of minimal degree dn . As before, we have d1 ≤ d2 ≤ · · · ≤ dn−1 ≤ dn , with
In = In−1 + fn (x)R[x] and Jn := Jn−1 + an R. If there exists m such that I = Im , then we
have that I is generated by (f1 (x), . . . , fn (x)), and we are done. Thus we may assume that
I1 ( I2 ( I3 ( · · · . But since R is Noetherian, the chain J1 ⊆ J2 ⊆ · · · must terminate
in R. Hence there must exist m such that Jm = Jm+1 = · · · . We claim that this implies
Im = Im+1 , thereby deriving a contradiction.
Recall that we picked fm+1 (x) ∈ I \ Im of minimal degree so that deg fm+1 ≥ deg fm . Now
am+1 ∈ Jm+1 = Jm = a1 R + · · · + am R. So there exists r1 , . . . , rm ∈ R such that am+1 =
r1 a1 + · · · + rm am . Since r1 f1 (x), r2 f2 (x), · · · rm fm (x) ∈ Im Note that dm+1 ≥ d1 , . . . , dm . So
fm+1 (x) − r1 xdm+1 −d1 f1 (x) − r2 xdm+1 −d2 f2 (x) − · · · − rm xdm+1 −dm fm (x) =: g(x). Hence g(x)
has degree at most dm+1 − 1. Note that g(x) ∈ Im . Since fm+1 (x) ∈ I \ Im was chosen as an
element of smallest degree in I \ Im and deg(g(x)) < dm+1 = deg(fm+1 ) we have g(x) ∈ Im .
d −d
But this implies that fm+1 (x) = g(x) + r1 m+1 1 f1 (x) + · · · + rm fm (x)xdm+1 −dm ∈ Im , a
contradiction. Hence I = Im . 

12.1. Jacobson radicals.

Definition 12.2. Let R be a ring. We define the Jacobson radical of R to be
\ \ p
J(R) := M⊇ = (0).
M maximal ideal P prime
T
Example 12.3. J(Z) = pZ = (0). In this case, the nilradical and Jacobson radical are
p prime
equal.
18
Example 12.4. Let R = {ab−1 : a ∈ Z, b ∈ Z, b ≡ 1 (mod 2)}. Then the maximal ideal
of R is P = {ab−1 : a ≡ 0 (mod 2), b ≡ 1 (mod 2)}. Then we have a homomorphism
φ : R → Z/2Z defined by φ(ab−1 ) = a + 2Z = ab−1 + 2Z, and ker φ = P . In fact, P is the
only maximal ideal of R. If Q ( R is a proper ideal, then Q ⊆ P . Suppose otherwise. Then
there exists ab−1 ∈ R \ P . We know that a must be odd since ab−1 ∈/ P . But then ba−1 ∈ R
−1 −1 −1
so ab is a unit. Therefore 1 = (ab )(ba ) ∈ R(a/b) ⊆ Q. So P is the maximal ideal So
J(R) = P = 2R.
Now we prove some propositions that give alternate definitions of Jacobson radicals
Proposition 12.5. x ∈ J(R) ⇔ 1 + ax is a unit of R for all a ∈ R.
Proof. (⇒) Suppose that x ∈ J(R) and a ∈ R. Then ax ∈ J(R). Suppose that 1 + ax is not
a unit. Then R(1 + ax) ( R. So there must exist a maximal ideal M such that 1 + ax ∈ M .
But since ax ∈ J(R), by definition ax ∈ M . Therefore 1 ∈ M . Contradiction!
(⇐) Suppose that 1 + ax is a unit for all a ∈ R but x ∈
/ J(R). Then there must exist a
maximal ideal M such that x ∈ / M . So x + M ∈ R/M , and x + M 6= 0 + M . Clearly R/M
is a field. So there must exist a + M ∈ R/M so that (−a + M )(x + M ) = 1 + M . Therefore
(1 + ax) + M = 0 + M , so 1 + ax ∈ M . Contradiction! 
Definition 12.6. A ring R is called a Jacobson ring if for every prime ideal P of R we have
J(R/P ) = (0).
Example 12.7. Any field is a Jacobson ring. Z is another example of a Jacobson ring.

13. January 30
Remark 13.1. If R is a Jacobson ring, then by the correspondence theorem
\
M = P,
M ⊃P
M maximal
so \ \
M= P.
M maximal P prime

Theorem 13.1 (Nakayama’s lemma). Let R be a ring and M be a finitely generated R-

module. If J(R)M = M , then M = (0).
Example 13.2. Let R = {f (x)/g(x), f (x), g(x) ∈ C[x], g(0) 6= 0} be a ring. Then J(R) = xR,
and xR is a maximal ideal of R and xR is the maximal ideal. Since the map givens by
f (x)/g(x) 7→ f (0)/g(0) is a surjective homomorphism with kernel xR, by the first isomor-
phism theorem we have R/xR ∼ = C. J(R) = xR since xR is the unique maximal ideal.
Now consider M = C(x), and r ∈ R, f (0)/g(0) ∈ M . We say that, if rf (x)/g(x) ∈ M
then J(R)M = M .
If a(x)/b(x) ∈ M then a(x)/b(x) = x[a(x)/(b(x)x)]. Therefore J(R)M = M . So we really
need M to be finitely generated.
Remark 13.2 (Usefulness of Nakayama’s lemma). If R is a local ring with the unique maximal
ideal P and if M is a finitely-generated R-module then M/P M is an R/P -module. Since
R/P is a field, we can view M/P M as an R/P (= F )-vector space.
19
 If m1 , . . . , md ∈ M have the property that m1 , . . . , md ∈ M/P M form a basis for M/P M
as an F -vector space, then M = Rm1 + Rm2 + · · · + Rmd . We will try to prove this claim.
Let N = Rm1 + · · · + Rmd , which is indeed a submodule of M . Define A = M/N .
Then J(R)A = P A = P (M/N ) = (P M + N )/N . But P M + N = M because M/P M =
hm1 , . . . , md i meaning that M/P M = N , so equivalently P M + N = M . So by Nakayama’s
lemma, A = 0. Therefore M = N .
Proof of Theorem 13.1. Suppose that M is finitely generated and J(R)M = M with M 6=
(0). Then because M is finitely generated and non-zero, there exists d ≥ 1 and m1 , . . . , md ∈
M such that M = Rm1 +·+Rmd . Morevover, we may assume that d is minimal with respect
to there being a generating set of size d.
J(R)M = M , so M = {j1 m1 + · · · + jd md : j1 , . . . , jd ∈ J(R)}. We have md = j1 m1 + · · · +
jd md for some j1 , . . . , jd ∈ J(R). Write (1 − jd )md = j1 m1 + · · · + jd−1 md−1 . But then 1 − jd
is a unit, so md ∈ (1 − jd )−1 j1 m1 + · · · + (1 − jd )−1 jd−1 md−1 ∈ Rm1 + · · · + Rmd−1 . Hence
M = Rm1 + Rm2 + · · · + Rmd−1 , but this contradicts the minimality of d. So M = (0). 

13.1. Localization.
Example 13.3. We start with a specific example, to give an idea on what localization is
about. Think of R = Z and S = Z \ {0}. Then Q = S −1 R = {(a, b) : a ∈ R, b ∈ S}/ ∼
where (a, b) ∼ (c, d) iff a/b = c/d. This is an example of localization.
Definition 13.4. Let R be a ring. A subset S ⊆ R \ {0} is said to be multiplicatively closed
if:
• 1∈S
• s1 , s2 ∈ S ⇒ s1 s2 ∈ S
Given a ring R and a multiplicatively closed subset S ⊆ R, we can define the localization of
R with respect to S, which we denote S −1 R.
As a set, S −1 R = R × S/ ∼ where (r1 , s1 ) ∼ (r2 , s2 ) iff there exists some s0 ∈ S such that
0
s (s1 r2 − s2 r1 ) = 0. We claim that ∼ is an equivalence class. We will only prove transitivity
since the other two are evident. Suppose that (r1 , s1 ) ∼ (r2 , s2 ) and (r2 , s2 ) ∼ (r3 , s3 ). So
there exist s0 , s00 ∈ S such that
s0 (s2 r1 − s1 r2 ) = 0 (2)
s00 (s3 r2 − s2 r3 ) = 0. (3)
So we need to construct s000 so that s000 (s3 r1 − s1 r3 ) = 0:
s00 s3 × (2) − s1 s0 × (3) = s00 s3 s2 r1 − s00 s3 s0 s1 r2 + s00 s3 s0 s1 r2 − s00 s2 sa s0 r3
= s0 s2 s00 (r1 s3 − r3 s1 ) = 0.
So letting s000 = s0 s2 s00 proves the claim.
We see that R × S/ ∼ is a ring, with addition and multiplication being
s−1 −1 −1
1 r1 · s2 r2 = (s1 s2 ) (r1 r2 )

s−1 −1 −1
1 r1 + s2 r2 = (s1 s2 ) (r1 s2 + r2 s1 ).

You can check if it’s well-defined and satisfies ring axioms.

20
 14. February 3
We know that Z ⊂ Q,but it is not necessarily true in general that R embeds into S −1 R.
Consider the following example:
Example 14.1. Let R = Z × Z and S = ((Z \ {0}) × {0}) ∪ {(1, 1)}. Then S is multiplicatively
closed, and 0 ∈ / S. Let (a, b), (c, d) ∈ R and (s, t), (s0 , t0 ) ∈ S. Then ((a, b), (s, t)) ∼
((c, d), (s0 , t0 )) ⇔ ∃(v, 0) ∈ S such that (v, 0) · (a, b) · (s0 , t0 ) = (v, 0) · (c, d) · (s, t). This
happens iff as0 = cs. Thus, S −1 R → Q, with embedding [(a, b), (s, t)] 7→ as−1 .
In particular, R → S −1 R defined by r 7→ 1−1 r is not an embedding since the map is not
injective. To see why, note that (0, b) is in the kernel for any b ∈ Z.
Definition 14.2. We say that S is regular if S does not contain zero divisors.
Proposition 14.3. If S is regular, then R → S −1 R defined as r 7→ 1−1 r is an embedding.
Proof. If 1−1 r = 1−1 r0 , then there must exist s ∈ S such that s(r − r0 ) = 0. But since S is
regular, s cannot be a zero-divisor. Hence r = r0 . 
Remark 14.1. If I ⊂ R is an ideal and S ⊂ R is multiplicatively closed, then S −1 I = {s−1 r :
s ∈ S, r ∈ R} is an ideal of S −1 R.
14.1. Universal property of localization.
Theorem 14.4 (Universal property of localization). Let R be a ring and S be a multi-
plicatively closed regular subset of R. Then if T is another ring, and φ : R → T is a ring
homomorphism such that φ(S) ⊂ T × (units of T ), then φ extends to a homomorphism from
S −1 R into T , and this extension is unique: if φ is injective, then the extension is injective
as well.
Proof. We first prove the existence. We define ψ : S −1 R → T by ψ(s−1 r) = φ(s)−1 φ(r).
We will show that this extension works. Note that φ(s)−1 exists since φ(s) ∈ T × . First, we
need to show that ψ is well-defined. Let s−1 −1
1 r1 = s2 r2 . Then there exists s3 ∈ S such that
s3 (s1 r2 −s2 r1 ) = 0. Since s3 is not a zero-divisor, s1 r2 = s2 r1 . Hence φ(s1 )φ(r2 ) = φ(s2 )φ(r1 ),
from which φ(s1 )−1 φ(r1 ) = φ(s2 )−1 φ(r2 ). Therefore, ψ is well-defined. If r ∈ R, then
ψ(1−1 r) = (φ(1))−1 φ(r) = φ(r), so ψ indeed extends φ.
We need ψ to be a homomorphism:
ψ(s−1 −1 −1 −1
1 r1 s2 r2 ) = ψ((s1 s2 ) (r1 r2 )) = φ(s1 s2 ) φ(r1 r2 )

= φ(s1 )−1 φ(r1 )φ(s2 )−1 φ(r2 ) = ψ(s−1 −1

1 r1 )ψ(s2 r2 ).

Also,
ψ(s−1 −1 −1 −1
1 r1 + s2 r2 ) = ψ((s1 s2 ) (s2 r1 + s1 r2 )) = φ(s1 s2 ) φ(s2 r1 + s1 r2 )

= φ(s1 )−1 φ(s2 )−1 (φ(s1 )φ(r2 ) + φ(s2 )φ(r1 )) = ψ(s−1 −1

1 r1 ) + ψ(s2 r2 ).

This proves existence.

Suppose that f : S −1 R → T is any extension of φ to a homomorphism. Then 1 =
f (1) = f (s−1 s) = f (s−1 )f (s) = f (s−1 )φ(s). Therefore f (s−1 ) = φ(s)−1 . So f (s−1 r) =
f (s−1 )F (R) = φ(s)−1 φ(r) = ψ(s−1 r), proving uniqueness.
Finally, assume that φ is injective and let s−1 r ∈ ker ψ. Then φ(s)−1 φ(r) = 0, so r ∈ ker φ.
But since φ is injective, we have r = 0. Consequently, ψ is injective. 
21
Remark 14.2. Uniqueness gives that S −1 R is a unique ring up to isomorphism with the
universal property. Suppose B1 , B2 both have this universal property. Then note that
φ0 ◦ φ = idB1 . Reverse the role of B1 and B2 to get φ ◦ φ0 = idB2 . Hence B1 ∼
= B2 as desired.

>
B1
∃!φ
/  
R o / B2
∃!φ0

B1
Example 14.5. Suppose that f ∈ R is not nilpotent, and let S = {1, f, f 2 , . . . }. Let Rf :=
S −1 R.
If P is a prime ideal in R and S = R \ P , then S is multiplicatively closed. Write
RP := S −1 R. Note that RP has an ideal P Rp = {s−1 : S ∈ / P, r ∈ P }. We shall show that
P RP is a prime ideal.
Let T := Frac(R/P ). Define φ : R → T by r 7→ (r + P )/1. If s ∈ S = R \ P , then
φ(s) ∈ T × . Let ψ : RP → T be the unique extension of φ (universal property).
Note that ψ is surjective. If x = (a+P )/(b+P ), where b ∈/ P , then b ∈ S and ψ(b−1 a) = x.
Hence ker(ψ) is a maximal ideal of RP , since T is a field. But ker ψ = {s−1 r, s ∈ S, r ∈ P } =
P RP . So P RP is a maximal ideal of RP a fortiori.
15. February 5
Definition 15.1. A ring R is a local ring if it has a unique maximal ideal.
Recall that if RP is a local ring (R a ring and P a prime ideal) then P RP is a (in fact the)
maximal ideal.
Proposition 15.2. If R is a ring and M is a maximal ideal, then M is the unique maximal
ideal of R if and only if 1 + x is a unit for all x ∈ M.
T
Proof. If M is unique, then J(R) = P = M, so 1 + x is a unit for all x ∈ J(R) = M.
P maximal
If M is not unique, then there exists a maximal ideal Q 6= M. Then Q + M = R. So there
exists q ∈ Q and x ∈ M such that q − x = q + (−x) = 1. So q = 1 + x. But q is not a unit
since q ∈ Q. But x ∈ M so there exists M such that 1 + x is not a unit. Contradiction! 
So PRP is the unique maximal ideal of RP . To see why, start with x ∈ PRP . Then
x = s−1 a where s ∈ S = R \ P and a ∈ P. So 1 + x = s−1 s + s−1 a = s−1 (s + a), so letting
s + a = t ∈ P we have (s−1 t)−1 = t−1 s so t ∈ S.
Definition 15.3. Let R be a ring and S ⊆ R be a multiplicatively closed subset that has
no zero divisors (regular). Given an ideal J of R, we say that J is S-saturated if whenever
s ∈ S and x ∈ R are such that sx ∈ J we necessarily have x ∈ J.
Example 15.4. If R = Z and S = {1, 2, 22 , 23 , · · · } then 3Z is S-saturated but 4Z is not.
Proposition 15.5. Let R be a ring and let S be a multiplicatively closed set of regular
elements. Then there exists an inclusion-preserving bijection between the poset of proper
ideals of S −1 R and the poset of S-saturated ideals of R that intersects with S trivially.
22
Example 15.6. Let R = Z and S = {1, 2, 22 , 23 , . . . }. Then S −1 R has proper ideals. The
proper ideals of S −1 R are S −1 (nR) where n > 1 and n odd.
Proof. Let I be an ideal of S −1 R and J an ideal of R such that J ∩ S = ∅. Consider the
mappings f : I 7→ f (I) = I ∩ R and g : J 7→ S −1 J, and verify that they are bijections.
First, if I is a proper ideal of S −1 R, then what is g ◦ f (I)? We claim that g ◦ f (I) =
?
S −1 (I ∩R) = I. Notice that I ∩R ⊆ I and since I i an ideal, we have S −1 (I ∩R) = S −1 I = I,
as desired. Conversely, to see that I ⊆ S −1 (I ∩ R), let x ∈ I. Then x = s−1 a for some s ∈ S
and a ∈ R. Since I is an ideal of S −1 R, sx = a ∈ I so a ∈ I ∩R. Thus x = s−1 a ∈ S −1 (I ∩R).
Thus I ⊆ S −1 (I ∩ R).
If J is an S-saturated ideal of R and J ∩ S = ∅, then what is f ∩ g(J)? We claim that
f ∩ g(J) = S −1 J ∩ R = J. The inclusion J = 1−1 J ∩ R ⊆ S −1 J ∩ R is clear. For the other
way, start with x ∈ S −1 J ∩ R. Then x ∈ S−1J, so there exist s ∈ S and j ∈ J such that
x = s−1 j. So sx = j ∈ J. So x ∈ J since J is S-saturated. Thus x ∈ R. So S −1 J ∩ R ⊆ J.
Notice that if I is a proper ideal of S −1 R, then f (I) is S-saturated. Notice that if I is
a proper ideal of S −1 R, then f (I) is S-saturated. Next, if J an S-saturated ideal of R and
J ∩ S = ∅ then g(J) = S −1 J is a proper ideal. Otherwise, then 1 ∈ S −1 J so s−1 j = 1 for
some s ∈ S and j ∈ J, hence j = s, which is a contradiction. 
Corollary 15.7. Let R be a ring and let S ⊆ R be a multiplicative closed subset of regular
elements. If R is Noetherian, then S −1 R is Noetherian.
Remark 15.1. Notice that the converse does not hold. Consider R = C[x1 , x2 , . . . ] and
S = R \ {0}. Then S −1 R = C(x1 , x2 , . . . ). S −1 R is a field of fraction of R, so S −1 R is
automatically Noetherian. But R is not Noetherian.
Proof. Let J1 ⊆ J2 ⊆ J3 ⊆ · · · be a chain of ideals in S −1 R. If Jn = S −1 R for some
n, then Jn = Jn+1 = Jn+2 = · · · . Otherwise, we can apply the map f to get a chain
f (J1 ) ⊆ f (J2 ) ⊆ f (J3 ) ⊆ · · · of ideals in R. Since R is Noetherian, there exists n such
that f (Jn ) = f (Jn+1 ) = · · · . Hence Jn = g(f (Jn )) = Jn+1 = g(f (Jn+1 )) = · · · . So S −1 R is
Noetherian as well, as required. 
Theorem 15.8. Suppose that F is a field and K/F is a finitely generated (not necessarily
finite) field extension. If L is an intermediate field between F and K (i.e., F ⊆ L ⊆ K),
then L/F is also a finitely-generated field extension.
Remark 15.2. The above theorem need not hold when it comes to algebras. Note that C[x, y]
is a finitely-generated C-algebra, but C[xi y i+1 : i ≥ 1] is not a finitely generated subalgebra
of C[x, y].
Remark 15.3. Let’s come back to the correspondence mappings f and g we discussed in the
proof of Proposition 15.5. These bijections restrict to bijections between prime ideals of
S −1 R and {P ⊆ R : P prime, P ∩ S = ∅}.
16. February 6
Definition 16.1. Given a ring R, we let Spec(R) denote the set of prime ideals of R.
Remark 16.1. Spec(R) is a poset with respect to ⊆.
Example 16.2. Spec(Z) = {2Z, 3Z, 5Z, 7Z, · · · }.
23
 Let R be a ring, and S ⊆ R a multiplicatively closed regular set. Last time, we gave
an inclusion-preserving bijection between the set of proper ideals of S −1 R and the set of
f
S-saturated ideals of R that intersect with S trivially, with the map given by I 7→ I ∩ R and
g
S −1 J ←[ J.
These bijections send prime ideals to prime ideals. If P is prime in S −1 R, then f (P ∩ R) ⊆
R is prime in R. To see why, suppose that ab ∈ f (P ) with a, b ∈ R. Then ab ∈ P ∩ R ⊆ P ,
so a ∈ P or b ∈ P . Therefore a ∈ P ∩ R = f (P ) or b ∈ P ∩ R = f (P ).
Conversely, if Q ≤ R with Q prime, then Q ∩ S = ∅: notice that this implies that Q is
S-saturated. Suppose s ∈ S, x ∈ R and sx ∈ Q. Thus s ∈ Q or x ∈ Q, hence x ∈ Q.
Then g(Q) is prime in S −1 R. Suppose that s−1 −1
1 a, s2 b ∈ S
−1
R and s−1 −1
1 as2 b ∈ g(Q). Thus
(s1 s2 )−1 (ab) ∈ g(Q), hence ab ∈ g(Q) ∩ R = Q. Thus a ∈ Q or b ∈ Q, which implies that
s−1 −1
1 a ∈ g(Q) or s2 b ∈ g(Q). Hence the maps f and g restrict to:

Spec(S −1 R) ←→ {Q ∈ Spec(R) : Q ∩ S = ∅}.

Example 16.3 (One special case). Let R be a ring and x a non-zero divisor. Let S =
{1, x, x2 , · · · }. Then there is a correspondence between
Spec(Rx ) ←→ {Q ∈ Spec(R) : Q ∩ {1, f, f 2 , · · · } = ∅} = {Q ∈ Spec(R) : f ∈
/ Q}.
Example 16.4. Let R be a integral domain, and let P be a prime ideal. Let S = R \ P . Then
S −1 R = RP . Then there is a correspondence
Spec(RP ) ←→ {Q ∈ Spec(R) : Q ∩ S = ∅} = {Q ∈ Spec(R) : Q ⊂ P }.
This gives another proof that RP is a local ring.
Example 16.5 (Spec(R) vs Spec(RP ) vs Spec(R/P )). Spec(R) denotes the set of all prime
ideals of R. Recall that Spec(RP ) and the set of prime ideals of R contained in P have
one-to-one correspondence. On the other hand, by the correspondence theorem of ideals,
we see that Spec(R/P ) and the set of prime ideals of R containing P a prime ideal have
one-to-one correspondence.
Remark 16.2. In Assignment #2, you will show that if K is a field extension of a field F then
K is finitely generated as an extension of F . This can be deduced by showing that K ⊗F K
is Noetherian. In fact, in the assignment you will show that if K ⊗F K is Noetherian, then
any field L with F ⊆ L ⊆ K is finitely generated over F .
Theorem 16.6. Let F ⊆ K be a finitely-generated field extension. If F ⊆ L ⊆ K and L is
a field, then L/F is finitely generated as well.
Proof. By Assignment #2, it suffices to show that K ⊗F K is Noetherian.
Step 1. Since K/F is finitely generated, there exist a1 , a2 , . . . , ad ∈ K such that K =
F (a1 , a2 , . . . , ad ). Let A = F [a1 , a2 , . . . , ad ] = { αi1 ,...,id ai11 · · · aidd : αi1 ,...,id = 0 for all but
P
finitely many (i1 , . . . , id )}. Then A is a finitely generated F -algebra. So by the Hilbert basis
theorem, A is Noetherian.
Step 2. On Assignment #2, you will show that if A and B are finitely generated F -algebra,
then A ⊗F B is ale finitely generated. So since A = F [a1 , . . . , ad ] is finitely generated as an
F -algebra, so is A ⊗F A. Again by the Hilbert basis theorem, A ⊗F A is Noetherian.
Step 3. Let S = A \ {0}, which is multiplicatively closed and regular. Let T = {s1 ⊗ s2 :
s1 , s2 ∈ S}. We will show that T −1 (A ⊗F A) ∼ = S −1 A ⊗F S −1 A = K ⊗F K and that T is
24
regular. So, if A ⊗F A is Noetherian, then T −1 (A ⊗F A) is Noetherian, which shows that
K ⊗F K is Noetherian (then we are home free). 
Proposition 16.7. Let A, B be F -algebras and let S, T be multiplicatively closed, regular
subsets of A and B respectively. Then if U = {s ⊗ t : s ∈ S, t ∈ T }, then U −1 (A ⊗F B) ∼
=
S −1 A ⊗F T −1 B.
Proof. Let f : A → S −1 A given by a 7→ 1−1 a and g : B → T −1 B given by b 7→ 1−1 b. SO
we have a homomorphism f ⊗ g : A ⊗F B → S −1 A ⊗F T −1 B. Notice that if s ⊗ t ∈ U 7→
f (s) ⊗ g(t) = s ⊗ t, then s ⊗ t is a unit in S −1 A ⊗ T −1 B with inverse s−1 ⊗ t−1 . So f ⊗ g
extends to a homomorphism. SInce f ⊗ g : U −1 (A ⊗F B) → S −1 A ⊗F T −1 B by the universal
property of localization, all that remains is to show f ⊗ g is an isomorphism:
(1) f ⊗ g is surjective
S −1 A ⊗ T −1 B is generated by things of the form s−1 a ⊗ t−1 b = f ⊗ g((s ⊗ t)−1 (a ⊗ b)).
(2) f ⊗ g is injective
One has to be careful here: even though f : A → C and g : B → D are injective,
f ⊗ g : A ⊗R B → C ⊗R D need not be injective. Consider f : Z/2Z → Z/2Z defined
by [a] 7→ [2a]. But f ⊗ f has the non-trivial kernel, since (f ⊗ f )(1 ⊗ 1) = 2 ⊗ 2 = 0.

√ √
Example 16.8. If F = Q and K = Q(π, 2), then A = Q[π, 2] is finitely generated F -
algebra. Note that
nX √ j o
A= cij π i 2 : cij ∈ Q, cij = 0 for all but finitely many (i, j) .

17. February 10
Proposition 17.1. If A, B, C, D are F -vector spaces and f : A → C and g : B → D are
injective, then f ⊗ g : A ⊗F B → C ⊗F D is also injective.
Proof. Let {Xα }α∈I be a basis for A and {Yβ }β∈J a basis for B. Since f and g are injective,
the subsets {f (Xα )}α∈I ⊆ C, {g(yβ )}β∈J ⊆ D are both linearly independent. So we can
extend {f (Xα )}α∈I to a basis {zγ }γ∈I 0 for C and similarly extend {g(yβ )}β∈J to a basis
{wδ }δ∈J 0 for D. Recall that {xα ⊗ yβ }(α,β)∈I×J form an F -basis for A ⊗F B. So if f ⊗ g is not
injective, then there exists cα,β ∈ F notP all zero such that cα,βP= 0 for all but finitely many
(α, β) ∈ I × J and such that (f ⊗ g) (Xcα,β xα ⊗ yβ ) = 0, or cα,β (f (xα ) ⊗ g(yβ )) = 0.
Notice that we may write this as dγ,δ zγ ⊗wδ = 0, where dγ,δ = 0 for all but finitely
(γ,δ)∈I 0 ×J 0
many (γ, δ), and dγ,δ are not all zero. But this is a contradiction, since {zγ ⊗ wδ }(γ,δ)∈I 0 ×J 0
form a basis for C ⊗F D. 
Corollary 17.2. S −1 A ⊗F T −1 B ∼
= (S ⊗ T )−1 (A ⊗F B).
Remark 17.1. If S and T are regular in A and B respectively, then S ⊗ T = {s ⊗ t : s ∈
S, t ∈ T } ⊆ A ⊗F B is regular also. Since S and T have no zero divisors, it follows that
f : A → A and g : B → B given by f (a) = sa and g(b) = tb are injective. Therefore
f ⊗ g : A ⊗F B → A ⊗F B is injective also, so s ⊗ t cannot be a zero divisor.
25
17.1. Hilbert’s Nullstellensatz (“zero-locus theorem” or “theorem of zeroes”).
Definition 17.3. Recall that R is a Jacobson ring if for any P ∈ Spec(R) we have J(R/P ) =
(0).
Example 17.4. Any field F , the polynomial ring of a field F [x], and the ring of (rational)
integers Z are all Jacobson rings.
Definition 17.5. Recall that if R is a ring and S is a ring, then S is an R-algebra if there
exists a (not necessarily injective) homomorphism from R to S such that α(1R ) = 1S . We
say S is finitely generated as an R-algebra if there exists n ≥ 1 and s1 , . . . , sn ∈ S such that
every x ∈ S can be expressed as a polynomial p(s1 , . . . , sn ) with the coefficients of p in R.
Remark 17.2. Equivalently, if S is finitely generated by s1 , . . . , sn as an R-algebra, then
there exists a surjective homomorphism φ : R[x1 , . . . , xn ] → S given by p(x1 , . . . , xn ) 7→
p(s1 , . . . , sn ). Therefore if I = ker φ, then by the first isomorphism theorem we have S ∼ =
R[x1 , . . . , xn ]/I.
Theorem 17.6 (General Nullstellensatz). Let R be a Jacobson ring and let S be a finitely
generated R-algebra. Then
(1) S is also a Jacobson ring.
(2) if M ⊆ S is a maximal ideal of S, then α(R) ∩ M =: N is a maximal ideal of α(R)
and S/M =: F is a finite extension of α(R)/N.
Let’s first consider a special case: when R = k = k, an algebraically closed field. Write S =
k[x1 , . . . , xn ]/I where I is a proper ideal. Notice that there is a correspondence between the
set of maximal ideals of S and ideals M of k[x1 , . . . , xn ] containing I. So S/(maximal ideal) ∼ =
k[x1 , . . . , xn ]/M of which the latter is a field and is a finite extension over k. Then N := M∩k
is a maximal ideal of k, so N = (0) and k/N = k.
If k is algebraically closed, then a finite extension of k must be k itself. Therefore
k[x1 , . . . , xn ]/M ∼ = k so there exist λ1 , . . . , λn such that M = (x1 − λ1 , . . . , xn − λn ), with
the isomorphism φ given by p(x1 , . . . , xn ) 7→ p(λ1 , . . . , λn ). Thus ker φ = M.
So far, we haven’t seen any “zero locus”. So how does this fit in? Start with f1 (x1 , . . . , xn ),
. . . , fd (x1 , . . . , xn ) ∈ k[x1 , . . . , xn ] and k algebraically closed (i.e., k = k). If I = (f1 , . . . , fd )
is a proper ideal of k[x1 , . . . , xn ], then there exists M = (x1 − λ1 , . . . , xn − λn ) ⊇ I. This
means that f1 , . . . , fd are in the kernel of φ. Therefore fi (λ1 , . . . , λd ) = 0 for all 1 ≤ i ≤ d.
Thus (λ1 , . . . , λd ) is the so-called “zero locus”.

18. February 12
To do the general Nullstellensatz, we will use the so-called “Rabinowitsch trick”, which
gives a useful characterization of Jacobson rings.
Theorem 18.1 (Rabinowitsch trick). Let R be a ring. Then the following are equivalent:
(1) R is a Jacobson ring.
(2) J(R/P ) = (0) for all P ∈ Spec(R)
(3) For all P ∈ Spec(R), P is the intersection of all the maximal ideals containing P .
(4) Whenever P ∈ Spec(R), and T := R/P has the property that there exists a non-zero
b ∈ T such that Tb := {1, b, b2 , . . . }−1 T = T [b−1 ] is a field, then T is already a field.
26
Proof. ((1) ⇒ (4)) Suppose that R is Jacobson and let P ∈ Spec(R) and let T := R/P .
Suppose that there exists b ∈ T \ {0} such that Tb is a field. Our goal is to show that T
is a field. Recall that there is a one to one correspondence between the prime ideals of Tb
and the prime ideals of T that do not contain b. But since we assume that Tb is a field, it
follows that Spec(Tb ) = {(0)}. Therefore by the correspondence every non-zero prime ideal
of T contains b. Now if T is not afield, then every maximal iaedl of T contains b. Therefore,
we have \
J(T ) = M 3 b,
M maximal

which implies J(T ) 6= (0). But this is a contradiction since R is Jacobson and T = R/P .
((4) ⇒ (1)) Suppose that whenever T = R/P has the property that there exists a non-zero
b ∈ T such that Tb is a field we must have T is a field. We must show that J(R/Q) = (0)
for all Q ∈ Spec(R). Towards a contradiction, suppose that J(R/P ) 6= 0 for some prime
ideal P . Let S = R/P . Pick 0 6= b ∈ J(S). Then every maximal ideal of S contains b.
Consider the ring Sb := S[ 1b ]. Then there exists a maximal ideal Q of Sb . Recall that there
is a bijective correspondence between the two following sets:
{prime ideals of Sb } ←→ {prime ideals of S that do not contain b}
Q 7−→ S ∩ Q
Sb Q ←−[ Q0
So then there exists Q0 ∈ Spec(S) such that Sb Q0 = Q and b ∈ / Q0 . Notice that Q0 is not
maximal since b ∈ / Q0 . So S/Q0 is not a field. Let 0 6= b be the image of b in S/Q0 . Then
0 1 ∼
(S/Q )[ b ] = S[ b ]/S[ 1b ]Q0 = Sb /Q, but S/Q0 is not a field. Since R  S  S/Q, we have
1

S/Q ∼ = R/I for some prime ideal I of R. This is a contradiction because (R/I)[ 1b ] is a field
but R/I is not.
((1) ⇔ (2) ⇔ (3)) These directions are immediate from the definition of the Jacobson
ring. 
Theorem 18.2 (Full Nullstellensatz). Suppose that R is a Jacobson ring, and S is a finitely-
generated R algebra, i.e., S ∼
= R[x1 , . . . , xd ]/I for some ideal I. Suppose α(R) = R/(R ∩ I).
Then:
(1) S is Jacobian.
(2) If M is a maximal ideal of S, then N := M ∩ α(R) is a maximal ideal of α(R) and
S/M is a finite field extension of α(R)/N.
Proof. Suppose that R is Jacobson. Then R/J is also Jacobson for any proper ideal J. If
Q is a prime ideal of R/J, then there exists P ⊃ J in R such that R/P = (R/J)/Q – since
J(R/P ) = (0), then J((R/J)/Q) = (0), implying that R/J is Jacobson also. Notice that S
is a finitely generated α(R)-algebra
S∼= (α(R))[x1 , . . . , xd ]/I, I ∩ α(R) = (0).
Therefore, without loss of generality we may assume that α(R) = R.
The key steps of the proof are the following:
(i) Show that the theorem is true when S = R[x]. This is the hardest step.
(ii) Use induction to show that the theorem is true for S = R[x1 , . . . , xd ]
27
 (iii) Use correspondence to show that the theorem is true fro S = R[x1 , . . . , xd ]/I with
I ∩ R = (0).
We will assume (i) for now and prove (ii) and (iii).
(ii) Assuming (i), we have that R[x] is Jacobson where R is Jacobson. Therefore, by
induction we have R[x1 , . . . , xd ] Jacobson for any d. Write S := R[x1 , . . . , xd ]. Next,
let M be the maximal ideal of S = R[x1 , . . . , xd ]. Write S = T [xd ], where T =
R[x1 , . . . , xd−1 ]. So S/M is a finite extension of T /N where N := T ∩ M by (i). Also
note that N is a maximal ideal of T .
Argue by induction on d, we then have T /N is a finite extension of R/(R ∩ N).
Since [S/M : T /N], [T /N, R/(R ∩ M)] < ∞, the claim follows.
(iii) We have by (ii) that R[x1 , . . . , xd ] Jacobson ⇒ S := R[x1 , . . . , xd ]/I is Jacobson and
I ∩ R = (0). Let M be a maximal ideal of S. Note that there is a correspondence
between
{maximal ideals of S} ←→ {maximal ideals of R[x1 , . . . , xd ] containing I}
M ⊆ R[x1 , x2 , . . . , xd ]/I 7−→ M0 ⊆ R[x1 , . . . , xd ].
Then N := M0 ∩ R is maximal by (ii), and R[x1 , . . . , xd ]/M0 ∼ = S/M, and
[R[x1 , . . . , xd ]/M0 : R/N], [S/M, R/N] < ∞, and M ∩ R = M0 ∩ R = N.
So it remains to prove the Nullstellensatz when S = R[x]. Let’s look at special case R =
k, S = k[x]. Consider the case when |k| = ∞. Why is S Jacobson? To answer this,
we need to consider what Spec(S) looks like. If P 6= (0), then P = (f (x)) where f (x)
is irreducible, making P maximal. So J(k[x]/P ) = (0) since k[x]/P is a field. And so
k[x]/P ∼ = k[x]/(f (x)), and [k[x]/(f (x)) : k/(k ∩ P )] < ∞. Note that k = k/(k ∩ P ) and
[k[x]/(f (x)) : k/(k ∩ P )] = deg f (x).
Why is J(k[x]) = (0)? Observe that
\ \
J(k[x]) = (f (x)) ⊆ (x − λ) = (0),
f (x) irred. λ∈k

if |k| = ∞. 

19. February 13
Proof of Theorem 18.2 cont’d. To show that S = R[x] is Jacobson where R is Jacobson and
P a prime ideal of R, let T = S/P = R[x]/P and let R0 = R/(R ∩ P ). We must show that
if Tb is a field and b is non-zero, then T is a field. Notice that R[x]/P ∼ = R0 [x]/Q for some
prime ideal Q of R0 [x] and Q ∩ R0 = (0). Assume that Tb = Frac(T ). We claim that Q 6= (0).
Suppose otherwise. Then T = R[x]/Q = R0 [x]. If K = Frac(R0 ), then T ⊆ K[x] ⊆ Frac(T ),
hence Tb = Frac(T ) ⊆ K[x]b ⊆ Frac(T )b = Frac(T ). Therefore K[x]b = Frac(T ). We
showed that K[x] is Jacobson, so by the Rabinowitsch trick, K[x]b is a field, making K[x]
a field, which is a contradiction. If T = R0 [x]/Q and Q ∩ R0 = (0) and Q 6= (0) with
Tb = Frac(T ) a field, we need to show that T is in fact a field. Since Q ∩ R0 = (0), we
have an injection T 7→ K[x]/K[x]Q. Note that K[x]/K[x]Q = S −1 R0 [x]/S −1 Q, where
S = R0 \ {0}, image in R0 [x]/Q. Now Tb is a field, so (K[x]/K[x]Q)b i a field too. Therefore
Tb is a localization of Tb . Thus K[x] is Jacobson, so Q1 := K[x]Q has the property that
K[x]/Q1 is a field. Hence Q1 is a maximal ideal in K[x]. Also, since d = [k[x]/Q : k] < ∞, we
have Q1 = (xd + bd−1 xd−1 + · · · + b0 ) with bi ∈ K = Frac(R0 ). So clearing the denominators,
28
we get ad xd + · · · + a0 ∈ Q = k[x] ∩ Q1 with ad 6= 0. So just inverts ad to form Ra0 d . Then
Q0 := QRa0 d [x] 3 (xd + ad−1 a−1
d x
d−1
+ · · · + a0 a−1
d ).
0
Then the image x of x ∈ Rad [x]/Q0 satisfies a monic polynomial
xd + ad−1 a−1
d x
d−1
+ · · · + a0 a−1
d = 0. (†)
Recall that R0 [x]/Q = T so Ra0 d [x]/Q0 = Tad and that Ra0 d ⊆ Tad . Since Tad is generated by
x over Ra0 d , we have Tad ⊆ Ra0 d + Ra0 d x + · · · + Ra0 d xd−1 by (†). By assumption, Tb is a field,
so Tad b = (Tb )ad is a field. Therefore (Ra0 d [x])b is a field. So Tad is a finite module over Ra0 d
spanned by 1, x, . . . , xd−1 . Now we claim that there exists m ≥ 1 so that c0 , c1 , . . . , cm ∈ Ra0 d
not all zero satisfies c0 + c1 b + · · · + cm bm = 0. To see why, write
1=1
b = α1,0 + α1,1 x + · · · + α1,d−1 xd−1 (αij ∈ Ra0 d )
b2 = α2,0 + · · · + α2,d−1 xd−1 .
So think of bi → (αi,0 , . . . , αi,d−1 ) ∈ (Ra0 d )d−1 ⊆ K d−1 .
If m > d, then 1, b, b2 , b3 , · · · , bm are linearly dependent over K. So there exist γ0 , . . . , γm ∈
K not all zero such that γ0 + γ1 b + γ2 b2 + · · · + γm bm = 0. Now clear the denominator to
get c0 + c1 + · · · + cm bm = 0.
Without loss of generality, let c0 6= 0 and cm 6= 0. So c0 + c1 b + · · · + cm bm = 0 Invert c0 to
get 1 = b(−c1 c−10 − · · · − cm c0 b
−1 m−1
) in (Ra0 d )c0 [x]/Ra0 d ,c0 Q. Write Q
e := Ra0 ,c Q. This means
d 0

that b is a unit in Ra0 d ,c0 [x]/Q e = Ta ,c0 .

d
Now Tb is a field, so (Tad ,c0 )b is a field. But b is a unit, so Tad ,c0 os a field. So Tad ,c0 is
a finite module over Ra0 d ,c0 . By “the black box” from the last lecture, Tad ,c0 if a field then
Rad ,c0 is a field. And then by the Rabinowitsch trick, R0 is a field, whence R0 = K. So
S = R0 [x]/Q = K[x]/Q. K[x]/Q is Jacobson already, so indeed S = R[x] is Jacobson.
Now the last step is to show that M is an ideal of S where M ∩ R =: N is maximal in
R and [S/M : R/f N ] < ∞. But we just showed that if M ⊆ S is a maximal ideal, then
R0 = R/R ∩ M is a field. To see why, if S/M = R0 [x]/P and P ∩ R0 = (0). Note R0 [x]/P is
a field and R0 ∩ P = (0). And everything in R0 is a unit, so R = K. 

20. February 24: Integral extensions

Let R ⊆ S be rings. Then 1R = 1S . If i : R ,→ S is an inclusion map, then i gives an
R-algebra structure.
Definition 20.1. We say that S is an integral extension of R if every s ∈ S satisfies a monic
polynomial equation with coefficients in R, i.e., there exists n ≥ 1 and r0 , . . . , rn−1 ∈ R such
that sn + rn−1 sn−1 + · · · + r0 = 0.
Example 20.2. Q is not an integral extension of Z. Take s = 21 ∈ Q. If s were integral over Z,
then sn +rn−1 sn−1 +· · ·+r0 = 0 for some ri ∈ Z. Therefore 2−n +rn−1 2−(n−1) +· · ·+r1 2−1 +r0 =
0. Thus we see that the sum of an integer and 12 is 0, which is impossible.
√ √ √
Example
√ 20.3. Z[ 2] = {a + b 2 : a, b ∈ Z} is integral over Z, since every s = a + b 2 ∈
Z[ 2] satisfies x2 − 2ax + a2 − 2b2 = 0.
29
Definition 20.4. An element s ∈ S is called integral over R if it satisfies a monic polynomial
equation with coefficients in R.
Proposition 20.5. If R ⊆ S, and S is a finitely-generated R-module, then S is an integral
extension of R.
Proof. Write S = Ra1 + · · · + Rad , where a1 , . . . , ad ∈ S. Let s ∈ S. We need to find a monic
polynomial. Notice that there exist rij ∈ R with 1 ≤ i, j ≤ d such that
d
X
s · ai = rij aj .
j=1

Then
11 r12 · · ·
  r 
r1d 
a1 a1
 
.
 a2    r21 . . r2d   a2 

s  ..  =  .
 
.. .. 
 ..  .
.  .. . .   . 
ad rd1 rd2 · · · rdd ad
Thus
a1 0
   
 a2   0 
(sI − A) 
 ...  =  ...
  ,

ad 0
where A = (rij ) ∈ Md (R) and I : S d → S d the identity map. If we multiply by (sI − A)adj :=
(−1)i+j det((sI −A)ji ) (where (sI −A)ij denotes the (d−1)×(d−1) matrix with the j-th row
and the i-th column of sI − A removed), then indeed det(sI − A)ai = 0 for all i. Therefore
det(sI − A)(Ra1 + · · · + Rad ) = 0, so det(sI − A) = 0. Note that det(sI − A) is a polynomial;
let det(sI − A) =: pA (s). Note that pA (s) is a monic polynomial with coefficients in R. 
We will now show that if S ⊇ R then T := {s ∈ S : s integral over R} forms a ring with
R ⊆ T ⊆ S.
Example 20.6. Let R = Z and S = Q. Then T = A = {s ∈ Q : s is a root of a monic
integer polynomial} is a ring, and A is said to be the set of algebraic integers.
Proposition 20.7. Let R ⊆ S be rings and let s ∈ S. Then the following are equivalent:
(1) s is integral over R;
(2) there exists a finitely-generated R-submodule M of S such that sM ⊆ M and sM 6=
(0).
Proof. ((1) ⇒ (2)) If s is integral over R, then there exists n ≥ 1 and ri ∈ R such that
sn + rn−1 sn−1 + · · · + r1 + r0 = 0. Let M = R + Rs + Rs2 + · · · + Rsn−1 . Then M is a
finitely-generated R-module, with 1 ∈ M hence sM 6= (0). Also, note that
sM = s(R + Rs + · · · + Rsn−1 ) = Rs + Rs2 + · · · + Rsn
= Rs + Rs2 + · · · + R(−rn−1 sn−1 − · · · − r1 s − r0 )
⊆ R + Rs + · · · + Rsn−1 = M.
30
 ((2) ⇒ (1)) Suppose that 1 ∈ M = Ra1 + · · · + Rad with sM 6= (0) and sM ⊆ M . As
before,
d
X
sai = rij aj , rij ∈ R.
j=1
If A = (rij ) such that
a1 0
   
 a2   0 
(sI − A) 
 ...  =  ...  ,
  

ad 0
then det(sI − A)ai = 0 for all i hence det(sI − A) = 0, which is enough to show that s is
integral over R (see the proof of Proposition 20.5). 
Corollary 20.8. If T = {s ∈ S : s integral over R}, then T is a ring with R ⊆ T ⊆ S.
Proof. Let x, y ∈ T . Then x and y are integral over R. So we have xn + rn−1 xn−1 + · · · +
0
r1 x + r0 = 0 and y m + rm−1 y m−1 + · · · + r10 y + r00 = 0. Now let
n−1 m−1
X X
M= Rxi y j ⊆ S.
i=0 j=0

Then M is finitely generated and 1 ∈ M . Notice that xM ⊆ M . To see why, note that
(
i j xi+1 y j (if i < n − 1)
x(x y ) = n−1 n−2 j
(−(rn−1 x + rn−2 x + · · · + r1 x + r0 )y ) (if i = n − 1).
So xM ⊆ M , and similarly yM ⊆ M . So (x + y)M ⊆ xM + yM ⊆ M + M ⊆ M , and
similarly xyM ⊆ x(yM ) ⊆ xM ⊆ M . THerefore x + y and xy are integral over R and hence
are in T . Also, T ⊇ R: if r ∈ R then r satisfies x − r = 0. Hence T is a ring containing
R. 
Definition 20.9. Given R ⊆ S, the ring
T := {s ∈ S : s is integral over R}
is called the integral closure of R in S. Specifically, if R is an integral domain, then the
integral closure of R is the integral closure of R in Frac(R), the field of fractions of R.
Example 20.10. Let R = C[t2 , t3 ] ⊆ C[t]. What is the integral closure of R? In fact it’s C[t].
First, notice that t is a root of x2 − t2 ∈ R[x]. So t is integral over R. Thus C[t] is contained
in the integral closure of R.
If s ∈ C(t) is integral over R, then it is also integral over C[t]. Thus R ⊆ C[t]. So it suffices
to show that C[t] is integrally closed. For this, it suffices to prove the following theorem,
which we shall prove on Thursday.

31
 21. February 26
Theorem 21.1. Let R be a UFD. Then R is integrally closed.
Proof. Let ab−1 ∈ Frac(R) with gcd(a, b) = 1 with b =6 0 and ab−1 integral over R. Then
−1
because ab is integral over R, there exist n ≥ 1 and r0 , . . . , rn−1 ∈ R such that
(ab−1 )n + rn−1 (ab−1 )n−1 + · · · + r0 = 0.
Multiply both sides by bn to get
an + rn−1 an−1 b + · · · + r1 abn−1 + r0 bn = 0.
| {z }
multiple of b

So b | an . But gcd(a, b) = 1, hence gcd(an , b) = 1. Therefore b is a unit, so ab−1 ∈ R. Thus

the integral elements of Frac(R) over R are precisely the elements of R. 

21.1. Lying over & going up.

Remark 21.1. The theorems we shall cover in this section will tell us about prime ideals in
S in terms of prime ideals in R when S is an integral extension over R.
Theorem 21.2. Let R ⊆ S with 1R = 1S be an integral extension. If p ∈ Spec(R), then
there exists q ∈ Spec(S) (q is in general not unique) such that q ∩ R = p. Moreover, if
q1 ∈ Spec(S) is such that q1 ∩ R ( p, then there exists q ∈ Spec(S) such that q ) q1 and
q ∩ R = p.
√ √
Example 21.3. If S = Z[ 2] and R = Z, and p = (2), then q = ( 2).
√
Example 21.4. If R = Z and S = Z[ 3] and p = (5), then q can be (1 + 2i) or (1 − 2i).
Proof. Let p1 = q1 ∩ R be an ideal of R. We claim that p1 is a prime ideal of R, since q1 is
prime.
First reduction: we may replace S by S/q1 and R by R/p1 = R/(R ∩ q1 ). Then S is still
integral over R, so we may assume that q1 = (0) – and it suffices to show that there must
exist q so that q ∩ R = p. Now let U := R \ p ⊆ S. Recall that this set is multiplicatively
closed. Time for the second reduction.
Second reduction: Replace S by U −1 S and R by U −1 R = Rp . Then by a question in
Assignment #3, U −1 S is integral over U −1 R. We can reduce to this case by results on
localization.
So now R is a local ring with the unique maximal ideal p (which is pRp based on the
notation used in the second reduction). And S is still integral over R. Consider the ideal
pS ⊆ S. If pS ( S, then there must exist a maximal ideal k such that k ⊇ pS so that k
is prime. Note that k ∩ R ( R since 1 ∈ k. So we see that k ∩ R ⊆ p, since p is a unique
maximal ideal. But then k ∩ R ⊇ pS ∩ R ⊇ p, so k ∩ R = p. So take q = k and we are done.
So we may assume that pS = S. In particular, assume that 1 ∈ pS. Thus there exist
p1 , p2 , . . . , pd ∈ p and s1 , . . . , sd ∈ S such that p1 s1 + · · · + pd sd = 1. Let S 0 be the R-algebra
generated by s1 , . . . , sd .
Then pS 0 = S 0 . Note that p1 s1 + · · · + pd sd = 1 ∈ pS 0 . So S 0 ⊆ (p1 s1 + · · · + pd sd S 0 ⊆
pS 0 S 0 ⊆ pS 0 . Also, by the integrality of S 0 (over R), S 0 is also a finitely generated R-module,
32
which is stronger than being finitely generated as an R-algebra. Let’s try to explore why.
Note that X X
S0 = Rsi11 si22 · · · sidd ⊆ Rsi11 · · · sidd =: M.
i1 ,...,id 0≤ij ≤nj −1
0≤j≤d

But then S is integral over R so there exist ni ≥ 1 such that sni i ∈ Rsni i −1 + · · · + Rsi + R.
Also, M is a finitely generated R-module. So we have pM = M and p = J(R). So by
Nakayama’s lemma it follows M = (0). But this is a contradiction, since 1 ∈ S 0 means
S 0 ⊇ R. 
One corollary to this theorem is incomparability:
Corollary 21.5. Suppose R ⊆ S is an integral extension with 1R = 1S . If q 6= q0 ∈ Spec(S)
and q ∩ R = q0 ∩ R = p, then q and q0 are incomparable: that is, q 6⊂ q0 and q0 6⊂ q.
Proof. Suppose that q0 ( q. Then q0 ∩R = p. So if we mod out by q0 then we can replace S by
S/q0 and R by R/(R ∩ q0 ) = R/p. Notice that S is still integral over R. Now this reduces to
the case when p = (0), q0 = (0), q 6= (0), q ∩ R = (0). Now pick x ∈ q \ {0}. Then x is integral
over R so there exists n ≥ 1 and rn−1 , . . . , r0 ∈ R such that xn +rn−1 xn−1 +· · ·+r1 x+r0 = 0.
Without loss of generality, let r0 6= 0, which we can do since S is an integral domain and
x 6= 0.
But now, note r0 = −xn − rn−1 xn−1 − · · · − r1 x, so r0 ∈ R ∩ q since each monomial on
the RHS is in q. That is, r0 ∈ R ∩ q = p = (0), so r0 = 0, hence a contradiction. Thus
it is impossible to have (0) ( q ⊆ S such that q ∩ R = (0) with R integral domain, as
required. 

22. February 27
Definition 22.1. Given a ring R, we define the Krull dimension of R to be
Kdim(R) := sup{n : there exist a chain P0 ( P1 ( · · · ( Pn of prime ideals in R}.
Example 22.2. Every field F has Kdim(F ) = 0. We have Kdim(Z) = 1, since (0) ( (p) is the
only chain available, where p is prime. Kdim(F [x]) = 1, since the only prime ideal chains
can be (0) ( (p(x)) where p(x) is irreducible.
Theorem 22.3. If R ⊆ S and S an integral extension of R with 1R = 1S , then Kdim(R) =
Kdim(S).
Example 22.4. Let K be a finite field extension of Q, i.e., [K : Q] < ∞. Then if we take
R = Z then S = {s ∈ K : s integral over Z} = A ∩ K =: OK (where A is the set of algebraic
numbers), and OK is said to be the ring of integers or a number ring. Since Kdim(Z) = 1,
it follows that Kdim(OK ) = 1.
Proof. First suppose that Kdim(R) ≥ n. So there exists a chain P0 ( P1 ( P2 ( · · · ( Pn
in Spec(R). Now we use lying over and going up, so that for any prime ideals P0 ( P1 ( R
and Q0 ∩ R = P where Q0 is a prime ideal in S, then there exists Q1 an ideal of S such that
Q1 ∩ R = P1 . Do this for any chain of ideals of R, there exists a chain Q0 ( Q1 ( Q2 (
· · · ( Qn in Spec(S) with Qi ∩ R = Pi . So Kdim(S) ≥ n.
If Kdim(S) ≥ n, then there exists a chain Q0 ( Q1 ( · · · ( Qn in Spec(S). Take
Pi := Qi ∩ R prime ideals. Then we have P0 ⊆ P1 ⊆ · · · ⊆ Pn . How do we know that the
33
containment is strict? Suppose otherwise. But this contradicts the fact that Q0 and Q1 must
be incompatible, by Corollary 21.5. Thus the strict containment follows. 
Proposition 22.5. Suppose R is Noetherian. Then the following are equivalent:
(1) R has Krull dimension 0
(2) All prime ideals of R are maximal.
(0) = {x : x nilpotent}), then R/N ∼
p
(3) If N is the nilradical of R (i.e., N = =
F1 × F2 × · · · × Fs for s ≥ 1, where each Fi is a field.
Proof. ((1) ⇔ (2)) Take P ∈ Spec(R). If P is not maximal then there must exist a maximal
ideal M such that M ) P , so Kdim(R) ≥ 1, which is a contradiction. The other direction
is immediate.
((3) ⇒ (1)) We claim that F1 × · · · × Fs has Krull dimension 0. First, if P is a prime ideal,
then we claim that there exists j ∈ {1, . . . , s} such that P = F1 × · · · × Fj−1 × (0) × Fj+1 ×
· · · × Fs . If we let e1 = (1F1 , . . . , 0Fs ), e2 = (0F1 , 1F2 , · · · , 0Fs ), · · · , es = (0, 0, · · · , 1Fs ), then
we have ek · el = 0 whenever k 6= l. Thus ek el ∈ P , so either ek or el is in P . Therefore there
must exist j such that e1 , . . . , ej−1 , ej+1 , . . . , es ∈ P . So there must exist P such that
P ⊇ F1 × · · · × Fj−1 × (0) × Fj+1 × · · · × Fs =: Ij .
So if we let F1 × F2 × · · · × Fs /Ij → Fj such that (a1 , a2 , . . . , as ) 7→ aj , then the kernel of
this map is Ij . Therefore Ij is maximal, so P = Ij . Therefore all the prime ideals of R are
maximal, which implies that Kdim(R) = 0.
((1) ⇒ (3)) Suppose that R has Kdim(R) = 0. Since R is Noetherian, there exist
P1 , P2 , . . . , Ps ∈ Spec(R) such that (0) ⊆ Pi for all 1 ≤ i ≤ s, and such that if Q is another
prime in R containing (0) then Q ⊇ Pi for some i. This implies that Spec(R) = {P1 , . . . , Ps },
and if Q ∈ / Spec(R) and Q ⊇ Pi for some i, then Q = Pi , which is a contradiction. This
proves that every Pi is maximal. Therefore Pi + Pj = R whenever i 6= j. By the Chinese
Remainder theorem, we have
s s s
R/N = R/ Pi ∼ ∼
\ Y Y
= R/P i = Fi ,
i=1 i=1 i=1
proving the desired direction. 
Lemma 22.6. Let R be a ring with Kdim(R) = d. Then d + 1 ≤ Kdim(R[x]) ≤ 2d + 1.
Proof. Let P0 ( P1 ( P2 ( · · · ( Pd be a chain in Spec(R). Let Qi = Pi R[x] = {a0 +
a1 x + · · · + am xm : m ≥ 0, a0 , · · · , am ∈ Pi }. Then R[x]/Qi ∼= (R/P )[x]. Note that
(R/Pi )[x] is an integral domain. So Q0 ( Q1 ⊆ · · · ( Qd is a chain in Spec(R[x]). Then
R[x]/Qd ∼ = (R/Pd )[x] and (x) is a prime ideal in this ring. So by the correspondence of
chains, we have Q0 ( Q1 ( · · · ( Qd ( (Qd , x) is a chain of length d + 1.
For the other bound, suppose that there is a chain
Q0 ( Q1 ( Q2 ( · · · ( Q2d+1 ( Q2d+2
in Spec(R[x]). Let Pi = Qi ∩ R. Then P0 ⊆ P1 ⊆ · · · P2d+2 is a chain in Spec(R). So we have
a chain of length 2d+3. But note that this is impossible unless some ideals are equal, since we
cannot have more than d + 1 distinct ideals in this chain. By the pigeonhole principle, there
exists some i so that Pi = Pi+1 = Pi+2 . Thus we have Qi ( Qi+1 ( Qi+2 in Spec(R[x]) where
all three of them contain P . So by correspondence there exists a chain Q fi ( Q ]i+1 ( Qi+2
]
34
in Spec((R/P )[x]) = Spec(R[x]/P R[x]). Let S = R/P so we have Q fi ( Q] i+1 ( Qi+2 in
]
Spec(S[x]). Moreover Q fi ∩ S = Q
] i+1 ∩ S = Qi+2 ∩ S = (0).
]
Let T = S \ {0}, which is multiplicatively closed since S is an integral domain. Thus
−1 fj ∩ T = ∅ for j = i, i + 1, i + 2. So by the results from
T S = Frac(S) =: K. Then Q
[ [ −1
fi ⊆ Q
localization, there exist prime ideals Q ]i+1 ⊆ Qi+2 in T
] (S[x]) = (T −1 S)[x] = K[x].
c
But then Kdim(K[x]) = 1 since K is a field. This is a contradiction since we just constructed
a chain of length 2. 

23. March 3: Noether normalization

Theorem 23.1 (Noether normalization theorem). Let R be a finitely generated k-algebra
for some field k. Then there exists a k-subalgebra S of R such that
(1) S ∼
= k[x1 , x2 , . . . , xd ], where d = Kdim(R);
(2) R is a finitely-generated S-module.
Remark 23.1. Hard part of the proof of the normalization theorem is proving that d =
Kdim(R). This is the part we will get to later.
Proposition 23.2. If R is a finitely generated k-algebra then Kdim(R) < ∞.
Proof. Recall that R ∼ = k[x1 , . . . , xm ]/I. So Kdim(R) = Kdim(k[x1 , . . . , xm ]/I) ≤ Kdim(k[x1 ,
. . . , xm ]) by the correspondence. We showed that if Kdim(R) = s then s+1 ≤ Kdim(R[x]) ≤
2s + 1. By induction we have Kdim(k[x1 ]) = 1, and Kdim(k[x1 , x2 ]) ≤ Kdim k[x1 ][x2 ] ≤ 3,
and so forth. So by induction we have Kdim(k[x1 , . . . , xm ]) ≤ 2m − 1. 

Proof of the normalization theorem. Let m be the number of generators for the k-algebra R.
We will do this by induction on m. Let’s say R = k[a1 , . . . , am ] (not necessarily a polynomial
ring). Then {a1 , . . . , am } is a set of generators.
Start with the base case m = 1. Then R = k[a1 ]. So R ∼ = k[x]/I, where I = (0) or
I = (p(x)) where p(x) is the minimal polynomial of a1 . If I = (0) then R ∼ = k[x]. So
S = k[a1 ] ∼ = k[x] and R = S so R is a finitely generated S-module. If I = (p(x)) with
p(x) 6= 0, then R ∼ = k[x]/(p(x)) is e-dimensional k-vector space where e = deg p(x). In this
case, take S = k; and dimk R < ∞, so R is a finitely generated S-module.
Induction hypothesis: let’s assume that the claim holds whenever R is generated by fewer
than m elements. Consider the case when R = k[a1 , . . . , am ], the k-algebra generated by
a1 , . . . , am . First, suppose that a1 , . . . , am are algebraically independent over k. In this case,
we have R ∼ = k[x1 , . . . , xd ]. To see why, consider the map φ : k[x1 , . . . , xm ] → R defined by
xi 7→ ai . Note that ker φ = (0), so in this case S = R ∼ = k[x1 , . . . , xm ] and S is clearly a
finitely generated R-module as S = R.
Now suppose that a1 , . . . , am are not algebraically independent over k. So there exists a
non-trivial polynomial relation q(a1 , . . . , am ) = 0. Before proceeding, we turn to an exercise
which will appear in Assignment #4: prove that there exist natural numbers A1 , . . . , Am−1 >
0 such that q(x1 +xA m , . . . , rm−1 +rm
1 Am−1
, xm ) = CxD
m +p(x1 , . . . , xm ) where C 6= 0 is a constant
and p(x1 , . . . , xm ) is a polynomial such that the degree of xm is less than D.
We can find u1 , . . . , um such that ai = ui + uA m for all i = 1, . . . , m − 1 and am = um ,
1

A1 Ai
as ui = ai − um = ai − am ∈ R. Notice also that k[u1 , . . . , um ] = k[a1 , . . . , am ] = R. One
35
direction (⊇) follows since ai = ui + uA
m for all i < m and the other one (⊆) follows since
i

Ai
ui = ai − am .
Notice that (by the aforementioned exercise)
D−1
X
0 = q(a1 , . . . , am ) = q(u1 +uA Am−1
m , . . . , um−1 +um
i
, um ) = C ·uD
m+ pi (u1 , . . . , um−1 )uim (∗)
i=0

Now let T = k[u1 , . . . , um−1 ]. Then R = T [um ] = T + T um + · · · + T uD−1 m by (∗). So R

is a finite T -module. But T is generated by fewer than m elements, so by the induction
hypothesis there exists S ∼ = k[x1 , . . . , xd ] such that T is a finite S-module. So since R is
a finitely generated T -module and T a finitely generated S-module, then R is a finitely
generated S module. Thus we are done. 

To show that Kdim k[x1 , . . . , xd ] = d and that R is a finitely generated k[x1 , . . . , xd ]-module
implies that Kdim(R) = d, we will need the Gelfand-Kirillov dimension and integrality.
Let k be a field, and let A be a finitely generated k-algebra. And let V be a finite-
dimensional k-vector space with V ⊆ A such that 1 ∈ V and V contains a set of generators
of A. Define V 2 := spank {vw : v, w ∈ V } ⊃ V since 1 ∈ V . If x1 , . . . , xm is a basis for V
then {xi xj : i, j ∈ {1, . . . , m}} span V 2 . Similarly, we get the chain V ⊆ V2 ⊆ V 3 ⊆ · · · and
dimk V n < ∞ for all n ≥ 1.
We now define the Gelfand-Kirillov dimension:
Definition 23.3. The Gelfand-Kirillov dimension of a finitely generated k-algebra GKdim(A)
is defined to be
log(dim V n )
GKdim(A) := lim sup .
n→∞ log n
Remark 23.2. We will see that this does not depend on V . That is, we get the same result
for any finite-dimensional k-subspace of A that contains 1 and a set of generators. Also,
intuitively speaking one can think that if dim V n ∼ Cnd for some C > 0 then d = GKdim(A).
Example 23.4. GKdim(k[x]) = 1. Take V = k+kx = span{1, x}, and V n = span{1, x, x2 , . . . ,
xn } with dim V n = n + 1. Then
log(n + 1)
GKdim(k[x]) = lim sup = 1.
n→∞ log n
Example 23.5. What about GKdim(k[x, y])?
Let V = k + kx + ky = span{1, x, y}. Then
2
V n = span{xi y j : i + j ≤ n} so dim V n = n+2
2
∼ n2 .

24. March 5
Suppose that k is a field, A a finitely generated k-algebra, and V ⊆ A a finite-dimensional
k-vector subspaces of A with 1 ∈ V and V contains a set of generators for A (such V is said
to be a generating subspace for A). We defined the Gelfand-Kirillov dimension last time.
Proposition 24.1. The definition of the Gelfand-Kirillov dimension is independent of choice
of generating subspace.
36
Proof.
S n Suppose that V and W are two generating subspaces. Then V ⊆ V 2 ⊆ V 3 ⊆ · · · ⊆
2 3
S n
V = A, and W ⊆ W ⊆ W ⊆ · · · ⊆ W = A. It follows that there exists p, q ≥ 1 such
that V ⊆ W p and W ⊆ V q . Therefore it follows that V n ⊆ W pn for all n ≥ 1. Therefore
log(dim V n ) log(dim(W pn ))
lim sup ≤ lim sup
n→∞ log n n→∞ log n
log(dim(W pn )) log(pn)
= lim sup ·
n→∞ log(pn) log n
n
log(dim W )
≤ lim sup .
n→∞ log n
And similarly since W ⊆ V q , we see that
log(dim W n ) log(dim V n )
lim sup ≤ lim sup . 
n→∞ log n n→∞ log n
Proposition 24.2. If A is a finitely generated k-algebra then GKdim(A) = 0 if and only if
dimk A < ∞; and if dimk A > 0 then GKdim(A) ≥ 1.
Proof. If dimk A < ∞, then V = A is a generating subspace. But now V 2 = V 3 = · · · = A
for all n ≥ 1, so
log(dimk (A))
GKdim(A) = lim sup = 0,
n→∞ log n
as log(dimk (A)) is a finite quantity and log n → ∞. Conversely, suppose that dimk (A) = ∞,
and let V be a generating subspace for A. Thus V ( V 2 ( V 3 ( · · · . If V i = V i+1 for some
i then V i · V = V i+1 · V = V i+2 , so by induction we have V n = V i for all n ≥ i. This means
dimk A < ∞, a contradiction. SInce V ( V 2 ( V 3 ( · · · , we have dim V n ≥ n, and so
log(dim V n ) log n
lim sup ≥ lim sup = 1.
n→∞ log n n→∞ log n


Theorem 24.3. Let d ≥ 1. Then GKdim(k[x1 , . . . , xd ]) = d.
Proof. Let V = k + kx1 + kx2 + · · · + kxd . Then
V n = span{xi11 · · · xidd : i1 + · · · + id ≤ n}.
So dim V n = n+d . To see why, note that we can create a bijection between {xi11 · · · xidd :


d
i1 + · · · + id ≤ n} and the set of all ways of placing d X’s into n + d slots, and there are n+d


d
ways to do this. Since
nd
    
n+d (n + d) · · · (n + 1) 1
= = 1+O .
d d! d! n
d
log(dim V n ) log nd! (1 + O( n1 ))
=
log n log n
d log n − log d! + log(1 + O(n−1 )) d log n log d! O(n−1 )
= = − + →d
log n log n log n log n
as n → ∞. Thus GKdim(k[x1 , . . . , xd ]) = d. 
37
 Our main theorem that we hope to prove is the following:
Theorem 24.4. Let k be a field and let A be a finitely generated k-algebra. Then GKdim(A) =
Kdim(A).
Corollary 24.5. Kdim(k[x1 , . . . , xd ]) = d.
The proof of the main theorem will use two facts, one from Assignment #3 and the other
from Assignment #4:
(1) (from Assignment #3) If R ⊆ S and S is a finitely generated R-module, then S is
integral over R; hence Kdim(S) = Kdim(R) if S is a finite R-module.
(2) (from Assignment #4) If R ⊆ S and S is a finitely-generated R-module then
GKdim(R) = GKdim(S).
Proof of the main theorem. Notice that if A is a finitely generated k-algebra then by the
Noether normalization theorem there exists B ⊆ A such that B ∼ = k[y1 , . . . , yd ] such that A
is a finite B-module. Since A is finite as a B-module, we see that A is integral over B hence
Kdim(A) = Kdim(B) ≥ d.
Also A is finitely generated as a B-module, so GKdim(A) = GKdim(B) = d. So we sat
least know that Kdim(A) ≥ GKdim(A). Let
α = inf{GKdim(A) : A finitely generated such that Kdim(A) > GKdim(A)}.
So there exists A such that Kdim(A) > GKdim(A) and GKdim(A) < α + 21 . So now by
the Noether normalization theorem that there exists d and B ∼ = k[y1 , . . . , yd ] with B ⊆ A
such that [A : B] < ∞ so Kdim(A) = Kdim(B) and GKdim(A) = GKdim(B) = d. Since
Kdim(A) > GKdim(A), we have Kdim(A) ≥ d + 1. So we are in the situation where we may
assume Kdim(k[x1 , . . . , xd ]) ≥ d + 1. That means that there exists a chain (0) ( P1 ( · · · (
Pd+1 in Spec(k[x1 , . . . , xd ]). Then we claim that Kdim(k[x1 , . . . , xd ]/P1 ) ≥ d, and that if B
is a finitely generated k-algebra that is also an integral domain and I 6= (0) is an ideal of B
then GKdim(B/I) ≤ GKdim(B) − 1.
Once we have the claim, we see that Kdim(k[x1 , . . . , xd ]/P1 ) ≥ d and
GKdim(k[x1 , . . . , xd ]/P1 ) ≤ d − 1 < α. But by the definition of α, we see that
GKdim(k[x1 , . . . , xd ]/P1 ) = Kdim(k[x1 , . . . , xd ]/P1 ), which is a contradiction. We will prove
the necessary claims tomorrow and move on to transcendence degree. 
25. March 6
We need the following claim to finish off the proof:
Claim. Let A be a finitely generated k-algebra that is an integral domain. If I is a non-zero
ideal of A then GKdim(A/J) ≤ GKdim(A) − 1.
Proof. Let π : A → A/I be the canonical surjection. Let V be a generating space for A.
Then π(V ) =: V is a generating space for A/J. For each n ≥ 1, pick a subspace Wn of V n
n n
such that π(Wn ) = V and dim Wn = dim(V ).
Pick f ∈ V \{0} such that f ∈ I. Then we claim that the sum Wn +f Wn−1 +· · ·+f n−1 W1 ⊆
V n is a direct sum. Suppose that is not the case. Then there exists wi ∈ Wi , i = 1, . . . , n
not all zero so that wn + f wn−1 + f 2 wn−2 + · · · + f n−1 w1 = 0. So there exists a largest m ≤ n
such that wm 6= 0. So f n−m wm + · · · + f n−1 w1 = 0 with wm 6= 0. Since A is an integral
domain, we see that f wm−1 + · · · + f m−1 w1 ∈ I. Now apply π to see that π(wm ) = 0 and
38
 m m
π(I) = (0). But dim(Wm ) = dim(V ) and π(Wm ) = V . Hence π is injective meaning that
wm = 0, but this is a contradiction.
Now we want to show that GKdim(A/I) ≤ d − 1, where d = GKdim(A). Suppose that
n
GKdim(A/I) > d − 1. Then GKdim(A/I) ≥ d. Let ε < 1. This means that dim(V ) ≥ nd−ε
n
for infinitely many n. To see why, assume that this is not true. Then dim(V ) < nd−ε , hence
n
log dim(V ) (d − ε) log n
< =d−ε
log n log n
for all sufficiently large n. Hence
n
log dim V
GKdim(A/I) = lim sup ≤ d − ε < d,
n→∞ log n
n
which is a contradiction. By assumption, dim Wn = dim V . So dim Wn ≥ nd−ε for infinitely
n n+1 2n
many n. Now V ⊆ V ⊆ · · · ⊆ V , so dim Wn ≤ dim Wn+1 ≤ · · · ≤ dim W2n . Thus there
are infinitely many n such that dim W2n , · · · dim Wn ≥ nd−ε . Recall that W2n ⊕ f W2n−1 ⊕
f 2 W2n−2 ⊕ · · · ⊕ f 2n−1 W1 ( V 2n . So
dim V 2n ≥ dim(W2n ) + dim(f W2n−1 ) + · · · + dim(f n Wn ) + · · · + dim(f 2n−1 W1 )
≥ dim(W2n ) + dim(W2n−1 ) + · · · + dim(Wn ) ≥ nd−ε · n = nd+1−ε .
Therefore
log dim V 2n (d + 1 − ε) log n
≥ ,
log(2n) log 2n
so
log dim V n
d = GKdim(A) ≈ lim sup ≥ d + 1 − ε > d,
n→∞ log n
which is a contradiction. 
Corollary 25.1. Let A be a finitely generated k-algebra. Then Kdim(A[x]) = Kdim(A) + 1.
Proof. We already showed that Kdim(A[x]) ≥ Kdim(A) + 1. Now by Noether normalization
there exists B ⊆ A such that B ∼ = k[x1 , . . . , xd ], with d = Kdim(A) and A a finitely generated
B-module.
Write A = Ba1 + · · · + Bas . Then A[x] = B[x]a1 + · · · + B[x]as . Thus A[x] is a finitely
generated B[x]-module. And we know by Assignment #3 that Kdim(A[x]) = Kdim(B[x]) =
Kdim(k[x1 , . . . , xd ][x]) = d + 1 (note B[x] ∼ = k[x1 , . . . , xd ][x]). 
25.1. Transcendence degree.
Definition 25.2. Suppose that K is a field extension of a field k. Then a set S ⊆ K
is algebraically independent if every finite subset {x1 , . . . , xd } of S has the property that
if p(x1 , . . . , xd ) = 0 and p(t1 , . . . , td ) ∈ k[t1 , . . . , td ] then p(t1 , . . . , td ) = 0. We define the
transcendence degree of K/k to be
trdegk K = sup{#S : S algebraically independent subset of K}.
We will only worry about finite transcendence degree.

39
Theorem 25.3. Let k be a field and let A be a finitely generated k-algebra that is an integral
domain and let K = Frac(A). Then Kdim(A) = trdegk K.
Proof. Let d = Kdim(A). By Noether normalization, there exists B ∼ = k[x1 , . . . , xd ] with
B ⊆ A and A a finitely generated B-module. Let b1 , . . . , bd be a set of generators for B. So
B = k[b1 , . . . , bd ] ∼
= k[x1 , . . . , xd ]. Then the set {b1 , . . . , bd } ⊆ K is algebraically independent
over k. So trdegk K ≥ d.
Now suppose that trdegk K > d. Then there exists c1 , . . . , cd+1 ∈ K algebraically inde-
pendent over k. Since K = Frac(A), we can write each ci in the fraction form. Take the
least common multiple of the denominators. So there exist b ∈ A \ {0} and a1 , . . . , ad+1 ∈ A
such that c1 = a1 /b, . . . , cd+1 = ad+1 /b. Thus k[y1 , . . . , yd+1 ] ∼ = k[c1 , . . . , cd+1 ] ⊆ A[b−1 ]. Now
Kdim(A[b−1 ]) ≤ Kdim(A) = d. A[b−1 ] is finitely generated so GKdim(A[b−1 ]) ≤ d as well.
Thus k[c1 , · · · , cd+1 ] has G-K dimension d + 1 but A[b−1 ], which contains k[c1 , . . . , cd+1 ] has
G-K dimension d. This is a contradiction, according to the claim below.
Claim. If R ⊆ S and R, S finitely generated k-algebra, then GKdim(R) ≤ GKdim(S).
To see why the claim is true, start by picking V a generating space for R. Now add in a set
of generators for S to obtain a generating space W ⊃ V for S. Then dim(V n ) ≤ dim(W n )
for all n. Thus GKdim(R) ≤ GKdim(S). 

26. March 10
Definition 26.1. Given a ring R and P ∈ Spec(R), we define the height of P to be
ht(P ) := Kdim(RP ) = sup{n : Q0 ( Q1 ( · · · ( Qn , Qn ⊆ P, Qi ∈ Spec(R)}.
Definition 26.2. For any R, we define M-Spec(R) to be the set of maximal ideals of R, i.e.,
M-Spec(R) := {M : M a maximal ideal of R}.
Now we shall put a topology on Spec(R). Then M-Spec(R) will inherit the subspace
topology. Particularly, we will put the Zariski topology:
Definition 26.3. We define the closed sets of the Zariski topology as follows: if I is an ideal
of R, then
CI := {P ∈ Spec(R) : P ⊇ I}.
Let’s verify if this topology is indeed a topology. Clearly, CR = ∅ and Spec(R) = C(0) .
Let’s check if CI1 ∪ · · · ∪ CIn = CT Ij . This is clear since
P ∈ CI1 ∪ · · · ∪ CIn ⇔ P ⊇ I1 or P ⊇ I2 or · · · P ⊇ In
\n
⇔P⊇ Ij
j=1
T
For the last ⇔, the ⇒ is obvious, so let’s just show the ⇐ direction. Suppose that P ⊇ Ij
T does not contain any Ij . Then there exists aj ∈ Ij \P for j = 1, 2, . . . , n. So a1 a2 · · · an ∈
but
Ij ⊆ P. But then P is a prime ideal so aj ∈ P for some j. Contradiction, so we proved
what we wanted to show.
Notice that an arbitrary union of CIj ’s need not be closed.
40
Example 26.4. Let R = C[x], so Kdim(R) = 1, since (0) ( (x − λ) are the only available
chains of prime ideals. Suppose I = (p(x)). Then
CI = {(x − λ) : (x − λ) ⊇ (p(x))},
so there is a bijection between CI and {λ ∈ C : p(λ) = 0}, as (x − λ) ⊇ (p(x)) ⇔ x − λ | p(x).
Thus CI is a finite set. So in this case,

Spec(R)
 if I = (0)
CI = finite set if I 6= (0)

(0) if I = R.
S
Then Z := C(x−n) is not closed, since Z is countably infinite while Spec(R) is uncountably
n∈Z
infinite.
Finally, it remains to show that the arbitrary intersection of the CI ’s is closed. Namely,
we show
S
Proposition 26.5. CIα = CP Iα .
α
S P
Proof. P ∈ CIα ⇔ P ∈ CIα for all α ⇔ P ⊇ Iα for all α ⇔ P ⊇ Iα . 
α
√
Remark 26.1. We actually only need to consider CI when I = I, since CI = C√I . Note
√
that I ⊆ I, so if P ⊇ I then P ⊇ I. But if P ⊇ I, then
\ √
P ⊇ Q = I,
Q⊇I
Q prime
√
so P ⊇ I ⇔ P ⊇ I. Hence CI = C√I .
Remark 26.2. Clearly, if I ⊇ J, then CI ⊆ CJ .
We consider some specific examples of Spec(R).
Example 26.6. Spec(Q) = {(0)}. Spec(Z) = {(p) : p prime number}.
Let P ∈ Spec(R). Then what is the closure of P ? Recall that the closure P = CP = {Q ∈
Spec(R) : Q ⊇ P }. Then {P } =p{P } if and only if P is maximal. Also, P is dense (i.e.,
P = Spec(R)) if and only if P = (0).
Example 26.7. In Spec(Z), every point is closed; the dense point is (0). If I is an ideal in
Z, then either I = (0) ⇒ CI = Spec(Z), or for n ≥ 2, I = (n) ⇒ CI = {(p) : (p) ⊇ (n)} =
{(p) : p | n}. Clearly, if I = Z then CI = ∅. Thus this topology is a cofinite topology since
the complement of an open set is finite.
Example 26.8. Let R = Z(2) . Then there is a correspondence between Spec(Z(2) ) and the
set {P ∈ Spec(Z) : P ⊆ (2)}. (2) is a closed point in Z and (0) is the dense point in Z.
Similarly, by this correspondence, 2Z(2) is a closed point and (0) is the dense point. We will
elaborate on this point more in the following theorem:
41
Theorem 26.9. Let R be an integral domain. Suppose also that P ∈ Spec(R), and that
X = {Q ∈ Spec(R) : Q ⊆ P }
with the subspace topology. We showed that there is a correspondence:
bij’n
X ←→ Spec(RP )
f
Q 7−→ QRP
g
J ∩ R ←−[ J.
Then f and g are continuous bijections and so X is homeomorphic to Spec(RP ). If C is
a closed subset of Spec(RP ), then C = CI where I is an ideal of RP . That is, C = {Q ∈
Spec(RP ) : Q ⊇ I}.
Proof. Note that
f −1 (C) = {L ∈ Spec(R) : L ⊆ P, f (L) = LRP ⊇ I} = {L ∈ Spec(R) : L ⊇ I ∩ R} ∩ X,
which is closed in X. Also, note that LRP ⊇ I ⇔ L = LRP ∩ R ⊇ I ∩ R, as required.
Conversely, if C ⊆ X is closed, then
C = X ∩ {L ∈ Spec(R) : L ⊇ I}
for some radical ideal I. So
g −1 (C) = {LRP : L ⊇ I; L ⊆ P } = {LRP : L ⊇ I}
= {LRP : LRP ⊇ IRP } = {Q ∈ Spec(RP ) : Q ⊇ IRP } = CIRP ⊆ Spec(RP ). 

27. March 12

√ CI = {P√∈
Let R be a ring and X = Spec(R), where the closed sets are of the form
Spec(R) : P ⊇ I}. Recall that we can only consider the cases when I = I and J = J
hence CI = C√I and CJ = C√J . If CI ⊇ CJ , then {P : P ⊇ I} ⊇ {P : P ⊇ J}, hence I ⊆ J.
Conversely, if CI ⊇ CJ then whenever P ⊇ J then P ⊇ I. Hence
\ \
P ⊇ P.
P prime P prime
P ⊇J P ⊇I
√ √
So J⊇ I so J ⊇ I as required.
Remark 27.1. CI with the subspace topology on Spec(R) is homeomorphic to Spec(R/I).
This claim follows from correspondence. Let f : CI → Spec(R/I) such that f (P ) = P where
J is the image of J under the natural map J 7→ J/I. Then f is a bijection, since by the
correspondence theorem we have
CI = {P : P ⊇ I} ←→ Spec(R/I)
1−1
P ←→ P
To see that f isa homeomorphism, note that if J is an ideal of R/J then there exists an
ideal L ⊇ I of R such that L = J, by correspondence. So we have
f −1 (CJ ) = f −1 ({Q ∈ Spec(R/I) : Q ⊇ J}) = {P ∈ Spec(R) : P ⊇ I} = CL ⊆ CI ,
42
so f is indeed continuous. Conversely, if C is a closed subset of CI , then C = CL for some
L ⊇ I. Then note that f (CL ) = {Q ∈ Spec(R/I) : Q ⊇ I} = CL . Therefore f −1 is
continuous so f is a homeomorphism.
This example fits more generally into the following framework. Consider the following
problem from Assignment #4. In Assignment #4, you will show that if φ∗ : R → S is
a ring homomorphism, then an induced map φ : Spec(S) → Spec(R) defined as φ(P ) 7→
(φ∗ )−1 (P ) =: Q is a continuous map. In this setting, if φ∗ : R → R/I given by r 7→ r + I
gives a continuous map, then the image of φ : Spec(R/I) → Spec(R) is CI , and this φ = f −1 .
So far, we have been talking about closed sets. What about open sets then? Start with
f ∈ R. Let’s define U (f ) := Spec(R) \ C(f ) = {P ∈ Spec(R)} \ {P : P ⊇ (f ) ⇔ f ∈ P } =
{P ∈ Spec(R) : f ∈ P }. Note that {P ∈ Spec(R) : f ∈ P } ∼ = Spec(Rf ). Then U (f ) is
indeed open.
Definition 27.1. Such U (f ) is said to be a principal open set.
Remark 27.2. The sets U (f ) form a basis for the Zariski topology. Notice also that if U is
an
p open √set with U = Spec(R) \ CI and if f ∈ I then U (f ) ⊆ U . Note that if f ∈ I then
(f ) ⊆ I, so CI ⊆ C(f ) . Therefore Spec(R) \ C(f ) ⊆ Spec(R) \ CI , or U (f ) ⊆ U as desired.
Theorem 27.2. Let R be a ring. Then the following are equivalent:
(1) Spec(R) is disconnected
(2) R has an idempotent e 6= 0, 1
(3) R ∼
= R1 × R2 where both R1 and R2 are non-trivial.
Proof. ((1) ⇒ (2)) Suppose that Spec(R) is disconnected. Then there exist I, J ideals of R
such that Spec(R) = CI t CJ , where t denotes the disjoint union. So if CI ∩ CJ = ∅, then
CI+J = ∅. On the other hand, if CI ∪ CJ = Spec(R) then CI∩J = CIJ p = Spec(R). Note
T
that CI+J = ∅ ⇔ I + J = R, and that CIJ = Spec(R) ⇔ P prime P = (0) ⊇ IJ. Since
p
I + J = R, there exist x ∈ I and y ∈ J such that x + y = 1 and xy ∈ IJ ⊆ (0), i.e.,
(xy)n = 0 for some n ≥ 1. Now
   
2n 2n 2n 2n−1 2n n n
1 = (x + y) = x + x y + ··· + x y
1 n
| {z }
=:e
   
2n n−1 n+1 2n
+ x y ··· + xy 2n−1 + y 2n .
n+1 2n − 1
| {z }
=1−e

It is easy to see that e(1 − e) = 0. Why? note that e ∈ xn R and 1 − e ∈ y n R so

e(1 − e) ∈ (xy)n R = (0). So e2 = e. So we are done once we show that e 6= 0, 1. But
this is easy: since e ∈ xn R ⊆ I ( R and 1 − e ∈ J ( R, we see that e 6= 1 and 1 − e 6= 1.
Therefore e 6= 0, 1 as required.
((2) ⇒ (3)) Suppose that there exists e ∈ R, e 6= 0, 1 such that e2 = e. Let R1 = Re and
R2 = R(1 − e). Since e 6= 0, 1 neither R1 nor R2 is the zero ring. So R1 anD R2 are rings
with identities as e and 1 − e respectively. Now the following claim comes in handy:
Claim. If f is an idempotent in R then Rf is a ring with unit f .
43
 If r ∈ R then rf ∈ Rf . Thus (rf )f = rf 2 = rf , and f (rf ) = rf 2 = rf . Also,
(rf ) · (sf ) = rsf and rf + sf = (r + s)f as desired.
Define φ : R → R1 × R2 by φ(r) = (re, r(1 − e)). We claim that φ is an isomorphism. φ
is a homomorphism since
φ(r1 + r2 ) = ((r1 + r2 )e, (r1 + r2 )(1 − e))
= (r1 e, r1 (1 − e)) + (r2 e, r2 (1 − e)) = φ(r1 ) + φ(r2 )
φ(r1 r2 ) = (r1 r2 e, r1 r2 (1 − e))
= (r1 r2 e2 , r1 r2 (1 − e)2 ) = (r1 er2 e, r1 (1 − e)r2 (1 − e))
= φ(r1 )φ(r2 ).
For injectivity, suppose φ(r) = (0, 0). That is, re = 0 and r(1 − e) = 0. therefore re +
r(1 − e) = r = 0. As for surjectivity, given any ae ∈ Re and b(1 − e) ∈ R(1 − e), we have
φ(ae + b(1 − e)) = (ae, b(1 − e)). So R ∼
= R1 × R2 as claimed.
((3) ⇒ (1)) This one is quite immediate. If R = R1 ×R2 then I = R1 ×{0} is an ideal, and
similarly J = {0} × R2 is an ideal. Clearly IJ = (0, 0) and I + J = R so CI t CJ = Spec(R)
as required. 
Definition 27.3. A topological space X is reducible if there exist proper closed subsets
C1 and C2 (not necessarily disjoint) such that X = C1 ∪ C2 . Otherwise, X is said to be
irreducible.
Remark 27.3. Evidently, if X is disconnected, then X is automatically reducible. However,
the converse is false.
28. March 13
Can we find a ring R such that Spec(R) is connected but reducible?
Example 28.1. Let R = C[x, y]/(x, y). Let x, y denote the images of x and y in R. Then
Spec(R) is reducible since C(x) ∪ C(y) = Spec(R): x̄ȳ = 0 so if P ∈ Spec(R) then x̄ȳ ∈ P . So
x̄ ∈ P or ȳ ∈ P so P ∈ C(x̄) or P ∈ C(ȳ) . Now we need to show that Spec(R) is connected.
Notice
R = {c + x̄p(x̄) + ȳq(ȳ) : c ∈ C, p(t), q(t) ∈ C[t]}.
We showed that Spec(R) is disconnected if and only if there is an idempotent e 6= 0, 1. If
e = c + x̄p(x̄) + ȳq(ȳ) is an idempotent, so (c + x̄p(x̄) + ȳq(ȳ)2 = c2 + 2cx̄p(x̄) + 2cȳq(ȳ) +
x̄2 p(x̄)2 + ȳ 2 q(ȳ)2 . Hence deg(x̄p(x̄)) = deg(ȳq(ȳ)) = 0. Thus e = c ∈ C so e = 0, 1. So
Spec(R) is connected.
p
Theorem 28.2. Spec(R) is irreducible if and only if N := (0) is a prime ideal. Therefore,
Spec(R) is irreducible if and only if R/N is an integral domain.
Proof. Spec(R) ∼
p = Spec(R/N ), so we may assume without loss of generality that R is reduced
(i.e., (0) = (0)). Now if R is not an integral domain, then there exist a, b ∈ R \ {0} such
that ab = 0. Then Spec(R) = C(a) ∪ C(b) = C(a)∩(b) = C(ab) = C(0) = Spec(R). So if R is not
an integral domain, then Spec(R) is reducible.
If Spec(R) is reducible, then Spec(R) = CI ∪ CJ = CIJ . So IJ = (0) and I, J 6= (0).
Pick a ∈ I \ {0} and b ∈ J \ {0} such that ab = 0. Thus R is not an integral domain, as
desired. 
44
 √
Corollary 28.3. CI ⊆ Spec(R) is irreducible if and only if I is a prime ideal.
√ √
Proof. Spec(R/I) ∼ = C I = C √I ∼ = Spec(R/ I), and S = R/ I is reduced. Thus C is
√ I
irreducible if and only if S is an integral domain. This is equivalent to saying that I is a
prime ideal. 
Theorem
S 28.4. Let R be a ring. Then Spec(R) is quasi-compact. That is, if Spec(R) =
Uα where Uα open sets, then there exists a finite collection of open sets {Uα1 , . . . , Uαs }
s
S
such that Spec(R) = Uαi .
i=1
S
Proof. Suppose that Spec(R) = Uα , where Uα = Spec(R) \ CIα for some ideal Iα of R.
S α T
Then Uα = Spec(R)
P if and only if CIα = ∅ if and only if CP Iα = ∅; and this is equivalent
to saying that Iα = R. So there exist iαk ∈ Iαk such that iα1 + · · · + iαs = 1. This would
s
T s
T
mean that Iα1 + · · · + Iαs = R. Thus CP Iαi = ∅, so CIαi = ∅. So Uαi = Spec(R). 
i=1 i=1

Let R be Noetherian. When is Spec(R) Hausdorff?

Theorem 28.5. Suppose that R is a Noetherian ring. Then the following are equivalent:
(1) Kdim(R) = 0
(2) Spec(R) is compact and Hausdorff
(3) Spec(R) is finite and discrete p
(4) R/N ∼= F1 × · · · Fs where N = (0), s ≥ 1 and Fi fields.
Proof. Earlier we showed (1) ⇔ (4), and we just showed (3) ⇒ (2) ⇒ (1). Thus we only
need to show (4) ⇒ (3). If R/N ∼ = F1 × Fs then Spec(R) ∼= Spec(R/N ) ∼ = Spec(F1 × Fs ).
But since F1 × Fs is Noetherian and has Krull dimension 0, it has only finitely many prime
ideals, all of which are maximal. So | Spec(R)| < ∞ and each point is closed. Therefore the
topology is discrete. 
28.1. Noetherian topological spaces.
Definition 28.6. Let X be a topological space. We say that X is a Noetherian topological
space if whenever C1 ⊇ C2 ⊇ C3 ⊇ · · · is a descending chain of closed subsets of X then
there exists n such that Cn = Cn+1 = Cn+2 = · · · .
Definition 28.7. If X is a topological space, then we can define the Krull dimension of X
to be
Kdim(X) := sup{there exists a chain C0 ) C1 ) · · · ) Cn , Ci irreducible subsets}.
n
Example 28.8. If R is a ring then Kdim(R) = Kdim(Spec(R)).
28.2. Artinian rings.
Definition 28.9. A ring R is Artinian if every descending chain of ideals terminates, i.e.,
if I1 ⊇ I2 ⊇ · · · then there exists n such that In = In+1 = In+2 = · · · . An R-module M is
Artinian if it satisfied the descending chain condition on submodules.
Proposition 28.10. R is Artinian if and only if R is Artinian as an R-module.
Proof. Just as with the Noetherian case, the same proof shows that R is Artinian iff every
non-empty subset of ideals has a minimal element with respect to inclusion. 
45
 29. March 17
Lemma 29.1. If R is a ring and
π
0 → M1 → M → M2 → 0
is a short exact sequence of R-modules, then M is Artinian if and only if M1 and M2 are
Artinian.
Proof. (⇒) This is clear, since M2 ∼ = M/M1 and M1 ⊆ M .
(⇐) Suppose that M1 and M2 are Artinian, and let N1 ⊇ N2 ⊇ · · · be a descending chain
of submodules in M . Without loss of generality, let M1 ⊆ M . Then N1 ∩M1 ⊇ N2 ∩M1 ⊇ · · ·
is a descending chain in M , and since M1 is Artinian there exists i such that Ni ∩ M1 =
Ni+1 ∩ M1 = · · · . Similarly, we have that π(N1 ) ⊇ π(N2 ) ⊇ · · · , which must stabilize as M2
is Artinian (i.e., exists j such that π(Nj ) = π(Nj+1 ) = · · · .
Let n ≥ max(i, j). Then we claim that Nn = Nn+1 implies Nmax(i,j) = Nmax(i,j)+1 = · · · .
Thus the chain terminates so M is Artinian. Since n ≥ j we have π(Nn ) = π(Nn+1 ). We
also know that Nn+1 ⊆ Nn , so we need to show that Nn+1 ⊇ Nn . Suppose x ∈ Nn . Then
π(x) ∈ π(Nn ) = π(Nn+1 ), so there exists y ∈ Nn+1 such that π(x) = π(y). Hence π(x−y) = 0
so x, y ∈ M1 . But then x − y ∈ Nn , so x − y ∈ Nn ∩ M1 = Nn+1 ∩ M1 since n ≥ i. Hence
x − y ∈ Nn+1 so x ∈ y + Nn+1 ⊆ Nn+1 . 
Proposition 29.2. If R is a ring in which (0) = M1 M2 · · · Ms , where Mi ’s are maximal
ideals in R, then R is Artinian if and only if R is Noetherian.
Proof. (⇐) Let A1 = M1 , A2 = M1 M2 , A3 = M1 M2 M3 , . . . , As = M1 M2 M3 · · · Ms = (0) and
A0 = R. Suppose that R is Noetherian but not Artinian. Then A0 ∼ = R is not Artinian as
an R-module. However, As = (0) is Artinian as an R-module. Therefore, there must exist
some largest i such that Ai = M1 M2 · · · Mi is not Artinian but Ai+1 = M1 M2 · · · Mi+1 is.
Note that
0 → Ai+1 → Ai → Ai /Ai+1 → 0
is a short exact sequence. Our goal is to show that Ai /Ai+1 is Artinian, since that would
imply that Ai is Artinian, when R is Artinian (hence a contradiction). So it suffices to
show that Ai /Ai+1 = M1 M2 · · · Mi /M1 M2 · · · Mi+1 is Artinian. Notice that Mi+1 annihilates
Ai /Ai+1 so Ai /Ai+1 inherits the structure of an F -module, where F := R/Mi+1 . Since
R is Noetherian, Ai is Noetherian also. Therefore Ai /Ai+1 is Noetherian as an R-module.
Therefore Ai /Ai+1 is Noetherian as an F -module or equivalently as an F -vector space. Recall
that an F -vector space is Noetherian as an F -module if and only if the vector space is finite-
dimensional. Therefore dimF Ai /Ai+1 < ∞. Another fact: suppose that F is a field and V is
an F -vector space. Then V is Artinian as an F -module if and only if V is finite-dimensional.
Hence Ai /Ai+1 is an Artinian F -module, so Ai /Ai+1 is Artinian as an R-module, as Mi+1
annihilates Ai /Ai+1 . Contradiction!
(⇒) Let A1 = M1 , A2 = M1 M2 , A3 = M1 M2 M3 , . . . , As = M1 M2 M3 · · · Ms = (0) and
A0 = R. Suppose that R is Artinian but not Noetherian. Then A0 ∼ = R is not Noetherian as
an R-module. However, As = (0) is Noetherian as an R-module. Therefore, there must exist
some largest i such that Ai = M1 M2 · · · Mi is not Noetherian but Ai+1 = M1 M2 · · · Mi+1 is.
Note that
0 → Ai+1 → Ai → Ai /Ai+1 → 0
46
is a short exact sequence. Our goal is to show that Ai /Ai+1 is Noetherian, since that
would imply that Ai is Noetherian, whence R is Noetherian (hence a contradiction). So it
suffices to show that Ai /Ai+1 = M1 M2 · · · Mi /M1 M2 · · · Mi+1 is Noetherian. Notice that Mi+1
annihilates Ai /Ai+1 so Ai /Ai+1 inherits the structure of an F -module, where F := R/Mi+1 .
Since R is Artinian, Ai is Artinian also. Therefore Ai /Ai+1 is Artinian as an R-module.
Therefore Ai /Ai+1 is Artinian as an F -module or equivalently as an F -vector space. Recall
that an F -vector space is Artinian as an F -module if and only if the vector space is finite-
dimensional. Therefore dimF Ai /Ai+1 < ∞. Another fact: suppose that F is a field and
V is an F -vector space. Then V is Noetherian as an F -module if and only if V is finite-
dimensional. Hence Ai /Ai+1 is an Noetherian F -module, so Ai /Ai+1 is Noetherian as an
R-module, as Mi+1 annihilates Ai /Ai+1 . Contradiction! 

30. March 19
Theorem 30.1. R is Artinian if and only if R is Noetherian and has Kdim(R) = 0.
Proof. (⇐) If R is Noetherian and has Krull dimension 0, then by Noether’s result, there
exist prime ideals P1 , . . . , Ps such that P1 P2 · · · Ps = (0). Since Kdim(R) = 0, each Pi is
maximal so (0) is a product of maximal ideals and R is Noetherian. Thus by Proposition
29.2, R is Artinian.
(⇒) For this direction, we need to prove a few claims first.
Claim (Claim 1). Let R be Artinian and let P ∈ Spec(R). Then P is maximal. In particular,
Kdim(R) = 0.
Proof of Claim 1. Let S = R/P . By correspondence, S is Artinian and S is an integral
domain. We will then show that S is a field. Let x ∈ S \ {0}. Consider the chain xS ⊇
x2 S ⊇ x3 S ⊇ · · · . Since S is Artinian, there exists n such that xn S = xn+1 S. So if
xn ∈ xn+1 S, then there exists y ∈ S such that xn = xn+1 y. Divide both sides by xn to get
xy = 1. Therefore x has an inverse meaning that S is a field. 
Claim (Claim 2). Let R be Artinian. Then Spec(R) is finite.
Proof of Claim 2. Suppose that we have distinct prime ideals P1 , P2 , . . . . Note that by Claim
1, we know all of them are maximal. Consider the chain
P1 ⊇ P 1 ∩ P 2 ⊇ P1 ∩ P2 ∩ P3 ⊇ · · · .
This chain terminates, so there exists n such that
n
\ n+1
\
Pi = Pi .
i=1 i=1

Therefore P1 · · · Pn ⊆ Pn+1 . Recall that the Pi ”s are all distinct and all maximal by Claim
1. Therefore there exists ai ∈ Pi \ Pn+1 for all i = 1, . . . n. So a1 · · · an ⊆ P1 · · · Pn ⊆ Pn+1 .
But this contradicts the fact that Pn+1 is prime and none of ai ’s are in Pn+1 . The claim
follows. 
Claim (Claim 3). If R is Artinian then J(R) is nilpotent.
47
Proof. Let J = J(R). Consider the chain J ⊇ J 2 ⊇ J 3 ⊇ · · · . Since R is Artinian,
there exists n such that J n = J n+1 = · · · . So J n = J 2n . Trick: Let S := {I ⊆ R :
I ideal of R, I ⊆ J n , IJ n 6= (0)}. Note that S 6= ∅, since J n ∈ S : notes that J n J n =
J 2n = J n = (0). Since R is Artinian, there exists a minimal elements L ∈ S . So LJ n 6= (0)
but if L0 ( L then L0 J n = (0). Since LJ n 6= (0), there exists x ∈ L such that xJ n 6= (0).
We claim that L = Rx. Since x ∈ L, clearly Rx ⊆ L. Since xJ n 6= (0), we have
J n Rx 6= (0). Thus by minimality L = Rx. Now by assumption we have J n L 6= (0)., and
J n J n L = J n L 6= (0). Thus J n L = L by minimality. But then since J n L ⊆ JL = L, by
Nakayama’s lemma L = (0) (recall that L = Rx is finitely generated). Contradiction! 
Claim 1 gives us Kdim(R) = 0. Claim 2 says that Spec(R) = {P1 , . . . , Pk }, k ≥ 1.
Moreover, by Claim 1, P1 , . . . , Pk are maximal so
k
\
J(R) = Pi ⊇ P 1 · · · P k .
i=1

So by Claim 3, we know there exists m ≥ 1 such that J(R)m = (0). Therefore J(R)m =
(0) ⊇ (P1 P2 · · · Pk )m , which is a finite product of maximal ideals. So (0) is the finite project
of maximal ideals, and since R is Artinian, Lemma 29.2 give us that R is Noetherian. 
Corollary 30.2. If R is Artinian and J(R) = (0), then R ∼ = F1 × · · · × Fs with s ≥ 1 and
Fi fields.
Proof. (0) = J(R) = P1 ∩ P2 ∩ · · · ∩ Pk where each Pi is maximal. Therefore Pi ’s are pairwise
co-maximal. So by the Chinese remainder theorem we have
s s
∼
\ Y
R = R/ Pi = R/Pi .
i=1 i=1

Letting Fi := R/Pi yields the result. 

Definition 30.3. Let R be a noncommutative ring. Then we say that R is (left-)Artinian
if ever descending chain of left ideals L1 ⊇ L2 ⊇ · · · terminates. The Jacobson radical of R
J(R) is defined to be \
J(R) = M.
M max. left ideals

Theorem 30.4 (Artin-Wedderburn theorem). Let R be a ring (not necessarily commuta-

tive). If R is left-Artinian with J(R) = (0), then
Ys
R= Mni (Di ),
i=1

where Mni (Di ) is the matrix ring over the division ring Di .

48
30.1. Primary decomposition.
Remark 30.1 (Motivation).
√ If I is an ideal of R and I ⊆√Pi for i = 1, . . . , s are the maximal
ideals so that I = P1 ∩ · · · ∩ Ps . In particular, if I = I then I is a finite intersection of
prime ideals.
Definition 30.5. Let I a proper ideal of R. We say that I is primary if whenever xy ∈ I we
have either x ∈ I or there exists n such that y n ∈ I. Equivalently, I is primary if whenever
xy ∈ I at least one of the following holds:
• x∈I
• y∈I
• there exists n such that xn , y n ∈ I.
Example 30.6. Let R = Z. Then 6Z is not primary since if x = 3, y = 2 then xy ∈ 6Z but
x∈ / 6Z and y n ∈ / 6Z for all n ≥ 1. 8Z is primary: if xy ∈ 8Z then 8 | xy. Therefore either
8 | x or 2 | y so 8 | y 3 . Thus either x ∈ 8Z or y 3 ∈ 8Z. 3Z is primary, since if xy ∈ 3Z (i.e.,
3 | xy) then either x ∈ 3Z or y 1 ∈ 3Z.
Lemma 30.7. Let n > 1. Then nZ is primary if and only if n = pk for some p prime and
and k ≥ 1.
Proof. If n 6= pk then n = ab with a, b > 1 and gcd(a, b) = 1. So n - a, n - b hence n - ak , n - bk .
Conversely, if n = pk Z and xy ∈ pk Z then either pk | x, pk | y or (p | x and p | y). Therefore
pk | xk and pk | y k . Thus either x ∈ pk Z, y ∈ pk Z or xk , y k ∈ pk Z. 
√
Proposition 30.8. Let Q be primary. Then Q is a prime ideal.
√ √
Proof. Suppose
√ that Q is not
√ prime. THen there exist x, y such that xy ∈ Q but
n n
x, y ∈/ Q. Because xy ∈ Q we have x y ∈ Q for some n ≥ 1. Now no power of
x can be in Q, and the same holds for y. This means that Q is not primary, which is a
contradiction. 

31. March 20
Recall that last time we proved that
(1) If P is prime then P √is primary.
(2) If P is primary then P is prime.
In general, the converse of (2) does not hold.
Example 31.1. Let R := C[x, y, z]/(xy − z 2 ) = C[x̄, ȳ, z̄] where x̄ȳ = z̄ 2 . Let P = (x̄, z̄) ⊆ R.
Notice that R/P = C[x̄, ȳ, z̄/(x̄, z̄) ∼ = C[x, y, z]/(xy − z 2 , x, z) = C[x, y, z]/(x, z) ∼ = C[y]
which √ is an integral domain. Since R/P is an integral domain, P is prime. Let Q = P 2 .
Then Q = P prime but we claim that Q is not primary. Clearly x̄ · ȳ ∈ Q, since x̄ · ȳ =
z̄ · z̄ ∈ P 2 = Q. If Q is primary, either x̄ ∈ Q or ȳ ∈ Q for some n ≥ 1.
Notice that if ȳ n ∈ Q = P 2 then ȳ n ∈ P . But R/P ∼ = C[t] (let x̄, z̄ 7→ 0 and ȳ 7→ t) so
n
ȳ ∈ / P because tn 6= 0. Also, x̄ ∈ / Q. Why? Note that P 2 = (x̄, x̄z̄, z̄ 2 ). This means that
x̄ ∈ P 2 if and only if x ∈ (x2 , xz, z 2 , xy − z 2 ), but this cannot happen. Hence ȳ n ∈ / Q for all
n ≥ 1 and x̄ ∈ Q so Q is not primary.
There is a partial converse, however.
49
 √
Proposition 31.2. Let Q be an ideal of R and suppose that P := Q is a maximal ideal.
Then Q is primary.
Proof. Let S = R/Q. Then S is a local ring with the unique maximal ideal P/Q =: M .
Since S is local, if x ∈ S \ M is a unit in S. Suppose that xy ∈ Q. This means x̄ · ȳ = 0 ∈ S.
If x̄ ∈
/ M , then x̄ is a unit. So ȳ = 0 in S so y ∈
/ Q. If ȳ ∈
/ M , then ȳ is a unit. Therefore
x̄ = 0 ∈ S so x ∈ Q. If x̄, ȳ ∈ M then there exists n such that x̄n = ȳ n = 0 so xn and y n are
in Q. 
Definition 31.3. Let I be a proper ideal of R. We say that I is reducible if I = J ∩ K
for some ideals J, K where J ) I and K ) I. I is irreducible if whenever I = J ∩ K and
J, K ⊇ I ideals, we have J = I or K = I.
Proposition 31.4. Let R be a Noetherian ring. Then every proper ideal is a finite inter-
section of irreducible ideals.
Proof. Suppose otherwise. Then S := {I : I ideal; not a finite intersection of irreducible
ideals} is non-empty. Since R is Noetherian, there exists a maximal element J of S . J is
not irreducible, so there exists I.K ) J such that J = I ∩ K. Now I, K ∈ / S by maximality
of J, so I = L1 ∩ L2 ∩ · · · ∩ Ls , and K = N1 ∩ · · · ∩ Kt where each of the Li ’s and Nj ’s is
irreducible. So J = I ∩ K = L1 ∩ · · · ∩ Ls ∩ N1 ∩ · · · ∩ Nt , so J is a finite intersection of
irreducible ideals. So S = ∅, as desired. 
Theorem 31.5. Let R be a Noetherian ring. Then every ideal I of R has a decomposition
I = Q1 ∩ Q2 ∩ · · · ∩ Qs where each Qi is primary.
This theorem follows immediately from the following lemma:
Lemma 31.6. Let R be a Noetherian ring. Then every irreducible ideal is primary.
Proof. Let I be an irreducible ideal, and let S := R/I, and suppose that x, y ∈ R and xy ∈ I.
We need to show that either x ∈ I or y n ∈ I for some n ≥ 1. In S, we have x̄ · ȳ = 0 and
we must show that either x̄ = 0 or ȳ n = 0 for some n ≥ 1. We also know that (0) is an
irreducible ideal in S. We know this because of correspondence. Suppose that x̄ 6= 0. We
will show that ȳ n − 0 for some n ≥ 1 and we are done. For m ≥ 1, let
Jm = {a ∈ S : aȳ m = 0},
which is an ideal of S; and we also hate J1 ⊆ J2 ⊆ J3 ⊆ · · · . Therefore, since R is Noetherian,
there exists n such that Jn = Jn+1 . In other words, if aȳ n+1 = 0, then aȳ n = 0. From this
we claim that
Claim. (0) = (x̄) ∩ (ȳ n ).
Suppose that a ∈ (x) ∩ (ȳ n ). Then a = bx̄ and a = cȳ n . So aȳ = bx̄ȳ = 0. Hence
aȳ = cȳ n · ȳ = cȳ n+1 = 0. So c ∈ Jm+1 = Jm . Thus cȳ n = a = 0. So (0) = (x̄) ∩ (ȳ n ). But
(0) is irreducible, so ȳ n = 0 since we assumed that x̄ 6= 0. So (0) is primary, which implies
that I is primary also. 
Remark 31.1. This is quite nice for R Noetherian
√ integral domain of Krull dimension 1. Here
Q is primary√if and only if Q = (0) or Q is maximal. Then every non-zero I = Q1 ∩ · · · ∩ Qs
where each Qi is maximal.
50
 32. March 24: Valuation rings
Definition 32.1. Let K be a field. A valuation ν on K is a map ν : K → Z ∪ {∞} such
that:
(1) ν(a) = ∞ ⇔ a = 0,
(2) ν(ab) = ν(a) + ν(b) where ∞ + n = n + ∞ = ∞ + ∞ = ∞,
(3) ν(a + b) ≥ min(ν(a), ν(b)).
Example 32.2 (p-adic valuation). Let K = Q and p a prime number. For n ∈ Z non-zero,
write n = pk n0 with p - n0 . Define νp (n) = k; and if m/n ∈ Q with m ∈ Z and n non-zero
integer, define ν(m/n) := ν(m) − ν(n). One can check easily that this is well-defined and is
a valuation.
Example 32.3. Let K be the field of rational functions on C, C(x). Given f (x) ∈ K \ {0},
we shall define

n
 if f (x) has a zero at x = 0 of order n ≥ 0
ν(f (x)) = −n if f (x) has a pole at x = 0 of order n ≥ 0

0 if f (x) is analytic at x = 0 and f (0) = 0
Time for some comparison between the two:
K = C(x) K=Q
Starting ring C[x] Z
Valuation ν(p(x)) = order of zero at x = 0 ν(n) = k where pk k n
Associated ideal (x) pZ
k k+1
ν(p(x)) = k ⇔ p(x) ∈ (x) \ (x) ν(n) = k ⇔ n ∈ (pZ)k \ (pZ)k+1
Definition 32.4. Let K be a field and let ν : K → Z ∪ {∞} be a valuation. We define the
valuation ring of ν:
Oν = {a ∈ K : ν(a) ≥ 0}.
As it turns out, Oν is a ring. If a, b ∈ Oν then ν(ab) = ν(a) + ν(b) ≥ 0 so ab ∈ Oν ; similarly,
ν(a + b) ≥ min(ν(a), ν(b)) ≥ 0 so a + b ∈ Oν . Finally, since ν(0) = ∞ and ν(1) = 0, it
follows that 0, 1 ∈ Oν . But there is more.
Remark 32.1. Oν is a local ring with the maximal ideal Mν := {a ∈ Oν : ν(a) > 0}. It
is straightforward to see that Mν is an ideal. If x ∈ Oν \ Mν then ν(x) = 0 (and x is a
unit). Therefore, ν(x−1 · x) = ν(1) = 0. Therefore, 0 = ν(x) + ν(x−1 ), so ν(x−1 ) = 0. Hence
x−1 ∈ Oν so x ∈ Oν∗ . Hence Mν is the unique maximal ideal of Oν .
Example 32.5. Let K = Q, p = 2, and ν be the 2-adic valuation. What is Oν ? Let
na o
Oν = : a, b ∈ Z, b ≡ 1 (mod 2) , M = 2Oν .
b
Example 32.6. K = C(x), and ν be the valuation as defined before.Then
 
p(x)
Oν = : q(0) 6= 0 .
q(x)
Then Mν = xOν .
51
 33. March 26
Definition 33.1. A ring of the form Oν is called a discrete valuation ring (DVR).
Remark 33.1. Recall that Oν is a local ring with the unique maximal ideal Mν := {a : ν(a) >
0} ⊆ Oν .
Proposition 33.2. Let R be a DVR. Then R is a PID.
Proof. Let I be an ideal of R. If I = (0) or I = R, then I is evidently principal. So without
loss of generality assume that (0) 6= I ⊆ Mν . Pick a ∈ I with ν(a) minimal. We say that
I = (a). To see this, suppose that ν(y) ∈ I \ {0} where ν(y) ≥ ν(a). Let K = Frac(R).
Then ya−1 ∈ K and ν(ya−1 ) = ν(y) − ν(a) ≥ 0. Therefore ya−1 ∈ R. So y = (ya−1 )a ∈ R,
so Ra ⊆ I ⊆ Ra. Thus I = (a). 
Question. What are the possible Krull dimensions of a PID?
Solution: 0 and 1, and that’s it. Let’s see why. Suppose that R is a PID. Consider the
chain (0) ( P ( Q with P, Q prime ideals. Since R is a PID, there exist x, y ∈ R such that
P = (x) and Q = (y). x ∈ (y) so x = ay for some a ∈ R. So ay ∈ (x) so a ∈ (x). Thus
a = xb for some b, so x = ay = xby. Hence 1 = by so 1 ∈ (y). But this is a contradiction
since (y) is proper. Therefore Kdim(R) ≤ 1.
Corollary 33.3. If R is a DVR with a non-trivial valuation, then R is a PID, so R is
Noetherian with Krull dimension 1.
Proof. We saw that R is a PID, so it is Noetherian and is finitely generated. We know that
R has the (non-zero) unique maximal ideal Mν which is clearly a prime ideal. Therefore
Kdim(R) ≥ 1. Hence Kdim(R) = 1. 
Theorem 33.4. Let R be a DVR. Then R is integrally closed (that is, if x ∈ K be integral
over R (that is, if K = Frac(R) and x ∈ K and x is a solution to some monic polynomial
over R, then x ∈ R).
Proof. Let K = Frac(R) and let x ∈ K be integral over R. Then there exists n ≥
1, r0 , r1 , . . . , rn−1 ∈ R such that
xn + rn−1 xn−1 + · · · + r1 x + r0 = 0.
To show that x ∈ R, we need to show that ν(x) ≥ 0. Suppose toward a contradiction
that ν(x) = c < 0. Then ν(xn ) = nc and ν(ri xi ) = ν(ri ) + ν(xi ) = ν(ri ) + ic ≥ ic for all
i < n. Now xn = −rn−1 xn−1 − rn−2 xn−2 − · · · − r1 x − r0 . Note that ν(a1 + a2 + · · · + an ) ≥
min(ν(a1 ), . . . , ν(an )), since ν is a valuation. So we have
nc = ν(xn ) = ν(−rn−1 xn−1 − · · · − r1 x − r0 ) ≥ min(ν(rn−1 xn−1 ), . . . , ν(r0 ))
≥ min((n − 1)c, (n − 2)c, . . . , 0) = (n − 1)c.
So nc ≥ (n − 1)c, hence c ≥ 0 as required. 

52
 34. March 27
Theorem 34.1. Let A be a Noetherian local domain of Kdim(A) = 1. Let P be its maximal
ideal of A. Then the following are equivalent:
(1) A is a DVR
(2) A is integrally closed
(3) P is principal
(4) dimk P/P 2 = 1, where k := A/P , also called a residue field
(5) every non-zero ideal of A is of the form P n for some n ≥ 0
(6) there exists a non-zero x in A such that every non-zero ideal of A is of the form (xm )
for some m ≥ 0.
Remark 34.1 (Why DVRs are useful). To see why DVRs are useful, we need to consider the
general strategy in commutative algebra.
(1) Input: Some Noetherian integral domain R and some problem
(2) Step 1: Show that you can reduce to the case where R is integrally closed by consid-
ering its integral closure.
(3) Step 2: For each prime P of height 1 in R, we have R ,→ RP , which is a DVR
(Noetherian and has Krull dimension 1).
Example 34.2. Let p be a prime. Then consider X = Z/pZ × Z/p2 Z × · · · . Note
that Zp ⊆ X, and that Z/pi Z can be given the discrete, compact topology. So X is
a compact topological space and one can show that Zp ⊆ X is a closed subset of X.
Thus Zp is a compact topological space under the subspace topology – and it is also
a ring., with the usual multiplication and addition. Clearly Z ,→ Zp with injection
given by n 7→ ([n]p , [n]2p , . . . ). Also the image of this injection is dense. Thus Zp is a
DVR with the maximal ideal pZp .
Now let’s move to the proof of Theorem 34.1.
Proof. ((1) ⇒ (2)) We just did it today.
((2) ⇒ (3)) Pick x ∈ P \ P 2 . Why can we do this? By Nakayama’s p lemma if P = P 2 then
P = J(A)P = P 2 , so P = (0) since P is finitely-generated. Now (x) = P , so there exists
m ≥ 1 such that P m ⊆ (x). Pick the smallest n so that P n ⊆ (x). If n = 1 then we are done,
since P = (x). So let n > 1, without loss of generality so that P n−1 ( (x). Pick b ∈ P n−1 \(x).
Let y = b/x. Then y ∈ / A. Otherwise, we would have b = yx ∈ (x), a contradiction. On the
other hand, P y = (P b)/x because since b ∈ P n−1 , we have P b/x ⊆ P n x−1 ⊆ A recall that
P n ⊆ (x). Thus P y ⊆ A so P y is an ideal of A. Thus, either P y ⊆ P or P y = A. If P y = A,
then P b/x = A so P b = Ax. But since n > 1, b ∈ P n−1 ⊆ P . Thus Ax ⊆ P · P = P 2 . But
this contradicts the fact that a ∈ P \ P 2 . If P y = P , then P ⊆ A is a finitely-generated
A-module with A an integral and P y ⊆ P . So y is integral over A. But since A is integrally
closed, y ∈ A. This is a contradiction! Therefore n = 1 and P = (x).
((3) ⇒ (4)) Suppose that P = (x), x 6= 0. We need to show that dimk P/P 2 = 1. Write
P = Ax. So P/P 2 = Ax/P 2 . Now if a ∈ A then ax + P 2 = ā(x + P 2 ), with ā ∈ k = A/P .
So P/P 2 = k(x + P 2 ). The claim follows.
((4) ⇒ (5)) We observe that we can get ((4) ⇒ (3)) by applying Nakayama’s lemma. Notice
that P/P 2 is one-dimensional over k, so there exists x ∈ P \ P 2 such that x + P 2 generates
P/P 2 as an A/P -module. Therefore P/P 2 is a finitely-generated A/P -module. We claim
53
that P = (x). Why? Suppose M = P/(x). Then P M = (P 2 + (x))/(x) = P/(x) = M .
Recall that P = J(A), so by Nakayama’s lemma we have √ P/(x) = M = (0). Therefore
P = (x). Now let I 6= (0) be an ideal of A. Again, let I = P so that there exists n such
that P n ⊆ I. In particular, there exists some largest natural number m such that I ⊆ P m .
Otherwise, we would have I ⊆ P n and P n ⊆ I, meaning I = P n and we will be done.
So we have I 6⊆ P m+1 = (xm+1 ). Pick y ∈ I \ (xm+1 ). But then y ∈ (xm ) so y = axm ,
where a ∈/ (x) = P . So a ∈ A \ P . Therefore a is a unit so (y) = (xm ) ⊆ I ⊆ P m = (xm ) so
in fact P m = I.
((5) ⇒ (6)) Pick x ∈ P \ P 2 . Now (x) = P m for some m ≥ 0. Notice that 1 ≤ m < 2,
since x ∈ P \ P 2 . Hence m = 1 so P = (x). Thus P n = (xn ) for all n ≥ 0, as required.
((6) ⇒ (1)) Define a map ν : A \ {0} → N≥0 by ν(a) = m, where m ≥ 0 is the unique
non-negative integer such that a ∈ P m \ P m+1 . But here is one question we need to address
before proceeding: why can’t 0 6= a be in (xm ) for all m ≥ 0? You will see why in Assignment
#4. For this reason, ν is well-defined.
We extend ν : K → Z ∪ {∞} by ν(a/b) = ν(a) − ν(b) for a, b ∈ A with b 6= 0 (as in the
example with Qp ) and ν(0) = ∞. Then we claim that ν is a valuation and that A is the
valuation ring of ν.
Since A is local, we have P = (x). If a ∈ A \ {0}, then (a) = (xm ) so a = uxm for some
unit u (i.e., ν(u) = 0). Similarly, if a, b ∈ A \ {0} then a = uxm and b = u0 xn where u, u0
units. Thus a/b = u(u0 )−1 xm−n , so ν(a/b) = m − n.
Notice that ν(a/b) ≥ 0 ⇔ m − n ≥ 0 ⇔ a/b = u(u0 )−1 xd with d = m − n ≥ 0.
Hence a/b ∈ A. So A = {a/b ∈ K ∗ : ν(a/b) ≥ 0} ∪ {0}. Now we should check that
ν is a valuation. Indeed, if α = u1 xd1 , β = u2 xd2 with u1 , u2 ∈ A∗ and d1 , d2 ∈ Z, then
ν(αβ) = d1 +d2 = ν(α)+ν(β). Also, note that α+β = u1 xd1 +u2 xd2 = xd1 (u1 +u2 xd2 −d1 ) ∈ A.
If d1 ≤ d2 , then ν(α + β) = d1 + ν(u1 + u2 xd2 −d1 ) ≥ d1 . Similarly, ν(α + β) ≥ d2 if d2 ≤ d1 .
So ν is indeed a valuation, as desired. 
35. March 31: Dedekind domains
Definition 35.1. A Dedekind domain is a Noetherian integral domain A that is integrally
closed (in Frac(A)) and has Kdim(A) = 1.
Remark 35.1. If P is a maximal ideal of A, then AP is a DVR.
Example 35.2. Z, k[t] (k field) are Dedekind domains. If [K : Q] < ∞ then OK , the integral
closure of Z in K, is a Dedekind domain.
Theorem 35.3. In a Dedekind domain R, every non-zero ideal I has a factorization into
prime ideals
I = P1m1 · · · Prmr .
Moreover, this factorization is unique up to permutation of factors.
Historically speaking, Dedekind domains arose in number theory with FLT. It used to be
thought that by some that Z[e2πi/n ] is a UFD – until Kummer pointed out that this is wrong.
Remark 35.2 (Strategy for proving Theorem 35.3). Our strategy is as follows:
(1) Use primary decomposition: I = Q1 ∩ · ∩ Qs , with Qi primary
(2) Show that in a Dedekind domain, (0) 6= Q primary implies that Q = P m where P is
maximal and m ≥ 1.
54
 (3) Use (1) and (2) to show that I = P1m1 ∩ · · · ∩ Psms with each Pi distinct
(4) Show that P1m1 ∩ · · · ∩ Psms = P1m1 · · · Psms = I.
(5) Use local rings to prove uniqueness.
Note that Step 4 follows from the following fact: if I1 , . . . , Is are pairwise comaximal ideals
of R, then I1 ∩ I2 ∩ · · · ∩ Is = I1 I2 · · · Is . The inclusion I1 I2 · · · Is ⊆ I1 ∩ · · · ∩ Is is immediate.
Fro the reverse inclusion, let’s consider the s = 2 case for insight. If I1 + I2 = R, then there
exist a ∈ I1 , b ∈ I2 such that a + b = 1. If x ∈ I1 ∩ I2 , thenTx = x · 1 = x · a + x · b ∈ I1 I2 . In
general, if I1 , . . . , Is are pairwise comaximal, then Ii and Ij are comaximal.
j6=i
Without loss of generality, consider I1 and I2 ∩ I3 ∩ · · · ∩ Is . We know for each j, there
exist aj ∈ I1 and bj ∈ Ij such that aj + bj = 1. Thus I1 and I2 · · · Is are comaximal.
b3 ) · · · (as + bs ) = x + b2 b3 · · · bs ∈ I2 ∩ · · · Is . For each j there
So 1 = (a2 + b2 )(a3 + Q
exist cj ∈ Ij and d ∈ Ik such that cj + dj = 1. Therefore if x ∈ I1 ∩ · · · ∩ Is then
k6=j
x = x · 1 = x(c1 + d1 ) · · · (cs + ds ) ∈ I1 I2 · · · Is , so I1 ∩ · · · ∩ Is = I1 I2 · · · Is , as required.
Proposition 35.4. Let A be a Noetherian integral domain of Krull dimension 1. Then the
following are equivalent:
(1) A is integrally closed (equivalently, A is a Dedekind domain)
(2) Every primary ideal in A is of the form P m where m ≥ 1 and P is prime
(3) If P is a prime ideal of A, then AP is a DVR.
Proof. ((1) ⇒ (3)) Suppose A is integrally closed. If P is a prime ideal of A then AP is also
integrally closed. Let’s see why this must be the case. In fact, a more general fact holds:
if R is integrally closed in K = Frac(R) and is S is multiplicatively closed, then S −1 R is
integrally closed in K. We will prove this more general fact instead. Suppose that x ∈ K is
integral over S −1 R. So there exists a monic polynomial xn +(rn−1 s−1 n−1 )x
n−1
+· · ·+r0 s−1
0 = 0.
0
Write s = sn−1 sn−2 · · · s1 s0 . Then there exist r00 , · · · , rn−1 0
∈ R so that xn + (rn−1 s−1 )xn−1 +
· · · + (r10 s−1 )x + r00 s−1 = 0. Multiply by sn to get
0 0
(sx)n + rn−1 (sx)n−1 + (rn−2 s)(sx)n−2 + · · · + (r10 sn−2 )(sx) + r00 sn−1 = 0.
Therefore, sx is integral over R. Since R is integrally closed, sx ∈ R. Therefore x ∈ S −1 R,
so S −1 R is indeed integrally closed.
From this we see that AP is also integrally closed. So AP is a local Noetherian ring with
Krull dimension one, so AP is a DVR.
((3) ⇒ (1)) Suppose that P ∈ Spec(A) and AP is a DVR. Assume that C ⊆ K = Frac(A)
is the integral closure of A. So A ⊆ C ⊆ K. Our goal is to show that A = C. Let f : A → C
be the inclusion map. We will show that f is surjective. Suppose that there exists c ∈ C \ A,
and let P be the maximal ideal of A. Then AP ⊆ S −1 C where S = A \ P ⊆ C. Now AP is
integrally closed and observe that S −1 C is integral over AP .
Suppose s−1 c ∈ S −1 C. If cn + an−1 cn−1 + · · · + a0 = 0, then multiplying by s−n we see that
s−n cn +an−1 s−n cn−1 +· · ·+a0 s−n = 0. So we have (s−1 c)n +an−1 s−1 (s−1 c)n−1 +· · ·+a0 s−n = 0,
so s−1 c is indeed integral over AP .
Since AP is integrally closed and S −1 C is integral over AP , we see that AP = S −1 C 3 c
for all P . So if c ∈ C \ A then c ∈ S −1 C so c = ax−1 where x ∈ A \ P . So for each maximal
ideal P , there is ap ∈ A and xp ∈ A \ P such that c = ap x−1 p , i.e., xp c = ap ∈ A. Let
J = {r ∈ A : rc ∈ A}. Then J is an ideal of A. Notice that if J 6= A then J ⊆ P for some
55
maximal ideal P . But then xp ∈ J and xp ∈
/ P so J = A. Hence 1 ∈ J so c ∈ A. This is a
contradiction! 
36. April 2
Theorem 36.1. Let A be a Noetherian integral domain with Krull dimension 1. Then the
following are equivalent:
(1) A is integrally closed
(2) Every primary ideal is a prime power.
(3) Every local ring AP with P 6= (0) prime ideal, is a DVR.
Proof. We showed (1) ⇔ (3) last class. Let’s show (2) ⇔ (1,3).
Suppose that (2) holds. Let P be a non-zero prime ideal (so P is maximal). Consider the
local ring AP . We will show that AP is a DVR. Let J be a non-zero ideal of AP . Then there
is a correspondence (by the correspondence√theorem) J ↔ (J ∩ A)AP . What is the radical
of J an ideal of AP ? Note that we can let J = P AP . Since AP is Noetherian √ there exists
n such that (P AP )n ⊆ J. √ Letting I = J ∩ A, we see that P n ⊆ I. Then I = P since P
is maximal. Recall that if I is maximal then I is primary. Therefore there is m such that
P m = I. But then J = IAP = P m AP = (P AP )m , so AP is a DVR (by Theorem √ 34.1(6)).
Conversely, suppose that (1) holds. Let I be a primary ideal and P = I (P maximal).
Since AP is a DVR, IAP = (P AP )m = Pm AP for some m ≥ 1. We are done once we show
that P m = I. We know that we have a bijection between the proper ideals of AP and S-
saturated (i.e., if J is an ideal and xs ∈ S with s ∈ S then x ∈ J) ideals of A contained in
P , where S = A \ P . The bijection is J ↔ J ∩ A (where J is an ideal of AP and J ∩ A an
ideal of A).
Since IAP = P m AP , we have IAP ∩ A = P m AP ∩ A. We claim that IAP ∩ A = I and
P AP ∩ A = P m , hence P m = I. We will show that I and P m are S-saturated. Suppose
m

that rs ∈ I. Since I is primary, either x ∈ I or sn ∈ I, but we know that sn ∈ / I since

sn ∈/ P to begin with (thus s ∈ / P since P is a prime ideal). So I is S-saturated. Similarly,
P m is also S-saturated. Thus by the bijection we see that IAP = P m AP ⇒ I = P m , as
required. 
Now we are ready to prove Theorem 35.3.
Proof of Theorem 35.3. Let I be a proper non-zero ideal of A. Then I has a primary decom-
position I = Q1 ∩· · ·∩Qs . By our last theorem, A is a Dedekind domain, so Theorem 36.1(1),
we have that every primary ideal is a prime power. It implies that there exist P1 , . . . , Ps
prime (maximal) ideals such that Qi = Pimi . Now write I = P1m1 ∩ · · · ∩ Psms . Without
loss of generality, we can let P1 , . . . , Ps be pairwise distinct, with s taken to be minimal
with respect to this property. Then P1m1 , · · · Psms are pairwise comaximal (as to why we can
do this, see Remark 36.1). Then we have P1m1 ∩ P2m2 ∩ · · · ∩ Psms = P1m1 P2m2 · · · Psms = I.
So we have at least one factorization into prime ideals. To see uniqueness, suppose that
I = P1m1 · · · Psms = Qn1 1 · · · Qnt t , where P1 , . . . , Ps are pairwise distinct and Q1 , . . . , Qt are
pairwise distinct (and mi , ni ≥ 1).
Claim. {P1 , . . . , Ps } = {Q1 , . . . , Qt }.
/ {P1 , . . . , Ps } then P1m1 · · · Psms = Qn1 1 · · · Qnt t ⊆ Qj . So P1m1 · · · Psms ∈
To see why, if Qj ∈
Qj . If no Pi ⊆ Aj , then for each i there exists ai ∈ Pi \ Qj such that am 1 m2
1 a2 · · · as
ms
∈
56
P1m1 · · · Psms ⊆ Qj . But this is a contradiction so there exists i such that Pi ⊆ Qj . But since
Pi is maximal, indeed Pi = Qj . This proves {Q1 , . . . , Qt } ⊆ {P1 , . . . , Ps } and by symmetry
the reverse inclusion follows.
So now we can consider I = P1m1 · · · Psms = P1n1 · · · Psns with mi , nj ≥ 1. Now it suffices to
who what mi = ni for all i. Let’s look at this in the local ring APi :
IAPi = (P1m1 · · · Psms )APi = (P1n1 · · · Psns )APi
Ys
m1 ms
(P1 · · · Ps )AP = (Pi APi )mi = (Pi APi )mi
i=1
(P1n1 · · · Psns )APi = (P1 AP1 )n1 · · · (Pi APi )ni · · · (Ps APs )ns = (Pi APi )ni .
Note that, if J an ideal of A and ∅ = 6 J ∩ S 3 s then S −1 J is an ideal of S −1 A. Then
s ∈ A so 1 = ss−1 ∈ S −1 A. Thus if P and Q are distinct maximal ideals then P AQ = AQ .
Therefore we have (Pi APi )mi = (Pi APi )ni . We just need to show that mi = ni . If not, assume
mi < ni without loss of generality. Then we would have that (Pi APi )mi = (Pi APi )mi +1 =
· · · = (Pi APi )ni . But by Nakayama’s lemma, (Pi APi )ni = (0), which is a contradiction since
Pi 6= (0) and A an integral domain. So mi ≥ ni ; similarly, by symmetry we can show that
mi ≤ ni . Therefore mi = ni as desired. 
m max(m ,m )
Remark 36.1. Take s to be minimal. If some Pi = Pj then take Pimi ∩ Pj j = Pi i j
.
m1 ms
So we would get a shorter expression. Now P1 , . . . , Ps are pairwise comaximal. To show
this, we will prove the following general claim: if P and Q are comaximal then P n and Qm
are comaximal. If P and Q are comaximal, then there exist x ∈ P, y ∈ Q such that x + y = 1
hence P + Q = A. Now (x + y)m+n = 1, and
   
m+n m+n m + n m+n−1 m+n n m
(x + y) =x + x y + ··· + x y
1 m
| {z }
∈P n
   
m + n n−1 m+1 m + n m+n
+ x y + ··· + y = 1,
m+1 m+n
| {z }
∈Qm

so P n + Qm = A.
This is the official end of the course materials for commutative algebra!

37. April 6
Theorem 37.1. Let A be an integral domain and suppose that every non-zero ideal factors
into prime ideals. Then A is a Dedekind domain.
Proof. We will prove a special case of this theorem, namely when A is Noetherian.
Suppose that (0) ( Q ( P , where Q, P ∈ Spec(A). Then notice that AP has the same
factorization property as A. So in particular, the ideal QAP + (P AP )2 must factor into
prime ideals, say QAP + (P AP )2 = L1 · · · Ls where s ≥ 1 and Li ∈ Spec(AP ). Notice that
s ≤ 1 also, since otherwise each Li ∈ P AP , which would mean that L1 · · · Ls ⊆ (P AP )s ⊆
(P AP )2 . One can show that Q 6⊆ P 2 (exercise!). So this gives us a contradiction, because
QAP + (P AP )2 ) (P AP )2 . Therefore s = 1, so QAP + (P AP )2 = L, where L is prime.
57
 Also, if x ∈ P AP then x2 ∈ (P AP )2 ⊆ L. Because L is prime, it follows that x ∈ L. So
L ⊇ P AP so L = P AP since P AP is the maximal ideal of AP . Now QAP +(P AP )2 = P AP , so
if M = P AP /QAP , then (P AP )M = ((P AP )2 + QAP )/QAP = P AP /QAP = M . Therefore,
by Nakayama’s lemma, M = (0). Thus P AP = QAP , hence P = Q, a contradiction.
Therefore A has Kdim(A) = 1.
To see that A is integrally closed, let x ∈ P \ P 2 with P ∈ Spec(A) and P 6= (0). Let
J = Ax + P 2 . Then arguing as before, we see J = P . So Ax + P 2 , hence M = P/Ax and
P M = M . So by Nakayama, M = (0). Hence P = Ax, so A is integrally closed by Theorem
34.1. 

One important skills to have as a mathematician is knowing lots of examples. Also, it is

important to think of analogies to get an insight on an unknown but seemingly similar setting.
What is equally important is to ask questions about weakening hypotheses, generalizing, and
study converses.
Question. Give an example of examples of integral domain(s) such that:
(1) is Noetherian and integrally closed but does not have Krull dimension 1.
(2) is Noetherian and has Krull dimension 1 but not integrally closed.
(3) has Krull dimension 1 and is integrally closed but is not Noetherian.
Solution: For the first part, note that C and C[x, y] are the examples. For the second part,
note that the integral closure of C[t] is the integral√ closure
√ of C[t2 , t3 ]. The third part is
slightly√harder, but here is an example: R := C[x, x, x, . . . ]. This is not Noetherian since
4

(x) ( ( x) ( · · · is an strictly increasing chain of ideals. Note that R has Krull dimension
1 since R is integral over C[x] and C[x] has Krull dimension √ 1. Finally,
√ to see that R is
2m 2n
integrally closed, write K := Frac(R). If θ ∈ K, then θ = p x/q x for some n ≥ 1 with
p, q ∈ C[t]. If θ is integral over S, then there exists a monic polynomial such that
√
L √
L
θm + cm−1 ( 2 x)θm−1 + · · · + c0 ( 2 x) = 0
√ √
for some L ≥ 1. Let N = max(L, n). Then θ ∈ C( 2 x) and is integral over C[ 2 x] ∼
N N

√ = C[t],
N
and C[t] is an UFD hence integrally closed. So θ ∈ C[ 2 x] ⊆ S.
Theorem 37.2 (“Fermat’s last theorem”). Let a(t), b(t), c(t) ∈ C[t], relatively prime and
non-constant. Let n > 2. Then a(x)n + b(x)n 6= c(x)n .
Proof. First, we can assume that deg(a(x)) ≥ max(deg b(x), deg c(x)). Also, we can assume
that gcd(a, b) = gcd(a, c) = gcd(b, c) = 1. Suppose that
a(x)n + b(x)n = c(x)n . (∗)
Differentiate both sides to get
na(x)n−1 a0 (x) + nb(x)n−1 b0 (x) = nc(x)n−1 c0 (x). (†)
Then nc0 (x) × (∗) − c(x) × (†) gives us a(x)n−1 [nc0 (x)a(x) − na0 (x)c(x)] + b(x)n−1 [nb(x)c0 (x) −
nb0 (x)c(x)] = 0.
Divide both sides by n to get
a(x)n−1 [c0 (x)a(x) − a0 (x)c(x)] = −b(x)n−1 [b(x)c0 (x) − b0 (x)c(x)].
58
But since gcd(a, b) = 1, indeed a(x)n−1 | b(x)c0 (x)−c(x)b0 (x). Now deg(b(x)c0 (x)−c(x)b0 (x)) ≤
deg b(x) + deg c(x) − 1 ≤ 2 deg a(x) ≤ (n − 1) deg a(x) since n ≥ 3. So bc0 − cb0 = 0 hence
 
d b
= 0.
dx c
So b(x) = λc(x) for some λ ∈ C. But this means b(x), c(x) ∈ C but this is a contradiction
as gcd(b, c) = 1. 

Department of Pure Mathematics, University of Waterloo, 200 University Avenue West,

Waterloo, ON, Canada N2L 3G1
E-mail address: [email protected]

59