MATH 210B: Modern Algebra II

Lecturer: Professor Daniel Bump

Notes by: Andrew Lin

Winter 2023

Introduction
Professor Bump’s office hours will be held before class from 12:15 to 1:15 (on Mondays, Wednesday, and Fridays),
and Ben Church (the CA) will also hold office hours on Tuesdays and Thursdays. Information about the course can
be found at [Link] (and also linked on Canvas).
This class covers two areas of mathematics that are pretty disjoint from each other (because they both appear
on the qualifying exam), so we’ll have five weeks of one topic and five weeks of another. We’ll have some topics in
commutative algebra (specifically, affine algebraic geometry) and also some in group representation theory. For more
specifics on the first half, we’ll start by discussing integral dependence and transcendence degree, leading into basic
facts from algebraic geometry (primes in integral extensions, going-up and going-down), and then we’ll talk about
dimension theory and primary decomposition. A useful general reference (just like for 210A) is Lang’s Algebra, and
Atiyah and Macdonald’s Introduction to Commutative Algebra will be useful for this first half of the course as well.
Our first homework assignment is already posted (due next week), and we should submit it on Gradescope.

1 January 8, 2023
For today, all rings will be commutative with a unit (denoted 1).

Definition 1
Let A ⊂ B be two rings, and let α ∈ B. We say that α is integral over A if f (α) = 0 for some monic polynomial
f ∈ A[x] (that is, f (x) = x n + an−1 x n−1 + · · · + a0 for some ai ∈ A).

This is similar to the definition for fields, but it is a bit more delicate.

Proposition 2
The following are equivalent:
1. α is integral over A,

2. The ring A[α] generated by A and α is finitely generated as an A-module,

3. There exists a faithful A[α]-module which is finitely generated as an A-module.

1
Proof. For (1) implies (2), if we have some monic polynomial f with f (α) = 0 as before, then A[α] = A ⊕ Aα ⊕ · · · ⊕
Aαn−1 . And this is closed under multiplication, because αn is a linear combination of lower powers −an−1 αn−1 −· · ·−a0 ,
and we can write similar relations for higher powers of α. So we are indeed finitely generated as an A-module.
For (2) implies (3), first recall that an R-module M is faithful if r m = 0 for all m ∈ M, then r = 0. So we can
take M = A[α] (thought of as the direct sum above), and this is faithful because βM = 0 means that β1 = 0 and
thus β = 0.
Finally, for (3) implies (1), let M be a faithful A[α]-module finitely generated as an A-module. We’ll do this with
linear algebra: suppose x1 , · · · , xm generate M as an A-module, so αxi is a linear combination of the xi s as well. Write
αxi = m
P
j=1 aij xj for aij ∈ A; then the matrix T with entries aii − α on the diagonal and aij off the diagonal maps
 
x1
 . 
 .  to the zero vector. We want to deduce that the determinant of T is then zero, because then the determinant
 . 
xm
will be a monic polynomial in α. Indeed, T has an adjugate matrix T ∗ such that T ∗ T = det(T )Id (explicitly, the(i , j) 
x1
 . 
entry is the determinant of the minor obtained by deleting the i th row and jth entry), and applying both sides to  . 
 . 
xm
yields zero on the left side. Thus det(T ) annihilates all xi s, and since the module is faithful that means det(T ) = 0,
giving us the desired polynomial proving integrality of α.

Proposition 3
If A ⊂ B are rings and α, β ∈ B are integral over A, then so are α + β and αβ. Thus, the elements of B that are
integral over A form a ring.

Proof. Suppose f (α) = 0 with f (x) = x n + an−1 x n−1 + · · · + a0 and g(β) = 0 with g(x) = x m + bm−1 x m−1 + · · · + b0 .
Then A[α, β] (the ring generated by A, α, and β) can be written as
X
A[α, β] = Aαi β j
i≤n−1,j≤m−1

(not necessarily a direct sum) because this is closed under multiplication by α and β (again by reducing higher powers
with f and g). This is a faithful module because it contains the identity, and it is finitely generated as an A-module,
so by criterion 3 of Proposition 2 the sum and product are both integral over A. (Here, note that we’re thinking of
A[α, β] as an A[α + β]-module in one case and an A[αβ] module in the other.)

Definition 4
Let A ⊂ B be rings. We say that B/A is integral (in other words, an integral extension of rings) if every element
of B is integral over A.

Proposition 5
Let A ⊂ B ⊂ C be rings. If B/A is integral and C/B is integral, then C/A is integral.

Proof. Suppose γ ∈ C. By integrality of C/B, there is some monic polynomial f ∈ B[x] with f (γ) = 0, which we may
write as x n +bn−1 +· · ·+b0 . Note that it’s not true that the ring B is finitely generated over A. However, we can say that

2
γ is integral over the ring B0 = A[b0 , · · · , bn−1 ], and B0 is finitely generated as an A-module because the bi are integral
kn−1
over A. (In particular, if bi satisfy a monic polynomial of degree di , then A[b0 , b1 , · · · , bn−1 ] = ki <di Ab0k0 b1k1 · · · bn−1
P

is closed under multiplication by any bi , so it is a ring.) Now M = B0 [γ] is finitely generated as a B0 -module and
kn−1 t
thus as an A-module (we can write it as the sum ki <di ,t<n Ab0k0 b1k1 · · · bn−1
P
γ ), so by criterion 3 of Proposition 2 γ
is integral over A.

Definition 6
Let A ⊂ B be rings. We say that A is integrally closed in B if for any β ∈ B integral over A, β is actually an
element of A.

In other words, we can look at the ring of elements A′ of B integral over A (which contains A), and if A′ = A then
we are integrally closed. And this is consistent with the terminology in topology, where being “closed” is the same as
being equal to the integral closure:

Definition 7
For any rings A ⊂ B, we call the set {β ∈ B : β integral over A} the integral closure of A in B.

To ensure this terminology is good, we must check the following fact:

Proposition 8
The integral closure of A in B is integrally closed in B.

Proof. Let A′ be the integral closure of A in B, and let A′′ be the integral closure of A′ in B. Since A′ /A is integral and
A′′ /A′ is integral, then A′′ /A is integral by Proposition 5. So by definition A′′ is contained in A′ and thus A′′ = A′ .

Definition 9
Let A be an integral domain and F be its field of fractions. We say that A is integrally closed if A is integrally
closed in F .

(In other words, if we omit the larger ring, then we are looking with respect to the field of fractions.)

2 January 11, 2023

Last time, we defined what it means for an element of a ring α ∈ B ⊃ A to be integral over A (for rings A, B
commutative with unit), namely that α is the root of a monic polynomial in A[x]. We found that these integral
elements form a ring, called the integral closure of A in B, and that for any integral elements α1 , · · · , αn , A[α1 , · · · , αn ]
is a finitely generated A-module (not just a finitely generated A-algebra) because we can write it as a (not necessarily
Aαk11 · · · αknn over all ki < di (where di is the degree of the polynomial fi for integrality of αi ), which is
P
direct) sum
closed under multiplication by applying fi s.
We then said that B/A is an integral extension of rings if all elements of B are integral over A and proved that if
C/B and B/A are integral, then so is C/A. This makes it possible to define the integral closure of A in B (the set of
integral elements over A) and show that “the closure of the closure is again the closure.”

3
 Today, we’ll start by discussing how localization relates to these concepts. Recall that if A is a ring, a subset S is
multiplicatively closed if it contains 1 and is closed under multiplication. We can then define the localization S −1 A to
be the set of symbols { as : a ∈ A, s ∈ S} under the equivalence relation where a
s = b
t if u(at − bs) = 0 for some u ∈ S
(if A is an integral domain it’s enough just to check that at − bs = 0). This localization is the “smallest ring that
contains A in which the elements of S become invertible,” – in other words, there is a homomorphism i : A → S −1 A
a
sending a to 1 which is injective if A is an integral domain, and there is a universal property that any other such
homomorphism from A to a ring in which element of S become units factors through A.

Proposition 10
Suppose B is integral over A, and S ⊂ A is multiplicatively closed. Then S −1 B is integral over S −1 A. (More
a
precisely, there is a homomorphism f : S −1 A → S −1 B which maps s to as , such that S −1 B is integral over the
image of f . This is just to avoid issues with f possibly having a kernel.)

b
Proof. Any element of S −1 B can be written in the form s for some b ∈ B and s ∈ S. Since B is integral over A, we
n n−1
have b + an−1 b + · · · + a0 = 0 for some ai ∈ A. But then we see that
 n
an−1 b n−1
 
b a0
+ + · · · + n = 0,
s s s s
b
which is a monic polynomial in S −1 A[x] for which s is a root.

Proposition 11
If A is an integral domain which is integrally closed (in its field of fractions F ), and S ⊂ A is multiplicatively closed,
then S −1 A is also integrally closed.

Proof. In this case A is an integral domain, so we don’t need to worry about the inclusion of S −1 A in F . Notice that
a
S −1 A and A have the same field of fractions F – our goal is to show that if an arbitrary element b of F is integral
over S −1 A, then it is in S −1 A. The issue is basically that b is not assumed to be in S, but by integrality we get an
equation of the form   
 a n cn−1 a n−1 c0
+ + ··· + = 0,
b sn−1 b s0
where the si s are in S. Multiplying everything by (s0 s1 · · · sn−1 )n , we get a monic polynomial with coefficients in
as0 s1 ···sn−1 a
A for which b is a root. But because A is integrally closed, this must be an element of A, and thus b =
1 as0 s1 ···sn−1 −1
s0 s1 ···sn−1 · b is an element of S A.

Proposition 12
Any unique factorization domain A is integrally closed.

a
Proof. Suppose b ∈ F is integral over A for some a, b ∈ A, such that a and b are coprime. Then we have a relation
of the form  a n  a n−1
+ · · · + c0 = 0.
+ cn−1
b b
Clearing denominators, we get an equation cn−1 an−1 b +· · ·+c0 bn = −an . Thus any irreducible that divides b (meaning
it divides every term on the left side) also divides an , meaning that it divides a because A is a UFD. Since we assumed
a and b are coprime, this means b is indeed a unit.

4
 Example 13
For any field F , the polynomial ring A = F [x] is a principal ideal domain, thus a UFD and thus integrally closed.
Then B = F [x, x1 ] is a localization of A, specifically S −1 A for S = {1, x, x 2 , · · · }, so it has the same field of
fractions as A and thus cannot be integral over A (because no other elements of A’s field of fractions are integral
over it).

Example 14
Consider the ring B generated by two elements x, y given by the polynomial ring quotient F [x, y ]/(y 2 − x 2 (x + 1))
(cautionary note: we may write F [x, y ] for the ring generated by F, x, and y , even though it’s not a polynomial
ring), and take A = F [x]. Then B is integral over A, because y 2 − x 2 (x + 1) = 0, and A and B are both integral
domains. (Assume that the characteristic of F is not 2.)

However, we will show that B is not integrally closed (even though it is an integral extension of an integrally closed
ring). Indeed, the field of fractions of A is K = F (x), and the field of fractions of B is E = F (x, y ), the splitting
field of the irreducible polynomial y 2 − x 2 (x + 1). This is a quadratic extension of K, meaning that [E : K] = 2, and
y y2
t = x ∈ E is integral over A because t 2 = x2 = x + 1 but is not an element of K, thus also over B. However, t is
not integral over B. To see that, notice that C = F [t] contains B (since t 2 = x + 1 so x ∈ C, and then y = tx so
y ∈ C), and it has the same field of fractions as B. Assume without loss of generality that C = B. Then there is a
homomorphism φ : C → F sending f (t) to f (1) (the evaluation map) which sends t to 1 and thus x = t 2 − 1 to zero,
meaning that y is also sent to zero. Thus ker(φ) contains x and y , and (x, y ) generates a maximal ideal in B = C so
ker(φ) = Bx + By (because it can’t be the whole ring). But this same argument works if we use a homomorphism
ψ : C → F sending f (t) to f (−1), meaning ker(ψ) = Bx + By as well. This is a contradiction because t + 1 is in one
kernel but not the other.
This example will become useful when we analyze prime ideals and transcendence degree next time.

3 January 13, 2023

Last time, we proved some properties related to integrality: specifically, we showed that S −1 B is integral over S −1 A
if B is integral over A, and we also showed that S −1 A is integrally closed (in the field of fractions F ) if A is integrally
closed. (A simpler way to phrase the proof from last time is that if x ∈ F is integral over S −1 A, then we get an
cn−1 n−1 c0
equation x n + sn−1 x + ··· + s0 = 0 with si ∈ S for all i . But this yields a polynomial equation in A[x] for which
s0 · · · sn−1 x is a root, so that element must be in A and thus x ∈ S −1 A.)
Today, we’ll start our discussion of transcendence bases, transcendence degree, and Noether normalization. But
first we’ll prove one more result related to integrality:

Proposition 15
Let A ⊂ B be rings with B integral over A, and let ℘ be a prime ideal. Then ℘B is a proper ideal of B, and there
is a prime ideal P of B such that P ∩ A = ℘.

Proof. We’ll first prove the special case where A is a local ring and ℘ is its (unique) prime ideal. Suppose for the
sake of contradiction that ℘B is not proper. Then we can write 1 = b1 p1 + · · · + bn pn for bi ∈ B and pi ∈ ℘. Define
B0 = A[b1 , · · · , bn ]; since the bi are integral over A, B0 is finitely generated as an A-module. Since 1 ∈ ℘B0 , we also

5
have that ℘B0 = B0 . But by Nakayama’s lemma, since A is local with maximal ideal ℘ and M is a finitely generated
A-module, ℘M = M implies that M = 0. Since B0 is nonzero, this gives a contradiction. Now for the general case,
we can localize: let S = A − ℘ and consider the localizations A℘ = S −1 A and B℘ = S −1 B (here thinking of B as an
A-module).

A℘ B℘

A B

From last lecture, we know that B℘ is integral over A℘ , and we know that (℘A℘ ) · B℘ ̸= B℘ . So if P′ is a maximal
ideal of B℘ that contains ℘B℘ , then we must have P′ ∩ A℘ = ℘A℘ . (Note that because the homomorphism A℘ → B℘
is not necessarily injective, we’re really saying that the preimage of P′ under that homomorphism is ℘A℘ ). So now
taking the preimage of P′ in B (using the correspondence of ideals under localization) gives us the desired prime ideal
P, which must indeed satisfy P ∩ A = ℘. (We’re basically completing the diagram above along the bottom right
path.)

We’ll state the condition from the previous proposition more explicitly:

Definition 16
If A ⊂ B are rings and ℘ ⊂ A and P ⊂ B are prime ideals, we say that P lies above ℘ if ℘ = P ∩ A. A similar
definition can be made if θ : A → B is a homomorphism with ℘ = θ−1 (P).

This terminology is motivated by affine algebraic geometry, which is a topic we’ll be discussing in this class:

Definition 17
Let F be a field. An affine algebraic variety over F is the set of solutions in F n of some set of polynomial
equations Σ = {f1 , · · · , fN }. Specifically, the variety V (Σ) is the set {x ∈ F n : fi (x) = 0 for all i }. This variety is
also denoted V (I), where I is the ideal generated by the elements fi ∈ F [X1 , · · · , Xn ]. We say that V = V (Σ) is
irreducible if it is not the union of two proper subvarieties.

Example 18
The union of the x- and y -axis V (X1 X2 ) is not irreducible – in fact, irreducibility is equivalent to the radical r (I) of
the corresponding ideal I = (X1 X2 ) being prime. (The radical of an ideal I is the set {f ∈ R : f n ∈ I for some n}
– the binomial theorem proves that this is indeed an ideal.)

We can assume that I is equal to its radical, because V (Σ) = V (I) = V (r (I)) for some ideal I. (Indeed, if f n ∈ I,
then f n vanishes on V (I), which is the same as f vanishing on V (I) because we’re working in a field. In such a situation
where we assume irreducibility (so I = r (I) is prime), R/I will be an integral domain. (If we don’t assume irreducibility
we just get a reduced ring.) In fact, R/I is the space of all polynomial functions on V because two functions f , g are
the same function on V if their difference is zero on all of I.

Example 19
We can now see an example where Proposition 15 doesn’t hold if we don’t have integrality of the extension.
Consider the ideal I = ⟨XY − 1⟩ ⊂ R = F [X, Y ] – then the variety V (I) is a hyperbola H, and it is irreducible.

6
 The space of polynomial functions on this hyperbola H is then B = R/I = F [x, y ], where x and y are X and Y mod
I, respectively, with xy = 1. In other words, B is basically F x, x1 , the affine algebra or coordinate ring O(H). If we
 

then consider the polynomial map φ from H to the x-axis sending (X1 , X2 ) to X1 , then we can compose a polynomial
on the x-axis (thought of as just F [x]) with φ. Specifically, we get a ring homomorphism φ∗ from A = F [x] (functions
on the x-axis) to B = F [x, 1/x] (functions on H) sending f to f ◦ φ, which is the inclusion map.
But now the algebraic geometry idea is that prime ideals roughly correspond to points – if F is an algebraically
closed field, there is a bijection between V (any affine variety) and maximal ideals of the ring of polynomial functions
on V . (This is the nullstellensatz.) And the point is that the mapping is not surjective, since no point on the
hyperbola projects to the origin, and thus there is no prime ideal of B lying above a prime ideal in A. And the reason
the proposition from earlier doesn’t apply is that F [x, 1/x] is not integral over F [x] (as we proved last time).

Example 20
On the other hand, if we take the example curve y 2 = x 2 (x +1) from last time and consider A = F [x], B = F [x, y ],
then we proved last time that B is integral over A. So if we believe the philosophy about prime ideals corresponding
to points, then this corresponds to the fact that projecting the curve y 2 = x 2 (x + 1) on the x-axis is surjective
(remembering that we are working over C, not R).

4 January 18, 2023

We’ll briefly talk about transcendence degree today just to discuss the main facts, and then we’ll move on to Noether
normalization.

Definition 21
Let Ω be a field. A set of elements {y1 , · · · , yn } ∈ Ω is algebraically independent if there is no nonzero
polynomial relation φ(y1 , · · · , yn ) = 0 between the yi s with coefficients in Ω (that is, Φ ∈ F [X1 , · · · , Xn ] is not
the zero polynomial).

Definition 22
Suppose F ⊂ K ⊂ Ω are fields (we’re going to put everything inside a big field Ω so that rings are all integral
domains). A set of elements x1 , · · · , xn ∈ K ⊂ Ω span K over F if K is algebraic over F (x1 , · · · , xm ) (here
parentheses means the field generated by F , x1 , · · · , xm , rather than just the algebra).

The word “span” here is technically incorrect (it’s not like in the sense of linear algebra), but it’s meant to provide
some intuition similar to that of linear algebra. And this next result shows a further similarity (comparing the size of
linear independent sets to spanning sets):

Theorem 23
Suppose y1 , · · · , yn ∈ K are algebraically independent over F , and x1 , · · · , xm ∈ K span K (meaning that
K/F (x1 , · · · , xm ) is algebraic), then n ≤ m.

In particular, this leads to the notion of a “trascendence basis” with well-defined size. And we can say something
similar when our sets {xi } and {yi } are infinite, just using cardinality instead of size – however, this requires transfinite
induction.

7
Proof. Since y1 is algebraic over F (x1 , · · · , xm ) (by definition of the spanning set), there is a relation φ(y1 , x1 , x2 , · · · , xm ) =
0 for some nonzero polynomial φ in the polynomial ring over m + 1 variables. This relation involves one of the xi s
(because y1 isn’t algebraic over F by algebraic independence, so our polynomial can’t just be in y1 ). Without loss of
generality say that φ involves x1 . Then this polynomial relation tells us that x1 is algebraic over F (y1 , x2 , · · · , xm ) – in
particular, K is algebraic over F (x1 , y1 , x2 , · · · , xm ), which is algebraic over F (y1 , x2 , · · · , xm ), and algebraic extensions
stack. Thus y1 , x2 , · · · , xm also span K.
We can then repeat this process: since y2 is algebraic over y1 , x2 , · · · , xm , there is some nonzero ψ ∈ F [Y1 , Y2 , X2 , · · · , Xm ]
such that ψ(y1 , y2 , x2 , · · · , xm ) = 0. Again ψ must involve one of the xi s because {y1 , y2 } are algebraically independent.
Letting it be x2 without loss of generality, we see that K is algebraic over F (y1 , y2 , x2 , · · · , xm ), which is algebraic over
F (y1 , y2 , x3 , · · · , xm ) (because x2 is algebraic over the other variables). This means {y1 , y2 , x3 , · · · , xm } also spans
K. This same process shows that we can replace n of the xi s by yi s, so n ≤ m (if we had n > m then we’d have
{y1 , y2 , · · · , ym } spanning K and then ym+1 is algebraic over F (y1 , · · · , ym ), contradicting linear independence).

Remark 24. There’s a general notion of a matroid (where independence and span are concepts) which generalizes
this kind of proof.

Definition 25
Let K ⊃ F be fields. A transcendence basis of K/F is a (possibly infinite) subset {y1 , y2 , · · · } such that the yi s
are algebraically independent and span K (in the sense mentioned before that K/F (y1 , y2 , · · · ) is algebraic).

Transcendence bases always exist by a Zorn’s lemma argument – if we let Ω be the set of all algebraically independent
subsets of K (over F ), partially ordered by inclusion, then Ω is nonempty (because it contains {∅}) and every chain
has a maximal element (the union of the elements), there is some maximal element in the poset Ω. The point is
that for any such maximal subset B, K must be algebraic over F (B) because otherwise we could add a non-algebraic
element to B. Thus B is a transcendence basis, and in fact by Theorem 23, any two transcendence bases have the
same cardinality, and we will call that the transcendence degree and denote it [K : F ]TD . One important fact (which
was an exercise for us) is that
[E : F ]TD = [E : K]TD + [K : F ]TD ,

so transcendence degree is additive.

Theorem 26 (Noether normalization lemma)

Suppose A = F [x1 , · · · , xn ] is an integral domain with field of fractions K = F (x1 , · · · , xn ) (remember that these
are not necessarily polynomial rings), and suppose the transcendence degree of K over F is r . Then there is some
transcendence basis {y1 , · · · , yr } such that yi ∈ A and A is integral over F [y1 , · · · , yr ]. (Note that F [y1 , · · · , yr ]
is actually a polynomial ring because the yi s are algebraically independent.)

There are two proofs of this in our textbook, but we’ll do the first one which doesn’t require us to treat the
finite-field case separately.

Proof. Induct on the number of generators n. For the base case, if x1 , · · · , xn are already algebraically independent
over F , then n = r is the transcendence degree and we can just take xi = yi for all i . Otherwise, suppose there is
an algebraic relation φ ∈ F [X1 , · · · , Xn ] such that φ(x1 , · · · , xn ) = 0. We will now introduce variables z1 = x1 , z2 =
x2 − x1N , z3 = x3 − x12N , and so on (with N to be determined). Writing out the polynomial as φ = j∈Nn aj x (j) , where
P

8
x (j) = x1j1 x2j2 · · · and aj ∈ F , we can now rewrite this polynomial in terms of Zs. We know that
X
0 = φ(x1 , · · · , xn ) = φ(z1 , z2 + z1N , z3 + z12N , · · · ) = aj z1j1 (z2 + z1N )j2 (z3 + z12N )j3 · · · ,
j

and for sufficiently large N the degree of any particular summand j is j1 +Nj2 +2Nj3 +· · · , coming from aj z1j1 +Nj2 +2Nj3 +··· .
Running through all of the summands, we can now cause a single term to dominate in degree. Specifically, we can
say that (j) > (j ′ ) if there is some k such that jk > jk ′ and jk+1 = jk+1
′ ′
, jk+2 = jk+2 , · · · , and this is a total order
(essentially lexicographic). Picking the j which is maximal under this ordering (across all j with nonzero coefficient
aj ), we find that φ, as a polynomial in the zi s, has a leading term aj z1j1 +Nj2 +··· . Since aj ∈ F , this means z1 = x1 is
integral over F [z2 , · · · , zn ], and remember that F [z1 , z2 , · · · , zn ] is the same ring as F [x1 , · · · , xn ]. But by induction,
there is a transcendence basis {y1 , · · · , yr } ⊂ F [z2 , · · · , zn ] (since we only have (n − 1) generators this time) such
that F [z2 , · · · , zn ] is integral over F [y1 , · · · , yr ]. (Importantly, the transcendence degree doesn’t change because z1
is algebraic over F [z2 , · · · , zn ].) Thus A = F [x1 , · · · , xn ] = F [z1 , z2 , · · · , zn ] is integral over F [z2 , · · · , zn ], which is
integral over F [y1 , · · · , yr ], and this is what we wanted to prove.

5 January 20, 2023

Today’s topic is valuation rings:

Definition 27
Let F be a field. A valuation ring is a subring R of F such that for any x ∈ F , either x ∈ R or x −1 ∈ R.

Example 28
A discrete valuation ring is a principal ideal domain with a unique maximal ideal. For example, Z(p) for a prime
p is a discrete valuation ring in Q, because any rational number as a reduced fraction has either numerator or
denominator not divisible by p.

Example 29
O, the ring of germs of holomorphic functions at 0 ∈ C (that is, the set of power series ∞ i
P
i=0 ai x with a positive
P∞
radius of convergence) is a discrete valuation ring in the field of fractions { i=−N ai x i converging near zero}).

Not all valuation rings are discrete valuation rings, but they are similar “in spirit,” as we’ll see in these next few
results.

Proposition 30
Any valuation ring R is a local ring.

Proof. The set of non-units is easily seen to be closed under multiplication. For closure under addition, suppose
x, y ∈ ℘ = R − R× are both non-units – we can just consider the case where x and y are nonzero. Then by definition
x y x x x+y
either y or x is in R; without loss of generality say that y ∈ R. Then 1 + y = y is also in R. But then if x + y is a
1
unit, then multiplying by its inverse shows that y is in R, contradicting that y is a unit.
In particular, the set of non-units yields a maximal ideal (adding any other element of R would give us the whole
ring) and it must be the unique maximal ideal. Thus R is a local ring.

9
 Proposition 31
Any valuation ring R is integrally closed. In other words, if R is a valuation ring of its field of fractions F , then R
is integrally closed in F .

Proof. Suppose x ∈ F is integral over R. Then there is some polynomial relation x n + an−1 x n−1 + · · · + a0 = 0, with
ai ∈ R. Suppose for the sake of contradiction that x ̸∈ R. By definition of a valuation ring, we must have x −1 ∈ R,
and in fact x −1 must be in ℘. But then we can rewrite the equation above as 1 = − an−1 x −1 + · · · + a0 x −n , where

the right-hand side is in ℘. Since ℘ is not the whole ring, this is a contradiction.

Theorem 32 (Extension theorem for valuation rings)

Let F be a field and let A ⊆ F be a ring. Suppose Ω is an algebraically closed field with φ : A → Ω a ring
homomorphism. Then φ can be extended to a valuation ring of F containing A.

Proof. Let Σ be the set of pairs (R, Φ), where R is a ring A ⊂ R ⊂ F and Φ : R → Ω is a ring homomorphism
extending φ. By Zorn’s lemma, Σ has some maximal element (which we’ll denote (R, Φ)), meaning that we cannot
extend Φ : R → Ω to a larger ring.
The first step is to prove that R is local. Because Ω is a field, the image of Φ in it must be an integral domain
and thus ker(Φ) is a prime ideal which we will call ℘ (this will be the ℘ we’re looking for). Then Φ extends to R℘ , via
a

Φ(a)
Φ s = Φ(s) – this definition makes sense because s is not in ℘, so Φ(0) is nonzero and can be inverted in Ω, and it
a b
is well-defined because s = t implies that at = bs, implying that Φ(a)Φ(t) = Φ(b)Φ(s); multiplying by the inverses
Φ(a) Φ(b)
of Φ(t) and Φ(s) shows that Φ(s) = Φ(t) . But by maximality this means R℘ = R (since we can’t extend beyond R),
and this means R is a local ring (because it is the localization at a prime ideal).
Next, we want to show that R is indeed a valuation ring, meaning that for all nonzero x ∈ F we want either x ∈ R
or x −1 ∈ R. The idea is to show that we can extend Φ to either R[x] or R[x −1 ], since again by maximality that
would mean x ∈ R or x −1 ∈ R, respectively. Without loss of generality, we can assume that x is algebraic over R
(otherwise R[x] is a polynomial ring, so Φ can be extended to R[x] by the universal property of the polynomial ring
sending Φ(a0 + a1 x + · · · + an x n ) = Φ(a0 )).

Lemma 33
We cannot have both ℘R[x] = R[x] and ℘R[x −1 ] = R[x −1 ].

Proof of lemma. Suppose that both of those equalities held. Then we could write

1 = a0 + a1 x + · · · + an x n = b0 + b1 x −1 + · · · + bn x −m , ai , bi ∈ ℘.

Choose such polynomials to minimize m and n; without loss of generality we may assume n ≥ m. Notice that 1 − b0
must be a unit; otherwise it would be in the ideal of all non-units ℘ (here is where we use that R is local) and thus 1
would be in ℘. Thus we have

(1 − b0 )x m = b1 x m−1 + · · · + bm =⇒ x m = b1′ x m−1 + · · · + bm

′

where bi′ = (1 − b0 )−1 bi ∈ ℘. But then multiplying this by an x n−m and subtracting it from the first relation allows us
to reduce n by cancelling out the leading coefficient, which is a contradiction by minimality.

10
 So now without loss of generality we may assume that ℘R[x] is a proper ideal of R[x]. Let P be a maximal ideal
of R[x] containing ℘R[x]; then P ∩ R = ℘ (because by construction the left-hand side contains ℘, it is an ideal of
R, and ℘ is maximal in R). Then we have R/℘ contained in R[x]/P – these are both fields because the ideals are
maximal in their respective rings, and in fact R[x]/P is a finite extension of R/℘ because x is assumed to be algebraic.
From Φ we then get an induced injective homomorphism Φ : R/℘ → Ω (since we define ℘ = ker(Φ)). We can thus
think of R/℘ as being a subfield of the algebraically closed Ω, and thus we can extend to the finite extension and get
′
a map Φ : R[x]/℘ → Ω (because the finite extension comes from adjoining algebraic elements, which will also lie in
Ω). This then pulls back to a map R[x] → Ω, and that can only happen (by maximality) if x ∈ R.

Fact 34
On a valuation ring R ⊂ F , we can introduce an ordered group Γ = F × /R× , where “ordered” means that there
is a subset Γ+ closed under multiplication (that is, a submonoid). Setting Γ− = {z : z −1 ∈ Γ+ }, we can then
have Γ+ ∩ Γ− = {1}, and we can think of this as the “positive reals” and the “negative reals.” The ordering is then
that x < y if y −1 x ∈ Γ+ ; in a discrete valuation ring Γ will be isomorphic to Z. We then have R/R× = Γ+ , and
a valuation is then a mapping sending F × to this group – specifically, we have ν : F 7→ Γ ∪ {−∞} so that 0 can
have valuation −∞. But we’ll probably talk more about this later.

We’ll see how this relates to the nullstellensatz next time.

6 January 23, 2023

We’ll discuss the nullstellensatz today – this works best over an algebraically closed field, and that’s what we’ll do.
There are actually two versions that we’ll go over – recall that the variety of an ideal a, denoted V (a), is the locus of
zeros of a.

Theorem 35 (Weak nullstellensatz)

Let F be an algebraically closed field, and let R = F [x1 , · · · , xn ] be a polynomial ring. If a ⊂ R is a proper ideal,
then V (a) (the set of (a1 , · · · , an ) ∈ F n with f (a1 , · · · , an ) = 0 for all f ∈ a) is nonempty.

In other words, every proper ideal has a zero. For the other statement, recall that for any set X ⊂ F n (usually a
variety or algebraic set), we can define

I(X) = {f ∈ F [X1 , · · · , Xn ] : f = 0 on X}.

By definition, we know that I(V (a)) ⊃ a, but we could have a a strict subset of I(V (a)). On the other hand, we can
define the radical
r (a) = {f ∈ R : f n ∈ a for some n}.

To see that the radical is closed under addition, notice that that by the binomial theorem, we know that (f + g)N =
P N
i n−i
i f g , and then if f , g are in the radical then for sufficiently large N f i g n−i is always zero, so the sum is also
zero.

Theorem 36 (Strong nullstellensatz)

Again let a ⊂ F [x1 , · · · , xn ] be any ideal. Then I(V (a)) = r (a).

11
 These results also apply to quotients of polynomial rings, so the nullstellensatz also applies to any finitely generated
ring over an algebraically closed field – this is the fundamental connection of commutative algebra with affine algebraic
geometry.
With our definitions, it is easy to see that r (a) ⊂ I(V (a)). Indeed, if f ∈ r (a), then f n ∈ a and thus f n = 0 on
V (a). Then f vanishes wherever f n vanishes so f ∈ I(V (a). But the other inclusion is harder.

Proof of equivalence of weak and strong nullstellensatz. To show that the strong nullstellensatz implies the weak null-
stellensatz, first notice that we can assume without loss of generality that a is maximal, since enlarging the set of
polynomials only makes the set smaller. Then by maximality, r (a) = a. But if V (a) were empty, then we would have
I(V (a)) = R, which is a contradiction with the strong nullstellensatz because the two sides are not equal.
It turns out there is an implication the other way as well, which uses the “Rabinowitsch trick” (the proof was
published by Rainich, who used a pseudonym to publish it). Suppose the weak nullstellensatz is true, and let f be some
polynomial in I(V (a)). Introduce another indeterminate Y , so that (we’ll use capital letters since we have a polynomial
ring)
R = F [X1 , · · · , Xn ] ⊂ F [Y, X1 , · · · , Xn ].

Now consider the ideal A in R generated by a and the polynomial 1−Y f (X1 , · · · , Xn ). We claim this ideal has no zeros
in F n+1 – that is, we claim that V (A) is empty. Indeed, if (b, a1 , · · · , an ) were a zero, then either (1) (a1 , · · · , an ) is in
V (a) (which would mean f (a1 , · · · , an ) = 0), meaning that 1 − Y f (X1 , · · · , Xn ) would have value 1 at (b, a1 , · · · , an )
and thus not everything in A vanishes at this point, or (2) (a1 , · · · , an ) is not in V (a), meaning there is some φ ∈ a
with φ(a1 , · · · , an ) ̸= 0, meaning again that (b, a1 , · · · , an ) cannot lie in V (A). So V (A) is empty, and that means
that by the weak nullstellensatz, A must be the entire ring R[Y ] = F [X1 , · · · , Xn , Y ].
P
In particular, this means that 1 ∈ A, so we can write 1 = bi ai + b0 (1 − Y f (X1 , · · · , Xn )), where bi ∈ R[Y ] and
ai ∈ a. Working in the field of fractions – that is, the ring of rational functions in these variables – we can substitute
1
Y = f (X1 ,··· ,Xn ) and find that
 
X 1
1= bi , X1 , · · · , Xn ai (X1 , · · · , Xn ) ,
f (X1 , · · · , Xn )

still with ai ∈ A and bi ∈ R[Y ]. But then if we multiply by a sufficiently large power of f to clear denominators in the
bi terms, we find that f N is equal to a sum of terms which are each some polynomial times some element of a. Thus
f N ∈ a and thus f (which we originally assumed was in I(V (a)) is also in r (a).

So the two statements are equivalent, and it remains for us to prove the weak nullstellensatz. We’ll first prove the
“algebraic nullstellensatz,” also called Zariski’s lemma:

Proposition 37 (Algebraic nullstellensatz)

Let F be a field, and let K be a field containing F that is finitely generated as an F -algebra (that is, K =
F [x1 , · · · , xn ] not as a polynomial ring but as an F -algebra). Then K is a finite (in particular algebraic) extension
of F .

Proof. Here is where we use the extension theorem for valuation rings, as well as Noether normalization – the latter
can be avoided with an alternative proof, though. By Noether normalization, there is a transcendence basis y1 , · · · , yr
of K over F such that F [x1 , · · · , xn ] is integral over F [y1 , · · · , yr ] (the latter of which is isomorphic to a polynomial
ring in r variables). We can thus define the homomorphism φ : F [y1 , · · · , yn ] → F which plugs in 0 for each yi . This
homomorphism then extends to a valuation ring R of K; furthermore, R ⊃ F [y1 , · · · , yr ] and R is integrally closed
because it is a valuation ring. Thus R contains the integral closure of F [y1 , · · · , yr ], which is K. Thus using this

12
homomorphism we can map K to F ; this is a map between fields so it is an embedding. In particular, this means K is
algebraic over F , and it’s also finitely algebraically generated by definition, so it is a finite extension. (And this means
r = 0.)

To get from here to the weak nullstellensatz is easy: if m is any maximal ideal of F [X1 , · · · , Xn ], then F [X1 , · · · , Xn ]/m
satisfies the conditions of the algebraic nullstellensatz, so it is a finite extension of F . We’ll talk more in detail about
this next time!

7 January 25, 2023

We’ll continue our discussion of algebraic geometry concepts today – the idea is that affine algebraic geometry and
commutative algebra are almost the same, in that they talk about the same concepts. (There’s three sets of lecture
notes about the Zariski topology, varieties, and the going-up and going-down lemmas – the last of these will be covered
soon in class.) This next definition should look familiar from our homework:

Definition 38
Let F be a field, which we assume to be algebraically closed (so that the nullstellensatz holds). An affine algebraic
set X ⊂ An (where An is the affine n-space Fn as a set) is the set of all solutions of some set Σ of polynomial
equations. We will write this as X = V (Σ) = {x ∈ An : f (x) = 0 for all f ∈ Σ}.

If we let a be the ideal generated by Σ, then clearly V (Σ) = V (a), and in fact this is even equal to V (r (a)) (as
we’ve discussed previously), where the radical of a is the set r (a) = {f ∈ F [x1 , · · · , xn ] : f N ∈ a for some N}. So
we may assume Σ is an ideal, even a radical ideal (meaning that Σ is its own radical), since the radical of r (a) is
again r (a). And as we showed, the set of subsets of An of the form V (a) for some a forms a set of closed sets of a
topology (it is closed under finite unions and arbitrary intersections), which we call the Zariski topology. Indeed, we
have V (a) ∪ V (b) = V (ab) (since the product of two ideals is generated by products of polynomials in a and b), and
P
T
V i ai = i V (ai ).
This topology on An is a bit strange in that it’s not Hausdorff – in fact, any two nonempty open sets have nonempty
intersection (so we can’t separate any open sets at all). So some of our intuition with ordinary topological spaces may
break down, but not all of it. The idea is that closed subsets of An correspond to radical ideals (with correspondence
order-reversing), and the nullstellensatz says that two radical ideals with the same closed set are actually the same
ideal. Furthermore, because R = F [x1 , · · · , xn ] is noetherian, that carries over to An as well via the correspondence –
any descending chain of closed sets X1 ⊋ X2 ⊋ · · · must terminate. Such a topological space is then called noetherian
as well:

Definition 39
A topological space is noetherian if any descending chain of closed subsets terminates, and it is called irreducible
if it is not the union of two proper closed subsets.

In particular, An is both noetherian and irreducible. On the other hand, the union of the coordinate axes X =
{(x, 0) : x ∈ A} ∪ {0, y : y ∈ A} in A2 is not irreducible – it’s a closed set, but it’s the union of its two parts, each of
which is closed.

13
 Proposition 40
Any noetherian space is the finite union of irreducible closed subspaces.

Proof. Let x ∈ X be some arbitrary element, and let Y be a minimal closed subset of X containing x. (This exists
by the nooetherian property and a Zorn’s lemma argument on the closed subsets containing x, or alternatively we can
just take the intersection of all closed subsets containing x.) Then Y must be irreducible, since we would otherwise
be able to write Y = Y1 ∪ Y2 , and then either x is in Y1 or Y2 ; whichever it is, we contradict minimality of Y . So x is
in some irreducible closed subset; we can take the union of the corresponding subsets over all x.
But we can in fact say that this union can be taken to be finite – this now actually uses the Noetherian property.
Indeed, if we had an infinite union and none of them could be thrown away, we could find an infinite descending chain.
(We’ll talk more about this next time.)

This decomposition is unique, since there is a unique smallest set of closed irreducible subsets (which we call the
components of X). This means that for any closed set X = V (a) ⊂ An (where we take a = r (a)), we can consider
the topology on X induced by the Zariski topology on An (which we also call the Zariski topology). To see whether
X is irreducible, notice that the ring of polynomials on X is R/a (here is where we’ve used the nullstellensatz – f1 , f2
have the same restriction to X if and only if f1 − f2 = 0 on X, which occurs if and only if f1 − f2 ∈ r (a) = a.)
We will call O(X) = R/a the affine algebra or coordinate ring. Notice that X is irreducible if and only if O(X)
is an integral domain – if we have f1 , f2 ∈ O(X) with f1 f2 = 0, then Xi = {x : fi (x) = 0} for i ∈ {1, 2} are closed
sets with X = X1 ∪ X2 , so irreducibility breaks unless either f1 or f2 is zero. So in fact irreducibility corresponds to
primeness of the ideal a, and we will call irreducible affine algebraic sets affine varieties.
However, we have to dispense of the embedding into affine space if we want to understand the connection with com-
mutative algebra more clearly – it shouldn’t depend on the ambient space. The idea is that affine algebraic sets form a
category, with morphisms given by polynomial maps: if we have X ⊆ An and Y ⊆ Am , then a map f : X → Y is a mor-
phism if there exist polynomials f1 , · · · , fn ∈ F [X1 , · · · , Xn ] with f (x1 , · · · , xn ) = (f1 (x1 , · · · , xm ), · · · , fn (x1 , · · · , xm )).
Such a morphism then induces a ring homomorphism f ∗ : O(Y ) → O(X) given by post-composition with f , and
conversely any ring (F -algebra) homomorphism O(Y ) → O(X) gives rise to a morphism. To see that, suppose
Y = F [X1 , · · · , Xm ]/b for some radical ideal b. Take generators y1 , · · · , ym which are cosets modulo b and look at
their images under the homomorphism, and then we can interpret φ(yi ) as some polynomial fi (x1 , · · · , xn ) (or specifi-
cally the corresponding cosets). The point is that if these affine algebraic sets form a category, then isomorphic objects
may correspond to different affine embeddings, which we do want to identify together.

Definition 41
A morphism f : X → Y between varieties is dominant if f (X) is dense in Y (though not necessarily surjective).

Proposition 42
A morphism f is dominant if and only if the corresponding map O(Y ) → O(X) is injective. In particular, for
any dominant morphism we can identify O(Y ) as a subring of O(X); if O(X) is integral over O(Y ), then f is
surjective.

(This fact is proved in next lecture.)

14
 Example 43
Consider X = V (X1 X2 − 1), which is the hyperbola in A2 , and Y = A1 the x-axis. Then O(X) can be thought of
as F [X1 , X2 ]/(X1 X2 − 1) = F [X, X1 ] Then the map f : X → Y which is the projection onto the x-axis is dominant
(because F [X] injects into F [X, X1 ], and it has O(X) not integral over O(Y ). And indeed f is not surjective (this
is a similar example as we’ve seen in a previous class).

This whole argument can also be formulated in terms of prime ideals, which would motivate the next topic of
“going-up” and “going-down.” But we’ll see that in the future!

8 January 27, 2023

Last time, we started discussing affine varieties and some relevant properties of their constructions. In the notes on
the class website, some additional information about the irreducible decomposition of a noetherian topological space
is posted – recall that a topological space is noetherian if every descending sequence X1 ⊃ X2 ⊃ · · · eventually
terminates. Specifically, we saw that An (F ) is noetherian, and more generally any algebraic set (affine variety) X is
also closed. (Indeed, O(X) = R/a is noetherian by the Hilbert basis theorem, so the ascending chain condition for
ideals yields the corresponding descending chain condition on closed sets.) The important takeaway is that if X = V (a)
for some radical ideal a, then X is irreducible if and only if a is prime.
Since we’re going to turn to dimension theory soon, we’ll go back to the proof that any noetherian space is the
finite union of irreducible closed subspaces, slightly restating the statement:

Proposition 44
If X is a noetherian topological space, then we may write X = X1 ∪ · · · ∪ Xn with Xi irreducible closed subsets.
Furthermore, this decomposition is unique if we discard redundant factors, meaning that Xi ̸⊆ Xj for any i , j.

Proof. From our discussion last lecture, a Zorn’s lemma argument tells us that (using lower bounds instead of upper
bounds, and using the descending chain condition) a nonempty set of closed subsets always has a minimal element.
Thus, if X does not have such a finite decomposition, let Σ be the set of all closed subsets Z ⊆ X with Z not
having a finite irreducible decomposition. By assumption Σ is nonempty because it contains X, so it has a minimal
element Z. Z is also noetherian (since it’s a closed subset of X), and Z is not irreducible (or else it would work
as its own decomposition into irreducible factors). Thus we have Z = Z1 ∪ Z2 , and by minimality Z1 and Z2 do
have finite irreducible decompositions. But putting those together, that means Z does have a finite decomposition, a
contradiction.
The uniqueness is written up in notes and is relatively routine (just comparing components in two different repre-
sentations).

We’re now ready to turn to dimension, for which there are two different definitions that will turn out to be
equivalent. But to prove their equivalence, we’ll need some of the results from earlier in this class, as well as the
going-up and going-down lemma.

15
 Definition 45
Let X be an affine variety (recall that for us this means it is an irreducible affine algebraic set). Then O(X) is an
integral domain, meaning that it has some field of fractions K. The dimension of X is then the transcendence
degree of K/F .

The idea is that for a dimension n variety, there should be n algebraically independent functions. And the way to
interpret this dimension is to say that O(X) is rational over F [y1 , · · · , yd ], so we sort of have “rational functoins in d
dimensions.”

Definition 46
Again let X be an affine variety. The combinatorial dimension is the maximal d such that we have a chain
X0 ⊊ X1 ⊊ X2 ⊊ · · · Xd = X of nonempty closed irreducible subspaces.

The idea is that the smallest possible chain starts with a point (which is 0-dimensional), then something like a line
(which is 1-dimensional), and so on up until the entire variety X. The point is that if X and Y are irreducible subsets
with X ⊊ Y , then we want dim(Y ) > dim(X) (we don’t want a situation where X is just the x-axis and Y is the union
of the x- and y -axes). And we can make a corresponding definition on the commutative algebra side:

Definition 47
Let R be a ring. The Krull dimension of R, denoted dim(R), is the maximal d such that there is a chain of prime
ideals ℘0 ⊊ ℘1 ⊊ · · · ⊊ ℘d . (If R is an integral domain, then ℘0 can be taken to be the zero ideal.)

Specifically, notice that the Krull dimension of O(X) is the combinatorial dimension of X, and the deeper result
is that this is also the dimension of X in the transcendence degree definition. And remember that ℘d corresponds to
X0 , ℘d−1 corresponds to X1 , and so on.
Recall from last time that a morphism f : X → Y of varieties is dominant if f (X) is dense in Y , and we stated that
f is dominant if and only if the corresponding F -algebra homomorphism f ∗ : O(Y ) → O(X) (given by precomposition
with f ) is injective. We’ll now do the proof of this:

Proof. For any nonzero φ ∈ O(Y ), consider the set

Yφ = {y ∈ Y : φ(y ) ̸= 0}.

This is the complement of the closed set V ((φ)), so it is open. In fact, Yφ form a basis of the topology, since the
complement of V (a) (for any a) is the union of Yφ s for any φ ∈ a. (Indeed, if x ̸∈ V (a), then φ(x) ̸= 0 for some φ ∈ a,
S
which is the same as saying that x ∈ Yφ and thus x ∈ φ∈a Yφ .) These Yφ s are sometimes called the principal open
sets.
So now if f ∗ is not injective, then this is equivalent to f ∗ (φ) = 0 for some nonzero φ ∈ O(Y ), which happens if
and only if f (X) ∩ Yφ = ∅ (in other words, f ∗ (φ) = φ ◦ f is the zero map). But that’s the same as saying that f (X)
is not dense (since density requires intersecting every basic open set).

We’ve already proved an easy version of the going-up theorem in this class (though a more complicated version
can be seen in the lecture notes, which we should read). The result was as follows: let B be integral over A, and let
℘ be a prime ideal of A. Then ℘B ̸= B, and B contains a prime ideal above A. (Recall that we proved ℘B ̸= B by
a Nakayama lemma argument when B is a finite A-module, and even when it isn’t we can still make the argument

16
work.) Now as an application, recall that we stated last time that if f : X → Y is a dominant morphism, then the
map O(Y ) ,→ O(X) is injective, and if O(X) is integral over O(Y ) then f is surjective. We’ll prove that as well:

Proof. Let y ∈ Y and consider my = {φ ∈ O(Y ) : φ(y ) = 0}. This is a maximal ideal, and thus there is some ideal P
of O(X) lying above my which turns out to be maximal. So by the nullstellensatz there is some x ∈ X with {x} = V (P)
(the key point is that points are in bijection with maximal ideals), and we can check that f (x) = y .

9 January 30, 2023

We’ll continue our discussion of dimension with the going-up and going-down theorems today. Recall that a (commu-
tative) ring A’s Krull dimension is defined by looking at the maximal length of a chain of prime ideals ℘0 ⊆ · · · ⊆ ℘d ,
and that if A is an algebra over some ground field the transcendence degree of the field of fractions of A agrees with
this. (Note that there are rings that are even finitely generated but still have infinite Krull dimension. But the rings in
algebraic geometry and algebraic number theory are typically finite dimension.)

Definition 48
A Dedekind domain is an integrally closed Noetherian domain of dimension 1.

In other words, every nonzero prime ideal is maximal (since the longest possible chain would be (0) ⊆ ℘, here using
that (0) is a prime ideal because we have a domain). A principal ideal domain is always a Dedekind domain, since any
nonzero prime ideal is (f ) for some irreducible element f ∈ A, and these are already maximal. But in general there are
lots of rings that are Dedekind domains but not PIDs that we still care about.

Theorem 49
Let L/K be a finite separable extension, and suppose A ⊆ K is a Dedekind domain. Then the integral closure B
of A in L is also a Dedekind domain.

This is left to us as an exercise, and we should note that this result does not hold for principal ideal domains:

Example 50
√ √
If L = Q( −5) and K = Q, and we take A = Z (which is a PID), then the integral closure of A in L is Z[ −5].
√ √
This is not a principal ideal domain because unique factorization fails (6 = 2 · 3 = (1 − −5)(1 + −5)), meaning
it’s not a UFD, but on the other hand it is a Dedekind domain.

The point is that Dedekind domains are a bit pathological, but localizing them gets us a discrete valuation ring
and thus we’re back in the world of PIDs.

Example 51
Suppose X is an affine algebraic curve (that is, an affine variety of dimension 1), then O(X) is a Dedekind domain
if and only if X is nonsingular (we will define this later).

For illustration, C[X, Y |Y 2 = X 2 (X +1)] is not integrally closed, so it is not a Dedekind domain. (The corresponding
y
curve has a singularity at 0, and indeed t = x, which captures the “different slopes” near 0 for the curve, is in the

17
integral closure.) so it is not a Dedekind domain, but C[X, Y |Y 2 = X(X 2 − 1)] is an elliptic curve which is nonsingular,
so the corresponding ring is a Dedekind domain.
We’re now ready to talk about the going-up and going-down theorems, which tell us how chains of primes behave
with respect to integral extensions. The original results come from a paper of Cohen and Seidenberg which can be
found here.

Theorem 52 (Going up)

Let B/A be an integral extension of commutative rings. (Sometimes we require that A and B are integral domains.)
Recall that a prime ideal P of B is “above” a prime ideal ℘ of A if P ∩ A = ℘. Then if ℘1 ⊂ ℘2 are prime ideals of
A, and P1 is a prime ideal of B above ℘1 , then there is a prime ideal P2 of B above ℘2 with P1 ⊂ P2 . In fact,
if ℘1 ⊊ ℘2 then we may choose P2 to be strictly larger than P1 .

Visually, we can imagine that we are trying to fill in the blue part of this diagram:
P1 ⊂ P2

℘1 ⊂ ℘2
Proof. We’ve proven previously that for any prime ideal ℘ of A, there is some prime of B that lies above it. We will
let A = A/℘1 and B = B/P1 – we then get a natural inclusion A ,→ B (this is injective because anything in the kernel
of the map A → B would be in P1 ∩ A = ℘1 ). Then B is still integral over A, so we can apply that previous result to
the prime ideal ℘2 of A. Furthermore, in the previous proof it is clear that ℘ ̸= 0 implies P ̸= 0, so we now get an
ideal P2 of B above ℘2 , and pulling that back to B yields an ideal P2 above ℘2 that contains ℘1 . And we do indeed
have P2 ∩ A = ℘2 because taking these quotients commutes with inclusions.

Proposition 53
If B/A is an integral extension with P and ℘ prime ideals of B and A respectively with P ∩ A = ℘, then P is
maximal if and only if ℘ is maximal.

Proof. Again consider A = A/℘ and B = B/P as in the previous proof – we still have an inclusion A ,→ B, and B
is integral over A. We must prove that A is a field if and only if B is a field – we’ll drop the bars from here on for
notational convenience.
This proof only relies on the integrality of the extension and the fact that A and B are integral domains (since
we mod out by a prime ideal). First of all, if A is a field and some nonzero x ∈ B is integral over A, then 0 =
x n + an−1 x n−1 + · · · + a0 for some n; we can divide this relation by powers of x until the constant term is nonzero
because we have an integral domain. But then −x(x n−1 + an−1 x n−2 + · · · + a1 )/a0 = 1, so x is invertible. Thus B is
a field since any nonzero element must be invertible. For the other direction, suppose A is not a field. Then there is
some nonzero maximal ideal ℘, meaning that there is some prime ideal P of B above ℘ which is also nonzero. But
the only proper ideal of a field is the zero ideal, so B cannot be a field.

The going-down theorem is similar but more subtle – we’ll need an additional hypothesis:

Theorem 54 (Going down)

Let A and B be integral domains with B/A an integral extension and A integrally closed. Suppose ℘1 ⊆ ℘2 are
prime ideals of A and P2 is a prime ideal of B above ℘2 . Then there exists a prime ideal P1 of B such that
P1 ⊆ P2 and P1 lies above ℘1 .

18
 In other words, we are trying to fill in a different corner of the diagram:

P1 ⊂ P2

℘1 ⊂ ℘2

We’ll talk about the proof next time and show an illustrative example, and we’ll just introduce an important concept
here (and start next time with the proof of the subsequent result):

Definition 55
Let A be a domain, let a be an ideal of A, and let E ⊃ A be a field. An element x ∈ E is integral over a if there
is some equation x n + an−1 x n−1 + · · · + a0 = 0 with ai ∈ a.

Proposition 56
If A is an integrally closed integral domain with field of fractions F , E/F is a finite extension, and B is the integral
closure of A in E, then {x ∈ B : x integral over a} is the radical r (aB).

10 February 1, 2023
Last time, we proved the going-up theorem, which basically says that given ℘1 ⊂ ℘2 and P1 lying over ℘1 , we can
find a corresponding P2 containing ℘1 lying over ℘2 . And as we stated, the going-down theorem is similar but going
from P2 to P1 , and it requires an additional assumption.
Recall that if X is a variety (assume irreducible) over an algebraically closed field, so that O(X) is an integral
domain, we can let K be its fraction field (which we also call the function field of X). Then the dimension of X is
defined as the transcendence degree of K over F , and the combinatorial dimension is the d corresponding to any
maximal chain of nonempty irreducible closed subspaces X0 ⊆ X1 ⊆ · · · ⊆ Xd . But because we have a dictionary
between irreducible closed subspaces and prime ideals of O(X) (which is an order-reversing bijection), this is the
same as the Krull dimension of X, which is the corresponding d for a maximal length of a chain of prime ideals
0 = ℘0 ⊊ ℘1 ⊊ · · · ⊊ ℘d . But we can now make the final connection:

Theorem 57
Let X be a variety (a subspace of Fn ). Then the combinatorial dimension of X is the same as the dimension of
X (with the definitions above).

Proof. First we will prove that O(X) has minimal nonzero prime ideals (in other words that we have some ℘1 , or
equivalently that there are maximal proper irreducible subsets). Choose a transcendence basis (y1 , · · · , yd ) of K
(where d is the dimension of X), so that B = O(X) is integral over A = F [y1 , · · · , yd ] (by Noether normalization).
Since the yi s are algebraically independent, A is a polynomial ring and a unique factorization domain, and thus the
minimal nonzero primes are (f ) for some irreducible polynomial f . So in A we have (0) ⊂ (f ), so by the going-up
theorem we have (0) ⊂ ℘1 in B such that ℘1 lies above (f ). We claim that ℘ is a minimal prime ideal of B – indeed,
we know that if B/A is integral and P1 ⊇ P2 are primes of B, then ℘1 = P1 ∩ A being the same as ℘2 = P1 ∩ P2
implies that P1 = P2 . So we can’t squeeze any other prime ideal between 0 and ℘1 , and furthermore any minimal
prime of B arises in this way because intersecting it with A will yield some nonzero prime in A which is also minimal.

19
 Now define Y = V (℘1 ) ⊂ X be the corresponding maximal proper irreducible subset. We claim that dim(Y ) =
dim(X) − 1. Indeed, O(Y ) = O(X)/℘1 is then integral over F [y 1 , · · · , y d ] = F [y1 , · · · , yd ]/(f ), and (without loss of
generality) the polynomial f must involves yd , so y d is algebraic over F [y 1 , · · · , y d−1 ]. So the transcendence degree
of the field of fractions of O(Y ) is indeed less than d. But y 1 , · · · , y d−1 are also algebraically independent, since
any polynomial relation g(y 1 , · · · , y d−1 ) = 0 corresponds to having g ∈ F [y1 , · · · , yd ] being in the kernel (f ) of the
quotient, and f involves yd so this is not possible unless g = 0. So the transcendence degree indeed drops by one.
On the other hand, the combinatorial dimension of Y is also one less than the combinatorial dimension of X,
because the images ℘i of prime ideals ℘i ⊆ O(X) in O(Y ) will satisfy 0 = ℘1 ⊊ ℘2 ⊆ · · · ⊊ ℘d . And because every
minimal prime ideal of O(X) arises this way, there’s no other chain that is longer. So by induction this proves the
result.

We’ll now talk more about the proof of the going-down theorem, where we should recall that we’re filling in the
top left corner of this diagram:

P1 ⊂ P2

℘1 ⊂ ℘2

Example 58
We do need the assumption that A is integrally closed for this result to hold. An instructive counterexample goes
as follows: consider the curve C = {(x, y ) : y 2 = x 2 (x + 1)}, which has a singularity at the origin.

y
As we’ve already discussed, A = F [x, y : y 2 = x 2 (x + 1)] is not integrally closed, since t = x is integral over A but
2 2
not in A. Then F [t] contains F [x, y ], since it contains x = t − 1 and y = t(t − 1), so we have an integral extension
F [t] over F [x, y ]. Then F [t] = O(D) is the coordinate ring for the affine line, and thus F [x, y ] ←- F [t] corresponds
to a morphism A1 → C which sends T to (T 2 − 1, T (T 2 − 1)) (so it finds the point on the curve of slope T ). This
map is not injective, since T = ±1 both map to the origin.

Remark 59. More generally, we can prove (by looking at local rings) that for any curve C (meaning any variety
of dimension 1), we have O(C) integrally closed if and only if C is nonsingular. Then the integral closure can be
interpreted as O(D) for some other curve D, yielding a morphism D → C which resolves the singularities (meaning
that the preimage of a singularity will come apart into several points). For higher-dimensional varieties, though, the
integral closure might still be singular for a general algebraic surface. We then get a normal variety, which is less
singular, but it’s still harder to work with and was only really done in the 1960s.

So to get the counterexample we’re looking for, consider the cylinder C × A1 , for which the function field is
F [x, y , z : y 2 = x 2 (x + 1)]. This cylinder contains a similarly parameterized curve z 7→ (z 2 − 1, −z(z 2 − 1), z) which
does not self-intersect but includes the two points (0, 0, 1) and (0, 0, −1) – let this curve be W . We thus have a
mapping A2 → C × A1 given by
(T, Z) 7→ (T 2 − 1, T (T 2 − 1), Z),

and this corresponds to F [t, z] sitting above F [x, y , z ]. The point now is that under this, W ′ = {(t, z) : t + z = 0}
will map to W .
Now the image of W is an irreducible curve, so W corresponds to some prime ideal ℘1 in F [x, y , z]: specifically
℘1 = (xz + y , z 2 − 1 + x). That ideal corresponds above to the unique prime ideal P1 = (t + z ) in F [t, z ]. Then we

20
also have the maximal ideal ℘2 = (x, y , z − 1), but there are two prime ideals of F [t, z] that lie above this, namely
P2 = (t − 1, z − 1) and P′2 = (t + 1, z − 1). And now if we choose P2 , corresponding to a point not actually on W ′ ,
then there is no P1 that will satisfy the going-down theorem.

11 February 3, 2023
Last time, we showed that the going-down theorem does not always hold without the necessary assumption of being
integrally closed. (We can take a look at the paper of Cohen and Seidenberg if we want more details – it’s linked on
the course website.) Thus, the statement we are instead trying to prove is that for any integral extension B/A with
A an integrally closed integral domain, for any ℘1 ⊂ ℘2 prime ideals of A, and for any q2 prime ideal of B above ℘2
(changing notation here), there is some prime q1 of B above ℘1 with q1 ⊂ q2 . Recall that the geometry of this setup
is that if we have a dominant morphism X → Y with A = O(Y ) and B = O(X), and we have varieties V (℘2 ) ⊂ V (℘1 )
of Y , then if V (q2 ) maps to V (℘2 ) we want to find a variety V (q1 ) containing it that maps to V (℘1 ). And the point
is that Y has to satisfy some additional conditions if we have the integrally closed suggestion – in the case where it’s
one-dimensional it’s just saying that the curve is nonsingular.
To prepare for the proof, notice that if ℘ is a prime ideal of A (not necessarily maximal), then A℘ is a local ring
which has a unique maximal ideal ℘A℘ and with ℘A℘ ∩ A = ℘. This has some useful applications:

Proposition 60
If B/A is an integral extension and ℘ is any prime ideal of A, then ℘B ∩ A = ℘.

Note here that ℘ is not assumed to be maximal here, so ℘B may not even be a prime ideal – for example if A = Z
and B = Z[i ], then (3)B is maximal but not (5)B, since (2 + i )(2 − i ) = 5. So instead (5)B factors into a product of
prime ideals (since we have a Dedekind domain).

Proof. By the Nakayama lemma, ℘B is a proper ideal of B. We know that the prime ideals of S −1 B are in bijection
with the prime ideals of B that don’t meet S (by applying intersection and extension). Applying this to S = A − ℘, we
know that A℘ is a local ring, and B℘ = S −1 B is not necessarily a local ring but ℘B does not meet S so it is contained
in some maximal ideal P ′ of B℘ . Then P = P ′ ∩ B is prime, and P ∩ A = P ′ ∩ A℘ ∩ A ⊂ ℘A℘ ∩ A = ℘.

So the point is that passing to a local ring can allow us to make arguments of this type by taking advantage of the
unique maximal ideal, and we’re basically drawing the following picture:

B B℘ = S −1 B

A A℘

Here is another application:

Proposition 61
Let ℘ be a prime ideal of A, and suppose A ⊂ B are rings (with B not necessarily integral over A). Assume that
℘B ∩ A = ℘. Then there is some prime ideal of B above ℘.

Proof. Again look at a square as above. If we start with ℘B in the top left corner, then we claim that ℘B℘ must
P Q
be proper – indeed, if 1 = pi bi /si , then si would be in ℘B ∩ A = ℘, which is a contradiction. So ℘B℘ is

21
contained in some maximal ideal P ′ of B℘ , and now as before let P = P ′ ∩ B be the prime ideal in B. Then
P ∩ A = P ′ ∩ A℘ ∩ A = ℘A℘ ∩ A = ℘ because we have a local ring.

So the point is that localizing is magical because we can “pretend that ideals are maximal even if they aren’t.”
We’re now ready to return to the proof of going-down, starting by restating a definition and result from a previous
lecture:

Definition 62
Let A be an integral domain, E be a field containing A ,and a an ideal of A. We say that x ∈ E is integral over
a if it is the root of a monic polynomial with coefficients in a.

Proposition 63
Let B be integral over A, with field of fractions F, E for A, B respectively. Then {x ∈ B : x integral over a} =
r (aB).

Proof. First we prove that if x ∈ B is integral over A, then x is in the radical of aB. Indeed, we have
X
x n + an−1 x n−1 + · · · + a0 = 0 =⇒ x n = − ai x i ∈ aB,
i

and thus x is in r (aB). For the other direction, it suffices to show that elements of aB are integral over a, because
then if x ∈ r (aB), then x n ∈ aB and thus x n is integral over a – the polynomial equation for which x n is a root then
also shows that there is a monic polynomial for which x is a root.
To do that, suppose x ∈ aB – clearly this is true for x = 0, so we’ll assume x ̸= 0. We work with the ring A[x −1 ],
and we define the ideal
b = {y ∈ A[x −1 ] : xy ∈ aA[x −1 ]}

If we can show that b is not a proper ideal (so in fact b = A[x −1 ], then it must contain 1 and that will give us
an integrality relation for x. Suppose otherwise, so that b is contained in some maximal ideal m of A[x −1 ]. By the
extension theorem, the homomorphism A[x −1 ] → A[x −1 /m extends to a homomorphism Φ : V → A[x −1 ]/m, where V
is a valuation ring of E containing A[x −1 ]. Then x ∈ V , because V ⊃ A and is integrally closed because it’s a valuation
ring, and x ∈ B so it is integral over A. And x −1 ∈ A[x −1 ⊂ V , so both x and x −1 are in V . But then Φ(x −1 a) maps
to zero because we’re extending a homomorphism that mods out by m, and Φ(x) is nonzero (because x is a unit in
V ), hence a unit, meaning that Φ(a) = 0 as well, meaning Φ(aB) = 0. But x ∈ aB means that Φ(x) = 0, which is a
contradiction.
So b is indeed all of A[x −1 ], hence containing 1, and in particular that means that means x ∈ aA[x −1 ] by definition
of the ideal. Thus we have x = a0 + a1 x −1 + · · · + an x −n for some ai ∈ a, and rearranging yields a monic polynomial
showing that x is integral over a, as desired.

Proposition 64
Let A be integrally closed with field of fractions F , E ⊃ F a field, and ℘ a prime ideal of A. Let x ∈ E be integral
over ℘, and suppose x n + an−1 x n−1 + a0 be the (minimal) monic irreducible polynomial in F [x]. Then ai ∈ ℘.

(The idea is that the polynomial showing integrality of x over ℘ may not be this irreducible one, but the irreducible
one also has coefficients in ℘.)

22
Proof. Without loss of generality we may enlarge E and assume it contains all Galois conjugates αi of x. If B is the
integral closure of A in E, then x n + an−1 x n−1 + · · · + a0 factors as i (x − αi ), with α1 = x and potentially with
Q

repetition if the extension is not separable. But then the ai are all in r (℘B) ⊆ r (℘B ∩ A) = r (℘) = ℘, as desired.

We’ll prove going-down using this result next time and then start dimension theory, including primary decomposition
(and for more details we can read the book).

12 February 6, 2023
We’ll start today by proving the going-down theorem. We’ll use the notation mentioned last lecture, where we have
℘1 ⊂ ℘2 prime ideals in A and q2 lying above ℘2 in B (where B/A is an integral extension of integral domains, and A
is integrally closed). Our goal is to show that we can find q1 contained in q2 and lying above ℘1 .

Proof. Let F and E be the field of fractions for A and B, respectively. Last time, we proved that if E/F is a finite
extension of fields and A is an integral domain with field of fractions F , then if x ∈ E is integral over some prime ideal
℘ of A (meaning that some monic polynomial with coefficients in ℘ has x as a root), then the minimal polynomial
x n + an−1 x n−1 + · · · + a0 of x over F also has coefficients in ℘. (The idea is that we can assume E/F is normal without
loss of generality, and we can use that x being integral over ℘ is equivalent to having x ∈ r (B℘).)
To apply that to this theorem, first we’ll prove that ℘1 Bq2 ∩ A = ℘1 . It is clear that ℘1 ⊂ ℘1 Bq2 ∩ A. For the
y
other direction, let x ∈ ℘1 Bq2 ∩ A, so that we can write x = s for y ∈ ℘1 B and s ∈ B − q2 . By the criterion for
integral dependence, since we know that y ∈ ℘1 B ⊂ r (℘1 B), we also know that y is integral over ℘, and thus the
minimal polynomial y n + a1 y n−1 + an = 0 has all coefficients in ℘1 . Now because s = yx , we have (dividing through by
x n)
ai
s n + b1 s n−1 + · · · + bn = 0,
, bi =
xi
and (by our homework problem) these coefficients are also in A. But now if (for the sake of contradiction) x were not
in ℘1 , then ai = x i bi is in ℘ but x ̸∈ ℘, meaning bi ∈ ℘ for all i . So then s would need to be integral over ℘, meaning
that (again by the criterion for integral dependence) s ∈ r (B℘1 ) ⊂ r (B℘2 ) ⊂ r (q2 ) = q2 (last step because we have a
prime ideal). This is a contradiction because s is in the complement of q2 . So in fact x must be in ℘1 and the boxed
equality is proved.
So now ℘1 Bq2 is contained in some prime ideal q of Bq2 , and we may define q1 = q ∩ B. (So we start with ℘1 ,
extend it to Bq2 , and pull it back to B.) Here we’re using the fact that if ℘B ∩ A = ℘, then there is a prime of B
above ℘ (via localizing). We know that q1 does not meet B − q2 , since elements of the latter are units in Bq2 . This
ideal then lies above p1 , so it is the desired one.

We’ll now begin our discussion of dimension theory, focusing on singularities (particularly of curves) and discrete
valuation rings. We’ll fix a ground field F which is algebraically closed.

Definition 65
Let X ,→ A2 be a plane curve, corresponding to some irreducible f (X, Y ) ∈ F [X, Y ]. A point (a, b) ∈ X is
∂f ∂f
singular if ∂X and ∂Y vanish at (a, b).

We’ve mentioned the example curve f (X, Y ) = Y 2 − X 2 (X + 1) before, and we can indeed see that (0, 0) is a
singular point because there are no degree-1 terms.

23
 Proposition 66
Any curve has only finitely many singular points.

∂f ∂f
Proof. The only way for ∂X and ∂Y to both be identically zero is if we’re in characteristic p and all monomials have
exponent a multiple of p, but it turns out that in that case we won’t have an irreducible polynomial. Furthermore, if
∂f
∂X is nonzero, then it is not a multiple of f because f is irreducible. So we can eliminate one variable and see that
∂f
∂X and F can only have finitely many common zeros.

Theorem 67
A point (a, b) is not a singularity if and only if the local ring O(a,b) (of the coordinate ring O(X)) is a discrete
valuation ring.

Proof. We’ll just prove the forward direction. We can assume without loss of generality by translation that (a, b) =
∂f ∂f
(0, 0). Since the point is not singular, we can’t have both ∂X and ∂Y vanish, so we may assume without loss of
∂f
generality that ∂Y (0, 0) is nonzero. We will show that the maximal ideal of this local ring is principal and generated
by X; to do this, we’ll show that if g(X, Y ) vanishes at (0, 0) and if x, y are the images of X, Y in the coordinate ring
(which is F [x, y ] ∼
= F [X, Y ]/(f )), then g(x, y ) is a multiple of x in the local ring. We know that g(0, Y ) vanishes at
Y = 0, so we have g(0, Y ) = Y g1 (Y ) for some polynomial g1 ∈ F [Y ], and similarly we have f (0, Y ) = Y f1 (Y ). If we
now consider f1 (Y )g(X, Y )−g1 (Y )f (X, Y ), this polynomial vanishes when X = 0 because both terms are Y f1 (Y )g1 (Y ),
so we have
f1 (Y )g(X, Y ) − g1 (Y )f (X, Y ) = Xh(X, Y ).

Substituting in x, y for X, Y , we find that f1 (y )g(x, y ) = xh(x, y ) ∈ (x) in the coordinate ring. But f1 (y ) ̸= 0 because
∂f
∂Y (0, 0) is nonzero by assumption (indeed, this comes from differentiating f (0, Y ) = Y f1 (Y ) at Y = 0 and using
the product rule), so f1 (y ) is a unit in A(0,0) . Thus g(x, y ) = xh(x, y )f −1 (x, y ) is indeed a multiple of x, as desired.
There’s a bit more work to showing that we do have a discrete valuation ring, but we can read up on that on our own.
(There’s a bit of Nakayama lemma involved, for example showing that if mAm is principal and generated by x, then
Am is a discrete valuation ring.)

We can now make the more general definition:

Definition 68
More generally, let A be an F -algebra of Krull dimension n. If m is a maximal ideal, then the localization Am is a
regular local ring if mAm can be generated by n elements. a variety is nonsingular at a point if the local ring at
that point is a regular local ring.

For example, we have a local ring of dimension 1 in our case, and the ideal for the discrete valuation ring is
generated by a single element. So for a curve, a necessary and sufficient condition is that we have a regular local ring
of dimension 1, which is equivalent to having a discrete valuation ring by Theorem 67.

13 February 8, 2023
Primary decomposition is a necessary tool for dimension theory, so that’s what our topic will be today. Recall that in
Nk
a Dedekind domain, any nonzero ideal is of the form a = ℘N
1 · · · ℘k for prime ideals ℘1 , · · · , ℘k . Assuming that the
1

24
 ℘N
T
℘i s are all distinct (so the repetition is encoded in the Ni s), we then also have a = i i . The point of the primary
i

decomposition is to generalize this fact to at least Noetherian rings.

Definition 69
Let A be a (commutative) ring. An ideal a of A is primary if xy ∈ a implies that either x ∈ a or y n ∈ a for some
n.

This condition may look a bit strange because it’s not symmetric in x and y , but it does turn out to be a useful
notion. The idea is that these primary ideals will take the role of the prime ideals in the Dedekind domain case.

Proposition 70
If a is primary, then r (a) = {x ∈ A : x n ∈ a for some n} is prime.

Proof. If xy ∈ r (a), then we know that x n y n ∈ a for some integers m, n. This means that either x n ∈ a or (y n )N ∈ a
for some N, but in the former case x ∈ r (a) and in the latter y ∈ r (a).
Nk Ni
And notice that when we have a = ℘N
1 · · · ℘k for prime ℘i in the Dedekind domain case, each ℘i is actually
1

primary. So the analogous statement to show is that for any Noetherian ring A, every ideal is a finite intersection
of primary ideals, and we’ll be working towards that goal.

Definition 71
If a is primary with corresponding prime ideal ℘ = r (a), then we say that a is ℘-primary.

Proposition 72
If q1 , q2 are ℘-primary, then so is q1 ∩ q2 .

Proof. Suppose that we have xy ∈ q1 ∩ q2 but x ̸∈ q1 ∩ q2 . Without loss of generality say that x ̸∈ q1 , which means
that y ∈ r (q1 ) = ℘, which is also r (q2 ) by assumption. But this means that y ∈ r (q1 ∩ q2 ) = ℘ (by taking a high
enough power of y ), as desired.

It is not true that r (a) = ℘ being prime implies that a is primary, and we’ll show an example in our homework.
However, it is true that if r (a) is maximal, then a is primary (also an exercise for us), and that implies the following
result:

Proposition 73
Let A be noetherian. If m is a maximal ideal, then an ideal a is m-primary if and only if m ⊇ a ⊇ mn for some n.

Proof. For one direction, suppose a is primary with a ⊂ m and r (a) = m. Since m is finitely generated, we may let
m = ⟨x1 , · · · , xN ⟩, and then xik ∈ a for all i for sufficiently large k. Then any element of m is of the form
P
ai xi ,
and raising this to the Nk power yields a sum where all monomials have some degree-k power of one of the xi s, so
mNk ⊆ a.
For the other direction, we use the exercise above: if m ⊇ a ⊇ mn , then m = r (m) contains r (a), which contains
r (mn ) = m. Thus r (a) = m and thus a is m-primary.

25
 Definition 74
A primary decomposition represents an ideal a as a finite intersection of primary ideals a = q1 ∩ · · · qr .

If we group together (intersect) ideals that are primary with respect to the same prime, we may assume that all
corresponding ℘i are distinct by Proposition 72 (meaning there is only one ℘-primary ideal of each ℘), and we may
T
also assume that qi is not contained in j̸=i qj (or else we could just throw away qi ). Then the minimal prime ideals
pi = r (qi ) that appear are unique, but overall this decomposition may still not be unique:

Example 75
Consider the polynomial ring F [X, Y ]. Then we claim (X) ∩ (X 2 , XY, Y 2 ) is the same ideal as (X) ∩ (X 2 , Y ).

Indeed, (X 2 , Y ) is clearly contained in (X 2 , XY, Y 2 ). On the other hand, an element α of (X 2 , XY, Y 2 ) is any
polynomial whose monomials all have degree at least 2, so the intersection (X) ∩ (X 2 , XY, Y 2 ) is a sum of monomials
aij X i Y j with i + j ≥ 2 and i ̸= 0 – thus it is also a monomial in (X 2 , Y ). But now r (X 2 , XY, Y 2 ) = r (X 2 , Y ) = (X, Y )
is a maximal ideal, and (X) is prime, so everything here is primary. So the uniqueness does hold at the level of the
primes, but not at the level of primary ideals. And now we’ll show existence:

Theorem 76
In a noetherian ring A, any ideal is the intersection of primary ideals.

Proof. Call an ideal irreducible if it is not the finite intersection of two larger ideals (that is, if a = b ∩ c, then either
a = b or a = c). First note that every ideal is an intersection of irreducible ideals by a Zorn’s lemma argument.
Indeed, if there is some ideal with no irreducible decomposition, then choose a maximal such counterexample (here
we use Noetherianness). Then a is not irreducible, so we can write a = b ∩ c with b, c larger, meaning that each of
them is an intersection of irreducibles. But this gives us a way to write a as an intersection of irreducibles, which is a
contradiction.
So now it suffices to show that every irreducible ideal is actually primary. Let a be irreducible, and let A = A/a, so
that the zero ideal is not the intersection of nonzero ideals. We’ll drop the bars in the notation and just prove that for
any ring where 0 is not the intersection of two nonzero ideals, the ideal (0) is primary (in other words, zero divisors
are nilpotent). Indeed, (0) being primary in A is exactly the same condition as a being primary in A.
Suppose that xy = 0 but x ̸= 0; we wish to show that y n = 0 for some n. Define the ideals

an = {z ∈ A : z y n = 0}.

If z annihilates y n , then it annihilates y n+1 as well, so a1 ⊂ a2 ⊂ · · · , so by the ascending chain condition we have
aN = aN+1 for some N. In other words, z y N+1 = 0 implies that z y N = 0. We claim that y N = 0; suppose otherwise.
Then (x) and (y N ) are both nonzero ideals, so (x) ∩ (y N ) is not the zero ideal and the intersection contains some
a ∈ A. Then a = z1 x = z2 y N for some z1 , z2 ∈ A, so

0 = z1 xy = ay = y N+1 z2 =⇒ a = y N z2 = 0,

which is a contradiction. Thus irreducible ideals are indeed primary and we always have a finite decomposition.

26
14 February 10, 2023
Today’s topics are the Hilbert polynomial and the statement of the dimension theorem, which is the equivalence
of three different definitions of dimension for rings in algebraic geometry (due to Krull). A good reference for this
material is the set of lecture notes on the website or the last chapter of Atiyah and Macdonald.

Definition 77
L∞
A graded ring is a ring G = i=0 Gi in which G0 is a ring, each Gi is a G0 abelian group module, the multiplication
respects the grading (meaning that Gi Gj ⊆ Gi+j ) and is bilinear and associative.

Here, G0 will be a ring, and its unit will be a unit for the entire graded ring. But the higher Gi s don’t necessarily
have to be rings anymore (in particular they are not closed under multiplication). We often take G0 = F a field and
Gi abelian group modules for G0 .

Example 78
A polynomial ring is graded by degree, and more generally we can often replace an affine algebra by a related
graded ring. For example, if we take y 2 = x(x 2 − 1), which is an elliptic curve, then the complex points on this
curve actually form a torus with a point missing (which is the “point at infinity”).

To recover this point at infinity, we can make the equation homogeneous by adding appropriate powers of an extra
variable z , so that the equation becomes y 2 z = x 3 − xz 2 . We can now think of (x, y , z) as projective coordinates,
meaning we identify (x, y , z) with (λx, λy , λz ) for any nonzero λ; that is, we define

P2 = {(x, y , z )not all zero : (x, y , z ) = (λx, λy , λz ) for all λ ̸= 0}.

This contains A2 = {(x, y , 1)}, so most of the points in P2 are the ones in A2 . But there’s also an additional line
at infinity of points of the form (x, y , 0), and out of these the point (0, 1, 0) is on our elliptic curve. And now the
polynomial f (x, y , z) = y 2 z − x 3 + xz 2 is homogeneous of degree 3, and that allows us to associate to this elliptic
curve the graded ring of the projective variety F[X, Y, Z]/(f ), with grading again given by degree.
L
The idea is that when G = i Gi , then the elements within a given Gi are called homogeneous of degree i , and
those are the ones that actually have meaning in the algebraic geometry sense here. (But we should think of them as
sections of a line bundle instead of as functions.)

Example 79
Let A be a commutative ring and a an ideal (which we should think of as being maximal). Then set G0 = A/a
and Gi = ai /ai+1 for all i > 0.

To check that this gives us a graded ring, we must check that multiplication respects the grading. Specifically, we
should check that the multiplication map induces a map ai ⊕ aj → ai+j , and that (ai+1 , 0) and (0, ai+j+1 ) map into
ai+j+1 , so that we get an induced map ai /ai+1 × aj /aj+1 → ai+j /ai+j+1 by the universal property of the quotient.

Definition 80
L
Let G = i Gi be a Noetherian graded ring, and assume that G0 is a field. A graded module of G is a module
L
M = i Mi such that Gi Mj ⊆ Mi+j .

27
 In particular, G is a graded module over itself.

Definition 81
P∞
For M a graded module of G (where G0 is a field), the Hilbert series of M is PM (t) = i=0 dimG0 (Mi )t i .

Theorem 82
Let G be a graded Noetherian ring. Suppose G is generated as an algebra by X1 , · · · , Xn , where each Xi is
homogeneous of degree di (meaning that Xi ∈ Gdi ). Then for any finitely generated graded module M of G, we
have
f (t)
PM (t) =
(1 − t d1 ) · · · (1 − t dn )
for some polynomial f (t).

Proof. We will assume G0 is a field – the case where G0 is not a field we’ll do later on. We induct on the number
of generators n – the base case n = 0 is true because G = F and we just have a finite-dimensional vector space.
Notice that G0′ = F [x1 , · · · , Xn−1 ] is also a graded ring, and M can also be thought of as a G0′ -module. Consider the
multiplication map Mi → Mi+dn given by multiplication by xn . Defining Q = M/Xn M, the map Mi+dn → Qi+dn is
surjective, so we have an exact sequence

0 → Ki → Mi → Mi+dn → Qi+dn → 0

where Ki is the kernel of the multiplication-by-Xn map in Mi . We claim that this means

dim(Ki ) − dim(Mi ) + dim(Mi+dn ) − dim(Qi+dn ) = 0

(we can do this by introducing another term in the short exact sequence with 0 → Ki → Mi → Ai → 0 and
0 → Ai → Mi+di → Qi+dn → 0). Multiplying this equation by t i+dn and summing over i (starting from −dn ; here we’re
defining the dimension of Mi to be zero for i < 0 and that still makes the equation hold), we find that

t dn PK (t) − t dn PM (t) + PM (t) − PQ (t) = 0,

PQ (t)−t dn PK (t)
and rearranging this shows that PM (t) = 1−t dn . But Q and K are both annihilated by xn , and this same result
′
holds whether we regard them as G-modules or G -modules. So by the inductive hypothesis both of those terms are
1
(1−t d1 )···(1−t dn−1 )
times a polynomial. Plugging that in yields the result.
L∞ i i+1
Suppose A is a Noetherian ring and m is a maximal ideal of A. We then have G = Gm = i=0 (m /m ), and
2
G1 = m/m is then important in connection with singularities and the Zariski tangent space, which we can read about
on our own.

Theorem 83
There is a polynomial χm (k) such that for all sufficiently large k, the length ℓ(A/mk ) of a module is equal to
χm (k).

The idea is that A/mk might not be a vector space even though mk−1 /mk is, so we need this more general definition:

28
 Definition 84
Suppose a module M has a composition series 0 = M0 ⊊ M1 ⊊ · · · ⊊ Md = M, where all Mi /Mi−1 are simple
modules (that is, they have no nonzero proper submodules). By Jordan-Hölder (proof is identical as the one for
groups), all such series have the same length and quotients up to ordering. The length of M is then the value of
d (or ∞ if not).

If we have a short exact sequence 0 → M ′ → M → M ′′ → 0, we can put the composition series for M ′ and M ′′
together and get one for M, so ℓ(M) = ℓ(M ′ ) + ℓ(M ′′ ). We’ll prove this last result about the Hilbert polynomial next
time!

15 February 13, 2023

Our goal today is to state the Krull dimension theorem (showing equivalence of several definitions of dimension). Recall
that we previously defined the Hilbert series of a graded module M of a graded ring G (currently defining only when G0
is a field) to be PM (t) = ∞ i
P L L
i=0 dimG0 (Mi )t (where the gradedness here means that G = Gi and M = Mi with
Gi Gj ⊂ Gi+j and Gi Mj ⊂ Mi+j ). Furthermore, we proved that if G is noetherian and finitely generated by homogeneous
1
elements of degree di , and M is finitely generated, then PM (t) is a polynomial times (1−t d1 )···(1−t dn )
.
We then mentioned one way of constructing graded rings: let A be a noetherian local ring with maximal ideal m.
Then may define
∞
M
Gm = (mk /mk+1 ),
k=0

where the degree-0 part Gm,0 = m0 /m = A/m is a field. Then the multiplication in the ring mk × mℓ → mk+ℓ induces a
multiplication Gm,k × Gm,ℓ → Gm,k+ℓ , so we do get a noetherian graded ring. In fact, this ring is generated by elements
of degree 1 (that is, by m/m2 ).

Lemma 85
If X1 , · · · , Xn generate m/m2 as a vector space over F = A/m, then they generate m as an ideal.

Proof. This is basically a Nakayama’s lemma argument. Consider the submodule m′ of m generated by X1 , · · · , Xn .
Since we generate m/m2 as a vector space, mm′ = m2 (since anything in m can be expressed as a linear combination
2
P
i ci Xi with ci ∈ A, plus some correction term φ ∈ m . Then multiplying any such element by m makes everything
on the right-hand side in m2 . But then m annihilates m/m′ , so by Nakayama’s lemma this means m/m′ = 0 and thus
m = m′ . (Here we do need that A is local.)
L∞
If we now apply the same proof to the case Gm = k=0 Gm,k with Gm,k = mk /mk+1 , then a basis of mk /mk+1
yields a set of generators of mk . Furthermore, if we take the dual space of (m/m2 ), we will get the Zariski tangent
space at a point a if A is a local ring, and thus a variety of X at a. The intuitive understanding (which does hold
for curves) is that if X has dimension d, then the dimension of m/m2 as a vector space (where the maximal ideal
corresponds to some point a) is at least d and is exactly d when X is not singular at a. For example, for the curve
y 2 = x 2 (x + 1), we see that dim(m/m2 ) = 2 at the origin because of the intersection point, but dim(X) = 1.

29
 Definition 86
For a local ring A with maximal ideal m, let δ(A) be the minimum number of generators needed for an m-primary
ideal of A.

Remember that q is m-primary if and only if m ⊇ q ⊇ mn for some n, and in this case we claim that the ideal
q = (x) is m-primary because y 2 = x 2 (x + 1) ⊂ q, meaning all of m2 = (x, y )2 = (x 2 , xy , y 2 ) is contained in q = (x).
But then because q is generated by just a single element, δ(A) = 1 but dim(m/m2 ) = 2 – the point is that we may
want to look at other ideals besides just m itself.
f (t)
Since Gm is generated by elements of degree 1, its Hilbert polynomial then has the form (1−t)d , where d =
2
dim(m/m ). We can look at the power series expansion
∞
1 X
d
= C(d, k)t k ,
(1 − t)
k=0

d+k−1


where C(d, k) = d−1 is a binomial coefficient (specifically, a degree (d − 1) polynomial in k). Last time, we stated
that there is a polynomial χm (k) such that for all sufficiently large k, χm (k) = ℓ(A/mk ) is the length of the module
A/mk (that is, the number of factors in its composition series). In particular, if M is semisimple (for instance since
mk /mk+1 is a vector space, this is just the dimension of M). This is called the Hilbert-Samuel polynomial, and a
key fact is that the length agrees with the dimension (and thus the corresponding Hilbert series) when we have a
semisimple module, for example if mk /mk+1 is a vector space. Furthermore, the length has the nice property that
ℓ(M ′ ) = ℓ(M) + ℓ(M ′′ ) for a short exact sequence 0 → M → M ′ → M ′′ → 0. However, A/mk may not be a vector
space and thus its dimension does not make sense in general (this is why we need to use length instead).

Proof of existence of Hilbert-Samuel polynomial. We claim that

k−1
X
ℓ(A/mk ) = dim(mi /mi+1 ) ,
i=0

since we can calculate length inductively via the short exact sequence 0 → mk /mk+1 → A/mk+1 → A/mk → 0. This
then means that ℓ(A/mk ) = k−1
P
i=0 dim Gm,i , and we can combine this with the expression for the Hilbert polynomial
P k f (t) P j
dim(Gm,k )t = Q(1−t d ) . Specifically, if the numerator is of the form f (t) = cj t , then the series expansion above
i
tells us that
X k − j + d − 1 X k − j + d − 1
dim(Gm,k ) = cj = cj
k −j d −1
j j

is a polynomial in k of degree at most (d − 1) since each term is of degree at most (d − 1), as long as k is large enough
(larger than the degree of f ). Now accumulating within the boxed equation above makes ℓ(A/mk ) a polynomial as
well (here we’re saying that if p(x) is a degree (d − 1) polynomial, then the cumulative sums p(1) + · · · + p(x) yield a
degree d polynomial).

This proof may not be very satisfying because it is not specific enough to show a direct connection, but next time,
we’ll prove the dimension theorem. The statement is that if A is a noetherian local ring, then the following are equal:
(1) δ(A) as defined today, (2) the actual degree of χm , and (3) the dimension of A. The proof is nice – it basically
involves proving the circular implications, each using a different technique.

30
16 February 15, 2023
Since we’re slightly behind schedule with lectures, we’ll start the proof of the dimension theorem today, then discuss
group representation theory next lecture, and then return to the dimension theorem next week.

Theorem 87
Let A be a noetherian local ring with maximal ideal m. Let d(A) be the degree of the Hilbert-Samuel polynomial,
δ(A) be the smallest number of generators in an m-primary ideal of A, and dim(A) be the Krull dimension of A
(the maximum length of a saturated chain of prime ideals). Then d(A) = δ(A) = dim(A).

Recall that part of the magic here is that some m-primary ideal will have fewer generators than m itself if the
coordinate ring corresponding to m is not a regular local ring (that is, a local ring of a singular point). And remember
that we’ve already shown dim(A) is equivalent to the transcendence degree of the fraction field for the corresponding
coordinate ring.
We will first prove that (1) d(A) ≤ δ(A). We have already almost finished proving that d(A) is at most the number
of generators of m, but we need something stronger (in case some m-primary ideal has fewer generators).
Gm,i , where Gm,i = mi /mi+1 and Gm,0 =
L
Let’s first explain that weaker statement. Let Gm be the graded ring
A/m = F is our ground field. Recall that the Hilbert-Samuel polynomial satisfies χm (k) = ℓ(A/mk ) as long as k is at
least the degree of the largest generator of the graded Gm , and it is actually a polynomial. Indeed, remember that the
Hilbert series PGm (t) = ∞ i
)t is of the form (1−t d1g(t)
i+1 i
P
i=0 ℓ(m /m )···(1−t dr )
if G is generated as an F -algebra by x1 , · · · , xr
with xi ∈ Gm,i with degree d1 , · · · , dr . But Gm is generated by elements of m/m2 (that is, elements of degree 1), we
g(t) 1
P r +i−1
i
in fact have Hilbert series (1−t) r . Using that (1−t)r = r −1 t , which has coefficients of degree r − 1 (in i ), and
i i+1
P
that χm (k) = i<k ℓ(m /m ), the accumulation must have at most degree r . But in fact the degree can be less
than r because of potential cancellations:

Example 88
Let m be the local ring of y 2 = x 3 at the origin (0, 0). Then ℓ(mi /mi+1 ) is 1 if i = 0 and 2 if i ≥ 1, but
the Hilbert-Samuel polynomial is χm (k) = 2k + 1, which has degree 1. The idea is that in the expression of
1−t 2
PGm (t) = (1−t)2 , we can in fact cancel a power of 1 − t in the denominator.

So we’re now going to go into the proof more carefully but working with an m-primary ideal.

Proof that d(A) ≤ δ(A). Recall that if q is an m-primary ideal, then m ⊇ q ⊇ mn for some n. Define the graded ring
M
Gq = Gq,i , Gq,i = qi /qi+1 ,

where Gq,0 = A/q is no longer a field and thus we must use length in the definition of

PGq = ℓ(qi /qi+1 )t i .

But the same theory goes through with this new definition, and now χq (k) = ℓ(A/qk ) will be a polynomial of degree at
most the number of generators s needed for q. Note that an element thought of as an element of Gm or of Gq may be of
different degrees, but since m ⊇ q ⊇ mn we know that mk ⊇ qk ⊇ mnk – in particular ℓ(A/mk ) ≤ ℓ(A/qk ) ≤ ℓ(A/mnk )
means that χm (k) ≤ χq (k) ≤ χm (Nk), which can only happen if in fact the degrees of χm and χq are the same. So
the degree of χm is in fact at most the number of generators needed for q as well.

31
 Next, we’ll show that δ(A) is at most the Krull dimension, which will use some properties of the primary decom-
position:

Proposition 89
TN
Let A be a noetherian ring, a be an ideal, and a = i=1 qi be a primary decomposition of a. Let ℘i = r (qi ) be the
associated primes for the primary ideals. Then any prime ideal containing a contains some ℘i .

Q T
Proof. Notice that if ℘ ⊇ a, then ℘ ⊇ qi for some i ; otherwise let xi ∈ qi \ ℘ and notice that xi ∈ qi = a but no
xi is in ℘. So ℘ = r (℘) contains r (qi ) = ℘i .

(Basically, even though the qi s are not canonical, the minimal primes are uniquely determined, and containment
does reflect this fact.)

Proposition 90
SN
Let a be an ideal and ℘i prime ideals of A. If a ⊂ i=1 ℘i (note that this is not the ideal formed by all of the ℘i s),
then a ⊂ ℘i for some i .

Proof. We prove this by induction on N. The base case N = 1 is clear, and now assume the statement holds for
N − 1. Suppose for the sake of contradiction that a ⊂ N
S
i=1 ℘i but a is not contained in any ℘i . Then for any fixed
S
index i there is xi ∈ a \ j̸=i ℘j (here we’ve used the inductive hypothesis). But since a is contained in the union of all
℘j s, that means xi ∈ ℘i . Then the element
XY
x= xi
j i̸=j

is in a because each xi is in a, so x ∈ ℘k for some k. But now all terms except the one where j = k have a copy
Q Q
of xk , so all terms except i̸=k xi are in ℘k , meaning i̸=k xi must be in ℘k as well. But no component is in ℘k ,
contradicting primeness of ℘k . Thus the inductive hypothesis is proved.

Definition 91
The height of a prime ideal ℘ is the maximal k corresponding to a chain of primes ℘0 ⊊ ℘1 ⊊ · · · ⊊ ℘k = ℘.

In particular, the height of a maximal ideal m is exactly the Krull dimension dim(A) (since every chain ends with a
maximal ideal).

Proposition 92
Suppose A is a local noetherian ring. Then there is a chain of elements x1 , · · · , xr of m such that any ideal
containing (x1 , · · · , xi ) has height at least i and such that (x1 , · · · , xr ) is m-primary.

In particular, this means that dim(A) = height(m) ≥ r , which is the number of generators needed for the particular
m-primary ideal (x1 , · · · , xr ). So this implies that δ(A) ≤ dim(A).

Proof. Assume x1 , · · · , xi have already been constructed. If r (x1 , · · · , xi ) = m, then we’re done; otherwise (x1 , · · · , xi )
is not m-primary. There are a finite number of minimal primes containing (x1 , · · · , xi ) (by using the primary decom-
position and Proposition 89), each of which is a proper subideals of m. Since m is not the union of these minimal
primes, we can then find some xi+1 ∈ m not in any of them. Now if ℘ is a prime containing (x1 , · · · , xi ), its height is

32
at least i by the inductive hypothesis, and now if it also contains xi+1 its height must be bigger than i (since it is not
among the minimal primes containing (x1 , · · · , xi )), which is what we wanted to show. And we finish when we get a
m-primary ideal with radical equal to m.

The last part of the proof of the dimension theorem requires the Artin-Rees lemma, which we’ll show on our
homework. And we’ll discuss that last part in the lecture after the next one.

17 February 17, 2023

We’ll discuss semisimplicity today, for which the most relevant example comes from group representation theory.

Definition 93
Let G be a finite group. A representation of G is a homomorphism π : G → GL(V ) for some vector space V
(usually in this class over C and usually finite dimensional).

There is a branch of representation theory in which V is of positive characteristic (say p) where p is a prime
dividing the order of the group G. This is called modular representation theory, which can get information that can’t
be obtained otherwise, but it’s outside the scope of this course. So we will avoid that here and just work over C, since
that gives us the same information as any characteristic p not dividing the order of G and it’s good to work over an
algebraically closed field.

Definition 94
Let F be a field and G be a finite group. The group algebra F [G] is the free vector space on G, which is the set
P
of formal summations g∈G ag g with ag ∈ F .

In particular, F [G] has a ring structure given by

X  X  X X
ag · g bg · g = cg · g, cg = ax by
x,y :xy =g

(in other words, extend the group multiplication by linearity), and it is commutative if and only if G is abelian. (So in
particular we now have to introduce noncommutative rings into the class.)

Theorem 95 (Maschke)
If F is a field of characteristic zero, or of characteristic p with p not dividing |G|, then F [G] is a semisimple group
algebra.

Before we define what semisimplicity is, we’ll first establish a relationship between representations and modules
over the group algebra:

Theorem 96
There is a bijection between representations of G on F -vector spaces and left modules over F [G].

This bijection is given as follows: given π : G → GL(V ), we can define the F [G] module structure given by
X X
( ag · g)v = ag π(g)v .

33
So the point is that π is a homomorphism exactly when this gives us a module structure – we need to check that
if ξ, η are in the group algebra, then ζ(ηv ) = (ζη)v , but that reduces by linearity to the case where ζ, η ∈ G, and
having π(ξ)(π(m)g) = π(ξη)g is equivalent to having π(ξ)π(η) = π(ξη). And if we have a module V , we may define
π : G → GL(V ) by mapping gv = (π(g))v , and these constructions are inverses of each other. Sometimes we say
that V is a G-module instead of saying that it is an F [G]-module.
We may ask whether this vector space V has invariant subspaces, meaning that we have a subspace W ⊆ V
with π(g)W ⊆ W for all g ∈ G. (Being invariant under the group then means we are invariant as a submodule of V
regarded as an F [G]-module, so we can think of it that way as well.) In particular this also means π(g −1 )W ⊆ W and
thus we must have π(g)W = W .

Definition 97
A vector space V is irreducible as a G-module if 0 and V are the only invariant subspaces.

We will prove soon that there are only finitely many isomorphism classes of such irreducibles for any G.

Definition 98
For any ring R, a left R-module M is simple if the only submodules of M are 0 and M.

So in particular, V is irreducible if and only if it is simple as an F [G]-module, so these terms are almost synonymous.

Proposition 99
Let R be a (not necessarily commutative) ring and M an R-module, possibly infinite-dimensional. Then the
following are equivalent:
1. M is a sum of simple submodules,

2. M is a direct sum of simple submodules,

3. (Complete reducibility) For any N ⊆ M, there is some submodule P (not necessarily unique) with M = N ⊕P .
(We say that N is complemented.)
A module satisfying any of these properties is called semisimple.

There are indeed modules that are not semisimple:

Example("100 #) " #
a b 2
1
Let R = ⊆ Mat2 (F ) acting on the module M = F . Then the span N of is a submodule that is
0 d 0
not complemented.

P
Proof. If M is a sum of simple modules, that means that we can write M = j∈J Ej for some (not necessarily finite)
P
index set J, such that any element can be written as x = j xj with xj ∈ Ej and only finitely many xj nonzero. We
then have a direct sum if the representation of x is always unique, and the strategy for showing that (1) implies (2)
is that we can discard some of the Ej s and get a sum that is direct. But this is a Zorn’s lemma argument – there is
P P
some subset I ⊆ J such that M = j∈I Ej is a direct sum, though we do not yet know that j∈I Ej = M. (Note that
P
being direct also means that if i ∈ I, then Ei ∩ j∈I,j̸=i Ej = 0.) We choose a maximal such subset I by Zorn’s lemma,
P L
and we claim that M = j∈I Ej . It suffices to show that every Ej is contained in i∈I Ei , since the Ej s generate the

34
 P
module M. Suppose otherwise for some Ej – then Ej ∩ i∈I Ei is a proper submodule of Ej , so by simplicity of Ej we
P L
must have Ej ∩ i∈I Ei = 0. Thus I ∪ {j} would also yield a direct sum M = i∈I+{j} Ei , contradicting maximality.
L L
(Here j is not in I because otherwise Ej would be contained in i∈I Ei in the first place.) So i∈I Ei must be a direct
sum for M.
L
(2) implies (3) is very similar: if we can write M = i∈J Ei and we have a submodule N ⊂ M, then choose a subset
P P
I ⊂ J maximal within the condition that N ∩ i∈I Ei = 0 by Zorn’s lemma. We can then show that N ⊕ j∈I Ej = M
(otherwise we could add another j to I), and so we’ve found our P .
Finally, (3) implies (1) is the tricky part – it suffices to show that any nonzero submodule E of M contains a simple
module. (Otherwise we can take the complement of the sum of all simple submodules of M, which we can do by
complete reducibility, and arrive at a contradiction.) Let v be some nonzero element of E, and consider the submodule
Rv . The map a 7→ av is a homomorphism R → Rv , and we can let m be a maximal ideal containing the kernel. The
image mv of m is then a maximal submodule of Rv , so by complete reducibility we can let M ′ be an R-module such
that M = mv ⊕ M ′ . We claim that Rv = mv ⊕ (M ′ ∩ Rv ); it’s clear if we put + instead of ⊕, and to show we have
a direct sum we can write x ∈ Rv as x = av + m′ for m′ ∈ M ′ and then notice that m′ = x − av is in Rv because
x, av ∈ Rv , and thus m′ ∈ M ′ ∩ Rv and thus M ′ ∩ Rv ∼
= Rv /mv is simple because we’ve modded out by a maximal
submodule. So we’ve exhibited a simple submodule of Rv , so in particular that is a simple submodule of E.

18 February 22, 2023

We’ll start by proving the last part of the dimension theorem today – recall that we wish to prove that for a local
noetherian ring A with maximal ideal m, d(A) (the degree of the Hilbert-Samuel polynomial), σ(A) (the smallest
number of generators of any m-primary ideal), and dim(A) (the Krull dimension) are equal. All that’s left is to show
that dim(A) ≤ d(A).

Proof. Suppose we have a chain of prime ideals ℘0 ⊊ ℘1 ⊊ · · · ⊊ ℘d = m. We wish to show that d is at most the
degree of the Hilbert-Samuel polynomial χm (which is the polynomial with χm (n) = ℓ(a/mn ) for sufficiently large n).
Notice that we can replace A by A/℘0 , which does not change d but may decrease the values of χm . Thus doing this
cannot increase the degree of the polynomial (or else the values of the polynomial will eventually grow bigger), and
thus we may assume without loss of generality that ℘0 = 0 and that A is an integral domain.
We’ll prove this by induction on d – the base case is a ring of dimension 0, but this is trivial. For the inductive
step, let a be some nonzero element of ℘1 , let a = (a), and let A = A/a. Now A has a chain of ideals ℘1 ⊊ · · · ⊊ ℘d ,
with ℘i = ℘i /a, and we know by induction that d − 1 ≤ deg χm . It suffices now to show that deg χm > deg χm .
There is a homomorphism A/mn → A/mn , whose kernel is (mn + a)/mn ∼= a/(a ∩ mn ). Thus the length can be written
as ℓ(A/mn ) = ℓ(A/mn ) + f (n), where f (n) = ℓ(a/a ∩ mn ). We will prove that f is of the same degree and leading
coefficient as χm , so that subtracting will reduce the degree. f is indeed a polyomial for sufficiently large n, since
it’s the difference of two such terms which are polynomials for large enough n. But by the Artin-Rees lemma (from
homework), we know that
a ∩ mn = mn−r (a ∩ mr )

for some r depending only on a and for all n ≥ r . Then we have amr ⊆ a ∩ mr ⊆ a, and multiplying both sides by mn−r
shows that amn ⊆ (a ∩ mr )mn−r ⊆ amn−r , but the middle term is now a ∩ mn . So this means that (dividing the ideal
a by each of these terms)
ℓ(a/amn−r ) ≤ ℓ(a/a ∩ mn ) ≤ ℓ(a/amn ).

35
But now a is principal, and A is an integral domain, so a is isomorphic to A as an A-module. In other words, we also
have
ℓ(a/mn a) = ℓ(A/mn ) = χm (n)

for large enough n, and thus the inequality is actually saying that

χm (n − r ) ≤ f (n) ≤ χm (n)

for sufficiently large n. So that indeed means f must also be a polynomial of the same degree as χm and of the same
leading order, yielding the result.

(So the dimension theorem is a nice showcase of lots of techinques in commutative algebra. And the Artin-Rees
lemma is also used to prove an important fact about a-adic topologies, which we are also seeing on our homework.)
We’ll now return to semisimplicity again – last time, we proved that for a (not necessarily commutative) ring R,
the following are equivalent for an R-module M: (1) M is a sum of simple R-modules (which are nonzero modules
with no proper nonzero submodules), (2) M is a direct sum of simple R-modules, and (3) M is completely reducible,
meaning that for any submodule N, we have some submodule P such that M = N ⊕ P .

Definition 101
A ring R is semisimple if R is semisimple as a left R-module.

Lemma 102
If M is a semisimple R-module and N is a submodule of M, then N is semisimple.

Proof. By semisimplicity of M, N is complemented and thus we can write M = N ⊕ P . Now for any submodule T of
N, we have M = T ⊕ U for some R-module U. Since N is isomorphic to M/P (from the direct sum), we can define
the image U of U in M = M/P , and M ∼
= U ⊕ T . Then T = (T + P )/P ∼
= T because T ∩ P = 0. Pulling back, since
N is isomorphic to M, we get a complement of T in N. Thus N is completely reducible, hence semisimple.
P
Alternatively, we can say that M is a (direct) sum of simple modules M = Ei , and define the homomorphism
∼
p : M → N be the homomorphism M → M/P = N. This is surjective, so N is the sum of the modules p(Ei ). Since
Ei is simple, each p(Ei ) will either be zero or isomorphic to a simple module, and thus N is also a sum of simple
modules.

Lemma 103
If M is semisimple and Q = M/K for some submodule K of M, then Q is semisimple.

Proof. By complete reducibility, M ∼

= K ⊕ P for some submodule P and Q is isomorphic to P , meaning that every
quotient module is isomorphic to a submodule, hence semisimple.

Lemma 104
L
An arbitrary direct sum M = i∈I Mi of semisimple modules is semisimple.

Proof. Each Mi is a direct sum of simple modules, and thus M is also such a sum.

36
 Proposition 105
If R is semisimple, then every R-module is semisimple.

L
Proof. If M is an R-module, then M is isomorphic to some free module i∈I R modulo an ideal, which is a quotient
of a semisimple module, hence semisimple.

Theorem 106 (Wedderburn)

Lk
A semisimple algebra over a field F must take the form i=1 Matdi (Di ), where Di s are division algebras over F ,
and conversely such a ring is always semisimple.

(Remember that we’re asking these rings to be semisimple over themselves.) In particular, if F is algebraically
closed, then Di = F , which is why doing representation theory over algebraically closed fields is nice. We may consult
Lang for the proof – we won’t go through it here. There’s also the Jacobson density theorem, which we should take a
look at on our own reference, and Burnside’s theorem, which states that if |G| = p a q b for primes p, q, then G is not
a nonabelian simple group (which is easiest to prove using representation theory).
Next time, we’ll prove Maschke’s theorem, which states that whenever we have a field of characteristic zero or of
characteristic p not dividing |G|, the group algebra F [G] is semisimple. (A counterexample is done in the homework
where F is a field of characteristic p and G is the cyclic group of order p.) Then we’ll move on to the Schur
orthogonality relations. The theme we’ll see is that there are methods from group representation theory that have
essential consequences in group theory but cannot be proved without the representation theory tools.

19 February 24, 2023

We’ve previously stated Maschke’s theorem, which can be phrased in the following way: let G be a finite group, F a
field, and char(F ) either zero or a prime not dividing |G|. Then F [G] is semisimple, meaning that every F [G]-module is
completely reducible. Since representations π : G → GL(V ) (for some F -vector space V ) correspond to F [G]-modules,
and completely reducible modules are direct sums of simple modules, this implies that representations can be written
as direct sums of irreducible representations – that is, if π : G → GL(V ) is a representation, then V is a direct sum
of irreducible G-invariant subspaces (that is, there is no subspace that is also G-invariant, which is equivalent to being
an F [G]-submodule).
Towards proving Maschke’s theorem, let M be a module over R, and let N be a submodule. Consider the R-module
homomorphisms p : M → M such that p(M) = N and p acts as the identity on N. We call such an operator p a
projection operator, and notice in particular that p 2 = p (since applying p twice gets into N and then does nothing).
In fact, being a projection operator is equivalent to the conditions that N = p(M) and p 2 = p, and we can visualize
this in the usual linear algebraic sense, and such a projection operator always yields a complement:

Lemma 107
Let p : M → M be a projection onto N. Then M = N ⊕ H, where H = ker(M).

Proof. We may write any element of M as m = n + h, where n = p(m) and h = m − p(m). Then clearly we have
n ∈ N = Im(p) and h ∈ ker(p), since p(h) = p(m − p(m)) = p(m) − p 2 (m) = p(m) − p(m) = 0. Thus this is a valid
decomposition for any m. Furthermore N and H only intersect trivially, since for any x ∈ N ∩ H we have x = p(y ) for

37
some y and thus p(x) = p(p(y )) = p(y ) = x but p(x) = 0 by being in the kernel. Thus we do have a direct sum, and
N and H are both submodules because p is a homomorphism.

So in the proof of Maschke’s theorem, we will construct such a projection operator:

Proof of Maschke’s theorem. We must prove that any submodule N of an F [G]-module M is complemented as an
F [G]-module. It is indeed complemented as an F -vector space, so we can write M = N ⊕ H, where H is a vector
subspace of M. We then have a vector space projection p ′ : M → N with kernel H (that is, write m uniquely in the
direct sum n + h and take p ′ (m) = n), but the problem is that this may not be a module homomorphism. Instead, the
trick is to modify p ′ by averaging and define
1 X ′ −1
p(x) = gp (g x)
|G|
g∈G

1
(here is where we use the hypothesis that the characteristic is not a prime dividing |G|, so that |G| is invertible).
We wish to show that p is an F [G]-module homomorphism and that p is actually a projection onto N. (Applying
Lemma 107 would then allow us to take H = ker(p) and we’d be done.)
First we check that p(x) ∈ N and that if x ∈ N, then p(x) = x. The first part is true because p ′ (g −1 x) is in
N for each g, and N is a G-invariant subspace so gp ′ (g −1 x) is in N for each g; taking the average of these values
then again gives us something in N. And the second part is true because if x ∈ N, then g −1 (x) ∈ N as well, so
gp ′ (g −1 x) = g(g −1 x) = x for all g ∈ G (since p ′ is a projection onto N). Averaging this over all g again gives us x.
Thus p is a projection onto N.
Finally, this map is actually a G-module homomorphism – it suffices to check that γp(x) = p(γx) for any γ ∈ F [G],
and by linearity we can just check this when γ ∈ G. We have
 
1 X ′ −1 1 X 1 X ′ −1
p(γx) = gp (g γx) = γgp ′ (g −1 γ −1 γx) = γ gp (g (x))
|G| g |G| |G|

where in the middle equality we’ve made a substitution g 7→ γg, and the right-hand side is indeed γp(x). So we’ve
turned the projection into a G-module homomorphism, and thus the proof is complete.

Proposition 108
Let R be a semisimple ring. Then R has a finite number of isomorphism classes of simple modules – in particular,
this means that over a field of characteristic zero, any finite group G has a finite number of equivalence classes
of irreducible representations (by applying this to R = F [G]).

(This last part is true even if we were working in characteristic p, but we’ll make use of semisimplicity so our proof
wouldn’t work.)
L
Proof. Since R is semisimple, it is a direct sum of simple submodules – in other words, we can write R = i∈J Li for
P
some simple left ideals Li . A priori, J could be infinite, but we can write 1 = i∈J ℓi for ℓi zero for all but finitely
P
many i . Thus we can write 1 = i∈I ℓi for a finite index set I.

Lemma 109
If L is a simple left ideal and M is a simple module, then either LM = 0 or M ∼
= L as R-modules.

38
Proof of lemma. Suppose LM ̸= 0. Then pick some m ∈ M such that Lm ̸= 0, and consider the module homomor-
phism φ : L → M which sends x to xm. We have φ(r x) = r φ(x), so φ is a nonzero module homomorphism, and we
claim φ is an isomorphism. Indeed, the kernel of φ is a proper submodule of L, hence must be zero because L is simple,
and then the image of φ is a nonzero submodule of M, hence must be M itself. Thus we’ve shown that M ∼ = L since
φ is both injective and surjective.
(Part of this argument is Schur’s lemma, which we’ll discuss in more detail after this proof.)
L
So returning to 1 = we can show that every irreducible module is isomorphic to one of the Li s where i ∈ I.
i∈I ℓi ,
Indeed, 1 · M = M, meaning that ℓi M ̸= 0 for some i ∈ I, and then we can apply the lemma to show that M ∼ = Li .

Proposition 110 (Schur’s lemma)

If M and N are simple modules over R, then either HomR (M, N) = 0 or M ∼
= N. Furthermore, if they are
isomorphic we may assume M = N, and every nonzero element of EndR (M) is invertible.

Proof. This is the same as before: suppose φ : M → N is any element of HomR (M, N) with φ ̸= 0. Then ker φ is a
proper submodule of M, hence must be zero, and the image of φ is a nonzero submodule of N, hence must be all of
N, and we’ve constructed an isomorphism. In particular, every nonzero element is invertible by taking the inverse of
this isomorphism, which is also a module homomorphism.

Definition 111
Let π : G → GL(V ) be a representation. The character of π, denoted χV or χπ , is defined via χπ (g) = tr(π(g)).

Example 112
For any homomorphism θ : G → C× , we can define a corresponding one-dimensional representation πθ : G →
GL(C) where πθ (g) = θ(g)IdC . So θ is its own character, since one-dimensional matrices can be thought of as
just numbers. Thus we will call a homomorphism θ : G → C× a linear character (and this is terminology often
used in group theory).

Going forward, we’ll mostly assume that F = C (mostly we just need characteristic zero, and later on we will
want algebraically closed). Looking ahead to next class, we may define an inner product on the set of functions on G,
given by
1 X
⟨f1 , f2 ⟩ = f1 (g)f2 (g).
|G|
g∈G

This gives us a finite-dimensional Hilbert space structure, and the big result (which we’ll prove next time) is that if
∼ M2 and 0 otherwise. So in particular
χ1 , χ2 are characters of irreducible modules M1 , M2 , then ⟨χ1 , χ2 ⟩ = 1 if M1 =
by Schur’s lemma this is the dimension of HomG (M1 , M2 ) – then by bilinearity this relation between the inner product
and dimension will hold for general modules.

20 February 27, 2023

Today, we’ll discuss permutation representations, the regular representation, and a bit about characters in preparation
for Schur orthogonality. All representations will be over the base field C today.

39
 First, let G be a finite group acting on a set X (meaning we have a map G × X → X sending (g, x) → g · x
such that (g1 g2 )x = g1 (g2 x)). If we then let FX be the free C-vector space on X, then we have a representation
π : G → GL(FX ) in which we just extend the group action by linearity.
!
X X
π(g) ax · x = ax (gx).
x∈X

This is called the permutation representation.

Proposition 113
The character of the the permutation representation is

χX (g) = number of fixed points of x 7→ gx.

Indeed, we just choose the elements of X as a basis of FX , and then all entries of the matrix will be 1 or 0 (with
a 1 in the (x, y ) entry if gy = x). Then we have 1s on the diagonals corresponding to fixed points.

Example 114  
0 0 1 0
 
0 1 0 0
Let G = S4 act on the set {1, 2, 3, 4}. Then if g is the element (143)(2) in cycle notation, then 
0

 0 0 1
1 0 0 0
is the corresponding matrix, with a trace of 1 because of the sole fixed point 2.

P
This permutation representation is usually not irreducible, since FX has an invariant vector ξ = x∈X x. Then
gξ = ξ for all g, so Cξ is a one-dimensional copy of the trivial representation (which sends everything to the identity).
So now by complete reducibility (using Maschke’s theorem), FX splits as

FX = Cξ ⊕ Fx0 ,

where FX0 is a complementary subspace to ξ. In fact we don’t really need Maschke’s theorem to see this – the subspace
should be nX o
X
FX 0 = ax x : ax = 0 .

Since the character of the trivial representation is just identically 1, the character χ is the number of fixed points minus
1.. It turns out this representation being irreducible is equivalent to the group action being doubly transitive, but we
won’t go into detail for that just yet.

Example 115
If G = S3 acts on {1, 2, 3}, then there are three conjugacy classes, namely id, (123), and (12). We then have
χ0X (id) = 2, χ0X ((123)) = −1, χ0X ((12) = 0, and it turns out this is an irreducible representation (we’ll see why
later).

40
 Definition 116
There is always one permutation representation that we can exhibit for any finite group G, in which we have G
act on itself by left multiplication (so sending (g, x) to gx for any g ∈ G, x ∈ G, where gx is the ordinary group
multiplication). The corresponding permutation representation (where we do not remove the one-dimensional
trivial representation) is then called the regular representation.

Then FG can be identified with C[G] – both of them are the free vector space over the group G. And the
computation of the character ρ for this representation is easy – a fixed point gx = x occurs only if g = id, and thus
id has |G| fixed points and everything else has no fixed points, meaning ρ(id) = |G| and ρ = 0 otherwise.

Proposition 117
For any representation π : G → GL(V ) with corresponding character χ, we have χ(g −1 ) = χ(g) (this is complex
conjugation; in particular, this is why we are using C as the base field)

Proof. Since g is a member of a finite group, we know that g N = 1 for some N. Thus π(g)N = 1, meaning
that all eigenvalues are Nth roots of unity with absolute value 1. The fact that π(g)N = 1 also implies that π(g) is
diagonalizable, so we can choose a basis of eigenvectors for g (v1 , · · · , vd ) with π(g)vi = εi vi . Then π(g)−1 vi = ε−1
i vi ,
but for roots of unity ε−1
i = εi . Since the trace is the sum of the eigenvalues, the desired relation holds for χ(g −1 )
and χ(g).

In a previous homework assignment, we considered the following situation:

Proposition 118
Let V1 , V2 be vector spaces and T1 : V1 → V1 and T2 : V2 → V2 be two linear transformations. If we then let
Ω = HomC (V1 , V2 ), we can define a map τ : Ω → Ω via τ (φ) = T2 ◦ φ ◦ T1 . Then tr(τ ) = tr(T1 )tr(T2 ).

P P
Proof. Say that T1 , T2 have bases x1 , · · · , xn and y1 , · · · , ym respectively, so that T1 xi = j aij xj and T2 xk = ℓ bkℓ yℓ .
With respect to these bases, we can identify Hom(V1 , V2 ) with Matn×m (C), with the basis of elementary matrices Ei,ℓ
(where Ei,ℓ (xj ) = xℓ if i = j and 0 otherwise). Then we can also identify endomorphisms of T1 as End(V1 ) ∼
= Matn (C)
and End(V2 ) ∼
= Matm (C), so that τ maps φ to BφA. Then for any elementary matrix, thought of as a basis element,
we have
X
τ (Eiℓ ) = aij bk,ℓ Ejk ,
j,k
P P P
meaning that tr(τ ) is the sum of the diagonal entries i=j,k=ℓ aij bkℓ = i aii k bkk = tr(A)tr(B) = tr(T1 )tr(T2 ), as
desired.

Also note the following fact:

Proposition 119
Let Ω be a complex vector space and π : G → GL(Ω) be a (not necessarily irreducible) representation. Letting
ΩG be the group of invariants – that is, the set of x ∈ Ω fixed by all π(g) – we have
1 X
dim ΩG = χ(g).
|G|
g∈G

41
 1
π(g). Then p is a projection with image ΩG , since
P
Proof. Define the map p : Ω → Ω via p = |G| g∈G

1 X 1 X
π(g)p(v ) = π(g) π(h)v = π(h)v = p(v )
|G| |G|
h∈G h∈V

1
v = v for any v ∈ ΩG . (So this is very similar to the
P
by relabeling the order of the elements, and since p(v ) = |G|
proof of Maschke’s theorem.) Thus p is idempotent, and Ω splits as ΩG ⊕ (ker p). With respect" to a basis # whose
I ΩG 0
first dim ΩG elements are exactly the elements of ΩG , and the rest are in ker(p), p has matrix in block
0 0
form, and the trace of this matrix is just the dimension of ΩG . But by definition of p, the trace of p will be exactly
1
P
|G| g∈G tr(π(g)), which is the right-hand side.

We will now apply the logic above to a pair of representations.

Theorem 120 (Schur orthogonality)

Let π1 , π2 be two representations with corresponding G-modules V1 , V2 and characters χ1 , χ2 . Then the dimension
1
P
of HomG (V1 , V2 ) (G-module morphisms, not just linear transformations) is |G| g∈G χ1 (g)χ2 (g). (This is an inner
product which we will denote ⟨χ1 , χ2 ⟩.) In particular (by Schur’s lemma) if V1 , V2 are irreducible, then the inner
product is 1 if V1 ∼
= V2 and 0 otherwise.

Proof. Let Ω = HomC (V1 , V2 ) be the space of vector space transformations, and define the representation Π : G →
End(Ω) via
Π(g)φ = π2 (g)φπ1 (g)−1 .

We claim φ is a G-module homomorphism V1 → V2 if and only if φ ∈ ΩG for Π. Indeed, being a G-module homomor-
phism means that φπ1 (g) = π2 (g)φ (applying φ can be done before or after applying π), and rearranging that exactly
1
yields Π(g)φ = φ. Now we know that the dimension of HomG (V1 , V2 ) is the dimension of ΩG , which is |G|
P
χΠ (g).
But now χΠ (g) is the trace of φ 7→ π2 (g)φπ1 (g)−1 , which is χ2 (g)χ1 (g −1 ) = χ2 (g)χ1 (g); plugging in yields the
result.

We’ll see some applications of this fact next time!

21 March 1, 2023
Last time, we showed that for any finite group G, there are finitely many irreducible representations (or equivalently,
finitely many distinct isomorphism classes of C[G]-modules). We can let these representations be V1 , · · · , Vk (cor-
responding to maps πi : G → GL(Vi )). We proved last time that for any two characters χ, θ of representations
π : G → GL(V ) and σ : G → GL(W ), we have
1 X
⟨χ, θ⟩ = χ(g)θ(g) = dim HomC[G] (V, W ).
|G| g

So combining with Schur’s lemma and applying this to our irreducible representations, we find that

⟨χi , χj ⟩ = δij .

Noticing that the character of a representation is called a class function, meaning that it is constant on conjugacy
classes (since the matrix π(gxg −1 ) is conjugate to π(x), the traces χ(gxg −1 ) and χ(x) are the same). We can thus

42
 1
define the function space L2 (G) to be the set of functions on G with the inner product ⟨f1 , f2 ⟩ =
P
|G| x∈G f1 (x)f2 (x),
2 2
and then we have the subspace L (G)class (which is the subset of L (G) which is constant on each conjugacy class).
It will turn out that χi ∈ L2 (G)class are not just orthonormal but also a basis of L2 (G)class , and since dim L2 (G)class is
the number of conjugacy classes of G this means the number of irreducible representations is equal to the number of
conjugacy classes. (And in fact, we can go from this to getting an orthonormal basis on all of L2 (G).)
Last time, we also showed that if G acts on a set X, then the action extends to a representation of G on the free
vector space FX and yields a character χX in which χX (g) is the number of fixed points of g. (And we can construct
the “reduced permutation character” χ0X by removing the invariant subspace spanned by x∈X x, which is often useful
P

too.) When G is acting on itself by left multiplication, we then get the (left) regular representation (which has
character ρ(g) = |G| for g = id and 0 otherwise).

Theorem 121
If χi are the characters of the irreducible representations, then the decomposition of the regular representation is
P
given by ρ = di χi .

In other words, the multiplicity of a representation Vi is just its degree di .

P
Proof. By Maschke’s theorem, we know that C[G] is isomorphic to some direct sum of the Vi s, yielding ρ = ni χi
for some integers ni . Then we can use the orthogonality relations to find that
1 X 1
ni = ⟨ρ, χi ⟩ = ρ(g)χi (g) = · |G| · χi (id),
|G| |G|

since ρ is zero for everything except g = id. But χi (id) is the trace of the identity matrix, which is exactly the dimension
di of the representation Vi .

Remark 122. We could go deeper into what’s going on behind the scenes here, but we’ll just talk about it briefly
– we stated Wedderburn’s theorem earlier without proof, and as a corollary of that (because C[G] is a semisimple
algebra over an algebraically closed field) C[G] is a finite direct sum of matrix rings Matdi (C). Each such matrix ring
is a two-sided ideal with a unique simple module Cdi . We then get a representation of G by injecting G → C[G] and
then restricting to Matdi (C), which acts on Cdi . And this is where the irreducible representation theorems come from
– decomposing the matrix ring into copies of the simple module, we get that Matdi (C) is a direct sum of di copies
of Cdi , denoted L1 ⊕ · · · ⊕ Ldi , where we could make Li the set of matrices only nonzero in the i th column. So the
module of the regular representation, C[G], will indeed be a direct sum of di copies of Vi , summed over all irreducible
representations Vi . And such an interpretation also allows us to deduce Schur orthogonality.
We may think of group representation theory as a nonabelian version of Fourier analysis, and in the coming lectures
we’ll prove two results of that type (though there are many others) which don’t require representation theory in their
statements but for which it is the right language for the proofs.

Theorem 123 (Burnside’s p a q b theorem)

Any group of order p a q b , where p, q are (distinct) primes, is solvable. Thus, any nonabelian simple group has |G|
divisible by at least three primes.

(Here note that p-groups are nilpotent because they have a normal subgroup of index p, so we do have a chain
of normal subgroups.) This result was proved without representation theory methods in the 1960s by Thompson and
Bender, but the original proof remains easier. And this is in fact one of the big steps towards the classification of finite
simple groups.

43
 Theorem 124
Suppose G is a permutation group which acts on X transitively, and assume that only the identity element of G has
more than one fixed point (this is called a Frobenius group). Then the set {g ∈ G : g = id or g has no fixed points}
is a normal subgroup of G.

This result has actually never been proved without representation theory methods.

Example 125 (" # ) " #

a b a b
An example of a Frobenius group is G = : a ∈ F×q , b ∈ Fq acting on Fq via : x 7→ ax + b.
0 1 0 1
Then the normal subgroup in question is the set of translations.

Both of those proofs will come later in the class, and for now we’ll turn to the topic of computing all irreducible
representations and values of their characters. It turns out that there is a computer program called Gap (included
in Sage) that can do this, and there is an atlas of finite simple groups which we can find online as well. The idea is
to let C1 , · · · , Ck be the conjugacy classes of G and to choose a representation gi of each Ci . For example, for S3 , we
can start a table as follows:
1 2 3
id (123) (12)
χ1
χ2
χ3
Here the numbers 1, 2, 3 indicate the number of elements in each conjugacy class, and a representative of
each conjugacy class is written below them. Now there are two linear (one-dimensional) representations, namely
the trivial one (which acts as the identity matrix on every element and thus has χ(g) = 1 for all g) and the sign
permutation Sn → {±1} which sends even permutations to 1 and odd permutations to −1. The point here is that
any homomorphism χ : G → C× is the character of a one-dimensional representation πχ : G → GL(C) sending v to
χ(g)v , with trace again equal to χ. So we can fill in the first two rows of the character table as follows:

1 2 3
id (123) (12)
χ1 = χtriv 1 1 1
χ2 = χsign 1 1 −1
χ3
We’re now wanting to fill in the third row to find the last irreducible representation. Recall that we have the
reduced permutation representation for any Sn , where Sn acts on {1, 2, · · · , n} with χ0X (g) equal to one less than the
number of fixed points of g. Thus we claim that this is the last entry in our table:

1 2 3
id (123) (12)
χ1 = χtriv 1 1 1
χ2 = χsign 1 1 −1
χ3 = χ0{1,2,3} 2 −1 0
To see this, we can prove the following lemma:

44
 Lemma 126
If χ is any character with ⟨χ, χ⟩ = 1, then χ is irreducible.

In particular, we have ⟨χ3 , χ3 ⟩ = 61 (χ(id)2 + 2χ((123))2 + 3χ((12))2 ) (since we have to take the multiplicity of
each conjugacy class into account), which is 16 (4 + 2 · 1 + 3 · 0) = 1. So we have indeed found all of the irreducible
representations of S3 .

Proof. Suppose χ is a character of (π, V ). Writing V as a sum of copies of irreducible representations, such that
P
χ = ni χi for nonnegative integers ni , we get
X X
⟨χ, χ⟩ = ni nj ⟨χi , χj ⟩ = ni2 .

If this is equal to 1, then all but one of the ni s is zero and the remaining one is 1, so χ is indeed just a single copy of
some irreducible representations.

We’ll do another computation next time and talk a bit more about linear characters, and we’ll also prove that
indeed the number of irreducible representations is indeed exactly equal to the number of conjugacy classes.

22 March 3, 2023
Today’s lecture will go over a few more basics of representation theory, so that we can discuss induced representations,
Mackey theory, and the Frobenius theorem next week. We’ve previously shown that the irreducible characters satisfy
⟨χ1 , χ2 ⟩ = 1 if π1 ∼
= π2 and 0 otherwise (by thinking about the dimension of the Hom-space HomG (π1 , π2 ) where G
acts on HomC via gφ = π2 φπ1−1 and using a formula from homework where we computed the dimension of the G-fixed-
points, and by using Schur’s lemma), so we have an orthonormal set. We also showed, by Schur orthogonality relations,
that the regular representation is a direct sum of di = dim(Vi ) copies of the corresponding irreducible representations
Vi , and in fact the simple modules of C[G] can be identified with irreducible representations via Wedderburn’s theorem.
The reason we go through all of this alternate reasoning is that we now want to show that the set of characters
of irreducible representations is actually a basis of L2 (G)class , the set of functions on G constant on each conjugacy
class:

Theorem 127
The number of irreducible representations of G is equal to the number of conjugacy classes of G.

We’ll do a proof here following the textbook:

P
Proof. We can compute the dimension of the center Z(C[G]) in two ways: first of all, ξ = g∈G ag · g is in the center
if and only if ag is constant on conjugacy classes (since conjugating will permute the ag s within a class). So letting
P
the conjugacy classes be C1 , · · · , Ck , we get a basis via the functions κi = x∈Ci x, and thus the dimension is k. On
the other hand, we can compute by writing (by Wedderburn’s theorem) C[G] = ∼ Lnumber of irreducibles Matd (C), and
i=1 i

the center is one-dimensional for each summand and spanned by Idi . So the dimension must also be the number of
irreducible representations.

We’ll now turn to the mechanics of computing character tables: the first step is to find the linear (one-dimensional)
representations. Then after that, we can often find the di s because we know how many there are (the number of

45
 di2 = |G|, which limits the
P
conjugacy classes), we know how many are 1s (from the previous step), and we know
number of total possibilities. Sometimes there will be tricks for producing some irreducible characters, and in fact
induced representations allow us to produce all of them as long as we know the characters of subgroups of G.
But we’ll start from the beginning here – let G ′ be the derived group or commutator subgroup, which is the
subgroup generated by all elements of the form xy x −1 y −1 . Notice that any homomorphism G → A for an abelian
group A contains G ′ in its kernel (since xy x −1 y −1 will always be sent to the identity in A, as everything commutes
there). So any one-dimensional representation χ : G → C× factors through the abelianization G/G ′ .
Now notice that if G and H are two groups with representations (π, V ) and (σ, W ) representations of G, H
respectively, then we have a representation π ⊗ σ : G × H → GL(V ⊗ W ), which has character χV ⊗W : (g, h) 7→
χV (g)χW (h). Furthermore, V ⊗ W will be irreducible if V and W are irreducible – this holds because
1 1 X
⟨χV ⊗W , χV ⊗W ⟩ = |χV (g)|2 |χW (g)|2
|G| |H|
g∈G×H

has the sum factor into |G||H|, so this inner product is indeed 1. And in fact the number of conjugacy classes of
G × H is the product of the number of conjugacy classes of G and of H, and the χV ⊗W s are all orthogonal, so that
gives us all irreducible representations of G × H. So now because G/G ′ is a finite abelian group, it is a product of
cyclic groups, and we know the characters of a cyclic group ⟨σ : σ n = 1⟩ are of the form χa (σ m ) = e 2πiam/n , where
a ∈ {0, 1, · · · , n − 1}. So that tells us that the one-dimensional irreducible representations of G are bijection with the
irreducible representations of G/G ′ .

Example 128
Consider the dihedral group D4 = ⟨x, y : x 4 = y 2 = 1, y xy −1 = x −1 ⟩ of order 8. The conjugacy classes of this
group are {1}, {x, x −1 }, {x 2 }, {y , x 2 y }, {xy , x −1 y }.

Thus we will have five irreducible representations, and one will always be the trivial character. Furthermore, the
commutator subgroup G ′ = Z(D4 ) is generated by x 2 . Indeed, ⟨x 2 ⟩ is normal, and |G/Z(G)| = 4 so G/Z(G) is an
abelian group, meaning G ′ ⊆ Z(G) (here we use that for N ◁ G, we have G/N abelian if and only if G ′ is contained in
N). Furthermore xy x −1 y −1 = x 2 , so ⟨x 2 ⟩ is contained in G ′ as well.
So G/G ′ is generated by the cosets x, y satisfying the relations x 2 = y 2 = (xy )2 = 1. So this group is isomorphic
to the Klein four group Z2 × Z2 , whose characters on 1, x, y , xy are either (1, 1, 1, 1), (1, 1, −1, −1), (1, −1, 1, −1)
or (1, −1, −1, 1). So pulling all of these back to the dihedral group lets us fill in most of the table:

1 2 1 2 2
1 x x2 y xy
χ1 1 1 1 1 1
χ2 1 1 1 −1 −1
χ3 1 −1 1 1 −1
χ4 1 −1 1 −1 1
χ5

There is then one more representation to find: we know that 8 = |D4 | = d12 +d22 +d32 +d42 +d52 , so d5 = 2. And now
there are different ways to proceed: one is to note that the regular representation satisfies χ1 +χ2 +χ3 +χ4 +2χ5 = χreg ,
where χreg is 8 on the identity and 0 everywhere else. That yields the following table:

46
 1 2 1 2 2
1 x x2 y xy
χ1 1 1 1 1 1
χ2 1 1 1 −1 −1
χ3 1 −1 1 1 −1
χ4 1 −1 1 −1 1
χ5 2 0 −2 0 0

Alternatively, we can make use of the fact that D4 acts on the vertices of a square via rotation (x) and reflection
(y ), and thus we can take the reduced permutation character χ0□ to get a representation of degree 3 which takes
values 3, −1, −1, −1, 1 on the correpsonding cnojugacy classes. And now we can calculate its inner product with each
of χ1 , χ2 , χ3 , χ4 , and we will see that ⟨χ0□ , χ4 ⟩ = 1. We thus have χ0□ = χ4 + χ for some other representation χ of
degree 2, and then when we compute ⟨χ, χ⟩ we get 1 and thus this is the last irreducible representation.

23 March 6, 2023
We’ll discuss induced representations, Frobenius groups, and Mackey theory this week. The idea is that for any
subgroup H ⊆ G of a group, we get a functor sending representations of G to representations of H via restriction,
and we wish to show that this has an adjoint functor (actually both a left and right adjoint) sending representations
of H to representations of G. This has a close relationship with character theory – a representation (π, V ) of H with
character χ leads us to a representation (π G , V G ) with character χG characterized as follows. For any G-module W ,
we have
HomG (W, V G ) ∼
= HomH (W, V )

(so this is the “right adjoint” relation – the isomorphism commutes with homomorphisms VV → V ′ and we can form
a commutative square, but we won’t care about that too much in this class), and here all we’re saying is that the
dimensions of the two vector spaces are the same. Indeed, if θ is the character of W , then the isomorphism above
can be formulated as ⟨θ, χG ⟩G = ⟨θ, χ⟩H , and this is equivalent (if we don’t care about the naturality) from the inner
product formula saying that ⟨ψ, χ⟩ is the dimension of the corresponding Hom space between G-modules. We’ll first
prove that we indeed do have this equality:

Proof of the character form. Suppose ψ1 , · · · , ψk are the irreducible characters of G. Then any class function (func-
P
tion on G constant on conjugacy classes) is a linear combination θ = di ψi , and θ is a character if and only if the
di s are all nonnegative integers. So now if we have a character χ of H, we can produce such a character of G by
setting di = ⟨χ, ψi ⟩H (which are nonnegative integers, since both χ and ψi restricted to H are representations and
thus can be written as linear combinations of irreducible representations in H), then letting χG be
P
di ψi . We then
have to prove the boxed relation, which is called Frobenius reciprocity. By linearity, we can just check this for θ = ψi
for some arbitrary i , but then this holds because ⟨χ, ψi ⟩H = di = ⟨χG , ψi ⟩G – first equality by definition and second
equality by orthogonality of the ψi s.

We can also sketch an alternate strategy, which generalizes to Lie groups and gives us a bit more to work with.

Alternate proof sketch. We will work with modules – let (π, V ) be a representation of H with character χ. We define

V G = {space of functions f : G → V |f (hg) = π(h)f (g) ∀h ∈ H}

47
(that is, we ask for a property under left translation), and now define the action π G : G → End(V ) via right translation,
meaning that (π G (g)f )(x) = f (xg). By associativity of the group action this endomorphism does indeed produce
another element of V G , and we need to check that HomG (W, V G ) ∼
= HomH (W, V ). We will only sketch this part: to
define the forward map, take an element Φ ∈ HomG (W, V G ) and map it to φ ∈ HomH (W, V ) via

φ(w ) = Φ(w )(idG )

(since Φ(W ) is a function G → V ). We may check that φ is an H-module homomorphism, since

φ(hw ) = Φ(hw )(idG ) = π G (h)Φ(w )(idG ) = Φ(w )(h) = π(h)Φ(w )(idG ) = π(h)φ(w )

where we use the definition of π G and then the property of invariance under right translation for the two blue equalities.
And then we just need to check that Φ 7→ φ is actually an isomorphism of vector spaces.

Remark 129. It turns out more generally we can also require that f has compact support on G/H, which is “compact
induction.” Then we get a left adjoint in one case and a right adjoint in the other, but these two ideas coincide for
finite groups, which is the case we’re considering here.

Theorem 130
Let χ be a character of H and extend χ to a function χ̇ on G which is equal to χ on H and zero otherwise. Then
the induced character χG is obtained via the formula
1 X
χG (x) = χ̇(g −1 xg).
|H|
g∈G

(If H were normal, then the averaging process still only keeps the nonzero part within H. So the induced character
from a normal subgroup actually still has support inside H.)

Proof. The right-hand side is a class function because we’ve averaged over all conjugates, and thus it is sufficient to
1 −1
show that if τ (x) = |H| χ̇(g xg), then ⟨τ, θ⟩G = ⟨τ, θ⟩H for any character θ of G (since χG is the only class function
satisfying this inner product relation). We have
1 X 1 1 XX
⟨τ, θ⟩G = τ (x)θ(x) = · χ̇(g −1 xg)θ(x).
|G| |G| |H| x∈g
x∈G g∈G

Swapping the order of summation and reparameterizing x 7→ gxg −1 , this simplifies to

1 1 XX
= · χ̇(x)θ(gxg −1 ).
|G| |H| g∈g
x∈G

But since θ is a class function, this is also equal to

1 1 XX
= · χ̇(x)θ(x),
|G| |H| g∈g
x∈G

and now the nonzero contribution comes only from x ∈ H and the sum doesn’t depend on the value of g, so this is
1
P
just |H| x∈H χ̇(x)θ(x) = ⟨χ, θ⟩H , as desired.

This formula can be rewritten – χ̇ is invariant under conjugation by H, so we in fact have χ̇(h−1 xh) = χ̇(x) for
any h ∈ H. So χG is constant on cosets gH, and if we choose a set of left coset representatives g1 , · · · , gn so that

48
 S
G= gi H, then
X X
χG (x) = χ̇(gi−1 xgi ) = χ̇(g −1 xg)
i G/H

(the last equality is just notation for choosing one representative from each coset).
Induced representations are sometimes irreducible (but sometimes not, even if χ is irreducible), and there is in fact
a criterion coming from Mackey theory that tells us when it will be. We’ll do that later on, and for now we’ll just do
some examples:

Example 131
Consider the group G of order 21 generated as ⟨x, y : x 7 = y 3 = 1, y xy −1 = x 2 ⟩. We can construct most of the
character table below, where ω is a third root of unity:

1 3 3 7 7
−1
1 x x y y2
χ1 1 1 1 1 1
χ2 1 1 1 ω ω2
χ3 1 1 1 ω2 ω
χ4 3 ? ? ? ?
χ5 3 ? ? ? ?

Recall that we compute this character table by first noting that G ′ = ⟨xr angle is a normal 7-Sylow subgroup,
∼ Z/3Z is abelian we have G ′ ⊂ ⟨x⟩, while since y xy −1 x −1 = x we have ⟨x⟩ ⊂ G ′ . Then the first
and since G/G ′ =
three characters are linear characters coming from Z3 , and then χ4 and χ5 ’s degrees are computed by noting that
d42 + d52 = 18 can only occur if we have a three-dimensional representation. But because inducing H → G multiplies
the degree by [G : H] (that is, χG (1) = [G : H]χ(1)), we may try to realize χ4 and χ5 as induced representations
from linear characters if we can find a subgroup of index 3. Indeed, the coset representations for G/G ′ are 1, y , y 2 ,
and we can consider the character χ : H → C∗ sending χ(x k ) = ζ k , where ζ is a seventh root of unity. We then have

χ4 = χG (x) = χ̇(x) + χ̇(y xy −1 ) + χ̇(y 2 xy −2 ).

Since H is a normal subgroup of G, this will be supported only within G ′ , and we have χ4 (x) = ζ + ζ 2 + ζ 4 and
χ4 (x −1 ) = ζ −1 + ζ −2 + ζ −4 . Taking the conjugate then gives us χ5 as well.
We cannot always obtain all representations by inducing in this way, but there is a theorem of Brauer that for any
character χ of G there are elementary subgroups (that is, a product of a cyclic group and a p-group) E1 , · · · , Em ,
di ψiG . So we just need to supplement this
P
linear characters ψ1 , · · · , ψm , and integers d1 , · · · , dm , such that χ =
process with an additional step of taking linear combinations.

24 March 8, 2023
We’ll discuss Frobenius groups today, which we introduced a few lectures ago but only briefly:

Definition 132
A Frobenius group is a group G along with a faithful group action of G on X which is transitive, such that no
group element other than the identity fixes more than one point.

49
 Example 133
Suppose X = Fq is a finite field,
" and #G is the set of transformations a 7→ ax + b, which is isomorphic to the matrix
a b
group of elements of the form . Then |G| = q(q − 1), and only the identity element fixes two points (since
0 1
the values of two points determine the line x 7→ ax + b).

Notice in particular that G has two notable subgroups. The first is the isotropy subgroup

H = {g ∈ G : gx0 = x0 }

for some fixed base point x0 – for example, if x0 = 0 then we get the subgroup x 7→ ax (with b = 0), and this
always works for any permutation representation. But there is also a normal subgroup in our case, which is the set of
translations x 7→ x + b (with a = 1), and this is called the Frobenius kernel.

Theorem 134 (Frobenius)

Let G (acting on some set X) be a Frobenius group. Then the set {idG } ∪ {g ∈ G with no fixed points} is a
normal subgroup of G.

As we’ve mentioned previously, this result has always required representation theory to prove. The hard part here is
proving that this is actually a subgroup – once we do that it’ll be normal, since the subgroups Hx = {g ∈ G : gx = x}
(sometimes called the Frobenius complements) are all conjugate because the action is transitive, and if γx = y then
γHy γ −1 = Hx .
Fix x0 = X and H = Hx0 (with the definition above), and let
 
[
K = {idG } ∪ G − γHγ −1  .
γ∈G

Then being in this set and not the identity means we have no fixed points. So if γ ̸∈ H, then γHγ −1 ∩ H = idG (since
γx0 ̸= x0 , and γHγ −1 = Hγx0 , and Hγx0 ∩ Hx0 = 1 by definition of a Frobenius group).

Proposition 135
Let G be a group, H a nontrivial subgroup of G, and assume that if x ∈ G, then either x ∈ H or xHx −1 ∩H = {idG }.
Then G is a Frobenius group.

Proof. We want to construct a group action – let X = G/H be the set of cosets gH of G. Then the stabilizer of
xH is xHx −1 , so by our hypothesis no element fixes both 1 · H and x · H; it’s easy to go from this to seeing that no
element has two fixed points.

Before we prove Frobenius’s theorem, we can extrapolate and deduce some implications. Then the set of elements
K = {idG } ∪ x∈G xHx −1 satisfies HK = G and H ∩ K = idG , so G is in fact a semidirect product. We then get
S

maps H → G → G/K, given by inclusion and then projection, such that the composite map is an isomorphism. So in
particular, if (π, V ) is a representation of H (in fact this works for general homomorphisms as well), we can extend
it to a representation of G in the following way: use the isomorphism to get a representation on G/K, and then pull
it back to G. This way of representation is always valid because the composite map G → G/K = ∼ H is the identity
map, so it will agree with π. So the strategy for Frobenius’s theorem will in fact be to prove this fact, showing that a
representation can be extended.

50
 Lemma 136
Let G be a Frobenius group acting on X, x0 some element of X, and H = Hx0 . Then any representation of H can
be extended to a representation of G.

P
Proof. Induction is part of the bag of tricks here – we know that if χ is a character of H, then we can write χ = di ψi ,
where ψi are the irreducible characters of H and di are nonnegative integers. We can relax this condition and say
P
that a class function χ on H is a generalized character if χ = di ψi , now allowing di to be any integers (so in
particular it is the difference of two ordinary characters). For any such character χ of H, we know from last lecture that
1
χG (g) = |H| χ̇(xgx −1 ), with the Frobenius reciprocity relation ⟨χG , ψ⟩ = ⟨χ, ψ⟩H for any ψ; this relation continues
P

to hold for generalized characters by linearity.

So this allows us to also induce generalized characters in the same way as ordinary characters as follows. Define
d = χ(idG ), so that χ = χ0 + d · 1H (where 1H is the trivial character sending everything to 1). Then χ0 (idG ) = 0,
χ0 is a generalized character, and then we can define χ̂ = χG
0 + d · 1G (getting the induced representation and then
adding back d copies of the trivial character). We must show that χ̂ agrees with χ on H; indeed for x = 1 both sides
are equal to d (and this is why we had to subtract off d1H in the first place), since we can recall that the formula for
the induced representation is
1 X
χ̂G
0 (x) = χ̇0 (gxg −1 )
|H|
g∈G

so if x = idG then all terms on the right are 0. Now if we suppose x ∈ H is not the identity. We know that for any
g ̸∈ H, we have H ∩ gHg −1 = idG , so in fact
1 X 1 X
χ̂G
0 (x) = χ̇0 (gxg −1 ) = χ̇0 (gxg −1 ) = χ0 (x).
|H| |H|
g∈G g∈H

Thus χ0 = χ̂G
0 , and adding d to both sides yields χ̂ = χ for all non-identity x ∈ H.
To finish the proof, we must now show that ⟨χ̂, ψ̂⟩G = ⟨χ, ψ⟩H for any generalized character χ of H. (In other
words, the mapping is an isometry under the inner product.) We may write χ = χ0 + d · 1H and ψ = ψ0 + c · 1H
(where d = χ(idG ) and c = ψ(idG )). By linearity, it suffices to handle the two cases where χ0 = 0 and where d = 0
separately, and the same two cases for ψ. In the case where χ0 , ψ0 = 0, we have χ = d, ψ = c everywhere, which
leads to χ̂ = d, ψ̂ everywhere, and thus both sides are just dc. Otherwise, we may assume without loss of generality
that χ = χ0 (and d = 0). Then because χ(idG ) = 0, we have by Frobenius reciprocity that

⟨χ̂, ψ̂⟩G = ⟨χ, ψ̂⟩H = ⟨χ, ψ⟩H

(the last part because ψ̂ agrees with ψ on H), which proves the result. (The key point here is that χ̂ is not the induced
representation of χ, so we need to get rid of the d · 1H part.)

Corollary 137
Any character χ of H can be extended to a character of G.

P
Proof. Without loss of generality, we may assume that χ is irreducible, so that χ̂ satisfies ⟨χ̂, χ̂⟩ = 1. Write χ̂ = d i θi
P 2
as a sum of irreducible representations of G; then di = 1 so in fact χ̂ is either an irreducible representation of G
or its negative. Then χ(1) = χ̂(1) > 0, so we rule out the latter possibility.

The result of Frobenius’ theorem now follows immediately:

51
Proof of Theorem 134. Take a faithful representation of H, such as the regular representation, and extend it to a
representation of G. Its kernel K is a normal subgroup, and it is easy to see that K is in fact {1}∪{G − xHx −1 }.
S

25 March 10, 2023

Mackey theory is a powerful method which makes use of double cosets to get information about induced representations.
For example, if G is a finite group, H is a subgroup, and π is a representation of H, we may want to know when
the induced representation π G = IndG
H (π) is irreducible (so we can produce irreducible representations). Or we may
be interested in seeing what representations occur if we induce a representation from H1 and then restrict to some
subgroup H2 .

Example 138
Let π be a representation of Sn and let λ = (λ1 , · · · , λk ) be a partition of n. (For example, the partitions of
5 are (5), (4, 1), (3, 2), (3, 1, 1), (2, 2, 1), and (1, 1, 1, 1, 1).) Then we can consider the Young subgroup Sλ =
Sλ1 × · · · × Sλk (that is, permuting only among the first λ1 elements, the next λ2 , and so on).

One result that Mackey theory is then able to tell us (along with some combinatorics) is that if we start from the
trivial representation on Sλ and induce a representation on Sn , and we start from the sign representation on Sµ and
also induce a representation on Sn , then Hom(IndSSnλ (1), IndSSnµ (sgn)) is one-dimensional if λ, µ are conjugate partitions
(meaning that we draw the Young diagrams for the two partitions and they are reflections of each other over the
diagonal, for example (3, 2) and (2, 2, 1)). In particular, this means that the image of this Hom space is an irreducible
representation of Sn , and thus all irreducible representations are constructed this way.
G G
More generally, we want to know how to compute Hom(πH 1
, πH 2
) for representations π1 , π2 of subgroups H1 , H2
of G. Let χ1 , χ2 be the corresponding characters – the dimension of the Hom space is then ⟨χG G
1 , χ2 ⟩G , and we will
need to talk about double cosets to answer this question.

Definition 139
A double coset of G is a subset of the form H2 γH1 , where H1 , H2 are subgroups of G.

For fixed H1 , H2 , any two double cosets are either equal or disjoint, meaning that we may write G = ki=1 H2 γi H1
S
S
for double coset representatives γi . (Alternatively, we can use the notation G = γ∈H2 \G/H1 H2 γH1 , where again we
sum over one representative γ per equivalence class.) Now for any γ ∈ G we may define

Hγ = H2 ∩ γH1 γ −1 ,

which is a subgroup of H2 and also conjugate to a subgroup of H1 (since γ −1 Hγ γ ⊆ H). So Hγ can be thought of as
a subgroup of both H1 and H2 (just with an extra conjugation in the former case).

Lemma 140
S
Let γ1 , · · · , γk be a set of double coset representatives for G, so that G = H2 γi H1 . For each γ = γi , let
S
δi1 , δi2 , · · · be a set of coset representatives for H2 /Hγi . Then we have that G = δij γi H1 as a disjoint union.

In other words, we’ve obtained a set of coset representatives for G/H1 : we can alternatively write
[ [
G= δγH1 .
γ∈H2 \G/H1 δ∈H2 /Hγ

52
Proof. Every element of G can be written as h2 γh1 for some h1 ∈ H1 , h2 ∈ H2 , and γ ∈ H2 \ G/H1 . We want to ask
when h2 γh1 = h2′ γ ′ h1 ; we must have γ = γ ′ , but we have some flexibility for h2 , since h2 γH1 = h2′ γH1 implies that
h2−1 h2′ γH1 = γH1 and thus γ −1 h2−1 h2′ γ ∈ H1 =⇒ h2−1 h2′ ∈ γH1 γ −1 . Since h2 , h2′ ∈ H2 , that means h2−1 h2′ ∈ Hγ . So
h2 , h2 ” produce the same coset h2 γH1 if and only if h2 , h2′ are in the same coset of H2 /Hγ . (And this reasoning is all
reversible.)

Theorem 141 (Mackey theory, version 1)

We have
X
⟨χG G
1 , χ2 ⟩G = ⟨ γ χ1 , χ2 ⟩Hγ ,
γ∈H2 \G/H1
γ
where χ1 corresponds to a representation π1γ and is the character given by

γ
χ1 (x) = χ1 (γ −1 xγ).

Proof. By Frobenius reciprocity, we know that ⟨χG G G

1 , χ2 ⟩G = ⟨χ1 , χ2 ⟩H2 . By the formula for the induced representation,
G −1
P
we know that χ1 (x) = G/H1 χ̇1 (γ xγ), so the left-hand side of the theorem statement is (plugging into Frobenius
reciprocity)
1 X G 1 X X
χ1 (x)χ2 (x) = χ̇1 (τ −1 xτ )χ2 (x).
|H2 | |H2 |
x∈H2 x∈H2 τ ∈G/H1

If we now split into double coset representatives, this becomes

1 X X X
= χ̇1 (γ −1 δ −1 δγ)χ2 (x).
|H2 |
x∈H2 γ∈H2 \G/H1 δ∈H2 /Hγ

But now if we swap the order of sums, the inner sum becomes a sum over x and x is mapped to δxδ −1 , and
χ2 (δxδ −1 ) = χ2 (x) because χ2 is a class function. So we get rid of dependence on one of the variables δ, and we’re
just left with
1 X X
= [H2 : Hγ ] χ̇1 (γ −1 xγ)χ2 (x).
|H2 |
γ∈H2 \G/H1 x∈H2

But now this sum is zero unless γ −1 xγ is in H1 , so in fact x ∈ H2 ∩ γH1 γ −1 = Hγ . So now we can remove the dot
on χ1 , and we’re left with
X 1 X
= χ1 (γ −1 xγ)χ2 (x),
|Hγ |
γ∈H2 \G/H1 x∈Hγ

which is exactly the expression on the right that we were looking for.

In the special case where H1 = H2 is a normal subgroup N of G, we can prove directly that
X
χG
1 (x) =
γ
χ(x)
γ∈G/N

for any x ∈ N, and that χG

1 (x) = 0 if x ̸∈ N. (Note that if N is normal, then N\G/N = N.) This doesn’t require
Mackey theory to prove:

Proposition 142
If N is a normal subgroup of G, then G/N acts on representations of N by conjugation. Let χ be an irreducible
character of N. Then χG is irreducible if and only if χ is not fixed by any non-identity coset ω ∈ G/N.

53
 In particular, γ χ(g) = χ(γ −1 gγ) only depends on the coset γN, since χ is a class function. So to prove this result,
by Frobenius reciprocity we have * +
X
G G G γ
⟨χ , χ ⟩G = ⟨χ , χ⟩N = χ ,χ ,
γ∈G/N

which is the number of γ ∈ G/N such that γ χ = χ. So if χ is not fixed by any nontrivial coset, then this calculation
shows that ⟨χG , χG ⟩ = 1.
We’ll do some more examples next time!

26 March 13, 2023

Last time, we started discussing Mackey theory – one version of this takes two subgroups H1 , H2 of G with represen-
tations (π1 , V1 ) and (π2 , V2 ) of those subgroups. We can then consider the homomorphisms HomG (V1G , V2G ), and we
showed using characters that we can compute
X
dim HomG (V1G , V2G ) = ⟨χG G
1 , χ2 ⟩ = ⟨ γ χ1 , χ2 ⟩Hγ ,
γ∈H2 \G/H1

where Hγ = H2 ∩ γH1 γ −1 and where the “twisted” character χ1 is given by γ χ1 (x) = χ1 (γ −1 xγ). (So here we are
γ
F
summing over a set of double coset representatives R, such that G = γ∈R H2 γH1 . In particular, we can change γ by
γ
an element of H1 on the right or by an element of H2 on the left, and the construction of χ1 and Hγ does depend
on the choice of representative up to H2 but we will end up getting the same values of inner products because we end
up in a conjugate subgroup.)
We’ve done a few examples in lecture where the subgroups have been normal, making the computations easier,
but in our homework we saw a more interesting case:

Example 143
Let G = GL(2, F ) for some finite field F = Fq . Then |G| = (q 2 − 1)(q 2 − q), and we may let B be the
Borel subgroup of upper-triangular matrices with nonzero (invertible) entries on the diagonal. We then have
|B| = (q − 1)2 q and [G : B] = q + 1. So now if χ1 , χ2 are two linear characters of F × , we can define the linear
character χ : B → C∗ via " #!
y1 x
χ = χ1 (y1 )χ2 (y2 ).
0 y2
We can then induce χ from B to G, and it turns out that if χ1 , χ2 are distinct then the induced representation
IndG
B (χ) is irreducible. (This turns out to compute about half of the irreducible representatives for GL(2, F ).

This was on our homework, but we’ll discuss it in some more detail here. We’ll compute more generally by letting
ψ1 , ψ2 be two other linear characters of F× and similarly define ψ : B → C∗ in the same way as χ. We then want to
compute ⟨χG , ψ G ⟩. The double cosets of G are given by the Bruhat decomposition
" #
0 1
G = B · I · B ∪ B · ω · B, ω = .
1 0

(The same argument works for GL(n, F ), just with n! double cosets corresponding to the n! elements of the symmetric
group.) Then Bγ = B ∩γBγ −1 , where γ is chosen to be either the identity matrix or ω, so Bγ is either B itself (for I) or
the diagonal subgroup (for ω). So to compute using Mackey theory we need to compute two terms ⟨χ, ψ⟩B +⟨ ω χ, ψT ⟩.

54
Since these are one-dimensional representations, each term will be 1 if the two characters agree and 0 otherwise –
the former is only 1 if χ1 = ψ1 , χ2 = ψ2 , and the latter is only 1 if χ1 = ψ2 , χ2 = ψ1 . So if χ1 ̸= χ2 but ψ = χ we
have ⟨ψ, ψ⟩G = 1 and thus χ is irreducible. On the other hand, if χ1 ̸= χ2 then χ is isomorphic to the corresponding
representation where we swap χ1 , χ2 .
All of this also extends (with nuances) to irreducible infinite-dimensional representations of GL(2, F ) for any
nonarchimedean local field. (We can search up “principal series representations” for more, and this is relevant to
automorphic forms.)
There is actually another theorem of Mackey theory which we can use for other computations, which involves
computing the induced representation of some representative (π, V ) of H1 and then restricting to another subgroup
of H2 :

Theorem 144 (Mackey theory, version 2)

Take Hγ = H2 ∩ γH1 γ −1 as before. Then we have

∼
M
IndG
H1 (V )|H2 = IndH2 γ
Hγ ( π),
γ∈H2 \G/H1

where γ π is defined via γ π(h) = π(γ −1 hγ) (where in particular γ −1 hγ ∈ H1 so this makes sense).

The idea is that the sum over double cosets is like embedding H1 into G in different ways. So instead if inducing
and then restricting, we can restrict to Hγ and then induce, but we must take a direct sum over different possibilities.

Proof. We’ll show that we have equality of characters

X
χG |H2 = = IndH 2 γ
Hγ ( χ),
γ∈H2 \G/H1

and to do this we can take the inner product with an arbitrary character ψ of H2 . We have by Frobenius reciprocity
that
⟨χG |H2 , ψ⟩H2 = ⟨χG , ψ G ⟩G ,

and now by version 1 of Mackey theory this simplifies to

X
= ⟨ γ χ, ψ⟩Hγ ,
γ∈H2 \G/H1

γ
χG , ψ⟩G , which is indeed the character of the right-hand
P
but by Frobenius reciprocity again this simplifies to γ⟨
side. Thus the two representatives are indeed isomorphic.

We’ll do one more example using the earlier form of Mackey theory to show some interesting computations:

Example 145
Let G = S5 , H1 = S3 × S2 , and H2 = S2 × S2 × S1 (so H1 only permutes among the first three and last two
elements separately, and similarly for H2 ). We claim that

HomS 5 IndG G


H1 (1), IndH2 (sgn)

is one-dimensional.

55
 We must find a set of double coset representatives forH2 \G/H1 , and we
 will do so by thinking of permutations as
0 0 1
 
1 0 0 
 
matrices. For example, h1 = (123)(45) corresponds to 0 1 0 , with zeros also in the remaining entries.
 
 

 0 1
1 0
 
1 1 1 2 2
 
1 1 1 2 2
 
If we now think of γ as being in blocks labeled by 3 3 3 4 4 (where 1 through 6 label groups, not matrix
 
 
3 3 3 4 4
 
5 5 5 6 6
entries), we see that we can permute entries in those six blocks using S3 × S2 on the right (using column operations)
and S2 × S2 × S1 on the left (using row operations). So the idea is that we can rearrange columns and rows within
blocks, but the blocks themselves must stay the same. In other words, each block has entries of 0 and 1, and the rank
gi of block i must stay constant.
Additionally, we also see that g1 + g2 = 2 (because we must have one nonzero entry per row of the big matrix),
and similarly g3 + g4 = 2, g5 + g6 = 1. By the same reasoning on columns, g1 + g3 + g5 = 3, g2 + g4 + g6 = 2. And we
can check that those are the only possible constraints using some combinatorics. So we can now basically compute
all possibilities: we have
           
g1 g2  2
 0 2 0 1 1 1 1 0 2 

           
g3
 g4 
 ∈ 0

 ,
2 1   ,
1 2  ,
0 1  ,
1 2
  0 ,


 

g5 g6  1 0 0 1 0 1 1 0 1 0 
 
1 0 0 00
 
0 1 0 0 0
 
where for example the first can correspond to the coset representative 0 0 0 1 0, which represents the
 
 
0 0 0 0 1
 
0 0 1 0 0
permutation (354). But now we claim we will have ⟨ γ 1, sgn⟩Hγ = 0 as long as Hγ contains a transposition. Indeed,
γ
χ1 = 1 on Hγ is orthogonal to (χ2 )Hγ = sgn, and we will check that only the blue one has no transpositions. Thus
the Hom space is indeed one-dimensional. And the corresponding generalization (working exactly the same way) is
that for any partitions λ, µ of n, we can compute Hom(IndSSnλ (1), IndSSnµ (sgn)) (the idea is that the matrix of gi s must
only contain 0s and 1s); in particular if λ is the conjugate partition of µ, this will always be one-dimensional with
irreducible image. In other words, partitions characterize irreducible representations of Sn .

27 March 15, 2023

Our first topic for today is to prove that the degree of an irreducible representation of G divides the order of the group
|G|. First, recall a fact that we have proved on our homework, which is that for any character χ of an irreducible
|C|χ(g)
representation (π, V ), then for any g ∈ G in some conjugacy class C, we have an algebraic integer. Indeed, this
χ(1)
P
can be proved by considering the center of Z[G], which has a basis of elements of the form Ki = g∈Ci g (one for each
conjugacy class). Then we get a representation of C[G] by extending π by linearity; since Ki is in the center of C[G],
it acts as a scalar by Schur’s lemma (since it commutes with the action of G). We then have π(Ki ) = ωπ (Ki )IdV for

56
some ωπ (Ki ) ∈ C.
P
Each ωπ (Ki ) is an algebraic integer, since we have Ki Kj = k aijk Kk for some coefficients aijk ∈ Z and thus we
P
also have ωπ (Ki )ωπ (Kj ) = aijk ωπ (Kk ). That means the Ki s generate a ring which is finitely generated over Z,
so by our usual criterion those elements must be algebraic integers. Remembering that having a faithful Z[α]-module
finitely generated as a Z-module proves that α is integral, we see that M = Z[ωπ (Ki )] is finitely generated, so each
ωπ Ki is an algebraic integer. But
X
ωπ (Ki ) = π(gi ) = ωπ (ki )IdV ,
g∈Ci

and taking traces on both sides yields |Ci |χ(g) = ωπ (Ki )χ(1), which yields the result that we want.
So now we can use this:

Proposition 146
For any irreducible representation of G with character χ, we have χ(1)||G|.

Proof. We have
X k
X
1 = ⟨χ, χ⟩ =⇒ |G| = χ(g)χ(g) = |Ci |χ(g)χ(g),
g∈G i=1

where we have broken up the sum into conjugacy classes. But this means that
k
|G| X |Ci |χ(g)
= · χ(gi ).
χ(1) χ(1)
i=1

Now the left-hand side is rational, and the right-hand side is an algebraic integer (since χ(gi ) is a sum of roots of
unity, each of which is an algebraic integer, and the other term we’ve also proved is an algebraic integer). Since the
only rational numbers that are algebraic integers are integers, this proves the claim.

In general, for an irreducible character χ of g corresponding to a representation (π, V ), we can define

Z(χ) = {g ∈ G : π(g) is a scalar endomorphism of V }.

The idea is that we can recognize whether g ∈ Z(χ) just by looking at the character table:

Proposition 147
With the notation above, we have g ∈ Z(χ) if and only if |χ(g)| = χ(1).

Proof. Since π(g) is diagonalizable, the matrix is similar to a diagonal matrix with entries (ε1 , · · · , εd ) with all |εi | = 1.
|αi | if and only if the αi are all proportional over R+ , which can
P P
Then the triangle inequality says that | αi | =
only occur if they are all equal because they are of unit length.

Also, notice that Z(χ) is a normal subgroup (since conjugating g ∈ Z(χ) corresponds to conjugating a scalar
matrix, which just recovers the original scalar matrix again) – this can be useful in various situations.

Theorem 148
Let C be a conjugacy class of G and χ be an irreducible character, and suppose that χ(1) and |C| are coprime.
Then either g ∈ Z(χ) or χ(g) = 0.

(This result is particularly useful if we know that G is a simple group.)

57
 χ(g)
Proof. By Bezout’s lemma, we can find a, b ∈ Z such that aχ(1) + b|C| = 1. Multiplying this by χ(1) , we see that

|Cχ(g) χ(g)
aχ(g) + b = .
χ(1) χ(1)
χ(g)
Now both terms on the left-hand side are algebraic integers, so χ(1) is also an algebraic integer. In particular, it is an
χ(g) ε1 +···+εd
algebraic integer of a particular type: the eigenvalues of π(g) are roots of unity ε1 , · · · , εd , so χ(1) = d . Any
Galois conjugate of this is also of this form for some other roots of unity ε′1 , · · · , ε′d , meaning all Galois conjugates
χ(g)
have absolute value at most 1. Thus the norm of χ(1) in Q(χ(g)) over Q has absolute value at most 1, and the norm
must lie in Z. So if the norm has absolute value 1, then |χ(g)| = χ(1) and thus (by Proposition 147) it is in Z(χ);
otherwise the norm is 0 and thus χ(g) = 0.

Theorem 149
Let G be a nonabelian simple group. Then the only way for a conjugacy class to have order a power of p is if
C = {idG }.

(Both this result and the previous one are likely due to Burnside.) In particular, this means that the non-identity
conjugacy classes of a simple group each have order divisible by at least two primes – indeed, for G = A5 , there are
conjugacy classes of size 1 (from the identity), 20 (from 3-cycles), 15 (from pairs of transpositions), and two of size
12 (from (12345) and (13524)).

Proof. Consider the regular representation of G, which has di copies of each irreducible representation χ. In particular,
this means that 
X |G| g = idG ,
χ(1)χ(g) =
0 otherwise.
χ

Suppose the conjugacy class Cg containing g has order a power of p. If χ is not the trivial representation, then Z(χ) is
trivial (because it is a normal subgroup that is not all of G, since π is faithful and G is nonabelian). By Theorem 148,
this means that either χ(1) is divisible by p or χ(g) = 0 (since in the latter case χ(1) and |Cg | are relatively prime).
So now for any fixed g ̸= idG , we can take the equation from above and write it as
X
0=1+ χ(g)χ(1).
χ nontrivial

But now every term in the sum has either χ(g) = 0 or p|χ(1), so it is divisible by p. Thus p divides 1, a contradiction.

Theorem 150 (Burnside’s p a q b theorem (1904))

If |G| = p a q b , then G is not a nonabelian simple group. (In particular, this means G must be solvable.)

Proof. Let P be a p-Sylow subgroup of G. Then P has nontrivial center, since P can be written as a union of Z(G)
and noncentral conjugacy classes. For any x ∈ Z(P ), the index of the centralizer |Cx | = [G : C(x)] contains P , so
[G : C(x)] must divide q b . This is a contradiction to Theorem 149.

It turns out that z ∈ Z(G) if and only if z ∈ Z(χ), so we can look at the character table to deduce which elements
are in the center. And we can also similarly pick out the derived group just by looking at characters, so the moral is
that those characters do encode a lot of information about the group.

58
28 March 17, 2023
In this last lecture, we’ll see a bit of how commutative algebra and representation theory can be tied together – we’ll
discuss Galois group representations, stating some results without proof. We’ll start with a bit more finite group
theory:

Definition 151
A finite group E is elementary if it is the direct product of a p-group and an abelian group.

Proposition 152
If E is elementary, then every irreducible representation of E is induced from a one-dimensional representation by
some subgroup (called an M-group).

The M in M-group stands for “monomial” – it means that we can choose a basis of the representation so that the
group elements are represented by monomial matrices.

Theorem 153 (Brauer, theorem A)

Any generalized character of a finite group G is a Z-linear combination of characters induced from elementary
subgroups.

In particular, by Proposition 152, we can always induce from one-dimensional characters of elementary subgroups.
And this result has applications to Artin L-functions (which we’ll see later on).

Theorem 154 (Brauer, theorem B)

A class function on G is a generalized character if and only if χ|E is a generalized character for every elementary
subgroup E of G.

This result has different-looking applications than the previous version – in particular, it can be used to compute
the irreducible representations of GLn . Brauer proved these results early on, and then Brauer and Tate came up with
a clever proof afterward – the idea is to consider the ring X of generalized characters on G and let U be the set of all
class functions χ with χ|E a generalized character for every elementary subgroup E and V be the set of all generalized
characters that are Z-linear combinations of characters induced by elementary subgroups. We have U ⊇ X ⊇ V –
theorem A then claims that X = V and theorem B then claims that X = U. But V is an ideal in U, so if we can prove
that 1 ∈ V then V = U and both theorems are implied.
We’ll now turn to some commutative algebra, considering a familiar situation: let E/F be a finite separable
extension of degree n, A ⊂ F a Dedekind domain, and B the integral closure of A in E (which is also a Dedekind
domain). If we let ℘ be a maximal ideal of A, then we can factor ℘B into prime ideals in B as

℘B = Pe11 · · · Per r ,

where the Pi are the prime ideals of B above A and we call ei the ramification index of Pi over ℘. If we then define
the residue class degree fi = [(B/Pi ) : (A/℘)], it turns out (we can consult Lang’s algebraic number theory book)
X
that ei fi = n . In the case where B = A[α] (for example if A = Z and E is a cyclotomic field) this is easy to

59
prove, though some combination of localization and working with completions gives us the result in general as well.
The idea is to let f ∈ A[x] be an irreducible polynomial with root α; we can factor its image f ∈ A/℘[x] into irreducible
polynomials giei , where the ei s are again the ramification indices from above and where deg(gi ) = fi . And the idea
Q

is that if we look at a root βi of gi in some integral extension of A/℘, then we can consider the homomorphisms
B : A[α] → A/℘ sending α to βi , which has kernel Pi . So we can use this to get a handle on the decomposition of
℘B as a product of primes.

Example 155
Let A = Z and B = A[α] for some root α of x 3 − 2. Then if p = (5), then (x 3 − 2) factors mod 5 into
(x + 2)(x 2 + 3x + 4), so P1 is the kernel of the homomorphism sending α to −2 (with f1 = 1) and P2 is the kernel
of the homomorphism sending α to x 2 + 3x + 4 in F52 (with f2 = 2). Then indeed f1 + f2 = 3 with e1 , e2 = 1 (no
ramification).

Proposition 156
(αi − αj )2 ∈ A. Since
Q
We will have all ei = 1 unless f has a multiple root, meaning ℘ divides the discriminant
this discriminant only has a finite number of prime factors, this means there are always only finitely many ℘ that
ramify, and other than that we just have ei = 1 (“unramified primes”).

It turns out that having one extension of degree 1 and one extension of degree 2, as in our example above, can’t
happen if we have a Galois extension:

Theorem 157
Suppose E/F is Galois. Then all ei are equal and all fi are equal.

The idea is that we can reduce to the case where B = A[α], and we notice that all roots of f will be in B. Since
f is irreducible, the Galois group of E/F acts transitively on the roots of f . Thus the action is also transitive on the
Pi s, and thus the residue class degrees and ramification indices must be all equal.

Example 158
Modifying the example above (with A = Q and B = Q(α)), if we adjoin all cube roots of 2 instead of just α, we
get C = Q(α, e 2πi/3 ) (which is Galois over Q); we have ℘ = P1 P2 as before (of f1 = 1, 2 respectively). But then
P1 lifts to P′1 in C with degree 2, and P2 lifts to P′2 , P′3 each with degree 1. Thus 5C = P′1 P′2 P′3 with all ei = 1
and all fi = 2.

So if E/F is Galois and we assume all ei = 1 and all fi = f , we have

℘B = P1 · · · Pr , r f = [E : F ] = n.

As we vary the prime ℘, we can get various behavior (for example, we may have f = 1 and the prime splits completely,
or r = 1 and the prime stays prime). But if we let DP1 ⊂ Gal(E/F ) be the stabilizer of P1 = P, also known as
the decomposition group, we see that DP1 is conjugate to DPi for all i , meaning that the decomposition group is
determined up to conjugacy. (In particular, this means it is unique for an abelian extension.)
If we now assume that A/℘ is finite, meaning we also have B/P finite with [(B/P) : (A/℘)] = F , then the Galois
group Gal((B/P)/(A/℘)) cyclic and generated by the Frobenius element σ 7→ σ q , where q = |A/℘| is the size of the

60
finite field (and thus B/P = q f ). We then get a homomorphism D → Gal((B/P)/(A/℘)) = Gal(Fq r /Fq ), which is
surjective. Furthermore, having a nontrivial kernel comes from a ramification index not being 1, which we’ve assumed
to not be the case, so we in fact have an isomorphism D ∼ = Gal(Fq r /Fq ).

Example 159
Suppose A = Z and E = Q(ζN ), where ζN is a primitive Nth root unity. Then it can be shown that B = A[ζ],
and the generator of the Galois group is p any prime not dividing N (that is, any unramified prime).

Then because Gal(E/F ) is abelian and the decomposition group and Frobenius element are independent of the
choice of prime above ℘ = (p), we have φ℘ ∈ Gal(E/F ) given by φp (ζ) = ζ p . So the Galois group is isomorphic to
(Z/nZ)× (containing φ(N) elements), f is the order of p in (Z/NZ)× , and r = nf . And in the general case where we
don’t assume that the Galois group is abelian (but still avoid ramification), the Frobenius element in the Galois group
is only determined up to conjugacy by φ(x) = x q mod Pi .
So in short, every prime of the ground ring yields a conjugacy class of elements in the Galois group. And now we
can see an application of Brauer’s theorem:

Definition 160
Let F be a number field (that is, a finite extension of Q), E/F a finite Galois extension, and A be the integral
closure of Z in F . Let G = Gal(E/F ) and π : G → GL(V ) a complex representation of G. The Artin L-function
is defined via
χ(φ℘ ) −1
Y  
L(s, π) = 1− ,
N℘s
℘ primes of A

where N℘ = |A/℘|. (The definition should be modified at ramified primes, but we’re ignoring that for now.)

In the simplest case where E = F = Q, this is just the usual Riemann zeta function. It was believed by Artin that
just like the ordinary Riemann zeta function, we have functional equations and analytic continuations. Cases of this
did eventually get proved, but the first important result was due to Brauer:

Theorem 161 (Brauer)

L(s, π) is meromorphic.

In the cyclotomic case, this works out rather elegantly, using the fact that 1 − x f =
Q
ε (1 − εx) where we take the
product over the f th roots of 1. And being entire is a consequence of the Langlands conjecture, in the case where
V G = 0 – the idea is that this should agree with the L-function of an automorphic form.
To sketch the idea here, the case where π is one-dimensional is known by work of Dirichlet and Hecke of 1919
(expressing in an integral form and then using Poisson summation). Then the idea is that induction does not change
the Artin L-function (that is, if we have some subgroup H of G, which by the Galois correspondence corresponds to
an intermediate field K, then L(s, π) over K is the same as L(s, π G ) over F ). So now we can use Brauer’s theorem
to write χ as a linear combination of induced representations from one-dimensional representations, which proves the
result.

61