Professional Documents
Culture Documents
Andrew Kobin
Contents
2 Integral Dependence 16
2.1 Integral Extensions of Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2 Integrality and Field Extensions . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.3 Integrality, Ideals and Localization . . . . . . . . . . . . . . . . . . . . . . . 25
2.4 Normalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.5 Valuation Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.6 Dimension and Transcendence Degree . . . . . . . . . . . . . . . . . . . . . . 38
i
Contents Contents
10 Homology 176
10.1 Chain Complexes and Homology . . . . . . . . . . . . . . . . . . . . . . . . . 177
10.2 Derived Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
10.3 Tor and Ext . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
10.4 Universal Coefficient Theorems . . . . . . . . . . . . . . . . . . . . . . . . . 202
ii
Contents Contents
16 Cohomology 373
16.1 Direct and Inverse Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374
16.2 Acyclic Sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383
16.3 Noetherian Scheme Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . 388
16.4 Cohomology of Projective Space . . . . . . . . . . . . . . . . . . . . . . . . . 391
17 Curves 394
17.1 Divisors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395
17.2 Morphisms Between Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . 398
17.3 Linear Equivalence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401
17.4 Répartitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
17.5 Differentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407
17.6 Serre Duality and the Riemann-Roch Theorem . . . . . . . . . . . . . . . . . 411
17.7 The Riemann-Hurwitz Formula . . . . . . . . . . . . . . . . . . . . . . . . . 414
17.8 The Canonical Map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 416
iii
Contents Contents
iv
Contents Contents
v
Contents Contents
33 Sites 686
33.1 Grothendieck Topologies and Sites . . . . . . . . . . . . . . . . . . . . . . . 687
33.2 Sheaves on Sites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 690
33.3 The Étale Site . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 694
34 Cohomology 701
34.1 Direct and Inverse Image Functors . . . . . . . . . . . . . . . . . . . . . . . 702
34.2 Étale Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 705
37 Descent 721
37.1 Galois Descent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 723
37.2 Fields of Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 734
37.3 Galois Descent for Varieties and Schemes . . . . . . . . . . . . . . . . . . . . 736
vi
Part I
Commutative Algebra
vii
Chapter 1
Preliminaries
The contents of Part I come from an Algebra IV course taught by Dr. Peter Abramenko
at the University of Virginia in Spring 2016. The main theory of commutative rings are
covered, with an eye toward applications to algebraic geometry. Topics include:
Integral dependence
Dimension theory
Affine algebras
Hilbert’s Nullstellensatz
The commutative algebra may be found in Atiyah and MacDonald’s Introduction to Com-
mutative Algebra.
We will adhere to several conventions throughout these notes:
All rings are commutative with unity 1. A will always denote such a ring.
Several topics also come from a course on commutative algebra taught by Dr. Craig
Huneke at UVA in Spring 2019. The main topics I’ve included from this course are: associ-
ated primes, some dimension theory and notions of singularity in rings.
1
1.1. Radicals Chapter 1. Preliminaries
1.1 Radicals
There are three important types of radicals that we may define in a commutative ring A.
The first two are defined below.
Definition. The nil radical of A is defined as the intersection of all prime ideals of A,
written \
N (A) = P.
prime ideals
P ⊂A
Lemma 1.1.1. For any x ∈ A, x ∈ J(A) if and only if 1 − xy is a unit in A for all y ∈ A.
Proof. ( =⇒ ) Suppose A(1 − xy) 6= A for some y ∈ A. Then A(1 − xy) ⊂ M for a maximal
ideal M . By definition, x ∈ J(A) ⊆ M and since J(A) is an ideal, xy ∈ M as well. This
means 1 = xy + (1 − xy) ∈ M , a contradiction. Therefore A(1 − xy) = A so there is some
a ∈ A such that a(1 − xy) = 1; that is, 1 − xy is a unit.
( ⇒= ) Conversely, suppose x 6∈ M for some maximal ideal M ⊂ A. Then M + Ax = A
by maximality of M , so b + xy = 1 for some b ∈ M , or 1 − xy = b ∈ M , meaning 1 − xy
cannot be a unit in A.
Proposition 1.1.2. N (A) = {x ∈ A | xn = 0 for some n ≥ 1}. That is, the nil radical
consists of all nilpotent elements in A.
Proof. (⊇) Suppose xn = 0. Then for any prime ideal P , xn = 0 ∈ P which implies x ∈ P .
Thus x ∈ N (A).
(⊆) If x ∈ A is not nilpotent, we will produce a prime ideal not containing x. Let Σ
denote the set of ideals I ⊂ A such that xn 6∈ I for any n ≥ 1. Notice that 0 is an element
of Σ since we are assuming x is not nilpotent; thus Σ is nonempty. Therefore we may apply
Zorn’s Lemma to choose an ideal P ∈ Σ which is maximal among this collection of ideals.
We claim that P is prime. Clearly P 6= A since A is not an element of Σ. Suppose y, z ∈ A
but y, z 6∈ P . Since P is maximal in Σ, the ideals P + Ay and P + Az are both strictly larger
ideals than P . Thus P + Ay, P + Az 6∈ Σ so there exist m, n ∈ N such that xm ∈ P + Ay and
xn ∈ P + Az. Then xm+n ∈ (P + Ay)(P + Az) ⊆ P + Ayz, but xm+n 6∈ P by assumption,
so we must have yz 6∈ P . This proves P is prime.
Definition. For an ideal I ⊂ A, the radical of I √is r(I) = {x ∈ A | xn ∈ I for some n ≥ 1}.
Alternate notations for the radical ideal include I and rad(I).
2
1.1. Radicals Chapter 1. Preliminaries
Corollary 1.1.3. For any ideal I ⊂ A, r(I) equals the intersection of all prime ideals of A
containing I. In particular, r(I) is an ideal.
Proof. Consider the quotient ring Ā = A/I. For x ∈ A we have the following equivalent
statements:
(a) I ⊆ r(I).
(b) I = r(I) if and only if A/I is reduced. In particular, every prime ideal is radical.
3
1.1. Radicals Chapter 1. Preliminaries
r(I) + r(J). Therefore z ∈ r(r(I) + r(J)). On the other hand, if w ∈ r(r(I) + r(J)) then
wm ∈ r(I) + r(J) for some m ≥ 1. This means wm = x + y for some x ∈ r(I), y ∈ r(J). In
turn this says that xk ∈ I and y ` ∈ J for some k, ` ≥ 1. Using the binomial formula, write
m(k+`) k+` k+` k+`−1 k + ` k+`−q q
w = (x + y) = x + (k + `)x y + ... + x y + . . . + y k+` .
q
Note that q ranges from 0 to k + `. When 0 ≤ q ≤ `, the qth term (the first term in the sum
being the 0th term) is divisible by xk and so it lies in I. Otherwise when 1 + ` ≤ q ≤ k + `,
the qth term is divisible by y ` and so it lies in J. Therefore wm(k+`) is of the form a + b
where a ∈ I and b ∈ J. Hence w ∈ r(I + J).
(g) Suppose r(I) = A. Then in particular 1 ∈ r(I) so for some n ≥ 1, 1 = 1n ∈ I.
Thus I = A. On the other hand, if I = A then part (a) says A ⊆ r(A), which implies
r(A) = A.
Example 1.1.5. Let A = Z and I = J = 2Z. Then IJ = 4Z so r(IJ) = 2Z, but I and J
are each prime and thus radical, so r(I)r(J) = IJ = 4Z. Hence r(IJ) ) r(I)r(J), showing
that the containment in (e) may be strict.
Example 1.1.6. Let k be an algebraically closed field and consider A = k[x, y]. Take
I = (y) and J = (y − x2 ). Then I = r(I) since A/I = k[x] which is a PID and hence
reduced. Likewise, k[x, y]/(y − x2 ) ∼ = k[x] by the first isomorphism theorem applied to the
map k[x, y] → k[x], f (x, y) 7→ f (x, x2 ). Thus J = r(J), so r(I) + r(J) = I + J. However,
I + J = (y) + (y − x2 ) = (y, x2 ) and the radical of this ideal is r((y, x2 )) = (y, x), a strictly
larger ideal than I + J. This shows the containment r(I + J) ⊂ r(I) + r(J) is sometimes
strict, so the second radical on the right side of property (f) is necessary for equality.
Lemma 1.1.7. Let J1 , . . . , Jn be ideals of a ring A and J = ni=1 Ji . Then r(J) = ni=1 r(Ji ).
T T
Proof. It suffices to prove the case when n = 2. Suppose x ∈ r(J1 ∩J2 ). Then xn ∈ J1 ∩J2 for
some n ∈ N, so in particular xn ∈ J1 , implying x ∈ r(J1 ), and xn ∈ J2 , implying x ∈ r(J1 ).
Hence x ∈ r(J1 ) ∩ r(J2 ).
On the other hand, if y ∈ r(J1 ) ∩ r(J2 ), then y ∈ r(J1 ) so y k ∈ J1 for some k ∈ N.
Similarly, y ∈ r(J2 ) implies y ` ∈ J2 for some ` ∈ N. Set m = max(k, `). Then y m =
y m−k y k ∈ J1 and y m = y m−` y ` ∈ J2 by the setup, so y m ∈ J1 ∩ J2 . Hence y ∈ r(J1 ∩ J2 ).
This proves r(J1 ∩ J2 ) = r(J1 ) ∩ r(J2 ).
Example 1.1.8. The conclusion of Lemma 1.1.7 does not hold for arbitrary intersections
of ideals. Let kTbe a field and A = k[t] and consider the family of ideals Jn = (tn ) for
`
n ∈T N. Then n∈N Jn = 0 since if t | f (t) for all ` ≥ 1 then necessarily f = 0. So
rT n∈N Jn = r(0) = N (A) = 0 since A is a PID. However, for each n ∈ N, r((tn )) = (t) so
n∈N r(Jn ) = (t) 6= 0.
4
1.1. Radicals Chapter 1. Preliminaries
5
1.2. Nakayama’s Lemma and Consequences Chapter 1. Preliminaries
6
1.3. Localization Chapter 1. Preliminaries
1.3 Localization
One of the most important constructions in commutative algebra is the localization of a ring
at a multiplicatively closed subset.
Examples. Two of the most important examples of multiplicatively closed subsets are:
Proposition 1.3.2. For the A-module A, S −1 A is a ring under the multiplication law as · bt =
ab
st
, with unity 1S −1 A = 11AA .
Proof. We will verify that multiplication is well-defined. The other ring axioms are easily
checked.
Example.
j ϕ0
S −1 A
7
1.3. Localization Chapter 1. Preliminaries
ϕ(a)ϕ(s)−1 . This implies uniqueness. Now define ϕ0 in this way: ϕ0 as = ϕ(a)ϕ(s)−1 for
any a ∈ A, s ∈ S. To see that ϕ0 is well-defined, suppose as = bt in S −1 A. Then there is an
element u ∈ S such that u(at − bs) = 0. Applying ϕ, we have ϕ(u)ϕ(at − bs) = 0. Since
ϕ(u) is a unit in B, it follows that
f : S −1 A ⊗A M −→ S −1 M
a
s
⊗ m 7−→ am
s
.
S −1 A × M −→ S −1 M
a
, m 7−→ am
s s
.
It is easy to check that is A-bilinear, so by the universal property of tensor products, we get a
well-defined homomorphism of A-modules f : S −1 A ⊗A M → S −1 M . Clearly f is surjective.
8
1.3. Localization Chapter 1. Preliminaries
Thus 1s ⊗ m = stt ⊗ m = st1 ⊗ tm = st1 ⊗ 0 = 0. This proves f is injective and thus we have
an isomorphism S −1 A ⊗A M ∼ = S −1 M as A-modules. Finally, it’s easy to see that the same
map is also an isomorphism of S −1 A-modules.
Given an ideal I ⊂ A, we can define the subset S −1 I ⊆ S −1 A by viewing I as an
−1 a
A-module. Specifically, S I = s : a ∈ I, s ∈ S .
(a) S −1 I ⊆ S −1 A is an ideal.
(a) Every proper ideal J ⊂ S −1 A is of the form S −1 I for some ideal I not intersecting S.
is an inclusion-preserving bijection.
9
1.3. Localization Chapter 1. Preliminaries
Corollary 1.3.9. If P ⊂ A is prime then AP is a local ring with unique maximal ideal
S −1 P , where S = A r P . Moreover, there is an inclusion-preserving bijection
S −1 A r S −1 P = as : a ∈ A r P, s ∈ S = as : a ∈ S, s ∈ S = A×
P,
the set of units in AP . This implies S −1 P is the unique maximal ideal of S −1 A. The bijection
follows from (c) of Proposition 1.3.8, using the fact that S ∩ Q = ∅ if and only if Q ⊆ P .
Lemma 1.3.10. Let S be a multiplicatively closed subset and I, J ideals of the ring A. Then
(b) S −1 (I ∩ J) = (S −1 I) ∩ (S −1 J).
10
1.3. Localization Chapter 1. Preliminaries
where p runs over the primes of A containing I and q runs over the primes of S −1 A containing
S −1 I. Hence we have S −1 r(I) = r(S −1 I) in all cases.
Lemma 1.3.11. For a prime ideal P ⊂ A, the quotient field AP /S −1 P is isomorphic to the
field of fractions of A/P .
π i
Proof. Let ϕ : A → − A/P → − Frac(A/P ) be the canonical quotient map composed with
ā
i : ā 7→ 1 . Then P ⊆ ker ϕ, so S = ArP gets mapped to nonzero elements in Frac(A/P ), but
these are all units since Frac(A/P ) is a field. Therefore the universal property of localization
gives us a unique homomorphism f : AP → Frac(A/P) such that f ◦ j = ϕ. Notice that if
a
s
∈ S −1 P then a ∈ P so ϕ(a) = 0 and therefore f as = 0. So S −1 P ⊆ ker f . Clearly f is
onto, so by the first isomorphism theorem, we have AP /S −1 P ∼ = Frac(A/P ).
Example.
6 Let A = Z. Since Z is an integral domain, (0) is a prime ideal and its localization is
A(0) = Q, the field of fractions of Z. For any nonzero prime p ∈ Z, the localization
Z(p) = as : a, s ∈ Z, p - s is called the ring of p-adic integers. Each Z(p) contains only
11
1.4. Transcendence Degree Chapter 1. Preliminaries
Definition. We say E is algebraically independent over k if for any finite subset {x1 , . . . , xn } ⊆
E and any function F ∈ k[t1 , . . . , tn ] for which F (x1 , . . . , xn ) = 0, it must be that F = 0.
When E = {x}, we say the element x is transcendental (or transcendent) over k.
Note that the empty set E = ∅ is a transcendence basis if and only if K/k is an algebraic
extension.
Example 1.4.1. Let K = k(t1 , . . . , tn ) be the field of fractions of the polynomial ring
k[t1 , . . . , tn ]. Then K is a purely transcendental extension of K, with transcendence basis
E = {t1 , . . . , tn }.
ϕ : k[t1 , . . . , tn ] −→ k[x1 , . . . , xn ]
ti 7−→ xi .
Lemma 1.4.3. Let K/k be a field extension and suppose E ⊆ K with |E| = n. Then
(b) If E is finite and K = k(E) then the elements of E may be numbered such that
k(x1 , . . . , xr )/k is purely transcendental for some 0 ≤ r ≤ n and K/k(x1 , . . . , xr ) is
algebraic.
(c) E is a transcendence basis for K/k if and only if K/k(E) is algebraic and K/k(E1 )
is not algebraic for any proper subset E1 ( E.
12
1.4. Transcendence Degree Chapter 1. Preliminaries
Proposition 1.4.4. Assume E1 , E2 ⊆ K are finite subsets with K/k(E1 ) algebraic and E2
algebraically independent over k. Then |E1 | ≥ |E2 |.
F (y1 , . . . , y` , x` , . . . , xm ) = 0 (∗)
13
1.4. Transcendence Degree Chapter 1. Preliminaries
f (x1 , . . . , xn , y1 , . . . , ym ) = 0.
14
1.4. Transcendence Degree Chapter 1. Preliminaries
where each fi1 ,...,im ∈ k[t1 , . . . , tn ], aj ≥ 0 and the sum is over all tuples (i1 , . . . , im ) ∈
Nm . Notice that f (E1 , tn+1 , . . . , tn+m ) ∈ K[tn+1 , . . . , tn+m ] so because E2 is algebraically
independent over K and f (E1 , E2 ) = 0, we must have f (E1 , tn+1 , . . . , tn+m ) ≡ 0. Therefore
each fi1 ,...,im (E1 ) = 0 but because E1 is algebraically independent over k, each fi1 ,...,im ≡ 0.
Hence f ≡ 0, showing E is algebraically independent over k.
To finish the proof, note that K/k(E1 ) and L/K(E2 ) are algebraic by Lemma 1.4.3(c),
and we have a tower k(E) ⊂ K(E2 ) ⊂ L in which each intermediate extension is algebraic,
so L/k(E) is algebraic by transitivity. Hence E is a transcendence basis for L/k. It follows
that tr degk L = |E| = |E1 | + |E2 | = tr degk K + tr degK L.
Proposition 1.4.9. If k is a perfect field, K = k(E) and tr degk K = r, then the elements
of E can be numbered such that k(x1 , . . . , xr )/k is purely transcendental and K/k(x1 , . . . , xr )
is a separable algebraic extension.
Proof. If char k = 0 then this follows from Corollary 1.4.6 and (b) of Lemma 1.4.3. If
char k = p > 0 then k = k p and the result follows from Proposition 4.1 in Lang.
15
Chapter 2
Integral Dependence
A critical concept in algebraic number theory and algebraic geometry is the idea of integrality:
the property that an element (or a collection of elements) satisfies a polynomial identity with
coefficients in a ring. In this chapter, we define integrality and explore the basic properties
of integral extensions of rings, eventually proving the ‘going up’ and ‘going down’ theorems
(Section 2.3) and Noether’s Normalization Lemma (Section 2.4). These results have vital
consequences in later chapters.
16
2.1. Integral Extensions of Rings Chapter 2. Integral Dependence
Lemma 2.1.1. If M is a B-module which is finitely generated as an A-module, then for any
x ∈ B, there exists a monic polynomial f (t) ∈ A[t] of degree at least 1 such that f (x)M = 0.
Pn
Proof. Let M = i=1 Ami . Since M is a B-module, there exist elements aij ∈ A for
1 ≤ i, j ≤ n, such that x satisfies the equations
n
X
xmi = aij mj for each 1 ≤ i ≤ n.
j=1
Pn
On the other hand, x also satisfies xmi = j=1 xδij mj for each i, so subtracting these
equations gives us a system
n
X
(xδij − aij )mj = 0 for each 1 ≤ i ≤ n.
j=1
Set cij = xδij − aij ∈ B and consider the matrix C = (cij ) ∈ Mn (B). Then the system of
equations above can be written
m1 0
.. ..
C . = . (∗)
mn 0
From linear algebra, there exists an adjoint matrix adj(C) ∈ Mn (B) such that adj(C)C =
(det C)In . Multiplying through (∗) on the left by adj(C) gives us
m1 0
.. ..
(det C) . = .
mn 0
Since the mi generated M , this proves that (det C)M = 0. Now, expanding det C using the
Leibniz formula for a determinant yields
X
det C = sgn(σ)cσ(1)1 · · · cσ(n)n
σ∈Sn
Yn
= (x − aii ) + a0n−2 xn−2 + . . . + a00 for some a0i ∈ A
i=1
= x + an−1 xn−1 + . . . + a0 for some ai ∈ A,
n
17
2.1. Integral Extensions of Rings Chapter 2. Integral Dependence
(iii) There exists a subring C ⊆ B containing A[x] such that C is finitely generated as an
A-module.
Proof. (i) =⇒ (ii) If xn +an−1 xn−1 +. . .+a1 x+a0 = 0 for ai ∈ A then xn ∈ M := n−1 i
P
i=0 Ax
which is a finitely generated A-module. By induction, for all m ≥ n, xm ∈ M which proves
A[x] = M . Hence A[x] is finitely generated.
(ii) =⇒ (iii) Let C = A[x]. Then C is finitely generated as an A-module by assumption.
(iii) =⇒ (iv) Let M = C. Then since C is a subring, 1A ∈ C and hence C is a faithful
A-module.
(iv) =⇒ (i) Lemma 2.1.1 implies that f (x)M = 0 for some monic polynomial f (t) ∈ A[t],
but if M is faithful, f (x) = 0. Hence x is integral over A.
Corollary 2.1.3. Let B/A be a ring extension and suppose x1 , . . . , xn ∈ B are integral over
A. Then A[x1 , . . . , xn ] is a finitely generated A-module.
Proof. We prove this by induction on n. The base case for n = 1 follows from (ii) of
Lemma 2.1.2. For n ≥ 2, suppose C := A[x1 , . . . , xn ] is finitely generated over A. Then
there exist elements y1 , . . . , y` ∈ C such that
`
X
C = A[x1 , . . . , xn−1 ] = Ayi .
i=1
Pm−1
By hypothesis, xn is integral over A and therefore over C as well. Then C[xn ] = j=0 Cxjn
for some m ∈ N. Thus
m−1
XX `
A[x1 , . . . , xn ] = C[xn ] = Ayi xjn ,
j=0 i=1
Definition. For a ring extension B/A, the set A0 of elements x ∈ B that are integral over
A is called the integral closure of A in B. If A0 = A, we say A is integrally closed in
B. Further, if A0 = B then B/A is called an integral extension.
Example 2.1.4. By Corollary 2.1.3, A[x1 , . . . , xn ] is an integral extension of A for any set
of algebraic integers x1 , . . . , xn over A.
18
2.1. Integral Extensions of Rings Chapter 2. Integral Dependence
Lemma 2.1.5. Let B/A be a ring extension. Then A0 , the integral closure of A in B, is a
subring of B which contains A.
Proof. It is trivial that A ⊆ A0 , for every a ∈ A is a root of t − a ∈ A[t]. Suppose x, y ∈ A0 .
Then by Corollary 2.1.3, A[x, y] is a finitely generated A-module which is a subring of B.
Clearly x + y, xy ∈ A[x, y] so applying Lemma 2.1.2(iii) shows that x + y and xy are integral
over A.
Proposition 2.1.6. If B/A and C/B are integral extensions of rings, then C/A is also an
integral extension.
Proof. Take x ∈ C. We must show that x is integral over A. First, x is integral over B so
there exist elements b0 , . . . , bn−1 ∈ B such that
xn + bn−1 xn−1 + . . . + b0 = 0.
Then since B/A is integral, x is integral over B 0 := A[b0 , . . . , bn−1 ]. By Corollary 2.1.3, B 0
is finitely generated as an A-module since all the bi are integral over A. Then there are
elements y1 , . . . , y` ∈ B 0 which generate B 0 as an A-module. Also, B 0 [x] is a B 0 -module
which is finitely generated by xj , 0 ≤ j ≤ m − 1. Then we have
m−1
X m−1
XX `
0 0 j
A[x] ⊆ B [x] = Bx = Ayi xj .
j=0 j=0 i=1
Thus by Lemma 2.1.2(iii), x is integral over A. This proves C/A is an integral extension.
Corollary 2.1.7. If A0 is the integral closure of A in B then A0 is integrally closed in B.
Proof. Take x ∈ B which is integral over A0 . Then A0 [x]/A0 is integral by Corollary 2.1.3 and
A0 /A is integral by definition of A0 . By Proposition 2.1.6, this implies A0 [x]/A is integral, so
in particular x is integral over A. Thus x ∈ A0 .
Definition. An integral domain A is called integrally closed (or alternatively, normal)
if A is integrally closed in its field of fractions.
Proposition 2.1.8. Every UFD is integrally closed.
Proof. Let A be a UFD and denote the field of fractions of A by K. Take x ∈ K and write
x = ab for a, b ∈ A and b 6= 0; we may assume a and b are relatively prime, that is, there
is no prime p ∈ A that divides both a and b. Suppose x is integral over A. Then there are
elements a0 , . . . , an−1 ∈ A such that
a n a n−1
+ an−1 + . . . + a0 = 0.
b b
Multiplying through by bn gives us
an + an−1 b + . . . + a0 bn = 0.
Subtracting an to the other side shows that b | an in A, but since a and b are relatively
prime, and therefore an and b are also relatively prime, this is only possible when b is a unit.
Hence b−1 exists in A, so x = ab = ab−1 ∈ A, proving A is integrally closed.
19
2.1. Integral Extensions of Rings Chapter 2. Integral Dependence
Example 2.1.9. Both the integers Z and the Gaussian integers Z[i] are Euclidean domains,
so in particular they are UFDs and therefore integrally closed by Proposition 2.1.8.
Example 2.1.10. Let K be √ a quadratic extension of Q, i.e. a field extension such that
[K : Q] = 2. Then K = Q( d) for some squarefree integer d ∈ Z r {0, 1}, meaning d has
prime factorization d = ±p1 p2 · · · pr for primes pi . We want to compute the integral closure
of Z in K. Since K/Q is Galois, there is one nontrivial Q-automorphism σ ∈ Gal(K/Q):
σ : K −→ K
√ √
a + b d 7−→ a − b d.
Using the norm and trace maps for K/Q, we have that
To verify this, suppose f (α) = 0 for a monic polynomial f (t) ∈ Z[t]. Then 0 = σ(f (α)) =
f (σ(α)), so σ(α) is integral over Z. Since the integral elements form a ring, N (α) = ασ(α)
and Tr(α) = α + σ(α) are also integral over Z. But the norm and trace maps take values in
Q, so N (α), Tr(α) ∈ Q and they’re integral, so both lie in Z.
Conversely, if ασ(α), α + σ(α) ∈ Z then α is a root of the monic polynomial
Therefore α is integral.
For any number field K/Q, we let OK denote the integral closure of Z in K, called the
ring of integers of K. In the quadratic case, our work above shows that
√
OK = {a + b d | a, b ∈ Q and a2 − b2 d, 2a ∈ Z}
( √
Z[h d] i if d ≡ 2, 3 (mod 4)
= √
Z 1+2 d if d ≡ 1 (mod 4).
√
1+ d
To see how the second case arises, suppose d ≡ 1 (mod 4). Then ω = 2
satisfies the
integral expression ω 2 − ω + 1−d
4
= 0.
20
2.2. Integrality and Field Extensions Chapter 2. Integral Dependence
Lemma 2.2.1. If B/A is an integral extension then A is a field if and only if B is a field.
x−1 := −a−1
0 (x
n−1
+ . . . + a1 ).
Lemma 2.2.2. Let A be integrally closed in its fraction field K, let L/K be a finite extension
and suppose α ∈ L is algebraic over K. Then α is integral over A if and only if α has a
minimal polynomial f (t) ∈ A[t].
σi : K(α) −→ K(αi ), α 7→ αi ,
Thus αi is integral over A for all 1 ≤ i ≤ n. It follows that f (t) = ni=1 (t − αi ) ∈ K[t]
Q
has coefficients which are integral over A. Finally, since A is integrally closed, this proves
f ∈ A[t].
21
2.2. Integrality and Field Extensions Chapter 2. Integral Dependence
Remark. The most important rings in algebraic number theory are rings of integers: for a
finite extension K/Q, OK denotes the integral closure of Z in K. By Lemma 2.2.2, we can
write OK = {α ∈ K | fα ∈ Z[t]} where fα denotes the minimal polynomial of α over Q.
Definition. For any finite field extension L/K of degree n = [L : K], there is a K-algebra
homomorphism
Tr = TrL/K : L −→ K
α 7−→ tr(ϕα ).
By properties of the matrix trace tr, the trace form TrL/K is K-linear.
Lemma 2.2.3. A finite extension L/K is separable if and only if TrL/K 6= 0.
Proof. We will only show the forward direction. Suppose L/K is separable. Then L = K(α)
for some α ∈ L by the primitive element theorem. We claim that for some 0 ≤ i ≤ n − 1,
Tr(αi ) 6= 0. Let
Yn
f (t) = (t − αj )
j=1
22
2.2. Integrality and Field Extensions Chapter 2. Integral Dependence
23
2.2. Integrality and Field Extensions Chapter 2. Integral Dependence
Corollary 2.2.6. If A is a PID and L/K is a finite separable extension, where K is the
fraction field of A, then the integral closure B of A in L is a finitely generated free A-module
of rank n = [L : K].
Proof. First, by Proposition 2.1.8, every PID is integrally closed so we may apply Proposi-
tion 2.2.5 and its proof to produce K-bases {u1 , . . . , un } and {v1 , . . . , vn } of L such that
n
X n
X
Aui ⊆ B ⊆ Avi .
i=1 i=1
The
Pn vi form a K-basis, so in particular they are linearly independent over A and thus
i=1 Avi is a free A-module of rank n. By the theory of modules over a PID, we have that
B is free of rank at most n over A, but reversing the roles of the ui and vi , we see that the
rank of B equals n.
Example 2.2.7. If K/Q is finite, then there exists a Q-basis of K, say {α1 , . . . , αn }, which
is a Z-basis of OK . Such a basis is called an integral basis of K (or of OK ). In particular,
Corollary 2.2.6 shows that OK is a free abelian group of rank
√ n = [K : Q].
In the special case of a quadratic extension, K = Q( d), the ring of integers is OK =
Z ⊕ Zω, with (√
d if d ≡ 2, 3 (mod 4)
ω = 1+√d
2
if d ≡ 1 (mod 4).
Thus an integral basis of OK is {1, ω}.
Remark. Proposition 2.2.5 and Corollary 2.2.6 are not true in general if L/K is not a
separable extension, as we will show in Section 4.5.
24
2.3. Integrality, Ideals and Localization Chapter 2. Integral Dependence
Lemma 2.3.1. Assume B/A is a ring extension with I ⊂ A and J ⊂ B ideals such that
J ∩ A = I. Then
(b) If I and J are prime ideals then I is maximal in A if and only if J is maximal in B.
The coefficients of the second equation are in A/I, so we see that x̄ is integral over A/I.
(b) If I and J are prime ideals then A/I and B/J are integral domains. By (a), B/J is
integral over A/I, and by Lemma 2.2.1, B/J is a field exactly when A/I is a field. Hence J
is maximal in B if and only if I is maximal in A.
Lemma 2.3.2. Let B/A be a ring extension and S ⊆ A be a multiplicatively closed subset.
xn + an−1 xn−1 + . . . + a0 = 0 in B.
25
2.3. Integrality, Ideals and Localization Chapter 2. Integral Dependence
is integral. If x
s
∈ S −1 B is integral over S −1 A, there exist fractions a0
s0
, . . . , asn−1
n−1
∈ S −1 A such
that x n a x n−1
n−1 a0
+ + ... + = 0.
s sn−1 s s0
Multiplying through by sn and the common denominator, we can write
Lemma 2.3.3. Suppose B/A is an integral extension of rings. If Q, Q0 ⊂ B are prime ideals
with Q ⊆ Q0 and Q ∩ A = Q0 ∩ A, then Q = Q0 .
Lemma 2.3.4. Suppose B/A is an integral extension of rings and p ⊂ A is a prime ideal.
Then there exists a prime ideal Q ⊂ B with Q ∩ A = p.
i j
Ap = S −1 A S −1 B
26
2.3. Integrality, Ideals and Localization Chapter 2. Integral Dependence
Choose any maximal ideal m ⊂ S −1 B and set Q = j −1 (m) which is a prime ideal of B by
Proposition 1.3.8. We claim that Q ∩ A = p. Since B is integral over A, Lemma 2.3.2(a)
implies S −1 B is integral over S −1 A = Ap . Then by Lemma 2.3.1(b), m maximal in S −1 B
implies m ∩ Ap is maximal in Ap . But Ap is a local ring, so m ∩ Ap = S −1 p. Since the first
diagram commutes, we have another commutative diagram:
Q∩A Q
i j
S −1 p = m ∩ Ap m
Theorem 2.3.5 (Going Up). Suppose B/A is an integral extension of rings and
p0 ( p1 ( · · · ( pn
is a chain of prime ideals in A, in which each inclusion is strict. Then for any chain of
prime ideals
Q0 ( Q1 ( · · · ( Qm−1
in B such that 0 ≤ m ≤ n and Qi ∩ A = pi for each 0 ≤ i ≤ m − 1, there exists a prime
ideal Qm ⊂ B such that Qm−1 ( Qm and Qm ∩ A = pm .
Proof. When m = 0, Lemma 2.3.4 says there exists a prime ideal Q0 ⊂ B with Q0 ∩ A = p0 .
Now to induct, suppose the theorem holds for some m, 1 ≤ m ≤ n. Then Qm−1 ∩ A = pm−1
so consider the extension of quotient rings
27
2.3. Integrality, Ideals and Localization Chapter 2. Integral Dependence
Examples.
3 If A is a PID but not a field, then all nonzero prime ideals are of the form (π) for a
prime element π ∈ A. This shows that dim A = 1.
Q0 ( Q1 ( · · · ( Q` (∗)
in A such that each Qi ⊇ p. Of course we can always add p to the bottom of such a chain,
so if ` is maximal, we must have Q0 = p. Now for any chain of prime ideals p0 ( p1 ( · · · (
pn−1 ( p in A, we can extend (∗) by this new chain, giving
p0 ( p1 ( · · · ( pn−1 ( p = Q0 ( Q1 ( · · · ( Q` ,
a chain of prime ideals in A. This shows dim A ≥ n + `. In particular, this holds for any
such n ≤ ht(p), so we have
28
2.3. Integrality, Ideals and Localization Chapter 2. Integral Dependence
Q0 ∩ A ( Q1 ∩ A ( · · · ( Q` ∩ A
is a chain of prime ideals in A – the strictness of each containment comes from Lemma 2.3.3.
This implies dim B ≤ dim A. On the other hand, if p0 ( p1 ( · · · ( p` is a chain of prime
ideals in A, then by the going up theorem, we get a chain of prime ideals
Q0 ( Q1 ( · · · ( Q`
Lemma 2.3.8. For a ring extension B/A and any prime ideal p ⊂ A, there exists a prime
ideal Q ⊂ B with the property that Q ∩ A = p if and only if pB ∩ A ⊆ p.
Lemma 2.3.9. Suppose A and B are integral domains with B/A an integral extension of
rings, K the field of fractions of A and A integrally closed (in K). If p ⊂ A is a prime ideal
of A and α ∈ pB is any element, then the nonleading coefficients of the minimal polynomial
f of α over K lie in p.
Theorem 2.3.10 (Going Down). Suppose B/A is an integral extension of integral domains
with A integrally closed. If
p0 ) p1 ) · · · ) pn
29
2.3. Integrality, Ideals and Localization Chapter 2. Integral Dependence
Q0 ) Q1 ) · · · ) Qm−1
1 1
n
f (as) = n f (α) = 0.
a a
It follows that a1n f (at) is the minimal polynomial of s over K. Thus each c0i lies in A for
0 ≤ i ≤ n − 1. Suppose that a 6∈ p. Then since p is prime, c0n−i ai = cn−i ∈ p implies c0n−i ∈ p
for each 0 ≤ i ≤ n − 1. Since s ∈ B, we get
by hypothesis. Then s ∈ Qm−1 since Qm−1 is a prime ideal, but this contradicts s ∈ S =
B r Qm−1 . Hence a ∈ p, implying pB 0 ∩ A ⊆ p after all.
For the remainder of the section, let A be an integral domain with field of fractions K.
Whenever S ⊆ Ar{0}, we can consider the local ring S −1 A as a subring of K. In particular,
Ap ⊆ K for every prime p ⊂ A.
\
Lemma 2.3.11. Any integral domain A can be written A = Am .
maximal ideals
m⊂A
Proof. Let D be the intersection of Am over all maximal ideals m ⊂ A. Then D ⊆ K. For
a fixed x ∈ D, consider the ideal I = {a ∈ A | ax ∈ A} ⊆ A. If m is a fixed maximal ideal
of A, by assumption x ∈ Am so there exist elements a, b ∈ A with a 6∈ m such that x = ab .
It follows that ax = b ∈ A, so a ∈ I. Since we took a 6∈ m, this implies I 6⊂ m. Now m was
arbitrary, so we must have I = A. Thus 1 ∈ I, implying x = 1 · x ∈ A and therefore A = D
as required.
30
2.3. Integrality, Ideals and Localization Chapter 2. Integral Dependence
31
2.4. Normalization Chapter 2. Integral Dependence
2.4 Normalization
Let k be a field and A a finitely generated k-algebra, but not necessarily an integral domain.
Definition. Elements a1 , . . . , an ∈ A are said to be algebraically dependent over k if
there exists a nonzero polynomial f ∈ k[t1 , . . . , tn ] such that f (a1 , . . . , an ) = 0. Otherwise
the ai are said to be algebraically independent over k.
If a1 , . . . , an ∈ A are algebraically independent, then the k-algebra homomorphism
ϕ : k[t1 , . . . , tn ] −→ A
ti 7−→ ai
is injective. As a consequence, k[a1 , . . . , an ] ∼
= k[t1 , . . . , tn ] as k-algebras.
Theorem 2.4.1 (Noether’s Normalization Lemma). If A is a finitely generated k-algebra
with A = k[x1 , . . . , xn ] for xi ∈ A, then there exist algebraically independent elements
y1 , . . . , yr ∈ A, 0 ≤ r ≤ n, such that A/k[y
Pn 1 , . . . , yr ] is an integral extension. Moreover,
if k is infinite, we may choose the yi in j=1 kxj .
Proof. We prove the theorem by induction on n. Starting with n = 1, there are two possi-
bilities:
x1 is transcendental over k, in which case we may take r = 1 and y1 = x1 .
x1 is algebraic over k, in which case we may take r = 0, so that k[x1 ]/k is integral.
Now suppose n ≥ 2. If x1 , . . . , xn are already algebraically independent, take r = n and
xi for each 1 ≤ i ≤ n. If not, we claim that there exist elements z1 , . . . , zn−1 ∈ A (and
yi = P
zi ∈ nj=1 kxj if k is infinite) such that the following hold:
(i) k[z1 , . . . , zn−1 , xn ] = A.
(ii) xn is integral over k[z1 , . . . , zn−1 ].
Once we prove the claim, the proof finishes as follows. Set B = k[z1 , . . . , zn−1 ] ⊆ A. Then
A = B[xn ] and A/B by (i) and (ii). By induction, there exist y1 , . . . , yr ∈ B (and
Pis integral P
if k is infinite, yi ∈ n−1 j=1 kzj ⊆ n
j=1 kxj ) such that the yi are algebraically independent over
k and B is integral over k[y1 , . . . , yr ]. It will then follow that A/k[y1 , . . . , yr ] is an integral
extension by Proposition 2.1.6.
To prove the existence of z1 , . . . , zn−1 satisfying (i) and (ii), assume k is infinite. By
assumptionP there is a nonzero polynomial f ∈ k[t1 , . . . , tn ] such that f (x1 , . . . , xn ) = 0.
Write f = di=0 fi (t1 , . . . , tn ) where each fi is homogeneous of (total) degree i and fd 6= 0.
(Since f0 ∈ k, we know d ≥ 1.) We can write fd in two ways:
X
fd (t1 , . . . , tn ) = cJ tj11 · · · tjnn where cJ ∈ k
J=(j1 ,...,jn )∈Nn
0
j1 +...+jn =d
d
X
fd (t1 , . . . , tn ) = Gj (t1 , . . . , tn−1 )tnd−j where each Gj is homogeneous of degree i.
j=0
32
2.4. Normalization Chapter 2. Integral Dependence
Since fd 6= 0, not all of the Gj are 0. This implies fd (t1 , . . . , tn−1 , 1) 6= 0. Since k is infinite,
there exist scalars λ1 , . . . , λn−1 ∈ k such that fd (λ1 , . . . , λn−1 , 1) 6= 0. Plugging these into
the top expression for fd , we get
X jn−1 j1 +...+jn
fd (λ1 xn , . . . , λn−1 xn , xn ) = cJ λj11 · · · λn−1 xn
J
X j
= cJ λj11 · · · λn−1
n−1 d
xn = fd (λ1 , . . . , λn−1 , 1)xdn .
J
where fd (λ1 , . . . , λn−1 , 1)xdn ∈ k × and each Hj ∈ k[t1 , . . . , tn−1 ]. This shows xn satisfies
a nontrivial monic polynomial equation with coefficients in B = k[z1 , . . . , zn−1 ], since in
particular we can divide out by fd (λ1 , . . . , λn−1 , 1) 6= 0. Thus xn is integral over B.
If k is finite, we need a slightly different argument. By the assumption on x1 , . . . , xn ,
there is a nonzero polynomial F ∈ k[t1 , . . . , tn ] such that F (x1 , . . . , xn ) = 0. Write
X j
F (t1 , . . . , tn ) = cj t11 · · · tjnn
j∈Nn
0
for cj ∈ k for all j = (j1 , . . . , jn ) ∈ Nn0 . Set J = {j | cj 6= 0}, which is a finite set since F is a
polynomial. We claim there exists a tuple m = (m1 , . . . , mn ) ∈ Nn0 such that for any distinct
j, j 0 ∈ J, jm 6= j 0 m. To justify such a claim, let N ∈ N be greater than any component
ji of any tuple j = (j1 , . . . , jn ) ∈ J. Then m = (N n−1 , N n−2 , . . . , N, 1) satisfies the desired
conclusion, since for N large enough, every base N expansion of a tuple j ∈ J is unique.
For each 1 ≤ i ≤ n − 1, set zi = xi − xm n . Then k[z1 , . . . , zn−1 , xn ] = k[x1 , . . . , xn ] so (i)
i
33
2.4. Normalization Chapter 2. Integral Dependence
where the lower terms at the end have total degree strictly bounded by m. By our choice
of m, we have that the powers xjm n over all j ∈ J are distinct, so in particular there is a
maximum exponent j m, j ∈ J and for this j ∗ , cj ∗ 6= 0. This proves xn satisfies a nonzero
∗ ∗
polynomial in k[z1 , . . . , zn−1 ][t] which may be made monic by dividing out by cj ∗ 6= 0. Hence
xn is integral over k[z1 , . . . , zn−1 ].
Remark. If A is an integral domain with field of fractions K = k(x1 , . . . , xn ), the r in the
normalization lemma equals tr degk K.
We use Noether’s normalization lemma to prove the fundamental theorem in algebraic ge-
ometry, Hilbert’s Nullstellensatz. This first version is stated in terms of algebraic extensions
of fields; we will prove other versions later.
Theorem 2.4.2 (Hilbert’s Nullstellensatz, Algebraic Version). If K/k is a field extension
such that K is a finitely generated k-algebra, then K/k is a finite algebraic extension.
Proof. Noether’s normalization lemma says that there are yi ∈ K such that K/k[y1 , . . . , yr ]
is integral. Then by Lemma 2.2.1, K a field implies k[y1 , . . . , yr ] is also a field. However,
k[y1 , . . . , yr ] ∼
= k[t1 , . . . , tr ] which is not a field unless r = 0. Thus no such yi exist, so K/k
is an integral extension, which is obviously equivalent to K/k being algebraic. Finally, the
extension is of finite degree because K is finitely generated as a k-algebra.
Definition. For k a field and a = (a1 , . . . , an ) ∈ k n , the ideal associated to a is
ma := (t1 − a1 , . . . , tn − an ) ⊂ k[t1 , . . . , tn ].
ϕa : k[t1 , . . . , tn ] −→ k
F (t1 , . . . , tn ) 7−→ F (a1 , . . . , an ).
34
2.4. Normalization Chapter 2. Integral Dependence
One can check that ϕa is a k-algebra homomorphism. Clearly ma ⊆ ker ϕa , and by k[t1 , . . . , tn ] =
k + ma , we get ma = ker ϕa and this is a maximal ideal of k[t1 , . . . , tn ]. Also, if a 6= b ∈ k n
then ma 6= mb . This defines an injection
k n ,−→ MaxSpec(k[t1 , . . . , tn ])
a 7−→ ma
where MaxSpec(A) = {maximal ideals of A} for any ring A. This mapping is not surjective
in general, but in the case that k is algebraically closed, the Nullstellensatz (2.4.2) implies
that it is surjective.
Corollary 2.4.3. If k is algebraically closed, then the mapping
k n −→ MaxSpec(k[t1 , . . . , tn ])
is bijective.
Proof. We saw that if a 6= b then ma 6= mb , so it remains to show surjectivity. Let m be
a maximal ideal of A := k[t1 , . . . , tn ]. Then the quotient ring K = A/m is a field which is
finitely generated as a k-algebra by images of the monomials t̄i = ti + m in A/m. By the
Nullstellensatz (2.4.2), K/k is algebraic, but since k was assumed to be algebraically closed,
this implies K = k. Therefore for all 1 ≤ i ≤ n, there exists an ai ∈ k such that t̄i = ai = āi
in K. Equivalently, ti ≡ ai (mod m) so ti − ai ∈ m for all i. This proves ma ⊆ m for
a = (a1 , . . . , an ), but we saw that ma is always maximal. Thus ma = m. We conclude that
k n → A is both surjective and injective.
By Corollary 2.4.3, if k is algebraically closed, every proper ideal I ⊂ k[t1 , . . . , tn ] has
a common zero in k n , called a Nullstelle in German. Explicitly, I ⊆ ma for some maximal
ideal ma defined by a point a ∈ k n .
Lemma 2.4.4. If A is a ring in which every prime ideal P which is not maximal is an
intersection of prime ideals properly containing P , then every prime ideal is an intersection
of maximal ideals.
Proof. Suppose to the contrary that p ⊂ A is a prime ideal that is not an intersection of
maximal ideals. In particular, p is not maximal so by hypothesis,
\
p = {Q ⊂ A | Q prime, Q ) p}.
Consider the integral domain B = A/p. Then J(B) 6= 0, where J(B) denotes the Jacobson
radical of B, so there is a nonzero element x ∈ J(B). Let S = {xn | n ≥ 0} be the
multiplicatively closed subset of powers of x in B, and consider the localization Bx = S −1 B.
Since p is prime, A/p is reduced, i.e. N (A/p) = 0, but this means x 6∈ N (B). In particular,
x is not nilpotent, so Bx is nonzero. Let m ⊂ Bx be a maximal ideal and let Q be its
corresponding prime ideal in B. If Q were itself maximal, we would have x ∈ Q but
then Q ∩ S 6= ∅. Therefore Q is not maximal, but for any prime ideal Q0 ) Q, there is no
corresponding prime in Bx , so Q0 ∩S must be nonempty. In particular, x ∈ Q0 for every prime
ideal Q0 ) Q. However, the assumption on p implies every non-maximal
T 0 prime in B =0 A/p is
an intersection of prime ideals properly containing it. Thus Q = {Q ⊂ B prime | Q ) Q},
but then x ∈ Q, a contradiction. Hence no such p exists in A, so every prime ideal in A is
an intersection of maximal ideals.
35
2.4. Normalization Chapter 2. Integral Dependence
Corollary 2.4.5. If A is a finitely generated k-algebra, for k a field, every prime ideal of A
is an intersection of maximal ideals.
Proof. By Lemma T 2.4.4, it suffices to show that for every prime ideal p ⊂ A which is not
maximal, p = {Q T ⊂ A | Q prime, Q ) p}. Suppose to the contrary that there exists such
a prime p with p ( {Q ⊂ A | Q prime, Q ) p}. Let B = A/p. Then B is not a field, and
by assumption the intersection of all nonzero prime ideals in B is nonzero. In particular this
intersection contains a nonzero element f , which
n as in Lemma o 2.4.4 is not nilpotent.
b
Consider the (nonzero) localization Bf = f n : b ∈ B, n ≥ 0 . Then Bf is finitely gener-
ated as a k-algebra since A and B are. Suppose Bf is a field. Then by Nullstellensatz (2.4.2),
Bf is an algebraic extension of k. In particular, Bf /k is integral, so Bf is integral over B
as well. Then Lemma 2.2.1 implies Bf and B are fields simultaneously, but this contradicts
the fact that B is not a field. Therefore Bf cannot be a field, so it has a nonzero prime ideal
m ⊂ Bf . This pulls back along j : B → Bf to a prime ideal m0 ⊂ B. Then f 6∈ m0 but
clearly this is impossible, since f was chosen to lie in the intersection of all prime ideals of
B. Thus no such prime p ⊂ A can exist, so every non-maximal prime ideal of A is equal to
the intersection of all prime ideals properly containing it.
36
2.5. Valuation Rings Chapter 2. Integral Dependence
Example 2.5.1. If A is a UFD and p ∈ A is a prime element, then the localization A(p) =
a
: a, b ∈ A, p - b is a valuation ring. To see this, take dc ∈ K × with c, d ∈ A r {0}. By
b
unique factorization, we may assume dc is in lowest terms, i.e. gcd(c, d) = 1. Then if p - d,
−1 d
we have dc ∈ A(p) . On the other hand, if p | d, we must have p - c so dc = c ∈ A(p) .
Theorem 2.5.2. Every valuation ring A is a local ring and is integrally closed.
Proof. We first prove A is a local ring. Let A× be the set of units in A and let m = A r A× .
One can prove that m is an ideal of A, and since every element outside of m is a unit, m
must be the unique maximal ideal of A.
To prove A is integrally closed, assume x ∈ K × is integral over A. Then there exist
a0 , . . . , an−1 ∈ A such that xn + an−1 xn−1 + . . . + a0 = 0. If x−1 ∈ A, we can multiply the
polynomial expression through by x−(n−1) to obtain
Since x−1 ∈ A, the right side lies in A and thus x ∈ A. This shows A is integrally closed.
Theorem 2.5.3. For K a field and A ⊆ K a nonzero subring, the integral closure of A in
K is equal to \
A= B.
valuation rings B
A⊆B⊆K
We will see in Section 4.1 how valuation rings arise naturally as subrings of a field K.
37
2.6. Dimension and Transcendence Degree Chapter 2. Integral Dependence
38
2.6. Dimension and Transcendence Degree Chapter 2. Integral Dependence
p0 ( p1 ( · · · ( p`
with all kernels nonzero since pi ( pi+1 for all i. By Lemma 2.6.2,
Definition. We say a ring A of finite Krull dimension has the strong dimension property
(SDP) if whenever p0 ( · · · ( p` is a maximal (non-refinable) chain of prime ideals in A,
we have ` = dim A.
p0 ( · · · ( pm−1 ( p
0 = Q0 ( Q1 ( · · · ( Qd
39
2.6. Dimension and Transcendence Degree Chapter 2. Integral Dependence
p0 ( · · · ( pm−1 ( p = Q0 ( Q1 ( · · · ( Qd
where Qi ⊂ A are the prime ideals such that Qi = Qi /p. This is a maximal, non-refinable
chain of primes in A, so by SDP, dim A = m + d = ht(p) + dim A/p.
(2) follows immediately from (1) since A/m is a field and dim F = 0 for any field F .
(3) follows from a similar argument to the proof of (1). Given such a chain of primes
p0 ( · · · ( p`
Q0 ( · · · ( Qj = p0 ( · · · ( P` ( p`+1 ( · · · ( pm .
Then SDP implies n = j + m. When such a chain is chosen so that j = ht(p0 ) and
m − ` = dim A/p` , we must have
0 ( p1 /p ( · · · ( p` /p
Proposition 2.6.6. Let A be a finitely generated k-algebra that is an integral domain. Then
for any nonzero prime ideal p ⊂ A such that ht(p) = 1, we have dim A/p = dim A − 1.
40
2.6. Dimension and Transcendence Degree Chapter 2. Integral Dependence
Corollary 2.6.8. For any n ≥ 1, k[t1 , . . . , tn ] has the strong dimension property. In partic-
ular, dim k[t1 , . . . , tn ]/p = n − ht(p) for any prime ideal p ⊂ k[t1 , . . . , tn ].
41
Chapter 3
42
3.1. Ascending and Descending Chains Chapter 3. Noetherian and Artinian Rings
f ∗ : ΣM 0 −→ ΣM g ∗ : ΣM 00 −→ ΣM
N 0 7−→ f (N 0 ) N 00 7−→ g −1 (N 00 ).
Then f ∗ and g ∗ are both order-preserving and injective maps, so the posets ΣM 0 and ΣM 00
inherit the chain conditions from ΣM .
( ⇒= ) We prove this for the artinian property; the proof for noetherian modules is
similar. Let C : M1 ⊇ M2 ⊇ M3 ⊇ · · · be a descending chain of submodules Mi ⊂ M . These
correspond to chains of submodules
43
3.1. Ascending and Descending Chains Chapter 3. Noetherian and Artinian Rings
By the DCC on ΣM 0 and ΣM 00 , there exists a single N ∈ N such that f −1 (Mn ) = f −1 (MN )
and g(Mn ) = g(MN ) for all n ≥ N . For all n ≥ N , we already have Mn ⊇ MN . On the
other hand, given x ∈ MN , there exists a y ∈ Mn such that g(x) = g(y). This implies
g(x) − g(y) = g(x − y) = 0 by A-linearity. So x − y ∈ ker g which equals im f by exactness.
Thus there is some z ∈ M 0 such that f (z) = x − y ∈ MN . Then z ∈ f −1 (MN ) = f −1 (Mn )
so x − y = f (z) ∈ Mn . Hence x = y + f (z) ∈ Mn so MN ⊆ Mn . This proves the DCC on
ΣM , so M is artinian.
Ln
Corollary 3.1.4. If M1 , . . . , Mn are noetherian (resp. artinian) A-modules then i=1 Mi
is also noetherian (resp. artinian).
Let A be a ring. Then the A-submodules of A are exactly the ideals of A. Then by
Proposition 3.1.2, A is noetherian if and only if all ideals of A are finitely generated.
Corollary 3.1.5. If A is noetherian (resp. artinian), then any finitely generated A-module
M is also noetherian (resp. artinian).
An −→ M
n
X
(a1 , . . . , an ) 7−→ ai mi .
i=1
Lemma 3.1.6. For a field k and a k-vector space V , the following are equivalent:
44
3.1. Ascending and Descending Chains Chapter 3. Noetherian and Artinian Rings
Proof. (i) =⇒ (ii), (iii) follow from the linear algebra fact that every subspace of a finite
dimensional vector space is finite dimensional, and thus has a finite basis.
(ii), (iii) =⇒ (i) If dimk V = ∞ then there is a countable linear independent subset
{vj }∞
j=1 ⊂ V . Let Vi = Span{vj | j ≤ i} for each i ∈ N. Then V1 ( V2 ( · · · is an
ascending chain in which each subspace is finitely generated, so it cannot stabilize. Likewise,
let Wi = Span{vj | j ≥ i} for each i ∈ N. Then W1 ) W2 ) · · · is a descending chain which
doesn’t stabilize. Hence dimk V = ∞ implies is V is neither noetherian nor artinian.
Proposition 3.1.7. Suppose A has maximal ideals m1 , . . . , mn , not necessarily distinct, with
m1 · · · mn = 0. Then A is noetherian if and only if A is artinian.
45
3.2. Composition Series Chapter 3. Noetherian and Artinian Rings
M = M0 ) M1 ) · · · ) Mn = 0.
One can view the length function `(·) as a generalization of the dimension of a vector
space.
Proposition 3.2.1. Assume M is an A-module that has a composition series. Then
(a) For every submodule N ⊂ M , `(N ) ≤ `(M ) and if N is a proper submodule, `(N ) <
`(M ).
M = M0 ) M1 ) · · · ) M` = 0.
N = N0 ⊇ N1 ⊇ · · · ⊇ N` = 0.
46
3.2. Composition Series Chapter 3. Noetherian and Artinian Rings
M = M0 ) M1 ) M2 ) · · ·
By maximality in each step, Mi /Mi+1 is simple for all i ≥ 0. Since M is artinian, this process
must terminate, else the DCC is violated. Then at some point we have
M = M0 ) M1 ) M2 ) · · · ) Mn ) 0
47
3.2. Composition Series Chapter 3. Noetherian and Artinian Rings
Now since g is surjective, g(g −1 (Mi00 )) = Mi00 so the inclusions are strict between the g −1 (Mi00 ),
0 ≤ i ≤ n. Similarly, since f is injective, f −1 (f (Mi0 )) = Mi0 so the inclusions between the
f (Mj0 ), 0 ≤ j ≤ `, are strict. Therefore (∗) is a strictly descending chain of submodules of
M of total length ` + n. Consider each quotient in the first half; since g is surjective, we get
an isomorphism
'
g −1 (Mi00 )/g −1 (Mi+1
00
→ Mi00 /Mi+1
)− 00
for 0 ≤ i ≤ n.
Then each g −1 (Mi00 )/g −1 (Mi+1
00
) is simple. Similarly, injectivity of f gives us an isomorphism
'
0
Mi0 /Mi+1 → f (Mi0 )/f (Mi+1
− 0
) for 0 ≤ j ≤ `,
0
which shows the f (Mi0 )/f (Mi+1 ) are all simple. Hence (∗) is a composition series for M , so
by Proposition 3.2.1(c), `(M ) = ` + n = `(M 0 ) + `(M 00 ).
48
3.3. Noetherian Rings Chapter 3. Noetherian and Artinian Rings
49
3.3. Noetherian Rings Chapter 3. Noetherian and Artinian Rings
which lies in I, showing an +bm ∈ L. Finally, if c ∈ A then can is the leading term of cf ∈ I, so
L is an ideal of A. Now, since A is noetherian, L is finitely generated, say by a1 , . . . , an ∈ A.
For each 1 ≤ i ≤ n, let fi ∈ I be a polynomial having ai as its leading coefficient. For each
of these, write fi = ai tei + . . . so that deg fi = ei . Set E = max{e1 , . . . , en }.
For each 0 ≤ d ≤ E − 1, let Ld be the set of all leading coefficients of polynomials in I
of degree d, together with 0:
( d
)
X
Ld = α ∈ A : there is some ai ti ∈ I with ad = α .
i=0
The proof that each Ld is an ideal of A is similar to the proof above for L. Then each
Ld is finitely generated, say by bd1 , . . . , bdnd ∈ A. Let fd1 , . . . , fdnd ∈ I be polynomials
such that the leading coefficient of fdr is bdr , 1 ≤ r ≤ nd . We claim that f1 , . . . , fn and
fdr , 0 ≤ d ≤ E − 1, 1 ≤ r ≤ nd , are a generating set for I. Let Q denote the ideal generated
by these polynomials. By construction, Q ⊆ I since the generators of Q are elements of I.
If Q ( I, choose f ∈ I r Q of minimum degree d with leading coefficient a.
First suppose d ≥ E. Then a ∈ L so there exist elements c1 , . . . , cn ∈ A such that
a = c1 a1 + . . . + cn an . Observe that g = c1 td−e1 f1 + . . . + cn td−en fn is a polynomial in Q
of degree d and leading coefficient a, so f − g ∈ I has strictly smaller degree than f . By
minimality, we must have f − g = 0, that is, f = g ∈ Q, a contradiction. On the other
hand, if d < E then f ∈ Ld so we have f = c1 bd1 + . . . + cnd bdnd for some cr ∈ A. Then
g = c1 fd1 + . . . + cnd fnd is a polynomial in Q of degree d and leading coefficient a, producing
the same contradiction. It follows that Q = I, so I is finitely generated. This proves that
A[t] is noetherian.
Remark. The converse of Hilbert’s basis theorem is trivially true: A[t] noetherian implies
A noetherian since A ∼
= A[t]/(t) as rings.
Corollary 3.3.3. If A is noetherian, then
Proof. (a) Induct on the number of generators. The base case is Hilbert’s basis theorem.
(b) We can write B = A[b1 , . . . , bn ] for some bi ∈ B. Then there is a (unique) surjective
ring homomorphism
ϕ : A[t1 , . . . , tn ] −→ B
ti 7−→ bi .
So B ∼
= A[t1 , . . . , tn ]/ ker ϕ and we can apply (a) and Lemma 3.3.1.
Corollary 3.3.4. For a field k, any finitely generated k-algebra is noetherian.
Proof. Every field is noetherian (it admits a trivial composition series) so Lemma 3.3.3(b)
may be applied.
50
3.3. Noetherian Rings Chapter 3. Noetherian and Artinian Rings
51
3.3. Noetherian Rings Chapter 3. Noetherian and Artinian Rings
Corollary 3.3.8. Any matrix ring GLn (K), n ≥ 2, over an infinite field K cannot be finitely
generated.
The proof of Hilbert’s basis theorem can be generalized to prove that power series rings
over a noetherian ring are also noetherian.
Theorem 3.3.9. If A is noetherian, then the power series ring A[[t]] is also noetherian.
Proof. For a power series f ∈ A[[t]], where f = ∞ i
P
i=0 ai t , let j be the smallest index such
that aj 6= 0. We say aj is the coefficient of lowest degree of f . If ai = 0 for all i ≥ 0, i.e. f
is the zero power series, we will say its coefficient of lowest degree is 0.
Let I ⊂ A[[t]] be an ideal and define L ⊂ A to be the set of all coefficients of lowest
degree of elements in I:
( ∞
)
X
L = aj : ai ti ∈ I for some j ∈ N0 and ai , i > j .
i=j
The proof that each Jd is an ideal of A is similar to the proof for L. In particular, since A is
noetherian we get that Jd is finitely generated for each 0 ≤ d ≤ E − 1. Say Jd is generated
by βd1 , . . . , βdnd ∈ A. ForP each 0 ≤ d ≤ E − 1 and for each 1 ≤ r ≤ nd , let fdr ∈ A[[t]] be
such that fdr = βdr td + ∞ a
i=d+1 i t i
∈ I. Let Q be the ideal of A[[t]] finitely generated by all
fk , 1 ≤ k ≤ n and fdr , 0 ≤ d ≤ E − 1, 1 ≤ r ≤ nd . We claim that Q = I. Since all of the
generators of Q lie in I, it is clear that Q ⊆ I. On the other hand, take f ∈ I and suppose
the lowest degree term of f is atdP , for some a ∈ L and d ≥ 0. If d ≤ E − 1, then a ∈ Jd so
nd
there exist cr ∈ A such that a = r=1 cr βdr , which implies
∞
nd nd
! nd ∞ X nd
X X X X X
d i d
f− cr fdr = f − cr βdr t + ai t = f − cr βdr t − cr ai ti
r=1 r=1 i=d+1 r=1 i=d+1 r=1
52
3.3. Noetherian Rings Chapter 3. Noetherian and Artinian Rings
which is a power series in A[[t]] of lowest degree at least d + 1. Notice that the ideals Jd
are ascending: J0 ⊆ J1 ⊆ J2 ⊆ · · · . Since A is noetherian, this chain stabilizes, or in other
words, we can always find power series
∞
X
gr = c`r t`−E for 1 ≤ r ≤ nE
`=d
53
3.4. Primary Decomposition Chapter 3. Noetherian and Artinian Rings
If we assume Σ is nonempty, then since A is noetherian, Lemma 3.1.1 guarantees that Σ has
a maximal element I0 . In particular, I0 is not irreducible. Then there exist ideals J, K ⊂ A
such that I0 = J ∩ K but I0 ( J and I0 ( K. Since I0 is maximal in Σ, neither J nor K lie
in Σ. Thus J and K are each a finite intersection of irreducible ideals, but then I0 = J ∩ K is
also a finite intersection of irreducible ideals, a contradiction. Hence Σ must be empty.
Definition. A proper ideal I ⊂ A is primary if for every x, y ∈ A, xy ∈ I implies x ∈ I
or y ∈ r(I). Equivalently, I is primary if xy ∈ I implies that x ∈ I, y ∈ I or x, y ∈ r(I).
Lemma 3.4.5. If a proper ideal I ⊂ A is irreducible, then I is primary.
Proof. Consider B = A/I. Then if I is irreducible, (0) is irreducible in B, i.e. there do not
exist two nonzero ideals J, K ⊂ B with J ∩ K = 0. Assume x, y ∈ B such that xy = 0 and
x 6= 0. We will show that y n = 0 for some n ∈ N, i.e. y is nilpotent. Consider the ascending
chain
Ann(y) ⊆ Ann(y 2 ) ⊆ Ann(y 3 ) ⊆ · · ·
By the ACC, there is some n ∈ N such that Ann(y i ) = Ann(y n ) for all i ≥ n. Now we show
(x) ∩ (y n ) = 0. If b ∈ (x) ∩ (y n ) then b = rx for some r ∈ B, so by = rxy = 0. Also, b = cy n
54
3.4. Primary Decomposition Chapter 3. Noetherian and Artinian Rings
for some c ∈ B, so cy n+1 = cy n y = by = 0. This shows c ∈ Ann(y n+1 ) = Ann(y n ), and thus
0 = cy n = b. Hence (x) ∩ (y n ) = 0 but since (0) is irreducible and x 6= 0, we must have
(y n ) = 0.
Finally, this implies I ⊂ A is primary, since if x0 , y 0 ∈ A such that x0 y 0 ∈ I but x0 6∈ I,
we have x0 + I 6= I and (x0 + I)(y 0 + I) = x0 y 0 + I = I. Then by the first part of the proof,
x0 + I 6= I means (y 0 + I)n = (y 0 )n + I = I for some n ∈ N, and thus (y 0 )n ∈ I.
Theorem 3.4.6 (Primary Decomposition). For any proper ideal I ⊂ A, there exist finitely
many primary ideals Q1 , . . . , Qn such that I = Q1 ∩ · · · ∩ Qn .
Lemma 3.4.7. If Q ⊂ A is a primary ideal, then r(Q) is the smallest prime ideal of A
containing Q.
Examples.
1 If A is a UFD and p ∈ A is a prime element, then (pn ) is p-primary (the usual notation
for (p)-primary) for any n ∈ N.
2 In general, a p-primary ideal need not be a power of p. Let A = k[x, y] and Q = (x, y 2 ).
Then A/Q ∼ = k[y](y 2 ) and in this quotient, the zero divisors are the multiplies of
y, hence nilpotent. This implies Q is primary, and clearly its radical is p = (x, y).
However, p2 ( Q ( p are strict inclusions, so Q is not a power of a prime ideal.
3 If p is a prime ideal, pn need not always be primary. Let A = k[x, y, z]/(xy − z 2 ) and
let x̄, ȳ, z̄ ∈ A be the respective images of x, y, z in the quotient. Set p = (x̄, z̄). Then
p is a prime ideal of A since A/p ∼ = k[y] is an integral domain. However, x̄ȳ = z̄ 2 ∈ p2 ,
but x̄ 6∈ p2 and ȳ 6∈ r(p2 ) = p. So p2 is not primary.
The next result shows that the counterexamples in 2 and 3 do not occur when we
consider maximal ideals.
55
3.4. Primary Decomposition Chapter 3. Noetherian and Artinian Rings
Proof. (a) Assume r(I) is maximal. Then by Corollary 1.1.3, the set of all prime ideals in
A containing I just consists of r(I). Thus B = A/I is a local ring with unique maximal
ideal r(I)/I. If x, y ∈ B such that xy = 0 but y 6∈ r(I)/I, then since r(I)/I is the unique
maximal ideal, y ∈ B × . So xy = 0 =⇒ xyy −1 = 0 =⇒ x = 0. Now if x0 , y 0 ∈ A such that
x0 y 0 ∈ I but y 0 6∈ r(I), then by the previous statement,
(x0 + I)(y 0 + I) = I =⇒ x0 + I = I =⇒ x0 ∈ I.
n
\ n
\
p = r(p) ⊇ r(I) ⊇ r(Qi ) = pi .
i=1 i=1
56
3.4. Primary Decomposition Chapter 3. Noetherian and Artinian Rings
By Theorem 3.4.6 and Lemma 3.4.11, minimal primary decompositions exist for all ideals
I ⊂ A.
Example 3.4.12. Let A = k[x, y, z] and consider the ideals p1 = (x, y), p2 = (x, z) and
m = (x, y, z). Then A/p1 ∼ = k[z] and A/p2 ∼ = k[y] which are integral domains, so p1 and p2
∼
are prime. Also, A/m = k which is a field, so m is a maximal ideal of A. We claim that
p1 ∩ p2 = (x, yz). Clearly x and yz both lie in (x, y) ∩ (x, z), so (x, yz) ⊆ p1 ∩ p2 . On the
other hand, for any f ∈ p1 ∩ p2 , f ∈ (x, y) implies f = gx + hy for g, h ∈ A. But f ∈ (x, z) is
possible if and only if h = h0 z for some h0 ∈ A, in which case we have f = gx+h0 yz ∈ (x, yz).
Hence p1 ∩ p2 = (x, yz).
Now we show that p1 p2 = p1 ∩ p2 ∩ m2 and this is a minimal primary decomposition
of p1 p2 . The left can be written p1 p2 = (x2 , xy, xz, yz). Notice that x2 , xy, xz and yz all
lie in m2 , and since p1 p2 ⊆ p1 ∩ p2 , this shows that x2 , xy, xz, yz ∈ p1 ∩ p2 ∩ m2 . Hence
p1 p2 ⊆ p1 ∩ p2 ∩ m2 . Going the other way, suppose F ∈ p1 ∩ p2 ∩ m2 . Then F ∈ m2 so
F = ax2 + bxy + cxz + dy 2 + eyz + f z 2 for some a, b, c, d, e, f ∈ A. Notice that by the
paragraph above, ax2 + bxy + cxz + eyz ∈ p1 p2 so to show that F ∈ p1 p2 , it will suffice to
show that dy 2 + f z 2 ∈ p1 p2 . By the previous statement, ax2 + bxy + cxz + eyz ∈ p1 ∩ p2
so because F ∈ p1 ∩ p2 , we must have dy z + f z 2 ∈ p1 ∩ p2 . Observe that f z 2 ∈ p2 so it
follows that dy 2 ∈ p2 . But y 2 6∈ p2 and this ideal is prime, so we must have d ∈ p2 . A similar
argument shows that f ∈ p1 . Thus we have dy 2 ∈ p2 p1 and f z 2 ∈ p1 p2 , so dy 2 + f z 2 ∈ p1 p2 .
This proves F ∈ p1 p2 , so p1 ∩ p2 ∩ m ⊆ p1 p2 .
Since p1 and p2 are prime, they are primary (3.4.5). Also, since m is maximal, m2 is
primary (3.4.8(b)). Each term in the primary decomposition has a distinct radical, so we
need only check that none of them contains the intersection of the remaining two. Notice:
x ∈ p1 ∩ p2 but x 6∈ m2 ,
y 2 ∈ p1 ∩ m2 but y 2 6∈ p2 ,
and z 2 ∈ p2 ∩ m2 but z 2 ∈
6 p1 .
(I : J) = {y ∈ A | yJ ⊆ I}.
The following lemma collects some basic facts about ideal quotients.
Lemma 3.4.13. Let I, J ⊂ A be ideals and x ∈ A. Then
(a) (I : J) is an ideal of A and I ⊆ (I : J).
57
3.4. Primary Decomposition Chapter 3. Noetherian and Artinian Rings
⇐⇒ xy ∈ Ij for each Ij
⇐⇒ y ∈ (Ij : x) for each j
\
⇐⇒ y ∈ (Ij : x).
j
58
3.4. Primary Decomposition Chapter 3. Noetherian and Artinian Rings
Proof. By Theorem 3.4.6, (0)Tnhas a primary decomposition and we may assume it is minimal
by Lemma 3.4.11. Set 0 = i=1 Qi where Qi ⊂ A is a pi -primary ideal for some prime ideal
pi . By Proposition 3.4.10, the minimal prime ideals of A are elements of {pi | 1 ≤ i ≤ n}
and by Proposition 3.4.15, each pi ⊆ D. Thus all minimal prime ideals of A are contained
in D.
Therefore the statement holds when n = 2. To induct, we need to show that Q1 · · · Qn−1 +
Qn = A. By hypothesis, for each 1 ≤ i ≤ n − 1 there are elements xi ∈ Qi , yi ∈ Qn−1 such
that xi + yi = 1. Then
n−1
Y
1= (xi + yi ) ∈ x1 · · · xn−1 + Qn ⊆ Q1 · · · Qn−1 + Qn .
i=1
Therefore Q1 · · · Qn−1 + Qn = A. Finally, using the base case and inductive hypothesis, we
have ! ! !
\n n−1
\ n−1
\ n−1
Y
Qi = Qi ∩ Qn = Qi ∩ Qn = Qi Qn .
i=1 i=1 i=1 i=1
(a) If S ∩ p 6= ∅ then S −1 Q = S −1 p.
59
3.4. Primary Decomposition Chapter 3. Noetherian and Artinian Rings
60
3.5. Artinian Rings Chapter 3. Noetherian and Artinian Rings
61
3.5. Artinian Rings Chapter 3. Noetherian and Artinian Rings
Theorem 3.5.5 (Hopkins). For a (commutative) ring A, the following are equivalent:
(i) A is artinian.
N (A) = J(A) = m1 ∩ · · · mn ⊇ m1 · · · mn .
By Proposition 3.5.4, there is some k ∈ N such that N (A)k = 0, but (m1 · · · mn )k ⊆ N (A)k =
0 so by Proposition 3.1.7, A is noetherian. On the other hand, assuming (ii) implies that
A has finitely many minimal prime ideals, by Proposition 3.4.10. But if dim A = 0, each
minimal prime ideal is maximal as well. Thus by Corollary 3.4.2, N (A)k = 0 for some k ∈ N.
Then once again the hypothesis of Proposition 3.1.7 is satisfied, so A is artinian.
Theorem 3.5.6. Let A be a local noetherian ring with unique maximal ideal m. Then one
of the following conditions holds:
m ⊇ m2 ⊇ m3 ⊇
If mn 6= mn+1 for all n ∈ N, the chain is not stationary so A is not artinian. Clearly in this
case mn 6= 0 for any n ∈ N, or else the chain would stabilize. Alternatively, if mn = mn+1
for some n, the locality condition implies mn+1 = J(A)mn . Since A is noetherian, mn is a
finitely generated submodule of A, so Nakayama’s Lemma (1.2.1) implies mn = 0. Now if
p ⊂ A is a prime ideal, 0 = mn ⊆ p so we must have m ⊆ p; otherwise, if x ∈ m r p then
xn ∈ p, which is impossible. By maximality, m = p so m is in fact the unique prime ideal of
A. Hence dim A = 0, so by Theorem 3.5.5, A is artinian.
Examples.
62
3.5. Artinian Rings Chapter 3. Noetherian and Artinian Rings
Proposition 3.5.7. Let A be a local artinian ring with maximal ideal m and residue field
k = A/m. Then the following are equivalent:
(a) Every ideal of A is principal.
(b) m is principal.
63
3.5. Artinian Rings Chapter 3. Noetherian and Artinian Rings
Theorem 3.5.9. Any artinian ring A is the finite direct product of local artinian rings, and
the factors are uniquely determined by A up to isomorphism.
Proof. By Theorem 3.5.9, A is the direct product of finitely many local artinian rings, and
each of these has finite dimension by Proposition 3.5.8.
64
3.6. Associated Primes Chapter 3. Noetherian and Artinian Rings
0 = M0 ⊆ M1 ⊆ · · · ⊆ Mn = M
By Lemma 3.6.2, there exists a prime p ∈ Ass(M ) so the submodule A/p ,→ M has trivial
prime filtration 0 ⊆ A/p. Thus Γ is nonempty. Let N be a maximal element of Γ, say with
prime filtration
0 ⊆ N1 ⊆ · · · ⊆ Nk = N.
If N 6= M , use Lemma 3.6.2 again to choose p ∈ Ass(M/N ) with p = AnnA (x + N ) for some
x ∈ M r N . Then N + Ax has a prime filtration:
0 ⊆ N1 ⊆ · · · ⊆ Nk = N ( N + Ax
with (N + Ax)/N ∼
= AnnA (x + N ) = p, contradicting maximality of N . Hence M = N .
65
3.6. Associated Primes Chapter 3. Noetherian and Artinian Rings
66
3.6. Associated Primes Chapter 3. Noetherian and Artinian Rings
Remark. The proof of (2) shows that if p1 , . . . , pn are the primes corresponding to a prime
filtration of M , then Ass(M ) ⊆ {p1 , . . . , pn }.
Example 3.6.8. Let A = k[x, y](x2 − y 3 ) and define an A-module M by the free resolution
2x
−3y 2
0 → A −−−−−−→ A2 → M → 0.
(M is called the module of Kähler differentials of A; see Section 15.2 and specifically Exam-
ple 15.4.3 for more details). We claim the only associated primes of M are 0 and (x, y). Let
p be a prime ideal of A and consider the localized sequence
0 → Ap → A2p → Mp → 0
Notice that
2
2x4 2x4 2 2x
xm = 5 = 2 2 =x =0
−3xy −3x y −3y 2
3
2x y 2x3 y 2 2x
while ym = = =x y = 0.
−3y 6 −3x2 y 3 −3y 2
Proof. Let Z be the set of all a ∈ A for which ax = 0 for some x 6= 0 in M . It is clear that
any p ∈ Ass(M ) is contained in Z, so one inclusion is true. On the other hand, suppose
a ∈ Z with ax = 0 for some nonzero x ∈ M . The set
Λ = {AnnA (bx) | b ∈ A, b 6= 0}
is nonempty since AnnA (ax) ∈ Λ, so it has a maximal element, say q = AnnA (bx) for some
b 6= 0. If we can show q is prime, then we are done by Lemma 3.6.1. Suppose uv ∈ q
but u 6∈ q. Then ubx 6= 0 and q ⊆ AnnA (ubx), so by maximality, q = AnnA (ubx). Hence
v ∈ AnnA (ubx) = q which shows q is prime as desired.
67
Chapter 4
68
4.1. Discrete Valuation Rings Chapter 4. Discrete Valuations and Dedekind Domains
2 Let k be a field and consider the field of rational polynomials K = k(t). The so-
v∞: K × → Z defined by the degree
called “infinite place” of K is a discrete valuation
function: v∞ (f ) = − deg f for f ∈ k[t] and v∞ fg = deg g − deg f for fg ∈ K × .
For A = k[t−1 ], the prime element t−1 induces the same valuation on Frac(A) = k(t):
vt−1 = v∞ .
3 For k a field, consider the field of rational power series
K = k((t)) = {t−n f | n ∈ N0 , f ∈ k[[t]]}.
Define a function v on K by v(x) = m ∈ Z if x = ∞ i
P
i=m ai t for ai ∈ k and am 6= 0.
Formally let v(0) = ∞ so that v : K → Z is a discrete valuation. As in 2 , the prime
element t in the UFD A[[t]] induces this valuation.
69
4.1. Discrete Valuation Rings Chapter 4. Discrete Valuations and Dedekind Domains
For a prime p and the p-adic valuation vp : Q → Z, the valuation ring O(vp ) =
1
a
b
: a, b ∈ Z, b 6= 0, p - b coincides with Z(p) , the p-adic integers described in Exam-
ple 1.2.2.
(2) O is a local ring with unique maximal ideal m = O r O× = {x ∈ K | v(x) > 0}.
(3) O is a DVR.
(6) O is a PID with unique prime elements of the form uπ, where u is a unit.
70
4.1. Discrete Valuation Rings Chapter 4. Discrete Valuations and Dedekind Domains
Proof. Exercise.
Definition. For a valuation ring O = O(v), an element π ∈ O such that m = (π) is the
unique maximal ideal of O is called a uniformizer of O.
(1) If dim A = 1 then the fact that A is local and a domain implies Spec(A) = {0, m}, where
m is the maximal ideal of A.
(2) Further, if I ⊂ A is a nonzero proper ideal then Spec(A/I) = {m/I} by the corre-
spondence theorem. Thus dim A/I = 0. By Lemma 3.1.3, A noetherian implies A/I is
noetherian as well. Thus by Theorem 3.5.5, A/I is also artinian.
(3) Now m/I = N (A/I) so by Proposition 3.5.4, (m/I)` = 0 for some ` ∈ N, and hence
m` ⊆ I ⊆ m. This shows that r(I) = m, i.e. that I is m-primary by Lemma 3.4.8.
(4) Finally, since dim A = 1, Theorem 3.5.5 says that A is not artinian, so by Theorem 3.5.6,
m` 6= m`+1 for any ` ∈ N0 .
Theorem 4.1.6. Let A be a local noetherian domain with maximal ideal m and residue field
k = A/m, such that dim A = 1. Then the following are equivalent:
(i) A is a DVR.
(v) For all nonzero proper ideals I ⊂ A, there exists an n ∈ N such that I = mn .
(vi) There exists an x ∈ A such that for any nonzero proper ideal I ⊂ A, I = (xn ) for
some n ∈ N.
71
4.1. Discrete Valuation Rings Chapter 4. Discrete Valuations and Dedekind Domains
by the isomorphism theorems. By Proposition 3.5.7, this implies I/m` = (m/m` )n = mn /m`
for some n ≤ `. However, I ⊇ m` and mn ⊇ m` so we must have I = mn .
(v) =⇒ (vi) By Remark (4), m 6= m2 so we can find some element x ∈ m r m2 . By
assumption, there is some n ∈ N such that (x) = mn . Since x 6∈ m2 , n has to equal 1, so
m = (x). Hence for any nonzero ideal I ⊂ A, I = m` = (x` ) for some ` ∈ N.
(vi) =⇒ (i) Suppose m = (xn ) for some x ∈ A and n ∈ N. Since m is maximal, x ∈ m
so we have (x) ⊆ m = (xn ) ⊆ (x) and thus m = (x), that is, m is principal. By Remark (4),
(x` ) 6= (x`+1 ) for any ` ∈ N0 . This means for any nonzero element a ∈ A, there is a unique
` ∈ N0 such that (a) = (x` ) by assumption. Define v : A r {0} → N0 by v(a) = ` where ` is
chosen as described. This map is surjective because x` 7→ ` for any ` ∈ N0 . If a, b ∈ A such
that (a) = (x` ) and (b) = (xm ), then
Moreover, suppose v(a) = ` and v(b) = m. Then (a) = (x` ) and (b) = (xm ) so there exist
u, w ∈ A× such that a = ux` and b = wxm . Without loss of generality assume ` ≤ m. Then
a + b = ux` + wxm = x` (u + vxm−` ) and u + vxm−` ∈ A, so this shows a + b ∈ (x` ) and
hence (a + b) ⊆ (x` ). If (a + b) = (xr ), we have (xr ) ⊆ (x` ) so in particular r ≥ ` and thus
v(a + b) ≥ ` = v(a).
Now by Lemma 4.1.1, v extends uniquely to a discrete valuation v : K × → Z. It
remains to verify that O(v) = A and we will be finished. By construction A ⊆ O(v) since
v|Ar{0} : A r {0} → N0 . On the other hand, if y ∈ O(v) is nonzero then y = ab for some
a, b ∈ A r {0}. Then 0 ≤ v(y) = v(a) − v(b), that is, v(a) ≥ v(b). If v(a) = ` and v(b) = m
as above, then (a) = (x` ) and (b) = (xm ), and since ` ≥ m, we have (a) ⊆ (b). Hence a ∈ (b)
so y = ab ∈ A. This completes the proof that A = O(v), i.e. A is a valuation ring.
72
4.2. Dedekind Domains Chapter 4. Discrete Valuations and Dedekind Domains
73
4.2. Dedekind Domains Chapter 4. Discrete Valuations and Dedekind Domains
Theorem 4.2.4. If A is a Dedekind domain with field of fractions K and L/K is a finite
separable field extension, then the integral closure B of A in L is a Dedekind domain.
Proof. First, B is integrally closed by Corollary 2.1.7. Moreover, since B/A is an integral
extension, dim B = dim A = 1 by Theorem 2.3.7. Finally, since L/K isPseparable, Proposi-
tion 2.2.5 says that there is a K-basis {v1 , . . . , vn } of L such that B ⊆ ni=1 Avi . Since A is
noetherian, we get that B is noetherian as well. Hence B is a Dedekind domain.
74
4.3. Fractional, Invertible Ideals Chapter 4. Discrete Valuations and Dedekind Domains
75
4.3. Fractional, Invertible Ideals Chapter 4. Discrete Valuations and Dedekind Domains
Lm = Im S −1 (A :K I) = Im (S −1 A :K S −1 I) by Lemma 4.3.2
= Im (Am :K Im ) = Im Im−1 = Am .
76
4.3. Fractional, Invertible Ideals Chapter 4. Discrete Valuations and Dedekind Domains
Our results in the previous two sections combine to give us the following characterization
of Dedekind domains.
Theorem 4.3.6. For an integral domain A which is not a field, the following are equivalent:
Theorem 4.3.7. If A is a Dedekind Qdomain then every nonzero ideal I ⊂ A factors uniquely
m
into a product of prime ideals I = i=1 pi , with repetitions allowed.
p1 p2· · · pm = p1 q2 · · · qn
−1
=⇒ p1 p1 p2 · · · pm = p−1
1 q2 · · · qn
=⇒ p2 · · · pm = q2 · · · qn .
77
4.3. Fractional, Invertible Ideals Chapter 4. Discrete Valuations and Dedekind Domains
Qm min(ei ,fi ) f
(c) I + J = i=1 pi . In particular, pei i + pj j = A for all ei and fj when i 6= j.
Proof. (a) One direction is clear: if ei ≥ fi for all iQthen pei i ⊆ pfi i for all i, and hence I ⊆ J.
On the other hand, suppose I ⊆ J. Then I ⊆ m fi f1
i=1 pi ⊆ p1 . By Theorem 4.3.6, p1 is
invertible, so we may multiply by p−11 to obtain:
78
4.3. Fractional, Invertible Ideals Chapter 4. Discrete Valuations and Dedekind Domains
Remark. Let A be a Dedekind domain and take a nonzero prime (maximal) ideal Q p ∈
Spec(A), so that Ap is a DVR. For 0 6= x ∈ A, we have a prime factorization (x) = m ei
i=1 pi
with p1 = p and ei ≥ 0. The ei are determined by (x), so define a function
vp : A r {0} −→ N0
x 7−→ e1 .
A/p21 p2 · · · pn ∼
= A/p21 × A/p2 × · · · × A/pn .
In particular, for any fixed x1 ∈ p1 r p21 , there exists y ∈ A that satisfies y + p21 = x1 + p21
and y + pi = 1 + pi for all 2 ≤ i ≤ n. Thus y − x1 ∈ p21 ⊂ p1 so it follows that y ∈ p1 . We
have (y) ⊆ p1 but (y) 6⊂ p21 and (y) 6⊂ pi for any 2 ≤ i ≤ n. Since A is a Dedekind domain,
(y) has a prime factorization
Y n
(y) = pei i for ei ≥ 0.
i=1
By Proposition 4.3.8(a) and the above work, we conclude that 1 ≤ e1 < 2 and 0 ≤ ei < 1
for each 2 ≤ i ≤ n. In other words, e1 = 1 and e2 = . . . = en = 0, so (y) = p1 . Repeating
the proof for each pi , 1 ≤ i ≤ n shows that every prime ideal of A is principal. Suppose
pi = (yi ). Then any ideal I ⊂ A has a prime factorization
n
Y n
Y
I= pai i = (yiai ) = (y1a1 · · · ynan ).
i=1 i=1
Therefore I is principal.
Corollary 4.3.11. Let I be a nonzero ideal in a Dedekind domain A. Then for any nonzero
a ∈ I, there is some b ∈ I such that I = (a, b).
Proof. We first prove A/I is a PID. By the correspondence theorem, every prime ideal of A/I
along the quotient map A → A/I to a unique prime ideal of A which contains I.
pulls back Q
a
Write I = rj=1 pj j for prime ideals pj ⊂ A and integers aj ≥ 1. Then Proposition 3.4.10(a)
and the fact that dim A = 1 imply that the only prime ideals containing I are p1 , . . . , pr . It
79
4.3. Fractional, Invertible Ideals Chapter 4. Discrete Valuations and Dedekind Domains
follows that A/I has finitely many prime ideals. Therefore by Proposition 4.3.10, A/I is a
PID.
Now take 0 6= a ∈ I. Then by the first paragraph, A/(a) is principal and therefore the
ideal I/(a) can be generated by a single element b+(a), b ∈ I. (If b ∈ (a), this means I = (a)
so I was principal to begin with.) On one hand, it’s immediate that (a, b) ⊆ I. On the other
hand, every y ∈ I r(a) has reduction y+(a) ⊆ I/(a) in A/(a), so y+(a) = r(b+(a)) = rb+(a)
for some r ∈ A, and thus y − rb ∈ (a), so y = rb + sa ∈ (a, b). This shows I = (a, b).
80
4.4. The Class Group Chapter 4. Discrete Valuations and Dedekind Domains
CA = IA /PA .
The order of this group, denoted h(A), is called the class number of A.
In general, class numbers can be arbitrarily large, but a highly important fact for number
fields is that their rings of integers have finite class number:
Theorem 4.4.1. For an algebraic extension K/Q, with ring of integers OK , the class number
hK := h(OK ) is finite.
(i) h(A) = 1.
(ii) A is a PID.
(iii) A is a UFD.
Proof. (i) ⇐⇒ (ii) Suppose h(A) = 1. Then for any ideal I ⊂ A, I is a fractional ideal
(e.g. by Proposition 4.3.3), but since the class group of A is trivial, I must be principal.
Hence A is a PID. Conversely, suppose every ideal I ⊂ A is principal.
Qr Any fractional ideal
ei
J ∈ IA may be written as a formal product of prime ideals J = i=1 pi with ei ∈ Z for each
i. Then each prime ideal is principal: pi = (xi ) for an element xi ∈ A, 1 ≤ i ≤ r, so we have
r
Y r
Y
J= pei i = (xi )ei = Axe11 · · · xerr .
i=1 i=1
Therefore J is a principal fractional ideal, and it follows that PA = IA , that is, h(A) =
|IA /PA | = 1.
(ii) ⇐⇒ (iii) It is well-known that every PID is a UFD. For the converse, suppose p ⊂ A
is a (proper) nonzero prime ideal. Choose 0 6= x ∈ p (which is not a unit since p 6= A) and
write x = pe11 · · · perr for distinct irreducible elements pi ∈ A and exponents ei ∈ N. Since p
is a prime ideal, pi ∈ p for some 1 ≤ i ≤ r. Then (pi ) ⊆ p. Since pi is irreducible, (pi ) is a
prime ideal, so the fact that dim A = 1 implies (pi ) = p. Thus every nonzero prime ideal of
A is principal. This implies the result by unique factorization of ideals.
81
4.4. The Class Group Chapter 4. Discrete Valuations and Dedekind Domains
√
Example√ 4.4.3. Let K = Q( −5) be the quadratic number field with ring of integers
OK = Z[ −5]. Then OK is a Dedekind domain, but it is not a UFD, since for example
√ √
6 = 2 · 3 = (1 + −5)(1 − −5)
and each of the elements in the two factorizations is irreducible but not prime. However,
since OK is Dedekind, the ideal (6) factors uniquely into a product of prime ideals:
(6) = p21 p2 p3 ,
√ √
where p1 = (2, 1 + −5) = (2, 1 − −5)
√
p2 = (3, 1 + −5)
√
p3 = (3, 1 − −5).
√ √
(2) = p21 , (3) √
In particular, √ = p2 p3 , (1 +
√ −5) = p1 p2 and
√ (1 − −5) = p1 p3 . We have
(2) ( p1 ( Z[ −5] and [Z[ −5] : 2Z[ −5]] = 4, so [Z[ −5] : p1 ] = 2 and hence p1 is a
maximal ideal. Similar arguments show that p2 and p3 are also √ maximal (thus prime) ideals.
Moreover, notice that each of p1 , p2 , p3 is not principal, so h(Z[ −5]) > 1 (in fact, the class
number is 2 in this case).
82
4.5. Dedekind Extensions Chapter 4. Discrete Valuations and Dedekind Domains
Definition. A field extension L/K is purely inseparable if for each α ∈ L, the minimal
polynomial of α over K has a single distinct root.
Lemma 4.5.1. For a finite extension L/K, the following are equivalent:
Lemma 4.5.2. Let L/K be any algebraic extension of fields. Then there is a unique subfield
K ⊆ K sep ⊆ L such that K sep /K is separable and L/K sep is purely inseparable.
Theorem 4.5.3. Let A be a Dedekind domain with field of fractions K, L/K a finite ex-
tension and B the integral closure of A in L. Then B is a Dedekind domain.
Proof. It remains to prove the theorem when L/K is inseparable. Assume char K = p > 0.
By Lemma 4.5.2, there is a unique subextension K ⊂ L0 ⊂ L such that L0 /K is separable
and L/L0 is purely inseparable. Let B 0 denote the integral closure of A in L0 . Since L0 /K is
separable, Theorem 4.2.4 shows B 0 is Dedekind. Thus it suffices to consider the case when
K = L0 and A = B 0 .
After this reduction, L/K is purely inseparable, so by Lemma 4.5.1, there is some n ∈ N
n
so that for every α ∈ L, αp ∈ K. Set q = pn . Let M be the splitting field of all polynomials
of the form tq − α, with α ∈ K. Alternatively,
M = {α1/q ∈ K | α ∈ K}
A B C
83
4.5. Dedekind Extensions Chapter 4. Discrete Valuations and Dedekind Domains
n
Define the Frobenius map ϕ : M → K by ϕ(x) = xq = xp . Notice that ϕ is well-defined
by Lemma 2; injective since char K = p; and surjective by the construction of M . It is
clear that ϕ is a homomorphism, so it is an isomorphism (of rings) M → K. Moreover, we
claim A = ϕ(C). Indeed, if y ∈ A then any root of tq − y lies in C and maps to y via ϕ.
Conversely, if x ∈ C then xq − α = 0 for some α ∈ K, i.e. xq ∈ K, but xq ∈ C as well, so
ϕ(x) = xq ∈ C ∩ K = A. We have thus established that A ∼ = C as rings, so C is in particular
a Dedekind domain.
By Theorem 4.3.6, B is a Dedekind domain if and only if every ideal of B is invertible.
For an ideal I ⊂ B, letP J = IC be the corresponding ideal of C. Then J is invertible
m −1
since C is Dedekind, Pm soq q i=1 ai bi = 1q for some elements ai ∈ I, bi ∈ J . Since we are in
q−1 q
characteristic
Pm p, a b
i=1 i i = 1 and b i ∈ K. For each 1 ≤ i ≤ m, set c i = a i b i so that
q−1 −q q −q q −q
i=1 ai ci = 1 and ci ∈ I J ⊆ L. Now ci I ⊆ I J ⊆ J J = C but the previous
statementPshows that ci I ⊆ L. So ci I ⊆ C ∩ L = B, and thus ci ∈ (B : I). In summary,
we have m i=1 ai ci = 1 for elements ai ∈ I and ci ∈ (B : I), so it follows that I(B : I) = B.
Hence every ideal of B is invertible, so B is a Dedekind domain.
In fact, Theorem 4.5.3 can be obtained as a corollary to the much more powerful Krull-
Akizuki Theorem, which we don’t quite have the tools to prove yet.
For the remainder of the section, we discuss counterexamples to Proposition 2.2.5 and
discuss a generalization of this result for affine algebras.
Example 4.5.5. (Kaplansky) Let k be a field of characteristic 2 and set C = k[[t]]. Take
some u ∈ C r k and let K = k(t, u2 ) and L = k(t, u). If u is chosen so that u and t are
algebraically independent over k, for example, then we will have [L : K] = 2. Consider the
rings A = C ∩ K, B = C ∩ L and the field of fractions M = Frac(C) = k((t)).
k K L M
A B C
84
4.5. Dedekind Extensions Chapter 4. Discrete Valuations and Dedekind Domains
Since C is a DVR (Example 3 in Section 4.1), A and B are also DVRs with Frac(A) = K
and Frac(B) = L. In particular, B is the integral closure of A in L – if B 0 is this integral
closure, then B ⊆ B 0 because B is integral over A, but on the other hand B is integrally
closed (using the fact that char k = 2), so it must be that B = B 0 . Suppose B were a finitely
generated A-module. Then we would have B ⊆ Aα1 + . . . + Aαr + Aβ1 u + . . . + Aβs u for
some αi , βj ∈ K. By clearing denominators, there is some m ∈ N such that tm αi ∈ A and
tm βj ∈ A for all 1 ≤ i ≤ r, 1 ≤ j ≤ s. So tm B ⊆ A[u]. Write u = ∞ `
P
`=0 a` t with a` ∈ k. Set
∞
X
m −m−1 −m−1
v = (u − a0 − a1 t − . . . − am t )t =t a` t` .
`=m+1
Theorem 4.5.6 (Noether). Let k be a field and suppose A is a finitely generated k-algebra
that is also an integral domain with field of fractions K. If L/K is any finite extension and
B is the integral closure of A in L, then B is a finitely generated A-module.
A B0 B
Since L/L0 is finite and separable, Proposition 2.2.5 gives us that B is a finitely generated
B 0 -module, so it’s enough to prove B 0 is a finitely generated A-module to obtain the result.
If L0 = K, we’re done so assume L0 6= K. This a nontrivial purely inseparable extension, so
char K = p > 0 and by Lemma 4.5.1, there is some n ∈ N with q = pn and αq ∈ K for every
α ∈ L0 . In particular, if L0 = K(z1 , . . . , zn ) for z1 , . . . , zn ∈ L0 , then for each 1 ≤ i ≤ n,
ziq ∈ K, so there exist fi , gi ∈ A such that ziq = fgii . Since A = k[y1 , . . . , yr ], we have
X X
fi = cJ,i y1j1 · · · yrjr and gi = dJ,i y1j1 · · · yrjr
J=(j1 ,...,jr )∈Nr0 J=(j1 ,...,jr )∈Nr0
85
4.5. Dedekind Extensions Chapter 4. Discrete Valuations and Dedekind Domains
for cJ,i , dJ,i ∈ k, finitely many of which are nonzero. Let S be the finite set of all nonzero
cJ,i and dJ,i , where J ranges over all tuples (j1 , . . . , jr ) ∈ Nr0 and 1 ≤ i ≤ n. Then the field
k 0 := k(s1/q | s ∈ S) is a finite extension of k, and consequently L e := k 0 (y11/q , . . . , yr1/q ) is
a finite extension of L0 . If B e is the integral closure of A in L, e then Be contains B 0 so it’s
0 0
enough to prove the result when L = L e and B = B.e
0 1/q 1/q 0
Now consider C := k [y1 , . . . , yr ] ⊆ L . Then C is integrally closed (e.g. it’s a UFD)
1/q 1/q 1/q
and Frac(C) = L0 , so B 0 ⊆ C. But each yj is integral over A, so C = k 0 [y1 , . . . , yr ] ⊆ B 0
since B 0 is the integral closure of A in L0 . Hence B 0 = C, which is finitely generated over A
by definition. This finishes the proof.
86
Chapter 5
87
5.1. Topological Abelian Groups and Completion Chapter 5. Completion and Filtration
88
5.1. Topological Abelian Groups and Completion Chapter 5. Completion and Filtration
Proposition 5.1.6. For any topological group G, G b is a topological group via pointwise
addition and inversion: [(xn )] + [(yn )] = [(xn + yn )] and −[(xn )] = [(−xn )].
Proof. Consider Cauchy sequences (xn ) and (yn ) in G. We first prove pointwise addition
and inversion of Cauchy sequences are invariant on equivalences classes. If (x0n ) ∼ (xn ) and
(yn0 ) ∼ (yn ) then for all U (0), there are numbers N1 , N2 ∈ N such that x0n − xn ∈ U (0) for
all n ≥ N1 and yn0 − yn ∈ U (0) for all n ≥ N2 . Taking N = max{N1 , N2 }, we have that for
all n ≥ N , (x0n + yn0 ) − (xn + yn ) = (x0n − xn ) + (yn0 − yn ) ∈ U (0) since U (0) is a subgroup.
Thus (x0n + yn0 ) ∼ (xn + yn ), so pointwise addition is well-defined. Similarly, (−x0n ) ∼ (−xn )
so inversion is well-defined.
To prove that G b is a group under these actions, consider two Cauchy sequences (xn ) and
(yn ) in G. Then for a neighborhood U (0), there are N1 , N2 ∈ N such that xm − xn ∈ U (0)
for all m, n ≥ N1 and ym − yn ∈ U (0) for all m, n ≥ N2 . Then for all n ≥ N = max{N1 , N2 },
(xm + ym ) − (xn − yn ) = (xm − xn ) + (ym − yn ) ∈ U (0) since U (0) is a subgroup. Hence
(xn + yn ) is a Cauchy sequence. The zero sequence (0) represents the identity class in G b and
it’s easy to check that [(xn )] + [(−xn )] = [(0)] for all Cauchy sequences (xn ), so we have that
Gb is an abelian group.
Finally, for each neighborhood U (0) in G, define a subgroup U b (0) in G
b by
Then {U
b (0) : U (0) is a neighborhood of 0 in G} forms a basis for a topology on G
b which is
compatible with the group structure on G.b Hence G b is a topological group.
89
5.1. Topological Abelian Groups and Completion Chapter 5. Completion and Filtration
90
5.2. Inverse Limits Chapter 5. Completion and Filtration
Definition. An inverse system is a family of groups (Am )m∈I , for some directed index set
I, together with homomorphisms θm` : Am → A` for every m ≥ ` such that when m ≥ ` ≥ k,
the following diagram commutes:
θm`
Am A`
θmk θ`k
Ak
Definition. Given an inverse system Q (Am , θm` ) over a directed set I, we define a coherent
sequence to be a sequence (ym ) ∈ m∈I Am such that θm` (ym ) = y` for all m ≥ `. The set
of all coherent sequences of (Am , θm` ) is called the inverse limit of the inverse system:
( )
Y
lim Am = (ym ) ∈ Am : θm` (ym ) = y` for all m ≥ ` .
←−
m∈I
91
5.2. Inverse Limits Chapter 5. Completion and Filtration
When (G/Gn , θn )n∈N is the inverse system of quotients in G, the inverse limit coincides
with the completion of G from Section 5.1.
Proposition 5.2.4. For a topological abelian group G with inverse system (G/Gn , θn )n∈N ,
b∼
there is a canonical isomorphism of topological abelian groups G = lim G/Gn .
←−
where kn is the smallest N (or any N ) such that xk + Gn = xN + Gn for all k ≥ N . Such a
choice is possible because (xk ) is Cauchy. Clearly the ζn are homomorphisms, and one can
verify that they are well-defined on equivalence classes of Cauchy sequences. Now define a
map into the inverse limit by
b −→ lim G/Gn
ζ:G
←−
[(xk )] 7−→ (ζn ([(xk )]))n∈N .
since kn+1 ≥ kn , and therefore θn+1 (ζn+1 [(xk )]) = ζn [(xk )]. Now given (yn ) ∈ lim G/Gn ,
←−
construct a Cauchy sequence mapping to (yn ) by taking a coset representative xn for each
n ∈ N such that yn = xn + Gn . Since (yn ) is a coherent sequence, for each n ∈ N we have
So xn+1 − xn ∈ Gn and hence (xn ) is a Cauchy sequence. Clearly ζ([(xn )]) = (yn ) so ζ is
surjective.
To see that ζ is continuous, suppose [(xk )] ∈ G.
b Then using the notation from the proof
of Lemma 5.2.3, for each m ∈ N,
[(xk )] ∈ ζ −1 (S(0,
e m)) ⇐⇒ xkm + Gm = 0 + Gm in G/Gm ⇐⇒ [(xk )] ∈ G
bm .
Since G
bm is open for all m, we have that ζ is continuous.
Finally, to show ζ is an isomorphism, we define an inverse function by
η : lim G/Gn −→ G
b
←−
xn + Gn 7−→ [(xn )].
We showed above that this map is well-defined and a homomorphism of topological abelian
groups, and clearly ζ ◦ η = idlim G/G by surjectivity of ζ. Finally, for any [(xk )] ∈ G,
b we
n
←−
have η ◦ ζ([(xk )]) = η(xkn + Gn ) = [(xkn )] = [(xk )] since subsequences of Cauchy sequences
are equivalent to their parent Cauchy sequence. This proves that ζ is an isomorphism of
topological abelian groups.
92
5.2. Inverse Limits Chapter 5. Completion and Filtration
Definition. When (Gn )n∈N is the system of all subgroups of finite index in G, the inverse
limit G
b = lim G/Gn is called the profinite completion of G.
←−
Definition. A surjective inverse system is one in which every map θm` is surjective.
Definition. Let (Am , αm` ) and (Bm , βm` ) be two inverse systems indexed over the same set
I. A morphism of inverse systems is a set of maps (ϕ• ) such that ϕm : Am → Bm is a
homomorphism for each m ∈ I that is compatible with the αm` and βm` , i.e. the following
diagrams commute for all m ≥ ` ∈ I:
ϕm
Am Bm
αm` βm`
A` B`
ϕ`
Definition. Let (Am , αm` ), (Bm , βm` ) and (Cm , γm` ) be inverse systems with the same index
set I. A short exact sequence of inverse systems is a sequence of maps of inverse
(im ) (pm ) m i pm
systems 0 → (Am ) −−→ (Bm ) −−→ (Cm ) → 0 such that 0 → Am −→ Bm −→ Cm → 0 is an
exact sequence for all m ∈ I.
Proposition 5.2.5. For any short exact sequence of inverse systems 0 → (Am ) → (Bm ) →
(Cm ) → 0, there is an exact sequence of topological abelian groups
In other words, inverse limit is a left exact functor from the category of inverse systems to
the category of topological abelian groups. Furthermore, if (Am ) is a surjective inverse system
then 0 may be added on the right, i.e. lim(−) is an exact functor.
←−
Proof. Let A = Am and define a map dA : A → A by (am ) 7→ (am − θm+1,m (am+1 )). Define
Q
B, dB , C and dC similarly for (Bm ) and (Cm ). Then ker dA = lim Am , ker dB = lim Bm and
←− ←−
kerdC = lim Cm . We have a commutative diagram with exact rows:
←−
0 A B C 0
dA dB dC
0 A B C 0
The first three terms of this are 0 → lim Am → lim Bm → lim Cm so we have our exact
←− ←− ←−
sequence. For the final statement, suppose (Am , αm` ) is a surjective system. We need to prove
coker dA = 0, i.e. for each (am ) ∈ A, we want to find (xm ) ∈ A such that dA (xm ) = (am ).
93
5.2. Inverse Limits Chapter 5. Completion and Filtration
Let x0 = 0. Then a0 = x0 − θ1 (x1 ) so θ1 (x1 ) = −a0 and since θ1 is surjective, this allows us
to choose some x1 ∈ A1 so that this holds. Proceed to inductively define (xm ). This shows
dA is surjective, so we are done.
p
Corollary 5.2.6. Let 0 → G0 → G → − G00 → 0 be a short exact sequence of topological
abelian groups having systems of open subgroups {Gn }, {G0n } and {G00n }, where G0n = G0 ∩ Gn
and G00n = p(Gn ) for each n ∈ N. Then the sequence of completions
b0 → G
0→G b00 → 0
b→G
is exact.
Proof. This is done by considering the short exact sequence of surjective inverse systems
0 → (G0 /G0n ) → (G/Gn ) → (G00 /G00n ) → 0 and applying Proposition 5.2.5.
Corollary 5.2.7. For a topological abelian group G with system of open subgroups Gn and
b for each n ∈ N, G
completion G, bn is a topological subgroup of G
b and G/
b Gbn ∼
= G/Gn .
Proof. Apply Corollary 5.4.9 to the sequence 0 → Gn → G → G/Gn → 0. Notice that
G/Gn ∼ \n ∼
= G/G = G/
b G bn because each has the discrete topology.
bb ∼ b
Corollary 5.2.8. For any topological abelian group G, G = G.
Definition. A topological abelian group G is complete if the canonical map ϕ : G → G,
b x 7→
[(x)], is an isomorphism.
94
5.3. Topological Rings and Module Filtrations Chapter 5. Completion and Filtration
Definition. A topological ring is a ring A which is a topological abelian group such that
the ring multiplication A × A → A, (a, b) 7→ ab, is a continuous map. Topological fields are
defined analagously.
Given an ideal J ⊂ A, one can always give A the structure of a topological ring:
A = J0 ⊇ J1 ⊇ J2 ⊇ J3 ⊇ · · ·
forms a basis of additive subgroups of (A, +) as a topological abelian group. This topology,
called the J-adic topology, makes A into a topological ring.
Examples.
1 Let A = k[t] be the polynomial ring in a single variable and take J = (t). Then
A
b = k[[t]], the power series ring.
2 When A = Z and J = (p) for a prime integer p, the completion Zb is called the p-adic
completion of Z, denoted Zp . Alternatively, Zp = lim Z/pn Z.
←−
M = M0 ⊇ M1 ⊇ M2 ⊇ · · ·
95
5.3. Topological Rings and Module Filtrations Chapter 5. Completion and Filtration
Lemma 5.3.2. If (Mn ) and (Mn0 ) are stable J-filtrations of M , then there exists an n0 ∈ N
0
such that Mn+n0 ⊆ Mn0 and Mn+n 0
⊆ Mn for all n ≥ 0.
Proof. Let n0 ∈ N such that JMn = Mn+1 and JMn0 = Mn+1 0
for all n ≥ n0 . Then
JMn0 = Mn0 +1 and JMn0 = Mn0 +1 . By induction, J Mn0 = Mn0 +n and J n Mn0 0 = Mn0 0 +n .
0 0 n
Then we have
J n M ⊆ J n−1 M ⊆ · · · ⊆ JM ⊆ M
and J n M 0 ⊆ J n−1 M 0 ⊆ · · · ⊆ JM 0 ⊆ M 0 for all n ≥ n0 .
From this we obtain Mn+n0 = J n Mn0 ⊆ J n M ⊆ Mn0 . Reversing the roles of the Mn and Mn0
gives the other containment.
Remark. T∞The ncompletion A along an ideal J can also be described in terms of a metric, as
b
long as n=1 J = 0: if x, y ∈ A, define
1
d(x, y) = where v(x, y) = sup{k ≥ 1 | x − y ∈ J k }.
2v(x,y)
Formally, we set d(x, y) = 0 if v(x, y) = ∞. Completion of the induced metric topology
on A, that is, by formally adding in limits of Cauchy sequences in this topology, yields a
topological ring isomorphic to A.b Moreover, this description makes it easier to see that the
ring operations + and · on A are continuous with respect to the metric topology and extend
to continuous operations on A, b and that the canonical embedding A ,→ A b has dense image.
Proposition 5.3.3. Suppose A is a ring and J ⊂ A is an ideal such that ∞ n
T
n=1 J = 0. Let
Ab be the completion of A along J. Then J A b ⊆ J(A),
b the Jacobson radical of A. b
Proof. By Lemma 1.1.1 it suffices to show 1 − x is a unit in A b for all x ∈ J. Consider the
formal expansion
1
= 1 + x + x2 + . . .
1−x
Set sn = 1 + x + . . . + xn ∈ A. Then for each n ≥ 1, sn − sn+1 = −xn+1 ∈ J n+1 so (sn ) is a
1
Cauchy sequence with respect to the J-adic metric and hence its limit 1−x exists in A.
b
96
5.4. Graded Rings and Modules Chapter 5. Completion and Filtration
Notice
L∞ that A0 is always a subring of A, so each An is an A0 -module. We also define
A+ = n=1 An , which is an ideal of A.
Definition. If A is aL
graded ring, an abelian group M is a graded A-module if M is an
∞
A-module and M = n=0 Mn for subgroups Mn ≤ M such that Am Mn ⊆ Mm+n for all
m, n ∈ N0 .
Example 5.4.1. If A = k[t1 , . . . , tn ] then A is a graded ring with respect to total degree:
∞
M X
A= An where A0 = k and An = kti11 · · · tirr
n=0 (i1 ,...,ir )∈Nr0
i1 +...+ir =n
This is called the standard grading on the polynomial ring A. The ideal A+ here is (t1 , . . . , tn ).
(i) A is noetherian.
97
5.4. Graded Rings and Modules Chapter 5. Completion and Filtration
Remark. If A is a ring and I ⊂ A is an ideal, there are two canonical graded rings we can
associate to I: ∞ ∞
M M
∗ n
A := I and GI (A) := I n /I n+1 .
n=0 n=0
M = M0 ⊇ M1 ⊇ M2 ⊇ · · ·
such that IMn ⊆ Mn+1 for all n ∈ N0 . The standard I-filtration on M is when Mn = I n M ,
which defines the I-topology (or I-adic topology) on M . For any filtration (Mn )n∈N0 , we
define a graded module from M by
∞
M
∗
M = Mn .
n=0
Lemma 5.4.4. Let A be a noetherian ring, I ⊂ A and ideal and M a finitely generated
A-module. Then for an I-filtration (Mn )n∈N0 of M , the following are equivalent:
(i) M ∗ = ∞ ∗
L
n=0 Mn is a finitely generated A -module.
Proof. Note that by Corollary 3.1.5, the assumptions that A is noetherian and M is finitely
generated imply that M is a noetherian A-module. Thus each submodule Mn ⊆ M is finitely
generated, so the modules
Mn
Qn := Mi
n=0
are finitely generated for each n ∈ N, by Corollary 3.1.4. For each n, Qn generates the
following A∗ -submodule of M ∗ :
∞
M
Q∗n := Qn ⊕ I m Mn .
m=1
98
5.4. Graded Rings and Modules Chapter 5. Completion and Filtration
(That Q∗n is an A∗ -submodule follows from the fact that I m Mn ⊆ Mm+n for all m ∈ N0 .)
Moreover, Q∗n is a finitely generated A∗ -module generated by the A-module generators of
Q
Sn∞. By∗ construction, Q∗n ⊆ Q∗n+1 for all n, since I m Mn ⊆ I m−1 Mn+1 for all m ≥ 1. Also,
∗
n=0 Qn = M . With these observations, we now proceed to the main proof.
(i) =⇒ (ii) If M ∗ is a finitely generated A∗ -module then it follows (3.1.5) that M ∗ is
noetherian as an A∗ -module, i.e. M ∗ satisfies the ACC on A∗ -submodules. In particular,
the ascending chain Q∗n ⊆ Q∗n+1 stabilizes,
S∞ so there is some n0 ∈ N0 such that Q∗n = Q∗n0 for
all n ≥ n0 . By the above, M = n=0 Qn = Q∗n0 so Q∗n0 is a finitely generated A∗ -module.
∗ ∗
Then ∞ n0 ∞
M M M
Mn = Mn ⊕ I m Mn0 =⇒ Mn0 +m = I m Mn0 for all m ∈ N0 .
n=0 n=0 m=n0
99
5.4. Graded Rings and Modules Chapter 5. Completion and Filtration
c0 → M
0→M c00 → 0
c→M
is also exact.
100
5.4. Graded Rings and Modules Chapter 5. Completion and Filtration
(b) For any finitely generated A-module M with stable I-filtration (Mn )n∈N0 , GI (M ) is
finitely generated as a GI (A)-module.
Now each Mi /Mi+1 is a finitely generated A-module and even an A/I-module, and A/I ⊆
GI (A). It follows that GI (M ) is a finitely generated GI (A)-module.
101
Chapter 6
Dimension Theory
102
6.1. Hilbert Functions Chapter 6. Dimension Theory
where for each j, gj (x1 , . . . , xs ) is an A0 -linear combination of monomials xe11 · · · xess such
that k1 e1 + . . . + ks es = n − rj . This shows Mn is finitely generated as an A0 -module: the
explicit generators are all elements of the form xe11 · · · xess mj for 1 ≤ j ≤ t and ei ≥ 0 such
that k1 e1 + . . . + ks es = n − rj .
Let C be the class of finitely generated A0 -modules.
Examples.
2 In the special case that A0 = k is a field, C is the class of all finite dimensional k-vector
spaces. Here, the dimension function λ(M ) = dimk M is an additive function.
f1 f2 fn fn+1
Lemma 6.1.2. Suppose 0 → M0 − → M1 − → · · · −→ Mn −−→ 0 is an exact sequence in C.
Then for any additive function λ : C → Z,
n
X
(−1)i λ(Mi ) = 0.
i=0
Proof. For each 1 ≤ i ≤ n + 1, set Ni = im fi which also equals ker fi+1 by exactness. Then
we have short exact sequences
fi+1
0 → Ni → Mi −−→ Ni+1 → 0
103
6.1. Hilbert Functions Chapter 6. Dimension Theory
Finally, additivity implies λ(0) = 0 so we have λ(N0 ) = λ(Nn+1 ) = 0. The result then
follows.
Definition. Let A be a graded ring, C the class of finitely generated A0 -modules and λ an
additive function on C. Then for any module M ∈ C, we define the Poincaré series of M
with respect to λ as
∞
X
PM (t) = λ(Mn )tn ∈ Z[[t]].
n=0
Formally, the Poincaré series of any module is a power series with coefficients in Z. A
surprising result is that for finitely generated A0 -module M , PM (t) is in fact a rational
function.
Theorem 6.1.3 (Hilbert, Serre). For any M ∈ C and additive function λ, there is some
polynomial f (t) ∈ Z[t] such that
f (t)
PM (t) = Qs ki
∈ Q(t).
i=1 (1 − t )
where Ān = An /xs An−ks and x̄i = xi + xs Aki −ks for all n, i. Since ki ≥ 1 for all i, we actually
have Ā0 ∼
= A0 , so the finitely generated modules over Ā0 are precisely the elements of C. In
a moment, this will allow us to apply the induction hypothesis to K and L.
For n ∈ N0 , let fn denote the restriction of f to Mn . We get the following exact sequences:
fn
0 → Kn → Mn −→ Mn+ks → Ln+ks → 0 for n ∈ N0 .
104
6.1. Hilbert Functions Chapter 6. Dimension Theory
since both power series have 0 coefficients for indices less than ks , and the second line implies
that the power series agree in every remaining term. This can be rearranged to read
By the induction hypothesis, there exist polynomials g(t), h(t) ∈ Z[t] such that
s−1
Y s−1
Y
(1 − tki )PL (t) = g(t) and (1 − tki )PK (t) = h(t).
i=1 i=1
It follows that s
Y
(1 − tki )PM (t) = g(t) − tks h(t) ∈ Z[t].
i=1
105
6.1. Hilbert Functions Chapter 6. Dimension Theory
Let λ(M ) = dimk M be the additive dimension function for k-vector spaces M . Notice that
n+s−1
by Lemma 6.1.5, dimk An = s−1 for each n ≥ 1, so the Poincaré series for A as a module
over itself is ∞ ∞
X
n
X n+s−1 n 1
PA (t) = (dimk An )t = t = s
.
n=0 n=0
s − 1 (1 − t)
Hence d(A) = s so the notion of vector space dimension coincides with the dimension of a
graded ring. Notice that in this example, k1 = k2 = · · · = ks , i.e. the generators t1 , . . . , ts
are all homogeneous of degree 1. The following proposition generalizes this and introduces
the Hilbert polynomial of an A-module.
Definition. The function g is called the Hilbert polynomial of M with respect to the
additive function λ.
106
6.1. Hilbert Functions Chapter 6. Dimension Theory
(c) The degree and leading coefficient of g(t) are independent of the choice of Q-filtration.
Notice that G0 (A) is noetherian and dim A/Q = 0 since r(Q) = m is the only prime ideal
of A containing Q. L∞Thus by Theorem 3.5.5, G0 (A) is artinian. The associated graded
module G(M ) = n=0 Mn /Mn+1 is a graded G(A)-module. By Lemma 5.4.10(b), G(M )
is finitely generated as a G(A)-module. Moreover, for each j ∈ N0 , Gj (M ) = Mj /Mj+1
is a finitely generated G0 (A)-module, hence artinian and noetherian by Corollary 3.1.5.
Hence
Pn−1 by Corollary 3.2.2, each quotient Mj /Mj+1 has a composition series, so `(M/Mn ) =
j=0 `(Mj /Mj+1 ) is finite.
(b) We have that Q = (x1 , . . . , xs ) as an ideal of A. For each xi , define x̄i = xi + Q2 ∈
Q/Q2 = G1 (A). Then G(A) = G0 (A)[x̄1 , . . . , x̄s ]. Note that k1 = . . . = ks = 1 in the
notation of Theorem 6.1.3. Consider the Poincaré series
∞
X
PG(M ) (t) = `(Mn /Mn+1 )tn+1 .
n=0
By Proposition 6.1.7, there exist n0 ∈ N and a polynomial f (t) ∈ Q[t] of degree d−1 ≤ s−1,
where d = d(G(M )), such that `(Gj (M )) = f (j) for each j ≥ n0 . Then by (a), for all n ≥ n0
we have
n−1
X 0 −1
nX n−1
X
`(M/Mn ) = `(Mj /Mj+1 ) = `(Mj /Mj+1 ) + f (j).
j=0 j=0 j=n0
Pn−1
Now we show that j=n0 f (j) is a polynomial in n of degree d ≤ s with rational coefficients.
It is a factP that if f ∈ Q[t] with deg f = d − 1, then there exists g̃ ∈ Q[t] P with deg g̃ = d
such that nj=1 f (j) = g̃n for all n. To see this explicitly, write f (t) = `r=0 ar tr so that
Pn P` Pn r Pn r
j=1 f (j) = r=0 a r j=1 j . For all r ∈ N0 , there exists an hr ∈ Q[t] satisfying j=1 j =
hr (n); indeed, these arise as certain Bernoulli polynomials:
If r = 0, h0 (t) = t.
If r = 1, h1 (t) = 21 t + 12 t2 .
If r = 2, h2 (t) = 61 t + 12 t2 + 13 t3 , etc.
107
6.1. Hilbert Functions Chapter 6. Dimension Theory
In any case, we have nj=1 f (j) = `r=0 ar hr (n) which is now a polynomial in n as claimed.
P P
This proves (b).
(c) If (Mn ) and (Mn0 ) are two stable Q-filtrations of M , then by Lemma 5.3.2, there is
some m0 ∈ N0 such that Mn+m0 ⊆ Mn0 and Mn+m 0
0
⊆ Mn for all n ∈ N0 . By (b), we get
integers n0 , n0 ∈ N0 and polynomials g, g ∈ Q[t] such that for all n ≥ max{n0 , n00 },
0 0
g(n)
It follows that lim = 1 so g and g 0 must have the same degree and leading coefficients.
n→∞ g 0 (n)
In the next section, we will prove that when A is local noetherian with maximal ideal m,
the dimension of A may be computed by
108
6.2. Local Noetherian Rings Chapter 6. Dimension Theory
Proof. This follows immediately from Corollary 6.1.9 and Proposition 6.1.10.
Lemma 6.2.2. If Q is an m-primary ideal and M is a finitely generated A-module, then for
any x ∈ A satisfying xz =
6 0 for all z ∈ m r {0},
0
deg χM M
Q (t) ≤ deg χQ (t) − 1
where M 0 = M/xM .
109
6.2. Local Noetherian Rings Chapter 6. Dimension Theory
is a chain of prime ideals in A0 /(x0 ). Since d(A0 /(x0 )) ≤ d(A) − 1, and dim A0 /(x0 ) ≤
d(A0 /(x0 )) ≤ d−1 by the inductive hypothesis, we get r −1 ≤ dim A0 /(x0 ) ≤ d−1. Therefore
r ≤ d. Taking the supremum over all such chains of prime ideals, we get dim A ≤ d.
Corollary 6.2.5. Let A be any noetherian ring. Then
(a) If A is local, dim A < ∞.
Taking the minimum over s on both sides, it follows that δ(m) = dimk m/m2 .
Definition. A local noetherian ring A is said to be a regular ring if dim A = dimk m/m2 .
By the above, A is regular if and only if m = (x1 , . . . , xd ), where d = dim A.
Theorem 6.2.8. Let (A, m, k) be a local noetherian ring of dimension d = dim A. Then the
following are equivalent:
(a) Gm (A) ∼
= k[t1 , . . . , td ] for indeterminates ti .
110
6.2. Local Noetherian Rings Chapter 6. Dimension Theory
Lemma 6.2.9. Let A be a regular ring. Then A is an integral domain which is integrally
closed.
(a) ht(p) ≤ 1.
Proof. (a) is immediate from Krull’s height theorem. For (b), suppose (x) ⊆ p and ht(p) = 0.
Then all elements of p are zero divisors by Corollary 3.4.16, so it must be that ht(p) = 1.
Proof. We have already proved dim A ≤ d := δ(A) in Corollary 6.2.6. For the other inequal-
ity, we construct a sequence of elements x1 , . . . , xd ∈ m satisfying the following condition:
For each 1 ≤ j ≤ d, every prime p ∈ Spec(A)
(∗)
containing (x1 , . . . , xj ) has height ht(p) ≥ j.
Given such x1 , . . . , xd , let Ij = (x1 , . . . , xj ) for each 1 ≤ j ≤ d. Then if p ⊇ Id , ht(p) ≥ d =
ht(m) but this is only possible if p = m. Hence Id is m-primary by Lemma 3.4.8(a), so we
get
dim A = d ≥ δ(Id ) ≥ δ(A).
Hence the result follows once we prove (∗).
We verify (∗) inductively. For 1 ≤ j ≤ d, let x1 , . . . , xj−1 be constructed. If p ∈ Spec(A)
such that p ⊇ Ij−1 , then ht(p) ≥ j − 1 by the inductive hypothesis. By Proposition 3.4.10,
111
6.2. Local Noetherian Rings Chapter 6. Dimension Theory
there are finitely many minimal elements of the set C = {p ∈ Spec(A) | p ⊇ Ij−1 }. Suppose
S1n, . . . , pn are the elements of C S
p with height j − 1 < d. Then pi 6= m for all 1 ≤ i ≤ n, so
n
i=1 i p ( m. Choose x j ∈ m r i=1 pi . It now follows that (∗) holds for these x1 , . . . , xd ,
and the theorem is proven.
We finish the section by proving a partial converse to Theorem 6.2.10. First we need a
technical lemma.
Lemma 6.2.14 (Prime Avoidance). Suppose S I1 , . . . , In ⊂ A are ideals of which at least n−2
are prime. Then for any ideal J ⊂ A, if J ⊆ nk=1 Ik then J ⊆ Ik for some 1 ≤ k ≤ n.
Theorem 6.2.15. Let A be a local noetherian ring of dimension d, with maximal ideal m.
Then there exist d elements x1 , . . . , xd ∈ m such that r(x1 , . . . , xd ) = m.
Proof. If d = 0, we take (x1 , . . . , xd ) = 0. In this case, Theorems 3.5.5 and 3.5.6 say that
A is artinian and mn = 0 for some n ≥ 1. Thus r(0) = m by definition. Now induct on
d > 0. Since A is noetherian, the set of minimal primes of A is finite (Proposition 3.4.10),
say {p1 , . . . , pk }. Clearly m 6⊂ pi for any 1 ≤ i ≤ k, or else we would be in the base case. By
the prime avoidance lemma, m 6⊂ p1 ∪ · · · pk . Choose x1 ∈ m such that x1 6∈ p1 ∪ · · · ∪ pk .
Then dim(A/(x1 )) ≤ d − 1 so by induction, there exist d − 1 elements x̄2 , . . . , x̄d ∈ A/(x1 )
such that r(x̄2 , . . . , x̄d ) = m̄ in A/(x1 ). Lifting back to A, we see that r(x1 , . . . , xd ) = m as
desired.
Proof. If dim(A/(x)) < d − 1 then we could choose a smaller number of elements to generate
m up to radical, but this would contradict Krull’s height theorem.
112
6.3. Complete Local Rings Chapter 6. Dimension Theory
Set Gn+1 = Gn + i=1 yi Bi and Hn+1 = Hn + `i=1 yi Ai . Then since each yi ∈ mn , we see
P` P
that Gn − Gn+1 , Hn − Hn+1 ∈ mn A[t]. Moreover,
`
X X
F − Gn+1 Hn+1 = F − Gn Hn − yi (Ai Gn + Bi Hn ) − yi yj Bi Aj
i=1 i,j
X
2n
=− yi yj Bi Aj ∈ m A[t]
i,j
which implies F − Gn+1 Hn+1 ∈ mn+1 A[t] since 2n ≥ n + 1. Hence (Gn ) and (Hn ) are Cauchy
sequences in A[t] so they converge to some G, H ∈ A[t]. By construction, F − GH ∈ mn A[t]
for all n ≥ 1, so by hypothesis, F = GH.
113
6.3. Complete Local Rings Chapter 6. Dimension Theory
Rp = {xp | x ∈ R}
114
6.3. Complete Local Rings Chapter 6. Dimension Theory
Corollary
T∞ 6.3.7. Let (A, m, k) be a complete local ring of characteristic p > 0 such that
n
n=1 m = 0 and k is perfect. Then
∞
e
\
L= Ap
e=1
Lemma 6.3.8. Let A T∞= An be the completion along an ideal I ⊆ A and suppose M is an
b
A-module such that n=1 I M = 0. If x1 , . . . , xr ∈ M such that M/IM is generated by
x̄i = xi + IM, 1 ≤ i ≤ r, then M is generated by the xi .
r
X
m0 = ai0 xi + m1
i=1
115
6.3. Complete Local Rings Chapter 6. Dimension Theory
k[[t1 , . . . , tr ]] −→ A
sending ti 7→ xi .
, . . . , tr ]] → A by sending ti 7→ xi as suggested. This means for an
Proof. Define π : k[[t1P
arbitrary power series αn1 ,...,nr tn1 1 · · · tnr r in k[[t1 , . . . , tr ]],
X X
π αn1 ,...,nr tn1 1 · · · tnr r = αn1 ,...,nr xn1 1 · · · xnr r
and the partial sums of this series form a Cauchy sequence in A, so they converge. Hence π
is well-defined and is a homomorphism by construction. It remains to show π is surjective.
But A is a module over k[[t1 , . . . , tr ]] and A/(t1 , . . . , tr )A = A/m is generated by 1, so by
Lemma 6.3.8, A is generated by 1 as a k[[t1 , . . . , tr ]]-module. This implies π is surjective.
Corollary 6.3.10. If A is a complete local ring containing a field with ∞ n
T
n=1 m = 0 and m
is finitely generated, then A is noetherian.
Proof. Apply Theorem 3.3.9 and Lemma 3.3.1(a).
Definition. For a local ring A with maximal ideal m, a system of parameters for A is a
set of elements x1 , . . . , xn ∈ m such that r((x1 , . . . , xn )) = m.
Recall (Theorem 6.2.12) that if x1 , . . . , xn is a system of parameters for A, then n =
dim A.
Theorem 6.3.11 (Cohen’s Third Structure Theorem). Let (A, m, k) be a complete local
noetherian ring of dimension d containing a field and suppose x1 , . . . , xd ∈ m is a sytem of
parameters for A. Then the map
ϕ : k[[t1 , . . . , td ]] −→ A
ti 7−→ xi
identifies k[[t1 , . . . , td ]] ∼
= im ϕ ⊆ A and the resulting ring extension A/k[[t1 , . . . , td ]] is finite.
116
6.3. Complete Local Rings Chapter 6. Dimension Theory
Example 6.3.12. Consider the 1-dimensional local ring A = R[x](x2 +1) with maximal ideal
m = (x2 + 1)A. Then the residue field of A is C but C does not embed into A. By Cohen’s
first structure theorem (6.3.6), C does embed into the completion A
b but it’s not immediately
obvious how this embedding is defined. Note that understanding C ,→ A b comes down to
2 2 2
finding an element u ∈ A b such that u = −1. Since x + 1 ∈ m, 1 − (x + 1) is a unit and
we have
x2
= −1.
1 − (x2 + 1)
Applying Hensel’s lemma (6.3.1) produces a square root of 1 − (x2 + 1), so set
x
u= p .
1 − (x2 + 1)
117
Part II
Homological Algebra
118
Chapter 7
Category Theory
119
7.1. Categories and Functors Chapter 7. Category Theory
For every A ∈ obj(C) there is an identity morphism 1A ∈ HomC (A, A) such that for all
f ∈ HomC (A, B), f ◦ 1A = f and likewise for HomC (B, A).
Examples.
1 Veck is a category consisting of vector spaces over a field k as the objects and linear
transformations as the morphisms. This is the same category as k-Mod (see below).
2 For a ring R with identity, R-Mod is the category of (left) R-modules with R-linear
maps as the morphisms. In this case we will denote HomR-Mod (A, B) as HomR (A, B).
3 Groups (or Grps) is the category consisting of groups together with group homomor-
phisms. Even more generally, we can define the category Sets: objects are sets and
morphisms are just regular functions. Other algebraic categories arise in a similar fash-
ion: AbGrps or Ab is the category of abelian groups with the usual homomorphisms;
Rings is the category of rings with ring homomorphisms, where we assume all rings
have unity; likewise, ComRings are commutative rings.
4 For a topological space X, we can form a category by taking the open subsets of X as
objects and for any open U, V ⊂ X,
(
∅ if U 6⊆ V
HomX (U, V ) =
{iU,V } if U ⊆ V
where iU,V is the inclusion U ,→ V . This category can be generalized to any space X
which has a partial ordering .
5 As another topological example, Top is the category of topological spaces together with
continuous functions between them.
How do we compare two categories?
Definition. A functor F from C to D is an assignment
120
7.1. Categories and Functors Chapter 7. Category Theory
Remark. The functor defined above is sometimes called a covariant functor: the image of
a morphism f : A → B is a morphism F (A) → F (B). On the other hand, a contravariant
functor takes f : A → B to F (f ) : F (B) → F (A), and F (gf ) = F (f )F (g).
Examples.
Forget : Veck −→ Ab
V 7−→ (V, +)
f 7−→ f
takes a vector space V and “forgets” its vector space structure. We can go further and
forget a group structure, giving a functor Ab → Sets.
π1 : Top −→ Grps
X 7−→ π1 (X)
f :X→Y f ∗ : π1 (X) → π1 (Y ).
3 The nth homology group of a topological space is a functor Hn (·) : Top → Ab.
4 Let C be the category of field extensions K/k with morphisms ϕ : K → L such that
ϕ|k = idk , i.e. the set of k-homomorphisms. Then Aut is the functor defined by
Aut : C −→ Grps
K/k 7−→ Aut(K/k).
Another category can be formed by taking obj(C) to be all Galois extensions K/k.
Then all automorphism groups are called Galois groups, and the functor is
Gal : C −→ Grps
where the last map is the quotient map. Notice that both of these are contravariant
functors.
121
7.1. Categories and Functors Chapter 7. Category Theory
( )∗ : Veck −→ Veck
f :V →W f∗ : W∗ → V ∗
g 7→ gf
6 Abelianization is a functor:
(·)ab : Groups −→ Ab
The main functors of interest in this course will be for the categories R-Mod:
− ⊗R N , a covariant functor.
Then a contravariant functor from C is simply a covariant functor from C op . For example,
we see that the category of left R-modules is isomorphic (to be defined) to the category of
right Rop -modules, where Rop = R as sets and a ·Rop b = b ·R a.
122
7.1. Categories and Functors Chapter 7. Category Theory
so we see that
T op (g op f op ) = T op ((f g)op ) = T (f g) = T (f )T (g)
since T is covariant. But this last part is just T op (f op )T op (g op ) so we have shown that T op is
contravariant.
Finally T op (1A ) = T (1A ) = 1T (A) = 1T op (A) so in particular T op is a contravariant functor.
Just as homomorphisms compare algebraic objects and functors compare categories, there
is a way to compare functors themselves.
S(f ) T (f )
S(A0 ) T (A0 )
τA0
Proof omitted.
123
7.1. Categories and Functors Chapter 7. Category Theory
Theorem 7.1.5. If A ∼ = B then HomC (A, −) and HomC (B, −) are naturally isomorphic
functors. Similarly, HomC (−, A) ∼
= HomC (−, B).
Proof. Let A ∼= B via the isomorphism α : A → B and let α−1 be its inverse. Define a
transformation τ : HomC (A, −) → HomC (B, −) by
f f
Hom(A, C 0 ) Hom(B, C 0 )
τC 0
If h ∈ Hom(A, C) then it maps h 7→ hα−1 7→ f hα−1 via the top map, and h 7→ f h 7→ f hα−1
via the bottom map, so indeed the diagram commutes. Thus τ is a natural transformation.
Finally we need to check that each τC is an isomorphism. To do this, construct a trans-
formation σC : HomC (B, C) → HomC (A, C) taking g 7→ gα. Then for any f ∈ HomC (A, C)
and g ∈ HomC (B, C) we have
124
7.1. Categories and Functors Chapter 7. Category Theory
Proof. Let τ : HomC (−, A) → G be a natural transformation. Then for each B ∈ obj(C) and
morphism ϕ : A → B, the following diagram commutes:
τA
Hom(A, A) G(A)
ϕ∗ G(ϕ)
Hom(B, A) G(B)
τB
Hom(A, C) G(C)
τC
Hence τ is a natural transformation and by construction, τA (1A ) = G(1A )(x) = 1G(A) (x) = x.
This proves the bijection.
Corollary 7.1.7. If HomC (−, A) and HomC (−, B) are naturally isomorphic, then A ∼
= B.
Proof. Suppose τ : HomC (−, A) → HomC (−, B) is a natural isomorphism. Applying Yoneda’s
Lemma to G = HomC (−, B), we get a unique element x ∈ G(A) = HomC (A, B). Applying
Yoneda to the inverse of τ yields a unique inverse to x, proving A ∼
= B.
For any category C, there is a functor C → Fun(C op , Sets) given by A 7→ HomC (−, A).
The Yoneda Lemma shows that this functor is fully faithful, i.e. it embeds C as a full
subcategory of Fun(C op , Sets). More precisely:
Corollary 7.1.8 (Yoneda Embedding). There is a functor C → Fun(C op , Sets), A 7→ HomC (−, A),
which is an isomorphism on HomC (B, C) for all B, C ∈ C.
Lemma 7.1.9 (Yoneda’s Lemma, Covariant Version). Let C be a category, A ∈ obj(C) and
G : C → Sets a covariant functor. Then there is a bijection
125
7.1. Categories and Functors Chapter 7. Category Theory
Corollary 7.1.10 (Yoneda Embedding, Covariant Version). There is a fully faithful functor
C → Fun(C, Sets), A 7→ HomC (A, −).
126
7.2. Exactness of Sequences and Functors Chapter 7. Category Theory
is called a short exact sequence. In particular, this means that f is injective, g is surjective
and ker g = im f . This is sometimes called an “extension of A by C”.
f g
If 0 → A →
− B→ − C → 0 is a short exact sequence of R-modules, the first isomorphism
theorem for modules tells us that A ∼
= im f and B/ im f ∼
= C.
Example 7.2.1. For a commutative ring R, take the set S of all R-module short exact
sequences:
S = {0 → A → B → C → 0 exact | A, B, C ∈ R-Mod}
Then there is an equivalence relation ∼ such that
S/ ∼∼
= Ext1R (C, A).
is exact. If it is both left and right exact, then F is simply called exact.
Proposition 7.2.2. HomR (X, −) is a left exact functor on the category R-Mod.
f g
Proof. Let 0 → A → − B → − C → 0 be an exact sequence of left R-modules and take
X ∈ obj(R-Mod). If we apply HomR (X, −), the sequence becomes
f∗ g∗
0 → HomR (X, A) −
→ HomR (X, B) −
→ HomR (X, C).
127
7.2. Exactness of Sequences and Functors Chapter 7. Category Theory
First, take h ∈ HomR (X, A) such that f h = 0. Note that f is one-to-one ⇐⇒ there
exists a morphism f˜ : im f → A such that f˜f = idA . In this case
h = idA h = f˜f h = f˜0 = 0
which shows that h = 0, i.e. f∗ is one-to-one.
Now let j ∈ HomR (X, B) such that gj = 0, i.e. j ∈ ker g∗ . We need to construct a
function k : X → A such that f k = j. Let x ∈ X. Then g(j(x)) = 0 so j(x) ∈ ker g. But
ker g = im f so there exists a (unique) element a ∈ A such that f (a) = j(x). Now we can
define k : X → A by mapping x to this unique a. Then j(x) = f (a) = f (k(x)) so j = f k as
desired. Hence HomR (X, −) is left exact.
Note that HomR (X, −) may not always be right exact. In fact, even in the category of
abelian groups, HomZ (X, −) is not right exact, as the following example shows.
Example 7.2.3. Consider the exact sequence
0 → 2Z → Z → Z/2Z → 0.
Let X = Z/2Z. Then Proposition 7.2.2 tells us that
0 → Hom(Z/2Z, 2Z) → Hom(Z/2Z, Z) → Hom(Z/2Z, Z/2Z)
is exact. But up to isomorphism, this sequence is
0 → 0 → 0 → Z/2Z.
Clearly adding a zero on the right makes the sequence not exact, since the kernel of Z/2Z → 0
is necessarily Z/2Z.
Two important examples in the theory of modules and their homological properties are
Definition. An R-module P is projective if HomR (P, −) is exact.
Definition. An R-module E is injective if HomR (−, E) is exact.
The proof of Proposition 7.2.2 may be adapted to show that the contravariant functor
HomR (−, Y ) is always left exact, so proving that a module is projective or injective comes
down to proving that the associated Hom is right exact. Example 7.2.3 shows that Z/2Z is
not a projective Z-module. We will explore these special modules in Chapter 8.
Proposition 7.2.4. P is a projective module if and only if for every surjection f : A → B,
f∗ : HomR (P, A) → HomR (P, B) is surjective. This in turn is equivalent to saying for every
diagram of the form shown below, where the bottom row is exact, there exists a g̃ that makes
the diagram commute.
P
g̃
g
A B 0
f
128
7.2. Exactness of Sequences and Functors Chapter 7. Category Theory
Theorem 7.2.5 (Hom-Tensor Adjointness). Given rings R and S and modules AR ,R BS and
CS , there is an isomorphism
The term adjointness comes from the idea that Hom may be viewed as an inner product
on R-Mod:
h·, ·i = HomR (−, −) :R Mod ×R Mod −→ Ab
which has ⊗R as its adjoint functor. In particular, we have functors
FB := − ⊗R B : ModR −→ ModS
GB := HomS (B, −) : ModS −→ ModR .
Then Theorem 7.2.5 says that FB and GB are adjoint functors. FB is sometimes called the
left adjoint and GB the right adjoint, and together they are called an adjoint pair.
Notice that the above setup uses right actions on A ⊗R B and HomS (B, C). There is a
similar setup involving left actions, which is not discussed here.
(The other properties are proven similarly.) By the universal mapping property, this induces
a linear map
f ⊗ 1 : A ⊗R B 7−→ A0 ⊗R B
a ⊗ b 7−→ f (a) ⊗ b.
It is clear that im(1 ⊗ i) ⊆ ker(1 ⊗ p) because pi = 0, but it is not as obvious in the other
direction. We will prove exactness at A ⊗ B another way in a moment.
129
7.2. Exactness of Sequences and Functors Chapter 7. Category Theory
X
To show 1 ⊗ p is onto, let ai ⊗ b00i ∈ A ⊗ B 00 . By exactness of the original sequence at
B 00 , for each b00i ∈ B 00 there exists some bi ∈ B such that p(bi ) = b00i . Then
X X X
(1 ⊗ p) ai ⊗ b i = ai ⊗ p(bi ) = ai ⊗ b00i .
Hence 1 ⊗ p is onto.
Finally, notice that if we prove that (A ⊗ B)/ im(1 ⊗ i) is a solution to the universal
mapping property in the definition of A ⊗ B 00 then this will show that (A ⊗ B)/ im(1 ⊗ i) ∼
=
00
A ⊗ B which implies exactness at A ⊗ B. Given a biadditive map ϕ : A × B → G, we need
to produce a biadditive map ψ and a linear map ϕ̄ making the following commute:
A × B 00
ϕ
ψ
(A ⊗ B)/ im(1 ⊗ i) G
ϕ̄
Define ψ(a, b00 ) = (a ⊗ b) + im(1 ⊗ i) where p(b) = b00 . We need to be carefull about defining
maps out of A × B 00 . In particular, if p(b1 ) = p(b2 ), we must show that (a ⊗ b1 ) + im(1 ⊗ i) =
(a⊗b2 )+im(1⊗i). But p(b1 ) = p(b2 ) implies there exists some b0 ∈ B 0 such that i(b0 ) = b1 −b2 .
Tensoring with a yields
(a ⊗ b1 ) − (a ⊗ b2 ) = a ⊗ (b1 − b2 ) = a ⊗ i(b0 )
which shows that ker Φ = im(1 ⊗ i). Therefore we may define the quotient map out of A ⊗ B:
ϕ̄ : (A ⊗ B)/ im(1 ⊗ i) −→ G
which inherits linearity from Φ. This completes the proof of right exactness.
Example 7.2.7. For an integer n > 0,
n
0→Z→
− Z → Z/nZ → 0
130
7.2. Exactness of Sequences and Functors Chapter 7. Category Theory
is a short exact sequence. Tensoring with the Z-module Z/nZ gives a sequence
n
Z/nZ →
− Z/nZ → Z/nZ ⊗ Z/nZ → 0.
However, multiplication by n is the zero map on Z/nZ, so this shows that − ⊗ Z/nZ is not
left exact. As a bonus, the tensored exact sequence splits into
So Z/nZ ⊗ Z/nZ ∼
= Z/nZ as abelian groups.
Definition. A short exact sequence of R-modules
f g
0→A→
− B→
− C→0
is split, then B ∼
= A ⊕ C.
Proof. We will show that B = im f ⊕ im h where h : C → B satisfies gh = 1C . If b ∈ B,
then g(b) ∈ C and b − h(g(b)) ∈ ker g. To see this, note that
Lemma 7.2.9 (3×3 Lemma). Consider the following commutative diagram in R-Mod having
exact columns.
131
7.2. Exactness of Sequences and Functors Chapter 7. Category Theory
0 0 0
0 A0 A A00 0
0 B0 B B 00 0
0 C0 C C 00 0
0 0 0
If the bottom two rows are exact, the top row is exact, and if the top two rows are exact, the
bottom row is exact.
Proof. Picking a00 ∈ A00 , let b00 be its image in B 00 . Then b00 must map to zero by exactness
of the third column. Thus there is a b ∈ B mapping to b00 since the map is surjective; let c
be the image of b in C. By commutativity, c 7→ 0. Therefore there is some c0 ∈ C 0 mapping
to c, using exactness of the third row.
Now by surjectivity of B 0 → C 0 , there is some b0 ∈ B 0 mapping to c0 ; denote its image in
B by b̄. By exactness, this b̄ maps to 0 in B 00 . This further implies that b − b̄ 7→ b00 . Since
the lower left square commutes, b̄ 7→ c =⇒ b − b̄ 7→ 0 =⇒ for some a ∈ A, a 7→ b − b̄. Let
a map to ā00 ∈ A00 . Then since A00 ,→ B 00 is injective, ā00 = a00 so it is guaranteed that there
will be an a mapping to a00 . Hence A → A00 is surjective.
Next, take some a0 ∈ A0 which maps to 0 in A. Then a0 7→ b0 ∈ B 0 and by commutativity
around the square, b0 7→ 0 ∈ B. But A0 ,→ B 0 is injective, so a0 = 0 and A0 → A is seen to
be injective.
Now pick another a0 ∈ A0 with image a ∈ A and subsequently a 7→ x ∈ A00 . We need to
show x = 0. Well a0 maps to some b0 ∈ B 0 and b0 7→ b ∈ B, so by commutativity a 7→ b. Since
b is in the image of B 0 → B and the second row is exact, b 7→ 0 ∈ B 00 . By commutativity of
the upper right square, x 7→ 0 ∈ B 00 . But since this map is injective, this forces x = 0.
We have thus shown that the top row is exact. The second statement is proven similarly.
The proof above is an example of a variety of proofs which are common to homological
algebra, called diagram chasing. Another example of this is
Lemma 7.2.10. Consider the following commutative diagram in R-Mod having exact rows
and columns.
132
7.2. Exactness of Sequences and Functors Chapter 7. Category Theory
A0 A A00 0
B0 B B 00 0
C0 C C 00 0
0 0 0
Proof. First suppose A00 ,→ B 00 and B 0 ,→ B are injections. The following diagram serves as
a useful blueprint for the diagram chase to follow.
A0 (7) A A00 0
0 a 00
a a =0 (6)
(8)
(4)
0
β
B0 B B 00 0
=
0 0
(9) b (3)
b (5)
(2)
C0 (1) C C 00 0
0 0
c
0 0 0
133
7.2. Exactness of Sequences and Functors Chapter 7. Category Theory
0
α0 α
0 A A A00 0
f g h
0
β β
0 B0 B B 00 0
Proof. To begin, note that the left square commutes so α0 restricts to α0 : ker f → ker g.
Likewise α restricts to α : ker g → ker h and these both inherit injectivity. Consider the
β0
projection B 0 − → B → B/ im g = coker g. For any b0 ∈ im f , there is some a0 ∈ A0 such that
f (a0 ) = b0 , but also gα0 (a0 ) = β 0 (b0 ) by commutativity. So β 0 (b0 ) ∈ im g which induces a map
β̄ 0 : coker f → coker g – in other words β 0 factors through f . Likewise, β induces a map β̄
and both β̄ 0 and β̄ inherit surjectivity. We pause to label the sequence:
α0 α ð β̄ 0 β̄
0 → ker f −
→ ker g −
→ ker h →
− coker f −
→ coker g →
− coker h → 0.
The interesting part of the sequence is at ker h → coker f ; we have labelled this with the
character ð. How can we define ð? Start with a00 ∈ ker h, so that h(a00 ) = 0. By exactness,
there is some a ∈ A such that α(a) = a00 . If b is the image of a under g, b 7→ 0 in B 00 and so
there is some b0 ∈ B 0 such that b0 7→ b 7→ 0.
With this notation, we define
ð : ker h −→ coker f
a00 7−→ b0 + im f.
Since all maps in this diagram are linear, it suffices to check well-definedness. Once we do
so, we will have all maps defined and can then proceed to check exactness.
Suppose we have α(a1 ) = α(a2 ) = a00 . Then there is some a0 ∈ A0 such that α0 (a0 ) = a1 −a2
since this difference lies in ker α. Let b01 and b02 be the unique (by injectivity) elements of B 0
134
7.2. Exactness of Sequences and Functors Chapter 7. Category Theory
such that β 0 (b01 ) = g(a1 ) and β 0 (b02 ) = g(a2 ). We want to show that b01 − b02 = f (a0 ). But this
is easily seen, since commutativity of the left square gives us a0 7→ a1 − a2 7→ g(a1 ) − g(a2 ) :=
b1 − b2 around the top, and a0 7→ f (a0 ) 7→ b1 − b2 around the bottom. Then by injectivity,
f (a0 ) must equal b01 − b02 as desired. This shows that ð maps into B 0 / im f = coker f , i.e. it
is well-defined.
We now have all our maps, so we can proceed to show exactness. First α0 α = 0 implies
im(α0 |ker f ) ⊆ ker(α |ker g ). For the other containment, take a ∈ ker α that maps to 0 under
g. Then by exactness, there exists an a0 7→ a but the left square commutes, so a0 7→ b0 7→ 0
and injectivity of β 0 shows b0 = 0.
Next suppose b ∈ coker g maps to b00 ∈ coker h = B/ im h. Then there is some a00 mapping
to b00 under h, and α is onto to there also exists a ∈ A mapping to a00 . Let g(a) = b̃ ∈ B;
by commutativity b̃ 7→ b00 . Then b̃ − b 7→ 0 ∈ B 00 implies there exists b0 ∈ B 0 such that
β 0 (b0 ) = b̃ − b = g(a) − b. So b = β 0 (b0 ) − g(a) =⇒ β̄ 0 (b0 + im g) = β 0 (b0 ) + im g = b + im g.
The hard part is showing exactness around the connecting map ð. For a00 ∈ ker h, we can
lift back to some a ∈ A which maps to a00 ∈ A00 and has image g(a) = b. But b 7→ 0 under
β so there is some b0 ∈ B 0 such that b0 7→ b. Recall that we defined ð(a00 ) = b0 + im f . Then
im(α |ker g ) ⊂ ker ð because g(a) = 0. On the other hand, suppose ð(a00 ) = 0. Then in the
diagram chase for the definition of ð, choose a0 ∈ A0 such that f (a0 ) = b0 . Set ã = α0 (a0 ). By
commutativity, g(ã − a) = 0 and by exactness, α(ã − a) = α(a) = a00 .
Finally let b0 ∈ B 0 such that β̄ 0 (b0 + im f ) = im g, so β 0 (b0 ) ∈ im g. The fact that
im ð ⊂ ker β̄ 0 is just by definition of the connecting map. Conversely, suppose b0 7→ b such
that there is some a ∈ A with g(a) = b. If we set α(a) = a00 , we have already constructed
the connecting map for a00 . Moreover we know b0 7→ b 7→ 0 so a00 must map to 0 along h, i.e.
a00 ∈ ker h. This shows exactness at every part of the sequence.
Lemma 7.2.12 (Five Lemma). Suppose the following diagram is commutative with exact
rows.
A1 A2 A3 A4 A5
f1 f2 f3 f4 f5
B1 B2 B3 B4 B5
135
Chapter 8
Special Modules
136
8.1. Projective Modules Chapter 8. Special Modules
Definition. A left R-module P is projective if for every diagram with exact rows and solid
arrows, shown below, there is a dashed arrow making the diagram commute.
M N 0
In other words, projectives allow you to lift along surjections. We will prove
Theorem 8.1.1. Over a ring R, a left R-module P is projective if and only if it is a direct
summand of a free R-module.
A long-standing open question, proposed by Serre and proven by Quillen and Suslin, is:
If P is a projective module over k[x1 , . . . , xn ] then must P be free?
We now show that the earlier definition of projectives agrees with our definition in this
chapter.
To get the 0 on the right, note that by definition of projective, every f : P → M 00 factors
through α : M → M 00 :
P
f˜
f
α
M 00 M 0
137
8.1. Projective Modules Chapter 8. Special Modules
0 −→ M −→ N −→ P −→ 0.
0 −→ K −→ F −→ P −→ 0
where F is free and the map F → R takes a generating set {fi } of F to a generating set
{pi } of P . By hypothesis the sequence splits, so F ∼
= P ⊕ K, and by the earlier (unproven)
characterization of projectives, P is projective.
138
8.1. Projective Modules Chapter 8. Special Modules
In the above proof we used the fact that direct summands of free modules are projective.
We will now prove an even stronger result which implies this.
Theorem 8.1.7. In the category of R-modules for a ring R,
(1) Direct summands of projective modules are projective.
π i
f˜ P0
g̃
g
M N 0
f
Given g surjective, we want to construct a map g̃ making the smaller triangle commute.
Note that P = P 0 ⊕ P 00 induces π : P → P 0 the natural projection and i : P 0 ,→ P the
inclusion map. Since P is projective, there exists a map f˜ : P → M making the larger
triangle commute. It suffices to show that g̃ = f˜i is what we are looking for, but this is
obvious since for any p0 ∈ P 0 ,
Hence P 0 is projective.
(2) If P 0 and P 00 are projective then HomR (P 0 , −) and HomR (P 00 , −) are exact functors.
Thus we have a natural isomorphism of functors
Hom(P 0 ⊕ P 00 , −) ∼
= Hom(P 0 , −) ⊕ Hom(P 00 , −).
Since direct sums of exact functors are exact, this shows Hom(P 0 ⊕ P 00 , −) is exact and
therefore P 0 ⊕ P 00 is projective. In the infinite case, note that
!
Ai , B ∼
M Y
Hom = Hom(Ai , B).
i∈I i∈I
Then the result follows from the fact that direct products of exact functors are exact.
Corollary 8.1.8.
(a) Every module over a field is projective.
139
8.1. Projective Modules Chapter 8. Special Modules
Proof. (a) A module over a field is a vector space which is free, hence projective by Propo-
sition 8.1.4.
(b) For modules over a PID, every submodule of a free module is free.
Projective modules have many desirable properties.
Proposition 8.1.9. Every projective left R-module P has a free complement, that is, a free
left R-module F such that P ⊕ F is free.
Proof. Since P is projective, it’s the direct summand of a free module, i.e. there exists a Q
so that P ⊕ Q is free. The trick is to consider
F = Q ⊕ P ⊕ Q ⊕ ···
which is free since each pair in the sum is free. But notice that P ⊕ F = P ⊕ Q ⊕ P ⊕ Q ⊕ · · ·
is free for the same reasons; hence P has a free complement.
Proof. Since P, Q are projective, there exist free modules F1 , F2 and modules G1 , G2 such
that
F1 = P ⊕ G1 and F2 = Q ⊕ G2 .
Tensoring these two expressions gives us
F1 ⊗ F2 = (P ⊕ G1 ) ⊗ (Q ⊕ G2 ) = (P ⊗ Q) ⊕ (P ⊗ G2 ) ⊕ (G1 ⊗ Q) ⊕ (G1 ⊕ G2 ).
Hence it suffices to prove F1 ⊗ F2 is free which will imply P ⊗ Q is projective. This is easily
seen from the fact that tensor distributes across direct sums (we already used this above):
! !
M M M
F1 ⊗ F2 = R ⊗ R = (R ⊗ R).
i j i,j
Recall that a left R-module M is finitely generated if M is the quotient of a free left
R-module F ∼
= Rn for n < ∞. We may then write M ∼ = F/K where K is a submodule of F .
Definition. A left R-module M is finitely presented if it is finitely generated and M ∼
=
F/K where K is finitely generated.
0 −→ K −→ Rn −→ M −→ 0.
Rm −→ Rn −→ M −→ 0.
140
8.1. Projective Modules Chapter 8. Special Modules
Proposition 8.1.11. Every finitely generated projective left R-module P is finitely presented.
0 −→ K −→ F −→ M −→ 0
141
8.2. Injective Modules Chapter 8. Special Modules
Definition. A left R-module E is injective if for every diagram with exact rows and solid
arrows, shown below, there is a dashed arrow making the diagram commute.
0 A B
Notice the difference between the diagrams for projectives and injectives. We remarked
that projectives allow you to lift along surjections. Similarly, injectives allow you to extend
injections.
In Section 7.2 we defined injectives in terms of exactness of the contravariant Hom functor.
We now prove that these two definitions coincide.
is exact. We proved in Proposition 7.2.2 that HomR (−, E) is always left exact, so it remains
to show exactness at HomR (A, E). In other words we must prove i∗ is surjective if and
only if i is injective. On one hand, if f ∈ Hom(A, E) there exists g ∈ Hom(B, E) with
f = i∗ (g) = gi; that is,
E
f g
0 A B
i
Proposition 8.2.2. E is injective ⇐⇒ every short exact sequence beginning with E splits.
142
8.2. Injective Modules Chapter 8. Special Modules
id g
0 E B
i
Since E is injective, the identity on E induces g : B → E such that gi = idE . Hence the
sequence splits.
( ⇒= ) Suppose E → M → M 00 → 0 is exact, so that by hypothesis M ∼ = E ⊕ M 00 .
We will see shortly that direct summands of injectives are injective, so it follows that E is
injective.
There is a useful characterization of injectives that allows us to restrict our attention to
ideals of R, rather than all module injections A ,→ B. Before stating and proving Baer’s
Criterion, we need
Zorn’s Lemma. Let X be a poset. If every chain in X has an upper bound in X then X
has maximal elements.
Theorem 8.2.3 (Baer’s Criterion). Let E be a left R-module. E is left injective ⇐⇒ every
R-linear map f : I → E, where I is an ideal of R, can be extended to R:
f g
0 I R
i
0 A B
i
143
8.2. Injective Modules Chapter 8. Special Modules
h h∗
0 I R
i
Let A1 = A0 + Rb and define g1 (a0 + rb) = g0 (a0 ) + rh∗ (1). Note that g1 may not be well-
defined since a0 + rb is not necessarily the unique way to write an element of A1 . However,
if g1 is well-defined, we will have shown (A0 , g0 ) ≺ (A1 , g1 ) contradicting maximality of
(A0 , g0 ). Thus to finish the proof we show well-definedness. Suppose a0 + rb = a00 + r0 b.
Then (r − r0 )b = a00 − a0 ∈ A =⇒ r − r0 ∈ I. This gives us
Now we can’t take the constant r − r0 out of h, since this would leave 1 behind, and 1 6∈ I.
But we do have h(r − r0 ) = h∗ (r − r0 ) and h∗ is defined on R, so h∗ (r − r0 ) = (r − r0 )h∗ (1).
It follows easily that g1 is well-defined.
We also have
Proposition 8.2.4. Direct sums and direct summands of injectives are injective.
Example 8.2.5. Q is a divisible Z-module. In fact, this holds for the field of fractions of
any domain.
144
8.2. Injective Modules Chapter 8. Special Modules
A y
x -
↑ 1
n
0 nZ Z
Proof. (2) follows from (1), while the proof of (1) is similar to the first part of the proof of
Baer’s Criterion (8.2.3).
Our next goal is to show that every left R-module can be realized as a submodule of an
injective left R-module. We begin by proving this for Z-modules (abelian groups). Let M
be a Z-module. Then M ∼ = F/S where F is some free abelian group
M and S is L the module
of relations. By the fundamental theorem of abelian groups, M ∼= Z/S and Z can be
L i∈I
embedded in Q so we have a composition
M∼
M M
= Z/S ,→ Q/S.
i∈I i∈I
L L
Now Q is divisible so Q is also divisible. Then by Theorem 8.2.7, Q/S is injective.
Before proving the property for R-modules in general, we will need
Proof. First let’s convince ourselves that HomR (S, E) is a left S-module at all. The map
ϕ : R → S induces a left R-action on S given by r · s = ϕ(r)s. S is also a right module over
itself, so there is an available action for S to act on HomR (S, E) on the left.
Now, note that HomR (S, E) is an injective left S-module ⇐⇒ HomS (−, HomR (S, E))
is an exact functor. By Hom-Tensor Adjointness,
and HomR (−, E) is exact precisely when E is injective. In fact we have proven that the
converse of the proposition holds as well.
145
8.2. Injective Modules Chapter 8. Special Modules
Corollary 8.2.10. For any divisible abelian group D, HomZ (R, D) is an injective left R-
module.
Definition. For an abelian group A, the group A∗ := HomZ (A, Q/Z) is called the dual of
A.
Theorem 8.2.11. Let A be an abelian group and A∗ its dual. Then
(1) If A is a left (resp. right) R-module, then A∗ is a right (resp. left) R-module.
(2) Given a surjection (resp. injection) of R-modules A → B, the induced map B ∗ → A∗
is injective (resp. surjective).
(3) If P is projective, P ∗ is injective. Likewise, if E is injective then E ∗ is projective.
One might hope for a “duality theorem” that says A ∼ = A∗ for any abelian group A.
Unfortunately, this is false in general. We will however prove that A embeds into A∗∗ .
Example 8.2.12. If A is finite, then A ∼ = A∗ . Indeed, since Hom splits over finite direct
sums, it suffices to consider A = Z/nZ, in which case the map
is an isomorphism. Hence there is a canonical isomorphism between A and its double dual:
On the other hand, if A = Z, we have Z∗ = Q/Z so that Z∗∗ = HomZ (Q/Z, Q/Z). One
can show that HomZ (Q/Z, Q/Z) ∼ b the profinite completion of Z. There is a natural
= Z,
embedding Z ,→ Z.
b
This shows we can embed a module into some injective module, but HomZ (R, D) is often
way bigger than necessary. The natural follow-up question is: What is the smallest injective
E such that M is a submodule of E? To answer this, first we need
146
8.2. Injective Modules Chapter 8. Special Modules
Examples.
1 If R is a domain, the field of fractions Q of R is the injective hull of R.
2 Let R = k[x1 , . . . , xn ] and consider the ideal m = (x1 , . . . , xn ). Let E(R/m) denote the
injective hull of R/m. E(R/m) has a k-basis given by monomials x−k 1
1
· · · x−k
n
n
where
ki ≥ 0. Further, R acts on E(R/m) via
(
rx1−k1 · · · x−k n
if for all i, powers of xi in rx−k1
· · · x−k n
are ≤ 0
r · (x−k
1
1
· · · x −kn
n ) = n 1 n
0 otherwise.
147
8.2. Injective Modules Chapter 8. Special Modules
R/m ,→ E(R/m)
1̄ 7→ 1
148
8.3. Flat Modules Chapter 8. Special Modules
∼
= ∼
=
R ⊗R A R ⊗R B
1A ⊗ i
showing 1A ⊗ i is injective.
Proposition 8.3.2. For a ring R,
(i) Every projective right (or left) R-module is flat.
(ii) A direct sum of right (or left) R-modules is flat if and only if each summand is flat.
Proof. (i) follows from (ii) since a free right R-module is the direct sum of copies of R, which
is flat. Moreover, P is projective if and only if it is the direct summand of a free module,
which, combined with the first statement Msays that projectives are always flat.
To prove (ii), consider a direct sum Mk of right R-modules Mk . For any family of R-
k M M
linear maps {fk : Mk → Nk } there is a map f : Mk → Nk taking (mk ) 7→ (fk (mk )),
k k
and clearly f is injective if and only if each fk is injective. So if i : A → B is a left R-module
injection, there is a commutative diagram
L 1⊗i L
( Mk ) ⊗R A ( Mk ) ⊗R B
∼
= ∼
=
L L
(Mk ⊗R A) (Mk ⊗R B)
f
149
8.3. Flat Modules Chapter 8. Special Modules
(a) M is flat.
Corollary 8.3.4. If every finitely generated submodule of a right R-module M is flat, then
M is flat.
Example 8.3.5. For a field k, let R = k[x, y]. Then R is flat as a module over itself (it is
free) and the ideal M = (x, y) is a finitely generated submodule of R, but M is not flat.
Proof. Let Q be the field of fractions of R. Since A is flat, the functor − ⊗R A is exact, so
the exact sequence 0 → R → Q induces an exact sequence
0 → R ⊗R A → Q ⊗R A.
Proof. The theory of modules over a PID says that every finitely generated R-module M
that is torsion-free is also free. Thus every finitely generated submodule M ⊆ B is free,
hence projective, hence flat. By Corollary 8.3.4, this implies that B is also flat.
Example 8.3.8. For any multiplicative set S ⊆ R, the localization S −1 R is a flat module.
Definition. A left R-module M is faithfully flat provided that for all left R-modules A
and B, the sequence 0 → A → B is exact if and only if 0 → A ⊗R M → B ⊗R M is exact.
One can define faithful flatness similarly for right modules.
150
8.3. Flat Modules Chapter 8. Special Modules
Lemma 8.3.10. A left R-module M is faithfully flat if and only if M is flat and for all left
R-modules A, A ⊗R M = 0 implies A = 0.
Example 8.3.12. Any localization S −1 R of a ring R is flat by Example 8.3.8, but is not
faithfully flat in general. For example, if R is an integral domain with field of fractions K –
the localization of R at the multiplicative set R r {0} – then for any nonzero torsion module
M , M ⊗R K = 0.
Example 8.3.13. In general, projective modules are not faithfully flat. For instance, sup-
pose R has a nontrivial idempotent element e ∈ R, that is, e2 = e. Then e(1 − e) = 0 and
1 − e is also an idempotent, and we have R ∼ = eR ⊗ (1 − e)R. So eR is a projective R-module,
but it is not faithfully flat since (1 − e)R ⊗R eR = (1 − e)R ⊗R e2 R = e(1 − e)R ⊗R eR = 0.
Theorem 8.3.14. Let R be a commutative ring and M an R-module. Then the following
are equivalent:
(a) M is flat.
(b) For all prime ideals p ⊂ R, the localization Mp is flat as an R-module (equivalently,
as an Rp -module).
(c) For all maximal ideals m ⊂ R, the localization Mm is flat as an R-module (equivalently,
as an Rm -module.
Recall that an R-module M is finitely presented if there exists a short exact sequence
0 → K → Rn → M → 0 with n finite.
f g id
0 L Rm M 0
151
8.3. Flat Modules Chapter 8. Special Modules
Then since Rm is free, we get a map g : Rn → Rm making the right square commute. Call
f the restriction of g to K, which gives a commutative square on the left. By the Snake
∼
Lemma, 0 = ker id → coker f − → coker g → coker id = 0 is exact, so the middle arrow is
an isomorphism. Now coker g is finitely generated as it is a quotient of Rm , so this implies
coker f = L/f (K) is finitely generated. Since K was finitely generated, f (K) is also finitely
generated and it now follows that L is finitely generated.
We now relate finite presentability to flatness. First, we need a lemma.
(c) Suppose F is free. For any finite set of elements {n1 , . . . , nr } ⊆ N , there exists a
morphism f : F → N such that f (ni ) = ni for all 1 ≤ i ≤ n.
I ⊗R N → I ⊗R F → I ⊗R M → 0.
ψ ∼
= ∼
=
0 N ∩ IF IF IM 0
152
8.3. Flat Modules Chapter 8. Special Modules
To induct, assume the statement holds for all finite subsets of N consisting of r − 1 elements.
For a given {n1 , . . . , nr } ⊆ N , the inductive hypothesis constructs for us two maps f1 , f2 :
F → N satisfying
f1 (n1 ) = n1 and f2 (ni − f1 (ni )) = ni − f1 (ni ) for all 2 ≤ i ≤ r.
Set f = f1 + f2 − f2 ◦ f1 , where f2 ◦ f1 is defined by changing targets, i.e. considering
f1 : F → N ⊆ F . Then f (n1 ) = f1 (n1 ) = n1 and for each 2 ≤ i ≤ r,
f (ni ) = f1 (ni ) + f2 (ni − f1 (ni )) = f1 (ni ) + ni − f1 (ni ) = ni .
Therefore the statement holds for all r by induction.
Theorem 8.3.17. For a left R-module M , the following are equivalent:
(1) M is finitely generated and projective.
(2) M is finitely presented and flat.
(3) M is finitely presented and for all maximal ideals m ⊂ R, the localization Mm is free.
Proof. (1) =⇒ (2) is Propositions 8.3.15 and 8.3.2.
(2) =⇒ (1) If M is finitely presented, then there exists a short exact sequence 0 →
N → F → M → 0 where F is finitely generated and free and N is finitely generated, say by
n1 , . . . , nr ∈ N . Then since M is flat, Lemma 8.3.16 says there is a morphism f : F → N
with f (ni ) = ni for all i, but since the ni generate N , this means f |N = idN . Hence the short
exact sequence splits, so in particular M is a direct summand of F and thus projective.
(1) =⇒ (3) By the above, M is finitely presented (and flat), so there exists a presentation
Rm → Rn → M → 0 with m, n finite. Flatness implies that tensoring with Rm for any
maximal ideal m yields an exact sequence of Rm -modules
m n
Rm → Rm → Mm → 0.
Therefore Mm is finitely presented. Theorem 8.3.14 also shows Mm is flat. By finite presenta-
tion, Mm /m is a finite dimensional vector space over the residue field κ(m) = Rm /m. Choose
m /m which are the images of x1 , . . . , xr ∈ Mm . Then by Nakayama’s
a basis x̄1 , . . . , x̄r of MP
Lemma (1.2.1), Mm = ri=1 Rm xi . Thus Mm has a finite presentation of Rm -modules
N → F → Mm → 0
where F is free with basis y1 , . . . , yr whose P
images in Mm are x1 , . . . , xP
r . Suppose n =
P r r r
i=1 ri yi ∈ N . By exactness, n maps to i=1 ri xi = 0 in Mm , so i=1 r̄i x̄i = 0 in
Mm /m. Since the x̄i are k(m)-linearly independent, this implies each Prr̄i = 0, that is, ri
lies in ker(Rm → κ(m)), and so ri ∈ m. Hence we see that n = i=1 ri yi ∈ mF . By
Lemma 8.3.16(a), N = N ∩ mF = mF , so by Nakayama’s Lemma (1.2.1), N = 0. This
shows Mm ∼ = F.
(3) =⇒ (1) will be proven later.
153
8.3. Flat Modules Chapter 8. Special Modules
is faithfully flat. Indeed, M is flat since it is a direct sum of flat R-modules (use Proposi-
tion 8.3.2(ii) and Theorem 8.3.14). On the other hand, suppose N ⊗R M = 0. Then
!
M M M
0 = N ⊗R Rm = (N ⊗R Rm ) = Nm .
m m m
ϕ ∼
= ∼
=
0 HomS −1 R (S −1 M, S −1 N ) (S −1 N )n (S −1 N )m
154
8.3. Flat Modules Chapter 8. Special Modules
ModB −→ ModA
N 7−→ N, with a · n = f (a)n,
and ModA −→ ModB
M 7−→ MB := B ⊗A M.
Proof. This follows from the fact that there is a canonical isomorphism of S −1 A-modules
T −1 B ∼
= S −1 A ⊗A B.
On the other hand, we have:
A/I ,→ A/I ⊗A B ∼
= B/IB.
In general, the kernel of this map is (IB∩A)/I but because the map is injective, (IB∩A)/I =
0, that is, IB ∩ A = I.
155
8.3. Flat Modules Chapter 8. Special Modules
(c) Take a prime ideal p ⊂ A. We want to find a prime ideal q ⊂ B with p = q ∩ A. Let
Ap be the localization at p, which is a local ring with maximal ideal pAp . Since A → B is
faithfully flat, Corollary 8.3.22 shows that Ap → Bp is also faithfully flat, and hence injective
by (a). Applying (b) to the ideal I = pBp ⊂ Bp , we get pBp ∩ Ap = pAp 6= Ap , so pBp 6= Bp .
Thus there is some maximal ideal m ⊂ Bp containing pBp . Now m ∩ Ap ⊇ pAp but 1 6∈ m, so
m ∩ Ap is a proper ideal of Ap containing its unique maximal ideal pAp ; hence m ∩ Ap = pAp .
By Proposition 1.3.8, there is a prime ideal q ⊂ B extending to m ⊂ Bp , and by construction
q ∩ A = pAp ∩ A = p.
Remark. It turns out that f : A → B is faithfully flat if and only if f is flat and f ∗ :
Spec B → Spec A is surjective.
Theorem 8.3.26 (Generic Faithful Flatness). Let A be an integral domain with field of
fractions K, B/A a finitely generated ring extension and assume B ,→ K ⊗A B is injective.
Then there exist nonzero elements a ∈ A and b ∈ B such that the induced morphism Aa → Bb
is faithfully flat.
Proof. Suppose B = A[b1 , . . . , bn ] with bi ∈ B. By Noether’s normalization lemma (Theo-
rem 2.4.1), there exist elements x1 , . . . , xm ∈ BK = K ⊗A B such that BK /K[x1 , . . . , xm ] is
an integral extension. After multiplying through by a common denominator, we may assume
x1 , . . . , xm ∈ B. In particular, b1 , . . . , bn are integral over K[x1 , . . . , xm ], so they are integral
over Aa0 [x1 , . . . , xm ] for some nonzero element a0 ∈ A. This implies Ba0 /Aa0 [x1 , . . . , xm ] is a
finite ring extension. Thus we may replace A with Aa0 and B with Bb0 to have a finite ring
extension B/A[x1 , . . . , xm ].
Set C = A[x1 , . . . , xm ], E = K(x1 , . . . , xm ) and consider the diagram
B BK E ⊗C B
C K[x1 , . . . , xm ] E
ϕ : Cqr −→ Bq
r
X
(c1 , . . . , cr ) 7−→ cj zj
j=1
156
8.3. Flat Modules Chapter 8. Special Modules
Am ⊗A A
b An ⊗A A
b M ⊗A A
b 0
∼ ∼ α
bm
A bn
A M
c 0
with exact rows. Then by the five lemma, there exists a map α which fills in the diagram
and is an isomorphism.
Corollary 8.3.28. Completion is flat, i.e. for any local noetherian ring A, the natural map
A→A b is flat.
The following is an analogue of the going down theorem (2.3.10) for flat morphisms.
Theorem 8.3.29. Let ϕ : A → B be a flat morphism of noetherian rings. Suppose q ⊂ B
is a prime ideal, p = ϕ−1 (q) and there is a chain of prime ideals
p ) p1 ) p2 ) · · · ) pn
in A. Then there exists a chain of prime ideals
q ) q1 ) q2 ) · · · ) qn
in B such that ϕ−1 (qi ) = pi for all 1 ≤ i ≤ n.
157
8.3. Flat Modules Chapter 8. Special Modules
158
8.3. Flat Modules Chapter 8. Special Modules
Corollary 8.3.32. If A is a noetherian ring, then dim A[x1 , . . . , xn ] = dim A + n for every
n ≥ 1.
159
Chapter 9
Categorical Constructions
160
9.1. Products and Coproducts Chapter 9. Categorical Constructions
Definition. The union A0 ∪ B 0 is called the formal disjoint union of A and B. We denote
this by A t B.
There are two functions associated with this construction, α : A → A0 ∪ B 0 which takes
a 7→ (a, 0), and β : B → A0 ∪ B 0 which maps b 7→ (b, 1). In fact, given functions f : A → X
and g : B → X, there is a unique function θ : A t B → X that extends both f and g:
(
f (a) if u = (a, 0) ∈ A0
θ(u) =
f (b) if u = (b, 1) ∈ B 0 .
This is a well-defined map since A0 ∩ B 0 = ∅. This describes disjoint unions in terms that
are easily compatible with categories; notice that we didn’t use an element at all.
A
f
α
θ
AtB X
β
g
B
This shows that the coproduct is a solution to a universal mapping problem. Below we
see that some familiar objects together with their most closely-associated morphisms can be
realized as coproducts.
Examples.
1 We showed above that in the category Sets, the coproduct of two sets A and B is their
disjoint union.
2 In the category R-Mod of left R-modules, two modules A and B have a coproduct: their
direct sum A ⊕ B together with the natural injections α : a 7→ (a, 0) and β : b 7→ (0, b).
161
9.1. Products and Coproducts Chapter 9. Categorical Constructions
3 Let k be a commutative ring and suppose A and B are commutative k-algebras. Then
the tensor product A⊗k B is their coproduct in the category of commutative k-algebras.
4 Coproducts may not exist in every category. For example, let X be a set and define C
to be the category of objects as subsets of X and morphisms as inclusion maps between
them. For an object A ( X, its complement is Ac = X r A. Then A and Ac do not
have a coproduct.
Examples.
7 Not all categories have initial objects. For example, the the partially-ordered set of
integers Z does not have an initial object.
Proposition 9.1.1. If A and A0 are initial objects in a category C, then they are isomorphic
and hence initial objects, if they exist, are unique.
Proof. By definition there is only one morphism f : A → A0 and only one in the other
direction, g : A0 → A. It’s easy to see that gf = idA and f g = idB , so f and g are
isomorphisms.
A
f
p
θ
X AtB
β
g
B
Examples.
162
9.1. Products and Coproducts Chapter 9. Categorical Constructions
8 For two sets A, B ∈ Sets, their product is the regular cartesian product A×B, together
with the usual coordinate projections.
9 If R is a ring and A and B are left R-modules, their product is the direct product
A ⊕ B. Thus the product and coproduct coincide in R-Mod.
10 In Groups, the coproduct of two groups G and H is called the free product, denoted
G ∗ H, which is related to the construction of free groups. In any case, G ∗ H is distinct
from G × H, which is the product in the category Groups.
11 Recall that in topology, the fundamental group is a functor π1 : Top → Groups. On one
side, the coproduct of two topological spaces X and Y is their wedge, X ∨ Y . On the
other side, the Seifert-van Kampen Theorem tells us that π1 (X ∨ Y ) ∼= π1 (X) ∗ π1 (Y ).
So this functor preserves coproducts.
As we did with initial objects and coproducts, we use terminal products to prove that
products are unique (up to isomorphism).
Proposition 9.1.3. If B and B 0 are terminal objects in a category C, then they are isomor-
phic and hence terminal objects, if they exist, are unique.
Proof. Similar to the proof for initial objects; just reverse the arrows!
163
9.2. Limits and Colimits Chapter 9. Categorical Constructions
Definition. Let F : C → A be an inverse system in A. The limit (also called the projective
or inverse limit) of F is an object lim F ∈ obj(A) such that for all objects C ∈ C, there
are morphisms ϕC : lim F → F(C) making the diagrams
lim F
ϕC ϕC 0
F(C) F(C 0 )
commute whenever HomA (F(C), F(C 0 )) 6= ∅, and such that lim F is universal among all
such objects in A.
Limits are sometimes also written lim F. By the universal property, limits are unique up
←−
to unique isomorphism.
Example 9.2.1. Let C be the category consisting of two objects {1, 2} and only identity
morphisms. An inverse system F : C → A is just defined by specifying two objects, F(1) = a1
and F(2) = a2 . Then lim F is the product of these elements, a1 u a2 .
Definition. Let F : C → A be a direct system in A. The colimit (also called the injective
or direct limit) of F is an object colim F ∈ obj(A) such that for all C ∈ C, there are
morphisms ψC : F(C) → colim F making the diagrams
F(C) F(C 0 )
ψC ψC 0
colim F
commute whenever HomA (F(C), F(C 0 )) 6= ∅, and such that colim F is universal among all
such objects in A.
Example 9.2.2. Let C be the same category as in Example 9.2.1. Then a direct system
F : C → A is once again defined by specifying F(1) = a1 and F(2) = a2 , but colim F is the
coproduct of these elements, a1 t a2 .
164
9.2. Limits and Colimits Chapter 9. Categorical Constructions
There are plenty of other important examples of limits and colimits in category theory.
Definition. Let C be the diagram category
The colimit of a direct system F : C → A is called the pushout of F and the limit of an
inverse system F : C → A is called the pullback of F.
B
Explicitly, if a direct system F has image A then we write the pushout Q =
C
colim F as a square diagram
A B
C Q
B
Similarly, if F is an inverse system with image A then we write the pullback P = lim F
C
as a square diagram
P B
C A
Definition. Suppose A is a category with an object 0 ∈ obj(A) that is both initial and
f B
terminal. For any morphism f : A → B, the pushout of the diagram A is called
0
the cokernel of f , written coker f . Similarly, for such a morphism f , the pullback of the
0
diagram B is called the kernel of f , written ker f .
A f
Example 9.2.3. The category AbGps of abelian groups has the trivial group 0 as a zero
object, i.e. an object that is both initial and terminal. Then for a homomorphism of abelian
groups f : A → B, the kernel and cokernel of f coincide exactly with these notions from
abstract algebra:
ker f = {a ∈ A | f (a) = 0}
and coker f = {[b] | b ∈ B and [b] = [b0 ] if b0 = b + f (a)}.
165
9.2. Limits and Colimits Chapter 9. Categorical Constructions
This also holds in ModR for any ring R, and indeed in any abelian category as we shall see.
In the category ModR , notice that the functor HomR (C, −) does not preserve colimits since
it doesn’t preserve cokernels (this functor is not right exact; see Example 7.2.3). However,
by Proposition 7.2.2, HomR (C, −) does preserve kernels. We will show in general that
this condition is nearly equivalent to the property of preserving all limits. Similarly, recall
(Proposition 7.2.6) that C ⊗R − is a right exact functor, so it preserves cokernels, but does not
preserve kernels. We will also show that the property of right exactness is nearly equivalent
to preserving all colimits.
HomB (F−, −) ∼
= HomA (−, G−).
Example 9.2.4. Let R and S be rings and fix a left R-module M , a left S-module N and
an (S, R)-bimodule L. Then by Theorem 7.2.5, there is a natural isomorphism
HomS (L ⊗R M, N ) ∼
= HomR (M, HomS (L, N )).
Proof. ( =⇒ ) Suppose (F, G) is an adjoint pair. Then for any objects X ∈ A and Y ∈ B,
there is an isomorphism
∼
ΦX,Y : Hom(FX, Y ) −→ Hom(X, GY ).
∼
Applying this to Y = FX, we have ΦF X,F X : Hom(FX, FX) − → Hom(X, GFX) which takes
the identity idF X to some morphism ηX = ΦF X,F X (idF X ) : X → GFX. This defines the
natural transformation η : idA → GF. On the other hand, applying the natural isomorphism
∼
Φ to X = GY gives an isomorphism ΦF GY,Y : Hom(FGY, Y ) − → Hom(GY, GY ) under which
the identity idGY is mapped to by some εY = Φ−1
F GY,Y (idGY ) : FGY → Y . This defines the
natural transformation ε : FG → idB . To check the identity conditions, apply Φ to the
idF GY εY
sequence εY : FGY −−−→ FGY −→ Y to get:
ηGY Gε
GY −−→ GFGY −→ GY.
Thus Gε ◦ ηGY = ΦF GY,Y (εY ) which by definition equals idGY . The proof of the other identity
is similar.
166
9.2. Limits and Colimits Chapter 9. Categorical Constructions
Φ−1 −1
X,Y ΦX,Y (α) = ΦX,Y (G(α) ◦ η) = ε ◦ F(G(α) ◦ η)
= ε ◦ FG(α) ◦ Fη since F is a functor
= εF(G(α))Fη = εF ◦ Fη(α) = α
ΦX,Y Φ−1
X,Y (β) = ΦX,Y (ε ◦ F(β)) = G(ε ◦ F(β)) ◦ η
= Gε ◦ GF(β) ◦ η since G is a functor
= Gε ◦ ηG(β) = β
Definition. For an adjoint pair (F, G), the natural transformations η : idA → GF and
ε : FG → idB are called the unit and counit of the adjunction, respectively. The conditions
Gε ◦ ηG = id and εF ◦ Fη = id are called the triangle identities.
Theorem 9.2.7. If (F, G) is an adjoint pair of functors, then F preserves colimits and G
preserves limits.
HomB (F(X(C)), B) ∼
= HomA (X(C), G(B)).
colim(F ◦ X)
colim X
167
9.2. Limits and Colimits Chapter 9. Categorical Constructions
F ◦ ϕC
F(X(C))
In other words, every map F(X(C)) → B which is compatible with the F ◦ ϕC will factor
through F(colim X). Hence by the universal property of colimits, F(colim X) = colim(F ◦
X). The proof that G preserves limits is dual.
Corollary 9.2.8. For any (S, R)-bimodule L, L ⊗R − preserves colimits and HomS (L, −)
preserves limits.
Theorem 9.2.9. Let F : ModR → AbGps be an additive functor. Then the following are
equivalent:
(1) F ∼
= L ⊗R − for some R-module L.
(2) F preserves colimits.
where X and Y are some sets. Note that coker ϕ = M . Since F preserves direct sums,
F(RY ) = (F(R))Y = LY ∼ = L ⊗R RY and likewise F(RX ) ∼ = L ⊗R RX . Then applying F to
the exact sequence above gives the top row in the following diagram with exact rows:
LY LX F(M ) 0
∼
= ∼
=
L ⊗R R Y L ⊗R RX L ⊗R M 0
Then there is a map F(M ) → L⊗R M which by the Snake Lemma (7.2.11) is an isomorphism.
Naturality is easy to check.
Theorem 9.2.10. Let F : ModR → AbGps be an additive functor. Then the following are
equivalent:
168
9.2. Limits and Colimits Chapter 9. Categorical Constructions
(1) F ∼
= HomR (L, −) for some R-module L.
(2) F preserves limits.
Proof. Similar.
169
9.3. Abelian Categories Chapter 9. Categorical Constructions
HomC (A, B) is an abelian group for all A, B ∈ obj(C). Note that we normally only
require Hom to be a set.
There exists a 0 object, which is both initial and terminal, meaning for all A ∈ obj(C),
there exist maps 0 → A and A → 0.
C has finite products ( ) and coproducts ( ) and they agree on finite-index sets.
Q L
Q L
The last condition, that and agree on finite sets, can be proven using the other
axioms. We can also define additive functors between additive categories, which are
functors that preserve the additive structure of the category.
To fully understand abelian categories, it requires us to redefine our concept of kernels
and cokernels. These are best understood in terms of monics and epics.
f
u
A B C
g
Definition. An epimorphism, or epic, is the dual notion, that is π is an epic if for every
f, g : B → C, f π = gπ implies f = g.
f
π
A B C
g
In general, if u is one-to-one then u is monic, and if π is onto then π is epic, but the
converse does not hold in a general category. The converse does hold however for abelian
categories, which we take to be our definition.
Definition. A category C is abelian if every one-to-one morphism is monic and every onto
morphism is epic.
170
9.3. Abelian Categories Chapter 9. Categorical Constructions
Example 9.3.1. Let Comm be the category of commutative rings, with ring homomorphisms
as morphisms. Consider the map i : Z → Q. This is an epimorphism in C but it is clearly not
f
onto. To see that it is epic, suppose we have Q R so that f i = gi, i.e. f (n) = g(n) for
g
all n ∈ Z. Clearly if f and g agree on Z they agree on Q, so i is an epimorphism. However,
i cannot be onto because Q is much bigger than Z. This shows that Comm is not an abelian
category.
Proof omitted.
From this we define
0
(2) If f : M → N is monic, then M is the kernel of the diagram coker f
N
M
(3) If f is epic, then N is the cokernel of the diagram ker f
0
Note that condition (2) implies that if f is monic, then there exists a short exact sequence
f
0→M →
− N→
− coker f → 0.
Likewise, condition (3) implies that if f is epic, there exists a short exact sequence
f
0 → ker f →
− M→
− N → 0.
Example 9.3.3. For any ring R, the category ModR is an abelian category. If R is (left)
noetherian, then the subcategory modR of finitely generated (left) R-modules is also an
abelian category.
Theorem 9.3.4. Any abelian category is naturally isomorphic to a full subcategory of ModR
for some ring R.
171
9.4. Projective and Injective Resolutions Chapter 9. Categorical Constructions
172
9.4. Projective and Injective Resolutions Chapter 9. Categorical Constructions
Lemma 9.4.2. (a) Every module over a field has a projective resolution of length 0.
(b) Every module over a PID has a projective resolution of length at most 1.
Proof. (a) If V is a module over a field, it is a vector space and therefore free and projective.
id
Hence 0 → V − → V → 0 is a projective resolution.
(b) Let P be a projective module with a surjection P → M , so that P is also free by
Corollary 8.1.8. Then the kernel K of this map is a submodule of P , so it’s free and therefore
projective by Corollary 8.1.8 again. Hence 0 → K → P → M → 0 is a projective resolution
of M .
Before proving some results for projective resolutions, we need
Proposition 9.4.4. Let A be a left R-module and suppose we have two exact sequences
i pn pn−1 p2 p1 p0
− Pn −→ Pn−1 −−−→ · · · −
0→K→ → P1 −
→ P0 −
→A→0
and
i0 qn qn−1 q2 q1 q0
0 → K0 −
→ Qn −→ Qn−1 −−→ · · · −
→ Q1 −
→ Q0 −
→ A → 0,
K ⊕ Qn ⊕ Pn−1 ⊕ · · · ∼
= K 0 ⊕ Pn ⊕ Qn−1 ⊕ · · ·
Proof. For all j = 0, . . . , n we will show that there are short exact sequences
173
9.4. Projective and Injective Resolutions Chapter 9. Categorical Constructions
In each case the kernel of the first map is contained in Pk+1 ⊕ Qk (flip these for the bottom
sequence) so the inductive step follows easily.
Now since the original sequence is exact at K, K embeds into Pn via i and therefore
K = ker pn . Hence
K ⊕ Qn ⊕ Pn−1 ⊕ · · · ∼ = K 0 ⊕ Pn ⊕ Qn−1 · · ·
as desired.
Definition. If A is a left R-module and · · · → P2 → P1 → P0 → A → 0 is a projective
resolution of A, we say A has projective dimension n if there is a smallest n such that Kn
is projective. Otherwise A has infinite projective dimension (or no projective dimension).
If A has projective dimension n, we write pd(A) = n. The following result tells us that
the integer n does not depend on the projective resolution chosen for A.
Proposition 9.4.5. If one projective resolution of a module A has a projective nth syzygy,
then the nth syzygy of every projective resolution of A is projective.
Proof. Consider the projective resolutions
P = · · · → Pn → Pn−1 → · · · → P1 → P0 → A → 0
and
P = · · · → Pn0 → Pn−1
0 0
→ · · · → P10 → P00 → A → 0.
Make a soft truncation at the nth location of each to produce exact sequences
0 → Kn → Pn → Pn−1 → · · · → P1 → P0 → A → 0
Then by Schanuel’s Lemma (Proposition 9.4.3), Kn ⊕Pn0 ⊕Pn−1 ⊕· · · ∼ = Kn0 ⊕Pn ⊕Pn−10
⊕· · · .
0 0 0
Let Qn = Pn ⊕ Pn−1 ⊕ · · · and Qn = Pn ⊕ Pn−1 ⊕ · · · . These are both projective since they
are direct sums of projectives, and so we see that Kn ⊕ Q0n ∼ = Kn0 ⊕ Qn . Notice that we
haven’t yet used the hypothesis that some Kn is projective, so this proof works for all n.
Now if Kn is projective, the isomorphism Kn ⊕ Q0n ∼ = Kn0 ⊕ Qn shows that Kn0 is a direct
summand of a projective, so it too is projective. Hence if the nth syzygy in a projective
resolution of A is projective, the nth syzygy is projective in all projective resolutions of A,
reassuring us that the term ‘projective dimension’ is well-defined.
There is a dual notion for injectives.
174
9.4. Projective and Injective Resolutions Chapter 9. Categorical Constructions
Notice that when the indices increases along the arrows, we write them as superscripts,
and when they decrease along arrows we write them as subscripts. This notational convention
will be seen throughout homological algebra.
175
Chapter 10
Homology
176
10.1. Chain Complexes and Homology Chapter 10. Homology
Proof omitted.
∂n+1 ∂n
··· An+1 An An−1 ···
fn+1 fn fn−1
··· Bn+1 0
Bn Bn−1 ···
∂n+1 ∂n
Note that B(A• ) ⊆ Z(A• ). This allows us to define the central object in homological algebra:
Intuitively, the nth homology measures the failure of exactness of a complex at the nth
location in the chain. Since A is abelian, we also have H(A• ) = Z(A• )/B(A• ). Notice that
A• is exact at An ⇐⇒ Hn (A• ) = 0, and the entire complex A• is exact ⇐⇒ H(A• ) = 0.
177
10.1. Chain Complexes and Homology Chapter 10. Homology
Proposition 10.1.2. For every abelian category A, the nth homology Hn : Comp(A) → A is
an additive functor for each n ∈ Z.
Proof. We prove this for A = Ab, the category of abelian groups. The definition above
defines the functor Hn : Comp(Ab) → Ab on objects by A• 7→ Hn (A• ) so it remains to define
homology on morphisms. Let f : C → C 0 be a chain map and define
Hn (f ) : Hn (C) −→ Hn (C 0 )
zn + Bn (C) 7−→ fn (zn ) + Bn (C 0 ).
We must show that fn (zn ) is a cycle and that Hn (f ) is well-defined, i.e. independent of the
choice of cycle zn . Since f is a chain map, we have the following commutative diagram
∂n+1 ∂n
··· Cn+1 Cn Cn−1 ···
fn+1 fn fn−1
0 0
··· Cn+1
0
Cn0 Cn−1 ···
∂n+1 ∂n
showing (by a quick diagram chase) that indeed fn (zn ) ∈ Zn (C 0 ), and if zn + Bn (C) =
wn + Bn (C) then fn (zn − wn ) must land in Bn (C 0 ).
Next we prove Hn is a functor. Obviously Hn (1C ) must be the identity on Hn (C). If f
and g are chain maps such that gf is defined in some Hom set, then for all z ∈ Zn (C) we
have
Hence Hn is a functor.
The first big result we will prove is a statement for short exact sequences of complexes.
We will make extensive use of the Snake Lemma (7.2.11). Note that by the equivalence of
abelian categories and module categories R-Mod for rings R, we are able to state things in
terms of R-modules.
i p
Theorem 10.1.3 (The Long Exact Sequence). Let 0 → C 0 → − C → − C 00 → 0 be an exact
sequence in Comp(R), the category of complexes of left R-modules. Then there is a long exact
sequence in R-Mod
178
10.1. Chain Complexes and Homology Chapter 10. Homology
.. .. ..
. . .
0
in+1 pn+1 00
0 Cn+1 Cn+1 Cn+1 0
0 00
∂n+1 ∂n+1 ∂n+1
in pn
0 Cn0 Cn Cn00 0
∂n0 ∂n ∂n00
0
in−1 pn−1 00
0 Cn−1 Cn−1 Cn−1 0
.. .. ..
. . .
How do we define Hn (i) : Hn (C 0 ) → Hn (C)? In the proof of the Snake Lemma (7.2.11), we
in
proved that Zn (C 0 ) −
→ Zn (C) induces
i
Zn (C 0 ) −
→n
Zn (C) → Hn (C)
so we just need to check that if z ∈ Bn (C 0 ) then in (z) ∈ Bn (C). This follows from the
commutativity of the left squares in the diagram above. The dual argument works for
pn
Hn (C) → Zn (C) −→ Zn (C 00 ). Hence we have well-defined maps Hn (i) : Hn (C 0 ) → Hn (C)
and Hn (p) : Hn (C) → Hn (C 00 ).
The Snake Lemma (7.2.11) is also a good place to consult when defining a connecting
map ð : Hn (C 00 ) → Hn−1 (C 0 ). In that proof, we saw that there is a map
Zn (C 00 ) −→ Cn−1
0
/Bn−1 (C 0 ) = coker ∂n0
z 00 7−→ c0 + im ∂n0 .
0
By the Correspondence Theorem, the submodules of Cn−1 /Bn−1 (C 0 ) are in correspondence
with submodules of Cn−1 containing Bn−1 , and one of these if Zn−1 (C 0 ). Thus we want to
0 0
prove c0 + im ∂n0 = z 0 + im ∂n0 for some z 0 ∈ ker ∂n−1 , which is equivalent to c0 − z 0 ∈ im ∂n0 .
A diagram chase actually reveals that c0 itself lies in ker ∂n−1
0
. So by the lifting property of
kernels, we have a map
Zn (C 00 ) −→ Zn−1 (C 0 )/Bn−1 (C 0 ) = Hn−1 (C 0 )
z 00 7−→ c0 + im ∂n−1
0
.
Another diagram chase shows that if z 00 ∈ Bn (C 00 ) then z 00 7→ c0 ∈ Bn−1 (C 0 ) in the above
map. This induces a map ð : Hn (C 00 ) → Hn−1 (C 0 ).
Now that we have all the maps in hand, it remains to show that the long sequence is
exact at each homology. The proof of this is very similar to the latter part of the proof of
the Snake Lemma (7.2.11).
179
10.1. Chain Complexes and Homology Chapter 10. Homology
Proposition 10.1.4. Let R and A be rings and suppose T : R Mod → A Mod is an exact
additive functor. Then T commutes with homology, that is, for every complex C ∈ Comp(R)
and for every n, there is an isomorphism
Hn (T (C)) ∼
= T (Hn (C)).
Proof. The nth homology Hn (C) defines two exact sequences in R Mod:
Since T is an additive exact functor, applying it to the sequences above gives two new exact
sequences
n ∂ ∂n−1
On the other side of things, expand the complex C = · · · → Cn −→ Cn−1 −−−→ · · · and apply
the exact functor T , giving
T (∂n ) T (∂n−1 )
T (C) = · · · → T (Cn ) −−−→ T (Cn−1 ) −−−−→ · · ·
Since T is additive, T (∂n )T (∂n−1 ) = T (∂n ∂n−1 ) = T (0) = 0 (the zero map) so we see that
T (C) is a complex in Comp(A). Therefore we can define its nth homology Hn (T (C)) =
Zn (T (C))/Bn (T (C)). As above, this defines two sequences in A Mod:
Now consider the diagram whose rows are exact sequences (6) and (4).
0 Zn (T (C)) T (Cn ) Bn−1 (T (C)) 0
f id g
Since the middle arrow is the identity (i.e. a boring isomorphism), if we can show that there
exist maps f and g where either one is an isomorphism, then the Five Lemma (7.2.12) proves
the other is an isomorphism. This will nearly finish the problem.
To prove that there is an isomorphism f , note that any exact additive functor T commutes
ϕ
with kernels, since applying T to an exact sequence 0 → K → A − → B produces an exact
T (ϕ)
sequence 0 → T (K) → T (A) −−→ T (B). Then T (K) is the kernel of T (ϕ), and kernels
are unique up to isomorphism so T (ker ϕ) ∼ = ker T (ϕ). In our context this proves that
Zn (T (C)) = ker T (∂n ) ∼
= T (ker ∂n ) = T (Zn (C)). Now we can fill one arrow:
180
10.1. Chain Complexes and Homology Chapter 10. Homology
∼
= id g
Clearly the left square commutes since both vertical arrows are isomorphisms.
Now we turn our attention to constructing g. If b ∈ Bn−1 (T (C)) then exactness of the
top row says there is an element c ∈ T (Cn ) mapping to b. Since the middle vertical arrow is
the identity, c 7→ c ∈ T (Cn ) in the bottom row, which has image b0 ∈ T (Bn−1 (C)). We use
this to define g : Bn−1 (T (C)) → T (Bn−1 (C)), b 7→ b0 .
The only concern is that g may not be well defined, so we take a moment to verify this
with a diagram chase:
c̃ (1)
(2) c
z c − c̃ b
0 Zn (T (C)) T (Cn ) Bn−1 (T (C)) 0
(3) ∼
= id g
Suppose b pulls back to c and c̃ in T (Cn ). Then their difference c − c̃ is in the kernel of
T (Cn ) → Bn−1 (T (C)) so c−c̃ pulls back to some z ∈ Zn (T (C)). The left vertical arrow takes
z 7→ z 0 isomorphically, so z 0 7→ c − c̃ along the bottom left by commutativity. By exactness
of the bottom row, pushing z 0 along both maps on the bottom takes z 0 7→ c − c̃ 7→ 0, which
shows that g is well defined after all.
So g exists and by construction it makes the right square commute. We can apply the
Five Lemma (7.2.12) to see that g must be an isomorphism, and this holds for any n so if
we consider the diagram with (5) and (3) as its rows we can put in isomorphisms on the left
and middle arrows.
0 Bn (T (C)) Zn (T (C))) Hn (T (C)) 0
∼
= ∼
=
The proof above generalizes to any diagram of this form, showing there exists an isomorphism
(the dashed arrow) Hn (T (C)) ∼= T (Hn (C)) as desired.
181
10.1. Chain Complexes and Homology Chapter 10. Homology
The dual notion to homology is cohomology, which as with many ‘dual’ concepts in
homological algebra, is formed by reversing the arrows. We define things explicitly below.
Note that the subscripts of homology become superscripts in cohomology, and most
objects acquire a “co-” prefix. For example, ker ∂n = Z n (A• ) are called the n-cocycles of
A• , im ∂n+1 = B n (A• ) are the n-coboundaries of A• and these allow us to define
182
10.2. Derived Functors Chapter 10. Homology
183
10.2. Derived Functors Chapter 10. Homology
K1 K2
Then T n (M ) ∼
= T n−1 (K1 ) ∼
= ··· ∼
= T 1 (Kn−1 ) and this last term is
T 0 (Kn ) ∼ ker(T 0 (En ) → T 0 (En+1 ))
T 1 (Kn−1 ) ∼
= = = H n (T 0 (E• )).
im T 0 (En−1 ) im(T 0 (En−1 ) → T 0 (En ))
This proves Theorem 10.2.3 for left exact functors. Dually, we have:
Definition. A homological δ-functor (Tn ) is coeffaceable if for any n ≥ 0 and any object
M , there is an epimorphism N → M such that the induced map Tn (N ) → Tn (M ) is 0.
Proposition 10.2.7. If A is an abelian category with enough projectives, then a δ-functor
is coeffaceable if and only if it vanishes on all projectives.
Corollary 10.2.8. Coeffaceable δ-functors are unique up to unique isomorphism.
Corollary 10.2.9. If (Tn ) is a homological δ-functor on an abelian category A with enough
projectives, then there is a natural isomorphism Tn ∼
= Hn (T0 (−)) for all n ≥ 0.
This proves Theorem 10.2.3 for right exact functors.
Definition. Suppose T : A → C is a covariant, additive, right exact functor between abelian
categories, where A has enough projectives. The nth left derived functor of T is the
universal homological δ-functor of T :
Ln T : A −→ C
A 7−→ Hn (T (P• ))
184
10.2. Derived Functors Chapter 10. Homology
Derived functors in some sense measure the failure of exactness in the nth homological
dimension of the functor T .
Definition. In the category R-Mod of left R-modules, if TM = − ⊗R M then its left derived
functors are called Tor:
TorR
n (M, N ) := (Ln TM )(N ) = Hn (P•,N ⊗R N ).
0
Likewise for a right R-module M , the left derived functors of TM = M ⊗R − are called tor:
0
torR
n (N, M ) := (Ln TM )(N ) = Hn (M ⊗R P•,N ).
Rn S : A −→ C
A 7−→ H−n (S(E• ))
We summarize the different variances and types of derived functors in the table below.
Variance of T Category requirements Derived functor
covariant enough projectives Ln T
covariant enough injectives Rn T
contravariant enough projectives Rn T
contravariant enough injectives Ln T
Example 10.2.10. For all right R-modules A and left R-modules B, TorR ∼
0 (A, B) = A ⊗R B.
Likewise, if A and B are both left R-modules then Ext0R (A, B) ∼
= HomR (A, B).
In defining left and right derived functors, we have neglected the fact that we are choosing
a particular projective (or injective) resolution of A. The Comparison Theorem says that
unique chain maps exist between projective resolutions of M and N when M → N is a
module homomorphism, so we need not worry when defining Ln T and Rn T .
Theorem 10.2.11 (Comparison). Let P• : · · · → P2 → P1 → P0 be a projective chain
complex and suppose C• : · · · C2 → C1 → C0 is an acyclic chain complex. Then for any
homomorphism ϕ : H0 (P• ) → H0 (C• ), there is a chain map f : P• → C• whose induced map
on H0 is ϕ, and ϕ is unique up to chain homotopy.
185
10.2. Derived Functors Chapter 10. Homology
f2 f1 f0 ϕ
∂20 ∂10 ∂00
C2 C1 C0 H0 (C• ) 0
fn fn−1
∂n0
Cn Cn−1
0
Note that fn−1 ∂ has image lying in ker ∂n0 ⊆ Cn−1 , but C• is acyclic, so ∂n0 (Cn ) = ker ∂n−1 .
Since Pn is projective, we can lift fn−1 ∂n to the desired map fn : Pn → Cn . By construction,
f = {fn }∞
n=0 satisfies the desired properties.
For uniqueness, suppose g : P• → C• is another chain map restricting to ϕ on H0 . Since
f0 − g0 = 0 on H0 , it must be that (f0 − g0 )(P0 ) ⊆ ker ∂00 = im ∂10 , so by projectivity of P0
there exists s0 : P0 → C1 making the following diagram commute:
P0
s0 f0 − g0
C1 im ∂10 0
0
Inductively, given s0 , . . . , sn−1 satisfying fk − gk = ∂k+1 sk + sk+1 ∂k for all 0 ≤ k ≤ n − 1, we
have
∂n0 (fn − gn − sn−1 ∂n ) = (fn−1 − gn−1 )∂n − ∂n0 sn−1 ∂n since f, g are chain maps
= (∂n0 sn−1 − sn−2 ∂n−1 )∂n − ∂n0 sn−1 ∂n = 0.
sn fn − gn − sn−1 ∂n
Cn+1 0
ker ∂n0 0
∂n+1
0
This establishes the chain homotopy s : P• → C• such that fn − gn = ∂n+1 sn + sn+1 ∂n for
all n ≥ 0. Hence f is unique up to chain homotopy.
186
10.2. Derived Functors Chapter 10. Homology
Corollary 10.2.13. For R-modules M and N , the assignments (M, N ) 7→ Hn (P• ⊗ N ) and
(M, N ) 7→ H n (Hom(P• , N )) are independent of the deleted projective resolution P• → M
and are functorial in both M and N . In particular, Ext and Tor are well-defined functors.
Proof sketch. Use the fact that Hn is an additive covariant functor Comp(A) → A. Be careful
checking that Ln T is well-defined on chain maps.
Proof. This is immediate from the definition since the nth term of a resolution P of A is 0
whenever n is negative.
Notice that since TorR R R
• (L, −) is an effaceable δ-functor, Tor1 (L, M ) = 0 implies Torn (L, M ) =
0 for n ≥ 2 and for all modules M . Similarly, Ext1R (L, M ) = 0 implies ExtnR (L, M ) = 0 for
n ≥ 2 and all M .
Definition. The depth of an R-module M , denoted depth(M ), is the smallest n for which
ExtnR (R, M ) is nonzero.
The Krull dimension (see Section 2.3) of an R-module M is defined to be the supremum
of the lengths of all chains of prime ideals. An important result, which we won’t prove, is
187
10.2. Derived Functors Chapter 10. Homology
n ≤ depth(R) ≤ dim R.
188
10.3. Tor and Ext Chapter 10. Homology
Theorem 10.3.1. Let R be a ring, A a right R-module and B a left R-module. Then
(i) For all n ≥ 0, TorR ∼ R op
n (A, B) = Torn (B, A).
Proof. It suffices to prove (i) since (ii) will follow immediately. We know that every left
R-module is a right Rop -module and likewise every right R-module is a left Rop -module. Let
P• be a projective resolution of A. Then t : P• ⊗R B −→ B ⊗Rop P• is a chain map of
Z-complexes with tn : Pn ⊗R B 7−→ B ⊗Rop Pn defined by pn ⊗ b 7→ b ⊗ pn . It is easy to
see that each tn is an isomorphism, so the complexes are isomorphic and therefore have the
same homology:
TorR ∼ Rop
n (A, B) = Hn (P• ⊗R B) = Hn (B ⊗Rop P• ) = Torn (B, A).
The key property is that Torn vanishes on projective modules of homological degree n ≥ 1,
since a projective resolution P• of a projective module P is simply → 0 → 0 → P → 0.
Proposition 10.3.2. Let R be a ring and let M and N be R-modules.
(a) If M is a free R-module, then TorR n
n (M, N ) = ExtR (M, N ) = 0 for n > 0.
ending in · · · → A0 ⊗R B → A ⊗R B → A00 ⊗R B → 0.
Proof. Use the definition of Tor via homology and the long exact sequence from Theo-
rem 10.1.3.
This theorem is true more generally for any left derived functor, and we will see a similar
sequence with Ext and right derived functors. The long exact sequence for Tor is usually
summarized in the literature by saying that the Tor sequence repairs loss of exactness on
the left of tensoring a short exact sequence.
The next result says that TorR R
n (A, −) and Torn (−, B) can be computed using projective
resolutions (or alternatively, flat resolutions) in either slot.
189
10.3. Tor and Ext Chapter 10. Homology
Theorem 10.3.4. For all projective (or flat) resolutions P• of A and Q• of B and for all
n ≥ 0,
Hn (P• ⊗R B) ∼ = TorR ∼
n (A, B) = Hn (A ⊗R Q• ).
Tor1 (A, B) ∼ r s
M M
= (Tor1 (Z, B)) ⊕ (Tor1 (A, Z)) ⊕ Tor1 (Ai , B) ⊕ Tor1 (A, Bj ).
i j
Since Z is projective, Tor vanishes on the free parts in the first two summands. Then we’re
left with Tor of cyclic groups. By the work above, each Tor1 (Ai , B) = {b ∈ B | ni b = 0}
where ni = |Ai |. Thus Tor1 (Ai , B) is a cyclic group of order di = gcd(ni , |B|) and the same
holds for the Tor1 (A, Bj ).
Example 10.3.7. If R is an integral domain with field of fractions Q and residue field
K = Q/R, then Tor captures the idea of torsion submodules. In fact, if A is an R-module
and At is its torsion submodule, then TorR ∼
1 (K, A) = At . Moreover, there is an exact sequence
0 → At → A → Q ⊗R A → K ⊗R A → 0
which shows that A is torsion ⇐⇒ Q ⊗R A = 0. This explains the name Tor for in general,
TorR
n is a torsion R-module.
190
10.3. Tor and Ext Chapter 10. Homology
ε
··· Pn ··· P1 P0 A 0
i
δ0
Qn ··· Q1 Q0 B 0
Φ j
··· Rn ··· R1 R0 C 0
Our goal is to complete the middle row such that the diagram commutes. We begin by
constructing δ0 : Q0 → B. Since R0 is projective, there exists a map Φ : R0 → B lifting j,
given by the dashed arrow above. Then define δ0 on Q0 = P0 ⊕ R0 by δ0 = (i ◦ ε) ⊕ (−Φ).
Having constructed δ0 , . . . , δn−1 , we have the following portion of the above diagram:
··· Pn Pn−1 Pn−2 ···
δn−1
··· Qn Qn−1 Qn−2 ···
δn
Qn L
··· Rn M 0
191
10.3. Tor and Ext Chapter 10. Homology
Proof. Take projective resolutions P•0 → N 0 and P•00 → N 00 ; then Lemma 10.3.8 provides a
projective resolution P• for N and a short exact sequence of complexes
0 → P•0 → P• → P•00 → 0.
Since each term in the sequence is a complex of projective modules, applying M ⊗ − yields
a short exact sequence again. Then the desired long exact sequence is merely the long exact
sequence in homology for this induced short exact sequence.
(1) TorR
0 (M, N ) = M ⊗ N .
(2) For any short exact sequence 0 → M 0 → M → M 00 → 0 and any R-module N , there
is a long exact sequence
which is natural in N .
Moreover, any functor satisfying these three properties is naturally isomorphic to TorR
n.
Proof. We have proven that Torn satisfies the stated properties so it remains to show that
these in fact characterize Torn . We prove this inductively on n. For n = 0, uniqueness
follows from the universal property of the tensor product. For n ≥ 1, take modules M and
N and a free module F such that there is an exact sequence 0 → K → F → M → 0. Then
by (2), there is a long exact sequence
When n > 1, Torn−1 (F, N ) = 0 as well so Torn (M, N ) ∼ = Torn−1 (K, N ). When n = 1,
Tor1 (M, N ) ∼
= ker(K ⊗ N → F ⊗ N ). In all cases, induction implies that Torn is determined
as a functor by Torn−1 so uniqueness holds.
Corollary 10.3.11. For any n ≥ 0 and any modules M and N , there is a natural isomor-
phism
Torn (M, N ) ∼
= Torn (N, M ).
Proof. Consider the assignment (M, N ) 7→ torn (M, N ) := Torn (N, M ). Then
192
10.3. Tor and Ext Chapter 10. Homology
(2) torn has a long exact sequence in the first variable because Torn has a long exact sequence
in the second variable by Corollary 10.3.9.
(3) For a free module F , torn (F, N ) = Torn (N, F ) = 0 for each n > 0 since − ⊗ F preserves
exactness.
Hence by Theorem 10.3.10, Torn and torn are naturally isomorphic.
Next we shift our focus to Ext. This is in some ways the more interesting of the two
derived functors, since in any module category it naturally admits an R-algebra structure.
As we saw for Tor, Ext defines a long exact sequence which repairs the loss of exactness of
applying the Hom functor to a short exact sequence.
Theorem 10.3.12 (Long Exact Sequence for Ext). If 0 → B 0 → B → B 00 → 0 is a short
exact sequence in R Mod then for every left R-module A, ExtnR (A, −) induces a long exact
sequence
which is natural in N .
(3) For any free module F , ExtnR (F, N ) = 0 for all n > 0.
Moreover, any functor satisfying these three properties is naturally isomorphic to ExtnR .
Proof. Reverse the arrows in the proof of Theorem 10.3.10.
Ext is characterized by the property that Extn vanishes on all injective modules whenever
n ≥ 1, for the same reason that Torn vanishes on projectives. The next proposition shows
that Ext interacts with direct sums and products in the same way that Hom does.
193
10.3. Tor and Ext Chapter 10. Homology
Proposition 10.3.15. For R-modules B and {Ai }i∈I and for all n ≥ 0, there are natural
isomorphisms !
Ai , B ∼
M Y
ExtnR = ExtnR (Ai , B).
i∈I i∈I
Proposition 10.3.16. For R-modules A and {Bi }i∈I and for all n ≥ 0, there are natural
isomorphisms !
Bi ∼
Y Y
ExtnR A, = ExtnR (A, Bi ).
i∈I i∈I
Example 10.3.17. Fix n ≥ 0, m ≥ 1 and consider ExtnZ (Z/mZ, Z). A simple projective
resolution of Z/mZ is the short exact sequence
m
0→Z−
→ Z → Z/mZ → 0.
Ext1Z (Z/nZ, B) ∼
= B/nB.
(Compare to Example 10.3.6.) As before, one can use this to show that for any finitely
generated abelian groups A, B, Ext1 (A, B) is the direct sum of Ext1 (Ai , Bj ) where Ai and
Bj are finite cyclic groups.
194
10.3. Tor and Ext Chapter 10. Homology
The elements in Ext1R (C, A) can be thought of as ‘obstructions’ to the splitting property
of extensions of A by C. This explains where the name Ext comes from. An important
property of Ext is
Proposition 10.3.19. A left R-module P is projective ⇐⇒ Ext1R (P, B) = 0 for every
R-module B. Similarly, a left R-module E is injective ⇐⇒ Ext1R (A, E) = 0 for every
R-module A.
Proof. If P is projective, then Ext1R (P, B) = 0 for all B by Proposition 8.1.3. On the other
hand, if Ext1R (P, B) = 0 for all B then Lemma 10.3.18 shows that every exact sequence
ending in P splits and hence P is projective.
The second statement is proven similarly.
Likewise, we have a homological characterization of flat modules.
Proposition 10.3.20. For an R-module M , the following are equivalent:
(a) M is flat.
(b) TorR
n (N, M ) = 0 for all n ≥ 1 and R-modules N .
(c) TorR
1 (N, M ) = 0 for all R-modules N .
(d) TorR
1 (N, M ) = 0 for all finitely generated R-modules N .
(e) TorR
1 (R/p, M ) = 0 for all prime ideals p ⊂ R.
0 → Nk−1 → N → R/pk → 0
0 = TorR R R
1 (Nk−1 , M ) → Tor1 (N, M ) → Tor1 (R/pk , M ) = 0.
So TorR
1 (N, M ) = 0.
Theorem 10.3.21 (Local Criterion for Flatness). Suppose ϕ : (R, m) → (S, n) is a local
homomorphism of local noetherian rings and N is a finitely generated S-module. Then N is
a flat R-module if and only if TorR
1 (R/m, N ) = 0.
195
10.3. Tor and Ext Chapter 10. Homology
N0 ⊆ · · · ⊆ N` = N
0 → k = R/m → N → N 0 → 0
Then dim(R/(p, x)) < dim(R/p) ≤ d so by induction, Tor1 (R/(p, x), M ) = 0. Thus the long
exact sequence in Tor for the above sequence contains
x
Tor1 (R/p, M ) →
− Tor1 (R/p, M ) → Tor1 (R/(p, x), M ) = 0.
That is, Tor1 (R/p, M ) = x Tor1 (R/p, M ) but since this is a finitely generated R-module,
Nakayama’s lemma (Theorem 1.2.1) implies Tor1 (R/p, M ) = 0.
(a) R/aR ⊗ M ∼
= M/aM .
(b) Tor1 (R/aR, M ) ∼
= aM .
(c) Hom(R/aR, M ) ∼
= aM .
(d) Ext1 (R/aR, M ) ∼
= M/aM .
Proof. (a) Define a map φ : R/aR × M → M/aM by (r, m) 7→ rm. If r − r0 ∈ aR, then
r − r0 = as for some s ∈ R. Then (r − r0 )m = (as)m = a(sm) ∈ aM so the map is
well-defined on the quotient R/aR. It is also clearly bilinear. By the universal property of
tensor products, this determines a linear map Φ : R/aR ⊗ M → M/aM . If Φ(r ⊗ m) = 0 in
M/aM then Φ(r ⊗ m) = rm ∈ aM . Thus rm = am0 for some m0 ∈ M , and since the tensor
is over R, we can pass elements of R across the tensor:
r ⊗ m = 1 ⊗ rm = 1 ⊗ am0 = a ⊗ m0 = 0 ⊗ m0 = 0 in M/aM.
196
10.3. Tor and Ext Chapter 10. Homology
197
10.3. Tor and Ext Chapter 10. Homology
Example 10.3.24. Let m ≥ 2 and n ≥ 0 be integers. Then using the above results, we can
compute
(
Z/mZ, n = 0
Torn (Z, Z/mZ) = Torn (Z/mZ, Z) =
0, n>0
0,
n=0
n
Ext (Z/mZ, Z) = Z/mZ, n = 1
0, n>1
(
Z/mZ, n = 0
Extn (Z, Z/mZ) =
0, n > 0.
Example 10.3.25. Let R be a commutative ring and m ⊂ R a maximal ideal. Let k = R/m
be the residue field. An interesting Ext group in commutative algebra is Ext1R (k, k), as we
will see in more detail in Section 11.1. To compute this group, consider the short exact
sequence
0 → m → R → k → 0.
Since R is a free module over itself, by Theorem 10.3.14, the long exact sequence in ExtR (−, k)
becomes
∼
=
→ HomR (R, k) → HomR (m, k) → Ext1R (k, k) → 0.
0 → HomR (k, k) −
Ext1R (k, k) ∼
= HomR (m, k).
For any R-module N , every map N → R/m factors through N/mN , so we see that HomR (N, R/m) ∼
=
HomR (N/mN, R/m). Setting N = m, we then get HomR (m, k) ∼ = HomR (m/m2 , k) ∼ =
(m/m2 )∗ , the dual space of m/m2 . This is more commonly known as the tangent space to
Spec R at m, Tm Spec R = (m/m2 )∗ . Thus tangent spaces have a homological interpretation.
Example 10.3.26. For an R-module M , we can similarly consider Ext1R (M, M ), which
consists of R-modules M
f for which there is a short exact sequence
0→M →M
f → M → 0.
198
10.3. Tor and Ext Chapter 10. Homology
Then other deformations of M come from maps α̃ : R → End(M [ε]) such that α ≡ α̃ mod
ε. That is, α̃ = α + εµ for µ : R → End(M ). For any r ∈ R, m1 , m2 ∈ M ,
α̃(r)(m1 + εm2 ) = rm1 + ε(rm2 + µ(r)µ1 ).
Thus for α̃ to define an R-module structure on M [ε], we must have α̃(r1 )α̃(r2 ) = α̃(r1 r2 ),
i.e. for any m ∈ M ,
α̃(r1 )(r2 m + εµ(r2 )m) = r1 r2 m + εµ(r1 )(r2 m) + εr1 µ(r2 )m
= r1 r2 m + εµ(r1 r2 )m = α̃(r1 r2 )m.
Thus the necessary condition on µ is
µ(r1 )r2 + r1 µ(r2 ) − µ(r1 r2 ) = 0.
This is obviously a cocycle condition, so it can be measured with homology.
On the other hand, when are α̃ and α̃0 equivalent? i.e. when do they determine equivalent
deformations? Let α̃0 = α + εµ0 . For equivalence to be true, there must be some β =
1 + εγ ∈ End(M [ε]) such that β α̃ = α̃0 β. (Notice that β −1 = 1 − εγ since ε2 = 0.) Then
α̃ = β −1 α̃0 β = (1 − εγ)α̃0 (1 + εγ), so the necessary condition on µ and µ0 is
µ = µ0 + ε(αγ − γα).
This proves there is a bijection
{µ : R → End(M ) | µ(r1 )r2 + r1 µ(r2 ) − µ(r1 r2 ) = 0}
Ext1R (M, M ) ←→ .
hαγ − γα | γ ∈ End(M )i
One can also check that this correspondence is compatible with the group operations, where
addition on the right is pointwise addition of the maps µ : R → End(M ).
Now assume R is a finitely presented ring with generators x1 , . . . , xn , and M is both an
R-module and a finite dimensional vector space over some field k (e.g. k is the residue field
of a local ring as in the previous example). A potential R-action on M is specified by a map
Span{x1 , . . . , xn } −→ Endk (M ) ∼
= Mn (k), xi 7−→ Xi
such that the relations on the xi are mapped to 0, i.e. there are polynomials pj for which we
have pj (X1 , . . . , Xn ) = 0. This defines an algebraic set Y , the space of all potential R-actions
on M , which has an action of GL(M ). The quotient space ModR (M ) = Y /GL(M ) is called
the moduli space of R-actions on M . Under the quotient map Y → Y /GL(M ), a tangent
vector to M maps to a tangent vector modulo a tangent vector to GL(M ), which is just a
matrix. These tangent vectors on ModR (M ) are precisely given by certain cocycle conditions
like the one above.
Example 10.3.27. Let Γ be a quiver, that is, an oriented graph.
3
1 2
4
199
10.3. Tor and Ext Chapter 10. Homology
The graph algebra of Γ is defined to be the span of all oriented paths in Γ over some commu-
tative ring R (usually a field), with multiplication defined by concatenation where possible
(and 0 otherwise) and addition given by formal sums over R. This algebra is written Π(Γ).
The representation theory of Π(Γ) is quite important in modern P representation theory. If 1v
denotes the trivial path at a vertex v ∈ V (Γ), notice that 1 = v∈V (Γ) 1v and 1v 1v0 = δvv0 1v ,
that is, the elements 1v are orthogonal idempotents in Π(Γ). Therefore for any module M
over the path algebra, there is a decomposition
M∼
M
= 1v M.
v∈V (Γ)
Set Mv = 1v M . We can label the vertices of the graph with these submodules:
M3
M1 M2
M4
This allows us to study the representations of Π(Γ) in more detail by making choices at each
vertex. For example, let R = k be a field and consider the following module:
0
0 k
0
This depicts a simple Π(Γ)-module since it is 1-dimensional. (In fact, if Γ has no oriented
cycles, then these are the only simple modules.) Let Sv denote the simple module with k at
vertex v, e.g. S1 is depicted above. We are interested in computing the group ExtΠ(Γ) (Sv , Sv0 )
for different v, v 0 ∈ V (Γ). Consider the short exact sequence
0 → K → Π(Γ) → Sv → 0.
Here, 1w 7→ δvw for any idempotent 1w ∈ Π(Γ). The kernel K is spanned by all paths of
length at least one, together with all 1w for w 6= v. Decompose the path algebra as:
M
Π(Γ) = Pw
w∈V (Γ)
where Pw := Π(Γ)1w . Then each summand Pw is a projective module. Note that HomΠ(Γ) (Pw , M ) ∼
=
∼
Mw for any module M – in general, this is the fact that HomR (Re, M ) = eM for an idem-
potent e. Thus we have a short exact sequence
0 → K → Pv → Sv → 0
200
10.3. Tor and Ext Chapter 10. Homology
which is easier to understand. Applying ExtΠ(Γ) (−, Sv0 ) yields an exact sequence
Ext(Sv , Sv0 ) ∼
= Hom(K, Sv0 ).
Further, one can see that K is in canonical bijection with the set of paths in Γ starting at
any w adjacent to v, counted with multiplicity given by the number of edges v → w. That
∼
is, K = v→w Pwjw , where jw is this multiplicity of paths. Thus we can write
L
!
Ext(Sv , Sv0 ) ∼
= Hom(K, Sv0 ) ∼ P jw , Sv0 ∼
M
= Hom w = k `vv0 ,
v→w
where `vv0 represents the number of paths v → v 0 . In practice, these Ext groups contain
enough information (number of paths) to reconstruct the quiver from just a set of vertices.
201
10.4. Universal Coefficient Theorems Chapter 10. Homology
where (2) is split. Tensoring with M , we get a diagram with exact rows and columns:
0 0
0
i q
Bn ⊗ M Zn ⊗ M Hn ⊗ M 0
g
j
∂ ∂
Cn+1 ⊗ M Cn ⊗ M Cn−1 ⊗ M
f
∂
0
0 0
The bottom row is the long exact sequence in Tor coming from the short exact sequence (1),
but it terminates after Tor1 because Zn−1 is free. Also, since (2) is split we get the zeroes at
the top of the middle column and bottom of the right column. By diagram chasing, one can
define f and g and show that the sequence is exact. Moreover, the sequence is split due to
the fact that sequence (2) is split. Finally, naturality follows from naturality of tensor, Hn
and Tor1 .
202
10.4. Universal Coefficient Theorems Chapter 10. Homology
Theorem 10.4.2 (Universal Coefficient Theorem for Cohomology). For a free chain complex
C• over a PID and a coefficient module M , for each n ≥ 0 there is a short exact sequence
Proof. Replacing − ⊗ M with Hom(−, M ), Tor1 with Ext1 and reversing the arrows in the
proof of Theorem 10.4.1 gives a diagram:
0 0
0
i q
Hom(Bn , M ) Hom(Zn , M ) Hom(Hn , M ) 0
g
j
∂ ∂
Hom(Cn+1 , M ) Hom(Cn , M ) Hom(Cn−1 , M )
f
∂
0
0 0
Lemma 10.4.3. Let G = F (G) ⊕ T (G) be a finitely generated abelian group, where F (G)
is the free part of G and T (G) is the torsion part of G. Then Hom(G, Z) ∼= F (G) and
Ext1 (G, Z) ∼
= T (G).
Proof. Use the universal coefficient theorem, Example 10.3.24 and additivity.
H n (C• , M ) ∼
= Fn ⊕ Tn−1 ,
203
Chapter 11
Ring Homology
In this chapter we use homological algebra to study several types of singularities in commu-
tative algebra. The strongest type of nonsingularity is regularity: as defined in Section 6.2, a
local ring (A, m, k) is regular if its Krull dimension equals the dimension of the k-vector space
m/m2 . By Theorem 6.2.8, this is equivalent to m being generated by dim A elements. The
dimension theory of Section 6.2 can be weakened in several ways to yield different notions
of singularity, including the Cohen-Macauley and Gorenstein conditions.
Note to the reader: in the first two sections, I use R and S to denote rings and A and B
for modules, in contrast to later sections. Sorry for the confusion!
204
11.1. Dimensions of Rings Chapter 11. Ring Homology
(3) Extn+1
R (A, B) = 0 for all left R-modules B.
(4) There exists a projective resolution P• of A such that the (n − 1)st syzygy Kn−1 is
projective.
Extn+1 ∼ 1
R (A, B) = ExtR (Kn−1 , B) = 0.
0 → Kn−1 → Pn−1 → · · · → P1 → P0 → A → 0
205
11.1. Dimensions of Rings Chapter 11. Ring Homology
Proof omitted.
The proof of the next proposition shows that the long exact Ext sequence can be useful
when dealing with projective dimension.
Proposition 11.1.3. Let 0 → M 0 → M → M 00 → 0 be an exact sequence of left R-modules.
(i) If pd(M 0 ) < pd(M ) then pd(M 00 ) = pd(M ).
(ii) If pd(M 0 ) > pd(M ) then pd(M 00 ) = pd(M 0 ) + 1.
(iii) If pd(M 0 ) = pd(M ) then pd(M 00 ) ≤ pd(M 0 ) + 1.
Proof. Let 0 → M 0 → M → M 00 → 0 be exact, take A to be a left R-module and consider
the long exact sequence for Ext:
0 → Hom(M 00 , A) → Hom(M, A) → Hom(M 0 , A)
→ Ext1 (M 00 , A) → Ext1 (M, A) → Ext1 (M 0 , A) → · · ·
· · · → Extn (M 00 , A) → Extn (M, A) → Extn (M 0 , A)
→ Extn+1 (M 00 , A) → Extn+1 (M, A) → Extn+1 (M 0 , A) → · · ·
(i) Let pd(M ) = n and suppose pd(M 0 ) < n. Then Extn (M 0 , A) = 0 and Extn+1 (M, A) =
0 so exactness implies Extn+1 (M 00 , A) = 0. Thus pd(M 00 ) ≤ n. Notice that since n is the
projective dimension of M , Extn (M, A) 6= 0. Then we cannot have Extn (M 00 , A) = 0 since
the right arrow of then nth row
· · · → Extn (M 00 , A) → Extn (M, A) → 0 → · · ·
is a surjection and hence nonzero. Therefore pd(M 00 ) = n.
(ii) Now suppose pd(M 0 ) > n. Let pd(M 0 ) = N so that ExtN +1 (M 0 , A) = 0. Since
N > n, ExtN (M, A) and ExtN +1 (M, A) are both 0 so the N th and (N + 1)st rows of the
long sequence look like
· · · ExtN (M 00 , A) → 0 → ExtN (M 0 , A) → ExtN +1 (M 00 , A) → 0 → 0 → · · ·
By exactness, this shows that ExtN (M 0 , A) ∼ = ExtN +1 (M 00 , A) and in fact this will hold for
Extk (M 0 , A) and Extk+1 (M 00 , A) for all k ≥ N (all of the Exts on M in between will be
0). Hence ExtN +1 (M 00 , A) 6= 0 and Extk (M 00 , A) = 0 for all k > N + 1 so by definition
pd(M 00 ) = N + 1.
(iii) Finally, suppose pd(M 0 ) = n. Then Extk (M 0 , A) = 0 = Extk (M, A) for all k ≥ n + 1.
Consider
· · · → Extn (M 00 , A) → Extn (M, A) → Extn (M 0 , A)
→ Extn+1 (M 00 , A) → 0 → 0
→ Extn+2 (M 00 , A) → 0 → 0 → · · ·
As before, the 0’s on either side show that Extn+2 (M 00 , A) = 0 which is enough to prove that
pd(M 00 ) ≤ n + 1.
The proofs when any of the projective dimensions are infinite follows from Proposi-
tion 11.1.2.
206
11.1. Dimensions of Rings Chapter 11. Ring Homology
Example 11.1.4. For (R, p, k) a local ring, pick a projective resolution P• of k over R.
Since − ⊗R Rp is exact, P• ⊗R Rp is a projective resolution of k ⊗R Rp . Then we can compute
so ExtiR (M, k) ∼
= k rank Fi as well. This proves (2) of Theorem ??.
In practice, we apply this to an exact sequence 0 → M 0 → M → M 00 → 0 to produce
where n = max{pd(M ), pd(M 0 ), pd(M 00 )}. The above shows that the Tor are all vector
spaces, and rank is additive along exact sequences, so
n
X n
X
i
(−1) βi (M ) = (−1)i (βi (M 0 ) + βi (M 00 ))
i=0 i=0
X
where βi denotes the ith Betti number of the module in its parentheses. The sum (−1)i βi (M )
is called the Euler characteristic of M , denoted χ(M ). We have therefore shown that
χ(M ) = χ(M 0 ) + χ(M 00 ), i.e. Euler characteristic is additive along short exact sequences.
207
11.2. Hilbert’s Syzygy Theorem Chapter 11. Ring Homology
Definition. Given a graded R-module M , its Hilbert series is the formal power series
X
HM (t) = (dim Mi )ti .
i∈Z
Hilbert’s question can be restated by asking if there is a recurrence relation among the
coefficients of HM (t). This in turn is equivalent to asking if HM (t) is rational.
0 → M 0 → M → M 00 → 0
where the arrows are graded homomorphisms. Then HM (t) = HM 0 (t) + HM 00 (t).
0 → Mi0 → Mi → Mi00 → 0
K1
0 0
208
11.2. Hilbert’s Syzygy Theorem Chapter 11. Ring Homology
The ring R(−bij ) is called a shifting of the graded ring R, which is used to line up the
degrees of the mi with the graded pieces of the ring. One can prove
tbij
In particular, HR(−bij ) (t) = tbij HR (t) = for all i, j. Hilbert’s proof of the Syzygy
(1 − t)n
Theorem boiled down to showing
X (−1)i+1 tbij
HM (t) =
i,j
(1 − t)n
whenever pd(M ) ≤ n. In modern language, this can be stated in terms of global dimension.
(i) For any R-module A, pd(A) = max{n ≥ 0 | ExtnR (A, M ) 6= 0 for some M }.
One defines injective and flat dimension in analogy with projective dimension:
(1) id(A) ≤ n.
(3) Extn+1
R (B, A) for all left R-modules B.
(4) There exist an injective resolution E• of A such that the (n − 1)st cosyzygy Kn−1 is
injective.
209
11.2. Hilbert’s Syzygy Theorem Chapter 11. Ring Homology
0 → N → E0 → E1 → · · · → Ee−1 → C
rational? That is, is there some recurrence on the size of the free modules in a minimal
resolution of M ?
210
11.2. Hilbert’s Syzygy Theorem Chapter 11. Ring Homology
(x y) 2
( x0 y0 −y
x )
0 0 x x 0
0 0 0 0 x
0 ← k ← S ←−−− S ←−−−− −− S 3 ←−−−−−−−− S 5 ← S 8 ← · · ·
This is a minimal resolution of k and we can see that the ith map relates the (i − 1)st
map in a recursive way. Interestingly, the ranks of the free modules are the Fibonacci
numbers (starting at the second 1). Then the Poincaré-Betti series is rational:
1
PSk (t) = 1 + 2t + 3t2 + 5t3 + 8t4 + . . . = .
1 − t + t2
√
Amazingly, PSk (t) is an analytic function with radius of convergence (lim sup n βn)−1 .
This gives a nice characterization of the growth of the Betti numbers βi .
√
Definition. The curvature of an R-module M is curv(M ) = lim sup n βn , where βi are
the Betti numbers for M .
211
11.3. Regular Sequences and the Koszul Complex Chapter 11. Ring Homology
(1) x1 is a nonzero divisor for M and for each 2 ≤ i ≤ n, xi is a nonzero divisor for
M/(x1 , . . . , xi−1 )M .
(2) (x1 , . . . , xn )M 6= M .
Remark. If M is finitely generated and each xi lies in the Jacobson radical J(A), then
condition (2) is guaranteed.
Example 11.3.2. If (A, m) is a 1-dimensional local integral domain, then every nonzero
element is a nonzero divisor, so A is Cohen-Macauley.
but x3 y 6= 0 in A/(x4 ).
x
Lemma 11.3.4. An element x ∈ A is a nonzero divisor if and only if 0 → A →
− A is exact,
and in this case the free resolution
x
0→A→
− A→
− A/(x) → 0
is exact.
Proof. Easy.
More generally, we will show that (x1 , . . . , xn ) is a regular sequence on A if and only if
the induced free resolution
(x1 ··· xn )
0 → Amn → · · · → Am1 −−−−−−→ A → A/(x1 , . . . , xn ) → 0
is exact.
212
11.3. Regular Sequences and the Koszul Complex Chapter 11. Ring Homology
213
11.3. Regular Sequences and the Koszul Complex Chapter 11. Ring Homology
(4) If A is noetherian and x1 , . . . , xn ∈ J(A) then the converse of (3) is true: if K• (x1 , . . . , xn ; M )
is acyclic then (x1 , . . . , xn ) is a regular sequence on M .
Proof. Write Hi (x1 , . . . , xn ; M ) := Hi (K• (x1 , . . . , xn ; M )). If n = 1, then (1) – (3) are
immediate for the Koszul complex
1 x
0 → M −→ M → 0.
H0 (x1 , . . . , xn ; M ) ∼
= M/(x1 , . . . , xn ; M ).
as desired.
(4) Again, the short exact sequence in Corollary 11.3.7 implies Hi (C• )/xn Hi (C• ) = 0 for
all i ≥ 1 and Nakayama’s lemma then implies Hi (C• ) = 0 for all i ≥ 1. By the inductive
hypothesis, (x1 , . . . , xn−1 ) is a regular sequence and we are assuming H1 (x1 , . . . , xn ; M ) =
AnnH0 (C• ) (xn ) = 0, so it follows that xn is a nonzero divisor on M/(x1 , . . . , xn−1 )M and we
are done.
214
11.3. Regular Sequences and the Koszul Complex Chapter 11. Ring Homology
TorB ∼ ∼
1 (k, A) = H1 (K• (x1 , . . . , xd ; B) ⊗B A) = H1 (K• (x1 , . . . , xd ; A)) = 0
since (x1 , . . . , xd ) is a regular sequence on A. By Theorem ??, this shows A is a flat B-module.
Conversely, if A is flat over B, then Tori (B)(k, A) = 0 for all i ≥ 1 which means
Hi (x1 , . . . , xd ; A) = 0 for all i ≥ 1. Thus by Theorem 11.3.10(4), (x1 , . . . , xd ) is a regu-
lar sequence. This shows that the Cohen-Macauley property (in this context) is equivalent
to being a flat B-module, where B is a power series ring. Even better, a finitely generated
flat module over any local noetherian ring is free, so the Cohen-Macauley property can be
detected locally by a certain freeness condition.
Example 11.3.13. Let B = k[[x4 , y 4 ]] and A = k[[x4 , y 4 , x3 y, xy 3 ]] be the completions of
the polynomial rings in Example 11.3.3. Then
but A is not free as a B-module, since for example y 4 (x6 y 2 ) − x4 (x2 y 6 ) = 0. Hence A is not
Cohen-Macauley.
Definition. A minimal free resolution of an A-module M is a free resolution
ϕ2 ϕ1 ϕ0
· · · → F2 −→ F1 −→ F0 −→ M → 0
215
11.3. Regular Sequences and the Koszul Complex Chapter 11. Ring Homology
The Betti numbers of a module have a homological characterization: note that TorA
i (k, M )
is the ith homology of the chain complex
ϕi+1 ϕ
· · · → k bi+1 −−→ k bi −
→i
k bi−1 → · · ·
216
11.4. Projective Dimension Chapter 11. Ring Homology
Proof. Since projective modules are flat and k ⊗A − is exact, we have automatically that
pdA (M ) ≤ sup{i | TorA i (k, M ) 6= 0}. Conversely, set n = pdA (M ) and induct on n; the base
ϕ2 ϕ1 ϕ0
case follows from Nakayama’s lemma. When n ≥ 1, let · · · → P2 −→ P1 −→ P0 −→ M → 0 be
a projective resolution of M and set Ki = ker ϕi . Then since P• is also a flat resolution of M ,
we have TorA ∼ A A
i+2 (k, M ) = Tor1 (k, Ki ). The base case shows that Ki is free, so Tori (k, M ) = 0
for all i > n, proving the opposite inequality.
Corollary 11.4.3. For a local noetherian ring (A, m, k) and a finitely generated A-module
M , pdA (M ) = n < ∞ if and only if M has a minimal free resolution of length n.
We will need the following lemma to establish Serre’s inequality for dimk TorA
i (k, k).
ϕ2 ϕ1 id id
m
··· A( 2 ) Am A k 0
217
11.4. Projective Dimension Chapter 11. Ring Homology
ψi ψi−1
d
Ki Ki−1
Assume inductively that ψi−1 : Fi−1 → Ki−1 has been constructed so that ψi−1 ◦ ϕi−1 = 1.
To show ϕi splits, by Lemma 11.4.4 it suffices to show that each ϕi : K• ⊗A k → F• ⊗A k
is injective. Take z ∈ Ki such that ϕi (z) ∈ mFi and note that δ(Fi ) ⊆ mFi−1 implies
δ ◦ ϕi (z) ∈ m2 Fi−1 . Thus ψi−1 ◦ δ ◦ ϕi (z) ∈ m2 Ki−1 , so we see that
d : Ki −→ Ki−1
X
eI 7−→ εj xj eIr{j}
j∈I
P
where εj = ±1. Write z = |I|=i zI eI ∈ Ki . Then
!
X X X X
d(z) = εj xj zI eIr{j} = εj xj zJ∪{j} eJ ∈ m2 Ki−1 .
I⊆{1,...,m} j∈I J⊆{1,...,m} j6∈J
|I|=i |J|=i−1
Since Ki−1 is a free A-module, this means each term j6∈J ±xj zJ∪{j} lies in m2 . However, in
P
m/m2 the xj form a basis so we must have zI ∈ m for all |I| = i. Hence d(z) ∈ mKi−1 , so ϕi
is one-to-one as required.
Remark. The following is an open problem independently posed by Buchsbaum-Eisenbud
and Harack. If (A, m) is a local noetherian ring of dimension d and M is a finitely generated,
artinian A-module with pdA (M ) < ∞, it is not known whether the Betti numbers bi =
d
dimk TorAi (k, M ) satisfy Serre’s lemma: bi ≥ i . However, Walker proved the following
weak form of the problem:
X d
bi ≥ 2d .
i=0
218
11.4. Projective Dimension Chapter 11. Ring Homology
Auslander, Buchsbaum and Serre gave a fundamental description of regular local rings
using projective dimension.
Theorem 11.4.6 (Auslander-Buchsbaum-Serre). For a local noetherian ring (A, m, k) of
dimension d, the following are equivalent:
(1) A is a regular local ring.
(2) m is generated by a regular sequence.
(3) pdA (k) < ∞.
(4) For all finitely generated A-modules M , pdA (M ) < ∞.
Proof. (1) =⇒ (2) Let m = (x1 , . . . , xd ). By Lemma 6.2.9, A is a domain and by induction
so is A/(x1 , . . . , xi ) for each 1 ≤ i ≤ d. Hence (x1 , . . . , xd ) is a regular sequence.
(2) =⇒ (3) Suppose m = (x1 , . . . , xt ) where (x1 , . . . , xt ) is a regular sequence on A.
Then K• (x1 , . . . , xt ; A) is a minimal free resolution of H0 (x1 , . . . , xt ; A) = k, so we have
pdA (k) = t < ∞.
(3) =⇒ (4) If n = pdA (k) then TorA i (k, k) = 0 for all i > n, so there is a minimal free
resolution of k with length n:
0 → Abn → · · · → Ab1 → A → k → 0.
Tensoring with M shows TorA i (k, M ) = 0 for all i > n, so pdA (M ) ≤ pdA (k).
(4) =⇒ (1) Assume pdA (k) = n < ∞. We claim n ≤ d = dim A. Let t be the length of
the longest regular sequence in A, say (y1 , . . . , yt ); we will see that t is equal to the depth of A.
By Proposition 3.6.9, m ∈ Ass(A/(y1 , . . . , yt )). That is, R/m = k sits inside of A/(y1 , . . . , yt )
and in fact we have an exact sequence
0 → k → A/(y1 , . . . , yt ) → M → 0.
If n > t, then tensoring with k gives a long exact sequence
· · · → TorA A A
n+1 (M, k) → Torn (k, k) → Torn (A/(y1 , . . . , yt ), k) → · · ·
but (4) implies TorA n+1 (M, k) = 0. On the other hand, since K• (y1 , . . . , yt ; A) has length t, we
A
get Torn (A/(y1 , . . . , yt ), k) = 0. Therefore TorA n (k, k) = 0, contradicting Proposition 11.4.2.
This Sproves n ≤ t. Next, we claim t ≤ d. If t = 1, A contains a nonzero divisor y1 such that
y1 6∈ p∈Ass(A) p. In particular y1 does not lie in any minimal prime of A by Theorem 3.6.7(3).
Thus d ≥ 1. Now suppose t ≥ 2 and note by Corollary 6.2.3 that dim(A/(y1 )) = d − 1.
Since (ȳ2 , . . . , ȳt ) is a regular sequence in A/(y1 ), so induction on t proves t − 1 ≤ d − 1 and
thus t ≤ d. Finally, let δ = δ(A) be the minimal number of generators of m. We know from
Corollary 6.2.6 that d ≤ δ so to complete the proof it is enough to show δ ≤ n. But this just
follows from Lemma 11.4.5.
Remark. In the course of the proof of Auslander-Buchsbaum-Serre, we established that if
(A, m, k) is a local noetherian ring with pdA (k) < ∞, then
δ(A) = pdA (k) = depth(A) = dim A
which strengthens the conclusion in Section 6.2. Further, this gives a new criterion for a
local noetherian ring to be regular: dimk m/m2 ≤ pdA (k).
219
11.4. Projective Dimension Chapter 11. Ring Homology
0 → Abr → · · · → Ab1 → A → M → 0,
so Corollary 11.4.3 implies pdAp (k(p)) < ∞. Hence (3) of Theorem 11.4.6 says that Ap is
regular.
Corollary 11.4.9. If A is the completion of a regular local ring, then Ap is regular for all
prime ideals p ⊂ A.
The Auslander-Buchsbaum-Serre theorem also allows us to prove the ‘fibrewise criterion
for regularity’.
Theorem 11.4.10. Let ϕ : (A, m) → (B, n) be a flat, local homomorphism of local rings. If
A and B/mB are regular, then so is A.
Proof. Set d = dim A and e = dim B/mB, so that m = (x1 , . . . , xd ) and n/mB = (ȳ1 , . . . , ȳe )
for some yi ∈ B having images ȳi ∈ B/mB. Then n = (ϕ(x1 ), . . . , ϕ(xd ), y1 , . . . , ye ) so we
are done by Theorem 8.3.30.
Corollary 11.4.11. If A is a regular local ring then A[[x1 , . . . , xn ]] is regular for any n ≥ 1.
Proof. The map A → A[[x1 , . . . , xn ]] is flat since it is the direct limit of a system of maps
to free modules. Moreover, the fibre of this map is k[[x1 , . . . , xn ]] which is regular by Theo-
rem 6.2.8(c). Hence Theorem 11.4.10 applies.
Corollary 11.4.12. If A is a regular local ring then A[x1 , . . . , xn ] is also regular for any
n ≥ 1, meaning for all prime ideals q ⊂ A[x1 , . . . , xn ], the local ring A[x1 , . . . , xn ]q is regular.
Proof. By induction, it’s enough to prove the A[x] case. Choose a prime ideal q ⊂ A[x]
and set p = q ∩ A. Then the map Ap → A[x]q factors through Ap → Ap [x] and thus is
flat. Meanwhile, the fibre is (Ap /pAp [x])q = k(p)[x]q which is regular of dimension 1 since
k(p)[x] is a PID, for example, or using Noether normalization (Theorem 2.4.1). Therefore
Theorem 11.4.10 implies A[x]q is regular.
Remark. Let A = k[x1 , . . . , xn ]. Since Ap is regular by the above, for any finitely generated
A-module M and any prime p ⊂ A, pdAp (Mp ) ≤ dim Ap ≤ n. So if
0 → Q → Fn−1 → · · · → F1 → F0 → M → 0
220
11.4. Projective Dimension Chapter 11. Ring Homology
Corollary 11.4.13. For a regular local ring (A, m) and any nonzero x ∈ m, the quotient
A/(x) is regular if and only if x 6∈ m2 .
Theorem 11.4.14 (Jacobian Criterion). Let k be a perfect field and set A = k[x1 , . . . , xn ].
For a prime p = (F1 , . . . , Fm ) ⊂ A of height h = ht(p), let B = A/p, choose a prime q ⊂ B,
set ξi = x̄i ∈ B/q and define the Jacobian
∂Fi
J = J(ξ1 , . . . , ξn ) := (ξ1 , . . . , ξn ) .
∂xj
Example 11.4.15. Let A = k[x, y, s, t] and p = (xy − st), which has height 1. Then for the
dimension 3 ring B = A/p,
y
x
J =
−t
−s
so clearly (1) is satisfied modulo any prime q. Moreover, rank J = 1 if and only if J 6= 0
mod q, so we see that there is only an isolated singularity at q = (x, y, s, t).
Remark. The Jacobian matrix J determines a presentation of the module of Kähler differ-
entials of B from Example 3.6.8:
J
Bn −
→ B m → Ω1A/k → 0.
221
11.5. Depth and Cohen-Macauley Rings Chapter 11. Ring Homology
Proof. Suppose HomA (A/I, M ) = Ext0A (A/I, M ) 6= 0. Choose a nonzero map f : A/I → M
and set x = f (1) ∈ M . Then for any y ∈ I, 0 = f (y) = yf (1) = yx so Ix = 0. Thus
every element of I is a nonzero divisor for M and therefore every regular sequence on M
has length
S 0. Conversely, if every regular sequence has length 0, then by Proposition 3.6.9,
I ⊆ p∈Ass(M ) p. By prime avoidance (Lemma 6.2.14), I ⊆ p for some p ∈ Ass(M ). This
means A/I → A/p ,→ M determines a nonzero element of HomA (A/I, M ). So the equality
holds when either side is zero.
Now induct on t; let (y1 , . . . , yt ) be a regular sequence in I for M of maximal length.
Applying HomA (A/I, −) to the short exact sequence
y1
0→M −
→M →
− M1 → 0
Note that (ȳ2 , . . . , ȳt ) is a maximal length regular sequence on M1 = M/y1 M , so by induction
ExtiA (A/I, M1 ) = 0 for i < t − 1 and ExttA (A/I, M1 ) 6= 0. Therefore the long exact sequence
implies ExtiA (A/I, M ) = 0 for i < t − 1 and the map
t−1 y1
ExtA → Extt−1
(A/I, M ) − A (A/I, M )
Definition. If (A, m) is a local noetherian ring and M is a finitely generated A-module, the
depth of M is defined to be the m-depth, depth(M ) = depthm (M ).
222
11.5. Depth and Cohen-Macauley Rings Chapter 11. Ring Homology
Remark. For any regular sequence (y1 , . . . , yt ) in I on M , iterating the above argument
yields an isomorphism
ExttA (A/I, M ) ∼
= HomA (A/I, M/(y1 , . . . , yt )M ).
Corollary 11.5.2. For any ideal I ⊂ A and any collection of finitely generated A-modules
{Mj }j∈J , with IMj 6= Mj for any j ∈ J, we have
!
M
depthI Mj = min{depthI (Mj ) | j ∈ J}.
j∈J
Proof. First assume depth(A) = 0. We claim M is free – by Corollary 11.5.2, this would
then imply depth(M ) = depth(A) = 0. Suppose that n = pdA (M ) > 0. Given a minimal
free resolution of M , say
ϕn ϕn−1 ϕ1 ϕ0
0 → Fn −→ Fn−1 −−−→ · · · −→ F0 −→ M → 0,
we know ϕi (Fi ) ⊆ mFi−1 for all i. Now depth(A) = 0 if and only if HomA (k, A) = 0, the
latter of which is only possible if there is some nonzero z ∈ A with mz = 0. Consider the
element
z
0
u = .. ∈ Fn .
.
0
Then ϕn (u) = zϕn (e1 ) ⊆ zmFn−1 = (mz)Fn−1 = 0, but since n > 0 this contradicts
injectivity of ϕn . Therefore n must be 0 in the first place, but this of course implies M ∼ = F0
is free and as a consequence depth(M ) = 0 as explained above.
Now assume depth(M ) = 0, so that HomA (k, M ) 6= 0. Let t = depth(A) and choose a reg-
ular sequence (y1 , . . . , yt ) in m for A. Notice that since n = pdA (M ), TorA
i (A/(y1 , . . . , yt ), M ) =
A
0 for all i > n and Torn (A/(y1 , . . . , yt ), M ) 6= 0. For a minimal free resolution F• → M as
above, TorA n (A/(y1 , . . . , yt ), M ) can be computed as the nth homology of
TorA
n (A/(y1 , . . . , yt ), M ) 6= 0.
223
11.5. Depth and Cohen-Macauley Rings Chapter 11. Ring Homology
Since (y1 , . . . , yt ) is a regular sequence, this is acyclic by Theorem 11.3.10. That is,
TorA
i (A/(y1 , . . . , yt ), M ) = 0 for all i > t.
Since depth(M ) = 0, there is some nonzero z ∈ M with mz = 0, so the same trick shows
TorA
t (A/(y1 , . . . , yt ), M ) 6= 0. Hence t = n.
The remaining cases are when d = depth(M ) > 0 and t = depth(A) > 0. Note that
HomA (k, M ) = 0 implies m 6∈ Ass(M ). By prime avoidance (Lemma 6.2.14), there is
some y ∈ M lying outside the union Ass(M ) ∪ Ass(A) – note that these sets are finite
by Theorem 3.6.7. Then Proposition 3.6.9 implies y is a nonzero divisor for M and for A.
Applying − ⊗A M to the short exact sequence
y
0→A→ − A→0
− A→
and the corresponding long exact sequence in Tor shows that TorA
i (A, M ) = 0 for all i > 0.
On the other hand, let F• → M be a minimal free resolution and apply − ⊗A A:
0 → F n → · · · → F 1 → F 0 → M ⊗A A → 0
where F i = Fi ⊗A A is a free A-module. Note that the Tor vanishing condition on A means
this sequence is exact. We have thus shown that pdA (M ) = pdA (A ⊗A M ). By induction,
but since A = A/(y) for a nonzero divisor y, depth(A) = depth(A)−1 and depth(M ⊗A A) =
depth(M ) − 1. This proves the formula.
Corollary 11.5.5. Let A be a regular local ring and I ⊂ A an ideal. Then B = A/I is
Cohen-Macauley if and only if pdA (B) = dim A − dim B.
Proof. By Theorem 11.5.3, depth(B)+pdA (B) = depth(A) = dim A, which can be rewritten
pdA (B) = dim A − depth(B). Therefore B is CM if and only if depth(B) = dim B if and
only if pdA (B) = dim A − dim B
The following is an analogue of Theorem 8.3.30 for depth, and also provides a version of
Theorem 11.4.10 for the Cohen-Macauley condition. In this way it justifies the inclusion of
depth on our list of types of dimension in ring theory.
224
11.5. Depth and Cohen-Macauley Rings Chapter 11. Ring Homology
Proof. Set k = A/m, t = depth(A) and suppose t > 0. Choose a maximal regular sequence
(x1 , . . . , xt ) in A. Consider the short exact sequence
1x
0 → A −→ A→
− A1 → 0.
It remains to show the base case, when t = depth(A) = 0. Fix some y ∈ B whose image
ȳ ∈ B/mB is a nonzero divisor. We’ll be done by induction, since dim B/mB < dim B, if
we can show A → B/yB is flat and y is a nonzero divisor in B. We begin with the second
statement. We claim that y is a nonzero divisor on B/mn B for all n ≥ 1. Indeed, the base
case is by hypothesis and in general, applying − ⊗A B to the exact sequence
0 → k` ∼
= mn /mn+1 → A/mn+1 → A/mn → 0
(for some ` < ∞) yields the exact sequence in both rows of the following commutative
diagram:
0 (B/mB)` B/mn+1 B B/mn B 0
ȳ y y
Then by the Snake Lemma and induction, the kernel of the middle column is 0, so y is a
nonzero divisor on B/mn+1 B. Suppose yz = 0 for some z ∈ B. T Then ȳz̄ = 0 in B/mn B
for all n ≥ 1 so by the claim, z̄ = 0 in every B/mn B. Thus z ∈ ∞ n
n=1 m B but by Krull’s
intersection theorem (Corollary 5.4.8), this is 0 so y is a nonzero divisor in B. Finally, to
show flatness, by Theorem ?? it suffices to prove TorA 1 (k, B/yB) = 0. Since y is a nonzero
divisor in B, the sequence
y
0→B→ − B→ − B/yB → 0
is exact, so applying − ⊗A k and taking Tor yields an exact sequence
ȳ
0 = TorA A
1 (k, B) → Tor1 (k, B/yB) → B/mB →
− B/mB.
(The 0 on the left follows from Theorem ??.) By assumption, multiplication by ȳ is injective
on B/mB, so exactness implies TorA 1 (k, B/yB) = 0.
225
11.5. Depth and Cohen-Macauley Rings Chapter 11. Ring Homology
0 = M` ⊆ M`−1 ⊆ · · · ⊆ M0 = M
of M are all isomorphic to A/m = k. By definition of depth, ExtiA (k, N ) = 0 for all
i < depth(A), so for each short exact sequence
the long exact sequence in Ext implies ExtiA (Mj , N ) = 0. In particular, ExtiA (M, N ) = 0.
Now induct; if dim M > 0 then by the same argument it’s enough to show ExtiA (A/p, N ) = 0
for all i < depth(N ) − dim M and all primes p in a prime filtration of M . Fix such a p.
Then by definition, dim A/p ≤ dim M . Choose x ∈ m which does not lie in p and consider
the short exact sequence
x
0 → A/p →
− A/p →
− A/(p, x) → 0.
By Corollary 6.2.16, dim A/(p, x) = dim A/p − 1 so by induction, ExtiA (A/(p, x), N ) = 0 for
all i < depth(N ) − dim A/p + 1. Applying HomA (−, N ) to the short exact sequence above
yields
x
0 = ExtiA (A/(p, x), N ) → ExtiA (A/p, N ) →
− ExtiA (A/p, N ) → ExtiA (A/(p, x), N ) = 0
whenever i < depth(N ) − dim A/p. Since x ∈ m r p, Nakayama’s lemma (1.2.1) implies
ExtiA (A/p, N ) = 0 for these i. Since p was any prime in a prime filtration of M , we are
done.
Corollary 11.5.10. If (A, m, k) is a local noetherian ring and M is a finitely generated
A-module, then for any p ∈ Ass(M ), depth(M ) ≤ dim A/p.
Corollary 11.5.11. If A is Cohen-Macauley, then
226
11.5. Depth and Cohen-Macauley Rings Chapter 11. Ring Homology
Example 11.5.12. By Corollary 11.5.11, the ring k[x, y, z]/(x)∩(y, z) is not Cohen-Macauley
since it has two minmal primes, (x) and (y, z), of different codimensions.
Example 11.5.13. Similarly, Corollary 11.5.11 implies that k[x, y]/(x2 , xy) is not Cohen-
Macauley because there is an associated prime, (x, y), which is not minimal – (x) is the only
minimal prime in this ring.
Proposition 11.5.14. Let (A, m) → (B, n) be a local homomorphism of local noetherian
rings such that A is regular and B is a finite A-module. Then B is Cohen-Macauley if and
only if B is a free A-module.
Proof. Since A is regular, Theorem 11.4.6 shows pdA (B) < ∞. Then by the Auslander-
Buchsbaum formula (Theorem 11.5.3), pdA (B) + depth(B) = depth(A). Next, since A → B
is finite, Theorem 2.3.7 implies dim A = dim B. But A is also Cohen-Macauley (Corol-
lary 11.4.7) so depth(A) = dim A by Corollary 11.5.4. Thus
We can now show that the definition of Cohen-Macauley is independent of the choice of
regular sequence on M .
Proposition 11.5.15. Let (A, m) be a d-dimensional Cohen-Macauley ring containing a field
and suppose x1 , . . . , xd is a system of parameters. Then (x1 , . . . , xd ) is a regular sequence.
Proof. To begin, note that A is CM if and only if A b is CM – this follows from Theorem 11.5.6.
Also note that (x1 , . . . , xd ) is a system of parameters if and only if dim A/(x1 , . . . , xd ) = 0.
Then A/(x b∼
b 1 , . . . , xd )A = A/(x1 , . . . , xd ) so because dim A
b = dim A, x1 , . . . , xd is a system of
parameters for A. Applying Theorem 11.3.10, we have:
b
b ⇐⇒ K(x1 , . . . , xd ; A)
(x1 , . . . , xd ) is a regular sequence on A b is acyclic
⇐⇒ K(x1 , . . . , xd ; A) ⊗A A b is acyclic (A → A
b is flat)
⇐⇒ K(x1 , . . . , xd ; A) is acyclic
⇐⇒ (x1 , . . . , xd ) is a regular sequence on A.
Thus we may reduce to the case when A is a complete CM ring. By Cohen’s third structure
theorem (6.3.11), there is a finite ring extension B = k[[x1 , . . . , xd ]] → A, where k ⊆ A is a
227
11.5. Depth and Cohen-Macauley Rings Chapter 11. Ring Homology
coefficient field. By Corollary 11.4.11, B is regular and because A is CM, Proposition 11.5.14
implies A is a free B-module (beware that A and B have switched roles here compared
to that proposition). On the other hand, K(x1 , . . . , xd ; B) is acyclic by Theorem 11.3.10,
but since B → A is flat, K(x1 , . . . , xd ; A) = K(x1 , . . . , xd ; B) ⊗B A is still acyclic. Hence
Theorem 11.3.10 again implies that (x1 , . . . , xd ) is a regular sequence on A.
The last result holds in general but requires a more flexible version of Cohen’s structure
theorems when A does not contain a field.
Proposition 11.5.16. Let A be a local Cohen-Macauley ring. Then every non-refinable
chain of prime ideals
p0 ( p1 ( · · · ( pk = m
has length k = dim A.
Proof. We induct on d = dim A. If d = 0, there’s only one prime ideal, m, so every chain of
primes is trivial. If d > 0, suppose p0 ( · · · ( pk = m is a chain of primes. Since p1 is not a
minimal prime, the prime avoidance lemma (6.2.14) says that
[
p1 6⊂ q.
q⊂A
q minimal
Pick x ∈ p1 that does not lie in any minimal prime q. By Corollary 11.5.11, the minimal
primes of A are exactly the associated primes, so Proposition 3.6.9 says x is not a zero divisor
on A. Thus A/(x) is also CM and dim A/(x) = d − 1. Consider the chain of primes
p1 ( p2 ( · · · ( pk = m
in A/(x), where pi = pi /(x) for each 1 ≤ i ≤ k. By induction, this chain must have length
k − 1 = dim A/(x) = d − 1 so k = d as desired.
For an ideal I in a noetherian ring A, we extend the definition of height to I by:
The proof of Lemma 2.3.6(b) goes through for any ideal I, so that ht(I) + dim A/I ≤ dim A.
Moreover, Proposition 11.5.16 gives us:
Corollary 11.5.17. If A is a Cohen-Macauley ring, then for any ideal I ⊂ A, ht(I) +
dim A/I = dim A.
Example 11.5.18. Let B = k[x, y1 , . . . , yn ](x,y1 ,...,yn ) , let J = (x) ∩ (y1 , . . . , yn ) and set
A = B/I. Then the minimal primes of A are (x̄) and I = (ȳ1 , . . . , ȳn ) and ht(I) = 0, but
dim A/I = dim k[x̄] = 1 so ht(I) + dim A/I = 1 < dim A = n as long as n ≥ 2. Therefore A
is not Cohen-Macauley.
Theorem 11.5.19 (Unmixedness). Let A be a local Cohen-Macauley ring and suppose I =
(x1 , . . . , xn ) ⊆ m such that ht(I) = n. Then (x1 , . . . , xn ) is a regular sequence and A/I is
Cohen-Macauley.
228
11.5. Depth and Cohen-Macauley Rings Chapter 11. Ring Homology
Proof. First note that in general, Krull’s height theorem (6.2.10) implies ht(I) ≤ n. As-
suming for the moment that (x1 , . . . , xn ) is a regular sequence, Proposition 11.5.15 says
that every maximal regular sequence in A has length depth A = dim A, so it follows that
A/I is also CM. Thus it suffices to show (x1 , . . . , xn ) is a regular sequence. If n = 1, then
ht((x1 )) = 1 is equivalent to saying x1 is not contained in any minimal prime of A, but since
A is CM, Corollary 11.5.11 and Proposition 3.6.9 show x1 is not a zero divisor. For n > 1,
we induct. Suppose x1 lies in some minimal prime p ⊂ A. Then for any prime pk which is
minimal over I, we have the following non-refinable chain of primes in A/(x1 ):
p/(x1 ) ( p1 /(x1 ) ( · · · ( pk /(x1 ).
Then k = ht(pk ) = n, but pk /(x1 ) is minimal with respect to primes of A/(x1 ) containing
(x2 , . . . , xn )/(x1 ), contradicting Krull’s height theorem (6.2.10). Hence x1 cannot be con-
tained in any minimal prime of A, but since A is CM, the argument in the base case shows
x1 is a nonzero divisor on A. That is, A/(x1 ) is CM and ht((x̄2 , . . . , x̄n )) = n − 1. Induction
finishes the proof.
Corollary 11.5.20. If A is a local Cohen-Macauley ring then Ap is Cohen-Macauley for all
primes p ⊂ A.
Proof. By Lemma 2.3.6(a), dim Ap = ht(p). Set n = ht(p). We claim there are x1 , . . . , xn ∈ p
such that ht((x1 , . . . , xn )) = n. Assuming this is possible, the unmixedness theorem then
implies (x1 , . . . , xn ) is a regular sequence in A, and because A → Ap is flat, (x1 , . . . , xn )
will also be a regular sequence in Ap . Hence dim Ap ≥ depth(Ap ) ≥ n = dim Ap so by
Corollary 11.5.4, Ap is CM. Finally, we can find such elements x1 , . . . , xn ∈ p as follows. By
the prime avoidance lemma (6.2.14), there exists an x1 ∈ p that does not lie in any minimal
prime of A. Moreover, there are a finite number of minimal primes containing (x1 ) and by
Krull’s height theorem (6.2.10), each has height 1. So choose x2 ∈ p avoiding these primes
and repeat the argument to construct the rest of the sequence (x1 , . . . , xn ).
Definition. A ring A is catenary if for all primes p ⊂ q in A, every non-refinable chain
of prime ideals between them,
p ( p1 ( · · · ( p` ( q,
has the same length.
Theorem 11.5.21. Suppose A = B/I for some ideal I ⊂ B. Then if B is Cohen-Macauley,
A is catenary.
Proof. For any prime ideal p = q/I ⊂ A, Corollary 11.5.20 shows that Bq is still CM, but
by Proposition 11.5.16, every maximal chain of primes in Bq has the same length. Therefore
the same holds in Ap = Bq /Iq , so A is catenary.
The next result says that the property of being Cohen-Macauley is an ‘open condition’
on the set of primes of a regular ring. This will be made precise once we define Spec(A) and
the Zariski topology in Section 14.1, but for now, define
V (I) := {p ⊂ A | I ⊆ p is prime}
for any ideal I ⊂ A.
229
11.5. Depth and Cohen-Macauley Rings Chapter 11. Ring Homology
Proposition 11.5.22. Suppose B is a regular ring and A = B/q for a prime ideal q ⊂ B.
Then the set
C = {p ⊂ A | p is prime and Ap is not CM}
is equal to V (I) for some ideal I ⊂ A.
where c = ht(q) = dim B −dim A by Corollary 11.5.17. If ϕ is the zero map then pdB (A) = c
so the Auslander-Buchsbaum formula (Theorem 11.5.3) and Corollary 11.5.4 show A is
CM. Otherwise, it’s enough to show that after localizing at an arbitrary prime p ⊂ B, the
r
corresponding map ϕp : Bp c+1 → Bprc splits. After choosing a basis of these free Bp -modules,
ϕp is split if and only if the ideal of k × k minors, where k is the rank of ϕp , is equal to the
unit ideal. Let I be this minor ideal. Then we have shown that the set of primes p ⊂ A for
which Ap is not CM is precisely V (I).
To finish, we list some nontrivial examples and properties of Cohen-Macauley rings.
Details can be found in papers by the cited authors.
Proposition 11.5.24 (Hochster-Huneke). Suppose A and B are local noetherian ring such
that A contains a field and B is regular. If there exists a split local homomorphism A ,→ B,
then A is Cohen-Macauley.
230
11.6. Gorenstein Rings Chapter 11. Ring Homology
We will show that this definition is equivalent to the more common definition of A having
finite injective dimension.
Proof. By Theorem 6.2.8, A is regular if and only if m = (x1 , . . . , xd ) for a system of param-
eters x1 , . . . , xd , but in general a maximal ideal is irreducible, so A is Gorenstein.
Suppose (A, m) is a local noetherian ring and x1 , . . . , xd is a system of parameters. Then A
is Gorenstein if and only if 0 is an irreducible ideal in the 0-dimensional ring A/(x1 , . . . , xd ).
So to begin our study of Gorenstein rings, let’s characterize when a 0-dimensional local
noetherian ring is Gorenstein. Recall that a local noetherian ring has dimension 0 if and
only if it’s artinian (this is Hopkins’ theorem, 3.5.5).
Example 11.6.2. Let k be a field. Then A = k[x, y]/(x, y)2 is 0-dimensional but not
Gorenstein since (x, y)2 = (x, y 2 ) ∩ (x2 , y) is not irreducible in k[x, y].
is Gorenstein.
Remark. One might hope to classify all Gorenstein rings of any dimension by the tech-
niques which we will describe for local artinian rings. However, it turns out that there is no
classification of Gorenstein rings of the form k[x1 , . . . , xn ]/I for n ≥ 4.
Recall the following definitions and basic properties from ring theory.
Definition. The socle of a local ring (A, m) is the set soc(A) = {x ∈ A | xm = 0}.
231
11.6. Gorenstein Rings Chapter 11. Ring Homology
Proof. (a) and (b) are standard. For (c), take a nonzero ideal I ⊂ A. Then for a nonzero
element x ∈ I, either xm = 0, so x ∈ soc(A) ∩ I, or by Theorem 3.5.6, xmn = 0 for some
minimal n ≥ 2. In the latter case, xmn−1 6= 0 but xmn−1 ⊆ soc(A) ∩ I. Thus soc(A) is an
essential ideal of A.
Theorem 11.6.6. A local artinian ring (A, m, k) is Gorenstein if and only if dimk soc(A) =
1.
0 → B bd → · · · → B b1 → B → A → 0,
is necessarily of length d since pdB (A) = dim B − dim A by Theorem 11.5.19 and Corol-
lary 11.5.5. Then tensoring with k produces a resolution with all zero maps, so TorB ∼
d (k, A) =
bd
k . On the other hand, the Koszul complex for A as a B-module is
ϕ
→ Ad → · · · → Ad → A → k ⊗B A → 0
0→A−
so by definition,
TorB
d (k, A) = ker ϕ = ker(a 7→ (±xi )a)
= {a ∈ A | axi = 0 for all 1 ≤ i ≤ d}
= soc(A).
This proves:
Proposition 11.6.7. If A is a local artinian ring that is a quotient of a regular local ring
of dimension d with residue field k, then dimk soc(A) = bd , the dth Betti number of A.
Corollary 11.6.8. A local artinian ring A that is a quotient of a regular local ring of
dimension d is Gorenstein if and only if bd = 1.
232
11.6. Gorenstein Rings Chapter 11. Ring Homology
Theorem 11.6.10. Let (B, n, k) be a regular local ring of dimension n, I ⊂ B an ideal and
suppose A = B/I is Cohen-Macauley of dimension d. Set c = ht(I) = n − d. Then A is
Gorenstein if and only if dimk TorB
c (k, A) = 1.
so bc = dimk TorB
c (k, A) is the last Betti number in a minimal free resolution of A:
0 → B bc → B bc−1 → · · · → B b1 → B → A → 0.
equals the last Betti number in K• (y1 , . . . , yd ; B) ⊗ F• . The resolutions K• (y1 , . . . , yd ; B) and
F• have lengths d and c, respectively, and the last free module in the tensor complex, in
dimension n = d + c, is exactly Kd (y1 , . . . , yd ; B) ⊗B Fc = B ⊗B Fc ∼ = Fc . Therefore we have
TorB (k, A) ∼
= Tor B
(k, A/(ȳ 1 , . . . , ȳ d )) so we conclude b c = 1 if and only if A/(ȳ1 , . . . , ȳd ) is
c n
Gorenstein, by Theorem 11.6.6. However, the same argument shows that bc = 1 if and only
if dimk TorB c (k, A/(z1 , . . . , zd )) = 1 for any system of parameters z1 , . . . , zd in A, which is
equivalent to (z1 , . . . , zd ) being an irreducible ideal in A for any such zi .
233
11.6. Gorenstein Rings Chapter 11. Ring Homology
b
a
with vertices labeled with variables a, . . . , e. Let I = (ab, bc, cd, de, ea) be the edge ideal in
B = k[a, b, c, d, e](a,b,c,d,e) . Then ht(I) = 3 so A = B/I is a local ring of dimension 2 with
free B-resolution
0 → B → B 5 → B 5 → B → A → 0.
Hence by the Auslander-Buchsbaum formula (Theorem 11.5.3), A is Cohen-Macauley. Fur-
ther, the last Betti number of A is 1 so Theorem 11.6.10 shows A is Gorenstein. However,
A is not a complete intersection since I is 5-generated.
Example 11.6.14. Let k = C and choose a polynomial f ∈ C[y1 , . . . , yn ] such that f (0) = 0.
The polynomial ring C[X1 , . . . , Xn ] acts on such f by partial differentiation: Xi · f = ∂y∂ i f .
Set
(f )⊥ := {G ∈ C[X1 , . . . , Xn ] | G · f = 0}.
Then A = C[X1 , . . . , Xn ]/(f )⊥ is a 0-dimensional Gorenstein ring and one can prove that
every 0-dimensional Gorenstein ring is isomorphic to such a quotient. For example, if f =
y12 + y22 + y32 then (f )⊥ = (X1 X2 , X1 X3 , X2 X3 , X12 − X22 , X12 − X32 ). On the other hand, for
g = y1 y2 y3 the corresponding ideal is (g)⊥ = (X12 , X22 , X32 ) and A = C[x1 , x2 , x3 ]/(g)⊥ is a
complete intersection in this case. It is an open question which f give quotients that are
complete intersections.
Theorem 11.6.15. Let (A, m, k) → (B, n, `) be a flat, local homomorphism of local rings.
Then B is Gorenstein if and only if A and B/mB are Gorenstein.
Proof. If dim A or dim B > 0, one can induct using Theorem 11.6.10 so we will reduce to
the case that dim A = dim B = 0. By Theorem 11.6.6, we must show dim soc(B) = 1 if and
only if dim soc(A) = 1 and dim soc(B/mB) = 1. First assume A and B/mB are Gorenstein.
Choose a ∈ A such that soc(A) = (a) ∼= k and b ∈ B such that soc(B/mB) = (b̄) ∼= k. We
claim soc(B) = (ab). On one hand, we have
234
11.6. Gorenstein Rings Chapter 11. Ring Homology
Thus x ∈ aB, say x = ay for some y ∈ B. Then ayn = 0 implies yn ⊆ AnnB (a) =
AnnA (a)B = mB. Hence y ∈ soc(B/mB), so by hypothesis, y = zb + π for some z ∈ B and
π ∈ mB. This implies
(2) ǩ ∼
= k.
(3) If λ(M ) denotes the length of a prime filtration of M , then λ(M ) = λ(M̌ ).
ˇ ∼
(4) For any A-module M , we have M̌ = M.
Proof. (1) E is injective.
(2) Note that HomA (k, E) ∼ = soc(E) as k-vector spaces. Moreover, if dim soc(E) ≥ 2,
consider the essential extension
k ,−→ E
1 7−→ x ∈ soc(E).
Then there is a y ∈ soc(E) linearly independent from x so (y) ∩ (x) = 0, contradicting the
fact that E is an essential extension of k. Hence dim soc(E) = 1.
(3) Inducting on λ(M ), note that AssA (M ) = {m}, so there exists an embedding k ,→ M .
Let
0 → k → M → M1 → 0
be the corresponding short exact sequence. Since λ is additive, λ(M ) = λ(k) + λ(M1 ) =
1 + λ(M1 ) and by (1),
0 → ǩ → M̌ → M̌ 1 → 0
is also exact and thus λ(M̌ ) = λ(ǩ) = λ(M̌ 1 ). So by induction, λ(M̌ 1 ) = λ(M1 ). Moreover,
(2) implies λ(ǩ) = λ(k) so it follows that λ(M̌ ) = λ(M ).
235
11.6. Gorenstein Rings Chapter 11. Ring Homology
0 → A → E → E/A → 0
yields λ(E) = λ(A) = λ(E/A), but Proposition 11.6.16(4) says λ(A) = λ(Ǎ) = λ(HomA (A, E)) =
λ(E) since E is injective. Thus λ(E/A) = 0, which implies E/A = 0.
(b) =⇒ (c) =⇒ (d) are trivial.
(d) =⇒ (a) It is well-known that over a local noetherian ring, every injective A-module
is isomorphic to a direct sum of copies of E. In particular, a finite injective resolution of A
over itself may be taken to be of the form
0 → A → E b0 → E b1 → · · · → E bn → 0.
so in particular pdA (E) < ∞. Further, the Auslander-Buchsbaum formula (Theorem 11.5.3)
implies pdA (E)+depth(E) = depth(A), but depth(E) and depth(A) are both 0 so pdA (E) =
0 as well. In particular, E ∼= Ab for some b ≥ 1. On the other hand, λ(E) = λ(A) by
additivity of λ, so we must have b = 1. Hence A is Gorenstein.
Theorem 11.6.18. Let f1 , . . . , fd be a system of parameters in the power series ring k[[x1 , . . . , xd ]]
and set A = k[[x1P
, . . . , xd ]]/(f1 , . . . , fd ). Then A is a 0-dimensional complete intersection.
d
Further, if fi = j=1 aij xj for some aij ∈ k, then soc(A) is generated by the image of
det(aij ) in k.
236
Part III
Algebraic Geometry
237
Chapter 12
Algebraic Varieties
One of the principal mathematical offshoots of commutative algebra is the study of algebraic
geometry. Many of the definitions and results in Part I have geometric interpretations which
give further insight into their meaning and importance. However, the beauty of algebraic
geometry lies in how it transports ideas in both directions: the algebraic to the geometric
and the geometric to the algebraic.
In this chapter, we begin the study of algebraic geometry by exploring algebraic varieties.
For awhile, this will illustrate enough of the spirit of algebraic geometry for our purposes,
but eventually we will want to generalize to schemes; this is done in Chapter 14.
238
12.1. Affine Algebraic Varieties Chapter 12. Algebraic Varieties
As sets, An = k n , but the new notation carries with it the implication that An is viewed
geometrically.
Remark. Alternatively, for any field k ⊆ K ⊆ k̄, one can define Ank (K) to be the fixed points
of Ank under the action of the Galois group Gal(k̄/K). In particular, Ank = Ank (k) = (k̄ n )Gk
where Gk = Gal(k̄/k) is the absolute Galois group of the field k.
Definition. For a polynomial f ∈ A, define its zero set (or zero locus) to be
Z(f ) = {P ∈ An | f (P ) = 0}.
Notice that if I = (F) is the ideal of A generated by F, then Z(F) = Z(I). By Hilbert’s basis
theorem (3.3.2), there exists a finite subset {f1 , . . . , fr } ⊆ F such that Z(F) = Z(f1 , . . . , fr ).
By the remark, it is equivalent to say X is a zero set if X = Z(I) for some ideal I ⊂ A.
Thus the operation Z(·) takes a subset of a ring and assigns to it a geometric space. There
is a dual notion:
239
12.1. Affine Algebraic Varieties Chapter 12. Algebraic Varieties
5 The algebraic set Z(y, y − x2 ) = Z(x, y) consists of just the point (0, 0) in A2k :
Z(y − x2 )
Z(y)
Z(y, y − x2 )
Definition. If X = Z(S) ⊆ Ank (k̄) is an algebraic set and K is a field such that k ⊆ K ⊆ k̄,
define the K-points of X by X(K) := X ∩ Ank (K) = X GK , where GK = Gal(k̄/K).
Moreover, we say X is defined over K if J(X) has a generating set consisting of elements
of K[t1 , . . . , tn ].
240
12.1. Affine Algebraic Varieties Chapter 12. Algebraic Varieties
(c) Z(J(X)) ⊇ X.
(d) J(Z(I)) ⊇ I.
(f ) J(Z(J(X))) = J(X).
241
12.1. Affine Algebraic Varieties Chapter 12. Algebraic Varieties
(b) Z(J(X)) = X.
Proof. (a) Since X ⊆ X, we immediately get J(X) ⊇ J(X) by Lemma 12.1.2(a). On the
other hand, if f ∈ J(X), f (P ) = 0 for all P ∈ X. In other words, X ⊆ Z(f ) but Z(f ) is
closed by definition, so Z(f ) ⊇ X. Thus f ∈ J(X).
(b) X is algebraic by definition so there exists some ideal I ⊂ A such that Z(I) = X. Now
by (a), Z(J(X)) = Z(J(X)) = Z(J(Z(I))) which by Lemma 12.1.2(e) equals Z(I) = X. So
Z(J(X)) = X as required.
Example 12.1.4. Let k = R and consider f = t12e1 + . . . + tn2en + 1 ∈ R[t1 , . . . , tn ] for some
ei ∈ N. Then Z(f ) = ∅ so J(Z(f )) = A and this holds for all choices of ei ∈ N. However,
f does not generate the whole ring A, so this shows JZ is not always the identity on ideals
of A. Hilbert’s Nullstellensatz, proven below, says that this correspondence is bijective on
radical ideals when k is an algebraically closed field.
Theorem 12.1.5 (Hilbert’s Nullstellensatz, Geometric Version). If k is algebraically closed,
then J(Z(I)) = r(I) for every ideal I ⊂ A.
Proof. By Lemma 12.1.2(d), J(Z(I)) ⊇ I and by Lemma 12.1.1, J(Z(I)) is radical. To-
gether, these imply J(Z(I)) ⊇ r(I). For the reverse containment, take f ∈ J(Z(I)). Since
r
P∈
A is noetherian, I = (f1 , . . . , fm ) for a finite set of polynomials f1 , . . . , fm
m
I. We must
show that there exist g1 , . . . , gm ∈ A and r ∈ N so that we can write f = i=1 fi gi .
Consider the ideal I 0 = (f1 , . . . , fm , 1 − tn+1 f ) ⊂ A[tn+1 ] = k[t1 , . . . , tn=1 ]. If P 0 =
(P, an+1 ) is a point in An+1 , with P ∈ An , assume P 0 ∈ Z(I 0 ). Then fi (P 0 ) = fi (P ) = 0 for
all 1 ≤ i ≤ m. Since I = (f1 , . . . , fm ), this means P ∈ Z(I). Then f ∈ J(Z(I)) implies
f (P ) = 0, but 1 − tn+1 f (P ) = 1 6= 0, contradicting P 0 ∈ Z(I 0 ). Therefore it must be that
Z(I 0 ) = ∅, so I 0 6⊂ mP 0 for any P 0 ∈ An+1 , where mP 0 is the maximal ideal of A[tn+1 ] given
by
mP 0 = (t1 − a1 , . . . , tn+1 − an+1 ).
By Corollary 2.4.3, the algebraic version of Hilbert’s Nullstellensatz for algebraically closed
fields, {mP 0 | P 0 ∈ An+1 } = MaxSpec k[t1 , . . . , tn+1 ]. We have therefore shown that I 0
is not contained in any maximalPideal of k[t1 , . . . , tn+1 ], but this is only possible if I 0 =
k[t1 , . . . , tn+1 ]. In particular, 1 = m
i=1 fi hi +(1−tn+1 f )hm+1 for some h1 , . . . , hm+1 ∈ A[tn+1 ].
1
Substituting tn+1 = f , we obtain
m
X
1= fi (t1 , . . . , tn )hi t1 , . . . , tn , f1 .
i=1
Hence f ∈ r(I).
242
12.1. Affine Algebraic Varieties Chapter 12. Algebraic Varieties
We have seen that J and Z establish a correspondence, though not always bijective,
between the ideals of A and the closed subsets of An . We next explore how concepts like the
noetherian property and prime ideals translate into geometry.
Example 12.1.6. Affine n-space An equipped with the Zariski topology is a noetherian
space. Indeed, if X1 ⊇ X2 ⊇ X3 ⊇ · · · is a descending chain of Zariski-closed subsets
of An , applying the vanishing ideal operation J to each set gives an ascending chain of
ideals J(X1 ) ⊆ J(X2 ) ⊆ J(X3 ) ⊆ · · · in A = k[t1 , . . . , tn ]. By Hilbert’s basis theorem
(3.3.2), A is noetherian so there is some n ∈ N for which we have J(Xm ) = J(Xn ) for
all m ≥ n. Now by Lemma 12.1.3, Z(J(Xi )) = Xi since each Xi is closed, so we get
Xm = Z(J(Xm )) = Z(J(Xn )) = Xn for all m ≥ n. Hence the chain of subsets stabilizes.
(a) X is quasi-compact.
which is strictly ascending because no finite union of the Ui can equal all of X. Therefore
X is not noetherian.
(b) Let Y1 ⊇ Y2 ⊇ Y3 ⊇ · · · be a descending chain of closed subsets of Y . TThen there
exists a closed set Xn0 ⊆ X such that Yn = Xn0 ∩ X for each n ∈ N. Set Xn = nj=1 Xj0 for
each n. Then each Xn is closed in X and
n
\ n
\
Xn ∩ Y = Xj0 ∩Y = Yj = Yn .
j=1 j=1
Definition. A nonempty topological space X is said to be irreducible if for any two closed
subsets X1 , X2 ⊆ X such that X1 ∪ X2 = X, we have X = X1 or X = X2 .
Proof. Obvious.
243
12.1. Affine Algebraic Varieties Chapter 12. Algebraic Varieties
Proof. (a) For a nonempty open set U ⊆ X, X r U is a proper closed subspace and X =
(X r U ) ∪ U is a decomposition into closed subsets. If X is irreducible, then X 6= X r U
implies X = U . Conversely, suppose X = X1 ∪X2 where X1 , X2 ⊆ X are closed and X1 6= X.
Then X r X1 is a nonempty open subset of X, so by assumption X r X1 = X. Then
X = X1 ∪ X2 =⇒ X r X1 ⊆ X2 =⇒ X r X1 ⊆ X2 =⇒ X ⊆ X2 =⇒ X = X2 .
Therefore X is irreducible.
(b) By Lemma 12.1.8, Y is irreducible if and only if whenever Y ⊆ X1 ∪ X2 for closed
X1 , X2 ⊆ X, we have Y ⊆ X1 or Y ⊆ X2 . Notice that this is equivalent to Y ⊆ X1 ∪ X2
implying Y ⊆ X1 or Y ⊆ X2 . So Y is irreducible precisely when Y is irreducible.
(c) Suppose X = U ∪ V for disjoint, nonempty open sets U, V ⊂ X. Then U = X r V
and V = X r U are both closed, but neither U nor V is equal to the whole space X since
they are disjoint and nonempty.
(d) Suppose f (X) = Y1 ∪ Y2 for closed subsets Y1 , Y2 ⊆ f (X). By continuity, X1 :=
f −1 (Y1 ) and X2 := f −1 (Y2 ) are both closed subsets of X. Then X1 ∪ X2 = X so by
irreduciblity of X, X = X1 or X = X2 . Suppose without loss of generality that X = X1 .
Then f (X) = f (X1 ) = Y1 , so it follows that f (X) is irreducible.
Example 12.1.11. Consider the affine plane A2 . Take f = t1 t2 in k[t1 , t2 ]. Then V(f ) is
the union of the t1 and t2 axes, each of which is an irreducible subspace of A2 :
244
12.1. Affine Algebraic Varieties Chapter 12. Algebraic Varieties
t2 A2
t1
Observe that M is closed under finite unions. We claim that N is empty. If Y ∈ N then Y
is not irreducible (since every irreducible set is an element of M), so there exist Y 0 , Y 00 ⊆ Y
closed such that Y 0 ∪ Y 00 = Y and Y 0 , Y 00 6= Y . Also, at least one of Y 0 , Y 00 is in N in
light of the observation that M is closed under unions. Without loss of generality, assume
Y 0 ∈ N . Continuing this construction inductively, with Y1 = Y and Y2 = Y 0 , we get a
strictly descending chain
Y = Y1 ) Y2 ) Y3 ) · · ·
contradicting the fact that X is noetherian. Therefore N must be empty.
Sm X ∈ M, so there exist closed, irreducible subsets X1 , . . . , Xm ⊆ X such
In particular,
that X = i=1 Xi . We may assume Xi 6⊂ Xj for any i 6= j. Now any irreducible subset
Y ⊆ X is contained in at least one Xi , hence only X1 , . . . , Xm can be irreducible components
of X. In fact, each Xi is an irreducible component since if Y is another irreducible subset
of X such that Xi ⊆ Y , there exists some 1 ≤ j ≤ m for which Xi ⊆ Y ⊆ Xj , but
by assumption this is only possible when i = j. In this case, Xi = Y , proving Xi is an
irreducible component of X for each 1 ≤ i ≤ m.
Definition. Assume X is a noetherian space. If X is irreducible, the dimension of X is
defined by
` ∈ N0 | there exists a chain of closed, irreducible subsets
dim X = sup .
Y0 ( Y1 ( · · · ( Y` with Yi ⊆ X
245
12.1. Affine Algebraic Varieties Chapter 12. Algebraic Varieties
Recall that all algebraic subsets of An are noetherian spaces (with the Zariski topology). We
now define one of the fundamental objects of algebraic geometry.
Definition. An affine algebraic variety over k is an irreducible algebraic subset of An .
Definition. An quasi-affine variety if a nonempty, open subset of an affine variety.
By Proposition 12.1.10, we have that if X is an affine variety, then any nonempty, open
subset U ⊆ X is dense in X. In particular, the Zariski topology on X is not Hausdorff in
general.
Remark. Suppose k is an algebraically closed field. Set A = k[t1 , . . . , tn ]. Then we have
the following correspondence:
246
12.1. Affine Algebraic Varieties Chapter 12. Algebraic Varieties
Definition. For an affine variety X ⊆ An , we define its function field over k to be the field
of fractions Frac k[X], denoted k(X). An element of k(X) is called a rational function on
X. If X is defined over some k ⊆ K ⊆ k̄, then the field K(X) := Frac K[X] is called the
field of K-rational functions on X.
Lemma 12.1.15. For any tower k ⊆ K ⊆ k̄ over which X is defined, K[X] = k̄[X]GK and
K(X) = k̄(X)GK , where GK = Gal(k̄/K) is the absolute Galois group of K.
Proof. (a) By Proposition 3.4.10(b), the set {p ∈ Spec(A) | p ⊇ I} has only finitely many
minimal elements, p1 , . . . , pm . ThenTCorollary 1.1.3 allows us to write r(I) = m
T
i=1 i . By
p
m
Hilbert’s Nullstellensatz, J(Z(I)) = i=1 pi and so by Lemma 12.1.2,
m
! m
\ [
Z(I) = Z(J(Z(I))) = Z pi = Z(pi ).
i=1 i=1
247
12.1. Affine Algebraic Varieties Chapter 12. Algebraic Varieties
Lemma 12.1.9 then says that since J(Z(pi )) = pi is prime, Xi := Z(pi ) is an irreducible
set. We have Xi 6⊂ Xj for any i 6= j since J(Xi ) = pi 6⊃ pj = J(Xj ). Hence by the proof
of Proposition 12.1.13, X1 , . . . , Xm are the irreducible components of J(I) = X. Now by
Krull’s height theorem (6.2.10), ht(pi ) ≤ s for each 1 ≤ i ≤ m. We now exploit the fact that
the polynomial ring has the strong dimension property. One has for each i,
248
12.2. Morphisms of Affine Varieties Chapter 12. Algebraic Varieties
M(X, Y ) = {ϕ : X → Y | ϕ is a morphism}
with ring operations given by pointwise addition and composition. A morphism ϕ ∈ M(X, Y )
is an isomorphism if ϕ is bijective and ϕ−1 is a morphism in the ring M(Y, X).
Example 12.2.1. Let k be a perfect field of characteristic p > 0 and consider the p-power
map
ϕ : A1 −→ A1
x 7−→ xp .
Then ϕ is a bijective morphism (it is explicitly given by a polynomial map and is known to
be an automorphism of any field of characteristic p) but it is not an isomorphism of varieties,
since taking the pth root is not a polynomial map.
Example 12.2.2. For any affine variety X over k, k[X] = M(X, A1 ).
Proposition 12.2.3. For an affine variety X over a field k, the points of X are in bijective
correspondence with Homk (k[X], k).
Proof. The correspondence is given by sending a point P ∈ X to the evaluation homomor-
phism evP : f 7→ f (P ), and sending a k-algebra homomorphism f : k[X] → k to the point
(f (t̄1 ), . . . , f (t̄n )) ∈ k[X], where k[X] = k[t̄1 , . . . , t̄n ].
249
12.2. Morphisms of Affine Varieties Chapter 12. Algebraic Varieties
Lemma 12.2.4. Let X and Y be affine varieties. Then every morphism ϕ ∈ M(X, Y ) is
continuous with respect to the Zariski topologies on X and Y .
ϕ∗ : k[Y ] −→ k[X]
g 7−→ (g ◦ ϕ : X → A1 ).
α∗ : X −→ Y
(t̄1 , . . . , t̄n ) 7−→ (α(t̄1 ), . . . , α(t̄n )).
(d) For any ϕ ∈ M(X, Y ), ϕ∗∗ = ϕ. Likewise, for any α ∈ Homk (k[Y ], k[X]), α∗∗ = α.
Remark. Let ϕ : X → Y be a finite morphism. Then the injection ϕ∗ : k[Y ] ,→ k[X] induces
a field extension k(Y ) ,→ K(X) which is finite and algebraic. Further, if k is algebraically
closed, Proposition 12.1.16 and Theorem 2.3.7 imply that dim X = dim Y .
Example 12.2.7. Let ϕ : A1 → A1 be the map sending x 7→ xn for some n ≥ 2. Then for
all infinite fields k, ϕ is a finite morphism but usually not surjective.
250
12.2. Morphisms of Affine Varieties Chapter 12. Algebraic Varieties
Assuming ϕ is finite, the extension k[X]/ϕ∗ (k[Y ]) is integral and in particular ϕ∗ : k[Y ] →
ϕ∗ (k[Y ]) is an isomorphism of k-algebras. For any given y ∈ Y , Lemma 2.3.4 shows there
exists a prime ideal m ⊂ k[X] with m ∩ ϕ∗ (k[Y ]) = ϕ∗ (my ). Since ϕ∗ (my ) is maximal, this m
must be maximal in k[X] by Lemma 2.3.1(b). Then by Hilbert’s Nullstellensatz (algebraic
version, Theorem 2.4.2), there exists x ∈ X such that m = mx . Thus ϕ∗ (my ) ⊆ m = mx . By
the above equivalences, this implies ϕ(x) = y so we have proven ϕ is a surjection.
Remark. If ϕ : X → Y is a finite morphism, the fibre ϕ−1 (y) is a finite set for all y ∈ Y .
In particular, dim ϕ−1 (y) = 0 as a variety.
251
12.2. Morphisms of Affine Varieties Chapter 12. Algebraic Varieties
HomAffk (X, Y ) ∼
= Homk-alg (k̄[Y ], k̄[X]).
In particular, there is an equivalence of categories between Affk , the affine varieties over k
together with variety morphisms, and (k-alg)op , the opposite category of finitely generated
k-algebras together with k-algebra homomorphisms.
Remark. Suppose k ⊆ K ⊆ k̄ and X and Y are defined over K. If ϕ = (ϕ1 , . . . , ϕm ) : X →
Y is a morphism such that each ϕi ∈ K[t1 , . . . , tn ], we say the morphism is defined over K. In
particular, any ϕ : X → Y induces a morphism of K-rational points, ϕK : X(K) → Y (K),
that is defined over K.
Example 12.2.11. (Interpreting Noether normalization) Assume k is algebraically closed.
Take an affine variety X ⊆ An with associated affine algebra k[X] = k[t1 , . . . , tn ]/J(X) =
k[t̄1 , . . . , t̄n ]. By Corollary 12.1.17 and Theorem 2.6.4, d := dim X = tr degk k[X] =
Pn k[X]. Then Noether normalization (Theorem 2.4.1) prescribes elements∼s1 , . . . , sd ∈
dim
i=1 k t̄i such that k[X]/k[s1 , . . . , sd ] is an integral extension and k[s1 , . . . , sd ] = k[t1 , . . . , td ]
as k-algebras. Consider the embedding α : k[s1 , . . . , sd ] ,→ k[X]. Viewing k[s1 , . . . , sd ] =
k[t1 , . . . , td ] = k[Ad ], we get a comorphism ϕ = α∗ : X → Ad . Since k[X]/k[t1 , . . . , td ] is
integral, ϕ is a finite morphism and hence surjective by Proposition 12.2.8. This says that
any affine variety can be viewed as a finite-sheeted cover of Ad for some d.
Suppose U ⊆ X is a nonempty, open subset of an affine variety X ⊆ An . Then Y = X rU
is a closed set in X, so Y = Z(I) for some ideal I ⊂ k[X]. Conversely, for every ideal I ⊂ k[X]
we associate to it an open set D(I) = X r Z(I).
Definition. A principal open set in X is D(f ) := D((f )) for some f ∈ k[X] r {0},
sometimes also denoted Xf . Explicitly, D(f ) = {x ∈ X | f (x) 6= 0}.
If I = (f1 ,S. . . , fm ) for f1 , . . . , fm ∈ k[X], then by Lemma 12.1.2(h), Z(I) = m
T
i=1 Z(fi ) so
m
that D(I) = i=1 D(fi ). This shows that every open subset in X can be covered by finitely
many principal open subsets.
For f ∈ k[X] r {0}, choose a lift F ∈ k[t1 , . . . , tn ] such that f = F + J(X) ∈ k[X]. Set
A = k[t1 , . . . , tn ] and B = k[t1 , . . . , tn+1 ].
Lemma 12.2.12. Define W ⊆ An+1 by W = Z(L)where L is the ideal (J(X), tn+1 F − 1) ⊂
B. Let ϕ : D(f ) → W be the map given by x 7→ x, f (x) with inverse ϕ−1 : W → D(f ).
1
Then
(a) ϕ is a Zariski-homeomorphism.
252
12.2. Morphisms of Affine Varieties Chapter 12. Algebraic Varieties
As a consequence, the principal open subsets of X can be identified with affine subvarieties
of An+1 , giving rise to the quasi-affine varieties mentioned previously.
Remark. If k is an arbitrary field, k[X]f does not only depend on the principal open set
Xf , but also on the choice of f . However, if k is algebraically closed, we have
Xf = Xg ⇐⇒ Z(f ) = Z(g)
⇐⇒ r((f )) = r((g)) by Hilbert’s Nullstellensatz
=⇒ k[X]f = k[X]g by Proposition 1.3.5.
253
12.3. Sheaves of Functions Chapter 12. Algebraic Varieties
h : U −→ k
fx (x)
x 7−→
gx (x)
(a) OX (X) = k[X], that is, the coordinate ring of X consists of regular k-valued functions
X → k.
254
12.3. Sheaves of Functions Chapter 12. Algebraic Varieties
Proof. (a) By the Nullstellensatz, the maximal ideals of k[X] are precisely mx for x ∈ X.
Since k[X] is a domain, Lemma 2.3.11 allows us to write
\ \ \
k[X] = k[X]m = k[X]mx = OX,x = OX (X).
maximal ideals x∈X x∈X
m⊂k[X]
(b) This follows from a similar proof to (a), using the fact that the maximal ideals of
k[X]f are precisely those of the form S −1 mx , where x ∈ Xf and S = {f m | m ∈ N0 }.
(ii) There exist nonzero functions f ∈ k[X] and g ∈ k[Y ] such that Xf ∼
= Yg as affine
varieties.
(iii) There exist nonzero functions f ∈ k[X] and g ∈ k[Y ] such that k[X]f ∼
= k[Y ]g as
k-algebras.
Proof. (i) ⇐⇒ (ii) By Proposition 12.1.10(a), the principal open sets Xf and Yg are dense
in X and Y , respectively, so k(X) = k(Xf ) ∼
= k(Yg ) = k(Y ). Thus X and Y are birationally
equivalent. The converse comes from the fact that the principal open sets form bases of the
Zariski topologies on X and Y .
(ii) ⇐⇒ (iii) follows from the identification k[Xf ] = k[X]f .
A major area of interest in algebraic geometry is the classification of varieties up to
birational equivalence. For curves, there is a canonical invariant called the genus which
completely classifices curves up to birational equivalence over the algebraic closure k̄ of a
field k.
255
12.4. Finite Morphisms and the Fibre Theorem Chapter 12. Algebraic Varieties
(a) For any closed set E ⊆ X, ϕ(E) ⊆ Y is closed and the restriction ϕ|E : E → ϕ(E) is
also a finite morphism.
Proof. (a) Set A = k[Y ] and B = k[X], so that B/A is an integral extension. Given a closed
set E ⊆ X, set I = J(E) to be its vanishing ideal in B. By Lemma 2.3.1(a), (B/I)/(A/A∩I)
is also an integral extension. From Lemma 12.1.3, we have Z(I) = Z(J(E)) = E = E.
Let I 0 = A ∩ I and set E 0 = Z(I 0 ). Then ϕ(E) = E 0 so in particular ϕ(E) is closed.
Define ψ : E → E 0 by x 7→ ϕ(x). We claim ψ is finite. First, since I = J(E) is a
radical ideal by Lemma 12.1.1, I 0 = A ∩ I must be radical, too, e.g. by Lemma 1.1.9.
Thus by the Nullstellensatz, I 0 = r(I 0 ) = J(Z(I)) = J(E 0 ). Viewing J(E) ⊆ B, we have
B/I = k[X]/J(E), but by isomorphism theorems, this can be written
Similarly, A/I 0 ∼
= k[E 0 ], so we get a comorphism ψ ∗ : k[E 0 ] → k[E] which is injective by the
above comments. Therefore Lemma 12.2.9 shows that ψ is dominant. Moreover, k[E]/k[E 0 ]
is integral by the above, so ψ is finite.
(b) Since each component Ei is closed in X, (a) says that each ϕ(Ei ) is closed in Y .
Further, irreducibility of Ei implies ϕ(Ei ) is irreducible by Proposition 12.1.10(d). By (a),
ϕ|Ei : Ei → ϕ(Ei ) is a finite morphism, so Theorem 2.6.4 implies dim Ei = dim ϕ(Ei ). In
particular, Corollary 12.1.17 implies that since ϕ(Ei ) is a subvariety of W , ϕ(Ei ) = W if
and only if dim W = dim ϕ(Ei ) = dim Ei .
To finish, we show dim ϕ−1 (W ) = dim W . Corollary 12.1.17 and the previous paragraph
give us dim Ei = dim ϕ−1 (Ei ) ≤ dim W , so it follows from the definition of dimension for
ϕ−1 (W ) that dim ϕ−1 (W ) = max{dim Ei | 1 ≤ i ≤ `} ≤ dim W . Conversely, since ϕ is
surjective we get that
256
12.4. Finite Morphisms and the Fibre Theorem Chapter 12. Algebraic Varieties
We now prove the powerful Fibre Theorem for dominant morphisms. Recall from Exam-
ple 12.1.19 that if X is an affine variety, so is X × Ar , and dim X × Ar = dim X + r.
Theorem 12.4.2 (Fibre Theorem). Suppose ϕ : X → Y is a dominant morphism of affine
varieties. Set r = dim X − dim Y . Then there exists a nonempty, open subset U ⊆ Y such
that
(a) U ⊆ ϕ(X).
(b) If W ⊆ Y is closed and irreducible and W ∩ U 6= ∅, then dim(ϕ−1 (W ) ∩ ϕ−1 (U )) =
dim W + r. In particular, dim ϕ−1 (y) = r for all y ∈ U .
Proof. Suppose X ⊆ Ap and Y ⊆ Aq for p, q ≥ 1. Let B = k[X] and A = k[Y ] so that
A ,→ B. Set K = k(Y ) and L = k(X).
K L
A B
For the multiplicative subset D = A r {0}, we have that K = D−1 A and L = Frac(D−1 B).
Write B = k[t̄1 , . . . , t̄p ] = A[t̄1 , . . . , t̄p ] and note that B is a finitely generated A-module.
Then K[t̄1 , . . . , t̄p ] = D−1 A[t̄1 , . . . , t̄p ] = D−1 B so D−1 B is a finitely generated K-algebra.
By Noether’s normalization lemma (Theorem 2.4.1), there exist elements h1 , . . . , h` ∈ D−1 B
which are algebraically independent over K such that D−1 B/K[h1 , . . . , h` ] is an integral
extension. Without loss of generality, we may assume h1 , . . . , h` ∈ B. Then we have the
following diagram:
K K(h1 , . . . , h` ) L
A K[h1 , . . . , h` ] D−1 B
257
12.4. Finite Morphisms and the Fibre Theorem Chapter 12. Algebraic Varieties
The final statement follows by applying this to the irreducible subset W = {y} ⊆ U .
258
12.5. Projective Varieties Chapter 12. Algebraic Varieties
Z(f ) = {P ∈ Pn | f (P ) = 0},
259
12.5. Projective Varieties Chapter 12. Algebraic Varieties
P T
(b) For any family (Iα )α∈A of homogeneous ideals of S, α∈A Iα and α∈A Iα are also
homogeneous ideals.
Proof. (a) is routine and (b) follows from applying (a) and the definitions of sums and
intersections of ideals.
(c) Write S = ∞
L L∞
j=d Sj−d . Then f S = j=d f Sj−d and note that (f S)j = f Sj−d makes
f S into a homogeneous ideal. L
(d) If I is homogeneous, I = ∞
S∞
d=0 (I ∩Sd ) so I is generated by the set d=0 (I ∩Sd ) which
consists of homogeneous elements. Finite generation
S∞ follows from Hilbert’s basis theorem
(3.3.2). Conversely, if I = (B) where B = P d=0 Bd and Bd ⊆ Sd for each d ∈ N0 , then for
any a ∈ I, there exist rb ∈ S such that a = b∈B rb b and finitely many of the rb are nonzero.
Then ∞ X ∞
X X M
a= rb b = rb b ∈ (I ∩ Sd ).
b∈B d=0 b∈Bd d=0
L∞ L∞
Therefore we have d=0 (I ∩ Sd ) ⊆ I ⊆ d=0 (I ∩ Sd ) and hence I is homogeneous.
(e) If f = m
P
f
d=0 d ∈ r(I) then by induction on 0 ≤ d ≤ m, it follows that fd ∈ r(I)
as well. For the base case, f ∈ r(I) implies there is some r ∈ N such that f r ∈ I. Then
(f r )0 = f0r ∈ I0r so f0 ∈ r(I). Then f − f0 ∈ r(I) which implies (f − f0 )` ∈ I for some ` ∈ N,
so f1` ∈ I` . Inductively, we get fd ∈ r(I).
(f) One direction is trivial. For the other, assume the statement holds for homogeneous
elements
P` in S and suppose Pm f g ∈ p and g 6∈ p for arbitrary elements f, g ∈ S. Write
f = d=0 fd and g = d=j gd with gj 6∈ p and fd , gd ∈ Sd . Then f0 gj ∈ pj = p ∩ Sj and so
f0 ∈ p. As in (e), we can continue inductively to get fd ∈ p for all d ≥ 0, so f ∈ p. Hence p
is prime.
260
12.5. Projective Varieties Chapter 12. Algebraic Varieties
As in the affine case, the sets Z(I) form the closed sets in the Zariski topology on Pn .
Proof. (a) By Lemma 12.1.9(b) and Corollary 1.1.3, r(I) is a homogeneous ideal and clearly
r(I) ⊆ J(Z(I)). For the other direction, suppose f ∈ J(Z(I)) ∩ Sd and f 6= 0. Since
Z(I) 6= ∅, d must be at least 1, so f ∈ (t0 , . . . , tn ). In particular, f (0) = 0, i.e. f vanishes
at the origin in An+1 . Set
Then the above shows f ∈ J(Za (I)). Notice that [a0 , . . . , an ] ∈ Z(I) ⊆ Pn if and only
if (a0 , . . . , an ) ∈ Za (I) r {0} ⊆ An+1 . By the affine Nullstellensatz (Theorem 12.1.5),
J(Za (I)) = r(I) so by the above, we must have f ∈ r(I). Therefore J(Z(I)) = r(I).
(b) If r(I) ⊇ (t0 , . . . , tn ) then Z(r(I)) ⊆ Z(t0 , . . . , tn ) = ∅, implying Z(I) = ∅. Con-
versely, Z(I) = ∅ implies Za (I) ⊆ {0}, where Za (I) is as defined in part (a). But Za (I) ⊆ {0}
impies J(Za (I)) ⊇ J({0}) = (t0 , . . . , tn ) by Lemma 12.1.2(a), so we have r(I) ⊇ (t0 , . . . , tn )
by the affine Nullstellensatz (Theorem 12.1.5). This completes the proof.
Corollary 12.5.7. For an algebraically closed field k, the algebraic subsets of Pn are in
bijective correspondence with the homogeneous radical ideals I ⊂ S, I =
6 (t0 , . . . , tn ).
261
12.5. Projective Varieties Chapter 12. Algebraic Varieties
Ih = {fh | f ∈ I} ⊆ k[t0 , . . . , tn ].
For a projective algebraic set Y ⊂ Pnk , where Y = Z(J) for an ideal J ⊆ k[t0 , . . . , tn ],
we get n + 1 affine algebraic sets Yi = ϕ−1i (Y ∩ Ui ) = Z(J(i) ). These are called the deho-
mogenizations of Y . Conversely, for an affine algebraic set X ⊆ Akn , with X = Z(I), the
projective closure of X in Pnk is the Zariski closure in Pnk of ϕ0 (X), denoted X. Note that
X = Z(I(ϕ0 (X))) = Z(Ih ).
Using the affine patches Ui as charts on Pnk , we can define regular functions and morphisms
on a projective variety X as follows.
Definition. A function on X ⊆ Pnk is regular if it pulls back along Ui ,→ Pnk (i.e. restricts)
to a regular function on each affine patch Xi = Ui ∩ X.
262
12.5. Projective Varieties Chapter 12. Algebraic Varieties
F (x0 , . . . , xn )
f=
G(x0 , . . . , xn )
F1 F2
for F, G ∈ k[t0 , . . . , tn ], G 6∈ J(X), where we say f = G1
and g = G2
are equivalent if
F1 G2 − F2 G1 ∈ J(X).
Definition. The function field of X ⊆ Pnk is the set of rational functions on X, denoted
k(X).
Proposition 12.5.12. A projective variety X ⊆ Pnk is a ringed space with structure sheaf
OX : U 7→ OX (U ) defined on open sets U ⊆ X by
We can now define morphisms between projective varieties using this ringed space struc-
ture.
The definition of rational maps between affine varieties extends to projective varieties in
the following way.
Note that a quasi-projective set with the Zariski topology is not Hausdorff in general.
Indeed, if X is irreducible, then any nonempty open set is dense. Thus we need a notion to
replace the Hausdorff condition for algebraic sets.
263
12.5. Projective Varieties Chapter 12. Algebraic Varieties
Definition. For a function f ∈ k[X], define the principal open subset of f by D(f ) :=
{P ∈ X | f (P ) 6= 0}.
264
12.6. Nonsingular Varieties Chapter 12. Algebraic Varieties
where derivatives of polynomials are just the formal power rule tkj 7→ ktjk−1 . We say X is
nonsingular, or smooth, at x ∈ X if rank Jx = n−d. Otherwise, X is said to be singular
at x.
(a) A/M ∼
= AM /m.
(b) A/M 2 ∼
= AM /m2 .
(c) M/M 2 ∼
= m/m2 as A/M -vector spaces.
The following asserts that the geometric condition of smoothness is equivalent to the
algebraic condition of a ring being regular local. This is a fundamental connection in algebraic
geometry.
θ : A −→ k n
∂f ∂f
f 7−→ (x), . . . , (x) .
∂t1 ∂tn
Notice that θ is linear, and θ(ti − xi ) = (0, . . . , 1, . . . , 0) (with 1 in the ith coordinate), so θ
is surjective. In fact, this shows that θ|mx is surjective. We claim that ker θ|mx = m2x . On
one hand, for any 1 ≤ i, j, k ≤ n, we have
∂((ti − xi )(tj − xj ))
= δik (tj − xj ) + δjk (ti − xi )
∂tk
265
12.6. Nonsingular Varieties Chapter 12. Algebraic Varieties
2
P at x. Hence mx ⊆ ker θ|mx . On the other
which is zero hand, suppose f ∈ ker θ|mx r m2x .
Then f = i gi (ti − xi ) for some gi ∈ A. Since f 6∈ m2x , we must have gj 6∈ mx for some j.
Thus
∂f ∂ X ∂gi
= (gj (tj − xj )) + (ti − xi )
∂tj ∂tj i6=j
∂t j
Thus rank Jx = dimk θ(J(X)) = dimk (J(X) + mx /m2x ). From Section 12.3, we know that
OX,x = k[X]mx and this is a local ring because mx is a maximal ideal. Let m be the unique
maximal ideal of OX,x and let m be the corresponding ideal in k[X]. By Lemma 12.6.1,
m/m2 ∼ = mx /m2x as k-vector spaces, so m/m2 ∼ = mx /m2x + J(X). As a result, we get
2
dimk m/m + rank Jx = n. Since k is algebraically closed, dim X = dim k[X] by Propo-
sition 12.1.16, so dim OX,x = dim k[X]mx = ht(mx ) by Lemma 2.3.6(a). Since X is affine,
k[X] is both a finitely generated k-algebra and an integral domain (Lemma 12.1.14), and
k[X]/mx ∼ = k as k-algebras, so Lemma 2.6.5 and Theorem 2.6.7 imply that ht(mx ) = d.
Putting everything together, we get
One can define analogs of the sheaf of regular functions on an arbitrary variety X (includ-
ing affine, projective, quasi-affine and quasi-projective). The notion of smoothness generalize
to these cases:
266
12.6. Nonsingular Varieties Chapter 12. Algebraic Varieties
Proof. Let K = k(X) be the fractional field of X. Since k[X]/k is a finitely gener-
ated algebra, K/k is a finitely generated field extension. Moreover, assuming k is alge-
braically closed and therefore perfect, Proposition 1.4.9 says that K/k can be written as a
tower K ⊃ k(t1 , . . . , td ) ⊃ k such that K/k(t1 , . . . , td ) is finite and separable (using that
tr degk K = dim X = d from Corollary 12.1.17) and k(t1 , . . . , td ) is purely transcendental.
By the primitive element theorem, we may write K = k(t1 , . . . , td )(g) for some g ∈ K. In
particular, there exists an irreducible polynomial f ∈ k(t1 , . . . , td )[t] such that f (g) = 0 and
K∼ = k(t1 , . . . , td )[t]/(f ). Clearing denominators (e.g. using Gauss’s lemma), we get an irre-
ducible polynomial f˜ = λf ∈ k[t1 , . . . , td ][t] for some λ ∈ k[t1 , . . . , td ]. Let Y = Z(f˜) ⊆ Ad+1 .
Then k[Y ] = k[t1 , . . . , td ][t]/(f˜) so applying Lemma 12.6.3 with A = k[t1 , . . . , td ] and its field
of fractions k(t1 , . . . , td ), we get
k(X) = K ∼
= k(t1 , . . . , td )[t]/(f ) ∼
= Frac(k[t1 , . . . , td ][t]/(f˜)) = Frac(k[Y ]) = k(Y ).
Proof. We first prove Sing X is closed. In the proof of Theorem 12.6.2, we saw that for any
x ∈ X, rank Jx = n − dimk m/m2 where m is the maximal ideal of the local ring OX,x . By
Example 6.2.7, dimk m/m2 ≥ dim OX,x = d and so rank Jx ≤ n − d (justifying the above
description of Sing X). By linear algebra,
267
12.6. Nonsingular Varieties Chapter 12. Algebraic Varieties
∂fi
Let J(X) = (f1 , . . . , fr ) and consider the operator J = . Then any determinant of a
∂tj
square minor of J is an element of A = k[t1 , . . . , td ]. Thus the set
D = {det C | C is an (n − d) × (n − d) minor of J }
Xg ∼
= Yh =⇒ k[Xg ] ∼
= k[Yh ] =⇒ OX,x ∼
= OXg ,x ∼
= OYh ,y ∼
= OY,y
Corollary 12.6.6. Let X be any variety. Then Sing X is a proper closed subset of X.
268
12.7. Abstract Curves Chapter 12. Algebraic Varieties
269
12.7. Abstract Curves Chapter 12. Algebraic Varieties
Corollary 12.7.3. Any DVR of the field extension K/k is the local ring of a variety Y at
some point P ∈ Y .
The next result shows that abstract nonsingular curves properly generalize nonsingular
quasi-projective curves as defined in Section 12.6.
Proof. Let K = k(X) and recall that Corollary 12.1.17(a) implies tr degk K = 1. Define a
map ϕ : X → CK that sends P 7→ OP . If we set U = ϕ(X), changing targets gives us a
270
12.7. Abstract Curves Chapter 12. Algebraic Varieties
Thus the abstract theory of curves presented in this section captures the more concrete
notions of affine, projective and quasi-affine/projective varieties detailed in earlier sections.
271
12.8. Products of Varieties Chapter 12. Algebraic Varieties
We thus make X × Y into a ringed space with OX×Y (U × V ) defined for all open sets
U ⊆ X, V ⊆ Y by stipulating that anything of the form
X
∗
f= (πX gi )(πY∗ hi ), for gi ∈ OX (U ) and hi ∈ OY (V ),
∗
is regular on U × V . If g ∈ OX (U ), we must have πX g ∈ OX×Y (U × V ) and likewise, if
∗
h ∈ OY (V ), then πY h ∈ OX×Y (U × V ). Thus for such an f as above, D(f ) is an open subset
of X × Y that would not be open in the usual product topology.
Example 12.8.1. Under the above description of products of affine varieties, An × Am ∼
=
n+m 2
A for any n, m ∈ N. Note that even for n = m = 1, the Zariski topology on A is not
equivalent to the product topology on A1 × A1 .
Lemma 12.8.2. If X and Y are affine varieties, then
(a) X × Y is an affine variety.
σn,m : Pn × Pm −→ P(n+1)(m+1)−1
([x0 , . . . , xn ], [y0 , . . . , ym ]) 7−→ [xi yj ]i,j
such that the image Σn,m := σn,m (Pn × Pm ) has the structure of an algebraic subset that
coincides with the Zariski topology of the product Pn × Pm .
Proof. (Sketch) Viewing P(n+1)(m+1)−1 as a space of (n + 1) × (m + 1) matrices, we have that
Then clearly Σn,m = Z((zij zk` − zkj zi` )i,j,k,` ), so Σn,m is an algebraic set. The fact that σn,m
is a bijection is obvious. One can now verify that the induced topology corresponds to the
topology on Pn × Pm .
272
12.8. Products of Varieties Chapter 12. Algebraic Varieties
Definition. Let V be a vector space over k. The set of lines, i.e. 1-dimensional subspaces,
of V is called the projective space over V , denoted P(V ).
Example 12.8.4. If V = k n is finite dimensional, then P(V ) can be identified with Pnk .
gives X × Y the structure of a ringed space which coincides with the previous description of
the product of two varieties.
G(x0 , . . . , xn , y0 , . . . , ym ),
Example 12.8.7. Consider the Segre embedding P1k × P1k ,→ P3k and set Q = Σ1,1 =
Z(z00 z11 − z01 z10 ). The polynomial z00 z11 − z01 z10 is called a quadric and the embedded
image Q is called a quadric surface. For each α, β ∈ P1k , one gets lines on the quadric surface
realized by {α} × P1k ,→ Q and P1k × {β} ,→ Q. Note that lines of these forms cover Q, for
which reason Q is called a ruled surface.
273
12.9. Blowing Up Chapter 12. Algebraic Varieties
12.9 Blowing Up
We now have a working notion of products of varieties, so consider the space An × Pn−1 .
Coordinates in this space are (P, [`]), where P ∈ An is a point and [`] ∈ Pn−1 is the class of
some line through the origin ` in An . Consider the set B ⊆ An × Pn−1 defined by
On the other hand, if P = 0, the set π −1 (0) = {(0, [`]) ∈ An × Pn−1 } is isomorphic to Pn−1 .
In the dimension 2 case, B is covered by the following affine patches:
Definition. The set B is called the blowup of An at the point 0, denoted B = Bl0 An . The
set E0 An := π −1 (0) ∼
= Pn−1 is called the exceptional divisor of the blowup.
Definition. Let X ⊆ An be an affine variety and π : An × Pn−1 → An the canonical
projection. The pullback π −1 (X) is called the total transform of X, while the proper (or
strict) transform of X is defined as
Bl0 X := π −1 (X r {0}).
As the notation suggests, this set is also called the blowup of X at 0. The set
E0 X := Bl0 X ∩ E0 An
274
12.9. Blowing Up Chapter 12. Algebraic Varieties
Remark. More generally, for any subvariety Z ⊆ X, one can define the blowup of X along
Z, a variety BlZ X that is birationally equivalent to X, such that Z is a codimension 1
subvariety of BlZ X.
Note that this variety has a singularity at the point (0, 0). Using the blowup of A2 defined
above, Bl0 A2 , we can blowup X to ‘remove the singularity’ at 0. Let U1 be the first affine
patch and ϕ : U1 → A2 the standard isomorphism. We make the substitution y = xt, so
that ϕ(π −1 (X)) = Z(x2 (t2 − x − 1)). The x2 factor of this polynomial corresponds to the
exceptional divisor E0 X under this blowup, so the proper transform of X at 0 looks like
Blowing up allows us to replace singular curves (or more generally, varieties) with non-
singular curves by a sequence of blowups, such that in each step the birational equivalence
class of the curve is preserved. The problem of finding such a nonsingular blowup is known
as resolution of singularities. Much progress has been made on this problem (e.g. Hironaka’s
theorem says that nonsingular blowups exist for any finite dimensional variety over a field
of characteristic zero), but there is still much to be done (e.g. in finite characteristic cases).
275
12.10. Prevarieties and Varieties Chapter 12. Algebraic Varieties
276
12.10. Prevarieties and Varieties Chapter 12. Algebraic Varieties
Example 12.10.4. Part (b) of Proposition 12.10.3 does not hold in the category of preva-
rieties. For instance, take the affine line with a double point, i.e. A e 1 := A1 ∪ {00 } where
e 1 r {0} ∼
A =A e 1 r {00 } ∼
= A1 . Then A
e 1 is a prevariety but not a variety, and there are maps
ϕ, ψ : A1 → A e 1 with ϕ(0) = 0 and ψ(0) = 00 . The maps agree on a dense subset, namely
e 1 r {0, 00 }, but they are not the same map on the whole space.
A
Proposition 12.10.5. Let X be a variety. If U and V are nonempty, open affine subvarieties
of X then U ∩ V is an open affine variety.
Remark. If U and V are nonempty, open affine subvarieties of a variety X then the ring of
k-valued functions on U ∩ V , OX (U ∩ V ), is generated by the restrictions of all functions in
OX (U ) and OX (V ) to U ∩ V .
277
12.11. Complete Varieties Chapter 12. Algebraic Varieties
Proof. (1) Let X 0 ⊆ X be a closed subvariety and Y any variety, and consider π 0 : X 0 × Y →
Y . Suppose Z ⊆ X 0 × Y is a closed subset. In general X 0 × Y is closed in X × Y so the
diagram
i
Z ⊆ X0 × Y X ×Y
π0 π
278
12.11. Complete Varieties Chapter 12. Algebraic Varieties
2 If V is a finite dimensional vector space over k, then the projective space P(V ) consist-
ing of one-dimensional subspaces of V is complete. By (1) of Proposition 12.11.1 below,
every closed subvariety of P(V ) is complete, so all projective varieties are complete.
We now prove an important property of projective varieties.
Theorem 12.11.2. Every projective variety is complete.
Proof. We proved that every closed subvariety of a complete variety is complete, so it suffices
to prove Pn is complete for all n ≥ 1. In other words we will show that if π : Pn × Y → Y
is the projection map and C ⊂ Pn × Y is closed then π(C) ⊆ Y is closed. Set A = k[Y ] and
B = A[T0 , T1 , . . . , Tn ]. Then B is a ring of k-valued functions on k n+1 × Y . For every proper
homogeneous ideal I ⊂ B, define
V ∗ (I) = {(x∗ , y) | f (x, y) = 0 for all f ∈ I} ⊆ Pn × Y.
Then the V ∗ (I) are the closed subsets of Pn × Y so it suffices to prove π(V ∗ (I)) is closed for
all proper homogeneous ideals I ⊂ B. We may assume V ∗ (I) is irreducible, i.e. I is prime.
We may also assume π|V ∗ (I) is dominant (changing the target to π(V ∗ (I)) if necessary). Then
we must show for every y ∈ Y , there exists x∗ ∈ Pn so that (x∗ , y) ∈ V ∗ (I), since then we
will have π(V ∗ (I)) = π(V ∗ (I)).
Take M ⊂ A to be the maximal ideal that vanishes at y. Then J = M B + I is a
homogeneous ideal so V ∗ (J) is defined, and if we show V ∗ (J) is nonempty, we’ll be done.
Assume to the contrary that V ∗ (J) = ∅. Then there is a k > 0 such that Tik ∈ J for each Ti .
Equivalently, there is an m > 0 so that Bm , the set of all degree m homogeneous polynomials
in B, is contained in J. Set N = Bm /(Bm ∩ I). This is a finitely generated A-module in the
obvious way. Moreover, notice that M N = N . Then by Nakayama’s Lemma (1.2.1), this
implies N = 0. But then Bm = Bm ∩ I so it follows that V ∗ (I) = ∅, which is impossible for
a proper ideal I ⊂ B. Hence V ∗ (J) 6= ∅ so the theorem is proved.
There are examples of complete, non-projective varieties but they are difficult to come
up with. However, we have:
Theorem 12.11.3. (Chow’s Lemma) For any complete irreducible variety X, there exists a
projective variety X and a surjective birational morphism f : X → X.
Recall from general topology that a topological space X is compact if and only if for
all spaces Y , X × Y → Y is a closed map. In this way complete varieties are an algebraic
analogue of compact spaces. This analogy runs quite deep:
Proposition 12.11.4. Let X be a connected complete variety. Then the sheaf OX (X) =
k[X] equals k. That is, every regular k-valued function on X is constant.
Proof. Take f ∈ OX (X). Then f is a map f : X → k = A1 . Extend this to a map
g : X → A1 ,→ P1 ,
so g is not surjective onto P1 . By completeness of X, g(X) is closed in P1 , but the only
proper closed subsets of P1 are point-sets. Therefore g(X) = {x0 }, or in other words, g is
constant. This implies f is constant.
279
12.11. Complete Varieties Chapter 12. Algebraic Varieties
This fact is analagous to the theorem in complex analysis that every holomorphic function
on a connected compact domain is constant.
Proof. Let X be an affine variety of dimension at least 1. View X as a proper subset of affine
n-space An , which has coordinate algebra k[T1 , . . . , Tn ]. Then some coordinate function Ti
does not vanish on X, so Ti ∈ OX (X) is a nonconstant regular function on X.
280
Chapter 13
Sheaf Theory
In this chapter, we describe the theory of sheaves on topological spaces. The basic intuition
is that for a topological space X, one wishes to study a given property defined on open sets
U ⊆ X that behaves well under union and intersection of open sets. Sheaves will allow us
in the following chapter to construct a generalization of a variety: a scheme.
281
13.1. Sheaves and Sections Chapter 13. Sheaf Theory
Γp (U ) −→ Γp (V )
σ 7−→ σ|V .
That is, Γp (−) is a contravariant functor on the category TopX of open sets on X.
S
(b) Let {Ui } be a collection of open sets in X, with U = Ui , and suppose there exist
local sections σi ∈ Γp (Ui ) such that σi |Ui ∩Uj = σj |Ui ∩Uj for all i, j. Then there exists a
unique section σ ∈ Γp (U ) for which σ|Ui = σi holds for all i.
This generalizes in the following way.
Definition. For a topological space X, a presheaf on X is a contravariant functor F :
TopX → C, where C is any category. For each inclusion of open sets V ,→ U , the induced
morphism FU V : F(U ) → F(V ) is called restriction, written FU V (σ) = σ|V .
Definition. Let C be a set category. A presheaf F : TopX → C is called a sheaf on X
provided it satisfies the following ‘descent conditions’:
(1) For any open covering U = Ui , if there exist sections σ, σ 0 ∈ F(U ) such that σ|Ui =
S
σ 0 |Ui for all i, then σ = σ 0 .
S
(2) For any open covering U = Ui admitting sections σi ∈ F(Ui ) such that σi |Ui ∩Uj =
σj |Ui ∩Uj for all i, j, then there exists a unique section σ ∈ F(U ) such that σ|Ui = σi
for all i.
Notice that the uniqueness in (2) is guaranteed by (1) so it is not strictly necessary to
include it in the definition.
Definition. A morphism of sheaves is simply a natural transformation of functors F →
G. Likewise, an isomorphism of sheaves is a natural isomorphism of the underlying functors.
Example 13.1.2. For a (pre)sheaf F on X and an open set U ⊆ X, the functor F|U : V 7→
F(V ) is a (pre)sheaf. The operation F 7→ F|U is called restriction of a sheaf.
282
13.1. Sheaves and Sections Chapter 13. Sheaf Theory
Remark. The descent conditions in the definition of a sheaf are stated for set categories,
S category C by saying that for every open set
but they can be reinterpreted in an arbitrary
U ∈ TopX and every open covering U = Ui , the object F(U ) is a limit of the following
diagram (also called an equalizer):
Y Y
F(U ) F(Ui ) F(Ui ∩ Uj )
i i,j
Example 13.1.3. The classic example of a sheaf on a topological space X assigns to each
open U ⊆ X the set F(U ) of all continuous (or differentiable, smooth, holomorphic, etc.)
functions on U , with restriction
S maps given by restriction of functions. To see that this is
a sheaf, suppose U = Ui is an open cover in X and f, g : U → Y are continuous maps
such that f |Ui = g|Ui . This means f (x) = g(x) for all x ∈ Ui . Since the Ui cover U , this
means f (x) = g(x) for all x ∈ U and hence f = g. If instead we have continuous functions
fi : Ui → Y for each i with fi |Ui ∩Uj = fj |Ui ∩Uj , then fi (x) = fj (x) for all x ∈ Ui ∩ Uj . For
x ∈ U , define f : U → Y by f (x) = fi (x) if x ∈ Ui . Then the previous statement implies f
is well-defined and continuous on U . Hence f ∈ F(U ) and f |Ui = fi for all i.
Example 13.1.4. Let F(U ) be the set of bounded functions U → R. Then F : U 7→ F(U )
is a presheaf on X, with the usual restriction maps, but F is not a sheaf. For example,
1 4
when X = R, consider the open cover of X given by Un = n − 3 , n + 3 . Then fn (x) = x
defines a sequence of bounded functions on each Un which agree on intersections, but the
only possible lift of {fn } to X is f (x) = x, an unbounded function. Hence axiom (2) fails.
Example 13.1.5. Let S be a set and define a functor F : TopX → Set by F(U ) = S for all
U ⊆ X, with the restriction maps given by σ|V = σ ∈ S for any inclusion V ,→ U . Then F
is a presheaf, called the constant presheaf on S, but if X has two disjoint open sets, then F
is not a sheaf.
Example 13.1.6. The right notion of a constant sheaf is constructed as follows. Let S be a
set considered with the discrete topology. For all open sets U ⊆ X, let FS (U ) = C(U, S) be
the set of all continuous (i.e. locally constant) functions U → S. Then FS defines a sheaf
on X. Notice that if U is a connected open set, then FS (U ) ∼= S, hence the name constant
sheaf.
Definition. For a sheaf F on X and a point x ∈ X, the stalk of F at x is the direct limit
Fx := lim F(U )
−→
Example 13.1.8. For the constant sheaf FS , the stalks are all the same: FS,x = S.
283
13.1. Sheaves and Sections Chapter 13. Sheaf Theory
Example 13.1.10. Let X = C be the space of complex numbers and for each open U ⊂ C,
let O(U ) denote the set of all holomorphic functions U → C. Then O is a sheaf of rings
with stalks given by Oz = C[t](t−z) , the ring of Taylor series at z with positive radius of
convergence.
Example 13.1.11. For an affine algebraic variety V ⊆ An , the structure presheaf OV (see
Section 12.3) is a sheaf. The proof is similar to the argument in Example 13.1.3. In fact, we
already proved axiom (2) in Lemma 12.3.1. The stalk of the structure sheaf OV at a point
x ∈ V is
OV,x = K[V ]mx ,
the localization of K[V ] at the maximal ideal mx = {f ∈ K[V ] | f (x) = 0}. Equivalently,
OV,x consists of all rational functions f ∈ K(V ) which are defined at x, meaning on some
neighborhood U of x, f = hgxx for some gx , hx ∈ OV (U ) with hx (x) 6= 0.
(1) The functor Γp (−) : TopX → Set, U 7→ Γp (U ), is a sheaf on X with restriction maps
Γp (U ) → Γp (V ) for V ⊆ U given by restriction of sections s 7→ s|V .
(2) Sets of the form s(U ), where U ⊆ X is open and s ∈ Γp (U ), form a basis for the
topology on E.
284
13.1. Sheaves and Sections Chapter 13. Sheaf Theory
Proof. (1) follows from Lemma 13.1.1 and (2) is routine. Here’s a proof of (3):
For each neighborhood U ⊆ X of x, define a map
νU : Γp (U ) −→ Ex
s 7−→ s(x).
Ex
νV
Γp (V )
where U runs over all open sets in X. The étale cover pF : EF → X is naturally defined
as (x, f |Fx ) 7→ x.
Example 13.1.13. For the sheaf of holomorphic functions O on C, the open sets VU,f are
given by
Proof. We have seen, courtesy of Proposition 13.1.12, that Γp is a sheaf on X. On the other
hand, it follows from the definition of the topology on EF that the map pF : EF → X is
continuous. Therefore it remains to show that the assignments p 7→ Γp and F 7→ pF define
an equivalence of categories. That is, we must show that for a sheaf F on X, there is an
isomorphism ΓpF ∼ = F, and for a map p : E → X, there is a homeomorphism EΓp ∼ = E
making the diagram
285
13.1. Sheaves and Sections Chapter 13. Sheaf Theory
∼
EΓp E
commute. First let F be a sheaf, let U ⊆ X be open, take σ ∈ F(U ) and define s ∈ ΓpF (U )
by s(x) = (x, σ|x ). Consider s−1 (VU,f ) ⊆ U for some f ∈ F(U ). Then
For such an x we have W ⊆ s−1 (VU,f ), so the morphism F → ΓpF sending σ 7→ s is well-
defined and continuous. SupposeSs0 ∈ ΓpF (U ) such that s(x) = s0 (x) for all x ∈ U 0 . Then
there exists an open cover U = Ui with s|Ui = s0 |Ui for all Ui , so by the sheaf axioms,
s = s0 . Therefore F → ΓpF is one-to-one.
On the other hand, suppose s : U → EΓp is a section of pF . Then
On the open set s−1 (VU 0 ,s0 ) ⊆ U , s comes from s0 uniquely. Therefore if U 0 is a neighborhood
of x, then s−1 (VUS 0
0 ,s0 ), where s (x) = s(x), is a neighborhood of x. This means there is an
open cover U = Ui with σi ∈ F(Ui ) such that σi (x) = s(x) for all x ∈ U . Moreover,
since F → ΓpF is one-to-one, we have σi |Ui ∩Uj = σj |Ui ∩Uj for all i, j. Therefore by the sheaf
axioms, there is a unique section σ ∈ F(U ) with σ|Ui = σi for all i, which then satisfies
σ(x) = s(x) for all x ∈ U . Thus F → ΓpF is onto, hence an isomorphism.
Now let p : E → X be a continuous map. Define E → EΓp as follows. For y ∈ E, set
x = p(y) ∈ X. Then for all open sets U ⊆ X containing X, y ∈ p−1 (U ), so y in fact lies
in the stalk Γp,x = lim Γp (U ). Thus the assignment E → EΓp , y 7→ y is well-defined. It is
−→
clearly one-to-one and onto, and continuity follows easily from the definition of the topology
on EΓp .
Definition. A sheaf F on X is locally constant if for all x ∈ X, there is a neighborhood
U ⊆ X of x such that F|U is isomorphic to a constant sheaf on U .
Example 13.1.15. If p : E → X is a topological covering, then Γp is a locally constant
sheaf on X. Indeed, for each x ∈ X, choose a connected, locally trivial neighborhood U
of X such that p−1 (U ) ∼= U × p−1 (x). Then for a local section σ : U → E, p maps σ(U )
homeomorphically onto U , so σ(U ) must be a connected component of p−1 (U ). It follows
that Γp (U ) ∼
= p−1 (x) and therefore Γp |U is isomorphic to the constant sheaf Fp−1 (x) .
Conversely, for every locally constant sheaf F, the étale cover EF → X is a topological
cover. To see this, note that for each locally constant neighborhood U ⊆ X, there is a fixed
discrete set S such that Fx = S for all x ∈ U . Thus p−1 (U ) ∼= U × S, so EF is a cover of X.
Thus we have proven:
Corollary 13.1.16. There is an equivalence of categories between covering spaces of X and
locally constant sheaves on X.
286
13.1. Sheaves and Sections Chapter 13. Sheaf Theory
Theorem 13.1.17 (Sheafification). For every presheaf (of sets) F on X, there exists a
presheaf F sh together with a morphism of presheaves θ : F → F sh such that for every x ∈ X,
θx : Fx → Fxsh is a bijection. Moreover, for any morphism ϕ : F → G where G is a sheaf,
there is a unique morphism of sheaves ψ : F sh → G such that the diagram
θ
F F sh
ϕ ψ
commutes.
Proof. This will from Proposition 13.2.8(2) and the fact that Fx → Fxsh is a bijection for all
x ∈ X.
π
Proof of 13.1.17. Let E = EF →− X be the étale space of F and define F = Γπ (−), the sheaf
of sections of π. Then the morphism θ : F → F sh is defined by
θU : F (U ) −→ F sh (U )
s 7−→ (s̃ : U → E, x 7→ s|x ).
θU
F (U ) F sh (U )
F (V ) F sh (V )
θV
287
13.1. Sheaves and Sections Chapter 13. Sheaf Theory
θU
F (U ) F sh (U )
ϕU ψU
G(U )
commutes, and once again this is compatible with restriction of sections along V ⊆ U . The
fact that such a ψ is unique follows from the bijection Fx → Fxsh and Proposition 13.2.8.
sh
Presh(C) Sh(C)
forget
which means sheafification is right exact and the forgetful functor is left exact. In a more
general setting, such as when we consider the categories of presheaves and sheaves of abelian
groups, rings, algebras, etc., we can take this as our definition of the sheafification functor:
it is the left adjoint to the (left exact) forgetful functor ShX ,→ PreshX .
Alternatively, we can equip an associated sheaf F sh with the structure of a sheaf of
π
abelian groups (or rings, algebras, etc.) as follows. Let E → − X be the étale cover for the
presheaf F and consider the fibre product
F sh (U ) × F sh (U ) = Γπ (U ) × Γπ (U ) ∼
= Γπ×X π (U ) −→ Γπ (U ) = F sh (U )
288
13.2. The Category of Sheaves Chapter 13. Sheaf Theory
The presheaf kernel ker ϕ is defined by (ker ϕ)(U ) = ker(F(U ) → G(U )).
The presheaf cokernel coker ϕ is defined by (coker ϕ)(U ) = coker(F(U ) → G(U )).
Lemma 13.2.1. If ϕ : F → G is a morphism of sheaves, then ker ϕ is a sheaf.
S
Proof. Suppose U = i Ui is an open cover in X and s, t ∈ (ker ϕ)(U ) such that s|Ui = t|Ui
for all Ui . Then since (ker ϕ)(U ) ⊆ F(U ), the first sheaf axiom for F implies directly that
s = t. Now suppose there are sections si ∈ (ker ϕ)(Ui ) such that si |Ui ∩Uj = sj |Ui ∩Uj for all
overlapping Ui , Uj . Since (ker ϕ)(Ui ) ⊆ F(Ui ), the second sheaf axiom for F implies there
exists a section s ∈ F(U ) such that s|Ui = si for all i. We must show s ∈ (ker ϕ)(U ) ⊆ F(U ).
Note that ϕU (s)|Ui = ϕUi (s|Ui ) = ϕUi (si ) = 0 for each Ui , so by the first sheaf axiom for G,
we must have ϕU (s) = 0.
Theorem 13.2.2. If C is an abelian category, then the category PreshX (C) of presheaves
TopX → C is an abelian category.
Example 13.2.3. For a morphism ϕ of sheaves, coker ϕ is not a sheaf in general. For exam-
ple, take X = C and let F = O be the sheaf of holomorphic functions from Example 13.1.10.
If O× denotes the sheaf of nonvanishing holomorphic functions on C, then there is a mor-
phism of sheaves ϕ : O → O× given locally by ϕU : O(U ) → O(U )× , f 7→ e2πif . In fact,
one can see that ϕ is an isomorphism on stalks (using the complex logarithm). However, on
U = C× , the map O(C× ) → O(C× )× is not surjective, since f (z) = z is not in the image.
Thus on the level of stalks we have exact sequences
ψx ϕx
Zx −→ Ox −→ Ox×
where Z denotes the constant sheaf, but the cokernel of the map ψ : Z → O is not O× , but
rather (O/Z)sh .
Definition. The sheaf cokernel of a morphism of sheaves ϕ : F → G is the sheafification
of the presheaf cokernel:
289
13.2. The Category of Sheaves Chapter 13. Sheaf Theory
Lemma 13.2.5. Let F be a sheaf on X and U ⊆ X an open set with sections s, t ∈ F(U ).
Then s =Qt if and only if for all x ∈ U , s|x = t|x in Fx . In other words, the morphism
F(U ) → x∈U Fx is injective.
Therefore by Lemma 13.2.5, ϕU (s) = ψU (s). Since this holds for all U , we get ϕ = ψ.
Definition. Let F be a sheaf of abelian groups on X and s ∈ Γ(X, F) a global section. Then
the support of s is the set
supp(s) = {x ∈ X | s|x 6= 0 in Fx }.
(1) The morphisms ϕx : Fx → Gx are injective for all x ∈ X if and only if the morphisms
ϕU : F(U ) → G(U ) are injective for all open sets U ⊆ X.
(2) The morphisms ϕx : Fx → Gx are bijective for all x ∈ X if and only if the morphisms
ϕU : F(U ) → G(U ) are bijective for all open sets U ⊆ X.
Proof. (1) Assume ϕx are injective for all x ∈ X and suppose s, t ∈ F(U ) are sections such
that ϕU (s) = ϕU (t) in G(U ). Then for all x ∈ X,
since ϕ is a morphism, so injectivity implies s|x = t|x . Finally, Lemma 13.2.5 shows that
s = t in F(U ).
Conversely, assume that each ϕU is injective. For a fixed x ∈ X, take sx , tx ∈ Fx such
that ϕx (sx ) = ϕx (tx ) in Gx . Then there is a neighborhood U of x such that sx = s|x and
290
13.2. The Category of Sheaves Chapter 13. Sheaf Theory
tx = t|x for some s, t ∈ F(U ). Since Gx = lim G(U ) over all neighborhoods U of x, the
−→
condition that ϕx (sx ) = ϕx (tx ) in Gx means there is some smaller neighborhood V ⊆ U
containing x on which ϕU (s)|V = ϕU (t)|V . Since ϕ is a morphism, this is equivalent to
ϕV (s|V ) = ϕV (t|V ), but by injectivity of ϕV , we get s|V = t|V . Passing to the direct limit,
we have sx = tx in Fx . Hence ϕx is injective.
(2) The ( ⇒= ) implication is trivial. For ( =⇒ ), suppose each ϕx is bijective. Fix
U ⊆ X and t ∈ G(U ); we must find s ∈ F(U ) so that ϕU (s) = t. For any x ∈ U ,
consider the image tx of t in the stalk Gx . By hypothesis, there is some sx ∈ Fx such that
ϕx (sx ) = tx . Then there is a neighborhood Ux ⊆ U of x on which sx = sUx |x for some
sUx ∈ F(Ux ). By construction, ϕUx (sUx ) = t in Gx , so there is a neighborhood Vx ⊆ Ux of x
with ϕVx (sUx |Vx ) = ϕUxS(sUx )|Vx = tVx . Set sVx = sUx |Vx . Letting x range over the points of
U , we get a cover U = x∈U Vx and moreover, if x, y ∈ U , then
ϕVx ∩Vy (sVx |Vx ∩Vy ) = ϕVx (sVx )|Vx ∩Vy since ϕ is a morphism
= t|Vx ∩Vy
= ϕVy (sVy )|Vx ∩Vy
= ϕVx ∩Vy (sVy |Vx ∩Vy ) for the same reason.
We are assuming each ϕx is bijective, so in particular (1) implies each ϕVx ∩Vy is injective, and
thus sVx |Vx ∩Vy = sVy |Vx ∩Vy . Now the second sheaf axiom guarantees that there is a section
s ∈ F(U ) such that s|Vx = sVx for all x ∈ U . Finally, observe that
Remark. Notice that the proof of surjectivity in (2) crucially relies on the fact that the ϕx
and ϕU are already injective. It is not true in general that the ϕx are all surjective if and
only if the ϕU are all surjective.
Proposition 13.2.9. For any x ∈ X, the stalk functor PreshX → Ab, F 7→ Fx , is exact.
ϕ ψ
Proof. Suppose F 0 − →F − → F 00 is an exact sequence of presheaves on X. We must show the
ϕx ψx
induced sequence of stalks Fx0 −→ Fx −→ Fx00 is exact. First, since ψ ◦ ϕ = 0 globally, on
each stalk we get ψx ◦ ϕx = 0 so im ϕx ⊆ ker ψx . On the other hand if sx ∈ ker ψx , then for
some neighborhood U of x, there is a section s ∈ F(U ) such that s|x = sx and ψU (s)|x =
ψx (sx ) = 0. Thus on some smaller neighborhood V ⊆ U of x, ψV (s|V ) = ψU (s)|V = 0, so
s|V ∈ ker ψV . By exactness, this means s|V = ϕV (s0 ) for some s0 ∈ F 0 (V ). Then we have
sx = ϕV (s0 )x = ϕx (s0x ) for s0x = s0 |x in Fx . Hence ker ψx ⊆ im ϕx so the sequence is exact.
This gives us a good definition for what it means for a sequence of sheaves to be exact.
291
13.2. The Category of Sheaves Chapter 13. Sheaf Theory
ϕ ψ
Theorem 13.2.10. Let 0 → F 0 − →F − → F 00 → 0 be a short exact sequence of sheaves on
X. Then for every open set U ⊆ X, the sequence
0 → F 0 (U ) → F(U ) → F 00 (U )
is exact. That is, the forgetful functor ShX → PreshX is left exact.
ϕx ψx
Proof. By definition, 0 → Fx0 −→ Fx −→ Fx00 → 0 is exact for all x ∈ X, so Propo-
sition 13.2.8(1) implies 0 → F 0 (U ) → F(U ) is exact for all open sets U ⊆ X. Set
K(U ) = ker(F(U ) → F 00 (U )). Then by Lemma 13.2.1, K : U 7→ K(U ) is a sheaf and it is
easy to see that the stalks are Kx = ker(Fx → Fx00 ). Now we have (ψ ◦ ϕ)x = ψx ◦ ϕx = 0 for
all x ∈ X by exactness of the sequence of stalks, so (ψ ◦ϕ)U = ψU ◦ϕU = 0 for all U , and thus
im ϕU ⊆ K(U ). Finally, ϕ can be written as a morphism F → K, but on the level of stalks
∼
this is a bijection ϕx : Fx −
→ Kx = im ϕx , so by Proposition 13.2.8(2), ϕU : F(U ) → K(U )
is also bijective. Hence F ∼ = im ϕ = K so the sequence 0 → F 0 (U ) → F(U ) → F 00 (U ) is
exact.
In general, if F → F 00 is surjective, it need not be true that F(U ) → F 00 (U ) is surjective.
By definition, surjectivity of sheaves means surjectivity of stalks, so an element s00x ∈ Fx00 is
the image of some sx ∈ Fx but this sx need only be defined on some neighborhood of x, not
necessarily on arbitrary U . Example 13.2.3 gives a counterexample.
Definition. A sheaf F on X is called flasque (or flabby) if for every open inclusion V ⊆ U ,
the restriction map F(U ) → F(V ) is surjective.
ϕ ψ
Theorem 13.2.11. If 0 → F 0 − → F 00 → 0 is a short exact sequence of sheaves with F 0
→F −
flasque, then the sequence
0 → F 0 (U ) → F(U ) → F 00 (U ) → 0
of open sets which admit an extension of t. Then C is a partially ordered set via (U1 , s1 )
(U2 , s2S) if and only if U1 ⊆ U2 and s2 |U1 = s1 . For S
a linearly ordered subset (Uα , sα ) in C, the
∗ ∗
pair ( Uα , s ) is an upper bound, where s ∈ F ( Uα ) is defined using the sheaf conditions
on F. Thus by Zorn’s Lemma, C contains a maximal element, say (U, s). Assume x ∈ X rU .
Then there is a neighborhood V of x and a section sV,0 ∈ F(U ) with ψV (sV,0 ) = t|V by
surjectivity on stalks. Note that
292
13.2. The Category of Sheaves Chapter 13. Sheaf Theory
0 F 0 (V ) F(V ) F 00 (V ) 0
If F is flasque, then the middle column is surjective, so it follows that the right column is
also surjective. Hence F 00 is also flasque.
(1) ϕ is surjective over all open sets U ⊆ X if and only if ϕ is an epimorphism in the
category PreshX .
(2) If F and G are sheaves, then ϕ is surjective on stalks if and only if ϕ is an epimorphism
in the category ShX .
Proof. (1) Let coker ϕ be the presheaf cokernel of ϕ. Then ϕ is an epimorphism if and only
if coker ϕ = 0. For any open U ⊆ X, (coker ϕ)(U ) is the cokernel of ϕU : F (U ) → G(U )
in the category of abelian groups, so it is 0 if and only if ϕU is surjective. On the level of
sheaves, coker ϕ = 0 if and only if (coker ϕ)(U ) = 0 for all U , thus ϕ is surjective on open
sets if and only if it’s an epimorphism.
(2) First suppose ϕ is surjective on stalks. If
ϕ f
F G H
g
fx ◦ ϕx = (f ◦ ϕ)x = (g ◦ ϕ)x = gx ◦ ϕx
293
13.2. The Category of Sheaves Chapter 13. Sheaf Theory
for any family of objects (Ax )x∈X in C. Since C has enough injectives, for each x ∈ X we
may choose a monic ϕx : Gx ,→ Ax with Ax injective, so that the induced morphism
Y
G −→ x∗ Ax
x∈X
Q
is monic and x∈X x∗ Ax is injective.
Thus for a category C with products and enough injectives, the right derived functors of
the section functor Γ : ShX (C) → C are defined.
294
13.3. Sheaf Cohomology Chapter 13. Sheaf Theory
H i (X, F ) ∼
= H i (Γ(X, L• )).
where ei is the inclusion of a subsheaf and the gi satisfy ei+1 ◦ gi = fi . Since Γ is left exact
(Theorem 13.2.10), each sequence
Γe Γgi
0 → Γ(X, K i ) −−→
i
Γ(X, Li ) −−→ Γ(X, K i+1 )
is exact. Note that Γei identifies Γ(X, K i ) with ker Γgi , which is equal to ker Γfi since Γei is
injective. Likewise, im Γgi−1 = im Γfi−1 . Therefore Γ(X, K i )/ im Γgi−1 ∼ = ker Γfi / im Γfi−1 =
H i (Γ(X, L• )). On the other hand, there is a long exact sequence
Γgi
0 → Γ(X, K i ) → Γ(X, L• ) −−→ Γ(X, K i+1 ) → H 1 (X, K i ) → H 1 (X, Li ) = 0
H 1 (X, K i ) ∼
= Γ(X, K i+1 )/ im Γgi = H i+1 (Γ(X, L• )).
The same argument shows that there is an isomorphism H j−1 (X, K i+1 ) ∼
= H j (X, K i ) for each
j ≥ 2. In particular, H (X, K ) = H (X, K ) = H (K, F ) so we obtain H i+1 (X, F ) ∼
1 i ∼ i+1 0 i+1
=
H i+1 (Γ(X, L• )) for all i ≥ 0.
In fact, Theorem 13.3.1 holds in much greater generality, namely for the derived functors
of a left exact functor out of an abelian category with enough injectives, but we will only
need it for sheaf cohomology computations in these notes.
295
13.3. Sheaf Cohomology Chapter 13. Sheaf Theory
Example 13.3.2. Let X be a smooth, paracompact manifold and define the sheaf of rings
U 7−→ OX (U ) = {f : U → R | f is smooth}.
The pair (X, OX ) is called a ringed space. A sheaf of abelian groups F on X is called an
OX -module if for all U ⊆ X, F(U ) is an OX (U )-module such that the module structure
OX (U ) × F(U ) → F(U ) is compatible with restrictions F(U ) → F(V ). Examples of OX -
modules include the sheaves ΩkX (differential k-forms), VF X (smooth vector fields), Γ(E)
(sections of any vector bundle E), etc. (See Section 14.4 for more details.) Using this
language, we have our first of several ‘vanishing theorems’:
Theorem 13.3.3. For a smooth, paracompact manifold X and any OX -module F,
H i (X, F) = 0
for all i > 0.
Q We induct on i > 0. For an OX -module F, the proof of Theorem 13.2.15 shows that
Proof.
I = x∈X Fx is an injective sheaf and F ,→ I is injective. It’s easy to check that I is also
an OX -module. Thus we obtain a short exact sequence of OX -modules
0 → F → I → I/F → 0.
By definition H 1 (X, F) = coker(Γ(X, I) → Γ(X, I/F)). Take a section s ∈ Γ(X, I/F).S By
the sheaf axioms and paracompactness, we may choose a locally finite open cover Ui = X
and sections ti ∈ I(Ui ) such that ti 7→ s|Ui . Fix a partition of unity fi subordinate to the
Ui , i.e. functions fi ∈ OX (X) such that for each Ui ,
X
fi |XrUi = 0 and fj ≡ 1.
j
Then fi ti extendsPto a global section ofPI vanishing away from Ui , so by the sheaf axioms,
the section t = f t
j j j maps to s = j fj s. This shows that Γ(X, I) → Γ(X, I/F) is
1
surjective, so H (X, F) = 0. The inductive step follows from analyzing the same short exact
sequence, which gives H k (X, F) ∼= H k−1 (X, I/F).
In fact, the proof of Theorem 13.3.3 shows:
Corollary 13.3.4. If F is a flasque sheaf on a smooth, paracompact manifold X, then
H i (X, F) = 0 for all i > 0.
This shows that flasque sheaves are acyclic, so we may compute H i (X, G) using a flasque
resolution of G. (We will generalize this in Section 16.2.)
In general, cohomology is hard to compute from the derived functors definition. However,
we may begin by recovering ordinary (singular) cohomology with coefficients:
Theorem 13.3.5. Let X be a topological space, A an abelian group and AX the constant
sheaf on X with stalks A. Then
H i (X, AX ) ∼
= H i (X; A)
where H i (X; A) denotes the ordinary cohomology of X with coefficients in A.
296
13.3. Sheaf Cohomology Chapter 13. Sheaf Theory
Proof. For each open set U ⊆ X, let Cn (U ; A) denote the A-module spanned by singular
simplices on U and let ∂ : Cn (U ; A) → Cn−1 (U ; A) denote the simplicial boundary map.
Dualizing gives us the modules of singular cosimplices C n (U ; A) := HomA (Cn (U ; A), A) with
coboundary map given by the adjoint, d = ∂ ∗ : C n (U ; A) → C n+1 (U ; A). The assignment
U 7→ C n (U ; A) defines a presheaf C n on X for each n ≥ 0. In fact by construction, for any
inclusion V ,→ U , the restriction map C n (U ; A) → C n (V ; A) is surjective, so C n is a flasque
presheaf on X. Note that C n satisfies the gluing axiom of a sheaf, but it is not separated.
However, sheafifying a flasque presheaf which satisfies the gluing axiom yields a flasque sheaf
in general, so setting C n := (C n )sh , we get a complex of flasque sheaves C • . The differential
d : C n → C n+1 is induced by the coboundary maps dU : C n (U ; A) → C n+1 (U ; A). In fact,
one can show that C n is isomorphic to the quotient sheaf C n /C0n , where
C0n (U ) = {ϕ ∈ C n (U ) | there exists an open cover U = {Ui } of U such that ϕ|Ui ≡ 0}.
H i (X, AX ) ∼
= H i (Γ(X, C • ))
for all i ≥ 0.
Finally, singular cohomology is computed by the complex of abelian groups C • (X) but
we have a short exact sequence of complexes of presheaves
0 → C0• → C • → C • → 0
where the limit is taken over all open covers U of X. This identification is compatible with
differentials, so we get C0• = lim CU• ,0 . By a result in algebraic topology, there exists a cover
−→
U for which CU• ,0 is acyclic, which finishes the proof.
Now the strategy for computing general sheaf cohomology is clear: we may take an in-
jective resolution of AX and compute the cohomology of the resulting complex. Of course,
different ways of resolving a constant sheaf can give different ways of computing cohomol-
ogy, and these observations will produce beautiful comparisons between different flavors of
cohomology theory.
297
13.3. Sheaf Cohomology Chapter 13. Sheaf Theory
298
13.4. Čech Cohomology Chapter 13. Sheaf Theory
Set tij = fi |Ui ∩Uj − fj |Ui ∩Uj . Then if the above equation is satisfied, we have tij ∈ O(Ui ∩ Uj ),
where O is the sheaf of holomorphic functions on X and
299
13.4. Čech Cohomology Chapter 13. Sheaf Theory
Definition. The first Čech cohomology group of the cover U with coefficients in F is
the quotient group
1 1 1
Ȟ (U, F ) := Ž (U, F )/B̌ (U, F ).
Now to solve the Mittag-Leffler problem, it’s enough to show that for a Riemann surface
1
X with cover U, Ȟ (U, O) = 0 where O is the sheaf of holomorphic functions on X. Note
that these Čech cohomology groups depend on the cover U. To package together all of
the information about holomorphic functions on covers of X, we introduce the notion of a
refinement of covers.
Definition. Let U = {Ui }i∈I and V = {Vj }j∈J be open covers of X. We say V is a refine-
ment of U if there exists a function τ : J → I such that for all j ∈ J, Vj ⊆ Uτ (j) .
1 1
Lemma 13.4.1. If τ is a refinement from U to V, then τ 1 (B̌ (U, F )) ⊆ B̌ (V, F ).
1 1
Thus there is an induced map on Čech cohomology groups, Ȟ (U, F ) → Ȟ (V, F ) and
one can check that this does not depend on the choice of map τ : J → I.
Definition. The first Čech cohomology of X with coefficients in F is the direct limit
1 1
Ȟ (X, F ) = lim Ȟ (U, F )
−→
300
13.4. Čech Cohomology Chapter 13. Sheaf Theory
1
Proof. Let U = {Ui }i∈I be an open cover of X. If g = (gik )i,k∈I ∈ Ȟ (U, F ) becomes 0 in
1
Ȟ (X, F ), this means there is some open cover V = {Vj }j∈J which is a refinement of U,
1
say by τ : J → I, such that g 0 = (gj`
0
)j,` = gτ (j)τ (`) |Vj ∩V` j,` lies in B̌ (V, F ). Since g 0 is a
0
Q
1-coboundary, there is a family (hj ) ∈ j F (Vj ) such that gj` = hj − h` on Vj ∩ V` . For a
fixed index i ∈ I, observe that
0
gj` = gτ (j)τ (`)
= gτ (j)i + giτ (`) by the cocycle condition
= giτ (`) − giτ (j) by the cocycle condition again
= hj − h` on Vj ∩ V` .
Consider the cover {Ui ∩ Vj }j∈J of Ui . By the sheaf axioms for F , there exists a section
ti ∈ F (Ui ) such that for each Vj , ti |Ui ∩Vj = (giτ (j) + hj )|Ui ∩Vj . Now let i, k ∈ I be arbitrary
and fix some j ∈ J. Then
gik = giτ (j) + gτ (j)k = giτ (j) − gkτ (j) by the cocycle condition
= (ti − hj ) − (tk − hj ) by the above
= ti − tk on Ui ∩ Uk ∩ Vj .
Since j was arbitrary and {Ui ∩ Uk ∩ Vj }j∈J covers Ui ∩ Uk , the sheaf axioms imply gik − ti − tk
1 1
on Ui ∩ Uk . Hence gik ∈ B̌ (U, F ) so g = 0 in Ȟ (U, F ).
Definition. Let U = {Ui }i∈I be an open cover of X and let F be a sheaf of abelian groups
on X. The Čech complex of F with respect to U is the cochain complex C • (U, F ) defined
by Y
C p (U, F ) = F (Ui0 ∩ · · · ∩ Uip )
i0 ,...,ip ∈I
with differential
301
13.4. Čech Cohomology Chapter 13. Sheaf Theory
302
13.4. Čech Cohomology Chapter 13. Sheaf Theory
Proposition 13.4.8. For an open cover U and a sheaf F , there is an exact sequence of
sheaves
0 → F → C 0 (U, F ) → C 1 (U, F ) → · · ·
Proof. The sheaf axioms on F imply F → C 0 (U, F ) is injective. To prove exactness at
C p (U, F ), we need to check that the sequence of stalks
α β
C p−1 (U, F )x −
→ C p (U, F )x →
− C p+1 (U, F )x
is exact for all x ∈ X. Since U covers X, we know x ∈ Uj for some Uj . Take fx ∈ C p (U, F )x
and let V be a neighborhood of x and f ∈ Γ(V, C p (U, F )) with f |x = fx . We may assume
V ⊆ Uj . Then for each p ≥ 1,
is well-defined and independent of the choice of V and Uj . Now with the same fx and f as
above, we have
Hence θp+1 d + dθp is the identity on C p (U, F )x , which shows the identity map on this stalk
is chain homotopic to the zero map and thus ker β = im α, proving exactness.
Definition. For a sheaf F on X, the exact sequence
0 → F → C 0 (U, F ) → C 1 (U, F ) → · · ·
for all p ≥ 0.
Proof. Applying the global sections functor to the Čech resolution of F yields
0 1
0 → F (X) → Č (U, F ) → Č (U, F ) → · · ·
Clearly the cohomology groups of this sequence are precisely the Čech cohomology groups
p
Ȟ (U, F ). Therefore there is a chain map
303
13.4. Čech Cohomology Chapter 13. Sheaf Theory
id
0 F E0 E1 ···
p
for any injective resolution E • of F . This induces the maps on cohomology, Ȟ (U, F ) →
H p (X, F ), and one can check that they commute with the maps coming from refinements of
open covers. The second statement follows.
The sheafification functor PreshX → ShX (Theorem 13.1.17) can alternatively be con-
structed using the Čech resolution. In fact, this construction works in a much more general
context known as Grothendieck topology. Let F be a presheaf on a space X, U = {Ui } an
open cover of X and
0
n Y o
Ȟ (U, F ) = f = (fi ) ∈ F (Ui ) : fi |Ui ∩Uj = fj |Ui ∩Uj for all i, j .
0
where the limit is over all open covers U 0 of U . Define the set F # (U ) = Ȟ (U, F |U ). For
0
an inclusion of open sets V ,→ U in X, there is a natural restriction map Ȟ (U, F |U ) →
0
Ȟ (V, F |V ) coming from restriction of an open cover. This makes U 7→ F # (U ) into a presheaf
on X. Moreover, Proposition 13.4.8 shows that F → F # is injective.
Definition. We call a presheaf
S F on X a separated presheaf if it satisfies the first sheaf
axiom; that is, if U = Ui is an open cover and there are sections s, t ∈ F (U ) such that
s|Ui = t|Ui for all Ui , then s = t.
Theorem 13.4.10. Let F be a presheaf on X. Then
(1) F # is a separated presheaf on X.
(3) F ## is a sheafification of F .
Proof. (1) Suppose s, t ∈ F # (U ) and there is an open cover {Vj } of U such that s|Vj = t|Vj
for all Vj . We must show s = t in F # (U ). Since they are elements of a direct limit, s and t
have representatives
0
sU = (si ), tU = (ti ) ∈ Ȟ (U, F |U )
on some open cover U of U – a priori there may be two different covers supporting s and t,
but after a common refinement of the covers we may assume this takes place on just a single
cover. By hypothesis, there is an open cover Wj = {Wjk }k of each Vj that is a refinement of
{Vj ∩ Ui }i such that si |Wjk = ti |Wjk for all k. Then W = {Wjk }j,k is an open cover of U on
304
13.4. Čech Cohomology Chapter 13. Sheaf Theory
0
which si |Wjk = ti |Wjk on every element. Therefore the images of sU and tU in lim Ȟ (U, F |U )
−→
are equal, so s = t as required.
(2) Now assume F itself is separated; we must show that F # satisfies the second sheaf
axiom. As an intermediate step, we claim that for any open cover U = {Ui }i∈I of U , the
map
0 0
Ȟ (U, F |U ) −→ Ȟ (U, F |U )
0
is injective (this is conditioned on F being separated). Notice that if s, t ∈ Ȟ (U, F |U )
0
become equal in the direct limit Ȟ (U, F |U ), then there exists a cover V = {Vj }j∈J of U
such that s|Vj = t|Vj for all j ∈ J. We may assume V is a refinement of U, say by a map
τ : J → I. Then sτ (j) |Vj = tτ (j) |Vj for all j ∈ J. Since s and t are Čech 0-cocycles, we have
si |Ui ∩Uτ (j) = sτ (j) |Ui ∩Uτ (j) and ti |Ui ∩Uτ (j) = tτ (j) |Ui ∩Uτ (j)
for all i, j. Fixing i ∈ I, the collection {Ui ∩Vj }j is an open cover of Ui and si |Ui ∩Vj = ti |Ui ∩Vj ,
0
so by the separated condition, si = ti on Ui . This implies s = t in Ȟ (U, F |U ), proving the
claim.
Now take si ∈ F # (Ui ) for some cover U = {Ui }i∈I of U such that si |Ui ∩Uj = sj |Ui ∩Uj for
all i, j ∈ I. We must construct a section s ∈ F # (U ) restricting to si on each Ui . Each si is
0
represented by some sVi = (sik ) ∈ Ȟ (Vi , F |Ui ) for an open cover Vi = {Vik }k of Ui . Then
V = {Vik }i,k is an open cover of U . Fixing i, j ∈ I, consider the cover W = {Vik ∩ Vj` }k,` of
Ui ∩ Uj . The sections
0
s̃i = (sik |Vik ∩Vj` ), s̃j = (sj` |Vik ∩Vj` ) ∈ Ȟ (W, F |Ui ∩Uj )
0
become equal in Ȟ (Ui ∩ Uj , F |Ui ∩Uj ) by assumption, so because F is separated, the previous
0
paragraph shows s̃i = s̃j in Ȟ (W, F |Ui ∩Uj ). Therefore the s̃i give a well-defined element
0
s ∈ Ȟ (U, F |U ) restricting to si on each Ui . Hence F # is a sheaf.
(3) Suppose F → G is a morphism of presheaves, where G is a sheaf. This determines a
chain map
C • (U, F ) −→ C • (U, G)
for any open cover U of X. Hence there is a map on Čech cohomology
p p
Ȟ (U, F ) −→ Ȟ (U, G)
305
13.4. Čech Cohomology Chapter 13. Sheaf Theory
F (U ) G(U )
∼
=
0 0
Ȟ (U, F |U ) Ȟ (U, G|U )
The right column is an isomorphism since G is a sheaf. Thus we may defined the dashed
0
arrow Ȟ (U, F |U ) → G(U ) which induces a morphism of presheaves F # → G through which
F → G factors. Repeating the argument shows that F → G further factors through a
morphism F ## → G. Uniqueness follows from a similar argument as the one in the proof
of Theorem 13.1.17.
Changing topics, we next describe some useful applications of sheaf cohomology (using
Čech complexes for computation).
Theorem 13.4.11. For any paracompact space X and sheaf F, there exists an injective
sheaf I on X such that
H n (X, F) ∼
= H n−1 (X, I/F)
for all n.
Let us illustrate this for n = 1. Note that H 1 (X, F) = Ext1ShX (Ab) (ZX , F) where ZX is a
constant sheaf. Explicitly, elements of Ext1 (ZX , F) are isomorphism classes of short exact
sequences of sheaves
0 → F → G → ZX → 0.
Assume X is connected, so that Γ(X, ZX ) ∼ = Z. For any short exact sequence 0 → F →
G → ZX → 0, we have a long exact sequence
δ
− H 1 (X, F) → · · ·
0 → Γ(X, F) → Γ(X, G) → Z →
Set ε = δ(1). Then ε = 0 if and only if there is a splitting Z → Γ(X, G) of δ. There are open
sets {Ui } in X such that Γ(Ui , G) → Γ(Ui , ZX ) is surjective for all i (although this need not
be true for global sections). So on Ui , the short exact sequence
splits, i.e. there is some σi ∈ G(Ui ) mapping to 1 ∈ Z. For each i, j, set τij = σi − σj ∈
G(Ui ∩ Uj ). Then τij maps to 1 − 1 = 0 in Γ(Ui ∩ Uj , ZX ), so by exactness τij ∈ F(Ui ∩ Uj ).
Also note that each σi is unique up to adding any bi ∈ F(Ui ), but if σi0 = σi + bi , we get
τij0 = τij + bi − bj . Viewing (τij ) and (τij0 ) as Čech 1-cocycles, the previous statement shows
1
that [(τij )] = [(τij0 )] in Ȟ (X, F). Q
Conversely, given a 1-cocycle (τij ) in F(Ui ∩ Uj ), we show how to reconstruct the
extension 0 → F → G → ZX → 0. We define G by its éspace étale:
where (x, f, n, Ui ) ∼ (x, f +nτij , n, Uj ) for every x ∈ Ui ∩Uj . Thus locally, G ét |Ui ∼
= F ét |Ui ×Z.
Over Ui , the section σi : x 7→ (x, 0, 1, Ui ) gives a splitting of δ : Γ(X, G) → Z, since
306
13.4. Čech Cohomology Chapter 13. Sheaf Theory
making the diagram commute. The data of a principal bundle is equivalent to specifying
an open cover {Ui } of X and “transition functions” τij : Ui ∩ Uj → G (which preserve the
extra structure on the manifold). Explicitly, if x ∈ Ui ∩ Uj then (x, g) 7→ (x, gτij (x)) is an
isomorphism on (Ui ∩ Uj ) × G as a trivial bundle over Ui ∩ Uj . It follows from this description
that τji = τij−1 and τij τjk = τik for all i, j, k. So (τij ) is a Čech cocycle for some sheaf, but
which one? Let ΓG be the sheaf on X given locally by
ΓG (U ) = {σ : U → U × G | σpσ = idU }.
Then the above description shows that principal G-bundles up to isomorphism are in one-
to-one correspondence with elements of H 1 (X, ΓG ).
Example 13.4.13. Let F be a field and G = GLn (F ). Then principal G-bundles are in
one-to-one correspondence with rank n vector bundles over X, so the isomorphism classes
of rank n vector bundles are determined by classes in H 1 (X, ΓGLn (F ) ). If F = R and n = 1,
so that GL1 (R) = R× , we have a short exact sequence of abelian groups
exp sgn
0 → R −−→ R× −−→ Z/2Z → 0.
shows that complex line bundles up to isomorphism are determined by classes in H 1 (X, ΓC× ) ∼
=
H 2 (X; Z), called Chern classes, denoted c(L).
307
Chapter 14
Scheme Theory
308
14.1. The Spectrum of a Ring Chapter 14. Scheme Theory
Lemma 14.1.2. Let E be a subset of A and set a = (E), the ideal generated by E. Then
(i) If a is the ideal generated by E then V (E) = V (a) = V (r(a)).
309
14.1. The Spectrum of a Ring Chapter 14. Scheme Theory
E. Thus p ∈TV (a) so we have V (E) = V (a). For the second inequality, Corollary 1.1.3 says
that r(a) = {P primes | P ⊇ a}, which directly implies V (a) = V (r(a)).
(ii) Every prime ideal p ⊂ A contains 0, so V (0) = X. On the other hand, if p is a prime
ideal containing 1 then p = A, which is impossible. Therefore V (1) = ∅.
(iii) For each Ei , let ai = (E
S Si ) be the
ideal generated by Ei , and let a be the ideal generated
by i∈I Ei . Then by (i), V T each i. Clearly ai ⊆ a
i∈I i = V (a) and V (Ei ) = V (ai ) for
E
for each i, soTV (a) ⊆ V (ai ), and since this holds for all i, V S(a) ⊆ i∈I V (ai ). On the other
hand, if p ∈ i∈I V (ai ) then p ⊇ aiT⊇ Ei for each i, so p ⊇ i∈I Ei . By definition of a, this
implies p ⊇ a, so p ∈ V (a). Hence i∈I V (ai ) ⊆ V (a) so we have equality.
(iv) By Lemma 1.1.4, r(a)r(b) ⊆ r(ab) so (i) gives us
(c) More generally, if B/f (A) is an integral ring extension, f ∗ is a closed map.
Proof. (a) By Lemma 14.1.2(i), it suffices to show that for any ideal I ⊂ A, (f ∗ )−1 (V (I)) =
V (f (I)). Note that for a prime p ∈ Spec(B),
Therefore (f ∗ )−1 (V (I)) = V (f (I)) so closed sets pull back to closed sets. This implies f ∗ is
continuous.
(b) Let J ⊂ B be any ideal. We claim that f ∗ (V (J)) = V (f −1 (J)). By Lemma 14.1.2(i),
this is enough to conclude that f ∗ is closed. Set I = f −1 (J) ⊂ A. Since f is surjective,
the map B/J → A/I is also surjective. Then V (J) = Spec(B/J) and V (I) = Spec(A/I)
so replacing A with A/I and B with B/J, it suffices to show that f ∗ (Spec(B)) = Spec(A).
If p ∈ f ∗ (Spec(B)) then p = f −1 (Q) for some prime ideal q ⊂ B. By Lemma 14.1.1,
310
14.1. The Spectrum of a Ring Chapter 14. Scheme Theory
⊇ {p ∈ Spec(A) | p ∈ Y } = Y
311
14.1. The Spectrum of a Ring Chapter 14. Scheme Theory
( ⇒= ) Conversely, assume r(I) is prime and V (I) = Y1 ∪Y2 for closed sets Y1 , Y2 ⊆ V (I).
Then there exist closed sets X1 , X2 ⊆ Spec(A) with Y1 = X1 ∩ V (I) and Y2 = X2 ∩ V (I).
For some ideals a1 , a2 ⊂ A, we have X1 = V (a1 ) and X2 = V (a2 ), so V (I) ⊆ X1 ∪ X2 =
V (a1 ) ∪ V (a2 ) = V (a1 a2 ) by Lemma 14.1.2(iv). Applying J, we get r(I) ⊇ J(V (a1 a2 )) =
r(a1 a2 ) ⊇ a1 a2 . Since r(I) is prime, a1 ⊆ r(I) or a2 ⊆ r(I); without loss of generality, assume
a1 ⊆ r(I). Then by (a), V (I) ⊆ V (r(I)) ⊆ V (a1 ) = X1 . This implies Y1 = X1 ∩V (I) = V (I),
so V (I) is irreducible.
Proposition 14.1.5. Let A be a noetherian ring and X = Spec(A) its prime spectrum.
Then
(a) X is noetherian.
Since A is noetherian, there is an n ∈ N such that r(an ) = r(an+1 ) = · · · . Then since each
Xj was taken to be closed, Proposition 14.1.4(a) implies V (J(Xj )) = V (r(aj )) = Xj and so
we get
312
14.1. The Spectrum of a Ring Chapter 14. Scheme Theory
the quotient to get A = k[tn | n ∈ N]/I. The ideal J = (tn | n ∈ N) strictly contains I,
so J = J/I is an ideal of A which is clearly not finitely generated. So A is not noetherian.
However, k[tn | n ∈ N]/J ∼= k is a field, so J is maximal and hence J is maximal in A. On
the other hand, since every generator of J is nilpotent, every element of J is nilpotent, so
J must also be a minimal prime ideal. This implies J is the unique prime ideal of A, so
Spec(A) is finite and in particular a noetherian space.
Next, for any prime ideal p ⊂ A, let Ap denote the localization at p. For any open set
U ⊆ Spec A, we define
( )
a f
O(U ) = s : U → Ap s(p) ∈ Ap , ∃ q ∈ V ⊆ U such that s(q) = for all q ∈ V, f, g ∈ A .
p∈U
g
313
14.1. The Spectrum of a Ring Chapter 14. Scheme Theory
Example 14.1.8. For any field k, Spec k is a single point ∗ corresponding to the zero ideal,
with sheaf O(∗) ∼
= k.
Example 14.1.9. Let A = k[t1 , . . . , tn ] be the polynomial ring in n variables over k. Then
Spec A = Ank , the affine n-space over k. For example, when A = k[t] is the polynomial ring
in a single variable, Spec k[t] = A1k , the affine line.
When k = C, Hilbert’s Nullstellensatz (Corollary 2.4.3) tells us that all the closed points
of A1k correspond to maximal ideals of the form (t − α) for α ∈ C. But there is also a
non-closed, ‘generic point’ corresponding to the zero ideal which was not detected before.
closed points
−2 0 1+i generic point
Spec C[t]
(t + 2) (t) (t − (1 + i)) (0)
On the other hand, if k = Q or another non-algebraically closed field, the same closed
points corresponding to linear ideals (t − α) show up, as well as the generic point cor-
responding to (0), but there are also points corrresponding to ideals generated by higher
degree irreducible polynomials like t2 + 1. Thus the structure of Spec Q[t] is much different
than the algebraically closed case.
closed points
−2 0 ?? generic point
Spec Q[t]
(t + 2) (t) (t2 + 1) (0)
Example 14.1.10. Let X be an algebraic variety over a field k, x ∈ X a point and consider
the affine scheme Y = Spec(k[ε]/(ε2 )). We can think of Y as a “big point” with underlying
space ∗ corresponding to the zero ideal, along with a “tangent vector” extending infinites-
imally in every direction around ∗. Then any map Y → X determines a unique tangent
vector in Tx X, the tangent space of X at x. This idea is useful in intersection theory. For
example, consider the tangency of the x-axis and the parabola y = x2 in A2k :
y − x2
y
(0, 0)
314
14.1. The Spectrum of a Ring Chapter 14. Scheme Theory
As a variety, this point (0, 0) corresponds to the quotient of k-algebras k[x, y]/r(y, y − x2 ) =
k[x]/r(x2 ) = k[x]/(x) = k. Thus the information of tangency is lost. However, as an affine
scheme, (0, 0) corresponds to Spec(k[x, y]/(y, y − x2 )) = Spec(k[x]/(x2 )) so the intersection
information is preserved.
Example 14.1.11. Every affine algebraic variety X may be viewed as an affine scheme. In-
deed, let I(X) ⊆ k[t1 , . . . , tn ] be the vanishing ideal of X and set A = k[X] = k[t1 , . . . , tn ]/I(X).
Then (X, OX ) determines an affine scheme (Spec A, O). Moreover, the assignment
315
14.2. Schemes Chapter 14. Scheme Theory
14.2 Schemes
In this section we define a scheme and prove some basic properties resulting from this defi-
nition. Recall that a ringed space is a pair (X, F) where X is a topological space and F is
a sheaf of rings on X.
Definition. A locally ringed space is a ringed space (X, F) such that for all P ∈ X,
there is a ring A such that FP ∼
= Ap for some prime ideal p ⊂ A.
Example 14.2.1. Any affine scheme Spec A is a locally ringed space by (1) of Theo-
rem 14.1.7. We will sometimes denote the structure sheaf O by OA .
Definition. The category of locally ringed spaces is the category whose objects are
locally ringed spaces (X, F) and whose morphisms are morphisms of ringed spaces (X, F) →
(Y, G) such that for each P ∈ X, the induced map fP# : OY,f (P ) → OX,P is a morphism of
local rings, i.e. (fP# )−1 (mP ) = mf (P ) where mP (resp. mf (P ) ) is the maximal ideal of the local
ring OX,P (resp. OY,f (P ) ).
We are now able to define a scheme.
Definition. A scheme is a locally ringed space (X, OX ) that admits an open covering {Ui }
such that each Ui is affine, i.e. there are rings Ai such that (Ui , OX |Ui ) ∼
= (Spec Ai , OAi ) as
locally ringed spaces.
The category of schemes Sch is defined to be the full subcategory of schemes in the
category of locally ringed spaces. Denote the subcategory of affine schemes by AffSch. Also
let CommRings denote the category of commutative rings with unity.
Proposition 14.2.2. There is an isomorphism of categories
∼
AffSch −→ CommRingsop
(X, OX ) 7−→ OX (X)
(Spec A, O) →−7 A.
Proof. (Sketch) First suppose we have a homomorphism of rings f : A → B. By Lemma 14.1.1
this induces a morphism f ∗ : Spec B → Spec A, p 7→ f −1 (p) which is continuous since
f −1 (V (a)) = V (f (a)) for any ideal a ⊂ A. Now for each p ∈ Spec B, define the localiza-
tion fp : Af ∗ p → Bp using the universal property of localization. Then for any open set
V ⊆ Spec A, we get a map
f # : OA (V ) −→ OB ((f ∗ )−1 (V )).
One checks that each is a homomorphism of rings and commutes with the restriction maps.
Thus f # : OA → OB is defined. Moreover, the induced map on stalks is just each fp , so the
pair (f ∗ , f # ) gives a morphism (Spec B, OB ) → (Spec A, OA ) of locally ringed spaces, hence
of schemes.
Conversely, take a morphism of schemes (ϕ, ϕ# ) : (Spec B, OB ) → (Spec A, OA ). This
induces a ring homomorphism Γ(Spec A, OA ) → Γ(Spec B, OB ) but by (2) of Theorem 14.1.7,
Γ(Spec A, OA ) ∼ = A and Γ(Spec B, OB ) ∼
= B so we get a homomorphism A → B. It’s easy
to see that the two functors described give the required isomorphism of categories.
316
14.2. Schemes Chapter 14. Scheme Theory
Example 14.2.3. We saw in Example 14.1.8 that for any field k, Spec k = ∗ is a point
` O(∗)
with structure sheaf ` = k. If A = L1 × · · · × Lr is a finite étale k-algebra, then
Spec A = Spec L1 · · · Spec Lr is (schematically) a disjoint union of points.
Example 14.2.4. Let A be a DVR with residue field k. Then Spec A = {0, mA }, a closed
point for the maximal ideal m and a generic point for the zero ideal. There are two open
subsets here, {0} and Spec A, and we have OA ({0}) = k and OA (Spec A) = A.
Example 14.2.5. If k is a field and A is a finitely generated k-algebra, then the closed
points of X = Spec A are in bijection with the closed points of an affine variety over k with
coordinate ring A (see Example 14.1.11).
Example 14.2.6. Let A = Z (or any Dedekind domain). Then dim A = 1 and it turns out
that dim Spec A = 1 for some appropriate notion of dimension (see Section 14.3). Explicitly,
Spec Z has a closed point for every prime p ∈ Z and a generic point for (0):
closed points
generic point
Spec Z
2 3 5 7 11 (0)
Example 14.2.7. Let k be a field, X1 = X2 = A1k two copies of the affine line and U1 =
U2 = A1k r {0}, where 0 is the closed point of A1k corresponding to (x) in k[x]. Then we can
glue together X1 and X2 along the identity map U1 → U2 to get a scheme X which looks
like the affine line with the origin “doubled”. Note that X is not affine!
A1k r {0}
317
14.3. Properties of Schemes Chapter 14. Scheme Theory
C = {P ∈ U | fP ∈ mP ⊂ OX,P }
D = {P ∈ U | gP ∈ mP ⊂ OX,P }.
Proposition 14.3.2. Let A be a noetherian ring. Then dim Spec A = dim A, the Krull
dimension of A.
Note that by commutative algebra, the codimension of P is equal to the height of the
prime ideal p ⊂ A associated to P for any choice of affine open neighborhood P ∈ U =
Spec A.
318
14.3. Properties of Schemes Chapter 14. Scheme Theory
An integral scheme X is normal if each stalk OX,P is integrally closed in its field of
fractions.
X is regular if each OX,P is regular as a local ring, that is, dim OX,P = dim mP /m2P
as OX,P /mP -vector spaces.
Definition. Let U ⊆ X be an open subset. Then (U, OX |U ) is a scheme which we call an
open subscheme of X. The natural morphism j : U ,→ X, j # : OX → j∗ OX |U is called
an open immersion.
Example 14.3.3. For X = Spec A, let f ∈ A and recall the open set D(f ) defined in
Theorem 14.1.7. Then D(f ) is an open subscheme of X and the open immersion D(f ) ,→ X
corresponds to the natural inclusion of prime ideals Spec Af ,→ Spec A (this is a property of
any localization).
Definition. Let A → A/I be a quotient homomorphism of rings. Then the induced mor-
phism Spec(A/I) → Spec A is called an affine closed immersion. For a general morphism
of schemes f : X → Y , f is called a closed immersion if f is injective, f (X) ⊆ Y is closed
and there exists a covering of X by affine open sets {Ui } such that each f |Ui : Ui → f (Ui ) is
an affine closed immersion. The set f (X) is called a closed subscheme of Y .
Definition. Let X be a scheme. The category of schemes over X, denoted SchX , consists
p
of objects Y →
− X, where Y is a scheme and p is a morphism, and morphisms Y → Z making
the following diagram commute:
Y Z
{x} X
We next construct fibre products in the category SchX and use these to construct the alge-
braic analogue of a fibre.
Definition. Let X be a scheme and Y, Z schemes over X. A fibre product of Y and Z
over X, denoted Y ×X Z, is a scheme over both Y and Z such that the diagram
319
14.3. Properties of Schemes Chapter 14. Scheme Theory
Y ×X Z
Y Z
commutes and Y ×X Z is universal with respect to such diagrams, i.e. for any scheme W
over both Y and Z, the following diagram can be completed uniquely:
∃!
Y ×X Z
Y Z
Theorem 14.3.5. For any schemes Y, Z over X, there exists a fibre product Y ×X Z which
is unique up to unique isomorphism.
Proof. (Sketch) First suppose X, Y and Z are all affine; write X = Spec A, Y = Spec B
and C = Spec Z. Then Spec(B ⊗A C) is a natural candidate for the fibre product in this
case. Indeed, the tensor product satisfies the universal property conveyed by the following
diagrams:
320
14.3. Properties of Schemes Chapter 14. Scheme Theory
∃!
B ⊗A C B ⊗A C
B C B C
A A
Applying the functor Spec yields the right diagrams with arrows reversed, by Proposi-
tion 14.2.2, so the fibre product exists in the affine case.
In general, note that once we construct any fibre product, it will be unique up to unique
isomorphism by the universal property, just as in every proof of the solution to a universal
mapping problem. Now suppose S X and Z are affine and Y is arbitrary. Write Y as a union
of affine open subschemes Y = Yi . Then by the affine case, Yi ×X Z exists for each Yi .
For each pair of overlapping open sets Yi ∩ Yj , set Uij = p−1
Yi (Yi ∩ Yj ) ⊆ Yi ×X Z, where pYi
is the morphism Yi ×X Z → Yi . Then it’s easy to verify that Uij = (Yi ∩ Yj ) ×X Z (that
is, Uij satisfies the definition of the fibre product for Yi ∩ Yj and Z over X), and by the
universal property, there are unique isomorphisms ϕij : Uij → Uji for each overlapping pair,
commuting with all projections. Therefore we may glue together the fibre products Yi ×X Z
along the isomorphisms ϕij to get a scheme Y ×X Z which then satisfies the definition of the
fibre product for Y and Z over X. Now, covering Z by affine open subschemes and repeating
this process will construct Y ×X Z for any schemes Y, ZSover an affine scheme X.
Finally, let X be an arbitrary scheme and write X = Xi for affine open subschemes Xi .
Let q : Y → X and r : Z → X be the given morphisms and for each Xi , set Yi = q −1 (Xi )
and Zi = r−1 (Xi ). By the affine case, each Yi ×Xi Zi exists, but any morphisms f : W → Yi
and g : W → Z making the diagram
W
Yi Z
commute must satisfy g(W ) ⊆ Zi . It follows that Yi ×Xi Z = Yi ×X Zi and the gluing
procedure from above allows us to construct Y ×X Z from these. Hence the fibre product
exists in every case.
321
14.3. Properties of Schemes Chapter 14. Scheme Theory
(b) The underlying topological space of Yx is homeomorphic to the set p−1 (x) of preimages
of x.
p
(c) The assignment (Y →
− X) 7→ Yx is functorial.
Example 14.3.7. Let A be a DVR and consider the affine scheme X = Spec A. We saw
in Example 14.2.4 that X has a closed point m = mA and a generic point (0). For any
morphism p : Y → X, there are two fibres:
The generic fibre Y(0) , which is an open subscheme of Y
∆
id id
Y ×X Y
Y Y
322
14.3. Properties of Schemes Chapter 14. Scheme Theory
Example 14.3.9. One can show that the affine line with the origin doubled (Example 14.2.7)
is not separated as a scheme over Spec k.
One perspective on separatedness is that it is a suitable replacement for the Hausdorff
condition in algebraic geometry. In the Zariski topology on any scheme, there are always
proper open subsets that are dense, so the Hausdorff property usually fails to hold.
Definition. A morphism f : Y → X is of finite type if there exists an affine covering
Ui , with Ui = Spec Ai , such that each f −1 (Ui ) has an open covering f −1 (Ui ) =
S
X =
S ni
j=1 Spec Bij for ni < ∞ and Bij a finitely generated Ai -algebra. Further, we say f is
a finite morphism if each ni = 1, i.e. f −1 (Ui ) = Spec Bi for some finitely generated
Ai -algebra Bi .
Definition. A separated morphism f : Y → X is proper if it is of finite type and for every
morphism Z → X, the base change morphism Y ×X Z → Z is closed.
Lemma 14.3.10. Let X, Y, Z be noetherian schemes. Then for any morphism f : Y → X,
(a) If f is an open immersion, then f is separated.
(b) If f is a closed immersion, then f is separated and proper.
(c) If g : Z → Y is separated (resp. proper) then the composition f ◦ g : Z → X is
separated (resp. proper).
(d) If Z is a scheme over X, the base change Y ×X Z → Z is separated and proper.
(e) If f is finite, then f is proper.
Example 14.3.11. (Projective line over a scheme) Let X = Spec A be an affine scheme.
Then the “affine line” A1X = Spec A[t] is an affine scheme over X. Set X1 = A1X = Spec A[t]
and X2 = A[t−1 ]. Then each contains an open subscheme isomorphic to U = Spec A[t, t−1 ],
coming from applying Spec to the diagram of A-algebras
A[t, t−1 ]
A[t] A[t−1 ]
Gluing along these isomorphic open subschemes gives us a scheme P1X = X1 ∪U X2 , called
the projective line over X. In the affine case, we will write P1X = P1A .
S When X is an arbitrary scheme, X has a covering by 1open affine subschemes X =
Ui and a gluing construction defines the projective line PX . We will generalize this in
Section 14.6.
323
14.4. Sheaves of Modules Chapter 14. Scheme Theory
OX (U ) × F(U ) F(U )
OX (V ) × F(V ) F(V )
If F(U ) ⊆ OX (U ) is an ideal for each open set U , then we call F a sheaf of ideals on X.
The most important of these constructions for our purposes will be the direct sum oper-
ation.
324
14.4. Sheaves of Modules Chapter 14. Scheme Theory
Remark. The rank of a locally free sheaf of OX -modules is constant on connected com-
ponents. In particular, the rank of a locally free OX -module is well-defined whenever X is
connected.
Definition. A locally free OX -module of rank 1 is called an invertible sheaf.
Let A be a ring, M an A-module and set X = Spec A. To extend module theory to
the language of schemes, we want to define an OX -module M f on X. To start, for each
p ∈ Spec A, let Mp = M ⊗A Ap be the localization of the module M at p. Then Mp is an
Ap -module consisting of ‘formal fractions’ ms where m ∈ M and s ∈ S = A r p. For each
open set U ⊆ X, define
( )
f(U ) = h : U →
a m
M Mp s(p) ∈ Mp , ∃ p ∈ V ⊆ U, m ∈ M, s ∈ A with s(q) = for all q ∈ V .
p∈U
s
(Compare this to the construction of the structure sheaf OA on Spec A in Section 14.1. Also,
note that necessarily the s ∈ A in the definition above must lie outside of all q ∈ V .)
A-Mod −→ OX -Mod
M 7−→ M
f
325
14.4. Sheaves of Modules Chapter 14. Scheme Theory
326
14.4. Sheaves of Modules Chapter 14. Scheme Theory
(a) Ω
e B/A is a quasi-coherent sheaf on Spec B.
e B/A (D(f )) ∼
(b) For any element f ∈ B, Ω = ΩBf /A where Bf is the localization of B at
powers of f .
B −→ ΩB/k
ti 7−→ dti .
(a) O∆(Y ) ∼
= OY ×X Y /I as sheaves on ∆(Y ).
(b) I/I 2 is an O∆(Y ) -module.
Identifying Y with its image ∆(Y ) in the fibre product Y ×X Y allows us to define a
sheaf analogue of the module of differentials by pulling back I/I 2 .
327
14.4. Sheaves of Modules Chapter 14. Scheme Theory
ΩY /X I/I 2
Y ∆(Y )
∆
328
14.4. Sheaves of Modules Chapter 14. Scheme Theory
ωX is an invertible sheaf on X.
One can define the geometric genus of X by g(X) := dimk Γ(X, ωX ). Then when X
is a curve (a smooth scheme of dimension 1), g(X) is equal to the arithmetic genus of
X, another important algebraic invariant. These genera are not equal in general.
If X is a curve over C, then the genus of the corresponding Riemann surface X(C)
is precisely g(X), so the canonical bundle carries important topological information
about X(C).
329
14.5. Group Schemes Chapter 14. Scheme Theory
µ : G × G −→ G (multiplication)
e : {e} ,−→ G (identity)
i : G −→ G (inverse)
satisfying associativity, identity and and inversion axioms. This generalizes to the notion of
a group object in an arbitrary category C. We state the definition for scheme categories here.
p
Definition. A group scheme over a scheme X is a scheme G →
− X with morphisms
µ : G × G −→ G
e : {e} ,−→ G
i : G −→ G
Remark. When G is a finite group scheme over X, flat is equivalent to locally free.
Proposition 14.5.1. Let G be a scheme over X. Then a choice of group scheme structure
on G is equivalent to a compatible choice of group structure on the sets HomX (Y, G) for all
schemes Y over X. That is, a group scheme structure is a functor SchX → Groups such
that the composition with the forgetful functor Groups → Sets is representable.
Example 14.5.2. Let G be a finite group of order Qn n and let X be any scheme. The constant
group scheme on G over X is defined as GX := i=1 X, with projection map induced by the
identity on each disjoint copy of X. Multiplication QnGX ×X GX → GX is given by sending
(P, Q), where P = x ∈ X in the gi th component of i=1 X and Q = x in the gj th component
(gi , gj ∈ G) to the corresponding point P Q = x in the gi gj th component of the disjoint union.
(Note that P and Q must correspond to the same point x ∈ X by definition of the fibre
product – draw the diagram!) Similarly, the identity is the morphism Qtaking X onto the
copy of X indexed by the identity element eG ∈ G, e : X → XeG ⊆ ni=1 X. Finally, the
330
14.5. Group Schemes Chapter 14. Scheme Theory
Example 14.5.3. Let X = Spec A be affine and recall the affine line A1X = Spec A[t]
constructed in Example 14.3.11. Then A1X is an affine group scheme over X, denoted Ga ,
called the additive group scheme over X, with morphisms induced by the following ring
homomorphisms:
Notice that these are just the axioms of a Hopf algebra! This construction generalizes to the
affine line over a non-affine scheme as well.
e∗ : A[t, t−1 ] −→ A
t, t−1 7−→ 1
i∗ : A[t, t−1 ] −→ A[t, t−1 ]
t 7−→ t−1
t−1 7−→ t.
Note that when A = k is a field, these are just the schematic versions of the algebraic groups
Ga,k and Gm,k . This shows that group schemes are a direct generalization of algebraic groups.
Example 14.5.5. For X = Spec A, the nth roots of unity form a group scheme defined by
µn = Spec(A[t, t−1 ]/(tn − 1)). This is a finite group subscheme of Gm .
Example 14.5.6. If char A = p > 0, then αp = Spec(A[t]/(tp )) defines a group scheme over
Spec A which is isomorphic as a scheme to Spec A, but not as a group scheme!
Example 14.5.7. The Jacobian of a curve is a group scheme. In particular, an elliptic curve
(a dimension 1 scheme with a specified point O) is a group scheme with identity O.
331
14.5. Group Schemes Chapter 14. Scheme Theory
p q
Definition. Let G → − X be a finite, flat group scheme. A left G-torsor is a scheme Y → − X
with q finite, locally free and surjective, together with a group action ρ : G ×X Y → Y , which
satisfies:
Remark. The idea is that a G-torsor is exactly the same as G, except we have forgotten
the “identity point” e (which is a morphism, not a point).
Example 14.5.9. Let k be a field and m an integer such that char k - m. Let µm =
Spec(k[t, t−1 ]/(tm − 1)) be the group scheme of mth roots of unity √ over Spec k. Take a ∈
× × m
k /(k ) (that is, a is not an mth power in k) and set L = k( a), which is a finite field
m
This defines a morphism ρ : µm ×Spec k Y → Y and one can prove that it satisfies axioms (1)
and (2) of a torsor by checking the corresponding properties for ρ∗ . When char k - m, µm
is a reduced scheme over Spec k and it’s easy to see that µm ∼ = (Z/mZ)Spec k , the constant
group scheme on Z/mZ over Spec k. Moreover, L/k is a Galois extension with Gal(L/k) ∼ =
∼ Qm
Z/mZ, L ⊗k L = i=1 L and this has a corresponding Galois action which induces the
∼
isomorphism L ⊗k L − → k[t, t−1 ]/(tm − 1) ⊗k L. Applying Spec again, we get the isomorphism
∼
µm ×Spec k Y −
→ Y ×Spec k Y so Y is indeed a µm -torsor.
Using Artin-Schreier
√ theory, one can show that every µm -torsor arises in this way, i.e. as
Spec L for L = k( a).
m
332
14.6. The Proj Construction Chapter 14. Scheme Theory
(i) V (0) = Proj(S), V (1) = ∅ and V (a) = ∅ for any homogeneous ideal a ⊇ S+ .
Proof. Same as the proof of Lemma 14.1.2, using the properties in Lemma 12.5.3.
As in the affine case, this shows that the sets V (a form the closed sets of a Zariski
topology on Proj(S). Next, we define a sheaf of rings O on Proj(S). If p is a homogeneous
prime ideal of S, let S(p) be the subring of elements of degree 0 in the localization T −1 S,
where T = (S r p) ∩ S h and S h is the set of homogeneous elements of S. Define S(f ) similarly
for f ∈ S h . For an open set U ⊆ Proj(S), define
( )
a f h
O(U ) = s : U → S(p) s(p) ∈ S(p) , ∃ q ∈ V ⊆ U such that s(q) = for all q ∈ V, f, g ∈ S .
p∈U
g
Definition. For a graded ring S, the scheme (Proj S, O) is called a projective scheme.
333
14.6. The Proj Construction Chapter 14. Scheme Theory
Example 14.6.3. Let A be a commutative ring and S = A[x0 , . . . , xn ] be the graded poly-
nomial ring in n + 1 variables over A. The projective n-space over A is the projective scheme
If k is an algebraically closed field, then the closed points of Pnk are in bijection with the
points of the projective variety defined in Section 12.5.
Proposition 14.6.4. Suppose ϕ : S → T is a morphism of graded rings and set
Then
(a) U is an open subset of Proj T .
(b) ϕ induces a morphism of schemes ϕ∗ : U → Proj S.
(c) If ϕ|Sd : Sd → Td is an isomorphism for all d, then U = Proj T . Moreover, ϕ∗ :
Proj T → Proj S is an isomorphism.
Example 14.6.5. More generally, if ϕ : S → T is merely surjective, then U = Proj T and
ϕ : Proj T ,→ Proj S is a closed immersion. One can show that every closed subscheme of
Proj S is of the form Proj(S/I) for some homogeneous ideal I ⊆ S.
Example 14.6.6. Suppose f : A → B is a homomorphism of rings and f ∗ : Spec B →
Spec A is the corresponding morphism of affine schemes, then
PnB ∼
= PnA ×Spec A Spec B.
Example 14.6.7. One defines projective n-space over an arbitrary scheme X, written PnX ,
by
PnX := PnZ ×Spec Z Y.
Alternatively, PnX may be constructed by gluing together n + 1 copies of affine n-space
A1X = Spec A[t1 , . . . , tn ] along the isomorphic open subsets Xi = {ti 6= 0} (when X is affine;
in the general case, glue affine subschemes together as in the previous example). The natural
morphism PnX → X generalizes in the following way.
Definition. A morphism of schemes Y → X is projective if it factors through a closed
immersion Y → PnX for some n ≥ 1.
Example 14.6.8. For any ring A and any graded ring S such that S0 = A, the morphism
Proj S → Spec A is a projective morphism.
Theorem 14.6.9. If f : Y → X is a projective morphism of noetherian schemes, then f is
proper.
The idea behind the proof of Theorem 14.6.9 is to first prove Pnk → Spec k is proper
for any n, which is a straightforward adaptation of the proof when Pnk is considered as a
projective algebraic variety. One can then modify this proof for PnZ → Spec Z and then use
the properties of proper morphisms in Lemma 14.3.10 to obtain the general result.
334
14.6. The Proj Construction Chapter 14. Scheme Theory
Let S be a graded ring and X = Proj S. As with affine schemes, there is a functor
GrModS → ShX which sends a graded S-module M to a sheaf of OX -modules M
f defined in
the same way as in Proposition 14.4.2.
f is a sheaf of OX -module on
Proposition 14.6.11. Let M be a graded S-module. Then M
X = Proj S and
(1) For any p ∈ X, Mfp ∼ = M(p) , where the latter denotes the submodule of degree 0
−1
elements in T M , as above.
f|D+ (f ) ∼
(2) For any f ∈ S h ∩ S+ , M =M(f ) as O(D+ (f ))-modules.
g
(3) M
f is quasi-coherent, and if S is noetherian and M is finitely generated, then M
f is
coherent.
Now let A be a noetherian ring and S = A[x0 , . . . , xn ] so that X = Proj S = PnA is the
projective n-space over A defined above. Recall the “shifted” graded ring S(d) for any d ∈ Z,
defined by S(d)k = Sd+k , which is naturally a graded S-module.
Definition. For d ∈ Z, the dth twisting sheaf of X = PnA is the OX -module OX (d) = S(d).
g
For any OX -module F , write F (d) = F ⊗ OX (d).
335
Chapter 15
Morphisms of Schemes
336
15.1. Flat Morphisms Chapter 15. Morphisms of Schemes
337
15.1. Flat Morphisms Chapter 15. Morphisms of Schemes
Proof. The flat condition implies that OX,x is a flat OY,y -module, and by Lemma 8.3.18 it is
even faithfully flat, so by Proposition 8.3.25(c), fx : Spec OX,x → Spec OY,y is surjective.
Proof. Cover Y with affine open sets V1 , . . . , Vr and for each 1 ≤ j ≤ r, cover f −1 (Vj ) with
affine open sets U1j , . . . , Usj j . Then it’s enough to prove each f (Uij ) is constructible in Vj , so
we may reduce to the case of a finite morphism of affine schemes f : Spec B → Spec A. Now
any irreducible closed subset of Spec A is of the form Z = V (p) for some prime ideal p ⊂ A.
Suppose f (Spec B) ∩ V (p) is dense in V (p). Then f (Spec B) ∩ V (p) may be identified with
the image of the morphism
ϕ : Spec(B/pB) −→ Spec(A/p).
Thus we may reduce to the case when A is a domain, so that B is a finitely generated
A-algebra. Then B = A[x1 , . . . , xn , xn+1 , . . . , xm ] where x1 , . . . , xn are algebraically indepen-
dent over A and xn+1 , . . . , xm are algebraic over A[x1 , . . . , xn ]. Thus for each n + 1 ≤ i ≤ m,
there exist polynomials pij ∈ A[x1 , . . . , xn ] such that
d d −1
p0j xj j + p1j xj j + . . . + pdj j = 0
Then for any prime ideal p ⊂ A, the ideal P = p[x1 , . . . , xn ] is prime in A[x1 , . . . , xn ] and
Q 6∈ P. Hence BP is integral over A[x1 , . . . , xn ]P and the result follows from the going up
theorem.
338
15.1. Flat Morphisms Chapter 15. Morphisms of Schemes
339
15.1. Flat Morphisms Chapter 15. Morphisms of Schemes
(a) There exists an element g ∈ B r q such that (M/pM )g is a flat A/p-module and
TorA
1 (M, A/p)g = 0.
is open in X.
Proof. It is enough to prove the theorem when X and Y are affine, say X = Spec B and
Y = Spec A. Note that Corollary 8.3.22 implies the set U contains the generalizations of all
of its points. Also, it follows from Lemma 15.1.13(b) that for all x ∈ U , the set U ∩ {x}
contains an open set of {x}. Hence the conditions in Theorem 15.1.10 are satisfied, so U is
open.
340
15.2. Kähler Differentials Chapter 15. Morphisms of Schemes
d
g
Ω1B/A
As with any universal object, if it exists, Ω1B/A is unique up to isomorphism. For its
existence, we have:
Proposition 15.2.2. For any A-algebra B, the module of relative differentials of B may be
defined by
Ω1B/A = Zhdb | b ∈ Bi/N
where N is the submodule generated by elements of the form da for a ∈ A and d(b+b0 )−db−db0
and d(bb0 ) − b(db0 ) − (db)b0 for b, b0 ∈ B.
Proof. The map d : B → Ω1B/A is simply b 7→ db + N , which is a derivation by definition of
N . Suppose e : B → M is an A-derivation. Then there is a map g : Ω1B/A → M obtained
P P
by setting g ( rb db + N ) = rb e(b). Note that g(N ) = 0 since e is a derivation, so g is
well-defined. Moreover, the diagram for the universal property of differentials commutes by
construction of g, and uniqueness is proven as usual.
341
15.2. Kähler Differentials Chapter 15. Morphisms of Schemes
which is natural in M .
Example 15.2.4. If B = A[x1 , . . . , xn ], then Ω1B/A is a free B-module generated by {dx1 , . . . , dxn }
and the derivation d : B → Ω1B/A is given by
n
X ∂f
df = dxi
i=1
∂x i
∂f
where ∂xi
denote the formal partial derivatives of the polynomial f .
Lemma 15.2.5. Suppose A is a commutative ring. Then
(a) If I ⊂ A is an ideal, then Ω1(A/I)/A = 0.
Proof. (a) For any a + I ∈ A/I, d(a + I) = ad(1 + I) since d is A-linear, and as we saw in
the proof of Lemma 15.2.1, d(1 + I) = 0.
(b) For any b ∈ S −1 A, there is some s ∈ S so that sb ∈ A. Then by Lemma 15.2.1,
0 = d(sb) = s(db) + (ds)b = s(db) since s ∈ S ⊂ A. But since s 6= 0, we must have
db = 0.
Now let ϕ : B → C be an A-algebra homomorphism. This induces a homomorphism of
C-modules
dϕ : Ω1B/A ⊗B C −→ Ω1C/A
db ⊗ c 7−→ cd(ϕ(b)).
On the other hand, by Lemma 15.2.3, the canonical derivation C → Ω1C/A induces a C-
module map βϕ : Ω1C/A → Ω1C/B which is just dc 7→ dc. (One can define βϕ directly this
way and then check it is well-defined and preserves the C-module structure, but there is a
universal property for a reason.)
Proposition 15.2.6. Let B be an A-algebra. Then
(a) For any other A-algebra A0 , set B 0 = B ⊗A A0 . Then there is a natural isomorphism
of B 0 -modules Ω1B 0 /A0 ∼
= Ω1B/A ⊗B B 0 .
342
15.2. Kähler Differentials Chapter 15. Morphisms of Schemes
(d) Suppose C = B/I for an ideal I ⊂ B. Then there is an exact sequence of C-modules
δ dϕ
I/I 2 →
− Ω1B/A ⊗B C −→ Ω1C/A → 0
The map here sends dbs ∈ S −1 Ω1B/A to 1s ·d 1b which is clearly injective, hence an isomorphism.
(d) Again by Yoneda’s Lemma and Lemma 15.2.3, it’s enough to show
343
15.2. Kähler Differentials Chapter 15. Morphisms of Schemes
(b) If S = R[x1 , . . . , xn ]/I for an ideal I, let ρ : R → S be the restriction of the quotient
map to R ⊆ R[x1 , . . . , xn ] and let α = dρ : Ω1R/A ⊗R S → Ω1S/A and δ : I/I 2 →
Ω1R[x1 ,...,xn ]/A ⊗R[x1 ,...,xn ] S be the maps defined above. Then there is a surjective B-
module homomorphism ker δ → ker α.
= Ω1S/A ⊗S B and Ω1B/S ∼
Proof. (a) From Proposition 15.2.6(a), Ω1B/R ∼ = Ω1R/A ⊗R B, so (b) of
the same proposition gives us exact sequences
ϕ
Ω1R/A ⊗R B −
→ Ω1B/A → Ω1S/A ⊗S B (15.1)
ψ
and Ω1S/A ⊗S B −
→ Ω1B/A → Ω1R/A ⊗R B. (15.2)
Then the desired map is ϕ ⊕ ψ which is of the desired form by construction and is an
isomorphism because the last maps in (1) and (2) are sections of ψ and ϕ, respectively.
(b) Set A0 = A[x1 , . . . , xn ] and R0 = R[x1 , . . . , xn ]. Then the fact that R0 = A0 ⊗A R
implies by (a) that
Ω1R0 /A ∼
= (Ω1A0 /A ⊗A0 R0 ) ⊕ (Ω1R/A ⊗R R0 ) = Ω1R0 /R ⊕ (Ω1R/A ⊗R R0 ).
Ω1R0 /A ⊗R0 S ∼
= (Ω1R0 /R ⊗R0 S) ⊕ (Ω1R/A ⊗R S) (15.3)
344
15.2. Kähler Differentials Chapter 15. Morphisms of Schemes
δ0
I/I 2 Ω1R0 /A ⊗R0 S Ω1S/A 0
δ ∼
= α
p i
0 Ω1R0 /R ⊗R0 S (Ω1R0 /R ⊗R0 S) ⊕ (Ω1R/A ⊗R S) Ω1R/A ⊗R S 0
where the top row is the exact sequence from Proposition 15.2.6(d) and the bottom row is
the canonical split sequence for the direct sum. A quick diagram chase reveals there is a
surjection ker δ → ker α.
We next study relative differentials for field extensions. Let k be a field.
Lemma 15.2.10. Let E/k be an arbitrary field extension and K = E[x]/(p(x)) a simple
algebraic extension of E. Then
(a) If K/E is separable, then Ω1K/E = 0 and there is an isomorphism of K-vector spaces
Ω1E/k ⊗E K ∼
= Ω1K/k .
so if the extension is separable, p0 (x) is relatively prime to p(x). Thus K/(p0 (x)) = 0. The
isomorphism Ω1E/k ⊗E K ∼ = Ω1K/k then follows from Proposition 15.2.6(d).
(b) When K/E is inseparable, p0 (x) = 0 so Ω1K/E = Khdxi/(p0 (x) dx) ∼ = K. From the
isomorphism (3) in the proof of Lemma 15.2.9, we obtain
345
15.2. Kähler Differentials Chapter 15. Morphisms of Schemes
from Proposition 15.2.6(b) implies Ω1K/E = 0 since the first arrow is an isomorphism. By
Lemma 15.2.10, K/E must be separable and hence so is K/k.
346
15.3. Sheaves of Relative Differentials Chapter 15. Morphisms of Schemes
d : B −→ I/I 2
b 7−→ 1 ⊗ b − b ⊗ 1
d
g
2
I/I
347
15.3. Sheaves of Relative Differentials Chapter 15. Morphisms of Schemes
commute. Uniqueness follows from a similar proof using the explicit description of I/I 2 .
Hence I/I 2 is universal with respect to A-derivations, so it is isomorphic to Ω1B/A .
Now let f : X → Y be a morphism of schemes. Then the diagonal morphism ∆ : X →
X ×Y X maps X to a closed subscheme of an open set U ⊆ X ×Y X. Let I be the ideal
sheaf of ∆(X) in OU .
Ω1X/Y := ∆∗ (I/I 2 ).
Ω1X/Y |U ∼
=Ωe1
OX (U )/OY (V )
Ω1X/Y,x ∼
= Ω1OX,x /OY,f (x) .
Example 15.3.6. Let A be a ring, put Y = Spec A and let X = AnY be affine n-space over
n
A. Then by Example 15.2.4, we have Ω1X/A = OX .
Example 15.3.7. With Y = Spec A again, let X = PnY be projective n-space over A. Let us
compute Ω1X/Y carefully. Recall that X has an affine cover {U0 , U1 } where U0 = Spec A[t] and
U1 = Spec A[t−1 ]. Then Ω1U0 /A is the sheafification of the A-module A[t]dt; likewise, Ω1U1 /A
is the sheafification of the A-module A[t−1 ]d 1t = A[t−1 ] −1 dt. We claim that Ω1X/Y ∼
t2 =
O(−2), the twisting sheaf of degree −2 – recall that this is defined as O(−2) = M f where
M = A[t](−2) is the −2 shift of the graded ring A[t]. On the open sets U0 and U1 , we have
isomorphisms
ϕ0 : Ω1U0 /A −→ M
f|U0
dt 7−→ 1
and ϕ1 : Ω1U1 /A −→ M f|U1
1 1
d 7−→ − 2
t t
So Ω1X/A is a line bundle on X (a locally free sheaf of rank 1). Moreover, on the overlap
U0 ∩ U1 = Spec A[t, t−1 ], we have d 1t = − t12 dt, so we see that ϕ0 and ϕ1 determine a global
348
15.3. Sheaves of Relative Differentials Chapter 15. Morphisms of Schemes
Example 15.3.8. Let X be an affine curve defined over a field k by an equation F (x, y) = 0
for F ∈ k[x, y] such that at any point on X, the formal partial derivatives ∂F ∂x
and ∂F
∂y
do not simultaneously vanish. Such a curve is said to be smooth in the classical sense,
though we will soon define smoothness for general schemes and recover this very definition.
Since X = Spec(k[x, y]/(F )) is affine, Ω1X/k is the sheafification of the module Ω1B/k where
B = k[x, y]/(F ), and Example 15.2.8 shows that
so Ω1X/k is isomorphic to OX = B.
e That is, Ω1 is a free OX -module of rank 1 with basis
X/k
n o n o
1 1
∂F/∂x
dy or ∂F/∂y dx .
(c) For any open set U ⊆ X, there is an isomorphism Ω1X/Y |U ∼ = Ω1U/Y of sheaves on U .
In particular, for any x ∈ X, Ω1X/Y,x ∼
= Ω1OX,x /OY,f (x) .
(d) Let Z be a closed subscheme of X with ideal sheaf I. Then there is an exact sequence
of sheaves on Z,
I/I 2 → Ω1X/Y ⊗OX OZ → Ω1Z/Y → 0.
Example 15.3.10. Let E be an elliptic curve over a field k, i.e. a nonsingular projective
curve in P2k with distinguished point O ∈ E(k) and affine Weierstrass equation
y 2 + a1 xy + a3 y = x3 + a2 x2 + a4 x + a6 for ai ∈ k.
349
15.4. Smooth Morphisms Chapter 15. Morphisms of Schemes
Note that a locally noetherian scheme is regular if and only if it is regular at all of its
closed points.
Example 15.4.1. Let X = Ank = Spec k[x1 , . . . , xn ] be affine n-space over a field k. Since X
is an affine variety, for any closed point x ∈ X we have dim OX,x = dim X = n (Lemma 2.6.5,
Corollary 2.6.3 and Proposition 14.1.5), and if x = (α1 , . . . , αn ) then the maximal ideal in
OX,x is mx = (x1 − α1 , . . . , xn − αn ). Thus mx is generated by n elements, so OX,x is regular.
Example 15.4.2. Likewise, Pnk is regular at all of its (closed) points since each one is
contained in an affine open subscheme isomorphic to Ank .
Example 15.4.3. Let char k 6= 2, 3 and consider the affine curve X ⊆ A2k defined by
P = (0, 0)
Then B ∼ = k[t3 , t2 ] and it’s easy to compute the dimensions of tangent spaces by localization
and see that X is regular at all P 6= (0, 0) and singular at P = (0, 0), but here’s a way of
seeing it using the results of the preceding sections. The local ring at any P = (x0 , y0 ) is
∗
OX,P = BmP where mP = (x − x0 , y − y0 ). Then TX,P = mP /m2P has dimension equal to
350
15.4. Smooth Morphisms Chapter 15. Morphisms of Schemes
The last example is also easy to verify using the Jacobian condition for regularity, which
we recall now. Let X ⊆ Ank be an (irreducible) affine variety with vanishing ideal I ⊆
k[x1 , . . . , xn ]. Then I = (f1 , . . . , fm ) for some polynomials fj ∈ k[x1 , . . . , xn ]. For each point
x ∈ X, the Jacobian of X at x is the m × n matrix
∂fi
Jx = (x) ∈ Mm×n (k).
∂xj
Corollary 15.4.5. The rank of the Jacobian matrix of X at x is independent of the choice
of generators f1 , . . . , fm for the vanishing ideal of X.
(ii) f is flat at x.
Denote by Xsm the set of smooth points of a Y -scheme X → Y and by Xsing the set of
singular points of X. Note that in general, the fibres of a morphism need not have the same
dimension.
Remark. To define smoothness for an arbitrary morphism of schemes, one replaces condition
(i) with “f is locally of finite presentation at x”. When Y is locally noetherian, this condition
is equivalent to being locally of finite type. We will assume for the rest of the section that
Y is always locally noetherian.
351
15.4. Smooth Morphisms Chapter 15. Morphisms of Schemes
Proof. Take x ∈ X and set y = f (x) ∈ Y , m = dim OX,x and n = dim OY,y . Since f is flat,
Theorem 8.3.30 says that
Proof. These are mostly straightforward from the definitions. For (a), note that finite gener-
ation and flatness for rings are preserved under tensor product (the latter by Lemma 8.3.21).
Moreover, the fibre of a base change is precisely the original fibre, so all parts of the definition
of smoothness hold for X 0 → Y 0 .
For (b), finite type is a property preserved in towers of rings. Moreover, flatness is
preserved under composition by Lemma 15.1.2. To prove the regularity condition for g ◦ f ,
set y = f (x) and z = g(y). Then
Example 15.4.8. For a scheme Y , one can define affine n-space over Y to be the affine
scheme
AnY := Spec(OY (Y )[x1 , . . . , xn ]).
Then the ring map OY (Y ) → OY (Y )[x1 , . . . , xn ] induces a natural projection morphism
π : AnY → Y . We claim π is always smooth of relative dimension n. By Hilbert’s Basis
Theorem, π is of finite type. Flatness is trivial. Finally, each fibre of π is Ank(y) where k(y)
is the residue field at y ∈ Y , and these are regular by Example 15.4.1 and have dimension
n. Thus π is smooth of relative dimension n.
The next result says that a smooth Y -scheme is étale-locally a subscheme in AnY . The
notion of ‘étale-locally’ can be made precise using Grothendieck topologies (namely, the étale
topology on Y ), which will be introduced in Section 33.1.
352
15.4. Smooth Morphisms Chapter 15. Morphisms of Schemes
OY [x1 , . . . , xn ] −→ f∗ OU
353
15.4. Smooth Morphisms Chapter 15. Morphisms of Schemes
If (b) holds for the map g, we could extend this with a zero on the left and, since rank
is additive along short exact sequences, this would imply rank Ω1X/Y = rank g ∗ Ω1An /Y +
Y
rank Ω1X/An . Further, Lemma 15.4.11(a) below shows that rank Ω1An /Y = n and, since g ∗ is
Y Y
exact (Corollary 16.1.13), rank g ∗ Ω1An /Y = n as well. Then if (a) holds for g, rank Ω1X/An = 0
Y Y
so we would get rank Ω1X/Y = n. Moreover, consider the commutative diagram
f
The top row is our desired sequence for X →− Y → Z. The bottom row comes from applying
g ∗ to the sequence
0 → π ∗ Ω1Y /Z → Ω1AnY /Y → Ω1AnY /Y → 0.
We claim that it is enough to prove that (a) and (b) hold for π and g. Indeed, if they hold
for g then the short exact sequences of relative differentials associated to the compositions
g g
X → − AnY → Z and X → − AnY → Y become isomorphisms g ∗ Ω1An /Z ∼ = Ω1X/Z and g ∗ Ω1AnY /Y ∼
=
Y
1
ΩX/Y so the middle and right columns of the diagram are isomorphisms. Then (b) for π
along with exactness of g ∗ imply the bottom row is exact, so we can deduce exactness of
the top row. To summarize, the Corollary follows from establishing (a) and (b) for g and π.
This is done in the following lemma.
Lemma 15.4.11. Let Y → Z be a morphism of locally noetherian schemes. Then
(a) For every n ≥ 1, let π : AnY → Y be the canonical projection. Then Ω1An /Y is a free
Y
OY -module of rank n and there is a short exact sequence
0 → π ∗ Ω1Y /Z → Ω1AnY /Z → Ω1AnY /Y → 0.
g×g
g
∆Y
Y Y ×Z Y
354
15.4. Smooth Morphisms Chapter 15. Morphisms of Schemes
where ∆X and ∆Y are the diagonal maps. Let I be the OY ×Z Y -ideal corresponding to
the image of ∆Y , so that Ω1Y /Z = ∆∗Y (I/I 2 ). Then since g and therefore g × g is flat,
J := (g × g)∗ I is the OX×Z X -ideal corresponding to the image of q. We will see that since
g is étale, ∆X is an open immersion. Assuming this, Ω1X/Z ∼
= (q ◦ ∆X )∗ (J /J 2 ) and we have
Ω1X/Z ∼
= (q ◦ ∆X )∗ (J /J 2 ) = ∆∗X ◦ q ∗ (J /J 2 )
= ∆∗X ◦ q ∗ ◦ (g × g)∗ (I/I 2 )
∼
= g ∗ ◦ ∆∗ (I/I 2 ) by commutativity
Y
∗ 1
=g ΩY /Z by definition.
Therefore g ∗ Ω1Y /Z ∼
= Ω1X/Z .
In fact, the converse of (a) is also true. We omit the proof.
Theorem 15.4.12. Let f : X → Y be a flat morphism of finite type with fibres of dimension
n. Then f is smooth if and only if Ω1X/Y is locally free of rank n.
There are two more important properties of smoothness to discuss. The first says that
every morphism is generically smooth, generalizing Theorem 12.6.5.
Theorem 15.4.13 (Generic Smoothness). For a morphism of finite type f : X → Y between
locally noetherian schemes, the smooth locus Xsm is a nonempty open subset of X.
g π
Proof. By Proposition 15.4.9, f |Xsm factors locally as Xsm → − AnY →− Y for some étale g.
n
Since the smooth locus of π is all of AY (Example 15.4.8), it’s enough to prove the theorem
for étale morphisms. This will be done in the next section.
Let Y = Spec A be an affine scheme and Y0 ⊆ Y a closed subscheme defined by an ideal
I ⊂ A. Then Y0 is called an infinitesimal subscheme of Y if I 2 = 0 (or equivalently, if I is
nilpotent).
Theorem 15.4.14 (Infinitesimal Lifting). Let f : X → S be a morphism of finite type over
a locally noetherian scheme S. Then f is smooth if and only if for every affine S-scheme Y
and every infinitesimal subscheme Y0 ⊆ Y ,
is surjective.
The second condition in the theorem is called formal smoothness. Informally, this result
says that smoothness is detected by the property of being able to lift tangent vectors. Another
way to view formal smoothness is as an abstraction of Hensel’s Lemma. In the case of an
arbitrary scheme, one has:
Theorem 15.4.15. A morphism of schemes f : X → S is smooth if and only if f is formally
smooth and locally of finite presentation.
One can use the infinitesimal lifting criterion to prove the following generalization of the
Jacobian condition.
355
15.4. Smooth Morphisms Chapter 15. Morphisms of Schemes
356
15.5. Unramified Morphisms Chapter 15. Morphisms of Schemes
(b) Ω1X/Y,x = 0.
Bq ⊗Ap k(p) ∼
= Bq /pBq = k(q).
357
15.5. Unramified Morphisms Chapter 15. Morphisms of Schemes
but this vanishes by Corollary 15.2.11. Further, since f is of finite type (locally, but we may
assume globally on Spec B → Spec A), Corollary 15.2.7 says Ω1Bq /Ap is a finitely generated
Bq -module so Nakayama’s lemma together with the above calculation imply Ω1Bq /Ap = 0.
(b) =⇒ (c) We may again assume X = Spec B, Y = Spec A and f is of finite type. In
the affine case, ∆ = ∆X is a closed immersion, say with ideal sheaf I. Then
0 = Ω1X/Y,x ∼
= (I/I 2 )∆(x) = I∆(x) /I∆(x)
2
.
It is routine to check that I∆(x) is a coherent sheaf, i.e. corresponds to a finitely generated
B ⊗A B-module via Theorem 14.4.7. Then by Nakayama’s lemma, I∆(x) = 0 and since it is
coherent, I then vanishes in some neighborhood V ⊆ X ×Y X of ∆(x). Thus ∆ restricts to
an isomorphism on U = ∆−1 (V ) ⊆ X.
(c) =⇒ (a) Since the question is again local, we may assume U = X = Spec B,
Y = Spec A and ∆ = ∆X is an open immersion. Consider the pullback diagram
Spec(Bp /pBp ) Spec B
Let q̄ be the image of qBp in Bp /pBp . The localization of Bp /pBp at q̄ is Bq /pBq and the
residue field of this local ring is (Bq /pBq )/q̄ ∼
= k(q). To prove f is unramified at x = q,
we must show qBq /pBq = 0 and k(q)/k(p) is a finite, separable extension. Since statement
(c) is preserved under base change, we may in fact assume Y = Spec k for a field k. Then
Lemma 15.5.1 shows that we can assume k is algebraically closed. Let π1 , π2 : X ×Y X → X
be the two coordinate projections and consider the morphism
h : X −→ X ×Y X
r 7−→ (r, p).
358
15.5. Unramified Morphisms Chapter 15. Morphisms of Schemes
(b) The Kähler differential condition in (b) says that at every point in the fibre over x,
there are the same number of tangent directions as at x.
(c) Condition (c) says that ramification points are precisely the ‘limit points’ of ∆X (X)
which are not in the image of the diagonal.
Proof. This follows from Theorem 15.5.2(b) since the support of a coherent sheaf on a locally
noetherian scheme is closed.
The unramified condition is preserved under composition and base change, as we saw for
flat and smooth morphisms in Lemmas 8.3.21/15.1.2 and Proposition 15.4.7, respectively.
Example 15.5.6. Let Y be the cuspidal curve from Example 15.4.3 defined by y 2 − x3 = 0
over a field k of characteristic different from 2, 3. There is a normalization map f : A1k → Y
corresponding to the ring map A → B where A = k[t2 , t3 ] → k[t]. Then Ω1A1 /Y is the
sheafification of Ω1B/A , but notice that B = A[s]/(s2 − t2 ) so by Example 15.2.8,
Ω1B/A = B dx/2B dx ∼
= k[t]/(2t) ∼
= k[t]/(t).
This shows that the support of Ω1A1 /Y is the point corresponding to (t), so f is unramified over
A1 r {0}. This illustrates the ‘number of tangent directions’ interpretation of ramification
in the remark above.
Example 15.5.7. Let us show that the definition of ramification in this section agrees
with the notion in algebraic number theory. Let B/A be an integral extension of Dedekind
domains. Then for any nonzero prime p ⊂ A, the extension pB factors as a product of prime
ideals g
Y
pB = Pegg
i=1
359
15.5. Unramified Morphisms Chapter 15. Morphisms of Schemes
where each Pi is a prime of B lying over p and ei ≥ 1. In algebraic number theory, one says
that p is ramified if ei > 1 for some i and unramified if e1 = · · · = eg = 1. Fix 1 ≤ j ≤ g
and consider the morphism of affine schemes f : Spec B → Spec A. Then
g
Y
pBPj = (pB)Pj = (Pi BPj )ei = (Pj BPj )ej .
i=1
where Ω1B/A is the module of Kähler differentials of B/A. Note that this reinforces the
connection to ramification in number theory, since unramified is equivalent to Ω1 vanishing
after localization (Theorem 15.5.2).
360
15.6. Étale Morphisms Chapter 15. Morphisms of Schemes
361
15.6. Étale Morphisms Chapter 15. Morphisms of Schemes
g f
Proposition 15.6.5. Suppose a composition X →
− Y →
− Z is étale and f is unramified.
Then g is also étale.
Γg ∆
X ×Z Y Y ×Z Y
g×1
where ∆ = ∆Y is the diagonal morphism. Since f is unramified, Theorem 15.5.2 says that
∆ is an open immersion, hence étale by Corollary 15.6.2. Then Proposition 15.6.4(a) implies
Γg is étale, too.
On the other hand, p2 is also a pullback:
p1
X ×Z Y X
p2 f ◦g
g
Y Z
Example 15.6.7. Let L/k be a finite extension of fields. The morphism f : Spec L → Spec k
is always flat, so f is étale if and only if it is unramified. On the other hand, Theorem 15.5.2
and Corollary 15.2.11 say that f is unramified if and only if L/k is separable. Note that L
is a trivial example of a finite étale k-algebra. For any k-algebra A, the structure morphism
f : Spec A → Spec k is étale if and only if A is a finite étale k-algebra. If A is of the
form A = k[t]/(p(t)), where p(t) is a monic polynomial over k, then the étale condition is
equivalent to p(t) being separable for the same reason as the case of L/k.
Example 15.6.8. Let p(t) be a monic polynomial over an arbitrary commutative ring A
and consider B = A[t]/(p(t)). Then B is a free A-module of rank n = deg p and for any
prime p ∈ Spec B,
B ⊗A k(p) ∼
= k(p)[t]/(p̄(t))
362
15.6. Étale Morphisms Chapter 15. Morphisms of Schemes
where p̄(t) is the image of p(t) in k(p)[t]. Then Corollary 15.5.4 shows that B is an étale
A-algebra (or equivalently Spec B → Spec A is étale) if and only if (p(t), p0 (t)) = 1, where p0
is the formal derivative of p. The same statement holds when we replace B by a localization
Bb at a nonzero element b ∈ B; namely, Bb is étale over A if and only if p0 (t) is a unit in
Bb [t].
Definition. Let B = A[t]/(p(t)) for a monic polynomial p(t) ∈ A[t] and let b ∈ B be nonzero
such that p0 (t) is a unit in Bb [t]. Then Bb is called a standard étale A-algebra.
In a moment we will prove that every étale morphism is locally standard étale, meaning
isomorphic to Spec Bb → Spec A for a standard étale A-algebra Bb . We will need Zariski’s
Main Theorem, which we state but do not prove below.
Definition. A morphism f : X → Y is called quasi-finite if f −1 (y) is discrete for all
y ∈Y.
Theorem 15.6.9 (Zariski’s Main Theorem). Let Y be a noetherian scheme and f : X → Y
a separated, quasi-finite morphism. Then f factors as
j g
− X0 →
f :X→ − Y
Set n = [k(q) : k(p)]. Then {1, γ̄, . . . , γ̄ n−1 } is a basis for the field extension k(q)/k(p),
so Nakayama’s lemma (version 1.2.4) implies {1, γ, . . . , γ n−1 } generates C as an A-algebra.
Thus there is a monic polynomial p(t) ∈ A[t] such that
B := A[t]/(p(t)) −→ A[γ] = C
is surjective. Note that the image p̄(t) of p(t) in k(q)[t] is exactly the characteristic polynomial
of γ̄ over k(p), so p̄(t) is separable. Therefore for some b ∈ B not landing in q ⊂ C,
363
15.6. Étale Morphisms Chapter 15. Morphisms of Schemes
Example 15.6.8 shows that Bb is a standard étale algebra over A. For the corresponding
choice of c ∈ C, we get a surjection
h : Bb −→ Cc .
where r and h∗ ◦ r are both étale. Therefore Proposition 15.6.5 shows that h∗ is also étale.
But since h is surjective, h∗ is an étale closed immersion and hence an open immersion by
Proposition 15.6.6. This proves h is an isomorphism.
364
15.7. Henselian Rings Chapter 15. Morphisms of Schemes
365
15.7. Henselian Rings Chapter 15. Morphisms of Schemes
Theorem 15.7.3. For a local ring (A, m, k), the following are equivalent:
(1) A is henselian.
(2) Every pointed étale neighborhood (A, m) → (B, n) is an isomorphism of local rings.
does not lie in m, then there exist elements b1 , . . . , bn ∈ A with bi ≡ ai mod m and
Fi (b1 , . . . , bn ) = 0 for all 1 ≤ i ≤ n.
(4) If F (x) ∈ A[x] is monic and its reduction mod m factors as F = gh for some relatively
prime (necessarily monic) polynomials g, h ∈ k[x], then F = GH for some monic
polynomials G, H ∈ A[x] with G = g and H = h.
Proof. (1) =⇒ (2) Write B = (A[x]/(f (x)))q as above, with q = (x − α, mB) for α ∈ A
such that f (α) ∈ m but f 0 (α) 6∈ m. By assumption, there exists a lift a ∈ A such that a ≡ α
mod m and f (a) = 0. Then f (x) = (x − a)g(x) for some g ∈ A[x]; since f 0 (a) 6= 0, g is not
divisible by x − a. Thus g 6∈ q, so it follows that q = (x − a, mB). Localizing at q then gives
366
15.7. Henselian Rings Chapter 15. Morphisms of Schemes
367
15.7. Henselian Rings Chapter 15. Morphisms of Schemes
so J(α, β) = 0 precisely when g and h are not relatively prime. Hence (3) guarantees the
existence of such a solution and consequently a factorization F = GH lifting the equation
f = gh to A[x].
(4) =⇒ (5) Notice that condition (4) implies that every A-algebra of the form A[x]/(f (x)),
for f (x) monic, is decomposable. Given a finite ring extension A → B, write B = B ⊗A k =
B 1 × · · · × B ` using Proposition 15.7.1(3). By Lemma 15.7.2, it’s enough to lift the orthog-
onal idempotents ē1 , . . . , ē` ∈ B; since the argument is identical for each one, we will just
construct the lift for ē = ē1 . For any lift b ∈ B of ē, let f (x) ∈ A[x] be the minimal monic
polynomial having b as a root. Set R = A[x]/(f (x)) and consider the diagram
A A[b]
ϕ
R B
368
15.7. Henselian Rings Chapter 15. Morphisms of Schemes
We will show that every local ring has a henselization. To construct Ah , we need the
following facts about pointed étale neighborhoods.
Proposition 15.7.5. Suppose B and C are pointed étale neighborhoods of a local ring
(A, m, k). Then
(1) If Q is the kernel of B ⊗A C → k ⊗A k = k, then Q is a prime ideal and (B ⊗A C)Q
is a pointed étale neighborhood of A.
(2) There is at most one local homomorphism B → C and if one exists, C is a pointed
étale neighborhood of B.
(3) If there exist local homomorphisms B → C and C → B, then they are both isomor-
phisms.
Proof. (1) By Theorem 15.6.10, we may write
for monic polynomials f (x) and g(y) and prime ideals q1 and q2 . Then B⊗A C is a localization
of A[x, y]/(f, g) and f 0 (x), g 0 (y) 6∈ Q, so (B ⊗A C)Q is a localization of a finite étale A-algebra.
Moreover, k(Q) = k, so (B ⊗A C)Q is a pointed étale neighborhood.
(2) If there exists a map α : B → C then Proposition 15.6.5 implies α is also étale.
Since the residue fields of A, B and C are all k, this implies B → C is a pointed étale
neighborhood so the second statement holds. For uniqueness, suppose β : B → C is another
homomorphism. Then the map B ⊗A B → C given by b ⊗ b0 7→ α(b)β(b0 ) takes Q from
(1) to the maximal ideal of C and hence induces a map (B ⊗A B)Q → C, with b⊗1 1
7→ α(b)
and 1⊗b 1
→
7 β(b). On the other hand, the natural map B ⊗ A B → B takes Q to the
maximal ideal of B, so induces a map (B ⊗A B)Q → B. If I is the kernel of B ⊗A B → B, by
Proposition 15.3.2 we know I/I 2 ∼ 1
= ωB/A = 0 since A → B is étale. Therefore IQ ⊂ (B⊗A B)Q
is finitely generated, lies in the maximal ideal and satisfies IQ2 = IQ so it is 0 by Nakayama’s
lemma (Theorem 1.2.1). That is, (B ⊗A B)Q → B is an isomorphism, but this means that
under (B ⊗A B)Q → C, both b⊗1 1
and 1⊗b
1
map to the same element, so α(b) = β(b) and we
conclude that α = β.
(3) It follows from (2) that the only local homomorphism B → B is the identity, so any
compositions B → C → B and C → B → C must be the identity.
Corollary 15.7.6. The isomorphism classes of pointed étale neighborhoods of A form a
directed system.
Theorem 15.7.7. Let (A, m, k) be a local ring. Then
(1) There exists a henselization of A, namely Ah = lim B where the limit is taken over
−→
some choice of representatives of the isomorphism classes of pointed étale neighbor-
hoods A → B.
369
15.7. Henselian Rings Chapter 15. Morphisms of Schemes
370
15.7. Henselian Rings Chapter 15. Morphisms of Schemes
Corollary 15.7.11 (Artin Approximation). Let (A, m, k) be a henselian local ring, fix N ≥ 1
and suppose there are polynomials F1 , . . . , Fn ∈ A[x] and elements â1 , . . . , ân ∈ Ab such
that Fj (â1 , . . . , ân ) = 0 for all 1 ≤ j ≤ n. Then there exist a1 , . . . , an ∈ A such that
Fj (a1 , . . . , an ) = 0 for all 1 ≤ j ≤ n and ai ≡ âi mod mN A
b for all 1 ≤ i ≤ n.
Proof. Let mN = (u1 , . . . , ur ) for elements u` ∈ A. For each i, choose any bi ∈ A such that
bi ≡ âi mod mN A
b and write
n
X
bi − âi = âi` u`
`=1
F1 (x1 , . . . , xn ) = 0
..
.
Fn (x1 , . . . , xn ) = 0
X n
x 1 − b1 = y1` u`
`=1
..
.
n
X
x n − bn = yn` u` .
`=1
371
15.7. Henselian Rings Chapter 15. Morphisms of Schemes
Theorem 15.7.12. If (A, m, k) is a local noetherian ring which is excellent and normal,
then
Ah = Amb ∩A
where A is the integral closure of A in its completion A.
b
Proof. By Theorem 15.7.7 and Proposition 15.7.8, Ah and B := Amb ∩A are henselian and Ah
is the smallest henselian extension of A, so it suffices to show B ⊆ Ah . In fact, we just need
to show A ⊆ Ah since then its localization at m b ∩ A will also be contained in Ah . Take
a∈A b integral over A. Then since A is normal and excellent, there is a monic polynomial
f (x) ∈ A[x] with f (a) = 0 and fintely many roots in A b which can be distinguished mod
m A for any sufficiently large N . Applying Artin approximation to Ah , along with the
N b
(2) Ash is universal with respect to such local homomorphisms: if f : A → B is any local
homomorphism to a strictly henselian ring B, then there is a unique map g : Ash → B
such that f = g ◦ j.
Similar to the properties of Ah , one can prove:
Theorem 15.7.14. Let (A, m, k) be a local ring. Then
(1) There exists a strict henselization Ash , namely Ash = lim B where the limit is over all
−→
local rings (B, mB ) admitting local étale maps A → B such that B/mB ∼ = k sep .
(2) If msh is the maximal ideal of Ash , then mAsh = msh and Ash /msh ∼
= k sep .
(3) The natural maps A → Ash and Ah → Ash are étale.
372
Chapter 16
Cohomology
373
16.1. Direct and Inverse Image Chapter 16. Cohomology
374
16.1. Direct and Inverse Image Chapter 16. Cohomology
Proof. Set Vj0 ,...,jp = Vj0 ∩ · · · ∩ Vjp and Uj0 ,...,jp = f −1 (Vj0 ,...,jp ). Then (f∗ F )(Vj0 ,...,jp ) =
F (Uj0 ,...,jp ) so it follows that C p (U, F ) = C p (V, f∗ F ) for all p ≥ 0. Hence the cohomologies
of these complexes are the same.
(f −1 F )(U ) = lim F (V )
−→
where the direct limit is taken over all open sets V ⊆ Y containing f (U ). The inverse
image (or pullback) of F along f , denoted f ∗ F , is the sheafification of f −1 F .
375
16.1. Direct and Inverse Image Chapter 16. Cohomology
Remark. The inverse image sheaf may also be defined topologically on the level of étale
spaces as follows. Let EF → Y be the étale space of F . Then f ∗ F is the sheaf of sections of
the map E = Ef ∗ F → X whose total space is the pullback in the diagram
E X
EF Y
λU : (f −1 F )(U ) −→ Fy .
376
16.1. Direct and Inverse Image Chapter 16. Cohomology
It’s easy to check that Φ is continuous with respect to the topology on the étale spaces.
Moreover, the diagram
Ef −1 F EF
f
X Y
377
16.1. Direct and Inverse Image Chapter 16. Cohomology
over all open sets V 0 ⊆ Y containing f (f −1 (V )). But since V itself is an open set containing
f (f −1 (V )), we get a map G(V ) → lim G(V 0 ) which is compatible with restriction. This
−→
defines a natural transformation
η : idShY −→ f∗ f −1
which will be the unit of our adjunction. Next, we construct a counit of the adjunction. For
any open U ⊆ X, we have
(f −1 f∗ F )(U ) = lim(f∗ F )(V ) = lim F (f −1 (V ))
−→ −→
where the limit is over all open V ⊆ Y containing f (U ). Well for every such V , f −1 (V ) ⊇ U
so the restriction maps F (f −1 (V )) → F (U ) assemble to give a map lim F (f −1 (V )) → F (U ).
−→
This defines the counit
ε : f −1 f∗ −→ idPreshX .
Finally, by Lemma 9.2.6 it remains to show that η and ε satisfy the triangle identities. For
an open set V ⊆ Y , we have
f∗ εF ◦ ηf∗ F (V ) : f∗ F (V ) → f∗ f −1 f∗ F (V ) → f∗ F (V )
which is F (f −1 (V )) → lim F (f −1 (V 0 )) → F (f −1 (V ))
−→
where the limit in the middle is over all open V 0 ⊆ Y containing f (f −1 (V )). Since V ⊇
f (f −1 (V )), this composition is clearly the identity on F (f −1 (V )). Therefore f∗ εF ◦ ηf∗ F is
the identity on f∗ F . Likewise, for U ⊆ X open,
εf −1 G ◦ f −1 ηG (U ) : f −1 G(U ) → f −1 f∗ f −1 G(U ) → f −1 G(U )
which is lim G(V ) → lim G(V 0 ) → lim G(V )
−→ −→ −→
where the outer limits are over all open V ⊆ Y containing f (U ) and the middle limit is
over all open V 0 ⊆ Y containing f (f −1 (V )) for any V containing f (U ). Any such V in
the first limit also appears in the second limit, so the composition is given by the trivial
restrictions G(V ) → G(V ) for all V . That is, the composition is the identity on lim G(V ) =
−→
f −1 G(U ).
378
16.1. Direct and Inverse Image Chapter 16. Cohomology
f g
Corollary 16.1.18. For any morphisms X → − Z, (g ◦ f )∗ = f ∗ ◦ g ∗ .
− Y →
∼
= ∼
=
Hom(f ∗ G, E) Hom(f ∗ F, E) 0
with isomorphisms on the columns. Since E is injective, the bottom row is exact. Therefore
the top row is also exact, which proves f∗ E is injective.
Remark. More generally, any functor with an exact left adjoint preserves injectives.
Since each f∗ E n is defined locally by a direct limit lim E n (f −1 (V )), there is an induced map
−→
of global sections Γ(X, E n ) → Γ(Y, f∗ E n ) for each n ≥ 0. This induces the desired maps on
cohomology: H p (X, F ) −→ H p (Y, f∗ F ).
379
16.1. Direct and Inverse Image Chapter 16. Cohomology
H p (Y, G) −→ H p (X, f ∗ G)
for all p ≥ 0. In principle these can be defined directly from the definitions of sheaf coho-
mology and f ∗ , but the derived categorical perspective makes this straightforward.
One thing to note about the direct image functor is that it does not preserve stalks.
Indeed, if f : X → Y is an arbitrary morphism and y ∈ Y , then for any sheaf F on X,
(f∗ F )y = Fx if y = f (x) ∈ f (X), but if y 6∈ f (X), then (f∗ F )y need not be 0 (the value of F
over the empty set). One way to remedy this is to define a new type of pushforward functor
that respects stalks. We start by giving the definition for open immersions.
Proof. Let F be a sheaf on U and G a sheaf on X. First recall that the inverse image functor
j ∗ is simply restriction: j ∗ G = G|U . Then for the presheaf P defined above, we have
HomShX (j! F, G) ∼
= HomPreshX (P, G) by the universal property of sheafification
∼
= HomShU (F, G|U ) by definition of P
= HomShU (F, j ∗ G).
Proof. Exactness of j! can be checked on stalks, and the second statement is a consequence
of the remark following Corollary 16.1.19.
If j : U ,→ X is an open immersion, the above gives us a triple of functors (j! , j ∗ , j∗ ) in
which each consecutive pair is an adjunction and j ∗ is exact. In a similar fashion, we can
extend the adjunction (i∗ , i∗ ) for a closed embedding i : Z ,→ X.
380
16.1. Direct and Inverse Image Chapter 16. Cohomology
ξ i∗ j∗ (χ)
G0 0
i∗ j∗ H 0
ϕ
381
16.1. Direct and Inverse Image Chapter 16. Cohomology
(1) W is open in W .
Proposition 16.1.28. The adjunctions (i! , i∗ , i∗ , i! ) exist for any locally closed immersion
i : W ,→ X.
0 → j! j ∗ F → F → i∗ i∗ F → 0
is exact.
This explains another common name, ‘sections with compact support’, for j! in the topolog-
ical setting. For a morphism of schemes f : X → Y , this suggests defining the pushforward
with proper support f! : ShZ → ShX by
(Recall that a morphism is proper if it is separated, of finite type and is a closed morphism
under all base changes. This is in contrast to the topological definition of properness: the
preimage of every compact set is compact.)
382
16.2. Acyclic Sheaves Chapter 16. Cohomology
HomOX (OU , F ) ∼
= HomOX |U (OX |U , j ∗ F ) = F (U ).
where the last term is zero since G is injective, and thus acyclic. Therefore exactness implies
H 1 (X, F ) = 0. Now for i ≥ 2, inductively assume that H i−1 (X, Q) = 0 for all flasque sheaves
Q on X. Part of the long exact sequence looks like
383
16.2. Acyclic Sheaves Chapter 16. Cohomology
H i (X, F ) = 0
0 → FU → F → FZ → 0.
By the long exact sequence in sheaf cohomology, it’s enough to prove H i (X, FU ) = 0 and
H i (X, FZ ) = 0 for all i > n. Note that FU extends to a sheaf on U , which is a closed
subset with one less irreducible component than X. Applying Proposition 16.2.1 to U ,→ X
and noting that Z is closed and irreducible, we may reduce to the case where X is itself
irreducible.
We induct on n = dim X. If n = 0, Γ : ShX → Ab is an equivalence of categories (X is
irreducible) and in particular exact, so Ri Γ is trivial for i > 0. For n > 0, note that we have
F = lim FB
−→
S
where the limit is over all finite subsets B ⊆ U ⊆X F (U ), directed by inclusion, and FB is
the subsheaf of F generated by the sections in B. Since cohomology commutes with direct
limits, it suffices to prove H i (X, FB ) = 0 for all B. Suppose B 0 ⊆ B and set q = |B r B 0 |.
Then there is an exact sequence
0 → FB 0 → FB → G → 0
where G is a sheaf generated by q sections over some open subsets of X. Using the long
exact sequence coming from this sequence, we may reduce to the case where q = 1. In this
case we have a short exact sequence
0 → K → ZU → F → 0
for some open set U ⊆ X, where Z denotes the (locally) constant sheaf on X with values
in Z. Thus we may prove the theorem by showing that H i (X, ZU ) = 0 for any open U and
384
16.2. Acyclic Sheaves Chapter 16. Cohomology
H i (X, K) = 0 for any nonzero subsheaf K ⊆ ZU . For such a subsheaf K, we have that Kx
is a subgroup of ZU,x for all x ∈ U , so there is some open set V ⊆ U for which K|V ∼
= dZ|V .
∼
Abstractly, dZV = ZV so there is a short exact sequence
0 → ZV → K → Q → 0.
The quotient sheaf Q is supported on U r V which has dimension < n since X is irreducible.
Therefore H i (X, Q) = 0 for all i ≥ n by the inductive hypothesis. Thus it remains to show
H i (X, ZU ) = 0 for i > n and all U (which implies vanishing for ZV if V ⊆ U ). Set Z = X rU .
Since X is irreducible, dim Z < n so by induction, H i (X, ZZ ) = 0 for i ≥ n. On the other
hand, Z is obviously flasque so Theorem 16.2.3 shows that H i (X, Z) = 0 for i > 0. Thus the
long exact sequence induced by the short exact sequence
0 → ZU → Z → ZZ → 0
gives us H i (X, ZU ) ∼
= H i−1 (X, ZZ ) which is 0 for all i − 1 ≥ n. This finishes the proof.
Suppose U = {Ui } is an open covering of a space X and F is a sheaf of abelian groups
p
on X. By Theorem 13.4.9, there is a natural map Ȟ (U, F ) → H p (X, F ) for every p ≥ 0.
p
Proposition 16.2.5. If F is a flasque sheaf on X, then Ȟ (U, F ) = 0 for all open covers
U of X and all p > 0.
Proof. Let C • (U, F ) be the Čech resolution of F with respect to U. Recall that C p (U, F ) is
defined for any open V ⊆ X by
Y
Γ(V, C p (U, F )) = F (V ∩ Ui0 ,...,ip )
i0 ,...,ip
where Ui0 ,...,ip = Ui0 ∩ · · · ∩ Uip . For each of these Ui0 ,...,ip , the sheaf V 7→ F (V ∩ Ui0 ,...,ip ) is
flasque and since products of flasque sheaves are flasque, the entire sheaf C p (U, F ) is flasque.
Therefore C p (U, F ) is acyclic by Theorem 16.2.3, so the Čech resolution can used to compute
sheaf cohomology:
p
Ȟ (U, F ) = H p (C • (U, F )) = H p (X, F ) = 0.
Definition. A sheaf F on X is acyclic for an open cover U = {Ui } if for all p > 0,
H p (Ui0 ,...,ip , F |Ui0 ,...,ip ) = 0.
Such covers exist in many common situations, e.g. for manifolds with a Riemannian
metric. The following result allows for relatively easy computations of sheaf cohomology
when an acyclic cover is available.
Theorem 16.2.6 (Leray). If F is a sheaf on X which is acyclic for an open cover U, then
the maps
p
Ȟ (U, F ) −→ H p (X, F )
are isomorphisms for all p ≥ 0.
385
16.2. Acyclic Sheaves Chapter 16. Cohomology
0
Proof. We prove this by induction on p. For p = 0, this is clear since both Ȟ (U, F ) and
j
H 0 (X, F ) are Γ(X, F ). Now assume Ȟ (U, G) → H j (X, G) is an isomorphism for all j ≤ p
and all acyclic sheaves G. For our given F , take a short exact sequence of sheaves
0→F →E→Q→0
where E is injective. Set U = Ui0 ,...,ip . Then by the hypothesis on F , H i (U, F |U ) = 0 for all
i > 0. Since E is injective, and thus flasque by Lemma 16.2.2, we have H i (U, E|U ) = 0 as
well. Thus in the long exact sequence induced by the above sequence, we can see
0 = H i (U, E|U ) → H i (U, Q) → H i+1 (U, F ) = 0.
This shows H i (U, Q) = 0 for all i ≥ 1, so by taking the product over all such Ui0 ,...,ip , we
conclude that Q is acyclic for U. Therefore, the sequence of complexes of sheaves
0 → C • (U, F ) → C • (U, E) → C • (U, Q) → 0
is exact and induces a long exact sequence
p−1 p−1 p p
· · · → Ȟ (U, E) → Ȟ (U, Q) → Ȟ (U, F ) → Ȟ (U, E) → · · ·
p−1 p
Now since E is flasque, Proposition 16.2.5 shows that Ȟ (U, E) = Ȟ (U, E) = 0. Therefore
we have a diagram with exact rows
p−1 p
0 Ȟ (U, Q) Ȟ (U, F ) 0
id id
386
16.2. Acyclic Sheaves Chapter 16. Cohomology
After passing to the direct limit over all U, it’s enough to show that
0
is an isomorphism. Injectivity is easy. For surjectivity, take s ∈ H 0Q (X, Q) = Ȟ (X, Q).
Then there exists an open cover U = {Ui } of X such that s = (si ) ∈ E(Ui ) and these si
satisfy si − sj ∈ ker(F (Ui ∩ Uj ) → E(Ui ∩ Uj )) = F (Ui ∩ Uj ) for all overlaps Ui ∩ Uj . By
0
definition this (si ) defines a Čech 0-cocycle in Ȟ (U, E) whose image in H 0 (K• (U)) maps to
s along α. Hence α is an isomorphism.
Definition. The support of a morphism of sheaves ϕ : F → G is the closed set
supp(ϕ) = {x ∈ X | ϕx 6= 0}
387
16.3. Noetherian Scheme Cohomology Chapter 16. Cohomology
Proof. We prove this using Baer’s criterion (Theorem 8.2.3); let J ⊂ A be an ideal and
ϕ : J → ΓI (E) a homomorphism. Since J is finitely generated, there is some N ∈ N such
that I N ϕ(J) = 0, or ϕ(I N J) = 0 since ϕ is a homomorphism. By Corollary 5.4.7, there is
some n ≥ N such that I N J ⊇ J ∩ I n , so ϕ(J ∩ I n ) = 0. Consider the diagram
A A/I n
ψ0
J J/(J ∩ I n ) ΓI (E) E
ϕ
Proof. Consider the A-module M = Γ(X, F ) and take an injective resolution E • of M . Then
0→M e0 → E
f→E e1 → · · ·
388
16.3. Noetherian Scheme Cohomology Chapter 16. Cohomology
S
Proof. Write X = Ui where Ui = Spec Ai for a noetherian ring Ai . For a quasi-coherent
sheaf F , there is an Ai -module Mi such that F |Ui = M fi for each I. Choose an injective
L
Ai -module Ei for which Mi ,→ Ei . Set E = (fi )∗ E ei where fi : Ui ,→ X is the canonical
inclusion. The morphisms F |Ui ,→ Ei glue to give a morphism F → (fi )∗ E
e ei , so by taking
the direct sum, there is a morphism F → E which is injective. For each i, Corollary 16.1.19
shows that (fi )∗ E
ei is an injective sheaf on X, hence flasque (Lemma 16.2.2). Moreover,
Lemma 14.4.8 shows each (fi )∗ E ei is quasi-coherent, so taking the direct sum gives a flasque,
quasi-coherent sheaf E into which F embeds.
Corollary 16.3.5. Let X be a noetherian scheme, U = {Ui } an affine open cover of X and
F a quasi-coherent sheaf on X. Then for all p ≥ 0, the natural map
p
Ȟ (U, F ) −→ H p (X, F )
is an isomorphism.
Proof. The p = 0 case was Lemma 13.4.5. For p ≥ 1, use Corollary 16.3.4 to embed F in
a flasque quasi-coherent sheaf G and let H = G/F be the quotient sheaf. Since F and G
are quasi-coherent and each Ui0 ,...,ip = Ui0 ∩ · · · ∩ Uip is affine, Theorem 16.3.3 implies that
H i (Ui0 ,...,ip , F ), H i (Ui0 ,...,ip , G) and H i (Ui0 ,...,ip , H) are all zero for i ≥ 1. Therefore the short
exact sequence 0 → F → G → H → 0 stays exact over each overlap:
Taking the product over all such overlaps, we see that the sequence of Čech complexes
with the final term vanishing since G is flasque (Proposition 16.2.5). Also note that for the
i i+1
same reason there will be isomorphisms Ȟ (U, H) ∼ = Ȟ (U, F ) for all i ≥ 1. Since H is
1 ∼
quasi-coherent and Ȟ (U, F ) −→ H 1 (X, F ) is an isomorphism by Corollary 16.2.7, the result
for all i ≥ 1 follows by induction.
Theorem 16.3.6 (Serre). For a noetherian scheme X, the following are equivalent:
(a) X is affine.
(d) There exist elements fi ∈ Γ(X, OX ) such that each D(fi ) = {x ∈ X | fi (x) 6= 0} is
affine and the ideal generated by the fi is Γ(X, OX ).
389
16.3. Noetherian Scheme Cohomology Chapter 16. Cohomology
and thus Γ(X, I) → Γ(X, I/J) is surjective. This means there is some f ∈ Γ(X, I) such that
f (x) 6= 0, so x ∈ D(f ) ⊆ U . Note that D(f ) = DU (f |U ) is indeed affine. Repeating this
for all closed pointsLof X, we get a family (fi )i∈S ⊂ Γ(X, OL X ). These define a morphism
of OX -modules ϕ : i∈S OX → OX such that each ϕD(fi ) : i∈S OX (D(fi )) → OX (D(fi ))
is surjective, L
but since the D(fi ) cover X, ϕ itself is surjective. The short exact sequence
0 → ker ϕ → i∈S OX → OX → 0 induces a long exact sequence
!
Γϕ
M
0 → Γ(X, ker ϕ) → Γ X, OX −→ Γ(X, OX ) → H 1 (X, ker ϕ) = 0
i∈S
390
16.4. Cohomology of Projective Space Chapter 16. Cohomology
S∼
M
= H 0 (X, O(d)).
d∈Z
for all d ∈ Z.
L
Proof. By (3) of Proposition 14.6.11, each O(d) is quasi-coherent, so the sheaf F = d∈Z O(d)
is quasi-coherent as well. Moreover, since cohomology comutes with direct sums, we have
M
H p (X, O(d)) = H p (X, F )
d∈Z
for all p ≥ 0.
(a) For 0 ≤ i ≤ n, set Ui = D+ (xi ), which is an affine open subset of X and X = ni=0 Ui .
S
p
Therefore by Corollary 16.3.5, H p (X, F ) ∼ = Ȟ (U, F ) for all p ≥ 0. For each i0 , . . . , ip ,
Ui0 ,...,ip = D+ (xi0 · · · xip ) so by (3) of Theorem 14.6.2, F (Ui0 ,...,ip ) ∼
= S(xi0 ···xip ) . This shows
that the Čech complex for F with respect to U is
Y Y
C • (U, F ) : 0 → S(xi ) → S(xi xj ) → · · · → S(x0 ···xn )
i i,j
and this is compatible with the grading on F . The kernel of the first map is on one hand
0
Ȟ (U, F ) but on the other hand one can easily see that this kernel is precisely S.
(b) We may assume n > 1 and induct on n. Localizing the above Čech complex C • (U, F )
at the element xn (as a graded module) yields
Y Y
0 → S(xn ) × S(xi xn ) → S(xn ) × S(xi xj xn ) → · · · → S(x0 ···xn )
i<n i,j<n
which is precisely the Čech complex C • (U 0 , F |Un ) where U 0 = {Ui ∩ Un } is an open cover of
p
Un . Taking cohomology of this yields Ȟ (U 0 , F |Un ) ∼
= H p (Un , F |Un ) by Corollary 16.3.5. But
391
16.4. Cohomology of Projective Space Chapter 16. Cohomology
since Un is affine and F is quasi-coherent, H p (Un , F |Un ) = 0 for p > 0 by Theorem 16.3.3,
and since localization is exact, H p (X, F )xn = 0 for all 0 < p < n. To finish, consider the
short exact sequence of graded S-modules
n x
0 → S(−1) −→ S→
− S/(xn ) → 0.
Applying the sheaf construction in Proposition 14.6.11 gives a short exact sequence of OX -
modules
0 → O(−1) → OX → OH → 0
where H ⊂ X is the hyperplane defined by xn = 0. Shifting the entire sequence by d yields
a short exact sequence
0 → O(d − 1) → O(d) → OH (n) → 0
and taking the direct sum over all d ∈ Z gives
0 → F (−1) → F → FH → 0
where FH = d∈Z OH (d). In the long exact sequence for this sequence, the map H p (X, F (−1)) →
L
H p (X, F ) is just multiplication by xn and, up to one degree of grading, H p (X, F (−1)) =
H p (X, F ). Since H ∼= Pn−1 p
A , induction gives H (X, FH ) = 0 for 0 < p < n − 1. The end of
the long exact sequence can be written as
δ x
0 → H n−1 (X, FH ) →
− H n (X, F (−1)) −→
n
H n (X, F ) → 0
but this may not be exact. However, it is clear from the description below of H n (X, F )
as a free A-module that the second map is surjective and δ is injective, so we have that
∼
multiplication by xn is an isomorphism H p (X, F (−1)) − → H p (X, F ) for all 0 < p < n. Since
we showed H p (X, F )xn = 0, this implies H p (X, F ) = 0 for the same values of p.
(c) The last map in the Čech complex,
Y
α: S(x0 ···x̂k ···xn ) −→ S(x0 ···xn ) ,
k
n
has kernel Ȟ (U, F ) which is isomorphic to H n (X, F ) by Corollary 16.3.5. The localization
Sx0 ···xn is a free A-module with basis {xj00 · · · xjnn | ji ∈ Z}, and
Thus ker α is the free A-module generated by the basis {xj00 · · · xjnn | ji < 0 for all i}. In
degree −n−1, the only monomial with this condition is x−1 −1
0 · · · xn , so (ker α)−n−1 is a free A-
module of rank 1. But the (−n−1)st graded piece of H (X, F ) is precisely H n (X, O(−n−1)).
n
(d) If d < 0, then H 0 (X, O(d)) = 0 by (a). To see that H n (X, O(−d − n − 1)) = 0, note
that there are no negatively graded monomials of degree > −n − 1 so a similar argument to
that in (c) works. So the statement holds trivially for d < 0. If d ≥ 0,
( n
)
H 0 (X, O(d)) ∼ j
X
= A x 0 · · · xjn | ji ≥ 0 and
0 n ji = n
i=0
392
16.4. Cohomology of Projective Space Chapter 16. Cohomology
The kernel of the pairing is any monomials for which ji + ki ≥ 0 for any i, so we can see that
the dual bases {xj00 · · · xjnn } and {x−j
0
0 −1
· · · x−j
n
n −1
} induce a perfect pairing.
The remaining results in this section are stated without proof but can be found in
Hartshorne.
Definition. An invertible sheaf L on a scheme Y is called very ample (or very ample
relative to a base X) if there exists an n ∈ N and an immersion f : Y → Pn (or to PnX )
such that f ∗ O(1) ∼
= L. We say L is ample if there is some m ∈ N such that L⊗m is very
ample.
Theorem 16.4.3 (Serre). Let A be a noetherian ring and X a projective A-scheme. Suppose
O(1) is a very ample sheaf on X and F is any coherent sheaf on X. Then there exists an
N ∈ N such that for all n ≥ N , F (n) is generated by finitely many global sections, i.e.
there exists a collection of sections {s1 , . . . , sr } ⊂ Γ(X, F (n)) such that for every x ∈ X, the
elements si,x ∈ F (n)x generate F (n)x as an OX,x -module.
Theorem 16.4.5 (Serre’s Theorem A). Let A be a noetherian ring and X a projective A-
scheme, with very ample sheaf O(1) relative to Spec A. Then for any coherent sheaf F on
X,
(b) There exists an N ∈ N such that for each n ≥ N , H i (X, F (n)) = 0 for all i > 0.
393
Chapter 17
Curves
394
17.1. Divisors Chapter 17. Curves
17.1 Divisors
Definition. Let X be a variety. An irreducible divisor on X is a closed, irreducible
k-subvariety x of X of codimension 1.
When X is a curve over k, an irreducible divisor is a closed point of MaxSpec k[X ∩ Ui ]
for some affine patch Ui , or alternatively, a Gk -orbit of points in X(k̄).
Definition. The degree of an irreducible divisor x on X is the size of the Gk -orbit in X(k̄)
corresponding to x, i.e. deg(x) = [κ(P ) : k] for any P ∈ x.
Example 17.1.1. Let X = P1 . On an affine patch A1 ,→ Ui ⊆ P1 , the irreducible divisors
correspond to irreducible polynomials in k[A1 ] = k[t].
Definition. Let X be a curve over k. The divisor group on X, Div(X), is the free abelian
group on the set of irreducible divisors on X:
( )
X
Div(X) = D = nx x : nx ∈ Z, nx 6= 0 for finitely many x .
x∈X
P
The elements of Div(X) are called
P divisors on X. For a divisor D = x∈X nx x ∈ Div(X),
the degree of D is deg(D) = x∈X nx deg(x).
Example 17.1.2. If k is algebraically closed, then the irreducibleP
divisors are the points of
X, so each D ∈ Div(X) is a weighted sum of points of P X: D = x∈X nx x. The degree of
such a divisor is just the sum of the weights: deg(D) = x∈X nx .
∗
P Now assume X is a nonsingular curve. For f ∈ k(X) , we can define a divisor D(f ) =
x∈X ordx (f )x, called the principal divisor of f . This defines a map
D : k(X)∗ −→ Div(X)
whose image is denoted PDiv(X), the group of principal divisors on X.
Definition. The Picard group, or divisor class group, of X is the quotient group
Pic(X) = Div(X)/ PDiv(X).
This defines an equivalence relation on divisors: D1 ∼ D2 if D1 = D2 + D(f ) for some
f ∈ k(X)∗ .
Example 17.1.3. Consider the variety E = Z(y 2 − x3 − 3x2 − 2x). This is the elliptic curve
defined by y 2 = f (x) where f = x3 + 3x2 + 2x = x(x + 1)(x + 2). The projective closure of
E is E = Z(fh ), where
fh = ZY 2 − X 3 − 3X 2 Z − 2XZ 2 .
Y
Setting y = Z
, we can compute its divisor on E:
X
D(y) = ordP (y)P.
P ∈X
On the affine part, there are only zeroes of y, and they occur precisely at P = (−2, 0), (−1, 0)
and (0, 0).
395
17.1. Divisors Chapter 17. Curves
Note that t ∈ OE,P is a uniformizer whenever dP t 6= 0. Viewing t ∈ k[x, y], i.e. as a lift of
[t] ∈ OE,P , we have that
∂f
t = x is a uniformizer as long as dP x = x|TP E 6= 0, which is equivalent to (P ) 6= 0.
∂y
∂f
t = y is a uniformizer as long as dP y = y|TP E 6= 0, that is, (P ) 6= 0.
∂x
In particular, we can always find a uniformizer! For P = (−2, 0), (−1, 0) and (0, 0), t = y is
a uniformizer. It follows that ordP (y) = 1 at each of these points, and ordQ (y) = 0 for any
other point Q ∈ E. Thus the divisor for y is
g = ζ − (ξ 3 + 3ξ 2 ζ + 2ξζ 2 ).
∂g
Notice that ∂ζ
(0, 0) = 1, so ξ is a uniformizer on this patch. Now
ordP (ζ) = ordP (ξ 3 + 3ξ 2 ζ + 2ξζ 2 ) ≥ min{ordP (ξ 3 ), ordP (3ξ 2 ζ), ordP (2ξζ 2 )}.
396
17.1. Divisors Chapter 17. Curves
We have ordP (ξ 3 ) = 3 and ordP (3ξ 2 ζ), ordP (2ξζ 2 ) ≥ 3. If all three orders are equal to 3, then
by the ultrametric inequality ordP (ζ) must be strictly greater than the minimum, which is
3 in this case. But then ordP (3ξ 2 ζ) = ordP (ξ 2 ) + ordP (ζ) > 2 + 3 > 3, so in fact we cannot
have all three orders equal to 3. Hence ordP (ζ) = 3. We have thus calculated the divisor of
y on the elliptic curve E:
397
17.2. Morphisms Between Curves Chapter 17. Curves
Proof. A more general result is that if Y is a normal variety, i.e. the local rings OY,P are
normal rings, then the locus of nondeterminacy of such a rational map ϕ : Y 99K X is a
subvariety of codimension at least 2. For Y = C a curve, this means there are no points
where ϕ fails to be regular.
A nonconstant rational map ϕ : C1 99K C2 between curves induces a field extension
k(C2 ) ,→ k(C1 ). Since both function fields have transcendence degree 1, this is in fact a
finite field extension.
Definition. For curves C1 and C2 and a rational map ϕ : C1 99K C2 , define the degree
of ϕ by deg ϕ = [k(C1 ) : k(C2 )]; the separable degree of ϕ by degs ϕ = [k(C1 ) : k(C2 )]s ;
and the inseparable degree of ϕ by degi ϕ = [k(C1 ) : k(C2 )]i . We say ϕ is separable if
k(C1 ) ⊇ k(C2 ) is a separable extension.
Definition. Any finitely generated field extension of k with transcendence degree 1 over k
is called a function field of degree 1 over k.
Proof. (Sketch) The assignment X 7→ k(X) determines one direction: we have seen that
k(X) is indeed a function field over k. Conversely, for a function field K/k, we associate
an abstract algebraic curve XK to K by putting a Zariski topology on theTmaximal ideals
of the valuation rings O ⊂ K. The structure sheaf is given by OXK (U ) = P ∈U OP where
U ⊆ XK is open and OP is the valuation ring corresponding to P . This determines the
reverse assignment K 7→ XK . One now checks that these assignments are inverse and
preserve categorical structure.
Now fix nonsingular curves X and Y over k and a morphism ϕ : X → Y defined over k.
Then an irreducible divisor y ∈ Div(Y ) corresponds to a maximal ideal mY (on some affine
patch) with uniformizer ty ∈ k(Y ).
Example 17.2.3. Let X be the plane curve defined by y 2 − x and Y = P1 the projective
line, and let ϕ : X → Y be the x-coordinate projection.
398
17.2. Morphisms Between Curves Chapter 17. Curves
x2
X
x0
x1
Y
y0 y1
Then ϕ∗ y0 = 2x0 + ord∞ (ϕ∗ ty0 )∞ and ϕ∗ y1 = x1 + x2 + ord∞ (ϕ∗ ty1 )∞.
ϕ∗ x = [κ(x) : κ(ϕ(x))]ϕ(x)
399
17.2. Morphisms Between Curves Chapter 17. Curves
But on the other hand, this means deg(f ) = deg(f −1 (0) − f −1 (∞)) = [k(X) : k(f )] − [k(X) :
k(f )] = 0.
Let Div0 (X) be the subgroup of Div(X) consisting of divisors of degree zero. Then
Corollary 17.2.6 shows that PDiv(X) ⊆ Div0 (X). Set
If X is defined over the algebraic closure k̄, write Pic(X/k̄) for Pic(X(k̄)). Consider Div(X/k̄)Gk .
Then we have an embedding
400
17.3. Linear Equivalence Chapter 17. Curves
deg : Pic(X) −→ Z.
P
Definition. A divisor D = nx x on X is called effective if nx ≥ 0 for all x ∈ X. In this
case we will write D ≥ 0. Also, if D1 , D2 ∈ Div(X) and D1 − D2 is an effective divisor, we
write D1 ≥ D2 . This defines an ordering on Div(X).
Definition. Let D be an effective divisor on X. Then the Riemann-Roch space associated
to D is the k-vector space
is called the complete linear system of D on X. Any projective subspace of |D| is called
a linear system of D on X.
Note that D is linearly equivalent to an effective divisor if and only if L(D) 6= 0.
Theorem 17.3.3. For any D ∈ Div(X), L(D) is finite dimensional.
Lemma 17.3.4. If D1 , D2 ∈ Div(X) are linearly equivalent, say D1 − D2 = (g) for some
g ∈ k(X)× , then there is an isomorphism
L(D1 ) −→ L(D2 )
f 7−→ gf.
401
17.3. Linear Equivalence Chapter 17. Curves
Next, we explore how much less than deg(D) + 1 the dimension `(D) can be. This
culminates with the Riemann-Roch theorem in Section 17.6. Set γ(D) = deg(D) + 1 − `(D).
The Riemann-Roch spaces are useful for constructing maps X → PN and in particular
embeddings into projective space. Given a rational map ϕ = (ϕ0 , . . . , ϕN ) : X 99K PN with
ϕi ∈ k(X), define the divisor of ϕ to be
Definition. If the complete linear system |D| is basepoint-free and the morphism ϕ|D| : X →
PN is an embedding, we say |D| is very ample. If for some m > 0, the complete linear
system |mD| is very ample, then we say |D| is ample.
P
Theorem 17.3.9. Let X be a curve and D = nx x an effective divisor on X. Then
(1) D is basepoint-free if and only if for all x ∈ X such that nx 6= 0, `(D − x) < `(D).
402
17.3. Linear Equivalence Chapter 17. Curves
(2) |D| is very ample if and only if `(D − P − Q) < `(D − P ) < `(D) for all P, Q ∈ X.
U 7→ OX (U )
defines a sheaf OX on X, called the structure sheaf of X. Note that for each P ∈ X,
OX,P = OP is the local ring at P , and for any open set U ⊆ X,
\
OX (U ) = OP .
P ∈U
For any sheaf F on X, let H i (X, F) be the ith cohomology of X with coefficients in F.
Theorem 16.2.4 implies the following:
Then U 7→ L(D)(U ) defines a sheaf of k-vector spaces on X, denoted L(D). In fact, L(D)
is a subsheaf of the constant sheaf on X associated to k(X) (which by abuse of notation we
will also write as k(X)). Note that H 0 (X, L(D)) = L(D), the Riemann-Roch space for D.
Proposition 17.3.11. For any divisor D on X, H i (X, L(D)) = 0 for i ≥ 2 and is a finite
dimensional k-vector space for i = 0, 1.
Proof. First suppose D = 0. Then L(0) = OX is the sheaf of regular functions on X and
since X is complete,
Hence h0 (L(0)) = 1. We will prove dimk H 1 (X, L(0)) < ∞ later. For the rest, it’s enough
to prove that the statements hold simultaneously for D and D0 = D + P , for any divisor
D and any point P ∈ X. For any point Q ∈ X, we have L(D)Q ⊆ L(D + P )Q so we may
regard L(D) as a subsheaf of L(D + P ). Consider the exact sequence of sheaves
0 → L(D) → L(D + P ) → F → 0
where F is the quotient sheaf. Then FQ = 0 for all Q 6= P , and for Q = P we get
L(D + P )P
dimk FP = dimk = 1, so H 1 (X, F) = 0 and H 0 (X, F) = k. Thus the long exact
L(D)P
sequence in sheaf cohomology is
403
17.3. Linear Equivalence Chapter 17. Curves
404
17.4. Répartitions Chapter 17. Curves
17.4 Répartitions
To interpret h1 (L(D)) in Theorem 17.3.12, we want to find a ‘canonical divisor’ for which
h1 (L(D)) = `(K − D) for all divisors D on X. We first interpret H 1 (X, L(D)) using
répartitions and then relate the dual of this space to the space of meromorphic differentials
on our curve. Finally, this will allow us to define the canonical divisor.
Remark. R may also be identified as the ring of adèles of k, as in adèlic number theory.
P
For a divisor D = nP P ∈ Div(X), define the subspace
As with the Riemann-Roch spaces, we have R(D0 ) ⊇ R(D) when D0 ≥ D. Also, note that
[
R= R(D).
D∈Div(X)
Then we have H 0 (X, k(X)) ∼ = k(X) since k(X) is a constant sheaf, and H 1 (X, k(X)) = 0,
either by computing using Čech cohomology or using the fact that k(X) is a flasque sheaf.
In any case, the long exact sequence becomes:
Moreover, Q has finite support (it vanishes off of a finite set, namely the support of D), so
it is a skyscraper sheaf and thus
H 0 (X, Q) ∼ (k(X)/L(D))P ∼
M
= = R/R(D)
P ∈X
Now since k(X) is a field, the natural k-algebra map k(X) → R/R(D), f 7→ {f } is injective,
so by exactness we get H 1 (X, L(D)) ∼= R/(R(D) + k(X)) as desired.
405
17.4. Répartitions Chapter 17. Curves
Set I(D) = R/(R(D) + k(X)) and let J(D) = I(D)∗ denote the algebraic dual. By
duality, we have J(D0 ) ⊆ J(D) if D0 ≥ D. Set
[
J= J(D).
D∈Div(X)
f α : R −→ k
r 7−→ α(f r).
Thus dim J(D − ∆n ) ≥ 2`(∆n ), but by the first form of Riemann-Roch (Theorem 17.3.12),
406
17.5. Differentials Chapter 17. Curves
17.5 Differentials
Definition. For a curve X, the space of meromorphic differentials on X is the k(X)-
vector space ΩX consisting of formal differentials df for each f ∈ k(X)× satisfying
d(f + g) = df + dg,
dα = 0 if α ∈ k,
d(f g) = f dg + g df .
That is, ΩX is the k-algebra
ϕ∗ : ΩY −→ ΩX
X X
∗
ϕ fi dti 7−→ ϕ∗ fi d(ϕ∗ ti ).
(2) For any s ∈ k(X) such that s(P ) = 0, ordP (f ds) = ordP (f ) + ordP (s) − 1 if p -
ordP (s), and ordP (f ds) ≥ ordP (f ) + ordP (s) if p | ordP (s).
407
17.5. Differentials Chapter 17. Curves
Definition. The canonical class on a curve X is the class KX = [(ω)] in Pic(X) for any
nonzero differential ω ∈ ΩX .
Lemma 17.5.5. The canonical class is well-defined, i.e. does not depend on the choice of
ω ∈ ΩX .
Note that Ω[X] is a k-vector space but need not be a k(X)-vector space.
Definition. The geometric genus of X is defined as g(X) := `(KX ), the dimension of the
Riemann-Roch space L(KX ) of the canonical class.
Proof. The map is f 7→ f ω for any fixed ω ∈ Ω[X] defining the canonical class.
Remark. For any divisor D ∈ Div(X), `k (D) = `k̄ (D) implies g(X(k)) = g(X(k̄)), so the
geometric genus is unchanged when passing to the algebraic closure k̄. Moreover, g(X) is a
birational invariant of X.
Example 17.5.8. Let X = P1 and let t be a coordinate function on some affine patch U of
P1 . We claim that (dt) = −2∞. Indeed, for any α ∈ U ∼ = A1 , t − α is a local uniformizer at
α. Thus ordα(dt) = ordα (d(t − α)) = 0. At infinity, 1t is a local uniformizer so we can write
dt = −t2 d 1t . Hence
So (dt) = −2∞ as claimed. Now for any ω ∈ ΩP1 , deg(ω) = −2 so we see that `(KP1 ) =
`(−2∞) = 0. Hence the genus of the projective line is g(P1 ) = 0.
Proof. By Corollary 17.5.7, g(P1 ) = dimk Ω[P1 ] but by the calculations above, the genus of
P1 is zero.
408
17.5. Differentials Chapter 17. Curves
There is another useful local invariant of a meromorphic differential, which we define now.
For P ∈ X, let ObP = k[[T ]] be the topological completion of the local ring OP , which comes
equipped with a map OP → O bP , t 7→ T for a local uniformizer t; this map is an embedding.
Then the fraction field of this ring is FbP := Frac(O bP ) = k((T )) so via the embedding just
defined, we can view functions f ∈ OP as formal Laurent series in T ,
∞
X
f= ai T i .
i=−n
Definition.
P∞ The residue of ω ∈ ΩX at P ∈ X is resP (ω) := a−1 if ω = f dt and f =
i
i=−n ai T .
Fix P ∈ X and let (O, m, v) be the DVR of the local field FbP and let Dk (FbP ) be the ring
of k-differentials on FbP . Set
\∞
Q= mn dO ⊆ Dk (FbP )
n=1
dt is a basis for D0 .
Proposition 17.5.12. The residue resP (ω) is well-defined, i.e. does not depend on the
choice of uniformizer t.
409
17.5. Differentials Chapter 17. Curves
Proof. We first establish this for X = P1 . In this case, k(X) = k(t) so for any differential
ω ∈ ΩP1 , ω = f dt for some f ∈ k(t). Since resP is k-linear by Proposition 17.5.11, we may
assume that f = tn or f = (t − a)−n for some n > 0 and a ∈ k. If f = tn , the only pole of
ω = tn dt is at P = ∞. At ∞, u = 1t is a uniformizer, so we get
Tr : ΩX −→ ΩP1
ω 7−→ TrF/E (f ) dϕ
where the product isPover all Q ∈ X for which ϕ(Q) = P . Under this isomorphism, for any
f ∈ F , TrF/E (f ) = Q→P TrQ (f ), which implies
X
resP (Tr(f ) dϕ) = resP (TrQ (f )) dϕ,
Q→P
which is 0 by the P1 case proven above. Therefore the residue formula holds for X.
410
17.6. Serre Duality and the Riemann-Roch Theorem Chapter 17. Curves
Ω(X) × R −→ k
X
(ω, r) 7−→ hω, ri := resP (rP ω)
P ∈X
if r = {rP }. Note that if r ∈ k(X) ⊆ R, then by the residue theorem (17.5.13), hω, ri = 0 for
all ω ∈ Ω(X). Further, if D is a divisor, r = {rP } ∈ R(D) and ω ∈ Ω(D), then rP ω ∈ Ω(X)P
for all P ∈ X, since vP (rP ) ≥ −vP (D) and vP (ω) ≥ vP (D). Hence the pairing passes to the
quotient I(D) = R/(R(D) + k(X)):
Then vP (rP ω) = −1, so resP (rP ω) 6= 0. Thus hω, ri 6= 0 since the pairing is 0 everywhere
else. However, r ∈ R(D) implies hω, ri = 0, a contradiction.
Theorem 17.6.2 (Serre Duality). For any divisor D ∈ Div(X), θ : Ω(D) → J(D) is an
isomorphism.
Proof. Suppose θ(ω) = 0. Then θ(ω) ∈ J(D) for every D, so by Lemma 17.6.1, ω ∈ Ω(D)
for every D. However, \
Ω(D) = 0
D∈Div(X)
since X is complete, so ω = 0. This proves θ is injective. On the other hand, dimk(X) Ω(X) =
1 and dimk(X) J ≤ 1 by Proposition 17.4.2, so the map Ω(X) → J must be surjective.
Suppose α ∈ J(D). Then there is some ω ∈ Ω(X) with θ(ω) = α, but by Lemma 17.6.1,
this means ω ∈ Ω(D). Therefore θ is surjective and hence an isomorphism.
Let D be a divisor on X. By Theorem 17.3.12, we already have
where `(D) = dimk L(D) = dimk H 0 (X, L(D)) and i(D) = dimk Ω(D) = dimk H 1 (X, L(D)).
Let ω, ω 0 ∈ Ω(X). Since Ω(X) has dimension 1, we may write ω = f ω 0 for f ∈ k(X). Then
(ω) = (ω 0 ) + (f ), so in particular every meromorphic differential form on X belongs to the
same nontrivial linear equivalence class in Pic(X), call it K.
411
17.6. Serre Duality and the Riemann-Roch Theorem Chapter 17. Curves
Definition. The class K ∈ Pic(X) such that [(ω)] = K for any ω ∈ Ω(X) is called the
canonical divisor class of X.
Fix a differential form ω0 ∈ Ω(X), so that K = [(ω0 )]. For any D ∈ Div(X) and
ω ∈ Ω(X), (ω) ≥ D is equivalent to (f ) ≥ D − (ω0 ), ω = f ω0 for f ∈ k(X). This proves:
Lemma 17.6.3. ω ∈ Ω(D) if and only if f ∈ L(K − D), where ω = f ω0 .
Theorem 17.6.4 (Riemann-Roch). For an algebraic curve X with canonical divisor K and
any divisor D ∈ Div(X),
Proof. Lemma 17.6.3 shows that i(D) = `(K − D) so this is just a reinterpretation of
Theorem 17.3.12.
Corollary 17.6.5. If KX is the canonical divisor on X, then deg(KX ) = 2g − 2.
Proof. Set D = K = KX . Then the Riemann-Roch theorem says that
but `(K) = g by definition and `(0) = 1. Solving for deg(K) we get deg(K) = 2g − 2.
Corollary 17.6.6. Suppose deg(D) > 2g − 2 for some divisor D ∈ Div(X). Then `(D) =
deg(D) + 1 − g.
The genus is a discrete invariant of nonsingular curves. There are two natural questions
that arise:
(1) What are the curves with genus g for a particular g ∈ N0 ?
(2) How do we describe the structure of the collection of all genus g curves?
We will see that one can put the structure of a variety on the collection of genus g curves.
Lemma 17.6.7. Let X be an algebraic curve. Then X ∼ = P1 if and only if there is some
divisor D ∈ Div(X) such that deg(D) = 1 and `(D) ≥ 2.
Proof. ( =⇒ ) If X ∼
= P1 then g(X) = g(P1 ) = 0 by Example 17.5.8. Take a point P ∈ X
and set D = P ∈ Div(X); of course deg(D) = 1. Then by the Riemann-Roch theorem,
412
17.6. Serre Duality and the Riemann-Roch Theorem Chapter 17. Curves
(1) X ∼
= P1 .
(2) g = 0 and there exists a divisor D ∈ Div(X) with deg(D) = 1.
413
17.7. The Riemann-Hurwitz Formula Chapter 17. Curves
d(ϕ∗ t) = se g ds + euse−1 ds
=⇒ ordP (d(ϕ∗ t)) = ordP (se g + euse−1 )
= min{ordP (se g), ordP (euse−1 )}.
If char k - e, then this minimum is e − 1; otherwise, when char k | e the minimum is at least
e.
Definition. If ϕ is ramified and char k - eϕ (P ) for all P ∈ X, we say ϕ is tamely ramified.
Otherwise ϕ is wildly ramified.
Remark. If ϕ is tamely ramified, then ordP (d(ϕ∗ t)) = eϕ(P ) − 1 for each P . If ϕ is wildly
ramified at P , then ordP (d(ϕ∗ t)) ≥ eϕ (P ).
Definition. For a morphism ϕ : X → Y , define the ramification divisor
X
Rϕ = ordP (d(ϕ∗ t))P.
P ∈X
Now for ω ∈ ΩY , the canonical classes on X and Y can be defined by KY = [(ω)] and
KX = [(ϕ∗ ω)]. On the other hand, the pullback defines a divisor ϕ∗ KY ∈ Div(X). We want
to determine the relation between these three divisors.
Lemma 17.7.1. If ϕ : X → Y is a morphism of curves, then KX = ϕ∗ KY + [Rϕ ], where
Rϕ is the ramification divisor of ϕ.
Proof. If ω = f dt ∈ ΩY , then
ordP (ϕ∗ ω) = ordP (ϕ∗ f d(ϕ∗ t)) = ordP (ϕ∗ f ) + ordP (d(ϕ∗ t)),
so we see that ordP (ϕ∗ ω) gives the coefficient in KX , ordP (ϕ∗ f ) gives the coefficient in
ϕ∗ KY and ordP (d(ϕ∗ t)) gives the coefficient in Rϕ . Summing over P ∈ X gives the desired
equality.
P
Taking ϕ to be tamely ramified, Rϕ = P ∈X (eϕ (P ) − 1)P so the degree function applied
to the equation in Lemma 17.7.1 gives
X
deg(KX ) = deg(ϕ∗ KY ) + (eϕ (P ) − 1).
P ∈X
414
17.7. The Riemann-Hurwitz Formula Chapter 17. Curves
415
17.8. The Canonical Map Chapter 17. Curves
Proof. This follows from the Riemann-Roch theorem (17.6.4) and Theorem 17.3.9, taking
D = KX .
Thus |KX | determines a regular map into projective space.
Definition. The canonical map of a genus g ≥ 1 curve X is the map ϕ|KX | : X → Pg−1 .
Example 17.8.2. When char k 6= 2, a hyperelliptic curve is of the form X = Z(y 2 − f (x))
for a polynomial f ∈ k[x]. More generally, the minimal degree of a nonconstant morphism
X → P1 is called the gonality of X. Thus, a hyperelliptic curve is a curve of gonality 2.
We have for the most part dealt completely with the cases of curves of genus g = 0 and
g ≥ 2, so the most interesting work remains to be done for curves of genus g = 1.
416
17.9. Bézout’s Theorem Chapter 17. Curves
Q
P
If L is the linear form representing the line in the figure, then (X · L)P = 1, (X · L)Q = 2 and the
intersection number is (X · L) = 1 + 2 = 3.
Proof. Set f = F/F1 ∈ k(X). Then divX (F ) ∼ divX (F1 ), so deg(divX (F )) = deg(divX (F1 )),
and thus the intersection number is well-defined.
Corollary 17.9.2. If deg F = m and L is any linear form such that L 6∈ J(X), then
(X · F ) = m(X · L).
Lemma 17.9.3. For any form F 6∈ J(X) and any point P ∈ X ∩ Z(F ), (X · F )P = 1 if
and only if F (P ) = 0 and TP X 6⊂ TP Z(F ).
Stated another way, Lemma 17.9.3 says that the intersection multiplicity at P is 1 if and
only if X and Z(F ) meet transversely.
Lemma 17.9.4. For any smooth curve X, there exists a linear form L such that (X ·L)P ≤ 1
for all P ∈ X ∩ Z(L).
417
17.9. Bézout’s Theorem Chapter 17. Curves
(X · F ) = (deg G)(deg F )
418
17.10. Rational Points of Conics Chapter 17. Curves
(b) The pairing bq (v, w) = 12 (q(v + w) − q(v) − q(w)) is symmetric and k-bilinear.
such that F (X, Y, Z) = (X Y Z)MF (X Y Z)t . The determinant deg MF is called the
discriminant of F .
The crucial Hasse-Minkowski theorem says that a plane conic having a Q-rational point
is equivalent to the conic having a rational point over every completion of Q.
419
17.10. Rational Points of Conics Chapter 17. Curves
This theorem is the classic example of Hasse’s “local-to-global principle”: points over the
local fields Qv determine points over Q. Note that the Hasse-Minkowski theorem does not
hold for general varieties X, nor for general fields k.
Example 17.10.3. For a conic X, X(R) 6= ∅ if and only if there is a change of sign among
the coefficients ai in the form F defining X. This condition is easily checked as long as one
can diagonalize MF .
Thus to find rational points of a conic, we need only ask if there is an algorithm for
checking whether X has points over each p-adic field Qp .
Example 17.10.4. Let X = Pn . Then Pn (Q) = Pn (Z) and for any prime p, Pn (Qp ) =
Pn (Zp ), so it’s enough to look for integer solutions. If P = [α0 , . . . , αn ] ∈ Pn (Qp ), then
we can clear denominators so that P = [β0 , . . . , βn ] for βi ∈ Zp and some βj ∈ Z× p . The
n
reduction mod p of P is then given by Pe = [β̄0 , . . . , β̄n ] ∈ P (Fp ).
It turns out that quadratic forms always have points over finite fields. To prove this, we
will need the following counting lemma.
Lemma 17.10.5. For a sum s = α∈Fqn α1k1 · · · αnkn , where α = (α1 , . . . , αn ) and ki ∈ Z≥0 ,
P
Proof. Write
X n
Y X
s= α1k1 · · · αnkn = ak i .
α∈Fqn i=1 a∈Fq
generator of the cyclic group Fq× and write ψ = φki . If ki is not a positive multiple of q − 1,
then ψ 6= 1. Now we have
q−2
X X X
ki ki
a = a = (φm )ki
a∈Fq a∈Fq× m=0
q−2
X 1 − ψ q−1 1−1
= ψm = ≡ = 0.
m=0
1−ψ 1−ψ
Therefore s = 0.
420
17.10. Rational Points of Conics Chapter 17. Curves
Now we have r n
X X
deg P = deg fi (q − 1) < n(q − 1)
i=1 i=1
Corollary 17.10.7. Every quadratic form in at least three variables has a point over each
finite field.
The theory of Hasse-Minkowski extends more generally to number fields K/Q, with
similar local-global principles at work.
We next determine when solutions to quadratic equations F = 0 over finite fields lift to
solutions in Zp , similar to Hensel’s Lemma. To do so, we introduce the notion of an integral
model for a variety over Q.
Note that we may assume the set of all coefficients of an integral model X = {F1 , . . . , Fm }
is coprime.
XFp = Z(F 1 , . . . , F m ) ⊆ PN
Fp ,
421
17.10. Rational Points of Conics Chapter 17. Curves
Definition. We say an integral model X has good reduction mod p if XFp is geometrically
reduced and nonsingular, and bad reduction mod p otherwise.
Lemma 17.10.8. An integral model X = {F1 , . . . , Fm } has good reduction mod p if and only
if Z[X0 , . . . , XN ]/(F1 , . . . , Fm ) ⊗ Fp is a regular ring.
Example 17.10.9. If X ⊆ P2Q is a plane conic and X is an integral model of X over Q given
by a primitive quadratic form F ∈ Z[X0 , X1 , X2 ], then X has bad reduction at a prime p if
and only if p divides the discriminant ∆(F ).
Corollary 17.10.10. A primitive quadratic form F ∈ Z[X0 , X1 , X2 ] has bad reduction at
only finitely many primes.
The following is a version of Hensel’s Lemma that we will need for lifting solutions of
quadratic forms.
Theorem 17.10.11. Let (R, v) be a complete DVR, f ∈ R[x1 , . . . , xN ] and suppose (a1 , . . . , aN ) ∈
RN such that
∂f
v(f (a1 , . . . , aN )) > 2v (a1 , . . . , aN )
∂xi
for some 1 ≤ i ≤ N . Then f has a root in RN .
Corollary 17.10.12. If X = Z(F ) is an integral model over Zp and P is a smooth point of
X (Fp ), then P lifts to a point of X (Zp ).
Corollary 17.10.13. Let β ∈ Z× 2 2
p . Then x = β has a solution in Zp if and only if x ≡ β
mod pε has a solution, where ε = 3 when p = 2 and ε = 1 otherwise.
Proof. Let f (x) = x2 − β ∈ Z[x] so that f 0 (x) = 2x. Suppose α0 ∈ Zp is a solution to
f (x) ≡ 0 mod pε , i.e. v(f (α0 )) ≥ ε. Then α02 ≡ β 6= 0 mod p so since Zp is a DVR, α0
must be a unit, i.e. v(α0 ) = 0. Now we have
2vp (f 0 (α0 )) = 2vp (2a0 ) = 2(vp (a0 ) + vp (2))
(
2, p = 2
= 2vp (2) =
0, p 6= 2
< v(f (α0 )) in all cases.
Therefore Hensel’s Lemma applies.
This leaves the question of lifting singular points.
Pn 2
Theorem 17.10.14. Let F = i=0 ai Xi be a primitive quadratic form over Zp and set
X = Z(F ) ⊆ P2Qp . Suppose β0 , . . . , βn ∈ Zp such that ordp (βj ) = 0 for some 0 ≤ j ≤ n, with
F (β0 , . . . , βn ) = 0 mod pε+1 , where
(
1, p 6= 2
ε=
3, p = 2.
Then there exists a nontrivial root of F in Zp , that is, α = (α0 , . . . , αn ) ∈ Znp , with α` 6= 0
for some 0 ≤ ` ≤ n, and F (α0 , . . . , αn ) = 0.
422
17.10. Rational Points of Conics Chapter 17. Curves
Theorem 17.10.15. Let F be a nondegenerate, primitive quadratic form over Z and let
X = Z(F ) be the corresponding conic over Q. Then X(Q) 6= ∅ if and only if
(3) F = 0 has a primitive solution mod p2 for all primes p > 2 – i.e. X(Qp ) 6= ∅.
In practice, one need only check (2) and (3) for primes at which X has bad reduction,
and by Corollary 17.10.10 there are only finitely many of these.
423
Chapter 18
Complex Varieties
This chapter is a survey of the theory of complex manifolds and varieties. This part of
the notes comes from a course on 4-manifolds taught by Dr. Tom Mark at the University
of Virginia in Spring 2019. I stuck around for the beginning of the course to take notes
on complex manifolds, complex surfaces and surface classification since these topics are
developed in parallel to the algebro-geometric theory of surfaces. The main reference for
this course is Griffiths-Harris, Principles of Algebraic Geometry.
424
18.1. Complex Manifolds Chapter 18. Complex Varieties
∂f ∂f
where ∂zj
and ∂ z̄j
are defined in the same way as their single-variable counterparts above.
¯ = 0.
Then f is holomorphic if ∂f
425
18.1. Complex Manifolds Chapter 18. Complex Varieties
ϕij := ϕ−1
j ◦ ϕi : ϕi (Ui ∩ Uj ) −→ ϕj (Ui ∩ Uj )
and its inverse are holomorphic. A complex structure on M is the choice of such a
complex atlas, up to holomorphic equivalence of charts, defined by a similar condition to the
above.
Real tangent space: the underlying real vector space of Tp M , which has real dimen-
sion 2n;
Complex tangent space: the complex vector space TpC M := Tp M ⊗ C, which has
complex dimension 2n;
n on
Holomorphic tangent space: the subspace Tp0 M := SpanC ∂z∂ j of TpC M ;
j=1
n on
Antiholomorphic tangent space: the subspace Tp00 M := SpanC ∂
∂ z̄j
of TpC M .
j=1
426
18.1. Complex Manifolds Chapter 18. Complex Varieties
We also define the real and complex cotangent spaces Tp∗ M and TpC,∗ M , as well as their
holomorphic and antiholomorphic counterparts Tp1,0 M := (Tp0 M )∗ and Tp0,1 M := (Tp00 M )∗ ,
respectively.
(2) Complex conjugation acts C-linearly on TpC M and exchanges Tp0 M and Tp00 M .
(4) Complex conjugation acts C-linearly on TpC,∗ M and exchanges Tp1,0 M and Tp0,1 M .
∂xj ∂yj
which defines a linear map JR f : Tp M → Tf (p) N for any p ∈ U . Extending scalars allows
us to define the complex Jacobian JC f : TpC M → TfC(p) N given by the same matrix JR f with
n o n o
∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂
respect to the real bases ∂xj , ∂ x̄j , ∂yj , ∂ ȳj and ∂uk , ∂ ūk , ∂vk , ∂v̄k of these tangent spaces.
n o n o
With respect to the complex bases ∂z∂ j , ∂∂z̄j and ∂w∂ k , ∂ w̄∂ k , the matrix is
∂wk ∂wk
∂zj ∂ z̄j
JC f =
∂ w̄k ∂ w̄k .
∂zj ∂ z̄j
(1) f is holomorphic.
∂wk ∂ w̄k
(2) For all 1 ≤ j ≤ m and 1 ≤ k ≤ n, ∂ z̄j
= ∂zj
= 0.
427
18.1. Complex Manifolds Chapter 18. Complex Varieties
Corollary 18.1.5. Suppose M and N are complex manifolds of the same dimension and
f : M → N is holomorphic. Then
2
∂w k
≥ 0.
det(JR f ) = det(JC f ) = det
∂zj
Example 18.1.7. Note that the operators ∂z∂ j and ∂∂z̄j act as expected on the coordinates
{zj , z̄j }. For example, consider the map f : C2 → C, f (z, w) = z 2 + w3 . Then
JC f = 2z 3w
which is surjective whenever (z, w) 6= (0, 0) so for any constant c ∈ C r {0}, the equation
z 2 + w3 = c defines a smooth submanifold of C2 .
More generally, there are holomorphic versions of the inverse and implicit function theo-
rems from differential geometry:
To prove these, one need only prove that f −1 (in the inverse function theorem) and g
(in the implicit function theorem) are holomorphic. The rest follows from real differential
geometry.
Remark. In the above example, the complex submanifold defined by the equation z 2 +w3 = c
is not compact. In fact, a complex submanifold of Cn is never compact. To see this, fix an
embedding M ,→ Cn of a complex submanifold. Then the composition f : M ,→ Cn → C
with projection onto the ith component is a holomorphic function, so by the maximum
principle, f must be constant. A version of this argument can be used to prove the following
important fact.
Theorem 18.1.10. If M is a closed complex manifold, then there are no nonconstant holo-
morphic functions M → C.
Example 18.1.11. Let Pn be complex projective n-space, defined as the quotient Cn+1 r
{0}/ ∼ where (z0 , . . . , zn ) ∼ (w0 , . . . , wn ) if wj = λzj for some λ ∈ C r {0}. Alternatively,
Pn is the space of complex lines through the origin in Cn+1 . Write the coordinates on Pn as
428
18.1. Complex Manifolds Chapter 18. Complex Varieties
[z0 , . . . , zn ]. There is a natural complex structure on Pn via the open sets Uj = {[z0 , . . . , zn ] ∈
Pn | zj 6= 0} and charts
ϕj : Uj −→ Cn
z0 zj−1 zj+1 zn
[z0 , . . . , zn ] 7−→ ,..., , ,..., .
zj zj zj zj
One can check that the transition maps are holomorphic, so this gives Pn the structure of an
n-dimensional complex manifold. For any homogeneous polynomial f ∈ C[z 0 , . . . , zn ], the
n ∂f ∂f
equation f (z0 , . . . , zn ) = 0 defines an algebraic set V (f ) in P . If JC f = ∂z 0
· · · ∂z n
is
nonzero at every point in V (f ) except for the point corresponding to 0 ∈ Cn+1 , then V (f )
is a complex submanifold of Pn .
429
18.2. Dolbeault Cohomology Chapter 18. Complex Varieties
Each graded piece p,q T C,∗ M has basis {dzi1 ∧· · ·∧dzip ∧dz̄j1 ∧· · ·∧dz̄jq | 1 ≤ i1 , . . . , ip , j1 , . . . , jq ≤
V
n} which we can write more compactly as {dzI ∧ dz̄J : |I| = p, |J| = q} where I, J ⊂ N and
dzI = dzi1 ∧ · · · ∧ dzip for I = {i1 , . . . , ip } (likewise for dz̄J ). Passing to global sections of
this vector bundle gives a type decomposition on the vector space of differential k-forms:
^ M ^ M
Ωk M := Γ M, k T C,∗ M = Γ M, p,q T C,∗ M =: Ωp,q M.
p+q=k p+q=k
In the last expression, the first term is a differential form of type (p + 1, q) and the second
term is of type (p, q + 1). Put
X X
∂ω = ∂fIJ ∧ dzI ∧ dz̄J and ∂ω ¯ = ¯ IJ ∧ dzI ∧ dz̄J .
∂f
|I|=p,|J|=q |I|=p,|J|=q
Lemma 18.2.1. For any complex manifold M , there are C-linear operators
430
18.2. Dolbeault Cohomology Chapter 18. Complex Varieties
for all k ≥ 0. In the case when M is a Kähler manifold (see Section 18.3),
Pthen the Hodge-de
Rham spectral sequence collapses on the second page, in which case p+q=k hp,q = bk for
all k. Moreover, when M embeds holomorphically into Cn for some n ≥ 1, then M has
the cohomology of an affine complex variety, so H p,q (M ) = 0 for all q > 0 and there is an
isomorphism HdR p
(M ) ∼
= H p (H •,0 (M ), ∂) for all p ≥ 0.
Example 18.2.2. Let E be a complex torus given by C/Λ for a lattice Λ ⊆ C. We may
write Λ = Z ⊕ Zτ for some τ = a + bi ∈ C such that b > 0. Then E carries the structure of
a complex elliptic curve (see Section 23.2). Using the above techniques, one may show that
431
18.2. Dolbeault Cohomology Chapter 18. Complex Varieties
where |·| is the usual Euclidean norm on Cn+1 . Then ωF S is a (1, 1)-form and is real, meaning
ω̄F S = ωF S . Moreover, dωF S = ∂ωF S = ∂ω ¯ F S = 0, so ωF S is a closed 2-form and defines a
class [ωF S ] ∈ H 1,1 (Pn ). On the other hand, one can show that ωF S is Poincaré dual to the
class of a hyperplane H ⊆ Pn representing Pn−1 ,→ Pn , so ωF S is not exact and therefore
[ωF S ] generates H 1,1 (Pn ). Since H 2 (Pn ; C) = H 2,0 (Pn ) ⊕ H 1,1 (Pn ) ⊕ H 0,2 (Pn ), it follows that
432
18.3. Kähler Manifolds Chapter 18. Complex Varieties
J : T M −→ T M
gC : TxC M ⊗ TxC M −→ C
for each x ∈ M such that if v, w ∈ Tx0 M , then gC (v, w) = 0. This is because Tx0 M may be
written Tx0 M = {X − iJX | X ∈ Tx M }.
Definition. For a complex Riemannian manifold M with compatible metric g, the associ-
ated Hermitian form on M is the assignment h of an R-bilinear map h : Tx0 M ⊗Tx0 M → C
to each x ∈ M , defined by h(v, w) = gC (v, w̄) for all v, w ∈ Tx0 M .
Lemma 18.3.2. Let (M, g) be a complex manifold with compatible metric and associated
Hermitian form h. Then
(a) For all v, w ∈ Tx0 M , h(v, w) = h(w, v). In particular, h(v, v) ∈ R.
433
18.3. Kähler Manifolds Chapter 18. Complex Varieties
Example 18.3.3. On Pn , the Fubini-Study form ωF S is closed and of the form ω(X, Y ) =
−g(X, JY ) where g is the standard metric inherited from Pn+1 . Thus Pn is a Kähler manifold.
Example 18.3.4. On Cn , the usual inner product may be expressed as a tensor field
n n
X 1X
g= (dxj ⊗ dxj + dyj ⊗ dyj ) = (dzj ⊗ dz̄j + dz̄j ⊗ dzj )
j=1
2 j=1
so the ∂z∂ j are mutually Hermitian-orthogonal of length √12 . Meanwhile, the associated 2-form
for this metric is determined by
∂ ∂ ∂ ∂
ω , = −g ,i =0
∂zj ∂zk ∂zj ∂zk
∂ ∂ ∂ ∂
ω , = −g ,i =0
∂ z̄j ∂ z̄k ∂ z̄j ∂ z̄k
∂ ∂ ∂ ∂ i
ω , = −g ,i = δjk .
∂zj ∂ z̄k ∂zj ∂ z̄k 2
as expected.
A short calculation shows that ω n = n! ε1 ∧ · · · ∧ ω2n = n! volM , where volM is the volume
form on M with respect to g. In particular this is easy to see from the calculation at the
end of Example 18.3.4 for Cn .
Proposition 18.3.7. Let (M, g) be a Kähler manifold and N ⊆ M any submanifold. Then
restricting g to N makes N a Kähler manifold.
434
18.3. Kähler Manifolds Chapter 18. Complex Varieties
Proof. Let i : N ,→ M be the subspace inclusion and write ωM and ωN for the associated
(1, 1)-forms on M and N , respectively. Then ωN = i∗ ωM and since d commutes with pullback,
we get dωN = 0. Therefore N is Kähler.
Corollary 18.3.8. On a complex manifold M of dimension n with Hermitian inner product
h and associated (1, 1)-form ω, the volume form on M is given by volM = n!1 ω n . In particular,
for any complex submanifold N ⊆ M of dimension d,
Z
1
volM (N ) = ωd.
d! N
Corollary 18.3.9. If (M, g) is a Kähler manifold, then
(1) The volume of any closed complex submanifold N ⊆ M depends only on the homology
class [N ] ∈ H • (M ).
(M ) ∼
M
k
HdR = H p,q (M ).
p+q=k
∆ : Ω• (M ) −→ Ω• (M )
ω 7−→ dd∗ ω + d∗ dω
435
18.3. Kähler Manifolds Chapter 18. Complex Varieties
k
cohomology class [ω] ∈ HdR (M ). Let Hk (M ) be the vector space of all harmonic k-forms on
M . The Hodge theorem says that the natural inclusion
Hk (M ) ,−→ HdR
k
(M )
is in fact an isomorphism for all k ≥ 0, so every de Rham cohomology class has a unique
harmonic representative.
On the other hand, since M is Kähler, its metric is compatible with the complex structure
on M in aLway that ensures ∆ preserves the type decomposition of differential k-forms
Ωk (M ) = p,q
p+q=k Ω (M ), and therefore this type decomposition is also compatible with
the ‘anti-holomorphic Laplace operator’
Hence the Hodge theorem, suitably applied to ∆∂¯, says that every Dolbeault cohomology
class has a unique ∆∂¯-harmonic representative, so there are isomorphisms
(M ) ∼ Hp,q (M ) ∼
M M
k
HdR = Hk (M ) = = H p,q (M ).
p+q=k p+q=k
C∼ 2
= HdR (Σ) = H 1,1 (Σ)
C2g ∼ 1
= HdR (Σ) = H 1,0 (Σ) ⊕ H 0,1 (Σ) ∼
= Cg ⊕ Cg since H 1,0 (Σ) ∼
= H 0,1 (Σ)
C∼ = H 0 (Σ) = H 0,0 (Σ).
dR
436
18.3. Kähler Manifolds Chapter 18. Complex Varieties
Similarly,
1
HdR (M ) = H 1,0 (M ) ⊕ H 0,1 (M )
= (H 1,0 (Σg ) ⊗ H 0,0 (Σh )) ⊕ (H 0,0 (Σg ) ⊗ H 1,0 (Σh ))
⊕ (H 0,1 (Σg ) ⊗ H 0,0 (Σh )) ⊕ (H 0,0 (Σh ) ⊗ H 0,1 (Σh ))
∼
= Cg ⊕ Ch ⊕ Cg ⊕ Ch .
Applying the symmetry described above, we get the following Hodge diamond for M :
1
g+h g+h
gh 2(gh + 1) gh
g+h g+h
1
437
18.4. Analytic Varieties Chapter 18. Complex Varieties
V ∩ U = V (f1 , . . . , fk )
in a neighborhood of (u1 , . . . , un ) and this power series converges on that neighborhood. One
useful consequence of this fact is that if f, g : U → C are holomorphic, U is connected and
f and g agree on some open subset of U , then f ≡ g on all of U . (This is sometimes called
the principle of analytic continuation.)
Definition. A Weierstrass polynomial on an open set U ⊆ Cn is a holomorphic function
g : U → C of the form
438
18.4. Analytic Varieties Chapter 18. Complex Varieties
where disc(g) is the discriminant of g. Therefore the branch locus of a hypersurface is itself
an analytic variety. For example, take n = 2 and g(z, w) = w2 − z. Then the projection
V → C, (z, w) 7→ z 2 gives the classic degree 2 branched cover of C with branch locus {0}.
439
18.4. Analytic Varieties Chapter 18. Complex Varieties
Corollary 18.4.8. If g ∈ OM,P is irreducible and h ∈ OM,P such that h ≡ 0 on V (g), then
g | h.
These results show that for any P ∈ M , OM,P more or less behaves like a polynomial
ring. This makes the connection to algebraic geometry even more explicit.
440
18.5. Divisors, Line Bundles and Sections Chapter 18. Complex Varieties
(2) If V1 , . . . , Vm are the connected components of Vsm , then their closures V 1 , . . . , V m are
the irreducible components of V .
Corollary 18.5.3. Every hypersurface V ⊆ M has a unique decomposition
V = V1 ∪ · · · ∪ Vm
441
18.5. Divisors, Line Bundles and Sections Chapter 18. Complex Varieties
where V (f ) = V ∩ U .
Lemma 18.5.4. For a hypersurface V ⊆ M and g ∈ OM (U ), the order of vanishing ordP (g)
is independent of the point P ∈ V ∩ U , that is, P →
7 ordP (g) is locally constant.
As a result, the order of vanishing ordV (g) of g along any hypersurface V ⊆ M is well-
defined.
Example 18.5.5. Let Σ be a Riemann surface. Then a divisor on Σ is just a formal linear
combination of points of Σ: X
D= nP P.
P ∈X
Example 18.5.6. The above situation is atypical, as many complex manifolds have no
divisors at all. However, projective manifolds, i.e. those admitting complex embeddings
M ,→ Pn , contain many hypersurfaces. For example, if V ⊆ Pn is a hypersurface then
M ∩ V is a hypersurface in M . Examples of such hypersurfaces in Pn can be found by taking
the vanishing set of any homogeneous polynomial f ∈ C[z0 , . . . , zn ].
Here, V (−z02 + z12 + z22 ) is called the divisor of zeroes of f sometimes written (f )0 , while
V (z1 z2 ) is called the divisor of poles and is denoted (f )∞ . Further, in this example V (z1 z2 )
is reducible (it’s the union of the z0 z2 - and z0 z1 -planes in P2 ) so we have
442
18.5. Divisors, Line Bundles and Sections Chapter 18. Complex Varieties
where ordV (f ) := ordV (gα ) − ordV (hα ) for a local expression f = hgαα on some open Uα ⊆ M
with Uα ∩ V 6= ∅. Write
X X
(f )0 = ordV (gα ) and (f )∞ = ordV (hα )
V ⊆M V ⊆M
(Here, OM (U )× denotes the nonvanishing holomorphic functions on U .) Set gαβ = ffαβ . Notice
that if gαβ = 1 on all overlaps, then the fα glue together to define a global meromorphic
function f such that D = (f ). However, in general we have gαβ gβγ gγα = 1 on all triple
overlaps Uα ∩ Uβ ∩ Uγ . Then the maps gαβ : Uα ∩ Uβ → C× are the transition maps of a line
bundle on M , denoted OM (D) or just O(D) if the underlying complex manifold is clear.
Definition. A holomorphic vector bundle on a complex manifold M is a holomorphic
map π : E → M which is a complex vector bundle such that the transition functions are
holomorphic.
Lemma 18.5.8. For any divisor D ∈ Div(M ), O(D) is a holomorphic line bundle.
Definition. The set of isomorphism classes of holomorphic line bundles on M , which is a
group under ⊗, is called the Picard group of M , written Pic(M ).
Theorem 18.5.9. For any complex manifold M , the assignment
Φ : Div(M ) −→ Pic(M )
D 7−→ O(D)
is a group homomorphism.
Proof. From the above description of O(D) using transition functions, it is clear that O(D1 +
D2 ) ∼
= O(D1 ) ⊗ O(D2 ) for any divisors D1 , D2 ∈ Div(M ).
443
18.5. Divisors, Line Bundles and Sections Chapter 18. Complex Varieties
fα2 fβ2
= 1 on Uα ∩ Uβ .
fα1 ◦ hα fβ ◦ hβ
n 2 o
This latter condition says that the collection f 1f◦h α
α
defines a global meromorphic function
α
f on M such that
X
D2 − D1 = (ordV (fα2 ) − ordV (fα1 ))V
V ⊆M
fα2
X
= ordV V
V ⊆M
fα1
fα2
X
= ordV V since hα is holomorphic
V ⊆M
fα1 ◦ hα
= (f ).
Uj = {[z0 , . . . , zn ] ∈ Pn | zj 6= 0}.
444
18.5. Divisors, Line Bundles and Sections Chapter 18. Complex Varieties
E = {(`, v) | ` ∈ Pn , v ∈ `} ⊂ Pn × Cn+1 ,
called the tautological line bundle. Over the same open cover {Uj }, E is trivialized by
ϕ−1j : Uj −→ E|Uj
z0 zn
([z0 , . . . , zn ], λ) 7−→ [z0 , . . . , zn ], λ , . . . , 1, . . . ,
zj zj
λ
= [z0 , . . . , zn ], (z0 , . . . , zn ) .
zj
Then the transition maps for E are:
λ
(ϕj ◦ ϕ−1
k )([z0 , . . . , zn ], λ) = ϕj [z0 , . . . , zn ], (z0 , . . . , zn )
zk
λzj
= [z0 , . . . , zn ],
zk
E z
i.e. gjk = zkj , the inverse of the transition map on Uj ∩ Uk for O(H). This proves that
E∼ = O(H)∗ = O(−H).
Definition. For a holomorphic vector bundle π : E → M , a holomorphic section (resp.
meromorphic section) of π is a holomorphic (resp. meromorphic) function s : M → E
such that π ◦ s = idM . The group of all holomorphic (resp. meromorphic) sections of π is
denoted Γh (M, π) or Γh (M, E) (resp. Γ(M, π) or Γ(M, E)).
Locally, a holomorphic section s ∈ Γh (M, π) is determined by sections sα : Uα → E|Uα
such that sα = gαβ sβ on each overlap Uα ∩ Uβ .
P
Example 18.5.13. Suppose D = nV V ∈ Div(M ) is a divisor with local defining functions
fα on Uα . By definition, the transition maps of O(D) are gαβ = ffαβ on Uα ∩ Uβ so the fα glue
together to give a global section fD of O(D). If D is an effective divisor, fD is a holomorphic
section of O(D), but in general fD need only be meromorphic. Conversely, if L → M is a
line bundle and s : M → L is a nonvanishing meromorphic section, there is a natural divisor
associated to the pair (L, s), namely
X
(s) = ordV (s)V.
V ⊆M
445
18.5. Divisors, Line Bundles and Sections Chapter 18. Complex Varieties
as in Section 17.3.
Proposition 18.5.16. Let L = O(D) be the line bundle defined by a divisor D ∈ Div(M )
and let sD be a meromorphic section of L defined by a set of local defining functions fα for
D. Then there is a one-to-one correspondence
L(D) ←→ Γh (M, L)
f 7−→ f sD .
Example 18.5.17. For M = Pn and L = O(H) the hyperplane bundle, this correspondence
gives
Γ(Pn , O(H)) ∼
= L(H) = {f ∈ M(Pn ) | f has a pole of order ≤ 1 along H}.
p(z0 , . . . , zn )
f (z0 , . . . , zn ) =
z0
Γ(Pn , O(dH)) ∼
= Symd (Cn+1 )∗ .
446
18.6. Cohomology and Chern Classes Chapter 18. Complex Varieties
with differential
where Ui0 ,...,ip denotes the intersection Ui0 ∩ · · · ∩ Uip . We will use the fact that for any sheaf
0
F on X and any open cover U of X, Ȟ (U, F ) = F (X) = H 0 (X, F ) (Lemma 13.4.5) and
1
Ȟ (X, F ) → H 1 (X, F ) is always an isomorphism (Corollary 16.2.7).
Example 18.6.1. If M is a complex manifold, for each open U ⊆ M let M(U ) be the
vector space of meromorphic functions on U . Then U 7→ M(U ) defines a sheaf on M and
H 0 (M, M) = M(M ) by the above. Moreover, there is an injective morphism of sheaves
×
OM ,→ M and we have
Div(M ) ∼
= H 0 (M, M/O× ) and Pic(M ) ∼ ×
= H 1 (M, OM ).
This in fact coincides with the long exact sequence in sheaf cohomology coming from the
short exact sequence of sheaves
× ×
0 → OM → M → M/OM → 0.
Recall that by de Rham’s theorem (Corollary 13.3.8), the singular cohomology of any real
manifold can be computed by de Rham cohomology. Dolbeault’s theorem is an analogous
statement for complex manifolds.
Theorem 18.6.2 (Dolbeault). For any complex manifold M and every p, q ≥ 0, there is an
isomorphism
H p,q (M ) ∼
= H q (M, Ωp ).
Consider the short exact sequence of sheaves
× exp
0→Z→
− OM −−→ OM →0
447
18.6. Cohomology and Chern Classes Chapter 18. Complex Varieties
where Z is the constant sheaf and exp denotes the exponential map f 7→ e2πif . This induces
a long exact sequence in sheaf cohomology:
× δ
· · · → H 1 (M, OM ) → H 1 (M, OM − H 2 (M, Z) → · · ·
)→
×
Again note that H 1 (M, OM ) = Pic(M ).
Definition. The (first) Chern class of a holomorphic line bundle L on M is the cohomology
×
class c1 (L) := δ([L]) ∈ H 2 (M, Z), where [L] denotes the class in H 1 (M, OM ) corresponding
to L.
The fact that the connecting map δ in a long exact sequence is a homomorphism implies
the following basic facts about Chern classes.
The same arguments define the first Chern class of any smooth (i.e. topological) line
bundle on M . Explicitly, replacing OM with the sheaf A of smooth functions on M , there
is also a short exact sequence of sheaves
exp
− A −−→ A× → 0.
0→Z→
Then H 1 (M, A× ) may be identified with the set of isomorphism classes of smooth line bundles
on M and for any such bundle L, we take c1 (L) to be the class δ([L]) ∈ H 2 (M, Z).
Proposition 18.6.4. For any holomorphic line bundle L → M , c1 (L) only depends on the
underlying smooth line bundle of L and this is classified by its first Chern class.
id
exp
0 Z A A× 0
448
18.6. Cohomology and Chern Classes Chapter 18. Complex Varieties
× δ
H 1 (M, OM ) H 2 (M, Z)
id
∼
0 = H 1 (M, A) H 1 (M, A× ) H 2 (M, Z) H 2 (M, A)
449
18.6. Cohomology and Chern Classes Chapter 18. Complex Varieties
χh (M ) = χh (M, OM ).
0 → E ⊗ O(−V ) → E → E|V → 0.
450
18.6. Cohomology and Chern Classes Chapter 18. Complex Varieties
Alternatively, this may be written χh (Σ, L) = deg(L) − g + 1 for any line bundle L ∈
Pic(Σ). We might want to know more about the dimensions h0 (L) and h1 (L) for a given
line bundle, but it turns out that these are not determined solely by the topology of L → Σ.
When M is compact, a version of Serre duality (Theorem 17.6.2) tells us that h1 (L) may be
computed as h0 of a different line bundle.
H p (M, E) ∼
= H n−p (M, E ∗ ⊗ O(KM ))∗
where (−)∗ denotes vector space dual and KM is a distinguished divisor on M which is defined
up to linear equivalence.
Here’s a sketch of the proof; along the way we will define KM ∈ Div(M ). Compare this
to its counterpart in Section 17.6. For any vector bundle E → M , define the vector spaces
^ ^
k k ∗ p,q p,q ∗
Ω (M, E) := Γ M, T M ⊗E and Ω (M, E) := Γ M, T M ⊗E .
Ωk (M, E) ∼
M
= Ωp,q (M, E).
p+q=k
α ∧ ?β = hα, βi · d volM
for any α, β ∈ Ωp,q (M ) and volM a Hermitian volume form on M . Locally, given a Hermitian
n-frame ε1 , . . . , εn for T 1,0 M with dual basis ε̄1 , . . . , ε̄n for T 0,1 M , the operator ? acts by
?(ε1 ∧ ε2 ) = ε3 ∧ ε̄1 ∧ ε̄2 ∧ ε̄3 , ?(ε2 ∧ ε̄1 ) = −ε1 ∧ ε3 ∧ ε̄2 ∧ ε̄3 , and so on.
451
18.6. Cohomology and Chern Classes Chapter 18. Complex Varieties
For nontrivial E, this local definition extends to the whole bundle using the Hermitian metric
on E.
Now by the theory of harmonic functions (see Section 18.3 and the proof of Hodge
decomposition), we may identify H p,q (M, E) with the space Hp,q (M, E) of harmonic functions
with values in E. Explicitly, α ∈ Hp,q (M, E) if and only if ∆E ∂¯
α = 0, where
∆E ¯ ¯∗ ¯∗ ¯ • •
∂¯ = ∂ ∂ + ∂ ∂ : Ω (M, E) −→ Ω (M, E)
as in the proof of Hodge decomposition. Let ∆∂¯ be the Laplace operator on Ωbullet (M ).
H p (M, E) ∼
= H 0,p (M, E) by Dolbeault’s theorem
∼
= H0,p (M, E) by the proof of Hodge decomposition
∼
= Hn,n−p (M, E ∗ )∗ by the Lemma
^ ∗
∼
=H n,n−p ∗ ∗ ∼
(M, E ) = H n−p
M, n ∗
T M ⊗E ∗
.
Definition.
Vn ∗ ∼ For a compact complex manifold M of dimension n, any divisor KM such that
T M = O(KM ) is called a canonical divisor on M .
Example 18.6.18. When Σ is a Riemann surface and L → Σ is a line bundle, Serre duality
says that
H 1 (Σ, L) ∼
= H 0 (Σ, L∗ ⊗ O(KΣ )).
452
18.6. Cohomology and Chern Classes Chapter 18. Complex Varieties
Note that deg(KΣ ) = 2g − 2 is precisely −χ(Σ) where χ denotes the topological Euler
characteristic. To see this directly, let e(Σ) = e(T Σ) be the Euler class of Σ and observe
that O(KΣ ) = T ∗ Σ, so we have
−χ(Σ) = −he(T Σ), [Σ]i = h−c1 (T Σ), [Σ]i = hc1 (T ∗ Σ), [Σ]i = hc1 (KΣ ), [Σ]i = deg(KΣ ).
453
18.6. Cohomology and Chern Classes Chapter 18. Complex Varieties
First note that B|L| = ∅ if and only if for all P ∈ Σ, there is a section s ∈ H 0 (Σ, L) such
that s(P ) 6= 0. Equivalently, we are asking if the natural map H 0 (Σ, L) → H 0 (P, L|P ) is
surjective for all P . Consider the short exact sequence of sheaves
0 → IP → L → L|P → 0.
0 → IP,Q → L → L|{P,Q} → 0
where IP,Q now denotes the sheaf of sections of L vanishing at P and Q simultaneously.
Also note that H 0 ({P, Q}, L|{P,Q} ) ∼= H 0 (P, L|P ) ⊕ H 0 (Q, L|Q ) = L(P ) ⊕ L(Q). Then ϕ|L|
is one-to-one if and only if H 0 (Σ, L) → L(P ) ⊕ L(Q) is onto and to show this condition, it
suffices to prove H 1 (Σ, IP,Q ) = 0. By Serre duality,
Thus we see that if deg(L) > 2g, deg(KΣ + P + Q) − deg(L) = 2g − deg(L) < 0 which implies
H 1 (Σ, IP,Q ) = 0.
Finally, we might ask when ϕ|L| is an immersion, i.e. when dP ϕ|L| : TP Σ → Tϕ|L| (P ) PN is
an injective linear map for all P ∈ Σ. Dualizing, this is the same as dP ϕ∗|L| : Tϕ∗|L| (P ) PN →
TP∗ Σ being surjective. For a given P ∈ Σ, we may choose a basis {s0 , . . . , sN } for H 0 (Σ, L)
with s0 (P ) 6= 0 and s1 (P ) = · · · = sN (P ) = 0. Then in a neighborhood of P , ϕ|L| is given
by
sj
ϕ|L| (Q) = (t1 (Q), . . . , tN (Q)) ∈ CN ⊂ PN where tj = .
s0
Then the differential is given by (dtj ) so the adjoint is the row vector dP ϕ∗|L| = dt1 · · · dtN .
Thus dP ϕ∗|L| is surjective precisely when for any α ∈ TP∗ Σ, there exist a section s ∈ H 0 (Σ, IP )
such that ds = α. Consider the exact sequence of sheaves
d
0 → L ⊗ O(−2P ) → − (T ∗ Σ ⊗ L)|P → 0
− L→
454
18.6. Cohomology and Chern Classes Chapter 18. Complex Varieties
Proof. Let g be the genus of Σ and take L → Σ to be a line bundle with degree deg(L) > 2g.
Then the above argument shows that the complete linear system |L| is basepoint free and
hence ϕ|L| : Σ → PN is defined, where h0 (L) = N + 1. Moreover, since deg(L) > 2g, the
above also shwos ϕ|L| is an embedding.
We have seen that for a line bundle L on a Riemann surface Σ, the conditions that ϕ|L|
is well-defined and an embedding are related to the vanishing of H 1 (Σ, −) for various line
bundles built from L. Since dim Σ = 1, there is a nice topological criterion for the vanishing of
such cohomology groups: the degree of those line bundles need only be sufficiently positive
(this is essentially because irreducible divisors are merely points). This nice situation is
special to the dimension 1 case.
If M is a complex manifold of dimension greater than 1, there still exists a criterion for
1
H (M, L) to vanish for a particular line bundle L → M , but it is geometric and not merely
topological in nature. To introduce this criterion, called positivity of a line bundle, we first
discuss the theory of connections on topological vector bundles.
455
Part IV
Algebraic Groups
456
Chapter 19
Introduction
The notes in Part IV come from two courses on Lie groups taught by Dr. Brian Parshall
at the University of Virginia in the fall of 2015 (Lie groups and algebraic groups) and the
spring of 2018 (finite groups of Lie type). The topics covered are:
The primary texts for this course are Procesi’s Lie Groups: An Approach through Invariants
and Representations, T.A. Springer’s Linear Algebraic Groups, Malle-Testerman’s Linear Al-
gebraic Groups and Finite Groups of Lie Type, Geck’s An Introduction to Algebraic Geometry
and Algebraic Groups and Jacobson’s Basic Algebra I.
Given a group G, we have many ways of assigning further structure to study its connection
to different areas of mathematics:
(4) As an algebraic variety – these are algebraic groups, which are covered in Chapters 20
and 21;
457
19.1. Chevalley’s Theorem Chapter 19. Introduction
To prove the theorem, we first need the following lemma, which we won’t prove.
f
F
k
Chevalley’s Theorem says that the image of any dominant morphism of varieties contains
an open subset of the codomain.
458
Chapter 20
Algebraic Groups
2 The nonzero elements of the field, k × , form an affine algebraic group under multiplica-
tion, denoted Gm . In this case, the coordinate ring is k[G] = k[t, t−1 ], a one-dimensional
torus. Note that dim Ga = dim Gm = 1.
3 Given affine groups G and H, G×H is an affine group so products exist in the category
AlgGroups. We could take Ga × · · · × Ga = Gna which is an affine group constructed
| {z }
n
from the underlying space An . Similarly, Gm × · · · × Gm = Gnm is an affine group called
| {z }
n
the n-dimensional torus.
4 The general linear group GLn (k) is an affine group. Here
A = k[GLn (k)] = k[Tij | 1 ≤ i, j ≤ n][T ]/(1 − T D)
where D is the determinant function and (1 − T D) is the ideal generated by 1 − T D.
5 The special linear group SLn (k) is an affine group with coordinate algebra A =
k[Tij ]/(D − 1). We can continue defining algebraic groups for the orthogonal linear
group, projective general/special linear groups, etc.
Later we will prove:
Theorem 20.0.1. Every irreducible one-dimensional affine algebraic group G is isomorphic
to either Ga or Gm .
459
20.1. Hopf Algebras Chapter 20. Algebraic Groups
m × 1G
G×G×G G×G
1G × m m
G×G m G
∆⊗1
A ⊗k A ⊗k A A ⊗k A
1⊗∆ ∆
A ⊗k A A
∆
Examples.
1 Recall that Ga is the additive group k as an algebraic group. Its coordinate algebra is
460
20.1. Hopf Algebras Chapter 20. Algebraic Groups
∆ : A −→ A ⊗k A
T 7−→ T ⊗ 1 + 1 ⊗ T ;
γ : A −→ A
T 7−→ −T ;
and ε : A −→ k
T 7−→ 0.
∆ : A −→ A ⊗k A
T 7−→ T ⊗ T
T 7−→ T −1 ⊗ T −1 ;
−1
γ : A −→ A
T 7−→ T −1 ;
and ε : A −→ k
T 7−→ 1.
3 For GLn (k), the Hopf algebra structure consists of ∆ : A → A ⊗k A which maps
Tij 7→ n`=1 Ti` ⊗ T`j and T 7→ D−1 ⊗ D−1 ; and γ : Tij 7→ D−1 adjTji by Cramer’s Rule,
P
461
20.2. Topological Properties Chapter 20. Algebraic Groups
(1) G has a unique irreducible component which is closed and contains e, the identity.
This component is denoted G0 .
(2) G0 is a normal subgroup of finite index in G and the other irreducible components of
G are the cosets aG0 .
G×m G
The map along the diagonal is continuous, so it takes irreducible components to irreducible
components. Thus Y1 · · · Ym is irreducible in G and so is its closure Y1 · · · Ym . However
Yi ⊆ Y1 · · · Ym for each Yi , but Yi is irreducible. This forces m = 1, so there is a single
irreducible component containing the identity.
(2) For G0 ≤ G, note that the multiplication and inverse maps are homeomorphisms, so
they take a 7→ e and G0 → G0 ; therefore G0 is closed under these operations. Since inner
automorphisms are also homeomorphisms, we can see that for any x ∈ G, xG0 x−1 = G0
so G0 is a normal subgroup. Moreover, since the components of G are mutually disjoint, it
follows that these must be the cosets xG0 .
(3) follows from the uniqueness part of (1), and the fact that morphisms map irreducible
components into irreducible components.
Examples.
1 GLn (k) is connected (irreducible) because GLn (k) = GLn (k)0 . This can be seen from
the fact that the ideal (1 − T D) is irreducible in k[Tij , T ], and since the coordinate al-
gebra for GLn (k) is A = k[Tij , T ]/(1 − T D), irreducibility follows. The most important
subgroup, SLn (k), is also irreducible.
2 Recall the orthogonal group On (k) = {g ∈ GLn (k) | g T g = 1}. An important subgroup
is the special orthogonal group SOn (k) = {g ∈ On (k) | det g = 1}. It turns out that
SOn (k) = On (k)0 so the special orthogonal group is a connected component of On (k).
462
20.2. Topological Properties Chapter 20. Algebraic Groups
From this we define the nth symplectic group Spn (k) = {g ∈ GLn (k) | g T Jg = J}.
The algebraic groups in 1 through 3 comprise what are known as the classical groups.
Let G be an affine group and take x ∈ G. The left multiplication map g 7→ xg is in fact
a homeomorphism of topological spaces G → G. In general, how might we show that a map
between algebraic groups is continuous?
Definition. The map ϕ is continuous if ϕ−1 (V(J)) is closed in X for every ideal J ⊂ k[Y ].
Notice that in the case of left multiplication ϕ(g) = xg, we have a commutative diagram
ϕ∗
k[G] k[G]
g ϕ(g)
where g and ϕ(g) are viewed as maps k[G] → k. Then for an ideal J ⊂ k[G], ϕ(g) ∈ V(J) if
and only if g vanishes on the ideal of k[G] generated by ϕ∗ (J). Hence ϕ−1 (J) is closed in G
and we see that left multiplication is continuous. Similar proofs show the continuity of right
multiplication, conjugation, etc.
Lemma 20.2.2. If U and V are dense, open subsets of an affine group G then G = U V .
Proof. (a) Let x ∈ H. Then H = xH ⊆ xH. This shows that H ⊆ xH, or x−1 H ⊆ H.
Therefore HH ⊆ H. So if h̄ ∈ H then H h̄ ⊆ H. Now right multiplication is continuous so
it maps the closure of H h̄ into H, meaning H h̄ ⊆ H. Since h̄ ∈ H was arbitrary, HH ⊆ H.
Hence H is closed under multiplication. Inverses are proven similarly. Finally, one can check
that multiplication and inverses on H are the restrictions of their counterparts on G, thus
they are variety morphisms. Hence H ≤ G and H is an affine algebraic group.
463
20.2. Topological Properties Chapter 20. Algebraic Groups
5 Let G = Spn (k), the symplectic group. The matrix J gives us a k-vector space V of
dimension n = 2m. Given a nondegenerate, alternating bilinear form b on V ,
b : V × V −→ k,
for every nonzero vector v ∈ V , there exists a u ∈ V such that b(u, v) 6= 0 and
b(u, u) = 0. Whenever char k 6= 2, b(u, v) = −b(v, u) for all u, v ∈ V . Thus
For each u ∈ V and c ∈ k, define τu,c ∈ Spn (k) by τu,c = x − cb(x, u)u; these are called
symplectic transvection. For each u, this corresponds to a morphism
464
20.2. Topological Properties Chapter 20. Algebraic Groups
Theorem 20.2.7. Let G be an affine algebraic group and suppose {Xi , ϕi }i∈I is a collection
of irreducible varieties Xi and morphisms ϕi : Xi → G such that e ∈ Yi := ϕi (Xi ) for every
i. Let H be the closed subgroup generated by the Yi , i.e. H = hYi ii∈I . Then there exist
elements a(1), a(2), . . . , a(m) ∈ I and a choice ε(h) = ±1 for each h = 1, 2, . . . , m such that
ε(1) ε(2) ε(m)
H = Ya(1) Ya(2) · · · Ya(m) =: Yaε .
τ : Xa(1) × · · · × Xa(m) −→ G
(xa(1) , . . . , xa(m) ) 7−→ ϕa(1) (xa(1) ) · · · ϕa(m) (xa(m) ).
4 (continued) Let G = GLn (k) again. Recall the maps ϕij (λ) for i, j = 1, . . . , n that are
homomorphisms A1 → G. Notice that when i < j, ϕji (λ) = ϕij (λ)T . The ϕij (λ) gen-
erate GLn (k) and from this we can deduce that H = GLn (k) so GLn (k) is irreducible
(connected).
Corollary 20.2.8. Let {Gi }i∈I be a family of closed, connected subgroups of an affine group
G. Then the subgroup generated by the Gi , hGi ii∈I is closed and connected.
Corollary 20.2.9. Let H, K ≤ G with H closed and connected. Then the commutator
subgroup [H, K] := hhkh−1 k −1 | h ∈ H, k ∈ Ki is closed and connected.
Proof. Let I = K and consider the collection {Hx , ϕx }x∈K where Hx = xHx−1 and ϕx :
Hx → G, xhx−1 7→ hxh−1 x−1 . It is clear that the images im ϕx over all x ∈ K generate
[H, K], so apply Theorem 20.2.7 to deduce that [H, K] is closed and irreducible.
465
20.3. G-spaces Chapter 20. Algebraic Groups
20.3 G-spaces
Let G be an affine algebraic group over k.
(2) e · x = x.
Examples.
1 Take X = G. Then G acts on itself in several ways: (g, x) 7→ gx (the left action);
(g, x) 7→ xg −1 (the right action); (g, x) 7→ gxg −1 (conjugation). These all make G into
a G-space.
2 Let V be a finite dimensional vector space over k and suppose G acts on V by linear
transformations, (g, v) 7→ g · v, so that the G-space axioms hold. Then there exists a
homomorphism
ϕ : G −→ GL(V ) ∼ = GLn (k).
(Here the isomorphism depends on choice of basis.) We say the mapping G → GLn (k)
is rational (formally, a rational representation) if it is a morphism of algebraic groups.
In this case, V is said to be a rational G-module and this structure makes V into a
G-space. However, not every G-space is a rational module since the G-action might
not be linear.
To expand this notion a bit, suppose V is a (possibly infinite dimensional) k[G]-module.
We say V is a rational G-module if it is the union of finite dimensional rational G-
submodules as defined above. Equivalently, V is rational if every v ∈ V lies in a finite
dimensional rational submodule.
Every G-module V is a comodule over the coordinate ring A = k[G]: there is a linear
map
∆V : V −→ V ⊗k A
such that (1 ⊗ ∆)∆V = (∆V ⊗ 1)∆V and (1 ⊗ ε)∆V = idV for ∆ the comultiplication of
A and ε the counit. For example, A = k[G] is a rational G-module from this viewpoint.
Pn∆A = ∆, 0we see that coassociativity holds, and for f ∈ A, g ∈ G such that
By setting
∆(f ) = i=1 (fi ⊗ fi ), we set
n
X
g·f = fi fi0 (g).
i=1
466
20.3. G-spaces Chapter 20. Algebraic Groups
F : 0 = V0 ⊆ V1 ⊆ · · · ⊆ Vm = V.
is another flag in V . This action makes Flag(V ) into a G-space. An important example
of this object is when G = GLn (k). Let V be the n-dimensional vector space over k
with standard basis {e1 , . . . , en } and set Vi = Span{e1 , . . . , ei } for each i = 1, . . . , n.
Then F : 0 = V0 ⊆ V1 ⊆ · · · ⊆ Vn = V is a complete flag in V and StabG (F) = B, the
subspace of upper triangular matrices in GLn (k). Hence we can view the flag variety
as a quotient: Flag(V ) ∼= G/B.
The following is perhaps the most important theorem in the study of G-spaces. We will
prove this in Section 12.11.
Examples.
467
20.3. G-spaces Chapter 20. Algebraic Groups
Then G also acts on Xn by conjugation, but what are the orbits this time? An n × n
matrix x is called unipotent if 1−x is nilpotent. Let Yn = {x ∈ Mn (k) | x is unipotent}.
There is an obvious bijection Xn ↔ Yn . Observe that Xn is closed in Mn (k) since the
points are given by a polynomial expression on their components. So Xn is an affine
algebraic variety and consequently so is Yn . Now x ∈ Xn if and only if x is similar to
an upper diagonal matrix with 0’s along the diagonal.
Next define
1
1
*
Un = ∈ Mn (k) ≤ GLn (k).
...
0
1
Theorem 20.3.3. Let H be a closed, irreducible subgroup of GLn (k) consisting entirely
of unipotent matrices. Then there exists a g ∈ GLn (k) such that gHg −1 ≤ Un .
Proof. Let’s induct on n. If n = 1, the only 1×1 unipotent matrix is [1] so the base case
is trivial. Suppose there is some n ∈ N where for all m < n, there exists a g ∈ GLm (k)
such that gHg −1 ≤ Um . View GLn (k) as GL(V ) where V is the vector space of n × 1
column vectors. We may assume the action of H on V is irreducible; otherwise there
is a subspace 0 6= W ( V such that W is stabilized by H, and thus for some g ∈ G,
" #
gHg −1 =
W *
0 V /W
In this case dim W and dim V /W are both strictly less than n so we’re done by induc-
tion.
Now, assuming V is an irreducible H-space, consider the subalgebra E of Endk (V )
spanned by the elements of H. Then by Wedderburn-Artin theory, E = Endk (V ). For
g, h ∈ H, tr(g) = tr(h) = n = dim V . Hence tr((1 − g)h) = tr(h) − tr(gh) = n − n = 0.
Fixing g ∈ H and letting h vary, we can write all elements of E as linear combinations
of (1 − g)h. Then tr((1 − g)r) = 0 for all r ∈ Endk (V ). This means 1 − g = 0, so
H = {e}.
468
20.3. G-spaces Chapter 20. Algebraic Groups
469
20.3. G-spaces Chapter 20. Algebraic Groups
Pn Pn
Thus ∆(fi ) = j=1 fj ⊗ gji so we see that ρ(g)fi = j=1 gji (g)fj ∈ V . This implies that
ρ(G)V = V .
The following result is critical in the general theory of affine algebraic groups.
It follows that the map ϕ : G → GLn (k), g 7→ (aij (g)), is a group homomorphism and hence
a morphism of algebraic groups. The comorphism for ϕ is
ϕ∗ : k[GLn (k)] −→ A
p 7−→ p ◦ ϕ.
∗
Pn xij ∈ k[GLn (k)] be the ijth entry of p. Then ϕ (xij ) = aij ∈ A, but fj (g) = (ρ(g)fj )(e) =
Let
i=1 aij (g)fi (e). Thus
Xn n
X
fj = fi (e)aij = fi (e)ϕ∗ (xij ).
i=1 i=1
470
20.4. Jordan Decomposition Chapter 20. Algebraic Groups
where λi are the distinct eigenvalues of a. We will need the following lemmas.
Lemma 20.4.1. If a, b ∈ Endk (V ) are semisimple and commute, then a + b and ab are
semisimple, and a and b can be simultaneously diagonalized.
Lemma 20.4.2. If a and b are semisimple, commuting, nilpotent (resp. unipotent) operators
then a + b and ab are also semisimple, commuting and nilpotent.
By the Chinese remainder theorem, there exists a polynomial p(T ) satisfying p(T ) ≡ 0
mod T and p(T ) ≡ ai mod (T − λi )ni for all i =L1, . . . , m. Define as = p(a) and an = a − as .
m
We can decompose V as a direct sum V = i=1 Vi where a(Vi ) ⊆ Vi and cha|Vi (T ) =
(T − λi )ni . (In other words, Vi is a generalized eigenspace for λi .) We claim that as is
semisimple and an is nilpotent, as an = an as , and a = as + an is a unique decomposition of
a having these properties.
Suppose a = a0s + a0n for a0s , a0n satisfying the above conditions. Then as + an = a0s + a0n .
Note that a commutes with a0s and a0n , so a0s commutes with p(a) = as ; likewise a0n commutes
with a − p(a) = an . Now subtract: as − a0s = a0n − an . By Lemma 20.4.1, any two commuting
semisimple operators can be simultaneously diagonalized, but this diagonalization turns
a0n − an into 0. Hence as − a0s = 0 = a0n − an so a0s = as and a0n = an .
Definition. The unique choice of semisimple as and nilpotent an such that a = as + an is
called the Jordan decomposition of a.
We have thus proved the following theorem.
Theorem 20.4.3 (Jordan Decomposition). Every linear operator a ∈ Endk (V ) on a finite
dimensional vector space V has a unique Jordan decomposition a = as + an where as is
semisimple and an is nilpotent.
Corollary 20.4.4. Let a ∈ GL(V ) for a finite dimensional vector space V . Then there is a
unique factorization a = as au = au as , where as is semisimple and au is unipotent.
Proof. Write a = as + an by the Jordan decomposition theorem. Since as an = an as , a−1 s an
−1 −1
is also nilpotent so we can write a = as + an = as (1 + as an ) = as au where au = 1 + as an
is unipotent.
We next develop some of the properties of the Jordan decomposition.
Proposition 20.4.5. Let a ∈ Endk (V ) for a finite dimensional vector space V and suppose
W ⊂ V is a subspace which is a-invariant.
471
20.4. Jordan Decomposition Chapter 20. Algebraic Groups
Proposition 20.4.6. Let V, W be finite dimensional vector spaces over k and suppose a ∈
Endk (V ), b ∈ Endk (W ) and ϕ ∈ Homk (V, W ) such that the following diagram commutes:
ϕ
V W
a b
V ϕ W
Then the semisimple and nilpotent parts of a and b also make the diagram commute:
ϕ ϕ
V W V W
as bs an bn
V ϕ W V ϕ W
Proof. Proposition 20.4.5 implies that the property holds in the special case that ϕ is an
injection or a surjection. In general, every morphism ϕ : V → W can be factored through
the direct sum:
σ π
V −−−→ V ⊕ W −−−→ W
v 7−→ (v, ϕ(v))
(v, w) 7−→ w.
Corollary 20.4.7. Let a ∈ Endk (V ) and b ∈ Endk (W ) be linear operators and consider the
linear operators
a ⊕ b : V ⊕ W −→ V ⊕ W
(v, w) 7−→ (a(v), b(w));
and a ⊗ b : V ⊗k W −→ V ⊗k W
v ⊗ w 7−→ a(v) ⊗ b(w).
472
20.4. Jordan Decomposition Chapter 20. Algebraic Groups
(2, 3)
∆V ⊗k W (V ⊗k W ) ⊗k (A ⊗k A)
idV ⊗ m
(V ⊗k W ) ⊗k A
473
20.4. Jordan Decomposition Chapter 20. Algebraic Groups
ϕ
V W
a b
V ϕ W
commutes, then as in the finite dimensional case, the following diagrams all commute:
ϕ
V W
as , an , au bs , bn , bu
V ϕ W
(1) There exist unique elements gs , gu ∈ G that commute such that ρ(g) has Jordan de-
composition ρ(g) = ρ(g)s ρ(g)u = ρ(gs )ρ(gu ).
(3) If G = GLn (k) then for every g ∈ G, gs and gu are the semisimple and unipotent parts
of g.
Proof. (1) Let m denote the multiplication map A ⊗k A → A, where A = k[G]. Then the
following diagram commutes:
m
A⊗k A
A⊗k A
m
(ρ(g)s f )(x) = (λ(x−1 )ρ(g)s f )(e) = (ρ(g)s λ(x−1 )f )(e) since λ, ρ commute
= (λ(x−1 )f )(gs ) by the defining property of gs
= f (xgs ) = (ρ(gs )f )(x).
474
20.4. Jordan Decomposition Chapter 20. Algebraic Groups
Since they agree at every x ∈ A, ρ(g)s and ρ(gs ) must be the same operator. The proof is
the same for ρ(gu ). Then we have ρ(g) = ρ(g)s ρ(g)u = ρ(gs )ρ(gu ) = ρ(gs gu ) = ρ(gu gs ) so the
elements gs , gu must commute.
To see that gs and gu are unique, suppose gs0 , gu0 ∈ G such that g = gs0 gu0 and ρ(gs0 ) =
ρ(g)s , ρ(gu0 ) = ρ(g)u . Let V be a G-space. Then for any nonzero f ∈ V ∗ = Homk (V, k),
the map f˜ : V → A sending v to f˜(v)(g) = f (gv) is an injective G-module homomorphism.
Indeed, if f˜ = 0 then f (gv) = 0 for all g ∈ G, v ∈ V which implies v = 0, so f˜ is injective.
On the other hand, for g, h ∈ G, v ∈ V , f˜(hv)(g) = f (ghv) = f˜(v)(gh) = (ρ(h)f˜(v))(g) so
f˜(hv) = ρ(h)f˜(v) and thus f˜ is a map of G-modules. For all g ∈ G, the diagram
f˜
V A
g ρ(g)
V A
f˜
commutes, so by the work above, gs0 and ρ(gs0 ) also make the diagram commute, as do gu0
and ρ(gu0 ). Since ρ(gs0 ) = ρ(g)s = ρ(gs ) and ρ(gu0 ) = ρ(g)u = ρ(gu ) and f ∈ V ∗ was arbitrary,
this identifies gs = gs0 and gu = gu0 .
(2) Given ϕ : G → G0 , ϕ is surjective onto im ϕ, which is a closed subgroup of G0 by (2)
of Theorem 20.2.4. Then we can factor
ϕ : G im ϕ ,→ G0 .
475
20.4. Jordan Decomposition Chapter 20. Algebraic Groups
Theorem 20.4.14. Let G be a commutative affine algebraic group and let Gs (resp. Gu )
be the subset of semisimple (resp. unipotent) elements of G. Then Gs and Gu are closed
subgroups of G and Gs × Gu ∼ = G. Further, if G is irreducible then so are Gs and Gu .
Proof. Assume G is a closed subgroup of some GLn (k). View GLn (k) as GL(V ) where
V is the vector space k n . Then x ∈ GL(V )u ⇐⇒ (x − 1)m = 0 for some m ≥ 1.
Therefore GL(V )u is closed since the elements satisfy polynomial equations, so in particular
Gu = G ∩ GL(V )u is closed in G. If x, y ∈ Gu then x − 1, y − 1 are both nilpotent and
xy − 1 = xy − y + y − 1 = (x − 1)y + y − 1
(3) If G = Gu then any element x ∈ G has finite order dividing char k. (When char k = 0,
this is interpreted to mean that G has no elements of finite order.)
Proof. (1) Let g ∈ G and define the morphism
ϕ : G −→ G
x 7−→ xgx−1 .
476
20.4. Jordan Decomposition Chapter 20. Algebraic Groups
over k has finite image consisting of the characteristic polynomials of g and the finitely many
characteristic polynomials of the elements of G r im ϕ. Since G is connected (irreducible),
the image of χ is a finite, irreducible subset of k[t], hence one point, so it must be that χ(G)
is a single element, which must be (x − 1)n since this is the characteristic polynomial of the
identity. This shows every element of G is unipotent, so by Theorem 20.3.3, G is similar to
a subgroup of GLn (k) consisting of upper triangular unipotent matrices. In particular, G
is solvable, so the derived subgroup G0 = [G, G] is a proper subgroup of G. We know from
Corollary 20.2.9 that G0 is closed and connected, but since G is one-dimensional, this means
G0 = {e}. Thus g −1 ϕ(x) = g −1 xgx−1 ∈ G0 = {e} for all x ∈ X, contradicting the hypothesis
that im ϕ = G. Hence im ϕ = {g} which implies that G is commutative.
(2) Theorem 20.4.14 shows that G ∼ = Gs × Gu but since G is connected, each of these
components is connected. Moreover, 1 = dim G = dim Gs + dim Gu so it must be that either
G = Gu or G = Gs .
(3) Assume G = Gu and char k = p > 0. Let x ∈ G be a nonidentity element. For
h h
some h ∈ N, define G(p ) ≤ G to be the image of the map G → G, y 7→ y p , which is a
h
morphism of affine algebraic groups since G is abelian. Then G(p ) is a closed subgroup of G
h h
which is connected. Since G is one-dimensional, either G(p ) = {e} or G(p ) = G. Suppose
ph > n. Since G = Gu , each x ∈ G may be written as x = 1 + λ for a nilpotent element λ,
h h h h
so xp = (1 + λ)p = 1 + λp = 1. Thus when ph > n, G(p ) = {e}. If G(p) = G then for all
h h−1 h
h > 0, G(p ) = (G(p ) )(p) but this contradicts the fact that G(p ) = {e} for large enough h.
Therefore G(p) = {e} so every element of G has order dividing the characteristic of k.
Example 20.4.16. Let G = GLn (k). Define the following subgroups of G:
1 ∗
∗ ∗
∗ 0
.. .. ..
Un = , = , = . .
. B n . Tn
0 1 0 ∗ 0 ∗
Then Un is a nilpotent, solvable subgroup of GLn (k). Bn (called a Borel subgroup; see
Chapter 21) is also nilpotent but it is not solvable, since [Bn , Bn ] = Un . The subgroup Tn is
known as an n-dimensional torus; it is abelian, solvable and nilpotent. It’s also not hard to
see that Tn ∼
= G×n
m = Gm × · · · × Gm . We can regard Tn and Un as closed subgroups of Bn .
We will see in Chapter 21 that Bn ∼ = Tn n Un .
We will prove the following characterization in Section 21.4, which says that every con-
nected, solvable, affine algebraic group behaves in precisely the same way as in the example
above.
Theorem 20.4.17. Let G be a connected, solvable, affine algebraic group. Then
(1) For any s ∈ Gs , ZG (s) is connected.
(2) Gs lies in a maximal torus in G, i.e. a maximal closed subgroup isomorphic to a
product of copies of Gm .
(3) Any two maximal tori in G are conjugate.
(4) Gu is a closed, connected subgroup of G and if T is a maximal torus, then T ×Gu → G
is an isomorphism of varieties (not necessarily of algebraic groups).
477
20.5. Lie Algebra of an Algebraic Group Chapter 20. Algebraic Groups
We can also view this tangent space as Te G = (Me /M2e )∗ where Me = {f ∈ k[G] | f (e) =
0}. Note that since G is smooth, dim g = dim G = tr degk k(G) (using Theorem 12.1.17 for
the second equality). To prove g is indeed a Lie algebra, consider
Then for any X, Y ∈ Der(A), [X, Y ] := XY − Y X ∈ Der(A) which defines a Lie algebra
structure on Der(A). However, dimk Der(A) = ∞ so this space is too big.
g −→ g
x 7−→ x[p] .
This satisfies
There is a finite dimensional algebra U(g)− , called the restricted enveloping algebra of g,
that contains g as a Lie subalgebra and which satisfies x[p] = xp ∈ U(g)− for all x ∈ g. The
dimension of U(g)− is pdim g . This algebra U(g)− is actually what’s known as a Frobenius
algebra, meaning it’s similar in nature to the group algebra of a finite group. Further, U(g)−
is also a Hopf algebra.
478
20.5. Lie Algebra of an Algebraic Group Chapter 20. Algebraic Groups
In our setting, Der(A) is a restricted enveloping algebra, but it’s not finite dimensional so
we will relate Te G to a finite dimensional subalgebra of Der(A) that satisfies the properties
of the Lie algebra we’re looking for.
Therefore D is a derivation.
479
20.5. Lie Algebra of an Algebraic Group Chapter 20. Algebraic Groups
ρg : A → A,f →
7 g·f
λg : A → A,f →7 f · g −1 ,
So we see that (1 ⊗ δ)∆ and (εx ⊗ 1)∆ commute. Hence D(f · x) = D(f ) · x.
This space α(Te G) has a Lie algebra structure with Lie bracket
X
[σ, τ ] = (σ(fi )τ (fi0 ) − τ (fi )σ(fi0 )).
i
ϕ∗ : g −→ g0
δ 7−→ δ ◦ ϕ∗ ,
480
20.5. Lie Algebra of an Algebraic Group Chapter 20. Algebraic Groups
Examples.
1 Let G = Ga with coordinate algebra A = k[t]. To compute g, note that any derivation
is determined by its value on t. For all x, y ∈ Ga , we have (λx t)(y) = y − x so every
left translation of t may be written λx t = t − x. Suppose D is a derivation on k[t]
commuting with all λx for x ∈ Ga . Then D(t) = ∞ i
P
a
i=0 i t for some ai ∈ k, all but
finitely many of which are zero, and we have
∞
X ∞
X
i
λx D(t) = λx ai t = ai (t − x)i
i=0 i=0
∞
X
while Dλx (t) = D(t − x) = D(t) − D(x) = D(t) = ai ti .
i=0
d d
(t + x)n = (tn ) · x.
dt dt
Furthermore, when char k = p > 0,
p
d
tn = n(n − 1) · · · (n − p + 1)tn−p = 0.
dt
2 Let G = Gm with coordinate algebra A = k[t, t−1 ]. Then if D is a derivation on k[t, t−1 ]
commuting with all λx , x ∈ Gm , we may write D(t) as a Laurent series,
∞
X
D(t) = ai ti
i=−∞
This implies x−1 ai = ai x−i for all i, which is only possible if ai = 0 for all i 6= 1 and
D(t) = at for a nonzero scalar a = a1 ∈ k × . Thus D = at dtd , which shows that the
Lie algebra g is spanned by the basis t dtd . Note that t dtd (tn ) = ntn . In characteristic
p > 0, p
d
t tn = tn
dt
481
20.5. Lie Algebra of an Algebraic Group Chapter 20. Algebraic Groups
p
so t dtd = t dtd , in which case the [p]-operator is just the identity. Once again, g is a
p-restricted Lie algebra.
The previous two examples show that the Lie algebras of Ga and Gm are abstractly
isomorphic (they are one-dimensional k-algebras in every characteristic), but not as
restricted Lie algebras. Interestingly, we may view the map
Ga −→ Gm
x 7−→ xn
as a covering space in characteristic zero, and thus the tangent spaces are preserved.
However, in characteristic p > 0, x 7→ xp is not a covering space (in particular, it is
not étale; see Part VI).
3 Let G = GLn (k). The coordinate algebra may be written k[xij ] ∇1 where ∇ = det
as an operator. Then g is spanned by dxdij and the Lie bracket is simply [X, Y ] =
x=e
XY − Y X as one should expect. In characteristic p > 0, g is a restricted Lie algebra
under the operator (aij )[p] = (apij ).
Theorem 20.5.7 (Chevalley). Let H be a closed subgroup of an algebraic group G, with Lie
algebras h ⊆ g. Then there exists a representation ρ : G → GL(V ) and a nonzero vector
v ∈ V such that
482
Chapter 21
483
21.1. Quotients of Algebraic Groups Chapter 21. Classifying Algebraic Groups
We have thus defined a topological space G/H, which inherits the Zariski topology from
the quotient map π : G → G/H, and a sheaf of functions OG/H .
(1) G/H is an algebraic variety with sheaf of functions OG/H as defined above.
(4) G/H is universal with respect to G-spaces which are H-trivial. That is, for any G-
space Y and any y ∈ Y such that H ⊆ Gy , there is a unique G-equivariant morphism
ϕ : G/H → Y making the diagram
αy
G/H Y
ϕ
commute, where αy : g 7→ g · y.
Given G and H as above, let A = k[G] be the coordinate algebra of G. Recall that A is a
rational G-module (via translation). We saw that every finite dimensional subspace W ⊂ A
is contained in a finite dimensional G-stable subspace V ⊂ A.
Lemma 21.1.3. For any closed subgroup H ≤ G, there exists a finite dimensional subspace
V ⊂ k[G] and a subspace W ⊂ V such that
484
21.1. Quotients of Algebraic Groups Chapter 21. Classifying Algebraic Groups
Proof. (1) Choose a basis {v1 , . . . , vd } of W and a basis {v`+1 , . . . , v`+d } of xW . These
may overlap since xW ∩ W is a subspace of V . For any x ∈ GL(V ), x · (v1 ∧ · · · ∧ vd ) ∈
k(v`+1 , . . . , v`+d ). If ϕ(x)L = L then v1 ∧· · ·∧vd and v`+1 ∧· · ·∧v`+d are linearly independent,
but this can only happen if ` = 0. This means v1 , . . . , vd also span xW , but dim xW = d =
dim W so we must have xW = w.
(2) is proven similarly.
Theorem 21.1.5. Given G an algebraic group and H a closed subgroup, there exists a
quasi-projective variety X such that
(5) X ∼
= G/H as varieties.
Proof. (Sketch) By Lemma 21.1.3, there exists a finite dimensional rational G-module V
containing an element v for which Gv = H. Form the projective space P(V ) consisting of
all one-dimensional subspaces of V . Then G acts on P(V ), and v corresponds to a point
x ∈ P(V ). Let X = G · x be the orbit of x under this action. One may check that X satisfies
(1) through (5).
485
21.2. Parabolic and Borel Subgroups Chapter 21. Classifying Algebraic Groups
π π
Theorem 21.2.2. Let G be connected. Then G has a proper parabolic subgroup if and only
if G is not solvable.
Remark. Let P ≤ G be a parabolic subgroup. Then for every variety X there is a morphism
πX : G × X −→ G/P × X
(g, x) 7−→ (gP, x).
−1
Let A ⊆ G/P ×X and set A0 = πX (A) ⊆ G×X. Observe that if (g, x) ∈ A0 then (gp, x) ∈ A0
for all p ∈ P . If some closed subset B ⊆ G × X satisfies (gP, x) ⊆ B then πX (B) is closed.
This is another way of phrasing the completeness condition.
Proof. Let X be a variety, take A ⊆ G/Q × X and set A0 = p−1 (A), where p : G × X →
G/Q × X. Consider the following diagram:
486
21.2. Parabolic and Borel Subgroups Chapter 21. Classifying Algebraic Groups
α
P ×G×X G × X ⊇ A0
σ p
P/Q × G × X G/Q × X ⊇ A
πG×X πX
e0 ⊆ G × X
A X
πX
e0 = πG×X (σα−1 (A0 )) ⊆ G × X. Then following the diagram around counterclockwise
Let A
e0 ) = πX (A), which is closed by completeness of P/Q. Therefore G/Q is also
shows that πX (A
complete.
Lemma 21.2.6. Let Q ≤ P ≤ G be closed subgroups of an algebraic group G. Then P is
parabolic if Q is parabolic.
Proof. Suppose G/Q is complete. There is a surjective morphism of varieties G/Q → G/P ,
so by (3) of Proposition 12.11.1, G/P is complete.
We will now prove the characterization of connected solvable groups in terms of parabolic
subgroups.
Theorem 21.2.7. A connected algebraic group G has a proper parabolic subgroup if and
only if G is not solvable.
Proof. We will prove the theorem when G is affine. We may then take G to be a closed
subgroup of GL(V ) for some finite dimensional k-vector space V . There is a mapping
V r {0} −→ P(V )
v 7−→ {cv | c ∈ k}.
The action of G on V induces an action on P(V ) which makes P(V ) into a homogeneous
G-space. Take an orbit O = G · x under this action and consider its closure O ⊆ P(V ). We
may assume O has minimal dimension, so that O = O. Since P(V ) is complete, Proposi-
tion 12.11.1(1) implies O is also complete. Thus there is a bijective morphism G/Gx ↔ O.
Let P = Gx = StabG (x). By Lemma 21.2.1, G/P is complete so P is parabolic. If P 6= G
then we’re done. Otherwise suppose P = G. Then we can replace V with the quotient space
V /x and repeat the argument to find x0 ∈ P(V /x) such that Gx0 is parabolic. Continue
this process, inductively defining x0 = x, x1 = x0 , . . . and V0 = V, V1 = V /x0 , . . . Either this
process terminates with a proper parabolic subgroup Pk = Gxk 6= G for some xk , in which
case we’re done, or G stabilizes a flag of subspaces,
0 = Vn ⊆ Vn−1 ⊆ · · · ⊆ V1 ⊆ V0 = V,
such that dim Vi /Vi+1 for i = 0, . . . , n − 1. Choose an ordered basis {v0 , . . . , vn } such that
the subgroup hv0 , . . . , vi i is G-stable for each i = 0, . . . , n − 1. Then relative to this basis, G
is isomorphic to a group of upper triangular matrices which is solvable.
487
21.2. Parabolic and Borel Subgroups Chapter 21. Classifying Algebraic Groups
Example 21.2.8. The most important example of a Borel subgroup is the group Bn ≤
GLn (k) of upper triangular matrices. In this case GLn (k)/Bn ∼
= Pn , projective n-space.
n
By Theorem 12.11.2, P is complete, which implies that B is also a parabolic subgroup of
GLn (k).
Theorem 21.2.9 (Borel’s Fixed Point Theorem). Let G be a connected solvable group acting
on a complete variety X. Then there exists an element x ∈ X that is fixed by the action of
G.
G × G −→ G
(g, x) 7−→ σ(g)xg −1 .
488
21.2. Parabolic and Borel Subgroups Chapter 21. Classifying Algebraic Groups
G/B G
f
commute, where α(g) = σ(g)g −1 . Now since G/B is complete and G is affine, by Propo-
sition 12.11.1 f must be constant, so for all g ∈ G, σ(g)g −1 = 1. That is, σ(g) = g as
required.
Proof. (1) Suppose P is parabolic and take a Borel subgroup B ≤ G. Consider the action
of B on G/P via b · xP = bxP . Then by Borel’s fixed point theorem, the action has a fixed
point x ∈ G/P , i.e.
Clearly x−1 Bx is also a Borel subgroup, so the forward implication holds. For the converse,
we prove directly that every Borel subgroup B ≤ G is parabolic and use Lemma 21.2.6. If G
itself is connected and solvable, then we’re done. Otherwise, by Theorem 21.2.7, G contains
a proper parabolic subgroup P . Take a Borel subgroup B ≤ G. By the argument above, we
may assume B ≤ P . Then dim G − dim B > dim P − dim B. By induction, we can show that
B is parabolic in P . Thus by Lemma 21.2.6, B is parabolic in G so every Borel subgroup is
parabolic.
(2) Let B, B 0 ≤ G be two Borel subgroups. Then by (1), B 0 is parabolic so by the
fixed point theorem, B is conjugate to a subgroup of B 0 , i.e. there is an x ∈ G such that
xBx−1 ⊆ B 0 . But B is maximal among connected, solvable subgroups of G so we must have
xBx−1 = B 0 . Hence every pair of Borel subgroups is conjugate.
Proof. Let B ≤ G be a Borel subgroup. Then ϕ(B) is also connected and solvable. Consider
the induced map ϕ : G/B → G0 /ϕ(B). Since B is also parabolic, G/B is complete so
ϕ(G/B) = G0 /ϕ(B) is complete. Thus ϕ(B) is parabolic in G0 , so it is Borel. The proof for
parabolic subgroups follows this proof closely.
489
21.2. Parabolic and Borel Subgroups Chapter 21. Classifying Algebraic Groups
Proof. Z(G)0 is closed and connected, and solvable since it’s abelian. So Z(G)0 ⊆ B 0 for
some Borel subgroup B 0 ≤ G. Since B 0 is conjugate to B and Z(G)0 ⊆ Z(G), we may
assume Z(G)0 ⊆ B. This implies Z(G)0 ⊆ Z(B). Now take g ∈ Z(B) and consider the
commutator map
f : G −→ G
x 7−→ gxg −1 x−1 .
Since g ∈ Z(B), this map is constant on left cosets of B:
f (xb) = gxbg −1 b−1 x−1 = gxg −1 bb−1 x−1 = gxg −1 x−1 = f (x).
By the universal property of quotient varieties, this gives us a map
τ : G/B −→ G
xB 7−→ f (x) = gxg −1 x−1 .
Sicne G/B is complete and G is affine, τ is a constant map, which forces gxg −1 x−1 = 1 for
all x ∈ G. Hence g ∈ Z(G).
Corollary 21.2.14. Let G be connected and B ≤ G be a nilpotent Borel subgroup. Then
B = G. That is, G is nilpotent.
Proof. Since B is nilpotent, B contains a closed abelian subgroup Z ≤ B with Z ⊆ Z(B)
and Z 6= {1}. By Corollary 21.2.13, Z ⊆ Z(G). Also, the surjection G → G/Z shows
that G/Z has B/Z as a Borel subgroup. Since B/Z is nilpotent and dim B/Z < dim G,
by induction we obtain B/Z = G/Z. Therefore G/Z is nilpotent, so G is nilpotent. This
implies B = G by maximality of B.
Examples.
3 Let G = GLn (k) and consider the Borel subgroup B of upper triangular matrices.
Then B is in fact the point-stabilizer in the projective variety Flagn (k) of complete
flags
0 = V0 ( V1 ( · · · ( Vn = V
where V = k n . Using the fact that Flagn (k) is complete, this proves that G/B is
complete so B is parabolic.
4 Consider B − ≤ GLn (k), the subgroup of lower triangular matrices. Then B − is a Borel
subgroup of GLn (k) since it is conjugate to B:
0 1
wBw−1 = B − where w =
.
..
1 0
The set B − B is a dense open subvariety of G, called the big cell. Let W be the subgroup
of permutation matrices in GLn (k), so that W ∼ = Sn . One can prove that
[
G= BwB
w∈W
490
21.2. Parabolic and Borel Subgroups Chapter 21. Classifying Algebraic Groups
is a disjoint union. This is called the Bruhat decomposition of G, mimicking the decom-
position of the same name for Lie algebras. We will discuss the Bruhat decomposition
further in Section 21.5.
Then GLn (k) acts on the group On (k) = {g ∈ G | hgx, gyi = hx, yi} of orthogonal
linear transformations of V . Recall that a subspace W ⊆ V is isotropic if for all
x, y ∈ W , hx, yi = 0. The Borel subgroups of On (k) are all conjugate to the point-
stabilizer of the variety of isotropic flags of V , i.e. flats
0 = V0 ( V1 ( · · · ( Vn = V
Lemma 21.2.15. Let X, Y, Z be irreducible varieties with X complete, and suppose there is
a morphism ϕ : X × Y → Z. If for some a ∈ Y , ϕ(·, a) : X → Z is a constant map then
ϕ(·, y) is constant for all y ∈ Y .
491
21.3. Diagonalizable Groups Chapter 21. Classifying Algebraic Groups
(2) X ∗ (G) is an abelian group of finite type, and therefore isomorphic to Zn ⊕ F where F
is finite, and the elements of X ∗ (G) form a basis for k[G].
492
21.3. Diagonalizable Groups Chapter 21. Classifying Algebraic Groups
for linear operators Aχ : V → V . (This is possible since by hypothesis, the characters form
a basis of k[G].) Since ϕ is a homomorphism, for any x, y ∈ G we have
X X X X
ϕ(xy) = ψ(xy)Aψ = θ(x)Aθ µ(y)Aµ = θ(x)µ(y)Aθ Aµ .
ψ∈X ∗ (G) θ∈X ∗ (G) µ∈X ∗ (G) θ,µ
By linear independence, we must have Aθ Aµ = δθ,µ Aθ where δθ,µ is the Kronecker delta. Now
if x = 1, ϕ(1) = idV and the expression (∗) becomes
X
idV = Aχ (∗∗)
χ∈X ∗ (G)
Thus the nonzero Aχ form a complete system of orthogonal idempotent linear operators on
V . If Aχ 6= 0, set Vχ = Aχ (V ). Then (∗∗) implies that we can decompose V into these
subspaces: M
V = Vχ .
χ∈X ∗ (G)
It remains to show the Vχ are one-dimensional. Notice that G acts on each Vχ via
X
x · v = ϕ(x)v = θ(x)Aθ (v) for x ∈ G, v ∈ Vχ
θ∈X ∗ (G)
X
= θ(x)Aθ Aχ (v 0 )
θ∈X ∗ (G)
= χ(x)Aχ (v 0 ) = χ(x)v.
Therefore G acts by the character χ on Vχ , so it follows that each Vχ is one-dimensional.
(3) =⇒ (1) Finally, by the work above we have a basis of eigenvectors for the image
of G in GL(V ), so by linear algebra G is diagonalizable in the linear sense. Then G ⊆ Tn
under this basis.
493
21.3. Diagonalizable Groups Chapter 21. Classifying Algebraic Groups
Proof. Suppose χ ∈ X ∗ (G) satisfies χp = 1. Then for any x ∈ G, 1 = χp (x) = χ(x)p which
implies χ(x) = 1, so χ is identically the operator 1. Hence the p-torsion part of X ∗ (G) is
trivial.
We can push this characterization further. Let M be an abelian group of finite type with
no p-torsion, where char k = p > 0. Define the k-algebra k[M ] having basis elements e(m)
for all m ∈ M ; these are formal exponentiation symbols, which distinguish the operation
e(m)e(m0 ) = e(m+m0 ) in k[M ] from the addition operation in M itself. Define multiplication
in this way, e(m)e(m0 ) = e(m+m0 ) and extend linearly. One can put a Hopf algebra structure
on k[M ] by defining
Theorem 21.3.6. k[M ] is an affine algebra which is the coordinate algebra for a diagonal-
izable algebraic group G(M ) for which k[G(M )] = k[M ] as algebras and M ∼
= X ∗ (G(M )) as
abelian groups.
Proposition 21.3.9. Let M be a finitely generated abelian group with submodules {Mi }ni=1 .
Then the following are equivalent:
(1) M = ni=1 Mi .
L
494
21.3. Diagonalizable Groups Chapter 21. Classifying Algebraic Groups
495
21.4. Connected Solvable Algebraic Groups Chapter 21. Classifying Algebraic Groups
(1) X = Xs + Xn .
(3) If ϕ : G → GLn (k) is an embedding with differential dϕ : g → gln (k), then dϕ(Xs ) =
dϕ(X)s and dϕ(Xn ) = dϕ(X)n .
(4) If G ≤ GLn (k) and g ⊆ gln (k) then X = Xs + Xn is the usual Jordan decomposition
from linear algebra.
Lemma 21.4.2. Let T = G×n m be an n-dimensional torus. Then T consists entirely of
semisimple elements, that is Ts = T . If H ≥ T is a closed, connected subgroup then H = G×r
m
for some r ≤ n, that is, every closed, connected subgroup of a torus is also a torus.
496
21.4. Connected Solvable Algebraic Groups Chapter 21. Classifying Algebraic Groups
W be the span of all common eigenvectors for [G, G]. Then for all g ∈ G, x ∈ [G, G], v ∈ W ,
say with xv = λv for some λ ∈ k, we have
Corollary 21.4.4 (Structure Theorem for Connected Nilpotent Groups). Assume G is con-
nected and nilpotent. Let Gs be the subset of all semisimple elements in G and let Gu be the
closed subvariety of all unipotent elements of G. Then
(2) Gs ∼
= Gm × · · · × Gm , that is, Gs is a torus.
χ : G −→ G
x 7−→ sxs−1 x−1 .
de χ : g −→ g
X 7−→ ad(s)(X) − X.
497
21.4. Connected Solvable Algebraic Groups Chapter 21. Classifying Algebraic Groups
r
M
V = Vχi where Vχi = {v ∈ V | sv = χi (s)v for all s ∈ Gs }.
i=1
(One can think of the Vχi as generalized eigenspaces with respect to the χi .) Now each Vχi
is a G-submodule of V , so by the Lie-Kolchin Theorem (21.4.3), we may simultaneously
triangularize G so that xGx−1 ⊆ Bn , the subgroup of upper triangular matrices in GLn (k).
Then with respect to this choice of basis, Gs consists of diagonal matrices in GLn (k). In
particular, Gs is a closed subgroup of G. The proof is similar for Gu , so we have proven (1).
Now by Jordan decomposition, the map
Gs × Gu −→ G
(s, u) 7−→ su
is a bijection. Since Gs , Gu are closed subgroups and varieties, this is in fact an isomorphism
of groups and varieties, so (2) holds.
Finally, since G is connected, Gs and Gu are connected. We proved in part (1) that Gs
consists of diagonal matrices in GLn (k) so by Lemma 21.4.2, Gs is isomorphic to a torus.
With the next result, we complete our description of solvable algebraic groups.
(1) The commutator subgroup G0 = [G, G] is a closed, connected and unipotent subgroup
of G.
Proof. (1) For each x ∈ G, set [G, x] = {gxg −1 x−1 | g ∈ G}. This is a closed, irreducible
subvariety of G containing the identity. The commutator G0 = [G, G] is generated by these
sets [G, x] as a subgroup of G, so G0 is also closed and connected.
(2) View G ≤ Bn by the Lie-Kolchin Theorem (21.4.3). By matrix multiplication, G0 ⊆
Un , the subgroup of Bn with 1’s along the diagonal, so in particular G0 consists of unipotent
matrices. This means G0 ⊆ Gu so Gu is normal. Moreover, Gu = G ∩ Un so Gu is closed and
connected.
(3) By (3) of Corollary 21.4.4, G/Gu ∼= Gs so G/Gu consists of semisimple elements, that
0
is, (G/Gu )s = G/Gu . Since G ⊆ Gu by part (2), G/Gu is abelian. This means G/Gu is
solvable so by (2) of Corollary 21.4.4, G/Gu is isomorphic to a torus.
498
21.4. Connected Solvable Algebraic Groups Chapter 21. Classifying Algebraic Groups
σs : G −→ G
x 7−→ sxs−1 .
Then σs has a differential, dσs = ad(s) : g → g. For example, if G = GLn (k) then
ad(s) = sXs−1 , the adjoint action of G on g. Let ZG (s) = {g ∈ G | gs = sg} be the
centralizer of s. This is a closed subgroup, so we may ask what the Lie algebra is for this
algebraic group.
(2) If char k = p > 0 and s is semisimple, then the same holds. In this case, we simply
write the Lie algebra of ZG (s) as Zg (s).
that when G is connected, all the subgroups Cn and Dn are closed and connected.
Recall T
Set C∞ = ∞ n=0 Cn . Since Cn ⊇ Cn+1 for all n, there is some n ≥ 0 such that C∞ = Cn .
499
21.4. Connected Solvable Algebraic Groups Chapter 21. Classifying Algebraic Groups
500
21.4. Connected Solvable Algebraic Groups Chapter 21. Classifying Algebraic Groups
Proof. Let P = yQy −1 for some y ∈ G. Then B and yBy −1 are both Borel subgroups of
P , so there exists x ∈ P such that B = xyBy −1 x−1 . Thus xy ∈ NG (B) = B ⊆ P by
Theorem 21.4.13, so y ∈ P and hence P = Q.
Corollary 21.4.16. Let G be an affine algebraic group and let B be the set of all Borel
subgroups of G. Then B is a projective G-space via the map
ϕ : G/B −→ B
xB 7−→ xBx−1
ϕ : G −→ G
x 7−→ gxg −1 x−1
is constant, so g ∈ Z(G). Thus {1} ( ZG (B) ⊆ Z(G) which implies dim G/Z(G) ≤ 1, so
G/Z(G) is solvable and hence so is G.
501
21.5. Semisimple Algebraic Groups Chapter 21. Classifying Algebraic Groups
The sl2 subgroups hgα , g−α i ∼ = sl2 (C), each spanned by eα , hα , fα for α ∈ X, which
satisfy [hα , eα ] = 2eα , [hα , fα ] = −2fα , [eα , fα ] = hα
The classification of semisimple algebraic groups is modelled after this setup. Suppose
S ≤ G is a torus, ϕ : S → GL(V ) is a finite dimensional representation and χ ∈ X ∗ (S) is a
character. Then we define the submodule
Definition. Let T be a maximal torus in a connected, affine algebraic group G. Then dim T
is called the rank of G.
Proof. For well-definedness, use the fact that any two tori in G are conjugate. Invariance is
obvious.
502
21.5. Semisimple Algebraic Groups Chapter 21. Classifying Algebraic Groups
ϕg : G −→ G
x 7−→ gxg −1 .
dϕg = Ad(g) : g −→ g.
In this way g is a rational G-module. Moreover, for any maximal torus T ≤ G, this action
restricts to a T -action on g.
Lemma 21.5.2. There exists a unique maximum connected, normal, solvable subgroup of
every affine algebraic group G.
Proof. Uniqueness follows from the fact that two connected, normal, solvable subgroups
generate a larger connected, normal, solvable subgroup.
Definition. The unique maximum connected, normal, solvable subgroup of G is called the
solvable radical of G, denoted Rs (G). An algebraic group is semisimple if Rs (G) = {1}.
There is an analagous definition for unipotent subgroups; existence and uniqueness are
proven in a similar fashion.
Definition. The unique maximum, connected, normal, unipotent subgroup of G is called the
unipotent radical of G, denoted Ru (G). An algebraic group is reductive if Ru (G) = {1}.
Example 21.5.3. GLn (k) is not semisimple, since Rs (GLn (k)) consists of the upper trian-
gular scalar matrices:
t
∗
t
×
Rs (G) = :t∈k .
..
0 .
t
One can check that Ru (GLn (k)) = {I}, so GLn (k) is reductive.
On the other hand, SLn (k) is semisimple. To see this, suppose H C SLn (k) is a con-
nected, normal, solvable subgroup. By the Lie-Kolchin Theorem (21.4.3), H can be put in
upper triangular form, so assume H ⊆ Bn . Conjugating H by the matrix
0 1
.
X=
..
±1 0
makes H lower triangular, where the sign in the lower left is chosen to make det X = 1.
Then XHX −1 = H since H C SLn (k), and hence H ⊆ Tn , the maximal torus of SLn (k).
So H consists of scalar matrices, so |H| < ∞. Since H is connected, this can only occur if
H = {1}. Therefore SLn (k) is semisimple.
503
21.5. Semisimple Algebraic Groups Chapter 21. Classifying Algebraic Groups
Remarks.
1 If G is reductive then G0 = [G, G] is semisimple.
1 → K → G0 × S → G → 1
with K a finite group. In other words, G is an almost direct product. By this token,
every reductive group is an almost direct product of semisimple groups so it’s enough
to study semisimple groups to capture all of the theory.
Example 21.5.5. Let k be a field of characteristic p > 0 and suppose (n, p) = 1. If
G = GLn (k) with S = {tI : t ∈ k × } and G0 = SLn (k) then
ξ
0
..
. :ξ =1 ∼
n
SLn (k) ∩ S = = Z/nZ.
0
ξ
We will need a good definition of the projective special linear group P SL2 (k) in the next
section. If char k 6= 2, then we define P SL2 (k) = SL2 (k)/h−Ii. However, if char k = 2, then
the subgroup h−Ii is trivial, and therefore this construction does not work. Instead, recall
the coordinate algebra of SL2 (k):
k[SL2 (k)] = k[T11 , T12 , T21 , T22 ]/hT11 T22 − T12 T21 − 1i = k[tij ].
504
21.5. Semisimple Algebraic Groups Chapter 21. Classifying Algebraic Groups
Define a subalgebra B of A by
B = k[tij tk` | 1 ≤ i, j, k, ` ≤ 2] ⊆ A.
i π
A B SL2 (k) P SL2 (k)
X X ◦ i = π(X)
k k
Here SL2 (k) is the space of k-maps A → k and for each X ∈ SL2 (k), π(X) = X ◦ i : B → k.
Suppose ϕ : B → k is a k-homomorphism, i.e. ϕ ∈ P SL2 (k). Then ϕ(tij tk` ) ∈ k, so for
example ϕ(t211 ) = ϕ(t11 )2 = ξ11 ∈ k. In characteristic 2, the equation ϕ(t11 )2 − ξ11 has a
unique solution which allows us to define Φ(tij ) for all tij such that ϕ = Φ ◦ i. This is how
to exhibit the isomorphism π : SL2 (k) → P SL2 (k) in the characteristic 2 case.
Definition. An automorphism σ ∈ Aut(G) is called semisimple if the comorphism σ ∗ :
k[G] → k[G] is semisimple as an operator.
Note that any σ ∈ Aut(G) induces an automorphism of the Lie algebra g of G by taking
the differential.
Lemma 21.5.7. An automorphism σ ∈ Aut(G) is semisimple if and only if there is an
isomorphism of algebraic groups ϕ : G → G ≤ GLn (k) and a semisimple matrix s ∈ GLn (k)
such that for all x ∈ G, ϕ(σ(x)) = sϕ(x)s−1 .
Theorem 21.5.8. Let G be an affine algebraic group over a field k. If σ ∈ Aut(G) is
semisimple, then the Lie algebra of Gσ = {x ∈ G | σx = x} is gσ = {X ∈ g | dσ(X) = X}.
Moreover, if char k = 0 then the same holds for any σ ∈ Aut(G).
Example 21.5.9. Suppose char k = 2 and G = SL2 (k). Then the Lie algebra of G is
505
21.5. Semisimple Algebraic Groups Chapter 21. Classifying Algebraic Groups
1 1
Let σ be the automorphism of SL2 (k) given by conjugation by the matrix . Then
0 1
one can check that
1 b 1 0 0 1
Gσ = :b∈k , gσ = ,
0 1 0 1 0 0
and dim Gσ = 1 and dimk gσ = 2. So the Lie algebra of Gσ cannot be gσ . However, this does
not present a counterexample to Theorem 21.5.8 since σ is not a semisimple operator.
Corollary 21.5.10. Let D be a diagonalizable group acting on an affine group G and define
Proof. Use induction on dim g and apply Theorem 21.5.8 to the automorphisms G → G, x 7→
dx for each fixed d ∈ D.
506
21.6. Root Data Chapter 21. Classifying Algebraic Groups
507
21.6. Root Data Chapter 21. Classifying Algebraic Groups
Let R0 be the set of roots α ∈ R such that Gα is not solvable. Also set Wα =
NG (S)/ZG (S). This will be an analogue of the Weyl group. Note that Wα ⊆ Aut(Gm ) ∼
=
Z/2Z, so |W | ≤ 2.
Theorem 21.6.6. If α ∈ R0 then Wα ∼ = Z/2Z.
To prove Theorem 21.6.6, assume T ≤ G is a torus and T ⊆ GL(V ) for a rational
G-module V . Choose a basis {v1 , . . . , vn } for V consisting of T -eigenvectors, so that T ⊆
diag(v1 , . . . , vn ) ⊆ GL(V ). Then for each t ∈ T and vi , we have tvi = γi (t)vi for some
γi (t) ∈ Gm . Thus γi ∈ X ∗ (T ), and in fact, X ∗ (T ) = hγ1 , . . . , γn i. For any cocharacter
λ : Gm → T , set mi = hλ, γi i ∈ Z for each 1 ≤ i ≤ n. The composition
λ γi
Gm → − T −
→ Gm
is given by γi ◦ λ(z) = z mi for all z ∈ Gm . Now if v = ni=1 ai vi is a nonzero vector in V , we
P
have n
X
λ(z)v = ai z mi vi .
i=1
On the other hand, Gm also acts on P(V ) via z[v] = [zv], so we can extend this to
" n #
X
λ(z)[v] = [λ(z)v] = ai z m i v i .
i=1
0
Let m0 and m be respectively the minimum and maximum of m1 , . . . , mn . Define
I = {1 ≤ i ≤ n | ai 6= 0}, I0 = {1 ≤ i ≤ n | mi = m0 }, I 0 = {1 ≤ i ≤ n | mi = m0 }.
Consider the map ϕ : Gm → P(V ) given by z 7→ λ(z)[v]. We extend ϕ to a map on P1k
as follows. Let P1k = U0 ∪ U1 where
U0 = {[x, y] : y 6= 0} and U1 = {[x, y] : x 6= 0}.
Viewing Gm = U0 ∩ U1 ⊂ P1k , we need only define ϕ for 0 = [0, 1] and ∞ = [1, 0]. In fact,
there are two ways to extend ϕ:
" #
X
ϕ0 (z) = ai z mi −m0 vi
i∈I
" #
−m0
X
and ϕ1 (z) = ai z m i vi .
i∈I
508
21.6. Root Data Chapter 21. Classifying Algebraic Groups
P
Call ϕ the extension by ϕ0 and ϕ1 to all of P1k . Notice that ϕ(0)[v] = i∈I 0
a i v i and
P
ϕ(∞)[v] = i∈I 0 ai vi are both fixed poitns for the Gm -action on P(V ). Also notice that
ϕ(0)[v] = ϕ(∞)[v] is equivalent to m0 = m0 , in which case all the mi are equal and hence [v]
is itself a fixed point of the Gm -action. So we may choose λ ∈ X∗ (T ) in the first place so that
mi 6= mj for all i 6= j, so that ϕ(0) 6= ϕ(∞). Since h·, ·i : X ∗ (T ) × X∗ (T ) → Z is a perfect
pairing, this shows that T and λ(Gm ) ⊆ T have the same eigenspaces in V . Theorem 21.6.6
follows from the next three results.
Theorem 21.6.8. Let T be a torus in GL(V ) acting on P(V ) and let X ⊆ P(V ) be a closed,
irreducible subvariety stabilized by the action of T . If X does not lie in a hyperplane in P(V ),
then
Proof. (1) Assume dim X ≥ 1. If T fixes all of X, the statement holds trivially. Otherwise,
there is some [v] ∈ X which is not fixed by T . But by the above, for a suitable choice of
λ ∈ X∗ (T ), λ(0)[v] and λ(∞)[v] are distinct and fixed by the action of T .
(2) Apply Lemma 21.6.7.
Proof. By Theorem 20.3.6, we may view G ⊆ GL(V ) so that G/P is an orbit of a point in
P(V ). Apply Theorem 21.6.8 to this orbit to get at least 2 fixed points.
This completes the proof of Theorem 21.6.6, since if α ∈ R0 , Gα has a proper Borel
subgroup B, so dim Gα /B ≥ 1 implies |Wα | = |NGα (T )/ZGα (T )| ≥ 2.
To classify semisimple algebraic groups, we next generalize the notion of a root system
for a complex Lie algebra.
and sets R and R∨ , called the roots and coroots of Ψ, respectively, which satisfy
509
21.6. Root Data Chapter 21. Classifying Algebraic Groups
(b) For each α ∈ R, there are maps sα : X → X and s∨α : X ∨ → X ∨ such that sα (x) =
x − hx, α∨ iα and s∨α (y) = y − hα, yiα∨ for all x ∈ X, y ∈ X ∨ .
510
21.6. Root Data Chapter 21. Classifying Algebraic Groups
X × X ∨ −→ Z
(α, χ) 7−→ hα, χi := m where α ◦ χ(t) = tm .
Given a root α ∈ Φ, there is a homomorphism ϕ : SL2 (k) → G such that ϕ(SL2 (k)) = Gα
is isomorphic to either SL2 (k) or P SL2 (k). Also, T normalizes Gα so the Lie algebra of Gα
may be written
h̄ ⊕ gα ⊕ g−α
for h̄ a one-dimensional subalgebra of h. This mimics the classical decomposition of sl2 (k)
in the Lie algebra setting. Consider the map
t 0
ν : Gm −→ SL2 (k), t 7−→ .
0 t−1
W = NG (T )/ZG (T ) −→ B T := {C ∈ B | T ⊆ C}
x̄ 7−→ xBx−1
is a bijection. This allows us to study the Weyl group in terms of the algebraic geometry of
the projective variety B.
Theorem 21.6.12. For a connected algebraic group G over an algebraically closed field k,
with maximal torus T , Borel subgroup B and Weyl group W = NG (T )/ZG (T ), the following
are equivalent:
511
21.6. Root Data Chapter 21. Classifying Algebraic Groups
512
21.6. Root Data Chapter 21. Classifying Algebraic Groups
(c) When S = (ker α)0 for a root α ∈ R, Wα = W (Gα , T ) = W (Gα /S, T /S).
Theorem 21.6.15. Let G be a connected semisimple algebraic group with maximal torus T ,
Borel subgroup B, Weyl group W = NG (T )/ZG (T ) and root system Φ. If E = X ⊗Z R, then
(1) E admits an inner product with respect to which W acts by orthogonal linear trans-
formations. In particular, W is a finite Coxeter group.
(2) Φ is a classical root system in the Euclidean space E and there is an embedding X ∨ ,→
E given by
2α
α∨ 7−→ .
hα, αi
(4) For any ordered subset Γ ⊆ Φ+ which is closed under the inner product, multiplication
defines an isomorphism
∼
Y
A|Γ| = Uα −→ UΓ := hUα | α ∈ Γi.
α∈Γ
Since Φ is a root system, we may define a set T = {α1 , . . . , αr } of simple roots, and
subsets Φ+ , Φ− ⊂ Φ of positive and negative roots, with respect to the bilinear form h·, ·i.
Theorem 21.6.16. G is simply connected if and only if its weight lattice X contains a set
of fundamental dominant weights.
513
21.6. Root Data Chapter 21. Classifying Algebraic Groups
$1 = ε1 , $2 = ε1 ε2 , . . . , $n = ε1 · · · εn .
Theorem 21.6.18. Let G and G0 be two simply connected, semisimple algebraic groups over
k with the same root system. Then G ∼ = G0 as algebraic groups. Conversely, given a classical
root system Φ, there exists a simply connected, semisimple algebraic group with Φ as its root
system.
This is an analogue to Theorem 21.6.18 in the Lie algebra case. For an arbitrary (meaning
not necessarily simply connected) semisimple algebraic group, we have the following.
Corollary 21.6.19. A semisimple algebraic group G is uniquely determined by its root datum
(X, Φ, X ∨ , Φ∨ ). Moreover, if G
b is a simply connected, semisimple algebraic group with root
system Φ then G is a homomorphic image of G b with finite kernel.
Theorem 21.6.20. For every irreducible root system Φ, Λ(Φ) is a finite group. Moreover,
there is a bijective correspondence
Example 21.6.21. The irreducible root systems are classified by their Dynkin diagrams,
which are labelled by their type: infinite families An , Bn , Cn , Dn (which splits into the even
and odd cases) and exceptional types G2 , F4 , E6 , E7 , E8 . For an irreducible root system
Φ of one of these types, there is a semisimple algebraic group Gsc corresponding to the
trivial subgroup {1} ⊆ Γ(Φ) – the notation is suggestive, as Gsc is simply connected in the
sense of our definition above. Likewise, there is a group GAd corresponding to the whole
fundamental group Γ(Φ). Below are listed some data about the irreducible root systems and
their corresponding semisimple algebraic groups.
514
21.6. Root Data Chapter 21. Classifying Algebraic Groups
Let Φ be an irreducible root system and take Π = {α1 , . . . , αr } to be a set of simple roots.
Note that there are many choices of Π, but if Π0 is another set of simple roots, there is a unique
element w ∈ W such that w(Π) = Π0 . We have the positive roots Φ+ = (Nα1 ⊕· · ·⊕Nαr )∩Φ
and the negative roots Φ− = −Φ+ , so that Φ = Φ+ ∪ Φ− . Notice that −Π = {−α1 , . . . , −αr }
is also a set of simple roots, so by the above, there is a unique element w0 ∈ W such that
w0 (Π) = −Π.
Definition. Let I ⊆ Π be a subset of simple roots and define the subgroup LI = hT, Uα iα∈±I
of G, where T is a maximal torus and the Uα are one-parameter subgroups. Then LI is called
a Levi subgroup, or Levi factor, corresponding to I.
Theorem 21.6.23. For a semisimple algebraic group G with maximal torus T , Borel sub-
group B and irreducible root system Φ containing simple roots Π,
(1) The subgroups PI := hLI , Bi, where LI is the Levi factor corresponding to some subset
I ⊆ Π, are parabolic subgroups of G. Moreover, the correspondence I ↔ PI is bijective.
Theorem 21.6.24 (Bruhat). Any semisimple algebraic group G may be written as a disjoint
union of double cosets [
G= BwB.
w∈W
515
21.6. Root Data Chapter 21. Classifying Algebraic Groups
Set Xw = BwB/B ⊆ G/B. Then X w is called a Schubert variety. It turns out that
each Xw is an affine subvariety of the complete variety G/B, with projective closure X w .
Specifically, Xw ∼
= A`(w) where `(w) is the length of w in the Weyl group.
We can also write G as a non-disjoint union
[
G= BwB.
w∈W
516
21.7. Representations of Algebraic Groups Chapter 21. Classifying Algebraic Groups
There is a unique element λ ∈ X ∗ (T ) which is maximal with respect to this ordering, and
such that Lλ 6= 0. One can prove dim Lλ = 1 and there is a nonzero eigenvector vλ ∈ Lλ such
that L = kG · vλ . Moreover, λ is dominant, i.e. hλ, α∨ i ∈ Z≥0 . Following the Lie algebra
case, we call λ the highest weight of L. This determines an assignment
τ : Irr(G) −→ X+
L 7−→ λ.
517
21.7. Representations of Algebraic Groups Chapter 21. Classifying Algebraic Groups
Chevalley proved that the mapping is also onto, i.e. for every λ ∈ X+ there is an (irreducible)
G-module L(λ). Borel later gave the following elementary proof.
It’s enough to consider λ = $i a fundamental dominant weight. Let Qi be the subgroup
nj αj ∈ Φ+ , where ni = 0. Then each Qi is closed
P
of G generated by B and U−α for α =
and parabolic. By Theorem 21.1.5, there exists a finite dimensional G-module V containing
a one-dimensional subspace ` ⊆ V such that Qi = StabG (`). Take a basis {v1 , . . . , vn } of V
such that kv1 = `. Then each x ∈ Qi corresponds to a matrix of the form
∗ ∗ ··· ∗
0 ∗ · · · ∗
.. .. . . ..
. . . .
0 ∗ ··· ∗
518
Chapter 22
In this chapter we study an important class of finite groups arising from linear algebraic
groups over a finite field F = Fq , where q = pm is a power of a prime p. First, we prove
an analogy of the fact from complex analysis that P GL2 (C) is equal to the automorphism
group of the Riemann sphere C ∪ {∞}.
V r {0} −→ P1
(x, y) 7−→ [x, y].
1 x x
Viewing points in V and P as column vectors and , respectively, GL2 (k) acts on
y y
V r {0} by left multiplication and this action descends to an action on P1 , written
a b x ax + by
= .
c d y cx + dy
Notice that Gm ∼
= Z(GL2 (k)) acts trivially on P1 , so we get an induced action of P GL2 (k) =
GL2 (k)/Z(GL2 (k)) on P1 , which we write as
a b x ax + by
= .
c d y cx + dy
1 a b
Consider the points 0 = [0, 1], 1 = [1, 1] and ∞ = [1, 0] in P . For each g = , define
c d
the triple
ϕ(g) := (g(0), g(1), g(∞)) = ([b, d], [a + b, c + d], [a, c]).
This defines a map ϕ : P GL2 (k) → P1 ×P1 ×P1 which is clearly injective (look at the first and
third components in particular). In fact, P GL2 (k) acts doubly transitively on P1 ×P1 ×P1 via
this map, with trivial isotropy subgroup. Therefore, for any (x, y, z) ∈ P1 ×P1 ×P1 with x, y, z
519
Chapter 22. Finite Groups of Lie Type
all distinct, there exists an element g ∈ P GL2 (k) such that (g(x), g(y), g(z)) = (0, [u, v], ∞)
with [u, v] 6= 0, ∞. Thus neither of u, v is 0, so we can write
−1
u 0 u
0, , ∞ = (0, 1, ∞).
0 v −1 v
This shows that P GL2 (k) acts triply transitively on the middle factor of P1 . Thus the map
is injective. On the other hand, for σ ∈ Aut(P1 ), there exists an element g ∈ P GL2 (k) such
that (σ(0), σ(1), σ(∞)) = (g(0), g(1), g(∞)) by triple-transitivity, so τ = σ −1 g fixes ∞ and
restricts to an automorphism of A1 = P1 r {∞}. Since k[A1 ] = k[t], the only invertible
endomorphisms are linear, so τ |A1 has the form x 7→ ax + b for some a, b ∈ k. But we also
know τ (0) = 0 and τ (1) = 1, which imply b = 0 and a = 1, so τ = 1 after all. Hence σ = g,
so P GL2 (k) ∼= Aut(P1 ).
520
22.1. Jordan’s Theorem Chapter 22. Finite Groups of Lie Type
This tells us that q n(n−1)/2 is the order of a Sylow p-subgroup of GLn (q), namely the nilpotent
subgroup
1 ∗
..
Un = . .
0
1
On the other hand, (q − 1)n is the order of the maximal torus of GLn (q),
∗ 0
..
Tn = . .
0 ∗
The remaining term has an interpretation in terms of topology, as we will see later.
Example 22.1.2. For SLn (q) = SLn (Fq ), consider the short exact sequence of groups
1 → SLn (q) → GLn (q) → F× q → 1.
i.e. C consists of diagonal matrices diag(ζ, . . . , ζ) where ζ is an nth root of unity. There-
fore |C| = (n, q − 1), the greatest common divisor, so the projective special linear group
P SLn (q) = P SLn (Fq ) = SLn (Fq )/C has order
|SLn (q)| |GLn (q)|
|P SLn (q)| = = .
(n, q − 1) (q − 1)(n, q − 1)
For example, P SL2 (2) = SL2 (2) is a finite group of order 6 which turns out to be isomorphic
to S3 (since it is nonabelian). Also, P SL2 (3) is a nonsimple group of order 12 which is
isomorphic to A4 . In what can be regarded as the origin of the classification of finite simple
groups, the following important result by Jordan gives an infinite class of finite simple groups
different from the alternating groups.
521
22.1. Jordan’s Theorem Chapter 22. Finite Groups of Lie Type
Theorem 22.1.3 (Jordan). Let F be any field. For n > 2, or for n = 2 and |F | =
6 2, 3, the
group |P SLn (F )| is simple.
Consider the set of n × n matrices Mn (F ) with basis {eij }1≤i,j≤n consisting of elementary
matrices. For b ∈ F and i 6= j, define
1
Tij (b) = 1 + beij = . . .
b 1
where b appears in the ith row and the jth column. Then Tij (b)Tij (c) = Tij (b + c), so
Tij : b 7→ Tij (b) defines a group homomorphism F → Mn (F ). Left (resp. right) multiplication
by each Tij (b) defines a map Mn (F ) → Mn (F ) which adds b times the jth row (resp. column)
to the ith row (resp. column). Also, for i 6= j, define
Pij = 1 + eij + eji − eii − ejj = (1 + eij )(1 − eji )(1 + eij )(1 − 2eii ).
Then left (resp. right) multiplication by Pij defines a map Mn (F ) → Mn (F ) which switches
the ith and jth rows (resp. columns). Note that
1
...
1 − 2eii = −1
. ..
Then P may be written with all the 1 − 2eii on the left and factors of Tij (b) otherwise (and
likewise for Q, with the 1 − 2eii on the right). Multiplying these factors across maintains
a diagonal matrix on the right, so we may assume P AQ = diag(d1 , . . . , dn ) where P and Q
are each a product of some Tij (b). Thus det(P ) = det(A) = det(Q) = 1 so d1 · · · dn = 1.
For d ∈ F × , there is a sequence of elementary matrix operations taking
−1
d 0 1 0
S(d) = to = T21 (d−1 ).
0 d d−1 1
522
22.1. Jordan’s Theorem Chapter 22. Finite Groups of Lie Type
In particular, S(d) is a product of the Tij (b). The same argument applies to
d−1 0 0 ··· 0
0 d
0 · · · 0
diag(d−1 , d, 1, . . . , 1) = 0 0
1 · · · 0
.. .. .. . . ..
. . . . .
0 0 0 ··· 1
for any number of 1’s on the diagonal. Next, our diagonal matrix diag(d1 , . . . , dn ) satisfies
diag(d−1
1 , d1 , 1, . . . , 1) diag(d1 , d2 , . . . , dn ) = diag(1, d1 d2 , d3 , . . . , dn ).
Proof. By Lemma 22.1.4, it suffices to show Tij (b) ∈ SLn (F )0 for every i 6= j and b ∈ F . For
n > 2, let 1 ≤ k ≤ n, k 6= i, j. Then for any b ∈ F , we can write
which is a commutator. Thus the lemma holds in this case. When n = 2, consider the
commutator
d 0 1 c 1 c(d2 − 1)
, = .
0 d−1 0 1 0 1
Since F 6= F2 , F3 , there exists a d ∈ F with d2 6= 1. Taking c = (d2 − 1)−1 b, the above
identity gives us
1 (d2 − 1)−1 b
d 0
T12 (b) = , ∈ SLn (F )0 .
0 d−1 0 1
Theorem 22.1.6 (Iwasawa). Suppose G acts on a set S with kernel K = ker(G → Aut(S)).
Then G/K is simple provided the following conditions hold:
(1) G acts primitively on S, i.e. S cannot be written as a disjoint union of subsets which
are permuted transitively by G.
523
22.1. Jordan’s Theorem Chapter 22. Finite Groups of Lie Type
(3) There exists an element x ∈ S such that the stabilizer Gx has a normal abelian sub-
group Ax and G is generated by its conjugates {gAx g −1 | g ∈ G}.
Example 22.1.7. Every 2-transitive group action is primitive. (Recall that a group action
is n-transitive if any subsetset of n elements can be taken to any other subset of n elements.)
Example 22.1.8. The natural action of G on itself by left (or right) multiplication is not
primitive since for any subgroup H ≤ G, the cosets gH are permuted transitively by G.
Proof of Jordan’s Theorem (22.1.3). Take G = SLn (F ) and let S = Pn−1 be projective
(n − 1)-space over F , i.e. the set of all lines through the origin in F n . Then G acts on S
with kernel C, the center of SLn (F ) so to prove that SLn (F ) is simple, we need only check
the conditions of Theorem 22.1.6. It is a standard fact that the action of SLn (F ) on lines
through the origin is 2-transitive, so by Example 22.1.7, the action is primitive. Further,
Lemma 22.1.5 showed that SLn (F ) = SLn (F )0 holds under our conditions on n and F , so
it remains to show condition (3) in Iwasawa’s theorem. Fix an ordered basis (e1 , . . . , en ) of
F n and let ` be the line spanned by e1 , so ` ∈ Pn−1 . The stabilizer of ` in SLn (F ) is
∗ ∗ · · · ∗
0 ∗ · · · ∗
G` = .. .. . . .. : ∗ ∈ F
. .
. .
0 ∗ ··· ∗
1 b
Then A` is a normal abelian subgroup of the stabilizer and since T12 (b) = , the
0 1
conjugates of A` generate SLn (F ). Hence SLn (F )/C = P SLn (F ) is simple.
524
22.2. Restricted Lie Algebras Chapter 22. Finite Groups of Lie Type
Derk (A) = {D : A → A | D is k-linear and D(xy) = D(x)y + xD(y) for all f, g ∈ A}.
Then [x, y] = xy − yx endows Derk (A) with the structure of a Lie algebra. The Leibniz rule
D(xy) = D(x)y + xD(y) can be extended inductively to the following formula for all n ≥ 1:
n
n
X n
D (xy) = Di (x)Dn−i (y).
i=0
i
which shows that Dp ∈ Derk (A) as well. In other words, the Lie algebra Derk (A) is closed
under the p-mapping D 7→ Dp . The notion of a restricted Lie algebra generalizes this.
where fi (x, y) is the coefficient of ti−1 in the formal expression (ad(tx + y))p−1 (x).
525
22.2. Restricted Lie Algebras Chapter 22. Finite Groups of Lie Type
so it suffices to show isi (x, y) = fi (x, y). Taking the formal derivative of the expressions for
f (t) with respect to t, we get
p−1 p−1
X X
0 i p−i−1
f (t) = (tx + y) x(tx + y) = isi (x, y)ti−1 .
i=0 i=1
On the other hand, (∗) shows that f 0 (t) = (ad(tx + y))p−1 (x), so we have isi (x, y) = fi (x, y)
as desired.
Definition. Let L and L0 be restricted Lie algebras over k.
A [p]-restricted homomorphism is a Lie algebra homomorphism ϕ : L → L0 such
that ϕ(x[p] ) = ϕ(x)[p] for all x ∈ L.
A [p]-restricted ideal of L is a Lie ideal I ≤ L such that x[p] ∈ I for all x ∈ I. Note
that [p]-restricted ideals are precisely kernels of [p]-restricted homomorphisms.
526
22.2. Restricted Lie Algebras Chapter 22. Finite Groups of Lie Type
(1) Does every restricted Lie algebra admit a finite dimensional, faithful, restricted repre-
sentation?
(2) When does an ordinary Lie algebra admit the structure of a restricted Lie algebra?
Starting with (2), notice that for L to be a restricted Lie algebra, we must have
(ad(x))p = ad(x[p] ) for all x ∈ L,
i.e. each (ad(x))p is an inner derivation on L. We will show that in the right circumstances
this condition is also sufficient for L to admit a restricted structure.
For an ordinary Lie alebra L with k-basis {ui }, recall the Poincaré-Birkhoff-Witt theorem:
Theorem 22.2.2 (PBW). The universal enveloping algebra U = U(L) has basis
{uki11 · · · ukirr | r ≥ 1, i1 < . . . < ir , kj ≥ 0}.
We will adapt this slightly to construct an analogue of the universal enveloping algebra
for restricted Lie algebras.
Theorem 22.2.3. Let L be a Lie algebra with basis {ui } such that every (ad(ui ))p is an
[p] [p]
inner derivation, say (ad(ui ))p = ad(ui ) for some ui ∈ L. Then there exists a unique
[p]-mapping L → L, x 7→ x[p] extending this definition and defining a p-restricted structure
on L.
[p]
Proof. Let U = U(L) be the universal enveloping algebra of L. Note that each zi := upi − ui
[p]
lies in the center of U since ad(zi ) = ad(ui )p − ad(ui ) = 0. Slightly adapting the proof of
the PBW theorem, one can prove that U has basis
{zih11 · · · zihrr uki11 · · · ukirr | r ≥ 1, i1 < . . . < ir , hj ≥ 0, 0 ≤ kj ≤ p − 1}.
Define the ideal I = hzi i ⊆ U and the quotient algebra U = U/I. Then the images of the
monomials uki11 · · · ukirr form a basis for U. Moreover, x 7→ x̄ := x + I ∈ U/I restricts to a
homomorphism L → L := (L + I)/I, but since the ūi = ui + I are linearly independent, this
mapping is an isomorphism of Lie algebras L ∼ = L.
Next, the associative algebra U is a restricted Lie algebra by Proposition 22.2.1 and by
construction L is a restricted subalgebra of U:
[p]
ūpi = (ui + I)p = upi + I p = upi + I = ui + I.
0
For uniqueness, suppose L admits another mapping x 7→ x[p] satisfying the axioms of a
0
restricted Lie algebra. Then x[p] − x[p] lies in the center of L for all x ∈ L. In other words,
0
the map f : x 7→ x[p] − x[p] is a p-semilinear mapping. Then ker f is a vector subspace of L,
[p] [p]0
so we need only observe that f is trivial on a basis of L: f (ui ) = ui − ui = upi − upi = 0.
Hence [p] = [p]0 .
Corollary 22.2.4. If L is a finite dimensional Lie algebra with nondegenerate Killing form,
then there is a unique p-restricted structure on L. Moreover, there is a restricted universal
enveloping algebra U as above which has dimension pdim L and is universal with respect to all
maps
527
22.2. Restricted Lie Algebras Chapter 22. Finite Groups of Lie Type
ψ
ϕ
L A
p
d
tn = n(n − 1) · · · (n − p + 1)tn−p = 0
dt
so in this case D[p] = Dp = D, i.e. D 7→ Dp is the identity on gm . This shows that gm and ga
are not isomorphic as restricted Lie algebras, even though they are isomorphic as abstract
Lie algebras over k.
Remark. Geometrically, the (restriction to k r {0} of the) map
Ga → Gm , x 7→ xn
may be viewed as a covering space (i.e. a local homeomorphism, or a locally trivial surjec-
tion) as long as char k does not divide n. In particular, this is always a covering space in
528
22.2. Restricted Lie Algebras Chapter 22. Finite Groups of Lie Type
characterstic 0. However, when char k = p | n, the above examples show that the tangent
space structure of Ga does not push forward to Gm under x 7→ xp . So this map cannot be a
local homeomorphism. It turns out that the correct analogue of the cyclic cover x 7→ xn in
characteristic p is the Artin-Schreier cover
Ga → Ga , x 7→ xp − x.
(See Part VI for the full theory of algebraic fundamental groups.) The general case of a
cover of degree n = pr m, (m, p) = 1 is given by Artin-Schreier-Witt theory.
The classical simple Lie algebras are completely classified by their Dynkin diagram: An ,
Bn , Cn , Dn , E6 , E7 , E8 , F4 , G2 . Each of these admits a model over Z (given by the
theory of Chevalley bases, which are related to Chevalley groups; see Section 22.4) which
therefore descends to a Lie algebra over Fp for any prime p. The reduction mod p of one
of the classical simple Lie algebras is semisimple in general, but by taking quotients, one
obtains all “classical” simple restricted Lie algebras. However, there are simple restricted
Lie algebras not arising from reduction mod p:
529
22.2. Restricted Lie Algebras Chapter 22. Finite Groups of Lie Type
530
22.3. The Frobenius Chapter 22. Finite Groups of Lie Type
GLn (k)F = {(aij ) ∈ GLn (k) | (aqij ) = (aij ) for all 1 ≤ i, j ≤ n} = GLn (Fq ).
Thus we recover our first finite (though non-simple) group of Lie type, GLn (q).
Let A = k[GLn (k)] be the coordinate algebra of GLn (k) and let B be any commutative
k-algebra. Then setting GLn (B) := Homk (A, B) defines a functor
G1 : Algk −→ Group
B 7−→ ker(F : GLn (B) → GLn (B)),
called the first infinitesimal subgroup of GLn . Note that k[G1 ] is a finite dimensional, commu-
tative Hopf algebra over k. In fact, its dual k[G1 ]∗ turns out to be the restricted enveloping
algebra U = U(gln ) of the Lie algebra gln = Lie(GLn (k)), in the sense of Section 22.2.
For any algebraic group G over k, Theorem 20.3.6 says that G is isomorphic to a closed
subgroup of GLn (k) for some n ≥ 1.
531
22.3. The Frobenius Chapter 22. Finite Groups of Lie Type
Example 22.3.4. When G = GLn (k), the map σ : (aij ) 7→ ((aqij )−1 )T satisfies σ 2 (aij ) =
2
(aqij ), so σ 2 = F : GLn (q 2 ) → GLn (q 2 ). Thus σ is Steinberg. The group GLn (k)σ ≤ GLn (q 2 )
is called the general unitary group, written GUn (q). Explicitly,
Also, the subgroup SUn (q) := GUn (q) ∩ SLn (k) is called the special unitary group.
Example 22.3.5. When G = Gm is the multiplicative group, all endomorphisms are of the
form
σm : Gm −→ Gm
γ 7−→ γ m , for some m ∈ Z.
σh : Ga −→ Ga
x 7−→ h(x)
is an endomorphism precisely when h(x) = f (xp ) for some f ∈ k[x]. Then for σhr = F , we
i
must have h(x) = xp for some i ≥ 1. Setting q = pi , σxq is a Steinberg endomorphism, but
the fixed points are just Gσaxq = Fq , so we don’t obtain any new groups this way.
532
22.3. The Frobenius Chapter 22. Finite Groups of Lie Type
Theorem 22.3.7 (Steinberg). Let G be a simple algebraic group over k = Fq . For any
endomorphism σ : G → G, exactly one of the following holds:
(1) σ is an automorphism.
One can construct Steinberg endomorphisms for all of the classical semisimple groups
over Fq (in the sense of Chevalley’s classification; see Theorem 21.6.18) except for types
B2 , C2 , G2 and F4 in small characteristic. This realizes “most” of the finite simple groups,
but misses types B2 = C2 in characteristic p = 2 (the Ree group), G2 in characteristic p = 3
(the Suzuki group) and F4 in characteristic p = 2 (the Tits group).
Proof. We give the proof when σ = F is the Frobenius, which is due to Lang. The general-
ization by Steinberg isn’t too much harder.
Let g be the Lie algebra of G. For L : G → G, g 7→ F (g)g −1 , the differential acts on g is
dL = d(F × (·)−1 ) = dF × (·)−1 − 1 = −1, so dL is a Lie algebra isomorphism. This means
L(G) is dense in G. Similarly, for x ∈ G, we may define Lx : G → G by L(g) = F (g)xg −1 ,
and since Tx G ∼
= g as k-vector spaces, dLx : g → Tx G is also an isomorphism. Thus Lx (G) is
dense in G, so by Chevalley’s theorem (19.1.1), Lx (G) ∩ L(G) 6= ∅. Take y ∈ Lx (G) ∩ L(G).
Then F (g1 )xg1−1 = y = F (g2 )g2−1 for some g1 , g2 ∈ G, so
This shows that F (B 0 ) = B 0 and F (T 0 ) = T 0 , so B 0 and T 0 work for the first statement of
the corollary. The second statement also follows from the above computation.
This allows one to construct Steinberg endomorphisms, and thus finite groups of Lie type,
and descend them to simple subgroups. To study representations of such groups, we have
the following construction, which we won’t discuss in detail.
533
22.3. The Frobenius Chapter 22. Finite Groups of Lie Type
ρ ◦ F : G −→ GL(V ).
This allows one to extend the representation theory of Section 21.7 to characteristic p.
For example, recall that for a semisimple algebraic group G, the irreducible G-modules are
in one-to-one correspondence with the set of dominant weights λ ∈ X ∗ (T ) = Hom(T, Gm ),
where T is a maximal torus in G.
Example 22.3.11. For G = SLn (k) with highest weight module L = Lλ , if λ is q-restricted
then Lλ |SLn (Fq ) gives an irreducible SLn (Fq )-module.
More generally, if Lξ is simple when restricted to the fixed points of the Frobenius, GF ,
then ξ is q-restricted. The converse also holds when q = p but not in general.
534
22.4. Chevalley Groups Chapter 22. Finite Groups of Lie Type
UZ = ZhXα(n) | α ∈ Φ, n ≥ 0i
(n) n
where Xα = Xn!α . Then UZ is a free abelian group and UZ ⊗ C = U. Thus for any finite
dimensional, faithful g-module V , there exists a free abelian Z-submodule VZ ⊆ V satisfying:
VZ ⊗ C = V
UZ · VZ ⊆ VZ .
Since UZ and VZ are defined over the integers, we can now define
U k = UZ ⊗Z k and V k = VZ ⊗Z k
ρ : U k −→ End(V k )
535
22.4. Chevalley Groups Chapter 22. Finite Groups of Lie Type
For the triple (g, V, k), the Chevalley group is then defined as
Example 22.4.1. For g = sln (C) with root system Φ = {εi − εj | i 6= j}, the root groups
are generated by Xεi −εj = Eij , the ijth matrix unit in SLn (k). Taking V = Cn to be the
defining representation of sln (C), we have that V k = k n for any field k. Then
so by Lemma 22.1.4, we see that GkV = SLn (k) for any field k. This construction even works
for any commutative ring R in place of k.
ΛV = SpanZ {µ ∈ Λr | Vµ 6= 0}.
Proposition 22.4.2. The quotient Λw /Λr is finite, and for any intermediate lattice Λr ⊆
Λ ⊆ Λw , there is a finite dimensional, faithful g-module V for which ΛV = Λ.
Fix a Lie algebra g, a finite dimensional, faithful g-module V and a Chevalley group
GV = GkV for some field k.
Example 22.4.4. The weight lattice Λw corresponds to the universal group Gsc , also known
as the simply connected group (in the sense of Example 21.6.21). Likewise, the Cheval-
ley group GAd corresponding to the root lattice Λr is GAd , the adjoint group from Ex-
ample 21.6.21. Thus, given any g-module V with Chevalley group GV , there is a pair of
α β
surjections Gsc −
→ GV →
− GAd .
α β
Proposition 22.4.5. Let V be a finite dimensional, faithful g-module and Gsc −
→ GV →
− GAd
the corresponding morphisms of Chevalley groups. Then
(b) When k = C, GV is a Lie group, ker β ∼ = ΛV /Λr and ker α ∼ = Λw /ΛV ∼= π1 (GV ), the
topological fundamental group of GV . In particular, π1 (Gsc ) = {1} so Gsc is simply
connected as a manifold.
536
22.4. Chevalley Groups Chapter 22. Finite Groups of Lie Type
Example 22.4.6. For the semisimple Lie algebra g = sln (C), the universal group is Gsc =
SLn (k), while the adjoint group is GAd = Gsc /Z(Gsc ) = SLn (k)/Z(SLn (k)) = P SLn (k).
Theorem 22.4.7. Assume Φ is an irreducible root lattice and k is a field such that Φ is not
of type A1 , B2 or G2 if k = F2 , and Φ is not of type A1 if k = F3 . Then for any g-module V
with root lattice Φ, GAd = GV /Z(GV ) is simple as an abstract group.
Theorem 22.4.8. Let k be an algebraically closed field and V a g-module with Chevalley
group GV . Then GV is a semisimple, connected algebraic group over k. Furthermore, every
semisimple, connected algebraic group over k arises in this way.
It turns out that if k is any field of characteristic p > 0, then every Chevalley group can
be defined over the prime subfield Fp . This describes a large portion of the theory of finite
simple groups, although as noted at the end of Section 22.2, there are exceptional cases.
537
Part V
Generalized Jacobians
538
Chapter 23
Classical Jacobians
The notes in Part V come from the 2016-2017 Galois-Grothendieck Seminar at the University
of Virginia. The general theme of the year’s seminar was “generalized Jacobians”: their
construction and application for algebraic curves over an arbitrary field.
Let C be a curve over a field k. We will almost always assume C is irreducible, projective
and smooth. For the first chapter, k will be the complex numbers C. Our goal is to construct
a new object J, the Jacobian of C, which is a projective algebraic variety over k of dimension
equal to the genus of C. We will see that J is also a group, and in fact is an algebraic group.
This means that since J is a projective (hence complete) algebraic group, it is abelian. Such
an algebraic group is called an abelian variety. When the genus of C is greater than 0, there
will be a canonical embedding C ,→ J.
The class field theory of k(C) = K will also be of interest. In general, for a field K,
class field theory describes the abelian extensions L/K, i.e. finite Galois extensions with
Gal(L/K) abelian, in terms of C itself. In the case of K = k(C) the function field of a
curve, such an extension will be of the form L = k(C 0 ) for a curve C 0 . Then the field
embedding K ⊆ L corresponds to a ramified cover C 0 → C. It turns out that if C 0 → C is
unramified, then it corresponds to a finite index subgroup of the surface group
g
* +
Y
Σg = x1 , y1 , . . . , xg , yg : [xi , yi ] = 1
i=1
where g is the genus of C. Even better, these covers of curves can be realized by covers of
algebraic varieties. When C 0 → C is a ramified cover, the usual notion of Jacobian will be
replaced by a generalized Jacobian, which is isomorphic to a direct product of the Jacobian
of C and another algebraic group.
We will see in the first chapter that when k = C, the theory has powerful connections to
the theory of complex manifolds. In particular, the Jacobian J can be realized as a complex
torus Cg /Λ where Λ is a lattice, and we will have π1 (C)ab ∼
= π1 (J) = Z2g .
539
23.1. Elliptic Functions Chapter 23. Classical Jacobians
ω1
ω2
Π
Lemma 23.1.1. For any choice of basis [ω1 , ω2 ] of Λ, Π(ω1 , ω2 ) is fundamental for Λ.
Lemma 23.1.2. Let Λ be a lattice. Then
(a) If Π is the fundamental domain of Λ, then for any α ∈ C, Πα := Π + α is fundamental
for Λ.
[
(b) If Φ is fundamental for Λ, then C = Φ + `.
`∈Λ
Corollary 23.1.3. Suppose f is an elliptic function with lattice of periods Λ and Φ funda-
mental for Λ. Then f (C) = f (Φ).
540
23.1. Elliptic Functions Chapter 23. Classical Jacobians
γ4 Πα γ2
γ1
α
R R R R
We show that γ1 f (z) dz+ γ3 f (z) dz = 0 and leave the proof that γ2 f (z) dz+ γ2 f (z) dz = 0
for exercise. Consider
Z Z Z 1 Z 1
f (z) dz + f (z) dz = f (α + tω1 )(ω1 dt) + f (α + (1 − t)ω1 + ω2 )(−ω1 dt)
γ1 γ3 0 0
Z 1 Z 0
= ω1 f (α + tω1 ) dt + ω1 f (α + sω1 ) ds since f is elliptic
0 1
Z 1 Z 1
= ω1 f (α + tω1 ) dt − f (α + sω1 ) ds = 0.
0 0
541
23.1. Elliptic Functions Chapter 23. Classical Jacobians
Corollary 23.1.6. Any elliptic function has either a pole of order at least 2 or two poles on
the fundamental domain of its lattice of periods.
Proof. For a pole z0 , we can write f (z) = (z − z0 )m g(z) for some holomorphic function g(z),
with g(z0 ) 6= 0. Then
f 0 (z) g 0 (z)
−1
= (z − z0 ) m + (z − z0 ) .
f (z) g(z)
0
Hence Res ff ; z0 = m. Then the statement follows from Proposition 23.1.5.
but f (k) (z) is also elliptic for all k ≥ 1. If z0 is a pole of f , the same result can be obtained
using f1 instead of f .
If Φ1 and Φ2 are any two fundamental domains for Λ, then for all a1 ∈ Φ1 , there is a
unique a2 ∈ Φ2 such that a2 = a1 + ` for some ` ∈ Λ. Thus Propositions 23.1.5 and 23.1.7
hold for any fundamental domain of Λ, so it follows that ord(f ) is well-defined on the quotient
C/Λ.
Now given any meromorphic function f (z) on C, we would like to construct an elliptic
function F (z) with lattice Λ. Put
X
F (z) = f (z + `).
`∈Λ
There are obvious problems of convergence and (in a related sense) the order of summation.
It turns out we can do this construction with f (z) = z1m , m ≥ 3 though. First, we need the
following result from complex analysis, which can be proven using Cauchy’s integral formula
and Morera’s theorem.
542
23.1. Elliptic Functions Chapter 23. Classical Jacobians
Lemma 23.1.8. Let U ⊆ C be an open set and suppose (fn ) is a sequence of holomorphic
functions on U such that fn → f uniformly on every compact subset of U . Then f is
holomorphic on U and fn0 → f 0 uniformly on every compact subset of U .
Proposition 23.1.9. Let Λ be a lattice with basis [ω1 , ω2 ]. Then the sum
X 1
|ω|s
ω∈Λr{0}
Λ Λ
∆
Λ Λ
Then ∆ is compact, so there exists c > 0 such that |z| ≥ c for all z ∈ ∆. We claim that for
all m, n ∈ Z,
|mω1 + nω2 | ≥ c · max{|m|, |n|}.
The cases when m = 0 or n = 0 are trivial, so without loss of generality assume m ≥ n > 0.
Then n
|mω1 + nω2 | = |m| ω1 + ω2 ≥ |m|c.
m
Hence the claim holds. Set M = max{|m|, |n|} and arrange the sum in question so that the
1
|ω|s
are added in order of increasing M values. Then the sum can be estimated by
∞ ∞
X 1 X 8M X 1
≤ ∼ .
|ω|s M =1 cs M s M =1 M s−1
ω∈Λr{0}
Then Fn (z) is holomorphic on C r Λ and has poles of order n at the points of Λ. Moreover,
Fn is doubly periodic and hence elliptic.
543
23.1. Elliptic Functions Chapter 23. Classical Jacobians
Proof. Fix r > 0 and let Br = Br (0) be the open complex r-ball centered at the origin in C.
Let Λr = Λ ∩ B r be the lattice points contained in the closed r-ball. Then the function
X 1
Fn,r (z) =
ω∈ΛrΛr
(z − ω)n
1 C
is holomorphic on Br . To see this, one has |z−ω| n ≤ |ω|n for some constant C and for all
since the series is absolutely convergent and we can rearrange the terms.
This shows that elliptic functions exist and more specifically that for each n ≥ 3, there
is at least one elliptic function of order n. Unfortunately the previous proof won’t work
to construct an elliptic function of order 3. However, Weierstrass discovered the following
elliptic function.
1 X 1 1
℘(z) = 2 + − .
z (z − w)2 ω 2
ω∈Λr{0}
Theorem 23.1.11. For any lattice Λ, ℘(z) is an elliptic function with poles of order 2 at
the points of Λ and no other poles. Moreover, ℘(−z) = ℘(z) and ℘0 (z) = −2F3 (z).
1 X 1 1
℘(−z) = + −
(−z)2 (−z − ω)2 ω 2
ω∈Λr{0}
1 X 1 1
= 2+ − = ℘(z)
z (z − (−ω))2 (−ω)2
−ω∈Λr{0}
544
23.1. Elliptic Functions Chapter 23. Classical Jacobians
Finally, proving ℘(z) is doubly periodic is difficult since we don’t necessarily have absolute
convergence. However, one can reduce to proving ℘(z + ω1 ) = ℘(z) = ℘(z + ω2 ). Then using
the formula for ℘0 (z), we have
d
[℘(z + ω1 ) − ℘(z)] = −2F3 (z + ω1 ) + 2F3 (z)
dz
= −2F3 (z) + 2F3 (z) = 0
Lemma 23.1.12. Let ℘(z) be the Weierstrass ℘-function for a lattice Λ ⊆ C and let Π be
the fundamental domain of Λ. Then
(1) For any u ∈ C, the function ℘(z) − u has either two simple roots or one double root
in Π.
(2) The zeroes of ℘0 (z) in Π are simple and they only occur at ω21 , ω22 and ω1 +ω
2
2
.
(3) The double roots occur exactly when ℘0 (u) = 0, so use (2).
We now prove that any elliptic function can be written in terms of ℘(z) and ℘0 (z).
Theorem 23.1.13. Fix a lattice Λ ⊆ C and let E(Λ) be the field of all elliptic functions
with lattice of periods Λ. Then E(Λ) = C(℘, ℘0 ).
Proof. Take f (z) ∈ E(Λ). Then f (−z) ∈ E(Λ) as well and thus we can write f (z) as the
sum of an even and an odd elliptic function:
545
23.1. Elliptic Functions Chapter 23. Classical Jacobians
f (z)
Λ, since then by Corollary 23.1.6, ϕ(℘(z)) is holomorphic and then by Proposition 23.1.4
it is constant. Suppose f (a)
= 0 for a some zero of order m. Consider ℘(z) = u. If
ω1 ω2 ω1 +ω2
u 6= ℘ 2 , ℘ 2 , ℘ 2
then ℘(z) = u has precisely two solutions in the fundamental
parallelogram, z = a and z = a∗ where
ω1 + ω2 − a if a ∈ Int(Π)
∗
a = ω1 − a if a is parallel to ω1
ω2 − a if a is parallel to ω2 .
(Notice that since f is even, f (a) = 0 implies f (a∗ ) = 0 as well.) Moreover, if orda f = 0 then
orda∗ f = m. Note that a = a∗ holds precisely when a is in the set Θ := 0, ω21 , ω22 , ω1 +ω
2
2
.
Let Z (resp. P ) be the set of zeroes (resp. poles) of f (z) in Π. Then the assignment
a 7→ a∗ is in fact an involution on Z and P , so we can write
where the Zi0 and Pi0 are the 2-element orbits of the involution and the Zj00 and Pj00 are the
1-element orbits. Of course then s, v ≤ 3. For a0i ∈ Zi0 , set orda0i f = m0i and for a00j ∈ Zj00 ,
set orda00i f = m00i , which is even. Likewise, for b0i ∈ Pi0 , set ordb0i f = n0i and for b00j ∈ Pj00 , set
ordb00i f = n00i which is even. Then we define ϕ(℘(z)) by
0 m0i 00 m00
Qr Qs
j /2
i=1 (℘(z) − ℘(ai )) j=1 (℘(z) − ℘(aj ))
ϕ(℘(z)) = Qu 0 n0i
Qv 00 nj
.
i=1 (℘(z) − ℘(bi )) j=1 (℘(z) − ℘(bj ))
546
23.2. Elliptic Curves Chapter 23. Classical Jacobians
Let ℘(z) be the Weierstrass ℘-function for Λ. Then ℘0 (z)2 is an even elliptic function, so
by Theorem 23.1.13, ℘0 (z)2 ∈ C(℘). On a small enough neighborhood around z0 = 0,
1 X 1 1
℘(z) − 2 = −
z (z − ω)2 ω 2
ω∈Λr{0}
1 1 2z 3z 2
= + + 4 + ...
(z − ω)2 ω2 ω3 ω
2
1 1 2z 3z
=⇒ 2
− 2 = 2 + 4 + ...
(z − ω) ω ω ω
1
P
where Gm = Gm (Λ) := ω∈Λr{0} ω m .
These Gm are examples of modular forms.
1
P
Definition. The series Gm (Λ) = ω∈Λr{0} ω m is called the Eisenstein series for Λ of
weight m.
547
23.2. Elliptic Curves Chapter 23. Classical Jacobians
u3 = ℘ ω1 +ω
2
2
are distinct roots.
Thus (x, y) = (℘(z), ℘0 (z)) determine an equation y 2 = 4x3 −g2 x−g3 which is the defining
equation for an elliptic curve E0 over C. Let E = E0 ∪ {[0, 1, 0]} ⊆ P2 be the projective
closure of E0 . The point [0, 1, 0] is sometimes denoted ∞.
Theorem 23.2.4. The map
ϕ : C/Λ −→ E(C)
(
[℘(z), ℘0 (z), 1], z ∈
6 Λ
z + Λ 7−→ ϕ(z + Λ) =
[0, 1, 0], z∈Λ
is a bijective, biholomorphic map.
Proof. Assume z1 , z2 ∈ C are such that z1 + Λ 6= z2 + Λ. Without loss of generality we may
assume z1 , z2 ∈ Π, the fundamental domain of Λ (otherwise, translate). If ℘(z1 ) = ℘(z2 ) and
℘0 (z1 ) = ℘0 (z2 ), then
with the notation of Theorem 23.1.13, we must have z2 = z1∗ 6= z1 and
ω1 ω2 ω1 +ω2
. Since ℘0 (z) is odd, we get ℘0 (z1 ) = ℘0 (z2 ) = −℘0 (−z2 ) =
thus z1 , z2 6∈ Θ = 0, 2 , 2 , 2
−℘0 (z1 ), but this implies ℘(z1 ) = 0, contradicting z1 6∈ Θ. Therefore ϕ is one-to-one.
Next, we must show that for any (x0 , y0 ) ∈ E(C), x0 = ℘(z) and y0 = ℘0 (z) for some
z ∈ C. If ℘(z1 ) = x0 , then it’s clear that ℘0 (z1 ) = y0 or −y0 . Now one shows as in the
previous paragraph that we must have ℘0 (z1 ) = y0 .
Now consider F (x, y) = y 2 − p(x), where p(x) = 4x3 − g2 x − g3 . If (x0 , y0 ) satisfies
F (x0 , y0 ) = 0 and y0 6= 0, then ∂F∂y
(x0 , y0 ) 6= 0 and thus the assignment (x, y) 7→ x is a local
chart about (x0 , y0 ). Likewise, (x, y) 7→ y defines a local chart about (x0 , y0 ) when x0 6= 0.
Finally, we conclude by observing that a locally biholomorphic map is biholomorphic.
548
23.2. Elliptic Curves Chapter 23. Classical Jacobians
E : ZY 2 = aX 3 + bX 2 Z + cXZ 2 + dZ 3 .
The single point at infinity, [0, 1, 0], can be studied by dehomogenizing via the coordinates
z̃ = YZ and x̃ = X
Y
, which yield
We have shown that a lattice Λ ⊆ C determines elliptic functions ℘(z) and ℘0 (z) that satisfy
℘0 (z)2 = 4℘(z)3 − g2 ℘(z) − g3 and that this polynomial expression has no multiple roots.
Therefore the mapping z 7→ (℘(z), ℘0 (z)) determines a bijective correspondence C/Λr{0} →
E(C) r {∞} which can be extended to all of C/Λ → E(C) (this is Theorem 23.2.4). There
is a natural group structure on C/Λ induced from C, but what is not so obvious is that E(C)
also possesses a group structure, the so-called “chord-and-tangent method”.
Let E be an elliptic curve over an arbitrary field k, let O ∈ E(k) be the point at infinity
and fix two points P, Q ∈ E(k). In the plane P2k , there is a unique line containing P and Q;
call it L. (If P = Q, then take L to be the tangent line to E at P .) By Bézout’s theorem,
E ∩ L = {P, Q, R} for some third point R ∈ E(k), which may not be distinct from P and Q
if multiplicity is counted. Let L0 be the line through R and O and call its third point R0 .
Q
P
P +Q
549
23.2. Elliptic Curves Chapter 23. Classical Jacobians
α β
C/Λ E(C)
ϕ
where α and β are the respective group operations. Since C/Λ × C/Λ is a topological group,
it’s enough to show the diagram commutes on a dense subset of C/Λ × C/Λ. Consider
e = {(u1 , u2 ) ∈ C2 | u1 , u2 , u1 ± u2 , 2u1 + u2 , u1 + 2u2 6∈ Λ}.
X
Then X e ∼
= C2 so X = X e mod Λ × Λ is dense in C/Λ × C/Λ. Take (u1 + Λ, u2 + Λ) ∈ X
and set u3 = −(u1 + u2 ). Then u1 + u2 + u3 = 0 in C/Λ. Set P = ϕ(u1 ), Q = ϕ(u2 ) and
R = ϕ(u3 ) ∈ E(C). By the assumptions on X, the points P, Q, R are distinct. We want to
show ϕ(u1 + u2 ) = ϕ(u1 ) + ϕ(u2 ) = P + Q. Since ℘(z) is even and ℘0 (z) is odd, we see that
ϕ(−z) = −ϕ(z) for all z ∈ C/Λ. Thus ϕ(u1 + u2 ) = −ϕ(−(u1 + u2 )) = −R so we need to
show P + Q + R = O, i.e. P, Q, R are colinear. Since u1 6= u2 , the line P Q is not vertical,
so there exist a, b such that ℘0 (ui ) = a℘(ui ) + b for i = 1, 2. Consider the elliptic function
Then on the fundamental domain Π, f only has a pole at 0, so ord0 f = −3. Also, u1 and u2
are distinct zeroes of f , so there is a third point ω ∈ Π such that deg(f ) = u1 +u2 +ω−3·0 = 0,
i.e. u1 + u2 + ω = 0. Solving for ω, we get ω = −(u1 + u2 ) = u3 . It follows that R = ϕ(u3 )
is on the same line as P and Q, so we are done.
The compatibility of the group operations of C/Λ and E(C) is highly useful. For example,
fix N ∈ N and let
E[N ] = {P ∈ E(C) | [N ]P = O},
where [N ]P = P
| + .{z
. . + P}. The points of E[N ] are called the N -torsion points of E. For
N
N = 2, the points P such that P = −P are exactly the intersection points of E with the
x-axis along with O = [0, 1, 0]:
550
23.2. Elliptic Curves Chapter 23. Classical Jacobians
In general, one can show that #E[N ] = N 2 . This is hard to see from the geometric
picture, but working with the isomorphism E(C) ∼ = C/Λ from Theorem 23.2.6, we see
that since C/Λ = R/Z × R/Z as an abelian group, the N -torsion is given by (C/Λ)[N ] =
1
N
Z/Z × N1 Z/Z. This is a group of order N 2 .
A morphism in the category of elliptic curves is called an isogeny. Explicitly, ϕ : E1 → E2
is an isogeny between two elliptic curves if it is a (nonconstant) morphism of schemes that
takes the basepoint O1 ∈ E1 to the basepoint O2 ∈ E2 .
Proposition 23.2.7. Suppose Λ1 , Λ2 ⊆ C are lattices and f : C/Λ1 → C/Λ2 is a holomor-
phic map. Then there exist a, b ∈ C such that aΛ1 ⊆ Λ2 and
f (z mod Λ1 ) = az + b mod Λ2 .
Proof. As topological spaces, C/Λ1 and C/Λ2 are complex tori with the same universal
covering space C, so any f : C/Λ1 → C/Λ2 lifts to F : C → C making the diagram
commute:
F
C C
π1 π2
C/Λ1 C/Λ2
f
Since covers are local homeomorphisms, it follows that F is holomorphic as well. Thus for
any z ∈ C, ` ∈ Λ1 ,
551
23.2. Elliptic Curves Chapter 23. Classical Jacobians
Corollary 23.2.8. Any holomorphic map f : C/Λ1 → C/Λ2 is, up to translation, a group
homomorphism. In particular, if f (0) = 0 then f is a homomorphism.
Corollary 23.2.9. For any elliptic curve E, the group of endomorphisms End(E) has rank
at most 2.
End(E) = {f : E → E | f is an isogeny}
= {f : C/Λ → C/Λ | f is holomorphic and f (0) = 0} by Corollary 23.2.8
= {z ∈ C | zΛ ⊆ Λ}
= {z ∈ C | z(Z + Zτ ) ⊆ (Z + Zτ )}
⊆ Z + Zτ.
End(E) = Z.
End(E) is an order O in some imaginary quadratic number field K/Q. In this case, E
is said to have complex multiplication.
552
23.3. The Classical Jacobian Chapter 23. Classical Jacobians
For an elliptic curve E defined by the equation y 2 = f (x), fix a holomorphic differential
form ω on E(C). (In general, the space of holomorphic differential forms on a curve has
dimension equal to the genus of the curve, so in the elliptic curve case, there is exactly one
such ω, up to scaling.)
Example 23.3.3. Under the map ϕ : C/Λ → E(C), z 7→ (x, y) = (℘(z), ℘0 (z)), we see that
dx = ℘0 (z) dz = y dz
so ω = dx
y
is a differential form on E(C). In fact, ω = f dx 2
0 (x) , where E is defined by y = f (x),
is holomorphic because f 0 (x) 6≡ 0. This differential form is also holomorphic at O = [0, 1, 0],
so up to scaling, this is the unique holomorphic form on E.
553
23.3. The Classical Jacobian Chapter 23. Classical Jacobians
For a more functorial description, let VE = Γ(E, ΩE ) be the space of all holomorphic
differential forms on E. If γ is a curve in E(C), there is an associated linear functional
ϕγ ∈ VE∗ defined by
ϕγ : VE −→ C
Z
ω 7−→ ω.
γ
Fixing the basepoint O ∈ E(C), the lattice of periods for E can be written
Definition. The Jacobian of an elliptic curve E is the quotient J(E) = VE∗ /Λ.
For each point P ∈ E(C), the coset ϕγ + Λ is an element of the Jacobian, where γ is a
path from O to P . This defines an injective map i : E ,→ J(E).
i1 i2
J(E1 ) J(E2 )
τ
Thus σ ∗∗ ϕγ1 = ϕσ(γ1 ) . If γ1 is a closed curve through O1 , then σ(γ1 ) is a closed curve passing
through O2 = σ(O1 ). Hence if ΛE1 , ΛE2 are the lattices of periods for E1 , E2 , respectively,
we have σ ∗∗ (λE1 ) ⊆ ΛE2 . So σ ∗∗ factors through the quotients, defining τ :
Lemma 23.3.5. For any elliptic curve E, the inclusion i : E ,→ J(E) induces an isomor-
phism
i∗ : π1 (E, O) −→ π1 (J(E), i(O)).
554
23.3. The Classical Jacobian Chapter 23. Classical Jacobians
Unfortunately, the construction of the Jacobian given so far is not algebraic so it would
be hard to carry over to curves over an arbitrary ground field. To construct Jacobians
algebraically, we will prove Abel’s theorem:
Theorem 23.3.6 (Abel). Suppose Λ ⊆ C is a lattice with fundamentalP domainPΠ and take
any set {ai } ⊂ Π such that there are integers mi ∈ Z satisfying mi = 0 and mi ai ∈ Λ.
Then there exists an elliptic function f (z) whose set of zeroes and poles is {ai } and whose
orders of vanishing/poles are ordai f = mi .
2 2
One has |eπi(n τ +2nz) | = e−π(n im τ +2n im z)
for any z ∈ C, which implies that the above
series converges absolutely.
Properties (2) and (3) together say that θ(z) is what’s known as a semielliptic function.
For our purposes, this will be good enough. Notice that for z = 1+τ 2
, we have
1+τ 1+τ
θ =θ − + (1 + τ )
2 2
πi(τ +2(− 1+τ )) 1 + τ
=e 2 θ −
2
πi 1+τ 1+τ
=e θ − = −θ .
2 2
1+τ
Thus z = 2
is a zero of θ(z).
Lemma 23.3.8. All zeroes of θ(z, t) are simple and are of the form 1+τ
2
+ ` for ` ∈ Λ.
555
23.3. The Classical Jacobian Chapter 23. Classical Jacobians
Proof. Given such a set {ai } ⊂ Π, let x1 , . . . , xn be the list of all ai with mi > 0, listed with
repetitions corresponding to the number mi . For example, if m1 = 2 then x1 = x2 = a1 .
Likewise, letPy1 , . . . , yn be the list of all ai with mi < 0, once again with repetitions. By the
hypothesis mi = 0, there are indeed an equal number of each. Set
Qn (xi )
θ (z)
f (z) = Qi=1
n (yi ) (z)
.
i=1 θ
Then by Lemma 23.3.9, f (z + 1) = f (z). On the other hand, the lemma also gives
Qn (xi )
i=1 θ (z + τ )
f (z + τ ) = Q n (y i ) (z)
i=1 θ
Pn Pn
= e2πi( i=1 xi − i=1 yi ) f (z)
P
= e2πi mi ai
f (z)
X
= f (z) since mi ai = 0.
θ0 (z)
Z
1
dz
2πi ∂Π θ(z)
This implies
556
23.3. The Classical Jacobian Chapter 23. Classical Jacobians
Corollary 23.3.10. The map Ψ : Div0 (E) → C/Λ induces an isomorphism Pic0 (E) ∼
= C/Λ.
Proof. One can prove that Ψ is a surjective group homomorphism. Moreover, Abel’s theorem
(23.3.6) implies that ker Ψ = PDiv(E).
Consider the map iO : E → Div0 (E) that sends P 7→ P −O. This fits into a commutative
diagram:
Div0 (E)
Ψ
iO C/Λ
ψO
E
557
23.3. The Classical Jacobian Chapter 23. Classical Jacobians
Pic0 (E)
Ψ
iO C/Λ
ψO
E
558
23.4. Jacobians of Higher Genus Curves Chapter 23. Classical Jacobians
Pic0 (C)
Ψ
iO J(C)
ψO
C
However, this time not every map is a bijection. In particular, dim C = 1 < g = dim J(C).
To remedy this, let C g be the g-fold product of C and consider the map
ψ g : C g −→ J(C)
(P1 , . . . , Pg ) 7−→ ψ(P1 ) + . . . + ψ(Pg )
559
23.4. Jacobians of Higher Genus Curves Chapter 23. Classical Jacobians
There is still work to do to show that the natural map C g → Pic0 (C) is surjective.
It turns out that J(C) is birationally equivalent to the symmetric power C (g) = C g / ∼,
where (P1 , . . . , Pg ) ∼ (Pσ(1) , . . . , Pσ(g) ) for any permutation σ ∈ Sg . Jacobi proved that this
birational equivalence is enough to endow Pic0 (C) ∼ = J(C) with the structure of an algebraic
group.
560
Chapter 24
561
24.1. Moduli and Local Symbols Chapter 24. Maps From Curves
562
24.1. Moduli and Local Symbols Chapter 24. Maps From Curves
Now if g, g ∈ k(X) with associated functions gP , gP0 as above, the product gP gP0 satisfies
0 ×
563
24.1. Moduli and Local Symbols Chapter 24. Maps From Curves
564
24.1. Moduli and Local Symbols Chapter 24. Maps From Curves
by (d) for the local symbol (f, −)P . Hence (Trπ f, −)P 0 is a local symbol on X 0 r S 0 .
Let π : X → X 0 be a ramified covering of curves and suppose f 0 : X 0 r S 0 → G0 is a
morphism to a commutative algebraic group G0 .
Definition. For g ∈ k(X), the norm of g is defined to be its norm in the field extension
k(X)/π(k(X 0 )):
Nπ g := Nk(X)/π(k(X 0 )) (g).
Proof. By Proposition 24.1.1, it’s enough to show the formula above defines a local Q symbol.
Note that the norm map is continuous and satisfies a local-global principle N = P ∈X NP ,
where NP is the norm of the extension of complete local fields at P → P 0 . Thus if a
modulus m supported on S exists P and g ≡ 1 mod m then P Nπ g ≡ 1 mod m0 . Moreover,
for any g ≡P1 mod m, if (g) = vP (g)P then π(g) = vP (g)π(P ). We must show
vP 0 (Nπ g) = P →P 0 vP (g) for all g ≡ 1 mod m. If w1 , . . . , wd are the distinct extensions of
vP from k(X 0 )vP to k(X) ⊗k(X 0 ) k(X 0 )vP , then we have
d
! d
Y X
vP 0 (Nπ g) = vP 0 Ni (g) = vP 0 (Ni (g))
i=1 i=1
d
X
= wi (g) by algebraic number theory
i=1
= π(vP (g)).
565
24.1. Moduli and Local Symbols Chapter 24. Maps From Curves
(f 0 , Nπ g)P 0 = (f 0 , Nπ h)P 0
X
=− (f, Nπ h)Q0 by condition (d) for the local symbol
Q0 6∈S 0
X
=− (f 0 ◦ π, h)P by the above
Q0 6∈S 0
X
= (f 0 ◦ π, h)P by condition (d) again
P →P 0
X
= (f ◦ π, g)P .
P →P 0
566
24.2. The Main Theorem Chapter 24. Maps From Curves
Example 24.2.2. Let G = Ga = k, the additive group of the field. We can show Theo-
rem 24.2.1 holds for any curve X over k and any map f : X → Ga . For this, it suffices
to prove the existence
of a local symbol (by Proposition 24.1.1), which we claim is given
by (f, g)P = resP f gdg where resP denotes the residue at P , as in Section 17.5. For each
P ∈ S, set nP = 1−vP (f ) and define m = P ∈S nP P . Let’s verify that (f, g)P = resP f gdg
P
(d)
follows
immediately from the residue formula (Theorem 17.5.13). Hence (f, g)P =
f dg
resP g is a local symbol and m is a modulus for f .
One
consequence
of this construction on G = Ga is that the local symbol (f, g)P =
f dg
resP g commutes with the Frobenius:
Proposition 24.2.3. If char k = p > 0, then for any map f : X r S → k, g ∈ k(X)× and
P ∈ X, p p
f dg f dg
resP = resP .
g g
Proof. Let θ : Ga → Ga , x 7→ xp be the Frobenius map. Then f factors through θ:
f
X rS Ga
Ga
567
24.2. The Main Theorem Chapter 24. Maps From Curves
Example 24.2.4. Let G = Gm = k × be the multiplicative group of the field and suppose
f : X 99K Gm is a rational map and S is the finite set ofPpoints of X at which f is not
regular, i.e. the set of zeroes and poles of f . We claim m = P ∈S P is a modulus for f , that
is, nP = 1 for all P ∈ S works. To show this, we show the assignment
fn
(f, g)P = (−1)mn (P ) where n = vP (g), m = vP (f )
gm
n
is a local symbol for f . Note that vP gfm = mn − mn = 0 so this symbol is well-defined.
Also, (f, g)P satisfies the following properties:
(ii) (−f, f )P = 1
(iii) (1 − f, f )P = 1.
P
To prove this is a local symbol corresponding to the modulus m = P ∈S P , keep in mind
that the abelian group structure on Gm is multiplicative with identity 1.
(a) becomes (f, gg 0 )P = (f, g)P (f, g 0 )P in multiplicative notation, which is clear from the
definition of the symbol.
(b) Suppose vP (1−g) ≥ nP = 1. Then vP (g) = 0 and g(P ) = 1, so (f, g)P = g(P )−m = 1,
the multiplicative identity in Gm .
(c) Take P 6∈ S. Then by the definition of S, vP (f ) = 0 and we have (f, g)P = f (P )n =
f (P )vP (g) , the correct equality in multiplicative notation.
(d) View g ∈ k(X)× as a rational function g : X 99K P1 . If g ≡ a 6= 0 is constant, then
for all P ∈ X, (f, g)P = g(P )−vP (f ) = a−vP (f ) so we have
Y Y P
(f, g)P = a−vP (f ) = a P ∈X −vP (f ) = a− deg((f ))
P ∈X P ∈X
which is a0 = 1 by Corollary 17.2.6. So the formula holds for constant functions. More
generally, any g ∈ k(X)× may be viewed as a branched cover g : X → P1 . Set E =
k(P1 ) = k(t) and F = k(X) so that g induces a field extension F/E with norm map
N = NF/E : F × → E × .
Lemma 24.2.5. Let t denote the identity map on E. Then for all P ∈ P1 ,
Y
(f, g)Q = (N f, t)P .
Q→P
×
Proof. Extend the definition of (f, g)P to the completions E d and Fb× = k(X)
bP = k(t) [ .
P Q Q
It is immediate that Q
(f, g)P is still multiplicative in both arguments. By algebraic number
theory, N = NF/E = Q→P NQ where NQ = NFbQ /EbP . Thus we have
Y
(N f, t)P = (NQ f, t)P
Q→P
568
24.2. The Main Theorem Chapter 24. Maps From Curves
so it suffices to show (f, g)Q = (NQ f, t)P for all Q → P . Fix such a Q ∈ g −1 (P ). By
multiplicativity of the local norm NQ , we may assume f and g are uniformizers of FbQ and
EbP , respectively, i.e. that vQ (f ) = vP (g) = 1, FbQ = k((f )) and E
bP = k((g)). Then
(NQ f )n NQ f
(NQ f, g)P = (−1)nm m
(P ) = − (P )
g g
fe ef
e
(f, g)Q = (−1)em (Q) = (−1) (Q).
gm g
Proposition 24.2.7. Suppose f, g ∈ k(X)× and Sf , Sg are, respectively, their sets of zeroes
and poles. If Sf ∩ Sg = ∅, then f ((g)) = g((f )).
569
24.2. The Main Theorem Chapter 24. Maps From Curves
n
Proof. Let (f, g)P = (−1)nm gfm (P ) be the local symbol from Example 24.2.4. Then
Y Y
f ((g)) = f (P )vP (g) = (f, g)P
P 6∈Sf P 6∈Sf
Y Y Y 1
and g((f )) = g(P )vP (f ) = (g, f )P = P 6∈ Sg .
P 6∈Sg P 6∈Sg
(f, g)P
Also, for all P outside Sf ∪ Sg , (f, g)P = 1 so putting these together, we have
Y Y Y 1
1= (f, g)P = (f, g)P (f, g)P = g((f )).
P ∈X P ∈Sf P ∈Sg
f ((g))
π ∗ : X 0 −→ X (n)
X
P 0 7−→ eP P.
P →P 0
F : Y n −→ G
n
X
(P1 , . . . , Pn ) 7−→ f (Pi ).
i=1
Clearly the sum on the right is invariant under the symmetric group action, so F descends
to a map F 0 : Y (n) → G. Consider the diagram
Y Yn
F
π
π∗ F0
Y0 Y (n) G
Trπ f
570
24.2. The Main Theorem Chapter 24. Maps From Curves
P
Proposition 24.2.9. Let m = P ∈S nP P be a modulus and f : X r S → G a regular map.
Then m is a modulus for f if and only if for all g ∈ k(X)× such that g ≡ 0 mod m, Trg f
is constant.
Proof. For a 6= ∞, let (g)a be the divisor of zeroes of g − a and let (g)∞ be the divisor of
poles of g. Then for all a ∈ P1 r {0},
!
X
Trg f (a) = f eP P = f ((g)a ).
P →a
Proof. Fix g ≡ 0 mod m. By Proposition 24.2.9 it is enough to show that the function
f 0 = Trg f is constant. To do this, we will show (f 0 )∗ ωi = 0 for all 1 ≤ i ≤ r. To see why this
implies the trace function is constant, notice that 0 = (f 0 )∗ ωi (P ) = ωi (f 0 (P )) so the map of
tangent spaces induced by f 0 is the zero map and thus f 0 is constant.
Our first step is to show (f 0 )∗ ωi = Trg ωi for all 1 ≤ i ≤ r. Let π : Y → X be a finite
Galois cover and consider the Galois, separable map g ◦ π : Y → P1 . Since it is separable, the
induced map ω 7→ (g ◦ π)∗ ω is injective. Thus if we can show that π ∗ g ∗ (f 0 )∗ ωi = π ∗ g ∗ Trg ωi ,
it will follow that (f 0 )∗ ωi = Trg ωi . To do this, we observe that all lifts along π are Galois
conjugate, or more generally:
571
24.2. The Main Theorem Chapter 24. Maps From Curves
π g
Proposition 24.2.11. Suppose Y → − X → − X 0 are Galois covers of curves with G =
Gal(k(Y )/k(X 0 )) and H = Gal(k(Y )/k(X)) ≤ G. Then for all P ∈ X,
X
g −1 (g(P )) = [k(X) : k(X 0 )]i π ◦ σj (Q)
j
where [k(X) : k(X 0 )]i denotes the inseparable degree, π(Q) = P and {σj } is a transversal of
G/H.
Proof. Set h = g ◦ π and note that
X X
h−1 (g(P )) = nσ σ(Q) = nσ(Q)
σ∈G σ∈G
for a fixed n ∈ N, since π is Galois. On the other hand, h(h−1 (P 0 )) = [k(Y ) : k(X 0 )]P 0 for
all P 0 ∈ X 0 , so we must have h−1 (g(P )) = σ∈G σ(Q). Now to get g −1 (g(P )), apply π:
P
X
π(h−1 (g(P ))) = π ◦ σ(Q)
σ∈G
XX
= π ◦ α ◦ σj (Q)
α∈H j
XX
= π ◦ σj (Q) since α ∈ H
α∈H j
X
= [k(Y ) : k(X)] π ◦ σj (Q).
j
On the other hand, π(h−1 (g(P ))) = [k(Y ) : k(X)]g −1 (g(P )) so dividing by [k(Y ) : k(X)],
we get the result.
Corollary 24.2.12. For any f ∈ k(X) and g ∈ k(X)× , (Trg f ) ◦ g ◦ π = j f ◦ π ◦ σj .
P
π g
− P1 , we have f 0 ◦ g ◦ π = j f ◦ π ◦ σj , so
P
Applying this to our maps Y →− X→
X
π ∗ g ∗ Trg αi = σj∗ π ∗ αi .
j
572
24.2. The Main Theorem Chapter 24. Maps From Curves
573
24.3. A Word About Symmetric Products Chapter 24. Maps From Curves
Proposition 24.3.1. If G is a finite group acting on a variety V such that every orbit of G
is contained in any affine open subset of V , then the quotient V /G is a variety.
Proof. Let S1 , . . . , Sk denote the orbits. Since each orbit is invariant under the G-action, we
can cover V by a collection of affine open sets {Ui0 } with Si ⊆ Ui0 . Set Ui = Ui0 /G. Then Ui
is affine: if Ui0 = Spec Ai for a ring Ai , then Ui ∼= Spec AG i , the affine scheme on the ring of
G-invariants of Ai . It is easy to see that V /G is covered by these affine open sets Ui .
T
Finally, we must show the diagonal map ∆V /G is closed. It suffices to show ∆V /G U
i,j i × Uj
is closed in Uk × U` for all k, `. Consider the commutative diagram
π
V V /G
∆V ∆V /G
V ×V V /G × V /G
π×π
T
0 0
Since V is a variety, ∆V U
i,j i × U j is closed in Uk0 × U`0 for all k, `. Since π : V → V /G
T
0 0
is a finite map, (π × π) ◦ ∆V U i × Uj is closed in Uk × U` , but by commutativity, this
T i,j
is precisely ∆V /G i,j Ui × Uj .
Remark. If dim V > 1, V /G may be a singular variety. For example, when X is a curve
and dim X > 1, the symmetric product X (n) is always singular.
574
Chapter 25
575
25.1. Normalization Chapter 25. Curves with Singularities
25.1 Normalization
To discuss singular curves, we introduce a process of “normalization” to get rid of singularities
in some way. The goal will be to prove a Riemann-Roch theorem and describe a canonical
divisor by way of meromorphic differentials on a singular curve.
As a motivating example, consider the irreducible algebraic curve X 0 defined by the
equation y 2 − x3 = 0 on an affine patch and compactified by adding a point O at infinity.
Let O0 be the structure sheaf of X 0 , which has local rings
(
k[x2 , x3 ], P = O
OP0 =
k[x]P , P 6= O.
Then k[x2 , x3 ] is not a regular local ring (the dimension of the tangent space at P = O is 2;
see Section 6.2), so X 0 is singular at O.
In general, let X 0 be an irreducible algebraic curve with singular locus S 0 and field of
rational functions K = k(X 0 ). For each Q ∈ X 0 , let OQ denote the integral closure of OQ
0
in
K.
For any Q ∈ S 0 , note that OQ = P ∈ϕ−1 (Q) OP where OP is the local ring of X at P .
T
576
25.1. Normalization Chapter 25. Curves with Singularities
Example 25.1.4. For X 0 defined by y 2 − x3 = 0 and the point Q = O, the chain above
(with n = 2) is
0
k + r2Q ⊆ k + cQ ⊆ OQ ⊆ k + rQ ⊆ OQ
2 2 3 2 3
k + hxi ⊆ k[x , x ] = k[x , x ] ⊆ k + hxi ⊆ k[x].
Proof. The sheaf axioms for O0 are easily checked. To prove X 0 is a curve, it suffices to
prove the case when X is affine, say X = Spec A. In this case let A0 = Q∈X 0 OQ 0
T
and set
0 0 0
Y = Spec A . It is clear that A ⊆ A. Moreover, by the chain condition on OQ , there exists
an f ∈ A such that f A ⊆ A0 ⊆ A so in particular A0 is finitely generated as a k-algebra.
Since A is an A0 -algebra and is finitely generated, it follows from Lemma 2.1.2 that A/A0 is
an integral extension. Hence dim A0 = dim A = 1 by Theorem 2.3.7 so Y is a curve. In fact,
since A is integrally closed (X is nonsingular), A is precisely the integral closure of A0 in
its fraction field, which shows X = Spec A is the normalization of Y . Finally, one can show
that X 0 ∼= Y so X 0 is also an affine curve, finishing the proof.
577
25.1. Normalization Chapter 25. Curves with Singularities
is a bijection.
Therefore we may treat singular curves formally as nonsingular curves equipped with
‘singular data’ in the form of S ⊆ X with a relation R on S.
Example
P 25.1.7. Let X be a complete, irreducible, nonsingular curve over k. For a modulus
m= nP P on X with support S = {P ∈ X | nP 6= 0} (we will assume |S| ≥ 2), define
the singular curve X 0 associated to the data (X, S, R) where R identifies the entire set S to
a point Q ∈ X 0 . Let cQ = {f ∈ K | f ≡ 0 mod m} and OQ 0
= k + cQ , so that c is the
0
conductor of O in O. We denote the resulting singular curve by Xm , which has a unique
singular point Q.
In our example X 0 : y 2 − x3 = 0, the singular curve X 0 is isomorphic to Xm where X is
the normalization of X 0 and m = 2O.
578
25.2. Singular Riemann-Roch Chapter 25. Curves with Singularities
L0 (D) = H 0 (X 0 , L0 (D))
I 0 (D) = H 1 (X 0 , L0 (D))
`0 (D) = h0 (X 0 , L0 (D)) = dim L0 (D)
i0 (D) = h1 (X 0 , L0 (D)) = dim I 0 (D).
When X 0 = Xm for a modulus m, we will also write these as Lm (D), Im (D), `m (D) and im (D),
respectively.
Proposition 25.2.1. For any divisor D on X, `0 (D) and i0 (D) are finite.
0
Proof. Both L(D) and OQ are coherent sheaves, so this follows from the proof of Proposi-
tion 17.3.11.
Theorem 25.2.3 (Riemann-Roch for Singular Curves). Let X 0 be a singular curve with
normalization X. Then for any divisor D on X prime to S,
Proof. For a divisor D prime to S and any point P 6∈ S 0 , let Q0 = L0 (D + P )/L0 (D) be the
cokernel sheaf in the exact sequence
0 → L0 (D) → L0 (D + P ) → Q0 → 0.
Then as in the proof of Theorem 17.3.12, Q0 is 0 away from P and has dimension 1 at
P . Thus the same argument as in that proof shows that χ0 (D + P ) = χ0 (D) + 1, where
χ0 (D) = `0 (D) − i0 (D). Thus it suffices to prove the theorem for D = 0. In this case, we have
L0 (D) = O0 and since O0 is the pushforward of O along the normalization map X → X 0 ,
there is a short exact sequence
0 → O0 → O → O/O0 → 0.
579
25.2. Singular Riemann-Roch Chapter 25. Curves with Singularities
Thus by additivity, χ0 (O0 ) + χ0 (O/O0 ) = χ0 (O). Further, as in the proof of Theorem 17.3.12,
O/O0 is a skyscraper sheaf concentrated on S 0 , so
X X
dim H 0 (X 0 , O/O0 ) = dim(OQ /OQ0
)= δQ = δ.
Q∈S Q∈S
This shows that χ0 (O0 ) + δ = χ0 (O), but by the standard Riemann-Roch theorem, χ(O) =
1 − g so to finish, it suffices to observe χ0 (O) = χ(O) by Proposition 25.2.2.
580
25.3. Differentials on Singular Curves Chapter 25. Curves with Singularities
Ω0Q = {ω ∈ Dk (OX,Q
0 0
) | hf, ωiQ = 0 for all f ∈ OQ },
P
where hf, ωiQ = P ∈ϕ−1 (Q) resP (f ω) is the residue pairing at Q.
(a) ΩQ ⊆ Ω0Q .
0
(b) hOQ , Ω0Q iQ = 0.
(c) hOQ , ΩQ iQ = 0.
0
Thus the vector spaces OQ /OQ and Ω0Q /ΩQ are dual to each other with respect to h·, ·iQ .
This defines a sheaf of differentials Ω0 (D) on X 0 for which Ω0 (D) = H 0 (X 0 , Ω0 (D)). As with
Serre duality, we have:
Theorem 25.3.4. For any divisor D on X prime to S, Ω0 (D) is isomorphic to the dual of
I 0 (D) = H 1 (X 0 , L0 (D)).
581
25.3. Differentials on Singular Curves Chapter 25. Curves with Singularities
As in Section 17.5, Dk (K) is isomorphic to the dual of R/K, so for any ω ∈ Dk (K), ω ∈ Ω0 (D)
if and only if ω|R0 (D) = 0. Hence Ω0 (D) ∼
= (R/(k(X) + R0 (D)))∗ . Consider the short exact
sequence of sheaves
0 → L0 (D) → k(X) → Q → 0
where k(X) is the constant sheaf and Q is the sheafification of k(X)/L0 (D). Note that
since X 0 is irreducible, H 0 (X 0 , k(X)) = k(X) and H 1 (X 0 , k(X)) = 0. Taking the long exact
sequence in sheaf cohomology, we get
k(X) → H 0 (X 0 , Q) → I 0 (D) → 0.
H 0 (X 0 , Q) ∼ (k(X)/L0 (D))Q ∼
M
= = R0 /R0 (D).
Q∈X 0
Further, k(X) → R0 /R0 (D) is injective since k(X) is a field, so the second exact sequence
gives us I 0 (D) ∼
= R0 /(k(X) + R0 (D)). One finishes by showing R/(k(X) + R0 (D)) and
R /(k(X) + R (D)) have the same dual, from which it follows that Ω0 (D) ∼
0 0
= I 0 (D)∗ .
Corollary 25.3.6. If Ω0 is the vector space of regular differential forms on X 0 , then dimk Ω0 =
π = g + δ.
`0 (D) − `0 (K 0 − D) = deg(D) + 1 − π.
582
Chapter 26
Constructing Jacobians
Recall that every algebraic curve X over the complex numbers has a Jacobian J(X) = Cg /Λ
where Λ is a full lattice in Cg . In this chapter, we construct Jacobians for more general
algebraic curves, including so-called generalized Jacobians for singular curves.
583
26.1. The Jacobian of a Curve Chapter 26. Constructing Jacobians
Theorem. There exists an abelian variety J(X) of dimension g over k and a rational map
ϕ : X → J(X) such that the group law of J(X) corresponds to a ‘rational group law’ on the
gth symmetric power of X.
Assume for the moment k is algebraically closed, although the results will hold in
general.
Fix a universal domain Ω, which is an algebraically closed field extension Ω/k of infinite
transcendence degree within which all other extensions we consider will be contained.
The field of definition k(P ) for a point P ∈ X is the finite subextension k(P )/k of Ω/k
generated by the coordinates of P .
Lemma 26.1.1. There exists a smallest field extension K/k such that a is generated by
polynomials with coefficients in K.
Definition. Such a field K is called the field of rationality of D, written K = k(D). For
any field L/k, we say D is rational over L if k(D) ⊆ L.
Lemma 26.1.2. If D1 , D2 ∈ Div(X) are rational over L/k, then D1 − D2 is also rational
over L.
Proof. This follows from the fact that k(D1 − D2 ) ⊆ k(D1 ) ∩ k(D2 ).
k(D) = k(P1 , . . . , Pn )s ,
584
26.1. The Jacobian of a Curve Chapter 26. Constructing Jacobians
Lemma 26.1.4. Suppose `(C − D) 6= 0 and D is rational over some L/k. Then for any
generic point P ∈ X which is L-rational, we have `(C − (D + P )) = `(C − D) − 1.
Proof. If this were not the case, Riemann-Roch would imply `(C − (D + P )) = `(C − D)
but then every f ∈ L(C − D) vanishes at P , which is impossible since P is generic.
Lemma 26.1.5. Fix a degree 0 divisor D0 ∈ Div0 (X) and let P1 , . . . , Pg be P independent
generic points of X defined over some L/k for which P D0 is rational. Set D = Pi . Then
there is a unique effective divisor of the form E = Qi , where Q1 , . . . , Qg are independent
generic points, such that E ∼ D0 + D. Moreover,
L(P1 , . . . , Pg )s = L(Q1 , . . . , Qg )s .
585
26.1. The Jacobian of a Curve Chapter 26. Constructing Jacobians
P
positive divisor E = Mi which is unique for D1 , D2 . Set q1 = ϕ(D1 ), q2 = ϕ(D2 ) and
q3 = ϕ(E). We define the additive composition law on V by
V × V −→ V
(q1 , . . . , q2 ) 7−→ q1 + q2 := q3 .
Proposition 26.1.6. Let q0 = ϕ(D0 ) for a divisor D0 ∈ Div(X) and let q1 , q2 , q3 as above.
Then
(2) q1 + q2 = q2 + q1 .
Proof. (1) and (2) are straightforward. For (3), consider the divisor class [D0 +D1 +D2 −2D].
Then associativity follows easy.
Theorem 26.1.7 (Weil, FAC). There is an algebraic group G/k and a birational isomor-
phism V → G such that
Thus for an effective divisor D ∈ Div(X) of degree deg(D) = g, we may choose a model
over k for k(D) which is an algebraic group.
Definition. The algebraic group thus defined is called the Jacobian of X, denoted J =
J(X).
Theorem 26.1.8. For a complete, nonsingular curve X/k of genus g ≥ 1, the Jacobian J =
J(X) is an abelian variety of dimension g equipped with a rational map f : X → J such that
P f is defined, and for independent generic points P1 , . . . , Pg ∈ X
for any field L/k over which
over L, the point Q = f (Pi ) is a generic point of J such that L(Q) = L(P1 , . . . , Pg )s .
Moreover, every generic point of J is of this form and the Pi are uniquely determined up to
permutation.
f
X J(X)
for abelian varieties A. We will show this in Section 26.3 in the case of generalized Jacobians.
586
26.2. Generalized Jacobians Chapter 26. Constructing Jacobians
Lemma 26.2.1. For any modulus m, ∼m is an equivalence relation on Div(X) which forms
a finer partition than ∼.
Let Cm be the group of m-equivalence classes of divisors prime to S and let Cm0 be the
subgroup of degree 0 divisor classes in Cm . Following the construction in Section 25.2, for a
divisor D ∈ Div(X) prime to S, let Lm (D) be the k-vector space Lm (D) = H 0 (Xm , Lm (D))
for the sheaf (
0
OQ , Q∈S
Lm (D)Q =
L(D)Q , Q 6∈ S.
Lemma 26.2.3. Let D ∈ Div(X) be prime to S and suppose L/k is a field containing k(D).
Then
(2) Im (D)∗ = Ωm (D) has a basis of differential forms which are rational over L.
Corollary 26.2.4. Fix D ∈ Div(X) prime to S. Suppose there is a unique positive divisor
D0 ∼m D. Then D0 is rational over k(D).
587
26.2. Generalized Jacobians Chapter 26. Constructing Jacobians
Proof. Since D0 is unique, dim Lm (D) = 1 and dim L(D − m) = 0 so by Lemma 26.2.3,
we can choose g ∈ Lm (D) which is rational over k(D). Write D0 = D + (g) and apply
Lemma 26.1.2.
Fix D ∈ Div(X) prime to S and set
Lemma 26.2.5. Let K be a field over which D is rational and let P ∈ X be a generic point
over K. Then im (D + P ) = im (D) − 1.
Proof. Basically the same as Lemma 26.1.4, as that proof only needed L(C −D) ∼
= I(D).
Lemma 26.2.6. Let D0 ∈ Div0 (X) be rational over K and suppose P1 , . . . , Pπ are inde-
pendent generic
Pπ points over a field K/k. Then there exists a unique effective divisor of the
form E = i=1 Qi , wherePQ1 , . . . , Qπ are independent generic points over K, such that
E ∼m D0 + D, where D = πi=1 Pi . Moreover, K(D) = K(E).
Proof. This is similar to the proof of Lemma 26.1.5, which is the m = 0 case. For general m,
note that since deg(D) = 0, `m (D) ≤ 1 so by singular Riemann-Roch, im (D) ≤ π. Applying
Lemma 26.2.5 multiple times, we see that im (D0 + D) = 0 implies `m (D0 + D) = 1. Thus
there is a unique (up to scaling) function g ∈ Lm (D0 + D). When m 6= 0, we must also
check L(D0 + D − m) = 0 so that g 6∈ L(D0 + D − m). Suppose `(D0 + D − m) ≥ 1. Since
deg(D0 +D−m) = g−1, ordinary Riemann-Roch implies i(D0 +D−m) = `(D0 +D−m) ≥ 1.
On the other hand, i(D0 +D −m) = i(D0 −m)−π by Lemma 26.2.5, so we get i(D0 −m) > π.
However, deg(D0 − m) < 0 implies `(D0 − m) = 0 so by Riemann-Roch, i(D0 − m) = π, a
contradiction. Thus E can be chosen to be the unique effective divisor in the complete linear
system for D0 + D and the above shows that E ∼m D0 + D.
Finally, Lemma 26.2.3 allows us to choose g rational over K(D0 + D) = K(P1 , . . . , Pπ )s .
By Corollary 26.2.4, E is rational over K(P1 , . . . , Pπ )s . On the other hand, the above para-
graph implies `m (D) = 1, so D is also the unique effective divisor that is m-equivalent to
E − D0 . Thus by the same argument, K(D) ⊆ K(E − D0 ) = K(E). This completes the
proof.
Let X (π) be the symmetric product of π copies of X. We now construct a birational
composition on an open subset of X (π) as we did in the nonsingular case. Fix a field K/k
and a point P0 ∈ X which is rational over K.
Lemma 26.2.7. Let P, Q ∈ X (π) be independent generic points over K. Then there exists
a unique divisor R ∈ Div(X) such that R ∼m P + Q − πP0 as divisors and K(P, Q) =
K(R, P ) = K(R, Q).
588
26.2. Generalized Jacobians Chapter 26. Constructing Jacobians
Proof. One need only check that π is defined over k – for this, look at the Gal(Ω/k)-action
– and that π is associative – as in Proposition 26.1.6, this can be done by considering the
divisor P + P 0 + P 00 − 2πP0 .
Thus by Weil’s theorem (26.1.7), there exists a unique (up to birational equivalence)
algebraic group G over k birationally equivalent to X (π) with group law induced from the
one constructed above on X (π) .
Definition. The algebraic group Jm = Jm (X) associated to the described birational group
law on X (π) is called the generalized Jacobian of X for the modulus m.
589
26.3. The Canonical Map Chapter 26. Constructing Jacobians
Lemma 26.3.1. Let K/k be a field, suppose D, D0 ∈ Div(X) are rational over K and set
D00 = D + D0 . Then if M, M 0 , M 00 are independent generic points of X (π) defined over K,
then
θM 00 (D00 ) = θM (D) + θM 0 (D0 ).
Proof. Lemma 26.2.6 allows us to write down three unique points N, N 0 , N 00 ∈ X (π) for
D, D0 , D00 , respectively, satisfying
Then ϕ(N 00 ) − ϕ(M 00 ) = ϕ(N ) − ϕ(M ) + ϕ(N 0 ) − ϕ(M 0 ), and so ϕ(N ) + ϕ(N 0 ) + ϕ(M 00 ) =
ϕ(N 00 ) + ϕ(M ) + ϕ(M 0 ). Since k(N ) = k(M ), k(N 0 ) = k(M 0 ) and k(N 00 ) = k(M 00 ), we see
that N, N 0 , N 00 are independent over k. Therefore N + N 0 + M 00 ∼m N 00 + M + M 0 , which
implies θM 00 (D00 ) = θM (D) + θM 0 (D0 ).
Lemma 26.3.2. For D ∈ Div(X), the definition of θM (D) is independent of the choice of
M ∈ X (π) .
by definition of ϕ.
590
26.3. The Canonical Map Chapter 26. Constructing Jacobians
Proposition 26.3.5. Let DivS (X) be the divisors on X which are prime to S. Then θ :
DivS (X) → Jm is a surjective group homomorphism with kernel
Proof. Lemmas 26.3.1 and 26.3.3 prove that θ is a homomorphism. For surjectivity, note
that Lemma 26.3.3 shows that the image of θ is open and hence generates Jm . Since θ is a
homomorphism, we have im θ = Jm . Lastly, θ(D) = 0 if and only if ϕ(N ) = ϕ(M ) for all
(equivalently, for any) generic points M ∈ X (π) , which is equivalent to N ∼m D−deg(D)P0 +
M . Hence θ(D) = 0 if and only if D ∼m deg(D)P0 .
We next describe the universal property of ϕ and θ.
n+1
X
g −1 (g(P )) = π ◦ σi (Q)
i=1
Pn
for a fixed Q ∈ π −1 (P ). Set HP = i=1 π ◦ σi (Q), which is an effective divisor on X (n) .
Next, define a map
h : Y −→ X n
Q 7−→ (π ◦ σ1 (Q), . . . , π ◦ σn (Q)).
Composing this with X n → X (n) , we get a map h0 : Y → X (n) which is regular. But
explicitly, X n → X (n) takes (P1 , . . . , Pn ) to the point of X (n) that identifies with the divisor
P n
i=1 Pi , so passing to the quotient Y /h = X, we get precisely the map P 7→ HP .
The map θ : DivS (X) → Jm defines a rational map θ : X r S → Jm . We will prove that
θ is in fact regular.
Lemma 26.3.7. For any generic point M of X (π) over K, there exists a rational map
θ0 : X → Jm defined over K(M ) and regular over X r S which is rational over K, and such
that θ0 = θ on X r S over K.
591
26.3. The Canonical Map Chapter 26. Constructing Jacobians
Jm
ϕm
F
X G
f
592
26.3. The Canonical Map Chapter 26. Constructing Jacobians
Proof. Assume 1G = 0. Then for D ∈ Div0 (X) which is prime to S, Theorem 26.3.10(2)
says Cm0 ∼
= Jm , so the natural map
Cm0 −→ G
D 7−→ f (D)
Proof. Suppose P00 ∈ X is another basepoint with associated canonical map ϕ0m : X → Jm .
(By Weil’s theorem, Jm is unique already.) Then by Theorem 26.3.10, there are unique maps
F, F 0 : Jm → Jm making the following diagram commute:
ϕm Jm
X F F0
ϕ0m Jm
Proof. By Theorem 24.2.10, such an f has a modulus m. Now apply Theorem 26.3.10.
593
26.4. Invariant Differential Forms Chapter 26. Constructing Jacobians
g Trg (ϕ)
P1
594
26.4. Invariant Differential Forms Chapter 26. Constructing Jacobians
By the proof of Theorem 24.2.10, Trg (α) = h∗ (ω) where α = ϕ∗ (ω). On the other hand,
since m is a modulus for ϕm , Proposition 24.2.9 implies h is constant. Thus h∗ (ω) = 0 as
required.
Corollary 26.4.3. If deg(m) ≥ 2, then the map ϕm is not regular at any point P ∈ S.
595
26.5. The Structure of Generalized Jacobians Chapter 26. Constructing Jacobians
Proposition 26.5.1. Let m and m0 be moduli on X such that m ≥ m0 and fix a point P0 ∈ X.
Then there exists a unique surjective map F : Jm → Jm0 with connected kernel Hm/m0 such
that ϕm0 = F ◦ ϕm .
Proof. Since m0 is a modulus for ϕm0 , m ≥ m0 implies m is also a modulus for ϕm0 . Thus by
Theorem 26.3.10, there exists a unique map F : Jm → Jm0 such that F ϕm + ϕm0 (P0 ) = ϕm0 .
0
To show F is onto, we construct a birational section of F . Consider the map X π → X π given
0
by (M1 , . . . , Mπ0 ) 7→ (M1 , . . . , Mπ0 , P0 , . . . , P0 ). This descends to a map S : X (π ) → X (π) ,
which in turn determines S 0 : Jm0 → Jm via the birational isomorphisms X (π) → Jm and
0
X (π ) → Jm0 . By construction, F ◦ S 0 = 1, so F is surjective.
As a result, anytime m ≥ m0 , the generalized Jacobian for m0 can be decomposed as
Jm0 ∼
= Jm × Hm/m0 . For the special case m0 = 0, J0 = J is the ordinary Jacobian and we write
Lm = Hm/0 . A point d ∈ Jm corresponds to a linear equivalence class of divisors D prime to
S. When D = (g) for some g ∈ k(X), vP (g) = 0 for all P ∈ S, so the image of d in J via the
map F is 0. On the other hand, the class of D determines the origin 0 ∈ Jm when D = (h)
for some h ≡ 1 mod m, so g = λh for some λ ∈ k × .
For a fixed point P ∈ X, define the groups
(n)
UP = {f ∈ k(X)× | vP (f ) = 0} and UP = {f ∈ UP | vP (1 − f ) ≥ n}.
(n)
For each P ∈ S and g ∈ k(X), we get a function gP ∈ UP /UP for some n ≥ 0, and
Q (n )
conversely, every collection of cosets (gP )P ∈S ∈ P ∈S UP /UP P is the image of such a g ∈
(n )
k(X)× . Set Rm = P ∈S UP /UP P , let ∆ be the image of k × in Rm via λ 7→ (λ, . . . , λ) and
Q
define
Hm = Rm /∆.
We have proven the following.
We next describe an algebraic group structure on Lm . To begin, let k((t)) be the field
of formal (Laurent) power series over k, which is a local field when equipped with the usual
power series valuation v. Let
Then any f ∈ U (n) can be written f = 1 + an tn + an+1 tn+1 + . . . so that the elements of the
quotient U/U (n) are truncated power series: f = a0 + a1 t + . . . + an−1 tn−1 , a0 6= 0. Therefore
U/U (n) has the structure of a variety, and pointwise multiplication, inversion and change of
uniformizer t are all regular functions on U/U (n) . Thus:
596
26.5. The Structure of Generalized Jacobians Chapter 26. Constructing Jacobians
Proposition 26.5.3. U/U (n) is an algebraic group under pointwise multiplication and in-
version.
Now let V(n) = U (1) /U (n) ⊆ U/U (n) be the subgroup of truncated power series of the form
1 + a1 t + . . . + an−1 tn−1 . This too has the structure of an algebraic group.
Lemma 26.5.4. There is an isomorphism of algebraic groups U/U (n) ∼
= Gm × V(n) .
Proof. Every function f ∈ U/U (n) is of the form f = a0 f 0 for some f 0 ∈ U (1) /U (n) and
a0 ∈ k × , so we have the abstract isomorphism U/U (n) ∼
= Gm × V(n) . Since multiplication by
−1
a0 and a0 are regular, the isomorphism preserves the algebraic group structure.
Lemma 26.5.5. For each 1 ≤ i ≤ n − 1, let gi ∈ k((t)) be a formal power series with
v(gi ) = i. Then every element g ∈ V(n) can be written uniquely as
for a1 , . . . , an−1 ∈ k.
b1
Proof. Write g = 1 + b1 t + . . . + bn−1 tn−1 and g1 = c1 t + c2 t2 + . . .. Then setting a1 = c1
and
h1 = 1+ag1 g1 , we see that v(1 − h1 ) ≥ 2. The subsequent ai are defined similarly.
Rm ∼ UP /UP P ∼
Y (n )
Y
= = G|S|
m × Gna P −1 .
P ∈S P ∈S
Then since ∆ ∼
= Gm , we may write
Hm = Rm /∆ ∼
Y
= G|S|−1
m × Gna P −1 .
P ∈S
597
26.5. The Structure of Generalized Jacobians Chapter 26. Constructing Jacobians
Let g, h ∈ k(X) and for every z ∈ P1k , let Dz be the principal divisor defined by Dz = (g +zh)
for z 6= ∞ and D∞ = (h). Let
T = {z ∈ P1k | Dz ∩ S 6= ∅}.
Proof. Notice that ϕ(Dz ) ∈ Lm since Dz is the divisor of a rational function. To prove
regularity, we may assume Dz = (g − z). Let ψ = Trg (ϕ), which is regular on P1k r T by
Proposition 24.2.8. Then ϕ(Dz ) = ψ(z) − ψ(∞), so z 7→ ϕ(Dz ) is regular off of T .
Proof. Write Hm ∼
|S|−1 Q
= Gm × P ∈S Gna P −1 . Then it is enough to check regularity on each com-
ponent of the product. For each P ∈ S, let Gm,P be the multiplicative group corresponding
to P in the first product. Let v be a function with v ≡ 1 mod m at P and v ≡ 0 mod m
on S r {P }. Then for λ ∈ Gm,P , uλ = (λ − 1)v + 1 satisfies uλ ≡ λ mod m at P and uλ ≡ 1
mod m on S r {P }. Moreover, θ(λ) = ϕ((uλ )) for (uλ ) the principal divisor associated to uλ ,
so by Lemma 26.5.9, θ is regular on Gm,P . A similar proof works for each factor Gna P −1 .
Remark. The proof in the case char k = p > 0 requires a bit more care, as a bijective,
birational isomorphism is not necessarily a global isomorphism (it may correspond to an
inseparable extension). However, an argument using the trace map and residue formula
show the result; see Serre for the details.
Corollary 26.5.12. For any curve X with modulus m, the generalized Jacobian Jm is a
group extension of the ordinary Jacobian J of X, i.e. there is a short exact sequence of
groups
1 → Lm → Jm → J → 1.
Example 26.5.13. For X = P1k and the modulus m = 0 + ∞, J(P1k ) = 0 so the generalized
Jacobian is Jm = P1k r {0, ∞} ∼
= Gm . More generally, if m = (P ) + (Q) for distinct points
1 ∼
P, Q ∈ Pk , then Jm = Gm . Let us explicitly describe the multiplication on Jm . First, we
have
Div0m (X) = {D = (g) | g : P1 → k, ordP (g) = ordQ (g) = 0}.
598
26.5. The Structure of Generalized Jacobians Chapter 26. Constructing Jacobians
Thus the map ϕm : Cm0 → Gm = P1 r {P, Q} is given by [(g)] 7→ [g(P ), g(Q)]. It is clear that
this ϕm is both well-defined and injective. To see that it is surjective, take [α, β] ∈ P1 r{0, ∞}.
Then the function α
β
(t − Q) t−P
g(t) = −
P −Q Q−P
defines an m-equivalence class [(g)] ∈ Cm0 and ϕ([(g)]) = [α, β] by construction.
For D = nP P ∈ Div0 (E), there is some function f ∈ k(E)× such that D = D − (O) + (f ).
P
Consider the modulus m = (M ) + (N ) for distinct points M, N ∈ E r {O}. If D ∈ Div0m (E),
then either
D = (M ) or D = (N ); or
D 6= (M ), (N ).
599
26.5. The Structure of Generalized Jacobians Chapter 26. Constructing Jacobians
Repeating this for two divisors D1 , D2 ∈ Div0m (E), we can write Di = (Si +Ri )−(Ri )+(fi )
for i = 1, 2. Then
ϕm : Cm0 (E) −→ Jm = Gm × E
f (M )
[D] 7−→ ,S if D = (S + R) − (R) + (f ).
f (N )
We need the following result to describe the group structure of Jm in this case.
A × G −→ A
(a, σ) 7−→ aσ := x ⊕ a x
1 → Gm → Jm → E → 0
600
26.5. The Structure of Generalized Jacobians Chapter 26. Constructing Jacobians
and Jm = Gm × E as sets, with (b1 , P ) ⊕ (b2 , Q) = (b1 b2 · cm (P, Q), P + Q) for all b1 , b2 ∈
Gm , P, Q ∈ E and cm : E × E → Gm a 2-cocycle. To determine this cocycle, let D1 =
(P1 ) − (O) + (f1 ) and D2 = (P2 ) − (O) + (f2 ) be two divisors on E prime to the support
of m. Then under the canonical map ϕm : Cm0 (E) → Jm , we have ϕm ([Di ]) = (bi , Pi )
for i = 1, 2 and (b1 , P1 ) ⊕ (b2 , P2 ) = (b3 , P3 ) for some b3 ∈ Gm , P3 ∈ E. Notice that
if D3 = (P1 ) + (P2 ) − 2(O) + (f1 f2 ), then it will be enough to determine f3 such that
D3 = (P3 ) − (O) + (f3 ). By the Abel-Jacobi isomorphism, we have
f3 (M ) `P ,P (M ) `P1 ,P2 (M )
b3 = = b 1 b2 1 2 and cm (P1 , P2 ) = .
f3 (N ) `P1 ,P2 (N ) `P1 ,P2 (N )
We claim (`P1 ,P2 ) = (P1 ) + (P2 ) − (P1 + P2 ) − (O) will work. Let LP1 ,P2 be the linear form
defining the line in E ⊆ P2 containing P1 , P2 and −(P1 + P2 ).
−(P1 + P2 )
P2
P1
LP1 ,P2
P1 + P2
LP1 +P2 ,O
Then dividing out by z gives a function with the following zeroes and poles:
LP1 ,P2 (z)
= (P1 ) + (P2 ) + (−(P1 + P2 )) − 3(O).
z
On the other hand, the linear form LP1 +P2 ,O defining the line containing P1 + P2 , −(P1 + P2 )
and O yields
LP1 +P2 ,O (z)
= (P1 + P2 ) + (−(P1 + P2 )) − 2(O).
z
LP1 ,P2 `P1 ,P2 (M )
So taking `P1 ,P2 = LP1 +P2 ,O
, we get cm (P1 , P2 ) = `P1 ,P2 (N )
as the desired 2-cocycle.
601
Chapter 27
The goal of class field theory is to classify all abelian extensions L/K in terms of the
arithmetic of K itself. To this end, let I(K) be the set of fractional ideals of K and set
GL/K = Gal(L/K) for any abelian Galois extension.
Definition. The Artin map of an unramified abelian extension L/K is the map on frac-
tional ideals given by
where the product runs over all places v of K, Frobv is the Frobenius element at v and ordv (a)
is the valuation of a at v.
Theorem 27.0.1. For any principal ideal (α) ∈ I(K) and any unramified abelian ex-
tension L/K, θL/K ((α)) = 1. Therefore θL/K descends to a map on the class group:
CK = I(K)/P (K) → GL/K .
Class field theory for number fields is summarized in the following theorem.
Theorem 27.0.2. Let L/K be a finite abelian extension with Galois group GL/K . Then
(1) (Artin Reciprocity) The Artin map θL/K : CK → GL/K is surjective with kernel
NL/K (CL ), where NL/K is the norm map.
(2) For any normal subgroup N ≤ CK , there exists a finite abelian extension L/K for
which N = ker θL/K . In this case, CK /N ∼
= GL/K .
(3) The Artin map is natural, i.e. for all finite unramified abelian extensions M ⊇ L ⊇ K,
there is a commutative diagram
602
Chapter 27. Class Field Theory for Curves
θM/L
CL GM/L
NL/K
CK GM/K
θM/K
Theorem 27.0.3. Let H any connected commutative algebraic group with principal homo-
geneous space H 0 admitting a regular map f : X r S → H 0 . Then
ϕm
X rS Jm0
fm0
f
H0
where Jm0 is a principal homogeneous space for the generalized Jacobian Jm and ϕm :
X r S → Jm0 is the restriction of the canonical map.
(2) There exists a map fm : Jm → H with respect to which fm0 : Jm0 → H 0 is an equivariant
map of principal homogeneous spaces.
(3) For every modulus m of f , the map fm0 is unique and thus there is a unique modulus
f dividing all such m.
(4) Any geometrically connected abelian cover of H 0 pulls back along f to give a geomet-
rically connected cover of X which is unramified outside S.
603
27.1. Coverings and Isogenies Chapter 27. Class Field Theory for Curves
X 0 r S0 G1
X Jm1
ϕm
commutes, where S 0 = π −1 (S) and Jm1 is the principal homogeneous space for Jm given by
Theorem 27.0.3. Moreover, there exists a unique modulus f dividing all moduli m for π, and
for this f, the cover G1 → Jm1 is geometrically connected if and only if X 0 is geometrically
connected.
Suppose π : X 0 → X is an abelian cover such that X 0 (k) 6= ∅. Then for any y0 ∈ X 0 (k),
with image x0 = π(y0 ) ∈ X(k), all elements of π −1 (x0 ) are k-points of X 0 since π is abelian.
Let
ϕ : X −→ J = J(X)
x 7−→ x − x0
be the ordinary Abel map. Then by Theorem 27.1.1, there is a finite, connected abelian
cover G → J such that the diagram
X0 G
X ϕ J
604
27.1. Coverings and Isogenies Chapter 27. Class Field Theory for Curves
θ : Div0 (X) −→ J
X X
D= nξ ξ 7−→ nξ ξ.
605
27.2. Class Field Theory Chapter 27. Class Field Theory for Curves
f ∗W X
f
π
W V
In fact, we have the following amazing result, which can be found in Serre.
Theorem 27.2.1. Every abelian cover W → V is the pullback of an isogeny f : G0 → G of
smooth, abelian varieties with kernel ker f ∼
= Gal(W/V ).
Given a finite group N , let A(N ) be the group algebra of N (over some universal domain
Ω, e.g.) and let G(N ) be the subgroup of invertible elements:
Then N embeds as a subgroup of G(N ) via s 7→ (at ) where at = δst for all t ∈ N .
Proposition 27.2.2. If π : W → V is a Galois cover with Galois group N , then there exist
a rational map f : V → G(N )/N such that W ∼
= f ∗ G(N ). Moreover, if W and π are defined
over k, then both f and the isomorphism W ∼
= f ∗ G(N ) are also defined over k.
Proof. Define g : W → A(N ) by x 7→ (ϕs (x)), where ϕ ∈ k̄(W ) is any fixed rational function.
Then since det(ϕst ) 6= 0, g has image in G(N ). Consider the diagram
g
W G(N )
π
f
V G(N )/N
Then by the normal basis theorem from Galois theory, there exists an element ψ ∈ k(W )
such that {ϕs }s∈N is a basis of the field extension k(W )/k(V ). It follows that f is equal to
g after taking quotients, so the diagram commutes.
606
27.2. Class Field Theory Chapter 27. Class Field Theory for Curves
Proof. To prove this, we must find a normal basis {ψ s }s∈N where ψ is regular on the set
π −1 (P ). Let OP be the local ring at P in K = k(V ) with maximal ideal mP and as in
Chapter 25, let OP0 be the integral closure of OP in L = k(W ), so that
\
OP0 = OQ .
Q∈π −1 (P )
Set k 0 (P ) = OP0 /mP OP0 and k(P ) = OP /mP . Since P is unramified, k 0 (P ) is an étale algebra
of degree n = [L : K] over k(P ). Thus by the normal basis theorem, we can choose λ ∈ k 0 (P )
such that {λs }s∈N is a normal basis of k 0 (P )/k(P ). Now λ may be lifted to OP0 for all P ,
giving a rational function ψ on W which is regular on π −1 (P ).
This proves Theorem 27.2.1.
Example 27.2.4. Let N = Z/nZ where (n, p) = 1. Then A(N ) ∼ = k[t]/(tn − 1) and if k
contains a primitive nth root of unity, then G(N ) ∼
= G×n
m . Let θn : Gm → Gm be the isogeny
n
x 7→ x (called the nth Kummer cover). Then G(N ) → G(N )/N is the pullback of this
isogeny:
G(N ) Gm
θn
G(N )/N Gm
Example 27.2.5. Let N = Z/pZ. Then A(N ) ∼ = k[t]/(tp − 1), G(N ) ∼= Ga and G(N ) →
G(N )/N is the pullback of the Artin-Schreier cover ℘ : Ga → Ga , x 7→ xp − x:
G(N ) Ga
G(N )/N Ga
Now let X be a complete, irreducible, nonsingular curve over k with field of rational
functions K = k(X).
607
27.2. Class Field Theory Chapter 27. Class Field Theory for Curves
Y G
π
f
X H
608
Part VI
609
Chapter 28
Introduction
The following notes are taken from a reading course on étale fundamental groups led by Dr.
Lloyd West at the University of Virginia in Spring 2017. The contents were presented by stu-
dents throughout the course and mostly follow Szamuely’s Galois Groups and Fundamental
Groups. Main topics include:
Applications.
610
28.1. Topology Review Chapter 28. Introduction
f
Y1 Y2
p1 p2
This defines a category CovX of covers over X. In this category, we will abbreviate
HomCovX (Y, Z) by HomX (Y, Z).
Example 28.1.1. The unit interval [0, 1] ⊆ R has no nontrivial covers. However, S 1 ⊆ C
does: for each n ∈ Z, the map
pn : S 1 −→ S 1 , z 7−→ z n
π : R −→ S 1 , t 7−→ e2πit
R S1
π pn
S1
611
28.1. Topology Review Chapter 28. Introduction
f
X
e Y
π p
It is equivalent to say that a universal cover is any simply connected cover of X, and
one shows easily that universal covers are unique up to equivalence of covers. An important
result is that a universal cover exists, under certain mild conditions on X.
In topology, the topological fundamental group is defined using homotopy:
top homotopy classes of loops
π1 (X, x) := .
in X based at x
The universal cover has important connections to this fundamental group. In particular,
consider an automorphism α ∈ AutX (X)e = HomX (X, e Fix x ∈ X and a lift x̃ ∈ X
e X). e of x.
e is simply connected, any path x̃ → αx̃ is unique up to
Then πα(x̃) = x. Moreover, since X
homotopy. This determines a map AutX (X) e → π1top (X, x).
Unfortunately, such a space X e does not exist in the algebraic world. Thus we describe a
slightly different interpretation of the fundamental group.
Definition. The fibre functor over x ∈ X is the assignment
where the Hom set consists of morphisms based at x̃. This fibre functor is constructible in
algebraic categories, though it fails to be representable.
Going further, there is a natural left action of AutX (X)
e on X;
e however, it will be more
convenient to view this as a right action of AutX (X)e op on X.
e This induces a left action of
top
AutX (X)
e = π1 (X, x) on HomX (X, e Y ):
612
28.1. Topology Review Chapter 28. Introduction
This action is called the monodromy action. Often, one views this as an action of π1top (X, x)
on the fibre Fibx (Y ) given by lifting paths. In any case, we get a map
where Aut(Fibx ) is the automorphism group of the fibre functor in the following sense. For
any functor F : C → D, an automorphism of F is a natural transformation of F that
has a two-sided inverse. The set Aut(F ) of all automorphisms of F is then a group under
composition. Moreover, Aut(F ) has a natural action on F (C) for any object C ∈ C.
Theorem 28.1.4. Let X be a connected, locally simply connected space and fix x ∈ X. Then
the fibre functor Fibx defines an equivalence of categories
∼
CovX −
→ {left π1 (X, x)-sets}
with connected covers corresponding to transitive π1 (X, x)-sets and Galois covers to coset
spaces of X
ex by normal subgroups.
where N is some finite index kernel. This shows that the monodromy action factors through
a finite quotient. As a result, this action can be defined on the level of a profinite group,
namely the profinite completion of π1top (X, x):
π1top
\ (X, x) := lim π1top (X, x)/N,
←−
where the inverse limit is over all finite index subgroups N ≤ π1top (X, x).
613
28.2. Finite Étale Algebras Chapter 28. Introduction
Proof. By Galois theory, # Homk (L, ks ) = [L : k], so this is finite when the extension is
assumed to be finite. The Gk -action is defined by σ·f = σ◦f for σ ∈ Gk and f ∈ Homk (L, ks );
it is routine to verify that this is indeed a group action. Now to show the action is continuous,
since Homk (L, ks ) is discrete, this is equivalent to showing the stabilizer StabGk (f ) is open
for each f ∈ Homk (L, ks ). Notice that
and this is open by Galois theory / the topology of the profinite group Gk . Finally, since
L/k is separable, we may pick a minimal polynomial h(t) for a primitive element of L/k.
Then Gk permutes the roots of h(t) transitively, so it follows that Gk acts transitively on
Homk (L, ks ).
Corollary 28.2.2. There exists an open subgroup H ≤ Gk such that Homk (L, ks ) ∼ = Gk /H
as Gk -sets. Further, if L/k is Galois, H may be chosen to be an open normal subgroup.
Proof. We may pick H = StabGk (h), the stabilizer of the minimal polynomial of a primitive
element of L/k. The Galois case follows from the fundamental theorem of Galois theory.
Proof. For any finite, continuous, transitive Gk -set S, define a finite separable extension
LS /k by taking the subfield of ks /k fixed by the stabilizer StabGk (s) of any point s ∈ S.
One now checks that this is an inverse functor to the Homk (−, ks ) functor.
To study all finite continuous Gk -sets, we replace separable field extensions with a slightly
more general object. Recall (Section 15.5) that a k-algebra A is a finite étale algebra if
A∼
= L1 × · · · × Lr for finite separable extensions Li /k.
Corollary 28.2.4 (Grothendieck). The assignment A/k 7→ Homk (A, ks ) is an anti-equivalence
of categories
∼
{finite étale k-algebras} −
→ {finite continuous Gk -sets}
which reduces to the above case when A = L is a finite separable extension of k.
614
28.2. Finite Étale Algebras Chapter 28. Introduction
Topologically, we may view k as a point space covering the ‘smaller’ point space ks , and
any intermediate extension L/k as an intermediate cover. In this setting, Gk plays the role
of the deck transformations of the ‘universal cover’ k → ks and Homk (L, ks ) plays the role
of the fibre of a cover. Also, under this analogy, a finite separable extension L represents
a connected cover while a finite étale algebra A may be viewed as a disconnected cover.
Finally, the choice of a algebraic (and separable) closure of k is analagous to the choice of a
basepoint of a topological space, which also determines a universal cover. We will see that
this subtlety conceals a wealth of information about the algebraic fundamental group.
615
28.3. Étale Morphisms Chapter 28. Introduction
(1) π1ét (X, x̄) is a profinite group which acts continuously on each fibre Fibx̄ (Y ) for Y → X
a cover.
(2) Fibx̄ : FétX → {continuous, finite π1ét (X, x̄)-sets} is an equivalence of categories.
Further, we will see that π1ét (X, x̄) = lim AutX (Fibx̄ (Y )), where the inverse limit may be
←−
taken over all Galois covers p : Y → X.
616
28.3. Étale Morphisms Chapter 28. Introduction
Example 28.3.2. Let k be a field and consider the scheme X = Spec k which is a point. A
geometric point x̄ : Spec Ω → X is a choice of algebraic closure Ω ⊇ k; such a choice also
defines a separable closure ksep ⊇ k. In this case, the correspondence of Theorem 28.3.1 is:
FétX ←→ {Spec A | A is a finite étale k-algebra}.
Here, the functor Fibx̄ is representable by Homk (A, Ω) ∼ = Homk (A, ksep ). Moreover, we
have a Galois action of Gk := Gal(ksep /k) on Homk (A, ksep ) given by σ · f = σ ◦ f for any
f : A → ksep and σ ∈ Gk . In particular, π1 (X, x̄) = Gk and this action is the monodromy
action as described in Section 28.1.
Theorem 28.3.3. For a field k, Homk (−, ksep ) induces an equivalence of categories
∼
{finite étale k-algebras} −
→ {finite continuous Gk -sets}.
Example 28.3.4. Let (A, m, k) be a complete DVR and set X = Spec A. There is an
equivalence between étale covers
FétA ←→ Fétk .
(One direction is obvious; the other is an application of Hensel’s Lemma.) Therefore
π1ét (Spec A, x̄) ∼
= π1ét (Spec k, x̄) for any geometric point x̄ of A.
Theorem 28.3.5. Suppose X is connected and normal, K = k(X) is the function field of X
and L/K is a finite, separable field extension. Let Y be the normalization of X in L. Then
p : Y → X is a ramified cover and for some set S ⊂ Y , p : Y r S → X is étale. Further,
every étale cover of X arises in this way.
Corollary 28.3.6. If X is a connected, normal scheme with function field K and K ur is
the compositum of all extensions of K in which normalization of X is unramified, then
π1ét (X) ∼
= Gal(K ur /K).
Example 28.3.7. A famous result in number theory says that Q has no nontrivial unramified
extensions. Therefore π1ét (Spec Z) = 1. However, ‘removing some points’
1 from Spec Z, e.g.
ét
localizing, yields nontrivial fundamental groups, such as π1 Spec Z n .
Example 28.3.8. We will show that π1ét (AC ) = 1 (compare to the topological case!), but
for char k > 0, we may have π1ét (Ak ) 6= 1.
There is a classic and important connection between curves and Riemann surfaces which
we will explored further in Chapter 30. The arithmetic version of this correspondence is
borne out by the following theorem.
Theorem 28.3.9. Let X be a normal curve of genus g over k. If char k = p > 0, then
g
* +
π ét (X)(p ) ∼
0
Y
1 = x1 , y1 , . . . , xg , yg : [xi , yi ] = 1 [p]
i=1
0
where (·)(p ) denotes the prime-to-p part of a profinite group. Further, if k = C, then
π ét (X) ∼
1 1
top
= π\ (X).
617
Chapter 29
Let A be a finitely generated k-algebra which is an integral domain with field of fractions K,
and suppose tr degk K = 1. Consider the space X = Spec A = {p ⊂ A | p is a prime ideal}.
Taking closed sets to be of the form V (I) = {p ∈ Spec A | I ⊇ p} for ideals I ⊂ A defines
a Zariski topology on X. The zero ideal 0 defines a generic point of X: for any nonempty
open subset U ⊆ X, 0 ∈ U . It follows that A0 = K. On the other hand, any nonzero prime
p is maximal (since dim A = tr degk K = 1), so V (p) = {p}. Thus prime ideals correspond
to closed points in X. In particular, all proper closed subsets of X are finite.
Definition. For P ∈ X, let OX,P := AP be the local ring at P . For any open U ⊆ X, set
\
OX (U ) := OX,P .
P ∈U
Spec A →−7 A
618
Chapter 29. Fundamental Groups of Algebraic Curves
Let X = Spec A be an integral affine curve. Suppose L/k is a field extension such that
A ⊗k L is still a domain. Then XL := Spec(A ⊗k L) is an integral affine curve over L, called
the base change of X to L. The natural map A → A ⊗k L, a 7→ a ⊗ 1 induces a morphism
of ringed spaces XL → X.
Definition. If A⊗k k̄ is an integral domain, we say X = Spec A is geometrically integral.
In this case, XL is integral for all intermediate extensions k̄ ⊇ L ⊇ k. For any of these
L, there is a functor
geometrically integral integral affine
FL : −→ .
affine curves over k curves over k
Example 29.0.3. Let X = Spec(R[x, y]/(x2 + y 2 + 1)). Then closed points of X are
in correspondence with maximal ideals of the ring R[x, y]/(x2 + y 2 + 1). These in turn
correspond with Galois orbits of XC . Explicitly, each closed point (x2 , y 2 + 1) of X is covered
by exactly two points (x, y − i) and (x, y + i) via XC → X.
Recall that a curve X is normal at P ∈ X if O(X)P is integrally closed and X is normal
if it is normal at every P ∈ X.
Lemma 29.0.4. Let ϕ : Y → X be a finite morphism. Then
(1) O(Y ) is integral over O(X).
(2) ϕ is surjective.
(3) ϕ# is injective.
Proof. Given an embedding i : k(X) ,→ L, let B be the integral closure of O(X) in L. Then
by commutative algebra, B is a finitely generated k-algebra which is integrally closed, and
since L/k(X) is a finite extension, tr degk L = 1. Hence by Proposition 29.0.2, Y = Spec B
is a normal affine curve with a morphism ϕ = i∗ : Y → X. One now checks that these
assignments are natural inverses.
Definition. For an integral affine curve X and a field extension L/k(X), the curve Y
constructed above is called the normalization of X in L.
Remark. If X is geometrically integral, then XL is equal to the normalization of X in L
for any L ⊇ k(X).
619
29.1. Proper Normal Curves Chapter 29. Fundamental Groups of Algebraic Curves
Lemma 29.1.1. The set {OX,P | P 6= 0} is equal to the set of DVRs of K containing O(X).
Proof. On one hand, (⊆) is obvious. For the other containment, let R be such a DVR and
call its maximal ideal m. Then p := m ∩ O(X) is a nonzero maximal ideal in O(X). This
means O(X)p ⊆ R but each of these is a DVR with the same function field, so we must have
OX,P = R.
Example 29.1.2. Let X = A1k so that O(A1k ) = k[t], the polynomial ring in a single
indeterminate. If Y is also A1k , we may take O(Y ) = k[t−1 ]. Let R be a DVR with Frac(R) =
k(X) ∼= k(t) ∼
= k(t−1 ). Then either R ⊃ O(X) or R ⊃ O(Y ). In fact, the only DVR not
containing O(X) is R = k[t−1 ](t−1 ) , while for O(Y ) the exception is R0 = k[t](t) . We want
some way of “gluing” these two DVRs together so that all DVRs of the function field over k
are considered.
Now assume X is a normal affine curve. By Noether normalization, there exists a function
−
f ∈ O(X) such that O(X) is a finitely generated
h i k[f ]-module. Define X to be the normal
curve associated to the integral closure of k f1 in K. Then every DVR R with fraction field
K is a local ring for X + = X or for X − , corresponding to whether f ∈ R or f1 ∈ R.
More abstractly, for a finitely generated field extension K/k of transcendence degree 1,
let XK be the set of DVRs of K/k. Define a topology on XK by declaring the complements
T
of finite subsets to be open. Define a sheaf of rings on XK by U 7→ OXK (U ) := R∈U R.
Theorem 29.1.3. (XK , OXK ) is a ringed space which is “locally affine” in the sense that
+ − + −
there is a decomposition XK = XK ∪ XK where XK and XK are affine curves.
Definition. The ringed space (XK , OXK ) is called an integral proper normal curve.
We will now drop the subscript on XK denoting the field unless it becomes convenient
to have it. Take a morphism of proper normal curves ϕ : Y → X with affine covers
X = X + ∪ X − and Y = Y + ∪ Y − . Then there are induced morphisms of affine curves
ϕ+ : Y + → X + and ϕ− : Y − → X − .
620
29.1. Proper Normal Curves Chapter 29. Fundamental Groups of Algebraic Curves
Lemma 29.1.5. Let X be a proper normal curve. Then a subset U ⊆ X is open if and only
if U ∼
= Spec A for some finitely generated k-algebra A.
Proof. (Sketch) Given a normal affine curve X, set X + = X and embed X into X 0 =
X + ∪ X − as shown above. This yields a proper normal curve, and to show the bijection is an
equivalence of categories, one need only check that a morphism of affine curves ϕ : Y → X
extends uniquely to ϕe : Y 0 → X 0.
Notice that for any affine set U ⊆ X and finite morphism ϕ : Y → X, the restriction
ϕ|ϕ−1 (U ) : ϕ−1 (U ) → U is a finite morphism of affine curves. Thus Lemma 29.0.4 yields:
Conversely, we have:
Proof. Let U ⊆ X be affine. Then ϕ−1 (U ) = {R | R is a DVR of L containing O(U )}. For
any R ∈ ϕ−1 (U ), the integral closure B of O(U ) is contained in R since R is itself integrally
closed. But by Theorem 29.0.5, this B is precisely the integral closure of the normalization
V of U in L. Hence V = ϕ−1 (U ) so (1) is satisfied. It also follows from the identification of
V as the normalization of U in L that ϕ∗ O(U ) = O(ϕ−1 (U )) = O(V ) is finitely generated
over O(U ).
Corollary 29.1.9. For any proper normal curve X, the assignment Y 7→ k(Y ) induces an
anti-equivalence of categories
proper normal curves with finite ∼ finitely generated field extensions k(X) ,→ L
−
→ .
morphisms Y → X with tr degk L = 1
621
29.2. Finite Branched Covers Chapter 29. Fundamental Groups of Algebraic Curves
for distinct closed points Qi ∈ Y and integers ei . By the Chinese remainder theorem,
r
O(Y )/P O(Y ) ∼
Y
= (O(Y )/Qei i ).
i=1
622
29.2. Finite Branched Covers Chapter 29. Fundamental Groups of Algebraic Curves
The next two results were proven in a more general context in Section 15.6; however,
concrete proofs are given here to illustrate some of the techniques we will need in future
discussions of étale and branched covers.
ψ ϕ
Lemma 29.2.4. If Z − →Y −
→ X are finite separable morphisms of affine curves and P ∈ X
is a closed point, then
(2) If X, Y and Z are normal affine curves, then the converse holds.
Proof. Let X = Spec A, Y = Spec B and Z = Spec Z for k-algebras A, B and C. Then we
ϕ∗ ψ∗
have ring extensions A −→ B −→ C as κ(P )-algebras. By the above comments,
C/P C ∼
= C ⊗A κ(P )
∼
= C ⊗B (B ⊗A κ(P ))
!
∼
Y
= C ⊗B κ(Q)
Q→P
∼
Y
= κ(R).
R→P
Y → Xr−1 → · · · → X2 → X1 → X
623
29.2. Finite Branched Covers Chapter 29. Fundamental Groups of Algebraic Curves
where F 1 , . . . , F s are the irreducible factors of F in κ(P )[t]. The work above shows that each
κ(P )[t]/(F i ) is separable, so B/P B is a finite étale κ(P )-algebra. The result follows.
In light of Proposition 29.2.5, the following definitions make sense:
If, in addition to ϕ being a finite Galois cover, X and Y are normal curves, then
Gal(k(Y )/k(X)) acts transitively on the fibres ϕ−1 (P ) of ϕ, just as in the case of a Ga-
lois topological cover.
Proof. Recall the definitions of the decomposition and inertia groups for a point Qi ∈ ϕ−1 (P ):
Since k is perfect, we know from algebraic number theory that ei = [κ(Qi ) : κ(Qi )IQi ], where
κ(Qi )IQi is the subfield of κ(Qi ) fixed by the inertia group, but this index is precisely the order
of IQi . Hence ei = 1 if and only IQi = {1}, so the result follows by Proposition 29.2.1.
For a normal affine curve X and a tower of fields (finite over k(X))
k(X) ⊂ K1 ⊂ K2 ⊂ · · ·
set L = ∞
S
j=1 Kj and let Aj be the integral closure of O(X) in Kj for each j ≥ 1. From this
we get a sequence of morphisms of curves
· · · → X2 → X1 → X.
Fix P ∈ X. Choose maximal ideals {Pj }j≥1 with Pj ∈ Aj such that Pj ∩ O(X) = P and
Pj+1 ∩ Aj = Pj for all j ≥ 1. If Ij is the inertia subgroup of Pj in Gj := Gal(Ki /k(X)), then
624
29.2. Finite Branched Covers Chapter 29. Fundamental Groups of Algebraic Curves
the Ij form an inverse system with the natural maps coming from each quotient of Galois
groups. Define the inertia subgroup of P to be the inverse limit
IP := lim Ij .
←−
Corollary 29.2.7. Let ϕ : Y → X be a Galois branched cover of normal affine curves over
a perfect field and assume k(Y ) ⊆ L where L is as above. Then ϕ is étale at P ∈ X if and
only if the image of IP in Gal(k(Y )/k(X)) = G/ Gal(L/k(Y )) is trivial.
Note that by Lemma 29.2.1, this definition does not depend on the neighborhood U of
P chosen. The last few results for affine curves generalize to proper normal curves as an
immediate consequence of the definition. Again, these also follow from results in Section 15.6.
(1) There exists a nonempty open set U ⊆ X such that ϕ is étale over U .
(2) If ϕ is a Galois branched cover and k is perfect, then ϕ is étale at P ∈ X if and only
if the inertia group is trivial for each point Qi over P .
Example 29.2.9. For a proper normal curve X over C, X(C) has the structure of a compact
Riemann surface such that the cover X(C) → P1 (C) corresponding to X → P1 is proper and
holomorphic.
Proposition 29.2.10. For a proper normal curve X over C, the following categories are
anti-equivalent:
compact connected Riemann
proper normal curves Y finite extensions
←→ ←→ surfaces Y with proper .
with finite morphism Y → X k(X) ,→ L
holomorphic maps Y → X(C)
625
29.2. Finite Branched Covers Chapter 29. Fundamental Groups of Algebraic Curves
ρ2 : P1R −→ P1R , t 7→ t2
corresponding to the ring extension R[t2 ] ,→ R[t]. The maximal ideals of R[t] are of the form
(1) (t − a) for a ∈ R
Example 29.2.12. Let E be an elliptic curve over an algebraically closed field k of char-
acteristic char k = p. Take m ∈ Z such that p - m and consider the multiplication-by-m
isogeny:
[m] : E −→ E, P 7→ mP.
We know that [m] is separable when p - m, and moreover that E[m] = (Z/mZ)2 and thus
#E[m] = m2 . For each P ∈ E, we have
m 2 m 2
O(E)/[m]P ∼ κ(P ) ∼
Y Y
= = k
i=1 i=1
and hence [m] is étale. In general, any degree-m isogeny ϕ : E1 → E2 of elliptic curves over
an algebraically closed field of characteristic not dividing m is separable and therefore étale
by the same argument.
Example 29.2.13. Let k be any field of characteristic p > 0 and consider the field
626
29.2. Finite Branched Covers Chapter 29. Fundamental Groups of Algebraic Curves
B/(x − a)B ∼
= B/(y p )B.
In the first case, B/(y − f (a)1/p )p is not a field extension of κ(x − a) = k, while in
the second, B/(y p − f (a)) is an inseparable field extension of k. Thus ϕ is not étale
at any of these (x − a).
Thus ϕ is étale at (x − a) in this case. On the other hand, if there does not exist such
a c, then (x − a) remains a prime ideal in B and thus B/(x − a) is a finite separable
extension of k.
627
29.3. Fundamental Group of CurvesChapter 29. Fundamental Groups of Algebraic Curves
Proof. To show KU /K is a Galois extension, it’s enough to see that KU is stable under the
Gal(Ks /K)-action, but this follows from the definition of the field KU . To prove the second
statement, it’s enough to show for a finite subextension L/K of KU /K that:
(i) If L comes from a cover of X which is étale over U , then so does any subfield L ⊇ L0 ⊇
K.
(ii) If M/K is any other finite subextension of KU /K coming from a cover which is étale
over U , then LM/K also comes from a cover étale over U .
Then the result will follow since KU is the compositum of a tower of finite extensions L1 ⊆
L1 L2 ⊆ L1 L2 L3 ⊆ · · ·
ϕ
For (i), let L/K correspond to a cover Y − → X which is étale over U . Then for any
ϕ0
L ⊇ L0 ⊇ K, we get a composition of covers Y → Y 0 − → X. Applying Lemma 29.2.4,
0
we get that ϕ is étale over U . For (ii), fix P ∈ U and consider the inertia group IP ≤
Gal(Ks /K) as defined in Section 29.2. Then by (3) of Theorem 29.2.8, IP is trivial in
both Gal(L/K) and Gal(M/K). Since Gal(L/K) = Gal(Ks /K)/ Gal(Ks /L), this means
IP ⊆ Gal(Ks /L); likewise IP ⊆ Gal(Ks /M ). Hence IP ⊆ Gal(Ks /LM ), which means IP is
trivial in Gal(LM/K), so by Theorem 29.2.8 again, the curve corresponding to LM/K is
étale at P .
Definition. For a proper normal curve X over a perfect field k and a nonempty open subset
U ⊆ X, we define the algebraic fundamental group over U to be
π1 (U ) := Gal(KU /K).
Note that π1 (U ) is a profinite group which depends on the choice of separable closure
Ks . Here, this choice of Ks acts as the “basepoint” in analogy with the topological case (see
Section 28.1).
Definition. Extending the notion of curves from the last section, we define a proper nor-
mal curve to be a finite disjoint union of integral proper normal curves.
`
The ring of rational functions on a proper normal curve X L = Xi is defined to be the
direct sum of the function fields of each component, O(X) = k(Xi ); this is naturally a
finite dimensional algebra over any of the components k(Xi ). In general, we say a morphism
Y → X, with X integral, is separable if k(Y ) is étale over k(X).
628
29.3. Fundamental Group of CurvesChapter 29. Fundamental Groups of Algebraic Curves
Theorem 29.3.2. Suppose X is an integral proper normal curve over a perfect field k and
U ⊆ X is a nonempty open subset. Then there is an equivalence of categories
ϕ
covers of proper normal curves Y − →X ∼
−
→ {finite continuous left π1 (U )-sets}.
where ϕ is finite, separable, étale over U
Ks
KU A
e
K(V ) O(V )
K A
Note that for any maximal ideal m ⊆ A, e m ∩ O(V ) is a closed point in V . Although Ae is
not itself a finitely generated ring over A, it is a compositum of such rings, i.e. the O(V ) are
finite over A. Define U e = MaxSpec A e and equip this with the inverse limit topology with
respect to the inverse system V → V 0 → U . We define a sheaf of rings OUe on U e stalkwise
by setting OUe ,Qe = A e∈U
e e for any Q
Q
e . Then for any open set Ve ⊆ Ue , we put
\
OUe (Ve ) = OUe ,Qe .
Q∈
e Ue
629
29.3. Fundamental Group of CurvesChapter 29. Fundamental Groups of Algebraic Curves
Definition. U
e is called the pro-étale cover of U .
e , O e ) is a locally ringed space.
Lemma 29.3.4. (U U
Remark. Note that U e is not an object in the category of proper normal curves (or even
in the category of schemes, as we shall see!) However, constructing a “universal cover” is
useful to illustrate the analogy with the topological case. For example, we have the following
theorem which will be proven in Section 30.3.
Theorem 29.3.5. Let X be an integral proper normal curve over C and U ⊆ X a nonempty
open subset. Then π1 (U ) is isomorphic to π1top\
(U (C)), the profinite completion of the topo-
logical fundamental group of the Riemann surface U (C). In particular, π1 (U ) is the profinite
completion of
ha1 , b1 , . . . , ag , bg , γ1 , . . . , γn | [a1 , b1 ] · · · [ag , bg ]γ1 · · · γn = 1i,
where n is the number of points in X r U and g is the genus of the Riemann surface U (C).
Let X/C and U be as above. If G is any finite quotient of π1 (U ), then G corresponds
to a finite Galois branched cover Y → X. Here, the image of each generator γi ∈ π1 (U ) in
G is a cyclic generator of the stabilizer of a point Qi → Pi , where Pi ∈ X r U is the point
corresponding to γi .
Let X be an integral proper normal curve over an arbitrary (perfect) field k. If L/k is
a finite extension, then k(X) ⊗k L is a finite dimensional algebra over L(t) which is not, in
general, a field. Assume k(X) ⊗k L is a finite direct product of fields L1 , . . . , Ln , each of
which is a finitely generated extension of L(t) of transcendence degree 1 over L. Then each
Li corresponds to an ` integral proper normal curve Xi /L by Proposition 29.1.4. Define the
base change XL = Xi . This comes equipped with a natural morphism of proper normal
curves XL → X. We will prove:
Theorem 29.3.6. If k is an algebraically closed field of characteristic 0, L/k is a finite
extension, X is an integral proper normal curve over k and U ⊆ X is open, then the base
change functor X 7→ XL induces an equivalence of categories
finite branched Galois covers Y → X ∼ finite branched Galois covers YL → XL
−
→ .
étale over U étale over UL
Corollary 29.3.7. For U ⊆ X as above, there is an isomorphism π1 (UL ) ∼
= π1 (U ) for all
finite extensions L/k.
Theorem 29.3.5 generalizes to curves over an algebraically closed field k of characteristic
0 in the following sense.
Corollary 29.3.8. If k is algebraically closed of characteristic 0 and U ⊆ X is a nonempty
open subset of an integral proper normal curve X over k, then π1 (U ) is isomorphic to the
profinite completion of a group with presentation
ha1 , b1 , . . . , ag , bg , γ1 , . . . , γn | [a1 , b1 ] · · · [ag , bg ]γ1 · · · γn = 1i,
where g = g(X) is the genus of X and n is the number of points in X r U .
630
29.3. Fundamental Group of CurvesChapter 29. Fundamental Groups of Algebraic Curves
2g(X) − 2 = n(0 − 2) + 0
but this is only possible if g(X) = 0 and n = 1. Thus X → P1k is a birational isomorphism,
but since X is complete, it must be a regular isomorphism. Hence there are no nontrivial
extensions L of k(P1k ) = k which proves π1 (P1k ) = 1.
Example 29.3.10. A similar argument shows π1 (A1k ) = 1 when k is algebraically closed and
char k = 0. Both this and the previous result also follow easily from Theorem 29.3.5 and
Corollary 29.3.8.
Example 29.3.11. Theorem 29.3.5 and Corollary 29.3.8 also imply that π1 (P1k r {0, ∞}) ∼ =
Z, the profinite completion of the integers. Using Theorem 29.3.2, we get a finite étale cover
b
Xn → P1k r {0, ∞} for each n ≥ 1 having Galois group Z/nZ.
Example 29.3.12. By Theorem 29.3.5 and Corollary 29.3.8, π1 (P1k r {0, 1, ∞}) is the free
profinite group on two generators.
631
29.4. The Outer Galois Action Chapter 29. Fundamental Groups of Algebraic Curves
KU
π1 (U ) K k̄
Gal(k̄/k)
Proposition 29.4.1. For X a geometrically integral, proper normal curve over a perfect
field k and for any nonempty open subset U ⊆ X, there is a short exact sequence of profinite
groups
1 → π1 (Uk̄ ) → π1 (U ) → Gal(k̄/k) → 1.
π1 (Uk̄ ) K0
K k̄
This also determies a finite branched cover Y0 → Xk̄ which is étale over Uk̄ ⊆ Xk̄ . Let
f ∈ K k̄[t] be the minimal polynomial of K0 /K k̄. Then we can find a finite extension L/k
such that KL/k contains all the coefficients of f . Let L0 /KL be the finite extension with
Gal(L0 /KL) = G.
632
29.4. The Outer Galois Action Chapter 29. Fundamental Groups of Algebraic Curves
K0
G
π1 (U ) L0 K k̄
G
KL k̄
K L
Definition. For an open subset U of a geometrically integral proper normal curve X over a
perfect field k, the subgroup π1 (Uk̄ ) ≤ π1 (U ) is called the geometric fundamental group
of U . The map pU : Gal(k̄/k) → Out(π1 (Uk̄ ) is called the outer Galois action on the
geometric fundamental group.
the residue field κ(P ) = OX,P /P OX,P . We have a natural surjection DQe → Gal(κ(Q)/κ(P
e ))
with kernel denoted IQe , called the inertia group of Q.e We may alternatively view X e as
a profinite Galois cover of Xk̄ with Galois group π1 (Uk̄ ), corresponding to the short exact
sequence from Proposition 29.4.1:
633
29.4. The Outer Galois Action Chapter 29. Fundamental Groups of Algebraic Curves
X
e
π1 (Uk̄ )
π1 (U ) Xk̄
Gal(k̄/k)
634
29.4. The Outer Galois Action Chapter 29. Fundamental Groups of Algebraic Curves
One of the most important objects in arithmetic geometry is π1 (P1Q r {0, 1, ∞}). If
U = P1Q r {0, 1, ∞}, there is an outer Galois action ρ : Gal(Q/Q) → Out(π1 (UQ )). We will
prove that this action is faithful, but first we need:
Theorem 29.4.6 (Belyi). Suppose X is an integral proper normal curve defined over an
algebraic closed field k of characteristic 0. Then there exists a morphism X → P1k which is
étale over P1k r {0, 1, ∞} if and only if X is defined over Q.
Proof. ( =⇒ ) follows from Theorem 29.3.6.
( ⇒= ) If X is defined over Q, there exists a map π : X → P1Q which is étale over P1Q r S
1
for some finite set S ⊆ PQ . We first show S consists of Q-rational points. Since S is finite,
we can find a point P ∈ S with [κ(P ) : Q] = n maximal among the points in S. Let
f ∈ Q[t] be the minimal polynomial of κ(P )/Q. Define ϕf : P1Q → P1Q by x 7→ F (x), where
F is the homogenization of f . Then ϕf is defined over Q and is étale over P1Q r Sf , where
Sf = {R ∈ AQ 1
| f 0 (R) = 0}. Notice that deg f 0 = n − 1 so for all R ∈ Sf , [κ(R) : Q] < n.
The composition ϕf ◦ π : X → P1Q is now étale away from S 0 := ϕf (S) ∪ {∞} ∪ ϕf (Sf ). But
now ∞ has degree 1, points in ϕf (S) have degree at most n and points in ϕf (Sf ) have degree
at most n − 1, so because ϕf (P ) = 0, there are strictly less points of degree n in S 0 than in
S. Repeat this procedure until n = 1, at which time all points in S 0 will be Q-rational.
Next, suppose S has more than three points. We may assume 0, 1, ∞ ∈ S and take
α ∈ S r {0, 1, ∞}. Define the Belyi function
1
ϕ : PQ −→ P1Q
x 7−→ xA (x − 1)B
635
29.4. The Outer Galois Action Chapter 29. Fundamental Groups of Algebraic Curves
π1 (UQ ) is a free profinite group on 2 generators, and such a group has trivial center, so we
must have π1 (UQ ) ∩ C = {1}. Thus π1 (UL ) ∼ = π1 (UQ ) × C as a direct product, or in other
words, π1 (UL ) → π1 (UQ ) is a continuous retraction of profinite groups. This says that any
continuous left π1 (UQ )-set comes from a continuous left π1 (UL )-set, but by Theorem 29.3.6,
any finite cover of UQ is obtained by base change from some finite cover of UL . Finally,
Belyi’s theorem says that any such cover defined over Q can be defined over L, but we will
demonstrate that this is impossible unless L = Q, in which case ker ρ is trivial.
If L 6= Q, take x ∈ Q r L. Let E 0 be an elliptic curve over Q with j-invariant j(E 0 ) = x.
If E 0 can be obtained by base change from some genus 1 curve X defined over L, as Belyi’s
theorem indicates, then X has a Jacobian E/L which is an elliptic curve. There exists an
embedding φ : E → P2L sending the distinguished L-point of E to [0, 1, 0] ∈ P2L . Moreover,
it is known that j(E) ∈ L and this j-invariant is preserved by any Q-isomorphism. Since E 0
is the base change of X, we have E 0 ∼ = XQ ,→ E, but this implies j(E) = j(E 0 ) = x 6∈ L, a
contradiction. Hence L = Q, so the outer Galois representation of U is faithful.
Example 29.4.8. One can similarly prove that for any elliptic curve E over Q (or any
number field K), the outer Galois representation ρ : Gal(Q/Q) → π1 (E r {0}) is faithful.
636
29.5. The Inverse Galois Problem Chapter 29. Fundamental Groups of Algebraic Curves
Question (Inverse Galois Problem). Which groups arise as Galois groups Gal(L/Q) for
some finite Galois extension L/Q?
Example 29.5.1. Let Cn = Z/nZ be a cyclic group of order n. We know that the cyclotomic
extension Q(ζp ) has Galois group Gal(Q(ζp )/Q) ∼
= Z/(p−1)Z, so by group theory, if n | p−1,
this Galois group has a subgroup H of order (p−1)/n. Hence the subextension Q(ζp )H /Q has
group Gal(Q(ζp )H /Z) = Z/nZ. Thus the inverse Galois problem for cyclic groups reduces to
finding a prime p such that p ≡ 1 (mod n), but by Dirichlet’s theorem, there are infinitely
many such primes p. Thus all cyclic groups arise as Galois groups over Q. Using the theory
of cyclotomic extensions, this can be generalized to give a positive solution to the inverse
Galois problem for all finite abelian groups.
Combined with the Kronecker-Weber theorem (every finite abelian extension of Q lies
in some cyclotomic extension Q(ζm )/Q), the conclusion of Example 29.5.1 gives a complete
description of abelian extensions of Q, namely that S the maximal abelian extension of the
rationals Qab is equal to the compositum Qcyc := m≥1 Q(ζm ). The study of Gal(Qab /Q)
and its subgroups is the content of global class field theory.
If a group G arises as a Galois group over Q, we may then ask for all G-extensions of Q.
This may be a much harder question to answer.
Example 29.5.2. One can show that the splitting field of f (x) = x3 + x2 − 2x − 1 defines
a C3 -extension of Q. How do we determine all C3 -extensions? Well notice that C3 embeds
as a subgroup of P GL2 (Q) = Aut(P1Q ) via
C3 = hσi −→ Aut(P1 )
1
σ 7−→ x 7→ .
1−x
P1 −→ P1 /C3 ∼
= P1
x3 − 3x − 1
x 7−→ T = .
x2 − x
637
29.5. The Inverse Galois Problem Chapter 29. Fundamental Groups of Algebraic Curves
By construction, Gal(Q(x)/Q(T )) ∼
= C3 and this extension has minimal polynomial
f (x, T ) = x3 − T x2 + (T − 3)x + 1.
A polynomial f (x, T ) that satisfies these conditions for some fixed F is called a versal family
for G-extensions of F . Thus a generic polynomial for K is one that is a versal family for
all extensions of K.
Example 29.5.4. For C8 -extensions of Q, there is a versal family for Q, but there is no
generic polynomial for all extensions of Q.
Example 29.5.5. By the Remark, the fact that C4 does not embed as a subgroup of
P GL2 (Q) implies that there does not exist a 1-parameter generic polynomial for C4 -extensions
of Q. However, there does exist an embedding C4 ,→ P GL2 (Q(i)), so there may be a 1-
parameter generic polynomial over Q(i).
Question (Noether’s Problem). For which groups G is the field E = Q(x1 , . . . , xn )G tran-
scendental over Q?
The inverse Galois problem is known to be a consequence of the so-called regular inverse
Galois problem:
Question (Regular Inverse Galois Problem). Which groups G arise as Galois groups of
finite Galois extensions K/Q(t) which are regular over Q?
638
29.5. The Inverse Galois Problem Chapter 29. Fundamental Groups of Algebraic Curves
(A regular extension K/Q(t) is one which does not contain a subextension of the form
L(t) for L/Q a nontrivial extension.) To see this implication, we need Hilbert’s irreducibility
theorem:
Theorem 29.5.6 (Hilbert Irreducibility). Let G be a finite group acting on {x1 , . . . , xn }. If
AnQ /G is rational over Q, then there exist infinitely many points P ∈ AnQ /G with Gal(Q(Q)/Q) ∼ =
−1
G for each Q ∈ π (P ).
Example 29.5.7. Let G = Sn be the symmetric group on n symbols. Let F = Q(x1 , . . . , xn ).
Then E = F Sn = Q(σ1 , . . . , σn ) where σi is the ith elementary symmetric polynomial and
the extension F/E has minimal polynomial
for σ = (σ1 , . . . , σn ). This defines an Sn -cover AnQ → AnQ /Sn . By Hilbert’s irreducibility
theorem, f (x, s) = xn − s1 xn−1 + . . . + (−1)n sn is irreducible for infinitely many tuples
s = (s1 , . . . , sn ) ∈ Qn . It is a fact that
Using this, one can prove that this polynomial f (x, σ) is a generic polynomial for Sn -
extensions of Q.
Similar proofs using Hilbert’s irreducibility theorem can be used to verify generic poly-
nomials for other classes of G-extensions. A question related to Noether’s Problem and the
regular IGP is:
Question. For which groups G and fields K is AnK /G rational?
The answer to this was demonstrated to be no the general case: Lenstra provided coun-
terexamples for K = Q and G = C8 , and Saltman even showed a counterexample for K = C.
The question is even unknown for G = An , n > 5.
Here is a strategy to solving the regular Inverse Galois Problem for a given group G.
(1) If G has a generating set consisting of n − 1 elements, then G arises as a quotient of
the topological fundamental group π1 (P1C r {P1 , . . . , Pn }), since this is the free group on
n − 1 elements. Set π top (n) = π1 (P1C r {P1 , . . . , Pn }). In particular, suppose we can find
g1 , . . . , gn ∈ G such that G = hg1 , . . . , gn i, g1 · · · gn = 1 and create a map ϕ : π top (n) → G
with ϕ(γi ) = gi for 1 ≤ i ≤ n. We call (g1 , . . . , gn ) ∈ G a generating n-tuple for G.
1
(2) Set π(n) = π1 (PQ top (n). We may
r {P1 , . . . , Pn }) so that by Corollary 29.3.8, π(n) = π\
even pick P1 , . . . , Pn ∈ P1Q (Q).
(3) Also set Π(n) = π1 (P1Q r {P1 , . . . , Pn }) so that Proposition 29.4.1 gives us a short exact
sequence of profinite groups
1 → π(n) → Π(n) → GQ → 1.
639
29.5. The Inverse Galois Problem Chapter 29. Fundamental Groups of Algebraic Curves
γi π(n) Π(n)
ϕ
ϕ
b
gi G
N Γ
where N is a normal subgroup of a profinite group Γ and G is any finite group. The set
Homcts (N, G) comes equipped with two actions:
Lemma 29.5.8. Let S ⊆ Homcts (N, G) be a set of maps which are stable under the G- and
Γ-actions and such that G acts freely and transitively on S. Then any ϕ ∈ S extends to a
continuous homomorphism ϕ b : Γ → G.
640
29.5. The Inverse Galois Problem Chapter 29. Fundamental Groups of Algebraic Curves
Now to access Lemma 29.5.8, we want to construct such a set S ⊆ Homcts (π(n), G).
First, we need S to be stable under the action of π(n). Note that for ϕ ∈ S, having ϕσ ∈ S
for all σ ∈ π(n) is equivalent to having ϕσ (n) = ϕ(σ)ϕ(n)ϕ−1 = (ϕ(σ) · ϕ)(n), under the
left action of G on ϕ. Let (g1 , . . . , gn ) be a generating n-tuple for G. Then we must have
gi = ϕ(γi ) stable under the conjugacy action of G. In particular, we will specify conjugacy
classes C1 , . . . , Cn in G and consider sets
By construction, this S is stable under the π(n)- and G-actions. Now we impose further
conditions on the Ci to force
(i) transitivity;
(iii) Π(n)-stability.
Definition. Let G be a finite group. We say a set of conjugacy classes C1 , . . . , Cn in G is a
rigid system if there exists a generating n-tuple (g1 , . . . , gn ) ∈ Gn for G with gi ∈ Ci and
G acts transitively on the set of all such tuples.
To ensure (i), we pick a rigid system of conjugacy classes in G represented by the ϕ(γi ).
To guarantee (ii), it is enough to assume that Z(G) = 1 (G has trivial center). Next, set
Σ = {(g1 , . . . , gn ) ∈ Gn | g1 · · · gn = 1 and gi ∈ Ci }
Σ = {(g1 , . . . , gn ) ∈ Σ | hg1 , . . . , gn i = G}.
641
29.5. The Inverse Galois Problem Chapter 29. Fundamental Groups of Algebraic Curves
Proof. Topologically, each γi generates an inertia group IQei in π(n) for some pro-point Qei →
ei ) is another pro-point over Pi and σγi σ −1 generates I e .
Pi . Also, for each σ ∈ Π(n), σ(Q σ(Qi )
1 1
But Pi ∈ P (Q) so Qi and σ(Qi ) are both pro-points above the same point P i ∈ P which
Q
e e
Q
1
lies over Pi . Thus IQei and Iσ(Qei ) are both stabilizers in π1 (PQ ) of points above some P i .
This implies these groups are conjugate in π(n). Write Iσ(Qei ) = αIQei α−1 for α ∈ π(n). Then
ϕ(Iσ(Qei ) ) = ϕ(α)ϕ(IQei )ϕ(α)−1 so ϕ(IQei ) and ϕ(Iσ(Qei ) ) are conjugate in G and cyclically
generated by ϕ(γi ) and ϕ(σγi σ −1 ), respectively. So for some g ∈ G and m prime to |G|, we
have ϕ(σγi σ −1 ) = gϕ(γi )m g −1 . Finally, since Ci is rational, this means ϕ(γi )m ∈ Ci and so
ϕ(σγi σ −1 ) ∈ Ci . Hence ϕ is Π(n)-stable.
Theorem 29.5.11. For a finite group G with Z(G) = 1, suppose there exists a rigid system
of rational conjugacy classes C1 , . . . , Cn in G. Then for any P1 , . . . , Pn ∈ P1Q (Q), there is a
continuous, surjective homomorphism
ϕ : π1 (P1Q r {P1 , . . . , Pn }) −→ G
Example 29.5.12. For Sn , we saw that {nA, 2A, C (1) } is a strictly rigid system. Also, any
conjugacy class in Sn is rational since g 7→ g m preserves cycle type. In general, one can
prove this holds for {nA, 2A, C (k) }, 1 ≤ k ≤ n. This choice of rigid system corresponds to
the cover
P1 −→ P1
x 7−→ xk (x − 1)n−k
642
Chapter 30
643
30.1. Riemann Surfaces Chapter 30. Riemann’s Existence Theorem
ϕij := ϕj ◦ ϕ−1
i : ϕi (Ui ∩ Uj ) −→ ϕj (Ui ∩ Uj )
and its inverse are holomorphic. A complex structure on X is the choice of a complex
atlas, up to holomorphic equivalence of charts, defined by a similar condition to the above.
A connected surface which admits a complex structure is called a Riemann surface.
Example 30.1.1. The complex plane C is a trivial Riemann surface. Any connected open
subset U in C is also a Riemann surface via the given embedding U ,→ C.
Example 30.1.2. The complex projective line P1 = P1C = C ∪ {∞} admits a complex
structure defined by the open sets U0 = P1 r {∞} = C and U1 = P1 r {0} = C× ∪ {∞},
together with charts
1
ϕ0 : U0 → C, z 7→ z and ϕ1 : U1 → C, z 7→ ,
z
1
where ∞ = 0 by convention. Note that ϕ1 ◦ ϕ−1
0 is the function z 7→ 1
z
on C× which is
holomorphic.
Example 30.1.3. Let Λ ⊆ C be a lattice with basis [ω1 , ω2 ].
ω1
ω2
644
30.1. Riemann Surfaces Chapter 30. Riemann’s Existence Theorem
Then the quotient C/Λ admits a complex structure as follows. Let π : C → C/Λ be the
quotient map and suppose Π ⊆ C is a fundamental domain for Λ, meaning no two points in
Π are equivalent mod Λ. Set U = π(Π) ⊆ C/Λ. Then π|Π : Π → U is a homeomorphism,
so let ϕ : U → Π be its inverse. Letting {Ui } be the collection of all images under π of
fundamental domains for Λ, we get a complex atlas on C/Λ (one can easily check that the
transition functions between the Ui are locally constant, hence holomorphic). Topologically,
C/Λ is homeomorphic to a torus.
Lemma 30.1.4. For any open set U of a Riemann surface X, O(U ) is a commutative
C-algebra.
More generally, we can define holomorphic maps between two Riemann surfaces.
f ∗ : O(U ) −→ O(f −1 (U ))
ψ 7−→ f ∗ ψ := ψ ◦ f.
Proof. The fact that f ∗ ψ is an element of O(f −1 (U )) follows from the above definitions of
O and a holomorphic map between Riemann surfaces. The ring axioms are also easy to
verify.
645
30.1. Riemann Surfaces Chapter 30. Riemann’s Existence Theorem
Proof. Let U ⊆ X be the set of all x ∈ X with an open neighborhood W on which f |W = g|W .
Then U is open and a ∈ U ; we will show it is also closed. If x ∈ ∂U , we have f (x) = g(x) since
f and g are continuous. Choose a neighborhood W ⊆ X of x and charts ϕ : W → ϕ(W ) ⊆ C
and ψ : W 0 → ψ(W 0 ) ⊆ C in Y with f (W ) ⊆ W 0 and g(W ) ⊆ W 0 . Consider
Theorem 30.1.11. Suppose X is a Riemann surface. Then the set of meromorphic func-
tions M(X) is in bijection with the set of holomorphic maps X → P1 .
Proof. If f ∈ M(X) is a meromorphic function, then setting f (p) = ∞ for every pole p of
f defines a holomorphic map f : X → P1 . Indeed, it is clear that f is continuous. Let P be
the set of its poles. If ϕ : U → ϕ(U ) ⊆ C is a chart on X and ψ : V → ψ(V ) ⊆ C is a chart
on P1 with f (U ) ⊆ V , then since f is holomorphic on X r P , ψ ◦ f ◦ ϕ−1 : ϕ(U ) → ψ(V ) is
holomorphic on ϕ(U ) r ϕ(P ). By Proposition 30.1.5, ψ ◦ f ◦ ϕ−1 is actually holomorphic on
ϕ(U ), so f is a holomorphic map of Riemann surfaces.
Conversely, if g : X → P1 is holomorphic, then by Theorem 30.1.8, either g(X) = {∞}
or g −1 (∞) is a set of isolated points in X. It is then easy to see that g : X r g −1 (∞) → C
is meromorphic.
Corollary 30.1.12 (Meromorphic Continuation). For any open set U ⊆ X and any x ∈ U ,
if f ∈ M(U r {x}) is bounded in a neighborhood of x, then f extends uniquely to some
f˜ ∈ M(U ).
Proof. Apply Proposition 30.1.5 and Theorem 30.1.11.
Corollary 30.1.13. Any nonzero function in M(X) has only isolated zeroes. In particular,
M(X) is a field.
646
30.1. Riemann Surfaces Chapter 30. Riemann’s Existence Theorem
647
30.1. Riemann Surfaces Chapter 30. Riemann’s Existence Theorem
648
30.1. Riemann Surfaces Chapter 30. Riemann’s Existence Theorem
Proof. Let ϕ : V → C be a chart of X. Then there exists an open subset U ⊆ V over which
p|U : p−1 (U ) → U is a homeomorphism. Set U e = p−1 (U ) and ϕ
e = ϕ ◦ p|U : Ue → C. Then
e is a complex chart on Y and the collection {U
ϕ e , ϕ}
e obtained in this way forms a complex
atlas on Y . Since p : Y → X is locally biholomorphic by construction, it is a holomorphic
map between Riemann surfaces. Uniqueness is easy to check.
Example 30.1.28. Now that we can view nonconstant holomorphic maps as local homeo-
morphisms, and in most cases covering spaces, we can rephrase the language of branch cuts
as a lifting problem. For example, let exp : C → C× be the exponential map and suppose
f : X → C× is a holomorphic map of Riemann surfaces, with X simply connected. Then by
covering space theory, for each fixed x0 ∈ X and z0 ∈ C such that f (x0 ) = ez0 , there exists
a unique lift F : X → C making the diagram
C
F exp
X C×
f
commute. Theorem 30.1.27 can be used to show that any such F is holomorphic. Moreover,
any other lift G of f differs from F by 2πin for some n ∈ Z. For the special case of a simply
connected open set X ⊆ C× , any lift F is a branch of the complex logarithm on X.
Example 30.1.29. Similarly, one can construct the complex root functions z 7→ z 1/n , n ≥ 2,
as lifts along the cover pn : C× → C.
Let f : Y → X be a nonconstant holomorphic map that is proper, i.e. the preimage of
any compact set in X is compact in Y . For each x ∈ X, define the multiplicity of f at x to
be X
ordx (f ) = vy (f )
y∈f −1 (x)
649
30.1. Riemann Surfaces Chapter 30. Riemann’s Existence Theorem
Corollary 30.1.33. Any complex polynomial f (z) ∈ C[z] of degree n has exactly n zeroes,
counted with multiplicity.
650
30.2. The Existence Theorem over P1C Chapter 30. Riemann’s Existence Theorem
γ1 γ2 γr
P1 P2 ··· Pr
Specify a generating r-tuple g1 , . . . , gr ∈ G – recall from Section 29.5 that this means
hg1 , . . . , gr i = G and g1 · · · gr = 1. Then G is obtained as a quotient of Γ by mapping
γi 7→ gi for each 1 ≤ i ≤ r. We have seen that the inertia groups at the points Pi ∈ P are
cyclicly generated, IPi = hgi i. The following lemma allows us to dispense with worries about
“canonical” generators γ1 , . . . , γr for Γ.
Lemma 30.2.2. If Γ1 and Γ2 are two finitely generated groups with the same homomorphic
images, then their profinite completions Γ
b1 and Γ
b2 are isomorphic.
On the other hand, with K = C(t), we have that Gal(K/K) = Fbω is free profinite of
(uncountably) infinite rank. By definition Gal(K/K) = lim Gal(L(P )/K) where the inverse
←−
limit is taken over all finite sets P ⊂ P1 , ordered by P ⊆ P 0 which induces L(P ) ⊆ L(P 0 )
and thus Gal(L(P 0 )/K) → Gal(L(P )/K). In particular, for each containment of finite sets
P ⊆ P 0 we have a diagram
651
30.2. The Existence Theorem over P1C Chapter 30. Riemann’s Existence Theorem
∼
= ∼
=
The topological side of the proof of Theorem 30.2.1 is essentially contained in the following
theorem.
Theorem 30.2.3. Let P ⊂ P1C be a finite set and π : R → P1C r P a finite-sheeted Galois
covering of topological spaces. Then there exists a smooth projective curve W with a regular
map f : W → P1C which is unramified outside P and such that the covering
W r f −1 (P ) → P1C r P
Ui
q R
P2 B(z, ε) P3
P1 z P1
Take q ∈ R r π −1 (P ) and set z = π(q). Since P is finite, we can choose a ball B(z, ε) which
does not contain any point of P . Then π −1 (B(z, ε)) = U1 q · · · q Un for disjoint open sets Ui
for which π|Ui : Ui → B(z, ε) is a homeomorphism. One of these Ui contains q; let this Ui be
the coordinate chart defining the surface structure on R. For two of these charts, U = Ui and
V = Vj , the transition maps are just given by lifting the transition maps B(z, ε) → B(z 0 , ε0 )
652
30.2. The Existence Theorem over P1C Chapter 30. Riemann’s Existence Theorem
along the local homeomorphism π, and these are analytic on the disks B(z, ε), B(z 0 , ε0 ) so
they are analytic on the U, V . This gives R the structure of a Riemann surface, but it may
not be compact.
In the next step, we show that there exists a compact Riemann surface R with an analytic
map π̄ : R → P1 and an open embedding R ,→ R such that the diagram
R R
π π̄
P1 r P P1
commutes. For any ball B = B(z, ε) ⊆ P1 , let B 0 = B 0 (z, ε) = B r {z} be the disk
punctured at its center. Then π1top (B 0 ) ∼
= Z and in particular B 0 has (up to isomorphism)
only one connected, degree d cover for each d ≥ 1. Call this cover µd . When B = B(0, 1) is
the unit ball, this degree d cover is explicitly given by
µd : B 0 −→ B 0
z 7−→ z d .
Moreover, µd extends to a branched cover of the entire disk B:
µd
B0 B0
µd
B B
The deck transformations of this cover are also easy to write down: they are generated by
τ : z 7→ ζz where ζ = ζd is a primitive dth root of unity.
Let z0 ∈ P and take a disk D = B(z0 , ε) around z0 which does not contain any other
points in P . Let D0 = D r {z0 }. Then as above, the preimage of D0 along π can be written
as a disjoint union of connected components,
π −1 (D0 ) = D10 q · · · q Dm
0
where m ≤ n, such that for each 1 ≤ i ≤ m, π|Di0 : Di0 → D0 is a covering map. We now
construct a Riemann surface R1 together with an analytic map π1 : R1 → P1 r (P r {z0 })
and an embedding R ,→ R1 such that |R1 r R| = m and the diagram
R R1
π π1
P1 r P P1 r (P r {z0 })
653
30.2. The Existence Theorem over P1C Chapter 30. Riemann’s Existence Theorem
commutes. After scaling by some λ ∈ C, we may replace D0 with B 0 = B(0, 1) r {0}. Then
there exists an integer di and a homeomorphism τi : Di → B 0 making the following diagram
commute:
τi
Di0 B0
π µdi
λ
D0 B0
It is an easy exercise to show that τi is in fact biholomorphic. Notice that while τi is not
unique, it is defined up to a deck transformation of µdi . We now construct R1 by taking m
0
copies of B, say B1 , . . . , Bm , and the corresponding punctured disks B10 , . . . , Bm and setting
T = B1 q · · · q Bm . Obtain R1 by gluing R to T along the homeomorphisms
D10 q · · · q Dm
0
−→ B10 q · · · q Bm
0
.
Repeating this process, we obtain R after a finite number of steps, i.e. whenever all points
of π −1 (P ) have been “filled in”.
σ
Bi0 Bj0
commutes but by Proposition 30.1.5, since σ has bounded image it extends to Bi → Bj and
hence to σ̄ : Di → Dj . Now for any z0 ∈ P , let D0 = D r {z0 }. Then each Di0 in the disjoint
union π −1 (D0 ) = D10 q · · · q Dm
0
corresponds to a point qi ∈ R such that π(qi ) = z0 , and
which we glue to R to obtain R. Moreover, for each qi , qj , we may choose points xi ∈ Di0 and
xj ∈ Dj0 such that xi 6= qi and xj 6= qj . Now π is Galois, so there exists a deck transformation
σ ∈ ∆(π) mapping σ(xi ) = xj . Then we must have σ(Di0 ) = Dj0 and hence by the above, the
extension σ̄ maps σ̄(qi ) = qj . This proves π̄ is a Galois covering.
Suppose π : R → P1 r P is a Galois cover and set G = ∆(π) = Aut(π), the group of deck
transformations. For any analytic curve Y , let M(Y ) be the field of meromorphic functions
on Y . For Y = R as above, we have an action of G on M(R) by (σf )(x) = f (σ −1 (x)).
654
30.2. The Existence Theorem over P1C Chapter 30. Riemann’s Existence Theorem
Lemma 30.2.6. For any f ∈ M(R) which is invariant under the G-action described above,
then f ∈ M(P1 ) ⊆ M(R). In other words, f = g ◦ π for some g ∈ M(P1 ).
Proof. By the construction in Lemma 30.2.4, there exists a function g : P1 → P1 such that
f = g ◦ π. Thus we need only show that g is meromorphic. Take p ∈ R and consider
q = π(p) ∈ P1 . If q 6∈ P , the set of branch points of π, then locally f is given by f = g ◦ π :
D(p) → D(q) on local trivializations D(p) ⊆ R and D(q) ⊆ P1 of p, q, respectively. But
locally, π is biholomorphic on these neighborhoods so we can invert π and write g = f ◦ π −1
and it follows that g is meromorphic.
If q ∈ P , look at the punctured neighborhoods D(p) = D(p) r {0}, D(q) = D(q) r {0}
and the cover π : D(p) → D(q). This is not holomorphic, but we know that it’s a degree
m map that in fact just corresponds to the multiplication-by-m map on the unit punctured
disk D(0, 1):
π
D(p) D(q)
∼
= ∼
=
[m]
D D
P∞
Write f as a Laurent series f (z) = j=−N aj z j for aj ∈ C and N ≥ 0. Then we must have
f (ζm z) = f (z) for any root of unity ζm . Thus aj is nonzero only when m | j and we can
rewrite ∞
X
f (z) = akm z km .
k=−K
Lemma 30.2.6 shows that there exists a finite group G acting on M(R) such that
M(R)G = M(P1 ). By Serre’s GAGA principle, M(P1 ) = C(t), so therefore we have proven
that [M(R) : C(t)] < ∞.
655
30.2. The Existence Theorem over P1C Chapter 30. Riemann’s Existence Theorem
Proof. The paragraph above implies, by Artin’s Lemma, that M(R)/C(t) is Galois. Further,
Gal(M(R)/C(t)) = G/N where N is the kernel of the action of G on M(R). To prove this,
pick a point p ∈ P1 and write π −1 (p) = {p1 , . . . , pn }. By the analytic version of Riemann’s
existence theorem (30.2.7), there exists a meromorphic g ∈ M(R) such that the values
g(p1 ), . . . , g(pn ) are all distinct. Suppose σ ∈ N is nontrivial. Then σ(pi ) = pj 6= pi for some
i 6= j, but σ ∈ N implies σ(g) = g, or g ◦ σ = g. Thus g(pi ) = g(σ(pi )) = g(pj ), which is
impossible. Therefore N = {1} so Gal(M(R)/C(t)) = G as claimed.
The final step in the proof of Theorem 30.2.3 is contained the following lemma.
Lemma 30.2.9. There exists a smooth projective curve W over C and a biholomorphic
isomorphism R ∼
= W (C).
Proof. We know M(R) is a transcendence degree 1 extension of C, so by Corollary 29.1.9
there exists a proper normal curve W with C(W ) = M(R). Abstractly, there is a map
ϕ : R → W which sends a point r ∈ R to the associated valuation vr on the field M(R),
which is now identified with a valuation v on C(W ) that is trivial on C, and hence determines
a point of W .
We may assume W is smooth and projective over C. Let W ,→ PN for large enough N .
Write PN = {[x0 , x1 , . . . , xN ] : xi ∈ C, some xi 6= 0}. Then for any x ∈ R, there is an affine
patch on which the map ϕ : R → W (C) is defined, so on this patch, ϕ = (f1 , . . . , fN ) where
fi = xx0i . But by properness, this must in fact be holomorphic. Moreover, Theorem 30.2.7
implies ϕ is also injective. Further, properness also implies ϕ(R) ⊆ W is closed. Therefore
to prove ϕ is a biholomorphic isomorphism, it suffices to show W (C) is connected.
This is a rather standard property, but we repeat the proof here. Given that W is
irreducible, smooth and projective over C, let us assume W (C) = M1 ∪ M2 for some open
sets M1 , M2 ⊂ W . Pick x0 ∈ M1 . Then by the Riemann-Roch theorem, there exists a
rational function f ∈ C(W ) that has a pole at x0 and is holomorphic everywhere else. View
f : M2 → P1 ; this cannot be constant since M2 is infinite and in particular dense in W (C).
On the other hand, f (M2 ) ⊆ P1 is open and W (C) is compact, so M2 is compact and
therefore f (M2 ) is also closed. Thus f (M2 ) = P1 so some point m ∈ M2 has f (m) = ∞,
contradicting the fact that f was holomorphic on M2 . Hence W (C) is connected, so we are
finished.
We derive the following important consequences.
Proof. By Theorem 30.2.3, ∆(π) is a subgroup of Aut(E/K) but in general we have | Aut(E/K)| ≤
[E : K]. Therefore Corollary 30.2.10 implies Aut(E/K) = ∆(π) and therefore the extension
is Galois.
656
30.2. The Existence Theorem over P1C Chapter 30. Riemann’s Existence Theorem
657
30.3. The General Case Chapter 30. Riemann’s Existence Theorem
(ϕ∗ an )f n + . . . + ϕ∗ a0 = 0.
(Here, n ≤ d.) Suppose one of the ai has a pole at x. Then we may choose a small enough
neighborhood about x not containing any branch points such that f is holomorphic and
assigns distinct values to all preimages of each point. Replacing x with any of the points of
this neighborhood, we may assume each ai is holomorphic at x. Finally, take g ∈ M(Y ).
Then by the primitive element theorem, M(X)(f, g) = M(X)(h) for some h ∈ M(Y ). But
then M(X)(f ) ⊆ M(X)(h) and by the first paragraph, h has degree ≤ d over M(X), so
M(X)(f ) = M(X)(h). In particular, M(Y ) = M(X)(f ) so the proof is complete.
We now generalize the results from Section 30.2 to arbitrary covers of Riemann surfaces.
Lemma 30.3.2. Suppose X is a Riemann surface, S ⊂ X a discrete set of points and
ϕ : Y → X r S a proper holomorphic cover of Riemann surfaces. Then there is a Riemann
surface Y and proper holomorphic map ϕ̄ : Y → X such that Y r ϕ−1 (S) = Y .
Proof. This is only a slight generalization of the proof of Lemma 30.2.4. For details, see
Forster.
Lemma 30.3.3. If ϕ : Y → X r S is a Galois covering, then ϕ̄ : Y → X is Galois as well.
Proof. A generalization of Lemma 30.2.5.
658
30.3. The General Case Chapter 30. Riemann’s Existence Theorem
Theorem 30.3.4. Let X be a connected, compact Riemann surface with field of meromorphic
functions M(X). Then for every finite étale M(X)-algebra A, there is a compact Riemann
surface Y with M(Y ) ∼
= A and a holomorphic map Y → X.
Corollary 30.3.5. For any connected, compact Riemann surface X, there is an anti-equivalence
of categories
proper holomorphic maps of ∼
−
→ {finite étale M(X)-algebras}.
Riemann surfaces Y → X
659
30.4. The Analytic Existence Theorem Chapter 30. Riemann’s Existence Theorem
compactness of X implies there is a finite subcover of each B(ai , εi ) and therefore each Ui .
Taking a finite subcover consisting of such sets, we obtain a cover W = {Wj }N j=1 of X such
that each W j is compactly contained in some Ui . We say the cover W is compactly contained
in U.
Proposition 30.4.2. If W is compactly contained in U then the restriction map Ȟ 1 (U, O) →
Ȟ 1 (W, O) has finite-dimensional image.
Proof. Consider the sheaf E of smooth functions on X. It is a fact that H 1 (B, E) = 0 for
any disk B ⊆ C – in technical terms, E is a flabby sheaf. Given a cocycle (fij ) ∈ Z 1 (B, O),
we can resolve it using smooth functions: fij = gj − gi for gi , gj ∈ C 0 (B, E). Now these may
not be analytic, but this won’t be a problem. Consider the Cauchy-Riemann operator
∂ 1 ∂f ∂f
: f 7→ +i .
∂ z̄ 2 ∂x ∂y
∂
By the Cauchy-Riemann equations, ∂ z̄
is 0 on analytic functions, so in particular
∂fij ∂gj ∂gi
0= = − .
∂ z̄ ∂ z̄ ∂ z̄
Now the functions ∂g
i
∂ z̄
agree on any Ui ∩ Uj , so we get a lift h ∈ Ž 1 (B, E). By Dolbeault’s
Lemma from complex analysis, there is some g ∈ E(B) such that ∂g ∂ z̄
= h. By the above,
∂(gj −g
∂ z̄
= 0 for all j. Writing fij = (gj − g) − (gi − g), we now see that (fij ) is a coboundary.
Hence Ȟ 1 (B, O) = 0.
To extend this to covers, suppose W is compactly contained in U. Take U ∈ U and
W ∈ W with W ⊆ U . Then for any f ∈ O(U ), f may be unbounded but by compactness,
f |W is bounded. Let || · || be the L2 -norm, so that
ZZ
||f || = |f (x + iy)|2 dx dy.
U
660
30.4. The Analytic Existence Theorem Chapter 30. Riemann’s Existence Theorem
Lemma 30.4.3. If W 0 is compactly contained in U 0 ⊆ C, then for every ε > 0 there exists
a closed subspace A ⊆ L2 (U, OC ) of finite codimension such that for all f ∈ A,
We next construct a norm on the Čech cochain groups of X. First, each Ui ∈ U is biholo-
morphic to some disk Di ⊆ C via a map zi . For g ∈ O(Ui ), set ||g||L2 (Ui ) := ||zi (g)||L2 (Di ) .
For a 0-cocycle η = (fi ) ∈ Č 0 (U, O), where fi ∈ O(Ui ), define
n
!1/2
X
||η|| := ||fi ||2L2 (Ui ) .
i=1
Continue in this fashion, we get a norm for every element of Č q (U, O), q ≥ 0. Write
ČLq 2 (U, O) for the subspace of Č q (U, O) where this norm is finite. Now for the compactly con-
tained covers W ⊆ U and for any ε > 0, Lemma 30.4.3 provides a closed, finite-codimension
subspace A ⊆ ŽL1 2 (U, O) such that for all ξ ∈ A, ||ξ||L2 (W) ≤ ε||ξ||L2 (U ) .
On the other hand, suppose we have a containment of covers W ⊆ V ⊆ U, with compact
containments Wi ⊆ Vi ⊆ Ui for each i.
Lemma 30.4.4. Given compactly contained covers W ⊆ V ⊆ U, there exists c > 0 such that
for every cocycle ξ ∈ ŽL1 2 (V, O), there is some ζ ∈ ŽL1 2 (U, O) such that
ζ = ξ + d(0) η
661
30.4. The Analytic Existence Theorem Chapter 30. Riemann’s Existence Theorem
Ȟ 1 (U, O) Ȟ 1 (U, O)
By Leray’s theorem (30.4.1), the vertical arrows are isomorphisms but by Proposition 30.4.2,
the restriction map along the bottom has finite dimensional image. This implies the identity
has finite dimensional image, i.e. dim H 1 (X, O) < ∞.
As a preliminary step to proving the analytic existence theorem, we have:
Theorem 30.4.6. Let X be a compact Riemann surface. Then for any a0 ∈ X, there exists
a meromorphic function f ∈ M(X) that has a pole at a0 and is holomorphic elsewhere.
Proof. Fix a0 ∈ X and choose an open cover {Up }p∈X such that p ∈ Up and a0 6∈ Up for
any p 6= a0 . Since X is compact, there is a finite subcover X = U0 ∪ U1 ∪ · · · ∪ Ur with
z
a0 ∈ U0 = Ua0 . Set U = {Ui }ri=0 . We know the map U0 →
− B(0, 1) is biholomorphic.
(m)
Now for each m ∈ N, we define a cocycle ξ (m) = (fij ) ∈ Ž 1 (U, O) by
(m)
fij = 0 when i, j 6= 0
(m)
f00 = 0
(m) (m)
f0j = −fj0 = z −m .
662
30.4. The Analytic Existence Theorem Chapter 30. Riemann’s Existence Theorem
where η = (gi ) ∈ Č 0 (U, O) with gi ∈ O(Ui ) and d(0) is the 0th boundary operator on the
Čech complex. Define f : X r {a0 } → C by
(
gi (z), z ∈ Ui and i ≥ 1
f (z) = Pn −m
g0 (z) + m=1 cm z , z ∈ U0 .
By construction, f (z) will be meromorphic so we need only check that the definition is
consistent on overlapping elements of the cover U. Suppose i, j 6= 0 and take z ∈ Ui ∩ Uj .
(m)
Then by definition the ijth part of the left side of (∗) is 0 since (ξ (m) )ij = fij = 0, whereas
on
Pnthe right it’s gi − gj (or the reverse). Hence gi = gj on Ui ∩ Uj . When j = 0, we have
−m
c
m=1 m z = gi − g0 on Ui ∩ U0 so again the definition of f (z) is consistent.
We now prove Theorem 30.2.7.
Proof. Suppose we can find meromorphic functions h1 , . . . , hn ∈ M(X) such that hi (pj ) = δij
for each 1 ≤ i, j ≤ n. Then the function g(z) = c1 h1 (z)+. . .+cn hn (z) will satisfy the desired
property. To produce these hi , first use Theorem 30.4.6 to find meromorphic functions
f1 , . . . , fn ∈ M(X) such that for each i, fi has a pole at pi and is holomorphic everywhere
else. Choose dij ∈ C such that fi (z) − fi (pi ) + dij is nonzero whenever we evaluate at z = pj .
Then define
Y fi (z) − fi (pi )
hi (z) = .
j6=i
fi (z) − fi (pi ) + dij
By the choice of dij , these functions are well-defined, meromorphic and clearly hi (pi ) = 1
and hi (pj ) = 0 for all j 6= i. Therefore the theorem is proved.
663
Chapter 31
In this chapter we define the étale fundamental group of a scheme. This will generalize the
construction for curves in Section 29.3 and allow us to fully translate the topological theory
of covering spaces to an algebraic setting. We first recall the so-called Galois theory for
covering spaces.
Let p : Y → X be a topological cover and let Aut(Y /X) be the group of automorphisms
of the cover, i.e. the homeomorphisms Y → Y making the diagram
Y Y
commute. If p is a finite, degree n cover, then # Aut(Y /X) ≤ n. When # Aut(Y /X) = n,
we call p a regular, or Galois cover. In this case, we write Gal(Y /X) = Aut(Y /X).
Example 31.0.2. Let X = Y = C r {0}, the punctured complex plane, with complex
parameters x and y, respectively. Then an example of a Galois cover of degree n is the nth
power map f : Y → X, y 7→ y n . For example, when n = 2 the real coordinates of ths cover
take the form of a parabola covering the line in all but one point:
664
Chapter 31. Fundamental Groups of Schemes
665
31.1. Galois Theory for Schemes Chapter 31. Fundamental Groups of Schemes
where Q1 , . . . , Qr are prime ideals of B. Then YP is a finite étale algebra if and only if ei = 1,
i.e. the prime P is unramified in the language of algebraic number theory. In the general
case, a finite morphism of normal schemes of dimension 1 is étale if and only if each affine
piece corresponds to a finite, unramified extension of Dedekind rings.
666
31.1. Galois Theory for Schemes Chapter 31. Fundamental Groups of Schemes
We next translate the ‘local triviality’ condition from covering space theory to schemes.
Proof. ( =⇒ ) The assumption that X is connected means every fibre of ϕ has the same car-
dinality, say n. If n = 1, the property is trivial, so we may induct on n.`By Theorem 15.5.2,
the diagonal Y → Y ×X Y induces a decomposition Y ×X Y = ∆(Y ) Y 0 for Y 0 an open
and closed subscheme of Y ×X Y . The maps Y 0 ,→ Y ×X Y (obvious) and Y ×X Y → Y
(Proposition 15.6.4(a)) are finite étale. Thus Proposition 15.6.4(b) implies their composition
Y 0 → Y is finite étale as well. But by construction the fibres of Y 0 → Y have cardinality
n − 1 so by induction, there exists a finite, locally free, surjective morphism ψ 0 : Z → Y
such that Y 0 ×Y Z consists of n − 1 disjoint copies of Z. Then finally the composition
ψ = ϕ ◦ ψ 0 : Z → X is finite, locally free and surjective (it is the composition of two such
maps) and (Y ×X Y ) ×Y Z ∼ = Y ×X Z is the disjoint union of n copies of Z.
( ⇒= ) Since ψ is locally free, each P ∈ X has an open neighborhood U = Spec A such
that ψ −1 (U ) = Spec C for some finitely generated free A-algebra C. Let ϕ−1 (U ) ⊆ Y be
ϕ−1 (U ) = Spec B for B an A-module. Then the base change of ϕ to ψ over ϕ−1 (U ) is given
by Spec(B ⊗A C) → Spec B. By linear algebra, B ⊗A C is a finitely generated, free C-module
so it is also finitely generated and free as an A-module. Hence B must be finitely generated
and free as an A-module, too.
We next show ϕ is étale using Proposition 31.1.2. Let z̄ : Spec Ω → Z be a geometric
point of Z; then x̄ = ψ ◦ z̄ : Spec Ω → X is a geometric point of X. There is a natural
isomorphism of fibres
Yx̄ = Spec Ω ×X Y ∼
= Spec Ω ×Z (Y ×X Z) = (Y ×X Z)z̄
but the hypothesis implies (Y ×X Z)z̄ is a disjoint union of copies of Spec Ω. Thus Yx̄ is a
disjoint union of points, and since ψ is surjective, every geometric fibre of ϕ is as well. Hence
Proposition 31.1.2 implies ϕ is étale.
Recall from Section 14.3 that the codimension of a point P ∈ X is defined as the di-
mension of the local ring OX,P (or the height of the prime associated to P in any affine
neighborhood). The powerful Zariski-Nagata purity theorem says that to show a map is
étale, it suffices to check the étale property at all codimension 1 points of the base scheme.
FétX −→ FétU
Y 7−→ Y ×X U.
667
31.1. Galois Theory for Schemes Chapter 31. Fundamental Groups of Schemes
ϕ ψ
X
commutes. Then
(1) Y → Z is a Galois cover and Z ∼
= Y /G for some subgroup G ≤ Aut(Y /X).
(2) There is a bijection
{subgroups G ≤ Aut(Y /X)} ←→ {intermediate covers Y → Z → X}.
(3) The correspondence is bijective on normal subgroups of Aut(Y /X) and Galois covers
Z → X, and in this case Aut(Z/X) ∼ = Aut(Y /X)/G as groups.
668
31.2. The Étale Fundamental Group Chapter 31. Fundamental Groups of Schemes
be the fibre functor over x̄. By Lemma 14.3.6(c), Fibx̄ is indeed a functor. Using the
definition of the automorphism group of a functor from Section 28.1, we can now define the
algebraic, or étale, fundamental group of a scheme.
By Theorem 28.1.3, this coincides with the topological fibre functor and fundamental
group. We will also see below that this definition for schemes coincides with the definition
given for curves in Section 29.3.
(1) π1 (X, x̄) is a profinite group and its action on Fibx̄ (Y ) is continuous for all Y ∈ FétX .
Example 31.2.2. Let k be a field and consider X = Spec k. Then finite étale covers
Y → Spec k are precisely Y = Spec A for A = L1 × · · · × Lr a finite étale k-algebra. Here,
the fibre functor over any x̄ : Spec k̄ → Spec k (equivalent to a choice of algebraic closure k̄
of k) is exhibited by
Moreover, there is an identification (via Prop. 14.2.2) Homk (A, ks ) = HomSpec k (Spec ks , Spec k).
669
31.2. The Étale Fundamental Group Chapter 31. Fundamental Groups of Schemes
The above shows that although Fibx̄ is not a representable functor – Spec ks is not a
finite étale k-scheme – it is pro-representable in FétSpec k . Explicitly,
where the direct limit is over all finite étale Galois covers Y 0 → Spec k ordered by the
existence of a Spec k-morphism Y 0 → Y 00 .
In fact, there was nothing special about X being Spec k or even affine in the last para-
graph.
Proposition 31.2.3. For a connected scheme X and any geometric point x̄ : Spec Ω → X,
the fibre functor Fibx̄ is pro-representable. Explicitly, for any finite étale cover Y → X,
where the direct limit is over all finite étale Galois covers Y 0 → Y .
where again Y 0 ranges over all finite Galois covers. Since Corollary 31.1.9 says that each
Aut(Y 0 /X) is a finite group, we see that π1 (X, x̄) = Aut(Fibx̄ ) is a profinite group. Note
that π1 (X, x̄) has a natural action on each Fibx̄ (Y ) for any finite étale cover Y → X.
It remains to show this action is continuous. Take a geometric point ȳ ∈ Fibx̄ (Y ) lying
over x̄ : Spec Ω → X. If ȳ comes from an element of HomX (Y 0 , Y ) by the direct limit in
Proposition 31.2.3, then the action of π1 (X, x̄) factors through Aut(Y 0 /X). This implies
continuity.
(2) Take a finite, continuous, left π1 (X, x̄)-set S. The action of π1 (X, x̄) is transitive on
each orbit of S, so we may assume it is transitive on S to begin with. Let H be the stabilizer
of any point in S. Then H is an open subgroup of π1 (X, x̄), so it contains the open normal
subgroup corresponding to the kernel of π1 (X, x̄) → Aut(Y 0 /X) for some finite Galois cover
Y 0 → X. Let H be the image of H in Aut(Y 0 /X). Then one proves S ∼ = Y 0 /H. This proves
essential surjectivity; fully faithfulness is routine.
The following shows that our definition of the fundamental group for schemes properly
captures the notion we described for curves in Section 29.3.
Theorem 31.2.5. Let X be an integral normal scheme with function field K and fix a
separable closure Ks /K. Let KX be the compositum of all finite subextensions L/K in Ks
such that the normalization XL of X in L is étale over X. Then
670
31.2. The Étale Fundamental Group Chapter 31. Fundamental Groups of Schemes
(1) KX /K is Galois.
f : Y −→ X
y 7−→ πy n .
(Here, we use π to ensure there are no algebraic relations on this map.) As in Example 31.1.3,
f is a finite étale Galois cover of degree n, with Galois group G = hσi ∼= Z/nZ, where σ is
the automorphism σ : y 7→ ζn y for ζn a primitive nth root of unity. Notice that by setting
z = π 1/n y, we may define f over Q. We make use of the following important result.
Theorem 31.2.7. If a scheme X/C is defined over Q, then any finite étale cover of X is
also defined over Q.
In particular, the finite cover f : y 7→ z n is defined over some number field K/Q. A
classic question asks for a description of K in terms of the topology of this cover.
Set GQ = Gal(Q/Q) and fix an automorphism ω ∈ GQ . Let G be any group and suppose
Y → X is a G-Galois cover which is defined over a number field K. Then Gal(Q/K) ≤ GQ
and if ω lies in Gal(Q/K), then ω carries this cover Y → X to itself (the action on Y being
the action of ω on the coefficients of the map). The field of moduli for the cover Y → X is
defined to be the subfield M of Q/Q fixed by all ω taking Y → X to itself.
Proposition 31.2.8. If G is an abelian group, then the unique maximal field of definition
for Y → X is precisely M.
Note that Proposition 31.2.8 is completely open when G is nonabelian. In any case, the
consequences for abelian covers is that we can see the Galois theory of finite étale covers in
two parallel ways:
(Geometric) Every Galois cover of schemes has a group of automorphisms and a cor-
responding field extension for which this is the Galois group.
(Arithmetic) Every cover of schemes has a field of definition over Q, and in the abelian
case there is even a unique minimal such field of definition.
As in Section 29.4, these related pieces are governed by a short exact sequence
(here, the base point can be omitted and k can be any field). Accordingly, π1 (X) is called
the geometric fundamental group of X, π1 (X) the algebraic fundamental group and Gal(k̄/k)
the arithmetic fundamental group. As in Corollary 29.4.4, the sequence splits if and only if
X contains a k-rational point.
671
31.3. Properties of the Fundamental Group Chapter 31. Fundamental Groups of Schemes
Proposition 31.3.1. For any two geometric points of X, x̄ : Spec Ω → X and x̄0 : Spec Ω0 →
∼
X, there is a natural isomorphism Fibx̄ −
→ Fibx̄0 .
Proof. We saw in Proposition 31.2.3 that each fibre functor is pro-representable by an inverse
system (Pα ) of finite étale Galois covers of X, but the choice of X-morphisms in each case
depends on the choices of basepoints in each Pα lying over x̄ or x̄0 . Let the covering morphisms
be denoted ϕαβ , ψαβ : Pβ → Pα for x̄, x̄0 , respectively. Then by Proposition 31.2.3, for any
Y ∈ FétX , Fibx̄ (Y ) ∼
= lim Hom(Pα , Y ) where the connecting homomorphisms in the direct
−→
limit are those induced by the ϕαβ ; and similarly Fibx̄0 (Y ) ∼
= lim Hom(Pα , Y ) with the maps
−→
induced by the ψαβ . Therefore is suffices to exhibit an isomorphism of inverse systems
(Pα , ϕαβ ) ∼
= (Pα , ψαβ ).
Let α ≤ β and take λβ ∈ AutX (Pβ ). Let xα ∈ Pα and xβ ∈ Pβ be the specified points
over x̄ and set x0α = ψαβ ◦ λβ (xβ ):
xβ λβ λβ (xβ )
Pβ Pβ
ϕαβ ψαβ
Pα Pα
xα λα x0α
Corollary 31.3.2. For any pair of geometric points x̄ : Spec Ω → X and x̄0 : Spec Ω0 → X,
there is a continuous isomorphism of profinite groups π1 (X, x̄0 ) → π1 (X, x̄).
∼
Proof. Any natural isomorphism λ : Fibx̄ − → Fibx̄0 from Proposition 31.3.1 determines an
isomorphism λ : π1 (X, x̄ ) → π1 (X, x̄) by λ∗ (ϕ) = λ−1 ◦ ϕ ◦ λ.
∗ 0
Note that the isomorphism λ∗ : π1 (X, x̄0 ) → π1 (X, x̄) depends on the choice of path λ,
but only up to inner automorphism.
672
31.3. Properties of the Fundamental Group Chapter 31. Fundamental Groups of Schemes
Now let X and X 0 be two connected schemes, take a geometric point in each, x̄ : Spec Ω →
X and x̄0 : Spec Ω → X 0 , and suppose there is a covering morphism ϕ : X 0 → X such that
ϕ ◦ x̄0 = x̄. Then ϕ induces a base change functor
Recall from the proof of Proposition 31.1.5 that there is a natural isomorphism Fibx̄ ∼
=
Fibx̄0 ◦Bx̄,x̄0 . This induces a map
(1) ϕ∗ is trivial if and only if for all connected Y ∈ FétX , the base change Y ×X X 0 → X 0
is a trivial cover.
(2) ϕ∗ is surjective if and only if for all connected Y ∈ FétX , the base change Y ×X X 0 →
X 0 is a connected cover.
Proof. (1) First note that an X 0 -cover is trivial if and only if π1 (X 0 , x̄0 ) acts trivially on its
fibre over x̄0 . Thus if, for any λ ∈ π1 (X 0 , x̄0 ) and Y ∈ FétX , λ acts trivially on Fibx̄0 (Y ×X
X 0) ∼
= Fibx̄ (Y ), then ϕ∗ (λ) must be trivial. Conversely, if some Y ×X X 0 is a nontrivial
cover, there exists λ ∈ π1 (X 0 , x̄0 ) acting nontrivially on Fibx̄ (Y ) ∼
= Fibx̄0 (Y ×X X 0 ). Hence
ϕ∗ (λ) is nontrivial.
(2) Suppose Y ∈ FétX (connected) is such that Y ×X X 0 is not connected. Then there
exist x1 , x2 ∈ Fibx̄ (Y ) such that ϕ∗ (λ)(x1 ) 6= x2 for any λ ∈ π1 (X 0 , x̄0 ). However, the fact
that Y is connected implies there exists µ ∈ π1 (X, x̄) such that µ(x1 ) = x2 . Thus µ is not
in the image of ϕ∗ , so ϕ∗ is not surjective. Going the other direction, if ϕ∗ is not surjective
then im ϕ∗ is a proper, closed subgroup of the profinite group π1 (X, x̄), so there exists a
proper, open subgroup U ⊆ π1 (X, x̄) such that U ⊇ im ϕ∗ . Thus π1 (X 0 , x̄0 ) acts trivially on
the coset space π1 (X, x̄)/U , or in other words, the connected cover Y → X corresponding
to π1 (X, x̄)/U from Theorem 31.1.12 base changes to a trivial cover Y ×X X 0 → X 0 . But if
U 6= π1 (X, x̄), then Y ×X X 0 6= X 0 and is therefore a disconnected cover.
Remark. Suppose Y → X is connected, ȳ is a geometric point over x̄ and U = Stabπ1 (X,x̄) (ȳ).
Then U is an open subgroup of π1 (X, x̄) and Fibx̄ (Y ) ∼
= π1 (X, x̄)/U as left π1 (X, x̄)-sets.
As above, let ϕ : X 0 → X be a cover of connected schemes such that ϕ ◦ x̄0 = x̄.
Proposition 31.3.5. Let U ⊆ π1 (X, x̄) be any open subgroup and let Y → X be the cor-
responding connected cover; let ȳ ∈ Fibx̄ (Y ) be the geometric point such that U = Stab(ȳ).
Then U ⊇ im ϕ∗ if and only if the base change Y ×X X 0 → X 0 has a section such that
y 0 := x̄0 (Spec Ω) ⊆ X 0 maps to y := ȳ(Y ×X Spec Ω) ⊂ Y ×X X 0 .
673
31.3. Properties of the Fundamental Group Chapter 31. Fundamental Groups of Schemes
Fibx̄ (Y0 ) ∼
= π1 (X 0 , x̄0 )/U 00 where U 00 = Stab(ȳ).
Now an X 0 -morphism Y0 → Y 0 exists if and only if there is an equivariant map π1 (X 0 , x̄0 )/U 00 →
π1 (X 0 , x̄0 )/U 0 . This in turn occurs if and only if U 00 ⊆ U 0 . If λ ∈ ker ϕ∗ , then λ fixes ȳ in
Fibx̄ (Y ) = Fibx̄0 (Y ×X X 0 ), hence λ fixes ȳ in Fibx̄0 (Y0 ). This means λ ∈ U 00 and so we see
that U 00 ⊇ ker ϕ∗ . By hypothesis, U 0 ⊇ ker ϕ∗ as well, so by the correspondence theorem,
U 00 ⊆ U 0 if and only if ϕ∗ (U 00 ) ⊆ ϕ∗ (U 0 ), but this holds by construction. Therefore an
X 0 -morphism Y0 → Y 0 exists.
( ⇒= ) We must show U 0 ⊇ ker ϕ∗ . Applying the fibre functor Fibx̄0 to the morphism
Y0 → Y 0 gives a map Fibx̄0 (Y0 ) → Fibx̄0 (Y 0 ). Note that ȳ 0 ∈ Fibx̄0 (Y 0 ); choose any lift
ȳ ∈ Fibx̄0 (Y0 ) and set U = Stab(ȳ). Identifying Y0 → Y 0 with a subgroup U 00 ⊆ π1 (X 0 , x̄0 ),
we get U 00 ⊆ U 0 by the above. Then once again, ker ϕ∗ has to fix ȳ, so ker ϕ∗ ⊆ U 00 ⊆ U 0 .
Corollary 31.3.7. ϕ∗ is injective if and only if for all connected covers Y 0 → X 0 , there
exists a cover Y → X and a morphism Y0 → Y 0 over X 0 such that Y0 is isomorphic to a
connected component of Y ×X X 0 .
Corollary 31.3.8. If every connected cover Y 0 → X 0 arises as a base change Y 0 = Y ×X X 0
of some Y → X, then ϕ∗ is injective.
ψ ϕ
Corollary 31.3.9. Let X 00 − → X0 −→ X be a pair of morphisms of connected schemes and
let x̄ : Spec Ω → X, x̄ : Spec Ω → X 0 and x̄00 : Spec Ω → X 00 be geometric points such that
0
674
31.3. Properties of the Fundamental Group Chapter 31. Fundamental Groups of Schemes
675
31.4. Structure Theorems Chapter 31. Fundamental Groups of Schemes
Theorem 31.4.1 (Bertini). Let k be an algebraically closed field and let X ⊆ Pnk be a smooth,
closed subscheme. Then there exists a hyperplane H ⊂ Pnk not containing X such that X ∩ H
is regular.
Lemma 31.4.2. Let k be algebraically closed and let X ⊆ Pnk be a smooth, connected, closed
subscheme of dimension dim X ≥ 2 and let Y → X be a connected, finite étale cover. Then
there exists a hyperplane H ⊆ Pnk not containing X so that X ∩ H is smooth and connected
and Y ×X (X ∩ H) is connected.
Proof. Let H be as in Bertini’s theorem. When dim X ≥ 2, one can use sheaf cohomology
(cf. Hartshorne, III.7.9.1) to show that X ∩ H is not just regular but also smooth and
connected. Then the same proof applies to the closed immersion Y ×X (X ∩ H) ,→ Y to
show that Y ×X (X ∩ H) is connected.
Theorem 31.4.3. For any smooth, projective, integral scheme X over C and every geometric
point x̄ : Spec C → X, π1 (X, x̄) is finitely generated.
Proof. Suppose dim X ≥ 2. Then by Lemma 31.4.2, for any connected Y → X there
exists a hyperplane H ⊆ PnC such that X ∩ H is smooth, connected and Y ×X (X ∩ H) is
connected. By Proposition 15.6.4(a), Y× (X ∩ H) → X ∩ H is a (connected) finite étale cover,
so Proposition 31.3.4 implies π1 (X ∩ H, x̄) → π1 (X, x̄) is surjective. Now note that X ∩ H
is smooth, projective and dim X ∩ H < dim X, so repeating this process, we eventually
obtain a surjection π1 (C, x̄) → π1 (X, x̄) for some smooth projective curve C over C. By
Theorem 29.3.5, we know π1 (C, x̄) is finitely generated so this implies π1 (X, x̄) is as well.
Even stronger than the surjections π1 (X ∩ H) → π1 (X) in the above proof, we have the
following result of Lefschetz.
Hk (X ∩ H) −→ Hk (X)
H k (X ∩ H) −→ H k (X)
πk (X ∩ H) −→ πk (X)
are all isomorphisms for k < dim X − 1 and are surjective for k = dim X − 1.
676
31.4. Structure Theorems Chapter 31. Fundamental Groups of Schemes
To compare π1 and π1top as we did in Chapter 30, we associate to any C-scheme X of finite
an
S
type a complex analytic space X as follows. Write X = Spec Ai for C-algebras Ai of finite
type. Then Ai ∼ = C[x1 , . . . , xn ]/(f1 , . . . , fr ) for polynomials f1 , . . . , fr ∈ C[x1 , . . . , xn ]. Then
each fj may be regarded as a holomorphic function on Cn , so the zero set of {f1 , . . . , fr }
in Cn is a ringed space (Yi , OYi ), where OYi is the ring of holomorphic functions on Yi (in
the sense of complex geometry). We regard this (Yi , OYi ) as the basic model S of a complex
analytic space. Now glue the Yi together using the same gluing as for Spec Ai to get a
ringed space (X an , OX
an
). Note that for any morphism of schemes Y → X, there is a natural
induced morphism of complex analytic spaces Y an → X an . The complex analytic space X an
has the following properties relative to X.
Proposition 31.4.5. Let X/C be a scheme of finite type. Then
(1) X is separated if and only if X an is Hausdorff.
(5) A morphism Y → X is proper (i.e. of finite type and the base change is closed, as
in Section 14.3) if and only if Y an → X an is proper (i.e. compact sets pull back to
compact sets).
Therefore, for any geometric point x̄ : Spec C → X with image x = x̄(Spec C), the induced
map
π1top\
(X an , x) −→ π1 (X, x̄)
is an isomorphism.
Just like in Chapter 30, the difficult part of this proof is establishing the essential sur-
jectivity of the functor Y 7→ Y an . This is essentially the main idea behind Serre’s GAGA
principle.
Question (Serre). Does there exist a scheme X over C of finite type such that π1 (X) = 1
but π1top (X an ) 6= 1?
677
31.4. Structure Theorems Chapter 31. Fundamental Groups of Schemes
id id
1 π1 (X) π1 (X) × π1 (Y ) π1 (Y ) 1
with the 1 on the left in the top row coming from the natural section X ,→ X ×k Y of
the projection morphism X ×k Y → X, and the entire bottom row representing the direct
product of profinite groups. Since the left and right vertical arrows are identities, the Five
Lemma implies the middle vertical arrow is an isomorphism.
Corollary 31.4.9. Let K ⊇ k be an extension of algebraically closed fields, X a proper
integral scheme over k and XK its base change. Then for any geometric points x̄ : Spec k →
X, x̄0 : Spec K → XK , the induced map π1 (XK , x̄0 ) → π1 (X, x̄) is an isomorphism.
Proof. We first show that π1 (XK , x̄0 ) → π1 (X, x̄) is surjective. Suppose Y → X is connected.
Then since k is algebraically closed, the K-algebra k(Y )⊗k K is a field, but since this is equal
to the function field of YK , we see that YK is connected. Therefore by Proposition 31.3.4,
the induced map on fundamental groups is surjective.
For injectivity, suppose Y → XK is a connected cover. It is possible (see Szamuely) to find
a subfield k 0 ⊆ K such that k 0 /k is finitely generated, as well as an integral affine k-scheme T
such that k(T ) = k 0 and there is a connected finite étale cover Y 0 → X ×k T . Fix a geometric
point t̄ : Spec k → T . By Corollary 31.4.8, there is an isomorphism π1 (X ×k T, (x̄, t̄)) ∼ =
π1 (X, x̄) × π1 (T, t̄), so Corollary 31.3.7 implies there exist connected covers Z → X and
T 0 → T and Z ×T T 0 → Y 0 a morphism of T -schemes. Take a k-point t ∈ T ; then the fibre
(Z ×T T 0 )t of the morphism Z ×T T 0 → T is a finite étale cover of X and there is a morphism
Z ×T T 0 → Y of schemes over XK . Thus by Corollary 31.3.7, π1 (XK , x̄0 ) → π1 (X, x̄) is
injective.
Corollary 31.4.10. Let k be an algebraically closed field of characteristic 0, X a smooth,
connected, projective scheme over k and x̄ : Spec k → X a geometric point. Then π1 (X, x̄)
is finitely generated.
678
31.4. Structure Theorems Chapter 31. Fundamental Groups of Schemes
Proof. We know the result for k = C by Theorem 31.4.3, so applying Corollary 31.4.9
to the extension C/Q shows that any Q-scheme has finitely generated fundamental group.
Finally, any algebraically closed field of characteristic 0 contains Q, so one more application
of Corollary 31.4.9 finishes the proof.
679
31.5. Results in Characteristic p Chapter 31. Fundamental Groups of Schemes
π1 (Yx , ȳ 0 ) −→ π1 (Y, ȳ 0 )
is an isomorphism.
(2) Further, if k is algebraically closed, ϕ is flat and Yη̄ and Yx̄ are geometrically reduced,
then for any geometric point ȳ ∈ Yη̄ , the natural map
is an isomorphism.
Lemma 31.5.2 (Abhyankar). Let (A, m, k) be a DVR with fraction field K and L/K and
M/K finite Galois extensions in which A has integral closures AL and AM , respectively, and
integral closure B in the compositum LM . Suppose pL ⊂ AL and pM ⊂ AM are maximal
ideals lying over m such that the inertia degrees e(pL | m) and e(pM | m) are relatively prime
to char k and e(pL | m) | e(pM | m). Then Spec B → Spec AM is a finite étale morphism of
affine schemes.
680
31.5. Results in Characteristic p Chapter 31. Fundamental Groups of Schemes
Proof. Set p = char k. Fix a maximal ideal P ⊂ B lying over m. We may assume P∩AL = pL
and P ∩ AM = pM . Then the injection Gal(LM/K) ,→ Gal(L/K) × Gal(M/K) restricts to
an injection of inertia groups IP ,→ IpL × IpM . Moreover, the induced projections IP → IpL
and IP → IpM are surjective. Since |Ip | = e(p | m) for any prime p lying over m, then
by hypothesis the inertia groups IP , IpL and IpM all have prime-to-p order and hence are
cyclic. Further, since |IpL | divides |IpM |, every element of IP ⊆ IpL × IpM has order dividing
e(pM | m), but since there is a surjection IP → IpM and the target has order e(pM | m), this
implies the map is really an isomorphism IP ∼ = IpM . By transitivity of inertia degrees, that
is e(P | m) = e(P | pM )e(pM | m), we get e(P | pM ) = 1, so P is unramified over M . Since
P was arbitrary, it follows from Proposition 29.2.6 that Spec B → Spec AM is étale.
Theorem 31.5.3 (Grothendieck). If k is algebraically closed of characteristic p > 0, ϕ :
Y → X = Spec A is smooth, proper and has geometrically connected fibres, then for any
geometric points ȳ ∈ Yη̄ and ȳ 0 ∈ Yx , the specialization map descends to an isomorphism of
prime-to-p parts
0 ∼ 0
π1 (Yη̄ , ȳ)(p ) −→ π1 (Yx̄ , ȳ 0 )(p ) .
0 0
Proof. By definition of sp, it’s enough to show π1 (Yη̄ , ȳ)(p ) → π1 (Y, ȳ)(p ) is an isomorphism
of prime-to-p profinite groups. To begin, suppose Z → Y is a connected finite étale cover
with base change Zη̄ = Z ×X Spec K over the generic point. Then
where the inverse limit is over all finite extensions L/K contained in a fixed algebraic closure
K corresponding to η̄. Then each Z ×X Spec L is an open subscheme of the base change
ZL = Z ×X XL , where XL = Spec AL and AL is the integral closure of A in L. We claim
ZL is connected. If ZL = Z1 ∪ Z2 for two closed subsets Z1 , Z2 , then the special fibre
Zx = Z ×X Spec k lies in one of them, say Z1 . Under the map ZL → XL , Zx is taken
to Yx , so Z2 must be open, contradicting the assumption that Y → X, and therefore also
ZL → XL , is proper and in particular closed. Thus ZL is connected, so Z ×X Spec L is as
well. Since L was arbitrary and the inverse limit of connected spaces is connected, we have
shown that Zη̄ is connected. Now (2) of Proposition 31.3.4 implies that π1 (Yη̄ , ȳ) → π1 (Y, ȳ)
is surjective and in particular it is surjective on prime-to-p parts.
For injectivity (on prime-to-p parts), Corollary 31.3.7 says that it’s enough to show that
every finite étale cover Z 0 → Yη̄ of prime-to-p degree is the base change of some cover Z → Y
of prime-to-p degree. In fact, the remark at the end of Section 31.3 implies that we can check
the condition for Galois covers Z 0 → Yη̄ . Applying Theorem 31.5.1, we have an isomorphism
π1 (Y ×X XL , z̄) ∼= π1 (Y, z̄) so replacing X with XL , we may assume Z 0 is obtained by base
change from a finite Galois cover Zη → Yη . Let Y be the normalization of Y in the function
field K(Zη ) of Zη . Then K(Zη )/K(Y ) is a finite Galois extension, say of degree d, and by the
Zariski-Nagata purity theorem (31.1.6), it’s enough to find a finite Galois extension K 0 /K
such that Y ×X XK 0 → Y ×X XK 0 is étale over all codimension 1 points. Further, since all
codimension 1 points of Y lie in Yη except for the generic point ξ of Yx , we need only check
that Y ×X XK 0 is étale over every point of Y ×X XK 0 lying over ξ. If A has maximal ideal
m = (π), then π also generates the √ maximal ideal of the local ring OY,ξ . It follows from
Kummer theory that K 0 = K(Y )( d π) is a finite Galois extension of K(Y ) of degree d and
681
31.5. Results in Characteristic p Chapter 31. Fundamental Groups of Schemes
since A is a DVR, ramification theory implies that there is a unique point ξ 0 ∈ YK 0 lying
above ξ, with inertia group I(ξ 0 | ξ) ∼= Z/dZ. Now Abhyankar’s lemma shows that YK(Zη )K 0
0 ∼
is étale over ξ . Moreover, YK(Zη )K 0 = Y ×X XK 0 by construction. Finally, this shows that
the base change of Z := Y ×X XK 0 → Y to Yη̄ is precisely Z 0 , completing the proof.
Now let X be any locally noetherian scheme and Y → X a proper morphism with
connected geometric fibres. Suppose x0 , x1 ∈ X such that x0 is a specialization of x1 , i.e.
x0 ∈ {x1 }. Let x̄0 , x̄1 be geometric points of X with images x0 , x1 , respectively, and fix
geometric points ȳ0 , ȳ1 ∈ Y lying over x̄0 , x̄1 , respectively. Set Z = {x1 }, consider the local
ring OZ,x0 and note that the geometric fibres Yx̄0 and Yx̄1 over X remain the same over
Spec OZ,x0 . Denote by A the completion of the localization at a height 1 prime ideal of
OZ,x0 . Then A is a complete DVR and we have a specialization map
for the generic point η̄ of Spec A. On the other hand, by Corollary 31.4.9, the morphism
Yη̄ → Yx̄0 induces an isomorphism
∼
π1 (Yη̄ , ȳ0 ) −→ π1 (Yx̄0 , y¯0 ).
π1 (Yx̄1 , ȳ1 ) ∼
= π1 (Yη̄ , ȳ0 ) −→ π1 (Yx̄0 , ȳ0 ).
Grothendieck used this specialization result to prove the following analogue of Theo-
rem 31.4.6.
Theorem 31.5.5. Let X be an integral, proper, normal curve of genus g over an algebraically
closed field k of characteristic p > 0. Then for any geometric point x̄ : Spec k → X,
682
31.5. Results in Characteristic p Chapter 31. Fundamental Groups of Schemes
Proof. Same as the proof in characteristic 0, using the same reduction to the case of curves
as in Theorem 31.4.3.
Example 31.5.7. This result is false in characteristic p when X is not assumed to proper.
For example, Raynaud proved that for X = A1k = P1k r {∞}, π1 (X, x̄) is infinitely generated.
(Compare this to Example 29.3.10.) Any Artin-Schreier equation y p − y = f (x), where
f (x) ∈ k[x] has prime-to-p degree, defines a nontrivial Galois cover Yf → P1k that is étale
over A1k and has Galois group Z/pZ. If k is algebraically closed, this gives an infinitely family
of finite Galois covers of A1k . More generally, π1 (X, x̄) is infinitely generated for any affine
curve X over k.
Definition. Let ϕ : Y → X be a finite separable morphism of normal integral schemes.
Then ϕ is tamely ramified at a codimension 1 point P ∈ X if ϕ is ramified at P
and for each Q ∈ ϕ−1 (P ), the ramification index e(Q | P ) of ϕ at Q is relatively prime to
the characteristic of κ(P ). We say ϕ is tamely ramified, or simply tame, if it is tamely
ramified at every codimension 1 point of X.
Let X be an integral normal scheme, let U ⊆ X be an open subscheme and denote by
FéttU
the full subcategory of FétU consisting of finite separable covers ϕ : Y → X which are
tamely ramified over X r U . Note that the fibre functor Fibx̄ restricts to a functor on FéttU .
Definition. For a geometric point x̄ : Spec Ω ,→ X with image x ∈ U , the tame funda-
mental group of U at x̄ is
π1t (U, x̄) := Aut(Fibx̄ |FéttU ).
Theorem 31.5.8. Let x̄ be a geometric point in X and U ⊆ X an open subscheme containing
x̄. Then
(1) π1t (U, x̄) is a quotient of π1 (U, x̄) and in particular a profinite group and its action on
Fibx̄ (Y ) is continuous for all Y ∈ FéttU .
(2) The restriction of Fibx̄ to FéttU induces an equivalence of categories
∼
FéttU −→ {finite, continuous, left π1t (U, x̄)-sets}
Y 7−→ Fibx̄ (Y ).
(3) Let KUt be the compositum inside a fixed algebraic closure of K = k(U ) of all finite
extensions L/K such that the normalization XL of X in L is étale over U and tamely
ramified over X r U .
(4) The tame fundamental group π1t (U, x̄) is finitely generated as a profinite group.
Statements (1) – (3) can be derived from similar arguments as for the full étale funda-
mental group. The proof of (4) relies on the deep fact that tame covers in characteristic
p lift to p-tame covers in characteristic 0, where a p-tame cover is one defined over a field
of characteristic 0 in which p does not divide the order of any inertia groups of the cover.
However, this more or less allows one to compute the tame fundamental group by lifting to
characteristic 0.
In contrast, the pro-p-part of the fundamental group in characteristic p is not completely
understood and is the subject of much current research. But we will end the discussion here.
683
Part VII
Étale Cohomology
684
Chapter 32
Introduction
Grothendieck topologies
Galois cohomology
Cohomology of curves
685
Chapter 33
Sites
This chapter covers the basic definitions and results in Grothendieck’s theory of sites, a
useful generalization of a topological space. The main motivation is to develop a working
sheaf theory on schemes that can detect the features of étale morphisms and more general
properties.
686
33.1. Grothendieck Topologies and Sites Chapter 33. Sites
When X is a scheme with the Zariski topology, Top(X) is called the (small) Zariski site on
X.
Example 33.1.2. The category Top of all topological spaces with continuous maps between
them is a site, called the big topological site, whose coverings are defined by
( )
[
Cov(X) = {fi : Xi ,→ X} : fi is an open embedding and X = Xi .
i
Example 33.1.3. Similarly, for a scheme X, let SchX be the category of X-schemes (the
category Sch of all schemes can be viewed in this framework by setting X = Spec Z since
this is a terminal object in Sch). Then SchX is a site, called the big Zariski site on X, with
coverings
( )
[
Cov(Y ) = {ϕi : Yi → Y } : ϕi is an open embedding and Y = Yi .
i
687
33.1. Grothendieck Topologies and Sites Chapter 33. Sites
Example 33.1.4. Let C be a site and X ∈ C be an object. Define the localized site (or
the slice category) C/X to be the category with objects Y → X ∈ HomC (Y, X), morphisms
Y → Z in C such that
Y Z
688
33.1. Grothendieck Topologies and Sites Chapter 33. Sites
689
33.2. Sheaves on Sites Chapter 33. Sites
` We say F is separated if for every collection of maps {Xi → X}, the map
Definition.
F (X) → i F (Xi ) is injective.
Definition. Let C be a site. A sheaf on C is a presheaf F : Cop → Set such that for every
object X ∈ C and every covering {Xi → X} ∈ Cov(X), the sequence of based sets
Y Y
F (X) F (Xi ) F (Xi ×X Xj )
i i,j
where, for a, b ∈ F (X), a ∼ b if there is a covering {Xi → X} of X such that a and b have
the same image under the map a
F (X) → F (Xi ).
i
By construction, F sep
is a separated presheaf on C and for any other separated presheaf F 0 ,
any morphism of presheaves F → F 0 factors through F sep uniquely. Hence F 7→ F sep is left
adjoint to the forgetful functor SepC → PreShC so it remains to construct a sheafification of
every separated presheaf on C.
For a separated presheaf F , define F a to be the presheaf
690
33.2. Sheaves on Sites Chapter 33. Sites
and ({Xi → X}, {αi }) ∼ ({Yj → Y }, {βj }) if αi and βj have the same image in F (Xi ×X Yj )
for all i, j. Then as above, F a is a sheaf which is universal with respect to all morphisms of
sheaves F → F 0 . Thus F 7→ F a defines a left adjoint to the forgetful functor ShC → SepC
and composition with the first construction proves the theorem.
Definition. A topos is a category T which is equivalent to ShC for some site C. The plural
of ‘topos’ is ‘topoi’.
Remark. Grothendieck held the view that the topos of a site is more important than the site
itself. In practice, many naturally occurring sites induce the same topos (that is, their sheaf
theories are equivalent). For example, the Zariski sites on Sch and AffSch (the categories
of schemes and affine schemes, respectively) induce the same topos. In general, AffSch is
easier to study since it is equivalent to the opposite category of commutative rings.
Definition. Let T be a topos and A ∈ T an object. Then A is called a group object (or
a group in T ) if it possesses distinguished morphisms
m : A × A → A, e : ∗ → A, i:A→A
1A × m
A×A×A A×A
m × 1A m
A×A m A
commutes.
(i, 1A )
A A×A
∗ A
e
691
33.2. Sheaves on Sites Chapter 33. Sites
commutes.
A×A
m
p A
m
A×A
commutes, where p flips the factors of A × A. In this case, we will often write the map m
as a to denote addition.
m : A × A → A and 1:∗→A
satisfying the usual axioms of associativity, identity and left and right distributivity. Further,
if m satisfies the axiom of commutativity, A will be called a commutative ring in T .
Remark. To make this more compatible with the language of categorical homotopy theory,
one might instead call a ring object a monoid or monoid object in T , meaning with respect
to the symmetric monoidal structure on T induced by ×.
Proof. Let T = SchC , T 0 = SchC0 and define the morphism f∗ : T → T 0 for each object
X 0 ∈ C0 and sheaf F ∈ T by
(f∗ F )(X 0 ) = F (f (X 0 )).
If {Xi0 → X 0 } is a covering in C0 , we have a commutative diagram
Y Y
(f∗ F )(X )0 (f ∗ F )(X 0
i ) (f∗ F )(Xi0 ×X 0 Xj0 )
i i,j
= = ∼
=
Y Y
0
F (f (X )) F (f (Xi0 )) F (f (Xi0 ) ×f (X 0 ) f (Xj0 ))
i i,j
692
33.2. Sheaves on Sites Chapter 33. Sites
Here, the bottom row is exact since F is a sheaf and f is continuous; the right vertical
arrow is an isomorphism since f is continuous; and the other vertical arrows are equalities
by definition. Hence by the Five Lemma, the top row is exact, so f∗ F defines a sheaf on C0 .
To define the left adjoint f ∗ : ShC0 → ShC , take an object X ∈ C and define a category
IX with objects (X 0 , ρ) where X 0 ∈ C0 and ρ ∈ HomC (X, f (X 0 )), and with morphisms
(X 0 , ρ) → (Y 0 , σ) given by a morphism g : X 0 → Y 0 in C0 making the following diagram
commute:
f (X 0 )
ρ
X f (g)
σ
f (Y 0 )
(f ∗ F )(X) = lim F (X 0 )
−→
op
where the limit is over all objects (X 0 , ρ) in the opposite category IX . If h : X → Y is a
∗ ∗
morphism in C, then there is an induced map (f F )(Y ) → (f F )(X) given by the functor
IY −→ IX
(Y 0 , ρ) 7−→ (Y 0 , ρ ◦ h).
One can show that it is an isomorphism, which establishes that (f ∗ , f∗ ) is an adjoint pair.
Hence we have a morphism of topoi T → T 0 .
693
33.3. The Étale Site Chapter 33. Sites
Y × G −→ Y ×X Y
(y, g) 7−→ (y, yg)
is an isomorphism.
Lemma 33.3.1. ϕ : Y → X is a G-cover if and only if ϕ is surjective, finite, étale and
deg ϕ = |G|.
Definition. A Galois cover ϕ : Y → X is said to be generically Galois if k(Y )/k(X) is
a Galois extension of fields.
Example 33.3.2. Let A be a ring, B an A-algebra and consider the corresponding morphism
of affine schemes
ϕ : Spec B −→ Spec A.
Then ϕ is a Galois cover with Galois group G if and only if A → B is faithfully flat and G
acts on B such that
Y
B ⊗A A −→ B × G = B
g∈G
0 0
b ⊗ b 7−→ (bgb )g
is an isomorphism.
Example 33.3.3. Let k be a field and f ∈ k[t] a monic irreducible polynomial. Set K =
k[t]/(f ). Then f = f1e1 · · · frer in K[t] and by the Chinese remainder theorem,
r
K ⊗k K ∼
= K[t]/(f ) ∼
Y
= K[t]/(fiei ).
i=1
Thus Spec K → Spec k is a Galois cover if and only if each fi is linear, fi 6= fj for any i 6= j
and ei = 1 for all i. That is, Spec K → Spec k is Galois if and only if f is separable with
splitting field K, just as in classical Galois theory.
Proposition 33.3.4. Suppose ϕ : Y → X is a G-cover and F is a presheaf on the étale
site Xét taking disjoint unions to products. Then F is a sheaf on Xét if and only if F (ϕ) :
F (X) → F (Y ) is an isomorphism onto the fixed set F (Y )G ⊆ F (Y ).
Proof. Consider the two maps Y × G Y given by (y, g) 7→ y and (y, g) 7→ yg. These
fit into a commutative diagram with the two coordinate projections Y ×X Y Y :
694
33.3. The Étale Site Chapter 33. Sites
Y ×G Y X
∼
= id id
Y ×X Y Y X
where the maps F (Y ) are given by s 7→ (s)g and s 7→ (gs)g . Then these
Q
g∈G F (Y )
maps agree precisely when gs = s for all g ∈ G, i.e. F (X) is the equalizer in the top row if
and only if it identifies with F (Y )G in the bottom row.
Proposition 33.3.5. Suppose F is a presheaf on Xét which satisfies the condition that
Y Y
F (U ) F (Ui ) F (Ui ×U Uj )
i i,j
is an equalizer diagram for all covers {Ui → U } of U ∈ XZar in the Zariski site on X and
for all étale affine covers {V → U } of U ∈ Xét consisting of a single morphism in the étale
site. Then F is a sheaf on Xét .
`
Proof. If U = i Ui for schemes Ui ∈ Xét , then the first condition implies that F (U ) =
a covering {Ui → U 0 } in Xét , the sequence
Q
i F (Ui ). Thus for
Y Y
0
F (U ) F (U i ) F (Ui ×U 0 Uj )
i i,j
for the covering {U → U 0 }, using the fact that ( i Ui ) ×U 0 ( i Ui ) = i,j (Ui ×U 0 Uj ). Since
` ` `
the equalizer condition is assumed to hold for all étale affine covers, this argument shows the
condition holds for all {Ui → U }i∈I with I finite `and each Ui affine.
Let {Uj → U }j be a covering and set U 0 = j Uj and f : U 0 → U . Write U = i Vi
S
for open affine subschemes Vi ⊆ U and for each i, write f −1 (Vi ) = k Wik for open affine
S
subschemes Wik ⊆ U 0 . Fix one of the Vi . Then each f (Wik ) is open in Vi , so by quasi-
compactness, we may reduce to a finite cover {Wik → Vi }K k=1 . Now consider the diagram
695
33.3. The Étale Site Chapter 33. Sites
F (U ) F (U 0 ) F (U 0 ×U U 0 )
Y YY YY
F (Vi ) F (Wik ) F (Wik ×U Wi` )
i i k i k,`
Y YY
F (Vi ×U Vj ) F (Wik ×Vi Wj` )
i,j i,j k,`
The two columns correspond to the coverings {Vi → U }i and {Wik → U 0 }i,k which are
all coverings in the Zariski site on X and hence these columns are exact by hypothesis.
Moreover, the middle row corresponds to the coverings {Wik → f (Wik ) ⊆ Vi }k which for
each i is finite and affine, so by the above paragraph this row is exact. An easy diagram
chase then implies the top row of the diagram is also exact, which is what we want.
0 → A → B → B ⊗A B (*)
is exact, where the second map is b 7→ 1 ⊗ b − b ⊗ 1. First note that the map g : B →
B ⊗A B, b 7→ b ⊗ 1 has a section s : b ⊗ b0 7→ bb0 . Then consider
h : (B ⊗A B) ⊗B (B ⊗A B) −→ B ⊗A B, x ⊗ y 7−→ xgs(y).
0 A B B ⊗A B
0 B B ⊗A B (B ⊗A B) ⊗B (B ⊗A B)
696
33.3. The Étale Site Chapter 33. Sites
0 → A → B → B ⊗A B
(Note that this is precisely the same as the equalizer condition since the map B → B ⊗A B
is b 7→ 1 ⊗ b − b ⊗ 1 and Γ(Y, OY ) are abelian groups.)
Example 33.3.8. Any scheme Z → X defines a presheaf
FZ : Xét −→ Set
Y 7−→ HomX (Y, Z).
Then FZ is a sheaf for XZar so once again, to show it is a sheaf for the étale topology, by
Proposition 33.3.5 it will suffice to check the sheaf condition for single étale affine covers.
For such a cover {Spec B → Spec A}, the map A → B is faithfully flat with
0 → A → B → B ⊗A B
exact. Take an open affine subscheme U = Spec C ,→ Z. Applying HomA (C, −) to the
above sequence gives
which is exact since HomA (C, −) is a left-exact functor. Generalizing to HomX (−, Z) is
straightforward using patching, so ultimately we conclude that FZ is a sheaf on Xét . As
usual, by Yoneda’s Lemma the assignment Z 7→ FZ is injective so we will often write FZ
simply by Z.
Example 33.3.9. Let n ≥ 1 be an integer and let µn be the group scheme defined (locally)
by tn − 1 = 0. Then for any Y ∈ Xét , µn (Y ) coincides with the set of nth roots of unity in
Γ(Y, OY ).
Example 33.3.10. Let X be a k-scheme and let Ga be the affine group scheme defined by
the additive group of k. Then for each Y ∈ Xét , Ga (Y ) = Γ(Y, OY ).
Example 33.3.11. Similarly, when Gm is the affine multiplicative group scheme defined by
k × , then for each Y ∈ Xét , Gm (Y ) = Γ(Y, OY )× .
Example 33.3.12. When char k = p > 0, let αp denote the group scheme defined by tp = 0.
Note that αp is not an étale group scheme (but it is flat). For Y ∈ Xét , αp (Y ) corresponds
to the set of nilpotent elements in Γ(Y, OY ).
697
33.3. The Étale Site Chapter 33. Sites
Example 33.3.13. Consider the ring k[ε] = k[t]/(t2 ). Write T = Spec k[ε]. The functor
T = Homk (−, T ) is called the (étale) tangent space functor since for any Y ∈ Xét , T (Y ) is
the tangent space to Y , which is locally given by T (Y )x = Tx Y := (mx /m2x ).
Example 33.3.14. Let R be a set and let FR be the sheaf on Xét defined by
Y
FR : Y 7−→ FR (Y ) := R
π0 (Y )
where π0 (Y ) denotes the set of connected components of Y . Then FR is called the constant
sheaf on Xét associated to R.
Example 33.3.15. Let M be a sheaf of coherent OX -modules on the Zariski site XZar .
This gives us an étale sheaf Mét as follows. If ϕ : Y → X is an étale morphism, then ϕ∗ M
is a coherent OY -module on YZar which on affine patches U = Spec A ⊆ X, V = Spec B ⊆ Y
takes the form
M ⊗A B M
B A
Let Mét be the presheaf Y 7→ Γ(Y, ϕ∗ M) on Xét . By a similar proof to the one in Exam-
ple 33.3.7 for OXét , one can show that Mét is then a sheaf on Xét . As a special case, note
that (OXZar )ét = OXét .
Example 33.3.17. Let k be a field, X = Spec k and consider the étale site Xét . If G-Modd
denotes the category of discrete G-modules, then there is an equivalence of categories
Sh(Xét ) ←→ G-Modd
F 7−→ MF
FM →−7 M
where MF = lim F (L) is the direct limit over all finite, Galois extensions L/k and FM is the
−→
sheaf A 7→ HomG (Homk (A, k sep ), M ).
We next describe the category Sh(Xét ). To begin, first consider the category Presh(Xét )
of presheaves on the étale site.
698
33.3. The Étale Site Chapter 33. Sites
(c) For every geometric point x̄ : Spec k̄ → X, the map on stalks Tx̄ : Fx̄ → Fx̄0 is
surjective.
Proposition 33.3.20. For a sequence of sheaves 0 → F 0 → F → F 00 → 0 in Sh(Xét ), the
following are equivalent:
(a) The sequence is exact.
(c) For all geometric points x̄ : Spec k̄ → X, the sequence of stalks 0 → Fx̄0 → Fx̄ → Fx̄00 →
0 is exact.
Corollary 33.3.21. For any scheme X, Sh(Xét ) is an abelian category.
Example 33.3.22. Let n ∈ N and assume n is invertible on X, i.e. char k - n for any residue
fields k of X. Consider the following sequence of sheaves, called the Kummer sequence for
X:
n
0 → µn →
− Gm → − Gm → 0
n
where Gm → − Gm is the morphism induced by t 7→ tn . We claim the Kummer sequence is
exact. By Proposition 33.3.20, it suffices to show exactness on stalks at geometric points.
For such a point x̄, set A = OX,x̄ . Then µn,x̄ = µn (A) and Gm,x̄ = A× and moreover,
0 → µn (A) → A× is clearly exact. For the right map, notice that tn − a splits in A[t] for
every a ∈ A× , since dtd (tn − a) = ntn−1 6= 0 when n is invertible on X. Thus every a is an
nth root in A× , and it follows that the sequence
n
− A× →
0 → µn (A) → − A× → 0
is exact as required.
699
33.3. The Étale Site Chapter 33. Sites
Example 33.3.23. When char k | n for some residue field k of X, the above example fails
since étale locally, we have
d p
(t − a) = ptp−1 = 0 in characteristic p > 0.
dt
Note however that the equation t 7→ tp does define a (locally) flat covering Gm → Gm , so
the sequence
p
0 → µp →− Gm →− Gm → 0
is exact in the flat topology on X. On the étale site, the appropriate characteristic p
replacement for the Kummer sequence is the Artin-Schreier sequence
℘
0 → Z/pZ →
− Ga →
− Ga → 0
where Z/pZ is the constant group scheme defined by the same group and ℘ : Ga → Ga is
induced by the map t 7→ tp − t. To see that the AS sequence is exact, it suffices once again to
check exactness on stalks at geometric points. For a geometric point x̄, again let A = OX,x̄
so that Z/pZx̄ = Z/pZ(A) and Ga,x̄ = A. As before, 0 → Z/pZ(A) → A is exact and the
kernel of t 7→ tp − t : A → A is precisely Z/pZ(A). For surjectivity, note that for any a ∈ A,
d p
(t − t − a) = ptp−1 − 1 = −1 6= 0
dt
so tp − t − a splits in A[t] for all a ∈ A. Thus the end of the sequence
℘
0 → Z/pZ(A) →
− A→
− A→0
700
Chapter 34
Cohomology
The étale site on a scheme has a sheaf cohomology theory which mimics the ordinary (sin-
gular) cohomology for schemes X/C with their natural complex structure. To define coho-
mology, we must first show that Sh(Xét ) has enough injectives. Along the way, we develop
a few useful algebraic constructions on sheaf categories that will play a role in our study of
cohomology and beyond. We will assume all sheaves have values in abelian groups.
701
34.1. Direct and Inverse Image Functors Chapter 34. Cohomology
where the sum is over all points z ∈ Y mapping to x and [k(z) : k(x)]s denotes the
separable degree of the residue field extension at z. In particular, if f is finite étale of
degree n, then (f∗ F )x̄ = Fȳ⊕n .
Corollary 34.1.4. If f : Y → X is finite or a closed immersion, then f∗ : Sh(Yét ) → Sh(Xét )
is an exact functor.
g f
Proposition 34.1.5. Suppose Z →
− Y →
− X are morphisms of schemes. Then (f ◦ g)∗ =
f∗ ◦ g∗ .
The direct image functor on presheaves is exact, so f∗ : Sh(Yét ) → Sh(Xét ) is left exact –
in general, it is not right exact. Thus we can define:
Definition. Let f : Y → X be a morphism with pushfoward functor f∗ : Sh(Yét ) → Sh(Xét ).
The inverse image functor (or pullback) along f is the left adjoint f ∗ : Sh(Xét ) → Sh(Yét )
to f∗ .
702
34.1. Direct and Inverse Image Functors Chapter 34. Cohomology
Lemma 34.1.6. For a morphism f : Y → X and any sheaf F on Xét , f ∗ F is the sheafifi-
cation of the presheaf f −1 F on Yét defined by
f −1 F (V ) := lim F (U )
−→
where V → Y is étale and the direct limit is over all U → X étale such that
V Y
U X
commutes.
Proof. It is easy to show that f −1 is left adjoint to f∗ on presheaves. Explicitly, for any
presheaf Q on Yét , there are bijections
HomPresh(Yét ) (f −1 F, Q) ←→ lim HomPresh(Xét ) (F (U ), Q(U ×X Y )) ←→ HomPresh(Xét ) (F, f∗ Q)
−→
703
34.1. Direct and Inverse Image Functors Chapter 34. Cohomology
A slight annoyance in Proposition 34.1.3 is that the direct image of a sheaf along an open
immersion is not necessarily supported on the image of the immersion. Indeed, consider the
following example.
Example 34.1.11. Let X = A1 be the affine line over an algebraically closed field k and
consider the open immersion j : A1 r {0} ,→ A1 . Recall (Example 29.3.11) that G :=
π1ét (A1 r {0}, ȳ) ∼
=Zb for any geometric point ȳ. Let S be a module over this fundamental
group, which determines a locally constant sheaf F = FS . Then (j∗ F )0 = S G , the submodule
of G-invariants of S. This shows that in (1) of Proposition 34.1.3, (f∗ F )x need not be 0.
However, we can define a version of the direct image which plays nicely with open im-
mersions.
Definition. Let X be a scheme and j : Y ,→ X an open immersion. For a sheaf F on Yét ,
define the extension along j of F to be the sheafification of the presheaf P on Xét which
sends an étale scheme U → X to
(
F (U ), if im(U ) ⊆ Y
P (U ) =
0, otherwise.
This defines a functor j! : Sh(Yét ) → Sh(Xét ).
Proposition 34.1.12. For any open immersion j : Y ,→ X, j! is exact and j ∗ preserves
injectives.
Proof. We first observe that (j! , j ∗ ) is an adjoint pair: for all F ∈ Sh(Yét ) and G ∈ Sh(Xét ),
HomSh(Xét ) (j! F, G) ∼
= HomPresh(Xét ) (P, G) where P is as above
∼
= HomSh(Yét ) (F, G|Yét )
= HomSh(Yét ) (F, j ∗ G) by Example 34.1.8.
Now take a geometric point x̄ : Spec k → X with image x and note that
(
Fx , if x ∈ Y
(j! F )x =
0, if x 6∈ Y
by definition of j! F . Then the first statement follows from Proposition 33.3.20 and the second
follows from a similar categorical statement as in the proof of Proposition 34.1.9.
Proposition 34.1.13. Let j : Y ,→ X be an open immersion, Z = X r j(Y ) and i : Z ,→ X
the corresponding closed immersion. Then for any sheaf F on Xét , the sequence
0 → j! j ∗ F → F → i∗ i∗ F → 0
is exact.
Proof. By Proposition 33.3.20, it suffices to check this on stalks, where for x ∈ X we have
( id
0 → Fx − → Fx →− 0 → 0, x ∈ Y
id
0→0→
− Fx −
→ Fx → 0, x ∈ Z.
704
34.2. Étale Cohomology Chapter 34. Cohomology
Note the following properties which are immediate from the definition/the theory of
derived functors:
For an étale morphism f : Y → X, the right derived functors of the composite functor
f∗ Γ(Yét ,−) i
Γ(Y, −) : Sh(Xét ) −→ Sh(Yét ) −−−−−→ Ab are Hét (Y, (−)|Y ), which we will simply write
i
Hét (Y, −) when the context is clear.
•
We now verify that the Eilenberg-Steenrod axioms hold for Hét (X, −), showing it is a
•
cohomology theory. Note that additivity is guaranteed since Hét is constructed as a right
derived functor.
705
34.2. Étale Cohomology Chapter 34. Cohomology
Theorem 34.2.3 (Exactness). Let Z ⊆ X be a closed subscheme and F any sheaf on Xét .
Then there is a long exact sequence
i−1
· · · → Hét (X r Z, F ) → HZi (X, F ) → Hét
i i
(X, F ) → Hét (X r Z, F ) → · · ·
where HZi (X, −) denotes the ith right derived functor of F 7→ ΓZ (X, F ) := ker(Γ(Xét , F ) →
Γ((X r Z)ét , F )). Moreover, the sequence is functorial in the pair (X, X r Z) and in F .
Proof. Let i : Z ,→ X and j : X r Z ,→ X be the (closed and open, resp.) immersions. For
the constant sheaf Z on X, Proposition 34.1.13 gives us a short exact sequence
0 → j! j ∗ Z → Z → i∗ i∗ Z → 0.
(2) π(Y r W ) ⊆ X r Z.
706
34.2. Étale Cohomology Chapter 34. Cohomology
Y rW Y W
π ∼
=
X rZ X Z
By Proposition 34.1.9 and Example 34.1.8, π ∗ is exact and preserves injectives, so it’s enough
to prove the isomorphism for i = 0, that is, to show α is an isomorphism.
Take s ∈ ΓZ (X, F ) and suppose α(s) = 0. Then s|XrZ = 0 by exactness of the bottom
row and α(s) = 0 in Γ(Y, π ∗ F ), but since {Y → X, X r Z → X} is an étale covering of X,
the sheaf axioms imply s = 0. Therefore α is injective.
On the other hand, if t ∈ ΓW (Y, π ∗ F ), we may view it as an element of Γ(Y, π ∗ F ). Let
t0 ∈ Γ(Y, π ∗ F ) be the zero section. Note that (X r Z) ×X Y = π −1 (X r Z) ⊆ Y r W and
we have t|Y rW = 0 by exactness of the top row and t0 |Y rW = 0 because it’s the zero section.
Thus the sheaf condition on π ∗ F says that both t and t0 come from a section s ∈ Γ(X, F )
which must satisfy s|XrZ = 0 by commutativity. Hence by exactness of the bottom row,
s ∈ ΓZ (X, F ) and α(s) = t, proving surjectivity.
Corollary 34.2.5. For any closed point x ∈ X and any sheaf F on Xét , there is an isomor-
phism
i ∼ i h
H{x} (X, F ) −→ H{x} (Spec OX,x ,F)
h
for all i ≥ 0, where OX,x is the henselization of OX,x .
Proof. Let π : U → X be an étale neighborhood of X with u = π −1 (x). Taking Z = {x}
and W = {u}, Theorem 34.2.4 gives us isomorphisms
i ∼ i
H{x} (X, F ) −→ H{u} (U, F |U )
h
for all i ≥ 0. Since OX,x = lim Γ(U, OU ) where the limit is over all étale neighborhoods of x,
−→
and since cohomology commutes with direct limits, we get
i
H{x} h
(Spec OX,x ,F) ∼ i
= lim H{u} (U, F |U ) ∼ i
= H{x} (X, F )
−→
as desired.
Finally, to prove the homotopy axiom, we need a suitable analogue of “homotopic maps”
in the category of schemes. There are various concepts of homotopy that can be introduced
in algebraic geometry, some giving quite different theories, but we will focus on one that
gives an analogue of the homotopy axiom for étale cohomology.
707
34.2. Étale Cohomology Chapter 34. Cohomology
Let k be a field (perhaps algebraically closed), X a scheme over k and suppose Z ⊆ X ×P1
is a closed subscheme whose image under the projection π : X × P1 → P1 is dense. For each
closed point t ∈ P1 , set Z(t) = Z ×X (X × {t}), so that each Z(t) may be viewed as an
algebraic cycle on X. If Z and Z 0 are algebraic cycles on X, we say that they are rationally
equivalent, denoted Z ∼ Z 0 , if there are finitely many such families of algebraic cycles
Z1 (t), Z2 (t), . . . , Zr (t) interpolating Z1 and Z2 in the following sense: for each 1 ≤ i ≤ r − 1,
there are closed points ti , t0i ∈ P1 such that Zi (ti ) = Zi+1 (t0i ), and in addition there are
t, t0 ∈ P1 such that Z = Z1 (t) and Z 0 = Zr (t0 ). We say two morphisms f, g : Y → X are
rationally equivalent if their graphs Γf ∼ Γg as algebraic cycles in X ×k Y .
708
Part VIII
709
Chapter 35
Introduction
Part VIII introduces algebraic stacks (following Olsson’s Algebraic Spaces and Stacks) and
is taken from part of my dissertation research in 2017 – 2020 at the University of Virginia.
The following topics are considered:
Grothendieck topologies
Algebraic spaces
Algebraic stacks
Deligne-Mumford stacks
Moduli spaces
Stacky curves.
For the necessary background in algebraic geometry, the reader can consult Hartshorne’s
Algebraic Geometry.
710
Chapter 36
One of the main motivations for Grothendieck’s use of fibred categories in the study of
algebraic spaces and stacks is to allow for the construction of universal objects. Here’s an
example to keep in mind. Suppose f : X → Y is a morphism in the category of topological
spaces (this problem will also arise in a category of schemes). Then for any sheaf F on Y , one
way to define a pullback sheaf f ∗ F on X is as a solution to the universal mapping problem
F → f∗ G of sheaves on Y , where G is a sheaf on X. This object f ∗ F is not unique, it is only
defined up to canonical isomorphism. Similar problems occur where universal objects are
present (any direct limit construction might pose a problem) so we must find a way around.
Here’s a brief discussion of how nontrivial automorphisms can get in the way of solving
a “moduli problem”, which is loosely defined as a classification problem with some natural
notion of geometry attached to the collection of all objects to be classified. A more rigorous
definition is that a moduli problem is a functor F : Schop → Set. If F is representable, say
F ∼= Hom(−, M ) for some scheme M , we say M is a (fine) moduli space for F . In general
no such space exists.
Example 36.0.1. There are many interesting moduli problems in classical geometry which
aren’t represented by schemes. For example, the problem of classifying all circles in R2 is
represented by R2 × R>0 , while circles up to isometry are parametrized by R>0 , neither of
which is a scheme.
Example 36.0.2. In differential geometry, all complex (or equivalently, hyperbolic) struc-
tures on a surface of genus g ≥ 1 are specified by Fenchel-Nielsen coordinates, and these
form a moduli space homeomorphic to R6g−6 .
Example 36.0.3. One can think of algebraic varieties themselves (over an algebraically
closed field) as moduli spaces whose points correspond to solutions to a given set of polyno-
mial equations.
Example 36.0.4. The classical Cayley-Salmon theorem says that the moduli problem of
finding lines on a smooth cubic surface is represented by a 0-dimensional scheme with 27
components.
Example 36.0.5. Let k be a field. The moduli problem of finding r-dimensional vector
subbundles of a rank n bundle on X ∈ Schk is represented by the Grassmannian variety
711
Chapter 36. Categories Fibred in Groupoids
Gr(r, n). In particular, Pnk = Gr(1, n) classifies line bundles inside rank n bundles. The
moduli problem of rank n bundles themselves has the infinite Grassmannian Gr(n, ∞) as a
moduli space.
Example 36.0.6. More generally, there is a scheme BG which is a moduli space for iso-
morphism classes of principal G-bundles.
Example 36.0.7. Let M1,1 : Schop C → Set be the moduli problem parametrizing complex
elliptic curves, i.e. M1,1 (S) is the set of isomorphism classes of smooth curves E → S whose
geometric fibres are all complex curves of genus 1 with a marked point. For a morphism
S → T , we get a functor M1,1 (T ) → M1,1 (S) defined by pullback: for a family of elliptic
curves E → T , f ∗ (E → T ) is the family E 0 → S which is the pullback in the diagram
E0 E
f
S T
Suppose M1,1 were representable, say M1,1 ∼= Hom(−, M ) for some scheme M . Let E0 ∈
M1,1 (M ) correspond to the identity idM ∈ Hom(M, M ). Then every elliptic curve E → S
would be the pullback
E E0
S M
Hom(−, S)
commute. In fact, M1,1 is none other than the j-line, A1j = Spec C[j].
712
Chapter 36. Categories Fibred in Groupoids
Example 36.0.8. Consider the moduli problem parametrizing Riemann surfaces of genus
g ≥ 2, or smooth proper complex curves X over C with X(C) ∼ = Cg /Λ for a full lattice Λ.
Let Mg : SchC → Set be the functor sending S to the set of isomorphism classes of smooth
op
schemes X → S whose geometric fibres are all Riemann surfaces of genus g. Such an X
is often called a family of Riemann surfaces, parametrized by the base S. As above, for a
morphism f : S → T we get a pullback functor f ∗ : Mg (T ) → Mg (S). By a similar proof to
the elliptic curve case, Mg is not a representable functor – that is, there is no (fine) moduli
space of genus g Riemann surfaces. However, as above, there is a moduli space Mg for Mg
which is of considerable complexity and has inspired much research in algebraic geometry.
Example 36.0.9. Let G be a group. Then G determines a category – usually also denoted
by G – in which there is only one object ∗ and for each g ∈ G, there is an automorphism
g
∗→− ∗. Thus G is a groupoid.
In Examples 36.0.7 and 36.0.8, the pullback functor f ∗ was crucial for studying the moduli
problems of curves of genus g. In order to allow for pullbacks to play a role in our study of
moduli spaces, we will replace a functor Schop → Set with a category fibred in groupoids. In
order to make this precise, we need to introduce fibred categories.
713
36.1. Fibred Categories Chapter 36. Categories Fibred in Groupoids
h π(ϕ)
π(y) π(x) π(x0 )
π(ψ)
714
36.1. Fibred Categories Chapter 36. Categories Fibred in Groupoids
Example 36.1.2. If X ∈ C is any object, then the slice category C/X is naturally a
category fibred in groupoids over C, where the projection C/X → C is just (Y → X) 7→ Y .
A category F fibred in groupoids over C is called representable if it is equivalent (as a category
fibred in groupoids) to a slice category C/X for some object X, in which case F is said to
be represented by X.
Example 36.1.3. Let SchX be the category of X-schemes and define a category π : F →
SchX whose objects are pullback squares
P Y
T X
ϕ∗ x ψ∗x x
ξx (α)
α ψ
Y Z X
715
36.1. Fibred Categories Chapter 36. Categories Fibred in Groupoids
f
x x0
is an equivalence of categories.
Corollary 36.1.7. For any category C, there is a fully faithful embedding of 2-categories
where CFG(C) is the 2-category of categories fibred in groupoids over C. In a similar way,
there is a fully faithful embedding of 2-categories
Example 36.1.8. Let Schk be the category of schemes over a field k. Then for each k, n ≥ 1,
we define a category fibred in groupoids Gr(k, n) → Schk called the (k, n)th Grassmannian
category, by putting Gr(k, n)(X) as the category of vector bundles E → X of rank k, together
with embeddings of bundles E ,→ AnX . A morphism in Gr(k, n) from E ∈ Gr(k, n)(X) to
E 0 ∈ Gr(k, n)(X 0 ) is a morphism of bundles E → E 0 covering a morphism of schemes X → X 0
and commuting with the natural map AnX → AnX 0 .
Example 36.1.9. For each X ∈ Schk , let M1,1 (X) be the category consisting of pairs (E, O)
where E → X is a smooth curve, O : X → E is a section and for every geometric point
x̄ : Spec k̄ ,→ X, the pullback (Ex̄ , Ox̄ ) is an elliptic curve over k̄. Then M1,1 → Schk is a
fibred category. Morphisms (E, O) → (E 0 , O0 ) in M1,1 are given by pullback diagrams
716
36.1. Fibred Categories Chapter 36. Categories Fibred in Groupoids
F
E E0
f
X X0
Example 36.1.10. Likewise for each g = 0, g ≥ 2 and X ∈ Schk , define Mg (X) to be the
category consisting of smooth curves C → X whose geometric fibres Cx̄ are smooth curves
of genus g over k̄. Then Mg is a fibred category over Schk . As above, morphisms in Mg are
given by pullbacks, so Mg is a category fibred in groupoids.
Example 36.1.11. The following construction will be important in the definition of stacks.
Let π : F → C be a category fibred in groupoids. For X ∈ C and x, x0 ∈ F(X), define a
presheaf Isom(x, x0 ) : (C/X)op → Set by
where ϕ∗ x, ϕ∗ x0 ∈ F(Y ) are pullbacks of x and x0 , respectively, along ϕ. Moreover, any other
ψ : Z → Y induces a map
717
36.2. Fibre Products Chapter 36. Categories Fibred in Groupoids
ϕ(α) ϕ0 (β)
ϕ0
π1 (X 0 ) π2 (Y 0 )
p1 π2
π1
G1 G
q1 π2
π1
G1 G
' '
π2 (λ2 )
π 2 ◦ q2 π 2 ◦ p2 ◦ t
718
36.2. Fibre Products Chapter 36. Categories Fibred in Groupoids
commute.
This says that G1 ×G G2 is a 2-categorical fibre product of G1 and G2 . Now let C be any
category and π : F → C, π1 : F1 → F and π2 : F2 → F categories fibred in groupoids over
C. The construction of a 2-categorical fibre product F1 ×F F2 in the 2-category CFG(C) of
categories fibred in groupoids over C is similar to the construction above, but we give it here
for completeness.
is an isomorphism.
π1 ◦ λ1
π1 ◦ q 1 π1 ◦ p 1 ◦ t
τ σ◦t
π2 ◦ λ2
π2 ◦ q 2 π2 ◦ p 2 ◦ t
commute.
Proof. (Sketch; see Olsson 3.4.13) Define G = F1 ×F F2 to be the category with objects
(x, y, ϕ) consisting of x ∈ F1 , y ∈ F2 and an isomorphism ϕ : π1 (x) → π2 (y) in F. A
morphism (x, y, ϕ) → (x0 , y 0 , ϕ0 ) in G is a pair of morphisms α : x → x0 in F1 and β : y → y 0
in F2 making the diagram
ϕ
π1 (x) π2 (y)
ϕ(α) ϕ0 (β)
0
ϕ0
π1 (x ) π2 (y 0 )
719
36.2. Fibre Products Chapter 36. Categories Fibred in Groupoids
it is clear that η goes to (ξ1 , ξ2 , γ). On the other hand, if η ∈ HomC (H, G) then η is precisely
the image under the above morphism of the triple (p1 ◦ η, p2 ◦ η, σ ◦ η). Therefore
is an isomorphism of groupoids.
(ii) Now suppose qi : G 0 → Fi , i = 1, 2, are morphisms and τ : π1 ◦ q1 ' π2 ◦ q2 a natural
isomorphism such that
720
Chapter 37
Descent
The notion of descent can be phrased in a quite general context. Fix an object S in a
category C and recall that the localized or slice category at S is the category C/S whose
objects are arrows X → S in C and whose morphisms are commutative triangles
X Y
Given any other object S 0 ∈ C and a morphism S 0 → S, there is a base change functor
C/S → C/S 0 given by (X → S) 7→ (X 0 → S 0 ) where X 0 is the fibre product
X 0 = X ×S S 0 X
S0 S
This answers the question “when does an object over S determine an object over S 0 ?” The
opposite question, namely when an object over S 0 determines an object over S, is much
harder to answer in general. In fact there are two interesting questions one might ask: given
a morphism S 0 → S and an arrow (X 0 → S 0 ) ∈ C/S 0 , (1) is there an arrow (X → S) ∈ C/S
completing the diagram
X0 X
S0 S
and if so, (2) how many ways are there to complete the diagram? The process of answering
both of these questions is known as descent. In the first three sections of this chapter, we
outline some of the basics of descent theory in the algebraic geometry of schemes, following
721
Chapter 37. Descent
722
37.1. Galois Descent Chapter 37. Descent
Example 37.1.1. obj(k) could be the class of k-vector spaces, or more specifically, the class
of Lie algebras over k, central simple algebras over k, etc. Notice that many examples like
this admit morphisms, and so form a category over k.
Example 37.1.2. A quadratic space over k is a k-vector space V together with a quadratic
form q, that is, a symmetric bilinear function q : V → k. The set of quadratic spaces
(V, q) may be an interesting class of objects over k. These too admit morphisms, where
(V, q) → (V 0 , q 0 ) consists of a k-linear isomorphism V → V 0 which commutes with q and q 0 .
Example 37.1.3. Important choices of obj(k) in algebraic geometry include the class of
algebraic varieties or algebraic groups over k, the class of schemes over Spec k, and more
generally things like algebraic spaces and algebraic stacks over k.
Many of these examples admit a base change functor as in the chapter introduction, i.e.
an assignment
obj(k) −→ obj(K)
X 7−→ XK .
Most of the time this functor can be built using the tensor product or fibre product in the
right category.
Example 37.1.4. For some of the examples above, we have the following base change
functors:
object over k base change to K
vector space V VK = V ⊗k K
central simple algebra A AK = A ⊗k K
quadratic space (V, q) (VK , qK ) where qK extends q
variety/scheme X XK = X ×k K (defined by ⊗ on affine patches)
algebraic group G GK = G ×k K
In the pattern of the introduction, there are two natural questions we would like to answer
about descending objects defined over K to objects over k:
(1) Given an object A ∈ obj(K), does there exist an object X ∈ obj(k) such that XK = A?
723
37.1. Galois Descent Chapter 37. Descent
√
Example 37.1.6. In other situations, descent is √ impossible. Let k = Q, K = Q( 2) and
2 2 2
consider the quadratic form qK (x, y) = x − y 2 on V = K . Is there a quadratic form
qQ on VQ = Q2 which extends to qK on VQ ⊗Q K = V ? The answer in this case is no – we
will see soon that there are some elementary conditions that are necessary for descent to be
possible, and they are not satisfied in this situation.
Example 37.1.7. (Motivation) The Inverse Galois Problem asks: for which finite groups
G can one construct a field extension `/Q with Gal(`/Q) ∼ = G? By Hilbert’s Irreducibility
Theorem, it is enough to construct an extension L/Q(t) with Gal(L/Q(t)) ∼
= G. Over C, one
can construct Galois extensions L/C(t) by instead constructing branched covers of Riemann
surfaces X → P1C . It is a general fact that every such cover X is defined over Q, so the
question of constructing a G-extension of Q(t) comes down to being able to descend the
cover X → P1Q to a cover XQ → P1Q . A general solution to this would solve the Inverse
Galois Problem.
Let K/k be a Galois extension with Galois group G = Gal(K/k) and suppose Vk and Wk
are objects over k (e.g. k-algebras) that are isomorphic over K, that is, VK = Vk ×k K ∼ =
Wk ×k K = WK . Moreover, assume that under the natural Galois action on VK and WK ,
we have Vk = VKG and Wk = WKG (this is true e.g. for k-algebras). When there is a notion of
maps between objects over k (e.g. k-linear algebra homomorphisms), the G-action extends
to MapK (VK , WK ) by (σf )(v) = σ(f (σ −1 v)) for all v ∈ VK , σ ∈ G.
f
For example, let f be an equivalence VK →− WK , e.g. a K-algebra isomorphism. Set
ξ(σ) = f −1 ◦ σf ∈ GK := AutK (VK ). Notice that if ξ(σ) = 1 for all σ ∈ G, then σf = f σ
∼
so f descends to an isomorphism fk : Vk = VKG −
→ WKG = Wk . That is, ξ = 1 is a necessary
condition for two (K/k)-forms of VK to be isomorphic over k. In general, the automorphism
ξ(σ) = f −1 · σf satisfies
Such a map ξ : G → GK is called a 1-cocycle, and the set of all 1-cocycle is denoted Z 1 (G, GK ).
If K/k is an infinite extension, we are guaranteed to have f (Vk ) ⊆ Wk ⊗k ` ⊆ Wk ⊗k K = WK
for some finite extension `/k. If σ ∈ H := Aut(K/`), then for all v ∈ Vk ,
since f (v) ∈ Wk ⊗k `. Thus σf = f , so ξ(σ) = 1 for all σ ∈ H. Said another way, the
1-cocycle ξ is constant on cosets of H, which means ξ is in fact a continuous 1-cocycle on
G = lim Aut(`/k)
−→
over all finite extensions `/k, when this direct limit is equipped with the Krull topology
induced by discrete topologies on each Aut(`/k).
∼
Turning back to our K-isomorphism f : VK − → WK , the definition of ξ may depend on
∼
this f , but suppose f 0 were a different isomorphism VK −→ WK . Then f 0 = f ◦ g for some
g ∈ GK and the cocycle ξ 0 defined by f 0 has the form
724
37.1. Galois Descent Chapter 37. Descent
Fix an object VK over K, set GK = AutK (VK ) and let F (K/k, VK ) be the set of k-
isomorphism classes of (K/k)-forms of Vk . Then our work above shows that there is a map
Θ : F (K/k, VK ) −→ H 1 (G, GK ).
Then defining σ ∗v = ξ(σ)σ(v), i.e. twisting by the cocycle ξ, gives a new k-semilinear action
of G on Vk ⊗k K. Note that for any σ, τ ∈ G,
725
37.1. Galois Descent Chapter 37. Descent
It will be useful to define group cohomology and Galois cohomology in the nonabelian
case, so we recall the definitions here. Suppose G acts continuously on a group A (which
may be nonabelian) with the discrete topology – remember that this means the stabilizers
are open in G. Then a 1-cocycle of the action is a function ξ : G → A that is continuous and
satisfies
ξ(στ ) = ξ(σ)σξ(τ ) for all σ, τ ∈ G.
Let Z 1 (G, A) denote the set of all 1-cocycles ξ : G → A. We say two cocycles ξ, ξ 0 ∈ Z 1 (G, A)
are equivalent if there exists an a ∈ A such that
Let H 1 (G, A) be the set of equivalence classes of 1-cocycles. Then H 1 (G, A) is a pointed set
with distinguished element [1], where 1 is the trivial cocycle.
Suppose f : A → B is a morphism of G-groups, that is, a group homomorphism which
commutes with the G-action. Then there is a map of pointed sets
Set H 0 (G, A) = AG .
Here, H 2 (G, A) is the ordinary 2nd group cohomology of G with coefficients in an abelian
G-group A. Suppose that α ∈ Z 1 (G, A). Define the twist of A by α to be the G-group Aα
whose underlying group is A but with G-action defined by
for any g ∈ G and a ∈ A. (Here, x · y denotes the inner action of A on itself.) Let AG∗ denote
the fixed points of this action.
∼
Lemma 37.1.11. The map ξ 0 7→ ξ := ξ 0 · α yields a bijection tα : Z 1 (G, Aα ) −
→ Z 1 (G, A)
which descends to a bijection of pointed sets
∼
τα : H 1 (G, Aα ) −→ H 1 (G, A).
α β
Let 1 → A −
→B→
− C → 1 be a short exact sequence of G-groups and let
f
1 → H 0 (G, A) → H 0 (G, B) → H 0 (G, C) → H 1 (G, A) →
− H 1 (G, B)
726
37.1. Galois Descent Chapter 37. Descent
be the corresponding long exact sequence. For ξ ∈ H 1 (G, A), there is a twisted sequence
1 → Aξ → Bα(ξ) → Cξ → 1
fξ
[1] H 1 (G, Aξ ) H 1 (G, Bα(ξ) )
τξ τα(ξ)
f
ξ H 1 (G, A) H 1 (G, B)
whose columns are bijections by Lemma 37.1.11. This shows that the fibre of f : H 1 (G, A) →
H 1 (G, B) over f (ξ), namely f −1 (f (ξ)), may be identified with the fibre of fξ over [1], which
G
is ker fξ . Explicitly, ker fξ = Bα(ξ) /CξG which can be computed in many cases.
Example 37.1.12. Let K = k sep be the separable closure of k and consider the matrix
algebra Ak = M2 (k) ∈ Algk . Suppose Bk is a (K/k)-form of AK = M2 (K), with k-linear
∼
isomorphism f : M2 (K) − → BK . Then Bk is a central simple k-algebra of dimension 4 over
k. Assuming char k 6= 2, it is well-known that any such Bk is a quaternion algebra given by
a basis {1, i, j, k} satisfying the relations i2 = a, j 2 = b, k = ij, ij = −ji
for some a, b ∈ k.
Denote the quaternion algebra with this presentation by Bk = a,b k
. For Ak = M2 (k),
×
we have G K = AutK (M2 (K)) = GL2 (K)/K = P GL2 (K). Given a quaternion algebra
Bk = a,bk
, set L = k(i) and define an embedding
Bk ,−→ M2 (L)
x0 + x1 i b(x2 + x3 i)
x0 + x1 i + x2 j + x3 k 7−→ .
x2 − x3 i x 0 − x1 i
∼
This extends to an isomorphism h : BK = Bk ⊗k K − → M2 (K) = AK ; call its inverse f . For
x = x0 + x1 i + x2 j + x3 k ∈ Bk and σ ∈ Gal(L/k), we have
727
37.1. Galois Descent Chapter 37. Descent
In fact, (
1, σ|L = id
ξ(σ) =
g, σ|L = 6 id.
Then the extended term in the long exact sequence from Proposition 37.1.10 is
˜
where δ(σ, τ ) = ξ(σ)σ ˜ )ξ(στ
ξ(τ ˜ )−1 for σ, τ ∈ G and ξ˜ is any lift of ξ to GL2 (K). Notice
that H 2 (G, K × ) = Br(k) is the Brauer group of k, whose elements are equivalence classes
of central simple k-algebras. For the cocycle ξ coming from the quaternion algebra a,b
k
as
above, (
b, σ|L , τ |L 6= id
δ(σ, τ ) =
1, otherwise.
This identifies each element of F (K/k, M2 (k)) with a (Brauer class of a) central simple k-
algebra of degree 2. In general, the (K/k)-forms of Mn (k) may be identified with the Brauer
classes of central simple algebras of degree n.
Conversely, start with a quadratic extension L/k and a 1-cocycle ξ ∈ H 1 (G, P GL2 (K)).
One can also construct a central simple algebra corresponding to ξ by taking BL = M2 (L)
as an algebra and defining a new Galois action on BL according to σ ∗ a = ξ(σ)σ(a) for all
x y
σ ∈ G and a ∈ M2 (L). In particular, if M = then
z w
0 b σ(x) σ(y) 0 1 σ(w) bσ(z)
ξ(σ)σ(M ) = = −1
1 0 σ(z) σ(w) b−1 0 b σ(y) σ(x)
for some b ∈ k. Thus we can identify Bk = BLG∗ as the matrices M ∈ M2 (L) for which
σ ∗ M = M , i.e.
σ(w) bσ(z) x y
= .
b−1 σ(y) σ(x) z w
That is,
x y
Bk = x, y ∈ k .
b−1 σ(y) σ(x)
Example 37.1.13. Assume char k 6= 2 and consider the collection of quadratic spaces
(V, q) over k, with morphisms f : V1 ,→ V2 such that q2 |V1 = q1 . Fix a quadratic space
(V, q). Then GK = Aut(VK , qK ) = O(K, 2)q , the 2 × 2 orthogonal group of K preserving
the orientation induced by q. The set F (K/k, (VK , qK )) in this case classifies all quadratic
forms of a given rank (up to isomorphism). By Lemma 37.1.8, there is an injective map
728
37.1. Galois Descent Chapter 37. Descent
Θ : F (K/k, (VK , qK )) ,→ H 1 (G, O(K, 2)q ). Descent theory shows when this is a bijection.
Let q = x2 + y 2 be the quadratic form on V = and q 0 = ax2 + by 2 the quadratic form on V 0 .
Define
f : VK −→ VK0
x y
(x, y) 7−→ √ , √ .
a b
The corresponding 1-cocycle ξ ∈ H 1 (G, O(K, 2)q ) is
√ !
a
√
σ( a)
0
ξ(σ) = f −1 ◦ σ(f ) = √ .
0 √b
σ( b)
Conversely, for a cocycle ξ one can construct a quadratic space (V 0 , q 0 ) with V 0 = VKG∗ and
q 0 = qK |V 0 by defining the action σ ∗ v = ξ(σ)σ(v) for all σ ∈ G, v ∈ VK . Explicitly, for
v = (x, y), √ !
a
√ σ(x) 0
σ( a)
σ ∗ (x, y) = √ .
0 √b σ(y)
σ( b)
The fixed points under this action are (x, y) ∈ VK such that σ a = a and σ √yb = √yb ,
√x √x
√ √
so V 0 has k-basis {( a, 0), (0, b)}. The quadratic form in this case is precisely q 0 = ax2 +by 2 .
Thus all Galois twists of the quadratic form q = x2 + y 2 are of the form q 0 = ax2 + by 2 for
some a, b ∈ k.
Finally, consider the short exact sequence of G-groups
det
1 → SO(K, 2)q → O(K, 2)q −→ Z/2Z → 1.
For the quadratic form ax2 +by 2 over k, let ξa,b be the corresponding cocycle in H 1 (G, O(K, 2)q ).
By Proposition 37.1.10, there is a map
Further, by Kummer theory (see below), H 1 (G, Z/2Z) ∼ = k × /(k × )2 , the set of nonzero ele-
ments of k modulo squares in k. Under this correspondence, ξa,b is identified with the coset √
ab(k × )2 . Similar analysis
√ justifies the assertion in Example 37.1.6 that q K = x 2
− y 2
2
defined over K = Q( 2) does not descend to a quadratic form on Q.
Let K/k be a Galois extension with Galois group G = Gal(K/k) and suppose G is a
linear algebraic group over k on which G acts continuously. Then
where the limit is over all finite index open subgroups U ⊆ G and K U is the subfield of K
fixed by U.
729
37.1. Galois Descent Chapter 37. Descent
Theorem 37.1.14 (Hilbert’s Theorem 90). For a Galois extension K/k with group G,
(2) H 1 (G, K × ) = 1.
where the last term is 0 by Hilbert’s Theorem 90. Thus H 1 (G, Fp ) ∼ = k/℘(k). On the
1 ∼
other hand, H (G, Fp ) = Hom(G, Fp ) which classifies cyclic extensions of degree p over k
(up to k-isomorphism). This demonstrates the Artin-Schreier correspondence between cyclic
p-extensions of k and elements of k/℘(k); explicitly, every cyclic p-extension L/k is of the
form L = k[x]/(xp − x − a) for some a ∈ k/℘(k). The general case of cyclic extensions of a
characteristic p field of order divisible by p is given by Artin-Schreier-Witt theory.
by Hilbert’s Theorem 90. Thus H 1 (G, µn ) ∼ = k × /(k × )n . At the same time, if k contains the
nth roots of unity µn (k), then H 1 (G, µn ) ∼
= Hom(G, µn ) classifies cyclic n-extensions, and
every such an extension L/k is of the form L = k[x]/(xn − b) for b ∈ k × /(k × )n .
Example 37.1.17. The additive and multiplicative groups of a field k are often written
k = Ga and k × = Gm . Consider the semidirect product
a b
G = Ga o Gm = ∈ GL2 (k) .
0 1
730
37.1. Galois Descent Chapter 37. Descent
Lemma 37.1.18. Let W be a vector space over K and assume that G acts on W by semi-
linear transformations. Set W0 = W G . Then W0 ⊗k K ∼
= W.
Proof. By passing to the direct limit, we may assume K/k is finite. Enumerate G =
{σ1 , . . . , σn } and pick a basis {a1 , . . . , an } for K/k. The matrix A = (σi (aj ))ij ∈ Mn (K)
satisfies At A = (TrK/k (ai aj ))ij which is the Gram matrix of the trace form TrK/k and hence
invertible since the trace form is nondegenerate. This shows A ∈ GLn (k). Suppose w ∈ W
and set n n
X X
λj (w) = σi (aj w) = σi (aj )σi (w)
i=1 i=1
Theorem 37.1.19. For any Galois extension K/k with group G, H 1 (G, GLn (K)) = 1.
Proof. In this case GLn (K) = Aut(VK ) where V = k n . Suppose ξ ∈ Z 1 (G, GLn (K)). This
defines a twisted action of G on WK := K n by σ ∗ w = ξ(σ)σ(w) for any σ ∈ G. Setting
∼
Wk = WKG∗ , Lemma 37.1.18 shows that Wk ⊗k K = WK . Thus the isomorphism f : k n − → Wk
∼
extends to an isomorphism fK : K n − → WK and additionally, f and fK each commutes with
the Galois action:
f (σ(v)) = σ ∗ f (v) = ξ(σ)σ(f (v))
for all v ∈ k n and σ ∈ G (likewise for fK and v ∈ K n ). Replacing v with σ −1 (v), we get
f (v) = ξ(σ)σf (σ −1 (v)) for all v, so f = ξ(σ)σ(f ) which may be written f −1 ξ(σ)σ(f ) = 1.
This shows that ξ is equivalent to 1 in H 1 (G, GLn (K)).
731
37.1. Galois Descent Chapter 37. Descent
Here, the proof of Theorem 37.1.19 essentially demonstrates that F (K/k, VK ) = 1 and then
the bijection Θ is used to conclude that H 1 (G, GLn (K)).
Corollary 37.1.20. For any Galois extension K/k with group G, H 1 (G, SLn (K)) = 1.
Proof. Apply the long exact sequence in Galois cohomology to the sequence
det
1 → SLn (K) → GLn (K) −→ K × → 1
Example 37.1.21. Let K = k sep and fix a quadratic space (V, q) over K. We use de-
scent to show that Θ : F (K/k, (V, q)) → H 1 (G, Oq (K, n)) is always a bijection. Given
ξ ∈ Z 1 (G, Oq (K, n)), define a twisted action of G on WK := V ⊗k K by σ ∗ w = ξ(σ)σ(w).
Then Wk = WKG∗ is an n-dimensional vector space over k, so it remains to produce a quadratic
form on Wk to which q restricts. Such a form q 0 must satisfy q(w) = q 0 (w) ∈ k for all w ∈ Wk .
Indeed, if w ∈ Wk , σ ∗ w = w so
since ξ takes values in Oq0 (K, n) and q 0 is defined over k. This implies q restricts to q 0 on
Wk , so descent is always possible here.
Theorem 37.1.22. Let K/k be a Galois extension with group G, let A be a finite dimensional
k-algebra and set AK = A ⊗k K. Then H 1 (G, A×
K ) = 1.
Proof. Since H 1 (G, A×K ) is a direct limit over finite Galois extensions, we may assume K/k
is finite. Set d = [K : k]. View A and AK as right A-modules. For a cocycle ξ ∈ Z 1 (G, A× K ),
define the twisted action of G on AK by σ ∗a = ξ(σ)σ(a) and put B = AG∗ K . As vector spaces,
AK ∼ = B ⊗k K, so A ∼ = B as vector spaces. Note that B is also a right A-module. We claim
A∼ = B as A-modules. By the Krull-Schmidt theorem, we may write
A ⊗k K ∼
= I1m1 d ⊕ · · · ⊕ Irmr d and B ⊗k K ∼
= I1n1 d ⊕ · · · ⊕ Irnr d ,
732
37.1. Galois Descent Chapter 37. Descent
Nrd = NrdK/k : A −→ k
called the reduced norm map. Further, there is a short exact sequence of algebraic groups
det
1 → SL1 (AK ) → GL1 (AK ) −→ K × → 1.
By Theorem 37.1.22, H 1 (G, GL1 (AK )) = 1 so k × / Nrd(A× ) ∼ = H 1 (G, SL1 (AK )).
To interpret this cohomology set, recall that for an algebraic group G, the set H 1 (G, G(K))
classifies torsors for G defined over k, up to isomorphism over K. That is, an equivalence
class in H 1 (G, G(K)) is represented by a G-space X for which the map
G ×k X −→ X ×k X
(g, x) 7−→ (x, gx)
is an isomorphism. In general, a torsor X is defined over k but need not have any rational
points over k.
Example 37.1.23. For G = SL1 (AK ), the G-torsors are given by solutions to the algebraic
equation Nrd(x) = a for some a ∈ k × . Further, such a torsor has a rational point over k
if and only if a ∈ Nrd(A× ). This exhibits the bijection k × / Nrd(A× ) ∼ = H 1 (G, SL1 (AK ))
explicitly. For an example of this correspondence, let H be the quaternion algebra over R.
Then Nrd(x0 + x1 i + x2 j + x3 k) = x20 + x21 + x22 + x23 and the equation
X t QX = B
for X = (xij ) and some B ∈ Mn (k). This equation has a k-rational point if and only if Q
and B are similar over k.
733
37.2. Fields of Definition Chapter 37. Descent
for some gβγ ∈ K. Let k0 be the subfield of K generated by {gβγ }β∈B,γ∈C . Take f ∈ I and
write it
X
f= aα Mα for aα ∈ K
α∈A
X X
= aβ Mβ + aγ Mγ .
β∈B γ∈C
Since f ∈ I, we have
!
X X X
f= a0β Mβ + aγ Mγ − gβγ Mβ
β∈B γ∈C β∈B
with a0β ∈ K. Each Mγ − β∈B gβγ Mβ lies in I ∩ k[x1 , . . . , xn ] = Ik0 , so β∈B a0β Mβ ≡ 0
P P
mod I. Hence by definition of B, we have a0β = 0 for each β ∈ B, so f is a linear combination
of elements in Ik0 . This shows k0 is a field of definition for I.
To show k0 is minimal, let L ⊆ K be any field of definition of I and write IL = (p1 , . . . , ps )
for polynomials pi ∈ L[x1 , . . . , xn ]. Given the same notation as above, we must show that
gβγ ∈ L for all β ∈ B, γ ∈ C. For a fixed γ0 ∈ C, we have
X
Mγ0 − gβγ0 Mβ = p1 q1 + . . . + ps qs
β∈B
P
for some qi ∈ K[x1 , . . . , xn ], say qi = α∈A yiα Mα for yiα ∈ K. For 1 ≤ i ≤ s, define
X
qi∗ = Yiα Mα ∈ K[Yiα | 1 ≤ i ≤ s, α ∈ A][x1 , . . . , xn ]
α∈A
734
37.2. Fields of Definition Chapter 37. Descent
where ϕα (Y ) are linear functions of Y with coefficients in L itself. Consider the following
linear system over L:
Then {yiα }1≤i≤s,α∈A gives a solution to the system over K, but by linear algebra, there also
exists a solution {ȳiα }1≤i≤s,α∈A over L. Evaluating the system at this solution yields
X
Mγ0 − ḡβγ0 Mβ = p1 q̄1 + . . . + ps q̄s ∈ I
β∈B
735
37.3. Galois Descent for Varieties and Schemes Chapter 37. Descent
736
37.3. Galois Descent for Varieties and Schemes Chapter 37. Descent
In general, we define:
Definition. A k-structure on a scheme X over K is the data of a k-topology on X,
i.e. a collection of open sets Topk (X) ⊆ Top(X), and for each k-open set U ∈ Topk (X), a
k-structure on the K-algebra OX (U ), which are subject to the following conditions:
(1) For every inclusion of k-open sets V ,→ U , the morphism OX (U ) → OX (V ) is defined
over k.
S
(2) For any affine open cover X = Ui , the subspace k-topology on each Ui and the
specified k-structure on each OX (Ui ) agree with the k-structure on the affine scheme
(Ui , OUi ,k ) defined above.
Set G = Gal(K/k). For a K-scheme X, let F (K/k, X) denote the collection of k-
structures on X. Then there is a map
Θ : F (K/k, X) −→ H 1 (G, AutK (X)).
(To ensure the G-actions are continuous, we will always assume X is a scheme of finite type
over K.) By Lemma 37.1.8, Θ is always injective. To show Θ is surjective, one might try to
construct a Galois twist of X(K), but in general the fixed points of this space may be too
small to recover X under base change – in many cases X(K)G is even empty! The solution for
affine schemes at least is to define a new G-action on the algebra K[X] of regular functions
on X. Explicitly, for f ∈ K[X], σ ∈ G and ξ ∈ H 1 (G, AutK (K[X])), set
(σ ∗ f )(x) = σ(f (σ −1 ∗ x)) = σf (σ −1 (ξ(σ)−1 x)) = σ(f )ξ(σ)−1 x.
737
37.3. Galois Descent for Varieties and Schemes Chapter 37. Descent
x + x−1 x − x−1
s= and t= √ .
2 2 a
So K[T ]G∗ = k[s, t]. Noting that s2 − at2 = 1, we can rewrite this algebra as
x + bx−1 x − bx−1
s= and t= √ .
2 2 a
Note that s2 − at2 = b, so we may instead write L[U ]G∗ = k[S, T ]/(S 2 − aT 2 − b). Further,
note that the affine equation S 2 − aT 2 = b has a solution over k if and only if A splits over
k. The homogeneous equation
X 2 − aY 2 − bZ 2 = 0
defines a smooth, projective curve SB(A) over k, which is the aforementioned Severi-Brauer
curve for A. Note that SB(A) is a k-structure for P1L and SB(A) has a k-rational point
exactly when A splits over k. More broadly:
738
37.3. Galois Descent for Varieties and Schemes Chapter 37. Descent
Theorem 37.3.7 (Amitsur). Suppose A and B are quaternion algebras over k with Severi-
Brauer curves SB(A) and SB(B), respectively. Then A ∼
= B over k if and only if SB(A)
and SB(B) are birationally equivalent over k.
Example 37.3.8. For a central simple algebra A of degree n > 2, one can similarly construct
a Severi-Brauer variety SB(A). The most general form of Amitsur’s Theorem says that if
SB(A) and SB(B) are birational as k-varieties, then [A] and [B] generate the same cyclic
subgroup in Br(k). Whether the converse to this statement holds is an open question.
Question. Suppose X is an algebraic variety over K and k ⊂ K is a subfield. Is there a
variety Y over K such that X ∼
= Y as K-varieties and Y is defined over K?
In the case of an affine algebraic variety X ⊆ AnK , if there exists such a variety Y ⊆ Am
K (a
∼
priori m 6= n) that has a k-structure, then we can twist K[X] = K[Y ] to obtain a k-structure
on X. In general, this is not possible, but there are some conditions one can place on the
variety to allow for descent. The following examples illustrate these conditions for quadratic
forms.
Example 37.3.9. First consider the quadratic form x2 + 2iy 2 over Q(i). Notice that the
linear change of variables u = x, v = (1 + i)y
√ yields form uv + v 2 over Q. On
a quadratic √
2 2
the other hand, one can show that q = x − 5y defined over Q( 5) is not equivalent to a
quadratic form q0 over Q. Let
√
1 0
√
Q= ∈ GL2 (Q( 5))
0 − 5
be the symmetric matrix corresponding to q. If q were equivalent to a quadratic form
over Q, with √matrix Q0 ∈ GL2 (Q), then we would be able√ to write Q = B t Q0 B for some
B ∈ GL2 (Q( √ 5)). Taking determinants, we see that − 5 =√rs where r = det(B)2 is
a√square in Q(√ 5) and s = det(Q0 ) ∈ Q. Write r = (α + β 5)2√for α, β ∈ Q. Then
− 5 = (α + β 5)2 s and applying the nontrivial element τ ∈ Gal(Q( 5)/Q) yields
√ √
5 = (τ (α) − τ (β) 5)2 s.
√ 2 √ 2
Combining these
√ expressions, we get that (α + β 5) = −(τ (α) − τ (β) 5) , which is im-
possible in Q( 5). Therefore no such Q0 exists.
More generally, suppose K/k is a Galois extension with group G and Q is a symmetric
matrix in GLn (K) for which Q = B t Q0 B for some Q0 ∈ GLn (k) and B ∈ GLn (K). Then
for any σ ∈ G,
Qσ = (B σ )t Q0 B σ = A(σ)t QA(σ)
for A(σ) = B −1 B σ . Notice that the A(σ) satisfy A(στ ) = A(σ)σ(A(τ )) for all σ, τ ∈ G,
so they define a 1-cocycle in Z 1 (G, GLn (K)). By Theorem 37.1.19, any such cocycle is a
coboundary, which reflects the fact that A(σ) = B −1 B σ .
Example 37.3.10. For the quadratic form q = x2 + 2iy 2 over Q(i) above, its symmetric
matrix
1 0
Q= ∈ GL2 (Q(i))
0 2i
739
37.3. Galois Descent for Varieties and Schemes Chapter 37. Descent
Example 37.3.11. The previous example shows that having a similarity Q ∼ Qσ for some
Galois conjugate of the symmetric matrix is a necessary condition for descent of a quadratic
√ 2
2
√ sufficient. Consider the same quadratic form q = x − 5y
form, but it turns out that it is not
but now over the field K = Q( 5, i). Set k = Q(i) and G = Gal(K/k) = hτ i ∼ = Z/2Z. Then
Q and Qτ are similar via the cocycle
1 0
A(τ ) = ∈ GL2 (K).
0 i
Suppose q0 is a quadratic form which is equivalent to q over Q(i). Then taking determinants
of Q = B t Q0 B, we once again have equations
√ √
− 5 = (α + βi)2 s and 5 = (τ (α) + τ (β)i)2 s
√
for α, β ∈ Q( 5). These imply (α + βi)2 = −(τ (α) + τ (β)i)2 , or
In the first case, we must have τ (α) = β and τ (β) = −α, which implies τ 2 = −1, a
contradiction. The second case produces the same contradiction. Therefore descent is not
possible in this case.
The correct characterization of descent for quadratic forms is contained in the following
proposition.
740
37.3. Galois Descent for Varieties and Schemes Chapter 37. Descent
Proposition 37.3.12. Let K/k be a Galois extension with Galois group G, let Q ∈ GLn (K)
be a symmetric matrix representing a quadratic form q over K and assume that for every
σ ∈ G, there exists a 1-cocycle A ∈ Z 1 (G, GLn (K)) such that Qσ = A(σ)t QA(σ). Then q
descends to a quadratic form on k.
741