Exercises from Vakil

Ian Coley
February 6, 2015

2
2.1

Sheaves
Motivating example: The sheaf of differentiable functions

Exercise 2.1.A.
Show that the only maximal ideal of the germ of differentiable functions Op is mp , the ideal
of functions vanishing at p.
Solution.
We take for granted that mp is maximal. We will show that Op = Op \ mp . Suppose that
f Op does not vanish at p, and say f (p) = . Then since f is differentiable, it is continuous,
so it is nonzero in some neighbourhood U of p. Let K U be a compact neighbourhood of p,
which exists since we are working in locally Euclidean space. Then we may define g : U R
by
(
1/f (x) x K
g(x) =
.
0
xU \K
We see that g(x) is certainly continuous in K, but it may not be differentiable if f 0 (x) = 0
somewhere in K. However in this case, we must shrink U and try again until we get it right.
(CHECK) Having established that, we may define g on an open subset of p K to obtain
the appropriate germ.
Exercise 2.1.B.
Prove that mp /m2p is naturally isomorphic to the cotangent space at p as an R-module.
Solution.
We would like to establish the exact sequence
0 m2p mp Tp X 0.
The first map is clearly given by inclusion, but the second map is not as obvious. We will
take a function f and send it to its derivation df , which is a member of the cotangent space.
This map d is surjective by facts from differential geometry. It is clear that m2p ker d by
Leibniz rule: d(f 2 ) = 2f df , so d(f 2 )p = 0 since f (p) = 0. (TO BE FINISHED)

2.2

Definition of sheaf and presheaf

Exercise 2.2.A.
Prove: a presheaf is the same thing as a contravariant functor.
Solution.
This is reasonably clear. Suppose we have a contravariant functor F : X C, where C
is some (concrete) category and X the category associated to a topological space X. Then
we know that given an inclusion i : V U , we have a morphism F (i) : F (U ) F (V ).
Further, if we have the identity inclusion idU : U U , then F (idU ) : F (U ) F (U ) must
also be the identity. Further, we know that F respects (and reverses) compositions. This is
all the data that we needed.
Exercise 2.2.B.
Show that the following are presheaves on C but not sheaves: (a) bounded functions and (b)
holomorphic functions admitting a holomorphic square root.
Solution.
It is clear that both are presheaves by the same reasoning that differentiable functions were
a presheaf; functions can be restricted. For (a), we see this is not a presheaf since it violates
the gluability axiom. Suppose we have any open cover of C by bounded open sets. Define
fi : Ui C be given by fi (x) = |x|. Then on any Ui , we have a bounded function. However,
we cannot glue all these functions together, since eventually |x| .
For (b), since hes said this exercise is unimportant and Im not sure what a holomorphic
square root is, I will skip it.
Exercise 2.2.C.
S
The identity and gluability axioms may be rephrased to say that F ( i Ui ) is what limit?
Solution. S
We need F ( i Ui ) to be the limit S
of the diagram J with objects F (Ui ) and maps the
restriction maps resUi ,Uj . Let U = i U . If this is the case, then the universal property
of limits tells us that, given two f, g F (U ) which restricts onto each of the Ui in the
appropriate way, we have a unique morphism F (U ) F (U ) sending f to g.
f F (U )
res

J
res

g F (U )
Since the identity map works for this unique morphism, this must be our choice, so f = g.
This proves identity. Gluability follows from the existence of the limit at all, i.e. there is
some element in F (U ) restricting to each of the F (Ui ) in the appropriate ways.
Exercise 2.2.D.
2

(a) Verify that differentiable functions, continuous functions, smooth functions, and ordinary functions are shaves on a manifold X Rn .
(b) Show that real-valued continuous functions on (open sets of) a topological space X
form a sheaf.
Solution.
We assume that everything is a presheaf already. The method for (b) will also work for (a),
so we will just do that.
In a topological space, we do not have coordinate charts, but everything works fine still.
The identity axiom is clear. The gluability axiom also works since the function obtained by
gluing is still continuous. If we were dealing with smooth functions on a manifold, then the
result would still be smooth, as the functions are glued together on open sets and not on
closed sets (where we might have problems).
Exercise 2.2.E.
Let F (U ) be maps from U to a set S that are locally constant. Show that this is a sheaf.
Solution.
We take the description that F (U ) is the set of continuous maps U S, where S is
endowed with the discrete topology. This is clearly a presheaf, so we need only check that it
is a sheaf. But this is clear, as the identity axiom is trivial and the gluability axiom follows
by the following: glue together fi : Ui S into one function f : U S. Then for each
p U , choose the constant open neighbourhood around p to be the one guaranteed to exist
in Ui U .
Exercise 2.2.F.
Suppose Y is a topological space. Show that continuous maps to Y form a sheaf of sets
on X.
Solution.
That it is a presheaf and the identity axiom are obvious. To test gluability, suppose that
we have an appropriate cover {Ui }iI and functions fi : Ui Y . Define f : U Y in the
obvious way. Then let V Y be an open set. We know that fi1 (V ) is open in Ui for all
i I. Thus
[
f 1 (V ) =
fi1 (V )
iI

is still open, since an open subset of an open subset is still open in U . Hence the function is
continuous. We can apply this proof to the above cases.
Exercise 2.2.G.
(a) Suppose we are given a continuous map : Y X. how that sections of form a
sheaf.
(b) Suppose that Y is a topological group. Show that continuous maps to Y form a sheaf
of groups.
3

Solution.
(a) The identity axiom is clear as it ever was, since sections are a type of continuous
functions to a topological space. Suppose now that we have our set {si : Ui Y }.
Glue together the functions in the necessary way to create s : U Y . Then we need
s to be injective to be left invertible, but this is easily the case. Suppose s(p) = s(q)
with p 6= q. Then p and q cannot lie in the same Ui , since the si are all injective.
Therefore let p Ui and q Uj . We could not have ( s)(p) = ( s)(q) since p and
q lie in different open sets, which implies that s(p) 6= s(q), a contradiction.
(b) We can put a group structure on the set of continuous maps by pointwise multiplication.
The presheaf axioms are still clearly satisfied, and the identity element of any F (U ) is
the constant map sending every element to eY , so we have a contravariant functor to
the category of groups. There is no reason any of the other axioms should fail because
we have added the rest of this structure, and indeed they do not.

Exercise 2.2.H.
Suppose : X Y is a continuous map, and F is a presheaf on X. Then define F by
F (V ) = F ( 1 (V )), where V Y is open. Show F is a presheaf on Y , and is a sheaf
of F is.
Solution.
If we have such a presheaf, then we can define the restriction maps in the following way: let
U V Y be open subsets. Then since 1 (U ) = U 0 and 1 (V ) = V 0 are open subsets
in X satisfying U 0 V 0 , there is a restriction map r0 : F (V 0 ) F (U 0 ). As such, we can
construct the restriction map r : F (V ) F (U ) with this in the only sensible way.
Thus we have a presheaf structure on F . If F is a sheaf, then taking 2.2.C, everything
still works out fine.
Exercise 2.2.I.
Suppose : X Y is a continuous map, and F is a sheaf of sets on X. If (p) = q, describe
the natural morphism of stalks ( F )q Fp .
Solution.
We will take the definition of the stalk as a colimit. We have that
( F )q = lim F (U ) = lim F ( 1 (U ))

Since each U is an open set containing p, we have the natural map we wanted by the map
induced by 1 (U ) U , i.e. itself.
Exercise 2.2.J.
If (X, OX ) is a ringed space, and F is an OX -module, describe for each p X how Fp is
an OX,p -module.

Solution.
We have that
OX,p = lim OX (U )

and

Fp = lim F (U ).

Each F (U ) has a well-defined structure as a OX (U )-module. Further, if V U , we can

turn F (U ) into a OX (V )-module by moving forward in the limit diagram, since we know
everything must commute.

2.3

Morphisms of presheaves and sheaves

Exercise 2.3.A.
If : F G is a morphism of sheaves on X and p X, described an induced morphism of
stalks p : FP Gp .
Solution.
Since we have, for p U open,
Fp = lim F (U ),

we can define p by elements by appealing to (U ) : F (U ) G (U ). Taking the germ

definition, can take a representative for an equivalence class (f, U ) and map it according to
(U ), same as above.
Exercise 2.3.B.
Suppose : X Y is a continuous map of topological spaces. Show that the pushforward
gives a functor : SetsX SetsY , where Sets is an arbitrary choice.
Solution.
We have seen that the pushforward gives us a map of objects, but we need to check that it
respects morphisms. Let F , G SetsX , and let : F G be a sheaf morphism. Then we
see that must fit in the diagram below (for every open set (U)):
F

F G
Dissecting this, suppose we have an element F (U ) = F ( 1 (U )). Then we must
have (U )() = ( 1 (U ))(), which is an element of G ( 1 (U )). Hence under it is
send to an element of G (U ), as required. It is clear from this description that if is the
identify on F that is the identity on F so we are done.
Exercise 2.3.C.
Suppose F and G are two sheaves of sets on X, though it suffices they be presheaves. Let
H om (F , G ) be the collection of data
H om (F , G )(U ) := Mor(F |U , G |U ),
where F |U is the restriction of F to the open set U . Show that this is a sheaf of sets on X.
5

Solution.
Let us look at H om (F , G ) as a functor X Sets. Given an inclusion V U , we can define
the following map on the morphism sets: if f : F |U G |U , then we have a natural restriction
of F |U to F |V , which will in turn map onto G |V . This is how we map H om (F , G )(U )
H om (F , G )(V ). This gives us a presheaf structure on H om (F , G ). The sheaf axioms
follow from the limit definition of sheaf axioms on F and G . The H om functor that we
could develop now, but Vakil hints will be developed soon after, should commute properly
with limits.
Exercise 2.3.D.
(a) If F is a sheaf of sets on X, then show that H om ({p}, F )
= F , where {p} is the
constant sheaf associated with the one element set {p}.
(b) If F is a sheaf of abelian groups on X, then show that H om AbX (Z, F )
= F as sheaves
of abelian groups.
(c) If F is an OX -module, then show that H om OX (OX , F )
= F as sheaves of OX modules.
Solution.
For (a), we need to check what is happening on any open set U X. Consider H om ({p}, F )(U ).
We have
H om ({p}, F )(U ) = Mor({p}|U , F |U ) = { : {p} F (U )}
= F (U ).
The Mor set is just the inclusion of one point into F (U ), so these are isomorphic as sets. As
described above, given V U , the restriction maps would then just map : {p} F (V ),
and the identity and gluability should follow easily too.
We are now going to generalise the above, since all we are considering is the sheaf based
around the object in a category representing the identity functor. We can address (b) and (c)
simultaneous by generally considering sheaves of R-modules. In general, HomR-Mod (R, M )
=
M as R-modules. Therefore we have for every open U ,
H om R-ModX (R, F )(U ) = { : R F (U )}
= F (U ).
Applying this to the case of R = Z, OX is what we want.
Exercise 2.3.E.
Check that, for a morphism of presheaves of abelian groups : F G , kerpre is a presheaf.
Solution.
To write down what this object actually is, we have (kerpre )(U ) := ker((U ) : F (U )
G (U )). We will take Vakils hint and use the diagram, for V U ,
(U )

0 kerpre (U ) F (U ) G (U )

res
res
U,V
U,V
?y
y
y
(V )

0 kerpre (V ) F (V ) G (V )
6

The map ? should be the restriction map for kerpre if we manage to endow it with a
presheaf structure. Indeed, we can take resU,V and restrict its domain to kerpre (U ), but
it is not clear that it lands in kerpre (V ). However, we have kerpre (U ), we have
(U )() = 0, hence resU,V (U )() = 0 as well. By commutativity, (V ) resU,V () = 0,
whence resU,V () ker (V ) as we wanted.
Exercise 2.3.F.
Show that the presheaf cokernel cokerpre satisfies the universal property of cokernels.
Solution.
This is clear when we look at what is happening on open sets. In that case, we have the
picture
F (U )

G (U )

p(U )

cokerpre (U )

0.

Suppose that we had some sheaf T and F G T with q = 0. Then on every open
set, we have the picture
0

F (U )

(U )

p(U )

G (U )

cokerpre (U )
q(U )
!

T (U )
Since cokerpre (U ) is the cokernel of (U ), we have a unique map shown above. This allows
us to define a unique map cokerpre T globally.
Exercise 2.3.G.
Show (or observe) that for a topological space X with open set U , F 7 F (U ) gives a
functor from presheaves of abelian groups on X, Abpre
X , to Ab. Show that this functor is
exact.
Solution.
We have now taken for granted that Abpre
X is an abelian category, otherwise this question
would make no sense. Also, I have observed the fact above, so we move on to exactness.
Suppose that
a
b
0 F G H 0
is an exact sequence of abelian group presheaves on X. Since a and b are defined on open
sets, this means that
a(U )

b(U )

0 F (U ) G (U ) H (U ) 0
must still be exact. That is all that we wanted to show.

Exercise 2.3.H.
Show that a sequence of presheaves 0 F1 F2 Fn 0 is exact if and only if
0 F1 (U ) F2 (U ) Fn (U ) 0 is exact for all U .
Solution.
I think based on Vakils statement that Homological algebra (exact sequences and so forth)
works, and also works open set by open set makes this trivial. The forward direction is
obvious (and we have been using it this entire time), and we can build up the appropriate
kernels and cokernels to show exactness open set by open set.
Exercise 2.3.I.
Suppose : F G is a morphism of sheaves. Show that the presheaf kernel kerpre is in
fact a sheaf. Show that it satisfies the universal property of kernels.
Solution.
As the problem suggests, assuming that kerpre is a sheaf, then it must satisfy the universal
property as AbX is a full subcategory of Abpre
X .
To show it is a sheaf, we just need to figure out identity and gluability. Since kerpre is
a subobject of F , we know exactly how to do this.
Exercise 2.3.J.
Let X be C with the classical topology, and let Z be the constant sheaf on X associated
to Z, OX the sheaf of holomorphic functions, and F the presheaf of functions admitting a
holomorphic logarithm. Describe an exact sequence of presheaves on X
0 Z OX F 0
where Z OX is the natural inclusion and OX F is given by f 7 exp(2if ). Show
that F is not a sheaf.
Solution.
The natural inclusion is clearly injective. Suppose a function g F (U ) admits a holomorphic
1
logarithm log g. Then we may let f = 2i
log g OX (U ). The map is surjective on open
sets, so is overall surjective.
Suppose exp(2if ) = 0 identically on U . Then we must have f (z) Z for all z U ,
hence f comes from the inclusion Z(U ) OX (U ). This proves exactness.
We are told that F does not satisfy gluability, so we check this. If we let Dr be the open
unit disc centred at the origin of radius r, let U = D2 \ D1 , an open annulus. Take an open
cover of U by
A = {(x, y) U : x (1.5, 2)},

B = {(x, y) U : x (2, 1.5)}.

Let fA , fB be the constant function. Since on both these sets we may make a branch cut
either along arg z = 0 or arg z = to define a holomorphic logarithm, but it cannot be done
on the entire annulus, so gluability fails.

2.4

Properties determined at the level of stalks, and sheafification

Exercise 2.4.A.
Prove that a section of a sheaf of sets is determined by its germs, i.e., the natural map
Y
F (U )
Fp
pU

is injective.
Solution.
According to the hints, we will need only the identity axiom to prove this. First, the natural
map f above maps an element F (U ) to its germ (, U ) over p. Indeed, this map is
induced by the restriction map from F (U ) to some F (V ) for some neighbourhood p Vp
U . Since the identity axioms tells us that two sections in F (U ) restricting identically to an
open cover of U (which we have here) are in fact equal, this proves injectivity.
Exercise 2.4.B.
Show that supp s := {p X : sp 6= 0 in Fp } is a closed subset of X.
Solution.
We will show its complement is open. Suppose that we have p X so that sp = 0 in Fp .
Then there must be some neighbourhood V of p so that s|V = 0 constantly. Therefore s is
locally constantly 0, so {p X : sp = 0} is open.
Exercise 2.4.C.
Prove that any choice of compatible germs for a sheaf of sets F over U is the image of a
section of F over U .
Solution.
We are pushed towards using gluability. Suppose that we have such a bunch of compatible
germs. Then from the definition, we obtain an open cover {Up } of p and sections s0p so that
we have sections (s0p , Up ). Then we may glue these together to obtain a section on U , i.e.
some (s, U ) so that resU,Up s = s0p . This implies that the value on each of the stalks is als
correct so we are done.
Exercise 2.4.D.
If 1 and 2 are morphisms from a presheaf of sets F to a sheaf of sets G that induce the
same maps on each stalk, show that 1 = 2 .
Solution.
We use the diagram
1

F (U )
Q

pU Fp

G (U )

1 =

pU

Gp

We may do even better: since the right vertical map is injective, it is invertible from its
image. Therefore we need only check that taking 1 on a stalk yields a compatible germ.
But this is obvious. Therefore since we have equality on the bottom maps, we have equality
of the top maps.
Exercise 2.4.E.
Show that a morphism of sheaves of sets is an isomorphism if and only if it induces an
isomorphism of all stalks.
Solution.
Using the above picture again, let : F (U ) G (U ) be a morphism of sheaves of sets, and
let F and G be the natural inclusions of the set of sections into the product of stalks.

F (U )

G (U )

pU

G
Q

Fp

pU

Gp

We refine this to

F (U )

im F

G (U )

p |im F

im G

If either the top or the bottom of this diagram is an isomorphism, then the other one is
an isomorphism as well. As stated above, it is clear that the bottom map takes compatible
germs to compatible germs, so the bottom map is well-defined, so we are done.
Exercise 2.4.F.
(a) Show that Exercise 2.4.A is false for general presheaves.
(b) Show that Exercise 2.4.D is false for general presheaves.
(c) Show that Exercise 2.4.E is false for general presheaves.
Solution.
(a) The hint overall is to use the two-point space X = {p, q} with the discrete topology.
Define F as follows: we let F ({p} = F ({q}) = {}, some one point set, and let
F (X) = {a, b}, some two point set. This is a presheaf, since we let the restriction
maps resX,{p} and resX,{q} be the constant maps with value (as they must since one
point sets are a final object in Sets).
Now, the map F (X) Fp Fq = {(, )} is a set map from a two point set to a
one point set, which cannot be injective.
(b) We will continue using the above sheaf. Let G be very similar to F : we will let
G (X) = {c, d} and G ({p}) = G ({q}) = {}. Let 1 , 2 : F G be defined as follows:
we must have 1 () = on Fp and Fq . On F (X), let 1 (a) = 1 (b) = c and let
2 (a) = 2 (b) = d. Then we have two different sheaf morphisms which agree on stalks,
but are not equal.
10

(c) Take 1 above. By construction, the bottom map is an isomorphism on stalks, but we
do not have an isomorphism overall as 1 (X) is not surjective.

Exercise 2.4.G.
Show that sheafification is unique up to unique isomorphism, assuming it exists. Show that
if F is a sheaf, then the sheafification is id : F F .
Solution.
Suppose that F1sh and F2sh are two sheafifications of F . Then these are both sheaves, so we
have a diagram
F1sh
sh

id

sh
sh

F2sh

F1sh

The two dashed morphisms must compose to be the identity, since that is the unique map
making the top triangle commute, so we have a unique isomorphism from F1sh to F2sh .
Further, if F is itself a sheaf, F satisfies the universal property of F sh by letting sh : F F
be given by the identity, so it must be its own sheafification.
Exercise 2.4.H.
Assume for now that sheafification exists. Use the universal property to show that for
any morphism of presheaves : F G , we get a natural induced morphism of sheaves
sh : F sh G sh . Show that sheafification is a functor from presheaves on X to sheaves on
X.
Solution.
We use the diagram below:
F

shF

sh

F sh

shG

G sh

We know there is a composite map F G sh by shG . By the universal property of

sheafification, there is a unique map F sh G sh making the diagram commute, which we
denote by the dashed morphism sh .
Since we have assumed how sh acts on objects of Abpre
X and now have shown how it
works on morhpisms, we have a good candidate for a functor. That composition works is
clear, and that sh takes the identity to the identity follows from 2.4.G. There fore we have
a functor sh : Abpre
X AbX .
Exercise 2.4.I.
Show that F sh forms a sheaf.
11

Solution.
We already know what the restriction maps should be, so we have a presheaf. Suppose that
{Ui } is a open cover of U , and suppose that f, g F sh (U ) are two sections that agree on
each restriction onto F sh (Ui ), which we denote fi , gi respectively. We know that we have
fi = (fi,x Fx : x Ui , Vi Ui and si F (Vi ) s.t. si,y = fi,y y Vi )
and similar for gi . Because of this constructions, we see that all the fi,x and gi,x must be the
same, which would make f = g overall.
Now suppose that we have fi F sh (Ui ) with the intersections agreeing appropriately.
Then suppose we take the section f F sh (U ) obtained by taking each member of the stalks
fi,x , since we know that if x Ui Uj , then by assumption fi,x = fj,x . We know that this
collection is a set of compatible germs because, at each x U , there is some Ui with x Ui .
As such there is an open Vi Ui such that the set of stalks is the image of a section. Since
Vi U is still open, we may take this as our neighbourhood. Thus we have glued our sections
together as required, so we have a sheaf.
Exercise 2.4.J.
Describe a natural map of presheaves sh : F F sh .
Solution.
We know that we have a natural maps
F (U )

Fp F sh (U ).

pU

Q
This should be our natural map of presheaves. Since F sh (U ) is a subobject of pU Fp , the
second map will make sense in abelian categories (or Sets), so we are fine. For it to be a
map of presheaves, it would have to play well with restriction maps, i.e. for V U
F (U )

pU

Fp

resU,V

resU,V

F (V )

F sh (U )

qV

Fq

F sh (V )

It is clear that the left and right squares commute individually, so we have the required
commutative diagram above.
Exercise 2.4.K.
Show that the map sh satisfies the universal property of sheafification.
Solution.
Suppose that G is a sheaf and : F G a map of presheaves. We will first construct the
map : F sh G then show it is unique. Let f F sh (U ). Construct an open cover Ui of
U so that on each Ui we have a section fi F (Ui ) and fx = fi,x for all x Ui .
Now, consider gi = (fi ) G (Ui ). Since G is a sheaf and it is clear that the gi agree on
intersections by the above work, we may glue the gi together into a unique g G (U ) which
12

agrees on each of the restrictions. We want to define (f ) := g for this g. It is also clear that
sh = from this, and it is unique because we were forced by commutativity to define
this way.
Exercise 2.4.L.
Show that the sheafification functor is left adjoint to the forgetful functor from sheaves on
X to presheaves on X.
Solution.
pre
Let u : CX CX
be the forgetful functor. Then we need to show that, for a presheaf F
and sheaf G ,
H om (sh(F ), G )
= H om (F , u(G )).
We know that given any map from a presheaf to a sheaf, we have a unique map from its
sheafification to that sheaf. This gives an inclusion of the right set in the left. Now, given
a map from F sh G , we can precompose with the natural map (abusing the hell out of
notation) sh : F F sh to obtain a map F G , which means the opposite inclusion also
holds. Therefore we have a bijection between these two sets. Depending on the category C
we will want a stronger isomorphism, but this follows easily.
Exercise 2.4.M.
Show F F sh induces an isomorphism of stalks.
Solution.
We will take the set of germs definition of stalks. Recall that Fp is the set of equivalence
classes (f, U ) where p U is open, f F (U ), and (f, U ) (g, V ) if there is an open
W U V with resU,W (f ) = resV,W (g).
We know that, locally, each f F sh (U ) is some tuple of stalk elements (fx ) which
(locally enough) are the image a section F (V ) with V U . Therefore associated to a stalk
element fx , x U , we have a section s F (V ) so that sx = fx . That is, fx is representable
by (s, V ). Therefore we have an injection Fxsh
QFx . Further, given some germ (s, V ) in
Fx , we may construct the element (sy , y V ) pU Fp . Then this is a compatible germ,
hence is an element of F sh (U ). Therefore sx is an element of the stalk Fxsh . This is the
bijection which implies an isomorphism.
Exercise 2.4.N.
Suppose : F G is a morphism of sheaves of sets on a topological space X. Show that
the following are equivalent:
(a) is a monomorphism in the category of sheaves.
(b) is injective on the level of stalks.
(c) is injective on the level of open sets.

13

Solution.
We will take Vakils recommendations: since we have shown that morphisms are determined
by the action on stalks for sheaves, (b) implies (a) and (b) implies (c), since we can view
stalks either globally or locally. (c) also implies (a). We therefore need only show that (a)
implies (c).
Vakil suggests for this to use the indicator sheaf for U , which we will denote TU for
lack of any other notation. being a monomorphism means that, for any two maps 1 , 2 :
T F , we have
1 = 2 = 1 = 2 .
Therefore let us try this with the indicator sheaf TU . Suppose that (U ) : F (U ) G (U ) is
not injective, so that (U )(s) = (U )(t) for some s 6= t F (U ). Then let 1 , 2 : TU F
be defined as follows: if {a} is the set associated to each TU (V ) for V U open, then we
would like
(
resU,V s i = 1
i (V )(a) =
,
resU,V t i = 2
where we include the possibility V = U so that resU,U = id. This does define two sheaf
morphisms such that 1 = 2 . However, we have 1 6= 2 since s 6= t by assumption,
so we have a contradiction. Therefore (U ) must be injective on each U .
Exercise 2.4.O.
Continuing the notation of the previous exercise, show that the following are equivalent:
(a) is an epimorphism in the category of sheaves.
(b) is surjective on the level of stalks.
Solution.
(b) implies (a) is clear, as it was above. For (a) implies (b), we take Vakils hint and use the
skyscraper sheaf with value {a, b, c} at p, which we denote Tp . If is an epimorphism then
for any 1 , 2 : G T such that 1 = 2 , then we have 1 = 2 .
Suppose p : Fp Gp is not surjective, and let g Gp not be in the image of p . Then
let us define maps 1 , 2 : G Tp by defining what happens on stalks: since Tp,q , that is
the stalk of Tp at q, is equal to {a, b, c} for q = p and the final object in Sets elsewhere, let
us define for s Gp ,

a s = g, i = 1
i,p (s) = b s = g, i = 2 .

c s 6= g
There are sheaf morphisms 1 and 2 corresponding to these sheaf maps, and these satisfy
1 = 2 . However by construction 1 6= 2 , so cannot be an epimorphism. This
proves the contrapositive of the converse, so we are done.
Exercise 2.4.P.
Let X = C with the classical topology, let OX be the sheaf of holomorphic functions, and let
exp

OX
be the sheaf of invertible (nowhere zero) holomorphic functions. Show that OX OX

describes OX as a quotient sheaf of OX . Find an open set on which this map is not surjective.
14

Solution.
We know that the exp sheaf map is an epimorphism if and only if expp is surjective for all
p X. Therefore we will show this. From Exercise 2.3.J, we know that we have an exact

sequence of (pre)sheaves if we replace OX

by holomorphic functions admitting a holomorphic
logarithm. From complex analysis, we can define the log of any function whose codomain is
a subset of C \ {0} by
Z 0
f (z)
dz,
(log f )(z) =
f (z)
where is a path between 0 and z (which one does not matter since C is simply connected).
Therefore we can write f = exp(log f ). Since the stalks in this case are just locally-defined
functions supported at p, this shows that every expp is surjective.
We will run into the same problem if we try to define functions on an annulus. Let U be
the open annulus centred at 0 constructed in Exercise 2.3.J. Then the constant function has
no logarithm, so the map exp(U ) : OX (U ) O (U ) cannot be surjective.

2.5

Sheaves of abelian groups, and OX -modules, form abelian categories

Exercise 2.5.A.
Show that the stalk of the kernel is the kernel of the stalks.
Solution.
We want a natural isomorphism
(ker(F G ))x
= ker(Fx Gx ).
Let : F G for ease of notation. Let ker denote the object of the kernel of in the
category of sheaves. Suppose we have an element f (ker )x (we will let ker x be the right
hand side of the above). Then since we have a natural monomorphism ker F , we have
an injection (ker )x Fx (assuming we are working over a nice category, otherwise we just
have a monomorphism). As such we can map (ker )x Gx via x . We summarise this in
the picture
0

ker

(ker )x

Fx

Gx

The top row of this diagram is exact. By diagram chasing, we can see the bottom is exact too.
Therefore (ker )x satisfies the requirements of ker x , so the two must be isomorphic.
Exercise 2.5.B.
Show that the stalk of the cokernel is naturally isomorphic to the cokernel of the stalk.
Solution.
Take the above proof and look at cokernels instead. Using additionally the fact that the
presheaf cokernel and the sheaf cokernel have isomorphic stalks (Poblem 2.4.M), we are
done.
15

Exercise 2.5.C.
Suppose that : F G is a morphism of sheaves of abelian groups. Show that the image
sheaf im is the sheafification of the image presheaf. Show that the stalk of the image is the
image of the stalk.
Solution.
We have now shown that we are working in an abelian category. We know that im =
ker coker in this case. Taking kernels will always give us a sheaf, but we do not know that
coker is a sheaf a priori. Writing pre when necessary, we see

kerpre (cokerpre )

cokerpre

sh

coker

ker(coker )

We want ? to be the sheafification map. As such, we will show that ker(coker ) satisfies the
universal property of the sheafification of kerpre (cokerpre ). Write H pre for kerpre (cokerpre )
and write H for its sheafification. By the above, the composite map H pre coker is
zero, which is a map from a presheaf into a sheaf. Therefore it factors uniquely through to
H coker . However, since kerpre (coker ) = ker(coker ), we know that H pre must also
factor uniquely through ker(coker()). Since this is another map between H pre and a sheaf,
we have another factorisation through H . This gives

H pre
H

cokerpre

coker

ker(coker )

Since this diagram commutes, if the starred dashed arrow is not the identity, then we have
two different ways to move from H pre through H to coker . Thus it must be the identity,
so ker(coker ) is the sheafification of the presheaf image.
That the stalk of the image is the image of the stalk follows from the previous two
exercises.
Exercise 2.5.D.
Show that talking the stalk of a sheaf of abelian groups is an exact functor.
Solution.
Consider the sheaf category AbX . Let p X be a point. Suppose that

0F G H 0
16

is an exact sequence of sheaves. We want to show that

p

0 F p G p Hp 0
is also exact. A criterion for exactness in the first sequence is that im = ker . We know
that
(im )p = im p ,
(ker )p = ker p
by the above problems. Therefore since (im )p = (ker )p , the sequence in Ab is exact since
its maps satisfy the above criterion.
Exercise 2.5.E.
To check the exponential sequence
exp

2i

0 Z Ox OX
1

is exact, we know that we may check it on stalks. Let p C be any point and let = 2i.
We have shown before that the first map is a monomorphism and the second map is an
epimorphism, i.e. p is injective on stalks and expp is surjective on stalks. We need only
show that im p = ker expp .
We know that im p ker expp since exp(2im) = 0 for all m Z, so the function
described is the zero function. Further, let f OX such that exp(f ) = 0. Then we know
that f (x) 2iZ for all x U , where U is some small open set containing p. Since f takes
values on a discrete set, we know that there is some smaller neighbourhood of p so that f is
constant, with value f (p) = 2in (for some n). This neighbourhood can be described as the
connected component containing p of f 1 (2in). Because we are working on stalks, f and
the constant function 2in represent the same germ, so we know that f comes from n Zp .
This completes the proof.
Exercise 2.5.F.
Suppose U X is an open set, and 0 F G H is an exact sequence of sheaves of
abelian groups. Show that
0 F (U ) G (U ) H (U )
is exact. Show that the section functor need not be exact.
Solution.

For the second part of this problem, we showed in 2.4.P that OX (U ) OX

(U ) is not always
surjective, e.g. for U the open annulus. Therefore we would not have right exactness for
that section functor.
We will do this the long way, not using the fact that all right adjoints are left exact. We
started this process in Exercise 2.4.N: if F G is a monomorphism, then F (U ) G (U ) is
injective. Therefore this sequence is exact at F (U ). Therefore we need only show that the
image of F (U ) in G (U ) is the kernel of G (U ) H (U ). We know that any exact sequence
can be factored into a short exact sequence. Exactness at G follows is equivalent to

for F G H ,

0 im G im 0 is short exact.

Since we have shown that im = ker ,we need to show that ker (U ) G (U ) im (U )
is short exact. But this is obvious.
17

Exercise 2.5.G.

Suppose 0 F G H is an exact sequence in AbX . If : X Y is a continuous

map, show that


0 F G H
is exact.
Solution.
We need to show that im = ker . It would be nice to check this on stalks, but there
is no clear way of writing down what the stalks in this case are, so we will check it on open
sets instead. We can check this on open stalks because we do not need right exactness of ;
that would cause problems only at exactness at H . Now, we know that
(U ) = ( 1 (U )),

(U ) = ( 1 (U )).

The condition we have to check boils down thus:

im (U ) = ker (U ) im ( 1 (U )) = ker ( 1 (U )).
But this is obviously true by assumption.
Exercise 2.5.H.
Suppose F is a sheaf of abelian groups on a topological space X. Show that H om (F , )
is a left exact covariant functor AbX AbX . Show that H om (, F ) is a left exact
contravariant functor.
Solution.
Let G and H be two other sheaves, and let : G H be a sheaf morphism. Since
it is already obvious that H om (F , ) takes one sheaf to the homomorphism sheaf, we
need to demonstrate its behaviour on morphisms. We see that H om (F , ) induces a map
: H om (F , G ) H om (F , H ) given by 7 . We see that H om (, F ) induces a
map : H om (H , F ) H om (G , F ) given by 7 .
We need only show left-exactness. We cannot use stalks again, because H om (F , G )p 6=
Hom(Fp , Gp ) in general. We can use the same open set argument as we did above, though,
since we do not need to worry about the last place. The proof is virtually identical.
Exercise 2.5.I.
Show that if (X, OX ) is a ringed space, then OX -modules form an abelian category.
Solution.
This is almost immediate since we know that sheaves of abelian groups form an abelian category. We only need to notice that subobjects, quotient objects, coproducts, and products of
OX -modules have a natural OX -module structure, so the sheaves of abelian groups satisfying
various universal properties actually lie in the category of OX -modules.
Exercise 2.5.J.
18

(a) Suppose OX is a sheaf of rings on X. Define (categorically) what should mean by the
tensor product of two OX -modules. Give an explicit construction, and show that it
satisfies your categorical definition.
(b) Show that the tensor product of stalks is the stalk of the tensor product.
Solution.
(a) What it should be is the following: let F and G be two OX -modules. Then suppose we
have a map : F G T of OX -modules so that on each U X, is OX (U )-linear
in both arguments. There may be a better way to define OX -linearity in this case, but
I do not see what it is.
We want to define the tensor product as the morphism F G F OX G such
that for any such as above, there is a unique map such that the diagram below
commutes:
F G
F G

T
Our guess for this explicit construction is, for p X, define the stalks of the tensor
sheaf to be
(F G )p = Fp Gp .
There is absolutely no guarantee what we obtain is a sheaf, so we sheafify for good
measure. We will keep the same notation. We need to show that this definition works.
Since morphisms are defined on stalks, we know that there is a unique morphism of
sheaves that corresponds to the stalk morphisms we propose. Therefore it satisfies the
universal property we described.

2.6

The inverse image sheaf

Exercise 2.6.A.
Let : X Y be a continuous map and let G be a sheaf on Y . Then let
1 G pre (U ) = lim
G (V ).

V (U )

Note that (U ) is not open in general. Show that this is a presheaf, but not necessarily a
sheaf, on X.
Solution.
We will show that we have the structure of a contravariant functor. Fortunately, the restriction maps are easy: suppose that V U X. Then clearly (V ) (U ). Therefore if we
have f 1 G pre (U ), we can represent it by an element (x, W ) for some open W (U ).
19

Then W (V ) is still an open set, so in fact (x, W ) represents an element of 1 G pre (V ).

We need only figure out what element of the colimit it corresponds to, and this is the value
of resU,V (f ).
To show it is not a sheaf in general, consider Y = {p} and X = {a, b}, where is the
constant map. Give X the discrete topology. If G is the constant sheaf S on Y , where S is
some set, then since the only (nontrivial) open set in Y is Y itself, we must have
1 G pre (U ) = G (Y ) = S
for any open U in X. This makes 1 G pre the constant sheaf on X. We know from previous
work that this is not a sheaf if we choose S to be a set with more than one element.
Exercise 2.6.B.
If : X Y is a continuous map, and F is a sheaf on X and G is a sheaf on Y , describe
a bijection
MorX ( 1 G , F ) MorY (G , F ).
Observe that this bijection is functorial in both F and G (i.e. natural). Thus 1 satisfies
the universal property of the left adjoint of .
Solution.
We will do this using Vakils hint. Consider U X and V Y open, and let V U : G (V )
F (U ) be any map. Then we call a collection = {U V : U X, V Y open} compatible
if the following holds: for all open U 0 U and V 0 V with (U 0 ) V 0 , we require the
diagram
G (V )

V U

resV,V 0

G (V 0 )

F (U )
resU,U 0

V 0 U 0

F (U 0 )

to commute. We let MorY X (G , F ) be the set of all compatible collections of maps. Vakil
suggest we show that both of the above sets are equal to this one we have just created.
Now, suppose we have a map : 1 G F . We want to show that we can turn this
map into a compatible collection. Let U X. Since 1 G (U ) is a direct limit of G (V ) for
V (U ) open, we can construct a map V U for all of these U : each x G (V ) corresponds
to some element y 1 G (U ), so let V U (x) = (U )(y). This commutes with restriction
maps because we know that (U ) does. This process clearly works backwards as well.
Suppose we have a map : G F . Let V Y be open. Since F (V ) = F ( 1 (V )),
we can construct a map V 1 (V ) in the obvious way. For x G (V ), let V 1 (V ) (x) =
(V )(x). This process is clearly reversible as well. Therefore we have shown these two
sets are the same. Showing functoriality should be trivial, because we require the nice
commutativity of these diagrams (which would be the only snag).
Exercise 2.6.C.
Show that the stalks of 1 G are the same as the stalks of G . More precisely, if (p) = q,
describe a natural isomorphism Gq
= ( 1 G )p .
20

Solution.
Using adjointness should be the easiest way. We know that left adjoints commute with
colimits, and that stalks are defined as colimits. Therefore since
1 G pre (U ) = lim
G (V ),

V (U )

we have
1 G pre (U ) = 1 lim
G pre (U ) = 1 (G pre )p = (G pre )q .
( 1 G pre )p = lim

pU

pU

Since presheaves and sheaves have the same stalks, this completes the proof.
Exercise 2.6.D.
If U is an open subset of Y , i : U Y is the inclusion, and G is a sheaf on Y , show that
i1 G is naturally isomorphic to G |U .
Solution.
Let us look at the definition. Suppose that W is an open set in U . Then by definition,
i1 G pre (W ) = lim
G (V ).

V i(W )

In particular, if we look at the diagram implied by the colimit, it has a terminal object,
namely W itself. Since the whole diagram must commute through the arrow W i(W ), we
have that the colimit itself must be G (W ). Therefore the data of i1 G pre is the sheaf data
of every open set contained in U , which is just G |U .
Exercise 2.6.E.
Show that 1 is an exact functor from sheaves of abelian groups on Y to sheaves of abelian
groups on X.
Solution.
Taking the hint, we know that we can check exactness on stalks, and by 2.6.C, we have an
isomorphism of stalks under 1 . Therefore suppose we have an exact sequence
0F G H 0
in AbY . Then at every point p X with (p) = q, we have
0

Fq

Gq

Hq

0 ( 1 F )p ( 1 G )p ( 1 H )p 0
Since the top row is exact, the bottom row is also exact. Therefore 1 is an exact functor.

Exercise 2.6.F.
21

(a) Suppose Z Y is a closed subset, and i : Z Y is the inclusion. If F is a sheaf of

sets on Z, then show that the stalk (i F )q is a one element set if q
/ Z, and Fq if
q Z.
(b) Define the support of a sheaf G of sets, denoted supp G , as the locus where the stalks
are not the one-element set:
supp G := {p X : |Gp | =
6 1}.
Suppose supp G Z where Z is closed. Show that the natural map G i i1 G is an
isomorphsm. Thus a sheaf supposed on a closed subset can be considered a sheaf on
that closed subset.
Solution.
(a) We have, for V open in Y ,
(i F )q = lim
i F (V ) = lim
F (i1 (V )).

qV

qV

Since i1 (V ) = Z V , which is open in Z, then we have two cases. If q Z, then

every open set (in Z) containing q can be described as Z V for some V open in Y .
Therefore we lose no data here, and we are taking the colimit over (essentially) the
same diagram, whence (i F )q
/ Z, then V Z will be empty for sufficiently
= Fq . If q
small V (because Z is a closed set). In this case, the colimit is being taken over empty
sets at the end of it, whence (i F )q must be the final object of Sets, i.e. a one-element
set.
(b) This is asking if the unit is a natural isomorphism. This occurs when the left adjoint
is fully faithful, which is provable in an abstract categorical way. To see this: let L
and R be an adjoint pair between C and D. If L is fully faithful, then we have for any
A, B C,
MorC (A, B)
= MorD (L(A), L(B))
= MorC (A, R L(B)),
where the second isomorphism follows from adjunction. Since this holds for all A, B
C, we must have R L naturally isomorphic to 1C .
In our case, we just need to show that i1 is fully faithful, i.e. that the first isomorphism
as above actually holds. But in this case, since supp G Z, taking i1 does not affect
the data of any important open sets; any sheaf map can only be defined in one way
outside of supp G since the one-point set is final. Therefore i1 is fully faithful in this
case, so we have the required natural isomorphism.

Exercise 2.6.G.
Suppose i : U Y is the inclusion of an open set into Y . Define the extension of i by zero
i! : ModOU ModOY as follows. Suppose F is an OU -module. For open W Y , define
(ipre
! F )(W ) = F (W ) if W U and 0 otherwise. This is clearly a presheaf OY -module.
sh
Define i! as (ipre
! ) . Note that i! F is an OY -module, and that this defines a functor.
22

(a) Show that ipre

! F need not be a sheaf.
(b) For q Y , show that (i! F )q = Fq if q U and 0 otherwise.
(c) Show that i! is an exact functor.
(d) If G is an OY -module, describe an inclusion i! i1 G G .
(e) Show that (i! , i1 ) is an adjoint pair, so there is a natural bijection
HomOY (i! F , G ) HomOU (F , G |U )
for any OU -module F and OY -module G .
Solution.
Even though Vakil says we shouldnt do this question right now, its worth a shot.
(a) The recommendation here is to try the sheaf of continuous functions on R to find a
counterexample. Let F be this sheaf. Let U be the disjoint union (0, 1) (2, 3). Then
ipre
! F (1, 2) = 0. However, we can find continuous functions on (0, 1) and (2, 3) that
cannot be glued together, since the restriction of the gluing on (1, 2) must be 0, which
is not necessarily true. Concretely, the constant function 1 on (0, 1) and (2, 3) cannot
be glued (continuously) to be 0 on (1, 2).
(b) If we dont need this point to prove (e), then it follows because left adjoints commute
with colimits. As it stands, however, I think we should prove this directly. Recalling
that the stalks of a presheaf and its sheafification are isomorphic,
(ipre
ipre F (V ).
! F )q = lim
!
qV

If q U , then eventually all open V with q V are also contained in U , so that

pre

ipre
/ U , then sufficiently
! F (V ) = F (V ). Then it is clear that (i! F )q = Fq . If q
pre
many V with q V yields i! F (V ) = 0 to force Fq = 0.
(c) (b) shows that i! is exact since it is easily seen to be exact on the level of stalks. If
q U , then it follows from the reasoning of 2.6.E. If q
/ U , then the sequence on
stalks is identically zero, which is trivially exact.
(d) We can try to define this on the level of open sets. Let W Y be open. We have
(
G |U (W ) W U
i! i1 G (W ) = i! G |U (W ) =
0
else
The inclusion G |U (W ) G (W ) or 0 G (W ) is perfectly well defined.

23

(e) We will establish this bijection as best we can. Suppose that : i! F G is an

OY -module homomorphism. Suppose that we have W U open. Then we have a
map
(W ) : i! F (W ) = F (W ) G (W ).
Since W U , we have G |U (W ) = G (W ), so (W ) is a perfectly well defined map on
F (W ) G |U (W ) as well. If W is not contained in U , then i! F (W ) = 0. Therefore
(W ) must be the (initial) zero map. But fortunately, this makes G |U (W ) = 0 too,
so (W ) : F (W ) G |U (W ) is the (final) zero map. Therefore we have an inclusion
of HomOY (i! F , G ) HomOU (F , G |U ). The reverse inclusion follows identically. That
we have a bijection follows easily as well: the pairs i! F and F and G and G |U have
the same stalks at the applicable open sets, so we can check to see if we get the same
map back by checking stalks.

2.7

Recovering sheaves from a sheaf on a base

Exercise 2.7.A.
How can you recover a sheaf F from the partial information on some base {Bi } of the
topology on X?
Solution.
We know how to recover the stalks now. The stalk Fp is obtained by taking the colimit over
all open sets containing p. But we know for every arbitrary open U containing p, there is
a base element B so that p B U . Put another way, a germ over p can be represented
by some (f, U ), but it can also be represented as (g, B) where B is a base element. This is
enough information to give us the data of the stalks. We can reconstruct
S compatible germs
by gluing the stalks back together for each open U X and, if U = i IBi , using the
known data of F (Bi ). This is probably precise enough.
Exercise 2.7.B.
Verify that F (B) F (B) is an isomorphism, likely by showing that it is injective and
surjective.
Solution.
We have defined F to be a sheaf (of sets) on the base {Bi }. To write back down the formal
definition of F :
F (U ) := {(fp Fp )pU : p U, B with p B U, s F (B), with sq = fq q B}.
It is the set of all compatible germs where we make sure that the open subset of U we pick
is in the base. That sentence alone pretty much proves what wed like.
Let us look at the natural map F (B) F (B). Let us first show it is surjective. Given
some set (fp ) of the above form, we may examine the elements (sq ) guaranteed by the
definition. This is an element of F (B 0 ) for some B 0 B. As such, there is an element of
F (B) which restricts down to (sq ), and this would then be a preimage of (fp ).
24

Injectivity is also clear: if we have (gp ) and (hp ) mapping to the same (fp ) F (B), then
we know that (gp ) and (hp ) must agree on all subbase elements of F (B), which (by identity)
would make them the same element. This gives us the required isomorphism.
Exercise 2.7.C.
Suppose {Bi } is a base for the topology of X. A morphism F G of sheaves on the base
is a collection of maps F (Bk ) G(Bk ) such that the diagram
F (Bi ) G(Bi )

resB ,B
resBi ,Bj
y
y i j
F (Bj ) G(Bj )
commutes for all Bj Bi .
(a) Verify that a morphism of sheaves is determined by the induced morphism of sheaves
on the base.
(b) Show that a morphism of sheaves on the base gives a morphism of the induced sheaves.
Solution.
We may now assume Theorem 2.7.1, which is very good. I think we can be very clever about
this. We know that H om (F , G ) is itself a sheaf on X, and so it is uniquely determined by
its data on a base, i.e. the restriction maps and the data of
H om (F , G )(B) = Hom(F (B), G (B))
= Hom(F (B), G(B)),
where the last isomorphism comes from 2.7.B. Therefore it seems that this proves (a) and
(b) simultaneously. This might be a little too clever, and in particular incorrect.
Exercise 2.7.D.
S
Suppose X = Ui is an open cover of X, and we have sheaves Fi on Ui along with isomorphisms ij : Fi |Ui Uj Fj |Ui Uj that agree on triple overlaps, i.e. jk ij = ik on
Ui Uj Uk . Show that these sheaves can be glued together into a sheaf F on X (unique
up to unique isomorphism) such that Fi
= F |Ui , and the isomorphisms over Ui Uj are the
obvious ones.
Solution.
We know that each Fi (for i I) is determined by its information on a base {Bji } for j Ji
some index set. Then we claim that
[[
B=
Bji
iI jJi

S
is a base for X. Indeed, let V S X be open. Then we can write V = Vi , where Vi = Ui V
is open. In turn, each Vi = Bki since each of those sets are a base for Ui . Therefore we
have
[[
V =
Bki ,
iI kJi

25

so we indeed have a base for X. The conditions above guarantee that we can come up with
a well defined sheaf F on the base B. To be more explicit, we gluability and identity hold
because they hold on each Ui and they agree on double and triple intersections. Therefore
we have a sheaf F on X from F on B. Everything else, including unique up to unique
isomorphism, follows from the theorem.
Exercise 2.7.E.
Suppose a morphism of sheaves F G on a base Bi is surjective for all Bi . Show that the
corresponding morphism of sheaves is surjective (or more precisely, an epimorphism). The
converse is not true, unlike the case for injectivity.
Solution.
Suppose we had some (U ) : F (U ) G (U ) which was not surjective. Let q G (U ) be an
element not in the image of (U ). Then q G (B) for some basis element B. As such q is in
the image of (B). This is a flagrant contradiction.

3
3.1

Toward affine schemes: the underlying set, and topological space

Toward schemes

Exercise 3.1.A.
Suppose that : X Y is a continuous map of differentiable manifolds (as topological
spaces). Show that is differentiable if differentiable functions pull back to differentiable
functions, i.e. if pullback by gives a map OY OX .
Solution.
Since we are working with differentiable manifolds, let V be an open subset of Y diffeomorphic to Rn (for whatever appropriate n). Then we shall construct the map ] : OY OX
on V , using the terminology from 3.1.1.
First, what is OX (V )? It is precisely OX ( 1 (V )), 1 (V ) being (for a properly chosen
V ) a chart on X. Therefore what is the map ] ? Given a function f OY , we should have
] (f ) = f . If this map is differentiable, then we know by the chain rule that itself must
be differentiable, as
(f )0 (x) = f 0 ((x)) 0 (x).
In particular, we can extract 0 (x) using only the data from the derivatives of f and f
that we are guaranteed exist. Since we have shown what we must on a base of the topology
for Y , we are done because now we need only glue everything together.
Exercise 3.1.B.
Show that a morphism of differentiable manifolds : X Y with (p) = q induces a
morphism of stalks ] : OY,q OX,p . Show that ] (mY,q ) mX,p .

26

Solution.
Consider a germ (f, V ) OY,q . Then since is a morphism of differentiable manifolds, we
are to assume that it is a differentiable function, so we have the map ] described above. We
such, we can take (f, V ) 7 (f , V ). Then we know that this is a perfectly fine differentiable
map in this neighbourhood V from 3.1.A, so this is how we get a map on stalks. If we had
f (p) = 0, then we would have f ((q)) = f (p) = 0, so ] (f, V ) would also vanish. This shows
the inclusion of the maximal ideals.

3.2

The underlying set of affine schemes

Exercise 3.2.A.
(a) Describe the set Spec k[]/(2 ). The ring k[]/(2 ) is called the ring of dual numbers.
You should think of as a very small number, so small that its square is 0 (although
it itself is not 0).
(b) Describe the set Spec k[x](x) .
Solution.
(a) The ring consists of all a + b, where a, b k. We know that all choices where b = 0
force the ideal (a + b) = k. If b 6= 0, then we can represent (a + b) by (a/b + ), so
we can just write (a + ). We see that
(a + )(b + ) = ab + (a + b),
since the 2 term vanishes. In particular, if a 6= 0, then we can choose b = a so that
(a + )(b + ) k.
All this is to say that the only prime ideal seems to be (). Therefore Spec k[]/(2 ) =
{()}.
(b) Recall what k[x](x) means: we invert all elements outside of (x), since this is a multiplicatively closed set. Since the only prime ideals in k[x] were of the form (x a) for
a k, this means the only prime ideals we have left are (x) and (0). This spectrum
therefore also has only two points.

Exercise 3.2.B.
Show that for prime ideals in R[x] of the form (x2 + ax + b) with that polynomial irreducible,
the quotient R[x]/(x2 + ax + b)
= C always.
Solution.
Up to a linear change of variables, (in particular, x 7 y a/2), we can rewrite
x2 + ax + b = y 2 + c
for some other constant c. This c will necessarily be positive, because a linear change of
variables does not change the discriminant of the polynomial. Since = b 4ac
and we
require < 0 for irreducibility, we have c > 0. We know that R[y]/(y 2 + c) = R[ c]
= C.
Since the linear change of variables is also an isomorphism on the quotient spaces, we are
done.
27

Exercise 3.2.C.
Describe the set A1Q .
Solution.
What are the prime ideals here? We are still in a PID, so we need only find the irreducible
polynomials. We know we need only to consider monic polynomials in Q[x]. We know that
monic irreducible polynomials are completely determined by their roots. Further, we know
that theseroots must come inGalois-conjugate (to use Vakils term) tuples; e.g., we could
not have 2 a root without 2 also a root.
Therefore it seems that every irreducible polynomial f corresponds to some set {a1 , . . . , an }
of Galois-conjugate tuples of complex numbers. Further, given any particular ai , we can
recover the rest of the set of conjugates because they must be the roots of the minimal
polynomial of ai in Q[x], which is the f we started with.
A1 , and the kernel of this map is the action of
This means we have a surjection Q
Q
Q). Trying to make this analogous to A1 , in that case
the absolute Galois group Gal(Q,
R
we had to glue together all complex conjugates. In this case, we have to glue together all
Galois-conjugates. Indeed, we had
Gal(R,
R) = C/(x = x).
A1R = R/
This situation here is analogous.
Exercise 3.2.D.
If k is a field, show that Spec k[x] has infinitely many primes.
Solution.
We know that prime ideals correspond to irreducible polynomials. If k is a field of characteristic zero, it is plain to see that (x p) for each p N prime are an infinite set of
prime ideals. If k has characteristic p, then the existence of finite fields of order pn for all n
guarantees irreducible polynomials of every degree in k[x]. These too are an infinite set of
prime ideals.
Exercise 3.2.E.
Show that the only prime ideals of C[x, y] are of the form (x a, y b) or (f (x, y)) for an
irreducible polynomial f C[x, y].
Solution.
We might not have a PID, but C[x, y] is a UFD. We know that every (f (x, y)) is a prime
ideal, so assume that p is an ideal that is not principal. Suppose that no f (x, y) and g(x, y)
in p were relatively prime. Then there would be some highest degree polynomial h(x, y) p
dividing every element of p, which implies that p ((h(x, y)). Since h(x, y) p, this shows
that p is principal, a contradiction.
Now, let f (x, y) and g(x, y) be these relatively prime polynomials. Then (f (x, y), g(x, y))
p. As implied by the hint, since C[x, y] = C[x][y], we can use obtain the greatest common
divisor of f (x, y) and g(x, y) in C[x], which is some h(x) 6= 0. Since h(x) p and p is a
prime ideal, and because h(x) splits over C, some prime divisor (i.e. irreducible factor) of
28

h(x) is contained in p, say (x a). The same process for C[x, y] = C[y][x] shows that there
is a (y b) p. But (x a, y b) is a maximal ideal, so if p is a proper ideal of C[x, y] and
(x a, y b) p, this must actually be an equality.
Exercise 3.2.F.
Show that the Nullstellensatz implies the Weak Nullstellensatz
Solution.
Writing down what they want, we assume that if k is a field, then every maximal ideal of
k[x1 , . . . , xn ] has a residue field a finite extension of k.
Since C is an algebraically closed field, it has no finite extensions. Therefore every
maximal ideal of C[x1 , . . . , xn ] must have residue field isomorphic to C itself. This only
occurs in the stated case, where m = (x1 a1 , . . . , xn an ). It suffices to check what happens
if we take one generator of a maximal ideal to be a non-linear irreducible polynomial. In
this case, the residue field clearly contains a nontrivial residue class for some xi , so it cannot
be isomorphic to C. For example, if we had f (x, y) = y 2 x3 , then x C[x, y]/(f (x, y)) is
nontrivial, so C[x, y]/(f (x, y)) is not isomorphic to C.
Exercise 3.2.G.
Prove that any integral domain A which is a finite k-algebra must be a field (not requiring
the Nullstellensatz).
Solution.
Were almost all of our way to a field anyway. We need only show that we are working in a
division algebra, i.e. every element is invertible. This is implied by showing, taking the hint
that `x : A A, y 7 x y is an isomorphism for all nonzero x A.
We certainly know that `x is injective; since A is an integral domain, we have cancellation.
If `x (a) = `x (b), then
`x (a) `x (b) = `x (a b) = x(a b) = 0.
Since we took x 6= 0, then we must have ab = 0, i.e. a = b. Since `x is a ring homomorphism
(not hard to see), it is a fortiori a vector space homomorphism. An injective map between
vector spaces of the same dimension must be a vector space isomorphism. In particular, the
map is surjective. Therefore it is also surjective as a ring homomorphism, and therefore a
ring isomorphism. This shows that there exists c A so that `x (c) = 1A for every x A\{0}.
Therefore A is a commutative division domain, so it is a field.
Exercise 3.2.H.

Describe the maximal ideal of Q[x, y] corresponding

to
(
2,
2)
and
(
2,

2). Describe

the maximal ideal of Q[x, y] corresponding to ( 2, 2) and ( 2, 2). What are the
residue fields in both cases?
Solution.

From what was discussed in 3.2.C, we know that for A1Q , 2 and 2 correspond to the

ideal generated by x2 2 in Q[x]. Similarly, the ideal corresponding to ( 2, 2) should be
29

just (xy 2, x2 2), which is definitely maximal

because we see its quotient is a field. When

2
we mod
out by this ideal, we
get K =Q( 2) from thex 2 contribution, and y is forced
to be 2 when we let x = 2 and 2 when x = 2. Therefore the whole quotient is
just K.
In the other case, we want to pick the maximal ideal (xy + 2, x2 2). We get the same
field extension but y is forced to be the opposite sign as x for the two choices of x.
Exercise 3.2.I.
Consider the map of sets : C2 A2Q defined as follows: (z1 , z2 ) is sent to the prime ideal
of Q[x, y] consisting of polynomials vanishing at (z1 , z2 ).
(a) What is the image of (, 2 )?
(b) Show that is surjective.
Solution.
(a) The image should be the zero ideal. It is known that is not an algebraic number.
2 under the gluing of
We proved above that prime ideals of Q[x, y] correspond to Q
we must map (, 2 ) trivially to (0).
Galois-conjugates. Since
/ Q,
(b) Vakil says we need to use other machinery to solve this, but it seems doable regardless.
We know that prime ideals of Q[x, y] are going to correspond to irreducible polynomials
in x, in y, or in both. We know that every irreducible polynomial in Q[x] has a root
in C. If we have a prime ideal of the form (f (x), g(y)), then choosing a root z1 of f (x)
and a root z2 of g(y) will make a preimage (z1 , z2 ) for this ideal. This is because any
polynomial (in x) vanishing at z1 must contain the factor
Y
(x (z1 ))

Gal(Q,Q)

for all Galois conjugates of z1 (without superfluous overrepresentation). This is the

minimal polynomial of z1 , hence irreducible, so it must be the f (x) we started with.
The same goes for g(y).
For ideals of the form (f (x, y)) for an irreducible f with nontrivial degree in both x
and y, we need to be slightly more creative. Suppose that f (x, y) did not have a root
in C2 . Then for any fixed b C, the (degree at least one) polynomial f (x, b) C[x]
would have no root. But this is a contradiction. Therefore f (x, y) has a root (a, b) in
C2 , and this root is a preimage. This is probably enough work to satisfy the solution
at this point in time.

Exercise 3.2.J.
Suppose A is a ring, and I an ideal of A. Let : A A/I. Show that 1 gives an
inclusion-preserving bijection between primes of A/I and primes of A containing I.

30

Solution.
This is a standard commutative algebra problem. Let p A/I be a prime ideal. Then 1 (p)
is prime for the following reason: suppose that ab 1 (p). Then (ab) = (a)(b) p,
hence (without loss of generality) (a) p. Therefore a 1 (p). Clearly 1 (p) contains
1 ((0)) = I, so this proves one direction.
Conversely, if q is a prime ideal of A, then (q) is prime in A/I. This generally holds for
any surjective map. Suppose that ab (q). Then 1 (a) and 1 (b) are both nonempty,
since is surjective, so let a0 1 (a) and b0 1 (b). Then a0 b0 q, so we have (without
loss of generality) a0 q, whence a (q). If q does not contain I, then taking 1 ((q))
gives a prime ideal containing I, so we get back something different from what we began
with. Therefore if we demand that q I, we have the required bijection. That inclusion is
preserved is completely obvious.
Exercise 3.2.K.
Suppose S is a multiplicative subset of A. Describe an order-preserving bijection of the
primes of S 1 A with the primes of A that do not meet S.
Solution.
We have a natural inclusion (though strictly speaking it may not be injective) : A S 1 A
such that a 7 a/1. Suppose that p is a prime ideal of S 1 (A). Then we know that p does
not contain any invertible elements. In particular, p does not contain any elements of the
form s/1 since these are invertible. Since every noninvertible element of S 1 A is in the image
of , we can take 1 (p), which (as discussed above) is a prime ideal that necessarily avoids
S.
Now suppose that q is a prime ideal of A. If q S 6= , then (q) contains an invertible
element, whence the ideal generated by this set cannot be a prime ideal. If q S = , then
we claim the ideal generated by (q) is prime ideal. Let (q) = p. Suppose that
a b
p.
s t
Then we have st as bt = ab p. Then since ab is in the image of q, we know that (without
loss of generality) a q. Therefore a/s p. This is the desired bijection.
Exercise 3.2.L.
Show that (C[x, y]/(xy))x
= C[x]x (in some natural way).
Solution.
C[x]x should have a pretty easy spectrum, since from previous problems we know Spec C[x]x
is just those prime ideals in C[x] avoiding {1, x, x2 , . . .}. Since (nonzero) prime ideals of C[x]
are of the form (x a) for any a C, we just need to avoid the choice a = 0. Any other
choice of a gives an ideal not containing x, for if x (x a) then a (x a) which implies
(x a) = C[x], so it is not a maximal ideal. Therefore we can realise Spec C[x]x = C \ {0}.
We know from the last few problems that Spec C[x, y]/(xy) can be viewed as a subset of
Spec C[x, y], which we have discussed. Specifically, Spec C[x, y]/(xy) contains those prime
ideals of Spec C[x, y] which contain (xy). Since prime ideals of C[x, y] are (by 3.2.E) just of
31

the form (x a, y b) or (f (x, y)), we know what to ignore. Since xy is irreducible, it never
shows up in anything of the form (f (x, y)) unless f (x, y) = xy, and this corresponds to the
zero element in Spec C[x, y]/(xy). Therefore the nonzero prime ideals in Spec C[x, y]/(xy)
are limited to those of the form (x a, y b). But we know that the element
(x a)(y b) = xy bx ay + ab = ab bx ay (x a, y b).
We do some more manipulation to show
ab bx ay + b(x a) + a(y b) = ab (x a, y b).
This shows in particular that we must have a = 0 or b = 0. Therefore we want ideals of the
form (x, y b) or (x a, y) for a, b C. Now, to deal with the localisation, we need to throw
out all choices of a = 0 and everything of the form (x, y b). Therefore the primes we have
left are of the form (x a, y) for a C \ {0}. That is precisely what we concluded above,
and we can see how the correspondence works now:
Spec C[x]x 3 (x a) (x a, y) Spec(C[x, y]/(xy))x

Exercise 3.2.M.
If : B A is a map of rings, and p is a prime ideal of A, show that 1 (p) is a prime ideal
of B.
Solution.
Im not sure how we couldve gotten away without proving this by this point. See Exercise
3.2.J.
Exercise 3.2.N.
Let B be a ring.
(a) Suppose I B is an ideal. Show that the map Spec B/I Spec B is the inclusion
map implied by 3.2.J.
(b) Suppose S B is a multiplicative set. Show that the map Spec S 1 B Spec B is the
inclusion implied by 3.2.K.
Solution.
Im reasonably sure I proved this well enough above to satisfy both myself and Vakil.
Exercise 3.2.O.
Consider the map of complex manifolds sending C C via x 7 y = x2 . We interpret the
domain as the x-line and the codomain as the y-line. Interpret the corresponding map of
rings given by
C[y] C[x] given by y 7 x2 . Verify that the fibre above the point a C is
the point(s) a C, using the definition given above.

32

Solution.
I think I know what theyre asking for here, but Im unclear since this is mostly a thought
exercise. The spectra of C[x] and C[y] can be identified with C (plus the mysterious (0)).
Therefore looking at the map : C[y] C[x] where y 7 x2 gives us the map on spectra
which is the first map described above. Im not sure what else to write down.
Exercise 3.2.P.
Suppose k is a field, and f1 , . . . , fn k[x1 , . . . , xm ] are given. Let : k[y1 , . . . , yn ]
k[x1 , . . . , xm ] be the ring morphism defined by yi 7 fi .
(a) Show that induces a map Spec k[x1 , . . . , xm ]/I Spec k[y1 , . . . , yn ]/J for any ideals
I k[x1 , . . . , xm ] and J k[y1 , . . . , yn ] such that (J) I.
(b) Show that the map of part (a) sends the point (a1 , . . . , am ) k m (i.e. (x1 a1 , . . . , xn
an ) Spec k[x1 , . . . , xm ]) to
(f1 (a1 , . . . , am ), . . . , fn (a1 , . . . , am )) k n .
Solution.
(a) This is clearly true if I = 0 because we can use the canonical map. Now, we know that
prime ideals of Spec k[x1 , . . . , xm ]/I are in bijective correspondence with prime ideals of
Spec k[x1 , . . . , xm ] containing I. A prime ideal p containing I will be mapped to a prime
ideal containing 1 (I). For 1 (p) to translate back to an ideal in k[y1 , . . . , yn ]/J we
need J 1 (p). This occurs if (J) p, which is true because (J) I p.
Therefore this map is well defined.
(b) If we think about maximal ideals as the vanishing set at a given point (which is
the correspondence between the Spec and points of affine space), then this shouldnt
be too bad. If we consider an element g k[y1 , . . . , yn ], then (g(y1 , . . . , yn )) =
g(f1 (x1 , . . . , xm ), . . . , gn (x1 , . . . , xm )). Therefore if p k[x1 , . . . , xm ] is the vanishing
set of a1 , . . . , am , we would like 1 (p) to be the vanishing set of the point (b1 , . . . , bn )
so that
(g(b1 , . . . , bn )) = g(f1 (a1 , . . . , am ), . . . , fn (a1 , . . . , am )).
Therefore these bi should correspond to fi (a1 , . . . , am ).

Exercise 3.2.Q.
Consider the map of sets : AnZ Spec Z given by the ring map Z Z[x1 , . . . , xn ]. If p is
prime, describe a bijection between the fibre 1 ([(p)]) and AnFp . Can you interpret the fibre
over [(0)] as Ank for some field k?
Solution.
Let be the ring map above. Then we have
1 ([(p)]) = {p Z[x1 , . . . , xn ] : 1 (p) = (p)}.
33

What does this mean? When we take the preimage of p under , all that remains are
the constant polynomials in that prime ideal. There may not be any such; in this case
(p) = (0). Otherwise, we have some minimal constant polynomial n in p. As such, the
ideal nZ[x1 , . . . , xn ] p. However, it is clear that n must be a prime integer: otherwise we
would have ` m = n p for some other constant polynomials ` and m, nullifying primeness.
Now, we know that AnFp = Spec Fp [x1 , . . . , xn ] = Spec Z/pZ[x1 , . . . , xn ]. Since prime ideals
of Z/pZ[x1 , . . . , xn ] are in bijection with ideal of Z[x1 , . . . , xn ] containing pZ[x1 , . . . , xn ], we
see the bijection starting to emerge. An ideal p Spec[x1 , . . . , xn ] mapping to (p) contains
the constant polynomial p. Therefore pZ[x1 , . . . , xn ] p, so p correspond to an ideal in
Z/pZ[x1 , . . . , xn ] = Fp [x1 , . . . , xn ]. This gives us a point in finite affine space AnFp .
If p is an ideal without any such constant polynomial p, as we said above it goes to the
zero ideal. We can probably picture this as a point in AnQ , and we might even have a bijection
there. We certainly know that for every monic irreducible polynomial in Q[x1 , . . . , xn ], we
can associate to it an irreducible polynomial in Z[x1 , . . . , xn ] by clearing denominators.
Indeed, these polynomials have the same roots, and the association is unique. We can use
this to come up with a bijection of closed points between AnZ lying over [(0)] and AnQ .
Exercise 3.2.R.
(a) Show that if I is an ideal of nilpotents, then the inclusion Spec B/I Spec B is a
bijection.
(b) Show that the nilpotents of a ring B form an ideal, the nilradical which we denote N.
Solution.
These are pretty standard commutative algebra exercises. It suffices to show that the nilradical is the intersection of all prime ideals for both parts. Let x N, which we only assume is
a subset for now, and that xn = 0. Then we have for any prime ideal p that xn = xn1 x p,
whence either xn1 or x p. If we then have xn1 p, then we can write this xn2 x p,
whence xn2 or x p. By induction this shows x p. Therefore N p for every prime
ideal p, so
\
N
p.
pB

Now we need to show that this is everything. Suppose x

/ N. Then we will show that x
/p
n
for some prime p. Let S be the set of ideals not containing x for any n N. Since (0) S
this is a nonempty set which has a partial ordering on it, hence by Zorns lemma S has a
maximal element m.
We claim that m is a prime ideal. Suppose that we have a, b B such that a, b
/ m but
ab m. Then consider the ideal
a = {z B : a z B}.
Then m ( a, hence xn a for some n. By the same reasoning,
b = {z B : xn z B}
34

is an ideal strictly bigger than m, so this ideal contains some xm . But then this implies that
xn xm = xn+m m, which is a contradiction. Therefore m is prime and hence x does not
belong to some prime ideal.
Having shown this, it is clear that the inclusion above is a bijection: since primes in B/I
correspond to primes in B containing I, this is all primes. Further, the intersection of ideals
is again an ideal, so N is an ideal.
Exercise 3.2.S.
Prove that the nilradical is the intersection of all prime ideals.
Solution.
Oops, I didnt read ahead. See above.
Exercise 3.2.T.
Suppose we have a polynomial f (x) k[x]. Instead, we work in k[x, ]/(2 ). What then is
f (x + )?
Solution. P
Let f (x) = ni=0 ai xi . Then we have
f (x + ) =

n
X

ai (x + )i =

n
X

i=0

i=0

ai

i  
X
i ij j
x
j
j=0

Most of those terms, however, vanish since 2 = 0. Hence

=

n
X

ai (xi + ixi1 ) = f (x) + f 0 (x).

i=0

This looks like it will be very convenient.

3.4

The underlying topological space of an affine scheme

Exercise 3.4.A.
Check that the x-axis is contained in V (xy, yz).
Solution.
Recall that
V (s) := {[p] Spec A : S p}.
Since the x-axis is given by y = z = 0, we know that it correspond to the prime ideal (y, z).
Certainly xy and yz are contained in this ideal.
Exercise 3.4.B.
Show that if (S) is the ideal generated by S, then V (S) = V ((S)).
Solution.
This is trivial. If S p, then (S) p. The converse is even more trivial.
35

Exercise 3.4.C.
(a) Show that and Spec A are both open.
(b) If Ii is a collection of ideals (as i runs over some index set), show that
\
X
V (Ii ) = V (
Ii ).
i

Hence the union of any collection of open sets is open.

(c) Show that V (I1 ) V (I2 ) = V (I1 I2 ). Hence the intersection of any finite number of
open sets is open.
Solution.
(a) We know that V () = Spec A, hence is open. Further, V (A) = , since no prime
ideal contains anything, therefore Spec A is open.
T
(b) We will show double inclusion. Suppose that p i V (Ii ). Then Ii p for all Ii ,
therefore P
the ideal generated by all P
of these Ii must also be a subset of p. This is
V ( i Ii ), then we have p V (Ii ) for each i as well,
precisely Pi Ii . Conversely, if p T
since Ii i Ii . Therefore p i V (Ii ). This completes the equality.
(c) Suppose that p V (I1 )P
V (I2 ). Then without loss of generality, I1 p. As such,
any linear combination nj=1 ai xi p, where xi I1 and ai A. These combinations
contain all A-linear combinations of elements of I1 and I2 , so I1 I2 p, hence p
V (I1 I2 ). Conversely, suppose that p
/ V (I1 ) V (I2 ). This means that I1 6 p and
I2 6 p. Let a I1 \ p and b I2 \ p. Then if we had p V (I1 I2 ), then in particular we
must have ab p. But a, b
/ p, which contradicts that p was a prime ideal. Therefore
we have V (I1 ) V (I2 ) V (I1 I2 ) and (V (I1 ) V (I2 ))c (V (I1 I2 ))c , which shows
double inclusion.

Exercise 3.4.D.
If I A is an ideal, then define its radical by

I := {r A : rn I for some n N}.

Show that I is anpideal. Show that V ( I) = V (I). We say that an ideal is radical if

I = I. Show that
I = I and that prime ideals are radical.
Solution.

Certainly
since
I

I,
we
have
V
(
I) V (I). We need only show that if p V (I), then

p V ( I). Consider some r I with rn I. Then since rn I p, we have either rn1

or r p. Going with the same
procedure as dealing with the nilradical above, by induction
this shows that r p, whence I p. This proves the first part of this problem.

36

We
need
only
to
show
that
I

I, since the other inclusion is obvious. Suppose

p

r
I. Then there exists m N sothat rm I. In turn, there is an n N such that
(rm )n I, i.e. rmn I, Therefore r I. This proves the second thing.

Again, we need only to show that p p. Suppose that r p and rm p. Then

going by the same exact argument by induction, we have r p. Therefore prime ideals are
radical.
Exercise 3.4.E.
If I1 , . . . , In are ideals of a ring A, show that
v
un
n p
\
u\
t Ii =
Ii .
i=1

i=1

Solution.
Suppose
that r is an element of the lefthand
side. Then for some m N, we have rm

Tn
m
Ii for all i, hence r is in the righthand side.
i=1 Ii , i.e. r Ii for all i. Therefore r
Conversely, if r is in the righthand side, then for each i there is an mi so that rmi Ii .
Let m = max mi (which exists because we only dealing with a finite intersection). Then
rm Ii for all i, hence r is in the lefthand side. Note that without finiteness this might not
have worked.
Exercise 3.4.F.

Show that I is the intersection of all prime ideals containing I.

Solution.
The hint is to use that the nilradical of A is the intersection of all prime ideals of A. Consider
the ring A/I. Then the nilradical of A/I is all r A/I such that rn = 0 for some n, i.e.
rn I for some n. The representatives for N(A/I) is precisely the radical of I. Since we
know that prime ideals in A/I correspond to prime ideals in A containing I, we have
\
\

I = N(A/I) =
p=
q.
pA/I

IqA

This completes the solution. Note that while we are swapping around somewhat wildly
between the two rings, there is no actual ambiguity because of the previous work weve done
on the spectra of these two spaces
Exercise 3.4.G.
Describe the topological space A1k .
Solution.
We know that prime ideals in k[x] correspond to irreducible polynomials, since this space
is a PID. We showed earlier that this space has infinitely many points no matter which
field k we consider. The open sets here are complements of (finite unions of) vanishing sets
of irreducible polynomials p(x) k[x]. It is pretty much the same as A1C in terms of its
description.
37

Exercise 3.4.H.
By showing that closed sets pull back to closed sets, show that : Spec A Spec B induced
from : B A is a continuous map. Interpret Spec as a contravariant functor RingsTop.
Solution.
It suffices to check this on a basis for the closed sets of Spec B. Let V (I) Spec B be a such
a closed set. Then we have
1 (V (I)) = {p Spec A : (p) V (I)} = {p : I 1 (p)} = V ((I)).
Therefore closed sets pull back to closed sets, so this map is indeed continuous. The interpretation of the contravariant functor follows readily: a ring morphism : B A induces a
continuous map of topological spaces : Spec A Spec B, with the necessary axioms about
the identity morphism easily seen to be satisfied.
Exercise 3.4.I.
Suppose that I, S B are an ideal and multiplicative set, respectively.
(a) Show that Spec B/I is naturally a closed subset of Spec B. If S = {1, f, f 2 , . . .} for
some f B, show that Spec S 1 B is naturally an open subset of Spec B. Show that
for arbitrary S, Spec S 1 B need not be open or closed.
(b) Show that the Zariski topology on these subsets is the subspace topology induced by
the inclusion.
Solution.
(a) We know that primes in B/I correspond to primes in B containing I. Therefore
Spec B/I = V (I) Spec B, so this is a naturally closed subset. We know that
primes in S 1 B correspond to primes in B avoiding S. If we take S as in the problem
statement, then Spec S 1 B = Spec B \ V ((S)) = Spec B \ V ((f )). As such, Spec S 1 B
is open. This should also hold whenever S or (S) satisfies some finitely-generated
criterion, so that V ((S)) is necessarily closed as it is a union of finitely many closed
sets.
To use the recommended counterexample, if we take the quotient field localisation
Z Q, then S is not finitely generated as a multiplicative set; it is generated by the
set of (positive) primes in Z. We know that Spec Q = {(0)}, and this point is not
closed in Spec Z. If it were, then we would have (0) = V (I) for some I Z. But the
only choice for I is (0) itself, but V (I) = Spec Z in this case. The issue arose here
because Spec Z was a countable union of points, and we took for V ((S)) a countable
union of closed sets, which is not necessarily closed.
(b) Let us compared closed subsets, as was suggested. If we look at a closed set V (a)
Spec B/I (in the inherent Zariski topology), then since we can pull back a to an ideal
of B, we have V (a) = V (a + I) in the subspace topology. Since V (a + I) is closed
Exercise 3.4.C, we see that the basis of the Zariski topology on Spec B/I comes from
the basis elements of the Zariski topology on Spec B contained in Spec B/I. Therefore
the topologies agree.
38

Similarly, if we have V (a) Spec S 1 (B), then we know that a is still an ideal in
Spec B, so the correspondence is even clearer in this case. All we require is that V (a)
in the localised spectrum ignores the prime ideals intersecting S, i.e. we intersect with
Spec S 1 B.

Exercise 3.4.J.

Suppose I B is an ideal. Show that f vanishes on V (I) if and only if f I.

Solution.

T
We know that V (I) = V ( I), and that I = IpA p. Therefore if f vanishes on V (I),

then we have f p for all primes p I, so that f I. Conversely, if f I, then f p

for all I p, so f vanishes on V (I).
Exercise 3.4.K.
Describe the topological space Spec k[x](x) .
Solution.
We know that the set of points we have is just (x) and (0). We know that the point (x) is
closed, so the point (0) is open. The point (0) is not closed, however, so we know that (x) is
not open. Therefore the only open sets are Spec k[x](x) , (0), and . This is a 2-point space
with neither the indiscrete nor the discrete topology.

3.5

A base of the Zariski topology on Spec A: Distinguished open

sets

Exercise 3.5.A.
Show that the distinguished open sets form a base for the (Zariski) topology.
Solution.
We define
D(f ) = {[p] Spec A : f
/ p}.
This is equivalent (as a subset of Spec A) to Spec Af , as we saw in Exercise 3.4.I. Take some
S A a subset. Suppose that x V (S). Then for each f S, we must have x
/ D(f ).
Rephrasing this in terms of open sets, since we take the sets Spec A \ V (S) = V (S)c to be
our topology, we have x V (S)c if and only if x D(f ) for some (perhaps many) f S.
That is,
[
V (S)c =
D(f ).
f S

Therefore the D(f ) are a good base for our topology.

Exercise 3.5.B.
S
Suppose fi A for i J some index set. Show that iJ D(fi ) = Spec A if and only if
(fi ) = A, P
or equivalently and very usefully, there are ai for i J, all but finitely many 0,
such that iJ ai fi = 1.
39

Solution.
Let S be the set of all the fi . Then we know
[
D(fi ) = V (S)c = Spec A \ V (S).
iJ

p
Then this union is Spec A if and only if V (S) = . Since V (S) = V ( (S)) (to be as general
as possible), this means that S must generate all of Spec A. We know that S generates all
of Spec A if and only if 1 (S), which is our equivalent criterion above since the span of S
contains only finite linear combinations.
Exercise 3.5.C.
S
Show that if Spec A is an infinite union of distinguished open sets jJ D(fj ), then in fact
0
it is a union
S of a finite number of these, i.e. there is a finite subset J J such that
Spec A = jJ 0 D(fj ).
Solution.
Because we know that 1 A is expressible as a finite linear combination of some fj1 , . . . , fjn
(with nonzero coefficients), then in fact the ideal generated by J 0 = {fji : i = 1, . . . , n} is all
of Spec A.
Exercise 3.5.D.
Show that D(f ) D(g) = D(f g).
Solution.
Suppose p D(f ) D(g). Then we have f
/ p and g
/ p. If f g p, then this contradicts
the primality of p, so we have p D(f g). Conversely, suppose f g
/ p. Then if f p, we
would need f g p since p pA. The same problem holds for g p. This shows a double
inclusion.
Exercise 3.5.E.
Show that D(f ) D(g) if and only if f n (g) for some n 1, if and only if g is an invertible
element of Af .
Solution.
We can show these in order. We know that
p we can identify D(f ) with Spec A \ V ({f }). We
know that V ({f }) = V (If ), where If = (f ). We define Ig similarly. Hence suppose that
D(f ) D(g). Then this means V (If ) V (Ig ), which occurs when If Ig . In particular,
f Ig so there exists n N so that f n (g).
Now supposing that f n (g), i.e. f Ig , consider the element g Af . Let f n = ag.
Then since we are inverting {1, f, f 2 , . . .}, we have that
1=

fn
a
= n g,
n
f
f

so g A
f.
40

Finally, if g is invertible in Af , then let g 1 = a/f n for some n. Then since g (a/f n ) = 1,
we have g a = f n . Then suppose that r If . Then there is m N such that rm = bf (f )
for some b A. Since (bf )n = bn f n (g), we have
rmn = (bf )n = abn g (g),
then r Ig . Hence If Ig , which completes the proof.
Exercise 3.5.F.
Show that D(f ) = if and only if f N.
Solution.
If f N, then f p for all primes p by the oft-quoted theorem about the nilradical, so
D(f ) = . Conversely, if D(f ) = , then f p for all primes p, whence f N.

3.6

Topological (and Noetherian) properties

Exercise 3.6.A.
If A = A1 A2 An , describe a homeomorphism Spec A1 t Spec A2 t t Spec An
Spec A for which each Spec Ai is mapped onto a distinguished open subset D(fi ) of Spec A.
Thus
n
n
Y
G
Spec
Ai =
Spec Ai .
i=1

i=1

Solution.
Solving this seems to imply that Spec commutes with coproducts, which is pretty nice.
We certainly could reduce to the case where n = 2, since the general case follows by
induction on this general principal. We need to understand what the images of Spec A1 and
Spec A2 are in Spec A. Specifically, we map primes p A1 to p A2 and primes q A2 to
A1 q; this is the map induced by the projection of A onto its components. Further, this
map as defined is injective, so we need only show it is surjective.
Since we know that A Ai is surjective, a prime in A maps onto a prime in Ai (for
each i). Therefore the restriction to each component is a prime ideal, so it must be of the
form p q for a prime in A1 and a prime in A2 . But the ring A/(p q) is not a domain for
two proper prime ideals p and q: we have (1, 0) (0, 1) = (0, 0). The only way to eliminate
such a possibility is if p = A1 or q = A2 , which are not prime ideals per se in A1 and A2 but
their product is a prime in A. Hence the inclusion above is actually surjective as well, so
we have an isomorphism.
The hint given in the question is that Spec A1 = D(1, 0) and Spec A2 = D(0, 1). This is
certainly true: for the first case, we are allowed any prime ideal in the first coordinate and
allowed only A2 in the second coordinate. This completes the solution.
Exercise 3.6.B.
(a) Show that in an irreducible topological space, any nonempty open set is dense.
(b) If X is a topological space and Z (with the subspace topology) is an irreducible subset,
then the closure Z in X is irreducible as well.
41

Solution.
(a) Let U X be a nonempty open set. Then we know that X \ U is closed, and further
we can write
X = U (X \ U ).
Since X is irreducible, we must have either X \ U = X or U = X. Since U 6= by
assumption, we must have U = X, i.e. U is dense.
(b) Suppose that Z is reducible in X. Then let Z = Y1 Y2 for Yi closed. Since these
are closed subsets of a closed set in X, they are themselves closed in X. Therefore
Y1 Z and Y2 Z are closed in the subspace topology, showing that Z is reducible, a
contradiction.

Exercise 3.6.C.
If A is an integral domain, show that Spec A is irreducible.
Solution.
Suppose that Y and Z are closed sets which span Spec A. Then we have [(0)] Y without
loss of generality. Further, we know that we can write Y = V (I). Then since [(0)] Y , we
must have I (0), i.e. I = (0). Therefore Y = V ((0)) = Spec A, so we do not have a proper
union of Spec A as closed subsets.
Exercise 3.6.D.
Show that an irreducible topological space is connected.
Solution.
Suppose that Spec A was disconnected. Then there exist Y, Z Spec A which are open and
satisfy Y Z = Spec A, Y Z = . Since we have Y = Spec A \ Z, this means that Y is
closed, and similarly Z is closed. Hence Spec A = Y Z is a union of proper closed sets, so
it is reducible. This proves the contrapositive.
Exercise 3.6.E.
Give an example of a ring A where Spec A is connected but reducible.
Solution.
Consider A = C[x, y]/(xy). Its spectrum is a subspace of A2C whose points are the two
coordinate axes (points satisfying xy = 0). It is certainly reducible, since it is the union of
Spec C[x, y]/(x = 0) and Spec C[x, y]/(y = 0).
Now we will show it is connected. Because the Zariski topology is strictly coarser than
the analytic topology when it comes to the closed points of A2C (
= C2 ), a disconnection here
2
would give rise to a disconnection in C . But the coordinate axes are not disconnected in
C2 , so we must be connected in A2C as well.
Exercise 3.6.F.
(a) Suppose I = (wz xy, wy x2 , xz y 2 ) k[w, x, y, z]. show that Spec k[w, x, y, z]/I
is irreducible by showing that k[w, x, y, z]/I is an integral domain.
42

(b) Note that the generators of the ideal of part (a) may be written as the equations
ensuring that


w x y
rank
1,
x y z
i.e. as the determinants of the 2 2 submatrices. Generalise part (a) to the idea of
rank one 2 n matrices.
Solution.
(a) Vakil recommends we show that k[w, x, y, z]/I is isomorphic to a subring of k[a, b]
generated by monomials of degree 3. Let us attempt that. Call this subring S, and we
have S = k[a3 , a2 b, ab2 , b3 ]. Let : k[w, x, y, z] S be a map to be defined shortly,
and we will show that ker = I. Let (w) = a3 , (x) = a2 b, (y) = ab2 , and (z) = b3 .
Then we see (wz xy) = a3 b3 a2 b ab2 = 0, (wy x2 ) = a3 ab2 a4 b2 = 0,
and (xz y 2 ) = a2 b b3 a2 b4 = 0. Therefore I ker . To show the converse, we
claim that the three relations making up I are the only (linearly independent) relations
between the generators of S. For instance, wxz 3 y 5 = 0, but since we may substitute
y 2 = xz, we have
wxz 3 y 5 = wxz 3 x2 yz 2 = xz 2 (wz xy),
which we have already counted. So supposing that this is sufficient, we have ker = I.
Since S is a domain, we have k[w, x, y, z]/I is a domain as well. This makes its spectrum
irreducible.
(b) I believe it wants us to check the following situation:


x0 x1 xn1
rank
1.
x1 x2 xn
Perhaps this should be = 1. Let I be the ideal generated by the determinants of the
2 2 submatrices of the above matrix. Then k[x0 , . . . , xn ]/I
= Sn k[a, b], where Sn
is generated by degree n monomials. Is this enough? Maybe?

Exercise 3.6.G.
(a) Show that Spec A is quasicompact.
(b) Show that in general Spec A can have nonquasicompact open sets.
Solution.
(a) Note that Vakils definition of compact is that X is quasicompact if every open cover
of X has a finite subcover. Exercise 3.5.C tells us that if Spec A is a infinite union of
distinguished open sets D(fi ) for some index set i I, then there is a finite subset
S I such that Spec A is the union of the D(fi ) with i S.

43

S
Now consider any open cover of Spec
A
=
iI Ui . Since the distinguished open sets
S
are a base for Spec A, we have Ui = jIi D(fi,j ). Therefore we can write
[[
Spec A =
D(fi,j ).
iI jIi

Therefore there is a finite set S

iI

Ii which covers Spec A. The elements are

S = {fi1 ,j1 , . . . , fin ,jn }.

If we take those original open sets corresponding to i1 , . . . , in , then we have
Spec A =

n
[

D(fik ,jk )

k=1

n
[

Uik Spec A.

k=1

Therefore this choice is a finite subcover.

(b) The hint is to consider the non-Noetherian polynomial ring A = k[x1 , . . .] in countably
many variables, and the maximal ideal m = (x1 , . . .). Then the claim is that V (m)c is
not quasicompact.
Consider the open cover by the distinguished open sets D(xi ) for each i N. Certainly
we have xi m for each i, so we are not including the (only) closed point we must forbid.
Further, consider any point (ai ) in A
k (for lack of better notation). This corresponds
to the maximal ideal (xi ai ), which is in each of the D(xi ) such that ai 6= 0. Since
always some ai 6= 0 (because we are ignoring the choice (0, 0, . . .)), we have that D(xi )
covers V (m)c . We know this is sufficient because any higher-dimensional points (e.g.
curves) can be described in terms of their closed points.
This cover does not have a finite subcover, however. Suppose we remove (without
loss of generality) D(x1 ). Then using the above reasoning, we know that the point
(1, 0, . . .) does not lie in any other D(xi ). Therefore we must keep all D(xi ) in the
original description. Graphically, what is going on is that we are removing exactly one
coordinate axis in each of our open sets, and thus every open set is necessary.

Exercise 3.6.H.
(a) If X is a topological space that is a finite union of quasicompact spaces, show that X
is quasicompact.
(b) Show that every closed subset of a quasicompact topological space is quasicompact.
Solution.
S
(a) Let X = ni=1 Xi , where the Xi are quasicompact.
Then consider an open cover Uj of
S
X for j J some index set. Then certainly Uj is an open cover for Xi , whence we
get a finite subcover Ui1 , . . . , Uik . Doing this finitely many times, we get a collection
{U11 , . . . , U1k , U21 , . . . , Unk }.
In particular, this is a finite subcover of the original Uj which covers all the Xi . Therefore it also covers their union, so X is quasicompact.
44

(b) Let Z X be a closed subset, and let X be quasicompact. Then let Ui be an open
cover of Z X for i I some index set. Then we have an open cover of X of the form
S
Ui Z c . Therefore there exists a finite subcover V0 , V1 , . . . , Vn for X. We know that
one of these Vi must be Z c , as none of the Ui cover any points in Z c X. We take
this in the labelling above to be V0 . Then V1 , . . . , Vn cover (Z c )c = Z. Since these Vi
must come from the original Ui , we have our finite subcover, so X is quasicompact.

Exercise 3.6.I.
Show that the closed points of Spec A correspond to the maximal ideals.
Solution.
Let m A be a maximal ideal, and consider the closed set V (m) = {p Spec A : m p}.
Since m is a maximal ideal, m p implies that p = m or p = A. Since A is not in Spec A,
we know that V (m) = {m}, i.e. this is a closed point.
Conversely, suppose that {p} is not a maximal ideal. Then p m for some maximal
ideal m by Zorns lemma. If {p} were closed, then we would be able to write {p} = V (I) for
some ideal I. Since p m, we know that p V (I) implies that m V (I), so any closed set
containing p must also contain m. Therefore {p} is not closed.
Exercise 3.6.J.
(a) Suppose that k is a field, and A is a finitely generated k-algebra. Show that closed
points of Spec A are dense by showing that if f A and D(f ) is a nonempty (distinguished) open subset of Spec A, then D(f ) contains a closed point of Spec A.
(b) Show that if A is a k-algebra that is not finitely generated, the closed points need not
be dense.
Solution.
(a) First, we may suppose that f
/ N, otherwise D(f ) = . In this case, Af (the
localisation of A at the set {1, f, f 2 , . . .} is nonzero, so it is itself a finitely generated
k-algebra. Further, Af has some maximal ideal pAf , where p D(f ). We claim
that this p is actually a maximal ideal, which would complete the proof. Consider the
quotient field
Af /pAf
= (A/p)f,
where f is the image of f in A/p. This is a finite field extension of k. We know that
A/p (A/p)f is a domain since p is prime. As such, A/p must actually be a field,
whence p is a maximal ideal of A.
(b) The hint is to examine Spec k[x](x) as in Exercise 3.4.K. This is not finitely generated
as a k-algebra, because we must invert (at least) all elements of the form (x a), which
is an infinite set of k is an infinite field. It consists of two points (x) and (0), where (x)
is closed but (0) is not closed. (x) is therefore the only closed points, and in particular
the open set (0) does not contain a closed point, whence {(x)} Spec k[x](x) is not
dense.
45

Exercise 3.6.K.
Suppose k is an algebraically closed field, and A = k[x1 , . . . , xn ]/I is a finitely generated
k-algebra with N(A) = {0}. Consider the set X = Spec A as a subset of Ank . The space Ank
contains the classical points k n . Show that functions on X are determined by their values
on closed points.
Solution.
We want to show that for different functions f and g on X, then f g is nonzero on some
closed point, and therefore we can distinguish different functions. We know that on the open
set D(f g), f g is nonzero by definition. Further, since N(A) = 0, we know that D(f g)
is nonempty for any choice of f 6= g. By part (a) of Exercise 3.6.J, there is some closed point
(of k[x1 , . . . , xn ] perhaps) p D(f g). Therefore f g is nonzero at p. But this point p
can be viewed in Spec A, so this proves what we would like.
Exercise 3.6.L.
If X = Spec A, show that [q] is a specialisation of [p] if and only if p q. Hence show that
V (p) = {[p]}.
Solution.
We know that
\

U=

V.

U V
V closed

In our case, we have that

{[p]} =

V (I).

Ip

By Exercise 3.4.C, we know that

P
V (Ij ) = V ( Ij ), so we need only to get a handle on
X
I=a
Ip

Certainly p p, so p a. Conversely, since elements of a are linear combinations of elements

in p, we have a p, so there are equal. Therefore we have proved that {[p]} = V (p).
As such, we have p q if and only if q V (p) = {[p]}, and we are done. This also gives
an easy proof of the fact that the closed points correspond to maximal ideals.
Exercise 3.6.M.
Verify that [(y x2 )] A2 is a generic point for V (y x2 ).
Solution.
We know that this point is generic if {[(y x2 )]} = V (y x2 ). But this is totally obvious
from the previous problem.
Exercise 3.6.N.
Show that every point x of a topological space X is contained in an irreducible component
of X.
46

Solution.
We have the description of irreducible components as maximal among the irreducible closed
subsets of X. Let F be the family of irreducible closed subsets of X containing x. Then to
apply Zorns lemma (and thus conclude that x is in some irreducible component), we need
to show that F is nonempty and that chains in F have an upper bound in F.
We know that x is contained in {x}, which is irreducible. If it were reducible, then there
would be some proper closed Z {x} with x Z, a contradiction. Therefore F is nonempty.
Therefore consider a chain Z1 Z2 with Zi F. Then
Z=

Zi

i=1

is still in F. To see this, suppose that Z were reducible for some A B. Then consider
Ai = A Zi and Bi = B Zi . These are closed sets such that Z = Ai Bi , whence we have
Ai = or Bi = . Since this must hold for all i, we see that A = or B = . Therefore
Z is not actually reducible. Therefore this chain has an upper bound in F, so Zorns lemma
tells us that x is in some irreducible component.
Exercise 3.6.O.
Show that A2C is a Noetherian topological space.
Solution.
We need to show that A2C satisfies the descending chain condition on closed subsets. Recall
that V (I) V (J) if and only if I J. Hence given a sequence of closed subsets Z1 Z2
, we may rewrite this
V (I1 ) V (I2 ) I1 I2 .
Therefore we ask if C[x, y] satisfies the ascending chain condition on ideals, i.e. is it a
Noetherian ring. It is: if R is a Noetherian ring, then R[x] is Noetherian. Since C is
Noetherian (because it has only finitely many ideals), C[x] and hence C[x][y] = C[x, y] are
both Noetherian. (You can also do this explicitly, but why bother when you can preempt
the proceeding discussion about Noetherian rings?)
Exercise 3.6.P.
Show that every connected component of a topological space X is the union of irreducible
components. Show that any subset of X that is simultaneously open and closed must be the
union of the connected components of X. If X is a Noetherian topological space, show that
the union of any subset of the connected components of X is always open and closed in X.
Solution.
We know that irreducible topological space is connected from Exercise 3.6.D. Therefore the
connected component of any element x must contain every irreducible component it belongs
to, say some Zi . Then if we let Zx,i be the irreducible components containing x in some
index set Ix , we have
[ [
U=
Zx,i .
xU iIx

47

Therefore U is the union of these irreducible components.

Now, suppose that U is an open and closed set, and for any x U let Cx be the connected
component of x. We have the following two subsets of Cx : A = Cx U and B = Cx \ A.
Since Cx is closed and U is closed, A is closed, hence B is open. Further, since we can write
B = Cx U c , and U c is closed as well, then A is open. Therefore we have a disconnection
of Cx unless A = or B = . By assumption, x A so we must have B = , and hence
Cx U . Therefore we can write
[
U=
Cx .
xU

Now, if X is a Noetherian topological space, then we know we can write

X = Z1 Zn
for its (finitely many) irreducible components Zi . We know that each of these Zi are both
open and closed. Since we have shown that every connected component is a union of irreducible components, then every connected component is open and closed (since these must
be finite unions). Finally, every union of connected components is actually a finite union, so
it is still open and closed.

Exercise 3.6.Q.
Show that a ring A is Noetherian if and only if every ideal of A is finitely generated.
Solution.
Suppose that A is Noetherian. Suppose a A is an ideal that can only be generated by
infinitely many elements, say a1 , a2 , . . .. Then we have an ascending chain of ideals
(a1 ) ( (a1 , a2 ) ( (a1 , a2 , a3 ) (
that does not stabilise. This is a contradiction.
Conversely, if every ideal A is finitely generated, consider a chain
a1 a2
P
of ideals. Then a =
ai is also an ideal, hence is finitely generated by some elements
a1 , . . . , an . We know that ai ami for some indices mi , and so let m = max mi . Then
a = (a1 , . . . , an ) am , and since a am as well, we must have an equality. Therefore this
chain stabilises at am , so A is Noetherian.
Exercise 3.6.R.
Show that if A is Noetherian, then so is
A[[x]] := lim A[x]/xn

the ring of power series in x.

48

Solution.
We take the possible hint. Let a A[[x]] be an ideal. Let an a be the coefficients of
xn which appear in the elements of a. Then we claim that an is an ideal in A. Indeed, if
a, b an appear as coefficients of xn in a, then they correspond to some power series f (x)
and g(x). Then the coefficient of xn in (f + g)(x) (which is still a power series) is a + b, so
a + b an . Similarly, if r A is a scalar, then since r f (x) a, we know that the coefficient
of xn in this new power seris is r a an . Therefore an is an ideal.
Now, it is clear that an an+1 in the following way: since a is closed under left multiplication, we know that xa a. As such, any coefficient of xn in f (x) is also a coefficient of
xn+1 in the power series x f (x). Further, we know that the ideals ai completely determine
a: we have that

[
a=
ai xi .
i=0

Now, since a0 a1 is an ascending chain in A, it stabilises at some an . Further, each

of the ai are finitely generated for some ai,1 , . . . , ai,mi . Hence we see that a is generated by
{a0,1 , . . . , a0,m0 , a1,1 x, . . . , a1,m1 , . . . , ai,ji xi , . . . , an,mn xn }.
In particular, a is finitely generated. Since a was an arbitrary ideal, this shows that A[[x]]
is Noetherian.
Exercise 3.6.S.
If A is Noetherian, show that Spec A is a Noetherian topological space. Describe a ring A
such that Spec A is not a Noetherian topological space.
Solution.
I think we have shown this question sufficiently. Since descending chains of closed subsets
correspond to ascending chains of ideals, the stabilisation of one implies the other. The
ring C[x1 , . . .] will not give a Noetherian topological space because the ascending chain
of ideals (x1 ) (x1 , x2 ) corresponds to a nontrivial descent of closed subspaces in
Spec C[x1 , . . .].
Exercise 3.6.T.
Show that every open subset of a Noetherian topological space is quasicompact. Hence if A
is Noetherian, every open subset of Spec A is quasicompact.
Solution.
Let V U be an open subset, where U is a Noetherian topological space. Then consider
any chain of closed subsets V1 V2 in V . Then since V c is closed in U , we see
that Ui := Vi V c is closed in U . Hence we have a descending chain of closed subsets
U1 U2 in U . Since U is Noetherian, this terminates at some Un . That is, for any
N > n, we have Un = UN , i.e. V c Vn = V c VN . Therefore
V (V c Vn ) = Vn = VN = (V c VN ) V
for all N > n. Therefore the chain in V also stabilises, so V is Noetherian as well.
49

Exercise 3.6.U.
Show that if M is a Noetherian A-module, then any submodule of M is a finitely generated
A-module.
Solution.
This is the statement equivalent to A is Noetherian if and only if all its ideals are finitely
generated.
Consider N M a submodule that is not finitely generated. Let n1 N be any element.
Then An1 ( N is a proper ideal, so we have some n2 N so that An1 ( An1 + An2 ( N is a
proper chain of submodules. We may continue this process infinitely to obtain an ascending
chain of submodules in M that never terminates, else N would be finitely generated. This
is contradiction. Hence no such N can exist, and we have proved what we want.
Exercise 3.6.V.
If
0 M 0 M M 00 0
is exact, show that M 0 and M 00 are Noetherian if and only if M is Noetherian.
Solution.
Suppose that M is Noetherian. Then any ascending chain of submodules in M 0 injects
to an ascending chain of submodules in M , which must terminate. Hence M 0 is Noetherian. Further, any ascending chain of submodules in M 00 pulls back to an ascending chain
of submodules in M , which must terminate. Hence submodules and quotient modules of
Noetherian modules are Noetherian.
Now, suppose that M 0 and M 00 are Noetherian. Let N1 N2 be an ascending
chain in M . We will reconsider this chain in the following way. Since M/M 0
= M 00 , we know
0
0
0
00
0
that the submodules Ni := Ni M are viewable in M and Ni := Ni /Ni are viewable in
00
.
M 00 . These chains both terminate, say at some Nk0 and Nm
This shows the following: only finitely many of our original Ni are contained in M 0 , and
this first part of the sequence terminates. The second part of the sequence can be realised
by Ni = Ni00 + M 0 , and this sequence also terminates. Therefore ascending chain Ni must
stabilise somewhere.
Exercise 3.6.W.
Show that if A is a Noetherian ring, then An is a Noetherian A-module.
Solution.
By the above exercise, we have
0AAAA0
is a (split) exact sequence (viewed as A-modules), hence A2 is Noetherian. By induction
we have An is Noetherian for any n.
Exercise 3.6.X.
Show that if A is a Noetherian ring and M is a finitely generated A-module, then M is a
Noetherian module. Hence by Exercise 3.6.U, any submodule of a finitely generated module
over a Noetherian ring is finitely generated.
50

Solution.
If M is finitely generated, suppose that it can be generated by n elements m1 , . . . , mn . Then
we have an exact sequence

0 ker An M 0,
where (a1 , . . . , an ) = a1 m1 + + an mn . Then by the previous two exercises, An is
Noetherian, hence its quotient object M is Noetherian.

3.7

The function I(), taking subsets of Spec A to ideals of A

Exercise 3.7.A.
Let A = k[x, y]. If S = {[(x)], [(x 1, y)]}, then I(S) consists of those polynomials vanishing
on the y-axis, and at the point (1, 0). Give generators for this ideal.
Solution.
We know by definition that I(S) = (x) (x 1, y). Since these ideals are comaximal, we
have (x) (x 1, y) = (x) (x 1, y). Hence I(S) is generated by x(x 1) and xy.
Exercise 3.7.B.
Suppose S A3C is the union of the three axes. Give generators for the ideal I(S).
Solution.
Since A3C
= Spec C[x, y, z], it will be easiest to prove it using that. We know that the x-axis
corresponds to y = z = 0, so it is generated by y and z. Similarly, the y-axis is generated
by x and z and the z-axis by x and y. We therefore have
I(S) = (y, z) (x, z) (x, y) = (xy, xz, yz).
First, we see that (xy, xz, yz) I(S). Second, suppose that we had any polynomial in only
one variable as a generator, say f (x). Then f cannot vanish on the entire x-axis by basic
properties of polynomials, so this is a contradiction. All generators must be in at least two
variables, and indeed what we have is sufficient.
Exercise 3.7.C.
Show that V (I(S)) = S.
Solution.
Going strictly from the definitions, we know that

\
\
pq .
V (I(S)) = V
p A = q Spec A :

{p}S

{p}S

V (I(S)) is closed by definition and certainly S V (I(S)), so S V (I(S)). Now suppose

that a
/ S. Then because Spec A \ S is open, so we can construct a function f that is
nonzero at a but is zero on S. Then f I(S), so a
/ V (I(S)). Therefore S = V (I(S)).
51

Exercise 3.7.D.

Prove that if a A is an ideal, then I(V (a)) = a.

Solution.

The huge hint is Exercise 3.4.J, which says that f vanishes on V (a) if and only if f a.
But that is all we want to show, so weve done it already.
Exercise 3.7.E.
Show that V () and I() give a bijection between irreducible closed subsets of Spec A and
prime ideals of A. From this conclude that in Spec A there is a bijection between points
of Spec A and irreducible closed subsets of Spec A (where a point determines an irreducible
closed subset by taking the closure). Hence each irreducible closed subset of Spec A has
precisely one generic point.
Solution.
The theorem preceding this exercise gives a bijection between closed subsets and radical
ideals. Now, suppose that we take a prime ideal p A. Then V (p) = {q Spec A : p q}
is irreducible because we have shown V (p) = {[p]}. If we could write V (p) = V (a) V (b),
then we would have [p] V (a) (without loss of generality) and hence {[p]} V (a) and thus
V (a) = V (p). Therefore V (p) is irreducible.
T
Conversely, let Z be a closed irreducible subset. Then I(Z) = [p]Z p A. If this ideal
were not prime, then we have ab I(Z) such that both a, b
/ I(Z). Then we claim
Z = (V (a) Z) (V (b) Z),
which makes Z reducible, a contradiction. Indeed, we know that V (a) Z and V (b) Z are
both nontrivial since ab I(Z), but they cannot be all of Z since a, b
/ I(Z). Therefore we
have our contradiction.
The bijection between points of Spec A and irreducible closed subsets is clear.
Exercise 3.7.F.
A prime of a ring A is a minimal prime if it is minimal with respect to inclusion. If A is
any ring, show that the irreducible components of Spec A are in bijection with the minimal
primes of A. In particular, Spec A is irreducible if and only if A has only one minimal prime
ideal.
Solution.
We showed earlier that a point [q] is a specialisation of a point [p] if and only if p q.
We know too that the sets {p} determine all irreducible closed subsets (not components) of
Spec A. Therefore given any chain of prime ideals p1 p2 , the irreducible component
corresponding to all of them is determined by the bottom of the chain, i.e. the minimal
prime p1 . In particular, given a minimal prime p, any prime q containing it will have its
corresponding irreducible closed subset contained in the irreducible component given by
{[p]}. This is the bijection we would like.
Exercise 3.7.G.
What are the minimal primes of k[x, y]/(xy), where k is a field?
52

Solution.
The primes of k[x, y]/(xy) are those primes of k[x, y] containing xy. We know the primes of
k[x, y] are either the closed points (x a, y b), (f (x, y)) for an irreducible f k[x, y], and
the generic point (0). Now, since xy = 0 in k[x, y]/(xy), any prime ideal must contain either
x or y, i.e. (x) p or (y) p. Since (x) and (y) are themselves prime ideals in k[x, y] that
contain xy, these are prime ideals in k[x, y]/(xy). There are no other candidates for minimal
primes, for any nonzero p (x) (without loss of generality) would need to contain x or y.
y p is a contradiction and x p implies p = (x). Therefore (x) and (y) are the minimal
primes.

4
4.1

The structure sheaf, and the definition of schemes in

general
The structure sheaf of an affine scheme

Exercise 4.1.A.
Show that the natural map Af OSpec A (D(f )) is an isomorphism.
Solution.
The given hint is that D(f ) D(g) if and only if g is invertible in Af , which is Exercise
3.5.E. This is great for us. Since we have defined OSpec A (D(f )) to be the localisation at all
g A with D(f ) D(g), the natural map is clearly an isomorphism.
Exercise 4.1.B.
Make tiny changes to the proof of Theorem 4.1.2 to show base identity for any distinguished
open D(f ).
Solution.
We see that D(f ) = Spec Af pretty readily. We have D(f ) = {p Spec A : f
/ p}. Primes
in Af are precisely those in Af avoiding f . Therefore we just need to apply the above
argument to the ring Af , the main consequence being that Spec Af is quasicompact (which
is not true for a general open subset of Spec A!). The same argument then applies.
Exercise 4.1.C.
Alter the argument appropriately to show base gluability for any distinguished open set
D(f ).
Solution.
I think this works the same was as the previous exercise. The replacement of A by Af should
make everything follow pretty reasonably.
Exercise 4.1.D.
f on
Suppose M is an A-module. Show that the following construction describes a sheaf M
f(D(f )) to be the localisation of M at the multiplicative
the distinguished base. Define M
set of all functions that do not vanish outside V (f ). Define restriction maps resD(f ),D(g) in
the analogous way to OSpec A . Show that this defines a sheaf on the distinguished base, and
hence a sheaf on Spec A. Show that this is an OSpec A -module.
53

Solution.
f(D(f )) in a more coherent way. Indeed, we should have a natural
We hope to describe M
f(D(f )), where
isomorphism of A-modules Mf M
Mf := {m/f n : m M, n N}.
This is the usual localised module. We should able to prove identity and gluability on
f(Spec A) and let the specialisation to each D(f ) follow as they did in the previous two
M
S
exercises. Consider Spec A = D(fi ) and the finite refinement Spec A = D(f1 ) D(fn )
f(Spec A) such that resSpec A,D(f ) s = 0 for all fi . Since
(i.e. (f1 , . . . , fn ) = A). Suppose s M
i
N
s = 0 in Mfi implies that there is some N N such
that
f
s
=
0
for all i {1, . . . , n}.
i
Pn
N
N
N
Because (f1 , . . . , fn ) = A, we have ri A so that i=1 ri fi = 1, so that
!
n
n
X
X
N
s=
ri fi s =
ri (fiN s) = 0.
i=1

i=1

This is basically the same argument, because nothing depended on A being a ring, just an
A-module. The same is true for the gluability axioms, which I dont feel the need to re-type.
Now, we need to see why this is an OSpec A -module. However, this is perfectly clear: since
f(D(f ))
for each distinguished open set D(f ) we have OSpec A (D(f ))
= Af and M
= Mf , we
have a natural structure of Mf and as Af -module. This defines the appropriate OSpec A module structure.

4.3

Definition of schemes

Exercise 4.3.A.
Describe a bijection between the isomorphisms Spec A Spec A0 and the ring isomorphisms
A0 A.
Solution.
Abbreviate the structure sheaves by OA and OA0 . Let : Spec A Spec A0 be an isomorphism of schemes. This comes with a homeomorphism of topological spaces (which we denote
again) and an isomorphism OA0 OA . In particular, we know that we can recover our
rings from a global section: (D(10 ), OA0 ) = A0 (where 10 A0 is the multiplicative identity).
gives us an isomorphism
(D(10 ), OA0 ) (D(10 ), OA )
where D(10 ) Spec A0 . But what is the righthand side? By definition,
(D(10 ), OA ) = ( 1 (D(10 )), OA ) = (D(1), OA ).
Therefore gives us an isomorphism ] : A0 A on these global sections. This is one
direction of the correspondence.
We now need to show that the scheme isomorphism corresponding to the ring isomorphism corresponding to the scheme isomorphism is what we started with; i.e., yields a ring
54

isomorphism, and that ring isomorphism yields right back. A priori, let be the scheme
isomorphism coming from the ring isomorphism ] corresponding to . We need to show
that = on points of Spec A. Let p be a prime ideal in A. Then we need to show that
(p) = (p). Certainly (p) = ( ] )1 (p). Additionally, if we consider the stalk of OA corresponding to p, must have yielded an isomorphism of stalks OA0 ,p OA,p
= OA,1 (p) . In
1
particular, the point (p) (where we view as a homeomorphism of topological spaces)
should be the same point as ( ] )1 (p). Hopefully this is sufficient to prove = on points.
(FINISH LATER)

Exercise 4.3.B.
Suppose f A. Show that under the identification of D(f ) in Spec A with Spec Af , there is
a natural isomorphism of ringed spaces (D(f ), OSpec A |D(f ) )
= (Spec Af , OSpec Af ).
Solution.
The fact that D(f )
= Spec Af as topological spaces is already clear. We just need to show
that we have the appropriate isomorphism of sheaves. Consider a distinguished open set
D(g) Spec Af . Then since Spec Af is open in Spec A, D(g) is a distinguished open subset
of Spec A. As such, (D(g), OSpec A |D(f ) )
= (D(g), OSpec Af ). But thats all we wanted to
show. Im not sure what further details are necessary here.
Exercise 4.3.C.
If X is a scheme, and U is any open subset, prove that (U, OX|U ) is also a scheme.
Solution.
We can attack this problem locally. Its enough to consider the problem on affine schemes.
Further, we know that this type of restriction works for U = D(f ) for some distinguished
open set. Since any open subset U is just a union of D(fi ) for i I some index set, we can
glue together the reasoning above for D(f ) to make (U, OX|U ) a scheme.
Exercise 4.3.D.
Show that if X is a scheme, then the affine open sets form a base for the Zariski topology.
Solution.
Let U X be open. We know that every x U has a neighbourhood Ux that is isomorphic
to Spec Ax for some ring Ax . We know that this Ux (and its appropriate rings) is an affine
open subscheme of X, and we can realise
[
U=
Ux .
Therefore every open set is a union of these Ux . Therefore the set of all affine open subschemes
(which may be larger than this collection or all U ) is also a base for the topology.
Exercise 4.3.E.
(a) Show that the disjoint union of a finite number of affine schemes is also an affine
scheme.
55

(b) Show that an infinite disjoint union of (nonempty) affine schemes is not an affine
scheme.
Solution.
(a) The hint is to use Exercise 3.6.A, in which we proved that
n
G

Spec Ai
= Spec

i=1

n
Y

Ai .

i=1

So indeed, the disjoint union of finitely many affine schemes is also an affine scheme.
(b) We proved in Exercise 3.6.G(a) that affine schemes are quasicompact, so it is sufficient
to show that

G
Spec Ai = X
i=1

is not quasicompact. Indeed, we basically did this already. We know that Spec Ai X
is open for each i purely topologically, and thus we have an open cover of X
X=

Spec Ai .

i=1

Omitting any one of these will yield something less than X, so it cannot be refined to
a finite cover, whence X is not quasicompact. Therefore it cannot be an affine scheme.

Exercise 4.3.F.
Show that the stalk of OSpec A at the point [p] is the local ring Ap .
Solution.
It suffices to build up the stalk by looking at the base of the topology. We know that
[p] D(f ) if and only if f
/ p, and that OSpec A (D(f )) is A localised at all those g A
such that D(f ) D(g). Then we are inverting all elements that eventually are identified by
g
/ p, which is just the local ring Ap . This can probably be made more specific by saying
that we have a representative for each germ for every element not in p.
Exercise 4.3.G.
(a) If f is a function on a locally ringed space X, show that the subset of X where f
doesnt vanish is open.
(b) Show that if f is a function on a locally ringed space that vanishes nowhere, then f is
invertible.
Solution.

56

(a) Taking the hint, we would like to show that if f is a function on a ringed space X,
then the subset of X on which the germ of f is invertible is open. Indeed, if the germ
of f is invertible at a point p, then since we know that f is actually some (f, U ) for
p U , we know that f is invertible on all of U , making this an open condition.
If X is now a locally ringed space, we know that the value of f at p is its residue in
(p) = OX,p /mp . f being nonzero at p is equivalent to f
/ mp . In the case of a local

ring, we know that OX,p

= OX,p \ mp , which means f is nonvanishing at p if and only
if its germ is invertible in OX,p . By the first paragraph, we are done.
(b) If f vanishes nowhere, then we know that locally at any point p that f is invertible (by
the above). We have done something like this much earlier in this document as well.
As such, we should be able to glue together the inverses for f in around each point p
into a global inverse for f by the general properties of schemes.

4.4

Three examples

Exercise 4.4.A.
Show that you can glue an arbitrary collection of schemes together. That is, given a collection of schemes Xi for i I some index set, open subschemas Xij Xi with Xii = Xi ,
isomorphisms (of schemes) fij : Xij Xji with fi i = idXi , and the cocycle condition, show
that there is a unique scheme X up to unique isomorphism with open subsets isomorphic to
Xi which respects this gluing data.
Solution.
This is an absolutely miserable question to ask anyone. But if Vakil says it is essential, so
be it. We know that by Exercise 2.7.D, we can perform this exact process on sheaves. In
particular, we can do it on the structure sheaves of the schemes, so we probably dont have
to do a whole lot of work on that end. For the topological space that goes along with this,
however, we know that we want it to be
G .
X=
Xi (Xij = Xji )i,jI .
iI

This X is certainly a unique choice. We know that given our data, we can define each
sheaf OXi on the topological space X, which is induced by the Xi X as an open subset.
Throwing Exercise 2.7.D at this should do all the hard work for us, so we have the unique
up to unique isomorphism scheme that we wanted all along.
Exercise 4.4.B.
Show that the affine line with doubled origin is not affine.
Solution.
Let X be A1 with doubled origin. Then if it were affine, we would have X
= Spec A for
some A, and thus
A = (X, OX )
57

We need to figure out what the ring of global sections is. We construct X from two copies
of the affine line from k[t] and k[u] via the gluing k[t, 1/t]
= k[u, 1/u] where u t. We
know that we can identify Spec k[u] X with D(t) and Spec k[t] X with D(u). Hence
X = D(u) D(t), and so global sections on X are those on D(u) and D(t) that agree on
D(u) D(t). Therefore really all we are looking at are the global sections on one of the
D(u), i.e.
(X, OX )
= (D(u), OX|D(u) ) = k[u].
Therefore if X were affine, it would be the spectrum of k[u], which is nonsense.
Exercise 4.4.C.
Do the same construction with A1 replaced by A2 . Describe two affine open subsets of this
scheme whose intersection is not an affine open subset.
Solution.
Well, should get that X = A2 with doubled origin can be written Spec k[x, y] t Spec k[s, t],
where we identify k[x, y, 1/x, 1/y]
= k[s, t, 1/s, 1/t] via x s and y t. Hence we have
D(s, t) = Spec k[x, y] and D(x, y) = Spec k[s, t], and as above we should be able to calculate
the ring of global sections to be k[x, y] as with the affine plane with no origin.
We further see that D(x, y) D(s, t) is isomorphic to the affine plane with no origin, so
we have the intersection of affine open subsets which is not affine, since Vakil has proven
that the affine plane with no origin is not affine itself.
Exercise 4.4.D.
Check that the gluing information in the construction of Pnk works on triple intersections.
Solution.
I think what we have to do is the following. We are trying to glue together open sets
U0 , . . . , Un which are made by
Ui := {[x0 , x1 , . . . , xn ] : xi 6= 0} viewed like {(x0/i , . . . , xc
i,i , . . . , xn/i )},
where xi/j = xi /xj . We know this works when xi 6= 0, and really the ratios are all that
matter in the projective case. We can identify Ui = Spec k[x0/i , . . . , xn/i /(xi/i 1) if we view
these as proper variables but we dont really care about xi/i since it will equal 1 for practical
purposes. The gluing in question is gluing (somewhat augmented for technical reasons) Ui
to Uj along D(xj/i )
= D(xi/j ) via xk/i = xk/j /xi/j and xk/i = xk/j /xj/i .
Now were expected to write this down for triple overlaps. First, what is the triple
intersection Ui Uj Uk ? It is those values such that xi , xj , xk 6= 0. Given all the dividing
were doing, it should just looks like the spectrum of k[x0 , . . . , xbi , xbj , xbk , . . . , xn ]. Therefore
any gluing information should do fine in the intersection as it is an affine scheme. This
is what Vakil claims, but I think this construction is nice enough that writing more is
counterproductive.
Exercise 4.4.E.
Show that the only functions on Pnk are constants ((Pnk , O)
= k), and hence that Pnk is not
affine if n > 0.
58

Solution.
It looks like were supposed to mostly talk our way through this. We know that we need
to look at functions defined on each Ui that agree on all intersections. But because of what
weve been doing this whole time, any rational functions we might thing to include have
to exclude one variable for each of the Ui , which ultimately exhausts all the variables at
our disposal. Hence (Pnk , O) is k[x0 , . . . , xn ] but without any polynomials containing any
variables, i.e. just k itself.
Exercise 4.4.F.
Show that if k is algebraically closed, the closed points of Pnk may be interpreted in the
traditional way: points are of the form [a0 , . . . , an ] where not all ai are zero and [a0 , . . . , an ]
is identified with [a0 , . . . , an ] for all k .
Solution.
We have glued together n + 1 copies of affine space Ank in a particular way to make Pnk , so
the only closed points in there should have been closed in affine space to begin with. By the
Nullstellensatz, the closed points in Ank for k algebraically closed are what we expected. The
gluing that we did identifies the orbits of the k action on this union of points. Therefore
our intuition is sound.

4.5

Projective schemes, and the Proj construction

Exercise 4.5.A.
Consider P2k with projective coordinates x0 , x1 , x2 . Think through how to define a scheme
that should be interpreted as x20 + x21 x22 = 0 in P2k .
Solution.
It would be great to have this question asked more firmly, but it is what it is. The hint
tells us that if we restrict to the affine scheme U2 P2k , this scheme will be cut out by
x20/2 + x21/2 1 = 0, whence it should be the scheme Spec k[x0/1 , x0/2 ]/(x20/2 + x21/2 1).
Similarly, we should have the restriction to U1 we should get Spec k[x0/1 , x2/1 ]/(x20/1 1+x22/1 )
and for U0 we should get Spec k[x1/0 , x2/0 ]/(1 + x21/0 x22/0 ). This gives us something that
looks vaguely circular in two copies of affine space and something that looks hyperbolic in
the third copy.
I have been advised not to deal with questions of gluing as much as I can, so I will avoid
this question.
Exercise 4.5.B.
Consider PnA with projective coordinates x0 , . . . , xn . Given a collection of homogeneous
polynomials fi A[x0 , . . . , xn ], make sense of the scheme cut out in PnA by the fi .
Solution.
The hint is to use the same procedure as 4.4.D, which (of course) we didnt really do. We
can realise the scheme cut out by these functions by restricting to the appropriate notion in
the affine spaces U0 , . . . , Un and note that we can still glue them together.
59

To try to make this precise, given a single function f (x0 , . . . , xn ), we can look at the
functions f (x0/i , . . . , xn/i ) which live in k[x0/i , . . . , xn/i ]. We then can take the quotient of
this ring by (xi/i 1) as in 4.4.9 to get us back to affine space, and now we know how to
look at the vanishing set of f (x0/i , . . . , xc
i/i , . . . , xn/i ) in this space. We then go through the
same effort of gluing everything together.
This is a pretty sketchy explanation, but it should suffice for the moment.
Exercise 4.5.C.
(a) Show that an ideal I is homogeneous if and only if it contains the degree n piece of
each of its elements for each n.
(b) Show that homogeneous ideals are closed under sum, product, intersection, and radical.
(c) Show that a homogeneous ideal I S is prime if I 6= S , and for any homogeneous
a, b S, if ab I, then a I or b I.
Solution.
(a) First, assume I is homogeneous. By definition, it is generated by homogeneous elements
{ai } for i A some index set. First, consider an
P element of the form s a for one of
the generators a. Assume that a Sm and s = jZ sj is the decomposition of s into
homogeneous elements. Then the degree n piece of s a is snm a. This is just a
multiple of one of the generators, so it is contained in I.
P`
An arbitrary
element
a

I
is
of
the
form
a
=
i=1 si ai , where ai Smi and
P
si = jZ si,j is the decomposition into homogeneous elements. Then the degree n
P
part of a is `i=1 si,nmi ai , which is contained in I as it is again a sum of multiples
of the generators. This proves the forward direction.
For the converse, suppose that I contains the degree n piece of each of its elements.
Consider a generating set {ai } of I for i A some index set. Then the degree n part
of each of the ai is also contained in I for each n Z, so we can construct a new
generating set
(
)
X
ai,n : i A, ai,n Sn , ai =
ai,n .
nZ

This is just taking each generator to its homogeneous decomposition. This is still a
generating set for I, so I is generated by homogeneous elements, i.e. I is homogeneous.
(b) Let I, J be homogeneous ideals. Let I be generated by {ai } and J by {bj }. Then I + J
is generated by {ai , bj } all of which are homogeneous so I + J is homogeneous. I J is
generated by {ai bj }, all of which are still homogeneous, so I J is homogeneous.
Now, consider I J. If a I J, then thePdegree n part of a is contained in both
I and J as well, so an I J (where a = nZ an in the usualway). By part (a),
this means that I J is homogeneous. Similarly, we know that I = {a S : am
I for some m N}. If am I, then we know that the degree n part of am is contained
in I, which we denote (am )n . It would be great to say (am )n = (an )m , but this is
patently false. However,
let ak be the highest degree
piece of a. Then we know that
m
m
(a )mk = (ak ) , so ak I. Therefore a a
I as well. By induction, this shows
k
that every homogeneous piece of a is also in I.
60

(c) Suppose that I satisfies ab I implies a I or b I for homogeneous elements. Let

a, b S be two arbitrary elements such that ab I. Since I is homogeneous, we know
that the homogeneous
piecesPof ab are all contained in I, which we denote (ab)n . If
P
we write a =
an and b =
bn as the individual decompositions into homogeneous
elements, we know that
X
(ab)n =
ai b j .
i+j=n

If a
/ I, then there exists some component ad of a that is not in I, and similarly if
b
/ I there is some be
/ I. Without loss of generality, we may take d and e to be the
maximal degree satisfying this property.
Now, consider
X

(ab)d+e =

ai b j = ad b e +

i+j=d+e

ai b j .

i6=d,j6=e

Then in the rightmost sum, we have i > d or j > e in each term, so each term is a
multiple of an element of I. Therefore
X
(ab)d+e
ai bj I.
i6=d,j6=e

But this is precisely ad be . However, we assumed that ad , be

/ I, which violates the
assumption on I. Hence we contradict that a
/ I and b
/ I, so a I or b I, so I is
prime.

Exercise 4.5.D.
Note: all graded rings are now assumed to contain no elements of negative degree henceforth,
i.e. they are all Z0 -graded.
(a) Show that a graded ring S over A is a finitely generated graded ring if and only if S
is a finitely generated graded A-algebra, i.e. generated over A = S0 by a finite number
of homogeneous elements of positive degree.
(b) Show that a graded ring S over A is Noetherian if and only if A = S0 is Noetherian
and S is a finitely generated graded ring.
Solution.
L
(a) If S is a finitely graded ring, then the ideal S+ = n>0 Sn is finitely generated. Let
s1 , . . . , sk be a generating set for this ideal. Without loss of generality, we can assume
that all these elements are homogeneous, with si Smi . We know further that mi > 0
for all i. The claim is that this set generates S as an A-algebra.
Let s S+ . Then we can write
s=

k
X
i=1

61

ci s i

for some elements ci S . For simplicity, we can assume that s is also a homogeneous
element, because we can always take the original problem and break it up into homogeneous pieces for consideration. Suppose s Sn . Then we can choose each ci to be
homogeneous of degree n mi . This is because of ci is not homogeneous, then ci si
will have support in a degree not equal to n, which must be zeroed by the rest of the
elements. If we just eliminate the parts of the ci which are not in degree n mi for
each i, then we do not change the net result.
If mi > n, then ci = 0, and if mi = n then ci A. If we have n > mi , then ci S+ , so
in turn we may write
`
X
ci =
di s i .
j=1

We now have the same situation: each di is (after trivial adjustment) homogeneous
of degree (n mi ) mi . If 2mi > n, then di = 0, and if 2mi = n then di A.
By induction, since mi > 0, we eventually express every coefficient as an A-linear
combination of products of the si . This is contained in the A-algebraic span of the
si , so this set si suffices to generate S+ as an A-algebra. Since we can get 1A by the
empty sum of the si , we can see that si span S as an A-algebra.
The converse is trivial. If S is finitely generated as an A-algebra, then any element of
S+ S may be written
k
X
s=
ci s i
i=1

for some generating set s1 , . . . , sk and elements ci A. This means that s1 , . . . , sk

generate S+ , so this ideal is finitely generated.
(b) If S is Noetherian, then S+ is an ideal of S , so is Noetherian. A = S0 = S /S+ ,
so it is a quotient object of a Noetherian ring, hence is Noetherian. Further, if S is
Noetherian, all its ideals are finitely generated, so S+ is finitely generated, hence S is
finitely generated as a ring.
Conversely, if S is finitely generated as a ring and A is Noetherian, by part (a) we
know that S is finitely generated as an A-algebra. We have shown before that a finitely
generated algebra over a Noetherian ring is still Noetherian, so S is Noetherian.

Exercise 4.5.E.
Suppose f S+ is homogeneous.
(a) Give a bijection between the primes of ((S )f )0 and the homogeneous prime ideals of
(S )f .
(b) Interpret the set of prime ideals of ((S )f )0 as a subset of Proj S .
Solution.
62

(a) Vakil immediately tells us to prove something else, so we will. Suppose A is a Z-graded
ring with a homogeneous invertible element f in positive degree. Then we should show
there is a bijection between prime ideals of A0 and homogeneous prime ideals of A.
This is actually what we want to do, but we are making our notation more polite.
Let : A0 A be the inclusion ring homomorphism. Then we know that if p A is a
homogeneous prime ideal, then 1 (p) is a prime ideal of A0 . We now need to reverse
the correspondence. We follow Vakils advice. Let p0 A0 be a prime ideal. Define
the subset p A (which wed like to be a homogeneous prime ideal eventually) as the
direct sum of qi , where
qi = {a Ai : adeg f /f i p0 }.
In particular, q0 = p0 , so if we are successful, 1 (p) = p0 , which would prove that we
have a bijection.
We will prove properties as Vakil suggests them: suppose that a qi . We claim that
a2 q2i . Indeed, a qi means that adeg f /f i p0 . Then
adeg f /f i adeg f /f i = a2 deg f /f 2i = (a2 )deg f /f 2i p0 .
Conversely, if a2 q2i , then
adeg f /f i

2

p0 = adeg f /f i p0

because prime ideals are radical.

Now, given a1 , a2 qi , we know that
(a1 + a2 )2 = a21 + 2a1 a2 + a22 q2i .
The square elements we have just shown are in q2i , and the mixed element satisfies
f
f
(a1 a2 )deg f /f 2i = adeg
/f i adeg
/f i p0 .
1
2

Therefore a1 + a2 qi . All this shows that qi Ai are ideals. As such, p = qi

is a homogeneous ideal of A. We claim that it satisfies the homogeneous prime ideal
criterion we have been using for a while. Suppose that a, b A are homogeneous
elements, so without loss of generality a Am , b An . Further suppose that ab p.
In particular, ab qn+m . Hence
(ab)deg f /f m+n = adeg f /f m bdeg f /f n p0 .
Since p0 is a prime ideal, either adeg f /f m or bdeg f /f n is in p0 . Therefore a qm or
b qn , so in particular a p or b p. This shows that p is a homogeneous prime
ideal, so we have our correspondence.
(b) The points of Proj S are those homogeneous prime ideals of S which do not contain
S+ . We know that prime ideals of ((S )f )0 correspond to homogeneous prime ideals of
(S )f . We know that (homogeneous) prime ideals of (S )f correspond to (homogeneous)
prime ideals of S which do not contain f . An ideal not containing f certainly cannot
contain S+ , so this is the subset we are looking for.
63

Exercise 4.5.F.
Show that D(f ) is (or more precisely, corresponds to) the subset Spec((S )f )0 from
4.5.E(b).
Solution.
We have defined D(f ) = Proj S \ V (f ), where V (f ) is the set of homogeneous prime ideals
of S containing f but not containing S+ . Therefore Proj S \V (f ) is the set of homogeneous
prime ideals of S which do not contain f (nor S+ ). By the above, this correspondence is
clear.
Exercise 4.5.G.
Verify that the projective distinguished open sets D(f ) (as f runs through the homogeneous
elements of S+ ) form a base of the Zariski topology.
Solution.
We want to show that any open set U Proj S is a union of the D(f ). We know that the
complement of U is some closed set V (I), where I is a homogeneous ideal contained in S+ .
We know I is generated by some positive degree homogeneous elements ai for i A some
index set. We know that
\
V (I) =
V (ai ),
iA

because any ideal containing all the ai will contain I and vice versa. Hence
\
[
[
U = Proj S \ V (I) = Proj S \
V (ai ) =
Proj S \ V (ai ) =
D(ai ).
iA

iA

iA

This completes the proof.

Exercise 4.5.H.
Fix a graded ring S .
(a) Suppose I is any homogeneous ideal of S contained in S+ , and f is a homogeneous
element of positive degree. Show that f vanishes on V (I) (i.e. V (I) V (f )) if and
only if f n I for some n.
(b) If Z Proj S , define I(Z) S+ . Show that it is a homogeneous ideal of S . For any
two subsets, show that I(Z1 Z2 ) = I(Z1 ) I(Z2 ).
(c) For any subset Z Proj S , show that V (I(Z)) = Z.
Solution.
(a) We have done this in the affine case, so we just need to bring it over to our new
situation. In the affine case,
showed that f vanishes on
which was Exercise 3.4.J, we
V (I) if and only if f I. In our case, we know that I is still the intersection
of all prime ideals containing I (since that result was general).
Thereforeif p is an

appropriate prime ideal containing

( I). Since
I, it will also contain I, i.e. V (I) = V
V () still reverses inclusion, V ( I) = V (I) V (f ) if and only if {f } I.
64

(b) Analogous to the affine case, we should have I(Z) S+ defined by

\
I(Z) =
p.
[p]Z

That is, it should be the intersection of homogeneous prime ideals not containing S+
on which Z vanishes. We know that homogeneity is preserved under intersection,
and the proof above (Exercise 4.5.C) works for arbitrary intersection, so this ideal is
homogeneous. Now,

!
\
\
\
I(Z1 Z2 ) =
p=
p
q = I(Z1 ) I(Z2 ).
[p]Z1 Z2

[p]Z1

qZ2

This is all we wanted to show.

(c) We have shown this in the affine case (Exercise 3.7.C). We know that V (I(Z)) is
/ Z, then because
a closed set which contains Z and so contains Z. Further, if a
Proj S \ Z is open, we can find some function f which is nonzero at a but zero on
Z, so f I(Z) and hence a
/ V (I(Z)). This gives the required double inclusion, so
Z = V (I(Z)).
This reasoning should still work even though we dont know were working with a
scheme yet.

Exercise 4.5.I.
Fix a graded ring S and a homogeneous ideal I. Show that the following are equivalent.
(a) V (I) = .
(b) For any fi (as i runs through some index set) generating I,

(c) I S+ .

D(fi ) = Proj S .

Solution.

We have
already shown V (I) = V ( I). If V (I) = , then there isno prime ideal which
contains I but does not contain S+ . This is obviously the case if I S+ , which shows
(c) implies (a).
If we assume (a), then we have shown above that if fi generate I, then
[
D(fi ) = Spec S \ V (I).
Since V (I) = , we have (b).
Finally, if (b) holds, then we know V (I) = 0. This shows that any prime ideal containing
I contains S+ . Therefore we have
\

I=
p S+ .
Ip

This completes the proof.

65

Exercise 4.5.J.
Suppose some homogeneous f S+ is given. Via the inclusion
D(f ) = Spec((S )f )0 , Proj S+
of Exercise 4.5.F, show that the Zariski topology on Proj S restricts to the Zariski topology
on Spec((S )f )0 .
Solution.
It is easy to check that they have the same distinguished open sets. Consider g ((S )f )0
and the open set D0 (g) in Spec((S )f )0 . We want to show this is compatible with D(g).
D0 (g) will be the set of prime ideals not containing g in ((S )f )0 , which corresponds to the
set of homogeneous prime ideals of (S )f not containing g. But this is just D(g) D(f ), so
we have the appropriate restriction.
Exercise 4.5.K.
If f, g S+ are homogeneous and nonzero, describe an isomorphism between Spec((S )f g )0
and the distinguished open subset D(g deg f /f deg g ) of Spec((S )f )0 .
Solution.
We know that Spec((S )f g )0 is the set of homogeneous prime ideals of S not containing
f g. Suppose p is a homogeneous prime ideal not containing f . Then p will also not contain
f g if and only if it does not contain g by the properties of primes. To describe D(g) in
Spec((S )f )0 requires changing the degree of g to zero, i.e. by examining instead g deg f /f deg g
so that deg(g deg f /f deg g ) = deg g deg f deg f deg g = 0. This at least gives us a bijection
on points. It should extend to an isomorphism of topological spaces trivially.
Exercise 4.5.L.
By checking that these gluings behave well on triple overlaps, finish the definition of the
scheme Proj S .
Solution.
Absolutely not. See Exercise 2.7.D if youd like to torture yourself.
Exercise 4.5.M.
(Re)interpret the structure sheaf of Proj S in terms of compatible stalks.
Solution.
(FINISH LATER)
Exercise 4.5.N.
Check that the definition PnA := Proj A[x0 , . . . , xn ] agrees with our earlier construction.
Solution.
It should agree fine. The distinguished open sets we will glue together are D(xi ), which are
just homogeneous prime ideals that avoid the various xi . They glue together here just as
they glued together before. The question Vakil gives is that how do we know this covers
everything? By construction before, we had homogenised our issue, so we know we need
only look at the homogeneous prime ideals.
66

Exercise 4.5.O.
Suppose that k is an algebraically closed field. We know from Exercise 4.4.F that the closed
points of Pnk are in bijection with the points of classical projective space. By Exercise
4.5.N, the scheme Pnk is isomorphic to Proj k[x0 , . . . , xn ]. Therefore, each point [a0 , . . . , an ] of
classical projective space corresponds to a homogeneous prime ideal of k[x0 , . . . , xn ]. Which
homogeneous prime ideal is it?
Solution.
In the affine case, we had the correspondence
(a1 , . . . , an ) (x1 a1 ), . . . , (xn an ).
Of course these ideals are not homogeneous. Instead, if we assume we have chosen a0 6= 0,
we can associate




a1
an
(a0 , . . . , an ) x1 x0 , . . . , xn x0 .
a0
a0
These are homogeneous prime ideals and should give us what we want.
Exercise 4.5.P.
If S is generated in degree 1, and if f S+ is homogeneous, explain how to define V (f )
in Proj S , the vanishing scheme of f . Hence define V (I) for any homogeneous ideal I of
S+ .
Solution.
I have no idea what Vakil wants here. It should be the set of points on which f vanishes. If
you imagine this with classical points, V (f ) should look like all those classical points such
that f (a0 , . . . , an ) = 0, albeit with more information.
To define V (I), take a generating set fi for I and then take unions.
Exercise 4.5.Q.
Suppose k is algebraically closed. Let V be a (n + 1)-dimensional k vector space. Describe
a natural bijection between one-dimensional subspaces of V and the closed points of PV .
Solution.
We define PV to be Proj(Sym V ), where Sym V is the symmetric algebra of V . We
are told that Sym V = k[x0 , . . . , xn ] (if we view x0 , . . . , xn as a dual basis to V ) which is
convenient for us.
We know that the closed points of PV will then be defined by homogeneous coordinates
[a0 , . . . , an ]. We can view this as a unique basis vector for a one-dimensional subspace of V ,
since no two different closed points will be scalar multiples. This is our bijection.

5
5.1

Some properties of schemes

Topological properties

Exercise 5.1.A.
Show that Pnk is irreducible.
67

Solution.
Suppose Pnk is reducible, so we could write Pnk = Y Z. Then we know Y = V (I) and
Z = V (J) for two homogeneous ideals I and J of elements of positive degree. Then since
[(0)] Y (without loss of generality), we know that I (0) so I = (0). But this means that
Y = V ((0)) = Pnk , which is a contradiction.
Exercise 5.1.B.
Show that there is a bijection between irreducible closed subsets and points for general
schemes.
Solution.
S
We showed this in Exercise 3.7.E for affine schemes. Let X = Ui , where the structure
sheaf restricted to each Ui is isomorphic to Spec Ai for some ring Ai . Since each point of X
lies in some Ui , we know that to each point p X we can associate and irreducible closed
subset of Ui by taking the closure of p in Ui . Let Zi be this set.
LATER LATER LATER.
Exercise 5.1.C.
We need to show that X satisfies the descending chain property on closed subsets. Suppose
that Z1 Z2 is a descending chain. Then we can write
Zj =

n
[

Zj Spec Ai .

i=1

Denote Zj Spec Ai = Yi,j . Then Yi,j for j = 1, 2, . . . is a descending chain of (relatively)

closed subsets of Spec Ai , so it stabilises at some Yi,mj . Let m = max{m1 , . . . , mn }. Then
we know that Yi,` = Yi,m for all ` > m and for all i. Hence
Z` =

n
[
i=1

Yi,` =

n
[

Yi,m = Zm

i=1

for all m > `, so our original sequence stabilises. Hence X is Noetherian.

Exercise 5.1.D.
Show that a scheme X is quasi compact if and only if it can be written as a finite union of
affine open subschemes.
Solution.
S
We know that any scheme X can be written X = Ui , where Ui
= Spec Ai are affine open
subsets. If X is quasicompact, then this open cover admits a finite subcover, hence X is
covered by finitely many affine open subschemes.
Conversely, an affine scheme is quasicompact. Therefore if X is covered by finitely many
affine open subschemes U1 , . . . , Un , consider an open cover Vi of all of X. Then the set Vi Uj
is an open cover of each Uj , hence admits a finite subcover. Therefore from our initial cover
Vi , take only those Vi appearing as Vi Uj in one of these finite subcovers. This is a finite
collection of open sets which must cover X. Hence X is quasicompact.
68

Exercise 5.1.E.
Show that if X is a quasicompact scheme, then every point has a closed point in its closure.
Show that every nonempty closed subset of X contains a closed point of X. In particular,
every nonempty quasicompact scheme has a closed point.
Solution.
From Exercise 5.1.B, we know that there is a bijection between irreducible closed subsets
and points, in the sense that a point determines an irreducible closed subset by taking the
closure. Therefore we should show that every irreducible closed subset contains a closed
point.
If X is quasicompact, then it is a finite union of open affine subschemes. Let Z be an
irreducible closed subset of X.

69