A Brief Introduction to Schemes and Sheaves

David Urbanik
Last Updated: June 2019

1 Introduction
Scheme theory, perhaps more than any other subject, has a reputation for being extremely
difficult and tedious to learn. One gets the impression that the subject involves many highly
technical and difficult constructions, is exceedingly vast and abstract, and that it takes
considerable time and energy before one is able to prove anything of value. Quite famously,
the subject originated from Grothendieck’s attempt to “simplify” an eighty page paper by
Serre into the thousand page document that was to become Les Éléments de géométrie
algébrique — a fact that is both oddly remarkable and offers little encouragement.
It is perhaps somewhat surprising, then, that there seems to be no shortage of graduate
students and even undergraduates eager to devote time to understand schemes. The usual
procedure is to sit down with a copy of Hartshorne, formally sift through a seemingly endless
series of complex definitions, and then grudgingly admit defeat. Usually absent from these
attempts at understanding schemes are good sources of intuition, motivation, and clear and
identifiable goals. The result is that students learning the subject this way have difficulty
explaining the “point” of a definition or a construction, and so don’t know what it’s related
to, why it’s there, and consequently can’t use it.
The purpose of this article is to give the basic definitions of scheme theory in context.
We will take the view that it is just as important, if not more so, to explain the definitions
themselves as it is to explain the lemmas and the proofs. In doing so, we hope to remedy a
common affliction that befalls those who read Hartshorne’s book: not having any idea what
is going on.

2 Schemes are like Manifolds

Our exposition of schemes will be according to the following mantra: “a scheme is to a
variety as an abstract manifold is to an embedded submanifold of Euclidean space.” As our
goal is to explain the term “scheme”, and not “variety”, “abstract manifold” or “embedded
submanifold of Euclidean space,” we will assume that the reader understands and is com-
fortable with the latter three terms.1 By assuming this background, we hope to explain and
motivate many of the constructions in schemes by analogy, and thereby make the motivation
for the subject clear.
Our mantra, as stated, already poses an interesting question: what is the difference
between an abstract manifold (say smooth), and a submanifold of Euclidean space? The
1 Here when we say “understand varieties” we mean in the ordinary sense of algebraic sets defined by

polynomial equations, not in the sense of scheme theory.

1
Witney embedding theorem, which tells us that any smooth real manifold of dimension n
can be embedded in a Euclidean space of dimension 2n, would seem to suggest that there
is none; with the usual notion of equivalence, diffeomorphism, smooth manifolds are the
same as smooth submanifolds of Rm . The answer, it turns out, is that the purpose of the
smooth manifold construction is not to study objects which are different from submanifolds
of Euclidean space, but to formulate the subject in a way which emphasizes the features
of interest and avoids clouding the picture with artifacts corresponding to any particular
embedding. That is, smooth manifold theory provides a language in which one can study
manifolds, which once internalized, becomes more expressive than the more immediately
accessible one which describes submanifolds of Rm .
An illustrative example of this phenomenon is Maxwell’s equations. Maxwell originally
formulated electromagnetism with twenty equations. Modern vector notation brings this
down to four. The language of smooth manifold theory requires just two: letting F be a
2-form (the electromagnetic field) and J a 3-form (corresponding to current density), the
Maxwell equations on a smooth 4-dimensional Lorentzian manifold read

dF = 0 d ? F = µ0 J,

where d is the exterior derivative, and ? is the Hodge star operator. Unlike the vector calculus
version of the Maxwell equations, this modern formulation does not privilege any particular
coordinate system, and thus despite being in some sense equivalent to the vector calculus
version, it is actually more powerful. Of course, despite the apparent simplicity of these
equations, the theory needed to formulate them is significantly more involved than what’s
needed to state the same equations in their original form. What has been achieved is a kind
of simplicity at the expense of efficiency, and it is in precisely this sense that Grothendieck’s
development of scheme theory can be said to be a “simplification” of classical developments
in geometry and algebra.

3 Core Definitions
3.1 Sheaves
The usual procedure when defining any kind of geometric space consists of two steps. In the
first, one specifies a topological space, and in the second, one specifies a class of functions
which give it its structure. For instance, one can speak of differentiable manifolds, smooth
manifolds, analytic manifolds, or simply topological manifolds; each construction starts by
specifying a second countable Hausdorff topological space, and the constructions differ due
to the nature of the functions defined on the space. In this regard, the situation with
schemes is no different. Just like a smooth manifold structure can be specified by giving a
topological space and describing which functions are smooth, a scheme can be presented by
giving an appropriate topological space and describing which functions are regular.
The tool we will use for managing the regular functions on the space is called a sheaf.
Sheaves are general tools whose purpose is to define collections objects in some category (e.g.
Sets, Groups, Rings, or Modules) which are stitched together topologically. In our case, we
will have a space X, and to each open set U ⊂ X we will have a ring OX (U ) giving the
regular functions on U . We will want these rings to be related to each other by restriction
maps and by the fact that we can “glue” families of functions defined over a cover to get a
function on a larger open set; sheaves will encode this idea. More generally, sheaves can be

2
used to define things like line and vector bundles by specifying their spaces of sections over
any open set, and describing how those sections restrict to one another and glue together.
Useful classes of sheaves also have the property that they form highly-structured categories:
maps of sheaves can have kernels, images, and correspond to quotients. For this reason, they
are also useful in developing various cohomology theories, and this application was also the
subject of the aforementioned paper of Serre.
Before defining sheaves, we begin with the notion of a presheaf.
Definition 3.1. A presheaf F of rings associated to a topological space X consists of the
following data:
(i) To each open set U ⊂ X, a ring F(U ).

(ii) To each inclusion of open sets U ,→ V a map resV,U : F(V ) → F(U ) called the
restriction map from F(V ) to F(U ). These maps satisfy the property that resU,U =
idF (U ) and resV,U ◦ resW,V = resW,U where U ⊂ V ⊂ W are open sets.
If s ∈ F(U ), we call s a section of F over U .2

As an equivalent but more sophisticated formulation, one can consider the topological
space X as a category, whose objects are open sets of X and whose maps are inclusions of
open sets, and then define a presheaf on X as simply a contravariant functor from X to the
category Ring. Replacing Ring with another category one can get presheaves of groups,
modules, etc., but we will largely limit ourselves to rings in this article. When not otherwise
specified, we will use the term “presheaf” (and later “sheaf”) to refer to the case of rings.
Examples of presheaves can easily be found in manifold theory. If M is a C k -manifold, the
object C k (−) which associates to any open set U ⊂ M the k-times differentiable functions
C k (U ) on U is a sheaf, with restriction maps corresponding to restrictions of functions. If
E is a bundle on M , then the object Γ(−, E), which associates to U the collection Γ(U, E)
of C k -times differentiable sections of E over U is also a sheaf (of modules) with the obvious
restriction maps. The reader can doubtless supply more examples by making the obvious
changes to the manifold structure.
The preceding examples all have the property that sections over some open set U can
be described as having been obtained by “gluing together” sections over some cover of U .
In general, the presheaf axioms impose no such requirement. It is easy to cook up contrived
examples by taking a two-point topological space with the discrete topology, and we will
see some more interesting examples later. For now, it suffices to say that we need some
additional properties on our presheaves, and presheaves with those properties will be called
sheaves.
Definition 3.2. A sheaf F of rings on X is a presheaf of rings satisfying the following
additional properties for any open cover {Ui }i∈I of any open set U ⊂ X:

(i) Suppose that fi ∈ F(Ui ) are a collection of sections which agree on overlaps (formally,
resUi ,Ui ∩Uj fi = resUj ,Ui ∩Uj fj whenever the intersection exists). Then they lift to a
f ∈ F(U ) which has the property that resU,Ui f = fi for all i ∈ I.
(ii) Suppose that f, f 0 ∈ F(U ) and that resU,Ui f = resU,Ui f 0 for all i ∈ I. Then f = f 0 .
2 The terminology comes from thinking of F as corresponding to some sort of bundle.

3
Less formally, property (i) says that given collections of sections that agree on overlaps, we
can glue them together to give sections over some larger set, and property (ii) says that
sections are determined by their restrictions. These properties are commonly called the
gluing and identity axioms respectively.
In usual mathematics style, we follow the description of a class of objects by a description
of their morphisms. To first approximation, a morphism of sheaves is something that can
be thought of as having been induced by a pullback map. For instance, if F : M → N is a
map of smooth manifolds, we have a pullback map map F ∗ : C ∞ (N ) → C ∞ (M ) given by
F ∗ (g) = g◦F . By abuse of notation, we also have a map F ∗ : C ∞ (V ) → C ∞ (F −1 (V )) where
V ⊂ N is an open set, and defined in the same way. The map F ∗ (or rather maps, since
there is in fact a different one for each open V ⊂ N ) gives a morphism between the sheaf
C ∞ (−) of smooth functions on N and the pushforward sheaf (yes the name is confusing!)
C ∞ (F −1 (−)), which is also a sheaf on N . Note that the ordinary pullback map F ∗ which
induces this map of sheaves is simply the case when V = N . As usual with morphisms, we
require that they preserve some extra structure. This next definition makes this precise.

Definition 3.3. Let X be a topological space, and let F and G be sheaves on X. A

morphism of sheaves η : F → G has the following properties:
(i) For each open set U ⊂ X, there is a morphism (of rings, in our case) ηU : F(U ) → G(U ).
(ii) If U ⊂ V ⊂ X, then the following diagram commutes:
ηV
F(V ) G(V )
resV,U resV,U
ηU
F(U ) G(U )

In plain language: restricting a section and then applying the map is the same as
applying the map and then restricting.
One easily checks that the pullback map described above satisfies the above definition. If we
use the functorial language (in which a presheaf was really a certain contravariant functor)
then a morphism of sheaves is a natural transformation between the two sheaf functors.
A useful construction connected with sheaves is that of a stalk. A stalk is an object
associated to any point p which captures the information of the sheaf which is defined in
any neighbourhood of p. For instance, in the manifold case, the tangent space at a point
is determined in any neighbourhood p, and we will find that there is a natural way of
defining tangent spaces on appropriate schemes using stalks. Note that the behaviour of a
function at a point can in some sense be regarded as something that’s not really determined
at the point but in any neighbourhood of it. For instance, if one wants to differentiate a
function, it does not suffice to know its value at a point, but it suffices to know its values
in any neighbourhood of that point, or its value at the level of stalks. It is also true that
morphisms of sheaves are determined by the morphisms they induce on stalks, a fact which
is often useful.
Definition 3.4. Let p ∈ X be a point, and F a sheaf on X. The stalk Fp at p is the ring
obtained by the following construction:

4
 (i) The set of Fp is
Fp := {(f, U ) : U ⊂ X open containing p, f ∈ F(U )}/ ∼,
where the equivalence relation ∼ is that (f, U ) ∼ (g, V ) if resU,U ∩V f = resV,U ∩V g.
(ii) The ring operations are defined via (f, U )+(g, V ) = (f +g, U ∩V ), and (f, U )·(g, V ) =
(f g, U ∩ V ). One easily checks this is well-defined.
Note that the above intersections are always non-empty since all open sets under consid-
eration contain p. We often omit the second entry of the ordered tuple and simply write
f ∈ Fp , with the understanding that we are only interested in the behaviour of f determined
in any open neighbourhood of p. We call f ∈ Fp the germ of the function f at p.
It is not too difficult to see that one may restrict the allowed open subsets to just those
that belong to a basis for the topology, since any open neighbourhood of p contains a basis
neighbourhood.
In some sense, the above definition is really too concrete. The reason is that the impor-
tance of stalks stems largely from their relationship with the spaces of sections from which
they are constructed; the exact construction of the stalk itself doesn’t matter so much,
provided that we have natural maps F(U ) → Fp (in our case given by f 7→ (f, U )) and
that these maps behave nicely with the sheaf structure. This will become particularly evi-
dent when we define the structure sheaf of a scheme, where the associated stalks, although
isomorphic to ones defined using the construction above, will typically have a much more
natural description using localizations of rings. For this reason, the categorical definition of
the same concept is often cleaner:
Fp := lim F(U ).
−→
p∈U

Given a morphism η : F → G of sheaves on X, one obtains a natural morphism of the

stalks Fp → Gp induced by η. This is easy to see from the usual universal property nonsense,
since Fp represents morphisms out of the spaces of sections F(U ) lying over p satisfying the
constraints imposed by the restriction maps, and the map η : F → G gives such a family of
morphisms into Gp after post-composing with the maps G(U ) → Gp . Of course, this has the
direct description of taking (f, U ) ∈ Fp and mapping it to (ηU (f ), U ) ∈ Gp , which can be
shown to be well-defined.
We mentioned before that one often obtains sheaves on a space X from looking at bundles
over X. In fact, there is a way of viewing all sheaves as coming from a kind of bundle. This
construction uses
F the so-called étalé space of a sheaf, which is constructed
F (as a set) as the
disjoint union p∈X Fp . The “bundle map” is then the map π : p∈X Fp → X which sends
fp ∈ Fp 7→ p, that is, which sends a germ at the point p to the point p in X. One then
thinks of Fp as the fibre of π above p in the usual way. In his original eighty-page paper,
Serre defined the notion of a sheaf using the étalé space construction. In F
his formulation, a
sheaf was a collection of stalks Fp with a topology on the disjoint union p∈X Fp (subject
to some restrictions). One would then define sections of F to be continuous maps from X
to F whose composition with the projection π was the identity (i.e., as ordinary sections).
Grothendieck, in his usual style, took Serre’s definition and functorialized it, but the étalé
space construction still plays an important role. The next definition illustrates this point.
Definition 3.5. Let F be a presheaf. The sheafification of a sheaf is in some sense the
simplest sheaf that can be made from F. We give two definitions:

5
 (i) The sheafification of F is a sheaf F sh and a map of presheaves3 α : F → F sh such
that for any other sheaf G and map of presheaves η : F → G, there is a unique map
β : F sh → G such that the following diagram commutes:

α
F F sh
η
β

(ii) F
Alternatively, we may give an explicit construction as follows. Define a topology on
p∈X Fp as follows.
F For each f ∈ F(U ), the set {fp : p ∈ U }, where fp is the germ of
f at p, is open in p∈X Fp . The topology is the one generated by these sets. We then
define:
 
 G 
F sh (U ) := continuous maps s : U → Fp .
 
p∈X

The map α : F → F sh is the one that sends f ∈ F(U ) to the map (p 7→ fp ) ∈ F(U )sh .

The idea of the sheafification is that it turns a presheaf into a sheaf by insisting that
the elements of the sheaf be genuine “sections” of some bundle-like space, and thereby
ensuring that the two sheaf axioms are satisfied (the sheaf axioms are easy to check for
sections of bundles). To understand why this may be necessary, we consider an illustrative
example. Let X be a Riemann Surface (a 2-dimensional real manifold whose charts are
analytic functions to open subsets of C), and let O× (−) be the sheaf which assigns to
U ⊂ X the set O× (U ) of non-vanishing holomorphic functions on U . Then one can define
a new sheaf (exp O× )(U ) := {exp ◦f : f ∈ O× (U )}. We then wish to consider a quotient
O× /(exp O× ). The naı̈ve thing to do is to define (O× / exp O× )(U ) := O× (U )/(exp O× )(U ),
but the result need only be a presheaf, and not a sheaf. To see why, consider a global section
f ∈ (O× / exp O× )(X). Then given p ∈ X, one can find a small enough neighbourhood U of
p such that it is possible to define a holomorphic branch of logarithm on f (U ), and hence
express f as exp ◦(log ◦f ). Thus there is an open cover of X such that f restricts to 0 on
that open cover, but f is not itself 0, violating the identity axiom. The solution is to take
the naı̈ve quotient, and then apply a sheafification.
In fact, the étalé space construction can do more than just turn a presheaf into a sheaf,
it can also help us create a sheaf from partial data. For instance, as we will see when
constructing affine schemes, it is often the case that we know what we want our sheaf to
look like on a distinguished base for the topology on X, but we don’t have a good description
for the sections F(U ) for arbitrary
F open sets U ⊂ X. The distinguished base will suffice
to generate the topologyFon p∈X Fp , and so we may define the sections of F(U ) to be
continuous maps U → p∈X Fp as before. This can be F viewed as a kind of “analytic
continutation,” in the sense that continuous sections of p∈X Fp are exactly ones which
locally look like p 7→ fp in some neighbourhood of p, and so one can get an element of
F(U ) by gluing together sections over the distinguished base. In fact, it suffices to specify
a morphism of sheaves on a distinguished base too, since the induced map over the other
open sets can be recovered from a similar gluing process.
3 This is the same as a map of sheaves.

6
3.2 Affine Schemes
We begin with a reminder of our general strategy. We began our section on sheaves by
observing that geometric constructions typically proceed via two steps: first one defines a
topological space, and then one specifies a class of functions on the space which give it its
structure. We then claimed that sheaves will be a necessary tool for specifying the functions
on schemes, and spent some time discussing their properties. We now turn our attention to
the first part of this construction, namely, the underlying topological space of a scheme.
There is a general principle of abstraction in mathematics which says the following: to
abstract a concept, describe not what it is, but what it does. The simplest example of this
phenomenon is the notion of a vector, which is commonly introduced in early mathematics
education as something akin to an “arrow in space,” and later abstracted as something be-
longing to a collection of objects satisfying the vector space axioms. As a related example,
tangent vectors in abstract manifolds are commonly defined as directional derivative opera-
tors, where the directional derivative operator is to be thought of as a kind of proxy for the
direction in which the functions are being differentiated.
When it comes to specifying the points of a scheme X, there is a similar sort of phe-
nomenon at play. The points of X will turn out to correspond to prime ideals of rings.
One interpretation of this is to simply think of a point of a space as something that can
be used to evaluate a function on that space. In the case of algebraic varieties, functions
of interest are polynomials f (x1 , . . . , xn ) ∈ k[x1 , . . . , xn ] in some variates x1 , . . . , xn . If one
wants to evaluate f at a point p = (p1 , . . . , pn ), the computational procedure is to “set
the variables xi = pi ” and then compute f (p1 , . . . , pn ). The mathematical construction
which “sets the variables,” or imposes the relations xi = pi , is simply a quotient of rings,
and the object being quotiented is the maximal ideal hx1 − p1 , . . . , xn − pn i. Thus we see
that the ideal hx1 − p1 , . . . , xn − pn i can be used as a proxy for the point p, as the map
k[x1 , . . . , xn ] → k[x1 , . . . , xn ]/hx1 − p1 , . . . , xn − pn i has a natural interpretation as a map
which evaluates elements of k[x1 , . . . , xn ] at p.
But thinking of ideals as points also gives additional advantages. For instance, a common
difficulty when working with classical varieties is that it is very difficult to say anything
interesting about varieties over finite fields, since the associated topological space gives very
few points to work with. To take the simplest possible example, over F2 , the polynomial
x2 + x is indistinguishable from the 0 polynomial when one considers its values on the
elements 0, 1 ∈ F2 , or on the ideals hx − 0i and hx − 1i. But if one instead considers the
generalized “point” hx2 i, one can detect the difference, since x2 + x ≡ x (mod x2 ), and
0 ≡ 0 (mod x2 ). The same phenomenon can be seen if one considers varieties over the reals,
where the vanishing set of x4 − 1 is enriched due to the presence of the generalized point
hx2 + 1i.
Much like the topological space of a manifold can be thought of as a union of open sets
(homeomorphic to) open subsets of Rm , the topological space of a scheme can be thought of
as a union of sets homeomorphic to the topological space of an affine scheme. Affine schemes
play the kind of primitive “building block” role that Rn plays in differential geometry. An
affine scheme can be built from any ring R, and its topological space (as a set) is SpecR, the
collection of all prime ideals of R. One then gives this topological space a scheme structure
by endowing it with a sheaf of functions, and defines general schemes as objects locally
isomorphic to affine schemes. We now give a precise definition of affine schemes.

Definition 3.6. Let R be a ring. An affine scheme is a pair (X, OX ), where X = Spec R,
satisfying:

7
 (i) X is a topological space generated by the open sets
D(f ) := {p ∈ SpecR : f 6≡ 0 (mod p)}.
where f ∈ R.
(ii) OX is a sheaf of rings, called the structure sheaf of X, determined on the base D(f )
for X by OX (D(f )) = Rf , with the restrictions resD(f ),D(g) : OX (D(f )) → OX (D(g))
given by the natural maps Rf → Rg given by further localizing Rf at g. For this to
make sense, it should be the case that inverting g also inverts f whenever D(g) ⊂ D(f );
this follows from the fact that when D(g) ⊂ D(f ),
\ p
g∈ p = hf i
p prime
f ∈p

since g vanishes wherever f does, and the equality holds by a theorem of commutative
algebra. Recall from the previous section that specifying OX on this base determines
the sheaf structure.
The definition of the structure sheaf of an affine scheme may seem somewhat peculiar in
two respects. Firstly, one needs to understand the role played by localization, and secondly,
one needs to justify why the localizations can be thought of as restrictions. To understand
the role of localization, we recall that if R is a ring, the localization ι : R ,→ Rf induces a
map SpecRf ,→ SpecR given by p 7→ ι−1 (p). Thus, if Rf is thought of as again determining
an affine scheme structure, the topological space Spec Rf will be naturally a subset of SpecR.
Under this identification, we also have Spec Rf = D(f ). An intuitive reason this works is
because by inverting f in R, one is forced to remove any ideals which set f to zero (since
then an “evaluation” could result in division by zero), and so the prime ideals in Spec Rf
are exactly those which don’t contain f , or exactly those ideals p such that p ∈ D(f ).
Secondly, we consider the role of the localization maps as restrictions. It may seem
strange, for instance, that when R is (say) an integral domain, the restriction maps R ,→ Rf
are actually inclusion maps. This would seem to contradict the intuition that when one
restricts a function to a smaller set, one ought to get less functions, since there are fewer
possible points on which to define their values. The key insight here is that algebraic (or
regular) functions really ought to be thought of as being “holomorphic”. Holomorphic
functions have the remarkable property that they are entirely determined on any open set –
once you describe a holomorphic function on an open set, no matter how small, its values on
any extension of its domain are determined. Therefore, restricting holomorphic functions to
a smaller set does not identify any functions that are distinct on the larger set, but it can
allow you to have “extra” functions which do not extend to the larger open set. To take a
simple example, consider the affine plane A2 . Over a field k, the affine plane can be thought
of as having the ring of regular functions k[x, y], where x is the function that assigns a
point to its x-coordinate, and y is the function that assigns a point to its y-coordinate. Now
suppose we remove the y-axis to obtain A2 \ V (x), where V (x) := {p ∈ Spec k[x, y] : x ≡ 0
(mod p)} is the vanishing set of x. Then the x-coordinate of every point in A2 \ V (x) now
has an inverse, so the function x1 is defined everywhere (holomorphic), and so it should be
no surprise that the resulting coordinate ring of D(x) = A2 \ V (x) is k[x, y, x1 ], which agrees
with our definition above.
For affine schemes, there is a description of the stalks that is more natural than the usual
construction. Recalling the categorical definition, we have

8
 (OX )p = lim OX (U ) = lim Af = Ap ,
−→ −→
p∈U p∈D(f )

where we have used the fact that the limit is determined on a base, and Ap is the ring
obtained by inverting every element not lying in the ideal p.
As usual, given some objects (affine schemes), we proceed to define a notion of morphism.
The usual way to do this for geometric spaces is to give a map of the underlying topological
spaces which respects the structure of the functions on the space. In our case, the structure
of the functions on the space is encoded in the form of the structure sheaf on the space, and
so we can expect that a map between two affine schemes will also give a map between their
respective structure sheaves. As we have alluded to earlier, a prototypical map of sheaves
can be thought of as being induced by a pullback map, and we will see that this case is no
exception.

Definition 3.7. A morphism of affine schemes π : (X, OX ) → (Y, OY ) consists of:

(i) A map π : X → Y of topological spaces (again denoted π by abuse of notation).
(ii) A morphism of sheaves π ] : OY → π∗ OX , where π∗ OX is the “pushforward” sheaf on
Y given by (π∗ OX )(U ) = OX (π −1 (U )).

(iii) If p ∈ X, the morphism π ] of sheaves should send the unique maximal ideal in (OY )π(p)
into the unique maximal ideal in (OX )p . Informally, this says that “functions which
vanishes at π(p) pullback to functions which vanish at p.” See also the definition of
morphism of ringed space in the next section.

We often write this morphism as π : X → Y , where the morphism of sheaves is understood.

In many ways, this definition is exceedingly redundant. Denote by Γ(U, F) the sections
of F over the set U . Then given a morphism of π : X → Y of affine schemes, where
X = Spec A and Y = Spec B, we get a morphism π ] : Γ(Y, OY ) → Γ(Y, π∗ OX ) which is
simply a morphism π ] : B → A. This map π ] : B → A then suffices to determine the entire
sheaf morphism: over the open set D(f ) ⊂ Y , we require the commutativity of the diagram

π]
B A

π]
Bf (π∗ OX )(D(f )),

and there is only one ring map that will do, since we must have π ] (1/f ) = 1/π ] (f ). Since
a morphism of sheaves is determined on a base, this tells us that the entire morphism
π ] : OY → π∗ OX is determined by the single “global” morphism π ] : B → A. The map
π : Spec A → SpecB can also be recovered using condition (iii). We have verified:
Proposition 3.1. Every morphism π : (X, OX ) → (Y, OY ) of affine schemes, where X =
Spec A and Y = Spec B, is determined by a unique ring morphism B → A, which is the
morphism Γ(Y, OY ) → Γ(Y, π∗ OX ) determined by the morphism of structure sheaves.
Using similar reasoning, it is also not difficult to show that:

9
Proposition 3.2. Every morphism π ] : B → A of rings induces a unique morphism
Spec A → Spec B of affine schemes, with the map on topological spaces given by p 7→
(π ] )−1 (p).
These two propositions suggest that the category Ring and the category AffSch are really
the “same.” Using more work, which is no more difficult but simply requires some routine
verifications, one can show that
Proposition 3.3. The functors Spec : Ring → AffSch and Γ(−, O− ) : AffSch → Ring
give an equivalence of categories between Ring and AffSch.
This is a very nice result, but it is almost disappointing! Could it be that the language
of affine schemes is really just a glorified way of talking about rings? (Yes, in fact it is.) One
could perhaps hope that maybe the notion of a general scheme is something more, and that
once we are able to glue together affine schemes to make general schemes we will unleash
scheme theory’s full power. This hope is a bit naı̈ve. Although it’s true that general schemes
are more than just rings per se, asking that schemes be more than “just rings” is a bit like
asking that abstract smooth manifolds be more than just submanifolds of Rm — it’s not the
point. The point is that the language of schemes (and it really is a language) allows for the
expression of concepts and notions that would be very difficult to express ordinarily, and in
doing so, it can be used to clarify much of algebraic geometry (what concepts and notions?
well, that’s still some ways off...).
An important part of becoming familiar with the scheme-theoretic language is to become
accustomed to translating algebraic facts in the ring-theoretic world to geometric facts in
the scheme-theoretic world. As an example, suppose that R1 and R2 are rings and consider
their direct product R = R1 × R2 . Let e1 = (1, 0) ∈ R1 and e2 = (0, 1) ∈ R2 . Both e1
and e2 are idempotent, and so satisfy the equation e2i − ei = ei (ei − 1) = 0. If p ∈ SpecR,
then R/p is an integral domain, and so evaluating the functions ei at p gives either 0 or 1.
Moreover, since 1 = e1 + e2 , we see that e1 is 0 at p precisely when e2 is 1, and vice versa.
This shows that Spec R = V (e1 ) t V (e2 ) = D(e1 ) t D(e2 ). Now, the localization Rei ∼ = Ri ,
and hence Spec R = Spec R1 t SpecR2 . Thus we see that if a ring R may be expressed as a
product R = R1 × R2 , its spectrum can be thought of as the disjoint union of the spectra of
the factors. Another way of looking at the same fact is that Spec : Ring → AffSch gives
a contravariant equivalence of categories, and so products in Ring become coproducts in
AffSch, and a coproduct of topological spaces is exactly a disjoint union.
A list of correspondences, by no means complete, between Ring and AffSch is as follows:

(i) Connected components of Spec R correspond bijectively to idempotent elements in R.

(ii) Primary ideals in a Noetherian ring R correspond to irreducible components in SpecR.

(iii) Affine schemes whose ring is a quotient of k[x1 , . . . , xn ] correspond to affine varieties
over k.
(iv) Let R be a ring, and m a maximal ideal (i.e., an “ordinary” point). The tangent space to
SpecR at m is the dual vector space (m/m2 )∨ , which is a vector space over R/m, where
the quotient m/m2 may be thought of as occuring in the local ring Rm = (OSpec R )m .
(v) The dimension of SpecR (as a topological space) is the length of a maximal chain
(ordered by inclusion) of prime ideals in R.

10
 (vi) Let R be an integral domain. Localization maps R ,→ R[S −1 ], where S is a multi-
plicative set correspond to open embeddings of affine schemes SpecR[S −1 ] ,→ SpecR.
(vii) Let R be a ring and I ⊂ R an ideal. Then the induced map SpecR/I ,→ SpecR is a
closed immersion of affine schemes (and all closed immersions of affine schemes look
like this).
(viii) The notions of dominant, finite, regular, and rational morphisms from classical alge-
braic geometry correspond to equivalent notions for affine schemes.
(ix) A tensor product A ⊗C B of rings corresponds to the fibre product X = SpecA ⊗ B =
SpecA ×Spec C SpecB of affine schemes, which is to be thought of as an object with two
morphisms X → SpecA and X → SpecB such that we have a commutative diagram
of the form:
X SpecB

SpecA SpecC.
The two arrows on the bottom and the right are the ones that come from the maps
C → A and C → B giving the tensor product. In fact, the fibre product diagram is
just the image of Spec in AffSch of the tensor product diagram:

A ⊗C B B

A C.

(x) A special case of the previous example is the fibre of a morphism π : Spec A → SpecC
over a subscheme Spec B ,→ SpecC, which is Spec A ⊗C B, and corresponds to the
usual notion of the fibre of a map in the case of varieties.
(xi) Suppose that M is a free R-module. Then M can be thought of (using a construction
we do not give here!) as a vector bundle over Spec R.
(xii) Local rings correspond to schemes whose topological space has a single “ordinary
point” (also called a “closed point”).
This list is a sampling of how scheme theory gives a geometric interpretation to constructions
in algebra, in particular how facts about rings correspond to facts about affine schemes. One
can create similar lists for the case of graded rings (which correspond to projective schemes),
or various types of modules, which correspond to various types of sheaves. Part of learning
scheme theory is to work enough with all these correspondences so that they become second
nature.

3.3 Schemes
We’ve been promising for some time that we’ll construct schemes as things which are locally
affine schemes. It’s time to make it happen! We will define schemes as topological spaces
with a sheaf of rings such that the sheaf of rings is locally isomorphic to one coming from
an affine scheme. To define what it means for a sheaf of rings to be “locally” something, we
need the next three definitions.

11
  and OX is a sheaf of rings on X.
Definition 3.8. Suppose that X is a topological space
Further suppose that U ⊂ X is open.
 Then define O X U as a sheaf on U (with the subspace

topology) on V ⊂ U open via OX U (V ) = OX (V ).
Definition 3.9. A pair (X, OX ) is a ringed space if X is a topological space and OX is a
sheaf of rings on X. A pair (X, OX ) is called a locally ringed space if in addition each stalk
is a local ring, i.e., it has a unique maximal ideal.
Definition 3.10. A morphism of ringed spaces π : (X, OX ) → (Y, OY ) consists of a map
π : X → Y of topological spaces, and a morphism of sheaves π ] : OY → π∗ OX , where
π∗ OX is the pushforward sheaf of Y , defined as before. A morphism of locally ringed space
is a morphism of ringed spaces where in addition, whenever q ∈ π −1 (p), the induced map4
(OY )p → (OX )q is a local ring homomorphism, i.e., it maps the unique maximal ideal of
(OY )p into the maximal ideal of (OX )q .

The last two definitions essentially just formalize what we did for affine schemes in a
more general context, since we will want to refer to morphisms of ringed spaces in situations
where our ringed spaces may not be affine schemes. We are now ready for the notion of a
scheme.
Definition 3.11. A scheme is a ringed space (X, OX ) such that
 for every p ∈ X there is
an open neighbourhood U ⊂ X containing p such that (U, OX U ) is isomorphic as a ringed
space to an affine scheme (SpecR, OSpec R ).
We now give (the simplest possible!) example of a scheme which is not an affine scheme.
Let k be a field, and consider the affine schemes X = Spec k[x] and Y = Spec k[y]. We would
like a scheme where the function x is the same as the function y1 . The common domain of
these functions should be a set that excludes the origin (the point hx−0i) in X and the origin
(the point hy − 0i) in Y . We would like this domain to be Spec k[x, 1/x] = Spec k[y, 1/y]. To
do this, we construct our scheme (which we will call P1k ) in two stages: first as a topological
space, and secondly we describe its sheaf.
For the topological space, we have a homeomorphism α : Spec k[x, 1/x] → Spec k[y, 1/y]
induced by the ring map x 7→ 1/y (which should be thought of as identifying x and 1/y),
and we may construct P1k as the topological gluing of Spec k[x] and Spec k[y] along α. We
may think of any primitive open D(f ) ⊂ X as being open in P1k via the identification
X = Spec k[x] ,→ P1k , and similarly for D(g) ⊂ Y . We claim that the primitive opens for
X and Y generate the quotient topology on P1k . To see this, note that any open set in P1k
corresponds to an open set U t V in X t Y via the quotient map τ , and the open set U t V
is a union of primitive opens by the definition of the disjoint union topology. Since maps
preserve unions, we see that τ (U t V ) is also a union of primitive opens (this time with the
identification induced by α), and so since τ (U t V ) was an arbitrary open set, we are done.
To describe the sheaf, we describe it on a base. This is also easy: we define OP1k to be
equal to OX (D(f )) when D(f ) ⊂ X, and OY (D(g)) when D(g) ⊂ Y , noting that when
D(h) is a subset of both X and Y we have an identification of OX (D(h)) and  OY (D(h))
induced by the map α. This gives P1k a sheaf of rings, and it is clear that OP1k  = OX and

 X
OP1k  = OY , so the ringed space (P1k , OP1k ) is in fact a scheme.

Y
4 We have not explicitly defined this map, but it’s easy to figure out what it means.

12
 We now compute the global sections of P1k (i.e., the ring Γ(P1k , OP1k )). By the identity
axiom, global sections of P1k can be described pairs of compatible (agreeing on overlaps)
sections over the open cover X ∪ Y . If (fx , fy ) is such a pair, then fx ∈ k[x], so fx is a
polynomial in x = 1/y, and fy ∈ k[y], so fy is a polynomial in y = 1/x. fx and fy agree on
the overlap (that is, in the ring k[x, 1/x] = k[y, 1/y], where x = 1/y), and so the polynomial
fx viewed as a polynomial in y = 1/x must not have any negative-degree terms, and so
must be constant, and similarly for fy . Thus we see that both fx and fy are elements of k,
and they must be the same element since they agree in the ring k[x, 1/x] = k[y, 1/y]. Hence
Γ(P1k , OP1k ) = k. Since the topological space of P1k is not homeomorphic to Spec k, we see
that P1k is not affine.
This construction might seem tedious (and this was the simplest possible example!),
but we will soon have a better way of describing P1k . This is via the projective scheme
construction (that construction is altogether more tedious, but it is at least general in the
sense that it describes a wide class of schemes). In much the same way as affine schemes are
built out of rings, projective schemes are built out of graded rings. In the affine case, the
construction mirrors the classical case of affine varieties, where the functions on the variety
come from an affine coordinate ring R. In the projective case, the functions on a variety are
instead described as ratios of homogeneous L functions from a graded coordinate ring S. By
∞
graded coordinate ring, we mean that S = i=0 Si , and that if fi ∈ Si and fj ∈ Sj , then
fi fj ∈ Si+j . The usual example is to think of S = k[x0 , x1 , . . . , xn ], and to think of Si as
homogeneous polynomials of degree i. In general, if f ∈ Si , we often say that f has degree
i.
The purpose of the grading on S is to keep track of which quotients of polynomials make
acceptable rational functions on the associated scheme. For instance, we can recall the case
of ordinary projective space Pnk (in the sense of classical algebraic varieties). This is the space
of lines through the origin of kn+1 , where a point p ∈ Pn can be written in homogeneous
coordinates p = [p0 : p1 : . . . : pn ], where [p0 : p1 : . . . : pn ] denotes the equivalence class
{(λp0 , λp1 , . . . , λpn ) : λ ∈ k}. If we want a rational function that we can evaluate on such
a point, we need our process of evaluating the function to be independent of the choice of
λ. If we take a ratio of two homogeneous polynomials fg(x (x0 ,x1 ,...,xn )
0 ,x1 ,...,xn )
, then evaluating this
expression at a projective point will be well-defined if and only if deg f = deg g, in which
case scaling the input by a factor of λ will result in a scaling the output by a factor of
λdeg f −deg g = 1.
To capture this notion of “allowable quotients,” we will need a projective version of
localization, which is the purpose of the next definition.
Definition 3.12. Let S be a graded ring, and p a homogeneous prime ideal5 in S. We
define

S(p) = {f /g : f, g homogeneous in S, deg f = deg g, g 6≡ 0 (mod p)}

which is a ring with the obvious operations.

We then construct projective schemes as follows.
Definition 3.13. Let S be a graded ring, and define X = ProjS (as a set!) to be the set
of all homogeneous
L∞ prime ideals of S which do not contain the so-called “irrelevant ideal”
S+ := i=1 Si (the purpose of this is so that evaluating at p ∈ ProjS does not set all
5A homogeneous ideal is one generated by homogeneous elements.

13
non-constant functions to zero, which is necessary because [0 : 0 : . . . : 0] is not a point in
projective space). We then define a scheme structure on X as follows:
(i) The topology of X is generated by the primitive open sets
D(f ) := {p ∈ X : f 6≡ 0 (mod p)},
where f ∈ S is a homogeneous element.
F
(ii) The étalé space of the structure sheaf on X is the disjoint union p∈X S(p) , with the
topology generated by the open sets {(f /g)p }p∈D(g) where f and g are homogeneous
elements of the same degree.
The sections OX (U ) where U ⊂ X is open are the continuous maps from U →
(iii) F
p∈X S(p) .

Note that property (iii) essentially just says that the regular functions
F over an open set U
locally look like homogeneous quotients f /g. Indeed, if s : U → p∈X S(p) is continuous,
then if V := {(f /g)p }p∈D(g) is open in the étalé space its inverse image under s must be
open. This either means that the inverse image of V is empty, or that D(g)∩U is non-empty,
in which case we must have s(p) = (f /g)p for all p ∈ D(g) ∩ U .
The definition we have given should mirror the usual idea of projective varieties and the
rational functions on them. It remains to show that this construction we have given satisfies
the definition of a scheme.
Proposition 3.4. With the construction as above, X = Proj S is a scheme.
Proof. It suffices to show that every point p ∈ X contains an open neighbourhood isomorphic
to an affine scheme. Since every p is contained in some D(g) (no homogeneous prime ideal
kills all functions,
 otherwise it would contain the irrelevant ideal), it suffices to show that
(D(g), OX D(g) ) is affine. We will show it is isomorphic (as an affine scheme) to Spec S(g) ,
where S(g) means the degree 0 piece of the localization Sg (i.e., the subring of quotients
where the numerator and denominator have the same degree).
The trickiest part of this exercise is to exhibit the homeomorphism between the set of
homogeneous prime ideals in D(g) and the ordinary prime ideals in S(g) . Note that we
have a localization map ι : S → Sg , and this gives a homeomorphism between prime ideals
in D(g) and homogeneous prime ideals in Sg , so it suffices to show that the homogeneous
prime ideals in Sg are in homeomorphic bijection to the prime ideals in S(g) .
Defining the map from homogeneous prime ideals in Sg to Spec S(g) is easy: simply
send p 7→ p ∩ S(g) . The inverse map is much trickier. Given q ∈ SpecS(g) we define the
deg g
homogeneous sets pi = {f : deg f = i, fgdeg f ∈ q}, and consider p0 := i∈Z pi . We make
L
a series of claims:
(i) If fi ∈ pi and fj ∈ pj , then fi fj ∈ pi+j . This is because

fjdeg g
! !
(fi fj )deg g fideg g
= ∈q
g deg fi +deg fj g deg fi g deg fj

(ii) If f 2 ∈ p2i , then f ∈ pi . This is because

2
(f 2 )deg g f deg g

= ∈ q.
g deg f 2 g deg f

14
 And so using the fact that q is prime, we see that f ∈ pi .
(iii) If f, f 0 ∈ pi , then f + f 0 ∈ pi . To see this, observe that

(f + f 0 )2 deg g
∈ q,
g 2 deg(f +f 0 )

since once the numerator is expanded each term will contain either a factor of f deg g
deg g
or f 0 . Hence (f + f 0 )2 ∈ p2i , and so f + f 0 ∈ pi .
(iv) p0 is an ideal of Sg . We observe that if r ∈ Sg , then rpi ⊂ pi+deg r . This is because if
f ∈ pi
(rf )deg g
 deg g   deg g 
r f
= ∈q
g deg rf g deg r g deg f
since q is an ideal and closed under multiplication from elements in S(g) .
(v) p0 is a homogeneous ideal. This is because it is generated by the homogeneous elements
in each pi .
(vi) p0 is a prime ideal. To show a homogeneous ideal is prime, it suffices to consider
homogeneous elements. Suppose that fi fj ∈ p0 , where deg fi = i and deg fj = j. Then
we have as in (i) that
fjdeg g
! !
fideg g
∈ q,
g deg fi g deg fj

but then one of the factors is in q, and so either fi or fj is in p0 .

(vii) The map p ∩ SpecS(g) =: q 7→ p0 gives an inverse to the
Pmap p 7→ p ∩ SpecS(g) .
We do this by simply showing two containments. If f = i fi ∈ p where each fi is
fideg g
homogeneous, then g deg fi
∈ q, and hence fi ∈ p0 , and thus f ∈ p0 . Alternatively, if
fideg g
f ∈ p0 , then each homogeneous element fi of f satisfies g deg fi
∈ q = p ∩ SpecS(g) , so
fideg g
in particular g deg fi
∈ p, hence fi ∈ p since p is prime, hence f ∈ p.

(viii) The two exhibited maps give homeomorphisms between D(g) and Spec S(g) . Rephrased,
this says that they map open sets to open sets (it would be inverse images of open
sets are open, but since they are bijections, one can say they map open sets to open
sets), which can be rephrased as saying that p ∈ D(f ) ∩ D(g) (a basis open in D(g)) if
and only if p ∩ SpecS(g) ∈ D((f deg g )/(g deg f )). This is equivalent to saying that f ∈ p
f deg g
if and only if g deg f
∈ p ∩ SpecS(g) , which is clear from the basic properties of prime
ideals.
We have succeeded in showing that D(g) is homeomorphic to Spec S(g) . To show that
their spaces of sections are isomorphic, it suffices to observe that the stalk at p of Proj S is
naturally isomorphic to the stalk at p ∩ SpecS(g) of Spec S(g) . Indeed, they are both ratios
of homogeneous elements of S, and by examining the correspondence between p and p ∩ S(g)
one can show that the condition on the denominators is the same for both (essentially as
we did in part (viii) above). One then gets an isomorphism between the structure sheaves
based on the fact that both are determined on the basis sets (which we gave a bijection

15
between) as continuous sections into their respective étalé spaces. We leave this verification
to the reader6 .
It is worth taking time to convince yourself that the construction in the preceding proposition
is really nothing more than a sophisticated version of the usual process of defining rational
functions on projective varieties from classical algebraic geometry — all that has changed
is the language.

4 Where to go from here?

In this article, which spans around fifteen pages, we have merely managed to define the core
notions needed to discuss the classical examples of affine and projective varieties as schemes.
This achievement is (so far) a linguistic one – nothing has really “happened”. To seriously
start using this theory, one then starts defining the sorts of things used in differential
geometry – line and vector bundles, (co)homology theories – and algebraic geometry (like
divisors), and uses the new abstract perspective to achieve results out of reach by ordinary
methods. This is not light work: there’s a reason why EGA is a thousand pages long,
and Hartshorne’s comparatively terse book is a few hundred. In order not to get lost, it’s
important to have a firm handle on the geometric intuition, since learning things just by
understanding the formalism really is hopeless. In this article, we have done our best to
provide the intuition and motivation for the definitions, with the hope that the reader can
start to see the “bigger picture” that lies behind the abstraction.

6 This may seem like somewhat of a cop out, but it could be worse! Hartshorne, in his book, replaces the

entire eight-point verification of the bijection between the two sets of prime ideals with the sentence “The
properties of localization show that [the map] is bijective as a map from [D(g) to Spec S(g) ]”. Indeed, and
the properties of elliptic curves show Fermat’s Last Theorem!

16