Algebraic Geometry

J.S. Milne

Version 6.10
November 11, 2024
These notes are an introduction to the theory of algebraic varieties emphasizing the
similarities to the theory of manifolds. In contrast to most such accounts they study
abstract algebraic varieties, and not just subvarieties of affine and projective space. This
approach leads more naturally into scheme theory.
Before learning scheme theory everyone should understand algebraic varieties over
algebraically closed fields: first the geometric intuition and then the abstractions.

BibTeX information
@misc{milneAG,
author={Milne, James S.},
title={Algebraic Geometry (v6.1)},
year={2024},
note={Available at www.jmilne.org/math/},
pages={231}
}

v2.01 (August 24, 1996). First version on the web.

v3.01 (June 13, 1998).
v4.00 (October 30, 2003). Fixed errors; many minor revisions; added exercises; added
two sections/chapters; 206 pages.
v5.00 (February 20, 2005). Heavily revised; most numbering changed; 227 pages.
v5.10 (March 19, 2008). Minor fixes; TEXstyle changed, so page numbers changed; 241
pages.
v5.20 (September 14, 2009). Minor corrections; revised Chapters 1, 11, 16; 245 pages.
v5.22 (January 13, 2012). Minor fixes; 260 pages.
v6.00 (August 24, 2014). Major revision; split off the basic first course from the topics;
223 pages.
v6.01 (August 23, 2015). Minor fixes; 226 pages.
v6.02 (March 19, 2017). Minor fixes; 221 pages.
v6.03 (November 2, 2023). Minor fixes; 223 pages.
v6.10 (November 11, 2024). Minor improvements; 231 pages.
Available at www.jmilne.org/math/
Please send comments and corrections to me at the address on my web page.

The curves are a tacnode, a ramphoid cusp, and an ordinary triple point.

Copyright © 1996–2024 J.S. Milne.

Single paper copies for noncommercial personal use may be made without explicit
permission from the copyright holder.
Table of Contents

Introduction 9

1 Preliminaries from commutative algebra 12

a Rings and ideals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
b Rings of fractions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
c Unique factorization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
d Integral dependence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
e Tensor Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
f Transcendence bases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

2 Algebraic Sets 36
a Definition of an algebraic set . . . . . . . . . . . . . . . . . . . . . . . . 36
b The Hilbert basis theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 37
c The Zariski topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
d The Hilbert Nullstellensatz . . . . . . . . . . . . . . . . . . . . . . . . . 39
e The correspondence between algebraic sets and radical ideals . . . . . . 40
f Finding the radical of an ideal . . . . . . . . . . . . . . . . . . . . . . . 44
g Properties of the Zariski topology . . . . . . . . . . . . . . . . . . . . . . 44
h Decomposition of an algebraic set into irreducible algebraic sets . . . . 45
i Regular functions; the coordinate ring of an algebraic set . . . . . . . . 48
j Regular maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
k Hypersurfaces; finite and quasi-finite maps . . . . . . . . . . . . . . . . 50
l Noether normalization theorem . . . . . . . . . . . . . . . . . . . . . . 51
m Dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

3 Affine Algebraic Varieties 59

a Sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
b Ringed spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
c The ringed space structure on an algebraic set . . . . . . . . . . . . . . . 61
d Morphisms of ringed spaces . . . . . . . . . . . . . . . . . . . . . . . . . 64
e Affine algebraic varieties . . . . . . . . . . . . . . . . . . . . . . . . . . 65
f The category of affine algebraic varieties . . . . . . . . . . . . . . . . . . 66
g Explicit description of morphisms of affine varieties . . . . . . . . . . . 67

Properties of the regular map Spm(𝛼) . . . . . . . .

h Subvarieties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
i . . . . . . . . . . . 72
j Affine space without coordinates . . . . . . . . . . . . . . . . . . . . . . 73
k Birational equivalence . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

3
 l Dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

4 Local Study 81
a Tangent spaces to plane curves . . . . . . . . . . . . . . . . . . . . . . . 81
b Tangent cones to plane curves . . . . . . . . . . . . . . . . . . . . . . . 83

Tangent spaces to algebraic subsets of 𝔸𝑚 .

c The local ring at a point on a curve . . . . . . . . . . . . . . . . . . . . . 85
d . . . . . . . . . . . . . . . . 86
e The differential of a regular map . . . . . . . . . . . . . . . . . . . . . . 88
f Tangent spaces to affine algebraic varieties . . . . . . . . . . . . . . . . 89
g Tangent cones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
h Nonsingular points; the singular locus . . . . . . . . . . . . . . . . . . . 94
i Nonsingularity and regularity . . . . . . . . . . . . . . . . . . . . . . . . 97
j Examples of tangent spaces . . . . . . . . . . . . . . . . . . . . . . . . . 97
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

5 Algebraic Varieties 100

a Algebraic prevarieties . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
b Regular maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
c Algebraic varieties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
d Maps from varieties to affine varieties . . . . . . . . . . . . . . . . . . . 103
e Subvarieties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
f Prevarieties obtained by patching . . . . . . . . . . . . . . . . . . . . . . 105
g Products of varieties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
h The separation axiom revisited . . . . . . . . . . . . . . . . . . . . . . . 111
i Fibred products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
j Dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
k Dominant maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
l Rational maps; birational equivalence . . . . . . . . . . . . . . . . . . . 117
m Local study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
n Étale maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
o Étale neighbourhoods . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
p Smooth maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
q Algebraic varieties as functors . . . . . . . . . . . . . . . . . . . . . . . 125
r Rational and unirational varieties . . . . . . . . . . . . . . . . . . . . . 128
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

Algebraic subsets of ℙ𝑛 . . . . . . . . . . . . . . . . . .
6 Projective Varieties 130

The Zariski topology on ℙ𝑛 . . . . . . . . . . . . . . . .

a . . . . . . . . . 130

Closed subsets of 𝔸𝑛 and ℙ𝑛 . . . . . . . . . . . . . . .

b . . . . . . . . . 134
c . . . . . . . . . 135

ℙ𝑛 is an algebraic variety . . . . . . . . . . . . . . . . .
d The hyperplane at infinity . . . . . . . . . . . . . . . . . . . . . . . . . . 136
e . . . . . . . . . 136
f The homogeneous coordinate ring of a projective variety . . . . . . . . . 138
g Regular functions on a projective variety . . . . . . . . . . . . . . . . . . 139
h Maps from projective varieties . . . . . . . . . . . . . . . . . . . . . . . 140
i Some classical maps of projective varieties . . . . . . . . . . . . . . . . . 142
j Projective space without coordinates . . . . . . . . . . . . . . . . . . . . 147
k The functor defined by projective space . . . . . . . . . . . . . . . . . . 147

4
 l Maps to projective space . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
m Grassmann varieties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
n Bézout’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
o Hilbert polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
p Dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
q Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159

7 Complete Varieties 161

a Definition and basic properties . . . . . . . . . . . . . . . . . . . . . . . 161
b Proper maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
c Projective varieties are complete . . . . . . . . . . . . . . . . . . . . . . 164
d Elimination theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
e The rigidity theorem; abelian varieties . . . . . . . . . . . . . . . . . . . 169
f Chow’s Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
g Analytic spaces; Chow’s theorem . . . . . . . . . . . . . . . . . . . . . . 173
h Nagata’s Embedding Theorem . . . . . . . . . . . . . . . . . . . . . . . 174
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175

8 Normal Varieties; (Quasi-)finite maps; Zariski’s Main Theorem 176

a Normal varieties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
b Regular functions on normal varieties . . . . . . . . . . . . . . . . . . . 179
c Finite and quasi-finite maps . . . . . . . . . . . . . . . . . . . . . . . . . 181
d The fibres of finite maps . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
e Zariski’s main theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
f Stein factorization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
g Blow-ups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
h Resolution of singularities . . . . . . . . . . . . . . . . . . . . . . . . . . 196
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197

9 Regular Maps and Their Fibres 198

a The constructibility theorem . . . . . . . . . . . . . . . . . . . . . . . . 198
b The fibres of morphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
c Flat maps and their fibres . . . . . . . . . . . . . . . . . . . . . . . . . . 204
d Lines on surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
e Bertini’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
f Birational classification . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219

Solutions to the exercises 221

Bibliography 227

Index 228

5
Notation
Throughout, 𝑘 is an algebraically closed field. Unadorned tensor products are over 𝑘. For
a 𝑘-algebra 𝑅 and 𝑘-module 𝑀, we often write 𝑀𝑅 for 𝑅 ⊗ 𝑀. The dual Hom𝑘-linear (𝐸, 𝑘)
of a 𝑘-vector space 𝐸 is denoted by 𝐸 ∨ . All rings are commutative with 1, and homomor-
phisms are required to map 1 to 1. Following Bourbaki, we require compact topological
spaces to be Hausdorff.

ℕ = {0, 1, 2, …}
ℤ = ring of integers
ℂ = field of complex numbers
𝔽𝑝 = ℤ∕𝑝ℤ = field of 𝑝 elements, 𝑝 a prime number.

Given an equivalence relation, [∗] denotes the equivalence class containing ∗. A family of
elements of a set 𝐴 indexed by a second set 𝐼, denoted (𝑎𝑖 )𝑖∈𝐼 , is a function 𝑖 ↦ 𝑎𝑖 ∶ 𝐼 → 𝐴.
We sometimes write |𝑆| for the number of elements in a finite set 𝑆.

𝑋 ⊂ 𝑌 𝑋 is a subset of 𝑌, not necessarily proper.

𝑋 = 𝑌 indicates that the equality in question is a definition.
𝑋 ≈ 𝑌 𝑋 and 𝑌 are isomorphic.
def

𝑋 ≃ 𝑌 𝑋 and 𝑌 are isomorphic by a specific isomorphism, usually canonical.

We use Gothic (fraktur) letters to denote ideals:

𝔞 𝔟 𝔠 𝔪 𝔫 𝔭 𝔮 𝔄 𝔅 ℭ 𝔐 𝔑 𝔓 𝔔
𝑎 𝑏 𝑐 𝑚 𝑛 𝑝 𝑞 𝐴 𝐵 𝐶 𝑀 𝑁 𝑃 𝑄

A reference “Section 3m” is to Section m in Chapter 3; a reference “3.45 ” is to item 45

in chapter 3; a reference “(67)” is to (displayed) equation 67 (often given with a page
reference).

Prerequisites
The reader is assumed to be familiar with the basic objects of algebra, namely, rings,
modules, fields, and so on.

References
CA: Milne, J.S., Commutative Algebra, v4.03, 2020.
FT: Milne, J.S., Fields and Galois Theory, Kea Books, 2022.
monnnn: Question nnnn on mathoverflow.net.
sxnnnn: Question nnnn on math.stackexchange.com.
We sometimes refer to the computer algebra programs
CoCoA (Computations in Commutative Algebra) website.
Macaulay2 website; web version.

6
Acknowledgements
I thank the following for providing corrections and comments for earlier versions of
these notes: Abhishek, Jorge Nicolás Caro Montoya, Sandeep Chellapilla, Rankeya
Datta, Umesh V. Dubey, Mark Faucette, Shalom Feigelstock, Tony Feng, B.J. Franklin,
Sergei Gelfand, Daniel Gerig, Darij Grinberg, Lucio Guerberoff, Isac Hedén, Guido
Helmers, Florian Herzig, Christian Hirsch, Cheuk-Man Hwang, Seonho Hwangbo,
Jasper Loy Jiabao, Dan Karliner, Lars Kindler, John Miller, Andrew Phillips, Devesh
Rajpal, Joaquin Rodrigues, Sean Rostami, David Rufino, Hossein Sabzrou, Jyoti Prakash
Saha, Tom Savage, Nguyen Quoc Thang, Bhupendra Nath Tiwari, Israel Vainsencher,
Soli Vishkautsan, Kai Wang, Dennis Bouke Westra, Felipe Zaldivar, Luochen Zhao.

There is almost nothing left to discover in geometry.

Descartes, March 26, 1619

Question: If we try to explain to a layman what algebraic geometry is, it seems to

me that the title of the old book of Enriques is still adequate: Geometrical Theory of
Equations . . . .
Grothendieck: Yes! but your “layman” should know what a system of algebraic
equations is. This would cost years of study to Plato.
Question: It should be nice to have a little faith that after two thousand years every
good high school graduate can understand what an affine scheme is . . .
From the notes of a lecture series that Grothendieck gave at SUNY at Buffalo in the
summer of 1973 (in 167 pages, Grothendieck manages to cover very little).

7
Introduction
I believe that you should begin by getting a solid foundation
in what I call “elementary algebraic geometry,” that is, the
theory of “Serre varieties” as defined in FAC.1 I think that
at the beginning you should should strictly limit yourself to
varieties over an algebraically closed field (but of arbitrary
characteristic).
Dieudonné, Letter to Ribenboim, 1972.
Just as the starting point of linear algebra is the study of the solutions of systems of

∑𝑛
𝑎𝑖𝑗 𝑋𝑗 = 𝑏𝑖 , 𝑖 = 1, … , 𝑚,
linear equations,

𝑗=1
(1)

the starting point for algebraic geometry is the study of the solutions of systems of
polynomial equations,

𝑓𝑖 (𝑋1 , … , 𝑋𝑛 ) = 0, 𝑖 = 1, … , 𝑚, 𝑓𝑖 ∈ 𝑘[𝑋1 , … , 𝑋𝑛 ].

theorems for linear equations do not depend on which field 𝑘 you are working over,2
One immediate difference between linear equations and polynomial equations is that

but those for polynomial equations depend on whether or not 𝑘 is algebraically closed
and (to a lesser extent) whether 𝑘 has characteristic zero.
A better description of algebraic geometry is that it is the study of polynomial func-
tions and the spaces on which they are defined (algebraic varieties), just as topology is
the study of continuous functions and the spaces on which they are defined (topological
spaces), differential topology the study of infinitely differentiable functions and the
spaces on which they are defined (differentiable manifolds), and so on:

algebraic geometry regular (polynomial) functions algebraic varieties

topology continuous functions topological spaces
differential topology infinitely differentiable functions differentiable manifolds
complex analysis analytic (power series) functions complex manifolds.

The approach adopted in this course makes plain the similarities between these different
areas of mathematics. Of course, the polynomial functions form a much less rich class
1
Serre, Jean-Pierre. Faisceaux algébriques cohérents. Ann. of Math. (2) 61, (1955). 197–278, commonly

For example, suppose that the system (1) has coefficients 𝑎𝑖𝑗 ∈ 𝑘 and that 𝐾 is a field containing
referred to as FAC.

𝑘. Then (1) has a solution in 𝑘𝑛 if and only if it has a solution in 𝐾 𝑛 , and the dimension of the space of
2

solutions is the same for both fields.

9
10 Introduction

than the others, but by restricting our study to polynomials we are able to do calculus

𝑑 ∑ ∑
over any field: we simply define

𝑎𝑖 𝑋 𝑖 = 𝑖𝑎𝑖 𝑋 𝑖−1 .
𝑑𝑋

Consider a nonzero differentiable function 𝑓(𝑥, 𝑦, 𝑧). In calculus, we learn that the
Moreover, calculations with polynomials are easier than with more general functions.

𝑓(𝑥, 𝑦, 𝑧) = 𝐶
equation

defines a surface 𝑆 in ℝ3 , and that the tangent plane to 𝑆 at a point 𝑃 = (𝑎, 𝑏, 𝑐) has
(2)

𝜕𝑓 𝜕𝑓 𝜕𝑓
( ) (𝑥 − 𝑎) + ( ) (𝑦 − 𝑏) + ( ) (𝑧 − 𝑐) = 0.
equation3

𝜕𝑥 𝑃 𝜕𝑦 𝑃 𝜕𝑧 𝑃
(3)

The inverse function theorem says that a differentiable map 𝛼 ∶ 𝑆 → 𝑆 ′ of surfaces is a

local isomorphism at a point 𝑃 ∈ 𝑆 if it is an isomorphism on the tangent planes.
Now consider a nonzero polynomial 𝑓(𝑥, 𝑦, 𝑧) with coefficients in a field 𝑘. In these
notes, we shall learn that the equation (2) defines a surface in 𝑘 3 , and we shall use
the equation (3) to define the tangent space at a point 𝑃 on the surface. However, and
this is one of the essential differences between algebraic geometry and the other fields,

difference is that 1∕𝑋 is not the derivative of any rational function of 𝑋, and nor is 𝑋 𝑛𝑝−1
the inverse function theorem does not hold in algebraic geometry. One other essential

in characteristic 𝑝 ≠ 0 — these functions cannot be integrated in the field of rational

functions 𝑘(𝑋).
These notes form a basic first course on algebraic geometry. Throughout, we require
the ground field to be algebraically closed in order to be able to concentrate on the
geometry. Additional chapters, treating more advanced topics, can be found on my
website.

The approach to algebraic geometry taken in these notes

In differential geometry it is important to define differentiable manifolds abstractly, i.e.,
not simply as submanifolds of some Euclidean space. For example, it is difficult even to
make sense of a statement such as “the Gauss curvature of a surface is intrinsic to the
surface but the principal curvatures are not” without the abstract notion of a surface.
Until the mid 1940s, algebraic geometry was concerned only with algebraic sub-
varieties of affine or projective space over algebraically closed fields. Then, in order
to give substance to his proof of the congruence Riemann hypothesis for curves and
abelian varieties, Weil was forced to develop a theory of algebraic geometry for “abstract”
algebraic varieties over arbitrary fields, but his “foundations” are unsatisfactory in two

⋄ Lacking a sheaf theory, his method of patching together affine varieties to form
major respects:

⋄ His definition of a variety over a base field 𝑘 is not intrinsic; specifically, he fixes
abstract varieties is clumsy.

some large “universal” algebraically closed field Ω and defines an algebraic variety
over 𝑘 to be an algebraic variety over Ω together with a 𝑘-structure.
Think of 𝑆 as a level surface for the function 𝑓, and note that the equation is that of a plane through
(𝑎, 𝑏, 𝑐) perpendicular to the gradient vector (▿𝑓)𝑃 of 𝑓 at 𝑃.
3
 11

In the ensuing years, several attempts were made to resolve these difficulties. In
1955, Serre resolved the first by borrowing ideas from complex analysis and defining an
algebraic variety over an algebraically closed field to be a topological space with a sheaf
of functions that is locally affine. Then, in the late 1950s Grothendieck resolved all such
difficulties by developing the theory of schemes.
In these notes, we follow Grothendieck except that, by working only over a base
field, we are able to simplify his language by considering only the closed points in the
underlying topological spaces. In this way, we hope to provide a bridge between the
intuition given by advanced calculus and the abstractions of scheme theory.

The following complementary material is available my website.

10 Algebraic Schemes. Explains how the the theory in these notes extends to arbitrary
base fields, nonreduced schemes, etc.
11 Surfaces. Develops enough of the theory of algebraic surfaces to explain what is still
the most illuminating proof of the Riemann hypothesis for curves over finite fields
(Weil, Mattuck, Tate, Grothendieck).
12 Divisors and Intersection Theory.
13 Coherent Sheaves and Vector Bundles.
14 Differentials (Outline).
15 Algebraic Varieties over the Complex Numbers.
17 Lefschetz Pencils.
Chapter 1

Preliminaries from commutative

algebra

Algebraic geometry and commutative algebra are closely intertwined. For the most
part, we develop the necessary commutative algebra in the context in which it is used.
However, in this chapter, we review some basic definitions and results from commutative
algebra.

a. Rings and ideals

Basic definitions
Let 𝐴 be a ring. A subring of 𝐴 is a subset that contains 1𝐴 and is closed under addition,
multiplication, and the formation of negatives. An 𝐴-algebra is a ring 𝐵 together
with a homomorphism 𝑖𝐵 ∶ 𝐴 → 𝐵. A homomorphism of 𝐴-algebras 𝐵 → 𝐶 is a
homomorphism of rings 𝜑 ∶ 𝐵 → 𝐶 such that 𝜑(𝑖𝐵 (𝑎)) = 𝑖𝐶 (𝑎) for all 𝑎 ∈ 𝐴.
Elements 𝑥1 , … , 𝑥𝑛 of an 𝐴-algebra 𝐵 are said to generate it if every element of 𝐵
can be expressed as a polynomial in the 𝑥𝑖 with coefficients in 𝑖𝐵 (𝐴), i.e., if the homo-
morphism of 𝐴-algebras 𝐴[𝑋1 , … , 𝑋𝑛 ] → 𝐵 acting as 𝑖𝐴 on 𝐴 and sending 𝑋𝑖 to 𝑥𝑖 is

When 𝐴 ⊂ 𝐵 and 𝑥1 , … , 𝑥𝑛 ∈ 𝐵, we let 𝐴[𝑥1 , … , 𝑥𝑛 ] denote the 𝐴-subalgebra of 𝐵

surjective.

generated by the 𝑥𝑖 .
A ring homomorphism 𝐴 → 𝐵 is said to be of finite-type, and 𝐵 is a finitely gener-
ated 𝐴-algebra if 𝐵 is generated by a finite set of elements as an 𝐴-algebra.
A ring homomorphism 𝐴 → 𝐵 is finite, and 𝐵 is a finite1 𝐴-algebra, if 𝐵 is finitely
generated as an 𝐴-module.
Let 𝑘 be a field, and let 𝐴 be a 𝑘-algebra. If 1𝐴 ≠ 0, then the map 𝑘 → 𝐴 is injective,
and we can identify 𝑘 with its image, i.e., we can regard 𝑘 as a subring of 𝐴. If 1𝐴 = 0,
then 𝐴 is the zero ring {0}.
A ring is an integral domain if it is not the zero ring and if 𝑎𝑏 = 0 implies that
𝑎 = 0 or 𝑏 = 0; in other words, if 𝑎𝑏 = 𝑎𝑐 and 𝑎 ≠ 0, then 𝑏 = 𝑐.
For a ring 𝐴, 𝐴× is the group of elements of 𝐴 with inverses (the units in the ring).
1
The term “module-finite” is also used.

12
 a. Rings and ideals 13

Ideals
Let 𝐴 be a ring. An ideal 𝔞 in 𝐴 is a subset such that
(a) 𝔞 is a subgroup of 𝐴 regarded as a group under addition;
(b) 𝑎 ∈ 𝔞, 𝑟 ∈ 𝐴 ⇒ 𝑟𝑎 ∈ 𝔞.
The ideal generated by a subset 𝑆 of 𝐴 is the intersection of all ideals 𝔞 containing 𝑆 —
∑
form 𝑎𝑖 𝑠𝑖 with 𝑎𝑖 ∈ 𝐴, 𝑠𝑖 ∈ 𝑆. The ideal generated by the empty set is the zero ideal
it is easy to see that this is in fact an ideal, and that it consists of the finite sums of the

{0}. When 𝑆 = {𝑠1 , 𝑠2 , …}, we write (𝑠1 , 𝑠2 , …) for the ideal it generates.
Let 𝔞 and 𝔟 be ideals in 𝐴. The set {𝑎 + 𝑏 ∣ 𝑎 ∈ 𝔞, 𝑏 ∈ 𝔟} is an ideal, denoted by 𝔞 + 𝔟.
The ideal generated by {𝑎𝑏 ∣ 𝑎 ∈ 𝔞, 𝑏 ∈ 𝔟} is denoted by 𝔞𝔟. Clearly 𝔞𝔟 consists of all
∑
finite sums 𝑎𝑖 𝑏𝑖 with 𝑎𝑖 ∈ 𝔞 and 𝑏𝑖 ∈ 𝔟, and if 𝔞 = (𝑎1 , … , 𝑎𝑚 ) and 𝔟 = (𝑏1 , … , 𝑏𝑛 ),
then 𝔞𝔟 = (𝑎1 𝑏1 , … , 𝑎𝑖 𝑏𝑗 , … , 𝑎𝑚 𝑏𝑛 ). Note that
𝔞𝔟 ⊂ 𝔞 ∩ 𝔟.
The kernel of a homomorphism 𝐴 → 𝐵 is an ideal in 𝐴. Conversely, for any ideal 𝔞
(4)

in 𝐴, the set of cosets of 𝔞 in 𝐴 forms a ring 𝐴∕𝔞, and 𝑎 ↦ 𝑎 + 𝔞 is a homomorphism

𝜑 ∶ 𝐴 → 𝐴∕𝔞 whose kernel is 𝔞. The map 𝔟 ↦ 𝜑−1 (𝔟) is a one-to-one correspondence
between the ideals of 𝐴∕𝔞 and the ideals of 𝐴 containing 𝔞.
An ideal 𝔭 is prime if 𝔭 ≠ 𝐴 and 𝑎𝑏 ∈ 𝔭 ⇒ 𝑎 ∈ 𝔭 or 𝑏 ∈ 𝔭. Thus 𝔭 is prime if and
only if 𝐴∕𝔭 is nonzero and has the property that
𝑎𝑏 = 0 ⇐⇒ 𝑎 = 0 or 𝑏 = 0,
i.e., 𝐴∕𝔭 is an integral domain. Note that if 𝔭 is prime and 𝑎1 ⋯ 𝑎𝑛 ∈ 𝔭, then at least
one of the 𝑎𝑖 ∈ 𝔭.
An ideal 𝔪 in 𝐴 is maximal if it is maximal among the proper ideals of 𝐴. Thus 𝔪
is maximal if and only if 𝐴∕𝔪 is nonzero and has no proper nonzero ideals, and so is a

𝔪 maximal ⇐⇒ 𝔪 prime.
field. Note that

If 𝔞 and 𝔟 are ideals in 𝐴 and 𝐵, then 𝔞 × 𝔟 is an ideal in 𝐴 × 𝐵, and all ideals in

𝐴 × 𝐵 are of this form. To see this, note that if 𝔠 is an ideal in 𝐴 × 𝐵 and (𝑎, 𝑏) ∈ 𝔠, then
(𝑎, 0) = (1, 0)(𝑎, 𝑏) ∈ 𝔠 and (0, 𝑏) = (0, 1)(𝑎, 𝑏) ∈ 𝔠. Therefore, 𝔠 = 𝔞 × 𝔟 with
𝔞 = {𝑎 ∣ (𝑎, 0) ∈ 𝔠}, 𝔟 = {𝑏 ∣ (0, 𝑏) ∈ 𝔠}.
Ideals 𝔞 and 𝔟 in 𝐴 are coprime (or relatively prime) if 𝔞 + 𝔟 = 𝐴. Assume that
𝔞 and 𝔟 are coprime, and let 𝑎 ∈ 𝔞 and 𝑏 ∈ 𝔟 be such that 𝑎 + 𝑏 = 1. For 𝑥, 𝑦 ∈ 𝐴, let
𝑧 = 𝑎𝑦 + 𝑏𝑥; then
𝑧 ≡ 𝑏𝑥 ≡ 𝑥 mod 𝔞
𝑧 ≡ 𝑎𝑦 ≡ 𝑦 mod 𝔟,

𝐴 → 𝐴∕𝔞 × 𝐴∕𝔟
and so the canonical map

is surjective. Clearly its kernel is 𝔞 ∩ 𝔟, which contains 𝔞𝔟. Let 𝑐 ∈ 𝔞 ∩ 𝔟; then

(5)

𝑐 = 𝑐1 = 𝑐𝑎 + 𝑐𝑏 ∈ 𝔞𝔟.
Hence, 𝐴 → 𝐴∕𝔞 × 𝐴∕𝔟 is surjective with kernel 𝔞𝔟. This statement extends to finite
collections of ideals.
14 1. Preliminaries from commutative algebra

Theorem 1.1 (Chinese Remainder Theorem). Let 𝔞1 , … , 𝔞𝑛 be ideals in a ring 𝐴. If

𝔞𝑖 is coprime to 𝔞𝑗 whenever 𝑖 ≠ 𝑗, then the canonical map
𝐴 → 𝐴∕𝔞1 × ⋯ × 𝐴∕𝔞𝑛
∏ ⋂
𝔞𝑖 = 𝔞𝑖 .
(6)
is surjective, with kernel
Proof. We have proved the statement for 𝑛 = 2, and we use induction to extend it to
𝑛 > 2. For 𝑖 ≥ 2, there exist elements 𝑎𝑖 ∈ 𝔞1 and 𝑏𝑖 ∈ 𝔞𝑖 such that
𝑎𝑖 + 𝑏𝑖 = 1.
∏
𝑖≥2
(𝑎𝑖 + 𝑏𝑖 ) lies in 𝔞1 + 𝔞2 ⋯ 𝔞𝑛 and equals 1, and so
𝔞1 + 𝔞2 ⋯ 𝔞𝑛 = 𝐴.
The product

𝐴∕𝔞1 ⋯ 𝔞𝑛 = 𝐴∕𝔞1 ⋅ (𝔞2 ⋯ 𝔞𝑛 )

Therefore,

≃ 𝐴∕𝔞1 × 𝐴∕𝔞2 ⋯ 𝔞𝑛 by the 𝑛 = 2 case

≃ 𝐴∕𝔞1 × 𝐴∕𝔞2 × ⋯ × 𝐴∕𝔞𝑛 by induction. 2

Noetherian rings
Proposition 1.2. The following three conditions on a ring 𝐴 are equivalent:
(a) every ideal in 𝐴 is finitely generated;
(b) every ascending chain of ideals 𝔞1 ⊂ 𝔞2 ⊂ ⋯ eventually becomes constant, i.e.,
𝔞𝑚 = 𝔞𝑚+1 = ⋯ for some 𝑚;
(c) every nonempty set of ideals in 𝐴 has a maximal element.
⋃
Proof. (a) ⇐⇒ (b): Let 𝔞1 ⊂ 𝔞2 ⊂ ⋯ be an ascending chain of ideals. Then 𝔞𝑖 is an
ideal, and hence has a finite set {𝑎1 , … , 𝑎𝑛 } of generators. For some 𝑚, all the 𝑎𝑖 belong
to 𝔞𝑚 , and then
⋃
𝔞𝑚 = 𝔞𝑚+1 = ⋯ = 𝔞𝑖 .
(b) ⇐⇒ (c): Let 𝛴 be a nonempty set of ideals in 𝐴. If 𝛴 has no maximal element, then

chain of ideals in 𝛴, contradicting (b).

the axiom of dependent choice2 implies that there exists an infinite strictly ascending

(c) ⇐⇒ (a): Let 𝔞 be an ideal, and let 𝛴 be the set of finitely generated ideals contained
in 𝔞. Then 𝛴 is nonempty because it contains the zero ideal, and so it contains a maximal
element 𝔠 = (𝑎1 , … , 𝑎𝑟 ). If 𝔠 ≠ 𝔞, then there exists an 𝑎 ∈ 𝔞 ∖ 𝔠, and (𝑎1 , … , 𝑎𝑟 , 𝑎) will be
a finitely generated ideal in 𝔞 properly containing 𝔠. This contradicts the definition of 𝔠,
and so 𝔠 = 𝔞.
A ring 𝐴 is noetherian if every nonempty set of ideals has a maximal element.
2

Applying this to the set of proper ideals containing a fixed ideal, we see that every proper
ideal in a noetherian ring is contained in a maximal ideal. This last assertion is, in fact,

A ring 𝐴 is local if it has exactly one maximal ideal 𝔪𝐴 . If 𝐴 is local, then 𝐴× =

true for all rings, but the proof for non-noetherian rings requires Zorn’s lemma (CA, 2.2).

𝐴 ∖ 𝔪𝐴 . A homomorphism 𝜑 ∶ 𝐴 → 𝐵 of local rings is local if 𝜑(𝔪𝐴 ) ⊂ 𝔪𝐵 , in which

case 𝜑−1 (𝔪𝐵 ) = 𝔪𝐴 .
This says the following: let 𝑅 be a binary relation on a nonempty set 𝑋, and suppose that, for each 𝑎 in
𝑋, there exists a 𝑏 such that 𝑎𝑅𝑏; then there exists a sequence (𝑎𝑛 )𝑛∈ℕ of elements of 𝑋 such that 𝑎𝑛 𝑅𝑎𝑛+1 for
2

all 𝑛. This axiom is strictly weaker than the axiom of choice (Wikipedia: Axiom of dependent choice).
 a. Rings and ideals 15

Proposition 1.3 (Nakayama’s Lemma). Let 𝐴 be a local ring with maximal ideal 𝔪,
and let 𝑀 be a finitely generated 𝐴-module.
(a) If 𝑀 = 𝔪𝑀, then 𝑀 = 0.
(b) If 𝑁 is a submodule of 𝑀 such that 𝑀 = 𝑁 + 𝔪𝑀, then 𝑀 = 𝑁.

Proof. (a) Suppose that 𝑀 ≠ 0. Choose a minimal set of generators {𝑒1 , … , 𝑒𝑛 }, 𝑛 ≥ 1,

for 𝑀, and write
𝑒1 = 𝑎1 𝑒1 + ⋯ + 𝑎𝑛 𝑒𝑛 , 𝑎𝑖 ∈ 𝔪.

(1 − 𝑎1 )𝑒1 = 𝑎2 𝑒2 + ⋯ + 𝑎𝑛 𝑒𝑛
Then

and, as (1−𝑎1 ) ∉ 𝔪, it is a unit, and so 𝑒2 , … , 𝑒𝑛 generate 𝑀, contradicting the minimality

(b) The hypothesis implies that 𝑀∕𝑁 = 𝔪(𝑀∕𝑁), and so 𝑀∕𝑁 = 0.

of the set.
2

Now let 𝐴 be a local noetherian ring with maximal ideal 𝔪. Then 𝔪 is an 𝐴-module,
and the action of 𝐴 on 𝔪∕𝔪2 factors through 𝑘 = 𝐴∕𝔪.
def

Corollary 1.4. Elements 𝑎1 , … , 𝑎𝑛 of 𝔪 generate 𝔪 as an ideal if and only if their

residues modulo 𝔪2 span 𝔪∕𝔪2 as a vector space over 𝑘. In particular, the minimum

𝔪∕𝔪2 .
number of generators for the maximal ideal is equal to the dimension of the vector space

Proof. If 𝑎1 , … , 𝑎𝑛 generate 𝔪, it is obvious that their residues span 𝔪∕𝔪2 . Con-

versely, suppose that their residues span 𝔪∕𝔪2 , so that 𝔪 = (𝑎1 , … , 𝑎𝑛 ) + 𝔪2 . Be-
cause 𝐴 is noetherian, 𝔪 is finitely generated, and Nakayama’s lemma shows that
𝔪 = (𝑎1 , … , 𝑎𝑛 ). 2

Definition 1.5. Let 𝐴 be a noetherian ring.

(a) The height ht(𝔭) of a prime ideal 𝔭 in 𝐴 is the greatest length 𝑑 of a chain of

𝔭 = 𝔭𝑑 ⊃ 𝔭𝑑−1 ⊃ ⋯ ⊃ 𝔭0 .
distinct prime ideals

(b) The Krull dimension, or simply dimension, dim(𝐴), of 𝐴 is sup{ht(𝔭) ∣ 𝔭 a prime

(7)

ideal in 𝐴}.

Thus, the Krull dimension of a noetherian ring 𝐴 is the supremum of the lengths of
chains of prime ideals in 𝐴 (the length of a chain is the number of gaps). For example, a
field has Krull dimension 0, and conversely an integral domain of Krull dimension 0 is a
field. The height of every nonzero prime ideal in a principal ideal domain is 1, and so
such a ring has Krull dimension 1 (unless it is a field).

of the ring may be infinite because it may contain a sequence of prime ideals 𝔭1 , 𝔭2 , 𝔭3 , …
The height of every prime ideal in a noetherian ring is finite, but the Krull dimension

such that ht(𝔭𝑖 ) tends to infinity (CA, p. 13).

Definition 1.6. A regular local ring 𝐴 is a noetherian local ring whose maximal ideal
can be generated by 𝑑 elements, where 𝑑 is the Krull dimension of 𝐴.

dimension is equal to the dimension of the vector space 𝔪∕𝔪2 .

It follows from Corollary 1.4 that a local noetherian ring is regular if and only if its
16 1. Preliminaries from commutative algebra

Lemma 1.7. In a noetherian ring, every set of generators for an ideal contains a finite
generating subset.

Proof. Let 𝔞 be an ideal in a noetherian ring 𝐴, and let 𝑆 be a set of generators for 𝔞. An
ideal maximal among those generated by a finite subset of 𝑆 must contain every element
of 𝑆 (otherwise it would not be maximal), and so equals 𝔞. 2

In the proof of the next theorem, we use that a polynomial ring over a noetherian

Theorem 1.8 (Krull Intersection Theorem). Let 𝐴 be a noetherian local ring with
ring is noetherian (Theorem 2.8).

⋂
maximal ideal 𝔪; then 𝑛≥1 𝔪𝑛 = {0}.

Proof. Let 𝑎1 , … , 𝑎𝑟 generate 𝔪. Then 𝔪𝑛 consists of all finite sums

∑
𝑐𝑖1 ⋯𝑖𝑟 𝑎11 ⋯ 𝑎𝑟𝑟 , 𝑐𝑖1 ⋯𝑖𝑟 ∈ 𝐴.
𝑖 𝑖

𝑖1 +⋯+𝑖𝑟 =𝑛

In other words, 𝔪𝑛 consists of the elements of 𝐴 of the form 𝑔(𝑎1 , … , 𝑎𝑟 ) for some
homogeneous polynomial 𝑔(𝑋1 , … , 𝑋𝑟 ) ∈ 𝐴[𝑋1 , … , 𝑋𝑟 ] of degree 𝑛. Let 𝑆𝑚 denote the
⋂
set of homogeneous polynomials 𝑓 of degree 𝑚 such that 𝑓(𝑎1 , … , 𝑎𝑟 ) ∈ 𝑛≥1 𝔪𝑛 , and
⋃
let 𝔞 be the ideal in 𝐴[𝑋1 , … , 𝑋𝑟 ] generated by the set 𝑚 𝑆𝑚 . According to the lemma, 𝔞
⋃
is generated by a finite subset {𝑓1 , … , 𝑓𝑠 } of 𝑚 𝑆𝑚 . Let 𝑑𝑖 = deg 𝑓𝑖 , and let 𝑑 = max 𝑑𝑖 .
⋂
If 𝑏 ∈ 𝑛≥1 𝔪𝑛 , then 𝑏 ∈ 𝔪𝑑+1 , and so 𝑏 = 𝑓(𝑎1 , … , 𝑎𝑟 ) for some homogeneous
polynomial 𝑓 of degree 𝑑 + 1. By definition, 𝑓 ∈ 𝑆𝑑+1 ⊂ 𝔞, and so

𝑓 = 𝑔1 𝑓1 + ⋯ + 𝑔𝑠 𝑓𝑠

for some 𝑔𝑖 ∈ 𝐴[𝑋1 , … , 𝑋𝑟 ]. As 𝑓 and the 𝑓𝑖 are homogeneous, we can omit from each
𝑔𝑖 all terms not of degree deg 𝑓 − deg 𝑓𝑖 , since these terms cancel out. Thus, we may
choose the 𝑔𝑖 to be homogeneous of degree deg 𝑓 − deg 𝑓𝑖 = 𝑑 + 1 − 𝑑𝑖 > 0. Then
𝑔𝑖 (𝑎1 , … , 𝑎𝑟 ) ∈ 𝔪, and so
∑ ⋂
𝑏 = 𝑓(𝑎1 , … , 𝑎𝑟 ) = 𝑔𝑖 (𝑎1 , … , 𝑎𝑟 ) ⋅ 𝑓𝑖 (𝑎1 , … , 𝑎𝑟 ) ∈ 𝔪 ⋅ 𝔪𝑛 .
𝑖 𝑛≥1
⋂ ⋂ ⋂
Thus, 𝔪𝑛 = 𝔪 ⋅ 𝔪𝑛 , and Nakayama’s lemma implies that 𝔪𝑛 = 0. 2

Aside 1.9. Let 𝐴 be the ring of germs of analytic functions at 0 ∈ ℝ (see p. 60 for the notion of
a germ of a function). Then 𝐴 is a noetherian local ring with maximal ideal 𝔪 = (𝑥), and 𝔪𝑛
consists of the functions 𝑓 that vanish to order 𝑛 at 𝑥 = 0. The theorem says (correctly) that only
the zero function vanishes to all orders at 0. By contrast, the function 𝑒−1∕𝑥 shows that the Krull
2

intersection theorem fails for the ring of germs of infinitely differentiable functions at 0 (this
ring is not noetherian).

b. Rings of fractions
A multiplicative subset of a ring 𝐴 is a subset 𝑆 with the property:

1 ∈ 𝑆, 𝑎, 𝑏 ∈ 𝑆 ⇐⇒ 𝑎𝑏 ∈ 𝑆.

Define an equivalence relation on 𝐴 × 𝑆 by

(𝑎, 𝑠) ∼ (𝑏, 𝑡) ⇐⇒ 𝑢(𝑎𝑡 − 𝑏𝑠) = 0 for some 𝑢 ∈ 𝑆.

 b. Rings of fractions 17

Write 𝑎𝑠 for the equivalence class containing (𝑎, 𝑠), and define addition and multiplication

𝑎 𝑏 𝑎𝑡 + 𝑏𝑠 𝑎𝑏 𝑎𝑏
of equivalence classes in the way suggested by the notation:

𝑠 + 𝑡 = 𝑠𝑡
, 𝑠 𝑡 = 𝑠𝑡 .
It is easy to check that these do not depend on the choices of representatives for the

{ }
𝑎
𝑆 −1 𝐴 = 𝑠 ∣ 𝑎 ∈ 𝐴, 𝑠 ∈ 𝑆
equivalence classes, and that we obtain in this way a ring

and a ring homomorphism 𝑎 ↦ 𝑎 1 ∶ 𝐴 → 𝑆 𝐴, whose kernel is

−1

{𝑎 ∈ 𝐴 ∣ 𝑠𝑎 = 0 for some 𝑠 ∈ 𝑆}.

If 𝐴 is an integral domain and 0 ∉ 𝑆, then 𝑎 ↦ 𝑎 1 is injective. On the other hand, if
0 ∈ 𝑆, then 𝑆 𝐴 is the zero ring.
−1

Write 𝑖 for the homomorphism 𝑎 ↦ 𝑎 1 ∶ 𝐴 → 𝑆 𝐴.

−1

Proposition 1.10. The pair (𝑆 −1 𝐴, 𝑖) has the following universal property: every element
𝑠 ∈ 𝑆 maps to a unit in 𝑆 −1 𝐴, and any other homomorphism 𝛼 ∶ 𝐴 → 𝐵 with this property
factors uniquely through 𝑖,
𝐴 → 𝑆 −1 𝐴
← 𝑖

∃! 𝛽
→
←

←
𝛼

𝐵
→

Proof. If 𝛽 exists, then

𝑎 𝑎 𝑎
𝑠 𝑠 = 𝑎 ⇐⇒ 𝛽(𝑠)𝛽( 𝑠 ) = 𝛽(𝑎) ⇐⇒ 𝛽( 𝑠 ) = 𝛼(𝑎)𝛼(𝑠)−1 ,

and so 𝛽 is unique. Define

𝑎
𝛽( 𝑠 ) = 𝛼(𝑎)𝛼(𝑠)−1 .

𝑎 𝑏
= ⇐⇒ 𝑠(𝑎𝑑 − 𝑏𝑐) = 0 some 𝑠 ∈ 𝑆 ⇐⇒ 𝛼(𝑎)𝛼(𝑑) − 𝛼(𝑏)𝛼(𝑐) = 0
Then

𝑐 𝑑
because 𝛼(𝑠) is a unit in 𝐵, and so 𝛽 is well-defined. It is obviously a homomorphism.2
As usual, the universal property determines the pair (𝑆 −1 𝐴, 𝑖) uniquely up to a unique

If 𝐴 is an integral domain and 𝑆 = 𝐴 ∖ {0}, then 𝐹 = 𝑆 −1 𝐴 is the field of fractions

isomorphism.

of 𝐴. In this case, for any other multiplicative subset 𝑇 of 𝐴 not containing 0, the ring
𝑇 −1 𝐴 can be identified with the subring { 𝑎𝑡 ∈ 𝐹 ∣ 𝑎 ∈ 𝐴, 𝑡 ∈ 𝑆} of 𝐹.

Example 1.11. Let ℎ ∈ 𝐴. Then 𝑆ℎ = {1, ℎ, ℎ2 , …} is a multiplicative subset of 𝐴, and

We shall be especially interested in the following examples.

we let 𝐴ℎ = 𝑆ℎ−1 𝐴. Thus every element of 𝐴ℎ can be written in the form 𝑎𝑚 , 𝑎 ∈ 𝐴, and
def

ℎ
𝑎 𝑏
= 𝑛 ⇐⇒ ℎ𝑁 (𝑎ℎ𝑛 − 𝑏ℎ𝑚 ) = 0, some 𝑁.
ℎ 𝑚 ℎ
If ℎ is nilpotent, then 𝐴ℎ = 0, and if 𝐴 is an integral domain with field of fractions 𝐹 and
ℎ ≠ 0, then 𝐴ℎ is the subring of 𝐹 of elements of the form 𝑎𝑚 , 𝑎 ∈ 𝐴, 𝑚 ∈ ℕ.
ℎ
18 1. Preliminaries from commutative algebra

Example 1.12. Let 𝔭 be a prime ideal in 𝐴. Then 𝑆𝔭 = 𝐴 ∖ 𝔭 is a multiplicative subset

of 𝐴, and we let 𝐴𝔭 = 𝑆𝔭−1 𝐴. Thus each element of 𝐴𝔭 can be written in the form 𝑎𝑐 ,
def

𝑐 ∉ 𝔭, and
𝑎 = 𝑏 ⇐⇒ 𝑠(𝑎𝑑 − 𝑏𝑐) = 0, some 𝑠 ∉ 𝔭.
𝑐 𝑑
{ }
The subset 𝔪 = 𝑎𝑠 ∣ 𝑎 ∈ 𝔭, 𝑠 ∉ 𝔭 is a maximal ideal in 𝐴𝔭 , and it is the only maximal
ideal, i.e., 𝐴𝔭 is a local ring.3 When 𝐴 is an integral domain with field of fractions 𝐹, 𝐴𝔭
is the subring of 𝐹 consisting of elements expressible in the form 𝑎𝑠 , 𝑎 ∈ 𝐴, 𝑠 ∉ 𝔭.
∑ ∑ 𝑎𝑖
Lemma 1.13. For any ring 𝐴 and ℎ ∈ 𝐴, the map 𝑎𝑖 𝑋 𝑖 ↦
ℎ𝑖
defines an isomorphism

𝐴[𝑋]∕(1 − ℎ𝑋) ,→ 𝐴ℎ .
≃

Proof. If ℎ = 0, then both rings are zero, so we may suppose that ℎ ≠ 0. Let 𝑥 be the
class of 𝑋 in the quotient ring 𝐴[𝑋]∕(1 − ℎ𝑋). Then 𝐴[𝑥] is generated by 𝑥 subject to
the relation 1 = ℎ𝑥, and so ℎ is a unit. Let 𝛼 ∶ 𝐴 → 𝐵 be a homomorphism of rings
∑ ∑
such that 𝛼(ℎ) is a unit in 𝐵. The homomorphism 𝑎𝑖 𝑋 𝑖 ↦ 𝛼(𝑎𝑖 )𝛼(ℎ)−𝑖 ∶ 𝐴[𝑋] → 𝐵
factors through 𝐴[𝑥] because 1 − ℎ𝑋 ↦ 1 − 𝛼(ℎ)𝛼(ℎ)−1 = 0, and, because 𝛼(ℎ) is a unit
in 𝐵, this is the unique extension of 𝛼 to 𝐴[𝑥]. Therefore 𝐴[𝑥] has the same universal
property as 𝐴ℎ , and so the two are (uniquely) isomorphic by an isomorphism that fixes
elements of 𝐴 and makes ℎ−1 correspond to 𝑥.
Let 𝑆 be a multiplicative subset of a ring 𝐴, and let 𝑆 −1 𝐴 be the corresponding ring of
2

fractions. Each ideal 𝔞 in 𝐴, generates an ideal 𝑆 −1 𝔞 in 𝑆 −1 𝐴. If 𝔞 contains an element of

𝑆, then 𝑆 −1 𝔞 contains a unit, and so is the whole ring. Thus some of the ideal structure
of 𝐴 is lost in the passage to 𝑆 −1 𝐴, but, as the next proposition shows, much is retained.
Proposition 1.14. Let 𝑆 be a multiplicative subset of the ring 𝐴. The map
𝔭 ↦ 𝑆 −1 𝔭 = (𝑆 −1 𝐴)𝔭
is a bijection from the set of prime ideals of 𝐴 disjoint from 𝑆 to the set of prime ideals of
𝑆 −1 𝐴. Its inverse sends a prime ideal of 𝑆 −1 𝐴 to its inverse image in 𝐴.
Proof. For an ideal 𝔟 of 𝑆 −1 𝐴, let 𝔟𝑐 denote the inverse image of 𝔟 in 𝐴, and for an ideal
𝔞 of 𝐴, let 𝔞𝑒 = (𝑆 −1 𝐴)𝔞 denote the ideal in 𝑆 −1 𝐴 generated by the image of 𝔞.
For an ideal 𝔟 of 𝑆 −1 𝐴, certainly, 𝔟 ⊃ 𝔟𝑐𝑒 . Conversely, if 𝑎𝑠 ∈ 𝔟, 𝑎 ∈ 𝐴, 𝑠 ∈ 𝑆, then
𝑎 ∈ 𝔟, and so 𝑎 ∈ 𝔟𝑐 . Thus 𝑎 ∈ 𝔟𝑐𝑒 , and so 𝔟 = 𝔟𝑐𝑒 .
1 𝑠
For an ideal 𝔞 of 𝐴, certainly 𝔞 ⊂ 𝔞𝑒𝑐 . Conversely, if 𝑎 ∈ 𝔞𝑒𝑐 , then 𝑎 1 ∈ 𝔞 , and so
𝑒

𝑎 = 𝑎′ for some 𝑎′ ∈ 𝔞, 𝑠 ∈ 𝑆. Thus, 𝑡(𝑎𝑠 − 𝑎′ ) = 0 for some 𝑡 ∈ 𝑆, and so 𝑎𝑠𝑡 ∈ 𝔞. If 𝔞

1 𝑠
is a prime ideal disjoint from 𝑆, this implies that 𝑎 ∈ 𝔞: for such an ideal, 𝔞 = 𝔞𝑒𝑐 .
If 𝔟 is prime, then certainly 𝔟𝑐 is prime. For any ideal 𝔞 of 𝐴, 𝑆 −1 𝐴∕𝔞𝑒 ≃ 𝑆̄ −1 (𝐴∕𝔞),
where 𝑆̄ is the image of 𝑆 in 𝐴∕𝔞. If 𝔞 is a prime ideal disjoint from 𝑆, then 𝑆̄ −1 (𝐴∕𝔞) is
a subring of the field of fractions of 𝐴∕𝔞, and is therefore an integral domain. Thus, 𝔞𝑒 is

We have shown that 𝔭 ↦ 𝔭𝑒 and 𝔮 ↦ 𝔮𝑐 are inverse bijections between the prime
prime.

ideals of 𝐴 disjoint from 𝑆 and the prime ideals of 𝑆 −1 𝐴.

First check 𝔪 is an ideal. Next, if 𝔪 = 𝐴𝔭 , then 1 ∈ 𝔪; but if 1 = 𝑎𝑠 for some 𝑎 ∈ 𝔭 and 𝑠 ∉ 𝔭, then
2

𝑢(𝑠 − 𝑎) = 0 some 𝑢 ∉ 𝔭, and so 𝑢𝑎 = 𝑢𝑠 ∉ 𝔭, which contradicts 𝑎 ∈ 𝔭. Finally, 𝔪 is maximal because

3

every element of 𝐴𝔭 not in 𝔪 is a unit.

 b. Rings of fractions 19

Lemma 1.15. Let 𝔪 be a maximal ideal of a ring 𝐴, and let 𝔫 = 𝔪𝐴𝔪 . For all 𝑛, the map

𝑎 + 𝔪𝑛 ↦ 𝑎
1 + 𝔫 ∶ 𝐴∕𝔪 → 𝐴𝔪 ∕𝔫
𝑛 𝑛 𝑛 (8)

is an isomorphism. Moreover, it induces isomorphisms

𝔪𝑟 ∕𝔪𝑛 → 𝔫𝑟 ∕𝔫𝑛

for all 𝑟 < 𝑛.

diagram (𝑟 < 𝑛):

Proof. The second statement follows from the first, because of the exact commutative

0 → 𝔪𝑟 ∕𝔪𝑛 → 𝐴∕𝔪𝑛 → 𝐴∕𝔪𝑟 → 0

← ← ← ←

≃ ≃
←

←
0 → 𝔫𝑟 ∕𝔫𝑛 → 𝐴𝔪 ∕𝔫𝑛 → 𝐴𝔪 ∕𝔫𝑟 → 0.
→

→
← ← ← ←

{ }
Let 𝑆 = 𝐴 ∖ 𝔪. Then 𝐴𝔪 = 𝑆 −1 𝐴 and 𝔫𝑛 = 𝔪𝑛 𝐴𝔪 = 𝑏𝑠 ∈ 𝐴𝔪 ∣ 𝑏 ∈ 𝔪𝑛 , 𝑠 ∈ 𝑆 . In

𝑎 = 𝑏 with 𝑎 ∈ 𝐴, 𝑏 ∈ 𝔪𝑛 , 𝑠 ∈ 𝑆 ⇐⇒ 𝑎 ∈ 𝔪𝑛 .
order to show that the map (8) is injective, it suffices to show that

1 𝑠

But if 𝑎 𝑏
1 = 𝑠 , then 𝑡𝑎𝑠 = 𝑡𝑏 ∈ 𝔪 for some 𝑡 ∈ 𝑆, and so 𝑡𝑎𝑠 = 0 in 𝐴∕𝔪 . The only
𝑛 𝑛

maximal ideal in 𝐴 containing 𝔪𝑚 is 𝔪 (because 𝔪′ ⊃ 𝔪𝑚 ⇐⇒ 𝔪′ ⊃ 𝔪), and so the

only maximal ideal in 𝐴∕𝔪𝑛 is 𝔪∕𝔪𝑛 . As 𝑠𝑡 is not in 𝔪∕𝔪𝑛 , it must be a unit in 𝐴∕𝔪𝑛 ,
and as 𝑠𝑡𝑎 = 0 in 𝐴∕𝔪𝑛 , 𝑎 must be 0 in 𝐴∕𝔪𝑛 , i.e., 𝑎 ∈ 𝔪𝑛 .
We now prove that the map (8) is surjective. Let 𝑎𝑠 ∈ 𝐴𝔪 , 𝑎 ∈ 𝐴, 𝑠 ∈ 𝑆. Because
the only maximal ideal of 𝐴 containing 𝔪𝑛 is 𝔪, no maximal ideal contains both 𝑠 and
𝔪𝑛 . It follows that (𝑠) + 𝔪𝑛 = 𝐴. Therefore, there exist 𝑏 ∈ 𝐴 and 𝑞 ∈ 𝔪𝑛 such that
𝑠𝑏 + 𝑞 = 1 in 𝐴. It follows that 𝑠 is invertible in 𝐴𝔪 ∕𝔫𝑛 , and so 𝑎𝑠 is the unique element
of this ring such that 𝑠 𝑎𝑠 = 𝑎. As 𝑠(𝑏𝑎) + 𝑞𝑎 = 𝑎, the image of 𝑏𝑎 in 𝐴𝔪 ∕𝔫𝑛 also has
this property and therefore equals 𝑎𝑠 in 𝐴𝔪 ∕𝔫𝑛 . 2

Proposition 1.16. In a noetherian ring, only 0 lies in all powers of all maximal ideals.

Proof. Let 𝑎 be an element of a noetherian ring 𝐴. If 𝑎 ≠ 0, then {𝑏 ∈ 𝐴 ∣ 𝑏𝑎 = 0} is a

proper ideal, and so is contained in some maximal ideal 𝔪. Then 𝑎 1 is nonzero in 𝐴𝔪 ,
𝑎
and so 1 ∉ (𝔪𝐴𝔪 ) for some 𝑛 (by the Krull intersection theorem 1.8), which implies
𝑛

that 𝑎 ∉ 𝔪𝑛 (by 1.15). 2

Let 𝐴 be an integral domain and let 𝑓 be an element of its field of fractions. If 𝐴 is

a unique factorization domain (see below), then there is a preferred expression 𝑓 = 𝑎
𝑏
for 𝑓, unique up to multiplying top and bottom by a unit. In general, there is no such
preferred expression.
20 1. Preliminaries from commutative algebra

Modules of fractions
Let 𝑆 be a multiplicative subset of the ring 𝐴, and let 𝑀 be an 𝐴-module. Define an
equivalence relation on 𝑀 × 𝑆 by

(𝑚, 𝑠) ∼ (𝑛, 𝑡) ⇐⇒ 𝑢(𝑡𝑚 − 𝑠𝑛) = 0 for some 𝑢 ∈ 𝑆.

Write 𝑚𝑠 for the equivalence class containing (𝑚, 𝑠), and define addition and scalar

𝑚 + 𝑛 = 𝑚𝑡 + 𝑛𝑠 , 𝑎 𝑚 = 𝑎𝑚 ,
multiplication by the rules:

𝑠 𝑡 𝑠𝑡 𝑠 𝑡 𝑠𝑡 𝑚, 𝑛 ∈ 𝑀, 𝑠, 𝑡 ∈ 𝑆, 𝑎 ∈ 𝐴.

equivalence classes, and that we obtain in this way an 𝑆 −1 𝐴-module

It is easily checked that these do not depend on the choices of representatives for the

{ }
𝑚
𝑆 −1 𝑀 = 𝑠 ∣ 𝑚 ∈ 𝑀, 𝑠 ∈ 𝑆

and a homomorphism 𝑚 ↦ 𝑚
𝑖𝑆
1 ∶ 𝑀 ,→ 𝑆 𝑀 of 𝐴-modules whose kernel is
−1

{𝑎 ∈ 𝑀 ∣ 𝑠𝑎 = 0 for some 𝑠 ∈ 𝑆}.

Proposition 1.17. The elements of 𝑆 act invertibly on 𝑆 −1 𝑀, and every homomorphism

from 𝑀 to an 𝐴-module 𝑁 with this property factors uniquely through 𝑖𝑆 ,

𝑀 → 𝑆 −1 𝑀
← 𝑖𝑆

→ ∃!
←
←

𝑁.
→

Proof. Similar to the proof of 1.10.

Proposition 1.18. The functor 𝑀 ⇝ 𝑆 −1 𝑀 is exact. In other words, if the sequence of

2

𝐴-modules
𝛼 𝛽
𝑀 ′ ,→ 𝑀 ,→ 𝑀 ′′
is exact, then so also is the sequence of 𝑆 −1 𝐴-modules
𝑆 −1 𝛼 𝑆 −1 𝛽
𝑆 −1 𝑀 ′ ,,,,→ 𝑆 −1 𝑀 ,,,,→ 𝑆 −1 𝑀 ′′ .

Proof. Because 𝛽◦𝛼 = 0, we have 0 = 𝑆 −1 (𝛽◦𝛼) = 𝑆 −1 𝛽◦𝑆 −1 𝛼. Therefore Im(𝑆 −1 𝛼) ⊂

Ker(𝑆 −1 𝛽). For the reverse inclusion, let 𝑚
𝑠 ∈ Ker(𝑆 𝛽), where 𝑚 ∈ 𝑀 and 𝑠 ∈ 𝑆.
−1

𝛽(𝑚)
Then 𝑠 = 0 and so, for some 𝑡 ∈ 𝑆, we have 𝑡𝛽(𝑚) = 0. Then 𝛽(𝑡𝑚) = 0, and so
𝑡𝑚 = 𝛼(𝑚′ ) for some 𝑚′ ∈ 𝑀 ′ . Now

𝑚 𝑡𝑚 𝛼(𝑚′ )
𝑠 = = ∈ Im(𝑆 −1 𝛼).
𝑡𝑠 𝑡𝑠 2

Proposition 1.19. Let 𝐴 be a ring, and let 𝑀 be an 𝐴-module. The canonical map
∏
𝑀→ {𝑀𝔪 ∣ 𝔪 a maximal ideal in 𝐴}

is injective.
 b. Rings of fractions 21

Proof. Let 𝑚 ∈ 𝑀 map to zero in all 𝑀𝔪 . The annihilator 𝔞 = {𝑎 ∈ 𝐴 ∣ 𝑎𝑚 = 0} of

𝑚 is an ideal in 𝐴. Because 𝑚 maps to zero 𝑀𝔪 , there exists an 𝑠 ∈ 𝐴 ∖ 𝔪 such that
def

𝑠𝑚 = 0. Therefore 𝔞 is not contained in 𝔪. Since this is true for all maximal ideals 𝔪,
𝔞 = 𝐴, and so it contains 1. Now 𝑚 = 1𝑚 = 0. 2

Corollary 1.20. An 𝐴-module 𝑀 is zero if 𝑀𝔪 is zero for all maximal ideals 𝔪 in 𝐴.

Proof. Immediate consequence of the proposition. 2

Proposition 1.21. Let 𝐴 be a ring. A sequence of 𝐴-modules

𝛼 𝛽
𝑀 ′ ,→ 𝑀 ,→ 𝑀 ′′ (*)

𝛼𝔪 𝛽𝔪
𝑀𝔪 ,→ 𝑀𝔪 ,→ 𝑀𝔪
is exact if and only if
′ ′′

is exact for all maximal ideals 𝔪.

(**)

to show that 𝑁 = Ker(𝛽)∕ Im(𝛼) is zero. Because the functor 𝑀 ⇝ 𝑀𝔪 is exact,

Proof. The necessity is a special case of Proposition 1.18. For the sufficiency, we have
def

𝑁𝔪 = Ker(𝛽𝔪 )∕ Im(𝛼𝔪 ).

If (**) is exact for all 𝔪, then 𝑁𝔪 = 0 for all 𝔪, and so 𝑁 = 0 (by 1.20). 2

Corollary 1.22. A homomorphism 𝑀 → 𝑁 of 𝐴-modules is injective (resp. surjective) if

and only if 𝑀𝔪 → 𝑁𝔪 is injective (resp. surjective) for all maximal ideals 𝔪.

Proof. Apply the proposition to 0 → 𝑀 → 𝑁 (resp. 𝑀 → 𝑁 → 0). 2

Direct limits
A directed set is a pair (𝐼, ≤) comprising a set 𝐼 and a partial order4 ≤ on 𝐼 such that for
all 𝑖, 𝑗 ∈ 𝐼, there exists a 𝑘 ∈ 𝐼 with 𝑖, 𝑗 ≤ 𝑘.
Let (𝐼, ≤) be a directed set, and let 𝐴 be a ring. A direct system of 𝐴-modules indexed
by (𝐼, ≤) is a family (𝑀𝑖 )𝑖∈𝐼 of 𝐴-modules together with a family (𝛼𝑗𝑖 ∶ 𝑀𝑖 → 𝑀𝑗 )𝑖≤𝑗 of
𝐴-linear maps such that 𝛼𝑖𝑖 = id𝑀𝑖 and 𝛼𝑘 ◦𝛼𝑗𝑖 = 𝛼𝑘𝑖 all 𝑖 ≤ 𝑗 ≤ 𝑘.5 An 𝐴-module 𝑀
𝑗

together with 𝐴-linear maps 𝛼𝑖 ∶ 𝑀𝑖 → 𝑀 such that 𝛼𝑖 = 𝛼𝑗 ◦𝛼𝑗𝑖 for all 𝑖 ≤ 𝑗 is the direct
limit (or colimit) of the system (𝑀𝑖 , 𝛼𝑖 ) if
𝑗
⋃
(a) 𝑀 = 𝑖∈𝐼 𝛼𝑖 (𝑀𝑖 ), and
(b) 𝑚𝑖 ∈ 𝑀𝑖 maps to zero in 𝑀 if and only if it maps to zero in 𝑀𝑗 for some 𝑗 ≥ 𝑖.
Direct limits of 𝐴-algebras are defined similarly.

Proposition 1.23. Let 𝑆 be a multiplicative subset of 𝐴. Then 𝑆 −1 𝐴 ≃ lim 𝐴ℎ , where ℎ

,,→
runs over the elements of 𝑆 (partially ordered by division).

Regard 𝐼 as a category with Hom(𝑎, 𝑏) empty unless 𝑎 ≤ 𝑏, in which case it contains a single element.
4
i.e., reflexive transitive antisymmetric.

Then a direct system is a functor from 𝐼 to the category of 𝐴-modules.

5
22 1. Preliminaries from commutative algebra

𝑎𝑔
Proof. When ℎ|ℎ′ , say, ℎ′ = ℎ𝑔, there is a canonical homomorphism 𝑎 ↦ ′ ∶ 𝐴ℎ →
ℎ ℎ
𝐴ℎ′ . and so the rings 𝐴ℎ form a direct system indexed by the set 𝑆. When ℎ ∈ 𝑆, the
homomorphism 𝐴 → 𝑆 −1 𝐴 extends uniquely to a homomorphism 𝑎 ↦ 𝑎 ∶ 𝐴ℎ → 𝑆 −1 𝐴
ℎ ℎ

Now it is easy to see that 𝑆 −1 𝐴 satisfies the conditions to be the direct limit of the 𝐴ℎ .2
(1.10), and these homomorphisms are compatible with the maps in the direct system.

c. Unique factorization
Let 𝐴 be an integral domain. An element 𝑎 of 𝐴 is irreducible if it is not zero, not a unit,

𝑎 = 𝑏𝑐 ⇐⇒ 𝑏 or 𝑐 is a unit.
and admits only trivial factorizations, i.e.,

An element 𝑎 is said to be prime if (𝑎) is a prime ideal, i.e.,

𝑎|𝑏𝑐 ⇐⇒ 𝑎|𝑏 or 𝑎|𝑐.

If 𝐴 is noetherian, then every nonzero nonunit element can be expressed as a finite

product of irreducible elements. To see this, suppose that 𝑎 cannot be so expressed
and that (𝑎) is maximal among the ideals generated by such elements; then 𝑎 is not
irreducible, so 𝑎 = 𝑏𝑐 with neither 𝑏 nor 𝑐 a unit; each of (𝑏) and (𝑐) properly contains
(𝑎), and at least one of 𝑏 or 𝑐 is not a finite product of irreducible elements, giving a

An integral domain 𝐴 is called a unique factorization domain (or a factorial

contradiction.

domain) if every nonzero nonunit in 𝐴 can be written as a finite product of irreducible

domains, for example, ℤ and 𝑘[𝑋], are unique factorization domains,

elements in exactly one way up to units and the order of the factors. Principal ideal

Proposition 1.24. Let 𝐴 be an integral domain, and let 𝑎 be an element of 𝐴 that is

neither zero nor a unit. If 𝑎 is prime, then 𝑎 is irreducible, and the converse holds when 𝐴
is a unique factorization domain.

Proof. Assume that 𝑎 is prime. If 𝑎 = 𝑏𝑐, then 𝑎 divides 𝑏𝑐 and so 𝑎 divides 𝑏 or 𝑐.

Suppose the first, and write 𝑏 = 𝑎𝑞. Now 𝑎 = 𝑏𝑐 = 𝑎𝑞𝑐, which implies that 𝑞𝑐 = 1
because 𝐴 is an integral domain, and so 𝑐 is a unit. We have shown that 𝑎 is irreducible.
For the converse, assume that 𝑎 is irreducible and that 𝐴 is a unique factorization
domain. If 𝑎|𝑏𝑐, then
𝑏𝑐 = 𝑎𝑞, some 𝑞 ∈ 𝐴.
On writing each of 𝑏, 𝑐, and 𝑞 as a product of irreducible elements, and using the
uniqueness of factorizations, we see that 𝑎 differs from one of the irreducible factors of
𝑏 or 𝑐 by a unit. Therefore 𝑎 divides 𝑏 or 𝑐.

Corollary 1.25. Let 𝐴 be a noetherian integral domain. If 𝐴 is a unique factorization

2

domain, then every prime ideal of height 1 is principal.

Proof. Let 𝔭 be a prime ideal of height 1. Then 𝔭 contains a nonzero element, and
hence an irreducible element 𝑎. We have 𝔭 ⊃ (𝑎) ⊃ (0). As (𝑎) is prime and 𝔭 has height
1, we must have 𝔭 = (𝑎).

Proposition 1.26. Let 𝐴 be a noetherian integral domain. If every irreducible element of

2

𝐴 is prime, then 𝐴 is a unique factorization domain.

 c. Unique factorization 23

𝑎1 ⋯ 𝑎𝑚 = 𝑏1 ⋯ 𝑏𝑛
Proof. Suppose that

with the 𝑎𝑖 and 𝑏𝑖 irreducible elements in 𝐴. As 𝑎1 is prime, it divides one of the 𝑏𝑖 , say,
(9)

𝑏1 . As 𝑏1 is irreducible, 𝑏1 = 𝑢𝑎1 for some unit 𝑢. On cancelling 𝑎1 from both sides of

(9), we obtain the equality

𝑎2 ⋯ 𝑎𝑚 = (𝑢𝑏2 )𝑏3 ⋯ 𝑏𝑛 .

Continuing in this fashion, we find that the two factorizations are the same up to units
and the order of the factors. 2

Aside. The converse to 1.25 is also true: let 𝑎 be an irreducible element of 𝐴, and let 𝔭 be
minimal among the prime ideals containing (𝑎); according to the principal ideal theorem (3.51;
CA, 21.3), 𝔭 has height 1, and so is principal, say, 𝔭 = (𝑏); now 𝑎 = 𝑏𝑐, and, because 𝑎 is
irreducible, 𝑐 is a unit; therefore (𝑎) = (𝑏) = 𝔭, and 𝑎 is prime. See 3.53.

Proposition 1.27 (Gauss’s Lemma). Let 𝐴 be a unique factorization domain with field
of fractions 𝐹. If 𝑓(𝑋) ∈ 𝐴[𝑋] factors into the product of two nonconstant polynomials
in 𝐹[𝑋], then it factors into the product of two nonconstant polynomials in 𝐴[𝑋].

Proof. Let 𝑓 = 𝑔ℎ in 𝐹[𝑋]. For suitable 𝑐, 𝑑 ∈ 𝐴, the polynomials 𝑔1 = 𝑐𝑔 and ℎ1 = 𝑑ℎ

have coefficients in 𝐴, and so we have a factorization

𝑐𝑑𝑓 = 𝑔1 ℎ1 in 𝐴[𝑋].

If an irreducible element 𝑝 of 𝐴 divides 𝑐𝑑, then, looking modulo (𝑝), we see that

0 = 𝑔1 ⋅ ℎ1 in (𝐴∕(𝑝)) [𝑋].

According to Proposition 1.24, (𝑝) is prime, and so (𝐴∕(𝑝)) [𝑋] is an integral domain.
Therefore, 𝑝 divides all the coefficients of at least one of the polynomials 𝑔1 , ℎ1 , say, 𝑔1 ,
so that 𝑔1 = 𝑝𝑔2 for some 𝑔2 ∈ 𝐴[𝑋]. Thus, we have a factorization

(𝑐𝑑∕𝑝)𝑓 = 𝑔2 ℎ1 in 𝐴[𝑋].

Continuing in this fashion, we can remove all the irreducible factors of 𝑐𝑑, and so obtain
a factorization of 𝑓 in 𝐴[𝑋]. 2

Let 𝐴 be a unique factorization domain. A nonzero polynomial

𝑓 = 𝑎0 + 𝑎1 𝑋 + ⋯ + 𝑎𝑚 𝑋 𝑚

in 𝐴[𝑋] is said to be primitive if the coefficients 𝑎𝑖 have no common factor other than

Every polynomial 𝑓 in 𝐹[𝑋] can be written 𝑓 = 𝑐(𝑓) ⋅ 𝑓1 with 𝑐(𝑓) ∈ 𝐹 and 𝑓1

units.

primitive. The element 𝑐(𝑓), which is well-defined up to multiplication by a unit in 𝐴,

is called the content of 𝑓. Note that 𝑓 ∈ 𝐴[𝑋] if and only if 𝑐(𝑓) ∈ 𝐴.

Lemma 1.28. The product of two primitive polynomials is primitive.

24 1. Preliminaries from commutative algebra

𝑓 = 𝑎0 + 𝑎1 𝑋 + ⋯ + 𝑎𝑚 𝑋 𝑚
Proof. Let

𝑔 = 𝑏0 + 𝑏1 𝑋 + ⋯ + 𝑏𝑛 𝑋 𝑛 ,

be primitive polynomials, and let 𝑝 be an irreducible element of 𝐴. Let 𝑎𝑖0 , 𝑖0 ≤ 𝑚, be

the first coefficient of 𝑓 not divisible by 𝑝, and 𝑏𝑗0 , 𝑗0 ≤ 𝑛, the first coefficient of 𝑔 not
∑
divisible by 𝑝. Then all the terms in the sum 𝑖+𝑗=𝑖 +𝑗 𝑎𝑖 𝑏𝑗 are divisible by 𝑝 except
𝑎𝑖0 𝑏𝑗0 , which is not divisible by 𝑝. Therefore, 𝑝 does not divide the (𝑖0 + 𝑗0 )th-coefficient
0 0

of 𝑓𝑔. We have shown that no irreducible element of 𝐴 divides all the coefficients of 𝑓𝑔,
which must therefore be primitive.

Proposition 1.29. Let 𝐴 be a unique factorization domain with field of fractions 𝐹. For
2

polynomials 𝑓, 𝑔 ∈ 𝐹[𝑋],
𝑐(𝑓𝑔) = 𝑐(𝑓) ⋅ 𝑐(𝑔);
hence every factor in 𝐴[𝑋] of a primitive polynomial is primitive.

Proof. Let 𝑓 = 𝑐(𝑓) ⋅ 𝑓1 and 𝑔 = 𝑐(𝑔) ⋅ 𝑔1 with 𝑓1 and 𝑔1 primitive. Then

𝑓𝑔 = 𝑐(𝑓) ⋅ 𝑐(𝑔) ⋅ 𝑓1 𝑔1

with 𝑓1 𝑔1 primitive, and so 𝑐(𝑓𝑔) = 𝑐(𝑓)𝑐(𝑔).

Corollary 1.30. An element 𝑓 ∈ 𝐴[𝑋] is irreducible if and only if either

2

(a) 𝑓 is constant, say, 𝑓 = 𝑎, with 𝑎 an irreducible element of 𝐴, or

(b) 𝑓 is a nonconstant primitive polynomial that is irreducible in 𝐹[𝑋].

Proof. ⇐: If 𝑓 is as in (a) and 𝑓 = 𝑔ℎ in 𝐴[𝑋], then 𝑔 and ℎ both lie in 𝐴 and one must
be a unit in 𝐴, and hence a unit in 𝐴[𝑋]. If 𝑓 is as in (b) and 𝑓 = 𝑔ℎ, then one of 𝑔 or
ℎ must be constant because otherwise 𝑓 would be reducible in 𝐹[𝑋]. If it is 𝑔 that is
constant, then, because 𝑓 is primitive, 𝑔 must be a unit in 𝐴, and hence in 𝐴[𝑋].
⇒: Let 𝑓 ∈ 𝐴[𝑋] be irreducible. If 𝑓 is a constant polynomial, say, 𝑓 = 𝑎, then 𝑎 is
obviously irreducible in 𝐴. If 𝑓 nonconstant, then it must be primitive because otherwise
𝑓 = 𝑐(𝑓) ⋅ 𝑓1 would be a nontrivial factorization in 𝐴[𝑋]. It must also be irreducible in
𝐹[𝑋], because otherwise it would have a nontrivial factorization in 𝐴[𝑋] (by 1.27). 2

Proposition 1.31. If 𝐴 is a unique factorization domain, then so also is 𝐴[𝑋].

Proof. We check that 𝐴 satisfies the conditions of Proposition 1.26.

Let 𝑓 ∈ 𝐴[𝑋], and write 𝑓 = 𝑐(𝑓)𝑓1 . Then 𝑐(𝑓) is a product of irreducible elements
in 𝐴, and 𝑓1 is a product of irreducible primitive polynomials. This shows that 𝑓 is a
product of irreducible elements in 𝐴[𝑋].
Let 𝑎 be an irreducible element of 𝐴. If 𝑎 divides 𝑓𝑔, then it divides 𝑐(𝑓𝑔) = 𝑐(𝑓)𝑐(𝑔).
As 𝑎 is prime (1.24), it divides 𝑐(𝑓) or 𝑐(𝑔), and hence also 𝑓 or 𝑔.
Let 𝑓 be an irreducible primitive polynomial in 𝐴[𝑋]. Then 𝑓 is irreducible in 𝐹[𝑋],
and so if 𝑓 divides the product 𝑔ℎ of 𝑔, ℎ ∈ 𝐴[𝑋], then it divides 𝑔 or ℎ in 𝐹[𝑋]. Suppose
the first, and write 𝑓𝑞 = 𝑔 with 𝑞 ∈ 𝐹[𝑋]. Then 𝑐(𝑞) = 𝑐(𝑓)𝑐(𝑞) = 𝑐(𝑓𝑞) = 𝑐(𝑔) ∈ 𝐴,
and so 𝑞 ∈ 𝐴[𝑋]. Therefore 𝑓 divides 𝑔 in 𝐴[𝑋].
We have shown that every element of 𝐴[𝑋] is a product of irreducible elements and
that every irreducible element of 𝐴[𝑋] is prime, and so 𝐴[𝑋] is a unique factorization
domain (1.26). 2
 d. Integral dependence 25

Polynomial rings
Let 𝑘 be a field. The elements of the polynomial ring 𝑘[𝑋1 , … , 𝑋𝑛 ] are finite sums
∑
𝑐𝑎1 ⋯𝑎𝑛 𝑋1 1 ⋯ 𝑋𝑛 𝑛 , 𝑐𝑎1 ⋯𝑎𝑛 ∈ 𝑘, 𝑎𝑗 ∈ ℕ,
𝑎 𝑎

monomials form a basis for 𝑘[𝑋1 , … , 𝑋𝑛 ] as a 𝑘-vector space.

with the obvious notions of equality, addition, and multiplication. In particular, the

The degree, deg(𝑓), of a nonzero polynomial 𝑓 is the largest total degree of a

monomial occurring in 𝑓 with nonzero coefficient. Since deg(𝑓𝑔) = deg(𝑓) + deg(𝑔),
𝑘[𝑋1 , … , 𝑋𝑛 ] is an integral domain and 𝑘[𝑋1 , … , 𝑋𝑛 ]× = 𝑘× . An element 𝑓 of 𝑘[𝑋1 , … , 𝑋𝑛 ]
is irreducible if it is nonconstant and 𝑓 = 𝑔ℎ ⇐⇒ 𝑔 or ℎ is constant.

Theorem 1.32. The ring 𝑘[𝑋1 , … , 𝑋𝑛 ] is a unique factorization domain.

𝑘[𝑋1 , … , 𝑋𝑛−1 ][𝑋𝑛 ] = 𝑘[𝑋1 , … , 𝑋𝑛 ].

Proof. Note that

This simply says that every polynomial 𝑓 in 𝑛 symbols 𝑋1 , … , 𝑋𝑛 can be expressed

uniquely as a polynomial in 𝑋𝑛 with coefficients in 𝑘[𝑋1 , … , 𝑋𝑛−1 ],

𝑓(𝑋1 , … , 𝑋𝑛 ) = 𝑎0 (𝑋1 , … , 𝑋𝑛−1 )𝑋𝑛𝑟 + ⋯ + 𝑎𝑟 (𝑋1 , … , 𝑋𝑛−1 ).

Since, as we noted, 𝑘[𝑋] is a unique factorization domain, the theorem follows by

induction from Proposition 1.31. 2

Corollary 1.33. A nonzero proper principal ideal (𝑓) in 𝑘[𝑋1 , … , 𝑋𝑛 ] is prime if and
only if 𝑓 is irreducible.

Proof. Special case of Proposition 1.24. 2

d. Integral dependence
Let 𝐴 be a subring of a ring 𝐵. An element 𝛼 of 𝐵 is said to be integral over 𝐴 if it is a
root of a monic6 polynomial with coefficients in 𝐴, i.e., if it satisfies an equation

𝛼𝑛 + 𝑎1 𝛼𝑛−1 + ⋯ + 𝑎𝑛 = 0, 𝑎𝑖 ∈ 𝐴.

More generally, if 𝑓 ∶ 𝐴 → 𝐵 is an 𝐴-algebra, then an element 𝛼 of 𝐵 is integral over 𝐴

if it if it is integral over the subring 𝑓(𝐴) of 𝐵. When every element of 𝐵 is integral over
𝐴, we say that 𝐵 is integral over 𝐴.
In the next proof, we shall need to apply a variant of Cramer’s rule: if 𝑥1 , … , 𝑥𝑚 is a
solution to the system of linear equations

∑
𝑚
𝑐𝑖𝑗 𝑥𝑗 = 0, 𝑖 = 1, … , 𝑚,
𝑗=1

with coefficients in a ring 𝐴, then

det(𝐶) ⋅ 𝑥𝑗 = 0, 𝑗 = 1, … , 𝑚, (10)
6
A polynomial is monic if its leading coefficient is 1, i.e., 𝑓(𝑋) = 𝑋 𝑛 + terms of degree less than 𝑛.
26 1. Preliminaries from commutative algebra

where 𝐶 is the matrix of coefficients. To prove this, expand out the left hand side of
∑
⎛ 𝑐11 … 𝑐1 𝑗−1 𝑐 𝑥 𝑐1 𝑗+1 … 𝑐1𝑚 ⎞
𝑖 1𝑖 𝑖
det ⎜ ⋮ ⋮ ⋮ ⋮ ⋮ ⎟=0
∑
𝑐
⎝ 𝑚1 … 𝑐𝑚 𝑗−1 𝑐 𝑥
𝑖 𝑚𝑖 𝑖
𝑐 𝑚 𝑗+1 … 𝑐𝑚𝑚 ⎠

An 𝐴-module 𝑀 is faithful if 𝑎𝑀 = 0, 𝑎 ∈ 𝐴, implies that 𝑎 = 0.

using standard properties of determinants.

Proposition 1.34. Let 𝐴 be a subring of a ring 𝐵. An element 𝛼 of 𝐵 is integral over 𝐴 if

and only if there exists a faithful 𝐴[𝛼]-submodule 𝑀 of 𝐵 that is finitely generated as an
𝐴-module.

Proof. ⇒∶ Suppose that

𝛼𝑛 + 𝑎1 𝛼𝑛−1 + ⋯ + 𝑎𝑛 = 0, 𝑎𝑖 ∈ 𝐴.

Then the 𝐴-submodule 𝑀 of 𝐵 generated by 1, 𝛼, ..., 𝛼 𝑛−1 has the property that 𝛼𝑀 ⊂ 𝑀,
and it is faithful because it contains 1.
⇐∶ Let 𝑀 be a faithful 𝐴[𝛼]-submodule of 𝐵 admitting a finite set {𝑒1 , … , 𝑒𝑛 } of
generators as an 𝐴-module. Then, for each 𝑖,
∑
𝛼𝑒𝑖 = 𝑎𝑖𝑗 𝑒𝑗 , some 𝑎𝑖𝑗 ∈ 𝐴.

(𝛼 − 𝑎11 )𝑒1 − 𝑎12 𝑒2 − 𝑎13 𝑒3 − ⋯ = 0

We can rewrite this system of equations as

−𝑎21 𝑒1 + (𝛼 − 𝑎22 )𝑒2 − 𝑎23 𝑒3 − ⋯ = 0

⋯ = 0.

Let 𝐶 be the matrix of coefficients on the left-hand side. Then Cramer’s formula tells
us that det(𝐶) ⋅ 𝑒𝑖 = 0 for all 𝑖. As 𝑀 is faithful and the 𝑒𝑖 generate 𝑀, this implies that
det(𝐶) = 0. On expanding out the determinant, we obtain an equation
𝛼𝑛 + 𝑐1 𝛼𝑛−1 + 𝑐2 𝛼𝑛−2 + ⋯ + 𝑐𝑛 = 0, 𝑐𝑖 ∈ 𝐴. 2

Proposition 1.35. An 𝐴-algebra 𝐵 is finite if it is generated as an 𝐴-algebra by a finite

set of elements each of which is integral over 𝐴.

Proof. We may replace 𝐴 with its image in 𝐵. Suppose that 𝐵 = 𝐴[𝛼1 , … , 𝛼𝑚 ] and that

𝛼𝑖 𝑖 + 𝑎𝑖1 𝛼𝑖 𝑖 + ⋯ + 𝑎𝑖𝑛𝑖 = 0, 𝑎𝑖𝑗 ∈ 𝐴, 𝑖 = 1, … , 𝑚.

𝑛 𝑛 −1

Any monomial in the 𝛼𝑖 divisible by some 𝛼𝑖 𝑖 is equal (in 𝐵) to a linear combination of

𝑛

monomials of lower degree. Therefore, 𝐵 is generated as an 𝐴-module by the finite set

of monomials 𝛼11 ⋯ 𝛼𝑚𝑚 , 1 ≤ 𝑟𝑖 < 𝑛𝑖 .
𝑟 𝑟

Corollary 1.36. An 𝐴-algebra 𝐵 is finite if and only if it is finitely generated and integral
2

over 𝐴.

Proof. ⇐: Immediate consequence of 1.35.

⇒: We may replace 𝐴 with its image in 𝐵. Then 𝐵 is a faithful 𝐴[𝛼]-module for all
𝛼 ∈ 𝐵 (because 1𝐵 ∈ 𝐵), and so 1.34 shows that every element of 𝐵 is integral over 𝐴. As
𝐵 is finitely generated as an 𝐴-module, it is certainly finitely generated as an 𝐴-algebra.2
 d. Integral dependence 27

Proposition 1.37. Consider rings 𝐴 ⊂ 𝐵 ⊂ 𝐶. If 𝐵 is integral over 𝐴 and 𝐶 is integral

over 𝐵, then 𝐶 is integral over 𝐴.

Proof. Let 𝛾 ∈ 𝐶. Then

𝛾𝑛 + 𝑏1 𝛾𝑛−1 + ⋯ + 𝑏𝑛 = 0
for some 𝑏𝑖 ∈ 𝐵. Now 𝐴[𝑏1 , … , 𝑏𝑛 ] is finite over 𝐴 (see 1.35), and 𝐴[𝑏1 , … , 𝑏𝑛 ][𝛾] is finite
over 𝐴[𝑏1 , … , 𝑏𝑛 ], and so it is finite over 𝐴. Therefore 𝛾 is integral over 𝐴 by 1.34. 2

Theorem 1.38. Let 𝐴 be a subring of a ring 𝐵. The elements of 𝐵 integral over 𝐴 form an
𝐴-subalgebra of 𝐵.

Proof. Let 𝛼 and 𝛽 be two elements of 𝐵 integral over 𝐴. Then 𝐴[𝛼, 𝛽] is finitely
generated as an 𝐴-module (1.35). It is stable under multiplication by 𝛼 ± 𝛽 and 𝛼𝛽 and
it is faithful as an 𝐴[𝛼 ± 𝛽]-module and as an 𝐴[𝛼𝛽]-module (because it contains 1𝐴 ).
Therefore 1.34 shows that 𝛼 ± 𝛽 and 𝛼𝛽 are integral over 𝐴. 2

Definition 1.39. The 𝐴-subalgebra of 𝐵 of elements integral over 𝐴 is called the inte-
gral closure of 𝐴 in 𝐵.

Proposition 1.40. Let 𝐴 be an integral domain with field of fractions 𝐹, and let 𝛼 be an
element of some field containing 𝐹. If 𝛼 is algebraic over 𝐹, then there exists a 𝑑 ∈ 𝐴 such
that 𝑑𝛼 is integral over 𝐴.

Proof. By assumption, 𝛼 satisfies an equation

𝛼𝑚 + 𝑎1 𝛼𝑚−1 + ⋯ + 𝑎𝑚 = 0, 𝑎𝑖 ∈ 𝐹.

Let 𝑑 be a common denominator for the 𝑎𝑖 , so that 𝑑𝑎𝑖 ∈ 𝐴 for all 𝑖, and multiply the
equation by 𝑑𝑚 :
(𝑑𝛼)𝑚 + 𝑎1 𝑑(𝑑𝛼)𝑚−1 + ⋯ + 𝑎𝑚 𝑑𝑚 = 0.
As 𝑎1 𝑑, … , 𝑎𝑚 𝑑𝑚 ∈ 𝐴, this shows that 𝑑𝛼 is integral over 𝐴. 2

Corollary 1.41. Let 𝐴 be an integral domain and let 𝐸 be an algebraic extension of the
field of fractions of 𝐴. Then 𝐸 is the field of fractions of the integral closure of 𝐴 in 𝐸.

Proof. In fact, the proposition shows that every element of 𝐸 is a quotient 𝛽∕𝑑 with 𝛽
integral over 𝐴 and 𝑑 ∈ 𝐴. 2

Definition 1.42. An integral domain 𝐴 is said to be integrally closed if it is equal to

its integral closure in its field of fractions 𝐹, i.e., if

𝛼 ∈ 𝐹, 𝛼 integral over 𝐴 ⇐⇒ 𝛼 ∈ 𝐴.

An integrally closed integral domain is called an integrally closed domain or normal

domain.

Proposition 1.43. Unique factorization domains are integrally closed.

28 1. Preliminaries from commutative algebra

Proof. Let 𝐴 be a unique factorization domain, and let 𝑎∕𝑏 be an element of its field of
fractions. If 𝑎∕𝑏 ∉ 𝐴, then we may suppose that 𝑏 is divisible by some prime element 𝑝
not dividing 𝑎. If 𝑎∕𝑏 is integral over 𝐴, then it satisfies an equation

(𝑎∕𝑏)𝑛 + 𝑎1 (𝑎∕𝑏)𝑛−1 + ⋯ + 𝑎𝑛 = 0, 𝑎𝑖 ∈ 𝐴.

On multiplying through by 𝑏𝑛 , we obtain the equation

𝑎𝑛 + 𝑎1 𝑎𝑛−1 𝑏 + ⋯ + 𝑎𝑛 𝑏𝑛 = 0.

The element 𝑝 then divides every term on the left except 𝑎𝑛 , and hence divides 𝑎𝑛 . Since
it does not divide 𝑎, this is a contradiction. 2

Let 𝐹 ⊂ 𝐸 be fields, and let 𝛼 ∈ 𝐸 be algebraic over 𝐹. The minimal polynomial of

𝛼 over 𝐹 is the monic polynomial of smallest degree in 𝐹[𝑋] having 𝛼 as a root. If 𝑓 is
the minimal polynomial of 𝛼, then the homomorphism 𝑋 ↦ 𝛼 ∶ 𝐹[𝑋] → 𝐹[𝛼] defines
an isomorphism 𝐹[𝑋]∕(𝑓) → 𝐹[𝛼], i.e., 𝐹[𝑥] ≃ 𝐹[𝛼], 𝑥 ↔ 𝛼.

Proposition 1.44. Let 𝐴 be an integrally closed domain, and let 𝐸 be a finite extension of
the field of fractions 𝐹 of 𝐴. An element of 𝐸 is integral over 𝐴 if and only if its minimal
polynomial over 𝐹 has coefficients in 𝐴.

Proof. Let 𝛼 ∈ 𝐸 be integral over 𝐴, so that

𝛼𝑚 + 𝑎1 𝛼𝑚−1 + ⋯ + 𝑎𝑚 = 0, some 𝑎𝑖 ∈ 𝐴, 𝑚 > 0.

Let 𝑓(𝑋) be the minimal polynomial of 𝛼 over 𝐹, and let 𝛼′ be a conjugate of 𝛼, i.e., a
root of 𝑓 in some splitting field of 𝑓. Then 𝑓 is also the minimal polynomial of 𝛼′ over
𝐹, and so there is an 𝐹-isomorphism

𝜎 ∶ 𝐹[𝛼] → 𝐹[𝛼′ ], 𝜎(𝛼) = 𝛼′ .

On applying 𝜎 to the above equation we obtain the equation

𝛼′𝑚 + 𝑎1 𝛼′𝑚−1 + ⋯ + 𝑎𝑚 = 0,

which shows that 𝛼′ is integral over 𝐴. As the coefficients of 𝑓 are polynomials in the
conjugates of 𝛼, it follows from Theorem 1.38 that the coefficients of 𝑓(𝑋) are integral
over 𝐴. They lie in 𝐹, and 𝐴 is integrally closed, and so they lie in 𝐴. This proves the
“only if” part of the statement, and the “if” part is obvious. 2

Corollary 1.45. Let 𝐴 ⊂ 𝐹 ⊂ 𝐸 be as in the proposition, and let 𝛼 be an element of 𝐸

integral over 𝐴. Then Nm𝐸∕𝐹 (𝛼) ∈ 𝐴, and 𝛼 divides Nm𝐸∕𝐹 (𝛼) in 𝐴[𝛼].

𝑓(𝑋) = 𝑋 𝑚 + 𝑎1 𝑋 𝑚−1 + ⋯ + 𝑎𝑚
Proof. Let

be the minimal polynomial of 𝛼 over 𝐹. Then Nm(𝛼) = (−1)𝑚𝑛 𝑎𝑚

𝑛
, where 𝑛 = [𝐸 ∶ 𝐹[𝛼]]
(FT, 5.45), and so Nm(𝛼) ∈ 𝐴. Because 𝑓(𝛼) = 0,

0 = 𝑎𝑚 (𝛼 + 𝑎1 𝛼𝑚−1 + ⋯ + 𝑎𝑚 )
𝑛−1 𝑚

= 𝛼(𝑎𝑚 𝛼
𝑛−1 𝑚−1
+ ⋯ + 𝑎𝑚
𝑛−1
𝑎𝑚−1 ) + (−1)𝑚𝑛 Nm(𝛼),

and so 𝛼 divides Nm𝐸∕𝐹 (𝛼) in 𝐴[𝛼]. 2

 d. Integral dependence 29

Corollary 1.46. Let 𝐴 be an integrally closed domain with field of fractions 𝐹, and
let 𝑓(𝑋) be a monic polynomial in 𝐴[𝑋]. Then every monic factor of 𝑓(𝑋) in 𝐹[𝑋] has
coefficients in 𝐴.

Proof. It suffices to prove this for an irreducible monic factor 𝑔 of 𝑓 in 𝐹[𝑋]. Let 𝛼 be
a root of 𝑔 in some extension field of 𝐹. Then 𝑔 is the minimal polynomial of 𝛼. As 𝛼 is
a root of 𝑓, it is integral over 𝐴, and so 𝑔 has coefficients in 𝐴.

Proposition 1.47. Let 𝐴 ⊂ 𝐵 be rings, and let 𝐴′ be the integral closure of 𝐴 in 𝐵. For
2

any multiplicative subset 𝑆 of 𝐴, 𝑆 −1 𝐴′ is the integral closure of 𝑆 −1 𝐴 in 𝑆 −1 𝐵.

Proof. Let 𝑏𝑠 ∈ 𝑆 −1 𝐴′ with 𝑏 ∈ 𝐴′ and 𝑠 ∈ 𝑆. Then

𝑏𝑛 + 𝑎1 𝑏𝑛−1 + ⋯ + 𝑎𝑛 = 0

for some 𝑎𝑖 ∈ 𝐴, and so

𝑏
𝑛
𝑎 𝑏 𝑛−1 𝑎𝑛
( 𝑠 ) + 𝑠1 ( 𝑠 ) + ⋯ + 𝑛 = 0.
𝑠

Therefore 𝑏∕𝑠 is integral over 𝑆 −1 𝐴. This shows that 𝑆 −1 𝐴′ is contained in the integral
closure of 𝑆 −1 𝐵.
For the converse, let 𝑏∕𝑠 (𝑏 ∈ 𝐵, 𝑠 ∈ 𝑆) be integral over 𝑆 −1 𝐴. Then

𝑏
𝑛
𝑎 𝑏 𝑛−1 𝑎𝑛
( 𝑠 ) + 𝑠1 ( 𝑠 ) + ⋯ + 𝑠 = 0.
1 𝑛

for some 𝑎𝑖 ∈ 𝐴 and 𝑠𝑖 ∈ 𝑆. On multiplying this equation by 𝑠𝑛 𝑠1 ⋯ 𝑠𝑛 , we find that

𝑠 ⋯𝑠 𝑏
𝑠1 ⋯ 𝑠𝑛 𝑏 ∈ 𝐴′ , and therefore that 𝑏𝑠 = 𝑠𝑠1 ⋯ 𝑛𝑠 ∈ 𝑆 −1 𝐴′ .
1 𝑛
2

Corollary 1.48. Let 𝐴 ⊂ 𝐵 be rings, and let 𝑆 be a multiplicative subset of 𝐴. If 𝐴 is

integrally closed in 𝐵, then 𝑆 −1 𝐴 is integrally closed in 𝑆 −1 𝐵.

Proof. Special case of the proposition in which 𝐴′ = 𝐴.

Proposition 1.49. The following conditions on an integral domain 𝐴 are equivalent:

2

(a) 𝐴 is integrally closed;

(b) 𝐴𝔭 is integrally closed for all prime ideals 𝔭;
(c) 𝐴𝔪 is integrally closed for all maximal ideals 𝔪.

to prove (c)⇒(a). If 𝑐 is integral over 𝐴, then it is integral over each 𝐴𝔪 , and hence lies
Proof. The implication (a)⇒(b) follows from 1.48, and (b)⇒(c) is obvious. It remains

in each 𝐴𝔪 . It follows that the ideal consisting of the 𝑎 ∈ 𝐴 such that 𝑎𝑐 ∈ 𝐴 is not
contained in any maximal ideal 𝔪, and therefore equals 𝐴. Hence 1 ⋅ 𝑐 ∈ 𝐴.

Let 𝐸∕𝐹 be a finite extension of fields. Then

2

(𝛼, 𝛽) ↦ Tr𝐸∕𝐹 (𝛼𝛽) ∶ 𝐸 × 𝐸 → 𝐹 (11)

is a symmetric bilinear form on 𝐸 regarded as a vector space over 𝐹.

Lemma 1.50. If 𝐸∕𝐹 is separable, then the trace pairing (11) is nondegenerate.
30 1. Preliminaries from commutative algebra

Proof. Let 𝛽1 , ..., 𝛽𝑚 be a basis for 𝐸 as an 𝐹-vector space. We have to show that the
discriminant det(Tr(𝛽𝑖 𝛽𝑗 )) of the trace pairing is nonzero. Let 𝜎1 , ..., 𝜎𝑚 be the distinct
𝐹-homomorphisms of 𝐸 into some large Galois extension Ω of 𝐹. Recall (FT, 5.45) that

Tr𝐿∕𝐾 (𝛽) = 𝜎1 𝛽 + ⋅ ⋅ ⋅ + 𝜎𝑚 𝛽 (12)

∑
By direct calculation, we have

det(Tr(𝛽𝑖 𝛽𝑗 )) = det( 𝑘 𝜎𝑘 (𝛽𝑖 𝛽𝑗 ))

∑
= det( 𝑘 𝜎𝑘 (𝛽𝑖 ) ⋅ 𝜎𝑘 (𝛽𝑗 ))
(by 12)

= det(𝜎𝑘 (𝛽𝑖 )) ⋅ det(𝜎𝑘 (𝛽𝑗 ))

= det(𝜎𝑘 (𝛽𝑖 ))2 .

Suppose that det(𝜎𝑖 𝛽𝑗 ) = 0. Then there exist 𝑐1 , ..., 𝑐𝑚 ∈ Ω such that

∑
𝑐𝑖 𝜎𝑖 (𝛽𝑗 ) = 0 all 𝑗.
𝑖
∑
By linearity, it follows that 𝑖 𝑐𝑖 𝜎𝑖 (𝛽) = 0 for all 𝛽 ∈ 𝐸, but this contradicts Dedekind’s
theorem on the independence of characters (FT, 5.14). 2

Proposition 1.51. Let 𝐴 be an integrally closed domain with field of fractions 𝐹, and let
𝐵 be the integral closure of 𝐴 in a separable extension 𝐸 of 𝐹 of degree 𝑚. There exist free
𝐴-submodules 𝑀 and 𝑀 ′ of 𝐸 such that

𝑀 ⊂ 𝐵 ⊂ 𝑀′. (13)

If 𝐴 is noetherian, then 𝐵 is a finite 𝐴-algebra.

Proof. Let {𝛽1 , ..., 𝛽𝑚 } be a basis for 𝐸 over 𝐹. According to Proposition 1.40, there
exists a 𝑑 ∈ 𝐴 such that 𝑑 ⋅ 𝛽𝑖 ∈ 𝐵 for all 𝑖. Clearly {𝑑 ⋅ 𝛽1 , … , 𝑑 ⋅ 𝛽𝑚 } is still a basis for 𝐸
as a vector space over 𝐹, and so we may assume to begin with that each 𝛽𝑖 ∈ 𝐵. Because
the trace pairing is nondegenerate, there is a dual basis {𝛽1′ , ..., 𝛽𝑚 ′
} of 𝐸 over 𝐹 with the
property that Tr(𝛽𝑖 ⋅ 𝛽𝑗 ) = 𝛿𝑖𝑗 for all 𝑖, 𝑗. We shall show that
′

𝐴𝛽1 + 𝐴𝛽2 + ⋯ + 𝐴𝛽𝑚 ⊂ 𝐵 ⊂ 𝐴𝛽1′ + 𝐴𝛽2′ + ⋯ + 𝐴𝛽𝑚

′
.

Only the second inclusion requires proof. Let 𝛽 ∈ 𝐵. Then 𝛽 can be written uniquely as
∑
a linear combination 𝛽 = 𝑏𝑗 𝛽𝑗′ of the 𝛽𝑗′ with coefficients 𝑏𝑗 ∈ 𝐹, and we have to show
that each 𝑏𝑗 ∈ 𝐴. As 𝛽𝑖 and 𝛽 are in 𝐵, so also is 𝛽 ⋅ 𝛽𝑖 , and so Tr(𝛽 ⋅ 𝛽𝑖 ) ∈ 𝐴 (1.44). But
∑ ∑ ∑
Tr(𝛽 ⋅ 𝛽𝑖 ) = Tr( 𝑏𝑗 𝛽𝑗′ ⋅ 𝛽𝑖 ) = 𝑏𝑗 Tr(𝛽𝑗′ ⋅ 𝛽𝑖 ) = 𝑏𝑗 ⋅ 𝛿𝑖𝑗 = 𝑏𝑖 .
𝑗 𝑗 𝑗

Hence 𝑏𝑖 ∈ 𝐴.
If 𝐴 is Noetherian, then 𝑀 ′ is a Noetherian 𝐴-module, and so 𝐵 is finitely generated
as an 𝐴-module. 2

Lemma 1.52. Let 𝐴 be a subring of a field 𝐾. If 𝐾 is integral over 𝐴, then 𝐴 is also a field.
 d. Integral dependence 31

Proof. Let 𝑎 be a nonzero element of 𝐴. Then 𝑎−1 ∈ 𝐾, and it is integral over 𝐴:

(𝑎−1 )𝑛 + 𝑎1 (𝑎−1 )𝑛−1 + ⋯ + 𝑎𝑛 = 0, 𝑎𝑖 ∈ 𝐴.

On multiplying through by 𝑎𝑛−1 , we find that

𝑎−1 + 𝑎1 + ⋯ + 𝑎𝑛 𝑎𝑛−1 = 0,

from which it follows that 𝑎−1 ∈ 𝐴. 2

Theorem 1.53 (Going-Up Theorem). Let 𝐴 ⊂ 𝐵 be rings with 𝐵 integral over 𝐴.

(a) For every prime ideal 𝔭 of 𝐴, there is a prime ideal 𝔮 of 𝐵 such that 𝔮 ∩ 𝐴 = 𝔭.
(b) Let 𝔭 = 𝔮 ∩ 𝐴; then 𝔭 is maximal if and only if 𝔮 is maximal.

Proof. (a) If 𝑆 is a multiplicative subset of a ring 𝐴, then the prime ideals of 𝑆 −1 𝐴 are
in one-to-one correspondence with the prime ideals of 𝐴 not intersecting 𝑆 (see 1.14).
It therefore suffices to prove (a) after 𝐴 and 𝐵 have been replaced by 𝑆 −1 𝐴 and 𝑆 −1 𝐵,
where 𝑆 = 𝐴 − 𝔭. Thus we may assume that 𝐴 is local, and that 𝔭 is its unique maximal
ideal. In this case, for all proper ideals 𝔟 of 𝐵, 𝔟 ∩ 𝐴 ⊂ 𝔭 (otherwise 𝔟 ⊃ 𝐴 ∋ 1). To
complete the proof of (a), we shall show that for all maximal ideals 𝔫 of 𝐵, 𝔫 ∩ 𝐴 = 𝔭.
Consider 𝐵∕𝔫 ⊃ 𝐴∕(𝔫 ∩ 𝐴). Here 𝐵∕𝔫 is a field, which is integral over its subring
𝐴∕(𝔫 ∩ 𝐴), and 𝔫 ∩ 𝐴 will be equal to 𝔭 if and only if 𝐴∕(𝔫 ∩ 𝐴) is a field. This follows

(b) The ring 𝐵∕𝔮 contains 𝐴∕𝔭, and it is integral over 𝐴∕𝔭. If 𝔮 is maximal, then
from Lemma 1.52.

Lemma 1.52 shows that 𝔭 is also. For the converse, note that any integral domain
integral over a field is a field because it is a union of integral domains finite over the
field, which are automatically fields (left multiplication by an element is injective, and
hence surjective, being a linear map of a finite-dimensional vector space). 2

Corollary 1.54. Let 𝐴 ⊂ 𝐵 be rings with 𝐵 integral over 𝐴. Let 𝔭 ⊂ 𝔭′ be prime ideals of
𝐴, and let 𝔮 be a prime ideal of 𝐵 such that 𝔮 ∩ 𝐴 = 𝔭. Then there exists a prime ideal 𝔮′ of
𝐵 containing 𝔮 and such that 𝔮′ ∩ 𝐴 = 𝔭′ ,

𝐵 𝔮 ⊂ 𝔮′

𝐴 𝔭 ⊂ 𝔭′ .

Proof. We have 𝐴∕𝔭 ⊂ 𝐵∕𝔮, and 𝐵∕𝔮 is integral over 𝐴∕𝔭. According to Theorem 1.53,
there exists a prime ideal 𝔮′′ in 𝐵∕𝔮 such that 𝔮′′ ∩ (𝐴∕𝔭) = 𝔭′ ∕𝔭. The inverse image 𝔮′
of 𝔮′′ in 𝐵 has the required properties. 2

Aside 1.55. Let 𝐴 be a noetherian integral domain, and let 𝐵 be the integral closure of 𝐴 in
a finite extension 𝐸 of the field of fractions 𝐹 of 𝐴. Is 𝐵 always a finite 𝐴-algebra? When 𝐴 is
integrally closed and 𝐸 is separable over 𝐹, or 𝐴 is a finitely generated 𝑘-algebra, then the answer

𝐴 whose integral closure in its field of fractions is not a finite 𝐴-algebra. F.K. Schmidt found
is yes (1.51, 8.3). However, in 1935, Akizuki found an example of a noetherian integral domain

another example at about the same time.7

7
According to Matsumura (1986, p. x), finding his example cost Akizuki a year’s hard struggle. For a
discussion of the examples of Akizuki and Schmidt, and generalizations, see Olberding, B., One-dimensional
bad Noetherian domains. Trans. Amer. Math. Soc. 366 (2014), no.8, 4067–4095.
32 1. Preliminaries from commutative algebra

e. Tensor Products
Tensor products of modules
Let 𝐴 be a ring, and let 𝑀, 𝑁, and 𝑃 be 𝐴-modules. A map 𝜙 ∶ 𝑀 × 𝑁 → 𝑃 of 𝐴-modules
is said to be 𝐴-bilinear if
𝜙(𝑥 + 𝑥 ′ , 𝑦) = 𝜙(𝑥, 𝑦) + 𝜙(𝑥′ , 𝑦), 𝑥, 𝑥 ′ ∈ 𝑀, 𝑦 ∈ 𝑁
𝜙(𝑥, 𝑦 + 𝑦 ) = 𝜙(𝑥, 𝑦) + 𝜙(𝑥, 𝑦 ),
′ ′ 𝑥 ∈ 𝑀, 𝑦, 𝑦 ′ ∈ 𝑁
𝜙(𝑎𝑥, 𝑦) = 𝑎𝜙(𝑥, 𝑦) = 𝜙(𝑥, 𝑎𝑦), 𝑎 ∈ 𝐴, 𝑥 ∈ 𝑀, 𝑦 ∈ 𝑁,
i.e., if 𝜙 is 𝐴-linear in each variable.
An 𝐴-module 𝑇 together with an 𝐴-bilinear map

𝜙∶ 𝑀 × 𝑁 → 𝑇
𝑀×𝑁 → 𝑇
𝜙

is called the tensor product of 𝑀 and 𝑁 over 𝐴 if it has

←

the following universal property: every 𝐴-bilinear map →

∃! linear

←
←
𝜙′

𝜙′ ∶ 𝑀 × 𝑁 → 𝑇 ′ 𝑇′ .

→
factors uniquely through 𝜙.

unique isomorphism. We denote it by 𝑀 ⊗𝐴 𝑁. It is determined by

As usual, the universal property determines the tensor product uniquely up to a

Hom𝐴-linear (𝑀 ⊗𝐴 𝑁, 𝑇) ≃ Hom𝐴-bilinear (𝑀 × 𝑁, 𝑇).

Let 𝑀 and 𝑁 be 𝐴-modules, and let 𝐴(𝑀×𝑁) be the free 𝐴-module with basis 𝑀 × 𝑁.
Construction

Thus each element 𝐴(𝑀×𝑁) can be expressed uniquely as a finite sum

∑
𝑎𝑖 (𝑥𝑖 , 𝑦𝑖 ), 𝑎𝑖 ∈ 𝐴, 𝑥𝑖 ∈ 𝑀, 𝑦𝑖 ∈ 𝑁.

Let 𝑃 be the submodule of 𝐴(𝑀×𝑁) generated by the following elements

(𝑥 + 𝑥′ , 𝑦) − (𝑥, 𝑦) − (𝑥 ′ , 𝑦), 𝑥, 𝑥′ ∈ 𝑀, 𝑦∈𝑁
(𝑥, 𝑦 + 𝑦′ ) − (𝑥, 𝑦) − (𝑥, 𝑦 ′ ), 𝑥 ∈ 𝑀, 𝑦, 𝑦 ′ ∈ 𝑁
(𝑎𝑥, 𝑦) − 𝑎(𝑥, 𝑦), 𝑎 ∈ 𝐴, 𝑥 ∈ 𝑀, 𝑦∈𝑁
(𝑥, 𝑎𝑦) − 𝑎(𝑥, 𝑦), 𝑎 ∈ 𝐴, 𝑥 ∈ 𝑀, 𝑦 ∈ 𝑁,

𝑀 ⊗𝐴 𝑁 = 𝐴(𝑀×𝑁) ∕𝑃.
and define

Write 𝑥 ⊗ 𝑦 for the class of (𝑥, 𝑦) in 𝑀 ⊗𝐴 𝑁. Then

(𝑥, 𝑦) ↦ 𝑥 ⊗ 𝑦 ∶ 𝑀 × 𝑁 → 𝑀 ⊗𝐴 𝑁
is 𝐴-bilinear — we have imposed the fewest relations necessary to ensure this. Every
element of 𝑀 ⊗𝐴 𝑁 can be written as a finite sum8
∑
𝑎𝑖 (𝑥𝑖 ⊗ 𝑦𝑖 ), 𝑎𝑖 ∈ 𝐴, 𝑥𝑖 ∈ 𝑀, 𝑦𝑖 ∈ 𝑁,
8
“An element of the tensor product of two vector spaces is not necessarily a tensor product of two vectors,
but sometimes a sum of such. This might be considered a mathematical shenanigan but if you start with
the state vectors of two quantum systems it exactly corresponds to the notorious notion of entanglement
which so displeased Einstein.” Georges Elencwajg on mathoverflow.net.
 e. Tensor Products 33

and all relations among these symbols are generated by the following relations

(𝑥 + 𝑥 ′ ) ⊗ 𝑦 = 𝑥 ⊗ 𝑦 + 𝑥 ′ ⊗ 𝑦
𝑥 ⊗ (𝑦 + 𝑦 ′ ) = 𝑥 ⊗ 𝑦 + 𝑥 ⊗ 𝑦 ′
𝑎𝑥 ⊗ 𝑦 = 𝑎(𝑥 ⊗ 𝑦) = 𝑥 ⊗ 𝑎𝑦.

The pair (𝑀 ⊗𝐴 𝑁, (𝑥, 𝑦) ↦ 𝑥 ⊗ 𝑦) has the correct universal property because any
𝐴-bilinear map 𝜙′ ∶ 𝑀 × 𝑁 → 𝑇 ′ extends uniquely to an 𝐴-linear map 𝐴(𝑀×𝑁) → 𝑇 ′ ,
which factors uniquely through 𝐴(𝑀×𝑁) ∕𝑃.

Tensor products of algebras

Let 𝐴 and 𝐵 be 𝑘-algebras. A 𝑘-algebra 𝐶 together with homomorphisms 𝑖 ∶ 𝐴 → 𝐶
and 𝑗 ∶ 𝐵 → 𝐶 is called the tensor product of 𝐴 and 𝐵 if it has the following universal
property: for every pair of homomorphisms (of 𝑘-algebras) 𝛼 ∶ 𝐴 → 𝑅 and 𝛽 ∶ 𝐵 → 𝑅,
there is a unique homomorphism 𝛾 ∶ 𝐶 → 𝑅 such that 𝛾◦𝑖 = 𝛼 and 𝛾◦𝑗 = 𝛽:

𝐴 → 𝐶 ← 𝐵
𝑖 𝑗

←
← →

𝛼 ∃! 𝛾 𝛽
→
←

𝑅.
→
→

by this property. We write it 𝐴 ⊗𝑘 𝐵. Note that

If it exists, the tensor product, is uniquely determined up to a unique isomorphism

Hom𝑘 (𝐴 ⊗𝑘 𝐵, 𝑅) ≃ Hom𝑘 (𝐴, 𝑅) × Hom𝑘 (𝐵, 𝑅)

(homomorphisms of 𝑘-algebras).

Form the tensor product 𝐴 ⊗𝑘 𝐵 of 𝐴 and 𝐵 regarded as 𝑘-vector spaces. There is a

Construction

multiplication map 𝐴 ⊗𝑘 𝐵 × 𝐴 ⊗𝑘 𝐵 → 𝐴 ⊗𝑘 𝐵 for which

(𝑎 ⊗ 𝑏)(𝑎′ ⊗ 𝑏′ ) = 𝑎𝑎′ ⊗ 𝑏𝑏′ .

This makes 𝐴 ⊗𝑘 𝐵 into a ring, and the homomorphism

𝑐 ↦ 𝑐(1 ⊗ 1) = 𝑐 ⊗ 1 = 1 ⊗ 𝑐

makes it into a 𝑘-algebra. The maps

𝑎 ↦ 𝑎 ⊗ 1 ∶ 𝐴 → 𝐶 and 𝑏 ↦ 1 ⊗ 𝑏 ∶ 𝐵 → 𝐶

are homomorphisms, and they make 𝐴 ⊗𝑘 𝐵 into the tensor product of 𝐴 and 𝐵 in the

Example 1.56. A 𝑘-algebra 𝐵, equipped with the given map 𝑘 → 𝐵 and the identity map
above sense.

𝐵 → 𝐵, has the universal property characterizing 𝑘 ⊗𝑘 𝐵, so 𝑘 ⊗𝑘 𝐵 ≃ 𝐵. In terms of the

constructive definition of tensor products, the isomorphism is 𝑐 ⊗ 𝑏 ↦ 𝑐𝑏 ∶ 𝑘 ⊗𝑘 𝐵 → 𝐵.
34 1. Preliminaries from commutative algebra

Example 1.57. The ring 𝑘[𝑋1 , … , 𝑋𝑚 , 𝑋𝑚+1 , … , 𝑋𝑚+𝑛 ], equipped with the obvious in-
clusions

𝑘[𝑋1 , … , 𝑋𝑚 ] → 𝑘[𝑋1 , … , 𝑋𝑚+𝑛 ] ← 𝑘[𝑋𝑚+1 , … , 𝑋𝑚+𝑛 ]

is the tensor product of 𝑘[𝑋1 , … , 𝑋𝑚 ] and 𝑘[𝑋𝑚+1 , … , 𝑋𝑚+𝑛 ]. To verify this we only have
to check that, for every 𝑘-algebra 𝑅, the map

Hom𝑘-alg (𝑘[𝑋1 , … , 𝑋𝑚+𝑛 ], 𝑅) → Hom𝑘-alg (𝑘[𝑋1 , …], 𝑅) × Hom𝑘-alg (𝑘[𝑋𝑚+1 , …], 𝑅)

induced by the inclusions is a bijection. But this map can be identified with the obvious

𝑅𝑚+𝑛 → 𝑅𝑚 × 𝑅𝑛 .
bijection

In terms of the constructive definition of tensor products, the isomorphism is

𝑓 ⊗ 𝑔 ↦ 𝑓𝑔 ∶ 𝑘[𝑋1 , … , 𝑋𝑚 ] ⊗𝑘 𝑘[𝑋𝑚+1 , … , 𝑋𝑚+𝑛 ] → 𝑘[𝑋1 , … , 𝑋𝑚+𝑛 ].

Remark 1.58. (a) If (𝑏𝛼 ) is a family of generators (resp. basis) for 𝐵 as a 𝑘-vector space,
then (1 ⊗ 𝑏𝛼 ) is a family of generators (resp. basis) for 𝐴 ⊗𝑘 𝐵 as an 𝐴-module.
(b) Let 𝑘 → Ω be fields. Then

Ω ⊗𝑘 𝑘[𝑋1 , … , 𝑋𝑛 ] ≃ Ω[1 ⊗ 𝑋1 , … , 1 ⊗ 𝑋𝑛 ] ≃ Ω[𝑋1 , … , 𝑋𝑛 ].

If 𝐴 = 𝑘[𝑋1 , … , 𝑋𝑛 ]∕(𝑔1 , … , 𝑔𝑚 ), then

Ω ⊗𝑘 𝐴 ≃ Ω[𝑋1 , … , 𝑋𝑛 ]∕(𝑔1 , … , 𝑔𝑚 ).

(c) If 𝐴 and 𝐵 are algebras of 𝑘-valued functions on sets 𝑆 and 𝑇 respectively, then
(𝑓 ⊗ 𝑔)(𝑥, 𝑦) = 𝑓(𝑥)𝑔(𝑦) realizes 𝐴 ⊗𝑘 𝐵 as an algebra of 𝑘-valued functions on 𝑆 × 𝑇.

f. Transcendence bases

1.59. Elements 𝛼1 , ..., 𝛼𝑛 of a 𝑘-algebra 𝐴 are said to be algebraically dependent over 𝑘

We review the theory of transcendence bases. For the proofs, see Chapter 9 of FT.

there exists a nonzero polynomial 𝑓(𝑋1 , ..., 𝑋𝑛 ) ∈ 𝑘[𝑋1 , ..., 𝑋𝑛 ] such that 𝑓(𝛼1 , ..., 𝛼𝑛 ) = 0.
Otherwise, the 𝛼𝑖 are said to be algebraically independent over 𝑘.

Now let Ω be a field containing 𝑘.

1.60. For a subset 𝐴 of Ω, we let 𝑘(𝐴) denote the smallest subfield of Ω containing 𝑘
𝑓(𝑥1 , … , 𝑥𝑚 }
and 𝐴. For example, if 𝐴 = {𝑥1 , … , 𝑥𝑚 }, then 𝑘(𝐴) consists of the quotients
𝑔(𝑥1 , … , 𝑥𝑚 }
with 𝑓, 𝑔 ∈ 𝑘[𝑋1 , … , 𝑋𝑚 ]. A subset 𝐵 of Ω is algebraically dependent on 𝐴 if each
element of 𝐵 is algebraic over 𝑘(𝐴).

1.61 (Fundamental Theorem). Let 𝐴 = {𝛼1 , ..., 𝛼𝑚 } and 𝐵 = {𝛽1 , ..., 𝛽𝑛 } be two sub-
sets of Ω. Assume that
(a) 𝐴 is algebraically independent (over 𝑘), and
(b) 𝐴 is algebraically dependent on 𝐵 (over 𝑘).
Then 𝑚 ≤ 𝑛.
 f. Transcendence bases 35

Note that this becomes the fundamental theorem of linear algebra when replace

1.62. A transcendence basis for Ω over 𝑘 is an algebraically independent set 𝐴 such

“algebraically” with “linearly” — the two topics are formally similar.

that Ω is algebraic over 𝑘(𝐴).

1.63. Assume that there is a finite subset 𝐴 ⊂ Ω such that Ω is algebraic over 𝑘(𝐴).

(a) every maximal algebraically independent subset of Ω is a transcendence basis;

Then

(b) every subset 𝑆 of 𝐴 minimal among those such that Ω is algebraic over 𝑘(𝑆) is a

(c) all transcendence bases for Ω over 𝑘 have the same finite number of elements
transcendence basis; in particular, a finite transcendence basis exists;

(called the transcendence degree, tr deg𝑘 Ω, of Ω over 𝑘).

1.64. Let 𝑘 ⊂ 𝐿 ⊂ Ω be fields. Then

tr deg𝑘 Ω = tr deg𝑘 𝐿 + tr deg𝐿 Ω.

Indeed, if 𝐴 is a transcendence basis for 𝐿∕𝑘 and 𝐵 is a transcendence basis for Ω∕𝐿,
then 𝐴 ∪ 𝐵 is a transcendence basis for Ω∕𝑘.

Exercises
1-1. Let 𝑘 be an infinite field (not necessarily algebraically closed). Show that an
𝑓 ∈ 𝑘[𝑋1 , … , 𝑋𝑛 ] that is identically zero on 𝑘𝑛 is the zero polynomial (i.e., has all its
coefficients zero).

1-2. Find a minimal set of generators for the ideal

(𝑋 + 2𝑌, 3𝑋 + 6𝑌 + 3𝑍, 2𝑋 + 4𝑌 + 3𝑍)

in 𝑘[𝑋, 𝑌, 𝑍]. What standard algorithm in linear algebra will allow you to answer this
question for any ideal generated by homogeneous linear polynomials? Find a minimal
set of generators for the ideal

(𝑋 + 2𝑌 + 1, 3𝑋 + 6𝑌 + 3𝑋 + 2, 2𝑋 + 4𝑌 + 3𝑍 + 3).

1-3. A ring 𝐴 is said to be normal if 𝐴𝔭 is an integrally closed domain for all prime
ideals 𝔭 in 𝐴. Show that a noetherian ring is normal if and only if it is a finite product of
normal integral domains.

1-4. Prove the statement in 1.64.

Chapter 2

Algebraic Sets

a. Definition of an algebraic set

An algebraic subset 𝑉(𝑆) of 𝑘 𝑛 is the set of common zeros of some collection 𝑆 of
polynomials in 𝑘[𝑋1 , … , 𝑋𝑛 ],

𝑉(𝑆) = {(𝑎1 , … , 𝑎𝑛 ) ∈ 𝑘 𝑛 ∣ 𝑓(𝑎1 , … , 𝑎𝑛 ) = 0 all 𝑓 ∈ 𝑆}.

We refer to 𝑉(𝑆) as the zero set of 𝑆. Note that

𝑆 ⊂ 𝑆 ′ ⇐⇒ 𝑉(𝑆) ⊃ 𝑉(𝑆 ′ );

Recall that the ideal 𝔞 generated by a set 𝑆 consists of the finite sums
— more equations means fewer solutions.

∑
𝑓𝑖 𝑔𝑖 , 𝑓𝑖 ∈ 𝑘[𝑋1 , … , 𝑋𝑛 ], 𝑔𝑖 ∈ 𝑆.
∑
Such a sum 𝑓𝑖 𝑔𝑖 is zero at every point at which the 𝑔𝑖 are all zero, and so 𝑉(𝑆) ⊂ 𝑉(𝔞),
but the reverse conclusion is also true because 𝑆 ⊂ 𝔞. Thus 𝑉(𝑆) = 𝑉(𝔞) — the zero
set of 𝑆 is the same as the zero set of the ideal generated by 𝑆. Therefore the algebraic
subsets of 𝑘 𝑛 can also be described as the zero sets of ideals in 𝑘[𝑋1 , … , 𝑋𝑛 ].
An empty set of polynomials imposes no conditions, and so 𝑉(∅) = 𝑘 𝑛 . Therefore
𝑘 is an algebraic subset. It is also the zero set of the zero ideal (0). We write 𝔸𝑛 for 𝑘𝑛
𝑛

regarded as an algebraic set.

Examples
2.1. If 𝑆 is a set of homogeneous linear equations,

𝑎𝑖1 𝑋1 + ⋯ + 𝑎𝑖𝑛 𝑋𝑛 = 0, 𝑖 = 1, … , 𝑚,

then 𝑉(𝑆) is a subspace of 𝑘 𝑛 . If 𝑆 is a set of nonhomogeneous linear equations,

𝑎𝑖1 𝑋1 + ⋯ + 𝑎𝑖𝑛 𝑋𝑛 = 𝑑𝑖 , 𝑖 = 1, … , 𝑚,

then 𝑉(𝑆) is either empty or is the translate of a subspace of 𝑘 𝑛 .

2.2. If 𝑆 consists of the single equation

𝑌 2 = 𝑋 3 + 𝑎𝑋 + 𝑏, 4𝑎3 + 27𝑏2 ≠ 0,

36
 b. The Hilbert basis theorem 37

then 𝑉(𝑆) is an elliptic curve. For example,

𝑌2 = 𝑋3 + 1 𝑌 2 = 𝑋(𝑋 2 − 1)

We generally visualize algebraic sets as though the field 𝑘 were ℝ, i.e., we draw the real
locus of the curve. However, this can be misleading — see the examples 4.11 and 4.17
below.

2.3. If 𝑆 consists of the single equation

𝑍2 = 𝑋2 + 𝑌2,

then 𝑉(𝑆) is a cone.

2.4. A nonzero constant polynomial has no zeros, and so the empty set is algebraic.

2.5. The proper algebraic subsets of 𝔸1 = 𝑘 are the finite subsets, because a polynomial
𝑓(𝑋) in one variable 𝑋 has only finitely many roots, and every finite set is the set of roots
of a polynomial.

2.6. Some generating sets for an ideal will be more useful than others for determining
what the algebraic set is. For example, the ideal

𝔞 = (𝑋 2 + 𝑌 2 + 𝑍 2 − 1, 𝑋 2 + 𝑌 2 − 𝑌, 𝑋 − 𝑍)

𝑋 − 𝑍, 𝑌 2 − 2𝑌 + 1, 𝑍 2 − 1 + 𝑌.
can be generated by1

The middle polynomial has (double) root 1, from which it follows that 𝑉(𝔞) consists of
the single point (0, 1, 0).

b. The Hilbert basis theorem

In our definition of an algebraic set, we did not require the set 𝑆 of polynomials to be
finite, but the Hilbert basis theorem shows that, in fact, every algebraic set is the zero

𝑘[𝑋1 , … , 𝑋𝑛 ] can be generated by a finite set of elements, and we have already observed
set of a finite set of polynomials. More precisely, the theorem states that every ideal in

that a set of generators of an ideal has the same zero set as the ideal.
1
This is, in fact, a Gröbner basis for the ideal.
38 2. Algebraic Sets

Theorem 2.7 (Hilbert Basis Theorem). The ring 𝑘[𝑋1 , … , 𝑋𝑛 ] is noetherian.

As we noted in the proof of 1.32,

𝑘[𝑋1 , … , 𝑋𝑛 ] = 𝑘[𝑋1 , … , 𝑋𝑛−1 ][𝑋𝑛 ].

Theorem 2.8. If 𝐴 is noetherian, then so also is 𝐴[𝑋].

Thus an induction argument shows that the theorem follows from the next statement.

Proof. We shall show that every ideal in 𝐴[𝑋] is finitely generated. Recall that for a

𝑓(𝑋) = 𝑎0 𝑋 𝑟 + 𝑎1 𝑋 𝑟−1 + ⋯ + 𝑎𝑟 , 𝑎𝑖 ∈ 𝐴, 𝑎0 ≠ 0,
polynomial

𝑎0 is called the leading coefficient of 𝑓.

Let 𝔞 be a proper ideal in 𝐴[𝑋], and let 𝔞(𝑖) denote the set of elements of 𝐴 that occur
as the leading coefficient of a polynomial in 𝔞 of degree 𝑖 (we also include 0). Clearly,
𝔞(𝑖) is an ideal in 𝐴, and 𝔞(𝑖) ⊂ 𝔞(𝑖 + 1) because, if 𝑐𝑋 𝑖 + ⋯ ∈ 𝔞, then 𝑋(𝑐𝑋 𝑖 + ⋯) ∈ 𝔞.
Let 𝔟 be an ideal of 𝐴[𝑋] contained in 𝔞. Then 𝔟(𝑖) ⊂ 𝔞(𝑖), and 𝔟 = 𝔞 if the two are
equal for all 𝑖. To see this, let 𝑓 be a polynomial in 𝔞. Because 𝔟(deg 𝑓) = 𝔞(deg 𝑓), there
exists a 𝑔 ∈ 𝔟 with the same leading coefficient as 𝑓, and so 𝑓 = 𝑔 + 𝑓1 with 𝑓1 ∈ 𝔞
and deg(𝑓1 ) < deg(𝑓). Similarly, 𝑓1 = 𝑔1 + 𝑓2 with 𝑔1 ∈ 𝔟 and deg(𝑓2 ) < deg(𝑓1 ).
Continuing in this fashion, we find that 𝑓 = 𝑔 + 𝑔1 + 𝑔2 + ⋯ ∈ 𝔟..
As 𝐴 is noetherian, the sequence

𝔞(1) ⊂ 𝔞(2) ⊂ ⋯ ⊂ 𝔞(𝑖) ⊂ ⋯

eventually becomes constant, say,

𝔞(𝑑) = 𝔞(𝑑 + 1) = ⋯

(and then 𝔞(𝑑) contains the leading coefficient of every polynomial in 𝔞). For each
𝑖 ≤ 𝑑, there exists a finite set of generators {𝑎𝑖1 , 𝑎𝑖2 , … , 𝑎𝑖𝑛𝑖 } for the ideal 𝔞(𝑖) (as 𝐴 is
noetherian), and we let 𝑓𝑖𝑗 denote a polynomial in 𝔞 with leading coefficient 𝑎𝑖𝑗 . The
ideal 𝔟 of 𝐴[𝑋] generated by the (finitely many) 𝑓𝑖𝑗 is contained in 𝔞 and has the property
that 𝔟(𝑖) = 𝔞(𝑖) for all 𝑖. Therefore 𝔟 = 𝔞, and 𝔞 is finitely generated. 2

Aside 2.9. One may ask how many elements are needed to generate a given ideal 𝔞 in 𝑘[𝑋1 , … , 𝑋𝑛 ],

set 𝑉. For 𝑛 = 1, the ring 𝑘[𝑋] is a principal ideal domain, and so every ideal is generated by a
or, what is not quite the same thing, how many equations are needed to define a given algebraic

single element. If 𝑉 is a linear subspace of 𝑘 𝑛 , then linear algebra shows that it is the zero set of
𝑛 − dim(𝑉) polynomials. All one can say in general, is that at least 𝑛 − dim(𝑉) polynomials are
needed to define 𝑉 (see 3.45), but often more are required. Determining exactly how many is an
area of active research — see 3.58.

c. The Zariski topology

Recall that, for ideals 𝔞 and 𝔟 in 𝑘[𝑋1 , … , 𝑋𝑛 ],

𝔞 ⊂ 𝔟 ⇐⇒ 𝑉(𝔞) ⊃ 𝑉(𝔟).

(a) 𝑉(0) = 𝑘𝑛 ; 𝑉(𝑘[𝑋1 , … , 𝑋𝑛 ]) = ∅;

Proposition 2.10. There are the following relations:
 d. The Hilbert Nullstellensatz 39

(b) 𝑉(𝔞𝔟) = 𝑉(𝔞 ∩ 𝔟) = 𝑉(𝔞) ∪ 𝑉(𝔟);

∑ ⋂
(c) 𝑉( 𝑖∈𝐼 𝔞𝑖 ) = 𝑖∈𝐼 𝑉(𝔞𝑖 ) for every family of ideals (𝔞𝑖 )𝑖∈𝐼 .

Proof. (a) Certainly, 𝑉(0) = 𝑘 𝑛 , and 𝑉(𝑘[𝑋1 , … , 𝑋𝑛 ]) is empty because 1 ∈ 𝑘[𝑋1 , … , 𝑋𝑛 ].

(b) Note that

𝔞𝔟 ⊂ 𝔞 ∩ 𝔟 ⊂ 𝔞, 𝔟 ⇐⇒ 𝑉(𝔞𝔟) ⊃ 𝑉(𝔞 ∩ 𝔟) ⊃ 𝑉(𝔞) ∪ 𝑉(𝔟).

For the reverse inclusions, observe that if 𝑎 ∉ 𝑉(𝔞) ∪ 𝑉(𝔟), then there exist 𝑓 ∈ 𝔞, 𝑔 ∈ 𝔟
such that 𝑓(𝑎) ≠ 0, 𝑔(𝑎) ≠ 0; but then (𝑓𝑔)(𝑎) ≠ 0, and so 𝑎 ∉ 𝑉(𝔞𝔟).
∑ ∑
(c) Recall that, by definition, 𝔞𝑖 consists of all finite sums of the form 𝑓𝑖 , 𝑓𝑖 ∈ 𝔞𝑖 .
Thus (c) is obvious. 2

The proposition shows that the algebraic subsets of 𝔸𝑛 satisfy the axioms to be the
closed subsets for a topology on 𝔸𝑛 : the empty set and the whole space are algebraic;

Thus, there is a topology on 𝔸𝑛 for which the closed subsets are exactly the algebraic
intersections of algebraic sets are algebraic; finite unions of algebraic sets are algebraic.

subsets — this is the Zariski topology on 𝔸𝑛 . The induced topology on a subset 𝑉 of

𝔸𝑛 is called the Zariski topology on 𝑉.

importance. For the Zariski topology on 𝔸1 , the closed subsets are the finite subsets
The Zariski topology has many strange properties, but it is nevertheless of great

and the whole space, and so the topology is not Hausdorff (in fact, there are no disjoint

of 𝔸2 are the unions of finitely many points and curves. Note that the Zariski topologies
nonempty open subsets at all). We shall see in 2.68 below that the proper closed subsets

on ℂ and ℂ2 are much coarser (have fewer open sets) than the complex topologies.

d. The Hilbert Nullstellensatz

Before examining the relation between the algebraic subsets of 𝔸𝑛 and the ideals of
𝑘[𝑋1 , … , 𝑋𝑛 ], we answer the question of when a collection 𝑆 of polynomials has a com-
mon zero, i.e., when the system of equations

𝑔(𝑋1 , … , 𝑋𝑛 ) = 0, 𝑔 ∈ 𝑆,

is “consistent”. Obviously, the system of equations

𝑔𝑖 (𝑋1 , … , 𝑋𝑛 ) = 0, 𝑖 = 1, … , 𝑚
∑
is inconsistent if there exist 𝑓𝑖 ∈ 𝑘[𝑋1 , … , 𝑋𝑛 ] such that 𝑓𝑖 𝑔𝑖 = 1, that is, if 1 is in the
ideal (𝑔1 , … , 𝑔𝑚 ) generated by the 𝑔𝑖 , which therefore equals 𝑘[𝑋1 , … , 𝑋𝑛 ]. The converse

Theorem 2.11 (Hilbert Nullstellensatz2 ). Every proper ideal 𝔞 in 𝑘[𝑋1 , … , 𝑋𝑛 ]

to this also holds.

has a zero in 𝑘 𝑛 .

A point 𝑃 = (𝑎1 , … , 𝑎𝑛 ) in 𝑘𝑛 defines a homomorphism “evaluate at 𝑃”

𝑓(𝑋1 , … , 𝑋𝑛 ) ↦ 𝑓(𝑎1 , … , 𝑎𝑛 ) ∶ 𝑘[𝑋1 , … , 𝑋𝑛 ] → 𝑘,

2
Nullstellensatz = zero-points-theorem.
40 2. Algebraic Sets

whose kernel contains 𝔞 if 𝑃 ∈ 𝑉(𝔞). Conversely, from a homomorphism 𝜑 ∶ 𝑘[𝑋1 , … , 𝑋𝑛 ] →

𝑘 of 𝑘-algebras whose kernel contains 𝔞, we obtain a point 𝑃 in 𝑉(𝔞), namely,

𝑃 = (𝜑(𝑋1 ), … , 𝜑(𝑋𝑛 )).

Thus, to prove the theorem, we have to show that there exists a 𝑘-algebra homomorphism
𝑘[𝑋1 , … , 𝑋𝑛 ]∕𝔞 → 𝑘.

prove this for a maximal ideal 𝔪. Then 𝐾 = 𝑘[𝑋1 , … , 𝑋𝑛 ]∕𝔪 is a field, and it is finitely
Since every proper ideal is contained in a maximal ideal (see p. 14), it suffices to

generated as a 𝑘-algebra. The next lemma shows that 𝐾 = 𝑘, which completes the proof.
def

Lemma 2.12 (Zariski’s Lemma). Let 𝑘 ⊂ 𝐾 be fields, not necessarily algebraically closed.
If 𝐾 is finitely generated as a 𝑘-algebra, then it is algebraic over 𝑘. (Hence 𝐾 = 𝑘 if 𝑘 is
algebraically closed.)

Proof. We begin by showing that 𝑘[𝑋] has infinitely many distinct monic irreducible
polynomials. When 𝑘 is infinite, the polynomials 𝑋−𝑎, 𝑎 ∈ 𝑘, are distinct and irreducible.
When 𝑘 is finite, we can adapt Euclid’s argument: if 𝑝1 , … , 𝑝𝑟 are monic irreducible
polynomials in 𝑘[𝑋], then 𝑝1 ⋯ 𝑝𝑟 + 1 is divisible by a monic irreducible polynomial
distinct from 𝑝1 , … , 𝑝𝑟 .
We prove the lemma by induction on 𝑟, the minimum number of elements required
to generate 𝐾 as a 𝑘-algebra. The case 𝑟 = 0 being trivial, we may suppose that

𝐾 = 𝑘[𝑥1 , … , 𝑥𝑟 ], 𝑟 ≥ 1.

If 𝐾 is not algebraic over 𝑘, then at least one 𝑥𝑖 , say, 𝑥1 , is not algebraic over 𝑘. Then,
𝑘[𝑥1 ] is a polynomial ring in one symbol over 𝑘, and its field of fractions 𝑘(𝑥1 ) is a
subfield of 𝐾. The induction hypothesis applied to 𝑘(𝑥1 ) ⊂ 𝐾 = 𝑘(𝑥1 )[𝑥2 , … , 𝑥𝑟 ] shows
that 𝐾 is algebraic over 𝑘(𝑥1 ). In particular, 𝑥2 , … , 𝑥𝑟 are algebraic over 𝑘(𝑥1 ), and so
(1.40) there exists a 𝑑 ∈ 𝑘[𝑥1 ] such that 𝑑𝑥2 , … , 𝑑𝑥𝑟 are integral over 𝑘[𝑥1 ].
Let 𝑓 ∈ 𝑘(𝑥1 ). Then 𝑓 ∈ 𝐾 = 𝑘[𝑥1 , … , 𝑥𝑟 ] and so, for a sufficiently large 𝑁, 𝑑𝑁 𝑓 ∈
𝑘[𝑥1 , 𝑑𝑥2 , … , 𝑑𝑥𝑟 ]. As the 𝑑𝑥𝑖 are integral over 𝑘[𝑥1 ], so also is 𝑑𝑁 𝑓 (by 1.38), and so it
lies in 𝑘[𝑥1 ] (by 1.43). In particular, for any monic irreducible polynomial 𝑓 ∈ 𝑘[𝑥1 ],
𝑑𝑁 ∕𝑓 ∈ 𝑘[𝑥1 ] for some 𝑁, but this contradicts the fact that there are infinitely many
distinct such 𝑓.

Let 𝑘 ⊂ 𝐾 be fields. The lemma shows that if 𝐾 is finitely generated as a 𝑘-algebra,

2

then it is finitely generated as a 𝑘-module (FT, 1.30).

e. The correspondence between algebraic sets and radical

ideals
The ideal attached to a subset of 𝑘 𝑛
For a subset 𝑊 of 𝑘 𝑛 , we write 𝐼(𝑊) for the set of polynomials that are zero on 𝑊:

𝐼(𝑊) = {𝑓 ∈ 𝑘[𝑋1 , … , 𝑋𝑛 ] ∣ 𝑓(𝑃) = 0 all 𝑃 ∈ 𝑊}.

𝑉 ⊂ 𝑊 ⇐⇒ 𝐼(𝑉) ⊃ 𝐼(𝑊).
Note that

Clearly, 𝐼(𝑊) is an ideal in 𝑘[𝑋1 , … , 𝑋𝑛 ]. There are the following relations:

 e. The correspondence between algebraic sets and radical ideals 41

(a) 𝐼(𝑘 𝑛 ) = {0}; 𝐼(∅) = 𝑘[𝑋1 , … , 𝑋𝑛 ];

⋃ ⋂
(b) 𝐼( 𝑊𝑖 ) = 𝐼(𝑊𝑖 ).
Only the statement 𝐼(𝑘𝑛 ) = 0 is (perhaps) not obvious. It says that every nonzero
polynomial in 𝑘[𝑋1 , … , 𝑋𝑛 ] is nonzero at some point of 𝑘𝑛 . This is true for any infinite
field 𝑘 (see Exercise 1-1).

Example 2.13. Let 𝑃 be the point (𝑎1 , … , 𝑎𝑛 ), and let

𝔪𝑃 = (𝑋1 − 𝑎1 , … , 𝑋𝑛 − 𝑎𝑛 ).

Clearly 𝐼(𝑃) ⊃ 𝔪𝑃 , but 𝔪𝑃 is a maximal ideal, because “evaluation at (𝑎1 , … , 𝑎𝑛 )” defines

𝑘[𝑋1 , … , 𝑋𝑛 ]∕(𝑋1 − 𝑎1 , … , 𝑋𝑛 − 𝑎𝑛 ) → 𝑘.
an isomorphism

As 𝐼(𝑃) is a proper ideal, it must equal 𝔪𝑃 .

Proposition 2.14. Let 𝑊 be a subset of 𝑘 𝑛 . Then 𝑉𝐼(𝑊) is the smallest algebraic subset
of 𝑘 𝑛 containing 𝑊. In particular, 𝑉𝐼(𝑊) = 𝑊 if 𝑊 is an algebraic set.

Proof. Certainly 𝑉𝐼(𝑊) is an algebraic set containing 𝑊. Let 𝑉 = 𝑉(𝔞) be another

algebraic set containing 𝑊. Then 𝔞 ⊂ 𝐼(𝑊), and so 𝑉(𝔞) ⊃ 𝑉𝐼(𝑊). 2

Radicals of ideals
The radical of an ideal 𝔞 in a ring 𝐴 is

rad(𝔞) = {𝑓 ∣ 𝑓 𝑟 ∈ 𝔞, some 𝑟 ∈ ℕ}.

def

Proposition 2.15. Let 𝔞 be an ideal in a ring 𝐴.

(a) The radical of 𝔞 is an ideal.
(b) rad(rad(𝔞)) = rad(𝔞).

Proof. (a) If 𝑎 ∈ rad(𝔞), then clearly 𝑓𝑎 ∈ rad(𝔞) for all 𝑓 ∈ 𝐴. Suppose that 𝑎, 𝑏 ∈
rad(𝔞), with, say, 𝑎𝑟 ∈ 𝔞 and 𝑏𝑠 ∈ 𝔞. When we expand (𝑎 + 𝑏)𝑟+𝑠 using the binomial
theorem, we find that every term has a factor 𝑎𝑟 or 𝑏𝑠 , and so lies in 𝔞.
(b) If 𝑎𝑟 ∈ rad(𝔞), then 𝑎𝑟𝑠 = (𝑎𝑟 )𝑠 ∈ 𝔞 for some 𝑠, and so 𝑎 ∈ rad(𝔞). 2

The radical of the ideal 0 is called the nilradical 𝔫 of 𝐴. Thus, 𝔫 consists of the
nilpotent elements of 𝐴. It is an ideal in 𝐴, and 𝐴∕𝔫 is is reduced, i.e., without nonzero

An ideal is said to be radical if it equals its radical. Thus 𝔞 is radical if and only if
nilpotent elements.

the ring 𝐴∕𝔞 is reduced. Since integral domains are reduced, prime ideals (a fortiori,
maximal ideals) are radical. Note that rad(𝔞) is radical (2.15b), and hence is the smallest
radical ideal containing 𝔞.
If 𝔞 and 𝔟 are radical, then 𝔞 ∩ 𝔟 is radical, but 𝔞 + 𝔟 need not be: consider, for
example, 𝔞 = (𝑋 2 − 𝑌) and 𝔟 = (𝑋 2 + 𝑌); they are both prime ideals in 𝑘[𝑋, 𝑌], but
𝑋 2 ∈ 𝔞 + 𝔟, 𝑋 ∉ 𝔞 + 𝔟. (See 2.22 below.)
42 2. Algebraic Sets

The strong Nullstellensatz

For a polynomial 𝑓 and point 𝑃 ∈ 𝑘 𝑛 , 𝑓 𝑟 (𝑃) = 𝑓(𝑃)𝑟 . Therefore 𝑓 𝑟 is zero at 𝑃 if and only
if 𝑓 is zero at 𝑃, and so, for any subset 𝑊 of 𝑘 𝑛 , the ideal 𝐼(𝑊) is radical. In particular,
𝐼𝑉(𝔞) ⊃ rad(𝔞). In fact, the two are equal.

Theorem 2.16 (Strong Nullstellensatz). For any ideal 𝔞 in 𝑘[𝑋1 , … , 𝑋𝑛 ],

𝐼𝑉(𝔞) = rad(𝔞);

in particular, 𝐼𝑉(𝔞) = 𝔞 if 𝔞 is a radical ideal.

Proof. We have already noted that 𝐼𝑉(𝔞) ⊃ rad(𝔞). For the reverse inclusion, we have
to show that if a polynomial ℎ vanishes on 𝑉(𝔞), then ℎ𝑁 ∈ 𝔞 for some 𝑁 > 0. We may
assume ℎ ≠ 0. Let 𝑔1 , … , 𝑔𝑚 generate 𝔞, and consider the system of 𝑚 + 1 equations in
𝑛 + 1 symbols,
𝑔𝑖 (𝑋1 , … , 𝑋𝑛 ) = 0, 𝑖 = 1, … , 𝑚,
{
1 − 𝑌ℎ(𝑋1 , … , 𝑋𝑛 ) = 0.
If (𝑎1 , … , 𝑎𝑛 , 𝑏) satisfies the first 𝑚 equations, then (𝑎1 , … , 𝑎𝑛 ) ∈ 𝑉(𝔞); consequently,
ℎ(𝑎1 , … , 𝑎𝑛 ) = 0, and (𝑎1 , … , 𝑎𝑛 , 𝑏) does not satisfy the last equation. The equations
are inconsistent, and so, according to the original Nullstellensatz, there exist 𝑓𝑖 ∈
𝑘[𝑋1 , … , 𝑋𝑛 , 𝑌] such that

∑
𝑚
1= 𝑓𝑖 ⋅ 𝑔𝑖 + 𝑓𝑚+1 ⋅ (1 − 𝑌ℎ)
𝑖=1

(in the ring 𝑘[𝑋1 , … , 𝑋𝑛 , 𝑌]). On applying the homomorphism

𝑋𝑖 ↦ 𝑋𝑖
{ ∶ 𝑘[𝑋1 , … , 𝑋𝑛 , 𝑌] → 𝑘(𝑋1 , … , 𝑋𝑛 )
𝑌 ↦ ℎ−1

∑
to the above equality, we obtain the identity
𝑚
1= 𝑓𝑖 (𝑋1 , … , 𝑋𝑛 , ℎ−1 ) ⋅ 𝑔𝑖 (𝑋1 , … , 𝑋𝑛 )
𝑖=1
(*)

in 𝑘(𝑋1 , … , 𝑋𝑛 ). Clearly
polynomial in 𝑋1 , … , 𝑋𝑛
𝑓𝑖 (𝑋1 , … , 𝑋𝑛 , ℎ−1 ) =
ℎ 𝑁𝑖
for some 𝑁𝑖 . Let 𝑁 be the largest of the 𝑁𝑖 . On multiplying (*) by ℎ𝑁 we obtain an

∑𝑚
ℎ𝑁 = (polynomial in 𝑋1 , … , 𝑋𝑛 ) ⋅ 𝑔𝑖 (𝑋1 , … , 𝑋𝑛 ),
equation

𝑖=1

which shows that ℎ𝑁 ∈ 𝔞.

Corollary 2.17. The map 𝔞 ↦ 𝑉(𝔞) defines a one-to-one correspondence between the
2

set of radical ideals in 𝑘[𝑋1 , … , 𝑋𝑛 ] and the set of algebraic subsets of 𝑘 𝑛 ; its inverse is 𝐼.

Proof. We know that 𝐼𝑉(𝔞) = 𝔞 if 𝔞 is a radical ideal (2.16), and that 𝑉𝐼(𝑊) = 𝑊 if 𝑊
is an algebraic set (2.14). Therefore, 𝐼 and 𝑉 are inverse bijections. 2
 e. The correspondence between algebraic sets and radical ideals 43

Corollary 2.18. The radical of an ideal in 𝑘[𝑋1 , … , 𝑋𝑛 ] is equal to the intersection of

the maximal ideals containing it.

Proof. Let 𝔞 be an ideal in 𝑘[𝑋1 , … , 𝑋𝑛 ]. Because maximal ideals are radical, every
maximal ideal containing 𝔞 also contains rad(𝔞), and so
⋂
rad(𝔞) ⊂ 𝔪.
𝔪⊃𝔞

For each 𝑃 = (𝑎1 , … , 𝑎𝑛 ) ∈ 𝑘𝑛 , the ideal 𝔪𝑃 = (𝑋1 − 𝑎1 , … , 𝑋𝑛 − 𝑎𝑛 ) is maximal in

𝑘[𝑋1 , … , 𝑋𝑛 ], and
𝑓 ∈ 𝔪𝑃 ⇐⇒ 𝑓(𝑃) = 0
(see 2.13). Thus 𝔪𝑃 ⊃ 𝔞 if 𝑃 ∈ 𝑉(𝔞). If 𝑓 ∈ 𝔪𝑃 for all 𝑃 ∈ 𝑉(𝔞), then 𝑓 is zero on 𝑉(𝔞),
and so 𝑓 ∈ 𝐼𝑉(𝔞) = rad(𝔞). We have shown that
⋂ ⋂
rad(𝔞) ⊃ 𝔪𝑃 ⊃ 𝔪.
𝑃∈𝑉(𝔞) 𝔪⊃𝔞 2

Remarks
2.19. Because 𝑉(0) = 𝑘 𝑛 ,
𝐼(𝑘 𝑛 ) = 𝐼𝑉(0) = rad(0) = 0.
in other words, only the zero polynomial is zero on the whole of 𝑘 𝑛 (which we knew
already from Exercise 1-1).

2.20. The one-to-one correspondence in Corollary 2.17 is order reversing. Therefore

But the maximal proper radical ideals are simply the maximal ideals in 𝑘[𝑋1 , … , 𝑋𝑛 ],
the maximal proper radical ideals correspond to the minimal nonempty algebraic sets.

𝐼((𝑎1 , … , 𝑎𝑛 )) = (𝑋1 − 𝑎1 , … , 𝑋𝑛 − 𝑎𝑛 )
and the minimal nonempty algebraic sets are the one-point sets. As

(see 2.13), we see that the maximal ideals of 𝑘[𝑋1 , … , 𝑋𝑛 ] are exactly the ideals (𝑋1 −
𝑎1 , … , 𝑋𝑛 − 𝑎𝑛 ) with (𝑎1 , … , 𝑎𝑛 ) ∈ 𝑘 𝑛 .

2.21. An algebraic set 𝑉(𝔞) is empty if and only if 𝔞 = 𝑘[𝑋1 , … , 𝑋𝑛 ] (Nullstellensatz,

2.22. Let 𝑊 and 𝑊 ′ be algebraic sets. As 𝑊∩𝑊 ′ is the largest algebraic subset contained
2.11).

in both 𝑊 and 𝑊 ′ , 𝐼(𝑊 ∩ 𝑊 ′ ) must be the smallest radical ideal containing both 𝐼(𝑊)
and 𝐼(𝑊 ′ ):
𝐼(𝑊 ∩ 𝑊 ′ ) = rad(𝐼(𝑊) + 𝐼(𝑊 ′ )).

For example, let 𝑊 = 𝑉(𝑋 2 − 𝑌) and 𝑊 ′ =

𝑉(𝑋 2 + 𝑌); then 𝑉(𝑋 2 − 𝑌)

𝐼(𝑊 ∩ 𝑊 ′ ) = rad(𝑋 2 , 𝑌) = (𝑋, 𝑌) ∙

(assuming char(𝑘) ≠ 2). Note that 𝑊 ∩ 𝑊′ =
{(0, 0)}, but when realized as the intersection of 𝑉(𝑋 2 + 𝑌)
𝑌 = 𝑋 2 and 𝑌 = −𝑋 2 , it has “multiplicity 2”.
44 2. Algebraic Sets

2.23. Let 𝒫 be the set of subsets of 𝑘 𝑛 and 𝒬 the set of subsets of 𝑘[𝑋1 , … , 𝑋𝑛 ]. Then
𝐼 ∶ 𝒫 → 𝒬 and 𝑉 ∶ 𝒬 → 𝒫 define a simple Galois correspondence between 𝒫 and 𝒬:
they are order reversing maps such that 𝑉𝐼(𝑊) ⊃ 𝑊 and 𝐼𝑉(𝔞) ⊃ 𝔞. It follows that 𝐼
and 𝑉 define a one-to-one correspondence between 𝐼 (𝒫) and 𝑉 (𝒬) (see FT, 7.19). But
the strong Nullstellensatz shows that 𝐼 (𝒫) consists exactly of the radical ideals, and (by
definition) 𝑉 (𝒬) consists of the algebraic subsets. Thus we recover Corollary 2.17.

Aside 2.24. The algebraic subsets of 𝔸𝑛 capture only part of the ideal theory of 𝑘[𝑋1 , … , 𝑋𝑛 ]

finer notion of an algebraic scheme over 𝑘 for which the closed algebraic subschemes of 𝔸𝑛 are
because two ideals with the same radical correspond to the same algebraic subset. There is a

in one-to-one correspondence with the ideals in 𝑘[𝑋1 , … , 𝑋𝑛 ] (see Chapter 10 on my website).

f. Finding the radical of an ideal

Typically, an algebraic set 𝑉 is defined by a finite set of polynomials {𝑔1 , … , 𝑔𝑠 }, and we
need to find 𝐼(𝑉) = rad(𝑔1 , … , 𝑔𝑠 ).

Proposition 2.25. A polynomial ℎ ∈ rad(𝔞) if and only if 1 ∈ (𝔞, 1 − 𝑌ℎ) (the ideal in
𝑘[𝑋1 , … , 𝑋𝑛 , 𝑌] generated by the elements of 𝔞 and 1 − 𝑌ℎ).

Proof. We saw that 1 ∈ (𝔞, 1 − 𝑌ℎ) implies ℎ ∈ rad(𝔞) in the course of proving 2.16.
Conversely, from the identities

1 = 𝑌 𝑁 ℎ𝑁 + (1 − 𝑌 𝑁 ℎ𝑁 ) = 𝑌 𝑁 ℎ𝑁 + (1 − 𝑌ℎ) ⋅ (1 + 𝑌ℎ + ⋯ + 𝑌 𝑁−1 ℎ𝑁−1 )

we see that, if ℎ𝑁 ∈ 𝔞, then 1 ∈ 𝔞 + (1 − 𝑌ℎ). 2

Given a set of generators of an ideal in 𝑘[𝑋1 , … , 𝑋𝑛 ], there is an algorithm for deciding

whether or not a polynomial belongs to the ideal, and hence an algorithm for deciding
whether or not a polynomial belongs to the radical of the ideal. There are even algorithms
for finding a set of generators for the radical. These algorithms have been implemented
in the computer algebra systems CoCoA and Macaulay2.

g. Properties of the Zariski topology

We now examine more closely the Zariski topology on 𝔸𝑛 and on its algebraic subsets.
Proposition 2.14 says that, for a subset 𝑊 of 𝔸𝑛 , 𝑉𝐼(𝑊) is the closure of 𝑊, and 2.17
says that there is a one-to-one correspondence between the closed subsets of 𝔸𝑛 and
the radical ideals of 𝑘[𝑋1 , … , 𝑋𝑛 ]. Under this correspondence, the closed subsets of an
algebraic set 𝑉 correspond to the radical ideals of 𝑘[𝑋1 , … , 𝑋𝑛 ] containing 𝐼(𝑉).

Proposition 2.26. Let 𝑉 be an algebraic subset of 𝔸𝑛 .

(a) The points of 𝑉 are closed for the Zariski topology.
(b) Every ascending chain of open subsets 𝑈1 ⊂ 𝑈2 ⊂ ⋯ of 𝑉 eventually becomes
constant. Equivalently, every descending chain of closed subsets of 𝑉 eventually

(c) Every open covering of 𝑉 has a finite subcovering.

becomes constant.
 h. Decomposition of an algebraic set into irreducible algebraic sets 45

Proof. (a) We have seen that {(𝑎1 , … , 𝑎𝑛 )} is the algebraic set defined by the ideal
(𝑋1 − 𝑎1 , … , 𝑋𝑛 − 𝑎𝑛 ).
(b) We prove the second statement. A sequence 𝑉1 ⊃ 𝑉2 ⊃ ⋯ of closed subsets of 𝑉
gives rise to a sequence of radical ideals 𝐼(𝑉1 ) ⊂ 𝐼(𝑉2 ) ⊂ …, which eventually becomes
constant because 𝑘[𝑋1 , … , 𝑋𝑛 ] is noetherian.
(c) Suppose given an open covering of 𝑉, and let 𝒰 be the collection of open subsets
of 𝑉 that can be expressed as a finite union of sets in the covering. If 𝒰 does not contain
𝑉, then every element of 𝒰 is properly contained in another element, and so there exists
an infinite ascending chain of sets in 𝒰 (axiom of dependent choice), contradicting (b).2

A topological space whose points are closed is said to be 𝑇1 ; the condition means
that, for any pair of distinct points, each has an open neighbourhood not containing

condition is equivalent to the following: every nonempty set of closed subsets of 𝑉 has a
the other. A topological space having the property (b) is said to be noetherian. The

minimal element. A topological space having property (c) is said to be quasi-compact.3

The proof of (c) shows that every noetherian space is quasi-compact. Since any open
subset of a noetherian space is again noetherian, it is also quasi-compact.

h. Decomposition of an algebraic set into irreducible

algebraic sets
A topological space is said to be irreducible if it is not the union of two proper closed
subsets. Equivalent conditions: every pair of nonempty open subsets has nonempty
intersection; every nonempty open subset is dense. By convention, the empty topological
space is not irreducible.
The closure of an irreducible space is irreducible and a nonempty open subset of an
irreducible space is irreducible.
A topological space is connected if it is not the union of two disjoint proper closed
subsets. Therefore, irreducible topological spaces are connected.
In a Hausdorff topological space, any two points have disjoint open neighbourhoods.

Proposition 2.27. An algebraic set 𝑊 is irreducible if and only if 𝐼(𝑊) is prime.

Therefore, the only irreducible Hausdorff spaces are those consisting of a single point.

Proof. Let 𝑊 be an irreducible algebraic set, and let 𝑓𝑔 ∈ 𝐼(𝑊) — we have to show
that either 𝑓 or 𝑔 is in 𝐼(𝑊). At each point of 𝑊, either 𝑓 is zero or 𝑔 is zero, and so
𝑊 ⊂ 𝑉(𝑓) ∪ 𝑉(𝑔). Hence

𝑊 = (𝑊 ∩ 𝑉(𝑓)) ∪ (𝑊 ∩ 𝑉(𝑔)).

As 𝑊 is irreducible, one of these sets, say, 𝑊 ∩ 𝑉(𝑓), must equal 𝑊. But then 𝑓 ∈ 𝐼(𝑊).
Let 𝑊 be an algebraic set such that 𝐼(𝑊) is prime, and let 𝑊 = 𝑉(𝔞) ∪ 𝑉(𝔟) with 𝔞
and 𝔟 radical ideals — we have to show that 𝑊 equals 𝑉(𝔞) or 𝑉(𝔟). The ideal 𝔞 ∩ 𝔟 is
radical, and 𝑉(𝔞 ∩ 𝔟) = 𝑉(𝔞) ∪ 𝑉(𝔟) (2.10); hence 𝐼(𝑊) = 𝔞 ∩ 𝔟. If 𝑊 ≠ 𝑉(𝔞), then there
exists an 𝑓 ∈ 𝔞 ∖ 𝐼(𝑊). Let 𝑔 ∈ 𝔟. Then 𝑓𝑔 ∈ 𝔞 ∩ 𝔟 = 𝐼(𝑊), and so 𝑔 ∈ 𝐼(𝑊) (here we
use that 𝐼(𝑊) is prime). We conclude that 𝔟 ⊂ 𝐼(𝑊), and so 𝑉(𝔟) ⊃ 𝑉(𝐼(𝑊)) = 𝑊. 2
3
Bourbaki’s terminology
46 2. Algebraic Sets

Summary 2.28. There are one-to-one correspondences,

radical ideals in 𝑘[𝑋1 , … , 𝑋𝑛 ] ↔ algebraic subsets of 𝔸𝑛

prime ideals in 𝑘[𝑋1 , … , 𝑋𝑛 ] ↔ irreducible algebraic subsets of 𝔸𝑛
maximal ideals in 𝑘[𝑋1 , … , 𝑋𝑛 ] ↔ one-point subsets of 𝔸𝑛 .

Example 2.29. Let 𝑓 ∈ 𝑘[𝑋1 , … , 𝑋𝑛 ]. We know that 𝑘[𝑋1 , … , 𝑋𝑛 ] is a unique factor-

ization domain (1.32), and so (𝑓) is a prime ideal if and only if 𝑓 is irreducible (1.33).

𝑓 is irreducible ⇐⇒ 𝑉(𝑓) is irreducible.

Thus

On the other hand, suppose that 𝑓 factors as

∏ 𝑚𝑖
𝑓= 𝑓𝑖 , 𝑓𝑖 distinct irreducible polynomials.

⋂( 𝑚𝑖 ) ( 𝑚𝑖 )
(𝑓) = 𝑓𝑖 𝑓𝑖
Then

⋂
rad(𝑓) = (𝑓𝑖 ) (𝑓𝑖 ) distinct prime ideals
distinct ideals

⋃
𝑉(𝑓) = 𝑉(𝑓𝑖 ) 𝑉(𝑓𝑖 ) distinct irreducible algebraic sets.

Lemma 2.30. Let 𝑊 be an irreducible topological space. If 𝑊 = 𝑊1 ∪ … ∪ 𝑊𝑟 with each

𝑊𝑖 closed, then 𝑊 is equal to one of the 𝑊𝑖 .

Proof. When 𝑟 = 2, the statement is the definition of “irreducible”. Suppose that 𝑟 > 2.
Then 𝑊 = 𝑊1 ∪ (𝑊2 ∪ … ∪ 𝑊𝑟 ), and so 𝑊 = 𝑊1 or 𝑊 = (𝑊2 ∪ … ∪ 𝑊𝑟 ); if the second,
then 𝑊 = 𝑊2 or 𝑊3 ∪ … ∪ 𝑊𝑟 , etc.

Proposition 2.31. Let 𝑉 be a nonempty noetherian topological space. Then 𝑉 is a finite

2

union of irreducible closed subsets, 𝑉 = 𝑉1 ∪ … ∪ 𝑉𝑚 . If the decomposition is irredundant

in the sense that there are no inclusions among the 𝑉𝑖 , then the 𝑉𝑖 are uniquely determined
up to order.

Proof. Suppose that 𝑉 cannot be written as a finite union of irreducible closed subsets.
Then, because 𝑉 is noetherian, there will be a nonempty closed subset 𝑊 of 𝑉 that
is minimal among those that cannot be written in this way. But 𝑊 itself cannot be
irreducible, and so 𝑊 = 𝑊1 ∪ 𝑊2 , with 𝑊1 and 𝑊2 proper closed subsets of 𝑊. Because
𝑊 was minimal, each 𝑊𝑖 is a finite union of irreducible closed subsets. Hence 𝑊 is also,
which is a contradiction.

𝑉 = 𝑉1 ∪ … ∪ 𝑉𝑚 = 𝑊1 ∪ … ∪ 𝑊𝑛
Suppose that

⋃
are two irredundant decompositions of 𝑉. Then 𝑉𝑖 = 𝑗 (𝑉𝑖 ∩ 𝑊𝑗 ), and so, because 𝑉𝑖 is
irreducible, 𝑉𝑖 = 𝑉𝑖 ∩ 𝑊𝑗 for some 𝑗. Consequently, there is a function 𝑓 ∶ {1, … , 𝑚} →
{1, … , 𝑛} such that 𝑉𝑖 ⊂ 𝑊𝑓(𝑖) for each 𝑖. Similarly, there is a function 𝑔 ∶ {1, … , 𝑛} →
{1, … , 𝑚} such that 𝑊𝑗 ⊂ 𝑉𝑔(𝑗) for each 𝑗. Since 𝑉𝑖 ⊂ 𝑊𝑓(𝑖) ⊂ 𝑉𝑔𝑓(𝑖) , we must have
𝑔𝑓(𝑖) = 𝑖 and 𝑉𝑖 = 𝑊𝑓(𝑖) ; similarly 𝑓𝑔 = id. Thus 𝑓 and 𝑔 are bijections, and the
decompositions differ only in the numbering of the sets.

The 𝑉𝑖 given uniquely by the proposition are called the irreducible components of
2

𝑉. They are exactly the maximal irreducible subsets of 𝑉. In Example 2.29, the 𝑉(𝑓𝑖 )
are the irreducible components of 𝑉(𝑓).
 h. Decomposition of an algebraic set into irreducible algebraic sets 47

A connected algebraic set with two irreducible components.

Corollary 2.32. The radical of an ideal 𝔞 in 𝑘[𝑋1 , … , 𝑋𝑛 ] is a finite intersection of prime

ideals, rad(𝔞) = 𝔭1 ∩ … ∩ 𝔭𝑛 . If there are no inclusions among the 𝔭𝑖 , then the 𝔭𝑖 are

𝔞).
uniquely determined up to order (and they are exactly the minimal prime ideals containing

⋃𝑛
Proof. Write 𝑉(𝔞) as a union of its irreducible components, 𝑉(𝔞) = 𝑖=1 𝑉𝑖 , and let
𝔭𝑖 = 𝐼(𝑉𝑖 ). Then rad(𝔞) = 𝔭1 ∩ … ∩ 𝔭𝑛 because they are both radical ideals and
⋃ ⋂
𝑉(rad(𝔞)) = 𝑉(𝔞) = 𝑉(𝔭𝑖 ) = 𝑉( 𝔭).
𝑖
2.10b

The uniqueness similarly follows from the proposition. 2

Remarks

not be irreducible. For example, the union of two surfaces in 3-space intersecting along
An irreducible topological space is connected, but a connected topological space need

2.33. An algebraic subset 𝑉 of 𝔸𝑛 is disconnected if and only if there exist radical ideals
a curve is reducible, but connected.

𝔞 and 𝔟 such that 𝑉 is the disjoint union of 𝑉(𝔞) and 𝑉(𝔟), so

𝑉 = 𝑉(𝔞) ∪ 𝑉(𝔟) = 𝑉(𝔞 ∩ 𝔟) ⇐⇒ 𝔞 ∩ 𝔟 = 𝐼(𝑉)
{
∅ = 𝑉(𝔞) ∩ 𝑉(𝔟) = 𝑉(𝔞 + 𝔟) ⇐⇒ 𝔞 + 𝔟 = 𝑘[𝑋1 , … , 𝑋𝑛 ].

𝑘[𝑋1 , … , 𝑋𝑛 ] 𝑘[𝑋1 , … , 𝑋𝑛 ]
𝑘[𝑉] ≃ 𝔞 ×
Then

𝔟
by Theorem 1.1.

2.34. A Hausdorff space is noetherian if and only if it is finite, in which case its irre-

2.35. In 𝑘[𝑋1 , … , 𝑋𝑛 ], a principal ideal (𝑓) is radical if and only if 𝑓 is square-free,

ducible components are the one-point sets.

in which case 𝑓 is a product of distinct irreducible polynomials, 𝑓 = 𝑓1 … 𝑓𝑟 , and

(𝑓) = (𝑓1 ) ∩ … ∩ (𝑓𝑟 ).
48 2. Algebraic Sets

Aside 2.36. Let 𝐴 be a noetherian ring. A proper ideal 𝔮 in 𝐴 is primary if every zero-divisor in
𝐴∕𝔮 is nilpotent. Every ideal 𝔞 in 𝐴 can be written as an intersection of primary ideals

𝔞 = 𝔮1 ∩ … ∩ 𝔮𝑛 .

Choose a minimal such decomposition, and let 𝔭𝑖 = rad(𝔮𝑖 ). Then each 𝔭𝑖 is prime, and

rad(𝔞) = 𝔭1 ∩ … ∩ 𝔭𝑛 .
∏
See CA, §19. For the ideal (𝑓) in 2.35, with 𝑓 = 𝑓𝑖 𝑖 , these decompositions become
𝑚

(𝑓) = (𝑓1 1 ) ∩ … ∩ (𝑓𝑛 𝑚 ) and

𝑚 𝑛

rad(𝑓) = (𝑓1 ) ∩ … ∩ (𝑓𝑛 ).

i. Regular functions; the coordinate ring of an algebraic set

Let 𝑉 be an algebraic subset of 𝔸𝑛 , and let 𝐼(𝑉) = 𝔞. The coordinate ring of 𝑉 is

𝑘[𝑉] = 𝑘[𝑋1 , … , 𝑋𝑛 ]∕𝔞.

def

This is a finitely generated 𝑘-algebra. It is reduced because 𝔞 is radical, but it is not

necessarily an integral domain. An 𝑓 ∈ 𝑘[𝑋1 , … , 𝑋𝑛 ] defines a function

𝑃 ↦ 𝑓(𝑃) ∶ 𝑉 → 𝑘.

Functions of this form are said to be regular. Two polynomials 𝑓, 𝑔 ∈ 𝑘[𝑋1 , … , 𝑋𝑛 ]

define the same function on 𝑉 if and only if they define the same element of 𝑘[𝑉], and
so 𝑘[𝑉] is the ring of regular functions on 𝑉. The coordinate function

𝑥𝑖 ∶ 𝑉 → 𝑘, (𝑎1 , … , 𝑎𝑛 ) ↦ 𝑎𝑖

is regular, and 𝑘[𝑉] = 𝑘[𝑥1 , … , 𝑥𝑛 ], so the coordinate ring of 𝑉 is the 𝑘-algebra generated
by the coordinate functions on 𝑉.
For an ideal 𝔟 in 𝑘[𝑉], set

𝑉(𝔟) = {𝑃 ∈ 𝑉 ∣ 𝑓(𝑃) = 0, all 𝑓 ∈ 𝔟}

— it is a closed subset of 𝑉. Let 𝑊 = 𝑉(𝔟). The quotient maps

𝑘[𝑋1 , … , 𝑋𝑛 ] 𝑘[𝑉]
𝑘[𝑋1 , … , 𝑋𝑛 ] ↠ 𝑘[𝑉] = 𝔞 ↠ 𝑘[𝑊] =
𝔟
send a regular function on 𝑘 𝑛 to its restriction to 𝑉 and then to its restriction to 𝑊.
Write 𝜋 for the quotient map 𝑘[𝑋1 , … , 𝑋𝑛 ] ↠ 𝑘[𝑉]. Then 𝔟 ↦ 𝜋−1 (𝔟) is a bijection
from the set of ideals of 𝑘[𝑉] to the set of ideals of 𝑘[𝑋1 , … , 𝑋𝑛 ] containing 𝔞, under
which radical, prime, and maximal ideals correspond to radical, prime, and maximal

𝑘[𝑋1 , … , 𝑋𝑛 ]∕𝜋−1 (𝔟) ≃ 𝑘[𝑉]∕𝔟). Clearly

ideals (because each of these conditions can be checked on the quotient ring, and

𝑉(𝜋−1 (𝔟)) = 𝑉(𝔟),

and so 𝔟 ↦ 𝑉(𝔟) is a bijection from the set of radical ideals in 𝑘[𝑉] to the set of algebraic
sets contained in 𝑉.
 j. Regular maps 49

Now 2.28 holds for ideals in 𝑘[𝑉] and algebraic subsets of 𝑉,

radical ideals in 𝑘[𝑉] ↔ algebraic subsets of 𝑉

prime ideals in 𝑘[𝑉] ↔ irreducible algebraic subsets of 𝑉
maximal ideals in 𝑘[𝑉] ↔ one-point sets of 𝑉.

Moreover (see 2.33), the decompositions of a closed subset 𝑊 of 𝑉 into a disjoint union
of closed subsets correspond to pairs of radical ideals 𝔞, 𝔟 ∈ 𝑘[𝑉] such that

𝑘[𝑊] = 𝑘[𝑉]∕𝔞 ∩ 𝔟 ≃ 𝑘[𝑉]∕𝔞 × 𝑘[𝑉]∕𝔟.

For ℎ ∈ 𝑘[𝑉], let

𝐷(ℎ) = {𝑎 ∈ 𝑉 ∣ ℎ(𝑎) ≠ 0}.
It is an open subset of 𝑉, because its complement is the closed set 𝑉((ℎ)). It is empty if
and only if ℎ is zero (2.19).

Proposition 2.37. The sets 𝐷(ℎ), ℎ ∈ 𝑘[𝑉], form a base for the topology on 𝑉: each 𝐷(ℎ)
is open and every open set is a (finite) union of sets 𝐷(ℎ).

Proof. We have already observed that 𝐷(ℎ) is open. Every open subset 𝑈 of 𝑉 is the
⋃
complement of a set 𝑉(𝔞), and if 𝑓1 , … , 𝑓𝑚 generate the ideal 𝔞, then 𝑈 = 𝐷(𝑓𝑖 ). 2

The 𝐷(ℎ) are called the basic (or principal) open subsets of 𝑉. We sometimes write
𝑉ℎ for 𝐷(ℎ). Note that

𝐷(ℎ) ⊂ 𝐷(ℎ′ ) ⇐⇒ 𝑉(ℎ) ⊃ 𝑉(ℎ′ )

⇐⇒ rad((ℎ)) ⊂ rad((ℎ′ ))
⇐⇒ ℎ𝑟 ∈ (ℎ′ ) some 𝑟
⇐⇒ ℎ𝑟 = ℎ′ 𝑔, some 𝑔.

Some of this should look familiar: if 𝑉 is a topological space, then the zero set of a
family of continuous functions 𝑓 ∶ 𝑉 → ℝ is closed, and the set where a continuous

If the algebraic set 𝑉 is irreducible, then 𝐼(𝑉) is a prime ideal, and 𝑘[𝑉] is an integral
function is nonzero is open.

domain. Its field of fractions, 𝑘(𝑉) is called the function field of 𝑉 or the field of
rational functions on 𝑉.

j. Regular maps
Let 𝑊 ⊂ 𝑘𝑚 and 𝑉 ⊂ 𝑘𝑛 be algebraic sets, and let 𝑥𝑖 denote the 𝑖th coordinate function

(𝑏1 , … , 𝑏𝑛 ) ↦ 𝑏𝑖 ∶ 𝑉 → 𝑘.

The 𝑖th component of a map 𝜑 ∶ 𝑊 → 𝑉 is

𝜑𝑖 = 𝑥𝑖 ◦𝜑.
def

Thus, 𝜑 is the map

𝑃 ↦ (𝜑1 (𝑃), … , 𝜑𝑛 (𝑃)) ∶ 𝑊 → 𝑉 ⊂ 𝑘𝑛 .
50 2. Algebraic Sets

Definition 2.38. A continuous map 𝜑 ∶ 𝑊 → 𝑉 of algebraic sets is regular if each of

its components 𝜑𝑖 is a regular function on 𝑊.

As the coordinate functions generate 𝑘[𝑉], a continuous map 𝜑 is regular if and only if
𝑓◦𝜑 is a regular function on 𝑊 for every regular function 𝑓 on 𝑉. Thus a regular map
𝜑 ∶ 𝑊 → 𝑉 of algebraic sets defines a homomorphism 𝑓 ↦ 𝑓◦𝜑 ∶ 𝑘[𝑉] → 𝑘[𝑊] of
𝑘-algebras, which we sometimes denote by 𝜑∗ .

k. Hypersurfaces; finite and quasi-finite maps

A hypersurface in 𝔸𝑛+1 is the algebraic set 𝐻 defined by a single nonzero nonconstant

𝐻 ∶ 𝑓(𝑇1 , … , 𝑇𝑛 , 𝑋) = 0.
polynomial,

We examine the regular map 𝐻 → 𝔸𝑛 defined by the projection

(𝑡1 , … , 𝑡𝑛 , 𝑥) ↦ (𝑡1 , … , 𝑡𝑛 ).

We can write 𝑓 in the form

𝑓 = 𝑎0 𝑋 𝑚 + 𝑎1 𝑋 𝑚−1 + ⋯ + 𝑎𝑚 , 𝑎𝑖 ∈ 𝑘[𝑇1 , … , 𝑇𝑛 ], 𝑎0 ≠ 0.

We assume that 𝑚 ≠ 0, i.e., that 𝑋 occurs in 𝑓 (otherwise, 𝐻 is a cylinder over a

hypersurface in 𝔸𝑛 ). The fibre of the map 𝐻 → 𝔸𝑛 over (𝑡1 , … , 𝑡𝑛 ) ∈ 𝑘 𝑛 is the set of
points (𝑡1 , … , 𝑡𝑛 , 𝑐) such that 𝑐 is a root of the polynomial

𝑎0 (𝑡)𝑋 𝑚 + 𝑎1 (𝑡)𝑋 𝑚−1 + ⋯ + 𝑎𝑚 (𝑡), 𝑎𝑖 (𝑡) = 𝑎𝑖 (𝑡1 , … , 𝑡𝑛 ) ∈ 𝑘.

def

Suppose first that 𝑎0 ∈ 𝑘, so that 𝑎0 (𝑡) is a nonzero constant independent of 𝑡. Then

the fibre over 𝑡 consists of the roots of the polynomial

𝑎0 𝑋 𝑚 + 𝑎1 (𝑡)𝑋 𝑚−1 + ⋯ + 𝑎𝑚 (𝑡), (14)

in 𝑘[𝑋]. Counting multiplicities, there are exactly 𝑚 of these. More precisely, let 𝐷 be
the discriminant of the polynomial4

𝑎0 𝑋 𝑚 + 𝑎1 𝑋 𝑚−1 + ⋯ + 𝑎𝑚 .

Then 𝐷 ∈ 𝑘[𝑋1 , … , 𝑋𝑚 ], and the fibre has exactly 𝑚 points over the open subset where
𝐷 ≠ 0, and fewer then 𝑚 points over the closed subset where 𝐷 = 0.5 We can picture it
schematically as follows (𝑚 = 3):

𝔸𝑛

I am ignoring the possibility that 𝐷 is identically zero. This case occurs when the characteristic is
4
See FT, p. 57 et seq. for discriminants.

𝑝 ≠ 0, and 𝑓 is a polynomial in 𝑇1 , … , 𝑇𝑛 , and 𝑋 𝑝 .

5
 l. Noether normalization theorem 51

Now drop the condition that 𝑎0 is constant. For certain 𝑡, the degree of (14) may

𝑓(𝑇, 𝑋) = 𝑇𝑋 − 1, then there is one point (𝑡, 1∕𝑡) in the fibre over 𝑡 when 𝑡 ≠ 0 but no
drop, which means that some roots have “disappeared off to infinity”. For example, if

point when 𝑡 = 0. Worse, for certain 𝑡 all coefficients may be zero, in which case the
fibre is a line. In general, there is a nested collection of closed subsets of 𝔸𝑛 such that
the number of points in the fibre (counting multiplicities) drops as you pass to a smaller

Definition 2.39. Let 𝜑 ∶ 𝑊 → 𝑉 be a regular map of algebraic subsets and 𝜑∗ ∶ 𝑘[𝑉] →

subset, except that over the smallest subset the fibre may be a full line.

𝑘[𝑊] the corresponding map 𝑓 ↦ 𝑓◦𝜑 on rings.

(a) The map 𝜑 is dominant if 𝜑(𝑊) is dense in 𝑉, i.e., every nonempty open subset
of 𝑉 intersects 𝜑(𝑊).
(b) The map 𝜑 is quasi-finite if 𝜑−1 (𝑃) is finite for all 𝑃 ∈ 𝑉.
(c) The map 𝜑 is finite if 𝑘[𝑊] is a finite 𝑘[𝑉]-algebra.

Finite maps are quasi-finite. To see this, note that the points of 𝑊 lying over a point
𝑃 of 𝑉 correspond to the maximal ideals 𝔪 of 𝑘[𝑊] such that 𝜑∗−1 (𝔪) = 𝔪𝑃 , and that
these correspond to the maximal ideals of 𝐴 = 𝑘[𝑊] ⊗𝑘[𝑉] (𝑘[𝑉]∕𝔪𝑃 ). If 𝜑 is finite,
then 𝐴 is a finite 𝑘-algebra. Let 𝔪1 , … , 𝔪𝑛 be maximal ideals in 𝐴. Then 𝔪𝑖 + 𝔪𝑗 = 𝐴
def

for 𝑖 ≠ 𝑗, and so the map

𝐴 → 𝐴∕𝔪1 × ⋯ × 𝐴∕𝔪𝑛
is surjective (1.1). Thus 𝑛 is at most the dimension of 𝐴 as a 𝑘-algebra.
As 𝑘[𝑊] is finitely generated as a 𝑘-algebra, hence as a 𝑘[𝑉]-algebra, to say that
𝑘[𝑊] is a finite 𝑘[𝑉]-algebra means that it is integral over 𝑘[𝑉] (1.36).
The map 𝐻 → 𝔸𝑛 considered above is finite if and only if 𝑎0 is constant, and quasi-
finite if and only if the polynomials 𝑎0 , … , 𝑎𝑚 have no common zero in 𝑘 𝑛 .

Proposition 2.40. A regular map 𝜑 ∶ 𝑊 → 𝑉 is dominant if and only if 𝜑∗ ∶ 𝑘[𝑉] →

𝑘[𝑊] is injective.

Proof. If 𝜑 is dominant and 𝑓 ∈ 𝑘[𝑉] is nonzero, then 𝐷(𝑓) intersects 𝜑(𝑊), and so
𝑓◦𝜑 ≠ 0. If 𝜑 is not dominant, then its image is contained in a proper closed subset of
𝑉, which is contained in 𝑉(𝑓) for some nonzero 𝑓 ∈ 𝑘[𝑉]; then 𝑓◦𝜑 = 0. 2

Proposition 2.41. A dominant finite map is surjective.

Proof. Let 𝜑 ∶ 𝑊 → 𝑉 be dominant and finite. Then 𝜑∗ ∶ 𝑘[𝑉] → 𝑘[𝑊] is injective,

and 𝑘[𝑊] is integral over the image of 𝑘[𝑉]. According to the going-up theorem (1.53),
for every maximal ideal 𝔪 of 𝑘[𝑉] there exists a maximal ideal 𝔫 of 𝑘[𝑊] such that
𝔪 = 𝔫 ∩ 𝑘[𝑉]. Because of the correspondence between points and maximal ideals, this
implies that 𝜑 is surjective. 2

l. Noether normalization theorem

Let 𝐻 be a hypersurface in 𝔸𝑛+1 . We show that, after a linear change of coordinates, the
projection map (𝑥1 , … , 𝑥𝑛+1 ) ↦ (𝑥1 , … , 𝑥𝑛 ) ∶ 𝔸𝑛+1 → 𝔸𝑛 defines a finite map 𝐻 → 𝔸𝑛 .
52 2. Algebraic Sets

𝐻∶ 𝑓(𝑋1 , … , 𝑋𝑛+1 ) = 0
Proposition 2.42. Let

be a hypersurface in 𝔸𝑛+1 . There exist 𝑐1 , … , 𝑐𝑛 ∈ 𝑘 such that the map 𝐻 → 𝔸𝑛 defined by

(𝑥1 , … , 𝑥𝑛+1 ) ↦ (𝑥1 − 𝑐1 𝑥𝑛+1 , … , 𝑥𝑛 − 𝑐𝑛 𝑥𝑛+1 )

is finite.

Proof. Let 𝑐1 , … , 𝑐𝑛 ∈ 𝑘. In terms of the coordinates 𝑥𝑖′ = 𝑥𝑖 − 𝑐𝑖 𝑥𝑛+1 , the hyperplane

𝐻 is the zero set of

𝑓(𝑋1 + 𝑐1 𝑋𝑛+1 , … , 𝑋𝑛 + 𝑐𝑛 𝑋𝑛+1 , 𝑋𝑛+1 ) = 𝑎0 𝑋𝑛+1

𝑚
+ 𝑎1 𝑋𝑛+1
𝑚−1
+ ⋯.

The next lemma shows that the 𝑐𝑖 can be chosen so that 𝑎0 is a nonzero constant. This
implies that the map 𝐻 → 𝔸𝑛 defined by (𝑥1 , … , 𝑥𝑛+1 ) ↦ (𝑥1′ , … , 𝑥𝑛′ ) is finite. 2

Lemma 2.43. Let 𝑘 be an infinite field (not necessarily algebraically closed), and let 𝑓 ∈
𝑘[𝑋1 , … , 𝑋𝑛 , 𝑇]. There exist 𝑐1 , … , 𝑐𝑛 ∈ 𝑘 such that

𝑓(𝑋1 + 𝑐1 𝑇, … , 𝑋𝑛 + 𝑐𝑛 𝑇, 𝑇) = 𝑎0 𝑇 𝑚 + 𝑎1 𝑇 𝑚−1 + ⋯ + 𝑎𝑚

with 𝑎0 ∈ 𝑘× and all 𝑎𝑖 ∈ 𝑘[𝑋1 , … , 𝑋𝑛 ].

Proof. Let 𝐹 be the homogeneous part of highest degree of 𝑓 and let 𝑟 = deg(𝐹). Then

𝐹(𝑋1 + 𝑐1 𝑇, … , 𝑋𝑛 + 𝑐𝑛 𝑇, 𝑇) = 𝐹(𝑐1 , … , 𝑐𝑛 , 1)𝑇 𝑟 + terms of degree < 𝑟 in 𝑇,

because the polynomial 𝐹(𝑋1 + 𝑐1 𝑇, … , 𝑋𝑛 + 𝑐𝑛 𝑇, 𝑇) is still homogeneous of degree 𝑟

in 𝑋1 , … , 𝑋𝑛 , 𝑇, and so the coefficient of the monomial 𝑇 𝑟 can be obtained by setting
each 𝑋𝑖 equal to zero in 𝐹 and 𝑇 to 1. As 𝐹(𝑋1 , … , 𝑋𝑛 , 𝑇) is a nonzero homogeneous
polynomial, 𝐹(𝑋1 , … , 𝑋𝑛 , 1) is a nonzero polynomial, and so we can choose the 𝑐𝑖 so that
𝐹(𝑐1 , … , 𝑐𝑛 , 1) ≠ 0 (Exercise 1-1). Now

𝑓(𝑋1 + 𝑐1 𝑇, … , 𝑋𝑛 + 𝑐𝑛 𝑇, 𝑇) = 𝐹(𝑐1 , … , 𝑐𝑛 , 1)𝑇 𝑟 + terms of degree < 𝑟 in 𝑇,

with 𝐹(𝑐1 , … , 𝑐𝑛 , 1) ∈ 𝑘 × , as required. 2

In fact, every algebraic set 𝑉 admits a finite surjective map to 𝔸𝑑 for some 𝑑.

Theorem 2.44. Let 𝑉 be an algebraic set. For some natural number 𝑑, there exists a finite
surjective map 𝜑 ∶ 𝑉 → 𝔸𝑑 .

This follows from the next statement applied to 𝐴 = 𝑘[𝑉]: the regular functions
𝑥1 , … , 𝑥𝑑 define a map 𝑉 → 𝔸𝑑 , which is finite and surjective because 𝑘[𝑥1 , … , 𝑥𝑑 ] → 𝐴

Theorem 2.45 (Noether Normalization Theorem). Let 𝐴 be a finitely generated

is finite and injective.

𝑘-algebra. There exist elements 𝑥1 , … , 𝑥𝑑 ∈ 𝐴 that are algebraically independent over 𝑘,

and such that 𝐴 is finite over 𝑘[𝑥1 , … , 𝑥𝑑 ].
 l. Noether normalization theorem 53

It is not necessary to assume that 𝐴 is reduced in Theorem 2.45, nor that 𝑘 is alge-
braically closed, although the proof we give requires it to be infinite (for the general

Let 𝐴 = 𝑘[𝑥1 , … , 𝑥𝑛 ]. We prove the theorem by induction on 𝑛. If the 𝑥𝑖 are alge-

proof, see CA, 8.1).

𝐴 is finite over a subring 𝐵 = 𝑘[𝑦1 , … , 𝑦𝑛−1 ]. By induction, 𝐵 is finite over a subring

braically independent, there is nothing to prove. Otherwise, the next lemma shows that

𝐶 = 𝑘[𝑧1 , … , 𝑧𝑑 ] with 𝑧1 , … , 𝑧𝑑 algebraically independent, and 𝐴 is finite over 𝐶.

Lemma 2.46. Let 𝐴 = 𝑘[𝑥1 , … , 𝑥𝑛 ] be a finitely generated 𝑘-algebra, and let {𝑥1 , … , 𝑥𝑑 }
be a maximal algebraically independent subset of {𝑥1 , … , 𝑥𝑛 }. If 𝑛 > 𝑑, then there exist
𝑐1 , … , 𝑐𝑑 ∈ 𝑘 such that 𝐴 is finite over 𝑘[𝑥1 − 𝑐1 𝑥𝑛 , … , 𝑥𝑑 − 𝑐𝑑 𝑥𝑛 , 𝑥𝑑+1 , … , 𝑥𝑛−1 ].
Proof. By assumption, the set {𝑥1 , … , 𝑥𝑑 , 𝑥𝑛 } is algebraically dependent, and so there
exists a nonzero 𝑓 ∈ 𝑘[𝑋1 , … , 𝑋𝑑 , 𝑇] such that
𝑓(𝑥1 , … , 𝑥𝑑 , 𝑥𝑛 ) = 0.
Because {𝑥1 , … , 𝑥𝑑 } is algebraically independent, 𝑇 occurs in 𝑓, and so
(15)

𝑓(𝑋1 , … , 𝑋𝑑 , 𝑇) = 𝑎0 𝑇 𝑚 + 𝑎1 𝑇 𝑚−1 + ⋯ + 𝑎𝑚
with 𝑎𝑖 ∈ 𝑘[𝑋1 , … , 𝑋𝑑 ], 𝑎0 ≠ 0, and 𝑚 > 0.
If 𝑎0 ∈ 𝑘, then (15) shows that 𝑥𝑛 is integral over 𝑘[𝑥1 , … , 𝑥𝑑 ]. Hence 𝑥1 , … , 𝑥𝑛 are
integral over 𝑘[𝑥1 , … , 𝑥𝑛−1 ], and so 𝐴 is finite over 𝑘[𝑥1 , … , 𝑥𝑛−1 ].
If 𝑎0 ∉ 𝑘, then, for a suitable choice of (𝑐1 , … , 𝑐𝑑 ) ∈ 𝑘, the polynomial
𝑔(𝑋1 , … , 𝑋𝑑 , 𝑇) = 𝑓(𝑋1 + 𝑐1 𝑇, … , 𝑋𝑑 + 𝑐𝑑 𝑇, 𝑇)
def

𝑔(𝑋1 , … , 𝑋𝑑 , 𝑇) = 𝑏𝑇 𝑟 + 𝑏1 𝑇 + ⋯ + 𝑏𝑟
takes the form

with 𝑏 ∈ 𝑘× (see 2.43). As

𝑔(𝑥1 − 𝑐1 𝑥𝑛 , … , 𝑥𝑑 − 𝑐𝑑 𝑥𝑛 , 𝑥𝑛 ) = 0
this shows that 𝑥𝑛 is integral over 𝑘[𝑥1 − 𝑐1 𝑥𝑛 , … , 𝑥𝑑 − 𝑐𝑑 𝑥𝑛 ], and so 𝐴 is finite over
(16)

𝑘[𝑥1 − 𝑐1 𝑥𝑛 , … , 𝑥𝑑 − 𝑐𝑑 𝑥𝑛 , 𝑥𝑑+1 , … , 𝑥𝑛−1 ] as before. 2

Remarks
2.47. For an irreducible algebraic subset 𝑉 of 𝔸𝑛 , the above argument can be modified

Let 𝑥1 , … , 𝑥𝑛 be the coordinate functions on 𝑉; after possibly renumbering

to prove the following more precise statement:

the coordinates, we may suppose that {𝑥1 , … , 𝑥𝑑 } is a maximal algebraically

independent subset of {𝑥1 , … , 𝑥𝑛 }; then there exist 𝑐𝑖𝑗 ∈ 𝑘 such that the map
∑
𝑛
∑
𝑛
(𝑥1 , … , 𝑥𝑛 ) ↦ (𝑥1 − 𝑐1𝑗 𝑥𝑗 , … , 𝑥𝑑 − 𝑐𝑑𝑗 𝑥𝑗 ) ∶ 𝔸𝑛 → 𝔸𝑑
𝑗=𝑑+1 𝑗=𝑑+1

induces a finite surjective map 𝑉 → 𝔸𝑑 .

Indeed, Lemma 2.46 shows that there exist 𝑐1 , … , 𝑐𝑛 ∈ 𝑘 such that 𝑘[𝑉] is finite over
𝑘[𝑥1 − 𝑐1 𝑥𝑛 , … , 𝑥𝑑 − 𝑐𝑑 𝑥𝑛 , 𝑥𝑑+1 , … , 𝑥𝑛−1 ]. Now {𝑥1 , … , 𝑥𝑑 } is algebraically dependent
on {𝑥1 − 𝑐1 𝑥𝑛 , … , 𝑥𝑑 − 𝑐𝑑 𝑥𝑛 }. If the second set were not algebraically independent,
we could drop one of its elements, but this would contradict 1.61. Therefore {𝑥1 −
𝑐1 𝑥𝑛 , … , 𝑥𝑑 − 𝑐𝑑 𝑥𝑛 } is a maximal algebraically independent subset of {𝑥1 − 𝑐1 𝑥𝑛 , … , 𝑥𝑑 −
𝑐𝑑 𝑥𝑛 , 𝑥𝑑+1 , … , 𝑥𝑛−1 } and we can repeat the argument.
54 2. Algebraic Sets

m. Dimension
The dimension of a topological space
Let 𝑉 be a noetherian topological space whose points are closed.

Definition 2.48. The dimension of 𝑉 is the supremum of the lengths of the chains

𝑉0 ⊃ 𝑉1 ⊃ ⋯ ⊃ 𝑉𝑑

of distinct irreducible closed subsets (the length of the displayed chain is 𝑑).

2.49. Let 𝑉1 , … , 𝑉𝑚 be the irreducible components of 𝑉. Then (obviously)

dim(𝑉) = max (dim(𝑉𝑖 )).

𝑖

2.50. Assume that 𝑉 is irreducible, and let 𝑊 be a proper closed subspace of 𝑉. Then
every chain 𝑊0 ⊃ 𝑊1 ⊃ ⋯ in 𝑊 extends to a chain 𝑉 ⊃ 𝑊0 ⊃ ⋯, and so dim(𝑊) + 1 ≤
dim(𝑉). If dim(𝑉) < ∞, then dim(𝑊) < dim(𝑉).

Thus an irreducible topological space 𝑉 has dimension 0 if and only if it is a point; it

has dimension ≤ 1 if and only if every proper closed subset is a point; and, inductively,
𝑉 has dimension ≤ 𝑛 if and only if every proper closed subset has dimension ≤ 𝑛 − 1.

The dimension of an algebraic set

Definition 2.51. The dimension of an algebraic set is its dimension as a topological
space.

Because of the correspondence between the prime ideals in 𝑘[𝑉] and irreducible
closed subsets of 𝑉,
dim(𝑉) = Krull dimension of 𝑘[𝑉].
Note that, if 𝑉1 , … , 𝑉𝑚 are the irreducible components of 𝑉, then

dim 𝑉 = max dim(𝑉𝑖 ).

𝑖

When the 𝑉𝑖 all have the same dimension 𝑑, we say that 𝑉 has pure dimension 𝑑. A

a surface; and an 𝑛-dimensional algebraic set is called an 𝑛-fold.

one-dimensional algebraic set is called a curve; a two-dimensional algebraic set is called

Let 𝑉 be an irreducible algebraic set and 𝑊 an algebraic subset of 𝑉. If 𝑊 is irre-

ducible, then its codimension in 𝑉 is

codim𝑉 𝑊 = dim 𝑉 − dim 𝑊.

def

Dimension and transcendent degree

Theorem 2.52. Let 𝑉 be an irreducible algebraic set. Then

dim(𝑉) = tr deg𝑘 𝑘(𝑉).

 m. Dimension 55

Let 𝐴 be an arbitrary commutative ring. Let 𝑥 ∈ 𝐴, and let 𝑆{𝑥} denote the multi-
The proof will occupy the rest of this subsection.

plicative subset of 𝐴 consisting of the elements of the form

𝑥𝑛 (1 − 𝑎𝑥), 𝑛 ∈ ℕ, 𝑎 ∈ 𝐴.

The boundary 𝐴{𝑥} of 𝐴 at 𝑥 is defined to be the ring of fractions 𝑆{𝑥}

−1
𝐴.
We write dim(𝐴) for the Krull dimension of 𝐴.

Proposition 2.53. Let 𝐴 be a ring and let 𝑛 ∈ ℕ. Then

dim(𝐴) ≤ 𝑛 ⇐⇒ for all 𝑥 ∈ 𝐴, dim(𝐴{𝑥} ) ≤ 𝑛 − 1.

Proof. We shall use the one-to-one correspondence between the prime ideals of 𝑆 −1 𝐴
and the prime ideals of 𝐴 disjoint from 𝑆 (1.14). We begin with two observations.
(a) For every 𝑥 ∈ 𝐴 and maximal ideal 𝔪 ⊂ 𝐴, 𝔪 ∩ 𝑆{𝑥} ≠ ∅. Indeed, if 𝑥 ∈ 𝔪, then
certainly 𝑥 ∈ 𝔪 ∩ 𝑆{𝑥} . On the other hand, if 𝑥 ∉ 𝔪, then it is invertible modulo
𝔪, and so there exists an 𝑎 ∈ 𝐴 such that 1 − 𝑎𝑥 ∈ 𝔪 (hence also 𝔪 ∩ 𝑆{𝑥} ).
(b) Let 𝔪 be a maximal ideal, and let 𝔭 be a prime ideal contained in 𝔪; for every
𝑥 ∈ 𝔪 ∖ 𝔭, we have 𝔭 ∩ 𝑆{𝑥} = ∅. Indeed, if 𝑥 𝑛 (1 − 𝑎𝑥) ∈ 𝔭, then 1 − 𝑎𝑥 ∈ 𝔭 (as
𝑥 ∉ 𝔭); hence 1 − 𝑎𝑥 ∈ 𝔪, and so 1 ∈ 𝔪, which is a contradiction.

is shortened when passing from 𝐴 to 𝐴{𝑥} , while statement (b) shows that a maximal
Statement (a) shows that every chain of prime ideals beginning with a maximal ideal

(i.e., nonrefinable) chain of prime ideals of length 𝑛 is shortened only to 𝑛 − 1 when 𝑥 is

chosen appropriately. From this, the proposition follows. 2

Proposition 2.54. Let 𝐴 be an integral domain, and let 𝑘 be a subfield of 𝐴. Then

dim(𝐴) ≤ tr deg𝑘 𝐹(𝐴),

where 𝐹(𝐴) is the field of fractions of 𝐴.

Proof. If tr deg𝑘 𝐹(𝐴) = ∞, there is nothing to prove, and so we suppose that tr deg𝑘 𝐹(𝐴) =
𝑛 ∈ ℕ. We argue by induction on 𝑛. We can replace 𝑘 with its algebraic closure in 𝐴
without changing tr deg𝑘 𝐹(𝐴).
Let 𝑥 ∈ 𝐴. If 𝑥 ∉ 𝑘, then it is transcendental over 𝑘, and so

tr deg𝑘(𝑥) 𝐹(𝐴) = 𝑛 − 1

by 1.64; since 𝑘(𝑥) ⊂ 𝐴{𝑥} , this implies (by induction) that dim(𝐴{𝑥} ) ≤ 𝑛 − 1. If 𝑥 ∈ 𝑘,
then 0 = 1 − 𝑥 −1 𝑥 ∈ 𝑆{𝑥} , and so 𝐴{𝑥} = 0; again dim(𝐴{𝑥} ) ≤ 𝑛 − 1. We deduce from
2.53 that dim(𝐴) ≤ 𝑛. 2

Corollary 2.55. The polynomial ring 𝑘[𝑋1 , … , 𝑋𝑛 ] has Krull dimension 𝑛.

Proof. The existence of the sequence of prime ideals

(𝑋1 , … , 𝑋𝑛 ) ⊃ (𝑋1 , … , 𝑋𝑛−1 ) ⊃ ⋯ ⊃ (𝑋1 ) ⊃ (0)

shows that 𝑘[𝑋1 , … , 𝑋𝑛 ] has Krull dimension at least 𝑛, and 2.54 shows that it has Krull
dimension at most 𝑛. 2
56 2. Algebraic Sets

Corollary 2.56. Let 𝐴 be an integral domain and let 𝑘 be a subfield of 𝐴. If 𝐴 is finitely

generated as a 𝑘-algebra, then

tr deg𝑘 𝐹(𝐴) = dim(𝐴).

Proof. According to the Noether normalization theorem (2.45), 𝐴 is integral over a

polynomial subring 𝑘[𝑥1 , … , 𝑥𝑛 ]. Hence 𝐹(𝐴) is algebraic over 𝑘(𝑥1 , … , 𝑥𝑛 ), and so
𝑛 = tr deg𝑘 𝐹(𝐴). The going up theorem (1.54), implies that a chain of prime ideals
in 𝑘[𝑥1 , … , 𝑥𝑛 ] lifts to a chain in 𝐴, and so dim(𝐴) ≥ dim(𝑘[𝑥1 , … , 𝑥𝑛 ]) = 𝑛. Now 2.54
shows that dim(𝐴) = 𝑛.

Corollary 2.57. Let 𝑉 be an irreducible algebraic set. Then 𝑉 has dimension 𝑛 if and
2

only if there exists a finite surjective map 𝑉 → 𝔸𝑛 .

Proof. The 𝑑 in Theorem 2.44 is the dimension of 𝑉. 2

Aside 2.58. In linear algebra, we justify saying that a vector space 𝑉 has dimension 𝑛 by proving
that its elements are parametrized by 𝑛-tuples. It is not true in general that the points of an
algebraic set of dimension 𝑛 are parametrized by 𝑛-tuples. All we can say is Corollary 2.57.

Aside 2.59. The inequality in Proposition 2.54 may be strict. Let 𝐴 and 𝑘 be as in Corollary
2.56, so that dim 𝐴 = tr deg𝑘 𝐹(𝐴). When we replace 𝐴 with 𝐴𝔭 , where 𝔭 is a nonmaximal prime
ideal, the Krull dimension will drop but the field of fractions will be unchanged.

Notes. The short proof of 2.55 is based on that in Coquand and Lombardi, Amer. Math. Monthly
112 (2005), no. 9, 826–829.

Examples
Example 2.60. Let 𝑉 = 𝔸𝑛 . Then 𝑘(𝑉) = 𝑘(𝑋1 , … , 𝑋𝑛 ), which has transcendence basis
𝑋1 , … , 𝑋𝑛 over 𝑘, and so dim(𝑉) = 𝑛.

Example 2.61. If 𝑉 is a linear subspace of 𝑘 𝑛 (or a translate of a linear subspace), then

the dimension of 𝑉 as an algebraic set is the same as its dimension in the sense of linear
algebra — in fact, 𝑘[𝑉] is canonically isomorphic to 𝑘[𝑋𝑖1 , … , 𝑋𝑖𝑑 ], where the 𝑋𝑖𝑗 are the
“free” variables in the system of linear equations defining 𝑉.
More specifically, let 𝔠 be an ideal in 𝑘[𝑋1 , … , 𝑋𝑛 ] generated by linear forms 𝓁1 , … , 𝓁𝑟 ,
which we may assume to be linearly independent. Let 𝑋𝑖1 , … , 𝑋𝑖𝑛−𝑟 be such that

{𝓁1 , … , 𝓁𝑟 , 𝑋𝑖1 , … , 𝑋𝑖𝑛−𝑟 }

is a basis for the linear forms in 𝑋1 , … , 𝑋𝑛 . Then

𝑘[𝑋1 , … , 𝑋𝑛 ]∕𝔠 ≃ 𝑘[𝑋𝑖1 , … , 𝑋𝑖𝑛−𝑟 ].

This is obvious if the forms are 𝑋1 , … , 𝑋𝑟 . In the general case, because {𝑋1 , … , 𝑋𝑛 }
and {𝓁1 , … , 𝓁𝑟 , 𝑋𝑖1 , … , 𝑋𝑖𝑛−𝑟 } are both bases for the linear forms, each element of one set

𝑘[𝑋1 , … , 𝑋𝑛 ] = 𝑘[𝓁1 , … , 𝓁𝑟 , 𝑋𝑖1 , … , 𝑋𝑖𝑛−𝑟 ],

can be expressed as a linear combination of the elements of the other. Therefore,

𝑘[𝑋1 , … , 𝑋𝑛 ]∕𝔠 = 𝑘[𝓁1 , … , 𝓁𝑟 , 𝑋𝑖1 , … , 𝑋𝑖𝑛−𝑟 ]∕𝔠

and so

≃ 𝑘[𝑋𝑖1 , … , 𝑋𝑖𝑛−𝑟 ].
 m. Dimension 57

Example 2.62. If 𝑊 is a proper algebraic subset of an irreducible algebraic set 𝑉, then

dim(𝑊) < dim(𝑉) (see 2.50).

irreducible algebraic set has dimension 0 if and only if it consists of a single point.
Example 2.63. A point in an algebraic set is a closed irreducible subset. Therefore an

Example 2.64. A hypersurface in 𝔸𝑛 has dimension 𝑛 − 1. It suffices to prove this for

an irreducible hypersurface 𝐻. Such an 𝐻 is the zero set of an irreducible polynomial 𝑓
(see 2.29). Let

𝑘[𝑥1 , … , 𝑥𝑛 ] = 𝑘[𝑋1 , … , 𝑋𝑛 ]∕(𝑓), 𝑥𝑖 = 𝑋𝑖 + (𝑓),

and let 𝑘(𝑥1 , … , 𝑥𝑛 ) be the field of fractions of 𝑘[𝑥1 , … , 𝑥𝑛 ]. As 𝑓 is not the zero polyno-
mial, some 𝑋𝑖 , say, 𝑋𝑛 , occurs in it. Then 𝑋𝑛 occurs in every nonzero multiple of 𝑓, and
so no nonzero polynomial in 𝑋1 , … , 𝑋𝑛−1 belongs to (𝑓). This means that 𝑥1 , … , 𝑥𝑛−1
are algebraically independent. On the other hand, 𝑥𝑛 is algebraic over 𝑘(𝑥1 , … , 𝑥𝑛−1 ),
and so {𝑥1 , … , 𝑥𝑛−1 } is a transcendence basis for 𝑘(𝑥1 , … , 𝑥𝑛 ) over 𝑘. (Alternatively, use
2.57.)

Example 2.65. Let 𝐹(𝑋, 𝑌) and 𝐺(𝑋, 𝑌) be nonconstant polynomials with no common
factor. Then each irreducible component of 𝑉(𝐹) has dimension 1 (by 2.64), and so
𝑉(𝐹) ∩ 𝑉(𝐺) has dimension 0 (by 2.62). Therefore, 𝑉(𝐹) ∩ 𝑉(𝐺) is a finite set.

Proposition 2.66. Let 𝑊 be an algebraic subset of codimension 1 in an algebraic set 𝑉.

If 𝑘[𝑉] is a unique factorization domain, then 𝐼(𝑊) = (𝑓) for some 𝑓 ∈ 𝑘[𝑉].
⋂
Proof. Let 𝑊1 , … , 𝑊𝑠 be the irreducible components of 𝑊; then 𝐼(𝑊) = 𝐼(𝑊𝑖 ), and
so, if we can prove 𝐼(𝑊𝑖 ) = (𝑓𝑖 ), then 𝐼(𝑊) = (𝑓1 ⋯ 𝑓𝑟 ). This allows us to assume that
𝑊 is irreducible. Let 𝔭 = 𝐼(𝑊); it is a prime ideal, and it is not zero because otherwise
dim(𝑊) = dim(𝑉). Therefore it contains an irreducible polynomial 𝑓. From (1.33) we
know (𝑓) is prime. If (𝑓) ≠ 𝔭 , then we have

𝔭 ⊃ (𝑓) ⊃ (0) (distinct prime ideals)

and hence

𝑊 = 𝑉(𝔭) ⊂ 𝑉(𝑓) ⊂ 𝑉 (distinct irreducible closed subsets).

dim(𝑊) < dim(𝑉(𝑓)) < dim 𝑉,

But then (2.62)

which contradicts the hypothesis. 2

Corollary 2.67. The algebraic sets of codimension 1 in 𝔸𝑛 are exactly the hypersurfaces.

Proof. Combine 2.64 and 2.66. 2

Example 2.68. We classify the irreducible algebraic sets 𝑉 of 𝔸2 . If 𝑉 has dimension

2, then it equals 𝔸2 (by 2.62). If 𝑉 has dimension 1, then 𝑉 = 𝑉(𝑓), where 𝑓 is any
irreducible polynomial in 𝐼(𝑉) (see 2.66 and its proof). Finally, if 𝑉 has dimension zero,

𝑘[𝑋, 𝑌]:
then it is a point. Correspondingly, the following is a complete list of the prime ideals in

(0), (𝑓) with 𝑓 irreducible, (𝑋 − 𝑎, 𝑌 − 𝑏) with 𝑎, 𝑏 ∈ 𝑘.

58 2. Algebraic Sets

Exercises
2-1. Find 𝐼(𝑊), where 𝑊 = 𝑉(𝑋 2 , 𝑋𝑌 2 ). Check that it is the radical of (𝑋 2 , 𝑋𝑌 2 ).

2-2. Identify 𝑘 𝑚𝑛 with the set of 𝑚 × 𝑛 matrices, and let 𝑟 ∈ ℕ. Show that the set of
matrices with rank ≤ 𝑟 is an algebraic subset of 𝑘𝑚𝑛 .

2-3. Let 𝑉 = {(𝑡, 𝑡2 , … , 𝑡𝑛 ) ∣ 𝑡 ∈ 𝑘}. Show that 𝑉 is an algebraic subset of 𝑘 𝑛 , and that
𝑘[𝑉] ≈ 𝑘[𝑇] (polynomial ring in one symbol). (Assume char(𝑘) = 0.)

2-4. Let 𝑓1 , … , 𝑓𝑚 ∈ ℚ[𝑋1 , … , 𝑋𝑛 ]. If the 𝑓𝑖 have no common zero in ℂ, prove that

there exist 𝑔1 , … , 𝑔𝑚 ∈ ℚ[𝑋1 , … , 𝑋𝑛 ] such that 𝑓1 𝑔1 + ⋯ + 𝑓𝑚 𝑔𝑚 = 1. (Hint: linear
algebra).

2-5. Let 𝑘 ⊂ 𝐾 be algebraically closed fields, and let 𝔞 be an ideal in 𝑘[𝑋1 , … , 𝑋𝑛 ]. Show
that if 𝑓 ∈ 𝐾[𝑋1 , … , 𝑋𝑛 ] vanishes on 𝑉(𝔞), then it vanishes on 𝑉𝐾 (𝔞). Deduce that
the zero set 𝑉(𝔞) of 𝔞 in 𝑘 𝑛 is dense in the zero set 𝑉𝐾 (𝔞) of 𝔞 in 𝐾 𝑛 . [Hint: Choose
∑
a basis (𝑒𝑖 )𝑖∈𝐼 for 𝐾 as a 𝑘-vector space, and write 𝑓 = 𝑒𝑖 𝑓𝑖 (finite sum) with 𝑓𝑖 ∈
𝑘[𝑋1 , … , 𝑋𝑛 ].]

2-6. Let 𝐴 and 𝐵 be (not necessarily commutative) ℚ-algebras of finite dimension over
ℚ, and let ℚal be the algebraic closure of ℚ in ℂ. Show that if there exists a ℂ-algebra
homomorphism ℂ ⊗ℚ 𝐴 → ℂ ⊗ℚ 𝐵, then there exists a ℚal -algebra homomorphism
ℚal ⊗ℚ 𝐴 → ℚal ⊗ℚ 𝐵. (Hint: The proof takes only a few lines.)

2-7. Let 𝐴 be finite dimensional 𝑘-algebra, where 𝑘 is an infinite field, and let 𝑀 and 𝑁
be 𝐴-modules. Show that if 𝑘 al ⊗𝑘 𝑀 and 𝑘al ⊗𝑘 𝑁 are isomorphic 𝑘al ⊗𝑘 𝐴-modules,
then 𝑀 and 𝑁 are isomorphic 𝐴-modules.

2-8. Show that the subset {(𝑧, 𝑒𝑧 ) ∣ 𝑧 ∈ ℂ} is not an algebraic subset of ℂ2 .

Chapter 3

Affine Algebraic Varieties

In this chapter, we define the structure of a ringed space on an algebraic set. In this way,

algebraic set with no preferred embedding into 𝔸𝑛 . This is in preparation for Chapter 5,
we are led to the notion of an affine algebraic variety — roughly speaking, this is an

where we define an algebraic variety to be a ringed space that is a finite union of affine
algebraic varieties satisfying a natural separation axiom.

a. Sheaves
Let 𝑘 be a field (in sections a, b, and d, the field 𝑘 need not be algebraically closed).

Definition 3.1. Let 𝑉 be a topological space, and suppose that, for every open subset
𝑈 of 𝑉 we have a set 𝒪𝑉 (𝑈) of functions 𝑈 → 𝑘. Then 𝑈 ⇝ 𝒪𝑉 (𝑈) is a sheaf of
𝑘-algebras if, for every open subset 𝑈 of 𝑉,
(a) 𝒪𝑉 (𝑈) is a 𝑘-subalgebra of the algebra of all 𝑘-valued functions on 𝑈, i.e., 𝒪𝑉 (𝑈)
contains the constant functions and 𝑓 + 𝑔 and 𝑓𝑔 whenever it contains 𝑓 and 𝑔.
(b) the restriction of an 𝑓 in 𝒪𝑉 (𝑈) to any open subset 𝑈 ′ of 𝑈 is in 𝒪𝑉 (𝑈 ′ );
⋃
(c) a function 𝑓 ∶ 𝑈 → 𝑘 lies in 𝒪𝑉 (𝑈) if there exists an open covering 𝑈 = 𝑖∈𝐼
𝑈𝑖
of 𝑈 such that 𝑓|𝑈𝑖 lies in 𝒪𝑉 (𝑈𝑖 ) for all 𝑖 ∈ 𝐼.

Let 𝑈 be a union of open subsets 𝑈𝑖 . If the 𝑈𝑖 are disjoint, then (b) and (c) require
∏
𝒪𝑉 (𝑈) ≃ 𝒪𝑉 (𝑈𝑖 ).
that
𝑖

{ ∏ | }
𝒪𝑉 (𝑈𝑖 ) ||| 𝑓𝑖 |𝑈𝑖 ∩ 𝑈𝑗 = 𝑓𝑗 |𝑈𝑖 ∩ 𝑈𝑗 for all 𝑖, 𝑗 .
In the general case, they require that

𝒪𝑉 (𝑈) ≃ (𝑓𝑖 ) ∈
𝑖 |

Examples
3.2. Let 𝑉 be a topological space, and, for an open subset 𝑈 of 𝑉, let 𝒪𝑉 (𝑈) be the set
of all continuous real-valued functions on 𝑈. Then 𝒪𝑉 is a sheaf of ℝ-algebras.

3.3. A function 𝑓 ∶ 𝑈 → ℝ on an open subset 𝑈 of ℝ𝑛 is said to be smooth (infinitely

dition is local, i.e., a function on 𝑈 is smooth if and only if it is smooth on an open

differentiable) if its partial derivatives of all orders exist and are continuous. This con-

neighbourhood of each point of 𝑈. As constant functions and sums and products of

59
60 3. Affine Algebraic Varieties

smooth functions are smooth, for any open subset 𝑉 of ℝ𝑛 , the smooth functions on the
open subsets of 𝑉 form a sheaf of ℝ-algebras.

3.4. A function 𝑓 ∶ 𝑈 → ℂ on an open subset 𝑈 of ℂ𝑛 , is said to be analytic if it is

described by a convergent power series in a neighbourhood of each point of 𝑈. This
condition is local, and so, for any open subset 𝑉 of ℂ𝑛 , the analytic functions on the
open subsets of 𝑉 form a sheaf of ℂ-algebras.

3.5. Let 𝑉 be a topological space, and, for an open subset 𝑈 of 𝑉, let 𝒪𝑉 (𝑈) be the set
of all constant functions 𝑈 → 𝑘. If 𝑉 is not connected, then 𝒪𝑉 is not a sheaf: let 𝑈1
and 𝑈2 be disjoint open subsets of 𝑉, and let 𝑓 be the function on 𝑈1 ∪ 𝑈2 that takes the
constant value 0 on 𝑈1 and the constant value 1 on 𝑈2 ; then 𝑓 is not in 𝒪𝑉 (𝑈1 ∪ 𝑈2 ),
and so condition (3.1c) fails. When “constant” is replaced with “locally constant”, 𝒪𝑉
becomes a sheaf of 𝑘-algebras (in fact, the smallest such sheaf).

3.6. Let 𝑉 be a topological space, and, for an open subset 𝑈 of 𝑉, let 𝒪𝑉 (𝑈) be the set
of all functions 𝑈 → 𝑘. Then 𝒪𝑉 is a sheaf of 𝑘-algebras. All our sheaves of 𝑘-algebras
are subsheaves of this one.

b. Ringed spaces
A pair (𝑉, 𝒪𝑉 ) comprising a topological space 𝑉 and a sheaf of 𝑘-algebras on 𝑉 will be
called a 𝑘-ringed space (or just a ringed space when the 𝑘 is understood). For historical
reasons, we sometimes write 𝛤(𝑈, 𝒪𝑉 ) for 𝒪𝑉 (𝑈) and call its elements the sections of
𝒪𝑉 over 𝑈.
Let (𝑉, 𝒪𝑉 ) be a 𝑘-ringed space. For any open subset 𝑈 of 𝑉, the restriction of 𝒪𝑉 to
the collection of open subsets of 𝑈 is a sheaf of 𝑘-algebras on 𝑈.
Let (𝑉, 𝒪𝑉 ) be a 𝑘-ringed space, and let 𝑃 ∈ 𝑉. A germ of a function at 𝑃 is an
equivalence class of pairs (𝑈, 𝑓) with 𝑈 an open neighbourhood of 𝑃 and 𝑓 ∈ 𝒪𝑉 (𝑈);
two pairs (𝑈, 𝑓) and (𝑈 ′ , 𝑓 ′ ) are equivalent if the functions 𝑓 and 𝑓 ′ agree on some open
neighbourhood of 𝑃 in 𝑈 ∩ 𝑈 ′ . The germs of functions at 𝑃 form a 𝑘-algebra 𝒪𝑉,𝑃 , called
the stalk of 𝒪𝑉 at 𝑃. In other words, 𝒪𝑉,𝑃 is the direct limit,

𝒪𝑉,𝑃 = lim 𝒪𝑉 (𝑈),

,,→
over the open neighbourhoods 𝑈 of 𝑃. In the interesting cases, 𝒪𝑉,𝑃 is a local ring with
maximal ideal the set 𝔪𝑃 of germs zero at 𝑃. We often write 𝒪𝑃 for 𝒪𝑉,𝑃 .

Example 3.7. Let 𝒪𝑉 be the sheaf of analytic functions on 𝑉 = ℂ, and let 𝑐 ∈ ℂ. A

∑
power series 𝑛≥0 𝑎𝑛 (𝑧 − 𝑐)𝑛 , 𝑎𝑛 ∈ ℂ, is said to be convergent if it converges on some
open neighbourhood of 𝑐. The set of such power series is a ℂ-algebra, and I claim that it
is canonically isomorphic to the stalk 𝒪𝑉,𝑐 of 𝒪𝑉 at 𝑐.
To prove this, let 𝑓 be a analytic function on a neighbourhood 𝑈 of 𝑐. Then 𝑓 has
∑
a unique power series expansion 𝑓 = 𝑎𝑛 (𝑧 − 𝑐)𝑛 in some (possibly smaller) open
neighbourhood of 𝑐 (Cartan 1963, II, 6). Another analytic function 𝑓 ′ on a neighbour-
hood 𝑈 ′ of 𝑐 has the same power series expansion if and only if 𝑓 and 𝑓 ′ agree on some
neighbourhood of 𝑐 contained in 𝑈 ∩ 𝑈 ′ (ibid., I, 4.3). Thus we have a well-defined
injective map from the ring of germs of analytic functions at 𝑐 to the ring of convergent
power series, which is obviously surjective.
 c. The ringed space structure on an algebraic set 61

c. The ringed space structure on an algebraic set

Let 𝑉 be an algebraic subset of 𝑘 𝑛 . Recall that the basic open subsets of 𝑉 are those of

𝐷(ℎ) = {𝑄 ∣ ℎ(𝑄) ≠ 0}, ℎ ∈ 𝑘[𝑉].

the form
def

A pair 𝑔, ℎ ∈ 𝑘[𝑉] with ℎ ≠ 0 defines a function

𝑔(𝑄)
𝑄↦ ∶ 𝐷(ℎ) → 𝑘.
ℎ(𝑄)

Definition 3.8. Let 𝑈 be an open subset of 𝑉. A function 𝑓 ∶ 𝑈 → 𝑘 is regular at

We say that a function is regular if it is locally of this form.

𝑃 ∈ 𝑈 if there exist 𝑔, ℎ ∈ 𝑘[𝑉] with ℎ(𝑃) ≠ 0 such that 𝑓(𝑄) = 𝑔(𝑄)∕ℎ(𝑄) for all 𝑄
in some neighbourhood of 𝑃. A function 𝑓 ∶ 𝑈 → 𝑘 is regular if it is regular at every
𝑃 ∈ 𝑈.

Let 𝒪𝑉 (𝑈) denote the set of regular functions on an open subset 𝑈 of 𝑉.

Proposition 3.9. The map 𝑈 ⇝ 𝒪𝑉 (𝑈) is a sheaf of 𝑘-algebras on 𝑉.

a constant function is regular. Suppose that 𝑓 and 𝑓 ′ are regular on 𝑈, and let 𝑃 ∈ 𝑈.
Proof. The condition to be regular is local, and so we only have to check 3.1(a). Clearly,

By assumption, there exist 𝑔, 𝑔′ , ℎ, ℎ′ ∈ 𝑘[𝑉], with ℎ(𝑃) ≠ 0 ≠ ℎ′ (𝑃) such that 𝑓 and
𝑔 𝑔′
𝑓 ′ agree with and ′ respectively on a neighbourhood 𝑈 ′ of 𝑃. Then 𝑓 + 𝑓 ′ agrees
ℎ ℎ
𝑔ℎ′ + 𝑔′ ℎ ′ is regular at 𝑃. Similarly, 𝑓𝑓 ′ agrees with 𝑔𝑔 on
′
𝑈 ′ 𝑓 + 𝑓
ℎℎ′ ℎℎ′
𝑈 , and so is regular at 𝑃.
′
with on , and so
2

Lemma 3.10. Let 𝑔, ℎ ∈ 𝑘[𝑉] with ℎ ≠ 0. The function

𝑃 ↦ 𝑔(𝑃)∕ℎ(𝑃)𝑚 ∶ 𝐷(ℎ) → 𝑘

is zero if and only if and only if 𝑔ℎ = 0 in 𝑘[𝑉].

Proof. If 𝑔∕ℎ𝑚 is zero on 𝐷(ℎ), then 𝑔ℎ is zero on 𝑉 because ℎ is zero on the comple-
ment of 𝐷(ℎ). Therefore 𝑔ℎ is zero in 𝑘[𝑉]. Conversely, if 𝑔ℎ = 0, then 𝑔(𝑃)ℎ(𝑃) = 0 for
all 𝑃 ∈ 𝑉, and so 𝑔(𝑃) = 0 for all 𝑃 ∈ 𝐷(ℎ). 2

Let 𝑘[𝑉]ℎ denote the ring 𝑘[𝑉] with ℎ inverted (1.11). The lemma shows that the

𝑔 𝑔(𝑃)
↦ (𝑃 ↦ ) ∶ 𝑘[𝑉]ℎ → 𝒪𝑉 (𝐷(ℎ)),
map

ℎ 𝑚
ℎ(𝑃)𝑚

Proposition 3.11. The map 𝑘[𝑉]ℎ → 𝒪𝑉 (𝐷(ℎ)) is an isomorphism of 𝑘-algebras.

is well-defined and injective.

Proof. It remains to show that every regular function 𝑓 on 𝐷(ℎ) arises from an element
⋃
of 𝑘[𝑉]ℎ . By definition, there exists an open covering 𝐷(ℎ) = 𝑉𝑖 of 𝐷(ℎ) and elements
𝑔
𝑔𝑖 , ℎ𝑖 ∈ 𝑘[𝑉] with ℎ𝑖 nowhere zero on 𝑉𝑖 such that 𝑓|𝑉𝑖 = 𝑖 . We may assume that
ℎ𝑖
62 3. Affine Algebraic Varieties

each set 𝑉𝑖 is basic, say, 𝑉𝑖 = 𝐷(𝑎𝑖 ) for some 𝑎𝑖 ∈ 𝑘[𝑉]. By assumption 𝐷(𝑎𝑖 ) ⊂ 𝐷(ℎ𝑖 ),
and so 𝑎𝑖𝑁 = ℎ𝑖 𝑔𝑖′ for some 𝑁 ∈ ℕ and 𝑔𝑖′ ∈ 𝑘[𝑉] (see p. 49). On 𝐷(𝑎𝑖 ),

𝑔𝑖 𝑔𝑖 𝑔𝑖′ 𝑔𝑖 𝑔𝑖′
𝑓= = = 𝑁.
ℎ𝑖 ℎ𝑖 𝑔𝑖′ 𝑎𝑖

Note that 𝐷(𝑎𝑖𝑁 ) = 𝐷(𝑎𝑖 ). Therefore, after replacing 𝑔𝑖 with 𝑔𝑖 𝑔𝑖′ and ℎ𝑖 with 𝑎𝑖𝑁 , we can
assume that 𝑉𝑖 = 𝐷(ℎ𝑖 ).
⋃ 𝑔
We now have that 𝐷(ℎ) = 𝐷(ℎ𝑖 ) and that 𝑓|𝐷(ℎ𝑖 ) = 𝑖 . Because 𝐷(ℎ) is quasi-
ℎ𝑖
𝑔 𝑔𝑗
compact, we can assume that the covering is finite. As 𝑖 = on 𝐷(ℎ𝑖 ) ∩ 𝐷(ℎ𝑗 ) =
ℎ𝑖 ℎ𝑗
𝐷(ℎ𝑖 ℎ𝑗 ),
ℎ𝑖 ℎ𝑗 (𝑔𝑖 ℎ𝑗 − 𝑔𝑗 ℎ𝑖 ) = 0, i.e., ℎ𝑖 ℎ𝑗2 𝑔𝑖 = ℎ𝑖2 ℎ𝑗 𝑔𝑗
— this follows from Lemma 3.10 if ℎ𝑖 ℎ𝑗 ≠ 0 and is obvious otherwise. Because 𝐷(ℎ) =
(*)

⋃ ⋃
𝐷(ℎ𝑖 ) = 𝐷(ℎ𝑖2 ),
𝑉((ℎ)) = 𝑉((ℎ12 , … , ℎ𝑚
2
)),
and so ℎ lies in rad(ℎ12 , … , ℎ𝑚
2
): there exist 𝑎𝑖 ∈ 𝑘[𝑉] such that

∑
𝑚
ℎ𝑁 = 𝑎𝑖 ℎ𝑖2 .
𝑖=1
(**)

∑
𝑎𝑖 𝑔𝑖 ℎ𝑖
for some 𝑁. I claim that 𝑓 is the function on 𝐷(ℎ) defined by .
ℎ𝑁
Let 𝑃 be a point of 𝐷(ℎ). Then 𝑃 will be in one of the 𝐷(ℎ𝑖 ), say 𝐷(ℎ𝑗 ). We have the
following equalities in 𝑘[𝑉]:
∑
𝑚
∑
𝑚
ℎ𝑗2 𝑎𝑖 𝑔𝑖 ℎ𝑖 = 𝑎𝑖 𝑔𝑗 ℎ𝑖2 ℎ𝑗 = 𝑔𝑗 ℎ𝑗 ℎ𝑁 .
(∗) (∗∗)

𝑖=1 𝑖=1
𝑔𝑗
But 𝑓|𝐷(ℎ𝑗 ) = , i.e., 𝑓ℎ𝑗 and 𝑔𝑗 agree as functions on 𝐷(ℎ𝑗 ). Therefore we have the
ℎ𝑗
following equality of functions on 𝐷(ℎ𝑗 ):

∑
𝑚
ℎ𝑗2 𝑎𝑖 𝑔𝑖 ℎ𝑖 = 𝑓ℎ𝑗2 ℎ𝑁 .
𝑖=1

Since ℎ𝑗2 is never zero on 𝐷(ℎ𝑗 ), we can cancel it, to find that, as claimed, the function
∑
𝑓ℎ𝑁 on 𝐷(ℎ𝑗 ) equals that defined by 𝑎𝑖 𝑔𝑖 ℎ𝑖 .

On taking ℎ = 1 in the proposition, we see that the definition of a regular function

2

on 𝑉 in this section agrees with that in Section 2i.

Corollary 3.12. For any 𝑃 ∈ 𝑉, 𝒪𝑃 ≃ 𝑘[𝑉]𝔪𝑃 , where 𝔪𝑃 is the maximal ideal 𝐼(𝑃).

Proof. In the definition of the germs of a sheaf at 𝑃, it suffices to consider pairs (𝑓, 𝑈)
with 𝑈 lying in a some basis for the neighbourhoods of 𝑃, for example, the basis provided
by the basic open subsets. Therefore,

𝒪𝑃 ≃ lim 𝒪𝑉 (𝐷(ℎ)) ≃ lim 𝑘[𝑉]ℎ ≃ 𝑘[𝑉]𝔪𝑃 .

,,→ ,,→
3.11 1.23

ℎ(𝑃)≠0 ℎ∉𝔪𝑃 2
 c. The ringed space structure on an algebraic set 63

Notes
3.13. Let 𝑉 be an algebraic set and let 𝑃 be a point on 𝑉. Proposition 1.14 says that there
is a canonical one-to-one correspondence between the prime ideals of 𝑘[𝑉] contained in
𝔪𝑃 and the prime ideals of 𝒪𝑃 . In geometric terms, this says that there is a one-to-one
correspondence between the irreducible closed subsets of 𝑉 passing through 𝑃 and the
prime ideals in 𝒪𝑃 . The irreducible components of 𝑉 passing through 𝑃 correspond
to the minimal prime ideals in 𝒪𝑃 . The ideal 𝔭 corresponding to an irreducible closed
subset 𝑍 consists of the elements of 𝒪𝑃 that can be represented by a pair (𝑈, 𝑓) with
𝑓|𝑍∩𝑈 = 0.

3.14. If 𝑃 lies on a single irreducible component of 𝑉, then 𝒪𝑉,𝑃 is an integral domain.

As 𝒪𝑉,𝑃 depends only on (𝑈, 𝒪𝑉 |𝑈) for 𝑈 an open neighbourhood of 𝑃, in proving this,
we may suppose that 𝑉 itself is irreducible, in which case the statement follows from
3.12. On the other hand, if 𝑃 lies on more than one irreducible component of 𝑉, then
𝒪𝑃 contains more than one minimal prime ideal (by 3.13), and so (0) is not prime.

3.15. Let 𝑉 be an algebraic subset of 𝑘 𝑛 , and let 𝐴 = 𝑘[𝑉]. Propositions 2.37 and 3.11
allow us to describe (𝑉, 𝒪𝑉 ) purely in terms of 𝐴:
⋄ 𝑉 is the set of maximal ideals in 𝐴.
⋄ For each 𝑓 ∈ 𝐴, let 𝐷(𝑓) = {𝔪 ∣ 𝑓 ∉ 𝔪}; the topology on 𝑉 is that for which the
sets 𝐷(𝑓) form a base.
⋄ For 𝑓 ∈ 𝐴ℎ and 𝔪 ∈ 𝐷(ℎ), let 𝑓(𝔪) denote the image of 𝑓 in 𝐴ℎ ∕𝔪𝐴ℎ ≃ 𝑘; this
identifies 𝐴ℎ with a 𝑘-algebra of functions 𝐷(ℎ) → 𝑘, and 𝒪𝑉 is the unique sheaf
of 𝑘-valued functions such that 𝛤(𝐷(ℎ), 𝒪𝑉 ) = 𝐴ℎ for all ℎ ∈ 𝐴.

3.16. When 𝑉 is irreducible, all the rings attached to it can be identified with subrings
of its function field 𝑘(𝑉). For example,
{𝑔 | }
𝛤(𝐷(ℎ), 𝒪𝑉 ) = 𝑁 ∈ 𝑘(𝑉) ||| 𝑔 ∈ 𝑘[𝑉], 𝑁 ∈ ℕ
ℎ |
{𝑔 || }
𝒪𝑃 = ∈ 𝑘(𝑉) || ℎ(𝑃) ≠ 0
ℎ |
⋂
𝛤(𝑈, 𝒪𝑉 ) = 𝒪𝑃
⋂𝑃∈𝑈 ⋃
𝛤(𝑈, 𝒪𝑉 ) = 𝛤(𝐷(ℎ𝑖 ), 𝒪𝑉 ) if 𝑈 = 𝐷(ℎ𝑖 ).

Note that every element of 𝑘(𝑉) defines a function on some dense open subset of 𝑉.
Following tradition, we call the elements of 𝑘(𝑉) rational functions on 𝑉.1

Examples
3.17. The ring of regular functions on 𝔸𝑛 is 𝑘[𝑋1 , … , 𝑋𝑛 ]. For a nonzero polynomial
ℎ(𝑋1 , … , 𝑋𝑛 ), the ring of regular functions on 𝐷(ℎ) is

𝑔 |
{ ∈ 𝑘(𝑋1 , … , 𝑋𝑛 ) ||| 𝑔 ∈ 𝑘[𝑋1 , … , 𝑋𝑛 ], 𝑁 ∈ ℕ} .
ℎ𝑁 |
1
The terminology is similar to that of “meromorphic function”, which is also not a function on the
whole space.
64 3. Affine Algebraic Varieties

For a point 𝑃 = (𝑎1 , … , 𝑎𝑛 ), the local ring at 𝑃 is

𝑔 |
𝒪𝑃 = { ∈ 𝑘(𝑋1 , … , 𝑋𝑛 ) ||| ℎ(𝑃) ≠ 0}
ℎ |
= 𝑘[𝑋1 , … , 𝑋𝑛 ](𝑋1 −𝑎1 ,…,𝑋𝑛 −𝑎𝑛 ) ,

and its maximal ideal consists of those 𝑔∕ℎ with 𝑔(𝑃) = 0.

3.18. Let 𝑈 = 𝔸2 ∖ {(0, 0)}. It is an open subset of 𝔸2 , but it is not a basic open subset
because its complement {(0, 0)} has dimension 0, and so is not of the form 𝑉((𝑓)) (see
2.64). Since 𝑈 = 𝐷(𝑋) ∪ 𝐷(𝑌), the ring of regular functions on 𝑈 is

𝛤(𝑈, 𝒪𝔸2 ) = 𝑘[𝑋, 𝑌]𝑋 ∩ 𝑘[𝑋, 𝑌]𝑌

(intersection inside 𝑘(𝑋, 𝑌)). Thus, a regular function 𝑓 on 𝑈 can be expressed

𝑔(𝑋, 𝑌) ℎ(𝑋, 𝑌)
𝑓= = .
𝑋𝑁 𝑌𝑀
We may assume that 𝑋 ∤ 𝑔 and 𝑌 ∤ ℎ. On multiplying through by 𝑋 𝑁 𝑌 𝑀 , we find that

𝑔(𝑋, 𝑌)𝑌 𝑀 = ℎ(𝑋, 𝑌)𝑋 𝑁 .

Because 𝑋 does not divide the left hand side, it cannot divide the right hand side either,
and so 𝑁 = 0. Similarly, 𝑀 = 0, and so 𝑓 ∈ 𝑘[𝑋, 𝑌]. We have shown that every regular
function on 𝑈 extends uniquely to a regular function on 𝔸2 :

𝛤(𝑈, 𝒪𝔸2 ) = 𝑘[𝑋, 𝑌] = 𝛤(𝔸2 , 𝒪𝔸2 ).

d. Morphisms of ringed spaces

Let (𝑉, 𝒪𝑉 ) and (𝑊, 𝒪𝑊 ) be 𝑘-ringed spaces. A continuous map 𝜑 ∶ 𝑉 → 𝑊 is a mor-
phism of 𝑘-ringed spaces if

𝑓 ∈ 𝒪𝑊 (𝑈) ⇐⇒ 𝑓◦𝜑 ∈ 𝒪𝑉 (𝜑−1 𝑈)

for all open subsets 𝑈 of 𝑊. If 𝜑 ∶ 𝑉 → 𝑊 is a morphism of 𝑘-ringed spaces and 𝑈 ′ and

𝑈 are open subsets of 𝑉 and 𝑊 such that 𝜑(𝑈 ′ ) ⊂ 𝑈, then

𝑓 ↦ 𝑓◦𝜑 ∶ 𝒪𝑊 (𝑈) → 𝒪𝑉 (𝑈 ′ ),

is a homomorphism of 𝑘-algebras, and these homomorphisms are compatible with

For example, when (𝑉, 𝒪𝑉 ) is a 𝑘-ringed space and 𝑈 is an open subset of 𝑉, the
restriction to smaller open subsets.

inclusion 𝑈 → 𝑉 is a morphism of 𝑘-ringed spaces (𝑈, 𝒪𝑉 |𝑈) → (𝑉, 𝒪𝑉 ).

precisely, a morphism 𝜑 ∶ (𝑉, 𝒪𝑉 ) → (𝑊, 𝒪𝑊 ) induces a 𝑘-algebra homomorphism

A morphism of ringed spaces maps germs of functions to germs of functions. More

𝒪𝑊,𝜑(𝑃) → 𝒪𝑉,𝑃

for each 𝑃 ∈ 𝑉, which sends the germ represented by (𝑈, 𝑓) to the germ represented
by (𝜑−1 (𝑈), 𝑓◦𝜑). In the interesting cases, 𝒪𝑉,𝑃 is a local ring with maximal ideal 𝔪𝑃
consisting of the germs represented by pairs (𝑈, 𝑓) with 𝑓(𝑃) = 0. Then 𝒪𝑊,𝜑(𝑃) → 𝒪𝑉,𝑃
maps 𝔪𝜑(𝑃) into 𝔪𝑃 , i.e., it is a local homomorphism of local rings.
 e. Affine algebraic varieties 65

Examples
3.19. Let 𝑉 and 𝑊 be topological spaces endowed with their sheaves 𝒪𝑉 and 𝒪𝑊 of con-

every continous map 𝑉 → 𝑊 is a morphism of ℝ-ringed spaces (𝑉, 𝒪𝑉 ) → (𝑊, 𝒪𝑊 ).

tinuous real valued functions (3.2). As composites of continuous maps are continuous,

3.20. Let 𝑉 and 𝑊 be open subsets of ℝ𝑛 and ℝ𝑚 respectively, and let 𝑥𝑖 be the coordinate
function (𝑎1 , … , 𝑎𝑚 ) ↦ 𝑎𝑖 ∶ 𝑊 → ℝ. Recall from advanced calculus that a map

𝜑 ∶ 𝑉 → 𝑊 ⊂ ℝ𝑚

is said to be smooth if each of its component functions 𝜑𝑖 = 𝑥𝑖 ◦𝜑 ∶ 𝑉 → ℝ is smooth.

Endow 𝑉 and 𝑊 with their sheaves of smooth functions (3.3), and let 𝜑 ∶ 𝑉 → 𝑊 be
def

a continuous map. If 𝜑 is smooth, then 𝑓◦𝜑 is smooth for every smooth function 𝑓 on
an open subset of 𝑊, and so 𝜑 is a morphism of ℝ-ringed spaces. Conversely, if 𝜑 is a
morphism of ℝ-ringed spaces, then, in particular, the component functions 𝑥𝑖 ◦𝜑 are
smooth, and so 𝜑 is smooth.

3.21. Same as 3.20, but replace ℝ with ℂ and “smooth” with “analytic”. A continuous
map 𝜑 ∶ 𝑉 → 𝑊 is analytic if and only if it is a morphism of ℂ-ringed spaces.

e. Affine algebraic varieties

We have just seen that every algebraic set 𝑉 ⊂ 𝑘𝑛 has the structure of a 𝑘-ringed space
(𝑉, 𝒪𝑉 ). A 𝑘-ringed space isomorphic to one of this form is called an affine algebraic
variety over 𝑘. We often shorten (𝑉, 𝒪𝑉 ) to 𝑉.
Let (𝑉, 𝒪𝑉 ) and (𝑊, 𝒪𝑊 ) be affine algebraic varieties. A map 𝜑 ∶ 𝑉 → 𝑊 is regular
(or a morphism of affine algebraic varieties) if it is a morphism of 𝑘-ringed spaces.
With these definitions, the affine algebraic varieties become a category. We usually
shorten “affine algebraic variety” to “affine variety”.

algebraic set. We now regard 𝔸𝑛 as an affine algebraic variety. The affine varieties we
In particular, the regular functions define the structure of an affine variety on every

have constructed so far have all been embedded in 𝔸𝑛 . We now explain how to construct

An affine 𝑘-algebra is a reduced finitely generated 𝑘-algebra. For such an algebra

affine varieties with no preferred embedding.

𝐴, there exist 𝑥𝑖 ∈ 𝐴 such that 𝐴 = 𝑘[𝑥1 , … , 𝑥𝑛 ], and the kernel of the homomorphism

𝑋𝑖 ↦ 𝑥𝑖 ∶ 𝑘[𝑋1 , … , 𝑋𝑛 ] → 𝐴

is a radical ideal. Therefore 2.18 implies that the intersection of the maximal ideals in 𝐴 is
0. Moreover, 2.12 implies that, for every maximal ideal 𝔪 ⊂ 𝐴, the map 𝑘 → 𝐴 → 𝐴∕𝔪
is an isomorphism. Thus we can identify 𝐴∕𝔪 with 𝑘. For 𝑓 ∈ 𝐴, we write 𝑓(𝔪) for the
image of 𝑓 in 𝐴∕𝔪 = 𝑘, i.e., 𝑓(𝔪) = 𝑓 (mod 𝔪). This allows us to identify elements of
𝐴 with functions {maximal ideals in 𝐴} → 𝑘.
We attach a ringed space (𝑉, 𝒪𝑉 ) to 𝐴 by letting

𝑉 = {maximal ideals in 𝐴}.

For 𝑓 ∈ 𝐴, let
𝐷(𝑓) = {𝔪 ∣ 𝑓(𝔪) ≠ 0} = {𝔪 ∣ 𝑓 ∉ 𝔪}.
66 3. Affine Algebraic Varieties

Since 𝐷(𝑓𝑔) = 𝐷(𝑓) ∩ 𝐷(𝑔), there is a topology on 𝑉 for which the 𝐷(𝑓) form a base. A
pair of elements 𝑔, ℎ ∈ 𝐴, ℎ ≠ 0, defines a function

𝑔(𝔪)
𝔪↦ ∶ 𝐷(ℎ) → 𝑘.
ℎ(𝔪)

For 𝑈 an open subset of 𝑉, we define 𝒪𝑉 (𝑈) to be the set of functions 𝑓 ∶ 𝑈 → 𝑘 that

are of this form in some neighbourhood of each point of 𝑈.

Proposition 3.22. The pair (𝑉, 𝒪𝑉 ) is an affine algebraic variety with 𝛤(𝐷(ℎ), 𝒪𝑉 ) ≃ 𝐴ℎ
for each ℎ ∈ 𝐴 ∖ {0}.

Proof. Represent 𝐴 as a quotient 𝑘[𝑋1 , … , 𝑋𝑛 ]∕𝔞 = 𝑘[𝑥1 , … , 𝑥𝑛 ]. Then (𝑉, 𝒪𝑉 ) is

isomorphic to the 𝑘-ringed space attached to the algebraic set 𝑉(𝔞) (see 3.15). 2

We write spm(𝐴) for the topological space 𝑉, and Spm(𝐴) for the 𝑘-ringed space (𝑉, 𝒪𝑉 ).
Aside 3.23. We have shown that we can recover an algebraic set from its ring of regular functions
as the set of maximal ideals in the ring (equipped with the Zariski topology). It may seem strange
to be describing a topological space in terms of maximal ideals in a ring, but the analysts have

Gel’fand and Kolmogorov (1939) prove that if 𝑆1 and 𝑆2 are completely regular
been doing this for more than 80 years.

morphic as rings, then 𝑆1 and 𝑆2 are homeomorphic. The proof begins by showing
spaces, and if their rings of real-valued continuous functions are algebraically iso-

that there is a one-to-one correspondence between the maximal ideals in the ring
of functions and the points in the underlying space. The space is recovered by
introducing a suitable topology on the set of maximal ideals.
Allen Shields, Banach Algebras, 1939–1989, Math. Intelligencer, Vol 11, no. 3, p15.

f. The category of affine algebraic varieties

For each affine 𝑘-algebra 𝐴, we have an affine variety Spm(𝐴), and for each affine variety
(𝑉, 𝒪𝑉 ), we have an affine 𝑘-algebra 𝑘[𝑉] = 𝒪𝑉 (𝑉). We make this correspondence into
def

Let 𝛼 ∶ 𝐴 → 𝐵 be a homomorphism of affine 𝑘-algebras. For any ℎ ∈ 𝐴, 𝛼(ℎ) is

an anti-equivalence of categories.

invertible in 𝐵𝛼(ℎ) , and so the homomorphism 𝐴 → 𝐵 → 𝐵𝛼(ℎ) extends to a homomor-

𝑔 𝛼(𝑔)
↦ ∶ 𝐴ℎ → 𝐵𝛼(ℎ) .
phism

ℎ 𝑚
𝛼(ℎ)𝑚
If 𝔫 is a maximal ideal in 𝐵, then 𝔪 = 𝛼 −1 (𝔫) is a maximal ideal in 𝐴 because 𝐴∕𝔪 →
𝐵∕𝔫 = 𝑘 is an injective map of 𝑘-algebras which implies that 𝐴∕𝔪 = 𝑘. Thus 𝛼 defines
def

𝜑 ∶ spm 𝐵 → spm 𝐴, 𝜑(𝔫) = 𝛼−1 (𝔫) = 𝔪.

a map

For 𝔪 = 𝛼−1 (𝔫) = 𝜑(𝔫), we have a commutative diagram:

𝐴 → 𝐵
← 𝛼
←

𝐴∕𝔪 → 𝐵∕𝔫.
≃
→

←
 g. Explicit description of morphisms of affine varieties 67

Recall that the image of an element 𝑓 of 𝐴 in 𝐴∕𝔪 ≃ 𝑘 is denoted by 𝑓(𝔪). The

commutativity of the diagram means that, for 𝑓 ∈ 𝐴,

𝑓(𝜑(𝔫)) = 𝛼(𝑓)(𝔫), i.e., 𝑓◦𝜑 = 𝛼◦𝑓. (*)

Since 𝜑−1 𝐷(𝑓) = 𝐷(𝑓◦𝜑) (obviously), it follows from (*) that

𝜑−1 (𝐷(𝑓)) = 𝐷(𝛼(𝑓)),

and so 𝜑 is continuous.
Let 𝑓 be a regular function on 𝐷(ℎ), and write 𝑓 = 𝑔∕ℎ𝑚 , 𝑔 ∈ 𝐴. Then, from (*) we
see that 𝑓◦𝜑 is the function on 𝐷(𝛼(ℎ)) defined by 𝛼(𝑔)∕𝛼(ℎ)𝑚 . In particular, it is regular,
and so 𝑓 ↦ 𝑓◦𝜑 maps regular functions on 𝐷(ℎ) to regular functions on 𝐷(𝛼(ℎ)). It
follows that 𝑓 ↦ 𝑓◦𝜑 sends regular functions on any open subset of spm(𝐴) to regular
functions on the inverse image of the open subset. Thus 𝛼 defines a morphism of ringed
spaces Spm(𝐵) → Spm(𝐴).
Conversely, by definition, a morphism of 𝜑 ∶ (𝑉, 𝒪𝑉 ) → (𝑊, 𝒪𝑊 ) of affine algebraic
varieties defines a homomorphism of the associated affine 𝑘-algebras 𝑘[𝑊] → 𝑘[𝑉].

Proposition 3.24. For all affine algebras 𝐴 and 𝐵,

Since these maps are inverse, we have proved the following proposition.

Hom𝑘-algebra (𝐴, 𝐵) ,→ Mor(Spm(𝐵), Spm(𝐴));

≃

for all affine varieties 𝑉 and 𝑊,

Mor(𝑉, 𝑊) ,→ Hom𝑘-algebra (𝑘[𝑊], 𝑘[𝑉]).

≃

Proposition 3.25. The functor 𝐴 ⇝ Spm 𝐴 is a contravariant equivalence from the

In terms of categories, Proposition 3.24 says the following.

category of affine 𝑘-algebras to the category of affine algebraic varieties over 𝑘, with quasi-
inverse (𝑉, 𝒪𝑉 ) ⇝ 𝒪𝑉 (𝑉).

g. Explicit description of morphisms of affine varieties

Proposition 3.26. Let 𝑉 ⊂ 𝑘𝑚 and 𝑊 ⊂ 𝑘𝑛 be algebraic subsets. The following condi-
tions on a map 𝜑 ∶ 𝑉 → 𝑊 are equivalent:
(a) 𝜑 is a morphism of ringed spaces (𝑉, 𝒪𝑉 ) → (𝑊, 𝒪𝑊 );
(b) the components 𝜑1 , … , 𝜑𝑚 of 𝜑 are regular functions on 𝑉;
(c) 𝑓 ∈ 𝑘[𝑊] ⇐⇒ 𝑓◦𝜑 ∈ 𝑘[𝑉].

Proof. (a) ⇐⇒ (b). By definition 𝜑𝑖 = 𝑦𝑖 ◦𝜑, where 𝑦𝑖 is the coordinate function

(𝑏1 , … , 𝑏𝑛 ) ↦ 𝑏𝑖 ∶ 𝑊 → 𝑘.

(b) ⇐⇒ (c). The map 𝑓 ↦ 𝑓◦𝜑 is a 𝑘-algebra homomorphism from the ring of all
Hence this implication follows directly from the definition of a regular map (2.38).

functions 𝑊 → 𝑘 to the ring of all functions 𝑉 → 𝑘, and (b) says that the map sends the
coordinate functions 𝑦𝑖 on 𝑊 into 𝑘[𝑉]. Since the 𝑦𝑖 generate 𝑘[𝑊] as a 𝑘-algebra, this
implies that it sends 𝑘[𝑊] into 𝑘[𝑉].
68 3. Affine Algebraic Varieties

(c) ⇐⇒ (a). The map 𝑓 ↦ 𝑓◦𝜑 is a homomorphism 𝛼 ∶ 𝑘[𝑊] → 𝑘[𝑉]. It therefore

defines a map spm (𝑘[𝑉]) → spm (𝑘[𝑊]), and it remains to show that this coincides
with 𝜑 when we identify spm (𝑘[𝑉]) with 𝑉 and spm (𝑘[𝑊]) with 𝑊. Let 𝑃 ∈ 𝑉, let
𝑄 = 𝜑(𝑃), and let 𝔪𝑃 and 𝔪𝑄 be the ideals of elements of 𝑘[𝑉] and 𝑘[𝑊] that are zero
at 𝑃 and 𝑄 respectively. Then, for 𝑓 ∈ 𝑘[𝑊],

𝛼(𝑓) ∈ 𝔪𝑃 ⇐⇒ 𝑓(𝜑(𝑃)) = 0 ⇐⇒ 𝑓(𝑄) = 0 ⇐⇒ 𝑓 ∈ 𝔪𝑄 .

Therefore 𝛼−1 (𝔪𝑃 ) = 𝔪𝑄 , which is what we needed to show. 2

The equivalence of (a) and (b) means that 𝜑 ∶ 𝑉 → 𝑊 is a regular map of algebraic
sets in the sense of Chapter 2 if and only if it is a regular map of the associated affine
algebraic varieties.
Consider equations

𝑌1 = 𝑓1 (𝑋1 , … , 𝑋𝑚 )
…
𝑌𝑛 = 𝑓𝑛 (𝑋1 , … , 𝑋𝑚 ).

On the one hand, they define a regular map 𝜑 ∶ 𝔸𝑚 → 𝔸𝑛 , namely,

(𝑎1 , … , 𝑎𝑚 ) ↦ (𝑓1 (𝑎1 , … , 𝑎𝑚 ), … , 𝑓𝑛 (𝑎1 , … , 𝑎𝑚 )).

On the other hand, they define a homomorphism 𝛼 ∶ 𝑘[𝑌1 , … , 𝑌𝑛 ] → 𝑘[𝑋1 , … , 𝑋𝑚 ] of

𝑘-algebras, namely, that sending 𝑌𝑖 to 𝑓𝑖 (𝑋1 , … , 𝑋𝑚 ). This map coincides with 𝑔 ↦ 𝑔◦𝜑,

𝛼(𝑔)(𝑃) = 𝑔(… , 𝑓𝑖 (𝑃), …) = 𝑔(𝜑(𝑃)).

because

Now consider closed subsets 𝑉(𝔞) ⊂ 𝔸𝑚 and 𝑉(𝔟) ⊂ 𝔸𝑛 with 𝔞 and 𝔟 radical ideals. I
claim that 𝜑 maps 𝑉(𝔞) into 𝑉(𝔟) if and only if 𝛼(𝔟) ⊂ 𝔞. Indeed, suppose 𝜑(𝑉(𝔞)) ⊂
𝑉(𝔟), and let 𝑔 ∈ 𝔟; for 𝑄 ∈ 𝑉(𝔞),

𝛼(𝑔)(𝑄) = 𝑔(𝜑(𝑄)) = 0,

and so 𝛼(𝑔) ∈ 𝐼𝑉(𝔞) = 𝔞. Conversely, suppose 𝛼(𝔟) ⊂ 𝔞, and let 𝑃 ∈ 𝑉(𝔞); for 𝑓 ∈ 𝔟,

𝑓(𝜑(𝑃)) = 𝛼(𝑓)(𝑃) = 0,

and so 𝜑(𝑃) ∈ 𝑉(𝔟). When these conditions hold, 𝜑 is the morphism of affine varieties
𝑉(𝔞) → 𝑉(𝔟) corresponding to the homomorphism 𝑘[𝑌1 , … , 𝑌𝑛 ]∕𝔟 → 𝑘[𝑋1 , … , 𝑋𝑚 ]∕𝔞
defined by 𝛼.
We have shown that the regular maps

𝑉(𝔞) → 𝑉(𝔟)

are all of the form

𝑃 ↦ (𝑓1 (𝑃), … , 𝑓𝑛 (𝑃)), 𝑓𝑖 ∈ 𝑘[𝑋1 , … , 𝑋𝑚 ].

In particular, they all extend to regular maps 𝔸𝑚 → 𝔸𝑛 .

 g. Explicit description of morphisms of affine varieties 69

Examples of regular maps

3.27. Let 𝑅 be a 𝑘-algebra. To give a 𝑘-algebra homomorphism 𝑘[𝑋] → 𝑅 is the same
as giving an element of 𝑅 (the image of 𝑋 under the homomorphism):

Hom𝑘-algebra (𝑘[𝑋], 𝑅) ≃ 𝑅.

Mor(𝑉, 𝔸1 ) ≃ Hom𝑘-algebra (𝑘[𝑋], 𝑘[𝑉]) ≃ 𝑘[𝑉].

Therefore
3.24

In other words, the regular maps 𝑉 → 𝔸1 are simply the regular functions on 𝑉 (as we
would hope).

3.28. Let 𝔸0 = Spm 𝑘. Then 𝔸0 consists of a single point and 𝛤(𝔸0 , 𝒪𝔸0 ) = 𝑘. The
regular maps 𝔸0 → 𝑉, where 𝑉 is an affine variety, are just the maps of sets, so
Mor(𝔸0 , 𝑉) ≃ 𝑉. Alternatively,

Mor(𝔸0 , 𝑉) ≃ Hom𝑘-algebra (𝑘[𝑉], 𝑘) ≃ 𝑉,

where the second map sends 𝛼 ∶ 𝑘[𝑉] → 𝑘 to the point corresponding to the maximal
ideal Ker(𝛼).

3.29. Let 𝑘 be of characteristic ≠ 2.

(a) The regular map 𝑡 ↦ (𝑡 2 , 𝑡3 ) ∶ 𝔸1 → 𝔸2 is bijective onto its image,

𝑉∶ 𝑌2 = 𝑋3,

In view of 3.25, to prove this it suffices to show that 𝑡 ↦ (𝑡2 , 𝑡3 ) does not induce an
but it is not an isomorphism onto its image because the inverse map is not regular.

isomorphism on the rings of regular functions.

𝑡 (𝑡2 , 𝑡3 )

𝔸1 𝑌2 = 𝑋3

We have 𝑘[𝔸1 ] = 𝑘[𝑇] and 𝑘[𝑉] = 𝑘[𝑋, 𝑌]∕(𝑌 2 − 𝑋 3 ) = 𝑘[𝑥, 𝑦]. The map on rings

𝑥 ↦ 𝑇2 , 𝑦 ↦ 𝑇3 , 𝑘[𝑥, 𝑦] → 𝑘[𝑇],
is

which is injective, but its image is 𝑘[𝑇 2 , 𝑇 3 ] ≠ 𝑘[𝑇]. In fact, 𝑘[𝑥, 𝑦] is not integrally
closed: (𝑦∕𝑥)2 − 𝑥 = 0, and so (𝑦∕𝑥) is integral over 𝑘[𝑥, 𝑦], but 𝑦∕𝑥 ∉ 𝑘[𝑥, 𝑦] (it maps
to 𝑇 under the inclusion 𝑘(𝑥, 𝑦) → 𝑘(𝑇)).
70 3. Affine Algebraic Varieties

(b) The regular map 𝑡 ↦ (1 − 𝑡2 , 𝑡(1 − 𝑡2 )) ∶ 𝔸1 → 𝔸2 is bijective onto its image,

𝑉 ∶ 𝑌2 = 𝑋2 − 𝑋3
except that both ±1 map to (0, 0).

𝑡
(1 − 𝑡2 , 𝑡(1 − 𝑡2 ))

𝔸1 𝑌2 = 𝑋2 − 𝑋3

3.30. Let char(𝑘) = 𝑝 ≠ 0. The regular map

(𝑎1 , … , 𝑎𝑛 ) ↦ (𝑎1 , … , 𝑎𝑛 ) ∶ 𝔸𝑛 → 𝔸𝑛
𝑝 𝑝

𝑋𝑖 ↦ 𝑋𝑖 ∶ 𝑘[𝑋1 , … , 𝑋𝑛 ] → 𝑘[𝑋1 , … , 𝑋𝑛 ],
𝑝
is a bijection, but it is not an isomorphism because the corresponding map on rings,

This is the famous Frobenius map. Let 𝑘 be the algebraic closure of 𝔽𝑝 , and write 𝐹
is not surjective.

for the map. For each 𝑚 ≥ 1, there is a unique subfield 𝔽𝑝𝑚 of 𝑘 of degree 𝑚 over 𝔽𝑝 ,
and that its elements are the solutions of 𝑋 𝑝 = 𝑋 (FT, 4.23). The fixed points of 𝐹 𝑚 are
𝑚

precisely the points of 𝔸𝑛 with coordinates in 𝔽𝑝𝑚 . Let 𝑓(𝑋1 , … , 𝑋𝑛 ) be a polynomial

with coefficients in 𝔽𝑝𝑚 , say,
∑
𝑓= 𝑐𝑖1 ⋯𝑖𝑛 𝑋11 ⋯ 𝑋𝑛𝑛 , 𝑐𝑖1 ⋯𝑖𝑛 ∈ 𝔽𝑝𝑚 .
𝑖 𝑖

If 𝑓(𝑎1 , … , 𝑎𝑛 ) = 0, then
(∑ )𝑝𝑚 ∑
0= 𝑐𝛼 𝑎11 ⋯ 𝑎𝑛𝑛 = 𝑐𝛼 𝑎1 ⋯ 𝑎𝑛 ,
𝑖 𝑖 𝑝𝑚 𝑖1 𝑝 𝑚 𝑖𝑛

and so 𝑓(𝑎1 , … , 𝑎𝑛 ) = 0. Here we have used that the binomial theorem takes the
𝑝𝑚 𝑝𝑚

simple form (𝑋 + 𝑌)𝑝 = 𝑋 𝑝 + 𝑌 𝑝 in characteristic 𝑝. Thus 𝐹 𝑚 maps 𝑉(𝑓) into itself,

𝑚 𝑚 𝑚

𝑓(𝑋1 , … , 𝑋𝑛 ) = 0
and its fixed points are the solutions of

in 𝔽𝑝𝑚 .
Aside 3.31. In one of the most beautiful pieces of mathematics of the second half of the twentieth
century, Grothendieck defined a cohomology theory (étale cohomology) and proved a fixed point

with coordinates in 𝔽𝑝𝑚 as an alternating sum of traces of operators on finite-dimensional vector

formula that allowed him to express the number of solutions of a system of polynomial equations

spaces, and Deligne used this to obtain very precise estimates for the number of solutions. See
my article The Riemann hypothesis over finite fields: from Weil to the present day and my notes
Lectures on Étale Cohomology.
 h. Subvarieties 71

h. Subvarieties
Let 𝐴 be an affine 𝑘-algebra. For any ideal 𝔞 in 𝐴, we define

𝑉(𝔞) = {𝔪 ∈ spm(𝐴) ∣ 𝑓(𝔪) = 0 all 𝑓 ∈ 𝔞}

= {𝔪 ∈ spm(𝐴) ∣ 𝔞 ⊂ 𝔪}.

This is a closed subset of spm(𝐴), and every closed subset is of this form.
Now let 𝔞 be a radical ideal in 𝐴, so that 𝐴∕𝔞 is again an affine 𝑘-algebra. Corre-
sponding to the homomorphism 𝐴 → 𝐴∕𝔞, we get a regular map

Spm(𝐴∕𝔞) → Spm(𝐴).

The image is 𝑉(𝔞), and spm(𝐴∕𝔞) → 𝑉(𝔞) is a homeomorphism. Thus every closed
subset of spm(𝐴) has a natural ringed structure making it into an affine algebraic variety.
We call 𝑉(𝔞) with this structure a closed subvariety of 𝑉.

Proposition 3.32. Let (𝑉, 𝒪𝑉 ) be an affine variety and let ℎ be a nonzero element of 𝑘[𝑉].

Spm(𝐴ℎ ) ≃ (𝐷(ℎ), 𝒪𝑉 |𝐷(ℎ)).

Then

Proof. The map 𝐴 → 𝐴ℎ defines a morphism spm(𝐴ℎ ) → spm(𝐴), and induces an

≃
Spm(𝐴ℎ ) ,→ ((𝐷(ℎ), 𝒪𝑉 |𝐷(ℎ)) ⊂ Spm(𝐴).
isomorphism
2

In particular, (𝐷(ℎ), 𝒪𝑉 |𝐷(ℎ)) is an affine variety. If 𝑉 is a closed subvariety of 𝔸𝑛 ,

say, 𝑉 = 𝑉(𝔞) ⊂ 𝔸𝑛 , then

(𝑎1 , … , 𝑎𝑛 ) ↦ (𝑎1 , … , 𝑎𝑛 , ℎ(𝑎1 , … , 𝑎𝑛 )−1 ) ∶ 𝐷(ℎ) → 𝔸𝑛+1 ,

defines an isomorphism of 𝐷(ℎ) onto 𝑉(𝔞, 1 − ℎ𝑋𝑛+1 ), thereby realizing 𝐷(ℎ) as a closed
subvariety of 𝔸𝑛+1 . For example,

𝑎 ↦ (𝑎, 1∕𝑎) ∶ 𝔸1 ∖ {0} → 𝑉 ⊂ 𝔸2 ,

is an isomorphism from 𝐷(𝑋) = 𝔸1 ∖ {0} ⊂ 𝔸1 onto the curve 𝑋𝑌 = 1 in 𝔸2 ,

(𝑎, 1∕𝑎)

𝑋𝑌 = 1

𝑎 𝔸1
72 3. Affine Algebraic Varieties

By an open affine (subset) 𝑈 of an affine algebraic variety 𝑉, we mean an open

subset 𝑈 such that (𝑈, 𝒪𝑉 |𝑈) is an affine algebraic variety. The proposition says that,
for all nonzero ℎ ∈ 𝛤(𝑉, 𝒪𝑉 ), the open subset of 𝑉, where ℎ is nonzero is an open affine.
An open affine subset of an irreducible affine algebraic variety 𝑉 is irreducible with the
same dimension as 𝑉 (2.52).

affine variety 𝑉 are again affine varieties with their natural ringed structure, but this
Remark 3.33. We have seen that all closed subsets and all basic open subsets of an

is not true for all open subsets of 𝑉. For an open affine subset 𝑈, the natural map
𝑈 → spm 𝛤(𝑈, 𝒪𝑉 ) is a bijection. However, for

𝑈 = 𝔸2 ∖ {(0, 0)} = 𝐷(𝑋) ∪ 𝐷(𝑌) ⊂ 𝔸2 ,

def

we know that 𝛤(𝑈, 𝒪𝔸2 ) = 𝑘[𝑋, 𝑌] (see 3.18), but 𝑈 → spm 𝑘[𝑋, 𝑌] is not a bijection,
because the ideal (𝑋, 𝑌) is not in the image. Clearly (𝑈, 𝒪𝔸2 |𝑈) is a union of affine
algebraic varieties, and in Chapter 5 we shall recognize it as a (nonaffine) algebraic
variety.

i. Properties of the regular map Spm(𝛼)

Proposition 3.34. Let 𝛼 ∶ 𝐴 → 𝐵 be a homomorphism of affine 𝑘-algebras, and let

𝜑 ∶ Spm(𝐵) → Spm(𝐴)

(a) The image of 𝜑 is dense for the Zariski topology if and only if 𝛼 is injective.
be the corresponding morphism of affine varieties.

(b) The morphism 𝜑 is an isomorphism from Spm(𝐵) onto a closed subvariety of Spm(𝐴)
if and only if 𝛼 is surjective.

Proof. (a) Let 𝑓 ∈ 𝐴. If the image of 𝜑 is dense, then

𝑓◦𝜑 = 0 ⇐⇒ 𝑓 = 0.

On the other hand, if the image of 𝜑 is not dense, then the closure of its image is a proper
closed subset of Spm(𝐴), and so there is a nonzero function 𝑓 ∈ 𝐴 that is zero on it.
Then 𝑓◦𝜑 = 0. (See 2.40.)
(b) If 𝛼 is surjective, then it defines an isomorphism 𝐴∕𝔞 → 𝐵, where 𝔞 is the kernel
of 𝛼. This induces an isomorphism of Spm(𝐵) with its image in Spm(𝐴). The converse
follows from the description of the closed subvarieties of Spm(𝐴) in the last section. 2

A regular map 𝜑 ∶ 𝑉 → 𝑊 of affine algebraic varieties is said to be a dominant if its

image is dense in 𝑊. The proposition then says that
𝑓↦𝑓◦𝜑
𝜑 ∶ 𝑉 → 𝑊 is dominant ⇐⇒ 𝛤(𝑊, 𝒪𝑊 ) ,,,,,,,→ 𝛤(𝑉, 𝒪𝑉 ) is injective.

A regular map 𝜑 ∶ 𝑉 → 𝑊 of affine algebraic varieties is said to be a closed immer-

sion if it is an isomorphism of 𝑉 onto a closed subvariety of 𝑊. The proposition then
says that
𝑓↦𝑓◦𝜑
𝜑 ∶ 𝑉 → 𝑊 is a closed immersion ⇐⇒ 𝛤(𝑊, 𝒪𝑊 ) ,,,,,,,→ 𝛤(𝑉, 𝒪𝑉 ) is surjective.
 j. Affine space without coordinates 73

j. Affine space without coordinates

Let 𝐸 be a vector space over 𝑘 of dimension 𝑛. The set 𝔸(𝐸) of points of 𝐸 has a natural
structure of an algebraic variety: the choice of a basis for 𝐸 defines a bijection 𝔸(𝐸) → 𝔸𝑛 ,
and the inherited structure of an affine algebraic variety on 𝔸(𝐸) is independent of the

automorphism of 𝔸𝑛 ).
choice of the basis (because the bijections defined by two different bases differ by an

We now give an intrinsic definition of the affine variety 𝔸(𝐸). Let 𝑉 be a finite-
dimensional vector space over a field 𝑘. The tensor algebra of 𝑉 is
⨁
𝑇∗ 𝑉 = 𝑉 ⊗𝑖 = 𝑘 ⊕ 𝑉 ⊕ (𝑉 ⊗ 𝑉) ⊕ (𝑉 ⊗ 𝑉 ⊗ 𝑉) ⊕ ⋯
def

𝑖≥0

with multiplication defined by

(𝑣1 ⊗ ⋯ ⊗ 𝑣𝑖 ) ⋅ (𝑣1′ ⊗ ⋯ ⊗ 𝑣𝑗′ ) = 𝑣1 ⊗ ⋯ ⊗ 𝑣𝑖 ⊗ 𝑣1′ ⊗ ⋯ ⊗ 𝑣𝑗′ .

It is a noncommutative 𝑘-algebra, and the choice of a basis 𝑒1 , … , 𝑒𝑛 for 𝑉 defines an

𝑒1 ⋯ 𝑒𝑖 ↦ 𝑒1 ⊗ ⋯ ⊗ 𝑒𝑖 ∶ 𝑘{𝑒1 , … , 𝑒𝑛 } → 𝑇 ∗ (𝑉)
isomorphism

to 𝑇 ∗ 𝑉 from the 𝑘-algebra of noncommuting polynomials in the symbols 𝑒1 , … , 𝑒𝑛 .

The symmetric algebra 𝑆 ∗ (𝑉) of 𝑉 is defined to be the quotient of 𝑇 ∗ 𝑉 by the
two-sided ideal generated by the elements

𝑣 ⊗ 𝑤 − 𝑤 ⊗ 𝑣, 𝑣, 𝑤 ∈ 𝑉.

This algebra is generated as a 𝑘-algebra by commuting elements (namely, the elements

of 𝑉 = 𝑉 ⊗1 ), and so is commutative. The choice of a basis 𝑒1 , … , 𝑒𝑛 for 𝑉 defines an

𝑒1 ⋯ 𝑒𝑖 ↦ 𝑒1 ⊗ ⋯ ⊗ 𝑒𝑖 ∶ 𝑘[𝑒1 , … , 𝑒𝑛 ] → 𝑆 ∗ (𝑉)
isomorphism

to 𝑆 ∗ (𝑉) from the commutative polynomial ring in the symbols 𝑒1 , … , 𝑒𝑛 . This shows
that 𝑆 ∗ (𝑉) is an affine 𝑘-algebra. The pair (𝑆 ∗ (𝑉), 𝑖) comprising 𝑆 ∗ (𝑉) and the natural
𝑘-linear map 𝑖 ∶ 𝑉 → 𝑆 ∗ (𝑉) has the following universal property: every 𝑘-linear map
𝑉 → 𝐴 from 𝑉 into a 𝑘-algebra 𝐴 extends uniquely to a 𝑘-algebra homomorphism
𝑆 ∗ (𝑉) → 𝐴:
𝑉 → 𝑆 ∗ (𝑉)
← 𝑖

𝑘-linear ∃! 𝑘-algebra
←

→
←

(17)

𝐴.
→

As usual, this universal property determines the pair (𝑆 ∗ (𝑉), 𝑖) uniquely up to a unique

We now define 𝔸(𝐸) to be Spm(𝑆 ∗ (𝐸 ∨ )), where 𝐸 ∨ is the dual vector space. For an
isomorphism.

affine 𝑘-algebra 𝐴,

Mor(Spm(𝐴), 𝔸(𝐸)) ≃ Hom𝑘-algebra (𝑆 ∗ (𝐸 ∨ ), 𝐴) (3.24)

≃ Hom𝑘-linear (𝐸 ∨ , 𝐴) (17)
≃ 𝐸 ⊗𝑘 𝐴 (linear algebra).
74 3. Affine Algebraic Varieties

𝔸(𝐸)(𝑘) ≃ 𝐸.
In particular,

The choice of a basis 𝑒1 , … , 𝑒𝑛 for 𝐸 determines a (dual) basis 𝑓1 , … , 𝑓𝑛 of 𝐸 ∨ , and hence

an isomorphism of 𝑘-algebras 𝑘[𝑓1 , … , 𝑓𝑛 ] → 𝑆 ∗ (𝐸 ∨ ). The map of algebraic varieties
𝔸(𝐸) → 𝔸𝑛 defined by this homomorphism is the isomorphism

𝑒 ↦ (𝑓1 (𝑒), … , 𝑓𝑛 (𝑒)) ∶ 𝐸 → 𝑘𝑛 .

k. Birational equivalence
Recall that if 𝑉 is irreducible, then 𝑘[𝑉] is an integral domain, and we let 𝑘(𝑉) denote
its field of fractions. If 𝑈 is an open affine subvariety of 𝑉, then 𝑘[𝑉] ⊂ 𝑘[𝑈] ⊂ 𝑘(𝑉),
and so 𝑘(𝑉) is also the field of fractions of 𝑘[𝑈].

Definition 3.35. Two irreducible affine algebraic varieties over 𝑘 are birationally
equivalent if their function fields are isomorphic over 𝑘.

Proposition 3.36. Two irreducible affine varieties 𝑉 and 𝑊 are birationally equivalent if
and only if there exist open affine subvarieties 𝑈𝑉 and 𝑈𝑊 of 𝑉 and 𝑊 such that 𝑈𝑉 ≈ 𝑈𝑊 .

Proof. Let 𝑉 and 𝑊 be birationally equivalent irreducible affine varieties, and let
𝐴 = 𝑘[𝑉] and 𝐵 = 𝑘[𝑊]. We use the isomorphism to identify 𝑘(𝑉) and 𝑘(𝑊). This
allows us to suppose that 𝐴 and 𝐵 have a common field of fractions 𝐾. Let 𝑥1 , … , 𝑥𝑛
generate 𝐵 as 𝑘-algebra. As 𝐾 is the field of fractions of 𝐴, there exists a 𝑑 ∈ 𝐴 such that
𝑑𝑥𝑖 ∈ 𝐴 for all 𝑖; then 𝐵 ⊂ 𝐴𝑑 . The same argument shows that there exists an 𝑒 ∈ 𝐵
such that 𝐴𝑑 ⊂ 𝐵𝑒 . Now

𝐵 ⊂ 𝐴𝑑 ⊂ 𝐵𝑒 ⇐⇒ 𝐵𝑒 ⊂ 𝐴𝑑𝑒 ⊂ (𝐵𝑒 )𝑒 = 𝐵𝑒 ,

and so 𝐴𝑑𝑒 = 𝐵𝑒 . This shows that the open subvarieties 𝐷(𝑑𝑒) ⊂ 𝑉 and 𝐷(𝑒) ⊂ 𝑊 are
isomorphic. We have proved the “only if” part, and the “if” part is obvious. 2

Theorem 3.37. Every irreducible affine algebraic variety of dimension 𝑑 is birationally

equivalent to a hypersurface in 𝔸𝑑+1 .

Proof. Let 𝑉 be an irreducible variety of dimension 𝑑. According to Proposition 3.38

below, there exist rational functions 𝑥1 , … , 𝑥𝑑+1 on 𝑉 such that 𝑘(𝑉) = 𝑘(𝑥1 , … , 𝑥𝑑 , 𝑥𝑑+1 ).
Let 𝑓 ∈ 𝑘[𝑋1 , … , 𝑋𝑑+1 ] be an irreducible polynomial satisfied by the 𝑥𝑖 , and let 𝐻 be the
hypersurface 𝑓 = 0. Then 𝑘(𝑉) ≈ 𝑘(𝐻) and so 𝑉 and 𝐻 are birationally equivalent. 2

We review some definitions from FT, Chapter 2. Let 𝐹 be a field. A polynomial

𝑓 ∈ 𝐹[𝑋] is separable if it is nonzero and has no multiple roots. Equivalent condition:
gcd(𝑓, ) = 1. When 𝑓 is irreducible, this just says that ≠ 0 because deg < deg 𝑓.
𝑑𝑓 𝑑𝑓 𝑑𝑓
𝑑𝑋 𝑑𝑋 𝑑𝑋
An element of an algebraic extension 𝐸 of 𝐹 is separable over 𝐹 if its minimal polynomial
over 𝐹 is separable, and 𝐸 is separable over 𝐹 if all its elements are separable over 𝐹.

Proposition 3.38. Let 𝐹 be a perfect field and 𝐸 a finitely generated field extension of 𝐹
of transcendence degree 𝑑.
(a) If 𝐸 = 𝐹(𝑥1 , … , 𝑥𝑑+1 ), then, after the 𝑥𝑖 have been renumbered, {𝑥1 , … , 𝑥𝑑 } will be a
transcendence basis for 𝐸 over 𝐹 and 𝑥𝑑+1 will be separable over 𝐹(𝑥1 , … , 𝑥𝑑 ).
 l. Dimension 75

(b) There exist 𝑥1 , … , 𝑥𝑑+1 ∈ 𝐸 such that 𝐸 = 𝐹(𝑥1 , … , 𝑥𝑑+1 ).

Proof. First observe that, if 𝐹 has characteristic 𝑝 ≠ 0, then, because 𝐹 is perfect, every
polynomial in 𝑋1 , … , 𝑋𝑛 with coefficients in 𝐹 is a 𝑝th power in 𝐹[𝑋1 , … , 𝑋𝑛 ]:
𝑝 𝑝

∑ ∑ 𝑝
𝑎𝑖1 ⋯𝑖𝑛 𝑋11 … 𝑋𝑛𝑛 = ( 𝑎𝑖 𝑋1 … 𝑋𝑛𝑛 ) .
𝑖 𝑝 𝑖 𝑝 1∕𝑝 𝑖 𝑖
1 ⋯𝑖𝑛 1
(18)

(a) Suppose 𝐸 = 𝐹(𝑥1 , … , 𝑥𝑑+1 ). Then 𝑓(𝑥1 , … , 𝑥𝑑+1 ) = 0 for some nonzero 𝑓 ∈
𝐹[𝑋1 , … , 𝑋𝑑+1 ], which we may take to be irreducible. If all the polynomials 𝜕𝑓∕𝜕𝑋𝑖 are
zero, then 𝐹 has characteristic 𝑝 ≠ 0 and 𝑓 is a polynomial in 𝑋1 , … , 𝑋𝑑+1 , and hence not
𝑝 𝑝

irreducble. Thus some polynomial 𝜕𝑓∕𝜕𝑋𝑖 , which (after renumbering) we may suppose
to be 𝜕𝑓∕𝜕𝑋𝑑+1 , is not zero. Now 𝑥𝑑+1 is separable over 𝐹(𝑥1 , … , 𝑥𝑑 ).
(b) Let {𝑥1 , … , 𝑥𝑛 } be a generating set for 𝐸 over 𝐹 with 𝑛 minimal. We assume that
𝑛 > 𝑑+1 and obtain a contradiction. After renumbering, we may suppose that {𝑥1 , … , 𝑥𝑑 }
is a transcendence basis for 𝐸 over 𝐹 (1.63). On applying (a) to 𝐹(𝑥1 , … , 𝑥𝑑+1 ), we see
that we may also suppose that 𝑥𝑑+1 is separable over 𝐹(𝑥1 , … , 𝑥𝑑 ). As 𝑥𝑑+2 is algebraic
over 𝐹(𝑥1 , … , 𝑥𝑑 ), the primitive element theorem (FT, 5.1), shows that 𝐹(𝑥1 , … , 𝑥𝑑+2 ) =
𝐹(𝑥1 , … , 𝑥𝑑 , 𝑦) for some 𝑦 ∈ 𝐸. Now 𝐸 = 𝐹(𝑥1 , … , 𝑥𝑑 , 𝑦, 𝑥𝑑+3 , … , 𝑥𝑛 ), contradicting the
minimality of 𝑛. 2

In particular, there exists a separating transcendence basis for 𝐸∕𝐹 if 𝐸 is finitely

generated of finite transcendence degree over 𝐹 and 𝐹 is perfect.

l. Dimension
Definition 3.39. The dimension of an affine algebraic variety is the dimension of the
underlying topological space (2.48).

Thus, the dimension of an affine variety 𝑉 is the maximum length of a chain

𝑉0 ⊃ 𝑉1 ⊃ ⋯ ⊃ 𝑉𝑑

of distinct closed irreducible subvarieties. Later in this section, we shall see that it is the
length of every maximal chain of closed irreducible subvarieties.

Definition 3.40. A regular map 𝜑 ∶ 𝑊 → 𝑉 of affine algebraic varieties is finite if the

homomorphism 𝜑∗ ∶ 𝑘[𝑉] → 𝑘[𝑊] makes 𝑘[𝑊] a finite 𝑘[𝑉]-algebra.

Theorem 3.41. Let 𝑉 be an affine algebraic variety of dimension 𝑛. Then there exists a
finite map 𝑉 → 𝔸𝑛 .

Proof. This is a geometric restatement of the Noether normalization theorem (2.45).2

Question. Let 𝐴 be a finitely generated 𝑘-algebra. Assume that 𝐴 is an integral domain,

and let 𝑑 be the transcendence degree of its field of fractions 𝐹. Does there exist a
transcendence basis {𝑥1 , … , 𝑥𝑑 } for 𝐹 over 𝑘 such that
(a) 𝐴 is finite over 𝑘[𝑥1 , … , 𝑥𝑑 ], and
(b) 𝐹 is separable over 𝑘(𝑥1 , … , 𝑥𝑑 ).
According 2.45 and 3.38 there exist transcendence bases satisfying (a) or (b). Does there
always exist one satisfying both?
76 3. Affine Algebraic Varieties

Theorem 3.42. Let 𝑉 be an irreducible affine algebraic variety and 𝑓 a nonzero regular
function on 𝑉. If 𝑓 has a zero in 𝑉, then its zero set is of pure codimension 1.

Proof. Let 𝑑 = dim(𝑉). When 𝑉 = 𝔸𝑑 , we proved this in 2.64, and an argument of

Let 𝑍1 , … , 𝑍𝑛 be the irreducible components of 𝑉(𝑓). We have to show that dim 𝑍𝑖 =

Tate allows us to deduce the general case from the Noether normalization theorem.

dim 𝑉 − 1 for each 𝑖. There exists a point 𝑃 ∈ 𝑍𝑖 not contained in any other 𝑍𝑗 . As the
𝑍𝑗 are closed, there exists an open affine neighbourhood 𝑈 of 𝑃 in 𝑉 not intersecting any
𝑍𝑗 with 𝑗 ≠ 𝑖. Then 𝑍𝑖 ∩ 𝑈 is irreducible, and it is the zero set of the regular function
𝑓|𝑈. We may replace 𝑉 and 𝑓 with 𝑈 and 𝑓|𝑈, and so assume that 𝑉(𝑓) is irreducible.
As 𝑉(𝑓) is irreducible, the radical of (𝑓) is a prime ideal 𝔭 in 𝑘[𝑉]. According to
Theorem 2.45, there exists an inclusion 𝑘[𝔸𝑑 ] → 𝑘[𝑉] realizing 𝑘[𝑉] as a finite 𝑘[𝔸𝑑 ]-

𝑓0 = Nm𝑘(𝑉)∕𝑘(𝔸𝑑 ) 𝑓
algebra. The norm
def

of 𝑓 lies in 𝑘[𝔸𝑑 ] and 𝑓 divides 𝑓0 in 𝑘[𝑉] (by 1.45). Hence 𝑓0 ∈ (𝑓) ⊂ 𝔭, and so
rad(𝑓0 ) ⊂ 𝔭 ∩ 𝑘[𝔸𝑑 ]. We claim that, in fact,

rad(𝑓0 ) = 𝔭 ∩ 𝑘[𝔸𝑑 ].

Let 𝑔 ∈ 𝔭 ∩ 𝑘[𝔸𝑑 ]. Then 𝑔 ∈ 𝔭 = rad(𝑓), and so 𝑔𝑚 = 𝑓ℎ for some ℎ ∈ 𝑘[𝑉], 𝑚 ∈ ℕ.

def

Taking norms, we find that

𝑔𝑚𝑒 = Nm(𝑓ℎ) = 𝑓0 ⋅ Nm(ℎ) ∈ (𝑓0 ), 𝑒 = [𝑘(𝑉) ∶ 𝑘(𝔸𝑛 )],

and so 𝑔 ∈ rad(𝑓0 ), as claimed.

The inclusion 𝑘[𝔸𝑑 ] → 𝑘[𝑉] therefore induces an inclusion

𝑘[𝔸𝑑 ]∕ rad(𝑓0 ) → 𝑘[𝑉]∕𝔭.

This makes 𝑘[𝑉]∕𝔭 into a finite algebra over 𝑘[𝔸𝑑 ]∕ rad(𝑓0 ), and so the fields of fractions
of these two 𝑘-algebras have the same transcendence degree over 𝑘. Hence (2.56)

dim 𝑉(𝔭) = dim 𝑉(𝑓0 ).

Clearly 𝑓 ≠ 0 ⇒ 𝑓0 ≠ 0, and 𝑓0 ∈ 𝔭 ⇒ 𝑓0 is nonconstant. Therefore dim 𝑉(𝑓0 ) = 𝑑 − 1

by 2.64. 2

We can restate Theorem 3.42 as follows: let 𝑉 be a closed irreducible subvariety of

𝔸𝑛 and let 𝐹 ∈ 𝑘[𝑋1 , … , 𝑋𝑛 ]; then

⎧ 𝑉 if 𝐹 is identically zero on 𝑉
𝑉 ∩ 𝑉(𝐹) = ∅ if 𝐹 has no zeros on 𝑉
⎨
⎩
pure codimension 1 otherwise.

Corollary 3.43. Let 𝑉 be an irreducible affine variety, and let 𝑍 be a maximal proper
irreducible closed subset of 𝑉. Then dim(𝑍) = dim(𝑉) − 1.

Proof. Because 𝑍 is a proper closed subset of 𝑉, there exists a nonzero regular function
𝑓 on 𝑉 vanishing on 𝑍. Let 𝑉(𝑓) be the zero set of 𝑓 in 𝑉. Then 𝑍 ⊂ 𝑉(𝑓) ⊂ 𝑉, and 𝑍
must be an irreducible component of 𝑉(𝑓) for otherwise it would not be maximal in 𝑉.
Thus Theorem 3.42 shows that dim 𝑍 = dim 𝑉 − 1. 2
 l. Dimension 77

Corollary 3.44. Let 𝑉 be an irreducible affine variety. Every maximal chain

𝑉 = 𝑉0 ⊃ 𝑉1 ⊃ ⋯ ⊃ 𝑉𝑑 (19)

of distinct irreducible closed subsets of 𝑉 has length 𝑑 = dim(𝑉).

Proof. As the chain is maximal, the last set 𝑉𝑑 must be a point and each 𝑉𝑖 must be
maximal in 𝑉𝑖−1 , and so, from 3.43, we find that

dim 𝑉0 = dim 𝑉1 + 1 = dim 𝑉2 + 2 = ⋯ = dim 𝑉𝑑 + 𝑑 = 𝑑. 2

Corollary 3.45. Let 𝑉 be an irreducible affine variety, and let 𝑓1 , … , 𝑓𝑟 be regular func-
tions on 𝑉. Every irreducible component 𝑍 of 𝑉(𝑓1 , … 𝑓𝑟 ) has codimension at most 𝑟:

codim(𝑍) ≤ 𝑟.

Proof. We use induction on 𝑟. Because 𝑍 is an irreducible closed subset of 𝑉(𝑓1 , … , 𝑓𝑟−1 ),

it is contained in some irreducible component 𝑍 ′ of 𝑉(𝑓1 , … 𝑓𝑟−1 ). By induction, codim(𝑍 ′ ) ≤
𝑟 − 1. Also 𝑍 is an irreducible component of 𝑍 ′ ∩ 𝑉(𝑓𝑟 ) because

𝑍 ⊂ 𝑍 ′ ∩ 𝑉(𝑓𝑟 ) ⊂ 𝑉(𝑓1 , … , 𝑓𝑟 )

and 𝑍 is a maximal irreducible closed subset of 𝑉(𝑓1 , … , 𝑓𝑟 ). If 𝑓𝑟 vanishes identically

on 𝑍 ′ , then 𝑍 = 𝑍 ′ and codim(𝑍) = codim(𝑍 ′ ) ≤ 𝑟 − 1; otherwise, the theorem shows
that 𝑍 has codimension 1 in 𝑍 ′ , and codim(𝑍) = codim(𝑍 ′ ) + 1 ≤ 𝑟. 2

Example 3.46. In the setting of 3.45, the components of 𝑉(𝑓1 , … , 𝑓𝑟 ) need not all have
the same dimension, and it is possible for all of them to have codimension < 𝑟 without
any of the 𝑓𝑖 being redundant. For example, let 𝑉 be the 3-dimensional cone

𝑋1 𝑋4 − 𝑋2 𝑋3 = 0

in 𝔸4 . Then 𝑉(𝑋1 ) ∩ 𝑉 is the union of two planes:

𝑉(𝑋1 ) ∩ 𝑉 = {(0, 0, ∗, ∗)} ∪ {(0, ∗, 0, ∗)}.

Both of these have codimension 1 in 𝑉 (as required by 3.42). Similarly,

𝑉(𝑋2 ) ∩ 𝑉 = {(0, 0, ∗, ∗)} ∪ {(∗, 0, ∗, 0)},

and so 𝑍 = 𝑉(𝑋1 , 𝑋2 ) ∩ 𝑉 consists of a single plane {(0, 0, ∗, ∗)}. Thus, 𝑍 still has
codimension 1 in 𝑉, but it requires two equations to define it.
def

Proposition 3.47. Let 𝑍 be an irreducible closed subvariety of codimension 𝑟 in an affine

variety 𝑉. Then there exist regular functions 𝑓1 , … , 𝑓𝑟 on 𝑉 such that 𝑍 is an irreducible
component of 𝑉(𝑓1 , … , 𝑓𝑟 ) and all irreducible components of 𝑉(𝑓1 , … , 𝑓𝑟 ) have codimen-
sion 𝑟.

Proof. We know that there exists a chain of irreducible closed subsets

𝑉 ⊃ 𝑍1 ⊃ ⋯ ⊃ 𝑍𝑟 = 𝑍
78 3. Affine Algebraic Varieties

with codim 𝑍𝑖 = 𝑖. We shall show that there exist 𝑓1 , … , 𝑓𝑟 ∈ 𝑘[𝑉] such that, for all
𝑠 ≤ 𝑟, 𝑍𝑠 is an irreducible component of 𝑉(𝑓1 , … , 𝑓𝑠 ) and all irreducible components of
𝑉(𝑓1 , … , 𝑓𝑠 ) have codimension 𝑠.
We prove this by induction on 𝑠. For 𝑠 = 1, take any nonzero 𝑓1 ∈ 𝐼(𝑍1 ), and apply
Theorem 3.42. Suppose that 𝑓1 , … , 𝑓𝑠−1 have been chosen, and let 𝑌1 , 𝑌2 , … , 𝑌𝑚 , be the
irreducible components of 𝑉(𝑓1 , … , 𝑓𝑠−1 ), numbered so that 𝑍𝑠−1 = 𝑌1 . We seek an
element 𝑓𝑠 that is identically zero on 𝑍𝑠 but is not identically zero on any 𝑌𝑖 — for such
an 𝑓𝑠 , all irreducible components of 𝑌𝑖 ∩ 𝑉(𝑓𝑠 ) will have codimension 𝑠, and 𝑍𝑠 will be
an irreducible component of 𝑌1 ∩ 𝑉(𝑓𝑠 ). But no 𝑌𝑖 is contained in 𝑍𝑠 because 𝑍𝑠 has
smaller dimension than 𝑌𝑖 , and so 𝐼(𝑍𝑠 ) is not contained in any of the ideals 𝐼(𝑌
(⋃𝑖 ). Now)
the prime avoidance lemma (see below) tells us that there exist an 𝑓𝑠 ∈ 𝐼(𝑍𝑠 ) ∖ 𝑖 𝐼(𝑌𝑖 ) ,
and this is the function we want.

Lemma 3.48 (Prime Avoidance Lemma). If an ideal 𝔞 of a ring 𝐴 is not contained in

2

any of the prime ideals 𝔭1 , … , 𝔭𝑟 , then it is not contained in their union.

Proof. We may assume that none of the prime ideals 𝔭𝑖 is contained in a second,
because then we could omit it. For a fixed 𝑖, choose an 𝑓𝑖 ∈ 𝔞 ∖ 𝔭𝑖 and, for each 𝑗 ≠ 𝑖,
def ∏𝑟
choose an 𝑓𝑗 ∈ 𝔭𝑗 ∖ 𝔭𝑖 . Then ℎ𝑖 = 𝑗=1 𝑓𝑗 lies in each 𝔭𝑗 with 𝑗 ≠ 𝑖 and 𝔞, but not in
∑𝑟
𝔭𝑖 (here we use that 𝔭𝑖 is prime). The element 𝑖=1 ℎ𝑖 is therefore in 𝔞 but not in any 𝔭𝑖
(e.g, ℎ2 , … , ℎ𝑟 ∈ 𝔭1 but ℎ1 ∉ 𝔭1 ).

Example 3.49. When 𝑉 is an affine variety whose coordinate ring is a unique factor-
2

ization domain, every closed subset 𝑍 of codimension 1 is of the form 𝑉(𝑓) for some
𝑓 ∈ 𝑘[𝑉] (see 2.66). The condition that 𝑘[𝑉] be a unique factorization domain is
definitely needed. Again consider the cone,

𝑉 ∶ 𝑋1 𝑋4 − 𝑋2 𝑋3 = 0

in 𝔸4 and let 𝑍 and 𝑍 ′ be the planes

𝑍 = {(∗, 0, ∗, 0)} 𝑍 ′ = {(0, ∗, 0, ∗)}.

Then 𝑍 ∩ 𝑍 ′ = {(0, 0, 0, 0)}, which has codimension 2 in 𝑍 ′ . If 𝑍 = 𝑉(𝑓) for some regular
function 𝑓 on 𝑉, then 𝑉(𝑓|𝑍 ′ ) = {(0, … , 0)}, which has codimension 2, in violation of
3.42. Thus 𝑍 is not of the form 𝑉(𝑓), and so

𝑘[𝑋1 , 𝑋2 , 𝑋3 , 𝑋4 ]∕(𝑋1 𝑋4 − 𝑋2 𝑋3 )

is not a unique factorization domain.

Restatement in terms of affine 𝑘-algebras

Let 𝐴 be a finitely generated 𝑘-algebra. Assume that 𝐴 is an integral domain with field
of fractions 𝐹.
3.50. The Krull dimension of 𝐴, dim 𝐴 = tr deg𝑘 𝐹.

3.51 (Principal Ideal Theorem). Let 𝑓1 , … , 𝑓𝑟 be elements of 𝐴. If 𝔭 is minimal among

See Corollary 2.56.

the prime ideals containing (𝑓1 , … , 𝑓𝑟 ), then ht(𝔭) ≤ 𝑟. In particular, ht(𝔭) ≤ 𝑟 if 𝔭 can be
generated by 𝑟 elements.
 l. Dimension 79

3.52. Let 𝔭 be a prime ideal in 𝐴. If 𝔭 has height 𝑟, then there exist 𝑓1 , … , 𝑓𝑟 ∈ 𝐴 such that
See Corollary 3.45.

(a) 𝔭 is minimal among the prime ideals containing (𝑓1 , … , 𝑓𝑟 ), and

(b) every prime ideal minimal among those containing (𝑓1 , … , 𝑓𝑟 ) has height 𝑟.

3.53. If every prime ideal of height 1 in 𝐴 is principal, then 𝐴 is a unique factorization

See Proposition 3.47.

domain.

In order to prove this, it suffices to show that every irreducible element 𝑓 of 𝐴 is

prime (1.26). Let 𝔭 be minimal among the prime ideals containing (𝑓). According to
3.51, 𝔭 has height 1, and so it is principal, say, 𝔭 = (𝑔). As (𝑓) ⊂ (𝑔), 𝑓 = 𝑔𝑞 for some
𝑞 ∈ 𝐴. Because 𝑓 is irreducible, 𝑞 is a unit, and so (𝑓) = (𝑔) = 𝔭 — the element 𝑓 is

3.54. Let 𝔭 be a minimal nonzero prime ideal in 𝐴. Then ht(𝔭) = 1.

prime.

According to 3.51, ht(𝔭) ≤ 1 and 3.43 says that it equals 1.

3.55. Every maximal chain of distinct prime ideals

𝔭0 ⊃ ⋯ ⊃ 𝔭𝑑

in 𝐴 has length dim(𝐴). In particular, all maximal ideals in 𝐴 have height dim(𝐴).

3.56. Let 𝔮 ⊃ 𝔭 be prime ideals in 𝐴. Any two maximal chains of distinct prime ideals
See Corollary 3.44.

𝔮 = 𝔭𝑑 ⊃ 𝔭𝑑−1 ⊃ ⋯ ⊃ 𝔭0 = 𝔭

have the same length.

Indeed, if follows from 3.54 that their length is ht(𝔮) − ht(𝔭).

Remark 3.57. The first four statements (3.50, 3.51, 3.52, 3.53) hold for all noetherian
rings, but with more difficult proofs. The remaining statements (3.54, 3.55, 3.56) may

but here is an example of a ring that fails 3.55. Let 𝐴 = 𝑅[𝑋], where 𝑅 = 𝑘[𝑇](𝑇) is
fail. Rings satisfying 3.56 are said to be catenary. Noncatenary rings are hard to find,

a discrete valuation ring with maximal ideal (𝑇). The Krull dimension of 𝐴 is 2, and
def

(𝑇, 𝑋) ⊃ (𝑇) ⊃ (0) is a maximal chain of prime ideals, but the ideal (1 − 𝑇𝑋) is (a)
maximal and (b) has height 1. To see (a), note that

𝐴∕(1 − 𝑇𝑋) ≃ 𝑅𝑇 = 𝑘(𝑇).

1.13

To see (b), note that the ideal (1 − 𝑇𝑋) in 𝑘[𝑇, 𝑋] has height 1, and that 𝐴 is the ring of
fractions of 𝑘[𝑇, 𝑋] obtained by inverting the elements of 𝑘[𝑇] ∖ (𝑇).
80 3. Affine Algebraic Varieties

Aside 3.58. Proposition 3.47 shows that an irreducible curve 𝐶 in 𝔸3 is an irreducible component
of 𝑉(𝑓1 , 𝑓2 ) for some 𝑓1 , 𝑓2 ∈ 𝑘[𝑋, 𝑌, 𝑍]. In fact 𝐶 = 𝑉(𝑓1 , 𝑓2 , 𝑓3 ) for suitable polynomials
𝑓1 , 𝑓2 , and 𝑓3 (exercise). Apparently, it is not known whether two polynomials always suffice to
define an irreducible curve in 𝔸3 .
More generally, one can ask whether an irreducible curve 𝐶 in 𝔸𝑛 can be defined by 𝑛 − 1
polynomials, i.e., do there exist 𝑓1 , … , 𝑓𝑛−1 ∈ 𝑘[𝑋1 , … , 𝑋𝑛 ] such that

𝐶 = 𝑉(𝑓1 , … , 𝑓𝑛−1 )?

(a) 𝑘[𝐶] is a Dedekind domain (Mohan Kumar);

A positive answer to this question is known in the following cases:

(b) 𝑘 is of nonzero characteristic (Cowsik and Nori).

For proofs, see Chapter 10 of Ischebeck and Rao, Ideals and Reality, Springer 2005.

Exercises
3-1. Show that a map between affine varieties can be continuous for the Zariski topology
without being regular.

3-2. Let 𝑉 = Spm(𝐴), and let 𝑍 = Spm(𝐴∕𝔞) ⊂ Spm(𝐴). Show that a function 𝑓 on
an open subset 𝑈 of 𝑍 is regular if and only if, for each 𝑃 ∈ 𝑈, there exists a regular
function 𝑓 ′ on an open neighbourhood 𝑈 ′ of 𝑃 in 𝑉 such that 𝑓 and 𝑓 ′ agree on 𝑈 ′ ∩ 𝑈.

3-3. Find the image of the regular map

(𝑥, 𝑦) ↦ (𝑥, 𝑥𝑦) ∶ 𝔸2 → 𝔸2

and verify that it is neither open nor closed.

3-4. Show that the circle 𝑋 2 +𝑌 2 = 1 is isomorphic (as an affine variety) to the hyperbola
𝑋𝑌 = 1, but that neither is isomorphic to 𝔸1 . (Assume char(𝑘) ≠ 2.)

3-5. Let 𝐶 be the curve 𝑌 2 = 𝑋 2 + 𝑋 3 , and let 𝜑 be the regular map

𝑡 ↦ (𝑡2 − 1, 𝑡(𝑡 2 − 1)) ∶ 𝔸1 → 𝐶.

Is 𝜑 an isomorphism?
Chapter 4

Local Study

Geometry is the art of drawing correct conclusions

from incorrect figures. (La géométrie est l’art de
raisonner juste sur des figures fausses.)
Descartes
In this chapter, we examine the structure of an affine algebraic variety near a point.
We begin with the case of a plane curve, since the ideas in the general case are the same
but the proofs are more difficult.

a. Tangent spaces to plane curves

Consider the curve 𝑉 in the plane defined by a nonconstant polynomial 𝐹(𝑋, 𝑌),
𝑉 ∶ 𝐹(𝑋, 𝑌) = 0.
We assume that 𝐹(𝑋, 𝑌) has no multiple factors, so that (𝐹(𝑋, 𝑌)) is a radical ideal
and 𝐼(𝑉) = (𝐹(𝑋, 𝑌)). We can factor 𝐹 into a product of irreducible polynomials,
∏ ⋃
𝐹(𝑋, 𝑌) = 𝐹𝑖 (𝑋, 𝑌), and then 𝑉 = 𝑉(𝐹𝑖 ) expresses 𝑉 as a union of its irreducible
components (see 2.29). Each component 𝑉(𝐹𝑖 ) has dimension 1 (by 2.64) and so 𝑉 has
pure dimension 1.
If 𝐹(𝑋, 𝑌) itself is irreducible, then
𝑘[𝑉] = 𝑘[𝑋, 𝑌]∕(𝐹(𝑋, 𝑌)) = 𝑘[𝑥, 𝑦]
is an integral domain. Moreover, if 𝐹 ≠ 𝑋 − 𝑐, then 𝑥 is transcendental over 𝑘, and 𝑦
is algebraic over 𝑘(𝑥), and so 𝑥 is a transcendence basis for 𝑘(𝑉) over 𝑘. Similarly, if
𝐹 ≠ 𝑌 − 𝑐, then 𝑦 is a transcendence basis for 𝑘(𝑉) over 𝑘.
Let (𝑎, 𝑏) be a point on 𝑉. If we were doing calculus, we would say that the tangent
space at 𝑃 = (𝑎, 𝑏) is defined by the equation
𝜕𝐹 𝜕𝐹
(𝑎, 𝑏)(𝑋 − 𝑎) + (𝑎, 𝑏)(𝑌 − 𝑏) = 0.
𝜕𝑋 𝜕𝑌
(20)

This is the equation of a line unless both 𝜕𝐹 (𝑎, 𝑏) and 𝜕𝐹 (𝑎, 𝑏) are zero, in which case
𝜕𝑋 𝜕𝑌

We are not doing calculus, but we can define 𝜕 and 𝜕 by

𝜕𝑋 𝜕𝑌
it is the equation of a plane.

(∑ ) ∑ (∑ ) ∑
𝜕 𝜕
𝑎𝑖𝑗 𝑋 𝑖 𝑌 𝑗 = 𝑖𝑎𝑖𝑗 𝑋 𝑖−1 𝑌 𝑗 , 𝑎𝑖𝑗 𝑋 𝑖 𝑌 𝑗 = 𝑗𝑎𝑖𝑗 𝑋 𝑖 𝑌 𝑗−1 ,
𝜕𝑋 𝜕𝑌

81
82 4. Local Study

Definition 4.1. The tangent space 𝑇𝑃 𝑉 to 𝑉 at 𝑃 = (𝑎, 𝑏) is the algebraic subset

and make the same definition.

If 𝜕𝐹 (𝑎, 𝑏) and 𝜕𝐹 (𝑎, 𝑏) are not both zero, then 𝑇𝑃 (𝑉) is a line through (𝑎, 𝑏), and
defined by equation (20).

𝜕𝑋 𝜕𝑌
we say that 𝑃 is a nonsingular or smooth point of 𝑉. Otherwise, 𝑇𝑃 (𝑉) has dimension
2, and we say that 𝑃 is singular or multiple. The curve 𝑉 is said to be nonsingular or
||
As in advanced calculus, we often write 𝜕𝐹 ||| for 𝜕𝐹 (𝑎, 𝑏).
smooth if all its points are nonsingular.

𝜕𝑋 |(𝑎,𝑏) 𝜕𝑋

Examples

that char(𝑘) ≠ 2, 3.
For each of the following examples, the reader is invited to sketch the curve. Assume

4.2. 𝑋 𝑚 + 𝑌 𝑚 = 1. The tangent space at (𝑎, 𝑏) is given by the equation

𝑚𝑎𝑚−1 (𝑋 − 𝑎) + 𝑚𝑏𝑚−1 (𝑌 − 𝑏) = 0.
All points on the curve are nonsingular unless the characteristic of 𝑘 divides 𝑚, in which
case 𝑋 𝑚 + 𝑌 𝑚 − 1 has multiple factors,
𝑋 𝑚 + 𝑌 𝑚 − 1 = 𝑋 𝑚0 𝑝 + 𝑌 𝑚0 𝑝 − 1 = (𝑋 𝑚0 + 𝑌 𝑚0 − 1)𝑝 .
4.3. 𝑌 2 = 𝑋 3 (sketched in 4.12 below). The tangent space at (𝑎, 𝑏) is given by the

−3𝑎2 (𝑋 − 𝑎) + 2𝑏(𝑌 − 𝑏) = 0.
equation

The only singular point is (0, 0).

4.4. 𝑌 2 = 𝑋 2 (𝑋 + 1) (sketched in 4.10 below). Here again only (0, 0) is singular.

4.5. 𝑌 2 = 𝑋 3 + 𝑎𝑋 + 𝑏. In 2.2 we sketched two nonsingular examples of such curves,

and in 4.10 and 4.11 we sketch two singular examples. The singular points of the curve

𝑌 2 − 𝑋 3 − 𝑎𝑋 − 𝑏, 2𝑌, 3𝑋 2 + 𝑎,
are the common zeros of the polynomials

which consist of the points (𝑐, 0) with 𝑐 a common zero of

𝑋 3 + 𝑎𝑋 + 𝑏, 3𝑋 2 + 𝑎.
As 3𝑋 2 + 𝑎 is the derivative of 𝑋 3 + 𝑎𝑋 + 𝑏, we see that 𝑉 is singular if and only if
𝑋 3 + 𝑎𝑋 + 𝑏 has a multiple root.

4.6. 𝑉 = 𝑉(𝐹𝐺), where 𝐹𝐺 has no multiple factors (so 𝐹 and 𝐺 are coprime). Then
𝑉 = 𝑉(𝐹) ∪ 𝑉(𝐺), and a point (𝑎, 𝑏) is singular if and only if it is
⋄ a singular point of 𝑉(𝐹),
⋄ a singular point of 𝑉(𝐺), or
⋄ a point of 𝑉(𝐹) ∩ 𝑉(𝐺).

𝜕(𝐹𝐺) 𝜕𝐺 𝜕𝐹 𝜕(𝐹𝐺) 𝜕𝐺 𝜕𝐹
This follows immediately from the product rule:

=𝐹⋅ + ⋅ 𝐺, =𝐹⋅ + ⋅ 𝐺.
𝜕𝑋 𝜕𝑋 𝜕𝑋 𝜕𝑌 𝜕𝑌 𝜕𝑌
 b. Tangent cones to plane curves 83

The singular locus

Proposition 4.7. The nonsingular points of a plane curve form a dense open subset of
the curve.

Proof. Let 𝑉 = 𝑉(𝐹), where 𝐹 is a nonconstant polynomial in 𝑘[𝑋, 𝑌] without multiple

irreducible component of 𝑉, and so we may suppose that 𝑉 (hence 𝐹) is irreducible.

factors. It suffices to show that the nonsingular points form a dense open subset of each

It suffices to show that the set of singular points is a proper closed subset. It is closed

𝜕𝐹 𝜕𝐹
because it is the set of common zeros of the polynomials

𝐹, , .
𝜕𝑋 𝜕𝑌

It will be proper unless 𝜕𝐹∕𝜕𝑋 and 𝜕𝐹∕𝜕𝑌 are both identically zero on 𝑉, and hence
both multiples of 𝐹, but, as they have lower degree than 𝐹, this is impossible unless they
are both zero. Clearly 𝜕𝐹∕𝜕𝑋 = 0 if and only if 𝐹 is a polynomial in 𝑌 (𝑘 of characteristic
zero) or is a polynomial in 𝑋 𝑝 and 𝑌 (𝑘 of characteristic 𝑝). A similar remark applies
to 𝜕𝐹∕𝜕𝑌. Thus if 𝜕𝐹∕𝜕𝑋 and 𝜕𝐹∕𝜕𝑌 are both zero, then 𝐹 is constant (characteristic
zero) or a polynomial in 𝑋 𝑝 , 𝑌 𝑝 , and hence a 𝑝th power (characteristic 𝑝, see (18), p. 75).
These are contrary to our assumptions. 2

Thus the singular points form a proper closed subset, called the singular locus.
Aside 4.8. In common usage, “singular” means uncommon or extraordinary as in “he spoke
with singular shrewdness”. Thus the proposition says that singular points (mathematical sense)
are singular (usual sense).

b. Tangent cones to plane curves

A polynomial 𝐹(𝑋, 𝑌) can be written (uniquely) as a finite sum

𝐹 = 𝐹0 + 𝐹1 + 𝐹2 + ⋯ + 𝐹𝑚 + ⋯ (21)

with each 𝐹𝑚 a homogeneous polynomial of degree 𝑚. The term 𝐹1 will be denoted

𝐹𝓁 and called the linear form of 𝐹, and the first nonzero term on the right of (21) (the
nonzero homogeneous summand of 𝐹 of least degree) will be denoted 𝐹∗ and called the
leading form of 𝐹.
If 𝑃 = (0, 0) is on the curve 𝑉 defined by 𝐹, then 𝐹0 = 0 and (21) becomes

𝐹 = 𝑎𝑋 + 𝑏𝑌 + higher degree terms,

and the equation of the tangent space is

𝑎𝑋 + 𝑏𝑌 = 0.

Definition 4.9. Let 𝐹(𝑋, 𝑌) be a polynomial without square factors, and let 𝑉 be the
curve defined by 𝐹. If (0, 0) ∈ 𝑉, then the geometric tangent cone to 𝑉 at (0, 0) is the
zero set of 𝐹∗ . The tangent cone is the pair (𝑉(𝐹∗ ), 𝑘[𝑋, 𝑌]∕𝐹∗ ). To obtain the tangent
cone at any other point, translate to the origin, and then translate back.
84 4. Local Study

1. While the tangent space tells you whether a point is nonsingular or not, the tangent
Note that the geometric tangent cone at a point on a curve always has dimension

In general we can factor 𝐹∗ as

cone also gives you information on the nature of a singularity.

∏
𝐹∗ (𝑋, 𝑌) = 𝑐𝑋 𝑟0 (𝑌 − 𝑎𝑖 𝑋)𝑟𝑖 .
𝑖
∑
Then deg 𝐹∗ = 𝑟𝑖 is called the multiplicity of the singularity. A multiple point is
ordinary if its tangents are nonmultiple, i.e., 𝑟𝑖 = 1 all 𝑖. An ordinary double point is
def

called a node. There are many names for special types of singularities — see any book,
especially an old book, on algebraic curves.

Examples

of 𝑘 is 0.
The following examples are adapted from Walker 1950. We assume that the characteristic

4.10. 𝐹(𝑋, 𝑌) = 𝑋 3 +𝑋 2 −𝑌 2 . The tangent cone at (0, 0)

is defined by 𝑌 2 − 𝑋 2 . It is the pair of lines 𝑌 = ±𝑋, and
the singularity is a node.

4.11. 𝐹(𝑋, 𝑌) = 𝑋 3 − 𝑋 2 − 𝑌 2 . The origin is an isolated

cone is defined by 𝑌 2 + 𝑋 2 , which is the pair of lines

point of the real locus. It is again a node, but the tangent

𝑌 = ±𝑖𝑋. In this case, the real locus of the tangent cone

is just the point (0,0).

4.12. 𝐹(𝑋, 𝑌) = 𝑋 3 − 𝑌 2 . Here the origin is a cusp.

The tangent cone is defined by 𝑌 2 , which is the 𝑋-axis
(doubled).

4.13. 𝐹(𝑋, 𝑌) = 2𝑋 4 −3𝑋 2 𝑌+𝑌 2 −2𝑌 3 +𝑌 4 . The origin

tangent cone is again defined by 𝑌 2 .

is again a double point, but this time it is a tacnode. The

4.14. 𝐹(𝑋, 𝑌) = 𝑋 4 +𝑋 2 𝑌 2 −2𝑋 2 𝑌 −𝑋𝑌 2 −𝑌 2 . The ori-

cusp. The tangent cone is again defined by 𝑌 2 .

gin is again a double point, but this time it is a ramphoid
 c. The local ring at a point on a curve 85

4.15. 𝐹(𝑋, 𝑌) = (𝑋 2 + 𝑌 2 )2 + 3𝑋 2 𝑌 − 𝑌 3 . The origin is

√ by
3𝑋 2 𝑌−𝑌 3 , which is the triple of lines 𝑌 = 0, 𝑌 = ± 3𝑋.
an ordinary triple point. The tangent cone is defined

4.16. 𝐹(𝑋, 𝑌) = (𝑋 2 + 𝑌 2 )3 − 4𝑋 2 𝑌 2 . The origin has

multiplicity 4. The tangent cone is defined by 4𝑋 2 𝑌 2 ,
which is the union of the 𝑋 and 𝑌 axes, each doubled.

4.17. 𝐹(𝑋, 𝑌) = 𝑋 6 − 𝑋 2 𝑌 3 − 𝑌 5 . The tangent cone

is defined by 𝑋 2 𝑌 3 + 𝑌 5 , which consists of a triple line
𝑌 3 = 0 and a pair of lines 𝑌 = ±𝑖𝑋.

curve itself is singular. Another example of such a curve is 𝑌 3 + 2𝑋 2 𝑌 − 𝑋 4 = 0. This is singular

Aside 4.18. Note that the real locus of the algebraic curve in 4.17 is smooth even though the

at (0, 0), but its real locus is the image of ℝ under the analytic map 𝑡 ↦ (𝑡3 + 2𝑡, 𝑡(𝑡3 + 2)), which
is injective, proper, and immersive, and hence an embedding into ℝ2 with closed image. See
mo98366 for a discussion of this question.

c. The local ring at a point on a curve

Proposition 4.19. Let 𝑃 be a point on a plane curve 𝑉, and let 𝔪 be the correspond-
ing maximal ideal in 𝑘[𝑉]. If 𝑃 is nonsingular, then dim𝑘 (𝔪∕𝔪2 ) = 1, and otherwise
dim𝑘 (𝔪∕𝔪2 ) = 2.

Proof. Assume first that 𝑃 = (0, 0). Then 𝔪 = (𝑥, 𝑦) in 𝑘[𝑉] = 𝑘[𝑋, 𝑌]∕(𝐹(𝑋, 𝑌)) =
𝑘[𝑥, 𝑦]. Note that 𝔪2 = (𝑥 2 , 𝑥𝑦, 𝑦 2 ), and
𝔪∕𝔪2 = (𝑋, 𝑌)∕(𝔪2 + 𝐹(𝑋, 𝑌)) = (𝑋, 𝑌)∕(𝑋 2 , 𝑋𝑌, 𝑌 2 , 𝐹(𝑋, 𝑌)).
In this quotient, every element is represented by a linear polynomial 𝑐𝑥 + 𝑑𝑦, and the
only relation is 𝐹𝓁 (𝑥, 𝑦) = 0. Clearly dim𝑘 (𝔪∕𝔪2 ) = 1 if 𝐹𝓁 ≠ 0, and dim𝑘 (𝔪∕𝔪2 ) = 2
otherwise. Since 𝐹𝓁 = 0 is the equation of the tangent space, this proves the proposition

The same argument works for an arbitrary point (𝑎, 𝑏) except that one uses the
in this case.

variables 𝑋 ′ = 𝑋 − 𝑎 and 𝑌 ′ = 𝑌 − 𝑏; in essence, one translates the point to the origin.2

We explain what the condition dim𝑘 (𝔪∕𝔪2 ) = 1 means for the local ring 𝒪𝑃 =
𝑘[𝑉]𝔪 . Let 𝔫 be the maximal ideal 𝔪 ⋅ 𝑘[𝑉]𝔪 of this local ring. The map 𝔪 → 𝔫 induces
an isomorphism 𝔪∕𝔪2 → 𝔫∕𝔫2 (see 1.15), and so we have
𝑃 nonsingular ⇐⇒ dim𝑘 𝔪∕𝔪2 = 1 ⇐⇒ dim𝑘 𝔫∕𝔫2 = 1.
Nakayama’s lemma (1.3) shows that the last condition is equivalent to 𝔫 being a principal
ideal. As 𝒪𝑃 has Krull dimension one (2.64), for its maximal ideal to be principal means
that it is a regular local ring of dimension 1 (see 1.6). Thus, for a point 𝑃 on a curve,
𝑃 nonsingular ⇐⇒ 𝒪𝑃 regular.
86 4. Local Study

Proposition 4.20. Every regular local ring of dimension one is a principal ideal domain.

Proof. Let 𝐴 be such a ring, and let 𝔪 = (𝜋) be its maximal ideal. According to the
⋂
Krull intersection theorem (1.8), 𝑟≥0 𝔪𝑟 = (0). Let 𝔞 be a proper nonzero ideal in 𝐴. As
𝔞 is finitely generated, there exists an 𝑟 ∈ ℕ such that 𝔞 ⊂ 𝔪𝑟 but 𝔞 ⊄ 𝔪𝑟+1 . Therefore,
there exists an 𝑎 = 𝑐𝜋𝑟 ∈ 𝔞 such that 𝑎 ∉ 𝔪𝑟+1 . The second condition implies that
𝑐 ∉ 𝔪, and so it is a unit. Therefore (𝜋𝑟 ) = (𝑎) ⊂ 𝔞 ⊂ (𝜋𝑟 ), and so 𝔞 = (𝜋𝑟 ) = 𝔪𝑟 . We
have shown that all ideals in 𝐴 are principal.
By assumption, there exists a prime ideal 𝔭 properly contained in 𝔪. Then 𝐴∕𝔭 is
an integral domain. As 𝜋 ∉ 𝔭, it is not nilpotent in 𝐴∕𝔭, and hence not nilpotent in 𝐴.
Let 𝑎 and 𝑏 be nonzero elements of 𝐴. There exist 𝑟, 𝑠 ∈ ℕ such that 𝑎 ∈ 𝔪𝑟 ∖ 𝔪𝑟+1
and 𝑏 ∈ 𝔪𝑠 ∖𝔪𝑠+1 . Then 𝑎 = 𝑢𝜋𝑟 and 𝑏 = 𝑣𝜋𝑠 with 𝑢 and 𝑣 units, and 𝑎𝑏 = 𝑢𝑣𝜋𝑟+𝑠 ≠ 0.
Hence 𝐴 is an integral domain. 2

conditions on a principal ideal domain 𝐴 are equivalent:

It follows from the elementary theory of principal ideal domains that the following

(a) 𝐴 has exactly one nonzero prime ideal;

(b) 𝐴 has exactly one prime element up to associates;
(c) 𝐴 is local and is not a field.
A ring satisfying these conditions is called a discrete valuation ring.1

Theorem 4.21. A point 𝑃 on a plane algebraic curve is nonsingular if and only if 𝒪𝑃 is

regular, in which case it is a discrete valuation ring.

Proof. The statement summarizes the above discussion. 2

d. Tangent spaces to algebraic subsets of 𝔸𝑚

Before defining tangent spaces at points of an algebraic subset of 𝔸𝑚 we review some
terminology from linear algebra.

For a vector space 𝑘 𝑚 , let 𝑋𝑖 be the 𝑖th coordinate function 𝐚 ↦ 𝑎𝑖 . Thus 𝑋1 , … , 𝑋𝑚 is

Linear algebra

∑
the dual basis to the standard basis for 𝑘 𝑚 . A linear form 𝑎𝑖 𝑋𝑖 can be regarded as an
element of the dual vector space (𝑘 𝑚 )∨ = Hom𝑘-linear (𝑘 𝑚 , 𝑘).
Let 𝐴 = (𝑎𝑖𝑗 ) be an 𝑛 × 𝑚 matrix. It defines a linear map 𝛼 ∶ 𝑘 𝑚 → 𝑘 𝑛 , by
def

∑𝑚
⎛ 𝑗=1 𝑎1𝑗 𝑎𝑗 ⎞
⎛ 𝑎1 ⎞ ⎛ 𝑎1 ⎞
⎜ ⎟.
⎜ ⋮ ⎟ ↦ 𝐴 ⎜ ⋮ ⎟ = ∑𝑚 ⋮
⎜ ⎟
⎝𝑎𝑚 ⎠ ⎝𝑎𝑚 ⎠ 𝑎𝑛𝑗 𝑎𝑗
⎝ 𝑗=1 ⎠
Let 𝐴 be a discrete valuation ring and 𝜋 a prime element of 𝐴. For an element 𝑎 of the field of fractions
𝐹 of 𝐴, let
1

𝑟 if 𝑎 = 𝑐𝜋𝑟 with 𝑐 ∈ 𝐴× ,
𝑣(𝑎) = {
∞ if 𝑎 = 0.
Then 𝑣 is an additive valuation on 𝐹 with discrete value group 𝑣(𝐹) = ℤ ⊔ {∞} and valuation ring 𝐴 =
{𝑎 ∈ 𝐹 ∣ 𝑣(𝑎) ≥ 0}. The discrete valuation rings are exactly the valuation rings of discrete valuations, which
explains the name.
 d. Tangent spaces to algebraic subsets of 𝔸𝑚 87

Let 𝑌1 , … , 𝑌𝑛 for the coordinate functions on 𝑘 𝑛 . Then

∑
𝑚
𝑌𝑖 ◦𝛼 = 𝑎𝑖𝑗 𝑋𝑗 .
𝑗=1

This says that the 𝑖th coordinate of 𝛼(𝐚) is

∑
𝑚
∑
𝑚
𝑎𝑖𝑗 (𝑋𝑗 𝐚) = 𝑎𝑖𝑗 𝑎𝑗 .
𝑗=1 𝑗=1

Definition 4.22. Let 𝑉 ⊂ 𝑘𝑚 be an algebraic subset of 𝑘 𝑚 , and let 𝔞 = 𝐼(𝑉). The

Tangent spaces

tangent space 𝑇𝐚 (𝑉) to 𝑉 at a point 𝐚 = (𝑎1 , … , 𝑎𝑚 ) of 𝑉 is the subspace of the vector

space with origin 𝐚 cut out by the linear equations
∑ 𝜕𝐹 ||||
(𝑋 − 𝑎𝑖 ) = 0, 𝐹 ∈ 𝔞.
𝑚

𝑖=1
𝜕𝑋𝑖 |||𝐚 𝑖
(22)

In other words, 𝑇𝐚 (𝔸𝑚 ) is the vector space of dimension 𝑚 with origin 𝐚, and 𝑇𝐚 (𝑉)
is the subspace of 𝑇𝐚 (𝔸𝑚 ) defined by the equations (22).
Write (𝑑𝑋𝑖 )𝐚 for (𝑋𝑖 − 𝑎𝑖 ); then the (𝑑𝑋𝑖 )𝐚 form a basis for the dual vector space
𝑇𝐚 (𝔸𝑚 )∨ to 𝑇𝐚 (𝔸𝑚 ) — in fact, they are the coordinate functions on 𝑇𝐚 (𝔸𝑚 )∨ . As in
advanced calculus, we define the differential of a polynomial 𝐹 ∈ 𝑘[𝑋1 , … , 𝑋𝑚 ] at 𝐚 by

∑ 𝜕𝐹 ||||
𝑚
(𝑑𝐹)𝐚 = (𝑑𝑋𝑖 )𝐚 .
the equation:

𝜕𝑋𝑖 |||𝐚
𝑖=1

It is again a linear form on 𝑇𝐚 (𝔸𝑚 ). In terms of differentials, 𝑇𝐚 (𝑉) is the subspace of

𝑇𝐚 (𝔸𝑚 ) defined by the equations:

(𝑑𝐹)𝐚 = 0, 𝐹 ∈ 𝔞. (23)

I claim that, in (22) and (23), it suffices to take the 𝐹 to lie in a generating subset for 𝔞.
∑
The product rule for differentiation shows that if 𝐺 = 𝑗 𝐻𝑗 𝐹𝑗 , then
∑
(𝑑𝐺)𝐚 = 𝐻𝑗 (𝐚) ⋅ (𝑑𝐹𝑗 )𝐚 + 𝐹𝑗 (𝐚) ⋅ (𝑑𝐻𝑗 )𝐚 .
𝑗

If 𝐹1 , … , 𝐹𝑟 generate 𝔞 and 𝐚 ∈ 𝑉(𝔞), so that 𝐹𝑗 (𝐚) = 0 for all 𝑗, then this equation
∑
(𝑑𝐺)𝐚 = 𝐻𝑗 (𝐚) ⋅ (𝑑𝐹𝑗 )𝐚 .
becomes

Thus (𝑑𝐹1 )𝐚 , … , (𝑑𝐹𝑟 )𝐚 generate the 𝑘-vector space {(𝑑𝐹)𝐚 ∣ 𝐹 ∈ 𝔞}.

Definition 4.23. A point 𝐚 on an algebraic set 𝑉 is nonsingular (or smooth) if it lies

on a single irreducible component 𝑊 of 𝑉 and the dimension of the tangent space at 𝐚
is equal to the dimension of 𝑊; otherwise it is singular (or multiple).
88 4. Local Study

Thus, a point 𝐚 on an irreducible algebraic set 𝑉 is nonsingular if and only if

dim 𝑇𝐚 (𝑉) = dim 𝑉. As in the case of plane curves, a point on 𝑉 is nonsingular if
and only if it lies on a single irreducible component of 𝑉 and is nonsingular on it.
Let 𝔞 = (𝐹1 , … , 𝐹𝑟 ), and let

⎛ 𝜕𝐹1 𝜕𝐹1 ⎞
, …,
𝜕𝐹𝑖 ⎜ 𝜕𝑋1 𝜕𝑋𝑚 ⎟
𝐽 = Jac(𝐹1 , … , 𝐹𝑟 ) = ( )=⎜ ⋮ ⋮ ⎟.
𝜕𝑋𝑗 𝜕𝐹𝑟 𝜕𝐹𝑟
⎜ , …, ⎟
⎝ 𝜕𝑋1 𝜕𝑋𝑚 ⎠
Then the equations defining 𝑇𝐚 (𝑉) as a subspace of 𝑇𝐚 (𝔸𝑚 ) have matrix 𝐽(𝐚). Therefore,

dim𝑘 𝑇𝐚 (𝑉) = 𝑚 − rank 𝐽(𝐚),

linear algebra shows that

and so 𝐚 is nonsingular if and only if the rank of Jac(𝐹1 , … , 𝐹𝑟 )(𝐚) is equal to 𝑚 − dim(𝑉).
For example, if 𝑉 is a hypersurface, say, 𝐼(𝑉) = (𝐹(𝑋1 , … , 𝑋𝑚 )), then
𝜕𝐹 𝜕𝐹
Jac(𝐹)(𝐚) = ( (𝐚), … , (𝐚)) ,
𝜕𝑋1 𝜕𝑋𝑚

and 𝐚 is nonsingular if and only if not all of the partial derivatives 𝜕𝐹 vanish at 𝐚.
𝜕𝑋𝑖
We can regard 𝐽 as a matrix of regular functions on 𝑉. For each 𝑟,
{𝐚 ∈ 𝑉 ∣ rank 𝐽(𝐚) ≤ 𝑟}
is closed in 𝑉, because it is the set where certain determinants vanish. Therefore, there
is an open subset 𝑈 of 𝑉 on which rank 𝐽(𝐚) attains its maximum value, and the rank
jumps on closed subsets. Later (4.37) we shall show that the maximum value of rank 𝐽(𝐚)
is 𝑚 − dim 𝑉, and so the nonsingular points of 𝑉 form a nonempty open subset of 𝑉.

e. The differential of a regular map

𝜑 ∶ 𝔸𝑚 → 𝔸𝑛 , 𝐚 ↦ (𝑃1 (𝑎1 , … , 𝑎𝑚 ), … , 𝑃𝑛 (𝑎1 , … , 𝑎𝑚 )).

Consider a regular map

We can think of 𝜑 as being given by the equations

𝑌𝑖 = 𝑃𝑖 (𝑋1 , … , 𝑋𝑚 ), 𝑖 = 1, … , 𝑛.
It corresponds to the map of rings 𝜑∗ ∶ 𝑘[𝑌1 , … , 𝑌𝑛 ] → 𝑘[𝑋1 , … , 𝑋𝑚 ] sending 𝑌𝑖 to
𝑃𝑖 (𝑋1 , … , 𝑋𝑚 ), 𝑖 = 1, … , 𝑛.
Let 𝐚 ∈ 𝔸𝑚 , and let 𝐛 = 𝜑(𝐚). Define (𝑑𝜑)𝐚 ∶ 𝑇𝐚 (𝔸𝑚 ) → 𝑇𝐛 (𝔸𝑛 ) to be the map such
∑ 𝜕𝑃𝑖 |||
(𝑑𝑌𝑖 )𝐛 ◦(𝑑𝜑)𝐚 = || (𝑑𝑋𝑗 )𝐚 ,
𝜕𝑋𝑗 |||𝐚
that

i.e., relative to the standard bases, (𝑑𝜑)𝐚 is the map with matrix

⎛ 𝜕𝑃1 𝜕𝑃1 ⎞
(𝐚), … , (𝐚)
𝜕𝑋1 𝜕𝑋𝑚
def ⎜ ⎟
Jac(𝑃1 , … , 𝑃𝑛 )(𝐚) = ⎜ ⋮ ⋮ ⎟.
⎜ 𝜕𝑃𝑛 𝜕𝑃𝑛 ⎟
(𝐚), … , (𝐚)
⎝ 𝜕𝑋1 𝜕𝑋𝑚 ⎠
 f. Tangent spaces to affine algebraic varieties 89

For example, suppose 𝐚 = (0, … , 0) and 𝐛 = (0, … , 0), so that 𝑇𝐚 (𝔸𝑚 ) = 𝑘 𝑚 and 𝑇𝐛 (𝔸𝑛 ) =
𝑘 𝑛 . If
∑
𝑚
𝑃𝑖 (𝑋1 , … , 𝑋𝑚 ) = 𝑐𝑖𝑗 𝑋𝑗 + (higher terms), 𝑖 = 1, … , 𝑛,
𝑗=1
∑
then 𝑌𝑖 ◦(𝑑𝜑)𝐚 = 𝑗 𝑐𝑖𝑗 𝑋𝑗 , and the map on tangent spaces is given by the matrix (𝑐𝑖𝑗 ),
i.e., it is simply 𝐭 ↦ (𝑐𝑖𝑗 )𝐭.
Let 𝐹 ∈ 𝑘[𝑋1 , … , 𝑋𝑚 ]. We can regard 𝐹 as a regular map 𝔸𝑚 → 𝔸1 , whose differen-
tial will be a linear map

(𝑑𝐹)𝐚 ∶ 𝑇𝐚 (𝔸𝑚 ) → 𝑇𝐛 (𝔸1 ), 𝐛 = 𝐹(𝐚).

When we identify 𝑇𝐛 (𝔸1 ) with 𝑘, we obtain an identification of the differential of 𝐹

(regarded as a regular map) with the differential of 𝐹 (regarded as a regular function).

Lemma 4.24. Let 𝜑 ∶ 𝔸𝑚 → 𝔸𝑛 be a regular map. If 𝜑 maps 𝑉 = 𝑉(𝔞) ⊂ 𝑘 𝑚 into

𝑊 = 𝑉(𝔟) ⊂ 𝑘𝑛 , then (𝑑𝜑)𝐚 maps 𝑇𝐚 (𝑉) into 𝑇𝐛 (𝑊), 𝐛 = 𝜑(𝐚).
def
def

𝑓 ∈ 𝔟 ⇒ 𝑓◦𝜑 ∈ 𝔞,
Proof. We are given that

and have to prove that

𝑓 ∈ 𝔟 ⇒ (𝑑𝑓)𝐛 ◦(𝑑𝜑)𝐚 is zero on 𝑇𝐚 (𝑉).

𝜕𝑓 ∑ 𝜕𝑓 𝜕𝑌𝑗
The chain rule holds in our situation:
𝑛
= , 𝑌𝑗 = 𝑃𝑗 (𝑋1 , … , 𝑋𝑚 ), 𝑓 = 𝑓(𝑌1 , … , 𝑌𝑛 ).
𝜕𝑋𝑖 𝑗=1 𝜕𝑌𝑗 𝜕𝑋𝑖

If 𝜑 is the map given by the equations

𝑌𝑗 = 𝑃𝑗 (𝑋1 , … , 𝑋𝑚 ), 𝑗 = 1, … , 𝑛,

then the chain rule implies that

𝑑(𝑓◦𝜑)𝐚 = (𝑑𝑓)𝐛 ◦(𝑑𝜑)𝐚 , 𝐛 = 𝜑(𝐚).

Let 𝐭 ∈ 𝑇𝐚 (𝑉); then

(𝑑𝑓)𝐛 ◦(𝑑𝜑)𝐚 (𝐭) = 𝑑(𝑓◦𝜑)𝐚 (𝐭),
which is zero if 𝑓 ∈ 𝔟 because then 𝑓◦𝜑 ∈ 𝔞. Thus (𝑑𝜑)𝐚 (𝐭) ∈ 𝑇𝐛 (𝑊).

We therefore get a map (𝑑𝜑)𝐚 ∶ 𝑇𝐚 (𝑉) → 𝑇𝐛 (𝑊). The usual rules from advanced
2

calculus hold. For example,

(𝑑𝜓)𝐛 ◦(𝑑𝜑)𝐚 = 𝑑(𝜓◦𝜑)𝐚 , 𝐛 = 𝜑(𝐚).

f. Tangent spaces to affine algebraic varieties

of the algebraic set into 𝔸𝑛 . In this section, we give an intrinsic definition of the tangent
The definition 4.22 of the tangent space at a point on an algebraic set uses the embedding

space at a point of an affine algebraic variety that makes clear that it depends only on
the local ring at the point.
90 4. Local Study

Dual numbers
For an affine algebraic variety 𝑉 and a 𝑘-algebra 𝑅 (not necessarily affine), we let

𝑉(𝑅) = Hom(𝑘[𝑉], 𝑅) (homomorphisms of 𝑘-algebras).

For example, if 𝑉 ⊂ 𝔸𝑛 , then

𝑉(𝑅) = {(𝑎1 , … , 𝑎𝑛 ) ∈ 𝑅𝑛 ∣ 𝑓(𝑎1 , … , 𝑎𝑛 ) = 0 for all 𝑓 ∈ 𝐼(𝑉)} .

A homomorphism 𝑅 → 𝑆 of 𝑘-algebras defines a map 𝑉(𝑅) → 𝑉(𝑆) of sets.

The ring of dual numbers is 𝑘[𝜀] = 𝑘[𝑋]∕(𝑋 2 ), where 𝜀 = 𝑋 + (𝑋 2 ). Thus 𝑘[𝜀] =
𝑘 ⊕ 𝑘𝜀 as a 𝑘-vector space, and
def

(𝑎 + 𝑏𝜀)(𝑐 + 𝑑𝜀) = 𝑎𝑐 + (𝑎𝑑 + 𝑏𝑐)𝜀, 𝑎, 𝑏, 𝑐, 𝑑 ∈ 𝑘.

Note that there is a 𝑘-algebra homomorphism 𝜀 ↦ 0 ∶ 𝑘[𝜀] → 𝑘.

Definition 4.25. Let 𝑃 be a point on an affine algebraic variety 𝑉 over 𝑘. The tangent
space to 𝑉 at 𝑃 is

𝑇𝑃 (𝑉) = {𝑃′ ∈ 𝑉(𝑘[𝜀]) ∣ 𝑃′ ↦ 𝑃 under 𝑉(𝑘[𝜀]) → 𝑉(𝑘)}.

def

Thus an element of 𝑇𝑃 (𝑉) is a homomorphism of 𝑘-algebras 𝛼 ∶ 𝑘[𝑉] → 𝑘[𝜀] whose

𝜀↦0
composite with 𝑘[𝜀] ,→ 𝑘 is the point 𝑃. To say that 𝑘[𝑉] → 𝑘 is the point 𝑃 means that
its kernel is 𝔪𝑃 , and so 𝔪𝑃 = 𝛼−1 ((𝜀)).

Proposition 4.26. Let 𝑉 be an algebraic subset of 𝔸𝑛 , and let 𝑉 ′ = (𝑉, 𝒪𝑉 ) be 𝑉 equipped

with its canonical structure of an affine algebraic variety. Let 𝑃 ∈ 𝑉. Then

𝑇𝑃 (𝑉) (as defined in 4.22) ≃ 𝑇𝑃 (𝑉 ′ ) (as defined in 4.25).

Proof. Let 𝐼(𝑉) = 𝔞 and let 𝑃 = (𝑎1 , … , 𝑎𝑛 ). On rewriting a polynomial 𝐹(𝑋1 , … , 𝑋𝑛 )

in terms of the variables 𝑋𝑖 − 𝑎𝑖 , we obtain the (trivial Taylor) formula,
∑ 𝜕𝐹 |||
𝐹(𝑋1 , … , 𝑋𝑛 ) = 𝐹(𝑎1 , … , 𝑎𝑛 ) + | (𝑋 − 𝑎𝑖 ) + 𝑅
𝜕𝑋𝑖 |||𝐚 𝑖

with 𝑅 a finite sum of products of at least two terms (𝑋𝑖 − 𝑎𝑖 ).

According to 4.25, 𝑇𝑃 (𝑉 ′ ) consists of the elements 𝐚 + 𝜀𝐛 of 𝑘[𝜀]𝑛 = 𝑘 𝑛 ⊕ 𝑘𝑛 𝜀 lying
in 𝑉(𝑘[𝜀]). Let 𝐹 ∈ 𝔞. On setting 𝑋𝑖 equal to 𝑎𝑖 + 𝜀𝑏𝑖 in the above formula, we get,

∑ 𝜕𝐹 |||
𝐹(𝑎1 + 𝜀𝑏1 , … , 𝑎𝑛 + 𝜀𝑏𝑛 ) = 𝜀 ( | 𝑏 ).
𝜕𝑋𝑖 |||𝐚 𝑖

Thus, (𝑎1 + 𝜀𝑏1 , … , 𝑎𝑛 + 𝜀𝑏𝑛 ) lies in 𝑉(𝑘[𝜀]) if and only if (𝑏1 , … , 𝑏𝑛 ) ∈ 𝑇𝐚 (𝑉). 2

We can restate this as follows. Let 𝑉 be an affine algebraic variety, and let 𝑃 ∈ 𝑉.
Choose an embedding 𝑉 → 𝔸𝑛 , and let 𝑃 map to (𝑎1 , … , 𝑎𝑛 ). Then the point

(𝑎1 , … , 𝑎𝑛 ) + (𝑏1 , … , 𝑏𝑛 )𝜀

of 𝔸𝑛 (𝑘[𝜀]) is an element of 𝑇𝑃 (𝑉) (definition 4.25) if and only if (𝑏1 , … , 𝑏𝑛 ) is an element

of 𝑇𝑃 (𝑉) (definition 4.22).
 f. Tangent spaces to affine algebraic varieties 91

Proposition 4.27. Let 𝑉 be an affine variety, and let 𝑃 ∈ 𝑉. There is a canonical isomor-
phism

𝑇𝑃 (𝑉) ≃ Hom(𝒪𝑃 , 𝑘[𝜀]) (local homomorphisms of local 𝑘-algebras).

Proof. By definition, an element of 𝑇𝑃 (𝑉) is a homomorphism 𝛼 ∶ 𝑘[𝑉] → 𝑘[𝜀] such

that 𝛼−1 ((𝜀)) = 𝔪𝑃 . Therefore 𝛼 maps elements of 𝑘[𝑉] ∖ 𝔪𝑃 into (𝑘[𝜀] ∖ (𝜀)) = 𝑘[𝜀]× ,
and so 𝛼 extends (uniquely) to a homomorphism 𝛼′ ∶ 𝒪𝑃 → 𝑘[𝜀]. By construction, 𝛼′
is a local homomorphism of local 𝑘-algebras, and clearly every such homomorphism
arises in this way from a (unique) element of 𝑇𝑃 (𝑉). 2

Derivations
Definition 4.28. Let 𝐴 be a 𝑘-algebra and 𝑀 an 𝐴-module. A 𝑘-derivation is a map
𝐷 ∶ 𝐴 → 𝑀 such that
(a) 𝐷(𝑐) = 0 for all 𝑐 ∈ 𝑘;
(b) 𝐷(𝑓 + 𝑔) = 𝐷(𝑓) + 𝐷(𝑔);
(c) 𝐷(𝑓𝑔) = 𝑓 ⋅ 𝐷𝑔 + 𝑔 ⋅ 𝐷𝑓 (Leibniz’s rule).

Note that the conditions imply that 𝐷 is 𝑘-linear (but not 𝐴-linear). The 𝑘-derivations
𝐴 → 𝑀 form a 𝑘-vector space Der𝑘 (𝐴, 𝑀).
For example, let 𝐴 be a local 𝑘-algebra with maximal ideal 𝔪, and assume that
𝐴∕𝔪 = 𝑘. For 𝑓 ∈ 𝐴, let 𝑓(𝔪) denote the image of 𝑓 in 𝐴∕𝔪. Then 𝑓 − 𝑓(𝔪) ∈ 𝔪,

𝑓 ↦ 𝑑𝑓 = 𝑓 − 𝑓(𝔪) mod 𝔪2
and the map
def

is a 𝑘-derivation 𝐴 → 𝔪∕𝔪2 because, modulo 𝔪2 ,

0 = (𝑓 − 𝑓(𝔪))(𝑔 − 𝑔(𝔪))
= −𝑓𝑔 + 𝑓(𝔪)𝑔(𝔪) + 𝑓 ⋅ (𝑔 − 𝑔(𝔪)) + 𝑔(𝑓 − 𝑓(𝔪))
= −𝑑(𝑓𝑔) + 𝑓 ⋅ 𝑑𝑔 + 𝑔 ⋅ 𝑑𝑓.

Proposition 4.29. Let (𝐴, 𝔪) be as above. There are canonical isomorphisms

Hom(𝐴, 𝑘[𝜀]) ≃ Der𝑘 (𝐴, 𝑘) ≃ Hom(𝔪∕𝔪2 , 𝑘).

local 𝑘-algebra homs 𝑘-linear maps

𝑐↦𝑐 𝑓↦𝑓(𝔪)
Proof. The composite of the maps 𝑘 ,,,,→ 𝐴 ,,,,,,,→ 𝑘 is the identity, and so, when
regarded as 𝑘-vector space, 𝐴 decomposes into

𝐴 = 𝑘 ⊕ 𝔪, 𝑓 ↔ (𝑓(𝔪), 𝑓 − 𝑓(𝔪)).

Let 𝛼 ∶ 𝐴 → 𝑘[𝜀] be a local homomorphism of 𝑘-algebras, and write 𝛼(𝑎) = 𝑎0 +

𝐷𝛼 (𝑎)𝜀. Because 𝛼 is a homomorphism of 𝑘-algebras, 𝑎0 = 𝑎(𝔪). We have

𝛼(𝑎𝑏) = (𝑎𝑏)0 + 𝐷𝛼 (𝑎𝑏)𝜀, and

𝛼(𝑎)𝛼(𝑏) = (𝑎0 + 𝐷𝛼 (𝑎)𝜀)(𝑏0 + 𝐷𝛼 (𝑏)𝜀) = 𝑎0 𝑏0 + (𝑎0 𝐷𝛼 (𝑏) + 𝑏0 𝐷𝛼 (𝑎))𝜀.

On comparing these expressions, we see that 𝐷𝛼 satisfies Leibniz’s rule, and therefore is
a 𝑘-derivation 𝐴 → 𝑘. Conversely, if 𝐷 ∶ 𝐴 → 𝑘 is a 𝑘-derivation, then

𝛼 ∶ 𝑎 ↦ 𝑎(𝔪) + 𝐷(𝑎)𝜀
92 4. Local Study

is a local homomorphism of 𝑘-algebras 𝐴 → 𝑘[𝜀], and all such homomorphisms arise in

A derivation 𝐷 ∶ 𝐴 → 𝑘 is zero on 𝑘 and on 𝔪2 (by Leibniz’s rule). It therefore

this way. This proves the first isomorphism.

defines a 𝑘-linear map 𝔪∕𝔪2 → 𝑘. Conversely, a 𝑘-linear map 𝔪∕𝔪 → 𝑘 defines a

2

𝑓↦𝑑𝑓
𝐴,,,,,→ 𝔪∕𝔪2 → 𝑘.
derivation by composition

Tangent spaces and differentials

4.30. Let 𝑉 be an affine algebraic variety, and let 𝑃 be a point on 𝑉. Write 𝔪𝑃 for the
We summarize the above discussion in the context of affine algebraic varieties.

corresponding maximal ideal in 𝑘[𝑉] and 𝔫𝑃 for the maximal ideal 𝔪𝑃 𝒪𝑉,𝑃 in the local
ring at 𝑃. There are canonical isomorphisms

𝑇𝑃 (𝑉) → Der𝑘 (𝑘[𝑉], 𝑘) → Hom(𝔪𝑃 ∕𝔪2𝑃 , 𝑘)

𝑘-linear homs
← ←
←

←
(24)

Hom(𝒪𝑃 , 𝑘[𝜀]) → Der𝑘 (𝒪𝑃 , 𝑘) → Hom(𝔫𝑃 ∕𝔫2𝑃 , 𝑘).

→
→

local 𝑘-algebra homs 𝑘-linear homs

← ←

In the term Der𝑘 (𝑘[𝑉], 𝑘) on the top row, 𝑘[𝑉] acts on 𝑘 through 𝑘[𝑉] → 𝑘[𝑉]∕𝔪𝑃 ≃ 𝑘
(so it depends on 𝑃), and in the term Der𝑘 (𝒪𝑃 , 𝑘) on the bottom row, 𝒪𝑃 acts on 𝑘 through
𝒪𝑃 → 𝒪𝑃 ∕𝔫𝑃 ≃ 𝑘. The maps have the following descriptions.
(a) By definition, 𝑇𝑃 (𝑉) is the fibre of 𝑉(𝑘[𝜀]) → 𝑉(𝑘) over 𝑃. To give an element of
𝑇𝑃 (𝑉) amounts to giving a homomorphism 𝛼 ∶ 𝑘[𝑉] → 𝑘[𝜀] such that 𝛼−1 ((𝜀)) =
𝔪𝑃 .
(b) The homomorphism 𝛼 in (a) can be written,

𝛼(𝑓) = 𝑓(𝔪𝑃 ) ⊕ 𝐷𝛼 (𝑓)𝜀, 𝑓 ∈ 𝑘[𝑉], 𝑓(𝔪𝑃 ) ∈ 𝑘, 𝐷𝛼 (𝑓) ∈ 𝑘.

Then 𝐷𝛼 is a 𝑘-derivation 𝑘[𝑉] → 𝑘, and 𝐷𝛼 induces a 𝑘-linear map 𝔪𝑃 ∕𝔪2𝑃 → 𝑘.

(c) The homomorphism 𝛼 ∶ 𝑘[𝑉] → 𝑘[𝜀] in (a) extends uniquely to a local homo-
morphism 𝒪𝑃 → 𝑘[𝜀]. Similarly, a 𝑘-derivation 𝑘[𝑉] → 𝑘 extends uniquely to a
𝑘-derivation 𝒪𝑃 → 𝑘.

phism 𝔪𝑃 ∕𝔪2𝑃 → 𝔫𝑃 ∕𝔫2𝑃 of (1.15).

(d) The two vector spaces at the right of the diagram are related through the isomor-

4.31. A regular map 𝜑 ∶ 𝑉 → 𝑊 defines a map

𝑇𝑃 (𝑉) → 𝑇𝑄 (𝑊)
𝑑𝜑
𝜑(𝑘[𝜀]) ∶ 𝑉(𝑘[𝜀]) → 𝑊(𝑘[𝜀]), which sends the fibre over
←

𝑃 to the fibre over 𝑄 = 𝜑(𝑃), i.e., it defines a map

←

def

𝑉(𝑘[𝜀]) → 𝑊(𝑘[𝜀])
𝜑
𝑑𝜑 ∶ 𝑇𝑃 (𝑉) → 𝑇𝑄 (𝑊).
→

𝜀↦0 𝜀↦0
This map of tangent spaces is called the differential of 𝜑
←

𝑉(𝑘) → 𝑊(𝑘)
𝜑
at 𝑃.
→

(a) When 𝑉 and 𝑊 are embedded as closed subvarieties of 𝔸𝑛 , 𝑑𝜑 has the description
←

in p. 89.
 g. Tangent cones 93

(b) As a map Hom(𝒪𝑃 , 𝑘[𝜀]) → Hom(𝒪𝑄 , 𝑘[𝜀]), 𝑑𝜑 is induced by 𝜑∗ ∶ 𝒪𝑄 → 𝒪𝑃 .

(c) As a map Hom(𝔪𝑃 ∕𝔪2𝑃 , 𝑘) → Hom(𝔪𝑄 ∕𝔪2𝑄 , 𝑘), 𝑑𝜑 is induced by the map 𝔪𝑄 ∕𝔪2𝑄 →
𝔪𝑃 ∕𝔪2𝑃 defined by 𝜑∗ ∶ 𝑘[𝑊] → 𝑘[𝑉].

Example 4.32. Let 𝐸 be a finite dimensional vector space over 𝑘. Then 𝑇0 (𝔸(𝐸)) ≃ 𝐸.
Aside 4.33. A map Spm(𝑘[𝜀]) → 𝑉 should be thought of as a curve in 𝑉 but with only the first
infinitesimal structure retained. Thus, the descriptions of the tangent space provided by the
terms in the top row of (24) correspond to the three standard descriptions of the tangent space
in differential geometry: via tangent curves, via derivations, via cotangent spaces (Wikipedia:
Tangent space).

g. Tangent cones
Let 𝑉 be an algebraic subset of 𝑘𝑚 , and let 𝔞 = 𝐼(𝑉). Assume that 𝑃 = (0, … , 0) ∈ 𝑉.
Define 𝔞∗ to be the ideal generated by the leading forms 𝐹∗ of the polynomials 𝐹 ∈ 𝔞.
The geometric tangent cone at 𝑃, 𝐶𝑃 (𝑉) is 𝑉(𝔞∗ ), and the tangent cone is the pair

☡ If 𝔞 is principal, say, 𝔞 = (𝐹), then 𝔞

(𝑉(𝔞∗ ), 𝑘[𝑋1 , … , 𝑋𝑛 ]∕𝔞∗ ). Obviously, 𝐶𝑃 (𝑉) ⊂ 𝑇𝑃 (𝑉).2

= (𝐹∗ ), but if 𝔞 = (𝐹1 , … , 𝐹𝑟 ), then it need

∗
not be true that 𝔞∗ = (𝐹1∗ , … , 𝐹𝑟∗ ). Consider for example 𝔞 = (𝑋𝑌, 𝑋𝑍 + 𝑍(𝑌 2 − 𝑍 2 )).
One can show that this is an intersection of prime ideals, and hence is radical. As the

𝑌𝑍(𝑌 2 − 𝑍 2 ) = 𝑌 ⋅ (𝑋𝑍 + 𝑍(𝑌 2 − 𝑍 2 )) − 𝑍 ⋅ (𝑋𝑌)

polynomial

lies in 𝔞 and is homogeneous, it lies in 𝔞∗ , but it is not in the ideal generated by 𝑋𝑌, 𝑋𝑍.
In fact, 𝔞∗ = (𝑋𝑌, 𝑋𝑍, 𝑌𝑍(𝑌 2 − 𝑍 2 ).

Let 𝐴 be a local ring with maximal ideal 𝔫. The associated graded ring is
⨁
gr(𝐴) = 𝔫𝑖 ∕𝔫𝑖+1 .
𝑖≥0
⨁
Note that if 𝐴 = 𝐵𝔪 and 𝔫 = 𝔪𝐴, then gr(𝐴) = 𝔪𝑖 ∕𝔪𝑖+1 (because of 1.15).

Proposition 4.34. The homomorphism of 𝑘-algebras

𝑘[𝑋1 , … , 𝑋𝑛 ]∕𝔞∗ → gr(𝒪𝑃 )

sending the class of 𝑋𝑖 in 𝑘[𝑋1 , … , 𝑋𝑛 ]∕𝔞∗ to the class of 𝑋𝑖 in gr(𝒪𝑃 ) is an isomorphism.

Proof. Let 𝔪 be the maximal ideal in 𝑘[𝑋1 , … , 𝑋𝑛 ]∕𝔞 corresponding to 𝑃. Then

∑
gr(𝒪𝑃 ) = 𝔪𝑖 ∕𝔪𝑖+1
∑
= (𝑋1 , … , 𝑋𝑛 )𝑖 ∕(𝑋1 , … , 𝑋𝑛 )𝑖+1 + 𝔞 ∩ (𝑋1 , … , 𝑋𝑛 )𝑖
∑
= (𝑋1 , … , 𝑋𝑛 )𝑖 ∕(𝑋1 , … , 𝑋𝑛 )𝑖+1 + 𝔞𝑖 ,

where 𝔞𝑖 is the homogeneous piece of 𝔞∗ of degree 𝑖 (that is, the subspace of 𝔞∗ consisting
of homogeneous polynomials of degree 𝑖). But

(𝑋1 , … , 𝑋𝑛 )𝑖 ∕(𝑋1 , … , 𝑋𝑛 )𝑖+1 + 𝔞𝑖 = 𝑖th homogeneous piece of 𝑘[𝑋1 , … , 𝑋𝑛 ]∕𝔞∗ . 2

2
There is a more natural definition of the tangent cone as an algebraic scheme — see Chapter 10.
94 4. Local Study

For an affine algebraic variety 𝑉 and 𝑃 ∈ 𝑉, we define the geometric tangent cone
𝐶𝑃 (𝑉) of 𝑉 at 𝑃 to be Spm(gr(𝒪𝑃 )red ), where gr(𝒪𝑃 )red is the quotient of gr(𝒪𝑃 ) by its
nilradical, and we define the tangent cone to be (𝐶𝑃 (𝑉), gr(𝒪𝑃 )).
As in the case of a curve, the dimension of the geometric tangent cone at 𝑃 is the
same as the dimension of 𝑉 (because the Krull dimension of a noetherian local ring is
equal to that of its graded ring; Matsumura 1989, Theorem 13.9). Moreover, gr(𝒪𝑃 ) is
a polynomial ring in dim 𝑉 variables if and only if 𝒪𝑃 is regular (ibid., Exercise 19.1).
Therefore, 𝑃 is nonsingular (see below) if and only if gr(𝒪𝑃 ) is a polynomial ring in
dim(𝑉) variables, in which case 𝐶𝑃 (𝑉) = 𝑇𝑃 (𝑉).
A regular map 𝜑 ∶ 𝑉 → 𝑊 sending 𝑃 to 𝑄 induces a homomorphism gr(𝒪𝑄 ) →
gr(𝒪𝑃 ), and hence a map 𝐶𝑃 (𝑉) → 𝐶𝑄 (𝑉) of the geometric tangent cones. We say that 𝜑
is étale at 𝑃 if gr(𝒪𝑄 ) → gr(𝒪𝑃 ) is an isomorphism. When 𝑃 and 𝑄 are nonsingular points,

☡ The map on the rings 𝑘[𝑋 , … , 𝑋 ]∕𝔞

this just says that the map 𝑑𝜑 ∶ 𝑇𝑃 (𝑉) → 𝑇𝑄 (𝑊) on tangent spaces is an isomorphism.

∗
1 𝑛 defined by a map of algebraic varieties is not
the obvious one, i.e., it is not necessarily induced by the same map on polynomial rings

necessary to work with the rings gr(𝒪𝑃 ).

as the original map. To see what it is, it is necessary to use Proposition 4.34, i.e., it is

h. Nonsingular points; the singular locus

Definition 4.35. A point 𝑃 on an affine algebraic variety 𝑉 is said to be nonsingular
or smooth if it lies on a single irreducible component 𝑊 of 𝑉 and dim 𝑇𝑃 (𝑉) = dim 𝑊;
otherwise the point is said to be singular. A variety is nonsingular if all of its points
are nonsingular. The set of singular points of a variety is called its singular locus.

Thus, on an irreducible variety 𝑉 of dimension 𝑑,

𝑃 is nonsingular ⇐⇒ dim𝑘 𝑇𝑃 (𝑉) = 𝑑 ⇐⇒ dim𝑘 (𝔫𝑃 ∕𝔫2𝑃 ) = 𝑑.

Proposition 4.36. Let 𝑉 be an irreducible variety of dimension 𝑑, and let 𝑃 be a nonsingu-

lar point on 𝑉. There exist 𝑑 regular functions 𝑓1 , … , 𝑓𝑑 defined in an open neighbourhood
𝑈 of 𝑃 such that 𝑃 is the only common zero of the 𝑓𝑖 on 𝑈.

Proof. Because 𝑃 is nonsingular, there exist regular functions 𝑓1 , … , 𝑓𝑑 on an open

neighbourhood 𝑈 of 𝑃 whose images in 𝒪𝑃 generate its maximal ideal 𝔫𝑃 . We show that
𝑃 is an irreducible component of the zero set of the 𝑓1 , … , 𝑓𝑑 in 𝑈. If not, there exists an
irreducible component 𝑍 of 𝑉(𝑓1 , … , 𝑓𝑑 ) properly containing 𝑃. Write 𝑍 = 𝑉(𝔭) with 𝔭
is a prime ideal in 𝑘[𝑈]. As 𝑉(𝔭) ⊂ 𝑉(𝑓1 , … , 𝑓𝑑 ) and 𝑍 properly contains 𝑃,

(𝑓1 , … , 𝑓𝑑 ) ⊂ 𝔭 ⫋ 𝔪𝑃 (ideals in 𝑘[𝑈]).

On passing to the local ring 𝒪𝑃 = 𝑘[𝑈]𝔪𝑃 , we find (using 1.14) that

(𝑓1 , … , 𝑓𝑑 ) ⊂ 𝔭𝒪𝑃 ⫋ 𝔫𝑃 (ideals in 𝒪𝑃 ).

This contradicts the assumption that the 𝑓𝑖 generate 𝔫𝑃 . Hence 𝑃 is an irreducible

component of 𝑉(𝑓1 , … , 𝑓𝑑 ). On removing the remaining irreducible components of
𝑉(𝑓1 , … , 𝑓𝑑 ) from 𝑈, we obtain an open neighbourhood of 𝑃 with the required property.2
 h. Nonsingular points; the singular locus 95

Let 𝑃 be a point (possibly singular) on an irreducible variety 𝑉. The argument in the

above proof shows that, if 𝑓1 , … , 𝑓𝑟 generate the maximal ideal 𝔫𝑃 in 𝒪𝑃 , then 𝑃 is an
irreducible component of 𝑉(𝑓1 , … , 𝑓𝑟 ), and so 𝑟 ≥ 𝑑 (by 3.45). Nakayama’s lemma (1.3)
shows that 𝑓1 , … , 𝑓𝑟 generate 𝔫𝑃 if and only if their images span 𝔫𝑃 ∕𝔫2𝑃 . Thus

dim 𝑇𝑃 (𝑉) ≥ dim 𝑉, with equality if and only if 𝑃 is nonsingular.

A point 𝑃 on 𝑉 is nonsingular if and only if there exists an open affine neighbour-

hood 𝑈 of 𝑃 and functions 𝑓1 , … 𝑓𝑑 on 𝑈 such that (𝑓1 , … , 𝑓𝑑 ) is the ideal of all regular
functions on 𝑈 zero at 𝑃.

Theorem 4.37. The set of nonsingular points of an affine algebraic variety 𝑉 is dense and
open.

Proof. We first show that the singular locus is closed. We may suppose that 𝑉 is affine,
say, 𝑉 = 𝑉(𝔞) ⊂ 𝔸𝑛 . Let 𝑃1 , … , 𝑃𝑟 generate 𝔞. Then the singular locus is the zero set of
the ideal generated by the (𝑛 − 𝑑) × (𝑛 − 𝑑) minors of the matrix

⎛ 𝜕𝑃1 𝜕𝑃1 ⎞
(𝐚) … (𝐚)
⎜ 𝜕𝑋1 𝜕𝑋𝑛 ⎟
Jac(𝑃1 , … , 𝑃𝑟 )(𝐚) = ⎜ ⋮ ⋮ ⎟.
⎜ 𝜕𝑃𝑟 𝜕𝑃𝑟 ⎟
(𝐚) … (𝐚)
⎝ 𝜕𝑋1 𝜕𝑋𝑛 ⎠

We next assume that 𝑉 is irreducible and prove that 𝑉sing ≠ 𝑉. According to 3.36
This is closed.

and 3.37 some nonempty open affine subset of 𝑉 is isomorphic to a nonempty open
affine subset of an irreducible hypersurface in 𝔸𝑑+1 , and so we may suppose that 𝑉
itself is an irreducible hypersurface in 𝔸𝑑+1 , say, equal to the zero set of the nonconstant
irreducible polynomial 𝐹(𝑋1 , … , 𝑋𝑑+1 ). The singular locus is the set of common zeros of

𝜕𝐹 𝜕𝐹
𝐹, ,…, ,
the polynomials

𝜕𝑋1 𝜕𝑋𝑑+1
and so it will be proper unless the polynomials 𝜕𝐹∕𝜕𝑋𝑖 are identically zero on 𝑉. As
in the proof of 4.7, if 𝜕𝐹∕𝜕𝑋𝑖 is identically zero on 𝑉(𝐹), then it is the zero polyno-
mial, and so 𝐹 is a polynomial in 𝑋1 , … , 𝑋𝑖−1 , 𝑋𝑖+1 , … 𝑋𝑑+1 (characteristic zero) or in
𝑋1 , … , 𝑋𝑖 , … , 𝑋𝑑+1 (characteristic 𝑝). Therefore, if the singular locus equals 𝑉, then 𝐹
𝑝

is constant (characteristic 0) or a 𝑝th power (characteristic 𝑝), which contradicts the

We now consider the general case, in which 𝑉 has irreducible components 𝑉1 , … , 𝑉𝑟 .

hypothesis.

Each of 𝑉𝑖 ∩𝑉𝑗 is a proper closed subset of 𝑉𝑖 , and we have proved that (𝑉𝑖 )sing is a proper
closed subset of 𝑉𝑖 . It follows that 𝑉𝑖 ∩ 𝑉sing is a finite union of proper closed subsets of
𝑉𝑖 , and so it is proper and closed in 𝑉𝑖 . Hence the points of 𝑉𝑖 that are nonsingular on 𝑉
form a nonempty open (hence dense) subset of 𝑉𝑖 . 2

Corollary 4.38. If 𝑉 is irreducible, then

dim 𝑉 = min dim 𝑇𝑃 (𝑉).

𝑃∈𝑉

Proof. We know that dim 𝑇𝑃 (𝑉) ≥ dim 𝑉, with equality if and only if 𝑃 is nonsingular.
As there exists a nonsingular 𝑃, dim 𝑉 is the minimum value of dim 𝑇𝑃 (𝑉). 2
96 4. Local Study

This formula can be useful in computing the dimension of a variety.

spaces 𝑇𝑃 (𝑉), 𝑃 ∈ 𝑉, have constant dimension.

Corollary 4.39. An irreducible algebraic variety is nonsingular if and only if the tangent

Proof. The constant dimension is the dimension of 𝑉, and so all points are nonsingu-
lar. 2

Corollary 4.40. Every variety on which a group acts transitively by regular maps is
nonsingular.

Proof. The group must act by isomorphisms, and so the tangent spaces have constant
dimension. 2

In particular, every group variety (see p. 110) is nonsingular.

Examples
4.41. For the surface 𝑍 3 = 𝑋𝑌, the only singular point is (0, 0, 0). The tangent cone at
(0, 0, 0) has equation 𝑋𝑌 = 0, and so it is the union of two planes intersecting in the
𝑧-axis.

4.42. For the surface 𝑉 ∶ 𝑍 3 = 𝑋 2 𝑌, the singular locus is the line 𝑋 = 0 = 𝑍, i.e., the
𝑦-axis. The singularity at (0, 0) is very bad: for example, it lies in the singular set of the
singular set.3 The intersection of the surface with the surface 𝑌 = 𝑐 is the cuspidal curve
𝑋 2 = 𝑍 3 ∕𝑐. In the picture, each curve lies in a plane 𝑌 = 𝑐 orthogonal to the 𝑦-axis, and
has its cusp on the 𝑦-axis.

4.43. Let 𝑉 be the union of the coordinate axes in 𝔸3 , and let 𝑊 be the zero set of
𝑋𝑌(𝑋 − 𝑌) in 𝔸2 . Each of 𝑉 and 𝑊 is a union of three lines meeting at the origin. Are
they isomorphic as algebraic varieties? Obviously, the origin 𝑜 is the only singular point
on 𝑉 or 𝑊. An isomorphism 𝑉 → 𝑊 would have to send the singular point 𝑜 to the
3
In fact, it belongs to the worst class of singularities (sx2848895, KReiser).
 i. Nonsingularity and regularity 97

singular point 𝑜 and map 𝑇𝑜 (𝑉) isomorphically onto 𝑇𝑜 (𝑊). But 𝑉 = 𝑉(𝑋𝑌, 𝑌𝑍, 𝑋𝑍),
and so 𝑇𝑜 (𝑉) has dimension 3, whereas 𝑇𝑜 𝑊 has dimension 2. Therefore, 𝑉 and 𝑊 are
not isomorphic.

i. Nonsingularity and regularity

Theorem 4.44. Let 𝑃 be a point on an irreducible variety 𝑉. Every generating set for the
maximal ideal 𝔫𝑃 of 𝒪𝑃 has at least 𝑑 elements, and there exists a generating set with 𝑑
elements if and only if 𝑃 is nonsingular.

Proof. If 𝑓1 , … , 𝑓𝑟 generate 𝔫𝑃 , then the proof of 4.36 shows that 𝑃 is an irreducible

component of 𝑉(𝑓1 , … , 𝑓𝑟 ) in some open neighbourhood 𝑈 of 𝑃. Therefore 3.45 shows
that 0 ≥ 𝑑 − 𝑟, and so 𝑟 ≥ 𝑑. The rest of the statement has already been noted.

Corollary 4.45. A point 𝑃 on a variety 𝑉 is nonsingular if and only if 𝒪𝑃 is regular.

2

Proof. If 𝑃 lies on a single irreducible component of 𝑉, then this is a restatement

integral domain. If 𝑃 lies on more than one irreducible component of a 𝑉, then 𝑃 is

of the second part of the theorem. According to CA, 22.3, a regular local ring is an

not nonsingular (by definition) and 𝒪𝑃 is not regular because not an integral domain
(3.14). 2

j. Examples of tangent spaces

when our variety is given to us in terms of its points functor. For example, let 𝑀𝑛 be
The description of the tangent space in terms of dual numbers is particularly convenient

the set of 𝑛 × 𝑛 matrices, and let 𝐼 be the identity matrix. Write 𝑒 for 𝐼 when it is to be
regarded as the identity element of GL𝑛 .
4.46. A matrix 𝐼 + 𝜀𝐴 has inverse 𝐼 − 𝜀𝐴 in 𝑀𝑛 (𝑘[𝜀]), and so lies in GL𝑛 (𝑘[𝜀]). In fact,

𝑇𝑒 (GL𝑛 ) = {𝐼 + 𝜀𝐴 ∣ 𝐴 ∈ 𝑀𝑛 }
≃ 𝑀𝑛 (𝑘).

4.47. On expanding det(𝐼 + 𝜀𝐴) as a sum of signed products and using that 𝜀2 = 0, we

det(𝐼 + 𝜀𝐴) = 𝐼 + 𝜀trace(𝐴).

find that

𝑇𝑒 (SL𝑛 ) = {𝐼 + 𝜀𝐴 ∣ trace(𝐴) = 0}
Hence

≃ {𝐴 ∈ 𝑀𝑛 (𝑘) ∣ trace(𝐴) = 0}.

4.48. Assume that the characteristic ≠ 2, and let O𝑛 be the orthogonal group:

O𝑛 = {𝐴 ∈ GL𝑛 ∣ 𝐴tr ⋅ 𝐴 = 𝐼}.

(𝐴tr denotes the transpose of 𝐴). This is the group of matrices preserving the quadratic
form 𝑋12 + ⋯ + 𝑋𝑛2 . The determinant defines a surjective regular homomorphism
det ∶ O𝑛 → {±1}, whose kernel is defined to be the special orthogonal group SO𝑛 . For
𝐼 + 𝜀𝐴 ∈ 𝑀𝑛 (𝑘[𝜀]),
(𝐼 + 𝜀𝐴)tr ⋅ (𝐼 + 𝜀𝐴) = 𝐼 + 𝜀𝐴tr + 𝜀𝐴,
98 4. Local Study

and so

𝑇𝑒 (O𝑛 ) = 𝑇𝑒 (SO𝑛 ) = {𝐼 + 𝜀𝐴 ∈ 𝑀𝑛 (𝑘[𝜀]) ∣ 𝐴 is skew-symmetric}

≃ {𝐴 ∈ 𝑀𝑛 (𝑘) ∣ 𝐴 is skew-symmetric}.

Aside 4.49. On the tangent space 𝑇𝑒 (GL𝑛 ) ≃ 𝑀𝑛 of GL𝑛 , there is a bracket operation

[𝑀, 𝑁] = 𝑀𝑁 − 𝑁𝑀
def

which makes 𝑇𝑒 (GL𝑛 ) into a Lie algebra. For any closed algebraic subgroup 𝐺 of GL𝑛 , 𝑇𝑒 (𝐺) is
stable under the bracket operation on 𝑇𝑒 (GL𝑛 ) and is a sub-Lie-algebra of 𝑀𝑛 , which we denote
Lie(𝐺). The Lie algebra structure on Lie(𝐺) is independent of the embedding of 𝐺 into GL𝑛 (it
has an intrinsic definition in terms of left invariant derivations), and 𝐺 ⇝Lie(𝐺) is a functor

This functor is not fully faithful, for example, every étale homomorphism 𝐺 → 𝐺 ′ defines an
from the category of linear group varieties to that of Lie algebras.

isomorphism Lie(𝐺) → Lie(𝐺 ′ ), but it is nevertheless very useful.

Assume that 𝑘 has characteristic zero. A connected group variety 𝐺 is said to be semisimple
if it has no closed connected solvable normal subgroup (except {𝑒}). Such a group 𝐺 may have a
finite nontrivial centre 𝑍(𝐺), and we call two semisimple groups 𝐺 and 𝐺 ′ locally isomorphic
if 𝐺∕𝑍(𝐺) ≈ 𝐺 ′ ∕𝑍(𝐺 ′ ). For example, SL𝑛 is semisimple, with centre 𝜇𝑛 , the set of diagonal
matrices diag(𝜁, … , 𝜁) with 𝜁 𝑛 = 1, and SL𝑛 ∕𝜇𝑛 = PSL𝑛 . A Lie algebra is semisimple if it has
no commutative ideal (except {0}). One can prove that

𝐺 is semisimple ⇐⇒ Lie(𝐺) is semisimple,

and the map 𝐺 ↦ Lie(𝐺) defines a one-to-one correspondence between the set of local isomor-
phism classes of semisimple group varieties and the set of isomorphism classes of Lie algebras.
The classification of semisimple group varieties can be deduced from that of semisimple Lie
algebras and a study of the finite coverings of semisimple group varieties arXiv:0705.1348— this
is quite similar to the relation between Lie groups and Lie algebras.

Exercises

(a) 𝑌 3 − 𝑌 2 + 𝑋 3 − 𝑋 2 + 3𝑌 2 𝑋 + 3𝑋 2 𝑌 + 2𝑋𝑌;
4-1. Find the singular points, and the tangent cones at the singular points, for each of

(b) 𝑋 4 + 𝑌 4 − 𝑋 2 𝑌 2 (assume that the characteristic is not 2).

4-2. Let 𝑉 ⊂ 𝔸𝑛 be an irreducible affine variety, and let 𝑃 be a nonsingular point

on 𝑉. Let 𝐻 be a hyperplane in 𝔸𝑛 (i.e., the subvariety defined by a linear equation
∑
𝑎𝑖 𝑋𝑖 = 𝑑 with not all 𝑎𝑖 zero) passing through 𝑃 but not containing 𝑇𝑃 (𝑉). Show
that 𝑃 is a nonsingular point on each irreducible component of 𝑉 ∩ 𝐻 on which it lies.
(Each irreducible component has codimension 1 in 𝑉 — you may assume this.) Give
an example with 𝐻 ⊃ 𝑇𝑃 (𝑉) and 𝑃 singular on 𝑉 ∩ 𝐻. Must 𝑃 be singular on 𝑉 ∩ 𝐻 if
𝐻 ⊃ 𝑇𝑃 (𝑉)?

4-3. Given a smooth point on a variety and a tangent vector at the point, show that
there is a smooth curve passing through the point with the given vector as its tangent
vector (see mo111467).

4-4. Let 𝑃 and 𝑄 be points on varieties 𝑉 and 𝑊. Show that

𝑇(𝑃,𝑄) (𝑉 × 𝑊) ≃ 𝑇𝑃 (𝑉) ⊕ 𝑇𝑄 (𝑊).

 j. Examples of tangent spaces 99

4-5. For each 𝑛, show that there is a curve 𝐶 and a point 𝑃 on 𝐶 such that the tangent
space to 𝐶 at 𝑃 has dimension 𝑛 (hence 𝐶 cannot be embedded in 𝔸𝑛−1 ).
( 0 𝐼)
4-6. Let 𝐼 be the 𝑛 × 𝑛 identity matrix, and let 𝐽 be the matrix −𝐼 0 . The symplectic
group Sp𝑛 is the group of 2𝑛 × 2𝑛 matrices 𝐴 with determinant 1 such that 𝐴tr ⋅ 𝐽 ⋅ 𝐴 = 𝐽.

tangent space to Sp𝑛 at its identity element, and also the dimension of Sp𝑛 .
(It is the group of matrices fixing a nondegenerate skew-symmetric form.) Find the

4-7. Find a regular map 𝛼 ∶ 𝑉 → 𝑊 which induces an isomorphism on the geometric

tangent cones 𝐶𝑃 (𝑉) → 𝐶𝛼(𝑃) (𝑊) but is not étale at 𝑃.

4-8. Show that the cone 𝑋 2 + 𝑌 2 = 𝑍 2 is a normal variety, even though the origin is
singular (characteristic ≠ 2). See p. 177.

4-9. Let 𝑉 = 𝑉(𝔞) ⊂ 𝔸𝑛 . Suppose that 𝔞 ≠ 𝐼(𝑉), and for 𝐚 ∈ 𝑉, let 𝑇𝐚′ be the subspace
of 𝑇𝐚 (𝔸𝑛 ) defined by the equations (𝑑𝑓)𝐚 = 0, 𝑓 ∈ 𝔞. Clearly, 𝑇𝐚′ ⊃ 𝑇𝐚 (𝑉), but need
they always be different?

4-10. Let 𝑊 be a finite-dimensional 𝑘-vector space, and let 𝑅𝑊 = 𝑘 ⊕ 𝑊 endowed with

the 𝑘-algebra structure for which 𝑊 2 = 0. Let 𝑉 be an affine algebraic variety over 𝑘.
Show that the elements of 𝑉(𝑅𝑊 ) = Hom𝑘-algebra (𝑘[𝑉], 𝑅𝑊 ) are in natural one-to-one
correspondence with the pairs (𝑃, 𝑡) with 𝑃 ∈ 𝑉 and 𝑡 ∈ 𝑊 ⊗ 𝑇𝑃 (𝑉) (cf. Mumford 1966b,
def

p. 25).
Chapter 5

Algebraic Varieties

An algebraic variety is a ringed space that is locally isomorphic to an affine algebraic

open subset of ℝ𝑛 . We require both to satisfy a separation axiom.

variety, just as a topological manifold is a ringed space that is locally isomorphic to an

a. Algebraic prevarieties

Definition 5.1. (a) A topological manifold of dimension 𝑛 is a ringed space (𝑉, 𝒪𝑉 )

As motivation, recall the following definitions.

such that 𝑉 is Hausdorff and every point of 𝑉 has an open neighbourhood 𝑈 for which
(𝑈, 𝒪𝑉 |𝑈) is isomorphic to the ringed space of continuous functions on an open subset
of ℝ𝑛 (cf. 3.2).
(b) A differentiable manifold of dimension 𝑛 is a ringed space such that 𝑉 is
Hausdorff and every point of 𝑉 has an open neighbourhood 𝑈 for which (𝑈, 𝒪𝑉 |𝑈) is
isomorphic to the ringed space of smooth functions on an open subset of ℝ𝑛 (cf. 3.3).
(c) A complex manifold of dimension 𝑛 is a ringed space such that 𝑉 is Hausdorff
and every point of 𝑉 has an open neighbourhood 𝑈 for which (𝑈, 𝒪𝑉 |𝑈) is isomorphic
to the ringed space of holomorphic functions on an open subset of ℂ𝑛 (cf. 3.4).

These definitions are easily seen to be equivalent to the more classical definitions in

imposed on 𝑉, for example, that it is connected or have a countable base of open subsets.
terms of charts and atlases (when stated correctly). Sometimes additional conditions are

Definition 5.2. An algebraic prevariety over 𝑘 is a 𝑘-ringed space (𝑉, 𝒪𝑉 ) such that
𝑉 is quasi-compact and every point of 𝑉 has an open neighbourhood 𝑈 for which
(𝑈, 𝒪𝑉 |𝑈) is isomorphic to the ringed space of regular functions on an algebraic set over
𝑘.

Thus, a 𝑘-ringed space (𝑉, 𝒪𝑉 ) is an algebraic prevariety over 𝑘 if there exists a finite
⋃
open covering 𝑉 = 𝑉𝑖 such that (𝑉𝑖 , 𝒪𝑉 |𝑉𝑖 ) is an affine algebraic variety over 𝑘 for all
𝑖. An algebraic variety will be defined to be an algebraic prevariety satisfying a certain

An open subset 𝑈 of an algebraic prevariety 𝑉 such that (𝑈, 𝒪𝑉 |𝑈) is an affine

separation condition.

algebraic variety is called an open affine (subvariety) in 𝑉. Because 𝑉 is a finite union

of open affines, and in each open affine the open affines (in fact the basic open subsets)

on 𝑉.
form a base for the topology, it follows that the open affines form a base for the topology

100
 b. Regular maps 101

Let (𝑉, 𝒪𝑉 ) be an algebraic prevariety, and let 𝑈 be an open subset of 𝑉. The functions
𝑓 ∶ 𝑈 → 𝑘 lying in 𝛤(𝑈, 𝒪𝑉 ) are called regular. Note that if (𝑈𝑖 ) is an open covering of
𝑉 by affine varieties, then 𝑓 ∶ 𝑈 → 𝑘 is regular if and only if 𝑓|𝑈𝑖 ∩ 𝑈 is regular for all
𝑖 (by 3.1(c)). Thus understanding the regular functions on open subsets of 𝑉 amounts

subvarieties fit together to form 𝑉.

to understanding the regular functions on the open affine subvarieties and how these

Example 5.3. (Projective space). Let ℙ𝑛 denote 𝑘 𝑛+1 ∖ {origin} modulo the equivalence
relation

(𝑎0 , … , 𝑎𝑛 ) ∼ (𝑏0 , … , 𝑏𝑛 ) ⇐⇒ (𝑎0 , … , 𝑎𝑛 ) = (𝑐𝑏0 , … , 𝑐𝑏𝑛 ) some 𝑐 ∈ 𝑘 × .

Thus the equivalence classes are the lines through the origin in 𝑘𝑛+1 (with the origin
omitted). Write (𝑎0 ∶ … ∶ 𝑎𝑛 ) for the equivalence class containing (𝑎0 , … , 𝑎𝑛 ). For each
𝑖, let
𝑈𝑖 = {(𝑎0 ∶ … ∶ 𝑎𝑖 ∶ … ∶ 𝑎𝑛 ) ∈ ℙ𝑛 ∣ 𝑎𝑖 ≠ 0}.
⋃
Then ℙ𝑛 = 𝑈𝑖 , and the map

𝑎 ˆ
𝑎 𝑎 𝑢𝑖
(𝑎0 ∶ … ∶ 𝑎𝑛 ) ↦ ( 𝑎0 , … , 𝑎𝑖 , … , 𝑎𝑛 ) ∶ 𝑈𝑖 ,→ 𝔸𝑛
𝑖 𝑖 𝑖

(the term 𝑎𝑖 ∕𝑎𝑖 is omitted) is a bijection. In Chapter 6 we shall show that there is a
unique structure of a (separated) algebraic variety on ℙ𝑛 for which each 𝑈𝑖 is an open
affine subvariety of ℙ𝑛 and each map 𝑢𝑖 is an isomorphism of algebraic varieties.

b. Regular maps
In each of the examples (5.1a,b,c), a morphism of manifolds (continuous map, smooth
map, holomorphic map respectively) is just a morphism of ringed spaces. This motivates

Let (𝑉, 𝒪𝑉 ) and (𝑊, 𝒪𝑊 ) be algebraic prevarieties. A map 𝜑 ∶ 𝑉 → 𝑊 is said to

the following definition.

be regular if it is a morphism of 𝑘-ringed spaces. In other words, a continuous map

𝜑 ∶ 𝑉 → 𝑊 is regular if 𝑓 ↦ 𝑓◦𝜑 sends a regular function on an open subset 𝑈 of 𝑊
to a regular function on 𝜑−1 (𝑈). A composite of regular maps is again regular (this is a
general fact about morphisms of ringed spaces).
Note that we have three categories:

(affine varieties) ⊂ (algebraic prevarieties) ⊂ (ringed spaces).

Each subcategory is full, i.e., the morphisms Mor(𝑉, 𝑊) are the same in the three cate-

Proposition 5.4. Let (𝑉, 𝒪𝑉 ) and (𝑊, 𝒪𝑊 ) be prevarieties, and let 𝜑 ∶ 𝑉 → 𝑊 be a

gories.

⋃
continuous map (of topological spaces). Let 𝑊 = 𝑊𝑗 be a covering of 𝑊 by open affines,
⋃
and let 𝜑−1 (𝑊𝑗 ) = 𝑉𝑗𝑖 be a covering of 𝜑−1 (𝑊𝑗 ) by open affines. Then 𝜑 is regular if

𝜑|𝑉𝑗𝑖 ∶ 𝑉𝑗𝑖 → 𝑊𝑗
and only if its restrictions

are regular for all 𝑖, 𝑗.

102 5. Algebraic Varieties

Proof. We assume that 𝜑 satisfies this condition, and prove that it is regular. Let 𝑓
be a regular function on an open subset 𝑈 of 𝑊. Then 𝑓|𝑈 ∩ 𝑊𝑗 is regular for each
𝑊𝑗 (sheaf condition 3.1(b)), and so 𝑓◦𝜑|𝜑−1 (𝑈) ∩ 𝑉𝑗𝑖 is regular for each 𝑗, 𝑖 (this is our
assumption). It follows that 𝑓◦𝜑 is regular on 𝜑−1 (𝑈) (sheaf condition 3.1(c)). Thus 𝜑
is regular. The converse is even easier. 2

Remark 5.5. A differentiable manifold of dimension 𝑛 is locally isomorphic to an open

subset of ℝ𝑛 . In particular, all manifolds of the same dimension are locally isomorphic.
This is not true for algebraic varieties, for two reasons.
(a) We are not assuming our varieties are nonsingular (see Chapter 4).

an isomorphism on the tangent space at a point 𝑃 need not induce an isomorphism in a

(b) The inverse function theorem fails in our context: a regular map that induces

neighbourhood of 𝑃. However, see 5.55 below.

c. Algebraic varieties
In the study of topological manifolds, the Hausdorff condition eliminates such bizarre
possibilities as the line with the origin doubled in which a sequence tending to the origin
has two limits (see 5.9 below).
It is not immediately obvious how to impose a separation axiom on our algebraic
varieties, because even affine algebraic varieties are not Hausdorff. The key is to restate
the Hausdorff condition. Intuitively, the significance of this condition is that it prevents a

should be determined by its values on a dense subset, i.e., if 𝜑1 and 𝜑2 are continuous
sequence in the space having more than one limit. Thus a continuous map into the space

maps 𝑍 → 𝑉 that agree on a dense subset 𝑈 of 𝑍, then they should agree on the whole
of 𝑍.1 Equivalently, the set where two continuous maps 𝜑1 , 𝜑2 ∶ 𝑍 ⇉ 𝑈 agree should be
closed. Surprisingly, affine varieties have this property, provided 𝜑1 and 𝜑2 are required

Lemma 5.6. Let 𝜑1 , 𝜑2 ∶ 𝑍 ⇉ 𝑉 regular maps of affine algebraic varieties. The subset of
to be regular maps.

𝑍 on which 𝜑1 and 𝜑2 agree is closed.

Proof. There are regular functions 𝑥𝑖 on 𝑉 such that 𝑃 ↦ (𝑥1 (𝑃), … , 𝑥𝑛 (𝑃)) identifies 𝑉
with a closed subset of 𝔸𝑛 (take the 𝑥𝑖 to be any set of generators for 𝑘[𝑉] as a 𝑘-algebra).
Now 𝑥𝑖 ◦𝜑1 and 𝑥𝑖 ◦𝜑2 are regular functions on 𝑍, and the set where 𝜑1 and 𝜑2 agree is
⋂𝑛
𝑖=1
𝑉(𝑥𝑖 ◦𝜑1 − 𝑥𝑖 ◦𝜑2 ), which is closed. 2

Definition 5.7. An algebraic prevariety 𝑉 is said to be separated if it satisfies the

for any pair of regular maps 𝜑1 , 𝜑2 ∶ 𝑍 ⇉ 𝑉 with 𝑍 an affine algebraic variety,

separation axiom:

the set {𝑧 ∈ 𝑍 ∣ 𝜑1 (𝑧) = 𝜑2 (𝑧)} is closed in 𝑍.

An algebraic variety over 𝑘 is a separated algebraic prevariety over 𝑘.2
Let 𝑧 ∈ 𝑍, and let 𝑧 = lim 𝑢𝑛 with 𝑢𝑛 ∈ 𝑈. Then 𝜑1 (𝑧) = lim 𝜑1 (𝑢𝑛 ) because 𝜑1 is continuous, and
lim 𝜑1 (𝑢𝑛 ) = lim 𝜑2 (𝑢𝑛 ) = 𝜑2 (𝑧).
1

2
These are sometimes called “algebraic varieties in the sense of FAC” (see the footnote p. 9). For

reduced separated schemes of finite type over 𝑘 (assumed to be algebraically closed) with the nonclosed
Grothendieck, they are the “espaces algébriques de Serre” (EGA I, Appendice); alternatively, they are

points omitted — we explain this in Chapter 10. Some authors use a more restrictive definition — they may
require a variety to be connected, irreducible, or quasi-projective — usually because their foundations do
not allow for a more flexible definition.
 d. Maps from varieties to affine varieties 103

Proposition 5.8. Let 𝜑1 and 𝜑2 be regular maps 𝑍 ⇉ 𝑉 from an algebraic prevariety 𝑍

to a variety 𝑉. The subset of 𝑍 on which 𝜑1 and 𝜑2 agree is closed.

Proof. Let 𝑊 be the set on which 𝜑1 and 𝜑2 agree. For any open affine 𝑈 of 𝑍, 𝑊 ∩ 𝑈
is the subset of 𝑈 on which 𝜑1 |𝑈 and 𝜑2 |𝑈 agree, and so 𝑊 ∩ 𝑈 is closed. This implies
that 𝑊 is closed because 𝑍 is a finite union of open affines. 2

Example 5.9. (The affine line with the origin doubled.)3 Let 𝑉1 and 𝑉2 be copies of
𝔸1 . Let 𝑉 ∗ = 𝑉1 ⊔ 𝑉2 (disjoint union), and give it the obvious topology. Define an
equivalence relation on 𝑉 ∗ by

𝑥 (in 𝑉1 ) ∼ 𝑦 (in 𝑉2 ) ⇐⇒ 𝑥 = 𝑦 and 𝑥 ≠ 0.

Let 𝑉 be the quotient space 𝑉 = 𝑉 ∗ ∕∼ with the quotient topology,

∙
∙

Then 𝑉1 and 𝑉2 are open subspaces of 𝑉, 𝑉 = 𝑉1 ∪ 𝑉2 , and 𝑉1 ∩ 𝑉2 = 𝔸1 − {0}. Define

a function on an open subset to be regular if its restriction to each 𝑉𝑖 is regular. This
makes 𝑉 into a prevariety but not a variety: it fails the separation axiom because the two

𝔸1 = 𝑉1 → 𝑉 ∗ , 𝔸1 = 𝑉2 → 𝑉 ∗
maps

agree exactly on 𝔸1 − {0}, which is not closed in 𝔸1 .

5.10. When 𝑉 is irreducible, all the rings attached to it have a common field of fractions
𝑘(𝑉) (see p. 115 below). Moreover,

𝒪𝑃 = {𝑔∕ℎ ∈ 𝑘(𝑉) ∣ ℎ(𝑃) ≠ 0}

⋂
𝒪𝑉 (𝑈) = {𝒪𝑉 (𝑈 ′ ) ∣ 𝑈 ′ ⊂ 𝑈, 𝑈 ′ open affine}
⋂
= {𝒪𝑃 ∣ 𝑃 ∈ 𝑈}.

d. Maps from varieties to affine varieties

Let (𝑉, 𝒪𝑉 ) be an algebraic variety, and let 𝛼 ∶ 𝐴 → 𝛤(𝑉, 𝒪𝑉 ) be a homomorphism
from an affine 𝑘-algebra 𝐴 to the 𝑘-algebra of regular functions on 𝑉. For any 𝑃 ∈ 𝑉,
𝑓 ↦ 𝛼(𝑓)(𝑃) is a 𝑘-algebra homomorphism 𝐴 → 𝑘, and so its kernel 𝜑(𝑃) is a maximal
ideal in 𝐴. In this way, we get a map

𝜑 ∶ 𝑉 → spm(𝐴)

which is easily seen to be regular. Conversely, from a regular map 𝜑 ∶ 𝑉 → Spm(𝐴),

we get a 𝑘-algebra homomorphism 𝑓 ↦ 𝑓◦𝜑 ∶ 𝐴 → 𝛤(𝑉, 𝒪𝑉 ). Since these maps are
inverse, we have proved the following result.
This is the algebraic analogue of the standard example of a non Hausdorff topological space. Let ℝ∗
denote the real line with the origin removed but with two points 𝑎1 ≠ 𝑎2 added. The subspace ℝ ∖ {0} has
3

its usual topology, and, for 𝑖 = 1, 2, a base for the neighbourhoods of 𝑎𝑖 is formed by the sets (𝑈 ∖ {0}) ∪ {𝑎𝑖 }
with 𝑈 an open neighbourhood of 0 in ℝ. Then ℝ∗ is not Hausdorff because 𝑎1 and 𝑎2 cannot be separated
by disjoint open sets. Every sequence that converges to 𝑎1 also converges to 𝑎2 . For example, 1∕𝑛 converges
to both 𝑎1 and 𝑎2 .
104 5. Algebraic Varieties

Proposition 5.11. For an algebraic variety 𝑉 and an affine 𝑘-algebra 𝐴, there is a canon-

Mor(𝑉, Spm(𝐴)) ≃ Hom𝑘-algebra (𝐴, 𝛤(𝑉, 𝒪𝑉 )).

ical bijection

Let 𝑉 be an algebraic variety such that 𝛤(𝑉, 𝒪𝑉 ) is an affine 𝑘-algebra. The proposi-
tion shows that the regular map 𝜑 ∶ 𝑉 → Spm(𝛤(𝑉, 𝒪𝑉 )) defined by id𝛤(𝑉,𝒪𝑉 ) has the
following universal property: every regular map from 𝑉 to an affine algebraic variety 𝑈
factors uniquely through 𝜑:

𝑉 → Spm(𝛤(𝑉, 𝒪𝑉 ))
← 𝜑
←
∃!
→

←
𝑈.

→
In particular, we recover 3.24: for affine 𝑘-algebras 𝐴 and 𝐵,

Hom𝑘 (𝐴, 𝐵) ≃ Mor(Spm(𝐵), Spm(𝐴)).

Let 𝖵𝖺𝗋 𝑘 denote the category of algebraic varieties over 𝑘 and regular maps. The functor
𝐴 ⇝ Spm (𝐴) ∶ 𝖠𝖿𝖿 𝑘 → 𝖵𝖺𝗋 𝑘 defines a contravariant equivalence of the first category

☡ For a nonaffine algebraic variety 𝑉, 𝛤(𝑉, 𝒪 ) need not be finitely generated as a

with the subcategory of the second whose objects are the affine algebraic varieties.

𝑉
𝑘-algebra.

e. Subvarieties
Let (𝑉, 𝒪𝑉 ) be an algebraic variety over 𝑘.

Open subvarieties
Let 𝑈 be an open subset of 𝑉. Then 𝑈 is a union of open affines, and it follows that
(𝑈, 𝒪𝑉 |𝑈) is a variety, called an open subvariety of 𝑉. A regular map 𝜑 ∶ 𝑊 → 𝑉 is
an open immersion if 𝜑(𝑊) is open in 𝑉 and 𝜑 defines an isomorphism 𝑊 → 𝜑(𝑊) of

Example 5.12. The open subspace 𝑈 = 𝔸2 ∖ {(0, 0)} of 𝔸2 becomes an algebraic variety
varieties.

when endowed with the sheaf 𝒪𝔸2 |𝑈.

Closed subvarieties
Every closed subset 𝑍 of 𝑉 has a canonical structure of an algebraic variety. Define a
function 𝑓 on an open subset 𝑈 of 𝑍 to be regular if, for every 𝑃 ∈ 𝑈, there exists a germ
(𝑈 ′ , 𝑓 ′ ) of a regular function at 𝑃 on 𝑉 such that 𝑓 ′ |𝑈 ′ ∩ 𝑈 = 𝑓|𝑈 ′ ∩ 𝑈. This defines
a ringed structure 𝒪𝑍 on 𝑍. To show that (𝑍, 𝒪𝑍 ) is a variety it suffices to check that,
for every open affine 𝑈 ⊂ 𝑉, the ringed space (𝑈 ∩ 𝑍, 𝒪𝑍 |𝑈 ∩ 𝑍) is an affine algebraic
variety, but this is an easy exercise (Exercise 3-2 to be precise). Such a pair (𝑍, 𝒪𝑍 ) is
called a closed subvariety of 𝑉. A regular map 𝜑 ∶ 𝑊 → 𝑉 is a closed immersion if
𝜑(𝑊) is closed in 𝑉 and 𝜑 defines an isomorphism 𝑊 → 𝜑(𝑊) of varieties.
 f. Prevarieties obtained by patching 105

Subvarieties
A subset 𝑊 of a topological space 𝑉 is said to be locally closed if every point 𝑃 in 𝑊 has
an open neighbourhood 𝑈 in 𝑉 such that 𝑊 ∩ 𝑈 is closed in 𝑈. Equivalent conditions:
𝑊 is the intersection of an open and a closed subset of 𝑉; 𝑊 is open in its closure.
A locally closed subset 𝑊 of a variety 𝑉 has a canonical structure of an algebraic
variety. Write 𝑊 as the intersection 𝑊 = 𝑈 ∩ 𝑍 of an open and a closed subset. Now
𝑍 is a closed subvariety of 𝑉 and 𝑊 is an open subvariety of 𝑍. Alternatively, 𝑈 is an
open subvariety of 𝑉 and 𝑊 is closed subvariety of 𝑈. Either way, the structure on 𝑊 is
characterized by having the following property: the inclusion map 𝑊 → 𝑉 is regular,
and a map from a variety to 𝑊 is regular if and only if it is regular as a map to 𝑉.
With this structure, 𝑊 is called a subvariety of 𝑉. A regular map 𝜑 ∶ 𝑊 → 𝑉 is an
immersion if it induces an isomorphism of 𝑊 onto a subvariety of 𝑉. Every immersion

A subvariety of an affine variety is said to be quasi-affine. For example, 𝔸2 ∖ {(0, 0)}

is the composite of an open immersion with a closed immersion (in both orders).

is quasi-affine but not affine. Note that every quasi-affine variety is an open subvariety
of some affine variety.

Application
Proposition 5.14. A prevariety 𝑉 is separated if and only if two regular maps from a
prevariety to 𝑉 agree on the whole prevariety whenever they agree on a dense subset of it.

Proof. If 𝑉 is separated, then the set on which a pair of regular maps 𝜑1 , 𝜑2 ∶ 𝑍 ⇉ 𝑉

agree is closed (5.8), and so must be the whole of the 𝑍 if it contains a dense subset.
Conversely, consider a pair of maps 𝜑1 , 𝜑2 ∶ 𝑍 ⇉ 𝑉, and let 𝑆 be the subset of 𝑍 on
which they agree. We assume that 𝑉 has the property in the statement of the proposition,
and show that 𝑆 is closed. Let 𝑆̄ be the closure of 𝑆 in 𝑍. Then 𝑆̄ has the structure of a
closed prevariety of 𝑍 and the maps 𝜑1 |𝑆̄ and 𝜑2 |𝑆̄ are regular. Because they agree on a
dense subset of 𝑆̄ they agree on the whole of 𝑆,
̄ and so 𝑆 = 𝑆̄ is closed. 2

⋃
f. Prevarieties obtained by patching
Proposition 5.15. Suppose that the set 𝑉 is a finite union 𝑉 = 𝑖∈𝐼 𝑉𝑖 of subsets 𝑉𝑖 and
that each 𝑉𝑖 is equipped with ringed space structure. If the following “patching” condition

for all 𝑖, 𝑗, 𝑉𝑖 ∩ 𝑉𝑗 is open in both 𝑉𝑖 and 𝑉𝑗 and 𝒪𝑉𝑖 |𝑉𝑖 ∩ 𝑉𝑗 = 𝒪𝑉𝑗 |𝑉𝑖 ∩ 𝑉𝑗 ,
holds:

then there is a unique structure of a ringed space on 𝑉 for which

(a) each inclusion 𝑉𝑖 → 𝑉 is a homeomorphism of 𝑉𝑖 onto an open set, and
(b) for each 𝑖 ∈ 𝐼, 𝒪𝑉 |𝑉𝑖 = 𝒪𝑉𝑖 .
If every 𝑉𝑖 is an algebraic prevariety, then so also is 𝑉, and to give a regular map from 𝑉
to a prevariety 𝑊 is the same as giving a family of regular maps 𝜑𝑖 ∶ 𝑉𝑖 → 𝑊 such that
𝜑𝑖 |𝑉𝑖 ∩ 𝑉𝑗 = 𝜑𝑗 |𝑉𝑖 ∩ 𝑉𝑗 .

Proof. One checks easily that the subsets 𝑈 ⊂ 𝑉 such that 𝑈 ∩𝑉𝑖 is open for all 𝑖 are the
open subsets for a topology on 𝑉 satisfying (a), and that this is the only topology to satisfy
(a). Define 𝒪𝑉 (𝑈) to be the set of functions 𝑓 ∶ 𝑈 → 𝑘 such that 𝑓|𝑈 ∩ 𝑉𝑖 ∈ 𝒪𝑉𝑖 (𝑈 ∩ 𝑉𝑖 )
for all 𝑖. Again, one checks easily that 𝒪𝑉 is a sheaf of 𝑘-algebras satisfying (b), and that
it is the only such sheaf.
106 5. Algebraic Varieties

For the final statement, if each (𝑉𝑖 , 𝒪𝑉𝑖 ) is a finite union of open affines, so also
is (𝑉, 𝒪𝑉 ). Moreover, to give a map 𝜑 ∶ 𝑉 → 𝑊 amounts to giving a family of maps
𝜑𝑖 ∶ 𝑉𝑖 → 𝑊 such that 𝜑𝑖 |𝑉𝑖 ∩ 𝑉𝑗 = 𝜑𝑗 |𝑉𝑖 ∩ 𝑉𝑗 (obviously), and 𝜑 is regular if and only
𝜑|𝑉𝑖 is regular for each 𝑖. 2

Clearly, the 𝑉𝑖 may be separated without 𝑉 being separated (see, for example, 5.9).
In 5.29 below, we give a condition on an open affine covering of a prevariety sufficient to
ensure that the prevariety is separated.

g. Products of varieties
Let 𝑉 and 𝑊 be objects in a category 𝖢. A triple

(𝑉 × 𝑊, 𝑝 ∶ 𝑉 × 𝑊 → 𝑉, 𝑞 ∶ 𝑉 × 𝑊 → 𝑊)

is said to be the product of 𝑉 and 𝑊 if it has the following universal property: for every
pair of morphisms 𝑍 → 𝑉, 𝑍 → 𝑊 in 𝖢, there exists a unique morphism 𝑍 → 𝑉 × 𝑊

𝑍
making the diagram

←
∃!
→
←

𝑉 ← 𝑉×𝑊 → 𝑊
𝑝 𝑞
→
→

→ ←

commute. In other words, the triple is a product if the map

𝜑 ↦ (𝑝◦𝜑, 𝑞◦𝜑) ∶ Hom(𝑍, 𝑉 × 𝑊) → Hom(𝑍, 𝑉) × Hom(𝑍, 𝑊)

is a bijection. The product, if it exists, is uniquely determined up to a unique isomor-

phism.
For example, the product of two sets (in the category of sets) is the usual carte-
sian product of the sets, and the product of two topological spaces (in the category of
topological spaces) is the product of the underlying sets endowed with the product
topology.

the moment, that 𝑉 × 𝑊 exists. For any prevariety 𝑍, Mor(𝔸0 , 𝑍) is the underlying set
We shall show that products exist in the category of algebraic varieties. Suppose, for

of 𝑍; more precisely, for any 𝑧 ∈ 𝑍, the map 𝔸0 → 𝑍 with image 𝑧 is regular, and these
are all the regular maps (cf. 3.28). Thus, from the definition of products we have

(underlying set of 𝑉 × 𝑊) ≃ Mor(𝔸0 , 𝑉 × 𝑊)

≃ Mor(𝔸0 , 𝑉) × Mor(𝔸0 , 𝑊)
≃ (underlying set of 𝑉) × (underlying set of 𝑊).

Hence, our problem can be restated as follows: given two prevarieties 𝑉 and 𝑊, define
on the set 𝑉 × 𝑊 the structure of a prevariety such that
(a) the projection maps 𝑝, 𝑞 ∶ 𝑉 × 𝑊 ⇉ 𝑉, 𝑊 are regular, and
(b) a map 𝜑 ∶ 𝑇 → 𝑉 × 𝑊 of sets (with 𝑇 an algebraic prevariety) is regular if its
components 𝑝◦𝜑, 𝑞◦𝜑 are regular.
There can be at most one such structure on the set 𝑉 × 𝑊. We first consider the affine
case.
 g. Products of varieties 107

Products of affine varieties

Example 5.16. Let 𝔞 and 𝔟 be ideals in 𝑘[𝑋1 , … , 𝑋𝑚 ] and 𝑘[𝑋𝑚+1 , … , 𝑋𝑚+𝑛 ] respectively,
and let (𝔞, 𝔟) be the ideal in 𝑘[𝑋1 , … , 𝑋𝑚+𝑛 ] generated by the elements of 𝔞 and 𝔟. Then

𝑘[𝑋1 , … , 𝑋𝑚 ] 𝑘[𝑋𝑚+1 , … , 𝑋𝑚+𝑛 ] 𝑘[𝑋1 , … , 𝑋𝑚+𝑛 ]

there is an isomorphism

𝑓 ⊗ 𝑔 ↦ 𝑓𝑔 ∶ 𝔞 ⊗𝑘 → .
𝔟 (𝔞, 𝔟)
Again this comes down to checking that the natural map

Hom𝑘-alg (𝑘[𝑋1 , … , 𝑋𝑚+𝑛 ]∕(𝔞, 𝔟), 𝑅)

←
Hom𝑘-alg (𝑘[𝑋1 , … , 𝑋𝑚 ]∕𝔞, 𝑅) × Hom𝑘-alg (𝑘[𝑋𝑚+1 , … , 𝑋𝑚+𝑛 ]∕𝔟, 𝑅)
→

𝑉(𝔞, 𝔟) = zero set of (𝔞, 𝔟) in 𝑅𝑚+𝑛 ,

is a bijection. But the three sets are respectively

𝑉(𝔞) = zero set of 𝔞 in 𝑅𝑚 ,

𝑉(𝔟) = zero set of 𝔟 in 𝑅𝑛 ,
and so this is obvious.

The tensor product of two 𝑘-algebras 𝐴 and 𝐵 has the universal property to be a
product in the category of 𝑘-algebras, but with the arrows reversed. Because of the
category anti-equivalence (3.25), this shows that Spm(𝐴 ⊗𝑘 𝐵) will be the product of
Spm 𝐴 and Spm 𝐵 in the category of affine algebraic varieties once we have shown that
𝐴 ⊗𝑘 𝐵 is an affine 𝑘-algebra.

Proposition 5.17. Let 𝐴 and 𝐵 be 𝑘-algebras with 𝐴 finitely generated.

(a) If 𝐴 and 𝐵 are reduced, then so also is 𝐴 ⊗𝑘 𝐵.
(b) If 𝐴 and 𝐵 are integral domains, then so also is 𝐴 ⊗𝑘 𝐵.
∑𝑛
Proof. Let 𝛼 ∈ 𝐴 ⊗𝑘 𝐵. Then 𝛼 = 𝑖=1 𝑎𝑖 ⊗ 𝑏𝑖 , some 𝑎𝑖 ∈ 𝐴, 𝑏𝑖 ∈ 𝐵. If one of the 𝑏𝑗
∑𝑛−1
is a linear combination of the remaining 𝑏𝑖 , say, 𝑏𝑛 = 𝑖=1 𝑐𝑖 𝑏𝑖 , 𝑐𝑖 ∈ 𝑘, then, using the
bilinearity of ⊗, we find that

∑
𝑛−1
∑
𝑛−1
∑
𝑛−1
𝛼= 𝑎𝑖 ⊗ 𝑏𝑖 + 𝑐𝑖 𝑎𝑛 ⊗ 𝑏𝑖 = (𝑎𝑖 + 𝑐𝑖 𝑎𝑛 ) ⊗ 𝑏𝑖 .
𝑖=1 𝑖=1 𝑖=1

Thus we may suppose that in the original expression of 𝛼, the 𝑏𝑖 are linearly independent
over 𝑘.
Now assume 𝐴 and 𝐵 to be reduced, and suppose that 𝛼 is nilpotent. Let 𝔪 be a
maximal ideal of 𝐴. From 𝑎 ↦ 𝑎̄ ∶ 𝐴 → 𝐴∕𝔪 = 𝑘 we obtain homomorphisms
≃
𝑎 ⊗ 𝑏 ↦ 𝑎̄ ⊗ 𝑏 ↦ 𝑎𝑏
̄ ∶ 𝐴 ⊗𝑘 𝐵 → 𝑘 ⊗𝑘 𝐵 → 𝐵.
∑
The image 𝑎̄ 𝑖 𝑏𝑖 of 𝛼 under this homomorphism is a nilpotent element of 𝐵, and hence
is zero (because 𝐵 is reduced). As the 𝑏𝑖 are linearly independent over 𝑘, this means that
the 𝑎̄ 𝑖 are all zero. Thus, the 𝑎𝑖 lie in all maximal ideals 𝔪 of 𝐴, and so are zero (see
2.18). Hence 𝛼 = 0, and we have shown that 𝐴 ⊗𝑘 𝐵 is reduced.
108 5. Algebraic Varieties

Now assume that 𝐴 and 𝐵 are integral domains, and let 𝛼, 𝛼′ ∈ 𝐴 ⊗𝑘 𝐵 be such
∑ ∑
that 𝛼𝛼′ = 0. As before, we can write 𝛼 = 𝑎𝑖 ⊗ 𝑏𝑖 and 𝛼′ = 𝑎𝑖′ ⊗ 𝑏𝑖′ with the sets
{𝑏1 , 𝑏2 , …} and {𝑏1′ , 𝑏2′ , …} each linearly independent over 𝑘. For each maximal ideal 𝔪 of
∑ ∑ ∑ ∑
𝐴, we know ( 𝑎̄ 𝑖 𝑏𝑖 )( 𝑎̄ 𝑖′ 𝑏𝑖′ ) = 0 in 𝐵, and so either ( 𝑎̄ 𝑖 𝑏𝑖 ) = 0 or ( 𝑎̄ 𝑖′ 𝑏𝑖′ ) = 0. Thus
either all the 𝑎𝑖 ∈ 𝔪 or all the 𝑎𝑖′ ∈ 𝔪. This shows that

spm(𝐴) = 𝑉(𝑎1 , … , 𝑎𝑚 ) ∪ 𝑉(𝑎1′ , … , 𝑎𝑛′ ).

As spm(𝐴) is irreducible (see 2.27), it follows that spm(𝐴) equals either 𝑉(𝑎1 , … , 𝑎𝑚 ) or
𝑉(𝑎1′ , … , 𝑎𝑛′ ). In the first case 𝛼 = 0, and in the second 𝛼′ = 0. 2

Remark 5.18. The proof of 5.17 fails when 𝑘 is not algebraically closed, because then
𝐴∕𝔪 may be a finite extension of 𝑘 over which the 𝑏𝑖 become linearly dependent. The

(a) Suppose that 𝑘 is nonperfect of characteristic 𝑝, so that there exists an element 𝛼

following examples show that the statement of 5.17 also fails in this case.

in an algebraic closure of 𝑘 such that 𝛼 ∉ 𝑘 but 𝛼𝑝 ∈ 𝑘. Let 𝑘′ = 𝑘[𝛼], and let 𝛼𝑝 = 𝑎.

Then (𝛼 ⊗ 1 − 1 ⊗ 𝛼) ≠ 0 in 𝑘′ ⊗𝑘 𝑘 ′ (in fact, the elements 𝛼𝑖 ⊗ 𝛼𝑗 , 0 ≤ 𝑖, 𝑗 ≤ 𝑝 − 1,
form a basis for 𝑘 ′ ⊗𝑘 𝑘 ′ as a 𝑘-vector space), but

(𝛼 ⊗ 1 − 1 ⊗ 𝛼)𝑝 = (𝑎 ⊗ 1 − 1 ⊗ 𝑎)
= (1 ⊗ 𝑎 − 1 ⊗ 𝑎) (because 𝑎 ∈ 𝑘)
= 0.

Thus 𝑘′ ⊗𝑘 𝑘′ is not reduced, even though 𝑘 ′ is a field.

(b) Let 𝐾 be a finite separable extension of 𝑘 and let 𝐸 be a second field containing 𝑘.
By the primitive element theorem (FT, 5.1),

𝐾 = 𝑘[𝛼] = 𝑘[𝑋]∕(𝑓(𝑋)),

for some 𝛼 ∈ 𝐾 and its minimal polynomial 𝑓(𝑋). Assume that 𝐸 is large enough to split
∏
𝑓, say, 𝑓(𝑋) = 𝑖 (𝑋 − 𝛼𝑖 ) with 𝛼𝑖 ∈ 𝐸. Because 𝐾∕𝑘 is separable, the 𝛼𝑖 are distinct,
and so

𝐸 ⊗𝑘 𝐾 ≃ 𝐸[𝑋]∕(𝑓(𝑋))
∏
≃ 𝐸[𝑋]∕(𝑋 − 𝛼𝑖 ),
(1.58(b))
(1.1)

which is not an integral domain. For example,

ℂ ⊗ℝ ℂ ≃ ℂ[𝑋]∕(𝑋 − 𝑖) × ℂ[𝑋]∕(𝑋 + 𝑖) ≃ ℂ × ℂ.

Definition 5.19. The product of the affine varieties 𝑉 and 𝑊 is

The proposition allows us to make the following definition.

(𝑉 × 𝑊, 𝒪𝑉×𝑊 ) = Spm(𝑘[𝑉] ⊗𝑘 𝑘[𝑊])

with the projection maps 𝑝, 𝑞 ∶ 𝑉 × 𝑊 → 𝑉, 𝑊 defined by the homomorphisms

𝑓 ↦ 𝑓 ⊗ 1 ∶ 𝑘[𝑉] → 𝑘[𝑉] ⊗𝑘 𝑘[𝑊]

𝑔 ↦ 1 ⊗ 𝑔 ∶ 𝑘[𝑊] → 𝑘[𝑉] ⊗𝑘 𝑘[𝑊].

Proposition 5.20. Let 𝑉 and 𝑊 be affine varieties.

 g. Products of varieties 109

(a) The variety (𝑉 × 𝑊, 𝒪𝑉×𝑊 ) is the product of (𝑉, 𝒪𝑉 ) and (𝑊, 𝒪𝑊 ) in the category
of affine algebraic varieties; in particular, the set 𝑉 × 𝑊 is the product of the sets 𝑉
and 𝑊 and 𝑝 and 𝑞 are the projection maps.
(b) If 𝑉 and 𝑊 are irreducible, then so also is 𝑉 × 𝑊.

Proof. (a) As noted at the start of the subsection, the first statement follows from
5.17(a), and the second statement then follows by the argument on p. 106.
(b) This follows from 5.17(b) and 2.27. 2

Corollary 5.21. Let 𝑉 and 𝑊 be affine varieties. For every prevariety 𝑇, a map 𝜑 ∶ 𝑇 →
𝑉 × 𝑊 is regular if 𝑝◦𝜑 and 𝑞◦𝜑 are regular.

Proof. If 𝑝◦𝜑 and 𝑞◦𝜑 are regular, then 5.20 implies that 𝜑 is regular when restricted
to any open affine of 𝑇, which implies that it is regular on 𝑇. 2

The corollary shows that 𝑉 × 𝑊 is the product of 𝑉 and 𝑊 in the category of prevari-

Example 5.22. (a) It follows from 1.57 that 𝔸𝑚+𝑛 endowed with the projection maps
eties (hence also in the categories of varieties).

𝑝 𝑞 𝑝(𝑎1 , … , 𝑎𝑚+𝑛 ) = (𝑎1 , … , 𝑎𝑚 )

𝔸𝑚 ←
, 𝔸𝑚+𝑛 ,
→ 𝔸𝑛 , {
𝑞(𝑎1 , … , 𝑎𝑚+𝑛 ) = (𝑎𝑚+1 , … , 𝑎𝑚+𝑛 ),

is the product of 𝔸𝑚 and 𝔸𝑛 .

(b) It follows from 5.16 that
𝑝 𝑞
𝑉(𝔞) ←
, 𝑉(𝔞, 𝔟) ,
→ 𝑉(𝔟)

☡ When 𝑉 and 𝑊 have dimension > 0, the topology on 𝑉 × 𝑊 is strictly finer than
is the product of 𝑉(𝔞) and 𝑉(𝔟).

product topology. For example, for the product topology on 𝔸2 = 𝔸1 × 𝔸1 , every proper
closed subset is contained in a finite union of vertical and horizontal lines, whereas 𝔸2
has many more closed subsets (see 2.68).

Products in general
We now define the product of two algebraic prevarieties 𝑉 and 𝑊.
⋃
Write 𝑉 as a union of open affines 𝑉 = 𝑉𝑖 , and note that 𝑉 can be regarded as
the variety obtained by patching the (𝑉𝑖 , 𝒪𝑉𝑖 ); in particular, this covering satisfies the
⋃
patching condition (5.15). Similarly, write 𝑊 as a union of open affines 𝑊 = 𝑊𝑗 .
⋃
𝑉×𝑊 = 𝑉𝑖 × 𝑊𝑗
Then

and the (𝑉𝑖 × 𝑊𝑗 , 𝒪𝑉𝑖 ×𝑊𝑗 ) satisfy the patching condition. Therefore, we can define
(𝑉 × 𝑊, 𝒪𝑉×𝑊 ) to be the variety obtained by patching the (𝑉𝑖 × 𝑊𝑗 , 𝒪𝑉𝑖 ×𝑊𝑗 ).

Proposition 5.23. With the sheaf of 𝑘-algebras 𝒪𝑉×𝑊 just defined, 𝑉 × 𝑊 becomes the
product of 𝑉 and 𝑊 in the category of prevarieties. In particular, the structure of prevariety
⋃ ⋃
on 𝑉 × 𝑊 defined by the coverings 𝑉 = 𝑉𝑖 and 𝑊 = 𝑊𝑗 are independent of the
coverings.
110 5. Algebraic Varieties

Proof. Let 𝑇 be a prevariety, and let 𝜑 ∶ 𝑇 → 𝑉 × 𝑊 be a map of sets such that 𝑝◦𝜑 and
𝑞◦𝜑 are regular. Then 5.21 implies that the restriction of 𝜑 to 𝜑−1 (𝑉𝑖 × 𝑊𝑗 ) is regular.
As these open sets cover 𝑇, this shows that 𝜑 is regular.

Proposition 5.24. If 𝑉 and 𝑊 are separated, then so also is 𝑉 × 𝑊.

2

Proof. Let 𝜑1 , 𝜑2 be two regular maps 𝑈 → 𝑉 × 𝑊. The set where 𝜑1 , 𝜑2 agree is the
intersection of the sets where 𝑝◦𝜑1 , 𝑝◦𝜑2 and 𝑞◦𝜑1 , 𝑞◦𝜑2 agree, which is closed.

Proposition 5.25. If 𝑉 and 𝑊 are connected, then so also is 𝑉 × 𝑊.

2

Proof. For 𝑣0 ∈ 𝑉, we have continuous maps

𝑊 ≃ 𝑣0 × 𝑊 ← → 𝑉 × 𝑊.
closed

Similarly, for 𝑤0 ∈ 𝑊, we have continuous maps

𝑉 ≃ 𝑉 × 𝑤0 ← → 𝑉 × 𝑊.
closed

The images of 𝑉 and 𝑊 in 𝑉 × 𝑊 intersect in (𝑣0 , 𝑤0 ) and are connected, which shows
that (𝑣0 , 𝑤) and and (𝑣, 𝑤0 ) lie in the same connected component of 𝑉 × 𝑊 for all 𝑣 ∈ 𝑉
and 𝑤 ∈ 𝑊. Since 𝑣0 and 𝑤0 were arbitrary, this shows that any two points lie in the
same connected component. 2

Group varieties
A group variety is an algebraic variety 𝐺 together with a group structure 𝑚 (map of
sets 𝐺 × 𝐺 → 𝐺 satisfying the group axioms) such that the maps

𝑚 ∶ 𝐺 × 𝐺 → 𝐺, inv ∶ 𝐺 → 𝐺, 𝑒 ∶ 𝔸0 → 𝐺

are regular. A homomorphism of group varieties is a regular map that is also a homo-
morphism of groups.

𝑘[𝑋11 , 𝑋12 , … , 𝑋𝑛𝑛 ]

The algebraic variety,

SL𝑛 = Spm
{ (det(𝑋𝑖𝑗 ) − 1)
SL𝑛 (𝑘) = {𝑀 ∈ 𝑀𝑛 (𝑘) ∣ det 𝑀 = 1}

∑𝑛
becomes a group variety when endowed with its usual group structures. Matrix multi-

(𝑎𝑖𝑗 ) ⋅ (𝑏𝑖𝑗 ) = (𝑐𝑖𝑗 ), 𝑐𝑖𝑗 = 𝑙=1 𝑎𝑖𝑙 𝑏𝑙𝑗 ,

plication

𝐴−1 as polynomials in the entries of 𝐴. The only affine group varieties of dimension 1
is given by polynomials, and Cramer’s rule gives an explicit expression of the entries of

over 𝑘 are
𝔾𝑚 = Spm 𝑘[𝑋, 𝑋 −1 ] and 𝔾𝑎 = Spm 𝑘[𝑋].
Every finite group 𝑁 can be made into a group variety by setting

𝑁 = Spm(𝐴)

with 𝐴 the 𝑘-algebra of all maps 𝑓 ∶ 𝑁 → 𝑘.

 h. The separation axiom revisited 111

h. The separation axiom revisited

By way of motivation, consider a topological space 𝑉 and the diagonal ∆ ⊂ 𝑉 × 𝑉,
∆ = (𝑥, 𝑥) ∣ 𝑥 ∈ 𝑉. If ∆ is closed for the product topology, then every pair of points
(𝑥, 𝑦) ∉ ∆ has an open neighbourhood 𝑈 × 𝑈 ′ such that (𝑈 × 𝑈 ′ ) ∩ ∆ = ∅. In other
def

words, if 𝑥 and 𝑦 are distinct points in 𝑉, then there are open neighbourhoods 𝑈 and 𝑈 ′
of 𝑥 and 𝑦 respectively such that 𝑈 ∩ 𝑈 ′ = ∅. Thus 𝑉 is Hausdorff. Conversely, if 𝑉 is
Hausdorff, the reverse argument shows that ∆ is closed.
For a variety 𝑉, we let ∆ = ∆𝑉 (the diagonal) be the subset {(𝑣, 𝑣) ∣ 𝑣 ∈ 𝑉} of 𝑉 × 𝑉.

Proposition 5.26. An algebraic prevariety 𝑉 is separated if and only if ∆𝑉 is closed.4

Proof. We shall use the criterion 5.8: 𝑉 is separated if and only if regular regular maps
to 𝑉 agree on a closed subset of their source.
Suppose that ∆𝑉 is closed. The map

(𝜑1 , 𝜑2 ) ∶ 𝑍 → 𝑉 × 𝑉, 𝑧 ↦ (𝜑1 (𝑧), 𝜑2 (𝑧))

is regular because its components 𝜑1 and 𝜑2 are regular (definition of a product). In

particular, it is continuous, and so (𝜑1 , 𝜑2 )−1 (∆𝑉 ) is closed, but this is exactly the subset
on which 𝜑1 and 𝜑2 agree.
Conversely, ∆𝑉 is the set on which the two projection maps 𝑉 × 𝑉 → 𝑉 agree, and
so it is closed if 𝑉 is separated. 2

Corollary 5.27. For any prevariety 𝑉, the diagonal is a locally closed subset of 𝑉 × 𝑉.

Proof. Let 𝑃 ∈ 𝑉, and let 𝑈 be an open affine neighbourhood of 𝑃. Then 𝑈 × 𝑈 is an

open neighbourhood of (𝑃, 𝑃) in 𝑉 × 𝑉, and ∆𝑉 ∩ (𝑈 × 𝑈) = ∆𝑈 , which is closed in
𝑈 × 𝑈 because 𝑈 is separated (5.6). 2

𝑊
Thus ∆𝑉 is always a subvariety of 𝑉 × 𝑉, and it 𝛤𝜑
is closed if and only if 𝑉 is separated. The graph
𝛤𝜑 of a regular map 𝜑 ∶ 𝑉 → 𝑊 is defined to be
𝜑(𝑣) (𝑣, 𝜑(𝑣))
{(𝑣, 𝜑(𝑣)) ∈ 𝑉 × 𝑊 ∣ 𝑣 ∈ 𝑉}.

𝑣 𝑉

Corollary 5.28. For any morphism 𝜑 ∶ 𝑉 → 𝑊 of prevarieties, the graph 𝛤𝜑 of 𝜑 is

locally closed in 𝑉 × 𝑊, and it is closed if 𝑊 is separated. The map 𝑣 ↦ (𝑣, 𝜑(𝑣)) is an
isomorphism of 𝑉 onto 𝛤𝜑 (as algebraic prevarieties).

(𝑣, 𝑤) ↦ (𝜑(𝑣), 𝑤) ∶ 𝑉 × 𝑊 → 𝑊 × 𝑊
Proof. The map

4
Thus does not contradict the fact that 𝑉 is not Hausdorff because the Zariski topology on 𝑉 × 𝑉 is not
the product topology.
112 5. Algebraic Varieties

is regular because its composites with the projections are 𝜑 and id𝑊 , which are regular.
In particular, it is continuous, and as 𝛤𝜑 is the inverse image of ∆𝑊 under this map, this

𝑝
map 𝛤𝜑 → 𝑉 × 𝑊 → 𝑉 is an inverse to 𝑣 ↦ (𝑣, 𝜑(𝑣)) ∶ 𝑉 → 𝛤𝜑 .
proves the first statement. The second statement follows from the fact that the regular
2

Theorem 5.29. The following three conditions on a prevariety 𝑉 are equivalent:

(a) 𝑉 is separated;
(b) for every pair of open affines 𝑈 and 𝑈 ′ in 𝑉, 𝑈 ∩ 𝑈 ′ is an open affine and the map

𝑓 ⊗ 𝑔 ↦ 𝑓|𝑈∩𝑈 ′ ⋅ 𝑔|𝑈∩𝑈 ′ ∶ 𝑘[𝑈] ⊗𝑘 𝑘[𝑈 ′ ] → 𝑘[𝑈 ∩ 𝑈 ′ ]

(c) the condition in (b) holds for the sets in some open affine covering of 𝑉.
is surjective;

Proof. Let 𝑈 and 𝑈 ′ be open affines in 𝑉. We shall prove that

(i) if ∆𝑉 is closed then 𝑈 ∩ 𝑈 ′ affine,
(ii) when 𝑈 ∩ 𝑈 ′ is affine,

(𝑈 × 𝑈 ′ ) ∩ ∆𝑉 is closed ⇐⇒ 𝑘[𝑈] ⊗𝑘 𝑘[𝑈 ′ ] → 𝑘[𝑈 ∩ 𝑈 ′ ] is surjective.

for the sets in some open affine covering (𝑈𝑖 )𝑖∈𝐼 of 𝑉. Then (𝑈𝑖 × 𝑈𝑗 )(𝑖,𝑗)∈𝐼×𝐼 is an open
Assume (a); then these statements imply (b). Assume that the condition in (b) holds

affine covering of 𝑉 × 𝑉, and ∆𝑉 ∩ (𝑈𝑖 × 𝑈𝑗 ) is closed in 𝑈𝑖 × 𝑈𝑗 for each pair (𝑖, 𝑗),

Proof of (i): The graph of the inclusion 𝑈 ∩ 𝑈 ′ → 𝑉 is the subset (𝑈 × 𝑈 ′ ) ∩ ∆𝑉

which implies (a). Thus, the statements (i) and (ii) imply the theorem.

of (𝑈 ∩ 𝑈 ′ ) × 𝑉. If ∆𝑉 is closed, then (𝑈 × 𝑈 ′ ) ∩ ∆𝑉 is a closed subvariety of an affine

variety, and hence is affine. Now 5.28 implies that 𝑈 ∩ 𝑈 ′ is affine.
Proof of (ii): Assume that 𝑈 ∩ 𝑈 ′ is affine. Then

(𝑈 × 𝑈 ′ ) ∩ ∆𝑉 is closed in 𝑈 × 𝑈 ′
⇐⇒ 𝑣 ↦ (𝑣, 𝑣) ∶ 𝑈 ∩ 𝑈 ′ → 𝑈 × 𝑈 ′ is a closed immersion
⇐⇒ 𝑘[𝑈 × 𝑈 ′ ] → 𝑘[𝑈 ∩ 𝑈 ′ ] is surjective (3.34).

Since 𝑘[𝑈 × 𝑈 ′ ] = 𝑘[𝑈] ⊗𝑘 𝑘[𝑈 ′ ], this completes the proof of (ii). 2

In more down-to-earth terms, condition (b) says that 𝑈 ∩ 𝑈 ′ is affine and every
regular function on 𝑈 ∩ 𝑈 ′ is a sum of functions of the form 𝑃 ↦ 𝑓(𝑃)𝑔(𝑃) with 𝑓 and
𝑔 regular functions on 𝑈 and 𝑈 ′ .

Example 5.30. (a) Let 𝑉 = ℙ1 , and let 𝑈0 and 𝑈1 be the standard open subsets (see
5.3). Then 𝑈0 ∩ 𝑈1 = 𝔸1 ∖ {0}, and the maps on rings corresponding to the inclusions
𝑈0 ∩ 𝑈1 → 𝑈𝑖 are

𝑓(𝑋) ↦ 𝑓(𝑋) ∶ 𝑘[𝑋] → 𝑘[𝑋, 𝑋 −1 ]

𝑓(𝑋) ↦ 𝑓(𝑋 −1 ) ∶ 𝑘[𝑋] → 𝑘[𝑋, 𝑋 −1 ].

Thus the sets 𝑈0 and 𝑈1 satisfy the condition in (b).

 i. Fibred products 113

(b) Let 𝑉 be 𝔸1 with the origin doubled (see 5.9), and let 𝑈 and 𝑈 ′ be the upper
and lower copies of 𝔸1 in 𝑉. Then 𝑈 ∩ 𝑈 ′ = (𝔸1 ∖ 0) is affine, but the maps on rings
corresponding to the inclusions 𝑈 ∩ 𝑈 ′ → 𝑈𝑖 are
𝑋 ↦ 𝑋 ∶ 𝑘[𝑋] → 𝑘[𝑋, 𝑋 −1 ]
𝑋 ↦ 𝑋 ∶ 𝑘[𝑋] → 𝑘[𝑋, 𝑋 −1 ].
Thus the sets 𝑈 and 𝑈 ′ fail the condition in (b) (and 𝑉 is not separated).
(c) Let 𝑉 be 𝔸2 with the origin doubled, and let 𝑈 and 𝑈 ′ be the upper and lower
copies of 𝔸2 in 𝑉. Then 𝑈 ∩ 𝑈 ′ is not affine (see 3.33).

i. Fibred products
Let 𝜑 ∶ 𝑉 → 𝑆 and 𝜓 ∶ 𝑊 → 𝑆 be regular maps of algebraic varieties. The set
𝑉 ×𝑆 𝑊 = {(𝑣, 𝑤) ∈ 𝑉 × 𝑊 ∣ 𝜑(𝑣) = 𝜓(𝑤)}
def

is closed in 𝑉 ×𝑊, because it is the set where 𝜑◦𝑝 and 𝜓◦𝑞 agree, and so it has a canonical
structure of an algebraic variety (see p. 104). The algebraic variety 𝑉 ×𝑆 𝑊 is called
the fibred product of 𝑉 and 𝑊 over 𝑆. Note that if 𝑆 consists of a single point, then
𝑉 ×𝑆 𝑊 = 𝑉 × 𝑊.
Writing 𝜑′ for the map (𝑣, 𝑤) ↦ 𝑤 ∶ 𝑉 ×𝑆 𝑊 → 𝑊 and 𝜓 ′ for (𝑣, 𝑤) ↦ 𝑣 ∶ 𝑉 ×𝑆 𝑊 →
𝑉, we get a commutative diagram:

𝑉 ×𝑆 𝑊 → 𝑊
← 𝜑′

𝜓′ 𝜓
←
←

𝑉 → 𝑆.
𝜑
→
→

The system (𝑉 ×𝑆 𝑊, 𝜑′ , 𝜓′ ) has the following universal property: for any regular maps
𝛼 ∶ 𝑇 → 𝑉, 𝛽 ∶ 𝑇 → 𝑊 such that 𝜑𝛼 = 𝜓𝛽, there is a unique regular map (𝛼, 𝛽) ∶ 𝑇 →
𝑉 ×𝑆 𝑊 such that the following diagram

𝑇
𝛽
(𝛼, 𝛽)

𝜑′
𝑉 ×𝑆 𝑊 𝑊
𝛼

𝜓′ 𝜓

𝜑
𝑉 𝑆

Hom(𝑇, 𝑉 ×𝑆 𝑊) ≃ Hom(𝑇, 𝑉) ×Hom(𝑇,𝑆) Hom(𝑇, 𝑊).

commutes. In other words,

Indeed, there is a unique such map of sets, namely, 𝑡 ↦ (𝛼(𝑡), 𝛽(𝑡)), which is regular
because it is as a map into 𝑉 × 𝑊.
The map 𝜑′ in the above diagrams is called the base change of 𝜑 with respect to 𝜓.
For any point 𝑃 ∈ 𝑆, the base change of 𝜑 ∶ 𝑉 → 𝑆 with respect to 𝑃 → 𝑆 is the map
𝜑−1 (𝑃) → 𝑃 induced by 𝜑, which is called the fibre of 𝑉 over 𝑃.
114 5. Algebraic Varieties

Example 5.31. If 𝑓 ∶ 𝑉 → 𝑆 is a regular map and 𝑈 is a subvariety of 𝑆, then 𝑈 ×𝑆 𝑉 is

the inverse image of 𝑈 in 𝑉, 𝜑−1 (𝑈) = 𝑈 ×𝑆 𝑉.

Notes
5.32. Since a tensor product of rings 𝐴 ⊗𝑅 𝐵 has the opposite universal property to that
of a fibred product, one might hope that

Spm(𝐴) ×Spm(𝑅) Spm(𝐵) = Spm(𝐴 ⊗𝑅 𝐵).

??

This is true if 𝐴 ⊗𝑅 𝐵 is an affine 𝑘-algebra, but in general it may have nonzero nilpotent
elements. For example, let 𝑘 have characteristic 𝑝, let 𝑅 = 𝑘[𝑋], and consider the
𝑘[𝑋]-algebras
𝑘[𝑋] → 𝑘, 𝑋↦𝑎
{
𝑘[𝑋] → 𝑘[𝑋], 𝑋 ↦ 𝑋 𝑝 .

𝐴 ⊗𝑅 𝐵 ≃ 𝑘 ⊗𝑘[𝑋 𝑝 ] 𝑘[𝑋] ≃ 𝑘[𝑋]∕(𝑋 𝑝 − 𝑎),

Then

which contains the nilpotent element 𝑥 − 𝑎1∕𝑝 .

The correct statement is

Spm(𝐴) ×Spm(𝑅) Spm(𝐵) ≃ Spm(𝐴 ⊗𝑅 𝐵∕𝔑), (25)

where 𝔑 is the ideal of nilpotent elements in 𝐴 ⊗𝑅 𝐵. To prove this, note that for any
algebraic variety 𝑇,

Mor(𝑇, Spm(𝐴 ⊗𝑅 𝐵∕𝔑)) ≃ Hom(𝐴 ⊗𝑅 𝐵∕𝔑, 𝒪𝑇 (𝑇))

≃ Hom(𝐴 ⊗𝑅 𝐵, 𝒪𝑇 (𝑇))
(5.11)

≃ Hom(𝐴, 𝒪𝑇 (𝑇)) × Hom(𝐵, 𝒪𝑇 (𝑇))

Hom(𝑅,𝒪𝑇 (𝑇))
≃ Mor(𝑇, Spm(𝐴)) × Mor(𝑇, Spm(𝐵)) (5.11).
Mor(𝑇,Spm(𝑅))

For the second isomorphism we used that the ring 𝒪𝑇 (𝑇) is reduced, and for the third
isomorphism, we used the universal property of 𝐴 ⊗𝑅 𝐵.

5.33. Fibred products exist also for prevarieties. In this case, 𝑉 ×𝑆 𝑊 is only locally
closed in 𝑉 × 𝑊.

Aside 5.34. Fibred products may differ depending on whether we are working in the category
of algebraic varieties or algebraic schemes. For example,

Spec(𝐴) ×Spec(𝑅) Spec(𝐵) ≃ Spec(𝐴 ⊗𝑅 𝐵)

𝜑
in the category of schemes. Consider the map 𝑥 ↦ 𝑥 2 ∶ 𝔸1 ,→ 𝔸1 (see 5.49). The fibre 𝜑−1 (𝑎)
consists of two points if 𝑎 ≠ 0, and one point if 𝑎 = 0. Thus 𝜑−1 (0) = Spm(𝑘[𝑋]∕(𝑋)). However,
the scheme-theoretic fibre is Spec(𝑘[𝑋]∕(𝑋 2 )), which reflects the fact that 0 is “doubled” in the
fibre over 0. This will be explained in Chapter 10.
 j. Dimension 115

j. Dimension
Recall p. 45 that, in an irreducible topological space, every nonempty open subset is

Let 𝑉 be an irreducible algebraic variety 𝑉, and let 𝑈 and 𝑈 ′ be nonempty open

dense and irreducible.

affines in 𝑉. Then 𝑈 ∩ 𝑈 ′ is also a nonempty open affine (5.29), which is dense in 𝑈,

and so the restriction map 𝒪𝑉 (𝑈) → 𝒪𝑉 (𝑈 ∩ 𝑈 ′ ) is injective. Therefore

𝑘[𝑈] ⊂ 𝑘[𝑈 ∩ 𝑈 ′ ] ⊂ 𝑘(𝑈),

where 𝑘(𝑈) is the field of fractions of 𝑘[𝑈], and so 𝑘(𝑈) is also the field of fractions of
𝑘[𝑈 ∩ 𝑈 ′ ] and of 𝑘[𝑈 ′ ].5 Thus, attached to 𝑉 there is a field 𝑘(𝑉), called the function
field of 𝑉 or the field of rational functions on 𝑉, which is the field of fractions of 𝑘[𝑈]
for any open affine 𝑈 in 𝑉. The dimension of 𝑉 is defined to be the transcendence degree
of 𝑘(𝑉) over 𝑘. Note the dim(𝑉) = dim(𝑈) for any open subset 𝑈 of 𝑉. In particular,
dim(𝑉) = dim(𝑈) for 𝑈 an open affine in 𝑉. It follows that some of the results in §2
carry over — for example, if 𝑍 is a proper closed subvariety of 𝑉, then dim(𝑍) < dim(𝑉).

Proposition 5.35. Let 𝑉 and 𝑊 be irreducible varieties. Then

dim(𝑉 × 𝑊) = dim(𝑉) + dim(𝑊).

Proof. We may suppose 𝑉 and 𝑊 to be affine. Write

𝑘[𝑉] = 𝑘[𝑥1 , … , 𝑥𝑚 ]
𝑘[𝑊] = 𝑘[𝑦1 , … , 𝑦𝑛 ],

where the 𝑥 and 𝑦 have been chosen so that {𝑥1 , … , 𝑥𝑑 } and {𝑦1 , … , 𝑦𝑒 } are maximal
algebraically independent sets of elements of 𝑘[𝑉] and 𝑘[𝑊]. Then {𝑥1 , … , 𝑥𝑑 } and
{𝑦1 , … , 𝑦𝑒 } are transcendence bases of 𝑘(𝑉) and 𝑘(𝑊) (see 1.63), and so dim(𝑉) = 𝑑 and
dim(𝑊) = 𝑒. Now6

𝑘[𝑉 × 𝑊] = 𝑘[𝑉] ⊗𝑘 𝑘[𝑊] ⊃ 𝑘[𝑥1 , … , 𝑥𝑑 ] ⊗𝑘 𝑘[𝑦1 , … , 𝑦𝑒 ],

def

which is a polynomial ring in the symbols 𝑥1 ⊗ 1, … , 𝑥𝑑 ⊗ 1, 1 ⊗ 𝑦1 , … , 1 ⊗ 𝑦𝑒 (see

1.57). In particular, the elements 𝑥1 ⊗ 1, … , 𝑥𝑑 ⊗ 1, 1 ⊗ 𝑦1 , … , 1 ⊗ 𝑦𝑒 are algebraically
independent in 𝑘[𝑉] ⊗𝑘 𝑘[𝑊]. Obviously 𝑘[𝑉 × 𝑊] is generated as a 𝑘-algebra by the
elements 𝑥𝑖 ⊗ 1, 1 ⊗ 𝑦𝑗 , 1 ≤ 𝑖 ≤ 𝑚, 1 ≤ 𝑗 ≤ 𝑛, and all of them are algebraic over
𝑘[𝑥1 , … , 𝑥𝑑 ] ⊗𝑘 𝑘[𝑦1 , … , 𝑦𝑒 ]. Thus the transcendence degree of 𝑘(𝑉 × 𝑊) is 𝑑 + 𝑒. 2

We extend the definition of dimension to an arbitrary variety 𝑉 as follows. An

⋃
Consequently (see 2.31), 𝑉 is a finite union 𝑉 = 𝑉𝑖 of its irreducible components,
algebraic variety is a finite union of noetherian topological spaces, and so is noetherian.

and we define dim(𝑉) = max dim(𝑉𝑖 ). When all the irreducible components of 𝑉 have
dimension 𝑛, 𝑉 is said to be pure of dimension 𝑛 (or to be of pure dimension 𝑛).
If 𝐴 is an integral domain and 𝑓 is a nonzero element of 𝐴, then 𝐴 ⊂ 𝐴𝑓 ⊂ FF(𝐴) and 𝐴 and 𝐴𝑓 have
the same field of fractions. Thus 𝑘(𝑈) does not change when we shrink 𝑈.
5

In general, it is not true that if 𝑀 ′ and 𝑁 ′ are 𝑅-submodules of 𝑀 and 𝑁, then 𝑀 ′ ⊗𝑅 𝑁 ′ is an 𝑅-

submodule of 𝑀 ⊗𝑅 𝑁. However, this is true if 𝑅 is a field, because then 𝑀 ′ and 𝑁 ′ will be direct summands
6

of 𝑀 and 𝑁, and tensor products preserve direct summands.

116 5. Algebraic Varieties

Proposition 5.36. Let 𝑉 and 𝑊 be closed subvarieties of 𝔸𝑛 ; for any (nonempty) irre-
ducible component 𝑍 of 𝑉 ∩ 𝑊,
dim(𝑍) ≥ dim(𝑉) + dim(𝑊) − 𝑛;

codim(𝑍) ≤ codim(𝑉) + codim(𝑊).

that is,

Proof. In the course of the proof of Theorem 5.29, we saw that 𝑉 ∩ 𝑊 is isomorphic
to ∆ ∩ (𝑉 × 𝑊), and this is defined by the 𝑛 equations 𝑋𝑖 = 𝑌𝑖 in 𝑉 × 𝑊. Thus the
statement follows from 3.45. 2

𝑋2 + 𝑌2 = 𝑍2
Remark 5.37. (a) The subvariety

{
𝑍 =0
of 𝔸3 is the curve 𝑋 2 + 𝑌 2 = 0, which is the pair of lines 𝑌 = ±𝑖𝑋 if 𝑘 = ℂ; in particular,
the codimension is 2. Note however, that real locus is {(0, 0)}, which has codimension 3.
Thus, Proposition 5.36 becomes false if one looks only at real points (and the pictures

(b) Proposition 5.36 becomes false if 𝔸𝑛 is replaced by an arbitrary affine variety.

we draw can mislead).

Consider for example the affine cone 𝑉

𝑋1 𝑋4 − 𝑋2 𝑋3 = 0.

𝑍 ∶ 𝑋2 = 0 = 𝑋4 ; 𝑍 = {(∗, 0, ∗, 0)}
It contains the planes,

𝑍′ ∶ 𝑋1 = 0 = 𝑋3 ; 𝑍 ′ = {(0, ∗, 0, ∗)}
and 𝑍 ∩ 𝑍 ′ = {(0, 0, 0, 0)}. Because 𝑉 is a hypersurface in 𝔸4 , it has dimension 3, and
each of 𝑍 and 𝑍 ′ has dimension 2. Thus
codim 𝑍 ∩ 𝑍 ′ = 3 ≰ 1 + 1 = codim 𝑍 + codim 𝑍 ′ .
The proof of 5.36 fails because the diagonal in 𝑉 × 𝑉 cannot be defined by 3 equations
(it takes the same 4 that define the diagonal in 𝔸4 ) — the diagonal is not a set-theoretic
complete intersection.

k. Dominant maps
As in the affine case, a regular map 𝜑 ∶ 𝑉 → 𝑊 is said to be dominant if its image is
dense in 𝑊.
Let 𝜑 ∶ 𝑉 → 𝑊 be a dominant map. For any open subset 𝑈 of 𝑊, the map
𝑓 ↦ 𝑓◦𝜑 ∶ 𝛤(𝑈, 𝒪𝑊 ) → 𝛤(𝜑−1 (𝑈), 𝒪𝑉 )
is injective. Now assume that 𝑉 and 𝑊 are irreducible. On passing to the direct limit in
(*)

𝑘(𝑊) → 𝑘(𝑉).
(*), we get a homomorphism of fields

If 𝑈𝑉 and 𝑈𝑊 are open affines of 𝑉 and 𝑊 such that 𝜑(𝑈𝑉 ) ⊂ 𝑈𝑊 , then

𝑘[𝑈𝑊 ] → 𝑘[𝑈𝑉 ]
is injective, so 𝜑|𝑈𝑉 ∶ 𝑈𝑉 → 𝑈𝑊 is dominant. An elementary, but nontrivial, argument
shows that 𝜑(𝑉) contains a dense open subset of 𝑊 (see Theorem 9.1 below).
 l. Rational maps; birational equivalence 117

l. Rational maps; birational equivalence

Loosely speaking, a rational map from a variety 𝑉 to a variety 𝑊 is a regular map from a
dense open subset of 𝑉 to 𝑊, and a birational map is a rational map admitting a rational

Let 𝑉 and 𝑊 be varieties over 𝑘, and consider pairs (𝑈, 𝜑𝑈 ), where 𝑈 is a dense open
inverse.

subset of 𝑉 and 𝜑𝑈 is a regular map 𝑈 → 𝑊. Two such pairs (𝑈, 𝜑𝑈 ) and (𝑈 ′ , 𝜑𝑈 ′ ) are
said to be equivalent if 𝜑𝑈 and 𝜑𝑈 ′ agree on 𝑈 ∩ 𝑈 ′ . An equivalence class of pairs is
called a rational map 𝜑 ∶ 𝑉 ⤏ 𝑊. A rational map 𝜑 is said to be defined at a point 𝑣 of
𝑉 if 𝑣 ∈ 𝑈 for some (𝑈, 𝜑𝑈 ) ∈ 𝜑. The set 𝑈1 of 𝑣 at which 𝜑 is defined is open, and there
⋃
is a regular map 𝜑1 ∶ 𝑈1 → 𝑊 such that (𝑈1 , 𝜑1 ) ∈ 𝜑 — clearly, 𝑈1 = (𝑈,𝜑 )∈𝜑 𝑈 and
we can define 𝜑1 to be the regular map such that 𝜑1 |𝑈 = 𝜑𝑈 for all (𝑈, 𝜑𝑈 ) ∈ 𝜑. Hence,
𝑈

in the equivalence class, there is always a pair (𝑈, 𝜑𝑈 ) with 𝑈 largest (and 𝑈 is called
“the open subvariety on which 𝜑 is defined”).

Proposition 5.38. Let 𝑉 and 𝑉 ′ be irreducible varieties over 𝑘. A regular map 𝜑 ∶ 𝑈 ′ →

𝑈 from an open subset 𝑈 ′ of 𝑉 ′ onto an open subset 𝑈 of 𝑉 defines a 𝑘-algebra homomor-
phism 𝑘(𝑉) → 𝑘(𝑉 ′ ), and every such homomorphism arises in this way.

Proof. The first part of the statement is obvious, so let 𝑘(𝑉) → 𝑘(𝑉 ′ ) be a 𝑘-algebra
homomorphism. We use it to identify 𝑘(𝑉) with a subfield of 𝑘(𝑉 ′ ). Let 𝑈 (resp. 𝑈 ′ )
be an open affine subset of 𝑉 (resp. 𝑉 ′ ). Let 𝑘[𝑈] = 𝑘[𝑥1 , … , 𝑥𝑚 ]. Each 𝑥𝑖 ∈ 𝑘(𝑉 ′ ),
which is the field of fractions of 𝑘[𝑈 ′ ], and so there exists a nonzero 𝑑 ∈ 𝑘[𝑈 ′ ] such that
𝑑𝑥𝑖 ∈ 𝑘[𝑈 ′ ] for all 𝑖. After inverting 𝑑, i.e., replacing 𝑈 ′ with basic open subset, we may
suppose that 𝑘[𝑈] ⊂ 𝑘[𝑈 ′ ]. The inclusion 𝑘(𝑉) → 𝑘(𝑉 ′ ) is induced by the inclusion
𝑘[𝑈] → 𝑘[𝑈 ′ ], hence by the corresponding dominant map 𝜑 ∶ 𝑈 ′ → 𝑈. The image of
𝜑 contains an open subset 𝑈0 of 𝑈 (see the preceding subsection), and the restriction of
𝜑 to 𝜑−1 (𝑈0 ) → 𝑈0 is the required map. 2

A rational (or regular) map 𝜑 ∶ 𝑉 ⤏ 𝑊 is birational if there exists a rational map

𝜑′ ∶ 𝑊 ⤏ 𝑉 such that 𝜑′ ◦𝜑 = id𝑉 and 𝜑◦𝜑′ = id𝑊 as rational maps. Two varieties 𝑉
and 𝑉 ′ are birationally equivalent if there exists a birational map from one to the other.
In this case, there exist dense open subsets 𝑈 and 𝑈 ′ of 𝑉 and 𝑉 ′ respectively such that
𝑈 ≈ 𝑈′.

Proposition 5.39. Two irreducible varieties 𝑉 and 𝑉 ′ are birationally equivalent if and
only if their function fields are isomorphic over 𝑘.

Proof. Assume that 𝑘(𝑉) ≈ 𝑘(𝑉 ′ ). We may suppose that 𝑉 and 𝑊 are affine, in which
case the existence of 𝑈 ≈ 𝑈 ′ is proved in 3.36. This proves the “if” part, and the “only
if” part is obvious. 2

Proposition 5.40. Every irreducible algebraic variety of dimension 𝑑 is birationally equiv-

alent to a hypersurface in 𝔸𝑑+1 .

Proof. Let 𝑉 be an irreducible variety of dimension 𝑑. According to Proposition

3.38, there exist 𝑥1 , … , 𝑥𝑑 , 𝑥𝑑+1 ∈ 𝑘(𝑉) such that 𝑘(𝑉) = 𝑘(𝑥1 , … , 𝑥𝑑 , 𝑥𝑑+1 ). Let 𝑓 ∈
𝑘[𝑋1 , … , 𝑋𝑑+1 ] be an irreducible polynomial satisfied by the 𝑥𝑖 , and let 𝐻 be the hyper-
surface 𝑓 = 0. Then 𝑘(𝑉) ≈ 𝑘(𝐻). 2
118 5. Algebraic Varieties

m. Local study
Everything in Chapter 4, being local, extends mutatis mutandis, to general algebraic

5.41. The tangent space 𝑇𝑃 (𝑉) at a point 𝑃 on an algebraic variety 𝑉 is the fibre of
varieties.

𝑉(𝑘[𝜀]) → 𝑉(𝑘) over 𝑃. There are canonical isomorphisms

𝑇𝑃 (𝑉) ≃ Der𝑘 (𝒪𝑃 , 𝑘) ≃ Hom𝑘-linear (𝔫𝑃 ∕𝔫2𝑃 , 𝑘),

where 𝔫𝑃 is the maximal ideal of 𝒪𝑃 .

5.42. A point 𝑃 on an algebraic variety 𝑉 is nonsingular (or smooth) if it lies on a single

irreducible component 𝑊 and dim 𝑇𝑃 (𝑉) = dim 𝑊. A point 𝑃 is nonsingular if and only
if the local ring 𝒪𝑃 is regular. The singular points form a proper closed subvariety, called
the singular locus.

5.43. A variety is nonsingular (or smooth) if every point is nonsingular.

n. Étale maps
An étale morphism is the analogue in algebraic geometry of a local isomor-
phism of manifolds in differential geometry, a covering of Riemann surfaces
with no branch points in complex analysis, and an unramified extension in

Definition 5.44. A regular map 𝜑 ∶ 𝑉 → 𝑊 of smooth varieties is étale at a point 𝑃

algebraic number theory.

of 𝑉 if the map (𝑑𝜑)𝑃 ∶ 𝑇𝑃 (𝑉) → 𝑇𝜑(𝑃) (𝑊) is an isomorphism; 𝜑 is étale if it is étale at

all points of 𝑉.

Examples
5.45. A regular map

𝜑 ∶ 𝔸𝑛 → 𝔸𝑛 , 𝑎 ↦ (𝑃1 (𝑎1 , … , 𝑎𝑛 ), … , 𝑃𝑛 (𝑎1 , … , 𝑎𝑛 ))

is étale at 𝐚 if and only if rank Jac(𝑃1 , … , 𝑃𝑛 )(𝐚) = 𝑛, because the map on the tangent
𝜕𝑃
spaces has matrix Jac(𝑃1 , … , 𝑃𝑛 )(𝐚). Equivalent condition: det ( 𝑖 (𝐚)) ≠ 0.
𝜕𝑋𝑗
∑
5.46. Let 𝑉 = Spm(𝐴) be an affine variety, and let 𝑓 = 𝑐𝑖 𝑋 𝑖 ∈ 𝐴[𝑋] be such that
𝐴[𝑋]∕(𝑓(𝑋)) is reduced. Let 𝑊 = Spm(𝐴[𝑋]∕(𝑓(𝑋))), and consider the map 𝑊 → 𝑉
corresponding to the inclusion 𝐴 → 𝐴[𝑋]∕(𝑓). Thus,

𝐴[𝑋]∕(𝑓) ← 𝐴[𝑋] 𝑊 ← → 𝑉 × 𝔸1
↠

← →
←
←

→
→

𝐴 𝑉.
←

The points of 𝑊 lying over a point 𝐚 ∈ 𝑉 are the pairs (𝐚, 𝑏) ∈ 𝑉 × 𝔸1 such that 𝑏 is
∑
a root of 𝑐𝑖 (𝐚)𝑋 𝑖 . I claim that the map 𝑊 → 𝑉 is étale at (𝐚, 𝑏) if and only if 𝑏 is a
∑
simple root of 𝑐𝑖 (𝐚)𝑋 𝑖 .
 n. Étale maps 119

To see this, write 𝐴 = 𝑘[𝑋1 , … , 𝑋𝑛 ]∕𝔞, 𝔞 = (𝑓1 , … , 𝑓𝑟 ), so that

𝐴[𝑋]∕(𝑓) = 𝑘[𝑋1 , … , 𝑋𝑛 ]∕(𝑓1 , … , 𝑓𝑟 , 𝑓).

The tangent spaces to 𝑊 and 𝑉 at (𝐚, 𝑏) and 𝐚 respectively are the null spaces of the

⎛ 𝜕𝑓1 𝜕𝑓1 ⎞
matrices

(𝐚) … (𝐚) 0 𝜕𝑓1 𝜕𝑓1

⎜ 𝜕𝑋1 𝜕𝑋𝑛 ⎟ ⎛ (𝐚) … (𝐚) ⎞
⎜ ⋮ ⋮ ⎟ ⎜ 𝜕𝑋1 𝜕𝑋𝑛 ⎟
⎜ 𝜕𝑓𝑟 𝜕𝑓𝑟 ⎟ ⎜ ⋮ ⋮ ⎟
(𝐚) … (𝐚) 0 𝜕𝑓𝑟 𝜕𝑓𝑟
⎜ 𝜕𝑋1 𝜕𝑋𝑛 ⎟ ⎜ (𝐚) … (𝐚) ⎟
⎜ 𝜕𝑓 𝜕𝑓 𝜕𝑓 𝜕𝑋1 𝜕𝑋𝑛
(𝐚) … (𝐚) (𝐚, 𝑏) ⎟ ⎝ ⎠
⎝ 𝜕𝑋1 𝜕𝑋𝑛 𝜕𝑋 ⎠
and the map 𝑇(𝐚,𝑏) (𝑊) → 𝑇𝐚 (𝑉) is induced by the projection map 𝑘 𝑛+1 → 𝑘 𝑛 omitting
𝜕𝑓
(𝐚, 𝑏)≠ 0, because
𝜕𝑋
the last coordinate. This map is an isomorphism if and only if

∑
then every solution of the smaller set of equations extends uniquely to a solution of the

𝜕𝑓 𝑑( 𝑖 𝑐𝑖 (𝐚)𝑋 𝑖 )
(𝐚, 𝑏) = (𝑏),
larger set. But

𝜕𝑋 𝑑𝑋
∑
which is zero if and only if 𝑏 is a multiple root of 𝑖 𝑐𝑖 (𝐚)𝑋 𝑖 . The intuitive picture is that
𝑊 → 𝑉 is a finite covering with deg(𝑓) sheets, which is ramified exactly at the points

5.47. Consider a dominant map 𝜑 ∶ 𝑉 → 𝑊 of smooth affine varieties, corresponding

where two or more sheets cross (see p. 50).

to a map 𝐴 → 𝐵 of rings. Suppose that 𝐵 can be written 𝐵 = 𝐴[𝑌1 , … , 𝑌𝑛 ]∕(𝑃1 , … , 𝑃𝑛 )

𝜕𝑃
𝜑 is étale if and only if det ( 𝑖 (𝐚)) is never zero.
(same number of polynomials as variables). A similar argument to the above shows that

𝜕𝑋𝑗

𝜑 ∶ 𝑉 → 𝑊 be étale at 𝑃 ∈ 𝑉. Then there exist a regular map 𝜑′ ∶ 𝑉 ′ → 𝑊 ′ of affine

5.48. The example in 5.46 is typical: in fact, locally, every étale map is of this form. Let

varieties with 𝑘[𝑉 ′ ] = 𝑘[𝑊 ′ ][𝑋]∕(𝑓(𝑋)), and a commutative diagram

𝑉 ← 𝑈𝑃 ← → 𝑉′
open
immersion
→

𝜑 𝜑′
←

←
←

𝑊 ← 𝑈𝑄 ← → 𝑊′
open
étale
→
→

immersion
→

with 𝑈𝑃 and 𝑈𝑄 open neighbourhoods of 𝑃 and 𝑄 = 𝜑(𝑃). See Milne 1980, I, 3.14, for
def

the proof, which uses the affine case of Zariski’s main theorem.

The failure of the inverse function theorem for the Zariski topology

function theorem says that a map 𝜑 that is étale at a point 𝐚 is a local isomorphism there,
5.49. In advanced calculus (or differential topology, or complex analysis), the inverse

i.e., there exist open neighbourhoods 𝑈 and 𝑈 ′ of 𝐚 and 𝜑(𝐚) such that 𝜑 induces an
isomorphism 𝑈 → 𝑈 ′ . This is not true in algebraic geometry, at least not for the Zariski
topology: a map can be étale at a point without being a local isomorphism. Consider, for

𝜑 ∶ 𝔸1 → 𝔸1 , 𝑎 ↦ 𝑎2 ,
example, the map
120 5. Algebraic Varieties

and assume the characteristic is ≠ 2. Then 𝜑 is étale at any point 𝑎 ≠ 0 because the
Jacobian matrix is (2𝑋), which has rank one for 𝑋 = 𝑎 ≠ 0 (alternatively, it is of the
form 5.46 with 𝑓(𝑋) = 𝑋 2 − 𝑇, where 𝑇 is the coordinate function on 𝔸1 , and 𝑋 2 − 𝑎 has
distinct roots for 𝑎 ≠ 0). Nevertheless, I claim that there do not exist nonempty open
subsets 𝑈 and 𝑈 ′ of 𝔸1 such that 𝜑 induces an isomorphism 𝑈 → 𝑈 ′ . If there did, then
𝜑 would define an isomorphism 𝑘[𝑈 ′ ] → 𝑘[𝑈] and hence an isomorphism of the fields
of fractions 𝑘(𝔸1 ) → 𝑘(𝔸1 ). But on the fields of fractions, 𝜑 corresponds to the map

𝑘(𝑋) → 𝑘(𝑋), 𝑋 ↦ 𝑋2,

which is not an isomorphism.

5.50. Let 𝑉 be the plane curve 𝑌 2 = 𝑋 and 𝜑 the map 𝑉 → 𝔸1 , (𝑥, 𝑦) ↦ 𝑥. Then 𝜑 is
2 ∶ 1 except over 0, and so we may view it schematically as

A1 |
0

However, when viewed as a Riemann surface, 𝑉(ℂ) consists of two sheets joined at a
single point 𝑂. As a point on the surface moves around 𝑂, it shifts from one sheet to the
other. Thus the true picture is more complicated. To get a section to 𝜑, it is necessary to
Die lliemaim'sche Windungsfläche erster Ordnung.

Tergl. Seite 162 168, 213-214ml 218-2 21

LiAdhsti\MSvyer, Zeinzu/.

remove a line in ℂ from 0 to infinity, which is not closed for the Zariski topology.

of the complex curve 𝑌 2 = 𝑋 into

Generated for jmilne (University of Michigan) on 2014-06-05 00:08 GMT / http://hdl.handle.net/2027/ucm.530650220x

It is not possible to fit the graph

3-space, but the picture at right is

an early depiction of it (from Neu-
mann, Carl, Vorlesungen über Rie-
Public Domain, Google-digitized / http://www.hathitrust.org/access_use#pd-google

mann’s theorie der Abel’schen inte-

grale, Leipzig: Teubner, 1865).

Étale maps of singular varieties

A regular map 𝜑 ∶ 𝑉 → 𝑊 induces a homomorphism gr(𝒪𝜑(𝑃) ) → gr(𝒪𝑃 ) for each

Using tangent cones, we can extend the notion of an étale morphism to singular varieties.

𝑃 ∈ 𝑉. We say that 𝜑 is étale at 𝑃 if this is an isomorphism. Note that then there

 o. Étale neighbourhoods 121

is an isomorphism of the geometric tangent cones 𝐶𝑃 (𝑉) → 𝐶𝜑(𝑃) (𝑊), but the map
on the geometric tangent cones may be an isomorphism without 𝜑 being étale at 𝑃.
Roughly speaking, to be étale at 𝑃, we need the map on geometric tangent cones to be

It is a fairly elementary result that a local homomorphism of local rings 𝜑 ∶ 𝐴 → 𝐵

an isomorphism and to preserve the “multiplicities” of the components.

the completions. Thus 𝜑 ∶ 𝑉 → 𝑊 is étale at 𝑃 if and only if the map 𝒪̂ 𝜑(𝑃) → 𝒪̂ 𝑃 is an

induces an isomorphism on the graded rings if and only if it induces an isomorphism on

isomorphism. Now 5.53 shows that the choice of a system of local parameters 𝑓1 , … , 𝑓𝑑
at a nonsingular point 𝑃 determines an isomorphism 𝒪̂ 𝑃 → 𝑘[[𝑋1 , … , 𝑋𝑑 ]].
We can rewrite this as follows: let 𝑡1 , … , 𝑡𝑑 be a system of local parameters at a
nonsingular point 𝑃; then there is a canonical isomorphism 𝒪̂ 𝑃 → 𝑘[[𝑡1 , … , 𝑡𝑑 ]]. For
𝑓 ∈ 𝒪̂ 𝑃 , the image of 𝑓 ∈ 𝑘[[𝑡1 , … , 𝑡𝑑 ]] can be regarded as the Taylor series of 𝑓.
For example, let 𝑉 = 𝔸1 , and let 𝑃 be the point 𝑎. Then 𝑡 = 𝑋 − 𝑎 is a local parameter
at 𝑎, 𝒪𝑃 consists of quotients 𝑓(𝑋) = 𝑔(𝑋)∕ℎ(𝑋) with ℎ(𝑎) ≠ 0, and the coefficients
∑
of the Taylor expansion 𝑛≥0 𝑎𝑛 (𝑋 − 𝑎)𝑛 of 𝑓(𝑋) can be computed as in elementary
calculus courses: 𝑎𝑛 = 𝑓 (𝑛) (𝑎)∕𝑛!.

Proposition 5.51. Let 𝜑 ∶ 𝑊 → 𝑉 be a map of irreducible affine varieties. If 𝑘(𝑊) is a

finite separable extension of 𝑘(𝑉), then 𝜑 is étale on a nonempty open subvariety of 𝑊.

Proof. After passing to open subvarieties, we may assume that 𝑊 and 𝑉 are nonsin-
gular, and that 𝑘[𝑊] = 𝑘[𝑉][𝑋]∕(𝑓(𝑋)), where 𝑓(𝑋) is separable when considered as a
polynomial in 𝑘(𝑉). Now the statement follows from 5.46. 2

Aside 5.52. There is an old conjecture that every étale map 𝜑 ∶ 𝔸𝑛 → 𝔸𝑛 is an isomorphism.
When we write 𝜑 = (𝑃1 , … , 𝑃𝑛 ), this becomes the statement:

𝜕𝑃𝑖
if det ( (𝐚)) is never zero (for 𝐚 ∈ 𝑘 𝑛 ), then 𝜑 has an inverse.
𝜕𝑋𝑗

𝜕𝑃𝑖 𝜕𝑃
The condition, det ( (𝐚)) never zero, implies that the polynomial det ( 𝑖 ) is a nonzero
𝜕𝑋𝑗 𝜕𝑋𝑗
𝜕𝑃
constant (by the Nullstellensatz 2.11 applied to the ideal generated by det ( 𝑖 )). This conjecture,
𝜕𝑋𝑗
which is known as the Jacobian conjecture, has not been settled even for 𝑘 = ℂ and 𝑛 = 2,
despite the existence of several published proofs and innumerable announced proofs. It has
caused many mathematicians a good deal of grief. It is probably harder than it is interesting. See
the Wikipedia: Jacobian conjecture.

o. Étale neighbourhoods
Let 𝑃 be a nonsingular point on a variety 𝑉 of dimension 𝑑. A system of local parameters
at 𝑃 is a family {𝑓1 , … , 𝑓𝑑 } of germs of regular functions at 𝑃 generating the maximal
ideal 𝔫𝑃 ⊂ 𝒪𝑃 . Equivalent conditions: the images of 𝑓1 , … , 𝑓𝑑 in 𝔫𝑃 ∕𝔫2𝑃 generate it as a
𝑘-vector space (see 1.4); or (𝑑𝑓1 )𝑃 , … , (𝑑𝑓𝑑 )𝑃 is a basis for the dual space to 𝑇𝑃 (𝑉). We
also say that 𝑓1 , … , 𝑓𝑛 are local parameters at 𝑃.7

Proposition 5.53. Let {𝑓1 , … , 𝑓𝑑 } be a system of local parameters at a nonsingular point

𝑃 of 𝑉. Then there is a nonsingular open neighbourhood 𝑈 of 𝑃 such that 𝑓1 , 𝑓2 , … , 𝑓𝑑 are
represented by pairs (𝑓̃1 , 𝑈), … , (𝑓̃𝑑 , 𝑈) and the map (𝑓̃1 , … , 𝑓̃𝑑 ) ∶ 𝑈 → 𝔸𝑑 is étale.
7
Sometimes also called local uniformizing parameters at 𝑃.
122 5. Algebraic Varieties

Proof. Obviously, the 𝑓𝑖 are represented by regular functions 𝑓̃𝑖 defined on a single
open neighbourhood 𝑈 ′ of 𝑃, which, because of 4.37, we can choose to be nonsingular.
The map 𝜑 = (𝑓̃1 , … , 𝑓̃𝑑 ) ∶ 𝑈 ′ → 𝔸𝑑 is étale at 𝑃, because the dual map to (𝑑𝜑)𝐚 is
(𝑑𝑋𝑖 )𝑜 ↦ (𝑑𝑓̃𝑖 )𝐚 . The next lemma then shows that 𝜑 is étale on an open neighbourhood
𝑈 of 𝑃. 2

Lemma 5.54. Let 𝑉 and 𝑊 be nonsingular varieties. If 𝜑 ∶ 𝑉 → 𝑊 is étale at 𝑃, then it is

étale at all points in an open neighbourhood of 𝑃.

Proof. The hypotheses imply that 𝑉 and 𝑊 have the same dimension 𝑑, and that their
tangent spaces all have dimension 𝑑. We may assume 𝑉 and 𝑊 to be affine, say, 𝑉 ⊂ 𝔸𝑚
and 𝑊 ⊂ 𝔸𝑛 , and that 𝜑 is given by polynomials 𝑃1 (𝑋1 , … , 𝑋𝑚 ), … , 𝑃𝑛 (𝑋1 , … , 𝑋𝑚 ). Then
𝜕𝑃
(𝑑𝜑)𝐚 ∶ 𝑇𝐚 (𝔸𝑚 ) → 𝑇𝜑(𝐚) (𝔸𝑛 ) is a linear map with matrix ( 𝑖 (𝐚)), and 𝜑 is not étale at
𝜕𝑋𝑗
𝐚 if and only if the kernel of this map contains a nonzero vector in the subspace 𝑇𝐚 (𝑊)
of 𝑇𝐚 (𝔸𝑛 ). Let 𝑓1 , … , 𝑓𝑟 generate 𝐼(𝑉). Then 𝜑 is not étale at 𝐚 if and only if the matrix

𝜕𝑓𝑖
⎛ (𝐚) ⎞
⎜ 𝜕𝑋𝑗 ⎟
⎜ 𝜕𝑃𝑖 (𝐚) ⎟
⎝ 𝜕𝑋𝑗 ⎠
has rank less than 𝑚. This is a polynomial condition on 𝐚, and so it fails on a closed
subset of 𝑊, which does not contain 𝑃. 2

Let 𝑉 be a nonsingular variety, and let 𝑃 ∈ 𝑉. An étale neighbourhood of a point

𝑃 of 𝑉 is a pair (𝑄, 𝜋 ∶ 𝑈 → 𝑉) with 𝜋 an étale map from a nonsingular variety 𝑈 to 𝑉
and 𝑄 a point of 𝑈 such that 𝜋(𝑄) = 𝑃.

Corollary 5.55. Let 𝑉 be a nonsingular variety of dimension 𝑑, and let 𝑃 ∈ 𝑉. There

is an open Zariski neighbourhood 𝑈 of 𝑃 and a map 𝜋 ∶ 𝑈 → 𝔸𝑑 realizing (𝑃, 𝑈) as an
étale neighbourhood of (0, … , 0) ∈ 𝔸𝑑 .

Proof. This is a restatement of the Proposition. 2

Aside 5.56. (a) Note the similarity to the definition of a differentiable manifold: every point 𝑃 on
a nonsingular variety of dimension 𝑑 has an open neighbourhood that is also a “neighbourhood”
of the origin in 𝔸𝑑 . There is a “topology” on algebraic varieties for which the “open neighbour-
hoods” of a point are the étale neighbourhoods. Relative to this “topology”, any two nonsingular
varieties are locally isomorphic (this is not true for the Zariski topology). The “topology” is called

(b) Smooth functions 𝑥1 , … , 𝑥𝑛 defined on an open neighbourhood 𝑈 of a point 𝑃 of a

the étale topology — see my notes Lectures on Étale Cohomology.

differential manifold 𝑀 are local coordinates at 𝑃 if they are zero at 𝑃 and the map 𝑥1 , … , 𝑥𝑛 ∶ 𝑈 →
ℝ𝑛 is an isomorphism (of manifolds) from 𝑈 onto an open submanifold of ℝ𝑛 .
Compare: regular functions 𝑓1 , … , 𝑓𝑛 defined on an open neighbourhood 𝑈 of a point 𝑃 of
a nonsingular algebraic variety 𝑉 are local parameters at 𝑃 if they are zero at 𝑃 and the map
𝑓1 , … , 𝑓𝑛 ∶ 𝑈 → 𝔸𝑛 is an étale map from 𝑈 onto an open subvariety 𝑈 ′ of 𝔸𝑛 . In general,
(𝑈; 𝑓1 , … , 𝑓𝑛 ) cannot be chosen so that the map 𝑈 → 𝑈 ′ is an isomorphism. However, when
𝑘 = ℂ, there exists a neighbourhood 𝑈 for the complex topology such that 𝑓1 , … , 𝑓𝑛 define an
isomorphism (of complex manifolds) from 𝑈 onto an open complex submanifold of ℂ𝑛 .
 o. Étale neighbourhoods 123

The inverse function theorem (for the étale topology)

Theorem 5.57 (Inverse Function Theorem). Let 𝜑 ∶ 𝑉 → 𝑊 be a regular map. If
𝜑 is étale at 𝑃 ∈ 𝑉, then there exists a commutative diagram

(𝑉, 𝑃) ← (𝑈, 𝑃) 𝑄 = 𝜑(𝑃)

←
open

𝑈 an open neighbourhood of 𝑃
→

𝜑
(𝑈, 𝑃) an étale neighbourhood of 𝑄.
←

(𝑊, 𝑄)
→
→

étale

Proof. According to 5.54, there exists an open neighbourhood 𝑈 of 𝑃 such that the
restriction 𝜑|𝑈 of 𝜑 to 𝑈 is étale. 2

The rank theorem

For vector spaces, the rank theorem says the following: let 𝜑 ∶ 𝑉 → 𝑊 be a linear map
of 𝑘-vector spaces of rank 𝑟; then there exist bases for 𝑉 and 𝑊 relative to which 𝜑 has
𝐼 0
matrix ( 𝑟 ). In other words, there is a commutative diagram
0 0

𝑉 → 𝑊
← 𝛼

≈ ≈
←

𝑘𝑚 → 𝑘𝑛 .
(𝑥1 ,…,𝑥𝑚 )↦(𝑥1 ,…,𝑥𝑟 ,0,…)
→

A similar result holds locally for differentiable manifolds. In algebraic geometry, there is

Theorem 5.58 (Rank Theorem). Let 𝜑 ∶ 𝑉 → 𝑊 be a regular map of nonsingular

the following weaker analogue.

varieties of dimensions 𝑚 and 𝑛 respectively, and let 𝑃 ∈ 𝑉. If rank(𝑇𝑃 (𝜑)) = 𝑛, then there
exists a commutative diagram

𝑈𝑃 → 𝑈𝜑(𝑃)
← 𝜑|𝑈𝑃

etale
́ etale
́
←

𝔸𝑚 → 𝔸𝑛
(𝑥1 ,…,𝑥𝑚 )↦(𝑥1 ,…,𝑥𝑛 )
→

in which 𝑈𝑃 and 𝑈𝜑(𝑃) are open neighbourhoods of 𝑃 and 𝜑(𝑃) respectively and the vertical
maps are étale.

Proof. Choose a system of local parameters 𝑔1 , … , 𝑔𝑛 at 𝜑(𝑃), and let 𝑓1 = 𝑔1 ◦𝜑, … , 𝑓𝑛 =

𝑔𝑛 ◦𝜑. Then 𝑑𝑓1 , … , 𝑑𝑓𝑛 are linearly independent forms on 𝑇𝑃 (𝑉), and there exist
𝑓𝑛+1 , … , 𝑓𝑚 such 𝑑𝑓1 , … , 𝑑𝑓𝑚 is a basis for 𝑇𝑃 (𝑉)∨ . Then 𝑓1 , … , 𝑓𝑚 is a system of local
parameters at 𝑃. According to 5.54, there exist open neighbourhoods 𝑈𝑃 of 𝑃 and 𝑈𝜑(𝑃)
of 𝜑(𝑃) such that the maps

(𝑓1 , … , 𝑓𝑚 ) ∶ 𝑈𝑃 → 𝔸𝑚
(𝑔1 , … , 𝑔𝑛 ) ∶ 𝑈𝜑(𝑃) → 𝔸𝑛

are étale. They give the vertical maps in the above diagram. 2
124 5. Algebraic Varieties

Aside 5.59. Tangent vectors at a point 𝑃 on a smooth manifold 𝑉 can be defined to be certain
equivalence classes of curves through 𝑃 (Wikipedia: Tangent space). For 𝑉 = 𝔸𝑛 , there is a
similar description with a curve taken to be a regular map from an open neighbourhood 𝑈 of 0
in 𝔸1 to 𝑉. In the general case there is a map from an open neighbourhood of the point 𝑃 in 𝑋
onto affine space sending 𝑃 to 0 and inducing an isomorphism from tangent space at 𝑃 to that at
0 (5.53). Unfortunately, the maps from 𝑈 ⊂ 𝔸1 to 𝔸𝑛 need not lift to 𝑋, and so it is necessary to
allow maps from smooth curves into 𝑋 (pull-backs of the covering 𝑋 → 𝔸𝑛 by the maps from 𝑈
into 𝔸𝑛 ). There is a description of the tangent vectors at a point 𝑃 on a smooth algebraic variety
𝑉 as certain equivalence classes of regular maps from an étale neighbourhood 𝑈 of 0 in 𝔸1 to 𝑉.

p. Smooth maps
Definition 5.60. A regular map 𝜑 ∶ 𝑉 → 𝑊 of nonsingular varieties is smooth at a
point 𝑃 of 𝑉 if (𝑑𝜑)𝑃 ∶ 𝑇𝑃 (𝑉) → 𝑇𝜑(𝑃) (𝑊) is surjective; 𝜑 is smooth if it is smooth at all
points of 𝑉.

Theorem 5.61. A map 𝜑 ∶ 𝑉 → 𝑊 is smooth at 𝑃 ∈ 𝑉 if and only if there exist open

neighbourhoods 𝑈𝑃 and 𝑈𝜑(𝑃) of 𝑃 and 𝜑(𝑃) respectively such that 𝜑|𝑈𝑃 factors into

𝑞
𝑈𝑃 ,,,,→𝔸dim 𝑉−dim 𝑊 × 𝑈𝜑(𝑃) ,→ 𝑈𝜑(𝑃) .
étale

Proof. Certainly, if 𝜑|𝑈𝑃 factors in this way, it is smooth. Conversely, if 𝜑 is smooth at

𝑃, then we get a diagram as in the rank theorem. From it we get maps

𝑈𝑃 → 𝔸𝑚 ×𝔸𝑛 𝑈𝜑(𝑃) → 𝑈𝜑(𝑃) .

The first is étale, and the second is the projection of 𝔸𝑚−𝑛 × 𝑈𝜑(𝑃) onto 𝑈𝜑(𝑃) . 2

Corollary 5.62. Let 𝑉 and 𝑊 be nonsingular varieties. If 𝜑 ∶ 𝑉 → 𝑊 is smooth at 𝑃,

then it is smooth on an open neighbourhood of 𝑉.

Proof. In fact, it is smooth on the neighbourhood 𝑈𝑃 in the theorem. 2

Separable maps
A transcendence basis 𝑆 of an extension 𝐸 ⊃ 𝐹 of fields is separating if the algebraic
extension 𝐸 ⊃ 𝐹(𝑆) is separable. A finitely generated extension 𝐸 ⊃ 𝐹 of fields is
separable if it admits a separating transcendence basis.

Definition 5.63. A dominant map 𝜑 ∶ 𝑊 → 𝑉 of irreducible algebraic varieties is

separable if 𝑘(𝑊) is a separable extension of 𝑘(𝑉).

Theorem 5.64. Let 𝜑 ∶ 𝑊 → 𝑉 be a regular map of irreducible varieties.

(a) If there exists a nonsingular point 𝑃 of 𝑊 such that 𝜑𝑃 is nonsingular and (𝑑𝜑)𝑃 is
surjective, then 𝜑 is dominant and separable.
(b) Conversely if 𝜑 is dominant and separable, then the set of 𝑃 ∈ 𝑊 satisfying (a) is
open and dense.

Proof. Replace 𝑊 and 𝑉 with their open subsets of nonsingular points. Then apply
the rank theorem. 2
 q. Algebraic varieties as functors 125

q. Algebraic varieties as functors

Let 𝑅 be an affine 𝑘-algebra and 𝑉 an algebraic variety over 𝑘. We define a point of
𝑉 with coordinates in 𝑅 (or an 𝑅-point of 𝑉) to be a regular map Spm(𝑅) → 𝑉. For
example, if 𝑉 = 𝑉(𝔞) ⊂ 𝔸𝑛 , then

𝑉(𝑅) = {(𝑎1 , … , 𝑎𝑛 ) ∈ 𝑅𝑛 ∣ 𝑓(𝑎1 , … , 𝑎𝑛 ) = 0 all 𝑓 ∈ 𝔞},

as the terminology suggests. For example, 𝑉(𝑘) = 𝑉 (as a set). In other words, 𝑉 (as a
set) can be identified with the set of points of 𝑉 with coordinates in 𝑘.
From the universal property of products, we see that

(𝑉 × 𝑊)(𝑅) = 𝑉(𝑅) × 𝑊(𝑅).

Caution 5.65. If 𝑉 is the union of two subvarieties, 𝑉 = 𝑉1 ∪ 𝑉2 , then it need not be

true that 𝑉(𝑅) = 𝑉1 (𝑅) ∪ 𝑉2 (𝑅). For example, for any polynomial 𝑓(𝑋1 , … , 𝑋𝑛 ),

𝔸𝑛 = 𝐷(𝑓) ∪ 𝑉(𝑓),

but, in general,

𝑅𝑛 ≠ {𝐚 ∈ 𝑅𝑛 ∣ 𝑓(𝐚) ∈ 𝑅× } ∪ {𝐚 ∈ 𝑅𝑛 ∣ 𝑓(𝐚) = 0}.

In fact, this need not be true even when 𝑉1 and 𝑉2 are open in 𝑉 because that would
require every regular map 𝑈 → 𝑉 with 𝑈 affine to factor through 𝑉1 or 𝑉2 , which is
nonsense. For example, 𝑉 = 𝔸2 ∖ {(0, 0)} = 𝐷(𝑋) ∪ 𝐷(𝑌), but the line 𝑋 + 𝑌 = 1 in 𝑉 is
not contained in 𝐷(𝑋1 ) or 𝐷(𝑋2 ).
def

Theorem 5.66. A regular map 𝜑 ∶ 𝑉 → 𝑊 of algebraic varieties defines a family of

maps of sets, 𝜑(𝑅) ∶ 𝑉(𝑅) → 𝑊(𝑅), one for each affine 𝑘-algebra 𝑅, such that for every
homomorphism 𝛼 ∶ 𝑅 → 𝑆 of affine 𝑘-algebras, the diagram

𝑉(𝑅) → 𝑊(𝑅)
← 𝜑(𝑅)

𝑉(𝛼) 𝑉(𝛽)
←

𝑉(𝑆) → 𝑊(𝛼1)
𝜑(𝑆)
(*)
→

commutes. Every family of maps with this property arises from a unique morphism of
algebraic varieties.

The first sentence just says that 𝑅 ⇝ 𝑉(𝑅) is a functor from affine 𝑘-algebras to sets,

Let 𝖵𝖺𝗋 𝑘 (resp. 𝖠𝖿𝖿 𝑘 ) denote the category of algebraic varieties over 𝑘 (resp. affine
which is obvious. We prove the second after restating it in terms of categories.

algebraic varieties over 𝑘). For a variety 𝑉, let ℎ𝑉aff denote the functor sending an affine
variety 𝑇 = Spm(𝑅) to 𝑉(𝑅) = Hom(𝑇, 𝑉). We can restate the second sentence of
Theorem 5.66 as follows.

Theorem 5.67. The functor

𝑉 ⇝ ℎ𝑉 ∶ 𝖵𝖺𝗋 𝑘 → 𝖥𝗎𝗇(𝖠𝖿𝖿 𝑘 , 𝖲𝖾𝗍𝗌)

aff

is fully faithful.
126 5. Algebraic Varieties

Proof. For an algebraic variety 𝑉 over 𝑘, let ℎ𝑉 denote the functor

𝑇 ⇝ Hom(𝑇, 𝑉) ∶ 𝖵𝖺𝗋 𝑘 → 𝖲𝖾𝗍.

𝑉 ⇝ ℎ𝑉 ∶ 𝖵𝖺𝗋 𝑘 → 𝖥𝗎𝗇(𝖵𝖺𝗋 𝑘 , 𝖲𝖾𝗍𝗌)

According to the Yoneda lemma (Wikipedia: Yoneda lemma) the functor

is fully faithful. Let 𝜑 be a morphism of functors ℎ𝑉aff → ℎ𝑉aff′ , and let 𝑇 be an algebraic
variety. Let (𝑈𝑖 )𝑖∈𝐼 be a finite affine covering of 𝑇. Each intersection 𝑈𝑖 ∩ 𝑈𝑗 is affine
(5.29), and so 𝜑 gives rise to a commutative diagram
∏ ∏
0 → ℎ𝑉 (𝑇) → ℎ𝑉 (𝑈𝑖 ) → ℎ (𝑈𝑖 ∩ 𝑈𝑗) )
𝑖 → 𝑖,𝑗 𝑉
← ← ←
←

𝜑(𝑈𝑖 ) 𝜑(𝑈𝑖 ∩𝑈𝑗 )

←
←

∏ ∏
0 → ℎ𝑉 ′ (𝑇) → ℎ𝑉 ′ (𝑈𝑖 ) → ℎ𝑉 ′ (𝑈𝑖 ∩ 𝑈𝑗) )
→
→

→
𝑖 𝑖,𝑗
← ← ←
←

in which the pairs of maps are defined by the inclusions 𝑈𝑖 ∩ 𝑈𝑗 → 𝑈𝑖 , 𝑈𝑗 . As the

rows are exact (5.15, last sentence), this shows that 𝜑𝑉 extends uniquely to a functor
ℎ𝑉 → ℎ𝑉 ′ , which (by the Yoneda lemma) arises from a unique regular map 𝑉 → 𝑉 ′ . 2
Corollary 5.68. To give an affine group variety over 𝑘 is the same as giving a functor
𝐺 from affine 𝑘-algebras to sets such that for some 𝑛 and finite set 𝑆 of polynomials in
𝑘[𝑋1 , 𝑋2 , … , 𝑋𝑛 ], 𝐺 is isomorphic to the functor sending 𝑅 to the set of zeros of 𝑆 in 𝑅𝑛 .

ditions imply that 𝐺 = ℎ𝑉 for an affine algebraic variety 𝑉 (unique up to a unique

Proof. Certainly an affine group variety defines such a functor. Conversely, the con-

isomorphism). The multiplication maps 𝐺(𝑅) × 𝐺(𝑅) → 𝐺(𝑅) give a morphism of func-
tors ℎ𝑉 × ℎ𝑉 → ℎ𝑉 . As ℎ𝑉 × ℎ𝑉 ≃ ℎ𝑉×𝑉 (by definition of 𝑉 × 𝑉), we see that they arise
from a regular map 𝑚 ∶ 𝑉 × 𝑉 → 𝑉. Similarly, the maps 𝑎 ↦ 𝑎−1 ∶ 𝐺(𝑅) → 𝐺(𝑅) arise
from a regular map inv ∶ 𝑉 → 𝑉. 2

Often a variety is most naturally defined in terms of its points functor. For example,

SL𝑛 ∶ 𝑅 ⇝ {𝑀 ∈ 𝑀𝑛 (𝑅) ∣ det(𝑀) = 1}

the group varieties

GL𝑛 ∶ 𝑅 ⇝ {𝑀 ∈ 𝑀𝑛 (𝑅) ∣ det(𝑀) ∈ 𝑅× }

𝔾𝑎 ∶ 𝑅 ⇝ (𝑅, +).

Which functors arise from algebraic varieties?

We now describe the essential image of ℎ ↦ ℎ𝑉 ∶ 𝖵𝖺𝗋 𝑘 → 𝖥𝗎𝗇(𝖠𝖿𝖿 𝑘 , 𝖲𝖾𝗍𝗌). The fibred
product of two maps 𝛼1 ∶ 𝐹1 → 𝐹3 , 𝛼2 ∶ 𝐹2 → 𝐹3 of sets is the set
𝐹1 ×𝐹3 𝐹2 = {(𝑥1 , 𝑥2 ) ∣ 𝛼1 (𝑥1 ) = 𝛼2 (𝑥2 )}.
When 𝐹1 , 𝐹2 , 𝐹3 are functors and 𝛼1 , 𝛼2 , 𝛼3 are morphisms of functors, there is a functor
𝐹 = 𝐹1 ×𝐹3 𝐹2 such that
(𝐹1 ×𝐹3 𝐹2 )(𝑅) = 𝐹1 (𝑅) ×𝐹3 (𝑅) 𝐹2 (𝑅)
for all affine 𝑘-algebras 𝑅.
To simplify the statement of the next proposition, we write 𝑈 for ℎ𝑈 when 𝑈 is an
affine variety.
 q. Algebraic varieties as functors 127

Proposition 5.69. A functor 𝐹 ∶ 𝖠𝖿𝖿 𝑘 → 𝖲𝖾𝗍𝗌 is in the essential image of 𝖵𝖺𝗋 𝑘 if and only
if there exists an affine variety 𝑈 and a morphism 𝑈 → 𝐹 such that
(a) the functor 𝑅 = 𝑈 ×𝐹 𝑈 is a closed affine subvariety of 𝑈 × 𝑈 and the maps 𝑅 ⇉ 𝑈
def

(b) the set 𝑅(𝑘) is an equivalence relation on 𝑈(𝑘), and the map 𝑈(𝑘) → 𝐹(𝑘) realizes
defined by the projections are open immersions;

𝐹(𝑘) as the quotient of 𝑈(𝑘) by 𝑅(𝑘).

Proof. Let 𝐹 = ℎ𝑉 for 𝑉 an algebraic variety. Choose a finite open affine covering
⋃ ⨆
𝑉 = 𝑈𝑖 of 𝑉, and let 𝑈 = 𝑈𝑖 . It is again an affine variety (Exercise 5-2). The functor
𝑅 is ℎ𝑈 ′ , where 𝑈 ′ is the disjoint union of the varieties 𝑈𝑖 ∩ 𝑈𝑗 . These are affine (5.29),
and so 𝑈 ′ is affine. As 𝑈 ′ is the inverse image of ∆𝑉 in 𝑈 × 𝑈, it is closed (5.26). This
proves (a), and (b) is obvious.
The converse is omitted for the present. 2

Aside 5.70. A variety 𝑉 defines a functor 𝑅 ⇝ 𝑉(𝑅) from the category of all 𝑘-algebras to 𝖲𝖾𝗍𝗌.
Again, we call the elements of 𝑉(𝑅) the points of 𝑉 with coordinates in 𝑅.
For example, if 𝑉 is affine,

𝑉(𝑅) = Hom𝑘-algebra (𝑘[𝑉], 𝑅).

More explicitly, if 𝑉 ⊂ 𝑘 𝑛 and 𝐼(𝑉) = (𝑓1 , … , 𝑓𝑚 ), then 𝑉(𝑅) is the set of solutions in 𝑅𝑛 of the

𝑓𝑖 (𝑋1 , … , 𝑋𝑛 ) = 0, 𝑖 = 1, … , 𝑚.
system equations

Note that, when we allow 𝑅 to have nilpotent elements, it is important to choose the 𝑓𝑖 to generate
𝐼(𝑉) (i.e., a radical ideal) and not just an ideal 𝔞 such that 𝑉(𝔞) = 𝑉.8 ⋃
For a general variety 𝑉, we write 𝑉 ∏ as a finite union of open affines 𝑉 = 𝑖 𝑉𝑖 , and we define
𝑉(𝑅) to be the set of families (𝛼𝑖 )𝑖∈𝐼 ∈ 𝑖∈𝐼 𝑉𝑖 (𝑅) such that 𝛼𝑖 agrees with 𝛼𝑗 on 𝑉𝑖 ∩ 𝑉𝑗 for all
𝑖, 𝑗 ∈ 𝐼. This is independent of the choice of the covering, and agrees with the previous definition
when 𝑉 is affine.

The functor defined by 𝔸(𝐸) (see p. 73) is 𝑅 ⇝ 𝑅 ⊗𝑘 𝐸.

A criterion for a functor to arise from an algebraic prevariety

5.71. By a functor we mean a functor from the category of affine 𝑘-algebras to sets. A
subfunctor 𝑈 of a functor 𝑋 is open if, for all maps 𝜑 ∶ ℎ𝐴 → 𝑋, the subfunctor 𝜑−1 (𝑈)
of ℎ𝐴 is defined by an open subvariety of Spm(𝐴). A family (𝑈𝑖 )𝑖∈𝐼 of open subfunctors
⋃
of 𝑋 is an open covering of 𝑋 if each 𝑈𝑖 is open in 𝑋 and 𝑋(𝐾) = 𝑈𝑖 (𝐾) for every field
𝐾. A functor 𝑋 is local if, for all 𝑘-algebras 𝑅 and all finite families (𝑓𝑖 )𝑖 of elements of
𝐴 generating 𝐴 as an ideal, the sequence of sets
∏ ∏
𝑋(𝑅) → 𝑋(𝑅𝑓𝑖 ) ⇉ 𝑋(𝑅𝑓𝑖 𝑓𝑗 )
𝑖 𝑖,𝑗

Let 𝔸1 denote the functor sending a 𝑘-algebra 𝑅 to its underlying set. For a functor
is exact.

𝑈, let 𝒪(𝑈) = Hom(𝑈, 𝔸1 ) — it is a 𝑘-algebra.9 A functor 𝑈 is affine if 𝒪(𝑈) is an

Let 𝔞 be an ideal in 𝑘[𝑋1 , …]. If 𝐴 has no nonzero nilpotent elements, then every 𝑘-algebra homomor-
phism 𝑘[𝑋1 , …] → 𝐴 that is zero on 𝔞 is also zero on rad(𝔞), and so
8

Hom𝑘 (𝑘[𝑋1 , …]∕𝔞, 𝐴) ≃ Hom𝑘 (𝑘[𝑋1 , …]∕rad(𝔞), 𝐴).

This is not true if 𝐴 has nonzero nilpotents.
Actually, one needs to be more careful to ensure that 𝒪(𝑈) is a set; for example, restrict 𝑈 and 𝔸1 to
the category of 𝑘-algebras of the form 𝑘[𝑋0 , 𝑋1 , …]∕𝔞 for a fixed family of symbols (𝑋𝑖 ) indexed by ℕ.
9
128 5. Algebraic Varieties

affine 𝑘-algebra and the canonical map 𝑈 → ℎ𝒪(𝑈) is an isomorphism. A local functor

𝑘.
admitting a finite covering by open affines is representable by an algebraic variety over

In the functorial approach to algebraic geometry, an algebraic prevariety over 𝑘 is

defined to be a functor satisfying this criterion. See, for example, Demazure and Gabriel
1970, I, §1, 3.11, p. 13.

r. Rational and unirational varieties

In this section, 𝑘 is an infinite field, not necessarily algebraically closed.

Definition 5.72. Let 𝑉 be a smooth projective variety over 𝑘 of dimension 𝑛.

(a) 𝑉 is unirational if there exists a dominant rational map ℙ𝑛 ⤏ 𝑉.
(b) 𝑉 is rational if there exists a birational map ℙ𝑛 ⤏ 𝑉.
(c) 𝑉 is stably rational if 𝑉 × ℙ𝑟 is rational for some 𝑟.

If 𝑉 is stably rational, then there exists a dominant rational map ℙ𝑛+𝑟 ⤏ 𝑉, and this
will restrict to a dominant rational map on some linear subspace ℙ𝑛 ⊂ ℙ𝑛+𝑟 . Therefore,

rational ⇐⇒ stably rational ⇐⇒ unirational.

An irreducible variety 𝑉 is rational if 𝑘(𝑉) is a pure transcendental extension of 𝑘,

and it is unirational if 𝑘(𝑉) is contained in a pure transcendental extension of 𝑘.
In 1876 (over ℂ), Lüroth proved that every unirational curve is rational (see FT, 9.19).
The Lüroth problem asks whether every unirational variety is rational.

and Severi proved that all unirational surfaces are rational, but in characteristic 𝑝 ≠ 0,
Already for surfaces, this is a difficult problem. In characteristic zero, Castelnuovo

Zariski showed that some surfaces of the form

𝑍 𝑝 = 𝑓(𝑋, 𝑌),

while obviously unirational, are not rational. Surfaces of this form are now called Zariski

Fano attempted to find counter-examples to the Lüroth problem in dimension 3

surfaces.

among the so-called Fano varieties, but none of his attempted proofs satisfies modern
standards. In 1971-72, three examples of nonrational unirational three-folds were found.

References.
Beauville, A., The Lüroth problem, Lecture Notes in Math., 2172, Fond. CIME/CIME
Found. Subser., Springer, Cham, 2016, 1–27, arXiv:1507.02476.
Hochenegger, A., Lehn, M., and Stellari, P. (Editors), Birational geometry of hypersur-
faces, Gargnano del Garda, Italy, 2018. Lectures from the school held in March 2018.
Lecture Notes of the Unione Matematica Italiana, 26. Springer, Cham, 2019.
 r. Rational and unirational varieties 129

A little history
In his first proof of the Riemann hypothesis for curves over finite fields, Weil made use
of the Jacobian variety of the curve, but initially he was not able to construct this as
a projective variety. This led him to introduce “abstract” algebraic varieties, neither
affine nor projective (in Weil 1946). Weil first made use of the Zariski topology when
he introduced fibre spaces into algebraic geometry (in 1949). For more on this, see my
article: The Riemann hypothesis over finite fields: from Weil to the present day.

Exercises
5-1. Show that the only regular functions on ℙ1 are the constant functions. [Thus
ℙ1 is not affine. When 𝑘 = ℂ, ℙ1 is the Riemann sphere (as a set), and one knows
from complex analysis that the only holomorphic functions on the Riemann sphere are
constant. Since regular functions are holomorphic, this proves the statement in this case.
The general case is easier.]

5-2. Let 𝑉 be the disjoint union of algebraic varieties 𝑉1 , … , 𝑉𝑛 . This set has an obvious
topology and ringed space structure for which it is an algebraic variety. Show that 𝑉 is
affine if and only if each 𝑉𝑖 is affine.

5-3. Show that an algebraic variety 𝐺 equipped with a group structure is a group variety
if the map (𝑥, 𝑦) ↦ 𝑥−1 𝑦 ∶ 𝐺 × 𝐺 → 𝐺 is regular.

5-4. Let 𝐺 be a group variety. Show:

(a) The neutral element 𝑒 of 𝐺 is contained in a unique irreducible component 𝐺 ◦ of
𝐺, which is also the unique connected component of 𝐺 containing 𝑒.
(b) The subvariety 𝐺 ◦ is a normal subgroup of 𝐺 of finite index, and every subgroup
variety of 𝐺 of finite index contains 𝐺 ◦ .

5-5. Show that every subgroup variety of a group variety is closed.

5-6. Show that a prevariety 𝑉 is separated if and only if it satisfies the following condi-
tion: a regular map 𝑈 ∖ {𝑃} → 𝑉 with 𝑈 a curve and 𝑃 a nonsingular point on 𝑈 extends
in at most one way to a regular map 𝑈 → 𝑉.

5-7. Prove the final statement in 5.71.

5-8. Let 𝑉 be an algebraic variety. Show that the Zariski topology on 𝑉 × 𝑉 agrees with
the product topology if and only dim(𝑉) = 0.
Chapter 6

Projective Varieties

Recall (5.3) that we defined ℙ𝑛 to be the set of equivalence classes in 𝑘𝑛+1 ∖ {origin} for

(𝑎0 , … , 𝑎𝑛 ) ∼ (𝑏0 , … , 𝑏𝑛 ) ⇐⇒ (𝑎0 , … , 𝑎𝑛 ) = 𝑐(𝑏0 , … , 𝑏𝑛 ) for some 𝑐 ∈ 𝑘 × .

the relation

Let (𝑎0 ∶ … ∶ 𝑎𝑛 ) denote the equivalence class of (𝑎0 , … , 𝑎𝑛 ), and let 𝜋 denote the map
𝑘 𝑛+1 ∖ {(0, … , 0)}
∼ → ℙ𝑛 .

Let 𝑈𝑖 be the set of (𝑎0 ∶ … ∶ 𝑎𝑛 ) ∈ ℙ𝑛 such that 𝑎𝑖 ≠ 0, and let 𝑢𝑖 be the bijection
𝑎 ˆ
𝑎 𝑎 𝑢𝑖
(𝑎0 ∶ … ∶ 𝑎𝑛 ) ↦ ( 𝑎0 , … , 𝑎𝑖 , … , 𝑎𝑛 ) ∶ 𝑈𝑖 ,→ 𝔸𝑛
𝑎𝑖
𝑖 𝑖 𝑖 𝑎𝑖
( omitted).

In this chapter, we show that ℙ𝑛 has a unique structure of an algebraic variety for which

a closed subvariety of ℙ𝑛 is called a projective variety, and a variety isomorphic to a

these maps become isomorphisms of affine algebraic varieties. A variety isomorphic to

locally closed subvariety of ℙ𝑛 is called a quasiprojective variety. Every affine variety is

quasi-projective, but not all algebraic varieties are quasi-projective. We study morphisms
between quasi-projective varieties.
Projective varieties are important for the same reason compact manifolds are im-
portant: results are often simpler when stated for projective varieties, and the “part at

theorem of Bezout (see 6.37 below) says that a curve of degree 𝑚 in the projective plane
infinity” often plays a role, even when we would like to ignore it. For example, a famous

intersects a curve of degree 𝑛 in exactly 𝑚𝑛 points (counting multiplicities). For affine

curves, one has only an inequality.

a. Algebraic subsets of ℙ𝑛
A polynomial 𝐹(𝑋0 , … , 𝑋𝑛 ) is said to be homogeneous of degree 𝑑 if it is a sum of terms
𝑎𝑖0 ,…,𝑖𝑛 𝑋00 ⋯ 𝑋𝑛𝑛 with 𝑖0 + ⋯ + 𝑖𝑛 = 𝑑; equivalently,
𝑖 𝑖

𝐹(𝑡𝑋0 , … , 𝑡𝑋𝑛 ) = 𝑡 𝑑 𝐹(𝑋0 , … , 𝑋𝑛 )

for all 𝑡 ∈ 𝑘. The polynomials homogeneous of degree 𝑑 form a subspace 𝑘[𝑋0 , … , 𝑋𝑛 ]𝑑

of 𝑘[𝑋0 , … , 𝑋𝑛 ], and ⨁
𝑘[𝑋0 , … , 𝑋𝑛 ] = 𝑘[𝑋0 , … , 𝑋𝑛 ]𝑑 ;
𝑑≥0

130
 a. Algebraic subsets of ℙ𝑛 131

∑
in other words, every polynomial 𝐹 can be written uniquely as a sum 𝐹 = 𝐹𝑑 with 𝐹𝑑
homogeneous of degree 𝑑.
Let 𝑃 = (𝑎0 ∶ … ∶ 𝑎𝑛 ) ∈ ℙ𝑛 . Then 𝑃 also equals (𝑐𝑎0 ∶ … ∶ 𝑐𝑎𝑛 ) for any 𝑐 ∈ 𝑘 × ,
and so we cannot speak of the value of a polynomial 𝐹(𝑋0 , … , 𝑋𝑛 ) at 𝑃. However, if 𝐹 is
homogeneous, then 𝐹(𝑐𝑎0 , … , 𝑐𝑎𝑛 ) = 𝑐𝑑 𝐹(𝑎0 , … , 𝑎𝑛 ), and so it does make sense to say
that 𝐹 is zero or not zero at 𝑃. An algebraic set in ℙ𝑛 (or projective algebraic set) is
the set of common zeros in ℙ𝑛 of some set of homogeneous polynomials.

Example 6.1. Consider the projective algebraic subset of ℙ2 defined by the homoge-

𝐸 ∶ 𝑌 2 𝑍 = 𝑋 3 + 𝑎𝑋𝑍 2 + 𝑏𝑍 3 .
neous equation

It consists of the points (𝑥 ∶ 𝑦 ∶ 1) on the affine curve 𝐸 ∩ 𝑈2

(26)

𝑌 2 = 𝑋 3 + 𝑎𝑋 + 𝑏

(see 2.2) together with the point “at infinity” (0 ∶ 1 ∶ 0). Note that 𝐸 ∩ 𝑈1 is the affine

𝑍 = 𝑋 3 + 𝑎𝑋𝑍 2 + 𝑏𝑍 3 ,
curve

and that (0 ∶ 1 ∶ 0) corresponds to the point (0, 0) on 𝐸 ∩ 𝑈1 :

𝑍 = 𝑋 3 + 𝑋𝑍 2 + 𝑍 3

As (0, 0) is nonsingular on 𝐸 ∩ 𝑈1 , we deduce from (4.5) that 𝐸 is nonsingular unless

𝑋 3 + 𝑎𝑋 + 𝑏 has a multiple root. A nonsingular curve of the form (26) is called an elliptic
curve.
132 6. Projective Varieties

An elliptic curve has a unique structure of a group variety for which the point at
infinity is the zero:

𝑃+𝑄

When 𝑎, 𝑏 ∈ ℚ, we can speak of the zeros of (26) with coordinates in ℚ. They also form
a group 𝐸(ℚ), which Mordell showed to be finitely generated. It is easy to compute the
torsion subgroup of 𝐸(ℚ), but there is at present no known algorithm for computing
the rank of 𝐸(ℚ). More precisely, there is an “algorithm” which works in practice, but

very beautiful theory surrounding elliptic curves over ℚ and other number fields, whose
which has not been proved to always terminate after a finite amount of time. There is a

origins can be traced back almost 1,800 years to Diophantus. (See my book on Elliptic
Curves for all of this.)

An ideal 𝔞 ⊂ 𝑘[𝑋0 , … , 𝑋𝑛 ] is said to be graded or homogeneous if it contains with

any polynomial 𝐹 all the homogeneous components of 𝐹, i.e., if
𝐹 ∈ 𝔞 ⇐⇒ 𝐹𝑑 ∈ 𝔞, all 𝑑.

⋄ an ideal is graded if and only if it is generated by (a finite set of) homogeneous

It is straightforward to check that

⋄ the radical of a graded ideal is graded;

polynomials;

⋄ an intersection, product, or sum of graded ideals is graded.

For a graded ideal 𝔞, we let 𝑉(𝔞) denote the set of common zeros of the homogeneous
polynomials in 𝔞. Clearly
𝔞 ⊂ 𝔟 ⇐⇒ 𝑉(𝔞) ⊃ 𝑉(𝔟).
If 𝐹1 , … , 𝐹𝑟 are homogeneous generators for 𝔞, then 𝑉(𝔞) is also the set of common zeros
of the 𝐹𝑖 . Clearly every polynomial in 𝔞 is zero on every representative of a point in
𝑉(𝔞). We write 𝑉 af f (𝔞) for the set of common zeros of 𝔞 in 𝑘 𝑛+1 . It is a cone in 𝑘𝑛+1 , i.e.,
together with any point 𝑃 it contains the line through 𝑃 and the origin, and
𝑉 aff (𝔞) ∖ {(0, … , 0)}
𝑉(𝔞) = ∼ .

The sets 𝑉(𝔞) in ℙ𝑛 have similar properties to their namesakes in 𝔸𝑛 .

 a. Algebraic subsets of ℙ𝑛 133

(a) 𝑉(0) = ℙ𝑛 ; 𝑉(𝔞) = ∅ ⇐⇒ rad(𝔞) ⊃ (𝑋0 , … , 𝑋𝑛 );

Proposition 6.2. There are the following relations:

(b) 𝑉(𝔞𝔟) = 𝑉(𝔞 ∩ 𝔟) = 𝑉(𝔞) ∪ 𝑉(𝔟);

∑ ⋂
(c) 𝑉( 𝔞𝑖 ) = 𝑉(𝔞𝑖 ).

Proof. For the second statement in (a), note that

𝑉(𝔞) = ∅ ⇐⇒ 𝑉 aff (𝔞) ⊂ {(0, … , 0)}

⇐⇒ rad(𝔞) ⊃ (𝑋0 , … , 𝑋𝑛 ) (strong Nullstellensatz 2.16).

relation between 𝑉(𝔞) and 𝑉 aff (𝔞).

The remaining statements can be proved directly, as in Proposition 2.10, or by using the
2

topology on ℙ𝑛 — this is called the Zariski topology on ℙ𝑛 .

Proposition 6.2 shows that the projective algebraic sets are the closed sets for a

If 𝐶 is a cone in 𝑘 𝑛+1 , then 𝐼(𝐶) is a graded ideal in 𝑘[𝑋0 , … , 𝑋𝑛 ]: if 𝐹(𝑐𝑎0 , … , 𝑐𝑎𝑛 ) = 0

for all 𝑐 ∈ 𝑘 × , then
∑
𝐹𝑑 (𝑎0 , … , 𝑎𝑛 ) ⋅ 𝑐𝑑 = 𝐹(𝑐𝑎0 , … , 𝑐𝑎𝑛 ) = 0,
𝑑
∑
for infinitely many 𝑐, and so 𝐹𝑑 (𝑎0 , … , 𝑎𝑛 )𝑋 𝑑 is the zero polynomial. For a subset 𝑆 of
ℙ𝑛 , we define the affine cone over 𝑆 in 𝑘𝑛+1 to be

𝐶 = 𝜋−1 (𝑆) ∪ {origin}

𝐼(𝑆) = 𝐼(𝐶).
and we set

Note that if 𝑆 is nonempty and closed, then 𝐶 is the closure of 𝜋−1 (𝑆) ≠ ∅, and that 𝐼(𝑆)
is spanned by the homogeneous polynomials in 𝑘[𝑋0 , … , 𝑋𝑛 ] zero on 𝑆.

Proposition 6.3. The maps 𝑉 and 𝐼 define inverse bijections between the set of algebraic
subsets of ℙ𝑛 and the set of proper graded radical ideals of 𝑘[𝑋0 , … , 𝑋𝑛 ]. An algebraic set
𝑉 in ℙ𝑛 is irreducible if and only if 𝐼(𝑉) is prime; in particular, ℙ𝑛 is irreducible.

Proof. We have bijections

{ } 𝑆↦𝐶 { }
algebraic subsets of ℙ𝑛 nonempty closed cones in 𝑘𝑛+1

𝑉 𝐼

{ }
proper graded radical ideals in 𝑘[𝑋0 , … , 𝑋𝑛 ]

Here the top map sends 𝑆 to the affine cone over 𝑆, and the maps 𝑉 and 𝐼 are in the
sense of projective geometry and affine geometry respectively. The composite of any

composite of the top map with 𝐼 is 𝐼 in the sense of projective geometry. Obviously, 𝑉 is
three of these maps is the identity map, which proves the first statement because the

irreducible if and only if the closure of 𝜋−1 (𝑉) is irreducible, which is true if and only if
𝐼(𝑉) is a prime ideal. 2
134 6. Projective Varieties

Note that the graded ideals (𝑋0 , … , 𝑋𝑛 ) and 𝑘[𝑋0 , … , 𝑋𝑛 ] are both radical, but

𝑉(𝑋0 , … , 𝑋𝑛 ) = ∅ = 𝑉(𝑘[𝑋0 , … , 𝑋𝑛 ])

and so the correspondence between irreducible subsets of ℙ𝑛 and radical graded ideals
is not quite one-to-one.

Bourbaki, Alg, II, §11. A graded ring is a pair (𝑆, (𝑆𝑑 )𝑑∈ℕ ) comprising a ring 𝑆 and a family of
Aside 6.4. In English “homogeneous ideal” is more common than “graded ideal”, but we follow

additive subgroups 𝑆𝑑 such that

⨁
𝑆= 𝑆𝑑 , 𝑆𝑑 𝑆𝑒 ⊂ 𝑆𝑑+𝑒 , all 𝑑, 𝑒 ∈ ℕ.
𝑑∈ℕ

An ideal 𝔞 in 𝑆 is graded if and only if

⨁
𝔞= (𝔞 ∩ 𝑆𝑑 );
𝑑∈ℕ

⨁ of (𝑆, (𝑆𝑑 )). The quotient of a graded ring 𝑆 by a graded

ideal 𝔞 is a graded ring 𝑆∕𝔞 = 𝑑 𝑆𝑑 ∕(𝔞 ∩ 𝑆𝑑 ).
this means that it is a graded submodule

b. The Zariski topology on ℙ𝑛

For a graded polynomial 𝐹, let

𝐷(𝐹) = {𝑃 ∈ ℙ𝑛 ∣ 𝐹(𝑃) ≠ 0}.

Then, just as in the affine case, 𝐷(𝐹) is open and the sets of this type form a base for the
topology of ℙ𝑛 . As in the opening paragraph of this chapter, we let 𝑈𝑖 = 𝐷(𝑋𝑖 ).
To each polynomial 𝑓(𝑋1 , … , 𝑋𝑛 ), we attach the homogeneous polynomial of the
(𝑋 )
𝑓 ∗ (𝑋0 , … , 𝑋𝑛 ) = 𝑋0 𝑓 1,…, 𝑛 ,
deg(𝑓) 𝑋
same degree

𝑋0 𝑋0

and to each homogeneous polynomial 𝐹(𝑋0 , … , 𝑋𝑛 ), we attach the polynomial

𝐹∗ (𝑋1 , … , 𝑋𝑛 ) = 𝐹(1, 𝑋1 , … , 𝑋𝑛 ).

Proposition 6.5. Each subset 𝑈𝑖 of ℙ𝑛 is open in the Zariski topology on ℙ𝑛 , and when
we endow it with the induced topology, the bijection

𝑈𝑖 ↔ 𝔸𝑛 , (𝑎0 ∶ … ∶ 1 ∶ … ∶ 𝑎𝑛 ) ↔ (𝑎0 , … , 𝑎𝑖−1 , 𝑎𝑖+1 , … , 𝑎𝑛 )

becomes a homeomorphism.

Proof. It suffices to prove this with 𝑖 = 0. The set 𝑈0 = 𝐷(𝑋0 ), and so it is a basic open
subset in ℙ𝑛 . Clearly, for any homogeneous polynomial 𝐹 ∈ 𝑘[𝑋0 , … , 𝑋𝑛 ],

𝐷(𝐹(𝑋0 , … , 𝑋𝑛 )) ∩ 𝑈0 = 𝐷(𝐹(1, 𝑋1 , … , 𝑋𝑛 )) = 𝐷(𝐹∗ )

and, for any polynomial 𝑓 ∈ 𝑘[𝑋1 , … , 𝑋𝑛 ],

𝐷(𝑓) = 𝐷(𝑓 ∗ ) ∩ 𝑈0 .

Thus, under the bijection 𝑈0 ↔ 𝔸𝑛 , the basic open subsets of 𝔸𝑛 correspond to the
intersections with 𝑈𝑖 of the basic open subsets of ℙ𝑛 , which proves that the bijection is a
homeomorphism. 2
 c. Closed subsets of 𝔸𝑛 and ℙ𝑛 135

Remark 6.6. It is possible to use this to give a different proof that ℙ𝑛 is irreducible. We

dense (see p. 45). Note that each 𝑈𝑖 is irreducible, and that 𝑈𝑖 ∩ 𝑈𝑗 is open and dense in
apply the criterion that a space is irreducible if and only if every nonempty open subset is

each of 𝑈𝑖 and 𝑈𝑗 (as a subset of 𝑈𝑖 , it is the set of points (𝑎0 ∶ … ∶ 1 ∶ … ∶ 𝑎𝑗 ∶ … ∶ 𝑎𝑛 )

with 𝑎𝑗 ≠ 0). Let 𝑈 be a nonempty open subset of ℙ𝑛 ; then 𝑈 ∩ 𝑈𝑖 is open in 𝑈𝑖 . For
some 𝑖, 𝑈 ∩ 𝑈𝑖 is nonempty, and so must intersect 𝑈𝑖 ∩ 𝑈𝑗 . Therefore 𝑈 intersects every
𝑈𝑗 , and so is dense in every 𝑈𝑗 . It follows that its closure is all of ℙ𝑛 .

c. Closed subsets of 𝔸𝑛 and ℙ𝑛

We identify 𝔸𝑛 with 𝑈0 , and examine the closures in ℙ𝑛 of closed subsets of 𝔸𝑛 . Note

ℙ𝑛 = 𝔸𝑛 ⊔ 𝐻∞ , 𝐻∞ = 𝑉(𝑋0 ).
that

With each ideal 𝔞 in 𝑘[𝑋1 , … , 𝑋𝑛 ], we associate the graded ideal 𝔞∗ in 𝑘[𝑋0 , … , 𝑋𝑛 ]

generated by {𝑓 ∗ ∣ 𝑓 ∈ 𝔞}. For a closed subset 𝑉 of 𝔸𝑛 , set 𝑉 ∗ = 𝑉(𝔞∗ ) with 𝔞 = 𝐼(𝑉).
With each graded ideal 𝔞 in 𝑘[𝑋0 , 𝑋1 , … , 𝑋𝑛 ], we associate the ideal 𝔞∗ in 𝑘[𝑋1 , … , 𝑋𝑛 ]
generated by {𝐹∗ ∣ 𝐹 ∈ 𝔞}. When 𝑉 is a closed subset of ℙ𝑛 , we set 𝑉∗ = 𝑉(𝔞∗ ) with
𝔞 = 𝐼(𝑉).

Proposition 6.7. (a) Let 𝑉 be a closed subset of 𝔸𝑛 . Then 𝑉 ∗ is the closure of 𝑉 in ℙ𝑛 ,

⋃
and (𝑉 ∗ )∗ = 𝑉. If 𝑉 = 𝑉𝑖 is the decomposition of 𝑉 into its irreducible components,
⋃
then 𝑉 ∗ = 𝑉𝑖∗ is the decomposition of 𝑉 ∗ into its irreducible components.
(b) Let 𝑉 be a closed subset of ℙ𝑛 . Then 𝑉∗ = 𝑉 ∩ 𝔸𝑛 , and if no irreducible component
of 𝑉 lies in 𝐻∞ or contains 𝐻∞ , then 𝑉∗ is a proper subset of 𝔸𝑛 , and (𝑉∗ )∗ = 𝑉.

Proof. Straightforward. 2

Examples

𝑉 ∶ 𝑌 2 = 𝑋 3 + 𝑎𝑋 + 𝑏,
6.8. For

𝑉 ∗ ∶ 𝑌 2 𝑍 = 𝑋 3 + 𝑎𝑋𝑍 2 + 𝑏𝑍 3 ,
we have

and (𝑉 ∗ )∗ = 𝑉.

6.9. Let 𝑉 = 𝑉(𝑓1 , … , 𝑓𝑚 ); then the closure of 𝑉 in ℙ𝑛 is the union of the irreducible
components of 𝑉(𝑓1∗ , … , 𝑓𝑚∗
) not contained in 𝐻∞ . For example, let

𝑉 = 𝑉(𝑋1 , 𝑋12 + 𝑋2 ) = {(0, 0)};

then 𝑉(𝑋0 𝑋1 , 𝑋12 + 𝑋0 𝑋2 ) consists of the two points (1 ∶ 0 ∶ 0) (the closure of 𝑉) and
(0 ∶ 0 ∶ 1) (which is contained in 𝐻∞ ).1

6.10. For 𝑉 = 𝐻∞ = 𝑉(𝑋0 ), we have 𝑉∗ = ∅ = 𝑉(1) and (𝑉∗ )∗ = ∅ ≠ 𝑉.

1
Of course, in this case 𝔞 = (𝑋1 , 𝑋2 ), 𝔞∗ = (𝑋1 , 𝑋2 ), and 𝑉 ∗ = {(1 ∶ 0 ∶ 0)}, and so this example does
not contradict the proposition.
136 6. Projective Varieties

d. The hyperplane at infinity

It is often convenient to think of ℙ𝑛 as being 𝔸𝑛 = 𝑈0 with a hyperplane added “at
infinity”. More precisely, we identify the set 𝑈0 with 𝔸𝑛 . The complement of 𝑈0 in ℙ𝑛 is

𝐻∞ = {(0 ∶ 𝑎1 ∶ … ∶ 𝑎𝑛 ) ∈ ℙ𝑛 },

which can be identified with ℙ𝑛−1 .

For example, ℙ1 = 𝔸1 ⊔ 𝐻∞ (disjoint union), with 𝐻∞ consisting of a single point,
and ℙ2 = 𝔸2 ∪ 𝐻∞ with 𝐻∞ a projective line. Consider the line

1 + 𝑎𝑋1 + 𝑏𝑋2 = 0

in 𝔸2 . Its closure in ℙ2 is the line

𝑋0 + 𝑎𝑋1 + 𝑏𝑋2 = 0.

This line intersects the line 𝐻∞ = 𝑉(𝑋0 ) at the point (0 ∶ −𝑏 ∶ 𝑎), which equals
(0 ∶ 1 ∶ −𝑎∕𝑏) when 𝑏 ≠ 0. Note that −𝑎∕𝑏 is the slope of the line 1 + 𝑎𝑋1 + 𝑏𝑋2 = 0,
and so the point at which a line intersects 𝐻∞ depends only on the slope of the line:
parallel lines intersect in one point at infinity. We can think of the projective plane ℙ2 as
being the affine plane 𝔸2 with one point added at infinity for each “direction” in 𝔸2 .
Similarly, we can think of ℙ𝑛 as being 𝔸𝑛 with one point added at infinity for each
direction in 𝔸𝑛 — being parallel is an equivalence relation on the lines in 𝔸𝑛 , and there

We can replace 𝑈0 with 𝑈𝑛 in the above discussion, and write ℙ𝑛 = 𝑈𝑛 ⊔ 𝐻∞ with

is one point at infinity for each equivalence class of lines.

𝐻∞ = {(𝑎0 ∶ … ∶ 𝑎𝑛−1 ∶ 0)}, as in Example 6.1. Note that in this example the point at
infinity on the elliptic curve 𝑌 2 = 𝑋 3 + 𝑎𝑋 + 𝑏 is the intersection of the closure of any
vertical line with 𝐻∞ .

e. ℙ𝑛 is an algebraic variety
For each 𝑖, write 𝒪𝑖 for the sheaf on 𝑈𝑖 ⊂ ℙ𝑛 defined by the homeomorphism 𝑢𝑖 ∶ 𝑈𝑖 →
𝔸𝑛 .
Lemma 6.11. Let 𝑈𝑖𝑗 = 𝑈𝑖 ∩ 𝑈𝑗 ; then 𝒪𝑖 |𝑈𝑖𝑗 = 𝒪𝑗 |𝑈𝑖𝑗 . When endowed with this sheaf,
𝑈𝑖𝑗 is an affine algebraic variety; moreover, 𝛤(𝑈𝑖𝑗 , 𝒪𝑖 ) is generated as a 𝑘-algebra by the
functions (𝑓|𝑈𝑖𝑗 )(𝑔|𝑈𝑖𝑗 ) with 𝑓 ∈ 𝛤(𝑈𝑖 , 𝒪𝑖 ), 𝑔 ∈ 𝛤(𝑈𝑗 , 𝒪𝑗 ).

Proof. It suffices to prove this for (𝑖, 𝑗) = (0, 1). All rings occurring in the proof will be
identified with subrings of the field 𝑘(𝑋0 , 𝑋1 , … , 𝑋𝑛 ).

(𝑎 𝑎 𝑎𝑛 )
𝑈0 = {(𝑎0 ∶ 𝑎1 ∶ … ∶ 𝑎𝑛 ) ∣ 𝑎0 ≠ 0} (𝑎0 ∶ 𝑎1 ∶ … ∶ 𝑎𝑛 ) ↔ 1 , 2 , … , ∈ 𝔸𝑛 .
Recall that

𝑎0 𝑎0 𝑎0
[𝑋 𝑋 ]
Let 𝑘 1 , 2 , … , 𝑛 be the subring of 𝑘(𝑋0 , 𝑋1 , … , 𝑋𝑛 ) generated by the quotients 𝑖
𝑋 𝑋
𝑋0 𝑋0 𝑋0 (𝑋 ) 0
𝑋
— it is the polynomial ring in the 𝑛 symbols 1 , … , 𝑛 . An element 𝑓 1 , … , 𝑛 ∈
𝑋 𝑋 𝑋

[𝑋 ] 𝑋0 𝑋0 𝑋0 𝑋0
𝑘 1 , … , 𝑛 defines a map
𝑋
𝑋0 𝑋0
(𝑎 𝑎𝑛 )
(𝑎0 ∶ 𝑎1 ∶ … ∶ 𝑎𝑛 ) ↦ 𝑓 1
,…, ∶ 𝑈0 → 𝑘,
𝑎0 𝑎0
 e. ℙ𝑛 is an algebraic variety 137

[𝑋 ]
and in this way 𝑘 ,
, … , 𝑛 becomes identified with the ring of regular functions
1 𝑋 𝑋2
𝑋0 𝑋0 ( [ 𝑋0 ])
on 𝑈0 , and 𝑈0 with Spm 𝑘 1 , … , 𝑛 .
𝑋 𝑋
𝑋0 𝑋0
Next consider the open subset of 𝑈0 ,
𝑈01 = {(𝑎0 ∶ … ∶ 𝑎𝑛 ) ∣ 𝑎0 ≠ 0, 𝑎1 ≠ 0}.
(𝑋 )
It is 𝐷 1 , and is therefore an affine subvariety of (𝑈0 , 𝒪0 ). The inclusion 𝑈01 → 𝑈0
𝑋0 [𝑋 ] [𝑋 ]
corresponds to the inclusion of rings 𝑘 1 , … , 𝑛 → 𝑘 1 , … , 𝑛 , 0 . An element
𝑋 𝑋 𝑋

(𝑋 ) [𝑋 ] 𝑋0 𝑋0 𝑋0 𝑋0 𝑋1 ( )
𝑓 ,…, , of 𝑘 , … , , defines the function (𝑎0 ∶ … ∶ 𝑎𝑛 ) ↦ 𝑓 1 , … , 𝑛 , 0
1 𝑋𝑛 𝑋0 1 𝑋𝑛 𝑋0 𝑎 𝑎 𝑎
𝑋0 𝑋0 𝑋1 𝑋0 𝑋0 𝑋1 𝑎0 𝑎0 𝑎1
on 𝑈01 .

(𝑎 𝑎𝑛 )
Similarly,

𝑈1 = {(𝑎0 ∶ 𝑎1 ∶ … ∶ 𝑎𝑛 ) ∣ 𝑎1 ≠ 0}; (𝑎0 ∶ 𝑎1 ∶ … ∶ 𝑎𝑛 ) ↔ ,…, ∈ 𝔸𝑛 ,

0
𝑎1 𝑎1
( [𝑋 𝑋 ]) (𝑋 ) [𝑋 ]
and we identify 𝑈1 with Spm 𝑘 0 , 2 , … , 𝑛 . A polynomial 𝑓 0 , … , 𝑛 in 𝑘 0 , … , 𝑛
𝑋 𝑋 𝑋
𝑋1 (𝑋0 𝑋1 ) 𝑋1 𝑋1 𝑋1 𝑋1
defines the map (𝑎0 ∶ … ∶ 𝑎𝑛 ) ↦ 𝑓 , … , ∶ 𝑈1 → 𝑘.
𝑎0 𝑎𝑛
𝑎1 𝑎1 (𝑋 )
When regarded as an open subset of 𝑈1 , 𝑈01 = 𝐷 0 , and is therefore an affine
𝑋1
[ 𝑋 of (𝑈 1], 𝒪1 ), [and the inclusion
] 𝑈01 → 𝑈1( corresponds)to the [ 𝑋inclusion of]
rings 𝑘 , … , → 𝑘 , … , , . An element 𝑓 ,…, , of 𝑘 , … , ,
𝑋𝑛 𝑋0 𝑋𝑛 𝑋1 𝑋0 𝑋𝑛 𝑋1 𝑋𝑛 𝑋1
subvariety
0 0
𝑋1 𝑋1 𝑋1 𝑋1 𝑋0 ( ) 𝑋1 𝑋1 𝑋0 𝑋1 𝑋1 𝑋0
defines the function (𝑎0 ∶ … ∶ 𝑎𝑛 ) ↦ 𝑓 0 , … , 𝑛 , 1 on 𝑈01 .
𝑎 𝑎 𝑎

[𝑋 ] 𝑎1 [ 𝑋 𝑎1 𝑎𝑋0 𝑋 ]
The two subrings 𝑘 , … , , and 𝑘 0 , … , 𝑛 , 1 of 𝑘(𝑋0 , 𝑋1 , … , 𝑋𝑛 ) are equal,
1 𝑋𝑛 𝑋0
𝑋0 𝑋0 𝑋1 𝑋1 𝑋1 𝑋0
and an element of this ring defines the same function on 𝑈01 regardless of which of the
two rings it is considered an element. Therefore, whether we regard 𝑈01 as a subvariety
of 𝑈0 or of 𝑈1 it inherits the same structure as an affine[algebraic variety (3.15). This
proves the first two assertions, and the third is obvious: 𝑘 1 , … , 𝑛 , 0 ] is generated by
𝑋 𝑋 𝑋

[𝑋 ] [𝑋 𝑋 ] 𝑋 𝑋0 𝑋1
its subrings 𝑘 , … , and 𝑘 , , … ,
0
1 𝑋𝑛 0 2 𝑋𝑛
𝑋0 𝑋0 𝑋1 𝑋1 𝑋1
. 2

Proposition 6.12. There is a unique structure of an algebraic variety on ℙ𝑛 for which

each 𝑈𝑖 is an open affine subvariety of ℙ𝑛 and each map 𝑢𝑖 is an isomorphism of algebraic
varieties. Moreover, ℙ𝑛 is separated.

Proof. Endow each 𝑈𝑖 with the structure of an affine algebraic variety for which 𝑢𝑖 is
⋃
an isomorphism. Then ℙ𝑛 = 𝑈𝑖 , and the lemma shows that this covering satisfies
the patching condition 5.15, and so ℙ𝑛 has a unique structure of a ringed space for
which 𝑈𝑖 → ℙ𝑛 is a homeomorphism onto an open subset of ℙ𝑛 and 𝒪ℙ𝑛 |𝑈𝑖 = 𝒪𝑈𝑖 .
Moreover, because each 𝑈𝑖 is an algebraic variety, this structure makes ℙ𝑛 into an
algebraic prevariety. Finally, the lemma shows that ℙ𝑛 satisfies the condition 5.29(c) to

Example 6.13. Let 𝐶 be the plane projective curve

be separated. 2

𝐶 ∶ 𝑌2𝑍 = 𝑋3
and assume that char(𝑘) ≠ 2. For each 𝑎 ∈ 𝑘 × , there is an automorphism
𝜑𝑎
(𝑥 ∶ 𝑦 ∶ 𝑧) ↦ (𝑎𝑥 ∶ 𝑦 ∶ 𝑎3 𝑧) ∶ 𝐶 ,→ 𝐶.
138 6. Projective Varieties

Patch two copies of 𝐶 × 𝔸1 together along 𝐶 × (𝔸1 − {0}) by identifying (𝑃, 𝑎) with
(𝜑𝑎 (𝑃), 𝑎−1 ), 𝑃 ∈ 𝐶, 𝑎 ∈ 𝔸1 ∖ {0}. One obtains in this way a singular surface that is
not quasi-projective (see Hartshorne 1977, Exercise 7.13). It is even complete — see

VI 2.3, there is an example of a nonsingular complete variety of dimension 3 that is not

below — and so if it were quasi-projective, it would be projective. In Shafarevich 1994,

projective. It is known that every irreducible separated curve is quasi-projective, and

every nonsingular complete surface is projective, and so these examples are minimal.

f. The homogeneous coordinate ring of a projective variety

Recall (p. 115) that attached to each irreducible variety 𝑉, there is a field 𝑘(𝑉) with the
property that 𝑘(𝑉) is the field of fractions of 𝑘[𝑈] for any open affine
[𝑋 𝑈 ⊂] 𝑉. We now
describe this field in the case that 𝑉 = ℙ𝑛 . Recall that 𝑘[𝑈0 ] = 𝑘 1 , … , 𝑛 . We regard
𝑋
𝑋0 𝑋0
this as a subring of 𝑘(𝑋0 , … , 𝑋𝑛 ), and wish to identify the field of fractions of 𝑘[𝑈0 ] as a
subfield of 𝑘(𝑋0 , … , 𝑋𝑛 ). Every nonzero 𝐹 ∈ 𝑘[𝑈0 ] can be written
(𝑋 ) 𝐹 ∗ (𝑋0 , … , 𝑋𝑛 )
𝐹 ,…, =
1 𝑋𝑛

𝑋0
𝑋0 𝑋0 deg(𝐹)

with 𝐹 ∗ homogeneous of degree deg(𝐹), and it follows that the field of fractions of 𝑘[𝑈0 ]

𝐺(𝑋0 , … , 𝑋𝑛 ) |||
is

𝑘(𝑈0 ) = { | 𝐺, 𝐻 homogeneous of the same degree} ∪ {0}.

𝐻(𝑋0 , … , 𝑋𝑛 ) |||
Write 𝑘(𝑋0 , … , 𝑋𝑛 )0 for this field (the subscript 0 is short for “subfield of elements of
degree 0”), so that 𝑘(ℙ𝑛 ) = 𝑘(𝑋0 , … , 𝑋𝑛 )0 . Note that for 𝐹 = in 𝑘(𝑋0 , … , 𝑋𝑛 )0 ,
𝐺
𝐻

𝐺(𝑎0 , … , 𝑎𝑛 )
(𝑎0 ∶ … ∶ 𝑎𝑛 ) ↦ ∶ 𝐷(𝐻) → 𝑘,
𝐻(𝑎0 , … , 𝑎𝑛 )
is a well-defined function, which is obviously regular (look at its restriction to 𝑈𝑖 ).
We now extend this discussion to any irreducible projective variety 𝑉. Such a 𝑉 can
be written 𝑉 = 𝑉(𝔭) with 𝔭 a graded radical ideal in 𝑘[𝑋0 , … , 𝑋𝑛 ], and we define the
homogeneous coordinate ring of 𝑉 (with its given embedding) to be

𝑘hom [𝑉] = 𝑘[𝑋0 , … , 𝑋𝑛 ]∕𝔭.

Note that 𝑘hom [𝑉] is the ring of regular functions on the affine cone over 𝑉; therefore its
dimension is dim(𝑉) + 1. It depends, not only on 𝑉, but on the embedding of 𝑉 into ℙ𝑛 ,
i.e., it is not intrinsic to 𝑉. For example,
𝜈
(𝑎0 ∶ 𝑎1 ) ↦ (𝑎02 ∶ 𝑎0 𝑎1 ∶ 𝑎12 ) ∶ ℙ1 ,→ ℙ2

is an isomorphism from ℙ1 onto its image 𝜈(ℙ1 ) ∶ 𝑋0 𝑋2 = 𝑋12 (see 6.23 below), but
𝑘hom [ℙ1 ] = 𝑘[𝑋0 , 𝑋1 ], which is the affine coordinate ring of the smooth variety 𝔸2 ,
whereas 𝑘hom [𝜈(ℙ1 )] = 𝑘[𝑋0 , 𝑋1 , 𝑋2 ]∕(𝑋0 𝑋2 − 𝑋12 ), which is the affine coordinate ring
of the singular variety 𝑋0 𝑋2 − 𝑋12 .
We say that a nonzero 𝑓 ∈ 𝑘hom [𝑉] is homogeneous of degree 𝑑 if it can be repre-
sented by a homogeneous polynomial 𝐹 of degree 𝑑 in 𝑘[𝑋0 , … , 𝑋𝑛 ], and we say that 0 is
homogeneous of degree 0.
 g. Regular functions on a projective variety 139

Lemma 6.14. Each element of 𝑘hom [𝑉] can be written uniquely in the form

𝑓 = 𝑓0 + ⋯ + 𝑓𝑑

with 𝑓𝑖 homogeneous of degree 𝑖.

Proof. Let 𝐹 represent 𝑓; then 𝐹 can be written 𝐹 = 𝐹0 + ⋯ + 𝐹𝑑 with 𝐹𝑖 homoge-

neous of degree 𝑖; when read modulo 𝔭, this gives a decomposition of 𝑓 of the required
∑
type. Suppose that 𝑓 also has a decomposition 𝑓 = 𝑔𝑖 , with 𝑔𝑖 represented by the
homogeneous polynomial 𝐺𝑖 of degree 𝑖. Then 𝐹 − 𝐺 ∈ 𝔭, and the homogeneity of 𝔭
implies that 𝐹𝑖 − 𝐺𝑖 = (𝐹 − 𝐺)𝑖 ∈ 𝔭. Therefore 𝑓𝑖 = 𝑔𝑖 . 2

It therefore makes sense to speak of homogeneous elements of 𝑘[𝑉]. For such an

element ℎ, we define 𝐷(ℎ) = {𝑃 ∈ 𝑉 ∣ ℎ(𝑃) ≠ 0}.
Since 𝑘hom [𝑉] is an integral domain, we can form its field of fractions 𝑘hom (𝑉).

𝑔 ||
Define

𝑘hom (𝑉)0 = { ∈ 𝑘hom (𝑉) ||| 𝑔 and ℎ homogeneous of the same degree} ∪ {0}.
ℎ ||

Proposition 6.15. The field of rational functions on 𝑉 is 𝑘(𝑉) = 𝑘hom (𝑉)0 .

def

Proof. Consider 𝑉0 = 𝑈0 ∩ 𝑉. As in the case of ℙ𝑛 , we can identify 𝑘[𝑉0 ] with a

subring of 𝑘hom [𝑉], and then the field of fractions of 𝑘[𝑉0 ] becomes identified with
def

𝑘hom (𝑉)0 . 2

g. Regular functions on a projective variety

Let 𝑉 be an irreducible projective variety, and let 𝑓 ∈ 𝑘(𝑉). By definition, we can write
𝑔
𝑓 = with 𝑔 and ℎ homogeneous of the same degree in 𝑘hom [𝑉] and ℎ ≠ 0. For any
ℎ
𝑃 = (𝑎0 ∶ … ∶ 𝑎𝑛 ) with ℎ(𝑃) ≠ 0,

𝑔(𝑎0 , … , 𝑎𝑛 )
𝑓(𝑃) =
ℎ(𝑎0 , … , 𝑎𝑛 )
def

is well-defined: if (𝑎0 , … , 𝑎𝑛 ) is replaced by (𝑐𝑎0 , … , 𝑐𝑎𝑛 ), then both the numerator and
denominator are multiplied by 𝑐deg(𝑔) = 𝑐deg(ℎ) .
𝑔
We can write 𝑓 in the form in many different ways,2 but if
ℎ

𝑔 𝑔′
𝑓= = ′ (in 𝑘(𝑉)0 ),
ℎ ℎ

𝑔ℎ′ = 𝑔′ ℎ (in 𝑘hom [𝑉])

then

𝑔(𝑎0 , … , 𝑎𝑛 ) ⋅ ℎ′ (𝑎0 , … , 𝑎𝑛 ) = 𝑔′ (𝑎0 , … , 𝑎𝑛 ) ⋅ ℎ(𝑎0 , … , 𝑎𝑛 ).

and so

Thus, if ℎ′ (𝑃) ≠ 0, the two representations give the same value for 𝑓(𝑃).
𝑔
Unless 𝑘hom [𝑉] is a unique factorization domain, there will be no preferred representation 𝑓 =
ℎ
2
.
140 6. Projective Varieties

Proposition 6.16. For each 𝑓 ∈ 𝑘(𝑉) = 𝑘hom (𝑉)0 , there is an open subset 𝑈 of 𝑉, where
𝑓(𝑃) is defined, and 𝑃 ↦ 𝑓(𝑃) is a regular function on 𝑈; every regular function on an
def

open subset of 𝑉 arises from a unique element of 𝑘(𝑉).

Proof. From the above discussion, we see that 𝑓 defines a regular function on 𝑈 =
⋃ 𝑔
𝐷(ℎ), where ℎ runs over the denominators of expressions 𝑓 = with 𝑔 and ℎ homo-
ℎ
geneous of the same degree in 𝑘hom [𝑉].
Conversely, let 𝑓 be a regular function on an open subset 𝑈 of 𝑉, and let 𝑃 ∈ 𝑈.
Then 𝑃 lies in the open affine subvariety 𝑉 ∩ 𝑈𝑖 for some 𝑖, and so 𝑓 coincides with the
function defined by some 𝑓𝑃 ∈ 𝑘(𝑉 ∩ 𝑈𝑖 ) = 𝑘(𝑉) on an open neighbourhood of 𝑃. If
𝑓 coincides with the function defined by 𝑓𝑄 ∈ 𝑘(𝑉) in a neighbourhood of a second
point 𝑄 of 𝑈, then 𝑓𝑃 and 𝑓𝑄 define the same function on some open affine 𝑈 ′ , and so
𝑓𝑃 = 𝑓𝑄 as elements of 𝑘[𝑈 ′ ] ⊂ 𝑘(𝑉). This shows that 𝑓 is the function defined by 𝑓𝑃
on the whole of 𝑈. 2

Remark 6.17. (a) The elements of 𝑘(𝑉) = 𝑘hom (𝑉)0 should be regarded as the algebraic

an open subset 𝑈 of 𝑉 are the “meromorphic functions without poles” on 𝑈. [In fact,
analogues of meromorphic functions on a complex manifold; the regular functions on

when 𝑘 = ℂ, this is more than an analogy: a nonsingular projective algebraic variety

over ℂ defines a complex manifold, and the meromorphic functions on the manifold are

on the Riemann sphere are the rational functions in 𝑧.]

precisely the rational functions on the variety. For example, the meromorphic functions

(b) We shall see presently (6.24) that, for any nonzero homogeneous ℎ ∈ 𝑘hom [𝑉],
𝐷(ℎ) is an open affine subset of 𝑉. The ring of regular functions on it is

𝑘[𝐷(ℎ)] = {𝑔∕ℎ𝑚 ∣ 𝑔 homogeneous of degree 𝑚 deg(ℎ)} ∪ {0}.

We shall also see that the ring of regular functions on 𝑉 itself is just 𝑘, i.e., any regular

𝑈 is an open nonaffine subset of 𝑉, then the ring 𝛤(𝑈, 𝒪𝑉 ) of regular functions can be
function on an irreducible (connected will do) projective variety is constant. However, if

almost anything — it need not even be a finitely generated 𝑘-algebra!

h. Maps from projective varieties

We describe the morphisms from a projective variety to another variety.

Proposition 6.18. The map

𝜋 ∶ 𝔸𝑛+1 ∖ {origin} → ℙ𝑛 , (𝑎0 , … , 𝑎𝑛 ) ↦ (𝑎0 ∶ … ∶ 𝑎𝑛 )

is an open morphism of algebraic varieties. A map 𝛼 ∶ ℙ𝑛 → 𝑉 with 𝑉 a prevariety is

regular if and only if 𝛼◦𝜋 is regular.

Proof. The restriction of 𝜋 to 𝐷(𝑋𝑖 ) is the projection

(𝑎 )
(𝑎0 , … , 𝑎𝑛 ) ↦ 0 ∶ … ∶ 𝑛 ∶ 𝑘 𝑛+1 ∖ 𝑉(𝑋𝑖 ) → 𝑈𝑖 ,
𝑎
𝑎𝑖 𝑎𝑖

which is the regular map of affine varieties corresponding to the map of 𝑘-algebras
[𝑋 ]
𝑘 0 , … , 𝑛 → 𝑘[𝑋0 , … , 𝑋𝑛 ][𝑋𝑖−1 ].
𝑋
𝑋𝑖 𝑋𝑖
 h. Maps from projective varieties 141

𝑋𝑗
(In the first algebra 𝑋 is to be thought of as a single symbol.) It now follows from
𝑖
Proposition 5.4 that 𝜋 is regular.
Let 𝑈 be an open subset of 𝑘 𝑛+1 ∖ {origin}, and let 𝑈 ′ be the union of all the lines
through the origin that intersect 𝑈, that is, 𝑈 ′ = 𝜋−1 𝜋(𝑈). Then 𝑈 ′ is again open
⋃
in 𝑘 𝑛+1 ∖ {origin}, because 𝑈 ′ = 𝑐𝑈, 𝑐 ∈ 𝑘 × , and 𝑥 ↦ 𝑐𝑥 is an automorphism of
𝑘 𝑛+1 ∖ {origin}. The complement 𝑍 of 𝑈 ′ in 𝑘 𝑛+1 ∖ {origin} is a closed cone, and the proof
of (6.3) shows that its image is closed in ℙ𝑛 ; but 𝜋(𝑈) is the complement of 𝜋(𝑍). Thus
𝜋 sends open sets to open sets.
The rest of the proof is straightforward.

Thus, the regular maps ℙ𝑛 → 𝑉 are just the regular maps 𝔸𝑛+1 ∖ {origin} → 𝑉
2

factoring through ℙ𝑛 as maps of sets.

Remark 6.19. Consider polynomials 𝐹0 (𝑋0 , … , 𝑋𝑚 ), … , 𝐹𝑛 (𝑋0 , … , 𝑋𝑚 ) of the same de-

gree. The map

(𝑎0 ∶ … ∶ 𝑎𝑚 ) ↦ (𝐹0 (𝑎0 , … , 𝑎𝑚 ) ∶ … ∶ 𝐹𝑛 (𝑎0 , … , 𝑎𝑚 ))

obviously defines a regular map to ℙ𝑛 on the open subset of ℙ𝑚 , where not all 𝐹𝑖 vanish,
⋃
that is, on the set 𝐷(𝐹𝑖 ) = ℙ𝑛 ∖ 𝑉(𝐹1 , … , 𝐹𝑛 ). Its restriction to any subvariety 𝑉 of ℙ𝑚
will also be regular. It may be possible to extend the map to a larger set by representing
it by different polynomials. Conversely, every such map arises in this way, at least locally.
More precisely, there is the following result.

Proposition 6.20. Let 𝑉 = 𝑉(𝔞) ⊂ ℙ𝑚 and 𝑊 = 𝑉(𝔟) ⊂ ℙ𝑛 . A map 𝜑 ∶ 𝑉 → 𝑊 is

regular if and only if, for every 𝑃 ∈ 𝑉, there exist polynomials

𝐹0 (𝑋0 , … , 𝑋𝑚 ), … , 𝐹𝑛 (𝑋0 , … , 𝑋𝑚 ),

homogeneous of the same degree, such that

𝜑 ((𝑏0 ∶ … ∶ 𝑏𝑛 )) = (𝐹0 (𝑏0 , … , 𝑏𝑚 ) ∶ … ∶ 𝐹𝑛 (𝑏0 , … , 𝑏𝑚 ))

for all points (𝑏0 ∶ … ∶ 𝑏𝑚 ) in some neighbourhood of 𝑃 in 𝑉(𝔞).

Proof. Straightforward.

Example 6.21. We prove that the circle 𝑋 2 + 𝑌 2 = 𝑍 2 is isomorphic to ℙ1 (char(𝑘) ≠ 2).

2

This equation can be rewritten (𝑋 + 𝑖𝑌)(𝑋 − 𝑖𝑌) = 𝑍 2 , and so, after a change of variables,
the equation of the circle becomes 𝐶 ∶ 𝑋𝑍 = 𝑌 2 . Define

𝜑 ∶ ℙ1 → 𝐶, (𝑎 ∶ 𝑏) ↦ (𝑎2 ∶ 𝑎𝑏 ∶ 𝑏2 ).

(𝑎 ∶ 𝑏 ∶ 𝑐) ↦ (𝑎 ∶ 𝑏) if 𝑎 ≠ 0
For the inverse, define

𝜓 ∶ 𝐶 → ℙ1 by { .
(𝑎 ∶ 𝑏 ∶ 𝑐) ↦ (𝑏 ∶ 𝑐) if 𝑏 ≠ 0

𝑐 𝑏
𝑎 ≠ 0 ≠ 𝑏, 𝑎𝑐 = 𝑏2 ⇐⇒ =𝑎
Note that,

𝑏
and so the two maps agree on the set where they are both defined. Clearly, both 𝜑 and 𝜓
are regular, and one checks directly that they are inverse.
142 6. Projective Varieties

i. Some classical maps of projective varieties

A hypersurface of degree 𝑚 in ℙ𝑛 is an algebraic subset defined by nonzero homoge-
neous of degree 𝑚 ≥ 1. When 𝑚 = 1, the hypersurface is called a hyperplane. As in
the affine case (2.67), the hypersurfaces in ℙ𝑛 are exactly the closed subvarieties of ℙ𝑛 of
codimension 1. The intersection of a projective variety 𝑊 ⊂ ℙ𝑛 with a hypersurface (of
degree 𝑚) is called a hypersurface section (of degree 𝑚) of 𝑊.
After proving that complements of hyperplane sections are affine, we study some
classical maps of projective varieties.

Hyperplane sections and complements

We show that the complement of a hyperplane section of a projective variety is an affine
variety, and deduce that any finite set of points of a projective variety is contained in an

∑
6.22. Let 𝐿 = 𝑐𝑖 𝑋𝑖 be a nonzero linear form in 𝑛 + 1 variables. Then the map
affine subvariety.

𝑎0 𝑎𝑛
(𝑎0 ∶ … ∶ 𝑎𝑛 ) ↦ ( ,…, )
𝐿(𝐚) 𝐿(𝐚)

is a bijection of 𝐷(𝐿) ⊂ ℙ𝑛 onto the hyperplane 𝐿(𝑋0 , 𝑋1 , … , 𝑋𝑛 ) = 1 of 𝔸𝑛+1 , with

(𝑎0 , … , 𝑎𝑛 ) ↦ (𝑎0 ∶ … ∶ 𝑎𝑛 ).
inverse

𝑋𝑗
functions ∑ . As 𝑉(𝐿 − 1) is affine, so also is 𝐷(𝐿), and its ring of regular functions
Both maps are regular — for example, the components of the first map are the regular

𝑐𝑖 𝑋𝑖
[ 𝑋 𝑋 ] 𝑋𝑗
is 𝑘 ∑ 0 , … , ∑ 𝑛 . In this ring, each quotient ∑
𝑐𝑖 𝑋𝑖 𝑐𝑖 𝑋𝑖 𝑐𝑖 𝑋𝑖
∑ 𝑋𝑗
is to be thought of as a single

symbol, and 𝑐𝑗 ∑ = 1; thus it is a polynomial ring in 𝑛 symbols; any one symbol

𝑐𝑖 𝑋𝑖
𝑋
∑ 𝑗 for which 𝑐𝑗 ≠ 0 can be omitted.
𝑐𝑖 𝑋𝑖
For a fixed 𝑃 = (𝑎0 ∶ … ∶ 𝑎𝑛 ) ∈ ℙ𝑛 , the set of 𝐜 = (𝑐0 ∶ … ∶ 𝑐𝑛 ) such that
def ∑
𝐿𝐜 (𝑃) = 𝑐𝑖 𝑎𝑖 ≠ 0

is a nonempty open subset of ℙ𝑛 (𝑛 > 0). Therefore, for any finite set 𝑆 of points of ℙ𝑛 ,

{𝐜 ∈ ℙ𝑛 ∣ 𝑆 ⊂ 𝐷(𝐿𝐜 )}

is a nonempty open subset of ℙ𝑛 (because ℙ𝑛 is irreducible). In particular, 𝑆 is contained

in an open affine subset 𝐷(𝐿𝐜 ) of ℙ𝑛 . Moreover, if 𝑆 ⊂ 𝑉, where 𝑉 is a closed subvariety
of ℙ𝑛 , then 𝑆 ⊂ 𝑉 ∩ 𝐷(𝐿𝐜 ): any finite set of points of a projective variety is contained in
an open affine subvariety.

The Veronese map embeds ℙ𝑛 in a higher dimensional projective space in such a way
The Veronese map

that hypersurface sections of subvarieties are transformed into hyperplane sections.

 i. Some classical maps of projective varieties 143

∑
𝐼 = {(𝑖0 , … , 𝑖𝑛 ) ∈ ℕ𝑛+1 ∣ 𝑖𝑗 = 𝑚}.
6.23. Let

Note that 𝐼 indexes the monomials of degree 𝑚 in 𝑛 + 1 variables. It has ( 𝑚+𝑛

𝑚 ) elements
(see 6.39 below). Write 𝜈𝑛,𝑚 = ( 𝑚+𝑛
𝑚 ) − 1, ℙ𝜈𝑛,𝑚 whose
coordinates are indexed by 𝐼; thus a point of ℙ𝜈𝑛,𝑚 can be written (… ∶ 𝑏𝑖0 …𝑖𝑛 ∶ …). The
and consider the projective space

𝑣 ∶ ℙ𝑛 → ℙ𝜈𝑛,𝑚 , (𝑎0 ∶ … ∶ 𝑎𝑛 ) ↦ (… ∶ 𝑏𝑖0 …𝑖𝑛 ∶ …), 𝑏𝑖0 …𝑖𝑛 = 𝑎00 … 𝑎𝑛𝑛 .

Veronese mapping is defined to be
𝑖 𝑖

In other words, the Veronese mapping sends an 𝑛 + 1-tuple (𝑎0 ∶ … ∶ 𝑎𝑛 ) to the set of
monomials in the 𝑎𝑖 of degree 𝑚. For example, when 𝑛 = 1 and 𝑚 = 2, the Veronese

(𝑎0 ∶ 𝑎1 ) ↦ (𝑎02 ∶ 𝑎0 𝑎1 ∶ 𝑎12 ) ∶ ℙ1 → ℙ2 .

map is

Its image is the curve 𝜈(ℙ1 ) ∶ 𝑋0 𝑋2 = 𝑋12 , and the map

(𝑏2,0 ∶ 𝑏1,1 ) if 𝑏2,0 ≠ 1
(𝑏2,0 ∶ 𝑏1,1 ∶ 𝑏0,2 ) ↦ {
(𝑏1,1 ∶ 𝑏0,2 ) if 𝑏0,2 ≠ 0
is an inverse 𝜈(ℙ1 ) → ℙ1 . (Cf. Example 6.22.)
When 𝑛 = 1 and 𝑚 is general, the Veronese map is
(𝑎0 ∶ 𝑎1 ) ↦ (𝑎0𝑚 ∶ 𝑎0𝑚−1 𝑎1 ∶ … ∶ 𝑎1𝑚 ) ∶ ℙ1 → ℙ𝑚 .
We shall show that, in the general case, the image of 𝜈 is a closed subset of ℙ𝜈𝑛,𝑚 and
that 𝜈 defines an isomorphism of projective varieties 𝜈 ∶ ℙ𝑛 → 𝜈(ℙ𝑛 ).
∑
𝑎𝑖 of a point 𝑃 of ℙ𝑛 as being the coefficients of a linear form 𝐿 = 𝑎𝑖 𝑋𝑖 (well-defined
First note that the map has the following interpretation: if we regard the coordinates

up to multiplication by nonzero scalar), then the coordinates of 𝜈(𝑃) are the coefficients
of the homogeneous polynomial 𝐿𝑚 with the binomial coefficients omitted.
As 𝐿 ≠ 0 ⇒ 𝐿𝑚 ≠ 0, the map 𝜈 is defined on the whole of ℙ𝑛 , that is,
(𝑎0 , … , 𝑎𝑛 ) ≠ (0, … , 0) ⇒ (… , 𝑏𝑖0 …𝑖𝑛 , …) ≠ (0, … , 0).

𝐿1 ≠ 𝑐𝐿2 ⇒ 𝐿1𝑚 ≠ 𝑐𝐿2𝑚

Moreover,

because 𝑘[𝑋0 , … , 𝑋𝑛 ] is a unique factorization domain. Therefore, 𝜈 is injective, and it is

obvious from its definition that it is regular.

regular map is closed, but here we can prove directly that 𝜈(ℙ𝑛 ) is defined by the system
We shall see in the next chapter that the image of any projective variety under a

𝑏𝑖0 …𝑖𝑛 𝑏𝑗0 …𝑗𝑛 = 𝑏𝑘0 …𝑘𝑛 𝑏𝓁0 …𝓁𝑛 , 𝑖ℎ + 𝑗ℎ = 𝑘ℎ + 𝓁ℎ , all ℎ.

of equations

Obviously ℙ𝑛 maps into the algebraic set defined by these equations. Conversely, let
(*)

𝑉𝑖 = {(… . ∶ 𝑏𝑖0 …𝑖𝑛 ∶ …) ∣ 𝑏0…0𝑚0…0 ≠ 0}.

Then 𝜈(𝑈𝑖 ) ⊂ 𝑉𝑖 and 𝜈−1 (𝑉𝑖 ) = 𝑈𝑖 . It is possible to write down a regular map 𝑉𝑖 → 𝑈𝑖
inverse to 𝜈|𝑈𝑖 : for example, define 𝑉0 → ℙ𝑛 to be
(… ∶ 𝑏𝑖0 …𝑖𝑛 ∶ …) ↦ (𝑏𝑚,0,…,0 ∶ 𝑏𝑚−1,1,0,…,0 ∶ 𝑏𝑚−1,0,1,0,…,0 ∶ … ∶ 𝑏𝑚−1,0,…,0,1 ).
⋃
Finally, one checks that 𝜈(ℙ𝑛 ) ⊂ 𝑉𝑖 .
𝜈
As ℙ𝑛 ,→ 𝜈(ℙ𝑛 ) is an isomorphism, for any closed subvariety 𝑊 of 𝑉, 𝜈(𝑊) is a
closed subvariety of 𝜈(ℙ𝑛 ) (hence of ℙ𝜈𝑛,𝑚 ) and 𝜈|𝑊 ∶ 𝑊 → 𝜈(𝑊) is an isomorphism.
144 6. Projective Varieties

6.24. The Veronese mapping has a very important property. Let 𝐻 be the hypersurface
in ℙ𝑛 of degree 𝑚 ∑
𝑎𝑖0 …𝑖𝑛 𝑋00 ⋯ 𝑋𝑛𝑛 = 0,
𝑖 𝑖

and let 𝐿 be the hyperplane in ℙ𝜈𝑛,𝑚 defined by

∑
𝑎𝑖0 …𝑖𝑛 𝑋𝑖0 …𝑖𝑛 .

Then 𝜈(𝐻) = 𝜈(ℙ𝑛 ) ∩ 𝐿, i.e.,

𝐻(𝐚) = 0 ⇐⇒ 𝐿(𝜈(𝐚)) = 0.

Thus for any closed subvariety 𝑊 of ℙ𝑛 , 𝜈 defines an isomorphism of the hypersurface

section 𝑊 ∩ 𝐻 of 𝑉 onto the hyperplane section 𝜈(𝑊) ∩ 𝐿 of 𝜈(𝑊). This observation
often allows us to replace questions about hypersurface sections with questions about

As one example of this, note that 𝜈 maps the complement of a hypersurface section
hyperplane sections.

of 𝑊 isomorphically onto the complement of a hyperplane section of 𝜈(𝑊), which we

know to be affine. Thus the complement of any hypersurface section of a projective
variety is an affine variety.

Automorphisms of ℙ𝑛
We show that the automorphisms of ℙ𝑛 are exactly the invertible changes of variables.

6.25. A Möbius transformation of ℙ1 is a regular map of the form

(𝑥 ∶ 𝑦) ↦ (𝑎𝑥 + 𝑏𝑦 ∶ 𝑐𝑥 + 𝑑𝑦) ∶ ℙ1 → ℙ1 ,

where 𝑎, 𝑏, 𝑐, 𝑑 ∈ 𝑘 are such that 𝑎𝑑 − 𝑏𝑐 ≠ 0. The Möbius transformations are exactly

the automorphisms of ℙ1 , and two quadruples 𝑎, 𝑏, 𝑐, 𝑑 define the same transformation
if and only if one is a nonzero multiple of the other.3 Thus,

1 0
Aut(ℙ1 ) = PGL2 (𝑘) = GL2 (𝑘)∕𝑘 × 𝐼, where 𝐼 = ( ).
0 1
def

A similar statement is true for ℙ𝑛 . An element 𝐴 = (𝑎𝑖𝑗 ) of GL𝑛+1 defines a regular

∑
(𝑥0 ∶ … ∶ 𝑥𝑛 ) ↦ (… ∶ 𝑎𝑖𝑗 𝑥𝑗 ∶ …) ∶ ℙ𝑛 → ℙ𝑛′
map

It is an automorphism with inverse defined by the inverse matrix. Scalar matrices act as

Let PGL𝑛+1 = GL𝑛+1 ∕𝑘 × 𝐼, where 𝐼 is the identity matrix, that is, PGL𝑛+1 is the
the identity map.

quotient of GL𝑛+1 by its centre. Then PGL𝑛+1 is the complement in ℙ(𝑛+1) −1 of the
2

hypersurface det(𝑋𝑖𝑗 ) = 0, and so it is an affine variety with ring of regular functions

𝑘[PGL𝑛+1 ] = {𝐹(… , 𝑋𝑖𝑗 , …)∕ det(𝑋𝑖𝑗 )𝑚 ∣ deg(𝐹) = 𝑚 ⋅ (𝑛 + 1)} ∪ {0}.

It is an affine group variety.

3
Therefore, when 𝑘 = ℂ, the automorphisms of ℙ1 coincide with the holomorphic automorphisms of
the Riemann sphere (Cartan 1963, VI.2.4).
 i. Some classical maps of projective varieties 145

The homomorphism PGL𝑛+1 → Aut(ℙ𝑛 ) is obviously injective. We sketch a proof

First note that the collection of lines in ℙ𝑛 has a natural structure of an algebraic
that it is surjective.4

variety and, in particular, a (Zariski) topology. Indeed, the lines in ℙ𝑛 correspond to

2-dimensional subspaces of 𝑘 𝑛+1 , and hence to the points on the Grassmann variety
𝐺2 (𝑘𝑛+1 ) (see section 6m below).

𝐻 ∶ 𝐹(𝑋0 , … , 𝑋𝑛 ) = 0
Fix a hypersurface

in ℙ𝑛 and consider a line

𝐿 = {(𝑎0 + 𝑡𝑏0 ∶ … ∶ 𝑎𝑛 + 𝑡𝑏𝑛 ) ∣ 𝑡 ∈ 𝑘}

in ℙ𝑛 . The points of 𝐻 ∩ 𝐿 are given by the solutions of

𝐹(𝑎0 + 𝑡𝑏0 ∶ … ∶ 𝑎𝑛 + 𝑡𝑏𝑛 ) = 0,

which is a polynomial of degree ≤ deg(𝐹) in 𝑡 unless 𝐿 ⊂ 𝐻. Therefore, if 𝐿 ⊄ 𝐻, then

𝐻 ∩ 𝐿 contains at most deg(𝐹) points, and it is not hard to show that, for the 𝐿 in an
open subset of the space of all lines, it will contain exactly deg(𝐹) points. Thus, the
hyperplanes are exactly the closed subvarieties 𝐻 of ℙ𝑛 such that
(a) dim(𝐻) = 𝑛 − 1,
(b) |𝐻 ∩ 𝐿| = 1 for an open set of lines 𝐿.
These are geometric conditions, and so any automorphism of ℙ𝑛 must map hyperplanes
to hyperplanes. But on an open subset of ℙ𝑛 , any automorphism takes the form

(𝑏0 ∶ … ∶ 𝑏𝑛 ) ↦ (𝐹0 (𝑏0 , … , 𝑏𝑛 ) ∶ … ∶ 𝐹𝑛 (𝑏0 , … , 𝑏𝑛 )),

where the 𝐹𝑖 are homogeneous of the same degree 𝑑 (see 6.20). Such a map will take
hyperplanes to hyperplanes if and only if 𝑑 = 1.

The Segre map

The Segre map embeds a product of projective spaces into a projective space, and allows
us to show that products of projective varieties are projective.

6.26. The Segre map is

((𝑎0 ∶ … ∶ 𝑎𝑚 ), (𝑏0 ∶ … ∶ 𝑏𝑛 )) ↦ (𝑎0 𝑏0 ∶ … ∶ 𝑎𝑖 𝑏𝑗 ∶ …)) ∶ ℙ𝑚 × ℙ𝑛 → ℙ𝑚𝑛+𝑚+𝑛 .

𝑤00 𝑤𝑖𝑗

The index set for ℙ𝑚𝑛+𝑚+𝑛 is {(𝑖, 𝑗) ∣ 0 ≤ 𝑖 ≤ 𝑚, 0 ≤ 𝑗 ≤ 𝑛}. Note that if we interpret the
∑ ∑
tuples on the left as the coefficients of two linear forms 𝐿1 = 𝑎𝑖 𝑋𝑖 and 𝐿2 = 𝑏𝑗 𝑌𝑗 ,
then the image of the pair is the set of coefficients of 𝐿1 𝐿2 , which is a homogeneous form
of degree 2. From this observation, it is obvious that the map is defined on the whole of
ℙ𝑚 × ℙ𝑛 ,
𝐿1 ≠ 0 ≠ 𝐿2 ⇐⇒ 𝐿1 𝐿2 ≠ 0,
and is injective. Its image is obviously contained the hypersurface

𝐻 ∶ 𝑊𝑖𝑗 𝑊𝑘𝑙 − 𝑊𝑖𝑙 𝑊𝑘𝑗 = 0.

4
This is related to the fundamental theorem of projective geometry. See Wikipedia: Fundamental
theorem of projective geometry or E. Artin, Geometric Algebra, Interscience, 1957, Theorem 2.26.
146 6. Projective Varieties

In fact, the Segre map is an isomorphism

≃
ℙ𝑚 × ℙ𝑛 ,→ 𝐻 ⊂ ℙ𝑚𝑛+𝑚+𝑛 .

To see this, note that the Segre map defines an isomorphism from the open affine ℙ𝑚 ×ℙ𝑛
where 𝑎0 𝑏0 ≠ 0 onto the open affine of 𝐻 where 𝑤00 ≠ 0, with inverse

(𝑤00 ∶ … ∶ 𝑤𝑖𝑗 ∶ …) ↦ ((𝑤00 ∶ … ∶ 𝑤𝑚0 ), (𝑤00 ∶ … ∶ 𝑤0𝑛 )),

and that a similar statement holds with 0, 0 replaced by 𝑖, 𝑗.

For example, the map

((𝑎0 ∶ 𝑎1 ), (𝑏0 ∶ 𝑏1 )) ↦ (𝑎0 𝑏0 ∶ 𝑎0 𝑏1 ∶ 𝑎1 𝑏0 ∶ 𝑎1 𝑏1 ) ∶ ℙ1 × ℙ1 → ℙ3

𝑤 𝑥 𝑦 𝑧

is an isomorphism from ℙ1 × ℙ1 onto the hypersurface

𝐻∶ 𝑊𝑍 = 𝑋𝑌,

(𝑤 ∶ 𝑥 ∶ 𝑦 ∶ 𝑧) ↦ ((𝑤 ∶ 𝑦), (𝑤 ∶ 𝑥))

with inverse

on the open affine of 𝐻 where 𝑤 ≠ 0.

In particular, we see that ℙ1 × ℙ1 is a projective variety. It is not isomorphic to ℙ2 ,
because, in ℙ2 , any two closed curves intersect (section 6p), whereas, in ℙ1 × ℙ1 , this is

If 𝑉 and 𝑊 are closed subvarieties of ℙ𝑚 and ℙ𝑛 , then the Segre map sends 𝑉 × 𝑊
not true (consider two vertical lines).

isomorphically onto a closed subvariety of ℙ𝑚𝑛+𝑚+𝑛 . Thus products of projective varieties

The product ℙ1 × ℙ𝑛 contains many disjoint copies of ℙ𝑛 as closed subvarieties.

are projective.

Thus finite disjoint unions of copies of ℙ𝑛 are projective, and so finite disjoint unions of

There is an explicit description of the topology on ℙ𝑚 × ℙ𝑛 : the closed subsets are

projective varieties are projective.

the sets of common solutions of families of equations

𝐹(𝑋0 , … , 𝑋𝑚 ; 𝑌0 , … , 𝑌𝑛 ) = 0

with 𝐹 separately homogeneous in the 𝑋𝑖 and in the 𝑌𝑗 .

Projections with given centre

Projections with a given centre allow us to map closed subvarieties of a projective space
onto closed subvarieties of a lower-dimensional projective space, possibly with the

6.27. Let 𝐿1 , … , 𝐿𝑛−𝑑 be linearly independent linear forms in 𝑛 + 1 variables. Their

introduction of singularities.

zero set 𝐸 in 𝑘𝑛+1 has dimension 𝑑 + 1, and so their zero set in ℙ𝑛 is a 𝑑-dimensional
linear space. Define 𝜋 ∶ ℙ𝑛 − 𝐸 → ℙ𝑛−𝑑−1 by 𝜋(𝑎) = (𝐿1 (𝑎) ∶ … ∶ 𝐿𝑛−𝑑 (𝑎)); such a map
is called a projection with centre 𝐸. If 𝑉 is a closed subvariety disjoint from 𝐸, then
𝜋 defines a regular map 𝜑 ∶ 𝑉 → ℙ𝑛−𝑑−1 . Its image is closed (7.22, 7.7) and the map
𝜑 ∶ 𝑉 → 𝜑(𝑉) is finite (8.53).
More generally, if 𝐹1 , … , 𝐹𝑟 are homogeneous forms of the same degree, and 𝑍 =
𝑉(𝐹1 , … , 𝐹𝑟 ), then 𝑎 ↦ (𝐹1 (𝑎) ∶ … ∶ 𝐹𝑟 (𝑎)) is a morphism ℙ𝑛 − 𝑍 → ℙ𝑟−1 .
 j. Projective space without coordinates 147

By carefully choosing the centre 𝐸, it is possible to linearly project any smooth curve
in ℙ𝑛 isomorphically onto a curve in ℙ3 , and nonisomorphically (but bijectively on an
open subset) onto a curve in ℙ2 with only nodes as singularities.5 For example, suppose
that we have a nonsingular curve 𝐶 in ℙ3 . To project to ℙ2 we need three linear forms
𝐿0 , 𝐿1 , 𝐿2 and the centre of the projection is the point 𝑃0 where all the forms are zero.
We can think of the map as projecting from the centre 𝑃0 onto some (projective) plane
by sending the point 𝑃 to the point where 𝑃0 𝑃 intersects the plane. To project 𝐶 to a
curve with only ordinary nodes as singularities, one needs to choose 𝑃0 so that it does
not lie on any tangent to 𝐶, any trisecant (line crossing the curve in 3 points), or any

Projecting a nonsingular variety in ℙ𝑛 to a lower dimensional projective space usually

chord at whose extremities the tangents are coplanar. See for example Samuel 1966.

introduces singularities. Hironaka proved that every singular variety arises in this way
in characteristic zero. See Chapter 8 below.

Proposition 6.28. Every finite set 𝑆 of points of a quasi-projective variety 𝑉 is contained

Application

in an open affine subset of 𝑉.

Proof. Regard 𝑉 as a subvariety of ℙ𝑛 , let 𝑉̄ be the closure of 𝑉 in ℙ𝑛 , and let 𝑍 = 𝑉̄ ∖ 𝑉.

Because 𝑆 ∩ 𝑍 = ∅, for each 𝑃 ∈ 𝑆 there exists a homogeneous polynomial 𝐹𝑃 ∈ 𝐼(𝑍)
such that 𝐹𝑃 (𝑃) ≠ 0. We may suppose that the 𝐹𝑃 have the same degree. An elementary
argument shows that some linear combination 𝐹 of the 𝐹𝑃 , 𝑃 ∈ 𝑆, is nonzero at each 𝑃.
Then 𝐹 is zero on 𝑍, and so 𝑉̄ ∩ 𝐷(𝐹) is an open affine of 𝑉, but 𝐹 is nonzero at each 𝑃,
and so 𝑉̄ ∩ 𝐷(𝐹) contains 𝑆. 2

j. Projective space without coordinates

Let 𝐸 be a vector space over 𝑘 of dimension 𝑛. The set ℙ(𝐸) of lines through zero in 𝐸 has
a natural structure of an algebraic variety: the choice of a basis for 𝐸 defines a bijection
ℙ(𝐸) → ℙ𝑛 , and the inherited structure of an algebraic variety on ℙ(𝐸) is independent

by an automorphism of ℙ𝑛 ). Note that in contrast to ℙ𝑛 , which has 𝑛 + 1 distinguished

of the choice of the basis (because the bijections defined by two different bases differ

hyperplanes, namely, 𝑋0 = 0, … , 𝑋𝑛 = 0, no hyperplane in ℙ(𝐸) is distinguished.

k. The functor defined by projective space

Let 𝑅 be a 𝑘-algebra. A submodule 𝑀 of an 𝑅-module 𝑁 is said to be a direct summand of
𝑁 if there exists another submodule 𝑀 ′ of 𝑀 (a complement of 𝑀) such that 𝑁 = 𝑀⊕𝑀 ′ .
Let 𝑀 be a direct summand of a finitely generated projective 𝑅-module 𝑁. Then 𝑀 is
also finitely generated and projective, and so 𝑀𝔪 is a free 𝑅𝔪 -module of finite rank for
every maximal ideal 𝔪 in 𝑅. If 𝑀𝔪 is of constant rank 𝑟, then we say that 𝑀 has rank 𝑟.
See CA, §12.

𝑃𝑛 (𝑅) = {direct summands of rank 1 of 𝑅𝑛+1 }.

Let

This is best possible: a nonsingular curve of degree 𝑑 in ℙ2 has genus (𝑑 − 1)(𝑑 − 2)∕2, and so, if 𝑔 is
not of this form, no curve of genus 𝑔 can be realized as a nonsingular curve in ℙ2 .
5
148 6. Projective Varieties

Then 𝑃𝑛 is a functor from 𝑘-algebras to sets. When 𝐾 is a field, every 𝐾-subspace of 𝐾 𝑛+1
is a direct summand, and so 𝑃𝑛 (𝐾) consists of the lines through the origin in 𝐾 𝑛+1 .
Let 𝐻𝑖 be the hyperplane 𝑋𝑖 = 0 in 𝑘𝑛+1 , and let

𝑃𝑖 (𝑅) = {𝐿 ∈ 𝑃𝑛 (𝑅) ∣ 𝐿 ⊕ 𝐻𝑖𝑅 = 𝑅𝑛+1 }.

Let 𝐿 ∈ 𝑃𝑖 (𝑅); then ∑

𝑒𝑖 = 𝓁 + 𝑎𝑗 𝑒𝑗 .
𝑗≠𝑖

𝐿 ↦ (𝑎𝑗 )𝑗≠𝑖 ∶ 𝑃𝑖 (𝑅) → 𝑈𝑖 (𝑅) ≃ 𝑅𝑛

Now

is a bijection. These combine to give an isomorphism 𝑃𝑛 (𝑅) → ℙ𝑛 (𝑅):

∏ ∏
𝑃𝑛 (𝑅) 𝑃𝑖 (𝑅) 𝑃𝑖 (𝑅) ∩ 𝑃𝑗 (𝑅)
0≤𝑖≤𝑛 0≤𝑖,𝑗≤𝑛

∏ ∏
ℙ𝑛 (𝑅) 𝑈𝑖 (𝑅) 𝑈𝑖 (𝑅) ∩ 𝑈𝑗 (𝑅).
0≤𝑖≤𝑛 0≤𝑖,𝑗≤𝑛

Let 𝑅 be a commutative ring, and let 𝐿 be a direct summand of rank 1 of 𝑅𝑛+1 . Then
𝐿 is a projective 𝑅-module of rank 1 and the images 𝑠0 , is a projective

l. Maps to projective space

In this section, we assume the reader is familiar with the definitions of coherent sheaves

To give a regular map from a variety 𝑉 to ℙ𝑛 is the same as giving an isomorphism

and vector bundles (Chapter 13).

class of pairs (𝐿, (𝑠0 , … , 𝑠𝑛 )) where 𝐿 is an invertible sheaf on 𝑉 and 𝑠0 , … , 𝑠𝑛 are global
sections of 𝐿
Let 𝑉 be a complete variety. A map 𝜑 ∶ 𝑉 → ℙ𝑛 is an isomorphism onto its (closed)

For 𝐷 a divisor on a variety 𝑉, we let

image if and only if it separates points and tangent directions (ZMT).

𝐿(𝐷) = {𝑓 ∈ 𝑘(𝑉) ∣ (𝑓) + 𝐷 ≥ 0} ∪ {0} = 𝐻 0 (𝑉, ℒ(𝐷)),

|𝐷| = {(𝑓) + 𝐷 ∣ 𝑓 ∈ 𝐿(𝐷)}.

Thus |𝐷| is the complete linear system containing 𝐷.

A projective embedding of an elliptic curve can be constructed as follows: let 𝐷 = 𝑃0 ,
where 𝑃0 is the zero element of 𝐴, and choose a suitable basis 1, 𝑥, 𝑦 of 𝐿(3𝐷); then the
map 𝐴 → ℙ2 defined by {1, 𝑥, 𝑦} identifies 𝐴 with the cubic projective curve

𝑌 2 𝑍 + 𝑎1 𝑋𝑌𝑍 + 𝑎3 𝑌𝑍 2 = 𝑋 3 + 𝑎2 𝑋 2 𝑍 + 𝑎4 𝑋𝑍 2 + 𝑎6 𝑍 3

(see Hartshorne 1977, IV, 4.6). This argument can be extended to every abelian variety.
Under construction.
 m. Grassmann varieties 149

m. Grassmann varieties

example, the lines in ℙ𝑛 , form a projective variety.

We show that the linear subvarieties of fixed dimension in a given projective space, for

Let 𝐸 be a vector space over 𝑘 of dimension 𝑛, and let 𝐺𝑑 (𝐸) be the set of 𝑑-dimensional
subspaces of 𝐸. When 𝑑 = 0 or 𝑛, 𝐺𝑑 (𝐸) has a single element, and so from now on we
assume that 0 < 𝑑 < 𝑛. Fix a basis for 𝐸, and let 𝑆 ∈ 𝐺𝑑 (𝐸). The choice of a basis
for 𝑆 then determines a 𝑑 × 𝑛 matrix 𝐴(𝑆) whose rows are the coordinates of the basis
elements. Changing the basis for 𝑆 multiplies 𝐴(𝑆) on the left by an invertible 𝑑 × 𝑑
matrix. Thus, the family of 𝑑 × 𝑑 minors of 𝐴(𝑆) is determined
( )
𝑛 −1
nonzero constant, and so defines a point 𝑃(𝑆) in ℙ
up to multiplication by a
𝑑
( )
.
𝑛 −1
( )The map 𝑆 ↦ 𝑃(𝑆) ∶ 𝐺𝑑 (𝐸) → ℙ 𝑑
𝑛 −1
closed subset of ℙ
Proposition 6.29. is injective, with image a
𝑑 .

( )
below. The maps 𝑃 defined by different bases of 𝐸 differ by an
𝑛 −1
automorphism of ℙ 𝑑 , and so the statement is independent of the choice of the basis
We give the proof

𝐺𝑑 (𝐸) as a projective algebraic variety called the Grassmann variety of 𝑑-dimensional

— later (6.34) we shall give a “coordinate-free description” of the map. The map realizes

subspaces of 𝐸.
Example 6.30. The affine cone over a line in ℙ3 is a two-dimensional subspace of 𝑘 4 .
Thus, 𝐺2 (𝑘 4 ) can be identified with the set of lines in ℙ3 . Let 𝐿 be a line in ℙ3 , and let
𝐱 = (𝑥0 ∶ 𝑥1 ∶ 𝑥2 ∶ 𝑥3 ) and 𝐲 = (𝑦0 ∶ 𝑦1 ∶ 𝑦2 ∶ 𝑦3 ) be distinct points on 𝐿. Then
| |
def || 𝑥 𝑥𝑗 |||
𝑃(𝐿) = (𝑝01 ∶ 𝑝02 ∶ 𝑝03 ∶ 𝑝12 ∶ 𝑝13 ∶ 𝑝23 ) ∈ ℙ5 , 𝑝𝑖𝑗 = ||| 𝑖 |,
|| 𝑦𝑖 𝑦𝑗 |||
depends only on 𝐿. The map 𝐿 ↦ 𝑃(𝐿) is a bijection from 𝐺2 (𝑘 4 ) onto the quadric
𝛱 ∶ 𝑋01 𝑋23 − 𝑋02 𝑋13 + 𝑋03 𝑋12 = 0
in ℙ5 . For a direct elementary proof of this, see (9.41, 9.42) below.
Remark 6.31. Let 𝑆 ′ be a subspace of 𝐸 of complementary dimension 𝑛 − 𝑑, and let
𝐺𝑑 (𝐸)𝑆′ be the set of 𝑆 ∈ 𝐺𝑑 (𝑉) such that 𝑆 ∩ 𝑆 ′ = {0}. Fix an 𝑆0 ∈ 𝐺𝑑 (𝐸)𝑆′ , so that
𝐸 = 𝑆0 ⊕ 𝑆 ′ . For any 𝑆 ∈ 𝐺𝑑 (𝑉)𝑆′ , the projection 𝑆 → 𝑆0 given by this decomposition is
an isomorphism, and so 𝑆 is the graph of a homomorphism 𝑆0 → 𝑆 ′ :
𝑠 ↦ 𝑠′ ⇐⇒ (𝑠, 𝑠′ ) ∈ 𝑆.
Conversely, the graph of any homomorphism 𝑆0 → 𝑆 ′ lies in 𝐺𝑑 (𝑉)𝑆′ . Thus,
𝐺𝑑 (𝑉)𝑆′ ≈ Hom(𝑆0 , 𝑆 ′ ) ≈ Hom(𝐸∕𝑆 ′ , 𝑆 ′ ).
The isomorphism 𝐺𝑑 (𝑉)𝑆′ ≈ Hom(𝐸∕𝑆 ′ , 𝑆 ′ ) depends on the choice of 𝑆0 — it is the
(27)

element of 𝐺𝑑 (𝑉)𝑆′ corresponding to 0 ∈ Hom(𝐸∕𝑆 ′ , 𝑆 ′ ). The decomposition 𝐸 = 𝑆0 ⊕𝑆 ′

End(𝑆0 ) Hom(𝑆 ′ , 𝑆0 )
gives a decomposition

End(𝐸) = ( ),
Hom(𝑆0 , 𝑆 ′ ) End(𝑆 ′ )
( )
1 0
and the bijections (27) show that the group Hom(𝑆 0 ,𝑆 ) 1
′

𝐺𝑑 (𝐸)𝑆′ .
acts simply transitively on
150 6. Projective Varieties

Remark 6.32. The bijection (27) identifies 𝐺𝑑 (𝐸)𝑆′ with the affine variety 𝔸(Hom(𝑆0 , 𝑆 ′ ))
defined by the vector space Hom(𝑆0 , 𝑆 ′ ) (cf. p. 73). Therefore, the tangent space to 𝐺𝑑 (𝐸)
at 𝑆0 ,
𝑇𝑆0 (𝐺𝑑 (𝐸)) ≃ Hom(𝑆0 , 𝑆 ′ ) ≃ Hom(𝑆0 , 𝐸∕𝑆0 ).
Since the dimension of this space does not depend on the choice of 𝑆0 , this shows that
(28)

𝐺𝑑 (𝐸) is nonsingular (4.39).

Remark 6.33. Let 𝐵 be the set of all bases of 𝐸. The choice of a basis for 𝐸 identifies
𝐵 with GL𝑛 , which is the principal open subset of 𝔸𝑛 where det ≠ 0. In particular,
2

𝐵 has a natural structure as an irreducible algebraic variety. The map (𝑒1 , … , 𝑒𝑛 ) ↦

⟨𝑒1 , … , 𝑒𝑑 ⟩ ∶ 𝐵 → 𝐺𝑑 (𝐸) is a surjective regular map, and so 𝐺𝑑 (𝐸) is also irreducible.
⋀ ⨁ ⋀𝑑
𝐸= 𝐸 of 𝐸 is the quotient of the tensor
⋀𝑑
𝑑≥0
algebra by the ideal generated by all vectors 𝑒 ⊗ 𝑒, 𝑒 ∈ 𝐸. The elements of 𝐸 are called
Remark 6.34. The exterior algebra

(exterior) 𝑑-vectors.The exterior algebra of 𝐸 is a finite-dimensional graded algebra

⋀0 ⋀1 ( )
over 𝑘 with 𝐸 = 𝑘, 𝐸 = 𝐸; if 𝑒1 , … , 𝑒𝑛 form an ordered basis for 𝑉, then the 𝑛𝑑

𝑒𝑖1 ∧ … ∧ 𝑒𝑖𝑑 (𝑖1 < ⋯ < 𝑖𝑑 )

wedge products

⋀𝑑 ⋀𝑛
𝐸. In particular, 𝐸 has dimension 1. For a subspace 𝑆 of
⋀𝑑 ⋀𝑑
𝐸 of dimension 𝑑, 𝑆 is the one-dimensional subspace of 𝐸 spanned by 𝑒1 ∧ … ∧ 𝑒𝑑
form an ordered basis for

for any basis 𝑒1 , … , 𝑒𝑑 of 𝑆. Thus, there is a well-defined map

⋀𝑑 ⋀𝑑
𝑆↦ 𝑆 ∶ 𝐺𝑑 (𝐸) → ℙ( 𝐸) (29)

which the choice of a basis for 𝐸 identifies with 𝑆 ↦ 𝑃(𝑆). Note that the subspace
spanned by 𝑒1 , … , 𝑒𝑛 can be recovered from the line through 𝑒1 ∧ … ∧ 𝑒𝑑 as the space of
vectors 𝑣 such that 𝑣 ∧ 𝑒1 ∧ … ∧ 𝑒𝑑 = 0 (cf. 6.35 below).

Fix a basis 𝑒1 , …(, 𝑒)𝑛 of 𝐸, and let 𝑆0 = ⟨𝑒1 , … , 𝑒𝑑 ⟩ and 𝑆 ′ = ⟨𝑒𝑑+1 , … , 𝑒𝑛 ⟩. Order the
First proof of Proposition 6.29.

𝑛 −1
coordinates in ℙ 𝑑 so that

𝑃(𝑆) = (𝑎0 ∶ … ∶ 𝑎𝑖𝑗 ∶ … ∶ …),

where 𝑎0 is the left-most 𝑑 × 𝑑 minor of 𝐴(𝑆), and 𝑎𝑖𝑗 , 1 ≤ 𝑖 ≤ 𝑑, 𝑑 < 𝑗 ≤ 𝑛, is the minor
obtained from the left-most 𝑑 × 𝑑 minor by replacing the
( )
𝑖th column with the 𝑗th column.
𝑛 −1
Let 𝑈0 be the (“typical”) standard open subset of ℙ 𝑑
nonzero zeroth coordinate. Clearly,6 𝑃(𝑆) ∈ 𝑈0 if and only if 𝑆 ∈ 𝐺𝑑 (𝐸)𝑆′ . We shall
consisting of the points with

prove the proposition by showing that 𝑃 ∶ 𝐺𝑑 (𝐸)𝑆′ → 𝑈0 is injective with closed image.
For 𝑆 ∈ 𝐺𝑑 (𝐸)𝑆′ , the projection 𝑆 → 𝑆0 is bijective. For each 𝑖, 1 ≤ 𝑖 ≤ 𝑑, let
∑
𝑒𝑖′ = 𝑒𝑖 + 𝑑<𝑗≤𝑛
𝑎𝑖𝑗 𝑒𝑗 (30)
If 𝑒 ∈ 𝑆 ′ ∩ 𝑆 is nonzero, we may choose it to be part of the basis for 𝑆, and then the left-most 𝑑 × 𝑑
submatrix of 𝐴(𝑆) has a row of zeros. Conversely, if the left-most 𝑑 × 𝑑 submatrix is singular, we can change
6

the basis for 𝑆 so that it has a row of zeros; then the basis element corresponding to the zero row lies in
𝑆 ′ ∩ 𝑆.
 m. Grassmann varieties 151

denote the unique element of 𝑆 projecting to 𝑒𝑖 . Then 𝑒1′ , … , 𝑒𝑑′ is a basis for 𝑆. Conversely,
for any (𝑎𝑖𝑗 ) ∈ 𝑘 𝑑(𝑛−𝑑) , the 𝑒𝑖′ defined by (30) span an 𝑆 ∈ 𝐺𝑑 (𝐸)𝑆′ and project to the
𝑒𝑖 . Therefore, 𝑆 ↔ (𝑎𝑖𝑗 ) gives a one-to-one correspondence 𝐺𝑑 (𝐸)𝑆′ ↔ 𝑘𝑑(𝑛−𝑑) (this is a

Now, if 𝑆 ↔ (𝑎𝑖𝑗 ), then

restatement of (27) in terms of matrices).

𝑃(𝑆) = (1 ∶ … ∶ 𝑎𝑖𝑗 ∶ … ∶ … ∶ 𝑓𝑘 (𝑎𝑖𝑗 ) ∶ …).

where 𝑓𝑘 (𝑎𝑖𝑗 ) is a polynomial in the 𝑎𝑖𝑗 whose coefficients are independent of 𝑆. Thus,
𝑃(𝑆) determines (𝑎𝑖𝑗 ) and hence also 𝑆. Moreover, the image of 𝑃 ∶ 𝐺𝑑 (𝐸)𝑆′ → 𝑈0 is the

( )
graph of the regular map
𝑛 −𝑑(𝑛−𝑑)−1
(… , 𝑎𝑖𝑗 , …) ↦ (… , 𝑓𝑘 (𝑎𝑖𝑗 ), …) ∶ 𝔸𝑑(𝑛−𝑑) → 𝔸 𝑑 ,
which is closed (5.28).

An exterior 𝑑-vector 𝑣 is said to be pure (or decomposable) if there exist vectors

Second proof of Proposition 6.29.

𝑒1 , … , 𝑒𝑑 ∈ 𝑉 such that 𝑣 = 𝑒1 ∧ … ∧ 𝑒𝑑 . According to 6.34, the image of 𝐺𝑑 (𝐸) in

⋀𝑑
ℙ( 𝐸) consists of the lines through the pure 𝑑-vectors.
Lemma 6.35. Let 𝑤 be a nonzero 𝑑-vector and let
𝑀(𝑤) = {𝑣 ∈ 𝐸 ∣ 𝑣 ∧ 𝑤 = 0};
then dim𝑘 𝑀(𝑤) ≤ 𝑑, with equality if and only if 𝑤 is pure.

Proof. Let 𝑒1 , … , 𝑒𝑚 be a basis of 𝑀(𝑤), and extend it to a basis 𝑒1 , … , 𝑒𝑚 , … , 𝑒𝑛 of 𝑉.

∑
𝑤= 𝑎𝑖1 …𝑖𝑑 𝑒𝑖1 ∧ … ∧ 𝑒𝑖𝑑 , 𝑎𝑖1 …𝑖𝑑 ∈ 𝑘.
Write

1≤𝑖1 <…<𝑖𝑑

If there is a nonzero term in this sum in which 𝑒𝑗 does not occur, then 𝑒𝑗 ∧ 𝑤 ≠ 0.
Therefore, each nonzero term in the sum is of the form 𝑎𝑒1 ∧ … ∧ 𝑒𝑚 ∧ …. It follows that
𝑚 ≤ 𝑑, and 𝑚 = 𝑑 if and only if 𝑤 = 𝑎𝑒1 ∧ … ∧ 𝑒𝑑 with 𝑎 ≠ 0.

For a nonzero 𝑑-vector 𝑤, let [𝑤] denote the line through 𝑤. The lemma shows that
2

⋀𝑑+1
[𝑤] ∈ 𝐺𝑑 (𝐸) if and only if the linear map 𝑣 ↦ 𝑣 ∧ 𝑤 ∶ 𝐸 ↦ 𝐸 has rank ≤ 𝑛 − 𝑑
(in which case the rank is 𝑛 − 𝑑). Thus 𝐺𝑑 (𝐸) is defined by the vanishing of the minors
of order 𝑛 − 𝑑 + 1 of this map.

⋀𝑑 ⋀𝑑+1
In more detail, the map

𝑤 ↦ (𝑣 ↦ 𝑣 ∧ 𝑤) ∶ 𝐸 → Hom𝑘 (𝐸, 𝐸)

⋀𝑑 ⋀𝑑+1
is injective and linear, and so defines an injective regular map

ℙ( 𝐸) → ℙ(Hom𝑘 (𝐸, 𝐸)).

⋀𝑑+1
The condition rank ≤ 𝑛 − 𝑑 defines a closed subset 𝑊 of ℙ(Hom𝑘 (𝐸, 𝐸)) (once a
basis has been chosen for 𝐸, the condition becomes the vanishing of the minors of order
⋀𝑑+1
𝑛 − 𝑑 + 1 of a linear map 𝐸 → 𝐸), and
⋀𝑑
𝐺𝑑 (𝐸) = ℙ( 𝐸) ∩ 𝑊.
152 6. Projective Varieties

Flag varieties
The discussion in the last subsection extends easily to chains of subspaces. Let 𝐝 =
(𝑑1 , … , 𝑑𝑟 ) be a sequence of integers with 0 < 𝑑1 < ⋯ < 𝑑𝑟 < 𝑛, and let 𝐺𝐝 (𝐸) be the set

𝐹 ∶ 𝐸 ⊃ 𝐸1 ⊃ ⋯ ⊃ 𝐸𝑟 ⊃ 0
of flags

with 𝐸 𝑖 a subspace of 𝐸 of dimension 𝑑𝑖 . The map

(31)

𝐹↦(𝐸 𝑖 ) ∏ ∏ ⋀𝑑
𝐺𝐝 (𝐸) ,,,,,,→ 𝑖 𝐺𝑑𝑖 (𝐸) ⊂ 𝑖 ℙ( 𝑖 𝐸)
∏
realizes 𝐺𝐝 (𝐸) as a closed subset7 𝐺 (𝐸), and so it is a projective variety, called
𝑖 𝑑𝑖
a flag variety. The tangent space to 𝐺𝐝 (𝐸) at the flag 𝐹 consists of the families of

𝜑𝑖 ∶ 𝐸 𝑖 → 𝐸∕𝐸 𝑖 , 1 ≤ 𝑖 ≤ 𝑟,
homomorphisms
(32)
that are compatible in the sense that

𝜑𝑖 |𝐸 𝑖+1 ≡ 𝜑𝑖+1 mod 𝐸 𝑖+1 .

Aside 6.36. A basis 𝑒1 , … , 𝑒𝑛 for 𝐸 is adapted to the flag 𝐹 if it contains a basis 𝑒1 , … , 𝑒𝑗𝑖 for each
𝐸 𝑖 . Clearly, every flag admits such a basis, and the basis then determines the flag. As in (6.33),
this implies that 𝐺𝐝 (𝐸) is irreducible. Because GL(𝐸) acts transitively on the set of bases for 𝐸, it
acts transitively on 𝐺𝐝 (𝐸). For a flag 𝐹, the subgroup 𝑃(𝐹) stabilizing 𝐹 is an algebraic subgroup
of GL(𝐸), and the map
𝑔 ↦ 𝑔𝐹0 ∶ GL(𝐸)∕𝑃(𝐹0 ) → 𝐺𝐝 (𝐸)
is an isomorphism of algebraic varieties. Because 𝐺𝐝 (𝐸) is projective, this shows that 𝑃(𝐹0 ) is a
parabolic subgroup of GL(𝐸).

n. Bézout’s theorem
Let 𝑉 be a hypersurface in ℙ𝑛 (that is, a closed subvariety of dimension 𝑛 − 1). For such
a variety, 𝐼(𝑉) = (𝐹(𝑋0 , … , 𝑋𝑛 )) with 𝐹 a homogenous polynomial without repeated
factors. We define the degree of 𝑉 to be the degree of 𝐹.
The next theorem is one of the oldest, and most famous, in algebraic geometry.8

Theorem 6.37. Let 𝐶 and 𝐷 be curves in ℙ2 of degrees 𝑚 and 𝑛 respectively. If 𝐶 and

𝐷 have no irreducible component in common, then they intersect in exactly 𝑚𝑛 points,
counted with appropriate multiplicities.

Proof. Decompose 𝐶 and 𝐷 into their irreducible components. Clearly it suffices to

prove the theorem for each irreducible component of 𝐶 and each irreducible component
of 𝐷. We can therefore assume that 𝐶 and 𝐷 are themselves irreducible.
We know from 2.62 that 𝐶 ∩ 𝐷 is of dimension zero, and so is finite. After a change
of variables, we can assume that 𝑎 ≠ 0 for all points (𝑎 ∶ 𝑏 ∶ 𝑐) ∈ 𝐶 ∩ 𝐷.
For example, if 𝑢𝑖 is a pure 𝑑𝑖 -vector and 𝑢𝑖+1 is a pure 𝑑𝑖+1 -vector, then it follows from (6.35) that
𝑀(𝑢𝑖 ) ⊂ 𝑀(𝑢𝑖+1 ) if and only if the map
7

⋀𝑑𝑖 +1 ⋀𝑑𝑖+1 +1
𝑣 ↦ (𝑣 ∧ 𝑢𝑖 , 𝑣 ∧ 𝑢𝑖+1 ) ∶ 𝐸 → 𝐸⊕ 𝐸

has rank ≤ 𝑛 − 𝑑𝑖 (in which case it has rank 𝑛 − 𝑑𝑖 ). Thus, 𝐺𝐝 (𝐸) is defined by the vanishing of many
minors.
8
Bézout 1779, but announced earlier by MacLaurin 1720.
 o. Hilbert polynomials 153

Let 𝐹(𝑋, 𝑌, 𝑍) and 𝐺(𝑋, 𝑌, 𝑍) be the polynomials defining 𝐶 and 𝐷, and write
𝐹 = 𝑠0 𝑍 𝑚 + 𝑠1 𝑍 𝑚−1 + ⋯ + 𝑠𝑚 , 𝐺 = 𝑡0 𝑍 𝑛 + 𝑡1 𝑍 𝑛−1 + ⋯ + 𝑡𝑛
with 𝑠𝑖 and 𝑡𝑗 polynomials in 𝑋 and 𝑌 of degrees 𝑖 and 𝑗 respectively. Clearly 𝑠𝑚 ≠ 0 ≠ 𝑡𝑛 ,
for otherwise 𝐹 and 𝐺 would have 𝑍 as a common factor. Let 𝑅 be the resultant (7.27
below; Wikipedia: Resultant) of 𝐹 and 𝐺, regarded as polynomials in 𝑍. It is either a
homogeneous polynomial of degree 𝑚𝑛 in 𝑋 and 𝑌 or it is identically zero. If the latter
occurs, then for every (𝑎, 𝑏) ∈ 𝑘 2 , 𝐹(𝑎, 𝑏, 𝑍) and 𝐺(𝑎, 𝑏, 𝑍) have a common zero, which
contradicts the finiteness of 𝐶 ∩ 𝐷. Thus 𝑅 is a nonzero polynomial of degree 𝑚𝑛. Write
𝑅(𝑋, 𝑌) = 𝑋 𝑚𝑛 𝑅∗ ( 𝑋𝑌 ), where 𝑅 (𝑇) is a polynomial of degree ≤ 𝑚𝑛 in 𝑇 = 𝑌 .
∗ 𝑋
Suppose first that deg 𝑅∗ = 𝑚𝑛, and let 𝛼1 , … , 𝛼𝑚𝑛 be the roots of 𝑅∗ (some of them
𝑏
may be multiple). Each such root can be written 𝛼𝑖 = 𝑎𝑖 , and 𝑅(𝑎𝑖 , 𝑏𝑖 ) = 0. According to
𝑖
7.28 this means that the polynomials 𝐹(𝑎𝑖 , 𝑏𝑖 , 𝑍) and 𝐺(𝑎𝑖 , 𝑏𝑖 , 𝑍) have a common root 𝑐𝑖 .
Thus (𝑎𝑖 ∶ 𝑏𝑖 ∶ 𝑐𝑖 ) is a point on 𝐶 ∩ 𝐷, and conversely, if (𝑎 ∶ 𝑏 ∶ 𝑐) is a point on 𝐶 ∩ 𝐷
(so 𝑎 ≠ 0), then 𝑎𝑏 is a root of 𝑅 (𝑇). Thus we see in this case, that 𝐶 ∩ 𝐷 has precisely
∗

𝑚𝑛 points, provided we take the multiplicity of (𝑎 ∶ 𝑏 ∶ 𝑐) to be the multiplicity of 𝑎 𝑏 as

a root of 𝑅∗ .
Now suppose that 𝑅∗ has degree 𝑟 < 𝑚𝑛. Then 𝑅(𝑋, 𝑌) = 𝑋 𝑚𝑛−𝑟 𝑃(𝑋, 𝑌), where
𝑃(𝑋, 𝑌) is a homogeneous polynomial of degree 𝑟 not divisible by 𝑋. Obviously 𝑅(0, 1) =
0, and so there is a point (0 ∶ 1 ∶ 𝑐) in 𝐶 ∩ 𝐷, in contradiction with our assumption. 2

Remark 6.38. The above proof has the defect that the notion of multiplicity has been

the theorem holds with the following more natural definition of multiplicity. Let 𝑃 be
too obviously chosen to make the theorem come out right. It is possible to show that

an isolated point of 𝐶 ∩ 𝐷. There will be an affine neighbourhood 𝑈 of 𝑃 and regular

functions 𝑓 and 𝑔 on 𝑈 such that 𝐶 ∩ 𝑈 = 𝑉(𝑓) and 𝐷 ∩ 𝑈 = 𝑉(𝑔). We can regard 𝑓
and 𝑔 as elements of the local ring 𝒪𝑃 , and clearly rad(𝑓, 𝑔) = 𝔪, the maximal ideal in
𝒪𝑃 . It follows that 𝒪𝑃 ∕(𝑓, 𝑔) is finite-dimensional over 𝑘, and we define the multiplicity
of 𝑃 in 𝐶 ∩ 𝐷 to be dim𝑘 (𝒪𝑃 ∕(𝑓, 𝑔)). For example, if 𝐶 and 𝐷 cross transversely at 𝑃,
then 𝑓 and 𝑔 will form a system of local parameters at 𝑃 — (𝑓, 𝑔) = 𝔪 — and so the
multiplicity is one.
The attempt to find good notions of multiplicities in very general situations motivated
much of the most interesting work in commutative algebra in the second half of the
twentieth century.

o. Hilbert polynomials
Recall that for a projective variety 𝑉 ⊂ ℙ𝑛 , the homogeneous coordinate ring
𝑘hom [𝑉] = 𝑘[𝑋0 , … , 𝑋𝑛 ]∕𝔟 = 𝑘[𝑥0 , … , 𝑥𝑛 ],
where 𝔟 = 𝐼(𝑉). The ideal 𝔟 is graded, and so 𝑘hom [𝑉] is a graded ring,
⨁
𝑘hom [𝑉] = 𝑘hom [𝑉]𝑚 ,
𝑚≥0

where 𝑘hom [𝑉]𝑚 is the 𝑘-subspace spanned by the monomials in the 𝑥𝑖 of degree 𝑚. In
particular, 𝑘hom [𝑉]𝑚 is a finite-dimensional 𝑘-vector space, and we define the Hilbert
function of 𝑉 to be
𝐻(𝑉, 𝑚) = dim𝑘 𝑘hom [𝑉]𝑚 .
def
154 6. Projective Varieties

It is a function ℕ → ℕ. Note that

𝑘hom [𝑉]𝑚 = 𝑘[𝑋0 , … , 𝑋𝑛 ]𝑚 ∕𝔟𝑚 ,

so 𝐻(𝑉, 𝑚) is the codimension, in the space of homogeneous polynomials of degree 𝑚

in the 𝑋𝑖 , of the subspace of those that vanish on 𝑉.

Example 6.39. By definition 𝑘hom [ℙ𝑛 ] = 𝑘[𝑋0 , … , 𝑋𝑛 ], so 𝑘hom [ℙ𝑛 ]𝑚 consists of the
homogeneous polynomials of degree 𝑚 in 𝑋0 , … , 𝑋𝑛 . There are ( 𝑚+𝑛 𝑛 ) monomials of
degree 𝑚 in 𝑛 + 1 variables, so
(𝑚 + 𝑛 ) (𝑚 + 𝑛) ⋯ (𝑚 + 1)
𝐻(ℙ𝑛 , 𝑚) = = .
𝑛 𝑛!

) are ( 𝑛 ) monomials can be proved by induction on 𝑚 + 𝑛. If 𝑚 = 0 = 𝑛,

That (there 𝑚+𝑛

then 0 = 1, which is correct. A general homogeneous polynomial of degree 𝑚 can be

0

written uniquely as

𝐹(𝑋0 , 𝑋1 , … , 𝑋𝑛 ) = 𝐹1 (𝑋1 , … , 𝑋𝑛 ) + 𝑋0 𝐹2 (𝑋0 , 𝑋1 , … , 𝑋𝑛 )

with 𝐹1 homogeneous of degree 𝑚 and 𝐹2 homogeneous of degree 𝑚 − 1. But

(𝑚 + 𝑛) (𝑚 + 𝑛 − 1) (𝑚 + 𝑛 − 1)
= +
𝑚 𝑚 𝑚−1
because they are the coefficients of 𝑋 𝑚 in

(𝑋 + 1)𝑚+𝑛 = (𝑋 + 1)(𝑋 + 1)𝑚+𝑛−1 ,

and this proves the induction.

Example 6.40. Let 𝑉 = {𝑃1 , 𝑃2 , 𝑃3 } be a set of three points in ℙ2 . There exists a nonzero
linear polynomial vanishing at all the points if and only if they are collinear. Thus

1 if the points are collinear

𝐻(𝑉, 1) = {
2 otherwise.

For 𝑚 ≥ 2, the map

𝑓 ↦ (𝑓(𝑃0 ), 𝑓(𝑃1 ), 𝑓(𝑃2 )) ∶ 𝑘[𝑋0 , 𝑋1 , 𝑋2 ]𝑚 → 𝑘 3

is surjective, and so its kernel has codimension 3. Thus

𝐻(𝑉, 𝑚) = 3 for 𝑚 ≥ 2.

Similarly, if 𝑉 is a set of 𝛿 points ℙ2 , then 𝐻(𝑉, 1) depends on the dimension of the

subspace spanned by the points, but

𝐻(𝑉, 𝑚) = 𝛿 for 𝑚 ≥ 𝛿 − 1.

The degree of a projective variety is the number of points in the intersection of the

𝑉 is the hypersurface in ℙ𝑛 defined by a homogeneous polynomial of degree 𝛿 and

variety and of a general linear variety of complementary dimension. For example, if

𝐻1 , … , 𝐻𝑛−1 are hyperplanes in ℙ𝑛 , then “in general”,

|𝑉 ∩ 𝐻1 ∩ … ∩ 𝐻𝑛−1 | = 𝛿.
 o. Hilbert polynomials 155

Theorem 6.41. Let 𝑉 ⊂ ℙ𝑛 be a projective variety. There exists a unique polynomial

𝑃(𝑉, 𝑇) ∈ ℚ[𝑇] such that
𝑃(𝑉, 𝑚) = 𝐻(𝑉, 𝑚)
for all sufficiently large 𝑚. Moreover,

𝛿 𝑑
𝑃(𝑉, 𝑇) = 𝑇 + terms of lower degree,
𝑑!
where 𝑑 is the dimension of 𝑉 and 𝛿 its degree.

Proof. Omitted (for the present). 2

The polynomial 𝑃(𝑉, 𝑇) in the theorem is called the Hilbert polynomial of 𝑉.

Despite the notation, it depends not just on 𝑉 but also on its embedding in projective

(𝑇 + 𝑛) (𝑇 + 𝑛) ⋯ (𝑇 + 1)
space.

𝑃(ℙ𝑛 , 𝑇) = = ,
For example,

𝑛 𝑛!
and if 𝑉 is a set of 𝛿 points in ℙ2 , then

𝑃(𝑉, 𝑇) = 𝛿.

Example 6.42. Let 𝑉 be the image of the Veronese map

(𝑎0 ∶ 𝑎1 ) ↦ (𝑎0𝛿 ∶ 𝑎0𝛿−1 𝑎1 ∶ … ∶ 𝑎1𝛿 ) ∶ ℙ1 → ℙ𝛿 , 𝛿 ∈ ℕ.

Then 𝑘hom [𝑉]𝑚 can be identified with the set of homogeneous polynomials of degree
𝑚 ⋅ 𝛿 in two variables (look at the map 𝔸2 → 𝔸𝛿+1 given by the same equations), which
is a space of dimension 𝛿𝑚 + 1, and so

𝑃(𝑉, 𝑇) = 𝛿𝑇 + 1.

Thus 𝑉 has dimension 1 (which we knew) and degree 𝛿.

Example 6.43. Let 𝑉 be the curve in ℙ2 defined by a homogeneous polynomial 𝐹 of

degree 𝛿. If 𝔟 is the ideal in 𝑘[𝑋0 , 𝑋1 , 𝑋2 ] corresponding to 𝔟, then 𝔟𝑚 consists of the
polynomials of degree 𝑚 divisible by 𝐹, and so
(𝑚 − 𝛿 + 2)
dim 𝔟𝑚 =
2
if 𝑚 ≥ 𝛿. Therefore, for 𝑚 ≥ 𝛿,
(𝑚 + 2 ) (𝑚 − 𝛿 + 2 ) 𝛿(𝛿 − 2)
𝐻(𝑉, 𝑚) = − = 𝛿𝑚 − .
2 2 2

𝛿(𝛿 − 2)
𝑃(𝑉, 𝑇) = 𝛿𝑇 − .
Hence

2
Macaulay2 knows how to compute Hilbert polynomials.
156 6. Projective Varieties

p. Dimensions

cation: if 𝑉 and 𝑊 are closed subvarieties of dimensions 𝑟 and 𝑠 in ℙ𝑛 and 𝑟 + 𝑠 ≥ 𝑛,

The results for affine varieties extend to projective varieties with one important simplifi-

then 𝑉 ∩ 𝑊 ≠ ∅. For example, any two closed curves in ℙ2 intersect.

Theorem 6.43. Let 𝑉 = 𝑉(𝔞) ⊂ ℙ𝑛 be a projective variety of dimension ≥ 1, and let

𝑓 ∈ 𝑘[𝑋0 , … , 𝑋𝑛 ] be homogeneous, nonconstant, and ∉ 𝔞; then 𝑉 ∩ 𝑉(𝑓) is nonempty and
of pure codimension 1.

Proof. Since the dimension of a variety is equal to the dimension of any dense open

𝑉 ∩𝑉(𝑓) is nonempty. Let 𝑉 af f (𝔞) be the zero set of 𝔞 in 𝔸𝑛+1 (that is, the affine cone over
affine subset, the only part that does not follow immediately from 3.42 is the fact that

𝑉). Then 𝑉 af f (𝔞) ∩ 𝑉 af f (𝑓) is nonempty (it contains (0, … , 0)), and so it has codimension
1 in 𝑉 af f (𝔞). Clearly 𝑉 af f (𝔞) has dimension ≥ 2, and so 𝑉 af f (𝔞) ∩ 𝑉 af f (𝑓) has dimension
≥ 1. This implies that the polynomials in 𝔞 have a zero in common with 𝑓 other than
the origin, and so 𝑉(𝔞) ∩ 𝑉(𝑓) ≠ ∅. 2

Corollary 6.44. Let 𝑓1 , … , 𝑓𝑟 be homogeneous nonconstant elements of 𝑘[𝑋0 , … , 𝑋𝑛 ];

and let 𝑍 be an irreducible component of 𝑉 ∩ 𝑉(𝑓1 , … 𝑓𝑟 ). Then codim(𝑍) ≤ 𝑟, and if
dim(𝑉) ≥ 𝑟, then 𝑉 ∩ 𝑉(𝑓1 , … 𝑓𝑟 ) is nonempty.

Proof. Induction on 𝑟, as before. 2

Proposition 6.45. Let 𝑍 be an irreducible closed subvariety of 𝑉; if codim(𝑍) = 𝑟, then

there exist homogeneous polynomials 𝑓1 , … , 𝑓𝑟 in 𝑘[𝑋0 , … , 𝑋𝑛 ] such that 𝑍 is an irreducible
component of 𝑉 ∩ 𝑉(𝑓1 , … , 𝑓𝑟 ).

Proof. Use the same argument as in the proof 3.47. 2

Proposition 6.46. Every pure closed subvariety 𝑍 of ℙ𝑛 of codimension one is principal,

i.e., 𝐼(𝑍) = (𝑓) for some 𝑓 homogeneous element of 𝑘[𝑋0 , … , 𝑋𝑛 ].

Proof. Follows from the affine case. 2

Corollary 6.47. Let 𝑉 and 𝑊 be closed subvarieties of ℙ𝑛 ; if dim(𝑉)+dim(𝑊) ≥ 𝑛, then

𝑉 ∩ 𝑊 ≠ ∅, and every irreducible component of it has codim(𝑍) ≤codim(𝑉)+codim(𝑊).

Proof. Write 𝑉 = 𝑉(𝔞) and 𝑊 = 𝑉(𝔟), and consider the affine cones 𝑉 ′ = 𝑉(𝔞) and
𝑊 ′ = 𝑉(𝔟) over them. Then

dim(𝑉 ′ ) + dim(𝑊 ′ ) = dim(𝑉) + 1 + dim(𝑊) + 1 ≥ 𝑛 + 2.

As 𝑉 ′ ∩ 𝑊 ′ ≠ ∅, 𝑉 ′ ∩ 𝑊 ′ has dimension ≥ 1, and so it contains a point other than the

origin. Therefore 𝑉 ∩ 𝑊 ≠ ∅. The rest of the statement follows from the affine case. 2

Proposition 6.48. Let 𝑉 be a closed subvariety of ℙ𝑛 of dimension 𝑟 < 𝑛; then there is a

linear projective variety 𝐸 of dimension 𝑛 − 𝑟 − 1 (that is, 𝐸 is defined by 𝑟 + 1 independent
linear forms) such that 𝐸 ∩ 𝑉 = ∅.
 q. Products 157

Proof. Induction on 𝑟. If 𝑟 = 0, then 𝑉 is a finite set, and the lemma below shows that
there is a hyperplane in 𝑘 𝑛+1 not intersecting 𝑉.
Suppose that 𝑟 > 0, and let 𝑉1 , … , 𝑉𝑠 be the irreducible components of 𝑉. By
assumption, they all have dimension ≤ 𝑟. The intersection 𝐸𝑖 of all the linear projective
varieties containing 𝑉𝑖 is the smallest such variety. The lemma below shows that there
is a hyperplane 𝐻 containing none of the nonzero 𝐸𝑖 ; consequently, 𝐻 contains none of
the irreducible components 𝑉𝑖 of 𝑉, and so each 𝑉𝑖 ∩ 𝐻 is a pure variety of dimension
≤ 𝑟 − 1 (or is empty). By induction, there is an linear subvariety 𝐸 ′ not intersecting
𝑉 ∩ 𝐻. Take 𝐸 = 𝐸 ′ ∩ 𝐻. 2

Lemma 6.49. Let 𝑊 be a vector space of dimension 𝑑 over an infinite field 𝑘, and let
𝐸1 , … , 𝐸𝑟 be a finite set of nonzero subspaces of 𝑊. Then there is a hyperplane 𝐻 in 𝑊
containing none of the 𝐸𝑖 .

Proof. Pass to the dual space 𝑉 of 𝑊. The problem becomes that of showing 𝑉 is not a
finite union of proper subspaces 𝐸𝑖∨ . Replace each 𝐸𝑖∨ by a hyperplane 𝐻𝑖 containing
∏
it. Then 𝐻𝑖 is defined by a nonzero linear form 𝐿𝑖 . We have to show that 𝐿𝑗 is not
identically zero on 𝑉. But this follows from the statement that a polynomial in 𝑛 variables,
with coefficients not all zero, cannot be identically zero on 𝑘 𝑛 (Exercise 1-1). 2

Let 𝑉 and 𝐸 be as in Proposition 6.48. If 𝐸 is defined by the linear forms 𝐿0 , … , 𝐿𝑟

then the projection 𝑎 ↦ (𝐿0 (𝑎) ∶ ⋯ ∶ 𝐿𝑟 (𝑎)) defines a map 𝑉 → ℙ𝑟 . We shall see later
that this map is finite, and so it can be regarded as a projective version of the Noether

In general, a regular map from a variety 𝑉 to ℙ𝑛 corresponds to a line bundle on

normalization theorem.

𝑉 and a set of global sections of the line bundle. All line bundles on 𝔸𝑛 ∖ {origin} are

regular maps 𝔸𝑛+1 ∖ {origin} → ℙ𝑚 are given by a family of homogeneous polynomials.

trivial (see, for example, Hartshorne II 7.1 and II 6.2), from which it follows that all

Corollary 6.50. Let 𝛼 ∶ ℙ𝑛 → ℙ𝑚 be regular; if 𝑚 < 𝑛, then 𝛼 is constant.

Assuming this, it is possible to prove the following result.

Proof. Let 𝜋 ∶ 𝔸𝑛+1 − {origin} → ℙ𝑛 be the map (𝑎0 , … , 𝑎𝑛 ) ↦ (𝑎0 ∶ … ∶ 𝑎𝑛 ). Then

𝛼◦𝜋 is regular, and there exist polynomials 𝐹0 , … , 𝐹𝑚 ∈ 𝑘[𝑋0 , … , 𝑋𝑛 ] such that 𝛼◦𝜋 is

(𝑎0 , … , 𝑎𝑛 ) ↦ (𝐹0 (𝑎) ∶ … ∶ 𝐹𝑚 (𝑎)).

the map

As 𝛼◦𝜋 factors through ℙ𝑛 , the 𝐹𝑖 must be homogeneous of the same degree. Note that

𝛼(𝑎0 ∶ … ∶ 𝑎𝑛 ) = (𝐹0 (𝑎) ∶ … ∶ 𝐹𝑚 (𝑎)).

If 𝑚 < 𝑛 and the 𝐹𝑖 are nonconstant, then 6.43 shows they have a common zero and so
𝛼 is not defined on all of ℙ𝑛 . Hence the 𝐹𝑖 must be constant. 2

q. Products
It is useful to have an explicit description of the topology on some product varieties.
158 6. Projective Varieties

The topology on ℙ𝑚 × ℙ𝑛 .
Suppose that we have a collection of polynomials 𝐹𝑖 (𝑋0 , … , 𝑋𝑚 ; 𝑌0 , … , 𝑌𝑛 ), 𝑖 ∈ 𝐼, each
of which is separately homogeneous in the 𝑋𝑖 and 𝑌𝑗 . Then the equations

𝐹𝑖 (𝑋0 , … , 𝑋𝑚 ; 𝑌0 , … , 𝑌𝑛 ) = 0, 𝑖 ∈ 𝐼,

define a closed subset of ℙ𝑚 × ℙ𝑛 , and every closed subset of ℙ𝑚 × ℙ𝑛 arises in this way
from a (finite) set of polynomials.

The topology on 𝔸𝑚 × ℙ𝑛
The closed subsets of 𝔸𝑚 × ℙ𝑛 are exactly those defined by sets of equations

𝐹𝑖 (𝑋1 , … , 𝑋𝑚 ; 𝑌0 , … , 𝑌𝑛 ) = 0, 𝑖 ∈ 𝐼,

with each 𝐹𝑖 homogeneous in the 𝑌𝑗 .

The topology on 𝑉 × ℙ𝑛
Let 𝑉 be an irreducible affine algebraic variety. We look more closely at the topology
on 𝑉 × ℙ𝑛 in terms of ideals. Let 𝐴 = 𝑘[𝑉], and let 𝐵 = 𝐴[𝑋0 , … , 𝑋𝑛 ]. Note that
𝐵 = 𝐴 ⊗𝑘 𝑘[𝑋0 , … , 𝑋𝑛 ], and so we can view it as the ring of regular functions on
𝑉 × 𝔸𝑛+1 : for 𝑓 ∈ 𝐴 and 𝑔 ∈ 𝑘[𝑋0 , … , 𝑋𝑛 ], 𝑓 ⊗ 𝑔 is the function

(𝑣, 𝐚) ↦ 𝑓(𝑣) ⋅ 𝑔(𝐚) ∶ 𝑉 × 𝔸𝑛+1 → 𝑘.

∑
The ring 𝐵 has an obvious grading — a monomial 𝑎𝑋00 … 𝑋𝑛𝑛 , 𝑎 ∈ 𝐴, has degree 𝑖𝑗
𝑖 𝑖

— and so we have the notion of a graded ideal 𝔟 ⊂ 𝐵. It makes sense to speak of the
zero set 𝑉(𝔟) ⊂ 𝑉 × ℙ𝑛 of such an ideal. For any ideal 𝔞 ⊂ 𝐴, 𝔞𝐵 is graded, and
𝑉(𝔞𝐵) = 𝑉(𝔞) × ℙ𝑛 .

Lemma 6.51. (a) For each graded ideal 𝔟 ⊂ 𝐵, the set 𝑉(𝔟) is closed, and every closed
subset of 𝑉 × ℙ𝑛 is of this form.
(b) The set 𝑉(𝔟) is empty if and only if rad(𝔟) ⊃ (𝑋0 , … , 𝑋𝑛 ).
(c) If 𝑉 is irreducible, then 𝑉 = 𝑉(𝔟) for some graded prime ideal 𝔟.

Proof. (a) In the case that 𝐴 = 𝑘, we proved this in 6.1 and 6.2, and similar arguments
apply in the present more general situation. For example, to see that 𝑉(𝔟) is closed, cover
ℙ𝑛 with the standard open affines 𝑈𝑖 and show that 𝑉(𝔟) ∩ 𝑈𝑖 is closed for all 𝑖.
The set 𝑉(𝔟) is empty if and only if the cone 𝑉 aff (𝔟) ⊂ 𝑉 × 𝔸𝑛+1 defined by 𝔟 is
contained in 𝑉 × {origin}. But
∑
𝑎𝑖0 …𝑖𝑛 𝑋00 … 𝑋𝑛𝑛 , 𝑎𝑖0 …𝑖𝑛 ∈ 𝑘[𝑉],
𝑖 𝑖

is zero on 𝑉 × {origin} if and only if its constant term is zero, and so

𝐼 aff (𝑉 × {origin}) = (𝑋0 , 𝑋1 , … , 𝑋𝑛 ).

Thus, the Nullstellensatz shows that 𝑉(𝔟) = ∅ ⇒ rad(𝔟) = (𝑋0 , … , 𝑋𝑛 ). Conversely, if

𝑋𝑖𝑁 ∈ 𝔟 for all 𝑖, then obviously 𝑉(𝔟) is empty.
For (c), note that if 𝑉(𝔟) is irreducible, then the closure of its inverse image in
𝑉 × 𝔸𝑛+1 is also irreducible, and so 𝐼𝑉(𝔟) is prime. 2
 q. Products 159

Exercises
6-1. Show that a point 𝑃 on a projective curve 𝐹(𝑋, 𝑌, 𝑍) = 0 is singular if and only if
𝜕𝐹∕𝜕𝑋, 𝜕𝐹∕𝜕𝑌, and 𝜕𝐹∕𝜕𝑍 are all zero at 𝑃. If 𝑃 is nonsingular, show that the tangent
line at 𝑃 has the (homogeneous) equation

(𝜕𝐹∕𝜕𝑋)𝑃 𝑋 + (𝜕𝐹∕𝜕𝑌)𝑃 𝑌 + (𝜕𝐹∕𝜕𝑍)𝑃 𝑍 = 0.

Verify that 𝑌 2 𝑍 = 𝑋 3 + 𝑎𝑋𝑍 2 + 𝑏𝑍 3 is nonsingular if 𝑋 3 + 𝑎𝑋 + 𝑏 has no repeated root,

and find the tangent line at the point at infinity on the curve.

6-2. Let 𝐿 be a line in ℙ2 and let 𝐶 be a nonsingular conic in ℙ2 (i.e., a curve in ℙ2

defined by a homogeneous polynomial of degree 2). Show that either
(a) 𝐿 intersects 𝐶 in exactly 2 points, or
(b) 𝐿 intersects 𝐶 in exactly 1 point, and it is the tangent at that point.

6-3. Let 𝑉 = 𝑉(𝑌 − 𝑋 2 , 𝑍 − 𝑋 3 ) ⊂ 𝔸3 . Prove

(a) 𝐼(𝑉) = (𝑌 − 𝑋 2 , 𝑍 − 𝑋 3 ),
(b) 𝑍𝑊 −𝑋𝑌 ∈ 𝐼(𝑉)∗ ⊂ 𝑘[𝑊, 𝑋, 𝑌, 𝑍], but 𝑍𝑊 −𝑋𝑌 ∉ ((𝑌 −𝑋 2 )∗ , (𝑍 −𝑋 3 )∗ ). (Thus,
if 𝐹1 , … , 𝐹𝑟 generate 𝔞, it does not follow that 𝐹1∗ , … , 𝐹𝑟∗ generate 𝔞∗ , even if 𝔞∗ is
radical.)

6-4. Let 𝑃0 , … , 𝑃𝑟 be points in ℙ𝑛 . Show that there is a hyperplane 𝐻 in ℙ𝑛 passing

through 𝑃0 but not passing through any of 𝑃1 , … , 𝑃𝑟 .

6-5. Is the subset

{(𝑎 ∶ 𝑏 ∶ 𝑐) ∣ 𝑎 ≠ 0, 𝑏 ≠ 0} ∪ {(1 ∶ 0 ∶ 0)}

of ℙ2 locally closed?

6-6. Show that the image of the Segre map ℙ𝑚 × ℙ𝑛 → ℙ𝑚𝑛+𝑚+𝑛 (see 6.26) is not
contained in any hyperplane of ℙ𝑚𝑛+𝑚+𝑛 .

6-7. Write 0, 1, ∞ for the points (0 ∶ 1), (1 ∶ 1), and (1 ∶ 0) on ℙ1 .

(a) Let 𝛼 be an automorphism of ℙ1 such that

𝛼(0) = 0, 𝛼(1) = 1, 𝛼(∞) = ∞.

Show that 𝛼 is the identity map.

(b) Let 𝑃0 , 𝑃1 , 𝑃2 be distinct points on ℙ1 . Show that there exists an 𝛼 ∈ PGL2 (𝑘) such

𝛼(0) = 𝑃0 , 𝛼(1) = 𝑃1 , 𝛼(∞) = 𝑃2 .

that

(c) Deduce that Aut(ℙ1 ) ≃ PGL2 (𝑘).

6-8. Show that the functor

𝑅 ⇝ 𝑃𝑛 (𝑅) = {direct summands of rank 1 of 𝑅𝑛+1 }

definition of ℙ𝑛 .)
satisfies the criterion 5.71 to arise from an algebraic prevariety. (This gives an alternative
160 6. Projective Varieties

6-9. (a) Let 𝑉 ⊂ 𝔸𝑛 and 𝑊 ⊂ ℙ𝑚 be algebraic varieties and 𝜑 ∶ 𝑉 → 𝑊 a map. Show

that 𝜑 is regular if and only if every point in 𝑉 has an open neighbourhood 𝑈 on which
there are regular functions 𝑓0 , … , 𝑓𝑚 such that

𝜑(𝑎1 , … , 𝑎𝑛 ) = (𝑓0 (𝑎1 , … , 𝑎𝑛 ) ∶ … ∶ 𝑓𝑚 (𝑎1 , … , 𝑎𝑛 ))

for all (𝑎1 , … , 𝑎𝑛 ) ∈ 𝑈.

(b) Show that, for a regular map 𝜑 as in (a), it may not be possible to take 𝑈 = 𝑉.
Hint: Let 𝑉 ⊂ 𝔸4 be the complement of (0, 0, 0, 0) in

𝑋𝑌 − 𝑍𝑊 = 0,

and let 𝜑 ∶ 𝑉 → ℙ1 send (𝑤, 𝑥, 𝑦, 𝑧) to (𝑥 ∶ 𝑧) if one of 𝑥 or 𝑧 is nonzero and (𝑤, 0, 𝑦, 0)

to (𝑤 ∶ 𝑦). See sx4626969 (Mohan).
Chapter 7

Complete Varieties

Complete varieties are the analogues in the category of algebraic varieties of compact

If 𝑉 is compact, then every continuous map 𝑉 → 𝑇 with 𝑇 Hausdorff sends compact

topological spaces in the category of Hausdorff topological spaces.

Hausdorff space 𝑉 is compact if and only if the map 𝑉 → {point} is universally closed,
sets to compact sets, hence closed sets to closed sets, i.e., it is a closed map. Moreover, a

i.e., for all topological spaces 𝑇, the projection map 𝑞 ∶ 𝑉 × 𝑇 → 𝑇 is closed (Bourbaki
TG, I, 10.2, Cor. 1 to Thm 1).

a. Definition and basic properties

Definition
Definition 7.1. A prevariety 𝑉 over 𝑘 is complete if

(b) for all algebraic varieties 𝑇, the projection map 𝑞 ∶ 𝑉 × 𝑇 → 𝑇 is closed.

(a) it is separated, and

We shall see (7.22) that projective varieties are complete.

Example 7.2. The projection map

(𝑥, 𝑦) ↦ 𝑦 ∶ 𝔸1 × 𝔸1 → 𝔸1 .

is not closed. For example, the variety 𝑉 ∶ 𝑋𝑌 = 1 is closed in 𝔸2 but its image in 𝔸1
omits the origin. On the other hand, the projection map ℙ1 × 𝔸1 → 𝔸1 is closed. The
closure of 𝑉 in ℙ1 × 𝔸1 is

𝑉̄ = {((𝑥 ∶ 𝑧), 𝑦) ∈ ℙ1 × 𝔸1 ∣ 𝑥𝑦 = 𝑧2 },
def

and the point ((𝑥 ∶ 0), 0) of 𝑉̄ projects to 0.

Properties
7.3. Closed subvarieties of complete varieties are complete.

Let 𝑍 be a closed subvariety of a complete variety 𝑉. For any variety 𝑇, 𝑍 × 𝑇 is closed in

𝑉 × 𝑇, and so the restriction of the closed map 𝑞 ∶ 𝑉 × 𝑇 → 𝑇 to 𝑍 × 𝑇 is also closed.

7.4. A variety is complete if and only if its irreducible components are complete.

161
162 7. Complete Varieties

Conversely, suppose that the irreducible components 𝑉𝑖 of a variety 𝑉 are complete. If 𝑍

Each irreducible component is closed, and hence complete if the variety is complete (7.3).

is closed in 𝑉 × 𝑇, then 𝑍𝑖 = 𝑍 ∩ (𝑉𝑖 × 𝑇) is closed in 𝑉𝑖 × 𝑇. Therefore, 𝑞(𝑍𝑖 ) is closed

⋃
in 𝑇, and so 𝑞(𝑍) = 𝑞(𝑍𝑖 ) is also closed.
def

(∏ )
7.5. Products of complete varieties are complete.

Let 𝑉1 , … , 𝑉𝑛 be complete varieties, and let 𝑇 be a variety. The projection 𝑖

𝑉𝑖 ×𝑇 → 𝑇

𝑉1 × ⋯ × 𝑉𝑛 × 𝑇 → 𝑉2 × ⋯ × 𝑉𝑛 × 𝑇 → ⋯ → 𝑉𝑛 × 𝑇 → 𝑇,
is the composite of the projections

7.6. If 𝜑 ∶ 𝑉 → 𝑊 is surjective and 𝑉 is complete, then 𝑊 is complete.

all of which are closed.

Let 𝑇 be a variety, and let 𝑍 be a closed subset of 𝑊 × 𝑇. Let 𝑍 ′ be the inverse image of
𝑍 in 𝑉 × 𝑇. Then 𝑍 ′ is closed, and its image in 𝑇 equals that of 𝑍.
7.7. Let 𝜑 ∶ 𝑉 → 𝑊 be a regular map of varieties. If 𝑉 is complete, then 𝜑(𝑉) is a complete
closed subvariety of 𝑊. In particular, every complete subvariety of a variety is closed.

Let 𝛤𝜑 = {(𝑣, 𝜑(𝑣))} ⊂ 𝑉× 𝑊 be the graph of 𝜑. It is a closed subset of 𝑉× 𝑊 (because 𝑉

is a variety, see 5.28), and 𝜑(𝑉) is the projection of 𝛤𝜑 into 𝑊. Therefore 𝜑(𝑉) is closed,
def

and 7.6 shows that it is complete. The second statement follows from the first applied to

7.8. A regular map 𝑉 → ℙ1 from a complete connected variety 𝑉 is either constant or

the inclusion map.

surjective.

The only proper closed subsets of ℙ1 are the finite sets, and such a set is connected if
and only if it consists of a single point. Because 𝜑(𝑉) is connected and closed, it must
either be a single point (and 𝜑 is constant) or ℙ1 (and 𝜑 is onto).
7.9. The only regular functions on a complete connected variety are the constant functions.

A regular function on a variety 𝑉 is a regular map 𝑓 ∶ 𝑉 → 𝔸1 ⊂ ℙ1 , to which we can

7.10. A regular map 𝜑 ∶ 𝑉 → 𝑊 from a complete connected variety to an affine variety

apply 7.8.

has image equal to a point. In particular, every complete connected affine variety is a point.

Embed 𝑊 as a closed subvariety of 𝔸𝑛 , and write 𝜑 = (𝜑1 , … , 𝜑𝑛 ), where 𝜑𝑖 is the

composite of 𝜑 with the coordinate function 𝑥𝑖 ∶ 𝔸𝑛 → 𝔸1 . Each 𝜑𝑖 is a regular function
on 𝑉, and hence is constant. (Alternatively, apply 5.11.) This proves the first statement,

7.11. In order to show that a variety 𝑉 is complete, it suffices to check that 𝑞 ∶ 𝑉 × 𝑇 → 𝑇

and the second follows from the first applied to the identity map.

is a closed mapping when 𝑇 is affine (or even an affine space 𝔸𝑛 ).

⋃
Every variety 𝑇 can be written as a finite union of open affine subvarieties 𝑇 = 𝑇𝑖 . If
𝑍 is closed in 𝑉 × 𝑇, then 𝑍𝑖 = 𝑍 ∩ (𝑉 × 𝑇𝑖 ) is closed in 𝑉 × 𝑇𝑖 . Therefore, 𝑞(𝑍𝑖 ) is closed
in 𝑇𝑖 for all 𝑖. As 𝑞(𝑍𝑖 ) = 𝑞(𝑍) ∩ 𝑇𝑖 , this shows that 𝑞(𝑍) is closed. This shows that it
def

suffices to check that 𝑉 × 𝑇 → 𝑇 is closed for all affine varieties 𝑇. But 𝑇 can be realized
as a closed subvariety of 𝔸𝑛 , and then 𝑉 × 𝑇 → 𝑇 is closed if 𝑉 × 𝔸𝑛 → 𝔸𝑛 is closed.
 b. Proper maps 163

Remarks
7.12. The statement that a complete variety 𝑉 is closed in every larger variety 𝑊 perhaps
explains the name: if 𝑉 is a complete subvariety of a connected variety 𝑊 and dim 𝑉 =
dim 𝑊, then 𝑉 = 𝑊. Contrast 𝔸𝑛 ⊂ ℙ𝑛 .

7.13. Here is another criterion: a variety 𝑉 is complete if and only if every regular map
𝐶 ∖ {𝑃} → 𝑉 extends uniquely to a regular map 𝐶 → 𝑉; here 𝑃 is a nonsingular point on
a curve 𝐶. Intuitively, this says that all Cauchy sequences have limits in 𝑉 and that the
limits are unique.

b. Proper maps
Definition 7.14. A regular map 𝜑 ∶ 𝑉 → 𝑆 of varieties is said to be proper if it is
“universally closed”, that is, if for all regular maps 𝑇 → 𝑆, the base change 𝜑′ ∶ 𝑉×𝑆 𝑇 → 𝑇
of 𝜑 is closed.

7.15. For example, a variety 𝑉 is complete if and only if the map 𝑉 → {point} is proper.

7.16. From its very definition, it is clear that the base change of a proper map is proper.

(a) if 𝑉 is complete, then 𝑉 × 𝑆 → 𝑆 is proper,

In particular,

(b) if 𝜑 ∶ 𝑉 → 𝑆 is proper, then the fibre 𝜑−1 (𝑃) over a point 𝑃 of 𝑆 is complete.
𝜑
7.17. If 𝜑 ∶ 𝑉 → 𝑆 is proper, and 𝑊 is a closed subvariety of 𝑉, then 𝑊 ,→ 𝑆 is proper.

Proposition 7.18. A composite of proper maps is proper.

Proof. Let 𝑉3 → 𝑉2 → 𝑉1 be proper maps, and let 𝑇 be a variety. Consider the diagram

𝑉3 ← 𝑉3 ×𝑉2 (𝑉2 ×𝑉1 𝑇) ≃ 𝑉3 ×𝑉1 𝑇

→

closed
←

𝑉2 ← 𝑉2 ×𝑉1 𝑇
→

closed
←

𝑉1 ← 𝑇.
→

Both smaller squares are cartesian, and hence so also is the outer square. The statement
is now obvious from the fact that a composite of closed maps is closed. 2

Corollary 7.19. If 𝑉 → 𝑆 is proper and 𝑆 is complete, then 𝑉 is complete.

Proof. Apply the proposition to 𝑉 → 𝑆 → {point}. 2

Corollary 7.20. The inverse image of a complete variety under a proper map is complete.

Proof. Let 𝜑 ∶ 𝑉 → 𝑆 be proper, and let 𝑍 be a complete subvariety of 𝑆. Then 𝑉 ×𝑆 𝑍 →

𝑍 is proper, and 𝑉 ×𝑆 𝑍 ≃ 𝜑−1 (𝑍). 2
164 7. Complete Varieties

Example 7.21. Let 𝑓 ∈ 𝑘[𝑇1 , … , 𝑇𝑛 , 𝑋, 𝑌] be homogeneous of degree 𝑚 in 𝑋 and 𝑌,

and let 𝐻 be the subvariety of 𝔸𝑛 × ℙ1 defined by

𝑓(𝑇1 , … , 𝑇𝑛 , 𝑋, 𝑌) = 0.

The projection map 𝔸𝑛 × ℙ1 → 𝔸𝑛 defines a regular map 𝐻 → 𝔸𝑛 , which is proper

(7.22, 7.15). The fibre over a point (𝑡1 , … , 𝑡𝑛 ) ∈ 𝔸𝑛 is the subvariety of ℙ1 defined by the
polynomial

𝑓(𝑡1 , … , 𝑡𝑛 , 𝑋, 𝑌) = 𝑎0 𝑋 𝑚 + 𝑎1 𝑋 𝑚−1 𝑌 + ⋯ + 𝑎𝑚 𝑌 𝑚 , 𝑎𝑖 ∈ 𝑘.

Assume that not all 𝑎𝑖 are zero. Then this is a homogeneous of degree 𝑚 and so the fibre
always has 𝑚 points counting multiplicities. The points that “disappeared off to infinity”
when ℙ1 was taken to be 𝔸1 (see p. 50) have literally become the point at infinity on ℙ1 .

c. Projective varieties are complete

The reader may skip this section since the main theorem is given a more explicit proof
in Theorem 7.31 below.

Theorem 7.22. A projective variety is complete.

Proof. After 7.3, it suffices to prove the Theorem for projective space ℙ𝑛 itself; thus we
have to prove that the projection map ℙ𝑛 × 𝑊 → 𝑊 is a closed mapping in the case that
𝑊 is an irreducible affine variety (7.11).
Write 𝑝 for the projection 𝑊 × ℙ𝑛 → 𝑊. We have to show that 𝑍 closed in 𝑊 × ℙ𝑛
implies that 𝑝(𝑍) closed in 𝑊. If 𝑍 is empty, this is true, and so we can assume it to be
nonempty. Then 𝑍 is a finite union of irreducible closed subsets 𝑍𝑖 of 𝑊 × ℙ𝑛 , and it
suffices to show that each 𝑝(𝑍𝑖 ) is closed. Thus we may assume that 𝑍 is irreducible,
and hence that 𝑍 = 𝑉(𝔟) with 𝔟 a graded prime ideal in 𝐵 = 𝐴[𝑋0 , … , 𝑋𝑛 ] (6.51).
If 𝑝(𝑍) is contained in some closed subvariety 𝑊 ′ of 𝑊, then 𝑍 is contained in
𝑊 ′ × ℙ𝑛 , and we can replace 𝑊 with 𝑊 ′ . This allows us to assume that 𝑝(𝑍) is dense in
𝑊, and we now have to show that 𝑝(𝑍) = 𝑊.
Because 𝑝(𝑍) is dense in 𝑊, the image of the cone 𝑉 aff (𝔟) under the projection
𝑊 × 𝔸𝑛+1 → 𝑊 is also dense in 𝑊, and so (see 3.34a) the map 𝐴 → 𝐵∕𝔟 is injective.
Let 𝑤 ∈ 𝑊: we shall show that if 𝑤 ∉ 𝑝(𝑍), i.e., if there does not exist a 𝑃 ∈ ℙ𝑛 such
that (𝑤, 𝑃) ∈ 𝑍, then 𝑝(𝑍) is empty, which is a contradiction.
Let 𝔪 ⊂ 𝐴 be the maximal ideal corresponding to 𝑤. Then 𝔪𝐵 + 𝔟 is a graded ideal,
and 𝑉(𝔪𝐵 + 𝔟) = 𝑉(𝔪𝐵) ∩ 𝑉(𝔟) = (𝑤 × ℙ𝑛 ) ∩ 𝑉(𝔟), and so 𝑤 will be in the image of 𝑍
unless 𝑉(𝔪𝐵 + 𝔟) ≠ ∅. But if 𝑉(𝔪𝐵 + 𝔟) = ∅, then 𝔪𝐵 + 𝔟 ⊃ (𝑋0 , … , 𝑋𝑛 )𝑁 for some 𝑁
(by 6.51b), and so 𝔪𝐵 + 𝔟 contains the set 𝐵𝑁 of homogeneous polynomials of degree 𝑁.
Because 𝔪𝐵 and 𝔟 are graded ideals,

𝐵𝑁 ⊂ 𝔪𝐵 + 𝔟 ⇐⇒ 𝐵𝑁 = 𝔪𝐵𝑁 + 𝐵𝑁 ∩ 𝔟.

In detail: the first inclusion says that an 𝑓 ∈ 𝐵𝑁 can be written 𝑓 = 𝑔 + ℎ with 𝑔 ∈ 𝔪𝐵

and ℎ ∈ 𝔟. On equating homogeneous components, we find that 𝑓𝑁 = 𝑔𝑁 + ℎ𝑁 .
∑ ∑
Moreover: 𝑓𝑁 = 𝑓; if 𝑔 = 𝑚𝑖 𝑏𝑖 , 𝑚𝑖 ∈ 𝔪, 𝑏𝑖 ∈ 𝐵, then 𝑔𝑁 = 𝑚𝑖 𝑏𝑖𝑁 ; and ℎ𝑁 ∈ 𝔟
because 𝔟 is homogeneous. Together these show 𝑓 ∈ 𝔪𝐵𝑁 + 𝐵𝑁 ∩ 𝔟.
Let 𝑀 = 𝐵𝑁 ∕𝐵𝑁 ∩𝔟, regarded as an 𝐴-module. The displayed equation says that 𝑀 =
𝔪𝑀. The argument in the proof of Nakayama’s lemma (1.3) shows that (1+𝑚)𝑀 = 0 for
 d. Elimination theory 165

some 𝑚 ∈ 𝔪. Because 𝐴 → 𝐵∕𝔟 is injective, the image of 1 + 𝑚 in 𝐵∕𝔟 is nonzero. But

𝑀 = 𝐵𝑁 ∕𝐵𝑁 ∩ 𝔟 ⊂ 𝐵∕𝔟, which is an integral domain, and so the equation (1 + 𝑚)𝑀 = 0
implies that 𝑀 = 0. Hence 𝐵𝑁 ⊂ 𝔟, and so 𝑋𝑖𝑁 ∈ 𝔟 for all 𝑖, which contradicts the
assumption that 𝑍 = 𝑉(𝔟) is nonempty. 2

Notes
7.23. Every complete curve is projective.

7.24. Every nonsingular complete surface is projective (Zariski), but there exist singular
complete surfaces that are not projective (Nagata).

7.25. There exist nonsingular complete three-dimensional varieties that are not projec-
tive (Nagata, Hironaka).

7.26. A nonsingular complete irreducible variety 𝑉 is projective if and only if every finite
set of points of 𝑉 is contained in an open affine subset of 𝑉 (Conjecture of Chevalley;
proved by Kleiman1 ; see 6.22 for the necessity).

d. Elimination theory
When given a system of polynomial equations to solve, we first use some of the equations
to eliminate some of the variables; we then find the solutions of the reduced system, and
go back to find the solutions of the original system. Elimination theory does this more

The fact that ℙ𝑛 is complete has the following explicit restatement: for each system
systematically.

of polynomial equations

⎧ 𝑃1 (𝑋1 , … , 𝑋𝑚 ; 𝑌0 , … , 𝑌𝑛 ) = 0
(∗) ⋮
⎨
𝑃𝑟 (𝑋1 , … , 𝑋𝑚 ; 𝑌0 , … , 𝑌𝑛 ) = 0
⎩
such that each 𝑃𝑖 is homogeneous in the 𝑌𝑗 , there exists a system of polynomial equations

⎧ 𝑅1 (𝑋1 , … , 𝑋𝑚 ) = 0
(∗∗) ⋮
⎨
𝑅 (𝑋 , … , 𝑋𝑚 ) = 0
⎩ 𝑠 1
with the following property; an 𝑚-tuple (𝑎1 , … , 𝑎𝑚 ) is a solution of (**) if and only if
there exists a nonzero 𝑛-tuple (𝑏0 , … , 𝑏𝑛 ) such that (𝑎1 , … , 𝑎𝑚 , 𝑏0 , … , 𝑏𝑛 ) is a solution of
(*). In other words, the polynomials 𝑃𝑖 (𝑎1 , … , 𝑎𝑚 ; 𝑌0 , … , 𝑌𝑛 ) have a common zero if and
only if 𝑅𝑗 (𝑎1 , … , 𝑎𝑚 ) = 0 for all 𝑗. The polynomials 𝑅𝑗 are said to have been obtained
from the polynomials 𝑃𝑖 by elimination of the variables 𝑌𝑖 .
Unfortunately, the proof we gave of the completeness of ℙ𝑛 , while short and elegant,
gives no indication of how to construct (**) from (*). The purpose of elimination theory
is to provide an algorithm for doing this.
1
Kleiman, Steven L., Toward a numerical theory of ampleness. Ann. of Math. (2) 84 1966 293–344
(Theorem 3, p. 327, et seq.). See also, Hartshorne, Robin, Ample subvarieties of algebraic varieties. Lecture
Notes in Mathematics, Vol. 156 Springer, 1970, I §9 p45.
166 7. Complete Varieties

Elimination theory: special case

Let 𝑃 = 𝑠0 𝑋 𝑚 + 𝑠1 𝑋 𝑚−1 + ⋯ + 𝑠𝑚 and 𝑄 = 𝑡0 𝑋 𝑛 + 𝑡1 𝑋 𝑛−1 + ⋯ + 𝑡𝑛 be polynomials. The
resultant of 𝑃 and 𝑄 is defined to be the determinant
|| 𝑠 𝑠1 … 𝑠𝑚 ||
|| 0 || 𝑛 rows
|| 𝑠0 … 𝑠𝑚 ||
|| ||
|| … … |||
|| ||
|| 𝑡 𝑡1 … 𝑡𝑛 ||
|| 0 ||
|| 𝑡0 … 𝑡𝑛 ||
|| | 𝑚 rows
|| … … |||
|
There are 𝑛 rows with 𝑠0 … 𝑠𝑚 and 𝑚 rows with 𝑡0 … 𝑡𝑛 , so that the matrix is (𝑚 + 𝑛) ×
(𝑚 + 𝑛); all blank spaces are to be filled with zeros. The resultant is a polynomial in the
coefficients of 𝑃 and 𝑄.

Proposition 7.27. The resultant Res(𝑃, 𝑄) = 0 if and only if

(a) both 𝑠0 and 𝑡0 are zero; or
(b) the two polynomials have a common root.

Proof. If (a) holds, then Res(𝑃, 𝑄) = 0 because the first column is zero. Suppose that 𝛼
is a common root of 𝑃 and 𝑄, so that there exist polynomials 𝑃1 and 𝑄1 of degrees 𝑚 − 1
and 𝑛 − 1 respectively such that

𝑃(𝑋) = (𝑋 − 𝛼)𝑃1 (𝑋), 𝑄(𝑋) = (𝑋 − 𝛼)𝑄1 (𝑋).

Using these equalities, we find that

𝑃(𝑋)𝑄1 (𝑋) − 𝑄(𝑋)𝑃1 (𝑋) = 0. (33)

On equating the coefficients of 𝑋 𝑚+𝑛−1 , … , 𝑋, 1 in (33) to zero, we find that the coeffi-
cients of 𝑃1 and 𝑄1 are the solutions of a system of 𝑚 + 𝑛 linear equations in 𝑚 + 𝑛
unknowns. The matrix of coefficients of the system is the transpose of the matrix

⎛ 𝑠0 𝑠1 … 𝑠𝑚 ⎞
⎜ 𝑠0 … 𝑠𝑚 ⎟
⎜ … … ⎟
⎜ 𝑡 𝑡1 … 𝑡𝑛 ⎟
⎜ 0 ⎟
𝑡0 … 𝑡𝑛
⎜ ⎟
… …
⎝ ⎠

that Res(𝑃, 𝑄) = 0.
The existence of the solution shows that this matrix has determinant zero, which implies

Conversely, suppose that Res(𝑃, 𝑄) = 0 but neither 𝑠0 nor 𝑡0 is zero. Because the

𝑃1 and 𝑄1 satisfying (33). A root 𝛼 of 𝑃 must be also be a root of 𝑃1 or of 𝑄. If the former,

above matrix has determinant zero, we can solve the linear equations to find polynomials

cancel 𝑋 − 𝛼 from the left hand side of (33), and consider a root 𝛽 of 𝑃1 ∕(𝑋 − 𝛼). As
deg 𝑃1 < deg 𝑃, this argument eventually leads to a root of 𝑃 that is not a root of 𝑃1 , and
so must be a root of 𝑄. 2
 d. Elimination theory 167

The proposition can be restated in projective terms. We define the resultant of two
homogeneous polynomials

𝑃(𝑋, 𝑌) = 𝑠0 𝑋 𝑚 + 𝑠1 𝑋 𝑚−1 𝑌 + ⋯ + 𝑠𝑚 𝑌 𝑚 , 𝑄(𝑋, 𝑌) = 𝑡0 𝑋 𝑛 + ⋯ + 𝑡𝑛 𝑌 𝑛 ,

Proposition 7.28. The resultant Res(𝑃, 𝑄) = 0 if and only if 𝑃 and 𝑄 have a common
exactly as in the nonhomogeneous case.

zero in ℙ1 .

Proof. The zeros of 𝑃(𝑋, 𝑌) in ℙ1 are of the form:

(a) (1 ∶ 0) in the case that 𝑠0 = 0;
(b) (𝑎 ∶ 1) with 𝑎 a root of 𝑃(𝑋, 1).
Since a similar statement is true for 𝑄(𝑋, 𝑌), 7.28 is a restatement of 7.27. 2

Now regard the coefficients of 𝑃 and 𝑄 as indeterminates. The pairs of polynomials

(𝑃, 𝑄) are parametrized by the space 𝔸𝑚+1 × 𝔸𝑛+1 = 𝔸𝑚+𝑛+2 . Consider the closed subset
𝑉(𝑃, 𝑄) in 𝔸𝑚+𝑛+2 × ℙ1 . The proposition shows that its projection on 𝔸𝑚+𝑛+2 is the set
defined by Res(𝑃, 𝑄) = 0. Thus, not only have we shown that the projection of 𝑉(𝑃, 𝑄)
is closed, but we have given an algorithm for passing from the polynomials defining the
closed set to those defining its projection.
Elimination theory does this in general. Given a family of polynomials

𝑃𝑖 (𝑇1 , … , 𝑇𝑚 ; 𝑋0 , … , 𝑋𝑛 ),

homogeneous in the 𝑋𝑖 , elimination theory gives an algorithm for finding polynomials

𝑅𝑗 (𝑇1 , … , 𝑇𝑚 ) such that the 𝑃𝑖 (𝑎1 , … , 𝑎𝑚 ; 𝑋0 , … , 𝑋𝑛 ) have a common zero if and only if
𝑅𝑗 (𝑎1 , … , 𝑎𝑚 ) = 0 for all 𝑗. (Theorem 7.22 shows only that the 𝑅𝑗 exist.)
Macaulay2Web can find the resultant of two polynomials in one variable: for example,
entering

R=ZZ[x,a,b] (and then)

resultant((x+a)^5,(x+b)^5,x)

gives the answer (−𝑎 + 𝑏)25 . Explanation: the polynomials have a common root if and
only if 𝑎 = 𝑏, and this can happen in 25 ways.

Elimination theory: general case

more explicit than that given above. Throughout, 𝑘 is a field (not necessarily algebraically
In this subsection, we give a proof of Theorem 7.22, following Cartier and Tate2 , that is

closed) and 𝐾 is an algebraically closed field containing 𝑘.

Theorem 7.29. For any graded ideal 𝔞 in 𝑘[𝑋0 , … , 𝑋𝑛 ], exactly one of the following state-

(a) there exists an integer 𝑑0 ≥ 0 such that 𝔞 contains every homogeneous polynomial of
ments is true:

degree 𝑑 ≥ 𝑑0 ;
(b) the ideal 𝔞 has a nontrivial zero in 𝐾 𝑛+1 .
2
Cartier, P., Tate, J., A simple proof of the main theorem of elimination theory in algebraic geometry.
Enseign. Math. (2) 24 (1978), no. 3-4, 311–317.
168 7. Complete Varieties

Proof. Statement (a) says that the radical of 𝔞 contains (𝑋0 , … , 𝑋𝑛 ), and so the theorem
is a restatement of 6.2(a), which we deduced from the strong Nullstellensatz. For a direct

⨁
proof of it, see the article of Cartier and Tate. 2

Theorem 7.30. Let 𝑅 = 𝑑∈ℕ 𝑅𝑑 be a graded 𝑘-algebra such that 𝑅0 = 𝑘, 𝑅 is generated

as a 𝑘-algebra by 𝑅1 , and 𝑅𝑑 is finite-dimensional for all 𝑑. Then exactly one of the following

(a) there exists an integer 𝑑0 ≥ 0 such that 𝑅𝑑 = 0 for all 𝑑 ≥ 𝑑0 ;

statements is true:

(b) no 𝑅𝑑 = 0, and there exists a 𝑘-algebra homomorphism 𝑅 → 𝐾 whose kernel is not

def ⨁
equal to 𝑅+ = 𝑑≥1 𝑅𝑑 .

Proof. The hypotheses on 𝑅 say that it is a quotient of 𝑘[𝑋0 , … , 𝑋𝑛 ] by a graded ideal.

Therefore 7.30 is a restatement of 7.29. 2

Let 𝑃1 , … , 𝑃𝑟 be polynomials in 𝑘[𝑇1 , … , 𝑇𝑚 ; 𝑋0 , … , 𝑋𝑛 ] with 𝑃𝑗 homogeneous of de-

gree 𝑑𝑗 in the variables 𝑋0 , … , 𝑋𝑛 . Let 𝐽 be the ideal (𝑃1 , … , 𝑃𝑟 ) in 𝑘[𝑇1 , … , 𝑇𝑚 ; 𝑋0 , … , 𝑋𝑛 ],
and let 𝔄 be the ideal of polynomials 𝑓 in 𝑘[𝑇1 , … , 𝑇𝑚 ] with the following property: there
exists an integer 𝑁 ≥ 1 such that 𝑓𝑋0𝑁 , … , 𝑓𝑋𝑛𝑁 all lie in 𝐽.

Theorem 7.31. Let 𝑉 be the zero set of 𝐽 in 𝔸𝑛 (𝐾) × ℙ𝑛 (𝐾). The projection of 𝑉 into
𝔸𝑛 (𝐾) is the zero set of 𝔄.

Consider the ring 𝐵 = 𝑘[𝑇1 , … , 𝑇𝑚 ; 𝑋0 , … , 𝑋𝑛 ] and its subring 𝐵0 = 𝑘[𝑇1 , … , 𝑇𝑚 ].

Then 𝐵 is a graded 𝐵0 -algebra with 𝐵𝑑 the 𝐵0 -submodule generated by the monomials of
⨁
degree 𝑑 in 𝑋0 , … , 𝑋𝑛 , and 𝐽 is a homogeneous (graded) ideal in 𝐵. Let 𝐴 = 𝑑∈ℕ 𝐴𝑑
⨁
be the quotient graded ring 𝐵∕𝐽 = 𝑑∈ℕ 𝐵𝑑 ∕(𝐵𝑑 ∩ 𝐽). Let 𝔖 be the ideal of elements 𝑎
of 𝐴0 such that 𝑎𝐴𝑑 = 0 for all sufficiently large 𝑑.

Theorem 7.32. A ring homomorphism 𝜑 ∶ 𝐴0 → 𝐾 extends to a ring homomorphism

def ⨁
Ψ ∶ 𝐴 → 𝐾 not annihilating the ideal 𝐴+ = 𝑑≥1 𝐴𝑑 if and only if 𝜑(𝔖) = 0.

Following Cartier and Tate, we leave it to reader to check that 7.32 is equivalent to
7.31.

⨁
Proof of Theorem 7.32
We shall prove 7.32 for any graded ring 𝐴 = 𝑑≥0
𝐴𝑑 satisfying the following two

(a) as an 𝐴0 -algebra, 𝐴 is generated by 𝐴1 ;

conditions:

(b) for every 𝑑 ≥ 0, 𝐴𝑑 is finitely generated as an 𝐴0 -module.

In the statement of the theorem, 𝐾 is any algebraically closed field.
The proof proceeds by replacing 𝐴 with other graded rings with the properties (a)
and (b) and also having the property that no 𝐴𝑑 is zero.
Let 𝜑 ∶ 𝐴0 → 𝐾 be a homomorphism such that 𝜑(𝔖) = 0, and let 𝔓 = Ker(𝜑). Then
𝔓 is a prime ideal of 𝐴0 containing 𝔖.
Step 1. Let 𝐽 be the ideal of elements 𝑎 of 𝐴 for which there exists an 𝑠 ∈ 𝐴0 ∖ 𝔓 such
that 𝑠𝑎 = 0. For every 𝑑 ≥ 0, the annihilator of the 𝐴0 -module 𝐴𝑑 is contained in 𝔖,
hence in 𝔓, and so 𝐽 ∩ 𝐴𝑑 ≠ 𝐴𝑑 . The ideal 𝐽 is graded, and the quotient ring 𝐴′ = 𝐴∕𝐽
has the required properties.
 e. The rigidity theorem; abelian varieties 169

Step 2. Let 𝐴′′ be the ring of fractions of 𝐴′ whose denominators are in 𝛴 = 𝐴0′ ∖ 𝔓.
Let 𝐴𝑑′′ be the set of fractions with numerator in 𝐴𝑑′ and denominator in 𝛴. Then
def

⨁
𝐴′′ = 𝑑≥0 𝐴𝑑′′ is a graded ring with the required properties, and 𝐴0′′ is a local ring with
maximal ideal 𝔓′′ = 𝔓′ ⋅ 𝐴0′ .
Step 3. Let 𝑅 be the quotient of 𝐴′′ by the graded ideal 𝔓′′ ⋅ 𝐴′′ . As 𝐴𝑑′′ is a nonzero
def

finitely generated module over the local ring 𝐴0′′ , Nakayama’s lemma shows that 𝐴𝑑′′ ≠
𝔓′′ 𝐴𝑑′′ . Therefore 𝑅 is graded ring with the required properties, and 𝑘 = 𝑅0 = 𝐴0′′ ∕𝔓′′
def

Step 4. At this point 𝑅 satisfies the hypotheses of Theorem 7.30. Let 𝜀 be the composite
is a field.

𝐴 → 𝐴′ → 𝐴′′ → 𝑅.
of the natural maps

In degree 0, this is nothing but the natural map from 𝐴0 to 𝑘 with kernel 𝔓. As 𝜑 has
the same kernel, it factors through 𝜀0 , making 𝐾 into an algebraically closed extension
of 𝑘. Now, by Theorem 7.30, there exists a 𝑘-algebra homomorphism 𝑓 ∶ 𝑅 → 𝐾 such
that 𝑓(𝑅+ ) ≠ 0. The composite map Ψ = 𝑓◦𝜀 has the required properties. 2

For more on elimination theory, see Cox et al. 2015, Chapter 8, Section 5.
Aside 7.33. Elimination theory became unfashionable several decades ago — one prominent
algebraic geometer went so far as to announce that Theorem 7.22 eliminated elimination theory
from mathematics,3 provoking Abhyankar, who prefers equations to abstractions, to start the
chant “eliminate the eliminators of elimination theory”. With the rise of computers, it has
become fashionable again.

e. The rigidity theorem; abelian varieties

First an observation: for any point 𝑤 ∈ 𝑊, the projection map 𝑉 × 𝑊 → 𝑉 defines an

The paucity of maps between complete varieties has some interesting consequences.

isomorphism 𝑉 × {𝑤} → 𝑉 with inverse 𝑣 ↦ (𝑣, 𝑤) ∶ 𝑉 → 𝑉 × 𝑊 (this map is regular

Theorem 7.34 (Rigidity Theorem). Let 𝜑 ∶ 𝑉 × 𝑊 → 𝑇 be 𝑉×𝑊

because its components are).

a regular map, and assume that 𝑉 is complete, 𝑉 and 𝑊 are

𝑞
→
irreducible, and 𝑇 is separated. If 𝜑(𝑣, 𝑤0 ) is independent of 𝑣
←
←

𝜑 𝑊
for one 𝑤0 ∈ 𝑊, then 𝜑(𝑣, 𝑤) = 𝑔(𝑤) with 𝑔 a regular map ←
𝑔 ∶ 𝑊 → 𝑇. 𝑔
𝑇
→

Proof. Choose a 𝑣0 ∈ 𝑉, and consider the regular map

𝑔 ∶ 𝑊 → 𝑇, 𝑤 ↦ 𝜑(𝑣0 , 𝑤).

We shall show that 𝜑 = 𝑔◦𝑞. Because 𝑉 is complete, the projection map 𝑞 ∶ 𝑉 × 𝑊 → 𝑊

is closed. Let 𝑈 be an open affine neighbourhood 𝑈 of 𝜑(𝑣0 , 𝑤0 ); then 𝑇 ∖ 𝑈 is closed in
𝑇, 𝜑−1 (𝑇 ∖ 𝑈) is closed in 𝑉 × 𝑊, and

𝐶 = 𝑞(𝜑−1 (𝑇 ∖ 𝑈))
def

3
Weil 1946, p. 31: “The device that follows, which, it may be hoped, finally eliminates from algebraic ge-
ometry the last traces of elimination-theory, is borrowed from C. Chevalley’s Princeton lectures.” Demazure
credits Dieudonné with saying: “Il faut éliminer la théorie de l’élimination.”
170 7. Complete Varieties

is closed in 𝑊. By definition, 𝐶 consists of the 𝑤 ∈ 𝑊 such that 𝜑(𝑣, 𝑤) ∉ 𝑈 for some

𝑣 ∈ 𝑉, and so
𝑊 ∖ 𝐶 = {𝑤 ∈ 𝑊 ∣ 𝜑(𝑉 × {𝑤}) ⊂ 𝑈}.
As 𝜑(𝑉, 𝑤0 ) = 𝜑(𝑣0 , 𝑤0 ), we see that 𝑤0 ∈ 𝑊 ∖ 𝐶. Therefore 𝑊 ∖ 𝐶 is nonempty, and
so it is dense in 𝑊. As 𝑉 × {𝑤} is complete and 𝑈 is affine, 𝜑(𝑉 × {𝑤}) must be a point
whenever 𝑤 ∈ 𝑊 ∖ 𝐶 (see 7.10); in fact

𝜑(𝑉 × {𝑤}) = 𝜑(𝑣0 , 𝑤) = 𝑔(𝑤).

We have shown that 𝜑 and 𝑔◦𝑞 agree on the dense subset 𝑉 × (𝑊 ∖ 𝐶) of 𝑉 × 𝑊, and
therefore on the whole of 𝑉 × 𝑊. 2

Corollary 7.35. Let 𝜑 ∶ 𝑉 × 𝑊 → 𝑇 be a regular map, and assume that 𝑉 is complete,

that 𝑉 and 𝑊 are irreducible, and that 𝑇 is separated. If there exist points 𝑣0 ∈ 𝑉, 𝑤0 ∈ 𝑊,
𝑡0 ∈ 𝑇 such that
𝜑(𝑉 × {𝑤0 }) = {𝑡0 } = 𝜑({𝑣0 } × 𝑊),
then 𝜑(𝑉 × 𝑊) = {𝑡0 }.

Proof. With 𝑔 as in the proof of the theorem,

𝜑(𝑣, 𝑤) = 𝑔(𝑤) = 𝜑(𝑣0 , 𝑤) = 𝑡0 . 2

In more colloquial terms, the corollary says that if 𝜑 collapses a vertical and a hor-
izontal slice to a point, then it collapses the whole of 𝑉 × 𝑊 to a point, which must
therefore be “rigid”.

Definition 7.36. An abelian variety is a complete connected group variety.

Theorem 7.37. Every regular map 𝛼 ∶ 𝐴 → 𝐵 of abelian varieties is the composite of a

homomorphism with a translation; in particular, a regular map 𝛼 ∶ 𝐴 → 𝐵 such that
𝛼(0) = 0 is a homomorphism.

Proof. After composing 𝛼 with a translation, we may suppose that 𝛼(0) = 0. Consider

𝜑 ∶ 𝐴 × 𝐴 → 𝐵, 𝜑(𝑎, 𝑎′ ) = 𝛼(𝑎 + 𝑎′ ) − 𝛼(𝑎) − 𝛼(𝑎′ ).

the map

Then 𝜑(𝐴 × 0) = 0 = 𝜑(0 × 𝐴) and so 𝜑 = 0. This means that 𝛼 is a homomorphism. 2

Corollary 7.38. The group law on an abelian variety is commutative.

map taking an element to its inverse is a homomorphism: if (𝑔ℎ)−1 = 𝑔−1 ℎ−1 , then, on
Proof. Commutative groups are distinguished among all groups by the fact that the

taking inverses, we find that 𝑔ℎ = ℎ𝑔. Since the negative map, 𝑎 ↦ −𝑎 ∶ 𝐴 → 𝐴, takes
the identity element to itself, the theorem shows that it is a homomorphism. 2

Aside. Abelian varieties arose out of the study of Abelian integrals, whence their name.
 f. Chow’s Lemma 171

f. Chow’s Lemma
The next theorem is a useful tool in extending results from projective varieties to complete

Theorem 7.39 (Chow’s Lemma). Let 𝑉 be a complete irreducible variety. There exists
varieties. It shows that a complete variety is not far from a projective variety.

a projective algebraic variety 𝑉 ′ and a surjective regular map 𝑓 ∶ 𝑉 ′ → 𝑉 such that 𝑓

induces an isomorphism 𝑓 −1 (𝑈) → 𝑈 for some dense open subset 𝑈 of 𝑉 (in particular, 𝑓

𝑓 −1 (𝑈) → 𝑈
isomorphism
is birational),
←

∩ ∩

𝑉′ → 𝑉.
onto
𝑓
←

Write 𝑉 as a finite union of nonempty open affines, 𝑉 = 𝑈1 ∪ … ∪ 𝑈𝑛 , and let

⋂
𝑈 = 𝑈𝑖 . Because 𝑉 is irreducible, 𝑈 is a dense in 𝑉. Realize each 𝑈𝑖 as a dense open
def ∏
subset of a projective variety 𝑃𝑖 . Then 𝑃 = 𝑖 𝑃𝑖 is a projective variety (6.26). We shall
construct an algebraic variety 𝑉 ′ and regular maps 𝑓 ∶ 𝑉 ′ → 𝑉 and 𝑔 ∶ 𝑉 ′ → 𝑃 such

(a) 𝑓 is surjective and induces an isomorphism 𝑓 −1 (𝑈) → 𝑈;

that

(b) 𝑔 is a closed immersion (hence 𝑉 ′ is projective).

Let 𝜑0 (resp. 𝜑𝑖 ) denote the given inclusion of 𝑈 into 𝑉 (resp. into 𝑃𝑖 ), and let

𝜑 = (𝜑0 , 𝜑1 , … , 𝜑𝑛 ) ∶ 𝑈 → 𝑉 × 𝑃1 × ⋯ × 𝑃𝑛 ,

be the diagonal map. We set 𝑈 ′ = 𝜑(𝑈) and 𝑉 ′ equal to the closure of 𝑈 ′ in 𝑉×𝑃1 ×⋯×𝑃𝑛 .
The projection maps 𝑝 ∶ 𝑉 × 𝑃 → 𝑉 and 𝑞 ∶ 𝑉 × 𝑃 → 𝑃 restrict to regular maps
𝑓 ∶ 𝑉 ′ → 𝑉 and 𝑔 ∶ 𝑉 ′ → 𝑃. Thus, we have a commutative diagram

𝜑0 → 𝑉
𝑓 →
𝑝
→

𝑈 → 𝑉 ′ ← → 𝑉×𝑃
𝜑
← ←
←

← (34)

𝑞
←

→
←

𝑃.
→

In the upper-left triangle of the diagram (34), the maps 𝜑 and 𝜑0 are isomorphisms from
Proof of (a)

𝑈 onto its images 𝑈 ′ and 𝑈. Therefore 𝑓 restricts to an isomorphism 𝑈 ′ → 𝑈. Note that

𝑈 ′ = {(𝑢, 𝜑1 (𝑢), … , 𝜑𝑛 (𝑢)) ∣ 𝑢 ∈ 𝑈},

which is the graph of the map (𝜑1 , … , 𝜑𝑛 ) ∶ 𝑈 → 𝑃. Therefore, 𝑈 ′ is closed in 𝑈 × 𝑃

𝑈 ′ = 𝑉 ′ ∩ (𝑈 × 𝑃) = 𝑓 −1 (𝑈).
(5.28), and so

The map 𝑓 is dominant, and 𝑓(𝑉 ′ ) = 𝑝(𝑉), which is closed because 𝑃 is complete.
Hence 𝑓 is surjective.
172 7. Complete Varieties

We first show that 𝑔 is an immersion. As this is a local condition, it suffices to find

Proof of (b)

⋃ 𝑔
open subsets 𝑉𝑖 ⊂ 𝑃 such that 𝑞−1 (𝑉𝑖 ) ⊃ 𝑉 ′ and each map 𝑉 ′ ∩ 𝑞−1 (𝑉𝑖 ) ,→ 𝑉𝑖 is an
immersion.

𝑉𝑖 = 𝑝𝑖−1 (𝑈𝑖 ) = 𝑃1 × ⋯ × 𝑈𝑖 × ⋯ × 𝑃𝑛
We set

where 𝑝𝑖 is the projection map 𝑃 → 𝑃𝑖 .

We first show that the sets 𝑞−1 (𝑉𝑖 ) cover 𝑉 ′ . The sets 𝑈𝑖 cover 𝑉, hence the sets
𝑓 −1 (𝑈𝑖 ) cover 𝑉 ′ , and so it suffices to show that

𝑞 −1 (𝑉𝑖 ) ⊃ 𝑓 −1 (𝑈𝑖 )

for all 𝑖. Consider the diagrams

𝑈 → 𝑈𝑖
𝜑0
𝑞−1 (𝑉𝑖 ) → 𝑈𝑖 𝑓 −1 (𝑈𝑖 ) → 𝑈𝑖
← ← 𝑓 ←

𝜑𝑖 𝜑𝑖 𝜑 𝜑𝑖

←→

←→
←→

←→
←→

←→

𝑉×𝑃 → 𝑃𝑖 .
𝑝𝑖 ◦𝑞
𝑉×𝑃 → 𝑃𝑖 𝑉×𝑃 → 𝑃𝑖
𝑝𝑖 ◦𝑞
← ←𝑝𝑖 ◦𝑞 ←

The diagram at left is cartesian, i.e., it realizes 𝑞−1 (𝑉𝑖 ) as the fibred product

𝑞−1 (𝑉𝑖 )_ = (𝑉 × 𝑃) ×𝑃𝑖 𝑈𝑖 ,

and so it suffices to show that the middle diagram commutes. But 𝑈 ′ is dense in 𝑉 ′ ,
hence in 𝑓 −1 (𝑈𝑖 ), and so it suffices to prove that the middle diagram commutes with
𝑓 −1 (𝑈𝑖 ) replaced by 𝑈 ′ . But then it becomes the diagram at right, which obviously
commutes.
𝑔
𝑉 ′ ∩ 𝑞−1 (𝑉𝑖 ) ,→ 𝑉𝑖
We next show that

is an immersion for each 𝑖. Recall that

∏
𝑉𝑖 = 𝑈𝑖 × 𝑃𝑖 , where 𝑃𝑖 = 𝑃𝑗 .
𝑗≠𝑖

𝑞−1 (𝑉𝑖 ) = 𝑉 × 𝑈𝑖 × 𝑃𝑖 ⊂ 𝑉 × 𝑃.
and so

Let 𝛤𝑖 denote the graph of the map

𝑝𝑖
(𝑈𝑖 × 𝑃𝑖 ,→ 𝑈𝑖 → 𝑉) .
( ) ( )
Being a graph, 𝛤𝑖 is closed in 𝑉× 𝑈𝑖 × 𝑃𝑖 and the projection map 𝑉× 𝑈𝑖 × 𝑃𝑖 → 𝑈𝑖 ×𝑃𝑖
restricts to an isomorphism 𝛤𝑖 → 𝑈𝑖 × 𝑃𝑖 . In other words, 𝛤𝑖 is closed in 𝑞−1 (𝑉𝑖 ), and
the projection map 𝑞 −1 (𝑉𝑖 ) → 𝑉𝑖 restricts to an isomorphism 𝛤𝑖 → 𝑉𝑖 . As 𝛤𝑖 is closed in
𝑞−1 (𝑉𝑖 ) and contains 𝑈 ′ , it contains 𝑉 ′ ∩𝑞 −1 (𝑉𝑖 ), and so the projection map 𝑞−1 (𝑉𝑖 ) → 𝑉𝑖
restricts to an immersion 𝑉 ′ ∩ 𝑞−1 (𝑉𝑖 ) → 𝑉𝑖 .
Finally, 𝑉 × 𝑃 is complete because 𝑉 and 𝑃 are, and so 𝑉 ′ is complete (7.3). Hence
𝑔(𝑉) is closed (7.7), and so 𝑔 is a closed immersion.
 g. Analytic spaces; Chow’s theorem 173

Notes
7.40. Let 𝑉 be a complete variety, and let 𝑉1 , … , 𝑉𝑠 be the irreducible components of 𝑉.
Each 𝑉𝑖 is complete (7.4), and so there exists a surjective birational regular map 𝑉𝑖′ → 𝑉𝑖
⨆
with 𝑉𝑖′ projective (7.39). Now 𝑉𝑖′ is projective 6.26, and the composite
⨆ ⨆
𝑉𝑖′ → 𝑉𝑖 → 𝑉

is surjective and birational.

7.41. Chow (1956, Lemma 1)4 proved essentially the statement 7.42 by essentially the
above argument. He used the lemma to prove that all homogeneous spaces are quasi-
projective. See also EGA II, 5.6.1.

g. Analytic spaces; Chow’s theorem

We summarize a little of Serre, J-P., Géométrie algébrique et géométrie analytique. Ann.
Inst. Fourier, Grenoble 6 (1955–1956), 1–42, commonly referred to as GAGA.

variety 𝑉, there exists a projective algebraic variety 𝑉 ′ and a birational regular map 𝜑
7.42. The following statement is more general than Theorem 7.39: for every algebraic

from an open dense subset 𝑈 of 𝑉 ′ onto 𝑉 whose graph is closed in 𝑉 ′ × 𝑉; the subset 𝑈
equals 𝑉 ′ if and only if 𝑉 is complete (GAGA, p. 12).

Proposition 7.43. An algebraic variety 𝑉 over ℂ is complete if and only if 𝑉(ℂ) is compact
in the complex topology.

Proof. The proof uses Chow’s lemma (GAGA, Proposition 6, p. 12). 2

A subset 𝑉 of ℂ𝑛 is analytic if every 𝑃 ∈ 𝑉 admits an open neighbourhood 𝑈 in

ℂ𝑛 such that 𝑉 ∩ 𝑈 is the zero set of a finite collection of holomorphic functions on 𝑈.

Let 𝑉 ′ be an open subset of an analytic set 𝑉. A function 𝑓 ∶ 𝑉 ′ → ℂ is analytic if,

Analytic subsets are locally closed.

for every 𝑃 ∈ 𝑉 ′ , there exists an open neighbourhood 𝑈 of 𝑃 in ℂ𝑛 and a holomorphic

function ℎ on 𝑈 such that 𝑓 = ℎ on 𝑉 ′ ∩ 𝑈. The holomorphic functions on open subsets
of 𝑉 define on 𝑉 the structure of a ℂ-ringed space.

Definition 7.44. An analytic space is a ℂ-ringed space (𝑉, 𝒪𝑉 ) satisfying the follow-

⋃
(a) there exists an open covering 𝑉 = 𝑉𝑖 of 𝑉 such that, for each 𝑖, the ℂ-ringed
ing two conditions:

space (𝑉𝑖 , 𝒪𝑉 |𝑉𝑖 ) is isomorphic to an analytic set equipped with its sheaf of analytic

(b) the topological space 𝑉 is Hausdorff.

1 functions;

Let 𝑉 be an algebraic variety over ℂ. Then 𝑉(ℂ) has a natural structure of a complex
analytic space, and so there is a canonical functor 𝑉 ⇝ 𝑉 an from algebraic varieties over
ℂ to complex analytic spaces (GAGA, §2).
4
Chow, Wei-Liang. On the projective embedding of homogeneous varieties. Algebraic geometry and
topology. A symposium in honor of S. Lefschetz, pp. 122–128. Princeton University Press, Princeton, N. J.,
1957.
174 7. Complete Varieties

module ℱ on a algebraic variety 𝑉 over ℂ defines a coherent module ℱ an = ℱ ⊗𝒪𝐵 𝒪𝑉 an

We refer the reader to Chapter 13 for the notion of a coherent module. Every coherent

on 𝑉 an .
def

Theorem 7.45. Let 𝑉 be a projective variety over ℂ. The functor ℱ ⇝ ℱ an is an equiva-

lence from the category of coherent 𝒪𝑉 -modules to the category of coherent 𝒪𝑉 an -modules,
under which locally free modules correspond to locally free modules. Moreover,

𝛤(𝑉 an , 𝒪𝑉 an ) ≃ 𝛤(𝑉, 𝒪𝑉 ).

Proof. This summarizes the main results of GAGA (Théorémes 2,3, p. 19, p. 20). 2

Theorem 7.46 (Chow’s Theorem). Every closed analytic subset of a projective variety
is algebraic.

Proof. Let 𝑉 be a projective space, and let 𝑍 be a closed analytic subset of 𝑉 an . A

theorem of Henri Cartan states that 𝒪𝑍 an is a coherent analytic sheaf on 𝑉 an , and so
there exists a coherent algebraic sheaf ℱ on 𝑉 such that ℱ an = 𝒪𝑍 an . The support of ℱ
is Zariski closed, and equals 𝑍 (GAGA, p. 29). 2

In particular, projective analytic spaces are projective algebraic varieties.

Theorem 7.47. Every compact analytic subset of an algebraic variety is algebraic.

Proof. Let 𝑉 be an algebraic variety, and let 𝑍 be a compact analytic subset of 𝑉 an .

By Chow’s lemma (7.42), there exists a projective variety 𝑉 ′ , a dense open subset 𝑈 of
𝑉 ′ , and a surjective regular map 𝜑 ∶ 𝑈 → 𝑉 whose graph 𝛤 is closed in 𝑉 × 𝑉 ′ . Let
𝛤 ′ = 𝛤 ∩ (𝑍 × 𝑉 ′ ). As 𝑍 and 𝑉 ′ are compact and 𝛤 is closed, 𝛤 ′ is compact, and so its
projection 𝑉 ′′ on 𝑉 ′ is also compact. On the other hand, 𝑉 ′′ = 𝑓 −1 (𝑍), which shows that
it is an analytic subset of 𝑈, and therefore also of 𝑉 ′ . According to Chow’s theorem, it is a
Zariski closed subset of 𝑉 ′ (hence an algebraic variety). Now 𝑍 = 𝑓(𝑉 ′′ ) is constructible
(Zariski sense; see 9.7 below), and therefore its Zariski closure coincides with its closure
for the complex topology, but (by assumption) it is closed.

Corollary 7.48. Let 𝑉 and 𝑊 be algebraic varieties over ℂ. If 𝑉 is complete, then every
2

analytic map 𝑓 ∶ 𝑉 an → 𝑊 an is algebraic.

Proof. Apply Theorem 7.47 to the graph of 𝑓.

Example 7.49. The graph of 𝑧 ↦ 𝑒𝑧 ∶ ℂ → ℂ is closed in ℂ × ℂ but it is not Zariski

2

closed.

h. Nagata’s Embedding Theorem

A necessary condition for a prevariety to be an open subvariety of a complete variety is
that it be separated. An important theorem of Nagata says that this condition is also

Theorem 7.50. Every variety 𝑉 admits an open immersion 𝑉 → 𝑊 into a complete

sufficient.

variety 𝑊.

If 𝑉 is affine, then one can embed 𝑉 → 𝔸𝑛 → ℙ𝑛 , and take 𝑊 to be the closure of

𝑉 in ℙ𝑛 . The proof in the general case is quite difficult. See:
 h. Nagata’s Embedding Theorem 175

Nagata, Masayoshi. Imbedding of an abstract variety in a complete variety. J.

Math. Kyoto Univ. 2 1962 1–10; A generalization of the imbedding problem
of an abstract variety in a complete variety. J. Math. Kyoto Univ. 3 1963
89–102.
For a modern exposition, see:
Lütkebohmert, W. On compactification of schemes. Manuscripta Math. 80
(1993), no. 1, 95–111.
In the 1970s, Deligne translated Nagata’s work into the language of schemes. His personal
notes are available in three versions.
Deligne, P., Le théorème de plongement de Nagata, Kyoto J. Math. 50,
Number 4 (2010), 661-670.
Conrad, B., Deligne’s notes on Nagata compactifications. J. Ramanujan
Math. Soc. 22 (2007), no. 3, 205–257.
Vojta, P., Nagata’s embedding theorem, 19pp., 2007, arXiv:0706.1907.
See also:
Temkin, Michael. Relative Riemann-Zariski spaces. Israel J. Math. 185
(2011), 1–42.

A little history
As noted earlier (p. 129), initially Weil was unable to construct the Jacobian variety of
a curve as projective variety, which led him to introduce “abstract varieties” and also
the notion of a complete abstract variety. Later he (and others) showed that Jacobian
varieties are in fact projective.

∑
Exercises
7-1. Identify the set of homogeneous polynomials 𝐹(𝑋, 𝑌) = 𝑎𝑖𝑗 𝑋 𝑖 𝑌 𝑗 , 0 ≤ 𝑖, 𝑗 ≤ 𝑚,
with an affine space. Show that the subset of reducible polynomials is closed.

7-2. Let 𝑉 and 𝑊 be complete irreducible varieties, and let 𝐴 be an abelian variety. Let
𝑃 and 𝑄 be points of 𝑉 and 𝑊. Show that any regular map ℎ ∶ 𝑉 × 𝑊 → 𝐴 such that
ℎ(𝑃, 𝑄) = 0 can be written ℎ = 𝑓◦𝑝 + 𝑔◦𝑞 where 𝑓 ∶ 𝑉 → 𝐴 and 𝑔 ∶ 𝑊 → 𝐴 are regular
maps carrying 𝑃 and 𝑄 to 0 and 𝑝 and 𝑞 are the projections 𝑉 × 𝑊 → 𝑉, 𝑊.
Chapter 8

Normal Varieties; (Quasi-)finite

maps; Zariski’s Main Theorem

We begin by studying normal varieties. Nonsingular varieties are normal, and normal
varieties have some of the good properties of nonsingular varieties, but it is easy to show
that every variety is birationally equivalent to a normal variety. After studying finite and
quasi-finite maps, we discuss the celebrated Zariski’s Main Theorem (ZMT), which says
that every quasi-finite map of algebraic varieties can be obtained from a finite map by
removing a closed subset from the source variety. In its original form, the theorem says

only at points where the fibre has dimension > 0.

that a birational regular map to a normal algebraic variety fails to be a local isomorphism

a. Normal varieties

closed in its field of fractions. Moreover (1.49), that an integral domain 𝐴 is integrally
Recall (1.42) that an integrally closed domain is an integral domain that is integrally

closed if and only if 𝐴𝔪 is integrally closed for every maximal ideal 𝔪 in 𝐴.

Definition 8.1. A point 𝑃 on an algebraic variety 𝑉 is normal if 𝒪𝑉,𝑃 is an integrally
closed domain. An algebraic variety is said to be normal if all of its points are normal.

Since the local ring at a point lying on two irreducible components cannot be an integral
domain (3.14), a normal variety is a disjoint union of its irreducible components, which

Proposition 8.2. The following conditions on an irreducible variety 𝑉 are equivalent.

are therefore its connected components.

(a) The variety 𝑉 is normal.

(b) For all open affine subsets 𝑈 of 𝑉, the ring 𝒪𝑉 (𝑈) is an integrally closed domain.
(c) For all open subsets 𝑈 of 𝑉, a rational function on 𝑉 that satisfies a monic polynomial
equation on 𝑈 whose coefficients are regular on 𝑈 is itself regular on 𝑈.

(a) ⇐⇒ (c). Let 𝑈 be an open subset of 𝑉, and let 𝑓 ∈ 𝑘(𝑉) satisfy

Proof. The equivalence of (a) and (b) follows from 1.49.

𝑓 𝑛 + 𝑎1 𝑓 𝑛−1 + ⋯ + 𝑎𝑛 = 0, 𝑎𝑖 ∈ 𝒪𝑉 (𝑈),

(equality in 𝑘(𝑉)). Then 𝑎𝑖 ∈ 𝒪𝑉 (𝑈) ⊂ 𝒪𝑃 for all 𝑃 ∈ 𝑈, and so 𝑓 ∈ 𝒪𝑃 for all 𝑃 ∈ 𝑈.

This implies that 𝑓 ∈ 𝒪𝑉 (𝑈) (5.10).

176
 a. Normal varieties 177

(c) ⇐⇒ (b). The condition applied to an open affine subset 𝑈 of 𝑉 implies that
𝒪𝑉 (𝑈) is integrally closed in 𝑘(𝑉). 2

Regular local rings are unique factorization domains (Matsumura 1989, Theorem
20.3), hence normal (1.43). Conversely, a normal local domain of dimension one is regular.
Thus nonsingular varieties are normal, and normal curves are nonsingular. However, a

𝑋2 + 𝑌2 − 𝑍2 = 0
normal surface need not be nonsingular: the cone

is normal, but it is singular at the origin — the tangent space at the origin is 𝑘 3 .
The singular locus of a normal variety 𝑉 must have dimension ≤ dim 𝑉 − 2 (see

singular locus cannot contain a curve. In particular, the surface 𝑍 3 = 𝑋 2 𝑌 (see 4.42) is
8.12 below). For example, a normal surface can only have isolated singularities — the

not normal.

The normalization of an algebraic variety

Let 𝐸 ⊃ 𝐹 be a finite extension of fields. The extension 𝐸∕𝐹 is said to be normal if the
minimal polynomial of every element of 𝐸 splits in 𝐸. Let 𝐹 al be an algebraic closure
of 𝐹 containing 𝐸. The composite in 𝐹 al of the fields 𝜎𝐸, 𝜎 ∈ Aut(𝐸∕𝐹), is normal
over 𝐹 (and is called the normal closure of 𝐹 in 𝐹 al ). If 𝐸 is normal over 𝐹, then 𝐸
is Galois over 𝐸 Aut(𝐸∕𝐹) (FT, 3.10), and 𝐸 Aut(𝐸∕𝐹) is purely inseparable over 𝐹 (because
Hom𝐹 (𝐸 Aut(𝐸∕𝐹) , 𝐹 al ) consists of a single element).
Proposition 8.3. Let 𝐴 be a finitely generated 𝑘-algebra. Assume that 𝐴 is an integral
domain, and let 𝐸 be a finite field extension of its field of fractions 𝐹. Then the integral
closure 𝐴′ of 𝐴 in 𝐸 is a finite 𝐴-algebra (hence a finitely generated 𝑘-algebra).

Proof. According to the Noether normalization theorem (2.45), 𝐴 contains a polyno-

mial subalgebra 𝐴0 and is finite over 𝐴0 . Now 𝐸 is a finite extension of 𝐹(𝐴0 ) and 𝐴′
is the integral closure of 𝐴0 in 𝐸, and so we only need to consider the case that 𝐴 is a
polynomial ring 𝑘[𝑋1 , … , 𝑋𝑑 ].
Let 𝐸̃ denote the normal closure of 𝐸 in some algebraic closure of 𝐹 containing 𝐸, and
let 𝐴̃ denote the integral closure of 𝐴 in 𝐸.
̃ If 𝐴̃ is finitely generated as an 𝐴-module, then
so is its submodule 𝐴 (because 𝐴 is noetherian). Therefore we only need to consider
′

the case that 𝐸 is normal over 𝐹.

According to the above discussion, 𝐸 ⊃ 𝐸1 ⊃ 𝐹 with 𝐸 Galois over 𝐸1 and 𝐸1 purely
inseparable over 𝐹. Let 𝐴1 denote the integral closure of 𝐴 in 𝐸1 . Then 𝐴′ is a finite
𝐴1 -algebra (1.51), and so it suffices to show that 𝐴1 is a finite 𝐴-algebra. Therefore we
only need to consider the case that 𝐸 is purely inseparable over 𝐹.
In this case, 𝑘 has characteristic 𝑝 ≠ 0, and, for each 𝑥 ∈ 𝐸, there is a power 𝑞(𝑥) of
𝑝 such that 𝑥𝑞(𝑥) ∈ 𝐹. As 𝐸 is finitely generated over 𝐹, there is a single power 𝑞 of 𝑝
such that 𝑥𝑞 ∈ 𝐹 for all 𝑥 ∈ 𝐸. Let 𝐹 al denote an algebraic closure of 𝐹 containing 𝐸.
For each 𝑖, there is a unique 𝑌𝑖 ∈ 𝐹 al such that 𝑌𝑖 = 𝑋𝑖 . Now
𝑞

𝐹 = 𝑘(𝑋1 , … , 𝑋𝑑 ) ⊂ 𝐸 ⊂ 𝑘(𝑌1 , … , 𝑌𝑑 )

𝐴 = 𝑘[𝑋1 , … , 𝑋𝑑 ] ⊂ 𝐴′ ⊂ 𝑘[𝑌1 , … , 𝑌𝑑 ]
and

because 𝑘[𝑌1 , … , 𝑌𝑑 ] contains 𝐴 and is integrally closed (1.32, 1.43). Obviously 𝑘[𝑌1 , … , 𝑌𝑑 ]
is a finite 𝐴-algebra, and this implies, as before, that 𝐴′ is a finite 𝐴-algebra. 2
178 8. Normal Varieties; (Quasi-)finite maps; Zariski’s Main Theorem

Corollary 8.4. Let 𝐴 be as in 8.3. If 𝐴𝔪 is normal for some maximal ideal 𝔪 in 𝐴, then
𝐴ℎ is normal for some ℎ ∈ 𝐴 ∖ 𝔪.
Proof. Let 𝐴′ be the integral closure of 𝐴 in its field of fractions. Then 𝐴′ = 𝐴[𝑓1 , … , 𝑓𝑚 ]
( ) 1.47
for some 𝑓𝑖 ∈ 𝐴′ . Now 𝐴′ 𝔪 = (𝐴𝔪 ) = 𝐴𝔪 , and so there exists an ℎ ∈ 𝐴 ∖ 𝔪 such
′

that, for all 𝑖, ℎ𝑓𝑖 ∈ 𝐴. Now 𝐴ℎ′ = 𝐴ℎ , and so 𝐴ℎ is normal.

The proposition shows that if 𝐴 is an integral domain finitely generated over 𝑘, then
2

the integral closure 𝐴′ of 𝐴 in a finite extension 𝐸 of 𝐹(𝐴) has the same properties.
Therefore, Spm(𝐴′ ) is an irreducible algebraic variety, called the normalization of
Spm(𝐴) in 𝐸. This construction extends without difficulty to nonaffine varieties.
Proposition 8.5. Let 𝑉 be an irreducible algebraic variety, and let 𝐾 be a finite field
extension of 𝑘(𝑉). Then there exists an irreducible algebraic variety 𝑊 with 𝑘(𝑊) = 𝐾
and a regular map 𝜑 ∶ 𝑊 → 𝑉 such that, for all open affines 𝑈 in 𝑉, 𝜑−1 (𝑈) is affine and
𝑘[𝜑−1 (𝑈)] is the integral closure of 𝑘[𝑈] in 𝐾.
The map 𝜑 (or just 𝑊) is called the normalization of 𝑉 in 𝐾.
Proof. For each 𝑣 ∈ 𝑉, let 𝑊(𝑣) be the set of maximal ideals in the integral closure
⨆
of 𝒪𝑣 in 𝐾. Let 𝑊 = 𝑣∈𝑉 𝑊(𝑣), and let 𝜑 ∶ 𝑊 → 𝑉 be the map sending the points of
𝑊(𝑣) to 𝑣. For an open affine subset 𝑈 of 𝑉,
𝜑−1 (𝑈) ≃ spm(𝑘[𝑈]′ ),
where 𝑘[𝑈]′ is the integral closure of 𝑘[𝑈] in 𝐾. We endow 𝑊 with the 𝑘-ringed space

(𝜑−1 (𝑈), 𝒪𝑊 |𝜑−1 (𝑈)) ≃ Spm(𝑘[𝑈]′ ).

structure for which

A routine argument shows that (𝑊, 𝒪𝑊 ) is an algebraic variety with the required prop-

Example 8.6. (a) The normalization of the cuspidal cubic 𝑉 ∶ 𝑌 2 = 𝑋 3 in 𝑘(𝑉) is the
erties. 2

map 𝔸1 → 𝑉, 𝑡 ↦ (𝑡 2 , 𝑡3 ) (see 3.29).

(b) The normalization of the nodal cubic 𝑉 ∶ 𝑌 2 = 𝑋 3 + 𝑋 2 (4.10) in 𝑘(𝑉) is the map
𝔸 → 𝑉, 𝑡 ↦ (𝑡2 − 1, 𝑡3 − 𝑡).
1

Proposition 8.7. The normal points in an irreducible algebraic variety form a dense open
subset.
Proof. Corollary 8.4 shows that the set of normal points is open, and so it remains to
show that it is nonempty. This follows from 4.37 and the fact (difficult to prove) that

Let 𝑉 be an irreducible algebraic variety. According to (3.37, 3.38), 𝑉 is birationally

nonsingular points are normal, but we shall give a direct proof.

equivalent to a hypersurface 𝐻 in 𝔸𝑑+1 , 𝑑 = dim 𝑉,

𝐻∶ 𝑎0 𝑋 𝑚 + 𝑎1 𝑋 𝑚−1 + ⋯ + 𝑎𝑚 , 𝑎𝑖 ∈ 𝑘[𝑇1 , … , 𝑇𝑑 ], 𝑎0 ≠ 0, 𝑚 ∈ ℕ;
moreover, 𝑇1 , … , 𝑇𝑑 can be chosen to be a separating transcendence basis for 𝑘(𝑉) (=
𝑘(𝐻)) over 𝑘. Therefore the discriminant 𝐷 of 𝑘(𝐻) over 𝑘(𝑇1 , … , 𝑇𝑑 ) (an element of
𝑘[𝐻]) is nonzero.1 We shall see that open subset of 𝐻 where 𝐷 is nonzero is normal.
Let 𝐵 ⊃ 𝐴 be rings. Assume that 𝐵 is free of rank 𝑚 as an 𝐴-module, and let 𝛽1 , ..., 𝛽𝑚 be a basis for 𝐵
as an 𝐴-module. We call 𝐷(𝛽1 , ..., 𝛽𝑚 ) = det(Tr𝐵∕𝐴 (𝛽𝑖 𝛽𝑗 )) the discriminant of 𝐵∕𝐴 (it is well-defined up
1

to a unit in 𝐴). The discriminant of a finite separable extension of fields is nonzero (Proposition 2.26 of my
def

notes on Algebraic Number Theory).

 b. Regular functions on normal varieties 179

Let 𝐴 = 𝑘[𝑇1 , … , 𝑇𝑑 ]; then 𝑘[𝐻] = 𝐴[𝑋]∕(𝑎0 𝑋 𝑚 + ⋯ + 𝑎𝑚 ) = 𝐴[𝑥]. Let

𝑦 = 𝑐0 + ⋯ + 𝑐𝑚−1 𝑥 𝑚−1 , 𝑐𝑖 ∈ 𝑘(𝑇1 , … , 𝑇𝑑 ),
be an element of 𝑘(𝐻) integral over 𝐴. For each 𝑗 ∈ ℕ, Tr𝑘(𝐻)∕𝐹(𝐴) (𝑦𝑥𝑗 ) is a sum of
(35)

conjugates of 𝑦𝑥 𝑗 , and hence is integral over 𝐴 (cf. the proof of 1.44). As it lies in 𝐹(𝐴),
it is an element of 𝐴. On multiplying (35) with 𝑥 𝑗 and taking traces, we get a system of

𝑐0 ⋅ Tr(𝑥𝑗 ) + 𝑐1 ⋅ Tr(𝑥1+𝑗 ) + ⋯ + 𝑐𝑚−1 ⋅ Tr(𝑥𝑚−1+𝑗 ) = Tr(𝑦𝑥 𝑗 ), 𝑗 = 0, … , 𝑚 − 1.

linear equations

det(Tr(𝑥𝑖+𝑗 )) ⋅ 𝑐𝑙 ∈ 𝐴, 𝑙 = 0, … , 𝑚 − 1.
By Cramer’s rule (p. 25),

But det(Tr(𝑥𝑖+𝑗 )) = 𝐷, and so 𝑐𝑙 ∈ 𝐴[𝐷 −1 ]. Hence 𝑘[𝐻] becomes normal once we invert
the nonzero element 𝐷. We have shown that 𝐻 contains a dense open normal subvariety,
which implies that 𝑉 does also.
Proposition 8.8. For every irreducible algebraic variety 𝑉, there exists a surjective regular
2

map 𝜑 ∶ 𝑉 ′ → 𝑉 from a normal algebraic variety 𝑉 ′ to 𝑉 such that, for some dense open
subset 𝑈 of 𝑉, 𝜑 induces an isomorphism 𝜑−1 (𝑈) → 𝑈 (in particular 𝜑 is birational).
Proof. Proposition 8.7 shows that the normalization of 𝑉 in 𝑘(𝑉) has this property. 2
8.9. More generally, for a dominant map 𝜑 ∶ 𝑊 → 𝑉 of irreducible algebraic varieties,
there exists a normalization of 𝑉 in 𝑊. For each open affine 𝑈 in 𝑉 we have
𝑘[𝑈] ⊂ 𝛤(𝜑−1 (𝑈), 𝒪𝑊 ) ⊂ 𝑘(𝑊).
The integral closure 𝑘[𝑈]′ of 𝛤(𝑈, 𝒪𝑉 ) in 𝛤(𝜑−1 (𝑈), 𝒪𝑊 ) is a finite 𝑘[𝑈]-algebra (be-
cause it is a 𝑘[𝑈]-submodule of the integral closure of 𝑘[𝑈] in 𝑘(𝑊)). The normalization
of 𝑉 in 𝑊 is a regular map 𝜑′ ∶ 𝑉 ′ → 𝑉 such that, for every open affine 𝑈 in 𝑉,
(𝜑′−1 (𝑈), 𝒪𝑉 ′ ) = Spm(𝑘[𝑈]′ ).
In particular, 𝜑′ is an affine map. For example, if 𝑊 and 𝑉 are affine, then 𝑉 ′ =
Spm(𝑘[𝑉]′ ), where 𝑘[𝑉]′ is the integral closure of 𝑘[𝑉] in 𝑘[𝑊]. There is a commutative

𝑊 → 𝑉′
𝑗
triangle

←
←

𝜑 → 𝜑′
←

𝑉.
→

b. Regular functions on normal varieties

Definition 8.10. An algebraic variety 𝑉 is factorial at a point 𝑃 if 𝒪𝑃 is a factorial
domain. The variety 𝑉 is factorial if it is factorial at all points 𝑃.
When 𝑉 is factorial, it does not follow that 𝒪𝑉 (𝑈) is factorial for all open affines 𝑈 in 𝑉.
A prime divisor 𝑍 on a variety 𝑉 is a closed irreducible subvariety of codimension
1. Let 𝑍 be a prime divisor on 𝑉, and let 𝑃 ∈ 𝑉; we say that 𝑍 is locally principal
at 𝑃 if there exists an open affine neighbourhood 𝑈 of 𝑃 and an 𝑓 ∈ 𝑘[𝑈] such that
𝐼(𝑍 ∩ 𝑈) = (𝑓); the regular function 𝑓 is then called a local equation for 𝑍 at 𝑃. If
𝑃 ∉ 𝑍, then 𝑍 is locally principal at 𝑃 because then we can choose 𝑈 so that 𝑍 ∩ 𝑈 = ∅,
and 𝐼(𝑍 ∩ 𝑈) = (1).
180 8. Normal Varieties; (Quasi-)finite maps; Zariski’s Main Theorem

Proposition 8.11. An irreducible variety 𝑉 is factorial at a point 𝑃 if and only if every

prime divisor on 𝑉 is locally principal at 𝑃.

only if every prime ideal of height 1 is principal (1.24, 3.53).

Proof. Recall that an integral domain is factorial (finitely generated over a field) if and
2

Proposition 8.12. The codimension of the singular locus in a normal variety is at least 2.

Proof. Let 𝑉 be a normal algebraic variety of dimension 𝑑, and suppose that its singular
locus has an irreducible component 𝑊 of codimension 1. After replacing 𝑉 with an
open subvariety, we may suppose that it is affine and that 𝑊 is principal, say, 𝑊 = (𝑓)
(see 8.11). There exists a nonsingular point 𝑃 on 𝑊 (4.37). Let (𝑈, 𝑓1 ), … , (𝑈, 𝑓𝑑−1 ) be
germs of functions at 𝑃 (on 𝑉) whose restrictions to 𝑊 generate the maximal ideal in
𝒪𝑊,𝑃 (cf. 4.36). Then (𝑈, 𝑓1 ), … , (𝑈, 𝑓𝑑−1 ), (𝑈, 𝑓) generate the maximal ideal in 𝒪𝑉,𝑃 ,
and so 𝑃 is nonsingular on 𝑉. This contradicts the definition of 𝑊. 2

Summary 8.13. For an algebraic variety 𝑉,

nonsingular ⇐⇒ factorial ⇐⇒ normal ⇐⇒ singular locus has codimension ≥ 2.

⋄ The variety 𝑋12 + ⋯ + 𝑋52 is factorial but singular.

⋄ The cone 𝑍 2 = 𝑋𝑌 in 𝔸3 is normal but not factorial (see 9.39 below).
⋄ The variety Spm(𝑘[𝑋, 𝑋𝑌, 𝑌 2 , 𝑌 3 ]) is a surface in 𝔸4 with exactly one singular
point, namely, the origin. Its singular locus has codimension 2, but the variety is
not normal (the normalization 𝑘[𝑋, 𝑋𝑌, 𝑌 2 , 𝑌 3 ] is 𝑘[𝑋, 𝑌]).
⋄ Every singular curve has singular locus of codimension 1 (hence fails all condi-
tions).

Let 𝑉 be a normal irreducible variety. A divisor on 𝑉 is an element of the free abelian

Zeros and poles of rational functions on normal varieties

group Div(𝑉) generated by the prime divisors. Thus a divisor 𝐷 can be written uniquely

∑
as a finite (formal) sum

𝐷= 𝑛𝑖 𝑍𝑖 , 𝑛𝑖 ∈ ℤ, 𝑍𝑖 a prime divisor on 𝑉.

The support |𝐷| of 𝐷 is the union of the 𝑍𝑖 corresponding to nonzero 𝑛𝑖 . A divisor is said
to be effective (or positive) if 𝑛𝑖 ≥ 0 for all 𝑖. We get a partial ordering on the divisors by
defining 𝐷 ≥ 𝐷 ′ to mean 𝐷 − 𝐷 ′ ≥ 0.
Because 𝑉 is normal, there is associated with every prime divisor 𝑍 on 𝑉 a discrete
valuation ring 𝒪𝑍 . This can be defined, for example, by choosing an open affine subvari-
ety 𝑈 of 𝑉 such that 𝑈 ∩ 𝑍 ≠ ∅; then 𝑈 ∩ 𝑍 is a maximal proper closed subset of 𝑈, and
so the ideal 𝔭 corresponding to it is minimal among the nonzero ideals of 𝑅 = 𝛤(𝑈, 𝒪);
so 𝑅𝔭 is an integrally closed domain with exactly one nonzero prime ideal 𝔭𝑅𝔭 — it is
therefore a discrete valuation ring (4.20), which is defined to be 𝒪𝑍 . More intrinsically
we can define 𝒪𝑍 to be the set of rational functions on 𝑉 that are defined an open subset
𝑈 of 𝑉 intersecting 𝑍.
Let ord𝑍 be the valuation 𝑘(𝑉)× ,→ ℤ with valuation ring 𝒪𝑍 ; thus, if 𝜋 is a prime
element of 𝒪𝑍 , then
onto

𝑎 = unit × 𝜋ord𝑍 (𝑎) .

 c. Finite and quasi-finite maps 181

The divisor of a nonzero element 𝑓 of 𝑘(𝑉) is defined to be

∑
div(𝑓) = ord𝑍 (𝑓) ⋅ 𝑍.

The sum is over all the prime divisors of 𝑉, but in fact ord𝑍 (𝑓) = 0 for all but finitely
many 𝑍. In proving this, we can assume that 𝑉 is affine (because it is a finite union of
affines), say, 𝑉 = Spm(𝑅). Then 𝑘(𝑉) is the field of fractions of 𝑅, and so we can write
𝑓 = 𝑔∕ℎ with 𝑔, ℎ ∈ 𝑅, and div(𝑓) = div(𝑔) − div(ℎ). Therefore, we can assume 𝑓 ∈ 𝑅.
⋃
The zero set of 𝑓, 𝑉(𝑓) either is empty or is a finite union of prime divisors, 𝑉 = 𝑍𝑖
(see 3.42) and ord𝑍 (𝑓) = 0 unless 𝑍 is one of the 𝑍𝑖 .

𝑓 ↦ div(𝑓) ∶ 𝑘(𝑉)× → Div(𝑉)

The map

is a homomorphism. A divisor of the form div(𝑓) is said to be principal, and two divisors
are said to be linearly equivalent, denoted 𝐷 ∼ 𝐷 ′ , if they differ by a principal divisor.
When 𝑉 is nonsingular, the Picard group Pic(𝑉) of 𝑉 is defined to be the group of
divisors on 𝑉 modulo principal divisors. (The definition of the Picard group of a general
algebraic variety agrees with this definition only for nonsingular varieties; it may differ

Theorem 8.14. Let 𝑉 be a normal variety, and let 𝑓 be rational function on 𝑉. If 𝑓 has
for normal varieties.)

no zeros or poles on an open subset 𝑈 of 𝑉, then 𝑓 is regular on 𝑈.

Proof. We may assume that 𝑉 is connected, hence irreducible, and apply the following
statement (see Chapter 12): if a noetherian integral domain 𝐴 is normal, then 𝐴 =
⋂
𝐴 (intersection in the field of fractions of 𝐴).
ht(𝔭)=1 𝔭 2

codimension ≥ 2, is regular everywhere.

Corollary 8.15. A rational function on a normal variety, regular outside a subset of

Proof. This is a restatement of the theorem. 2

Corollary 8.16. Let 𝑉 and 𝑊 be affine varieties with 𝑉 normal, and let 𝜑 ∶ 𝑉 ∖ 𝑍 → 𝑊
be a regular map defined on the complement of a closed subset 𝑍 of 𝑉. If codim(𝑍) ≥ 2,
then 𝜑 extends to a regular map on the whole of 𝑉.

Proof. We may suppose that 𝑊 is affine, and embed it as a closed subvariety of 𝔸𝑛 .

The map 𝑉 ∖ 𝑍 → 𝑊 → 𝔸𝑛 is given by 𝑛 regular functions on 𝑉 ∖ 𝑍, each of which
extends to 𝑉. Therefore 𝑉 ∖ 𝑍 → 𝔸𝑛 extends to 𝔸𝑛 , and its image is contained in 𝑊. 2

c. Finite and quasi-finite maps

Finite maps
Definition 8.17. A regular map 𝜑 ∶ 𝑊 → 𝑉 of algebraic varieties is finite if there exists
⋃
a finite covering 𝑉 = 𝑖 𝑈𝑖 of 𝑉 by open affines such that, for each 𝑖, the set 𝜑−1 (𝑈𝑖 ) is
affine and 𝑘[𝜑−1 (𝑈𝑖 )] is a finite 𝑘[𝑈𝑖 ]-algebra.

Example 8.18. Let 𝑉 be an irreducible algebraic variety. The normalization 𝜑 ∶ 𝑊 → 𝑉

of 𝑉 in a finite extension of 𝑘(𝑉) is finite. This follows from the definition 8.5 and
Proposition 8.3.
182 8. Normal Varieties; (Quasi-)finite maps; Zariski’s Main Theorem

The next lemma shows that, for maps of affine algebraic varieties, the above definition

Lemma 8.19. A regular map 𝜑 ∶ 𝑊 → 𝑉 of affine algebraic varieties is finite if and only
agrees with Definition 2.39.

if 𝑘[𝑊] is a finite 𝑘[𝑉]-algebra.

assume in the proof that 𝑉 and 𝑊 are irreducible. Let (𝑈𝑖 )𝑖 be a finite family of open
Proof. The necessity being obvious, we prove the sufficiency. For simplicity, we shall

affines covering 𝑉 and such that, for each 𝑖, the set 𝜑−1 (𝑈𝑖 ) is affine and 𝑘[𝜑−1 (𝑈𝑖 )] is a
finite 𝑘[𝑈𝑖 ]-algebra.
Each 𝑈𝑖 is a finite union of basic open subsets of 𝑉. These are also basic open subsets
of 𝑈𝑖 , because 𝐷(𝑓) ∩ 𝑈𝑖 = 𝐷(𝑓|𝑈𝑖 ), and so we may assume that the original 𝑈𝑖 are
basic open subsets of 𝑉, say, 𝑈𝑖 = 𝐷(𝑓𝑖 ) with 𝑓𝑖 ∈ 𝐴.
Let 𝐴 = 𝑘[𝑉] and 𝐵 = 𝑘[𝑊]. We are given that (𝑓1 , … , 𝑓𝑛 ) = 𝐴 and that 𝐵𝑓𝑖 is a
finite 𝐴𝑓𝑖 -algebra for each 𝑖. We have to show that 𝐵 is a finite 𝐴-algebra.
Let {𝑏𝑖1 , … , 𝑏𝑖𝑚𝑖 } generate 𝐵𝑓𝑖 as an 𝐴𝑓𝑖 -module. After multiplying through by a
power of 𝑓𝑖 , we may assume that the 𝑏𝑖𝑗 lie in 𝐵. We shall show that the family of all 𝑏𝑖𝑗
generate 𝐵 as an 𝐴-module. Let 𝑏 ∈ 𝐵. Then 𝑏∕1 ∈ 𝐵𝑓𝑖 , so

𝑎𝑖1 𝑎𝑖𝑚𝑖
𝑏= 𝑏𝑖1 + ⋯ + 𝑏𝑖𝑚𝑖 , some 𝑎𝑖𝑗 ∈ 𝐴 and 𝑟𝑖 ∈ ℕ.
𝑓𝑖 𝑖 𝑓𝑖 𝑖
𝑟 𝑟

The ideal (𝑓11 , … , 𝑓𝑛𝑛 ) = 𝐴 because any maximal ideal containing (𝑓11 , … , 𝑓𝑛𝑛 ) would
𝑟 𝑟 𝑟 𝑟

have to contain (𝑓1 , … , 𝑓𝑛 ) = 𝐴. Therefore,

1 = ℎ1 𝑓11 + ⋯ + ℎ𝑛 𝑓𝑛𝑛 , some ℎ𝑖 ∈ 𝐴.

𝑟 𝑟

Now

𝑏 = 𝑏 ⋅ 1 = ℎ1 ⋅ 𝑏𝑓11 + ⋯ + ℎ𝑛 ⋅ 𝑏𝑓𝑛𝑛
𝑟 𝑟

= ℎ1 (𝑎11 𝑏11 + ⋯ + 𝑎1𝑚1 𝑏1𝑚1 ) + ⋯ + ℎ𝑛 (𝑎𝑛1 𝑏𝑛1 + ⋯ + 𝑎𝑛𝑚𝑛 𝑏𝑛𝑚𝑛 ),

as required. 2

Lemma 8.20. Let 𝜑 ∶ 𝑊 → 𝑉 be a regular map with 𝑉 affine, and let 𝑈 be an open affine
in 𝑉. Then
𝛤(𝑊, 𝒪𝑊 ) ⊗𝑘[𝑉] 𝑘[𝑈] ≃ 𝛤(𝜑−1 (𝑈), 𝒪𝑊 ).

Proof. Let 𝑈 ′ = 𝜑−1 (𝑈), so we have to prove

𝛤(𝑊, 𝒪𝑊 ) ⊗𝑘[𝑉] 𝑘[𝑈] ≃ 𝛤(𝑈 ′ , 𝒪𝑊 )

The map is defined by the 𝑘[𝑉]-bilinear pairing

(𝑓, 𝑔) ↦ 𝑓|𝑈 ′ ⋅ 𝑔◦𝜑|𝑈 ′ ∶ 𝛤(𝑊, 𝒪𝑊 ) × 𝑘[𝑈] → 𝛤(𝑈 ′ , 𝒪𝑊 ).

When 𝑊 is affine, the statement is proved in 5.32.

Let 𝑆 = {𝑓 ∈ 𝑘[𝑉] ∣ ∀𝑃 ∈ 𝑈, 𝑓(𝑃) ≠ 0}. Then 𝑘[𝑈] = 𝑆 −1 𝑘[𝑉], and so, for any
𝑘[𝑉]-module 𝑀, 𝑀 ⊗𝑘[𝑉] 𝑘[𝑈] ≃ 𝑆 −1 𝑀. Hence the functor − ⊗𝑘[𝑉] 𝑘[𝑈] is exact.
 c. Finite and quasi-finite maps 183

⋃
Let 𝑊 = 𝑊𝑖 be a finite open affine covering of 𝑊, and consider the commutative

∏ ∏
diagram:

0 𝛤(𝑊, 𝒪𝑊 ) ⊗𝑘[𝑉] 𝑘[𝑈] 𝛤(𝑊𝑖 , 𝒪𝑊 ) ⊗𝑘[𝑉] 𝑘[𝑈] 𝛤(𝑊𝑖𝑗 , 𝒪𝑊 ) ⊗𝑘[𝑉] 𝑘[𝑈]

𝑖 𝑖,𝑗

∏ ∏
0 𝛤(𝑈 ′ , 𝒪𝑊 ) 𝛤(𝑈 ′ ∩ 𝑊𝑖 , 𝒪𝑊 ) 𝛤(𝑈 ′ ∩ 𝑊𝑖𝑗 , 𝒪𝑊 ).
𝑖 𝑖,𝑗

Here 𝑊𝑖𝑗 = 𝑊𝑖 ∩ 𝑊𝑗 . The rows are exact because 𝒪𝑊 is a sheaf. The varieties 𝑊𝑖 and
𝑊𝑖 ∩𝑊𝑗 are all affine, and so the two vertical arrows at right are products of isomorphisms.
This implies that the first is also an isomorphism. 2

Proposition 8.21. Let 𝜑 ∶ 𝑊 → 𝑉 be a regular map of algebraic varieties. If 𝜑 is finite,

then, for every open affine 𝑈 in 𝑉, 𝜑−1 (𝑈) is affine and 𝑘[𝜑−1 (𝑈)] is a finite 𝑘[𝑈]-algebra.

Proof. Let 𝑉𝑖 be an open affine covering of 𝑉 (which we may suppose to be finite) such
that 𝑊𝑖 = 𝜑−1 (𝑉𝑖 ) is an affine subvariety of 𝑊 for all 𝑖 and 𝑘[𝑊𝑖 ] is a finite over 𝑘[𝑉𝑖 ].
Let 𝑈 be an open affine in 𝑉, and let 𝑈 ′ = 𝜑−1 (𝑈). Then 𝛤(𝑈 ′ , 𝒪𝑊 ) is a subalgebra
def

∏
of 𝑖 𝛤(𝑈 ′ ∩ 𝑊𝑖 , 𝒪𝑊 ), and so it is an affine 𝑘-algebra finite over 𝑘[𝑈].2 We have a
morphism of varieties over 𝑉

𝑈′ Spm(𝛤(𝑈 ′ , 𝒪𝑊 ))
canonical

(36)

which we shall show to be an isomorphism. We know that each of the maps

𝑈 ′ ∩ 𝑊𝑖 → Spm(𝛤(𝑈 ′ ∩ 𝑊𝑖 , 𝒪𝑊 ))

is an isomorphism. But Spm(𝛤(𝑈 ′ ∩𝑊𝑖 , 𝒪𝑊 )) is the inverse image of 𝑉𝑖 in Spm(𝛤(𝑈 ′ , 𝒪𝑊 )).

Therefore the canonical morphism is an isomorphism over each 𝑉𝑖 , and so it is an iso-
morphism. 2

Summary 8.22. Let 𝜑 ∶ 𝑊 → 𝑉 be a regular map, and consider the following condition
on an open affine subset 𝑈 of 𝑉:
(*) 𝜑−1 (𝑈) is affine and 𝑘[𝜑−1 (𝑈)] is a finite over 𝑘[𝑈].
The map 𝜑 is finite if (*) holds for the open affines in some covering of 𝑉, in which case
(*) holds for all open affines of 𝑉.

Proposition 8.23. (a) Closed immersions are finite.

(b) The composite of two finite morphisms is finite.
(c) The product of two finite morphisms is finite.
2
Recall that a module over a noetherian ring is noetherian if and only if it is finitely generated, and
that a submodule of a noetherian module is noetherian. Therefore, a submodule of a finitely generated
module over a noetherian ring is finitely generated.
184 8. Normal Varieties; (Quasi-)finite maps; Zariski’s Main Theorem

Proof. (a) Let 𝑍 be a closed subvariety of a variety 𝑉, and let 𝑈 be an open affine
subvariety of 𝑉. Then 𝑍 ∩ 𝑈 is a closed subvariety of 𝑈. It is therefore affine, and the
map 𝑍 ∩ 𝑈 → 𝑈 corresponds to a map 𝐴 → 𝐴∕𝔞 of rings, which is obviously finite.
This proves (a). As to be finite is a local condition, it suffices to prove (a) and (b)
for maps of affine varieties. Then the statements become statements in commutative

(b) If 𝐵 is a finite 𝐴-algebra and 𝐶 is a finite 𝐵-algebra, then 𝐶 is a finite 𝐴-algebra.

algebra.

To see this, note that if {𝑏𝑖 } is a set of generators for 𝐵 as an 𝐴-module, and {𝑐𝑗 } is a set of
generators for 𝐶 as a 𝐵-module, then {𝑏𝑖 𝑐𝑗 } is a set of generators for 𝐶 as an 𝐴-module.
(c) If 𝐵 and 𝐵′ are respectively finite 𝐴 and 𝐴′ -algebras, then 𝐵 ⊗𝑘 𝐵′ is a finite
𝐴 ⊗𝑘 𝐴′ -algebra. To see this, note that if {𝑏𝑖 } is a set of generators for 𝐵 as an 𝐴-module,
and {𝑏𝑗′ } is a set of generators for 𝐵′ as an 𝐴′ -module, then {𝑏𝑖 ⊗ 𝑏𝑗′ } is a set of generators
for 𝐵 ⊗𝐴 𝐵′ as an 𝐴 ⊗ 𝐴′ -module. 2

𝔸1 ∖ {0} → 𝔸1 is not finite because the ring 𝑘[𝑇, 𝑇 −1 ] is not finitely generated as a
By way of contrast, open immersions are rarely finite. For example, the inclusion

𝑘[𝑇]-module.
Theorem 8.24. Finite maps of algebraic varieties are closed.

Proof. It suffices to prove this for affine varieties. Let 𝜑 ∶ 𝑊 → 𝑉 be a finite map of
affine varieties, and let 𝑍 be a closed subset of 𝑊. The restriction of 𝜑 to 𝑍 is finite (by
8.23a and b), and so we can replace 𝑊 with 𝑍; we then have to show that Im(𝜑) is closed.
The map corresponds to a finite map of rings 𝐴 → 𝐵. This will factors as 𝐴 → 𝐴∕𝔞 → 𝐵,

Spm(𝐵) → Spm(𝐴∕𝔞) → Spm(𝐴).

from which we obtain maps

The second map identifies Spm(𝐴∕𝔞) with the closed subvariety 𝑉(𝔞) of Spm(𝐴), and so
it remains to show that the first map is surjective. This is a consequence of the going-up
theorem (1.53). 2

The base change of a finite map

Recall that the base change of a regular map 𝜑 ∶ 𝑉 → 𝑆 is the map 𝜑′ in the diagram:

𝑉 ×𝑆 𝑊 → 𝑉
← 𝜓′

𝜑′ 𝜑
←
←

𝑊 → 𝑆.
𝜓
→
→

Proposition 8.25. The base change of a finite map is finite.

becomes: if 𝐴 is a finite 𝑅-algebra, then 𝐴⊗𝑅 𝐵∕𝔑 is a finite 𝐵-algebra, which is obvious.2
Proof. We may assume that all the varieties concerned are affine. Then the statement

Proposition 8.26. Finite maps of algebraic varieties are proper.

Proof. The base change of a finite map is finite, and hence closed.

Corollary 8.27. Let 𝜑 ∶ 𝑉 → 𝑆 be finite; if 𝑆 is complete, then so also is 𝑉.

2

Proof. Combine 7.19 and 8.26. 2

 c. Finite and quasi-finite maps 185

Quasi-finite maps
Recall that the fibres of a regular map 𝜑 ∶ 𝑊 → 𝑉 are the closed subvarieties 𝜑−1 (𝑃)
of 𝑊 for 𝑃 ∈ 𝑉. As for affine varieties (2.39), we say that a regular map of algebraic
varieties is quasi-finite if all of its fibres are finite.

Proposition 8.28. A finite map 𝜑 ∶ 𝑊 → 𝑉 is quasi-finite.

Proof. Let 𝑃 ∈ 𝑉; we wish to show that 𝜑−1 (𝑃) is finite. After replacing 𝑉 with an
affine neighbourhood of 𝑃, we may suppose that it is affine, and then 𝑊 will be affine
also. The map 𝜑 then corresponds to a map 𝛼 ∶ 𝐴 → 𝐵 of affine 𝑘-algebras, and a point
𝑄 of 𝑊 maps to 𝑃 if and only 𝛼−1 (𝔪𝑄 ) = 𝔪𝑃 . But this holds if and only if 𝔪𝑄 ⊃ 𝛼(𝔪𝑃 ),
and so the points of 𝑊 mapping to 𝑃 are in one-to-one correspondence with the maximal
ideals of 𝐵∕𝛼(𝔪𝑃 )𝐵. Clearly 𝐵∕𝛼(𝔪𝑃 )𝐵 is generated as a 𝑘-vector space by the image of
any generating set for 𝐵 as an 𝐴-module, and so it is a finite 𝑘-algebra. The next lemma
shows that it has only finitely many maximal ideals. 2

Lemma 8.29. A finite 𝑘-algebra 𝐴 has only finitely many maximal ideals.

Proof. Let 𝔪1 , … , 𝔪𝑛 be maximal ideals in 𝐴. They are coprime in pairs, and so

Theorem 1.1 shows that the map

𝐴 → 𝐴∕𝔪1 × ⋯ × 𝐴∕𝔪𝑛 , 𝑎 ↦ (… , 𝑎𝑖 mod 𝔪𝑖 , …),

∑
is surjective. It follows that

dim𝑘 𝐴 ≥ dim𝑘 (𝐴∕𝔪𝑖 ) ≥ 𝑛

— here dim𝑘 means dimension as a 𝑘-vector space. 2

Finite and quasi-finite maps of prevarieties are defined as for varieties.

Examples
8.30. The projection from the curve 𝑋𝑌 = 1 onto the 𝑋 axis (see p. 71) is quasi-finite
but not finite — its image is not closed in 𝔸1 , and 𝑘[𝑋, 𝑋 −1 ] is not finite over 𝑘[𝑋].

𝑡 ↦ (𝑡2 , 𝑡3 ) ∶ 𝔸1 → 𝑉(𝑌 2 − 𝑋 3 ) ⊂ 𝔸2
8.31. The map

from the line to the cuspidal cubic is finite because the image of 𝑘[𝑋, 𝑌] in 𝑘[𝑇] is
𝑘[𝑇 2 , 𝑇 3 ], and {1, 𝑇} is a set of generators for 𝑘[𝑇] as a 𝑘[𝑇 2 , 𝑇 3 ]-module (see 3.29).

8.32. The map 𝔸1 → 𝔸1 , 𝑎 ↦ 𝑎𝑚 is finite.

8.33. The obvious map

(𝔸1 with the origin doubled ) → 𝔸1

is quasi-finite but not finite (the inverse image of 𝔸1 is not affine).

186 8. Normal Varieties; (Quasi-)finite maps; Zariski’s Main Theorem

8.34. The map 𝔸2 ∖ {origin} → 𝔸2 is quasi-finite but not finite, because the inverse
image of 𝔸2 is not affine (see 3.33). The map

𝔸2 ∖ {(0, 0)}⊔ ∗→ 𝔸2

sending ∗ to (0, 0) is bijective but not finite (here ∗= Spm(𝑘) = 𝔸0 ).

8.35. The map in 8.31 and the Frobenius map

(𝑡1 , … , 𝑡𝑛 ) ↦ (𝑡1 , … , 𝑡𝑛 ) ∶ 𝔸𝑛 → 𝔸𝑛
𝑝 𝑝

in characteristic 𝑝 ≠ 0, are examples of finite bijective regular maps that are not isomor-
phisms.

8.36. Let 𝑓 be the regular map

(𝑥, 𝑦) ↦ (𝑥, 𝑥𝑦 2 + 𝑦 + 1) ∶ 𝔸2 → 𝔸2 .

Then 𝑓 is (obviously) quasi-finite, but it is not finite. For this we have to show that
𝑘[𝑋, 𝑌] is not integral over its subring 𝑘[𝐴, 𝐵], where

𝐴=𝑋
𝐵 = 𝑋𝑌 2 + 𝑌 + 1.

The minimal polynomial of 𝑌 over 𝑘[𝐴, 𝐵] is

𝐴𝑌 2 + 𝑌 + 1 − 𝐵 = 0,

which shows that it is not integral over 𝑘[𝐴, 𝐵] (see 1.44). Alternatively, one can show
directly that 𝑌 can never satisfy an equation

𝑌 𝑠 + 𝑔1 (𝐴, 𝐵)𝑌 𝑠−1 + ⋯ + 𝑔𝑠 (𝐴, 𝐵) = 0, 𝑔𝑖 (𝐴, 𝐵) ∈ 𝑘[𝐴, 𝐵],

by multiplying the equation by 𝐴.

8.37. Let 𝑉 be the hyperplane

𝑋 𝑛 + 𝑇1 𝑋 𝑛−1 + ⋯ + 𝑇𝑛 = 0

in 𝔸𝑛+1 , and consider the projection map

(𝑎1 , … , 𝑎𝑛 , 𝑥) ↦ (𝑎1 , … , 𝑎𝑛 ) ∶ 𝑉 → 𝔸𝑛 .

The fibre over a point (𝑎1 , … , 𝑎𝑛 ) ∈ 𝔸𝑛 is the set of solutions of

𝑋 𝑛 + 𝑎1 𝑋 𝑛−1 + ⋯ + 𝑎𝑛 = 0,

and so it has exactly 𝑛 points, counted with multiplicities. The map is certainly quasi-
finite; it is also finite because it corresponds to the finite map of 𝑘-algebras,

𝑘[𝑇1 , … , 𝑇𝑛 ] → 𝑘[𝑇1 , … , 𝑇𝑛 , 𝑋]∕(𝑋 𝑛 + 𝑇1 𝑋 𝑛−1 + ⋯ + 𝑇𝑛 ).

See also the more general example p. 51.

 d. The fibres of finite maps 187

8.38. Let 𝑉 be the hyperplane

𝑇0 𝑋 𝑛 + 𝑇1 𝑋 𝑛−1 + ⋯ + 𝑇𝑛 = 0

in 𝔸𝑛+2 . The projection map

𝜑
(𝑎0 , … , 𝑎𝑛 , 𝑥) ↦ (𝑎0 , … , 𝑎𝑛 ) ∶ 𝑉 ,→ 𝔸𝑛+1

has finite fibres except for the fibre above 𝑜 = (0, … , 0), which is 𝔸1 . Its restriction to
𝑉 ∖ 𝜑−1 (𝑜) is quasi-finite, but not finite. Above points of the form (0, … , 0, ∗, … , ∗) some
of the roots “vanish off to ∞”. (Example 8.30 is a special case of this.) See also the more
general example p. 51.

𝑃(𝑋, 𝑌) = 𝑇0 𝑋 𝑛 + 𝑇1 𝑋 𝑛−1 𝑌 + ⋯ + 𝑇𝑛 𝑌 𝑛 ,
8.39. Let

and let 𝑉 be its zero set in ℙ1 ×(𝔸𝑛+1 ∖{0}). In this case, the projection map 𝑉 → 𝔸𝑛+1 ∖{0}
is finite.

d. The fibres of finite maps

Let 𝜑 ∶ 𝑊 → 𝑉 be a finite dominant morphism of irreducible varieties. Then dim(𝑊) =
dim(𝑉), and so 𝑘(𝑊) is a finite field extension of 𝑘(𝑉). Its degree is called the degree of
the map 𝜑. The map 𝜑 is said to be separable if the field 𝑘(𝑊) is separable over 𝑘(𝑉).
Recall that |𝑆| denotes the number of elements in a finite set 𝑆.

Theorem 8.40. Let 𝜑 ∶ 𝑊 → 𝑉 be a finite surjective regular map of irreducible varieties

with 𝑉 normal.
| |
(a) For all 𝑃 ∈ 𝑉, |||𝜑−1 (𝑃)||| ≤ deg(𝜑).
| |
(b) The set of points 𝑃 of 𝑉 such that |||𝜑−1 (𝑃)||| = deg(𝜑) is an open subset of 𝑉, and it is
nonempty if 𝜑 is separable.

Before proving the theorem, we give examples to show that we need 𝑊 to be separated
and 𝑉 to be normal in (a), and that we need 𝑘(𝑊) to be separable over 𝑘(𝑉) for the
second part of (b).

Example 8.41. (a) The map

{𝔸1 with origin doubled } → 𝔸1

(b) Let 𝐶 be the curve 𝑌 2 = 𝑋 3 + 𝑋 2 , and consider the map

has degree one and is one-to-one except over the origin where it is two-to-one.

𝑡 ↦ (𝑡 2 − 1, 𝑡(𝑡 2 − 1)) ∶ 𝔸1 → 𝐶.

It is one-to-one except that the points 𝑡 = ±1 both map to 0. On coordinate rings, it

𝑥 ↦ 𝑇2 − 1
corresponds to the inclusion

𝑘[𝑥, 𝑦] → 𝑘[𝑇], {
𝑦 ↦ 𝑇(𝑇 2 − 1),

and so is of degree one. The ring 𝑘[𝑥, 𝑦] is not integrally closed — in fact 𝑘[𝑇] is the
integral closure of 𝑘[𝑥, 𝑦] in its field of fractions 𝑘(𝑥, 𝑦) = 𝑘(𝑇).
188 8. Normal Varieties; (Quasi-)finite maps; Zariski’s Main Theorem

(c) The Frobenius map

(𝑎1 , … , 𝑎𝑛 ) ↦ (𝑎1 , … , 𝑎𝑛 ) ∶ 𝔸𝑛 → 𝔸𝑛
𝑝 𝑝

in characteristic 𝑝 ≠ 0 is bijective on points, but has degree 𝑝𝑛 . The field extension

corresponding to the map is

𝑘(𝑋1 , … , 𝑋𝑛 ) ⊃ 𝑘(𝑋1 , … , 𝑋𝑛 )
𝑝 𝑝

which is purely inseparable.

Lemma 8.42. Let 𝑄1 , … , 𝑄𝑟 be distinct points on an affine variety 𝑉. Then there is a regular
function 𝑓 on 𝑉 taking distinct values at the 𝑄𝑖 .

Proof. We can embed 𝑉 as closed subvariety of 𝔸𝑛 , and then it suffices to prove the
statement with 𝑉 = 𝔸𝑛 — almost any linear form will do. 2

Proof (of 8.40). In proving (a) of the theorem, we may assume that 𝑉 and 𝑊 are
affine, and so the map corresponds to a finite map of 𝑘-algebras, 𝑘[𝑉] → 𝑘[𝑊]. Let
𝜑−1 (𝑃) = {𝑄1 , … , 𝑄𝑟 }. According to the lemma, there exists an 𝑓 ∈ 𝑘[𝑊] taking distinct
values at the 𝑄𝑖 . Let
𝐹(𝑇) = 𝑇 𝑚 + 𝑎1 𝑇 𝑚−1 + ⋯ + 𝑎𝑚
be the minimal polynomial of 𝑓 over 𝑘(𝑉). It has degree 𝑚 ≤ [𝑘(𝑊) ∶ 𝑘(𝑉)] = deg 𝜑,
and it has coefficients in 𝑘[𝑉] because 𝑉 is normal (see 1.44). Now 𝐹(𝑓) = 0 implies
𝐹(𝑓(𝑄𝑖 )) = 0, i.e.,

𝑓(𝑄𝑖 )𝑚 + 𝑎1 (𝑃) ⋅ 𝑓(𝑄𝑖 )𝑚−1 + ⋯ + 𝑎𝑚 (𝑃) = 0.

Therefore the 𝑓(𝑄𝑖 ) are all roots of a single polynomial of degree 𝑚, and so 𝑟 ≤ 𝑚 ≤
deg(𝜑).
In order to prove the first part of (b), we show that, if there is a point 𝑃 ∈ 𝑉 such that
𝜑 (𝑃) has deg(𝜑) elements, then the same is true for all points in an open neighbourhood
−1

of 𝑃. Choose 𝑓 as in the last paragraph corresponding to such a 𝑃. Then the polynomial

𝑇 𝑚 + 𝑎1 (𝑃) ⋅ 𝑇 𝑚−1 + ⋯ + 𝑎𝑚 (𝑃) = 0 (*)

has 𝑟 = deg 𝜑 distinct roots, and so 𝑚 = 𝑟. Consider the discriminant disc 𝐹 of 𝐹.

Because (*) has distinct roots, disc(𝐹)(𝑃) ≠ 0, and so disc(𝐹) is nonzero on an open
neighbourhood 𝑈 of 𝑃. The factorization
𝑇↦𝑓
𝑘[𝑉] ,,,,→ 𝑘[𝑉][𝑇]∕(𝐹) ,,,,→ 𝑘[𝑊]

𝑊 → Spm(𝑘[𝑉][𝑇]∕(𝐹)) → 𝑉.
gives a factorization

Each point 𝑃′ ∈ 𝑈 has exactly 𝑚 inverse images under the second map, and the first map

proves that 𝜑−1 (𝑃′ ) has at least deg(𝜑) points for 𝑃′ ∈ 𝑈, and part (a) of the theorem
is finite and dominant, and therefore surjective (recall that a finite map is closed). This

then implies that it has exactly deg(𝜑) points.

that 𝜑−1 (𝑃) has deg 𝜑 elements. Because 𝑘(𝑊) is separable over 𝑘(𝑉), there exists an
We now show that if the field extension is separable, then there exists a point such
 e. Zariski’s main theorem 189

𝑓 ∈ 𝑘[𝑊] such that 𝑘(𝑉)[𝑓] = 𝑘(𝑊). Its minimal polynomial 𝐹 has degree deg(𝜑) and
its discriminant is a nonzero element of 𝑘[𝑉]. The diagram

𝑊 → Spm(𝑘[𝑉][𝑇]∕(𝐹)) → 𝑉

shows that |𝜑−1 (𝑃)| ≥ deg(𝜑) for 𝑃 a point such that disc(𝑓)(𝑃) ≠ 0.

Let 𝐸 ⊃ 𝐹 be a finite extension of fields. The elements of 𝐸 separable over 𝐹 form

2

a subfield 𝐹 ′ of 𝐸, and the separable degree of 𝐸 over 𝐹 is defined to be the degree of

𝐹 ′ over 𝐹. The separable degree of a finite surjective map 𝜑 ∶ 𝑊 → 𝑉 of irreducible
varieties is the separable degree of 𝑘(𝑊) over 𝑘(𝑉).

Theorem 8.43. Let 𝜑 ∶ 𝑊 → 𝑉 be a finite surjective regular map of irreducible varieties,

and assume that 𝑉 is normal.
| |
(a) For all 𝑃 ∈ 𝑉, |||𝜑−1 (𝑃)||| ≤ sep deg(𝜑), with equality holding on a dense open subset.
(b) For all 𝑖,
| |
𝑉𝑖 = {𝑃 ∈ 𝑉 ∣ |||𝜑−1 (𝑃)||| ≤ 𝑖}
is closed in 𝑉.

Proof. If 𝜑 is separable, this was proved in 8.40. If 𝜑 is purely inseparable, then 𝜑 is

𝐹
one-to-one because, for some 𝑞, the Frobenius map 𝑉 (𝑞 ) ,→ 𝑉 factors through 𝜑. In
−1

the general case, 𝜑 factors through the normalization of 𝑉 in 𝐹 ′ , which realizes 𝜑 as the
composite of a purely inseparable map with a separable map. 2

Aside 8.44. A finite map from a variety onto a normal variety is open (hence both open and
closed). See 8.52.

e. Zariski’s main theorem

In this section, we explain a fundamental theorem of Zariski.

Statement and proof

An obvious way of constructing nonfinite quasi-finite map is to take a finite map 𝑊 → 𝑉
and remove a closed subset of 𝑊. Zariski’s Main Theorem (ZMT) shows that, for algebraic
varieties, every quasi-finite map arises in this way.

𝑗 𝜑′
eties 𝜑 ∶ 𝑊 → 𝑉 factors into 𝑊 → 𝑉 ′ → 𝑉 with 𝜑′ finite and 𝑗 an open immersion:
Theorem 8.45 (Zariski’s Main Theorem). Every quasi-finite map of algebraic vari-

𝑊 ← → 𝑉′
←
𝑗
open immersion

𝜑 𝜑′
→
←

𝑉.
quasi-finite → finite

When 𝜑 is a dominant map of irreducible varieties, the statement is true with 𝜑′ ∶ 𝑉 ′ → 𝑉

equal to the normalization of 𝑉 in 𝑊 (in the sense of 8.9).

algebra. For a ring 𝐴 and a prime ideal 𝔭 in 𝐴, 𝜅(𝔭) denotes the field of fractions of 𝐴∕𝔭.
The key result needed to prove 8.45 is the following statement from commutative
190 8. Normal Varieties; (Quasi-)finite maps; Zariski’s Main Theorem

Theorem 8.46 (local version of ZMT). Let 𝐴 be a commutative ring, and let 𝑖 ∶ 𝐴 →
𝐵 be a finitely generated 𝐴-algebra. Let 𝔮 be a prime ideal of 𝐵, and let 𝔭 = 𝑖 −1 (𝔮). Finally,
let 𝐴′ denote the integral closure of 𝐴 in 𝐵. If 𝐵𝔮 ∕𝔭𝐵𝔮 is a finite 𝜅(𝔭)-algebra, then there
exists an 𝑓 ∈ 𝐴′ not in 𝔮 such that the map 𝐴𝑓′ → 𝐵𝑓 is an isomorphism.

Proof. The proof is quite elementary, but intricate — see §17 of my notes CA. 2

Recall that a point 𝑣 in a topological space 𝑉 is isolated if {𝑣} is an open subset of 𝑉.

The isolated points 𝑣 of an algebraic variety 𝑉 are those such that {𝑣} is both open and
closed. Thus they are the irreducible components of 𝑉 of dimension 0. Thus, if {𝑣} is
isolated in 𝑉, then 𝑉 = {𝑣} ⊔ 𝑉 ′ , and, if 𝑉 is affine, then 𝑘[𝑉] ≃ 𝑘 × 𝑘[𝑉 ′ ].
Let 𝜑 ∶ 𝑊 → 𝑉 be a continuous map of topological spaces. We say that 𝑤 ∈ 𝑊 is
isolated in its fibre if it is isolated in the subspace 𝜑−1 (𝜑(𝑤)) of 𝑊. Let 𝜑 ∶ 𝐴 → 𝐵 be
a homomorphism of finitely generated 𝑘-algebras, and consider spm(𝜑) ∶ spm(𝐵) →
spm(𝐴); then 𝔫 ∈ spm(𝐵) is isolated in its fibre if and only if 𝐵𝔫 ∕𝔪𝐵𝔫 is a finite 𝑘-
algebra; here 𝔪 = 𝜑−1 (𝔫).

Proposition 8.47. Let 𝜑 ∶ 𝑊 → 𝑉 be a regular map of algebraic varieties. The set 𝑊 ′ of

points of 𝑊 isolated in their fibres is open in 𝑊.

Proof. Let 𝑤 ∈ 𝑊 ′ . Let 𝑊𝑤 and 𝑉𝑣 be open affine neighbourhoods of 𝑤 and 𝑣 = 𝜑(𝑤)

such that 𝜑(𝑊𝑤 ) ⊂ 𝑉𝑣 , and let 𝐴 = 𝑘[𝑉𝑣 ] and 𝐵 = 𝑘[𝑊𝑤 ]. Let 𝔫 = {𝑓 ∈ 𝐵 ∣ 𝑓(𝑤) = 0}
— it is the maximal ideal in 𝐵 corresponding to 𝑤.
Let 𝐴′ be the integral closure of 𝐴 in 𝐵. Theorem 8.46 shows that there exists an
𝑓 ∈ 𝐴′ not in 𝔪 such that 𝐴𝑓′ ≃ 𝐵𝑓 . Write 𝐴′ as the union of the finitely generated
𝐴-subalgebras 𝐴𝑖 of 𝐴′ containing 𝑓:
⋃
𝐴′ = 𝐴𝑖 .
𝑖

Because 𝐴′ is integral over 𝐴, each 𝐴𝑖 is finite over 𝐴 (see 1.35). We have

⋃
𝐵𝑓 ≃ 𝐴𝑓′ = 𝐴𝑖𝑓 .
𝑖

Because 𝐵𝑓 is a finitely generated 𝐴-algebra, 𝐵𝑓 = 𝐴𝑖𝑓 for all sufficiently large 𝐴𝑖 . As the
𝐴𝑖 are finite over 𝐴, 𝐵𝑓 is quasi-finite over 𝐴, and spm(𝐵𝑓 ) is an open neighbourhood of
𝑤 consisting of quasi-finite points. 2

Proposition 8.48. Every quasi-finite map of affine algebraic varieties 𝜑 ∶ 𝑊 → 𝑉 factors

𝑗 𝜑′
into 𝑊 ,→ 𝑉 ′ ,→ 𝑉 with 𝑗 a dominant open immersion and 𝜑′ finite.

Proof. Let 𝐴 = 𝑘[𝑉] and 𝐵 = 𝑘[𝑊]. Because 𝜑 is quasi-finite, Theorem 8.46 shows
that there exist 𝑓𝑖 ∈ 𝐴′ such that the sets spm(𝐵𝑓𝑖 ) form an open covering of 𝑊 and
𝐴𝑓′ ≃ 𝐵𝑓𝑖 for all 𝑖. As 𝑊 quasi-compact, finitely many sets spm(𝐵𝑓𝑖 ) suffice to cover
𝑊. The argument in the proof of (8.47) shows that there exists an 𝐴-subalgebra 𝐴′′ of
𝑖

𝐴′ , finite over 𝐴, which contains 𝑓1 , … , 𝑓𝑛 and is such that 𝐵𝑓𝑖 ≃ 𝐴𝑓′′ for all 𝑖. Now the
map 𝑊 = Spm(𝐵) → Spm(𝐴′′ ) is an open immersion because it is when restricted to
𝑖

Spm(𝐵𝑓𝑖 ) for each 𝑖. As Spm(𝐴′′ ) → Spm(𝐴) = 𝑉 is finite, we can take 𝑉 ′ = Spm(𝐴′′ ).2

A regular map 𝜑 ∶ 𝑊 → 𝑉 is said to be affine if 𝜑−1 (𝑈) is an open affine subset of

𝑊 whenever 𝑈 is an open affine subset of 𝑉.
 e. Zariski’s main theorem 191

Proposition 8.49. Let 𝜑 ∶ 𝑊 → 𝑉 be an affine map of irreducible algebraic varieties.

Then the map 𝑗 ∶ 𝑊 → 𝑉 ′ from 𝑊 into the normalization 𝑉 ′ of 𝑉 in 𝑊 (8.9) is an open
immersion.

Proof. Let 𝑈 be an open affine in 𝑉. Let 𝐴 = 𝑘[𝑈] and 𝐵 = 𝑘[𝜑−1 (𝑈)]. The integral
closure 𝐴′ of 𝐴 in 𝐵 is finite over 𝐴 because it is contained in the integral closure of 𝐴 in
𝑘(𝑊), which is finite over 𝐴 (8.3)). Thus, in the proof of 8.48 we can take 𝐴′′ = 𝐴′ , and
then 𝜑−1 (𝑈) → Spm(𝐴′ ) is an open immersion. As Spm(𝐴′ ) is an open subvariety of 𝑉 ′
and the sets 𝜑−1 (𝑈) cover 𝑊, this implies that 𝑗 ∶ 𝑊 → 𝑉 ′ is an open immersion. 2

As 𝑉 ′ → 𝑉 is finite, this proves Theorem 8.45 in the case that 𝜑 is an affine map of
irreducible varieties. To deduce the general case of Theorem 8.45 from 8.44 requires
an additional argument. See Theorem 12.83 of Görtz, U. and Wedhorn, T., Algebraic
Geometry I., Springer Spektrum, Wiesbaden, 2020.

8.50. Let 𝜑 ∶ 𝑊 → 𝑉 be a quasi-finite map of algebraic varieties. In 8.45, we may

Notes

replace 𝑉 ′ with the closure of the image of 𝑗. Thus, there is a factorization 𝜑 = 𝜑′ ◦𝑗

with 𝜑′ finite and 𝑗 a dominant open immersion.

maps of prevarieties. A regular map 𝜑 ∶ 𝑉 → 𝑆 of algebraic prevarieties is separated if

8.51. Theorem 8.45 is false for prevarieties (see 8.33). However, it is true for separated

the image ∆𝑉∕𝑆 of the map 𝑣 ↦ (𝑣, 𝑣) ∶ 𝑉 → 𝑉 ×𝑆 𝑉 is closed.

8.52. If 𝑉 is normal in 8.45, then 𝜑′ is open (8.44), and so 𝜑 is open. Thus, every
quasi-finite map to a normal algebraic variety is open.

Aside. The normalization map of an affine cuspidal cubic 𝑌 2 = 𝑋 3 is a universal homeomor-

phism without being an isomorphism. The normalization map of an affine nodal cubic with one
of the points lying over the node removed is a homeomorphism but not a universal homeomor-
phism. In particular, a regular map may be étale and bijective, without being a homeomorphism,
much less an isomorphism. See mo479766.

Applications to quasi-finite maps

Zariski’s main theorem allows us to give a geometric criteria for a regular map to be

Proposition 8.53. A quasi-finite map 𝜑 ∶ 𝑊 → 𝑉 of algebraic varieties is finite if 𝑊 is

finite.

complete.

Proof. The map 𝑗 ∶ 𝑊 → 𝑉 ′ in 8.45 is an isomorphism of 𝑊 onto its image 𝑗(𝑊) in

𝑉 ′ . If 𝑊 is complete, then 𝑗(𝑊) is closed (7.7), and so the restriction of 𝜑′ to 𝑗(𝑊) is
finite. 2

Proposition 8.54. A quasi-finite map 𝜑 ∶ 𝑊 → 𝑉 of algebraic varieties is finite if (and

only if) it is proper.
𝑗 𝛼
Proof. Factor 𝜑 into 𝑊 → 𝑊 ′ → 𝑉 with 𝛼 finite and 𝑗 an open immersion. Factor 𝑗

𝑤↦(𝑤,𝑗𝑤) (𝑤,𝑤′ )↦𝑤 ′

𝑊 ,,,,,,,,,→ 𝑊 ×𝑉 𝑊 ′ ,,,,,,,,,→ 𝑊 ′ .
into
192 8. Normal Varieties; (Quasi-)finite maps; Zariski’s Main Theorem

The image of the first map is 𝛤𝑗 , which is closed because 𝑊 ′ is a variety (see 5.28; 𝑊 ′ is
separated because it is finite over a variety — exercise). Because 𝜑 is proper, the second
map is closed. Hence 𝑗 is an open immersion with closed image. It follows that its
image is a connected component of 𝑊 ′ , and that 𝑊 is isomorphic to that connected
component. 2

8.55. When 𝑊 and 𝑉 are curves, every surjective map 𝑊 → 𝑉 is closed. Thus it is easy
Notes

( 1 )
to give examples of closed surjective quasi-finite, but nonfinite, maps. Consider, for

𝔸 ∖ {0} ⊔ 𝔸0 → 𝔸1 ,
example, the map

sending each 𝑎 ∈ 𝔸1 ∖ {0} to 𝑎 and 𝑂 ∈ 𝔸0 to 0. This does not violate the Proposition
8.54, because the map is only closed, not universally closed.

Applications to birational maps

Recall (p. 117) that a regular map 𝜑 ∶ 𝑊 → 𝑉 of irreducible varieties is said to be
birational if it induces an isomorphism 𝑘(𝑉) → 𝑘(𝑊) on the fields of rational functions.

8.56. One may ask how a birational regular map 𝜑 ∶ 𝑊 → 𝑉 can fail to be an isomor-
phism. Here are three examples.

(b) The map (8.31) from 𝔸1 to the cuspidal cubic,

(a) The inclusion of an open subvariety into a variety is birational.

𝔸1 → 𝐶, 𝑡 ↦ (𝑡2 , 𝑡3 ),

is birational. Here 𝐶 is the cubic 𝑌 2 = 𝑋 3 , and the map 𝑘[𝐶] → 𝑘[𝔸1 ] = 𝑘[𝑇]
identifies 𝑘[𝐶] with the subring 𝑘[𝑇 2 , 𝑇 3 ] of 𝑘[𝑇]. Both rings have 𝑘(𝑇) as their

(c) For any smooth variety 𝑉 and point 𝑃 ∈ 𝑉, there is a regular birational map
fields of fractions.

𝜑 ∶ 𝑉 ′ → 𝑉 such that the restriction of 𝜑 to 𝑉 ′ ∖ 𝜑−1 (𝑃) is an isomorphism onto

𝑉 ∖ 𝑃, but 𝜑−1 (𝑃) is the projective space attached to the vector space 𝑇𝑃 (𝑉). See
the section on blow-ups below.

The next result says that, if we require the target variety to be normal (thereby
excluding example (b)), and we require the map to be quasi-finite (thereby excluding

Proposition 8.57. Let 𝜑 ∶ 𝑊 → 𝑉 be a birational regular map of irreducible varieties. If

example (c)), then we are left with (a).

𝑉 is normal and the map 𝜑 is quasi-finite, then 𝜑 is an isomorphism from 𝑊 onto an open
subvariety of 𝑉.

Proof. Factor 𝜑 as in the Theorem 8.45 (so, in particular, 𝜑′ ∶ 𝑉 ′ → 𝑉 is the normaliza-

tion of 𝑉 in 𝑊). For each open affine subset 𝑈 of 𝑉, 𝑘[𝜑′−1 (𝑈)] is the integral closure of
𝑘[𝑈] in 𝑘(𝑊). Because 𝜑 is birational, the inclusion 𝑘(𝑉) ⊂ 𝑘(𝑉 ′ ) = 𝑘(𝑊) is an equality.
Now 𝑘[𝑈] is integrally closed in 𝑘(𝑉) (because 𝑉 is normal), and so 𝑈 = 𝜑′−1 (𝑈) (as
varieties). We have shown that 𝜑′ ∶ 𝑉 ′ → 𝑉 is an isomorphism locally on the base 𝑉,
and hence an isomorphism. 2
 e. Zariski’s main theorem 193

8.58. In topology, a continuous bijective map 𝜑 ∶ 𝑊 → 𝑉 need not be a homeomor-

phism, but it is if 𝑊 is compact and 𝑉 is Hausdorff. Similarly, a bijective regular map of

(a) In characteristic 𝑝, the Frobenius map

algebraic varieties need not be an isomorphism. Here are three examples:

(𝑥1 , … , 𝑥𝑛 ) ↦ (𝑥1 , … , 𝑥𝑛 ) ∶ 𝔸𝑛 → 𝔸𝑛
𝑝 𝑝

is bijective and regular, but it is not an isomorphism even though 𝔸𝑛 is normal.

(b) The map 𝑡 ↦ (𝑡2 , 𝑡3 ) from 𝔸1 to the cuspidal cubic (see 8.56b) is bijective, but not

(c) Consider the regular map 𝔸1 → 𝔸1 sending 𝑥 to 1∕𝑥 for 𝑥 ≠ 0 and 0 to 0. Its graph
an isomorphism.

𝛤 is the union of (0, 0) and the hyperbola 𝑥𝑦 = 1, which is a closed subvariety of

𝔸1 × 𝔸1 . The projection (𝑥, 𝑦) ↦ 𝑥 ∶ 𝛤 → 𝔸1 is a bijective, regular, birational
map, but it is not an isomorphism even though 𝔸1 is normal.

If we require the map to be birational (thereby excluding example (a)), 𝑉 to be normal

(thereby excluding example (b)), and the varieties to be irreducible (thereby excluding

Proposition 8.59. Let 𝜑 ∶ 𝑊 → 𝑉 be a bijective regular map of irreducible algebraic

example (c)), then the map is an isomorphism.

varieties. If the map 𝜑 is birational and 𝑉 is normal, then 𝜑 is an isomorphism.

Proof. The hypotheses imply that 𝜑 is an isomorphism of 𝑊 onto an open subset of 𝑉

(8.57). Because 𝜑 is bijective, the open subset must be the whole of 𝑉. 2

In fact, example (a) can be excluded by requiring that 𝜑 be generically separable

Proposition 8.60. Let 𝜑 ∶ 𝑊 → 𝑉 be a bijective regular map of irreducible varieties. If

(instead of birational), and (b) can be excluded by requiring that the map be étale.

𝑉 is normal and 𝑘(𝑊) is separably generated over 𝑘(𝑉), then 𝜑 is an isomorphism.

Proof. Because 𝜑 is bijective, dim(𝑊) = dim(𝑉) (see Theorem 9.9 below) and the
separable degree of 𝑘(𝑊) over 𝑘(𝑉) is 1 (apply 8.40 to the variety 𝑉 ′ in 8.45). Hence 𝜑 is
birational, and we may apply 8.59. 2

𝜑 ∶ 𝑊 → 𝑉 between two Banach spaces has a closed graph 𝛤 = {(𝑤, 𝜑𝑤) ∣ 𝑤 ∈ 𝑊},
8.61. In functional analysis, the closed graph theorem states that, if a linear map

then 𝜑 is continuous (Wikipedia: Closed graph theorem). One can ask whether
def

a similar statement is true in algebraic geometry. Specifically, if 𝜑 ∶ 𝑊 → 𝑉 is a map

(in the set-theoretic sense) of algebraic varieties 𝑉, 𝑊 whose graph is closed (for the
Zariski topology), then is 𝜑 a regular map? The answer is no in general. For example,
even in characteristic zero, the map (𝑡 2 , 𝑡3 ) → 𝑡 ∶ 𝐶 → 𝔸1 inverse to that in 8.56(b) has
closed graph but is not regular. In characteristic 𝑝, the inverse of the Frobenius map
𝑥 ↦ 𝑥 𝑝 provides another counterexample. For a third counterexample, see 8.58(c). The
projection 𝜋 from 𝛤 to 𝑊 is a bijective regular map, and so 𝜑 will be regular if 𝜋 is an
isomorphism. According to 8.60, 𝜋 is an isomorphism if the varieties are irreducible,
𝑊 is normal, and 𝜋 is generically separable. In particular, a map between irreducible
normal algebraic varieties in characteristic zero is regular if its graph is closed.
194 8. Normal Varieties; (Quasi-)finite maps; Zariski’s Main Theorem

A condition for an algebraic monoid to be a group

Recall (p. 110) that we defined an algebraic variety 𝑉 with a group structure 𝑉 × 𝑉 → 𝑉
to be a group variety if both the multiplication map 𝑎, 𝑏 ↦ 𝑎 ⋅ 𝑏 ∶ 𝑉 × 𝑉 → 𝑉 and
inversion map 𝑎 ↦ 𝑎−1 ∶ 𝑉 → 𝑉 are regular. Here we show that the second condition

A monoid variety is an algebraic variety 𝐺 together with the structure of a monoid

is unnecssary.

defined by regular maps

𝑚 ∶ 𝐺 × 𝐺 → 𝐺, 𝑒 ∶ 𝔸0 → 𝐺.

Lemma 8.62. Let (𝐺, 𝑚, 𝑒) be a monoid variety. The map

(𝑑𝑚)(𝑒,𝑒)
𝑇𝑒 𝐺 ⊕ 𝑇𝑒 𝐺 ≃ 𝑇(𝑒,𝑒) (𝐺 × 𝐺) ,,,,,,,→ 𝑇𝑒 (𝐺)

is addition.

Proof. The first isomorphism is (𝑋, 𝑌) ↦ (𝑑𝛼)𝑒 (𝑋) + (𝑑𝛽)𝑒 (𝑌), where 𝛼 is the map
𝑥 ↦ (𝑥, 𝑒) ∶ 𝐺 → 𝐺 × 𝐺 and 𝛽 is 𝑥 ↦ (𝑒, 𝑥). To compute (𝑑𝑚)(𝑒,𝑒) ((𝑑𝛽)𝑒 (𝑋) + (𝑑𝛼)𝑒 (𝑌)),
note that 𝑚◦𝛼 = id𝐺 = 𝑚◦𝛽. 2

Proposition 8.63. Let 𝐺 be an algebraic variety with a group structure 𝑚 ∶ 𝐺 × 𝐺 → 𝐺.

If 𝑚 is regular, then (𝐺, 𝑚) is a group variety, i.e., the map 𝑎 ↦ 𝑎−1 is regular.

Proof. Let 𝑎 ∈ 𝐺(𝑘). The translation map 𝐿𝑎 ∶ 𝑥 ↦ 𝑎𝑥 is an isomorphism 𝐺 → 𝐺

because it has an inverse 𝐿𝑎−1 . Therefore 𝐺 is homogeneous as an algebraic variety: for
any two points in 𝐺, there is an isomorphism 𝐺 → 𝐺 mapping one to the other. It follows
that 𝐺 normal (8.7).

(𝑥, 𝑦) ↦ (𝑥, 𝑥𝑦) ∶ 𝐺 × 𝐺 → 𝐺 × 𝐺

The map

is regular, bijective, and induces an isomorphism on the tangent spaces at (𝑒, 𝑒) (apply
the lemma). It is therefore an isomorphism of algebraic varieties over 𝑘. Therefore, its
inverse (𝑥, 𝑦) ↦ (𝑥, 𝑥−1 𝑦) is regular, and so

(𝑥, 𝑦) ↦ 𝑥−1 𝑦 ∶ 𝐺 × 𝐺 → 𝐺

is regular. This implies that (𝐺, 𝑚) is an algebraic group. 2

Variants of Zariski’s main theorem

Original form (8.57) Let 𝜑 ∶ 𝑊 → 𝑉 be a birational regular map of irreducible varieties.

Mumford 1966a, III, §9, lists the following variants of ZMT.

If 𝑉 is normal and 𝜑 is quasi-finite, then 𝜑 is an isomorphism of 𝑊 onto an open

subvariety of 𝑉.
Topological form Let 𝑉 be a normal variety over ℂ, and let 𝑣 ∈ 𝑉. Let 𝑆 be the
singular locus of 𝑉. Then the complex neighbourhoods 𝑈 of 𝑣 such that 𝑈 ∖ 𝑈 ∩ 𝑆
is connected form a base for the system of complex neighbourhoods of 𝑣.
Power series form Let 𝑉 be a normal variety, and let 𝒪𝑉,𝑍 be the local ring attached to
an irreducible closed subset of 𝑉 (cf. p. 180). If 𝒪𝑉,𝑍 is an integrally closed integral
domain, then so also is its completion.
 f. Stein factorization 195

Grothendieck’s form (8.45) Every quasi-finite map of algebraic varieties factors as the

Connectedness theorem Let 𝜑 ∶ 𝑊 → 𝑉 be a proper birational map, and let 𝑣 be a

composite of an open immersion with a finite map.

(closed) normal point of 𝑉. The 𝜑−1 (𝑣) is a connected set (in the Zariski topology).
The original form of the theorem was proved by Zariski using a fairly direct argument
whose method does not seem to generalize.3 The power series form was also proved
by Zariski, who showed that it implied the original form. The last two forms are much
deeper and were proved by Grothendieck.
Notes. The original form of the theorem (8.57) is the “Main theorem” of Zariski, O., Foundations
of a general theory of birational correspondences. Trans. Amer. Math. Soc. 53, (1943). 490–542.

f. Stein factorization
The following important theorem shows that the fibres of a proper map are disconnected

Theorem 8.64 (Stein factorization). Every proper map 𝜑 ∶ 𝑊 → 𝑉 of algebraic

only because the fibres of finite maps are disconnected.

𝜑1 𝜑2
varieties factors into 𝑊 → 𝑊 ′ → 𝑉 with 𝜑1 proper with connected fibres and 𝜑2 finite.

When 𝑉 is affine, this is the factorization

𝑊 → Spm(𝒪𝑊 (𝑊)) → 𝑉.

The first major step in the proof of the theorem is to show that 𝜑∗ 𝒪𝑊 is a coherent
sheaf on 𝑉 (see Chapter 13). Here 𝜑∗ 𝒪𝑊 is the sheaf of 𝒪𝑉 -algebras on 𝑉,

𝑈 ⇝ 𝒪𝑊 (𝜑−1 (𝑈)).

To say that 𝜑∗ 𝒪𝑊 is coherent means that, on every open affine subset 𝑈 of 𝑉, it is the
sheaf of 𝒪𝑈 -algebras defined by a finite 𝑘[𝑈]-algebra. This, in turn, means that there
exists a regular map 𝜑2 ∶ Spm(𝜑∗ 𝒪𝑊 ) → 𝑉 that, over every open affine subset 𝑈 of 𝑉,
is the map attached by Spm to the map of 𝑘-algebras 𝑘[𝑈] → 𝒪𝑊 (𝜑−1 (𝑈)).
The Stein factorization is then
𝜑1 𝜑2
𝑊 ,→ 𝑊 ′ = Spm(𝜑∗ 𝒪𝑊 ) ,→ 𝑉.
def

By construction, 𝜑2 is finite and 𝜑1 ∶ 𝑊 → 𝑊 ′ has the property that 𝒪𝑊 ′ → 𝜑1∗ 𝒪𝑊

is an isomorphism. That its fibres are connected is a consequence of the following

Theorem 8.65. Let 𝜑 ∶ 𝑊 → 𝑉 be a proper map such that the map 𝒪𝑉 → 𝜑∗ 𝒪𝑊 is an

extension of Zariski’s connectedness theorem to non birational maps.

isomorphism. Then the fibres of 𝜑 are connected.

See Hartshorne 1977, III, §11.

Notes. The Stein factorization was originally proved (in 1956) by Stein for complex spaces
(Wikipedia: Stein factorization).
3
See Lang, S., Introduction to Algebraic Geometry, 1958, V 2, for Zariski’s original statement and proof
of this theorem. See Springer, T.A., Linear Algebraic Groups, 1998, 5.2.8, for a direct proof of (8.59).
196 8. Normal Varieties; (Quasi-)finite maps; Zariski’s Main Theorem

g. Blow-ups

Let 𝑃 be a nonsingular point on an algebraic variety 𝑉, and let 𝑇𝑝 (𝑉) be the tangent
Under construction.

space at 𝑃. The blow-up of 𝑉 at 𝑃 is a regular map 𝑉̃ → 𝑉 that replaces 𝑃 with the

projective space ℙ(𝑇𝑃 (𝑉)). More generally, the blow-up at 𝑃 replaces 𝑃 with ℙ(𝐶𝑃 (𝑉)),
where 𝐶𝑃 (𝑉) is the geometric tangent cone at 𝑃.

Blowing up the origin in 𝔸𝑛

Let 𝑂 be the origin in 𝔸𝑛 , and let 𝜋 ∶ 𝔸𝑛 ∖ {𝑂} → ℙ𝑛−1 be the map (𝑎1 , … , 𝑎𝑛 ) ↦
˜𝑛 be the closure of 𝛤𝜋 in 𝔸𝑛 × ℙ𝑛−1 .
(𝑎1 ∶ … ∶ 𝑎𝑛 ). Let 𝛤𝜋 be the graph of 𝜋, and let 𝔸
The map 𝜎 ∶ 𝔸 ˜ → 𝔸 defined by the projection map 𝔸𝑛 × ℙ𝑛−1 → 𝔸𝑛 is the blow-up
𝑛 𝑛

of 𝔸𝑛 at 𝑂.

Blowing up a point on a variety

Examples
8.66. The nodal cubic

8.67. The cuspidal cubic

h. Resolution of singularities
Let 𝑉 be an algebraic variety. A desingularization of 𝑉 is birational regular map
𝜋 ∶ 𝑊 → 𝑉 such that 𝑊 is nonsingular and 𝜋 is proper; if 𝑉 is projective, then 𝑊 should
also be projective, and 𝜋 should induce an isomorphism

𝑊 ∖ 𝜋−1 (Sing(𝑉)) → 𝑉 ∖ Sing(𝑉).

In other words, the nonsingular variety 𝑊 is the same as 𝑉 except over the singular
locus of 𝑉. When a variety admits a desingularization, then we say that resolution of
singularities holds for 𝑉.
Note that with “nonsingular” replaced by “normalization”, the normalization of 𝑉
(see 8.5) provides such a map (resolution of abnormalities).
Nagata’s embedding theorem 7.50 shows that it suffices to prove resolution of sin-
gularities for complete varieties, and Chow’s lemma 7.39 then shows that it suffices to
prove resolution of singularities for projective varieties. From now on, we shall consider
only projective varieties.
Resolution of singularities for curves was first obtained using blow-ups (see Chapter

of a variety, and observed that the normalization 𝜋 ∶ 𝑉̃ → 𝑉 of a curve 𝑉 in 𝑘(𝑉) is a

7 of Fulton’s book, Algebraic Curves). Zariski introduced the notion of the normalization

desingularization of 𝑉.
There were several proofs of resolution of singularities for surfaces over ℂ, but

1935). For a surface 𝑉, normalization gives a surface with only point singularities (8.12),
the first to be accepted as rigorous is that of Walker (patching Jung’s local arguments;

which can then be blown up. Zariski showed that the desingularization of a surface in
characteristic zero can be obtained by alternating normalizations and blow-ups.
The resolution of singularities for three-folds in characteristic zero is much more
difficult, and was first achieved by Zariski (Ann. of Math. 1944). His result was extended
 h. Resolution of singularities 197

to nonzero characteristic by his student Abhyankar and to all varieties in characteristic

zero by his student Hironaka.
The resolution of singularities for higher dimensional varieties in nonzero character-

de Jong proved a weaker result in which, instead of the map 𝜋 being birational, 𝑘(𝑊) is
istic is one of the most important outstanding problems in algebraic geometry. In 1996,

allowed to be a finite extension of 𝑘(𝑉).

The article Wikipedia: Resolution of singularities is excellent.

A little history
Normal varieties were introduced by Zariski in a paper, Amer. J. Math. 61, 1939, p. 249–

least 2 and that the system of hyperplane sections of a normal variety relative to a
194. There he noted that the singular locus of a normal variety has codimension at

projective embedding is complete (i.e., is a complete rational equivalence class). Zariski’s

introduction of the notion of a normal variety and of the normalization of a variety was
an important insertion of commutative algebra into algebraic geometry. It is not easy
to give a geometric intuition for “normal”. One criterion is that a variety is normal if
and only if every surjective finite birational map onto it is an isomorphism (8.57). See
mo109395 for a discussion of this question.

Exercises
8-1. Prove that a finite map is an isomorphism if and only if it is bijective and étale. (Cf.
Harris 1992, 14.9.)

8-2. Give an example of a surjective quasi-finite regular map that is not finite (different

8-3. Let 𝜑 ∶ 𝑊 → 𝑉 be an affine map. Show that 𝑊 is separated if 𝑉 is separated.

from any in the notes).

8-4. For every 𝑛 ≥ 1, find a finite map 𝜑 ∶ 𝑊 → 𝑉 with the following property: for all
1 ≤ 𝑖 ≤ 𝑛,
𝑉𝑖 = {𝑃 ∈ 𝑉 ∣ 𝜑−1 (𝑃) has ≤ 𝑖 points}
def

is a nonempty closed subvariety of dimension 𝑖.

Chapter 9

Regular Maps and Their Fibres

Consider again the regular map 𝜑 ∶ 𝔸2 → 𝔸2 , (𝑥, 𝑦) ↦ (𝑥, 𝑥𝑦) (Exercise 3-3). The line
𝑌 = 𝑐 maps to the line 𝑌 = 𝑐𝑋. As 𝑐 runs over the elements of 𝑘, this line sweeps out
the whole 𝑥, 𝑦-plane except for the 𝑦-axis, and so the image of 𝜑 is

𝐶 = (𝔸2 ∖ {𝑦-axis}) ∪ {(0, 0)},

(𝑎, 𝑏) is
which is neither open nor closed, and, in fact, is not even locally closed. The fibre over

point (𝑎, 𝑏∕𝑎) if 𝑎 ≠ 0

𝜑−1 (𝑎, 𝑏) = { 𝑌-axis if (𝑎, 𝑏) = (0, 0)
∅ if 𝑎 = 0, 𝑏 ≠ 0.
From this unpromising example, it would appear that it is not possible to say anything
about the image of a regular map or its fibres. However, it turns out that almost everything
that can go wrong already goes wrong in this example. We shall show:
(a) the image of a regular map is a finite union of locally closed sets;
(b) the dimensions of the fibres can jump only over closed subsets (upper semiconti-
nuity)
(c) the number of elements (if finite) in the fibres can drop only on closed subsets

and 𝑘 has characteristic zero.

(lower semicontinuity), provided the map is finite, the target variety is normal,

a. The constructibility theorem

Theorem 9.1. Let 𝜑 ∶ 𝑊 → 𝑉 be a dominant map of irreducible affine algebraic varieties.
Then 𝜑(𝑊) contains a nonempty open subset of 𝑉.

Proof. Because 𝜑 is dominant, the map 𝑓 ↦ 𝑓◦𝜑 ∶ 𝑘[𝑉] → 𝑘[𝑊] is injective (3.34).
According to Lemma 9.4 below, there exists a nonzero 𝑎 ∈ 𝑘[𝑉] such that every ho-
momorphism of 𝑘-algebras 𝛼 ∶ 𝑘[𝑉] → 𝑘 such that 𝛼(𝑎) ≠ 0 extends to a homomor-
phism 𝛽 ∶ 𝑘[𝑊] → 𝑘. In particular, for any point 𝑃 in 𝐷(𝑎) ⊂ 𝑉, the homomorphism
𝑔 ↦ 𝑔(𝑃) ∶ 𝑘[𝑉] → 𝑘 extends to a homomorphism 𝛽 ∶ 𝑘[𝑊] → 𝑘. The kernel of 𝛽 is a
maximal ideal of 𝑘[𝑊] whose zero set is a point 𝑄 of 𝑊 such that 𝜑(𝑄) = 𝑃. 2

Before beginning the proof of Lemma 9.4, we should look at an example.

198
 a. The constructibility theorem 199

Example 9.2. Let 𝐴 be an affine 𝑘-algebra, and let 𝐵 = 𝐴[𝑇]∕(𝑓) with 𝑓 = 𝑎𝑚 𝑇 𝑚 +

⋯ + 𝑎0 , 𝑚 > 0. When does a homomorphism 𝛼 ∶ 𝐴 → 𝑘 extend to 𝐵? The extensions of
𝛼 correspond to roots of the polynomial 𝛼(𝑎𝑚 )𝑇 𝑚 + ⋯ + 𝛼(𝑎0 ) in 𝑘, and so there exists
an extension unless this is a nonzero constant polynomial. In particular, 𝛼 extends if
𝛼(𝑎𝑚 ) ≠ 0.

Lemma 9.3. Let 𝐴 and 𝐵 = 𝐴[𝑇]∕𝔞 be affine 𝑘-algebras. Assume that 𝐴 and 𝐵 are integral
domains, and let 𝔠 ⊂ 𝐴 be the ideal of leading coefficients of the polynomials in 𝔞. Then
every homomorphism 𝛼 ∶ 𝐴 → 𝑘 such that 𝛼(𝔠) ≠ 0 extends to a homomorphism 𝐵 → 𝑘.

Proof. If 𝔞 = 0, then every homomorphism 𝛼 extends, and so we may suppose that

𝔞 ≠ 0. Let 𝛼 ∶ 𝐴 → 𝑘 be a homomorphism such that 𝛼(𝔠) ≠ 0, and choose a polynomial
𝑓 = 𝑎𝑚 𝑇 𝑚 + ⋯ + 𝑎0 in 𝔞 of least degree such that 𝛼(𝑎𝑚 ) ≠ 0. Then 𝑚 ≥ 1 otherwise
𝐵 = 0. We shall use induction on 𝑚.
Extend 𝛼 to a homomorphism 𝛼̃ ∶ 𝐴[𝑇] → 𝑘[𝑇] by sending 𝑇 to 𝑇. Then 𝛼(𝔞) ̃
ideal in 𝑘[𝑇].
is an

̃
If 𝛼(𝔞) ≠ 𝑘[𝑇], then it has a zero 𝑐 in 𝑘 (2.11). This means that the homomorphism

𝐴[𝑇] → 𝑘[𝑇] → 𝑘
← 𝛼̃ ℎ↦ℎ(𝑐)
←

is zero on 𝔞, and so it factors through a homomorphism 𝐵 = 𝐴[𝑇]∕𝔞 → 𝑘. This is an

extension of 𝛼 to 𝐵.
̃
If 𝛼(𝔞) = 𝑘[𝑇], then 𝔞 contains a polynomial
𝑔(𝑇) = 𝑏𝑛 𝑇 𝑛 + ⋯ + 𝑏0 , 𝑛 > 0, 𝛼(𝑏𝑛 ) = ⋯ = 𝛼(𝑏1 ) = 0, 𝛼(𝑏0 ) ≠ 0.
On dividing 𝑓(𝑇) into 𝑔(𝑇), we find that
𝑎𝑚
𝑑
𝑔(𝑇) = 𝑞(𝑇)𝑓(𝑇) + 𝑟(𝑇), 𝑑 ∈ ℕ, 𝑞, 𝑟 ∈ 𝐴[𝑇], deg 𝑟 < 𝑚.
On applying 𝛼̃ to this equation, we obtain
̃ 𝛼(𝑓)
𝛼(𝑎𝑚 )𝑑 𝛼(𝑏0 ) = 𝛼(𝑞) ̃ ̃
+ 𝛼(𝑟).
̃
Because 𝛼(𝑓) has degree 𝑚 ≥ 0, we must have 𝛼(𝑞) ̃ ̃
= 0, and so 𝛼(𝑟)
constant. We may replace 𝑔(𝑇) with 𝑟(𝑇), and so suppose that 𝑛 < 𝑚. If 𝑚 = 1, such a
is a nonzero

𝑔(𝑇) cannot exist, so we may suppose that 𝑚 > 1.

For a polynomial ℎ(𝑇) = 𝑐𝑟 𝑇 𝑟 + ⋯ + 𝑐0 , we let ℎ′ (𝑇) = 𝑐𝑟 + ⋯ + 𝑐0 𝑇 𝑟 . The 𝐴-module
generated by the polynomials 𝑇 𝑠 ℎ′ (𝑇) with 𝑠 ∈ ℕ and ℎ ∈ 𝔞 is an ideal 𝔞′ in 𝐴[𝑇]. If
𝔞′ ∩ 𝑘 ≠ {0}, then 𝔞 contains a nonzero polynomial 𝑐𝑇 𝑟 , so 𝐵 is generated over 𝐴 by a
nilpotent, which implies that 𝐴 = 𝐵 (recall that 𝐵 is an integral domain). Otherwise, we
let 𝐵′ = 𝐴[𝑇]∕𝔞′ , and note that 𝔞′ contains the polynomial
𝑔 ′ = 𝑏𝑛 + ⋯ + 𝑏0 𝑇 𝑛 , 𝑛 < 𝑚, 𝛼(𝑏0 ) ≠ 0.
Because deg 𝑔′ < 𝑚, the induction hypothesis implies that 𝛼 extends to a homomorphism
𝐵′ = 𝐴[𝑇]∕𝔞′ → 𝑘. Let 𝑐 denote the image of 𝑇 in 𝑘. Then, for all ℎ = 𝑐𝑟 𝑇 𝑟 + ⋯ + 𝑐0 ∈ 𝔞,
𝛼(𝑐𝑟 ) + ⋯ + 𝛼(𝑐0 )𝑐𝑟 = 0. On applying this with ℎ = 𝑔, we find that 𝑐 = 0. On applying it
with ℎ = 𝑓, we find that 𝛼(𝑎𝑚 ) = 0, which is a contradiction. This completes the proof.2

Lemma 9.4. Let 𝐴 ⊂ 𝐵 be affine 𝑘-algebras, and assume that 𝐴 and 𝐵 are integral do-
mains. For any nonzero 𝑏 ∈ 𝐵, there exists a nonzero 𝑎 ∈ 𝐴 with the following property:
every homomorphism 𝛼 ∶ 𝐴 → 𝑘 from 𝐴 into 𝑘 such that 𝛼(𝑎) ≠ 0 extends to a homomor-
phism 𝛽 ∶ 𝐵 → 𝑘 with 𝛽(𝑏) ≠ 0.
200 9. Regular Maps and Their Fibres

Proof Suppose first that 𝐵 is generated by a single element, say, 𝐵 = 𝐴[𝑥]. Let 𝔞 be the
kernel of the homomorphism 𝑇 ↦ 𝑥, 𝐴[𝑇] → 𝐴[𝑥].
If 𝔞 = (0), write
𝑏 = 𝑓(𝑥) = 𝑎0 𝑥𝑛 + 𝑎1 𝑥𝑛−1 + ⋯ + 𝑎𝑛 , 𝑎𝑖 ∈ 𝐴,
and take 𝑎 = 𝑎0 . If 𝛼 ∶ 𝐴 → 𝑘 is such that 𝛼(𝑎0 ) ≠ 0, then there exists a 𝑐 ∈ 𝑘 such that
∑ ∑
𝑓(𝑐) ≠ 0, and we can take 𝛽 to be the homomorphism 𝑑𝑖 𝑥𝑖 ↦ 𝛼(𝑑𝑖 )𝑐𝑖 .
If 𝔞 ≠ (0), let
𝑓(𝑇) = 𝑎𝑚 𝑇 𝑚 + ⋯ + 𝑎0 , 𝑎𝑚 ≠ 0,
be an element of 𝔞 of smallest possible degree. Let ℎ(𝑇) ∈ 𝐴[𝑇] represent 𝑏. As 𝑏 is
nonzero, ℎ ∉ 𝔞. Because 𝑓 is irreducible over the field of fractions of 𝐴, it and ℎ are
coprime over that field. Hence there exist 𝑢, 𝑣 ∈ 𝐴[𝑇] and 𝑐 ∈ 𝐴 ∖ {0} such that
𝑢ℎ + 𝑣𝑓 = 𝑐.
It follows now that 𝑐𝑎𝑚 satisfies our requirements, for if 𝛼(𝑐𝑎𝑚 ) ≠ 0, then 𝛼 can be
extended to 𝛽 ∶ 𝐵 → 𝑘 by the preceding lemma, and 𝛽(𝑢(𝑥) ⋅ 𝑏) = 𝛽(𝑐) ≠ 0, and so
𝛽(𝑏) ≠ 0.
In the general case, we can write 𝐵 = 𝐴[𝑥1 , … , 𝑥𝑛 ]. There exists an element 𝑏𝑛−1 ∈
𝐴[𝑥1 , … , 𝑥𝑛−1 ] with the following property: every homomorphism 𝛼 ∶ 𝐴[𝑥1 , … , 𝑥𝑛−1 ] →
𝑘 such that 𝛼(𝑏𝑛−1 ) ≠ 0 extends to a homomorphism 𝛽 ∶ 𝐵 → 𝑘 with 𝛽(𝑏) ≠ 0. Then
there exists a 𝑏𝑛−2 ∈ 𝐴[𝑥1 , … , 𝑥𝑛−2 ] etc. Continuing in this fashion, we obtain an element
𝑎 ∈ 𝐴 with the required property. 2

Aside 9.5. For an alternative proof of Theorem 9.1 using the generic flatness theorem, see 9.28
below.

of a constructible set. Let 𝑊 be a noetherian topological space. A subset 𝐶 of 𝑊 is said

In order to generalize 9.1 to regular maps of arbitrary varieties, we need the notion

to constructible if it is a finite union of sets of the form 𝑈 ∩ 𝑍 with 𝑈 open and 𝑍 closed,
⋃
𝐶= 𝑈𝑖 ∩ 𝑍𝑖 .
1≤𝑖≤𝑛

⋃
𝐶′ = 𝑈𝑖′ ∩ 𝑍𝑖′ ,
On passing to the complements, we find that

1≤𝑖≤𝑛

finite intersections of constructible sets are constructible. Obviously, if 𝐶 is constructible

so the complement of a constructible set is constructible. It follows that finite unions and

in 𝑊 and 𝑉 ⊂ 𝑊, then 𝐶 ∩ 𝑉 is constructible in 𝑉, and it is constructible in 𝑊 if 𝑉 is

A constructible subset of 𝔸𝑛 is one that is definable by a finite number of polynomials.

open or closed.

𝑓(𝑋1 , … , 𝑋𝑛 ) = 0, 𝑔(𝑋1 , … , 𝑋𝑛 ) ≠ 0
More precisely, it is defined by a finite number of statements of the form

combined using only “and” and “or” (or, better, statements of the form 𝑓 = 0 combined
using “and”, “or”, and “not”). The next proposition shows that a constructible set 𝐶 that
is dense in an irreducible variety 𝑉 must contain a nonempty open subset of 𝑉. Contrast
ℚ, which is dense in ℝ (real topology), but does not contain an open subset of ℝ, or an
infinite subset of 𝔸1 that omits an infinite set.
 b. The fibres of morphisms 201

Proposition 9.6. Let 𝐶 be a constructible set whose closure 𝐶̄ is irreducible. Then 𝐶

contains a nonempty open subset of its closure 𝐶.̄
⋃
Proof. We are given that 𝐶 = (𝑈𝑖 ∩ 𝑍𝑖 ) with each 𝑈𝑖 open and each 𝑍𝑖 closed. We
⋃
may assume that each set 𝑈𝑖 ∩ 𝑍𝑖 in this decomposition is nonempty. Clearly 𝐶̄ ⊂ 𝑍𝑖 ,
and as 𝐶̄ is irreducible, it must be contained in one of the 𝑍𝑖 . For this 𝑖,

𝑈𝑖 ∩ 𝑍𝑖 ⊃ 𝑈𝑖 ∩ 𝐶̄ ⊃ 𝑈𝑖 ∩ 𝐶 ⊃ 𝑈𝑖 ∩ (𝑈𝑖 ∩ 𝑍𝑖 ) = 𝑈𝑖 ∩ 𝑍𝑖 .

̄ which is a nonempty open subset of 𝐶̄ contained in 𝐶.

Thus 𝑈𝑖 ∩ 𝑍𝑖 = 𝑈𝑖 ∩ 𝐶, 2

Theorem 9.7. Every regular map 𝜑 ∶ 𝑊 → 𝑉 sends constructible sets to constructible

sets.

Proof We first show that it suffices to prove the theorem with 𝑊 and 𝑉 affine. Write
𝑉 as a finite union of open affines, and then write the inverse image of each of the
⋃
affines as a finite union of open affines. In this way, we get 𝑊 = 𝑖∈𝐼 𝑊𝑖 with each 𝑊𝑖
open affine and 𝜑(𝑊𝑖 ) contained in an open affine of 𝑉. If 𝐶 is a constructible subset of
⋃
𝑊, then 𝜑(𝐶) = 𝑖∈𝐼 𝜑(𝐶 ∩ 𝑊𝑖 ), and so 𝜑(𝐶) is constructible if each set 𝜑(𝐶 ∩ 𝑊𝑖 ) is

Now assume that 𝑊 and 𝑉 are affine, and let 𝐶 be a constructible subset of 𝑊. Let 𝑊𝑖
constructible.

be the irreducible components of 𝑊. They are closed in 𝑊, and so 𝐶 ∩ 𝑊𝑖 is constructible

⋃
in 𝑊. As 𝜑(𝑊) = 𝜑(𝐶 ∩ 𝑊𝑖 ), it is constructible if the 𝜑(𝐶 ∩ 𝑊𝑖 ) are. Hence we may
suppose that 𝑊 is irreducible. Moreover, 𝐶 is a finite union of its irreducible components.
As these are closed in 𝐶, they are constructible in 𝑊. We may therefore assume that 𝐶 is
also irreducible; 𝐶̄ is then an irreducible closed subvariety of 𝑊.
We prove the theorem by induction on the dimension of 𝑊. If dim(𝑊) = 0, then
the statement is obvious because 𝑊 is a point. If 𝐶̄ ≠ 𝑊, then dim(𝐶) ̄ < dim(𝑊), and
𝜑(𝐶) is constructible by the induction hypothesis applied to 𝐶̄ ,→ 𝑉. We may therefore
𝜑

assume that 𝐶̄ = 𝑊. Replace 𝑉 with 𝜑(𝐶). According to Proposition 9.6, 𝐶 contains

𝜑
a dense open subset 𝑈 ′ of 𝑊, and Theorem 9.1 applied to 𝑈 ′ ,→ 𝑉 shows that 𝜑(𝐶)
contains a dense open subset 𝑈 of 𝑉. Write

𝜑(𝐶) = 𝑈 ∪ 𝜑(𝐶 ∩ 𝜑−1 (𝑉 ∖ 𝑈)).

Then 𝜑−1 (𝑉 ∖ 𝑈) is a proper closed subset of 𝑊 (the complement of 𝑉 ∖ 𝑈 is dense

in 𝑉 and 𝜑 is dominant). As 𝐶 ∩ 𝜑−1 (𝑉 ∖ 𝑈) is constructible in 𝜑−1 (𝑉 ∖ 𝑈), the set
𝜑(𝐶 ∩ 𝜑−1 (𝑉 ∖ 𝑈)) is constructible in 𝑉 by induction, which completes the proof. 2

Aside 9.8. Let 𝑋 be a subset of ℂ𝑛 . If 𝑋 is constructible for the Zariski topology on ℂ𝑛 , then the
closure of 𝑋 for the Zariski topology is equal to its closure for the complex topology.

b. The fibres of morphisms

We examine the fibres of a regular map 𝜑 ∶ 𝑊 → 𝑉. After replacing 𝑉 with the closure
of the image of 𝜑, we may suppose that 𝜑 is dominant.

Theorem 9.9. Let 𝜑 ∶ 𝑊 → 𝑉 be a dominant map of irreducible algebraic varieties.

(a) dim(𝑊) ≥ dim(𝑉).
202 9. Regular Maps and Their Fibres

(b) If 𝑃 ∈ 𝜑(𝑊), then

dim(𝜑−1 (𝑃)) ≥ dim(𝑊) − dim(𝑉)
with equality holding exactly on a nonempty open subset 𝑈 of 𝑉.
(c) For each 𝑖 ∈ ℕ, the set

𝑉𝑖 = {𝑃 ∈ 𝑉 ∣ dim(𝜑−1 (𝑃)) ≥ 𝑖}

is closed in 𝜑(𝑊).

In other words, for 𝑃 in a dense open subset 𝑈 of 𝑉, the dimension of the fibre 𝜑−1 (𝑃)
has the expected value dim(𝑊) − dim(𝑉), and it jumps on the closed complement of 𝑈
(possibly empty). It may jump further on closed subsets of the closed complement of 𝑈.
Before proving the theorem, we look at an example.

Example 9.10. Consider a system of linear equations

∑
𝑛
𝑎𝑖𝑗 𝑋𝑗 = 0, 𝑖 = 1, … , 𝑚,
𝑗=1

with coefficients in a field 𝑘 (not necessarily algebraically closed). The quotient

𝑘[𝑋1 , … , 𝑋𝑛 ]
∑ ≃ 𝑘[𝑋𝑗1 , … , 𝑋𝑗𝑑 ],
( 𝑎𝑖𝑗 𝑋𝑗 )

where 𝑋𝑗1 , … , 𝑋𝑗𝑑 are the “free” variables for the system of equations (cf. 2.61). Thus the
∑
field of fractions of 𝑘[𝑋1 , … , 𝑋𝑛 ]∕( 𝑎𝑖𝑗 𝑋𝑗 ) has transcendence degree 𝑑 over 𝑘, where
𝑑 = 𝑛 − rank(𝑎𝑖𝑗 ).
Now consider the subvariety 𝑊 ⊂ 𝑉 × 𝔸𝑛 defined by a system of linear equations

∑
𝑛
𝑎𝑖𝑗 𝑋𝑗 = 0, 𝑖 = 1, … , 𝑚,
𝑗=1

with coefficients 𝑎𝑖𝑗 ∈ 𝑘[𝑉]. The projection map 𝜑 ∶ 𝑊 → 𝑉 is surjective, and the above
discussion shows that 𝑘(𝑊) has transcendence degree 𝑑 over 𝑘(𝑉), where

𝑑 = 𝑛 − rank 𝐴, 𝐴 = (𝑎𝑖𝑗 ).
def

dim 𝑊 = dim 𝑉 + 𝑛 − rank(𝐴).

Thus,

The fibre 𝜑−1 (𝑃) over 𝑃 ∈ 𝑉 is the subvariety of 𝔸𝑛 defined by the system of linear

∑
𝑛
𝑎𝑖𝑗 (𝑃)𝑋𝑗 = 0, 𝑖 = 1, … , 𝑚,
equations

𝑗=1

with coefficients in 𝑘. It has dimension

dim 𝜑−1 (𝑃) = 𝑛 − rank 𝐴(𝑃), 𝐴(𝑃) = (𝑎𝑖𝑗 (𝑃)).

Let 𝑟 = rank(𝐴). Then rank 𝐴(𝑃) = 𝑟 on the open subset 𝑈 of 𝑉 where some 𝑟 × 𝑟-minor
of 𝐴 does not vanish, and it drops on the closed complement of 𝑈. Correspondingly,
dim 𝜑−1 (𝑃) = dim 𝑊 − dim 𝑉 on 𝑈, and it jumps on the closed complement of 𝑈.
 b. The fibres of morphisms 203

𝑘(𝑉) → 𝑘(𝑊), and obviously tr deg𝑘 𝑘(𝑉) ≤ tr deg𝑘 𝑘(𝑊) (an algebraically independent
Proof (of Theorem 9.9). (a) Because the map is dominant, there is a homomorphism

subset of 𝑘(𝑉) remains algebraically independent in 𝑘(𝑊)).

(b) In proving the first part of (b), we may replace 𝑉 with an open neighbourhood of
𝑃. In particular, we can assume 𝑉 to be affine. Let 𝑚 and 𝑛 be the dimensions of 𝑉 and
𝑊. From Proposition 3.47 we know that there exist regular functions 𝑓1 , … , 𝑓𝑚 such
that 𝑃 is an irreducible component of 𝑉(𝑓1 , … , 𝑓𝑚 ). After replacing 𝑉 by a smaller open
neighbourhood of 𝑃, we may suppose that 𝑃 = 𝑉(𝑓1 , … , 𝑓𝑚 ). Then 𝜑−1 (𝑃) is the zero
set of the regular functions 𝑓1 ◦𝜑, … , 𝑓𝑚 ◦𝜑, and so (if nonempty) has codimension ≤ 𝑚
in 𝑊 (by 3.45). Hence
dim 𝜑−1 (𝑃) ≥ dim 𝑊 − 𝑚 = dim(𝑊) − dim(𝑉).
In proving the second part of (b), we can replace both 𝑊 and 𝑉 with open affine
subsets. Since 𝜑 is dominant, 𝑘[𝑉] → 𝑘[𝑊] is injective, and we may regard it as an inclu-
sion (we identify a function 𝑥 on 𝑉 with 𝑥◦𝜑 on 𝑊). Then 𝑘(𝑉) ⊂ 𝑘(𝑊), and 𝑘(𝑊) has
transcendence degree dim 𝑊 − dim 𝑉 = 𝑛 − 𝑚 over 𝑘(𝑉). Let 𝑘[𝑉] = 𝑘[𝑥1 , … , 𝑥𝑀 ] and
let 𝑘[𝑊] = 𝑘[𝑦1 , … , 𝑦𝑁 ]. Then {𝑥1 , … , 𝑥𝑀 } contains a transcendence basis for 𝑘(𝑉) over
𝑘, which we may suppose to be {𝑥1 , … , 𝑥𝑚 }. Similarly, after renumbering, we may sup-
pose that {𝑦1 , … , 𝑦𝑛−𝑚 } is a transcendence basis for 𝑘(𝑊) over 𝑘(𝑉). Now {𝑥1 , … , 𝑦𝑛−𝑚 }
is a transcendence basis for 𝑘(𝑊) over 𝑘, and so, for each 𝑖 > 𝑛 − 𝑚, there is a nonzero
polynomial 𝐹𝑖 (𝑋1 , … , 𝑋𝑚 , 𝑌1 , … , 𝑌𝑛−𝑚 , 𝑌𝑖 ) such that
𝐹𝑖 (𝑥1 , … , 𝑥𝑚 , 𝑦1 , … , 𝑦𝑛−𝑚 , 𝑦𝑖 ) = 0.
Let 𝑃 ∈ 𝑉 and let 𝑦̄ 𝑖 denote the restriction of 𝑦𝑖 to 𝜑−1 (𝑃). Then
(37)

𝑘[𝜑−1 (𝑃)] = 𝑘[𝑦̄ 1 , … , 𝑦̄ 𝑁 ].

The equation (37) is an algebraic relation among the functions 𝑥1 , … , 𝑦𝑖 on 𝑊. When
restricted to 𝜑−1 (𝑃), it becomes
𝐹𝑖 (𝑥1 (𝑃), … , 𝑥𝑚 (𝑃), 𝑦̄ 1 , … , 𝑦̄ 𝑛−𝑚 , 𝑦̄ 𝑖 ) = 0.
If this is a nontrivial algebraic relations for all 𝑖, i.e., if none of the polynomials
𝐹𝑖 (𝑥1 (𝑃), … , 𝑥𝑚 (𝑃), 𝑌1 , … , 𝑌𝑛−𝑚 , 𝑌𝑖 )
is the zero polynomial, then tr deg𝑘 (𝑘(𝑦̄ 1 , … , 𝑦̄ 𝑁 ) ≤ 𝑛 − 𝑚, so dim 𝜑−1 (𝑃) ≤ 𝑛 − 𝑚.
Regard 𝐹𝑖 (𝑥1 , … , 𝑥𝑚 , 𝑌1 , … , 𝑌𝑛−𝑚 , 𝑌𝑖 ) as a polynomial in the 𝑌 with coefficients poly-
nomials in the 𝑥. Let 𝑈𝑖 be the open subset of 𝑉 where some coefficient of the polynomial
is nonzero — this is nonempty because 𝐹𝑖 is a nonzero element of 𝑘[𝑉][𝑌1 , … , 𝑌𝑛−𝑚 , 𝑌𝑖 ].
⋂
The last remark shows that, for 𝑃 ∈ 𝑈𝑖 , dim 𝜑−1 (𝑃) ≤ 𝑛 − 𝑚, hence = 𝑛 − 𝑚 by (a).
Finally, if for a particular point 𝑃, dim 𝜑−1 (𝑃) = 𝑛 − 𝑚, then we can modify the

of 𝑃.
above argument to show that the same is true for all points in an open neighbourhood

(c) We prove this by induction on the dimension of 𝑉 — it is obviously true if

dim 𝑉 = 0. We know from (b) that there is an open subset 𝑈 of 𝑉 such that
dim 𝜑−1 (𝑃) = 𝑛 − 𝑚 ⇐⇒ 𝑃 ∈ 𝑈.
Let 𝑍 be the complement of 𝑈 in 𝑉; thus 𝑍 = 𝑉𝑛−𝑚+1 . Let 𝑍1 , … , 𝑍𝑟 be the irreducible
components of 𝑍. On applying the induction to the restriction of 𝜑 to the map 𝜑−1 (𝑍𝑗 ) →
𝑍𝑗 for each 𝑗, we obtain the result. 2
204 9. Regular Maps and Their Fibres

Recall that a regular map 𝜑 ∶ 𝑊 → 𝑉 of algebraic varieties is closed if, for example,
𝑊 is complete (7.7).
Proposition 9.11. Let 𝜑 ∶ 𝑊 → 𝑉 be a regular surjective closed map of varieties, and let
𝑛 ∈ ℕ. If 𝑉 is irreducible and all fibres 𝜑−1 (𝑃) of 𝜑 are irreducible of dimension 𝑛, then 𝑊
is irreducible of dimension dim(𝑉) + 𝑛.

Proof. Let 𝑍 be an irreducible closed subset of 𝑊, and consider the map 𝜑|𝑍 ∶ 𝑍 → 𝑉;
it has fibres (𝜑|𝑍)−1 (𝑃) = 𝜑−1 (𝑃) ∩ 𝑍. There are three possibilities.
(a) 𝜑(𝑍) ≠ 𝑉. Then 𝜑(𝑍) is a proper closed subset of 𝑉.
(b) 𝜑(𝑍) = 𝑉, dim(𝑍) < 𝑛 + dim(𝑉). Then (b) of (9.9) shows that there is a nonempty
open subset 𝑈 of 𝑉 such that for 𝑃 ∈ 𝑈,

dim(𝜑−1 (𝑃) ∩ 𝑍) = dim(𝑍) − dim(𝑉) < 𝑛.

Thus, for 𝑃 ∈ 𝑈, the fibre 𝜑−1 (𝑃) is not contained in 𝑍.

(c) 𝜑(𝑍) = 𝑉, dim(𝑍) ≥ 𝑛 + dim(𝑉). Then 9.9(b) shows that

dim(𝜑−1 (𝑃) ∩ 𝑍) ≥ dim(𝑍) − dim(𝑉) ≥ 𝑛

for all 𝑃; thus 𝜑−1 (𝑃) ⊂ 𝑍 for all 𝑃 ∈ 𝑉, and so 𝑍 = 𝑊; moreover dim 𝑍 =
dim 𝑉 + 𝑛.
Now let 𝑍1 , … , 𝑍𝑟 be the irreducible components of 𝑊. I claim that (c) holds for
at least one of the 𝑍𝑖 . Otherwise, there will be an open subset 𝑈 of 𝑉 such that for
𝑃 in 𝑈, 𝜑−1 (𝑃) is contained in none of the 𝑍𝑖 ; but 𝜑−1 (𝑃) is irreducible and 𝜑−1 (𝑃) =
⋃ −1

☡ It is possible for all the fibres of regular map 𝑊 → 𝑉 to be reducible without 𝑊

(𝜑 (𝑃) ∩ 𝑍𝑖 ), and so this is impossible. 2

being reducible. The subvariety of 𝔸2 × 𝔸2 with equation 𝑥12 𝑦1 − 𝑥22 𝑦2 = 0 is irreducible,

but the fibres of the projection to the first factor (obtained by fixing the values of 𝑦1 and
𝑦2 ) are all reducible. To extend this to ℙ2 × ℙ2 , pass to the projective closure.

c. Flat maps and their fibres

A flat map is the algebraic analogue of a map whose fibres form a contin-
uously varying family. For example, a surjective regular map of smooth
varieties is flat if and only if all fibres have the same dimension. A finite
map is flat if and only if, over every connected component, all fibres have
the same number of points (counting multiplicities). Flat maps of algebraic
varieties are open.

Flat homomorphisms of rings

Let 𝐴 be a ring and 𝐵 an 𝐴-algebra. If the sequence of 𝐴-modules
𝛼 𝛽
0 → 𝑁 ′ ,→ 𝑁 ,→ 𝑁 ′′ → 0

is exact, then the sequence of 𝐵-modules

𝐵 ⊗𝐴 𝑁 ′ → 𝐵 ⊗𝐴 𝑁 → 𝐵 ⊗𝐴 𝑁 ′′ → 0
1⊗𝛼
← ← 1⊗𝛽 ←
(*)
 c. Flat maps and their fibres 205

is exact, but 𝐵⊗𝐴 𝑁 ′ → 𝐵⊗𝐴 𝑁 need not be injective. For example, when we tensor the
exact sequence of 𝑘[𝑋]-modules

𝑓↦𝑋⋅𝑓 𝑓↦𝑓 mod (𝑋)

0 → 𝑘[𝑋] ,,,,,,→ 𝑘[𝑋] ,,,,,,,,,,,,→ 𝑘[𝑋]∕(𝑋) → 0

with 𝑘, we get the sequence

0 id
𝑘 ,→ 𝑘 ,→ 𝑘 → 0.
We prove that (*) is exact. The surjectivity of 1 ⊗ 𝛽 is obvious. Let 𝑞 ∶ 𝐵 ⊗𝐴 𝑁 → 𝑄
be the cokernel of 1 ⊗ 𝛼. As (1 ⊗ 𝛽)◦(1 ⊗ 𝛼) = 1 ⊗ (𝛽◦𝛼) = 0, the map 1 ⊗ 𝛽 factors
through 𝑞,
𝐵 ⊗𝐴 𝑁 ′ → 𝐵 ⊗𝐴 𝑁 → 𝐵 ⊗𝐴 𝑁 ′′ → 0.
← 1⊗𝛼 ← 1⊗𝛽 ←

𝑓
𝑞 →
←

→
𝑄

←
We construct an inverse 𝑔 to 𝑓. If 𝑛1 , 𝑛2 ∈ 𝑁 have the same image in 𝑁 ′′ , then they
differ by an element of 𝛼(𝑁 ′ ), and so 𝜙(𝑏 ⊗ 𝑛1 ) = 𝜙(𝑏 ⊗ 𝑛2 ) for all 𝑏 ∈ 𝐵. Hence the
𝐴-bilinear map
𝐵 × 𝑁 → 𝑄, (𝑏, 𝑛) ↦ 𝜙(𝑏 ⊗ 𝑛)
factors through 𝐵 × 𝑁 ′′ , and so defines an 𝐴-linear map 𝑔 ∶ 𝐵 ⊗𝐴 𝑁 ′′ → 𝑄. This is
inverse to 𝑓.

Definition 9.12. An 𝐴-algebra 𝐵 is flat if

𝑀 → 𝑁 injective ⇐⇒ 𝐵 ⊗𝐴 𝑀 → 𝐵 ⊗𝐴 𝑁 injective.

It is faithfully flat if, in addition,

𝐵 ⊗𝐴 𝑀 = 0 ⇐⇒ 𝑀 = 0.

A homomorphism 𝛼 ∶ 𝐴 → 𝐵 of rings is flat (resp. faithfully flat) if it makes 𝐵 into a

flat (resp. faithfully flat) algebra.

Therefore, an 𝐴-algebra 𝐵 is flat if and only if the functor 𝑀 ⇝ 𝐵 ⊗𝐴 𝑀 from

𝐴-modules to 𝐵-modules is exact.

Example 9.13. If 𝑆 is a multiplicative subset of 𝐴, then 𝑆 −1 𝐴 is a flat 𝐴-algebra (1.18).

exact, an 𝐴-algebra is flat if it is free as an 𝐴-module (and faithfully flat if it also nonzero).
As tensor products commute with direct sums, and direct sums of exact sequences are

Proposition 9.14. Let 𝐴 → 𝐴′ be a homomorphism of rings. If 𝐴 → 𝐵 is flat, then so

also is 𝐴′ → 𝐵 ⊗𝐴 𝐴′ .

Proof. For any 𝐴′ -module 𝑀,

(𝐵 ⊗𝐴 𝐴′ ) ⊗𝐴′ 𝑀 ≃ 𝐵 ⊗𝐴 (𝐴′ ⊗𝐴′ 𝑀) ≃ 𝐵 ⊗𝐴 𝑀.

In other words, tensoring an 𝐴′ -module 𝑀 with 𝐵 ⊗𝐴 𝐴′ is the same as tensoring 𝑀

(regarded as an 𝐴-module) with 𝐵. Therefore it preserves exact sequences. 2
206 9. Regular Maps and Their Fibres

Proposition 9.15. Let 𝛼 ∶ 𝐴 → 𝐵 be a homomorphism of rings. If 𝛼 ∶ 𝐴 → 𝐵 is flat, then

𝐴𝛼−1 (𝔮) → 𝐵𝔮 is flat for all prime ideals 𝔮 of 𝐵; conversely, 𝛼 ∶ 𝐴 → 𝐵 is flat if 𝐴𝛼−1 (𝔫) → 𝐵𝔫
is flat for all maximal ideals 𝔫 of 𝐵

Proof. Let 𝔮 be a prime ideal of 𝐵, and let 𝔭 = 𝛼−1 (𝔮) — it is a prime ideal in 𝐴. If
𝐴 → 𝐵 is flat, then 𝐴𝔭 → 𝐴𝔭 ⊗𝐴 𝐵 ≃ 𝑆𝔭−1 𝐵 (9.14). The map 𝑆𝔭−1 𝐵 → 𝑆𝔮−1 𝐵 = 𝐵𝔮 is flat
(9.13a), and so the composite 𝐴𝔭 → 𝐵𝔮 is flat (9.13c).
For the converse, let 𝑁 ′ → 𝑁 be an injective homomorphism of 𝐴-modules, and let
𝔫 be a maximal ideal of 𝐵. Then 𝐴𝔪 ⊗𝐴 (𝑁 ′ → 𝑁) is injective (9.13). Therefore, the map

𝐵𝔫 ⊗𝐴 (𝑁 ′ → 𝑁) ≃ 𝐵𝔫 ⊗𝐴𝔪 (𝐴𝔪 ⊗𝐴 (𝑁 ′ → 𝑁))

is injective, and so the kernel 𝑀 of 𝐵 ⊗𝐴 (𝑁 ′ → 𝑁) has the property that 𝑀𝔫 = 0. Let

𝑥 ∈ 𝑀, and let 𝔞 = {𝑏 ∈ 𝐵 ∣ 𝑏𝑥 = 0}. For each maximal ideal 𝔫 of 𝐵, 𝑥 maps to zero in
𝑀𝔫 , and so 𝔞 contains an element not in 𝔫. Hence 𝔞 = 𝐵, and so 𝑥 = 0. 2

Proposition 9.16. A flat homomorphism 𝛼 ∶ 𝐴 → 𝐵 is faithfully flat if and only if every

maximal ideal 𝔪 of 𝐴 is of the form 𝛼−1 (𝔫) for some maximal ideal 𝔫 of 𝐵, i.e., if and only

spm(𝛼) ∶ spm(𝐵) → spm(𝐴)

if the map

is surjective.

Proof. ⇒: Let 𝔪 be a maximal ideal of 𝐴, and let 𝑀 = 𝐴∕𝔪; then

𝐵 ⊗𝐴 𝑀 ≃ 𝐵∕𝛼(𝔪)𝐵.

As 𝐵 ⊗𝐴 𝑀 ≠ 0, we see that 𝛼(𝔪)𝐵 ≠ 𝐵. Therefore 𝛼(𝔪) is contained in a maximal

ideal 𝔫 of 𝐵. Now 𝛼−1 (𝔫) is a proper ideal in 𝐴 containing 𝔪, and hence equals 𝔪.
⇐: Let 𝑀 be a nonzero 𝐴-module. Let 𝑥 be a nonzero element of 𝑀, and let 𝔞 =
ann(𝑥) = {𝑎 ∈ 𝐴 ∣ 𝑎𝑥 = 0}. Then 𝔞 is an ideal in 𝐴, and 𝑀 ′ = 𝐴𝑥 ≃ 𝐴∕𝔞. Moreover,
𝐵 ⊗𝐴 𝑀 ′ ≃ 𝐵∕𝛼(𝔞) ⋅ 𝐵 and, because 𝐴 → 𝐵 is flat, 𝐵 ⊗𝐴 𝑀 ′ is a submodule of 𝐵 ⊗𝐴 𝑀.
def def

Because 𝔞 is proper, it is contained in a maximal ideal 𝔪 of 𝐴, and therefore

𝛼(𝔞) ⊂ 𝛼(𝔪) ⊂ 𝔫

for some maximal ideal 𝔫 of 𝐴. Hence 𝛼(𝔞)⋅𝐵 ⊂ 𝔫 ≠ 𝐵, and so 𝐵 ⊗𝐴 𝑀 ⊃ 𝐵 ⊗𝐴 𝑀 ′ ≠ 0.2

Corollary 9.17. A flat local homomorphism 𝐴 → 𝐵 of local rings is faithfully flat.

Proof. Let 𝔪 and 𝔫 be the (unique) maximal ideals of 𝐴 and 𝐵. By hypothesis, 𝔫𝑐 = 𝔪,

and so the statement follows from the proposition. 2

Properties of flat homomorphisms of rings

Lemma 9.18. Let 𝐵 be an 𝐴-algebra, and let 𝔭 be a prime ideal of 𝐴. The prime ideals
of 𝐵 contracting to 𝔭 are in natural one-to-one correspondence with the prime ideals of
𝐵 ⊗𝐴 𝜅(𝔭).

Proof. Let 𝑆 = 𝐴 ∖ 𝔭. Then 𝜅(𝔭) = 𝑆 −1 (𝐴∕𝔭). Therefore we obtain 𝐵 ⊗𝐴 𝜅(𝔭) from

𝐵 by first passing to 𝐵∕𝔭𝐵 and then making the elements of 𝐴 not in 𝔭 act invertibly.
def

After the first step, we are left with the prime ideals 𝔮 of 𝐵 such that 𝔮𝑐 ⊃ 𝔭, and after
the second step only with those such that 𝔮𝑐 ∩ 𝑆 = ∅, i.e., such that 𝔮𝑐 = 𝔭. 2
 c. Flat maps and their fibres 207

Proposition 9.19. Let 𝐵 be a faithfully flat 𝐴-algebra. Every prime ideal 𝔭 of 𝐴 is of the
form 𝔮𝑐 for some prime ideal 𝔮 of 𝐵.

Proof. The ring 𝐵 ⊗𝐴 𝜅(𝔭) is not zero, because 𝜅(𝔭) ≠ 0 and 𝐴 → 𝐵 is faithfully flat,
and so it has a prime (even maximal) ideal 𝔮. For this ideal, 𝔮𝑐 = 𝔭.

Summary 9.20. A flat homomorphism 𝛼 ∶ 𝐴 → 𝐵 is faithfully flat if the image of

2

spec(𝛼) ∶ spec(𝐵) → spec(𝐴)

includes all maximal ideals of 𝐴, in which case it includes all prime ideals of 𝐴.

Theorem 9.21 (Going-down theorem for flat maps). Let 𝐵 be a flat 𝐴-algebra.
Let 𝔭 ⊃ 𝔭′ be prime ideals in 𝐴, and let 𝔮 be a prime ideal in 𝐵 such that 𝔮𝑐 = 𝔭. Then 𝔮
contains a prime ideal 𝔮′ such that 𝔮′𝑐 = 𝔭′ :

𝐵 𝔮 ⊃ 𝔮′

𝐴 𝔭 ⊃ 𝔭′ .
Proof. Because 𝐴 → 𝐵 is flat, the homomorphism 𝐴𝔭 → 𝐵𝔮 is flat, and because
𝔭𝐴𝔭 = (𝔮𝐵𝔮 )𝑐 , it is faithfully flat (9.16). The ideal 𝔭′ 𝐴𝔭 is prime (1.14), and so there
exists a prime ideal of 𝐵𝔮 lying over 𝔭′ 𝐴𝔭 (by 9.19). The contraction of this ideal to 𝐵 is
contained in 𝔮 and contracts to 𝔭′ in 𝐴.

For example, let 𝛼 ∶ 𝐴 → 𝐵 be a flat local homomorphism of local rings. By definition,

2

𝛼−1 (𝔫) = 𝔪, where 𝔪 and 𝔫 are the maximal ideals of 𝐴 and 𝐵. The theorem says that,
for every prime ideal 𝔭 in 𝐴, there exists a prime ideal 𝔮 in 𝐵 such that 𝛼−1 (𝔮) = 𝔭, i.e.,

Spec 𝐵 → Spec 𝐴
the map

is surjective.

Flat maps of algebraic varieties

Definition 9.22. A regular map 𝜑 ∶ 𝑊 → 𝑉 of algebraic varieties is flat if, for all
𝑃 ∈ 𝑊, the local homomorphism 𝒪𝑉,𝜑(𝑃) → 𝒪𝑊,𝑃 is flat, and it is faithfully flat if it is
flat and surjective.

Proposition 9.23. A regular map 𝜑 ∶ 𝑊 → 𝑉 of affine algebraic varieties is flat (resp.

Open immersions are flat and composites of flat maps are flat.

faithfully flat) if and only if the map 𝑓 ↦ 𝑓◦𝜑 ∶ 𝑘[𝑉] → 𝑘[𝑊] is flat (resp. faithfully flat).

Theorem 9.24. Let 𝜑 ∶ 𝑊 → 𝑉 be a flat map of affine algebraic varieties. Let 𝑆 ′ ⊃ 𝑆 be

Proof. Apply 9.15 and 9.16. 2

closed irreducible subsets of 𝑉, and let 𝑇 be a closed irreducible subset of 𝑊 such that 𝜑(𝑇)
is a dense subset of 𝑆. Then there exists a closed irreducible subset 𝑇 ′ of 𝑊 containing 𝑇
and such that 𝜑(𝑇 ′ ) is a dense subset of 𝑆 ′ :

𝑊 ⊃ 𝑇′ ⊃ 𝑇
𝜑 𝜑 𝜑

𝑉 ⊃ 𝑆′ ⊃ 𝑆.
208 9. Regular Maps and Their Fibres

als (2.28), this is just a geometric restatement of Theorem 9.21. Let 𝔭 = 𝐼(𝑆), 𝔭′ = 𝐼(𝑆 ′ ),
Proof. In view of the correspondence between closed irreducible subsets and prime ide-

and 𝔮 = 𝐼(𝑇). Then 𝔭 ⊃ 𝔭′ because 𝑆 ⊂ 𝑆 ′ . As 𝑇 → 𝑆 is dominant, the homomorphism

𝑘[𝑆] = 𝑘[𝑉]∕𝔭 → 𝑘[𝑇]∕𝔮 = 𝑘[𝑇]

is injective (2.40), and so 𝔮𝑐 = 𝔭. According to Theorem 9.21, there exists a prime ideal
𝔮′ in 𝑘[𝑊] contained in 𝔮 and such that 𝔮′𝑐 = 𝔭′ . Now 𝑇 ′ = 𝑉(𝔮′ ) has the required
def

properties. 2

Corollary 9.25. Let 𝜑 ∶ 𝑊 → 𝑉 be a flat map of algebraic varieties. Let 𝑤 ∈ 𝑊 and

let 𝑣 = 𝜑(𝑤). For any closed irreducible subset 𝑆 of 𝑉 through 𝑣, there exists a closed
𝜑
irreducible subset 𝑇 through 𝑤 such that 𝜑(𝑇) ⊂ 𝑆 and the map 𝑇 ,→ 𝑆 is dominant.

Proof. When 𝑊 and 𝑉 are affine, this is a special case of the theorem. The general case
can be proved by replacing 𝑊 and 𝑉 with suitable affine neighbourhoods of 𝑤 and 𝑣.2

Theorem 9.26 (Generic flatness). Let 𝜑 ∶ 𝑊 → 𝑉 be a dominant map of irreducible

algebraic varieties. There exist nonempty open subsets 𝑈 ⊂ 𝑉 and 𝑈 ′ ⊂ 𝑊 such that
𝜑
𝜑(𝑈 ′ ) ⊂ 𝑈 and 𝑈 ′ ,→ 𝑈 is faithfully flat.

Proof. In the proof we keep replacing 𝑊 and 𝑉 with smaller nonempty open subvari-
eties until they have the required property. First, we may replace 𝑊 and 𝑉 with open
affines, so that 𝜑 is the map Spm(𝐵) → Spm(𝐴) defined by a homomorphism 𝐴 → 𝐵
of integral domains (finitely generated over 𝑘). Let 𝐹 be the field of fractions of 𝐴, and
regard 𝐵 as a subring of 𝐹 ⊗𝐴 𝐵.
As 𝐹 ⊗𝐴 𝐵 is a finitely generated 𝐹-algebra, it contains elements 𝑥1 , … , 𝑥𝑚 alge-
braically independent over 𝐹 and such that 𝐹 ⊗𝐴 𝐵 is a finite 𝐹[𝑥1 , … , 𝑥𝑚 ]-algebra (2.45).
After multiplying each 𝑥𝑖 by an element of 𝐴, we may suppose that it lies in 𝐵. Let
𝑏1 , … , 𝑏𝑛 generate 𝐵 as an 𝐴-algebra. Each 𝑏𝑖 satisfies a monic polynomial equation with
coefficients in 𝐹[𝑥1 , … , 𝑥𝑚 ]. If 𝑎 ∈ 𝐴 is a common denominator for the coefficients of
these polynomials, then each 𝑏𝑖 is integral over 𝐴𝑎 [𝑥1 , … , 𝑥𝑚 ]. As the 𝑏𝑖 generate 𝐵𝑎 as
an 𝐴𝑎 -algebra, this shows that 𝐵𝑎 is a finite 𝐴𝑎 [𝑥1 , … , 𝑥𝑚 ]-algebra (1.36). After replacing
𝐴 with 𝐴𝑎 and 𝐵 with 𝐵𝑎 , we may suppose that 𝐵 is a finite 𝐴[𝑥1 , … , 𝑥𝑚 ]-algebra. We
have constructed the left-hand part of the diagram,

𝐵 → 𝐹 ⊗𝐴 𝐵 → 𝐸 ⊗𝐴[𝑥1 ,…,𝑥𝑚 ] 𝐵
← injective ←

f inite f inite f inite

→

𝐴[𝑥1 , … , 𝑥𝑚 ] → 𝐹[𝑥1 , … , 𝑥𝑚 ] → 𝐸 = 𝐹(𝑥1 , … , 𝑥𝑚 )

←
←

← ← def
→

𝐴 → 𝐹.
←

and we now construct the rest. Let 𝐸 = 𝐹(𝑥1 , … , 𝑥𝑚 ) be the field of fractions of
𝐹[𝑥1 , … , 𝑥𝑚 ]. It is also the field of fractions of 𝐴[𝑥1 , … , 𝑥𝑚 ]. Let 𝑒1 , … , 𝑒𝑟 be elements of
𝐵 forming a basis for 𝐸 ⊗𝐴[𝑥1 ,…,𝑥𝑚 ] 𝐵 as an 𝐸-vector space. Each element of 𝐵 can be
expressed as a linear combination of the 𝑒𝑖 with coefficients in 𝐸. Let 𝑞 be a common
 c. Flat maps and their fibres 209

denominator for the coefficients arising from a set of generators for 𝐵 as an 𝐴[𝑥1 , … , 𝑥𝑚 ]-
module. Then 𝑒1 , … , 𝑒𝑟 generate 𝐵𝑞 as an 𝐴[𝑥1 , … , 𝑥𝑚 ]𝑞 -module. In other words, the
∑
(𝑐1 , … , 𝑐𝑟 ) ↦ 𝑐𝑖 𝑒𝑖 ∶ 𝐴[𝑥1 , … , 𝑥𝑚 ]𝑟𝑞 → 𝐵𝑞
map

is surjective. This map becomes an isomorphism when tensored with 𝐸 over 𝐴[𝑥1 , … , 𝑥𝑚 ]𝑞 .
(*)

Thus its kernel 𝑀 is an 𝐴[𝑥1 , … , 𝑥𝑚 ]𝑞 -submodule that becomes zero when tensored with
the field of fractions of 𝐴[𝑥1 , … , 𝑥𝑚 ]𝑞 . As 𝐴[𝑥1 , … , 𝑥𝑚 ]𝑞 is an integral domain, 𝑀 = 0,
and so (*) is an isomorphism. Let 𝑎 be some nonzero coefficient of the polynomial 𝑞,
and consider the homomorphisms

𝐴𝑎 → 𝐴𝑎 [𝑥1 , … , 𝑥𝑚 ] → 𝐴𝑎 [𝑥1 , … , 𝑥𝑚 ]𝑞 → 𝐵𝑎𝑞 .

The first and third arrows are faithfully flat because their targets are free modules over
their sources, and the second arrow is flat because it is a localization (9.13). In sum, we

𝐷(𝑎𝑞) = Spm(𝐵𝑎𝑞 ) → Spm(𝐴𝑎 ) = 𝐷(𝑎)

have a flat map

from an open subvariety of 𝑊 to an open subvariety of 𝑉.

Let 𝔪 be a maximal ideal in 𝐴𝑎 . Then 𝔪𝐴𝑎 [𝑥1 , … , 𝑥𝑚 ] does not contain the poly-
nomial 𝑞 because the coefficient 𝑎 of 𝑞 is invertible in 𝐴𝑎 . Hence 𝔪𝐴𝑎 [𝑥1 , … , 𝑥𝑚 ]𝑞 is a
proper ideal of 𝐴𝑎 [𝑥1 , … , 𝑥𝑚 ]𝑞 . Any maximal ideal of 𝐴𝑎 [𝑥1 , … , 𝑥𝑚 ]𝑞 containing it will
intersect 𝐴𝑎 in 𝔪, and so the map 𝐴𝑎 → 𝐴𝑎 [𝑥1 , … , 𝑥𝑚 ]𝑞 is faithfully flat by 9.16. Hence
𝐴𝑎 → 𝐵𝑎𝑞 is faithfully flat, which completes the proof. 2

Theorem 9.27. Every flat map 𝜑 ∶ 𝑊 → 𝑉 of algebraic varieties is open.

Proof. It suffices to show that 𝜑(𝑊) is open. Let 𝑊 = 𝑉 ∖ 𝜑(𝑉), and let 𝑍1 , … , 𝑍𝑛 be
the irreducible components of the closure 𝑊 of 𝑊. It suffices to show that 𝑊 contains
every 𝑍𝑖 . Suppose not, and let 𝑣 ∈ 𝑍𝑗 ∩ 𝜑(𝑊). Then 𝑣 = 𝜑(𝑤) for some 𝑤 ∈ 𝑊, and
according to 9.25 there exists a closed irreducible subset 𝑆 of 𝑊 through 𝑤 such that the
map 𝑆 → 𝑍𝑗 is dominant. This means that there exists an open subset 𝑈 of 𝑉 such that
𝜑(𝑊) ⊃ 𝑈 ∩ 𝑍𝑗 ≠ ∅.
⋃ ⋃
Let 𝑈 ′ = 𝑉 ∖ 𝑖≠𝑗 𝑍𝑖 . As 𝑉 = 𝑖 𝑍𝑖 ∪ 𝜑(𝑊), we have 𝑈 ′ ⊂ 𝑍𝑗 ∪ 𝜑(𝑊), and so

𝜑(𝑊) ⊃ 𝑈 ∩ 𝑈 ′ .

Note that 𝑈 and 𝑈 ′ are both open subsets of 𝑉 meeting 𝑍𝑗 . As 𝑍𝑗 is irreducible, 𝑈 ∩ 𝑍𝑗

and 𝑈 ′ ∩ 𝑍𝑗 are both dense open subsets of 𝑍𝑗 . Hence 𝑈 ∩ 𝑈 ′ ∩ 𝑍𝑗 is nonempty. As
its elements are not in the closure of 𝑊, this contradicts the definition of 𝑍𝑗 . We have
shown that all 𝑍𝑖 are contained in 𝑊, as required. 2

Corollary 9.28. Let 𝜑 ∶ 𝑊 → 𝑉 be a dominant map of irreducible algebraic varieties.

There exists a dense open subset 𝑈 of 𝑊 such that 𝜑(𝑈) is open, 𝑈 = 𝜑−1 (𝜑𝑈), and
𝜑
𝑈 ,→ 𝜑(𝑈) is flat.
𝜑
Proof. According to 9.26, there exists a dense open subset 𝑈 of 𝑉 such that 𝜑−1 (𝑈) ,→
𝑈 is flat. In particular, 𝜑(𝜑−1 (𝑈)) is open in 𝑉 (9.27). Note that 𝜑−1 (𝜑(𝜑−1 (𝑈)) = 𝜑−1 (𝑈).
Let 𝑈 ′ = 𝜑−1 (𝑈). Then 𝑈 ′ is a dense open subset of 𝑊, 𝜑(𝑈 ′ ) is open, 𝑈 ′ = 𝜑−1 (𝜑𝑈 ′ ),
𝜑
and 𝑈 ′ ,→ 𝜑(𝑈 ′ ) is flat. 2
210 9. Regular Maps and Their Fibres

Fibres and flatness

Proposition 9.29. Let 𝜑 ∶ 𝑊 → 𝑉 be a dominant map of irreducible algebraic varieties,

The notion of flatness allows us to sharpen our earlier results.

and let 𝑃 ∈ 𝜑(𝑊). Then

( )
dim 𝜑−1 (𝑃) ≥ dim(𝑊) − dim(𝑉) (38)

with equality if 𝜑 is flat.

Proof. The inequality was proved in 9.9. Assume that 𝜑 is flat, and let 𝑍 be an irre-
ducible component of 𝜑−1 (𝑃).
After replacing 𝑉 with an open neighbourhood of 𝑃 and 𝑊 with an open subset
intersecting 𝑍, we may suppose that both 𝑉 and 𝑊 are affine. Let

𝑉 ⊃ 𝑉1 ⊃ ⋯ ⊃ 𝑉𝑚 = {𝑃}

be a maximal chain of distinct irreducible closed subsets of 𝑉 (so 𝑚 = dim(𝑉) by 3.44).

Now 𝜑(𝑍) = {𝑃}, and so, by Theorem 9.24, there exists a chain of irreducible closed

𝑊 ⊃ 𝑊1 ⊃ ⋯ ⊃ 𝑊𝑚 = 𝑍
subsets

such that 𝜑(𝑊𝑖 ) is a dense subset of 𝑉𝑖 . Let

𝑍 ⊃ 𝑍1 ⊃ ⋯ ⊃ 𝑍𝑛

be a maximal chain of distinct irreducible closed subsets of 𝑉 (so 𝑛 = dim(𝑍)). The

existence of the chain

𝑊 ⊃ 𝑊1 ⊃ ⋯ ⊃ 𝑊𝑚 ⊃ 𝑍1 ⊃ ⋯ ⊃ 𝑍𝑛

dim(𝑊) ≥ 𝑚 + 𝑛 = dim(𝑉) + dim(𝑍).

shows that

Together with the inequality (38), this gives the equality. 2

Proposition 9.30. Let 𝜑 ∶ 𝑊 → 𝑉 be a dominant map of irreducible algebraic varieties.

There exists a dense open subset 𝑈 of 𝑊 such that 𝜑(𝑈) is open in 𝑉, 𝑈 = 𝜑−1 (𝜑(𝑈)), and
( )
dim 𝜑−1 (𝑃) = dim(𝑊) − dim(𝑉).

for all 𝑃 ∈ 𝜑(𝑈).

Proof. According to Proposition 9.29, the open subset 𝑈 of 𝑊 in 9.28 has these proper-
ties. 2

Proposition 9.31. Let 𝜑 ∶ 𝑊 → 𝑉 be a dominant map of irreducible varieties. Let 𝑆 be

a closed irreducible subset of 𝑉, and let 𝑇 be an irreducible component of 𝜑−1 (𝑆) such that
𝜑(𝑇) is dense in 𝑆. Then

dim(𝑇) ≥ dim(𝑆) + dim(𝑊) − dim(𝑉)

with equality if 𝜑 is flat.

 c. Flat maps and their fibres 211

codim(𝑆) ≥ codim(𝑇),
In other words,

with equality if 𝜑 is flat.

Proof. When 𝑆 is a point, this becomes 9.9(b) and 9.29. As we now explain, the general

In proving the inequality, we may replace 𝑉 with any open subvariety intersecting 𝑆.
case can be proved by an easy modification of the proofs of those statements.

In particular, we can assume 𝑉 to be affine. Let 𝑚 = codim(𝑆). From Proposition 3.47 we

know that there exist regular functions 𝑓1 , … , 𝑓𝑚 such that 𝑆 is an irreducible component
of 𝑉(𝑓1 , … , 𝑓𝑚 ). After replacing 𝑉 by a smaller open subset, we may suppose that
𝑆 = 𝑉(𝑓1 , … , 𝑓𝑚 ). Then 𝜑−1 (𝑆) is the zero set of the regular functions 𝑓1 ◦𝜑, … , 𝑓𝑚 ◦𝜑,
and every irreducible component has codimension ≤ 𝑚 in 𝑊 by 3.45.
When 𝜑 is flat, we shall prove (more precisely) that, if 𝑍 is an irreducible component
of 𝜑−1 (𝑆), then
dim(𝑍) = dim(𝑆) + dim(𝑊) − dim(𝑉).
After replacing 𝑉 (resp. 𝑊) with an open subvariety that intersects 𝑆 (resp. 𝑍), we may
suppose that both 𝑉 and 𝑊 are affine. Let

𝑉 ⊃ 𝑉1 ⊃ ⋯ ⊃ 𝑉𝑚 = {𝑆}

be a maximal chain of distinct irreducible closed subsets (so 𝑚 = codim(𝑆) by 3.44).

Now 𝜑(𝑍) is a dense subset of 𝑆, and so (see 9.24) there exists a chain of irreducible

𝑊 ⊃ 𝑊1 ⊃ ⋯ ⊃ 𝑊𝑚 = 𝑍
closed subsets

such that 𝜑(𝑊𝑖 ) is a dense subset of 𝑉𝑖 . Let

𝑍 ⊃ 𝑍1 ⊃ ⋯ ⊃ 𝑍𝑛

be a maximal chain of distinct irreducible closed subsets of 𝑉 (so 𝑛 = dim(𝑍)). The

existence of the chain

𝑊 ⊃ 𝑊1 ⊃ ⋯ ⊃ 𝑊𝑚 ⊃ 𝑍1 ⊃ ⋯ ⊃ 𝑍𝑛

dim(𝑊) ≥ 𝑚 + 𝑛 = dim(𝑉) − dim 𝑆 + dim(𝑍),

shows that

codim(𝑍) ≥ codim(𝑆).
i.e.,

Together with the previous inequality, this implies that codim(𝑍) = codim(𝑆). 2

Proposition 9.32. Let 𝜑 ∶ 𝑊 → 𝑉 be a dominant map of irreducible algebraic varieties.

There exists a dense open subset 𝑈 of 𝑊 such that 𝜑(𝑈) is open in 𝑉, 𝑈 = 𝜑−1 (𝜑𝑈), and

dim(𝑇) = dim(𝑆) + dim(𝑊) − dim(𝑉)

for all closed irreducible subsets 𝑆 of 𝑉 intersecting 𝜑(𝑈) and all irreducible components 𝑇
of 𝜑−1 (𝑆) intersecting 𝑈.

Proof. According to Proposition 9.31, the open subset 𝑈 in 9.28 has these properties.2
212 9. Regular Maps and Their Fibres

Proposition 9.33. Let 𝑉 be an irreducible algebraic variety. A finite map 𝜑 ∶ 𝑊 → 𝑉 is

Finite maps

∑
dim𝑘 𝒪𝑄 ∕𝔪𝑃 𝒪𝑄
flat if and only if

𝑄↦𝑃

is independent of 𝑃 ∈ 𝑉.

Proof. It suffices to prove this with 𝑉 affine, in which case it follows from CA, 12.6
(equivalence of (d) and (e)).

The integer dim𝑘 𝒪𝑄 ∕𝔪𝑃 𝒪𝑄 is the multiplicity of 𝑄 in its fibre. The theorem says
2

that a finite map is flat if and only if the number of points in each fibre (counting

For example, let 𝑉 be the subvariety of 𝔸𝑛+1 defined by an equation

multiplicities) is constant.

𝑋 𝑚 + 𝑎1 𝑋 𝑚−1 + ⋯ + 𝑎𝑚 = 0, 𝑎𝑖 ∈ 𝑘[𝑇1 , … , 𝑇𝑛 ]

and let 𝜑 ∶ 𝑉 → 𝔸𝑛 be the projection map (see p. 50). The fibre over a point 𝑃 of 𝔸𝑛 is
the set of points (𝑃, 𝑐) with 𝑐 a root of the polynomial

𝑋 𝑚 + 𝑎1 (𝑃)𝑋 𝑚−1 + ⋯ + 𝑎𝑚 (𝑃) = 0.

The multiplicity of (𝑃, 𝑐) in its fibre is the multiplicity of 𝑐 as a root of the polynomial.
∑
Therefore 𝑄↦𝑃 dim𝑘 𝒪𝑄 ∕𝔪𝑃 𝒪𝑄 = 𝑚 for every 𝑃, and so the map 𝜑 is flat.

Criteria for flatness

Let 𝐴 be a local noetherian ring with maximal ideal 𝔪. A sequence of elements 𝑎1 , … , 𝑎𝑛
of 𝐴 is regular if 𝑎1 is a nonzerodivisor in 𝐴, 𝑎2 is a nonzerodivisor in 𝐴∕(𝑎1 ), etc.,
and 𝐴∕(𝑎1 , … , 𝑎𝑛 ) ≠ 0. According to a theorem of Rees, all maximal regular sequences
𝑎1 , … , 𝑎𝑛 , 𝑎𝑖 ∈ 𝔪, in 𝐴 have the same length, called the depth of 𝐴. According to the
Auslander–Buchsbaum formula, depth(𝐴) ≤ dim(𝐴). When the two are equal, the
ring is said to be Cohen–Macaulay. More generally, a noetherian ring 𝐴 is said to be
Cohen–Macaulay if it is zero or 𝐴𝔪 is Cohen–Macaulay for all maximal ideals 𝔪 of 𝐴.

Theorem 9.34. Let 𝜑 ∶ 𝐴 → 𝐵 be a local homomorphism of noetherian local rings, and

let 𝔪 be the maximal ideal of 𝐴. If 𝐴 is regular, 𝐵 is Cohen–Macaulay, and

dim(𝐵) = dim(𝐴) + dim(𝐵∕𝔪𝐵),

then 𝜑 is flat.

Proof. See Matsumura 1989, 23.1. 2

9.35. There are the following examples.

(a) Zero-dimensional and reduced one-dimensional noetherian rings are Cohen–
Macaulay (ibid. p. 139).

(c) Let 𝜑 ∶ 𝐴 → 𝐵 be a flat local homomorphism of noetherian local rings, and let 𝔪
(b) Regular noetherian rings are Cohen–Macaulay (ibid. p. 137).

be the maximal ideal of 𝐴. Then 𝐵 is Cohen–Macaulay if and only if both 𝐴 and

𝐵∕𝔪𝐵 are Cohen–Macaulay (ibid. p. 181).
 d. Lines on surfaces 213

Proposition 9.36. Let 𝜑 ∶ 𝐴 → 𝐵 be a finite homomorphism of noetherian rings with 𝐴

regular. Then 𝜑 is flat if and only if 𝐵 is Cohen–Macaulay.

Proof. Note that 𝐵∕𝔪𝐵 is zero-dimensional,1 hence Cohen–Macaulay, for every maxi-
mal ideal 𝔪 of 𝐴 (9.35a), and that ht(𝔫) = ht(𝔫𝑐 ) for every maximal ideal 𝔫 of 𝐵. If 𝜑 is
flat, then 𝐵 is Cohen–Macaulay by (9.35c). Conversely, if 𝐵 is Cohen–Macaulay, then 𝜑
is flat by (9.34). 2

Example 9.37. Let 𝐴 be a finite 𝑘[𝑋1 , … , 𝑋𝑛 ]-algebra (cf. 2.45). The map 𝑘[𝑋1 , … , 𝑋𝑛 ] →
𝐴 is flat if and only if 𝐴 is Cohen–Macaulay.

An algebraic variety 𝑉 is said to be Cohen–Macaulay if 𝒪𝑉,𝑃 is Cohen–Macaulay

for all 𝑃 ∈ 𝑉. An affine algebraic variety 𝑉 is Cohen–Macaulay if and only if 𝑘[𝑉] is

Theorem 9.38. Let 𝑉 and 𝑊 be algebraic varieties with 𝑉 nonsingular and 𝑊 Cohen–
Cohen–Macaulay (9.35c). A nonsingular variety is Cohen–Macaulay (9.35b).

Macaulay. A regular map 𝜑 ∶ 𝑊 → 𝑉 is flat if and only if

dim 𝜑−1 (𝑃) = dim 𝑊 − dim 𝑉 (39)

for all 𝑃 ∈ 𝑉.

Proof. Immediate consequence of (9.34). 2

Aside 9.39. The theorem fails with “nonsingular” weakened to “normal”. Let 𝐺 = ℤ∕2ℤ act
on 𝑊 = 𝔸2 by (𝑥, 𝑦) ↦ (−𝑥, −𝑦), and let 𝑉 ⊂ 𝔸3 be the quadric cone defined by 𝑇𝑉 = 𝑈 2 . The
def
def

𝜑 ∶ 𝑊 → 𝑉, (𝑥, 𝑦) ↦ (𝑡, 𝑢, 𝑣) = (𝑥 2 , 𝑥𝑦, 𝑦 2 ),

map

has as fibres the orbits for the action (it is the “quotient map” for the action). The variety 𝑊 is
nonsingular, and 𝑉 is normal because 𝑘[𝑉] = 𝑘[𝑋, 𝑌]𝐺 . Moreover 𝜑 is finite, and so its fibres
have constant dimension 0, but it is not flat because

∑ 3 if 𝑃 = (0, 0, 0)
dim𝑘 𝒪𝑄 ∕𝔪𝑃 𝒪𝑄 = {
2
𝑄↦𝑃
otherwise

contradicting Proposition 9.33. See mo117043.

d. Lines on surfaces
Every algebraic geometer knows the traditional

singular cubic surface in ℙ3

proof that there exists at least one line on a non-

Miles Reid.

the set of lines on a surface of degree 𝑚 in ℙ3 . To avoid possible problems, we assume

As an application of some of the above results, we consider the problem of describing

for the rest of this chapter that 𝑘 has characteristic zero.

We first need a way of describing lines in ℙ3 . Recall that we can associate with
each projective variety 𝑉 ⊂ ℙ𝑛 an affine cone over 𝑉̃ in 𝑘 𝑛+1 . This allows us to think
Note that 𝐶 = 𝐵∕𝔪𝐵 = 𝐵 ⊗𝐴 𝐴∕𝔪 is a finite 𝑘-algebra. Therefore it has only finitely many maximal
ideals. Every prime ideal in 𝐶 is an intersection of maximal ideals (2.18), but a prime ideal can equal a
1 def

finite intersection of ideals only if it equals one of the ideals.

214 9. Regular Maps and Their Fibres

of points in ℙ3 as being one-dimensional subspaces in 𝑘 4 , and lines in ℙ3 as being

two-dimensional subspaces in 𝑘 4 . To such a subspace 𝑊 ⊂ 𝑘 4 , we can attach a one-
⋀2 ⋀2 4
𝑊 in 𝑘 ≈ 𝑘6 , that is, to each line 𝐿 in ℙ3 , we can attach
point 𝑝(𝐿) in ℙ5 . Not every point in ℙ5 should be of the form 𝑝(𝐿) — heuristically, the
dimensional subspace

lines in ℙ3 should form a four-dimensional set. (Fix two planes in ℙ3 ; giving a line in
ℙ3 corresponds to choosing a point on each of the planes.) We shall show that there is
natural one-to-one correspondence between the set of lines in ℙ3 and the set of points
on a certain hyperspace 𝛱 ⊂ ℙ5 . Rather than using exterior algebras, I shall usually give

Let 𝐿 be a line in ℙ3 and let 𝐱 = (𝑥0 ∶ 𝑥1 ∶ 𝑥2 ∶ 𝑥3 ) and 𝐲 = (𝑦0 ∶ 𝑦1 ∶ 𝑦2 ∶ 𝑦3 ) be

the old-fashioned proofs.

distinct points on 𝐿. Then

| |
def || 𝑥 𝑥𝑗 |||
𝑝(𝐿) = (𝑝01 ∶ 𝑝02 ∶ 𝑝03 ∶ 𝑝12 ∶ 𝑝13 ∶ 𝑝23 ) ∈ ℙ5 , 𝑝𝑖𝑗 = ||| 𝑖 |,
|| 𝑦𝑖 𝑦𝑗 |||

depends only on 𝐿. The 𝑝𝑖𝑗 are called the Plücker coordinates of 𝐿, after Plücker (1801-

In terms of exterior algebras, write 𝑒0 , 𝑒1 , 𝑒2 , 𝑒3 for the canonical basis for 𝑘 4 , so that
∑ ∑ ⋀2 4
1868).

𝐱, regarded as a point of 𝑘 4 is 𝑥𝑖 𝑒𝑖 , and 𝐲 = 𝑦𝑖 𝑒𝑖 ; then 𝑘 is a 6-dimensional

∑
vector space with basis 𝑒𝑖 ∧ 𝑒𝑗 , 0 ≤ 𝑖 < 𝑗 ≤ 3, and 𝑥∧ 𝑦 = 𝑝𝑖𝑗 𝑒𝑖 ∧ 𝑒𝑗 with 𝑝𝑖𝑗 given by the

We define 𝑝𝑖𝑗 for all 𝑖, 𝑗, 0 ≤ 𝑖, 𝑗 ≤ 3 by the same formula — thus 𝑝𝑖𝑗 = −𝑝𝑗𝑖 .
above formula.

Lemma 9.40. The line 𝐿 can be recovered from 𝑝(𝐿) as follows:

∑ ∑ ∑ ∑
𝐿 = {( 𝑗 𝑎𝑗 𝑝0𝑗 ∶ 𝑗 𝑎𝑗 𝑝1𝑗 ∶ 𝑗 𝑎𝑗 𝑝2𝑗 ∶ 𝑗 𝑎𝑗 𝑝3𝑗 ) ∣ (𝑎0 ∶ 𝑎1 ∶ 𝑎2 ∶ 𝑎3 ) ∈ ℙ3 }.

Proof. Let 𝐿̃ be the cone over 𝐿 in 𝑘 4 — it is a two-dimensional subspace of 𝑘 4 — and

̃
let 𝐱 = (𝑥0 , 𝑥1 , 𝑥2 , 𝑥3 ) and 𝐲 = (𝑦0 , 𝑦1 , 𝑦2 , 𝑦3 ) be two linearly independent vectors in 𝐿.

𝐿̃ = {𝑓(𝐲)𝐱 − 𝑓(𝐱)𝐲 ∣ 𝑓 ∶ 𝑘 4 → 𝑘 linear}.

Then

∑
Write 𝑓 = 𝑎𝑗 𝑋𝑗 ; then
∑ ∑ ∑ ∑
𝑓(𝐲)𝐱 − 𝑓(𝐱)𝐲 = ( 𝑎𝑗 𝑝0𝑗 , 𝑎𝑗 𝑝1𝑗 , 𝑎𝑗 𝑝2𝑗 , 𝑎𝑗 𝑝3𝑗 ). 2

Lemma 9.41. The point 𝑝(𝐿) lies on the quadric 𝛱 ⊂ ℙ5 defined by the equation

𝑋01 𝑋23 − 𝑋02 𝑋13 + 𝑋03 𝑋12 = 0.

|| ||
Proof. This can be verified by direct calculation, or by using that

|| 𝑥0 𝑥1 𝑥2 𝑥3 ||
|| 𝑦0 𝑦1 𝑦2 𝑦3 ||
0 = ||| || = 2(𝑝 𝑝 − 𝑝 𝑝 + 𝑝 𝑝 )
|| 𝑥0 𝑥1 𝑥2 𝑥3 ||| 01 23 02 13 03 12
|| ||
|| 𝑦0 𝑦1 𝑦2 𝑦3 ||

(expansion in terms of 2 × 2 minors). 2

Lemma 9.42. Every point of 𝛱 is of the form 𝑝(𝐿) for a unique line 𝐿.
 d. Lines on surfaces 215

Proof. Assume 𝑝03 ≠ 0; then the line through the points (0 ∶ 𝑝01 ∶ 𝑝02 ∶ 𝑝03 ) and
(𝑝03 ∶ 𝑝13 ∶ 𝑝23 ∶ 0) has Plücker coordinates

(−𝑝01 𝑝03 ∶ −𝑝02 𝑝03 ∶ −𝑝03

2
∶ 𝑝01 𝑝23 − 𝑝02 𝑝13 ∶ −𝑝03 𝑝13 ∶ −𝑝03 𝑝23 )
⏟⎴⎴⎴⎴⏟⎴⎴⎴⎴⏟
−𝑝03 𝑝12

= (𝑝01 ∶ 𝑝02 ∶ 𝑝03 ∶ 𝑝12 ∶ 𝑝13 ∶ 𝑝23 ).

A similar construction works when one of the other coordinates is nonzero, and this
way we get inverse maps. 2

Thus we have a canonical one-to-one correspondence

{lines in ℙ3 } ↔ {points on 𝛱};

that is, we have identified the set of lines in ℙ3 with the points of an algebraic variety.
We may now use the methods of algebraic geometry to study the set. (This is a special

We next consider the set of homogeneous polynomials of degree 𝑚 in 4 variables,

case of the Grassmannians discussed in §6.)

∑
𝐹(𝑋0 , 𝑋1 , 𝑋2 , 𝑋3 ) = 𝑎𝑖0 𝑖1 𝑖2 𝑖3 𝑋00 … 𝑋33 .
𝑖 𝑖

𝑖0 +𝑖1 +𝑖2 +𝑖3 =𝑚

( 3+𝑚
Lemma 9.43. The set )
of homogeneous polynomials of degree 𝑚 in 4 variables is a vector
space of dimension 𝑚

( ) (𝑚 + 1)(𝑚 + 2)(𝑚 + 3)
Proof. See the 6.39. 2

Let 𝜈 = 3+𝑚𝑚 −1 = 6 − 1, and regard ℙ𝜈 as the projective

space attached to the vector space of homogeneous polynomials of degree 𝑚 in 4 variables
(p. 147). Then we have a surjective map

ℙ𝜈 → {surfaces of degree 𝑚 in ℙ3 },
∑
(… ∶ 𝑎𝑖0 𝑖1 𝑖2 𝑖3 ∶ …) ↦ 𝑉(𝐹), 𝐹= 𝑎𝑖0 𝑖1 𝑖2 𝑖3 𝑋00 𝑋11 𝑋22 𝑋33 .
𝑖 𝑖 𝑖 𝑖

The map is not quite injective — for example, 𝑋 2 𝑌 and 𝑋𝑌 2 define the same surface —
but nevertheless, we can (somewhat loosely) think of the points of ℙ𝜈 as being (possibly
degenerate) surfaces of degree 𝑚 in ℙ3 .
Let 𝛤𝑚 ⊂ 𝛱 × ℙ𝜈 ⊂ ℙ5 × ℙ𝜈 be the set of pairs (𝐿, 𝐹) consisting of a line 𝐿 in ℙ3 lying
on the surface 𝐹(𝑋0 , 𝑋1 , 𝑋2 , 𝑋3 ) = 0.

Theorem 9.44. The set 𝛤𝑚 is an irreducible closed subset of 𝛱 × ℙ𝜈 ; it is therefore a

𝑚(𝑚 + 1)(𝑚 + 5)
projective variety. The dimension of 𝛤𝑚 is 6 + 3.

Example 9.45. For 𝑚 = 1, 𝛤𝑚 is the set of pairs consisting of a plane in ℙ3 and a line
on the plane. The theorem says that the dimension of 𝛤1 is 5. Since there are ∞3 planes
in ℙ3 , and each has ∞2 lines on it, this seems to be correct.
216 9. Regular Maps and Their Fibres

Proof. We first show that 𝛤𝑚 is closed. Let

∑
𝑝(𝐿) = (𝑝01 ∶ 𝑝02 ∶ …) 𝐹= 𝑎𝑖0 𝑖1 𝑖2 𝑖3 𝑋00 ⋯ 𝑋33 .
𝑖 𝑖

From 9.40 we see that 𝐿 lies on the surface 𝐹(𝑋0 , 𝑋1 , 𝑋2 , 𝑋3 ) = 0 if and only if
∑ ∑ ∑ ∑
𝐹( 𝑏𝑗 𝑝0𝑗 ∶ 𝑏𝑗 𝑝1𝑗 ∶ 𝑏𝑗 𝑝2𝑗 ∶ 𝑏𝑗 𝑝3𝑗 ) = 0, all (𝑏0 , … , 𝑏3 ) ∈ 𝑘 4 .

Expand this out as a polynomial in the 𝑏𝑗 with coefficients polynomials in the 𝑎𝑖0 𝑖1 𝑖2 𝑖3
and 𝑝𝑖𝑗 . Then 𝐹(...) = 0 for all 𝐛 ∈ 𝑘 4 if and only if the coefficients of the polynomial

𝑃(… , 𝑎𝑖0 𝑖1 𝑖2 𝑖3 , … ; 𝑝01 ∶ 𝑝02 ∶ …)

are all zero. But each coefficient is of the form

with 𝑃 homogeneous separately in the 𝑎’s and 𝑝’s, and so the set is closed in 𝛱 × ℙ𝜈 (cf.

It remains to compute the dimension of 𝛤𝑚 . We shall apply Proposition 9.11 to the

the discussion in 6.51).

(𝐿, 𝐹) 𝛤 𝑚 ⊂ 𝛱 × ℙ𝜈
projection map

𝜑
↤

𝐿 𝛱.
→

For 𝐿 ∈ 𝛱, 𝜑−1 (𝐿) consists of the homogeneous polynomials of degree 𝑚 such that
𝐿 ⊂ 𝑉(𝐹) (taken up to nonzero scalars). After a change of coordinates, we can assume
that 𝐿 is the line
𝑋 =0
{ 0
𝑋1 = 0,
i.e., 𝐿 = {(0, 0, ∗, ∗)}. Then 𝐿 lies on 𝐹(𝑋0 , 𝑋1 , 𝑋2 , 𝑋3 ) = 0 if and only if 𝑋0 or 𝑋1 occurs
in each nonzero monomial term in 𝐹, i.e.,

𝐹 ∈ 𝜑−1 (𝐿) ⇐⇒ 𝑎𝑖0 𝑖1 𝑖2 𝑖3 = 0 whenever 𝑖0 = 0 = 𝑖1 .

Thus 𝜑−1 (𝐿) is a linear subspace of ℙ𝜈 ; in particular, it is irreducible. We now compute

its dimension. Recall that 𝐹 has 𝜈 +1 coefficients altogether; the number with 𝑖0 = 0 = 𝑖1
is 𝑚 + 1, and so 𝜑−1 (𝐿) has dimension
(𝑚 + 1)(𝑚 + 2)(𝑚 + 3) 𝑚(𝑚 + 1)(𝑚 + 5)
− 1 − (𝑚 + 1) = − 1.
6 6
We can now deduce from 9.11 that 𝛤𝑚 is irreducible and that
𝑚(𝑚 + 1)(𝑚 + 5)
dim(𝛤𝑚 ) = dim(𝛱) + dim(𝜑−1 (𝐿)) = + 3,
6
as claimed. 2

𝜓 −1 (𝐹) = {𝐿 ∣ 𝐿 lies on 𝑉(𝐹)}.

Now consider the other projection. By definition

Example 9.46. Let 𝑚 = 1. Then 𝜈 = 3 and dim 𝛤1 = 5. The projection 𝜓 ∶ 𝛤1 → ℙ3 is

surjective (every plane contains at least one line), and 9.9 tells us that dim 𝜓 −1 (𝐹) ≥ 2.
In fact of course, the lines on any plane form a 2-dimensional family, and so 𝜓−1 (𝐹) = 2
for all 𝐹.
 d. Lines on surfaces 217

Theorem 9.47. When 𝑚 > 3, the surfaces of degree 𝑚 containing no line correspond to
an open subset of ℙ𝜈 .

𝑚(𝑚 + 1)(𝑚 + 5) (𝑚 + 1)(𝑚 + 2)(𝑚 + 3)

Proof. We have

dim 𝛤𝑚 − dim ℙ𝜈 = +3− + 1 = 4 − (𝑚 + 1).

6 6
Therefore, if 𝑚 > 3, then dim 𝛤𝑚 < dim ℙ𝜈 , and so 𝜓(𝛤𝑚 ) is a proper closed subvariety
of ℙ𝜈 . This proves the claim. 2

We now look at the case 𝑚 = 2. Here dim 𝛤𝑚 = 10, and 𝜈 = 9, which suggests that
𝜓 should be surjective and that its fibres should all have dimension ≥ 1. We shall see
that this is correct.
A quadric is said to be nondegenerate if it is defined by an irreducible polynomial
of degree 2. After a change of variables, any nondegenerate quadric will be defined by

𝑋𝑊 = 𝑌𝑍.
an equation

This is just the image of the Segre mapping (see 6.26)

(𝑎0 ∶ 𝑎1 ), (𝑏0 ∶ 𝑏1 ) ↦ (𝑎0 𝑏0 ∶ 𝑎0 𝑏1 ∶ 𝑎1 𝑏0 ∶ 𝑎1 𝑏1 ) ∶ ℙ1 × ℙ1 → ℙ3 .

There are two obvious families of lines on ℙ1 × ℙ1 , namely, the horizontal family and
the vertical family; each is parametrized by ℙ1 , and so is called a pencil of lines. They
map to two families of lines on the quadric:

𝑡0 𝑋 = 𝑡1 𝑍 𝑡 𝑋 = 𝑡1 𝑌
{ and { 0
𝑡0 𝑌 = 𝑡1 𝑊 𝑡0 𝑍 = 𝑡1 𝑊.

quadric surface contains a line, that is, that 𝜓 ∶ 𝛤2 → ℙ9 is surjective. Thus (9.9) tells us
Since a degenerate quadric is a surface or a union of two surfaces, we see that every

that all the fibres have dimension ≥ 1, and the set where the dimension is > 1 is a proper
closed subset. In fact the dimension of the fibre is > 1 exactly on the set of reducible 𝐹’s,

It follows from the above discussion that if 𝐹 is nondegenerate, then 𝜓−1 (𝐹) is
which we know to be closed (this was a homework problem in the original course).

isomorphic to the disjoint union of two lines, 𝜓 −1 (𝐹) ≈ ℙ1 ∪ ℙ1 . Classically, one defines

lines. One can show that the set of reguli is, in a natural way, an algebraic variety 𝑅, and
a regulus to be a nondegenerate quadric surface together with a choice of a pencil of

that, over the set of nondegenerate quadrics, 𝜓 factors into the composite of two regular

𝛤2 − 𝜓 −1 (𝑆) = pairs (𝐹, 𝐿) with 𝐿 on 𝐹;

maps:

↓
𝑅 = set of reguli;
↓
ℙ9 − 𝑆 = set of nondegenerate quadrics.
The fibres of the top map are connected, and of dimension 1 (they are all isomorphic to
ℙ1 ), and the second map is finite and two-to-one. Factorizations of this type occur quite

We now look at the case 𝑚 = 3. Here dim 𝛤3 = 19; 𝜈 = 19 ∶ we have a map

generally (see the Stein factorization theorem, 8.64).

𝜓 ∶ 𝛤3 → ℙ19 .
218 9. Regular Maps and Their Fibres

Theorem 9.48. The set of cubic surfaces containing exactly 27 lines corresponds to an
open subset of ℙ19 ; the remaining surfaces either contain an infinite number of lines or a
nonzero finite number ≤ 27.

Example 9.49. (a) Consider the Fermat surface

𝑋03 + 𝑋13 + 𝑋23 + 𝑋33 = 0.

Let 𝜁 be a primitive cube root of one. There are the following lines on the surface,
0 ≤ 𝑖, 𝑗 ≤ 2:

𝑋0 + 𝜁 𝑖 𝑋1 = 0 𝑋0 + 𝜁 𝑖 𝑋2 = 0 𝑋0 + 𝜁 𝑖 𝑋3 = 0
{ { {
𝑋2 + 𝜁 𝑗 𝑋3 = 0 𝑋1 + 𝜁 𝑗 𝑋3 = 0 𝑋1 + 𝜁 𝑗 𝑋2 = 0.

There are three sets, each with nine lines, for a total of 27 lines.

𝑋1 𝑋2 𝑋3 = 𝑋03 .
(b) Consider the surface

𝑋0 ≠ 0 — here we can take the equation to be 𝑋1 𝑋2 𝑋3 = 1. A line in 𝔸3 can be written

In this case, there are exactly three lines. To see this, look first in the affine space where

in parametric form 𝑋𝑖 = 𝑎𝑖 𝑡 + 𝑏𝑖 , but a direct inspection shows that no such line lies on
the surface. Now look where 𝑋0 = 0, that is, in the plane at infinity. The intersection of
the surface with this plane is given by 𝑋1 𝑋2 𝑋3 = 0 (homogeneous coordinates), which
is the union of three lines, namely,

𝑋1 = 0; 𝑋2 = 0; 𝑋3 = 0.

Therefore, the surface contains exactly three lines.

𝑋13 + 𝑋23 = 0.
(c) Consider the surface

Here there is a pencil of lines:

𝑡0 𝑋1 = 𝑡1 𝑋0
{
𝑡0 𝑋2 = −𝑡1 𝑋0 .

(In the affine space where 𝑋0 ≠ 0, the equation is 𝑋 3 + 𝑌 3 = 0, which contains the line
𝑋 = 𝑡, 𝑌 = −𝑡, all 𝑡.)

We now discuss the proof of Theorem 9.48. If 𝜓 ∶ 𝛤3 → ℙ19 were not surjective, then
𝜓(𝛤3 ) would be a proper closed subvariety of ℙ19 , and the nonempty fibres would all
have dimension ≥ 1 (by 9.9), which contradicts two of the above examples. Therefore the
map is surjective, and there is an open subset 𝑈 of ℙ19 where the fibres have dimension
0; outside 𝑈, the fibres have dimension > 0.

is an open subset 𝑈 ′ where the cubics have exactly 27 lines. In fact, 𝑈 ′ can be taken
Given that every cubic surface has at least one line, it is not hard to show that there

to be the set of nonsingular cubics. According to 8.26, the restriction of 𝜓 to 𝜓 −1 (𝑈) is

finite, and so we can apply 8.40 to see that all cubics in 𝑈 − 𝑈 ′ have fewer than 27 lines.

Remark 9.50. The twenty-seven lines on a cubic surface were discovered in 1849 by
Salmon and Cayley, and have been much studied — see A. Henderson, The Twenty-
Seven Lines Upon the Cubic Surface, Cambridge University Press, 1911. For example, it
is known that the group of permutations of the set of 27 lines preserving intersections
 e. Bertini’s theorem 219

(that is, such that 𝐿 ∩ 𝐿′ ≠ ∅ ⇐⇒ 𝜎(𝐿) ∩ 𝜎(𝐿′ ) ≠ ∅) is isomorphic to the Weyl group of
the root system of a simple Lie algebra of type 𝐸6 , and hence has 25920 elements.
It is known that there is a set of 6 skew lines on a nonsingular cubic surface 𝑉. Let
𝐿 and 𝐿′ be two skew lines. Then “in general” a line joining a point on 𝐿 to a point on
𝐿′ will intersect the surface in exactly one further point. In this way one obtains an
invertible regular map from an open subset of ℙ1 × ℙ1 to an open subset of 𝑉, and hence
𝑉 is birationally equivalent to ℙ2 .

e. Bertini’s theorem
Let 𝑋 ⊂ ℙ𝑛 be a nonsingular projective variety. The hyperplanes 𝐻 in ℙ𝑛 form a
projective space ℙ𝑛∨ (the “dual” projective space). The hyperplanes 𝐻 not containing 𝑋
and such that 𝑋 ∩ 𝐻 is nonsingular, form an open subset of ℙ𝑛∨ . If dim(𝑋) ≥ 2, then
the intersections 𝑋 ∩ 𝐻 are connected. For a proof of a weak version of the theorem, see
11.45 (Chapter 11).

f. Birational classification
Recall that two varieties 𝑉 and 𝑊 are birationally equivalent if 𝑘(𝑉) ≈ 𝑘(𝑊). This
means that the varieties themselves become isomorphic once a proper closed subset has
been removed from each (3.36).
The main problem of birational algebraic geometry is to classify algebraic varieties
up to birational equivalence by finding a particularly good representative in each equiva-
lence class.

nonsingular projective curve (up to isomorphism). More precisely, the functor 𝑉 ⇝ 𝑘(𝑉)
For curves this is easy: in each birational equivalence class there is exactly one

curves over 𝑘 and dominant maps to the category of fields finitely generated and of
is a contravariant equivalence from the category of nonsingular projective algebraic

transcendence degree 1 over 𝑘.

For surfaces, the problem is already much more difficult because many surfaces,
even projective and nonsingular, will have the same function field. For example, every
blow-up of a point on a surface produces a birationally equivalent surface.
A nonsingular projective surface is said to be minimal if it cannot be obtained from
another such surface by blowing up. The main theorem for surfaces (Enriques 1914,
Kodaira 1966) says that a birational equivalence class contains either

(b) a surface of the form 𝐶 × ℙ1 for a unique nonsingular projective curve 𝐶.

(a) a unique minimal surface, or

In higher dimensions, the problem becomes very involved, although much progress
has been made — see Wikipedia: Minimal model program. For a beautiful article
on the minimal model program, see Notices AMS, Jan. 2024.

Exercises

9-1. Let 𝐺 be a connected group variety, and consider an action of 𝐺 on a variety 𝑉, i.e.,
a regular map 𝐺 × 𝑉 → 𝑉 such that (𝑔𝑔′ )𝑣 = 𝑔(𝑔′ 𝑣) for all 𝑔, 𝑔′ ∈ 𝐺 and 𝑣 ∈ 𝑉. Show
̄ and that 𝑂̄ ∖ 𝑂 is a union of orbits of
that each orbit 𝑂 = 𝐺𝑣 of 𝐺 is open in its closure 𝑂,
220 9. Regular Maps and Their Fibres

strictly lower dimension. Deduce that each orbit is a nonsingular subvariety of 𝑉, and
that there exists at least one closed orbit.

9-2. Let 𝐺 = GL2 = 𝑉, and let 𝐺 act on 𝑉 by conjugation. According to the theory of

(a) Characteristic polynomial 𝑋 2 + 𝑎𝑋 + 𝑏; distinct roots.

Jordan canonical forms, the orbits are of three types:

(b) Characteristic polynomial 𝑋 2 + 𝑎𝑋 + 𝑏; minimal polynomial the same; repeated

(c) Characteristic polynomial 𝑋 2 + 𝑎𝑋 + 𝑏 = (𝑋 − 𝛼)2 ; minimal polynomial 𝑋 − 𝛼.

roots.

of 𝑉), the closure of the orbit, and which other orbits are contained in the closure.
For each type, find the dimension of the orbit, the equations defining it (as a subvariety

the following (fairly difficult) result: for any closed subgroup 𝐻 of an group variety
(You may assume, if you wish, that the characteristic is zero. Also, you may assume

𝐺, 𝐺∕𝐻 has a natural structure of an algebraic variety with the following properties:
𝐺 → 𝐺∕𝐻 is regular, and a map 𝐺∕𝐻 → 𝑉 is regular if the composite 𝐺 → 𝐺∕𝐻 → 𝑉 is
regular; dim 𝐺∕𝐻 = dim 𝐺 − dim 𝐻.)
[The enthusiasts may wish to carry out the analysis for GL𝑛 .]

9-3. Find 3𝑑2 lines on the Fermat projective surface

𝑋0𝑑 + 𝑋1𝑑 + 𝑋2𝑑 + 𝑋3𝑑 = 0, 𝑑 ≥ 3, (𝑝, 𝑑) = 1, 𝑝 the characteristic.

9-4. (a) Let 𝜑 ∶ 𝑊 → 𝑉 be a quasi-finite dominant regular map of irreducible varieties.

Show that there are open subsets 𝑈 ′ and 𝑈 of 𝑊 and 𝑉 such that 𝜑(𝑈 ′ ) ⊂ 𝑈 and
𝜑 ∶ 𝑈 ′ → 𝑈 is finite.
(b) Let 𝐺 be a group variety acting transitively on irreducible varieties 𝑊 and 𝑉, and
let 𝜑 ∶ 𝑊 → 𝑉 be 𝐺-equivariant regular map satisfying the hypotheses in (a). Then 𝜑 is
finite, and hence proper.
Solutions to the exercises

1-1 Use induction on 𝑛. For 𝑛 = 1, use that a nonzero polynomial in one variable has
∑
suppose 𝑛 > 1 and write 𝑓 = 𝑔𝑖 𝑋𝑛𝑖 with each 𝑔𝑖 ∈ 𝑘[𝑋1 , … , 𝑋𝑛−1 ]. If 𝑓 is not the zero
only finitely many roots (which follows from unique factorization, for example). Now

polynomial, then some 𝑔𝑖 is not the zero polynomial. Therefore, by induction, there
exist (𝑎1 , … , 𝑎𝑛−1 ) ∈ 𝑘 𝑛−1 such that 𝑓(𝑎1 , … , 𝑎𝑛−1 , 𝑋𝑛 ) is not the zero polynomial. Now,
by the degree-one case, there exists a 𝑏 such that 𝑓(𝑎1 , … , 𝑎𝑛−1 , 𝑏) ≠ 0.
1-2 (𝑋 + 2𝑌, 𝑍); Gaussian elimination (to reduce the matrix of coefficients to row echelon
form); (1), unless the characteristic of 𝑘 is 2, in which case the ideal is (𝑋 + 1, 𝑍 + 1).
2-1 𝑊 = 𝑌-axis, and so 𝐼(𝑊) = (𝑋). Clearly,

(𝑋 2 , 𝑋𝑌 2 ) ⊂ (𝑋) ⊂ rad(𝑋 2 , 𝑋𝑌 2 )

and rad((𝑋)) = (𝑋). On taking radicals, we find that (𝑋) = rad(𝑋 2 , 𝑋𝑌 2 ).

2-2 The 𝑑 × 𝑑 minors of a matrix are polynomials in the entries of the matrix, and the
set of matrices with rank ≤ 𝑟 is the set where all (𝑟 + 1) × (𝑟 + 1) minors are zero.
2-3 Clearly 𝑉 = 𝑉(𝑋𝑛 − 𝑋1𝑛 , … , 𝑋2 − 𝑋12 ). The map

𝑋𝑖 ↦ 𝑇 𝑖 ∶ 𝑘[𝑋1 , … , 𝑋𝑛 ] → 𝑘[𝑇]

induces an isomorphism 𝑘[𝑉] → 𝑘[𝑇]. [Hence 𝑡 ↦ (𝑡, … , 𝑡𝑛 ) is an isomorphism of affine

varieties 𝔸1 → 𝑉.]
2-4 On tensoring the exact sequence of ℚ-vector spaces

0 → (𝑓1 , … , 𝑓𝑚 ) → ℚ[𝑋1 , … , 𝑋𝑛 ] → ℚ[𝑋1 , … , 𝑋𝑛 ]∕(𝑓1 , … , 𝑓𝑚 ) → 0

with ℂ, we get an exact sequence of ℂ-vector spaces

0 → (𝑓1 , … , 𝑓𝑚 ) → ℂ[𝑋1 , … , 𝑋𝑛 ] → ℂ[𝑋1 , … , 𝑋𝑛 ]∕(𝑓1 , … , 𝑓𝑚 ) → 0.

As the 𝑓𝑖 have no common zero in ℂ, the right-most term of the second sequence is zero,

2-6 The statement Hom𝑘−algebras (𝐴 ⊗ℚ 𝑘, 𝐵 ⊗ℚ 𝑘) ≠ ∅ can be interpreted as saying that

which implies that the same is true of the first sequence.

a certain set of polynomials has a zero in 𝑘.2 If the polynomials have a common zero in
ℂ, then the ideal they generate in ℂ[𝑋1 , …] does not contain 1. A fortiori, the ideal they
generate in ℚ[𝑋1 , …] does not contain 1, and so the Nullstellensatz (2.11) implies that
the polynomials have a common zero in 𝑘.
2
Choose bases for 𝐴 and 𝐵 as ℚ-vector spaces. Now a linear map from 𝐴 to 𝐵 is given by a matrix 𝑀.
The condition on the coefficients of the matix for the map to be a homomorphism of algebras is polynomial.

221
222 Solutions to the exercises

2-7 Regard Hom𝐴 (𝑀, 𝑁) as an affine space over 𝑘; the elements not isomorphisms are
the zeros of a polynomial; because 𝑀 and 𝑁 become isomorphic over 𝑘 al , the polynomial
is not identically zero; therefore it has a nonzero in 𝑘 (Exercise 1-1).
3-1 A map 𝛼 ∶ 𝔸1 → 𝔸1 is continuous for the Zariski topology if the inverse images of
finite sets are finite, whereas it is regular only if it is given by a polynomial 𝑃 ∈ 𝑘[𝑇], so
it is easy to give examples, e.g., any map 𝛼 such that 𝛼−1 (point) is finite but arbitrarily

3-3 The image omits the points on the 𝑌-axis except for the origin. The complement of
large.

(0, 0), and so it not closed. See the introduction to Chapter 9.

the image is not dense, and so it is not open, but any polynomial zero on it is also zero at

3-4 Let 𝑖 be an element of 𝑘 with square −1. The map (𝑥, 𝑦) ↦ (𝑥 + 𝑖𝑦, 𝑥 − 𝑖𝑦) from
the circle to the hyperbola has inverse (𝑥, 𝑦) ↦ ((𝑥 + 𝑦)∕2, (𝑥 − 𝑦)∕2𝑖). The 𝑘-algebra
𝑘[𝑋, 𝑌]∕(𝑋𝑌 − 1) ≃ 𝑘[𝑋, 𝑋 −1 ], which is not isomorphic to 𝑘[𝑋] (too many units).
3-5 No, because both +1 and −1 map to (0, 0). The map on rings is

𝑘[𝑥, 𝑦] → 𝑘[𝑇], 𝑥 ↦ 𝑇 2 − 1, 𝑦 ↦ 𝑇(𝑇 2 − 1),

which is not surjective (𝑇 is not in the image).

4-1 (b) The singular points are the common solutions to

⎧ 4𝑋 3 − 2𝑋𝑌 2 = 0 ⇐⇒ 𝑋 = 0 or 𝑌 2 = 2𝑋 2
4𝑌 − 2𝑋 𝑌 = 0
3 2 ⇐⇒ 𝑌 = 0 or 𝑋 2 = 2𝑌 2
⎨ 4
𝑋 + 𝑌 4 − 𝑋 2 𝑌 2 = 0.
⎩
Thus, only (0, 0) is singular, and the variety is its own tangent cone.
4-2 Directly from the definition of the tangent space, we have that

𝑇𝐚 (𝑉 ∩ 𝐻) ⊂ 𝑇𝐚 (𝑉) ∩ 𝑇𝐚 (𝐻).

dim 𝑇𝐚 (𝑉 ∩ 𝐻) ≥ dim 𝑉 ∩ 𝐻 = dim 𝑉 − 1 = dim 𝑇𝐚 (𝑉) ∩ 𝑇𝐚 (𝐻),

As

we must have equalities everywhere, which proves that 𝐚 is nonsingular on 𝑉 ∩ 𝐻. (In

The surface 𝑌 2 = 𝑋 2 +𝑍 is smooth, but its intersection with the 𝑋-𝑌 plane is singular.
particular, it cannot lie on more than one irreducible component.)

No, 𝑃 need not be singular on 𝑉 ∩ 𝐻 if 𝐻 ⊃ 𝑇𝑃 (𝑉) — for example, we could have

𝐻 ⊃ 𝑉 or 𝐻 could be the tangent line to a curve.
4-4 We can assume 𝑉 and 𝑊 to affine, say

𝐼(𝑉) = 𝔞 ⊂ 𝑘[𝑋1 , … , 𝑋𝑚 ]
𝐼(𝑊) = 𝔟 ⊂ 𝑘[𝑋𝑚+1 , … , 𝑋𝑚+𝑛 ].

If 𝔞 = (𝑓1 , … , 𝑓𝑟 ) and 𝔟 = (𝑔1 , … , 𝑔𝑠 ), then 𝐼(𝑉 × 𝑊) = (𝑓1 , … , 𝑓𝑟 , 𝑔1 , … , 𝑔𝑠 ). Thus,

𝑇(𝐚,𝐛) (𝑉 × 𝑊) is defined by the equations

(𝑑𝑓1 )𝐚 = 0, … , (𝑑𝑓𝑟 )𝐚 = 0, (𝑑𝑔1 )𝐛 = 0, … , (𝑑𝑔𝑠 )𝐛 = 0,

which can obviously be identified with 𝑇𝐚 (𝑉) × 𝑇𝐛 (𝑊).

 Solutions to the exercises 223

4-5 Take 𝐶 to be the union of the coordinate axes in 𝔸𝑛 . (Of course, if you want 𝐶 to be

4-6 A matrix 𝐴 satisfies the equations

irreducible, then this is more difficult. . . )

(𝐼 + 𝜀𝐴)tr ⋅ 𝐽 ⋅ (𝐼 + 𝜀𝐴) = 𝐼

𝐴tr ⋅ 𝐽 + 𝐽 ⋅ 𝐴 = 0.
if and only if

𝑀 𝑁
Such an 𝐴 is of the form ( ) with 𝑀, 𝑁, 𝑃, 𝑄 𝑛 × 𝑛-matrices satisfying
𝑃 𝑄

𝑁 tr = 𝑁, 𝑃tr = 𝑃, 𝑀 tr = −𝑄.

The dimension of the space of 𝐴’s is therefore

𝑛(𝑛 + 1) 𝑛(𝑛 + 1)
(for 𝑁) + (for 𝑃) + 𝑛2 (for 𝑀, 𝑄) = 2𝑛2 + 𝑛.
2 2

4-7 Let 𝐶 be the curve 𝑌 2 = 𝑋 3 , and consider the map 𝔸1 → 𝐶, 𝑡 ↦ (𝑡 2 , 𝑡3 ). The

corresponding map on rings 𝑘[𝑋, 𝑌]∕(𝑌 2 ) → 𝑘[𝑇] is not an isomorphism, but the map

4-8 The singular locus 𝑉sing has codimension ≥ 2 in 𝑉, and this implies that 𝑉 is normal.
on the geometric tangent cones is an isomorphism.

[Idea of the proof: let 𝑓 ∈ 𝑘(𝑉) be integral over 𝑘[𝑉], 𝑓 ∉ 𝑘[𝑉], 𝑓 = 𝑔∕ℎ, 𝑔, ℎ ∈ 𝑘[𝑉];
for any 𝑃 ∈ 𝑉(ℎ) ∖ 𝑉(𝑔), 𝒪𝑃 is not integrally closed, and so 𝑃 is singular.]
4-9 No! Let 𝔞 = (𝑋 2 𝑌). Then 𝑉(𝔞) is the union of the 𝑋 and 𝑌 axes, and 𝐼𝑉(𝔞) = (𝑋𝑌).
For 𝐚 = (𝑎, 𝑏),

(𝑑𝑋 2 𝑌)𝐚 = 2𝑎𝑏(𝑋 − 𝑎) + 𝑎2 (𝑌 − 𝑏)

(𝑑𝑋𝑌)𝐚 = 𝑏(𝑋 − 𝑎) + 𝑎(𝑌 − 𝑏).

If 𝑎 ≠ 0 and 𝑏 = 0, then the equations

(𝑑𝑋 2 𝑌)𝐚 = 𝑎2 𝑌 = 0
(𝑑𝑋𝑌)𝐚 = 𝑎𝑌 = 0

5-1 Let 𝑓 be regular on ℙ1 . Then 𝑓|𝑈0 = 𝑃(𝑋) ∈ 𝑘[𝑋], where 𝑋 is the regular function
have the same solutions.

(𝑎0 ∶ 𝑎1 ) ↦ 𝑎1 ∕𝑎0 ∶ 𝑈0 → 𝑘, and 𝑓|𝑈1 = 𝑄(𝑌) ∈ 𝑘[𝑌], where 𝑌 is (𝑎0 ∶ 𝑎1 ) ↦ 𝑎0 ∕𝑎1 .

On 𝑈0 ∩ 𝑈1 , 𝑋 and 𝑌 are reciprocal functions. Thus 𝑃(𝑋) and 𝑄(1∕𝑋) define the same
function on 𝑈0 ∩ 𝑈1 = 𝔸1 ∖ {0}. This implies that they are equal in 𝑘(𝑋), and must both

∏ ⨆
5-2 Note that 𝛤(𝑉, 𝒪𝑉 ) = 𝛤(𝑉𝑖 , 𝒪𝑉𝑖 ) — to give a regular function on 𝑉𝑖 is the same
be constant.

as to give a regular function on each 𝑉𝑖 (this is the “obvious” ringed space structure).
∏
Thus, if 𝑉 is affine, it must equal Specm( 𝐴𝑖 ), where 𝐴𝑖 = 𝛤(𝑉𝑖 , 𝒪𝑉𝑖 ), and so 𝑉 =
⨆
Specm(𝐴𝑖 ) (use the description of the ideals in 𝐴 × 𝐵 on in Section 1a). Etc..
5-5 Let 𝐻 be an algebraic subgroup of 𝐺. By definition, 𝐻 is locally closed, i.e., open in
̄ Assume first that 𝐻 is connected. Then 𝐻̄ is a connected algebraic
its Zariski closure 𝐻.
̄ In the general
group, and it is a disjoint union of the cosets of 𝐻. It follows that 𝐻 = 𝐻.
224 Solutions to the exercises

case, 𝐻 is a finite disjoint union of its connected components; as one component is

5-8 The diagonal in 𝑉 × 𝑉 is closed for the Zariski topology. Therefore, if the Zariski
closed, they all are.

topology on 𝑉 × 𝑉 equals the product topology, then 𝑉 is Hausdorff for the Zariski
topology, hence has dimension 0.
6-1 Let 𝑃 = (𝑎 ∶ 𝑏 ∶ 𝑐), and assume 𝑐 ≠ 0. Then the tangent line at 𝑃 = ( 𝑎𝑐 ∶ 𝑏𝑐 ∶ 1) is

( )
𝜕𝐹 𝜕𝐹 𝜕𝐹 𝑎 𝜕𝐹 𝑏
( ) 𝑋 + ( ) 𝑌 − (( ) 𝑐 + ( ) ( 𝑐 )) 𝑍 = 0.
𝜕𝑋 𝑃 𝜕𝑌 𝑃 𝜕𝑋 𝑃 𝜕𝑌 𝑃

Now use that, because 𝐹 is homogeneous,

𝜕𝐹 𝜕𝐹 𝜕𝐹
𝐹(𝑎, 𝑏, 𝑐) = 0 ⇐⇒ ( ) 𝑎 + ( ) + ( ) 𝑐 = 0.
𝜕𝑋 𝑃 𝜕𝑌 𝑃 𝜕𝑍 𝑃

(This just says that the tangent plane at (𝑎, 𝑏, 𝑐) to the affine cone 𝐹(𝑋, 𝑌, 𝑍) = 0 passes
through the origin.) The point at ∞ is (0 ∶ 1 ∶ 0), and the tangent line is 𝑍 = 0, the line
at ∞. [The line at ∞ intersects the cubic curve at only one point instead of the expected
3, and so the line at ∞ “touches” the curve, and the point at ∞ is a point of inflexion.]

After a linear change of variables, the equation will be of the form 𝑋 2 + 𝑌 2 = 𝑍 2 (this is
6-2 The equation defining the conic must be irreducible (otherwise the conic is singular).

proved in calculus courses). The equation of the line in 𝑎𝑋 + 𝑏𝑌 = 𝑐𝑍, and the rest is

is 2 in case (b).]
easy. [Note that this is a special case of Bezout’s theorem (6.37) because the multiplicity

6-3 (a) The ring

𝑘[𝑋, 𝑌, 𝑍]∕(𝑌 − 𝑋 2 , 𝑍 − 𝑋 3 ) = 𝑘[𝑥, 𝑦, 𝑧] = 𝑘[𝑥] ≃ 𝑘[𝑋],

which is an integral domain. Therefore, (𝑌 − 𝑋 2 , 𝑍 − 𝑋 3 ) is a radical ideal.

(b) The polynomial 𝐹 = 𝑍 − 𝑋𝑌 = (𝑍 − 𝑋 3 ) − 𝑋(𝑌 − 𝑋 2 ) ∈ 𝐼(𝑉) and 𝐹 ∗ = 𝑍𝑊 − 𝑋𝑌.

𝑍𝑊 − 𝑋𝑌 = (𝑌𝑊 − 𝑋 2 )𝑓 + (𝑍𝑊 2 − 𝑋 3 )𝑔,

If

then, on equating terms of degree 2, we would find

𝑍𝑊 − 𝑋𝑌 = 𝑎(𝑌𝑊 − 𝑋 2 ),

6-4 Let 𝑃 = (𝑎0 ∶ … ∶ 𝑎𝑛 ) and 𝑄 = (𝑏0 ∶ … ∶ 𝑏𝑛 ) be two points of ℙ𝑛 , 𝑛 ≥ 2. The

which is false.

∑
condition that the hyperplane 𝐿𝐜 ∶ 𝑐𝑖 𝑋𝑖 = 0 pass through 𝑃 and not through 𝑄 is that
∑ ∑
𝑎𝑖 𝑐𝑖 = 0, 𝑏𝑖 𝑐𝑖 ≠ 0.

The (𝑛 + 1)-tuples (𝑐0 , … , 𝑐𝑛 ) satisfying these conditions form a nonempty open subset
∑
of the hyperplane 𝐻 ∶ 𝑎𝑖 𝑋𝑖 = 0 in 𝔸𝑛+1 . On applying this remark to the pairs (𝑃0 , 𝑃𝑖 ),
we find that the (𝑛 + 1)-tuples 𝐜 = (𝑐0 , … , 𝑐𝑛 ) such that 𝑃0 lies on the hyperplane 𝐿𝐜 but
not 𝑃1 , … , 𝑃𝑟 form a nonempty open subset of 𝐻.

𝐶 = {(𝑎 ∶ 𝑏 ∶ 𝑐) ∣ 𝑎 ≠ 0, 𝑏 ≠ 0} ∪ {(1 ∶ 0 ∶ 0)}

6-5 The subset
 Solutions to the exercises 225

of ℙ2 is not locally closed. Let 𝑃 = (1 ∶ 0 ∶ 0). If the set 𝐶 were locally closed, then 𝑃
would have an open neighbourhood 𝑈 in ℙ2 such that 𝑈 ∩ 𝐶 is closed. When we look
in 𝑈0 , 𝑃 becomes the origin, and

𝐶 ∩ 𝑈0 = (𝔸2 ∖ {𝑋-axis}) ∪ {origin}.

The open neighbourhoods 𝑈 of 𝑃 are obtained by removing from 𝔸2 a finite number of

curves not passing through 𝑃. It is not possible to do this in such a way that 𝑈 ∩ 𝐶 is
closed in 𝑈 (𝑈 ∩ 𝐶 has dimension 2, and so it cannot be a proper closed subset of 𝑈; we
cannot have 𝑈 ∩ 𝐶 = 𝑈 because any curve containing all nonzero points on 𝑋-axis also

∑
6-6 Let 𝑐𝑖𝑗 𝑋𝑖𝑗 = 0 be a hyperplane containing the image of the Segre map. We then
contains the origin).

∑
𝑐𝑖𝑗 𝑎𝑖 𝑏𝑗 = 0
have

for all 𝐚 = (𝑎0 , … , 𝑎𝑚 ) ∈ 𝑘 𝑚+1 and 𝐛 = (𝑏0 , … , 𝑏𝑛 ) ∈ 𝑘 𝑛+1 . In other words,

𝐚𝐶𝐛𝑡 = 0

for all 𝐚 ∈ 𝑘 𝑚+1 and 𝐛 ∈ 𝑘 𝑛+1 , where 𝐶 is the matrix (𝑐𝑖𝑗 ). This equation shows that
𝐚𝐶 = 0 for all 𝐚, and this implies that 𝐶 = 0.
7-2 Define 𝑓(𝑣) = ℎ(𝑣, 𝑄) and 𝑔(𝑤) = ℎ(𝑃, 𝑤), and let 𝜑 = ℎ − (𝑓◦𝑝 + 𝑔◦𝑞). Then
𝜑(𝑣, 𝑄) = 0 = 𝜑(𝑃, 𝑤), and so the rigidity theorem (7.35) implies that 𝜑 is identically
zero.

𝑥↦𝑥𝑛
(𝔸1 ∖ {1}) → 𝔸1 → 𝔸1
8-2 For example, consider

for 𝑛 > 1 an integer prime to the characteristic. The map is obviously quasi-finite, but it
is not finite because it corresponds to the map of 𝑘-algebras

𝑋 ↦ 𝑋 𝑛 ∶ 𝑘[𝑋] → 𝑘[𝑋, (𝑋 − 1)−1 ]

which is not finite (the elements 1∕(𝑋 − 1)𝑖 , 𝑖 ≥ 1, are linearly independent over 𝑘[𝑋],
and so also over 𝑘[𝑋 𝑛 ]).
8-3 Assume that 𝑉 is separated, and consider two regular maps 𝑓, 𝑔 ∶ 𝑍 ⇉ 𝑊. We
have to show that the set on which 𝑓 and 𝑔 agree is closed in 𝑍. The set where 𝜑◦𝑓
and 𝜑◦𝑔 agree is closed in 𝑍, and it contains the set where 𝑓 and 𝑔 agree. Replace 𝑍
with the set where 𝜑◦𝑓 and 𝜑◦𝑔 agree. Let 𝑈 be an open affine subset of 𝑉, and let
𝑍 ′ = (𝜑◦𝑓)−1 (𝑈) = (𝜑◦𝑔)−1 (𝑈). Then 𝑓(𝑍 ′ ) and 𝑔(𝑍 ′ ) are contained in 𝜑−1 (𝑈), which
is an open affine subset of 𝑊, and is therefore separated. Hence, the subset of 𝑍 ′ on
which 𝑓 and 𝑔 agree is closed. This proves the result.
[Note that the problem implies the following statement: if 𝜑 ∶ 𝑊 → 𝑉 is a finite
regular map and 𝑉 is separated, then 𝑊 is separated.]
8-4 Let 𝑉 = 𝔸𝑛 , and let 𝑊 be the subvariety of 𝔸𝑛 × 𝔸1 defined by the polynomial
∏𝑛
(𝑋 − 𝑇𝑖 ) = 0.
𝑖=1
∏
The fibre over (𝑡1 , … , 𝑡𝑛 ) ∈ 𝔸𝑛 is the set of roots of (𝑋 − 𝑡𝑖 ). Thus, 𝑉𝑛 = 𝔸𝑛 ; 𝑉𝑛−1 is

𝑇𝑖 = 𝑇𝑗 , 1 ≤ 𝑖, 𝑗 ≤ 𝑛, 𝑖 ≠ 𝑗;
the union of the linear subspaces defined by the equations
226 Solutions to the exercises

𝑉𝑛−2 is the union of the linear subspaces defined by the equations

𝑇𝑖 = 𝑇𝑗 = 𝑇𝑘 , 1 ≤ 𝑖, 𝑗, 𝑘 ≤ 𝑛, 𝑖, 𝑗, 𝑘 distinct,

9-1 Consider an orbit 𝑂 = 𝐺𝑣. The map 𝑔 ↦ 𝑔𝑣 ∶ 𝐺 → 𝑂 is regular, and so 𝑂 contains

and so on.

an open subset 𝑈 of 𝑂̄ (9.7). If 𝑢 ∈ 𝑈, then 𝑔𝑢 ∈ 𝑔𝑈, and 𝑔𝑈 is also a subset of 𝑂 which

is open in 𝑂̄ (because 𝑃 ↦ 𝑔𝑃 ∶ 𝑉 → 𝑉 is an isomorphism). Thus 𝑂, regarded as a
topological subspace of 𝑂,̄ contains an open neighbourhood of each of its points, and so
must be open in 𝑂.̄
We have shown that 𝑂 is locally closed in 𝑉, and so has the structure of a subvariety.
From (4.37), we know that it contains at least one nonsingular point 𝑃. But then 𝑔𝑃 is
nonsingular, and every point of 𝑂 is of this form.
From set theory, it is clear that 𝑂̄ ∖ 𝑂 is a union of orbits. Since 𝑂̄ ∖ 𝑂 is a proper
̄ all of its subvarieties must have dimension < dim 𝑂̄ = dim 𝑂.
closed subset of 𝑂,
Let 𝑂 be an orbit of lowest dimension. The last statement implies that 𝑂 = 𝑂. ̄
9-2 An orbit of type (a) is closed, because it is defined by the equations

Tr(𝐴) = −𝑎, det(𝐴) = 𝑏,

𝛼 0
(as a subvariety of 𝑉). It is of dimension 2, because the centralizer of ( ), 𝛼 ≠ 𝛽, is
0 𝛽
∗ 0
{( )}, which has dimension 2.
0 ∗
An orbit of type (b) is of dimension 2, but is not closed: it is defined by the equations

𝛼 0
Tr(𝐴) = −𝑎, det(𝐴) = 𝑏, 𝐴≠( ), 𝛼 = root of 𝑋 2 + 𝑎𝑋 + 𝑏.
0 𝛼

𝛼 0
An orbit of type (c) is closed of dimension 0: it is defined by the equation 𝐴 = ( ).
0 𝛼

9-3 Let 𝜁 be a primitive 𝑑th root of 1. Then, for each 𝑖, 𝑗, 1 ≤ 𝑖, 𝑗 ≤ 𝑑, the following
An orbit of type (b) contains an orbit of type (c) in its closure.

equations define lines on the surface

𝑋0 + 𝜁 𝑖 𝑋1 = 0 𝑋0 + 𝜁 𝑖 𝑋2 = 0 𝑋0 + 𝜁 𝑖 𝑋3 = 0
{ { {
𝑋2 + 𝜁 𝑗 𝑋3 = 0 𝑋1 + 𝜁 𝑗 𝑋3 = 0 𝑋1 + 𝜁 𝑗 𝑋2 = 0.

There are three sets of lines, each with 𝑑2 lines, for a total of 3𝑑2 lines.
9-4 (a) Compare the proof of Theorem 9.9.
(b) Use the transitivity, and apply Proposition 8.26.
Hartshorne 1977
Bibliography

Bourbaki, N. TG. Topologie Générale. Éléments de mathématique. Hermann, Paris. Chap. I, II, Hermann
1965 (4th edition); Chap. III, IV, Hermann 1960 (3rd edition) (English translation available from Springer).

Cartan, H. 1963. Elementary theory of analytic functions of one or several complex variables. Éditions
Scientifiques Hermann, Paris; Addison-Wesley Publishing Co., Inc., Reading, Mass.-Palo Alto, Calif.-
London.

Cox, D. A., Little, J., and O’Shea, D. 2015. Ideals, varieties, and algorithms. Undergraduate Texts in
Mathematics. Springer, Cham, fourth edition. An introduction to computational algebraic geometry and
commutative algebra.

Demazure, M. and Gabriel, P. 1970. Groupes algébriques. Tome I: Géométrie algébrique, généralités,
groupes commutatifs. Masson & Cie, Éditeurs, Paris; North-Holland Publishing Co., Amsterdam.

Hartshorne, R. 1977. Algebraic geometry. Springer-Verlag, New York-Heidelberg.

Matsumura, H. 1989. Commutative ring theory, volume 8 of Cambridge Studies in Advanced Mathematics.
Cambridge University Press, Cambridge, first paperback edition. Translated from the Japanese by M.
Reid.

Milne, J. S. 1980. Étale cohomology, volume No. 33 of Princeton Mathematical Series. Princeton University
Press, Princeton, NJ.

Mumford, D. 1966a. Introduction to algebraic geometry. Mimeographed notes based on a 1965–66 course
at Harvard University. Reprinted (with the introduction of misprints) by Springer 1999.

Mumford, D. 1966b. Lectures on curves on an algebraic surface, volume No. 59 of Annals of Mathematics
Studies. Princeton University Press, Princeton, NJ. With a section by G. M. Bergman.

Samuel, P. 1966. Lectures on old and new results on algebraic curves, volume No. 36 of Tata Institute
of Fundamental Research Lectures on Mathematics. Tata Institute of Fundamental Research, Bombay.
Notes by S. Anantharaman.

Shafarevich, I. R. 1994. Basic algebraic geometry. 1, 2. Springer-Verlag, Berlin.

Walker, R. J. 1950. Algebraic Curves, volume vol. 13 of Princeton Mathematical Series. Princeton University
Press, Princeton, NJ. Reprinted by Dover 1962.

Weil, A. 1946. Foundations of Algebraic Geometry, volume Vol. 29 of American Mathematical Society
Colloquium Publications. American Mathematical Society, New York.

227
Index

algebra derivation, 91
affine, 65 desingularization, 196
faithfully flat, 205 differential, 87
finite, 12 dimension, 115
finitely generated, 12 of a topological space, 54
flat, 205 of an affine algebraic variety, 75
algebraically dependent, 34 of an algebraic set, 54

𝔸𝑛 , 36
algebraically independent, 34 pure, 54, 115
direct limit, 21
analytic space, 173 direct system, 21
axiom directed set, 21
separation, 102 discrete valuation ring, 86
divisor, 180
base change, 113 effective, 180
base for a topology, 49 locally principal, 179
basis positive, 180
transcendence, 35 prime, 179
birationally equivalent, 74, 117 principal, 181
boundary, 55 support of, 180
domain
codimension, 54
factorial, 22
Cohen–Macaulay ring, 212
integrally closed, 27
colimit, 21
normal, 27
component
unique factorization, 22
of a function, 49
cone, 132 element
affine over a set, 133 integral over a ring, 25
content of a polynomial, 23 irreducible, 22
convergent, 60 prime, 22
coordinate function, 48
Cramer’s rule, 25 fibre, 113
curve, 54 field of rational functions, 49, 115
elliptic, 37, 131, 136 function
analytic, 173
degree rational, 63
of a hypersurface, 152 regular, 48, 61, 101
of a map, 187 function field, 49, 115
of a projective variety, 154
depth generate, 12
of a local ring, 212 germ

228
 of a function, 60 local equation, 179
graph local parameters, 121
of a regular map, 111 local ring
group regular, 15
symplectic, 99 local uniformizing parameters, 121
group variety, 110
manifold
Hilbert function, 153 complex, 100
homogeneous, 138 differentiable, 100
homomorphism topological, 100
faithfully flat, 205 map
finite, 12 bilinear, 32
flat, 205 birational, 117
local, 14 Frobenius, 70
of algebras, 12 rational, 117
hyperplane, 142 Segre, 145
hypersurface, 50, 142 Veronese, 143
maximal chain, 55
ideal, 13 minimal surface, 219
generated by a subset, 13 morphism
graded, 132, 134 of affine algebraic varieties, 65

𝔪𝑃 , 41
homogeneous, 132 of ringed spaces, 64
maximal, 13
primary, 48 multiplicity, 212
prime, 13 of a point, 84

𝑛-fold, 54
radical, 41
immersion, 105
closed, 72, 104 neighbourhood
open, 104 étale, 122
integral closure, 27 nilradical, 41
integral domain, 12 node, 84
integrally closed, 27 nondegenerate quadric, 217
irreducible components, 46 normalization, 178, 179
isolated in its fibre, 190
isomorphic open affine, 72
locally, 98 open subset

𝜅(𝔭), 189
basic, 49
principal, 49

leading form pencil of lines, 217

of a polynomial, 83 Picard group, 181
lemma point
Gauss’s, 23 factorial, 179
Nakayama’s, 15 multiple, 87
prime avoidance, 78 nonsingular, 82, 87
Zariski’s, 40 normal, 176
linear form ordinary multiple, 84
of a polynomial, 83 singular, 87
linearly equivalent, 181 smooth, 82, 87
local condition, 59 with coordinates in a ring, 125

229
polynomial section of a sheaf, 60
Hilbert, 155 semisimple
homogeneous, 130 group, 98
monic, 25 Lie algebra, 98
primitive, 23 separable
prevariety field extension, 124
algebraic, 100 separable degree, 189
separated, 102 separating transcendence basis, 124
product set
fibred, 113 (projective) algebraic, 131
of algebraic varieties, 109 constructible, 200
of objects, 106 sheaf
tensor, 33 of algebras, 59

Spm(𝐴), 66
projection with centre, 146 singular locus, 83

radical spm(𝐴), 66
of an ideal, 41 stalk, 60
rational map, 117 subring, 12
real locus, 37 subset
regular map, 50, 101 algebraic, 36
affine, 190 analytic, 173
dominant, 51, 72, 116 multiplicative, 16
étale, 94, 118, 120 subspace
faithfully flat, 207 locally closed, 105
finite, 51, 75, 181, 185 subvariety, 105
flat, 207 closed, 71
of affine algebraic varieties, 65 open affine, 100
of algebraic sets, 50 surface, 54
proper, 163 system of local parameters, 121

𝑇1 space, 45
quasi-finite, 51, 185
separable, 124, 187
separated, 191 tangent cone, 83, 93
regular sequence, 212 geometric, 83, 93, 94
regulus, 217 tangent space, 82, 87
resolution of singularities, 196 tensor product
resultant, 166 of modules, 32
ring theorem
associated graded, 93 Bezout’s , 152
catenary, 79 Chinese Remainder, 14
coordinate, 48 going-up, 31
discrete valuation, 86 Hilbert basis, 38
graded, 134 Hilbert Nullstellensatz, 39
local, 14 Noether normalization, 52
noetherian, 14 Stein factorization, 195
normal, 35 strong Nullstellensatz, 42
of dual numbers, 90 Zariski’s main, 189
reduced, 41 topological space
regular local, 15 connected, 45
ringed space, 60 irreducible , 45

230
 noetherian, 45
quasi-compact, 45
topology
étale, 122
Zariski, 39, 133

𝖵𝖺𝗋 𝑘 , 104
variety
abelian, 170
affine algebraic, 65
algebraic, 102
Cohen–Macaulay, 213
complete, 161
factorial, 179
flag, 152
Grassmann, 149
group, 110
normal, 176
projective, 130
quasi-affine, 105
quasi-projective, 130
rational, 128
stably rational, 128
unirational, 128

zero set, 36

231