Methods of Algebraic Geometry in Control Theory - Part I - Scalar Linear Systems and Affine Algebraic Geometry (PDFDrive)

Modern Birkhäuser Classics
Peter Falb
Methods of
Algebraic Geometry
in Control Theory:
Part I
Scalar Linear Systems and Affine
Algebraic Geometry
Many of the original research and survey monographs, as well as textbooks,

in pure and applied mathematics published by Birkhäuser in recent
decades have been groundbreaking and have come to be regarded
as foundational to the subject. Through the MBC Series, a select number
of these modern classics, entirely uncorrected, are being re-released in
paperback (and as eBooks) to ensure that these treasures remain
accessible to new generations of students, scholars, and researchers.
Methods of A lgebraic
G eometry in Control
Theory:Part I
Scalar Linear Systems
and A ffine A lgebraic
G eometry
Peter Falb
Reprint of the 1990 Edition

Peter Falb
Division of Applied Mathematics
Brown University
Providence, Rhode Island, USA
ISSN 2197-1803 ISSN 2197-1811 (electronic)

ISBN 978-3-319-98025-6 ISBN 978-3-319-98026-3 (eBook)
https://doi.org/10.1007/978-3-319-98026-3
Library of Congress Control Number: 2018952484
© Springer Nature Switzerland AG 2018

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or
dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt
from the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained
herein or for any errors or omissions that may have been made. The publisher remains neutral with
regard to jurisdictional claims in published maps and institutional affiliations.
This book is published under the imprint Birkhäuser, www.birkhauser-science.com by the registered
company Springer Nature Switzerland AG part of Springer Nature.
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Peter Falb
Methods of Algebraic
Geometry in
Control Theory: Part I
Scalar Linear Systems and
Affine Algebraic Geometry
1990 Birkhauser
Boston · Basel · Berlin
Peter Falb
Division of Applied Mathematics
Brown University
Providence, Rhode Island 02912
USA
Library of Congress Cataloging-in-Publication Data

Falb, Peter L.
Methods of algebraic geometry in control theory I Peter Falb.
p. cm.-(Systems & control; v. 4)
Includes bibliographical references (p.
Contents: v. 1. Scalar linear systems and affine algebraic
geometry.
ISBN 0-8176-3454-1 (v. 1. : alk. paper)
1. Control theory. 2. Geometry, Algebraic. I. Title.
II. Series.
QA402.3.F34 1990
629.8'312--dc20 90-223
Printed on acid-free paper.
© Birkhauser Boston, 1990

All rights reserved. No part of this publication may be reproduced, stored in a retrieval
system, or transmitted, in any form or by any means, electronic, mechanical, photo•
copying, recording or otherwise, without prior permission of the copyright owner.
ISBN 0-8176-3454-1
ISBN 3-7643-3454-1
Camera-ready copy prepared by the author using TeX.

Printed and bound by Edwards Brothers, Inc., Ann Arbor, Michigan.
Printed in the U.S.A.
9 8 7 6 5 4 3 2 1
Preface
Control theory represents an attempt to codify, in mathematical terms, the principles

and techniques used in the analysis and design of control systems. Algebraic geometry
may, in an elementary way, be viewed as the study of the structure and properties of the
solutions of systems of algebraic equations. The aim of these notes is to provide access to
the methods of algebraic geometry for engineers and applied scientists through the
motivated context of control theory.
I began the development of these notes over fifteen years ago with a series of
lectures given to the Control Group at the Lund Institute of Technology in Sweden. Over
the following years, I presented the material in courses at Brown several times and must
express my appreciation for the feedback (sic!) received from the students. I have
attempted throughout to strive for clarity, often making use of constructive methods and
giving several proofs of a particular result. Since algebraic geometry draws on so many
branches of mathematics and can be dauntingly abstract, it is not easy to convey its beauty
and utility to those interested in applications. I hope at least to have stirred the reader to
seek a deeper understanding of this beauty and utility in control theory.
The first volume deals with the simplest control systems (i.e. single input, single
output linear time-invariant systems) and with the simplest algebraic geometry (i.e. affine
algebraic geometry). This represents the beginning of both system theory and algebraic
geometry. The classical frequency domain methods for control system design were
primarily for scalar linear systems and did not always extend readily to multivariable
systems. So, in a way, this first volume provides the algebraic geometry for understanding
the mathematical structure (in part) of classical scalar control systems. It also provides the
foundation for going further mathematically by developing the basic results of affine
algebraic geometry.
While affine algebraic geometry is quite satisfactory and natural for scalar systems,
the study of multi-input, multi-output linear time-invariant control systems requires the
introduction of projective algebraic geometry. Since an important role for algebraic
geometry to play in control theory is the description of the structure of multi variable linear
systems, it is quite necessary to go beyond the basic results of the first volume.
Consequently, the second volume of these notes is devoted to the study of multi variable
linear systems and projective algebraic geometry. A section, ("Interlude"), indicating in a
brief and heuristic fashion the flavor of Part II, is included in Part I.
While there are many friends, colleagues, teachers, and students, to whom
considerable thanks are owed, I should like especially to express my appreciation to the
vii
Vlll PREFACE
Laboratory for Information and Decision Systems at M.I.T. and its Director, Sanjoy K.
Mitter, for the use of a quiet office (without which this book would still not be finished.)
Thanks are also due to Ann Kostant for the excellent computer preparation of the
manuscript.
Finally, I dedicate this work to my daughters, Hilary and Alison.
'Iable of Contents
Preface vii
O. Introduction.................. . . 1
1. Scalar Linear Systems over the Complex Numbers . . 5
2. Scalar Linear Systems over a Field k . . 9
3. Factoring Polynomials 15
4. Affine Algebraic Geometry: Algebraic Sets 20
5. Affine Algebraic Geometry: The Hilbert Theorems 24
6. Affine Algebraic Geometry: Irreducibility . . . . . 29
7. Affine Algebraic Geometry: Regular Functions and Morphisms I 32
8. The Laurent Isomorphism Theorem . . . . . . . . . . . . . . 39
9. Affine Algebraic Geometry: Regular Functions and Morphisms II . 45
10. The State Space: Realizations . . . . . . . . . . . . . . 50
11. The State Space: Controllability, Observability, Equivalence 58
12. Affine Algebraic Geometry: Products, Graphs and Projections 65
13. Group Actions, Equivalence and Invariants 71
14. The Geometric Quotient Theorem: Introduction 76
15. The Geometric Quotient Theorem: Closed Orbits 83
16. Affine Algebraic Geometry: Dimension 87
17. The Geometric Quotient Theorem: Open on Invariant Sets . 103
18. Affine Algebraic Geometry: Fibers of Morphisms . . . . . . . 105
19. The Geometric Quotient Theorem: The Ring of Invariants . 113
20. Affine Algebraic Geometry: Simple Points . . 128
21. Feedback and the Pole Placement Theorem . 137
22. Affine Algebraic Geometry: Varieties . 145
23. Interlude . . 151
Appendix A: Tensor Products . . 168
Appendix B: Actions of Reductive Groups 175
Appendix C: Symmetric Functions and Symmetric Group Actions 179
Appendix D: Derivations and Separability . 185
Problems . . . . . . 190
References . . . . . . . . . . . . . . . 200
ix
0. Introduction
The overall goal of these notes is to provide an introduction to

the ideas of algebraic geometry in the motivated context of system
theory. While there are a number of excellent mathematical works
on algebraic geometry (e.g. [D-1], [H-2], [K-4], [M-3], [S-2] etc.), these
books do not deal with applications to system theory and are not,
by virtue of their abstraction, always as accessible to engineers and
applied scientists as the potential utility of the concepts warrants. We
seek to provide a bridge for those interested in applications.
While we do not assume considerable mathematical knowledge
beyond the basics oflinear algebra, some simple notions from topology,
and the elementary properties of groups, rings and fields, (see e.g. [B-
1], [J-1], (Z-3]), we do suppose some knowledge of system theory and
applications such as might be provided by a course on linear systems.
Thus, the motivation for such ideas as, say, controllability will not
be considered in detail here nor will we discuss questions of design.
In addition, we deal only with the algebraic side of the subject and
not with the analytic (or differential geometric) side. The comparable
analytic development is examined, for example, in [H-4], [H-6], etc.
Basic material on system theory can be found, for example, in [A-1],
[B-4], [K-2], [R-1], and (W-1].
Part I deals with scalar (i.e. single input, single output) linear
systems and affine algebraic geometry. This represents, in essence, the
beginning of both system theory and algebraic geometry and is the
subject of this volume.
We begin by introducing the four representations of a scalar linear
system, namely: (i) a strictly proper rational meromorphic function
(the transfer function); (ii) a pair of relatively prime polynomials (the
differential operator); (iii) the Hankel matrix (the impulse response);
and, (iv) a triple of matrices (the state space). A key complex of ideas
revolves around the transition theorems from one representation to
another. The first transition theorem (Hankel's theorem 1.4) relates
rationality of a meromorphic function to the finiteness of the rank of
a Hankel matrix.
1
2 METHODS OF ALGEBRAIC GEOMETRY IN CONTROL THEORY: PART I
In order to establish the second transition theorem, which is

known as the Laurent Isomorphism Theorem (theorem 8.12 and corol•
lary 9.15), we require some algebraic geometry. The basic notion of
algebraic set as the zeros of a collection of polynomials is introduced
(section 4. ). Viewing the algebraic sets as the closed sets, we can define
the Zariski topology on the affine N -space AN. This is the natural
topology for algebraic geometry. The crucial connection between the
geometric notion of an algebraic set and the algebraic notion of an
ideal is studied (section 5.). In particular, we prove both the Hilbert
Basis Theorem and the Hilbert Nullstellensatz. These theorems allow
us to develop the key idea of affine algebraic geometry which is to
associate with an algebraic set the ring of regular functions on that
set. This association provides the connection between algebra and
geometry and plays a vital role throughout the development. We can
then prove the Laurent Isomorphism Theorem.
We have, up to that point, examined the relationships between
the transfer function, differential operator, and Hankel representations
of a scalar linear system. We proved that these representations were
"algebraically" the same. We next consider the state space represen•
tation and its relation to the other representations. The realization
maps are introduced (section 10), their relationships studied, and the
set Sf 1 of linear systems of degree n is defined. The concepts of control•
lability and observability and minimality are defined. In addition, the
group action and equivalence resulting from coordinate changes in the
state space are examined. This leads to a theorem on the existence of
minimal realizations (theorem 10.18) and to the State Space Isomor•
phism Theorem (theorem 11.19) which essentially states that minimal
realizations are "equivalent" under coordinate change with a "unique
equivalence". The group action leads naturally to a consideration of
products of affine algebraic sets.
We then turn our attention to studying in a more general way
such group actions as coordinate change and the corresponding equi•
valence relations. Calling the equivalence classes, orbits, we see that
the main problem consists of two parts: first, to find the orbits; and
second, to parameterize the orbits via an algebraic object. Functions
which are constant on orbits are called invariants and we show that the
characteristic realization map !.Rx is an independent complete abstract
invariant (proposition 13.17). This, however, is not enough to have an
appropriate parameterization for state space equivalence. The better
INTRODUCTION 3
concept is that of a "geometric quotient" and we then state the Ge•

ometric Quotient Theorem (theorem 14.20). This may be viewed as
the transition theorem between the state space representation and the
other representations of a scalar linear system. The various proofs we
give for the Geometric Quotient Theorem require us to treat a number
of important algebraic ideas.
In particular, we start with an analysis of the notion of dimen•
sion. We give a topological definition. Then, we introduce the tran•
scendence degree and use it to define an algebraic notion of dimension.
An examination of the crucial concept of integral dependence follows
and culminates in the "Going Down" Theorem (theorem 16.40). We
also prove the Noether Normalization Lemma (lemma 16.43). We can
then show the equivalence of the topological and algebraic definitions
of dimension. Ultimately, this allows us to prove the first part of the
Geometric Quotient Theorem, namely, that the orbits are closed.
We then turn our attention to the second property of a geometric
quotient which we speak of as being "open on invariant sets". One
proof of this property leads us to consider fibers of morphisms (sec•
tion 18). We prove a theorem (18.5) which says that the dimension of
the fibers is not "too small". We then define finite morphisms which
are intimately related to integral dependence and prove the "Going
Up" theorem (theorem 18.11). By using the process of factoring a
morphism through a finite morphism (which is essentially a geometric
form of the Noether Normalization Lemma), we prove that the dimen•
sion of the fibers is "just right almost everywhere". Next we establish
a theorem of Chevalley (theorem 18.19) which states that morphisms
carry constructible sets into constructible sets. Finally we prove the
result required for the second property of a geometric quotient.
The third property of a geometric quotient involves showing that
the ring of invariants is indeed the coordinate ring of the quotient. We
first give an elementary proof based on a "canonical form". The sec•
ond proof involves invariants under the symmetric group and utilizes
results from Appendix C. The third proof is essentially a translation of
the methods of classical invariant theory to the system theory context
and generalizes readily to the multivariable case. With these proofs,
we complete the development of the Geometric Quotient Theorem
(theorem 14.20).
Next we focus on the final major result of scalar linear system
theory, namely: the pole placement theorem. We describe the state
feedback group and its action on the space of linear systems. Then
we solve the pole placement problem (and variants). In particular we
show that the "coefficient assignment" map is surjective if and only if
the system is controllable.
We have, at this point, established the four major algebra•
geometric results of scalar linear system theory, namely: (i) that the
Laurent map is an isomorphism; (ii) that minimal state space realiza•
tions exist and that any two such minimal realizations are uniquely
isomorphic; (iii) that a "geometric quotient" for equivalence under co•
ordinate change exists; and, (iv) that the "poles" are assignable under
state feedback if and only if the system is controllable. In order to
develop a similar theory for multivariable systems, we require a more
general approach. With this in mind, we conclude Part I with an ex•
amination of "abstract affine varieties" (section 22) and an "interlude"
(section 23) which indicates some of the flavor of Part II.
Several additional comments are in order here. First, we do not
strive for the greatest generality nor always for the simplest proof of a
result. Frequently, we use constructive methods with a view towards
application and also give several proofs of a particular theorem. We
believe this is helpful in an introductory work. Second, the exercises
are an integral part of the treatment and are used in the main body of
the text. Finally, we have, of course, drawn on many sources and we
acknowledge their considerable contribution (e.g. [A-2], [D-1], [F-2],
[H-2], [H-8], [M-2], [S-2], [Z-3], etc.) even if explicit mention of them
is not made at a particular point in the text.
Conventions
All rings are assumed commutative with an identity element 1.
A ring homomorphism maps 1 into 1. Neither an integral domain or
a field is the zero ring (i.e. 0 f. 1) and consequently, a prime (or
maximal) ideal is necessarily a proper ideal (i.e. is not the ring itself).
The notation A C B means A is contained in B and A may equal B
while the notation A < B means A is contained in B but A is not
equal to B (in other words, A is strictly contained in B). For sets, the
notation A - B means the complement of B in A.
1. Scalar Linear Systems Over the Complex Numbers
Let us begin by considering single input, single output linear sys•

tems over the complex numbers, C. Such systems may be character•
ized by either a transfer function f(z), or a pair of relatively prime
polynomials p(z) and q(z), or a Hankel matrix H = (hi+j-1)i,j= 1 (the
impulse response), or a triple (A, b, c) where A is a matrix and b, care
vectors. More precisely, we have the following:
Definition 1.1. A scalar linear system over e is any one of the

following: (a) a strictly proper rational meromorphic function J( z)
(i.e. f( oo) = 0) with Laurent series f(z) = '2:.'f= 1 hjz-j about oo; (b)
a pair of relatively prime polynomials p(z) = bo+b 1 z+· · ·+bm_ 1 zm- 1 ,
q(z) = a0 +a 1 z+· · ·+am_ 1 z"'- 1 +zm; (c) a matrix H = (h;+ 1 -di- ,J- 1
with finite rank; and, (d) a triple (A, b, c) with A ann X n matrix (i.e.
A E M(n, n; C)), ban n x 1 column vector (i.e. bE en) and c a 1 x n
row vector (i.e. c E en').
The first question, which arises quite naturally, is: what are the
relationships between these various concepts?
Clearly, if f(z) is a strictly proper rational meromorphic function,
then f(z) = p(z)/q(z) with p(z) and q(z) relatively prime polynomials
of the form given in (b) and conversely.
If (A, b, c) is a linear system then det[zi- A] = QA(z) is a monic
polynomial of degree n and (zi- A)- 1 = '2:.7= 1 q)j(z)Ai- 1 /qA(z)
where q)j(z) is a polynomial of degree n- j. It follows that f(z) =
c(zi- A)- 1 b = '2:.j'= 1 q)j(z)cAJ- 1 b/qA(z) is a strictly proper ratio•
nal meromorphic function since '2:.7= 1 q)j(z)cAi- 1 b is a polynomial of
degree less than n. However, it is not necessarily true that the polyno•
mials '2:.7= 1 q)j(z)c Aj- 1 b and qA(z) are relatively prime (this involves
the notion of minimality which shall be introduced in section 10.).
Example 1.2. Let A =h be the 2 X 2 identity matrix and let
b= [~],c=[1 0 ]. Then IJA(z) = (z-1)- and '2:.j=

0 2 . 1
1 q)j(z)cAJ- b =
(z- 1).
If (A, b, c) is a linear system, then the Hankel matrix H =
5
P. Falb, Methods of Algebraic Geometry in Control Theory: Part I,
Modern Birkhäuser Classics, https://doi.org/10.1007/978-3-319-98026-3_1
( cAi+i- 2 b)iJ= 1 has finite rank by virtue of the Cayley-Hamilt on theo•

rem. On the other hand, if H = (hi+i-diJ= 1 has finite rank n, then,
we have, from the form of the Hankel matrix H, a (unique) recurrence
relation
with ao,a1, ... ,an-1 inC, j = 1, .... If we let
0 1 0 0
0 0 1 0
A=
1
-ao -a1 -an-1
b="n[!],c =[l o ... 0]
where Hn = (hi+i-d~j= 1 , then (A,b,c) is a linear system with

cAi- 1 b = hj for j = 1, ....
The relation between strictly proper rational meromorphic f and
finite rank Hankel matrices is given by the following theorem.
Theorem 1.4 ( Hankel [G-1] ). Let f(z) = 2::_: 1 hiz-i be a

strictly proper meromorphic function and let H 1 = ( h;+ j-1 )i,j= 1 . Then
f is ratio lUll if and only if rank H 1 is finite.
Proof 1: Suppose that f(z) is rational. Then f(z) = p(z)jq(z) =

(bo + b1z + · · · + bn-1Zn- 1)/(ao + a1z + · · · + an-1Zn- 1 + zn) with
p(z), q(z) relatively prime. It follows that p(z) = (2::.: 1 hjz-i)q(z).
Comparing the coefficients of z-i, we get the recurrence relation
for j = 1, .... Hence, all columns of H 1 are in the span of the first n
columns and rank H 1 ::; n is finite.
SCALAR LINEAR SYSTEMS OVER THE COMPLEX NUMBERS 7
Conversely, if rank H f = n, then there is a unique recurrence

relation oft he form (1.5). Setting q( z) = a0 +a 1z+ · · ·+an_ 1zn- 1+zn,
we deduce that o=7=1hjz-j)q(z) = p(z) is a polynomial of degree
less than n and, hence, that f(z) = p(z)jq(z) is rational. Moreover,
p(z) and q(z) are relatively prime for if there were a common factor,
then a recurrence relation of length less than n would exist for the hj
contradicting rank H f = n.
Proof 2: Suppose that f(z) is rational with f(z) = p(z)jq(z) =

(bo + b1z + ··· + bn-1Zn- 1)j(ao + a1z + ··· + an-1Zn- 1 + zn).
Let
0 1 0 0
0 0 1 0
A=
1
-ao -a1 -an-1
0
0
b= , c = [ b0 b1 · · · bn-1 ]
0
1
Then c(zi- A)- 1b = f(z) for if we let x = (zi- A)- 1b, then (zi•
A)x = b, x 1 = 1/det [zi- A], and ZXj = Xj+ 1 for j = 1, ... , n- 1.
But c(zi- A)- 1b = z- 1c(I- Ajz)- 1b = 2::::~ 1 (cAi- 1 b)z-J = f(z) =
2::::;:.. 1 hjz-j and so, hj = cAJ- 1 b. By the Cayley-Hamilton theorem,
ao I + a 1 A + · · · + An = 0 so that c( a0 A j + a 1 A J+ 1 + · ·· + A J+ n )b = 0,
j = 0, 1, ... , or, in other words, aohj + a 1 hj+ 1 + · · · + hj+n = 0 and
rank II f is finite.
Conversely, if rank H f is finite, then, as before, there is a unique

recurrence relation of the form (1.5). If we let
0 1 0 0
0 0 1 0
A=
1
-ao -al -an-l
b~Hn[!1 rh,h2 1' c = [1

hn
0 OJ
then cA.i-lb = hj, j = 1, .... But c(zi- A)- 1 b = z- 1 (!- A/ z)- 1 b =

2::~ 1 (cA.i- 1 b)z-.i = 2::~ 1 h.iz-.i = f(z). Since (zf- A)- 1 = 2::7= 1
cPj(z)A.i- 1 jdet [zi -A] where ¢.i(z) is a polynomial of degree n- j and
cAi- 1 b = hj, we deduce that f(z) = c(zi- A)- 1 b = 2::7= 1 cPj(z)h.i/
det[zi- A]. But 2::7= 1 cPj(z)h.i has degree less than nand so, f(z) is
rational.
Now, by way of summary, we have shown that the concepts
(a), (b) and (c) of definition 1.1 are readily related and that there
are natural ways to go from a triple (A,b,c) to a transfer function
f(z) (i.e. f(z) = c(zi - A)- 1 b) and to a Hankel matrix H (i.e.
H = (cAi+.i- 2 b);j= 1 ). Filling in the gaps will require additional ideas
which we develop in the sequel. Also, we observe that the notions
(b), (c) and (d) of definition 1.1 make sense over any field k and the
various proofs are algebraic in tha.t these proofs do not depend on
convergence. To give an appropriate meaning to (a) and the general
development, we require some algebraic ideas which we develop in the
sequel.
2. Scalar Linear Systems Over a Field k
Based on the intuition developed for the complex number case,

we now turn our attention to the situation over a field k. Our first task
will be to give an algebraic analog of the notion of a. strictly proper
meromorphic function. To do this, we introduce formal power series.
Let R be a commutative ring with (multiplicative) identity 1. If
a, b E R, then we say that a divides b (or is a divisor of b) if there is
a c in R such that b = a c. An element a is irreducible (or prime) if
a is not a unit (i.e. a divisor of 1) and any divisor of a is of the form
ae where e is a. unit. R is an integral domain if there are no proper
divisors of zero.
Let R be an integral domain. Consider the set of ordered pairs
(T,s) with r,s E Rands "I 0. Call two such pairs (r,s) and (T 1,s 1)
equivalent if there is a. t "I 0 in R such that t( T1s- rsl) = 0 and write
( r, s) "" ( r 1 , s1 ). It is easy to see that "" is an equivalence relation. For
example,if(1·,s)"" (T 1,sl) and (1· 1,s 1) ""(r2,s2), then t(r1s-rsl) = 0
and t 1(T 1s2- r2sl) = 0 with t,t' "I 0 imply (tt's1)(rs2- T2s) =
(tt's2)(Ts 1 - T1s) + (tt's)(r1s2- T2s1) = 0 so that (r,s) "" (r2,s2),
We let K(R) be the set of equivalence classes written rjs and we
define addition and multi plication in J( ( R) in the natural way (e.g.
rjs · ri/s1 = rrifssr). K(R) is then a field called the quotient field of
R. We note also that R can be identified with the subring of elements
r/1 of K(R). For example, K(Z) = Q (i.e. the quotient field of the
integers is the rationals).
Let R[ x] be the polynomial ring in one variable over R (i.e. ele•
ments of R[x] are polynomials L:7=l ajxj with coefficients aj in R).
Recall that p(x) E R[x] is homogeneous of degree T if p(x) = axr with
a E R.
Definition 2.1. A formal power series in one variable over R is an

infinite sequence 1t = ( u 0 , u 1 , ... , 1ti, ... ) of homogeneous polynomials
1ti in R[x] with 1ti of degree i or 0.
If 1t = (1to, 1t1, ... ) , v = (vo, 1J1, . •• ) are formal power series, then
we set
9
U + V =(Uo + Vo, U1 + VI, ... )

(2.2)
uv ==(wo,wl, ... )
where Wr = l:i+j=r u;vj. With these definitions, the set of all for•
mal power series over R becomes a commutative ring with identity
( 1, 0, ... , 0, ... ). The ring of formal power series over R is denoted by
R[[x]]. If u = (u 0 ,u 1, ... ,) is an element of R[[x]] with u -:J 0, then the
first index r with Ur -:J 0 is called the order of u, v( u ), and Ur is called
the initial form of u. We note that v( u + v) ~ min { v( u ), v( v)} and
v(uv) ~ v(u) + v(v). We write u = 2:~ 0 uj for the formal power se•
ries u, or, since Uj = a1xi, u = 2:~ 0 ajxi. Note that R[x] is naturally
(identified with) a subring of R[[X]].
Definition 2.3. Let z- 1 be an indeterminate over R. An element

f( z- 1 ) of R[[ z- 1 ]] of positive order is called a strictly proper meromor•
phic function on R. Such an f(z- 1 ) is rational if there are relatively
prime polynomials p(z) and q(z) in R[z] such that f(z- 1 ) = p(z)jq(z).
If J( z- 1 ) is a strictly proper meromorphic function on R, then

f(z- 1 ) = 2:~ 1 hjz-i with hj E Rand, by abuse of notation, we write
f(z) = 2:~ 1 hjz-i and we call2:~ 1 hjz-i the (formal) Laurent series
of f(z).
Now let k be a field. We then have the following:
Definition 2.4. A scalar linear system over k is any one of the

following: (a) a strictly proper rational merom orphic function f( z)
with Laurent series f(z) = 2:~ 1 hjz-i, hj E k; (b) a pair ofrelatively
primepolynomials p(z) = bo+blz+ .. ·+bm-1Zm-I, q(z) = ao+a1z+
. ·· + am-lZm-l + zm in k[z]; (c) a Hankel matrix H = (hi+j-1)iJ= 1
of finite rank with hj E k; and, (d) a triple (A,b,c) with A ann X n
matrix (i.e. A E M( n, n; k)), ban n X 1 column vector (i.e. b E kn),
and c a 1 X n row vector (i.e. c E kn').
We observe that all that was done over C remains valid over the
field k and, in particular, we have:
Theorem 2.5 (Hankel). Let f(z) = 2:~ 1 hjz-i and let H f

(hi+i-I)i,j=l· Then f is rational if and only if Hf has finite rank.
SCALAR LINEAR SYSTEMS OVER A FIELD k 11
We call the rank of H 1 , the degree of f, and we note that if rank

H 1 = n, then the rank of (HJ)n+j,j = 0,1, ... , is also n. In general,
if H = ( h;+ j - l )iJ= 1 , then we say that H has degree n if H has rank
n. We now make the following:
Definition 2.6. Let Rat(n,k) = {f(z): f(z) a strictly proper ra•

tional merom orphic function over k of degree n} and let Hank( n, k) =
{ H : H a Hankel matrix over k of degree n }.
In view of Theorem 2.5, there is a natural map L : Rat(n,k)---+

Hank(n,k) given by L(f) = H 1 which is surjective.
Our goal will be to give an appropriate algebraic structure to both
Rat( n, k) and Hank( n, k) and to show that the mapping L is an "iso•
morphism" of the algebraic structure. We begin by observing that, al•
though on the face of it both Rat( n, k) and Hank( n, k) involve infinite
objects, there are actually only a finite number of parameters required
because of the degree n condition. For example, f(z) is an element
of Rat( n, k) if and only if there are elements bo, ... , bn-l, ao, . .. , an-l
in k such that the polynomials p(z) = bo + b1z + · ·· + bn_ 1zn-l,
q(z) = ao + a1z + · · · + an-lZn-l + zn are relatively prime. In other
words, Rat( n, k) can be viewed as a subset of the affine space AX,n
(= k 2 n). More precisely, Rat(n,k) = {(bo, ... ,bn-l,ao, ... ,an-1) E
k 2 n: f(z) = (bo+b1z+· · ·+bn-lZn-l )/(ao +a1z+· · ·+an-lZn-l +zn)
a strictly proper rational meromorphic function of degree n }. We can
also view Hank( n, k) as a subset of the affine space AX,n. For, if H has
rank n, then, in view of the recurrence relation (1.5), His an element
of Hank( n, k) if and only if there are elements h1, ... , h n, ao, ... , an-1
ink such that hj+n = - 2:::7;01 a;hi+j, j = 1, ... , and no recurrence of
order less than n exists. Alternatively, H is an element of Hank( n, ,'.,)
if and only if the linear equation
(2.7)
has a unique solution Xj = -aj, j = 0,1, ... ,n- 1 with the aj

in k and such that (1.5) is satisfied. More precisely, Hank( n, k) =
{(hl,···,h2n-l,hzn) E k 2n: Hn = (hi+i-dfj=l has rank n i.e. is
non-singular}. With Rat( n, k) and Hank( n, k) viewed in this way, the

map L: Rat(n,k)---. Hank(n,k) given by
(2.8)
is called the Laurent map. We shall eventually show that the Laurent
map is an isomorphism of algebraic structures.
We observe that Hank(n,k) = A~n - V(O) where V(O) =
{(h1, ... ,hzn) E k 2 n: O(h1, ... ,hzn-1) = det Hn = 0}. In other
words, Hank( n, k) is the complement of the "zero set" of a polyno•
mial. Such "zero sets" of polynomials are the basic building blocks
from which algebraic geometry is built. We shall show in the next
section that Rat( n, k) can also be described as the complement of the
"zero set" of a polynomial (called the resultant of p and q).
If Rat(n, k) and Hank(n, k) are viewed as subsets of A~n, then
we can ask for an explicit description of the Laurent map L. Suppose
that (bo, ... , bn-1, ao, ... , an-1) is an element of Rat( n, k) (i.e. f(z) =
(bo + b1z + · ··+ bn-1Zn- 1)/( ao + a1z +·· ·+ an-1 zn- 1 + zn) is rational
of degree n). Then, in view of the second proof of Hankel's Theorem,
we have L(f) = (h1(f), ... ,hzn(f)) where
hj(f) = hj(bo, ... ,bn-1,ao, ... ,an-1) = cAi- 1b (2.9)
and
,b~ m,,~ [bo, ...

0 1 0 0
0 0 1 0
A= ,b._t]
1
(2.10)
and j = 1, ... , 2n.
Example 2.11. Let n 2. Then the explicit formula (2.9)
becomes
h1 = b1
hz = bo- a1b1
h3 = -a1bo- aob1 + aib1 = -aoh1 - a1hz
h4 = -aobo + aibo + 2aoa1b1- a~b1 = -aohz - a1h3
SCALAR LINEAR SYSTEMS OVER A FIELD k 13
so that the hj are polynomials in (bo,b1,ao,at). If J(h1, ... ,h4;

b0 , b1 , a 0 , a 1 ) is the Jacobian matrix of the partial derivatives of the
hj with respect to the b; and a;, then detJ( h1, ... , h4; bo, b1, ao, a1) =
b5- a1b1bo + aobi = -det His the same as the determinant of either
of the matrices
b1
bo- a1 b1
] = O(A,c)
The significance of this observation will become clear in the sequel.
We can also give an explicit description of the map P (actually

L- 1 ) from Hank(n,k) into Rat(n,k). Suppose that (h1, ... ,h2n) E
Hank(n, k) and let H = (hi+i-d~j= 1 (so that det H ::J 0). Then (cf.
(2. 7)) the linear equation H x = h where x' = ( xo, x1, ... , Xn-d, h' =
(hn+1,···,h2n) has a unique solution Xj = -aj, j = 0,1, ... ,n -1.
In fact, if ( H+ )i denotes the j- th row of the adjoint of H, then
(2.12)
for j = 0, 1, ... , n - 1. Motivated by the second proof of Hankel's

Theorem, we let
0 1 0 0
,b~H [!1 ,,~[!

0 0 1 0
A= 0 ... OJ
1
-ao -a1 ... . .. -an-1
so that cAJ- 1b = hj for j = 1, .... We observe that det(zi- A) =

ao + a1z + · · · + an-1Zn- 1 + zn = q(z) and that c(zi- A)- 1b =
p(z)jq(z) = L:~ 1 (cAi- 1 b)z-j = L:~ 1 hjz-j wherep(z) = bo+b1z+
· · · + bn_ 1zn- 1 has coefficients given by the relations
n-r
br = L hjaj+r (2.13)
j=1
14 METHODS OF ALGEBRAIC GEOME1RY IN CON1ROL THEORY: PART I
for r = 0, 1, ... , n- 1, and where we have set an = 1 for convenience.

Since H has rank n, f(z) = p(z)jq(z) is an element of Rat(n,k) and
so the mapping F : Hank( n, k) - Rat( n, k) is given explicitly by
F(hl,···,h2n)=
(bo(hl, · · ·, h2n), · · ·, bn-l(hl, .. ·, h2n), ... , an-1 (h1, ... , h2n))
(2.14)
where the ai and bj are given by (2.12) and (2.13).
Example 2.15. Let n = 2. Then the explicit formula (2.14)

becomes
ao =- (h~- h2h4)j(h1h3- h~)

a1 =- (h1h4- h2h3)j(h1h3- h~)
bo = h1a1 + h2 = (2h1h2h3- hih4- hnj(h1h3- h~)
bl = hl
so that the ai and bi are polynomials in the hi divided by (at most)

det H. If J(bo,bbao,al;ht,h2,h3,h4) is the Jacobian matrix of the
partial derivatives of the bi and ai with respect to the hi, then the
determinant det J (bo, b1, ao, a1; h1, h2, h3, h4) = -1 Jdet H is the same
as the determinant of either of the matrices
(with the bi and ai as functions of the hj)· Again, this observation

will become clearer as we progress.
3. Factoring Polynomials
Now, in order to provide an appropriate algebraic structure for

Rat( n, k ), we need to have an algebraic criterion for the requirement
that two polynomials p(z) and q(z) be relatively prime i.e. have no
common factor. In fact, we would like to develop such a criterion in
terms of the coefficents (b 0 , ... , bn- 1, ao, ... , an-1) of the polynomials.
We shall do this by introducing the "resultant".
Definition 3.1. An integral domain R is a unique factorization

domain if (i) every non-unit in R is a finite product of primes, and,
(ii) the factorization is unique to within order and unit factors. A
polynomial P( x) in R[ x] is primitive if its coefficients have no (non•
unit) common divisor.
If f(x) E R[x] where R is a unique factorization domain, then

f(x) = c(f) h(x) where c(f) E R and h(x) is primitive. c(f) is
called the content of f.
Example 3.2. Let R = Z be the set of integers. The polynomial

2x 2+ 3x + 5 in Z[x] is primitive and if f(x) = 12x 2 + 18x + 30, then
c(f) = 6 (is the greatest common divisor of the coefficients of f(x)).
Lemma 3.3. (Gauss) Let R be a unique factorization domain. Iff( x ),

9(x) E R[x], then c(f9) = c(f)c(9) and hence, the productofprimitive
polynomials is primitive.
Proof: Since f = c(f)h, 9 = c(9)91 and !9 = c(f)c(9)f191, it is

enough to show that the product h91 of the primitive polynomials h, 9 1 is
primitive. If not, then there is a prime p in R which divides the coefficients
of h91· Let h = l:a;xi, 91 = l:flixi and suppose a; 0 ,(3j0 are the first
coefficients which are not divisible by p (note h ,9 1 primitive). Then the
coefficient of xio+ io in h9t is of the form a; 0 {3io + terms divisible by p.
Since this coefficient is divisible by p, we obtain the contradiction that p
divides a; 0 or (3 io.
15
Corollary 3.4. If g( x) divides af( x) with a E R and g primitive,

then g(x) divides f(x).
Proof: If af(x) = g(x)h(x), then ac(J) = c(g)c(h) = Ec(h) where

Eisa unit. Hence, a divides h(x) i.e. ah 2 (x) = h(x) (where h 2 (x) =
(c(J)c 1 )h 1 (x) and h(x) = c(h)h 1 (x)). Since R[x] is an integral domain,
f(x) = g(x)h2(x).
Lemma 3.5. (Euclidean Algorithm). Let R be an integral domain. If
f(x),g(x) are elements of R[x] with degrees m, n respectively and m ~ n,
then there are uniquely determined polynomials d( x ), r( x) such that
1'm-n+l f(x) = d(x)g(x) + r(x) (3.6)
where r( x) has degree less then n and 1' is the leading coefficient of g( x ). *
Proof: Let a be the leading coefficient of f(x). Then lf(x)•

axm-ng(x) has degree less than m and so, by induction, there are poly•
nomials dt(x),rt(x) such that 1'm-n(1'f(x)- axm-ng(x)) = dt(x)
g(x) + r1(x) with degree rt(x) < n. Let d(x) = a1m-nxm-n + dt(x),
r( x) = r 1 ( x) to obtain (3.6). As for uniqueness, if 1'm-n+l f( x) =
d'(x)g(x) + r'(x), then (d(x) - d'(x))g(x) = (r'(x)- r(x)). Since
r'(x)- r(x) has degree less than n, d(x) = d'(x) and r'(x) = r(x).
Theorem 3.7. If R is a unique factorization domain, then R[x] is a

unique factorization domain.
Corollary 3.8. k[x 1, ... , xn] is a unique factorization domain.
Proof (of theorem): We shall prove property (i) first using induction
on the degree. If f( x) has degree 0, then f( x) factors into primes in R.
If f(x) has degree n (with n > 0), then f(x) = c(f)ft(x) with ft(x)
primitive. If ft (x) is not irreducible, then ft (x) = g( x )h( x) and neither
g(x) nor h(x) is constant. But then both g(x) and h(x) have degree less
than n and factor by induction. As for property (ii), it is enough to show
that if a prime p( x) divides f( x )g( x ), then p( x) is a divisor of either f( x)
or g( x ). If p( x) has degree 0, then p( x) is a divisor of either c(f) or
c(g) and hence, of either f( x) or g( x ). If p( x) has positive degree, then
* If m < n, then f(x) = 0 · g(x) + f(x).

FACTORING POLYNOMIALS 17
p( x) is primitive. We suppose that p( x) is not a divisor of f ( x) and that

f( x) has degree m. We let M = R[x ]p( x) + R[x ]f( x ). If a( x) is a non•
zero element of M of lowest degree n with leading coefficient a, then,
by lemma 3.5, am-n+ 1 j(x) = d(x)a(x) + r(x) with degree r(x) <nor
r(x) = 0. Since d(x)a(x) and am-n+lj(x) are in M, r(x) = 0. Let
a(x) = c(a)a 1 (x) with a1 (x) primitive. In view of corollary 3.4, a1 (x)
is a divisor of f(x) and similarly, a1 (x) is a divisor of p(x). Since p(x)
is prime and and does not divide f( x ), a1 ( x) is a unit in R[ x] and so,
is an element of R. Hence a( x) = c( a )al E R and a E M so that
ag(x) = b1 (x)p(x)g(x) + b2(x)j(x)g(x). It follows that p(x) divides
ag( x) and, by corollary 3.4, that p( x) divides g( x ).
We can now prove a lemma which allows us to determine whether or

not two polynomials arc relatively prime.
Lemma 3.9. Let R be a unique factorization domain and let f( x) =

GmXm + Gm-lXm-l + ··· + ao, g(x) = f3nXn + f3n-lXn-l + · · · + f3o
be elements of R[ x J with am or f3n f; 0. Then f, g have a non-constant
common factor if and only if there are F( x), G (x) in R[ x J such that
f(x)G(x) = g(x)F(x) (3.10)
with degree F < m and degree G < n.
Proof: Let f( x) = J{"' ... J;'!'r, g( x) = g~' ... g~' be prime fac•
torizations of f and g. If (say) j 1 is a common factor, then we simply
let F(x) = f(x)ffi(x) and G(x) = g(x)fh(x). On the other hand, if
there is no common factor and (say) am f; 0, then f( x )G( x) = g( x )F( x)
implies f;m; divides F for i = 1, ... , r. Therefore,
r
degree F 2: L m; degree J; = m.
i=l
Now, let F( x ), G( x) be given by
F(x) = Am-lXm-l + · · · + Ao
(3.11)
G(x) = Bn-lXn-l +···+Eo
Then the polynomials J(x) and g(x) will have a non-constant common
factor if and only if there are Ao, ... ,Am-1,Bo, ... Bn-1 not all 0 such
that
(amxm + ··· +ao)(Bn-1Xn-l + ··· + Bo) =

(3.12)
(f3nxn + · · · + f3o)(Am-1Xm- 1 + · · · + Ao)
or, equating coefficients,
amBn-1 = f3nAm-1
am-1Bn-1 + amBn-2 = f3n-1Am-1 + f3nAm-2
(3.13)
Given f(x) and g(x), (3.13) is a set of m + n linear equations in the

m + n "unknowns" Ao, ... , Am-1, Bo, ... , Bn-1· The coefficient matrix
of these equations is
-ao 0 0
0 ··· f3n 0
and so, the system (3.13) will have a non-trivial solution if and only if the
determinant of the following matrix is not zero:
ao Ill elm
0 ao elm } n rows
0
p(f,g) = f3n
elm
(3.14)
f3o
0
f3t
f3o f3n }mmws
0 f3n
The determinant of 3.14 is called the resultant off and g and is denoted
by Res(!, g).
FACTORING POLYNOMIALS 19
Corollary 3.15. f(x) and g(x) have a non-constant common factor

if and only if Res(!, g) = 0.
Now, returning to the system theory context, we can see immedi•
ately that if f(z) = p(z)jq(z) then f(z) will be a strictly proper rational
meromorphic function if and only if Res(p, q) :j; 0. In other words, if
f(z) = p(z)jq(z) and Hf is the Hankel matrix generated by J, then p(z)
and q( z) are relatively prime if, and only if H f has the right rank. This
result may even be viewed as a system-theoretic treatment of the resultant.
If p(z) = bo + b1z + · · · + bn-1Zn- 1 and q(z) = ao + a1z + · · · + zn, then
bo b1 bn-1
0 bo bn-1
} n mw'
Res(p, q) = det (3.16)
ao a1 an-1 1
0 ao an-1 1 } n -1 cow'
so that Res(p,q) = '1/J(bo, ... ,bn-1,ao, ... ,an-d is in fact, a polynomial

in the coefficients of the polynomials p and q. Thus, Rat( n, k) = A~n -
V ( 'ljJ) is the complement of the "zero set" of a polynomial. Such "zero
sets" of polynomials are, as we shall soon see, the basic building blocks
of affine algebraic geometry.
We have given a definition of a scalar linear system over k (see defi•
nition 2.4). Let us denote an "abstract" scalar linear system by the symbol
A. The form f(z) of A is called the transfer function representation of
A; the form (p( z), q( z)) is called the polynomial (or differential operator)
representation of A; tlle form H = (hi+i-diJ= 1 is called the Hankel
(or input-output or impulse response) representation of A; and the form
(A, b, c) is called a state space representation of A. We have shown that
the transfer function and polynomial representations are indeed essentially
the same and tllat tllese representations of degree n are the elements of
Rat( n, k ). The Laurent map L, (2.8), gives a correspondence between tlle
transfer function (or polynomial) representation and tlle Hankel represen•
tation. In fact, L gives a correspondence between systems of degree n.
Investigation of tllis map will involve additional algebraic ideas. The re•
lation between tlle state space representation and tlle oilier representations
will be explored after our algebraic digression.
4. Affine Algebraic Geometry: Algebraic Sets
Let k be a field and let Af ={a= (a1 , ... , aN): ai E k} be the

affine N -space over k.
Definition 4.1. V C Af
is an affine algebraic set if V = {a =
(ai,···aN): fcx(a) = 0, fcx(x) E k[xi,···,xN]}. In other words, Vis
an affine algebraic set if V is the set of common zeros of a family of
polynomials.
Example 4.2. Let n = 2 and 'lj;(bo, b1, ao, a!)= b6- a1bob1 + aobi.
Then V('lj;) = {(bo,bi,ao,ai) EAt,: 'lj;(bo,bi,ao,ai) = 0} and so, for
example, (1,1,0,1) E V('lj;). In system theory terms, this means that
since p(z) = 1 + z and q(z) = z + z 2 = z(z + 1) are not relatively
prime, the degree of the transfer function 1 + zj z + z 2 is one, not two.
Example 4.3. Let N = 2 and f(x 1,x 2 ) =xi- x~. Then V(f) =
{(a 1 ,a 2 ) : ai = aD is a semi-cubic parabola in A~. Note that the
mapping t-+ (t 2 , t 3 ) of A1 into A~ is a bijective map of A1 onto V(f).
We shall in parts I and II assume that the field k is what is

called "algebraically closed" and now work toward making that notion
prense.
Let ]( be an extension field of k (i.e. k C K). An element ~ of
]( is algebraic over kif there is a non-zero polynomial g(x) in k[x]
such that g(~) = 0 (i.e. ~is a root of g). If every element of K is
algebraic over k, then ]( is an algebraic extension of k. We observe
that if ~ E ]( is algebraic over k, then there is a (unique) monic
polynomial rn~(x) of least degree such that m~(~) = 0 and hence, k[~]
is a finite dimensional vector space over k. Conversely, if the dimension
of k[~] over k is finite, then ~ is algebraic over k as 1, ~, ... , C, ... are
linearly dependent over k. Moreover, if ~ is algebraic over k and
rn~(x) = xn + an-IXn-l + · · · + ao with ai E k and (necessarily)
ao of 0, then~( -a 0 1 ~n-l - a 0 1 an-l~n- 2 - • • · - a 0 1a1) = 1 and so
k[~] is a field k(O. Suppose that k C K C L and that K is a finite
dimensional vector space over k with basis u 1 , ... , Un and Lis a finite
20
AFFINE ALGEBRAIC GEOMETRY: ALGEBRAIC SETS 21
vector space over J( with basis v1, ... , Vm. Then L is a finite dimen•
sional vector space over k with basis u;vj, i = 1, ... , n, j = 1, ... , m.
It follows that if K1 is an algebraic extension of k and K2 is an alge•
braic extension of K1, then K2 is an algebraic extension of k. This is
called the transitivity of algebraic extensions.
Definition 4.4. A field k is algebraically closed if every algebraic

extension ]( of k coincides with k.
We shall assume that k is algebraically closed. For example, k = C,

the complex numbers, but k -::J R, the real numbers.
Definition 4.5. An ideal a of a ring R is a subset of R which is an

additive subgroup and satisfies Ra C a (i.e. j,g E a implies f- g E a
and f E a, hE R imply hf E a).
We note that ideals correspond to the kernels of homomorphisms.

If a is an ideal, then Rja denotes the residue class ring modulo a.
Elements of R/ a are usually denoted r where r E R.
Definition 4.6. Let a be an ideal in k[x 1 , ... , xN] and let V( a) =

{a= (at, ... , an): f(a) = 0 for all fin a}. V( a) is called the zero set of
the ideal a. If W C Af, then I(W) = {! E k[x1, ... ,xn]: f(W) = 0}
is called the ideal of W (clearly, I(W) is an ideal).
Example 4.7. Let a= k[x1,x2]· (x1x2- 1) (i.e. a is the ideal

consisting of all multiples of x1x2 -1 ). Then V( a) = {(a 1 , a 2 ) : a 1 a 2 =
1} is a "hyperbola".
Example 4.8. If W = (0, 0) in AL then I(W) = {! E k[ x 1 , x 2 ) :

f has no constant term}. For example, x 1 x 2 + x~ is in I(W) but
(x1 + 3)(x2 + 4) is not.
We now have:
Proposition 4.9. Let a, a; be ideals in k[x 1 , ... , xN] and let W, W;

be subsets of Af. Then
(1) a C a1 implies V( a)=> V( a 1)
(2) W c W1 implies I(W) :) I(W1)

(3) V(Eai) = nV( ai)
(4) I(UWi) = ni(Wi)
(5) V(a n ai) = V(aai) = V(a) U V(a 1)
(6) V(I(W)):) Wand I(V(a)):) a
(7) V(I(W)) = W if and only if Wan affine algebraic set
(8) I(V( a))= a if and only if a = I(W) for some W.
Proof: Many are easy and we will only do (3), (5), and (7).
(3) V(Eai) c nV( ai) since ai c Eai. But if f E Eai, then
f = E/j (a finite sum) with fiE ai and so, f = 0 on nV(ai)·
(5) Since aa 1 C an a1 and an a1 is a subset of of both a and a1,
V(a) u V(a 1) c V(a n a 1) c V(aa 1). If c f/. V(a) U V(ai), then there
are f E a, g E a1 such that (fg)(c) = f(c)g(c) ::f 0. But fg E aa1 and
so, c fl. V(aa1). Thus, V(aai) C V(a) U V(a1).
(7) Obviously, V(I(W)) = W implies that W is an algebraic
set. If W is an algebraic set, then W = V(a) for some ideal a and
I(W) = I(V(a)):) a so that V(I(W)) c V(a) = W c V(I(W)).
Definition 4.10. W C A~ is (Zariski-) closed if W is an affine

algebraic set. An open subset of an affine algebraic set is called a
quasi-affine algebraic set.
We observe that A~ is a topological space under Zariski closure.

Intersections of closed sets are closed by (3); finite unions of closed
sets are closed by (5); and, <P = V(k[xb ... ,xn]) and A~ = V((O)).
We note that the Zariski topology is not Hausdorff.
The sets Rat(n, k) and Hank(n, k) are both Zariski open sets in
A1n. Moreover, the Zariski topology provides a notion of continuity
for maps such as the Laurent map.
Example 4.11. Let L: A1 --+ A1 be given by L(x, y) = (x, -xy).

We assert that Lis continuous (in the Zariski topology). For, if W =
{( u, v) : / 01 ( u, v) = 0} is closed and if we let V = {(x, y) : g01 (x, y) = 0}
where g(){(x, y) = f(){(x, -xy), then Vis closed and L- 1(W) = V. To
see that L- 1(W) = V we note that if (x,y) E V, then fOt(L(x,y)) =
AFFINE ALGEBRAIC GEOMETRY: ALGEBRAIC SETS 23
fo:(x, -xy) = go:(x,y) = 0 so that (x,y) E L- 1(W) and that if

(x,y) E L- 1(W), then 0 = fo:(L(x,y)) = fo:(x,-xy) = go:(x,y) so
that (x, y) E V. For instance, if W = {(u, v) : f(u, v) = u + v = 0},
then g(x,y) = x- xy and V = {(x,y): x(1- y) = 0} (note that here
L- 1 of the line W consists of two lines). As another illustration, let
W = {(u,v): uv- 1 = 0}. Then V = {(x,y): x 2 y + 1 = 0}. Now,
let n = 1 so that Rat(1,k) = {(b 0 ,a0 ) : Res(b0 ,a 0 + z) = b0 :I 0}
and Hank(1,k) = {(h1,h2): h1 :I 0}. We observe that L restricted
to Rat(1, k) is precisely the Laurent map and that L(Rat(1, k)) =
Hank(1,k) and that the Laurent map is injective. Thus, for n = 1,
the Laurent map is a continuous bijective map.
Recalling that Rat(n,k) = A~n- V('lf'7) and that Hank(n,k) =

A~n- V(O) where 'lf'7 = Res(p,q) and 0 = detH are single polynomials,
we have:
Definition 4.12. Let V C A);' be an affine algebraic set and let

f E k[x1, ... ,xn] with f f/_ I(V). The set V1 = {v E V: f(v) :I 0} is
called a principal affine open subset of V.
We shall show that the principal affine open sets form a base for
the Zariski topology and that algebraic sets can be determined by a
finite number of equations in the next section.
5. Affine Algebraic Geometry:
The Hilbert Theorems
Now, let us consider the question of whether or not algebraic sets

can be determined by a finite number of equations.
Definition 5.1. An ideal a in a ring R has afinite basis if there

are it, ... , fn in a such that iff E a, then f = L:~=l rdi with the r;
in R. We then call ft, ... ,Jn a basis of a and write a= (it, ... ,fn)·
If n = 1 so that a = (f) = Rf, then a is a principal ideal. If every
ideal in the ring R has a finite basis, then R is a Noetherian ring.
Proposition 5.2. A ring R is Noetherian if and only if every ascend•

ing chain a 1 C a 2 C · · · of ideals terminates after finitely many terms i.e.
a" = av+l = · · · (ascending chain condition).
Proof: If R is Noetherian and a 1 c a 2 c · · ·, then a = Ua; is

an ideal. But a = (it, ... , fn) implies all f; are elements of some a"
and hence, a" = av+t = · · · = a. Conversely, if there is an ideal a
without a finite basis, then there exist fi E a, i = 1, 2 . . . such that
(It) < (it, fz) < · · · is an infinite ascending chain of ideals.
We can now prove the Hilbert Basis Theorem which allows us to

conclude that any affine algebraic set is defined by a finite number of
polynomial equations.
Theorem 5.3. (Hilbert Basis) If R is Noetherian, then R[ x] is also

Noetherian.
Corollary 5.4. k[x 1 , ..• , XN] is Noetherian.
Proof (of theorem): Let a be an ideal in R[ x] and let €; (a) = {c:

there exists f E a of degree i with leading coefficient c or c = 0}. f; (a)
is an ideal in R for: Ct, c2 E €;( a) imply it - fz = (c1 - c2 )xi + ··· is
in a and rit = (rct)Xi + is in a for r in R. Let e = Uf;(a) so that e
0 0 0
is an ideal in R (as €;( a) C fi+ 1 (a)). Then e has a finite basis r1, ... , rn
24
AFFINE ALGEBRAIC GEOMETRY: THE HILBERT THEOREMS 25
and we let J; (x), i = 1, ... , n be elements of a with leading coefficients

r;. Let d = max (degree fi). If f E a with s = degree f ~ d, then
f = CX 5 + ... with c E e and c = L~=l a;r;. It follows that f(x)•
L a;f;( x )xs-d; E a (where d; = degree j;) and has degree less than s.
Continuing, we obtain <Pt ( x ), ... , </>v ( x) in (It, ... ,jn) such that f- 'E</> j
is an element of a of degree less than d. If f;(a) = (r;J, ... ,riv(i)),
i = 0, 1, ... , d - 1 and f;J ( x) are the corresponding elements of a, then
clearly a = (It, ... , f n.fot, ... , fov(o), ... , fd-lv(d-1)) has a finite basis.
Proposition 5.5. The principal affine open sets (Af)J. where f E

k[x 1 , •.. ,xN].form a base for the Zariski topology.
Proof: Let U be Zariski open so that U = Af- V with V an affine

algebraic set. Let I(V) = (ft, ... .fr). Then V = ni= 1 V((f;)) (by (3)
of proposition 4.9). We claim that U = U;=t (Af)k If v E Ui=t (Af)J;o
then J; (v) f:. 0 for some i and v f/. V. On the other hand, if v E U, then
v f/. V and j;(v) f. 0 for some i so that v E UL 1 (Af)J;.
Example 5.6. Let X = (x;J)L=l be a 2 x 2 matrix with inde-

terminate entries; let Y = [ ~~ ] be a 2 x 1 vector with indeterminate
entries; and, let Z = ( z 1 z 2 ) be a 1 x 2 vector with indeterminate entries.
If 1(X, Y, Z) = det[Y XY] = xnYi + (xzz - xu )YtYz - x12Yi and
w(X,Y,Z) = det [Z' X'Z'] = x12zi + (x22- xn)ztzz- xztzi, then
(A~)'Y, (A~)w and (A~)hw) = (An'Y n (A~)w are principal open sets (as
1 and w are elements of k[X, Y, Z] = k[xu, x12, x21, xn, Yl, yz, z1, zz]).
We now consider the following questions: (1) if a C k[x1, ... , x N] is

a proper ideal (i.e. a f:. k[ x 1 , ... , x N ]), is V (a) non-empty?, and (2) how
can the ideals I(W) for W an algebraic set be characterized i.e. when is
a = I (V (a))? These two seemingly unrelated questions are both answered
by the Hilbert Nullstellensatz.
Definition 5.7. If a is an ideal in a ring R, then the radical of a,

fo, is the set {!: fm E a for some m }.
Clearly, fo is an ideal. Moreover, if a is an ideal in k[x 1 , ... , xN].

then a C fo C I(V(a)) since f(xl,··. ,xN) f. 0 at~= (6, ... ,~N) E
Af implies fm(~) f:. 0 for all m.
Lemma 5.8. Let a be a proper ideal in k[x1, ... , XN]. Then V( a) -=J
¢ (note k is algebraically closed).
Let us assume, for the moment, that this lemma is true. We then have:
Theorem 5.9. (Hilbert Nullstellensatz) If a is an ideal in the ring

k[x1, ... ,xN]. then ..fo. = I(V(a)).
Proof: We must show that given /,!I, ... , fr in k[ x 1 , ... , x N] such

that f vanishes at every common zero of the fi, then there is an m
with fm = Egifi• 9i E k[x1, ... ,xN]. Let XN+l be a new indeter•
minate and consider fi, ... ,fr. 1- XN+d in k[xl, ... ,XN+I]· Then
V((h, ... ,Jr, 1- XNH/)) =¢(by lemma 5.8) and
r
1= L hi(x,xN+t)fi(x) + hr+l(x,xN+t)(1- XN+d(x)) (5.10)
i=l
Setting XN+l = 1/ f(x) and multiplying by a factor fm (to clear denomi•

nators), we get the result.
Now, let us return to lemma 5.8. Before proving the lemma, we

introduce the following:
Definition 5.11. An ideal m in a ring R is maximal if m -=J R and

m c a c R implies a = m or a = R.
We observe that every proper ideal is contained in a maximal ideal

by Zorn's Lemma. We also observe that an ideal m is maximal if and only
if R/m is a field (since the ideals in Rfa are in one-one correspondence
with the ideals of R which contain a). Thus, lemma 5.8 will follow from:
Lemma 5.12. Let m be a maximal ideal in k[ x 1 , ... , x N ]. Then

V(m) -=J ¢.
Proof: Let J( = k[x 1 , ... ,xN]/m and let ~i =Xi (the residue class
modulo m). Then J( is a field, (6, ... ~N) E J( and f E m implies
f( 6, ... , ~N) = 0. Thus, it will be enough to show that the ~i are algebraic
over k (as k is algebraically closed). This will follow from:
AFFINE ALGEBRAIC GEOMETRY: THE HILBERT THEOREMS 27
Lemma 5.13. ([Z-1]) If Lis afield and L[z1, ... ,zN] is afield, then
the z; are algebraic over L.
Proof: For N = 1, if z1 is not algebraic over L, then L[z1] is a

polynomial ring in one variable which is not a field. So we use induction
on N. Since L[z1, ... ,zN] is a field, L(zi) = L1 C L[z1, ... ,zN] and
L1 [zz, ... , zN] C L[z1, ... , ZN]. But L1 is a field and so, by induction,
z 2 , •.• ,zN are algebraic over L1. Thus, by the transitivity of algebraic
extensions, we need only show that z1 is algebraic over L. If not, then z;
algebraic over L1 = L( z1) implies there is a polynomial f( z1) such that
f( z1 )z; satisfies a monic equation over L[zl] (by clearing denominators).
Hence, if g E L[ z1 , ... , z N], then f (z1 ) m g satisfies a monic equation over
L[z1 ] for some m. But z1 is not algebraic over L so that L[z1] is a unique
factorization domain. Hence, if h E L 1 C L[ z1, . .. , z N ], then h would be
of the form h' ( z1 ) j f( z1 )m which is clearly false (by virtue of Proposition
5.14 which follows).
Proposition 5.14. Let R be a unique factorization domain and let

z = xjy be an element of K(R) (the quotient field of R) with x andy
relatively prime. If zn + rn-1zn-I + ·· ·+ ro = 0 with r; E R, then z E R
i.e. y is a unit in R.
Proof: If (xjy)n + rn-I(xjy)n- 1 + · ·· + ro = 0, then xn =

-y(rn-1Xn- 1 + · ·· + royn-I) means y is a divisor of xn. But then
x and y would have a common prime factor if y were not a unit.
Corollary 5.15. M is a maximal ideal in k[x 1, ... , XN] if and only

if M = (x1- 6, ... ,xN- ~n) with ~i E k (i.e. M = I(~),~ =
(6, · .. . ~N) E Af).
Proof: If M = (x1- 6, ... ,xN- ~N). then k[x 1, ... ,xN]/M = k,

a field, and M is maximal. On the other hand, if M is maximal, then
K = k[xi, ... ,xN]/M = k(6, ... ,~N) = k (by lemma 5.13). Hence,
k + M = k[xi, ... ,xN] and x; = ~i + m; with m; EM. Therefore,
(xi- 6, ... ,xN- ~N) C M. By the first part, (x1- 6, ... ,xN- ~N)
is maximal.
Corollary 5.16. If a is an ideal in k[xi. ... , xN] and R =

k[xi, ... ,xN]fa, then every maximal ideal min R is oftheform m =
(x1- 6, ... ,XN- ~N) with ~i E k and xi =a-residue of xi.
Corollary 5.15 means that points of Af and maximal ideals of k[xt,

... , x N] are essentially the same object. In other words, Af = { (6, ... ,
~N): ~i E k} = {M = (x1- 6, ... ,xN- ~N): M a maximal ideal in
k[xt,··· ,xN]} (this set is called the maximal spectrum of k[xt,· .. ,xN]
and is denoted Spm( k[ x 1 , ... , x N ]). Since any maximal ideal M is the
kernel of a k-homomorphism aM of k[xt, ... , XN] onto k and conversely,
we see that Af and the set Homk(k[xt, ... , xN], k) = {a : a is a k-
homomorphism of k[ Xt, ... , x N] into k} are also essentially the same
object. This point of view, as we shall ultimately see, is the most funda•
mental.
6. Affine Algebraic Geometry: Irreducibility
We saw in the last section that points and maximal ideals are
essentially the same object. Now a point is clearly the "smallest" non•
empty algebraic set and so cannot be broken-up into smaller algebraic
sets. This leads us to the following:
Definition 6.1. A non-empty subset V of a topological space X

is irreducible if V is not the union of two proper closed subsets of V
(i.e. V =f V1 u V2 with V1 < V, V2 < V and V1, V2 closed in V).
Example 6.2. Let V = V(a) where a= (X{- Xi) in k[X1,X2].

Then V = V1 U V2 where V1 = V(X1 + X2) and V2 = V(X1- X2) so
that Vis not irreducible. The algebraic sets V1 and V2 are irreducible.
Proposition 6.3. (i) V is irreducible if and only if any two non•

empty open sets in V have a nonempty intersection; (ii) a nonempty open
subset U of an irreducible V is irreducible and dense; and, (iii) if V is
irreducible, then the closure V of V is also irreducible.
Proof:
(i) If v is irreducible and ul' u2 are nonempty open sets in v' then
ul n u2 = ¢would imply v = vl u v2 where vl = v- ul,
V2 = V -U2 are proper closed sets in V. Conversely, ifV = V1 UV2
with vl' v2 proper closed subsets, then ul = v- vl' u2 = v- v2
are nonempty open subsets of v with ul n u2 = ¢.
(ii) Let U1, Uz be nonempty open subsets of U. Then U1 n U, U2 n U
are nonempty open subsets of V and so, U1 n U2 = ( U1 n U) n
(U2 n U) =f ¢. If U were not dense, then (V- U) n U = ¢would
contradict the irreducibility uf V.
(iii) IfV = V1UV2 , then V = (V1nV)U(1/2nV). Since Vis irreducible,
we have (say) V c V1. But V1 is closed in V implies V1 = V so
that vl is not a proper subset.
Recalling that an ideal pin a ringR is prime if fg E p implies that
29
either f E p or g E p (or equivalently, if Rjp is an integral domain),

we have:
Theorem 6.4. An algebraic set V C Af: is irreducible if and only if

I(V) is a prime ideal in k[x 1 , ... ,xN]·
Proof: If I(V) = p is a prime ideal and V = V(I(V)) = V(p) =

V1UVz. then p = I(V1)ni(Vz) implies p = I(Vi) (say) and V(p) = V1 =
V(I(Vl)). On the other hand, if V is irreducible and fg E J(V), then
V c V((fg)) = V(f) U V(g) implies that V = (V(f) n V) U (V(g) n V)
and that either V = V(f) n V or V = V(g) n V so that V c V(f) or
V C V (g). It follows that either f E I(V) or g E J(V).
Corollary 6.5. Af: is irreducible (since I(Af:) = (0) is prime in

k[xi,···,xN]).
Corollary 6.6. Rat( n, k) and Hank( n, k) are irreducible (since

Rat( n, k) and Hank( n, k) are nonempty open subsets of A~n ).
Corollary 6.7. If V is irreducible, then every principal affine open

subset of V is irreducible.
Corollary 6.8. If V C Af: is irreducible, then the residue class ring

k[x 1 , ... ,xN]/I(V) is an integral domain.
We shall, for the time being, call an irreducible affine algebraic set an
affine variety and an open subset of an affine variety, a quasi-affine variety.
Thus, Rat( n, k) and Hank( n, k) are quasi-affine varieties.
Example 6.9. Let GL(2,k) = {g: g E M(2,2;k), detg f' 0}.

Since M(2, 2; k) can be identified with A'f:, GL(2, k) may be identified
with the principal affine open subset defined by the nonvanishing of the
determinant and so, G L(2, k) is a quasi-affine variety. More explicitly, if
~=[~6111 6z]EM(2,2;
62
k)
then (~u,6z,6 1 ,~zz) E A'f: (and conversely). If we let X =

(Xu,X1z,Xz1,X22) with the Xii as indeterminates, then GL(2,k) =
(A'f,).0. where 6(X) = 6(Xu,X1z,X2 1,Xzz) = XuXzz- X12X21.
REGULAR FUNCTIONS AND MORPHISMS I 31
We now have:
Proposition 6.10. Every affine algebraic set V can be represented

as a finite union, V = UV;, of affine varieties V;. The representation is
unique (to within order) if no Vi is superfluous.
Proof: If false, then there is a V which is not decomposable. Such

a V is reducible and so V = V1 u V{ with V1 < V, V{ < V and (say)
V1 not decomposable. Continuing we obtain a chain V > V1 > · · · and
a chain I(V) < I(V1 ) < · · ·. But k[x 1, ... , x N] is Noetherian (Corollary
5.4) and we have a contradiction. Now suppose that
v = uv; = uv;
are two decompositions in which no V; or VJ is superfluous. Then Vi =
V n V; = uVj n Vi implies Vi c Vj for some j as Vi is irreducible.
Similarly, Vj C Vs for some s and so, Vi C Vj C V5 • But then Vi = Vs
(since no Vi is superfluous) and Vi = Vj. In other words, the set {Vi} and
the set {Vj} are the same.
Corollary 6.11. If V is an affine algebraic set, then I(V) = p 1 n · · ·n

Pr where the p; are prime ideals in k[xi, ... , XN] and no Pi is superfluous.
The sets Vi of Proposition 6.10 are called the irreducible components

(or simply components) of V. They play a role in dealing with the notion
of dimension.
7. Mfine Algebraic Geometry:
Regular Functions and Morphisms I
Now the key idea of affine algebraic geometry is to associate with

algebraic sets the "regular" functions on these sets. Let us consider
Af and let Xt,· .. ,xN be the coordinate functions i.e. Xi: Af---+ A1
is the mapping given by Xi(~)= ~i where~= (~~,6, ... ,~N) E Af.
Definition 7.1. A ring R is a k-algebra if k is a subring of R. A k•
alge bra R is finitely generated over k is there are elements r 1 , ••• , r n in
R such that R = k[r1 , •.. , rn]· A finitely generated k-algebra without
nilpotent elements is called an affine k-algebra.
Now, if x 1 , •.. , x N are the coordinate functions on Af, then

k[xt, ... , XN] is a finitely generated k-algebra isomorphic to the poly•
nomial ring inN-variables. Let V be an affine algebraic set and let
k[V] = k[xt, ... ,xN]/I(V) (7.2)
i.e. we define an equivalence relation by setting f equivalent to g if

f - g E I(V) and k[V] is the set of equivalence classes. We note that
k[V] = k[ut, ... , uN] where u; = x; is the equivalence class of x; so
that k[V] is a finitely generated k-algebra. In fact, k[V] is an affine k•
algebra since I(V) = JIM
(by Theorem 5.9) and therefore, = 0 -r
implies r E I(V) imples f E JIM
= I(V) so that f = 0. The
elements of k[V] are polynomial functions on V.
Definition 7 .3. k[V] is called the affine coordinate ring of V. The

elements of k[V] are called regular functions (on V).
Example 7.4. Let V C Ai be the affine algebraic set given by

V = {(~1,6): 6 = ~i}. Let F(X,Y) = ~a;ixiyi be a polynomial
in k[X,Y]
. 2.
and let f be the restriction ofF to V so that J(6,~i) =
~aii~i~1 3 . Then f E k[V]. For instance, if F(X,Y) = Y- X, then
J: V---+ A1 is given by f(6,~i) = ~t- 6- If G(X, Y) = X 2 - X, then
32
G defines the same element of k[VJ as F(X, Y). Note that k[VJ =
k[X, Y]j(Y - X 2 ) = k[X, YJ = k[XJ since Y = X 2 and so k[VJ ~
k[A1J.
One of the reasons that the ring k[VJ is important is that V can
be recovered from it. More precisely, if we denote the set {a : a is a
k-homomorphism of k[VJ into k} by Homk(k[V],k), then we have:
Theorem 7.5. V ~ Homk(k[VJ,k).
Proof: If v E V, then let av: k[V]-+ k be given by av(f) = f(v).

Clearly, avis an element of Homk(k[VJ,k) (e.g. av(fg) = (fg)(v) =
f(v)g(v) = av(J)av(g)). So we define a map '1/J: V-+ Homk(k[V],k)
by setting '1/J( v) = av. Let mv be the kernel of av. Then mv is a
maximal ideal in k[VJ (since av is a k-homomorphism) and so, by
corollary 5.16, has the form Mdi(V) where Me= (x1-6, ... ,xN-
~N) is a maximal ideal of k[x1, ... ,xN]· It follows that v = (6, ... ,~N)
and that 'ljJ is injective. On the other hand, if a E Hom k(k[VJ, k) and
m"' is the kernel of a, then k[V]jm"' = k is afield which means that m"'
is a maximal ideal in k[VJ. Letting M"' = {f: f E k[x1, ... , xN], f E
ma}, we have M"' :J I(V) and M"' is a maximal ideal in k[x1, ... , XN]
(since k C k[x1, ... , XN]/Ma ~ (k[x1, ... , XN]/ I(V))j(Ma/ I(V)) =
k[V]jm"' = k). But then M"' = (xl- 6, ... , XN- ~N) and V(Ma) =
(6, ... , ~N) = v E V so that a = av. In other words, 'ljJ is surjective.
Corollary 7.6. V ~ Spm( k[V]) = {m : m a maximal ideal in

k[V]}.
We also have :
Proposition 7.7. If R = k[fi, ... ,JN] is an affine k-algebra, then

R = k[W] for some affine algebraic set W.
Proof: Let 'ljJ: k[X1, ... ,XN] -+ R be the k-homomorphism given

by '1/J(Xi) = j; and let I = ker '1/J. Then I = VJ since
-n
r E I implies
f = 0 (in k[Xl, ... ,XN]/I) so that j = 0 (as R ~ k[Xl, ... ,XN]/1
-
has no nilpotent elements) and f E I. It follows that R = k[WJ where

W = V(I).
Another of the reasons k[VJ is important is that it allows us to define
the notion of morphism for affine algebraic sets. First, let h , ... , f m be
elements of k[x 1 , ... , x N ]. Then the fi define a polynomial mapping ¢

of A~ into Ak' as follows: if~= (6, .. ·~N) E A~, then ¢(0 =
(JI (0, · · · 'fm(O).
Definition 7.8. If V C A~ and W c Ak' are affine algebraic sets,

then a mapping ¢ : V -+ W is a k-morphism if, for every g in k[W], the
composite function go¢ is in k[V].
Definition 7.9. If V C A~ and W C Ak' are affine algebraic sets,

then a mapping ¢ : V _,. W is regular if ¢ is the restriction to V of a
polynomial mapping.
We shall show that the notions in these two definitions are the same.
The key to doing so is the fact that the regular functions on an affine
algebraic set are the restrictions of polynomial functions.
So, now let V c A~. W c Ak' be affine algebraic sets and let g be
a polynomial function on W (i.e. g E k[W]). If ¢ : V -+ W is a regular
map, then go¢ is clearly a polynomial function on V. Thus, every regular
map is a morphism and we have a mapping ¢* : k[W] -+ k[V] given by
<P*(g) = g 0 ¢ (7.10)
¢* is called the "lifting of ¢".
Proposition 7.11. If cp is regular, then cp* is a k-homomorphism.
Proof: Simply note that cp*(1) = 1 o tp = 1, cp*(f +g) =

(!+g) o cp = f o cp +go cp = cp*(f) + cp*(g), and cp*(fg) =(!g) o cp =
(! 0 cp)(g 0 cp) = cp*(f)cp*(g).
Suppose, on the other hand, () : k[W] -+ k[V] is a k-homomor•

phism. Then, we can define a mapping 1/J : V -+ W which is regular
(hence, a morphism) and for which 1/J* = 8. We do this as follows : if
W C Ak', then k[W] = k[Y1 , ... Ym]/ I(W) and let 'fli be the residue class
of Yi. Then O(y;) = fi E k[V] and J; is a polynomial function on V. Let
1/J: V _,. Ak' be given by
1/J(v) = (fi(v), ... , fm(v)) (7.12)
for v E V. We claim that 1/J(V) C W. Since W = V(I(W)), it will be

enough to show that if v E V and g E I(W), then g( 1/J( v )) = 0. But
g(1jJ(v)) = g(JI(v), ... .fm(v)) = O(g(]h,···,Ym))(v) = 0 since 0 is a

k-homomorphism and g E I(W). Thus, 1jJ: V- Wand 1/J*(h) = ho1/J =
0( h) is in k[V] if h E k(W] by the following :
Proposition 7.13. A mapping </> : V - W is regular if and only if

y; o </> is a regular function for each i, where YI, ... , Ym are the coordinate
functions on Ar.
Proof: If</> is regular, then y;o</> is a polynomial function by definition.
If the y; o </> are all regular functions, then, for any f E k(y~, ... , Ym].
f o </> = f(YI o </>, .. . , Ym o </>) is a regular function.
Let us denote by Hom k(V, W) the set of k-morphisms from V
into W, by Reg(V, W) the set of regular maps of V into W, and by
Hom k(k[W], k[V]) the set of k-homomorphisms of k[W] into k[V]. We
have shown that the natural map </> - </>* of Reg(V, W) into Hom k(k[W],
k[V]) is surjective. It is, in view of Proposition 7.13, also injective (since
</>I f: </>2 implies there is a v E V with </>I ( v) f: </>2 (v) and hence,
(y; o </>I)(v) f: (Yi o </>2)(v) for some i so that </>i(Yi) f: </>2(Yi)). In
other words, we have :
Proposition 7.14. Reg(V, W) ~Hom k(k[W],k[V]).
We shall next show that Hom(V, W) and Homk(k[W],k[V]) also

correspond naturally and so, for affine algebraic sets, morphisms and reg•
ular maps are the same thing.
A k-morphism </> : V - W defines naturally a map</>* : k[W] - k(V]

given by
</>*(g)= g 0 </> (7.15)
This map is called the comorphism of </>. Again, </>* is clearly a k•
homomorphism, and we have a natural map </> - </>* of Hom k(V, W)
into Homk(k[W],k[V]).
Proposition 7.16. Hom k(V, W) ~Hom k(k[W], k[V]).
Proof: It will be enough to show that there is a map 1 : Hom k

(k[W],k[V])- Homk(V,W) such that !(h)*= hand!(</>*)= <j>,
where h E Hom k(k[W], k[V]) and </> E Hom k(V, W). In view of the•
orem 7.5 and the fact that the map o: - o: o h carries Hom k(k[V], k)
into Homk(k[W],k), we can define a map 1(h) : V-+ W by setting

1(h)(v) = w where aw =avo h (avE Homk(k[V],k) corresponding to
v, etc.). If we show that h(g) = go1( h) = 1( h )*(g) for all g E k[W], then
1(h) is a morphism and !(h)*= h. But (go1(h))(v) = g(w) = aw(g) =
(avo h)(g) = av(h(g)) = (h(g))(v). As for showing that 1(¢*) =¢,we
need only prove that al'(¢•)(v) = aq,(v) for all v in V. But, if g E k[W],
then al'(¢•)(v)(g) = ctv(¢*(g)) = av(g o ¢) = g(¢(v)) = ctq,(v)(g).
We say that a k-morphism '¢ : V -+ W is a k-isomorphism if it is

bijective and '¢- 1 is a k-morphism. We then have:
Corollary 7.17. The morphism '¢ : V -+ W is an isomorphism if and

only if'¢* : k[W] -+ k[V] is an isomorphism.
Proposition 7.18. Let '¢ : V -+ W be a morphism and let '¢*
k[W] -+ k[V] be the comorphism of '¢. Then: (i) '¢* is injective if and
only if '1/J(V) = W i.e.'lj;(V) is dense in W; and, (ii) if'¢* is surjective,
then '1/J(V) is closed in W.
Proof: We note that '¢* is injective if and only if'¢*(!) = f o '¢ = 0

implies that f = 0. But this means that '¢* is injective if and only if
I('¢(V)) = 0. Since V(I('¢(V))) = '1/J(V), (i) follows.
If 'lj;* is surjective and a = ker 'lj;*, then k[W]/ a is isomorphic to
k[V]. We claim that '1/J(V) = V (a). If w E 'lj;(V) and f E a, then
f(w) = f('l/J(v)) = '1/J*(J)(v) = 0 where w = '1/J(v) so that 'lj;(V) C V(a).
If wE V( a), then the maximal ideal mw ~ a and there is a maximal ideal
mv in k[V] such that mw = mw/ a"="mv (as k[W]ja ~ k[V]). But then
'¢( v) = w so that w E 'lj;(V). Thus, (ii) is established.
Example 7.19. Let V = Al and W C A1 be given by W =

{(6,6);6 = ~i}. Then the map¢: v-+ W given by¢(~)= (~,e) is
an isomorphism of V and W since k[W] ~ k[AlJ (cf. example 7.4).
Example 7.20. (cf. example 4.11) Let L : A1 -+ A1 be given by

L(x, y) = (x, -xy). Lis a morphism. The algebraic set V(I(L(A1))) is
the closure of L(A1). Since 0 is the only polynomial which vanishes on
L(Ak) i.e. I(L(Ak)) = (0), the closure of L(Ak} is all of A1. However,
L(A1) does not contain the points (0,~). ~ :j:. 0 and so, the image under a
morphism need not be closed.
Example 7.21. Let L: A~n- A~n be given by L(~o, ... ,~n-1.

17o, · ·. , 17n-d = ( c( ~)b, c( ~)A( 17 )b, .. . , c( ~)A( 17 ) 2n-lb) where
0 1 0 0
0 0 1 0
A(17) =
1
-17o -171 ... . .. -17n-1
b= m ,c(n = [{o, ... ,en~I]

(Lis just the Laurent map "extended" to all of A~n). Then Lis a morphism.
Let ¢> : Af+ 2n - A%n be given by ¢(X, Y, Z) =

Example 7.22.
(ZY, ZXY, ... Then ¢> is a morphism. For instance, if
, ZX 2 n-l Y).
n = 2, then ¢>(Xn,Xt2,X21,X22,Yt,Y2,zl,z2) = (ZtYt + Z2Y2,
ZtXuYt + Z1X12Y2 + Z2X21Y1 + Z2X22Y2t, ... ).
Example 7.23. Let V = A1 and W C A~ be given by W =

{(6,6) : ~t- ~i = 0}. The map¢>: V - W with¢>(~)= (e,e)
is a morphism. Since ~ =j: e
implies e e e e
=j: 2 or =j: 3 , ¢> is in•
jective. Obviously, ¢>(0) = (0,0) E W. If (6,6) =I (0,0) is an ele•
ment of W, then, letting~ = 6/6. we have ¢>(0 = (~i/~i.~Va) =
(~rf~i.6~V~r) = (6,6)sothat¢>issurjective. Butk[W] = k[X 2,X 3 ]
(since k[WJ = k[Y1,Y2 ]/(Yl- Yl) ~ k[X 2,X3 ] under the map which
sends Y 1 - X 2 , Y 2 - X 3 ) and so ¢> is not an isomorphism. (In other
words, a bijective morphism need not be an isomorphism.)
The fact that morphisms of affine algebraic sets are restrictions of

morphisms of the ambient affine spaces (i.e. of polynomial maps) is special
and will not be true in general. However, it makes the continuity of
morphisms in the Zariski topology almost obvious.
Proposition 7.24. A regular mapping ¢>: V- W is continuous.
Proof: If W1 is a closed subset of W, then W1 = Wt n W where

wl is closed in Ak" and SO, wl is closed in A;;-. Let¢> be the restriction
of the polynomial map F = (F1, ... , F m) of A~ into A k where F; E

k[xl, ... ,xN]· Then </>- 1(Wl) = F- 1(W1) n V and so, </>- 1(W1) is
closed by the following lemma.
Lemma 7.25. IfF :A~ --+ Ak is a polynomial mapping, then F is

continuous.
Proof: Let W C Ak be an affine algebraic set and let G 1, ... , G r be

a basis of I(W). Then V = {~ E A~: (G; oF)(~)= O,i = l, ... ,r} is
an affine algebraic set. Moreover, if~ E V, then F( ~) E W and conversely
so that V = p- 1 (W).
8. The Laurent Isomorphism Theorem
Now, one of our goals is to show that the Laurent map L :

Rat( n, k) -7 Hank( n, k) is a k-isomorphism. However, Rat (n, k) and
Hank( n, k) are not affine algebraic sets but rather are irreducible prin•
cipal open sets. Thus, we must eventually extend the concepts of
regularity and morphism. We begin with:
Proposition 8.1. Let v1 be a principal affine open subset ofV c Af.

Then there is a natural bijection 1/J 1 between V1 and an affine algebraic
set in A~+ 1 .
Proof: Let h(xl, ... ,xN), ... ,Jr(xl, ... ,xN) be a basis of IkV)
and let h(x1, ... ,xN,XN+l) = 1- XN+tf(xt, ... ,xN)· Let W C Ak +1
be the zero set of the fi and h i.e. W = V ((!I, ... , fr, h)). We define the
map '1/J f by setting
for~= (6, ... ,~N) E vf. Clearly tPJ is injective.

If 1] = (1Jl, ... '1JN,
1JN+l) E W, then fi(1Jt, ... ,1JN) = 0, i = l, ... ,r and h(TJ) = 1-
1JN+d (1]1' ... '1JN) = 0 together imply that ( 1]1, ... '1JN) E vf. Thus 1/Jf
is surjective.
Recalling that Rat(n,k) = (A~n)P where pis the resultant and that
Hank(n,k) = (A~n)9 where f)= detH, we may view Rat(n,k) as the
algebraic set V((l- x 2n+lP)) in A~n+l and Hank(n, k) as the algebraic
set V((l- x 2n+ldet H)) in A~n+l. We shall eventually extend the no•
tion of a morphism in such a way that the mappings 1/J P : Rat( n, k) -7
V((l- X2n+1P)) and ,P 9 : Hank(n,k) -7 V((l- X2n+ldetH)) are k•
isomorphisms. For the moment, however, we shall adopt the point of view
that Rat(n, k) = V((l- X2n+1P)) and Hank(n, k) = V((l- x2n+ 10)).
From this point of view, we can describe the Laurent map L as follows:
let (xt, ... ,x2n,X2n+l) and (y1, ... ,Y2n, Y]n+l) be coordinate functions
on A!n+l and define a polynomial map L : Ak n+l -7 A!n+l by setting
39
for j = 1, 2, ... 2n and

(Y2n+l oL)(xl,···,X2n,X2n+I) = -X2n+l (8.4)
for n 2: 2*, where
0 1 0 0
0 0 1 0
A(x) =
1
(8.5)
b ~ m ~ [x~, ,C(x) ,x,.]
(cf. (2.9)). Then Lis simply the restriction of L to Rat (n,k) = V((1-
Xzn+ 1 p) ). Thus, L is a morphism and hence, is continuous. L is smjective
by Hankel's Theorem (2.5).
We can also study the map F : Hank( n, k) -+ Rat( n, k) from the
same point of view. Again, we let (YI, ... , Y2n, Y2n+l) and (x1, ... , X2n,
Xzn+l) be coordinate functions on Akn+l and we define a polynomial map
F : Akn+l -+ Akn+l by setting
n-s
Yn.:+ll
(xs 0 F')(Yl, · · ·, Y2n, Y2n+l) = Yn-s+l- L(Y+)s+j [ YjY2n+l
j=l
Y2n
(8.6)
for s = 1, ... , n - 1 and
( Xn o F)(Yl, · · ·, Yzn, Y2n+l) = Yl (8.7)
and
(8.8)
* For n = 1, y 3 o L= x3 .
THE LAURENT ISOMORPHISM THEOREM 41
for t = 1, ... , n and
( Xzn+l o F)(Yt, .. ·, Yzn, Y2n+!) = -Y2n+l (8.9)
l
for n ~ 2 *,(cf. (2.14)) where y+ is the formal adjoint of the matrix
Y= [~~ ~: Yn+l
Yn
(8.10)
~n Yn:+l Y2~-1
and yt is its t-th row. Then F is simply the restriction of F to Hank
( n, k) = V((1- Y2n+l detY)). Thus, F is also a morphism.
Example 8.11. Let n = 2 and let (x1, ... ,x 4 ,xs) and

(Yt, ... , Y4, Ys) be coordinates on A%. Then Rat(2, k) = V((l- x 5 (xj -
x1xzx4 + x~x3)) and Hank (2, k) = V((1- Ys(YtY3- yn)). The maps
L and F are given by
Yl L = X2
0
Y2 0 L = Xt - X2X4
Y3 0 L = -XtX4 - X2X3 + X2X4

- 2
-
Y4 o L = -x1x3 + XtX 42 + 2x2x3x4- x2x 43
Ys o L = -xs
- 2 3
x1 oF= (2YtY2Y3- YtY4- Y2)Ys
X2 0 F = Yl
- 2
X3 oF= (Y2Y4- Y3)Ys
X4 oF= ( -YlY4 + Y2Y3)Ys
xs oF= -ys
and L and F are their restrictions to Rat(2, k) and Hank(2, k) respectively.
We observe that (Yt oL )(y3oL)-(y2oL) 2 = -x1x2x4 -x~x3+x~x~ -xi+
2x1x2x4 -x~x~ = -xi +x1x2x4 -x~x3 so that {(y1 oL )(y3oL )-(yzoL ) 2 }
(Ys o L) = xs(xi - x1x2x4 + x~x3) = 1 i.e. L maps Rat(2, k) into
Hank(2, k). Similarly, F maps Hank(2, k) into Rat(2, k). Let us examine
* For n = 1, x3 oF= Y3·

the map F o L. We have x1 o (F o L) = {2xz(xl - x2x4)( -x1x4 -

XzX3 + XzX~)- X~( -X1X3 + X1X~ + 2XzX3X4- XzX~)- (x1- XzX4) 3}
(-xs) = x1(xi-x1x2x4+x~x3)xs = x1, xzo(FoL) = Xz, x3o(FoL) =
{(x1 - XzX4)( -X!X3 +XIX~+ 2x2X3X4 - XzxD- ( -X1X4 - XzX3 +
XzX~) 2 }( -Xs) = X3, X4 o (F o L) = {( -Xz )( -X!X3 + X1X~ + 2xzX3X4-
xzx~) + (x1- xzx4)( -x1x4- xzx3 + xzx~)}( -xs) = x4(xi + x~x3-
x 1x 2x 4 )x 5 = x 4, and x 5 o (F o L) = xs so that F o Lis the identity on
Rat(2, k ). Similarly, L o F is the identity on Hank (2, k ). Thus, L is a
k-isomorphism.
If we examine what we have done so far, we can see that the Laurent
map L is a surjective morphism and that the map F is a morphism. If
we can show that L is injective and that F = L -l, then we will have
demonstrated the following :
Theorem 8.12. The Laurent map L is a k-isomorphism between

Rat (n,k) = V((1- Xzn+1P)) and Hank(n,k) = V((1- Xzn+1B)).
In other words, the notions (a), (b), (c) of a scalar linear system are
essentially the same.
Lemma 8.13. Let u 1 , ... , Un be variables and let
1 0
0 1
-uz
Uz U3 Un 1
U3 u4 1 0
gu =
1
1 0 0
0 0
1 0 -uz
Au= 0 1
0 0
THE LAURENT ISOMORPHISM THEOREM 43
Proof: Simply compute.
Lemma 8.14. Let L be the Laurent map and let F be the associated
map of Hank (n,k) into Rat (n,k). Then LoF is the identity on Hank ( n, k).
Proof: Let h; be the restriction of y; to Hank( n, k) for i = 1, ... , 2n

and let H = (hi+ j-l )i,j=l be the restriction of Y to Hank ( n, k ). Then
1/Yzn+l = detH on Hank (n, k) and we may write the equations ofF as
follows:
H [-Xn~loF] = [hn:+ll
(8.15)
-Xzn oF h2n
[xl oF··· Xn oF]= [h1 · · · hn]9F
where gp is of the form 9u of lemma 8.13 with u1 = +xn+l oF, ... , Un =
+x 2 n oF. What we must then show is that
(8.16)
for j = 1, ... , 2n where c( x oF), A( x oF), and bare given by (8.5). How•
ever, in view of the previous lemma and the fact that g F is a nonsingular
n x n-matrix, we have
(8.17)
where Ap is of the form Au oflemma 8.13 with u1 = +xn+I oF, ... , Un =

+x2n oF. However, it follows from (8.17) and the form of Ap and
the Cayley-Hamilton Theorem and (8.15) that equation (8.16) holds for
j = 1, ... , 2n.
Lemma 8.18. Let u1, ... , Un be variables and let Au. 9u be of the
form given in lemma 8.3. If En is the unit column vector with I in the n-th
row, then [En Aufn · · -A~-lfn] = g;; 1.
Proof: Simply compute 9u[cn Aufn ···A~- len](= In)·
Example 8.19. Let n = 4. Then lemma 8.18 states that
["'
0 0
~][~
U3 U4
1 0 1
u3
u4
u4
1 0 1 -u4 -u3
I
-u4
+
2
u4
l
1 0 0 -u4 -u3 + u~ -u2 +
2u3u4- u~
Lemma 8.20. If L is the Laurent map and F is the associated map

of Hank ( n, k) into Rat (n, k ). Then F o L is the identity on Rat ( n, k ).
Proof: We must show that x 1 o F o L = x 1 on Rat ( n, k ). Now, for

j = 1, ... , n, we have
[xi 0 (F 0 L) ... Xn 0 (F 0 L)]= [cb ... cA:n-lb]gp
where gp is of the form 9u of lemma 8.13 with u 1 = Xn+j
o (F o L),
j = 1, ... , n. Suppose for the moment that x n+ j o ( F o L) = x j for
j= 1, ... , n. It will then follow from lemma 8.18 that
- -n-1
[x1 o (F o L) · · · Xn o (F o L )] =[xl · · · Xn][cnAcn ···A cn]9F
=[xl · · · Xn]
i.e. x 1 o ( F o L) = x j for j = 1, ... , n. Since L maps Rat ( n, k) into
Hank (n,k), we have [cA:n+ib]j= 1 = H(x)[-xn+ilJ=l where H(x) =
[cA:i+i- 1 b]~j=I· But [xn+j o (F o L)]j= 1 = H(x)- 1 [cA:n+ib]j= 1 and so
we have Xn+j o (F o L) = Xn+j for j = 1, ... ,n. Thus, the lemma is
established.
In view of lemmas (8.14) and (8.20), we have established the Laurent
Isomorphism Theorem (8.12). To fully complete the proof, we shall have
to extend the ideas in such a way that the mappings 1/J P and 1/; 8 are k•
isomorphisms. This we shall do in the next section.
We note that the rationale for our proof of the Laurent Isomorphism
Theorem is based on the structure of the state space representation and on
the relation of the state space representation to the other representations.
The fact that lemmas 8.13 and 8.18 can be done "globally" is special to
the scalar situation.
Regular Functions and Morphisms II
We now turn our attention to the issue of generalizing the notions

of regular function and morphism. We recall that a complex mero•
morphic function is said to be regular at a point if the point is not a
pole of the function. Similarly, a complex meromorphic function is
said to be regular on a set if it has no poles in the set.
Let V be an affine variety (i.e. V is irreducible) so that k(V] is
an integral domain and jJ = I(V) is a prime ideal. We denote the
quotient field of k[V] by k(V) and call it the function field of V. If~ is
a point of V and m~ = {!: f E k(V], !( 0 = 0}, then m~ is a maximal
ideal in k(V] and m~ = MdP where M~ = (xi - 6, ... , XN- ~N) is
the maximal ideal in the polynomial ring k[X1 , ... , XN] consisting of
those polynomials which vanish at ~.
Definition 9.1. The set O~,v = {f lg : f,g E k[V], g(~) =/= 0} is

called the local ring of V at ~.
We observe that O~,v is indeed a local ring (i.e. a ring with exactly
one maximal ideal) since the set of functions fIg with (!I g)( 0 = 0
forms a ma..ximal ideal and if(! lg)(O =/= 0, then gl f is an element
of O~,v (i.e. every element of O~,v which does not vanish at ~ is
invertible). We also observe that O~,v is an integral domain with the
same quotient field as k[V], namely the function field k(V) of V.
Now, 1 (j. m~ and if /I,h (j. m~, then /Ih (j. m~ so that O~,v =
{! lg : g (j. md. We have:
Definition 9.2. Let R be a ring. A subset M of R is multiplicatively

closed if 1 EM, 0 (j. M and m, m' EM implies mm' EM.
Some examples of multiplicatively closed sets are M 1 = {fn : f =J

O,n ~ 0} and Mp = R- p = {r: r (/. p} where jJ is a prime ideal.
If Af is a multiplicatively closed subset of R, then we can define an
equivalence relation on R xMas follows: (r,m) ,.._, (r',m') if there is
45
an m" such that m"(m'r- mr') = 0. We denote the set of equivalence

classes by RM and we let rim denote the equivalence class of (r,m).
By defining addition and multiplication of these fractions r lm in the
usual way i.e.
(rim)+ (r'lm') =(m'r + mr')l(mm')

(rlm)(r' lm') =(rr' lmm')
we make RM into a ring.
Definition 9.3. RM is called the quotient ring of R with respect to

M or the ring of fractions of R with respect to M.
If M = Mf = {r: f i- O,n 2 0}, then we write R 1 in place of

RM, and if M = Mp = {r : r ~ jJ} where jJ is a prime ideal, then we
wirte Rp in place of RMP.
Proposition 9.4. If R is an integral domain, then so is RM.
Proof: If (rlm)(r' lm') = 0 then rr' = 0 and so r orr' is 0.

Proposition 9.5. If p is a prime ideal, then Rp is a local ring.
Proof: The elements rIm with r E jJ form an ideal m P = p Rp in

Rp. If rl/m 1 is not in mp, then r 1 is not in 1J and ml/r 1 is in Rp i.e.
rl/m 1 is a unit in Rp. Thus mp is the unique maximal ideal in Rp.
Proposition 9.6. If R is an integral domain, then the natural homo•

morphism r --+ 1)1 of R into RM is an isomorphism.
Proof: It is enough to show that the map is injective. But r I 1- r' I 1 =

0 means that m( r - r') = 0 for some m( i- 0) in M. Since R is an integral
domain, r - r' = 0.
We now note that
where jJ = I(V). In other words, Ocv is the quotient ring of k[V] with
respect to the multiplicative set k[V] - m€ = {! : f ¢ md.
REGULAR FUNCTIONS AND MORPHISMS II 47
Definition 9.8. Let V be an affine variety and let ~ E V. Call a

function f : V ---+ A~ regular at ( is there is an open neighborhood U
(in V) of ( and polynomials F1, F2 in k[ X 1, ... , X N] such that F2 does
not vanish at any point of U and f(u) = F1(u)jF2(u) for all u in U. A
function is regular on W C V if it is regular at each point of W.
Proposition 9.9. A function f :V ---+ A~ is regular at ( if and

only iff E O€,v·
Proof: Iff E 0€.V• then f = gfh with h(~) =f 0. Let G and H be

polynomials whose restrictions to V are g, h respectively. If U = Vh is the
principal open subset of V defined by h, then ( E U C V, H (() = h( O =f
0, and f(u) = g(u)Jh(u) = G(u)JH(u) for all u in U. Conversely, if
f is regular at (, then f = Fl/ F2. F2 =f 0 at (. Letting !I and h be
the restrictions of F1 and F2, respectively, to V, we have f = h Jh with
f2(() =f 0 so that j E 0€,V·
This immediately leads to the following:
Definition 9.10. If U is open in V, then Ov(U) = n€EUO€,v is the

ring of regular functions on U.
Proposition 9.11. k[V] = Ov(V) = n€EVO€,V (i.e. the local rings

O€,v determine k[V] and the notion of regularity in definition 9.8 extends
our previous definition.)
Proof ([M-2]): Clearly k[V] c O€.V· So suppose that f E O€,v

for all ( E V. Let af = {g E k(Xl,···,XN] : gf E k[V] where
g = g mod p}. Then a f is an ideal in k(X1, ... ,XN]. Since f E O€,v
f = FdG€ where F€,G€ E k(XI,··. ,XN] and G€(0 =f 0. Hence,
G€ E OJ and~(/. V(aJ) for all ( E V. But p = I(V) C a1 implies
V (a 1) C V = V ( p) so that V (a f) = ¢. However, 1 E a 1 by the
Nullstellensatz and so, f E k[V].
Proposition 9.12. Ov(VJ) = k[V] 1 (i.e. the ring of regular func•

tions on Vf is the quotient ring of k[V] with respect to the multiplicatively
closed set M 1 = { r : n 2 O}J.
Proof: If hf r is an element of k[V] 1, then hf r E Ocv for~ E Vf.

Conversely, if h E Ov(V1) and if ah = {g : gh E k(V]}, then ah is

an ideal in k[V]. We claim that V( ah) C V((f)). If ~ E V1 , then
h = ht/h2 with h2(~) ::f 0. But h2 E ah and so ~ rJ. V( ah)· However,
V(ah) C V((f)) implies that f E Vah· It follows that rh E k[V) and,
hence, that h E k[V] 1 .
We are now ready to extend the notion of a morphism. In particular,

we have:
Definition 9.13. Let V and W be quasi -affine varieties. A mapping

¢ : V --+ W is a morphism (or k-morphism) if (i) ¢ is continuous and (ii)
¢*(g) = g o ¢ is a regular function on V if g is a regular function on W.
We also say that a k-morphism 'lj; : V --+ W is a k-isomorphism if it is
bijective and 'lj;- 1 is a k-morphism.
Proposition 9.14. Let Vf be a principal affine open subset of the

variety V c A.f/ and let 1/J 1 be the natural bijection between V1 and
the affine algebraic set Win At'+ 1 (see proposition 8.1). Then ¢ 1 is a
k-isomorphism.
Proof: Let F(X1 , ... ,XN) E k[X1 , ... ,XN) such that F restricted
to V is f. If g is an element of k[W] and G(X1, ... ,XN,XN+l) =
2.".::~= 0 G;(X1, ... , XN )Xfv+ 1 is a polynomial whose restriction toW is g,
then 1/Jj(g) = go,P 1 = H(X1, ... ,XN)/Fn(Xl, ... ,XN) restricted to Vf
for some polynomial Hand so, ,Pj(g) E Ov(V1 ) = k[V] 1 . lnotherwords,
,P 1 is a morphism. But we have k[W] = k[Xt, ... , XN, XN+l]/(I(V),
1-XN+ 1 F) which is k-isomorphic toR= k[V)[XN+ 1)/(1- fXN+!)k[V]
[X N+ 1 ]. So, it will be enough to show that R is k-isomorphic to k[V]J.
We define first a k-homomorphism a : k[V][XN+l] --+ k[V] 1 by setting
a equal to the identity on k[V] and a(XN+l) = 1/ f. Clearly a(1 -
f X N +1) = 0 and so we can extend a to a k-homomorphism a : R --+
k[V] 1. Since every element of k[V] 1 is of the form gJr. a is surjective
and in fact, every element of k[V]J is of the form a(gXN+l) where
g E k[V). If a(gXN+ 1) = gf r = 0, then, in view of proposition 9.4,
g = 0 and SO, a is injective.
Corollary 9.15. The Laurent map L : Rat( n, k) --+ Hank( n, k) is a

k-isomorphism.
REGULAR FUNCTIONS AND MORPHISMS II
49
Proof: The maps t/J P and '1/Je are k-isomorphisms.
Example 9.16. (cf. example 6.9). Now let GL(2, k) = {9 : 9 E

n)
M(2,2 ;k), det 9 i 0} =(At)~ where ~(X)= ~(X11,X12,X21,X
= XnXn - X12X2 1. Let (3: G--+ G be given by (3(9) = 9- 1 i.e. if
9= [911 912]
921 922
then
-912]
911 /(911922 - 921912)
(3 is
Then (3 is a k-isomorphism of GL(2, k) onto GL(2, k). Clearly
L(2,k) ) = k[Aib and that f3*(X 11 ) =
bijective. Observing that OA4(Gk
Xzz/~. ,8*(X12) = -X12j~. (3*(X21) = -Xn/~. (J*(Xz

z) =Xu/~.
(3* ( 1/ ~) == ~. we can readily see that (3* is a k-isom orphism .
10. The State Space: Realizations
We have, up to now, examined the relations between the transfer

function, polynomial, and Hankel representations of a scalar linear
system. We proved that these representations were "algebraically" the
same. In this section, we shall consider the state space representation
and examine its relation to the other representations. We begin by
introducing the following:
Definition 10.1. The triple (A,b,c) is a (state space) realization

of the transfer function f(z) if f(z) = c(zl- A)- 1b or, equivalently,
if f(z) = 2:~ 1 (cAi- 1 b)z-i. Similarly, the triple (A,b,c) is a (state
space) realization of the Hankel matrix H = (hi+i-diJ= 1 if h; =
cAi- 1 b fori= 1, ....
We observe that if (A, b, c) is a realization off( z) (or of H) with
A E M(n, n; k), then the degree of f(z) (or of H) is at most n by
the Cayley- Hamilton theorem. We shall view the triple (A, b, c) as a
2 2
point in A~ +Zn and we consider a polynomial ring k[A~ +Zn] which
we write k[(Xij),(Y;),(Zj)], i,j = 1, ... ,n or simply k[X,Y,Z]. We
will define some morphisms of A~ 2 + 2 n into A~n which we shall call
realization maps.
First, let B 0 ... , Bn_ 1, A 0 , ••. , An_ 1 denote the coordinate func-
tions on A~n and define a polynomial mapping !J\ 1 : Ak 2 +2 n-+ A~n by
setting !R 1 (X,Y,Z) = (Bo(X,Y,Z), ... ,Bn-1(X,Y,Z), Ao(X,Y,Z),
... , An-1(X, Y, Z)) where
J( X y Z) = Zadj(zl- X)Y ( 10. 2 )
z, ' ' det(zl- X)
Bo(X, Y, Z) + ·· · + Bn-1(X, Y, Z)zn- 1
Ao(X, Y, Z) + · · · + An-1(X, Y, Z)zn-1 + zn
We call !R f the transfer function realization map.
Example 10.3. Let n = 2. Then
zi-X= [z-Xu -X12 ]

-X21 z- Xn
50
THE STATE SPACE: REALIZATIONS 51
so that det(zl- X) = Z 2 -(Xu + X2z)z + XuXn - X12X21 and

Ao(X, Y, Z) = XnXn- X12X21, A1(X, Y, Z) = -(Xn + Xz2). Also
adj(zJ _X)= [z-X21X22 X12 ]

z-Xn
so that Z adj(zi -X)Y = {(Z1Y1)+(Z2Y2)}z+Z1( -X22Y1 +X12Y2)+

Z2(X21Y1- XuY2) and Bo(X,Y,Z) = ZY · (A1(X,Y,Z)) + ZXY,
B 1(X,Y,Z) = ZY. Thus IJ\1 is a morphism. Let (bo,b1,ao,a1) be
any element of At. If we let
A = [ -ao
0 1 ] , b = [0 ] , c
-a1 1 = [bo b1]
then 1Jt 1 (A,b,c) = (bo,b 1 ,ao,ar) so that IJ\1 is smjective. If g = (9ij)

is an element of GL(2, k), then det[zJ -gXg- 1] = det[g(zJ -X)g- 1] =
det[zJ- X] and (Zg- 1 )(gY) = ZY, (Zg- 1 )(gXg- 1 )(gY) = ZXY so
that 1Jt 1 (A,b,c) = 1Jt 1 (gAg-I,gb,cg- 1 ) for all (A,b,c). Thus, IJ\1 is
not injective.
Now, let H 1 , ••• , H 2 n denote the coordinate functions on A~n
and define a polynomial mapping IJth : A~ 2 +Zn ~ A~n by setting
!Jth(X, Y, Z) = (H1(X, Y, Z), ... ,H2n(X, Y, Z)) where H(X, Y, Z) =
(ZXi+i- 2Y)iJ= 1 so that
(10.4)
for j = 1, ... , 2n. We call IJth the Hankel matrix realization map.
Example 10.5. Let n = 2. Then Hi(X, Y, Z) = (ZXi- 1Y) for

j = 1, ... ,4 so that H1(X, Y, Z) = ZY = Z1Y1 + Z2Y2, H2(X, Y, Z) =
ZXY = Z1XnY1 + Z1X12Y2 + Z2XnY1 + ZzXnYz, etc. IJth is clearly
a morphism. We now observe that
so that
X 2 = (Xu+ X22)X + (X12X21 - XnX22) ·I
and hence that H 3 (X, Y, Z) = (ZX 2Y) = (X11 + X 22 )ZXY +
(X12X21-XnX22)ZY =(Xu +Xn)H2(X, Y, Z)+(X12X21-XnXn)
Ih(X, Y, Z). Similarly, we have H4(X, Y, Z) =(Xu +X2 z)H3 (X, Y, Z)

+(X12X21- XuXzz)Hz(X, Y, Z). Let (0, 0, ~' 1J) be an element of A'!
with ~' 1] i- 0. Then 9th" 1 ((0, 0,~, 1J)) = 4> is empty i.e. 9\h is not sur•
jective. We also note that 9\h(gAg- 1,gb, cg- 1) = 9\h(A, b, c) for all
A,b,c where g E GL(2,k) so that 9\h is not injective.
Finally, let ( U1, . .. , Un, . .. , Uzn) denote the coordinate functions
on A~n and define a polynomial mapping 9\x : AZ 2 +Zn -+ A~n by set•
ting 9\x(X, Y, Z) = (U1 (X, Y, Z), ... , Un(X, Y, Z), ... , Uzn(X, Y, Z))
where
U·(X
J
1
, Y , Z) =ZXj- Y j = 1, ... ,n (10.6)
Un+t(X, Y, Z) =- Xt(X) t = 1, ... ,n
where det(zJ- X)= zn- XI(X)zn-I- · · ·- Xn(X). We call 9\x the
characteristic function realization map.
Example 10.7. Let n Then U1(X, Y, Z) = ZY, Uz(X, Y, Z) =

= 2.
ZXY, U3(X, Y, Z) = -(Xu + Xzz)
and U4(X, Y, Z) = XuXzz -
X 21 X 12 • 9\x is clearly a morphism. Let (6,6,6,~4) be any element
of A'! and let
A= [ -~4 -~3 J , b = [n . c = [6 + 66 6]
Then X1(A) = 6, Xz(A) = ~4, U1(A,b,c) = 6, and Uz(A,b,c) = 6 so
that 9\x is surjective. As before 9\x is not injective since 9\x( A, b, c) =
9\x(gAg-I,gb,cg- 1 ) forgE GL(2,k).
Proposition 10.8. The realization maps 9\ f, 9\h and 9\x are mor•
phisms. Both 9\ f and 9\x are surjective but 9\h is not. None of the maps
is injective.
Proof: (Left to reader.)
Now let us raise the following seemingly unrelated questions: (1)

how are the realization maps related?, and (2) what is the structure of
the open sets 9tj 1 (Rat(n,k)) = 9tj 1 ((A~n)p) and 9th" 1 (Hank(n,k)) =
9\ h1 ( (A~ n) a)? To consider these questions, we introduce three maps 1/J f,
1/Jh and 1/Jx of A~n into itself. Let ( x 1, ... , x 2n) be the coordinate functions
on A~n. The map 1/Jf: A.~n-+ A~n is simply the Laurent map L i.e.
7J;1 (x) = (c(x)b, ... ,c(x)A:j- 1 (x)b, ... , c(x)A 2 n- 1 (x)b) (10.9)
where c( x ), A( x) and bare given by (8.5). Note that 'lj;1 is a morphism and
that ,p1 : (A~n)p--+ (A~n)e is an isomorphism. The map ,Ph: A~n --+ A~n
is given by ,Ph(xt, ... , X2n) = (ht(x), ... , h2n(x)) where
i = 1, ... ,n
n
(10.10)
hn+j(x) =- LX2n+l-ihi+j-t(x) j = 1, ... ,n
i=l
Note that ,Ph is a morphism. Finally, the map 'lj;x : A~n --+ A~n is given
by ,Px(x) = y where
n-r
Yr = LX jX2n+l-j-r + Xn-r+l r = 1, ... ,n- 1
j=l
(10.11)
Yn = Xt
Yn+j = X2n+l-j j = 1, ... ,n
Clearly 'lj;x is a morphism. The morphism 'lj; 1 is neither smjective nor

injective (since, say, for n = 2, Y3 = -x3y1 - x4y2, Y4 = -x3Y2 - x4y3
and ,P 1 (0,0,~,1J) = (0,0,0,0) for all ~.1J). Similarly, the morphism ,Ph
is neither surjective nor injective. However, we do have:
Proposition 10.12. 'lj;x is an isomorphism.
Proof: Define a morphism of A~n--+ A~n by setting = Yn, ... , XrXt
= Yn-r+l -I:;:: Xj(Y)Y2n+l+j-ro r = 2, ... 'n, Xn+j = Y2n+l-j• j =

1, ... , n. A straightforward calculation shows that this morphism is ,P;1 .
Example 10.13. Let n = 3. Then 'lj;x may be written
Y1 xs X4 1 0 0 0 Xt
Y2 X4 1 0 0 0 0 xz
Y3 1 0 0 0 0 0 X3
Y4
= 0 0 0 0 0 1 X4
Ys 0 0 0 0 1 0 xs
Y6 0 0 0 1 0 0 X6
If ,Px(~) = ,Px(e). then ~4 = Y6(~) Y6(e) = ~~. 6 = Ys(~) =

Ys(e) = ~~. ~6 = Y4(~) = Y4(e) = ~~. 6 = Y3(0 = Y3(e) = ~L
6 = Y2(0- ~16 = Y2(t)- ~~~i =~~.and 6 = Y1(0- ~56- ~16 =

Y1(t)- ~t~i- ~~~~ = ~~· Thus, '1/Jx is injective. Also, if 77 E A%, then,
setting 6 = 773, 6 = 772- 173176· 6 = 771 -775773 -ry6(772 -773776). ~4 = 176•
6 = 775, ~6 = 774, we have '1/Jx ( 0 = 77 so that '1/Jx is surjective. The map
'I/J;_ 1 may be written
X1 0 0 1 0 0 0 Y1
X2 0 1 -y6 0 0 0 Y2
X3 1 -y6 yg - Y5 0 0 0 Y3
X4 0 0 0 0 0 1 Y4
xs 0 0 0 0 1 0 Y5
X6 0 0 0 1 0 0 Y6
For instance, xz = Yz- Y3Y6 = X4X1 + x2 - X1X4 and X3 = Y1- Y6Y2 +

(yg- Ys)Y3 = X3 + x2x4 + x1xs- x4(x4x1 + x2) + (x~- xs)x1 = X3 so
that 'I/J;_ 1 o '1/Jx = I.
Proposition 10.14. The realization maps !Jl f• !Jlh, !Jlx and the as•
sociated morphisms 'I/J 1, '1/Jh. '1/Jx satisfy the following relations:
'1/Jx 0 9lx = 9l f
'1/Jjo!Jlf=!Jlh
'1/Jh o !Jlx = !Jlh
'1/Jh='I/Jjo'I/Jx
Proof: We first show that '1/Jx o !Jlx = !Jlf. Let ('1/Jx o !Jtx)(A,b,c) =
(Y1, ... , Y2n)· Then Yr = - I:j;:;( cAi- 1b)Xn+1-(j+r) + cAn-rb, for
r = l, ... ,n- 1, Yn = cb, and Yn+j = -Xn-j+l for j = 1, ... ,n.
If !JtJ(A,b,c) = (bo, ... ,bn-l,ao, ... ,an-!). then aj = -xn-j• j =
0 ' ... 'n - 1' bn-1 -- c b' and br -- "'n-r- L.Jj=1 ( c Ai- 1b) aj+r + c An-r- 1b'
1
r = n- 2, n- 1, ... , 0. It follows that aj = Yn+l+i• j = 0, ... , n- 1,

bn-1 = Yn· and br = Yr+1• r = 0, ... , n- 2. In other words, '1/Jx o 9lx =
!Jl 1 . To show that '1/Jho!Jlx = !Jlh, we observe that, by the Cayley-Hamilton
Theorem, A n+j-l -- "'n L.Ji= 1 Xn+1-•·Ai+j-2 , J· -- 1 , ... , n, an
d h ence, th at
'1/Jh(!Jtx(A,b,c)) = (cb, ... ,cAn-lb, ... ,I:~=lXn+l-ihi+j-1(x), ...) =
(cb, ... ,cAn-lb, ... cAi-lb, ... ,cA 2n-lb) = !Jth(A,b,c). If we show
that '1/Jh = 'lj;1 o '1/Jx, then '1/JJ o !Jl f = '1/JJ o '1/Jx o !Jlx = '1/Jh o !Jlx = !Jlh.
In view of the fact that Xi(A(y)) = -Y2n+l-i (where A(y) is given by

(8.5)), we have Xi(A('l/Jx(x))) = -xn+i• i = 1, ... ,n, and hence, that
n
c( '1/Jx(x)).Jn+j- 1 ('1/Jx(:r))b = LXn+I-i(A('l/Jx(x)))·
i=1
n
c('l/Jx(x)).Ai+i- 2('1/Jx(x))b =- L X2n+l-ic('l/Jx(x)).Ai+j- 2('1/Jx(x))b
i=l
for j = 1, ... , n. In other words, to show that '1/Jh = 'f/; 1 o '1/Jx it will be
enough to show that c('l/Jx(x)).Ai- 1 ('1/Jx(x))b = Xj for j = 1, ... ,n. But
c( '1/Jx(x )) = [x1 ... Xn]9u and A( '1/Jx(x )) = Au where u1 = X2n, ... 'Un =
Xn+I and gu. Au are of the fonn given in lemma 8.13. The result then
follows from lemma 8.18.
Corollary 10.15. Let A : A~n - k be given by A = '1/Ji(p) = p o

'1/Jx· Then '1/Jx : (A~n).\ -+ (A~n)p and '1/Jh : (Ain).\ -+ (Akn)9 are k•
isomorphisms. Thus, the diagram
is a commutative diagram of k-isomorphisms.
Corollary 10.16. The principal affine open sets 9t! 1 ((Ain)p),

9th" 1 ((A~n) 9 ) and 9t~ 1 ((A~n).\) are all the same.
Proof: 9t! 1 ((A~n)p) = {~: (p o 9tJ)(~) =f 0}, 9th" 1 ((Ain)9) = {~:

(8o9th)(~) ::j 0}, and 9t~ 1 ((A~n).\) = {~: (Ao9tx)(~) ::j 0}. But 9tf =
'1/Jx o 9tx and so, p o 9t f = p o '1/Jx o 9tx = A o 9tx as A = p o '1/Jx· Similarly,
9th= 'f/Jho9tx and Bo9th = (Bo'f/Jh)o9tx with 8o'f/Jh = (Bo'f/; 1)o'l/Jx,
- (p o '1/Jx) = -A since ( (} o '1/J1) = - p.
Example 10.17. Let n = 2. Then the maps 'f/;1, '1/Jh. '1/Jx are given
by:
'lf'>J(XI,Xz,X3,X4) = (xz,XI- X4X2, -X3X2- X4(X1- X4Xz),o:(x)),

o:(x) = -x3(x1- x4xz)- x4(-x3x2- x4(x1- x4xz)))
'!f'>h(XI, Xz, X3, X4) = (xi, Xz, -X4XI - X3X2, -X4X2 - X3( -x 4x 1 - X3Xz ))
'lf'>x(xi,xz,x3,x4) = (x1x3 + Xz,x1,x4,x3)
The functions p, B, >. are given by:
p(x1,xz,x3,x4) =xi- X4X1X2 + X3X~

B(x1,xz,x3,x4) =x1x3- x~
A(xl,Xz,X3,X4) =X~+ X1X2X3 + X4Xi
Clearly po'lf'>x =>..Now (Oo'!f'>h)(xi,xz,x3, x4) = B(xl,xz,-x4xl•
XJXz,-x4xz-x3(- x4xi-x3xz)) = x1(-x4x1-x3xz)- x§ = -x4xi•
x1xzx3- x§ = >.(xi,xz,x3,x4). Thus we can see that '!f'>h maps (A~n)>
into (A~n) 8 . Let us show directly that '!f'>h is bijective as a map of (A~n)>
into (A~n)a. If 1f;h(6,(z,6,(4) = '!f'>h((f,(L(~,(~). then 6 = (f, 6 =
(~. -(46 - 66 = -(~(i - (~(~. and -(46 - 6( -(46 - 66)
-(~(~ - (H -(~(i - (~(2 ). It follows that
((~- (4)6 -((~- 6)6 =0

((~- (4)6 -((~- 6)( -(46 - 66) = 0
and, hence, that(~= (4, (~ = 6 since >.(6,(z,6,(4) = (i + 666 +

(r(4 =J 0. In other words, '!f'>h is injective. If (ry1 ,ry2 ,ry3 ,ry4) E (A~n) 8
and if we let 6 = 'TJl• 6 = TJz, 6 = (TJzTJ3- TJITJ4)1(TJITJ3- ryi), (4 =
(TJz1]4- 17'JJI(TJITJ3- TJ~). then '!f'>h(6,6,6,(4) = (TJI,TJz,(-TJ!TJzTJ4 +
TJ1 TJ5 - TJi TJ3 +TJ1 TJz TJ4) I ( TJ1 TJ3- TJi), (- TJi TJ4 +TJz TJ5 - TJz TJ5 +TJ1 TJ3 TJ4) I ( TJl TJ3 -
ryi)) = ( TJ1, TJz, 1]3, TJ4) and 'If'> h is surjective. Since 'If'>;: 1 = 'If'>;_ 1 o 'If'> j \ 'If'> h
is a k-isomorphism.
WeletSf, 1 betheprincipalaffineopenset!J\;_ 1 (A~n):\ = IJ\j 1 (A~n)p
= IJ\h" 1 (A~n)e and we call Sf, 1 the set of linear systems of degree n.
We should like to characterize the elements of Sf 1 . Denoting (A~n):\
by Char( n, k ), we observe that if (A, b, c) is an ~lement of Sf, 1 , then
IJ\ 1 (A,b,c) is an element of Rat(n,k), IJ\h(A,b,c) is an element of
Hank(n,k), and IJ\x(A,b,c) is an element of Char(n,k). Since IJ\ 1 and
IJ\x are surjective and ,p 1 is an isomorphism between (A~n)p and (A~n)e.
we can see that, conversely, given an element f(z) ofRat(n,k) or H of

Hank(n,k) or~ of Char(n,k), there is an element (A,b,c) of Si\ such
that !.R 1(A,b,c)" = "f(z) or !.Rh(A,b,c)" = "H or !.Rx(A,b,c)'' = "~.
In other words, if f(z) is a transfer function of degree nor if His a Han•
kel matrix of degree n or if ~ is a characteristic representation of degree
n, then there is a linear system of degree n which is a realization of f ( z)
or of H or of~. We summarize this in the following:
Theorem 10.18. If X = (A, b, c) is an element of sr,l. then fx(z) =

c(zi- A)- 1 b is an element of Rat(n,k). Conversely, if f(z) is an
elementofRat(n,k), then there is an X= (A,b,c) in srl such that
fx(z) = f(z). Similarly, if X = (A,b,c) is an element oi srl. then
Hx = (cAi+i- 2 b)iJ= 1 is an element of Hank(n,k) and conve~sely, if
H is an element of Hank( n, k ). then there is an X = (A, b, c) in sr,l
with Hx = H. Finally, if x = (A, b, c) is an element of Sf 1 , then ~x =
( cb, ... , cAn- 1 b, -x 1 (A), ... , -xn(A)) is an element of Char( n, k) and
conversely, if~ is an element of Char( n, k ), then there is an x = (A, b, c)
in sr,1 with ~X=~. Such an X is called a minimal realization of f(z) or
of H or off
We have thus established that the set of linear systems of degree

n, Sf, 1 , is the set of minimal realizations for Rat( n, k) or Hank( n, k)
or Char( n, k ). In other words, !.R 1 : Sf, 1 --+ Rat( n, k ), !.Rh : Sf, 1 --+
Hank( n, k ). and !){X : sr,l --+ Char( n, k) are surjective morphisms.
11. The State Space:
Controllability, Observability, Equivalence
We noted in section 10 that the various realization maps were

never injective. We should like to examine this issue and to further
characterize the set of linear systems of degree n.
We again view the triple (A, b, c) as an element of A~ 2 + 2 n and
we consider the polynomial ring k[A~ 2 +2 n] = k[( Xij ), (Yi), ( Zi )] =
k[X, Y, Z]. We will define some morphisms of Af+ 2 n into Ak 2 which
we shall call the controllability and observability maps.
Let Y : A~ 2 + 2 n ----+ Ak 2 be the map given by
Y(X, Y, Z) = [Y XY · · · xn- 1Y] (11.1)
The morphism Y is called the controllability map. Similarly, let Z :
l
Af+Zn ----+ Ak 2 be the map given by
Z(X,Y,Z) = [~X (11.2)

zxn-1
The morphism Z is called the observability map. We note that Y is
independent of Z and that Z is independent of Y. Finally, we let
H : A~ 2 +Zn ----+ A~n be given by
H(X, Y, Z) = Z(X, Y, Z) · Y(X, Y, Z) ( 11.3)
and we call the morphism H the Hankel matrix map.
58
THE STATE SPACE: CONTROLLABILITY, OBSERVABILITY, EQUIVALENCE 59
Example 11.4. Let n = 2. Then
Y(X y Z) = [Y1 XnY1 + X12Y2]

' ' Y2 X21Y1 + X22Y2
Z(X y Z) = [ Z1 Z2 ]
' ' Z1X11 + Z2X21 Z1X12 + Zz2X22
and
zy ZXY ]
H(X, Y, Z) = [ ZXY zx2y
We note that rank[H(A,b,c)] :::; min[rank Y(A,b,c), rank Z

(A,b,c)] :::; n for all (A,b,c) in Af+ 2 n. This leads us to make the
following definition:
Definition 11.5. The scalar linear system (A,b,c) is control-

lable if rank Y(A,b,c) = nand observable if rank Z(A,b,c) = n. If
rank H(A,b,c) = n, then (A,b,c) is minimal.
We note that a system is minimal if and only if it is both con•

trollable and observable. Letting 'Y(X, Y, Z) = det Y(X, Y, Z) and
w(X, Y, Z) = det Z(X, Y, Z), we can see that')' and w are elements of
k[X, Y, Z]. If Ic ={!)and 10 = (w), then Vc = V(Ic) and V 0 = V(Io)
are affine algebraic sets and VcUV0 = V(Iclo) = V(Icnlo) = V(I(J) =
Vh where 8(X, Y, Z) = w(X, Y, Z) · 'Y(X, Y, Z) = det H(X, Y, Z). It
follows immediately that x = (A, b, c) is controllable if and only if
x cf. Vc; that x = (A,b,c) is observable if and only if x cf. V 0 ; and,
that X = (A, b, c) is minimal if and only if X C/. Vc u Vo = vh. In
other words, the controllable systems are the elements of the principal
affine open set (A~ 2 + 2 n),; the observable systems are the elements of
the principal affine open set (Af+ 2 n)wi and, the minimal systems are
the elements of the principal affine open set (Ak 2 +2 n)(J·
Proposition 11.6. (A~ 2 + 2 n)(J = Sf, 1 (so that both notions of a

minima/linear system are the same).
Proof: Simply observe that (0 o !:Rh)(X, Y, Z) = 8(X, Y, Z). Thus,

if (A,b,c) E Si\. then !:Rh(A,b,c) E (A1n)B and (0 o !:Rh)(A,b,c) =
'
6(A,b,c) -::j:. 0. Conversely, if 6(A,b,c) -::j:. 0, then !.Rh(A,b,c) E (A~n)e

and (A,b,c) E Sf, 1 .
Thus, we may characterize sr,l as the set of systems which are both
controllable and observable.
Now, let us ask the following question: when do two minimal triples
x = (A,b,c) and x 1 = (A 1,b1,ct) generate the same f(z) in Rat(n,k)
or H in Hank( n, k) or ~ in Char( n, k )? In order to answer this question,
we shall introduce an equivalence in A~ 2 + 2 n based on the "action" of a
group of transformations.
Definition 11.7. Let GL(n,k) = {g: g E M(n,n;k),detg -::j:. 0}.

G L( n, k) is called the general linear group over k.
We observe that since M( n, n; k) may be identified with Ak~ GL( n, k)

may be identified with the principal affine open subset defined by the
nonvanishing of the determinant which is a polynomial in the entries of a
matrix.
Proposition 11.8. Let x = (A,b,c) be an element of A~ 2 + 2 n and

let g be an element of G L( n, k ). If x 1 = (A1, b1, c1) where
A1 = gAg- 1, b1 = gb, c1 = cg- 1 (11.9)
then x 1 is an element of Af+Zn (obviously) and
(11.10)
for the realization maps !.R f, !.Rh and !.Rx·
Proof: To show that !.R 1 (x) = !.R 1(x 1), we simply note that fx(z) =
c(zi- A)- 1b = (cg- 1)g[zg- 1g- g- 1gAg- 1gt 1g- 1(gb) = c1 (zl•
A1)- 1 b1 = fx 1 (z). As for showing that !.Rh(x) = !.Rh(x1). we have
c1At 1b1 =( cg- 1)(gAg- 1)j- 1(gb)

=(cg- 1)gAj- 1g- 1 (gb) = cAj-Ib
for j = 1, 2, .... Finally, in view of what was just proved, we need only
observe that Xi( AI) = x;(gAg- 1 ) = Xi( A), i = 1, ... , n, in order to
establish that !.Rx(x) = !.Rx(xi).
We say that GL(n,k) "acts" on A~ 2 + 2 n via (11.9) and we write

g · x = g · (A,b,c) for this "action" so that
g · x = g · (A,b,c) = (gAg-l,gb,cg- 1 ) (11.11)
In effect, the action of G L( n, k) corresponds to coordinate change in the

state space. Now, noting that G L( n, k) is a quasi-affine variety and that
A~ 2 + 2 n is an affine variety, we may view the action (11.11) as a mapping
'1/Ja of GL(n,k) X A~ 2 + 2 n into Af+ 2n i.e., letting G = GL(n,k), we
have '1/Ja : G X Af+ 2 n --+ A~ 2 + 2 n given by
'1/Ja(g,x) = g · x = (gAg- 1 ,gb,cg- 1 ) (11.12)
where x = (A,b,c). Now, if G x A~ 2 + 2 n were an affine variety i.e. if

there were a notion of product, then we could ask whether or not '1/Ja was
a morphism and what were the algebraic properties of '1/Ja. For this and
many other reasons, it is necessary to have a notion of "product". We
shall postpone discussion of this concept until section 12, and now tum
our attention to examining the effect of the action of G L( n, k) = G on
the controllable, observable and minimal linear systems.
Proposition 11.13. If x = (A, b, c) is an element of A~ 2 + 2 n and g is

an element of G = G L( n, k ), then
Y(g · x) =gY(x)
(11.14)
Z(g · x) =Z(x)g- 1
where Y(x) = Y(A, b, c), Z(x) = Z(A, b, c) and Y,Z are the controllability
and observability maps respectively.
Proof: Simply note that
Y(g ·x) = [gbgAg- 1 (gb) .. ·(gAg- 1 t- 1 (gb)]

= [gb gAb··· gAn- 1 b] = g[b Ab · · · An- 1 b]
= gY(x)
and similarly, for Z(g · x ).
62 METHODS OF ALGEBRAIC GEOME1RY IN CON1ROL THEORY: PART I
Corollary 11.15. If x = (A, b, c) is controllable, or observable, or

minimal and g is an element of G, then g · x is controllable, or observable,
or minimal.
Corollary 11.16. If x = (A, b, c) is controllable (or observable) and

g is an element of G such that g · x = x, then g = I is the identity.
Proof: Since Y(x) = Y(g · x) = gY(x) and Y(x) is invertible by

controllability, we have g = I.
Example 11.17. Let n = 2 and let
A=[~ ~], b= [~], c=[l 0]
so that (A, b, c) is neither controllable nor observable. If t is any element

ofkandifgt= [! ~] SOthatg"t 1 = [!t n.thenclearlygt•X=X.
Corollary 11.18. If x = (A, b, c) is controllable (or observable ) and

g,g are elements ofG such that g · x = g · x, then g =g.
Proof: If g · x = g · x, then gAg- 1 = gAg- 1 , gb = gb, and cg- 1 =

cg- 1 so that (9- 1 g)A(g- 1 g)- 1 = A, (g- 1 g)b = band c(g- 1 g)- 1 = c.
In other words, (g- 1g) · x = x and so, g- 1 g =I or g =g.
We are now ready to state and prove a second major theorem, namely
the State Space Isomorphism theorem. This theorem essentially states that
minimal realizations of the same transfer function or Hankel matrix or
element of Char( n, k) are "equivalent" under the action of G L( n, k) with
a "unique equivalence". More precisely, we have:
Theorem 11.19. (State Space Isomorphism Theorem)

(1) Let f(z) be an element ofRat(n,k) and let x = (A,b,c) and x 1 =
( A1 , bt, ct) be minimal realizations of f( z ). Then there is a unique
g E GL(n,k) such that g · x = x 1 .
(2) Let H be an element of Hank(n,k) and let x = (A,b,c) and let
x 1 = (A 1 ,b 1 ,c1 ) be minimal realizations of H. Then there is a
unique g E GL(n,k) such that g · x = x 1 .
(3) Let~ be an element of Char(n,k) and let x = (A,b,c) and x 1 =

( A 1 , b1 , ci) be minimal realizations of f Then there is a unique
g E GL(n,k) such that g · x = x 1 .
Proof: We first note that, in all three cases, if g · x = x 1 and g · x = x 1

for g, g E G = G L( n, k ), then g = g by corollary 11.18. In other words,
uniqueness is established.
Next, we note that, in view of proposition 10.14 and proposition 10.12,
'.Rx( x) = '.Rx( xl) if and only if
'.RJ(x) = (1/Jx o '.Rx)(x) = (1/Jx o 9\x)(xi) = '.RJ(xl)·

Similarly, if '.R 1 (x) = '.R 1 (x 1 ), then '.Rh(x) = (¢ 1 o '.R 1 )(x)
( 1/J 1 o 9\ 1 )(xi) = '.Rh ( x1 ). If, on the other hand, '.Rh ( x) = '.Rh( x 1 )
and x, x 1 are elements of Sf, 1 , then
in view of corollary 10.15. In other words, it will be enough to establish

existence for the case (2).
So suppose that H is an element of Hank( n, k) and let x = (A, b, c)
and x 1 = (A 1 ,b1,cl) be minimal realizations of H. Then '.Rh(x) =
'.Rh(xl) so that
(11.20)
for j = 1, ... , and
Z(A,b,c)Y(A,b,c) =H(A,b,c) (11.21)
Let Z = Z(A,b,c), Y = Y(A,b,c), Z1 = Z(AI,bl,cl), and Y1

Y( A1, b1, c1 ). By minimality, Z, Y, Z 1 and Y 1 are all nonsingular. But
(11.20) and (11.21) imply that
(11.22)
It follows from (11.21) that
(11.23)
and so, letting g = Z1IZ, we have gAg-I = AI, gb = bi, and cg-I = ci
in view of (11.22). In other words, g · x =XI.
Example 11.24. Let n = 2 and let
A=[~ n, b=[~], c=[O OJ
AI =[ ~ n, ~]
bi = [ , ci = [0 0]
and x = (A,b,c), XI = (AI,bi,cl). Then 9lx(x) = (0,0,-2,1)

9lx(xi) and !.Rh(x) = (0,0,0,0) = !.Rh(xt). However, there does not
exist a gin GL(2,k) with g · x =xi.
We have shown that elements x, xi of Sf 1 correspond to the same

element f( z) of Rat( n, k) (or H of Hank( n, k) 'or~ of Char( n, k )) if and
only if there is a g in G L( n, k) such that g · x = XI. In particular, we
have observed that the group G = GL(n,k) acts on Af+
2n via
(11.25)
and that G · Sf I = Sf 1 so that Sf I is "invariant" under that action of

G. If we say that X' X~ in sr 1 are equivalent under G (or modulo G)
when there is aginG with g · ~ = x 1 , then the set of equivalence classes,
denoted by SftfG, can be identified with Hank(n,k) (or Rat(n,k) or
Char( n, k )). We shall eventually show that this identification can be made
algebraic. In order to do so, we again require some algebra.
Products, Graphs and Projections
We wish to define the product V X W of any two affine alge-

braic sets. We first observe that if~= (6, ... ,~N) E Af and
'fJ = jTJl>···,TJM) E A~, then (~,TJ) = (6, ... ,~N,T]l,···,TJM) E
Af:+ and Af:+M can be identified with the set of pairs (~,TJ) i.e.
Af:+M = Af x A~ as sets. However, the Zariski topology in Af:+M is
not the same as the product of the Zariski topologies in Af and A~.
Example 12.1. Let N = M = 1. Then the set e-

TJ = 0 is closed
in Ak but is not closed in the product of the topologies in Al as it is
not a finite union of lines parallel to the axes.
We note that we may identify k[Af] as a subring k[X1, ... ,XN]
of k[Af:+M] = k[X1, ... , XN, Y1, ... , YM] and k[A~] with the subring
k[Y1, ... , YM]· If V C Af and W C A~ are algebraic sets and if we
let a be the ideal in k[Xl,···,XN,Yl,···,YM] generated by I(V) U
I(W) = (I(V),I(W)), then V( a) is an affine algebraic set.
Proposition 12.2. V(a) = V x W (as sets).
Proof: We have V = {(: J;(f.) = O,j; E I(V)}, W = {TJ: gj(TJ) =

0, gj E I(W)} so that clearly 11 X W C 11( a). On the other hand, if
((,TJ) E 11(a), then/;(()= 0 and gj(TJ) = 0 so that~ E V and ryE W.
Definition 12.3. 11( a) is called the product of V and W and shall

be denoted by 11 X W.
Since 1(11( a)) = ft, the affine coordinate ring k[\1 X W] is given
by k[X,Y]/ft = k[Xl,···,XN,Yt, ... ,YM]/ft. We now have, in
view of the properties of tensor products (appendix A), the following:
Theorem 12.4. k[\1 x W] = k[\f]®k k[W].
Proof 1. Let us, by abuse of notation, let X = (X 1 , ... , X N ), Y =
65
(Y 1, ... , Y N) be the appropriate residue class (e.g. k[VJ = k[X]

k[X1, ... ,XN], X; being the J(V)-residue of X;, and k[V x WJ
k[X,YJ = k[X1, ... ,XN,Y1, ... ,Y N], xi being the I(V X W) yfa• =
residue of Xi). Consider the k-homomorphism 7/J : k[VJ ®k k[W] -+
k[V X WJ given by
(12.5)
Clearly 7/J maps k[VJ ®k k[W] into k[V X WJ. 7/J is surjective since
7jJ(X;®1) =X;, 7/J(1®Yj) = Yj and these generate the ring k[Vx WJ.
To show that 7/J is injective it will be enough to show that if J;(X),
gj(Y) are independent over k, then 7/J(f; ® 9j) are independent over
k. But, if L.cijfi(v)gj(w) = 0, Cij E k, for all v,w, then, for fixed v,
"L.c;jfi(v)gj(-) = 0 on W. This implies that "L.;c;jf;(·) = 0 on V and
so, c;j = 0 for all i, j.
Proof 2. Clearly k[V] ®k k[W] = (k[X]/I(V)) ®k (k[Y]/I(W)).

From appendix A, we have that k[X] ®k k[Y] = k[X, YJ and that
(k[X]/I(V)) ®k (k[Y]/I(W)) = k[X, Y]f(I(V) · k[Y] + k[XJ · I(W)).
Thus there is a natural k-homomorphism 7/J: k[V]®kk[WJ-+ k[Vx WJ
given as follows: if f E k[V] and g E k[WJ, then f is the restriction
of a polynomial F( X) to V and g is the restriction of a polynomial
G(Y) toW; we let 7/J(f®g) be the restriction of F(X)G(Y) to Vx W.
Since 7/J is obviously surjective, we need only show that 7/J is injective.
If I:~=l J; ® g; f- 0 is an element of the kernel of 7/J with r as small as
possible (!; ® g; f- 0 all i), then there is a v E V with J;( v) E k and
not all 0. It follows that 2:::~= 1 J;(v)g;(-) = 0 on W and hence that
(say) 91 = 2:::;= 2Cj9j· But then I:j= 2(fj + cjfr) ® 9j is an element of
the kernel of 7/J with fewer than r terms which is a contradiction.
Corollary 12.6. If V and W are irreducible, then V x W is irre•

ducible.
Proof 1 (Appendix A): Since k[V] and k[W] are integral domains,
so is k[V] ®k k[W].
Proof2 (direct). Suppose that VxW = Z1UZ2. Then V = V1UV2

where Vi = {v : {v} X W C Z;} for i = 1,2 (since {v} X W =
( { v} x W) n Z 1 U ( {v} x W) n Z2). But each V; is closed. For, say, let
V2 = {v: (v,w) E Zr}::::::: V x {w}n Z1. Then V2 is closed and v E V1
PRODUCTS, GRAPHS AND PROJECTIONS 67
if and only if ( v, w) E Z1 for all w E W i.e. if and only if v E nwEW V2.

Thus, V1, V2 are closed and so, V = V1 (say) and V X W = Z1.
Corollary 12.7. Suppose that V, Ware irreducible and that~ E V

TJ E W. Then
(12.8)
i.e. the local ring of V x W at(~, ry) is the "localization" of the tensor
product of the local ring of V at ~ and the local ring of W at TJ with
+
respect to the maximal ideal m{OTJ,w mTJO{,V (see section 9).
Proof: By definition,
Since k[V] 0k k[W] c O{,V 0k OT],W c a({,TJ),vxw. it will be enough

to show that the maximal ideal n(E,TJ) of functions in O{,v 0k OTJ,W
which vanish at (~, ry) is precisely the ideal m{OTJ,w + mTJO{,V = m.
Clearly m C n({,TJ)· On the other hand, if E(J; ® g;) E O{,V ®k OT),w
and /;(0 =a;, g;(TJ) = b;, then E(J; 0g;)- E(a; ® b;) = E(J;- a;) 0
g; + Ea; 0 (g; - b;) E m. So, if E(J; ® g;)(~, TJ) = 'Ef;(~) ® g;(TJ) =
Ef;(~)g;(TJ) = Ea;b; = Ea; ® b; = 0, then E(J; 0gi) Em.
Definition 12.9. Let 1rv : V x W- V be given by 1rv(~, ry) = ~

and 1rw : V x W - W be given by 1rw(~, ry) = ry. Then 1rv is the
projection of V x W on V and 1rw is the projection of V x W on W.
We observe that the projections are morphisms since, for example,

1rv is the restriction to V of the mapping1r1 : Af Ar - Ar
x given
by
(12.10)
which is clearly a polynomial map. We note that the comorphism 1ri
is the natural injection of k[xl, ... ,xN] - t k[xi.···,xN,Yl,····YM],
and that 1r\( is the natural injection of k[V] - k[V] ®k k[W]. Clearly
1rv is a surjective morphism.
The product is "categorical" in that given an algebraic set X and
morphisms ¢ : X - t V, 1/; : X - t W, there is a unique morphism
¢X 1/;: X-V X W such that¢= 1rv o (¢ x '1/;) and 7/J = 1rw o ( ¢ x '1/;).
Note that, necessarily,(</; X 1/J)(x) = (<f;(x),'l/J(x)). Put another way,

the diagram
v
X
¢/! rt>x'I/J Vx W
11"V
~j 11"W
(12.11)
is a commutative diagram of morphisms. It follows that a base for the

topology in V X W is given by the principal affine open sets (V X W) 1
where f(x, y) = 'f,ji(x)gi(Y) with fi in k[V] and 9i in k[W]. We use
this observation to prove the following proposition which will be useful
in section 17.
Proposition 12.12. The projection 1rw is an open mapping.
Proof: Let U be an open subset of V x W. Then U = U(V x W)J;

with fj E k[V X W]. Since 1rv(U) = U1rv[(V X W)J,], we may suppose
that U = (V X W)J = {(v,w): f(v,w) ::J 0} is a principal affine
open subset of V x W. Let F E k[X, Y] such that F restricted to
V X W is f and let (~o, TJo) be an element of U. It will be enough
to show that there is a Uo open in V with ~o E Uo and Uo C 1rv(U).
Let G(X) = F(X, TJo) and let g be the restriction of G to V. Since
G(~o) = f(~o,TJo) ::J 0, G(X) -::J- 0 and g(~o) ::J 0. Thus, Uo = V9 is
open in V and ~o E Uo. If~ E Uo, then g(~) = F(~, TJo) ::J 0 so that
(~, 'T]o) E (V x W)J = U and~ E 1rv(U).
Proposition 12.13. Let V C Af:, W c A~ be affine varieties

and let <P : V --+ W, 'ljJ : V --+ W be morphisms. Then the set{~ : ~ E
V, </;( 0 = 1/J( 0} is closed.
Proof: Let <jJ be the restriction of a polynomial map Fq, and let 'ljJ
be the restriction of a polynomial map F'I/J. Then Fq, : AJ: --+ A~, F11, :
PRODUCTS, GRAPHS AND PROJECTIONS 69
Af--. Ar and {~: ~ E V, ¢(~) = .,P(~)} = V n {~: (Fq,- F.p)(~) = 0}

is closed.
Corollary 12.14. Let b.: Af--. Af xAf be given by b.({)=(~,~).

If V is an algebraic set, then b.(V) is closed.
Proof: Simply note that b.(V) = (V x V)n {(~,17): X;- XN+i =

0, i = 1, ... , N}.
The map b. is called the diagonal morphism and b.(V) is the di•
agonal of V.
Corollary 12.15. Let <P : V --. W be a morphism and let gr( <P) =
{( ~, 1J) E V x W : 1J = ¢( ~)}. Then gr( <P) is closed and gr( <P) is isomor•
phic to V.
Proof: Consider the morphism 1/J : V X W -+ W X W given

by .,P(~,7J) = (¢({),7]). Then gr(¢) = ¢- 1 (b.(W)) is closed. Let
J> : V -+ gr( <P) be the morphism given by ¢( ~) = (~, ¢( 0 ). Then J>
is bijective. If 1rv : gr ( <P) -+ V is the restriction of the projection
on V to gr(¢), then 1fv is a bijective morphism and (1rv o ¢)(~) = ~,
(¢ o 1rv )(~,¢(~)) = (~, ¢({)) so that 1fv = ;J>- 1 •
Corollary 12.16. Let <P : V -+ W, 1/J : V -+ W be morphisms. If

<P = ¢on a dense subset of V, then <P = ¢on V.
Corollary 12.17. Let <P : V -+ W be a morphism. Then there is an

injective morphism i and a surjective morphism 7'i such that <P = 7'i o i.
Proof: Let 7'i = Tiw : V x W -+ W and let i : V -+ V x W be

given by i(~) =(~,¢(~))(so that i(V) = gr(¢)). Clearly 7'i o i = ¢.
We have defined, in a natural way, the product V x W of any two

affine algebraic sets and noted that the coordinate ring k[V x W] of the
product is simply the tensor product k[V] Q()k k[W] of the coordinate
rings of the factors. We now have:
Proposition 12.18. If V and W are irreducible, and f E k[V],

then V1 x W is a quasi-affine variety with k[V1 x W]s::! k[V] 1 ®k k[W].
Proof: Clearly, V1 x W = (V x W) 1 (viewing k[XI. ... , XN] as

a subring of k[X1, ... ,XN, Y1, ... , YM]) and k[V1 X W] = k[V X WlJ
( cf. section 9). Thus, it remains to show that k[V] 1 ®k k[W] is k•
isomorphic with k[V X W] 1 . Let 7/J : k[V1] ® k[W] --+ k[V x W] 1 be
given by
7/J(h/ f ®g)= itg/ f (12.19)
where it E k[V] and g E k[W]. Then 7/J is a k-homomorphism (by

properties of the tensor product). If hj r is an element of k[V X W]J
r
with h = Ef;gi, J; E k[V], g; E k[W], then 7/J(EJ;j ® g;) = hj r so
that 7/J is surjective. Since k[V X W] is an integral domain (corollary
12.6), 7/J(ft/ f ®g)= 0 implies either it = 0 or g = 0 so that (it/ f ®
g) = 0 and 7/J is injective.
Example 12.20. Let n = 2 and let G = GL(2, k) = (M(2, 2; k))det 9

= (A1,)det 9 . Consider the product G X A~ and note that k[G X A~]=
k[A! X A~]det 9 . If we let G 11 , G12, G21, G22 be the coordinate func•
tions in At, and Xu, X12, X21, Xn, Y1,Y2, Z1,Z2 be the coordi•
nate functions in AL then the map 7/Jc : G X A~ --+ A~ is given by
7/Jc(g,x) = (gAg-I,gb,cg- 1) (cf. (11.12)) or in coordinates by
Gn -G12]
[
G21 Gu
[ GnYl + G12Y2]
G21 Y1 + Gz2 Yz
[Z1G22- Z2G21 - Z1G12 + Z2Gu]/(GuG22- G12G2t)
so that 7/Jc is a morphism. We observe that the comorphism 7/J'G
k[X,Y,Z] --+ k[G,X,Y,Z]detG is given by 7/J(;(p(X,Y,Z)) =
p(GXadjGjdetG,GY,Zad jGjdetG) where pis an element of
k[X, Y, Z]. Clearly 7/Jc is surjective and 7/Ja is injective. As we have
shown many times, 7/Jc is not injective. Of course, all these observa•
tions hold for arbitrary n.
We shall have a number of uses for the notion of a "product" in

the sequel.
13. Group Actions, Equivalence and Invariants
We now turn our attention to examining in a general way such

group actions as (11.25) and the corresponding equivalence relations.
The main problem consists of two parts: first, to find the equivalence
classes under the action of a group; and second, to parameterize these
classes via an algebraic object. (See e.g. [D-2], [F-2], [M-3], [M-4]).
Definition 13.1. Let X be a set and let f be a group. Then f

acts on X if there is a map a : r x X -+ X such that
(i) a(1112,x) =a(/t,0(/2,x))

(ii) a(£,x) =x
where£ is the identity in f. The map a, which we frequently suppress,

is called an action of r on X ..
If G L( X) is the set of endomorphisms of X* and r acts on X,
then the map f..t-r : X -+ X given by f..t-y(x) = 1 · x is in GL(X) and
f..te =I, f..t-r 1 -r2 = f..t-r1 f..t-y2 so that the map 1 -+ f..t-r is a homomorphism
off into GL(X).
Definition 13.2. An element x in X is invariant under the action

of r if 'Y • X = X for all 'Y in r and a subset y c X is invariant if
1 · Y C Y for all 1 in f. If Y C X, then the stabilizer of Y, S(Y), is
the set {1 E f : 1 · Y = Y}.
We note that S(Y) is a subgroup of r since f..t-r E GL(X).

Let X, Y be sets and let M(X, Y) = {f: f maps X into Y}.
Proposition 13.3. If r acts on X andY, then

(13.4)
defines an action of ron M(X, Y).
* An endomorphism is a bijective map which preserves algebraic

structure.
71
Proof: Simply note that ( E · f)( x) = E/( EX) = f( x) and that

[( /1/2) . f]( X) = (/I/2 )/( (/1/2 )- 1X) = /1 b2f(l'z 1l't 1 X)] = /1 [(12 ·
f)(l'1 1x)] = 11 · [(12 · f)](x).
We note that f E M(X, Y) is invariant under the action (13.4)

if and only if f(ix) = i[f(x)] for all{ in r.
Definition 13.5. Let k be a field and let k* = k - {0} be the

multiplicative group of units in k. A homomorphism X : r -+ k*
is called a ( k )-character of r. If f E M (X, k ), then f is a relative
invariant ofweight X if f(ix) = x(l')f(x) for all1 and f is an absolute
invariant if X( 1) = 1.
If X is a character, then f acts on k via 1 · a = x(l' )a for a E k

since (1112)a = X(lll2)a = X(ll)X(I2)a = 11 · b2 ·a]. Thus f is a
relative invariant of weight X if and only iff is an invariant under this
"x -action".
Example 13.6. Let Q(u, v) au 2 + 2buv + cv 2 and let r
GL(2, k). For 1 E r, set
Then Q(u',v') = Q(l't)u 2 +2l'i [~ :] 12uv+Q(12)v 2 = a'u 2 +

2b'uv + c'v 2 where the li are the columns of I· Then b' 2 - a'c' =
( det l'? ( b2 - ac) and so the discriminant bz - ac is a relative invariant
of weight (det1)z. If r = 0(2,k) ={IE GL(2,k): 1'1 = J} is the
orthogonal group, then, since a' = a1i 1 + 2bl'nl'21 + c1i1 and c' =
a1? 2 + 2h12l22 + qi 2, we have a'+ c' = a+ cas 1?1 +liz = 1, l'n/21 +
l'12l'2Z = 0, 1i1 + 1iz = 1. In other words, the "trace" of Q is an
absolute invariant under 0(2, k). Now, let P(u, v) = Au 2 +2Buv+Cvz
be a second quadratic form. Then we could seek simultaneous invariants
of the pair (P, Q). Such an invariant would also be an invariant of the
"pencil" Q + >..P. For example, if 1 E G L(2, k ), then
b'+>..B' a'+>..A'] [b+>..B a+>..A]

det [ c' + >..C' b' + >..B' = (detl')2 det c + >..C b + >..B
GROUP ACTIONS, EQUIVALENCE AND INVARIANTS 73
using the (relative) invariance of the discriminant of Q+>..P. Equating

powers of>.., we find that a'C' +c' A'-2b'B' = (det-y) 2 (aC+cA-2bB)
is a (relative) simultaneous invariant.
Definition 13.7. If r acts on X and x E X, then the set Or ( x) =

{y : there is a 1' E r with 1' · x= y} is the called the orbit of x under
r ( or modulo f).
Proposition 13.8. The orbits Or( x) are the equivalence classes with
respect to the relation: xEry if there is a 1' E r with 1' · x = y.
Proof: Simply note that: (1) xErx as E • x = x; (2) if xEry, then

1' · x = y implies -y- 1 · y = x so that yErx; and (3) if xEry and yErz,
then 1'1 · x = y, 1'2 · y = z imply (1'21'1) · x = z so that xErz.
Lemma 13.9. If r acts on X andY and iff E M(X, Y) is r•

invariant (proposition 13.3 ), then
f(Or(x)) = Or(f(x)) (13.10)
for all x.
Proof: If y = 1' · x E Or(x), then f(y) =

f('Y · x) =
-y[f(x)]
as f is f-invariant so that f(y) E Or(f(x)). On the other hand, if
w = -y[f(x)] E Or(f(x)), then w = lf(x) = f(-yx) = f(y) where
Y=i·xEOr(x).
We shall say that the action of r is trivial on Y (or r acts trivially
on Y) if 1' · y = y for all 1 E f and y E Y.
Corollary 13.11. If r acts trivially on Y, then a f-invariant f is

constant on orbits.
Corollary 13.12. If r acts trivially on Y and iff is constant on

orbits (i.e. f( Or( x)) = {!( x)} ), then f is 1-invariant.
The corollaries provide a basis for defining the notion of an ab•

stract invariant.
Definition 13.13. Let E be an equivalence relation on X (e.g.

E = Er where r acts on X). Then j E M(X, Y) (where r acts
trivially on Y if E = Er) is an abstract invariant if f is constant on

orbits (i.e. if xEx 1, then f(x) = f(x!)). An abstract invariant is
complete if it is injective on orbits (i.e. if f(x) = f(x!), then xExi)
and a surjective abstract invariant is called independent.
We observe that there is always an independent complete abstract

invariant. For, let XE = X/E = {OE(x): x EX} be the quotient
space of X modulo E (if E = Er, we write Xr for XEr ). XE is called
the orbit space of X modulo E. We then have:
Proposition 13.14. The projection 7rE: X--+ XE given by 7rE(x) =

0 E ( x) is an independent complete abstract invariant.
Proof: Clearly 7rE is constant on orbits and is surjective. But

xEx1 if and only if 7rE(x) = OE(x) = OE(x!) = 7rE(xi) and so 7rE is
injective on orbits.
Theorem 13.15. (Universal Property). Let f E M(X, Y) be any

abstract invariant. Then there is a unique map f' : X/ E --+ Y such that
f = f' o 7rE i.e. the diagram
X _ _1r_E_ __.XE = X/E
~\!'
y
is commutative.
Proof: Simply set f' (0 E ( x)) = f (x) and note that f' is well-defined
since f is an invariant.
Now let us return to the system theory context. Recall that the char•
acteristic function realization map fRx : Af+ 2 n --+ A~n given by
fRx(A, b, c) = (cb, ... , cA n-Ib, -XI (A), ... , -xn(A)) (13.16)
is a smjective morphism and that fRx( Sf, 1 ) = ( A~n)-' = Char( n, k) is

isomorphic to Hank( n, k) or Rat( n, k ). In view of proposition 11.8, the
GROUP ACTIONS, EQUIVALENCE AND INVARIANTS 75
map 91x is an invariant for the action of G = GL(n,k) on A~ 2 + 2 n (given

by (11.11)) and the trivial action of G on A~n. Since 91x is surjective, it
is an independent invariant. Moreover, we have:
Proposition 13.17. On Sf, 1 , 91x is an independent complete ab•

stract invariant for equivalence modulo G L( n, k ).
Proof: Completeness is an immediate consequence of the State Space

Isomorphism Theorem 11.19.
We observe that Proposition 13.17 remains true if we replace 91x by
either 91 h or 91 f.
sr
Let Ef 1 = 1 I G and let 7r G : sr
1 -+ Ef 1 be the projection given
by 1rc(x) ~ Oc(x'). Then, by the Uni~ersal P~operty (Theorem 13.15),
there is a unique map <.p : Ef, 1 -+ Char( n, k) such that the following
diagram
Sf, I --71"-G----+ Ef,1
~ i~
Char(n, k)
is commutative. Moreover, since 91x and 1rc are both independent and
complete, the mapping <.pis bijective. Now S}, 1 and Char( n, k) are quasi•
affine varieties and the mapping 91x is a morphism. Since <.p identifies Ef, 1
and Char( n, k ), we can see that, at least, the possibility of giving the orbit
space an algebraic structure (e.g. that of a quasi-affine variety) exists and,
hopefully, in that case the mapping <.p will be an isomorphism. Of course,
similar remarks apply to 91 1 and Rat( n, k) and to 91h and Hank( n, k ).
In order to fully develop these ideas, we tum our attention to the study of
"quotients".
14. The Geometric Quotient Theorem:
Introduction
Now let us try to analyze the situation from a general standpoint.

We have a variety V (quasi-affine), an equivalence relation E on V
which may be generated by the action of a group r on v, and a
variety W (quasi-affine) with a map 'ljJ: V----+ W which is a morphism.
In the system theory• case, V = Sf,ll E = Ea where G = G L( n, k),
W = Char(n, k) with G acting trivially on W, and 'ljJ = fRx.
Definition 14.1. A regular function f E k[V] is an E-invariant

(on V) if v1Ev2 implies j(v1) = f(v2)· The set k[V]E = {!: f is an
E-invariant} is called the ring of E-invariants (on V).
It is clear that k[V]E is a ring. In the case E = Er, we write

k[V]r in place of k[V]Er and speak of the ring of r -invariants (or if
r is understood from the context, simply the ring of invariants). We
note then that f E k[ Sf 1 ]G if and only if
'
f9(x) = f(g · x) = f(x) (14.2)
for all gin G = GL(n, k) and x =(A, b, c) in Sf 1 . Since k[V]E C k[V],

the ring of invariants has no nilpotent eleme~ts; however, it is not
necessarily finitely generated ([N-1], [M-3], [M-4]) and so need not be
an affine k-algebra. If k[V]E is finitely generated, then k[V]E = k[W]
for some affine algebraic set W (proposition 7.7). However, W is not
necessarily in bijective correspondenc e with VE = V j E ([M-4]). If so,
then we can give an algebraic set structure to VE. Thus, one critical
issue is whether or not k[V]E is finitely generated. Even if this is so,
we must still seek conditions which insure that k[V]E" = "k[VE]·
Definition 14.3. An ideal a in k[V] is an E-invariant ideal if

v E V(a) implies that OE(v) C V(a). We set R = k[V] andRE=
76
THE GEOMETRIC QUOTIENT THEOREM: INTRODUCTION 77
k[V]E so that RE is a subring of R. If b is an ideal in RE, then

a = Rb is an E-invariant ideal in R for: if v E V(a) and v1 Ev,
then f E a means that f = L,rif; with r; E R, J; E b C RE so
that f(v!) = 'L,r;(vi)J;(vi) = 'L,r;(vi)f;(v) = 0 (i.e. v1 E V(a)). In
particular, if f E RE, then Rf is an invariant ideal. If, on the other
hand, a is any ideal in R, then a n RE is an ideal in RE. We call
an RE = ac the contraction of a to RE and be = Rb the extension of
b toR.
In view of these observations, we have the following definitions:
Definition 14.4. A subset Y of V is E-invariant if v E Y implies

that oE(v) C Y.
Definition 14.5. The equivalence E is proper if (i) k[V]E is finitely

generated and (ii) if a1, ... , Or (arbitrary r) are invariant ideals, then
(L,ai) n k[V]E = L,( Oi n k[V]E). *
We note first of all that if a is an invariant ideal, then V (a) is

a closed invariant subset of V. Conversely, if Y is a closed invariant
subset of V, then I(Y) is an invariant ideal. If E is proper, then k[V]E
is an affine k-algebra and so k[V]E = k[W] for some algebraic set W.
Since k[V]E C k[V], there is a natural injection 'lj;* : k[V]E --+ k[V]
which defines a natural map 'ljJ : V--+ W given by
'lj;(a) =a o 'lj;*, a E Homk(k[V],k) (14.6)
or, equivalently, in terms of ma.ximal ideals, by
'lj;(m) = m n k[W] = m n k[V]E, mE Spm(k[V]) (14.7)
(See Theorem 7.5 and Corollary 7.6). It follows that 'ljJ is constant
on orbits for: if vEv1 and f E k[V]E, then f(v) = f(v 1) so that
mv n k[V]E = mvl n k[V]E. Condition (ii) of definition 14.5 can
then be interpreted geometrically as follows: if v; are closed invariant
subsets of V, then
* This is equivalent to the seemingly more general eondi tion: if {a;}

is a family of invariant ideals, then (L,a;) n k[V]E = L,( a; n k[V]E).
-¢(nVi) = n-¢(Vi) (14.8)

where- indicates closure in W. We then have:
Lemma 14.9. If E is proper, then the natural morphism '¢ maps

closed invariant sets into closed sets and consequently,
'¢(nVi) = n'¢(Vi) (14.10)
where Vi are closed invariant subsets of V.
Proof: Let V1 be a closed invariant subset of V, let w be an

element of W, and let V{ = '¢- 1 (w). Note that V{ is a closed in•
variant set (since '¢is constant on orbits and is a morphism). Then
'¢(V1 n V{) = '¢(Vl) n { w }. Therefore, if w is an element of '¢(VI),
V 1 n V{ must be non-empty. In other words, w E '¢(V1 ) and so '¢(VI)
is closed.
Corollary 14.11. If E is proper, then the natural morphism '¢ is

surjective. In other words, '¢ is an independent invariant.
Proof: Since'¢* is injective, '¢(V) = W by proposition 7.18. But

V is a closed invariant subset of V and so, '¢(V) is closed. Thus,
'¢(V) = W and '¢is surjective.
Lemma 14.12. If E is proper and at. a2 are £-invariant ideals, then

V( a 1 ) n V( a 2 ) = ¢if and only if there is an fin k[V]E such that f = 0
on V(a1) and f = 1 on V(a2).
Proof: Clearly if there is such an J, then V(at) n V(a 2 ) = ¢. If

V( a 1 ) n V( a 2) = ¢,then, by the Nullstellensatz, a 1 + a 2 = k[V]. Since
E is proper and the ai are invariant, ( a 1 + a2)nk[V]E = (a 1 nk[V]E)+
( a2 n k[V]E) = k[V]E. But then 1 = ft + 12 with fi E ai n k[V]E so
= =
that fi(V(ai)) 0 and (say) 12 1 on V(a2)·
Corollary 14.13. If E is proper and the orbits under E are closed,

then vEv1 if and only if
(14.14)
(where m€ = {!: f(~) = 0, f E k[V]} ). In other words, 7/J is a complete

invariant.
Proof: If vEv1, then 7/J( mv) = 7/J( mv 1 ) since 1/J is an invariant.

On the other hand, since orbits are closed, oE(v) and oE(vi) are
closed invariant sets and I(oE(v)) and I(oE(vt)) are invariant ideals.
If oE(v) n oE(vl) =¢>,then there is an fin k[V]E with f(v) = 0
and f( vi) = 1 i.e. f E 7/J( mv) but f rf. 1/J( mv 1 ). It follows that
'lj;(mv) = 'lj;(mv 1 ) implies that oE(v) = oE(vl)·
Combining corollaries 14.11 and 14.13, we have:

then 7/J is a complete independent abstract invariant and there is a natural
bijection OE which makes the diagram
commutative.

then for each invariant open set U C V, there is an open set U0 c W
such that 'lj;- 1 (U0 ) = U.
Proof: If U is an invariant open set, then V - U is an invariant

closed set and U0 = W- 1/J(V- U) is an open set in W (by lemma 14.9)
with 7/J- 1 (Uo) C U. If V 0 is an element of U, then oE( v0 )n (V- U) = ¢>
(as U is invariant). Since oE(vo) is a closed invariant set, it follows
from lemma 14.9 that 1/J(oE(vo)) n 7/J(V- U) = ¢>so that OE(vo) c
1/J- 1 (U0 ). Thus, 1/J- 1 (Uo) = U.
The full import of these ideas will only become clear as we pro•
gress through the sequel.
Example 14.17. Let GL(2,C) act on M(2,2;C) by g·A = gAg- 1 •

If
Aa = [ ~ ~] , a EC
and
(i.e. a -:f 0)
then 9a·At = gaAtg,;- 1 = Aa so that Aa E oa(At). But lima ...... o Aa =

Ao = >..I and Ao rf. oa(At). In other words, the orbits are not dosed.
Let us now define the notion of "quotient".
Definition 14.18. ([F-2], [M-3]) Let E be an equivalence on V.

A quotient of V under E (or modulo E) is a pair (W, ¢) consisting of
an algebraic set W and a morphism 1,1; : V -+ W such that (i) 1,1; is an
abstract £-invariant (i.e. 1,1; is constant on orbits), and (ii) if X is an
algebraic set and 17 : V -+ X is an £-invariant morphism, then there
is a unique morphism ry' : W -+ X with 17 = ry' o ¢. If E = Er where
r acts on V, then r is assumed to act trivially on both Wand X and
(W, 1,1;) is called a quotient of V modulo r.
Clearly, the quotient is unique to within isomorphism. Moreover,

we have shown that, if E is proper and orbits are closed, then (VE, 7rE)
may be viewed as a quotient of V modulo E. In fact, we have actually
shown even more; namely, that if E is proper and orbits are closed,
then (VE, 7rE) may be viewed as a geometric quotient of V modulo E
in the sense of the following:
Definition 14.19. ([M-3]) Let E be an equivalence on V. A geo•

metric quotient of V modulo E is a pair (W, 1,1;) consisting of an algebraic
set W and a morphism 1,1; such that
(i) for each wE W, ¢- 1 (w) is a closed orbit;
(ii) for each invariant open set U C V, there is an open set U0 C W
such that ¢- 1 (Uo) = U;
(iii) the comorphism 'lj;* : k[W] ---+ k[V] is a surjective k-isomorphism

between k[W] and the ring of invariants k[V]E.
If E = Er where r acts on V, then we speak of a geometric quotient
of v modulo r.
We also note that, in view of the results in Appendix B, if G is a

reductive group which acts on V and if the orbits under G are closed,
then (Vo, Tic) may be viewed as a geometric quotient of V modulo G.
Now let us return again to the system theory context. We then
have the key result:
Theorem 14.20 (Geometric Quotient Theorem). Let G = GL(n,k)

act on Sf, 1 via state equivalence. Then (Char(n,k),fRx), where fRx is
sr
the characteristic function realization map, is a geometric quotient of 1
mo~Q '
We shall not prove this theorem immediately since the proof takes
place in stages and requires the introduction of considerable additional
algebra and algebraic geometry. Also, we shall give several proofs of
the various stages so that final completion of the result is a fair bit
away.
Before we begin, we observe that this theorem combined with
proposition 10.14 and corollary 10.15 shows that the diagram
sr.1
Hank(n, k)
is commutative (where Bx is the natural bijection of theorem 13.15)

and that (Rat( n, k ), 9l f) and (Hank( n, k ), 9lh) are also geometric quo•
tients of 51 1 modulo G. This means, of course, that the various rep•
resentation~ of a linear system of degree n are all algebraically the
same!
Now let us sketch the path that we will follow to obtain the geo•
metric quotient theorem. We first examine the orbits giving a direct
proof of closure and then giving a proof using the concept of dimen•
sion and the so-called "closed orbit lemma" which states that the
orbits of minimal dimension under the action of an algebraic group
are closed. We then examine the second property of a geometric
quotient. The first proof we give of this property depends on the
state space isomorphism theorem, the existence of minimal realiza•
tions, and dimension theoretic properties of morphisms. The sec•
ond proof simply involves the observation that Ec is proper since G
is reductive (Appendix B). Finally, we shall show that k[Sn 1 1 ] 0 =
k[Char(n,k)J"= "k[cb, ... ,cAn-lb,x1 (A), ... ,xn(A)J. The first proof
is simple and depends on, in effect, a "canonical form." The gener•
alization of this proof to the multivariable case is quite subtle since
no "global" canonical form exists. The second proof we give involves
invariants under the symmetric group and a good bit of algebra (Ap•
pendix C). The third proof is a translation of the methods of classical
invariant theory to the system theory context and generalizes rather
smoothly to the multivariable case. A fourth proof simply involves
invoking the Hilbert-Mumford theory (Appendix B). While such a
collection of proofs may seem overwhelming each provides a different
insight into the problem.
15. The Geometric Quotient Theorem: Closed Orbits
Consider the polynomial ring R = k[Af+ 2 n] which we write

k[(X;j),(Yj),(Z;)], i,j = 1, ... ,n or simply, k[X,Y,Z]. We let 7/J;
(X, Y, Z) be the polynomials in R given by
7/J;(X, Y, Z) =ZXi-ty i = 1, ... ,n

(15.1)
7/J;(X, Y, Z) =- Xi-n(X) i = n + 1, ... ,2n
where the x's are the characteristic polynomials and let 7/J : A~ 2 + 2 n ----*
A1n be the regular map ( 7/Jt, .•. , 7/J2n) (In other words, 7/J = !Rx to
simplify notation.). If U~, ... , U2 n are the coordinate functions on
A1n then we have 7/J*(Ui) = 7/J;(X, Y, Z) and
k[Char( n, k)] = k[Ut, ... , U2n]>.(U1 , .•• ,U2 ,.)
where >.(UJ, ... ,U2n) = (0 o 7/Jh)(U~, ... ,U2n) (7/Jh is given by 10.10
and() is the Hankel determinant). If we letS= k[Char(n,k)], then it
follows that
(15.2)
where 6(X, Y, Z) = w(X, Y, Z)I(X, Y, Z) = detH(X, Y, Z) (cf. section
11, especially 11.3). We recall that G = GL(n,k) acts on R via
(15.3)
and that this action preserves degrees and is by automorphisms of

R. If X0 = (Ao, b0 , c0 ) is an element of A~ 2 + 2 n, then oa(xo) is the
orbit of X0 under the action of G. Let Ixo be the ideal (7f;1 (X,Y,Z)-
7/Jt(Ao,bo,Co), ... ,?j;2n(X,Y,Z) -7/J2n(A 0 ,b 0 ,C 0 )). Then V(IxJ is an
affine algebraic set (i.e. is closed in Af+ 2 n) and we have:
Proposition 15.4. oa(xo) C V(IxJ·
Proof: If g ·X 0 = Xt = (At,bt,ct), then At= gA 0 g- 1 , b1 =

gbo, Ct = C0 g- 1 and C 0 A~- 1 bo = CtAt 1 bt, X;(At) = X;(Ao) so that
7/J;(At,bt,ct) = 7/J;(Ao,bo,co)·
83
Corollary 15.5. If Xo E sr,l, then oc(xa) = V(Ixa) n Sf,l·
Proof: If x1 E oo(xo), then x1 E V(lx 0 ) by the proposition and

Xl E sr,l by (say) corollary 11.15. On the other hand, if Xl E V(Ixa)n
sr,l, then '¢(xl) = '¢(xo) and 9'\x(xl) = 9'\x(xa)· It follows from the
state space isomorphism theorem that x 1 E o0 ( X 0 ).
Of course, corollary 15.5 means that orbits in Sf 1 are closed. Off

sr,l' there is trouble as the following examples show:
Example 15.6. Let n = 2, Ao n

= [z, ba = [ and Co = [1 0 ].
Then C 0 A~b 0 = Ofor allj so that '¢(x = (0,0,-2,1) = '¢(x1) where
0 )
A1 = ]z, b1 = [~], c1 = [1 OJ. But x1 is not in oc(xo) as b0 1 0

implies gb 0 1 0 for all gin G.
Example 15.7. Now let n = 3 and
Then c 0 A~b 0 = 1 for all j. If X0 = (A 0 ,b0 ,c 0 ), then we have '¢(xo) =

(1,1,1,-3,3,-1), fxo(z) = 1/(z-1) and rank Hxo = 1. If
A1 ~ [ i i ~l b1 ~ ~ [ l, ~ c1 I 1 o oI
then c1Aib1 = 1 for all j. If x1 = (AI,bl,cl), then we have '¢(x1) =

(1,1,1,-3,3,-1) = '¢(xo) and fx 1 (z) = 1/(z-1) = fxo(z) but x1 rf.
oo(xa) since gA 0 g- 1 = Aa 1 A1 for all gin G.
Now, since the orbits oo(xo) in Sf, 1 are closed, we have:
Proposition 15.8. Let~ be an element of Char( n, k ). Then '¢- 1 ( ~)

= oo( X) for Some X in Sl,l, and hence, requirement (i) for a geometric
quotient is satisfied.
THE GEOMETRIC QUOTIENT THEOREM: CLOSED ORBITS 85
Proof: By theorem 10.18 on the existence of minimal realizations,

is not empty and there is a.n X= (A,b,c) in sr,l with '1/J(x) = ~.
'1/J- 1 (~)
By the state space isomorphism theorem 11.19, if x 1 E '1/1- 1 (~), then
there is a. unique 9x 1 in G such that gx 1 • x1 = x and so x1 E oa( x ).
In other words, '1/J- 1 (~) C oa(x). On the other hand, if Xz E oa(x),
then '1/J(xz) = '1/J(x) =~by proposition 13.17 so that oa(x) C 'ljl- 1 (~).
Now let Xo be a.n element of Sf, 1 and let S(xo) = {g: g · X0 = X0 }

be the stabilizer of x 0 • Then S(xo) = {J} by the state space isomor•
phism theorem. Since the mapping G-----+ G·x 0 = oa(x 0 ) is a. bijective
morphism between G / S( x 0 ) and oa( x 0 ), a. reasonable notion of dimen•
sion for affine (or quasi-affine) algebraic sets would imply that the di•
mension of each orbit is the same as the dimension of GfS(xo) = G
(i.e. dimension is invariant under isomorphism). A second reasonable
requirement for the notion of dimension is that if vl is strictly con•
tained in V2 (with V2 irreducible), then the dimension of V1 should
be strictly less than the dimension of Vz' written dim vl < dim Vz.
We shall, in a. lengthy algebraic digression, develop such a. notion of
dimension and show that, in fact, dimoa(xo) = dimGL(n, k) = n 2 for
all X 0 in Sl,l.
Assuming the requisite properties of dimension, we have the fol•
lowing:
Lemma 15.9 (Closed Orbit Lemma). Orbits of minimal dimension

are closed.
Proof: Let Xo be a.n element of sr,l and v = G. Xo = oa(xo)·

Since oa(xo) = V(IxJnSf, 1 is a. locally closed set, it contains a. dense
open subset of its closure V in A~ 2 + 2 n. But G acts transitively on
V a.s G · x = oa(x) = oa(xo) = V for all x in oa(xo)· Moreover,
G · V C V (i.e. V is stable under the action of G). It follows that V
contains a. neighborhood in V of each of its points. In other words, V
is open in V and so, V = V n U with U open in A~ 2 + 2 n. Therefore,
V - V = V - (V n U) = W n V is closed where W = A~ 2 + 2 n - U
is the closed complement of U. If V - V is not empty, then V - V
is strictly contained in V and dim(V - V) < dimV. Since V - V
is stable under the action of G (i.e. G · (V - V) C V - V), V - V
contains orbits oflower dimension. Thus, if V = oa(xo) were a.n orbit
of minimal dimension, then V - V would be empty and so V = V

would be closed.
We note that lemma 15.9 holds in the more general context of an

algebraic group acting on a variety ([H-8]).
16. Affine Algebraic Geometry: Dimension
So let us now turn our attention to developing the notion of di•

mension. We begin with the definition of topological dimension. Then,
we introduce the transcendence degree and use it to define the dimen•
sion of a prime ideal. This provides an algebraic notion of dimension.
An analysis of integral dependence follows and culminates in the "Go•
ing Down" theorem. This allows us to relate the dimension of a prime
ideal to its depth and height. We can then show the equivalence of the
topological and algebraic definitions of dimension. This in turn leads
us to the result that dimkGL(n,k) = n 2 • We conclude with some
results on varieties of codimension r which will be useful in treating
fibers of morphisms. Of course, the lengthy development will be quite
useful throughout the sequel.*
Definition 16.1. If X is a topological space, then the (topological)

dimension of X, dimX, is the maximum integer n such that there is
a chain Vo < V1 · · · < Vn of distinct closed irreducible subsets of X.
If V C Af: is an affine or quasi-affine variety, then the (topological)
dimension of V, dimk V, is its dimension as a topological space (with
the Zariski topology). If V C Af: is an affine algebraic set, then the
(topological) dimension of V, dimk V, is the maximum of dimk Vi where
V = UVi is the decomposition of V into its irreducible components
(proposition 6.10).
Now let R = k[V] where V is an affine variety. Then R is an
integral domain since p = I(V) is a prime ideal. We let ]( ( R) be the
quotient field of R. The elements of K(R) are of the form f fg where
f, g E k[V] and g f. 0 and are called rational functions on V. In this
situation, we write K(V) in place of K(k[V]) = K(R) and call K(V)
thefunctionfield of V (see section 9).
We now make the following:
Definition 16.2. Let ]( be an extension field of k. Elements

6, ... , ~N of K are algebraically independent over k if f E k[ x 1 , • .• ,
* Our treatment relies on [A-2], [D-1], [H-2], [H-8], [M-1] and [Z-3].
87
XN], f(f.I,···,f.N) = 0 implies f = 0. A maximal set of algebraically

independent elements is called a transcendence basis of K over k. If
K = k(z 1 , ..• , ZN) is finitely generated, then the maximum number (fi•
nite) of algebraically independent elements is called the transcendence
degree of Kover k and is written tr. degKfk. If R = k[z~, ... ,zN]
is a finitely generated integral domain with quotient field K( R) =
( k( z1, ... , z N) ), then the transcendence degree of R over k, tr. deg
Rfk, is tr. deg K(R)fk. If p is a prime ideal in R, then the dimen•
sion of p over k, dimkp, is the transcendence degree of Rjp over k
i.e. dimkP = tr. deg(R/P)/k = tr. degK(Rfp)fk. In particular, if
R = k[Af] = k[X1, ... , XN] and p = I(V) is the ideal of an affine
or quasi-affine variety V, then the algebraic dimension of V over k,
written Dimk V, is dimkP = dimki(V).
We note that
DimkV = dimki(V) = tr. degk[V]/k = tr. degK(V)/k (16.3)
for an affine (or quasi-affine) variety V. In other words, the algebraic

dimension of Vis the transcendence degree of the function field K(V)
of V.
Now we have defined both a topological and an algebraic dimen•
sion for V and our goal will be to show that these are the same. Before
doing so, we develop some simple results which indicate that we are
on the right track.
Proposition 16.4. DimkAf = N.
Proof: k[Af] = k[X~, ... ,XN] has transcendence degree N.

Corollary 16.5. DimkGL(n, k) = n 2 •
Proof: The quasi-affine variety GL(n,k) is open and dense in

M(n,n;k) so that DimkGL(n,k) =
DimkM(n,n;k) =
DimkAk2 =
n2.
Corollary 16.6. DimkRat(n,k) = 2n, DimkHank(n,k) = 2n, and

DimkChar(n,k) = 2n.
Lemma 16.7. Let R' = k[y~, ... , Ym] be a finitely generated integral
AFFINE ALGEBRAIC GEOMETRY: DIMENSION 89
domain and let R be an integral domain over k. Let cp : R --+ R' be a

surjective k-homomorphism. Then tr. deg R' I k ~ tr. deg Rl k.
Proof: Let ri, ... , r~ be algebraically independent elements of R'

and let r 1 , ... , rs be elements of R with cp(ri) = r;. Iff E k[X1, ... ,Xs]
and f(rJ, ... ,rs) = 0, then f(cp(ri), ... ,cp(rs)) = f(rL ... ,r~) = 0
implies f = 0 so that the ri are algebraically independent.
Corollary 16.8. JfV and Ware k-isomorphic varieties then Dimk V

= DimkW.
Proof: Simply note that k[V] and k[W] are k-isomorphic.
Corollary 16.9./ftr.degR'Ik = tr.degRik, then cp is a k•

isomorphism.
Proof: Enough to show cp is injective. Let u E R with u # 0. If

r~, . .. , r~ is a transcendence basis of R' I k and if r 1 , ... , rv are elements
of R with cp( r i) = r~, then r 1, ... , r v is a transcendence basis of Rl k
(being algebraically independent by the proof of the lemma). Then u
is algebraic over k[r~, ... , 1'v] and there is an equation
ao ( r1, ... , r v )u t + · · · + at( r1, ... , rv) == 0

with ai(·) E k[xJ, ... ,xvJ, ao(rl,···,rv) # 0 and t 2::1 minimal (so
that at(rJ, ... ,rv) # 0 and at(xl,· .. ,xv) :f; 0). Applying cp, we get
ao ( r~ , ... , r~) [cp( u)] t + · · · + at ( r~, ... , r~) =0

If cp( u) = 0, then at( ri, ... , r~) = 0 which would imply the contradic•
=
tion at(XJ, ... ,xv) 0. Thus, cp(u) # 0.
Corollary 16.10. If p' < p, then dimkp' > dimklJ· In particular, if

V and W are affine varieties and V < W, then Dimk V < Dimk W.
Proof: The k-homomorphism cp : RIP' --+ RIP = (RIP')I(PIP')

is surjective but is not an isomorphism.
Corollary 16.11. Let R be a finitely generated integral domain over

k and let p =I (0) be a prime ideal in R. Then dimk(O) == tr. deg Rf k = v
and dimklJ < v.
Corollary 16.12. dimkJ:J = 0 if and only if p is maximal. In par•

ticular, if V is a variety, then Dimk V = 0 if and only if V = ~ is a
point.
Proof: If p = m is maximal, then R/m = k[z1 , ... , zm] is a field

and the Zi are algebraic over k (by lemma 5.13). On the other hand,
suppose dimk!J = 0. Then p ~ m where m is maximal. If p < m, then
dimk!J > dimk m = 0, a contradiction.
Corollary 16.13. If dimk!J = v - 1, then p is a minimal prime

ideal (i.e. if p' < p, then p' = (0)). In particular, if V C Af and
Dimk V = v-1, then V = V(f) where f is irreducible in k[X1 , ... ,XN].*
Proposition 16.14. Let V c Af and W c Azr be affine (or

quasi-affine) varieties. Then Dimk(V x W) = Dimk V + Dimk W.
Proof: Let n = Dimk V and m = Dimk W. Then k[V] contains

a subalgebra k[ x 1 , ... , Xn] isomorphic to a polynomial ring and k[W]
contains a subalgebra k[y1, ... , Ym] isomorphic to a polynomial ring.
Moreover,
k[V] = (k[x1, ... , Xn])[h, ·. ·, fs]

and
where the fi are algebraic over k(x1, ... , Xn) and the gj are alge•
braic over k(y1, ... , Ym)· Now k[V X W] = k[V] ®k k[W] (theorem
12.4) contains k[xl,···,xn]®k[yl ,···,Ym] = k[xl,···,Xn,Yl,···, Ym]
and so, K(V X l¥) contains k(xl,···,xn,Yl,···, Ym) which has tran•
scendence degree n + m over k. But K(V X W) is an extension of
k(xl, ... ,xn,Yl,···,Ym) generated by fi ® 1, 1 ® gj which areal•
gebraic over this subfield. It follows that tr. degK(V x W)/k
tr. degk(xl,··· ,xn,Yl,··· ,Ym)/k = n + m and hence that
DimkV x W = DimkV + DimkW.
Suppose Vis a variety (i.e. is irreducible) and that the topological

dimension of V, dimk V, is n. Then there is a chain V0 < V1 < · · · <
* Note ll is a variety here (i.e. is irreducible).

Vn = V of closed irreducible subsets of V and a corresponding chain

I(V) < I(Vn-d < · · · < I(Vo) < k[xt, ... , Xn]· Since Vo must be
a point, I(V0 ) is a maximal ideal. It follows from corollaries 16.10
and 16.12 that Dimk V > Dimk Vn- 1 > · · · > Dimk Vo = 0. In other
words, Dimk V 2: n. On the other hand, if there is a chain of prime
ideals, I(V) < Pd-1 < · · · < Po < k[x1, ... , Xn], then V(Po) < · · · <
V(Pd-t) <Vis a chain of closed irreducible subsets of V and dimkV
is equal to the length of a maximal chain of prime ideals containing
I(V). This leads us to make the following:
Definition 16.15. Let R be an integral domain and let p be a

prime ideal in R. Then h = h( jJ) is the height of p if there is a (strict)
chain (0) = Po < P1 < · · · < Ph = p of prime ideals and there does not
exist such a chajn with more than h+ 1 elements. Also, d = d(p) is the
depth of p if there is a (strict) chain p = Pd < Pd-1 < · · · < Po < R
of prime ideals and there does not exist such a chain with more than
d + 1 elements.
The key result which makes things work is a theorem which states
that, in a finitely generated integral domain R over k, if dimkP = s,
then d(p) =sand h(p) = v- s where v = tr. degRjk. This theorem
rests on a lemma which is a converse of corollary 16.13. In order to
prove the lemma we need the concept of integral dependence and some
of its properties.
Definition 16.16. Let R C S be rings. An element u of S is

integral over R if u satisfies a monic equation
(16.17)
with coefficients r; in R.
Definition 16.18 (cf. A.l6). Let R be a ring. An abelian group M

on which R acts linearly is called an R-module (i.e. Af is an R-"vector"
space). More explicitly, there is a mapping R x M ...._. M, ( r, m) ...._. rm,
such that r(mt + m2) = rm1 + rm2, (r + r 1 )m = rm + r 1m, (rr 1 )m =
r(rtm), and 1m= m. M is afinite (or finitely generated) R-module if
there are m1, ... , mt in M such that m E M implies m = I:~=l r;mi
with r; E R. M is a faithful R-module if r M = 0 implies r = 0.
Proposition 16.19. Let R C S be rings. Then the following are

equivalent: (i) u E S is integral over R; (ii) R[u] is a finite R-module;
(iii) R[u] c S1 a subring of Sand S1 is a finite R-module; and, (iv) there
is afaithful R[u]-module M which is afinite R-module.
Proof: Clearly (i) implies (ii) since un+r = -(r1 un+r-l + ... +
rnur) for r ~ 0 means R[u] is generated over R by 1,u, ... ,un-l. As
for (ii) implying (iii), simply let S1 = R[u]. If (iii) holds, let M = S 1
and note that p(u)M = 0 implies p(u) ·1 = 0 where p(u) E R[u] (i.e.
S 1 is a faithful R[u]-module). Suppose that (iv) holds forM and let
c/Ju be the map of M into itself given by
c/Ju(m) = um (16.20)
If m 1 , ... , mt are generators of M over R, then
(16.21)
so that
E(b·'J·u- r·'J·)m J· -- 0 (16.22)
fori= 1, ... ,t. Multiplying by the adjoint of (bijU- rij), we have

det( bijU- rij )M = 0. Since M is a faithful R[u]-module, det( bijU•
rij) = 0 which is an equation of integral dependence for u over R.
Corollary 16.23. If u 1, . .. , Un are integral over R, then the ring

R[ut, ... ,un] is afinite R-module.
Proof: Induction on n.
Corollary 16.24. Let S 1 = { u E S : u is integral over R}. Then S1

is a subring of S.
Proof: If u 1 ,u2 E S 1 , then R[u~,u 2 ] is a finite R-module and so,

u 1 + Uz, u1 Uz are integral over R.
Definition 16.25. S 1 = {u E S; u is integral over R} is the integral

closure of R in S. If S1 = R, then R is integrally closed in S, and if
S 1 = S, then S is integral over R. If R is an integral domain, then the
integral closure of R is its integral closure in K(R) (the quotient field

of R) and R is integrally closed if R equals its integral closure.
Corollary 16.26 (Transitivity of Integral Dependence). Let R C S C

T be rings with S integral over R and T integral over S. Then T is
integral over R.
Proof: If u E T, then un + 81 un- 1 + ... + Sn = 0 with Si E s.

Then R[s 1 , ... , sn] is a finite R-module and R[st, ... , sn][u] is a finite
R[s 1 , ... , sn]-module together imply that R[st, ... , sn][u] is a finite
R-module.
Corollary 16.27. The integral closure S 1 of R in S is integrally

closed inS.
For some examples, we have:
Proposition 16.28. A unique factorization domain R is integrally

closed.
Proof: (see proposition 5.14).
Corollary 16.29. The polynomial ring k[Xt, ... , XN] is integrally

closed.
Corollary 16.30. The ring of integers l is integrally closed.
Proposition 16.31. If R is integrally closed and M is a multiplica•

tively closed subset of R (definition 9.2), then RM is integrally closed.
Proof: Since K(RM) = K(R), let u = rtfr 2, r; E R, be integral

over RM. Then
(16.32)
with s; E Rand mE M. Multiplying by mn,
mr1 ) n
(--;:;- (mr1 )n-1 + ··· + Snm n- 1
+ 81 --;:;- =0 (16.33)
and since all the coefficients are in R, we have ( mri/ r 2 ) = r' E R so

that rt/r2 = r' fm E RM.
Corollary 16.34. If V is an affine or quasi-affine variety and k[V] is

integrally closed, then O.;,v is integrally closed for all~ in V. Conversely,
if Ocv is integrally closed for all~ in V, then k[V] is integrally closed.
Proof: For the first part, simply note that O.;,v = k[V]m~. As for
the second part, it follows immediately from proposition 9.11.
Corollary 16.35. Sf,t, Char(n,k), Hank(n,k), Rat(n,k) and

G L( n, k) all have coordinate rings which are integrally closed.
Now let us continue on the road to showing that Dimk V = dimk V.
Proposition 16.36. If S is integral over R, b is an ideal in S, and

a = b n R, then Sjb is integral over Rja (where Rja = Rj R n b is
viewed as a sub ring of S /b).
Let R C S and let a be an ideal in R. Then u E S is integral over

a if un +at un-t + ···+an = 0 with a; E a. The integral closure of a
in S is the set { u E S : u is integral over a}.
Lemma 16.37. Let R c S, a be an ideal in R, and S 1 be the integral

closure of R in S. If at = St a, then Vat is the integral closure of a in
S.
Proof: If u is integral over a, then un + a 1 un-t + · · · + an = 0

with a; E a. Thus, u E St and un E St a = at so that u E Vat.
Conversely, if u E y(i1 , then un = ~s;a; with s; E St and a; E a.
Since the s; are integral over R, M = R[s 1 , .• • , s J.L] is a finite R•
module. But un M C aM and letting m; be the generators of M, we
have unm; = ~a;jmj, a;j E a so that ~( D;jun- a;j )mJ = 0. It follows
that det(8;jun- a;J) = 0 (cf. proof of proposition 16.19). In other
words, un and hence, u, is integral over a.
Proposition 16.38. Let R, S be integral domains with R c S

and R integrally closed. If u is integral over the ideal a of R, then
u is algebraic over K(R) and the minimal polynomial f(X) xn+

a1xn- 1 +···+an of u has coefficients a1, ... , an in yla.
Proof: Clearly u is algebraic over K(R). Let L :J K(R) be a field

containing all the roots 6, ... , ~n of f(X). Then each ~i is integral
over a (as each ~i satisfies the same equation as u ). But the coefficients
a; are in K(R)[6, ... ,~n] and so are integral over a. Hence, by the
lemma, the a; are in yla.
Corollary 16.39. Let R, S be integral domains with R C S and

R integrally closed. If u is integral over R, then u is algebraic over
K = K(R) and the minima/polynomial f(X) = Xn+auyn- 1 +···+an
of u over J{ has coefficients in R. If S is integral over Rand L = K(S),
then L is an algebraic extension of K. If L / ]( is a finite algebraic
extension and u E S, then the field polynomial (cf exercise 47 [L-1], [Z-3])
ofu over]( has coefficients in R so that, in particular, uo = N LjK( u) E R
where NL;K(u) is the norm ofu. Furthermore, ifp = y'(U) (an ideal in
S), then p n R = ~- (We shall use this observation in the proof of
Krull's Theorem 16.54).
Proof: The first part is a direct consequence of the proposition.

To show that Lis algebraic over K, it is enough to show that ljv for v
inS is algebraic over J(. But vis integral over R so that vm+a 1 vm-l +
···+am= 0 with a; E R. Hence 0 = (ljv)m(vm+a1vm- 1+· ··+am)=
1+a 1 (1/v)m-l +· · ·+am(1/v)m and 1/v is algebraic over K. Since the
field polynomial of u over]( is a power of the minimal polynomial, ([L-
1], [Z-3]) it has coefficients in R. Let f(X) = Xn+a 1xn- 1+···+an be
the minimal polynomial of u over J( (with a; E R). If uo = NL/K(u),
then uo =a~ for some v. Thus, if p = y'(U) and un + a 1un- 1 + · ·. +
an= 0, then 0 = (a~- 1 un+· · ·+a~- 1 an-1u)+uo and uo E pas u E p.
On the other hand, if v E p n R, then v E y'(U) so that vn = us with
s E S. Taking norms, we have vn[L:K] = NL;I<(vn) = NL;K(us) =
NL;I<(u)NL;I<(s) = uoNL;I<(s) (where [L: K] is the degree of L/K).
But NL;I<(s) is an element of R since Sis integral over R. In other
words, v E ~-
vVe now turn our attention to the so-called "Going-Down" theo•

rem. Our treatment is based on [A-2].
Theorem 16.40. Let R C S be integral domains with R integrally

closed and S integral over R. Let Pn C · · · C p1 be a chain of prime
ideals in R and let q m C · · · C q1, m < n, be a chain of prime ideals in
S with q; n R = J:l;, i = 1, ... , m. Then the chain of q; can be extended to
a chain qn C · · · C qm C · · · C q1 of prime ideals such that q; n R = J:l;,
i = 1, ... ,n.
Proof: By induction, we may assume n = 2, m = 1. Let P be the

multiplicativel y closed subset of S containing elements of the form rs
with r E R- J:l2 and s E S- q1 (i.e. P" = "(R- P2)(S- qi)). Note
that R- J:l2 and S- q1 are contained in P as 1 E R, S. Suppose we
show that Spp 2 is a proper ideal in Sp i.e. that Sp 2 n P = ¢. Then
S pjJ 2 is contained in a prime ideal m of S p (e.g. a maximal ideal). But
q2 = m n Sis a prime ideal and q2 n P = 1> so that qz C q1. Moreover,
q2 n R :J 1J2- Since q2 n (R- P2) = 1> (as q2 n P = ¢), q2 n R = p2.
So, we now show that Sp2 n P = ¢. If u E Sp2, then, by lemma 16.37,
u is integral over J:lz and, by proposition 16.38, the minimal equation
of u over K(R) is f(u) = un + a1un- 1 +···+an= 0 with a; E p2 •
If u were also in P, then u = rs with r E R- J:lz, s E S- q1 so that
s = ur- 1 with r- 1 E K(R) and the minimal equation for s over K(R)
is
(16.41)
where r; b; = a; E .p 2 • Since s is integral over R, the b; are in R by

proposition 16.38. If r ~ .)Jz, then each b; E .P2 and sn E Sp2 C Sp1 C
qi. But qi is prime implies the contradiction s E q1.
Corollary 16.42. If q is a minimal prime ideal in Sand q n R = .p,

then jJ is a minimal prime ideal in R.
Proof: If .PI C jJ, then there is a q1 C q with ql nR =.PI· But

qi = q or (0) implies lJI = jJ or (0).
We shall use the corollary to prove a converse of corollary 16.13

and then, prove the key result relating the dimension of a prime ideal
jJ to its depth. We first need a lemma known as the Noether Normal•
ization lemma.
Lemma 16.43. Let R k[ ui, ... , uN] be a finitely generated in-

AfFINE ALGEBRAIC GEOMETRY: DIMENSION 97
tegral domain with transcendence degree v over k. Then there are al•
gebraically independent elements z1, ... , zv of R such that R is integral
over R' = k[z1, ... , zv].
Proof: If v = N, then take Zi = ui and so we use induction on N.

Suppose N > v and that the result is true for N- 1 generators. Since
N > v, u 1, ... , UN are algebraically dependent over k. Let f -=J 0 be a
polynomial with f( u 1 , ... , UN) = 0 and suppose that f is of degree d.
Set
(16.44)
. (d+l) (d+l)N- 1 .
fort=2, ... ,N. Thenj(u1,v2+u 1 , ... ,vN+u1 )=OLe.
. (d+l)
Ut' v2, ... 'VN are roots of the polynonual J(Xl' y2 + xl ' ... 'YN
(d+l)N-1
+X1 ). Each term of f gives rise to a monomial of the form
aeXf where e = e 1 + e2(d + 1) + · · · + eN(d + 1)N-l. As the ei's range
from 0 to d, the resulting e's are all distinct and one of them will
be the term of highest order. In other words, for some e, f(Xt, Y2 +
(d+l) (d+l)N- 1
' ... 'YN + xl
e .
xl ) = aeXI + terms of degree < e With
lYe -=J 0. It follows that a; 1J(u1,v2 + uid+l), ... ,) = 0 is an equa-
tion of integral dependence for u1 over k[v2, ... , vN] and hence, R is
integral over k[v 2 , ••• , vN]· The lemma follows by induction and the
transitivity of integral dependence.
Theorem 16.45. If p is a minimal prime ideal in the finitely gener•

ated integral domain R = k[x 1 , ... , XN] of transcendence degree v, then
dimkP = v- 1.
Proof: If v = N, then R is a polynomial ring and hence a unique

factorization domain. Therefore, p = (!)with f prime and degree f >
0. Say XN occurs in j, then g(x1, ... ,xN- 1 ) = 0 (xi in Rjp) would
imply that f divides g and XN occurs in g. This is a contradiction.
Therefore, tr. deg(R/P)/k = dimkP ~ v- 1. But dimkp ::; v- 1 by
corollary 16.11. If v < N, we use the Noether Normalization lemma
to giveR integral over R' = k[z1, ... ,zv]· Let p' = p n R'. Since R'
is a polynomial ring, R' is integrally closed. But p' is minimal as p is
(corollary 16.42) and, by what was just proved, dimkp' = v- 1. Since
R/p is integral over R' /P' (proposition 16.36), dimkP = v- 1.
Theorem 16.46. Let R be a finitely generated integral domain of

transcendence degree v over k and let p be a prime ideal in R. If dimkP =
s, then
h(P)=v-s, d(P)=s, h(p)+d(P)=v (16.47)
so that dimkP = d(p).

Proof: For d( p), we use induction from s - 1 to s. We may
assumes :j; 0 (otherwise pis maximal and corollary 16.12 applies). If
p = Pd < · · · < Po < R, then 0 = dimkPo < · · · < dimkp = s and so,
d(p) :::; s. Consider Rjp and let p' be a minimal prime ideal in Rjp
where p' > p is a prime ideal with p' /P = p'. Then dimklJ1 = s- 1 =
dimkp'. But then d(p') = s- 1 and d(p) 2: d(p') + 1 = s.
For h( p), we use induction from s + 1 to s and we may assume
p :j; (0). If (0) < P1 < · · · < Ph = p, then v = dimk(O) > dimkPl >
· · · > dimkP = s and so, h(p) :::; v - s. Since p :j; (0), there is
a prime ideal p' < p such that there are no prime ideals between
p' and p. In other words, P/P' is a minimal prime ideal in the fi•
nite integral domain Rjp'. Therefore, dimkP/P' = dimkp'- 1. But
dimkP/P' = tr. deg(R/p')/(P/P') = tr. degR/p (as (R/p')/(P/P') =
Rjp) = dimkP· Hence, dimkp' = s + 1 and h(p') = v- (s + 1). But
h(p') = h(p)- 1 and so, h(p) = v- s.
Corollary 16.48. Let V be an affine variety. Then
(16.49)
i.e. the topological and algebraic dimensions of V over k are the same.
Proof: Closed irreducible subsets of V correspond to prime ideals

in k[X1 , ... , XN] which contain I(V) and hence to prime ideals in
k[V]. Thus, dimkV = d(I(V)) = dimkJ(V) = tr. degk[V] = DimkV.
Corollary 16.50. lfV is a quasi-affine variety, then dimk V = dimk V

where V is the (Zariski) closure of V.
Proof: If Vo < Vi < · · · < VJL is a chain of irreducibles in V, then

V0 < V 1 < · · · < V JL is a chain of irreducibles in V by proposition 6.3
and so dimk V :S dimk V. Let v = dimk V and let Vo < · · · < V11 be
a maximal chain in V so that V0 = { v0 } is a point. Then, again by
proposition 6.3, {v0 } < V 1 < · · · < Vv is a maximal chain in V. Since
m 0 = I( {v0 }) n k[V] is a maximal ideal in k[V] and the V; correspond
to prime ideals in k[V], we have h(m 0 ) = v. Since m 0 is maximal,
d( mo) = 0. It follows from the theorem that h( mo) + d( mo) = v =
dimkV.
Corollary 16.51. dimkGL(n,k) 2n,

dimkHank(n,k) = 2n.
Corollary 16.52. An affine variety V C Af has dimension N - 1

(or, in other words, codimension 1) if and only if V = V(J) where f is a
prime in k[X1 , •.. , XN ].
Proof: The "if" has been established in corollary 16.13. On the

other hand, if f is a prime, then (f) is a minimal prime ideal and so,
dimk V(J) = dimk(J) = N- 1.
From now on, we shall speak of the dimension of V and use the
notation, dim V (or dimk V). We also note that dim V is invariant
under isomorphism (corollary 16.8) and that if vl < v2
with ir• v2
reducible, then dimV1 < dimV2 (corollary 16.10). Since oc(x 0 ) and
G = G L( n, k) are k- isomorphic for any xo in 51 1, we have, in view of
the "Closed Orbit Lemma" (lemma 15.9), the f~llowing:
Proposition 16.53. The orbits o0 (x 0 ), x 0 E 5f, 1 , are closed.
Thus, again, we have established the first property of a geometr:c

quotient for (Char(n,k),9lx)·
Now let us reflect for a moment on corollary 16.52 and corollar•
ies 5.15 and 5.16. The first states that a variety in Af defined by
a single equation has dimension N - 1 and the latter states that a
variety defined by N equations has dimension 0. Suppose we have a
variety defined by r equations or that Af is replaced by an irreducible
algebraic set of dimension N. What can we say under these circum•
stances? We begin with the following geometric version of Krull's
Principal Ideal Theorem ([M-2)).
Theorem 16.54. Let V be an affine variety, 0 -::} f E k[V] with f a

nonunit in k[V], and let Z be an irreducible component of V(f) = {v E

V: f(v) = 0}. Then dimZ = dimV- 1 so that the codimension of Z in
Vis 1.
Proof: Let J:l = I(Z) in k[V] and let Zi, i = 1, ... ,s be the
other irreducible components of V(f) with J:li = I(Zi)· Then .j(fj =
p n p1 n · · · n J:ls (by the Nullstellensatz). Since the decomposition into
irreducible components contains no superfluous elements, there is a
g E nf= 1 J:li, g (/. j:l. Let us replace V by V9 and k[V] by k[V9 ] = k[V] 9 •
Then Vg(f) = {v E V9 : f(v) = 0} = Z n V9 is irreducible and
I( Z n V9 ) in k[V] 9 is simply pk[V] 9 • In other words, we may assume
that Z = V(f) and that jJ = I(Z) = .j(fj. Let S = k[V]. Then
by the Noether Normalization Lemma (16.43), there are algebraically
independent elements z1, ... , Zv of S, v = dim V = tr. deg S/ k, such
that Sis integral over R = k[z 1 , ... , zv]· Since R is integrally closed
(corollary 16.29) and K(S) = k(V) is a finite algebraic extension of
K(R) (as R is finitely generated over k), we deduce immediately from
corollary 16.39 that jJ n R = vf(K;) where fo = NI<(S)/I<(R)(f). Since
R is a unique factorization domain, fo = af~t where a is a unit and
f!J is a prime in R. In other words, jJ n R = (!~). It follows from
corollary 16.52 that tr. degk Rj R n J:l = v - 1. Since S/J:l is integral
over Rj R n jJ (proposition 16.36), we have dimZ = tr. degk Sjp =
tr. degk R/R n p = v- 1 = dimV -1.
The theorem has some significant corollaries.
Corollary 16.55. Let Y be an affine variety and let Z be a closed

irreducible subvariety of V with codim Z = 1. Then if U is open in V
with Z n U f.¢ and iff E k[U], f f. 0 with f(Z) = 0, then Z n U is an
irreducible component of Y(f).
Proof: IfWis the irreducible component ofY(f) with ZnU c W,

then dimU > dimW ~ dim(Z n U) = dimU- 1 = dimV- 1 and so,
W = Z n U by corollary 16.10.
Corollary 16.56. Let Y be an affine variety and let W be a closed

irreducible subset of codimension r ~ 1. Then there exists a chain of
closed irreducible subsets zi. i = 1, ... 'r of v such that zl > z2 >
... > Zr = W and codim Z; = i.
Proof: Use induction on r. For r = 1, Zr = W. Since W < V,

there is an J =f 0 (J E k(V]) in I(W) such that W C Z1 where Z1 is
an irreducible component of V(J). By the theorem, codim Z1 = 1 and
by induction, there are Z;, i = 2, ... , r such that Z2 > · · · > Zr = W
and codim Z; = i.
Note that this corollary also shows that the topological and alge•
braic notions of dimension are the same.
Corollary 16.57. Let h, ... , fr E k[VJ and let Z be an irreducible

component of V (h, ... Jr ). Then co dim Z :::; r.
Proof: Use induction on r. For r = 1, the theorem applies. Since

Z in an irreducible subset of V(fi, ... , fr-l), Z c Z1 where Z1 is a
component of V(fi, ... , J,.._ 1). Since Z is a maximal irreducible subset
of F(h, ... , fr ), Z is a component of Z 1 n V(J,.. ). By induction codim
Z 1 :::; r - 1. If fr( ZI) = 0, then Z = Z1 and co dim Z :::; r - 1 :::; r.
If fr =f 0 on Z 1 , then dimZ = dimZ1 - 1 by the theorem and codim
Z = dimV- dimZ = dimV- dimZ1 + 1 = codim Z 1 + 1:::; r.
Corollary 16.58. Let V be an affine variety and let Z be a closed

irreducible subset of codimension r 2:: 1. Then there are h, ... , fr in k[V]
such that Z is a component of V(h, ... , fr ).
Proof: We shall actually prove a more general result. Let Z 1 >

· · · > Zr = Z be a chain with codim Z; = i (corollary 16.56). Thtn
there are fi, . .. , fr in k[V] such that Z; is a component ofF (!I, ... , j;)
and all components of V(h, ... , fi) have codimension i.
We use induction on i. For i = 1, let h E J(Z1) with fi =f
0 and apply corollary 16.54. Suppose then, that h, .. . , fi-l have
been determined and that Z;_ 1 = W 1, W 2 , ••• , Ws are the components
of V(fi, ... , J;_I). Since each Wj has codimension i - 1, none is
contained in Z;. Therefore, I(Wj) /J I( Z;) for j = 1, ... s. Since the
I(Wj) are prime, Uj= 1J(Wj) /J I(Z;) (exercise 48). Let J; E I(Z;)
such that j; tf. Uj= 1I(Wj)· If W is any component of V(/I, ... ,j;),
then W is a component of Wj n V (Ji) for some j (as in the proof
of corollary 16.57). Since J; =f 0 on Wj, dimW = dimWj - 1 and
codim W = i. But, Zi C V(JI, ... ,ji) as fi E I( Zi). Since Zi is

irreducible, Zi is contained in a component W of V(JI, ... , fi). But
codim Z; = i = co dim W and so, Z; = W.
We shall use these corollaries in studying fibers of morphisms.

Open On Invariant Sets
We now turn our attention to the second property of a geometric

quotient. In other words, we want to prove the following:
Theorem 17 .1. If U is an invariant open set in Sf 1 , then there is an

open set Uo in Char(n,k) such that 'lj;- 1 (Uo) = U (where 'If;= fflx as in
section 15.)
As noted earlier, we shall give several proofs of this theorem.
Definition 17 .2. A morphism t.p : V -+ W of varieties is dominant

if t.p(V) is dense in W i.e. t.p(V) = W.
We observe that, in view of proposition 7.18, t.p is dominant if

and only if t.p* is injective. Since the mapping 'If; : Sf, 1 -+ Char( n, k)
is surjective, 'If; is a dominant morphism. Furthermore, we have:
Proposition 17 .3. If~ E Char( n, k ), then the "fiber" 'lj;- 1 ( ~) =

{x E Sf, 1 : 'lf;(x) = 0 is a closed irreducible variety and dimk'lj;- 1 (~) =
n 2 . Hence, all components of'lf;- 1 (~) have dimension equal to dimkSf 1 -
dimkChar(n,k). '
Proof: Simply note that 'lj;- 1 (~) = oa(x) for some X in sr1 and
that oa(x) and G = GL(n,k) are k-isomorphic by the State 'space
Isomorphism theorem (11.19).
Let us, for the moment, call an affine (or quasi-affine) variety V
normal if its coordinate ring k[V] is integrally closed. Then Char( n, k)
is normal by virtue of corollary 16.35. Now, summarizing, we observe
that 'If; : Sf,1 -+ Char( n, k) is a dominant morphism into a normal variety
such that each fiber 'lj;- 1 (~) has all irreducible components of dimension
n 2 = dimkSf, 1 - dimkChar(n,k). It follows from proposition 18.22 of
the next section that 'If; is an open map. We now have:
103
Proof 1. (of Theorem 17.1). Suppose then that 7/J is an open

map and let U be an open invariant set in Sf 1 . Then U0 = 7/J(U) is
open in Char(n,k) and 7/J- 1 (U0 ) is an open in~ariant set in Sf 1 with
U C 7/J- 1 (Uo). We claim that U = 7/J- 1 (Uo). Let x 0 E 7/J- 1 (U0 ) ~o that
7/J(xo) = ~o E Uo = 7/J(U). Then there is an x in U with 7/J(x) = ~o
and hence, Xo E oa(x) (by Theorem 11.19). But U is invariant means
that oa(x) C U and hence that xo E U.
Of course, the difficult part of this proof is embodied in proposi•
tion 18.22 which shall be proved in the next section. Now, in view of
the results in Appendix B (and [F-2], [M-3], [M-4]), we note that the
equivalence relation Ea, G = GL(n,k), generated by the action
g ·(X, Y, Z) = (gXg- 1 ,gY, Zg- 1 ) (17.4)
on Ro = k[(X;j),(Yi) ,(Z;)Jo = k[X,Y,Z]o where O(X,Y,Z) =

det H(X, Y, Z), is proper. We then have:
Proof 2. (of Theorem 17.1). Simply apply corollary 14.16.
Of course, again the difficult part of this proof is embodied in the

results in Appendix B and in [F-2], [M-3] and [M-4].
Fibers of Morphisms
We shall in this section use the term, "variety", to mean either

an affine or quasi-affine variety. Our goal is to study the fibers of
morphisms and, in particular, to examine the dimensions of fibers.
We begin with
Definition 18.1. Let 1.j; : X -+ Y be a. morphism of varieties. If

y E Y, the closed set 'l.j;- 1 (y) = {x E X : ¢( x) = y} is called the fiber
of 1.j; over y. If W is a. closed irreducible subset of Y, then the closed
set 1.j;- 1 (W) = {x EX: 1.j;(x) E W} is called thefiberof'l.j; over W.
Example 18.2. Let L: A~ -+A~ be given by L(x, y) = (x, -xy).

Then L- 1 ((0,0)) = {(O,y),y E k} and so dimL-- 1 ((0,0)) = 1 >
dim A~- dimA~.
Example 18.3. Let 1.j;: Al _,A~ be given by 1.j;(t) = (t 2 -1, t(t 2 -

1) ). The image of 1.j; is the curve Y 2 = X 2 (X+ 1) and (0, 0) is a point
on this curve. But 'l.j;- 1 ((0,0)) = {1} U { -1} and so a. fiber may well
be reducible.
Definition 18.4. A morphism 1.j; : X -+ Y of varieties is dominant

if 1.j;(X) = Y i.e. 1.j;(X) is dense in Y. If Z is a component of 'l.j;- 1 (1¥),
then Z dominates W if 'l.j;(Z) is dense in W.*
We observe that if 1.j; : X -+ Y is any morphism of varieties,

then Z = 1.j;(X) is irreducible and the morphism -;jj : X -+ Z given
by ";jj( x) = 1);( x) is dominant. If i : Z -+ Y is the natural injection,
then 1.j; = i o -;jj and so, the study of fibers of morphisms is essentially
reduced to the case of dominant morphisms.
* When X, Y are not irreducible, we use the term dominant to mean

both 1.j;(X) = Y and 1.j; maps each component of X onto a dense subset
of a component of Y.
105
Theorem 18.5. Let 'ljJ :X --+ Y be a dominant morphism of varieties

and let W be a closed irreducible subset of Y. If Z is a component of
'lj;- 1 (W) which dominates W, then dimZ ~ dimW + dimX- dimY.
Proof: Since 'ljJ is dominant, 'lj;* is an injection of k[Y] into k[X]

so that dimX ~ dimY (proposition 7.18). If r = codimension of Win
Y, then W is a component of V(g1, ... ,gr) for 9i E k[Y] by corollary
16.58. Let fi = '1/J*(g;). Then Z C 'lj;- 1 (W) C V(fi, ... Jr ). Since
Z is irreducible, Z C Z' where Z' is a component of V(JI, ... , fr ).
But W = '1/J(Z) C 1/J(Z') C V(g1,···,9r)· Since W is a component of
Y(g1, ... ,gr ), W = 1/J(Z') and Z' C 'lj;- 1(W). But Z is a component
of '1/J- 1 (W) and so Z = Z' is a component of V (!I, ... Jr). It follows
from corollary 16.57 that codim Z :S r = codim W. In other words,
dimX- dimZ :S dimY- dimW.
Corollary 18.6. dim'lj;- 1 (y) ~ dimX- dimY,for y E Y.
Definition 18.7. A morphism 1/J: X--+ Y of varieties is finite if

k[X] is integral over 1/J*(k[Y]).
If 1/J is a dominant finite morphism then k(X) is a finite alge•

braic extension of 1/J*(k(Y)) and so, dimX - dimY = 0. The term
finite comes from the fact that the fibers 'lj;- 1 (y) of a finite dominant
morphism are finite sets.
Example 18.8. ( cf. example 7.23) Let X = Al and let Y C A%

be given by Y = {(6,6): ~{- ~~ = 0}. The morphism 1/J: X--+ Y
given by '1/J(~) = (e' e) is finite and dominant. This is so SlllCC
k[Y] ~ k[t 2 , t 3 ] and k[X] = k[t] is integral over k[Y].
Before developing the properties of fibers further, we shall need

some results on integral extensions. In particular, we prove the so•
called "Going Up" theorem. We begin with some lemmas.
Lemma 18.9. Let R be a subring of Sand suppose that S is integral

over R. If S is a field, then R is a field.
FIBERS OF MORPHISMS 107
Proof: Let r =f 0 be an element of R. Then 1/r E SasS is a field

and 1/r is integral over R. Thus, (1/r)n + r 1(1/rt- 1 + · · · + Tn = 0
with Ti E R. It follows that 1/r = -r1 - · · ·- rnrn- 1 E R.
Lemma 18.10. Let R be a local ring with m as unique maximal

ideal. Suppose that R is a subring of S and that S in integral over R. If
m' is a maximal ideal of S, then m' n R = m.
Proof: We note that m'nR C m and that R/m'nR may naturally

be viewed as a subring of Sjm'. Moreover, Sjm' is integral over
Rjm' n R. Since Sjm' is a field, Rjm' n R is a field by the previous
lemma. Hence, m' n R is a maximal ideal and so, m' n R = m.
Theorem 18.11 (Going Up). Let R be a subring of S and suppose

that S is integral over R. If jJ is a prime ideal of R, then there exist prime
ideals p' of S such that p' n R = jJ.
Proof ([M-2]). We reduce to the case of lemma 18.10 by local-

izing both RandS with respect to the multiplicative set M = R- jJ.
In other words, Rp = RM is a local ring which is a subring of SM and
SM is integral over RM. Moreover, if iR, is denote the natural maps
of R into RM and S into SM, then p = i]/(PRp) and if pis a prime
ideal of SM such that iJ nRM = jJRp, then p' = i,S 1(p) is a prime ideal
inS with p' n R = iR 1 (p n RM) = iR 1(pRp) = p.
We shall now prove a critical theorem on finite morphisms.
Theorem 18.12. Let 7./J : X --. Y be a dominant fznite morphism of

varieties. Then (a) 7./J is closed (and therefore surjective); (b) if W is a
closed irreducible subset of Y and Z is a component of ?.j;- 1(W), then
7./J(Z) is closed, dim?.j;(Z) = dimZ and, for some Z, 7./J(Z) = W; and,
(c) if k[Y] is integrally closed (i.e. Y is normal), then if;( Z) = W for all
components Z of 7./J- 1(W).
Proof ( [D-1], [H- 7]).

(a) Let R = k[Y] and S = k[X]. Since 7./J* is injective, we shall view R
as a subring of S with S integral over R. If a is an ideal inS, then
S /a is naturally an integral extension of R/ a n R. If Z = V (a) is
closed in X where a= I(Z) = y'O, then 7./J(Z) c V(a n R) = Z'
and I(Z') = an R (since a = y'O). But k[Z] = k[X]/I(Z)
and k[Z'] = k[Y]/I(Z'). It follows that k[Z] = Sja and that

k[ Z'] = R/ a n R. Hence, 1jj : Z ---+ Z' ("ijj is the restriction of
'1/J) is again finite and dominant. Thus, it is enough to show the
following:
(18.13) any finite dominant morphism is surjective.
But if y E Y and my is the maximal ideal of y in R, then, by
the Going Up theorem, there is a maximal ideal mx in S such that
mx n R =my. Thus, '1/J(x) = y and 'ljJ is surjective.
(b) Let Z be a component of 'lj;- 1 (W). Then, by part (a), '!j;(Z) is
closed. Morever, if 1jj is the restriction of 'ljJ to Z, then 'ljJ : Z ---+
'1/J(Z) is finite and dominant so that dimZ = dim'!j;(Z). Now let
p = I(W) in k(Y] =Rand let q = I('!j;- 1 (W)) in k[X] = S. Then
jJ is prime and qnR = p. Let q = q1 n· · ·nqr where q; = I(Z;) and
the Z; are the components of 'lj;- 1 (W). Then p = ni= 1 (q; n R)
and, since jJ is prime, jJ = q; n R for some i. We assert that
'lj;(Z;) = W. For if g E I('!j;(Z;)), then '1/J*(g) E I(Z;) = q; and
so, '1/J*(g) E q; n R = p. Since '1/J* is injective (actually here
'lj;*(g) =g), g E I(W) i.e. I('!j;(Z;)) c I(W). It follows that
'lj;(Z;) = W.
(c) Let p, q, q; be as in part (b). Then p = q nRc q; n R for all i.
By the Going Down Theorem, there is, for each i, a prime ideal
q\ C q; such that q\ n R = p. But 'lj;(V( qD) C Wand so, V( qD is
aclosedirreduciblesubsetof'!j;- 1 (W). SinceZ; = V(q;) C V(q:),
we must have Z; = V(q:) and q; = q;. In other words, q;nR = p
for each i. By the proof of part (b), it follows that '1/J(Z;) = W
for all i.
Corollary 18.14. A morphism 'ljJ is finite and dominant if and only if

'ljJ is finite and surjective
Corollary 18.15. If 'ljJ : X ---+ Y is a finite surjective morphism, if

Y is normal i.e. k[Y] is integrally closed, and W is a closed irreducible
subset of Y, then dimZ = dimW = dimW + dimX- dimY for every
component Z of 'lj;- 1 (W).
Now, theorem 18.5 says that the dimension of the fi hers is not
"too small". Example 18.2 indicates that the dimension of the fibers
of a morphism may not be "just right" (i.e. equal to dimX- dimY)
FIBERS OF MORPIDSMS 109
everywhere. We shall soon show that things work out "almost every•
where".
Lemma 18.16. Let R C S be finitely generated integral domains

over k. Then there are x 1 , ... , x v in S and f in R such that S 1 is integral
over RJ[Xt, ... ,xv]. v = tr. degK(S)/K(R).
Proof: (Exercise 49)
Theorem 18.17. Let ?j; : X -+ Y be a dominant morphism of va•

rieties. Then there is an open set U C Y such that (a) U C ?j;(X)
and (b) if W is a closed irreducible subset of Y with W n U -:f </> and
if Z is a component of ?j;- 1 (W) with Z n ?j;- 1 (U) -:f </>, then dimZ =
dimW + dimX- dimY.
Proof ([H-7]). Let R = k[Y] and S = k[X]. View Rasa subring

of S. By lemma 18.16, there are Xt, ... , Xv inS and fin R such that
Sf is integral over RJ[x 1 , ... ,xv] where v = tr. degk(X)Jk(Y) =
dimX- dimY. But Sf = k[XJ] and R1 = k[YJ]· Thus, k[YJ x
Ak] = RJ[xt, ... ,xv]· Let 1jj denote the restriction of ?j; to X!. Then
?j; = 7rt o 1.p where 1.p : X f -+ Yf X Ak and 7rt : YJ X Ak -+ Yf is the
projection. But 1.p (given by the injection of R 1 [x~, ... ,xv] into S1 )
is finite and dominant and so is surjective. Since 1r1 is also surjective,
1jj is surjective. Thus, if U = Y1, then ?j;- 1 (U) = X 1 and U C ?j;(X).
This establishes part (a). As for part (b), we may suppose already that
X = XJ, U = Y = YJ and ?j; = 7rt o ~.p. If W is a closed irreducible
subset of Y and Z is a component of ?j;- 1 (W), then Z is a component
of 1.p- 1 (W X Ak)· Since 1.p is finite and dominant, ~.p(Z) = W X Ak and
dimZ = dim~.p( Z) = dim W + v = dim W + dimX - dimY.
The process of factoring a morphism through a finite morphism

(which is essentially a geometric form of normalization lemma) is often
useful. We shall use the same idea in proposition 18.22. Now we make
the following:
Definition 18.18. A subset of a topological space is locally closed

if it is the intersection of an open set and a closed set. A subset is
constructible if it is a finite union of locally closed subsets.
The following theorem of Chevalley is often quite useful.
Theorem 18.19 (Chevalley). Let 1/J: X - t Y be a morphism. Then

1/J maps constructible sets into constructible sets and, in particular, 1/J(X)
is constructible.
Proof: Since a locally closed subset of an affine or quasi-affine

algebraic set is again an affine or quasi-affine algebraic set, it is enough
to prove that 1/J(X) is constructible. In addition, we may assume
that 1/J : X - t Y is a morphism of varieties. We use induction on
dimY as the case dimY = 0 is trivial. If 1/J is not dominant, then
1/J(X) C 1/J(X) < Y and dim'ljJ(X) < dimY so the induction hypothesis
applies. Thus, we suppose that 1/J is dominant. If U C Y is the non
empty open set of theorem 18.17 and W 1 , ••. , Wr are the components
of Y - U and Zij are the components of 'ljJ- 1 (Wi), then 1/J(Zij) is
constructible since dimWi < dimY. Hence, 1/J(X) = UuUi,j'ljJ(Zii) is
also constructible.
Corollary 18.20. Let '1/J : X _.,. Y be a dominant morphism of va•

rieties. Suppose that, for every closed irreducible W C Y and each
component Z of 'ljJ- 1 (W), dimZ = dimW+dimX- dimY. Then 1/J is an
open map.
Proof: If W = {y} is a single point, then 1/J- 1 (W) is not empty

by the hypothesis and so 'ljJ is surjective. Let W be a closed irreducible
subset of Y and let Z 1 , ••. , Zr be the components of 1/J- 1 (W). Then
Zi is also a component of 1/J- 1 (1/J(Zi)). It follows that dim'ljJ(Zi) +
dimX- dimY = dimZi = dimW + dimX- dimY and hence, that
dim'ljJ(Zi) = dimW. In other words, 1/J(Zi) =Wand Zi dominates W.
If x E X and Ux is an open neighborhood of x, then we must show
that 1/J(x) = y is in the interior of 1/J(Ux) = Vy· If not, then y E Y- Vy
which is constructible since Vy is constructible by the thoerem. Thus,
y E W where W is the closure of some locally closed set 0 n W C
Y- Vy (0 is open in Y and W may be assumed irreducible so that
0 n W = W). By the hypothesis, all the components Zi of 'ljJ- 1 (W)
have the same dimension and dominate W. Since 1/J- 1 (0) meets each
zi, 1/J- 1 (0) n 1/J- 1 (W) is dense in 1/J- 1 (W). But 1/J- 1 (0) n 1/J- 1 (W) =
1/J- 1 (0 n W) C X- Ux which is closed. Then, x E 'ljJ- 1 (W) C X- Ux
which is a contradiction.
FIBERS OF MORPHISMS Ill
Corollary 18.21. If 'lj; : X -+ Y is a finite surjective moprhism of

varieties andY is normal (i.e. k[Y] is integrally closed), then 'lj; is an open
map.
Proof: Apply corollary 18.15 and corollary 18.20.
We now prove the proposition which is the key to the first proof
of Theorem 17.1 (the second property of a geometric quotient).
Proposition 18.22 (Chevalley [D-1]). Let 'lj;: X-+ Y be a dominant

morphism of varieties and let r = dimX - dimY. Suppose that Y is
normal and that all the components of 'lj;- 1(y), y E Y, are r-dimensional.
Then 'lj; is an open map.
Proof: We shall reduce the proposition to corollary 18.21 and

proposition 12.12 by appropriately factoring 'lj;. Actually, we will show
that if y E Y and 'lj;( x) = y and Ux is an open neighborhood of x,
then 'lj;(Ux) is an open neighborhood of y. In fact, to do this it will
be enough to show that 'lj;(X) is an open neighborhood of 'lj;( x) = y
in Y (by restricting 'lj; and noting that the intersection of an open set
0 c X with any component of 'lj;- 1(y) is dense in that component).
So let R = k[Y] and S = k[X] and let us view R as a subring of S.
Since Y is normal, R is integrally closed. Set my = I( {y} ). Then
'lj;- 1(y) = V(myS) and k['lj;- 1(y)] = Sja where a= vrn;5. Since all
components of 'lj;- 1(y) have dimension r, there is a k-algebra S 1 c Sf a
with S1 = k[u1, ... , ur], the u; algebraically independent over k, and
s I a integral over sl (by the Noether Normalization Lemma). The u;
are elements of S and the natural injection 'lj;* : R -+ S factors as
'lj;* = 'lj;i o 'lj;z where 'lj;i : R[U1, ... Ur]-+ Sis given by
'lj;; ( U;) = u; , i = 1, ... , r (18.23)
(the U; being indeterminates) and 'lj;?, : R -+ R[U1, ... , Ur] is the

natural injection. In other words, 'lj; = 'lj;2 o 'lj;l where 'lj;I : X -+ Y X Ak
and 'lj;2: YxAk-+ Yis the projection 1ry. IfW = 'lj;2 1(y) = {y}xAk,
then 'lj;- 1(y) = 'lj;! 1(W) and, since k[W] ~ S1 , the restriction of 'lj;1
to 'lj;- 1(y) is a finite smjective morphism and hence, an open map.
Therefore, every point of 'lj;- 1 (y) is isolated in its fiber for 'lj; 1. But
X'= {x' EX: x' isolated in 'lj;! ('lj;1(x'))} is open in X and contains
'lj;- 1 (y). In view of theorem 18.17 and the fact that Y X Ak is normal
by the lemma which follows, 'lj;1 restricted to X' is an open mapping.

Since the projection '!f;2 is an open map, '!f;(X') = '!f;2('1/J1(X')) is open
in Y and '!f;( X) is a neighborhood of '!f;( x) = y.
Lemma 18.24. If V is a normal variety, then V X Ak is a normal

variety.
Proof: By induction we may assume that r = 1. Let R = k[V].

Then k[V x A1] = R ®k k[X] = R[X]. Let K = k(V) be the quotient
field of R so that K(X) is the quotient field of R[X]. Iff E K(X)
is integral over R[X], then f is a fortiori integral over the unique
factorizaton domain K[X] and hence, f E K[X] (proposition 16.28).
The result is then an immediate consequence of the following lemma.
Lemma 18.25. Let R be an integral domain with quotient field K

and let R' be the integral closure of R in K. Then R'[X] is the integral
closure of R[X] in K[X].
Proof ([B-2]). Iff E K[X] is integral over R[X], then
Tk E R[X] (18.26)
If m >nand the degrees of the r;, then -!I = xm- f is monic and
or, equivalently,
+ g1 Jn-1
f 1n 1 + ··· + gn = 0 (18.27)
where gn = (Xm)n + r1(Xmt- 1 + ·· · + rn. In other words,- fi and
g- 1 + giff- 2 + ··· + gn-1 are monic polynomials in K[X] whose
product is in R'[X]. Then (exercise 50.) fiE R'[X] and f E R'[X].
The Ring of Invariants
We now turn our attention to establishing the third property

of a geometric quotient. Namely, we want to show that the ring of
invariants k[Sft]G, G = GL(n,k), is k[Char(n,k)]. So, again as in
section 15, we ~onsider the polynomial ring R = k[A(+ 2 n] which we
write k[(Xij),(Yj),(Z;)], i,j = 1, ... ,n or simply k[X,Y,Z]. We let
'lji;(X, Y, Z) be given by
'lji;(X, Y, Z) =ZXi- 1 Y i = 1, ... ,n

(19.1)
'lji;(X, Y, Z) =- Xi-n(X) i = n + 1, ... , 2n
where the x~s are the characteristic polynomials and we let 'ljl : A~ 2 + 2 n
--+ Ak be the regular map ('ljl1, ... , 'ljlzn)· If Ro = k[Char(n, k)], then
(19.2)
where O(X,Y,Z) = w(X,Y,Z)I(X,Y,Z) = det H(X,Y,Z) and

w(X,Y,Z) = det Z(X,Y,Z) and 1(X,Y,Z) = det Y(X,Y,Z),
Z(X, Y, Z) being the observability map and Y(X, Y, Z) being the con•
trollablity map (see section 11). Then k[Sf,d = Ro and k[Char(n, k)]
" = "k['ljl1, ... , 'ljln]o· We recall that G = GL( n, k) acts on R via
(19.3)
and that this action preserves degrees. We also note that
Lemma 19.4. Xt(X), ... ,Xn(X), zxi-t Y i = 1, ... , are elements

of k[X, Y, z]G = RG. In other words, k['ljl1, ... ''ljlzn] c RG.
Proof: Since det(z/- X) = det(zl- gXg- 1 ), the X;(X) are

invariant. Also, since (gXg- 1 )i- 1 = gXi- 1 g-t, we have
113
Corollary 19.5. IJ(X, Y, Z) is an element of R0 .
Proof: Iff E k[S1, 1 f, then f = hfiJm with m 2 0 and h E

k[X, Y, Z] = R. Since IJm f = h and fJm f is invariant, !1 E R 0 .
Now, what we have shown so far is that k[1,61 , ... ,1,62n] is con•
tained in R 0 and hence, that 1,6*(Ro) ~ k[Char(n,k)] ~ k[1,61 , ... ,
1P2n]f) is contained in ( R 0 )(} = k[ Sf, 1 ] 0 . In other words, to establish
the third property of a geometric quotient, namely that k[Sf 1 f"="
k[Char( n, k )], it will be enough to prove the following. '
Theorem 19.7. LetS= R 0 = k[X, Y, Z] 0 and So= k[1,6 1 , ... , 1P2n]

= k[ZY, ... ,zxn-ty,Xl(X), ... ,xn(X)]. Then So= S.
We shall give several proofs of the theorem. The first involves a

"canonical form". The second involves invariants under the symmet•
ric group (Appendix C). A third proof, which actually deals with
Hank( n, k ), involves a classical determinantal approach. A fourth
proof which will not be developed here involves invoking the Hilbert•
Mumford Theory ([M-3], [M-4]).
We first note the following:
Proposition 19.8. Let w(X, Y, Z) det Z( X, Y, Z). Then

(A~ + 2 n)w is invariant under G.
2
Proof: If x = (A,b,c) E (A~ 2 + 2 n)w, then w(A,b,c) f. 0 and

w(g · x) = w( A, b, c)· (det g- 1 ) (by proposition 11.13) so that w(g · x) :f-
0.
Since (A~ 2 + 2 n)w is an open dense G-invariant subset of A~ 2 + 2 n,
an f in S = R 0 is determined by its restriction to this set (e.g. by
corollary 12.16).
Proposition 19.9. If Xc(X) = Xc is the "companion matrix" given

THE GEOMETRIC QUOTIENT THEOREM: THE RING OF INVARIANTS 115
by
0 1 0 0
0 0 1 0
Xc= (19.10)
0 1
-xn(X) -xn-l(X) -xi( X)
then
Z(X, Y, Z)X =XcZ(X, Y, Z)
Z(X, Y, Z)Y =(ZXi- 1Y)j= 1 (19.11)
[I 0 ··· O]Z(X,Y,Z)=Z
(as polynomial identities).
Proof: Since Z(X,Y,Z) = (ZXi- 1 )j= 1 (cf. 11.2), we clearly

have Z(X,Y,Z)Y = (ZXi- 1 Y)j= 1 and [1 0 ··· O]Z(X,Y,Z)
= Z. Now if M is an n X n matrix, let Mj denote the j-th col•
umn of M. Thus, X = [Xt, ... ,XN] and xr = [(Xr)t · · · (Xr)n] =
l
[xr-t X 1 • •• xr-l Xn]· It follows that
zx1
zxx1 ZXn
ZXXn
[
=
l
Z(X,Y,Z)X
zxn~lxl zxn~ 1 Xn
zx1
··· ZXn
= [ Z(~ 2 )t •· · Z(~ 2 )n
(19.12)
z(Xn)t Z(Xn)n
But, by the Cayley-Hamilton theorem, xn = -x1(X)xn-l - · · ·•

Xn(X)I and so, (Xn)j = -xl(X)(xn-l)j- · · ·- Xn(X)Ij. The as•
sertion that Z(X, Y, Z)X = XcZ(X, Y, Z) follows immediately.
Corollary 19.13. If (A, b, c) E (AZ 2 +2 n)w. then

Z(A,b,c)AZ(A,b,c)- 1 = Ac,
cZ(A,b,c)- 1 =[1 0 ··· 0],
and
Z(A,b,c)b = (cAi- 1 b)j= 1 •
Corollary 19.14. Let f(X, Y, Z) be an element of R = k[X, Y, Z]

andsetfw(X,Y,Z)=f(Xc,(ZXi- 1Y)j= 1,[1 0 ··· 0])./ffisan
element of S = R 0 , then f = fw·
Proof: It is enough to show that f = fw

on (A~ 2 + 2 n)w· But (!•
fw)(A,b,c) = f(A,b,c)- fw(A,b,c) = f(A,b,c)- f(gAg-l,gb,cg- 1 )
for (A,b,c) E (A~ 2 + 2 n)w where g = Z(A,b,c) in view of the previous
corollary. Since f is invariant, f(A, b, c)= f(gAg- 1 ,gb, cg- 1 ) and the
corollary follows.
Corollary 19.15. Iff E S, then f =!wE So.
Proof: Simply note that fw E So and apply corollary 19.14.
In other words, we have shown that S 0 = S and thus, have es•

tablished theorem 19.7 so that the third property of a geometric quo•
tient is satisfied. We note that a completely analogous proof can be
obtained using !(X, Y, Z) = det Y(X, Y, Z) where Y is the controlla•
bility map. In particular, we have the following:
Proposition 19.16. (A~ 2 + 2 n)'Y is G-invariant.
Proposition 19.17. If Xc(X) = xc is the "companion matrix"

given by
0 0 0 -xn(X)
1 0 -xn-1(X)
Xc(X) = 0 1 (19.18)
0
0 0 1 -x1(X)
then
Y(X,Y,Z)Xc =XY(X,Y,Z)
[zxi- 1 Y]j= 1 =ZY(X,Y,Z)
Y(X,Y,Z) m~Y (19.19)
(as polynomial identities).
Corollary 19.20. If (A, b, c) E (A~ 2 + 2 n),, then
Y(A,b,c)- 1 AY(A,b,c) = Ac,
Y(A,b,c)
-1
Y = [b1] ,
0
and
cY(A, b,c) = [cAi- 1 b]j= 1 .
Corollary 19.21. Let f(X, Y, Z) be an element of R = k[X, Y, Z]
and set / 0 ( X, Y, Z) ~ f(X', [ f] ,

[Zxi-t Y)j"' ). Iff is an elemenl of
S = R 0 , then f = !,.
Corollary 19.22. Iff E S, then f = !, E So.
We can interpret these results in terms of maps from A~ 2 + 2 n into

A~n and A~n into A~ 2 + 2 n. More precisely, let a 1 : A~ 2 + 2 n---+ A~n and
a2 : A~n ---+ A~ +Zn be given by
2
a 1 (A,b, c)= (Z(A, b,c)b, -x 1 (A), ... , -xn(A))

(19.23)
a2(ht, ... ,hn,Xl,···,Xn)=(Ac(X),h,[1 0 ··· 0])
where
[jJ
0 0 0
0 0 1 0
Ac(X) =
' h=
1
-xn ... . .. . .. -xt
(note that at= 9'tx)· Then it is easy to see that a1,a2 are morphisms
and that f E S = R 0 = k[X, Y, Z] 0 if and only if (a 1 o a 2 )*(f) = f
(i.e. f o a2 o a1 = f). But (at o a2)*(f) is an element of So and so,
S =So.
We now give a second proof of the third property of a geometric

quotient using the results of Appendix C and assuming that k has
characteristic zero. We let a( A) be the discriminant of A so that
a( X) is an element of k[X, Y, Z]. Then a( X) is an invariant under
G (proposition C.19) so that a(X) E S = R 0 . In fact, since a(X) =
6.(xt(X), ... ,Xn(X)), 6.(X) E So. We have:
2 2 2
Proposition 19.24. (A~ + n)a is open and dense in A~ +2 n and,
consequently, a regular function f in S is determined by its restriction to
(A~2+2n)6.·
Now let D(n, k) C M(n, n; k) be the set of nx n diagonal matrices

and let D = {(D(n,k),b,c)} viewed as a subset of Af+ 2 n. Then Dis
the affine variety in Af+ 2 n defined by the equations Xij = 0, i ::f j. If
x = (A,b,c) E (Af+ 2 n)a, then there is a Dx (= DA) in D(n,k) and
aginG with gAg- 1 = Dx such that (Dx,gb,cg- 1 ) ED n oa(A,b,c).
Iff is a G-invariant regularfunction, then f(A, b,c) = f(Dx,gb, cg- 1 )
and so, f is determined by its restriciton to (A~ 2 + 2 n)6. n D. Since
this set is open and dense in D, f E Sis determined by its restriction
fn to D. In other words, So will coincide with S i.e. Theorem 19.7
will hold, if we can prove the following:
Lemma 19.25. Iff E S, then there is an fo E So such that both f

and fo have the same restriction to D i.e. fn = /on.
We can immediately develop a proof of this lemma in the follow•

ing way: let qy: A~ 2 + 2 n---+ A~n be given by qy(A,b,c) = (cY(A,b,c),
-x1 (A), ... , -xn(A)) and, iff E R, let J : A~n ---+ Al be given by
[!J;
}(u1,···,un,X1,···,xn) = f(Ax,E1,[u1,···,un]) where Ax= Ac(x)
;, an ap p mpdate "companion matrix" and <, = if f E S =
RG, then f(A,b,c) = f(Ac,El,cY(A,b,c)) = (} o qy)(A,b,c) and so,

J( D A, gb, cg- 1) = (} o qy )( D A ,gb, cg- 1) when A has distinct eigenval•
ues; but (} o 4Y)D is an element of SoD and lemma 19.25 follows. Of
course, this approach is simply a variant on what we did before.
Now let us look at a different approach to the lemma. Let N =
Sn X D 0 (n,k) be the subgroup of G which stabilizes D (see appendix
C). We note that the action of Non Dis given by
(19.26)
where a E Sn is a permutation matrix, a = diag[a1, ... , an] with
a; =j:. 0 all i, d = diag[d1, ... , dn], d(/ = diag[d<7(1), ... , d<7(n)], b =
column (b 1, ... ,bn), b(/ =column (b<7(l), ... ,b<7(n)), c = row(c 1, ... ,
en) and c(/ = row( c(/( 1), ... , C<7(n) ). A fortiori N acts on k[D] = RD
which can be identified with the subring k[Xn, ... ,Xnn,Y,Z] of R.
Moreover, if f E S = RG, then
fD E R~ = k[Xll,Xn, ... ,Xnn,Y,Z]N.
If pis an element of Rlfy and p =!Dis the restriction of an invariant
fin S, then p- f viewed as an element of R vanishes on D. In other
words, p- f E I(D) = (Xij ), i =f. j, and therefore,
(19.27)
for g E G. However, since N < G, there may be functions in RfJ

which are not the restrictions of functions in S.
Example 19.28. Let n = 2 and let p(X11 ,X12 ,X21 ,X22 ,Y1,Y2 ,
Z 1,Z2 ) = Z 1 Y 1 Z 2 Y2 • Then pis an element of Rlfy but pis not the
restriction of an element of S. Let
so that
Y{ = (gY)t Z{ = (Zg-1 )1
Y{ = (gY)2 z~ = (Zg-1)2
and Z{Y{Z~Y{ = Z1(Y1 + Yz)(Zz- Z1)Yz = Z1Y1ZzYz- Z1Y1Z1Y2 +

Z1YzZzYz- Z1YzZ1Yz = p 9 (Xn,X12,Xzt.Xzz,Yt.Yz,Z1,Zz). If p
were the restriction of an element of S, then
would be an element ofthe ideal (X12,X21,X12+Xzz-Xn-X21, Xz1)

= (X12 , X 21 , X 22 - X 11 ) which is clearly false. Note, for example, that
if q(X, Y, Z) = Z1Y1 + ZzYz, then
Example 19.29. Let us now generalize the previous example. Sup•

pose that n > 2 and set W; = ZXi fori= 1, ... , n. Let 1/11 (Wt, ... , Wn),
tPz (W1, ... , W n ), ... , tPn (W1, ... , W n) be the elementary symmetric
polynomials in the W; and observe that the '1/Ji are elements of Rlj
(see appendix C). If j 2: 2, then tPj(W1, ... , Wn) is not the restriction
of an element of S. To see this, let g E G be given by
1
1
where 0 s,t is an s X t matrix with 0 entries. Then t/J j contains a

unique term involving W1 Wz and the remainder is invariant under
g. Thus, t/J]- '1/Jj is not an element of (X;j,X~) and so tPj is not
the restriction of an invariant to D. For example, if n = 4, then
'I/J 2 (W) = W1Wz+W1W3+W1W4+WzW3+WzW4+W3W4 = WtWz+
(W1 + Wz)(W3 + W4) + W3W4, 1/13 (W) = W1W2W3 + W1WzW4 +
WtW3W4 + WzW3W4 = (W1Wz)(W3 + W4) + (Wt + Wz)W3W4 and
t/J 4 (W) = (Wt Wz)(W3 W4) and Wt + Wz is invariant under g as are
w3 + w4 and w3w4.
Bearing these examples in mind, we let W; = Z;Yi, i = 1, 2, ... ,
n. If a = diag[a 1 , ... , an] E Do(n, k), then W;a = Z;ai 1 a;Yi =
THE GEOMETRIC QUOTIENT THEOREM: THE RING OF INVARIANTS
121
Z;Y; = W; so that the W; are invaria nt under Do(n,k ). More•
over, X{f = a;X;;a i 1 = X;; so that the X;; are also invaria nt un•
der Do(n,k ). Set X= (Xn, ... ,Xnn) and W = (Wn, ... ,Wnn)
so that Sn acts on k[X, WJ C RD via a· (X, W) = (X a, Wa) where
Xa = (Xa(1)a(1), ... ,Xa(n)a(n)) and Wa = (Wa(1)ak1), ... , Wa(n)a(n))·
Then k[X, WJ 5 n = k[Xn, ... , Xnn, Wu, ... , Wnn] n C Rfy. In fact,
we have:
Proposition 19.30. Rfy = k[X, WJ 5 n.
Proof: If p E Rfy, then p = Po + P1 + · · · + Pm where p; is

homoge neous of degree i (or 0). Since the action of N preserv es de•
grees, each p; E Rfy and so, we may suppose p homoge neous. Let
p = :Earst Xrzsyt = :Earst( XnY1 ···(Xnn YnZt 1 · ··Z~ny1t 1 ·· ·Y~n
with arst ::J 0. If a= diag[a1 , ... ,an] E Do(n,k ), then pa =
:Ea rs t<Xt1-s1 . atn-sn
n (X11 )rl. ··(Xnn )rnzsl ... zsny1ti ... ytn It fol-
1 1 n n ·
lows that pa = p implies s1 = t 1, ... , Sn = tn (for exampl e, take
a 2 = · · · = <Xn = 1 and equate coefficients in pa - p to obtain
( ai 1 - 81 )arst = 0). In other words, p E k[X, WJ 5 n.
Now let U1 =(Xu, WI), Uz = (Xzz, Wz), ... , Un = (Xnn, Wn) so
that k[X, W] = k[U]. Then Sn acts on k[U] via
(19.31)
and k[X,WJ 5 n = k[U] 5 n. Writing Uj = (Up,U p) and letting O(p)

be the sum of the elemen ts in the orbit of p under Sn where p is a
monom ial in the uj' we have:
Proposition 19.32. Suppose that k has characteristic zero. Then

k[U] 5 n is generated by the multilinear terms O(U1; 1 • • • Usi, ), s::; n.
Proof: See Append ix C, corollar y C.27.
Example 19.33. Let n = 2 so that U1 = (Xu,W l) and Uz =

(X 2 z, W 2 ). Then k[X, WJ 5 n is generat ed by the terms O(X ) =
11
Xn + X22, O(W1) = W1 + Wz, O(XuX zz) = 2XnX n, O(W1 Wz) =
W1 Wz, and O(Xn W!) =Xu W1 +XzzW z. We observe that O(Xn) =
XI(X)D , O(W1) = (ZY)D, O(XnX 22) = (2xz(X ))D, O(Xu W1) =
(ZXY)D and that W1 Wz ::J fD for any fin S. Thus, iff is an
invariant, then fD E k[Xn +X22, Wt + W2,XnX22,XnW1 +X22W2]

and hence, there is a fo in So with !D = foD·
Now let SD = {JD : f E S} and let SoD = {foD : f E So}.

Then SoD C SD C k[X, W] 5 " = R~. We observe that SoD is
generated by the 2n terms O(Xn), O(Xft), ... , 9(Xft), O(Wt),
O(XuWt), ... ,O(Xf1- 1 Wt)· Let Pii = O(Xl 1 W{), i < j be the
remaining generators of k[X, W] 5 " so that R~ = SoD[pij]· However,
application of equation 19.27 and the calculations of example 19.29,
show that none of the Pij are in SD and in addition, that if p E SD,
then p E SoD which establishes lemma 19.25. In other words, the sec•
ond proof of the third property of a geometric quotient is complete.
We now turn our attention to the third proof of the ring of invari•
ants property of a geometric quotient. If x = (A, b, c) is an element
of A~ 2 + 2 n, we let Y 1 : A~ 2 + 2 n __. M( n, n + 1) and Z 1 : Af+ 2 n __.
M ( n + 1, n) be the morphisms given by
Yt(x)=Yt(A,b,c)=[b Ab ··· Anb]=[Y(A,b,c) Anb]
Z1 (x) =Z (A,b,c) = [ c~] = [zc:A~,c)]

1 (19.34)
cAn
where Y, Z are the controllability and observability matrices. Letting
M*r(s, t) be the set of s X t matrices of rank r, we have:
Proposition 19.35. Y 1 (x) E M*n(n,n + 1) if an only if x is

controllable; and, Z 1 (x) E M*n(n + 1, n) if and only if xis observable.
Now let <p: A~ 2 + 2 n __. M(n+ 1, n) X M(n, n+ 1) be the morphism

given by
(19.36)
Then r.p(Sf, 1 ) C M*n(n + 1,n) M,.n(n,n + 1). If ME M(s,t) and
[MI .
X
z.1 , ... , Zr,

. Jt.
. ... , Jr. are In
. d.ICes, t h en det it ... ir] . t he r
. IS X r
Jt · · · Jr
minor of M with rows it, ... , ir and columns it. ... ,ir· We let a(M) =
det [ M I ~ : :::] for M in M ( n + 1, n) and /3( M) = det [Ml ~ : : ::]
forM in M(n,n + 1).
Proposition 19.37. Z 1(Sf, 1) = (M(n + 1,n))a and Yt(Sf, 1) =

(M(n,n + 1))13.
Proof: Enough to treat Yt(Sf 1). Clearly, Yt(Sf 1) C (M(n, n +

1))13. If [vt···Vn w] E M(n,n+'1),8, then det[vJ.:.,vn] i 0 and
w = l:j= 1 ajVj, ai E k. Let g E G be such that 9 · [v1 ... vn] = I. Set
0 0 al
1 0
A*= 0 b* = EI c* =En= (0 ... 0 1)
' '
0 1 an
· r10r J· -- 0 , ..• , n - 1 an d A*nb* -- L..Jj=l

so th a t A *ib* -- EJ+l ~n a3·E3· •
Let b = 9- 1 b* and A = g- 1 A*g and c = c*g. Since g · [vi··· vn] =
I= [E1 ···En], we can see that b = g- 1E1 = v1, Ab = 9- 1A*g9- 1 E1 =
g - 1"2
.c
-- v 2, ... , Aib-
1
- g- A*ib* -- g- "J+l
1 .... ·
-- vi+1 1'ror J. -- 0 ' 1' ... '
n - 1 and A nb = g- A *nb* = 'Eajg- Ej = Z:j= 1 ajVj = w. In other
1 1
words, Y1(A,b,c) = [vt, ... ,vn w]. Since
detZ(A,b,c) = detZ(A*,b*,c*) · detg i 0,
(A,b,c) is in S} 1 .
'
Corollary 19.38. cp(Sf, 1 ) C M(n + 1, n)a x M(n, n + 1) 13 .
Let 'ljJ: M(n + 1,n) x M(n,n + 1)--+ M(n + 1,n + 1) be the

morphism given by
'1/J(Z,Y) = ZY (19.39)
If O(M) = det [MI ~::::]forM in M(n + 1,n + 1), then '1/J(M(n +

1,n)a X M(n,n + 1),8) C M*n(n + 1,n + 1)~:~. We let .Cn+l be the
(linear*) variety in M(n + 1,n + 1) defined by the equations
Mii- Mii =0 i,j = 1, ... , n + 1

(19.40)
Min+l - Mi+1n =0 i = 1, ... ,n
* See definition 20.5.
and let 7-£ = .Cn+l n M*n(n + 1, n + 1)9. 7-£ is a quasi-affine variety

since it is an open subset of .Cn+l·
Proposition 19.41. ( 1/7 o cp)(Sf, 1 ) = 7-£.

Proof: If X E sr,l' then ( 1/7 0 tp )(X) = Z 1(X )Y1(X) clearly satisfies
(19.40). If H is an element of 7-£, then there are h1, ... , h2n+l and
a1, ... , an in k such that
and
n
hn+j = L at_hf.+j-1,
t'=1
for j = 1, ... , n + 1. Since det [HI~:::~] i' 0, there is, by realization

theory (theorem 10.18), an X= (A,b,c) in srl such that he= cA£-lb,
£ = 1, ... , 2n + 1(· · ·). In other words, (1/7 o ~)(x) =H.
Corollary 19.42. cp(Sf 1 ) = 1/7- 1 (7-£) so that cp(Sf 1 ) is a quasi-affine

' '
variety.
Proof: Clearly, cp(Sf 1 ) C 1/7- 1 (7-£). If (Z,Y) E 1/7- 1 (7-£), then

1/7( z, Y) = ZY = H E 7-£ ~nd, by the proposition, there is an X in 1 sr
with '!f1(cp(x)) =H. Let '
z = [ ~n] , Y = [Yn y] , Z 1 (x) = [~r] , Y1(x) = [Y1nYd

Since H = ZY = Z 1(x)Y1(x), we have ZnYn = z;
Y1n, zYn = z 1Y1n,
ZnY = z;yl, and zy = Z 1 YI· But ZnYn = z;Yln is non-singular so
that g = Yn Y;-; E G = GL(n, k) and Z = Z 1 (x)g-1, Y = gY 1 (x)
(as Z; 1 z;
= Yn Y;-;). If x = g · x, then cp(x) = (Z 1 (x),Y1 (x)) =
(Z 1 (x)g-l,gY 1 (x)) = (Z, Y). In other words, we have 'lj1- 1 (7-£) C
cp(Sf,I)·
Set Sf 1 = <p(Sf 1 ). <pis a morphlsm and as a map of Sf 1 into

sr'1 is obviously
' '
surjective.
'
Proposition 19.43. <p : Si\ -+ Sf 1 is injective.

' '
Proof: Suppose that <p(x) = <p(xi) where x = (A,b,c) and x 1 =
(A1.b1,c1)· The~ b = b1, c = c1, and (say) Y1(A,b,c) = Y1(A~,b,c)
so that Ai b = Ai b, j = 0, ... , n. Since Vj = Ai b, j = 0, ... , n - 1 is
a basis of kn, to show that A = A 1 , it will be enough to show that
Avj = A1vj, j = 0, ... ,n- 1. But Avj = Ai+lb = A1 · Aib = A1vj
for j = 0,1, ... ,n-1 since Y1(A,b,c) = Y1(A~,b,c).
Proposition 19.44. <p : sr' 1 - sr'1 is an isomorphism.
Proof: Since <pis bijective, we need only show that <p- 1 is a mor•
phism. If (Z 1(x), Y1(x)) = <p(x) E Sf 1, then <p- 1(Z 1 (x), Y1(x)) =
X= (A,b,c) and b = (Yl(x)h, c = (Z 1 (x)) 1 . Clearly, band care reg•
ular and so, we need only show that A(Z 1 (x), Y 1 (x)) is regular. Let
(say) Y1(x) = [Y1 · · · Yn Yn+1] with Yn+I = a1Y1 + · · · + anYn
and let g E G be ([Y1 · · · Yn])- 1 • Then, g,g- 1 are regular functions of
Y1, ... , Yn on Sf, 1 and A = g- 1 A *g where
Thus, it suffices to show that the a; are regular. But
is clearly regular and so, <p- 1 is a morphism.
Now we observe that G acts on M(n + l,n) x M(n,n + 1) via

g · (Z, Y) = (Zg- 1 ,gY) (19.45)
and that Sf 1 is invariant under this action. However, in view of

proposition i1.13, <p(g · x) = g · <p(x) so that <pis a G-isomorphism.
Thus,
Corollary 19.46. k[51 1 f = k[Sft] 0 .

' '
Hence, in order to establish the ring of invariants property of a
geometric quotient, we need only prove the following:
Theorem 19.47. k[Sf, 1 ] 0 = k[H]

Since k[H] ~ 7/J*(k[H]) c k[Sf, 1 ] and 7/J(g · (Z,Y)) = 1/J((Z,Y)),
it is clear that k[H] C k[Sf,tJ 0 and so, we must show that k[Sf,tJ° C
k[H] (or, more precisely, 1/J*(k[H])). Let V C S]\ be the set of ele-
ments of the form '
( 19.48)
i.e. Z(x) =I. We consider G XV and we observe that G acts on G XV

VIa
(19.49)
We have:
Lemma 19.50. k[G x V] 0 = k[V]

Proof: Let R = k[G], 5 = k[V]. Then R 0 = k[G] 0 = k. It
follows from lemma B.14 (say), that (R0Rc 5) 0 = S i.e. that (k[G]®k
k[V]) 0 = (k[G x V]) 0 = k[V]. This can also be shown directly. If
f = ~ri 0 Si E k[C] ®k k[V], then f(g, v) = ~ri(g)si(v). If f is
invariant, then f(g,v) = f(g ·I,v) = f(I,v) = ~ri(I)si(v) so that
f = ~ri(I)si E k[V].
Define a morphism 1 : G x V -. Sf, 1 by setting
l((g, v)) =g ·v (19.51)
Then:
Lemma 19.52. 1 is a G-invariant isomorphism and so, k[Sf 1 f =

k[G X V] 0 = k[V]. '
Proof: Clearly 1(g1 · (g, v)) = 1((g1g, v)) = (g1g) · v = g1 · (gv) =

g1 ·1((g, v)) so that 1 is C-invariant. Define 1- 1 : Sf, 1 -. G x V as
follows:
so that (obviously) 1 ·1- 1 =identity and ~- 1 o 1 =identity.
Corollary 19.53. Iff is an element of k[Sf 1 ] 0 , then f E k[zz-I,

ZY,Zy] (i.e. f is afunction of zz- 1 , ZY and Zy).
In view of this corollary, we need only prove the following:
Lemma 19.54. k[zz-1, ZY, Zy] c k[1t].
Proof: Since k[H] = k[ZY, Zy, zY, zy]det(ZY), it is enough to

show that zz- 1 is in k[1t]. But z- 1 = Y(ZY)- 1 = Yadj(ZY)/det
(Z Y) and so zz- 1 = (zY)adj(Z Y)/ det(Z Y) is in k[H].
We also note that, in view of proposition 19.41 and the results

of section 10, it is quite easy to show that Hank( n, k) and 1t are
isomorphic (Exercise 56.).
A fourth proof using the Hilbert-Mumford theory can also be
given but we shall not do so here.
Simple Points
We have shown that Char(n,k) (or Hank(n,k) or Rat(n,k)) is

a quasi-affine variety of dimension 2n which is a geometric quotient
of Sf 1 modulo GL(n,k). We shall now show, that, in addition,
Hank( n, k) or Char( n, k) or Rat( n, k) is what is called nonsingular
i.e. has no singular points. This will, of course, again require some
algebra.
Definition 20.1. A mapping a : Af --.. Af

is called an affine
transformation if there is a gin GL(n,k) and a ~o in such that Af
a(~)= 9~ + ~o ( 20.2)
for all~ in Af. If g =I (the identity), then a is called a translation.
Let A( N, k) be the set of all affine transformations of Af/ and

write a= (9,~o) for a in A(N,k). If a1 = (g1,6) and O'z = (92,6),
then ( 0'1 O'z )( ~) = (9192)~ +916 +6 and we may define a multiplication
in A(N,k) by setting
(20.3)
Then A(N, k) is an algebraic group with identity i = (I, 0) and (g, ~o )- 1

= (g-1, -g- 1 ~0 ) (see exercise 57.). Since A(N, k) acts on via Af
(20.2), A(N, k) also acts on k[X] = k[XI, ... , XN] via
(20.4)
and this action preserves degrees. If Z; = (gX + ~0 );, i = l, ... ,N,

then the Z; are transcendental over k and k[Z1 , ... ,ZN]= k[X1, ... ,XN]·
If V CAf is an affine algebraic set with I(V) = (h (X), ... , fr(X )),
128
AFFINE ALGEBRAIC GEOMETRY: SIMPLE POINTS 129
then a(V) is also an affine algebraic set with I(a(V)) a(I(V)).

Moreover, if Vis an affine variety so that I(V) is prime, then a(V) is
also an affine variety and k[V] ~ k[a(V)] so that dimk V = dimka(V).
Definition 20.5. Let £1, ... ,lm be independent linear forms (ho•
mogeneous polynomials of degree 1) in k{Xt, ... , XN)and let a1, ... , am
be elements of k. Then the affine algebraic set V( £1 - a1, ... , lm- am)
is called a linear variety.
Proposition 20.6. If V = V( £1 - a1, ... , Rm - am) is a linear

variety in A.f:. then Vis irreducible, dimk V = N -m, and Vis isomorphic
to a subspace of A. f:.
Proof: Since £1, ... , lm are independent, there is an affine trans•

formation a = (g,~o) such that if Z; = (gX + ~o);, then Z; =
l;(X) - a;, i = 1, ... , m. In other words, a(V) = V(Zt, ... , Zm)·
But V(Z 1 , ... ,Zm) is a subspace of Af: and is irreducible and has
dimension N- m. Since V and a(V) are isomorphic, the proposition
is established.
Definition 20.7. Let V be an affine variety so that p = I(V) is

prime and let ~ = (6, ... , ~N) E V. Then the (Zariski) tangent space
to V at ~ is the linear variety defined by
of
LN
i=l
[)X
t
(~)(X;- ~i) =0 (20.8)
for all f E p. We denote this variety by Tv,~.
We observe that, in view of proposition 20.6, Tv,~ may be viewed

as a (vector) subspace of Af: with origin ~. We also note that if
h, ... , fr are a basis of p = I(V) (i.e. p = (ft, ... , fr )), then
(20.9)
where J(ft' ... 'fr; xl' ... XN )(~) = ( %1~ (~)) is the Jacobian matrix
of ft, ... , fr with respect to X1, ... , XN at ~· We then have:
Proposition 20.10. If V = V(£ 1 - a 1 , ... ,.em- am) is a linear

variety, then dimkTv,.; = dimk V for all ~ in Y.
Proposition 20.11. If V is an affine variety and v is an integer,

then {~ E V : dimk Tv.~ ~ v} is Zariski closed in V.
Proof: In view of (20.9), {~ E V: dimkTV,{ ~ v} is the zero set of

jJ and the ideal of N- v+ 1 X N- v+ 1 minors of J(ft, ... Jri Xt, ... ,
XN)·
Definition 20.12. Let V be an affine variety and let~ E V. We

say that ~ is a simple point of V if dimkTV,{ = dimk V and that V is
nonsingular or smooth if all points of V are simple.
Corollary 20.13. A linear variety is nonsingular.
We shall show that the tangent space can be defined intrinsically

and that the definition is "local". We require the results on derivations
developed in Appendix D.
Definition 20.14. Let R beak-algebra and let M be an R-module.

A k-derivation of R into M is a k-linear map D : R -+ M such that (i)
D(a) = 0 for all a E k and (ii) D(r1r2) = r1D(r2) + r2D(rt) for all
r1, r2 E R.
We write Derk(R, M) for the set of all k-derivations of R into

M and note that Derk(R, M) is an R-module in a natural way. If
</> : R -+ S is a homomorphism of k-algebras and N is an S-module,
then N can be viewed as an R-module by setting
r · n = </>(r)n (20.15)
for r E R and n E N and there is a natural homomorphism 4>

Derk(S, N)-+ Derk(R, M) of R-modules given by
4>(D) =Do</> (20.16)
forD E Derk(S, N).

Now let V be an affine variety and let ~be an element of V. We
set R = k[V] and J( = K(R) = k(V). If M~ = I(~) is the maximal
ideal of~ in k[X1, ... , XN] = k[X] and m~ = Md I(V) is the maximal
ideal of~ in R, then k = Rfmf. = k[X]fME. so that k may be viewed

as an R-module (via the homomorphism R - t RjmE,.) or as a k[X]•
module (via the homomorphism k[X] - t k[X]/ME,.). We may thus
consider the spaces of derivations Derk(R,k) and Derk(k[X],k) at~·
In that case, we write Derk(R, kE,.) and Derk(k[X], kf.) and speak of
derivations centered at~·
Proposition 20.17. Tv,f. = Derk(R, kf.) = {D' E Derk(k[X], kf.) :

D'(I(V)) = 0}. Hence, if V C W, then Tv,f. c Tw,f.·
Proof: Clearly a derivation Din Derk( R, kE,.) is the same thing as a

derivation D' in Derk(k[X],kE.) such that D'(p) = 0 where p = I(V).
If D' E Derk(k[X], kE.), then D' is determined by its values ai =
D' (Xi), i = 1, ... , N, and conversely. Since D' f = 2::~ 1 88 ~)D' X;J; (
for all fin k[X], the derivations D' such that D'(P) = 0 correspond to
values a1, ... , aN such that 2::~ 1 a~; (~)a; = 0 for all fin p and the
proposition holds. (The last assertion follows from I(W) C I(V).).
If 1/J : V - t W is a morphism and 7/J* : k[W] - t k[V] is the

comorphism of 1/J, then there is an induced homomorphism (d't/J)f. :
Tv,f. - t Tw,.p(f.) of the tangent spaces given by
(d't/J)E.(D) =Do 1/J* (20.18)
where DE Derk(k[VJ,kd. We note that (d't/J)E,. is a linear map which

we call the tangent map at ~.
Lemma 20.19. Let 1/J : V - t W be an isomorphism of affine varieties.

Then (d't/J)E. is an isomorphism ofTv,f. onto Tw,.p(O·
Proof: Let R = k[V], S = k[W] and let 7/J* : S - t R be the

comorphism of 1/J. We let 7/J*: Sjm..pw - t Rfmf. be the map given by
1{;*(7) = 1/J*(f) = (! 0 1/J) (20.20)
for f E S. If g E mf., then there is an fin S with 1/J*(f) = g and so,

mf. C 7/J*(m.,p(f.))· Conversely, iff E m.,p(f.), then 7/J*(f)(~) = J('t/J(~)) =
0 and so, 1/J * is well-defined. Clearly 1{;* is an isomorphism and so,
kf." = "k.p(f.)· We write {/;* for (d't/J)E,. so that ({/;*)(D) = Do 1/J* for
D E Derk(R, k€)· If ;j;*(D) = 0, then (Dol/;*)(!)= 0 for all finS and

therefore, D(g) = 0 for all g in R as 1/;* is an isomorphism. In other
words, ;j;* is injective. On the other hand, if iJ E Derk(S,k..p(€)), then
D =Do 1/;*- 1 E Derk(R, k€) and ;j;*(D) = iJ so that ;j;* is surjective.
Thus, the lemma is established.
Corollary 20.21. Let '1/J : V -+ W be an isomorphism of affine vari•

eties. If V is nonsingular, then W is nonsingular.
Corollary 20.22. Let V and W be affine varieties and let '1/J be an

isomorphism ofV onto an open subvariety ofW. If~ E V, then (dl/;)€ is
an isomorphism of Tv,€ onto Tw,..p(€)·
Proof: Since the principal affine open sets form a base for the
Zariski topology, we may, in view of the lemma, assume that V = WI
for f E k[W] and that 1/; : WI --+ W is the natural injection. Then
k[WI] = k[W][Y]/(1 - fY) and a derivation D in Derk(k[W], k€)
extends to a derivation D' in Derk(k[WI],kd by setting D'(Y)
-(Df)(~)/ /(~) 2 • The result follows easily.
Corollary 20.23. If V is a principal affine open subset of a linear

variety, then V is nonsingular.
Corollary 20.24. Char( n, k ), Hank( n, k) and Rat( n, k) are all non•

singular.
We shall soon use corollary 20.23 to show that Hank( n, k) is non•

singular in a different way than simply observing that it is a principal
affine open subset of A~n. Before doing so, we shall show that a prod•
uct of nonsingular varieties is nonsingular and that the singular points
(i.e. non simple) lie in a closed set.
Proposition 20.25. Let V and W be affine varieties and let ( ~, 77)

be an element of V x W. Then
Tv,€ X Tw,ry ~ TvxW,(€,"1) (20.26)
and therefore, if~ is simple on V and 77 is simple on W, then (~, ry) is

simple on V x W.
Proof: Let a, f3 be the natural isomorphisms of V, W into V x W

given by
a.(v) = (v,ry) , f3(w) = ((,w) (20.27)
so that
a.*(f0g)=Jg (ry)' (3*(!0g)=f( ()g (20.28)
for the comorphisms. We note that a*, (3* are surjective. We define a
map (do.)~ X (df3)1J: Tv,~ X Tw,1J-+ Tvxw,(~,1J) by setting
(do.)~ x (df3)1J(D,D') =Do a*+ D' o (3* (20.29)
for derivations D, D'. If D o a* + D' o (3* = 0 and f is an element

of k[V], then f 0 1 is an element of k[V X W], a*(! 0 1) = f and
[Do a*+ D' o(3*](! 01) = D(f) = 0. In other words, D = 0. Similarly,
D' = 0 and so, (do.)~ x ( df3)1J is injective. Now suppose iJ is an element
of Derk(k[V X WJ,k(~,1J))· Set D(f) = D(f 0 1) for f E k[V] and
JZ'(g) = D(10g) forgE k[W]. T~en D(f0g) = D((f01)(10! ])) =
D(f01)(10g) ((, 77)+(!01)((, ry)D(10g) = D(f01)g(ry)+ f(()D(10
g)= (Do a*)(! 0 g)+ (D' o (3*)(! 0 g)= [Do a*+ D' o (3*](! 0 g)
so that (do.)~ x ( df3)1J is surjective. As for the final assertion, simply
note that dimk(Tv,~ X Tw,1J) = dimkTv,~ + dimkTw,w
We now make the following
Definition 20.30. Let Sing(V) = {( E V: (is singular}. Sing(V)

is called the singular locus of V.
We shall show that Sing(V) is a proper closed subset of V.
Proposition 20.31. Let V be an affine variety. Then there is a

nonempty open subset U of V such that dimkTv,~ = dimk V for ( in U.
Proof ([M-2]). Now dimk V = tr. deg ]( jk where k = k(V) is

the function field of V. Since ]( is separably generated over k ( Ap•
pendix D), we have tr. degK/k = dimkDerk(K, K) = dim 1(Derk
(k[V],K). But Derk(k[V],K) = {D' E Derk(k[Xl, ... ,XN],K) :
D'(p) = 0} where jJ = I(V). If jJ = (JI, ... ,fr), then we have
dimKDerk(k[V ],K) as the dimension over]( of the subspace of Afi.
defined by the vanishing of the terms 2::;': 1 ~_f, Xi, j = 1, ... , r. Thus,
it will be enough to show that rank J(fi, ... ,fr;Xl, ... ,XN) in]( is
the same as rank J(fi, ... , fr; X1, ... , XN )(0 for all~ in an open sub•
set U of V. If pis the rank of J(fi, ... ,fr;Xl,·· ·,XN) inK, then
there are P E GL(r, k) and Q E GL(N, K) such that
where J(fl,···,f r;Xl,···, XN) = J(ft, ... ,fr;Xl,·· ·,XN) modulo

jJ. Let P = P0 jp 0 , Q = Qo/qo where Po, Qo have entries in k(V] and
Po,qo E k[V]. Let f = poqodetP odetQo and U = V1 (a principal
affine open subset of V). If~ E U, then
so that rankJ(fi, 0 0 0 'fr; xl, 0 0 0 XN )(0 = p and the proposition holds.

Corollary 20.32. dimkTv.E ~ dimk V for all~ E V.
Proof: Since {~ : dimkTV,E ~ dimkV} is closed by propositio n

20.11, its compleme nt is open and hence, must be empty in view of
the propositio n and the irreducibi lity of V.
Corollary 20.33. Sing(V) is closed (as Sing(V) = { ~ : dimkTV,E

~ dimkV + 1}.
Now let us turn our attention to Hank( n, k ). Let Sym( n, k) be

the set of n x n symmetri c matrices with entries ink. Then Sym(n,k)
is a linear variety in A;; 2 as it is defined by the equations X;j- Xji =
0 fori -::J j. Moreover, dimkSym (n,k) = n(n + 1)/2. We identify
Sym( n, k) with A~( n+l) 12 by using the coordinat e functions Xij, i ~ j,
i,j = 1, ... ,n. Let H(n,k) be the subset of Sym(n,k) consisting of
matrices of the form
hl h2
[ h2 h3
H= . . (20.34)
h.n hn.+l
where ht, ... , h 2 n-l are elements of k. Since Hank( n, k) is isomorphic

with the set H(n,k)detH X Al, it will be enough, in view of corollary
20.23, to show that H(n,k) is a linear variety in Sym(n,k) in order
to prove that Hank( n, k) is nonsingular.
Example 20.35. Let n = 3. Then H(3, k) is defined by the

equation X 13 - X 22 = 0 in Sym(3,k) and so H(3,k) is a linear variety
of dimension 5.
Example 20.36. Let n = 4. Then H( 4, k) is defined by the

equations X 13 - X 22 = 0, X14- X23 = 0, X24- X33 = 0 in Sym(4, k)
and so H (4, k) is a linear variety of dimension 7.
Generalizing these examples, we have:
Proposition 20.37. H( n, k) is a linear variety of dimension n(n +

1) /2 - ( n- 1)( n - 2)/2 = 2n - 1 and hence, Hank( n, k) is nonsingular.
Proof: We use induction. Assume that H( n, k) is a linear variety

of dimension 2n- 1. If H = (hij)i,j=l, ... ,n+l E H(n + 1, k), then
Hn = (hij)i,j=l, ... ,n E H(n,k) satisfies (n-1)(n-2)/2linearequations
in Xij ( i ~ j, i, j = 1, 2, ... , n) and the elements hin+l of the last
column of H satisfy the n- 1linear equations
(20.38)
i = 1, ... , n - 1. Conversely, any solution of the ( n - 1)( n - 2)/2

equations of H(n,k) and then- 1 equations (20.38) determines an
element of H(n + 1, k). Since (n -1) + (n- 1)(n- 2)/2 = n(n -1)/2,
H( n+ 1, k) is a linear variety of dimension [( n+ 1)( n+2)-n( n-1))/2 =
2( n + 1) - 1 and the proposition is established.
Now let us give several more interpretations of the tangent space

in terms of the local ring O{,V of a point and the maximal ideal m{ of
a point. If V C Af is an affine variety, p = I(V), and~ E V, then we
recall (section 9) that O~,v = {! fg: f,g E k[V],g(~) ::j:. 0}. It follows
that if D E Derk(k[V], kd, then D extends uniquely to a derivation D'
in Derk(O~.v, k~) by setting D'(f fg) = (g(~)D f- f(~)Dg)fg(~) 2 • In
other words, Tv,{ = Der k( 0{, v, kd so that the tangent space at ~ can
be identified with the space of k-derivations of O~.v which are centered
at ~- Let R = k[V] and let m~ = {fER: J(O = 0} be the maximal

ideal of regular functions on V which vanish at r
Then mdm~ is a k
(= kdvectorspace. IfD E Tv,(= Derk(R,k€) = Derk(O(,v,k€),then
D(m~) = 0 so that D defines a linear functional on mdm~ i.e. D may
be viewed as an element of the (k-) dual space (mdm~)*. Conversely,
if e is a linear functional on mdm~, then the map De : R--. k€ given
by
De(!)= C(f- f(~)) (20.39)
where f- f(O is them~ residue off- J(€),is an element of Derk(R, k€}
= Tv,€· Thus, we have:
Proposition 20.40. Tv,E = (mdm~)*.
Summarizing we have shown that Tv,€ = V(:E :J;(€)(Xi- €;)) =

Derk(R,kE) = Derk(OE,v,kd = (mEfmV*·
Now we have established three of the major results of scalar
linear system theory, namely: (i) that the Laurent map L is a k•
isomorphism between Rat(n,k) and Hank(n,k) (Theorem 8.12 and
Corollary 9.15); (ii) that there exist minimal realizations (Theorem
10.18) and that any two minimal realizations are isomorphic with
a unique isomorphism (the state space isomorphism theorem 11.19);
and, (iii) that (Char(n,k),9lx) is a geometric quotient of 51, 1 mod•
ulo G L( n, k) where Char( n, k) is a nonsingular quasi-affine variety of
dimension 2n (theorem 14.20 and corollary 20.24). The fourth major
result is the so-called pole placement theorem which we prove in the
next section.
21. Feedback and the Pole Placement Theorem
We now turn our attention to the study of feedback. Let G =

GL(n,k) and let k* = GL(1,k) = k-{0} = Al-{0}. WenotethatG
and k* are (quasi-affine) varieties. Now let K be an element of ( kn )'.
Then K may be viewed as an element of Hom k(kn, k) which in turn
may be viewed as AJ: (written as a row). Consider the set r 1 (n, 1) =
G X Hom k( kn, k) X k* = G X AJ: X k* and define a multiplication in
r J( n, 1) as follows:
(21.1)
where g,g1 E G, K, K1 E AJ:, and a,a1 E k*. We then have:
Theorem 21.2. (i) r 1( n, 1) is a group and a quasi affine variety;

(ii) the mappings fL: f J(n, 1) X f J(n, 1)-+ f J(n, 1) and i: f J(n, 1)-+
f t(n, 1) given by J.L(/,/1) = //1 and i(!) = 1-1, where /,/1 E f t(n, 1),
are morphisms; (iii) the mapping r : r 1( n, 1) -+ GL( n + 1, k) given by
r([g,K,a]) = [J( ~] (21.3)
is an isomorphism off 1( n, 1) onto a closed subgroup f' 1(n, 1) ofGL( n+

1, k ); (iv) f J( n, 1) is a linear subvariety of GL( n + 1, k) (therefore, non•
singular) of dimension n 2 + n + 1 = (n + 1) 2 - n; and, (v) r 1 (n, 1) acts
2 2 2 2
on A~ + n via the morphism </> : f f x A~ +2 n -+ A~ +2 n given by
(21.4)
or, equivalently, f J( n, 1) acts on A~ 2 + 2 n via the morphism ¢ : f' 1 ( n, 1) x

A~ 2 + 2 n -+ A~ 2 + 2 n given by
forgE G, J( E Ak', and a E k*.
137
Proof: Since fJ(n,1) = G x Ak x k*, it is clearly a quasi affine

variety. The identity in fJ(n,1) is [1,0,1] and if 1 = [g,K,a], then
1 -1 = [g-1, -a- 1J(g- 1 , a- 1]. Moreover, the multiplication (21.1) is
clearly associative. Thus, (i) is established. As for (ii), both J.L and
i are given by regular functions and so, are morphisms. We note as
well that r is clearly a morphism. Since
and
!
r([g, K, a][g1, K1, a1]) = [ Kg 19 1aJ(1 a~J
- [i( ~] [ill ~J
for g E G, J( E Ak, and a E k*, r is a group homomorphism. More•
over, Tis obviously injective. Iff J(n, 1) = r(f J(n, 1)), then f J(n, 1)
is the linear subvariety of G L( n + 1, k) defined by the equations
(21.6)
r
and so, J( n, 1) is nonsingular (corollary 20.23) and has dimension
( n + 1)2 - n = n 2 + n + 1. If
g - = [g]( 0] E fJ(n,1)
a -
then r- 1 (9) = [g,K,a] and so, r- 1 is also a morphism. In other
words, (iii) and (iv) hold and r is an isomorphism of algebraic groups.
If")' = [g, K, a] and 1'1 = [g1, K1, a1], then

(1'1'!). (A,b,c) = [gg 1 ~] [ ~ ~]
[ (gg1)-l 0 ]
-(aa1)- 1(Kg1 + aK1)(gg1)- 1 (aal)- 1
= [~ ~] [g~ ~] [ ~ ~]
0 ] [
-1
Ql -Q
!!.1-1Ag -1
J>'
FEEDBACK AND THE POLE PLACEMENT THEOREM 139
and so, </> or 4> define actions of r 1 ( n, 1) or f' 1 ( n, 1) on A~ 2 + 2 n.

Since r 1 ( n, 1) and f' 1 ( n, 1) are isomorphic, we shall not distin•
guish between them and shall use either form interchangeably. We
then have:
Definition 21.7. The group f J(n, 1) is called the state feedback

group (of type (n,1)) and the action of r 1(n,1) on A~ 2 + 2 n is called
state feedback.
For simplicity, we writer fin place of r 1 (n, 1) in the remainder

of this section. If x = (A, b, c) E A~ 2 + 2 n and if 1 E r f• then </>(!, x) =
"'I • X E Ank + n and 'I'
,A is a morphism of r
f X Ank + n into Ank + n • We
2 2 2 2 2 2
want to study the properties of this morphism and of state feedback.
Proposition 21.8. Let Y(A,b,c) = [b Ab · ·· An- 1 b] be the

controllability map (section 11 ). Then the rank of Y (A, b, c) is invari•
ant under state feedback and so, controllability is invariant under state
feedback.
Proof: Let I= [g, K, a] E r f and let X= (A, b,c), Xt =(At' bt, C1)
= 1 · x so that A1 = g(A- ba- 1 K)g- 1 , b1 = gba- 1 and Ct = cg- 1 .
Then
and range Y(A~,b 1 ,c 1 ) =grange Y(A,b,c)a- 1 so that
rank Y(A~,b~,ct) =rank Y(A,b,c).
= [ ~ ~], b = [ ~],
n, u
Example 21.9. Let A c = [1 1 ]. Then
Y(A,b,c) = [~ Z(A,b,c) = ~] so that (A,b,c) is minimal.

Let 1 =[I, (1, 2), 1]. Then .
and Z( At, b1, Ct) = [~ ~] so that (At, b1, c1) is not observable. In
other words, observabili ty (a fortiori minimality) is not invariant under
state feedback.
Proposition 21.10. Let N f = {[I, K, 1] E r f} and let G f =

{[g,O,o:] E r 1}. Then (i) N 1 is a normal subgroup ojr 1; (ii) G 1 is
a subgroup of r 1 which acts on N 1 via inner automorphisms; and, (iii)
r 1 = N 1c 1 = c 1N 1.
Proof: Clearly Nf and G 1 are subgroups of r 1 . If 1 = [g, K, a:] E
rh then1 = [I,Kg- 1 ,1][g,O,o:] = [g,O,o:][I,o:- 1 K,1] so that r 1 =
N 1G 1 = G 1 N 1 . Since
r.-· ][I ,11.1,

[g,1~,o: r.-· 1][g - t ,-a -1 1.1.9 [I - t 1.1.19,
r.-· -1 ,a -1] =,a T." -1 1]
and
[g, 0, o:][I, Kt, 1][g- 1 , 0, a- 1 ] = [I, o:- 1 Ktg- 1 , 1],
N 1 is a normal subgroup on which G f acts via inner automorphisms.
Now let x = (A,b,c) and f(z) = fx(z) = c(zi- A)- 1 b. If

1 9 = [g,O,a] E Gt and x 9 = 1 9 • x, then fx.(z) = (cg- 1 )(zi•
gAg- 1 )- 1 gba- 1 = fx(z)a- 1 . It follows that if 1 E G f, then the poles
of f-y·x(z) and fx(z) are the same and are determined by the roots of
det(zi- A)= qA(z). On the other hand, if /K = [I, K, 1] E N 1 and
XJ( = /K ·x, then fxK(z) = c(zi -A+bi()- 1 b and the poles of fxK(z)
are determined by the roots of det(zi -A+bK) = qA-bK(z). Thus, to
analyze the effect of state feedback on the poles of fx(z) we need only
study the action of N 1 on the orbits under the action of G 1 . Put in
another way, if Pn(z) = {q(z) : q(z) is a monic polynomial of degree
n} and if we define a mapping .6.: r 1 X A~ 2 + 2 n---+ Pn(z) by setting
.6.(/,x) = det(zi- g(A- ba- 1 K)g- 1) (21.11)
where I= [g,K,a] E r,, X= (A,b,c) E A~ 2 + 2 n, then .6.(/,X) =

.6.(/e,/ · x) where le = [I,O, 1] so that
(21.12)
for all/1, /2 in r J· If Jg E G j, then ~(/g, x) = ~(/e, /gX) = det(zi•

A)= ~(/e,x) so that~ is constant on orbits under G1 (i.e.~ is
a Grinvariant). It follows from (21.12) that the range ~(r 1 ,x) is
the same as the range ~( N f, x) and the range ~(r 1 , 1 9 x) is the same
as the range ~(r f,x) for /g E Gf. Observing that Pn(z) can be
identified with Ak by the mapping ao + a 1z + · · ·+ an-lZn-l + zn ------+
( ao, a1, ... , an-1 ), we now have:
Problem 21.13 (The Coefficient Assignment Problem). Given

2 + 2 n, determine the range of the map ~x: Nf---* Ak given by
x E A~
~x(K) = ~(/K,x) = (XI(A- bK), ... ,Xn(A- bK)) (21.14)
where the xi(-) are the characteristic coefficients, and, in particular, de•
termine conditions for the surjectivity of ~x-
If k has characteristic zero, then ~x will be surjective if and only

if the map Tx: N1 ---* Ak given by
Tx(K) = (tr(A + bK), tr(A + bK) 2 , ••• , tr(A + bKt) (21.15)
is surjective in view of the results in Appendix C (e.g. proposition

C.14). The map Tx is called the trace-assignment map and we can
formulate the following:
Problem 21.16 (The Trace Assignment Problem). Given x E

2 +Zn,
A~ determine the range ofTx and, in particular, determine conditions
for the surjectivity of Tx.
vVe now turn our attention to solving these problems.

We first observe that since zi- A+ bJ( = (zi- A)[I + (zi•
A)- bK], det[zi- A+ bK] = det[zi- A] det[I + (zi- A)- 1 bK].
1
However,
det[I + (zi- A)- 1 bK] = 1 + K(zi- A)- 1 b (21.17)
by a well-known determinant identity. It follows that
~x(K) = ~x(O) + Kadj(zl- A)b (21.18)

and hence that the range of ~x(K) is determined by the range of the
linear map lx(K) : Ai: ----+ Ai: given by
lx(K) = Kadj(zi- A)b (21.19)
where a polynomial of degree n- 1 is identified with its coefficients.
Lemma 21.20. lx(K) = K Jx where

Jx = [b Ab + X1(A)b · · · An- 1b + XI(A)An- 2 b + · ·· + Xn-l(A)b]
(21.21)
and the Xi are the characteristic coefficients.
Proof: An immediate consequence of the formula adj(zJ- A) =

L~=1(zn-i + X1(A)zn-i-1 + ... + Xn-i(A))Ai-1.
Corollary 21.22. lx(K) = KY(A,b,c)Lx where
1 X1 X2
··· Xn-1]
0 1 X1 ... Xn-2
L X-- [
. . • (21.23)
0 0 0
0 0 0
0 0 0 1
and so, dim R(lx) = rankY(A, b, c).

Theorem 21.24 (Coefficient Assignment Theorem) ([W-2]). ~x is
surjective if and only if x is controllable.
Proof: ~x is surjective if and only if lx is surjective and lx is

surjective if and only if dim R(lx) = rankY(A,b,c) is n.
We also note that corollary 21.22 characterizes the range of ~x·

Now let us examine the trace assignment problem. We begin with
a lemma.
Lemma 21.25. (i) tr(A + bi()m = L:T=o ( j) tr(Am-i(bK)i);

(ii) (bK)J = (bK)[tr(bK)ji- 1 for j = 1, ... ; (iii) tr(Am-i(bK)i) =
tr(Am-ibK)[tr(bK)]i- 1 ; and, (iv) tr(bK)i = [tr(bK)]i.
Proof: (i) follows from the linearity of the trace and the fact that
tr(XY) = tr(Y X). (ii) follows by induction and the easy fact that
(bK) 2 = (bK) · tr(bK). (iii) follows from (ii) and the linearity of the
trace. (iv) is also an immediate result of (ii).
We let tx( K) be the linear map given by
tx(K) = (tr(bK), tr(AbK), ... , tr(An- 1 bK)) (21.26)
and let t~(K) = tr(Ai- 1 bK), i = 1, ... , n be the components of tx. In

view of lemma 21.25, we have
(21.27)
where the n X n matrix Nx(K) is given by
[H
0 (t!Y
Nx(K) = ~] + 0 0 (3t!,)
0 0 n 0 0
(21.28)
i.e. Nx(K) is the sum of a constant diagonal matrix and a matrix
which depends only on t~( K) and has zero entries on and below the
main diagonal.
Lemma 21.29. Suppose that the characteristic of k is zero or is

greater than n. Then Tx is surjective if and only if tx is surjective.
Proof: Clearly Tx is surjective if and only if txNx is surjective.

If tx is surjective and ( O!t, .•. , an) E Ak, then there is a ]( such
that tx(K) = (at, ... ,an)(Nx(K))- 1 as Nx(K) depends on a1 alone.
Conversely, if txNx is surjective and (!h, ... ,f3n) E Ak, then there is
a]( such that tx(K)Nx(K) = (f3t. ... ,f3n)Nx(K) as Nx(K) involves
f3t alone.
Theorem 21.30 (Trace Assignment Theorem). Suppose that the char•

acteristic of k is zero or is greater than n. Then Tx is surjective if and
only if x = (A,b,c) is controllable.
Proof: In view of the previous lema, it is enough to show that

ix is surjective if and only if x is controllable. Since ix is a linear
map, tx is surjective if and only if J(tx(K); K) is nonsingular. But
J(tx(K);K) = =
[b Ab· · ·An-lb] Y(A,b,c) by lemma 21.31 and the
result follows.
Lemma 21.31. Let X E M(n,n;k). Then J(Tr(XbK);K) = Xb.

Proof: Simply note that tr(XbK) = (X 1b)K1 + · · · + (Xnb)Kn
where Xi is the i-th row of X.
Example 21.32. Let A = [~ ~] , b = [ n, = [

c 1
that (A, b, c) is minimal. Suppose that k has characteristic 2. Then
0] so
tr(A + bK) = Kz and tr(A + b1() 2 = (K1 + K1) + Ki = Ki so that

Tx is not surjective.
Example21.33. LetA=[~~ ~],b= m,c=[l 0 OJ
so that (A,b,c) is minimal. Then tr(A + bK) = K 3, tr(A + b1() 2 =

(Kz+Kz)+Kj, and tr(A+bK) 3 = (K1 +K1 +K1)+(KzK3+K2K3+
K 2 K3) + ](~. Thus Tx is not surjective if k has characteristic 2 or 3.
The theorems 2.14 and 21.30 provide the solution to the so-called
"pole-placement problem".
We have thus established the four major algebra-geometric results

of scalar linear system theory, namely: (i) that the Laurent map L is
a k-isomorphism between Rat(n, k) and Hank(n, k); (ii) that minimal
realizations exist and that any two minimal realizations are uniquely
isomorphic; (iii) that (Char(n,k),9lx) is a geometric quotient of Sf, 1
modulo GL(n,k) with Char(n,k) a nonsingular quasi-affine variety of
dimension 2n; and, (iv) that the "poles" are assignable under state
feedback if and only if the system is controllable.
In order to develop a similar theory for multivariable linear sys•
tems, we shall require ideas from projective algebraic geometry and
ultimately, an even more general approach. With this in mind, we
examine "abstract affine varieties" in the next section.
22. Affine Algebraic Geometry: Varieties
Let R be an affine k-algebra and let X = Spm(R) = { m : m

is a maximal ideal in R} be the maximal spectrum of R. We recall
(theorem 7.5, corollary, 7.6) that X = Spm(R) ~ Hom k(R, k) and
that R may be viewed as the coordinate ring of some affine algebraic
set. What we wish to do here, is to show that X may be made into
a topological space, that the "regular" functions on X are simply
the elements of R, and that if U is open in X, then the "regular"
functions on U can be defined in a manner totally analogous to that
used in section 9 (definition 9.8).
Proposition 22.1. Let R, S be affine k-algebras and let 7/J be an

element of Hom k( R, S). If n E Spm( S), then 7/J- 1 ( n) E Spm( R).
Proof: Let an E Homk(S,k) with Kewn = n. Then an o'ljJ E

Hom k( R, k) and Ker an o 7/J E Spm( R). Thus it will be enough to
show that 7/J- 1 (n) = Ker(an o 7/J). If r E 7/J- 1 (n), then 7/J(r) E n
and r E Ker(an o 7/J). If r E Ker(an o 7/J), then an(7/J(r)) = 0 and
1/J(r) E Keran = n.
Definition 22.2. If E is a subset of R and (E) = a is the ideal

generated byE, then V(E) ={mE X: m :J E} is the closure of E.
If M is a subset of X= Spm(R), then I(M) = {r E R: rEm for all
m in M} is the ideal of M (I( M) is clearly an ideal).
Proposition 22.3. (i) V(E) = V(a) = V(fo) where a= (E); (ii)

V(O) = X, V(R) = ¢; (iii) V(UE;) = nV(E;); and, (iv) V( an b) =
V(ab) = V(a) U V(b)
Proof: Clearly, V(E) = V(a) 2 V(Ja). If m :J a, then m :J Ja
as r E a C m implies f E m. Thus, V( a) = V( Ja). Parts (ii) and
(iii) are obvious. If m E V (a) UV (b), then m E V( an b) 2 11 ( ab). On
the other hand, if m :J ab but m /J a and m /J b, then there are f E a,
145

f ¢ m and g E b, g ¢ m. But fg E ab C m gives a contradiction.
It is clear from the proposition that the sets V( a), if viewed as

the closed sets, define a topology on X = Spm( R). This topology
is called the Zariski topology. In view of the proposition, we have
I(V( a)) = I(V( y'(i)). Since y'(i = P1 n ... n Pr where the Pi are
prime ideals (corollary 6.11), we can see that I(V(a)) = y'(i and that
V(I(M)) = M are immediate consequences of the following:
Lemma 22.4. If .):l is a prime ideal in the affine k-algebra R, then

p = nm, mE Spm(R),P c m.
Proof: By passage to Rj.)J, we may suppose .):l = (0). Since, in

that case, .):l = (0) C m for all m, we must show that if r ::f 0, then
there is an m with r ¢ m or, equivalently, there is an a E Homk(R,k)
with a(r) ::f 0. If ]( = K(R) is the quotient field of R and S =
R[l/r] is an affine k-algebra (not the zero ring), then there is, by the
Nullstellensatz, an a 8 E Homk(S,k). But 1 = a 8 (1) = a 8 (r)as(ljr)
and so, if a = a 8 restricted to R, then a( r) ::f 0.
Since X= Spm(R) is a topological space with the Zariski topol•

ogy, we can seek to determine the irreducible subsets ( cf. section 6)
and to find a suitable base for the topology.
Definition 22.5. Let XI= X- V(f) = {m: f ¢ m}. Then XI is

a principal open subset of X (since V (f) is closed, X I is clearly open).
Proposition 22.6. The X I form a base for the Zariski topology on

X and (i) XI n X 9 = XI 9 so that Xr =XI; (ii) XI=</> if and only if
f = 0; (iii) XI= X if and only iff is a unit in R; (iv) XI= X 9 if and
only if y'(]) = .j[Yj; and, (v) X is quasi-compact as is each XI.
Proof: If V(a) is closed with a= y'(i ::f R, then, form¢ V(a),

there is an f with f ¢ m (take f E a) and so, m E XI. Thus, the
XI form a base. Since V(fg) = V(J) U V(g), (i) is clear. (ii) is an
obvious consequence of the proof of lemma 22.4 and (iii) is a result of
the fact that every proper ideal is contained in a maximal ideal. As for
(iv), observe that V(J) = V(.J(})) = V(g) = V(J(g)) if and only if
.J(}) = .j[Yj. To establish (v), note that X= UXJ; implies (ii) = R
AFFINE ALGEBRAIC GEOMETRY: VARIETIES 147
We recall that a topological space is Noetherian if any descending

sequence of closed sets eventually stops i.e. if Vi :J V2 :J · · · then
Vn = Vn 0 for n ~ no.
Proposition 22.7. (i) { m} is closed; (ii) X = Spm( R) is a Noethe•

rian topological space; and, (iii) V(a) is irreducible if and only if a is
prime and so, X is irreducible if and only if R is an integral domain.
Proof: Clearly, I(V ( m)) = m so that { m} is closed. If V1 =

V( a 1 ) :J V2 = V( a 2 ) :J · · · is a descending sequence of closed sets,
then a1 C a2 C · · · is an ascending chain of ideals in R. Since R is
Noetherian, an = ano for n ~ no and some no so that V (an) = V ( ano).
As for (iii), if V(a) is irreducible and fg E a, then V(a) C V(f)UV(g)
implies (say) V(a) C V(f) and f Ea. On the other hand if pis prime
and V(p) = V( a 1) U V( a2), then p = a1 n a2 implies p = a1 (say) and
V ( p) is irreducible.
Corollary 22.8. The (irreducible) components of X are the closed

sets V(Pi) where (0) = npi and the Pi are minimal prime ideals in R.
Now suppose that R and S are affine k-algebras and that X =

Spm(R), Y = Spm(S). If 1/J E Hom k(R, S), then 1/J induces a map
'¢ : Y --+ X given by
where n E Spm(S), in view of proposition 22.1.
Proposition 22.9. (i) ;j;- 1 (X 1 ) = Y1/!(f); (ii) ;j;- 1 (V( a)) =

V(,P(a)) where a is an ~deal in R; (iii) ~(V(b)) = V(,P- 1 (b)) where
b is an ideal in S; (iv) 1/;(Y) is dense in X if and only if 1/J is injective
(cf proposition 7.18); and (v) if'lj; is surjective, then Y ~ V(ker'lj;) (cf.
proposition 7.18).
Proof: Properties (iv) and (v) have been established in section 7.

If n E Y- V(,P(f)), then 1/;(f) (/nand f (/ ,p- 1 (n). On the other
hand, if n E ;j;- 1 (X- V(f)), then y(f) (/ n, and so, Q) is established.
As for (ii ), noting that 1/;(f) = f o 1/J, we see that n E ,p- 1 (V( a)) if and
only if '¢(n) E V(a) if and only if 'lj)- 1 (n) :J a if and only if n :J 'IJ)(a).
To demonstrate iii , we need only show that 'lj)- 1 (b) = I('¢(V(b)) as
'lj)- 1 (Vb) = 'lj)- 1 (b). But g E J('¢(V(b)) implies (go '¢)(n) = 0 for
n E V(b). In other words, 'IJ)(g)(n) = 0 and hence, 'IJ)(g) E J(V(b)) = b
or g E 'lj)- 1 (b).
Corollary 22.10. '¢is continuous.
Corollary 22.11. If X = Spm(R) is irreducible and f E R and

'lj) : R ---+ R 1 is the canonical injection, then '¢- 1 (X1 ) = Spm(R 1 ) and
there is a natural bijective homeomorphism of X 1 onto Spm( R f) (where
X 1 has the induced Zariski topology). In other words, X 1 = Spm(R 1 ).
Proof: Since '¢- 1 (X 1 ) = Y.p(f) = Y1 and f is a unit in R 1 , we have

Yf = Y = Spm( R 1 ). The rest of the corollary has been established in
propositions 8.1 and 9.14.
If m is an element of X = Spm(R), then Rm is a local ring and

we have:
Proposition 22.12. R = nRm where mE X= Spm(R).
Proof: Clearly, R C nRm. If h E nRm and ah = {g E R: gh E

R}, then ah is an ideal in R. If ah is a proper ideal, then (by the
Nullstellensatz) V(ah) -:f ¢and there is an mE X with ah C m. But
h E Rm implies h = hi/h2 with h2 rf. m. However, h2h = h1 E R so
that h2 E ah C m. This is a contradiction and so ah = R. In other
words, 1 E ah and h E R.
Corollary 22.13. If X is irreducible, then R 1 = nRm where m E

X 1 = Spm(R 1 ).
Proof: Simply apply the proposition and corollary 22.11.
Now let us write x E X in place of m E X. We then have:
Definition 22.14. A function f : X ---+ k is regular at x if f E Rx.

The set of regular functions at x is simply the local ring Rx which we
denote by Ox,X.
AFFINE ALGEBRAIC GEOMETRY: VARIETIES 149
Proposition 22.15. Iff is regular at x, then there are g, hE Rand a

neighhorlwod V ofx such that h(y) -:j: Ofor y E V and f(y) = g(y)lh(y)
for y E V. Conversely, if X is irreducible and there are g, h E R and a
neighhorlwod V of x such that h -:j: 0 on V and f = g I h on V, then f is
regular at x.
Proof: For the first part, simply take V = Xh. For the second
part, simply note that hf = g on V which is dense in X so that hf = g
on X and J = g I h E Rx.
Definition 22.16. If U is open in X and f : X -+ k, then f is

regular on U if f is regular for each x in X. If X is irreducible, then
all Rx C K(R), the quotient field of R, and f is regular on U if and
only iff E nxEURx. We let Ox(U) = nxEURx and call Ox(U) the
k-algebra of regular functions on U.
Proposition 22.17. Suppose that X is irreducible. Then (i) Ox(X1)=

R1 and Ox(X) = R; (ii) if V C U are open sets and pu,v : Ox(U)-+
Ox(V) is the restriction map i.e. Pu,v(J)(m) = f(m)for mE V and
f E Ox(U), then Pu,v is an element of Hom k(Ox(U), Ox(V)), pu,u =
identity, and if W C V C U, then Pu,v o Pv,w = Pu,w; (iii) if U =
UU;, !I,h E Ox(U) and Pu,u;(Jt) = Pu,u;(h) for all i, then h =
h; and, (iv) if U = UUa and fa E Ox(Ua) with PUa,Uanup(Ja) =
PUp,UpnUa (!13) all a, (3, then there is a (unique by (iii)) fin Ox(U) such
that PU,Ua (f) = fa·
Proof: (i) has been established in corollary 22.13, and (ii) is ob•
vious. As for (iii), we may suppose that pu,u1 (!1) = pu,u1 (h) for all
f. If h -:j: h, then g = h - h E Ox(U) and there is an x E U9
(as X 9 n U -:j: ¢). But PU,Ug(!J) =PU,Ug(h) implies !J(x) =h(x),
which is a contradiction. To prove (iv), we may assume that the Ua
are principal open sets and so reduce to the following: let Ufi cover
U and lets; E Ou(UJ;) with s; = Sj on U1; n U1;, then there is an s
in Ox(U) with Pu,u1 (s) = s;. Since X is quasi-compact, we actually
j
have u = Uf:l uf; and S; = r;f ft' i = 1, ... 'n. Since S; = Sj on

Uf; n Uh = Uf;f;, we have (r;Jj)f(J;fi)m = (rifr)f(f;fi)m in Uf;f;
and, therefore, (r;Jj- rJ;m)(J;fj)m;; = 0 in Ox(U). Since U is irre•
ducible, r;Jj =rift. Moreover, Uf; = U!;"' and so, (fr, ... ,j:::.") =
Ox(U). In other words, 1 = '£~= 1 9d;m· Lets= '£~= 1 gm. Then
s E Ox(U) and sfj = 'Egir;Jj = 'Egdtrj = rj. Thus, PU.UJ; (s) =

r3 j fj = s 3 and we are done.
Now, what we have is a topological space X and, for each non•

empty open subset U of X, a k-algebra Ox(U), and, a family of
k-homomorphisms pu,v: Ox(U)- Ox(V) for U :J V, such that (ii),
(iii), (iv) of proposition 22.17 are satisfied. The family 0 of Ox(U)
and maps Pu,v is called a sheaf (of functions) and the pair (X, 0) is
called a ringed space. We then have:
Definition 22.18. If R is an irreducible affine k-algebra, then

the ringed space (X, Ox) where X = Spm(R) and Ox is the sheaf
{Ox(U),Pu,v} is an affine algebraic variety over k.
If (X, 0 x) and (Y, Oy) are ringed spaces and 1/J : X - Y, then
we say that 1/J is a morphism of ringed spaces if (i) 1/J is continuous,
and (ii) for all open sets V in Y, f E Oy(V) implies 1/J*{f) = f o 1/J E
0 x( 1/J- 1 (V) ). Thus, in this context, a morphism of affine algebraic
varieties is simply a morphism of ringed spaces. We leave it to the
reader to verify the consistency of these ideas with what we have done
previously. This approach will turn out to be most fundamental.
23. Interlude
We seek, in this section to indicate, in a brief and heuristic fash•

ion, some of the flavor of Part II: Multivariable Linear Systems and
Projective Algebraic Geometry. While affine algebraic geometry is
quite satisfactory and natural for scalar systems, the study of multi•
input, multi-output systems requires projective geometric ideas. For
simplicity, we shall assume that our algebraically closed field k is the
complex numbers C throughout this section.
The classical mathematical reasons for introducing projective
space are to complete intersections and to compactify. Suppose we
consider the family of lines y = mx +bin A~. Given two distinct lines
y = m1 x + b1, y = m2 x + b2, these two lines will intersect in a point if
m1 :f. m2, but will not intersect if m1 = m2. In some sense, the lines
y = mx + bt, y = mx + b2 meet at "infinity". In other words, it would
be pleasing to assert that any two distinct lines meet in a point and
not have to deal with the exceptional case of parallel lines. Similarly,
if we consider a curve of degree 2, say: y 2 = x 2 -- 1, then lines of
the family y = mx will meet the curve in two points unless m = ± 1;
in that case, there is no intersection but (ylx) 2 ___., 1 as x ___., oo (or
-oo) and so, the curve has the lines as asymptotes. Again, in some
sense, the curve meets the asymptotic line at "infinity" and it would
be desirable to avoid the exceptional case if possible. We can do this
by extending the affine plane A~ to the projective plane P~ by adding
"points at infinity".
More precisely, consider the set A~ - {(0, 0, 0)} and define an
equivalence relation on it via multiplication by elements of C* i.e.
x = (xo,x1,x2) ,...., y = (Yo,YI,Y2) if y; = ax;,a E C*. The set
of equivalence classes is the projective plane P~ and is the same as
the set of all !-dimensional subspaces of A~ (i.e. the set of all lines
through the origin in A~). Elements of P~ are called points and any
representative (~o,6,6) of a point {is called a set of homogeneous
coordinates off Note that (~o,6, 6) :f. (0, 0, 0) and that if, say 6 :f. 0,
then ( ~o I 6, 6 I 6, 1) is also a set of homogeneous coordinates of { and
every point ( x, y) in A~ defines exactly one point with homogeneous
151
coordinates ( x, y, 1) in P~. The points (~o, 6, 0) (note either ~0 f= 0 or

6 f= 0) are called points at infinity and H 00 = {(~o,6,0)} (i.e. 6 = 0)
is called the line at infinity. A line in P~ may be defined as the set of
points which satisfy a homogeneous equation of the form
(23.1)
with (a 0 ,at,az) f= (0,0,0). Observe that if one set of homogeneous

e
coordinates of a point satisfies (23.1), then all sets do. The lines
y = mx + b in A~ correspond to the lines mxo - Xt + bxz = 0 in P~
and then, the two distinct (parallel) lines y = mx + bt, y = mx + bz
correspond to the lines mxo - Xt + b1 xz = 0, mxo - x1 + bzxz = 0
which intersect in the point (1, m, 0) at infinity! Similarly, the curve
y 2 = x 2 - 1 corresponds to the curve xi = xfi - x~ (homogeneous of
degree 2) and meets the line xo - x 1 = 0 (corresponds to y = x) in
the point (1, 1, 0) and the line x 0 + x 1 = 0 (corresponds toy = -x) in
the point (1, -1, 0). We remark that the line x 0 - x 1 = 0 is tangent
to the curve xfi - xi - x~ = 0 at (1, 1, 0) and that the curve and line
meet "twice" at this point. In other words, we are able to include all
intersections by working in projective space.
Now consider P~ as the set of !-dimensional subspaces of A~ =
C 2 • We claim that Pi: is a one-point compactification of C. A simple
classical way to see this is to identify P~ with the 2-sphere S 2 =
R2 U { oo} which is the one-point compactification of C. If we identify
S 2 with the set {(u,v,w) : u 2 + v2 + w 2 = 1} in R3 and if we let
P~ = (Pt - {€oo}) U {~oo} where {€ 00 } = (1,0) then Pi:- {€ 00 } =
{(€, 1): € E C} ~ C and the map t/;~"" :Pi: -+ S 2 given by
tP~oo(~oo) = (0,0,1)
'lj; ((€ l)) = ( 2r cos (J 2r sin (J r 2 - 1)
~"" ' r2 +1 ' r2 +1 ' r 2 +1
2a 2b a2 + b2 - 1
= ( a2 + b2 + 1' a2 + b2 + 1' a 2 + b2 + 1) ( 23 ·2)
where € = re; 9 = r cos (J + ir sin (J = a+ ib E C, identifies Pi: and S 2 •

An alternative way with, as we shall see, more relevance to control
involves subspaces. Let € be a point ofP~ and let V~ be € viewed as a!•
dimensional subspace of C 2 • Suppose W is any !-dimensional subspace
INTERLUDE 153
of C 2 with V~ EB W = C 2 and let a(W) = {V : V a !-dimensional

subspace of C2 and dime V n W ~ 1}. Note that here a(W) = {W}
is a single point. If V11 is any element of P~, then either V11 EB W = C 2
or V11 E a(W) so that V11 = W. In case V11 EB W = C 2 , there is a linear
map L : V~ ---. W with V11 as graph. Since the dimension is 1, L is
multiplication by an element of C. Thus, P~ - {0 is identified with
C and P~ = (P~- {0) U {0 = {P~- a(W)} U {a(W)} = C U {oo}
= one-point compactificati on of C.
Definition 23.3. Let cN+l = {(xo, ... ,xN) : X; E C} and call

x,y in cN+l- {0} equivalent if X= >..y,>.. E C*(= c - {0}). The set
of equivalence classes P~ is called projective N -space.
Since the equivalence identifies points on the same line through

the origin, P~ is the same as the set of !-dimensional subspaces
of A~+l. A point ~ E P~ can be represented by an N + 1-tuple
(~0 , .•. , ~N) -::f 0 and another N + 1-tuple (~~' ...• ~~)defines the same
point if and only if there is a >.. in C* such that ~j = >..~j ,j = 0, ... , N.
Any N + 1-tuple representing~ is called a set of homogeneous coordi•
nates of~-
Definition 23.4. A point ~ E P~ is a zero of a polynomial

F(~o, ... ,~N) = 0 for all sets of
F(Xo, ... ,XN) in C[Xo, ... ,XN] if
homogeneous coordinates of ~.
We observe that ifF = F0 + F1 + · · · + Fr, F; a homogeneous poly•

nomial of degree i, then F(>..~o, .. o,>..~N) Fo(~o, ... ,~N)+
>..Ft(~o,0 ,~N) + · · + >..r Fr(>..o, ... , AN) and so, F(>..~o, ... , >..~N) = 0
•• o
for all >..in C* if and only if F;(~o, 00. ,~N) = 0, i = 0, 0. 0, r.
Definition 23.5. V C P~ is a projective algebraic set if V is

the set of common zeros of a family of homogeneous elements of
C[Xo,•oo ,XN]o
Viewing the projective algebraic sets as the closed sets, we define

a topology, the Zariski topology, on P~. We show, of course, that this
is a topology in Part II.
Definition 23.6. An irreducible ( cf. Definition 6.1) projective

algebraic set in P~ is a projective variety and an open subset of a

variety is a quasi-projective variety.
We shall soon construct some examples with a control motiva•

tion. First, suppose that f(z) = bo + ··· + bn-1Zn- 1fao + a1z + · · · +
an-1Zn- 1 + zn = p(z)fq(z) with p,q relatively prime. Then we can
define a map 'lj;f: C--+ C XC= C 2 via
'l/Jt(z) = (p(z),q(z)) (23.7)
If we let ~oo = (0, 1) and C = {(1,~) : ~ E C} then we can "extend"

'lj;f to a map, also denoted 'lj;f of P~ --+ P~ via
'l/JJ((xo, X1 )) = (boXo +·· ·+bn-lXQX~-1, aoXo +•· ·+an-1XoX~- 1 +xr)

(23.8)
Noting that 'l/Jt(~oo) = ~oo and that 'l/Jt((1,~)) = (p(~), q(~)), we see
that 'lj;f does indeed map P~ into P~ as p, q are coprime. We also
observe that the "poles" of f(z) are the "points" (1,~) such that
-¢ 1 ((1,~)) = (1,0). Now let us consider a system with m = 2 in•
puts and p = 2 outputs. Such a system can be represented by a
transfer matrix F(z) which is 2 X 2. If each entry Fij(z) of F(z) is a
strictly proper rational function, then F( z) is a strictly proper rational
transfer matrix. In such a case, we can write
F(z) = P(z)Q- 1 (z) = P(z)AdjQ( z)/ detQ(z) (23.9)
where P(z),Q(z) are 2 x 2 matrices with polynomial entries and are

"coprime" in the sense that there are polynomial matrices X(z), Y(z)
such that
X(z)P(z) + Y(z)Q(z) h = (23.10)
(this generalizes the relation a( z)p(z)+b( z)q(z) = 1 of the scalar case).

We can then define a map '¢F: C--+ M(4,2) of C into the set of 4 X 2
matrices (with entries in C) by
(23.11)
and coprimeness means that the 4x2 matrix '¢F(z) has rank 2for all z
i.e. that the columns of '¢F(z) define a 2-dimensional subspace of C 4 .
INTERLUDE 155
Let us denote the set of all 2-dimensional subspaces of C 4 by Gr(2, 4)

(Gr for Grassmanman). By analogy, Gr(1,4) = P~. If we fix a basis in
C\ then every element of Gr(2,4) determines an element of M*(4,2),
the set of 4 x 2 matrices of rank 2, and conversely. Let vo, v~, Vz, V3
denote the rows of an element of M*(4,2) (the Vi are elements of C 2 )
and let [v, w] denote the determinant of the matrix [ ~]. Consider
the map P : M*( 4, 2) __. P~ given by
so that po = [vo, v1],p1 = [vo, vz], ... ,ps = [vz, v3]. Since the matrices
are of rank 2, at least one Pi =j:. 0. Let G = GL(2, C). Then G acts
on M*(4,2) by multiplication on the right and defines an equivalence
relation on M*( 4, 2) by
(23.13)
fori= 0, ... ,3 and some g E G. Clearly, M*(4,2)/G "=" Gr(2,4)

and Pis an invariant for the equivalence. Thus, P(M*(4, 2)) or better
P(M*(4,2)/G) may be identified with Gr(2,4). Now consider the
homogeneous polynomial equation
PoPs + P2P3 - P4Pl =0 (23.14)

which defines a projective variety in P~ which is of dimension 4. We
claim that Gr(2,4) is this variety. Since [vo, v1][vz, v:l]+[vo, v3][v1, vz]•
[v1,v3][v0,vz] = 0 holds for any 2-vectors vo,vl,vz,V3 as a determi•
nantal identity, Gr(2, 4) is contained in the variety. In fact, they are
equal. For example, suppose that { = (1,6, ... ,6) satisfies (23.14),
then vo = (1,0),vl = (0 1),vz = (-6 6),v3 = (-~4 6) gives an
element of M*( 4, 2) with image{. We can thus consider the map 1/JF
(abuse of notation) as a map 1/JF: P~ __. Gr(2,4) C P~.
We can also see that it is reasonable to view (Gr(2,4), P) as a
"geometric quotient" (in a projective sense) of M*( 4, 2) modulo the
action of G. If M is an element of M,.(4, 2) and Oa(M) is the orbit

under G, then dimOa(M) = dimG = 4 (since Sa(M) = {I}) and
so orbits are closed (cf. lemma 15.9). If we let G(po,Pt, ... ,ps) =
PoPs + PzPJ - P4Pt. then G is a form of degree 2 and Gr(2, 4) = { ~ =
(~o, ...• ~s) : G(~) = 0}. Since 8Gf8pi are not all 0 at any point
of Gr(2,4), the variety Gr(2,4) is non-singular (cf. section 20.) and
hence, normal (shown in Part II). Thus the map Pis an open map
( cf. proposition 18.22). As for making the ring of invariants part
of a geometric quotient plausible, we deal locally with the open set
Gr(2,4)o = {~ E Gr(2,4): ~o =f 0}. This set can be identified with the
hypersurface Xs + XzXJ- x4x1 = 0 in A~ (let ~o = 1 and Xi = ~i/~o)
which has the coordinate ring C[x1, ... , xs]/(xs + XzXJ- x4x1). It is
easy to see that p- 1(Gr(2,4) 0 ) "=" G XV where
V={MEM.(4,2):M= [f]}
and that there is an action of GonG xV given by g(g1, M) = (gg1, M)
(cf. (19.49)). Moreover, C[G x V] 0 = C[V] and so, as in section 19.,
we want P*(C[Gr(2,4)0 ]) "=" C[V]. But if,
is in V, then P(M) = (1,v22,v32, -v21, -v31,v21v32- v22v31) and so,

letting Xt = Vz2,X2 = V32,X3 = -v21,X4 = -V3t,X5 = V21V32- V22V31,
we see that our assertion about the ring of invariants is plausible. *
Example 23.15. Let
P(z) = [Pn P12] ,Q(z) = [zz ++ qu q12 ]

Pzl Pz2 qzt z + q22
(with det P =f 0). Set vo(xo, x1) = (pnxo p12xo), Vt(Xo, xt)
(p21xo pzzxo),vz(xo,xt) = (x1 + qnxo q12xo),v3(xo,x1) = (x1 +
q21xo x1 + qzzxo). Then
1/JF(xo,Xt) = ([vo(xo, Xt), Vt(Xo, xt)], ... , [vz(xo,Xt), v3(xo, x1)])
* Of course, these issues are dealt with in Part II.
INTERLUDE 157
and so, for example, Po(xo,xl) = x6detP,p3(xo,xi) = x6(P2lql2-

qupn )-P22xox1 ,ps ( xo, x1) = xi +(qu -q12 )xoxl +x6( qu q22 -q12q21 ).
Note that 1/JF(O, 1) = (0, ... , 0, 1) and that the "poles" of 1/JF are the
points where the "curve" 1/JF(PU meets the hyperplane Ps = 0 !
Now suppose that Z 01 is an element of Gr(2,4) and let u(Z01 ) =

{V E Gr(2,4) : dimc(V n Z 01 ) ~ 1} (note that either V EB Z 01 =
C 4 or V E u(Z 01 )). Let ~ 01 = (po(Z 01 ), ... ,ps(Z01 )) and consider the
homogeneous equation of degree 1
Po(ZOI)P5 -p1(Z01)p4 +pz(ZOI)P3+P3(ZOI)Pz -p4(Z01)P1 +Ps(ZOI)Po = 0

(23.16)
in P'~. Let H 01 be the projective variety defined by (23.16). Then
a(ZOI) =HOI n Gr(2,4) for if ZOI corresponds to VOI,O,VOI,l,v01,2,v01,3
and V corresponds to vo, VI, v2, V3, then V E u( Z 01 ) if and only if
(note this is a 4 X 4 matrix and expansion into determinants of 2 X 2

matrices shows that (23.16) holds). The variety u(Z 01 ) is called a
Schubert variety.
Now suppose that we have a system represented by a 2 x 2 transfer
matrix F(z) with associated map 1/JF : P~ -+ Gr(2, 4) given by
(23.17)
suitably "homogenized". The roots of det Q(z) are the "poles" ofF
and we wish to examine the effects of output feedback on the "poles".
In other words, given ]( E M(2, 2), what happens when Q is shifted
to Q- ]( P i.e. what are the roots of det(Q- J( P)? We observe that
and hence that
det(Q- J( P) = det ( [ J,- ~l)

Thus, if (1, 0 is to be a "closed loop pole" after "output feedback with

gain K", then det( Q ( ~) - K P( ~)) = 0 and the 2-dimensional subspace
in C 4 satisfies the condition dim(WI< n Z<) > 1 where Z~ is the 2-

dimensional subspace
z~ =span [Q(O
P(O]
i.e. WK E a( Z€ ). If we are given n = degree ( det Q) points 6, ... , ~n,

then the output feedback problem becomes: does there exist a K such
that WI< E nf= 1 a( Z~;) or, in view of (23.16), does there exist a K
such that WI< E Gr(2, 4) n nf= 1 H€;? If say, n ~ 4, then "in general",
dim nf= 1 H ~; 2: 4 - n and dim Gr(2, 4) n nf= 1 H ~; 2: 0 so that such a
WK exists. Observe that if ~oo = (0, 1), then
and so, WK ~ a(Zoo) for all K. Thus, a "non-degeneracy" condition

e.g. nf= 1 a( Z~;) n a( Zoo)) = 0 is required. This is a very brief in•
dication by simple example of some of the relevance of intersections,
Grassmannians and Schubert varieties to a control problem.*
Grassmannians and other projective ideas also impact multivari•
able control systems in the state-space and Hankel matrix represen•
tations. So let us again consider a system with m = 2 inputs, p = 2
outputs, and of "degree" n = 2. This system is described by a triple
(A,B,C) where A is n X n,B is n X m and Cis p X n. (We are,
for simplicity, assuming all are 2 X 2.) The transfer matrix associated
with (A,B,C) is given by
F(z) = C(zl- A)- 1 B (23.18)
* For much more detail, see the survey article: "Pole Assignment
by Output Feedback" by C. I. Byrnes.
INTERLUDE 159
and is, of course, rational. We may define controllability, observability,

and Hankel maps as follows:
l
Y(A,B,C) = [B AB···An-IB] (23.19)
Z(A,B,C) = [ ~ (23.20)
CAn-I
H(A,B,C) = Z(A,B,C)Y(A,B,C) (23.21)
so that Y is n x nm, Z is np x n, and H is np X nm. So, in our sample

case, Y is 2 X 4, Z is 4 X 2 and H is 4 X 4. IT B = (bi b2] and if we let
v0 = bt, VI = b2 , v2 = Abi, v3 = Ab2, and if we say that (A, B, C) is
controllable if rank(vo VI v2 v3] = rank[B AB] = rankY(A,B,C) = 2,
then (A, B, C) is controllable means that Y( A, B, C) is an element of
M,.(2,4). We can embed M,.(2,4) in Gr(2,4) C P~ via the map (again
denoted 7')
P([vo, VI, v2, v3]) = ([ vo vi], [Vo v2], [vo v3], [vi v3], [v2 v3]) (23.22)
and we have a natural map 'Py from the set of controllable sys•
tems into Gr(2,4) given by Py = 1' o Y. Similarly, there is a map
'Pz from the set of observable systems into Gr(2,4) given by 1' oZ.
Viewing the systems (A, B, C) as elements of A~2 , we can see that
(A~2 )c = {(A,B,C): (A,B,C) is controllable} is an open set (hence,
a quasi-affine variety) since (A~2 )c = uf= 0 (A~2 )p; wherepi(A,B,C) =
Pi(A,B) with Po(A,B) = det(bi b2],PI(A,B) = det[bi Abi],p2(A,B) =
det[bi Ab2],p3(A,B) = det[b2 Abi],P4(A,B) = det[b2 Ab2] (note that
Ps(A, B) = det[Abi Ab2] = det A· Po(A, B)). The map Y is a "mor•
phism" of the quasi-projective (no mistake as every affine or quasi•
affine variety is quasi-projective- Part II) variety (A~2 )c into the pro•
jective variety Gr(2, 4). In fact, Y((A~2 )c) ~ Uf= 0 Gr(2, 4)p; and does
not contain the point (0, 0, ... , 0, 1) which corresponds to the subspace
[00 I] (in fact, no point of the form (0,6, ... ,~4, 1)). Of course, things
are similar for observable systems.
Calling a system minimal if it is both controllable and observable
and letting Si, 2 (= s;,m) be the set of minimal systems, we have a
map 'Py,z: Si, 2 -+ Gr(2,4) X Gr(2,4) C P~ X P~ given by
'Py,z(A,B,C) = ('Pz(A,B,C),'Py(A,B,C)) (23.23)

(the projective notion of a product is somewhat more complicated

than its affine counterpart and will be treated in Part II). We note
that dimS?, 2 = 12 and that dimGr(2,4 )xGr(2,4) = 4 + 4 = 8. Let
G = G L( C, n) = G L( C, 2) be the group of "coordinate changes" in
state-space. Then G acts on Si 2 via the map 1 : G x Si 2 -+ S~ 2
~wn~ ' ' '
!((g, (A, B,C))) = (gAg-t,gB , Cg- 1 ) (23.24)
and this action is compatible with the action of G on M*( 4, 2) and

M*(2,4). In other words, Y(g · (A,B,C)) = gY(A,B,C ), etc. Thus,
PY,Z is an invariant and may be viewed as a map of Si, 2 /G -+
M*(4,2)/G x M*(2,4)fG .
Now the transfer matrix F(z) of (23.18) has an expansion
00
F(z) = ~Hiz-i, Hi= CAi- 1 B (23.25)

j=l
and an associated "block" Hankel matrix
(23.26)
which is of rank 2 (when Hj = GAi-l B). In view of the Cayley•

Hamilton theorem, the matrix
(23.27)
must also be of rank 2. In other words, if (A,B,C) E Si, 2 , then

H(A,B,C) = 1l~(A,B,C) is of rank 2. Moreover, H(g · (A,B,C)) =
H( A, B, C) so that H is an invariant for the action of G. Now, sup•
pose, on the other hand, that 1i = (Hi+j-I)lJ= I is a "block" Hankel
matrix of rank 2. Then we have:
Query 23.28. Does there exist an (A, B, C) in Si, 2 such that

Hi= CAi- 1 B?
INTERLUDE 161
Let 1-lcq, tla 2 ( a 1 < a 2 ) denote the first two independent rows of
1-l. Note that 1 ::; a1 ::; 2 (otherwise 1{ = 0). Then there is a 2 X 2 C
with
and since 1{i+ 2 is a subrow of 1-li for any i, there is a (unique) 2 x 2 A

with
If we let B be the 2 X 2 matrix consisting of the first two columns of

1-la where
i.e. if B = [1la]j=l,2, then it follows that CB = H1,CAB =Hz, ... ,

GAi-l B =Hi, ... , and that 1£; = Z(A,B,C)Y(A,B,C) = H(A,B,C).
We know that the rank of 1£~, p(?i~) ::; 2 and we shall show that, in
fact, p(?i~) = 2. This will insure that (A, B, C) E S5, 2 • Let
and let A be the least integer £ such that p(?i~) ::; 2.
Claim 23.29: A ::; 2.
To verify the claim, we first have the (obvious):
Observation 23.30: If 1-lr E span{Hi: j = 1, ... ,r- 1}, then

(1i~r E span{(1i~)i :j = 1, ... ,r-1}.
Observation 23.31: For all k, if (1i1r E span{('H;)i : j =

1, ... , r - 1} (where 1{1 is an appropriate su bmatrix of h), then
(1-l;+lr+z E span{('HI+ 1 )i :j = 1, ... ,r+2-1}. (This holds since
1{i+ 2 is a sub row of 1{i as h;,i+2 = hi+2,j.)
Let a = (a I, a2) be first two independent rows of 1i and note

1:::; ai:::; 2. If r f:. ai or a2, then 1ir E span{Jii: j = 1, ... ,r -1}
and, by (23.30), (1i~Y E span{(1t~)i : j = 1, ... , r - 1}. Let lt;x, =
{a; : i = 1,2 and a; :::; 2A} so that at least ai E a;x,. If r ft O;x,
and r :::; 2A, then (1i~Y E span{(1t~)i : j = 1, ... , r - 1} and so,
2 = p(1t~) :::; number of elements in a;x, :::; 2. In other words, a;x, = a
and 1 :::; OJ < a2 :::; 2A. Since A is minimal, (1t~Y' 2 E [H ;x, • • • H2A-I]
(otherwise p(1t~=~) = 2) and p(1t~=i):::; 1. Thus, 02 = 2(A- 1) + io
where j 0 = 1 or 2. If A > 1, then a 2 = 2(A- 2) + 2 + j 0 • However, in
view of 23.31, if r + 2 E a;x,, then r E a;x,. Thus, ai = 2(A- 2) + j 0
where jo = 1 or 2. It follows that A = 2 (if A > 1) since 1 :::; OJ :::; 2.
In other words, the claim is verified. *
Letting Hank(2, 2, 2) = {1i : 1i a block (2 X 2) Hankel matrix of
rank 2} and letting '1/J: Hank(2,2,2)----> M(4,4) be given by
'l/;(1i) = 1t~ (23.32)

then we have effectively shown that M E 'I/;(Hank(2, 2, 2)) if and only
if there is an (A,B,C) in Si 2 with H(A,B,C) = M. If M*2(4,4)
is the set {M E M(4,4) : r~nkM = 2} and if G acts trivially on
M*2(4,4), then it is easy to show that the map (Z, Y)----> ZY is a G•
isomorphism between M*(4, 2)xM*(2,4) and M* 2(4,4) i.e. an isomor•
phism between M*(4,2)/G X M*(2,4)/G and M*2(4,4)/G. This gen•
erates an isomorphism between Pv,z(S~, 2 /G) and PH(Si,dG) where
PH((A,B,C)) = H(A,B,C). In a "loose" sense, these generate a
geometric quotient for the action of G. (Details and rigor are in Part
II and reexamination of the "classical" proof of the ring of invariants
property of a geometric quotient for scalar systems given in section 19
might prove fruitful. )
Now let us suppose that our system (A,B,C) with m = 2,p = 2
and n = 2 has an A with distinct eigenvalues AI, A2 (i.e. AI f:. A2 ).
Then F(z) = C(zl- A)-I B has the "partial fraction" expansion
(23.33)
* This argument is developed in llissanen, J. "Basis of invariants

and canonical forms for linear dynamic systems", Automatica, Vol.
10, 1974.
INTERLUDE 163
or,
(23.34)
where Ct,C2 are the columns of C and B 1 ,B 2 are the rows of B.

In other words, (A, B, C) can be viewed as the "direct sum" of the
systems (.\ 1,Bl,Cl),(.\2,B 2,C2) which have m = 2,p = 2 and de•
gree n = 1. If, say, (A, B, C) is controllable, then so are ( .\ 1, B\ Cl)
and (.\ 2,B 2,C2). Conversely, if (.\1,B 1,Cl) and (.\2,B 2,C2) are con•
trollable, then so is their "direct sum". Of course, the same applies
Si
to observability. If .6. denotes the discriminant, then ( 2 ).a. "= "
S~ 2 EElS~ 2 • We shall use this idea shortly in proving a result concern•
ing state'feedback which is known as Heymann's lemma (and also in
Part II for one proof of the Geometric Quotient Theorem).
Systems of degree n = 1 are also of interest in other problems.
For example, in section 19, we used a "canonical form" for controllable
scalar systems under state equivalence in one proof of the ring of
invariants property of a geometric quotient. Such a global smooth
canonical form does not exist if m 2 2. Let n = 1, m = 2 and consider
systems of the form (a, [b 1 b2 ]) with a, b1 , b2 in C. Such a system is
controllable if and only if either b1 =I 0 or b2 =I 0. The action of
G = GL(1, C) = C - (0) = C* is multiplication on [b1 b2] and so the
orbit space is simply C x (C 2 - {0})/C* = C x P~. Thus, there is
no global canonical form as there is no global canonical form for the
homogeneous coordinates on P~. Naively, if ~o =I 0,6 =I 0, then we
must consider (1,6/~o) and (~o/6, 1). With more sophistication, we
call a finite formal sum~ niPh nj E 1 (the integers) and Pj E PL a
divisor and we let Div(PU be the (free abelian) group of divisors on
P~. Iff= fdh where ft,h E C[Xo,Xl] are homogeneous of the
same degree and relatively prime (i.e. f is a non-zero element of the
"function field", C(P~) of P~ ), then the divisor of j, (f) is given by
(23.35)
where f = f1p?' ,p; distinct irreducible homogeneous polynomials

(here Pi are linear forms) and P; is the zero of p;. If we define the
degree of a divisor D = ~ n;P; to be deg D = ~ n;, then deg(J) = 0.
Conversely, if D = ~ n;P; and deg D = 0, then D = (f) for some
f. Any such divisor is called a principal divisor. Div(P~)/ (princi•
pal divisors) is called the Picard Group of P~ and denoted Pic(PU.
Since the sequence 0---+ C(P~)*je*---+ Div(P~)d~ez---+ 0 is exact,

Pic(P~) = Z. A global canonical form would in a sense correspond
to a global generator of Div(P~) modulo principal divisors so that
Pic(P~) would not be Z. We shall see in Part II that "divisors" are
important in a number of multivariable control problems.
We now turn our attention to the final result of this interlude,
namely: Heymann's Lemma. Let A ben X nand let B be n X m
so that the system (A,B) E Af+nm. If G = GL(n,C) and if rf =
G X Hom c(en' em) = G X Acn' then r f becomes an algebraic group
under the multiplication
(23.36)
and r1 acts on Ac2 +nm via
[g, K]· (A, B)= (g(A + BK)g-l,gB) (23.37)
(cf. section 21). If Z(A,B) =[BAB···An- 1 B] is the "controlla•

bility" map, then the set of (A, B) which are not controllable is the
closed set (affine) determined by the vanishing of all n X n minors of
Z(A,B). Thus, S'j. m ={(A, B): (A, B) is controllable} is an open
set in Ac2 +nm. M~reover, it is invariant under the action of r 1 . We
now have:
Theorem 23.38. (Heymann's Lemma) The pair (A, B) is control•

lable if and only if there is an (I, K) in r f and a win Hom c(em, C)=
em such that the scalar system (A+ BK, Bw) is controllable.
We begin with a proposition.
Proposition 23.39. The pair (A, B) is controllable if and only if there

is no proper invariant subspace W of A containing the columns of B i.e.
(0) < W < Ac,AW c Wand Col(B) c W.
Proof: Let W(A, B) = Col(Z(A, B)) = span of the columns

of Z(A, B). Then AW(A, B) C W(A, B) and Col( B) C W(A, B).
If rank Z(A,B) < n, then W(A,B) is a subspace with the requisite
properties if B -j:. 0. If B = 0, a non-cyclic invariant subspace of A will
do. Conversely, given such a W, then Col(B) C Wand AW C W
INTERLUDE 165
together imply that W(A, B) C W < Ac which gives rank Z(A, B)<
n.
Corollary 23.40. If there is a J( in Acn and a w in Ac such that

(A+ BK,Bw) is controllable, then (A, B) is controllable.
Proof: Note that Bw E Col( B). If (A, B) were not controllable,

then there is a W with (0) < W < Ac,AW c Wand Col(B) c W.
But (A+BK)W c Wfor: ifv E W, then (A+BK)v = Av+B(Kv)
and Av E W (as AW C W),B(Kv) E W (as B(Kv) E Col(B)).
This contradicts the controllability of (A+ BK, Bw).
In other words, we have established half of Heymann's Lemma.
It is the other half which is more interesting from the point of view of
algebraic geometry.
Proposition 23.41. If A has distinct eigenvalues, then (A, B) is

controllable if and only if there is a w E Ac which is not an element of
U'i= 1 Ker(Bi) (i.e. (A,Bw) is controllable).
Proof: Let >. 1 , ... , An be the distinct eigenvalues of A. Then
F(z) = (zl- A)- 1 B has the "partial fraction" expansion
F(z) =(zl- A)- B ="'n (z-I·Bi>..)

1
L..,
j=l
1 .
J.
(23.42)
where Ij are the columns of In and Bi are the rows of B. In other

words, (A, B) is the "direct sum" of the systems ( Aj, Bi ), j = 1, ... , n.
It follows that (A, B) is controllable if and only if all (>.j, Bi) are
controllable i.e. if and only if all Bi ,j = 1, ... , n are non-zero. If
Bi 1 0 then Vi = Ker(Bi) = {x E Ac : Bix = 0} is a proper closed
subset of Ac. Hence, if all Bi 1 0, then Uj= 1 Vi < Ac and so, (A, B)
is controllable if and only if there is a w ~ Uj= 1 Vi. (Note also that
(zl- A)- 1 (Bw) = ~7= 1 Ii(Biw)jz- >.i)·
In view of Proposition 23.41, we need only prove the following

proposition to establish Heymann's Lemma.
Proposition 23.43. If (A, B) is controllable, then there is a J( such

that (A+ BK) has distinct eigenvalues.
Proof: Following an idea of Hermann and Martin ([H-6]), we

consider a map (A,B): r f - M(n,n) given by
<f>(A,B)(g, K) = g(A + BK)g- 1 (23.44)
for (g, K) E r 1 . Let V(~) be the set of zeros of the discriminant~ in

M(n, n) so that M(n, n)~ (matrices with distinct eigenvalues) is open
and dense ( cf. Appendix C). If the image of (A,B), Im (A,B) contains
an open set U, then U n M( n, n )~ ::f </> and the proposition holds.
If (A,B) is a dominant morphism, then, in view of proposition 18.17,
Im (A,B) contains an open set. Thus, we are reduced to proving the
following lemma.
Lemma 23.45. If (A, B) is controllable, then cP(A,B) is a dominant

morphism.
Proof: We first note that r 1 and M( n, n) = At

are both non•
singular. Thus, letting e = (I, 0) E r f, we see that (d¢(A,B))e is a
linear map of M(n,n) X M(m,n)(= A(t X AC'n) into M(n,n) and
so, (d<f>(A,B))e will be surjective if and only if (d<f>(A,B)): is injec•
tive. Note that ((d<f>(A,B)):(M), [g, K]} = ((d<f>(A,B))e([g, K]), M) =
tr. ([(d¢(A,B))e([g, K])]M). Suppose that
(d<f>(A,B))e([g,K]) = gA- Ag + BJ( (23.46)
Then, ( d¢(A,B) ):( M) = 0 would imply that

tr. ((gA- Ag + BK)M) = 0 (23.47)
for all g, J(. In other words,
tr. ((AM- M A)g) = 0 allg

tr. (M BK) = 0 all]( (23.48)
and so,
AM - M A = 0, MB =0 (23.49)
or,
MZ(A,B) =0 (23.50)
INTERLUDE 167
where Z(A,B) = [B AB · · · An- 1 B]. Thus, (d<f>(A,B))e is surjective if

and only if (A, B) is controllable (under the assumption that (23.36)
holds.) But
(d<f>(A,B))e(g, K) = :t (I+ tg)(A + tBK)(I + tg)-t=o

=[(I+ tg) :t {(A+ tBK)(I + tg)- 1 } + g(A + tBK)(I + tg)- 1 ]'t=O
=[(I+ tg){(A + tBK)( -g)(I + tg)- 2 } + BK(I + tg)- 1 ]'t=o + gA
= -gA + BK + gA (23.51)
which is (23.46) * In fact, if (go, Ko) = /'o E f f, then a similar calcu•

lation shows that ( d<f>( A,B) )1'0 is surjective if and only if (A+ B J(o, B)
is controllable. But (A+ BKo, B) is controllable, if and only if (A, B)
is controllable. So, we have shown that if (A, B) is controllable, then
(d<f>(A,B))/'O is surjective for allf'o in r f· But (d<f>(A,B))/O surjective for
allf'o in r 1 means that the Jacobian matrix (84>(A,B)/fh) has full rank
n 2 globally and so <f>(A,B) must be injective. In view of proposition
7.18, 4>( A, B) is dominant. *
We shall prove the following general result which implies propo•
sition 23.45 in Part II (see, e.g. [S-4]).
Proposition 23.52. Let 4> : V --. W be a morphism of varieties.

If~ E V,</>(~) E W are simple points and (d<f>)€ : Tv,f. --. Tw,cf>(€) is
surjective, then 4> is diJminant.
We hope that this brief "interlude" has provided an indication

of some of the spirit of Part II: Multivariable Linear Systems and
Projective Algebraic Geometry.
* This can be established less intuitively by algebraic means (see

Part II).
* This proof of Heymann's Lemma can be found in the survey ar•
ticle: "Pole Assignment by Output Feedback", by C. I. Byrnes.
Appendix A. Tensor Products
Let k be a field. All vector spaces and algebras will be over k

unless an explicit statement to the contrary is made. Let U, V, W be
vector spaces. A mapping f3 : U X V -+ W is bilinear if for each u E U
the mapping v -+ f3( u, v) of V into W is linear, and for each v E V
the mapping u -+ f3( u, v) of U into W is linear.
Definition A.l. Let U, V be vector spaces. The pair (T,j3) where

T is a vector space and f3 is a bilinear mapping of U X V -+ T is a
tensor product of U and V if: (i) span [f3(U x V)] = T; and, (ii) given
any vector space W and any bilinear mapping f3w : U X V -+ }V,
there is a linear mapping £ : T -+ W such that f3w = £ o f3. In other
words, all bilinear maps factor through the tensor product (T, (3).
We observe that the linear transformation £ of Definition A.l is

necessarily unique for if£,£' are linear mappings of T into W with
£ o f3 = £' o (3, then£-£' = 0 on f3(U x V), and, hence, by (i), on T.
This leads us to the following:
Proposition A.2. Let (T, f3) and (T', f3') be tensor products of U and
V. Then there is a unique isomorphism r: T-+ T' such that r o f3 = (3'.
Proof: In view of the definition and our observation, there is a

unique £ : T -+ T' such that £ o f3 = (3' and a unique £' : T' -+ T such
that £' o (3' = f3. But then£ o £' and £' o £ must be the identity and so
£ = T is the isomorphism.
In other words, the tensor product if it exists is essentially unique.
We now construct a tensor product to establish existence.
Proposition A.3. There exists a tensor product of U and V.
Proof: Let M denote the vector space kuxv. The elements of

M are formal finite linear combinations of elements of U X V with
coefficients ink i.e. expressions of the form ~ai(u, v) with ai E k. Let
168
Modern Birkhäuser Classics, https://doi.org/10.1007/978-3-319-98026-3
APPENDIX A. TENSOR PRODUCTS 169
N be the subspace of M generated by all elements of the forms:
(u1 + uz,v)- (u1,v)- (uz,v)

(u,v1 + vz)- (u,v1)- (u,vz)
(au,v)- a(u,v)
(u,av)- a(u,v)
where u, u 1 , uz E U, v, v1, v2 E V and a E k. Let T be the quotient

space M j N. If ( u, v) is a basis element of M, then let u 0 v denote its
equivalence class in T. If {3: U XV--+ Tis given by {3(u,v) = u 0 v,
then span [f3(U x V)] = T and {3 is bilinear since ( u1 + Uz) 0 v = u1 0
v+uz0v, u0(vl +vz) = u0v1 +u0vz, (au)0v = u0(av) = a(u0v).
If W is any vector space and f3w : U x V --+ W is bilinear, then f3w
extends by linearity to M and vanishes on N and so generates a well•
defined linear map f : T --+ W such that £( u 0 v) = f3w( u; v) i.e.
£of3=f3w-
We shall, in view of propositions A.2 and A.3, speak of the tensor
product of U and V which we denote by U 0 V (or U 0k V). It is
generated by the "products" u 0 v. If { ui} is a basis of U and {Vj}
is a basis of V, then {u; 0 Vj} is a basis of U 0 V. In particular,
if u =f. 0 and v =f. 0, then u 0 v =f. 0. We note also that U 0 V is
naturally isomorphic to V 0 U (via the map u 0 v ---+ v 0 u) and that
(U 0 V) 0 W is naturally isomorphic to U 0 (V 0 W) (via the map
(u 0 v) 0 w--+ u 0 (v 0 w)).
Lemma A.4. If u 1 , ... , Un are linearly independent over k and

v; = 0, then v; = 0 for i = 1, ... , n.
2:7= 1 u; 0
Proof: Let v~, ... , v~ be a basis of span [v1, ... , vn] so that v; =
2:;= 1 a;jvj. Then 2:~ 1 2:;= 1 a;j( uk 0 vj) = 0 and all the a;j = 0 so
that v; = 0 fori= 1, ... ,n.
If rp : u --+ u1 and 'lj; : v --+ v1 are linear maps, then the map
rp 0 'lj; : u 0 v --+ u1 0 Vi given by
(¢ 0 ?j;)(u 0 v) = rp(u) 0 ?j;(v) (A.5)
is linear. If rp and 'lj; are surjective, then so is rp 0 'lj;. If rp and 'lj; are
injective, then rp 0 'lj; is also injective (for if {ui} is a basis of U and
{vj} is a basis of V, then(¢ 0 ?j;)(u; 0 vj) = rp(u;)@ ?j;(vj) is linearly
independent in U1 0 Vi as '£aij</Y( ui) 0 '1/J( Vj) = '£i¢( Ui) 0 ( 'ljJ('£jaijVj))

and so(¢ 0 '1/J)('EaijUi 0 vj) = 0 implies '£aij¢(ui) 0 '1/J(vj) = 0 and
hence that aij = 0, for all i,j.)
Lemma A.6. Let U1 be a subspace of U and let V1 be a subspace

of V. Let <Pt : U --+ U/Ut. 'I/J1 : V --+ VjV1 be the natural linear
maps. Then (UfU1 ) 0 (V/V1) is isomorphic to (U 0 V)f[U1, V1 ] where
[Ut, Vi]= U1 0 V + U 0 Vt and hence, Ker(¢1 0 '1/Jt) = [U1, V1].
Proof: We shall show that ( U 0 V) /[ U1 , V1 ] is in fact a tensor

product of UjU1 and VfVt. We write <Pt(u) = u and 'I/J1(v) = v
and we let w : U 0 V --+ U 0 Vf[Ut, V1] be the natural linear map.
We observe that the map ( u, v) --+ w( u 0 v) is bilinear on U X V
and is 0 for either u in ul or v in vl (by the definitions of w and
(U~, V1]). Therefore, w( u 0 v) depends only on the classes u and v
i.e. w(u 0 v) may be written as w(u,v). Setting W = U 0 Vf[U1, V1 ],
the map w: (U/U1) x (VfVl)--+ W given by w(u,v) = w(u 0 v) is
bilinear and clearly satisfies W = span [w(U fU 1 x VfV1)]. Let W
be any vector space (over k) and let /J : U/ U1 X V /Vt --+ W be a
bilinear map. Then for any u E U, v E V, the map (3 : U x V --+ W
given by (J(u,_v) = /J(u,v) is bilinear and_there exists a lin~ar map
£: U 0 V--+ W with f(u 0 v) = (J(u,v) = (J(u,v). But then£ is 0 for
either u in ul or v in vl and so i is 0 on (Ul' Vi]. It follows that i = £ow
with£: W--+ W a linear map. Then /J(u, v) = £(w(u0v)) = f(w(u, v))
or in other words, /J =£ow. Thus, U 0 Vf[U1 , Vt] is a tensor product
of (UfU1 ) and (Vf'!t) and the result follows by proposition A.2.
Now let R and S be k-algebras i.e. k is a subring of R and of S
and the element 1 of k is also the unit 1R of Rand the unit 1s of S.
Then R0kS is defined as a k-vector space. However, we can introduce
a multiplication on the tensor product R 0k S: for elements (r 0 s)
and (r1 0 s1) it is given by
(A.7)
and in general by
It is easy to see that R0k S becomes a commutative ring with identity

1 0 1. We also observe that the mappings '1/JR : R --+ R 0 S and
'lj;s: S-+ R0S given by 'lj;R(r) = r01 and 'lj;s(s) = 10s are injective
k-homomorphisms so that R and S may be viewed as subrings of
R 0k S. We shall adopt this point of view here. Finally we note that
the mapping a -+ a( 1 0 1) = a 0 1 = 1 0 a is a natural imbedding of
k into R 0k S. Thus, R 0k Sis in fact a k-algebra which we call the
tensor product of the k-algebras R and S.
Example A.9. If k C J( with J( a field, then J( 0k S contains the

field J( and any basis of S over k is also a basis of J( ® k S over J(.
Thus k0kS = S. Alternatively, the map a0s-+ as is an isomorphism
between k ®k Sand S.
Example A.10. Let R = k[X1 , ... , XN] and S = k[Y1 , ... , YM] be
polynomial rings. Then R0k S ~ k[Xb···,XN,Yt,···,YM], for the
mapping X; 01 -+ X;, 10 Yj -+ Yj clearly generates a k-isomorphism.
Proposition A.ll. Let RandS be k-algebras and let n be an ideal

in Rand b be an ideal inS . .lf[n, b] = n0 S + R0 b, then (Rjn)0 (Sjb)
is isomorphic to R 0 S /[ n, b].
Proof: Let ¢ 1 : R -+ Rjn, 'lj;1 : S -+ Sjb be the natural maps.

Since <Pt, 'lj;1 are surjective k-homomorphisms, l/>1 0 'lj;1 : R 0 S -+
(Rjn) ® (Sjb) is a surjective k-homomorphism. However, lemma A.6
implies that Ker( l/>1 0 'lj;l) = [ n, b] and the proposition follows.
We note that if {ri} is a set of generators of Rand {sj} is a set
of generators of S, then the set { r; 0 1, 1 0 s j} is a set of generators
for R 0 S. In particular, if Rand S are finitely generated, then so is
their tensor product R ®k S.
Proposition A.12. If R and S are affine k-algebras, then R 0k S

is an affine k-algebra.
Proof: Since R0k Sis finitely generated, it will be enough to show

that there are no non-zero nilpotents. Let 2::7= 1 r;®s; be nilpotent and
assume without loss of generality that the s; are linearly independent
over k. Let a E Hom(R,k) and consider the map a® 1: R 0 S-+ S
given by (a® l)(r ® s) = a(r)s. Since a 01 E Homk(R 0 S,S),
2::7= 1 a(r;)s; is a nilpotent in S. But Sis an affine k-algebra and
so 2::~ 1 a( r; )s; = 0. Since the s; are linearly independent over k,
a(ri) =
0 fori= 1, ... ,n. It follows from the Nullstellensatz (cf.
problem 20) that Ti = 0 for i = 1, ... , n.
Proposition A.13. Let R be an affine k-algebra. Suppose that R

and S are integral domains. Then R ®k S is an integral domain.
Proof: Let u = E;r; 0 s; and u' = E irj 0 sj be elements of R 0 5

such that uu' = 0 and assume without loss of generality that the s;
and sj are linearly independent over k. Let a E Hom(R,k) and again
consider the map a 0 1: R 0 5--+ 5 given by (a 0 1)(r 0 s) = a(r)s.
Since a 0 1 E Hom k(R 0 S, 5), we have
(a 0 1)(uu') =(a 0 1)(u) ·(a® 1)(u') = (E;a(r;)si)(Eja(rj)sj) = 0.

But 5 is an integral domain and the s; and sj are linearly independent
= =
over k and so, either a(r;) 0 for all i or a(rj) 0 for all j. It follows
that a(r;rj) = 0 for all i,j and hence, as a was arbitrary, that r;rj = 0
for all i,j. If, say, u f 0, then some r; is not zero and rj = 0 for all
j. In other words, if u f 0, then u' = 0. Thus R ®k S is an integral
domain.
Corollary A.14. Let R and S be k-algebras and let a be a prime

ideal in Rand b be a prime ideal inS. Then [a, b] =a® S + R® b is a
prime ideal in R ® 5.
Corollary A.15. Let RandS be integral domains and let f f 0 be

an element of R. Then Rf ®k S ~ (R ®k 5)!01·
Proof: Consider the map 'lj; : R 1 ® S --+ (R ® S)u 01 ) given by

'lj;(rf f 0 s) = (r ® s)f(f ® 1) and extended in the natural way. 'lj; is
clearly a k-homomorphism. If h = Ei(ri®s;)/{!®1), then 'lj;(E;r;f J;®
s;) = h so that 'lj; is surjective. Since Rand 5 are integral domains, so
are R ® 5 and ( R ® 5)u 01 ). If 'lj;(E;r;/ f ® si) = 0 with the s; linearly
independent over k, then E;(r;®s;)/(!®1) = 0 and so E;(r;®s;) = 0.
The injectivity of 'lj; follows from lemma A.4.
Corollary A.16. Let R and S be integral domains and let f f 0,

g f 0 be elements of R, 5 respectively. Then (RJ ®k 5 9 )= (R®k 5) 109 .
Proof: Note that Rt0S9 = (R0S 9 )(!01) = [(R0S)(l0g)](f01) =

(R 0 S)u 0 1)(10g) = (R 0 S)(f0g)·
We shall now extend the notion of a tensor product to "modules".
Definition A.l7. Let R be a ring. An abelian group M on which

R acts linearly is called an R-module (i.e. M is an R "vector" space).
More explicitly, there is a mapping R X M-+ M, (r,m)-+ rm, such
that r(m1 + m2) = rm1 + rm2, (r + r1)m = rm + r1m, (rr1)m =
r(r 1m), and lm = m.
If L, M, N are R-modules, then a mapping f3 : M X N -+ L is

R-bilinear, if for each 'TTl- E M, the mapping n -+ f3( m, n) of N into L
is R-linear, and for each n E N, the mapping m -+ f3( m, n) of M into
L is R-linear. The tensor product of two R-modules M and N can
be developed in a manner similar to that used for vector spaces using
the universal property for factoring bilinear maps. More precisely, we
have:
Theorem A.l8. Let M, N be R-modules. Then there is an R-module

T and a bilinear mapping f3T : M x N -+ T such that : (*) if L is any
R-module and f3 is any bilinear mapping of M x N into L, then there is
a unique R-linear mapping £{3 : T-+ L such that f3 = £{3 o f3T. Moreover,
ifT' and f3T' satisfy(*), then there is a unique isomorphism T: T-+ T'
such that r o f3T = f3T'.
Proof: Uniqueness is proved in a manner entirely similar to that

used in proposition A.2. As for existence, we consider the free R•
module RMxN i.e. the expressions I:~=l r;(m;,n;). If Pis the sub•
module of RMxN generated by all elements of the following forms:
(m+m 1 ,n)- (m,n)- (m1,n)

(m, n + n1)- (m, n)- (m, n1)
(rm,n)- r(m,n)
(m,rn)- r(m,n)
then we let T be the quotient R-module RMxN j P and for each basis
element (m, n) in RMxN, we let m@n be its residue class in T. Then T
is generated by the elements m 0 n and the mapping f3T : M x N -+ T
defined by /h( m, n) = m 0 n and extended by linearity, is clearly

R-bilinear. IT Lis an R-module and f3: M X N-+ Lis bilinear, then
the map .efJ : T -+ L given by £fJ( m 0 n) = {3( m, n) satisfies ( *).
Definition A.l9. The R-module T of theorem A.l8 is called the

tensor product of M and N and is denoted by M 0R N or simply by
M0N.
We note that R is itself an R-module and that R 0R N" = "N in

the sense that the mapping r 0 n -+ rn is an isomorphism. We also
have
Proposition A.20. If M and N are R-modules, if M' is a submodule

of M and if N' is a submodule of N, then
(M/M')0R(N/N') ~ M0RN/[Im(M'0RN)+Im(M0RN')] (A.21)
under a canonical isomorphism.
Proof: (see [F-2]).

Appendix B ([M-4]). Actions of Reductive Groups
Let G be an affine algebraic group over k. (This means that G is

an affine algebraic set which is a group and that the group operations
are k-morphisms.) Let S = k[G] be the ring of regular functions on
G. We note that S is a k vector space. Let M be a k vector space
and let a be a linear map of M into k[G] ® M so that
a(m) = Ej; ® m; (B.l)

with J; E k[G]. Then a defines a mapping a(7: Gx M--+ M by setting
(B.2)
where a(m) is given by (B.l). We say that a defines an action of
G on M if a(7 is an action (i.e. a(7(g1g2,m) = a(7(g 1 ,a(7(g 2 ,m)) and
a(7(c:,m) = m), and, in that case, we say that G acts on M. If a
defines an action, then m E M implies m is contained in a finite
dimensional invariant subspace (Ekm;). Clearly, m is invariant if
and only if a(m) = l(·)m and M 1 C M is invariant if and only if
a(M1 ) C k[GJ ® M 1 . We usually suppress the linear map a.
Example B.3. Let M ben-dimensional with e1 , •.. , en as basis. If

a(e;) = Ejfij(·)ej, then (f;j(·)) E GL(n,k) and the map g--+ (f;j(g))
is a k-homomorphism of G into GL(n,k).
Definition B.4. Let G act on a finite dimensional vector space M.

The action is completely reduced if, for any invariant subspace M 1 of
M, there is an invariant complement (i.e. an invariant subspace M 2 of
M such that M = Mt EB M2). The group G is reductive if each action
on any M is completely reduced.
The group GL(n,k), k of characteristic 0, is an important exam•
ple of a reductive group ((D-2]). The importance of reductivity for
our purposes lies in the following results.
Proposition B .5. If G is reductive and acts on M, then the subspace

M = { m : g · m = m} of invariant elements has a unique invariant
0
complement Ma.
175
Proof: By Zorn's lemma, there is a maximal invariant subspace

Me such that Mg = Men Me = {0}. If N is any invariant subspace
of M and m E N, then there is a finite dimensional invariant subspace
N 1 C N with m E N 1, and hence, an invariant subspace N' C N 1 with
N 1 = (Nt nMe)r:FN' (as G is reductive and N1 is finite dimensional).
If Ne = {0}, then N'e = {0} and (Me ffi N')e =(Me ffi N') n Me =
{0}. By the maximality of Me, we have N' = {0} and so, N C Me
if Ne = {0}. In other words, Me is unique. Since Men Me= {0},
we need only show that M = Me ffi Me. So let m E M and let M 1
be a finite dimensional invariant subspace with m E M1. Since G is
reductive, there is an invariant M{ C M1 with M 1 = (M1 n Me) ffi M{
and M{e = {0}. This implies that M{ C Me and mE M 1 C Me ffi
Me.
Definition B .6. If G is reductive and acts on M, then the mapping

PM : M --+ Me is called the Reynolds operator on M.
We note that PM is a projection of M onto Me with kernel Me.
Proposition B. 7. If G is reductive and acts on M and N and if

L : M --+ N is a G-linear map (i.e. commutes with the action of G), then
(B.8)
i.e. L commutes with the Reynolds operator.
Proof: First note that L(Me) c Ne since, if g · m = m, then

L(m) = L(g·m) = g·L(m). So it is enough to show that L(Me) C Ne.
If m E Me, then there is an M1 C Me with m E M1 and M1 =
(M1 n KerL) 81 M{ where M{ is invariant. Since L(MI) = L(M{),
L(Mi)e = L(M{)e = L(M{e) = {0} (as M{e C Me n Me). It
follows that L(m) E L(M{) C Ne.
Corollary B.9. Let R be a k-algebra and suppose that the reductive

group G acts on R by k-automorphisms (i.e. the map r--+ g · r, g E G is
a k-automorphism of R). Then
PR(xy) = xPR(Y) (Reynolds Identity) (B.lO)
if x E Re and y E R.
APPENDIX B ACTIONS OF REDUCTIVE GROUPS 177
Proof: The mapping y --+ xy is G-linear if x is an element of R 0 .
Corollary B .11. Let R be a k-algebra and suppose that the reductive

group G acts on R by k-automorphisms. If a 1 , ... , ar are invariant ideals
in R, then o=~=l a;) n R 0 = l:::~=l(a; n R 0 ).
Proof: Since E( a; n R 0 ) c (Ea;) n R 0 , we need only show that

iff= Ef; with J; E a; and f E R 0 , then f = Eff with ff E a; n Ro.
Let PR be the Reynolds operator on R and let P; = PR restricted to
a; be the Reynolds operator on a; (noting that a; is invariant). Then
(B.12)
since f E R 0 . But P;J; = ff E ay
C a; n R 0 and the corollary is
established.
Let us suppose that R is a k-algebra and that the reductive group
G acts on R by k-automorphisms. If 5 is an R 0 -alg;ebra, then G also
acts on R 0Ra 5 via
g . ( 1' 0 s) = (g . 1') 0 s (B.13)
where g E G, r E Rand s E 5. We then have:
Lemma B.14. If 5 is an R 0 -algebra, then
(R 0RG 5) 0 = 5 (B.15)
i.e. the ring of invariants in (R 0Ra 5) is 5.
Proof: Let P = PR and P' = PR0 Ra s be the Reynolds operators.

Since 5 ~ R 0 0Ra 5 (appendix A) C (R0Ra 5) 0 , it will be enough
to show that if E(r; 0 s;) E (R 0Ra 5) 0 , then E(r; ~) si) E R 0 0 Ra 5.
But P'(E(r; 0 s;)) = P'(E(r; 01)(10 s;)) = E[P'(r,; 01)](10 5;) (by
Reynolds identity). Since P'(r; 0 1) = (Pri) 0 1 by proposition B.7,
we have E(r; 0 s;) = P'(E(r; 0 s;)) = E(Pr;) 0 s; E R 0 0Ra 5 and
the lemma holds.
Corollary B.16. If a is an ideal in R 0 , then Ran R 0 = a.
Proof: We observe that R 0 Ra R 0 /a = R/ Ra in view of propo•

sition A.20. It follows from the lemma that (R/ Ra) 0 = R 0 fa and
hence that Ra. n R0 = a..
Corollary B.17. If R is noetherian, then RG is noetherian.
Proof: By corollary B.16, the mapping a -+ ae = Ra is an injec•

tion and satisfies the condition that a < a 1 implies Ra < Ra 1 . Thus,
if {a;} is an ascending chain of ideals in RG, then { Ra;} is an ascend•
ing chain of ideals in R. Since R is noetherian, Ral/ = Ral/+l = · · ·
for some 11 and so, al/ = al/+1 = · · · .
A ring R is a graded ring if there is a family Rn, n = 0, 1, ... of
subgroups of the additive group of R such that (i) R = EBERn and
(ii) RmRn ~ Rm+n for all rn, n;::: 0. In that case, Ro is a subring of
R and each Rn is an R 0 module. Any element r of R can be written
uniquely as a sum Ern where Tn E Rn and all but a finite number of
rn are 0. The non-zero rn are called the homogeneous components of
r. Elements of Rn are called homogeneous of degree n.
Example B.18. Let R = k[X1 , .•• ,XN] and let Rn be the set of
all homogeneous polynomials of degree n.
Proposition B.19. If R is a graded noetherian ring, then Ro is noe•

therian and R is a finitely generated Ro-algebra (i.e.R = R 0 [!I, ... , f N ]).
Proof: Let R+ = EB L:n>O Rn. Then R+ is an ideal in R and

Ro ~ R/ R+. It follows that Ro is noetherian and that R+ has a finite
basis h, ... , JN which we may assume are homogeneous of degrees
d1, ... ,dN (all> 0). Let R' = Ro[fi, ... ,fN] CR. We claim that
R' = R. To verify the claim, we need only show that Rn C R' for
all n. We use induction on n. Clearly R 0 C R'. Let n > 0 and let
r E Rn· Since r E R+, r = L:~ 1 rif; where r; E Rn-d; or r; = 0 (if
n < d;). By induction, r; E R' and sorER'.
Let us suppose now that R is a graded noetherian k-algebra (with
Ro = k) and that the reductive group G acts homogeneously (i.e. G ·
Rn C Rn) on R by k-automorphisms. Then RG = EB I:~=O R~ is also
a graded noetherian k-algebra (with Rfj = k) and is, by proposition
B.19, a finitely generated k-algebra. In other words, in this situation,
the equivalence Ec on R is proper (definition 14.5). Thus, if k is of
characteristic 0, then the equivalence Ec, G = GL(n,k), on R is
proper. If k has characteristic p, then not every action on any M is
completely reduced for GL(n,k). Nonetheless, ([M-3], [H-1], [S-4]),
the equivalence Ec is still proper.
Appendix C. Symmetric Functions and
Actions of the Symmetric Group
Let us consider the set of integers {1, ... , n }. A bijective map

of this set onto itself is called a permutation on n letters. The set of
all such permutations forms a group Sn called the symmetric group of
degree n. The order of Sn (i.e. the number of elements) is n!. An
element a of Sn such that
(C.l)
and leaves the rest of {1, ... , n} fixed is called a cycle and we write
a = ( i1 i2 · · · ir ). Every a E Sn can be written as a product of dis•
joint cycles. A cycle of the form ( i1 i2) is called a transposition. Since
( i1 · · · ir) = (i1 i2 )( i1 i3) · · · ( i1 ir ), every a E Sn is a product of trans•
positions.
Let R = k[x 1, ... , xn]· Then we may define an action of Sn on R
via automorphisms as follows:
where f E R and a E Sn.
Definition C.3. A polynomial function f in R is symmetric if

a ·f= f for all a in Sn i.e. f is an invariant for the action of Sn on
R. We let R 8 n be the ring of symmetric functions. The polynomials
(C.4)
r = 1, ... , n are called elementary symmetric polynomials.
Proposition C.5. The tPr are symmetric.
Proof: Now let z be an indeterminate over R and let F(z) =

(z- x1)(z- x2) · · · (z- Xn)· The action of Sn on R can be extended
179
to an action on R[z] by setting cr(z) = z. Then cr E Sn permutes the

factors ofF so that cr · F = F. Thus, cr leaves the coefficients of F
invariant. But F(z) = zn -,P1(x)zn- 1 +¢2 (x)zn- 2 - • • ·+ ( -l)ntPn(x)
so that the tPr( x) are symmetric.
Proposition C.6. R is algebraic over k[t/11, ... , tPnl·
Proof: Simply note that F(xi) = xf-,P1 (x)x~- 1 +· · +tPn(x) = 0.

Proposition C. 7. R 8 n = k[ tPt, ... , tPn]·
Proof: If f E R 8 n, then f = fo + f1 + · · · + fm where fi is
homogeneous of degree i and, since the action of Sn preserves de•
grees, each fi is in R 8 n. Thus, we may suppose f is homogeneous
of degree m. Call a monomial x; 1 • • • x~" of degree m greater than
xf 1 ···X~" if k1 = £1, ... , ks = ls but ks+l > ls+1· For example,
xix2x3 > x 1 x~ > x 1 x~x3. Let ax; 1 ···X~" be the greatest term
in f. Since f is symmetric, kt ~ k2 ~ · · · ~ kn. The greatest
term in a,P~ 1 -k 2 ,P~ 2 -k 3 • • • ¢~" is thus the same as that of f. But
then ft = f - a?JI~ 1 - k 2 • • • t/1~" is a homogeneous symmetric poly•
nomial whose greatest term is less than that of f. By induction,
ft E k[t/11, · · ·, tPn]·
Proposition C.8. The tPr are algebraically independent over k.
Proof 1. If there is a non-trivial relation over k amongst the tPn

then distinct terms in the tPr have distinct greatest terms in the x j.
The greatest term in the x j occurs only once which gives a non-trivial
relation amongst the x j (a contradiction).
Proof 2. We observe that

tPt(Xt, · .. ,xn) =Xt + t/J~(x2, ... ,xn)
t/J2 ( Xt, · · ·, Xn) =Xt tP~ (X2, · · · , Xn) + tP; (X2, · · · , Xn)
(C.9)
tPn( Xt, · .. , Xn) =Xt tP~-1 (x2, · · ·, Xn)

where the¢~ are the elementary symmetric polynomial in x2, . .. ,xn.
If P( t/J1, ... , tPn) = 0 and we set x1 = 0, then P( ,P~, ... , t/J~_ 1 , 0) = 0.
By induction, the t/1~ are algebraically independent. Thus, P(zt , ... ,zn)
= ZnPt ( Zl, ... , Zn) with Pt of lower degree than P. Since t/1 n -=/= 0,
Pt(¢ 1 , . .• ,,Pn) = 0. By induction on the degree, P1 0. =
APPENDIX C SYMMETRIC FUNCTIONS AND SYMMETRIC GROUP ACTIONS 181
Proposition C.10. Let 8r( x1, ... , xn) = 2:~= 1 xi, r = 1, ... , n.
Then the 8r are symmetric and, for k of characteristic 0, satisfy New•
ton's identities
(C.ll)
for r = 1, ... , n.
Proof: Clearly the 8r are symmetric. As for the rest, simply use
induction and (C.9).
Example C.12. Let r = 3 and 8: = 2:7= 2 x~. Then 83 - t/J18z +

tPz8I - tP3 = xr +.J.'8j- (x1.J.' + t/J~)(xi + 8~) + (xit/J~ + t/J;)(XI + 8i)-
XI .J.'
'+' 2 - .J.'
'¥ 3 = 83 -
I
'+'I 8I2 + '+' 2 81I - .J.'
'+'3 - x1 ( 8 I2 -
.J,f
'+'I 8II ) = 0 (by m
' d uctwn
'
on n).
If A E M(n,n,k) is ann x n-matrix, then
det(zi- A)= zn- Xl(A)zn-l- · · ·- Xn(A) (C.l3)
where the X;(A) are the characteristic coefficients of A. The roots of

det( zi - A) = 0 are the eigenvalues of A. If AI, ... , An are the eigen•
values of A, then x;(A) = ( -1)i- 1t/J;(A1, ... , An)· The characteristic
coefficients can be computed directly from A and, in particular, we
have, for characteristic k = 0:
Proposition C.14. Let sr(A) = tr(Ar) (the trace of Ar) for r

1, ... , n. Then
8r(A)- XI(A)sr-I(A)- ... - rxr(A) =0 (C.15)
for r = 1, ... , n.
Proof: If AI, ... , An are the eigenvalues of A, then tr(Ar) = ~Ai
and apply proposition C.lO.
Definition C.l6. Let D(xi, ... ,xn) = ni~i<j~n(x;- Xj) 2 • Then

Dis symmetric and so D(xi, ... ,xn) = A(t/JI,-t/J 2 , ... ,
( -1)n- 1t/Jn)· The function A( A) = A(XI(A), ... ,Xn(A)) is called
the discriminant of A.
Proposition C.17. A E M(n,n,k) has distinct eigenvalues if and

only if A(A) =f 0.
Proof: A has distinct eigenvalues if and only if D(>.t, ... , >.n) =f 0
which holds if and only if A(A) =f 0.
Corollary C.18. The set {A EM(n, n, k) :A has distinct eigenvalues}

is open and dense.
Proof: Simply observe that the set is the principal affine open set
(Ak2)A·
Proposition C.19. If g E GL(n,k) and A1 = gAg- 1 , then

A(At) = A(A).
Proof: det(zi- At)= det(zgg- 1 - gAg- 1 )
= detg det(zi- A) detg- 1 = det(zi- A).
Now let D( n, k) be the set of n X n diagonal matrices with entries
ink and let
N = {g E GL(n,k): gD(n,k)g- 1 = D(n,k)} (C.20)
Clearly, N is a subgroup of G L( n, k) and is, in fact, the set of so•
called monomial matrices (i.e. matrices with exactly one non-zero entry
in each row and column). We also note that N contains a subgroup
which is isomorphic to the symmetric group Sn. To see this, we observe
that it is enough to consider the transpositions (ij) and that, if j > i,
then (ij) = (ii + 1)(i + 1 i + 2)···(j- 1 j) so that it is enough to
consider a transposition (i i+ 1). The element of N which corresponds
to (i i + 1) is given by:
z i+1
1
O"(i i+l) =z 0 1 (C.21)
i+1 1 0
1
1
APPENDIX C SYMMETRIC FUNCTIONS AND SYMMETRIC GROUP ACTIONS 183
Note that a(i\+ 1) = a(i i+1) and that a(i i+l)diag[dt, ... , dn]
a(i\+1) = diag[d1, ... , di-1, di+t, d;, di+2, ... , dn]· It follows that
adiag[dt, ... ,dn]a- 1 = diag[da(l),···,da(n)] (C.22)
for any a in Sn. It is, in fact, not hard to show that N = Sn X Do( n, k)
where D0 (n,k) = D(n,k) n GL(n,k).
Now let R = k[X1, ... , XN] where the X; are vectors of indepen-
dent variables, X;= (xil, ... ,Xim)· The symmetric group Sn acts on
R by permuting the vectors X;. In other words,
(a· J)(X1, ... ,Xn) = r(Xt, ... ,Xn) = f(Xa(t), ... ,Xa(n)) (C.23)
where f E Rand a E Sn. By analogy with proposition C.7, we want

to find the generators of R 5 n. Our treatment is based on [N-2]. If 1/J
is a monomial in the X;, we let 0(1/J) be the sum of the elements in
the orbit of 1/J.
Lemma C.24. Suppose that k has characteristic 0. Then R 5 n is

generated by elements of the form 0( 1/J) where 1/J = x~\ · · · x~;, is a
monomial in X1.
Proof: If 't/J 1, ... , 1/Js are monomials in X1, ... , Xs, respectively,
then the length of 0( 't/J 1 · · · 1/Js) is s. Clearly any element of length 1
can be obtained from monomials in X 1 . We now use induction on s.
Supposing the result for all elements of length s or less, we have f0r
any monomials 't/J1, ... , 't/Js+1 in Xt, ... , X s+l, respectively,
(C.25)
where cis a positive integer and 6 consists of elements of length :=::; s*.
The lemma follows by the induction hypothesis.
Lemma C.26. Suppose that k has characteristic 0. The elements

0( x1; 1 x2; 2 • • • x si. ), s ::; n, generate the vector subspace of R 5 n consist•
ing of elements with total degree ::; n.
* Forexample,if't/J1 = 't/J2 = ··· = 1/Jr,'t/Jr+j f' 1/Jt.J = l ... ,s+l-r,

then c = r.
Proof: By the previous lemma, it is enough to show that the

elements 0( 'lj; ), 'lj; a monomial in X1 of total deg,r~e ::; n, can be written
· ) • We observe that O(x1·t1 · .. x Sts· ) =
l· n terms of the O(x1·t1 x2·t2 .. · x SZs
Ct0(Xti 1 )0(x2i 2 • • ·Xsi.)- LJ=Z Cj0(x2i 2 • ··Xj-li;_ 1 Xji 1 Xji; · · ·Xsi,)
where the Cj are positive. Continuing we have, O(x1; 1 • • • Xsi,) =
8t + (-1)k-f.l:jc}O('Ij;jl'I/Jj2 ···'1/Jjl) where 8t is a sum of products of
shorter terms, the c} are positive, and '1/Jii is a monomial in Xi. Finally,
e
for = 1, we have
where 81 is a sum of products of shorter terms and c1 is positive. By

induction and the fact that c 1 is positive, the result follows.
Corollary C.27. Suppose that k has characteristic 0. Then R 5 n is

generated by the multilinear terms 0( x1; 1 • • · Xsi, ), s ::; n.
Proof: Let n 1 > n and let R 1 = k[X1 , ... , XN ]. There is a

homomorphism a : R 1 ""' R with a( X;) = X;, i = 1, ... , n and
a(Xn+j) = O,j = 1, .. ·,n1-n. Thisdeterminesah omomorphisma1 :
R;n 1 ""'R 5 n which sends O('lj;1 · · · '1/Js) into 0('1/Jl · · · '1/Js) for multilinear
monomials 'I/J1, ... , '1/Js with s ::; n and sends 0( 'I/J1 · · · '1/Js) into 0 for
monomials with s > n.
Example C.28. Let n = 2, m = 2 so that X 1 = (xu, x12),

Xz = (xn,x 22 ) and R = k[X 1,Xz]. Then R 52 is generated by the
terms O(xn) = Xn + xn, O(x12) = x12 + xn, O(xux21) = 2xux21,
O(x12 x22 ) = 2x 12 x22, O(xuxzz) = xuxzz + XztX12 = O(x12x 21 ).
Observe that O(xux12) = xux12 + xnxn =(xu+ x21)(x12 + x22)•
(xuxzz + x21x12) = O(xn)O(x12)- O(xuxzz) and that O(xnxu) =
xi 1 + x~ 1 =(xu+ xn) 2 - 2xux21 = O(xu)O(xu)- O(xuXzt)·
Appendix D. Derivations and Separability ([S-4], [Z-3])
Let k be a field and let R = k[x] where x is an indeterminate.

Let J( be an extension field of k and suppose that ~ E ]( is algebraic
over k. Then the set h(O = {! E R : f(~) = 0} is a prime ideal
in R (for (h/2)(0 = h(~)fz(~) = 0 implies h(~) or fz(~) is zero as
]( is a field). Hence, h(O = (p) where pis irreducible. Moreover,
apart from an arbitrary unit factor a E k*, p is unique and so there
is exactly one monic irreducible polynomial Pt. such that h(O = (pt.).
We call Pt. the minimal polynomial of~ over k.
Definition D.l. An irreducible polynomial fin k[X] is separable

if f'(x) 1 0 where f' is the derivative of f. An element~ of]( which
is algebraic over k is separable (or separably algebraic) if its minimal
polynomial ~is separable. If]( is an algebraic extension of k, then ](
is a separable extension if every element ~ of ]( is separable.
We say that the field of k is perfect if either the characteristic

of k is zero or if the characteristic of k is p =f 0, then kP = k where
kP = { aP : a E k }. An algebraically closed field k is perfect since the
equation xP- a has a root in k.
Proposition D.2. If k is perfect, then every irreducible polynomial in

k[x] is separable.
Proof: If k has characteristic zero and f(x) = a0 xn+a 1 xn- 1 +· · ·+

an, then J'(x) = naoxn-l · · · + an_ 1 = 0 if and only if (n- i)a; = 0,
i = 0, 1, ... , n- 1 i.e. if and only if a; = 0, i = 0, 1, ... , n- 1. Thus,
f' (x) 1 0 if n > 0 and so, every irreducible f is separable.
On the other hand, if k has characteristic p 1 0 and f( x) =
aoxn +···+an, then f'(x) = 0 if and only if n is divisible by p and
the coefficients a; for which n- i is not divisible by pare 0 i.e. f(x) is
a polynomial in xP i.e. f E k[xP]. Since k = kP, k[xP] = kP[xP] and so
f E kP[xP] which implies f = (fi)P where fi E k[x]. In other words,
f is not irreducible.
185
Definition D.3. Let ]( be a.n extension field of k. Then ]( is

a. vector space over k and we say that ]( is a. finite extension of k if
dimkK = [K: k] is finite i.e.]( is a. finite dimensional k-vector space.
We observe that if ]( is a. finite extension of k, then ]( is a.n

algebraic extension of k for if ~ E ]( and [K : k] = n < oo, then
1,~, ... ,~n are linearly dependent over k. We also observe that if~
is algebraic over k, then ]( = k(~) is a. finite extension of k since in
this case k(O = k[~] ([Z-3]) and 1,~, ... ,~n-t are a. basis of k[~] over
k where n is the degree of the minimal polynomial of~·
Definition D.4. Let ](be a.n algebraic extension of k. An element

~ of ]( is a. primitive element of ](I k if ]( = k( ~).
Theorem D.5 (Theorem of the Primitive Element). If k is an infinite

field* and ]( is a finite separable extension, then ](I k has a primitive
element.
Proof: Let J{ = k(6, ... ,~n) and let x,xt, ... ,xn be indeter-
minates. Set k = k(xt, ... ,xn) and K = K(x 1 , •.. ,xn)· Then
K = k( 6, ... , ~n) and K is a. finite separable extension of k. Let
[ = 6x1 + ·· · + ~nXn (D.6)
and let P(x) be the minimal polynomial of [in k[x]. Then
where g is a. common denominator of the coefficients of P. Let us set

G(xt, ... ,xn) = J([,xt,···,xn) so that G is in K[xt,···,xn]· Since
P( [) = 0, G = 0 and so 8GI OXi = 0 i = 1, ... , n. It follows that
for i = 1, ... , n. _!'fote that ~D.8) is a.n identity in Xt, ... , Xn· Since [
is separable, P' ( ~) =f 0 and M(
~, x 1 , •.. , x n) is a. non-zero polynomial
* If k is finite, then the theorem is still true since in that case a finite
separable extension J{ is also a finite field and](* is a cyclic group ([Z-3]).
APPENDIX D DERIVATIONS AND SEPARABILITY 187
in K[xt, ... ,xn]· Since k is infinite, there are ct, ... ,en ink such that
(D.9)
where~= 6c1 + ... + ~ncn, i = 1, .. . ,n. Hence,~ is a primitive

element of ](.
Definition D.10. Let R be a ring and Man R-module. A mapping

D : R---+ M is a derivation of R with values in M if (i) D(r1 + r2) =
D(rt) + D(r2) and (ii) D(r1r2) = r1D(r2) + r2D(r1) for all r1,r2 in
R. The R-module of all derivations of R with values in M is denoted
Der(R,M) and the R-module of derivations which vanish on a subring
R' of R is denoted DerR·(R,M).
We note that if R is an integral domain with quotient field ]( =

]( ( R) and L is a field containing R, then every D E Der( R, L) extends
in a unique way toaD' E Der(K,L) by setting D'(rfs) = (sD(r)•
rD(s))fs 2 for rfs E K, r,s E R, s 'f 0.
Proposition D.11. Let J( = k(~) be a simple separable algebraic

extension of K. Then every DE Der(k, L) (where Lis an extension field
of]() may be extended in a unique way to a D' E Der( ](, L ).
Proof: Let p be the minimal polynomial of ~ in k[x]. If D' is

an extension of D, then, for every polynomial fin k[x], D'(J(0) =
JD(O + (D'~)(M)(~) where JD(x) = l;D(ai)xi when f(x) = l;aixi.
Since D'(O) = 0, there is at most one derivation D' of k[~] (and hence
of k( ~)) extending D. Since ~ is separable, p' (~) 'f 0 and if we set
D'~ = -pD(~)fp'(~) and D' =Don k, then D' determines an exten•
sion of D to K.
Corollary D.12. If](/ k is a finite separable extension, then every

derivation of k extends in a unique way to K.
Corollary D.l3. If]( = k(6, ... ,~n) is a separable (algebraic)

extension, then every derivation of k extends in a unique way to ](.
Proof: It is enough to show that [K : k] is finite. This will follow

by induction on n and the lemma D.14.
Lemma D.14. Let k C J( C L be a tower of fields with [K: k] =n

and [L : K] = m. Then [L : k] = mn.
Proof: Let u1, ..• , Un, v1, ... , Vm be bases of J( I k and L I]( re-
spectively. Then u;vi> i = 1, ... , n, j = 1, ... , m clearly form a basis
of Llk.
Now let J( = k(x 1 , •.• , Xn) be a finitely generated extension of k
and suppose that tr. deg K I k = v.
Definition D.l5. If]( is a separable extension of k(zb···,zv),

then { z1, ... , Zv} is called a separating transcendence basis. We say
that K I k is separably generated (or separable) if there is a separating
transcendence basis.
We now have:
Theorem D.16. If Klk is separably generated, then v

tr. degKfk = dimgDerk(K,K).
Proof: Let { z1, ... , Zv} be a separating transcendence basis and

let K' = k(z1, ... , zv) so that K is a finite separable (algebraic) exten•
sion of K'. Since every D E Derk(K', K) is determined by its values
D(z;), we have dimK•Derk(K',K) = v.* In view of corollary D.13, ev•
ery such D extends in a unique way to K and so, dimK Der k( K, K) ~
v. Now let D E Derk(K, K) and note that K = k(z1, ... , zv,~) =
K'(~) by the theorem of the primitive element. Let D(z;) = a; and
let D; be the derivations of J( with D;(zj) = Dij· Then D-'L-a;D; van•
ishes on K' and so, is an extension of the 0 derivation on K'. It follows
from corollary D.12 that D- 'L-a;D; = 0 and so, dimKDer(K,K)::; v.
We conclude with the following:
Theorem D.17. If K = k(x1, ... , xn) and k is perfect, then Klk is
* A basis is given by the "partial derivatives" D; where D;(zj) =

o;i.
APPENDIX D DERIVATIONS AND SEPARABILITY 189
separably generated.
Corollary D.l8. If k is algebraically closed and J( = k(V) is the

function field of an affine variety V, then k(V)fk is separably generated.
A proof of the theorem can be found in [Z-3] and shall not be

given here.
Problems
1.Show that if A is ann x n matrix, then (zl- A)- 1 = 1 "L7=

cPj(z)An-j / det[zJ- A] where cPj(z) is a polynomial of degree n- j.
Compute cP.i(z).
2. Prove that if H = (hi+j-1)i,j= 1 has rank n, then there is a

unique recurrence relation of the form aohj+a1hj+1 +· · ·+an-1 hj+n-1 +
h .i+n = 0 with ao, a1, ... , an-l in C. j = 1,.. .. [Hint: consider the
columns of Hone by one.]
3. Let
0 1 0 0
m
0 0 1 0
A=
b=
1
-ao -al -an-l
and x = (zl- A)- 1 b. Show that zxj = xj+ 1 ,j 1, ... , n- 1 and

x1 = 1/ det[zJ- A]. Compute det[zJ- A].
4. Let R be an integral domain. Show that R[[x]] is an integral

domain.
5. Show that if H = (hi+j-I)i,j= 1 has rank n, then Hn+.i has

rank n for j = 0, 1, . . . , and conversely.
6. Let 7/J: Hank(n,k)----+ A~n be given by 7/J(H) = (h 1 , ... ,

h 2 n). and let H be the subset of A%n given by H = {(h 1 , ... ,h 2 n):
Hn = (hi+J-di~j=l is non-singular}. Show that 7/J is a bijective map of
Hank( n, k) onto H.
7. Let A, b, c be given by equation (2.10) and let h j =

hj(bo, ... ,bn-l,ao, ... ,an-1) be given by (2.9). Show that detJ(h1,
190
PROBLEMS 191
... ,h2n;bo, ... ,bn-1,ao, ... ,an-1) is not zero if and only if det
[c' A'c' ··· A'n- 1c'] :f. 0 and that detJ(hb···,h2n;bo, ... ,bn-1,
a 0 , ••. , an- 1 ) is not zero if and only if the determinant of the matrix
bo b1 bn-1
0 bo bn-1 } n rows
ao a1 an-1 1
0 ao an-1 1 }n-lrows
is not zero.
8. Let R be a unique factorization domain and let r, s E R. An

element d of R is a greatest common divisor of r and s if (i) d is a divisor
of r and s, and (ii) if d1 is a divisor of r and s, then d1 divides d. Show
that any two elements r, s of R have a greatest common divisor which is
unique to within a unit. Generalize to any finite number of elements of R.
An element m of R is a least common multiple of r and s if (i) m is a
multiple of r and s, and (ii) if m1 is a multiple of r and s, then m 1 is a
multiple of m. Show that any two elements r, s of R have a least common
multiple which is unique to within a unit. Generalize to any finite number
of elements of R. Show that if d is a greatest common divisor and m a
least common multiple of r, s, then m · d = rs.
9. Let R = Z be the integers and let f(x) = 3x 4 + x 3 + 7,

g( x) = 2x + x + 5. What are d( x ), r( x )? Develop several "algorithms"
2
for computing d(x),r(x) for general J(x),g(x) in Z[x]. Evaluate the

"algorithms" in terms of computational efficiency.
10. Let p(z) = bo + b1 + b1z + ·· · + bn-1Zn- 1, q(z) = ao +

a1z + · · ·+ an-1Zn- 1 + zn. Show that the resultant of p and q, Res(p, q),
is given by
bo bl bn-1
0 bo bn-1 } n rows
det
ao a1 an-1 1
0 ao an-1 1 } n- 1 rows
11. Let J(x) = amxm + ·· · +

ao, g(x) = bnxn + ·· · +
bo. Show
that Res(!, g) is a homogeneous polynomial in the coefficients a;, bj of
degree m + n.
12. Show that a field k is algebraically closed (i) if and only if

every prime in k [x] has degree 1; or, (ii) if and only if every f (x) in k [x]
factors completely into linear factors; or, (iii) if and only if every f( x) in
k[:r] of positive degree has a root in k.
13. Let R, S be rings and let 1/J : R --+ S be a homomorphism.

Show that Ker 1/J = a is an ideal (Ker 1/J is the kernel of 1/J). If R = Z,
S = Z and 1/J: R-+ Sis given by 1/J(r) = 0 or 1 according as r is even
or odd, show that ·lj; is a homomorphism. What is Ker 'lj;?
14. Let a, a1 be ideals in k[x1, ... , XN] and let W, W1 be subsets

of Af:. Show that if a C a 1 , then V(a) :J V(a 1 ) and if W C W1o then
I(W) :J J(W1 ).
15. Determine the Zariski closed sets in Al and in A~.
16. Show that every ideal in k[x] is principal.
17. Give a counterexample to lemma 5.8 when k is not alge-

braically closed.
18. Let R, S be rings and let 1/J : R --+ S be a homomorphism.

Show that if p is a prime ideal in S, then 1/J- 1 ( .p) is a prime ideal in
R. Suppose that S = Rl a where a is an ideal in R and 1/J is the natural
homomorphism. Show that the prime ideals of S are the ideals of the fom1
pI a where .p is a prime ideal in R and that the maximal ideals in S are
the ideals of the form m I a where m is a maximal ideal in R.
19. Let R be an affine k-algebra and let Spm(R) = { m : m is a

maximal ideal in R}. If a is an ideal in R, let V(a) ={mE Spm(R):
a c m}. Show that the V (a) define a topology (the Zariski topology) on
Spm(R).
20. Let a be an ideal in k[X 1 , •.. ,XN] and let V(a) = {M<:
i\1< :J a}. Show that Ja = nM< where M< E V( a). If R is an affine
PROBLEMS 193
k-algebra and f is an element of R such that a(J) 0 for all a E

Hom (R, k), then f = 0.
21. Let F: Af --+-A~ be a polynomial mapping. Show that if

V c Af is irreducible, then F(V) is irreducible. What does this mean
for morphisms?
22. Let L be given by equations (8.3), (8.4), (8.5). Show that

L(Rat( n, k)) ~ Hank( n, k ). Let F be given by equations (8.6)-(8.10).
Show that F(Hank(n,k)) ~ Rat(n, k).
23. Let V be an affine variety with lJ = I(V) and let ~ E

11. Show that k[X1, ... , XN]Me jpk[X1, ... , XN]Me is isomorphic to
k[V]me = (k[Xl, ... , XN]/P)Mefp· (In other words, both define o~.v).
24. Let V be an affine variety and let ~ E V. Consider the set

of pairs (J, U) where f is regular on U (in the sense of definition 9.8)
and U is an open subset of V with~ E U. Say that (h,U1)E~(h,U2)
if h = h on U1 n U2. Show that E~ is an equivalence relation and that
the set of equivalence classes, with the natural definitions of addition and
multiplication, is a local ring isomorphic to O{,V·
25. Let V c Af, W c A~ be quasi-affine varieties and let

¢ : V --+- W be a continuous map. Then ¢ is a morphism if and only if
¢* (0¢({),w) C O{,V for all~ E V.
26. Prove proposition 10.8.
27. Let Char(n,k) = {~ E A~n: A(~) f; 0} = 1/J~ 1 (Rat(n,k)).

Then show that if x = (A,b,c) is an element of Sf,1 , then !Rx(x) =
!Rx( A, b, c) is an element of Char( n, k) and conversely, if~ is an element
of Char(n,k), then there is an (A,b,c) in Sf, 1 with !Rx(A,b,c) = f
28. Let n = 2 and let X, Y, Z be the coordinate functions

on A%(i.e. X= (X;i)L=l• etc.). Set .\(X,Y,Z) = A(!Rx(X,Y,Z)),
8(X, Y, Z) = O(!Rh(X, Y, Z)), and p(X, Y, Z) = p(!R 1(X, Y, Z)). Show
by direct computation that A= p = -8.
29. Show that the controllability and observability maps are not
surjective. What about the Hankel matrix map?
30. Let k = C, the field of complex numbers. Suppose that

(A, b, c) is controllable and consider the differential equation x = Ax+ bu.
Show that given any x 0 in en, there is a u(t) such that the solution
x(t;x 0 ,u(·)) of the differential equation with x(O;x 0 ,u(·)) = x 0 passes
through 0 for some t > 0.
31. Let k = C, the field of complex numbers. Suppose that

(A, b, c) is observable and consider the system x = Ax + bu, y = ex.
Show that, given u( ·) and y( ·) on an interval ( 0, t] with t > 0, then the
initial condition x 0 such that y(-) = cx(·;x 0 ,u(·)) can be determined.
32. Let G = G L( n, k ). Show that G is a group and that the

mappings a : G X G --+ G and j3 : G --+ G given by a(91, 92) = 9192,
!3(9) = g- 1 are regular (where G x G is viewed as a subset of A~n\
33. Let V C Af: and W C A~ be affine varieties. Show that

{O x W is isomorphic to W and that V x {rt} is isomorphic to V. What
if V and W are algebraic sets?
34. Let n = 2 and G = G L(2, k ). Consider the action of G on

A~ given by 9 ·(A, b, c) = (9A9-1, 9b, c9- 1). Let xo = (Ao, bo, co) with
Ao = [ ~ ~] , bo = [ ~] , co = [1 0]
What is the stabilizer subgroup S ( x 0 )? Let x 1 = ( A 1, b1, c1) with
What is the stabilizer subgroup S ( x 1 )? Explain the results.
35. Let R = k [Sf 1 ] be the ring of regular functions on Sf 1 .

Viewing the elements of R ~s functions from 51', 1 into Al.
show that the
action of G defines a map (or action) from G x R --+ R and that the set
of invariant elements of R forms a subring R 0 of R. In fact, show also
2 2
that the action of G on A~ +2 n defines an action from G x k[A~ + n] --+
2
k[A~ 2 + 2 n] which preserves degrees.

PROBLEMS 195
36. Let E be an equivalence relation on V and let FE = {a : a

is an E-invariant ideal in k[V]}. Consider the set {V(a): a E FE}. Does
this set define a topology on V? If so, what can you say about fin RE?
37. Let E be an equivalence relation on V and let R = k[V],

RE = k[V]E. Then there is a natural map 1/J : Spm( R) --+ Id( RE) given
by 7/J( m) = m n RE where Id( RE) is the set of ideals in RE. Show that
1/J is an invariant and that Spm(RE) is contained in the range of 7/J.
38. Let E be an equivalence relation on a variety V and let

R = k[V], RE = k[V]E. Suppose that f E RE i.e. f is an invariant.
Show that (R 1 )E = (RE) 1. If a variety Wand a morphism 1/J determine a
geometric quotient of V modulo E, show that, for each open set U0 C W,
1/J* : Ow(Uo) --+ Ov(1/J- 1 (U0 )) is a surjective k-isomorphism between
Ow(Uo) and the ring of invariants on 7j;- 1 (U0 ), Ov('¢- 1 (U0 ))E. [Hint:
treat Uo = Wf with j E k[W] first.]
39. Let R = k[X,Y,Z] (X= (X;j),Y = (Yj),Z = (Z;))

and let G = GL(n,k) act on R via g · (X,Y,Z) = (gXg-I,gY,zg- 1 ).
Show that, for fixed g, the map 1/J9 : R--+ R given by 1j;9 (p(X, Y, Z)) =
p(gXg- 1 ,gY, Zg- 1 ) is a k-automorphism of R.
40. Develop analogs of proposition 15.4 and corollary 15.5 for

the cases of Rat( n, k) and Hank( n, k ).
41. (Linear and Algebraic dependence, [Z-3]). Let X be a set and

let 2x = {A : A c X (i.e. A is a subset of X)}. A mappings : 2x --+ 2x
is a span if: (i) A C B implies s(A) C s(B); (ii) if :r E X, x E s(A),
then there is a finite A' c A with x E s(A') ; (iii) A c s(A); (iv)
s(A) = s(s(A)); (v) if x E s(A U {a}),x ~ s(A), then a E s(A U {x}).
Say that A generates X if s(A) = X; that A is free (independent) if
a f/ s( A - {a}) for all a E A; and that A is a basis if A is free and
generates X. Show that A is a basis if and only if A is a minimal set of
generators or A is a maximal free subset. Show that (a) every free set A
extends to a basis; (b) every set of generators contains a basis; and, (c) if
X has a finite basis, then all bases are finite and have the same number of
elements. Show that if X is a k-vector space and s( A) = spand A}. then
s is a span. Show that if J( j k is field extension and, for A c J(, s( A) is
the algebraic closure of k( A) in J(, then s is a span. Interpret the results
in terms of transcendence for this case.
42. Give an example which shows that V1 < V2 does not imply
dimk V1 < dimk V2.
43. Let R be a ring. Show that R, R[x], R[[x]] are R-modules.
44. Let R be a ring and let Rn = REB·· ·EBR (n summands). Prove

that M is a finitely generated R-module if and only if M is isomorphic to
a quotient of Rn.
45. Let M be a finitely generated R-module, a be an ideal in R,

and ¢ : M ___. M an R-module morphism such that ¢( M) C aM. Show
that ¢n + a 1 ¢n-l + · · · + an = 0 with ai E a (cf. proof of proposition
16.19). Show that if b is an ideal in R with bM = M, then there is an
r E R such that r - 1 E b and r M = 0.
46. Let P be a multiplicatively closed set in the integral domain

R. If a is an ideal in R, then R p a is a proper ideal in R p if and only if
a n P = ¢ (i.e. a does not meet P).
47. (Field Polynomial, Norm and Trace [L-1], [Z-3]). Let k c J(

be fields and suppose that J( is a finite extension of k i.e. J( is a finite
dimensional vector space over k with [K : k] = dimkf( = n. In other
words, J( is an algebraic extension of k of finite degree. If ~ E J(, then
the mapping A~ : J( ___. J( given by A~( w) = ~ · w, w E J( is a k-linear
mapping. Let w 1 , ... , wn be a basis of J( / k and let A~ be the matrix of A€
with respect to this basis. Set F{ (X) = det ( x ·I- A~) and call F~ (X) the
field polynomial of~ over k. Show that F€ (X) is independent of the choice
of basis and that F€(0 = 0. Let TrK;k(~) =trace of A€ and NKfk(~) =
determinant of A~ and call TrK;k(~) the trace of~ over k and NI<;k(~)
the nonn of~ over k. Show that (i) TrK;k(~ + ry) = Tri<fk(~) +TrK/k(ry);
(ii) NK/k(~TJ) = NK/k(~)NK/k(TJ), and (iii) NI<fk(0 =j; 0 if~ =j; 0.
Suppose that k C J( C L is a tower of finite extensions with [K : k] = n,
[L : K] = m. Show that [L : k] = nm (cf. sec 4). If~ E K, then ~
has a field polynomial FcKfk(X) as an element of J( and ~has a field
polynomial Ff.,Lfk(X) as an element of L. Show that
PROBLEMS 197
and hence, that TrLfk(~) = mTrK/k(~) and NL/k(() (NK/k(~))m

= (NK/k(~))[L:k]_ Apply this to the situation k c k(() c L to show that
the field polynomial of ~ is a multiple of the minimal polynomial of ~
Interpret this for norms.
48. Let R be a ring. Let p1, ... , Pn be prime ideals in R and let
a be an ideal contained in U!P;. Show that a C Pi for some i. [Hint: use
induction on n in the form a ct. Pi. i = 1, ... , n implies that a ct. Uf= 1 p;.]
49. Prove lemma 18.16. [Hint: apply the Noether Normalization

lemma 16.43.]
50. Let R be an integral domain with quotient field J( and let

R' be the integral closure of R in K. If f, g are monic polynomials in
K[X] such that fg E R'[X], then j,g E R'[X]. [Hint: let K 1 be a field
with ]( C ](1 such that J,g split into linear factors i.e. f = f1(X- ai).
g = f1(X -j3j). a;, j3j E K1. Then O'i, j3j are integral over R' being roots
of fg and so, the coefficients off and g are integral over R'.] Can you
prove a generalization when R is not an integral domain? ([A-2], [B-2]).
51. Prove the results of 19.16-19.22.
52. Show that if f E S = RG, then ( Sf 1 ) f is an invariant open

set in Sf,1 . Show also that iff is a prime in R ~ k[X,Y,Z] and (Sf,dJ
is invariant, then f E S i.e. f is invariant (cf. [B-7]).
53. Show that the maps a 1 , a 2 of (19.23) are regular and that
f E S if and only if ( a1 o a2 )*(!) = f.
54. Let D C A~ 2 + 2 n be V(Xij : i =f j, i,j = 1, ... n) so
that I(D) = (Xij). i =f j. Show that RD = k[D] is isomorphic with
k[Xn, ... , Xnn, Y, Z]. Is D irreducible? What is the dimension of D?
55. Let U1 = (Xu, W1). Uz = (X22, W2). U3 = (X33, W3).

Determine the generators of k[U] 53 explicity. Show that SoD n k[U] 53 =f
k[U] 53 (using 19.27) and that SD =SoD by direct calculation. Show also
that SoD is generated by O(Xn), O(Xt1), O(Xr1), O(Wt), O(Xu W1).
and O(Xi 1 Wt). Show also that O(X11 W[) is not an element of SD.
56. Prove that Hank( n, k) and 1{ are isomorphic.
57. Show that A( N, k) is a group and that the maps p

A(N,k) x A(N,k) --+ A(N,k) and B : A(N,k) --+ A(N,k) given by
p(a 1 ,a 2 ) = a 1 a 2 , B(a) = a- 1 are regular. Show also that A(N,k) acts
on Af: (a fortiori on k[X1 , ... , X N]) and that the action on k[X1 , ... , X NJ
preserves degrees.
58. Show that if V is an affine algebraic set, then so is a(V) for

a E A(N, k) and that V and a(V) are isomorphic.
59. Let V = V(C 1 - a 1 , ••. ,Cm - am) be a linear variety in

A f:. Prove that there is an affine transformation a = (g, ~0 ) such that if
Z; = (gX + ~0 );, then Z; = C;(X)- a; fori= 1, ... , m.
60. Let V be an affine variety with jJ = I(V). Let fi, ... , fr be

a basis of J:l and let ~ E v. Show that rank J(!I' ... 'fr; x1' ... X N)
( ~) is independent of the choice of basis of J:l.
61. Let 'lj; : V --+ W be a morphism of affine varieties. Show that

the tangent map is a homomorphism. What form does the tangent map
take in terms of bases of I(V) and I(W)?
62. Determine the set Sing(V) for V = V(X[ - Xr) in A~.

Determine Tv,~ for all points ~of Y.
63. Prove proposition 20.40 in detail.
64. Let r be a closed subgroup of G L( n, k ). Show that there is

a unique irreducible component r e of r containing the identity I and that
r e is a closed normal subgroup of r. Use this to show that the feedback
group is irreducible. Suppose that r is a closed irreducible subgroup of
G L( n, k ). Show that r is nonsingular and hence, that the feedback group
is nonsingular.
65. Prove that range [b Ab · · · An- 1 b] range [b (A -

ba- 1 K)b · · · (A- ba- 1 I()n- 1 b].
66. Show that N f and G f are closed linear subvarieties of rf

PROBLEMS 199
and that dimNf = n, dimG f = n 2 + 1. If 1 = [I, K, 1] E Nf, then show

that lo = 1- [I, 0, 1] is nilpotent with ~~+l = 0 (as matrices). (Such a 1
is called unipotent ([H-8], [S-4]).
67. Show that .6.(f hx) = .6.(N1 ,x) = .6.(I'J>Igx) = .6.(N1 ,

1 9 x) for lg E G 1 .
68. Show that if k has characteristic 0, then .6.x is smjective if

and only if Tx is surjective. What is the situation if k has characteristic
pi: 0?
69. Prove that if X E M(n,1;k) andY E M(1,n;k), then
det[I +X · YJ = 1 + Y ·X using the Binet-Cauchy fonnula ([W-1]).
70. Prove that if X E M(n,1;k) andY E M(l,n;k), then

det[I +X· YJ = 1 + Y ·X by showing that X· Y is similar to a diagonal
matrix with entries [Tr(X · Y),O,···,OJ and noting that Tr(X · Y) =
Tr(Y ·X)= Y ·X ([W-1]).
71. Prove corollary 21.22.
72. Prove that Tx(K) = Tx(O) + tx(K)Nx(K) where Nx(K) is

given by 21.28.
73. Suppose that k has characteristic p -:j: 0 with p < n. Can Tx

be surjective?
74. Show that if k = C is the complex numbers, then the

map .6.x(K) = det[zi- A+ bKJ is surjective if and only if the map
Ax(K) = (A.l,···,A.n) is surjective where A.;= A.;(x,K) are the roots of
.6.x( K)( z) = 0. What is the situation when k = R is the real numbers?
75. Prove corollary 22.11 in detail ([D-1], [S-4]).

References
(A-1] Athans, M. and Falb, P.L., Optimal Control: An Introduction to the

Theory and its Applications, McGraw-Hill, New York, 1966.
(A-2] Atiyah, M.F. and MacDonald, I.G., Introduction to Commutative
Algebra, Addison- Wesley, Reading, MA, 1969.
(B-1] Birkhoff, G. and Maclane, S., A Survey of Modern Algebra, Revised
Edition, Macmillan, New York, 1953.
(B-2] Borel, A., Linear Algebraic Groups, Benjamin, New York, 1969.
(B-3] Bourbaki, N. Algebre Commutative, I-VII, Hermann, Paris, 1961-
65.
(B-4] Brockett, R. Finite-Dimensional Linear Systems, Wiley-Intersci•
ence, New York, 1970.
(B-5] Byrnes, C. "The moduli space for a linear dynamical system" in
Geometric Control Theory (C. Martin, R. Hermann, eds. ), Math.
Sci. Press, Brookline, MA, 1977.
(B-6] Byrnes, C. and Falb, P.L., "Applications of algebraic geometry in
system theory", Am. J. Math., Vol. 101, 1979.
(B-7] Byrnes, C. and Hurt, N. "On the moduli of linear dynamical
systems", Ad v. in Math: Studies in Analysis, Vol. 4, 1978.
(D-1] Dieudonne, J. Cours de Geometrie Algebrique, 1, 2, Presses Uni•
versi taires de France, 1974.
(D-2] Dieudonne, J. and Carrell, J ., "Invariant theory, old and new",
Advances in Math, 4, 1970.
(F-1] Falb, P.L., Linear Systems and Invariants, Lecture Notes, Control
Group, Lund University, Sweden, 1974.
(F-2] Fogarty, J., Invariant Theory, Benjamin, New York, 1969.
(F-3] Fulton, W., Algebraic Curves, Benjamin, New York, 1969.
(G-1] Gantmacher, F.R., Theory of Matrices I, II, Chelsea, New York,
1959.
200
REFERENCES 201
[H-1] Haboush, W.J., "Reductive groups are geometrically reductive",

Ann. of Math, Vol. 102, 1975.
(H-2] Hartshorne, R., Algebraic Geometry, Springer-Verlag, Berlin•
Heidelberg-NewYork, 1977.
(H-3] Hazewinkel, M., "Moduli and canonical forms for linear dynami•
cal systems III: the algebraic-geometric case", in Geometric Con•
trol Theory (C. Martin, R. Hermann, eds.), Math. Sci. Press,
Brookline, MA, 1977.
[H-4] Hazewinkel, M. and Kalman, R.E., "On invariants, canonical
forms and moduli for linear constant finite-dimensional, dynam•
ical systems" in Lecture Notes Economics-Math System Theory,
Springer- Verlag, Berlin-Heidelberg-New York, 1976.
(H-5] Hermann, R. Linear Systems Theory and Introductory Algebraic Ge•
ometry, Math. Sci. Press, Brookline, MA, 1974.
(H-6] Hermann, R. and Martin, C., "Applications of algebraic geomtry
to system theory-Part I", IEEE Trans. Aut. Cont., AC-22, 1977.
(H-7] Hochschild, G., Basic Theory of Algebraic Groups and Lie Algebras,
Springer-Verlag, Berlin-Heidelberg-New York, 1981.
[H-8] Humphreys, J., Linear Algebraic Groups, Springer-Verlag, Berlin•
Heidelberg-New York, 1975.
[J-1] Jacobson, N., Lectures in Abstract Algebra I, II, III, Van Nostrand,
Princeton, NJ, 1953.
(K-1] Kalman, R.E., "Global Structure of classes of linear dynamical
systems" in Lectures, NATO Adv. Study Inst., London, 1971.
[K-2] Kalman, R.E., Falb, P.L. and Arbib, M.A., Topics in Mathematical
System Theory, McGraw-Hill, New York, 1969.
[K -3] Kendig, K., Elementary Algebraic Geometry, Springer-Verlag, Berlin•
Heidelberg-New York, 1977.
[K-4] Kunz, E., Inroduction to Commutative Algebra and Algebraic Geom•
etry, Birkhauser-Boston, Boston, 1985.
[L-1] Lang, S., Algebra, Addison-Wesley, Reading, MA, 1971.
[M-1] Matsumura, H., Commutative Algebra, 2nd Ed., Benjamin, New
York, 1980.
(M-2] Mumford, D.B., Introduction to Algebraic Geometry (Preliminary
Version of First 3 Chapters), Lecture Notes, Harvard University,
1965.
[M-3] Mumford, D.B., Geometric Invariant Theory, Springer-Verlag,

Berlin-Heidelberg-N ew York, 1965. [2nd Enlarged Edition, with
J. Fogarty, 1982].
[M-4] Mumford, D. and Suominen, K., "Introduction to the theory of
moduli" in Proc Fifth Summer School in Math., Oslo, 1970.
[N-1] Nagata, M., "Invariants of a group in an affine ring", J. Math
Kyoto, 1964.
[N-2] Neeman, A., "Topics in Algebraic Geometry", Thesis, Harvard
University, 1983.
[N-3] Northcott, D. G., Affine Sets and Affine Groups, Cambridge Univ.
Press, Cambridge (Eng.), 1980.
[R-1] Rosenbrock, H., State-space and Multivariable Theory, Wiley•
Interscience, New York, 1970.
[S-1] Samuel, P., Methodes d'Algebre Abstraite en Geometrie Al-
gebrique, Springer-Verlag, Berlin-Heidelberg-New York, 1967.
[S-2] Shafarevich, I.R., Basic Algebraic Geomtery, Springer- Verlag,
Berlin-Heidelberg-N ew York, 1977.
[S-3] Sontag, E., "Linear systems over commutative rings: a survey",
Richerche di Automatica, Vol. 7, 1976.
[S-4] Springer, T.A., Linear Algebraic Groups, Birkhauser-Boston , Bos•
ton, 1981.
[W-1] Wolovich, W., Linear Multivariable Systems, Springer- Verlag, Ber•
lin-Heidelberg-New York, 1974.
[W-2] Wonham, W.M., "On pole assignment in multi-input controllable
linear systems", IEEE Trans. Aut. Cont., AC-12, 1967.
[W-3] Wonham, W.M. and Morse, A.S., "Feedback invariants of linear
multi variable systems", Automatica, Vol. 8, 1972.
[Z-1] Zariski, 0., "A new proof of the Hilbert Nullstellensatz", Bull.
Am. Math. Soc., Vol. 53, 1947.
[Z-2] Zariski, 0., "The concept of a simple point of an abstract alge•
braic variety", Trans. Am. Math. Soc., Vol. 62, 1947.
[Z-3] Zariski, 0 and Samuel, P., Commutative Algebra I, II, Van Nos•
trand, Princeton, NJ, 1958, 1960.

Methods of Algebraic Geometry in Control Theory - Part I - Scalar Linear Systems and Affine Algebraic Geometry (PDFDrive)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Methods of Algebraic Geometry in Control Theory - Part I - Scalar Linear Systems and Affine Algebraic Geometry (PDFDrive)

Uploaded by

Copyright:

Available Formats

Modern Birkhäuser Classics

Many of the original research and survey monographs, as well as textbooks,

Reprint of the 1990 Edition

ISSN 2197-1803 ISSN 2197-1811 (electronic)

Library of Congress Control Number: 2018952484

© Springer Nature Switzerland AG 2018

Library of Congress Cataloging-in-Publication Data

Printed on acid-free paper.

© Birkhauser Boston, 1990

Camera-ready copy prepared by the author using TeX.

Control theory represents an attempt to codify, in mathematical terms, the principles

The overall goal of these notes is to provide an introduction to

In order to establish the second transition theorem, which is

concept is that of a "geometric quotient" and we then state the Ge•

Let us begin by considering single input, single output linear sys•

Definition 1.1. A scalar linear system over e is any one of the

Example 1.2. Let A =h be the 2 X 2 identity matrix and let

b= [~],c=[1 0 ]. Then IJA(z) = (z-1)- and '2:.j=

( cAi+i- 2 b)iJ= 1 has finite rank by virtue of the Cayley-Hamilt on theo•

with ao,a1, ... ,an-1 inC, j = 1, .... If we let

b="n[!],c =[l o ... 0]

where Hn = (hi+i-d~j= 1 , then (A,b,c) is a linear system with

Theorem 1.4 ( Hankel [G-1] ). Let f(z) = 2::_: 1 hiz-i be a

Proof 1: Suppose that f(z) is rational. Then f(z) = p(z)jq(z) =

Conversely, if rank H f = n, then there is a unique recurrence

Proof 2: Suppose that f(z) is rational with f(z) = p(z)jq(z) =

Conversely, if rank H f is finite, then, as before, there is a unique

recurrence relation of the form (1.5). If we let

b~Hn[!1 rh,h2 1' c = [1

then cA.i-lb = hj, j = 1, .... But c(zi- A)- 1 b = z- 1 (!- A/ z)- 1 b =

Based on the intuition developed for the complex number case,

Definition 2.1. A formal power series in one variable over R is an

U + V =(Uo + Vo, U1 + VI, ... )

Definition 2.3. Let z- 1 be an indeterminate over R. An element

If J( z- 1 ) is a strictly proper meromorphic function on R, then

Definition 2.4. A scalar linear system over k is any one of the

Theorem 2.5 (Hankel). Let f(z) = 2:~ 1 hjz-i and let H f

We call the rank of H 1 , the degree of f, and we note that if rank

Definition 2.6. Let Rat(n,k) = {f(z): f(z) a strictly proper ra•

In view of Theorem 2.5, there is a natural map L : Rat(n,k)---+

has a unique solution Xj = -aj, j = 0,1, ... ,n- 1 with the aj

non-singular}. With Rat( n, k) and Hank( n, k) viewed in this way, the

hj(f) = hj(bo, ... ,bn-1,ao, ... ,an-1) = cAi- 1b (2.9)

,b~ m,,~ [bo, ...

so that the hj are polynomials in (bo,b1,ao,at). If J(h1, ... ,h4;

The significance of this observation will become clear in the sequel.

We can also give an explicit description of the map P (actually

for j = 0, 1, ... , n - 1. Motivated by the second proof of Hankel's

,b~H [!1 ,,~[!

so that cAJ- 1b = hj for j = 1, .... We observe that det(zi- A) =

for r = 0, 1, ... , n- 1, and where we have set an = 1 for convenience.

Example 2.15. Let n = 2. Then the explicit formula (2.14)

ao =- (h~- h2h4)j(h1h3- h~)

so that the ai and bi are polynomials in the hi divided by (at most)

(with the bi and ai as functions of the hj)· Again, this observation

Now, in order to provide an appropriate algebraic structure for

Definition 3.1. An integral domain R is a unique factorization

If f(x) E R[x] where R is a unique factorization domain, then

Example 3.2. Let R = Z be the set of integers. The polynomial

Lemma 3.3. (Gauss) Let R be a unique factorization domain. Iff( x ),

Proof: Since f = c(f)h, 9 = c(9)91 and !9 = c(f)c(9)f191, it is

Corollary 3.4. If g( x) divides af( x) with a E R and g primitive,

Proof: If af(x) = g(x)h(x), then ac(J) = c(g)c(h) = Ec(h) where