Professional Documents
Culture Documents
Linear Algebra
Linear Algebra
Linear Algebra
To
Contents
1 3 3 5 7 13 16 19 19 21 24 28 33 38 41 44 48 51 51 54 56 59 61 67
Two Linear Equations with Two Variables Basic Notations for Systems of Linear Equations Elementary Transformations of Systems of Linear Equations and Elementary Row Transformations of Matrices Methodes for Solving Homogeneous and Nonhomogeneous Systems of Linear Equations Two Problems Vector Spaces
Chapter 2. 1. 2. 3. 4. 5. 6. 7. 8. 9. Fields
Vector Spaces Linear Combinations and Basis of a Vector Space Linear Dependence and Existence of a Basis The Rank of a Finite System of Vectors The Dimension of a Vector Space Direct Sum and Linear Complements Row and Column Rank of a Matrix Application to Systems of Linear Equations Linear Maps
Denition and Simple Properties Isomorphisms and Isomorphism of Vector Spaces Dimension Formula for Linear Maps The Vector Space
HomF (V, W )
Linear Maps and Matrices The Matrix Product The Matrix Description of Isomorphisms (Again) Change of Bases Equivalence and Similarity of Matrices The General Linear Group Application to Systems of Linear Equations (Again) Determinants
EndF (V )
Chapter 4. 1. 2. 3. 4.
The Concept of a Determinant Function Proof of Existence and Expansion of a Determinant with Respect to a Row Elementary Properties of a Determinant The Leibniz Formula for Determinants Some Terminology about Sets and Maps
Appendix A. 1. Sets
iii
2.
Maps Fields with Positive Characteristic Zorn's Lemma and the Existence of a Basis A Summary of Some Algebraic Structures. About the Concept of a Rank
iv
Introduction
Linear algebra is the branch of mathematics concerned with the study of vectors, vector spaces (also called linear spaces), linear transformations, and systems of linear equations. Vector spaces are a central theme in modern mathematics; thus, linear algebra is widely used in both abstract algebra and functional analysis. Linear algebra also has a concrete representation in analytic geometry and it is generalized in operator theory. It has extensive applications in the natural sciences and the social sciences, since nonlinear models can often be approximated by a linear model. We will begin our studies by studying systems of linear equations. Without becomming to formal in the notation and language we will study the basic properties of the solutions for homogeneous and nonhomogeneous systems of linear equations. We will get known to the Gaussian algorithm for solving systems of linear equations. This algorithm will re-occur repeatedly in this lecture note. Towards the end of this rst chapter two problems will in a natural way catch our attention. In order to solve them we need to begin to formalize the observations we made sofar. The formalisation will begin by extracting the esential properties of the numbers we have used in the rst chapter. This will lead to the concept of a eld (Denition 2.1). Next we will formalize the properties which the solutions of a homogeneous system of equations posesses: the sum of two solutions of a homogeneous system of equations is again a solution this system of equations, and the same is true if we multiply a solution of such a system of equations by a number (that is an element of a eld). This will lead to the concept of a vector space (Denition 2.4). Roughly spoken a vector space over a eld is a set
arbitrary elements and where we can multiply any element by scalars of the eld and such that this addition and scalar multiplication satises certain rules which seem natural to us. After we have made these essential denitions we will begin to study vector spaces more in detail. We will encounter basic but very important concepts of linear algebra, amongst others:
linear combinations of vectors, basis of a vector spaces, linear dependence of vectors, the rank of a system of vectors (and related concepts), the dimension of a vector space.
Maybe one of the most essential results will be the theorem about the existence of a basis. It takes just
till the foundations of the mathematics: Every vector space has a basis. We will proof it only in the special case of nite dimensional vector spaces as the more general result will need some heavy machinery of axiomatic set theory. Though for completeness we have included this proof in the appendix of this lecture notes (together with an more detailed explanation about its importance; see Appendix C).
1
INTRODUCTION
Towards the the end of this second chapter we will nally be able to answer the problems answer the problems which did arise in the end of the rst chapter. In the third chapter we will then start to study the relation ship between. We will introduce the concept of a linear map between vector spaces (Denition 3.1). The whole chapter is devoted to the study of these kind of maps. One of the main theme of this chapter will be the matrix description of linear maps between nite dimensional vector spaces. We will explore the relation ship between matrices and linear maps and what we can all conclude from that. Two theorems will provide us with the necessary information:
HomF (V, W ) of all linear maps between a n dimenV and a m-dimensional vector space W is isomorphic m,n as vector spaces to the the vector space of all mn-matrices F . (Thesional vector space
orem 3.27) The endomorphism ring
isomorphic as
F -algebras
(Theorem 3.35) The proper understanding of these two theorems might take time but they are essential in order to really understand linear algebra. the third chapter will be: Other important topics of
isomorphism and isomorphism of vector spaces, rank of a linear map, dimension formla for linear maps, the general and the special linear group, equivalence and similarity of matrices, normal form of matrices upto equivalence.
During the third chapter we will encounter that every invertible can be written as a product
n n-matrix
A = SD
where
uniqueness of this decomposition. It will turn out that the theory developed upto the third chapter is not enough to give a proper answer to this problem. We will need a new concept and this will lead to the the denition of a determinant function. The fourth chapter will then be devoted to the rst studies of determinant functions. First we introduce the new concept and show what hypotetical properties a determinant function would have. It will turn out that the detreminant function in case it exists must be unique. This will be the reason why we will later be able to give an armative answer about the uniqueness of the above decomposition. But we will rst have to show that a determinant function exists and this is done in the second section of the chapter about determinants. When we are nally convinced about the existence of determinants we will study in the remaining part of the chapter some basic properties of the determinant function. The chapter will nish with the presentation of the Leibniz formula which shows the beauty and symmetries of the determinat function (Theorem 4.28).
CHAPTER 1
Following closely [
Lor92]
equations where we will encounter the rst time linear structures. chapter we will then study linear structures in a more general setting.
ax + by = e cx + dy = f.
Here which values of (1) concern is
a, b, c, d, e and f are numbers and x and y are variables. Our x and y satisfy the two equations above simultanously.
In order to avoid triviality we may assume that some of the numbers of the on the left hand side of (1) are not equal to zero. Let us assume that can substract get
a = 0.
Then we
c/a
times the rst equation of (1) from the second equation and we
ax + by = e dy=f ,
with (2)
d = d bc/a
and
f = f ec/a.
of linear equations again the rst one by adding equations. If some values for
c/a
second equation of (2). We say that (1) and (2) are equivalent systems of linear
and
then they also satisfy the linear equations of (2) simultaneously. And the converse is also true. In general, equivalent systems of linear equations have exact the same solutions. Note that the system of linear equations (2) is more easy to solve than the the rst system. This is because in the second equation only appeares one variable instead of two, namely equivalent with
y.
Since
a=0
(3)
Thus the solveability of (1) depends very much on the fact whether the number (4)
Case 1:
(a, b, c, d) = 0.
y=
y=
It follows after a short calculation that since we have made the assumption
a=0
that
x=
ed bf (e, b, f, d) = ad bc (a, b, c, d)
(3) follows that the
Case 2:
e and f of the righthand side of (3) can be choosen such that af ec = 0 and in this case there exist no y satisfying the equation (3). This happens for example if f = 1 and e = 0, because then af ec = a = 0 since we assumed that a = 0. But in the case that af ec = 0 we can choose the value for y freely and y together with e by x= a
constants is a solution for the system of linear equations (1). Thus the system of linear equations (1) has a solution but this solution is not unique. Collecting the above observations it is not dicult to prove the following result:
Proposition 1.1.
variables
Consider the system of linear equations (1) in the two unknown Then we have the following three cases:
and
y.
Case 1: Case 2:
a = b = c = d = 0. e=f =0
and in this case any pair
and
is a sollution.
a, b, c, d are equal 0, but (a, b, c, d) = 0. (a, e, c, f ) = af ec = 0 or (e, b, f, d) = (a, e, c, f ) = (e, b, f, d) = 0, then (1) is solveable and
we can always achive by exchanging the equations
or renaming the unknown variables) that all the solutions of (1) are given
x=
1 (e bt), a
number.
x=t
Case 3:
where
x=
and
y=
(a, e, c, f ) . (a, b, c, d)
The above considerations give already an exhaustive description of what can happen when solving a system of linear equations, even if it consists of more equations and more variables: either the system will have exactly one sollution, or many solutions or no solutions. It will turn out that whether there exist a unique solution or not will always solely depend on the left hand side of the equations, that is on the coecients of the variables (in the above case of the system of linear equations (1) these are the numbers
a, b, c
and
d).
(a, b, c, d)
equations in the
The numbers
the coecients of
(7)
Such a scheme of numbers is called a matrix. If one wants to emphasis the dimension of the above matrix, then one can also say that the scheme of numbers (7) is a
m n-matrix.
...
m 1-matrices.
m-tuple.
m-tuple
b1 , . . . , bm
b1 b2 b := . . . bm
Note the use of the symbol := in the previous two equations. It does not denote an equality but rather denotes the denition of what is on the left side of the symbol. For example
x := 2
means that
c1 d1 c1 + d1 c2 d2 c2 + d2 . + . := . . . . . . . cm dm cm + dm
and likewise the multiplication of an
2. m-tuples
of numbers by
by
the
eld R.
F.
When we talk in the following of number one may think of real numbers, that is elements of But all our considerations will be true in the case where
is replaced by an arbitrary
eld
(What a eld
Using this notation we can write the system of linear equations (6) in the following compact form:
x 1 v1 + x 2 v2 + + x n vn = b
We call the
(8)
n-tuple x1 x2 x := . . . xn
x1 , . . . , x n
of the system of linear equations. We will give the class of system of linear equations where the numbers b1 , . . . , bm are all equal to equations.
2
b1 , . . . , b m
by
the
0 0 0 := . . . 0
the
m-tuple
only.
m-tuple
Then the homogeneous system of linear equations which is assoziated with the system of linear equations (6) or equivalently (8) can be written as
x1 v1 + x2 v2 + + xn vn = 0.
(9)
Proposition 1.2.
ear equations (6) or equivalently (8) if one adds to a specic solution all the solutions of the homogeneous system (9) associated with it.
x of the system
Proof. In order to avoid triviality we may assume that the system (6) has a
solution
x.
Then if
x+x
x and x
x x
of the
M the set of all solutions of (6). In the case that (6) is M = (where denotes the empty set). If M0 denotes the
set of all solutions of the homogeneous system of linear equations (8) assoziated with (6), then we can write the content of Proposition 1.2 in the compact form
M = x + M0
2
(10)
nonhomogeneous.
3. ELEMENTARY TRANSFORMATIONS
where
x + M0 := {x + x : x M }. M0
Now using (9) one can make the following observations about the solutions of an given homogeneous system of linear equations: If and: If
x M0
and
x M0
then also
x + x M0 , cx M0 .
(11)
x M0
and
(12)
be just any set but must obey the requirements (11) and (12). If we denote by
n-tuples
M0 of some homogeneous system of linear equations cannot Fn n n then we call a subset U F a subspace of F if we have x, x U x + x U
(13)
and
x U cx U
the following compact form.
c).
(14)
Thus, if we use the above notation we write the observations (11) and (12) in
Proposition 1.3.
equations in
The set of M0 all solutions of a homogeneous system of linear n unknown variables is a subspace of F .
Note that the requirement Therefore we call the the following result.
n-tuple
M0 = is automatically satsied since the n-tuple 0 n-variables. 0 the trivial solution of a homogeneous system of
liear equations. Now one gets as a consequence of Proposition 1.2 or equation (10)
Proposition 1.4.
a unique solution if and only if the associated homogeneous system of linear equations has only the trivial solution.
3. Elementary Transformations of Systems of Linear Equations and Elementary Row Transformations of Matrices
In this section we shall introduce a simple way to solve systems of linear equations: the Gaussian elimination algorithm. in
3
n variables.
Our aim is similar as in Section 1 to convert this system of linear Of course one step should transform a
equations step by step into a system of linear equations which where the question about it solveability is easier to answer. given system of linear equations
which is
S.
x
for every possible
is a solution of
S x
is a solution of
n-tuple x.
terested in keeping the transformations in each step as simple as possible. We will allow the following three elementary transformations : (I) Adding a multiple of one equation to another equation. (II) Exchanging two equations with each other. (III) Multiplying one equation with a non-zero number.
Named after the German mathematician and scientist Carl Friedrich Gauss, 17771855.
One veries immediately that each of those transformations transform a system of linear equations into an equivalent one. Now the system of linear equations (6) is completely determined by the extended coecient matrix
a1n a2n
. . .
amn
b1 b2 . . . bm
(15)
m (n + 1)-matrix.
the above matrix, then we will call it the simple coecient matrix of the system of linear equations (6).) The transformations of type I, II and III will result in a natural way in row transformations of the extendet coecient matrix (15). To simplify notation we will consider in the following elementary row transformations for arbitrary matrices. Therefore let
(16)
ui
be
pq -matrix. We call the entries cij the coecients of the matrix C . u1 , u2 , . . . , up the rows of the matrix C . That is for i = 1, 2, . . . p the (horizontaly written) q -tuple ui := (ci1 , ci2 , . . . , ciq ).
An elementary row transformation of a matrix one of the following three transformations. (I) Adding a multiple of one row of (II) Exchanging two rows of (III) Multiplying one row of
In order to simplify notation we will subsequently also allow switching columns as an elementary operation on matrices. Note that switching columns of the extended coecient matrix (15) which do not involve the variables
bi
x1 , . . . , x n
p q -matrix C
as in (16).
are zero then there is nothing to do. Thus we may assume that some coecient In case this coecient is in the
i-th
i = 1,
we use a
elementary row transformation of type II to switch the i-th with the rst row. Thus in the rst row of the so optained matrix which we call again least one coecient
there is now at
c1k
k -th
with
the rst column and thus we may assume from the beginning that
c11 = 0.
Now we add for every
i2
the
ui .
u1
3. ELEMENTARY TRANSFORMATIONS
c11 0 C = . . . 0
Below the coecient
c11
of the matrix
below the rst row are zero, then there is nothing to do. Thus we may assume for similar reasons as above that
c22 = 0.
Now we apply the same procedure as we had applied to the matrix ginning to the columns away the rst row
C in the beu2 , . . . , up , that is the matrix we optain from C by leaving u1 = u1 . For every i 3 we add the (ci2 /c22 )-times of u2 to
which we optain, has now the rows:
ui .
The matrix
is of the form
c11 0 C = 0 . . . 0
form:
c12 c22 0 0
D=
d11 0 0 0
. . .
d22 0 0
. . .
d33 0
.. .
d44
.. . .. .
(17)
... 0
drr
where
0rp
and
and
0rq
(18)
dii = 0
(1 i r).
(19)
Note that the two zeros below the horizontal line of the matrix (17) denote zero-
matrices, that is matrices where every coecient is equal to zero. Likewise the
D which we do not In the case that r = 0 it means that D is the p q -zero-matrix. it is clear how the cases r = p or r = q have to be understood.
symbolizes arbitrary entries in the matrix
Proposition 1.5.
transform
Let
be an arbitrary
p q -matrix.
mentary row transformations of type I and type II and suitable column exchanges
into a matrix
10
dii = 0 (1 i r) d11 0 0 0
. . .
0 d22 0 0
. . .
0 0 d33 0
.. .
...
..
...
0 0
. . .
. .. .. . .
d44
.. .
. . .
(20)
... 0
0 drr
The
r r-matrix in left upper part of the matrix (20) has beside the coecients 1ir
on the diagonal only zeros. We call such a matrix a diagonal matrix. And if we use row transformations of type III and multiply for every the
i-th
1/dii
1 0 0 0
. . .
0 1 0 0
. . .
0 0 1 0
.. .
...
..
... 0 0
. . . .. .. . . . . .
1
.. .
(21)
... 0
0 1
where the left upper part of the matrix (21) is a diagonal ones on the diagonal. symbol We call this matrix the
only case
there is no danger of confusion we may also denote the identity matrix just by the
I .)
Consider for example the following
Example.
4 5-matrix, 0 2 1 1 C
1 3 5 2 3 9 10 1 C := 0 2 7 3 2 8 12 2
Here the upper left coecien
c11 = 1
of the matrix
0.
Thus we can begin to add suitable multiples of the rst row to the rows below. In our case this means we substract substract
3-times
2-times 1 3 0 2
3 9 2 8
5 10 7 12
2 1 3 2
0 2 1 1
1 3 0 0 0 2 0 2
5 2 0 5 5 2 7 3 1 2 2 1
3. ELEMENTARY TRANSFORMATIONS
11
1 3 0 2 0 2 0 0
5 2 0 2 2 1 7 3 1 5 5 2
Now we can substract the second row once from the third and then in the next step we add the third row to the last and we optain:
1 0 0 0
3 2 0 0
5 2 0 2 2 1 5 5 2 5 5 2
1 3 5 0 2 2 0 0 5 0 0 0
2 2 5 0
0 1 2 0
3/2-time
2/5-th
1 0 0 0 1/2
0 2 0 0
2 0 5 0
2/5-th 3 4 5 0
of the third row to the rst and then in the last step and the third row by
1 0 0 0 2 0 0 0 5 0 0 0
7/10 9/5 2 0
1 0 0 0
0 1 0 0
3 2 1 0
Thus we have optained nally the form (21), and this by the way without the need of column transformations. This concludes the example. Note that if one does not exchange columns than it is still possible to transform any
p q -matrix into a matrix 0 . . . 0 d1,j1 0 ... ... 0 ... 0 ... ... ... ... . . . 0 ... ... ... ...
...
(22)
where
di,ji = 0
for
1 i r.
In particular it is
Now if one applies Proposition 1.5 and the forth following considertations to the simple coecient matrix or (22) (with of a general system of linear equations (6), then we see that we can transform this matrix into a matrix of the form (17) or (20) or (21)
0 r m, n)
or III and column exchanges (in the rst three cases). If one applies the same row transformations to the extended coecient matrix (15) and translates the result to the language of systems of linear equations, then one gets the following result.
12
Proposition 1.6.
x1
ter renaming the variables one can always transform an arbitrary system of linear equations (6) into an equivalent one of the form
0 = bm
with the
0 r m, n.
then also the transformed system of linear equations (23) is homogeneous, that is
mr
last
Proposition 1.7.
solutions.
n>m
unknown
variables (that is more unknown variables then equations) has always non-trivial
Proof. We may assume without any loss of generality that the system of linear
r m < n.
Let
xr+1 , . . . , xn
be
arbitrary numbers with at least one number dierent from zero. Then set
1 i r.
Then the
n-tuple x
system (23).
Proposition 1.8.
tions (6) with as many unknown variables as equations (that is then the system (6) has a unique solution.
m = n).
If the
homogeneous part of the system of linear equations has only the trivial solution
Proof. Since the homogeneous part of the system of linear equations is as-
sumed to have only the trivial solution it means that we can transform the system of linear equations (6) with elementary row transformations of type I, II and III into a system of linear equations of the following form (after possible renaming of the unknown variables):
x1 x2
.. .
= b1 = b2
. . .
xn = bn
Now this system of linear equations has clearly only a unique solution and since it is equivalent with the given system of linear equations (6) it follows that also the system of linear equations (6) has only a unique solution.
13
4.1. Methodes for Solving Homogeneous Systems of Linear Equations. Consider the homogeneous system of linear equations
a11 x1 + a12 x2 + + a1n xn = 0 a21 x1 + a22 x2 + + a2n xn = 0
. . . . . . . . . . . . (24)
A := (aij ).
columns we can transform the matrix That is we have then
(25)
Using elementary row transformations of type I, II and III together with exchanging
into the
m n-matrix D
D=
where
Ir 0
B 0
(26)
Ir
is the
r r-identity matrix, B
is a certain
stand for zero matrices of the appropriate format. Therefore is the system (24) is equivalent (after possible renaming of the unknown variables) to the system
x1 x2
.. .
n = r,
then it is evident that this system has only the trivial solution
0.
Thus
n > r.
xr+1 , . . . , xn
x1 , . . . , x r
i = 1, 2, . . . , r,
then
x1
. . . x x := r xr+1 . . . xn
is a solution of (27). Thus we see from this the following result.
(29)
14
Proposition 1.9.
In the case
n>r
B Inr
where
(30)
...,
lnr
(31)
Then the following is true: every solution of (27) can be written in an unique way
x = t1 l1 + t2 l2 + . . . + tnr lnr
(32)
whith certain numbers t1 , t2 , . . . , tnr . On the other hand the expression (32) is for any numbers t1 , t2 , . . . , tnr a solution of (27).
Proof. Assume that
t1 := xr+1 ,
Since
t2 := xr+2 ,
...,
tnr := xn .
(33)
is a solution of (27) we have the relations (28). But these state together
with (33) nothing else but the relation between Assume now on the other hand, that the it follows that
form (32). Then necessarily the relation (33) and (28) are satised. From the later
the expression (32). Now Proposition 1.9 gives a very satisfying description of the subspace all solutions of a homogeneous system of linear equations. If exists elements l1 , . . . , lnr
M0
a unique way as a linear combination of the form (32). Such a system of elements
l1 , . . . , lnr of M0 is occasionally named a system of fundamental solutions. A set {l1 , . . . , lnr } which is made up of the elements of a system of fundamental solutions is also called a basis of M0 .
Example.
x1 x1 2x1 x1
with
+ +
+ + +
+ +
x4 + x5 = 0 2x4 + x5 = 0 x4 =0 2x4 x5 = 0
(34)
equations in
unknown variables.
1 1 2 1
2 1 1 1 2 2 2 1 4 3 1 0 2 2 2 1
1 2 0 0 0 0 0 0
1 1 1 1 3 2 1 3 2 1 3 2
15
1 0 0 0
1 2 1 1 1 0 3 2 1 0 3 2 1 0 3 2 0 1 0 0 2 0 0 0 4 3 0 0 3 2 0 0
1 0 0 0
1 1 0 0 2 0 0 0
2 0 0 0
1 3 0 0
1 2 0 0
1 0 0 0
1 0 0 1 0 0 0 0
4 3 3 2 0 0 0 0
Note that the third matrix is derived from the second one by exchanging the second and third column. If one takes this column exchange into account when applying Proposition 1.9 then one optains the following system
2 1 l1 := 0 , 0 0
4 0 l2 := 3 , 1 0
3 0 l3 := 2 0 1
of fundamental solutions for the homogeneous system of linear equations (34). This concludes the example.
4.2. Methodes for Solving Nonhomogeneous Systems of Linear Equations. We begin with a nonhomogeneous system of linear equations as in (6). Using
elementary row transformation of type I, II and III and possible column exchanges we can bring the simple coecient matrix of (6) into the form (26). As the last column one obtains Then one performs the same row transformations on the extended coecient matrix of (6).
b1
. . . . bm
Thus the system of linear equations (6) after possible renaming of the unknown variables is equivalent to the following system of linear equations:
x1 x2
.. .
0 = bm
Now this system of linear equations is exactly then solveable (and therefore also the system (6)) if
br+1 = br+2 = . . . = bm = 0
16
holds.
x r+ 1 = . . . = x n = 0 b1 . . . b x := r 0 . . . 0
n-tuple
of (6) that
M = x + M0
where
M0
associated with (6). Thus we have due to Proposition 1.9 the following result.
Proposition 1.10.
is solveable and that subspace form
M0
x is some solution of (6). If {l1 , . . . , lnr } is a basis of the of all solutions of the hogogeneous system of linear equations associx
of (6) can be written in a unique way in the
x = x + t1 l1 + . . . + tnr lnr
whith numbers t1 , . . . , tnr . And vice versa, any such expression is a solution of (6).
Example.
+ + + +
+ + + +
1 0 3 1
(36)
In the example on page 10 we have already considered the simple coecient matrix of this system of linear equations. We perform now the same row transformations on the extended coecient matrix and obtain:
1 3 0 2
3 9 2 8
5 10 7 12
2 1 3 2
0 1 2 0 1 3 1 1
1 3 5 0 2 2 0 0 5 0 0 0
2 2 5 0
0 1 1 1 2 4 0 1
Thus we see already at this stage that the system of linear equations (36) is not solveable: the last row of this extended coecient matrix transforms into the equation
0 = 1.
Note that the above example shows that a nonhomogeneous system of linear equations with more equations than unknown variables is not always solveable (even though one might not see this on the rst sight).
5. Two Problems
In the discussion so far we optained knowledge about the theory of solving systems of linear equations which one can describe as satisfying. After we gained insight in the general form of the solutions (Proposition 1.2 and Proposition 1.3) we have seen that using the Gauss algorithim we have a simple way to determine whether a given system of linear equations (6) is solveable or not. Further we have
4
r=m
5. TWO PROBLEMS
17
seen in the previous section that the Gauss algorithm is a useable tool to obtain all sollutions of a solveable system of linear equations. But on the other hand our studies give inevitable raise to some theoretical questions which we want to point out at this place: As we have seen we can transform any matrix ber
into a matrix of the form (21). Now we might ask about the nature of the num-
only depend on
C.
even if we
use dierent sequences of elementary row transformations and column exchanges to achive the general form (21). Let us write down this problem explicitely.
Problem 1.
r0
such that this number does not change under an elementary row transformation and under a column exchange and such that a matrix of the form (21) is assigned exactly the number
r? M0
of all solutions to a homogeneous
The other question concerns the set of sollutions of a homogeneous system of linear equations. We have seen that the set tuples system of linear equations (9) is always a subspace of the set of all possible
n-
F n.
This suggest the following opposing question: Exist for every subspace
Problem 2.
of
Fn
U?
It will now come with no surprise that if we want to nd an answer to the above problems then we need to extend our theoretical horizont. In the next chapter we will introduce in a systematically way the basic concepts of Linear Algebra, which we have already prepared in this chapter. Towards the end of next the next chapter we will be able to answer these problems.
5
Problem 1 will be answered on page 45 and Problem 2 will be answered on page 49.
CHAPTER 2
Vector Spaces
1. Fields
In the previous chapter we have constantly spoken of numbers without saying anything in particular what kind of numbers we actually deal with. We did not use any particular properties of the numbers but rather only assumed some well known properties of the addition and multiplication of numbers as we know them for example from the real numbers (which are known from calculus). Indeed,
most of Linear Algebra as for example the the treatment of systems of linear equations does not depend on any particular properties of the numbers used. In this section we shall introduce the concept of elds. In a nutshell a eld + and the other is called the multiplication and denoted by minimum of axioms which will be important for us.
1
is a set together with two operations one is called the addition and denoted by
satisfying a
Denition 2.1.
maps
A eld
F = (F, +, )
and two
+ : F F F, (x, y) x + y
and
: F F F, (x, y) xy
(the rst one called addition, the latter is called multiplication ) satisfying the following axioms: (A1) The addition is associative, that is
for every
x, y, z F .
(A2) The addition is commutative, that is
x, y F .
(A3) There exist exactly one element which we denote by 0 for which
x+0=x
holds for every (A4) For every denote by
and
x F . This element is also called the zero x F there exists exactly one element in F x such that x + (x) = 0
and
F.
which we will
x, y, z
F.
(M2) The multiplication is commutative, that is by 1 for which (M3) There exist exactly one element dierent from
x1 = x
holds for every
1
and
1x = x F.
x F.
See Appendix A for the notation used for sets and maps.
19
20
2. VECTOR SPACES
x=0
of
which we
) such that
x(1/x) = 1
and
(1/x)x = 1.
(D) The addition and multiplication are bound by the distributive law, that is
(x + y)z = xz + yz
for every
and
x(y + z) = xy + xz
x, y, z F .
Note that we could do with a smaller list of requirements. For example since the addition is required to be commutative (A2) it follows from necessarily
x+0 = x
that
0 + x = x,
and (M4) could have been left away. Note further that in the language of Algebra the axioms (A1) to (A4) mean, that that
F is an abelian group under the addition, and the axioms (M1) to (M4) mean F := F \ {0} is an abelian group under the multiplication. This terminology
is not of further importance here, but it will reappear Section 11 of Chapter 3. There we will give a more detailed denition. From the eld axioms one can easily other known calculation rules, for example
(37) (38)
y=0
for
(39)
y=0
and
v=0
(40)
x/y := x(1/y) = xy 1 .
exercise.
Examples.
Q :=
r :rZ s R.
and
s N+ .
C := {a + ib : a, b R}
where If
is a eld and
F F is a subset such that F is a eld under the same F , then F is said to be a subeld of F . For example Q( 3) := {x + y 3 : x, y Q} xF
and
is a subeld of Let
R. k >0
a positive integer, then we
be an arbitrary eld. If
shall dene
kx := x + . . . + x .
k
times
If k = 0 we dene kx := 0 and in the case that k < 0 is a negative integer we dene kx := (k)x.
2. VECTOR SPACES
21
Now observe that the eld axioms do not exclude the possibility that for the eld
holds
k1=0
for every integer
(41)
k>0
k = 0. F
If
is a eld where
k1 = 0
Denition 2.2
(Characteristic of a Field)
Let
be a eld.
If there exists a
k>0
such that
k1=0
then we say that we say that
k,
in symbols
0,
that is
char(F ) := k . char(F ) := 0.
Otherwise
Most of the time the characteristic of a eld does not aect the results in what follows. It is still important to remember that as long as a eld does not have characteristic and non-zero
0 we cannot be sure that k x = 0 is always true for non-zero x F k Z. In Appendix B examples for leds with positive characteristic
are presented. We shall conclude this section about elds with a related denition. In Linear Algebra (and other elds of mathematics) one encounters mathematical objects which satisfy nearly all requirements of a eld. tegers For example the set of inwith the well known addition and multiplication satises all eld axioms
3
except (M4). In Chapter 3 we will encounter mathematical objects which satisfy even less of the led axioms, namely all eld axioms except (M2) and (M4). observation motivates the following denition. This
(Ring)
said to be commutativ . A ring which satises (39) is called regular. The set of integers
Z := {0, 1, 2, 3, . . .}
is a commutative and regular ring under the usual addition and multiplication. It is called the ring of integers.
2. Vector Spaces
We are now ready to dene the central algebraic object of Linear Algebra: vector spaces over a eld
F.
is a set
where we can compute the sum of any two elements and where we can multiply
Denition 2.4
(or in short maps
(Vector Space)
Let
F -vector
space ) is a triple
F be a eld. A vector space over the eld F V = (V, +, ) consisting of a set V and two
+: V V V, (x, y) x + y
2
Note that this does not contradict with (39) since there we assumed that
in (41) we have
kZ
and
1 F.
x, y F Z as
whereas a subset
of the eld of rational, real or complex numbers. But in general we cannot assume that we can consider the integers as a subset of an arbitrary eld.
3
EndF (V )
Mn (F )
of degree
over
F.
22
2. VECTOR SPACES
and
: F V V, (a, x) ax
(the rst one is called addition, the latter is called scalar multiplication ) satisfying the following vector space axioms: (A1) The addition is associative, that is
x + (y + z) = (x + y) + z x+y =y+x
for every
for all
x, y, z V .
(A2) The addition is commutative, that is (A3) There exists exactly one element in
x, y V .
x+0=x
for every such that (A4) For every
xV. x V
x + (x) = 0.
(SM1) (SM2) (SM3) (SM4)
(ab)x = a(bx) for every a, b F and x V . 1x = x for every x V . a(x + y) = ax + ay for every a F and x, y V . (a + b)x = ax + bx for every a, b F and x V .
Note that the above axioms (A1) to (A4) are structural exactly the same as the axioms (A1) to (A4) in Denition 2.1 of a eld in the previous section. Again this means in the language of Algebra that group. Elements of the vector space the addition in
V.
0x = 0 ax = 0
and
a0 = 0
and in particular
(42)
(1)x = x
x=0
a(x y) = ax ay
where
a F , x, y V
and
x y := x + (y). F
The verication of these rules is and the zero vector of the vector
left as an exercise. Note that the calculation rule (42) ensures that the use of the symbol 0 for both the zero element of the eld space
will not cause confusion. (1) In the previous chapter we have already introduced the set all
Examples.
Fn
n-tuples x1 x2 x = . , . . xn xi F
F.
scalar multiplication as dened in Section 2 of the previous chapter this (2) Consider the set
I R
space.
with
F -vector space. C 0 (I) of all continuous maps f : I R from the interval values in R. This set becomes in a natural way a R vector
2. VECTOR SPACES
23
V E
F. E
Then
is also a vector
F F. E.
Then is in a natural way a vector
a subeld of
space over
F.
is in a natural way
Q-vector
space.
FI
is
I . Then F , that
F I := {f : f
The set of
is a map from
to
F }.
elements of
F I,
that is
and
is the map
F -vector space: If x and y are x: I F and y: I F are maps, then the sum x + y: I K dened by i I, aF
is the map (46)
with an element
ax: I F
(47)
(ax)(i) := ax(i).
In particular, if we set vector space as
I := {1, 2, . . . , n}
4
then
Fn
Denition 2.5.
eld
Let
and let
V V
scalar multiplication of
V,
then
V V
V.
of a
Proposition 2.6.
space
U V
F -vector
is a linear subspace of
satised:
x, y U
x+y U ax U
(48) (49)
a F, x U
Proof. If
V , then of course (48) and (49) are satised. Thus V satisfying the conditons (48) and (49). We need only to verify the vector space axioms (A3) and (A4). Let x U . Then x = (1)x U by (43) and (49). Thus (A4) is satised. Now since U is assumed to be not empty it follows that there exists an x U . Thus also x U due to (A4) and therefore 0 = x + (x) U due to (48).
is a subspace of assume that
is a non-empty subset of
dierence between the denition of a linear subspace The rst one de-
(Denition 2.5) and the Subspace Criterion (Proposition 2.6). whether a subset
nes what a linear subspace is, the latter gives a criterion to decide more easily
U V
of a vector space
V V
Examples.
{0} V . A 0 is also called the zero space. By abuse of notation we denote the zero space with the symbol 0, that is 0 := {0}. Again there is no danger of confusion of the zero space with the zero element of the eld F or with the zero vector of the vector space V if the reader is awake.
(1) Every vector space has the trivial subspace vector space which consists only of the zero vector
We will later dene precisely what we mean exactly by the term essentialy the same when
we introduce in Section 2 in the next chapter the concept of isomorphisms and isomorphic vector spaces.
24
2. VECTOR SPACES
M0
Fn
RI . the F -vector
space
FI
from the
(I)
the subset of
consisting of all
f (x) = 0
Then
for all
xI
F (I)
is a linear subspace of
FI
and
F (I) = F I F (I)
(I)
if and only if
is not
a nite set. Note that the vector spaces of the form tance. Every are of principal impor-
F -vector
I .5
This
is the fundamental theorem of the theory of vector spaces and we will later return to this subject. If
sum of
of a vector space
V,
then we denote by
U +W
the
U + W := {u + w : u U
and
w W }.
and let
Proposition 2.7.
subspaces of subspaces of
Let
V. V.
F U W
and
W be U +W
Proof. This is veried directly using the Subspace Criterion from Proposi-
tion 2.6.
3. Linear Combinations and Basis of a Vector Space Denition 2.8 (Linear Combination). Let V be a F -vector space and u1 , . . . , um
some vectors of
V.
u1 , . . . , um
of the vectors
v = a1 u1 + . . . + am um .
If
(50)
M =
is an arbitrary subset of
V,
combination of vectors in
is a linear combination of
u1 , . . . , um . 0
Note that a linear combination is always a nite sum! In Linear Algebra we do not consider innite sums. Note also that the zero vector way a linear combination of any collection is always in a trival
u1 , . . . , um ,
namely
0 = 0u1 + . . . + 0um .
Denition 2.9.
Let
subset of the
F -vector
space
V.
Then the
span M := {v V : v
If
M }.
(51)
M =
we dene
span := {0}
to be the trivial zero vector space
(52)
0.
0.
If one considers
F (I)
for
I=
25
u1 , . . . , um
such that
M = v can be
then
v span M
ui
are not
pairwise dierent vectors, then we can always transform the linear combination into one with pairwise dierent vectors Using the vector space the following way.
ui M . v span M
in
F (M )
Proposition 2.10.
vV
Let
is an element of
of the
exists a
x(u)u.
u1 , . . . um M ai 0
if
v = a1 u1 + . . . + am um .
x: M F
by
x(u) :=
u = ui
for some
1 i m,
otherwise.
x F (M )
and
then
is
Note that the notation in (53) is a very practical notation. If now and
x, x F (M )
(54)
a F,
x(u)u +
uM
and
x (u)u =
uM uM
(x + x )(u)u
a
uM
x(u)u =
uM
(ax)(u)u
(55)
due to the denition of the addition and scalar multiplication in (46) and (47) for vectors of the vector space
FM
(M )
. Now using Proposition 2.10 the above two equations translate into the two
observations:
v, v span M
and
v + v span M
v span M, a F
Since
av span M.
span M
Proposition 2.11.
span M
Let
F -vector
space
V.
Then
is linear subspace of
V.
26
2. VECTOR SPACES
that
M is an arbitrary subset of V . We claim that M span M . In the case M = this is clear. Thus let M = . If v M then v = 1v span M and thus M span M also in this case. On the other hand, if W is an arbitrary linear subspace of V with M W , then clearly span M W . Let U be another linear subspace which has the above property of span M , that is M U and U is containied in any linear subspace W which contains M . Then U span M and on the other hand also span M U . That is, U = span M . Thus we have shown the following usefull characterisation of the linear hull span M of M .
If
Proposition 2.12.
if
Let
M
of
V.
such that
M U
with
is a linear subspace of
M W
U W.
(56)
U = span M . M
is precisely the smalles
In short the above result means that the linear hull of linear subspace of
containing
M.
(57)
(58)
M = span M
The rst property we have already veried above, the proof of the remaining prop-
Examples.
V := C 0 (R) be the R-vector space of all continuous functions f : R R. Then the functions exp, sin and cos are vectors of the vector space V . We shall show that exp is not contained in the linear hull of M := {sin, cos}. We show this by assuming towards a contradiction that exp span{sin, cos}. By denition this means that there exists numbers a1 , a2 R such that exp = a1 sin +a2 cos. In other words this
(1) Let means
for all
x R.
0 < exp(0) = a1 sin(0)+a2 cos(0) = 0+a2 = a2 0 < exp() = a1 sin() + a2 cos() = 0 a2 = a2 , that is a2 > 0 and at the same time a2 < 0 which is a contradiction. Thus the assumption that exp is a linear combination of the functions sin and cos is shown to be wrong and it follows that exp span{sin, cos}. /
(2) Recall the setting and notation of the frist chapter. Consider a system of
F.
Denote by
with the system of linear equations. Then the system of linear equations
b span{v1 , . . . , vn }
where
27
Moreover the associated homogeneous system of linear equations has a non-trivial solution if and only if the zero vector linear combination of the vectors The meaning of
0 Fm
is a non-trivial
v1 , . . . , v n . v1 , . . . , v n
Denition 2.13.
of the
Let v1 , . . . , vn be some (not necessarily pairwise distinct) vectors F -vector space V . Then 0 is a non-trivial linear combination of the vectors v1 , . . . , vn if there exist elements a1 , . . . , an F such that
a1 v1 + . . . + an vn = 0
and
ai = 0
1 i n.
Note that the above denition is of essential importance! This is already suggested by the close relation with homogeneous systems of linear equations which did lead to the denition.
Example.
Consider the
R-vector
space
V = R4 .
Then
is a non-trivial linear
1 1 v1 := , 3 4
For example
4 10 v2 := 3 5
and
3 7 v3 := . 1 2
v1 2v2 + 3v3 = 0.
Denition 2.14
A subset
M V
of a vector space
V, M
if every vector
vV M
M.
contains only nitely many elements. of
vV
v=
(62)
Note that the above denition of a basis contains two properties: (1) We have that
V = span M .
vV
can be written as
M.
That is if
(2) For every vector there exists only one way to write
M.
v=
uM
then necessarily
x(u)u =
uM
that is
y(u)u
for every
x = y,
x(u) = y(u)
u M. n
elements, say
Examples.
M V
which consist of
M = {u1 , . . . , un },
is precisely then a basis of the
F -vector
space
vV
(63)
x1 u1 + . . . + xn un = v.
28
2. VECTOR SPACES
In the previous chapter we have seen that the uniqueness requirement is equivalent with the homogeneous part of (63),
x1 u1 + . . . + xn un = 0
having only the trivial solution (Proposition 1.8). And this is again equivalent with that there exists no non-trivial linear combination of the zero vector
0. M
is a basis of
if and only if
V = span M 0 by
M. V = F 2. a c , b d
Then every subset
(64)
where
ad bc = 0
is a basis of
V.
Then the set consisting of the canonical
F n.
unit vectors
1 0 . e1 := . , . . . . 0
forms a basis of exercise. (4) More general, if
0 1 e2 := 0 , . . . , . . . 0
0 . . . en := . . . 0 1 F n.
(65)
F n, I
ei
(66)
F (I)
dened by
ei (j) :=
for every
0 1
for for
j = i, j = i. F (I) .
iI
standard basis of
(I)
. It follows that
x=
iI
for every (5) Let
x(i)ei
(67)
x F (I) . {0}.
Because otherwise
and this would contradict to the uniqueness requirement in the denition of a basis. This motivates the agreement that the empty set basis of the zero vector space.
F. M V
Denition 2.15.
if In this case
We say that
V = span M. M
is called a (linear) generating system of
V.
We say that
is nitely
of
29
Examples.
V = span V .
(2) The vector space
Fn
{e1 , . . . , en }.
(3) The empty set is a generating system of the zero vector space. (4) The vectors
1 , 2
gives a generating system of (5) The ve vectors
3 , 4 R2 ,
5 6
1 3 , 0 2
ample on page 10. (6) If
3 9 , 2 8
5 10 , 7 12
2 1 , 3 2
0 2 1 1 R4 ,
compare with the ex-
F (I)
C (R)
to
is not
We are now interested in small generating systems in the sense that the generating system does not contain any vectors which are not necesarily needed to generate the vector space. This leads to the following denition.
Denition 2.16
generated by holds
(Linear Dependence)
A subset
if for every
M \ {u}
is a proper subspace of
M V of the vector space V is u M the linear subspace span M . That is, for every u M
M V is said to be a linear V. M uM
dependent subset of
if it is not a linear
that there
Examples.
(2) If
V.
M V
subset of
V. 1 , 2 3 , 4 R2 5 6
since any two of those vectors gen-
form a linear dependent subset of erate already (4) Assume that two non-zero dependent.
R . char F = 2 (that is 1 + 1 = 0 in F ). And let u, v V be vectors of V . Then the set {u, u + v, u v} is always linear
be a subset. Let
Lemma 2.17.
M {v}
Let
M V
v span M
such that
v M. /
Then
V.
30
2. VECTOR SPACES
M M {v} we have that span M span(M {v}). On the v span M we have that M {v} span M . But then also span(M {v}) span(span M ) = span M . Thus we have the inclusion in both directions and therefore span M = span(M {v}). Now since we assumed that v / M we have (M {v}) \ {v} = M . Therefore span((M {v}) \ {v}) = span(M {v}) and this shows that M {v} is a linear dependent subset of V .
Proof. Since
A subset
M V
is linear depenedent if and only if there exists nitely many pairsuch that
v1 , . . . , v m M v1 , . . . , v m .
exists a
Assume that M is a linear dependent subset of V . Then there v1 M such that span M = span(M \ {v1 }). Thus there exist pairwise distinct vectors v2 , . . . , vm M \ {v1 } such that v1 = a2 v2 + . . . + am vm for some numbers a2 , . . . , am F . But this means that
Proof. :
0 = 1v1 a2 v2 . . . am vm
is a non-trivial linear combinations of the zero vector
v1 , . . . , v m .
: Assume that there exists pairwise distinct vectors
v1 , . . . , v m M
such
that the zero vector is a non-trivial linear combination of the these vectors, say
0 = a1 v1 + . . . + am vm
with some number
ak = 0 .
a1 = 0 and a1 = 1. Then
in
v1 = a2 v2 + . . . + am vm span(M \ {v1 }). M span(M \ {v1 }) and therefore span M span(M \ {v1 }). On the other span(M \{v1 }) span M since M \{v1 } M . Therefore we have the inclusion in both directions and this shows that we have the equality span M = span(M \{v1 }) for v1 M . Therefore M is a linear dependent subset of V .
Thus hand Now using the fact that a if and only if
V. M
A subset
v1 , . . . , v m M
it follows from
a1 v1 + . . . + am vm = 0
that
a1 = . . . = am = 0.
Note that the above proposition is often used in literature to dene linear inde-
pendence. Further it follows from the same proposition that if of a linear independent subset subset of
M M
is a subset
of
V,
then
V.
Let
Denition 2.20.
that
M V
V.
Then we say
M V
if for every
u M \M u V \M
the set
M {u}
is linear dependent.
if for every
M {u}
is linear dependent.
31
Lemma 2.21.
(1) If (2)
Let
M V
V.
v V is a vector such that M {v} is a linear depenendent set, then v span M . If M is a maximal linear independent subset of M V , then M span M . (In particular span M = span M .)
(1) Assume that
Proof.
M {v}
is linear
dependent. It follows from Proposition 2.18 that there exists nitely many pairwise distinct vectors
v1 , . . . , v m M
a, a1 , . . . , am F
(68)
av + a1 v1 + . . . + am vm = 0.
If
a=0
is linear dependend,
a=0
1
a1
and get
v = (a1 a
Thus (2) Let
)v1 + . . . + (a1 a
)vm .
v span M as claimed. u M be an arbitrary element. If u M then clearly u span M . On the other hand, if u M , then M {u} is linear dependend (due / to the maximality of M ) and thus by the previous result it follows that u span M , too. Alltogether this means that M span M . M V
to be a basis of
We are now ready to characterize the property of a set the vector space
Proposition 2.22
F -vector
(1) (2) (3) (4) (5)
Let
M V
be a subset of the
space
V.
M M M M M
of
is a basis of generates
V. V. V. E
and
is linearly independent.
which contains
M.
E of V which contains M as a maximal E. 6 (M ) There exists a bijective map f : F V such that: (M ) (a) f (x + y) = f (x) + f (y) and f (ax) = af (x) for all x, y F and a F, (b) f (eu ) = u for every u M were {eu : u M } is the standard basis (M ) of F as dened in the example on page 28.
(3) (6)
(4) (5)
(3):
That the statements (2) and (3) are equivalent follows straight
from the denition of the linear independence. (1) (3): Assume that
32
2. VECTOR SPACES
that
is
not linearly independent. Then this means that there exists nite many vectors
v1 , . . . , v m M
a1 , . . . , am F
0 = a1 v1 + . . . + am vm .
On the other hand we can also express the zero vector in the following form:
0 = 0v1 + . . . + 0vm .
But this contradicts to the property of a basis which says that by denition there is only one unique way to write the zero vector as a linear combination of vectors of
M.
is linearly independent.
M is a linearly independent V . We have to show that for every v V which is not a vector of M the set M {v} becomes linearly dependent. But since M is assumed to be a generating system of V it follows that v span M and thus M {v} is linearly
(4): We assume that (3) holds, that is generating system of dependent by Lemma 2.17. (4)
(5):
If
then
a maximal linear independent subset of any subset (5) (6) (6): (3): Evident since
EV
which contains
M is M. M.
also
is a generating system of
containing
M is already assumed to be linearly indeV = span M . Since M is a maximal linear independent subset of E it follows from the second part of Lemma 2.21 that E span M . In particular this means that span E = span M and since E is a generating system of V we get that V = span M .
We assume (6). Since pendent it remains to show that (3)
(1):
We assume that
V = span M
(that is
is linearly independent.
M
are
satisfy also the second property of a basis. Assume therefore that two elements such that
x, y F (M )
x(u)u =
uM
Then
y(u)u.
uM
(x(u) y(u))u = 0
uM
and since for every means
M is assumed to be linearly independent this means that x(u) y(u) = 0 u M (Proposition 2.19). Thus x(u) = y(u) for every u M and this that x = y as required by the denition of a basis.
(7): We dene a function
(1)
f : F (M ) V x(u)u.
uM
by
f (x) :=
Then
is surjective due to the rst property of a basis and injective due to the
satises
the condition (a), see (54) and (55). Further we have by the denition of the map
ev ,
vM
f (ev ) =
uM
ev (u)u = v
33
(7) Since
(1):
x F (M )
such that
to the equation (67) one gets using the relations (54) and (55)
v = f (x) =
uM
and thus
f (x(u)eu ) =
uM
x(u)f (eu ) =
v span M .
follows that also the second property of a basis is satised. Now one very important result for Linear Algebra is the following theorem
about the existence of a basis for a vector space which we will state without a proof.
has a basis.
The idea of the proof is roughly the following: Due to the characterisation of a basis (Proposition 2.22) it is enough to show that every vector space
has a
maximal linear independent subset. The existence of such a set is proven using a
result from set theory, called Zorn's Lemma. For more details see the Appendix C.
Example.
R can be considered as a vector space over Q. Then the above theorem states that there exists a Q-linear independent subset B R such that for every x R there exists nitely many (!) elements b1 , . . . , bm B and numbers a1 , . . . , am Q such that
the eld of rational numbers
x = a1 b1 + . . . + am bm .
Note that the above theorem does not give an answer how this set looks like. The theorem only states that this subset of
exists.
For nitely generated vector space the existence of a basis is easier prove. We state therefore the bit weaker form of the above theorem.
Theorem 2.24
V
Let
contains a basis of
V. E
is nite it must is by Proposition 2.22, (2) a
M.
But then
V.
u1 , . . . , um
of a vector
V.
The rank of this collection shall be the maximal number of linear inde-
pendent vectors in this collection. More precisely this is formulate in the following
For example in the end of this chapter we will dene what we mean by the rank of a matrix
and in the next chapter we will assign to every linear map spaces a number which we will call the rank of
f.
34
2. VECTOR SPACES
Let
r,
in symbols
rank(u1 , . . . , um ) := r,
if the following two conditions are satised: (1) There exists a linear independent subset of of exactly
{u1 , . . . , um } r+1
which consists
vectors.
{u1 , . . . , um }
which consists of
vectors is linear
u1 , . . . , um of vectors of a vector space V m-tuple (u1 , . . . , um ) of vectors in V . One must dierentiate this from the set {u1 , . . . , um }. In the later case the order of the vectors is not important, the order in which the the vectors u1 , . . . , um appeare in the m-tuple
If we speak in the following of a system then we mean always the are essential! Note that we have clearly
m,
that is
rank(u1 , . . . , um ) m.
Equality holds, that is
rank(u1 , . . . , um ) = m,
if and only if the system of vectors
u1 , . . . , um
Proposition 2.26.
(2) We have (3) (4)
Let u1 , . . . , um a system of vectors of the Then the following statements are equivalent:
(1) The system
F -vector
space
V.
u1 , . . . , um is linearly independent. rank(u1 , . . . , um ) = m. The set {u1 , . . . , um } is a linear independent subset of V of m vectors. If there exists elemenst a1 , . . . , am F such that a1 u1 + . . . + am um = 0
consisting exactly
then necessarily
a1 = . . . = am = 0. 0
to a system of vectors
u1 , . . . , um ,
then
the rank of this system does not change. That is we have the equality
rank(u1 , . . . , um , 0) = rank(u1 , . . . , um ).
This observation generalizes to the following result.
Proposition 2.27.
Let
u span{u1 , . . . , um }.
and
rank(u1 , . . . , um , u) = rank(u1 , . . . , um ).
35
of vectors
u1 , . . . , um ,
that is we set
r := rank(u1 , . . . , um ).
(69)
We know already that rank(u1 , . . . , um , u) r . Thus it remains to show that also rank(u1 , . . . , um , u) r is true. In order to show this we need to prove that there exists no linear independent subset {u1 , . . . , um , u} which consists of r + 1 elements. Thus let us assume towards a contradiction that there actually exists a linear independent set linear we have that necessarily Lemma 2.21 that
T {u1 , . . . , um , u} which consists of r + 1 elements. Due to (69) u T and that the set M := T \ {u} is a maximal independent subset of {u1 , . . . , um }. Thus it follows from the second part of span M = span{u1 , . . . , um }.
(70)
u span{u1 , . . . , um }
we have that
u span M .
Thus
M {u} = T
is linearly independent.
{u1 , . . . , um , u}
r+1
rank(u1 , . . . , um , u) r.
always by the
Next we turn our attention to the question how we can determine the rank of a given system vectors replace the vector space
u1 , . . . , um of vectors of a F -vector space V . Since we can V by the nitely generated vector space spanned u1 , . . . , um we will assume in the following that V
is a nitely generated
F -vector V
space.
(71)
Thus Theorem 2.24 applies and we know that is such a nite basis of
V,
say
M = {b1 , . . . , bn }
consists of exactly as
(72)
distinct vectors of
V.
uV
can be written
u=
i=1
ai bi = a1 b1 + . . . + an bm
(73)
numbering
denition.
ai F . Observe that the n-tuple (a1 , . . . , an ) u only depend on the set M but also on the of the basis vectors b1 , . . . , bn . Therefore it is useful to make the following
(Ordered Basis). Let V be a nitely generated vector V is a n-tuple (b1 , . . . , bn ) of pairwise distinct vectors {b1 , . . . , bn } is a basis of V in the sense of Denition 2.14. space. An of
Denition 2.28
ordered basis of
that the set
such
It is a straight consequence of Theroem 2.24 that every nitely generated vector space
ordered basis of a vector space. In order to simplify the notation we will therefore agree to call an ordered basis simply just a basis. Therefore it will be the context which will tell whether the therm basis will refer to an set or an ordered system. When we speak in the following of the canonical basis of the vector space we will nearly always mean the ordered system vector (65) of the vector space
Fn
then
e1 , . . . , e n
F n. If V is a nitely generated F -vector space with b1 , . . . , bn being an (ordered) basis of V then every vector u V can be expressed in a unique way in the form (73). One calles the ai F the coordinates of the vector u with respect to
36
2. VECTOR SPACES
the basis of
b1 , . . . , b n .
The
We
Denition 2.29.
(u1 , . . . , um )
of a
F -vector ui
ations which transforms the system again into a system of (I) Replacing one in
(u1 , . . . , um ) ui
in
by
(II) Exchanging two vectors in the system. (III) Replacing one vector
(u1 , . . . , um )
aui
where
a F \ {0}.
Next we shall show, that we may use these transformations in order to determine the rank of a vector system.
Proposition 2.30.
Proof.
The rank of a system of vectors (u1 , . . . , um ) is not changed under the elementary transformations of Denition 2.29.
(I) We may assume with out any loss of generality that
i=1
and
system
u1 , . . . , um
we have
rank(u1 , . . . , um ) = rank(u1 , u2 , . . . , um )
as needed to be shown. (II) The claim is obvious for a transformation of type (II). (III) The proof is done analogous as in the case of a transformation of type (I).
space
u1 , . . . , um be a given system of vectors of a nitely generated F -vector V . Let b1 , . . . , bn be an ordered basis of V . Then every vector of the system u1 , . . . , um can be written in a unique way as a linear combination of the basis vectors b1 , . . . , bn :
Now let
ui =
j=1
with numbers
aij bj ,
i = 1, 2, . . . , m
(74)
aij F .
mn-matrix:
(75)
We shall call this matrix the coordinate matrix of the vector system
u1 , . . . , um
b1 , . . . , b n .
37
u1 , . . . , um
with respect
b1 , . . . , b n . u1 , . . . , um
is equivalent with an We have seen in
Chapter 1 how we can transform this matrix into a matrix of the form (21). To achive this form we possible need to exchange columns, but such a column exchange corresponds to re-numbering the elements irrelevant for computing
b1 , . . . , b n
rank(u1 , . . . , um ).
space and let
Let
be
F -vector
b1 , . . . , bn
be a basis of
V.
transform step by step using elementary transformation any nite system of vectors
u1 , . . . , um of V into a systme u1 , . . . , um such that the coordinate matrix of this transformed system with respect to a basis which diers from the basis b1 , . . . , bn at
most by a re-numeration is of the following form
0 1 0
0 0 1
0
. . . . . . . . .
0 0 . . . . . . 0
..
. .. .
0 1 0
matrix and
(76)
r r-identity
is a certain natural
0 r m, n.
Then
rank(u1 , . . . , um ) = r
Proof. We only have to show the last claim.
tion 2.30 that elementary transformations do not change the rank of a system of vectors we need just to show that if holds. Since the last in this case the vectors coordinate matrix with respect to some basis
u1 , . . . , um is a system of vectors such that its b1 , . . . , bn is of the form (76) then (77)
that
m r rows of the matrix (76) contains only zeros it follows ui = 0 for i r + 1. Thus we have that necessarily rank(u1 , . . . , ur ) r holds.
n
0ir
ui = bi +
j=r+1
where
aij bj = bi + u i ci F
are numbers such that
u span{br+1 , . . . , bn }. i
c1 u1 + . . . + cr ur = 0
is a linear combination of the zero vector. Then
c1 u1 + . . . + cr ur = c1 b1 + . . . + cr br + u =0
(with
u span{br+1 , . . . , bn })
38
2. VECTOR SPACES
c1 = . . . = cr = 0 since the b1 , . . . , bn are linearly independent 0 is only the trivial linear combination of these vectors. Thus the vectors u1 , . . . , ur are linearly independent and it follows rank(u1 , . . . , um ) = rank(u1 , . . . , ur ) r. Alltogether we have shown that rank(u1 , . . . , um ) r and rank(u1 , . . . , um ) r
if and only if and therefore there and thus equality holds.
Example.
V = R5 :
1 3 u1 := 5 , 2 0
3 9 u2 := 10 , 1 2
0 2 u3 := 7 , 3 1
2 8 u4 := 12 2 1
of
Its coordinate matrix with respect to the canonical basis row transformations as follows:
e1 , . . . , e 5
R5
is the
45-
matrix from the example on page 10 which can be transformed using elementary
1 3 0 2
3 9 2 8
5 10 7 12
2 1 3 2
0 2 1 1
2 2 5 0
0 1 2 0
1 0 0 3 0 1 0 2 0 0 1 1 0 0 0 0
rank(u1 , u2 , u3 , u4 ) = 4.
we can already see this from from the second matrix in the above chain of matrices, that is, we do not necessarily need to completely transform the coordinate matrix into the form required by Theorem 2.31.
Denition 2.32.
b1 , . . . , b n
We say that a
F -vector space V has dimension n if V n (distinct) vectors. V has dimension m and m = n (Theorem 2.34).
has a basis
dimension
and
n,
then necessarily
Theorem 2.33.
sion
Let
u1 , . . . , um
of dimen-
n.
Then
rank(u1 , . . . , um ) n.
Corollary.
Then
Let
n-dimensional
vector space
V.
has at most
elements.
M contains at least n + 1 pairu1 , . . . , un+1 . Then these vectors form a linearly independent system of vectors and we have rank(u1 , . . . , un+1 ) = n + 1. But this contradicts to Theorem 2.33 which states that rank(u1 , . . . , un+1 ) n! Thus the assumption was wrong and necessarily M has at most n elements.
Proof. Assume towards a contradiction that
39
Let
has the same number of elements. In other words, vector space consist precisely of
n-dimensional E V
elements.
M.
Then
M must be nite, too. Say has precisely n elements. Then V n by Denition 2.32. Now let M be another basis of V . Then M is a linear independent subset of V and thus M contains at most n elements by the corollary to Theorem 2.33. Denote by m the number of elements in M . We have then m n and V has dimension m by Denition 2.32. Thus M as a linear independent subset of V has at most m elements by the corollary to Theorem 2.33, that is n m. Therefore we have alltogether m = n.
is nite has dimension The above theorem justies that we can speak of the dimension of a vector space and that we denote the dimension of
M M
is a basis of
V. V
Apparently
has a
by Proposition 2.22.
in symbols by
dim V := n
in case that
has dimension
is a nite dimen-
if and only if
nitely generated. Note further that the trivial zero space since its only basis contains the empty set If
V =0
has dimension
dim V := .
The dimension of a on the eld
vector space
F.
R-vector
space
R.
Then
is innite dimensional (compare the example on page 33). Thus the notation
dimF V
might be used to emphasis the eld we have for example
of the
F -vector
dimR R = 1
and
dimQ R = .
Theorem 2.35.
space
V.
Let u1 , . . . , um be a system of vectors of a n-dimensional vector Then this system of vectors forms a basis of V if and only if
rank(u1 , . . . , um ) = n
Proof. :
and
m=n
dent.
Thus
rem 2.34
If u1 , . . . , um is a basis of V the vectors are linearly indepenrank(u1 , . . . , um ) = m by Proposition 2.26. It follows from Theothat m = n. assumption we have that Since
: By {u1 , . . . , um } is that
by Proposition 2.26.
m = n
by
Proposition 2.22.
40
2. VECTOR SPACES
If
it every subset
N M
N N
V,
of
V,
due
to Proposition 2.22. We shall prove this result for nite dimensional vector spaces:
V.
Then
of
such that
N be a N can be N M.
Let
V,
has at most
elements. Let
V.
E := B N
is a nite generating system of Since
V which contains the linear independent subset N . E is nite it must surely contain a maximal linear independent subset M with N M . Then M is a basis of V by Proposition 2.22 which has by construction
the properties required by the theorem.
Theorem 2.37.
Let
V.
Then
(78)
dim U dim V.
sional. If
U =V
if and only if
dim U = dim V .
Proof. If
dim V =
assume that
n := dim V .
Let M be a linear U . Then M is also a linear independent subset of V and has therefore at most n elements due to the corollary to Theorem 2.33. Let m be the maximal number such that there exists a linear independent subset of U with m elements (this number exists since n is a nite number). Let M be such a linear independent subset of U with m elements. Then M is a maximal linear independent subset of U and therefore a basis of U by Proposition 2.22. Thus U is nite dimensional and dim U = m n = dim V . This proves the rst part. Assume that V is nite dimensinal. If U = V , then clearly dim U = dim V . If on the other hand dim U = dim V , then U has a basis which consists of n := dim U elements. But then M is also a basis of V due to Theorem 2.35 and it follows that U =V. We rst need to show that then is nite dimensional. independent subset of We shall conclude this section with an application of the basis extension theorem from above which yield a nearly self-evident result:
Proposition 2.38.
sional) and let
Let V be an arbitrary vector space (nite or innite dimenu1 , . . . , um be an arbitrary nite system of vectors of V . Then
u1 , . . . , um is the equal to the dimension of the subspace spanned by the vectors of the system.
Proof. Let us use the following abreviation to simplify the notation:
U := span{u1 , . . . , um },
r := rank(u1 , . . . , um ),
k := dim U.
41
consists of exactly
of
and since
dim U = k
k = rank(u1 , . . . , uk ) rank(u1 , . . . , uk , . . . , um ) = r
On the other hand is space
and thus
u1 , . . . , um a system of vectors of the k -dimensional vector r k by Theorem 2.33. Alltogether we have therefore k = r.
Then
Corollary.
Let
uV.
u span{u1 , . . . , um }
if and only if
rank(u1 , . . . , um , u) = rank(u1 , . . . , um ).
Proof. :
:
Due to
rank(u1 , . . . , um , u) = rank(u1 , . . . , um )
we get
span{u1 , . . . , um } is a subspace of span{u1 , . . . , um , u} of the same nite span{u1 , . . . , um } = span{u1 , . . . , um , u} by Theorem 2.37.
theorem of the previous section and illustrate its powerfull application. If U and W are two subspaces of V , then we have seen already that their sum U + W and their intersection U W are subspace of V (see Proposition 2.7). We are in the following interested in the situation where U + W is the largest possible subspace of V , namely the whole space V , and where U W is the smallest possible subspace of V , namely the zero vector space 0.
Let
and
be two subspaces of
V.
We say that
is the (internal)
and
V =U +W
If
U W = 0.
V is the direct sum of U and W , then we may express this fact in symbols by V = U W .8 If W is a subspace of V such that U W = 0 then we say that W is a transversal space of U in V . If even V = U W , then we say that W is a linear complement of U in V .
Note that if
in
then also
in
V.
As an application of the basis extension theorem we can see that if we are given a subspace a basis complement
U of the nite dimensional vector space V then there always a linear W of U in V . To see this we choose a basis M of U and extend it to B of V . We set M := B \ M . Then W := span M is clearly a subspace of V := U W.
such that
Note there exists also an construction of an external direct sum which is normaly denoted
with the same symbol which very closely related to the concept of the internal direct sum. But we do not consider this construction here.
42
2. VECTOR SPACES
Theorem 2.40
Let
be a given subspace of
V.
of
of
U,
such that
V =U W
Note that the linear complement of a given subspace
U.
Proposition 2.41.
Let
and
W be subspaces of an arbitrary vector space V (nite V = U W if and only if for every vector v V u U and w W such that v = u + w.
(79) That we can nd vectors
Proof. :
We assume that
V = U W.
uU
and
w W
u U and w W some vectors (possible distinct from u and w) v = u + w . Then from u + w = v = u + w follows that u u = w w.
(80)
u u U and w w W and thus it follows from (80) that both u u w w are elemnts of U W = 0. Thus u u = 0 and w w = 0 or in other words u = u and w = w . Thus the decomposition (79) is seen to be unique.
Now and Assume that
V = U +W . Thus we need only to show that U W = 0. v = v + 0 and v = 0 + v are decompositions of the form (79). But due to the uniqueness requirement this means that v = 0. Therefore U W = 0 is the zero space. Alltogether this shows that V = U W .
By assumption
:
v U W.
Then both
Corollary.
Assume that
if and only if
u=0
and
V =U W w = 0.
and let
uU
and
w W. U
Then
u+w =0
be two -
Theorem 2.42
space
Let
and
F -vector
(81)
V.
Proof. Again we will use in this proof the basis extension theorem. We set
U W.
a1 , . . . , am , b1 , . . . , br
of the subspace
a1 , . . . , am , c1 , . . . , cs
of the subspace
(83)
W.
m + s.
a1 , . . . , am , b1 , . . . , br , c1 , . . . , cs
is a basis of
U + W.
43
U + W . It remains to show that a1 , . . . , am , b1 , . . . , br , c1 , . . . , cs are linearly independent because then they form a basis of U + W due to Proposition 2.22. Therefore we assume that the zero vector 0 is a linear combination of those vectors, that is we assume that
It is clear that (84) is a generating system of the vectors
i ai +
i=1
for some coecients sums of (85) by
i bi +
i=1 i=1
i ci = 0
(85)
i , i , i F .
and
necessarily equal to zero. To simplify the notation let us denote the three partial
a, b
c,
a + b + c = 0.
Then and therefore
(85 )
a U W , b U and c W and it follows from (85 ) that b = a c W b U W . But this means that the coecients 1 , . . . , i are all equal to zero due to the choice of the vectors b1 , . . . , br and thus b = 0. Similarly one deduces that also the coecients 1 , . . . , s are all equal to zero and thus c = 0. But then from (85 ) follows that a = 0 and thus the remaining coecients 1 , . . . , m
are necessarily all equal to zero, too. Thus we have seen that the vectors (84) are linearly independent and since they generate form a basis of
U +W
U +W U
and
Example.
Let
be two
vector space
R3
dim(U W ) = dim U + dim W dim R3 = 2 + 2 3 = 1. U W is a 1-dimensional subspace of R3 and in particular U W is not the zero space 0. That is there exists a non-zero vector v U W and U W = span{v} = {av : a R}. This set is a straight line in R3 passing through the 3 origin 0. The example shows that two planes in R which contain the origin 0 of 3 R here U and W intersect always in a straight line passes through the origin 3 of R here U W .
Thus As a consequence of Theorem 2.42 we shall note the following criterion for a subspace
in
V.
Let
Proposition 2.43.
sional) and let that
(1)
and
V.
Then we have
is a transversal space of
in
if and only if
is a linear complement of
in
if and only if
and
V:
The condition
U W = 0
is equivalent with
equivalent with
V = U +W
is
Note that the second requirement of the above proposition is equivalent with the vector space
W,
that is
V being the direct sum of the nite dimensional subspaces U and V = U W . In particular is V in this case nite dimensional, too.
44
2. VECTOR SPACES
Let
(86)
mn-matrix with coecients in a eld F . We denote by u1 , . . . , um A, considered as vectors of the vector space F m . Similarly we denote by v1 , . . . , vn the columns of the matrix, considered as vectors of the n vector space F . Then the row rank of the matrix A is the rank of the system of vectors u1 , . . . , um and is denoted in symbols by rankr A := rank(u1 , . . . , um ). A
to be the rank of the system
v1 , . . . , v n
in symbols by
rankc A := rank(v1 , . . . , vn ).
Note that the row rank of a matrix remains unchanged under elementary row transformations due to Proposition 2.30. Similarly the column rank remains invariant under elementray row transformations.
9
The natural question is how the row and column rank of a matrix relate to each other. The answer will be simple: both numbers are always equal. This will mean that it will make sense to assign a rank to a matrix using either ways of computation. column rank. Let The aim of this section will be to show this equality of row and In the next section we will then use the result of this section to
u1 , . . . , um be a system of vectors of a nite dimensional F -vector space V , b1 , . . . , bn be a basis of V . Let us for a while use the following notation: for every vector u V we shall denote by u the coordinate vector of u with respect to the basis b1 , . . . , bn . (This notation does express the dependency of the coodinate vectors on the basis b1 , . . . , bn but we do not mind this here cause we keep the basis xed for the current consideration.) Now the zero vectore 0 is a non-trivial linear combination of the vectors u1 , . . . , um if and only if 0 is a non-trivial linear combination of the corresponding coordinate vectors u1 , . . . , um . This is because for arbitrary elements 1 , . . . , m F we have
and let
i ui
i=1
and
=
i=1
i ui
v=0 A
is equivalent with
v = 0.
rank(u1 , . . . , um ) = rank(1 , . . . , um ). u
Let the be the coordinate matrix of the system of vectors the basis
u1 , . . . , um with respect to b1 , . . . , bn . Then by its denition the rows of the matrix A are precisely n vectors u1 , . . . , um F . Therefore we have shown the follwoing result.
9
Elementary column transformations of type I, II and III of a matrix are dened in the very
analogous way as we have dened elementary row transformations for matrices on page 8 in the previous chapter.
45
Proposition 2.45.
F -vector
space arbitrary basis
Let u1 , . . . , um be a system of vectors of the nite dimensional V . Let A the coordinat matrix of this system with respect to an b1 , . . . , bn of V . Then
is
u1 , . . . , um
b1 , . . . , b n . u1 , . . . , uk
is
unchanged under elementary transformations and that the elementary transformations are in a one-to-one correspondence to row transformations of the coordinate matrix and of the system
A of the system u1 , . . . , um with respect to the basis b1 , . . . , bn . But the rank u1 , . . . , um is by denition exactly the column rank of the matrix A thus we have shown in Section 5 that the column rank of a matrix A is left
invariant under elementary row transformations (Proposition 2.30). But what happens in the case of an elementary column transformation of a matrix. It appears that there is a beautiful analogy to the case of a row transformation. Without going into detail, the elementary column transformations of the coordinate matrix system of
of the system
u1 , . . . , um
b1 , . . . , bn
b1 , . . . , b n .
change the rank of the system (Proposition 2.30) it means that a basis
b1 , . . . , bn
Now the
b1 , . . . , b n
V.
A is derived from the coorA by an elementary column transformation, then there exists a basis b1 , . . . , bn such that the coordinate matrix of the system u1 , . . . , um with respect to the new basis b1 , . . . , bn is precisely equal to A . Thus it follows by Proposition 2.45
ommit this not too dicult proof here) means that if dinate matrix that
rank(u1 , . . . , um ).
the row rank is invariant under elementary column transformations. Combining this observation with Proposition 2.30 yields the following result about the row rank of a matrix.
Proposition 2.46.
the matrix
the end of Chapter 1: the row rank of the matrix satises the requirements stated by the problem.
Now there exists no essential dierence in the denition of row and column rank of a matrix. Let us denote by
which is the
n m-matrix
(88)
46
2. VECTOR SPACES
rankr tA = rankc A
and
rankc tA = rankr A
and that an elementary column (row) transformations of mentary row (column) transformation of
corresponds to a ele-
A.
Proposition 2.47.
from the matrix
m n-matrix A
into a
0 1
...
..
... 0
. . . . . . .. .
= 0
0 . . . . . . 0
...
... 0
0 1
Ir 0
(89)
by using only row transformations and possible column transformations of type II (that is exchanging two columns). Now it is evident how to continue from this form using elementary column transformations to transform this matrix into a matrix of the form
0 1
...
..
... 0
. . . . . . .. .
0 = 0
matrix
0 . . . . A := . . 0
...
... 0
0 1
Ir 0
0 0
(90)
r r-identity
Ir
0r
m, n.
way extends Proposition 1.5 from the the previous chaptere were we only allowed row transformations.
Theorem 2.48.
Let
m n-matrix
a matrix of the form (90) by using elementary row and column transformations.
47
in (90) holds
rankr A = rankc A = r.
Since the elementary row and column transformations do not change the row and column rank of a matrix this means that
Theorem 2.49.
the row rank of
Let
F.
Then
A,
that is
rankr A = rankc A.
Denition 2.50
eld
(Rank of a Matrix)
Let
F.
column rank) of
A,
in symbols
rank A := rankr A.
Let us collect the basic properties of the rank of a matrix: (1) The rank of a matrix pendent rows of (2) If
A. A is the coecient matrix of the system u1 , . . . , um of vectors of a vector space V with respect to an arbitrary basis of V , then
maximal number of linear independent columns of
rank A = rank(u1 , . . . , um ).
(3) The rank of a matrix is invariant under elementary row and column transformations. (4) Every matrix form
where
0 0 matrix A.
As an appli-
Ir 0
cation (and repetition) of our knowledge we have obtained so far we shall present an algorithm to compute the basis of a given subspace of a nite dimensional vector Let
V. V
F
of
of dimension
and let
e1 , . . . , en
be
a basis of
V.
u1 , . . . , um
of vectors in
V,
that is
U := span{u1 , . . . , um }.
Then the subspace
A = (aij ) of u1 , . . . , um
e1 , . . . , e n :
n
ui =
j=1
aij ej
(i = 1, . . . , m).
48
2. VECTOR SPACES
Chapter 1) using suitable elementary row transformations and column exchanges. By this we obtain naly a matrix of the form
C :=
where the matrix
Ir 0
B 0
over
(91)
is a certain
r (n r)-matrix
F.
It follows that
r =
rank A = dim U . U we have also also found a basis for U : b1 , . . . , br the vectors of V of which the coordinate vectors are precisely the rst r rows of the matrix C in (91). Then the system has rank r . If we revert the column exchanges which have been done to optain the matrix C in (91) then we get a system b1 , . . . , br of vectors from the subspace U which has rank r = dim U . Thus the system b1 , . . . , br is a basis of U . Note that b1 , . . . , br is indeed a system of vectors of U since the elementary transformations done to the system u1 , . . . , um do not lead out of the subspace U .
But in addition to the dimension of let us denote by
m equations
as in (6):
denote by
v1 , . . . , vn be the columns of the matrix A and b the right most column of the extendet coecient matrix. Then the mm tuples v1 , . . . , vm and b are vectors of F . Now the system of linear equation (92) is solveable in F if and only if there exists elements x1 , . . . , xn F such that
its extended coecent matrix. Let
x1 v1 + . . . + xn vn = b.
This is again equivalent that
b span{v1 , . . . , vn }.
tion 2.38 then states that this is the case if and only if
rank(v1 , . . . , vn , b) = rank(v1 , . . . , vn ).
Now the rst one is by denition the rank of the extended coecient matrix the latter is the rank of the simple coecent matrix
and
A.
Proposition 2.51.
only if
Let
cient matrix of the system of linear equations (92). Then (92) is solveable if and
rank C = rank A.
49
Proposition 2.52.
is exactly equal to
Let
be a eld and
F n.
Then
there exists a homogeneous system of linear equatinons such that its solution space
U. U
has a nite generating system
u1 , . . . , um .
of
u1 , . . . , u k Is 0
b1 , . . . , bn
C :=
with some
0 s m, n
where
is some
B 0 s (n s)-matrix.
We consider the
t
(Recall that
Is . B B
Reverting the column exchanges which have possible done to obtain the matrix we see that we may assume that the subspace form (93).
We consider now the homogeneous system of linear equations which has the coecent matrix
(t B, Ins ).
It follows from Proposition 1.9 from Chapter 1 that the columns of the matrix (93) is a basis after possible reordering of the unknown variables of the solution base. Therefor this solution space is equal to the given subspace
U.
CHAPTER 3
Linear Maps
Denition 3.1
eld
(Linear Map)
Let
and
F.
Then a map
f: V W
x, y V
and all
a F.
be given by
Examples.
(1) Let
f: V W
is linear since
f (x + y) = 0 = 0 + 0 = f (x) + f (y)
and
f (ax) = 0 = a0 = af (x) a F . This linear map, which maps every vector 0 of W is called the trivial linear map. Let V be a F -vector space and let c F be a xed element. Then the map f : V V given by f (x) := cx for every vector x V is linear since
for every of and
x, y V
(2)
x, y V
and
a, b, c, d F
x1 x2
:=
is linear as can be easily veried. (4) An example from calculus: dierential operator consider the innite many times dierentiable real valued function on
to its derivative
52
3. LINEAR MAPS
consider the
R-vector
space
I: C 0 ([0, 1]) C 0 ([0, 1]), f I(f ) := F f to the antiderivative F dened by F (x) := x f (t)dt is linear because the integration is linear. 0 (6) Let f : V W be a map between two F -vector spaces. Then f is a linear
which maps every function map if and only if
x, y V
and
a, b F .
In mathematics maps between objects which are compatible with the mathematical structure of those objects are often called homomorphism. Now the properties (94) and (95) state that a linear map is compatible with the linear structure of a vector space and thus linear maps are also called homomorphisms of vector spaces.
Lemma 3.2.
Let
and
W be vector spaces over the same eld F and let f : V W f (0) = 0 and f (x) = f (x) for every x V . f (0) = f (00) = 0f (0) = 0
eld
and
Proposition 3.3.
f: V V
F.
If
shown. Before we study linear maps in more detail we will classify them according to their mapping properties.
Let
V and W be two vector spaces over the same eld F . We call f : V W a monomorphism and a surjective linear map is
be a system of vectors of an arbitrary vector space
u1 , . . . , un
F.
f : F n V,
is a linear map from of
x1
. . .
x1 u1 + . . . + xn un
xn F to V . This map maps the canonical basis vectors e1 , . . . , en f (ek ) = uk (1 k n). The map f is apparently a monomorphism if and only if the vectors u1 , . . . , un are linearly independent. The map f is apparently an epimorphism if and only the u1 , . . . , un are a generating system of V . In other words f is an isomorphism if and only if u1 , . . . , un is a basis of V . Fn
to the vectors
53
Note that in the above example we have essentially dened the map by specifying the images of the standard basis vectors of following result holds.
f: Fn V
Proposition 3.5.
B
be a basis of
Let V and W be two vector spaces over the same eld F . Let V and f0 : B W an arbitrary map from the basis B to the vector space W . Then f0 extends in a unique way to a linear map f : V W . That is, there exists a unique linear map f : V W such that f (b) = f0 (b) for every b B .
Proof. Let us rst prove the uniqueness. Assume that
f (b) = f0 (b)
such that
for every
x F (B)
v=
Then by the linearity of
and
follows
f (v) =
bB
Therefore map ing
f (x(b)b) =
bB
x(b)f (b) =
bB
x(b)f (b) =
bB
and this means
f (x(b)b) = f (v). f =f
. Thus the linear extend-
f (v) = f (v)
for every
v V
is unique if it exists.
f: V W
f0 : B W dening f in the
vV
x F (B)
f (v) :=
bB
Due to the uniquensess of exercise.
x(b)f0 (b).
In other words the above proposition states that a linear map completely characterized by the images of an arbitrary basis
of
Example.
Assume that
m n matrix with coecients in a eld F . Denote by v1 , . . . , vn the A, which are then vectors of F m . Then there exists precisely f: Fn Fm
e1 , . . . , e n
Fn
to the vectors
v1 , . . . , v n .
If
xF
is an arbitrary vector
x1
x=
. xi ei = . . i=1 xn
54
3. LINEAR MAPS
f (x)
is given by
a11 x1 + a12 x2 + . . . + a1n xn n a21 x1 + a22 x2 + . . . + a2n xn f (x) = xi v i = . . . . . . . . . i=1 am1 x1 + am2 x2 + . . . + amn xn
2. Isomorphisms and Isomorphism of Vector Spaces Proposition 3.6. Let V , V and V be three vector spaces over the same eld F .
And let
f : V V and g: V V be two isomorphisms. Then the composite g f : V V and the inverse map f 1 : V V are isomorphisms.
Proof. Left as an exercise.
map
. Let V and W be two vector V is isomorphic to W (as F -vector spaces) if there exists an isomorphism f : V W . If V and W are isomorphic vector spaces then we denote this fact in symbols by V W . =
(Isomorphismus of Vector Spaces) spaces over the same eld
Denition 3.7
F.
Note the above dened relation has the following properties: If are vector spaces over the same eld (1) (2) (3)
V, V
and
F,
then
V V V
V, = V V V, = = V and V V = =
V V =
In mathematical terms the rst property means that the relation is reexive, the = means that is transitive. A relation which is reexive, symmetric and transitive = is also called a equivalence relation. The importance of the concept of isomorphic vector spaces is apparent: if second property means that the relation is symmetric and the third property =
f : V W is an V which is based
then
isomorphism of vector spaces then every property of objects of on the vector space structure is transfered automatically to the
W.
For example if
b1 , . . . , b n
is a basis of
proof since it is a consequence of the nature of an isomorphism. In future we will not justify such kind of claims by more then saying due to isomorphy reasons. For example due to isomorphy reason we have that if
u1 , . . . , un
are vectors of
V,
then
Examples.
in
v1 , . . . , vn be a linearly independent system of vectors f : V W is an isomorphism of F -vector spaces. Then f (v1 ), . . . , f (vn ) is a linearly independent system of vectors of W . Proof: Since v1 , . . . , vn is a linearly independent system of vectors it follos that they are all pairwise distinct vectors in V and since f is a monomorphism it follows that also f (v1 ), . . . , f (vn ) are pairwise distinct vectors. Let a1 , . . . , an F such that
(1) Let
a1 f (v1 ) + . . . + an f (vn ) = 0.
combination of the zero vector. Due to the lienarity of
(97)
We want to show that this linear combination is infact the trivial linear
we have that
55
Now
we get that
a1 v1 + . . . + an vn = 0.
f (0) = 0 and f (a1 v1 +. . .+an vn ) = 0 But since v1 , . . . , vn was assumed to of vectors it follows that a1 = . . . = f (v1 ), . . . , f (vn )
is a linearly
an = 0.
That is, the above linear combination (97) is actually the trivial
W. f
is only a monomorphism.
Note that we used in this proof only the fact that an isomorphism is a monomorphism! Thus the result is also true if (2) Let
v1 , . . . , v n W.
be a generating system of
an isomorphism of system of
F -vector spaces.
Then
Proof: We have to show that every combination of the vectors arbitrary vector. Since f y = f (x).
linear be an
xV
such that
it
x = a1 v1 + . . . + an vn
with some elements have that
a1 , . . . , an F .
we
yW
it follows that
W.
Note also here, that we needed not the full power of an isomorphism. We just used the fact that an isomorphism is an epimprphism. Thus the result is also true if (3) Let
phism of
v1 , . . . , vn be a basis of V and assume that f : V W is an F -vector spaces. Then f (v1 ), . . . , f (vn ) is a basis of W .
Proof: This follows now from the previous two results since a basis is a linear independent generating system. Note that here we need the full power of an isomorphism: a linear map
and
maps a basis of
to a basis of
if and only if
is an isomorphism.
Proposition 3.8.
B = (b1 , . . . , bn )
Let
be an ordered basis of
the eld
F.
Let
xn
is an isomorphism of vector spaces.
Proof. This follows straight from the considerations of the example on page 52
56
3. LINEAR MAPS
cB : V F n , x1 b1 + . . . + xn bn
is called the coordinate isomorphism of over the same eld
x1
. . .
B.
xn V
with respect to the basis We can now prove a very strong result about nite dimensional vector spaces
F,
Proposition 3.9.
same eld
Let
and
F.
Then
V W dim V = dim W. =
If
Proof. :
and
dim V = dim W .1
dim V = dim W , say both vector spaces are n-dimensional. B with n elements and W has a nite basis C with n n elements. Then we have the coordinate isomorphism cB : V F of V with respect n to the basis B and the basis isomorphism iC : F W of W with respect to the basis C . Its composite is apparently an isomorphism
Then
iC cB : V W.
Therefore
V W. =
But one can show that
V W. =
and a
V W = basis of W .
that
and
and let
is the set
ker f := {x V : f (x) = 0}
and the image of
im f := {y W : there
and
xV
such that
f (x) = y}. V
f: V W
W,
but share even more structure. We have namely the following result.
Proposition 3.11.
f: V W of W .
Let
and
and let
ker f
is a subspace of
and
im f
is a subspace
ker f is a subspace of V . First note that f (0) = 0 0 ker f . In partitcular ker f is not empty. If x1 , x2 ker f , then f (x1 + x2 ) = f (x1 ) + f (x2 ) = 0 + 0 = 0 and thus x1 + x2 ker f . If x ker f and a F , then f (ax) = af (x) = a0 = 0 and thus also ax ker f . Thus all the
Proof. We show rst that
and therefore
requirements of the supspace criterion of Proposition 2.6 are staised and it follows that
ker f
is a subspace of
V.
is a subspace of
im f
57
that
f (x1 ) = y1 and f (x2 ) = y2 . Then y1 + y2 = f (x1 ) + f (x2 ) = f (x1 + x2 ) im f x1 + x2 V . If y im f and a F then there exists a x V such that f (x) = y and we have that ay = af (x) = f (ax) im f since ax V . Thus all the
since
requirements of the supspace criterion of Proposition 2.6 are staised and it follows that
im f
is a subspace of
W.
and
Proposition 3.12.
f: V W (1) f (2) f
Let
and let
be a linear map. Then the following two statements hold: is a monomorphism if and only if is an epimorphism if and only if
ker f = 0. im f = W .
Proof. Note that the second claim is just the denition of an epimorphism.
x ker f .
Then
f (x) = 0 = f (0) but this means that x = 0 since ker f = {0} is the trival zero vector
space. : Assume that ker f = {0} and let x, y V such that f (x) = f (y). Then f (x y) = f (x) f (y) = 0 and thus x y ker f . But since ker f = {0} this means that x y = 0 and this again is equivalent with x = y . Therefore f is a monomorphism. Note that the above proposition simplies in a very essential way the verication whether a linear map is a monomorphism or not. monomorphism if for every
f : V W is a x = y . Now using the fact that f is a linear map it it enough to just verify that f (x) = 0 implies that x = 0!
By denition
x, y V
from
f (x) = f (y)
it follows that
Proposition 3.13.
that sets
is an
Let f : V W be a linear map of F -vector spaces. Assume n-dimensional vector space and that b1 , . . . , bn is a basis of V . If one
ui := f (bi )
for
1 i n,
then
dim(im f ) = rank(u1 , . . . , un ).
Proof. Due to the linearity of
a1 , . . . , an F
the equality
f
i=1
It follows that tion 2.38 of Chapter 2.
ai bi =
i=1
ai f (bi ) =
i=1
ai ui .
im f = span(u1 , . . . , un ).
Let
Denition 3.14.
of
and
W be arbitrary vector spaces (nite or innite diF . Let f : V W be a linear map. Then the rank rank f := dim(im f ).
is dened to be
Note that the above denition explicitely allows the case that map:
rank f = .
Further we have apparently the following two constraints on the rank of a linear
rank f dim W
If
and
rank f dim V.
is an epimorphism.
rank f = dim W f
58
3. LINEAR MAPS
If
is a nite dimensional vector space then one can show easily (left as an exercise)
that
rank f = dim V f
and the rank of a matrix
is a monomorphism.
We will soon see that there is a connection between the rank of a linear map
Theorem 3.15
. Let V and W be vector F . Assume that V is nite dimensional and let f : V W ker f and im f are nite dimensional subspaces of V and W
dim V = dim(im f ) + dim(ker f ).
(98)
ker f
of a nite dimensional space. Then there exists a linear complement Theorem 2.40, that is there exists a subspace
of
ker f
by
of
such that
V = U + ker f
Let
and
U ker f = 0. U,
that is let
g: U W
the restriction of
to the subspace
be the linear
map
following relations:
ker g = 0.
It follows that
dimensional space
g denes an isomorphism U = im f . Since U is a subspace of the nite V it follows that im f is nite dimensional, too. Furthermore we dim V = dim U + dim(ker f ) = dim(im f ) + dim(ker f ).
Corollary.
form
We can express the dimension formula for linear maps (98) also in the
f.
Proposition 3.16.
let
Let
and
f: V W
be a linear map. If
is a linear complement of
ker f
then
F and f maps
isomorphicaly onto
im f .
Again we will show that linear maps between nite dimensional have a very rigid behavior. We have the following result which is true for nite dimensional vector spaces but not for innite dimensonal vector spaces.
Proposition 3.17.
Assume that
(1) (2) (3)
Let
and
F.
and
f: V W
is a linear
f f f
HomF (V, W )
59
HomF (V, W )
In this section we will study the structure of linear maps which is essential to the deeper understanding of Linear Algebra. Linear maps are structure preserving maps between vector spaces which map vectors to vectors. But they can also be seen as vectors of suitable vector spaces. Let set
and
F.
WV
to
W.
FI
. If
f, g W
V W,
then we dene
f +g
(99)
v V.
Similar if
f WV
and
a F , then af
is given by
(100)
v V.
Apparently
f + g WV
Proposition 3.18.
af
Let
f, g: V W
u, v V
and
b F,
then
(f + g)(bv) = f (bv) + g(bv) = bf (v) + bg(v) = b(f (v) + g(v)) = b(f + g)(v)
and therefore
f +g
(af )(u + v) = af (u + v) = a(f (u) + f (v)) = af (u) + af (v) = (af )(u) + (af )(v)
and
af
is a linear map.
60
3. LINEAR MAPS
V W
is a subspace of
WV
follows now using the subspace criterion from the above considerations and
the fact that there exists always at least the trivial linear map
V W. F.
Then we
Denition 3.19.
denote by the
Let
and
F -vector
Note that if there is no danger of confusion about the underlying eld, then we might ommit the eld in the notation and just write
HomF (V, W ).
We come now to another essential point in this section. Let vector spaces over the same eld then the composite map
F.
If
f HomF (V, V )
and
V , V and V be g HomF (V , V )
gf
g f: V V ,
that is
g f HomF (V, V ).
g.
If there is
(g, f ) with f HomF (V, V ) and g HomF (V , V ) g f HomF (V, V ). We say that g f is the product of f no danger of confusion we write also gf instead of g f : gf := g f.
(101)
The so dened multiplication of linear maps satises rules which are very similar to the calculation rules we would expect from a multiplication. following distributive laws hold: First of all the
g (f1 + f2 ) = g f1 + g f2 (g1 + g2 ) f = g1 f + g2 f
where f, f1 , f2 HomF (V, V ) and g, g1 , g2 HomF (V , V ). HomF (V , V ), then the following associative law holds:
h (g f ) = (h g) f
the scalar multiplication. If
(104)
Moreover the multiplication of linear maps is in the following way compatible with
aF
F,
V we have the identity map idV : V V which is idV HomF (V, V ). For every linear map f : V V idV f = f idV .
If there is no danger of confusion then we may leave away the index the above relation (106) reads then
of
idV
and
id f = f id .
(107)
Note that the rules (104) and (106) are true in general for arbitrary maps. The rules (102), (103) and (105) are veried easily if one applies both sides of the equality to an arbitrary element of
xV
(g (f1 + f2 ))(x) = g((f1 + f2 )(x)) = g(f1 (x) + f2 (x)) = g(f1 (x)) + g(f2 (x)) = (g f1 )(x) + (g f2 )(x) = ((g f1 ) + (g f2 ))(x).
61
f: V V
and
g: W W
is only dened if
V = W! HomF (V, V ). In this case f, g HomF (V, V ) is always dened. HomF (V, V ) we can also dene a multiplication
(108)
Now let us turn to the special case of the vector space the the product on
f g := f g
of two elements
HomF (V, V ):
Proposition 3.20.
cation .
Let V be a vector space over the eld F . Then the HomF (V, V ) becomes a ring (with unit) under the addition + and the above dened multipli-
Proof. Recall that we have dened in the previous chapter on page 21 a ring
to be a set together with an addition and multiplication which satises all eld axioms except (M2) and (M4). Now the vector space axioms (A1) to (A4) are precisely the eld axioms (A1) to (A4). The calculation rule (104) veries the eld axiom (M1). The identity map
idV
is the identity element of the multiplication by the calculation rule (106) and Finally the eld axiom (D) is veried by the
this veries the eld axiom (M3). calculation rule (102) and (103).
Denition 3.21.
ring
An linear map
f: V V
HomF (V, V )
is also denoted by
F -vector
space
V.
End(V ) instead of EndF (V ). Note EndF (V ) is in general not commutative nor is EndF (V ) in general zero
EndF (V ) =
HomF (V, V )
Denition 3.22
Let
be a eld. A ring
F -vector
a F , g, f R,
or just an
FF-
algebra.
With this denition and since the endomorphism ring of this section in the follwing compact form.
EndF (V )
is also a
vector space which satises the calculation rule (105) we can summerize the results
Proposition 3.23.
phism ring
Let
F.
EndF (V )
is in a natural way an
F -algebra.
and
F,
say and
dim V = n
dim W = m
62
3. LINEAR MAPS
and
m.
Furthermore let
f: V W
be a linear map from ordered basis map
V to W . We x C = (c1 , . . . , cm ) of W . ai := f (bi )
an ordered basis
B = (b1 , . . . , bn )
of
and a
(i = 1, . . . , n) f.
We can describe the images
b1 , . . . , b n
a1 , . . . , an W as W . We have then
c1 , . . . , c m
of
f (bi ) = ai =
j=1
with some unique elements
aji cj
(i = 1, . . . , n)
(109)
aji F .
for
and a basis
for
f: V W
is completely determined by
the matrix
(110)
m n-matrix c1 , . . . , cm
W.
Denition 3.24
m n-matrix A dened by coordinate matrix of the linear map f : V W with respect (b1 , . . . , bn ) and C = (c1 , . . . , cm ) of V and W .
rows (see page 36). Since we use to write vectors in to call the matrix the basis
B =
Note that in the previous chapter we did arrange the coordinate coecients in
Fn
in columns it is customary
t A as we a1 , . . . , an with
c1 , . . . , cm . V
and
f : V W with respect to the bases B and C of the W respectively is nothing else than the coordinate matrix of images f (b1 ), . . . , f (bn ) with respect to the basis C of W . f : R5 R4 given x1 x1 + 3x2 + 5x3 x2 3x1 + 9x2 + 10x3 x3 2x2 + 7x3 x4 2x1 + 8x2 + 12x3 x5
is by
Example.
+ + + +
This map is apparently linear and its coordinate matrix with respect to the standard bases of
R5
and
R4
1 3 5 2 3 9 10 1 A= 0 2 7 3 2 8 12 2
0 2 1 1
63
Note that sometimes a too precise notation is more hindering than helpful. If we denote the coordinate matrix of the linear map
by
cB (f ). C
of the dependency of the coordinate matrix on the choice of bases!
But much more important (!) than this notation is that one is in principle aware
Example.
R5
1 3 c1 := , 0 2
calculates to
B be the standard C of R4 : 0 0 0 1 1 0 c2 := , c3 := , c4 := 0 1 1 0 0 1 f
with respect to the bases
basis of
and
1 3 5 0 2 2 A = 0 0 5 0 0 0
2 0 2 1 5 2 0 0
(compare this with the example on page 10 in Chapter 1). We are now able to explain the connection of the rank of a linear map dened in Denition 3.14 and the rank of a matrix as dened in Chapter 2.
as
Proposition 3.25.
and by
Let and
V W
and
respectively. Let
f: V W
matrix of
and
C.
Then
rank f = rank A.
Proof. We have
(ai = f (bi )
for
1 i n)
space
f : V W of an n-dimensional F -vector V to an m-dimensional F -vector space W is after choosing bases for V and W completely determined by a m n-matrix. In order to be able to describe
We have seen that a linear map the connection between linear maps and their coordinate matrices more precisely we shall rst study matrices a bit more in general. Sofar we have just said that a elements of
Denition 3.26
Denote by
(Matrix)
Let
and
the sets
matrix
over
we understand a map
A: M N F, (i, j) ai,j
which asigns each pair element
ai,j
of the eld
(i, j) F.
of natural numbers
1 i m
and
1 j n
an
64
3. LINEAR MAPS
m n-matrices
F .
is denoted by (112)
F
of
m,n
:= F
M N
aij
instead
ai,j .
From (112) follows that the set
F m,n
F -vector space
structure (see Example 5 on page 23 in the previous chapter). The addition and scalar multiplication of That is
m n-matrices over F
b11 b12 b1n a1n a2n b21 b22 b2n . . + . . . . . . . bm1 bm2 bmn amn a11 + b11 a12 + b12 a21 + b21 a22 + b22 = . . . am1 + bm1 am2 + bm2 a1n ca11 a2n ca21 . = . . . . . amn cam1 m n-matrix A ca1n ca2n . . . camn mn-tuple
and if
as in (110) the
F
In particular we have that
m,n
F -vector F mn . =
Theorem 3.27.
eld
F
If
W.
to the
Let V and W be nite dimensional vector spaces over the same dim V = n and dim W = m. Let B be a basis of V and C be a basis of we assign each linear map f : V W its coordinate matrix c(f ) with respect bases B and C , then we obtain an isomorphism with
F -vector m n-matrices
a linear map follows that the map
V W
onto the
F -vector
space of all
F.
describes the linear map
f: V W
c
If
Corollary.
of dimension
V m,
is a then
F -vector
space of dimension
and if
is a
F -vector
space
65
(113)
and
and
this isomorphism is not suitable for identifying linear maps bases over another one.
f: V W
with their
coordinate matrices. There is general simply no canonical way to choose one set of The situation is dierent if one considers linear maps from
Fn
to
F m.
For
those vector spaces exists a natural choice for a basis, namely the standard basis of
Fn
and
F m.
A := c(f )
F m.
HomF (F n , F m ) = F m,n .
Note the dierence between (113) and (114): the rst one is an isomorphism of vector spaces, the latter an equality (after identication). The equality of (114) is much stronger than the isomorphism of (113)! Using this identication we can express the denition of the coordinate matrix of a linear map in the special case of
V = Fn
and
W = Fm
Proposition 3.28.
The columns of a
mn-matrix A over F are in turn the images e1 , . . . , en of F n under the linear map A: F n F m .
We can visualize the content of Theorem 3.27 with the help of the following commutative diagram
VO iB Fn
bases
/ Fm
/ W O iC iB
and
(115)
iC
for
and
for
means that regardless which way one follows the arrows in (115) from corresponding composite maps agree. That is
Fn
to
the
f iB = iC A.
One proves this by applying both sides of this equality to an arbitrary vector the standard basis of
ei
of
Fn
A = i1 f iB C
and
f = iC A i1 . B
and if one use the coordinate isomorphisms
cB
and
cC ,
then
A = cC f c1 B
66
3. LINEAR MAPS
or if one wants to use the notation introduced by (111) then one has
cB (f ) = cC f c1 . C B m n-matrices over a eld F with HomF (F n , F m ) of all linear maps F n F m the natural questions n arises how to calculate the image of a vector x F under a given matrix A = (aij ).
Now since we identied the vector space of The answer to this question is given in the following result.
Proposition 3.29.
that
Let
A = (aij ) be a m n-matrix over the eld F . Then we A: F n F m explicitely in the following way: assume
with
y = A(x)
following formula
n
of
by the
yi =
j=1
Proof. Let
aij xj
(116)
e1 , . . . , en
F n.
A(ej ) =
If we apply the linear map
a1j
. . .
, j = 1, . . . , n.
amj A
to the element
x=
j=1
then we get due to the linearity of
xj ej
A
n n
y = A(x) = A
j=1
and thus
xj ej =
j=1
xj A(ej )
y1
xj
a1j
. . .
a1j xj
. . = . ym
j=1
amj
. . . j=1 amj xj
From this one can then read then the claim. Compare (116) with (109) on page 62! Note that we can write now a linear system of equations in a very compact way. Assume that we have given a non-homogeneous system of in
linear equations
unknown variables
x1 v1 + . . . + xn vn = b.
Let
A be the
m n-matrix
F.
A(x) = b.
Apparently the system is solveable if and only if solutions
b im A. In case b = 0 the M form a linear subspace of F n , namely M = ker A, and its dimension is dim(ker A). In the end of this chapter see Section 12 on page 88 we will have a
closer look on the content of Chapter 1 in this new view.
67
Let
be a eld and
A F l,m
and
B F m,n
composite map
AB := A B
which is an element of
l,n
Note that the content of the above denition can be visualized with the following commutative diagram:
F m? ?? ? ??A B ?? ?? / Fl Fn
AB
Again arises as a natural question how we can compute the the coecents of the product matrix
AB
and
B.
The answer
Theorem 3.31
formula
l,m (Formula for the Matrix Product) Let A = (aqp ) F and m,n l,n F be two given matrices. Let C := AB F be the product
cij
of the matrix
cij =
k=1
Proof. Let
aik bkj ,
1 i l, 1 j n. F n.
(117)
e1 , . . . , e n
By denition
we have then
j = 1, . . . , n.
of
(118)
C(ej )
ej
C = AB .
By
B(ej )
B(ej ) =
b1j
. . .
, j = 1, . . . , n.
(119)
bmj
We can now use the result of Proposition 3.29 to determine the coordinates of the images
C(ej )
of the vectors
B(ej )
m
A.
yields indeed
cij =
k=1
and this is true for
aik bkj ,
1il
1 j n.
68
3. LINEAR MAPS
AB
of
A with B
B.
A: F n F m
y = A(x)
with
x=
x1
. . .
n F ,
y1
(120)
xn
n
. y = . F m, . ym
yi =
i=1
One can consider the vectors
aij xk ,
i = 1, . . . , m.
(121)
can be considered as a
x and y in (120) as matrices of a special form, namely n 1-matrix and y can be considered as a m 1-matrix. product Ax of the matrices A and x is dened and (121) states y = Ax.
A special case of Theorem 3.31 is the multiplication of a called row vector ) with a have then
1 m-matrix
(also
m 1-matrix x1
. . .
In this case we
(a1 , . . . , am )
ai xi = a1 x1 + . . . + am xm .
i=1
(122)
xm
Thus we can also describe the content of Theorem 3.31 as follows: the coecient
cij of the matrix C = AB is the product of the i-th row of the matrix A with the j -th column of the matrix B . Here the product of the i-th row and j -th column of A and B is calculated by the formula (122).
Example.
2 1 6 3 2 3 1 = 5 3 22+61+33 12+31+53 = 19 20
Consider the linear map f : V W of the n-dimensional F -vector space V to m-dimensional F -vector space W . Let B = (b1 , . . . , bn ) be a basis of the vector space V and C = (c1 , . . . , cm ) be a basis of the vector space W . Let the
x=
i=1
be an arbitrary vector of coordinate vector
xi bi b1 , . . . , b n
the
V.
Then
x :=
that is map
x1
. . .
n F . y
the image of
xn x = cB (x) =
that is
i1 (x). B
Denote then by
f,
y := f x.
69
Denote by
y :=
y1
. . .
m F
ym
the coordinate vector of y with respect to the basis c1 , . . . , cm of W , that is y = cB (y) = i1 (y). Let A = (aij ) the coordinate matrix of f with respect to the bases B B and C . Then it follows from the commutative diagram (115) and the above considerations that
y = A. x
That is for the coordinates hold the equation (121), namely
yi =
i=1
aij xk ,
i = 1, . . . , m.
Apparently we have the following result about the composite map of two linear maps.
Proposition 3.32.
matrices.
tween nite dimensional vector spaces corresponds to the product of their coordinate More precisely: Let same eld
V, V
and
F,
say
dim V = n,
Furthermore let
dim V = m
and
dim V
= l.
f: V V
and
the composite
B be a basis of V , C a basis of V and D a basis of V . If g : V V are linear maps, then for the coordinate matrix A of map g f : V V with respect to the basis B of V and D of V A =AA
where
and
and
is
and
D.
We can visualize the content of the above proposition in a nice way with a commutative diagram, namely
VO iB Fn
/ Fm
/ V O iC
/ V O iD / Fl iB , iC
(123)
and
iD .
To be very precise, the previous proposition states that actually the diagram
VO iB Fn
gf
A =A A
/ V O i D / Fl
(124)
70
3. LINEAR MAPS
commutes, that is
(gf ) iB = iD (A A).
follows apparently from the commutativity of the diagram (123). Formaly we can convince oursel with the following calculation:
cB (gf ) = cC (g) cB (f ). D D C
One can use the follwing help to memorize this:
(125)
C B B = D D C
Now from (114) follows that the calculation rules for linear maps transfer in a natural way to matrices. In particular we have the following rules
in case the matrix product is dened, that is matrices have the right dimensions. Of course one can verify those rules for matrix multiplicaton just by using the formula for matrix multiplication as given in Theorem 3.31, The rules (126), (127) and (128) are then evident, and even the verication of (129) is not dicult but rather a bit cumbersom and as said unnecessary. Note that the role of the identity map is taken in the case of the matrices by the
In :=
1 1
0
.. .
0
whose colums are in turn the elements every
1 e1 , . . . , e n
of the canonical basis of
F n.
For
AF
m,n
holds
Im A = A = AIn .
Note that it is often customary to leave away the index identity matrx
(131)
In ,
also be written as
IA = A = AI.
(132)
n-dimensional F -vector
space
V,
that is endomorhpsism
f: V V
of
and their matrix description. It follows from Proposition 3.23 that the
n n-matrices
F
form a
n,n
= HomF (F , F ) = EndF (F n )
has numerous applications and this motivates
F -algebra.
This
F -algebra
EndF (V )
71
Denition 3.33.
Let
be a eld and
n n-matrices
is
Mn (F ) := F n,n . n n-matrices over F is sometimes also called n over F . Apparently the identity element of the multiplication in Mn (F ) is the n n-identity matrix In . The elements of Mn (F ) are also called square matrices (of degree n). As a F -vector space Mn (F ) has 2 dimension n . We use the notation of Theorem 3.31 but this time we set W = V and B = C . In particular this means that we will use one and the same basis B = (b1 , . . . , bn ) for the range and the image for the matrix description of any enodmorphism f : V V . The coordinate matrix A = (ars ) of the endomorphism f with respect to the basis B is then dened by
Note that the algebra of the the (full) matrix ring of degree
f (bi ) =
j=1
aji bj
(i = 1, . . . , n). A
(133)
is the endomorphism
of
Fn
V cB Fn
where the vertical arrows are the coordinate isomorphisms notation introduced in (111) we have
A = cB (f ). B
But we shall also use the notation
A = cB (f )
to denote the coordinate matrix the basis
A of the endomorphism f : V V with respect to B of V . If now f and g are endomorphisms of V which have the coordinate matrices A and A respectively with respect to the basis B , then the coordinate matrix of the endomorphism f g = f g with respect to the same basis B is given by the product AA . Using the above introduced notation we get therefore cB (f g) = cB (f )cB (g).
This is a straight consequence of Proposition 3.32, but compare the result also with (125)! Furthermore hold
cB (idV ) = In .
In order to summarize the matrix description of endomorphisms in a suitable compact way let us dene rst what we mean by an ismorphism of
F -algebras.
72
3. LINEAR MAPS
Denition 3.34.
: R S
(1) two conditions:
Let
R and S
F.
A linear map
is said to be a homomorphism of
F -algebras
(f g) = (f )(g) for all f, g R. (1) = 1. If is bijective, then is called an isomorphism of F -algebras (and then also the 1 inverse map : S R is an isomorphism of F -algebras). Two F -algebras R and S are said to be isomorphic (as F -algebras) if there exists an isomorphism : R S of F -algebras and in symbols this fact is denoted by R S . =
(2)
Theorem 3.35.
B = (b1 , . . . , bn )
Let
be a
be a basis of
and let
cB : EndF (V ) Mn (F ),
which assigns each endomorphism basis
of
V,
is an isomorphism of
EndF (V ) Mn (F ) =
Note that we did know already that
F -algebras).
EndF (V ) Mn (F ) as vector spaces but = now we also know that they are also isomorphic as F -algebras. This is a stronger result since a vector space isomorphism between F -algebras is not necessarily a isomorphism of F -algebras!
g: W V
such that
g f = idV
Proof. :
and
f g = idW .
(134) a is is
If
f: V W
is an isomorphism, then
is apparently bijective
and thus an isomorphism. For nite dimensional vector spaces we have the following result.
Proposition 3.37.
(1)
Let
be a
n-dimensional
and
F.
is an isomorphism.
m=n
and
rank f = n. V
and
is the coordinate matrix of f with respect to some bases of n m then the linear map A: F F is a isomorphism.
W,
m=n
(2): If f is an isomorphism, then W = im f and dim W = rank f = dim(im f ) = dim W = n. (2) (1): We have rank f = dim V and rank f = dim W . Then from the notes made after the Denition 3.14 it follows from the rst equality that f is a monomorphism and from the latter equality that f is an epimorphism. Thus alltogether we have shown that f is an isomorphism.
Proof. (1)
and
9. CHANGE OF BASES
73
(1) (3): From the commutativity of the diagram (115) we have that A = cC f iB where iB is the basis isomorphism with respect to the basis B of V 1 and cC = (iC ) is the coordinate isomorphism of W with respect to the basis C of W . Thus A is a isomorphism since it is the composite of three isomorphism (Proposition 3.6). (3)
(1):
f = iC AcB
as the com-
Now Proposition 3.37 states that a linear map if and only if coecents of
n = m
A has A1 . We
A1
A1
A.
Denition 3.38.
A: F
n
A F n,n
A1
of an invertible matrix
: F F
Note that apparently the inverse matrix of an invertible matrix is again invertible. Furthermore note that we have from the commutativity of diagram (115) that if
f: V W
is an isomorphism, then
cC (f 1 ) = (cC (f ))1 . B B
9. Change of Bases
Now we have already many times repeated that the coordinate matrix of a linear map closely. vector spaces
and
of the
F.
there exists
Let V be an n-dimensional vector space over B = (b1 , . . . , bn ) and B = (b1 , . . . , bn ) be two bases of V . Then unique elements sji F such that Let
bi =
j=1
Then the
sji bj ,
i = 1, . . . , n.
(135)
With the help of determinants we will be able to formulate the Cramer rule for matrix
inversion.
74
3. LINEAR MAPS
VO iB Fn
to the bases
idV
/ Fn
/ V O iB
(136)
and
B of the vector space V . Thus S is just the coordinate matrix idV with respect to the bases B and B (note the order in which S = cB (idV ). B
(137)
S is invertible, that is S: F n F n
is an isomorphism. Now from the commutativity of the diagram (136) one concludes
cB = i1 B
and
cB = i1 B
cB
Fn
commutes. Thus for every vector
V ? ?? ??cB ?? ? S / Fn
n
v=
i=1
in
xi bi =
i=1
x i bi
holds
that is
x1
x1
. . . =S . . . xn xn
So we have for the coordinates of mation equation
and
the transfor-
xi =
j=1
sij xj ,
i = 1, . . . , n.3
Note further that from the commutativity of the diagram (136) it follows that if
to
, then
S 1
matrix from
to
B.
Consider still the notations and conditions of Denition 3.39. precisely one linear map
f: V V
transition matrix
f (bi ) = bi for 1 i n (Proposition 3.5). It follows from (135) that S of B to B is at the same time the coordinate matrix of endomorphism f of V with respect to the basis B , that is
such that
the the
S = cB (f ) = cB (f ). B
3
sij
in this sum are indeed correct! Compare this with (135) where the
coecients
sji
are used!
9. CHANGE OF BASES
75
B S 1
to the basis
with respect to
B,
that is
S 1 = cB (g) = cB (g). B cB (f ) = I = B B (g) and if one applies f to (135) one obtains S = cB (f ). Moreover, the transition matrix from the canonical standart basis e1 , . . . , en of F n to an arbitrary basis s1 , . . . , sn of F n is apparently the matrix
Furthermore we obtain straight from the denition the equalities
B cB
S = (s1 , . . . , sn )
which columns are precisely the vectors
s1 , . . . , sn .
A = T 1 AS
where
(138)
to
S is C.
to
and
VO ? ?? ?? idV ?? ?? ? VO
iB iB n
/ W O
/ W O idW
iC A
iC
(139)
Fn
F ? S
/ Fm _?? ?? ??T ?? ?? / Fm
The small inner square of this di-
A.
A.
trapezoid is trivially commutative. The left and right trapezoids are the commutative diagrams (136). Now the verication of the commutativity of remaining lower trapezoid is simple:
A = (iC )1 f iB = T 1 i1 idW f iB C =T =T =T
1 1 1
(outer square) (right trapezoid) (upper trapezoid) (left trapezoid) (inner square)
i1 C i1 C
f idV iB f iB S
AS
76
3. LINEAR MAPS
If one uses the notation (111) we can using (137) write the content of Theorem 3.40 in the intuitive form
cB (f ) = cC (id) cB (f ) cB (id) C C C B
since
(140)
S = cB (id) B
and
T = cC (id). C
Applying Theorem 3.40 to matrices seen as linear maps we get the next result.
Proposition 3.41.
bases given by
A be a m n-matrix with coecients in the eld F . Then A of the linear map A: F n F m with respect to arbitrary (s1 , . . . , sn ) and (t1 , . . . , tm ) of the vector spaces F n and F m respectively is
Let
A = T 1 AS
where
(141)
is the
n n-matrix
with columns
s1 , . . . , sn
and
is the
m m-matrix
the coordinate Now the claim
with columns t1 , . . . , tm .
Proof. With respect to the canonical bases of
Fn
and
A: F n F m
Fm A.
of the proposition follows from Theorem 3.40 and the note on page 75 regarding the interpretation of the transition matrix from the standard basis to an arbitrary basis. We shall re-state Theorem 3.40 in the special case of an endomorphism and that we consider only coordinate matrices with respect to the same bases for domain and co-domain of the endomorphism.
Theorem 3.42
F B
with
dim V = n.
is given by
the coordinate
. Let V be a vector B and B be bases of V . If f has with matrix A, then the coordinate matrix A of f
Let
(142)
A = S 1 AS
where If phism
to
A is a n n-matrix over F , then the coordinate matrix A of the endomorA: F n F n with respect to an arbitrary basis s1 , . . . , sn of F n is A = S 1 AS
where
is the
n n-matrix
s1 , . . . , sn .
and
A = T 1 AS
then we may see the matrices other.
and
A and A to belong to one and the f : V W (with respect to dierent bases). Similar is true for n,n n,n matrices A and A in F . If we have for some invertible S F the A = S 1 AS
(144)
77
then
phism
A and A can be seen as coordinate matrices of one and the same endomorf : V V (with respect to dierent bases of V ). Now these considerations
Let
A, A F
n,m
is equivalent with
we
AA,
if there exists invertible matrices equality (143) is satised. (2) If
S F n,n
and to
T F m,m
in
A, A F n,n
A is similar
AA, S F n,n
Proposition 3.44.
(2)
A, A F m,n are equivalent if and only if there exists a linear map f : V W between an n-dimensional F -vector space V and an m-dimensional F -vector space W such that both A and A appear as coordinate matrices of f with respect to suitable bases for V and W . n,n Two matrices A, A F are equivalent if and only if there exists an endomorphism f : V V of an n-dimensional F -vector space V such that both A and A appear as coordinate matrices of f with respect to suitable bases for V .
(1) Two matrices and A are coordinate matrices of one and the A A according to Theorem 3.40. : We assume that A A , that is there exists invertible matrices S F n,n and T F m,m such that A = T 1 AS . Let (s1 , . . . , sn ) be the system of the columns of the matrix S and let (t1 , . . . , tm ) be the system of the columns of the matrix T . Then according to Proposition 3.41 the n linear map A: F F m has the coordinate matrix A with respect to n m the bases (s1 , . . . , sn ) and (t1 , . . . , tm ) of F and F respectively. But the same linear map has the coordinate matrix A with respect to the standard n m bases of F and F . Thus there exists a linear map such that both A and A appear as the coordinate matrix this linear map and this concludes the If same linear map (1) :
Proof.
then
proof of the rst part of the proposition. (2) The second part of the proposition is proven the same way as the rst part. Note that the relation on the following three statements.
F m,n
A, A , A F n,m
hold
A A (reexivity) A A A A (symmetry) (3) A A and A A A A (transitivity) n,n Similarly the similarity relation on F is an equivalence
(1) (2) matrices
relation.
Now the natural question is how can one decide easily whether two
m n-
78
3. LINEAR MAPS
Proposition 3.45.
Let
be a eld. Then the following two statements are true. matrix of the form
(145)
where the upper left part of this matrix is the is a certain natural number
(2) The matrices
and
of
r r-identity matrix and r 0 r m, n. In this case rank A = r. F m,n are equivalent if and only if rank A =
rank B .
Proof.
A: F n F m . Set r := dim(im A) = rank A. Then dim(ker A) = n r. Let br+1 , . . . , br be a basis of ker A. By the Basis Extension Theorem 2.36 we can extend this to a basis b1 , . . . , br n of F . Set ci := Abi for i = 1, . . . , r . Then c1 , . . . , cr is a linear indepenm dent subset of F and again using Theorem 2.36 we can extend this set m to a basis c1 , . . . , cm of F . Then we have
(1) Consider the linear map
A(bi ) = ci
(1 i r) Fn
and
and
A(bi ) = 0 (r + 1 i n). A
with respect to the above
Thus the coordinate matrix of the linar map constructed bases for that
Fm Ir 0
A
with (2) :
0 . 0
(146)
r = rank A.
If A B then by Proposition 3.44 there exists a linear map f : V W such that both matrices A and B appeare as coordinate matrices of f with respect to suitable bases for V and W . Then rank A = rank f = rank B by Proposition 3.25. In particular
A
if
Ir 0
0 0
r = rank A
: If
and this completes the proof of the rst part of the propo-
sition.
both the
r := rank A = rank B , then by the rst part it follows that B are equivalent to a matrix of the form (146). But then by transitivity of the relation it follows that A B . A
and
two
F m,n
are equivalent if and only if they have the same rank. More over Therefore every equivalence
the proposition states, that in every class of matrices which are equivalent to each other exists precisely one matrix of the form (145). class of such matrices can be labeled with a representative of such a class, which
79
Ir 0
the class of all
0 , 0 r
over the eld
a representative which has a very simple form. This simple matrix which represents
m n-matrices
with rank
A natural similarity.
How would a normal form for such a class look like? to give an answer to this problem.
harder to solve and in this lecture we will not develope the tools which are needed
aR
is invertible (in
R)
if there exists a
we say that
ab = 1
and
ba = 1. R
by
R ,
that is we dene
R := {a R : a
The elements of
is invertible}.
are called the units of the ring R. If a is a unit and b R ab = 1 and ba = 1, then we call b the (multiplicative) inverse of a. Note that if a R is invertible then its multiplicative inverse is uniquely dened. It is 1 customary to denote this unique multiplicative inverse element of a by a . such that
Examples.
(2) (3)
Z. The only two invertible ele1. Thus Z = {1, 1}. Let F be a eld. Then F = {x F : x = 0}. Let V be a F -vector space. Then EndF (F ) is a ring and EndF (V ) is precisely the set of all isomorphisms f : V V (see Proposition 3.47).
ments in
are
and
If a and b are units R, that is the set R is closed under 1 1 the multiplication of R. This is due to ab(b a ) = aa1 = 1 and likewise 1 1 1 1 ba(a b ) = bb = 1. In particular (ab) = b1 a1 (note the change in the 1 order of a and b!). Furthermore 1 R and apparently a R for every a R . Finally, since the multiplication on R is associative it follows that the multiplication restricted to the units of R is also associative. A set G together with a law of composition We make the following algebraic observations about then also their product of
R :
R,
ab
is a unit of
G G G, (x, y) xy
which staises precisely the above conditions is called in mathematics a group. More preciesly we have the following denition.
80
3. LINEAR MAPS
Denition 3.46
(Group)
A group is a tuple
G = (G, )
consisting of a set
: G G G, (x, y) x y
(called the law of composition ) if the follwoing three group axioms are satised: (G1)
x, y, z G. e G (called the identity element of G) such that x e = x and e x = x for all x G. For every x G there exists an element y G (called the inverse element of x) such that x y = e and y x = e.
for every for every
(x y) z = x (y z)
If
xy =yx
x, y G,
is called abelian .
Examples.
(2) Let (3) Let (4) Let
addition.
F V R
be a eld. Then
thermore
of vectors.
the multiplication of
R.
After this short excurse into the world of Algebra we return to Linear Algebra. Given an arbitrary vector space ring
over a eld
EndF (V )
of
V.
Proposition 3.47.
(1) (2)
Let
f: V V
V.
f f
EndF (V )
of
V.
g: V V
such that
being invertible in
f precisely then an isomorhphism if there exists g f = id and f g = id. But this is equivalent EndF (V ).
Denition 3.48
EndF (V )
of the
F -vector
is denoted by
GLF (V ) := EndF (V )
and is called the General Linear Group of the of
F -vector space V .
V = Fn
we set
GLn (F ) := Mn (F ) = EndF (F n )
and this group is called the General Linear Group (of degree An element
f GLF (V ), V.
Let
that is an isomorphism
eld
F.
called an
Proposition 3.49.
be an
n-dimensional
F.
then
GLF (V ) GLn (F ) =
4
(isomorphism of groups).
81
G and G are isomorphic (as groups) f : G G , that is a bijective map such that
cB : EndF (V ) Mn (F ) is an isomorphism of F -algebras. In particular cB maps isomorphism of V to invertible matrices of F . Apparently cB induces then an isomorphism of groups GLF (V ) GLn (F ).
Because of the above result it is now possible to restrict our attention without any loss of generality to the groups of nite dimensional columns
F -vector
spaces.
v1 , . . . , v n ,
that is the
an elementary column
A be a m n-matrix over a eld F with vi (1 i n) are all vectors of F m . We consider transformation of type I. Let us denote by Uij (a) precisely
(. . . , vi , . . . , vj , . . .) (. . . , vi , . . . , vj + avi , . . .)
of the system of vectors the vector
the vector
vj
in
(v1 , . . . , vn )
by
T ej = ej + aei ,
where
T e k = ek
(for
k = j ),
n
(147)
e1 , . . . , e n
k=j
we
F . Then we have for the AT by denition AT ej = A(ej +aei ) = get AT ek = Aek = vk . That is we have
(for
(AT )ek = vk
k = j ).
(148)
Now recall the following (compare this with Proposition 3.28): the columns of a
m n-matrix C over F are in turn the images of e1 , . . . , en under the linear map C: F n F m . From (148) it follows that the elementary column transformation Uij (a) is obtained by multiplying the matrix A from right with the n n-matrix . . 1 . . . 1 . 0 ... a (n) . .. (149) Tij (a) := . . . 1 . .. . . 0 .
. . .
1
That is, if the
which has on the main diagonal only ones 1 and otherwise only zeros except the entry in the i-th row and coecients of the matrix
j -th column where we have the element a. (n) Tij (a) are denoted by tkl then we have 1 if k = l, tkl = a if k = i and l = j , 0 otherwise.
82
3. LINEAR MAPS
Proposition 3.50
over a eld
Let
be an
m n-matrix
F.
Denote by
(v1 , . . . , vn )
(u1 , . . . , um ) the system of its rows. An elementary column transformation Uij (a) of A that is replacing of the j -th column vj of A by vj + avi is obtained by (n) multiplying the matrix A with the special matrix Tij (a) from the right. Likewise
the eect of multiplying the matrix row transformation
Uji (a),
that is
(m)
by
ui + auj .
Proof. We have already shown that the proposition is true for column trans-
formations.
Tij (a)A.
(m)
Denition 3.51
(149) for
(Elementary Matrices)
i=j
and
aF
are called
n n-elementary
One obtains directly from the denion (147) the following result about elementary matrices.
Proposition 3.52.
matrices over a eld
n n(150) (151)
are satised:
(a, b F )
1i=jn
and
n n-matrices is invertible, that is Tij (a) GLn (F ) a F . For the inverse of an elementary n n-matrix
(152)
If one applies successively elementary column transformation to a matrix then by Proposition 3.50 this corrsponds to successively multiplying
from the
T1 , . . . , Tr :
(153)
The set
n)
over
n n-matrices F. SLn (F )
is called the
Proposition 3.54.
linear group
GLn (F ).
83
H of G.5
a group
T1 T2 Tr
T1 T2 Tr
is apparently
SLn (F ).
SLn (F ).
the associativity of the matrix product. Due to (151) we have inverse element of
T1 T2 Tr
is given by
1 1 1 (T1 T2 Tr )1 = Tr T2 T1
and it is therefore also an element of
is a group under
A GLn (F )
rank A = n.
We try solve the task to transform the matrix
into a matrix of a as simple form as possible, and this only by using the fact (155) and row transformations of type I. To avoid triviality we may assume that Since the column rank of
A is not empty. Thus we can achive by using a suitable row transformation of type I that the second entry of the rst column of A that is a21 is dierent from 0. 1 Then by adding a21 (1 a11 )-times the second row to the rst row we get a matrix A with a11 = 1.
Now we can eliminate the coecients in the rst column below the element suitable row transformations and get a matrix
a11
by
of the form
1 0 A = . . . 0
with a
... C
n 1.
If (156)
(n 1) (n 1)-matrix C A
n 3,
then we
can apply the above described algorithm to the matrix that the row transformations of applied to the matrix
C.
1
.. . .. .
1 d
(d = 0).
(157)
Compare the denition of a subgroup with the denition of a linear subspace, see Deni-
84
3. LINEAR MAPS
The main diagonal of this matrix consist only of ones except the last entry, which is an element
dF
entries in this matrix below the main diagonal are all zero. Apparently we can only with the help of row transformations of type I transform this matrix into the following form
1 Dn (d) :=
1
.. . .. .
1 d
(d = 0),
(158)
which is a diagnoal matrix with only ones on the diagonal except that the last element of the diagonal is an element dierent from zero. We have that
d F from which we only know that Dn (d) GLn (F ). Thus we have shown: n n-matrix d F. S SLn (F )
over a eld
it is
Theorem 3.55.
If
is an invertible
F,
then we can
transform this matrix by row transformations of type I into a diagonal matrix of the special form (158) for some non-zero In other words, if
A GLn (F ),
and a non-zero
(159)
dF
were
Dn (d)
the rst part and Proposition 3.50 follows that there exist elementary matrices
T1 , . . . , Tr
such that
Tr T1 A = Dn (d).
Since all elementary matrices are invertible we get then
1 1 A = T1 Tr Dn (d)
and we see that (159) is satised with
1 1 S := T1 Tr SLn (F ).
Note that we can of course transform an invertible is, we get similar to the above result a decomposition
n n-matrix
using elemen-
tary column transformations of type I to a matrix of the special form (158). That
A = Dn (d )S
of
(160)
with
S SLn (F )
and a non-zero
d F.
The
The previous theorem shows that the matrices of the special linear group
SLn (F )
all
are by far not as special as one would imagine on the rst sight. upto a simple factor of the simple form
elements of
Dn (d)
already
Now a natural uniqueness question arises from the existence of the decomposition (159). And further, does in (159) and (160) hold every invertible matrix invertible
d = d.
over
a non-zero element
d = d(A) F
In other words: according to Theorem 3.55 we know that we can transform every
n n-matrix A
over F only by using elementary row transformations Dn (d) for some non-zero element d F . Ineviteable we
The mathematical precise result is that the general linear group is isomorphic (as groups) to
a semi-direct product of the special linear group and the multiplicative group in symbols
GLn (F ) SLn (F ) =
of the eld
F,
F .
85
have to ask ourself what kind of character this number has. Does it only depend on the original matrix we transform the matrix way.
A? Does always appear the same number d regardless how A with the help of elementary transformations of type I
into the form (158)? Let us formulate this question more precisely in the following
Problem 3.
matrix
n n-matrix
over a eld
an non-zero element
d = d(A)
of
a row transformation of type I and such that we assign this way to the (invertible)
Dn (d)
d?
The next theorem gives the following complete answer to this problem.
Theorem 3.56.
with
A GLn (F )
d F. d = d .7
But we are not yet able to proof this theorem because we will need more theoretical concepts. A direct proof of Theorem 3.56 would be desireable but the claim of this theorem is not evident. We have a similar situation as we had in the end of Chapter 1 with Problem 1. There we were not able to answer this problem before we introduced a new concept, namely the rank of a matrix in Chapter 2. Similar we will need to nd a new, suitable invariant for matrices to answer this problem. This will lead to the concept of determinants of a the solotion of Problem 3 will turn out to be very simple.
8
n n-matrix.
If we
have introduced this new concept, then the proof of Theorem 3.56 and with it
So far we have only shown how row and column transformations of type I are described using matrices. Now we shall complete the list of elementary transformations by showing how an elementary row or column transformation of type II and III are described with the help of matrix multiplication.
Proposition 3.57.
of the matrix
Let
0 1
1 (n) Di (a) :=
. . .
a
. . . . . . . . .
..
(161)
(the only non-zero entries are on the main diagonal and are all equal to entry at position
(i, i)
which is equal to
a F)
vi
of
by
S=S
See page 98 where we will nally carry out the proof of Theorem 3.56.
86
3. LINEAR MAPS
Similarly the same row transformation will be performed, if (m) the left with the matrix Di . ing the
is multiplied from
i-th
j -th
is
A
. . . . . .
:=
..
0
. . . . . . . . .
..
1
. . . . . . . . .
Rij
(n)
(162)
1
. . . . . .
0
. . . . . .
1
..
n n-identity
j -th
A
m n-matrix A
over
A .
can be formulated using suitable products of matrices of the form (149), (161) Recal for example Theorem 2.48 from the previous chapter. It states that we can transform any matrix
of rank
transformations to a matrix
of the form
A =
Ir 0
0 . 0
We can rewrite this theorem in our new language in the following way:
Theorem 3.58.
matrices
F.
P AQ =
holds. Here
Ir 0
0 0
matrix.
(163)
r = rank A
and
Ir
denotes the
r r-identity
Proof. From Theorem 2.48 we know that we can transform the matrix
using elementary row and column transformations into the desired form. there exists invertible matrices and (162) such that
Thus
P1 , . . . , P s
and
Q1 , . . . , Qt Ir 0 0 0
Ps P2 P1 AQ1 Q2 Qt =
(164)
Therefore (163) holds with P := Ps P2 P1 and Q := Q1 Q2 Qt . The claim that r = rank A follows from the observation that the rank of a matrix is invariant under elementary row and column transformations and that the matrix of the right hand side of (163) has apparently rank
r.
87
Note that the content of Theorem 3.58 is nothing else than what we have said before in Proposition 3.45 even though we gave there a complete dierent kind of proof. But now the theorem tells also how one can establish the equivalence
A
of matrices of
Ir 0
0 , 0
r = rank A
F m,n
nn-matrices. If one obtains after n n-identity matrix I then rank A = n and A is an invertible matrix. In this way we can not only verify whether a given nn-matrix A 1 is invertible or not, but we also get tool to compute the inverse matrix A for a given invertible matrix A. If (164) is satised with r = n, then
Now consider the special case of quadratic
Ps P2 P1 AQ1 Q2 Qt = I.
Therefore we get
9
1 1 1 A = P1 P2 Ps IQ1 Q1 Q1 t 2 1 1 1 1 = P1 P2 Ps Q1 Q1 Q1 t 2 1
= (Q1 Q2 Qt Ps P2 P1 )1 .
and thus we have for the inverse matrix
A1
of the matrix
the equality
A
Now in the case that
= (Q1 Q2 Qt Ps P2 P1 ) = Q1 Q2 Qt Ps P2 P1 .
1 1
(165)
row transformations are already enough to achive the transformation of That is we can in this case assume with out any loss of generality that
to
I.
Ps P2 P 1 A = I
where
(166)
P1 , . . . , Ps1
Ps = Dn (d
) is a diagonal matrix
10
A1 = Ps P2 P1 = Ps P2 P1 I.
of matrices we get the following result. (167) If one interpretes the equality (167) by means of elementary row transformations
F.
then
formations in the same order to the identity matrix 1 transforms to the inverse A of A.
I,
Example.
1 1 A := 1 1
9
1 1 0 0
1 0 1 0
1 0 0 1
g, h
of a
Using amongst othere the following easily to verify calculation rules for elements
group
G: (gh)1 = h1 g 1
and
(g 1 )1 = g .
10
Pi
in (165).
88
3. LINEAR MAPS
write blow this what we obtain by succesively applying the same elementary row transformations to both matrices:
1 1 1 1 1 0 0 0 1 0 0 0 1 0 0 0 1 0 0 0 1 0 0 0
1 1 0 0 1 0 1 1 1 1 1 0 1 1 0 0 1 1 0 0 1 1 0 0
1 0 1 0 1 1 0 1 1 0 1 1 1 0 1 1 1 0 1 0 1 0 1 0
. . .
1 0 0 1 1 1 1 0 1 1 0 1 1 1 1 1 1 1 1 2 1 1 1 1
1 0 0 0 1 1 1 1 1 1 1 1 1 1 0 1 1 1 0 1 1 1 0
1 2
0 1 0 0 0 1 0 0 0 0 0 1 0 0 0 1 0 0 0 1 0 0 0 1 2
. . .
0 0 1 0 0 0 1 0 0 1 0 0 0 1 1 0 0 1 1 1 0 1 1 1 2
1 2 1 2 1 2 1 2
0 0 0 1 0 0 0 1 0 0 1 0 0 0 1 0 0 0 1 1 0 0 1
1 2
1 0 0 0
0 1 0 0 A
0 0 1 0
0 0 0 1 1 2 1 2 1 2 1 2
1 2
1 2 1 2 1 2
1 2 1 2 1 2 1 2
1 2 1 2 1 2 1 2
A1
is:
A1 =
1 2 1 2 1 2 1 2
1 2 1 2 1 2 1 2
1 2 1 2 1 2 1 2
89
of
equations in the
unknown variables
x1 ,
...,
xn
over a eld
F.
Then to nd
all solution to this system of equations is equivalent to nd all solutions equation
to the (169)
Ax = b
where
bF
coecients
b1 , . . . , b m .
To nd all solution to the homogeneous part of (168) is equivalent to nd all solutions to
Ax = 0.
the kernel of the linear map have
(170)
Thus we see that the set of all solutions to the homogeneous part of (168) is precisely
A: F n F m ,
M0 = ker A, in particular since we know that ker A is a linear subspace n all solutions M0 is a subspace of F which is the content of
Chapter 1. Apparently the set
of
Fn
the set of
Proposition 1.3 in
M = x + M0
where
x F
Ax = b
b im A
and in this case the solution is unique if and only if the linear map phism which is equivalent with
(171)
A is a monomor(172)
ker A = 0.
linear maps: By this formula we have
dim(im A) dim F m .
Thus
dim(ker A) > 0
if
n>m
and in this case there must exists a non-trivial solution to (170) in this case. maps that the linear map
m = n we know from the dimension formula for linear A: F n F n is an epimorphism if and only if it is a b im A for every b F n if and only if ker A = 0, that is,
if and only if the homogeneous equation (170) has only the trivial solution. This proves Proposition 1.8 in Chapter 1.
CHAPTER 4
Determinants
the previous chapter the general linear group of a vector space we have encountered in a natural way the problem (see Problem 3 on page 85) whether there exists a function
d: GLn (F ) F
with the following two properties: (1) If
A, A GLn (F )
is obtained from
by an
d(A ) = d(A).
(2) For every matrix
Dn (a) GLn (F )
d(Dn (a)) = a.
One can show the proof is left as an exercise that the above two properties are equivalent with the following three properties: (1) If
A, A GLn (F )
is obtained from
by an
d(A ) = d(A).
(2) If
A, A GLn (F )
A is obtained a F , then
from
by multi-
d(A ) = ad(A).
(3) For the identity matrix
I GLn (F ) d(I) = 1.
holds
A map
d: F n,n F
is called a determinant function if it satises the following three properties: (1) If
A, A F n,n are two matrices and A is obtained from A by replacing the i-th column with the sum of the i-th and j -th column (1 i, j n, i = j ), then d(A ) = d(A). A, A F n,n
are two matrices and a column with a non-zero element
(2) If
d(A ) = ad(A).
91
92
4. DETERMINANTS
holds
d(I) = 1.
Note that if one identies a
n n-matrix A F,
over
with system
v1 , . . . , v n
of
its columns then a determinant function is nothing else then a map which maps
n-tuples
of vectors of
Fn
to the eld
in symbols
d: (F ) F,
and which satises with the following three properties: (1) For every system
n n
v1 , . . . , v n
of vectors of
Fn
and every
1 i, j n
with
i=j
holds
v1 , . . . , v n
of vectors of
F n,
every
a F
and every
1jn
holds
e1 , . . . , e n
of
Fn
holds
d(e1 , . . . , en ) = 1.
Depending on which point of view is more convenient we will consider in the following a determinant function either to be a map from all eld
n-tuples
of vectors of
n n-matrices F n.
to the
Proposition 4.2.
(1) If
Let
and
be
nn
matrices over
and let
d: F n,n F A
be
times the
j -th
column of
to the
(173)
(2) If
A A
d(A ) = d(A).
(3) If
is obtained from
a F,
then
(175)
d(A ) = ad(A).
Proof. We denote in this proof the colums of the matrix
A by v1 , . . . , vn . a = 0. Then
93
Denition 4.3.
if
Let A be a nn-matrix over the eld F . Then A is called singular A is not invertible. Likewise A is called non-singular (or regular ) if it is invertible, that is if A GLn (F ).
Proposition 4.4.
matrix. Then
Let
d: F n,n F
A F n,n
d(A) = 0. vi
Proof. :
We know that a matrix is singular if it doesn't have full rank, This is the case if and only if there exists one colum
that is if
rank A < n.
n1
v1 = a2 v2 + . . . + an vn
for some elements
a2 , . . . , an . = d(0, v2 , . . . , vn ) =0
d(v1 , . . . , vn ) = d(v1 a2 v2 . . . an vn , v2 , . . . , vn )
where the last equality follows from the second property of a determinant function. : can write We assume that
A = SDn (a)
for some
S SLn (F )
and a non-zero
aF
where
Dn (a)
matrix by multiplying the last column with transformations of type I. Thus Thus necessarily
must be
a and furthermore only elementary d(A) = a and since a = 0 it follows that d(A) = 0. singular if d(A) = 0.
Note that we still do not know whether determinant functions exist! In order to nd an answer to the question whether determinant functions exists in general we proceed in a way which is common to existence probelms in mathematics: often one studies the hypothetical properties of an object which is postulated to exists in detail until one is able to actually prove the existence of the object in question or until one gathers enough evidence which rules out the possibility that the object in question can exist. So let us continue with the studies.
d: F n,n F and d : F n,n F be two determinant functions. Then we n,n know already that d(A) = 0 = d (A) if A F is a singular matrix. On the other hand we have seen in the previous proof that if A = SDn (a), then d(A) = a and for the same reason d (A) = a, too. We can summarize this observation in the
Let following result.
d: F n,n F and d : F n,n F n,n tions, then d(A) = d (A) for every A F . In other n,n one determinant function on F .
Let A determinant function holds
d: F n,n F
1in
d(v1 , . . . , vi1 , avi + bwi , vi+1 , . . . , vn ) = = ad(v1 , . . . , vi1 , vi , vi+1 , . . . , vn ) + bd(v1 , . . . , vi1 , wi , vi+1 , . . . , vn )
for all
(176)
v1 , . . . , vi , wi , . . . , vn
in
and all
a, b F .
94
4. DETERMINANTS
Proof. Note rst that due to (175) it is enough to verify the claim for the
of generality that
a = b = 1. Due to (174) we can furthermore assume with out i = 1. n the vector space F has dimension n it follows that the system v1 , w1 , v2 , . . . , vn
any loss
of
n+1 vectors in F n
combination
cv1 + c w1 + c2 v2 + . . . + cn vn = 0
of the zero vector with Assume rst that the form
(177)
of
and thus
rank(v2 , . . . , vn ) < n 1.
and therefore the equality holds (compare with Proposition 4.4). Thus it remains to verify the case that one of the two possibilities, say also assume that in this case
c = 0 or c = 0. It is enough to verify c = 0. Without any loss of generality we may then c = 1. Thus from (177) follows that
v1 = c w1 + c2 v2 + . . . + cn vn .
Using this and repeatedly (173) we get
d(v1 + w1 , v2 , . . . , vn ) = d(c w1 + c2 v2 + . . . + cn vn + w1 , v2 , . . . , vn ) = d((c + 1)w1 + c2 v2 + . . . + cn vn , v2 , . . . , vn ) = d((c + 1)w1 + c2 v2 + . . . + cn1 vn1 , v2 , . . . , vn ) = ... = = d((c + 1)w1 + c2 v2 , v2 , . . . , vn ) = d((c + 1)w1 , v2 , . . . , vn ) = (c + 1)d(w1 , v2 , . . . , vn ) = c d(w1 , v2 , . . . , vn ) + d(w1 , v2 , . . . , vn )
and on the other hand
95
2. Proof of Existence and Expansion of a Determinant with Respect to a Row Theorem 4.7 (Existence of a Determinant). Let F be a eld. Then for every
natural number
n1
d: F n,n F .
We
det
or more precise
detn .
unique.
existence of a determinant function. We will carry out this proof by induction with respect to n
n.
If
= 1:
A = (a)
is a
1 1-matrix,
then
det1 (A) := a
denes apparently
the determinant function. n 1 n: We assume that n > 1 and that we have a determinant function detn1 : F n1,n1 F is given. We want to construct a map d: F n,n F using detn1 . For every n n-matrix A = (aij ) over F dene
d(A) :=
j=1
Here
(178)
Anj
denotes the
away the
n-th
row and
(n 1) (n 1)-matrix which is derived from A by leaving j -th column. We need to verify that the so dened map is
indeed a determinant function. Thus we need to verify all the three properties of a determinant function from Denition 4.1, one by one. (1) Assume that the matrix by replacing the column of
and
A = (ars ) is obtained from the matrix A = (ars ) i-th column vi of A by vi + vk where vk ist the k -th i k . Then
and
anj = anj
and if
for
j = i.
Ani = Ani
j = i, k
then
detn1 (Ank )
can be written
Ain
k -th
row past
|k i| 1
vk
1.
(1)|ki|1 = (1)ki1 )
and alltogether
d(A ) d(A) = (1)ni ani detn1 (Ani ) ani detn1 (Ani ) + (1)nk ank detn1 (Ank ) ank detn1 (Ank ) = (1)ni ank detn1 (Ani ) + (1)nk ank detn1 (Ani ) = 0
and therefore we have indeed
d(A ) = d(A).
96
4. DETERMINANTS
A = (ars )
and
is obtained from
A = (ars )
by multi-
i-th
a F.
Then
ani = aani
Furthermore we have
anj = anj
and if
for
j = i.
Ani = Ani
j=i
then
matrix
Anj
is obtained from
Anj
by multiplying
the element
a.
d(A ) = ad(A):
d(A ) =
=a
j=1
(3) If
A = (ars ) is the n n-identity matrix In over F , then an,1 = an,2 = . . . = an,n1 = 0 and ann = 1. Furthermore Ann = In1 . Therefore d(In ) = detn1 (In1 ) = 1 by (178). n,n Thus the map d: F F dened by (178) satises all properties of a determinant function. Therefore we dene detn := d. And this concludes the induction step n 1 n.
Now after this lengthish technical proof we know that for every eld every
and
n 1
F n,n F det
detn .
Denition 4.8.
If
Let
det : F A F
n,n
then the number determinant of
n,n
F.
the determinant of
is also denoted by
A.
The
a11
. . .
:= det(A).
an1
ann
But what kind of map is this? The answer will be given by the following
Proposition 4.9.
over
Let
be a eld and
n 1.
Let
F.
Then
detn (A)
is a polynomial of degree
in the
n
variable
= 1: Apparently det1 (A) = a11 is a polynomial of degree the single a11 . 2 n1 n: We assume that detn1 is a polynomial of degree n1 in (n1)
n
detn (A) =
j=1
follows that
detn1
in the
n2
variables
a11 , . . . , ann .
97
The rst three determinant functions are then explicitly the following:
det1 (A) = a11 det2 (A) = a11 a22 a12 a21 det3 (A) = a11 a22 a33 + a12 a23 a31 + a13 a21 a32 a11 a23 a32 a12 a21 a33 a13 a22 a31
n. n = 1 we have one summand, for n = 2 we have 2 = 1 2 summands, for n = 3 we have 6 = 1 2 3 summands. The general rule is that detn (A) is a polynomial with n! = 1 2 3 n, that is n-factorial summands. For example det10 (A) is a polynomial with 3 628 800 summands in 100 variables! Thus it becomes apparent
Note that the number of terms in those polynomials is increasing drastically with For that it is dicult to memorize those polynomials and that we need to develope calculation methodes to simplify the calculation of a determinant in the concrete case. One way to simplify the calculation can be obtained from the equation (178) which can also be written as
det(A) =
j=1
with the help of the equality
(1)nj = (1)n+j . If the last row of the matrix which are equal to 0, then the right hand side of the
In the extreme case that only one
above equation contains only few summands. coecient of the last row of
j -th
coeent
anj = 0,
Anj
is only a
(n 1) (n 1)-matrix
of this equation is easier to calculate than the left hand side. But can we gain a similar advantage if another row of the matrix
say the
i-th
from
is the
by shifting the
i-th
=
j=1
anj = aij
and
Anj = Aij
we get
=
j=1
Thus we have obtained the following general formula to expand the determinant
of
98
4. DETERMINANTS
Proposition 4.10.
have for every
Assume that
A
n
is an
n n-matrix
F.
Then we
1in
the equality
det A =
j=1
(179)
by leaving
Example.
0 2 0 4
10 4 0 0 3 1 5 =3 2 0 3 0 4 6 2 3
10 3 6
0 2 5 = 30 4 3
Here we have expanded the determinant in the rst step along the then in the next step we have expanded it along the the determinant shrinks in size. simplicity by applying previous proposition.
1-st
row.
Before we begin to study the determinant function more closely we shall give now the
Answer to Problem 3
A GLn (F ).
exists a decomposition of
of the form
A = Dn (a)S
with
(180)
S SLn (F ) and a F .
A = Dn (a )S
with
S SLn (F )
and
a F .
Then
Dn (a)S = Dn
and it follows
Dn (a) = Dn (a )S S 1
Since
S S 1 SLn (F ) and thus the above equation means that Dn (a) is obtained from Dn (a ) by elementary column transformations of type I. Therefore necessarily det Dn (a) = det Dn (a ). Since Dn (a) is obtained from the identity matrix I by multiplying the last column by a it follows that det Dn (a) = a det I = a. Likewise det Dn (a ) = a and thus we must have that a = a and therefore Dn (a) = Dn (a ). Multiplying the previous equation from the 1 left with Dn (a) and from the right with S yields then the equality SLn (F )
is a group it we have that
S=S.
Therefore the decomposition (180) is uniue. We still have to prove the second claim of Theorem 3.56, namely if have a decomposition of
of the form
A = S Dn (a )
then necessarily
(181)
a = a
where
is the element of
is obtained
99
a.
Therefore
det A = a det I = a .
Similarly, if we interprete (181) in terms of column transformations it follows that
det A = a
column by
since
a=a.
3. Elementary Properties of a Determinant Proposition 4.11. A n n-matrix A over the vield F is invertible if and only if
det A = 0.
In other words:
is singular
det A = 0.
Proposition 4.12.
A, B Mn (F )
det B = 0.
of (182) is equal to to
0.
Since
0. det B = 0
We have to show that in this case also the left side is equal it follows from the previous proposition that
rank B < n.
implies that the left hand side of (182) is equal to Thus we can consider the remanining case that there exists a the equality
S SLn (F )
and a non-zero
bF
AB
is
transformations of type I and nally by multiplying the last column by the ele-
b.
Thus
det AB = b det A.
Since apparently
det B = b
Corollary.
If moreover
If
A, B Mn (F )
det(A1 ) =
Proof. The rst claim is evident. If
1 . det A
A is invertible then it follows from Propodet A det A1 = det AA1 = det I = 1 and thus the claim is evident.
Note that the content of Proposition 4.12 can be interpreted in the language of Algebra as follows: the determinant function is a group homomorphism
1
det : GLn (F ) F .
1
A map
f: G G
for every
g, h G.
100
4. DETERMINANTS
where
of the eld
F.
(Characterisation of the Special Linear Group)
Proposition 4.13
linear group That is
The special
with
det A = 1.
Proof. :
Let
S SLn (F ).
Then
matrix
det S = det I = 1.
: that and
Assume that A Mn (F ) with det A = 1. By Proposion 4.11 we know A GLn (F ). Thus it follows by Theorem 3.55 that there exists a S SLn (F ) a non-zero c F such that
A = SDn (c) det A = det S det Dn (c). Since S SLn (F ) it det S = 1. Thus also det Dn (c) = 1. On the other hand we have apparently det Dn (c) = c det I = c and thus necessarily c = 1. Therefore Dn (c) = I and it follows that A = S is an element of SLn (F ).
and Proposition 4.12 follows that follows that Note that sofar we have introduced the concept of a determinant only by looking at the columns of a matrix and only by considering elementary column transformations. Denition 4.1 is in no way symmetric with respect to columns and rows. Assume that
d: Mn (F ) F
is a function which satises the analogous conditions We shall call such a function
for the moment a row determinant function. Likewise we shall mean for the moment by a column determinant function a function which satises the conditions of Denition 4.1. Furthermore, if by
d: Mn (F ) F
t
the function
d: Mn (F ) F, A t d(A) := d(tA) A
(see page 45). It is apparent that
where
terminant function.
determinant function remains valid for row determinant functions. Furthermore it is a consequence of Theorem 4.7 that there exists precisely one row determinant function
d: Mn (F ) F ,
namely
det.
holds the equality
Theorem 4.14.
For every
A Mn (F )
(1) Let
A, A Mn (F )
Then
is obtained from
A S SLn (F )
A = SA.
S SLn (F )
we
det S = 1
and thus
det A = det A.
101
A, A Mn (F ) and assume that A is obtained from A by multiplying (n) (n) i-th row by a F . That is A = Di (a)A where Di (a) is the det A = det Di (a)A = det Di (a) det A
(n) (n)
where the last equality is due Proposition 4.12. Since have then (3) Clearly
det Di (a) = a
(n)
we
Thus we have seen that we actually need not to distinguish between column and row determinant functions because they are actually the very same functions. Thus we shall abolish this notation again and we will from now on speak the determinant
det : Mn (F ) F.
As an consequence of the previous result we get a variation of the Proposition 4.10, namely how to expand a determinant with respect to an arbirary column. Precisely this is the following
Proposition 4.15.
have for every
Assume that
A
n
is an
n n-matrix
F.
Then we
1in
the equality
det A =
j=1
(183)
by leaving
Proof. Denote by
A.
Then
aij = aji
and
Aij = Aji .
det A = det tA
n
=
j=1 n
We can enrich the collection of identities given in Proposition 4.10 and 4.15 by two more identities which are less important but which then combined give a compact formula for matrix inversion.
Lemma 4.16.
Let
A Mn (F ).
n
1 i, k n
with
i=k
the
(184)
(185)
by replacing
the
k -th
row of
by the i-row of
A.
rows of A is linear dependent and thus det A = 0. If one expands A along the k -th row one obtains precisely (184). The equality (185) is veried in a similar way.
102
4. DETERMINANTS
to
is dened to be the
Let A Mn (F ). Then the complimenn n-matrix which has the coecients (186)
and
Theorem 4.18
Mn (F )
For any
AA = det(A) I
and
AA = det(A) I
Proof. Using the formula for the matrix product (see Theorem 3.31) we get
cik =
j=1 n
=
j=1
det A = 0
Lemma 4.16. But this means that that
if if
i = k, i = k,
where the rst case is due to Proposition 4.10 and the remaining case due to
C = det(A) I .
The second equality is shown in a similar way or follows straight from the fact
Mn (F ) A
Theorem 4.19
sume that by the formula
Let
A.
aij =
Proof. If
is invertible then
from
det(A)I = AA
the equation
A1 =
For
1 A. det A
A1 = (aij )
Note that the importance of the Cramer's rule for matrix inversion does lie so much in the ability to actually calculate the inverse of a matrix but rather in the theoretical content of it. Consider the case that
F =R
or
F = C.
Since we know
that the determinant is a polynomial in the coecients of the matrix we know that it is a continuous function. Thus (187) states that the coecients of the inverse matrix
A1
A.
103
Example.
A :=
with coecients in the eld
a c
b d det A = ac bd = 0
and we have
is invertible. Then
A1 =
In praticular if
1 det A
d b . c a
A SLn (F ) then det A = 1 and we get the even more simple formula A1 = d b . c a A B 0
is a square matrix over
of the form
A :=
where
C B
an arbitrary
B Mm (F ), B Mn (F )
and
C is B, B
m n-matrix
over
F.
If
0.
is regular .
and
are regular,
too (why?). Thus we can transform the matrix transformations of type I to the rst
A :=
with
Dm (b) C 0 B
b = det B
where
is a
m n-matrix over F . Applying elementary row n rows we can transform the matrix A into a Dm (b) C 0 Dn (b ) n rows to the rst
A :=
with
b = det B
A A
:=
Dm (b) 0 . 0 Dn (b )
I by n-th and the last column by the elements b and b . Therefore det A = bb det I = det B det B . Thus we have seen that we can transform the matrix A to a matrix A by using elementary row transformations of type I only. Thus det A = det A and it follows that (188) is also satised in the case that A is
Now is a diagonal matrix which is obtained from the identity matrix multiplying the a regular matrix.
Recall that a square matrix is called regular if it is not singular which is by denition
104
4. DETERMINANTS
Theorem 4.20.
Let B1 , . . . , Br be arbitrary square matrices over the same eld Then we have the equality
F.
B1 B2
.. .
=
i=1
|Bi |
(189)
Br
det B1 det Br . Bi A
Example.
are all
1 1-matrixes.
a11 A= 0
Then
a12 a22
... ...
.. .
Let us conclude this section with a note about one possible use of the determinant function. Recall that in Section 10 of the previous chapter we introduced the concept of similarity of square matrices: two matrices similar if there exists a regular
A, A Mn (F )
are called
n n-matrix S
over
such that
A = S 1 AS.
Thus in this case we have
S 1 S = I
always the same determinant. Thus we have a convenient way to verify whether two matrices are not (!) similar to each other. For example the real matrices
1 0
has determinant endomorphism
0 1
and
0 1
1 0
+1 and the latter matrix 1. So these two matrices cannot represent one and the same f : R2 R2 . It would be much more dicult to show this without n n-matrices
over the
the concept of a determinant. Note that the converse is not true in general: if two same eld have the same determinant they might still be not similar to each other!
105
det1 (A) = a11 det2 (A) = a11 a22 a12 a21 det3 (A) = a11 a22 a33 + a12 a23 a31 + a13 a21 a32 a11 a23 a32 a12 a21 a33 a13 a22 a31
(see page 97). These polynomials share a great amount of symmetries and the recursive formuly (178) hints that even the following determinant functions share similar symmetries. The natural question arises whether there is a compact formula for the determinant for an arbitrary
n n-matrix
And the answer is: yes! In this section which will be the last one about determinants in general we will derive this formula which is known as the Leibniz formula
for determinants.
behaviour of
But before we can state this formula we need to take again an The aim of this excursion is to study the under arbitrary permutations of the columns of
det A
A.
Assume that
SM
the set of all bijective maps
: M M .
permutation of
M.
, S ,
: M M, x ( (x)) SM . Therefore we obtain a multiliSM . The product of two elements of , SM is dened to be 4 the composite map . Forming the composite map is an associative operation. Apparently the identity map on M is contained in SM and id = and id = for every SM . Thus id SM is the identity element for this product and is 1 also called the trivial permutation. Furthermore the inverse map is dened 1 for every SM and is itself a bijective map : M M and thus an element 1 of SM . Clearly is the inverse element of with respect to the above dened 1 product on SM , that is = id and 1 = id. Therefore SM satises
is again a bijective map and thus an element of cation on the set with this product all the properties required by a group (see Denition 3.46). This group is called the group of permutations of
M. SM
is not abelian!
5
Therefore we
have to be carefull when switching the order of elements in a product of permutations as does mostly leave the result not unchanged. We are interested in the group of permutations in the following special case:
is called
Let n 1 be a natural number 1 i n, that is N := {1, . . . , n}. the symmetric group of n elements
Sn := SN .
3 4
Named after the German polymath Gottfried Wilhelm von Leibniz, 16461761. Compare this with the deniton of the multiplication of two endomorphisms of a vector
EndF (V )
But then again the restriction of this multiplication to the set of all
automorphisms yields a group, namely the One can show that the group
GLF (V ). M
has not more
SM
3 elements.
106
4. DETERMINANTS
We have to make some basic observations about the group the permutations of the set
Sn .
Amongst all
N = {1, . . . , n}
Denition 4.22.
such that and such that
Let
Sn .
1 i, j n, i = j
(i) = j (k) = k
for any other
(j) = i integer 1 k n
and
then
is called a transpo-
sition.
That is, a transposition of the set the other elements of in
i,j
the
ij
and
j i.
S1
N. 2 = id.
Lemma 4.23.
Sn
holds
We will use this result without further reference. Another very intuitive result is the following. It basically states that any nite collection objects can be ordered by exchanging switching nitely many times suitable objects (you might want to try this with a deck of cards). Precisely formulatet this is the following
Lemma 4.24.
Let Sn be a non-trivial permutation. Then can be written as a nite product of transpositions, that is there exists a decomposition
= 1 k
where the
i (1 i k )
N. n 2. Sn and
n
= 2:
this per-
mutation is equal to
1,2 .
all
we assume that by
the set
n > 2 and that the lemma is prooven for {1, . . . , n 1}. Let Sn . Then either
Then the restriction of
(n) = n.
to the set
Sn1
= 1 k
is a product of transpositions of product of transpositions of that
N.
(n) = n. (n) = n. Set i := (n) and N such that (n) = is the trivial permutation then = i,n and the claim Thus we may assume that is not the trivial permutation
Then
is a permutation of
We get a decomposition
= 1 k of into transpositions. But then due to i,n i,n = id we get that = i,n i,n = i,n = i,n 1 k is a decomposition of into transpositions of N . And this completes the remaining case of the induction step.
In the words of an Algebraist the above lemma states that the group is generated by the transpositions of the set
Sn (n 2)
N.
Note that the above lemma does not state that the decomposition of a permutation into transpositions is unique. The lemma just states that always such a
107
decomposition exists.
is
not unique there is a property which is uniquely dened: it will turn out and this is not a trivial result is that for a given permutation
Sn
the number
of transpositions needed for its decomposition into transpositions is either always even or always odd. We will see this soon. But before we are heading towards this result we shall still state the following not to dicult observation.
Lemma 4.25.
elements.
elements
Sn
has
n! = 1 2 (n 1) n
We begin to return from our excursion to the real of Algebra back to Linear Algebra . Let
be a eld and
n1
a xed integer
P ej = e(j) ,
where
(e1 , . . . , en )
P = (e(1) , . . . , e(n) ),
standard basis (Proposition 3.28). The matrix identity matrix by permutation of the columns.
(191)
since the columns of a matrix are precisely the images of the vectors of the canonical
Denition 4.26.
Let Sn . Then the matrix P as dened above is called the permutation matrix belonging to the permutation . In particular the special matrix position
Rij
chapter is then nothing else then the permutation matrix belonging to the trans-
ij . P P = P
(192)
Straight from the denition (190) follows that we have the equality
for every
, Sn .
a group which we shall for the moment denote by genral linear group
Pn (F ).
It is a subgroup of the
GLn (F ).
P : Sn Pn (F ), P
is a homomorphism of groups and it is apparent that this homomorphism is actually an isomorphism of groups. Thus the symmetric group as groups. Now let
Sn
is isomorphic to
Pn (F )
A Mn (F )
be an arbitrary
n n-matrix.
Denote by
A.
A := (v(1) , . . . , v(n) ).
Note that in particular
I = P
A = AP .
Thus we can interprete the permutation of the columns of the matrix multiplication of then
(193)
by a
(194)
108
4. DETERMINANTS
det A
we must calculate
det P .
det P = 1.
For an arbitrary non-trivial permutation there exists a decomposition of we get
Sn
(and therefore
)
is
Denition 4.27.
Let
be a eld with
char F = 2.
1 or the that is an
number
Sn
F.
Moreover
nothing hinders us to use the denition of a sign function also in the case that
char F = 2. But since the symbols +1 and 1 denote in this case the same element F we have that this function does not carry much (actually any) information when char F = 2. It is then just the constant 1 function.
of Note that with this notation we can now write the equation (194) as
n n-matrix A = (aij ) A.
F.
Let us denote by
v1 , . . . , v n
Then we
det A = det(v1 , . . . , vn )
n n
= det
i=1
ai1 ei , . . . ,
i=1
ain ei
=
(i1 ,...,in )N n
Here the last sum is taken over all an
(196)
n-tuples (i1 , . . . , in ) of elements in N . n-tuple two numbers coincide then the rank of the system ei1 , . . . , een less than n and therefore det(ei1 , . . . , een ) = 0. Thus we actually only sum in (196) only over all n-tuples (i1 , . . . , in ) for which hold {i1 , . . . , in } = {1, . . . , n}.7
6
(197)
But this does not inuence the
Note that we silently ignored the case that Note that this is an equality of
char F = 2.
following discussion.
7
sets !
109
(i1 , . . . , in ) is such an n-tuple, then (k) := ik denes an bijective : N N and is therefore an element of Sn . Vice versa, if Sn , ((1), . . . , (n)) is apparently an n-tuple which satises (197). Thus we get
Now if the equality (196) the formula
det A =
Sn
sgn()a(1),1 a(n),n .
(198)
Theorem 4.28
arbitrary
Let
A = (aij )
be an
n n-matrix
F.
Then
(199)
det A =
Sn
Notice the impressive beauty and simplicity of the Leibniz formula (199). Yet it is not important for the explicite computation of a determinant. chapter are the useful ones. The importance of the Leibniz formula lies in the theoretical insight it provides. It gives an explicite relationship between the coecients of the matrix value of its determinant For example, if To compute a determinant it the properties and results from Section 1 and Section 3 of this
and the
det A: F = R or F = C,
continuous function in the coecients of the matrix polynomial and polynomials are continuous.
A,
Or in case that
is a matrix with
Appendix
APPENDIX A
1. Sets
We shall collect in this section a few mathematical denitions about sets. We will restrict ourself to what is know as naive set theory (and will not touch the more complicated issues about axiomatic set theory).
Denition A.1.
If If
A set
A, then we denote this fact in symbols by x A. A, then we denote this fact by x A. If A and B /
are sets, then the sets are equal if they contain precisely the same elements. The set which does not contain any elements is called the empty set and denoted in symbols by
. A
is charactericed by its ele-
Note that the above denition means that a set ments. In order to show that two sets two seperate things: (1) For every (2) For every
A and B
xA xB
holds holds
x B. x A.
if the second statement
Note that the above statements are really two dierent statements. If the rst is true, then we say that is true, then we say that
Denition A.2.
xA
holds Thus
Let
is a subset of
if for every
x B.
A B.
Note that the empty set is
A = B
if and onyl if
A B
and
B A.
subset of every set. We may describe sets by listing there elements in curly bracets. That is, the empty set is given by
{}
and the set containing all integers from
to
is given by
{1, 2, 3, 4, 5}.
We might also list the elements of a set by its property. For example we may write the set
N := {i Z : i 0},
which reads
i 0.
This is
114
We can construct new sets out of given sets. The most common constructs are the following:
Denition A.3.
Let
and
AB
is the set
A B := {x : x A
and their intersection
x B}
AB
A B := {x : x A
The (set theoretic) dierence
x B}.
A\B
of
and
is the set
A \ B := {x : x A
Note that always true that
and
x B} /
but in general it is not (!)
AB = B A A \ B = B \ A.
and
AB = B A
Another common way to construct new sets out of given ones is to form the cartesian product.
Denition A.4.
of all pairs
Let with
(a, b)
cartesian product
AB
is the set
and
and that
n1
n-fold
usual denoted by
An ,
that is
An := A . . . A .
n-times
Elements in
are called
n-tuples.
We set
A = A A.
A0 := . n notation F in
Apparently
A1 = A
and
2. Maps
We shall collect a few basic mathematical denitions about maps in general, not specic to Linear Algebra.
Denition A.5.
the set
Let
and
from
to
B, f : A B, x f (x), x A precisely one element f (x) B . The set B is called the range (or co-domain ) of f . g: C D are equal if A = C , B = D and for every
xA
If
xA
the pre-image of
under
is the set
f (x) B is the image of x under f . If y B then f 1 (y) := {x A : f (x) = y} which is a of A. Similarly if A A, then the image of A under f (A) := {f (x) : x A}
is the set
One might relax the denition of equal maps by leaving away the requirement
B = D,
2. MAPS
115
B.
B B,
then by the
pre-image of
under
f 1 (B ) := {x A : f (x) B }
and this is a subset of the set in the range:
Denition A.6.
Let
f: A B
is an injective map
f 1 (y) f
1
contains at most
3
y B. y B.
(200)
is a surjective map
(y)
contains at least
(201)
A bijective map is a map which is both injective and surjective. Note that the statement (200) is equivalent with For every
x, y A
follows from
f (x) = f (y)
that
x = y.
f (A) = B.
Note further that every injective map
f: A B
A f (A).
We can construct new maps from given ones:
Denition A.7.
of
Let
f: A B
be a map and
A A.
f |A
to the set
is the map
f |A : A B, x f (x).
Furthermore, if
C
of
g: B C
composite map
gf
and
g f : A C, x g(f (x)).
Note that the operation of forming composite maps is associative. That is, if
h: C D
h (g f ) = (h g) f.
For every set
idA : A A, x x
which is called the identity map of the set then the identity map of the set
A.
is denoted just by
id.
If
A A,
i: A A, x x
is called the inclusion of of
into
A.
to the subset
If
f: A B
y B
the pre-image
f 1 (y)
f 1 : B A
2
116
y B
x A
f (x) = y .
f.
Note that
the inverse map is only dened for bijective maps! Apparently if map.
f 1
is again a bijective
f: A B
1 1
is again equal to
f.
If
(g f )1 = f 1 g 1 . It is useful to know that a map f : A B is a bijectvie map if and only if there 1 exist a map g: B A such that g f = idA and f g = idB . In this case g = f .
APPENDIX B
In Chapter 2, when we dened the algebraic structure of a eld, we have only given exapmles of elds with characteristic with only nite many elements. Dene for a given number
0.
m>0
the set
Fm
to be
Fm := {0, 1, 2, . . . , m 1}.
That is, the set
Fm
consists of exactly
m elements.
a multiplication on
Fm
which makes the this set into a ring and we will see under
which conditions this ring is also a eld. We need a result basic result from number theory: the division algorithm for
integers.
m > 0
and
an
and
x = qm + r
We say that
0 r < m.
r is the reminder
of the division of
in the following by
rm (x).
Now we dene the addition and the multiplication in
Fm
as follows: (203)
x + y := rm (x + y)
for every
and
x y := rm (xy)
x, y Fm .
m.
Proposition B.1.
(1) The set (2) The ring
Let
m>0
Fm together with the addition and multiplication modulo dened in (203) is a commutative ring with unit. Fm
is a eld if and only if
as
is a prime number.
the equality
char(Fp ) = p.
On
the other hand one has the following result which gives a strong constraint on the
Proposition B.2.
number.
Let
be a eld with
char(F ) > 0.
Then
char(F ) = p is a prime
x := me = 0
and
y := ne = 0.
xy = (me)(ne) = (mn)e = ke = 0
but this contradicts to (39). Thus a eld with
char(F ) > 0
then
Thus if
is
117
APPENDIX C
Recall that we have proven the existence of a basis for a vector space in the case that
only
proof was even relatively easy. If one wants to prove the existence of a basis in the general case where
machinery to conquer the problem. The key to prove the existence of a basis for vector spaces which are not nitely generated is Zorn's Lemma , which is a result of set theory and equivalent to the Axiom of Choice. Before we can state Zorn's Lemma we need some denitions.
1
Denition C.1.
Let
if the following three conditions hold: (1) The relation is reexive, that is from
xx
for every
x X. x, y X
it follows
xy
and
and
yx
that
that
x = y. x, y, z X
it follows from
xy
yz X
x z. xy
or
y x for every x, y X ,
bound of element
A X be a subset of a partialy ordered set. Then m X is an upper A if x m for every x A. An element m X is called a maximal if m x implies m = x for every x X .
Let
Example.
X.
Zorn's Lemma.
is totally ordered subset) has an upper bound contains at least one maximal element.
Let
F.
Denote by
V.
Then
the set of
C L
C :=
M C
Named after the American algebraist, group theorist and numelrical analyst Max August
Zorn, 19061993.
119
120
v1 , . . . , v n
()
a1 , . . . , an F
a1 v1 + . . . + an vn = 0
is a linear combination of the zero vector. Now there exists that
M1 , . . . , Mn C such i = 1, . . . , n. Since C is by assumption totaly ordered by incluseion there exists a 1 i0 n such that Mi Mi0 for every i = 1, . . . , n. In particular this means that () is a linear combination of the zero vector by vectors of the is linearly independent set Mi0 . Thus a1 = . . . = an = 0. Now since v1 , . . . , vn had been arbitrary, but pairwise distinct vectors in C this shows that C is a linearly independent subset of V and therefore an element of L and by construction for every M C holds M C . Thus C has an upper bound. Since C was an arbitrary chain of L this shows that every chain in L has an vi M i
for upper bound. Thus the can apply Zorn's Lemma which states that there exists a maximal element Then
B L. V
and thus by Proposi-
is a basis of
V.
The interesting issue about this theorem is that it is so strong that it can be proven to be equivalent with Zorn's Lemma (see [
Bla84]).
vector space has a basis, then it follows that the Axiom of Choice must hold and thus Zorn's Lemma holds, too. In this sense the existence of a basis is a very deep and strong result of Linear Algebra. Note the beauty of the above theorem! It takes only as little as six words we can state a theorem which relates to the very foundation of mathematics. Observe that with a slightly addapted proof we can verify that the basis extension theorem that is Theorem 2.36 holds true in the innite dimensional case, too. Precisely we have the following result.
Theorem
M
of
Let
vector space
V.
Then
such that
N can N M.
be extendet to a basis of
V,
APPENDIX D
Semigroup:
A semigroup
H = (H, )
is a non-empty set
together with
a binary operation
: H H H, (x, y) x y
which satises the associativity law. Note that a semigroup is not required to hava an idenitity element. A semigroup
is said to be commutative if
xy =yx
for every
x, y H .
Monoid:
A monoid
H = (H, ) is a semigroup which has a identity element. e H is said to be an identity element if xe = x and ex = x x H . The identity element of a monoid is always unique. A
Group:
and
inverse element. If
G = (G, ) is a monoid where every element of G has an x G, then y G is an inverse element of x if xy = e y x = e where e is the identity element of G. An inverse element
A group
is always unique, therefore one can speak of the inverse elment of an element
x G.
Ring:
R = (R, +, )
is a set
operations
+: R R R, (x, y) x + y, : R R R, (x, y) xy
satisfying the following three conditions: (1) (2) (3)
(addition) (multiplication)
(distributive law).
The identity of the addition is usually denoted by 0 and the identity of the multiplication is usually denoted by 1. In order to avoid triviality one usually requires that
and the identity element of the multiplication are dierent elements. A ring where the multiplication is commutative is called a commutative ring. A zero divisor of
is an element
0=xR
such that
xy = 0
for some
0 = y R.
Field:
A eld
a group
(F , )
is
is commutative and every non-zero element has a multiplicative inverse. (See page 19.)
121
122
Vector Space:
two maps
A vector space
V = (V, +, )
over a eld
is a set
with
+: V V V, (x, y) x + y, : F V V, (a, x) ax
satisfying the following conditions: (1) (2)
(F, +) is an abelian group. (ab)x = a(bx) for every a, b F and x V . (3) 1x = x for every x V . (4) a(x + y) = ax + ay for every a F and x, y V . (5) (a + b)x = ax + bx for every a, b F and x V . Elements of V are called vectors and elements of F are called scalars.
page 21.)
(See
Algebra:
over a eld
(or
F -algebra)
is a vector
with a multiplication
: V V V, (x, y) xy
satisfying the following conditions: (1)
together with the addition of vectors and the above dened mul-
V is x, y V
a ring. and
aF
holds
V!
Examples.
N := {0, 1, 2, 3, . . .}
is a commutative monoid under addition and under multiplication. (2) The set of strictly positive natural numbers
N+ := {1, 2, 3, . . .}
is a commutative semigroup under addition and a commutative monoid under multiplication. (3) The set of integers
Z := {0, 1, 2, 3, . . .}
is a abelian group under addition and a commutative monoid under multiplication. In particular
(Z, +, )
Q := { r : r Z s
(5) The set of all real numbers tion. (6) The set of all complex numbers
and
s N+ }
C := {a + ib : a, b R}
is a eld under the multiplication of complex numbers (note that for the
imaginary unit
(7) If
i2 = 1).
is a linear map}
is a
F -vector
EndF (V ) := {f : f : V V
is a
F -algebra
123
n n-matrices
n,n
over a eld
Mn (F ) := F
is a
F -algebra
GL(V ) := {f EndF (V ) : f
of the
is an isomorphism}
vector space
n n-matrices
n
over
GLn (F ) := GLF (F )
is a group, called the general linear group of degree
n.
APPENDIX E
In Section 5 we dened the important concept of a rank. The rank of a system of vectors is a typical example of how a simple but well crafted denition can result into a very fruitful concept in mathematics. This appendix shall recollect the dierent variants of the rank and point out some aspects of the concept. Recall Denition 2.25 were we dened what we mean by the rank of the nite system of vectors
u1 , . . . , um
is
r,
namely:
(1) There exist a linear independent subset of the set consists of exactly (2)
{u1 , . . . , um } which r vectors. Any subset of {u1 , . . . , um } which consists of r + 1 vectors is linear dependent.
We obtained straight away some simple results about the rank: the rank of a system of
m,
that is
rank(u1 , . . . , um ) m,
and equality holds if and only if
u1 , . . . , u m
are
form a linar independent set. Another simple result has been that the rank of a system of vectors is not decreasing if new vectors are added, that is
um+1
u1 , . . . , um
(Proposition 2.27).
The rank of a system of vectors is invariant under elementary transformations (Denition 2.29 and Proposition 2.30). And Theorem 2.31 described an algorithm how to compute the rank of a nite system of vectors using the Gauss Algorithm. The invariance of the dimension (Theorem 2.34) is the rst non-trivial result which we was obtained using the concept of a rank. The key to the proof of this result is Proposition 2.38 which states that the rank of a nite system of vectors is bounded by the dimension of the vector space, that is
rank(u1 , . . . , um ) dim V.
The so obtained theorem about the invariance of the dimension enabled us to dene the dimension of a vector space in a proper way. Proposition 2.38 gives intuitive interpretation of the rank of a system of vectors, namely
rank(u1 , . . . , um ) = dim span(u1 , . . . , um ). m n-matrix over a eld F can be seen as F n the concept of a rank extends in a natural way to matrices. We dened see Denition 2.44 the column rank of a m nmatrix A to be the rank of the n vectors u1 , . . . , un obtained from the colums of the matrix A, that is
Since the columns and rows of an elements of the vector space
Fm
and
126
A to be the A, that is
rang of the
vectors
v1 , . . . , v m
A,
that is
which is done in Denition 2.50. Using the interpretation given in Proposition 2.38 this means that the rank of a matrix spanned by the vectors obtained from the columns of obtained from the rows of
A is equal to the dimension of the vector space A and that this number is
equal to the dimension to the dimension of the vector space spanned by the vectors
A. f: V W
is introduced in Denition 3.14
f,
that is
rank f := dim(im f ).
Note that this allows the possibility of
rank f = .
an
are nite dimensional vector spaces Proposition 3.25 states the following relation between the rank of matrices and the rank of a linar map: for any coordinate matrix
rank f = rank A.
Thus the rank of a linear map turns out to be a very natural denition. From the rank of a linear map For example if
f: V W
rank f = dim V . Likewise, if W is nite and only if rank f = dim W . And from
nite dimensional vector spaces of same dimension in the case that a linear map:
if
From the dimension formula for linear maps that is Theorem 3.15 we get
0.
Index
algebra, 61, 122 homomorphism, 72 isomorphism, 72 algorithm for calculating the inverse matrix, 87 for solving homogeneuos systems of linear equations, 13 for solving nonhomogeneuos systems of linear equations, 15 Gaussian elimination algorithm, 7 to compute the basis of a subspace, 47 to compute the rank of a system of vectors, 36 automorphism, group basis, 27 cannonical basis, existence, 33 nite, 27 ordered, 35 standard basis of standard basis of
F -algebra, F I , 23 F (I) , 24
see
algebra
Fm , 117 F m,n , 64 F n , 22
function,
see
see
map
linear map
automorphism group,
see
general linear
see
see
standard basis
GLF (V ), see general linear group GLn (F ), see general linear group
group, 80, 121 abelian, 80 homomorphism, 99 isomorphism, 81
characterisation, 31
F (I) , 28 F n , 28
basis isomorphism, 56
C,
see
complex numbers
see
vector
HomF (V, W ),
integers, 122 isomorphism
60
commutative diagram, 65 complex numbers, 20, 122 coordinate isomorphism, 56 coordinate vector, Cramer's rule, 102 determinant, 96 Leibniz formula, 109 determinant function, 91 dimension,
of algebras, 72 of groups, 81 of vector spaces, 54, 72 Leibniz formula, seedeterminant109 linear combination, 24 non-trivial, 27 linear dependence, 29 linear hull,
see
vector
see
vector space
direct sum, 41 elementary matrix, of a matrix, 8 of a system of vectors, 36 of system of linear equations, 7
see
see
span
matrix
linear map, 51 automorphism, 80 coordinate matrix, 62 defect, 58, 126 dimension formula, 58 endomorphism, 61 epimorphism, 52 image, 56
127
elementary transformation
EndF (V ),
see
endomorphism ring
endomorphism,
see
linear map
128
INDEX
Q, R,
see see
see
rank
rank, 34, 125 linear map, 57, 63, 126 matrix, 47, 126 of a system of vectors, 34 row and collumn rank of a matrix, 125 row and column rank of a matrix, 44
see
vector space
see
domain
rational numbers, 20, 122 real numbers, 122 relation antisymmetric, 119 partial order, 119 reexive, 54, 119 symmetric, 54 total order, 119 transitive, 54, 119 ring, 21, 121 commutative, 21 invertible element, 79 ontegers unit, 79
composite map, 115 domain, 114 identity, 115 image, 114 inclusion, 115 injective, 115 inverse, 116 linear map, 51 one-to-one, onto,
see
see
injective
surjective
pre-image, 114 range, 114 restriction, 115 surjective, 115 matrix, 5, 63 calculating the inverse matrix, algorithm complimentary, 102 coordinate matrix of a linear map, 62 determinant, diagonal, 10 elementary, 82 equivalent, 77 formula for the matrix product, 67 identity matrix, 10 non-singular, 93 product, 67 seerank, 126 regular, 93 similar, 77, 104 singular, 93 square, 71 transition matrix, 73 transposed, 45 upper triangular, 104 maximal linear dependent subset, 30
Z,
21
regular, 21
see
scalar, 21 scalar multiplication, 21 semigroup, 121 set, 113 cartesian product, 114 dierence, 114 empty set, 113 equality, 113 intersection, 114 subset, 113 union, 114
see
determinant
sgn(),
see
sign function
a set
Mn (F ),
71
see
relation
permutation, 105
change of coordinates for linear maps, 75 dimension formula for linear maps, 58
INDEX
129
dimension formula for subspaces, 42 existence of a basis, 33, 120 existence of a determinant, 95 formula for the matrix product, 67 invariance of dimension, 39 total order,
see
relation
transposition, 105 vector, 21 addition of vectors, 21 canonical unit vectors, 28 coordinate vector, 36, 68 scalar multiplication, 21 zero vector, 22 vector space, 21, 122 basis, 27 dimension, 38, 39 nite dimensional, 39 nitely generated, 28 generating system, 28 intersection, 24 isomorphic, 54 linear dependent subset, 29 subspace, 23 sum, 24 zero space, 23
WV , Z,
59 integers
see
Bibliography
[Bla84]
Andreas Blass,
Linear algebra, AddisonWesley, 1966. Algebraic structures, AddisonWesley, 1967. Falko Lorenz, Lineare algebra 1, 3rd ed., Wissenschaftsverlag, 1992.
131