9.

Properties of Matrices

Block Matrices
It is often convenient to partition a matrix M into smaller matrices called
blocks, like so:
 
1 2 3 1 !
4 5 6 0 A B
M = =
 
7 8 9 1 C D
0 1 2 0
   
1 2 3 1  
Here A = 4 5 6, B = 0, C = 0 1 2 , D = (0).
   
7 8 9 1

• The blocks
! of a block matrix must fit
! together to form a rectangle. So
B A C B
makes sense, but does not.
D C D A

• There are many ways to cut up an n × n matrix into blocks. Often

context or the entries of the matrix will suggest a useful way to divide
the matrix into blocks. For example, if there are large blocks of zeros
in a matrix, or blocks that look like an identity matrix, it can be useful
to partition the matrix accordingly.

• Matrix operations on block matrices can be carried out by treating

the blocks as matrix entries. In the example above,
! !
2 A B A B
M =
C D C D
!
A2 + BC AB + BD
=
CA + DC CB + D2

1
 Computing the individual blocks, we get:
 
30 37 44
A2 + BC =  66 81 96 
 
102 127 152
 
4
AB + BD = 10
 
16
 
18
CA + DC = 21
 
24
CB + D2 = (2)

Assembling these pieces into a block matrix gives:

 
30 37 44 4
 66 81 96 10
 
102 127 152 16
 

4 10 16 2

This is exactly M 2 .

The Algebra of Square Matrices

Not every pair of matrices can be multiplied. When multiplying two matri-
ces, the number of rows in the left matrix must equal the number of columns
in the right. For an r × k matrix M and an s × l matrix N , then we must
have k = s.
This is not a problem for square matrices of the same size, though.
Two n × n matrices can be multiplied in either order. For a single matrix
M ∈ Mnn , we can form M 2 = M M , M 3 = M M M , and so on, and define
M 0 = In , the identity matrix.
As a result, any polynomial equation can be evaluated on a matrix.

Example Let f (x)!= x − 2x2 + 3x3 .

1 t
Let M = . Then:
0 1
! !
2 1 2t 1 3t
M = , M3 = ,...
0 1 0 1

2
 Then:
! ! !
1 t 1 2t 1 3t
f (M ) = −2 +3
0 1 0 1 0 1
!
2 6t
=
0 2

Suppose f (x) is any function defined by a convergent Taylor Series:

1 00
f (x) = f (0) + f 0 (0)x + f (0)x2 + . . .
2!
Then we can define the matrix function by just plugging in M :
1 00
f (M ) = f (0) + f 0 (0)M + f (0)M 2 + . . .
2!
There are additional techniques to determine the convergence of Taylor Se-
ries of matrices, based on the fact that the convergence problem is simple for
diagonal matrices. It also turns out that exp(M ) = 1+M + 21 M 2 + 3!1 M 3 +. . .
always converges.

Matrix multiplication does not commute. For generic n × n square

matrices M and N , then M N 6= N M . For example:
! ! !
1 1 1 0 2 1
=
0 1 1 1 1 1

On the other hand: ! ! !

1 0 1 1 1 1
=
1 1 0 1 1 2
Since n × n matrices are linear transformations Rn → Rn , we can see
that the order of successive linear transformations matters. For two linear
transformations K and L taking Rn → Rn , and v ∈ Rn , then in general
K(L(v)) 6= L(K(v))!
Finding matrices such that M N = N M is an important problem in
mathematics.

Trace
Matrices contain a great deal of information, so finding ways to extract
essential information is useful.

3
Definition The trace of a square matrice M = (mij ) is the sum of its
diagonal entries.
n
X
tr M = mii
i=1

Example  
2 7 6
tr 9 5 1 = 2 + 5 + 8 = 15
 
4 3 8

While matrix multiplication does not commute, the trace of a product

of matrices does not depend on the order of multiplication:
X
tr(M N ) = tr( Mli Njl )
l
XX
= Mli Nil
i l
XX
= Nil Mli
l i
X
= tr( Nil Mli )
i
= tr(N M ).
Thus, tr(M N ) = tr(N M ) for any square matrices M and N .
In the previous example,
! !
1 1 1 0
M= ,N = .
0 1 1 1
! !
2 1 1 1
MN = 6= N M = .
1 1 1 2
However, tr(M N ) = 2 + 1 = 3 = 1 + 2 = tr(N M ).
Another useful property of the trace is that:
tr M = tr M T
This is true because the trace only uses the diagonal entries, which are fixed
! ! !T
1 1 1 2 1 2
by the transpose. For example: tr = 4 = tr = tr
2 3 1 3 1 3
Finally, trace is a linear transformation from matrices to the real num-
bers. This is easy to check.

4
Linear Systems Redux Recall that we can view a linear system as a ma-
trix equation
M X = V,
with M an r × k matrix of coefficients, x a k × 1 matrix of unknowns, and V
an r × 1 matrix of constants. If M is a square matrix, then the number of
equations (r) is the same as the number of unknowns (k), so we have hope
of finding a single solution.
Above we discussed functions of matrices. An extremely useful function
1 1 1
would be f (M ) = M , where M M = I. If we could compute M , then we
1
would multiply both sides of the equation M X = V by M to obtain the
1
solution immediately: X = M V.
Clearly, if the linear system has no solution, then there can be no hope of
1
finding M , since if it existed we could find a solution. On the other hand, if
1
the system has more than one solution, it also seems unlikely that M would
1
exist, since X = M V yields only a single solution.
1
Therefore M only sometimes exists. It is called the inverse of M , and is
usually written M −1 .

References
Wikipedia:

• Trace (Linear Algebra)

• Block Matrix

Review Questions
!
1 2 0
1. Let A = . Find AAT and AT A. What can you say about
3 −1 4
matrices M M T and M T M in general? Explain.

2. Compute exp(A) for the following matrices:

!
λ 0
• A=
0 λ
!
1 λ
• A=
0 1

5
 !
0 λ
• A=
0 0
!
a b
3. Suppose ad − bc 6= 0, and let M = .
c d

(a) Find a matrix M −1 such that M M −1 = I.

(b) Explain why your result explains what you found in a previous
homework exercise.
(c) Compute M −1 M .
 
1 0 0 0 0 0 0 1
0 1 0 0 0 0 1 0
 
0 0 1 0 0 1 0 0
 
 
0 0 0 1 1 0 0 0
4. Let M =  . Divide M into named blocks,
0 0 0 0 2 1 0 0

0 0 0 0 0 2 0 0
 
0 0 0 0 0 0 3 1
 

0 0 0 0 0 0 0 3
and then multiply blocks to compute M 2 .