Maths for Signals and Systems

Problem Sheet 6

Problems

Consider a matrix A with eigenvalues 1  1 and 2  1 and eigenvectors x1  cos sin  

T
1.
and x2   sin  cos  . Show that A  AT , A2  I , det( A)  1 , A1  A .
T

Solution

The eigenvectors of this matrix are perpendicular to each other, since:

 sin  
x1 x2  cos sin    0.
T

 cos 
Furthermore, their magnitude is 1, since:
xi  cos2   sin 2   1, i  1,2
2

For the above reasons, we conclude that the eigenvectors of matrix A are orthogonal and
therefore, A is a symmetric matrix and A  AT .
Moreover, A can be diagonalised as A  QQT with Q a matrix which contains the
eigenvectors of A in its columns, i.e., Q  x1 x2 .
(1) 2 0  1 0
A2  QQT QQT  Q2QT (note that QT Q  I ) and 2   2
   I , so that
 0 1   0 1 
A  Q Q  QIQ  QQ  I .
2 2 T T T

det( A)  12  (1) 1  1 .

Finally from A2  I we see that A1  A .

2. Consider a matrix A with A3  0 . Find the eigenvalues of A . Give an example of a matrix of

any size that satisfies A3  0 , with A  0 . In case that a matrix A satisfies A3  0 and is also
symmetric, prove that A  0 .

Solution

The eigenvalues of A3 are 3i , where i are the eigenvalues of A . This can be seen from the
fact that if i is an eigenvalue of A with corresponding eigenvector xi then
A3 xi  A2 Axi  A2i xi  i AAxi  i Ai xi  3i xi . But A3 xi  3i xi  0 and therefore,
3i xi  0  3i  0  i  0 . An example of a matrix which is non-zero but has zero eigenvalues
0 1
is A   with A  0, n . In case where A is symmetric, the diagonalisation of A is
n

0 0
A  QQT with Q an orthogonal matrix that contains the eigenvectors of A.
A  QQ  Q0Q  0 .
T T

1
3. Show that the eigenvalues of a symmetric matrix A with real entries are real.

Solution

Suppose that  is an eigenvalue of A that corresponds to the eigenvector x . In that case we

have the following:
Ax  x  ( Ax)  (x)  A x   x 
Ax   x since A is real.
In the above we transpose both sides and we get:
( Ax )T  ( x )T  ( Ax )T   ( x )T
Now we multiply both sides from the right with x :
( Ax )T x   ( x )T x  ( x )T AT x   ( x )T x  ( x )T Ax   ( x )T x , since A is symmetric.
( x )T x   ( x )T x   ( x )T x   ( x )T x     . Therefore,  is real.

4. (i) A skew-symmetric (or antisymmetric) matrix B has the property BT   B . Show that the
eigenvalues of a skew-symmetric matrix B with real entries are purely imaginary.
(ii) Show that the diagonal elements of a skew-symmetric matrix are 0.

Solution

(i) Bx  x  ( Bx)  (x)  B x   x 

Bx   x since B is real.
In the above we transpose both sides and we get:
( Bx )T  ( x )T  ( Bx )T   ( x )T
Now we multiply both sides from the right with x :
( Bx )T x   ( x )T x  ( x )T BT x   ( x )T x  ( x )T Bx   ( x )T x , since B is skew-
symmetric.
 ( x )T x   ( x )T x   ( x )T x   ( x )T x     . Therefore,  is purely
imaginary. The only real eigenvalue that a skew-symmetric matrix might have is the zero
eigenvalue.
(ii) We know that BT   B and therefore if bij is a random element of B then bij  b ji .
Therefore, for the diagonal elements we have bii  bii and this gives bii  0 .

5. Consider the skew-symmetric matrix

0 1 1 1
 1 0  1 1 
1  .
M
3   1 1 0  1
 
 1  1 1 0
(i) Show that Mv  v , with v any 4-dimensional column vector. What observation can you
make out of this result?
(ii) Using the trace of M , the result of 5(i) and furthermore, the result of Problem 4 above, find
all four eigenvalues of M .

2
 Solution

(i)
0 1 1 1  x   yzw 
 1 0  1 1   y   
Mv 
1      1   x  z  w
3  1 1 0  1  z  3  x  y  w
    
 1  1 1 0   w  x  y  z 
2 1

Mv  ( y  z  w) 2  ( x  z  w) 2  ( x  y  w) 2  ( x  y  z ) 2
3

1 2

 ( y  z 2  w 2  2 yz  2 zw  2 yw)  ( x 2  z 2  w 2  2 xw  2 zw  2 xz)
3

1
 
 ( x 2  y 2  w 2  2 xw  2 yw  2 xy)  ( x 2  y 2  z 2  2 xz  2 yz  2 xy)
3
1
3
  1

 ( y 2  z 2  w2 )  ( x 2  z 2  w2 )  ( y 2  z 2  w2 )  ( x 2  z 2  w2 )
3
2
 x 2  y 2  z 2  w 2  v  Mv  v .
If we take an eigenvalue  of matrix M which correspond to the eigenvector v , we have
that Mv  v  Mv  v . Based on the result Mv  v we have
v  v   v    1 . Therefore, the eigenvalues of the given matrix have magnitude
1.
(ii) In Problem 4 we proved that the eigenvalues of a skew-symmetric matrix are purely
imaginary. For this particular case we also see that they have magnitude of 1. Therefore, for
matrix M we can say that the eigenvalues are i or  i or 0. The matrix has four
eigenvalues and they sum up to zero. The determinant of M is not zero (it is actually
1
9  1 ) which means that the matrix is full rank and therefore, it doesn’t have any zero
 
3
4

eigenvalues. Therefore, two of them are equal to i and the rest are equal to  i .

6. Consider matrices A and B shown below:

0 0 1  1 1 1
A  0 1 0 and B  1 1 1
  1
3
1 0 0 1 1 1
(i) In which of these classes do they belong to? Invertible, orthogonal, projection, permutation,
diagonalisable, Markov.
(ii) Which of the factorisations LU , QR, SS 1 , QQ 1 are possible for A and B ?

Solution

(i) For A we have:

Eigenvalues are -1, 1, 1 and corresponding eigenvectors are  1 0 1 , 1 0 1 and
T T

0 1 0 . The eigenvectors are independent and also orthogonal. The determinant is -1.
T

A is invertible (its determinant is non-zero), orthogonal (its rows are orthogonal to each
other), permutation (obvious), diagonalisable (any invertible matrix is diagonalisable),
Markov (satisfies the Markov properties – rows/columns have positive elements which sum
up to 1). It is not a projection matrix because it doesn’t satisfy the property A2  A .

3
 For B we have:
Eigenvalues are 1, 0, 0 and corresponding eigenvectors are 1 1 1 ,  1 0 1 and
T T

 1 1 0 . The eigenvectors are independent and also orthogonal. The determinant is 0.

T

B is projection ( B 2  B and is symmetric), diagonalizable (it has a set of independent

eigenvectors), Markov (satisfies the Markov properties – rows/columns have positive
elements which sum up to 1). It is not orthogonal or permutation.
(ii) For a 3 3 matrix the LU decomposition looks like:
 a11 a12 a13  l11 0 0  u11 u12 u13 
A  a21 a22 a23   l21 l22 0   0 u 22 u23   LU
a31 a32 a33  l31 l32 l33   0 0 u33 
For the given matrix A we have a11  0 and therefore, at least one of l11 and u11 has to be
zero. In that case either L or U is singular. This is not possible since A is not singular.
Therefore, A doesn’t have an LU decomposition. In order for A to have an LU
decomposition, we must reorder the rows of A , i.e., A must be multiplied from the left
with a permutation matrix P , and in that case we have PA  LU .
A has a QR decomposition as follows:
0 0 1 0 0 1 1 0 0
A  0 1 0  0 1 0 0 1 0
1 0 0 1 0 0 0 0 1

For the given matrix B we find the LU decomposition using elimination, as follows:
1 1 1 1 0 0 1 1 1
1  1
B  1 1 1  1 1 0 0 0 0 , and therefore, B does have an LU
3 3
1 1 1 1 0 1 0 0 0
decomposition.
1 1 1 0.577 0.816 0  0.577 0.577 0.577
B  1 1 1  0.577  0.408 0.707   0
1   0 0  where the matrix
3
1 1 1 0.577  0.408  0.707  0 0 0 
0.577 0.816 0 
0.577  0.408 0.707 
  is orthogonal, therefore B does have a QR decomposition.
0.577  0.408  0.707

Both matrices have SS 1 decomposition since they have a set of independent
eigenvectors.
Both matrices have QQ 1 decomposition since, due to their symmetry, we can choose a
set of independent and also orthogonal eigenvectors.