143

4.4 TRIDIAGONALIZATION OF A REAL SYMMETRIC MATRIX

Let A = (a
ij
) be a real symmetric nxn matrix. Our aim is to get a real symmetric
tridiagonal matrix T such that T is similar to A. The process of obtaining this T is called
the Givens Householder scheme. The idea is to first find a reduction process which
annihilates the off tridiagonal matrices in the first row and first column of A and
repeatedly use this idea. We shall first see some preliminaries.

Let
|
|
|
|
|
.
|

\
|
=
n
2
1
U
U
U
U
M
be a real nx1 nonzero vector.

Then, H = UU
T
is an nxn real symmetric matrix. Let be a real number (which we
shall suitably choose) and consider

( ) I . .......... UU I H I P
T
= =

We shall choose such that P is its own inverse. (Note that P
T
= P). So we need

P
2
= I

i.e.
(I - H) (I - H) = I
i.e.
(I - UU
T
) (I - UU
T
) = I


I 2 UU
T
+
2
UU
T
UU
T
= I

So we choose such that


144


2
UU
T
UU
T
= 2 UU
T


Obviously, we choose 0. Because otherwise we get P = I; and we dont get any new
transformation.
Hence we need
UU
T
UU
T
= 2 UU
T


But
U
T
U = U
2
1
+ U
2
2
+ .. + U
2
n
=
2
U

is a real number 0 (since U is a nonzero vector) and thus we have

(U
T
U). UU
T
= 2 UU
T


and hence


) II ..( ..........
U U
2
T
=


Thus if U is an nx1 nonzero vector and is as in (II) then P defined as

) III ....( .......... UU I P
T
=

is such that

) IV ....( .......... P P P
1 T
= =

Now we go back to our problem of tridiagonalization of A. Our first aim is to find a P
of the form (IV) such that PAP AP P
t
= has off tridiagonal entries in 1
st
row and 1
st

column as zero. We can choose the P as follows:

Let ( ) V a a a s n ...... .......... ..... 1
2
31
2
21
2 2
+ + + =
(the sum of the squares of the entries below the 1
st
diagonal in A)


145

Let s = nonnegative square root of s
2
.

Let


|
|
|
|
|
|
.
|

\
|
+
=
1
31
21 21
. sgn
0
n
a
a
a s a
U
M
. (VI)


Thus U is the same as the 1
st
column of A except that the 1
st
component is taken as 0 and
second component is a variation of the second component in the 1
st
column of A. All
other components of U are the same as the corresponding components of the1
st
column
of A.



Then


1
T
2
U U

(
(

=


( )
1
2
1 n
2
41
2
31
2
21 21
2
a ..... a a a . sgn s a

(
(

+ + + + +
=



1
2
1 n
2
31 21
2 2
21
2
a ..... a s a 2 s a

(
(

+ + + + +
=

==
( )
1
21
2 2
1 n
2
31
2
21
2
a s 2 s a ..... a a

(
(

+ + + + +



146


( )
1
21
2 2
1 n
2
31
2
21
2
a s 2 s a ..... a a

(
(

+ + + + +



[ ]
1
21
2
2
2

+
=
a s s
since
2
1 n
2
31
2
21
2
a a a s + + + = L


21
2
1
a s s +
=


21
2
1
a s s +
=
(VII)

Thus if is as in (VII) and U is as in (VI) where s is as in (V) then

P = I - UU
T


is s.t. P = P
T
= P
-1
, and it can be shown that

A
2
= PA
1
P = PAP ( letting A
1
= A)

is similar to A and has off tridiagonal entries in 1
st
row and 1
st
column as 0.

Now we apply this procedure to the matrix obtained by ignoring 1
st
column and 1
st
row
of A
2
.

Thus we now choose
2
2
42
2
32
2 2
..... n a a a s + + + =

(where now a
ij
denote entries of A
2
)



147
(i.e. s
2
is sum of squares of the entries below second diagonal of A
2
)

s = Positive square root of s
2



|
|
|
|
|
|
|
|
.
|

\
|
+
=
2
42
32 32
) . (
0
0
n
a
a
s a sign a
U
M




32
2
1
a s s +
=


P = I - UU
T


Then

A
3
= PA
2
P

has off tridiagonal entries 1n 1
st
, 2
nd
rows and columns as zero. We proceed similarly
and annihilate all off tridiagonal entries and get T, real symmetric tridiagonal and
similar to A.

Note: For an nxn matrix we get tridiagonalization in n 2 steps.

Example:

|
|
|
|
|
.
|

\
|
=
4 2 1 1
2 4 1 1
1 1 5 4
1 1 4 5
A



148

A is a real symmetric matrix and is 4 x 4. Thus we get tridiagonalization after (4 2)
i.e. 2 steps.

Step 1:


24264 . 4 18
18 1 1 4
2 2 2 2
= =
= + + =
s
s



( )( ) 97056 . 34
1
4 24264 . 4 18
1 1
21
2
=
+
=
+
=
a s s


= 0.02860


|
|
|
|
|
.
|

\
|
+
=
|
|
|
|
|
.
|

\
|
+
=
1
1
24264 . 4 4
0
. sgn
0
41
31
21 21
a
a
a s a
U



With this , U, we get


|
|
|
|
|
.
|

\
|



= =
97140 . 0 02860 . 0 23570 . 0 0
02860 . 0 97140 . 0 23570 . 0 0
23570 . 0 23570 . 0 94281 . 0 0
0 0 0 1
t
UU I P








149

|
|
|
|
|
.
|

\
|

= =
5 . 3 5 . 1 1 0
5 . 1 5 . 3 1 0
1 1 6 24264 . 4
0 0 24264 . 4 5
2
PAP A


Step 2


( ) ( )
41421 . 1 2
2 1 1
2 2 2
= =
= + =
s
s



( )( )
29289 . 0
41421 . 3
1
1 41421 . 1 2
1 1
32
2
= =
+
=
+
=
a s s



|
|
|
|
|
.
|

\
|

=
|
|
|
|
|
.
|

\
|

=
|
|
|
|
|
.
|

\
|
+
=
1
41421 . 2
0
0
1
41421 . 1 1
0
0
. sgn
0
0
42
32 32
a
a s a
U


P = I - UU
T




|
|
|
|
|
.
|

\
|

=
70711 . 0 70711 . 0 0 0
70711 . 0 70711 . 0 0 0
0 0 1 0
0 0 0 1


A
3
= PA
2
P


|
|
|
|
|
.
|

\
|

=
2 0 0 0
0 5 41421 . 1 0
0 41421 . 1 6 24264 . 4
0 0 24264 . 4 5




150
which is tridiagonal.
Thus the Givens Householder scheme for finding the eigenvalues involves two steps,
namely,

STEP 1: Find a tridiagonal T (real symmetric) similar to T (by the method
described above)

STEP 2: Find the eigenvalues of T (by the method of sturm sequences and
bisection described earlier)

However, it must be mentioned that this method is used mostly to calculate the
eigenvalue of the largest modulus or to sharpen the calculations done by some other
method.

If one wants to calculate all the eigenvalues at the same time then one uses the
Jacobi iteration which we now describe.