47
INTRODUCTION
(b)
and substituting in (a) lead to the following set of algebraic equations relating
the frequency, co, and the amplitudes, AI, A2:
2
(k1 + k 2 )A - k 2A2 = n1co A 1
2co
-k 2 Al + k 2A2 =
Characteristic-Value
Problems and
Quadratic Forms
)x
+ a,2x -2
)X2
kL + k2---
(2-1)
(2-2)
or
(2-3)
(a
- 21 2)X
A1
inYZ1n
+ kjy 2 - k 2 (y2 - Y) = 0
is a hybrid of the
* Also called "eigenvalue" problem in some texts. The term "eigenvalue"
German term Eigenwerte and English "value."
46
712
nrz
A, +
k2
/'2
4A1
(e)
A-2 =A2
M2
k2
The values of 2 for which nontrivial solutions of (2-1) exist are called the
characteristicvalues of a. Also, the problem of finding the characteristic values
and corresponding nontrivial solutions of(2-) is referred to as a second-order
characteristic-value problem.*
The characteristic-value problem occurs naturally in the free-vibration
analysis of a linear system. We illustrate for the system shown in Fig. 2-1.
The equations of motion for the case of no applied forces (the free-vibration
case) are
m2 y 2 + k2 (y 2 - Y) = 0
m2
(d)
I
)2
A = A 12,
A.2 = A 2
INTRODUCTION
(all
A2
We can transform (c) to a form similar to that of(2-1) by defining new amplitude
measures,*
1/1-1I1n
2-1.
CHARACTERISTIC-VALUE PROBLEMS
48
SEC. 2-2.
CHAP. 2
Solving (a),
Al = +6
= +1
(2)
-2
al
all -1]
x2 =
A1
;11,
2-2.
49
= +i
1
-I-
where i =
are possible only if the determinant of the coefficient matrix vanishes, that is,
I
I
Ia_l l
when
-U12
(2->)
Ia21
a22
(all -
- 0
22
(2-7)
f11 = all + a 2 2
al1 a2 2 - a
2 a2 1
al
(all -
= (all - a 2 2)
+ 4(a 12)
l)(a2 2 -
1)
- a1 2 a21
X C~~1)
1 011
X(t)
(d)
-~~~~
_ all - A
a1 2
(I)
+
2511
By definition,
(a)
(e)
Example 2-1
- (2)(2)= 67
fI = 2
6
(2)(5) - (2)(2)
fl
(c)
Since this quantity is never negative, it follows that the characteristic values for a
symmetrical second-order matrix are always real.
= 0
(b)
Since only one equation is independent and there are two unknowns, the solu
tion is not unique. We define x(?, x21) as the solution for A = Al. Assuming*
that a 2 # 0, the solution of the first equation is
a 21
(2-8)
The roots of (2-8) are the characteristic values of a. Denoting the roots by
il, A2,the solution is
(2-9)
42)2
A1 , 2 = (WIt + I-
2
A - 7
(a)
We let
pi - 4
-_
Expanding (2-5) results in the following equation (usually called the characteristic equation) for A:
\'-O
2
I-- o}
0
' - (all + a )A + (alla2 2 - a 2 1a 12 )
1f2
l)xl + al 2x
QTQ = 1
(2-10)
2
11
(2-11)
50
CHARACTERISTIC-VALUE PROBLEMS
SEC. 2-2.
aQ =
Following the same procedure for
=
A2,
(1)
we obtain
1
[2
(2-13)
- 2b2}
all -+-
-4x, + 2x 2 = 0
22
2x
Also,
Q2Q 2 = 1
2
(2-14)
These equations are linearly independent, and the two independent solutions
are
x(l) = {c 1 , 0}
(b)
X(2) = {0, C2 }
The corresponding characteristic vectors are
a 12
-A2)
{1,2}
= 2 = + 1, we find
and
(2)= c2{1, -}
2
1
Q2 = / {1, -} = -/5 {2, - 1}
i = 1,2
jQj
QIQ 2 = Q2Q 1 = 0
(2)
(all -- A,)(a
X1) = 2X(1 ) = 2c 1
Q, =
Repeating for
= 0
x( )= cl{1, 2}
(2-15)
- x
X(?) = C1
Then,
and
Q = {+1,0}
Q2
{0, + 1}
-c
i;2Q2
QTQ2
= +
= 1 = + 6 are
at2
aQ
2]
A1 = +6
where
(i\2
a:[1
3]
The characteristic values and corresponding normalized solutions for this matrix are
21 = +5
(a)
2 = -1
15
{4 , - 1}
Q =
al
- 22 = -(a
22
) =
-a
1
all -
(b)
1
and we see that Q Q 2 = 0. This result is also valid when the roots are equal.
In general, QQ2 # 0 when a is unsymmetrical. Two nth order column vectors
U, V having the property that
UT V=VTU = 0
51
Example 2-2
(2-12)
Q1
c2 {1
Q2 =
CHAP. 2
QTQ 2
-5
(3)
(2-16)
= +i
)2 = -i
We have included this example to illustrate the case where the characteristic values are
I IF PROBLEMS
A nrTRI .lTIC.A_\ A
rV,
52
CHAP. 2
SEC. 2-3.
I1v-., ,..
cI
(1 - i)xt - 2x 2 = 0
(1 + i)x 2 = 0
Note that the second equation is (1 - i) times the first equation. The general solution is
I-
Repeating for
= C2 {1, W
~~-~L'2
When the roots are complex, 22 is the complex conjugate of At,.Now, we take c2 = c.
(
Then, x(2) is the complex conjugate of x 1) We determine c, such that
2
(X( t))Tx( )=
QfQ 2 = Q2Q1 = 0
Also, by definition,
QtQ
Al,2-- +i
= QQ2 = 1
Q1 2-
In general, the characteristic values are complex conjugate quantities when the elements
of a are real. Also, the corresponding characteristic vectors are complex conjugates.
f
i
The characteristic vectors for the second-order system satisfy the following
relations:
(a)
aQ1 =AiQ1
aQ
=A 2 Q2
Q2] =
LQtJfI [Qt
q-1
2-3.
(2-19)
The matrix operation, p- ( )p where p is arbitrary, is called a similaritytransformation. Equation (2-19) states that the similarity transformation, q-l'( )q,
reduces a to a diagonal matrix whose elements are the characteristic values
of a.
If a is symmetrical, the normalized characteristic vectors are orthogonal,
that is,
= .2, we find
X(2)
q-laq
{1, 2j}
x(l)
(2-20)
qT
Now, we let
Q2 1
= [Q1 Q21'
(b)
Example 2-3
(1)
1, ==
[2 2]
Q2]
q = [Q
,
(2-17)
2, = +6
Q1 =
- {(1,2}
normalized
A2 =
+1
Q2 =1
{2,-1}
(2-18)
* This terminology has developed from dynamics, where the characteristic vectors define the
normal modes of vibration for a discrete system.
p'
is called an
53
We have shown that the characteristic vectors are always linearly indepen
dent when a is symmetrical. They are also independent when a is unsymA2. Then, ql # 0 except for the case where a is
metrical, provided that Al
unsymmetrical and the characteristic values are equal. If ql # 0, q-t exists
and we can express (2-18) as
=[A
q= [Q,
0]i= [+
+0
Q2 = ]= [ 2_ -1]
Q2
CHAP. 2
CHARACTERISTIC-VALUE PROBLEMS
54
2rl
X2
/5
5 2
qTaq
-1]
-1
1]
I [I - 1 126
1 5 2a
_2
01 = [I
[0
5
= 6
0
1]
2-4.
-1|
1]
=
IL
rql//
+ a2 X,, = AX 2
+ a2 x 2
4/X/
[l/6
2//7
...
(2-22)
+ an,,,x, = Ax,
ax = Ax
(a -
(2-23)
n)x =
-1/17
Q= L//
1/J17
85-
al 2 x 1 + a2 2 x2 +
q 1. Actually,
+ alnXn = AX 1
L 4,- 1
Q2=
[O
allxl + a 2 X2 +
an 1 ,
Q1 =-A={2, + 1}
- 5
-1
J=-I
1
A1, = +5
aq
(2)
A
2
55
21
1 r
-1|
0
[5
5[0
2]1
21
I r2
v/
2] [1
-1][2
T I[
5[2
(Iq
SEC. 2-4
2 5/3]
- ,/1/31
(n- _
3n-1
+ p 2,n -2 _
** + (1)
n) = O
where
(2-25)
0] = h
I1
q-aq=
q' 10
and /3j is the sum of all the jth order minors that can be formed on the diag,An denote the roots, and expressing the characteristic
onal.* Letting Al, A2,
equation in factored form, we see that
(3)
i -21
[=L+i
+ ,
fl = A1 + 2 +
+
... + An-1An
A21A
+
=
A21A
2
3
2
fn =
q= 2/3
al
i2 1
[-i
In this case, q involves complex elements. Since the characteristic vectors are complex
conjugates, they are linearly independent and q-' exists. We find q-', using the definition equation for the inverse (Equation (1-50)):
1I2 .i..
-Adj
Iqi
q= 3/4I +
We summarize below the theoretical results for the real symmetrical case.
The proofs are too detailed to be included here (see References 1 and 9):
1. The characteristic values A1, A2, ... , Ain, are all real.
. , Qn , are orthogonal:
2. The normalized characteristic vectors Q 1,Q,
QTQj = 6ij
1~lAdjq
I =
(2-26)
(2-26)
i,j = 1,2.
j2
* Minors having a diagonal pivot (e.g., delete the kth row and column). They are generally called
principal minors.
CHARACTERISTIC-VALUE PROBLEMS
56
3.
CHAP. 2
4) = 0
)2
= 3
23 = -1
22 =-3
Q1n
QtQ2 *
2)((1-
(3 -
where
57
QUADRATIC FORMS
SEC. 2-5.
= [iJj]
(1 - )x
+2x 2
2x
+(1 -
(a)
)x2
Example 2-4
(3 - i,)X 3
33
- 922 + 182 - 6 =
+0.42
+ 2.30
23,
+6.28
(1 - 2)X3 =
I=1,2,3
I - Aj
(2)
O-
j2
L+0.8
+0.68
-0.52
-0.84
+0.54
-0.10
{A2'
we
' }
{0,0,1}
xi
- -x
31)
Q3
x3' = 0
and
'
2,O]
This example illustrates the case of a symmetrical matrix having two equal characteristic
inde
values. The characteristic vectors corresponding to the repeated roots are linearly
roots.
repeated
the
for
I
rank
of
is
I13
pendent. This follows from the fact that a -
QUADRATIC FORMS
l lx
2 ral
La.2
.0
X3 = C2
2xl + 2x 2 = 0
2x, + 2 2 = 0
4x = 0
for 23 are
vector
characteristic
and
The general solution
2-5.
)x2
is
Solving the first and third equations for xt and x3 in terms of x2, the general solution
+0.50
Q2
=- 1,(a) reduces to
X2
X2 = C1
= C1
Ql
2x 2
When
_-X = -(3
q = [QtQ2Q3
We see from (b) that (a - 213) is of rank 1 when 2 = 3. The general solution of (b) is
(b)
By specializing the constants, we can obtain two linearly independent solutions for the
repeated root. Finally, the characteristic vectors for L = `2= 3 are
= +11 + 5 + 2= +18
,3 = 5(2) - (-2)(-2) - +6
,,
A,
2
= 0
(O)X3 = 0
2,
-2X2 =
2xt
il, fi2,
Since a is symmetrical, its characteristic values are all real. We first determine
25
+9
=
I
5 3+
using (2- ):
+2x
- 2x I
01
5 -2
3 - -2
0 -1 I1
a]
a22
{X
x2
xT
58
CHARACTERISTIC-VALUE PROBLEMS
CHAP. 2
SEC. 2-5.
F =
k=1 j=1
ajkxjxk = x ax
(2-27)
=Fall
F
Lal2
where ajk = akj, for j =A k, is said to be a quadratic form in xl, x2 ,..., x,,.
If F = xTax is nonnegative (2 0) for all x and zero only when x = 0, we call
F a positive definite quadratic form. Also, we say that a is a positive definite
matrix. If F > 0 for all x but is zero for some x :A 0, we say that F is positive
semidefinite. We define negative definite and negative semidefinite quadratic
forms in a similar manner. A quadratic form is negative definite if F < 0 for
all x and F = 0 only when x = 0. The question as to whether a quadratic form
is positive definite is quite important. For example, we will show that an
equilibrium position for a discrete system is stable when a certain quadratic
form is positive definite.
Consider the quadratic form
2
F = blx + b 2x2 +
21 + 22 =
'. '
xlA
b2
...
X2
LO=0
... -"
(2-28)
12>0
(b)
Ai >0
A2 >0
(c)
a,
x,
= all
= a
a l la 2
al 2 a1 2
31> 0
b >0
Then, letting
y = qTx
x = qy
(2-29)
A =
(2-30)
It follows that F is positive definite with respect to y when all the characteristic
values of a are positive. But y is uniquely related to x and y = 0 only when
x = 0. Therefore, F is also positive definite with respect to x. The problem of
establishing whether xTax is positive definite consists in determining whether
all the characteristic values of a are positive.
(2-31)
2 > 0
(2-32)
f >
f22>
,, > 0
(2-33)
where flj is the sum of all the jth-order principal minors. Equivalent conditions
can be expressed in terms of the discriminants. Let Aj represent the deter
minant of the array consisting of the first j rows and columns.
ai
f2 >
A, >0
= (XTq)[.ijijy](qX)
(d)
> 0.
b >0
(a)
Jai
al = alla22 - a 2 > 0
Since a > 0, it follows from the second requirement in (d) that a2 2
Therefore, (d) is equivalent to (b). We let
or
b 2 > 0"
all >0
#3I>o0
A2 =
bl
are equivalent to
al 1 a2 2 -
+ bx,
,
= all + a2 2
A;l2 =
bl
al2
a2
XJ
bl > 0
59
[XIX ...
=
QUADRATIC FORMS
al1
a1 2
:j12
22
alj
c2j
..
atj
a2
.*
(2-34)
ajj
The conditions,
ik,
N,
I
A1 > 0 A 2 > 0
are sufficient for a to be positive definite.*
* See Ref. 1 for a detailed proof. Also see Prob. 2-15.
A,, >0
(2-35)
CHARACTERISTIC-VALUE PROBLEMS
60
CHAP. 2
Example 2-5
(1)
9.
11.
A2 = 2 -1 = +1
A3 = 1(6-4)- 1(3 -2)
12.
13.
+ 2 + 3 = +6
P2 = (2 - 1) +-(3 - 1) + (6 - 4) = +5
PROBLEMS
f3 = A3 = +1
(2)
1
L
1
Since
A2
Since all the discriminants are positive, this matrix is positive definite. The corresponding
invariants are
=
8.
1
12 2
1 2 3
A1 =
2-1.
b = pTap = p- ap
B = bTb
= pTaTp = p ap
(2-36)
ax = AX
(2-37)
Now, b and a have the same characteristic values.* This follows from
I,)p = la - I,
(2-38)
that x1 and x2 are linearly dependent when one is a scalar multiple of the other.
Using (2-10) and (2-13), show that Q1 and Q2 are linearly independent when
;il
= 212-
2-3.
REFERENCES
1. HILDEBRAND, F. B.: Methods of Applied Mathematics, Prentice-Hall, New York,
1952.
2. BODEWIG, E.: Matrix Calculus, Interscience Publishers, New York, 1956.
3. SMIRNOV, V. I.: LinearAlgebra, Addison-Wesley Publishing Co., Reading, Mass.,
1964.
4. TURNBULL, H. W., and A. C. AITKEN: An Introduction to the Theory of Canonical
Matrices, Dover Publications, New York.
5. HADLEY, G.: Linear Algebra, Addison-Wesley Publishing Co., Reading, Mass., 1961.
* See Prob. 2-5.
(b)
AI, = p-'(a -
(a)
By
lb-
-2
61
PROBLEMS
(a)
(b)
r3 21
2
0
0 5 0
3 0 2
2-4. Following the procedure outlined in Prob. 2-1, determine the charac
teristic values and modal matrix for
12y, + 12y 2 = 4y,
12y + 63y2 = 9y2
CHARACTERISTIC-VALUE PROBLEMS
62
2-5.
Then,
Il = la - )1,l
lb -
k= 1,2,..., n
3(a)
/3(b) =
Note: F(a) = a2 -
Demonstrate for
x82,
1r -2
- 1
P=
'1
Show that
a-
(a)
(b)
2-10.
qTaq = -
for n
ia + /3
2 3)
8x 2x
:finite is
dc
a=2
7l
aa
2
a 3 = aa
aar =
= aar-1
aa'
If fal # O,a-
(a2 -
la -+ 212.
2 31
..
63
PROBLEMS
CHAP. 2
P(a) = c0 a
a22
ca
C Cn-
laC ,,-1
Show that b is positive definite only when the rank of C is equal to n. What
can we say about b when r(C) < n?
2-15. Consider the quadratic form
all
F[X1 X2
x]L
12
al
a
a1 2
a22
..
' xi
a2n
X2
ann-
X,
+ c2 a + c3 In
+
aln
a2n*
,u
knrlxv
64
o A rTI:ITl'.-\ALuIE
?[ IV
I L- IV
65
PROBLEMS
CHAP. 2
PROBLEMS
.......
We partition a symmetrically,
(pxp)
r[All
[X 1X2 AT
F
x
(qp)
(pxq)
(p
A1 2
(q
1)
form
where q = n - p. The expansion of F = XTaX has the
F = X A, 1 X
3 0 0
0 2 1
0 0 5 21
matrices,
2-18. Suppose we express a as the product of two quasi-triangular
for example,
2
X2
A22
(q x q)
1)
+ 2XTA 12 A2 + X2A 2 2 X
(pxp)
(nxn)
G 21
p = 12,...,n
that it
are necessary conditions for a to be positive definite. Note
conditions.
remains to show that they are also sufficient
(b) Discuss the case where Ap = 0.
2-16. Refer to Prob. t-25. Consider a to be symmetrical.
product of nonsingular
(a) Deduce that one can always express a as the
positive definite.
is
a
when
matrices
triangular
lower and upper
(b) Suppose we take
bll = b22 =
bn,,
+1
i\ =
lb 2 .
= bb 1=22
11
(qxp)
Since /A l1 is equal
For Fp > 0 for arbitrary X 1, All must be positive definite.
that A lll must be
to the product of the characteristic values of All, it follows
positive.
(a) By taking p = 1,2,..., n, deduce that
Ap = IAlll > 0
[G
b2PP
when a is
Show that the diagonal elements of b will always be real
positive definite.
partitioned, the
2-17. If a quasi-diagonal matrix, say a, is symmetrically
that
Establish
matrix.
submatrix All is also a quasi-diagonal
i, j = 1, 2, . . , N
a = [Ai
positive definite.
is positive definite only when Ai (i = 1, 2, . ., N) are
where p
(pxq)
G 2 2j
B1 2
B2 2 J
(qxq)
q = n. We take
B2 2 ' Iq
B 11 = Ip
for arbitrary p when
Show that the diagonal submatrices of g are nonsingular
a is positive definite.