Review of eigenvalues/eigenvectors in linear algebra

When a square matrix, A, is symmetrical, i.e.

A = AT or (Ax, y) = (x, Ay), (1)

the following properties are held:

1. All the eigenvalues, λi ’s, are real.

2. All the eigenvectors, en , are orthogonal, i.e. (en , em ) = 0 if n ̸= m. Although not a requirement, it is
more convenient if all the eigenvectors are normalized (||en || = 1).
3.  
λ1 0 0 ...
 0 λ2 0 ... 
 
P T AP =  0 0 λ3 ...  (2)
 
.. .. ..
. . .
where
P = (e1 , e2 , e3 , . . . en ) . (3)

i.e. A can be diagonalized. Note that P satisfies

P −1 = P T , (4)
1
and is called a unitary matrix. It is also called a transformation matrix as it is used to rotate a
vector. 2

Example 1
( )
2 1
A= (5)
1 2
The eigenvalues can be found by solving
 
 2 − λ, 1 
  = (2 − λ)2 − 1 = (λ − 1)(λ − 3) = 0 (6)
 1, 2−λ 

so (λ1 , λ2 ) = (1, 3) and the eigenvectors are (after normalizing)

( )
√1
λ1 = 1, e1 = 2 (7)
− √12

 
1 Proof:
e1 ( )
1, 0, ..., 0
P P =  (e1 , e2 , . . . , en ) =
T e2
0, 1, ..., 0 = I,
...
0, 0, ... 1
en
so
P T = P −1 .

2 Proof:

AP = A(e1 , e2 , . . . , en ) = (λ1 e1 , λ2 e2 , . . . , λn en ),
so
   λ 0 0 ...

1
e1
 0 λ2 0 ... .
P −1 AP =   (λ1 e1 , λ2 e2 , . . . , λn en ) =  0
e2
...
0 λ3 ... 
en .. .. ..
. . .
where ei · ej = δij was used.

1
 ( )
√1
λ2 = 3, e2 = 2 (8)
√1
2

The unitary matrix, P , can be constructed from e1 and e2 as

( )
√1 √1
P = (e1 , e2 ) = 2 2 . (9)
− √12 √1
2

Verify that
P −1 = P T . (10)
Thus, ( )
1 0
P −1 AP = . (11)
0 3
Example 2 – Coupled springs

Figure 1: Coupled springs.

The equations of motion for each mass are expressed as

d2 x1
m = −k1 x1 − k2 (x1 − x2 ), (12)
dt2
d2 x2
m 2 = −k2 (x2 − x1 ), (13)
dt
which can be rewritten in matrix-vector form as
( ) 2 ( ) ( )( )
m 0 d x1 k1 + k2 −k2 x1
=− , (14)
0 m dt2 x2 −k2 k2 x2
or
d2
M x = −Kx. (15)
dt2
Note that K is symmetrical. Invoke the transformation of

x = P y, (16)

and Eq.(15) becomes

d2 y
M = −P −1 KP y
dt2 ( )( )
λ1 0 y1
= − , (17)
0 λ2 y2
or
d2 y1
m = −λ1 y1 (18)
dt2
d2 y2
m 2 = −λ2 y2 . (19)
dt

2
Note that y1 and y2 are no longer coupled. The general solutions of Eqs.(18) and (19) for y1 and y2 are

√ √
λ1 λ1
y1 = A sin t + B cos t,
m m
√ √
λ2 λ2
y2 = C sin t + D cos t,
m m
where A ∼ D are integral constants that need to be determined from the initial conditions of x1 (0) and
x2 (0).
As
x = P y,
it follows
y = P T x,
i.e.
y1 (0) = p̃11 x1 (0) + p̃12 x2 (0),
y2 (0) = p̃21 x1 (0) + p̃22 x2 (0),
where p̃ij are the components of P T .
Therefore, the initial conditions of x can be translated into the initial condition of y and once y can be
solved, x can be obtained from
x = P y,
Example 3 – Stress

Stress components can be represented by a fully-populated matrix with shear components. A 2-D stress
state can be expressed as
( )
σx τxy
σij = , (20)
τxy σy
where σx and σy are the normal stresses and τxy is the shear stress. Note that σij is symmetrical because
τxy = τyx .
The traction force, t, acting on a surface with its normal, n, can be found from the stress tensor as

σn = t. (21)

When the orientation of the surface is varied, there must be a surface orientation where the normal and the
traction force are aligned as

When this happens, it follows

σn = λn, (22)
which implies that σ is the principal stress and n is the principal direction.