EIGENVALUES,

NATURAL FREQUENCIES &

MODAL ANALYSIS
 EIGENVALUES AND NATURAL FREQUENCIES

• The method of solution explored so far can be extended to

symmetric algebraic eigenvalue problem.
• This allows the use of mathematical software packages, and sets
the background needed for analysing systems with an arbitrary
number of degrees of freedom.
• In addition, the important concepts of mode shapes and natural
frequencies can be generalized by connecting the undamped
vibration problem to the mathematics of the algebraic eigenvalue
problem
• A number of ways can be used to connect the solution of the
vibration problem with that of the algebraic eigenvalue problem
• The most productive approach is to cast the structural dynamic
problem as a symmetric eigenvalue problem because of the special
properties associated with symmetry.
• Note that the physical nature of both the mass and stiffness matrices
is that they are usually symmetric.
• Hence preserving this symmetry is also a natural approach to
solving the vibration problem
• Since M is symmetric and positive definite, it may be factored into
two terms

Where:
• L is a special matrix with zeros in every position above the
diagonal (called a lower triangular matrix).
• A matrix M is positive definite if the scalar formed from the
product
Some matrix and vector reminders
éa b ù -1 1 é d -b ù
A=ê ú ÞA = ê ú
ë c c û ad - cb ë -c a û
x x = x1 + x2
T 2 2

é m1 0 ù
M=ê ú Þ x Mx = m1 x1 + m2 x2
T 2 2

ë 0 m2 û
M > 0 Þ x Mx > 0 for every value of x except 0
T

Then M is said to be positive definite

 for every nonzero choice of the vector x
• The factorization L is called the Cholesky decomposition and is
examined, along with the notion of positive definite
• In solving a single-degree-of-freedom system it was useful to
divide the equation of motion by the mass

• Hence, consider resolving the system of two equations described

in matrix form by det −𝜔2 𝑀 + 𝐾 = 0 by making a coordinate
transformation that is equivalent to dividing the equations of
motion by the mass in the system

• To that end, consider the matrix square root defined to be the matrix
M1/2 such that M1/2M1/2 = M, the mass matrix.
• For the simple example of the mass matrix,

the mass matrix is diagonal and the matrix square root becomes
simply
Eqn. 1

• The inverse of the diagonal matrix M1/2, denoted by M-1/2,

becomes simply

Eqn. 2
• The matrix of eqn. 1 provides a means of changing coordinate
systems to one in which the vibration problem is represented by
a single symmetric matrix

• To accomplish this transformation, or change of coordinates, let the

vector x in equation,

Eqn. 3

and multiply the resulting equation by M -1/2 yields:

Eqn. 4
 From Eqn 4;

Since 𝑀−1/2 𝑀𝑀−1/2 = 𝐼, the identity matrix, eqn (4) reduces

to
Eqn. 5

෩ = 𝑀−1/2 𝐾𝑀−1/2 , like the matrix 𝐾, is a symmetric

• The matrix 𝐾
matrix

• The matrix ෩
𝐾is called the mass-normalized stiffness and is
analogous to the single-degree-of freedom constant, 𝑘/𝑚.

• Equation 5 is solved, as before, by assuming a solution of the

form 𝐪(𝑡) = 𝐯𝒆𝒋𝝎𝒕

• where 𝐯 ≠ 𝟎 is a vector of constants.

Substitution of this form into equation 5 yields

Again, here v ≠ 0
 Eqn. 6

upon dividing by the nonzero scalar 𝑒 𝑗𝜔𝑡

• Here it is important to note that the constant vector v cannot be zero
if motion is to result.
• Next let 𝜆 = 𝜔2 in equation 6, which yields:

Eqn. 7
Again, here v ≠ 0

• The scalar 𝜆 in eqn 7 is called the eigenvalue and v is called the

(corresponding) eigenvector.
• Since the matrix K is symmetric, this is called the symmetric
eigenvalue problem. The eigenvector v generalizes the concept of a
mode shape u used in previous lecture.
• If the system being modelled has n degrees of freedom, each free to
move with a single displacement labelled 𝑥𝑖 (𝑡), the matrices M, K,
and hence 𝐾 ෩ will be 𝑛 × 𝑛, and the vectors x(t), q(t), and v will be
𝑛 × 1 in dimension.

• Each subscript 𝑖 denotes a single degree of freedom where 𝑖 ranges

from 1 to 𝑛, and the vector x(t) denotes the collection of the 𝑛
degrees of freedom.

• It is also convenient to label the frequencies ω and

eigenvectors v with subscripts 𝑖, so that 𝜔𝑖 and 𝐯𝒊 denote the
𝑖th natural frequency and corresponding 𝑖th eigenvector,
respectively

• In 2DOF system, 𝑛 = 2 but the notation is useful and valid for

any number of degrees of freedom.
 Review of Eigenvalue Problems
• The algebraic eigenvalue problem is the problem of computing
the scalar λ and the nonzero vector v satisfying

• Here, A is an 𝑛 × 𝑛 real valued, symmetric matrix

• The vector v is 𝑛 × 1

• There will be n values of the scalar λ, called the eigenvalues, and

n values of the corresponding vector v, one for each value of λ

• The vectors v are called the eigenvectors of the matrix A.

 Review of Eigenvalue Problems
• The eigenvalues of A are all real numbers

• The eigenvectors of A are real valued

• The eigenvalues of A are positive numbers if and only if A is

positive definite

• The eigenvectors of A can be chosen to be orthogonal, even for

repeated eigenvalues.
 Example 1
෩ for
Calculate the matrix 𝐾

Solution
෩ is computed from the following expression
𝐾
෩ = 𝑀−1/2 𝐾𝑀−1/2
𝐾

Let’s first compute 𝐾𝑀−1/2 and multiply the product by 𝑀−1/2

෩
𝐾
 Example 2

Solve the eigenvalue and vectors for the two-degree-of-freedom

system of Example 1 where:

solution

The eigenvalue problem is to calculate the eigenvalues λ and

eigenvectors v that satisfy eqn 7, which is:
 Rewriting equation 7 yields

• where v must be nonzero

• Hence the matrix coefficient must be singular and therefore its
determinant must be zero
This last expression is the characteristic equation and has the two roots

𝜆1 = 2 𝑎𝑛𝑑 𝜆1 = 4
෩
These are the eigenvalues of 𝐾

Note that these are also the squares of the natural frequencies,
𝜔𝑖2 as calculated in the examples of previous lecture

The eigenvector associated with λ1 is calculated from equation

This results in the two dependent scalar equations

Hence 11 = 21, which defines the direction of the vector v1

To fix a value for the elements of v1, the normalization condition of

equation applies:

This is used to force v1 to have a magnitude of 1. This results in

(setting 11 = 21 )
 Solving for 21 yields:

so that the normalized vector v1 becomes

ෙ = λ𝐯 solving for the elements

Similarly, substitution of λ = 4 into 𝐾𝐯
2

of v2, and normalizing the result yields

so that the set of vectors v1 and v2 are orthogonal as well as normal
KEY POINTS TO TAKE NOTE:

• Eigenvectors of a symmetric matrix are orthogonal and can

always be calculated to be normal.
• Such vectors are called orthonormal
• This fact can be used to decouple the equations of motion of any
order undamped system by making a new matrix P out of the
normalized eigenvectors, such that each vector forms a column.
• Thus the matrix P is defined by

Where:
n is the number of degrees of freedom in the system

Note the matrix P has the unique property that 𝑃𝑇𝑃 = 𝐼,

KEY POINTS TO TAKE NOTE:

From Previous example: Using the values for the orthonormal

vectors v1 and v2

Note that: 𝑃𝑇𝑃 = 𝐼,

 MODAL ANALYSIS
• The matrix of eigenvectors P introduced earlier can be used to
decouple the equations of vibration into two separate equations

• The two separate equations are then 2nd DOF equations that can be
solved and analysed using the typical 2DOF methods

• The matrices P and M-1/2 can be used again to transform the

solution back to the original coordinate system

• The matrices P and M-1/2 can also be called transformations, which

is appropriate in this case because they are used to transform the
vibration problem between different coordinate systems

• This procedure is called modal analysis, because the

transformation S = M-1/2P, often called the modal matrix, is related
to the mode shapes of the vibrating system
Consider the matrix form of the equation of vibration
Eqn. 8

subject to the initial conditions

and

vector of initial displacements and velocities respectively

1
−2
Substituting 𝐱 𝑡 = 𝑀 𝒒(𝑡) into Eqn 8 and multiplying from left
by 𝑀−1/2 yields:

Eqn. 9
 since the matrix M is constant, and where

ෙ = 𝑀−1/2 𝐾𝑀−1/2 as before

𝐾

The transformation 𝑀−1/2 simply transforms the problem from

the coordinate system defined by to a new
coordinate system defined by

Eqn. 10

Next, define a second coordinate system:

Eqn. 11

ෙ
where P is the matrix composed of the orthonormal eigenvectors of 𝐾
Substitution of the vector Eqn. 10 into Eqn. 9 and multiplying from the
left by the matrix 𝑃𝑇 yields:

Eqn. 12

Using the result 𝑃𝑇𝑃 = 𝐼 and ; this can be

reduced to

Eqn. 13

Equation 13 can be written out by performing the indicated matrix

calculations as

Eqn. 14

Eqn. 15
• The equality of the two vectors in this last expression implies the
two decoupled equations

Eqn. 16

Eqn. 17

• These two equations are subject to initial conditions which must

also be transformed into the new coordinate system r(t) from the
original coordinate system x(t)

• Following the preceding two transformations applied to the

initial displacement yields

Eqn. 18
Likewise the initial velocity in the decoupled coordinate system, r(t),
becomes

Eqn. 19

• Equations 14 and 15 are called the modal equations

• The coordinate system r(t) = [r1(t) r2(t)]T is called the modal

coordinate system.

• Denoting the initial conditions individually by 𝑟10 , 𝑟10

ሶ , 𝑟20 , 𝑟20
ሶ and
using SDOF solution eqns, the solution of each of the modal
equations 16 and 17 is:
 Eqn. 20

Eqn. 21

The solution stands provided 𝜔1 and 𝜔2 are nonzero

• Once the modal solutions 20 and 21 are known, the

transformations M1/2 and P can be used on the vector r(t) = [r1(t)
r2(t)]T to recover the solution x(t) in the physical coordinates x1(t)
and x2(t).
• To obtain the vector x from the vector r, substitute equations 10
into q(t) = Pr(t) to get
 Eqn. 22

The matrix product M1/2 P is again a matrix, which is denoted by

Eqn. 23

• This is the same size as M and P (2 x 2).

• The matrix S is called the matrix of mode shapes, each column of
which is a mode-shape vector.
• This procedure, referred to as modal analysis, provides a means
of calculating the solution to a 2DOF vibration problem by
performing a number of matrix calculations.
Fig below summarizes how computing the matrix of mode shapes S
transforms the vibration problem from a coupled set of equations of
motion into a set of single DOF problems.

Fig. 1: Schematic illustration of decoupling equations of motion using

modal analysis and the matrix of mode shapes S.
 EXAMPLE
Using Modal analysis, calculate the solution of the 2DOF system given
by:

Compare the result to that obtained in Example we did in previous

example for the same system and initial conditions.
 Solution

The next step is to calculate the matrix S and its inverse

As a check, it can be seen that:
𝑆𝑆 −1 = 𝐼
And that: 𝑆 𝑇 𝑀𝑆 = 𝐼

Equations 18 and 19 yield that the modal solutions are:

The solution in the physical coordinate system x(t) is calculated
from