4.

3 - Modal Analysis Review of the Eigenvalue Problem

• Physical coordinates are not always the easiest to work in Start with Mx ( t ) + Kx = 0 where x is a
• Eigenvectors provide a convenient transformation to modal
coordinates vector and M and K are matrices. Assume
– Modal coordinates are linear combination of physical
coordinates
– Say we have physical coordinates x and want to transform
initial conditions x 0 and x 0 . Rewrite as
to some other coordinates u 1 1
M 2 M 2 x + Kx = 0 and let
u1 = x1 + 3x 2 ⎡ u1 ⎤ ⎡1 3 ⎤ ⎡ x1 ⎤
⇒⎢ ⎥=⎢
(
⎥⎢ ⎥ q
u2 = x1 − 3x 2 ⎣u2 ⎦ ⎣1 −3⎦ ⎣ x2 ⎦
M 2 x = q ⇒ x = M − 2 q (coord. trans. #1)
1 1

1 2

Eigenvectors = Mode Shapes?

Eigenproblem (cont)
− 12 Mode shapes are solutions to Mω 2 u = Ku
Premultiply by M to get
−1 −1 −1
in physical coordinates. Eigenvetors are
M 2 MM 2 q + M 2 KM 2 q = q + Kq = 0
1

characteristics of matrices. The two are

(4.55) related by a simple transformation,
• Now we have a symmetric, real matrix
but they are not synonymous.
• Guarantees real eigenvalues and distinct, mutually
orthogonal eigenvectors
3 4
 Eigenvectors vs. Mode Shapes The Matrix of eigenvectors can be used to
decouple the equations of motion
The eigenvectors of the symmetric PD matrix K˜ If P orthonormal (unitary), P T P = I ⇒ P T = P −1
are orthonormal, i.e., P T P = I. Are the mode shapes Thus, P T KP = Λ = diagonal matrix of eigenvalues.
−1
orthonormal? Using the transformation x = M 2 q, Back to q + Kq = 0. Make the additional coordinate
− 12 1
the modes shapes U = M P ⇒ P = M U. Now, 2
transformation q =Pr and premultiply by P T
1 1
P T P = U T M 2 M 2U = U T MU = I. Thus, the mode P T P r + P T KP r = I r + Λr =0 (4.59)
shapes are orthogonal only w.r.t. the mass matrix. • Now we have decoupled the EOM i.e., we have n
− 12 − 12
Similarly, U T KU = P T M KM P = Λ independent 2nd-order systems in modal
K˜ coordinates r(t)
5 6

This transformation takes the problem from couple equations

Writing out equation (4.59) yields in the physical coordinate system in to decoupled equations in
the modal coordinates
⎡1 0 ⎤ ⎡ r1 ( t ) ⎤ ⎡ω1
2
0 ⎤ ⎡ r1 (t ) ⎤ ⎡ 0 ⎤
⎢0 1 ⎥ ⎢ r (t ) ⎥ + ⎢ 0 ⎥ =
ω22 ⎥⎦ ⎢⎣ r2 (t ) ⎥⎦ ⎢⎣ 0 ⎥⎦
(4.60)
r1
⎣ ⎦ ⎣ 2 ⎦ ⎢⎣ (ω1)2
x1 x2
k1
1
k2
r1 ( t ) + ω r (t ) = 0 (4.62)
2
1 1 m1 m2
⇒
r2 ( t ) + ω r (t ) = 0 (4.63)
2
2 2
r2
(ω2)2
We must also transform the initial conditions 1
⎡ r1 (0) ⎤ ⎡ r10 ⎤ Physical Coordinates.
r0 = ⎢ ⎥ = ⎢ ⎥ = P T q (0) = P T M 1/2 x (0) (4.64) Coupled equations
⎣ r2 (0) ⎦ ⎣ r20 ⎦
−1 Modal Coordinates.
⎡ r1 (0) ⎤ ⎡ r10 ⎤
r0 = ⎢ ⎥ = ⎢ ⎥ = P T q (0) = P T M 1/2 x (0) (4.65) x = M Pr 2 Uncoupled equations
⎣ r2 (0) ⎦ ⎣ r20 ⎦ Figure 4.5
7 8
 The free response is calculated for each
Modal Transforms to SDOF mode independently using the formulas
1
• The modal transformation P M 2
T
from chapter 1:
transforms our 2 DOF into 2 SDOF systems ri 0
ri ( t ) = sin ωi t + ri 0 cosωi t , i = 1,2
• This allows us to solve the two decoupled ωi
SDOF systems independently using the or (see Window 4.3 for a reminder)
methods of chapter 1
ri 20 ωi ri 0
• Then we can recombine using the inverse ri ( t ) = r + 2
i0 sin(ωi t + tan −1 ), i = 1,2
ωi2 ri 0
transformation to obtain the solution in
terms of the physical coordinates. Note, the above assumes neither frequency is zero

9 10

Once the solution in modal coordinates is

determined (ri) then the response in Physical Steps in solving via modal
Coordinates is computed: analysis (Window 4.4)
• With n DOFs these transformations are:
x (t) = S r(t)
n ×1 n × n n ×1
−1

S =M P
2
where
n × n nxn nxn
(where n = 2 in the previous slides) 11 12
 Modal and Physical Responses Section 4.4 More then 2 Degrees
Modal 4
Free response in modal coordinates

r1
of Freedom
Coordinates: 2 r2
Independent
oscillators 0

-2

λ1 = 2 ⇒ ω1 = 2 -4
0 1 2 3 4 5 6 7 8 9 10
⇒ T1 = 2π = 4.44 sec, π sec 4.44 sec
Fig 4.8
λ 2 = 4 ⇒ ω1 = 2 4
Free response in physical coordinates

⇒ T2 = π sec x1 Extending previous

2
x2
Physical 0 section to any number of
Coordinates:
Coupled
-2 degrees of freedom
oscillators -4
Note IC 0 1 2 3 4 5 6 7 8 9 10
Time (s)
13 Fig 4.714

A FBD of the system of figure 4.8 yields the relevant matrices and vectors are:
the n equations of motion o the form:
⎡ k1 + k 2 −k 2 0 0 ⎤
⎡ m1 0 0⎤
mi + ki (xi − xi −1 ) − ki +1 (xi −1 − xi ) = 0, i = 1, 2, 3L n (4.83) ⎢0 m2 0⎥
⎢ −k
⎢ 2
k2 + k3 −k3 0 ⎥
⎥
M =⎢ ⎥, K = ⎢ 0 −k 3 ⎥ (4.83)
⎢ ⎥ ⎢ ⎥
⎢ ⎥ ⎢ k n−1 + k n −k n ⎥
Writing all n of these equations and ⎣⎢ 0 0 mn ⎦⎥ ⎢ 0 0 −k n k n ⎥⎦
⎣
casting them in matrix form yields:
⎡ x1 ( t ) ⎤ ⎡ x1 ( t ) ⎤
Mx (t ) + Kx (t ) = 0, (4.80) ⎢ x (t ) ⎥ ⎢ x (t ) ⎥
x (t ) = ⎢ ⎥ , x (t ) = ⎢ 2 ⎥
2

⎢ ⎥ ⎢ ⎥
where: ⎢ ⎥ ⎢ ⎥
⎢⎣ x n ( t ) ⎥⎦ ⎢⎣ x n (t ) ⎥⎦
15 16
For such systems as figure 4.7 and The Mode Summation Approach
4.8 the process stays the same…just • Based on the idea that any possible time response
more modal equations result: is just a linear combination of the eigenvectors
Process stays the same as section 4.3 Starting with q ( t ) + Kq ( t ) = 0 (4.88)
r1 (t ) + ω12 r1 (t ) = 0
( )v
n n
let q (t )= ∑ q i (t ) = ∑ ai e
− j λi t j λi t
r2 (t ) + ω 22 r2 (t ) = 0 + bi e i
i =1 i =1
r3 (t ) + ω 32 r3 (t ) = 0 ⇒ two linearly independent solutions for each term.
Just get more modal
equations, one for each n

rn (t ) + ω n2 rn (t ) = 0
degree of freedom (n is can also write this as q (t ) = ∑ d i sin (ωi t + φi ) v i (4.92)
the number of dof) i =1

See example 4.4.2 for details 17 18

Mode Summation Approach (cont)

Mode Summation Approach (cont) Solve for d i and φi from the two IC equations
Find the constants d i and φi from the I.C.
n n v Ti q (0) ω v T
i i q (0)
q (0) = ∑ d i sin φi v i and q (0) = ∑ d iωi cosφi v i di = and φi = tan −1 T
i =1 i =1
sinφi v i q (0)
Assuming eigenvectors normalized such that v Tj v i = δ ij IMPORTANT NOTE about q (0)= 0
n n if you just crank it through the above expressions
T
v q (0)=v
j
T
j ∑ d sinφ v =∑ d sinφ ( v v ) = d sinφ
i =1
i i i
i =1
i i
T
j i j j
you might conclude that d i = 0, i.e., the trivial soln.
Similarly for the initial velocities, v Tj q (0)=d jω j cosφ j Be careful with q (0) = 0 as well.
19 20
Mode Summation Approach for zero initial Mode Summation Approach with rigid
displacement body modes (ω1 = 0)
if λ1 = 0,
If q (0) = 0, the return to
(
q1 (t) = a1e− j 0t
+ b1e j 0t
)v i = (a1 + b1 )v i
n
q (0) = ∑ d i sinφi v i
does not give two linearly independent solutions.
Now we must use the expansion
i =1
( )v
n
q(t) = (a1 + b1t )v 1 + ∑ ai e− j λi t λi t
and realize that φi = 0 instead of d i = 0. i=2
+ bi e j i

The compute d i from the velocity expression and adjust calculation of the constants from the
initial conditions accordingly.

v Ti q (0)=ωi d i cosφi Note that the underline term is a translational motion

21 22

Example 4.3.1 solved by the mode Example 4.3.1 constructing the summation of modes
summation method
⎡ q1 (t ) ⎤ 3 2 ⎛ π ⎞ 1 ⎡ −1 ⎤ 3 2 ⎛ π ⎞ 1 ⎡ −1 ⎤
⎡3 0 ⎤ 1 ⎡ −1 −1⎤ =
⎢ q (t ) ⎥ 2 sin ⎜ 2 t − ⎟ + sin ⎜ 2 t − ⎟
From before, we have M = ⎢ 2 ⎠ 2 ⎢⎣ −1⎥⎦ 2 2 ⎠ 2 ⎢⎣ 1 ⎥⎦
1/2
⎥ and V = ⎝ ⎝
⎣0 1 ⎦ 2 ⎢⎣ −1 1 ⎥⎦ ⎣ 2 ⎦
the first mode the second mode

⎡3⎤ ⎡0 ⎤
Appropriate IC are q 0 =M 1/2 x 0 = ⎢ ⎥ , q 0 =M 1/2 v 0 = ⎢ ⎥ Transforming back to the physical
⎣0 ⎦ ⎣0 ⎦
coordinates yields:
⎡− π ⎤
ωi v q (0)
T
i
T
i
φ
ωi v q (0) ⎡ 1 ⎤ ⎢ 2 ⎥ 3 2 ⎛ π ⎞ 1 ⎡ 1 3 0 ⎤ ⎡ −1⎤ 3 2 ⎛ π ⎞ 1 ⎡ 1 3 0 ⎤ ⎡ −1⎤
φi = tan −1 T = tan −1 ⇒⎢ ⎥=⎢ x(t) = M −1/2 q = sin ⎜ 2t −
⎟⎠ ⎢ ⎥⎢ ⎥ + sin ⎜ 2t − ⎟ ⎢ ⎥
v i q (0) 0 ⎣φ2 ⎦ π⎥ 2 ⎝ 2 2 ⎢⎣ 0 1 ⎥⎦ ⎣ −1⎦ 2 ⎝ 2 ⎠ 2 ⎢⎣ 0 1 ⎥⎦ ⎢⎣ 1 ⎥⎦
⎢− ⎥
⎣ 2⎦ 3 2 ⎛ π ⎞ 1 ⎡ −1⎤ 3 2 ⎛ π ⎞ 1 ⎡ −1⎤
= sin ⎜ 2t − ⎟ ⎢ ⎥ + sin ⎜ 2t − ⎟
2 ⎝ 2 ⎠ 3 2 ⎣ −1⎦ 2 ⎝ 2 ⎠ 3 2 ⎢⎣ 1 ⎥⎦
⎡3 2 ⎤
v Ti q (0) ⎡ d1 ⎤ ⎢ 2⎥
di = ⇒⎢ ⎥=
sinφi ⎣ d 2 ⎦ ⎢3 2 2 ⎥
⎣ ⎦ 23 24
 Example 4.3.1 a comparison of the Steps for Computing the
two solution methods shows they Response By Mode Summation
yield identical results 1. Write the equations of motion in matrix
1
free response of DOF 1
form, identify M and K
modal
0.5

0
MSA
2. Calculate M -1/2 (or L)
3. Calculate K˜ = M − KM −
1 1
-0.5
2 2

-1
0 1 2 3 4 5
time (sec)
6 7 8 9 10
4. Compute the eigenvalue problem for the
K˜ and get ω i2 and v i
free response of DOF 2
3

2 matrix
1

-1
5. Transform the initial conditions to q(t)
modal

q(0) = M 2 x(0) and q(0) = M 2 x(0)

-2 MSA 1 1
-3
0 1 2 3 4 5 6 7 8 9 10
time (sec)

25 26

Summary of Mode Summation Continued

6. Calculate the modal expansion
Nodes of a Mode Shape
coefficients and phase constants • Examination of the mode shapes in
⎛ T ⎞
−1⎜ ω i v i q(0) ⎟ vTi q(0) Example 4.4.3 shows that the third entry of
φi = tan , di = the second mode shape is zero!
⎜ vT q(0) ⎟ sin φi
⎝ i ⎠
• Zero elements in a mode shape are called
7. Assemble the time response for q nodes.
n
q(t) = ∑ di sin(ω i t + φ i )v i • A node of a mode means there is no motion
i=1
of the mass or (coordinate) corresponding to
that entry at the frequency associated with
8. Transform the solution to physical that mode.
n
x(t) = M 2 q(t) = ∑ di sin(ω i t + φ i )ui
coordinates − 1

i=1 27 28
 The second mode shape of A rigid body mode is the mode
Example 4.4.3 has a node associated with a zero frequency
• Note that for more then 2 DOF, a mode Fig 4.12
shape may have a zero valued entry
• This is called a node of a mode.
⎡ 0.2887 ⎤
⎢ 0.2887 ⎥
u2 = ⎢ ⎥ • Note that the system in Fig 4.12 is not constrained and can
move as a rigid body
⎢ 0 ⎥
⎢ ⎥ • Physically if this system is displaced we would expect it to
node ⎣ −0.2887 ⎦ move off the page whilst the two masses oscillate back and
forth
They make great mounting points in machines
29 30

Example 4.4.4 Rigid body motion Following the steps of Window 4.4
The free body diagram of figure 4.12 yields ⎡1 0⎤
1. M −1/2
=⎢ 1⎥
m1 x1 = k ( x 2 − x1 ) and m2 x 2 = − k ( x 2 − x1 ) ⎢0 ⎥
⎣ 2⎦
⎡ m1 0 ⎤ ⎡ x1 ⎤ ⎡ 1 −1⎤ ⎡ x1 ⎤ ⎡ 0 ⎤ ⎡1 0 ⎤ 1 −1 ⎡1 0 ⎤
⇒⎢ +k = ⎡ ⎤⎢ ⎡ 400 −200 ⎤
m2 ⎥⎦ ⎢⎣ x 2 ⎥⎦ ⎢⎣ −1 1 ⎥⎦ ⎢⎣ x 2 ⎥⎦ ⎢⎣ 0 ⎥⎦ 2. K = M KM = 400 ⎢ 1 ⎥ ⎢ ⎥
−1/2 −1/2
1 ⎢=
⎣0 ⎢0
⎣ 2⎦
⎥ ⎣ −1 1 ⎥⎦ ⎢0
⎣ 2⎦
⎥ ⎣ −200 100 ⎥⎦

Solve for the free response given: ⎛ ⎡ 4 − λ −2 ⎤ ⎞

3. det ( K − λ I ) = 100det ⎜⎜ ⎢ ⎥ ⎟⎟ = 100 ( λ − 5λ ) = 0
2

m1 = 1 kg, m2 = 4 kg, k = 400 N subject to ⎝⎣ −2 1 − λ ⎦⎠

⎡ 0.01⎤ ⇒ λ1 = 0 and λ2 = 5 ⇒ ω1 = 0, ω2 = 2.236 rad/s
x0 = ⎢ ⎥ m and v 0 = 0
⎣ 0 ⎦
Indicates a rigid body motion
31 32
Now calculate the eigenvectors and note in
particular that they cannot be zero even if the 5. Calculate the matrix of mode shapes
eigenvalue is zero ⎡1 0 ⎤ ⎡ 0.4472 −0.8944 ⎤ ⎡ 0.4472 −0.8944 ⎤
⎡ 4 − 0 −2 ⎤ ⎡ v11 ⎤ ⎡ 0 ⎤ S = M −1/2 P = ⎢ ⎥ ⎢0.8944 0.4472 ⎥ = ⎢ 0.4472 0.2236 ⎥
λ = 0 ⇒ 100 ⎢ ⎣ 0 1/ 2 ⎦⎣ ⎦ ⎣ ⎦
⎥ ⎢ ⎥ = ⎢ ⎥ ⇒ 4 v11 − 2 v21 = 0
⎣ −2 1 − 0 ⎦ ⎣ v21 ⎦ ⎣ 0 ⎦
⎡ 0.4472 1.7889 ⎤
⎡1 ⎤ ⎡ 0.4472 ⎤ ⇒ S −1 = ⎢ ⎥
⇒ v 1 = ⎢ ⎥ or after normalizing v 1 = ⎢ ⎥ ⎣ −0.8944 0.8944 ⎦
⎣2 ⎦ ⎣0.8944 ⎦ 7. Calculate the modal initial conditions:
⎡ −0.8944 ⎤ ⎡ 0.4472 −0.8944 ⎤
Likewise: v 2 = ⎢ ⎥ ⇒ P = ⎢ 0.8944 0.4472 ⎥ ⎡ 0.4472 1.7889 ⎤ ⎡ 0.01⎤ ⎡ 0.004472 ⎤
⎣ 0.4472 ⎦ ⎣ ⎦ r(0) = S −1 x 0 = ⎢
−0.8944 0.8944 ⎥ ⎢ 0 ⎥ = ⎢ −0.008944 ⎥
⎣ ⎦⎣ ⎦ ⎣ ⎦
As a check note that r(0) = S −1 x 0 = 0
P T P = I and P T KP = diag ⎡⎣0 5 ⎤⎦
33 34

7. Now compute the solution in modal Applying the modal initial conditions
coordinates and note what happens to to these two solution forms yields:
the first mode. r1 (0) = a = 0.004472
Since ω1 = 0 the first modal equation is r1 (0) = b = 0.0
r1 + (0)r1 = 0 ⇒ r1 (t ) = 0.0042
⇒ r1 (t) = a + bt Rigid body translation as in the past problems the initial conditions for r2 yield
And the second modal equation is r2 (t ) = −0.0089cos 5t
r2 (t) + 5r2 (t) = 0 ⎡ 0.0042 ⎤
⇒ r( t ) = ⎢ ⎥
⇒ r2 (t) = a2 cos 5t −0.0089cos 5t
Oscillation
35
⎣ ⎦ 36
8. Transform the modal solution to Order the frequencies
the physical coordinate system • It is convention to call the lowest frequency ω1 so
that ω1 < ω2 < ω3 < …
• Order the modes (or eigenvectors) accordingly
⎡ 0.4472 −0.8944 ⎤ ⎡ 0.0045 ⎤
x(t) = Sr(t) = ⎢ ⎥⎢ ⎥ • It really does not make a difference in computing
⎣ 0.4472 0.2236 ⎦ ⎣ −0.0089 cos 5t ⎦ the time response
⎡ x1 (t) ⎤ ⎡ 2.012 + 7.60 cos 5t ⎤ −3
• However:
⇒ x(t) = ⎢ ⎥ =⎢ ⎥ × 10 m – When we measuring frequencies, they appear lowest to
⎣ x2 (t) ⎦ ⎢⎣ 2.012 − 1.990 cos 5t ⎥⎦ highest
– Physically the frequencies respond with the highest
energy in the lowest mode (important in flutter
Each mass is moved a constant distance and calculations, run up in rotating machines, etc.)
then oscillates at a single frequency.
37 38

The system of Example 4.1.5 Transform coordinates

solved by Mode Summation ⎡9 0 ⎤ 1 ⎡3 0⎤ −1 ⎡1 / 3 0 ⎤
M =⎢ ⎥ ⇒ M 2
= ⎢0 1 ⎥ and M 2
= ⎢ 0 1⎥
From Example 4.1.6 we have: ⎣0 1 ⎦ ⎣ ⎦ ⎣ ⎦

⎡1 ⎤ −⎡ 1 ⎤ Thus the initial conditions become

ω 1 = 2, u1 = ⎢ 3 ⎥ , ω 2 = 2, u 2 = ⎢ 3 ⎥
⎢⎣ 1 ⎥⎦ ⎢⎣ 1 ⎥⎦
1 ⎡3 0 ⎤ ⎡ 1 3 ⎤ ⎡1 ⎤
q (0 ) = M x (0 ) = ⎢
2
=
⎣0 1 ⎥⎦ ⎢ 1 ⎥ ⎢⎣ 1 ⎥⎦
Use the following initial conditions and note that only one mode
⎣ ⎦
should be excited (why?)
⎡1 ⎤
1 ⎡3 0⎤ ⎡0⎤ ⎡0⎤
⎡0⎤ q (0 ) = M x (0 ) = ⎢
2
=
x (0 ) = ⎢ 3 ⎥ , x (0 ) = ⎢ ⎥
⎣0 1 ⎥⎦ ⎢⎣ 0 ⎥⎦ ⎢⎣ 0 ⎥⎦
⎢⎣ 1 ⎥⎦ ⎣0⎦

39 40
Transform Mode Shapes to Eigenvectors From Equation (4.92):
2 2
1 ⎡ 3 0 ⎤ ⎡ 1 3 ⎤ ⎡1⎤
v1 = M 2 u1 = ⎢ q(t ) = ∑ di sin(ωi t + φi ) v i ⇒ q(t ) = ∑ diωi cos(ωi t + φi ) v i
⎥⎢ ⎥ = ⎢ ⎥
⎣ 0 1 ⎦ ⎢⎣ 1 ⎥⎦ ⎣1⎦ i =1 i =1
eigenvectors
1 ⎡ 3 0 ⎤ ⎡ −1 3 ⎤ ⎡ −1⎤ Set t=0 and multiply by v1:
v 2 = M 2 u2 = ⎢ ⎥⎢ ⎥=⎢ ⎥
⎣ 0 1 ⎦ ⎢⎣ 1 ⎥⎦ ⎣ 1 ⎦ 2
Note that unlike the mode shapes, the eigenvectors q(0) = ∑ diωi cos φi v i
i =1
are orthogonal: 0
⎡ −1⎤ ⎡− 1 ⎤ 2 ⎡0 ⎤ 1 ⎡1⎤ 1 ⎡− 1⎤
Note that v1T v 2 = [1 1]⎢ ⎥ = 0, but u1T u 2 = ⎡⎣ 1 3 1⎤⎦ ⎢ 3 ⎥ = ≠ 0 ⇒ v1T ⎢ ⎥ = d1 2 cos φ1v1T ⎢ ⎥ + d 2 2 cos φ2 v1T ⎢ ⎥
⎣1⎦ ⎢⎣ 1 ⎥⎦ 3 ⎣0 ⎦ 2 ⎣1⎦ 2⎣1⎦
Normalizing yields: 1 ⎡1⎤ 1 ⎡ −1⎤ ⇒ 0 = d1 cosφ1 ⇒ φ1 = π /2
v1 = ⎢1⎥ and v 2 = ⎢ ⎥
2 ⎣ ⎦ 2⎣1⎦ Or directly from Eq. (4.97)
41 42

From the initial displacement: Transforming Back to Physical Coordinates:

T
1 ⎡1⎤
= [1 1] ⎢ ⎥ =
v q(0) 2
d1 = 1
−1 ⎡1 0⎤ ⎡1⎤
sin(π / 2) 2 ⎣1⎦ 2 x(t) = M 2
q(t) = ⎢ 3 ⎥ cos( 2t) ⎢ ⎥
(4.98) ⎢⎣ 0 1 ⎥⎦ ⎣1⎦
v T2 q(0) 1 ⎡1⎤
d2 = = [−1 1] ⎢ ⎥ = 0 ⎡1 ⎤
sin(π / 2) 2 ⎣1⎦ cos 2t
= ⎢3 ⎥
⎢ ⎥
thus 2
Eigenvector 1
⎢⎣ cos 2t ⎥⎦
q(t) = ∑ di sin(ωi t + φi )v i Eigenvector 2
i=1 1
⇒ x1 (t) = cos 2t and x 2 (t) = cos 2t
1 ⎡1⎤ ⎡1⎤ 3
= 2 cos( 2t) ⎢ ⎥ = cos( 2t)⎢ ⎥
2 ⎣1⎦ ⎣1⎦ So, the initial conditions generated motion only in the
first mode (as expected)
43 44
 Alternate Path to Symmetric Cholesky Decomposition
Let M = U T U where U is upper triangular
Single-Matrix Eigenproblem
Introduce the coordinate transformation
• Square root of matrix conceptually easy, but
computationally expensive Ux = q ⇒ x = U −1q ⇒ U T U x + K x = 0
q U −1q
M − 2 MM 2 q + M − 2 KM − 2 q = q + Kq = 0
1 1 1 1

premultiply by U − T to get
• More efficient to decompose M into product
of upper and lower triangular matrices I q + U − T KU −1q = q + K q = 0
note that: ⎡⎣U − T KU −1 ⎤⎦ = [U −1 ] K T [U − T ] = U − T KU −1
T T T
(Cholesky decomposition)

45 46