CHAPTER 5

NUMERICAL EVALUATION OF DYNAMIC RESPONSE

Expected outcome:
1. Students can determine response of SDF systems due to any excitation by numerical
methods.
2. Students can choose appropriate methods and use appropriate parameters for
analysis.

Analytical solution is usually not possible when excitation varies arbitrarily with time
or if the system is nonlinear. Such problems can be solved by numerical time-stepping
methods for integration of differential equations.

5.1 Time-Stepping Method

The equation of motion of an SDF system to be solved is

mu  cu  ku  p  t  for linearly elastic system (5.1)

or
mu  cu  f S  u , u   p  t  for inelastic system (5.2)

with given initial conditions:

u  0   u0 u  0   u0

The applied force is given by a set of discrete values pi  p  ti 

where i = 0, 1, 2, … N . The time interval

ti  ti 1  ti (5.3)
is usually constant, although this is not necessary.

In each step, the response is determined at discrete time instant ti . The displacement

u  ti  , velocity u  ti  and acceleration u  ti  at time ti , denoted by ui , ui , and ui ,

respectively, are assumed to be known and satisfy the equation

5-1
 mui  cui  kui  pi (5.4)

The numerical procedure to be presented will enable us to determine the response

quantities ui 1 , ui 1 , and ui 1 at time step i  1 which satisfy the equation of motion at time

ti 1 :

mui 1  cui 1  kui 1  pi 1 (5.5)

We first apply the procedure to time step i  0 to determine response at time step i  1 and
repeat the procedure again to determine response at time step i  2 and so on. Therefore, this
progressive calculation is called “time-stepping method.”

Figure 5.1 Notation for time-stepping methods

The response at time step i  1 determined from response at time step i is usually not
exact. Many approximate procedures implemented numerically are possible. The
requirements for a numerical procedure are
(1) Convergence—the numerical solution should approach the exact solution as the
time step decreases,

(2) Stability—the numerical solution should be stable even if there is some round-off
error or approximation,

(3) Accuracy—the numerical solution should provide results that are close enough to
the exact solution.

5-2
These issues are very important in numerical methods of solving equations. They will govern
the limitation of time-stepping procedures.

Three types of methods will be discussed:

1) Method based on interpolation of excitation

2) Method based on finite difference expression of velocity and acceleration

3) Method based on assumed variation of acceleration.

5.2 Method based on interpolation of excitation

This method is highly efficient by interpolation excitation during a time step as a

linearly varying function.
pi
p    pi   (5.6)
ti

where pi  pi 1  pi (5.7)

and the time variable  varies from 0 to ti .

Figure 5.2 Notation for linear interpolation of excitation

For simplicity, we will show derivation of this procedure for a system without
damping, although this procedure can be extended to damped systems. The equation to be
solved is

pi
mu  ku  pi   (5.8)
ti

5-3
The response u   over time 0    ti is the sum of three parts:

1) Free vibration due to initial displacement u i and velocity ui at   0

2) Response to step force pi with zero initial condition

pi
3) Response to ramp force  with zero initial condition
ti
Analytical solution derived in Chapters 2 and 4 can be used to determine the above
three parts of responses and we will get

ui pi p   sin n 

u    ui cos n  sin n  1  cos n   i    (5.9)
n k k  ti n ti 

u   ui pi p 1
 ui sin n  cos n  sin n  i 1  cos n  (5.10)
n n k k n ti

Evaluating these equations at   ti gives the displacement ui 1 and velocity ui 1 at time

i 1 :

ui pi p 1
ui 1  ui cos(n ti )  sin n ti   1  cos(n ti )  i n ti  sin n ti   (5.11)
n k k n ti 

ui 1 ui pi p 1
 ui sin n ti   cos n ti   sin n ti   i 1  cos n ti   (5.12)
n n k k n ti 

Substitute pi by Eq. 5.7. Then, these equations can be rewritten as recurrence formulas:

ui 1  Aui  Bui  Cpi  Dpi 1

(5.13)

ui 1  Aui  Bui  C pi  Dpi 1 (5.14)

where coefficients A, B, C , D, A, B, C , and D are constant in terms of k , n ,and t . If the
time step t is constant, the coefficients A , B , … D ' in this procedure need to be
computed only once. These parameters for damped SDF systems are summarized in Table
5.1.

These formulae were derived from exact solution of the equation of motion.
Therefore, the result is exact if the excitation actually varies linearly during each time step as
usually assumed for earthquake ground excitation, which is recorded at closely spaced time

5-4
intervals. Note that the exact solution used in deriving this procedure is available only if the
system is linear.

The only restriction on the size of time step is that it should provide a close
approximation to the excitation function and provide response results at closely spaced time
intervals so that the response peaks are not missed.

Table 5.1 Coefficients in recurrence formulas Eq. (5.13) and (5.14) for damped system

  
A  e n t  sin D t  cos D t 
 1  2 
 

 1 
B  e n t  sin  D t 
 D 

1  2  1  2 2    2   
C   e n t    sin D t  1   cos D t  
k  n t  D t 1  2 
  n t   
 

1 2 n t  2  1
2
2 
D 1   e  sin D t  cos D t  
k  n t  D t n t 

  
A  e n t  n
sin D t 
 1  2 
 

  
B  e n t  cos D t  sin D t 
 1  2 
 

1  1 n t
    1  
C     e  n
  sin D t  cos D t  
k  t  1   2 t 1   2 
 t  
 

1    
D  1  e n t  sin D t  cos D t  
k t   1  2 
  

5-5
Example 5.1

An SDF system has the following properties: m = 0.2533 kip-sec2/in., k = 10 kip/in.,

Tn = 1 sec (ωn = 6.283 rad/sec), and ζ = 0.05. Determine the response u(t) of this system to
p(t) defined by the half-cycle sine pulse force shown in Fig. E5.1 by (a) using piecewise
linear interpolation of p(t) with Δt = 0.1 sec, and (b) evaluating the theoretical solutions.

Solution

1. Initial calculations

e n t  0.9691  D  n 1   2 sin D t  0.5871 cos D t  0.8095

Figure E5.1 Half-cycle sine pulse force in Example 5.1

Substituting these in Table 5.1 gives

A  0.8129 B  0.09067 C  0.01236 D  0.006352

A  3.5795 B  0.7559 C   0.1709 D  0.1871

2. Apply the recurrence equations (Eq. 5.13 and 5.14). The results are summarized in
Table E5.1a and E5.1b.

3. Compute the theoretical response. Forced vibration solution can be computed by Eq.
(3.11) for response to harmonic force, which is valid for 0< t ≤ 0.6 sec. Then, Eq.
(2.17) free vibration of damped SDF system with initial displacement and velocity
equal to u  0.6  and u  0.6  from the end of forced vibration is used for t ≥ 0.6 sec.

Deformation at each time step ti are evaluated and shown in Table E5.1a. Then,
derivatives of Eq. (3.11) and Eq. (2.17) are also evaluated for each ti, and results are
given in Table E5.1b.

4. Check the accuracy of the numerical results. The numerical solution based on
piecewise linear interpolation of the excitation agrees reasonably well with the

5-6
 theoretical solution. The discrepancy arises because the half-cycle sine curve has been
replaced by the series of straight lines shown in Fig. E5.1. With a smaller Δt the
piecewise linear approximation would be closer to the half-cycle sine curve, and the
numerical solution would be more accurate.

Table E5.1a Numerical solution using linear interpolation of excitation

ti pi Cpi Dpi+1 B ui ui Aui ui Theoretical ui

0 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
0.1 5.0000 0.0618 0.0550 0.0848 0.9354 0.0258 0.0318 0.0328
0.2 8.6602 0.1070 0.0635 0.2782 3.0679 0.1849 0.2274 0.2332
0.3 10.0000 0.1236 0.0550 0.4403 4.8558 0.5150 0.6336 0.6487
0.4 8.6603 0.1070 0.0318 0.4290 4.7318 0.9218 1.1339 1.1605
0.5 5.0000 0.0618 0.0000 0.1753 1.9336 1.2109 1.4896 1.5241
0.6 0.0000 0.0000 0.0000 −0.2735 −3.0159 1.1771 1.4480 1.4814
0.7 0.0000 0.0000 0.0000 −0.6767 −7.4631 0.7346 0.9037 0.9245
0.8 0.0000 0.0000 0.0000 −0.8048 −8.8765 0.0471 0.0579 0.0593
0.9 0.0000 0.0000 0.0000 −0.6272 −6.9177 −0.6160 −0.7577 −0.7751
1 0.0000 -2.5171 -1.2432 −1.2718

Table E5.1b Numerical solution using linear interpolation of excitation

ti pi C pi Dpi 1 Aui ui Bui ui Theoretical ui

0 0.0000 0.0000 0.9354 0.0000 0.0000 0.0000 0.0000 0.0000

0.1 5.0000 0.8544 1.6201 −0.1137 0.0318 0.7071 0.9354 0.9567
0.2 8.6602 1.4799 1.8707 −0.8140 0.2274 2.3192 3.0679 3.1383
0.3 10.0000 1.7088 1.6201 −2.2679 0.6336 3.6708 4.8558 4.9674
0.4 8.6603 1.4799 0.9354 −4.0588 1.1339 3.5771 4.7318 4.8408
0.5 5.0000 0.8544 0.0000 −5.3320 1.4896 1.4617 1.9336 1.9783
0.6 0.0000 0.0000 0.0000 −5.1832 1.4480 −2.2799 −3.0159 −3.0848
0.7 0.0000 0.0000 0.0000 −3.2347 0.9037 −5.6418 −7.4631 −7.6346
0.8 0.0000 0.0000 0.0000 −0.2074 0.0579 −6.7103 −8.8765 −9.0808
0.9 0.0000 0.0000 0.0000 2.7124 −0.7577 −5.2295 −6.9177 −7.0771
1 0.0000 −1.2432 −2.5171 −2.5754

5-7
5.3 Central difference method

This method is based on a finite difference approximation of the time derivatives of

displacements, which are velocity and acceleration. Suppose ti is constant t . The central

difference expression for velocity and acceleration at time i are

ui 1  ui 1 ui 1  2ui  ui 1
ui  and ui  (5.15)
2t  t 
2

Substituting these in the equation of motion at time i , we get

ui 1  2ui  ui 1 ui 1  ui 1
m c  kui  pi (5.16)
 t  2 t
2

We assume that ui and ui 1 are known from previous steps. Transferring known quantities to

the right hand side, we get

 m c   m c   2m 
   ui 1  pi     ui 1   k  2 i
u (5.17)
  t  2t    t  2t    t  
2 2

or

ˆ
i 1  pi
ku ˆ (5.18)

where
m c
kˆi   (5.19)
 t  2t
2

 m c   2m 
pˆ i  pi     u   k  2 i
u (5.20)
  t  2t    t  
2 i 1

pˆ i
Then, the unknown ui 1 is then given by u i 1 
kˆ

Note that ui 1 is obtained without using equation of motion at time i  1 but from

equation of motion at time i (Eq. 5.16). And, ui 1 can be computed explicitly from the

known displacement ui and ui 1 . Such method is called an “explicit method.”

5-8
In the first step ( i  0 ), we need to use u1 to compute u1 , but we only have u0  u  0  and

u0  u  0  from the given initial conditions. So we need to find u1 by using finite difference

approximation of u0 and u0 :

u1  u1 u1  2u0  u1

u0  and u0  (5.21)
2t  t 
2

Using the first equation in Eq. (5.21) to eliminate u1 in the second equation in Eq. (5.21) and

arranging the terms to have only u1 on the left-hand side, we then have

 t 
2

u1  uo  t  u0   u0 (5.22)

2

Next, consider equation of motion at time step i  0

mu0  cu0  ku0  p0 (5.23)

We arrange the terms in Eq. (5.23) to have only u0 on the left-hand side and obtain

p0  cu0  ku0
u0  (5.24)
m

We can use Eqs. (5.24) and (5.22) for determining u1 by first computing u0 using Eq. (5.24)

from the given initial conditions u0  u  0  and u0  u  0  and then substituting u0 together

with u0 and u0 in Eq. (5.22) to obtain u1 . The whole procedure is summarized in Table 5.2.

5-9
Table 5.2 Central difference method

1. Initial calculations

p0  cu0  ku0
1.1 u0 
m
 t 
2

1.2 u1  u0  tu0  u0

2
m c
1.3 kˆ  
 t  2t
2

m c
1.4 a  
 t  2t
2

2m
1.5 b  k 
 t 
2

2. Calculations for time step i

2.1 pˆ i  pi  aui 1  bui

pˆ
2.2 ui 1  i
kˆ
ui 1  ui 1 ui 1  2ui  ui 1
2.3 If required: ui  ; ui 
2t  t 
2

3. Repetition for the next time step

Replace i by i+1 and repeat steps 2.1 through 2.3 to obtain ui 1 for the next time step.

This central difference method will give meaningless results, and become “unstable”, if the
time step is not short enough. The requirement for stability of this procedure is that

t 1
 (5.25)
Tn 

However, this requirement is never a constraint because the time step needs to be much
shorter, typically t / Tn  0.1 , to obtain acceptable accuracy of results. In analysis of

earthquake response, a time step about 0.005 to 0.02 sec is normally used to define and
record ground excitation.

5 - 10
Example 5.2

Solve Example 5.1 by the central difference method using t  0.1sec

Solution

1. Initial calculations

m  0.2533 kip-sec2 /in. k  10 kips/in. c  0.1592 u0  0 u0  0

p0  cu0  ku0
1.1 u0  0
m
 t 
2

1.2 u1  u0  tu0  u0  0

2
m c
1.3 kˆ    26.13
 t  2t
2

m c
1.4 a    24.53
 t  2t
2

2m
1.5 b  k   40.66
 t 
2

2. Calculations for each time step

2.1 pˆ i  pi  aui 1  bui  pi  24.53ui 1  40.66ui

pˆ i
2.2 ui 1 
26.13

3. Computation in Steps 2.1 and 2.2 are repeated for i = 0, 1, 2, 3, … leading to Table
E5.2, wherein the theoretical result (from Table E5.1a) is also included.

5 - 11
Table E5.2 Numerical solution by central difference method

ti pi ui 1 ui pˆ i ui 1 Theoretical ui 1

0 0.0000 0.0000 0.0000 0.0000 0.0000 0.0328

0.1 5.0000 0.0000 0.0000 5.0000 0.1914 0.2332
0.2 8.6600 0.0000 0.1914 16.4416 0.6293 0.6487
0.3 10.0000 0.1914 0.6293 30.8929 1.1825 1.1605
0.4 8.6600 0.6293 1.1825 41.2992 1.5808 1.5241
0.5 5.0000 1.1825 1.5808 40.2637 1.5411 1.4814
0.6 0.0000 1.5808 1.5411 23.8798 0.9140 0.9245
0.7 0.0000 1.5411 0.9140 -0.6463 -0.0247 0.0593
0.8 0.0000 0.9140 -0.0247 -23.4310 -0.8969 −0.7751
0.9 0.0000 -0.0247 -0.8969 -35.8591 -1.3726 −1.2718
1 0.0000 -0.8969 -1.3726 -33.8046 -1.2939 −1.2674

5.4 Newmark’s method

This method is developed by Nathan M. Newmark in 1959 based on the following equations:

ui 1  ui  1    t  ui   t  ui 1 (5.26)

ui 1  ui   t  ui   0.5    t   ui     t   ui 1

2 2
    (5.27)

The parameters  and  define the variation of acceleration over a time step and
determine the stability and accuracy characteristics of the method. Typical selection is  =0.5
and 1
6    14 .

Special cases

1. Average acceleration

If   1
2 and   1
4 are chosen, the above equations for ui 1 and ui 1 corresponds to

the special case that acceleration of the mass during the time step i is constant and equal to
the average of ui and ui 1 as shown in the left column of Table 5.3.

5 - 12
2. Linear acceleration

If   1
2 and   16 are chosen, the above equations for ui 1 and ui 1 corresponds to

the special case that acceleration of the mass during the time step i varies linearly between
ui and ui 1 as shown in the right column of Table 5.3.

Table 5.3 Average acceleration and linear acceleration methods of Newmark

Average Acceleration Linear Acceleration

1 
u     ui 1  ui  u    ui   ui 1  ui 
2 t
 2
u    ui   ui 1  ui  u    ui  ui   ui 1  ui 
2 2t
t t
ui 1  ui   ui 1  ui  ui 1  ui   ui 1  ui 
2 2
2 2 3
u    ui  ui   ui 1  ui  u    ui  ui  ui  ui 1  ui 

4 6t 2
 t 
2
21 1 
ui 1  ui  u t 
i  ui 1  ui  ui 1  ui  ui t   t   ui 1  ui 
4 6 3 

Time-Stepping Formula

Newmark’s method uses equilibrium equation at time i and time i  1 , which

involves response quantities at time ti 1 , i.e. ui 1 , ui 1 , and ui 1 . Such method is called an

“implicit method.”

Let us define the incremental form

ui  ui 1  ui ui  ui 1  ui ui  ui 1  ui pi  pi 1  pi (5.28)

5 - 13
From the basis equations of Newmark in Table 5.3,

ui  1    t  ui   t  ui 1   t  ui   t  ui (5.29)

and

ui   t  ui   0.5    t   ui     t   ui 1

2 2
   
 t 
2

  t  ui ui    t  ui

2
 (5.30)
2
Arrange terms in Eq. (5.30) to have ui on the left-hand side as

1 1 1
ui  ui  ui  ui (5.31)
  t  t 2
2

Then, substitute ui of Eq. (5.31) in Eq. (5.29) for ui

    
ui  ui  ui  t 1   ui (5.32)
t   2 

Substitute ui and ui [Eq. (5.32) and (5.31)] in the incremental form of equation of motion:

mui  cui  k ui  pi (5.33)

It can be written as kˆui  pˆ i (5.34)

pˆ i
We can compute ui  (5.35)
kˆ

 1
where k̂  k  c m (5.36)
t   t 
2

 1    1    
and pˆ i  pi   m c  ui   m  t   1 c  ui (5.37)
  t    2  2  

Once ui is known, ui , ui , ui 1 , ui 1 , and ui 1 can be computed.

ui 1  ui  ui ui 1  ui  ui ui 1  ui  ui (5.38)

Alternatively, ui 1 can be computed from

pi 1  cui 1  kui 1
ui 1  (5.39)
m

5 - 14
Table 5.4 Newmark’s method: Linear system

Special cases
(1) Average acceleration method (  1 2,   1 4)
(2) Linear acceleration method (  1 2,   1 6)
1. Initial calculations

p0  cu0  ku0
1.1 u0 
m
1.2 Select t
 1
1.3 kˆ  k  c m
t   t 
2

1  1   
1.4 a  m  c; and b  m  t   1 c
t  2  2 

2. Calculations for time step i

2.1 pˆ i  pi  aui  bui

pˆ
2.2 ui  i
kˆ
    
2.3 ui  ui  ui  t 1   ui
t   2 
1 1 1
2.4 ui  ui  ui  ui
  t  t 2
2

2.5 ui 1  ui  ui , ui 1  ui  ui , ui 1  ui  ui ,

3. Repetition for the next time step. Replace i by i+1 and repeat Steps 2.1 to 2.5 for the next time
step

Newmark’s method is stable if

t 1 1
 (5.40)
Tn  2   2 

For Average acceleration method

t
  12 and   14  (5.41)
Tn

5 - 15
 This implies that average acceleration method is stable for any t , although results
would not be accurate for large t . It is so called “unconditionally stable.” This quality can
be useful when the method is applied to higher modes which have very short natural period of
vibration.
For Linear acceleration method

t
  12 and   16  0.551 (5.42)
Tn

This requirement is not significant because a much smaller time step is required for accurate
representation of excitation and response.

Example 5.3

Solve Example 5.1 by the average acceleration method using t  0.1sec

Solution

1. Initial calculations

m  0.2533 k  10 c  0.1592 u0  0 u0  0 p0  0

p0  cu0  ku0
1.1 u0  0
m
1.2 t  0.1
2 4
1.3 kˆ  k  c  m  114.5
t  t 
2

4
1.4 a  m  2c  10.45; and b  2m  0.5066
t

2. Calculations for each time step

2.1 pˆ i  pi  aui  bui  pi  10.45ui  0.5066ui

pˆ pˆ i
2.2 ui  i 
kˆ 114.5
2
2.3 ui  ui  2ui  20ui  2ui
t
4
2 
2.4 ui  ui  tui   2ui  400  ui  0.1ui   2ui
 
 t
2.5 ui 1  ui  ui , ui 1  ui  ui , ui 1  ui  ui

3. Repetition for the next time step. Steps 2.1 and 2.5 are repeated for successive time
steps and are summarized in Table E5.3.

5 - 16
Table E5.3 Numerical solution by Newmark’s average acceleration method.

ti pi ui pi pˆ i ui ui ui ui ui Theoretical ui

0 0 0.000 5.000 5.000 0.044 0.873 17.467 0.000 0.000 0.000

0.1 5 17.467 3.660 21.635 0.189 2.032 5.713 0.873 0.044 0.0328
0.2 8.66 23.180 1.340 43.448 0.379 1.778 -10.807 2.906 0.233 0.2332
0.3 10 12.373 -1.340 53.870 0.470 0.043 -23.890 4.683 0.612 0.6487
0.4 8.66 -11.518 -3.660 39.893 0.348 -2.484 -26.643 4.726 1.083 1.1605
0.5 5 -38.161 -5.000 -0.902 -0.008 -4.642 -16.512 2.242 1.431 1.5241
0.6 0 -54.673 0.000 -52.774 -0.461 -4.419 20.972 -2.400 1.423 1.4814
0.7 0 -33.701 0.000 -88.326 -0.771 -1.791 31.579 -6.818 0.962 0.9245
0.8 0 -2.122 0.000 -91.046 -0.795 1.316 30.565 -8.609 0.191 0.0593
0.9 0 28.442 0.000 -61.810 -0.540 3.791 18.929 -7.293 -0.604 −0.7751
1 0 47.372 -3.503 -1.144 −1.2718

Example 5.4

Solve Example 5.1 by the linear acceleration method using t  0.1sec

Solution

1. Initial calculations

m  0.2533 k  10 c  0.1592 u0  0 u0  0 p0  0

p0  cu0  ku0
1.1 u0  0
m
1.2 t  0.1
3 6
1.3 kˆ  k  c  m  166.8
t  t 
2

6 t
1.4 a  m  3c  15.68; and b  3m  c  0.7679
t 2

2. Calculations for each time step

2.1 pˆ i  pi  aui  bui  pi  15.68ui  0.7679ui

pˆ pˆ i
2.2 ui  i 
kˆ 166.8
3 t
2.3 ui  ui  3ui  ui  30ui  3ui  0.05ui
t 2

5 - 17
 6
2.4 ui   ui  tui   3ui  600  ui  0.1ui   3ui
 t 
2

2.5 ui 1  ui  ui , ui 1  ui  ui , ui 1  ui  ui

3. Repetition for the next time step. Steps 2.1 and 2.5 are repeated for successive time
steps and are summarized in Table E5.4.

Table E5.4 Numerical solution by Newmark’s linear acceleration method.

ti pi ui pi pˆ i ui ui ui ui ui Theoretical ui

0 0 0.000 5.000 5.000 0.030 0.900 17.991 0.000 0.000 0.000

0.1 5 17.991 3.660 31.575 0.189 2.082 5.666 0.900 0.030 0.0328
0.2 8.66 23.656 1.340 66.247 0.397 1.790 -11.518 2.982 0.219 0.2332
0.3 10 12.138 -1.340 82.777 0.496 -0.030 -24.869 4.772 0.617 0.6487
0.4 8.66 -12.731 -3.660 60.897 0.365 -2.634 -27.212 4.742 1.113 1.1605
0.5 5 -39.942 -5.000 -2.622 -0.016 -4.799 -16.103 2.108 1.478 1.5241
0.6 0 -56.046 0.000 -85.220 -0.511 -4.456 22.975 -2.691 1.462 1.4814
0.7 0 -33.070 0.000 -137.424 -0.824 -1.629 33.559 -7.147 0.951 0.9245
0.8 0 0.488 0.000 -137.193 -0.823 1.622 31.461 -8.776 0.127 0.0593
0.9 0 31.950 0.000 -87.611 -0.525 4.103 18.164 -7.154 -0.695 −0.7751
1 0 50.113 -3.051 -1.221 −1.2718

5.5 Stability

Numerical procedures that give bounded results if time step is shorter than a certain
limit are called “conditionally stable.” Numerical procedures that give bounded results
regardless of time step size, no matter how large, are called “unconditionally stable.”

Stability of the method is important for multi-degree-of-freedom system where an

unconditionally stable method is sometimes necessary because the natural period of vibration
of high modes are very short and instability could happen with the response of higher modes.

5 - 18
Computational error

Error is inherent in any numerical method both from round-off error and
approximation of solution.

Let us consider solutions of free vibration using different procedures discussed

earlier; t  0.1Tn ; and compare to the exact analytical solution.

Figure 5.3 Free vibration solution by for numerical methods ( t  0.1Tn ) and

the theoretical solution (Source: Chopra, 2012)

All numerical methods give results that have amplitude decay, implying that these
procedures introduce numerical damping. Most methods make the period of vibration longer
except the central difference method, which gives result that has shorter period than the exact
result. Period shortening in the central difference method is highly significant when t / Tn is

close to its stability limit 1/  .

5 - 19
 Figure 5.4 (a) Amplitude decay versus t / Tn ; (b) definition of AD and PE;

(c) period elongation versus t / Tn (Source: Chopra, 2012)

The linear acceleration Newmark’s method seems to be most accurate with regard to
least period elongation error for these methods considered for linear SDF system. The choice
of methods would be different for MDF system or nonlinear response analysis.

The choice of time step also depends on the time variation of the dynamic excitation
and natural period of the system. t  0.1Tn gives reasonably accurate results, but it also has

to be small enough to avoid distortion of the excitation function. For earthquake excitation
t is usually less than 0.02 sec.

A useful technique for selecting the time step is to solve the problem with a time step
that seems reasonable and re-solve the problem with a small time step. The time step is
deemed small enough if results from two analyses are essentially the same, otherwise reduce
the time step and repeat such comparison until two successive solutions are close enough.

5 - 20
Exercises

1. The SDF frame system has the following properties: m  35 ton ; k  1400 kN/m ; and
damping   0.05 . Compute displacement at the top of the frame subjected to the blast
loading history shown in Fig. P5.1 with t  0.1sec by using

a. Linear interpolation of excitation

b. Central difference method
c. Average acceleration method
d. Linear acceleration method

m p (t) p (kN)
45 45
c 30
k (total)
20
5
t (sec)
6 @ 0.1=0.6 sec
(a) (b)
Figure P5.1 Structure and force function for Problems 1 to 3

2. The damping of SDF system in Problem 1 is now modified to   0.2 . Determine the
response of SDF system in Problem 1 by central difference method. Plot the response as a
function of time, compare with the solution of Problem 1, and comment on how damping
affects the peak response.

3. Repeat Problem 1 with t  0.05sec . How does time step size affect the accuracy of the
solution?

5 - 21