Appendix/Numerical evaluation of

response spectrum'

A.l Linear elastic systems

As explained in Chapter 1, the elastic response spectrum is obtained by integrating
the following equation of motion for a single-degree-of-freedom system:

in which all the variables are defined in Chapter 1, except the ground acceleration
history, which is now denoted by a ( t ) .A number of efficient methods are available for
integrating Equation A.I. An exact technique was developed by W.D. Iwan in an
unpublished study at the California Institute of Technology and was later reported by
Nigam and Jennings[2]. In this method, Equation A.l is solved analytically within
each successive time step assuming the ground acceleration varies linearly between
designated points.

where the ground acceleration, a ( t ) , has been replaced by its piecewise-linear

approximation. The solutions for the relative displacement, u , and velocity, u, are
U(t) =

[C, cos oD( t - t i )

e-F~(~-~i)

21; Aai
(t - t i ) + - - - - ,
o2At,
o3 At,
I Aa,

+ C2sin wD( t - t,)]

ai

- A -

o2

1 Aa,

- (C,
oD + 1;oC2)sin oD( t - ti)] - --.
o' At,

'

In these expressions oDis the damped circular natural frequency, o, = o J( I - 1; ) ,

and C, and C2 are constants. These constants are evaluated by defining
u ( f = t i ) = U , , u(t

ti)= u,.

Thus, C, and C2 are

c, = U i +

21; Aa,
- --

-2

o2 o3At, '

*Based on Nau and Hall [I].

u , + ~ o u , + -1;a , + - -1 -2C2 Aa,

0
o2 At,

(AS)

The relative displacement and velocity at the end of the time step, ui+!and ir,+, , may
be determined by substituting Equation A S into Equation A.3 and setting t = ti+,.
The resulting recursion formulae for u,+, and u,+, may be conveniently expressed in
matrix form as,
ui+I

+ B(5, 0 ,At,)

o ,At,)

= A([,

[u,+l}

(01

a;+I

},

where
A([, o , At,) = ['I'

B(4, o ,At,) =

a21 a22

bll b12
b21 b22

].

The elements of matrices A and B are functions of c , w and At, and are given by Nigam
and Jennings[2]. After simplifying elements b2,and b2*,the coefficients ofA and Bare
a,, = e

-<mAr,

cos wDAI, +

1;
4 1 - C2)

e-50ati

a I 2= -sin o, At,,
OD

a2]= a22= e

o
J(1

- c2)

e-w'l sin O D~ t, ,

-gwAr, cos oDAll -

c
J(1 - C 2 )

and
6I I

[ (e5 )
+

e-cwA'~

W'A~,

o~

OD

(o325At, + $)cos oD

25
-,
o3Ati

12

'

02At,

+-0321;A ti

sin o D A t i
WD

If the record is digitized at equal time intervals, the coefficients of A and B are constant for a given frequency. Hence, given the initial conditions for the single-degreeof-freedom system, usually u(0) = u(0) 0,response computations proceed rapidly
by applying the recursion relationships defined by Equation A.6. Monitoring the
response quantities as computation proceeds enables the determination of the
maximum relative displacement, i.e. the spectral displacement. The calculations are
repeated for a family of frequencies for each selected damping value. Thereby an
entire set of elastic response spectra is developed for the given earthquake record.
The procedure described above can, of course, be applied to accelerograms
digitized at unequal time intervals. However, the evaluation of matrices A and B at

N U M E R I C A L E V A L U A T I O N O F R E S P O N S E S P E C T R U M / 155

each step of integration, i.e. for each Ati, increases the computation time considerably. Experience has shown that this increase in computation time may be 100% or
more. To maintain computational efficiency for records digitized at unequal time
intervals, Nigam and Jennings[Z] recommend an approximate method involving
time coordinate rounding. However, with the development of uniform processing
and correction procedures, records are routinely digitized at equal time steps of 0.0 1
or 0.02 sec. Hence, it is unnecessary, insofar as the discussion here is concerned, to
consider the treatment of records digitized at unequal time intervals.
The time step used in the response computations is selected as the smaller of the
digitized interval of the earthquake accelerogram or some fraction of the period of
free vibration, for example TjlO. For systems whose natural period governs the
selection of Ati, i.e. for high frequencies, At, must be chosen so that an integral
number of time steps comprises the digitized interval of the accelerogram. This
restriction on At, preserves uniform time intervals and guarantees that response
quantities will be computed at times corresponding to those of the given earthquake
record. For example, suppose that the response of a system with T = 0.12 sec is to be
determined. In addition, assume that the earthquake accelerogram is digitized at
intervals of 0.02 sec. If the time step is not to exceed, say, T/10 or the digitized
interval, At, must be selected as 0.01 sec, providing two time steps between successive
digitized values of acceleration.
Aside from the uncertainties associated with the recording and processing of the
accelerogram itself, errors in spectral calculations result from approximations
employed in the numerical integration technique used for response computation. In
this sense, the method described herein is exact. However, error is introduced by discretization. That is, the true maximum displacement or velocity, i.e. the spectral
quantities, will not, in general, occur at one of the discrete times at which
computations are made. The maximum error results when the true maximum falls
midway between two consecutive time points, as depicted in Figure A.1. If the

Fig. A.l The true maximum and the computed values.

response within the time step is approximated by a sinusoid of frequency equal to the
natural frequency of the single-degree-of-freedom system[2], the maximum error is
maximum error, % =

(A. 10)

The true spectral quantities are greater than those computed at the discrete time
points. By appropriately selecting the time step, however, the maximum error in the
spectral ordinates may be controlled. For example, the expression above gives 4.9%
error for At, = TIIO, 1.2% for T/20, and 0.3% for T/40. Thus, a time step
corresponding to At, = TI20 is generally adequate.

A.2 Bilinear hysteretic systems[3]

The bilinear hysteretic load-deformation model is shown in Figure A.2. In this figure,
u, represents the initial yield level; u, and u,, are the current positive and negative
yield levels; s, the current set remaining after an excursion of yielding; k, the initial
elastic and unloading stiffness; and a, the ratio of the strain-hardening stiffness to the
elastic stiffness. Initially, of course, s = 0, u, = u,, and u,, = -u,. Note that
kinematic hardening for the bilinear system is shown, in which the current positive
and negative yield levels are separated by a region of linearly elastic deformation of
magnitude 224,.

Fig. A.2 Bilinear hysteretic load-deformation model.

N U M E R I C A L E V A L U A T I O N O F RESPONSE S P E C T R U M / 1 5 7

consider first the linear elastic response which follows unloading. For this case, the
equation of motion for ti s t S ti+,is

ii + 26o1i+ w2(u- s) = -ai

Aa,
-(t - r,),
A ti

(A.1 1 )

where all symbols are as previously defined. This equation may be more conveniently
expressed as

ii

+ 2@u+

0 2 u = -a',--([

Aii;
At,

- t,),

(A.12)

where
a',= a ; - 02s,

= a,,,

- w2s.

(A. 13)

The notation Aa', in Equation A. 12 is used for convenience since AZ,% hai.The solution for Equation A. 12 is given by Equation A.6 with the substitution of a', and a',+, for
a, and ai+t ,
(A.14)
in which the coefficients of matrices A and B are defined by Equations A.8 and A.9.
The set s required in Equation A.13 is computed a t the instant of unloading.
Following an excursion of positive yielding, the set is given by s = ( 1 - a) (uunl- u,);
following an excursion of negative yielding, s = (1 - a) (u,,,
u,). In these
equations, u,,, is the relative displacement computed at the instant of unloading. At
the same time, the current yield levels are updated. For example, following a positive
yield excursion, uyp= uunland uyn= u,,, - 2uy.
Now consider excursions of loading beyond the current yield levels for the bilinear
system. With reference to Figure A.2, the equation of motion for relative displacements greater than the current positive yield level uypis

ii

+ 26wu + 02(u, - s) + am2( u - u,)

-a,

Aa,

-(t - t,).
At,

(A.1 5 )

This differential equation applies for u > u,, until unloading is detected, when the
product ui X u,, < 0.Simplifying Equation A. 15 gives

ii

+ 2 ~ , o z l+i o:u = -a',

A&
At;

- -(t

- ti),

(A.16)

in which
(A.17)
and
Z i i 3 a,

+ w Z u y ( l- a ) ,

a,,,

+ o 2 u Y ( 1- a).

(A. 18)

Note that C,, and o,, equivalent properties associated with the strain-hardening
branch of the force-deformation model, are defined only for a > 0. For an extursion
of negative yielding, for u < u,,, Equation A.16 applies with the modification,
a',= a , - w2u,(1 - a ) ,

Zi+, = a,,,

- o 2 u Y ( l- a ) .

(A. 19)

The character of the solution of Equation A.16 may be underdamped (c2 c l),
critically damped (C, = I), or overdamped (4, > 1). However, for the majority of
bilinear systems of practical interest, the response is underdamped. For example,
when C, = 0.05 and a = 0.02, 0.05 and 0.10, the largest value of is 0.051J0.02 or
0.35. Thus, the solution as expressed by Equation A. 14 holds with the substitution of
and o2for C, and o in the elements of A and B given in Equation A.8 and A.9.

c,

c2

A.3 Elastoplastic systems [3]

The discussion regarding the linear elastic portions of the response for the bilinear
system also applies to the elastoplastic system. For yielding excursions, however, the
equation of motion for the elastoplastic system is
ii

AZi
+ 2cwu = - a', - ( I - ti)
Ati

where ii, and a',,, are computed with a = 0, in accordance with either Equation A. 18
for positive yielding or Equation A.19 for negative yielding. The solution for
Equation A.20 may also be expressed by Equation A. 14 in which the elements of
matrices A and B are:

a2, = 0, a,,

e-2Cma1,

(A.21)

For the special case of no viscous damping (6 = O), the coefficients of A and B are: