You are on page 1of 14

Bulletin o f the Sei smol ogi cal Soci et y o f Amer i ca. Vol . 59, No. 2, pp. 9 0 9 - 9 2 2 .

Apr i l , 1 9 6 9
CALCULATI ON OF RESPONSE SPECTRA FROM
STRONG- 5~OTI ON EARTHQUAKE RECORDS
BY NAVIN C. NIGA~ AND PAUL C. JENNINGS
ABSTRACT
A numerical method f or computing response spectra from strong-motion earthquake
records is devel oped, based on the exact solution to the governing di fferenti al
equation. The method gives a three to four-fol d saving in computing time compared
to a third order Runge-Kutta method of comparable accuracy. An analysis also i s
made of the errors introduced at various stages in the calculation of spectra so that
al l owabl e errors can be prescribed f or the numerical integration. Using the pro-
posed method of computing or other methods of comparable accuracy, exampl e
calculations show that the errors introduced by the numerical procedures are much
l e s s than the errors inherent in the di gi ti zati on of the acceleration record.
INTRODUCTION
Since their introduction (Benioff, 1934; Blot, 1941; Housner, 1941), response spectra
of strong-motion earthquakes have proved useful and informative in problems of
design and analysis of structures subj ect ed to strong earthquake motions. The spectra,
calculated from the recorded ground acceleration, are plots of the maximum response
to the eart hquake of a simple oscillator over a range of values of its natural period and
damping. These curves provide a description of t he frequency characteristics of t he
ground motion and give t he maximum response of simple structures to t he earthquake.
By superposition of different modes of response, spectrum techniques can be applied
~o the design and analysis of complex structures such as buildings and dams. Used in
this manner the spectrum technique represents an approach intermediate between a
design based on static loads and a complete integration of the equations of motion of
t he complex structure.
Strong-motion eart hquake records have been obtained infrequently in t he past and
the reduction t o digital form, or equivalent analog form, and subsequent calculation of
spectra have been performed on more or less an individual basis. However, in recent
years t he number of strong-motion instruments in the seismic regions of the world,
particularly California, Mexico and Japan, has increased t o t he point where a
major eart hquake in these areas will generate a large number of records. The potential
volume of t he dat a and the development of the tape-recording accelerograph indicate
clearly t hat rapid and aut omat ed dat a processing and spectrum c~lculation procedures
are needed. I n an effort to fulfill part of this need, this paper presents a rapid, accurate
met hod for comput at i on of response spectra from strong-motion eart hquake records.
Response spectra were first obtained by Bl ot (1941), using a direct mechanical
analog and later by Housner and MeCann (1949) using electric analog techniques. The
availability of digital computers and a progressive increase in t he speed of digital
comput at i on in recent years has led to an increasing use of digital computers in the
calculation of spectra. The extensive application of the response spectrum to problems
in eart hquake engineering has sustained interest in methods of calculating spectra and
has raised questions regarding accuracy, reproducibility and economy in such calcula-
tions (Hudson, 1962; Brady, 1966; Berg, 1963; Schiff and Bogdanoff, 1967).
The digital comput at i on of spectra requires the repeated numerical solution of the
9O9
910 :BULLETIN OF THE SEISMOLOGICAL SOCIETY OF AMERICA
response of a simple oscillator to a component of recorded ground acceleration. The
motion of the oscillator is described by a second order, linear, inhomogeneous differen-
tial equation, and if a digital description of the earthquake record is available, the
response can be obtained by numerical integration. Several numerical integration tech-
niques have been used for calculation of spectra, for example, the third order Runge-
Kut t a scheme of integration has been preferred by some investigators because of its
accuracy, long-range stability, self-starting feature and because it can be adapted
easily for use when the excitation is not defined at regular intervals (Jennings, 1963;
Hildebrand, 1956). The truncation error in this method is proportional to (A0) 4, where
A0 is the normalized interval of integration. Thus by a suitable choice of A0 the calcu-
lations can be performed to an acceptable degree of accuracy.
An alternative approach to calculation of spectra is based on obtaining the exact
solution to the governing differential equation for the successive linear segments of the
excitation, then using this solution to compute the response at discrete time intervals
in a purely arithmetical way (Hudson, 1962; Iwan, 1960). This method does not intro-
duce numerical approximations in the integration other than those inherent in round-
off, so in this sense it is an exact method.
In this p~per this method is uss:l to develop a computation technique which leads
to a three-to four-fold s~ving in computing time compared to a third order Runge-
Kut t a methoJ of comparable accurgcy. If the egrthqu~ke record is digitized at equal
time intervals, the proposed scheme gives spectral values which are exact in the sense
mentioned above. If the record is digitized at arbitrary time intervals, however, it is
necessary to introduce an approximation into the digitization. An analysis of the
errors introduced at various stages in the preparation of response spectra and the
results of digitization experiments show that the additional numerical approximations
are not detrimental.
The numerical techniques for calculating response spectra have been codified into
computer programs in Fortran IV; listings of these programs and details not included
herein are available in a recent report (Nigam and Jennings, 1968).
FORMULATION OF THE ]-\/[ETHOD
Spectra are defined by the maximum response of a simple oscillator subjected to
base acceleration a ( t ) as shown in Figure 1. The equation of motion of the oscillator is
2 -t- 2 f l ~ 2 + o~x = - - a ( t ) (1)
in which fl = the fraction of critical damping and ~o = the natural frequency of
vibrations of the oscillator.
Assuming t hat a ( t ) may be approximated by a segmentally linear function as
shown in Figure 2, equation (1) may be written as
A a ~ ( t - - t ~ ) ; t~ _-< t ~ ti+l (2)
with
A t l = ti+l -- tl
A a ~ = a ~ + l - - a . ( 3 )
CALCULATION OF RESPONSE SPECTRA
The solution of equation (2), for t~ < t < t~+~ is given by
X = e - f l w ( t - t i ) [ C ' l s i n 50 ~ / 1 - ~ ( t - - t l ) - ~ C2 c o s 50 ~ , / ~ ( t - - t,/)]
as 2,3 A a l 1 Aa ~ ( t - - t~)
w 2 -~- w 3 A t i w 2 A t i
911
(4)
I
m
k ? k
2 2
i - - - l - - - - a ( t )
0 0 0
/ / / ' / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / /
FI G. 1. A s i mp l e o s c i l l a t o r .
a ( t )
Oi+l
a i
o i - I
I I
i I
I I
t i - I ti ]i+l
FI G. 2. I d e a l i z e d b a s e a c c e l e r a t i o n .
in which C1 and C~ are constants of integration. Setting x = x ~ a n d 2 = 2 a t t = t ~
and solving for C1 and C2, it is found t hat
1 ( 2/~2-- 1Aal /3 / (5a)
2/3 Aa~ _f a~ (5b)
C2 = x~ ~3 Atl ~ "
Substitution of these values of C1 and C2 into equation (4) will show t hat x and 2
at t = t~+l are given by
XI+I = A ( ~ , 50, A t l ) x l "~- B ( ~ , 50, A ~ i ) d i ( 6 a )
912 BULLETIN OF THE SEISMOLOGICAL SOCI ETY OF AMERICA
i n wh i c h
21 ai+~
A _ _ | - a 2 l f . a l l a12 B = I bl l b12] .
a.~,._ 1 [.b21 b22.J
T h e e l e me n t s of ma t r i c e s A a n d B a r e g i v e n b y
a l l = . V / ~ s i n co %/ 1 - - #2 A t l "Jr- COS co % / 1 - - , ~ At i
e-#a,A t i
a t , - s i n co %/ 1 - #2 Z~h
co %/ 1 - , 2
a,1 = - - % / ~
e - # ~ s i n co %/ 1 - , 2 At,:
/
e - ~ , o A t i
| % /1 , ' A t,:
a22 = \ e s co -- %/ 1 - - , 2
b:: = e-'~""" Lkf(2" -- ico,At,: + ~ ) s i n . c o %/1%'/~ ---- ,2 '-? A t i
# __ si n co %/ 1 - , 2 At,:)
( 2 , 1 ) ] 2 #
+ ~ + ~ cos co %/i - # ~t,: co'~t~
b12 = - - e- #~V ( 2" - - ~l . ) sh : co% / 1- ' 2A t* 2fl ]
L \ co,At~ ~ - ~ i . ~ + ~ cos co %/i - , . zxt~
1 2 #
--co~ -I- co3Ah
L \ co2~t~ + e os ~o - z~ti % / 1 - . ~
( " 5)
_ ~ q_ (co V/ 1 _ #2 si n co %./[ - - , 2 Ati q- ,co cos co %./1 - - , ' At,:)
1
+ L
co2At,:
~ ,2 2 = - ~ - ~ ' ~ r o 2 1 2 . _ - _
L ~ , ~ t , :
1 ( c o s co %/ 1 - #2 At,:
X
s i n co %/ 1 - f12 At,:}
%/ 1 - #2
/
7
_ 2 # (co %/ 1 - - , ~ s i n co %/ 1 - - #2 Ah q- , w cos co %/ 1 - - #2 A h ) |
co3Ah
/
i
- - ~2At i "
( 6b)
(6c)
( 6 d )
( 6 e )
CALCULATI ON OF RESPONSE SPECTRA 913
From equation (2), it follows t hat the absolute acceleration, he, of the mass at time
t~ is given by
2
2i = 2i zr ai = - - ( 2 f l o ~ 2 i - ~ - o ~ x i ) . (7)
Hence, if the displacement and velocity of the oscillator are known at some time
to, the state of the oscillator at all subsequent times t~ can be computed exactly
by a step-by-step application of equations (6) and (7). The computational ad-
vantage of this approach lies in the fact t hat A and B depend only on 5, ~ and Ate.
fl and co are constant during the calculation of each spectrum value, and if At~ is
constant also, x~, 2~ and ~ can be evaluated by the execution of only ten multi-
plication operations for each step of integration. Matrices A and B, defined by the
rather complicated expressions equations (6d) and (6e) need to be evaluated only at
the beginning of each spectrum calculation.
If varying time intervals are used, it is necessary, in general, to compute A and B
at each step of integration. However, by rounding the time coordinates of the record,
as discussed below, the number of these matrices needed during the calculation
can be reduced to only a few. These, too can be computed at the beginning of the
calculation and called upon when needed, thereby saving computing time.
COMPUTATI ON OF SPECTRA
To construct the response spectra, it is necessary to find the maximum values of
the displacement, velocity and acceleration during a given excitation. This is done
by computing the response at discrete time intervals through equations (6) and (7) and
monitoring the response parameters to retain the maximum values. Thus, the
response spectra are given by
S ~ ( ~ , 5 ) = M a x ~( ~, #)] ( 8 a )
i=I,N
&(~, ~) = M a x [ 2~( ~, #) ] (8b)
I=I, N
&(~, ~) = M a x [~(~, #)] ( s o )
~=I,N
in which S~, S, and S~ are the spectral values of displacement, velocity, and accelera-
tion respectively, for selected values of damping and natural frequency; and N is the
total number of discrete points at which the response is obtained.
This process of obtaining the maximum response is approximate because the re-
sponse is found only at discrete points, whereas the true maxima may occur between
such points. This error, called the error of discretization, is inherent in all numerical
procedures, but can be bounded within acceptable limits by a suitable choice of the
time interval. The discretization error operates to give spectrum values lower than
the true values and the error will be a maximum if the maximum response occurs
exactly midway between two discrete points as shown in Figure 3. An estimate for
the upper bound of this error can be found by noting that at the time of maximum
displacement or velocity, the response of the oscillator is nearly sinusoidal at a fre-
quency near its natural frequency. Under this assumption the error can be related to
the maximum interval of integration (Ar),~ and the period of the oscillator as illus-
trated in Figure 3.
914 BULLETI N OF THE SEISMOLOGICAL SOCIETY OF AMERICA
Digitization of Earthquake Records. When eart hquake records are digitized at equal
time intervals, t he acceleration values are measured at regular time steps and are
assumed to be connected by straight line segments. If t he interval of digitization, At,
is sufficiently small, this procedure approximates t he actual earthquake record quite
closely. The choice of t he interval of digitization depends upon t he period range of
interest and the nat ure of t he earthquake record; commonly used values range from
0.01 to 0.04 seconds. Analytical methods for assessing the error induced by such a
sampling process are available (Schiff and Bogdanoff, 1967, Bendat and Piersol, 1966).
Digitization at equal time intervals will be a virtual necessity for efficient processing
of accelerograms recorded on magnetic tape, and in other instances where analog to
digital conversion is made automatically.
For records digitized at unequal time intervals the abscissae and ordinates of the
points where changes of slope are judged to occur are measured and t he points are
again assumed to be connected by straight line segments (Figure 2). This procedure
leads to variable time intervals between successive points. The accuracy of this ap-
( AT ) m M a x i m u m E r r o r
( p e r c e n t ) - o
X
--< T / 1 0 --< 4 . 9
T / Z 0 ~ I Z
--< T / 4 0 ! 0 3
j I
3-
FIG. 3. Er r or due t o di s er et i zat i on.
proach is difficult to determine analytically, but should be considerably greater t han
would be achieved by t he same number of sample points at equal intervals. This
method is t he most convenient for manually controlled digitization techniques.
Spectra Computation for Equal Time Intervals. Equat i ons (6) show t hat matrices
A and B are functions of ~0, ~ and At+. If At~ is constant, t hat is, the earthquake record
is digitized at equal time intervals, these matrices need to be computed only once for
each pair of co and ~ and the chosen value of At, t he interval of integration. Also, since
t he method does not involve truncation error, it is possible to use larger intervals of
integration t han in other methods. The choice of t he interval of integration, At, is
controlled by the interval of digitization (At =< At) and the error of diseretization.
Figure 3 shows t hat the error due to discretization is less t han 1.2 per cent
if Ar ~ T/20 (T = Nat ural Period). Considering other errors in t he computation of
spectra, this choice of interval of integration seems reasonable. In fact, AT =< T/IO
may be satisfactory for most purposes.
Since the third order Runge-Kut t a scheme of integration often has been used for t he
digital computation of spectra, the proposed method is compared to it wi t h regard t o
accuracy and computing time. To make the comparison, undamped velocity spectrum
values were obtained for artificial earthquake No. 7 (Jennings, 1963) using a t hi rd
order Runge-Kut t a scheme, for four values of the interval of integration: Ar =< T/IO,
T/20, T/40, and T/80. The spectrum values also were obtained by t he present met hod
for Ar _--< T/ 10 and T/20. The results of these computations are compiled in Table 1,
which shows t he spectral values and t he relative values of t he computation time for
different cases.
CALCULATION OF RESPONSE SPECTRA 915
The dat a i n Ta b l e 1 show t h a t i t is necessar y t o use Ar < T/80 t o get accur acy t o
t hr ee si gni fi cant fi gures usi ng t he t hi r d order Runge - Kut t a scheme of i nt egr at i on.
For t he exact s cheme of i nt egr at i on, however, Ar =< T/IO ma y be accept abl e for mos t
pr act i cal pur poses. I t is seen also f r om t he l ast t wo col umns of Tabl e 1 t ha t in some
cases t he spect r al val ues are t he s ame for mr _-< T/IO and Ar -<_ T/20. Thi s occurs
when t he ma x i mu m val ue has a t i me coor di nat e whi ch is a c ommon mul t i pl e of bot h
st ep l engt hs.
The r el at i ve t i mes gi ven in Tabl e 1 i ndi cat e t ha t for accur acy t o t hr ee significant
figures t he pr oposed me t hod wi t h AT --< T/20 is about 3 t o 4 t i mes as f ast as t he t hi r d
TABLE 1
~.TNDAMPED VELOCITY SPECTRA AND COMPUTING Wi~IES FOR P~ECORDS DIGITIZED
AT EQUAL TIME ~NTERVALS
(Artificial Earthquake No. 7, At = 0.025 sees)
Sv In Ins/Sec
Period In Secs third order Runge-Kutca exact
AT ~ T/IO Ar ~ T/20 Ar ~ T/40 AT ~ T/80 Ar ~ T/tO AT ~ T/20
0.05 0.126427 0. 151835 0. 167647 0. 204386 0.213610
0.1 1.175386 1. 700338 1. 938206 2. 000499 2.021378
0.2 5.892718 6. 974802 7. 951382 8. 077665 8.129632
0.3 8.149913 8. 262451 7. 951037 7. 951037 7. 927181 7.99255a
0.4 16. 399755 21. 987390 22. 985190 23. 097921 23. 174661 23.174610
0.5 20. 716002 20.716002 22. 897381 23. 164824 23. 352803 23.352803
0.7 25. 625738 25.625738 26. 732634 26. 861891 26. 966020 26.966020
0.9 14.911507 14. 911507 15. 019002 15. 024426 15. 059319 15.059319
1.0 13. 712175 13. 712175 13. 712175 13. 809191 13. 841665 13.841665
1.6 23. 740771 23. 740771 23. 740771 23. 757607 23.772265 23.772265
2.0 13. 137178 13. 137178 13. 137178 13. 137178 13. 148899 13.148899
0.213611
2.021378
8.142673
Relative com-
putation
time
1. 5 2 3 4 . 75 1
Note: Spectral values below dotted lines were obtained at Ar = 0.025 sees, the interval of
digitization.
or der Ru n g e - Ku t t a t echni que. Usi ng t he I BM 7094 Comput e r at t he Cal i f or ni a
I ns t i t ut e of Technol ogy Comput i ng Cent er , t he comput i ng t i me for t he pr oposed
me t hod wi t h Ar -< T/20 aver ages about 1.0 see execut i on t i me per s pect r um poi nt for
t he 30 see record used t o compi l e Tabl e 1.
Spectra Computation for Unequal Time Intervals. I f t he ear t hquake record is di gi t i zed
at unequal t i me i nt er val s, mat r i ces A (o, ~, Ate) and B(co, ~, Atg) in equat i ons (6)
will, i n general , change f r om st ep t o st ep, maki ng i t necessar y t o comput e t he m anew
at each st ep of i nt egr at i on. Fr om equat i ons (Cd) and (Ce), whi ch define t he el ement s
of t hese mat r i ces, i t is cl ear t h a t t hi s comput at i on woul d t ake a l arge a mount of t i me;
i n fact , i t woul d ma ke t he met hod slower t ha n a compar abl e t hi r d order Ru n g e - Ku t t a
scheme. However , t he i nher ent l i mi t at i ons of t he di gi t i zed ear t hquake record ma ke i t
possi bl e t o devel op an appr oxi mat e met hod of sat i sf act or y accur acy whi ch gi ves a
s avi ng i n comput i ng t i me. The essent i al f eat ur e of t hi s met hod is a r oundi ng of t he
t i me coor di nat es of t he ori gi nal record t o a pr edet er mi ned accur acy. Then, choosi ng an
916 BULLETI N OF THE SEISMOLOGICAL SOCIETY OF AMERICA
appropriate maximum interval of integration, (At)m, and subdividing each time
interval of the rounded time record into time intervals equal to or less than ( At ) ~,
the calculation proeeeds much as before, except t hat A(~o, 8, Ar~.) and B(o, f~, Ar~')
must be calculated for four or five different values of Arj. This procedure is illustrated
in Table 2.
After the time record is rounded and subdivided, the values of the ground aeeelera-
tion at the additional points created by subdivision are computed by linear interpola-
tion from the original record. The modified record now consists of a set of points along
the time axis, spaced at a limited number of known intervals, and the corresponding
values of the acceleration. The operations necessary to produee the modified record
can easily be programmed for computer execution.
The number and size of the time intervals in the modified record depend on the way
TABLE 2
ROUND-OFF AND SUBDIVISION OF TIME RECORD
[(~)~ = 0.04 secs]
Subdivision into Intervals
Step Original Time Rounded Time of Integration Remarks
(Ar)m = 0.04
t~ 10.4267 10.43 0.04 Round-off to . 01
ti+l 10. 5213 10.52 0.04
At~ 0.0946 0.09 0.01
II
0.09
( a ) ( b )
t~ 10. 4267 10. 425 0.04 0.04
ti+, 10. 5213 10. 520 0.04 0.04
Atl 0.0946 0.095 0.01 0.015
0. 005 0. 095
0.095 0.095
Round-off to .005
in which rounding of the time coordinates is carried out and the choice of ( kr ) ~. For
the example presented in Table 2, the possibilities are:
I Arj = 0.04, 0.03, 0.02 and 0.01 (4 time-intervals)
IIa Arj = 0.04, 0.03, 0.02, 0.01 and 0.005 (5 time-intervals)
IIb Arj = 0.04, 0.035, 0.03, 0.025, 0.02, 0.015, 0.01, 0.005 (8 time-intervals).
Therefore, if the original record is modified as indicated above, the exact method needs
only a limited number of matrices for each pair of o~ and fl, and these can be computed
before the integration is started. By indexing each possible Arj and the corresponding
matrices, the appropriate matrices can be cMled at each step of integration.
The procedure outlined above requires t hat the time coordinates of the record be
rounded to a predetermined decimal fraction. Assuming t hat the original digital
description of the earthquake record is exact, this is an approximatioI/which must
lead to random errors in the computation of spectra. However, a digital description of
an earthquake at unequal time intervals is obtained by manual, or manually con-
trolled, reading of a graphical record and is subject to certain unavoidable errors. If
the round-off is carried out in such a way that effects of the approximation so introduced
CALCULATI ON OF RESPONSE SPECTRA 917
ar e well wi t hi n t he errors i nher ent in t he process of di gi t i zat i on, t he process out l i ned
above will be accept abl e.
Digitization Experiment. Ber g (1963) and Br a dy (1966) have st udi ed t he errors
i nt r oduced dur i ng t he process of di gi t i zat i on and t hei r effect on t he comput at i on of
spect r a. I n t he absence of gross personal errors or bias, t he errors arise pr i mar i l y f r om
resol ut i on of t he scaling devi ce and f r om t he t hi ckness of t he t r ace whi ch makes t he
choice of poi nt s at whi ch a change of slope occurs somewhat ar bi t r ar y. To exami ne
t hes e errors an exper i ment on di gi t i zat i on was conduct ed by t hr ee persons ( X, Y, Z)
on t he Benson- Lehner Da t a Reducer 099, at t he Cal i forni a I ns t i t ut e of Technol ogy.
Th e first 5 seconds on t he Taf t , Cal i forni a N21E record were di gi t i zed i ndependent l y
by X (5 t i mes) , Y (2 t i mes) and Z (2 t i mes) , at t he hi ghest resol ut i on of t he Da t a
Reducer for t he record used (0.0015 secs and 0.0001 g). I n each case poi nt s were t aken
whe n a change in slope was j udged t o occur i n t he record. Taki ng one record of each
per son as a st andar d, correspondi ng t i me coordi nat es on t he ot her records were sub-
t r act ed f r om i t t o obt ai n what is called her e t he self-error of digitization. The mean and
TABLE 3
SELF- ERRoR OF DIGITIZ~4~TION IN TI ME COORDINA.TES
Error in Time X
Y Z
Coordinates 1 2 3 4 average
Mean (see) -0.00101 0 . 0 0 0 1 -0.00021 0. 00061 -0.00041 0. 00013 0.00045
St andard 0.00432 0 . 0 0 5 0 3 0 . 0 0 4 1 2 0 . 0 0 4 2 6 0 . 0 0 4 4 3 0. 00398 0.00556
Devia-
tion (sec)
s t andar d devi at i on for each case are shown in Tabl e 3. The s t andar d devi at i ons shown
i n Tabl e 3 i ndi cat e t ha t a careful r eader will be consi st ent wi t hi n 0.004 sec about 70
per cent of t he t i me, wi t h a mean er r or ve r y close t o zero.
To compar e t he di gi t i zed records obt ai ned by di fferent persons, one of t he records of
X was t aken as an overal l s t andar d and t he records of Y and Z were subt r act ed f r om
i t t o obt ai n what is called t he cross-error of digitization. The mean and st andar d
devi at i on of t he error are shown i n Tabl e 4. The st andar d devi at i ons f r om t he mean
shown i n Tabl es 3 and 4 i ndi cat e t ha t t he cross error is about four t i mes t he self error,
t hus showing t he effect of personal preferences i n reading. The consi st ent , negat i ve
means i ndi cat e a const ant shi ft i n t he t i me axis of t he record chosen as a s t andar d as
compar ed t o t he ot her records. Such a shi ft does not effect t he devi at i on about t he
mean.
A compar abl e measur e of t he error i nt r oduced by rounding-off t he t i me record can
be obt ai ned by subt r act i ng t he r ounded record f r om t he original record and comput i ng
t he mean and st andar d devi at i on. These were done for one of t he records di gi t i zed by
X, and t he resul t s are shown i n Tabl e 5, for r oundi ng t o 0.01 and 0.005 sec.
The resul t s of t he exper i ment descri bed i n Tabl e 5 show t ha t t he er r or i nt r oduced by
roundi ng-off t he t i me coordi nat es of an ear t hquake record t o 0.01 or 0.005 secs is well
wi t hi n t he error i nher ent i n t he process of di gi t i zat i on. For round-off t o 0.01 secs, t he
st andar d devi at i on of t he er r or due t o round-off is about hal f of t he st andar d devi at i on
of t he self-error of di gi t i zat i on and about one si xt h of t he st andar d devi at i on of t he
cross-error. For round-off t o 0.005 secs, t hi s mar gi n is f ur t her i ncreased by a f act or of 2.
918 BULLETI N OF THE SEI SMOLOGI CAL SOCI ETY OF AMERI CA
Fr om t hi s i t is concl uded t ha t a round-off t o 0.01 or 0.005 sees is an accept abl e ap-
proxi mat i on.
Effect of Rounding Upon Accuracy. Ber g (1963) has exami ned errors i n spect r a
caused by r andom errors in t i me and accel erat i on coordi nat es and has concl uded f r om
an appr oxi mat e analysis and f r om comput er exper i ment at i on t ha t a scat t er of t he
or der of 20 per cent in undamped spect ra is t o be expect ed f r om i dent i cal comput i ng
procedures appl i ed t o i ndependent l y pr epar ed di gi t i zat i ons of t he same accel erogram.
He at t r i but ed t he scat t er t o r andom errors in readi ng t he accel erogram. Usi ng Ber g' s
appr oxi mat e f or mul ae for expect ed error i n vel oci t y spect r a due t o r andom errors irt
TABLE 4
CROSS-~RI~OR OF DI GI TI ZATI ON IN TI ME COORDINATES
Error in Time Y Z
Coordinates 1 2 average 1 2 average
Mean (see) -0.02525 -0.02538 -0.02532 -0.02212 -0.02257 -0.02235
Standard De- 0 . 0 1 8 1 9 0 . 0 1 7 9 4 0 . 0 1 8 0 6 0 . 0 1 5 2 0 0 . 0 1 5 2 0 0. 01520
viation (see)
TABLE 5
ERROR DUE TO I~OUND-OFF IN TI ME COORDINATES
Error in Time Coordinates Roound-off to 0.01 sees Round-off to 0.005 sees
Mean (see) 0.00003 0.00004
Standard Deviation (see) 0.00285 0.00149
Maximum Error 4-0.005 4-0.0025
t he t i me coordi nat es of t he record, and assumi ng a = 0.045g, (S~) ave ~ 20 si n/ see
and At = 0.05 sees, t he following resul t s are f ound:
Round-off t o 0.01 sees
( a ~ 0.003 sees)
Round-off t o 0.005 sees
(~ ~ 0.0015 sees)
Expect ed Er r or i n S,
0. 12~JD
Expect ed Per cent age
Er r or in S~ ( D =
30 sees)
3.3
0. 06v/ D 1.6
Her e, a = t he r.m.s, val ue of t he gr ound accel erat i on, = st andar d devi at i on of
t he expect ed error i n t he t i me coordi nat es and D = t he dur at i on of t he ear t hquake.
The resul t s i ndi cat e t ha t t he errors i n spect ral val ues due t o round-off t o 0.01 or 0.005
sees are much less t ha n t he 17 per cent error Ber g (1963) f ound l i kel y t o occur as a
resul t of r andom errors in readi ng t he t i me coordi nat es dur i ng di gi t i zat i on.
The val i di t y of t he above analysis can be checked by compar i ng t he spect r a for
unr ounded and r ounded t i me records. Thi s was done for t he N 65E component of t he
st at i on 2 record of t he Parkfi el d ear t hquake of June 27, 1966 and t he N 8E component
of t he Li ma, Per u ear t hquake of Oct ober 17, 1966. The undamped vel oci t y spect ra for
t he original ( unr ounded) record and for t he records obt ai ned af t er round-off t he t i me
coordi nat es t o 0.01 sees and t o 0.005 sees are shown in Fi gures zi and 5. For t he Par k-
CALCULATI ON OF RESPONSE SPECTRA 919
field ear t hquake, t he t hr ee curves ar e al most indistinguishable. For t he Li ma, Per u
ear t hquake t he aver age per cent age errors ar e of t he order expect ed on t he basis of
Be r t ' s appr oxi mat e analysis.
4O
IM
20 /
i l
X
/
r
, / 3 = 0 . 0
i
1.0
J
2.0
[
LIMA-PERU EARTHQUAKE
OCT I ~ 1966
N8 E
- - U N R O U N D E D
- ~ - R OU N D E D TO . 01SEC.
- ~ - R O U N D E D TO . 0 0 5 SEC.
J
~ V
PERIOD IN SECS
FIG. 4. Effect of rounding the time coordinates on velocity spectra.
3.0
80
w
6O
Z
z
> -
I . -
S 4o
w
>
<
~ 2O
j
~
/
0 j
\ /
,B=O.O
]
I
I
PARKFIELD EARTHQUAKE STN. 2
JUNE 27, 1966
N6 5 E
i i
1.0 2.0
PERIOD IN SECS
I a. 5. Ef f e c t of r oundi ng t he t i me c oor di na t e s on ve l oc i t y s pe c t r a .
5.0
The comput i ng t i me for spect r a based on di gi t i zed accel erograms r ounded t o 0.005
~ecs was f ound t o be onl y' 10 t o 15 per cent mor e t ha n t he comput i ng t i me for records
r ounded t o 0.01 secs. Consi deri ng t he expect ed error in t he spect ral val ues in t he t wo
920 B UL L E T I N OF T HE S E I S MOL OGI C AL S OC I E T Y OF AME R I C A
cases, i t is concl uded t h a t t o be consi st ent wi t h t he choice of ( At ) ~ ( < T/ 2 0 ) , r ound-
off t o 0.005 secs woul d be adopt ed as s t andar d. Her e agai n i t ma y be poi nt ed out t h a t
t he use of (AT)~ ==_ T/ 10 and round-off t o 0.01 secs shoul d pr ove accept abl e for mos t
pur poses.
Intervals of Integration. The pr ocedur e for comput i ng spect r a out l i ned above re-
qui res t he choi ce of a ma x i mu m i nt er val of i nt egr at i on, ( Ar ) ~ . Since t r uncat i on
er r or is not i nvol ved, t he choi ce is gover ned b y t he er r or i nt r oduced b y di scret i zat i on,
and b y comput i ng t i me. As shown i n Fi gur e 3, if (hr)m < T/20, t he er r or due t o
TABLE 6
"UNDAMPED VELOCITY SPECTRA AND RELATIVE COMPUTING TIMES FOR I~ECORDS DIGITIZED AT
UNEQUAL TI ME I NTERVALS
(N 65E Component the Station 2 Accelerogram of the Parkfield Earthquake, June 27, 1966,
(Z~r)~ = 0.05 sees)
S~ I n Ins/ Sec
Peri od I n Secs Thi r d Or der Runge- Kut t a exact
Ar ~ T/IO A'r <=T/ 20 rA < T/ 40 Ar =< / ' / 80 Ar =< T/IO Ar -< T/ 20
0.05 0.404177 0. 547537 0. 690496 0. 781693 0.776551
0.1 1.864953 2. 749153 3. 177131 3. 391490 3.446407
0.2 8.662501 10. 587279 10. 905353 10. 929761 11.007311
0.3 14. 519660 20. 197662 20. 780575 20. 880682 20.724359
0.4 39. 198212 40. 554802 40. 660100 40. 677461 40.844893
0.5 46. 064539 50. 050687 51. 059478 51. 129117 51.329105
0.7 85. 185941 86. 373600 89. 262830 89. 655182 90.064254
0.9 57. 705933 59. 043586 60. 146103 60. 228546 60.508116
1.0 44. 625701 44. 625701 44. 980583 44. 990469 45.149522
1.6 63. 948129 63. 948129 63. 960555 63. 941084 64.027008
2.0 64. 084503 64. 084503 64. 084503 64. 134411 64.167577
0.802090
3.446375
11.007311
20.954274
40.844893
51.329105
90.063806
60.488801
45.149522
64.027008
64.167577
Relative com-
putation
time
1.25 1.5 2.25 3.5 0.75 1
Note: Spectrum values below dotted lines were calculated for (Ar)m = 0.05 secs.
di scr et i zat i on is less t ha n 1.2 per cent . Thi s ma y be t r eat ed as one const r ai nt on t he
choice of ( Ar ) ~ .
The effect of ( At ) ~ on t he comput i ng t i me is mor e compl ex i n t h a t a l ar ger val ue of
( At ) ~ t ends t o r educe t he numbe r of i nt egr at i on st eps but i ncreases t he numbe r of
mat r i ces t o be cal cul at ed. Smal l er val ues of ( Ar ) ~ i mpl y t he r ever se resul t s, so a
choi ce of ( At ) ~ shoul d s t r ~ e a bal ance bet ween t he t wo effects. The i deal choi ce
woul d depend on t he nat ur e of t he accel er ogr am used, but a compr omi se sat i sf act or y
for mos t uses was achi eved b y t he choice of ( At ) ~ = 0.05 for s pect r um peri ods equal
or exceedi ng one second and (AT)~ =< T/20 for peri ods less t h a n one second.
I f t he t i me- coor di nat es of t he record ar e r ounded t o 0.005 secs, t her e is mor e t h a n
one wa y i n whi ch t he i nt er val s of di gi t i zat i on can be subdi vi ded, as i l l ust r at ed in
Tabl e 2. The choice of subdi vi di ng met hod is gover ned b y t he comput i ng t i me, and t o
exami ne t hi s quest i on a few t ri al s wer e conduct ed usi ng bot h t he met hods i n Tabl e 2.
I t was f ound t ha t in mos t cases, subdi vi si on met hod I I a ( Ar s -- 0.04, 0.03, 0.02, 0.01
and 0.005) requi red less comput i ng t i me.
CALCULATION OF RESPONSE SPECTRA 921
Comparison to a Third Order Runge-Kutta Method. To indicate t he effect of trunca-
tion error on spectrum values obtained by a Runge- Kut t a method and to compare t he
accuracy and computing time for t hat method wi t h t he techniques developed above,
undamped velocity spectra were computed by bot h methods. The results of these
computations are compiled in Table 6.
The results in Tabl e 6 show again t hat the accuracy of the proposed method is only
achieved by t he Runge- Kut t a method with Ar =< T/80, and t hat there is a three-to-
four-fold savings in time by t he use of t he proposed technique.
Computer Programs. Comput er programs, in Fort ran IV, for the comput at i on of
spectra using t he method of comput at i on presented in this paper are included in a
recent report (Ni gam and Jennings, 1968). Fl ow charts and instructions for the use of
t he programs are given also.
CONCLUSIONS
This paper presents a numerical met hod for the calculation of response spectra from
strong-motion earthquake accelerograms. The method is based on the exact solution
of the differential equation governing the response of a simple oscillator to segmentally
linear excitation. To adapt t he exact solution to efficient computing, the solution is
written in t he form of two 2 X 2 matrices which operate upon the conditions at the
beginning of t he integration step and upon the acceleration at the beginning and end of
the integration step to produce the exact velocity and displacement at the end of the
time interval. Because the matrices are functions only of the damping and period of
the oscillator used to calculate spectra and of the magnitude of the time interval, only
a limited number of t he matrices need be evaluated. The method can be applied di-
rectly to accelerograms digitized at equal time intervals, and by suitably rounding the
time coordinates, t o records digitized at unequal time intervals. The errors introduced
by rounding t he time coordinates are much less t han other errors inherent in the
digitization of eart hquake records.
The choice of using equal or unequal time intervals for digitization of accelerograms
usually is not based on computational ease, but on t he t ype of equipment available for
t he work. However, t he choice of digitization met hod affects the calculation technique
and t he computing time. For t he same number of sample points, digitization at unequal
time intervals represents the original record bet t er t han does digitization at equal
intervals, and if digitization is done at equal intervals a small time step usually must
be used to avoid significant distortion of t he record. Because for t he usual integration
methods the interval of integration must be less t han or equal to the interval of digiti-
zation, a small digitization interval makes i t necessary to use a small integration step
where otherwise a larger interval could be used. This leads t o an increase in computing
time and is of particular significance to t he met hod presented herein, because this
technique does not involve truncation error and therefore can be used accurately with
time intervals as large as T/IO to T/20. Thus, an accelerogram digitized at unequal
intervals, wi t h its smaller number of coordinates, can be processed more rapidly by t he
proposed met hod t han can t he same record digitized at equal time intervals.
For the accelerograms used as examples in this study, t he proposed met hod showed
a three-to four-fold saving in time over a third order Runge- Kut t a met hod of com-
parable accuracy. This was found to be t he case for records digitized at bot h equal and
unequal time intervals. For accelerograms wi t h durations of 30 seconds, spectra for 30
periods and 4 values of damping were calculated and pl ot t ed by an I BM 7094 com-
puting syst em wi t h an average computing time of 120 seconds, of which 100 seconds
were used for execution of the calculations. If spectra for several earthquakes were to
922 BULLETI N OF THE SEISMOLOGICAL SOCIETY OF AMERICA
be computed in the same operation, a further savings in computing time could be
achieved by storing the matrices used in the integration met hod rather t han recom-
puting t hem at t he beginning of the processing of each accelerogram.
Because of the increasing interest in the response of yielding structures to earthquake
motions, it is appropriate to point out t hat the method presented herein can be adapt ed
also to the calculation of spectra for bilinear hysteretic and elasto-plastic oscillators. To
apply the met hod it would be necessary to compute two sets of matrices corresponding
to t he two stiffness coefficients of t he structure.
ACKNOWLEDGMENT
Appr eci at i on is ext ended to t he Nat i onal Science Foundat i on for par t i al suppor t of t hi s re-
search under NSF Gr ant 1197X.
REFERENCES
Alford, J ' . L' , G. W. Housner, G. W. and R. R. Mar t el (1951). Spectrum Anal ysi s of Strong-Motion
Earthquakes, Ear t hquake Engr. Res. Labl , Calif. I nst . of Tech., Pasadena. (Revised, Aug.
1 9 6 4 )
B e n d a t a n d A . G . P i e r s o l ( 1 9 6 6 ) . M e a s u r e m e n t a n d A n a l y s i s o f R a n d o m D a t a . J o h n W i l e y a n d
S o n s , N e w Y o r k .
B e n i o f f , H . ( 1 9 3 4 ) . C a l c u l a t i o n o f t h e r e s p o n s e o f a n o s c i l l a t o r t o a r b i t r a r y g r o u n d m o t i o n , B u l l .
S e i s m . S o e . o f A m . 3 1 , 3 9 ~ 4 0 3 .
B e r g , G . V . ( 1 9 6 3 ) . A s t u d y o f e r r o r s i n r e s p o n s e s p e c t r u m a n a l y s e s , J o r n a d a s C h i l e n a s d e S e s -
mologia e Ingenieria Ant i si smi ca 1, B1.3, 1-11.
Berg, G. V., G. W. Housner (1961). I nt egr at ed vel oci t y and di spl acement of st r ong ear t hquake
ground mot i on, Bull. Seism. Soc. of Am. 51,175-189.
Bl ot , M. A. (1941). A mechani cal anal yzer for t he pr edi ct i on of ear t hquake st resses, Bull. Seism.
So c. o f Am. 31,151-171.
Br ady, A. G. (1966), Studies of Response to Earthquake Ground Motion, Ear t hquake Engr. Res.
Lab. , CaliL Inst . of Tech., Pasadena.
Hi l debr and, F. B. (1956). Introduction to Numerical Anal ysi s, McGr aw- Hi l l Book Company Inc. ,
New York.
Housner, G. W. (1941). Cal cul at i on of t he response of an osci l l at or to ar bi t r ar y ground mot i on,
Bull. Seism. Soc. of Am. 32,143-149.
I~Iousner, G. W. and G. D. McCann (1949). The anal ysi s of st rong-mot i on ear t hquake records wi t h
el ect ri c analog comput er, Bull. Seism. Soc. of Am: 39, 47-56.
Hudson, D. E. (1962). Some probl ems in t he appl i cat i on of spect rum t echni ques to st r ong- mot i on
ear t hquake anal ysi s, Bull. Seism. Soc. of Am. 52,417-430.
Iwan, W. D. (1960). Digital Calculation of Response Spectra and Fourier Spectra, Unpubl i shed
Not e, Calif. I nst . of Tech. , Pasadena.
Jennings, P. C. (1963). Velocity Spectra of the Mexican Earthquakes of 11 May and 19 May 1962,
Ear t hquake Engr. Res. Lab. Calif. I nst . of Tech. , Pasadena.
Nigam, N. C. and P. C. Jenni ngs (1968). Digital Calculation of Response Spectra f rom Strong-
Motion Earthquake Records, Ear t hquake Engr. Res. Lab. , Calif. I nst . of Tech. , Pasadena.
Schiff, A. and J. L. Bogdanoff (1967). Anal ysi s of current met hods of i nt er pr et i ng st r ong mot i on
accelograms, Bull. Seism. Soc. of Am. 57,857-874.
DIVISION OF ENGINEERING AND APPLIED SCIENCE
CALIFORNIA INSTITUTE OF TECHNOLOGY
PASADENA, CALIFORNIA 91109
Manuscr i pt received Sept ember 12, 1968.