Inelastic Spectrum-Based Approach for Seismic

Design Spectra
Sittipong Jarenprasert1; Enrique Bazán2; and Jacobo Bielak, M.ASCE3

Abstract: A consistent approach is proposed for deriving inelastic design spectra and estimates of maximum displacement directly
through statistical studies of inelastic response spectra, without the need to resort first to elastic spectra. The main finding is that the
seismic response coefficient, or dimensionless yield strength, Cy, that will maintain the mean ductility ratio, ␮¯ , equal to a prescribed value
can be estimated by an expression of the form Cy共T , ␮ ¯ 兲 = C共T兲␮
¯ −n共T兲. C is interpreted as a mean unreduced inelastic spectrum, and n
depends only on the elastic natural period, T, of the structure. Explicit formulas for both C and n are obtained for a set of 87 accelero-
grams recorded in California. C differs from the mean 5% damped elastic spectrum. Another significant result is that Cy共T , ␮ ¯ = 2兲 can be
closely approximated by a highly damped mean elastic spectrum, i.e., Ce共T , ␰兲, with ␰, the critical damping ratio taken to be 30%. Based
on these two results, Cy can be conveniently written as Cy共T , ␮ ¯ 兲 = Ce共T , 30兲共␮
¯ / 2兲−n共T兲. This means that the seismic coefficient of an
inelastic system can also be expressed approximately in terms of a highly damped mean elastic spectrum divided by a reduction factor that
depends only on ␮ ¯ and T. With Cy determined, it is straightforward to approximate consistently the normalized mean maximum relative
displacement of the structure as the product of Cy and ␮ ¯ ; the approximate results differ by less than 10% from the corresponding exact
values, for elastic natural periods between 0.1 and 3.0 s and mean ductility ratios up to 5.
DOI: 10.1061/共ASCE兲0733-9445共2006兲132:8共1284兲
CE Database subject headings: Seismic response; Inelasticity; Earthquake resistant structures; Response spectra.

Introduction level. The key to an adequate design is to ensure that this reduc-
tion in required strength be consonant with the combined over-
Current seismic provisions allow the seismic design of building strength and global ductility capacity of the lateral force-resisting
structures to be based on static or dynamic analyses of damped system. The development and evolution of strength reduction fac-
elastic models of the structure. The seismic base shear force is tors have been based on the comparison of the response of simple
prescribed in terms of “elastic design spectra,” or, more precisely, elastic and inelastic models, and, most importantly, on consis-
design spectra with no reductions due to inelastic behavior. Most tency with previous design approaches and judgment and obser-
codes 关e.g., ASCE 7-02 共ASCE 2002兲; 2003 International Build- vation of the overall performance of different types of buildings
ing Code 共IBC兲 共ICC 2003兲兴 associate a 5% viscous damping to during actual earthquakes.
these unreduced design spectra. Recognizing, however, that it The pioneering work of Veletsos and Newmark 共1960兲 on the
would be unduly expensive to design buildings to remain within earthquake response of single-degree-of-freedom 共SDF兲 elasto-
the elastic range during severe earthquakes, the codes anticipate plastic systems provided the basis for the development of the first
that structures will undergo inelastic deformations under seismic inelastic design spectra by Newmark and Hall 共1982 and refer-
events of a similar intensity to that used in developing the design ences therein兲 from elastic spectra. Their procedure consists in
dividing the elastic design spectrum into several regions and then
spectra. Such inelastic behavior is usually incorporated into the
applying a separate reduction factor to each spectral region as a
design by dividing the “elastic” spectra by a factor that reduces
function of a specified ductility ratio. The strength reduction fac-
the spectrum from its original elastic demand level to a design
tor thus depends primarily on the natural period of the system and
1
the maximum tolerable displacement ductility demand. Many ad-
Staff Engineer, Paul C. Rizzo Associates, 105 Mall Blvd, Ste. 270-E, ditional studies based on the concept of a reduction factor from
Monroeville, PA 15146; formerly, Graduate Student, Dept. of Civil and
elastic spectra have been conducted over the years 共e.g., Riddell
Environmental Engineering, Carnegie Mellon Univ., Pittsburgh, PA
15213. E-mail: [email protected] and Newmark 1979; Elghadamsi and Mohraz 1987; Nassar and
2
Senior Staff Engineer, GAI Consultants, Inc., 385 East Waterfront Krawinkler 1991; Miranda 1993; Miranda and Bertero 1994;
Dr., Homestead, PA 15120. E-mail: [email protected] Vidic et al. 1994; Ordaz and Pérez-Rocha 1998; Riddell et al.
3
Professor, Dept. of Civil and Environmental Engineering, Carnegie 2002; Chopra and Chintanapakdee 2004兲 with the objective of
Mellon Univ., Pittsburgh, PA 15213. E-mail: [email protected] improving the agreement between elastic reduced and inelastic
Note. Associate Editor: Vinay Kumar Gupta. Discussion open until results, and incorporating the effects of parameters, such as type
January 1, 2007. Separate discussions must be submitted for individual of force-deformation relationship, earthquake magnitude, epicen-
papers. To extend the closing date by one month, a written request must
tral distance, and soil conditions.
be filed with the ASCE Managing Editor. The manuscript for this paper
was submitted for review and possible publication on September 2, 2004; There has also been a significant effort to develop simple rules
approved on October 5, 2005. This paper is part of the Journal of Struc- to evaluate approximately the maximum relative displacement,
tural Engineering, Vol. 132, No. 8, August 1, 2006. ©ASCE, ISSN without having to resort to detailed nonlinear analyses. The rea-
0733-9445/2006/8-1284–1292/$25.00. son is that the maximum relative displacement is the most widely

1284 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / AUGUST 2006

used structural response parameter for evaluating the inelastic
performance of structures 共e.g., Moehle 1992; Kowalski et al.
1995; Chopra and Goel 1999; Fajfar 2000兲. For SDF elastoplastic
and other simple systems the peak relative displacement can be
calculated simply by multiplying the ductility ratio by the yield
displacement. If the latter is obtained from the elastic spectrum
and the reduction factor, it is then said that the method for calcu-
lating the reduction factor and the peak displacement is consis-
tent. There are two other methods for evaluating the peak relative
displacement of inelastic SDF systems based on the elastic
spectrum and the ductility ratio or reduction factor. In the dis-
placement modification procedure, the inelastic displacement is
calculated by multiplying the maximum deformation of a linear
elastic system with the same lateral stiffness and damping coeffi- Fig. 1. Spectra for mean seismic elastic coefficient and
cient as the inelastic system by a modifying factor that depends corresponding coefficient of variation of the 87 California
on the elastic natural period and on the extent of expected non- accelerograms scaled by different criteria
linearity 共e.g., Newmark and Hall 1982; Miranda 2000; Goel and
Chopra 2001兲. In the second approach, based on equivalent lin-
earization, the peak inelastic deformation is estimated as the its elastic vibration period, T, linear damping ratio ␰ 共assumed to
maximum deformation of a linear elastic system with lower lat- be 5% in this study兲, yield displacement, uy, and slope ␣ of the
eral stiffness and higher damping coefficient than those of the second branch of the skeleton force-displacement relationship, as-
elastic portion of the inelastic system 共e.g., Rosenblueth and Her- sumed throughout to be 2%. The yield strength coefficient, or
rera 1964; Gulkan and Sozen 1974; Iwan 1980, 2002; Kwan and seismic coefficient, Cy, is defined in terms of the yield force,
Billington 2003兲. The reduced lateral stiffness and modified Vy = CyW, where W = mg is the weight of the structure; m⫽mass of
damping coefficient are functions of the target ductility. Miranda the system; and g⫽acceleration of gravity. The nondimensional
and Ruiz-García 共2002兲 have evaluated different approximate factor Cy can be expressed in terms of the yield displacement and
methods to estimate the peak inelastic displacement demands and vibration period
have reported on their advantages and disadvantages when ap- 4␲2 uy
plied to practical situations. Cy = 共1兲
T2 g
The main motivation for the present study is the observation
that mean inelastic spectra are smoother than, and have little Of the various quantities commonly used to measure the response
resemblance to, mean elastic spectra with the 5% of critical of an inelastic SDF system, this study focuses on the maximum
damping usually associated with design spectra. Whereas elastic relative displacement, umax, and on the ductility demand,
spectra often exhibit large distinct peaks at various natural peri- ␮ = umax / uy. From Eq. 共1兲 and the definition of ␮, umax can be
ods, these peaks tend to disappear for inelastic spectra, and their expressed in dimensionless form in terms of Cy and ␮
ordinates increase only mildly or decrease almost monotonically
4␲2 umax
with period. This smoother behavior suggests that it might be = ␮Cy 共2兲
possible to derive design inelastic spectra without first construct- T2 g
ing the corresponding elastic spectra. Hence, it is reasonable to The seismic excitation consists of 87 accelerograms recorded in
investigate whether simple rules can be established to derive in- California for different earthquakes at sites with consistencies that
elastic seismic design spectra directly from mean inelastic re- range from hard to medium hard, with a few soft sites, according
sponse spectra. With this objective in mind, this study uses a to the definitions of the IBC 2003 Code 共ICC 2003兲. A complete
sample of 87 accelerograms recorded during strong earthquakes list of the accelerograms, which were downloaded from the strong
that exhibit spectral coefficients of variation similar to those re- motion database of the Pacific Earthquake Engineering Research
ported in ground motion studies for Western North America Center, can be found in Jarenprasert 共2005兲. The 87 records are
共Boore et al. 1997兲. The 87 accelerograms are used to obtain regarded as a conditional sample of the ground motion that could
mean inelastic response spectra for SDF viscously damped bilin- be expected at the site if an earthquake were to occur with the
ear systems. These spectra are regarded as the “exact design” intensity implied in a design spectrum. All the records have been
spectra and simple rules are then developed to approximate these scaled to produce the same Arias Intensity 共Arias 1970兲. This
“inelastic design” spectra. These approximations are then used to measure of intensity is based on an energy criterion expressed as
derive simple consistent formulas for estimating the mean peak the integral of the acceleration squared; it has been recently found
relative displacements of the structural systems. Some calcula- to be the best parameter for deriving seismic vulnerability func-
tions are also performed with accelerograms recorded in the lake- tions for steel pipes 共Hwang et al. 2004兲. Kayen and Mitchell
bed region in Mexico City, in order to illustrate similarities and 共1997兲 have reported that the Arias Intensity is the most reliable
differences with respect to soft sites. measure of earthquake severity for the assessment of seismic soil
liquefaction. As an additional verification, the 87 records were
alternatively normalized to have the same peak ground velocity
Description of Model and Seismic Input 共PGV兲 or peak ground acceleration 共PGA兲. The mean spectra re-
sulting for the three normalization procedures are presented in
A SDF system with a bilinear hysteretic viscously damped Fig. 1. The three criteria lead practically to the same average
force-deformation relationship is used to investigate the statistical spectra. Fig. 1 also shows the coefficients of variation resulting
structural response for actual records. Using standard structural from the three alternative normalization procedures. Of the three,
dynamics theory, the SDF bilinear hysteretic system is defined by the Arias normalization produces the more uniform variation over

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / AUGUST 2006 / 1285

Fig. 2. Cumulative frequencies of the scaled PGA, PGV, and PGD of Fig. 3. Comparison between mean strength coefficient Cy for a
the 87 California accelerograms considered in this study. Thin solid prescribed target ductility ␮t and strength coefficient Cy with
lines show the corresponding lognormal plots. corresponding mean ductility demand ␮
¯

the entire range of periods. Such coefficients vary from approxi- assigned a different strength depending on the particular excita-
mately 0.3 at zero period to about 0.7 for T = 3 s. These values are tion. For an ensemble of earthquake records one then plots the
of the same order as those reported in ground motion models mean value Cy of Cy for each period to obtain the mean inelastic
共Boore et al. 1987兲. It is also noteworthy that the spectra in Fig. 1 response spectrum for each desired value of the ductility demand
are very similar to the corresponding mean elastic spectra ob- ratio ␮. Thus, in this approach, constant target ductility is attained
tained by Vidic et al. 共1994兲 关see their Fig. 6共a兲兴 using a total of at the expense of changing the yield strength of the structure from
20 standard records from California and Montenegro, and practi- one seismogram to the next. It should be pointed out that the
cally identical to the mean elastic spectra of Lin and Chang ductility factor ␮ for a given seismogram does not always in-
共2003兲, obtained using 1053 accelerograms from 102 earthquakes crease monotonically as the yield strength Cy decreases from the
recorded in different seismic regions of the USA. elastic limit. In fact, more than one yield strength is possible
Fig. 2 presents the cumulative frequencies of the scaled PGA, corresponding to a given ␮, as illustrated in Chopra 共1995兲. In
PGV, and displacement 共PGD兲. By plotting cumulative statistical cases where there is more than one value of Cy, it is customary to
frequencies of the PGA, PGV, and PGD, it was found that their choose the smallest value for defining the inelastic spectrum.
variability is closely represented by lognormal probability distri- In this study, the mean inelastic response spectra will be de-
butions. The coefficients of variation are 0.26, 0.30, and 0.68, fined somewhat differently in order to represent a scenario in
respectively. These values are of the same order as those reported which a structure with prescribed elastic period and strength is
in published attenuation laws 共Dowrick 1987; Boore et al. 1997兲. subjected to the 87 accelerograms. The procedure is as follows:
Even though the present study is largely confined to one par- 共1兲 For each structural period T, find the minimum dimensionless
ticular type of structure and one sample of California accelero- 共Ce兲global that keeps the response of the structure within the linear
grams, the qualitative results might be applicable to other situa- range for all 87 earthquake records. The corresponding force
tions, for as has been observed by Riddell and Newmark 共1979兲, Fe = 共Ce兲globalW may be referred to as the peak global elastic
and more recently by Nassar and Krawinkler 共1991兲, Riddell et al. earthquake-induced resisting force. 共2兲 Then, introduce Cy as a
共2002兲, and Chakraborti and Gupta 共2005兲, the effects of the type progressively smaller fraction of 共Ce兲global. 共3兲 For each value of
of force-deformation relationship of nondegrading nonlinear sys- Cy and each record, solve numerically the nonlinear governing
tem are not significant for the inelastic reduction factors. Simi- differential equation of motion to calculate umax and evaluate the
larly, Nassar and Krawinkler 共1991兲, Miranda 共1993兲, and most corresponding ductility demand, ␮. Then, 共4兲 for each T and each
recently Chopra and Chintanapakdee 共2004兲 concluded that mag- Cy, calculate the mean value ␮ ¯ . Note that the value of ␮ can be
nitude and epicentral distance do not affect the reduction factor. less than unity for an individual record. Initially, the structure
On the other hand, Tiwari and Gupta 共2000兲 reached the opposite with a period T will enter into the inelastic range only for some of
conclusion. In addition, Miranda 共1993兲 observed that soil condi- the records; as Cy becomes progressively smaller, the structure
tions may influence significantly the reduction factors, though will behave nonlinearly for an increasing number of records. This
Elgadhamsi and Mohraz 共1987兲 reached the opposite conclusion approach is physically consistent in that the mean maximum dis-
and Miranda and Bertero 共1994, Fig. 12兲 show only modest dif- placement is equal to the mean ductility demand times the yield
ferences between the reduction factors for rock and alluvium. displacement, whether the behavior is elastic or inelastic. Statis-
tical consistency is also found when fitting a probability distribu-
tion to the statistical data: The inclusion of ductility values below
Inelastic Response Spectra unity produces very good lognormal distribution fits 共Jarenprasert
2005兲. In line with this observation, the coefficient of variation of
The common procedure for determining inelastic response spectra ␮
¯ from all the records is calculated under the assumption that ␮ ¯
consists in varying Cy for each individual record until a pre- has a lognormal distribution.
scribed value of the ductility demand ␮ is attained. This means Since the proposed procedure is conceptually different from
that for each structure with a given natural elastic period, and the first one, it is useful to compare results from the two proce-
prescribed ductility factor, the corresponding inelastic structure is dures. Fig. 3 shows two sets of inelastic spectra, for natural peri-

1286 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / AUGUST 2006

Fig. 4. Comparison between mean strength coefficient Cy for a Fig. 5. Variability of mean ductility demand ␮
¯ with yield strength Cy
prescribed target ductility ␮t and strength coefficient Cy with for prescribed values of the elastic natural period T. Figure also
corresponding mean ductility demand ␮
¯ for accelerograms recorded shows linear logarithmic approximations by solid lines.
in the Mexico City lakebed region

figure so that interpolation only affects the third significant figure.

ods between 0.1 and 3.0 s. Solid lines depict spectra for Cy for This eliminates a potential problem associated with the interpola-
different values of ␮¯ between 1 and 5. The dashed lines in Fig. 3 tion, which has been discussed by Riddell et al. 共2002兲 and
represent the corresponding inelastic spectra for the mean yield Ascheim and Black 共2000兲. The regression results for C共T兲 and
strength Cy for target ductility ␮t. In both cases, the yield strength n共T兲 are shown in solid lines in Fig. 6. These curves can be
has been normalized with respect to the dimensionless mean peak approximated for the period interval from 0.1 to 3.0 s by the ex-
ground acceleration PGA/ g. The two sets of curves are remark- plicit formulas
ably similar, indicating that for the selected sample of earth- 2
quakes, there is little difference between computing mean values C共T兲 = e−0.8T + 0.534e共ln T + 1.2兲 共4兲
of the yield strength given a target ductility, or mean ductility
demands for fixed values of yield strength. This close agreement n共T兲 = 1.5T0.5e−共0.5T兲 共5兲
between the two sets of curves may also be interpreted as an These approximations are shown by the thin dashed and dashed-
indication that including ductility values below unity does not dotted lines, respectively, in Fig. 6. The values of C共T兲 and n共T兲
significantly affect the results. A similar correspondence between from these expressions are very close to those obtained directly
Cy and ␮ ¯ and Cy and ␮ is observed from a comparison of the through regression. Notice that n passes through the origin, as
results from an ensemble of 66 scaled accelerograms recorded in required by the condition that Cy共T , ␮ ¯ 兲 be equal to PGA/ g at
the lakebed region in Mexico City, as shown in Fig. 4. Except for T = 0 for all ␮
¯ s; it then increases monotonically, reaching a peak
the case of a unit ductility ratio, both sets of curves almost coin- near unity around T = 1.0 s and then decays gradually with T.
cide. Such behavior is consistent with the observation by Miranda Eq. 共3兲 shows that the inelastic response spectrum Cy共T , ␮
¯ 兲 for
共2001兲 that the constant target ductility approach provides a first- the yield strength can be obtained in terms of a reference unre-
order approximation of the constant relative strength approach, duced spectrum C共T兲 that depends only on the elastic natural
and that it provides a small but systematic underestimation of period of the structure, divided by a modifying factor R␮¯ , i.e.,
displacement and ductility demands. These observations have
been recently corroborated by Farrow and Kurama 共2003兲 and by
Ruiz and Miranda 共2003兲.
To examine how the mean ductility demand ␮ ¯ varies with the
yield strength, Fig. 5 shows, on a logarithmic scale, ␮ ¯ versus Cy
for selected values of T, for the California ensemble of records.
As mentioned previously, it is well-known that for individual ac-
celerograms there may be more than one Cy for a given ␮. By
contrast, Fig. 5 shows a one-to-one correspondence between ␮ ¯
and Cy. Moreover, for each T, log ␮ ¯ essentially varies linearly
with log Cy for ␮ ¯ within the range of 1.5 to 5.5. This suggests
that, for each T, the variability of Cy with ␮¯ can be approximated
as

Cy共T,␮
¯ 兲 = C共T兲␮
¯ −n共T兲, 1.5 艋 ␮
¯ 艋 5.5 共3兲
C共T兲 = intersection of each straight-line approximation for Cy ver-
sus ␮¯ in Fig. 5 with the horizontal axis ␮ ¯ = 1 for period T, and
−n共T兲, with n共T兲 艌 0, is its slope. C共T兲 and n共T兲 can be deter- Fig. 6. Pseudo-linear spectrum C and power n in Eq. 共3兲, determined
mined by regression of Cy on ␮ ¯ for different values of T, similar both from a regression fit 共solid lines兲 and from approximations given
to those shown in Fig. 5, but using a large number of values of T by Eqs. 共4兲, 共5兲, and 共8兲, respectively 共dashed-dotted, dashed, and
and many more pairs 共Cy , ␮ ¯ 兲 for each T than those shown in the dotted lines兲

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / AUGUST 2006 / 1287

 Fig. 8. Inelastic unreduced spectrum, inelastic spectrum for the
Fig. 7. Variability of mean ductility demand ␮
¯ with yield strength Cy mean ductility ␮¯ = 1 共solid line兲, and mean elastic spectrum for
for prescribed values of the elastic natural period T. Figure also accelerograms recorded in the Mexico City lakebed region
shows linear logarithmic approximations by lines for accelerograms
recorded in the Mexico City lakebed region.

smaller than the mean elastic spectrum over the entire range of
natural periods, but it is also flatter and does not exhibit the peak
C共T兲 at about 2 s observed in the elastic spectrum.
Cy共T,␮
¯兲 = 共6兲 The results described in the preceding paragraphs suggest a
R␮¯
new approach for defining inelastic design spectra. The most di-
rect procedure is to simply use Eq. 共3兲, or equivalently Eqs. 共6兲
in which
and 共7兲, with the values for C共T兲 and n共T兲 given by Eqs. 共4兲 and
共5兲 to calculate the inelastic design spectrum for a desired mean
R␮¯ = ␮
¯ n共T兲 共7兲 ductility demand. Notice that this approach is based entirely on
inelastic analysis and does not make use at all of elastic spectra.
accounts for the effect of inelastic behavior in the system, as This property and the fact that the unreduced spectrum C共T兲
measured by the mean ductility demand. Eq. 共6兲 has precisely the seems to provide a natural basis for the selection of the reference
format adopted in current seismic codes. In addition to depending spectrum, makes this direct inelastic spectrum approach worthy of
on ␮ ¯ , R␮¯ , known as the inelastic or ductility reduction factor, consideration for potential adoption in future seismic provisions.
varies implicitly with T through the power n. On the other hand, inelastic design spectra in current practice
For values of ␮ ¯ lower than 1.5, the lines ␮
¯ versus Cy in Fig. 5 are still based on elastic design spectra. To conform to this ap-
no longer lie on a straight line, but curve visibly to the right. This proach, it is desirable to express the preceding results in a more
deviation from logarithmic linearity seems to indicate that it may traditional format. It was mentioned already that there have been
be more convenient to use the straight line extension of each line a number of studies devoted to obtaining peak inelastic response
down to ␮ ¯ = 1 in order to derive inelastic spectra via a simple based on equivalent linear oscillators with modified natural peri-
reduction factor, as indicated by 共6兲, rather than using the mean ods and effective amounts of damping. Along these lines, the use
elastic spectrum corresponding to the exact starred values at of increased damping is explored in the following sections as a
␮
¯ = 1 as the reference spectrum. proxy for inelastic spectra.
It is important to emphasize that C共T兲 is not the mean elastic
spectrum. From Fig. 5, it can be seen that C共T兲 will generally be
smaller than or equal to the exact Cy for ␮ ¯ = 1. Since from Fig. 3 Inelastic Spectra Based on High-Damping Elastic
Cy共T , ␮¯ = 1兲 is similar to the mean elastic spectrum, then C共T兲 can Systems
also be expected to be smaller than the mean elastic spectrum.
The 66 scaled Mexico City accelerograms are now used again An alternative approach to using Eq. 共3兲, with Eqs. 共4兲 and 共5兲 for
to examine whether the variability of Cy with ␮ ¯ described by Eq. C and n, is to associate C共T兲 with a highly damped elastic mean
共3兲 can also characterize the response of structures subjected to response spectrum. To determine a suitable relationship, it is use-
very different ground motion excitation. The results, shown in ful to compare the elastic spectra with the inelastic spectra shown
Fig. 7, indicate that the curve log ␮ ¯ versus log Cy is also linear, as in Fig. 3. Such a comparison is presented in Fig. 9, which dis-
in Fig. 5, within a certain range of values of ␮ ¯ . In contrast with plays in solid lines the inelastic spectra Cy共T , ␮
¯ 兲 for values of the
the California case, however, the relationship deviates drastically mean ductility demand between 1.5 and 5 and, in dashed lines,
from linearity as ␮ ¯ approaches the unit value. Consequently, the the mean elastic spectra Ce共T , ␰兲 for damping ratios between 5
difference between the unreduced spectral values C共T兲, corre- and 30%. Notice that the shape of the elastic spectra for large
sponding to ␮ ¯ = 1, and the respective mean elastic values is quite damping ratios resembles that of the inelastic spectra with a ref-
pronounced. To appreciate more clearly this effect, the 5% mean erence linear damping of 5%. The striking resemblance between
elastic spectrum, the unreduced inelastic spectrum C共T兲, and the the inelastic spectrum for a mean ductility demand of 2 and the
exact inelastic spectrum, Cy, for a mean ductility demand ␮ ¯ of 1 elastic spectrum for a critical damping ratio of 30% suggests that
for the Mexico City accelerograms are presented in Fig. 8. Not Ce共T , 30兲 can be used as an approximation for Cy共T , 2兲. Thus, by
only is the inelastic unreduced spectrum C共T兲 significantly using Eq. 共3兲 with ␮¯ = 2, C共T兲 can be expressed approximately as

1288 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / AUGUST 2006

Fig. 9. Normalized spectra Cy共T , ␮ ¯ 兲 and Ce共T , ␰兲 for different mean Fig. 10. Normalized spectra Cy共T , ␮ ¯ 兲 and Ce共T , ␰兲 for different mean
ductility ratios, ␮
¯ , and critical damping ratios, ␰ ductility ratios, ␮
¯ , and critical damping ratios, ␰, for accelerograms
recorded in the Mexico City lakebed region

C共T兲 = 2n共T兲Ce共T,30兲 共8兲

inelastic spectra for the 87 California accelerograms, three sets of
To compare this approximation with the regressions fit for C共T兲, inelastic spectra are shown in Fig. 11. Solid lines correspond to
the plot of Eq. 共8兲 has been included in Fig. 6. The agreement is the exact spectra for several mean ductility demands obtained
excellent. Now, substituting back C共T兲 from Eq. 共8兲 into Eq. 共3兲 from the numerical solution of the governing nonlinear equation
results in of motion. Dashed lines represent inelastic spectra obtained from
Cy共T,␮
¯ 兲 = Ce共T,30兲共␮
¯ /2兲−n共T兲 共9兲 Eq. 共3兲, and the dashed-dotted lines are the values obtained from
Eq. 共9兲 based on the 30% mean elastic spectrum. For the values of
Eq. 共9兲 defines a new approach for developing inelastic design ␮
¯ 艌 1.5 shown, the agreement between exact and approximate
spectra. Standard procedures use lightly damped elastic oscilla- spectra is quite close. In fact, as shown in Fig. 12, the actual mean
tors, typically with 5% critical damping, as a basis. By contrast, ductility demands corresponding to the yield strengths defined by
Eq. 共9兲 provides a simple rule for inferring inelastic design spec- Eqs. 共3兲 and 共9兲 differ only moderately from the target ductility
tra from spectra for highly damped elastic spectra, with 30% ratios, for all values of ␮ ¯ between 1.5 and 5 and periods from
damping for the 87 accelerograms used in the present study. 0.1 to 3.0 s.
That inelastic spectra can be derived readily from highly
damped elastic spectra is a useful result, since due to their relative
smoothness, highly damped elastic spectra can be defined ap- Consistent Evaluation of Maximum Relative
proximately by fewer values of the natural period than those re- Displacement
quired for lightly damped systems. Moreover, using highly
damped elastic spectra as a basis for design opens the possibility With the simple rules for determining the yield strength spectra of
for developing directly mean inelastic 共and eventually design兲 the bilinear hysteretic system, it is straightforward to estimate the
spectra from observations that use inexpensive seismoscopes. system’s mean peak relative displacement, ūmax, in a consistent
These would need to be designed for only two or three natural manner from Eq. 共2兲. Thus, by using Eq. 共9兲, it is possible to
periods and a high damping ratio. approximate ūmax by
The new approach rests on the observation that a highly
damped elastic spectrum can serve as a surrogate for a particular
inelastic spectrum. It is thus here called SELIS for surrogate elas-
tic inelastic spectrum. To explore whether SELIS can be used also
for sites that experience very different seismic excitation than that
represented by California accelerograms, results for the 66 accel-
erograms recorded at sites in the soft, lakebed, region in Mexico
City are briefly considered next. Fig. 10 shows elastic and inelas-
tic spectra, in the same sense as Fig. 9. Whereas the 5% mean
elastic spectrum exhibits a characteristic peak around a period of
2.0 s, neither the highly damped elastic spectra nor the inelastic
spectra display this behavior. On the contrary, their behavior is
qualitatively similar and does not show any large peaks: Both
highly damped 共␰ 艌 30 percent兲 and inelastic 共␮ ¯ 艌 2兲 spectra re-
main essentially flat over a range of periods, and either increase
or decay gradually elsewhere. Moreover, it appears that just as for
the California earthquakes, for Mexico City a particular inelastic
spectrum may be approximated with a suitable highly damped Fig. 11. Normalized yield strength coefficient spectra, Cy, obtained
elastic spectrum. by numerical integration of equation of motion 共solid line兲 and by the
To assess how closely the direct inelastic spectra procedure approximate closed-form expressions 共3兲 共dashed line兲 and 共9兲
represented by 共3兲 and the SELIS approach approximate the exact 共dashed-dotted line兲

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / AUGUST 2006 / 1289

 Concluding Remarks

Two main conclusions can be drawn from this study of inelastic

spectra of SDF bilinear hysteretic viscously damped systems
共with 5% critical damping兲 subjected to a sample of California
accelerograms. The first finding is that the use of a pseudo-elastic
spectrum based solely on the analysis of inelastic systems, rather
than using the elastic 5% spectrum as a basis, leads to a single
consistent expression for the reduction factor from an unreduced
reference spectrum valid for all natural periods of interest and for
ductility demands down to unity.
Significantly, the peak of the unreduced inelastic spectrum is
approximately 15% smaller than that of the 5% damped mean
elastic spectrum. 共This difference is much greater for the Mexico
accelerograms.兲 The second conclusion stems from the observa-
Fig. 12. Mean ductility demand, ␮¯ , 共solid and dashed dotted lines兲 tion that a highly damped mean elastic spectrum 共with 30% criti-
corresponding to yield strength coefficient spectra Cy共T , ␮t兲, cal damping兲 can be used as a surrogate for the inelastic spectrum
evaluated from Eqs. 共9兲 and 共3兲, respectively, for prescribed target for a mean ductility demand of 2. This allows one to estimate the
ductility ␮t 共dashed lines兲 inelastic spectrum for mean arbitrary ductility demand ␮ ¯ as the
ratio of the highly damped elastic spectrum and a correction fac-
tor given by ␮ ¯ / 2 raised to a power that depends only on the
4␲2 ūmax elastic natural period of the structure. Using a highly damped
=␮
¯ Cy共T,␮
¯ 兲 = C共T兲␮
¯ 1−n共T兲 共10兲
T2 g elastic spectrum as the reference spectrum has the advantage of
simplicity and perhaps also accuracy, since highly damped spectra
for the direct inelastic analysis approach, and by
are much smoother than the usual 5% critical elastic spectrum, are
4␲2 ūmax less sensitive to accelerogram base line corrections, and have a
= 2Ce共T,30兲共␮
¯ /2兲1−n共T兲 共11兲 much closer resemblance to the target inelastic spectra. In addi-
T2 g
tion, one can conveniently estimate an inelastic spectrum from
for the SELIS approach. The factor 2Ce共T , 30兲 = approximation simple linear analyses.
for the normalized ūmax for ␮¯ = 2; and the remaining coefficient is Upon additional verification of the high damping elastic spec-
the modifying factor for other mean ductility demands between trum approach to derive inelastic design spectra, ground motion
1.5 and 5. Fig. 13 shows a comparison between the dimensionless models 共attenuation laws兲 could be directly developed for spectra
mean peak relative displacement spectrum estimates calculated with large damping. Such spectra are useful for structures on base
from Eqs. 共10兲 and 共11兲 and the corresponding values obtained isolation 共e.g., Skinner et al. 1993兲 or with other passive damping
directly from the numerical solution of the governing equation of systems 共e.g., Hanson and Soon 2001; Ramirez et al. 2002兲. It is
motion for a mean peak ground acceleration equal to g. These well known 共e.g., Wu and Hanson 1989兲 that high-damping sys-
results show that the difference between the three methods is less tems vary more uniformly with period and have smaller coeffi-
than 10% for all ductility ratios and periods considered. Notice cients of variation for a constant period, facilitating the statistical
that for elastic natural periods greater than 0.75 s, n共T兲 is close to analysis. Indeed, recent work toward the development of attenu-
unity and the normalized maximum inelastic displacement be- ation laws for high-damping spectra has been undertaken by
comes relatively insensitive to ␮. Atkinson and Pierre 共2004兲 for damping ratios up to 15% for the
eastern U.S. Optionally, the equations for mean elastic spectra
provided by existing ground motion models for 5% damping
could be transformed to higher damping ratios by using formulas
such as those by Arias and Husid 共1962兲, Newmark and Hall
共1982兲, and Jarenprasert 共2005兲. In addition, using highly damped
elastic spectra as a basis for design spectra opens the possibility
for developing directly mean inelastic 共and eventually design兲
spectra from observations that use inexpensive seismoscopes de-
signed for two or three natural periods and a high damping ratio,
as the mean inelastic spectra can be approximated quite accu-
rately by just a few values 共e.g., 0.2, 1.0, and 3.0 s兲. With this
application in mind, it is particularly encouraging that using
records from the 1971 San Fernando earthquake, Hudson 共1972兲
showed that the spectral ordinates read directly from
seismoscopes located at several sites within the epicentral region
provided very close approximations to the corresponding dis-
placement spectra derived from accelerograms located at the
Fig. 13. Dimensionless mean peak relative displacement 共pseudo- same sites. In another finding, the present study suggests that
acceleration兲 spectrum for prescribed mean target ductility ratios ␮t, design spectra in building codes can be interpreted as unreduced
obtained both by numerical integration of the equation of motion inelastic spectra rather than 5% elastic spectra. Eqs., such as 共4兲
共dashed lines兲, by the estimate provided by Eq. 共10兲 共dashed lines兲, and 共5兲, could then be used to evaluate inelastic reduction factors.
and by Eq. 共11兲 共solid lines兲 This conjecture, of course, needs to be examined carefully for

1290 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / AUGUST 2006

different earthquake ensembles and other seismic regions, before References
it can be applied in practice.
Finally, with the knowledge of the inelastic spectrum for a ASCE. 共2002兲. “Minimum design loads for buildings and other struc-
prescribed mean ductility demand, the mean peak relative dis- tures.” ASCE 7-02, ASCE, Reston, Va.
Arias, A. 共1970兲. “A measure of earthquake intensity.” R. J. Hansen, ed.
placement can be consistently approximated as the product of the
Seismic design for nuclear power plants, MIT Press, Cambridge,
yield displacement times the mean ductility demand. Approxi- Mass., 438–483.
mate results, for natural periods between 0.1 and 3.0 s and mean Arias, A., and Husid, R. 共1962兲. “Influencia del Amortiguamiento sobre
ductility ratios up to 5, show that estimates from either the direct la Respuesta de Estructuras Sometidas a Temblor.” Rev. IDIEM, 1共3兲,
inelastic approach or the SELIS approach differ by less than 10% 219–228.
Aschheim, M. A., and Black, E. F. 共2000兲. “Yield point spectra for seis-
from the corresponding values obtained from the numerical solu-
mic design and rehabilitation.” Earthquake Spectra, 16共2兲, 317–335.
tion of the governing nonlinear differential equation of motion of Atkinson, G. M., and Pierre, J. R. 共2004兲. “Ground-motion response spec-
the inelastic system, well within the accuracy required in practice. tra in eastern North America for different critical damping values.”
The results presented in this paper correspond to a single struc- Seismol. Res. Lett., 75共4兲, 541–545.
tural system and to only two particular ensembles of earthquake Boore, D. M., Joyner, W. B., and Fumal, T. E. 共1997兲. “Equations for
ground motion. Based on observations from previous studies and estimating horizontal response spectra and peak acceleration from
western North American earthquakes: A summary of recent work.”
because of the great differences between the two ensembles con-
Seismol. Res. Lett., 68共1兲, 128–153.
sidered herein, there is reason to expect that the new approach, Chakraborti, A., and Gupta, V. K. 共2005兲. “Scaling of strength reduction
either in its direct inelastic form or as the SELIS approach, might factors for degrading elasto-plastic oscillators.” Earthquake Eng.
be applicable also to more general systems. Struct. Dyn., 34, 189–206.
Chopra, A. K. 共1995兲. “Dynamic of structures.” Theory and Applications
to earthquake Engineering, W. J. Hall, ed., Prentice-Hall, Upper
Saddle River, N.J., 257–259.
Chopra, A. K., and Chintanapakdee, C. 共2004兲. “Inelastic deformation
Acknowledgments
ratios for design and evaluation of structures: Single-degree-of-
freedom bilinear systems.” J. Struct. Eng., 130共9兲, 1309–1319.
This research was partially supported by the National Science Chopra, A. K., and Goel, R. 共1999兲. “Capacity-demand-diagram methods
Foundation Division of Engineering Education and Centers under based on inelastic design spectrum.” Earthquake Spectra, 15共4兲,
Grant No. 01-21989. The writers are grateful for this support. 637–656.
They also thank the reviewers for their thoughtful comments, Dowrick, D. J. 共1987兲. Earthquake resistant design for engineers and
architects, 2nd Ed., Wiley, N.Y.
which led to significant improvements in the article.
Elghadamsi, F. E., and Mohraz, B. 共1987兲. “Inelastic earthquake spectra.”
Earthquake Eng. Struct. Dyn., 15共1兲, 91–104.
Fajfar, P. 共2000兲. “A nonlinear analysis method for performance-based
seismic design.” Earthquake Spectra, 16共3兲, 573–592.
Notation Farrow, K. T., and Kurama, Y. C. 共2003兲. “SDOF demand index relation-
shops for performance-base seismic design.” Earthquake Spectra,
The following symbols are used in this paper: 19共4兲, 799–838.
Goel, R., and Chopra, A. K. 共2001兲 “Improved direct displacement-based
C共T兲 ⫽ pseudo-elastic spectrum, Eqs. 共3兲 and 共4兲;
design procedure for performance-based seismic design of structures.”
Ce ⫽ elastic seismic coefficient;
Proc., Structures 2001—A Structural Engineering Odyssey, Structural
共Ce兲global ⫽ normalized peak global elastic earthquake Engineering Institute of ASCE, Reston, Va., Sec. 22, Chap. 4.
resisting force; Gulkan, P., and Sozen, M. 共1974兲. “Inelastic response of reinforced con-
Ce ⫽ mean elastic seismic coefficient; crete structures to earthquakes motions.” ACI J., 71共12兲, 604–610.
Cy ⫽ seismic coefficient; yield strength coefficient; Hanson, R. D., and Soon, T. T. 共2001兲, “Seismic design with supplemen-
inelastic spectra; tal energy dissipation devices.” MNO-8, Earthquake Engineering
Cy ⫽ mean seismic coefficient; Research Institute, Oakland, Calif.
Fe ⫽ peak global elastic earthquake resisting force; Hudson, D. E. 共1972兲. “Local distribution of strong earthquake ground
g ⫽ gravitational acceleration; motion.” Bull. Seismol. Soc. Am., 62共6兲, 1765–1786.
Hwang, H., Chiu, Y. H., Chen, W. Y., and Shih, B. J. 共2004兲. “Analysis of
m ⫽ mass of the structure;
damage to steel gas pipelines caused by ground shaking effects during
n共T兲 ⫽ coefficient for constructing inelastic spectra, the Chi-Chi, Taiwan, Earthquake.” Earthquake Spectra, 20共4兲,
Eqs. 共3兲 and 共5兲; 1095–1110.
PGA ⫽ mean peak ground acceleration; International Code Council 共ICC兲. 共2003兲. “International Building Code
R␮¯ ⫽ modifying factor or reduction factor for mean 2003.”
ductility demand ␮ ¯; Iwan, W. D. 共1980兲. “Estimating inelastic response spectra from elastic
T ⫽ elastic vibration period; spectra.” Earthquake Eng. Struct. Dyn., 24共12兲, 1623–1643.
umax ⫽ maximum inelastic displacement; Iwan, W. D. 共2002兲. “The use of equivalent linearization in performance
ūmax ⫽ mean maximum inelastic displacement; based engineering.” Proc., Int. Conf. on Advances and New Chal-
uy ⫽ yield displacement; lenges in Earthquake Engineering Research, Institute of Engineering
␮ ⫽ response ductility demand; Mechanics, China Seismological Bureau, Harbin, China, 15–17.
Jarenprasert, S. 共2005兲. “An inelastic design approach for asymmetric
␮¯ ⫽ mean response ductility demand;
structure-foundation systems.” Ph.D. dissertation, Carnegie Mellon
␮t ⫽ target response ductility demand; and Univ., Pittsburgh.
␰ ⫽ critical damping ratio. Kayen, R. E., and Mitchell, J. K. 共1997兲. “Assessment of liquefaction
potential during earthquakes by Arias Intensity.” J. Geotech. Geoen-

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / AUGUST 2006 / 1291

 viron. Eng., 123共12兲, 1162–1174. Ramirez, O. M., Constatinou, M. C., Whittaker, A. S., Kircher, C. A., and
Kowalski, M., Priestley, M. J. N., and McRae, G. A. 共1995兲. Chrysostomou, C. Z. 共2002兲. “Elastic and inelastic seismic response
“Displacement-based design of RC bridge columns in seismic of buildings with damping systems.” Earthquake Spectra, 8共3兲,
regions.” Earthquake Eng. Struct. Dyn., 24共12兲, 1623–1643. 531–547.
Kwan, W.-P., and Billington, S. L. 共2003兲. “Influence of hysteretic behav- Riddell, R., Garcia, J. E., and Garces, E. 共2002兲. “Inelastic deformation
ior on equivalent period and damping of structural systems.” J. Struct. response of SDOF systems subjected to earthquakes.” Earthquake
Eng., 129共5兲, 576–585. Eng. Struct. Dyn., 31共4兲, 515–538.
Lin, Y. Y., and Chang, K. C. 共2003兲. “Study on damping reduction factor Riddell, R., and Newmark, N. M. 共1979兲. “Statistical analysis of the
for buildings under earthquake ground motions.” J. Struct. Eng., response of nonlinear systems subjected to earthquakes.” Structural
129共2兲, 206–214. Research Series No., 468, Dept. of Civil Engineering, University of
Miranda E. 共1993兲. “Site-dependent strength reduction factors.” J. Struct. Illinois-Urbana.
Eng., 119共12兲, 2503–3519. Rosenblueth, E., and Herrera, I. 共1964兲. “On a kind of hysteretic damp-
Miranda, E. 共2000兲. “Inelastic displacement ratios for structures on firm ing.” J. Eng. Mech. Div., Am. Soc. Civ. Eng., 90共2兲, 37–48.
sites.” J. Struct. Eng., 126共10兲, 1150–1159. Ruiz-García, J., and Miranda, E. 共2003兲. “Inelastic displacement ratios for
Miranda, E. 共2001兲. “Estimation of inelastic deformation demands of evaluation of existing structures.” Earthquake Eng. Struct. Dyn.,
SDOF systems.” J. Struct. Eng., 127共9兲, 1005–1012. 32共8兲, 1237–1258.
Miranda, E., and Bertero, V. V. 共1994兲. “Strength reduction factors for
Skinner, I. R., Robinson, W. H., and McVerry, G. H. 共1993兲. An intro-
earthquake-resistant design.” Earthquake Spectra, 10共2兲, 357–375.
duction to seismic isolation, Wiley, N.Y.
Miranda, E., and Ruiz-García, J. 共2002兲. “Evaluation of approximate
methods to estimate maximum inelastic displacement demands.” Tiwari, A. K., and Gupta, V. K. 共2000兲. “Scaling of ductility and damage-
Earthquake Eng. Struct. Dyn., 32共4兲, 539–560. based strength reduction factors for horizontal motions.” Earthquake
Moehle, J. P. 共1992兲. “Displacement-base design of RC structures Eng. Struct. Dyn., 29, 969–987.
subjected to earthquakes.” Earthquake Spectra, 8共3兲, 403–428. Veletsos, A. S., and Newmark, N. M. 共1960兲. “Effect of inelastic behavior
Nassar, A., and Krawinkler, H. 共1991兲. “Seismic demands of SDOF and on the response of simple systems to earthquake motions.” Proc., 2nd
MDOF systems.” Rep. No. 95, The John A. Blume Earthquake World Conf. on Earthquake Engineering, Science Council of Japan,
Engineering Center, Stanford Univ., Calif. Tokyo, Japan, Vol. 2, 895–912.
Newmark, N. M., and Hall, W. J. 共1982兲. “Earthquake Spectra and Vidic, T., Fajfar, P., and Fischinger, M. 共1994兲. “Consistent inelastic de-
Design.” Earthquake Engineering Research Institute, Berkeley, Calif. sign spectra: Strength and displacement.” Earthquake Eng. Struct.
Ordaz, M., and Pérez-Rocha, L. E. 共1998兲. “Estimation of strength- Dyn., 23共5兲, 507–521.
reduction factors for elastoplastic systems: A new approach.” Earth- Wu, J., and Hanson, R. D. 共1989兲. “Study of inelastic spectra with high
quake Eng. Struct. Dyn., 27共9兲, 889–901. damping.” J. Struct. Eng., 115共6兲, 1412–1431.

1292 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / AUGUST 2006