Soil Dynamics and Earthquake Engineering 22 (2002) 611624

www.elsevier.com/locate/soildyn

Viscous damping formulation and high frequency motion propagation

in non-linear site response analysis
Youssef M.A. Hashash*, Duhee Park
Department of Civil and Environmental Engineering, University of Illinois at Urbana-Champaign, 205 N. Mathews Avenue, Urbana, IL 61801, USA
Accepted 1 January 2002

Abstract
Non-linear time domain site response analysis is widely used in evaluating local soil effects on propagated ground motion. This approach
has generally provided good estimates of field behavior at longer periods but has shortcomings at relatively shorter periods. Viscous damping
is commonly employed in the equation of motion to capture damping at very small strains and employs an approximation of Rayleigh
damping using the first natural mode only. This paper introduces a new formulation for the viscous damping using the full Rayleigh damping.
The new formulation represents more accurately wave propagation for soil columns greater than 50 m thick and improves non-linear site
response analysis at shorter periods. The proposed formulation allows the use of frequency dependent viscous damping. Several examples,
including a field case history at Treasure Island, California, demonstrate the significant improvement in computed surface response using the
new formulation. q 2002 Elsevier Science Ltd. All rights reserved.
Keywords: Site response; Viscous damping; Deep deposits; Non-linear analysis; Amplification

1. Introduction
One-dimensional site response analysis is used to solve the
problem of vertical propagation of horizontal shear waves (SH
waves) through a horizontally layered soil deposit. Horizontal
soil layer behavior is approximated as a KelvinVoigt solid
whereby elastic shear moduli and viscous damping characterize soil properties. Solution of wave propagation equations
is performed in the frequency or time domain (TD).
Seed, Idriss and co-workers introduced the equivalent
linear approximation method to capture non-linear cyclic
response of soil. For a given ground motion time series (T.S.
also referred to as time history) and an initial estimate of
modulus and damping values, an effective shear strain (equal
to about 65% of peak strain) is computed for a given soil layer.
Modulus degradation and damping curves are then used to
obtain revised values of shear modulus and damping. The
solution is performed in frequency domain (FD) and an
iterative scheme is required to arrive at a converged solution
(e.g. SHAKE , Ref. [1]). This approach provides results that
compare well with field measurements and is widely used in
engineering practice. More recently, Sugito et al. [2] and
* Corresponding author. Tel.: 1-217-333-6986; fax: 1-217-265-8041.
E-mail addresses: hashash@uicu.edu (Y.M.A. Hashash), dpark1@uiuc.
edu (D. Park).

Assimaki et al. [3] extended the equivalent linear approach to

include frequency and pressure dependence of soil properties.
Assimaki et al. [3] suggest that it is appropriate to assume soil
damping to be frequency dependent to truly represent nonlinear soil response in a FD analysis.
The equivalent linear approach is computationally easy
to use and implement but remains an approximation of nonlinear cyclic response of soils. Non-linear site response
analysis is employed by integrating the equation of motion
in TD. A non-linear constitutive relation is used to represent
the hysteretic behavior of soil during cyclic loading. The
simplest constitutive relations use a model relating shear
stress to shear strain, whereby the backbone curve is
represented by a hyperbolic function. Strain dependent
modulus degradation curves are used to define the backbone
curve. The Masing criteria [4] and extended Masing criteria
[5,6] define unloading reloading criteria and behavior
under general cyclic loading. Lee and Finn [7] developed
a one-dimensional seismic response analysis program using
the hyperbolic model. Matasovic [8] and Matasovic and
Vucetic [9] further extended the model with a modification
of the hyperbolic equation. Plasticity models have also been
used to represent cyclic soil behavior. For example, Borja et
al. [10] used a bounding surface plasticity model to
represent cyclic soil response at the Lotung Site in Taiwan.
Hashash and Park [11] introduced an extension of the

0267-7261/02/$ - see front matter q 2002 Elsevier Science Ltd. All rights reserved.
PII: S 0 2 6 7 - 7 2 6 1 ( 0 2 ) 0 0 0 4 2 - 8

612

Y.M.A. Hashash, D. Park / Soil Dynamics and Earthquake Engineering 22 (2002) 611624

modified hyperbolic model to capture the dependence of

modulus degradation and damping curves on confining
pressure.
A problem commonly noted in non-linear site response
analysis is that while it provides good estimate at relatively
long periods, the computed ground response underestimates
measured response at shorter periods [12].

C aR M bR K

2. Numerical implementation of non-linear onedimensional wave propagation analysis

In non-linear analysis, the following dynamic equation of
motion is solved:
M{u} C{_u} K{u} 2M{I}ug

than 1024 1022%) with practically no damping, which can

cause unrealistic resonance during wave propagation. These
models incorporate additional damping to the dynamic
equation in the form of the [C] matrix, as shown in Eq. (1).
Similarly, the model by Borja et al. [10] uses the viscous
damping matrix. The [C] matrix is derived from a
combination of the mass matrix and the stiffness matrix
[14]:

where [M], mass matrix; [C], viscous damping matrix; [K],

stiffness matrix; {u}; vector of nodal relative acceleration;
{u}; vector of nodal relative velocities; and {u}, vector of
nodal relative displacements. u g is the acceleration at the
base of the soil column and {I} is the unit vector. The [M],
[C] and [K] matrices are assembled using the incremental
properties of the soil layers. The properties are obtained
from a constitutive model that describes the cyclic behavior
of soil. The dynamic equilibrium equation, Eq. (1), is solved
numerically at each time step using the Newmark [13] b
method. The geologic column is discretized into individual
layers using a multi-degree-of freedom lumped parameter
model shown in Fig. 1.
Each individual layer i is represented by a corresponding
mass, non-linear spring, and a dashpot for viscous damping.
Lumping half the mass of each of two consecutive layers at
their common boundary forms the mass matrix. The
stiffness matrix is updated at each time increment to
incorporate non-linearity of the soil.
The geologic material (soil or rock) is represented either
as a linear elastic material with constant value of damping or
using a non-linear constitutive model such as the pressure
dependent modified hyperbolic model described by Hashash
and Park [11]. The base of the soil column can be modeled
as either an infinitely stiff or a visco-elastic half space.

3. Limitation of current viscous damping formulation

In a non-linear soil model, soil damping is captured through
hysteretic loadingunloading cycles in the soil model. The use
of the damping matrix [C] may become unnecessary but is
commonly used as a mathematical convenience or to include
damping at very small strains where response of many
constitutive models is nearly linear elastic.
Hysteretic damping of the soil model defined by Hashash
and Park [11], as well other models (e.g. Ref. [8]), can
capture damping at strains larger than 1024 1022%,
depending on the values of material properties. However,
the hyperbolic model is nearly linear at small strains (less

The damping matrix is assumed in current formulations to

be only stiffness proportional since the value of aR M is
small compared to bR K: Small strain viscous damping
effects are assumed proportional only to the stiffness of the
soil layers. This is further simplified to:
C bR K
where bR 2j/v1 and v1 is the frequency of the first
natural mode of the soil column.
The viscous damping matrix for a multi-layered soil is
expressed as [11]:
2
3
j1 K1
2j1 K1
7
2
26
C ji Ki  6
2j1 K1 j1 K1 j2 K2 2j2 K2 7
4
5
v
v
2j2 K2

3
where v is natural circular frequency of the first natural
mode and ji is the equivalent damping ratio for layer i at
small strains. The viscous damping matrix is dependent on
the first natural mode of the soil column and the soil column
stiffness, which are derived from the shear wave velocity
profile of the soil column. [C] is commonly taken as
independent of strain level and the effect of hysteretic
damping induced by non-linear soil behavior can be
separated from (but added to) viscous damping.
The value of the equivalent damping ratio j is obtained
from the damping ratio curves at small strains. A constant
small strain viscous damping is used in some non-linear
models with a recommended upper bound value of 1.5 4%
for most soils, independent of confining pressure [8,15].
Hashash and Park [11] propose a pressure dependent
equation for the viscous damping ratio j.
In order to assess the accuracy of the viscous damping
formulation approximation, a series of linear site response
analyses are conducted using four idealized soil columns 50,
100 and 500 m thick with constant stiffness and viscous
damping ratio profiles (1) shown in Fig. 2. The thick soil
columns with variable shear wave velocity and viscous
damping are representative of conditions in the Mississippi
Embayment in the Central US (New Madrid Seismic Zone).
The analyses compare linear TD wave propagation analysis
with linear FD wave propagation analysis. The FD analysis
represents the correct analysis as the solution of the wave
equations can be derived in closed form (e.g. Ref. [16]). Fig.
3 shows the computed surface response for a harmonic input

Y.M.A. Hashash, D. Park / Soil Dynamics and Earthquake Engineering 22 (2002) 611624

613

Fig. 1. Multi-degree-of freedom lumped parameter model representation of horizontally layered soil deposit shaken at the base by a vertically propagating
horizontal shear wave. The model is used in the solution of the dynamic equation of motion in TD.

motion with soil thickness up to 500 m. TD analyses provide

identical results to FD analysis for the analyses using zero
viscous damping.
TD analysis with viscous damping ratio of 1% gives
results similar to FD analysis for the 50 m soil column.
However, for 100 and 500 m soil columns the TD analysis
gives a response lower than FD analysis. The quality of the
computed response deteriorates with increasing soil column
thickness. The approximation of the viscous damping
matrix in TD analysis may be acceptable for soil columns
less than about 50 m thick and when the contribution of the
viscous damping is very small.
The simplified damping formulation in Eq. (3) introduces
excessive damping in the TD analysis that increases with

increasing column thickness. The contribution of higher

modes is small for relatively short soil columns but may
become important for deeper soil columns and when
propagating high frequency motion. The simplified damping formulation depends only on the first mode of the
deposit and is proportional to the stiffness matrix. If only
stiffness proportional damping is used [17,18], then the
effective damping ratio being used for higher modes is:

jn

bR vn
v
j n
2
v1

This implies that the effective damping ratio is increasing at

higher natural modes. This would explain the underestimate
of surface ground motion for TD analysis shown in Fig. 3.

Fig. 2. Shear wave velocity and viscous damping profiles used in analyses. The variable profile properties are representative of conditions encountered in the
Mississippi Embayment, Central US. Bedrock shear wave velocity is 2700 m/s.

614

Y.M.A. Hashash, D. Park / Soil Dynamics and Earthquake Engineering 22 (2002) 611624

Fig. 3. Computed surface ground motion, linear frequency and TD site response, Vs profile (1). Linear TD analysis uses first natural mode approximation only
of viscous damping formulation. Harmonic input motion, amplitude 0.3 g, period 0.2 s, duration 1 s.

4. Proposed extension of viscous damping formulation

In the original damping formulation proposed by
Rayleigh and Lindsay [14], also in Clough and Penzien
[17], Chopra [18], aR and bR coefficients of Eq. (2) can be
computed using two significant natural modes m and n:
"
#( ) "
#
jm
aR
1 1/vm vm

5
2 1/vn vn
jn
bR
This matrix can be solved for aR and bR:
!
vm jn 2 vn jm
aR 2vm vn
v2m 2 v2n
!
vm jm 2 vn jn
bR 2
v2n 2 v2n
If the damping ratio j is frequency independent then:
!
!
vm vn
1
aR 2 j
b R 2j
vm vn
vm vn

large value of viscous damping matrix. Therefore, vn 0

should be included to represent the contribution of higher
modes.
When choosing higher modes, the mass matrix component will counter-balance part of the contribution of the
stiffness matrix component. As higher modes are used, aR
increases and bR decreases. Contribution of the mass matrix
cannot be ignored as using only the stiffness proportional
damping will result in different damping ratio for the
corresponding modes as shown in Eq. (4). The Rayleigh
damping formulation using two significant modes has been
incorporated for example in the two-dimensional finite
element program QUAD4M by Hudson et al. [19]. For a
multi-layered soil with frequency independent damping
ratio, Eq. (2) can be expanded as follows:

C 2

vm vn
vm vn

2
! j1 M 1
6
6
4

7
7
5

j2 M 2

Eq. (3) is obtained from Eqs. (5) (7) by assuming m is the

first natural mode and vn 0; implying that the second
relevant mode occurs at zero circular frequency. This is
acceptable for short soil columns where only the first mode
dominates. For thicker columns, such an assumption will
filter out high frequency components due to the resulting

j1 K 1

6
1
6 2j K
vm vn 4 1 1

2j1 K1

j1 K1 j2 K2
2j2 K2

3
7
2j2 K2 7
5

with frequency dependent damping ratio [3], Eq. (2) can be

Y.M.A. Hashash, D. Park / Soil Dynamics and Earthquake Engineering 22 (2002) 611624

615

expanded as follows:

C

2vm vn
v2m 2 v2n

2
! vm j1n 2 vn j1m M1
6
6
4

3
7
72
5
v2m 2 v2n

vm j2n 2 vn j2m M2

vm j1m 2 vn j1n K1

6
6
4 2vm j1m 2 vn j1n K1

2vm j1m 2 vn j1n K1

kvm j1m 2 vn j1n K1 vm j2m 2 vn j2n K2 l
2vm j2m 2 vn j2n K2

The viscous damping matrix is therefore dependent on

stiffness, mass and the natural modes of the soil column. In
most current applications, [C] is taken as strain independent
and has a constant value throughout an analysis. The natural
modes and the soil column stiffness are derived from the
shear wave velocity profile of the soil column. In a nonlinear analysis, the mass matrix remains constant but the
stiffness matrix is related to the strain level in the soil
column. The natural periods or frequencies vary with the
stiffness variation of the soil column. In the Netwon-b
method, Eq. (1) is integrated or linearized over a given time
increment. All matrix quantities including the [C] matrix
should correspond to soil properties at that given time
increment and strain level. The [C] matrix has to be updated
to accurately represent the viscous damping ratio (j)
property of the soil layers. Therefore, the [C] matrix is
strain dependent and its strain dependent components

3
7
2vm j2m 2 vn j2n K2 7
5

including frequencies of natural modes and stiffness matrix

are updated at each time increment.
This formulation of the damping matrix is implemented
in the non-linear site response analysis program DEEPSOIL ,
developed by Hashash and Park [11]. The proposed viscous
damping formulation is necessary to improve the accuracy
of the solution of wave equations in TD. In addition, the
formulation allows for the use of frequency dependent
viscous damping ratio.
The Rayleigh viscous damping formulation represents an
approximate solution and has some important features and
limitations [17,18]. The use of the Rayleigh damping
formulation results in an effective frequency dependent
damping as shown in Fig. 4 even when using Eq. (8). In the
figure, the first mode is assumed to occur at 1.0 s and the
higher mode at 0.1 s with corresponding damping ratio of
1%. Fig. 4 shows that for 0.1 , T , 1.0 s the resulting
damping ratio is less than 1% while at T . 1 s or T , 0.1 s
the resulting damping ratio increases significantly. Therefore, the user has to carefully choose the relevant modes to
capture ground motion response in the desired period/
frequency range. The choice is significant for the frequencies higher than the higher mode used in the Rayleigh
damping. At frequencies higher than the frequency of the
higher mode, ground motion content can be filtered out.
However, this may not be significant from an engineering
point of view if the natural modes in the Rayleigh damping
cover the range of frequencies of interest. The following
sections illustrate the significance of the new viscous
damping formulation.

5. Validation of new viscous damping formulation

Fig. 4. Variation of viscous damping as a function of period and frequency

using Rayleigh damping formulation (after Clough and Penzien [17] and
Chopra [18]).

A series of analyses are presented to validate the new

viscous damping formulation and evaluate its impact on
non-linear site response analysis. The analyses use a range
of idealized as well as representative input motions and soil
profiles. Four typical soil columns, 50, 100, 500 and 1000 m
deep, are used in the analyses as shown in Fig. 2. Two
typical shear wave velocity and viscous damping profiles
are used. The non-linear material properties used are those
published in Ref. [11].

616

Y.M.A. Hashash, D. Park / Soil Dynamics and Earthquake Engineering 22 (2002) 611624

Fig. 5. Computed surface ground motion, linear frequency and TD site response, Vs and viscous damping profiles (1). Linear TD analysis uses first natural mode
approximation as well as proposed full Rayleigh viscous formulation. Harmonic input motion, amplitude 0.3 g, period 0.2 s, duration 1 s.

Harmonic input motion, constant shear wave velocity

and viscous damping profiles (1), linear analysis. In this set
of analyses the results of linear TD wave propagation
analysis (DEEPSOIL ) using a harmonic input motion with a
duration t 1 s, period T 0.2 s, and amplitude 0.3 g
are compared to those obtained from linear FD wave
propagation analysis. Fig. 5(a) shows that for a 50 m profile
the TD analysis using the first natural mode only
(conventional viscous damping formulation) matches the
result of the FD analysis. For soil profiles greater than 50 m
in thickness the use of TD analysis with first mode
approximation significantly underestimates the surface
response. For a 100 m profile, Fig. 5(b), TD first mode
analysis gives lower response than FD analysis. However,
TD analysis using first and second modes in the proposed
viscous damping formulation, Eq. (9), matches that of the
FD analysis. For the 500 m profile, Fig. 5(c), TD analysis
using first mode only as well as first and second modes
underestimates the surface response compared to FD
analysis. A good match is obtained when the first natural

mode and the second mode at T 0.2 s (corresponding to

the period of the ground motion) are used in the TD
analysis.
Recorded input ground motion, shear wave velocity and
viscous damping profiles (2) representative of the Mississippi Embayment, linear analysis. In this series of
analyses, the shear wave velocity and viscous damping
profiles representative of the conditions in the Mississippi
Embayment, profiles (2) in Fig. 2, are used. A transient
ground motion recording from Hector Mine earthquake
(1999) in California with a peak ground acceleration
PGA 0.0073 g is used as the input motion. Computed
5% damped surface response spectra obtained from linear
TD and FD analyses are plotted in Fig. 6. Computed surface
ground motions for t 10 15 s are plotted in Fig. 7. The
analyses show that with increasing soil column thickness the
use of TD analysis with first natural mode (conventional
formulation) results in an increasing underestimate of
surface motion.
The computed surface response spectra from FD and TD

Y.M.A. Hashash, D. Park / Soil Dynamics and Earthquake Engineering 22 (2002) 611624

617

Fig. 6. Computed 5% damped surface response spectra, linear frequency and TD site response, Vs and viscous damping profiles (2). Linear TD analysis uses
first natural mode approximation as well as proposed full Rayleigh viscous damping formulation. Input motion: Hector mine earthquake, PGA ,
duration 18 s.

first mode analyses compare well for the 50 m column, Fig.

6(a). The match between the two analyses can be improved
by using TD analysis with first and second modes. For the
100 m profile, Fig. 6(b), a good match is obtained using first
and second modes in TD analysis. For the 500 m column,
Fig. 6(c), a good match is obtained when using first and fifth

modes. For the 1000 m column, Fig. 6(d), match is obtained

when using first and eighth modes.
Fig. 7 shows that the use of the full Rayleigh viscous
damping formulation improves the results from the TD
analysis, but does not provide an exact match of the results
of the FD analysis.

Fig. 7. Computed surface ground motion, linear frequency and TD site response, Vs and viscous damping profiles (2). Linear TD analysis uses first natural mode
approximation as well as proposed full Rayleigh viscous damping formulation. Input motion: Hector mine earthquake, PGA 0.007 g, duration 18 s.

618

Y.M.A. Hashash, D. Park / Soil Dynamics and Earthquake Engineering 22 (2002) 611624

Fig. 8. Influence of confining pressure on modulus degradation and damping ratio curves in DEEPSOIL non-linear model used for modeling of site response in
the Mississippi Embayment. Data from Laird and Stokoe [20] shown for comparison.

Synthetic input ground motion, shear wave velocity and

viscous damping profiles (2) representative of the Mississippi Embayment, non-linear analysis. The analyses
provided so far focus on the performance of TD analysis
whereby soil response is assumed linear, i.e. soil at a given
depth has a constant stiffness (obtained from shear wave
velocity) and a constant viscous damping value. In nonlinear (TD) site response analysis, damping is primarily a
result of hysteretic soil response, and the contribution of the
viscous damping term may become relatively small.
In the model proposed by Hashash and Park [11] the
cyclic response of the soil is dependent on the in situ
confining pressure. Fig. 8 shows the modulus degradation
and total damping curves at a range of confining pressures.
The model was designed to fit the data from Laird and
Stokoe [20]. The total damping curves are the sum of
hysteretic damping from the non-linear soil model and the
viscous damping profile (2) shown in Fig. 2.
Fig. 9 12 show the surface ground motion and computed
5% damped surface response spectra for non-linear TD

analyses using the program DEEPSOIL . Ground motion input

used in the analyses are synthetic ground motions generated
using the program SMSIM [21] for M 5 R 20 km
(PGA 0.063 g) and M 7, R 20 km (PGA 0.59 g)
with input parameters appropriate for the New Madrid
Seismic Zone. The analyses for a given soil profile are
conducted in two steps, (a) a linear TD analysis, with
viscous damping only, is performed to select the relevant
modes (always consisting of the first mode and a higher
mode) to match a similar FD analysis within the range of
frequencies of interest, and (b) a non-linear analysis is then
performed with viscous damping using the selected modes
from the previous step. Step one analyses show that for a
given soil profile (column thickness, shear wave velocity
and viscous damping profile) the choice of relevant modes
that provides the correct linear response is not very sensitive
to the input ground motion within the range of frequencies
of engineering interest.
Figs. 9 and 10 show comparisons of the computed nonlinear response for the four soil columns using the

Y.M.A. Hashash, D. Park / Soil Dynamics and Earthquake Engineering 22 (2002) 611624

619

Fig. 9. Computed surface ground motion, non-linear TD site response, Vs and viscous damping profiles (2), non-linear soil properties. Non-linear TD analysis
uses first natural mode approximation as well as proposed full Rayleigh viscous damping formulation. Synthetic input motion, M 5, R 20, New Madrid
Seismic Zone parameters, PGA 0.063 g, duration 4 s.

conventional viscous damping formulation (first mode only)

and the proposed viscous damping (first and a higher mode
as described in the earlier paragraph). For the 100 m profile,
the contribution of the higher modes in non-linear analysis
is small. For the 500 and 1000 m profile the contribution is
significant for periods less than 1 s. Use of the conventional

viscous damping formulation will filter out a significant

portion of the high frequency content of the ground motion.
The analysis results in Figs. 11 and 12 correspond to
higher input motion (PGA 0.59 g) whereby the contribution of the non-linear model and hysteretic damping is
also higher relative to the viscous damping component. The

Fig. 10. Computed 5% damped surface response spectra, non-linear TD site response, Vs and viscous damping profiles (2), non-linear soil properties. Nonlinear TD analysis uses first natural mode approximation as well as proposed full Rayleigh viscous damping formulation. Synthetic input motion, M 5,
R 20, New Madrid Seismic Zone parameters, PGA 0.063 g, duration 4 s.

620

Y.M.A. Hashash, D. Park / Soil Dynamics and Earthquake Engineering 22 (2002) 611624

Fig. 11. Computed surface ground motion, non-linear TD site response, Vs and viscous damping profiles (2), non-linear soil properties. Non-linear TD analysis
uses first natural mode approximation as well as proposed full Rayleigh viscous damping formulation. Synthetic input motion, M 7, R 20, New Madrid
Seismic Zone parameters, PGA 0.59 g, duration 17 s.

results still show that for deeper soil columns the use of the
conventional viscous damping formulation will filter out
important components of ground motion at high frequencies/short periods compared to the proposed viscous
damping formulation. Fig. 12(a) includes an additional

analysis for the 50 m column using frequency dependent

damping whereby the damping ratio of the second modes is
taken as half the damping ratio of the first mode. The figure
shows that the computed response is higher for periods less
than 0.1 s.

Fig. 12. Computed 5% damped surface response spectra, non-linear TD site response, Vs and viscous damping profiles (2), non-linear soil properties. Nonlinear TD analysis uses first natural mode approximation as well as proposed full Rayleigh viscous damping formulation. Synthetic input motion, M 7,
R 20, New Madrid Seismic Zone parameters, PGA 0.59 g, duration 17 s.

Y.M.A. Hashash, D. Park / Soil Dynamics and Earthquake Engineering 22 (2002) 611624

621

Fig. 13. Computed surface Fourier spectra, non-linear TD site response, Vs and viscous damping profiles (2), non-linear soil properties. Non-linear TD analysis
uses proposed full Rayleigh viscous damping formulation with and without the update of the [C] matrix. Synthetic input motion, M 7, R 20, New Madrid
Seismic Zone parameters, PGA 0.59 g, duration 17 s.

6. Use of constant versus variable [C] matrix in nonlinear analysis

In the proposed damping formulation, Eq. (9), the [C]
matrix is updated at every time step. This implementation is
in contrast to the common implementation where [C] is
constant and based on the initial soil properties. All the
examples presented above use the updated formulation of
the [C] matrix. The same analyses were also performed
using a constant form of [C] dependent on the initial soil
properties only. The analysis results using variable and
constant [C] were very similar for most cases. The variable
[C] analyses give slightly higher response than the constant
[C] analyses. The difference was more noticeable for the
analyses with the largest ground motion input (M 7
analyses of Fig. 12). The main difference is in the high
frequency range of the ground motion and is best seen in
plots of the Fourier amplitude of computed surface ground
motion in Fig. 13. This difference is a result of the
significant non-linear effects and reduction in stiffness
experienced by the soil.

7. Application of new viscous damping formulation to

case history of Treasure Island and Yerba Buena Island
recordings
During the 1989 Loma Prieta Earthquake in Northern
California ground motion recordings were obtained on fill
material underlain by sediments at Treasure Island and on
rock at adjacent Yerba Buena Island. Several site response
studies using these recordings were made using the

equivalent linear analysis [22,23] and the non-linear

analysis [8,24] methods. In these studies, the computed
surface response spectra were in general agreement with
measured spectra. However, several of the analyses had
some difficulty in capturing the recorded response in the
short period/high frequency range.
A similar site response analysis using the Yerba BuenaTreasure Islands recordings is made using the non-linear site
response analysis program DEEPSOIL . The analysis is
presented to further illustrate the significance of viscous
damping formulation on computed site response. The soil
shear wave velocity profile used, Fig. 14, is based on data
from Gibbs et al. [25] and Pass [26]. Shear wave velocity of
the rock is taken as 2700 m/s. Non-linear hyperbolic model
parameters were obtained from modulus degradation and
damping curves published in Ref. [27]. Fig. 14 shows the
modulus degradation and damping curves obtained from the
calibrated modified hyperbolic model. The recording at
Yerba Buena Island is used as the input motion at the base of
the column. The site response analysis was conducted by
first obtaining a best estimate of soil parameters using
available data without any attempt to match model results to
recorded motions. Fig. 15 shows plots of the computed and
recorded response spectra of the E W and N S components. For both motion components the analysis results
using the first mode (conventional) viscous damping
formulation significantly underestimates ground motion
response at periods less than 0.5 s.
For the E W motion component using the first and
second mode (proposed) viscous damping formulation
significantly improves ground motion response at periods
below 0.5 s and captures the high frequency (low period)

622

Y.M.A. Hashash, D. Park / Soil Dynamics and Earthquake Engineering 22 (2002) 611624

Fig. 14. Soil profile and soil properties used in the non-linear analysis of the Treasure Island case history.

peaks in the response spectra. The use of the new

formulation also improves the response at the highest
peak of the response spectrum. Use of higher modes
(first and eighth natural modes) does not result in
significant improvement of the results and is indicative of
the convergence of the solution at higher modes.
For the N S motion component the proposed (first
and second natural modes) viscous damping formulation
results in a dramatic improvement of computed ground
motion response at periods below 0.5 s compared to
conventional viscous damping formulation. The results
nearly match computed ground response and capture the
peaks in this low period/high frequency range. At longer
periods, the computed results underestimate recorded
motion. This result was also reported in other studies and
is attributed by Finn et al. [24] to ground motion
incoherency between the Yerba Buena Island and
Treasure Island recordings.
Fig. 15(c) and (d) also shows the influence of
frequency dependent formulation on computed surface
response spectra. The frequency dependent formulation
produces slightly higher response in the short period
range. However, it does not significantly alter the overall
response.
The case history shows that the new damping formulation significantly improves computed ground motion
response, especially at short periods, for a soil column
less than 100 m thick.

8. Summary and conclusions

The use of the full form of the Rayleigh damping to
represent viscous damping significantly improves the
performance of non-linear site response analysis in TD.
The proposed formulation addresses a long-standing
problem whereby non-linear site response performance
appeared to underestimate ground motion response at short
periods/high frequencies. The proposed viscous damping
formulation suggests that in addition to the first mode of the
soil column, higher modes have an important contribution to
the viscous damping component. The use of the new
formulation ensures that when the soil behavior is linear the
computed ground motion response in TD analysis is similar
to that computed in the FD. The new formulation allows the
option of using a frequency dependent viscous damping
component in non-linear analysis.
The use of the proposed viscous damping procedure in
non-linear site response analysis consists of two steps:
(a) Viscous damping modes identification. For the selected
soil column perform a linear TD analysis using the
column shear wave velocity profile and viscous
damping component profile only. The two relevant
natural modes are identified in the viscous damping
formulation such that the computed response matches a
similar linear analysis in FD over the frequency range
of interest. One of the modes always corresponds to the

Y.M.A. Hashash, D. Park / Soil Dynamics and Earthquake Engineering 22 (2002) 611624

623

Fig. 15. Computed 5% damped surface response spectra for the Treasure Island case history. Non-linear TD analysis uses first natural mode approximation as
well as proposed full Rayleigh viscous damping formulation.

first natural mode of the column. The other relevant

mode corresponds to the second or a higher natural
mode.
(b) Non-linear site response analysis. A non-linear
analysis is then performed with viscous damping
formulation using the two selected modes from the
previous step.
Examples of site response calculations using the
proposed formulation show that for soil columns greater
than 50 m thick ground response in the low period/high
frequency range is higher than what would be obtained
using conventional viscous damping formulation. Analysis
of site response using Yerba-Buena-Treasure Island recordings demonstrate the significant improvement achieved in
computed surface response using the new viscous damping
formulation in non-linear analysis.

Acknowledgments
This work was supported primarily by the Earthquake
Engineering Research Centers Program of the National
Science Foundation under Award Number EEC-9701785;

the Mid-America Earthquake Center. The authors gratefully

acknowledge this support. All opinions expressed in this
paper are solely those of the authors.

References
[1] Schnabel PB, Lysmer JL, Seed HB. SHAKE : a computer program for
earthquake response analysis of horizontally layered sites. Berkeley,
CA: Engineering Research Center; 1972.
[2] Sugito M, Goda H, Masuda T. Frequency dependent equi-linearized
technique for seismic response analysis of multi-layered ground.
Doboku Gakkai Rombun-Hokokushu/Proc Japan Soc Civil Engng
1994;493(3-2):4958.
[3] Assimaki D, Kausel E, Whittle AJ. Model for dynamic shear modulus
and damping for granular soils. J Geotech Geoenviron Engng 2000;
126(10):859 69.
[4] Masing G. Eignespannungen und Verfestigung beim Messing. Second
International Congress on Applied Mechanics, Zurich, Switzerland;
1926.
[5] Pyke RM. Nonlinear soil models for irregular cyclic loadings.
J Geotech Engng Div 1979;105(GT6):71526.
[6] Vucetic M. Normalized behavior of clay under irregular cyclic
loading. Can Geotech J 1990;27:29 46.
[7] Lee MK, Finn WDL. DESRA-2, Dynamic effective stress response
analysis of soil deposits with energy transmitting boundary including
assessment of liquefaction potential. Soil mechanics series no. 36,

624

[8]

[9]

[10]

[11]

[12]
[13]
[14]
[15]
[16]

[17]
[18]

[19]

Y.M.A. Hashash, D. Park / Soil Dynamics and Earthquake Engineering 22 (2002) 611624
Vancouver, Canada: Department of Civil Engineering, University of
British Columbia; 1978.
Matasovic N. Seismic response of composite horizontally-layered soil
deposits. PhD Thesis, University of California, Los Angeles; 1993.
p. xxix, 452 leaves.
Matasovic N, Vucetic M. Seismic response of soil deposits composed
of fully-saturated clay and sand layers. First International Conference
on Earthquake Geotechnical Engineering, Tokyo, Japan; 1995.
Borja RI, Chao HY, Montans FJ, Lin CH. Nonlinear ground response
at Lotung LSST site. J Geotech Geoenviron Engng 1999;125(3):187
97.
Hashash YMA, Park D. Non-linear one-dimensional seismic ground
motion propagation in the Mississippi embayment. Engng Geol 2001;
62(13):185206.
Idriss IM. Personal communications; 2000.
Newmark NM. A method of computation for structural dynamics.
J Engng Mech Div 1959;85:6794.
Rayleigh JWS, Lindsay RB. The theory of sound, 1st American ed.
New York: Dover Publications; 1945.
Lanzo G, Vucetic M. Effect of soil plasticity on damping ratio at small
cyclic strains. Soils Foundations 1999;39(4):121 41.
Kramer SL. Geotechnical earthquake engineering. Prentice-Hall
international series in civil engineering and engineering mechanics,
Upper Saddle River, NJ: Prentice Hall; 1996.
Clough RW, Penzien J. Dynamics of structures, 2nd ed. New York:
McGraw-Hill; 1993.
Chopra AK. Dynamics of structures: theory and applications to
earthquake engineering. Prentice-Hall international series in civil
engineering and engineering mechanics, Englewood Cliffs, NJ:
Prentice Hall; 1995.
Hudson M, Idriss IM, Beikae M. University of California Davis,
Center for Geotechnical Modeling, and National Science Foundation
(US). QUAD4M : a computer program to evaluate the seismic response

[20]

[21]

[22]

[23]

[24]

[25]

[26]

[27]

of soil structures using finite element procedures and incorporating a

compliant base. Center for Geotechnical Modeling, Department of
Civil and Environmental Engineering University of California Davis:
Davis California; 1994.
Laird JP, Stokoe KH. Dynamic properties of remolded and
undisturbed soil samples test at high confining pressure. Electric
Power Research Institute; 1993.
Boore DM. SMSIM Fortran programs for simulating ground motions
from earthquakes: Version 1.87. Users manual, US Geological
Survey; 2000. p. 73.
Seed RB, Dickenson SE, Riemer MF, Bray JD, Sitar N, Mitchell JK,
Idriss IM, Kayen RE, Kropp A, Harder LF, Power MS. Preliminary
report on the principal geotechnical aspects of the October 17, 1989.
Loma Prieta Earthquake; 1990.
Hryciw RD, Rollins KM, Homolka M, Shewbridge SE, McHood M.
Soil amplification at Treasure Island during the Loma Prieta
earthquake. Second International Conference on Recent Advances
in Geotechnical Earthquake Engineering and Soil Dynamics, St Louis,
MO; 1991.
Finn WDL, Ventura CE, Wu G. Analysis of ground motions at
Treasure Island site during the 1989 Loma Prieta earthquake. Soil
Dynamics Earthquake Engng 1993;12:383 90.
Gibbs JF, Fumal TE, Boore DM, Joyner WB. Seismic velocities and
geologic logs from borehole measurements at seven strong-motion
stations that recorded the 1989 Loma Prieta earthquake. USGS: Menlo
Park 1992;139.
Pass DG. Soil characterization of the deep accelerometer site at
Treasure Island, San Francisco, California. MS Thesis in Civil
Engineering, University of New Hampshire; 1994.
Hwang SK, Stokoe KH. Dynamic properties of undisturbed soil
samples from Treasure Island, California. Geotechnical Engineering
Center, Civil Engineering Department, University of Texas at Austin;
1993.