You are on page 1of 22

Bulletin of the Seismological Society of America. Vol. 65, No. 5, pp. 1315-1336.

October 1975

C A L C U L A T I O N OF N O N L I N E A R G R O U N D RESPONSE IN E A R T H Q U A K E S BY WILLIAM B. JOYNER AND ALBERTT. F. CHEN

ABSTRACT

A method is presented for calculating the seismic response of a system of horizontal soil layers. The essential element of the method is a rheological model suggested by Iwan which takes account of the nonlinear hysteretic behavior of soils and has considerable flexibility for incorporating laboratory results on the dynamic behavior of soils. Finite rigidity is allowed in the underlying elastic medium, permitting energy to be radiated back into the underlying medium. Three alternate ways of integrating the equations of motion are compared, an implicit technique, an explicit technique, and integration along characteristics. An example is set up for comparing the different methods of integration and for comparing the nonlinear solution with a solution based on the widely used equivalent linear assumption. The example consists of a 200-m section of firm alluvium excited at its base by the N21E component of the Taft accelerogram multiplied by four to produce a peak acceleration of 0.7 g and a peak velocity of 67 cm/sec. The three techniques of integration give very similar results, but integration along characteristics has the advantage of avoiding spurious highfrequency oscillations in the acceleration time history at the surface. For the chosen example, which has a thick soil column and a strong input motion, the equivalent linear solution underestimates the intensity of surface motion for periods between 0.1 and 0.6 sec by factors exceeding two. The discrepancies, however, wonld probably be less for input motion of lower intensity. At longer periods the equivalent linear solution is in essential agreement with the nonlinear solution. For the same example both solutions show that, compared to a site with rock at the surface, motion at the surface of the soil is amplified for periods longer than 1.5 sec by as much as a factor of two. At shorter periods the amplitude is reduced.
INTRODUCTION

The problem we are concerned with is basically a very simple one. We postulate a system of horizontal soil layers bounded above by the free surface and below by a semiinfinite elastic medium representing the bedrock. We further postulate a vertically incident shear wave in the underlying medium, and we ask the question, "How will the overlying layers respond and in particular what will be the motion of a point on the free surface ?" This is a classical problem in engineering seismology. There is some disagreement concerning the range of applicability of the solution (Hudson, 1972; Newmark et al. 1972), but no one would deny the importance of solving this relatively simple problem correctly. The problem was solved by Kanai (1952) some years ago for the case of layers with linear viscoelasticity of the Voigt type. When we are dealing with input motion sufficiently intense to cause severe damage to structures, however, we cannot assume linear behavior over the entire range of strain. To do so would imply stresses many times greater than the strength of typical materials as measured in the laboratory. To circumvent this difficulty the method in common use currently is what we shall
1315

1316

WILLIAM B. JOYNER AND ALBERT T. F. CHEN

refer to as the "equivalent linear method" (Idriss and Seed, 1968; Schnabel et al., 1972). It is based on the assumption that the response can be approximated by the response of a linear model whose properties are chosen in accord with the average strain that occurs at each depth in the model during excitation. The calculation is iterative. First, trial values for average strain are chosen, then soil properties are determined in accordance with the trial values of strain, and finally the response of the model is ~calculated. If the calculated strains differ by too much from the trial values, the cycle is repeated. Idriss and Seed (1968) tested the validity of the equivalent linear assumption by comparing the equivalent linear solution for a test case with the solution obtained using a bilinear stress-strain relationship. Recently, a number of workers have described other methods that utilize nonlinear stress-strain relationships. The Ramberg-Osgood relationship was used by Streeter et al. (1974), by Constantopoulos (1973), and by Faccioli et al. (1973). An elasto-plastic relationship was used by Papastamatiou (written communication). A different method, described by Chen and Joyner (1974) and elaborated here, is based on a rheological model proposed by Iwan (1967). This model leads to a very simple and efficient method of calculation and offers considerable flexibility for incorporating laboratory data on soil behavior. The following section of this report describes the Iwan model and its implementation. After that, a boundary condition (Papastamatiou, written communication) is presented which takes account of finite rigidity in the elastic substratum. Next, three alternate ways are described for integrating the equations of motion. An example is then developed with material properties assigned as a function of depth in accordance with the laboratory results of Hardin and Drnevich (1972a, 1972b). The example is first used to compare the different techniques of integration. Then, the example is used to compare results from the equivalent linear method with those from the nonlinear method. Iwan (1967) extended his model to three dimensions. Application of the extended Iwan model to ground response calculations in two dimensions is described in a separate paper now in preparation.
CONSTITUTIVE RELATION

The basic requirement for a solution to the problem is a constitutive relation--in simple terms we need a rule that will tell each soil element how to find its way around the stress-strain plane. For this purpose we adopted a model (Figure 1) proposed by Iwan (1967). It is composed of simple linear springs and Coulomb friction elements arranged as shown. The friction elements remain locked until the stress on them exceeds the yield stress Yv Then, they yield, and the stress across them during yielding is equal to the yield stress. Generally, the yield stress of the first element Y1 is set to zero. By appropriate specification of the spring constants G~ and the yield stresses Y~, we can model a very broad range of material behavior as dictated by laboratory experiments. The accuracy of the modeling depends upon the number N of elements used. We have found it possible to use large numbers economically. For our typical problem N is 50; we have gone as high as 100 without unreasonable increases in computing time. There is one model of the kind diagrammed in Figure 1 for each soil layer in the system. In practice, the model is not described by specifying the Gi directly. It turns out that it is more efficient to work in terms of the tangent modulus of the whole model. We index the elements in order of increasing yield stress. At any given time, all the elements up to a certain index number will be yielding and all those above will not. (A proof of

CALCULATION OF NONLINEAR GROUND RESPONSE IN EARTHQUAKES

1317

this proposition is given in Appendix A.) We denote by the letter m the index of the element with the largest yield stress of all the elements that are yielding. Then the tangent modulus Sm of the model depends only upon m. It is related to the G i by the equation

i=1

We evaluate the Sm from the initial loading curve in simple shear. Alternatively, we could use the locus of the end points of hysteresis loops formed in cyclic loading tests at different peak strain levels. The two curves are the same insofar as the model is concerned. The second curve is used by Hardin and Drnevich (1972a, 1972b) in describing their experimental results on the stress-strain behavior of soils.

G2

Gi

GN

....
'(2 Yi

....7 2
YN

Fro. 1. Model used for constitutive relationship. Model consists of simple elastic springs with spring constants G~ and Coulomb friction elements with yield stresses Y~.

Given the initial loading curve, we proceed by selecting a set of yield stresses Y,, (m = 1, N + 1). The Yi are chosen to cover the range of stress the system is expected to encounter and are distributed so that the initial loading curve can be faithfully represented. From the initial loading curve a set of shear strain values e m (m = 1, N + 1) is obtained corresponding to the stress values Ym. The tangent moduli S,, are then given simply by
Y m + l - ]Zm

em+l --ern

To simplify the computations, the stress and strain are normalized in the manner used by Hardin and Drnevich (1972b). Stress is normalized by multiplying by 1/T. . . .

1318

WILLIAM

B . J O Y N E R A N D A L B E R T T . F. C H E N

where Zma x is the shear stress at failure, and strain is normalized by multiplying by Gm,x/z . . . . where Gin,x is the low-strain modulus. The stress-strain curve normalized in this way has a stress limit of 1.0 for high strain and a slope at the origin of 1.0. For the examples shown later in this report, the same normalized initial loading curve and therefore the same set of normalized S,, and Y,, are used for all of the soil layers in the model. The differences in soil behavior from one layer to another result from differences in the values of Zm,x and G~, x assigned to the different layers. The following set of values was adopted for Y~

Y~=O
Ym
----

0.025(0"5) 6-'n

2<m<6 7<m<44
m-44

Y~ = 0.025(m- 5)

Yrn
I.O-

1"0-0'025(0"5)

45 < rn < 51

0,0

-1.0

-4.o

-2'o

'

o'o
STRAIN

2'.o

'

41o

REDUCED

FIG. 2. Hysteresis loops for the model shown in Figure l. Stress and strain are scaled so that the maximum stress and the low-strain modulus are unity. A hyperbolic initial loading curve (Hardin and Drnevich, 1972b) was used and normalized strain was expressed in terms of normalized stress by the equation

Ym
em - -

1 - Y,,"

The basic method, however, does not depend on the hyperbolic relationship; a purely empirical stress-strain curve derived from laboratory measurements could be used just as well. The type of hysteresis loops that this model produces is shown in Figure 2, which illustrates the behavior of a single soil layer subject to cyclic loading of increasing amplitude. The "reduced stress" and "reduced strain" plotted in Figure 2 are the stress and strain normalized in the way used by Hardin and Drnevich (1972b). Denoting the upper branch of a closed hysteresis loop as the "reloading curve," it can readily be shown that for any hysteresis loops generated by the Iwan model the reloading curve is simply the initial loading curve with its origin translated and its scale expanded by a factor of two both vertically and horizontally. This rule for describing hysteresis loops is called Masing's criterion (Masing, 1926; for a further discussion of Masingtype systems see Newmark and Rosenblueth, 1971, p. 163, 345-346). The Iwan model can be used to represent, to any desired accuracy, the behavior of any material whose

CALCULATION OF NONLINEAR GROUND RESPONSE IN EARTHQUAKES

1319

hysteresis loops satisfy the Masing criterion and do not depend on the number of cycles of loading. There is some evidence that soils conform at least approximately to the Masing criterion. Hardin and Drnevich (1972b), commenting on experimentally recorded hysteresis loops for a wide variety of soils, stated that the slope of the stress-strain curves immediately after load reversal was approximately equal to the low-strain modulus, independent of the strain amplitude of the loop. Even if a soil does not exactly meet the Masing criterion, its behavior might still be approximately represented by an Iwan model. The best approach would probably be to select the hysteresis loop corresponding to the strain amplitude for which faithful representation was most desired. The reloading curve from that loop could then be used to evaluate the constants of the model. Even if the shape of the loop changed with the number of loading cycles, the Iwan model might still be used, provided the changes were not too drastic. In evaluating the constants of the model, one would simply use the loop generated after a certain number of cycles. The number would be chosen to be representative of the number expected during the postulated earthquake. It should be noted that the rheological model described here has no viscous damping, and as a result the stress depends on the strain (and strain history) but not on the strain rate. The energy dissipation per cycle, therefore, does not depend upon the frequency. There would be no difficulty in adding a dashpot in parallel with the model. As a matter of fact, that option was allowed in one of our computer programs (Chen and Joyner, 1974). We note, however, that the experimental data of Hardin and Drnevich (1972a, p. 620) indicate very little or no effect of frequency on damping in soils for frequencies between about 0.1 and 25 Hz, which covers the range of interest in engineering seismology. The Masing criterion is satisfied by the elasto-plastic stress-strain relationship used by Papastamatiou (written communication), by the bilinear relationship used by Idriss and Seed (1968), and by the Ramberg-Osgood relationship used by Streeter et al. (1974), Constantopoulous (1973), and Faccioli et al. (1973). All of these relationships could therefore be represented by a model of the Iwan type.
BOUNDARY CONDITION

In solving soil response problems of this kind it is common to prescribe the motion at the base of the soil column. This is an approximation that allows no energy to be radiated back into the underlying medium. The accuracy of the approximation depends upon the contrast in rigidity across the boundary and upon the energy dissipation within the soil column. If the energy dissipation is small, multiple reflection within the soil column will give rise to resonances whose strength will depend upon the rigidity contrast. Prescribing the motion at the base of the soil column is the equivalent of assuming infinite rigidity in the underlying medium. To take account of the finite rigidity of the underlying medium we use a method suggested by Papastamatiou (written communication). That method allows us to satisfy the boundary conditions exactly for a wave vertically incident on the boundary from below. The approach is similar to that of Lysmer and Kuhlemeyer (1969). We have assumed a vertically incident plane shear wave in the underlying elastic medium. That assumption allows us to obtain an expression for the shear stress in the underlying medium at the boundary in terms of the particle velocity of the incident wave and the particle velocity on the boundary. Particle displacement for the incident wave is a function of depth x and time t given by
u~ = v , ( x + v , t )

1320

WILLIAM B. JOYNERAND ALBERTT. F. CHEN

where vs is the shear velocity in the underlying medium. For the reflected wave, particle displacement is given by
UR = g . ( x - v / ) .

The shear stress at the boundary is

(av,+ev.
where/@ is the rigidity of the underlying medium. c~Ux ~X
-

1 V s
v,

where V~is the particle velocity for the incident wave and VR.

OX

Vs

The particle velocity on the boundary is the sum of the velocity of the incident and reflected waves,
v,, = vR + v , .

Solving for VR and substituting in the equation for shear stress gives

ZB =

/~e ( 2 I11 -- VB)

Us

or, equivalently,
z~ = p ~ G ( 2 V I - VB)

(1)

where pe is the density of the underlying medium. TECHNIQUES OF INTEGRATION Given a constitutive relation and a boundary condition at the base of the system, we need only integrate the equations of motion step by step in time to obtain the motions at the surface. We have experimented with three different techniques for integrating the equations of motion, an implicit technique, an explicit technique, and integration along characteristics. The implicit technique is described in detail elsewhere (Chen and Joyner, 1974). It is based on the approach used by Idriss and Seed (1967) for analyzing a system of soil layers with bilinear stress-strain relationships. Briefly, the soil profile is divided into layers, and the mass of each layer is lumped equally at the top and bottom. The equations of motion are written in the form of a set of simultaneous equations which is solved once for each time step. The rheological model used with the implicit technique is an Iwan model of the type shown in Figure 1 in parallel with a dashpot to give viscous damping. The dashpot was included to permit comparison with the earlier work of Idriss and Seed (1967).

CALCULATION OF NONLINEAR GROUND RESPONSE IN EARTHQUAKES

1321

For the example presented in this report, we use a dashpot coefficient small enough so that the effect of the dashpot is negligible. We do this because of the evidence cited previously indicating that damping in soils is not significantly dependent on frequency. The boundary condition allowing finite rigidity in the substratum was not implemented in the computer program for the implicit technique although it could probably be done without difficulty. In the explicit technique, also, the mass of the soil layers is lumped at the top and bottom of the layers. The state of the system is described in terms of the particle velocity Vi at the top of each layer i and the normalized shear stress si in each layer i. The layers are numbered from the top down. The computations proceed step by step in space from the bottom up and step by step in time. A simplified outline of the sequence of computations is as follows: 1. At each layer boundary the particle velocity Vi is known at time t, and in each layer the normalized shear stress s~ is known at time ( t - At/2), where At is the time increment. 2. The change in normalized shear strain Ae~ in layer i is computed for a time interval At centered about t, using the following formula

Aei = ( V i + l - Vi)kiAt/Axi
where Ax~ is the thickness of layer i and k~ is a normalizing constant which is equal to the ratio Gmax/Zmaxfor the layer. 3. The values of Ae~ are used in conjunction with the rheological model, in a manner described subsequently, to update the normalized shear stress to time (t At/2). 4. The values of normalized stress in the layers above and below a layer boundary at time (t + At/2) are used to update the particle velocity at the boundary to time (t + At) according to the following formula, based on Newton's second law

Vi+l(t+ At) = Vi+a(t ) + [('~max)i+lSi+l (Tmax)iSi] At/mi+ 1.


-

where mi+ 1 is the total mass per unit area lumped at the top of layer (i+ 1). The factors of Zmaxconvert normalized stress to actual stress. In step 4 special treatment is required for the layer boundaries at the top and bottom of the system. At the top the stress above the boundary is zero and the particle velocity is updated by

Vl(t+ At) = Vl(t ) + (Zm,x)lS 1At/m~.


If there is a total of N layers in the system, the index of the bottom boundary is (N+ 1). We can write

Vu+l(t + At) = Vu+l(t) + Au+l(t + At/2)At


where Au+ a is the acceleration at the boundary. Newton's second law gives

(2)

AN +l(t + At/2) = [z.(t + At/Z) -- ('Cmax)NS.~(t At/2)]/m N+1

(3)

where -c n is the shear stress in the underlying medium at the boundary. Using equation (1) we make the approximation zB(t+ At/2) ~ pEvs[2Vi(t+ A t ) - VN+I(t+ At)]. (4)

1322

WILLIAM B. JOYNER AND ALBERT T. F. CHEN

Substituting from equation (4) into equation (3) and from equation (3) into equation (2) and solving for VN+ 1(t + At) gives

VN+I(t+A t )

[1.0- 1.O/(pEvsAt/rnN+ 1 + 1.0)]. {2 V,( t + A t ) - [(ZmJNSN( t + At/2)


- VN+I(t)/(At/mN+I)]/(pEv~)}.

(5)

The approximation made in equation (4) leads to the form of equation (5) such that the limit of VN+a as (pEvs) goes to infinity is equal to 2VI, as it should be. Other approximations, although they may appear reasonable, do not lead to that limiting condition and result in instability for large (p~vs), in which the errors are magnified at each step in the computation. In step 3 we use the Iwan model to determine the stress changes from the strain changes. Each soil layer is represented by a model of the type shown in Figure 1. Throughout the computations we keep track of the stress in each of the springs of each model and the index m of the element with the highest yielding stress among all elements that are currently yielding. When we obtain the value of Ae~ for a layer, we first check to see if the direction of deformation has been reversed from the previous cycle. If it has, we set m equal to one. We then use the tangent modulus Sm to compute a trial value of Asi
As i = SmAei.

We then check to see if the new stress (si + Asi) will cause the (m + 1) element to yield. That will happen if

[s,+As,-H,.+,.,I--> Ym+l
where Hm+ 1, ~is the stress in the (m + 1) spring of layer i. If the (m + 1) element does yield, we set the stress sl to the value where yielding begins and correct the stress increment. We then subtract from Ae~ the strain increment corresponding to the corrected stress increment and increase the value of m by one. This process is repeated until we reach a value ofm such that the (m + 1) element does not yield. For integration along characteristics we followed the approach of Streeter et al. (1974). The Iwan model is used in the manner just described to keep track of the stressstrain history of each layer and provide a value of the tangent modulus for each layer at each time step. The tangent modulus is used to obtain the strain-dependent shear-wave velocity, which is required for integration along characteristics. The boundary condition allowing finite rigidity in the substratum is included.
EXAMPLE

For our example we chose the soil profile illustrated in Figure 3, representing a 200-m section of firm alluvium consisting predominantly of cohesive material. The key material parameters are the maximum shear stress ZmaX and the low-strain modulus Gma x. Both parameters are strongly dependent upon the state of effective stress prior to seismic excitation, which in turn depends upon depth. An attempt was made to assign realistic values to these parameters as functions of depth using the methods of Hardin and Drnevich (1972a, 1972b) with minor modifications. A density of 2.05 gm/cm 3 was assumed and the low-strain shear modulus was converted to shear velocity for plotting

CALCULATION OF NONLINEAR GROUND RESPONSE IN EARTHQUAKES

1323

in Figure 3. A past consolidation vertical stress of 2.94 bars was assumed. As a result, the material was overconsolidated above a depth of 29 m, causing kinks at that depth in the two curves of Figure 3. The material was assigned a plasticity index of 20 per cent, and complete water saturation was presumed at all depths. In the normally consolidated part of the section Zmax was assumed proportional to vertical effective stress, Zmax = CsPve where Pve is the vertical effective stress prior to seismic excitation. P~e can readily be calculated given the density and the depth. The coefficient C~ is evaluated on the assumption that no pore-pressure change occurs during shear using the equation

where ~ is the angle of drained shear resistance and Ko is the coefficient of earth pressure
VELOCITY (METERS 200 PER SECOND) 400

I00

300

500

50

DEPTH I00

(METERS)

150

200

2.0

4.0 STRESS

6.0 (BARS)

8.0

I0.0

FIo. 3. Dynamic properties for soil profile used in sample problem.

at rest. The equation is obtained from an equation of Hardin and Drnevich (1972b, p. 673) by assuming that the cohesion c for normally consolidated material is zero (Terzaghi and Peck, 1967, p. 112). Estimation on the basis of plasticity index gives a value of 31 for q5(Terzaghi and Peck, 1967, p. 112) and 0.55 for K o (Brooker and Ireland, 1965). Carrying out the computations gives a value of 0.33 for Cs. The assumption of no pore pressure change will tend to lead to an overestimate of the strength and thereby of the capacity of the material to transmit shear. Normally consolidated clays tend to contract with shear, increasing the pore pressure and reducing the strength. For the overconsolidated part of the section we assumed that we could represent the variation of'Cma x by an equation of the form
"Cma x = CsPve(OCR) T

1324

WILLIAM B. JOYNER AND ALBERT T. F. CHEN

where OCR is the overconsolidation ratio in terms of vertical effective stresses. Analysis of undrained strength data from Ladd and Edgers (1972) gave a value of 0.75 for T. Hardin and Black (1969) and Hardin and Drnevich (1972b) have shown that for a wide variety of undisturbed cohesive soils, and also sands, the low-strain shear modulus Gma x is given by an equation which can be rewritten in the form Gmax = C6P,nel/2(OCRrn)r where Pine is the mean effective stress, OCRm is the overconsolidation ratio in terms of mean effective stresses, C G is a constant depending on void ratio, and K depends on plasticity index. For a PI of 20 per cent K has a value of 0.18. It is more convenient in the present case to deal in terms of vertical stresses. Using K o values from Brooker and Ireland (1965) to convert from mean stress to vertical stress, it was found that the results from the preceding equation could be reproduced within a few percentage points over the range of 1.0 to 32 in OCR by the following equation Gmax = cmevea/2(OCR) Q where Q takes on a value of 0.28 for a PI of 20 per cent. A value of 0.9 10 6 (dynes/ cruZ) 1/2 was chosen for C m based on shear velocity measurements in a 180-m drill hole in alluvium in the San Francisco Bay Area (Warrick, 1974). COMPARISON OF INTEGRATIONTECHNIQUES Five different runs were made in comparing the three different techniques of integration. The soil column used was the one illustrated in Figure 3 and described in the preceding section. Infinite rigidity was assumed for the underlying medium, because the finite-rigidity boundary condition had not been incorporated into the computer program for the implicit technique. The input motion at the base was the first 20 sec of the N21E component of the Taft strong-motion record multiplied by a factor of four, giving a peak acceleration of 0.7 g and a peak velocity of 67 cm/sec. (The Taft accelerogram was recorded during the 1952 Kern County, California, earthquake.) Layer thicknesses and time increments for the five runs are given in Table 1. For integration along characteristics, once the time increment is chosen, the layer thicknesses are determined by the requirement that the travel time through a layer at the low-strain shear velocity be equal to the time increment. For run C1 a time increment of 0.01 sec was chosen in order to obtain good results for frequencies below about 10 Hz. This corresponds to a requirement of 10 points per wavelength for good resolution, a requirement suggested by the work of Boore (1972). The time increment was halved for run C2 in TABLE 1
VALUES OF AX AND At FOR METHODS USED Symbol Technique Ax (m) At (sec)

I E2 E1 C1 C2

Implicit Explicit Explicit Characteristics Characteristics

2.4-2.5 2.25 1.8-4.5 1.8-4.5 0.8-2.2

0.0025 0.005 0.01 0.01 0.005

CALCULATION OF NONLINEAR G R O U N D RESPONSE I N EARTHQUAKES

1325

order to show convergence. Run E1 was done by the explicit technique with the same time increment and layer thicknesses as were used for C1. Run E2 was done by the explicit technique using a constant layer thickness, and run I was done by the implicit technique. Layer thicknesses for runs E2 and I were chosen to give a frequency resolution comparable to that of C1 and El. Time increments for E2 and I were chosen to satisfy the requirements of stability. Stability with the implicit technique is discussed by Chen and Joyner (1974). In predicting stability for runs with the explicit technique, the stability criterion for the linear elastic case (Richtmyer and Morton, 1967, p. 263) was applied using the low-strain value of the velocity. No numerical evidence of instability was noted for runs in which the time increment satisfied that criterion.

IU
v

t-t3 O _3 la.t

TIME (SECONDS)

FIG. 4. Surface particle velocityby different techniques of integration. Surface particle velocity for the five runs is compared in Figure 4, and the agreement is quite good. Surface acceleration is compared in Figure 5. There the agreement is also good except for the high-frequency oscillations that appear in the explicit and implicit runs. These oscillations are discussed in greater detail by Chen and Joyner (1974). Their frequency depends upon the thickness of the layers into which the soil column is divided in the lumped mass procedure. The frequency can be made as high as desired by choosing the layer thicknesses sufficiently small. The oscillations can then be removed by applying a suitable digital low-pass filter. Removal by digital filtering is better than the use of viscous damping to suppress the oscillations, because it is easier to avoid distortion of the signal within the frequency band of interest. As mentioned previously,

1326

W I L L I A M B. JOYNER AND ALBERT T. F. CHEN

the damping characteristics of soils are essentially independent of frequency above 0.1 Hz indicating that viscous damping is not a valid representation of the material behavior. As a matter of fact, for many purposes there may be no need to remove the oscillations. Figure 6 shows the acceleration response spectra at five per cent damping for the five runs. The differences are small for periods greater than about 0.1 sec. We conclude that all three techniques give valid results. The run with the shortest computation time was El, which was done by the explicit technique with layer thicknesses chosen to give constant travel time at the low-strain velocity. Run C1, however, the comparable run done by integration along characteristics, required only 2 minutes on an IBM 370-155. Integration along characteristics may be preferred for many applications because it avoids the problem of the high-frequency oscillations.
I

,-,
z o
F--

E2

~:
b_l U.I (..) ~.) (12

E1

~
o'.oo e'.oo

C1

C2

4.0o

G'.oo

s'.oo

ib.oo

ib.oo

ib.oo

ib.oo

ib.oo

2b.oo

TIME

(SECONDS)

FIG. 5. Surface acceleration by different techniques of integration.


C O M P A R I S O N W I T H THE E Q U I V A L E N T L I N E A R M E T H O D

For comparing the results of the equivalent linear method with those of the nonlinear method, the same soil column (Figure 3) is used as before. The substratum is assigned a shear velocity of 2.0 km/sec and a density of 2.6 gm/cm 3. The input is the first 50 sec of the N21E component of the Taft strong-motion record multiplied by a factor of four, as before; so that the incident wave if incident on a free surface would produce a peak acceleration of 0.7 g and a peak velocity of 67 cm/sec. It should be noted that this example, with a thick soil column and strong input motion, was deliberately chosen to provide a severe test for the equivalent linear assumption. The nonlinear calculations were done by integration along characteristics. The equivalent linear calculations were done by a method similar to that described by Schnabel et al. (1972). Because of the high level of the input motion and the consequent large energy

CALCULATION OF NONLINEAR GROUND RESPONSE IN EARTHQUAKES


1.6.
m
co
I i

] 327

z
o F-y LIJ

1 .e,

-E 1

K
i t
B

II
t
A

___-~E2
0.8

Ld

-,

~.--

_J

\
/ ' | .... f~-.i

o o')

~ C 2 \C1
i i

O' 0

0.2

.#

O .6

0.8 SECONDS

1.0

I .e

NRTURRL PERIOD -

FIG. 6. Acceleration response at five per cent d a m p i n g for different techniques of integration.

"-

EOUIVRLENT

LINERR

v
"

z (23
I-.-t

NONL

IN E R R

(12 rY LLI __1 W (..3

INPUT

0,00

5,00

I0.00

15,00

9'0,00

25,00

30.00

35,00

40,00

q.5,00

50,00

TIME

(SECONDS)

FIG. 7. C o m p a r i s o n of surface acceleration for equivalent linear and n o n l i n e a r m e t h o d s .

1328

WILLIAM B. JOYNER AND ALBERT T. F. CHEN

dissipation in the soil column, special attention was required to the relationship between soil damping and the dissipation constant of the equivalent linear model. The parameters of the equivalent linear model as functions of strain were calculated from the properties of the Iwan model used for the nonlinear calculations, so that the only difference between the two sets of calculations is the equivalent linear assumption. Further description of the equivalent linear procedure is given in Appendix B. Figure 7 shows the input acceleration time history and compares the surface acceleration computed by the nonlinear method and the equivalent linear method. There are definite points of similarity, but it is clear that the equivalent linear approximation does not reproduce the short-period components of motion present in the nonlinear solution. Comparing the nonlinear solution with the input shows the effect of the postulated
"q
q

EQUI gRLENT L.INERR

NONLINERR
v

>I-o ._1 LLI >

INPUT

f3.00

5.00

10,00

15.00

20,00

25.00

30.00

35.00

q.o.00

q.5,00

50.00

TIME (SECONDS)
FIo. 8. Comparison of surface particle velocity for equivalent linear and nonlinear methods.

soil profile on ground motion. The peak acceleration is sharply reduced. The longer period components are amplified, however, and the overall effect may be a more damaging motion as will be illustrated subsequently. Figure 8 shows the corresponding velocity time histories for the same example. Comparing the nonlinear and equivalent linear solutions we see much better agreement, indicating that the equivalent linear approximation is adequate with respect to the longer period components that are dominant in the velocity time history. Comparing either of the solutions with the input shows clearly the amplifying effects of the soil profile for long-period motions. To illustrate the consequences of these results for structures we have computed response

CALCULATION OF NONLINEAR GROUND RESPONSE IN EARTHQUAKES

1329

spectra. Figure 9 shows the acceleration response at 5 per cent damping for the three motions between 0- and 3-sec period. The line represents the input, hexagons the nonlinear solution, and crosses the equivalent linear solution. It is clear that the equivalent linear method significantly underestimates the intensity of motion for periods between 0.1 and 0.6 sec in this case. This period range corresponds to buildings between one and about six stories. The importance is obvious. Comparing the input response with the nonlinear response in Figure 9, one might be tempted to conclude that for short-period structures, the motions on alluvium would be less damaging than on bedrock. Considerable caution is indicated here. For one thing, different soil materials, dense sands for example, might give more intense motions.
3,0
I

Z O
H

FEl: Ld Ld O C-) L~
I.--

2.0

jr
1.0
~ x ,~D~au~ ~ ,n ~ x O
^ ~

_J c~ (]:

0.5

t .0

t ,5
PERIOD -

2,0

2.5

3,0

NRTURRL

SECONDS

Fio. 9. Acceleration response at five per cent damping. The line represents the input, hexagons the nonlinear solution, and crosses the equivalent linear solution.

Possible lengthening of structural periods due to deformation beyond the linear range needs to be considered as well as the effects of duration not accounted for in the response spectrum. Consideration should also be given to the effects of ground deformation and ground failure. Figure 10 compares the relative velocity response spectra at 5 per cent damping for the range from 0- to 6-sec period. The results show that the equivalent linear approximation is adequate for the longer periods and that the soil site gives large amplification for longer periods. During the running of the nonlinear solution, we monitored the peak strain for each depth interval. The maximum was 6 x 10- 3 for the interval from 32 to 35 m. The comparative costs of the two methods is difficult to evaluate in the general case because it is possible to run the equivalent linear method using fewer layers, depending on the detail one wishes to represent in the soil profile. For the example presented here, however, using the same number of layers for both methods, the nonlinear solution required less than half as much computer time as the equivalent linear. So, we believe that in general it will be competitive, at least.

1330

WILLIAM B. JOYNER AND ALBERTT. F. CHEN CONCLUSIONS

The Iwan model leads to a simple and efficient method for calculating the seismic response of a system of horizontal soil layers. The method takes account of the nonlinear hysteretic behavior of soils and has considerable flexibility for incorporating laboratory results on the dynamic behavior of soils. Finite rigidity is allowed in the underlying elastic medium, permitting energy to be radiated back into the elastic medium. Integration by implicit or explicit techniques or integration along characteristics gives valid results. Integration along characteristics has the advantage of avoiding spurious highfrequency oscillations in the solution for the acceleration time history at the surface. Comparison with results obtained by the widely used equivalent linear assumption indicates that for a thick soil column and a high level of input motion the equivalent linear method may underestimate the short period components of surface motion by a factor of two or more. At longer periods (longer than 0.6 sec in the example) the two methods agree. Better agreement at short periods would be expected for lower levels of input motion. Comparative computer time requirements depend on the circumstances of the

d U3

300

O~

850
I )I-(J 0 -J IllJ bJ X X X

800

o o

o~

x
0 0 0

150
x x

100
I.tl "

50

~"

0.5

i .0

1.5

8.0

8.5

3.0
-

3.5

q.O

q.5

5.0

5.5

6.0

NATURAL PERIOD

SECONDS

Fla. 10. Relative velocity response at five per cent damping. The line represents the input, hexagons the nonlinear solution, and crosses the equivalent linear solution. particular problem. We believe that in many cases nonlinear methods will be competitive, at least, in this regard. In general, nonlinear methods are preferable because they have a more rigorous basis, but equivalent linear methods may be adequate for many purposes where accurate evaluation of response spectra for short periods is not required. Our results give an indication of the general nature of the effect one might expect for a thick soil column at a high level of input motion, compared to a site with bedrock at the surface. The model shows amplification by a factor of two or more at long periods and some reduction in amplitude at short periods. ACKNOWLEDGMENTS We are indebted to Harold W. Olsen for his advice on questions of soil dynamics and to Roger D. Borcherdt for assistance in the theory of linear viscoelasticity. David M. Boore and D. J. Andrews made helpful suggestions concerning numerical methods. William D. Stuart and Harold W. Olsen read the manuscript and made a number of valuable suggestions.

CALCULATION OF NONLINEAR GROUND RESPONSE IN EARTHQUAKES

1331

APPENDIX A

A Theorem Regarding the Rheological Model Given the rheological model illustrated in Figure 1, we prove the following theorem: I f element j is not yielding, then all elements i for which Y~ is greater than Yj are also not yielding. We denote the stress in the i spring by Hi and the stress in the i friction element by R~. Equilibrium requires that Hi+ Ri = H j + Rj = s where s is the stress for the model. The friction element i is fixed when (A1)

IR,I < Y,
and likewise forj. The friction element i is yielding to increasing stress when
Ri= Yi

and to decreasing stress when


Ri = - ~i

and likewise forj. We can now restate the theorem as follows If

Y i > Yj
and then

IR,I < Yi
From equation (A 1) we have
Ri = Rj + (Hj - Hi).

Since
]Rjl < Yj

we have

Ri < Y j + ( H j - H , )
and Ri > - Y j + ( H j - H , ) . It follows then that

R, < Yj+IH~-H,I

Ri >
So long as we have that

5-IHj=Hi[.

[Hi- H,I < (Yi- r~)


IR, I < r,.
So, we have proved a weak form of the theorem which states that

1332 if

WILLIAM B. JOYNER AND ALBERT T. F. CHEN

Y i > Yj, and

Igjl <
and

]Hi - Hil < ( Y , - Yj),


then IRi] < Yi. We now proceed to use the weak form of the theorem to prove that IHj-Hil can never exceed ( Y i - Yj), which completes the proof of the theorem in its original form. The initial value of (Hj-Hi) is zero. The quantity remains constant so long as both elements are fixed and also so long as both elements are yielding in either direction. So long as the absolute value remains less than ( Y i - Yj), the weak form of the theorem tells us that element i cannot yield while j is fixed. The only way, therefore, that the value of ( H j - Hi) can change is for elementj to yield with element i fixed. I f j is yielding under increasing stress and i is fixed, then ( H j - Hi) will increase. Since

R~=
substitution into equation (A1) gives

rj

Ri = (Hj-Hi)+ Yj. ( Y i - Yj), substitution into equation (A2) shows that


Ri= Yi

(A2)

As element j yields, the quantity ( H i - H i ) will increase, but when it reaches the value

and further increase of stress will cause element i to yield. With both elements yielding, the quantity ( H i - Hi) remains constant. If element j is yielding under decreasing stress and element i is fixed, then (Hi-Hi) will decrease. Since Rj = - Yj substitution into equation (A1) gives

Ri = ( H i - H i ) - Yr.
-

(A3)

As element j yields, the quantity (Hi-Hi) will decrease but when it reaches the value ( Y i - Yj), substitution into equation (A3) shows that

Ri = -- Yi
and further decrease of stress will cause element i to yield. As noted before, when both elements are yielding, the quantity ( H j - Hi) remains constant.
APPENDIX B

Equivalent Linear Procedure


The equivalent linear method for solving soil response problems is based on the assumption that a linear model will give a satisfactory approximation providing that the dynamic properties are chosen to accord with the average strain that occurs in the model during excitation. The method is iterative. Assumed values of average strain are used for the first run. After the run, the computed strain is compared to the assumed strain.

CALCULATION OF NONLINEAR GROUND RESPONSE IN EARTHQUAKES

1333

If necessary, adjustments are made and the process is repeated as many times as needed to obtain satisfactory agreement. The linear model we use is based on Kanai's (1952) solution to the horizontal layer problem, extended to general linear viscoelastic material. We postulate a system of N horizontal plane layers composed of homogeneous isotropic linear viscoelastic material all resting on an elastic half-space. We further postulate an incident plane S H wave in the half-space, propagating at an angle q5 with the vertical. For the computations presented in this paper q5 was taken to be zero, but the method was originally developed for arbitrary ~b. We use a Cartesian coordinate system, x, with the origin on the surface of the half-space and with an orientation such that x3 is vertical and the ray path for the incident S H wave lies in the x l - x 3 plane. Within each layer we look for solutions to the equation of motion in which displacement takes the form Ui = ui exp [itot] (B1)

where uj is a complex function of the spatial coordinates, to is angular frequency and t is time. The displacement Uj is complex, but when solutions for positive and negative to are added the result will be real. With this formulation the equation of motion for a homogeneous isotropic linear viscoelastic material is (Borcherdt, 1973)

2vj (k+#/3) VO+#V2Uj =


where
~U 1 ~U 2 ~U 3

(B2)

# = (ito/2) r~ k = (ito/3) r k p is the density and the functions r s and r k are the Fourier transforms of the relaxation functions characteristic of the shear and bulk behavior of the material. Substituting from equation (B 1) into equation (B2) gives (k + #/3) V0 + where
0 : 8ul 8/'/2 8~/3
# V 2 I , / j --~ -

(.O2pb/j

(B3)

a -i

"

The orientation of the coordinate system was chosen so that the displacement for the incident wave does not depend on x2. Since the geometric and material properties of the system do not depend on x2, there is no way that one value of x 2 could be distinguished from another. We theretbre assume that ~?uflc?x2 = 0 and consequently that c~O/c~x2 = 0. With that assumption the equation for the j = 2 component in (B3) is uncoupled from the other two and becomes
iJV2u = - ooZpu.

(B4)

The subscript is omitted in (B4) and in what follows, because we are concerned only with t h e j = 2 component.

1334

W I L L I A M B. J O Y N E R A N D ALBERT T. F. C H E N

Displacement for the incident wave in the underlying medium is represented by Ut = at exp (icot- ilxl - ibex3) where
l = cosin hi =

4)/Vs c o c o s 4/v~

and v~ is the shear velocity in the elastic half-space. The solution to equation (B4) in the nth layer is (u), = exp (-ilxl)[a . exp ( - i h n x 3 ) + b, exp (ihnx3)] where h, 2 = c o 2 p , / # , - I z. The first term in the brackets represents an upgoing wave and the second term a downgoing wave. In the general case, since/~, is complex, h, will be complex. At each boundary application of the conditions requiring continuity of stress and displacement leads to two relationships among the complex coefficients a, and b, for adjacent layers. With the assumption of a stress-free surface, these relationships are solved using a modification of Haskell's (1953) matrix method to give the coefficients a, and b, in terms ofa~. The Fourier coefficients at for a postulated input motion are calculated using the Fast Fourier Transform program H A R M from the IBM Scientific Subroutine Package. Given the at, the coefficients a, and b, are found. The motion at a point in the layers is then represented as a function of time by a Fourier series, which is evaluated using the H A R M program. For the purpose of comparing the equivalent linear with the nonlinear method, we determine the complex modulus # for each layer by requiring that the viscoelastic material show stress-strain behavior similar to that of the Iwan model. We define the damping ratio D by
D = w/(z~E;,)

(B5)

where zp is the peak stress and W is the energy per unit volume dissipated per cycle at the given peak strain level Ep. This corresponds to the definition of damping ratio commonly used in soils dynamics (e.g., Hardin and Drnevich, 1972a). Using elementary principles, Wand "cocan be calculated for an Iwan model given Ep. The hysteresis loops of viscoelastic material for sinusoidal loading are ellipses and, of course, are not identical to those of an Iwan model. We determine/~t//~, the ratio between the imaginary and real parts of the complex modulus, by requiring that the viscoelastic material have the same value of D as the Iwan model, realizing that in the viscoelastic material peak stress and peak strain will not occur at the same instant. In a viscoelastic material
w= ~
cycle

z~dt.

For sinusoidal loading


E = E . cos cot

z = Zp cos (cot+(5)

CALCULATION OF NONLINEAR GROUND RESPONSE IN EARTHQUAKES where 6 = arc t a n (/2~//zR). P e r f o r m i n g the integration and substituting into e q u a t i o n (B5) gives D = (sin 6)/2. W i t h some algebraic m a n i p u l a t i o n this expression can be solved to give /q/#R = 2D/(1 - 4D2) a/z. The d e t e r m i n a t i o n of/~R is straightforward. W e require t h a t

1335

f r o m which we easily o b t a i n
I~R = (zp/Ep)(1 + [Izr/pR]2) -1/2 .

In r u n n i n g the p r o b l e m described in the text, values o f E v were a s s u m e d at d e p t h s o f 16, 32, 64, 128, a n d 199 m a n d were i n t e r p o l a t e d in between. D u r i n g each iteration, strain was m o n i t o r e d at those depths. A strain value equal to 70 p e r cent o f the m a x i m u m absolute value was t a k e n for c o m p a r i s o n with the Ep assumed at the beginning o f the iteration. I f necessary, Ep was adjusted a n d a n o t h e r iteration m a d e . F o u r iterations were required for the e x a m p l e given in the text. The choice o f 70 p e r cent as the conversion factor f r o m m a x i m u m strain p e a k to average strain p e a k is a d m i t t e d l y s o m e w h a t arbitrary. W e believe, however, t h a t it is a reasonable choice a n d we d o u b t t h a t the results are very sensitive to the choice.

REFERENCES Boore, D. M. (1972). Finite difference methods for seismic-wave propagation in heterogeneous materials, in Methods in Computational Physics, Vol. 11, Seismology, Surface Waves and Earth Oscillations, Bruce A. Bolt, Editor, Academic Press, New York, 1-37. Borcherdt, R. D. (1973). Energy and plane waves in linear viscoelastic media, J. Geophys. Res. 78, 24422453. Brooker, E. W. and H. O. Ireland (1965). Earth pressures at rest related to stress history, Can. Geotech. J. 2~ 1-15. Chen, A. T. F. and W. B. Joyner (1974). Multiqinear analysis for ground motion studies of layered systems, Report No. USGS-GD-74-020, NTIS No. PB232-704/AS, Clearinghouse, Springfield, VA 22151. Constantopoulos, I. V. (1973). Amplification studies for a nonlinear hysteretic soil model, Research Report R73-46, Department of Civil Engineering, Massachusetts Institute of Technology, 204 pp. Faccioli, E., E. Santoyo V., and J. L. Le6n T. (1973). Microzonation criteria and seismic response studies for the city of Managua, Proc. Earthquake Eng. Res. Inst. Conf. Managua, Nicaragua, Earthquake of Dec. 23, 1972 1,271-291. Hardin, B. O. and W. L. Black (1969). Vibration modulus of normally consolidated clay (discussion), Proc. Am. Soc. Civil Eng.,J. Soil Mech. Found. Div. 95, 1531-1537. Hardin, B. O. and V. P. Drnevich (1972a). Shear modulus and damping in soils: Measurement and parameter effects, Proc. Am. Soc. Civil Eng., J. Soil Mech. Found. Div. 98, 603-624. Hardin, B. O. and V. P. Drnevich (1972b). Shear modulus and damping in soils: Design equations and curves, Proc. Am. Soc. Civil Eng., J. Soil Mech. Found. Div. 98, 667-692. Haskell, N. A. (1953). The dispersion of surface waves on multilayered media, Bull. Seism. Soc. Am. 43,
17-34.

Hudson, D. E. (1972). Strong motion seismology, Proc. Inter. Conf. Microzonation (Seattle, Washington), 1, 29-60. Idriss, I. M. and H. B. Seed (1967). Response of Horizontal Soil Layers During Earthquakes, Department of Civil Engineering, University of California, Berkeley.

1336

WILLIAM B. JOYNER AND ALBERT T. F. CHEN

Idriss, I. M. and H. B. Seed (1968). Seismic response of horizontal soil layers, Proc. Am. Soc. Civil Eng., J. Soil Mech. Found. Div. 94, 1003-1031. Iwan, W. D. (1967). On a class of models for the yielding behavior of continuous and composite systems, J. Appl. Mech. 34, 612-617. Kanai, K. (1952). Relation between the nature of surface layer and the amplitudes of earthquake motions, Bull. Earthquake Res. Inst., Tokyo Univ. 30, 31-37. Ladd, C. C. and L. Edgers (1972). Consolidated-undrained direct-simple shear tests on saturated clays, Research Report R72-82, Department of Civil Engineering, Massachusetts Institute of Technology, Cambridge, Massachusetts. Lysmer, J. and R. L. Kuhlemeyer (1969). Finite dynamic model for infinite media, Proc. Am. Soe. Civil Eng., J. Eng. Mech. Div. 95, 859-877. Masing, G. (1926). Eigenspannungen und Verfestigung beim Messing, Proe. Intern. Congr. Appl. Mech. 332-335. Newmark, N. M., A. R. Robinson, A. H.-S. Ang, L. A. Lopez, and W. J. Hall (1972). Methods for determining site characteristics, Proc. Intern. Conf. Microzonation, (Seattle, Washington) 1, 113-129. Newmark, N. M. and E. Rosenblueth (1971). Fundamentals of Earthquake Engineering, Prentice-Hall, Englewood Cliffs, N.J. Richtmyer, R. D. and K. W. Morton (1967). Difference Methods for Initial Value Problems, 2rid edition, Interscience, New York. Schnabel, P., H. B. Seed, and J. Lysmer (1972). Modification of seismograph records for effects of local soil conditions, Bull. Seism. Soc. Am. 62, 1649-1664. Streeter, V. L., E. B. Wylie, and F. E. Richart (1974). Soil motion computations by characteristics method, Proe. Am. Soc. Civil Eng., J. Geotech. Eng. Div. 100, 247-263. Terzaghi, K. and R.B. Peck (1967). SoilMechanics in EngineeringPraetice, 2nd edition, Wiley, New York. Warrick, R. E. (1974). Seismic investigation of a San Francisco Bay mud site, Bull. Seism. Soc. Am. 64, 375-385.
U.S. GEOLOGICALSURVEY 345 MIDDLEFIELDROAD MENLO PARK, CALIFORNIA94025

Manuscript received February 10, 1973

You might also like