SIMQKE-II Conditioned Earthquake Ground Motion Simulator

Users Manual, Version 2.1

Erik H. Vanmarcke Gordon A. Fenton Ernesto Heredia-Zavoni

Mar 19, 1999 Copyright c Princeton University 1996 All rights reserved.

SIMQKE-II Users Manual Abstract

Simqke-II generates an array of differing yet spatially correlated earthquake ground motions at an arbitrary set of points, optionally statistically compatible with known or prescribed motions at other locations. The simulated motions can be used for more realistic earthquake resistant design of extended structures such as bridges, power plants, dams, shopping malls, etc., where the input motions during an earthquake event are not expected to be identical at all points in contact with the ground. It can also be used for the post-analysis of earthquake damaged structures given recorded motions in the vicinity. Other potential applications include liquefaction analysis, slope stability evaluation, etc. Supplied with a target ground motion spectral density function, which may be evolutionary in nature, the program employs Covariance Matrix Decomposition in the frequency domain followed by Best Linear Unbiased Estimation and an inverse Fast Fourier Transform to efciently produce the nonstationary, spatially correlated, conditioned or unconditioned ground motions. The evolutionary nature of the motion is modeled by subdividing the motion into a sequence of time windows, within each of which a target spectral density function is specied. An averaging algorithm is applied to smoothly connect the sequential windows. The spatial correlation is based on the phase aligned motions and is conveniently specied through simple parameters supplied to one of a number of typical space-frequency correlation functions. Simqke-II can also act as a complementary product to the original Simqke program (see NISEE Computer Software for Earthquake Engineering, 1993). The original program has a number of pre-processing options related to target response spectra that are not currently part of the Simqke-II suite of capabilities. In particular, the original Simqke can be used to generate target spectral density functions from desired response spectra. These can then be used in Simqke-II to provide the simulated spatially variable ground motions.

SIMQKE-II Users Manual Table of Contents

Page 1. 2. 3. 4. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Theoretical Background Running the Program . . . . . . . . . . . . . . . . . . . . . . . . . . 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

Input Data Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

4.1 4.2 4.3 JCDAT: Job Control Data File . . . . . . . . . . . . . . . . . . . . GWDAT: Spectral Density Function Data File TMDAT: Known Time History Data File . . . . . . . . . . . . . . . .

. . . 9 . . . . . . . . . . . . . . . 12 . . 16 19 19 20 25

5. 6. 7. 8.

Output Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . System Specics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Example References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

SIMQKE-II Users Manual 1. Introduction

Large, spatially extended structures will experience different ground motions at different support points during an earthquake event. It is only in recent years that detailed studies of ground motion spatial variability have been undertaken (see, for example, Harichandran and Vanmarcke, 1986, and Boissi` eres and Vanmarcke, 1995) leading to estimates of typical space-frequency correlation functions. At the same time a number of ground motion generation methods have been developed, culminating in the computer program Simqke-II. This program is designed to take advantage of current progress in earthquake statistical estimation, providing the means to simulate realistic space-time elds of ground motion for use in structural analysis. The program operates in one of two modes; 1) conditional simulation: in this mode, generated ground motions are statistically compatible with, or conditioned by, recorded ground motions at nearby points. Simulated motions become increasingly similar to the recorded motions as they become more correlated, and conversely, become statistically independent as their correlation drops to zero. At the same time, simulated motions are properly correlated with one another. 2) unconditional simulation: in this mode the ground motions are simulated using only the user prescribed space-time statistics the simulated motions are still properly inter-correlated, but they are not conditioned on any recorded motions. In either mode, the task is to simulate ground motions at a set of spatial points where motions are unknown but desired. The ground motions are phase aligned, which is to say that the motions do not account for any possible travel time between points. This assumption corresponds to vertically propagating earthquake waves, shear or compressional. Only one component of motion may be generated at a time which implies that the program can be used to produce independent components of motion only. Point-wise cross-correlated motions can, in principle, be subsequently formed through linear combinations of the independent components. The program has been run on the following platforms; 1) DEC workstations running OSF/1 (DEC UNIX), 2) Sun workstations running Solaris, 3) HP workstations running HP-UX, 4) Linux using f2c and gcc compilers. 5) Windows 95/98/NT (under DOS command line window) Simqke-II is written in standard Fortran 77, except for a few extensions in the form of BSD type system calls (see the last Section for details) which are supported by virtually all Fortran compilers implemented on UNIX operating systems as well as by Digitals Visual Fortran for Win95/98/NT. 2. Theoretical Background

According to the spectral representation theorem, if the ground motion at a point i is a homogeneous mean-square continuous real time process Zi (t), then it can be expressed as a sum of independent sinusoidal processes as,

Zi(t) =

K 1 X k=0

[Aik cos(!k t) + Bik sin(!k t)];

(1)

SIMQKE-II Users Manual

at K discrete frequencies each separated by ! . The random coefcients Aik and Bik are mean zero, implying that the resultant process Zi (t) is also mean zero. A non-zero mean can always be added to the nal realizations. For a nite duration segment, the average of Zi (t) need not be equivalent to the prescribed mean, that is, Zi (t) is only asymptotically ergodic as t ! 1. This makes sense since one does not expect the average of earthquake ground motion over 1 millisecond to be zero, nor the average over 1 second. The rate at which the time average approaches the true mean depends on the spectral density function describing the process, in particular on the low frequency power content. From the point of view of simulation, it is convenient to discretize time and generate Zi (tj ) at the times tj = j t, j = 0; 1; : : :; K 1. In this case, the coefcients Aik and Bik are related to Zi (tj ) through the discrete Direct Fourier Transform (DFT),

Aik = K

K 1 1 X j =0

Zi(tj ) cos

kj ; K

Bik = K

K 1 1 X j =0

Zi(tj ) sin

kj K

(2)

where t = tf =(K 1), !k = k ! , ! = 2 =(K t), k = 0; 1; : : :; K 1, and tf is the time duration of the process Z (t). The basic concepts pertinent to simulation using Fourier transforms can be found in Fenton (1994). The following symmetry conditions about the Nyquist frequency, !K=2 = = t, apply to the Fourier coefcients Aik and Bik when the process Zi (tj ) is real,

Aik = Ai;K k ;

Bik = Bi;K

for

k = 1; 2; : : :; K=2

(3)

It can be shown (Fenton, 1990; Heredia-Zavoni, 1993) that by making use of Eq. (2) and the above symmetry conditions, the covariance Cij (!k ) = E Aik Ajk = E Bik Bjk (A and B are mutually independent) between coefcients at two spatial points i and xj can be written as follows,

where ij = i j is the relative position vector, G(! ) is the one-sided point spectral density function, and !k ( ij ) is the frequency-dependent spatial correlation function. Note that the above relationships implicitly assume that the ground motion is statistically homogeneous both in space and time. This is not a severe restriction since correlation structures with component scales larger than the space or time domains will result in apparently non-stationary realizations. Deterministic trends in the mean and/or variance can always be applied to the nal simulation in any case. Consider now the simulation of earthquake ground motion at a set of m target (unknown) points given that some motions have been recorded at a set of n = N m recording (known) points , where N is the total number of spatial locations under consideration. Using Eq. (4), the N N covariance matrix k = [Cij (!k )], i; j = 1; 2; : : :; n; : : :; n + m, for each Fourier frequency, !k , k = 0; 1; : : :; K=2, can be assembled and expressed as the symmetric matrix

!k ( ij )G(!k ) !; Cij (!k ) = > 1 4 f !k ( ij )G(!k ) + !K k ( ij )G(!K k )g : !k ( ij )G(!k ) !;

1 2

8 > <

x x r

r r

!;

for k = 0

for k = 1; 2; : : :; K=2 for k = K=2

(4)

x x

Ck = C C

"

C C CT C

(5)

where is the covariance matrix between known (recording) points, is the covariance matrix between unknown (target) points, and is the covariance matrix between known and unknown

SIMQKE-II Users Manual

points, all at frequency !k . Let s = f s ; s g denote the set of simulated Fourier coefcients at frequency !k , where the subsets s = fA1k ; A2k ; : : :; Ank g and s = fAn+1;k ; : : :; An+m;k g correspond to coefcients at known and unknown target points, respectively. The set of coefcients s can be dened similarly. Now, for admissible spatial correlation and spectral density functions, !k ( ij ) and G(!k ), the matrix k is positive and can be expressed as the product of a lower triangular matrix k and its transpose by means of a Cholesky decomposition (or an appropriate modication thereof in the case where k is singular):

A A

As and Bs can be generated from As = Lk Uk ; Bs = Lk Vk (7) T = fU ; U ; : : :; U g and V T = fV ; V ; : : :; V g are independent where each element of Uk k k Nk k k Nk k standard normal random variables. Under these denitions, As and Bs have the correct distributions, since, for example, E [As ] = 0, Cov [As ] = Lk fCov [Uk ]gLT k = Ck , since E [Uk ] = 0 and Cov [Uk ] = I , the identity matrix, etc. Thus, the generation of the correlated set of Fourier coefcients can proceed using K sets of independent random numbers and the Cholesky decomposition of Ck at each frequency !k , k = 0; 1; : : :K 1, to yield the correlated set of ground motions. To produce motions which are conditioned on the recordings at points x , a set of Best Linear Unbiased Estimators (BLUE) must be employed for each of the unknown points x (HerediaZavoni, 1993); As = C T C As ; Bs = C T C Bs ; (8) so that As and Bs are the best linear unbiased estimates of the simulated Fourier coefcients at the unknown target points. Similarly, given the actual Fourier coefcients at the known points (obtained by Fourier decomposition of the recordings), A and B , the BLUE coefcients at the
Assuming that Z (t) is normally distributed (Gaussian), then
1 2 1 2 1 1

Ck = Lk LT k

(6)

unknown points are given by

A = CT C A ;
1

B = CT C B ;
1

(9)

Finally, the conditioned Fourier coefcients may be obtained by combining all three forms;

Ac = A + As A B

As ;

Bc = B + Bs x

Bs

(10)

The use of the BLUE coefcients in Equation (AAI ) guarantees that the conditioned Fourier coefcients c and c have the proper covariance structure (Heredia-Zavoni, 1993). Conceptually, approaches one of the points , the last two terms in each equation cancel each other as a point leaving just the recorded Fourier coefcient. Thus, the conditioned random eld exactly matches the known motions at the known points and becomes increasingly random with distance from the known points (or, more precisely, with decreasing correlation with the known points). The Fourier coefcients of Eq. (10) are produced for each frequency !k , k = 0; 1; : : :; K=2. The remaining coefcients, up to frequency !K 1 , are then obtained using the symmetry relationships of Eq. (3). Once the entire suite of coefcients have been generated, an inverse FFT is applied to yield a set of (conditioned) ground motions at the unknown target points. The above concepts result in the following algorithm which has been implemented in Simqke-II; 1) For each frequency !k , k = 0; 1; : : :; K=2: - Assemble the frequency specic covariance matrix k as in Eqs (4) and (5).

SIMQKE-II Users Manual

- Simulate the sets of unconditioned Fourier coefcients s and Bs according to Eq. (7). If there are no known recordings skip to step (2). - Obtain from the recorded motions the known Fourier coefcients, and by means of an FFT as in Eq. (2). (This can be done for all frequencies simultaneously prior to this loop.) - Compute the BLUE estimators, and , using the known coefcients, and , according to Eq. (9). - Compute the BLUE estimators, s and s , using the simulated coefcients, s and s , at the known points according to Eq. (8). - Generate the conditional Fourier coefcients at the unknown points, c and c , using Eq. (10). 2) Generate the remaining Fourier coefcients at frequencies !k , k = K=2 + 1; : : :; K 1 using the symmetry conditions of Eq. (3). 3) Apply an inverse FFT to construct the eld of time histories at the unknown target points. To account for non-stationarity or evolving spectral density function, steps 1 through 3 may be repeated for a sequence of adjoining time windows (each having different spectral density and/or correlation functions) to generate time segments of ground motion. These segments are then pieced together by means of a linear interpolation algorithm (Fenton, 1990). Although not included in the program, the simulated motions may be post-processed to introduce non-stationarity in the mean and/or variance. This may be accomplished by a simple addition (mean) and multiplication (standard deviation) of the simulation by known functions of space and time.

A A

B B

3.

Running the Program

Simqke-II is invoked from the command line as follows simqke2 [-q] file1 [file2 ...] where optional arguments are enclosed in square brackets (do not include the brackets on the command line). The -q option simply tells Simqke-II not to send anything to standard output (the screen) except for terminal error messages. Without this option, a limited amount of progress information is reported to the screen. The lename argument(s) is the name of the governing, or job control, data le. If the data le is not in the current directory, its full pathname must be specied. If more than one input lename is specied, they are processed sequentially. For example, the command line simqke2 -q run4.dat run5.dat would quietly produce two sets of simulations, one for the problem specied in the data le run4.dat and the second for the problem specied in run5.dat. The limit on the number of input les given on the command line is system dependent, but typically a large number. In order to perform a simulation run, the following steps must be performed by the user; 1) create the job control data le, as discussed in Section 4.1, including the names of the following les (where appropriate), 2) set up the spectral density function data le as discussed in Section 4.2, 3) for a conditional simulation, create the known time history data le (see Section 4.3), 4) invoke Simqke-II with the name of the job control data le as an argument.

SIMQKE-II Users Manual 4. Input Data Files

The data required by this program is input through two or three data les which are described in more detail in the following subsections. The data les are referred to as follows, although they may actually have any name; JCDAT denes the job control parameters GWDAT contains the spectral density function data TMDAT contains the known, or recorded, time history data. This le is optional and will be absent/ignored if no known time histories are specied in JCDAT. The name of the JCDAT le is obtained from the command line argument. For example, if Simqke-II is invoked as simqke2 run4.dat then the job control parameters are read from the le run4.dat. The extension .dat is not necessary. Note, however, that output les having the basename run4 (as derived from the basename of the argument lename) are created having extensions *.stt job status/debug/error information *.sdf output spectral density function plots (PLOTPS format, optional) *.thp output time history plots (PLOTPS format, optional) *.tym output time histories (free format, ASCII, optional) and so it is not advisable to use any of these extensions for an input data le name since it may be overwritten if its basename is also the same as the job control data le. For example, therefore, it is inadvisable to name the input spectral density function data le run4.sdf if the JCDAT lename is run4.dat. See Output Files for more details on the above les.

SIMQKE-II Users Manual

4.1 JCDAT: Job Control Data File

The job control data le has the following format (optional quantities are shown in square brackets do not include the brackets); Description 1) job title 2) data echo ag 3) debug ag 4) no. of eld points 5) no. of known time histories 6) name of time hist le 7) name of SDF data le 8) time step increment 9) number of time steps 10) number of time windows [sizes] 11) name of correlation function 12) cor function parameters 13) plot time history ag 14) combine time history plots ag 15) plot SDF ag 16) generator seed (0 for random) 17) coordinate title 18) point coordinates Variable Name(s) job echo debug npt nkp tmdat gwdat dt nts nev [nv1, nv2 ...] cornm [P1 P2 P3 ] lplot lcombn lgplot kseed i x y z Type (character string) (logical t/f) (logical t/f) (integer) (integer) (character string) (character string) (real) (integer) (integers) (character string) (reals) (logical t/f) (logical t/f) (logical t/f) (integer) (int,reals,multi-line)

The le is set up in such a way that the rst 49 columns of each line (except the rst line and those containing the point coordinates) are informational and are ignored by this program. The lines must remain in the order listed above, and all reads beyond the 49th column are free-format. A typical example data le might appear as follows;
Four Point Simulation, No Known Points Echo job control data to stats file (t/f)? Dump debug data to stats file (t/f)? . . . Total number of field points . . . . . . . Number of known time histories . . . . . . Name of time history data file . . . . . . Name of spectral density data file . . . . Time step increment (seconds) . . . . . . Total number of time steps . . . . . . . . No. of time windows [sizes] . . . . . . . Correlation structure function name . . . parameters . . . . . . . . . . . . . . . Time histories to be plotted (t/f)?. . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

. . . . . . . . . . . .

t f 4 0 none 4step.sdf 0.02 2304 4 256 1024 512 512 exprw 8168.14 t

SIMQKE-II Users Manual

Time histories combined on single plot(t/f)? . . Power spectra to be plotted (t/f)? . . . . . . . Generator seed (0 = chosen internally) . . . . . Point X coordinate Y coordinate Z coordinate 1 0. 0. 0. 2 100.000 0. 0. 3 0. 100.000 0. 4 100.000 100.000 0. t f 0

10

(line ignored)

(npt entries)

Each record, according to line number, is described as follows; 1) The job title (job) is a character string which is echoed in its entirety to the *.stt and plot output les. 2) Set the logical ag echo to true (T) if the input data in the JCDAT le is to be echoed to the *.stt output le. This may be used to ensure that Simqke-II has properly read the input data. 3) Set the logical ag debug to true if debug information is to be logged to the *.stt output le. This may result in a lot of output, so debugging should be restricted to small runs. 4) The positive integer npt specied on this line represents the total number of spatial points of interest, including both the known (recording) and unknown (target) points. 5) The non-negative integer nkp specied on this line represents the number of known points at which ground motions have been recorded. If this number is non-zero, then the recorded time history data must appear in the le specied on line 6. 6) The character string tmdat contains the name (or pathname, if not in the current directory) of the le containing the recorded time history data. If the number specied on line 5 is zero, then this name is not used, however this record (line 6) must still be present in the JCDAT data le. 7) The character string gwdat contains the name (or pathname, if not in the current directory) of the le containing the spectral density data for this eld. 8) The time step increment dt is a positive real number which denotes the time increment to be used in the output time histories. This increment is taken to be constant over the duration of the simulation and is assumed to apply also to the known time histories. An error message is issued if the time step used in the known time histories does not agree with that provided here. 9) The number of time steps nts is a positive integer denoting the total number of time steps to take in the simulation. Thus, the simulation duration is the product of nts and dt. Because the simulation is obtained using an inverse FFT, the value of nts cannot be selected entirely arbitrarily it must be the sum of nev powers of 2 (see next). For example, if nev is 3, then nts must equal 2i + 2j + 2k for three positive integers i, j , and k . In addition, the number of time steps specied on this line must agree with the number of time steps found in the known time history data les. An error message will be issued if not. 10) To allow for a pseudo-evolving spectral density, the overall time interval [0; tf ] may be split into a sequence of subintervals within each of which a different spectral density function may be prescribed. On this line the rst integer specied, nev, is the number of subintervals to use. This value must be provided and cannot be less than one. Optionally following nev are up to nev more integers specifying the number of time steps to use in each subwindow. These numbers must all be powers of 2. If there are fewer values provided than nev, the remaining windows are evenly divided into the remaining number of time steps. Thus, for example, if lines 9 and 10 read

SIMQKE-II Users Manual

Total number of time steps . . . . . . . . . . . Number of time windows [sizes] . . . . . . . . . 2048 4 256

11

then the 4 time windows consist of 256, 597, 597, and 598 time steps each. Notice that the last window is adjusted so that the sum of window sizes is equal to that specied on line 9 (deviation in window sizes is due to integer truncation). However, these particular numbers will result in failure of the program since each time window must be a powers of 2 in length. This limitation is checked internally and error messages issued, where appropriate. 11) To provide for spatial correlation between time histories, the user must specify (and, for special cases, implement) a space-frequency correlation function, which computes the correlation coefcient between two points at a given separation distance and for a particular ground motion component frequency. The character string provided for cornm is the name (as known to the program) of the function to be used. At this time, two space-frequency correlation functions are provided with the program having names exprw alnrw and so one of the above strings would normally be specied on this line. See next line for a discussion of the parameters passed to these functions. Users with access to the source code for Simqke-II may write their own correlation function in which case its name must be added to the external list in the main routine simqke2.f and the program must be recompiled with the new module. Arguments to the correlation function are (r,w), where r is the separation distance between the two points for which the correlation is desired, and w is the wave frequency under consideration in rad/sec. 12) On this line, up to 3 real values may be specied. These values are passed to the spacefrequency correlation function via the labeled common block /evopar/ and are treated within the function as parameters controlling the correlation decay rate with respect to the two formal function arguments r and w (see above). Note that the units of w are rad/sec and the units of r are whatever units are used to specify the target point locations (see line 18). The parameters passed to the correlation function must have units which are consistent with the units of w and r. If echo is true (see line 2), then the values of all three parameters are reported in the *.stt le (any unset parameters are assumed to be zero). If one of the provided space-frequency correlation functions is selected on the above line (exprw or alnrw) then the parameters entered here have the following interpretation; for exprw parameter #1 (P1 ) is the correlation scale factor. The correlation returned is expf !r=P1 g where ! is the frequency of the particular component of motion under consideration, and r is the absolute distance between the two points for which the correlation is desired. The value of P1 can be expressed as 2 cs, where c is the shear wave velocity in the medium, and s is a dimensionless distance-scale parameter. The degree of correlation between points can be controlled by varying s. The other two parameters are unused. for alnrw parameters #1 and #2 (P1 and P2 ) are correlation scale velocities and parameter #3 (P3) is the local wave velocity through the medium. The correlation at frequency w between two points separated by r is given by ln

P1 + r! P
3

ln(P1 =P2 )

ln

P2 + r! P
3

where ln( ) is the natural logarithm.

SIMQKE-II Users Manual

12

13) Set the logical ag lplot to true if the simulated time histories are to be output to a le having plotps format. Plotps is a program which translates its input le into a PostScript graphics le. It can be obtained by anonymous FTP to random.field.tuns.ca:pub/progs/plotps 4.0.tar.gz (Avoid the plotps min program unless you already have the PSLIB library, found in pub/GAFlibs, installed. This library is already included directly in the plotps 4.0.tar.gz distribution.) Set the logical ag lcombn to true if all the time histories produced are to appear on the same plot axes with rst time history shown using a solid line, second with a dashed line, etc. For more than two time histories, or for rapidly varying time histories, this option is probably not appropriate. It is mainly useful for comparing motions visually. Set the logical ag lgplot to true if the provided spectral density function is to be plotted. The output le produced must be post-processed by plotps to convert it into PostScript. The specied generator seed, kseed, is used to initialize the random number generator randu (Press et al., 1992, from RAN1 program, pg 271). If two subsequent runs are invoked with identical generator seeds, then the sequence of random numbers which go into the simulation will be identical. However, the time histories may not be identical if the correlation structure is changed. If the random number generator seed is specied as 0, then a seed is produced internally using the current process system ID (see iseed.f). This line contains a title for the following coordinate data. It is ignored although it must be present. On this line and continuing for a total of npt records, provide the point number and spatial coordinates of each point in the eld. For conditioned simulations, read the following notes; i) the point number given in the rst column of each record of the spatial coordinate data must be unique and, overall, the point numbers must count from 1 to npt. ii) the point numbers associated with known points must count from 1 to nkp. The point numbers associated with the unknown points must count from (nkp + 1) to npt. In this way the coordinates of the known points and unknown points are determined. iii) the numbering of the known points must correspond to the ordering of the recorded time histories appearing in the TMDAT data le. This is the only link between the known points and their associated recordings.
GWDAT: Spectral Density Function Data File

14)

15) 16)

17) 18)

4.2

This data le contains the points dening the one-sided marginal spectral density function, Gt (! ), governing the temporal behaviour of the time histories. Since the eld is assumed homogeneous, the same spectral density function is used at all eld points. If the spectral density is evolving with time, a sequence of functions may be prescribed in this data le (see JCDAT, line 10) so that if the ith time window spans the time interval [ti; ti+1), then Gt (! ) = Gi (! ) for t 2 [ti; ti+1), i = 1; 2; nev. The individual functions are specied by a discrete sequence of real value pairs [!; Gi (! )], enabling the use of discrete spectral analyses of actual recordings. If the spectral density function is only known in functional form, it must rst be sampled at discrete values of ! the samples then placed in this data le. The values of ! need not be equispaced, nor do they need to be the same as the

SIMQKE-II Users Manual

13

Fourier frequencies used in the simulation. The value of Gi (! ) is obtained by linear interpolation at the Fourier frequency using the provided data. If the provided data is very coarsely spaced in the Fourier frequency range (0 to = t rad/sec), then the representation of the spectral density function by the simulation may be poor, so it is best to ensure reasonably ne resolution in the Fourier frequency range. The spectral ordinate data in the GWDAT le is converted to have length units of m2 . The units of the raw data provided in the le are specied in the le header as a character string. Recognizable length units are "m" (the default), "cm", "ft" or "feet", and "in". Length units other than one of these result in no conversion, in which case the user must ensure that spectral and (any) known time history units are consistent. For length units L, the spectral data has units L2 =sec3 for acceleration simulations, L2 =sec for velocity simulations, and L2 sec for displacement simulations. To ensure consistency, the length units of the known time histories are similarly converted to metres. The length units of the spatial coordinates of the eld points are not converted because they do not directly affect the simulation the coordinate units must, however, be consistent with the prescribed space/frequency correlation function parameters. The units specied in the spectral density function data le are also used to determine if the simulation is to be one of (which are searched for in the given order); 1) acceleration, in which case one of the strings sec3 or sec**3 must be found in the units character string, or 2) velocity, in which case the string /sec must be found, or 3) displacement, in which case the string *sec or sec must be found. The spectral data is read from a le having one of the following two forms (A or B); A) Free format (ASCII) with lines; for each j = 1; 2; : : :; nev 1) nev, nwj 2) type 3) units 4) wj (i), Gj (i), i = 1, 2, end for which is arranged in nev identically structured blocks. The blocks may be separated by blank lines if desired. The values of nev, type, and units are only actually recorded from the rst block that is read in subsequent blocks, lines 2 and 3 may be blank, although the lines themselves must be provided, and nev may take on other values since it is ignored except as a placeholder. The individual le components are described as follows; nev is the total number of spectral density functions provided in the le. This number must match that provided in the JCDAT data le on line 10. The value of nev in each of the 2nd to nevth block is ignored but must be present. nwj is the number of points dening the spectral density function specied in this (j th) block. For blocks 2 through nev this is the only pertinent piece of information, beyond the spectral data itself. This approach allows different spectral sample rates in the various time windows. type is a character string containing either Period or Frequency depending on whether the units of wj (i) are seconds (for period) or rad/sec (for frequency). If type is Period,

: : :,

nwj

SIMQKE-II Users Manual

14

then wj (i) is converted to the corresponding frequency 2 =wj (i) in rad/sec. The value of type encountered in the rst block is assumed to hold for all subsequent blocks. units is a character string containing the units of the following G(! ) data. Typically this is something like m2/sec3 but length units cm, ft or feet, and in are also acceptable. If any of the latter length units are found in the string units, then G(! ) is converted to units m2 =sec3 for acceleration simulations, m2 =sec for velocity simulations, or m2 sec for displacement simulations. (Note, however, that this means if type contains the string G(w) in m2/sec3 then, because of the substring in, the data will be converted from in2=sec3 to m2 =sec3 , which is probably not as intended.) the real number pairs (wj (i), Gj (i)), i = 1, 2, : : :, nwj may be input one pair per line or up to 128 pairs per line (maximum 1024 characters) with spaces, tabs, or newlines (carriage returns) separating the individual numbers. In each block, the program continues reading numbers until it has found nwj pairs in total. B) PLOTPS format: Here the input le is assumed to be in a format suitable also for processing by plotps, whose manual pages should be consulted in conjunction with the following description. The input le consists of two parts, the rst part being a header describing how the data is to be plotted and the second part being the raw data, that is the actual (!; G(! )) data pairs. In detail, the le contains; (11 + 2 nev) lines of header information. Only lines 1, 6, 10, and 11 contain information pertinent to this program; line 1: if the rst integer (nplot) provided is a 1, then the spectral data is assumed to follow model (1) or (2) below. Otherwise the rst integer should equal nev and model (3) below is followed. Note that if model (3) is used, the header consists of only 13 lines of plotting information. line 6: if the rst integer in line is a 1, then line 6 is checked to see how the data is provided. If line 6 is an integer (k) equal to negative nev, then model (2) is used, otherwise model (1) is used. line 10: denotes by the word Period or Frequency whether the frequency data, WG, has units seconds or rad/sec. The default is rad/sec. If line 10 contains the word Period, the data is converted to frequency (rad/sec). A period of 0.0 is assigned a frequency of 1 1015 rad/sec (arbitrarily). line 11: contains the units of G(! ). The default length unit is m2 . Only the length units are converted, and recognizable strings are cm, ft, feet, or in. If one of these strings are found on line 11, then G(! ) data is converted to length units m2 using the appropriate conversion factor. Models: In the following, the frequency data is denoted by WG(i,j), i = 1, 2, ..., nw(j), j = 1, 2, ..., nev and the corresponding spectral density values as G(i,j), i = 1, 2, ..., nw(j), j = 1, 2, ..., nev, where nev is the number of subwindows into which the simulation duration is divided, and nw(j) is the number of spectral data points provided in the jth subwindow. (1) if nplot = 1, k > 0, then the spectral data has the following form; (11 + 2*nev) lines of header information (see above)
WG(1) G(1,1) G(1,2) ... G(1,nev)

SIMQKE-II Users Manual

WG(2) G(2,1) G(2,2) ... G(2,nev) . . . . . . . . WG(nw) G(nw,1) G(nw,2) ... G(nw,nev) [ blank line or end-of-le]

15

Note that in this case, all spectral density functions are prescribed by the same number of discrete observations (nw). (2) if nplot = 1, k = nev, then the spectral data has the following form; (11 + 2*nev) lines of plotting information (see above)
WG(1,1) G(1,1) WG(2,1) G(2,1) . . . . WG(nw(1),1) G(nw(1),1) [ blank line] WG(1,2) G(1,2) WG(2,2) G(2,2) . . . . WG(nw(2),2) G(nw(2),2) [ blank line] . . [ blank line] WG(1,nev) G(1,nev) WG(2,nev) G(2,nev) . . . . WG(nw(nev),nev) G(nw(nev),nev) [ blank line or end-of-le]

Note that in this case, each spectral density function may be prescribed by a different number of discrete observations (nw(1), nw(2), ... nw(nev)). (3) if nplot = nev, then the spectral data has the following form; 13 lines of plotting information (see above)
WG(1,1) WG(2,1) . . WG(nw(1),1) [ blank line] WG(1,2) WG(2,2) . . WG(nw(2),2) [ blank line] G(1,1) G(2,1) . . G(nw(1),1)

13 lines of plotting information (ignored)

G(1,2) G(2,2) . . G(nw(2),2)

13 lines of plotting information (ignored)

SIMQKE-II Users Manual

. . [ blank line] 13 lines of plotting information (ignored) WG(1,nev) G(1,nev) WG(2,nev) G(2,nev) . . . . WG(nw(nev),nev) G(nw(nev),nev) [ blank line or end-of-le]

16

Again, in this case, each spectral density function may be prescribed by a different number of discrete observations (nw(1), nw(2), ... nw(nev)). The only basic difference between models (2) and (3) is that in model (3) the plot header information is repeated between blocks of data. NOTES: 1) The le type is determined by consulting the second line of the data le to see if it contains one or other of the strings Period or Frequency (if so, then the le is Free format, see above). 2) The number of spectral density function data sets provided in this data le must agree with the number of time windows specied in the control data le (nev). 3) The elements of WG as read in may be either frequency (rad/sec) or period (sec) as specied earlier in the data le by the word Period or Frequency appearing in the preamble. If WG is a period, then it is converted to frequency internally (0 period is converted to a frequency of 1:015). It is also assumed that the same units for G and WG are used in all spectral function data sets provided in the le.
4.3 TMDAT: Known Time History Data File

This le contains the time history data corresponding to the spatial locations at which recordings were made. The time histories are input as data pairs (TI, TH), where TI is the time in seconds, and TH is the corresponding motion value. The length units of TH are converted into metres from recognizable units of cm, ft or feet, or in. Length units other than one of these are left unconverted. Simulated motions will generally be one of displacement, velocity, or acceleration and the user must ensure that the provided spectral density function is consistent with the time history data. If the units of TH are U, then units of G must be U2 sec. For example, if TH is acceleration in m=sec2 , then G must have (or be converted to) units of m2=sec3 . The parameters nts (number of time steps) and nkp (number of known time histories) in this data le must agree with that specied in the job control data le. The known time history(s) are read from a data le having one of two forms (A or B); A) Free format (ASCII) with lines; 1) title (job title) 2) units (string containing length units) 3) nts, nkp (number of; time steps, known points) 4) (X(i,1), X(i,2), X(i,3), i = 1, nkp ) (currently ignored, coordinates of known points) TI(1) TH(1,1) TH(2,1) ... TH(nkp,1) 5)

SIMQKE-II Users Manual

TI(2) TH(1,2) TH(2,2) . . . . . . TI(nts) TH(1,nts) TH(2,nts) [ end of le] ... TH(nkp,2) ... . ... . ... TH(nkp,nts)

17

B) PLOTPS format: Here the input le is assumed to be in a format suitable also for processing by plotps, whose manual pages should be consulted in conjunction with the following description. The input le consists of two parts, the rst part being a header describing how the data is to be plotted and the second part being the raw data, that is the actual (TI,TH) data pairs, where TI is the time in seconds, and TH is the motion value. In detail, the le contains; (11 + 2 nkp) lines of header information. Only lines 1, 6, and 11 contain information pertinent to this program; line 1: if the rst integer (nplot) provided is a 1, then the time history data is assumed to follow model (1) or (2) below. Otherwise the rst integer should equal nkp and model (3) below is followed. Note that if model (3) is used, the header consists of only 13 lines of plotting information. line 6: if line 1 specied a 1, then line 6 is checked to see how the data is provided. If line 6 is an integer (k) equal to negative nkp, then model (2) is used, otherwise model (1) is used. line 11: contains the units of the time history, assumed to length unit with default m (metres). Only the length units may be changed, and recognizable strings are cm, ft, feet, or in. If one of these strings are found on line 11, then the length units are converted to m using the appropriate conversion factor. In all models to follow, the number of time steps must be the same and equal to the prescribed nts (which must be the same here and in the job control le). Models: (1) if nplot = 1, k > 0, then time history data has the following form; (11 + 2*nkp) lines of header information (see above)
TI(1) TH(1,1) TH(2,1) TI(2) TH(1,2) TH(2,2) . . . . . . TI(nts) TH(1,nts) TH(2,nts) [ blank line or end-of-le] ... TH(nkp,1) ... TH(nkp,2) ... . ... . ... TH(nkp,nts)

(2) if nplot = 1, k =

nkp, then the time history data has the following form;

(11 + 2*nkp) lines of plotting information (see above) TI(1) TH(1,1) TI(2) TH(1,2) . . . . TI(nts) TH(1,nts) [ blank line] TI(1) TH(2,1) TI(2) TH(2,2)

SIMQKE-II Users Manual

. . . . TI(nts) TH(2,nts) [ blank line] . . [ blank line] TI(1) TH(nkp,1) TI(2) TH(nkp,2) . . . . TI(nts) TH(nkp,nts) [ blank line or end-of-le]

18

3) if nplot = nkp, then the time history data has the following form; 13 lines of plotting information (see above)
TI(1) TH(1,1) TI(2) TH(1,2) . . . . TI(nts) TH(1,nts) [ blank line]

13 lines of plotting information (ignored)

TI(1) TH(2,1) TI(2) TH(2,2) . . . . TI(nts) TH(2,nts) [ blank line]

13 lines of plotting information (ignored)

. . [ blank line]

13 lines of plotting information (ignored)

TI(1) TH(nkp,1) TI(2) TH(nkp,2) . . . . TI(nts) TH(nkp,nts) [ blank line or end-of-le]

The only basic difference between models (2) and (3) is that in model 3 the plot header information is repeated between blocks of data. NOTES: 1) The time increment is assumed to be uniform throughout the sequences. 2) The le format is determined by consulting the third line of the data le to see if it contains two leading integers (nts and nkp). If not then the le is assumed to be in PLOTPS format.

SIMQKE-II Users Manual 5. Output Files

19

Output is made to the les; *.stt - job status/debug/error information *.sdf - spectral density function plots (PLOTPS format) *.thp - time history plots (PLOTPS format) *.tym - time histories (free format, ASCII) where the last 3 are optional, depending on user dened options in the job control data le. The basename * is obtained from the name of the JCDAT le. Ensure that basenames are unique if output les of the same name already exist, they are simply overwritten with no warning. The free format saved time history, *.tym (which is produced if the time history plotting option is not selected), will have the format 1) title (character string) 2) units (character string) 3) nts, npt, nkp (number of; time steps, points, known points) 4) (X(1,i),X(2,i), X(3,i), i = 1, npt) (spatial coordinates) 5) (TI(j), (TH(i,j), i = 1, npt), j = 1, nts) (time, time history) while all other plots are in PLOTPS format (see plotps(1)). The units are expressed in the form m=sec2 for acceleration time histories and m2=sec3 for the corresponding spectral density functions Similarly, m=sec and m are the units used for velocity and displacement time histories, respectively, with corresponding spectral density function plots in m2=sec and m2 sec. 6. System Specics

A number of system specic function calls are made by this program which may need to be replicated on certain platforms. In particular, these routines may not be available under certain Fortran compilers. These system routines are described as follows; 1) call fdate(rdate) returns current date/time in a 24 character string stored in rdate (called by simqke2.f) 2) call getarg(n,arg) returns the nth command line argument in the string arg (on HPs use the +U77 ag when compiling and linking) (called by simqke2.f) 3) getpid() returns current process ID. Called by iseed.f, this function is used to generate a pseudo-random number generator seed in the event that the user species a seed of 0. This function should at least return different integers on each invocation of Simqke-II. Wall clock time could also be used. (Unix version only.) 4) etime(t) returns the elapsed execution time of the running program. The argument t is a single precision real vector of length 2, however, it is not used. This function is called by second.f. 5) getdat(), gettim() returns the date and time in integer arguments. Used to internally generate a pseudo-random number generator seed, see (3) above. (Win95/98/NT only.)

SIMQKE-II Users Manual 7. Example

20

In this section an example is presented to illustrate the use of Simqke-II. The exponentially decaying isotropic frequency-dependent spatial correlation function exprw has been used,

!k ( ij ) = exp

where ij is the relative position vector between points i and j , c is the shear wave velocity in the medium, and s is a distance-scale parameter. The job control data le examp.dat used in the simulation is as follows
Three Point Simulation, One Known Point, Echo control data to stats file (t/f)? . Dump debug data to stats file (t/f)? . . Total number of field points . . . . . . Number of known time histories . . . . . Name of time history data file . . . . . Name of spectral density data file . . . Time step increment . . . . . . . . . . Total number of time steps . . . . . . . Number of subwindows [sizes] . . . . . . Correlation structure function name . . parameter(s) . . . . . . . . . . . . . Time histories to be plotted (t/f)?. . . Time history plots combined (t/f)? . . . Power spectra to be plotted (t/f)? . . . Pseudo-random number generator seed . . Point X coord. Y Coord. Z 1 0.0 0.0 2 10.0 0.0 3 500.0 0.0 scale = 5.0 . . . . t . . . . t . . . . 3 . . . . 1 . . . . comp1.tym . . . . psdc1.sdf . . . . 0.008 . . . . 3328 . . . . 3 256 1024 2048 . . . . exprw . . . . 4084.1 . . . . t . . . . f . . . . f . . . . 0 Coord. 0.0 0.0 0.0

!k jrij j 2 cs

(11)

which is a 3 point problem having points arranged along the x axis at 0.0, 10.0, and 500.0, with motion recorded at the rst point. The recorded time history is stored in the le comp1.tym. The spectral density function is dened in 3 unequal length subwindows as shown in Figure 1 and stored in the le psdc1.sdf. Figures 2, 3, and 4 show excerpts from the comp1.tym, psdc1.sdf, and the output examp.stt le respectively. A distance scale of s = 5 has been used in the example data le, so that a simulated motion at a distance of 10 m from the recording should look fairly similar, at least in the lower frequency components. On the other hand, a simulated motion at a distance of 500 m will be largely independent of the recording. The simulated motions for this problem are shown in Figure 5.

SIMQKE-II Users Manual

4

21

x10

For time window [0, 2.04] seconds

4 0 0 2

G(w) ( m /sec )

10

20

30

40

50

60

70

80

90

100

Frequency (rad/sec)
0.06

For time window [2.048, 10.232] seconds

G(w) ( m /sec )

0 0
4

0.02

0.04

10

20

30

40

50

60

70

80

90

100

Frequency (rad/sec) x10 For time window [10.24, 26.616] seconds

15 0 0 5 10

G(w) ( m /sec )

10

20

30

40

50

60

70

80

90

100

Frequency (rad/sec)

Figure 1.

Power spectral density function used in each of three subwindows. Note that the vertical scales are considerably different.

SIMQKE-II Users Manual

Horizontal Component 1 at 20.00 degrees clockwise from North m/sec2 3328 1 0.0 0.0 0.0 0.000000E+00 -0.218960E+00 0.800000E-02 -0.271767E+00 0.160000E-01 -0.328865E+00 0.240000E-01 -0.350127E+00 0.320000E-01 -0.301266E+00 0.400000E-01 -0.167549E+00 0.480000E-01 -0.299566E-01 0.560000E-01 0.747209E-01 0.640000E-01 0.126518E+00 0.720000E-01 0.128547E+00

22

Figure 2 Heading of prescribed time history le comp1.tym.

3 200 Frequency m2/sec3 0.0000E+00 0.3157E+01 0.6315E+01 0.9472E+01 0.1263E+02 0.1579E+02 0.1894E+02 0.2210E+02 0.2526E+02

0.2635E-03 0.2643E-03 0.2670E-03 0.2714E-03 0.2776E-03 0.2857E-03 0.2955E-03 0.3069E-03 0.3196E-03

Figure 3 Heading of spectral density function le psdc1.sdf.

SIMQKE-II Users Manual

This is SIMQKE2 (Version 2.6): Wed Feb 19 09:10:09 1997 Control file: examp.dat ============================ Job Control Data ================== Three Point Simulation, One Known Point, scale = 5.0 Echo control data to stats file (t/f)? . . . . . T Dump debug data to stats file (t/f)? . . . . . . T Total number of field points . . . . . . . . . . 3 Number of known time histories . . . . . . . . . 1 Name of time history data file . . . . . . . . . comp1.tym Name of spectral density data file . . . . . . . psdc1.sdf Time step increment . . . . . . . . . . . . . . 0.800000E-02 Total number of time steps . . . . . . . . . . . 3328 Number of subwindows [sizes] . . . . . . . . . . 3 256 1024 2048 Correlation structure function name . . . . . . exprw parameter(s) . . . . . . . . . . . . . . . . . 0.40841E+04 0.0 0.0 Time histories to be plotted (t/f)?. . . . . . . T Time history plots combined (t/f)? . . . . . . . F Power spectra to be plotted (t/f)? . . . . . . . F Pseudo-random number generator seed . . . . . . 0 Point X Coord. Y Coord. Z Coord. 1 0.000000E+00 0.000000E+00 0.000000E+00 2 0.100000E+02 0.000000E+00 0.000000E+00 3 0.500000E+03 0.000000E+00 0.000000E+00 ========================== End of Control Data ================ Reading spectral data file psdc1.sdf ... (ASCII format) (SDF units: m2/sec3) Target Variances: For Window 1 = 0.0220256 (m/sec2)2 For Window 2 = 0.7707890 (m/sec2)2 For Window 3 = 0.0144998 (m/sec2)2 Reading time history data file comp1.tym (ASCII format) (Time history units: m/sec2) For pseudo-random number generator seed 6459 Generating acceleration time histories ... window: initializing... -1% variance above w max in subwindow 1 -2% variance above w max in subwindow 2 0% variance above w max in subwindow 3 npt,nkp,nev = 3 1 3 nstw = 256 1024 2048 mstw = 8 10 11 nw = 200 257 257 dw = 3.06796 0.76699 0.38350 nv,dv = 32 0.01563 sg(1-->5) for each window; 0.00081 0.00020 0.00020 0.00021 0.00021 0.04001 0.00990 0.00979 0.00968 0.00943 0.00040 0.00010 0.00010 0.00010 0.00010 Inter-Point Distance Matrix: 0.0000000 10.0000000 500.0000000 10.0000000 0.0000000 490.0000000 500.0000000 490.0000000 0.0000000 generating window 1 generating window 2 generating window 3 took 0.215696 seconds. Time history plot written to file examp.thp

23

Figure 4 Output le examp.stt.

SIMQKE-II Users Manual

24

Known motion at recording point

2

Accel. ( m/sec )

-3 0

-2

-1

10

15

20

25

30

Time (sec)
3

Simulated motion at target point, r = 10 m

2

Accel. ( m/sec )

-3 0

-2

-1

10

15

20

25

30

Time (sec)
3

Simulated motion at target point, r = 500 m

2

Accel. ( m/sec )

-3 0

-2

-1

10

15

20

25

30

Time (sec)

Figure 5.

Conditionally simulated ground acceleration.

SIMQKE-II Users Manual 8. 1. References

25

motion: Non-parametric estimation, Soil Dynamics and Earthquake Engineering, 14, 2331, 1995. 2. Boissieres, H.P., Estimation of the Correlation Structure of Random Fields, thesis presented to Princeton University, at Princeton, New Jersey, in partial fulllment of the requirements for the degree of Doctor of Philosophy, 187 pg., Jun. 1992. 3. Fenton, G.A., Error evaluation of three random eld generators, ASCE Journal of Engineering Mechanics, 120(12), 24872497, 1994. 4. Fenton, G. A., Simulation and Analysis of Random Fields, thesis presented to Princeton University, at Princeton, New Jersey, in partial fulllment of the requirements for the degree of Doctor of Philosophy, 178 pg., Jan. 1990. 5. Harichandran, R.S., and Vanmarcke, E.H., Stochastic variation of earthquake ground motion in space and time, ASCE Journal of Engineering Mechanics, 112(2), 154174, 1986. 6. Heredia-Zavoni, E., Structural Response to Spatially Varying Earthquake Ground Motion, thesis presented to Princeton University, at Princeton, New Jersey, in partial fulllment of the requirements for the degree of Doctor of Philosophy, 178 pg., Jun. 1993. 7. Press, W.H., Teukolsky, S.A., Vetterling, W.T., and Flannery, B.P., Numerical Recipes in Fortran, 2nd Ed., Cambridge University Press, New York, NY, 1992. 8. Vanmarcke, E.H., Heredia-Zavoni, E., and Fenton, G.A., Conditional simulation of spatially correlated earthquake ground motion, ASCE Journal of Engineering Mechanics, 119(11), 23332352, 1993. 9. Vanmarcke, E. H., and Fenton, G. A., Conditioned simulation of local elds of earthquake ground motion, Structural Safety, Special Issue on Spatial Variation of Earthquake Ground Motion, Jan. 1991. 10. Vanmarcke, E.H., Random Fields: Analysis and Synthesis, The MIT Press, Cambridge, Massachusetts, 1984.

Boissieres, H.P., and Vanmarcke, E.H., Spatial correlation of earthquake ground