Bulletin of the Seismological Society of America. Vol. 57, No. 6, pp. 1299-1315.

December, 1967

FAST PROGRAMS FOR LAYERED HALF-SPACE PROBLEMS

BY M. J. RANDALL

ABSTRACT
Knopoff's matrix method for the solution of P-SV problems has been some-
what simplified and modified to take account of oceanic structures. Advantage
has been taken of a method of separating the frequency-dependent operations
from the matrix multiplications to obtain very fast computer programs for calcu-
lating Rayleigh dispersion, crustal reflection functions, and crustal transfer func-
tions. Applications include Raylelgh dispersion inversion, Q~ inversion, crustal
investigations using pP, crustal transfer corrections to amplitude observations, and
the construction of synthetic seismograms for investigation of the source time-
function.
~NTRODUCTION
Knopoff (1964a) has described an alternative to the procedure of Thomson
(1950) and Haskell (1953) for solving elastic wave problems in layered half-spaces.
This paper concerns the application and, to some degree, simplification of Knopoff's
matrix method for Rayleigh and P-SV problems. In particular, the theory can be
expressed in such a way as to show the possibility of separating those operations
that depend on frequency from the matrix operations. This has important reper-
cussions on the time required for computation.
The elastic wave equation requires that we express the motions in a region con-
sisting of N homogeneous isotropic elastic layers bounded above by a free surface
and below by a homogeneous isotropic elastic half-space in terms of P- and SV-
wave potentials, which, for the nth layer, take the form

(1)

( - - 1 ) ~ (~) = b~ei'(k~+,~z-~).~ d~ei~(~x-,.~-t) (2)

where k is the wave slowness, or the reciprocal of the phase velocity,

~n = (1/~: -- k~) ~/~ (3)

~/n = ( 1 / ~ n 2 - k2) 1/2 (4)

an and ~,, are the P- and S-wave velocities in the nth layer.
The boundary conditions on stress and displacement at successive interfaces
generate the matrix equation
1299

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/57/6/1299/2694885/BSSA0570061299.pdf

by Uppsala Universetet user
 1300 BULLETIN OF THE SEISMOLOGICAL SOCIETY OF AMERICA

!
al
b1'
!
¢1
dl'
]
~2

If etc
I d~ r

' l . bN+l.

0
0
l df
0 C N-t-1 ~r-t-1
0
]
I
I
: I
0 i

(5)

where the blocks represent submatrices as follows:

= LGl~'l -F1 -Gi~'l -Fi J (6)

f,~ K -f~ K

(7)

also K (~) = D(~)H (n), where D (~) is the diagonal matrix

=
[o e ~~n~
0 e~'~
0

0
0 0
0
0
e-i~¢~
0
0
0
e-i~r~z~
j (8)

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/57/6/1299/2694885/BSSA0570061299.pdf

by Uppsala Universetet user
 FAST PROGRAMS FOR LAYERED HALF-SPACEPROBI,EMS 1301

and h~ is the thickness of the nth layer, and

F~ = o~(1 - 2¢~2]c2) (9)

(1o)
also

t ~ n e i°~[nZn - i
an (11)

bn' = b~e i°'Tnz'~-I (12)

t --i~ofnZn-- 1
Cn ~ Cne (13)

d~' = d ~ e - ~ ' ~ - L (14)

We denote by the block

the submatrix formed by the ith column of H (~ followed by the j t h column of

H (~) in that order. This notation will be used below to denote analogous submatrices;
the absence of subscripts within the block denotes that all columns of the named
matrix are present in the conventional order. Thus, Rayleigh dispersion curves are
obtained by solving for the zeroes of the Rayleigh denominator:

AR ~-

(15)

]Reflection problems are solved by calculating the ratios of the form aN+l/c~+l.
The numerator determinant in Cramer's solution for this ratio has the same form
as the denominator A~, and may be evaluated in the same way.

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/57/6/1299/2694885/BSSA0570061299.pdf

by Uppsala Universetet user
 1302 BULLETIN OF THE SEISMOLOGICAL SOCIETY OF AMERICA

Crustal transfer problems are solved by calculating the ratios of the form a:/c~+:.
The numerator determinant in Cramer's solution for this ratio no longer has the
same form as AR and a separate method must be developed for its evaluation.

THE MATRIX METHOD FOR AR

Determinants of the form of AR (equation 15) may be reduced to a series of

6 N 6 matrix multiplications by noting that a horizontal partitioning at any of the
block boundaries results in the expansion of the determinant in a sum of just six
minor-eofactor products. The six minors corresponding to a new partitioning at the
block next below the original partitioning, just below the U block, results in the
expansion of 5R in terms of the I N 6 matrix formed by the sLx 2 X 2 minors of the
U block:

(16)

where

FI 2
1 3
4
3
2 4
3 (:7,:s)

We also introduce the notation

:31 4
4
3
Z=
1 4
3
2 (19, 20)

Thus
6

etc (21)

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/57/6/1299/2694885/BSSA0570061299.pdf

by Uppsala Universetet user
 FAST PROGRAMS FOR LAYERED HALF-SPACE PROBLEMS 1303

where

11
-11
11
8
II
11
1_1 (22)

The determinants on the right can themselves be expanded:

6 6

j=l 114, 8j

ete (23)
where the superscripts to the upper left indicate the removal of the corresponding
rows and, upon expanding again,

6 6
*SmjK(1) ~Sk
ill d=l k=l

etc
(24)
This form is similar to that of equation (21), except that the right hand deter-
minants begin one block further down and V is replaced by V R (1), where R (1) is
the 6 X 6 matrix of elements

ik ~ 8i 8j
ldm i (1) [
IJ. lkm/~
i=1

~" 81 --llml ]

=
(25)

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/57/6/1299/2694885/BSSA0570061299.pdf

by Uppsala Universetet user
 13OAx BULLETIN OF T H E S E I S M O L O G I C A L S O C I E T Y OF A M E R I C A

where the frequency dependent terms are wholly contained in the vector

-e-~eii([i+~i)-
-1
e-i~hi (~!-~i)
f(i) = e¢~hi(i'i--~i)
-1
(26)
The frequency independent matrix/~(*)

ik = (27')

need be calculated only once and applied repeatedly for each frequency.
The above reduction may be carried out until only the last block remains and
we have

Ale = { VF(l>[~°)FC~l ft (2) "'" F('V)/~(~)}l (28)

r (~? which are

where F (') is the 6 X 6 diagonal matrix whose elements are the six j~
defined by analogy to the f~(~).
Thus, once V and the /~(~) have been calculated for a given phase velocity,
A~(k) can be calculated for any desired frequencies by successive applications of
the two steps:
(i) replace each V~ by its product with f~
(ii) replace the vector V by its product with the matrix /~. Finally, after N
repetitions, AR is the first element of the resulting V.

A LIQum S~RFAC~ LAYER

The procedure described above cannot be applied in the presence of a liquid
layer simply by putting the corresponding ~ equal to zero, for the original deter-
minant then has columns containing only zeroes. However, a modification is
easily made to account for one liquid layer above the top solid layer, which is the
case most commonly considered: that of oceanic structure. We do this by noting
that the presence of the liquid layer changes the boundary condition from that at a
free surface. Now we have for the top of the solid layer, z0 = 0,

alF1 + blGl~?l Jr- c i F 1 - - d i g i t / 1 = K (29)

where K may be determined from the remaining boundary conditions, the con-
ditions, the continuity of normal stress and the displacement at z = 0 and the
vanishing of the normal stress at z = --hz where hz is the depth of the liquid.
That is

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/57/6/1299/2694885/BSSA0570061299.pdf

by Uppsala Universetet user
 FAST PROGRAMS FOR LAYERED HALF-SPACE PROBLEMS 1305

K = i tan (~0h£z)[(a~ - c~)f~/f~ q- (b~ q- d~)k/f~lpz (30)

Thus, we must replace U of equation (15) by

U< = rFl -~- t~l Gl ~l -~- tk F 1 - t~l - G l y l -~- []c1

L alr~ -F~ -GIfl -F1 J (31)

where t = -ip~/fl tan (fzhzw).

In particular, we must replace V of equations (16, 28) by

vl - tpl ~'(
V2

V~
V6 + tpl ~1 (32:)

This new vector V' is frequency dependent and the correction must be applied to
the starting V for each frequency to be calculated.

CRUSTAL I~EFLECTION CALCULATIONS

The reflection coefficient for a complex layered crustal structure is determined
by ratios of the form a~+l/C~.l. This particular ratio corresponds to P-P reflection.
P-SV, S V - P and S V - S V reflection coefficients are determined by the analogous
ratios. We will deal in detail only with P-P reflection.
The Cramer solution of equation (5) yields

? !
,~ a•+l/c•+1 =

ere

(33)

This determinant is of the same form as AR except for the appearance of the bottom
block

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/57/6/1299/2694885/BSSA0570061299.pdf

by Uppsala Universetet user
 1306 BULLETIN OF THE SEISMOLOGICAL SOCIETY OF AMERICA

instead of

Thus,

i.e., the reflection coefficient is given by the ratio of the fourth element to the first,
after the final matrix multiplication.

CRUSTAL TI~ANSFER ]d~UNCTIOiWS

Crustal transfer functions express the ratio of displacements; that of a component,

vertical or horizontal, of the surface motion to that of an incident wave, P or SV.
The transfer function is a linear combination of ratios of the type al/cN+l. We will
deal in detail only with the vertical-component/P-wave transmission function,
so we are concerned with the ratios with numerator at, bl, cl, d~ and denominator
c~+1. In particular, the ratio a//c'~+l is given by the Cramer solut~ion

! !
A~ al/c~+1 = --
0

etc

(35)

~__A = ¸

7q
etc

(36)

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/57/6/1299/2694885/BSSA0570061299.pdf

by Uppsala Universetet user
 FAST P ~ O G R A M S FOR L A Y E R E D HALF-SPACE PROBLEMS 1307

We evaluate A by e x p a n s i o n i n t e r m s o f t h e 2 X 2 m i n o r s o f t h e firstbloek

A'=V4I~ ~etc.I-V~~ ~etc.

etc. (37)

.e.,

A =
o] D (1)
1/
v, Ew~'~ (3s)
where D (~) is the diagonal matrix (equation 8) and W (~) is the vector with elements

w(.) ~!~) H(~

I

(39)

We may evaluate W
,, i (n) by expansion in terms of the 4 X 4 minors of the top block:

4
w~"~ =, F~ ~,~ _ i/f (n)
K(y+~) ~

ere (40)

~--(n)
where ~l//ij is the 4 × 4 determinant formed by replacing the j t h column in H (~+1)
by the ith column of H ('). Thus,

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/57/6/1299/2694885/BSSA0570061299.pdf

by Uppsala Universetet user
 1308 BULLETIN OF THE SEISMOLOGICAL SOCIETY OF AMERICA

W(~) = _ M(~)D(~+nW(~+i) (41)

IlOW

w(N-i)
i

71/l(iv-i) r~(iv) [ .~jr(iv),~

"~* - - ~ i j ..t.l~k \-- k4 ) (42)

with summation over the repeated indices.

Similarly,
... A = (-1) ~

{I°1 Vo
-V~
V,
D(1)M (~) ... D(~)M (N) }
(4a)

! !
% bi/cN+l = ( - - 1 ) ~r
(44)

and

! !
n~ cl/c~+~ = (-1) ~ -- D(nM a) ... D ( n M (~)
(45)
L
and

n~
, , =
A/c~+1 (-1) ~
V2 DmM(n • .. D(N)M(n}.
(46)

Now the vertical displacement at the free surface

u~° = --i(ze~(~-t)(ai'~i + bl'k -- ci'~i + di'k) (47)

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/57/6/1299/2694885/BSSA0570061299.pdf

by Uppsala Universetet user
 PAST PROGRAMS FOR LAYERED HALF-SPACE PROBLEMS 1309

while the longitudinal displacement in the incident wave

uz N+I = i~ (-- 1)z"+~ei~'(U'~v+l t)CN+I/aN+, . (48)

Thus, apart from a constant phase factor,

O. N+I _ GN+I
~tz/~z AR
{QD(~)M(~) • .. D(~)M(N)}4
(49)

where

- V 5 ~ 1 - (V4 + V~)]~
Q =
- V ~ h + ( V ~ - VOk
( V 4 - V O h - V~ ]~ (50)

aNOW

- F 1 2 - G12~'I ~I-

- F 1 2 -~ G12~l '171
V =
F12 - G12~'1~1.
- - 2 F 1 G1 ~1
-F12 - a12h ~1 (51)

Thus

Q = 2pl
/-Flrq
LF, rlJ (52)

Thus, apart from a constant~ phase factor,

u O/u~z+i 2m <~+1 | - F 1 tl[ DO)M a)

- A~ |G~ ~1 v ~ "'" D(z")M(N)
L rlr,j (53)

Again~ the frequency-dependent terms are restricted to the diagonal matrices

D (~) and the 4 X 4 matrices M (~) need be calculated only once.

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/57/6/1299/2694885/BSSA0570061299.pdf

by Uppsala Universetet user
 1310 B U L L E T I N OF THE SEISMOLOGICAL SOCIETY OF AMERICA

I t m a y be noted t h a t the results of this section conflict with Knopoff's expression

for the coefficients (equation 11 of Knopoff, 1964a), which was in error.

THE I~AYLEIGH D I S P E R S I O N IOROGRAM

The phase velocity dispersion curve for Rayleigh waves on a given crustal model
is the locus in velocity-frequency space of the zeroes of the Rayleigh denominator
AR. Since it is possible, for certain structures with low-velocity layers, t h a t the

TABLE 1
THE 17-LAYER•{ODEL OF .~ CRUST AND UPPER ]~¢[ANTLEWHICHWAS USED IN
THE CALCULATIONOF TABLE 2
Thickness (kin) P velocity (km/sec) S veloeity(km/sec) Density (gm/cra~)

22.0 6.030 3.530 2.78

15.0 6.700 3.800 3.00
13.0 7.960 4.600 3.37
25.0 7.850 4.500 3.39
25.0 7.850 4.410 3.42
5.0 5.000 2.000 3.24
20.0 7.850 4.410 3.42
75.0 8.000 4.410 3.45
50.0 8.200 4.500 3.47
100.0 8.400 4.600 3.50
100.0 9.000 4.950 3.63
100.0 9.630 5.310 3.89
100.0 10.170 5.630 4.13
100.0 10.535 5.915 4.33
100.0 10.955 6.140 4.49
10O.0 11.275 6.285 4.60
150.0 11.460 6.384 4.69
½-space 11.755 6.500 4.80
7 layer model

phase velocity is not a monotonic function of frequency, it would be natural to seek

the phase velocity root c(e) for each of a discrete set of frequencies. To derive
m a x i m u m benefit from the feature of the above matrix method t h a t allows the
separation of the frequency-dependent operations, however, we perform the inverse
procedure: for each of a specified set of phase velocities we seek the frequency root
~o(c). As ~ is changed during convergence towards the root, the matrices /~(~)
remain constant, whereas the Haskell procedure requires the repeated evaluation
of frequency-dependent 4 X 4 matrices. I t is this use of the separation feature that
explains the advantage in speed t h a t the present program has over t h a t described
b y Press et at (1961), who quote a time of 2 minutes for a 20-layer case with 20-
phase velocities, or a speed of 3.3 layer-points per second (lps). The present program
has a speed of 15.3 lps, determined from the execution time for the 17-layer model
which is specified b y Table 1. Table 2 presents a comparison of the results obtained
b y the present program (R), and b y a copy of the program ( H ) described b y Press

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/57/6/1299/2694885/BSSA0570061299.pdf

by Uppsala Universetet user
 FAST PROGRAMS FOR LAYERED ItALF-SPACE PROBLEMS 1311

et al, which was kindly supplied by the Seismological Laboratory, California Insti-
tute of Technology. The actual speed of this latter program on an IB1V[ 7094 was
only 1.0 lps for this case. The R program obtains the group velocities U ( T ) b y
using a 5-point differentiation formula, rather than b y perturbing the phase ve-
locities as in the H program. These group velocities agree tolerably well.

TABLE 2
COMPARISON OF IoHASE AND GROUP VELOCITIES OBTAINED FOR 3_ 17-LAYER MODEL
BY THE IORESENT METHOD AND BY THE THOMPSON-HAsKELL METHOD

Phase Velocity Periods (seconds) T(c) Group Velocity (kin/see) U(T)

(km/sec) C R H R H

6.00 45470.85 45410.02 5.998

5.93 3911.38 3910.84 5.868
5.86 1769.66 1769.63 5.754 5.757
5.79 1021.13 1021.12 5.638 5.646
5.72 707.96 707.96 5.483 5.489
5.65 566.48 566.48 5.281 5.287
5.58 489.11 489.11 5.063 5.072
5.51 439.48 439.48 4.851 4.859
5.44 404.05 404.05 4.650 4.658
5.37 376.83 376.83 4.466 4.473
5.30 354.76 354.76 4.299 4.305
5.23 336.13 336.13 4.150 4.155
5.16 319.90 319.89 4.019 4.024
5.09 305.36 305.36 3.905 3.909
5.02 292.06 292.06 3.808 3.811
4.95 279.67 279.67 3.725 3.728
4.88 267.92 267.92 3.656 3.658
4.81 256.64 256.64 3.600 3.602
4.74 256.65 245.64 3.555 3.556
4.67 234.82 234.82 3.520 3.524
4.60 224.04 224.04 3.495 3.498
4.53 213.20 213.20 3.478 3.480
4.46 202.20 202.20 3.469 3.470
4.39 190.95 190.94 3.466 3.466
4.32 179.32 179.32 3.468 3.468
4.25 167.23 167.23 3.475 3.476
4.18 154.56 154.56 3.485 3.485
4.11 141.21 141.20 3.497 3.497
4.04 127.03 127.03 3.511
3.97 111.84 111.84 3.529

Central to the Rayleigh dispersion program is a subroutine named I N D I V which

accepts as input the parameters of the crustal model, the phase velocity c, and an
initial angular frequency ~1 ; and which returns with the Rayleigh root o~(c) next
highest in frequency to ~1. Thus, I N D I V is used to find the fundamental mode by
setting ~1 to zero (or to some value known to be less than the frequency of the
fundamental); successively higher modes are found by setting ~01 to a value just
greater than the frequency of the previous mode. I n cases where the phase velocity

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/57/6/1299/2694885/BSSA0570061299.pdf

by Uppsala Universetet user
 1312 BULLETIN OF THE SEISMOLOGICAL SOCIETY OF AMERICA

is not a monotonic function of frequency, such as those described b y Thrower and

H;arkrider (1965), the multiple frequencies of the fundamental mode for certain
phase velocities are also to be found b y this technique.
•A ,generalized flow chart for I N D I V is shown in Figure 1. T h e "appropriate"
incrementing of ¢ varies in the current, version of the program with the progress
t h a t the subroutine makes towards finding the root, as does the method of testing
A~. At first, only the phase of the complex number AR is tested and ~ is incremented
b y a fixed quantity c02 until the phase changes b y approximately ~, which indicates

TABLE 3

CRUSTAL ~ O D E L S U S E D IN THE CALCULATION OF F I G U R E S 2 AND 3

Thickness (km) P Velocity (km/sec) S Velocity(k~sec) Density (gm/cm~)

S~continentalcrust

2. 4.5 2.6 2.2

10. 5.5 3.2 2.4
8. 6.3 3.7 2.6
-- 7.6 4.4 3.2

ContinentalShield

2. 4.5 2.6 2.2

14. 6.0 3.5 2.5
7. 6.7 3.9 2.7
7. 7.1 4.2 2.9
8. 7.6 4.4 3.2
-- 8.1 4.7 3.3

Average Oceanic Crust

: 4.50 1.50 -- 1.03

0.45 2.00 1.00 1.70
1.75 5.00 2.88 2.30
4.70 6.71 3.86 2.84
-- 8.09 4.65 3.25

that the root is bracketed. The bracketing interval is then successively halved until
the root is contained within ~ specified small range, whereafter the larger of the
real and imaginary parts of A~ is tested and closer estimates of the root are ob-
tained b y linear interpolation until ~ (c) is determined to within the desired accuracy
(;t=10 .6 in the current version). Considerable numerical cxperinaentation lies
behind this particular choice of methods to approach the root. Because of the be-
t~aviour of AR in regions not very close to the root., the more sophisticated root
predicting techniques encounter problems that are difficult to avoid i n advance,
)yhile the phase test with intervM division is virtually foolproof and is, in most
cases, faster. However, once the root is near, a~ becomes smooth and t h e linear
it}terpolation method converges very quickly onto the root.
The computation of matrix elements and the matrix multiplications are best

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/57/6/1299/2694885/BSSA0570061299.pdf

by Uppsala Universetet user
 FAST PROGRAMS FOR LAYERED HALF-SPACE PROBLEMS 1313

Subroutine !NDiV

-Enter
'[--- (given C ~l and She ~arame%ers of the model. )

I •
>
F!oa%ing point overflo~
I
I re~la~e ~_o~es~ ! a ~ r 1
!
I
[ wi~h half space I

t
I I
L___(~) iu , 1

I.~o--~,1 !
< (
il initialise V and correct
to V' if liquid layer

~--~1__
,
l, ~ >
, I
L
I
I
*I
Replace each
I
'
! 1
!

I
~ est ~X~ ~Jinore~ont ~
• g J c l o s e enough t o r o o t ? / ] - - ~ t ° converge on r o o t
a~rop, iaely

Re%urn
Fio. 1. Flow chart for the subroutine INDIV.

handled in machine assembly language (MAP for the 7094) to avoid the slow com-
plex arithmetic of FORTRAN. Additional advantages are that mueh redundant
computation is avoided since many of the numbers involved are either real or
imaginary and, most importantly, it is possible to modify the 7094% Floating Point
Trap feature. At sufficiently high frequencies, overflow can be avoided by modifying

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/57/6/1299/2694885/BSSA0570061299.pdf

by Uppsala Universetet user
 1314 BULLETIN OF T H E SEISMOLOGICAL SOCIETY OF AMERICA

the model as described by Press et al (I961). Using the modified Floating Point
Trap to reduce the number of layers only after an overflow is detected ensures that
the frequency is high enough for the approximation to be valid.
This program, based as it is on models with flat homogeneous layers over a half-
space, is designed for relatively shortperiod studies of crustal and upper mantle
structure, as in the study of Berry and Knopoff (1967). Corrections to take account
of the earth's sphericity, for example that of Alterman et al (1961), may be applied
to extend the range of usefulness to periods of the order of 300 seconds.
The inversion of Rayleigh dispersion data is not unique and investigation of the

4.0
SUBCONTINENTAL
CONTINENTAL S H I E L D

1.0 [ I I I ! I I_ i i t I i I I i i I i I I I I I t I
O.I 0.2
f (cps)
FIG. 2. Transfer functions for typical subcontinental and continental shield erustM models
for values of p = dt/dA = 5.0, 7.0, 9.0 see/degree. Tl, z is the amplitude ratio of the vertical
component of motion at the free surface to the incident P wave. At f = 0, Tpz decreases as
p increases.
permissible variations of model parameters is best performed using least squares
techniques, which require knowledge of partial derivatives or the form Oc/Oa~,
Oc/O~, Oc/Oh~. An experimental version of the program which will modify a given
model to give the best fit to a set of dispersion data and which estimates the uncer-
tainties of the derived parameters has been operated with fair success. The lack of
uniqueness requires that the variation of parameters be constrained in some way;
the choice of constraints requires some insight and there seems to be little hope of
automating the process completely.
Observations of the attenuation of Rayleigh waves as a function of frequency
may also be inverted to obtain a radial distribution of the specific attenuation
factor QR if the latter is assumed to be independent of frequency (Knopoff, 1964b).
H e r e again partial derivatives of the form Oc/Oo~, Oc/O#~ are needed. Numerical
values of these derivatives are readily obtained with the Rayleigh dispersion pro-
gram.

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/57/6/1299/2694885/BSSA0570061299.pdf

by Uppsala Universetet user
 FAST PROGRAMS FOR LAYERED HALF-SPACE PROBLEMS 1315

The matrix method for P-P reflection has been programmed, again using the
frequency separation feature, to form a "black box" subroutine which performs on
the Fourier transform of an input time series, the transformation effected by P:P
reflection from the crustal model. An analogous "black box" subroutine has been
based on the matrix method for crustal transfer. These subroutines are being used in
studies of crustal structure based on comparisons between the spectra of P and pP,
to investigate the corrections that must be made to observations of amplitudes of

0.8 - - 5 . o ~ ¢

o.6 LL_____ ~.~ ~o9

7.5

. . . . . 8.5
9.0
o.4
O. I 0.2

FIG. 3. Reflection coefficients for P-P f (cps)

reflection from b e n e a t h an average oceanic
s~ructure.

A B

i t ,NPuT

" - - I O SECONDS "1

FIG. 4. S y n t h e t i c seismogram representing the effect on a s t a n d a r d short-period Benioff
i n s t r u m e n t due to an originally s h o r t pulse after passing t h r o u g h an a t t e n u a t i n g m a n t l e and a
layered crust t h a t incorporates a low velocity channel. A corresponds in time to a phase re-
flected between the surface and the b o t t o m of the low velocity channel, while B corresponds
in time to a phase reflected between the surface and the b o t t o m of the crust.

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/57/6/1299/2694885/BSSA0570061299.pdf

by Uppsala Universetet user
 1316 BULLETIN OF THE SEISMOLOGICAL SOCIETY OF AMERICA

P waves to take account of crustal structure, and to generate synthetic seismograms

in order to study the constraints that are imposed on the source time function by
observed P pulses. Figures 2 to 4 exemplify some of the possibilities of these sub-
routines.
The Rayleigh dispersion program has now been written in PL1 for the IBFI 360
computer.
ACKNOWLEDGMENT

This research was supported by grant ~ AF-710-67 of the US Air Force Office
of Scientific Research. The author wishes to thank the New Zealand Department
of Scientific and Industrial Research whose National Research Fellowship has
made possible his stay in the United States.

REFERENCES

Alterma~l, Z., H. Jarosch and C. L. Pekeris (1961). Propagation of Rayleigh waves in the earth,
Geophys. J. 4, 219-241.
Berry, M. J. and L. Knopoff (1967). Structure of the upper mantle under the western Medi-
terranean basin, J. Geophys. Res. 72, 3613-3626.
Haskell, N. A. (1953). The dispersion of surface waves on multilayered media, Bull. Seism.
Soc. Am. 43, 17-34.
Knopoff, L. (1964a). A matrix method for elastic wave problems, Bull. Seism. Soc. Am. 54,431-
438.
Knopoff, L. (1964b). Q, Rev. Geophys. 2, 625-660.
Press, F, D. G. ttarkrider and C. A. Seafeldt (1961). A fast convenient program for compu-
tation of surface-wave dispersion curves in multilayered media, Bull. Seism. Soc. Am. 51,
495-502.
Thomson, W. T. (1950). Transmission of elastic waves through a stratified solid medium, J .
Appl. Phys. 21, 89-93.
Thrower, E. N. and D. G. ttarkrider (1965). A note on the existence of relative maxima and
minima on phase velocity curves, Bull. Sesim. Soc. Am. 55, 971-974.

INSTITUTE OF GEOPHYSICS AND PLANETARY PHYSICS

UNIVERSITY OF CALIFORNIA
Los ANGELES, CALIFORNIA
PUBLICATION I~UMBER 595

Manuscript received May 1, 1967.

Downloaded from https://pubs.geoscienceworld.org/ssa/bssa/article-pdf/57/6/1299/2694885/BSSA0570061299.pdf

by Uppsala Universetet user