Vincit Veritas: A Portrait of the Life and Work

Special Publications of Norman Abraham Haskell, 1905–1970 Vol. 30

Reprinted from. BULLET'INO1•THE SEISMOLOGICAL

SOCIETYOr A•EaIC.•
VOLUMe43 ß NUMSER1 ß JXNUAr•Y•1953
THE DISPERSION OF SURFACE WAVES ON
MULTILAYERED MEDIA*

By N. A. HASKELL
ABSTRACT

A matrix formalism developedby W. T. Thomson is used to obtain the phase velocity disper-
sionequationsfor elastic surfacewavesof Rayleigh and Love type on multilayered solid media.
The method is usedto computephaseand group velocities of Rayleigh waves for two assumed
three-layer modelsand one two-layer model of the earth's crust in the continents.The computed
group velocity curves are comparedwith published values of the group velocities at various
frequenciesof Rayleigh waves over continental paths. The scatter of the observed values is
larger than the difference between the three computed curves. It is believed that not all of this
scatter is due to observational errors, but probably representsa real horizontal heterogeneity
of the continental crusts.

INTRODUCTION

IN THE usual treatmenU of the dispersionof surfacewaves of Rayleigh type on a

layered medium, the dependenceof phasevelocity upon wave length is expressed
by the vanishingof a certaindeterminantwhoseelementsare functionsof the phase
velocity,the wavelength,and the densitiesand elastictoodullof the variouslayers.
If there are n layers (includingthe last, which is assumedto be semi-infinite),there
are 4n - 2 boundary conditionsto be satisfied' continuity of two displacement
componentsand two stresscomponentsat each interface, and vanishingof two
stresscomponentsat the free surface.These lead to 4n - 2 homogeneous simul-
taneousequationsto determinean equalnumberof unknownconstants.A solution
existsonly if the determinant of the coefficientsvanishes.Although many of the
elementsof this 4n - 2 rowed determinant are zero, the computationallabor in-
volved in determining the roots is so formidable that no attempt appearsto have
beenmadeto treat casesof morethantwo layersby thismethSd.
In the presentpaper the problemis reformulatedin terms of matrices,following
a method introducedby W. T. Thomson.2 Although this may be regardedas no
more than a changein notation, the matrix notation itself suggestsa systematic
computationalprocedurethat makesit possibleto handleat leasta three-layercase
on an ordinary deskcalculatorwithout an unreasonable expenditureof time.
In the caseof Love wavesthere are only two boundaryconditionsto be satisfied
at eachinterface,and the three-layerproblemcan be treated by straightforward
methodswithout excessivealgebraicmanipulation.However, for more than three
layersthe matrix methodmay be advantageousin this casealso.
In order to simplify the discussionas far as possible,we shall considersolutions
of the elasticequationsof motionin the form of planewavesrather than attempt to
treat the more complexcaseof wavesdivergingfrom a point source.So far as the
dispersionfunctionis concerned,
this involvesno lossof generality,sincethe point-
* Manuscriptreceivedfor publicationJune 5, 1951.
• K. Sezawa, "Dispersionof Elastic Waves Propagated on the Surfaceof Stratified Bodiesand
on Curved Surfaces,"Bull. Earthq.Res.Inst. Tokyo,3:1-18 (1927), esp.p. 16; A. W. Lee, "The
Effect of GeologicalStructure upon MicroseismicDisturbance,"Mon. Not. Roy. Astron.Soc.,
Geophys. Suppl.,3:83-105 (1932);C. Y. Fu, "RayleighWavesin a SuperficialLayer," Geophysics,
11:10-23 (1946).
• W. T. Thomson,"Transmission of Elastic Wavesthrougha StratifiedSolid Medium," Jour.
Appl. Phys.,21:89 (1950).
[17]

Copyright American Geophysical Union

 Vincit Veritas: A Portrait of the Life and Work
Special Publications of Norman Abraham Haskell, 1905–1970 Vol. 30

18 BULLETIN OF THE SEISMOLOGICAL SOCIETY OF AMERICA

sourcesolutionmay be developedby integrationof plane-wavesolutions.The plane-

wave solutionwill not, of course,determinethe way in which the relative excitation
of the variouspossiblenormalmodesof propagationvarieswith depth of sourceand
with frequency.
MATRIX FORMULATION OF THE PROBLEM FOR RAYLEIGH WAVES

Sincethere is an easilycorrectederror in ThornsOh'Sdevelopment,we shallrepeat

hisderivationin somedetail, with certainminor changesin notationthat will exhibit
the basicsymmetry of the final expressionssomewhatmore clearly.
We considerplane waves of angularfrequencyp and horizontalphasevelocity c
propagatedin a semi-infinitemediummadeup of n parallel,homogeneous, isotropic
-D• DIRECTION OF PROPAGATION

FREE SURFACE
(o) •4-X
I

(2)
-I-z

n-I
(n-•)
n

Fig. 1. Direction of axesand numberingof layersand interfaces.

layers.For thepresent,all layerswill beassumed

to besolid;thecaseof a fluidlayer
will be considered
later. The x axisis taken parallelto the layerswith the positive
sensein the directionof propagation. The positivez axisis taken as directedinto
the medium.The variouslayersand interfacesare numberedaway from the free
surface,
asshownin figure1. We confineourattentionto wavesof Rayleightype,
by whichwe meanthat thereis no displacement in the y directionand that the
amplitudediminishes
exponentially
in the •-z directionin thesemi-infinite
layer.
For the mth layer let
Pm= density
dm -- thickness
kin,Ym= Lambelasticconstants
Olin
-- [ (kin•- 2•m)/Pm] 1/2= velocityofpropagagionofdilatationalwaves
•m --- [ •m/•)m]1/2• velocityof propagation
of rotationalwaves
k = p/c = 2•/wave length(horizontal)
q-[(C/am)• -- 1] •/• c > a.•
r.m= --ill -- (C/e,m)•]l/•c
< e,.•
q-[ (C/fire)• -- 1]•/• C >•m
r•,• = --i[ 1 -- (½/•m)211/2C
< •m
=
•, to = displacement
components
in x andz directions
• = Z,. = normal stress
• = X,. = tangentialstress

87

Copyright American Geophysical Union

 Vincit Veritas: A Portrait of the Life and Work
Special Publications of Norman Abraham Haskell, 1905–1970 Vol. 30

DISPERSION OF SURFACE WAVES ON MULTILAYERED MEDIA 19

Then, as is well known, periodicsolutionsof the elasticequationsof motion for the

mthlayer may be found by combiningdilatational wave solutions.

Am= (Ou/Ox)
•- (Ow/Oz)
----exp[i(pt-- kx)][A'm
exp(--ikr,•z)-]- A• exp(ikr,mz)]
(2.1)
With rotational wave solutions,

•m=(1/2)[(Ou/Oz)-(Ow/Ox)]=
exp[i(pt-kx)][• exp(-ikr•z)•-•'• exp(ikr•z)]

where A m,
' Am" , (Din
' and (Din
" are constants.With the sign conventionsdefined
above,the term in Am'representsa planewave whosedirectionof propagationmakes
an angle cot-x r•m with the -]-z direction when r•m is real, and a wave propagated
in the -]-x direction with amplitude diminishingexponentiallyin the •-z direction
when r•mis imaginary. Similarly, the term in Am"representsa plane wave making
the same angle with the -z direction when r•m is real and a wave propagatedin
the -]-x directionwith amplitude increasingexponentiallyin the -]-z directionwhen
r•m is imaginary. The same remarks apply to the terms in •m with r•m
substituted for r•m.
The displacementsand the pertinent stress componentscorrespondingto the
dilatation and rotation given by (2.1) and (2.2) are,

u - - (2.3)

W ------(Olm/p)2(OAmlOZ)
+ 2(•m/p)2(O•m/OX) (2.4)

O' = pm[Olm2Am
+ 2•m2{(Olm/p)2(O2Am/OX
2) + 2(•m/p)(O2(Mm/OXOZ)
}] (2.5)

7* •- 2pm•m2[
-- (OIm/p)2(O2AmlOXOZ)
+ •3m/p)2{(O2(MmlOX2)
-- (O2•mlOZ2)}](2.6)

The boundaryconditionsto be met at an interfacebetweentwo layersare that these

four quantitiesshall be continuous.Continuity of the displacements
is assuredif
the correspondingvelocity componentsa and • are made continuousand, since
c is the samein all layers, we may take the dimensionless quantitiesa/c and •/c to
be continuous.Substituting the expressions(2.1) and (2.2) in equations(2.3) to
(2.6) and expressing the exponentialfunctionsof ikrz in trigonometricform, we find

= - + coskramz
- i(A'm
- A'•)sinkra•z]
--?mr•m[(•'m
-- •'m
•) coskr•mZ
-- i(•'mnu •) sinkr•,,z] (2.7)

•v/c= --(o•.•/c)•r•,•[--i(A•m
-]-A'm
•)sinkramZ
•- (A•m
-- z•:)COS
kr•mz]

sinkr•mz
+ ((M'
m+ •'m
l) COS (2.s)

• -- 1)[(Am-I-
• = -- Pm•'m(•m ' A•"
m)coskr•mz-- i( A'm-- Am)
' ' sin

Copyright American Geophysical Union

 Vincit Veritas: A Portrait of the Life and Work
Special Publications of Norman Abraham Haskell, 1905–1970 Vol. 30

2O BULLETIN OF THE SEISMOLOGICAL SOCIETY OF AMERICA

--pmc•T•r•m[(•
-- •o•)coskr•z -- i(•o•+ •o•)sinkr•mz] (2.9)

sin kr=•z q- ( m - m)coskr=•z]

. !
--pmCa7m(q•m
-- 1)[--t(•m-- •m)Sinkr$mZ
q- (0/•q-0•'•')
COSkr$,•z] (2.10)

Whenany of the r's areimaginary,the trigonometric

functionsareto be understood
as goingoverinto the correspondinghyperbolicfunctions.
Placingthe o•iginof z at the (m - 1)th interface,the linearrelationship
be-
tweenthe valuesof a/c, •v/c, •, and T at the (m - 1)thinterfaceand the constants
(Am
! + /•mlt),(Am
! -- Amtt),(OJm
! -- OJmtt),
and (OJm
! + OJm
it) may be represented
by the transformation
• / II / If I II I
('•m-l/d,'•l)m-,/C,
O'm-,Tm-l)
--' t•m(Am
-JI-Am,Am-- Am,O•m
-- O•m,
OJm-JI-O•m) (2.11)
where Em is the matrix

Zm • (2.12)

Settingz =dm in equations(2.7) to (2.10) givesthe valuesof a/c etc. at the mth
interface in terms of Am' q- Am"etc.

Al II I Alii I II I II (2.13)

where Dm is the matrix

- (am/c)2cosPm i(ctm/C) 2sin Pm --'ymT•m COS Qm g•fm•'•m sin

Q•
i(a=/c)•r•=
sinP= - (a=/c)"r• cos
P=--i7• sinQ• 7•cos Q•
--ip=a.?7•r•sinP•
p=a.,"(7•
--1)cosP•ip•a•a(7•--
1)sin
P=--p.,ca7=ar•
cos Q•ip•ca7•r• sinQ•
p•a•"7•r•=cosP= ip•caT•(7•-l) sinQ• -p=c"7•(7•-1) cosQ•
(2.14)
with Pm= kr, mdmand Qm = kr•mdm.

The constantsAm! + Am'! etc. may be eliminatedbetweenequations(2.11) and

(2.12),givinga linearrelationship
between
the valuesof U/c,•/c, a, and ß at the
topandbottomofthemt• layerthatmaybeexpressed symbolically by theequation,
(•n/•, •Om/C,
(Ym,
Tm)---D,•E;•(am_•/c,
•m-1/C,O'm--1,
Tin--l) (2.15)

89

Copyright American Geophysical Union

 Vincit Veritas: A Portrait of the Life and Work
Special Publications of Norman Abraham Haskell, 1905–1970 Vol. 30

DISPERSION OF SURFACE WAVES ON MULTILAYERED MEDIA 21

whereEm-• is the inverseof Emand isgivenby

-- 2(•m/•m)
2 0 (Pm•m2)
-1 0

0 C•(•ym--
I)/•m•ram 0 (pmam•ram)
-•
(2.16)
(Tin- 1)/Tmr6m 0 --(pmC•Tmr6m)
-• 0

_ 0 1 0 (OmC•ym)
-• _

Fromequations(2.14)and (2.16)the elements

of the matrixproductam= DmEm
-•
may be computedas follows:
(am)m1-- •YmCOS
Pm -- (•m -- 1) cosQm

(C/m)12
: i[(•ym
-- 1)ra•sinPm+ •ymr•m
sin
(am)m3
-- --(pm½2)-l(cos
Pm-- COS
Qm)
(C/m)14
---i(Dm½2)-l(r•
SinPm+ r•msinQm)
(am)•i= --i[Tmram
SinPm2ff(Tin-- 1)rt•-m
• sin
(am)• -- --(Tin -- 1) cosPm+ •/mCOS
Qm

(am)•S
-- i(pmC•)-l(ram
S•nPm+ r• sinQm)
(am)•4-- (am)iS

(am)s1----
pmC•Tm(Tm-
1)(COS
Pm- COS
Qm)
(am)s•-- ipmC•[('ym
-- 1)2r•m
-1 sinPmq- •mr•msin Qm]

(am)aS--(am)•

(a.•)s4-- (am)12

(am)41
= ipm½
2['•mr•m
2 sinPm+ (•m- 1)3r•m
-1 sin
(am)42-- (am)am

(am)4S
= (am)•

(am)44'- (am)11
Nowtheboundary conditions
requirethat thevaluesof a/c, •b/c,a, andr com-
putedat thetopofthemtnlayerbethesameasthevalues computed at thebottora
of the (m - 1)thlayer.Thismeansthat wemaywrite

(am/C,
•vm/c,
am,rm)= amam-•(am_•/C,
•Vm_•/C,
am-•,•m-•) (2.17)

Copyright American Geophysical Union

 Vincit Veritas: A Portrait of the Life and Work
Special Publications of Norman Abraham Haskell, 1905–1970 Vol. 30

22 BULLETIN OF THE SEISMOLOGICAL SOCIETY OF AMERICA

In Thomson's paper,the quantityr/2u is takeninsteadof r asthe fourthvariable

on whichthe matrix amoperates.This is merelya changein notationsofar as any
onelayeris concerned, but u will generallybe differentin differentlayers,and it is
the shearingstressr, and not the shearingstrain r/u, that is continuous acrossthe
interface.The iterative procedureindicatedby equation (2.17) thereforerequires
that r, or a constantmultipleof r, ratherthan r/2u, betakenasthe fourthvariable.
Thomson'smatrices a shouldthen be correctedby multiplying the fourth row by
2u = (2Gin Thomson'snotation),andthe fourthcolumnby (2U)-1.
By repeatedapplicationof equation(2.17) we have,

(•.•n--1/½l
•l)n--1/Cl
fin--l,Tn-1)----a,•_•a,•_a
ß ß ßa•(•o/C,•o/C, •o, •'o) (2.18)

andby applicationof the inverseof equation(2.11)for the ntnlayer,

(A' I -- All
n+ A'n',A,• • ,• I -- • II ,• I + •)II -- E•,-1a•_• aa_•' ß ß a•(•o/C,•o/C,
(2.19)

Sofar thedevelopment
hasb•n quitegeneral,andequation(2.19)is equallyappli-
cableto surfacewavesor to wavestransmittedthroughthe layeredmedium.The
casewith whichwe are particularlyconcerned
is that b whichthereare no stresses
across
the freesurface,sothat a0: r0: 0, andthereareno sources
at infinity,so
that • • •: 0. Writbg J for the matrix product E•-•a•_• a•_• ....... a•,
equation(2.19) becomes,
I

= 0, 0)
or, explicitly,
• : Zi•o/C• J12•0/c
= +
!
(2.20)
Wn-- J31•o/Cq- J3•Vo/C
!

w• = Ja1•to/C
q- Ja•dVo/C

By eliminatingA,•'ando•,•'wehave,

•to= Jaa-Jla
Wo Jll- Jal
= Jsx-
Jaa- Jsa
Jax
(2.21)
Sincethe elementsof the matrix J are functionsof the parametersc and k, equation
(2.21)provides
animplicitrelationship
between
c andk, whichisthedesired
phase
velocitydispersion
function.
SOME GENERAL PROPERTIESOF THE SOLUTION

SettingA = a•_•an-•ß ß ßaxandusingequation

(2.16)forEn-•, equation
(2.21)may
be written in the form
--(•to/•Vo)
= K/L = M/N (3.1)

91

Copyright American Geophysical Union

 Vincit Veritas: A Portrait of the Life and Work
Special Publications of Norman Abraham Haskell, 1905–1970 Vol. 30

DISPElSSIONOF SUlkFACEWAVES ON MULTILAYERED MEDIA 23

where

L = q,•r•A• q- (% - 1)A,• -r•,A3•/p•c '• q- A4•/p•c•

(a.=)

In the two-layer case,equations(3.1) and (3.2) may be shownto be equivalentto

the expressionsthat have previouslybeenderivedby Sezawa,Lee, and others.
In the expressionsfor the elementsof the matricesa• it will be observedthat the
quantitiessin P•, sin Qm,r•m,and r•m,which may be either real or imaginaryde-
pending
uponthevalueofc,occuronlyin thecombinations
r •-•
•m• sinP•, andr •-•
t•msinQ.
Sincesin Pm is real or imaginaryaccordingas r•mis real or imaginary,and sin Qmis
similarly relatedto rt•m
, thesecombinationsare alwaysreal for real valuesof c. With
regardto the real or imaginarypropertiesof its elementsthe matricesamthen have
the form -- --

R I R I

I R I R
am --
R I R I

I R I R

wherean R indicatesa real quantity (not, of course,the samequantity in all posi-

tions)and an I indicatesan imaginaryquantity. The productof any two matrices
of this form is alsoa matrix of the sameform; henceof the elementsof A occurring
in equations(3.2)'
Ajj, A •.=,A3•,andAa•.are real;
A•=,A •.•,A•., and Aa•are imaginary.

By definitiona "surface"waveis onewhoseamplitudediminishes

for largevalues
of z, which meansin our casethat r•, and r•, must be imaginary,that is, we are
concernedonly with valuesof c < •S•.Then, referringto equation(3.2), all terms
of K and N are real and all terms of L and M are imaginary.Thus the ratio
will alwaysbe imaginary,which meansa phasedifferenceof 90ø betweenthe hori-
zontal and vertical displacementsat the free surface.The particle motionis there-
fore an ellipsewhoseaxesare vertical and horizontal.The phasedifferencemay,
however,be of either sign,and hencethe senseof the motion aroundthe ellipse
is not necessarilyretrogradewith respectto the directionof propagation at all fre-
quencies, asis the casewith Rayleighwaveson a homogeneous medium.
If we weredealingwith a dissipativemedium,it wouldbe necessary to consider
complexvaluesof c and k. In that case,the ratio •o/tb0wouldnot necessarily be a

Copyright American Geophysical Union

 Vincit Veritas: A Portrait of the Life and Work
Special Publications of Norman Abraham Haskell, 1905–1970 Vol. 30

24 BULLETIN OF TIlE SEISMOLOGICAL SOCIETY OF AMERICA

pure imaginary, meaning that phase differencesof other than 4-90ø could occur
and the axes of the displacement ellipse could be inclined from the vertical. It is
thereforepossiblethat imperfectelasticity of the medium is the causeof the inclina-
tion of the axes that is very commonlyobservedin the caseof explosionexcited
surfacewaveson poorlyconsolidated
sediments.
It is obviousthat if two adjacentlayershave identicalphysicalproperties,they
must be equivalent to a singlelayer whosethicknessis equal to the sum of the thick-
nessesof the two layers. Thus, if we let am(d)be the matrix amcomputedfor a given
layer thicknessd, we must have

am(all)am(d2)
-- am(d,-• 42) (3.3)

This relation may be readily verified by direct multiplication. Also, since k occurs
in amonly asthe productkdm,equation(3.3) implies,

am(kl)am(k2)-- am(kl -• (3.4)

where kx and k2 are any two values of k and am iS computed for fixed values of c
and dm.
ASYMPTOTIC FORM FOR LONG WAVE LENGTHS

As the wave length becomesvery large, kdm-• 0 and all the matrices amapproach
the unit matrix. Thus J• -• E• -1 and equation (2.21) reducesto

•/•o -- -- (• -- 1)/w•r•n = 'ynr•n/('Yn-- 1) (4.1)

or

('Yn-- 1)• •- 'yn•ranr•n

= 0 (4.2)

which is the equation for the Rayleigh wave velocity on the semi-infinitelayer.
If we expandthe termsof the matricesamin powersof k and ignorepowershigher
than the first, the matrix A reducesto
n-1 n--1

1 ik •dmi
0 ik •dm/Pm•m
I
2
n-1 n-1

ik •'•dm[1
1
- 2(•m/(Xm)
2] 1 ik•dm/pme•m
1
2 0
A --> (4.3)
n-1

ikc2•dmpm i ik •'•dm
1

n-1 n-1

ikc2•dmpm[1
--2•[m-•2'ym(•m/Otm)
i
2] 0 ik •dm[1-2•m/e•m)
2] I

Thusforwavelengths
solongthatk2(•. dm)
• maybeignored,
theam's
commute
1

and the orderin whichthe upperlayersare arrangedis immaterial.To this orderof

approximationthe quantitiesK, L, M, andN are linearfunctionsof k. Hencefor a
givenvalueof c equation(3.1) is a quadraticin k and may be solvedexplicitly.

Copyright American Geophysical Union

 Vincit Veritas: A Portrait of the Life and Work
Special Publications of Norman Abraham Haskell, 1905–1970 Vol. 30

DISPERSION OF SURFACE WAVES ON MULTILAYERED MEDIA 25

ASYMPTOTICFORM FOR SHORTWAVE LENGTHS

It hasbeenshown
by Sezawa
andKanai• that in thetwo-layer
casethehigh-fre-
quencyasymptoticform of the phasevelocityequationmay be factored.One of
thesefactorshasa rootcorresponding to the Rayleighwavevelocityonthe free
surfaceofthefirstlayer;theotherisStoneley's4expression
forthevelocityofinter-
facewaves onthecontact between thetwolayers. Thelattermayormaynothave
a realroot,dependingupontherelativevaluesof p, a, and• in thetwolayers.It
seems evidentonphysical grounds that in themultilayercasethe phasevelocity
equationmustalsobefactorable at sufficiently
highfrequencies,
thevariousfactors
representingRayleighwavesonthefreesurface andStoneley wavesoneachinter-
face. In order to demonstratethis it will be convenientto write the matrix J in
the form,
J = b._•b._•. ß ßb•E•• (5.1)
where
bm -- Em+•
- 1 Dm (5.2)
Thatis,insteadofgroupingthematrixfactors ofJ bylayers,
wenowgroupthem
by interfaces.
Nowsuppose thatc < •.-1, sothatP.-1 andQ._•areimaginary
and
thesinesandcosines
representhyperbolicfunctions.
Thenforlargevaluesofkd._•,
sinPn-•-• -i cosP._• andsinQ._•-• -i cosQ._•.In thislimit,theelements
of
b._•approachthe followingvalues:

(bn-1)11-- -(bn-,)12= (Otn--l/Otn)2{')/'n

-- ('¾n--1
-- 1)(pn--l/pn)}COSP,_•
(b.-•)xa= - (b.-,)x•= (c/a.)•7._,r•(•,){•. -- •n--l(Pn--1/pn)
} COSQ---X
(b._•)•= -(b._•)•e= (a._•/a.)•(r•(._•)/r•.) {(y. - 1) - •.-i(p.-x/p.)} cosP._•
(bn-1)23= -- (bn-1)24
= -- (c/•.)•(•._•/r•.){(•. -- 1)- (•._•- 1)(p._•/p.)}cosQ._•
(b.-•)ax= -(b._x)ae = -(•._x/c)•(•.r•.)-•{(•. - 1)- (•.-1 - 1)(p._•/p.)}cosP._•
(bn-l)33= -(bn-•)34= (•n--lrfi(n--1)/•nrfin){(•n
-- 1) -- •n--I(Pn--•/Pn)}
cOSQn-•
(b.-,)4x= -(b._,)• = (a._,/c)•(ra(._•)/•.){•.- •._•(p._,/p.)}cosP._•

If wesetJ,•_•= b,•_2
b,•-aßßßb,E•
-•, thensince
(b._•)j•= - (bn_•);•
and(bn-x)•'a
=
-(b._•) •'4for highfrequencies,
wemaywrite

• - •,• = [ (•._•)•, - (b._,)•,1[(•._,).. - (•._•)•]

+ [(•_•)•- (•_,)•] [(•_,)•-

+ [(b._•),a- (b._x)e•][(J._,)•x- (J._•),x]

• - •,• = [(b._,)•,- (v._•),•][(•._•),, - (•._,)•,]

aK. Sezawa andK. Kanai,"Anomalous

Inst. Tokyo,16:683 (1938).
Dispersion
ofElastic
Surface
Waves,"
Bull.Earthq.
Res.
4R. S.Stoneley,
106:416 (1924).
"Elastic
Waves at theSurfaee
ofSeparation
ofTwoSolids,"
Proc.Roy.Soc.
'

94

Copyright American Geophysical Union

 Vincit Veritas: A Portrait of the Life and Work
Special Publications of Norman Abraham Haskell, 1905–1970 Vol. 30

26 BULLETII• OF TI-IE SEISMOLOGICAL SOCIETY OF AMERICA

Setting

(5.3)

R -- (bn-1)ll -- (bn-1)21
(5.4)

therelation(2.21)betweentheelements
ofJ maybewrittenin theform
RK' q- SM' TK' q- UM'
RL' q- SN' TL' q-

Cross-multiplication
andcancellation
reduces
thisequationto the factoredform,

(RU - ST) (K'N' - L'M') = 0 (5.5)

Equatingthe first factorof thisexpression

to zeroandusing•he valuesof the
elementsof b,_xgivenabove,the commonfactorcosP,-x cosQ,_• may be divided
out and the resultingexpression
put in the form,

whichis equivalentto Stoneley's equationfor the (n - 1)thinterface.

The secondfactor of equation(5.5), whenequatedto zero,is equivalentto the
originalexpression,exceptthat it refersto (n - l) insteadof • layers.The same
processmaytherefore berepeated, leadingto a productof factorseachcorrespond-
ingto Stoneleywavesononeof theinterfaces, anda finalfactorwhichhastheform
of equation(4.2) for the firstlayerandthusrepresents Rayleighwavesonthe free
surfaceof this layer.
It will be notedthat a completealgebraicfactorizationat highfrequencies
by the
foregoingprocess canbecarriedthroughonlywhenc is lessthanthesmallest value
of f•m.For valuesof c greaterthan the minimumvalueof firetherewill be at least
onelayerforwhichQmisa realquantity.In thiscase,theratiosK/L andM/N will
not apporach an asymptotic valuefor largevaluesof k, but will remainoscillatory
functionsof k. Thusfor a givenvalueof c greaterthan the smallestvalueof fireand
lessthan f•, there will generallybe an infinitenumberof valuesof k for which
KN - ML - 0. Theserootsrepresent the sequence
of normal-mode solutions.
Since
thereis onlyonerootat verylowfrequency,eachof thesehigherordermodesmust
have a low frequencycut-off.

95

Copyright American Geophysical Union

 Vincit Veritas: A Portrait of the Life and Work
Special Publications of Norman Abraham Haskell, 1905–1970 Vol. 30

DISPERSION OF SURFACE WAVES ON MULTILAYERED MEDIA 27

THE MATRIX am fOR A FLUID LAYER

In somecasesof seismological interest one of the upper layersmay be a fluid. If we
go directly to the limit •m- 0, a difficulty arisesbecausethe matrix Em as defined
by equation (2.12) becomessingular and Em-• doesnot exist. However, we may
definean effectiveinversetransformationas follows'
Setting
•m- 0, equation (2.11) becomes
•m--1/C--- --(Ogm/C)2(Arnt+
iVm_,/C-- -- (am/C)Sr•m(Am ' --
O'm--I --- pmOgm2(
Am! +
Tm-] = 0

Sincein an ideal fluid continuity of the tangential displacementat a solid boundary

is not a required condition, the first of these equationsis irrelevant and we may
write,
!

am+ am:
!

- am--
Also, sincerotational wavesdo not exist in the fluid, •oa' = •om"= 0. The transfor-
mation (am' q- a"m • Am' -- AmIt, OJm ' -- OJm
1!, CO
m! + OJm
It) • (Urn--l/C,
•Om--]/C,
O'm--1,
r•_•), which is the effectiveinverseof Em,thereforehasthe matrix,

0 0 (pro--m2)
-1 0

0 --(C/t•m)2r•m
• 0 0
F -1 (6.0
o o o o

o o o O _

For/•m = 0 the matrix Dm takes the form

(--am/C)•' COS
Pm i(a,•/c)•'sin Pm 0 0

i(am/c)Zr•sin Pm --(Crm/C)Zr•m
COSPm 0 0
Dm --- (6.2)
2 ß 2
PmamCOSPm --ZPmam
sin Pm 0 0

_ 0 0 0 O_

Copyright American Geophysical Union

 Vincit Veritas: A Portrait of the Life and Work
Special Publications of Norman Abraham Haskell, 1905–1970 Vol. 30

28 BULLETIN OF THE SEISMOLOGICAL SOCIETY OF AMERICA

from which we find

0 tr•msinPm -(pmC
a) • COS
Pm 0

0 cosPm ir•m(pmC•)
-• sinP,• 0
am= DmF••= (6.3)
ß 2 -1
0 tpc r•m sin P,• cos Pm 0

_0 0 0 0_

COMPUTATIONAL PROCEDURE

Numerical computation of the function c(k) from equation (3.1) must be carried
out by a processof trial and error. It will usually be convenientto take the thickness
of the first layer as the unit of length, o2as the unit of density, and/• (or a• if the
first layer is a fluid) as the unit of velocity, in which casethe result of the computa-
tion will be a relation betweenthe dimensionless quantities c/t•x and kd•. Since the
coe•cients of the functionsof Pm and Qmin the matricesamdependonly on c and
the constantsof the medium, and are independentof k, lesscomputationis required
to determinethe value of k for a given value of c than for the reverseprocess.The
limits of possiblevalues of c are given by the Rayleigh velocitiesfor the nth layer
and for the lowestvelocity layer present.A set of values of c at convenientintervals
betweenthese limits may then be chosenand the quantities r,m, r•m, and ym for
eachlayer computedfor eachvalue of c.
From previously computedc(k) functionsfor two-layer cases(e.g., Sezawa's
curves)it will usually be possibleto make a preliminary estimate of the value of k
for a givenvalue of c that will be at leastwithin an orderof magnitudeof the correct
value. With this as a first trial value, the elementsof the matricesam,the pertinent
elementsof the productmatrix A, and the ratiosK/L and M/N may be computed.
By repeatingthe computationwith otherassumedvaluesof k and plottingthe values
K/L - M/N againstk it will usuallybe possibleto determinethe root to one-tenth
of one per cent with four or five trial values.Sincethe computationsinvolve.taking
differencesbetween quantities of comparablemagnitude at several stages,it is
advisableto carry at least two more significantfiguresin the computation than is
desired in the final value of k.

ALTERNATIVE FORMULAS FOR THE Two-LAYER CASE

If the expressions(3.2) are substitutedin the equationKN - LM = O, certain

terms in the productscancelout and the equationmay be written in the form,

[ (•,, - 1)" q- y,,"r.,,r•] [A,.•A• - A {•A,.•] •- (p•c•)-i[ (y• - 1) q- y,,r•,,r•,,]

[A•A• + A,.,.Aa•- Ai{A• - A•.iAs•] q- r•,,(p,,c•)-I[A{•.As{- A•As•]
q-r•,•(p,•c'-)-•[A•A•- A,.,.A•] q- (p,•cZ)-•(1q-r•,•r•)[A•A4•.- A•A•] = 0
(8.1)

In the caseof morethan two layersthere doesnot appearto be any advantagefor

numericalcomputationsin writing the phasevelocity equationin this form, but in

Copyright American Geophysical Union

 Vincit Veritas: A Portrait of the Life and Work
Special Publications of Norman Abraham Haskell, 1905–1970 Vol. 30

DISPERSION OF SURFACE WAVES ON MULTILAYERED MEDIA 29

the two-layercase(n = 2, A - a•) insertion

of the explicitexpressions
for the
elementsofthematrixa leadsto themoreconvenientform,
-1 -1
C• - C•.cosP• cosQ• -•- Csr•r• sinP• sinQ• - C4r;•cosP• sinQ•

-C•r•-•• sinP• cosQ• - 0

wherethecoefficients
C•throughC5aregivenby thefollowing
expressions'
C• - 2•(• - 1)(A - D • E)
C• - A -]- C•
22 2x 3
as- A([w•- 1127ylr•r•)- 2B(p•//p2)([y•-
1]s-]
-
+ + - +
4 2

2 2 2
Ca - - +
2 2

A = (72- 1)2 •
B = (72- 1)-]-
D = (p•/p2)(2'•- 1)B
Z = (p•/p2)2(1
• r,•2r•2)'y•('y1-
1)
Equation(8.2)is particularly
usefulbecause
it canbesolved explicitly
forthe
higher-order
roots.For• • c • a•,P• isimaginary
andQjisreal,sothatsinP• and
cosP• becomeverylargeinabsolutevalueformoderately
largevaluesofkd•,while
I sinQ•Iand]cos
Q• remain
• 1.Dividing
equation
(8.2)through
bycos
P•cos
Q•,
dropping
thesmalltermC•/cos
P•,andnoting
thattanP•-• -i forlargevalues
of
kd•,we find,
ß -1
tan Q• -• r•l(•r• C5 -- Ce)/(•r,,
• ' -• Cs+ C4) (8.3)
or

kdl• r•' tan-•{r•(•r•

' -•C• -- C2)/(zr,•Cs
' -• + C4)}+ nwr;• (8.4)
where
n = 1,2, 3, ßßßetc.andtan-1isgivenitssmallest
positive
value.
Low W•ws

In thecaseof Lovewavestheboundawconditions
to besatisfied
at eachinterface
arecontinuity
ofthetransverse
component
ofdisplacement,
v,andofthetransverse
shear
stress,
Y,.Thepertinent
plane-wave
solution
oftheelastic
equations
ofmotion
for a homogeneous
layer is
u=w=O

v - exp[i(pt -- kx)][v'exp(-ikr•z) -]-v"exp(ikr•z)] (9.1)

where v' and v" are constants.

98

Copyright American Geophysical Union

 Vincit Veritas: A Portrait of the Life and Work
Special Publications of Norman Abraham Haskell, 1905–1970 Vol. 30

30 BULLETIN OF THE SEISMOLOGICAL

SOCIETYOF AMERICA

Thecorresponding
transverse
shearing
stress
is
Y, = •Ov/Oz
= ik•r•exp[i(pt-- kx)][--v'exp(-ikr•z) + v"exp(ikr•z)
]
At the (m - 1)thinterface
wethenhave

(•/C)m--]
---
ik(vm'
+•)m
it) }
(Yz)m-1- iklzmr•m(Vm"
-- Vm')
(9.3)

and at the m th interface

(•/C)m---(Um!
+ Vm")ik
cosQm-- (•)m
!! -- Umt)
k sinQm
(9.4)
(rz)m--' -- (Um!
+ Vm")klzrBm
sinQm+(Um
t! -- Vm')ik•zmr•m
COSQm
Byeliminating
v•! andUm
t!between
equations
(9.3)and(9.4),

(•m/C)m
=(•/C)m--1
COS
Qm
•-(Yz)m-ltZm-lr•
-li
sin
Q•
(r..)m-- (9/C)m--lilZmr•m
sinQm•- (rz)m--1
COSQm
(9.5)

The matrix amin this caseis therefore

am
----
I COSilz.•-lr•
Qm -1
sin
Qm
]
[ilzmr•msinQm cosQm

=cosQm
[ 1 itzm-lr-l•mtan
1 ilZmrt]m
tan Q• 1
(9.6)

Setting
an_,
an_•ßßßa•= A asbefore,
theanalog
ofequations
(2.18)
is

(•/C)n--1
---
All(•/C)o
+A•(r,.)o
}
(r..)n-• = A•,(i•/C)o
+ A::(r.)o
(9.7)

andusingequation(9.3) for m = n

vn! + v•" = A•(ik)-l(b/C)o+ A•:(ik)-l(Y,.)o

(9.8)
v•" -vn' = A•(ik•r•)-l(b/c)o + A•(ik•r•)-l(Y'.) o
Theconditions
fortheexistence
of freesurface
wavesare(Y,.)0= 0 andv•" = 0,
which,withequations
(9.8)leadto
A• = - •,,r•,,A,• (9.9)

Copyright American Geophysical Union

 Vincit Veritas: A Portrait of the Life and Work
Special Publications of Norman Abraham Haskell, 1905–1970 Vol. 30

DISPElSSION OF SUlkFACE WAVES ON MULTILAYERED MEDIA 31

In the two-layer case,A = a•, and equation (9.9) reducesto the Love wave dis-
persionequationin the familiar form.

(9.10)

APPLICATION TO THE RAYLEIGH WAVES OF EARTHQUAKES IN

COnTInenTAL AR•AS

Considerablymore detailed study of the dispersionof Rayleigh waves over conti-

nental paths will be required before it will be possibleto make any conclusive
quantitative interpretation of the observeddispersionin terms of continentalstruc-
ture. However, as a preliminary step in this direction, phase and group velocities
have beencomputedfor assumedlayered structuresas shownin table 1. The values
of a, •, and d have been sochosenthat the time-distancecurvesfor the first arrivals
of P and S wavesare the samein all casesexceptfor that portion of the curve that
corresponds to refraction from the intermediatelayer. Case III is intended to be
typical of the interpretationsthat have beenmade from blast and near earthquake
recordingsin areasof little or no low velocitysedimentarycover,in particularthose

TABLE 1

Layer a (kin/sec.) fi(km/sec.) p(gm/cm3) d (km.)

Case I ..................... 1 6.14 3.39 2.70 13.60

2 5.50 3.18 2.70 11.85
3 8.26 4.65 3. O0 •o

Case II .................... 1 6.14 3.39 2.70 28.38

2 8.26 4.65 3. O0 •o

Case III ................... 1 6.14 3.39 2.70 13.60

2 7. O0 4.04 2.70 21.21
3 8.26 4.65 3. O0 •o

TABLE 2

Layer a fi p d

Case I .................. 1 1.810 1.000 1.000 1.000

2 1.620 0.938 1.000 0.871
3 2.440 1.370 1.110 •

Case II ................. I 1.810 1.000 1.000 1.000

2 2.440 1.370 1.110 •

Case III ................ 1 1.810 1.000 1.000 1.000

2 2.060 1.190 1.000 1.560
3 2.440 1.370 1.110 •

lOO

Copyright American Geophysical Union

 Vincit Veritas: A Portrait of the Life and Work
Special Publications of Norman Abraham Haskell, 1905–1970 Vol. 30

'32 BULLETII• OF THE SEISMOLOGICALSOCIETY O1•AMERICA

TABLE 3

Case I

c/fi kd• k(d• • d2) •/•o

1.265 0 0 -0.670 i
1.250 0.0647 0.1211 -0.744 i
1.200 0.405 0.758 -0.849 i
1.150 0.729 1.364 -0.716 i
1.100 0.962 1.800 -0.645 i
1.050 1.186 2.219 -0.619 i
1.000 1.461 2.734 - 0.619 i
0.950 1.920 3.592 -0.638 i
0.938 2.126 3.978 -0.645 i
0.920 2.817 5.271 -0.662 i
0.920 5.00 9.36 -0.669 i
O. 924 •o •o -0. 667 i

Case II

Raleigh mode First mode

1.265 0 1.370 2.383

1.250 0.1339 1.350 2.957
1.200 0.895 1.300 4.303
1.150 1.550 1.250 5.510
1.100 2.029 1.150 7.716
1.050 2.518 1.100 9.397
1.000 3.167 1.050 12.766
0.950 4.489 1.030 15.882
0.930 6.381 1.010 25.54
0.924 • 1.000 •

Case III

1.265 0 0 --0.670 i
1.250 0.0717 0.1836 -0.710 i
1.200 0.5123 1.311 -0.828i
1.150 0.8634 2.210 -0.746 i
1.100 1.197 3.064 --0.699 i
1.050 1.551 3.971 -0.672 i
1.025 1.954 5.002 --0.656 i
1.000 2.365 6.054 -- 0.649 i
0.975 2.929 7.498 --0.647 i
0.950 3.837 9.823 -0.652 i
0.930 5.76 14.75 --0.662 i
0.924 • • -0.667 i

lol

Copyright American Geophysical Union

 Vincit Veritas: A Portrait of the Life and Work
Special Publications of Norman Abraham Haskell, 1905–1970 Vol. 30

DISPElSSIONOF SUlkFACEWAVES ON MULTILAYERED MEDIA 33

FIRST NORMAL MODE

RAYLEIGH
MODE •

1.o 1o.o

k(d,+d,)
Fig. 2. Phasevelocity of Rayleighwavesfor assumedcrustalstructures.

FIRST MODE

+ 4' 4' = 6UTENBERG AMDRICHTER(1936):N.AND$.AM,

I:1
, LYNCH
AND
DILLON
(1947):
ALASKA-FORDHAM
• ,&- WILSON
AND
$AYKAL
(1948)

I • •, S 7 10 •0 • ( sec ) SO 70 100 •C)0•OO •00 1'001000

¾•g. 3. •roup velocity o• R•yle•gh w•¾es for •ssumed crustal structures.

102

Copyright American Geophysical Union

 Vincit Veritas: A Portrait of the Life and Work
Special Publications of Norman Abraham Haskell, 1905–1970 Vol. 30

34 BULLETIN OF THE SEISMOLOGICAL SOCIETY OF AMERICA

of Leer, Hodgson,and Tuve.5 Case I has been computedto show the effect on
Rayleighwave dispersion of a low velocityzonewhoseexistencehasbeensuggested
by Gutenberg, 6 and II represents an intermediatecasebetweenI and III for com-
parisonpurposes.The computationshave been carried through in dimensionless
formwith f]•,p•,andd takenrespectively asthe unitsof velocity,density,andlength
in eachcase.The valuesof the ratiosactually usedare givenin table 2.
The computedvaluesof kd•,k(dx•- ds),and of •0/•v0for variousvaluesof c/f]• are
listed in table 3. With the signconventionsusedhere a negativeimaginary value of
•0/•v0 correspondsto retrogradeparticle motion. Only the mode of lowest order
(Rayleighmode)hasbeencomputedfor Cases! and III, but the next highermode
is alsogivenfor CaseII.
The phasevelocitiesof the Rayleighmodesfor the threecasesand the first normal
modefor CaseI! are plotted in dimensionless units in figure2. The curvefor CaseI
hasa minimum at c/f]• -- .918, k(d• •- ds) -- 7 approximately.
The corresponding groupvelocities,computedfrom the expressionU--c•-kdc/dk
by graphicaldifferentiation,are plotted againstthe periodT -- 2•-/kc in figure3.
Althoughthe computationof the rootscorresponding to the first normal modehas
not beencarriedthroughfor CasesI and III, roughestimatesindicatethat the low
frequencycut-offfor thesecaseswill not be very differentfrom that for CaseII.
Some observedvalues of group velocitiesat various frequenciesover continental
pathsarealsoplottedin figure3. Thesevalueshavebeentakenfromtabulations
publishedby Gutenberg,Gutenbergand Richter, Lynch and Dillon, and Wilson
and Baykal.7 In view of the scatterof the observedpointsit is not possibleto say
that any one of the assumedmodelsis conspicuously favored over the others,but
the data do not disprovethe hypothesisof the possibleexistenceof a low-velocity
layerunderat leastsomeparts of the continents.Someof the scatterof the observed
valuesis no doubt due to observationalerrorssuchas misidentificationof wave type
and erroneousdetermination of periodsdue to interferenceby the simultaneous
arrival of higher frequenciesin the higher-ordermodes,but someof the scatter is
probablydue to real horizontalinhomogeneity of the continentalcrusts.Certainly
the discordance between the crustal structures derived from P and S travel-time
data from blasts and near earthquakesin different areassuggestsa comparatively
high degreeof heterogeneityin the layersabovethe Mohorovi•i• discontinuity.
• L. D. Leer, "Trial Travel Times for NortheasternAmerica," Bull. Seism.Soc.Am., 31: 325-334
(1941); J. H. Hodgson, "Analysisof Travel Times from Rockburstsat Kirkland Lake, Ontario,"
Bull. Seism.Soc. Am., 37:5-17 (1947); M. A. Tuve et al., "Studies of Deep Crustal Layers by
ExplosiveShots," Trans. Am. Geophys.Union, 29:772 (1948).
6 B. Gutenberg, "The Structure of the Crust in the Continents," Science,111:29 (1950).
?B. Gutenberg,HandbuchderGeophysik, Vol. 4 (Berlin, 1932); B. Gutenbergand C. F. Richter,
On SeismicWaves. III, GetlandsBeitr. z. Geophysik,47:73-131 (1936); W. A. Lynch and V.
Dillon, Characteristicsof Alaskan Earthquake Recordsat Dista n cesof40to70,,Bull
" 0 0, . Sesi m.
Soc.Am., 37:181-195 (1947);J. T. Wilsonand O. Baykal, "CrustalStructureof the North Atlantic
BasinasDeterminedfrom RayleighWave Dispersion,"ibid., 38:41-53 (1948).

GEOPHYSICS RESEARCH DIRECTORATE

AIR FORCE CAMBRIDGE RESEARCH CENTER

103

Copyright American Geophysical Union