Advanced School On Direct and Inverse Problems of Seismology
Advanced School On Direct and Inverse Problems of Seismology
Advanced School On Direct and Inverse Problems of Seismology
T. Yanovskaya
St. Petersburg University
St. Petersburg
Russian Federation
0
Contents
1.Introduction………………………………………………………1
2. Equation of motion for solid elastic media………………………2
2.1.Homogeneous isotropic elastic medium. P and S waves…….3
2.2. Plane waves…………………………………………………3
2.3. Inhomogeneous plane waves………………………………..5
2.4. Energy flux…………………………………………………..8
2.5. Spherical waves…………………………………………….10
2.6. Cylindrical waves……………………………………… …13
2.7. Anisotropic medium……………………………………..….14
0
1
1. Introduction
Wave motion is one of the well-known scientific concepts. Behavior of the waves
on the water surface, as well as propagation of acoustic or light waves are known from
everyday experience. However, it is not easy to define the wave. In general we can say
that it is a form of propagation of a disturbance of some physical field. We know
seismic, electromagnetic, acoustic, gravitational waves. Though, there is no exact
general definition of the waves, because of a variety of their characteristic features in
different cases. For example, we may generally define the wave as a disturbance
(signal), which propagates in a space with a certain velocity, but a form of the signal,
as well as its velocity may vary. However, this definition involves propagation of heat
(disturbance of temperature), but it is well known that the heat is propagated in
another way – not by a wave. Therefore it is preferable to proceed from an intuitive
notion on a wave as on a signal propagating from one to another part of a medium
with a certain finite velocity. This signal may be distorted, may change its intensity
and velocity, but should remain distinguishable. A perturbation arising in a part of the
medium causes returning forces preventing this perturbation, and the forces are of
such kind that they lead to appearance of similar (in general not exactly the same)
perturbation in neighboring points.
Seismic waves arise in solid media due to elastic forces. A main peculiarity of
seismic waves is that there are at least two types of waves (in anisotropic media -
three types), with different velocities and different polarization. This fact is due to
existence of at least two different elastic modules: in isotropic media - compressible
and shear modules. Therefore returning forces are different for different types of
deformation.
1
2
u i +1 − u i u − u i −1
mui = K −K i
Δx Δx
In continuous case, when Δx→0, and m=ρΔx, u=u(x,t), we obtain
∂ 2u
ρu = K 2
(1)
∂x
This is the simplest one-dimensional wave equation. Its solution is as follows:
x x K
u = f (t − ) + g (t + ), c= ,
c c ρ
where f(ξ) and g(ξ) are arbitrary functions, and c is regarded as velocity of the
wave propagation.
If the masses deviate from the equilibrium in perpendicular direction (shear), it
would be the same, but the module K is different (it is less than compression module),
and the velocity of shear wave propagation is less than for the compressional wave.
In continuum (2D or 3D) there are both types of deformation (compression and
shear), therefore two types of waves may propagate.
Applying Gauss formula to the surface integral, and taking into account that the
stress tensor is symmetric, we finally obtain that
∂ 2u
ρ 2 = ∇7 + F
∂t (2)
or
∂ 2 u ∂τ x ∂τ y ∂τ z
ρ = + + +F
∂t 2 ∂x ∂y ∂z (3)
§τx ·
¨ ¸
where 7 is stress tensor, or a matrix formed by vector-rows τx, τy, τz: 7 = ¨ τ y ¸
¨ ¸
©τz ¹
2
3
Eqs. (2,3) are valid for all types of media: isotropic, anisotropic, inhomogeneous,
anelastic, differing by the relationship between stress and strain. Below we shall
consider some particular cases.
In homogeneous isotropic medium the relationship between stress and strain, which
in turn is related with spatial derivatives of displacement, is following:
§ ∂u ∂u j ·
τ ij = λdivuδ ij + μ ¨¨ i + ¸
¸
(4)
∂
© jx ∂ x i ¹
ρ ∂ 2ψ 1 ∂ 2ψ
Δψ = = (6b)
μ ∂t 2 b 2 ∂t 2
These are the wave equations (scalar and vector): they describe propagation of the
waves with two different velocities a and b. The scalar potential determines the
longitudinal (compressional), or P-wave, vector potential determines the share, or S-
wave.
To solve these equations we should know the initial conditions, i.e. the functions
ϕ(x) and ψ(x) at t=0.
It is clear that the solutions of (6) are additive, i.e. if ϕ 1 and ϕ 2 are two different
solutions of (6a), then ϕ 1 + ϕ 2 will also be a solution of this equation. It means that
that by superposition of different (elementary) solutions we can construct the
solution, which would fit the given initial conditions. The simplest elementary
solutions are the plane waves.
3
4
1
where f(ξ) and g(ξ) are arbitrary functions, and k =
2
. So the two terms in the
c2
right-hand side of (8) describe the waves propagating in opposite directions with the
velocity c. It is clear that at any moment the solution at the planes (k,x)=const is one
and the same.
For simplicity we shall represent the solution of (6a),(6b) by one term in (8):
1
ϕ (x, t ) = f (t − (k P , x)) kP =
a
1
ψ(x, t ) = nF (t − (k S , x)) kS =
b
Then
u = uP + uS
where
u P = ∇ϕ = −k P f ′(t − (k P , x))
(9)
u S = rotψ = (n × k S ) F ′(t − (k S , x))
The motion in P-wave is directed along the direction of propagation k, whereas in S-
wave it is orthogonal to k.
It is clear that in both cases (P and S waves) the displacement may be represented in
the form
§ (n, x) ·
u = lΦ¨ t − ¸ (10)
© c ¹
where n and l are unit vectors, and c is a velocity . Using the concept of plane waves
we can show that c may be equal a or b, and in case c=a polarization vector l=n, and
in case of c=b l is orthogonal to n.
Substituting (10) to (5) we obtain
[(λ + μ )n(l, n) + μl ]Φ ′′(t − (n, x) / c ) = ρc 2 lΦ ′′(t − (n, x) / c )
or
(λ + μ )n (l, n ) = ( ρc 2 − μ )l
ρc 2 − μ
Define θ = , then
λ+μ
n (l , n ) = θ l (11)
or Nl = θ l , where the matrix N = nn , so that θ and
T
l are eigenvalue and
eigenvector of the matrix N correspondingly.
It is easy to show that θ fits the equation
θ 3 −θ 2 = 0 ,
that has three roots
θ 1 = 1, θ 2 = θ 3 = 0
Otherwise
λ + 2μ μ
c1 = = a , c2 = c 3 = =b
ρ ρ
The eigenvector corresponding to the first root is l=n, and those corresponding to the
other two roots are mutually orthogonal unit vectors, both orthogonal to n, i.e to the
direction of propagation. Thus the first root corresponds to the longitudinal wave, and
4
5
to other ones – to two share waves propagated with one and the same velocity. As
shown below, in anisotropic medium these two roots are different, so that there are
two quasi-share waves propagating with different velocities.
(n1 , n 1 ) − (n 2 , n 2 ) + 2i (n1 , n 2 ) = 1
(n1 , n 1 ) − (n 2 , n 2 ) = 1
(n 1 , n 2 ) = 0
(l 1 , l 1 ) − (l 2 , l 2 ) = 1
(l 1 , l 2 ) = 0
As usual, since the displacement u is real, we take only the real part of the complex
solution:
u (x, t ) = l 1 f (t − ( x, n 1 ) / c, − ( x, n 2 ) / c) − l 2 g (t − ( x, n 1 ) / c, − ( x, n 2 ) / c) (13)
This formula describes inhomogeneous plane wave. The motion in the
inhomogeneous wave has the following meaning. Temporal behavior of
displacement is one and the same along straight lines determining by intersection of
the planes (x,n1)=const and (x,n2)=const. Direction of the wave propagation
coincides with the vector n1, the wave propagates along this direction with the
c 2
velocity V = . Since n1 = 1 + n 2 > 1 , velocity of the inhomogeneous wave is
n1
always less than c (a or b). The wave form and the wave amplitude are changing in
direction of the vector n2. Components of the displacement along vectors l1 and l2 are
changing differently, accordingly to the different functions f and g.
The vectors l1 and l2 in compressional wave coincide with the vectors n1 and n2. In
shear wave the vectors l1 and l2 satisfy the relations
(l 1 , n 1 ) − ( l 2 , n 2 ) = 0
(14)
(l 1 , n 2 ) + ( l 2 , n 1 ) = 0
Orientation of the vectors n1 , n2 , l1 , l2 are shown below.
5
6
l1
n
β 1
l2
n
2
6
7
∞
f (ξ ,η ) = ³ A(ω )e −ωη cos ωξdω
0
∞
g (ξ ,η ) = ³ A(ω )e −ωη sin ωξdω
0
The functions f and g are not finite in the infinite space due to the exponential term
e −ωη . Therefore they may be used to represent solutions of the wave motion either
in a finite volume, or in case of sources.
Any wave field may be represented by superposition of plane waves (both
homogeneous and inhomogeneous), which fits the equation of motion (5) and the
following boundary conditions:
• radiation condition, requiring the displacement not to increase at the infinity;
• boundary conditions at interfaces in the medium;
• conditions in the points where sources are located.
The 1st and the 3d (and sometimes the 2nd) conditions cannot be satisfied by only
homogeneous plane waves. In such cases the inhomogeneous waves should be
involved.
7
8
For plane wave u (x, t ) = lΦ (t − (x, n ) / c) the density of kinetic energy is expressed as
ρ
Wk = [Φ ′]2 - for homogeneous wave
2
Wk =
ρ
2
l1 f ′ − l 2 g ′ =
2 ρ
2
(l 1
2
( f ′ )2 + l 2 2 ( g ′ )2 ) - for inhomogeneous wave
∂f ∂g
(here and below f' and g' mean and ).
∂t ∂t
The density of potential energy is
1 λ + 2μ
Wp = ® [(l, n)Φ ′]2 + μ2 [ l × n Φ ′]2 ½¾ = ρ [Φ ′]2 = Wk -for homogeneous wave.
2¯ c 2
c ¿ 2
For inhomogeneous wave we must replace divu and ε ij in (15) by ℜedivu and
ℜeε ij . These expressions are different for P, SV and SH waves:
ρ ° ½
2 2
4μ n 1 n 2
W p = ®( f ′) +
2 °̄
2
ρa 2
[( f ′) 2 2 °
+ ( g ′) ¾ ] - for P -wave,
°¿
Wp =
2
ρ
{( f ′) 2
+ 4 n1 n 2
2 2
[( f ′) 2
+ ( g ′)
2
]} for SV-wave,
= {( f ′) [( f ′) ]}
ρ
+ ( g ′)
2
+ n2
2 2 2
Wp -for SH-wave.
2
Taking into account that for P wave n1 = l 1 , and n 2 = l 2 , as well as l2=0 for
SH-wave, we may write the expression for total energy density in common form:
8
9
{2( f ′) ]}+
2 2
2μ n 1 l 2
W =
ρ
2
2
+ n2
2
[( f ′) 2
+ ( g ′)
2
c 2
[( f ′) 2
+ ( g ′)
2
] (16)
ρ
The first term ( ( f ′) 2 ) is similar to that for homogeneous wave, the second one
2
describes the energy of elastic deformation due to amplitude variation in the direction
perpendicular to the wave propagation, and the third one includes the part of energy
due to non-linearity of polarization.
It should be noted that the total energy density is not constant as in case of
oscillation. However, it can be explained easily: the energy is transported within the
medium, and the energy conservation law is justified for the whole volume.
© c ¹¨© 2 ρc 2 ¸
¹
for inhomogeneous P, SV and SH waves.
So for homogeneous wave the energy oscillates within a half of period from 0 to
ρω 2 , whereas for inhomogeneous wave it never achieves 0. This can be easily
explained: if the amplitude does not change along the wave front, then at the moments
corresponding to maximum displacement both velocity and strain vanish, whereas this
is not so in case of elliptic polarization and variation of the amplitude in the direction
perpendicular to the wave propagation.
Here we shall derive the expression for the energy flux in general case.
For simplicity we assume that no external forces act in the medium, then the
equation of motion is
∇T = ρu (17)
Multiply (17) by u :
(∇T, u ) = ρuu = 1 ∂ ( ρu ) = k
2 ∂W
2 ∂t ∂t
(∇T, u ) = ∇(Tu ) − (T∇, u )
The last term in the right-hand side is
∂W
(T∇, u ) = τ ik εik = cikjl ε jl εik = ∂ §¨ 1 cikjl ε jl ε ik ·¸ = ∂ §¨ 1 τ ik ε ik ·¸ = p
∂t © 2 ¹ ∂t © 2 ¹ ∂t
∂W
Thus = ∇(Tu ) = −∇p p = −Tu - Poynting vector
∂t
Therefore, if E = ³³³WdΩ , then
Ω
∂E
= − ³³³ divpdΩ = − ³³ p n dS
∂t Ω S
9
10
Thus, variation of the energy within a volume Ω is equal (with opposite sign) to a flux
of the Poynting vector across its surface.
∂E
On the other hand, = − ³³ Wc n dS and p=Wc , where c is group velocity, i.e.
∂t S
velocity with which the energy transports: in case of homogeneous wave in isotropic
c n1
medium it is equal to cn, and in case of inhomogeneous wave it is .
n1 n1
The relationship Wc = −Tu is useful in case of anisotropic media (as will be shown
later), because it provides the expression for group velocity, which differs from phase
velocity.
It should be noted that a solution in the form of ‘pure’ spherical wave exists only
for compressional waves. For shear wave we cannot have spherical symmetry
because this is impossible for vectors tangential to a spherical surface.
10
11
Now we shall represent spherical wave as a superposition of plane waves. For this
purpose it is necessary at first to represent the function F(t) in a form of the Fourier
integral
∞
− F (t ) = ³ Fˆ (ω ) exp(iωt )dω
−∞
.
ω2
Δϕ + ϕ = −4πδ (x) (22)
a2
Let us represent the solution in a form of 3D spatial Fourier transform:
1
8π 3 ³³³
ϕ ( x) = Φ (k ) exp[i (k , x)]dk , (23)
a
But the because of (24) the variables k x , k y , k z are not independent. Therefore we can
integrate over one of the components of the wave vector, e.g. over kz. The integration
is performed by the use of the theory of residuals, and finally we obtain
exp(iωR / a ) 1 exp[i (k x x + k y y ) − γ z ]
R
=
2π ³³ γ
dk x dk y (25)
where
1/ 2
§ ω2 ·
γ = ¨¨ k x2 + k y2 − 2 ¸¸
© a ¹
and the sign at γ is chosen so that Re γ>0.
(25) is the Weyl integral, which represents spherical wave as superposition of plane
waves. Since − ∞ < k x < ∞, − ∞ < k x < ∞ , the integrand contains not only
homogeneous, but also inhomogeneous plane waves.
11
12
The integral (25) may be transformed to the Sommerfeld integral, which represents
spherical wave as a superposition of cylindrical waves. Let us replace the variables:
x = r cosη k x = k cos ϕ
y = r sin η k y = k sin ϕ
Then
exp(iωR / a ) 1
∞ 2π
exp[ikr cos(ϕ − η ) − γ z ]
2π ³0 ³0
= kdkdϕ
R γ
2π
1
Taking into account that
2π ³ exp[ikr cos(ϕ − η )]dϕ = J
0
0 (kr )
we obtain
exp(iωR / a )
∞
J 0 (kr ) exp(−γ z )
=³ kdk
R 0
γ
In case of exp(-iωR/a) it is necessary to take γ * (complex conjugate) instead of γ in
the right-hand side.
This relationship can be extended to a case of non-analitical signals with sharp onset.
If F(ω) is Fourier transform of such a signal f(t), then
∞
F (ω ) exp[iω (t − R / a )] ∞
f (t − R / a ) −³∞ J 0 (kr ) exp(−γ z )
∞
= = ³ F (ω ) exp(iωt ) ³ kdk
R R −∞ 0
γ
How to understand that a wave with discontinuity on the front can be
represented as a superposition of the waves including inhomogeneous waves,
which arise simultaneously along the whole vertical axis? It is easier to analyze
spherical wave as a superposition of cylindrical waves (Sommerfeld integral)
rather than of plane waves.
Let us take r=0. Then the integral represents the wave field at
the z-axis. The integral becomes
∞
exp(−γ z ) ω/a ∞
§ homogeneous · § inhomogeneous ·
³
0
γ
kdk = ³
0
+
ω
³
/a
= ¨¨ ¸¸ + ¨¨ ¸¸
© waves (Ȗ = iȟ ¹ © waves (Ȗ is real) ¹
Now we shall show that contribution of the inhomogeneous waves is compensated by a part
of contribution of homogeneous waves
12
13
If the wave field is symmetric in respect to a straight line (z-axis), and the field does
not depend on z-coordinate, the wave equation for potentials may be written in
cylindrical coordinates as follows:
1 ∂ § ∂ϕ · 1 ∂ 2ϕ
¨r ¸= , (c = a ) for scalar potential,
r ∂r © ∂r ¹ c 2 ∂t 2
and similar equations for components of the vector potential with c=b.
This case may be regarded as 2D case, i.e. corresponding to the wave propagation in
a plane z=const. In contradiction to the 1D case, to which a plane wave can be
reduced, and to the 3D case (spherical wave), in 2D case it is impossible to construct a
solution in a general form f(r,t), and it is necessary to express a solution as a function
of t in the form of Fourier integral, and consequently to solve the equation for the
harmonic wave:
1 ∂ § ∂ϕˆ · ω 2
¨r ¸+ ϕˆ = 0 (26a)
r ∂r © ∂r ¹ c 2
or
∂ 2ϕˆ 1 ∂ϕˆ
+ + k 2ϕˆ = 0 (26b)
∂r 2
r ∂r
where k = ω / c .
Solution of the eq.(26b) is a combination of the Bessel functions. If the time
function is taken in the form exp(iωt), and the wave is expanded from r=0, then
ϕˆ (r , ω ) = AH 0( 2) (kr ) ,
and the solution for ϕ ( r , t ) is following:
∞
1
2π −³∞
ϕ (r , t ) = A(ω ) H 0( 2) (ωr / c) exp(iωt ) dω
It is clear that, in contradiction to the 1D and 3D cases, the waveform is not remained
unchanged in the process of propagation. This peculiarity was noticed by Hadamard in
his classic studies of the wave equation: he pointed that the behavior of the solution is
different for odd and even numbers of spatial dimension.
§ ωr ·
At large distances ¨ >> 1¸ we can use the asympotic representation for the Hankel
© c ¹
function, so that the solution can be written as a wave with preserved form and with
1
the amplitude decaying as :
r
exp[iω (t − r / c)]
ϕ (r , t ) = C
r
Though it seems that cylindrical waves cannot be excited in reality, they are important
in analysis of surface waves and the waves with axial symmetry.
13
14
For anisotropic elastic medium the relationship between stresses and derivatives of
displacement is expressed by the Hooke’s law in the form:
τ ij = cijkl ε kl (27)
(summation over repeated subscripts is assumed here and below). In the similar
notation we may re-write the equation of motion (2):
∂τ ij ∂ 2u
= ρ 2i (28)
∂x j ∂t
Substitute (27) into (28):
1 ∂ § ∂u k ∂u l · ∂ 2u
cijkl ¨¨ + ¸¸ = ρ 2 i (29)
2 ∂x j © ∂xl ∂x k ¹ ∂t
A solution of this equation also may be represented in a form of plane waves:
§ nq xq ·
u i = l i Φ¨¨ t − ¸¸ (30)
© c ¹
where nq are components of the unit vector indicating the direction of propagation, and
li are components of the polarization (unit) vector. Substitute this to (29):
1
cijkl (l k n j nl + ll n k n j ) = ρc 2 l i (31)
2
This is a system of 3 linear equation respectively the components of the polarizarion
vector l1 , l2 , l3 . Determinant of the system should be equated to 0, so we obtain a
cubic equation for c2. All three roots of this equations are different (unlike the
isotropic case) and depend on n, i.e. the velocity is different in different directions.
The components of the vector l are not related to n as was in isotropic case.
In this case we have no pure ‘longitudinal’ and ‘shear ‘ waves: in so-called ‘quasi-
longitudinal’ wave the polarization vector l does not coincide with n, and in two
‘quasi-shear’ waves they are not orthogonal to n.
14
15
• n z = 0, n x2 + n y2 = 1
ξ 3 − ( A + N + L)ξ 2 + ( AN + LA + LN )ξ − LAN = 0 (33)
ξ 1 = A, ξ 2 = L, ξ3 = N
l 1 = n, l2 = ez , l 3 = (n × e z )
The most unusual property of the waves in anisotropic medium is that the energy is
transferred not in the direction of propagation. It is difficult to understand for plane
15
16
waves, but can be illustrated on the example of the waves propagating from a point
source. Surfaces of constant phase at two different moments are shown below:
k
q
Vector k=n/c is wave vector, and c is phase velocity. But the energy is transported
along the ray q=n’/u, where u is group velocity. The group velocity and the direction
of the energy transfer can be obtained using the vector of energy flux, which in
general case is expressed as follows.
16
17
τ nt(1) = τ nt( f 1)
u n(1) = u n( f 1)
u t(1) = u t( f 1)
At the interface 2 the displacements and tractions in the layer are
17
18
∂τ nn
( f 1)
τ nn
( f 2)
= τ nn
( f 1)
+ h → τ nn
( f 1)
∂n
∂τ nt( f 1)
τ nt( f 2 ) = τ nt +
( f 1)
h → τ nt( f 1)
∂n
∂u ( f 2)
u n( f 2 ) = u n( f 1) + n h → u n( f 1)
∂n
∂u t( f 2) h
u t( f 2 ) = ut +
( f 1)
h → u t( f 1) + lim τ nt( f 1)
∂n h →0 μ
μ →0
τ nt(1) = τ nt( 2)
(37)
u n(1) = u n( 2)
u t(1) + mτ nt(1) = u t( 2)
where m = lim () h
μ
.
h →0
μ →0
If m=0 we obtain the welded contact, and if m→∞ the contact is sliding.
Alternative conditions for unwelded contact may be derived if we assume the layer
between two solids as filled by a viscous fluid.
It is well known that if a plane wave impinges on a plane boundary, new derivative
waves result. The number of them depends on a type of the boundary and on
polarization of the wave. In all cases we shall assume the boundary to be horizontal
(z=0), and the waveform in the incident plane wave to be
§ x sin α − z cos α ·
F¨t − ¸
© c ¹
so that the plane of incidence is y=0.
1. Free surface.
Scheme of the incident and reflected waves is shown below.
x x x
αp α p αs α s
αs αp > αs α
> s
>
>
>
>
P
>
P P
SV SV SV SH SH
z z z
18
19
It is clear that to satisfy the boundary conditions we must assume the wave forms of
the reflected waves the same as for the incident wave, i.e. F(t). Also the argument of
the function F should be the same at any point x of the boundary z=0. This
requirement leads to the Snell’s law:
sin α P sin α s 1
= =
a b c
where the meaning of c is an apparent velocity along the boundary.
If P or SV waves impinges to the surface, the displacements in the reflected P and
SV waves are expressed as
x sin α P + z cos α P
u P = κ P F (t − )(e x sin α P + e z cos α P )
a
x sin α S + z cos α S
u S = κ S F (t − )(e x cos α S − e z sin α S )
b
where κP and κS are the reflection coefficients. They are determined from a linear
system derived from the two boundary conditions
τ xz = 0, τ zz = 0, at z = 0
The system may be written in the matrix notation:
§κ P ·
A¨¨ ¸¸ = b
©κ S ¹
where the matrix A is following:
§ sin 2α P γcos2α S · a
A = ¨¨ ¸¸, γ =
© − γ cos 2α S sin2α S ¹ b
19
20
2.5
2.0
1.5
1.0
SS
0.5
SP
0.0
0 20 40 60 80
x z x z ½
u P (t , x, z ) = Re®(κ 1 + iκ 2 )( f (t − , γ 2 sin 2 α S − 1) + ig (t − , γ 2 sin 2 α S − 1))(e x γ sin α S + ie z γ 2 sin 2 α S − 1¾
¯ c c c c ¿
For harmonic wave
ωz
x x ½ − γ 2 sin 2 α S −1 )
u P (t , x, z ) = Re ®(κ 1 + iκ 2 )(cos[ω (t − ) + i sin[ω (t − )])(e x γ sin α S + ie z γ 2 sin 2 α S − 1)¾e c
¯ c c ¿
Particle motion in P wave for different angles of incidence is shown in the next figure.
The motion is elliptic, prograde (as a rolling ball), and z-axis of the ellipse increases
with the angle of incidence.
38o
49o
o
64
Particle motion in
inhomogeneous
reflected SP-wave
75o
80o
If P wave is incident, reflection coefficients are always real. The behavior of the
coefficients with the angle of incidence is shown below.
20
21
1.5
1.0 PS
0.5
0.0 α
0 20 40 60 80 P
-0.5
-1.0 PP
-1.5
S
S (1)
P
P
β reflected/
incident α transmitted
waves waves
P
P
S
S (2)
21
22
and the vector bi in the right-hand side depends on the incident wave: it is formed by
the i-th column of the matrix A according to the rule b (ji ) = a ji (−1) j (i=1(2) if P(S)
wave is incident from the medium (1) , i=3(4) if P(S) wave is incident from the
medium 2); the angles α and β correspond to P and S waves.
VP=7.8, VS=4.5
1 VP=8.2, VS=4.8 1
P P
S wave reflection coefficient
P-wave reflection coefficient
0.8 0.8 S
P
0.6 0.6
10
8
0.4
10
0.4 5
5
2
0.2 2
8
0.2
0 0
0 0
0 20 40 60 80 0 20 40 60 80
angle of incidence, deg. angle of incidence, deg.
P
1.2 P
1 S
S-wave transmission coefficient
1 0
P-wave transmission coefficient
2 0.8
P
0.8
5 0.6
0.6
10 10
8
0.4 5
20
0.4 2
8
0.2
0.2 0
0
0 20 40 60 80
0 0 20 40 60 80
angle of inciidence, deg.
angle of incidence, deg.
22
23
a b
P1 P1
P1 MP P
1 S1 1 P2 P 1
1 S1 N 1
2
O
S
2 S2 2 P2
S2 P1
Q
P2 P2
(fig.b).
In this case a part of the boundary between points A and O turns to be disturbed. This
disturbance is radiated to the half-space 1 in a form of so-called head waves with
conical fronts. The waveform of the head wave (ψ(t)) is the integral of the waveform
of the incident wave (f(t)):
t t
ψ (t ) = ³ f (τ )dτ ψ (t ) = ³ f (τ )dτ
0 0
O R
(0)
r
< > r ( 0) + R l
l ψ (t − − ) tan α Pcrit
(1)
< > a1 a2
r U head = Γ
l 3/ 2 r
23
24
In a half space with free surface exists a specific solution, which represents a
superposition of inhomogeneous plane P and S waves. If we look for a solution in a
form of a plane wave, the plane of incidence being y=0 (for simplicity we assume the
dependence on time to be harmonic) then, according to the general representation of
the inhomogeneous waves, we may write
§a a2 · § 1 1 ·
u P (t , x, z ) = A¨ e x − ie z − 1 ¸ exp[iω (t − x / c)] exp¨¨ − ωz 2 − 2 ¸¸
¨c c 2 ¸ c a ¹
© ¹ ©
(38)
§ b 2
b · § 1 1 ·
u S (t , x, z ) = B¨ ie x 2 − 1 + e z ¸ exp[iω (t − x / c)] exp¨¨ − ωz 2 − 2 ¸¸
¨ c c ¸¹ c b ¹
© ©
Replacing these expressions to the boundary conditions at the free surface z=0 we
obtain a linear system for the amplitudes A and B of P and S waves. This system is
homogeneous, therefore if non-zero solution exists, the determinant of the system
should be equal to zero. This is
2
§ c2 · c2 c2
¨¨ 2 − 2 ¸¸ − 4 1 − 2 1 − 2 = 0
© b ¹ a b
Thus, the superposition of the waves (38) with A and B satisfying the boundary
conditions (up to a constant multiplier) represent a wave propagating along x-axis and
decaying exponentially along vertical direction.
Velocity of this wave along x-axis varies from 0.874b up to 0.956b for all possible
1
values of b/a (from to 0). It does not depend on frequency.
2
Motion in the Rayleigh wave is elliptic, retrograde at the surface and at shallow
depths, but becomes to be prograde at large depths. The figure shows variation of the
vertical (w) and horizontal (u) components with depth.
0 0.5 1 1.5
z/λ
u
24
25
Ratio of horizontal and vertical amplitudes in Rayleigh waves at the surface is equal
4
1 − (c / b ) 2 1
to . It depends also only on the ratio b/a. For b/a varying from to 0
4
1 − (c / a ) 2 2
1
this ratio varies from 0.786 to 0.541. For b/a= it is equal to 0.681.
3
Rayleigh wave can be generated by a point source in the half-space, because the
spherical wave radiated by such source may be represented as a superposition of both
homogeneous and inhomogeneous plane waves. So it contains the waves with the
apparent velocity equal to the velocity of Rayleigh wave.
25
26
1 1
μ2 2 − 2
§ 1 1· c b2
tan¨ ωH 2 − 2 ¸ = (39)
© b1 c ¹ 1 1
μ1 2 − 2
b1 c
This is dispersion equation for Love wave velocity: unlike to Rayleigh waves in a
half-space the velocity depends on frequency.
It is easy to show analytically that b1 ≤ c ≤ b2 (as concluded above from simple
physical consideration). In fact, only in this case left and right sides of the dispersion
equation are real.
It is also easy to show that for a given c there are infinite numbers of frequencies
satisfying the dispersion equation. In fact,
ª 1 1 º
« μ2 2 − 2 »
1 c b2 »
ωH = a tan « + kπ , k = 1,2,3....
1 1 « 1 1 »
− « μ1 2 − 2 »
b12 c 2 «¬ b1 c »¼
Also we can show that for any given ω there are finite number of c. The Eq. (39) may
be written as
f 2 (c , ω ) = f 1 (c )
A graph for the right-hand side is drawn number of roots increases with increase of
by solid line. And the graph of the left- ω.
hand side behaves as tan - at the values of
π
argument + kπ it tends to ±∞. But the
2
rate of change depends on ω. At the figure
below the graphs for the left-hand side are
shown for two different values of ω. It is
clear that for small ω there is only one f1(c)
ω
root, and the 1
f2(c)
ω >ω
Dispersion curves of phase ( c ) and group (u) velocities are
2 shown
1 below.
b c b2
1
26
27
b2
b1
frequency
Real solids are not perfectly elastic. This causes seismic processes (waves, oscillations) to
attenuate with time due to various energy-loss mechanisms. The most usual explanation of these
mechanisms is internal friction between microscopic particles of the material, which leads to
transformation of mechanical energy to heat.
The simplest description of attenuation due to ‘friction’ can be developed for an oscillating mass
on a spring: this is a phenomenological model for seismic attenuation.
m f
Let x be a deviation of the mass from the equilibrium. The force f is friction opposing the
motion of the mass. Denote K a measure of the spring’s stiffness.
The motion of the mass is
mx − F = 0
If friction is absent, and oscillation results only from elastic force, F=-kx,
mx + kx = 0 ,
and we obtain harmonic oscillation:
x = A sin(ωt + ϕ ), ω = K / m
27
28
However, if there is a friction between the moving mass and the underlying surface, and this force
is proportional to the velocity of mass, so that the total force is
F = − kx − γx ,
then the oscillation attenuate:
x = e − βt e i ω t
γ K β 2m
where β = , ω= 1−
2m m K
Motion in a solid fits the equation
∂ 2u
∇T = ρ 2
∂t
where T is stress tensor. To solve this equation for any particular case it is necessary to express
the stress in terms of displacement and its derivatives. In perfectly elastic medium this relation is
expressed by the Hooke’s law. As shown above, in the case of homogeneous isotropic medium
the equation of motion is reduced to the following
∂ 2u
( λ + 2 μ ) ∇divu − μrotrotu = ρ ,
∂t 2
28
29
The relationship between stress and strain may be written in another form:
§ dε ·
τ = μ ¨ε + Tε ¸
© dt ¹
τ ik = με ik − ³ ϕ (ξ )ε ik (t − ξ ) dξ (40)
0
ϕ(ξ) is the so-called creeping function. Various rheological models correspond to various
creeping function.
μ
If ϕ (ξ ) = exp( −ξ / Tτ ) (for pressure the Hooke's law is kept), we obtain the Maxwell model.
Tτ
Substituting this function to the formula (40) and integrating by parts, we obtain
dτ τ dε
+ =μ
dt Tτ dt
The constant Tτ is the relaxation time of stress under a constant strain:
τ = τ 0 exp(−t / Tτ )
The Maxwell model is valid only for shear strain. The figure below shows the mechanical
analogue of the Maxwell model, as well as behavior of strain under a constant stress,
29
30
Standard linear solid. This model combines the both dissipation mechanisms, so that the
relationship between stress and strain is following:
dτ dε
τ + Tτ = μ (ε + Tε )
dt dt
In this model the strain is relaxed under a constant stress, and the stress is relaxed under a
constant strain.
30
31
(1 + iωTτ )τ (r , t ) = μiωTτ ε (r , t )
iωTτ με (r , t )
τ (r , t ) =
(1 + iωTτ )
For standard linear solid:
(1 + iωTε )
τ (r , t ) = μ ε (r , t )
(1 + iωTτ )
This for all cases the relationship between stress and strain is formally coincides with the Hooke's
law, but the elastic modules are complex and depend on frequency. The frequency dependence is
different for different models. Therefore in analysis of wave propagation of harmonic waves in
anelastic media we may formally use the inferences obtained for perfectly elastic medium.
Consider propagation of a plane harmonic wave along x-axis:
A( x , t ) = A0 exp[iω (t − x / V )]
If the modules are complex, the wave velocity V should be also complex:
1 1 i
= − *
V V (ω ) V
Then
§ ωx ·
A( x , t ) = A0 exp¨ − * ¸ exp[iω ( t − x / V (ω ))] ( 41 )
© V ¹
This shows that the wave attenuates with distance, and its velocity depends on frequency:
V = V (ω ) . Attenuation and dispersion are the main properties of the waves propagating in
anelastic media.
Using the wave number k we can represent the plane wave in the form
A( x , t ) = A0 exp[i (ωt − kx )]
where the wave number k is complex: k = k − ik , so that the attenuation is determined by the
*
2π k *VT π
k= , then Q −1 = , and consequently, k * = . Thus, the term describing the
VT π QVT
π
attenuation is exp( − ) . In inhomogeneous medium, where both velocity and Q are functions
QVT
§ π ds ·
of coordinates, it is exp¨ − ³ ¸ .
© T QV ¹
31
32
Now we show how the quality factor Q is expressed in terms of the real and imaginary parts of
the complex modules, and how to relate it with the relaxation times. Consider a shear wave. The
complex velocity is expressed through the complex shear module as follows:
−1 ρ ρ§ μ* ·
V = ≈ ¨1 − ¸
μ + iμ * μ © 2μ ¹
2k *
−1 −1 2(V * ) −1 μ *
It follows from Q = that Q = = . Knowing the expressions for complex
k V −1 μ
modules for different rheological models we can write the quality factor as a function of
frequency and relaxation times:
It seems that Q should noticeably change with frequency. However seismological observations
indicate that Q does not practically depend on frequency over a large range of frequencies. This is
because of a variety and scale of attenuation processes in real materials. The most general model
is the standard linear solid, for which the frequency dependence of Q −1 is as follows:
-1
Q
ω
0
1
The peak in Q −1 is known as a Debye peak. It corresponds to the frequency ω0 = , and
Tε Tτ
1 § Tε T ·
the value of Q −1 at this frequency is equal to ¨ − τ ¸ . The superposition of numerous
2 © Tτ Tε ¹
Debye peaks for various relaxation processes , each with a different frequency range, produces a
broad, flattened absorption band.
32
33
-1
Q
5.1. Caustic
Caustic is an envelope of the rays. It divides the space to illuminates zone and shadow zone. The
rays cover the illuminated zone, where two rays cross a given point: coming to and off the caustic
(see the figure below)
illuminated zone
caustic
shadow zone
33
34
The wave field cannot be discontinuous; therefore we may expect a perturbation in the shadow
zone in the vicinity of the caustic. In fact, asymptotic solution of the wave equation (or equation
of motion of elastic medium) in a narrow vicinity of the caustic describes a wave propagating
along the caustic, whose amplitude varies in the direction perpendicular to the caustic as the Airy
function. The amplitude of such a wave depends on frequency, velocity and effective radius of
curvature of the caustic (effective curvature is a sum of curvatures of the caustic and the ray
tangent to the caustic). In 2D case the wave field in the vicinity of the caustic is determined by the
formula
§ § 2 ·1 / 3 2 / 3 ·
U ( n, s; t ) = CAi¨ − ¨ 2 ¸ ω n ¸ exp(ω (t − τ ( s ) ) (43)
¨ © Rc ¹ ¸
© ¹
where s is a distance along the caustic, n is a distance perpendicular to the caustic (n<0 in the
shadow zone) , R is effective radius of curvature of the caustic, τ(s) is travel time along he
caustic.
Airy function oscillates at negative
argument and decays approximately Ai(x)
exponentially at positive argument.
x
-10 -5 0 5
Now we shall analyze the waves diffracted at sharp ends of boundaries. An example is
the case when a plane wave impinges to an opaque boundary :
34
35
>
>
> diffracted
<
<
<
> < wave<
> >
> <
<
bo had
<
un ow
s
<
da
ry
of
Representation theorem
Since the diffracted waves depend on frequency we may consider only harmonic waves.
At S given are a displacement US and a traction Tn . The displacement US and the
traction Tn cause a wave field within Ω. For simplicity we suppose that no body forces
act within this volume. The wave field in Ω satisfies the equation
∇T + ρω 2 U = 0 (44)
Define the Green function g ( x, ξ) , which is a solution of the equation
(q)
∇τ q + ω 2 ρg q = −δ (x − ξ)e q (45)
where τq is the stress tensor corresponding to the Green function.
Let T and V correspond to any two different solutions of (44), T being a symmetric
tensor. Then according to the Gauss formula we obtain
35
36
Denote
( )
I 1 = ³ (∇τ q , U ) − (∇T, u q ) dS
S
( )
I 2 = ³ (τ q ∇, U ) − (T∇, u q ) dS
S
To apply the formulas (43) and (47) for determining the wave field it is necessary to
know the Green tensor g q (x, ξ, t ) . It should be noted that the Green tensor can be
determined in different ways, depending on the boundary conditions at S. In case of a
bounded volume it is convenient to assume either displacement or traction equal to zero
at S , depending on which characteristics of the field (displacement or traction) is given
at S. In case of unbounded medium it is sufficient to take into account the radiation
condition. In general case of the medium the Green function can be determined only
approximately, but in homogeneous isotropic medium the exact expression for the Green
function exists. It can be easily obtained from the Stokes’ formula for the wave field
excited by a point force located in the origin of coordinates. Remind that we consider
harmonic waves and omit the time factor exp(iωt).
1 ª 2 cos θ sin θ º cos θ
R/b
u ( x, t ) = « e + e θ » ³ τ exp(−iωτ ) dτ + exp(−iωR / a )e R −
4πρ ¬ R 4πρa 2 R
3 R 3
R ¼R / a
(48)
sin θ
exp(−iωR / b)eθ
4πρb 2 R
The Green function g q ( x , ξ , ) is obtained from (48) if we replace R = x − ξ ,
cosθ = (e q , e R ) .
36
37
The first term in the right-hand side of (48) decays with distance more rapidly than the
last two ones, so usually, if the source is far away from the point of observation, it is
sufficient to consider only the second and the third terms.
In this section we shall consider a simple example how the representation theorem is
applied to the problems of diffraction of the waves. We shall analyze the waves
diffracted at sharp ends of boundaries. The simplest example is the case when a plane
wave impinges to an opaque half-plane screen, which does not transmit P-wave, placed
along xy-plane at − ∞ < x, < − X , − ∞ < y < ∞ .
A plane wave is incident normally to this plane along positive direction of z-axis. We
shall determine the wave field in the point x=0, z=H.
e
r
-X O
r H
ez
37
38
To calculate τ nzz we take into account that for high frequencies (and for large k) it is
sufficient to differentiate in respect to r only the exponential term. Then
ik § H H3 ·
τ nzz = exp( ikr )¨ λ
¨ r + 2 μ ¸
3 ¸
4πρa 2 r © r ¹
Substituting all these expressions to (50) we obtain
1
∞ ∞
§ H2 § H H 3 · · exp(ikr )
¨ − ik (λ + 2 μ ) 2 − ik ¨¨ λ + 2 μ 3 ¸¸ ¸¸
4πρa 2 −³∞−³X ¨©
U z (M ) = dxdy
r © r r ¹¹ r
To estimate this integral we use the method of stationary phase. The stationary point is
x st = y st = 0 . In this point r=H. According to the method of stationary phase we
represent the phase function as series in the vicinity of the stationary point and keep only
terms of the second order. Then we obtain
∞ ∞
2ik (λ + 2 μ )
U z (M ) = − ³ ³ exp(ikrxx x 2 / 2)dx =
2
exp( ikH ) exp(ikr y / 2 ) dy
4πρa H2 yy
−∞ −X
∞
(51)
ik 2πiH
=− exp(ikH ) ³ exp(ikx 2 / 2 H )dx
2πH k −X
1
(It is taken into account that rxx = ryy = ).
H
∞
³ exp(ikx
2
The integral / 2 H )dx can be expressed through the Fresnel integral
−X
z
F ( z ) = ³ exp(iπt 2 / 2)dt
0
∞
πH § kX 2 ·
³ exp(ikx / 2 H )dx = ¨¨ F (∞) − F (− )¸
2
(52)
−X
k © πH ¸¹
i
Substituting (52) to (51), and taking into account that F (∞) = , F (− z ) = − F ( z ) we
2
finally obtain
ª 1 e − iπ / 4 º
U z ( M ) = exp(ikH ) « + F ( X / R F )»
¬2 2 ¼
πH
where RF = is the Fresnel radius. The modulus of this function is shown below.
k
1/2
X/RF
38
39
Thus, under the edge (X=0) the amplitude of the transmitted wave is twice as less of the
amplitude of the incident wave. When the screen is moved to the left (X>0) the
amplitude increases and exceeds that of the incident wave. If the screen is moved to the
right, the amplitude decreases gradually to zero.
It is also possible to estimate a phase of the total transmitted wave. The total field may
be represented as a superposition of ‘pure’ transmitted wave and diffracted wave. If the
transmitted field is deducted from the total field, we obtain a field of the diffracted wave.
It can be shown that a phase of this wave is approximately equal to
kR+π/4, where R is a distance from the point M to the edge of the screen. Thus the edge
of the screen may be regarded as a source of the diffracted wave.
U=Udiff U=Uinc+Udiff
39
40
in each point the rays go off the boundary, and propagate later according to the
geometrical seismic. Decay of the sliding wave amplitude at a segment from P to Q is
determined by the following factor
j ( P) § Q
ds ·
exp¨¨ − ξω 1 / 3 ³ 2 / 3 1 / 3 ¸
j (Q ) © P R c ¸¹
where j is geometrical spreading at the surface of the obstacle, R is effective radius of
curvature of the surface, c – wave velocity. Thus we may calculate the amplitude in the
point Q, and then use formulas of the ray theory.
Similar pattern of the rays, and consequently behavior of the diffraction wave field, is
observed when the rays are tangent to a wave guide (low velocity zone).
Consideration of these three examples shows that the diffraction effects can be
substantially different.
40