Professional Documents
Culture Documents
and Computing
Numerical
Simulation
in Applied
Geophysics
Lecture Notes in Geosystems Mathematics
and Computing
Series Editors
W. Freeden, Kaiserslautern
Z. Nashed, Orlando
O. Scherzer, Vienna
More information about this series at http://www.springer.com/series/15481
Juan Enrique Santos • Patricia Mercedes Gauzellino
Numerical Simulation
in Applied Geophysics
Juan Enrique Santos Patricia Mercedes Gauzellino
Universidad de Buenos Aires Departamento de Geofísica Aplicada
Facultad de Ingeniería Facultad de Ciencias Astronómicas y
Instituto del Gas y del Petróleo Geofísicas
Ciudad Autónoma de Buenos Aires, UNLP Universidad Nacional de La Plata
Argentina La Plata, Argentina
Universidad Nacional de La Plata
La Plata, Argentina
Department of Mathematics
Purdue University
West Lafayette, Indiana, USA
solid and fluid displacement vectors in 1-D, 2-D and 3-D wave propagation prob-
lems.
In the context of Numerical Rock Physics, this book presents several finite el-
ement up-scaling procedures, formulated in the space-frequency domain, to deter-
mine a long-wave equivalent viscoelastic medium to a Biot medium with multiscale
heterogeneities in the fluid and solid properties. These up-scaling procedures yield
the complex and frequency dependent stiffness coefficients defining the viscoelastic
model to be used to simulate wave propagation at the macro-scale.
The cases of patchy gas-brine saturation and a poroelastic matrix composed of
a fractal shale-limestone mixture are used to construct the corresponding equiva-
lent isotropic viscoelastic medium. The case of a Biot medium with aligned frac-
tures, modeled either as fine highly permeable and compliant layers or boundary
conditions is studied to determine an equivalent transversely isotropic viscoelastic
medium.
Wave propagation in the ultrasonic range of frequencies is illustrated for the case
of partially frozen porous media, where snapshots of the solid, ice and water phases
allow to identify all waves that can propagate in this type of medium.
The up-scaling procedures are used at the macro-scale to simulate 2-D seismic
monitoring of CO2 sequestration and 3-D wave propagation in transversely isotropic
media. The numerical simulators are based on a finite element solution of the vis-
coelastic wave equation in the space-frequency domain, with absorbing boundary
conditions at the artificial boundaries of the subsurface model, which are derived
for elastic, viscoelastic and Biot media. Due to the large number of degrees of free-
dom needed for the spatial discretization, a finite element domain decomposition
iteration is used to solve the algebraic problems at a set of frequencies of interest.
The time-domain solution is recovered by a discrete inverse Fourier transform.
The book is aimed at researchers and professionals working in the fields of Geo-
physics, Engineering, Physics and Applied Mathematics. Basic knowledge on anal-
ysis, elasticity, fluid mechanics and numerical analysis is assumed.
ix
Contents
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2 Constitutive relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.2.1 Relations to determine the two-phase elastic constants . . . 38
2.3 Inclusion of linear viscoelasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.4 The equations of motion. Low frequency range . . . . . . . . . . . . . . . . 39
2.5 The equations of motion. High frequency range . . . . . . . . . . . . . . . . 41
2.6 Plane wave analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.7 Application to a real sandstone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.7.1 Characterization of the compressional modes of
propagation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.7.2 Analysis of all waves in the purely elastic case . . . . . . . . . 46
2.7.3 Analysis of all waves as function of frequency in the
general dissipative case . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
xi
xii Contents
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.2 Constitutive relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.3 The equations of motion. Low frequency range . . . . . . . . . . . . . . . . 59
3.4 The equations of motion. High frequency range . . . . . . . . . . . . . . . . 61
3.4.1 Phase velocities and attenuation coefficients . . . . . . . . . . . 62
3.5 Numerical Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.5.1 Characterization of the four compressional modes of
propagation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.5.2 Behaviour of all waves in the purely elastic case . . . . . . . . 68
3.5.3 Behaviour of all waves as function of frequency . . . . . . . . 69
3.6 Appendix 1. Determination of the elastic coefficients. Inclusion
of linear viscoelasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.2 The strain energy of the composite system . . . . . . . . . . . . . . . . . . . . 80
4.3 Constitutive relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.4 Determination of the coefficients in the constitutive relations . . . . . 84
4.4.1 Inclusion of linear viscoelasticity . . . . . . . . . . . . . . . . . . . . 85
4.5 The equations of motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.5.1 Correction of the viscodynamic coefficients in the high
frequency range . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.6 Plane wave analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.7 Numerical Examples. Shaley sandstones . . . . . . . . . . . . . . . . . . . . . . 90
4.8 Appendix 1. Calculation of the elastic coefficients in the
stress-strain relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.8.1 The case of shaley sandstones . . . . . . . . . . . . . . . . . . . . . . . 94
4.8.2 The case of partially frozen porous media . . . . . . . . . . . . . 94
4.8.3 The Kuster and Toksöz theory . . . . . . . . . . . . . . . . . . . . . . . 95
Glossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
Chapter 1
Waves in poroelastic solid saturated by a
single-phase fluid
Abstract This chapter contains the derivation of Biot’s theory describing the prop-
agation of waves in a porous elastic solid saturated by a single-phase fluid. After
deriving the constitutive relations and the form of the potential and kinetic energy
densities and the dissipation function, the lagrangian formulation of the equations of
motion is given. Next, a plane wave analysis is performed showing the existence of
two compressional waves and one shear wave. An example showing the behaviour
of all waves as function of frequency for a sample of Nivelsteiner sandstone satu-
rated by water, oil and gas is included.
The equations governing the macroscopic behaviour of porous media can also be
obtained by means of homogenization methods, which consist on passing from the
microscopic scale description at the pore and grain scales to the mesoscopic and/or
macroscopic scale.
Contributions to the solution of this problem were given in [Sanchez Palencia, 1980]
and [Bensoussan, et al., 1978], who developed the so called two-space homogeniza-
tion technique. This method provides a systematic procedure for deriving macro-
scopic static and dynamic equations starting from the equations governing the be-
haviour of the medium at the micro-scale. It was successfully applied by different
authors to obtain a theoretical justification of Darcy’s law and Biot’s equations of
motion ([Levy, 1979, Burridge and Keller, 1981, Auriault et al., 1985]).
Let Ω be a porous medium saturated by a single–phase fluid, let φ (x) be the effective
u f ,T be the locally averaged solid and fluid displacements in
porosity, and let us,T ,
Ω . The physical meaning of u f ,T is as follows: take a unit cube Q of bulk material.
Then, for any face F of the cube, the quantity
f ,T · ν dF
φu
F
represents the amount of fluid displaced through F, where ν denotes the unit out-
ward normal to F.
Let τi j = τ i j + Δ τi j and σi j = σ i j + Δ σi j be the total stress tensor of the bulk
material and the stress tensor in the solid part, respectively, where Δ τi j and Δ σi j
represent changes in the corresponding stresses with respect to reference stresses
τ i j and σ i j in the initial equilibrium state. Also, let p f = p f + Δ p f denote the fluid
pressure, with Δ p f being the increment with respect to a reference pressure p f in
the initial equilibrium state. Also, let
σ f = −φ p f (1.1)
τi j = σi j + δi j σ f = σi j − φ p f δi j , (1.2)
gsi = σ i j ν j , gif = −φ p f δi j ν j .
1 Waves in poroelastic solid saturated by a single-phase fluid 3
Since the fluids are at rest, all fluid pressures are constant on Ω and the total stress
field is also in equilibrium. Hence, the fluid pressure and the total stress field satisfy
the conditions
∂ pf ∂ τi j
∇p f = = 0, = 0, in Ω. (1.3)
∂ xi ∂xj
Here and in what follows the Einstein convention of sum on repeated indices is used.
Let W denote the strain energy density for the fluid–solid system. Then, the
virtual work principle states that the variation of strain energy in a body Ω is equal
to the virtual work of the surface forces on ∂ Ω (body forces such as gravity are
neglected); i.e.,
δ W dΩ = (gsi δ usi + gif δ uif )d(∂ Ω ), (1.4)
Ω ∂Ω
Set
uif = φ (
uif − usi ),
which represents the displacement of the fluid relative to the solid measured in terms
of volume per unit area of bulk material, so that u̇if indicates the infiltration speed.
Then, using Gauss’s theorem (1.6) becomes
δ W dΩ = (Δ τi j δ usi ν j − Δ p f δi j δ uif ν j )d(∂ Ω ) (1.7)
Ω ∂Ω
∂ ∂
= (Δ τi j δ usi )d Ω − (Δ p f δi j δ uif )d Ω .
Ω ∂xj Ω ∂xj
4 1.2 Constitutive relations
Next, note that since the body remains in equilibrium, using the symmetry of τi j and
(1.3) we get
∂ ∂ Δ τi j s ∂ δ usi
(Δ τi j δ usi ) = δ ui + Δ τi j
∂xj ∂xj ∂xj
∂ δ ui
s 1 ∂ δ ui 1
s ∂ δ usi
= Δ τi j = Δ τi j + Δ τ ji = Δ τi j δ εi j (us ),
∂xj 2 ∂xj 2 ∂xj
where
∂ usi ∂ u j
s
1
εi j (u ) =
s
+
2 ∂ x j ∂ xi
denotes the strain tensor. Also,
∂ ∂Δ pf f ∂ δ uif
(Δ p f δi j δ uif ) = δ ui + Δ p f = Δ pf δ∇·uf .
∂xj ∂ xi ∂ xi
ξ = −∇ · u f . (1.8)
δ W = Δ τi j δ εi j (us ) + Δ p f δ ξ . (1.10)
Next, since δ W must be an exact differential of the strains εi j (us ) and ξ , W must
satisfy the conditions
∂W ∂W
= Δ τi j , = Δ pf , (1.11)
∂ εi j ∂ξ
∂ 2W ∂ 2W ∂ 2W ∂ 2W
= , = . (1.12)
∂ εi j δ ξ ∂ ξ ∂ εi j ∂ εi j ∂ εk ∂ εk ∂ εi j
The strain energy density can be expressed in terms of the invariants separating
dilatational and deviatoric effects as well as the coupling between the solid and the
fluid displacements. Also, it is convenient to use I2 = −4I2 instead of I2 :
I2 = 4(ε12
2
+ ε13
2
+ ε23
2
) − 4ε11 ε22 − 4ε22 ε33 − 4ε11 ε33
= 2(ε12
2
+ ε21
2
+ ε13
2
+ ε31
2
+ ε23
2
+ ε32
2
) − 4ε11 ε22 − 4ε22 ε33 − 4ε11 ε33 .
1
W = W (u) = (Eu (es )2 + μ I2 − 2Bes ξ + M ξ 2 ), (1.13)
2
where u = (us , u f ).
Using (1.11) we obtain
∂W
= Δ τ11 = Eu es + μ (−2ε33 − 2ε22 ) − Bξ , (1.14)
∂ ε11
∂W
= Δ τ22 = Eu es + μ (−2ε11 − 2ε33 ) − Bξ ,
∂ ε22
∂W
= Δ τ33 = Eu es + μ (−2ε22 − 2ε11 ) − Bξ ,
∂ ε33
∂W ∂W
= Δ τi j = 2μεi j , i = j, = Δ p f = −Bes + M ξ .
∂ εi j ∂ξ
Next we rewrite (1.14) introducing new elastic constants and relationships among
them. Later, the elastic constants will be determined as a function of the properties
of the solid and fluid phases. Set
Eu = λu + 2μ , λu = λ + α 2 M. (1.15)
In abbreviated form,
Sx
1
1
εi j = Δ τi j + δi j (DΔ τ − F Δ p f ), (1.18)
2μ
ξ = −F Δ τ + H Δ p f , (1.19)
Consider a cube of bulk material as in Figure 1.1. In the initial equilibrium state, V b ,
V s , V f are the bulk, solid, and fluid volumes, respectively. Since us is the averaged
solid displacement vector over the whole bulk material, es represents the change
Δ Vb = Vb −V b in bulk volume per unit volume of bulk material; i.e.,
Δ Vb
es = .
Vb
Therefore, es denotes the volumetric strain of the bulk material. Similarly, the volu-
metric strain of the pore space is defined as
V f −V f ΔVf
ep = = .
Vf Vf
The amount of fluid entering the face Sx1 is φ ( u1f (x1 ) − us1 (x1 ))Δ x2 Δ x3 , and the
f
amount of fluid leaving the face Sx1 +Δ x1 is φ (
u1 (x1 + Δ x1 ) − us (x1 + Δ x1 ))Δ x2 Δ x3 .
Then, the change in fluid content δ Fc is given by
Δ V f − Δ V fc Vf 1 ΔVf − ΔVf c
ξ= = (Δ V f − Δ V fc ) =φ . (1.21)
Vb Vb Vf Vf
For the analysis that follows, we consider a cube of bulk material immersed in a
container filled with the same fluid saturating the solid matrix. Therefore, in the
bulk material the same pressure is supported by the the rock matrix and the fluid.
Then, any tensional change Δ τi j is conveniently decomposed into the form
Δ τi j = −Δ p f δi j + Δ τi j , (1.22)
where τi j is the so-called residual or effective stress of the material. Following the
ideas in [Santos et al., 1990a] the elastic coefficients in the right-hand side of (1.16)
and (1.17) can be determined as follows. First, since the fluid does not support any
shear, μ is identical to the shear modulus of the dry matrix. To determine the re-
maining coefficients in (1.16)-(1.17) it is sufficient to consider tensional changes
Δ τi j such that
1
Δ τ11 = Δ τ22 = Δ τ33 = Δ τ = −Δ p, Δ p > 0, Δ τi j = 0, i = j.
3
Set
1
Δ τ ≡ Δ τ11 = Δ τ22 = Δ τ33 = −Δ p.
3
8 1.3 Determination of the elastic coefficients
es = −Δ p/G. (1.30)
Δp Δ p
Δ p f = 0, es = − =− , (1.32)
Km Km
so that the fluid pressure is held constant and the external applied pressure
is supported only by the solid matrix. Here Km denotes the bulk modulus of the dry
matrix.
In the jacketed compressibility test, a sample of bulk material is enclosed in an
impermeable jacket and immersed in a chamber filled with a fluid held at the same
reference pressure p f than the fluid inside the sample. Then an additional pressure
change Δ p is applied to the fluid in the chamber. To ensure that the fluid pressure in
the sample stays at the reference value p f , a tube T f is connected from the inside
of the sample to a container filled with fluid held at the reference pressure p f . Thus
there is no change in fluid pressure as in (1.32). This test is illustrated in Figure 1.2.
0110
1010 p
Tw
f
11001100
00000000000000000000000000000000000000000000
11111111111111111111111111111111111111111111
111111
000000 111111
000000
p pw
f p
pw
f
0110
10 p
Δ V f − Δ V fc ΔVf Δ (φ Vb ) Δ φ Vb + φ Δ Vb
ξ = = = =
Vb Vb Vb Vb
Δ φ (V b + Δ Vb ) + (φ + Δ φ )Δ Vb
= .
Vb
Then,
Δ Vb
ξ Δφ +φ . (1.33)
Vb
Now, according to [Zimmerman et al, 1986]
1 (1 − φ )
Δφ = − Δ p, (1.34)
Ks Km
10 1.3 Determination of the elastic coefficients
where Ks denotes the bulk modulus of the solid grains. Thus, using (1.32), (1.33)
and (1.34),
1 1 Δ p Δ Vb 1 1
ξ= − Δ p + φ +φ = − Δ p. (1.35)
Ks Km Km Vb Ks Km
Therefore,
1 1
3D + = . (1.36)
2μ 3Km
Also, from (1.19),
1 1
− Δ p = −F Δ τ = F 3 Δ p,
Ks Km
so that
1 1 1
F= − . (1.37)
3 Ks Km
Now using (1.29), (1.36), and (1.37) in (1.28) we obtain
Δp 1 1 1 1 1
es = − = Δ τ − 3F Δ p f = (−3Δ p) − 3 − Δ p f . (1.38)
Ku 3Km 3Km 3 Ks Km
Next we will derive a relation between Δ p and Δ p f valid for the closed system.
First note that since, for the closed system ξ = 0, from (1.12) and (1.20) we have
c
ΔVf ΔVf
0=φ − ,
Vf Vf
so that
ΔVf Δ V fc Δ pf
= =− . (1.40)
Vf Vf Kf
Next, using (1.34), up to first order terms, we have that
ΔVf Δ (φ Vb ) Δ Vb Vb V f Δ Vb (V b + Δ Vb )
= =φ + Δφ = + Δφ (1.41)
Vf Vf Vf Vf Vb V f Vf
1 Waves in poroelastic solid saturated by a single-phase fluid 11
(V f + Δ V f )Δ Vb (V b + Δ Vb )Δ φ Δ Vb Δ φ
= + +
VbV f Vf Vb φ
Δp 1 1 1−φ
=− + − Δ p.
Ku φ Ks Km
Thus,
1 1 1 (1 − φ ) 1 1 1 (1 − φ )
Δp − + − = Δ pf − + − .
Ku φ Ks Km K f φ Ks Km
Km + Ξ K f (Ks − Km )
Ku = Ks , Ξ= . (1.44)
Ks + Ξ φ (Ks − K f )
Thus,
Ku 1 1
1= +B − , (1.45)
Km Ks Km
12 1. .3 Determination of the elastic coefficients
B 1 1
0= +M − . (1.46)
Km ks km
Ks K f (Ks − Km )
B= . (1.47)
Ks φ (Ks − K f ) + K f (Ks − Km )
The elastic coefficients B and M can also be determined using the unjacketed
compressibility test [Biot and Willis, 1957] corresponding to a tensional state of the
form
Δ p = 0, Δ τ11 = Δ τ22 = Δ τ33 = −Δ p = −Δ p f .
In this test, a sample of bulk material is immersed in a container with the same
fluid as that inside the pore space and then subjected to a hydrostatic pressure change
Δ p.
Thus, in this case, the pressure change is supported by both the solid and fluid
parts of the bulk material, and the residual stress vanishes. Thus, according to (1.34),
Δ φ = 0. (1.52)
Δ Vs Δ ((1 − φ )Vb ) (1 − φ )Δ Vb − Δ φ Vb
= =
Vs Vs Vs
[1 − (φ + Δ φ )]Δ Vb Δ Vb
= ≈ ,
(1 − φ )V b Vb
ΔVf Δ (φ Vb ) φ Δ Vb Δ Vb
= = ≈ .
Vf Vf φ Vb Vb
1 Waves in poroelastic solid saturated by a single-phase fluid 13
Thus,
ΔVf Δ Vs Δ Vb
= = . (1.53)
Vf Vs Vb
Since
Δ Vs Δp
=− ,
Vs Ks
we conclude that
Δp
es = − . (1.54)
Ks
Also, using (1.20), (1.21) and (1.53)
c
ΔVf ΔVf Δ p Δ pf 1 1
ξ =φ − =φ − + =φ − Δ p. (1.55)
Vf Vf Ks Kf K f Ks
Now, from (1.56) and algebraic manipulations using the expression for Ku in (1.44)
we recover the expression for B and M given in (1.47) and (1.48).
2W = Ku (es )2 + 4μ (ε12
2
+ ε13
2
+ ε23
2
) (1.57)
2
+ μ (ε11 − ε22 )2 + (ε11 − ε33 )2 + (ε22 − ε33 )2 − 2Bes ξ + M ξ 2 .
3
es = ξ = 0
2W = Ku (es )2 − 2Bes ξ + M ξ 2
14 1.3 Determination of the elastic coefficients
Ku −B es
= (e s
ξ) .
−B M ξ
Ku M − B2 = Ku M − α 2 M 2 = (Ku − α 2 M)M.
Next, set
2
K = λ + μ.
3
Then,
2 2
Ku − K = λu + μ − λ − μ = λu − λ = α 2 M
3 3
so that
Ku M − B2 = (Ku − α 2 M)M = KM.
Therefore, for W to be nonnegative, we have the necessary and sufficient conditions
2
μ > 0, M > 0, K = λ + μ > 0. (1.58)
3
1
ξ= Δ p f + α es . (1.59)
M
Using (1.59) in (1.16) we obtain
Now using (1.22) in (1.60) to write the strain Δ τi j in terms of the residual stress
Δ τi j and the fluid pressure Δ p f the following relation is obtained:
Hence,
Δ p Δ p
es = − =− .
λ + 3μ
2 K
1 Waves in poroelastic solid saturated by a single-phase fluid 15
Thus, the requirement K = λ + 23 μ > 0 in (1.58) simply states the physically mean-
ingful condition that, for the open system , the inverse of the jacketed compressibil-
ity be positive.
It is well known that wave dispersion and attenuation phenomena in real saturated
rocks are higher than the associated to viscodynamic effects [Mochizuki, 1982,
Stoll and Bryan, 1970, Carcione, 2014]. This is mainly due to the complexity of
pore shapes, heterogeneities in the physical properties and in the distribution of the
fluids and the intrinsic anelasticity of the frame. These factors can be included in
the formulation by means of the theory of viscoelasticity. The theoretical basis for
this generalization was given by Biot (1956a,1962), who developed the general the-
ory of deformation of porous materials saturated by viscous fluids when the solid
phase exhibits linear viscoelastic behaviour. Using principles of irreversible thermo-
dynamics Biot established a general operational relationship between generalized
forces Qi and observed coordinates qi , of the form
Qi = Ti j q j ,
Hence, using Fourier transform in time we can state in the space–frequency do-
main the constitutive relations (1.16)-(1.17) as follows:
d ∂ Td ∂ Dd ∂ Vd
+ = − f , 1 ≤ i ≤ 3. (1.63)
dt ∂ u̇if ∂ u̇if
∂ ui
be the potential energy of the fluid–solid system where gsi , gif are given in (1.5).
Now we consider a perturbation of the system from the equilibrium state; i.e.,
the conditions (1.3) do not hold anymore. Then, using the argument leading to (1.7)
and the relation (1.10) we get
δV = δ W dΩ − ((Δ τi j + φ Δ p f δi j )δ usi ν j − φ Δ p f δi j δ uif ν j )d(∂ Ω )
Ω ∂Ω
= δ W dΩ − (Δ τi j δ usi ν j − Δ p f δi j δ uif ν j )d(∂ Ω ) (1.64)
Ω ∂Ω
∂ ∂
= δ W dΩ − (Δ τi j δ usi )d Ω + (Δ p f δi j δ uif )d Ω
Ω Ω ∂xj Ω ∂xj
∂ Δ τi j s
= (Δ τi j δ εi j + Δ p f δ ξ )d Ω − δ ui d Ω − Δ τi j δ εi j d Ω
Ω Ω ∂xj Ω
∂Δ pf f
+ δ ui d Ω + Δ p f (−δ ξ )d Ω .
Ω ∂ xi Ω
Hence,
∂ Δ τi j s ∂ Δ p f f
δV = − δ ui − δ ui d Ω = δ Vd d Ω .
Ω ∂xj ∂ xi Ω
1 Waves in poroelastic solid saturated by a single-phase fluid 17
Thus,
∂ Δ τi j s ∂ Δ p f f
δ Vd = − δ ui + δ ui .
∂xj ∂ xi
Assuming that Vd is an exact differential in the variables usi and uif , we see that
∂ Vd ∂ Δ τi j ∂ Vd ∂Δ pf
=− , = , i = 1, 2, 3. (1.65)
∂ ui
s ∂xj ∂ uif ∂ xi
Next, we will compute the kinetic energy density Td for the fluid–solid system.
Let us consider a unit cube Q of bulk material, and let Q p denote the porous part of
Q. Let ρ f and ρs be the mass densities of the fluid and solid phases, respectively.
Let (vi )1≤i≤3 be the relative micro-velocity field; i.e., the velocity of each fluid
particle with respect to the solid frame. Assuming that the relative flow inside the
pore space is of laminar type (i.e., we are in the low frequency range) we can write
vi = ai j u̇ fj ,
with the coefficients ai j depending on the coordinates of the pores and the pore
geometry. Let
ρ1 = (1 − φ )ρs
be the mass of solid per unit volume of bulk material. Then, on the solid part of Q
the kinetic energy is given by
1 1 1
ρs u̇si u̇si d(Q \ Q p ) = |Q \ Q p |ρs u̇si u̇si = ρ1 u̇si u̇si . (1.66)
2 Q\Q p 2 2
Here we have used that since usi is the average solid displacement over Q, usi is
constant over Q. In (1.66) |Q \ Q p | indicates the measure of the set Q \ Q p .
Next, on the porous part Q p , the velocity of any given particle is the relative
microvelocity plus the averaged solid velocity; i.e., u̇si + vi . Then the kinetic energy
in Q p is obtained by integration of (u̇si + vi )(u̇si + vi ) over Q p . Thus, the total kinetic
energy per unit volume of bulk material is given by
1 1
Td = ρ1 u̇si u̇si + ρ f (u̇si + vi )(u̇si + vi )dQ p . (1.67)
2 2 Qp
and that
ρf u̇si vi dQ p = ρ f u̇si vi dQ p = ρ f u̇si u̇if , (1.69)
Qp Qp
18 1.5 The equations of motion. Low frequency range
since the averaged relative fluid velocity is obtained by averaging the relative micro-
velocity field over Q p .
Next,
ρf vk vk dQ p = ρ f aki u̇if ak j u̇ fj dQ p (1.70)
Qp Qp
= ρf aki ak j dQ p u̇if u̇ fj = gi j u̇if u̇ fj ,
Qp
where
gi j = ρ f aki ak j dQ p .
Qp
Note that gi j = g ji .
Using (1.68), (1.69), and (1.70) in (1.67), we obtain
1 1
Td = ρ u̇si u̇si + ρ f u̇si u̇if + gi j u̇if u̇ fj , (1.71)
2 2
where
ρ = ρ1 + ρ f φ = (1 − φ )ρs + φ ρ f
is the mass density of bulk material.
Note that gi j must be positive definite, otherwise, we may have, for usi ≡ 0,
1
T = gi j u̇if u̇ fj = 0 for ufi = 0.
2
For an isotropic micro-velocity field, we have that
gi j = gδi j ,
must be satisfied.
Next, we will compute the form of the dissipation energy density function Dd .
Following [Biot, 1956a], we will assume that dissipation depends only on the rel-
ative flow between the fluid and the solid. Assuming that the relative flow is of
Poiseuille type, the microscopic flow pattern inside the pores is uniquely deter-
˙ f . The dissipation function vanishes
mined by the six generalized velocities u̇si , ui
f
˙ . Thus, we can write Dd in the form
when u̇s = u
i i
1 Waves in poroelastic solid saturated by a single-phase fluid 19
1
Dd = η ri j u̇if u̇ fj (1.73)
2
where η is the fluid viscosity and ri j is a symmetric positive definite matrix. Now,
from (1.71) and (1.73) we have that
∂ Td ∂ Td
= ρ u̇sk + ρ f u̇kf , = ρ f u̇sk + gk j u̇ fj , (1.74)
∂ u̇sk ∂ u̇kf
∂ Dd ∂ Dd
= 0, = η rk j u̇ fj .
∂ u̇sk ∂ u̇kf
Thus, combining (1.65) and (1.74) we see that the Lagrange equations (1.62)-(1.63)
become
∂ Δ τi j
ρ üsi + ρ f üif = , (1.75)
∂xj
∂Δ pf
ρ f üsi + gi j üif + η ri j u̇ fj = − , (1.76)
∂ xi
which are Biot’s equation of motion for the fluid–solid system.
Note that in the case of steady flow rate (u̇if = const) and vanishing solid accel-
erations from (1.76) we have that
∂Δ pf
η ri j u̇ fj = . (1.77)
∂ xi
Let κ = (κi j ) be the inverse of the matrix R = (ri j ). Then, (1.77) is Darcy’s Law
ηκ −1 u̇ f = ∇ p f ,
Equations (1.79)- (1.80) together with the constitutive relations given in (1.16)-
(1.17) completely determines the dynamic behaviour of the solid–fluid system in
the low–frequency range.
Let us write the equations of motion (1.79)-(1.80) and the constitutive relations
(1.16)-(1.17) in terms of usi , uif , in order to recover Biot’s equation in the original
form in [Biot, 1956a], which validity is restricted to constant porosity case. Using
(1.2), from (1.79) we have
20 1.5 The equations of motion. Low frequency range
¨ f − üs ) = ∂ Δ σi j + ∂ Δ σ .
f
ρ üsi + ρ f φ (u (1.81)
i i
∂xj ∂ xi
∂Δσ f ∂Δ pf
= −φ = φ ρ f üsi + φ güif + ηκ −1 φ u̇if (1.82)
∂ xi ∂ xi
¨ f − üs ) + ηκ −1 φ 2 (u
= φ ρ f üs + φ 2 g(u ˙ f − u̇s )
i i i i i
2 ¨f −1 2 ˙ f − u̇s ).
= (φ ρ f − φ 2
g)üi + φ gui
s
+ ηκ φ (ui i
∂ Δ σi j ¨ f + φ 2 ηκ −1 (u̇s − u
˙ f ). (1.83)
= (ρ − 2φ ρ f + φ 2 g)üsi + (φ ρ f − φ 2 g)u
∂xj i i i
Set
ρ11 = ρ − 2φ ρ f + φ 2 g, ρ12 = φ ρ f − φ 2 g,
ρ22 = φ 2 g, b = φ 2 ηκ −1 .
u f − us )) = φ (es − θ).
ξ = −∇ · u f = −∇ · (φ (
Using (1.2), (1.16) and (1.86) and that λu = λ + α 2 M (see (1.15)), we obtain
Δ σi j = Δ τi j − δi j Δ σ f (1.87)
= [λ + M(α − φ )2 ]es + φ M(α − φ )θ δi j + 2μεi j .
Setting
A = λ + M(α − φ )2 , P = φ (α − φ )M, R = φ 2 M, (1.88)
we can rewrite (1.85) and (1.87) in the form
1 Waves in poroelastic solid saturated by a single-phase fluid 21
The equations of motion (1.79)-(1.80) were derived under the assumption that the
flow inside the pore space is of Poiseuille type. This assumption breaks down if the
frequency exceeds a certain critical value ωc . This occurs when inertial and viscous
forces in (1.76) are of the same order, i.e., when g ω ≈ ηκ −1 , so that
ηκ −1 ηκ −1 φ
ωc = = , (1.91)
g ρf S
By analyzing the flow in cylindrical ducts and in plane slits [Biot, 1956b] con-
cludes that in the high frequency range the equations of motion (1.84)-(1.85) must
be modified employing a universal function
1 θ (ω )T (θ (ω ))
F(θ (ω )) = = FR (θ (ω )) + iFI (θ (ω )),
4 2
1− T (θ (ω ))
iθ ( ω )
ber (θ (ω )) + ibei (θ (ω ))
T (θ (ω )) = (1.93)
ber(θ (ω )) + ibei(θ (ω ))
that can be adopted to represent the frequency effect with a non-dimensional param-
eter
1
ω 2
θ (ω ) = a p
η
where a p is the pore–size parameter depending on size and pore geometry and
22 1.6 The equations of motion. High frequency range
= η /ρ f
η
−ω 2 ρ11 us (ω ) − ω 2 ρ12
u f (ω ) − iω b F(ω )(
u f (ω ) − us (ω )) (1.96)
= ∇·Δσ,
−ω ρ12 u (ω ) − ω ρ22
2 s 2
u (ω ) + iω b F(ω )(
f
u f (ω ) − us (ω )) (1.97)
= −φ ∇Δ p f ,
s
(ω )) = [A e (ω ) + P θ (ω )]δi j + 2μεi j (us (ω )),
Δ σi j (u (ω ), u f s
(1.98)
s
) = P e (ω ) + R θ (ω ).
−φ Δ p f (u (ω ), u f s
(1.99)
Next, after algebraic manipulations, we can write the equations (1.96)-(1.97) us-
ing the variables us (ω ) and u f (ω ) = φ (
u f (ω ) − us (ω )) as:
−ω 2 ρ us (ω ) − ω 2 ρ f u f (ω ) − ∇ · Δ τ = f(1) , (1.100)
−ω 2 ρ f us (ω ) − ω 2 g(ω )u f (ω ) + iω b(ω )u f (ω ) + ∇Δ p f = f(2) , (1.101)
where f(1) and f(2) are external forces in the bulk material and the fluid per unit bulk
volume and
Sρ f FI (ω ) −1
g(ω ) = + ηκ ,
φ ω
b(ω ) = ηκ −1 FR (ω ).
1 1
F(θ (ω )) → θ (ω ) √ (1 + i), as θ (ω ) → ∞
4 2
Also,
θ (ω ))2
F(θ (ω )) → 1 + i , as θ (ω ) → 0.
24
For the function in (1.94),
2κ S −iωρ f 1/2
F(ω ) → , as ω →∞
Λφ η
F(ω )) → 1 as ω → 0.
Thus at low frequencies the low–frequency coefficients are recovered, and at high
1
frequencies these correcting functions behave like ω 2 .
Remark. If viscoelasticity is included, the coefficients λ u , μ and M in (1.102)-
(1.103) become complex and frequency dependent.
−ω 2 ρ es − ω 2 ρ f e f = Eu ∇2 es + B ∇2 e f , (1.106)
−1 f
−ω ρ f e − ω g e + iωηκ
2 s 2 f
e = B∇ e +M ∇ e ,
2 s 2 f
(1.107)
where e f = ∇ · u f .
24 1.7 Plane wave analysis. Attenuation and dispersion effects
where
ηκ −1
g = g + i
ω
leads to the following generalized eigenvalue problem:
γ 2 A C (γ ) = E C (γ ) . (1.110)
Now from (1.110) it is seen that to determine the complex wave-numbers = r + ii
it is sufficient to solve the problem
det(S − γ 2 I) = 0, (1.111)
where
S = A −1 E .
()
Instead of the attenuation coefficient i , it is convenient to use another attenuation
coefficient defined as follows: from (1.108) and (1.109) we see that at x1 = 0, the
original wave amplitude amplitude for eθ , θ = s, f , is
(( j) )
eθ1 = Cθ .
2π v( j)
λ ( j) = ,
ω
after travelling one wavelength the wave has amplitude
( j) 2π v( j)
eθ2 = eθ0 e−i ω .
Thus,
eθ2 ( j) 2π
log10 = −i log10 (e).
eθ1 ( j)
r
We define the attenuation coefficient b( j) measured in dB by the formula
θ
e ( j) ( j)
b( j) = −20 log10 2θ = (2π )(8.685889)i /|r |.
e1
Hence this coefficient b( j) measures the wave attenuation after travelling one wave-
length. For example, an attenuation coefficient b( j) of 20 dB implies that after trav-
elling one wavelength the Pj -wave has reduced ten times its original amplitude.
Next we consider rotational waves. Let
ks = ∇ × us , kf = ∇×uf .
Then applying the curl operator to equations (1.104)-(1.105) we obtain the relations
governing the propagation of rotational waves:
ρ ks + ρ f k f = μΔ ks , (1.112)
−1 f
ρ f k + gk + iηκ
s f
k = 0. (1.113)
Let us consider a plane rotational wave of angular frequency ω and wave number
= r + ii travelling in the x1 –direction:
() ω
ks = C1 e−i x1 eir (x1 − r t) , (1.114)
() ω
kf = C2 e−i x1 eir (x1 − r t) . (1.115)
In this section we compute phase velocities and attenuation coefficients for a sample
of Nivelsteiner sandstone, a friable sandstone mainly composed of quartz with small
percentages of rock fragments and potash-feldspar [Kelder and Smeulders, 1997].
Its material properties, taken from [Arntsen and Carcione, 2001], and those of the
saturant fluids, water, oil and gas, are given in Table 1.1.
The gas properties correspond to a dry gas at a reference pressure of 5MPa, (at
a depth of 500 m, approximately ) using the calculations given in [Standing, 1977]
and [McCoy, 1983].
Figures 1.3, 1.4 and 1.5 show phase velocities for P1, S and P2 waves as function
of frequency, while Figures 1.6, 1.7 and 1.8 display the corresponding attenuation
coefficients. It is observed that for the three saturating fluids, P1 and shear waves
have phase velocities almost independent of frequency. Figure 1.3 shows that P1
1 Waves in poroelastic solid saturated by a single-phase fluid 27
waves have the highest and lowest velocities for the water and oil saturated cases,
respectively, while the gas saturated sample has intermediate velocity values. On the
other hand, Figure 1.4 shows that shear waves have the highest values for the gas
saturated case. Also, up to about 1 kHz, Figure 1.4 exhibits the lowest velocity for
the water saturated case, and the oil saturated case has intermediate values between
the water and gas cases. Above 1 kHz, both the water and oil curves show an increase
behaviour and at 100 kHz the oil saturated sample has higher velocities than the
water saturated one.
For P2 waves, Figure 1.5 shows that for all cases velocities almost vanish at low
frequencies and display an increasing behaviour. The water saturated case has the
highest velocities in all the frequency range. At high frequencies, the gas saturated
case exhibits the lowest velocities, and the oil case shows intermediate values be-
tween the water and gas cases. At low frequencies, the gas and oil curves show the
opposite behaviour.
Concerning attenuation for P1 waves, Figure 1.6 shows maximum and minimum
attenuations for the oil and gas saturated cases, respectively, and intermediate max-
imum attenuation for water saturated samples. Also, the attenuation peaks move to
higher frequencies as the fluid viscosity increases. For shear waves, the attenuation
peaks also move to higher frequencies with increasing fluid viscosity, the maximum
and minimum attenuation is observed for the water and gas saturated cases, repec-
tively, with the oil saturated case having intermediate maximum attenuation. Both
P1 and shear waves suffer negligible attenuation below 100 Hz, and shear wave
attenuation is always higher than P1 attenuation.
P2 waves attenuation exhibit a different behaviour than the fast P1 and shear
waves. Attenuation values are very high at low frequencies, showing that they are
diffusion-type waves. After 100 Hz, all curves have a decreasing behaviour, with P2
waves suffering the highest attenuation for the oil case, the lower attenuation for the
gas case, and the water case having an intermediate behaviour. After 1 MHz (ultra-
sonic range), P2 attenuation is negligible and P2 waves become truly propagating
waves.
28 1.8 Application to a real sandstone
2900
2800
2600
Fig. 1.3 Phase velocity of
P1 waves as function of fre-
quency for a sample of Nivel-
steiner sandstone saturated by 2500
0 1 2 3 4 5 6 7
water, oil and gas. Frequency (Hz) - Logarithmic Scale
1650
1600
Shear Wave Phase Velocity (m/s)
Water
1550 Oil
Gas
1500
800
Water
P2 Wave Phase Velocity (m/s)
Oil
Gas
600
400
200
Fig. 1.5 Phase velocity P2
waves as function of fre-
quency for a sample of Nivel- 0
steiner sandstone saturated by 0 1 2 3 4 5 6 7
water, oil and gas. Frequency (Hz) - Logarithmic Scale
1 Waves in poroelastic solid saturated by a single-phase fluid 29
0.4
0.3 Water
Oil
0.2
0.1
Fig. 1.6 Attenuation coeffi-
cient of P1 waves as function
of frequency for a sample
0
of Nivelsteiner sandstone
saturated by water, oil and 0 1 2 3 4 5 6 7
gas. Frequency (Hz) - Logarithmic Scale
Water
0.8 Oil
Gas
Shear Wave Attenuation (dB)
0.6
0.4
First recall that for any given complex and frequency dependent modulus M(ω ) the
quality factor is defined by
Re(M(ω ))
QM (ω ) = . (1.118)
Im(M(ω ))
Next, we define the Zener or standard linear solid model associated with a given
elastic modulus M.
The dimensionless Zener element can be written in the form
30 1.9 Appendix 1. Models of linear viscoelasticity
60
Water
50 Oil
Gas
30
20
1 + iω tε
Nz (ω ) = . (1.119)
1 + iω tσ
In (1.119) tε and tσ are relaxation times given by
t0 2t0
tε = 1 + Q20 + 1 , tσ = tε − ,
Q0 Q0
where to is a relaxation time such that 1/t0 is the center frequency of the relaxation
peak and Q0 is the minimum quality factor of the complex modulus
M(ω ) = M Nz (ω ).
Next we formulate a model that for given elastic modulus M yields constant
quality factors over a frequency range of interest.
Such behaviour is modeled by a continuous distribution of relaxation mecha-
nisms based on the standard linear solid (see [Liu et al., 1976] and [Ben-Menahem and
Sing, 1981], pp. 909).
The dimensionless complex moduli for a specific frequency can be expressed as
2 1 + i ω t2
Nl (ω ) = 1 + ln , (1.120)
π QM 1 + iω t1
where t1 and t2 are time constants, with t2 < t1 , and the quality factor Q(ω ) associ-
ated with the complex modulus
M(ω ) = M Nl (ω ) (1.121)
remains nearly constant and equal to QM over the selected frequency range. The com-
plex modulus in (1.121) can also be written in the equivalent form [Bourbie et al., 1987]
1 Waves in poroelastic solid saturated by a single-phase fluid 31
M
M(ω ) = (1.122)
β ( ω ) − iγ ( ω )
where
1 1 + ω 2t12 2 ω (t1 − t2 )
βl ( ω ) = 1 − ln , γl ( ω ) = tan−1 . (1.123)
M 1 + ω 2t22
πQ
π QM 1 + ω 2 t 1 t2
Chapter 2
A poroelastic solid saturated by two immiscible
fluids
2.1 Introduction
Theoretical formulations for the study of the deformation and elastic wave propa-
gation in porous rocks with partial, multi-phase, or segregate fluid saturation have
been presented in several papers (see [Dutta and Odé, 1979, Berryman et al., 1988,
Mochizuki, 1982] among other authors).
However, none of these models incorporates the capillary forces existing when
the pore fluids are immiscible. Consequently, the pressure variations induced by
wave propagation in the different fluid phases are considered almost equal, neglect-
ing possible changes in capillary pressure.
For the case of multi-phase fluids, we mention an analysis of wave propagation
in porous media saturated by immiscible fluids presented in [Corapcioglu, 1996].
Later, [Lo et al., 2005] derived a model for waves travelling in an elastic porous
solid permeated by two immiscible fluids incorporating both inertial and viscous
© Springer International Publishing AG 2016 33
J.E. Santos, P.M. Gauzellino, Numerical Simulation in Applied Geophysics, Lecture Notes
in Geosystems Mathematics and Computing, DOI 10.1007/978-3-319-48457-0_2
34 2.2 Constitutive relations
When two immiscible fluids occupy the voids of a poroelastic solid one of them
(depending on their adhesion tension), tends to preferentially wet the solid surface,
spreading over it. In this way we can distinguish a wetting phase and a non-wetting
one, which will be denoted with the subscripts (or superscripts) “w” and “n”, respec-
tively. Let Sw and Sn denote the averaged wetting and non-wetting fluid saturations,
respectively. Furthermore, let us assume that the two fluid phases completely satu-
rate the porous part of the bulk material so that
Sw + Sn = 1.
The symbols Srw and Srn will denote the residual wetting and non-wetting fluid
saturations, respectively. We assume that both fluids are allowed to move inside the
pore space, so that
Δ Sn = Sn − Sn = −Δ Sw .
uq − us ),
uq = φ ( q = n, w. (2.1)
Also set
ξ q = −∇ · uq , q = n, w.
Note that for uniform porosity Sn ξ n and Sw ξ w represent the change in the cor-
responding fluid contents per unit volume of bulk material [Santos et al., 1990b].
Thus, if Δ Vqc denotes the part of the total change in volume Δ Vq = Vq − V q due to
changes Δ pq = pq − pq in the corresponding fluid pressures, since in equilibrium
∇Sq = 0, we see that
Δ Vq − Δ Vqc Δ Vq − Δ Vqc
Sq ξ q = =φ , q = n, w. (2.2)
Vb Vf
Let Kn = Cn−1 and Kw = Cw−1 denote the bulk moduli of the non-wetting and
wetting fluids, respectively, Cn and Cw being the corresponding compressibilities.
Then, by definition
Δ Vqc Δ pq
=− , q = n, w.
Vq Kq
Also, neglecting second order terms,
Δ Vq = V f Δ Sq + Sq Δ V f , q = n, w,
and consequently
Δ Sq Δ V f Δ Vqc
ξ =φ
q
+ − , q = n, w.
Sq Vf Vq
Setting
ξ ∗ = Sn ξ n + Sw ξ w ,
it follows from (2.2) that
36 2.2 Constitutive relations
Δ V f − Δ Vnc − Δ Vwc
ξ∗ = φ .
Vf
The function Pca (Sn ) is a positive and strictly increasing function of the variable Sn ;
hysteresis effects are ignored.
Let us consider a volume Ω of bulk material of boundary ∂ Ω in static equi-
librium state under the action of the surface forces fis , fin , fiw , i = 1, 2, 3, acting on
the solid and fluid parts of ∂ Ω per unit area of bulk material. These forces can be
written in the form [Santos et al., 1990b]
The principle of complementary virtual work for our system, including capillary
forces, can be stated in the form [Santos et al., 1990b]:
δ W ∗ dΩ = (δ fis usi + δ fin uni + δ fiw uwi ) d(∂ Ω ) (2.6)
Ω ∂Ω
− [δ λnw (Δ pn − Δ pw − Δ Pca ) + λnw (δ Δ pn − δ Δ pw − δ Δ Pca )]d Ω ,
Ω
where δ denotes virtual changes and λnw is a Lagrange multiplier used to introduce
the capillary relation (2.3) as a constraint. Neglecting second order terms in the
different increments and using (2.4) we have [Santos et al., 1990b]
δ Δ Pca
δ fi = −φ Sn δ Δ pn + pn
n
δi j ν j , (2.7)
P ca (Sn )
2 A poroelastic solid saturated by two immiscible fluids 37
δ Δ Pca
δ fiw = −φ Sw δ Δ pw − pw δi j ν j , (2.8)
Pca (Sn )
δ Δ Pca
δ fis = −φ Sn δ Δ pn + Sw δ Δ pw + Pca (Sn ) δi j ν j + δ Δ τi j ν j . (2.9)
Pca (Sn )
where
Pca (Sn ) pw
βnw = , βnw
w
= .
P ca (Sn ) Pca (Sn )
Δ τi j (u) = 2μ εi j + δi j (λu es − B1 ξ n − B2 ξ w ),
Δ Tn (u) = Sn + βnw + βnw w
Δ pn − (βnw + βnw w
) Δ pw = −B1 es + M1 ξ n + M3 ξ w ,
Δ Tw (u) = Sw + βnw w
Δ pw − βnww
Δ pn = −B2 es + M3 ξ n + M2 ξ w , (2.11)
Δ Vb
es = εii = .
Vb
38 2.3 Inclusion of linear viscoelasticity
Ks (Km + Ξ ) K ∗f (Ks − Km )
Ku = , Ξ= , (2.12)
(Ks + Ξ (φ (Ks − K ∗f ))
−1
K ∗f = α γ SnCn + SwCw , α = 1 + (Sn + βnw )(γ − 1),
−1
γ = 1 + Pca (Sn )Sn SwCw 1 + Pca (Sn )Sn SwCn ,
where Km = Cm−1 and Ks = Cs−1 denote the bulk modulus of the empty matrix and
the solid grains, respectively.
In the equations above, K ∗f = (C∗f )−1 defines an effective bulk modulus for a
two–phase fluid taking into account capillary forces.
Here we are assuming that the solid matrix is composed of a homogeneous ma-
terial. In the case of mixed mineralogy an effective average coefficient for Ks can be
used (see [Mavko et al., 1998]).
Using the expression of Ku in (2.12), the remaining coefficients can be obtained
by using the following relations [Ravazzoli et al., 2003]:
where
−1
χ = Cs −Cm + φ (Cm −Cu ) α (Cs −Cm ) + φ Cm −C f ,
1
b = φ Cn + ,
P ca (Sn )Sn Sw
CmCu
a = (Sn + βnw )Cs + b B2 + (Sn + βnw ) 1 −CsCu−1 .
(Cm −Cu )
behaviour, we are able to extend the constitutive relations (2.11) by simply replacing
the real elastic moduli μ , Ku , M1 , M2 and M3 by appropriate viscoelastic operators.
Using Fourier transform in time we can state this relations in the space–frequency
domain as follows:
2
where μ (ω ), λu (ω ) = Ku (ω ) − μ (ω ), M1 (ω ), M2 (ω ) and M3 (ω ) are complex fre-
3
quency dependent poroviscoelastic moduli.
Also, as indicated in Chapter 1, possible choices to make the poroviscoelastic
moduli frequency dependent are given in Appendix 1.9.
First we will get an expression for the complementary potential energy density Vd∗
of the system. Consider a perturbation of the system from the equilibrium state.
According to (2.5)
δV ∗ = δ Vd∗ d Ω = (δ fis usi + δ fin ũni + δ fiw ũwi )d(∂ Ω ).
Ω ∂Ω
Since we are in the linear case, the complementary strain energy W ∗ equals the
strain energy W , also Vd∗ = Vd [Fung, 1965]. Then, assuming that Vd is an exact
differential we have
∂ Δ τi j ∂ ∂
Vd = −usi + uni Δ Tn + uwi Δ Tw + λ (Δ Pca − (Δ pn − Δ pw )).(2.16)
∂xj ∂ xi ∂ xi
∂ Vd ∂ Δ τi j ∂ Vd ∂ ∂ Vd ∂
=− , = Δ Tn , = Δ Tw , 1 ≤ i ≤ 3. (2.17)
∂ usi ∂xj ∂ uni ∂ xi ∂ uwi ∂ xi
Next, in the isotropic case the kinetic energy density Td and the dissipation density
function Dd have the form [Santos et al., 1990b]
40 2.4 The equations of motion. Low frequency range
1
Td = ρ u̇si u̇si + ρn Sn u̇si u̇ni + ρw Sw u̇si u̇wi (2.18)
2
1 1
+ gcn u̇ni u̇ni + gcw u̇wi u̇wi + gcnw u̇ni u̇wi ,
2 2
1 c n n 1 c w w
Dd = b u̇ u̇ + b u̇ u̇ + bcnw u̇ni u̇wi , (2.19)
2 n i i 2 w i i
where ρw and ρn are the mass densities of the wetting and the non-wetting fluids
and ρ is the density of the bulk material, i.e.,
ρ = (1 − φ )ρs + Sn ρn + Sw ρw ,
1 1
ε S ρn ρw S n S w 2 (ηn ηw ) 2 Sn Sw Krnw
gcnw = , bcnw = , (2.21)
φ A
The factor S is known as a structure factor and is related to the tortuosity of the pore
space; it can be estimated as in (1.92).
The constants ηn , ηw are the non-wetting and wetting fluid viscosities, respec-
tively. Also,
A = κ Krn Krw − Krnw 2
,
Krw Krn
An = , Aw = ,
A A
with κ , Krn , Krw and Krnw denoting the absolute and relative permeabilities, respec-
tively. The relative permeabilities are assumed to be functions of the non-wetting
saturation Sn [Bear, 1972, Scheidegger, 1974, Peaceman, 1977].
The Lagrange formulation of the equations of motion is given by
d ∂ Td ∂ Dd ∂ Vd
+ =− , 1 ≤ j ≤ 9. (2.22)
dt ∂ u̇ j ∂ u̇ j ∂uj
In the isotropic case, from (2.17),(2.18), (2.19) and (2.22) we obtain the equations
where f s , f n and f w indicate external forces in the solid, non-wetting and wetting
phases, respectively.
To obtain (2.23)-(2.25) we have applied the dimensional argument given in
[Santos et al., 1990b] using that Sn can be assumed independent of time.
Sq ηq φ Aq
ωcq = , q = n, w, (2.26)
(Sρq )
1 1
φ (ηn ηw ) 2 Sn Sw 2 Krnw
ωcnw = 1 . (2.27)
( ε A S ( ρn ρw ) 2 )
These frequencies are obtained by equating the inertial and viscous drag terms as-
sociated with each fluid phase and using (2.20) and (2.21). For frequencies bigger
than the minimum of the three characteristic frequencies defined above, the mass
and viscous coupling coefficients are defined as follows.
If we employ the frequency correction function F(ω ) = FR (ω ) + FI (ω ) defined
in Chapter 1, equation (1.93), then we define
ηq (Sq )2 Aq FI (θq (ω ))
gq (ω ) = gcq + , q = n, w,
ω
1
(ηn ηw ) 2 Sn Sw Krnw FI (θnw )
gnw (ω ) = gcnw + ,
(ω A )
bq (ω ) = bcq FR (θq ), q = n, w, nw.
ηq aqp FIq (ω )
gq (ω ) = gcq + , bq (ω ) = bcq FRq (ω ), q = n, w,
ω
Next, from (2.13)-(2.15) and (2.23)-(2.25) we can state the equations of motion in
the space-frequency domain (including viscoelasticity effects if needed) as follows
−ω 2 ρ us (ω ) + ρn Sn un (ω ) + ρw Sw uw (ω ) − ∇ · Δ τ (u(ω )) = f s (ω ),(2.28)
−ω 2 ρn Sn us (ω ) + gn un (ω ) + gnw uw (ω ) (2.29)
+iω bn un (ω ) + iω bnw uw (ω ) + ∇Δ Tn (u(ω )) = f n (ω ),
−ω 2 ρw Sw us (ω ) + gnw un (ω ) + gw uw (ω ) + iω bw uw (2.30)
+iω bnw u (ω ) + ∇Δ Tw (u(ω )) = f (ω ).
n w
For a spatially homogeneous medium and in the absence of body forces, the
equations of motion (2.28)-(2.30) take the form:
−ω 2 ρ us (ω ) + ρn Sn un (ω ) + ρw Sw uw (ω ) = Eu ∇∇ · us (ω )
−μ ∇ × ∇ × us (ω ) + B1 ∇(∇ · un (ω )) + B2 ∇(∇ · uw (ω )), (2.31)
−ω 2 ρn Sn us (ω ) + gn (ω ) un (ω ) + gnw (ω ) uw (ω )
+iω bn (ω ) un (ω ) + iω bnw (ω ) uw (ω )
= B1 ∇(∇ · us ) + M1 (ω )∇(∇ · un (ω )) + M3 ∇(∇ · uw (ω )), (2.32)
−ω 2 ρw Sw us (ω ) + gnw(ω ) un (ω ) + gw (ω ) uw (ω )
+iω bw (ω ) uw (ω ) + iω bnw (ω ) un (ω )
= B2 ∇(∇ · us (ω )) + M3 ∇(∇ · un )(ω ) + M2 (ω )∇(∇ · uw (ω )), (2.33)
where
4
Eu = Ku + μ.
3
2 A poroelastic solid saturated by two immiscible fluids 43
−ω 2 ρ es − ω 2 ρo Sn en − ω 2 ρw Sw ew (2.34)
= Eu ∇2 es + B1 ∇2 en + B2 ∇2 ew ,
−ω 2 ρn Sn es − ω 2 gn en − ω 2 gnw uw + iω bn en + iω bnw ew (2.35)
= B1 Δ es + M1 ∇2 en + M3 ∇2 ew ,
−ω 2 ρw Sw es − ω 2 gnw en − ω 2 gw ew + iω bnw en + iω bw ew (2.36)
= B2 ∇2 es + M3 ∇2 en + M2 ∇2 ew .
and setting
ω
γ=
leads to the eigenvalue problem
γ 2 A C (γ ) = E C (γ ) ,
where
⎛ ⎞ ⎛ ⎞ ⎛ γ⎞
ρ ρn ρw Eu B1 B2 Cs
γ
A = ⎝ ρn gn gnw ⎠ , E = ⎝ B1 M1 M3 ⎠ , C γ = ⎝ Cn ⎠ ,
γ
ρw gnw gw B2 M3 M2 Cw
and
bn bw bnw
gn = gn + i , gw = gw + i , gnw = gn + i .
ω ω ω
Hence, to determine the complex wave-numbers = r + ii it is sufficient to solve
the problem
44 2.7 Application to a real sandstone
det(S − γ 2 I) = 0,
where
S = A −1 E .
Then take the curl operator in (2.31)-(2.33) and substitute (2.37)-(2.39) in the re-
sulting equations to obtain
!−1
ρn S n gew ρw S n − genw ρw S w + ρw S w gen ρw S w − genw ρn S n
γ =µ ρ−
2
gen gew − (e
gnw )2
that allows to compute the phase velocity v(s) and attenuation coefficients b(s) for
the shear mode.
The phase velocities and attenuation coefficients for all waves are given by
ω (s) (s)
v( j) = ( j)
, b( j) = (2π ) · 8.685889(|i |/|r |), j = 1, 2, 3, s.
|r |
In this section we compute phase velocities and attenuation coefficients for a sample
of Nivelsteiner sandstone. The material properties of the sandstone and the saturant
2 A poroelastic solid saturated by two immiscible fluids 45
fluids, water, oil and gas, are given in Table 1.1. Water is considered the wetting
phase.
The relative permeability functions Krn (Sn ) and Krw (Sn ) and the capillary pres-
sure Pca (Sn ) used in the following examples are computed using the relations
[Douglas Jr. et al., 1993, Ravazzoli et al., 2003]:
1 − Sn 2 1 − Sn − Srw 2
Krn (Sn ) = 1 − , krw (Sn ) = , (2.40)
1 − Srn 1 − Srw
2
1 Srn
Pca (Sn ) = A − . (2.41)
(Sn + Srw − 1)2 Sn2 (1 − Srn − Srw )2
Also, we take
Krnw (Sn ) = ε Krn (Sn ) Krw (Sn ), (2.42)
with the capillary pressure amplitude coefficient A equals 3 kPa and ε = 0.01. The
residual saturation values are Srn = Srw = 0.002.
The relations (2.40)-(2.42) are based on laboratory experiments performed on
different porous rocks during imbibition and drainage processes (neglecting hys-
teresis effects). However, it must be pointed out that those functions depend on the
nature of the porous medium and the wettability of the pore fluids involved.
To study how the solid and fluid phases move in the three compressional modes,
following [Biot, 1956a] it is convenient to write the equations of motion for the
elastic case (zero viscosity) in terms of the solid displacement us and the absolute
fluid displacements un and
uw . Set
en = ∇ ·
un , ew = ∇ ·
uw .
Using (2.1) in (2.31)-(2.33), taking divergence in the resulting equations and replac-
ing a plane compressional wave of angular frequency ω and (real) wave number
travelling in the x1 –direction of the form
() ω () ω () ω
es = Cs ei(x1 − t) , en = Cn ei(x1 − t) , ew = Cw ei(x1 − t) ,
γ 2 A C() = E C(γ ) ,
where
ω t
γ= , Cγ = Csγ , Cnγ , Cwγ ,
46 2.7 Application to a real sandstone
⎛ ⎞ ⎛ ⎞
ρ − φ ( ρn S n + ρw S w ) φ ρn S n φ ρw S w Eu − φ (B1 + B2 ) φ B1 φ B2
A =⎝ ρn Sn − φ (gn + gnw ) φ gn φ gnw ⎠, E = ⎝ B1 − φ (M1 + M3 ) φ M1 φ M3 ⎠.
ρw Sw − φ (gw + gnw ) φ gnw φ gw B2 − φ (M2 + M3 ) φ M3 φ M2
⎡ ⎤ ⎡ ⎤
v(1) v(2) v(3) v(1) v(2) v(3)
⎢s + − + ⎥ ⎢s + + − ⎥
C1 = ⎢
⎣n +
⎥, C2 = ⎢
⎣n +
⎥. (2.43)
+ + ⎦ + + ⎦
w + + − w + − +
Let us analyze the behaviour of all waves in the elastic case (zero viscosity) as func-
tion of saturation of the non-wetting phase. The behaviour of P1 waves is shown in
Figure 2.1. The Figure shows the characteristic decreasing behaviour for oil-water
mixtures shown by the corresponding Gassmann velocity v pG defined in (2.44). For
gas-water mixtures, at very low gas saturation we see a sharp decreasing of veloc-
ities until a low gas saturation value at which a continuous increasing behaviour is
observed.
2 A poroelastic solid saturated by two immiscible fluids 47
2800
Gas-water
2700
2600
Figure 2.2 exhibits strong differences in P2 phase velocities for the gas-water
and oil-water mixtures, and in both cases show decreasing velocities as non-wetting
saturation increases. Besides, P2 waves for the gas-water mixture have slower ve-
locities than the oil-water mixture because the opposite motion of the fluid and solid
phases should be smaller for lower density fluids.
800
Gas-water
Oil-water
600
P2 Wave Phase Velocity (m/s)
400
200
Figure 2.3 shows that P3 waves phase velocities for gas-water mixtures are al-
most independent of gas saturation, while for oil-water mixtures they decrease as
oil saturation increases. The P3 wave is associated with the existence of capillary
forces, and it is due to the in-phase motion of the solid and non-wetting phases,
what explains its higher values for oil-water mixtures as compared with gas-water
mixtures.
50
P3 Wave Phase velocity (m/s)
40
Gas-water
Oil-water
30
20
10
Let us analyze next the behaviour of all waves for the general case in which dissipa-
tion due to the relative motion of the viscous fluids and the solid phase is considered.
Besides, high frequency corrections of the viscodynamic coefficients are made for
frequencies above the minimum of the critical frequencies defined in (2.26)-(2.27).
Figures 2.5, 2.6, 2.7 and 2.8 show phase velocities for P1, shear, P2 and P3 waves
as function of frequency, respectively, while 2.9, 2.10, 2.11 and 2.12 display the cor-
responding attenuation coefficients. The non-wetting fluids are gas and oil of satu-
2 A poroelastic solid saturated by two immiscible fluids 49
1600
1580 Gas-water
1560
1540
1520
4 μ
v pG = Ku + μ , vsG = . (2.44)
3 ρ
At fixed water saturation Sw = 0.9, P1 waves in Figure 2.5 show higher velocities
for oil-water than for gas-water mixtures, while shear waves in this case, Figure
2.6, exhibit very little differences in velocities. On the other hand, at fixed water
saturation Sw = 0.5, again P1 waves have higher velocities for oil-water than
for gas-water mixtures, but for shear waves in this case the opposite behaviour is
observed, due to the lower density of gas and that the shear modulus is independent
of the saturant fluids.
Phase velocities of P2 and P3 waves have a quite different behaviour than the fast
P1 and shear waves. These waves suffer noticeable dispersion on the whole range
of frequencies and velocities tend to zero at low frequencies. At ultrasonic frequen-
cies, the inertial terms are dominant and velocities stabilize at constant values. For
the highest saturation of the wetting phase (Sw = 0.9) P2 waves have the highest ve-
locities for the non-wetting fluid of higher viscosity (oil), while P3 waves show the
opposite behaviour. For equal proportions of wetting and non-wetting fluids, the P2
and P3 waves exhibit higher velocities when the non-wetting fluid is that of higher
viscosity (oil).
Figures 2.9 and 2.10 show that attenuation for P1 and shear waves is negligible
at low frequencies. Also, for P1 waves and equal proportions of wetting and non-
wetting fluids, attenuation is higher for the fluid of lower viscosity (gas). For water
saturation Sw = 0.9 attenuation is higher when gas is the non-wetting phase; this is
the case exhibiting the highest attenuation.
50 2.7 Application to a real sandstone
For shear waves and water saturation Sw = 0.9, Figure 2.10 shows that attenua-
tion is dependent of the non-wetting phase, with one attenuation peak for gas-water
mixtures and two peaks for oil-water mixtures, one lower than the gas-water peak
and the second at an ultrasonic frequency of about 1 MHz. For equal proportions
of wetting and non-wetting fluids, the attenuation peak moves to higher frequencies
for the fluid of higher viscosity (oil).
In Figures 2.11 and 2.12 can be observed that attenuation for P2 and P3 waves is
very high at low frequencies up to about 100 Hz, indicating the diffusive character
of these two modes in the seismic range. Above 100 Hz, attenuation decays to zero
at ultrasonic frequencies, where the P2 and P3 waves become propagating modes.
For P2 waves and water saturation Sw = 0.9, attenuation is independent of the
non-wetting fluids, and for equal proportions of wetting and non-wetting fluids, the
oil phase is the one suffering the highest attenuation of all cases.
For P3 waves and equal proportions of wetting and non-wetting fluids attenuation
is higher when oil is the non-wetting phase, and for water saturation Sw = 0.9 the
highest attenuation is seen when oil is the non-wetting phase; this case is the one
exhibiting the highest attenuation.
2800
So = 0.1, Sw = 0.9
Sg = 0.5, Sw = 0.5
So = 0.5, Sw = 0.5
2600
2500
Fig. 2.5 Phase velocity of
P1 waves as function of
frequency for two different
values of oil and gas satura- 2400
-1 0 1 2 3 4 5 6 7
tions. Frequency (Hz) - Logarithmic Scale
2 A poroelastic solid saturated by two immiscible fluids 51
1600
Sg = 0.1, Sw = 0.9
So = 0.1, Sw = 0.9
Sg = 0.5, Sw = 0.5
1500
700
600
Sg = 0.1, Sw = 0.9
P2 Wave Phase Velocity (m/s)
So = 0.1, Sw = 0.9
500
Sg = 0.5, Sw = 0.5
So = 0.5, Sw = 0.5
400
300
200
200
Sg = 0.1, Sw = 0.9
150 So = 0.1, Sw = 0.9
P3 Wave Phase Velocity (m/s)
Sg = 0.5, Sw = 0.5
So = 0.5, Sw = 0.5
100
50
Fig. 2.8 Phase velocity of
P3 waves as function of
frequency for two different 0
values of oil and gas satura- -1 0 1 2 3 4 5 6 7
tions. Frequency (Hz) - Logarithmic Scale
52 2.7 Application to a real sandstone
Sg = 0.1, Sw = 0.9
0.5 So = 0.1, Sw = 0.9
Sg = 0.5, Sw = 0.5
So = 0.5, Sw = 0.5
0.3
0.2
Sg=0.1, Sw=0.9
0.8 So=0.1, Sw=0.9
Sg=0.5, Sw=0.5
So=0.5, Sw=0.5
Shear Wave Attenuation (dB)
0.6
0.4
0.2
Fig. 2.10 Attenuation co-
efficient of shear waves as
0
function of frequency for two
different values of oil and gas 0 2 4 6
saturations. Frequency (Hz) - Logarithmic Scale
60
50
P3 Wave Attenuation (dB)
40
Sg = 0.1, Sw = 0.9
So = 0.1, Sw = 0.9
30
Sg = 0.5, Sw = 0.5
So = 0.5, Sw = 0.5
20
60
50
20
3.1 Introduction
θ ,T ,θ = o, w, g, denote the
Let φ = φ (x) be the effective porosity in Ω and let us,T , u
locally averaged solid and absolute θ -fluid displacements in Ω .
θ
Consider an initial state of equilibrium about displacements us , u and let us
define the infinitesimal displacements
θ
us = us,T − us , θ = u
u θ ,T − u , θ = o, w, g.
σθ = −φ Sθ pθ , θ = g, o, w, σ f = σo + σw + σg .
τi j = σi j + δi j σ f .
Next, following the ideas given in Chapter 2 for the case of two-phase fluids,
we proceed to derive the strain-stress relations for our system using the principle of
virtual complementary work [Fung, 1965] .
Let us consider a domain Ω of bulk material with boundary ∂ Ω , assumed to be
θ
initially in static equilibrium under the action of surface forces f i , θ = s, g, o, w,
θ
where f i represents the force in the θ −part of ∂ Ω per unit of surface area of bulk
material. Thus,
s θ
f i = σ i j ν j, f i = −φ Sθ pθ δi j ν j , θ = o, w, g,
where Sθ denotes the θ -fluid saturation in the initial equilibrium state. Now, con-
θ
sider a new system of surface forces fiθ superimposed on the original system f i
such that Ω remains in equilibrium under the action of the total surface forces
3 A poroelastic solid saturated by a three-phase fluid 57
θ
fiθ ,T = f i + fiθ , θ = s, o, w, g.
Since the fluids are at rest, all fluid pressures are constant on Ω and the total stress
field is in equilibrium. Hence,
∂ pθ ∂ τi j
∇pθ = = 0, θ = o, w, g, ∇·τ = = 0. (3.1)
∂ xi ∂xj
with the functions Pcow (So ) and Pcgo (Sg ) being strictly increasing functions of So
and Sg , respectively. Set
Next, note that it follows from (3.1) and (3.2) that in the initial equilibrium state
(at which Δ pg = Δ po = Δ pw = 0),
Pcgo (Sg ) ∇Sg = ∇pg − ∇po = 0, Pcow (So ) ∇So = ∇po − ∇pw = 0.
Since it is the case that Pcgo > 0 and Pcow > 0, we see that
∇Sg = 0, ∇So = 0.
uθ = φ (
uθ − us ), ξ θ = −∇ · uθ , θ = o, w, g, u = (us , uo , uw , ug ). (3.4)
For uniform porosity, Sθ ξ θ measures the amount of the θ -fluid entering or leaving
a unit cube of bulk material.
Next, computing the virtual forces δ fiθ , θ = s, o, w, g in terms of the virtual
stresses and using that Ω remains in equilibrium, from (3.3) we conclude that
δ W ∗ dx = εi j δ Δ τi j + (So ξ o − λow + λgo )δ Δ po + (Sw ξ w + λow )δ Δ pw
Ω Ω
+(Sg ξ g − λgo )δ Δ pg + (βow ξ o + βow
w
(ξ o − ξ w ) + λow )δ Δ Pcow (3.5)
+((βgo + βgo
ow
) ξ g + βgo
w
(ξ g − ξ w ) + λgo )δ
Δ Pcgo
+(Δ po − Δ pw − Δ Pcow )δ λow + (Δ pg − Δ po − Δ Pcgo )δ λgo ] d Ω
where
Pcow (So ) pw
βow = , βow
w
= ,
Pcow (So )
Pcow (So )
Pcgo (Sg ) Pcow (So ) pw
βgo = , βgo
ow
= , βgo
w
= .
Pcgo (Sg ) Pcgo (Sg )
Pcgo (So )
Since δ W ∗ must be an exact differential in the stresses(Δτ ij, Δ po, Δ pw, Δ pg , Δ Pcow , ΔPcgo),
it follows from (3.5) that (see [Santos and Savioli, 2015, Santos and Savioli, 2016])
= −F3 Δ τ + H5 Δ po + H6 Δ pw + H3 Δ pg .
Also, eliminating the Lagrange multipliers in (3.6) and using that in the linear
case W = W ∗ [Fung, 1965] we see that
1
W = (εi j Δ τi j + T o Δ po + T w Δ pw + T g Δ pg ) (3.11)
2
1
= ( Δ τ i j εi j + Δ F o ξ o + Δ F w ξ w + Δ F g ξ g ) ,
2
where
Next, consider a perturbation of system from the equilibrium state. The potential
energy V is
V = Vd d Ω = W dΩ − ( fis usi + fio uoi + fiw uwi + fig ugi )d(∂ Ω ),(3.16)
Ω Ω ∂Ω
where V and Vd denote the complementary energy and its density, respectively.
Since W = W ∗ , also V = V ∗ . Then transform the surface integral in (3.16) into a
volume integral and use the argument leading to (3.5) to see that
∂ ∂
δ Vd = −usi Δ τi j + ∑ uθi ΔFθ.
∂xj θ =o,w,g ∂ xi
60 3.3 The equations of motion. Low frequency range
Choosing usi , uoi , uwi , ugi as generalized coordinates and assuming that the system is
conservative we obtain the relations
∂ Vd ∂
=− Δ τi j ,
∂ ui
s ∂xj
∂ Vd ∂
θ
= Δ F θ , θ = o, w, g.
∂ ui ∂ xi
ρ = (1 − φ )ρs + φ ( ∑ ρθ Sθ )
θ =o,w,g
Sl St ρl ρt
glt = ε S
c
, (3.19)
φ
√
Sl St ηl ηt
blt = ε √ , (l,t) = (o, w), (o, g), (w, g). (3.20)
κ Krl Krt
ρo So üs + gco üo + gcow üw + gcog üg + bco u̇o + bcow u̇w (3.22)
+bcog u̇g + ∇Δ Fo (u) = f o ,
3 A poroelastic solid saturated by a three-phase fluid 61
ρw Sw üs + gcow üo + gcw üw + gcog üg + bcow u̇o + bcw u̇w (3.23)
+bcwg u̇g + ∇Δ Fw (u) = f w ,
ρg Sg üs + gcog üo + gcwg üw + gcg üg + bcog u̇o + bcwg u̇w (3.24)
+bcg u̇g + ∇Δ Fg (u) = f g ,
where f s , f o , f w and f g indicate external forces in the solid, oil, water and gas phases,
respectively.
As indicated for single phase fluids in Chapter 1 and for two-phase fluids in Chapter
2, the mass and viscous coupling coefficients need to be modified for frequencies
higher than the minimum of the following critical frequencies
1/2
2ηθ Sθ φ κ Krθ A0
ωθc = θ 2 , aθp =2 , θ = o, w, g, (3.25)
(a p ) ρθ φ Sθ
Sl St ρl ρt ηl ηt Sl St FIlt (ω ))
glt (ω ) = ε S + √ , (3.28)
φ κ Krl Krt ω
√
Sl St ηs ηt
blt (ω ) = ε √ FRlt (ω ), (l,t) = (o, w), (o, g), (w, g). (3.29)
κ Krl Krt
Using a straightforward generalization to the three-phase fluid case of the frequency
correction function defined in Chapter 1 in formula (1.94), the functions F θ (ω ) =
FRθ (ω ) + iFIθ (ω ) and F lt (ω ) = FRlt (ω ) + iFIlt (ω ) are given by
θ 4 i T 2 κ Krθ ηθ φ Sθ 8Sκ Krθ
F (ω ) = 1 − , xθ = , Λθ2 = ,
xθ Λθ2 φ ωκ Krθ ρθ φ Sθ
62 3.4 The equations of motion. High frequency range
Using (3.26)-(3.29) and the time Fourier transform to state the constitutive rela-
tions (3.12)-(3.15) in the space-frequency domain, the equations of motion (3.21)-
(3.24) become
−ω 2 ρ us (ω ) + ρo So uo (ω ) + ρw Sw uw (ω ) + ρg Sg ug (ω ) (3.30)
−∇ · Δ τ (u(ω )) = f (ω ),
s
−ω ρo So u (ω ) + go (ω ) u (ω ) + gow (ω ) uw (ω ) + gog (ω ) ug (ω ) (3.31)
2 s o
Next, following the arguments given in Chapters 1 and 2, assuming constant coeffi-
cients and in the absence of external sources from (3.30)-(3.33) we get
−ω 2 ρ us − ω 2 ρo So uo − ω 2 ρw Sw uw − ω 2 ρg Sg ug (3.34)
= (Eu ∇ es − μ ∇ × (∇ × us ) + B1 ∇eo + B2 ∇ew + B3 ∇eg
−ω 2 ρo So us − go uo − ω 2 gow uw − ω 2 gog ug + iω bo uo (3.35)
+ibow uw + iω bog ug
= B1 ∇ es + M1 ∇eo + M4 ∇ew + M5 ∇eg
−ω 2 ρw Sw us − ω 2 gow uo − ω 2 gw uw − ω 2 gwg ug + iω bow uo (3.36)
+iω bw uw + iω bwg ug
= B2 ∇ es + M4 ∇eo + M2 ∇ew + M6 ∇eg
3 A poroelastic solid saturated by a three-phase fluid 63
where
eθ = ∇ · uθ , θ = s, o, w, g,
and
Eu = λu + 2 μ .
To obtain the equations determining the propagation of compressional waves,
apply the divergence operator in (3.34)-(3.37) to obtain
−ω 2 ρ es − ω 2 ρo So eo − ω 2 ρw Sw ew − ω 2 ρg Sg eg (3.38)
= (Eu ∇ e + B1 ∇ e + B2 ∇ e + B3 ∇ e ,
2 s 2 o 2 w 2 g
−ω ρo So es − ω 2 go eo − ω 2 gow uw − ω 2 gog ug + iω bo eo
2
(3.39)
+iω bow ew + iω bog eg
= B1 Δ es + M1 ∇2 eo + M4 ∇2 ew + M5 ∇2 eg ,
−ω ρw Sw es − ω 2 gow eo − ω 2 gw ew − ω 2 gwg eg + iω bow eo
2
(3.40)
+iω bw ew + iω bwg eg
= B2 ∇2 es + M4 ∇2 eo + M2 ∇2 ew + M6 ∇2 eg ,
−ω 2 ρg Sg es − ω 2 gog eo − ω 2 gwg ew − ω 2 gg eg + iω bog eo (3.41)
+iω bwg ew + iω bg eg
= B3 ∇2 es + M5 ∇2 eo + M6 ∇2 ew + M3 ∇2 eg .
γ 2 A C (γ ) = E C (γ ) ,
where
64 3.4 The equations of motion. High frequency range
⎛ ⎞ ⎛ ⎞ ⎛ γ⎞
ρ ρo S o ρw S w S g ρg Eu B1 B2 B3 Cs
⎜ ρo S o gw gow gog ⎟ ⎜ B1 M1 M4 M5 ⎟ ⎜ Coγ ⎟
A =⎜
⎝ ρw S w
⎟, E =⎜ ⎟, Cγ = ⎜ ⎟
⎝ Cwγ ⎠ ,
gow gw gwg ⎠ ⎝ B2 M4 M2 M6 ⎠
γ
ρg S g gog gwg gg B3 M5 M6 M3 Cg
and
bo bw bg
go = go + i , gw = gw + i , gg = gg + i ,
ω ω ω
bow bog bwg
gow = gow + i gog = gog + i gwg = gwg + i .
ω ω ω
Hence, to determine the complex wave-numbers = r + ii it is sufficient to
solve the eigenvalue problem
det(S − γ 2 I) = 0, (3.44)
where
S = A −1 E .
The P1 wave is the analogue of the classical P1 wave in Biot theory. The P2, P3
and P4 waves are slow waves associated with the motion out of phase of the four
phases. The behavour of all four compressional waves is analyzed numerically in
Subsection 3.5.1.
To determine the phase velocities and attenuation coefficients of the rotational
waves we apply the curl operator in (3.34)-(3.37) and set
κ θ = ∇ × uθ , θ = s, o, w, g,
to obtain
−ω 2 ρκ s − ω 2 ρo So κ o − ω 2 ρw Sw κ w − ω 2 ρg Sg κ g = μ ∇2 κ s (3.45)
−ω ρo So κ − ω go κ − ω gow κ − ω gog κ + iω bo κ
2 s 2 o 2 w 2 g o
(3.46)
+iω bow κ + iω bog κ = 0,
w g
Ao = Co As , Aw = Cw As , Ag = Cg As , (3.53)
where the coefficients Co ,Cw ,Cg depend on the mass and viscous coupling coeffi-
cients in (3.50)-(3.52). Next, using (3.49) in (3.45) and using (3.53) and (3.43) we
get the equation
1
ρ + ρo SoCo + ρw SwCw + ρg SgCg = μ. (3.54)
γ2
The physically meaningful solution of (3.54) gives the shear phase velocity and
attenuation coefficient as
ω |i |
vs = bs = 2π .8.655588 .
|r | |r |
In this section we compute phase velocities and attenuation coefficients for a sample
of Nivelsteiner sandstone saturated by a three phase fluid, taken to be oil, water and
gas, water assumed to be the wetting phase. The material properties of the Nivel-
steiner sandstone are given in Table 1.1.
The gas properties correspond to a dry gas at a reference pressure of 5MPa, (at a
depth of 500 m, approximately ).
66 3.5 Numerical Examples
The capillary pressure functions Pcow (So ) and Pcgo (Sg ) used to determine the
elastic coefficients are [Ravazzoli et al., 2003]
Pcow (So ) = Aow 1/(So + Srw − 1)2 − Sro
2
/[So (1 − Sro − Srw )]2 ,
Sro ≤ So ≤ 1 − Srw − Srg ,
Pcgo (Sg ) = Aog 1/(Sg + Sro − 1) − Srg
2 2
/[Sg (1 − Srg − Sro )]2 ,
Srg ≤ Sg ≤ 1 − Sro − Srw .
In the examples we use Srw = Sro = Srg = 0.02, Aow = 40 kPa, Aog = 30 kPa.
Besides, the relative permeability functions krθ (Sθ ), θ = o, w, g, are [Baker, 1988]
Sθ − Srθ 2
Krθ (Sθ ) = ,
1 − Srθ
Srθ ≤ Sθ ≤ 1 − Srp − Srq , p = θ , q = θ , p = q.
We used ε = 0.01 in the definitions of the cross mass and viscous coupling coef-
ficients in (3.19)-(3.20) and (3.28)-(3.29).
γ 2 A C() = E C(γ ) ,
γ γ γ γ t
where γ is defined in (3.43), Cγ = Cs , Co , Cw , Cg and
3 A poroelastic solid saturated by a three-phase fluid 67
⎛ ⎞
ρ − φ ( ρo S o + ρw S w + ρg S g ) φ ρo S o φ ρw S w φ ρg S g
⎜ ρo So − φ (gco + gcow + gcog ) φ gco φ gcow φ gcog ⎟
A =⎜
⎝ ρw Sw − φ (gcw + gcow + gcwg ) φ gcow φ gcw gcwg ⎠ ,
⎟
B3 − φ (M3 + M5 + M6 ) φ M5 φ M6 φ M3
⎡ ⎤ ⎡ ⎤
v(1) v(2) v(3) v(4) v(1) v(2) v(3) v(4)
⎢s + − + − ⎥ ⎢s + − − − ⎥
⎢ ⎥ ⎢ ⎥
C3 = ⎢
⎢o + + − − ⎥ ⎥, C4 = ⎢
⎢o + + + − ⎥ ⎥ (3.56)
⎣w + + − + ⎦ ⎣w + + + + ⎦
g + + + − g + + − −
⎡ ⎤ ⎡ ⎤
v(1) v(2) v(3) v(4) v(1) v(2) v(3) v(4)
⎢s + − − + ⎥ ⎢s + − + + ⎥
⎢ ⎥ ⎢ ⎥
C5 = ⎢
⎢o + + + + ⎥ ⎥, C6 = ⎢
⎢o + + − + ⎥ ⎥ (3.57)
⎣w + + + − ⎦ ⎣w + + − − ⎦
g + + − + g + + + +
Next let us analyze the behaviour of all waves for the non-dissipative case (zero vis-
cosity) and several values of the reference pressure pw . The corresponding gas prop-
erties for the different pressures, computed using the calculations in [Standing, 1977]
and [McCoy, 1983] are given in Table 3.1.
Figures 3.1 and 3.2 show phase velocities of all waves for the zero viscosity
case as function of gas saturation at water saturation Sw = 0.25 and reference water
pressures pw equal to 20 and 30 MPa.
2590 1550
Type
P1, I, pw=20 MPa, Sw=0.25
Type
P1, I, pw=30 MPa, Sw=0.25 1540
2580 S-wave, pw=20 MPa, Sw=0.25
P-Wave Phase Velocity (m/s)
1520
2560
Type IP1
1510
2550
1500
2540 1490
0 0,1 0,2 0,3 0,4 0,5 0,6
Gas Saturation
Fig. 3.1 P1 and shear wave phase velocities as function of gas saturation at water saturation Sw =
0.25 and reference water pressures pw = 20 and 30.
Figure 3.1 shows that P1 waves have higher velocities for higher values of pw ,
displaying a characteristic decreasing behaviour as gas saturation increases until a
threshold value (which depends on pw ) at which velocities start to increase. This
behaviour is similar to that observed in Figure 2.1 for the gas-water case.
Besides, phase velocities of S-waves are increasing functions of gas saturation,
and show lower values for higher values of pw . Thus these fast P and S waves ve-
locities mimic the behaviour of the corresponding Gassmann velocities.
Figure 3.2 shows a decrease of the P2 and P3 waves as gas saturation increases,
behaving as the P2 wave for the gas-water case in Figure 2.2. Besides, they exhibit
higher values for increasing values of pw . Also, P4 waves phase velocities are almost
independent of gas saturation and pw .
Next we analyze the behaviour of all waves as function of frequency fixing the
saturation of oil, water and gas as indicated in the different figures.
70 3.5 Numerical Examples
500
Type
P2, II, pw=20 MPa, Sw=0.25
Type
P2, II, pw=30 MPa, Sw=0.25
Type
P3, III, pw=20 MPa, Sw=0.25
400
Type
P3, III, pw=30 MPa, Sw=0.25
Type
P4, IV, pw=20 MPa, Sw=0.25
Type
P4, IV, pw=30 MPa, Sw=0.25
Phase Velocity (m/s)
300
200
100
0
0 0.1 0.2 0.3 0.4 0.5 0.6
Gas Saturation
Fig. 3.2 P2, P3 and P4 wave phase velocities as function of gas saturation at water saturation
Sw = 0.25 and reference water pressures pw = 20 and 30.
In Figures 3.3 and 3.4 can be seen that P1 and shear wave phase velocities suf-
fer little dispersion in the whole range of frequencies, show constant values up to
about 1 kHz, where they start to increase, reaching constant values. Also, velocities
increase as oil saturation increases.
Phase velocitiy of P2, P3 and P4 waves in Figures 3.5, 3.6 and 3.7 vanish at low
frequencies and exhibit increasing behaviour until attaining constant values at ultra-
sonic frequencies. P2 wave velocities are increasing function of the oil saturation, is
the fastest of the three slow waves and the P4 wave is the slowest.
Figures 3.8 and 3.9 show a similar behaviour for attenuation of P1 and shear
waves, with negligible attenuation at low frequencies and attenuation peaks moving
to high frequencies with increasing oil saturation. Attenuation of shear waves is
slightly higher than for P1 waves.
Attenuation of P2, P3 and P4 waves shown in Figures 3.10, 3.11 and 3.12 display
the characteristic high attenuation at low frequencies and continuous decreasing be-
haviour above 100 kHz until reaching vanishing attenuation in the ultrasonic range.
The P4 wave is the one suffering the highest attenuation. P2 and P4 waves atten-
uation is seen to be almost independent of oil saturation, and P3 waves suffer the
strongest attenuation at the highest value of oil saturation.
3 A poroelastic solid saturated by a three-phase fluid 71
2550
Sw = 0.1, So = 0.8
2525 Sw = 0.4, So = 0.5
Sw = 0.8, So = 0.1
P1 Wave Phase Velocity (m/s)
2500
P1 waves as function of
frequency for a sample of 2450
Nivelsteiner sandstone with
water saturation Sw = 0.1, oil
saturation So = 0.8 and gas 2425
-1 0 1 2 3 4 5 6 7
saturation Sg = 0.1 Frequency (Hz) - Logarithmic Scale
1550
Sw = 0.1_So = 0.8
Sw = 0.4, So = 0.5
Shear Wave Phase Velocity (m/s)
Sw = 0.8, So = 0.1
1525
250
Sw = 0.1, So = 0.8
150
P2 waves as function of
frequency for a sample of 50
Nivelsteiner sandstone with
water saturation Sw = 0.1, oil 0
saturation So = 0.8 and gas -1 0 1 2 3 4 5 6 7
saturation Sg = 0.1 Frequency (Hz) - Logarithmic Scale
150
Sw = 0.1, So = 0.8
P3 Waves Phase Velocity (m/s)
Sw = 0.4, So = 0.5
Sw = 0.8, So = 0.1
100
50
40 Sw = 0.1, So = 0.8
P4 Waves Phase Velocity (m/s)
Sw = 0.4, So = 0.5
Sw = 0.8, So = 0.1
30
20
Fig. 3.7 Phase velocity of
P4 waves as function of
10
frequency for a sample of
Nivelsteiner sandstone with
water saturation Sw = 0.1, oil 0
0.5
0.3
0.2
Fig. 3.8 Attenuation of P1
waves as function of fre-
0.1
quency for a sample of Nivel-
steiner sandstone with water
saturation Sw = 0.1, oil sat- 0
0.5
Sw = 0.4, So = 0.5
Sw = 0.8, So = 0.1
0.3
0.2
Fig. 3.9 Attenuation of shear
waves as function of fre-
0.1
quency for a sample of Nivel-
steiner sandstone with water
saturation Sw = 0.1, oil sat- 0
Since the shear modulus μ is assumed to be known and equal to the shear modulus
of the dry matrix, as explained in Chapter 1, to determine the coefficients in (3.12)-
(3.15) it is sufficient to consider tensional changes Δ τi j of the form
1
Δ τ11 = Δ τ22 = Δ τ33 = Δ τ = −Δ p, Δ p > 0, Δ τi j = 0, i = j.
3
For these tensional changes, from (3.12) we get the relation
−Δ p = Ku es − Bo ξ o − Bw ξ w − Bg ξ g , (3.58)
74 3.6 Appendix 1. Determination of the elastic coefficients. Inclusion of linear viscoelasticity
60
50 Sw = 0.1, So = 0.8
Sw = 0.4, So = 0.5
Sw = 0.8, So = 0.1
30
P2 waves as function of
frequency for a sample of 10
60
Sw = 0.1, So = 0.8
50 Sw = 0.4, So = 0.5
Sw = 0.8, So = 0.1
P3 Waves attenuation (dB)
40
30
P3 waves as function of
frequency for a sample of 10
with Ku = λu + 23 μ .
Let Kθ and Cθ = 1/Kθ denote the bulk modulus and compressibility of the θ -
fluid, respectively, θ = o, w, g, u.
Considering the gedanken closed compressibility test as in Chapter 1, since for
the closed system there is conservation of mass of the fluid phases, ξ θ = 0, θ =
o, w, g, and from (3.58) we see that Ku is the Gassmann undrained bulk modulus of
our closed system, which is given by the relation: (see [Santos and Savioli, 2015])
Km + Ξ K ∗f (Ks − Km )
Ku = Ks , Ξ= , (3.59)
Ks + Ξ φ (Ks − K ∗f )
where
3 A poroelastic solid saturated by a three-phase fluid 75
60
50
Sw = 0.1, So = 0.8
Sw = 0.4, So = 0.5
30
P4 waves as function of
frequency for a sample of 10
250
200 pw = 5 MPa
P2 Wave Phase Velocity (m/s)
pw = 10 MPa
pw = 20 MPa
150
C∗f = 1/K ∗f = γow So Co + Sw Cw + γgw Sg Cu /α
, (3.60)
= So + βow − βgo − βgo γow + Sw − βow + Sg + βgo + βgo γgw ,
α ow ow
300
250
pw = 5 MPa
150
Fig. 3.14 Phase velocity
of P3 waves as function of 100
frequency for a sample of
Nivelsteiner sandstone at 50
pw = 5, 10 and 20 MPa with
water saturation Sw = 0.4, oil 0
saturation So = 0.5 and gas -1 0 1 2 3 4 5 6 7
saturation Sg = 0.1 Frequency (Hz) - Logarithmic Scale
80
pw = 5 MPa
pw = 10 MPa
P4 Wave Phase Velocity (m/s)
pw = 20 MPa
60
In (3.60) the expression for the undrained modulus Ku in (3.59) is formally iden-
tical to that for the single phase and the two-phase immiscible cases as given in
Chapters 1 and 2. The remaining coefficients can be determined from the following
relations [Santos and Savioli, 2015]
B1 = KuΘ So + βow + βow
w
γow − (βow + βow
w
) , (3.61)
w
B2 = KuΘ Sw + βow + γow βgo − βow − βgo γgw ,
w w w
(3.62)
B3 = KuΘ γgw Sg + βgo + βgo + βgo − γow βgo + βgo + βgo . (3.63)
w ow w ow
4.1 Introduction
are associated with propagation modes in the solid phase accompanying matrix with
infinite permeability in the rock-frame or zero viscosity in the fluid phase.
The verification that additional (slow) waves can be observed in laboratory ex-
periments was published in [Leclaire et al., 1985]. Later, this theory was extended
in [Carcione and Tinivella, 2000a] to include the interaction between the solid and
ice particles and grain cementation with decreasing temperature.
This Chapter presents a generalization to the case of non-uniform porosity of the
theory in [Leclaire et al., 1994, Carcione and Tinivella, 2000a]. This generalization,
appeared in [Santos et al., 2004a], can be used to perform numerical experiments or
fit laboratory data related to heterogeneous media.
As an application, we mention the work by [Rubino et al., 2008], where this
model was used to analyze conversions to Biot-type slow waves in highly heteroge-
neous gas-hydrate bearing sediments.
The non-welding condition between the two solid phases is assumed when defin-
ing the potential and kinetic energies, with proper interaction terms among the solid
and fluid phases. If the two solid phases would be welded, then additional slow
waves would not be present [Brown and Korriga, 1975].
The derivation is based on the energy formulation presented in Chapter 1, rather
than on volume averaging or homogenization methods. A formal derivation of this
model using homogenization techniques appeared in [Santos et al., 2005b].
Let u(1) , u(2) and u(3) be the averaged solid and fluid displacements over the bulk
material. Here u(2) is defined such that on any face F of the cube Ω
φ u(2) · ν dF
F
represent the displacements in the two solid parts of F, respectively. Here ν denotes
the unit outward normal to F,
(1) (3)
Let σi j and σi j denote the stress tensors in Ω1 and Ω3 averaged over the bulk
material Ω , respectively, and let p f denote the fluid pressure. These quantities de-
(1) (3)
scribe small changes with respect to reference values σ i j , σ i j and p f correspond-
ing to an initial equilibrium state. Let us also introduce the tensors
(1,T ) (1) (3,T ) (3)
σi j = σi j − S1 φ p f δi j , σi j = σi j − S3 φ p f δi j , (4.1)
associated with the total stresses in Ω1 and Ω3 , respectively, so that the total stress
tensor in the bulk material Ω is given by
(1,T ) (3,T )
σi j = σi j + σi j .
Next we shall derive the stress-strain relations for our system using the virtual
work principle, following the derivation for the case of a single solid phase given in
Chapter 1.
Let W be the strain energy density and V the total potential energy. Also let
(1) (3) (2)
Vd denote the total potential energy density. Then, if fi , fi , fi represent the
surface forces acting on the solid and fluid parts of the boundary of Ω , denoted by
∂ Ω , we have that
(1) (1) (2) (2) (3) (3)
V = Vd d Ω = W dΩ − ( fi ui + fi ui + fi ui )d(∂ Ω ),
Ω Ω ∂Ω
and the virtual work principle for the composite fluid-solid system can be stated in
the form:
(1) (1) (2) (2) (3) (3)
δV = 0 = δ W dΩ − ( fi δ (ui ) + fi δ (ui ) + fi δ ui )d(∂ Ω ), (4.2)
Ω ∂Ω
where
(1) (1) (3) (3) (2)
fi = σi j ν j , fi = σi j ν j , fi = − φ p f δi j ν j , (4.3)
and δ denotes virtual changes in the different quantities. Using (4.3) in (4.2) we get
(1,T ) (1) (3,T ) (3)
δV = 0 = δ W dΩ − σi j ν j δ (ui ) + σi j ν j δ (ui ) (4.4)
Ω ∂Ω
−p f δi j ν j δ wi d(∂ Ω ).
where
(2) (1) (3)
wi = φ ui − S1 ui − S3 ui . (4.5)
Then, transforming the surface integral in (4.4) into a volume integral we get
82 4.2 The strain energy of the composite system
∂ # (1,T ) (1) (3,T ) (3)
δV = 0 = δ W dΩ − σ δ (ui ) + σi j δ (ui )
Ω Ω ∂xj ij
−p f δi j δ wi d Ω .
where
ζ = −∇ · w
represents the change in fluid content and εi j (u(m) ), m = 1, 3 is the strain tensor in
Ωm with linear invariant e(m) = εii (u(m) ).
Thus from (4.1) we finally obtain the following expression for the variation in
strain energy density δ W :
(1) (3)
δ W = (σi j − S1 φ p f δi j )δ (εi j (u(1) )) + (σi j − S3 φ p f δi j )δ (εi j (u(3) )) (4.6)
+p f δ ζ
(1,T ) (3,T )
= σi j δ (εi j (u(1) )) + σi j δ (εi j (u(3) )) + p f δ ζ .
∂W (1,T ) (1)
= σi j = σi j − S1 φ p f δi j ,
∂ εi j (u(1) )
∂W (3,T ) (3)
(3)
= σi j = σi j − S3 φ p f δi j ,
∂ εi j (u )
∂W
= pf .
∂ζ
Also, it follows from (4.6) that
1 # (1) (3)
W = (σi j − S1 φ p f δi j )(εi j (u(1) )) + (σi j − S3 φ p f δi j )(εi j (u(3) )) (4.7)
2
+p f ζ .
4 Waves in a fluid-saturated poroelastic matrix composed of two weakly coupled solids 83
(m) 1 (m)
di j = εi j (u(m) ) − e δi j , m = 1, 3.
3
In the linear isotropic case the strain energy density W in (4.7) is a quadratic positive
(1) (1) (3) (3)
definite form in the invariants e(1) , e(3) , ζ , (d1 )2 = di j di j , (d3 )2 = di j di j and
(1) (3)
d1,3 = di j di j . Note that
1 1
(d1 )2 = εi j (u(1) ) εi j (u(1) ) − (e(1) )2 , (d3 )2 = εi j (u(3) ) εi j (u(3) ) − (e(3) )2 ,
3 3
and
1
d1,3 = εi j (u(1) ) εi j (u(3) ) − e(1) e(3) .
3
Then,
1 1
W = E1 (e(1) )2 + μ1 (d1 )2 + E3 (e(3) )2 + μ3 (d3 )2
2 2
1
−B1 e(1) ζ − B2 e(3) ζ + B3 e(1) e(3) + M(ζ )2 + μ1,3 d1,3 .
2
The B3 and μ13 terms represent elastic interaction between the two solid phases.
Thus, we obtain
∂W (1,T ) (1)
= σi j = σi j − S1 φ p f δi j (4.8)
∂ εi j (u(1) )
(1) (3)
= E1 e(1) − B1 ζ + B3 e(3) δi j + 2μ1 di j + μ1,3 di j ,
∂W (3,T ) (3)
= σi j = σi j − S3 φ p f δi j (4.9)
∂ εi j (u(3) )
(3) (1)
= E3 e(3) − B2 ζ + B3 e(1) δi j + 2μ3 di j + μ1,3 di j ,
∂W
= p f = −B1 e(1) − B2 e(3) + M ζ . (4.10)
∂ζ
(1,T ) (1)
Equations (4.8)-(4.9)-(4.10) express the generalized stresses σi j = σi j −S1φ p f δi j,
(3,T ) (3)
σi j = σi j − S3 φ p f δi j and p f in terms of the strains εi j (u(1) ), εi j (u(3) ) and ζ .
84 4.4 Determination of the coefficients in the constitutive relations
θ2 = ∇ · u(2) , θ3 = ∇ · u(3)
ζ = φ (S1 θ1 + S3 θ3 − θ2 ) (4.11)
&
(3) (1)
−(φ S3 M − φ B2 )θ2 δi j + 2μ3 di j + μ1,3 di j .
2
Set
K2 = M φ 2 ,
C12 = B1 φ − M φ 2 S1 ,
C23 = B2 φ − M φ 2 S3 ,
K1 = E1 + (S1 φ )2 M − 2S1 φ B1 , (4.15)
K3 = E3 + (S3 φ )2 M − 2S3 φ B2 ,
C13 = S1 S3 φ 2 M − S3 φ B1 − S1 φ B2 + B3 .
4 Waves in a fluid-saturated poroelastic matrix composed of two weakly coupled solids 85
Then, for constant porosity, the stress-strain relations (4.12)-(4.14) can be stated
as follows:
(1) (1) (3)
σi j = (K1 θ1 +C13 θ3 +C12 θ2 ) δi j + 2μ1 di j + μ1,3 di j , (4.16)
(3) (3) (1)
σi j = (K3 θ3 +C13 θ1 +C23 θ2 ) δi j + 2μ3 di j + μ1,3 di j , (4.17)
−φ p f = C12 θ1 +C23 θ3 + K2 θ2 . (4.18)
Relations (4.16)-(4.18) were derived in [Leclaire et al., 1994] for the case of uni-
form porosity and when the solid phase 3 is ice; it is also assumed in [Leclaire et al.,
1994] that there is no contact between the solid and ice phases. This situation corres-
ponds to the particular case in which C13 = μ1,3 = 0.
The nonsingular linear system of equations (4.15) relating the coefficients for
both formulations can always be solved to obtain expressions for the coefficients
of the variable porosity formulation in terms of the corresponding to the uniform
porosity case, with the following expressions:
E1 = K1 + (S1 )2 K2 + 2S1C12 ,
E3 = K3 + (S3 )2 K2 + 2S3C23 ,
S1 K2 +C12
B1 = , (4.19)
φ
S3 K2 +C23
B2 = ,
φ
B3 = C13 + S3C12 + S1C23 + S3 S1 K2 ,
K2
M = 2.
φ
In Appendix 4.8 we give the formulas for the computation of the coefficients
K1 , K2 , K3 , C12 , C13 , C23 , μ1 ,μ3 ,μ1,3 , which combined with (4.19) allows for the
evaluation of the moduli H1 , H3 , B1 , B2 , B3 , M, needed for this variable porosity
formulation.
The derivation of the kinetic energy and dissipation function densities is given in
[Santos et al., 2004a], leading to the following formulation of the equations of mo-
tion:
The coefficients mi j in (4.23) can be written in terms of the tortuosities a13 , a31 , a12 ,
and a32 defined in Refs. [Carcione and Tinivella, 2000a] and [Leclaire et al., 1994]
as follows:
ρ1 ρ3
n(1,3) = a13 , n(3,1) = a31 , (4.24)
φ1 φ3
(1,2) ρ2 (3,2) ρ2
m = a12 , m = a32 ,
φ φ
4 Waves in a fluid-saturated poroelastic matrix composed of two weakly coupled solids 87
φ1 ρ φ3 ρ
a12 = r12 + 1, a32 = r32 + 1,
φ ρ2 φ ρ2
φ3 ρ φ1 ρ
a13 = r13 + 1, a31 = r31 + 1,
φ1 ρ1 φ3 ρ3
where ri j are the geometrical aspects of the boundaries separating the phases i and
j (equal to 21 for spheres) and
φ ρ2 + φ3 ρ3 φ ρ2 + φ1 ρ1
ρ= , ρ = .
φ + φ3 φ + φ1
Using the relations (4.24) in (4.23) we get
m11 = ρ2 φ 1 + (S1 )2 a32 + (S3 )2 a12 − 2S3 − (S1 )2 + a13 ρ1 φ1 + (a31 − 1)ρ3 φ3 ,
m12 = ρ2 (1 + S1 (a32 − 1) − S3 a12 ) ,
m13 = ρ2 φ 1 − (S1 )2 a32 − (S3 )2 a12 − S1 S3 + ρ1 φ1 (1 − a13 ) + ρ3 φ3 (1 − a31 ),
ρ2
m22 = (a12 + a32 − 1) ,
φ
m23 = ρ2 (1 + S3 (a12 − 1) − S1 a32 ) ,
m33 = ρ2 φ 1 + (S1 )2 a32 + (S3 )2 a12 − 2S1 − (S3 )2 + a31 ρ3 φ3 + (a13 − 1)ρ1 φ1 .
The expressions for the dissipation coefficients f11 , f22 and f12 in (4.20)–(4.22)
are determined as follows.
For the case of frozen porous media, following [Carcione and Tinivella, 2000a],
we define the dissipation coefficients b12 , b23 and b13 as
η η
b12 = (φ )2 , b23 = (φ )2 ,
κ1 κ3
while b13 is a friction coefficient between the ice and the solid frames. Also, η
denotes the fluid viscosity and the permeability coefficients κ1 , κ3 are defined in
terms of the absolute permeabilities κ1,0 , κ3,0 of the two solid frames by the relations
(see also [Leclaire et al., 1994])
3
(φ )3 (1 − φ1 )2 φ
κ1 = κ1,0 , κ3 = κ3,0 . (4.25)
(1 − φ1 )3 φ32 φ1
For the case of shaley sandstones, following [Carcione et al., 2000] the coeffi-
cient b13 can be assumed to be zero and the friction coefficients b12 and b23 are
taken to be of the form:
where Rs1 , Rs3 denote the average radii of the sand and clay particles, respectively.
Next, the frequency independent friction coefficients for this model are given by
88 4.5 The equations of motion
In the high frequency range the set of inertial and friction coefficients are modified as
follows. Let F(ω ) = FR (ω ) + iFI (ω ), be the frequency correction function defined
in either (1.93) or (1.94) where the permeability is given by
1 1 1
= + .
κ κ1 κ3
Then we define the frequency dependent mass and viscous coupling coefficients in
the following fashion
FI (ω )c11
m11 (ω ) = m11 + , (4.28)
ω
FI (ω ) f12
m12 (ω ) = m12 − ,
ω
FI (ω )c11
m13 (ω ) = m13 − ,
ω
FI (ω ) f22
m22 (ω ) = m22 + ,
ω
FI (ω ) f12
m23 (ω ) = m23 + ,
ω
FI (ω )c11
m33 (ω ) = m33 + ,
ω
f11 (ω ) = FR (ω ) f11 + b13 , f12 (ω ) = FR (ω ) f12 ,
f22 (ω ) = FR (ω ) f22.
The coefficients d13 is left as a free parameter chosen so that the following condition
is satisfied
Since the coefficient b13 takes into account friction between the two solid phases,
a proper model based for example in Coulomb’ s friction theory may be used. For
simplicity in all the numerical examples we have set the coefficient b13 to be zero.
4 Waves in a fluid-saturated poroelastic matrix composed of two weakly coupled solids 89
To determine the different types of waves that can propagate in this type of media,
we use the argument given in Chapter 1 assuming plane waves of angular frequency
ω and wave number = r + i i travelling in the x1 -direction and satisfying the
equations of motion (4.20), (4.21) and (4.22).
The argument, (using the frequency dependent definition of the viscodynamic co-
efficients in (4.28) in the high frequency range) leads to the solution of the following
eigenvalue problem
det(S − γ 2 I) = 0, (4.30)
where
ω
γ= ,
S = A −1 E . (4.31)
and ⎡ ⎤
E1 + 43 μ1 0 0 B1 B3 + 23 μ13 0 0
⎢ μ ⎥
2 μ13 0 ⎥
1
⎢ 0 1 0 0 0
⎢ 0 0 μ1 0 0 0 2 μ13 ⎥
1
⎢ ⎥
E =⎢
⎢ B1 0 0 M B2 0 0 ⎥⎥,
⎢ B3 + 2 μ13 0 0 B2 E3 + 43 μ3 0 0 ⎥
⎢ 3 ⎥
⎣ 0 1
μ 13 0 0 0 μ3 0 ⎦
2
0 0 12 μ13 0 0 0 μ3
where
f11 f12 f11
11 = m11 − i
m , m 12 = m12 + i , , m 13 = m13 + i ,
ω ω ω
f22 f12 f11
22 = m22 − i , m
m 23 = m23 − i , m 33 = m33 − i ,
ω ω ω
12 )2
(m 12 m
m 23 m2
q1 = m 11 − , q2 = m 13 − , q3 = m 33 − 23 .
22
m 22
m m 22
90 4.7 Numerical Examples. Shaley sandstones
After solving (4.30) for the complex eigenvalues (γ j )2 , j = 1, · · · , 7 the phase veloc-
ities v( j) and the attenuation coefficients b( j) (in dB) of the compressional and shear
waves are obtained in terms of γ j from the relations
( j)
ω |i |
v( j) = ( j)
, b( j) = 2π · 8.685889 (s)
, j = 1, · · · , 7. (4.32)
|r | | j |
Of the seven eigenvalues of the matrix S , three of them, associated with the
first,fourth and fifth row and columns of S are related with the compressional
modes, while of the other four eigenvalues, associated with the second, third, sixth
and seventh rows and columns of S , only two of them are different and are related
with the two shear modes of propagation.
The slow wave modes are important to explain attenuation and dispersion effects
observed on the faster modes associated with scattering phenomena due to the pres-
ence of heterogeneities inside the composite poroelastic materials being analyzed.
Let us consider a shaley sandstone, and denote the sand fraction by S1 and the clay
fraction by S3 . As stated previously, the theory predicts three compressional waves
(P1, P2 and P3 waves) and two shear waves (S1 and S2 waves), whose phase veloc-
ities can be obtained by solving the eigensystem resulting from the equation of mo-
tion as indicated in Section 4.6 (cf. (4.30)-(4.32)); see also [Carcione et al., 2000].
In this example, we consider the faster P and S waves.
The bulk and shear moduli of the sand and clay (dry) matrices versus porosity
φ are obtained from a relationship proposed by [Krief et al., 1990] using formulas
(4.35)–(4.36) in Appendix 4.8.3
We consider the data set published by [Han et al., 1986]. obtained at a confining
pressure of 40 MPa. Han and his co-workers provide ultrasonic measurements of P-
and S-wave velocities for 75 sandstone samples with porosities ranging from 2 %
to 30 % and clay content from 0 to 50 %. One feature of this data set is that a small
amount of clay significantly softens the rock moduli, leading to reduced velocities.
Table I shows the properties of the different constituents.
The permeabilities and friction coefficients are calculated by using the equations
given in (4.25), (4.26) and (4.27)
Moreover, the geometrical aspects ri j of the boundaries separating the phases i
and j needed in (4.24) to determine de the mass coefficients pi j in (4.23) were taken
to be r12 = r32 = r13 = r31 = 1/2.
The predictions of the theory against the measurements obtained by [Han et al.,
1986] are shown in Figure 4.1.
We used the values A = 2 and a = 0.5 to determine the moduli Ks1,m , Ks3,m and
μs1,m in (4.35) and (4.36).
4 Waves in a fluid-saturated poroelastic matrix composed of two weakly coupled solids 91
A frequency of 5 kHz was assumed to fit the experimental data, without including
viscoelasticity. Strictly speaking, this is not correct since the data has been acquired
at ultrasonic frequencies of the order of hundreds of kilohertz, and Biot-type dis-
sipation mechanisms alone do not account for the level of attenuation observed in
rocks.
The figure shows the compressional and shear velocities versus porosity, where
each curve corresponds to a different value of the clay content S3 . The root-mean-
square deviation computed for all samples, apart from five outliers for P waves and
seven outliers for S waves, is 93 m/s for the P-wave velocity and 100 m/s for the
S-wave velocity. This results reflects the robustness of the model for this particular
example.
Fig. 4.1 Velocities of the faster compressional (a) and shear (b) waves versus porosity φ for differ-
ent values of clay content S3 , indicated by the numbers inside the boxes (1: S3 = 0%, 2: S3 = 10%,
3: S3 = 20%, 4: S3 = 30% and 5: S3 = 40%). The experimental data, represented with numbers,
correspond to the data set published by Han et al.[Han et al., 1986]. In this case, 1, 2, 3, 4 and 5
correspond to S3 values in the ranges [S3 , S3 + 5%], S3 = 0, ..., 40 %. The frequency is 5 kHz.
4 Waves in a fluid-saturated poroelastic matrix composed of two weakly coupled solids 93
The remaining elastic coefficients are given by the following expressions [Carcione
and Tinivella, 2000a]
K1 = [(1 − c1 )φ1 ]2 M + Ks1,m , K3 = [(1 − c3 )φ3 ]2 M + Ks3,m ,
where
indicate the procedure used in this paper to determine those moduli for the cases of
shaley sandstones and frozen porous media presented in this paper.
The porosity dependence of the sand and clay (dry) matrices is consistent with the
concept of critical porosity, since the moduli should vanish above a certain value
of the porosity (usually from 0.4 to 0.5). This dependence is determined by the
empirical coefficient A in equation (4.35).
In some rocks there is an abrupt change of rock matrix properties with the ad-
dition of a small amount of clay, attributed to softening of cements, clay swelling
and surface effects. That is, the wave velocities decrease significantly when the clay
content increases from zero to a few percentages. In order to model this effect, we
multiply the shear modulus of the sand matrix by a factor depending on the empiri-
cal coefficient a in equation (4.36) (this factor tends to 1 when a → ∞).
If Ω1 represents the sand and Ω3 the clay minerals, then the bulk and shear
moduli of the sand and clay (dry) matrices are assumed to satisfy
Following [Leclaire et al., 1994] and [Carcione and Tinivella, 2000a] we assume
that Ks1,m is known, and that the other modulus may be computed using a percolation-
type model with critical exponent 3.8 [Deptuck et al., 1985]. The percolation theory
is used here to describe the transition of a system from the continuous to the discon-
tinuous state, which is governed by a power law independent of the system material.
Hence, we obtain μs1,m , μs3,m and Ks3,m using the relations
(max) φ3 3.8
μs1,m = μs1,m − μs1,m
0
] + μs1,m
0
, (4.37)
1 − φ1
(max) φ3 3.8
μs3,m = μs3,m − μs3,m
0
] + μs3,m
0
,
1 − φ1
(max) φ3 3.8
Ks3,m = Ks3,m − Ks3,m
0
] + Ks3,m
0
,
1 − φ1
4 Waves in a fluid-saturated poroelastic matrix composed of two weakly coupled solids 95
The theory in [Kuster and Toksöz, 1974] is a widely accepted model for the estima-
tion of the effective elastic properties of an infinite solid with randomly distributed
inclusions. Let c denote the bulk volume concentration of inclusions with elastic
coefficients K and μ and K, μ the elastic moduli of the background medium. Then,
for the case of spherical inclusions, the effective elastic moduli K ∗ and μ ∗ take the
form:
K∗ − K K −K
=c ,
3K ∗ + 4μ 3K + 4μ
μ∗ − μ μ −μ
∗
=c
6μ (K + 2μ ) + μ (9K + 8μ ) 6μ (K + 2μ ) + μ (9K + 8μ )
In a frozen porous media, we use these formulae to obtain the maximum effective
(max) (max) (max)
moduli μs1,m , Ks3,m , μs3,m for each solid phase matrix, using the known values of
Ks1 , μs1 and Ks3 , μs3 for the background medium with inclusions of air, with prop-
erties Ka , μa .
For the solid matrix Ω1 the concentration is c = 1 − φ1 and for the ice matrix Ω3
we consider that the water is totally frozen, i.e. c = φ1 .
Chapter 5
Absorbing boundary conditions in elastic and
poroelastic media
We consider an elastic body Ω and follow the ideas in [Loeve, 1944] and [Santos et al.,
1988].
Consider a small disturbance originated in a restricted portion of an elastic solid
medium Ω = Ω1 ∪ Ω2 characterized by Lamé parameters λ and μ and density ρ .
We may assume that the disturbed portion is bounded at any instant by a surface
S. If the medium is isotropic and if the disturbance involves dilatation, we may
expect that the surface S moves normally to itself with velocity
λ + 2μ 1/2
( ) .
ρ
If the disturbance involves rotation without dilation, we may expect the velocity of
the surface to be
μ
( )1/2 .
ρ
Then, let us assume that the surface moves normally to itself with velocity c and let
us seek the conditions that must be satisfied at the moving surface.
Let uc = (uci )1≤i≤3 be the displacement vector. On one side Ω1 of the surface S at
time t, the medium is disturbed and uc = 0. On the other side, (Ω2 )uc ≡ 0. We take
the velocity c to be directed from the side Ω1 into the side Ω2 so that the disturbance
χ1
χ2
Disturbed
Region
Ω1
spreads into parts of the medium Ω2 that were previously undisturbed, see Figure
5.1.
Since uc is continuous across S, uc must vanish on S; i.e.,
uc ≡ 0 on S.
Consequently,
∂ uci
= ∇uci · χ = 0,
∂x
for any χ in the plane defined by
x · ν = 0,
∇uci = γν , (5.1)
so that
∇uci · ν = γν · ν = γ ;
i.e.,
∂ uci
γ= . (5.2)
∂ν
Now from (5.1) and (5.2),
c
∂ ui ∂ uci ∂ uci ∂ uc
, , = i · (ν1 , ν2 , ν3 ).
∂ x1 ∂ x2 ∂ x3 ∂ν
Thus,
5 Absorbing boundary conditions in elastic and poroelastic media 99
}c δ t
δS
(x1 + cδ t ν1 , x2 + cδ t ν2 , x3 + cδ t ν3 ,t + dt)
in (5.4). Thus,
Hence,
∂ uci ∂ uc ∂ uc
+ c∇uci · ν = i + c i = 0 on S. (5.6)
∂t ∂t ∂ν
Now from (5.3) and (5.6) we obtain the relations
∂ uci
∂xj 1 ∂ uci
=− , on S, 1 ≤ i ≤ 3, j not summed. (5.7)
νj c ∂t
In (5.7), the derivatives need to be computed from the side Ω1 where there is a
disturbance at time t. The dynamical conditions which hold at the surface S are
found by considering the change in momentum of a thin slice of the medium in a
neighborhood of S. We mark out a small area δ S of S and consider the prismatic
100 5.1 The Elastic Isotropic Case
Fs,i = −σi j ν j ,
and
∂W
ρ cu̇ci = −σi j ν j = − ν j, 1 ≤ i ≤ j. (5.10)
∂ εi j
Let χ 1 and χ 2 be two tangent vectors at the point 0 ∈ S. Then from (5.11) we get
the three equations
Set
1 1 1 1
vc1 = u̇c · ν = u̇ci νi , vc2 = u̇c · χ 1 = u̇ci χi1 , (5.13)
c c c c
1 1
vc3 = u̇c · χ 2 = u̇ci χi2 vc = (vc1 , vc2 , vc3 )t .
c c
In the new variables vci equations (5.12) become
Next we write the right–hand sides of (5.14) in terms of the variables vc1 , vc2 , and vc3 .
For that purpose, we first note that using (5.7) we can write εi j (uc ) on the surface S
in the form
5 Absorbing boundary conditions in elastic and poroelastic media 101
c
1 ∂ uci ∂ u j 1 1 1
εi j (uc ) = + = − ν j u̇ci + νi u̇cj . (5.15)
2 ∂ x j ∂ xi 2 c c
Consequently,
1 1
σ ν · ν = σi j νi ν j = −λ vc1 δi j νi ν j − μ u̇ci ν j νi ν j − μ u̇cj νi νi ν j
c c
= −λ vc1 − 2μ vc1 = −(λ + 2μ )vc1 , (5.19)
and
1 1
σ ν · χ 1 = σi j νi χ 1j = −λ vc1 δi j νi χ 1j − μ u̇ci ν j νi χ 1j − μ u̇cj νi νi χ 1j
c c
= −μ vc2 . (5.20)
Similarly,
σ ν · χ 2 = −μ vc3 . (5.21)
1 1 c 1
εi j νi χ 1j = − u̇i ν j + u̇cj νi νi χ 1j
2 c c
1 1 c 1 1
=− u̇i νi ν j χ 1j + u̇cj χ 1j νi νi = − vc2 ,
2 c c 2
and
1
εi j νi χ 2j = − vc3 .
2
102 5.1 The Elastic Isotropic Case
Let us compute the strain energy density W (εi j ) on the surface S in terms of the
variables (vci )1≤i≤3 . To simplify the calculations, let us assume that we have changed
coordinates so that ν = (1, 0, 0), χ 1 = (0, 1, 0), and χ 2 = (0, 0, 1). Then, from (5.13)
1 1 1 1 1 1
vc1 = u̇ci νi = u̇c1 , vc2 = u̇ci χi1 = u̇c2 , vc3 = u̇ci χi2 = u̇c3 .
c c c c c c
Also, from (5.15) and (5.18),
1 1 1
ε11 = − ν1 u̇c1 + ν1 u̇c1 = −vc1 ,
2 c c
1 1 1 1
ε12 = − ν2 u̇c1 + ν1 u̇c2 = − vc2 ,
2 c c 2
1 1 1 1
ε13 = − ν3 u̇c1 + ν1 u̇c3 = − vc3 ,
2 c c 2
1 1 1
ε23 = − ν3 u̇c2 + ν2 u̇c3 = 0,
2 c c
ε22 = ε33 = 0,
σ11 = λ ε11 + 2με11 = −(λ + 2μ )vc1 ,
1
σ12 = 2με12 = −2μ vc2 = −μ vc2 ,
2
σ13 = 2με13 = −μ vc3 ,
σ23 = σ22 = σ33 = 0.
Hence,
1
Π (vc ) = (vc )t E vc ,
2
where ⎛ ⎞
λ + 2μ 0 0
E =⎝ 0 μ 0 ⎠.
0 0 μ
Since
∂Π ∂Π ∂Π
= (λ + 2μ )vc1 , = μ vc2 , = μ vc3 ,
∂ vc1 ∂ vc2 ∂ vc3
from (5.19), (5.20) and (5.21), we see that (5.12) can also be written in the form
5 Absorbing boundary conditions in elastic and poroelastic media 103
∂Π ∂Π ∂Π
ρ c2 vc1 = − , ρ c2 vc2 = − , ρ c2 vc3 = − . (5.22)
∂ vc1 ∂ vc2 ∂ vc3
∂Π
c2 ρ vc = = −Fs = E vc . (5.23)
∂ vc
Next write (5.23) in the form
Set
vc = ρ 1/2 vc ,
S = ρ −1/2 E ρ −1/2 .
c2 vc = S vc . (5.24)
Also, in terms of vc , the strain energy density on the surface S can be written in the
form
1
Π (vc ) = Π (vc ) = (vc )t S vc (5.25)
2
1 1/2 c t −1/2 1
= [ρ (v ) ]ρ S ρ −1/2 (ρ 1/2 vc ) = vc S vc . (5.26)
2 2
Let (ci )1≤i≤3 be the three positive wave speeds satisfying (5.24); i.e., solutions of
det(S − c2 I) = 0.
so that
104 5.1 The Elastic Isotropic Case
S = Nt Λ N. (5.27)
Set
1
zci = ρ 1/2 zci = [Ni , ρ 1/2 z]e Ni , (5.28)
ci
where [·, ·]e denotes the Euclidean inner product. Then zci satisfies the equation (cf.
(5.24))
and the strain energy associated with zci satisfies the relation (cf. (5.26))
1
Π (zci ) = (zci )t S zci .
2
Now using (5.23) and (5.26) we see that the force Fi on S associated with zci satis-
fies the relation
∂Π
ρ 1/2 = ρ 1/2 S zci = ρ 1/2 ρ −1/2 E ρ −1/2 ρ 1/2 zci = E zci = −F (i) .
∂ z ci
Next we use that the total force Fs on the surface S is equal to the sum of the forces
Fi (see argument leading to the proof of this statement in (5.40)); i.e.,
3 3
Fs = ∑ F (i) = − ∑ ρ 1/2 S zci . (5.30)
i=1 i=1
Consequently, since S zci = c2i zci is a vector in the direction of Ni , from (5.30)
and (5.31) we have that
where
B = ρ S 1/2 .
Note that the matrix B in the right hand side of (5.36) is positive definite.
The absorbing boundary condition (5.36) is stated in the 3D case. In the 2-D case,
it has the same form that in (5.36) but changing the definition of the matrix E to
∗ λ + 2μ 0
E =
0 μ
and defining
S ∗ = ρ −1/2 E ∗ ρ −1/2 .
Thus the 2-D absorbing boundary condition is
with
B ∗ = ρ S ∗ 1/2 . (5.37)
Remark. The argument given above for the isotropic case, was extended in
[Lovera and Santos, 1988] to obtain an absorbing boundary condition for the gen-
eral anisotropic elastic case using the constitutive relation
106 5.1 The Elastic Isotropic Case
σ12
σ11
x1
x1 = 1
Remark. Let us for simplicity in the argument analyze the 2-D case and the
interface x1 = 1. The work done at any interface averaged over a period is given by
(see Figure 5.3)
2π
ω ω
F = (σ11 u̇1 + σ12 u̇2 )dt.
2π 0
F represents the energy flux across the interface and since σi j and u̇ j are con-
tinuous, we have conservation of energy. Since
1 11 1
ε11 = u̇1 = −vc1 , ε12 = − u̇2 = − vc2
c 2c 2
we may rewrite F in the form
2π 2π
1 ω ω 1 ω ω
F = − (σ11 ε11 + 2σ12 ε12 )dt = − Π (vc )dt.
c 2π 0 c 2π 0
ui = ui,p + ui,s
2π
ω ω (σ11,k u̇1,k + σ12,k u̇2,k )dt,
Fk,k = k = p, s,
2π 0
2π
ω ω
F j,k = (σ11, j u̇1,k + σ12, j u̇2,k )dt, j, k = p, s, j = k,
2π 0
5 Absorbing boundary conditions in elastic and poroelastic media 107
where σi j,p denotes the stress associated with the compressional mode and likewise
for σi j,s . It can be seen that F j,k = 0 for j = k ([Dutta and Odé, 1983]). This shows
that the energy flux on Γ can be written as the sum of the energy fluxes associated
with each type of wave, so that
3 3
Π (z) = ∑ Π (zci ), Π (z) = ∑ Π (zci ). (5.39)
i=1 i=1
where
⎛ ⎞
ρI ρ f I ⎠
A =⎝ ,
ρ f I gI
and I denotes the identity matrix in R3×3 . Equation (5.41) can also be written in the
form
∂W
c(ρ u̇s,c + ρ f u̇ f ,c ) = −τi j ν j = − ν j , on S, (5.42)
∂ εi j
∂W
c(ρ f u̇s,c + gu̇ f ,c ) = p f νi = νi , on S. (5.43)
∂ξ
As in the elastic case, let χ 1 and χ 2 be two tangent vectors at the point o ∈ S (see
Figure 5.1). Taking the inner product with χ 1 and χ 2 in (5.43) we see that
so that
Hence, taking the inner product with ν and χ 1 , χ 2 in (5.42) and using (5.44) we
obtain the equations
1 1
vc1 = u̇s,c · ν , vc2 = u̇s,c · χ 1 ,
c c
1 1
vc3 = u̇s,c · χ 2 , vc4 = u̇ f ,c · ν .
c c
Then in the new variables vci equations (5.45), (5.46) and (5.47) become
where
q = ρ − g−1 (ρ f )2 .
Next we use the constitutive equations (cf.(1.16)-(1.17)),
with
es = ∇ · us , ξ = −∇ · u f ,
and (5.15) to write the right hand side of (5.48) in terms of the variables (vci )1≤i≤4 .
First note that
ξ = −∇ · u f ,c = −εii (u f ,c )
1 1 1
= νi uif ,c + νi uif ,c = u f ,c · ν = vc4 .
2 c c
Thus,
Consequently,
1
τν · ν = τi j νi ν j = −(λu vc1 + B vc4 )δi j νi ν j − μ u̇is,c ν j ν j νi
c
1 s,c
−μ u̇ j νi νi ν j = −(λu + 2μ ) vc1 − B vc4 ,
c
1
τν · χ = τi j νi χ j = −(λu vc1 + B vc4 )δi j νi χ 1j − μ u̇is,c ν j νi χ 1j
1 1
c
1 2,c
−μ u̇ j νi νi x j = −μ v2 ,
1 c
c
τν · χ = −μ v3 ,
2 c
p f = B vc1 + M vc4 .
Set
t
F = τν · ν , τν · χ 1 , τν · χ 2 , −p f ,
⎛ ⎞ ⎛ ⎞
ρ 00 ρf λu + 2 μ 0 0 B
⎜ 0 q0 0 ⎟ ⎜ 0 μ 0 0⎟
⎜
Ap = ⎝ ⎟, ⎜
Ep = ⎝ ⎟.
0 0q 0 ⎠ 0 0 μ 0⎠
ρf 00 g B 0 0 M
c2 A p vc = E p vc = −F . (5.49)
Next, a calculation similar to that given for the elastic solid case shows that the
strain energy density W (εi j , ξ ) on the surface S can be written in terms of the vari-
ables (vci )1≤i≤4 in the form
1
Π (vc ) = (vc )t E p vc .
2
Thus, (5.49) can also be stated in the equivalent form
∂ Π (vc )
c2 A p vc = = E p vc = −F , on S. (5.50)
∂ vc
Set
1/2
vc = A p vc , (5.51)
−1/2 −1/2
S = Ap EpAp . (5.52)
110 5.2 The case of a porous elastic solid saturated by a single-phase fluid
S vc = c2 vc . (5.53)
Also, in terms of vc the strain energy density on S can be written in the form
1
Π (vc ) = Π (vc ) = (vc )t E p vc
2
1 1
= (vc )t (A p )1/2 (A p )−1/2 E p (A p )−1/2 (A p )1/2 vc = (vc )t S vc .
2 2
Let (ci )1≤i≤4 be the four positive wave speeds satisfying (5.53); i.e., solutions of the
equation
det(S − c2 I) = 0.
and they correspond to the shear modes of propagation. The other two roots are
associated with the compressional models of propagation; i.e., the P1 and P2 com-
pressional waves.
Next, let Ni , 1 ≤ i ≤ 4, be the set of orthonormal eigenvectors corresponding to
(ci )2 , 1 ≤ i ≤ 4, and let N be the matrix containing the eigenvectors Ni of S as rows
and Λ the diagonal matrix containing the eigenvalues (ci )2 , 1 ≤ i ≤ 4, of S , i.e.,
⎛ 2 ⎞
c1 0 0 0
⎜ 0 c2 0 0 ⎟
Λ =⎜ 2
⎝ 0 0 c2 0 ⎠
⎟ (5.54)
3
B 0 0 c4 2
so that
S = Nt Λ N.
Next, let
z = A p 1/2 z.
4 4
z = ∑ [Ni , z]e Ni = ∑ [Ni , A p 1/2 z]e Ni .
i=1 i=1
Set
1
zci = A p 1/2 zci ≡ [Ni , A p 1/2 z]e Ni , 1 ≤ i ≤ 4. (5.55)
ci
Since zci is a multiple of Ni , we see that
and
1
Π (zci ) = (zci )t S zci .
2
Also, using (5.50) we see that the force Fi on S associated with zci satisfies the
equation
∂Π
A p 1/2 = A p 1/2 S zci = A p 1/2 A p −1/2 E p A p −1/2 A p 1/2 zci (5.57)
∂ z ci
= E p zci = −F (i) .
It is known that the interaction among the different types of waves arriving at
an interface in a saturated porous medium is small compared with the total energy
involved ([Dutta and Odé, 1983, Santos et al., 1992]). Neglecting such interactions,
we can write the total strain energy density on S as the sum of the partial energies;
i.e.,
4
Π (z) = ∑ Π (zci ),
i=1
and the total force F on the surface S as the sum of forces associated with each zci
so that, according to (5.57),
4 4
F = ∑ F (i) = − ∑ A p 1/2 S zci .
i=1 i=1
Consequently,
where
# t $1/2
A p −1 E p
1/2
B p = [NA p 1/2 ]t Λ 1/2 [NA p ] = A p 1/2 S 1/2 A p 1/2 = A p . (5.61)
Equation (5.60) defines the first–order absorbing boundary conditions on the surface
S, with the matrix S defined in (5.52).
Note that NA p 1/2 is nonsingular and, consequently, B p in (5.61) is positive def-
inite.
The absorbing boundary condition (5.60) is stated in the 3-D case. For the 2-D
case, it has formally the same form:
− τν · ν , τν · χ 1 , −p f = B ∗p (u̇s · ν , u̇s · χ 1 , u̇ f · ν )t ,
where
and
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
ρ 0 ρf λ + 2μ 0 B c21 0 0
A p∗ = ⎝ 0 q 0 ⎠ , E p∗ = ⎝ 0 μ 0 ⎠ , Λ ∗ = ⎝ 0 c22 0 ⎠ .
ρf 0 g B 0M 0 0 c23
S ∗ = A p∗ −1/2 E p∗ A p∗ −1/2 .
5 Absorbing boundary conditions in elastic and poroelastic media 113
where
t 1 1 1
B p = [ A p−1 E p ] 2 A p = A p2 S 2 A p2
1
with
−1 −1
S = Ap 2 EpAp 2 ,
⎛ ⎞ ⎛ ⎞
ρ 0 0 S n ρn S w ρw λu + 2 μ 0 0 B1 B2
⎜ 0 ρ 0 0 0 ⎟ ⎜ 0 μ 0 0 0 ⎟
⎜ ⎟ ⎜ ⎟
Ap = ⎜
⎜ 0 0
ρ 0 0 ⎟ ⎟, Ep = ⎜
⎜ 0 0 μ 0 0 ⎟⎟.
⎝ S n ρn 0 0 gn gnw ⎠ ⎝ B1 0 0 M1 M3 ⎠
S w ρw 0 0 gnw gw B2 0 0 M3 M2
where
t 1
B ∗p = [ A p∗ −1 E p∗ ] 2 A p∗ = A p∗ 2 S ∗ 2 A p∗ 2
1 1 1
with
S ∗ = A p − 2 E p∗ A p∗ − 2 ,
1 1
and ⎡ ⎤ ⎡ ⎤
ρ 0 S̄n ρn S̄w ρw λu + 2 μ 0 B1 B2
⎢ 0 ρ 0 0 ⎥ ⎢ 0 μ 0 0 ⎥
Ap = ⎢
∗
⎣ S̄n ρn
⎥, Ep = ⎢
∗ ⎥.
0 gn gnw ⎦ ⎣ B1 0 M1 M3 ⎦
S̄w ρw 0 gnw gw B2 0 M3 M2
114 5.4 The case of a composite solid matrix saturated by a single-phase fluid
c2
m11 vc1 + m12 vc4 + m13 vc5 = −σ (1,T ) ν · ν , (5.68)
5 Absorbing boundary conditions in elastic and poroelastic media 115
c2 q1 vc2 + q2 vc6 = −σ (1,T ) ν · χ 1,
c q1 v3 + q2 v7 = −σ (1,T ) ν · χ 2,
2 c c
c m12 v1 + m22 v4 + m23 v5 = p f ,
2 c c c
c m13 v1 + m23 v4 + m33 v5 = −σ (3,T ) ν · ν ,
2 c c c
c q2 v2 + q3 v6 = −σ (3,T ) ν · χ 1 ,
2 c c
c q2 v3 + q3 v7 = −σ (3,T ) ν · χ 2 ,
2 c c
where
m212 m12 m23 m223
q1 = m11 − , q2 = m13 − , q3 = m33 − .
m22 m22 m22
Next we write the force
F = σ (1,T ) ν · ν , σ (1,T ) ν · χ 1 , σ (1,T ) ν · χ 2 , −p f ,
t
σ (3,T ) ν · ν , σ (3,T ) ν · χ 1 , σ (3,T ) ν · χ 2
on S associated with the arrival of the wave front travelling with speed c in terms of
the new variable vc = (vci ). For that purpose, note that on the surface S, the strain
tensor εi j (u(m,c) ) and the deviatoric tensor di j (u(m,c) ) can be written in the form
(m,c) (m,c)
(m,c) 1 ∂ ui ∂uj 1 1 (m,c) 1 (m,c)
εi j (u )= + = − ν j u̇i + νi u̇ j , (5.69)
2 ∂xj ∂ xi 2 c c
1
di j (u(m,c) ) = εi j (u(m,c) ) − εll (u(m,c) )δi j
3
1 1 (m,c) 1 (m,c) 1 1 (m,c)
= − ν j u̇i + νi u̇ j + u̇ νl δi j , m = 1, 2, 3.
2 c c 3c l
(1,T ) (1)
σi j = σi j − S1 φ p f δi j (5.70)
(1) (3)
= E1 θ1 − B1 ζ + B3 θ3 δi j + 2μ1 di j + μ1,3 di j ),
(3,T ) (3)
σi j = σi j − S3 φ p f δi j
116 5.4 The case of a composite solid matrix saturated by a single-phase fluid
(3) (1)
= E3 θ3 − B2 ζ + B3 θ1 δi j + 2μ3 di j + μ1,3 di j , (5.71)
p f = −B1 θ1 − B2 θ3 + M ζ , (5.72)
c2 A p vc = −F = E p vc . (5.73)
On the other hand, using (5.69) in (5.70)–(5.72) we conclude that the strain en-
ergy density on Π on S can be written in the form
1
Π (vc ) = (vc )t E p vc ,
2
so that (5.73) can be stated in the equivalent form
∂ Π (vc )
c2 A p vc = −F = . (5.74)
∂ vc
Set
−1 −1
S = Ap 2 EpAp 2 ,
1
vc = A p2 vc .
S vc = c2 vc ,
det(S − c2 I) = 0
are the squares of the wave speeds of the system. Four of them (only two are dif-
ferent) correspond to the shear modes of propagation, and the other three to the
compressional modes.
Let Ni , i = 1, · · · , 7 be the set of orthonormal eigenvectors associated with the
eigenvalues (ci )2 , i = 1, · · · , 7, and let N be the matrix containing as rows the eigen-
vectors Ni and let
Λ = diag(c21 , · · · , c27 )
be the diagonal matrix containing the eigenvalues (ci )2 . Thus,
S = Nt Λ N.
Let
z = u̇(1) · ν , u̇(1) · χ 1 , u̇(1) · χ 2 , ẇ · ν ,
t
u̇(3) · ν , u̇(3) · χ 1 , u̇(3) · χ 2
be a general velocity field on Γ due to the simultaneous arrival of waves with speeds
ci , i = 1, · · · , 7.
Set
1
z = A p2 z
Set
1 1 1
zci = A p2 zci = [Ni , A p2 z]e Ni . (5.76)
ci
and
1
Π (zci ) = (zci )t S zci .
2
Also, using (5.74) and (5.75), we see that the force Fi associated with zci satisfies
the equation
1 ∂ Π (zci ) 1
A p2 = A p2 S zci = E p zci = −Fi .
∂z ci
Assuming that the interaction among the different waves arriving at a given in-
terface Γ is small compared with the total energy involved (see [Santos et al., 1988]
for the validity of this assumption in the case of a single solid phase), the total strain
energy density on S is equal to the sum of the partial energies and the total force F
on S is equal to the sum of the forces associated with each type of wave, so that
Π (z) = ∑ Π (zci ),
i
and
F = ∑ Fi = − ∑ A p 2 S zci .
1
(5.78)
i i
−1
Next, decomposing A p 2 F in term of the eigenvectors Ni ,
−1 −1
A p 2 F = ∑[Ni , A p 2 F ]e Ni . (5.79)
i
−F = B p z (5.82)
where
1
1 1 1 2
Bp = Ap S Ap =
2 2 2
E p A p−1 Ap. (5.83)
Abstract The finite element method (FEM) is a useful tool to solve boundary value
problems of interest in applied geophysics. 1-D finite element spaces are first de-
fined and analyzed. The concept of continuous and discrete weak solutions is in-
troduced and a priori error estimates are stated. The FEM is used to solve wave
propagation problems and to characterize fine layered media in the 1-D case. Next,
2-D and 3-D conforming and non-conforming finite element spaces and defined
over partitions of a bounded domain into triangular or rectangular elements in 2-D
and tetrahedral or 3-rectangular elements in 3-D. These finite element spaces are
used in the following Chapters to represent solid or fluid vector displacements in
the boundary value problems to be formulated and solved using the FEM.
6.1 Introduction
This Chapter presents the use of the finite element technique to find approximate
solutions to boundary problems appearing in applied geophysics.
The basic idea of the finite element method is to use a partition the domain into a
set of sub-domains, the finite element partition, where the solution is being sought,
and using what is known as the weak formulation of the original problem to deter-
mine an approximate solution over such partition.
To illustrate the finite element approach to solve boundary value problems two
simple model problem are considered. The first is the propagation of waves in one-
dimensional viscoelastic media with absorbing boundary conditions. These bound-
ary conditions act to make the boundaries transparent for arriving waves, thus al-
lowing to simulate wave propagation in unbounded domains.
The second model problem consists of simulating the seismic response of a peri-
odic sequence of fine viscoelastic layers. Many geological systems can be modeled
as effective transversely isotropic and viscoelastic (TIV) media. Fine layering is a
typical example which refers to the case when the dominant wavelength of the trav-
elling waves is much larger than the average thicknesses of the single layers.
Backus [Backus, 1962] obtained the average elastic constants in the case when
the single layers are transversely isotropic and elastic (lossless), with the symmetry
axis perpendicular to the layering plane.
Backus averaging was verified numerically in [Carcione et al., 1991], and gener-
alized to the anelastic case in [Carcione, 1992] in what constitutes the Backus/Carcione
(BC) theory to describe anisotropic attenuation [Carcione, 2014].
In the second model problem we perform 1-D time-harmonic finite element ex-
periments to determine the effective stiffness coefficient for waves travelling nor-
mally to the periodic layering. The numerical results are validated against the BC
theory.
Next, a set of finite element spaces to be used in the following chapters to solve
boundary value problems in 2-D and 3-D are defined and analyzed.
The 1-D wave propagation problem in viscoelastic media is one of the simplest ex-
amples we can develop since it is represented by a second order differential equation
with boundary conditions at the two endpoints of the domain. For this purpose, let
u = u(x) denote the displacement of the viscoelastic medium and let
∂u
ε (u) = (6.1)
∂x
be the strain tensor.
Then, with σ (u) denoting the stress tensor and ω the angular frequency, the
stress-strain relations, stated in the space-frequency domain, are
where E(ω ) = λ (x, ω ) + 2μ (x, ω ) is the complex plane-wave modulus and the co-
efficients λ (x, ω ) and μ (x, ω ) are the complex and frequency dependent Lamé co-
efficients of the viscoelastic material. Also,
2
λ (x, ω ) = K(x, ω ) − μ (x, ω )
3
with K(x, ω ) denoting the bulk modulus. To represent the viscoelastic behaviour
of the material, we model the bulk and shear moduli using their relaxed (zero fre-
quency) values Kr and μr and the models described in Appendix 1.9.
Recall that for either the Zener or the Liu models of viscoelasticity, we have that,
for any ω > 0,
with ER (ω ) and EI (ω ) denoting the real and imaginary parts of E(ω ) and E (r) ,
E (u) the relaxed and unrelaxed values of E(ω ). Space and frequency are the natural
coordinates to express the equation of motion.
Next let us consider a domain Ω = (0, L) with boundary ∂ Ω = {0, L}; we for-
mulate the following boundary value problem in the space-frequency domain as
follows: find u(x, ω ) such that
where ρ = ρ (x) denotes the mass density and (6.4) is an absorbing boundary con-
dition. Also, f (x, ω ) is the time Fourier transform of an external real valued source
f(x,t) to be used to excite the medium and generate travelling waves. It will be as-
sumed that
f (x,t) = 0, t ≤ 0.
Note that since the solution u(x,t) is real, the
u(x, −ω ) = u(x, ω )
and consequently only the case ω > 0 needs to be considered. Here u denotes the
complex conjugate of u.
In this model problem we intend to simulate the propagation of waves in an
unbounded domain. In the boundary condition (6.4) the coefficient B(ω ) is used
here for this purpose, which acts eliminating spurious reflections from the artificial
boundaries. It has the form:
B(ω ) = iω α − ω β,
A B
=ρ
α β = ρ ,
+ (B)
(A) 2 2 + (B)
(A) 2 2
'
(
(1 c2 c 4 − c4
=
A ) R
+ R I
,
2 c4R + c4I c4R + c4I
1 cI
B = ,
2A cR + c2I
2
ER EI
cR = , cI = .
ρ ρ
124 6.3 A variational formulation for the 1-D wave propagation model problem
The form of the coefficient B(ω ) can be obtained considering plane waves and
asking that they satisfy (6.3) at ∂ Ω (see [Ravazzoli and Santos, 1995] for details).
In many cases of practical interest, the coefficients λ and μ are discontinuous,
such as piecewise constants, and the external source may not even be a function, but
instead Dirac distributions or derivatives in the space variable of Dirac distributions,
modeling the spatial location of point sources.
Consequently, the solution of (6.3)-(6.4) may not exist in the classical sense, i.e.,
asking that the solution be smooth and that such equations be valid at every point.
Thus, we formulate the problem in a weak form, in order to be able to solve prob-
lems not satisfying strong smooth requirements on their coefficients or the external
sources.
−ω 2 ρ u(x, ω )v(x)dx − ∇ · σ (u(x, ω ))v(x)dx
Ω Ω
= f (x, ω )v(x)dx (6.5)
Ω
be valid for test functions v belonging to a set to be specified later. Here v denotes
the complex conjugate of the function v. Also, we use ∇ · σ (u(x, ω )) to denote
∂ σ (u(x, ω ))
.
∂x
Using integration by parts in the second term in the left-hand side of (6.5) we
obtain
−ω 2 ρ u(x, ω )v(x)dx + σ (u(x, ω ))ε (v(x))dx − σ (u(x)) · ν v(x)d(∂ Ω )
Ω Ω ∂Ω
= f (x, ω )v(x)dx. (6.6)
Ω
Thus, applying the boundary condition (6.4) in (6.6) and the definitions of σ (u)
and ε (v) in (6.2) and (6.1) we get the equation
6 Solution of differential equations using the finite element method 125
∂ u(x, ω ) ∂ v(x)
−ω 2 ρ u(x, ω )v(x)dx + E(x, ω ) dx (6.7)
Ω Ω ∂x ∂x
+ B(x, ω )u(x, ω )v(x)d(∂ Ω )
∂Ω
∂ u(x, ω ) ∂ v(x)
= −ω 2 ρ u(x, ω )v(x)dx + E(x, ω ) dx
Ω Ω
∂x ∂x
+ iωα (x, ω )u(x, ω )v(x)d(∂ Ω ) − ωβ (x, ω )u(x, ω )v(x)d(∂ Ω )
∂ Ω ∂Ω
= f (x, ω )v(x)dx.
Ω
it is clear that in order that (6.7) holds, we need that the solution u and the test
function v as well as their first derivatives be square integrable.
Let C denote the set of complex numbers and let us introduce the space L2 (Ω )
of square integrable complex valued functions as
L2 (Ω ) = { f : Ω → C such that | f (x)|2 dx < ∞}.
Ω
The space of functions L2 (Ω ) can be provided with a natural inner product and
norm, denoted by (·, ·) and · 0 , respectively, defined as follows:
( f , g) = f (x)g(x) dx, f , g ∈ L2 (Ω ),
Ω
1/2
f 0 = | f (x)| dx
2
, f ∈ L2 (Ω ).
Ω
Using this notation, we can state the original model problem in weak or variational
form as follows: find u ∈ H 1 (Ω ) such that
+ ,
∂u ∂v
Λ (u, v) = −ω (ρ u, v) + E ,
2 u, v − ω β u, v
+ iω α (6.11)
∂x ∂x
= ( f , v) , ∀ v ∈ H 1 (Ω ).
Note that in the above formulation, the boundary condition (6.4) has been imposed
weakly, i.e., it has not been required that the test functions v satisfy (6.4).
The existence of a solution of (6.11) follows from a functional analysis argument
using the Fredholm alternative [Yosida, 1980]. The proof is outside the scope of this
book and is omitted.
Let us turn to analyze uniqueness of the solution of problem (6.11). For that
purpose, we assume there are two solution, u1 and u2 , then the difference is u =
u1 − u2 and satisfies (6.11) for f = 0. Thus set f = 0 and take v = u in (6.11) to
obtain
+ ,
∂u ∂u
−ω (ρ u, u) + (ER + iEI ) ,
2
+ iω α u, u − ω β u, u = 0. (6.12)
∂x ∂x
∂ u(·, ω )
0 = 0, u(x, ω ) = 0, x ∈ {0, L}. (6.14)
∂x
Next, from (6.14), assuming that u(x, ω ) is smooth, we can write,
x
∂ u(z, ω )
u(x, ω ) = dz (6.15)
0 ∂z
so that squaring the equation in (6.15) and using (6.8) with f = u and g = 1 leads to
the inequality
* * x L
* ∂ u(z, ω ) **2
x ∂ u(z, ω ) 2
|u(x, ω )|2 = ** dz* dz ≤ | | dz dz
0 ∂z 0 ∂z 0
L
∂ u(z, ω ) 2
≤L | | dz. (6.16)
0 ∂z
6 Solution of differential equations using the finite element method 127
∂ u(x, ω )
u(·, ω )0 ≤ L 0 . (6.17)
∂x
Inequality (6.17) is know as Poincare’ s inequality. Consequently
u(·, ω )0 = 0,
and we have uniqueness for the solution of (6.11) assuming that u(·, ω ) is smooth.
An argument involving the density-of smooth - functions in the space H 1 (Ω )
-∂v-
[Adams, 1975], it can be shown that - -
- ∂ x - defines a norm | · | equivalent to
0
the · 1 -norm defined in (6.9) in the space of functions
Hence, (6.14) says that u(·, ω ) ∈ H01 (Ω ) and |u(·, ω )| = 0, so that uniqueness
holds for the solution of (6.11).
Note that the space H 1 (Ω ) where we are seeking the solution of (6.11) is infinite-
dimensional. Thus we would need an infinite number of coefficients to represent the
solution u in any basis of H 1 (Ω ).
Consequently, we will consider the problem of finding an approximate solution
of (6.11) in a appropriate finite dimensional subspace of H 1 (Ω ) using the finite
element method.
M h = {v : v|Ω j ∈ P1 (Ω j )} ∩C 0 (Ω ),
∂v
Remark. Note that since elements v ∈ M h are continuous in Ω and are piece-
∂x
∂v
wise constant, both v and are functions in L2 (Ω ) and consequently M h ⊂
∂x
H 1 (Ω ).
Thus we are searching for an approximate solution of u in a finite dimensional
space of piecewise polynomials that is contained in the space H 1 (Ω ) where the
solution u of the continuous problem (6.11) belongs.
When this is the case, it is said that the space M h is a conforming finite element
space of H 1 (Ω ).
Now the finite element procedure to compute the approximate solution of (6.11)
is defined as follows: find uh ∈ M h such that
∂ uh ∂ v + , + ,
Λ (uh , v) = −ω 2 ρ uh , v + E , + iω α uh , v − ω β uh , v
∂x ∂x
= ( f , v) , ∀ v ∈ M h . (6.18)
Uniqueness for the solution of the (6.18) can be demonstrated with the same argu-
ment than for the solution of the continuous problem (6.11). Existence of the solu-
tion of (6.18) follows from finite dimensionality (see the argument in the Remark
following (6.23)).
Next, let
Nx +1
uh = ∑ u jϕ j. (6.19)
j=1
ϕ1 ϕ2 ϕ3 ϕNx ϕNx+1
1
x1 x2 x3 x4 x Nx−1 x x
Nx Nx+1
Ω1 Ω2 Ω3 ΩNx−1 ΩNx
Fig. 6.1 Finite element mesh for the 1-D model problem conformed by nodes and elements. More-
over, piecewise basis functions are shown.
dim M h = Nx + 1.
Since the functions {ϕ1 , · · · , ϕNx +1 } are linearly independent they are a basis for
M h . Hence (6.18) is equivalent to
Λ (uh , ϕk ) = ( f , ϕk ) , k = 1, · · · , Nx + 1, (6.21)
Λ (ϕ j , ϕk )u j = ( f , ϕk ), k = 1, · · · , Nx + 1. (6.22)
Au = b, (6.23)
To obtain the linear system associated with (6.21) we compute each integral assum-
ing that the coefficients ρ and E are piecewise constant. We denote by ρk and Ek to
the values of ρ and E in the subinterval (xk , xk+1 ).
Since
Nx +1 k+1
(ρ uh , ϕk ) = ρ ∑ u j ϕ j , ϕk = ρ ∑ u j ϕ j , ϕk
j=1 j=k−1
and
h
(ρϕk−1 , ϕk ) = ρk−1 (1 − δk,1 ),
6
h h
(ρϕk , ϕk ) = ρk−1 (1 − δk,1 ) + ρk (1 − δk,Nx +1 ),
3 3
h
(ρϕk+1 , ϕk ) = ρk (1 − δk,Nx +1 ),
6
we obtain
h h h
(ρ u , ϕk ) = ρk−1 (1 − δk,1 )uk−1 + ρk−1 (1 − δk,1 ) + ρk (1 − δk,Nx +1 ) uk
h
6 3 3
h
+ρk (1 − δk,Nx +1 )uk+1 . (6.26)
6
Here δk,1 and δk,Nx +1 denote the Kronecker’s delta, so that the term multipled by
that factor dissapears from the equation when k = 1 or k = Nx + 1.
Next, note that
6 Solution of differential equations using the finite element method 131
∂ uh ∂ ϕk k+1
∂ ϕ j ∂ ϕk
E ,
∂x ∂x
= E ∑ uj ,
∂x ∂x
,
j=k−1
and
∂ ϕk−1 ∂ ϕk 1
E , = − Ek−1 (1 − δk,1 ),
∂x ∂x h
∂ ϕk ∂ ϕk 1 1
E , = Ek−1 (1 − δk,1 ) + Ek (1 − δk,Nx +1 ),
∂x ∂x h h
∂ ϕk+1 ∂ ϕk 1
E , = − Ek (1 − δk,Nx +1 ).
∂x ∂x h
Thus
∂ uh ∂ ϕk 1
E , = − Ek−1 (1 − δk,1 )uk−1 (6.27)
∂x ∂x h
1 1
+ Ek−1 (1 − δk,1 ) + Ek (1 − δk,Nx +1 ) uk
h h
1
− (1 − δk,Nx +1 )Ek uk+1 .
h
Next, note that
+ , + ,
uh , ϕk · ν = 0,
iω α ω βuh , ϕk · ν = 0 only for k = 1 or k = Nx + 1.
Now collecting all terms in (6.26), (6.27) and (6.28) we get the following tridiagonal
linear system to be solved for each ω > 0:
h 1
−ω ρk−1 − Ek−1 (1 − δk,1 )uk−1
2
(6.29)
6 h
h h
+ −ω 2 ρk−1 (1 − δk,1 ) + ρk (1 − δk,Nx +1 )
3 3
1 1
+ Ek−1 (1 − δk,1 ) + Ek (1 − δk,Nx +1 )
h h $
+ iω α
k − ω βk δk,1 + iω α k − ω βk δk,Nx +1 uk
hk 1
+ −ω 2 ρk − Ek (1 − δk,Nx +1 )uk+1
6 hk
132 6.6 A numerical example for the 1-D wave propagation problem
= ( f , ϕk ), k = 1, · · · , Nx + 1
Equation (6.29) defines the coefficients of the tridiagonal matrix A and the right-
hand side vector b in (6.23).
In this numerical example the domain Ω is an open interval of size 1000 m. The ma-
terial correspond to Nivelsteiner sandstone saturated by a single-phase fluid, water
or gas; with properties indicated in Table 1.1. The interface between the two media
is located at the center of Ω .
The source function is given by the expression
∂ δ (x)
ξ (t − t0 )e−ξ (t−t0 ) ,
2
f (x,t) = C t ≥ 0,
∂x
where f0 =1/t0 =50 Hz is the central frequency of the source, ξ = 8 f02 and C is a con-
stant chosen so that max| f (t)|=1. The Dirac distribution, δ (x), determines the posi-
tion of the source, see Figure 6.2 (a) where we also indicate elastic compressional
velocities for the dry and saturated sandstone and the location of three receivers re-
ferred to as R1, R2 and R3. The saturated real plane wave moduli are calculated by
the Gassmann’s approximation (1.15) since frequency range is low: 0-100 Hz.
The mesh has 582 elements which allow to solve with 25 grid points per mini-
mum wavelength. The solution is computed at 600 discrete points of the frequency
range.
For the viscoelastic case we use a continuous distribution of relaxation mech-
anisms (see Appendix 1.9) where the values of t1 and t2 are such that the quality
factor Q is approximately equal to the constant Q M in the range t −1 ≤ 2π f ≤ t −1 .
1 2
In the example we choose Q M =148 for the saturated bulk modulus, Q M =100 for the
shear modulus, t1 =10−6 ms and t2 =103 ms. Figure 6.2 (b) shows traces of displace-
ment in the three receivers. The arrival time of the transmitted waves are designated
by T1, T2 and T3 for receivers 1, 2 and 3, respectively. The arrival time of the re-
flected waves at the interface are denoted by R1 and R2. Observe that the absorbing
boundary condition eliminates reflections at the boundaries of the domain. The am-
plitude decay observed in the traces represents physical attenuation corresponding
to phenomena included in the viscoelastic model.
We emphasize that in this algorithm the solution is computed separately for each
frequency and only a limited number of them are needed to recover correctly the
solution in the domain, therefore the numerical procedure is naturally suited to be
used in parallel computers. This property becomes a necessity when dealing with
realistic problems in two or three dimensions.
6 Solution of differential equations using the finite element method 133
Vp =1870 m/s
dry
Vp =2545 m/s
Vp =2415
sat
2415m/s sat 1545
Source
* R1 R2 R3
200 m
400 m
950 m
Water saturated sandstone Gas saturated sandstone
(a)
Receiver 1
Receiver 2
0,05 Receiver 3
T1 T2 T3
0
R2 R1
-0,05
(b)
Fig. 6.2 1-D geometrical model (a) and the traces of displacement at the three receivers (b).
134 6.8 Determination of the stiffness p 33
∂ u
e33 (
u) =
∂ x3
the strain tensor at the macro-scale.
If σ (
u) is the stress tensor of the equivalent 1-D medium at the macro-scale,
the corresponding stress-strain relation, stated in the space-frequency domain, is
[Carcione, 1992]
σ33 (
u) = p33 e33 (
u),
where
E = λ +2μ is the complex plane wave modulus of each single layer and · denotes
the thickness weighted average.
The p33 is the complex and frequency-dependent Voigt stiffnesses to be deter-
mined with the harmonic experiments.
where En = λn + 2μn is the complex plane wave modulus of the n-layer and ε (u) is
the strain tensor on each fine layer.
Then if ρ = ρ (x3 ) is the mass density, the equation of motion is
−ω 2 ρ u − ∇ · σ (u) = 0, (6.32)
For this set of boundary conditions the material is not allowed to move on the bottom
boundary Γ B and a uniform compression is applied on the boundary Γ T .
Denoting by V the original volume of the sample, its (complex) oscillatory vol-
ume change, Δ V (ω ), allows us to define the equivalent complex plane-wave modu-
lus p33 (ω ), by using the relation
Δ V (ω ) ΔP
=− ,
V p33 (ω )
−L Δ P
p33 (ω ) ≈ .
u(L, ω )
M0,B
h
= {v : v|Ω j ∈ P1 (Ω j ), v = 0 on Γ B } ∩C0 (Ω ).
6 Solution of differential equations using the finite element method 137
Then the finite element procedure to compute the approximate solution of (6.36)
is defined as follows: find uh ∈ M0,B
h such that
∂ uh ∂ v
−ω 2 ρ uh , v + E , = − Δ P, v · ν Γ T , ∀v ∈ M0,B
h
. (6.37)
∂x ∂x
Uniqueness for the solution uh of (6.37) follows from the same argument than for
the problem (6.18), while existence follows from finite dimensionality.
With the argument given for the wave propagation problem in Section (6.4) it can
be seen that
dim M0,B
h
= Nx .
Since the functions ϕk , 2, ≤ k ≤ Nx + 1 given in (6.20) belong to the space M0,B
h and
are linearly independent, they are a basis for the finite element space M0,B
h . Hence,
M0,B
h
= Span{ϕk , 2, ≤ k ≤ Nx + 1}.
Nx +1
uh = ∑ u jϕ j
j=2
∂ uh ∂ ϕk
−ω 2
ρ u , ϕk
h
+ E , = − Δ P, ϕk · ν Γ T , (6.38)
∂x ∂x
k = 2, · · · , Nx + 1.
To obtain the linear system associated with (6.38), we proceed as in the wave prop-
agation example, computing all integrals in (6.38) assuming piecewise constant co-
efficients.
In this fashion, we obtain the following tridiagonal linear system to be solved for
each ω > 0:
hk−1 1
−ω ρk−1
2
− Ek−1 uk−1
6 h k−1
138 6.12 A numerical example to determine the stiffness p 33
hk−1 hk
+ −ω ρk−1
2
+ ρk (1 − δk,Nx +1 )
3 3
1 1
+ Ek−1 + Ek (1 − δk,Nx +1 ) uk
h k−1 hk
2 hk 1
+ −ω ρk (1 − δk,Nx +1 ) − (1 − δk,Nx +1 )Ek uk+1
6 hk
= −Δ Pδk,Nx +1 , k = 2, · · · , Nx + 1.
Let the time constants in equation (6.39) be t1 = 0.16 s and t2 = 0.3 ms, so that the
quality factor of each single isotropic layer is nearly constant from about 10 Hz to
100 Hz.
Figure 6.3 shows the phase velocity v p (ω ) and the plot of 1000/Q p (ω ) with
v p (ω ) and Q p (ω ) computed using (6.35). The stiffness p33 (ω ) was determined us-
ing the finite element solution of the harmonic experiments and compared with the
corresponding analytical values of the BC theory in (6.30). A perfect fit can be ob-
served.
3800 30
Analytical
3750 Numerical
Compressional Wave Velocity (m/s)
25
3700
Analytical
Numerical
1000/Qp
3650 20
3600
15
3550
3500 10
0 25 50 75 100 0 25 50 75 100
Frequency (Hz) Frequency (Hz)
(a) (b)
Fig. 6.3 Analytical and numerical phase velocity (a) and inverse quality factor (b) for compres-
sional waves as function of frequency for the equivalent viscoelastic medium.
Here we define some of the finite element spaces that will be used in the next chap-
ters to solve problems in two and three dimensions.
enough triangles with vertices touching the boundary, the error will be small for the
problems to be considered here. Thus we consider our domain to be represented as
Ω = ∪Ni=1 T i .
Let us define the first finite element space over the triangulation as follows:
ϕ j (x, y) = p j + q j x + r j y, j = 1, 2, 3.
The values of the polynomial ϕ j (x, y) at the three nodes of the triangle T are the
local degrees of freedom of the finite element. These values uniquely define the
polynomial ϕ j (x, y).
Thus, the equations
ϕk (a j ) = δk j , j, k = 1, 2, 3
define uniquely three linear functions ϕk (x, y), k = 1, 2, 3 over the triangle T . They
are a local basis over T .
Then for any given complex (or real) valued function g(x, y) ∈ C0 (Ω ) we can
define the Lagrange interpolant gI (x, y) constructing a basis φs , s = 1, · · · , M for M h
using these local base s.
For this purpose, we glue together the local bases ϕ ki and ϕsl of two adjacent
triangles T i and T l having a common side Γ il and a common node ail to construct
one basis element φil (x, y) associated with the node ail , which is a global degree of
freedom. This basis function is such that
φil (ail ) = 1,
and φil vanishes at all other nodes of T i and T l . Note that this basis has small sup-
port, a very convenient property for computational purposes.
For the case of triangle with a boundary node, we use as global basis only the
local basis associated with such node.
6 Solution of differential equations using the finite element method 141
Then if (as )1≤s≤M are all the global degrees of freedom associated with this par-
(1)
tition, and (φs )1≤s≤M the associated global basis, the interpolant gI (x, y) is defined
as
M
(1)
gI (x, y) = ∑ g(as )φs (x, y). (6.40)
s=1
Then by construction
(1)
gI (as ) = g(as ) ∀ s.
Also, if T i and T l are adjacent triangles with a common side Γ , gI |T i − gI |T l is a
linear function on Γ in a single variable that vanish on two points of the edge Γ , so
that
(1) (1)
gI |T i − gI |T l = 0, on Γ
(1) (1)
and consequently gI ∈ C0 (Ω ) ⊂ L2 (Ω ). Also, since derivatives of gI are piece-
(1)
wise constant over Ω , we conclude that gI ∈ H 1 (Ω ).
Now if u (x, y) = (ux (x, y), uy (x, y)) ∈ [C 0 (Ω )]2 represents the solid displacement
s s s
using the global basis (φs )1≤s≤M to interpolate the components usx and usy of us
as indicated in (6.40) for the scalar function g(x, y). In particular, by construction,
Π (1) us ∈ M h .
a3
a4 a3
a1 a2 a1 a2
(a) (b)
Fig. 6.4 Triangular (a) and rectangular (b) element and degrees of freedom for the discretization
each component of the solid displacement.
Figure (6.4) (a) shows the triangular element and its degrees of freedom associ-
ated with the finite element space M h .
142 6.13 2-D finite element spaces
Next we define a second finite element space over Ω using bilinear polynomials
over a partition of Ω into rectangles Ri of diameter bounded by h as follows:
Let
P1,1 = Span {1, x, y, xy}
and set
ψ j (x, y) = p j + q j x + r j y + s j x y, j = 1, 2, 3, 4.
ψk (a j ) = δk j , j, k = 1, 2, 3, 4
As in the triangular case, the global basis constructed in this fashion has small sup-
port. For boundary nodes, we use as global basis the local basis associated with such
node.
Also, by construction,
(1,1)
gI (as ) = g(as ) ∀ as .
(1,1) (1,1)
Thus, if Ri and Rl are adjacent rectangles with a common side Γ , gI |Ri − gI |Rl
is a linear function of a single variable on Γ vanishing on two points of Γ . Thus
6 Solution of differential equations using the finite element method 143
(1,1) (1,1)
gI |Ri − gI |Rl = 0, on Γ
(1,1)
and consequently gI ∈ C0 (Ω ).
(1,1)
Also, since derivatives of gI are piecewise constant over Ω , we conclude that
(1,1)
gI ∈ H 1 (Ω ).
Now for a solid displacement vector us (x, y) = (usx (x, y), usy (x, y)) ∈ [C 0 (Ω )]2
proceeding as before we can construct another interpolant of us as
(1,1) (1,1)
Π (1,1) us = (usx )I , (usy )I ,
i.e., we interpolate the components usx and usy of us as indicated in (6.42) for the
scalar function g(x, y).
Figure (6.4) (b) shows the rectangular element and its degrees of freedom asso-
ciated with the finite element space N h .
To state the interpolation error associated with the above definitions of Π (1) us
and Π (1,1) us we introduce the spaces [H 1 (Ω )]2 and [H 2 (Ω )]2 in two dimensions as
follows.
First, for any scalar function v(x, y) defined over Ω we define the norms
* *2 * *2
1/2
*∂v* *∂v*
v1 = |v| + ** ** + ** ** dx
2
, (6.43)
Ω ∂x ∂y
* *2 * *2 * 2 *2 * 2 *2 * 2 *2
1/2
* * *∂v* *∂ v* *∂ v* * ∂ v *
2 *∂v*
v2 = |v| + * * + ** ** + ** 2 ** + ** 2 ** + ** * dx .
Ω ∂x ∂y ∂x ∂y ∂ x∂ y *
Note that the function u(x, y) is interpolated using linear function on each element
of the partition, and derivatives of u(x, y) are interpolated by derivatives of the inter-
polants Π (1) or Π (1,1) , which are piecewise constants. Then if u(x, y) is sufficiently
smooth (us ∈ [H 2 (Ω )]2 ) and the partition of Ω is sufficiently regular, the following
estimates hold [Ciarlet, 1976, Becker et al., 1981]
The finite element spaces M h and N h defined above are useful to approximate the
solid displacements of a viscoelastic or poroelastic medium Ω , because displace-
ments are continuous across any interior interface Γ of the domain Ω , unless Γ is a
fracture, where displacements are discontinuous across Γ .
But for computing approximate solutions of Biot’s equations in 2-D or 3-D me-
dia, where we must approximate both the solid and fluid vectors, it must be done taking
into account that only the normal component of the fluid displacement is continuous
across any interior interface Γ .
Consequently the spaces M h and N h are not adequate to represent the fluid
displacement vector and other finite element spaces need to be introduced.
For this purpose, we define first a new space of vector functions H(div, Ω ) as
follows
Let us denote by Pk,l to the polynomials of degree not greater than k in x and not
greater than l in y.
Consider a partition T h of Ω into rectangles Ri of diameter bounded by h.
Then our first finite element space to approximate the fluid displacement u f in
Biot’s equations is defined as follows:
where
V jh = P1,0 (R j ) × P0,1 (R j ) . (6.53)
In (6.52) v|R j indicates the function v as seen from Ω j , and similarly for v|Rk .
Remark. The space V h is the vector part of the 2-D Raviart-Thomas-Nedelec
space of zero order [Raviart and Thomas, 1977].
6 Solution of differential equations using the finite element method 145
In (6.54) ν jk denote the unit outer normal on Γjk from R j into Rk , with similar
notation for νk j .
Thus to construct a basis for V h we will patch the local polynomials in adjacent
rectangles using (6.54).
First, let R is a rectangle of the partition and Γ any side of R with outer unit
normal νΓ . Note that any v ∈ V h is of the form
p j · ν jk = 1, on Γ jk , p j · νΓ = 0, Γ = Γ jk ,
pk · νk j = 1, on Γ jk , pk · νΓ = 0, Γ = Γ jk ,
Let (as )1≤s≤M be the set of global degrees of freedom chosen in this fashion, and
(φsx )1≤s≤M the associated basis functions in V h .
Next we proceed similarly for the y-component vy of v. Assume that by transla-
tion and scaling R j = [0, 1] × [0, −1] and Rk = [0, 1] × [0, 1], so that Γ2jk = [0, 1]. In
this case ν jk = (0, 1) and νk j = (0, −1). Then
pj · ν jk = 1, on Γ jk pj · νΓ = 0, Γ = Γ jk ,
pk · νk j = 1, on Γ jk pk · νΓ = 0, Γ = Γ jk ,
M N
Q(1,1) u f = ∑ uxf (as )φsx (x, y), ∑ uyf (bs )φsy (x, y) .
s=1 t=1
The following interpolation error estimates hold [Raviart and Thomas, 1977,
Nedelec, 1980]
u f − Q(1,1) u f 0 ≤ Ch u f 0 (6.55)
(1,1) f
u − Q
f
u H(div,Ω ) ≤ Ch u H(div,Ω ) .
f
Next we construct a finite element space to approximate the fluid displacement over
a partition of T h of Ω into triangles T i of diameter bounded by h. We first define
the set
S2 = (P0 , P0 ) + (x P0 , y P0 ), (6.56)
If follows immediately from (6.57) that for any side Γ of a triangle T ∈ T h with
unit outer normal νΓ and w ∈ S2
6 Solution of differential equations using the finite element method 147
w · νΓ (6.58)
is constant.
The finite element space over the partition T h of Ω into triangles is defined as
Next, thanks to (6.58), on each triangle T with sides Γ i and unit outer normals
νΓ i , i=1,2,3 the following three equations define uniquely three local bases ϕ j , j =
1, 2, 3 in the space S2 :
ϕ j · νΓ k = δ jk , j, k = 1, 2, 3.
Also, we choose as global degrees of freedom the common mid point of two adjacent
triangles.
As in the rectangular case, we construct a global basis for the space W h by patch-
ing two adjacent triangles. The argument is identical to the rectangular case and is
omitted.
a4
a1 a3
a1 a3
a2
a2
(a) (b)
Fig. 6.5 Triangular (a) and rectangular (b) elements and degrees of freedom for the discretization
of the fluid displacement.
Figure (6.5) shows the triangular (a) and rectangular (b) elements and its degrees
of freedom associated with the finite element spaces W h and V h , respectively.
Here we briefly describe the finite element spaces used later to represent solid and
fluid vectors in three dimensions.
148 6.14 3-D Finite element spaces
ϕ j (x, y, z) = a j + b j x + c j y + d j z, j = 1, 2, 3, 4.
The values of the polynomial ϕ j (x, y) at the four nodes of the tetrahedron T are the
local degrees of freedom and they uniquely define the polynomial ϕ j (x, y, z).
Thus, the equations
ϕk (a j ) = δk j , j, k = 1, 2, 3, 4 (6.59)
uniquely define four linear functions ϕk (x, y, z), k = 1, 2, 3, 4 over the tetrahedron T ;
they are a local basis over T .
Now for an element v = (vx , vy , vz ) ∈ M h a global basis
Q1 = Span (1, x, y, z, x y, x z, y z, x y z) .
Since
dim Q1 = 8,
for any 3-rectangle R in the partition we choose as local degrees of freedom the eight
vertices a j of R.
Next, for any polynomial ψk ∈ Q1 , the equations
6 Solution of differential equations using the finite element method 149
ψk (a j ) = δk j , j, k = 1, · · · , 8 (6.60)
uniquely define eight functions ψk (x, y, z), k = 1, . . . , 8, over the 3-rectangle R; they
are a local basis for R.
Now we define the finite element space over the 3-rectangular finite element
partition as
As in the case of tetrahedral elements, a global basis for the space N h in (6.61)
is constructed by patching together the local basis of the form in (6.60) of adjacent
tetrahedra for each component of an element v = (vx , vy , vz ) ∈ N h in (6.61).
Remark. Both M h and N h are conforming finite element spaces of [H 1 (Ω )]3 .
Figure (6.6) shows the 3-rectangular elements (a) and tetrahedral (b) to discretize
the solid displacement vector associated with the spaces N h and M h , respectively.
a4
a8 a7
a5 a6
a3
a1
a4 a3
a1 a2 a2
(a) (b)
Fig. 6.6 3-rectangular (a) and tetrahedral (b) elements and degrees of freedom for the 3-D solid
displacement discretization associated with the spaces N h and M h , respectively.
Next we define the finite element space to approximate fluid displacement vectors
in R3 , as a natural extension to those in the 2-D case.
Let
Note if v ∈ Y h , for any face Γ with unit outer normal νΓ of a 3-rectangle R of the
partition, v · νΓ is constant on Γ.
Hence as in the 2-D case, we use (6.54) to construct a global basis function for
Y h by patching the local basis of adjacent 3-rectangles, choosing the mid points of
the faces of such 3-rectangles as global degrees of freedom.
The construction of the interpolation operator Q(1,1,1) u ∈ Y h follows the ideas
to construct Q(1,1) in the 2-D case and is omitted.
Next we construct a finite element space to approximate the fluid displacement
over a partition T h of Ω into tetrahedra T i of diameter bounded by h. Let
Then we define
Non-conforming finite element spaces will be used to represent the solid displace-
ment vector when solving wave propagation problems in two and three dimen-
sions. This choice is explained by the fact that they induce less numerical dis-
persion as compared with the conforming spaces described in the previous section
[Zyserman et al., 2003, Zyserman and Gauzellino, 2005].
An additional advantage is that non-conforming spaces require less amount of
information to be passed among processors in any parallel implementation of the
algorithms to be described in the next sections.
6 Solution of differential equations using the finite element method 151
a4
a4
a6 a3
a1 a3 a1
a5
a2
a2
(a) (b)
Fig. 6.7 3-rectangular (a) and tetrahedral (b) elements and degrees of freedom for the 3-D fluid
displacement discretization associated with the spaces Y h and Z h , respectively.
Thus we include here the description of the non-conforming spaces in 2-D and 3-
D to be used. These spaces have been defined and analyzed in [Douglas Jr. et al., 1999,
Douglas Jr. et al., 2001].
First consider the case that the domain Ω ⊂ Rn , n=2,3, is partitioned into triangles
or tetrahedra T i , referred to a n-simplices, of diameter bounded by h. Let
Γ j = ∂ Ω ∩ ∂ T j, Γ jk = ∂ T j ∩ ∂ T k .
Let ξ j and ξ jk denote the mid points of the faces of Γ j and Γ jk , respectively.
Then we define the local finite element polynomials as
The local degrees of freedom are chosen to be the three centers of the sides of the
triangle T i for n=2, and the four centers of the faces of the tetrahedron T i for n = 3.
Then the finite element space defined over the partition of Ω in n-simplices is
defined as
A local basis can be constructed by requiring at each local basis function to take the
value one at one nodal point ξ j or ξ jk and and zero at all other nodal points.
Then a global basis for N C h is obtained patching together the local basis of sim-
plices having a common side (triangles) or a common face (tetrahedron). Boundary
nodes are treated as in the 2-D case.
152 6.15 Non-conforming finite element spaces to represent solid displacements in 2-D and 3-D
Note that the elements in N C h are functions in [L2 (Ω )]n , but since continuity
across the interfaces Γ jk is required only at the mid points ξ jk of inter-element faces
Γ jk , functions in N C h do not have first derivatives in [L2 (Ω )]n , so that N C h is
NOT a subset of [H 1 (Ω )]n .
In this case it is said that N C h is a non-conforming finite element space of
[H (Ω )]n .
1
Next, for any vector function u sufficiently smooth defined over Ω , we may con-
struct the interpolation operator Π u ∈ N C h using the values of u at the nodal points
and the global basis in N C h .
Since Π reproduces constants and linears on each element T i of the finite element
partition, the interpolation error in the L2-norm can be estimated as [Douglas Jr. et al.,
1999]
u − Π u0 ≤ Ch2 u2 . (6.64)
In (6.64) the [H 2 (Ω )]3 -space is defined for vector functions defined over Ω ⊂ R3
asking that all first and second derivatives on the three space variables of all com-
ponents of the vector be square integrable. Then the norm in [H 2 (Ω )]3 is defined
similarly to that defined in (6.43) for 2-D vector functions.
Next consider the case in which the domain Ω ⊂ Rn , n=2,3, is partitioned into n-
rectangles Ri .
Consider first the case n = 2 and the reference rectangle R = [−1, 1]2 and the
reference basis
Note that the space M C h defined in (6.68) is also a non-conforming finite space of
[H 1 (Ω )]n for the same reason that the space over n-simplices defined in (6.63).
A global basis for M C h and an interpolation operator can be constructed as
indicated for the case of n-simplices. Also, the interpolation error stated in (6.64)
for the n-simplices case is still valid for the n-rectangular case.
Chapter 7
Modeling Biot media at the meso-scale using a
finite element approach
7.1 Introduction
For each realization of the stochastic parameters, the complex moduli are ob-
tained by solving Biot’s equations in the space-frequency domain in the diffusive
range of frequencies using a finite element procedure.
Appropriate boundary conditions are used to represent compressibility and shear
laboratory experiments.
These complex moduli in turn allow us to obtain the corresponding equivalent
compressional and shear phase velocities and quality factors for each frequency and
realization.
The statistical properties of the computed equivalent phase velocities and inverse
quality factors are obtained by averaging over the realizations of the stochastic pa-
rameters. The Montecarlo realizations are stopped when the variance of the com-
puted quantities stabilizes at an almost constant value.
In this fashion we determine the complex and frequency dependent moduli in the
constitutive relations of an isotropic effective viscoelastic medium behaving in the
average as our original highly-heterogeneous Biot medium.
The numerical procedure presented can be regarded as an up-scaling method to
obtain the effect of the mesoscopic scale heterogeneities at the macro-scale.
For the compressional case the algorithm was validated comparing the results
with those obtained using a White’s model which is valid for alternating layers of
two fluid-saturated porous sandstones ([White, 1975]). For the shear case, the algo-
rithm was checked to yield the real shear modulus for the case of uniform either gas
or water saturation at the zero frequency limit.
Biot’s equations in the diffusive range (ignoring acceleration terms) are (see
Chapter 1, equations (1.79)-(1.80) )
∇ · τ (u) = 0, (7.3)
−1 f
iωηκ u + ∇p f (u) = 0, (7.4)
Denote by ν the unit outer normal on Γ and let χ be a unit tangent on Γ oriented
counterclockwise so that {ν , χ } is an orthonormal system on Γ .
For obtaining the complex plane wave modulus of our fluid-saturated porous
medium, let us consider the solution of (7.3)-(7.4) with the following boundary
conditions
For this set of boundary conditions the solid is not allowed to move normally to the
lateral and bottom boundaries, the fluid is not allowed to flow out of the sample,
a uniform compression is applied on the boundary Γ T and no tangential external
forces are applied on the boundary Γ .
These boundary conditions can be associated with a laboratory experiment for
a periodic sample obtained by a mirror reflection with respect to the x1 -axis of the
domain Ω , enclosing the periodic sample in a thin impermeable jacket and applying
a uniform compression on the boundary Γ T and its corresponding image boundary
after the indicated reflection. In the case of periodic layered media, this experiment
mimics exactly the one described in [White et al., 1975].
Denoting by V the original volume of the sample, its (complex) oscillatory vol-
ume change Δ V (ω ) allows us to define the equivalent undrained complex plane-
wave modulus Eu (ω ), by using the relation
Δ V (ω ) ΔP
=− , (7.9)
V Eu (ω )
valid for a viscoelastic homogeneous medium in the quasi-static case.
160 7.2 Determination of the complex P-wave and shear moduli of the equivalent viscoelastic medium
After solving (7.3)-(7.4) with the boundary conditions (7.5)-(7.8), the average
vertical displacement us,T
3 (ω ) of the vertical solid displacements u3 (x1 , L, ω ) on Γ
s T
is computed.
Then, for each frequency ω , the complex volume change produced by the com-
pressibility test can be approximated by
Δ V (ω ) ≈ Lus,T
3 (ω ),
σν = (0,−ΔPeiwt)
For obtaining the equivalent complex shear modulus of our fluid-saturated porous
medium, let us consider the solution of (7.3)-(7.4) with the following boundary
conditions
−τ (u)ν = g, (x1 , x3 ) ∈ Γ T ∪ Γ L ∪ Γ R ,
us = 0, (x, y) ∈ Γ B ,
u f · ν = 0, (x, y) ∈ Γ ,
where ⎧
⎪
⎪ (0, Δ G), (x1 , x3 ) ∈ Γ L ,
⎪
⎪
⎨
g = (0, −Δ G), (x1 , x3 ) ∈ Γ R ,
⎪
⎪
⎪
⎪
⎩
(−Δ G, 0), (x1 , x3 ) ∈ Γ T .
The change in shape of the rock sample allows to recover its equivalent complex
shear modulus μ c (ω ) by using the relation
ΔT
tg(θ (ω )) = , (7.11)
μ c (ω )
where θ (ω ) is the departure angle between the original positions of the lateral
boundaries and those after applying the shear stresses (see, for example, [Kolsky, 1963]).
Equation (7.11) holds for this experiment in a viscoelastic homogeneous media in
the quasi-static approximation.
Computing the average horizontal displacement us,T 1 (ω ) of the horizontal dis-
placements us1 (x1 , L, ω ) at the top boundary Γ T yields an approximation to the
change in shape suffered by the sample, given by
tg(θ (ω )) ≈ us,T
1 (ω )/L.
and the equivalent shear phase velocity vs (ω ) and (inverse) quality factor Qs (ω ) are
estimated using the relations
−1
1 1 Im(vsc (ω )2 )
vs (ω ) = Re , = . (7.12)
vsc (ω ) Qs (ω ) Re(vsc (ω )2 )
Figure 7.2 shows and illustration for the undrained shear experiment.
162 7.3 A variational formulation
σν = (0,−ΔTe iwt)
σν = (0,ΔTe iwt)
Fig. 7.2 The harmonic shear
test applied to a representative
sample of saturated poroe-
lastic material. The figure
shows the application of shear
stresses on the left, right, and
top boundaries. The solid is
not allowed to move on the
bottom boundary, and no fluid
is permitted to flow into or out
of the sample.
First we introduce the following functional spaces that are closed subspaces of the
spaces [H 1 (Ω )]2 and H(div; Ω ) defined in (6.44) and (6.50).
H 1,P (Ω ) = {v ∈ [H 1 (Ω )]2 : v · ν = 0 on Γ L ∪ Γ R ∪ Γ B },
1,T
H0,B (Ω ) = {v ∈ [H 1 (Ω )]2 : v = 0 on Γ B },
2
Then multiply equation (7.3) by a test function vs ∈ H 1,P (Ω ) and equation (7.4)
by a test function v f ∈ H0 (div; Ω ), integrate over Ω , use integration by parts and
apply the boundary conditions (7.5)-(7.8). Adding the resulting equations we see
that the solution u(P) = (u(s,P) , u( f ,P) ) ∈ V (P) of (7.3)-(7.4) and (7.5)-(7.8) satisfies
the weak form:
Λ (u(P) , v) = − Δ P, vs · ν Γ T , ∀ v = vs , v f ∈ V (P) , (7.13)
In (7.13) for f , g either scalar or vector functions and Γ ∗ any subset of Γ we use the
notation
f , gΓ ∗ = f g d(Γ ∗ ).
Γ∗
Also, the matrix D and the column vector ε̃ ((u)) in (7.14) are defined by
⎛ ⎞ ⎛ ⎞
λu + 2 μ λu B 0 ε11 (us )
⎜ λu λu + 2 μ B 0 ⎟ ⎜ ε33 (us ) ⎟
D=⎜ ⎝ B
⎟, ε (u) = ⎜
⎝ ∇·uf ⎠.
⎟ (7.15)
B M 0 ⎠
0 0 0 4μ ε13 (us )
Existence of the solution of the boundary value problem (7.3)-(7.4) and either
(7.5)-(7.8) or (7.11)-(7.11) and its variational formulations (7.13) or (7.16) will be
assumed.
The proof of uniqueness of the solution of the variational problems (7.13) and
(7.16) is given in Appendix 7.10.
N h,P
= {v : v|R j ∈ [P1,1 (R j )]2 , v · ν = 0 on Γ L ∪ Γ R ∪ Γ B } ∩ [C0 (Ω )]2 ,
h,T
N0,B = {v : v|R j ∈ [P1,1 (R j )]2 , v = 0 on Γ B } ∩ [C0 (Ω )]2 .
164 7.4 The finite element procedures
Remark. Note that N h,P and N0h,T ,B are conforming finite element spaces of
[H 1 (Ω )]2 . Also note that the boundary conditions (7.7)-(7.8) and (7.11)-(7.11) have
been imposed strongly on the spaces N h,P and N0,B h,T
, respectively.
To approximate the fluid displacement a closed subspace V0h of the finite element
space V h defined in (6.52) is used. It is defined as
V (h,P) = N h,P
× V0h , V (h,T ) = N0,B
h,T
× V0h .
The a priori error estimates for the procedures (7.17) and (7.18) can be found in
[Santos and Carcione, 2015]. The result is stated in the following theorem.
Theorem 7.2. Assume that the coefficients in the constitutive relations (7.1)-(7.2)
satisfy the conditions stated in (1.58). Then for any ω > 0 the solutions u(h,P) , u(h,T )
of problems (7.17) and (7.18) satisfy the following a priori error estimates
u(s,h, j) − u(s, j) 1 + u( f ,h, j) − u( f , j) 0 + ∇ · u( f ,h, j) − u( f , j) 0 (7.19)
# $
≤ C(ω ) h1/2 u(s, j) 3/2 + h u( f , j) 1 + ∇ · u( f , j) 1 , j = P, T.
7 Modeling Biot media at the meso-scale using a finite element approach 165
Remark. In (7.19) we are using the fact that the solution u(s, j) belongs to the
Sobolev space [H 3/2 (Ω )]2 of functions having 3/2 derivatives in Ω . We refer to
[Adams, 1975] for a precise definition of the definition of Sobolev spaces of frac-
tional order.
Concerning choice of the mesh size h, it has to be small enough so that diffusion
process associated with the fluid pressure equilibration is accurately resolved.
Recall that the diffusion length is given by the relation length [Mavko et al., 1998]
2πκ K f
Ld = . (7.20)
ηω
For practical purposes, we take h so that the minimum diffusion length is discretized
with at least 3 mesh points at the highest frequency, which is sufficient to represent
a (smooth) diffusion-type process.
Besides, the size of the rock sample is not arbitrary: it has to be big enough to
constitute a representative part of the Biot medium but, at the same time, it has to
be much smaller than the wavelengths associated with each frequency.
To find an upper bound for the side length L of the sample, it was checked that
the compressibility and shear tests applied to homogeneous samples of side length
L, composed of any of the different materials forming the heterogeneous medium,
give negligible attenuation and velocity dispersion at the highest frequency being
considered.
Field measurements show that permeability values in reservoir rocks have a high
degree of spatial variability and exhibit long range correlations. It is also known that
these permeability fluctuations are well described by stochastic fractals [Douglas Jr.
et al., 1997].
It is also the case that in hydrocarbon reservoirs, regions of non-uniform patchy
saturation occur at gas-oil and gas-water contacts. By using CT scans it is pos-
sible to visualize the fluid distribution and spatial heterogeneities in real rocks
[Cadoret et al., 1995].
These are two examples of highly heterogeneous saturated porous media where
the size of the heterogeneities is small as compared with the wavelengths of the fast
compressional and shear waves.
On the other hand, a precise knowledge of the spatial distribution of the rock
heterogeneities at mesoscopic scales is in general not avalaible. Instead, averages
and variances of the material properties may be known.
This fact suggest to represent the heterogeneities using stochastic functions with
given spectral density distributions. In this sense, to obtain significant values for the
effective complex moduli in highly-heterogeneous fluid-saturated porous media, the
numerical experiments are applied in a Montecarlo fashion.
166 7.5 A Montecarlo approach for stochastic fractal parameter distributions
Thus, the compressibility and shear harmonic tests are applied to representa-
tive volumes of bulk material containing stochastic heterogeneities characterized by
fractal spectral density distributions, and the boundary-value problems are solved
for each realization. The means and variances of the phase velocities and inverse
quality factors associated with the complex moduli are obtained by averaging over
realizations of the stochastic parameters, and they represent the statistical behaviour
of the response of the porous rocks under consideration.
Following this stochastic approach, any given spatial property in the solid or fluid
P(x1 , x3 ) is considered as a sample (or realization) taken at random from an ensem-
ble of a physically meaningful random function P(x1 , x3 , γ ), where γ is the random
variable. Using this concept of ensemble allows us to define the statistical proper-
ties of the stochastic process P(x1 , x3 , γ ) and of any computed (random) quantity
obtained as output of our model. For example, the ensemble mean of P(x1 , x3 , γ )
could be interpreted as the average of repeated measurements of the property.
In our case, a finite set of realizations P(x1 , x3 , γn ), n = 1, · · · NR , is generated
using a given spectral density distribution. Next, for each temporal frequency ω
and realization γn , the phase velocities and inverse quality factors β n (ω ), β =
v p , 1/Q p , vs , 1/Qs , are random functions and they are determined by taking spatial
averages of the computed displacements at certain locations of the domain, as ex-
plained in the derivation of formulas (7.10) and (7.12).
In this procedure we exchange ensemble averages of the computed displacements
by their spatial averages and consequently the ergodic hypothesis must be used.
This hypothesis ([Lumley and Panofsky, 1964]) states that the statistical struc-
ture of P(x1 , x3 , γ ) can be obtained by substitution of ensemble averages by spatial
averages, and is of common use in geophysical applications. Finally, the means and
variances of the random functions β n (ω ), n = 1, · · · , NR represent the statistical be-
haviour of the response of the fluid-saturated porous rocks under consideration.
The generation of these kind of heterogeneities involves the use of a stochastic
fractal field, based on the so-called von Karman self-similar correlation functions.
These models are widely used in the statistical characterization of heterogeneities
for different applications.
Following [Frankel and Clayton, 1986, Santos et al., 2005a], we consider a par-
ticular case for which the spectral density of the stochastic field is given by:
1 NR n
β (ωm , NR ) = ∑ β (ωm ),
NR n=1
β = v p , vs , 1/Q p , 1/Qs ,
NR
1
σβ2 (ωm , NR ) = ∑
(NR − 1) n=1
[β n (ωm ) − β (ωm , NR )]2 .
The Montecarlo simulations were stopped when the variance (7.22) of the computed
quantities stabilized at an approximate constant value defined in terms of a given
tolerance.
To validate the compressibility test, the P-wave phase velocities and inverse of qual-
ity factors are computed for the case in which the sample is a periodic medium
consisting of alternating layers of equal thickness 0.2 m saturated with either gas or
brine.
The physical properties of the solid matrix are taken constant in all the domain,
and correspond to the sandstone 1 in Table 7.1 while the physical properties of the
fluids (brine and gas) are given in Table 7.2.
Phase velocities and inverse quality factors were obtained using the finite ele-
ment approach for 15 frequencies between 0 and 100 Hz and the corresponding
theoretical values calculated as indicated in Appendix 7.10 using White’s theory
[White et al., 1975]. For the theoretical values the domain is a periodic medium
composed of alternating layers of equal thickness 0.4 m saturated with either gas
or brine. This comparison is valid because the boundary conditions (7.5)-(7.8) for
the compressibility test can be associated with a compression similar to that pro-
posed by [White et al., 1975], but applied to a periodic sample obtained by a mirror
reflection of the domain Ω with respect to the x1 -axis.
As can be observed in Figure 7.3, the computed values are in excellent agreement
with those predicted by White’s theory.
168 7.6 Validation of the finite element procedure
2.66 0.07
2.64
0.06
2.62
0.05
2.6
Vp (Km/s)
2.58 0.04
1/Qp
2.56 0.03
2.54
0.02
2.52
0.01
2.5 Compressibility test Compressibility test
White model White model
2.48 0
0 20 40 60 80 100 0 20 40 60 80 100
Frequency (Hz) Frequency (Hz)
(a) (b)
Fig. 7.3 P-wave phase velocity (a) and inverse quality factor (b) obtained from the compressibility
test (dots) and using White’s theory (line) for frequencies lying between 0 and 100 Hz.
In the case of the shear modulus and for uniform either gas or brine saturation,
the finite element procedure was checked to yield the real shear modulus at the zero
limit frequency.
Table 7.1 Physical properties of the solid materials used in the numerical examples
Table 7.2 Physical properties of the fluids used in the numerical examples
Brine Gas
Kf 2.25 GPa 0.012 GPa
η 0.003 Pa · s 0.00015 Pa · s
In this section we use the time-harmonic tests to analyze the mesoscopic effects in
samples of Biot media having different kinds of heterogeneities in the rock and fluid
properties.
( j)
is fully saturated with gas, while if Sw > S∗ it is consider that R j is fully saturated
with brine.
In this way, a multi-scale gas-brine patchy-saturation model is constructed and
an overall brine saturation Sw is obtained for the synthetic rock sample.
In the examples fractal dimension is D = 2.2, the self-similarity coefficient is
H = 0.8 and correlation lengths are CL = 10 cm and CL = 5 cm.
Figure 7.4 displays the gas-brine distribution for a given realization (seed) and
correlation lengths CL = 10 cm (a) and CL = 5 cm (b). White zones correspond to
pure gas saturation and the black ones to pure brine saturation. The S∗ threshold
value was selected to have an overall gas saturation of 10 %.
50 50
Vertical side length (cm)
25 25
0 0
0 25 50 0 25 50
Horizontal side length (cm) Horizontal side length (cm)
(a) (b)
Fig. 7.4 Patchy gas-brine distribution for a given seed or realization for correlation lengths CL =
10 cm (a) and CL = 5 cm (b). White zones correspond to full gas saturation while black ones to
pure brine saturation. Overall gas saturation is 10 %. The sample is a square of side length 50 cm.
Figure 7.5 displays phase velocities and inverse quality factors for two different
realizations of the patchy saturation distributions, where the curves labeled Realiza-
tion 2 are associated with the patchy saturation distributions in Figure 7.4 (a), CL =
10 cm. It can be observed that phase velocities exhibits noticeable dispersion, while
quality factors Q p show strong mesoscopic attenuation effects, with values of Q p ≈
12 at about 60 Hz.
The next experiment analyzes the effect of varying the correlation length in the
patchy generation procedure. Figure 7.6 show the compressional phase velocity v p
(a) and inverse quality factor Q p (b) as function of frequency for correlation lengths
10 cm and 5 cm, indicated in the Figures by the labels CL = 10 cm and CL = 5 cm.
It is seen that as correlation length diminishes, the attenuation peak shifts to
higher frequencies. This effect can be explained by the fact that higher frequencies
see better the small-size gas patches associated with the shorter correlation length
CL = 5 cm, and consequently the relaxation frequency becomes higher.
7 Modeling Biot media at the meso-scale using a finite element approach 171
0.08
Realization 1
Realization 2
2.7 Realization 1
Realization 2
0.06
Vp (km/s)
1/Qp
2.6
0.04
2.5
0.02
2.4 0
1 10 100 1 10 100
Frequency (Hz) Frequency (Hz)
(a) (b)
Fig. 7.5 P-wave phase velocity v p (a) and inverse quality factor Q p (b) obtained from the com-
pressibility test for 2 seeds or realizations of the patchy saturation model. Overall gas saturation
is 10 %.
On the other hand, compressional phase velocities in Figure 7.6 are not sensitive
to this change in correlation length at low frequencies, while at higher frequencies
the differences are small, the order of 1.4 %.
0.08
2.7
CL = 10 cm CL = 10 cm
CL = 5cm CL = 5 cm
0.06
2.6
Vp (km/s)
1/Qp
0.04
2.5
0.02
2.4 0
1 10 100 1 10 100
Frequency (Hz) Frequency (Hz)
(a) (b)
Fig. 7.6 Compressional phase velocity v p (a) and inverse quality factor Q p (b) obtained from the
compressibility test for correlation lengths CL equal to 10 cm and 5 cm. Overall CO2 saturation is
10 %.
Figures 7.7 and 7.8 show the absolute value of the the fluid pressure distribution
at frequencies 10 Hz and 65 Hz, respectively, for the patchy distribution realization
in Figure 7.4(a) with correlation length CL = 10 cm.
It can be observed that pressure gradients take their highest values at the brine-
gas contact interfaces, with higher values at 65 Hz than at 10 Hz. This corresponds
to the higher mesoscopic attenuation values at 65 Hz as compared with those at 10
Hz observed in the black curve in Figure 7.6(b).
172 7.7 Analysis of the mesoscopic effects in heterogeneous Biot media
50 0.14
0.12
Vertical side length (cm)
0.1
25
0.06
0.04
0.02
0 0
Horizontal side length (cm)
Fig. 7.7 Absolute value of fluid pressure distribution (Pa) at 10 Hz and 10 % CO2 gas saturation.
The sample is a square of side length 50 cm. Correlation length is CL = 10 cm.
50 0.35
0.3
Vertical side length (cm)
0.25
Fluid Pressure (Pa)
0.2
25
0.15
0.1
0.05
0 0
Horizontal side length (cm)
Fig. 7.8 Absolute value of fluid pressure distribution (Pa) at 65 Hz and 10 % CO2 gas saturation.
The sample is a square of side length 50 cm. Correlation length is CL = 10 cm.
7 Modeling Biot media at the meso-scale using a finite element approach 173
The finite element procedures (7.17) and (7.18) were implemented to analyze the
response of a brine saturated rock sample with the solid matrix composed of two
different materials: shale and the sandstone 2 of Table 1. The domain Ω is a square
of side length 5 cm, and the partition T h is composed of 75 × 75 squares R j , so that
j
Ω = ∪Jj=1 R .
We assume a distribution of shale and sandstone in the form of irregular patches
using an stochastic fractal field based on the von Karman spectrum in (7.21).
The procedure to generate this rock sample is similar to that explained for the
case of patchy saturation. We assign to each sub-domain R j a pseudo-random num-
ber using a generator with uniform distribution. This random field is Fourier trans-
formed to the spatial wave-number domain and its amplitude spectrum is multi-
plied by the spectrum given in equation (7.21). The result is then transformed back
to the spatial domain, obtaining a micro-heterogeneous sandstone content model
S( j) , j = 1, · · · , J.
Next, to assign to the porous solid matrix in each R j pure shale or pure sandstone,
we choose a threshold value S∗ so that for each sub-domain R j where S( j) ≤ S∗ we
assume that such sub-domain matrix is pure shale, while if S( j) > S∗ we consider
that R j is pure sandstone.
In this way, a multi-scale highly-heterogeneous quasi-fractal model is con-
structed and an overall sandstone content S is obtained for the synthetic rock sample.
In this experiment we choose S∗ so that the overall sandstone content S is equal to
50 %.
We generated two realizations of the model for two different correlation lengths,
labeled as CL = 0.0025 cm and CL = 0.05 cm in the figures. The other parameters
of the fractal spectrum density are self-similarity coefficient H = 0.8 and fractal
dimension D = 2.2.
The distribution of shale and sandstone obtained in this fashion for the case of
correlation length CL = 0.05 cm is illustrated in Figure 7.9, where the black zones
correspond to pure shale while the white ones to pure sandstone.
These mesoscopic-scale heterogeneities produce non-negligible mesoscopic at-
tenuation effects as shown in Figure 7.10(b) for both correlation lengths. It can be
observed values of Qs of about 75 for frequencies between 20 and 40 Hz for both
correlation lengths. The attenuation peak moves to higher frequencies for the case
of the shorter correlation length. On the other hand, the shear phase velocity in Fig-
ure 7.10(a) shows very slight dispersion, with a 1.5 % increase between 0.3 and 100
Hz.
This mesoscopic attenuation effects are visualized in Figure 7.11 displaying the
absolute value of the fluid pressure. As in the case of patchy saturation, the pressure
gradients at the boundaries of the heterogeneities illustrate the induced fluid flow
mechanism.
174 7.7 Analysis of the mesoscopic effects in heterogeneous Biot media
2.5
0
0 2.5 5
Horizontal side length (cm)
Fig. 7.9 Fractal shale-sandstone 2 distribution. Black zones correspond to pure shale and white
ones to pure sandstone 2. The sample is a square of side length 5 cm. Correlation length CL is 0.05
cm.
0.015
CL= 0.025
0.05 cmcm
1.325 CL = 0.05
0.025cmcm
0.0125 0.05 cm
CL = 0.025 cm
CL = 0.05 cmcm
0.025
Shear Velocity (km/s)
1.3 0.01
1/Qp
0.0075
1.275
0.005
1.25
0.0025
1.225 0
1 10 100 1 10 100
Frequency (Hz) Frequency (Hz)
(a) (b)
Fig. 7.10 Shear phase velocity (a) and inverse quality factor (b) for correlation lengths CL equal
to 0.025 0.05 cm. The sample is a square of side length 5 cm. Shale percentage is 50 %.
7 Modeling Biot media at the meso-scale using a finite element approach 175
50 0.9
0.8
0.7
0.6
0.5
25
0.4
0.3
0.2
0.1
0 0
Horizontal side length (cm)
Fig. 7.11 Absolute fluid pressure distribution (Pa) at 30 Hz. Correlation lengths CL is 0.05 cm.
The sample is a square of side length 5 cm. Shale percentage is 50 %.
These lithological variations also induce wave-induce fluid flow and mesoscopic
loss effects in the case of the compressibility tests in the p33 -experiment. These ef-
fects are shown in Figure 7.12, with values of the quality factor Q p of compressional
waves of about 70 at 65 Hz. The corresponding compressional phase velocity shows
very slight dispersion, with an increase of 1.7 % between 0.1 and 100 Hz. The figure
is not included for brevity.
This result demonstrates that fast shear and compressional waves propagating
through Biot media with heterogeneities in the solid matrix are also affected by
wave-induce fluid flow effects.
0.015
1/Qs, CL = 0.05
1/Qp, CL = 0.05 cm
0.01
1/Q
Fig. 7.12 Shear inverse qual- 0.005
ity factors Qs and Q p for
correlation lengths is CL =
0.05 cm. The sample is a
square of side length 5 cm. 0
1 10 100
Shale percentage is 50 %. Frequency (Hz)
0.004
0.0035
0.0025
0.002
0.0015
0.001
2.7 0.1
0.09
2.65
0.08
0.07
2.6
0.06
Vp (Km/s)
1/Qp
2.55 0.05
0.04
2.5
0.03
0.02
2.45
0.01
2.4 0
0.1 1 10 100 0.1 1 10 100
Frequency (Hz) Frequency (Hz)
(a) (b)
Fig. 7.14 Average equivalent compressional phase velocity v p ((a), solid line) and inverse quality
factor Q p ((b), solid line) as function of frequency. Dotted lines indicate its corresponding standard
deviation interval.
Note also the lack of uncertainty at low frequencies, as expected from a viscoelastic-
type material ([Ferry, 1970]), since for long wavelengths the medium is seen as
homogeneous independently of the realization. As frequency increases, the shorter
wavelengths start to see the differences among realizations, with the increasing vari-
ances observed in Figure 7.14.
The effective shear modulus of this sample was checked to coincide numerically
with the the shear modulus of the dry matrix. This happens because shear tests ap-
plied on samples with fluid heterogeneities like patchy gas-brine saturation, induce
negligible fluid-pressure gradients and, consequently, negligible mesoscopic effects.
178 7.9 The algebraic problem associated with the determination of the plane wave modulus...
In this section we write the local equations associated with the computation of the
complex an frequency dependent plane wave modulus E u (ω ) using the finite ele-
ment method (7.17).
For simplicity in the explanation, let us consider a uniform partition of the square
domain Ω = [(0, L)]2 into squares Ω = [(0, h)]2 in the (x1 , x3 )-plane.
Take the x3 -axis pointing upwards and denote BL, BR, T R and T L to the nodes
in the corners (0, 0), (h, 0), (h, h) and (0, h), respectively.
Let us define the 4 local basis for each component of the solid vector us
x1 x3
ϕ BL (x1 , x3 ) = (1 − )(1 − ),
h h
x1 1 − x3
ϕ (x1 , x3 ) = ( )(
BR
),
h h
1 − x1 x3
ϕ T L (x1 , x3 ) = ( )( ),
h h
x1 x3
ϕ (x1 , x3 ) =
TR
,
h h
and the 4 local basis for the fluid u f
x1
ψ L (x1 , x3 ) = 1 − ,
h
x1
ψ R (x1 , x3 ) = ,
h
x3
ψ T (x1 , x3 ) = ,
h
x3
ψ B (x1 , x3 ) = 1 − .
h
Let us use the notation us = (U1 ,U3 ), u f = (W1 ,W3 ) and let us write U1 ,U3 ,W1
and W3 in terms of the local basis as follows:
U1 (x1 , x3 , ω ) = uBL
1 (ω )ϕ (x1 , x3 ) + u1 (ω )ϕ (x1 , x3 ) +
BL BR BR
U3 (x1 , x3 , ω ) = uBL
3 (ω )ϕ (x1 , x3 ) + u3 (ω )ϕ (x1 , x3 ) +
BL BR BR
u3 (ω )ϕ (x1 , x3 ) + u3 (ω )ϕ T L (x1 , x3 ),
TR TR TL
The coefficients
1 (ω ), u1 (ω ), u1 (ω ), u1 (ω ),
uBL BR TR TL
3 (ω ), u3 (ω ), u3 (ω ), u3 (ω ),
uBL BR TR TL
Also,
∂ ϕ BL (x1 , x3 )
ε11 ((ϕ BL (x1 , x3 ), 0)) = ,
∂ x1
1 ∂ ϕ BL (x1 , x3 )
ε13 ((ϕ BL (x1 , x3 ), 0)) = ,
2 ∂ x3
ε33 ((ϕ BL (x1 , x3 ), 0)) = 0,
∂ ϕ BL (x1 , x3 )
∇ · ((ϕ BL (x1 , x3 ), 0)) = .
∂ x1
∂ ϕ BL ∂ ϕ BL ∂ ϕ BL ∂ ϕ BL
(λu + 2 μ ) , + μ , uBL
∂ x1 ∂ x1 ∂ x3 ∂ x3 1
∂ ϕ BR ∂ ϕ BL ∂ ϕ BR ∂ ϕ BL
+ ( λu + 2 μ ) , + μ , uBR
∂ x1 ∂ x1 ∂ x3 ∂ x3 1
∂ ϕ T R ∂ ϕ BL ∂ ϕ T R ∂ ϕ BL
+ ( λu + 2 μ ) , + μ , uT1 R
∂ x1 ∂ x1 ∂ x3 ∂ x3
∂ ϕ T L ∂ ϕ BL ∂ ϕ T L ∂ ϕ BL
+ ( λu + 2 μ ) , + μ , uT1 L
∂ x1 ∂ x1 ∂ x3 ∂ x3
∂ ϕ BL ∂ ϕ BL ∂ ϕ BL ∂ ϕ BL
+ λu , + μ , uBL (7.24)
∂ x3 ∂ x1 ∂ x1 ∂ x3 3
∂ ϕ BR ∂ ϕ BL ∂ ϕ BR ∂ ϕ BL
+ λu , + μ , uBR
∂ x3 ∂ x1 ∂ x1 ∂ x3 3
180 7.9 The algebraic problem associated with the determination of the plane wave modulus...
∂ ϕ T R ∂ ϕ BL ∂ ϕ T R ∂ ϕ BL
+ λu , + μ , uT3 R
∂ x3 ∂ x1 ∂ x1 ∂ x3
∂ ϕ T L ∂ ϕ BL ∂ ϕ T L ∂ ϕ BL
+ λu , + μ , uT3 L
∂ x3 ∂ x1 ∂ x1 ∂ x3
∂ ψ L ∂ ϕ BL ψ R ∂ ϕ BL
+ B , W1L + B , W1R
∂ x1 ∂ x1 ∂ x1 ∂ x1
∂ ψ B ∂ ϕ BL ∂ ψ T ∂ ϕ BL
+ B , W3B + B , W3T
∂ x3 ∂ x1 ∂ x3 ∂ x1
≡ a11 uBL
1 + a12 u1 + a13 u1 + a14 u1 + a15 u3 + a16 u3 + a17 u3 + a18 u3
BR TR TL BL TR TR TL
4 5
+a19W1L + a1,10W1R + a1,11W3B + a1,12W3T = − Δ P, (ϕ BL (x1 , x3 ), 0) · ν Γ T = 0.
Remark. In this equation the right-hand side vanishes since the normal compo-
nent of (ϕ BL (x1 , x3 ), 0) on the top boundary vanishes.
In (7.24) the unknowns are numbered in the form:
1 ,
1 → uBL 2 → uBR
1 , 3 → uT1 R , 4 → uT1 L ,
3 ,
5 → uBL 6 → uBR
3 , 7 → uT3 R , 8 → uT3 L ,
∂ ϕ BL (x1 , x3 )
ε33 ((0, ϕ BL (x1 , x3 ))) = ,
∂ x3
1 ∂ ϕ BL (x1 , x3 )
ε13 ((0, ϕ BL (x1 , x3 ))) = ,
2 ∂ x1
ε1 ((0, ϕ BL (x1 , x3 ))) = 0,
∂ ϕ BL (x1 , x3 )
∇ · (0, ϕ BL (x1 , x3 )) = .
∂ x3
Then the following equation is obtained:
∂ ϕ BL ∂ ϕ BL ∂ ϕ BL ∂ ϕ BL
λu , + μ , uBL (7.25)
∂ x1 ∂ x3 ∂ x3 ∂ x1 1
7 Modeling Biot media at the meso-scale using a finite element approach 181
∂ ϕ BR ∂ ϕ BL ∂ ϕ BR ∂ ϕ BL
+ λu , + μ , uBR
∂ x1 ∂ x3 ∂ x3 ∂ x1 1
∂ ϕ T R ∂ ϕ BL ∂ ϕ T R ∂ ϕ BL
+ λu , + μ , uT1 R
∂ x1 ∂ x3 ∂ x3 ∂ x1
∂ ϕ T L ∂ ϕ BL ∂ ϕ T L ∂ ϕ BL
+ λu , + μ , uT1 L
∂ x1 ∂ x3 ∂ x3 ∂ x1
∂ ϕ BL ∂ ϕ BL ∂ ϕ BL ∂ ϕ BL
+ ( λu + 2 μ ) , + μ , uBL
∂ x3 ∂ x3 ∂ x1 ∂ x1 3
∂ ϕ BR ∂ ϕ BL ∂ ϕ BR ∂ ϕ BL
+ ( λu + 2 μ ) , + μ , uBR
∂ x3 ∂ x3 ∂ x1 ∂ x1 3
∂ ϕ T R ∂ ϕ BL ∂ ϕ T R ∂ ϕ BL
+ ( λu + 2 μ ) , + μ , uT3 R
∂ x3 ∂ x3 ∂ x1 ∂ x1
∂ ϕ T L ∂ ϕ BL ∂ ϕ T L ∂ ϕ BL
+ ( λu + 2 μ ) , + μ , uT3 L
∂ x3 ∂ x3 ∂ x1 ∂ x1
∂ ψ L ∂ ϕ BL ∂ ψ R ∂ ϕ BL
+ B , W1L + B , W1R
∂ x1 ∂ x3 ∂ x1 ∂ x3
∂ ψ B ∂ ϕ BL ∂ ψ T ∂ ϕ BL
+ B , W3B + B , W3T
∂ x3 ∂ x3 ∂ x3 ∂ x3
≡ a51 uBL
1 + a52 u1 + a53 u1 + a54 u1 + a55 u3 + a56 u3 + a57 u3 + a58 u3
BR TR TL BL TR TR TL
Next, taking the test functions vs = (0, ϕ BR (x1 , x3 )), vs = (0, ϕ T R (x1 , x3 )), vs =
(0, ϕ T L (x1 , x3 )), v f = (0, 0) in (7.23) we get the sixth, seventh and eighth equations
with coefficients a6 j , a7 j , a8 j , j = 1, · · · , 12, defined similarly than those in (7.25), re-
placing the function ϕ BL appearing before the right parenthesis in all inner products
in (7.25) by ϕ BR , ϕ T R and ϕ T L , respectively.
Next take the test function vs = (0, 0), v f = (ψ L (x1 , x3 ), 0) in (7.23) and note that
∂ ψ L (x1 , x3 )
ε33 (vs ) = ε13 (vs ) = ε11 (vs ) = 0, ∇ · (ψ L , 0) =
∂ x1
to get the 9th equation:
∂ ϕ BL ∂ ψ L ∂ ϕ BR ∂ ψ L
B , u1 + B
BL
, uBR
∂ x1 ∂ x1 ∂ x1 ∂ x1 1
∂ ϕTR ∂ ψL ∂ ϕTL ∂ ψL
+ B , uT1 R + B , uT1 L
∂ x1 ∂ x1 ∂ x1 ∂ x1
∂ ϕ BL ∂ ψ L ∂ ϕ BR ∂ ψ L
+ B , u3 + B
BL
, uBR (7.26)
∂ x3 ∂ x1 ∂ x3 ∂ x1 3
182 7.9 The algebraic problem associated with the determination of the plane wave modulus...
∂ ϕTR ∂ ψL ∂ ϕTL ∂ ψL
+ B , u3 + B
TR
, uT3 L
∂ x3 ∂ x1 ∂ x3 ∂ x1
∂ ψL ∂ ψL
+ iω η κ −1 ψ L , ψ L + M , W1L
∂ x1 ∂ x1
∂ ψR ∂ ψL
+ iω η κ −1 ψ R , ψ L + (M , W1R
∂ x1 ∂ x1
∂ ψB ∂ ψL ∂ ψT ∂ ψL
+ M , W3B + M , W3T
∂ x3 ∂ x1 ∂ x3 ∂ x1
≡ a91 uBL
1 + a92 u1 + a93 u1 + a94 u1 + a95 u3 + a96 u3 + a97 u3 + a98 u3
BR TR TL BL TR TR TL
Taking the the test functions vs = (0, 0), v f = (ψ R (x1 , x3 ), 0) in (7.23) to get the
the 10th- equation replacing ψ L by ψ R in all inner products appearing before the
right parenthesis in (7.26).
Next take the test function vs = (0, 0), v f = (0, ψ B (x1 , x3 )) in (7.23) and note that
∂ ψ B (x1 , x3 )
ε33 (vs ) = ε13 (vs ) = ε11 (vs ) = 0, ∇ · (0, ψ B ) =
∂ x3
to get the 11th equation:
∂ ϕ BL ∂ ψ B ∂ ϕ BR ∂ ψ B
+ B , u1 + B
BL
, uBR
∂ x1 ∂ x3 ∂ x1 ∂ x3 1
∂ ϕTR ∂ ψB ∂ ϕTL ∂ ψB
+ B , uT1 R + B , uT1 L
∂ x1 ∂ x3 ∂ x1 ∂ x3
∂ ϕ BL ∂ ψ B
+ B , uBL
∂ x3 ∂ x3 3
∂ ϕ BR ∂ ψ B ∂ ϕTR ∂ ψB
+ B , uBR + B , uT3 R (7.27)
∂ x3 ∂ x3 3
∂ x3 ∂ x3
∂ ϕTL ∂ ψB
+ B , uT3 L
∂ x3 ∂ x3
∂ ψL ∂ ψB ∂ ψR ∂ ψB
+ M , W1 + M
L
, W1R
∂ x1 ∂ x3 ∂ x1 ∂ x3
−1 B B ∂ ψB ∂ ψB
+ iω η κ ψ , ψ + M , W3B
∂ x3 ∂ x3
−1 T B ∂ ψT ∂ ψB
+ iω η κ ψ , ψ + (M , W3T
∂ x3 ∂ x3
≡ a11,1 uBL
1 +a11,2 u1 + a11,3 u1 +a11,4 u1 +a11,5 u3 +a11,6 u3 +a11,7 u3 +a11,8 u3
BR TR TL BL TR TR TL
Finally, take the test function vs = (0, 0), v f = (0, ψ T (x1 , x3 )) in (7.23) and re-
place ψ B by ψ T in all inner products appearing before the right parenthesis in (7.27)
to get the 12th equation.
Remark. Only the test functions (0, ϕ T R ), and (0, ϕ T L ) will contribute to the
right-hand side of (7.23) when the element has the top face contained in the top
boundary Γ T of the sample Ω .
All inner products in the twelve equations above were computed assuming con-
stant coefficients on each computational cell.
Next, the assembly of the global matrix can be performed using a global num-
bering connectivity matrix as usually done when constructing the global matrix of
the linear system in any finite element code (see for example [Becker et al., 1981]).
The boundary-value problems formulated in this Chapter are associated with second-
order operators having boundary data in L2 (Ω ). Existence will be assumed for the
solution of these problems. Also, following [Dahlberg et al., 1988] we will assume
that u(s,P) , u(s,T ) ∈ [H 3/2 ]2 ; we refer to [Adams, 1975] for the definition of Sobolev
spaces H s (Ω ) for s any real number.
On the other hand, it will be assumed that u( f ,P) , u( f ,T ) ∈ H 1 (div; Ω). These as-
sumptions will be used in the proof of the theorem stated below.
(f)
Theorem 7.3. Assume that u(s,P) , u(s,T ) ∈ [H 3/2 (Ω )]2 , u f ∈ H 1 (div; Ω). Also as-
sume that the coefficients in the constitutive relations (7.1)-(7.2) satisfy the condi-
tions stated in (1.58), so that the matrix D in (7.15) is positive definite. Then for any
ω > 0 the solution of problems (7.13) and (7.16) is unique.
Proof: To analyze the uniqueness of the solution of (7.13), set Δ P = 0 and choose
v = u(P) in (7.13) to obtain the equation
iω ηκ −1 u( f ,P) , u( f ,P) + D
ε (u(P) ),e(u(P) ) = 0. (7.28)
e(u(s,P) ),e(u(s,P) ) = 0,
D (7.30)
where
184 7.10 Appendix 1. Uniqueness of the solution of the variational problems
⎛ ⎞
λu + 2 μ λu 0
= ⎝ λu λu + 2 μ 0 ⎠ ,
D
0 0 4μ
⎛ ⎞
ε11 (u(s,P) )
e(u(s,P) ) = ⎝ ε33 (u(s,P) ) ⎠ .
ε13 (u(s,P) )
which
Next, since the matrix D in (7.15) is positive definite, so it is the matrix D,
from (7.30) allow us to conclude that
ε11 (u(s,P) )0 = ε33 (u(s,P) )0 = ε13 (u(s,P) )0 = 0. (7.31)
∂ f (x3 ) ∂ g(x1 )
2 ε13 (u(s,P) ) = + = 0, a.e. in Ω,
∂ x3 ∂ x1
which in turn implies
∂ f (x3 ) ∂ g(x1 )
=− = C = constant a.e. in Ω.
∂ x3 ∂ x1
Hence,
H 3/2 (Ω ) → C0 (Ω )
(s,P) (s,P)
so that u1 , u3 are uniformly continuous functions on Ω . Consequently (7.32)
(s,P) (s,P)
holds for all (x1 , x3 ) ∈ Ω as uniformly continuous functions, and u1 , u3 have
unique extensions to ∂ Ω . Hence,
7 Modeling Biot media at the meso-scale using a finite element approach 185
(s,P) (s,P)
u1 (x1 , x3 ) = f (x3 ), u3 (x1 , x3 ) = g(x1 ) ∀ (x1 , x3 ) ∈ Ω . (7.33)
On the other hand, the boundary condition (7.7) tells us that the normal compo-
nents of the traces of u(s,P) vanish on Γ B ∪ Γ L , so that
(s,P) (s,P)
u1 (0, x3 ) = 0, u3 (x1 , 0) = 0. (7.34)
u( f ,T ) 0 = 0, (7.35)
(s,T ) (s,T ) (s,T )
ε11 (u )0 = ε33 (u )0 = e13 (u )0 = 0.
u(s,T ) 1 = 0. (7.36)
From (7.35) and (7.36) we see that uniqueness holds for the solution of (7.16). This
completes the proof.
where −1
p1 p2
E0 = + , (7.37)
Eu1 Eu2
is the Reuss average, with pl = dl /(d1 + d2 ), l = 1, 2. Omitting the subindex l for
clarity, we have for each medium
4
Eu = Ku + μ , (7.38)
3
where Ku is the undrained or closed bulk modulus given by equation (1.51).
Moreover,
αM
r=
Eu
is the ratio of fast P-wave fluid tension to total normal stress, with M is given by
(1.50). Besides
η κ −1 kd
I= coth
k 2
is an impedance related to the slow P wave,
iωη κ −1
k= (7.39)
KE
where the subindex 1 refers to water for a layered medium alternately saturated
with water and gas. At this reference frequency, the Biot slow-wavelength equals the
mean layer thickness or characteristic length of the inhomogeneities [Gurevich et al.,
1997].
7 Modeling Biot media at the meso-scale using a finite element approach 187
It can be shown [Johnson, 2001] that E(ω = 0) is equal to the plane-wave modulus
(7.38) for a fluid whose composite modulus is given by Wood’s law in equation
(7.41).
On the other hand, at high frequencies, the pressure is not uniform but can
be assumed to be constant within each phase. In such a situation Hill’s theorem
[Hill, 1964] gives the high-frequency limit E(ω = ∞) = E0 with E0 given in equa-
tion (7.37).
Chapter 8
The meso-scale. Fractures as thin layers in Biot
media and induced anisotropy
8.1 Introduction
Fractured hydrocarbon reservoirs have been the subject of interest in exploration and
production geophysics, since generally, natural fractures control the permeability
of the reservoir [Gurevich et al., 2009]. In geophysical prospecting and reservoir
development, knowledge of orientation, densities and sizes of fractures is essential
since these factors control hydrocarbon production [Gurevich, 2003].
This knowledge and analysis is also important in the monitoring of injected
plumes of CO2 in geological storage. There might be faults and fractures that can
leak CO2 to the surface [Picotti et al., 2012].
One of the important mechanisms of seismic attenuation in fluid-saturated porous
media is wave-induced fluid flow, by which the fast compressional and shear waves
are converted to slow (diffusive) Biot waves at mesoscopic-scale heterogeneities,
which are larger than the pore size but much smaller than the predominant wave-
lengths of the fast compressional and shear waves [Carcione and Picotti, 2006,
Carcione, 2014]. These effects were analyzed in detail in Chapter 7.
White et al. [White et al., 1975, White, 1975] were the first to introduce the
mesoscopic-loss mechanism in the framework of Biot theory considering alternat-
ing thin poroelastic layers with symmetry axes perpendicular to the layering plane
[Carcione and Picotti, 2006]. Gelinsky and Shapiro [Gelinsky and Shapiro, 1997]
obtained the relaxed and unrelaxed (real) stiffnesses of the equivalent poro-viscoelastic
medium to a finely layered horizontally homogeneous material.
Krzikalla and Müller [Krzikalla and Müller, 2011] combined the previous mod-
els to obtain the five complex and frequency-dependent stiffnesses of the equivalent
transversely isotropic viscoelastic (TIV) medium. Their approach assumes a 1-D
character of the fluid pressure equilibration process which generates diffusive modes
from the fast compressional wave, i.e., the fluid-flow direction is perpendicular to
the fracture layering. As a consequence, compressional waves travelling horizon-
tally or vertically or shear waves will generate fluid pressure in such a way as to
maintain its distribution. Thus, the model considers only one relaxation function,
corresponding to the symmetry-axis compressional wave stiffness. These assump-
tions fail for heterogeneous layers, where the propagation of waves may depend on
direction.
A planar fracture embedded in a fluid-saturated poroelastic background is a par-
ticular case of the thin layer problem, when one of the layers is very thin, highly
permeable and compliant. A dense set of horizontal fractures in a fluid-saturated
poroelastic medium behaves as a TIV medium when the average fracture distance is
much smaller than the predominant wavelength of the travelling waves. This leads
to frequency and angular variations of velocity and attenuation of seismic waves.
An analysis of wave anelasticity and anisotropy in fractured poroelastic rocks can
be found in [Carcione et al., 2013].
This Chapter presents a generalization of the time-harmonic compressibility and
shear tests presented in Chapter 7 to this type of fractured Biot media.
In [Picotti et al., 2010] and [Santos et al., 2012] similar experiments to the ones
to be described here were applied to determine long-wave equivalent media to finely
layered viscoelastic materials.
Among other authors using numerical simulations to analyze dispersion, attenu-
ation and anisotropy in Biot media we mention the works in [Saenger et al., 2007]
and [Wenzlau et al., 2010].
In the context of Numerical Rock Physics, we present and analyze a collection of
time-harmonic experiments defined on fluid-saturated isotropic poroelastic samples
having a dense set of horizontal fractures modeled as very thin layers. The experi-
ments take into account the effects of fractures and interlayer fluid flow occurring
at the meso-scale. Each experiment is defined at the continuous level as a boundary
value problem (BVP) defined in the space-frequency domain, with boundary con-
ditions representing compressibility and shear tests. These BVP’s are solved using
the finite element (FE) method.
To discretize each component of the solid and fluid displacement vectors we
employ the same finite element spaces used in Chapter 7 but modified according to
the different BVP’s to be defined.
8 The meso-scale. Fractures as thin layers in Biot media and induced anisotropy 191
The time-harmonic experiments allow to determine the five complex and fre-
quency dependent coefficients of the TIV medium equivalent to the original frac-
tured Biot medium.
The methodology is validated by comparison with the analytical solution pro-
vided in [Krzikalla and Müller, 2011] (included in Appendix 8.8) for the case of a
brine saturated homogeneous sample having a dense set of horizontal fractures.
Then, the procedure is applied in two cases for which no analytical solutions are
available, namely patchy brine-CO2 saturated samples and brine-saturated samples
with fractures having fractal variations in the petrophysical properties.
We analyze the effect of these fluid and frame heterogeneities on the attenuation
and dispersion of the quasi-compressional wave (qP wave), the vertically polarized
quasi-shear wave (qSV wave) and the horizontally polarized shear wave (SH wave).
Let us consider isotropic fluid-saturated poroelastic layers and let us = (us1 , us2 , us3 )
and u f = (u1f , u2f , u3f ) indicate the time Fourier transform of the displacement vector
of the solid and fluid relative to the solid frame, respectively.
On each plane layer n in a sequence of N isotropic layers, the frequency-domain
stress-strain relations are given in (7.1)-(7.2). Also, Biot’s equations in the diffusive
range, stated in the space-frequency domain are given by (7.3)-(7.4) and stated here
again for convenience:
∇ · τ (u) = 0, (8.1)
iωηκ −1 u f + ∇p f (u) = 0, (8.2)
σ11 (
us ) = p11 e11 (
us ) + p12 e22 (
us ) + p13 e33 (
us ), (8.3)
σ22 (
us ) = p12 e11 (
us ) + p11 e22 (
us ) + p13 e33 (
us ), (8.4)
σ33 (s
u ) = p13 e11 ( s
u ) + p13 e22 ( s
u ) + p33 e33 (us ), (8.5)
σ23 (
us ) = 2 p55 e23 (
us ), (8.6)
σ13 (
us ) = 2 p55 e13 (
us ), (8.7)
σ12 (
us ) = 2 p66 e12 (
us ). (8.8)
σ = P e(
us )
where
⎛ ⎞ ⎛ ⎞
p11 p12 p13 0 0 0 us )
e11 (
⎜ p12 p11 p13 0 0 0 ⎟ ⎜ e22 (us ) ⎟
⎜ ⎟ ⎜ ⎟
⎜ p13 p13 p33 0 0 0 ⎟ ⎜ e33 (us ) ⎟
P=⎜
⎜ 0
⎟, e( ⎜
u )= ⎜
s ⎟. (8.9)
⎜ 0 0 p55 0 0 ⎟⎟ us ) ⎟
⎜ 2e23 ( ⎟
⎝ 0 0 0 0 p55 0 ⎠ ⎝ 2e13 (
us ) ⎠
0 0 0 0 0 p66 us )
2e12 (
In (8.3)-(8.8) we have assumed a closed system. This can be done for the
undrained composite medium, for which the variation of fluid content ζ = −∇ · u f
vanishes. This approach provides the complex velocities of the fast qP, qSV and SH
modes. The pIJ are the complex and frequency-dependent Voigt stiffnesses to be de-
termined with the time-harmonic experiments. In the next sections we present and
analyze a FE procedure to determine the coefficients in equations (8.3)-(8.8) and the
corresponding phase velocities and quality factors. These properties, which depend
on frequency and propagation direction, are given in Appendix 8.9.
Denote by ν the unit outer normal on Γ and let χ be a unit tangent on Γ so that
{ν , χ } is an orthonormal system on Γ .
i) To determine the complex coefficient p33 let us consider the solution of (8.1)-
(8.2) in Ω together with the following boundary conditions
Denoting by V the original volume of the sample, its (complex) oscillatory volume
change, Δ V (ω ), allows to define p33 . This is achieved by identifying p33 with the
plane wave modulus E u (ω ) in (7.9). Then using (7.9) and the procedure given in
Chapter 7 determines p33 .
ii) To determine p11 , we solve (8.1)-(8.2) in Ω together with the boundary con-
ditions
where ⎧
⎨ (0, Δ G), (x1 , x3 ) ∈ Γ L ,
g = (0, −Δ G), (x1 , x3 ) ∈ Γ R , (8.21)
⎩
(−Δ G, 0), (x1 , x3 ) ∈ Γ T .
The stiffness p55 can be identified with μ c (ω ) in (7.11), so that is obtained as in-
dicated in Chapter 7 using the average horizontal displacement u1s,T (ω ) suffered by
the boundary Γ T and equation (7.11).
v) To determine p13 solve (8.1)-(8.2) in Ω with the boundary conditions
us · ν = 0, (x1 , x3 ) ∈ Γ L ∪ Γ B , (8.24)
u f · ν = 0, (x1 , x3 ) ∈ Γ . (8.25)
us ) = ∇ ·
Thus, in this experiment e22 ( u f = 0, and from (8.3) and (8.5) we get
where e11 and e33 are the (macro-scale) strain components at the right lateral side
and top side of the sample, respectively. Then from (8.26) and the fact that τ11 =
τ33 = −Δ P on Γ R ∪ Γ T (cf.(8.22)) it follows that σ11 = σ33 and we obtain p13 (ω )
as
p11 e11 − p33 e33
p13 (ω ) = .
e11 − e33
iv) To determine p66 , let us consider the solution of (8.1)-(8.2) in Ω with the
following boundary conditions
where
⎧
⎨ (Δ G, 0), (x1 , x3 ) ∈ Γ B ,
g2 = (−Δ G, 0), (x1 , x3 ) ∈ Γ T , (8.30)
⎩
(0, −Δ G), (x1 , x3 ) ∈ Γ R .
Then, we proceed as indicated for p55 (ω ).
The stiffnesses coefficients pIJ allow us to to calculate the wave velocities and
quality factors of the effective TIV medium as explained in Appendix 8.9.
Figure 8.1 shows an schematic representation of the five Numerical Rock Physics
experiments.
In order to write the variational formulation let us introduce the following closed
subspaces of [H 1 (Ω )]2 :
W11 (Ω ) = {v ∈ [H 1 (Ω )]2 : v · ν = 0 on ΓB ∪ ΓT ∪ ΓL },
W33 (Ω ) = {v ∈ [H 1 (Ω )]2 : v · ν = 0 on ΓL ∪ ΓR ∪ ΓB },
W13 (Ω ) = {v ∈ [H 1 (Ω )]2 : v · ν = 0 on ΓL ∪ ΓB },
8 The meso-scale. Fractures as thin layers in Biot media and induced anisotropy 195
Fig. 8.1 Figures 1(a)-1(e) illustrate the five experiments needed to compute the stiffnesses compo-
nents. In Figures 1(a) and 1(b) we show how to compute p33 and p11 using the boundary conditions
(8.10)-(8.13) and (8.14)-(8.17), respectively. On the other hand, using the boundary conditions
(8.18)-(8.20), we obtain the stiffness p55 by performing the experiment shown in 1(c) and the stiff-
ness p66 with the experiment shown in 1(e)(boundary conditions (8.27)-(8.29)). Finally, Figure 1(d)
displays the experiment to determine p13 using the boundary conditions (8.22)-(8.25).
196 8.4 A variational formulation
W55 (Ω ) = {v ∈ [H 1 (Ω )]2 : v = 0 on ΓB },
W66 (Ω ) = {v ∈ [H 1 (Ω )]2 : v = 0 on ΓL }.
Also, let
and for (I, J) = (1, 1), (3, 3), (1, 3), (5, 5), (6, 6) set
Similarly, we obtain the variational formulations for the other pIJ ’s:
• p(11) : find u(11) = (u(s,11) , u( f ,11) ) ∈ Z11 (Ω ) such that
Λ (u(11) , v) = − Δ P, vs · ν Γ R , ∀ v = vs , v f ∈ Z11 (Ω ). (8.32)
The proof of uniqueness of these variational problems (8.31), (8.32), (8.33), (8.34)
and (8.35) is identical to that given in Appendix 7.10 for the problems (7.13) and
(7.16) in Chapter 7. Existence for the solution of these five problems will be as-
sumed.
and let
be the space to approximate the fluid displacement vector. Here Ps,t denotes the
polynomials of degree not greater than s in x1 and not greater than t in x3 .
Then, for (I, J) = (1, 1), (3, 3), (1, 3), (5, 5), (6, 6) set
ZIJh (Ω ) = WIJh (Ω ) × V h (Ω ).
The interpolation error estimates (6.48)-(6.49) are still valid for the finite element
spaces WIJh (Ω ), (I, J) = (1, 1), (3, 3), (1, 3), (5, 5), (6, 6), while the interpolating er-
ror (6.55) still holds for the space V h (Ω ).
Now, we formulate the FE procedures to determine the stiffnesses pIJ ’s as fol-
lows:
• p33 (ω ): find u(h,33) = (u(s,h,33) , u( f ,h,33) ) ∈ Z33
h (Ω ) such that
Uniqueness for the FE procedures (8.36)-(8.40) follows from uniqueness for the
continuous case. Existence follows from finite dimensionality.
This section includes the theorem stating the a priori error estimates associated
with the FE procedures (8.36)-(8.40). The proof of the theorem is presented in
[Santos and Carcione, 2015].
Theorem 8.1. Assume that u(s,IJ) ∈ [H 3/2 (Ω )]2 , u( f ,IJ) ∈ H 1 (div; Ω) for (I, J) =
(1, 1), (3, 3), (1, 3), (5, 5), (6, 6). Also assume that the entries in the matrix D in
(7.15) satisfy the conditions in (1.58) so that D is positive definite. Then for any
ω > 0 the following a priori error estimate holds:
8 The meso-scale. Fractures as thin layers in Biot media and induced anisotropy 199
rg2 φ 3
κ= (8.42)
45(1 − φ )2
1. Figure 8.2 shows plots of the energy velocities (a) and the dissipation factors
(b) of qP and qSV waves at 300 Hz, respectively, where it can be observed a very
good agreement between the theoretical and numerical results. In all polar plots,
0 degrees and 90 degrees correspond to waves arriving parallel and normal to the
fracture layering, respectively.
Next we proceed to analyze the results for Cases 1, 2 and 3. Figure 8.3 shows
polar plots of the qP energy velocities (a) and dissipation factors (b) of qP waves at
50 Hz for Cases 1, 2 and 3.
Loss is negligible in Case 1 along the direction of the fracture plane. It can be
observed strong velocity and Q anisotropy, with higher attenuation for patchy brine-
CO2 saturation.
Energy losses are much higher for angles between 60 and 90 degrees, i.e., for
waves travelling in the direction incident normal to the fracture layering.
On the other hand, velocity anisotropy caused by the fractures in Cases 1 and 3
is enhanced for the case of patchy saturation, with lower velocities when fractures
are present. The velocity behaves isotropically in Case 2.
8 The meso-scale. Fractures as thin layers in Biot media and induced anisotropy 201
90 FE 90 FE
Theory Theory
4.0 100
60 60
qP
80
3.0 qP
Vez (km/s)
1000/Q (Z)
60
30 30
2.0
qSV
40
qSV
1.0
20
0 0
1.0 2.0 3.0 4.0 20 40 60 80 100
Vex (km/s) 1000/Q (X)
(a) (b)
Fig. 8.2 Polar representation of qP and qSV energy velocity vectors (a) and dissipation factors (b)
of qP and qSV waves at 300 Hz for Case 1. The solid lines indicate the theoretical values.
qP Waves qP Waves
90 1. Brine saturated medium with fractures 90 1. Brine saturated medium with fractures
2. Patchy saturated medium without fractures 2. Patchy saturated medium without fractures
3. Patchy saturated medium with fractures 3. Patchy saturated medium with fractures
4.0 140
60 60
120
3.0
100
1000/Q (Z)
Vez (m/s)
30 80 30
2.0
60
40
1.0
20
0 0
1.0 2.0 3.0 4.0 20 40 60 80 100 120 140
Vex (m/s) 1000/Q (X)
(a) (b)
Fig. 8.3 Polar representation of the qP energy velocity vectors (a) and dissipation factors (b) at 50
Hz for Cases 1, 2 and 3.
Figure 8.4 shows the fluid pressure distribution at frequencies 50 Hz (a) and 300
Hz (b), respectively, for Case 3 and compressions normal to the fracture layering
(i.e., the p33 experiment). It can be observed that pressure gradients take their high-
est values at the fractures, and at 300 Hz remain always higher than at 50 Hz.
The next experiment analyzes the validity of an approximation commonly used
to relate the different attenuation mechanisms for qP waves analyzed in Cases 1, 2
and 3.
202 8.7 Numerical experiments
100 100
0.3
0.4
Pf (Pa)
Pf (Pa)
Z (cm)
Z (cm)
80 0.25 80
0.3
0.2
60 60
0.15 0.2
40 40
0.1
0.1
20 20
0.05
0 0
20 40 60 80 100 120 140 160 20 40 60 80 100 120 140 160
X (cm) X (cm)
(a) (b)
Fig. 8.4 Fluid pressure distribution at 50 (a) and 300 Hz (b). Compressibility test for p33 for Case
3.
Denoting by QP1 , QP2 and QP3 the qP-quality factors associated with Cases 1, 2
and 3 , Figure 8.5 shows the approximate validity of the commonly used approxi-
mation for the dissipation factors for qP waves
Q−1 −1 −1
P3 = QP1 + QP2 (8.43)
120
1000/QP3
1000/QP1 + 1000/QP2
100
80
1000/Q
60
40
20
0
0 10 20 30 40 50 60 70 80 90
Phase angle (degrees)
Fig. 8.5 Test of the approximate validity of (8.43) for Cases 1, 2 and 3 at 300 Hz.
8 The meso-scale. Fractures as thin layers in Biot media and induced anisotropy 203
Figure 8.6 shows polar plots of the qSV energy velocity vectors (a) and dissipa-
tion factors (b) of qSV waves at 50 Hz for the three cases. Case 2 is lossless, while
for a fractured sample brine or patchy saturated (Cases 1 and 3), Q anisotropy is
strong with high attenuation for angles between 30 and 60 degrees.
qSV Waves
qSV Waves
90 1. Brine saturated medium with fractures
2. Patchy saturated medium without fractures 90 1. Brine saturated medium with fractures
3. Patchy saturated medium with fractures 2. Patchy saturated medium without fractures
4.0 3. Patchy saturated medium with fractures
60 80
60
3.0
60
Vez (km/s)
1000/Q (Z)
30 30
2.0 40
1.0 20
0 0
1.0 2.0 3.0 4.0 20 40 60 80
Vex (km/s) 1000/Q (X)
(a) (b)
Fig. 8.6 Polar representation of the qSV energy velocity vectors (a) and dissipation factors (b) at
50 Hz for Cases 1, 2 and 3.
Figure 8.7 displays polar plots of the SH energy velocity at 50 Hz. Velocity
anisotropy is observed to be induced by fractures (Cases 1 and 3), while patchy
saturation , as expected, does not affect the anisotropic behaviour of the SH veloci-
ties. On the other hand, Case 2 shows isotropic velocity, with higher velocity values
than for the fractured cases. Also, SH waves are lossless since p55 and p66 are real
[Carcione et al., 2013].
In Figure 8.8 we analyze the variations in attenuation of qP and qSV waves for
Case 3 due to changes in CO2 saturation, with attenuation coefficients computed at
300 Hz. For qP waves, an increase of CO2 saturation from 10% to 50% induces a
noticeable decrease in attenuation for angles close to the normal orientation of the
fractures. Furthermore, for qSV waves the same decrease in attenuation is observed,
but for angles between 30 and 60 degrees.
Next, we analyze the behaviour of waves as a function of frequency in the range 1
Hz - 1 kHz at 10% CO2 saturation. Figure 8.9 displays dissipation factors for Cases
1, 2 and 3 for waves parallel (‘11’ waves) and normal (‘33’ waves) to the fracture
layering, respectively, while Figure 8.10 shows the corresponding velocities.
Figure 8.9 indicates that ‘11’ waves for Case 1 (brine-saturated homogeneous
background with fractures) are lossless, while the cases of patchy saturation with
and without fractures suffer similar attenuation, though there is a change from lower
204 8.7 Numerical experiments
SH Waves
3.0
Vez (km/s)
30
2.0
1.0
120
60
100
1000/Q (Z)
1000/Q (Z)
80
30 30
40
60
40
20
20
0 0
20 40 60 80 100 120 140 20 40 60 80
1000/Q (X) 1000/Q (X)
(a) (b)
Fig. 8.8 Dissipation factors for qP (a) and qSV (b) waves at 300 Hz for Case 3 and 10% and 50 %
CO2 saturation.
to higher attenuation for the patchy saturated case with fractures (Case 3) at a fre-
quency of about 40 Hz.
On the other hand, the curves for ‘33’ waves displayed in Figure 8.9(b) show
much higher attenuation than those for ‘11’ waves for the three cases. The case of
patchy saturation with fractures (Case 3) is the one exhibiting the highest attenu-
ation values for all frequencies. Besides, the case of brine-saturated homogeneous
background with fractures (Case 1) shows a much higher attenuation than the non-
fractured patchy saturated case (Case 2) starting at a frequency of about 26 Hz,
8 The meso-scale. Fractures as thin layers in Biot media and induced anisotropy 205
while below such frequency the opposite behaviour is observed, though with quite
similar attenuation.
40
1. Brine saturated medium with fractures
35 2. Patchy saturated medium without fractures
3. Patchy saturated medium with fractures
30
25
1000/Q11
20
15
10
−5 1 2 3
10 10 10
Frequency (Hz)
(a)
150
1. Brine saturated medium with fractures
2. Patchy saturated medium without fractures
3. Patchy saturated medium with fractures
100
1000/Q33
50
0 1 2 3
10 10 10
Frequency (Hz)
(b)
Fig. 8.9 Dissipation factor of waves parallel to the fracture plane (‘11’ waves) (a) and of waves
normal to the fracture plane (‘33’ waves) (b) as function of frequency.
Figure 8.10 shows that for the case of brine-saturated homogeneous background
with fractures (Case 1), ‘11’ velocities are essentially independent of frequency. In
the case of patchy saturation with fractures (Case 3), velocities are always smaller
than in Case 1, i.e., the presence of CO2 patches induces a velocity decay in the
whole range of frequencies analyzed. On the other hand, velocities for patchy satu-
ration with or without fractures show a similar increasing behaviour with frequency.
Figure 8.10(b) also shows that ‘33’ waves have the higher frequency dependent be-
haviour for the case of a patchy saturated fractured medium (Case 3), with lower
values than in the other two cases and increasing behaviour after 10 Hz. When the
sample is brine saturated with fractures (Case 1), velocities are higher than in case
3. On the other hand, the patchy saturated case without fractures exhibits the high-
est velocity values with moderate increase with frequency. Thus, for ‘33’ waves
206 8.7 Numerical experiments
3.9
1. Brine saturated medium with fractures
2. Patchy saturated medium without fractures
3.8 3. Patchy saturated medium with fractures
3.7
3.6
Velocity (km/s)
4.0
3.9
3.8
3.7
3.6 1 2 3
10 10 10
Frequency (Hz)
(a)
3.0
1. Brine saturated medium with fractures
2.5 2. Patchy saturated medium without fractures
3. Patchy saturated medium with fractures
4.0
3.5
Velocity (km/s)
3.0
2.5
4.0
3.5
3.0
2.5 1 2 3
10 10 10
Frequency (Hz)
(b)
Fig. 8.10 Velocity of waves parallel to the fracture plane (‘11’ waves) (a) and waves normal to the
fracture plane (‘33’ waves) (b) as function of frequency.
with f (x, z) being the fractal representing the spatial fluctuation of the permeability
field, chosen to be of fractal dimension D = 2.2, correlation length 2 cm and average
permeability 0.25 Darcy in the background and 4.44 Darcy in the fractures. In (8.44)
log κ denotes the average value of the function log κ . Porosity was obtained using
the Kozeny-Carman relation
8 The meso-scale. Fractures as thin layers in Biot media and induced anisotropy 207
κ (1 − φ )2
= constant.
φ3
Thus, the heterogeneous sample was constructed as a fractal perturbation of the
sample in Case 1. Figure 8.11 shows the Lamé coefficient λ u of the brine saturated
fractal sample used in Case 4.
160 7
’lambda_global_gnu_2.dat’
140 6.5
120
6
100
λ u (GPa)
5.5
Z (cm)
80
5
60
4.5
40
4
20
3.5
20 40 60 80 100 120 140 160
X (cm)
Fig. 8.11 Lamé coefficient (GPa) for the brine-saturated fractal porosity-permeability sample of
Case 4.
Figure 8.12 compares the qP (a) and qSV (b) dissipation factors of this case at 50
Hz with those of Case 1, while Figure 8.12 (c) and (d) compares the corresponding
energy velocities.
As in the patchy saturation case, frame heterogeneities induce a noticeable in-
crease in Q anisotropy for qP waves for angles normal to the fracture plane and for
qSV waves for angles between 30 and 60 degrees. From Figure 8.12 (a) and (b) we
conclude that both qP and qSV wave attenuation are highly increased when frame
heterogeneities are present. Also, from Figure 8.12 (c) and (d) we see the expected
energy velocity reduction in the heterogeneous case, and that velocity anisotropy
is less affected by frame heterogeneities than Q anisotropy. Concerning SH waves,
they show anisotropy and a moderate reduction in velocity and negligible attenua-
tion in the heterogeneous case. The SH energy velocity is similar to that of Figure
8.7 in Case 1 and the plot is omitted.
208 8.8 Appendix 1. Mesoscopic-flow attenuation theory for anisotropic poroelastic media
90 1. Brine saturated medium with fractures 90 1. Brine saturated medium with fractures
4. Fractal porosity−permeability medium 4. Fractal porosity−permeability medium
with fractures with fractures
80 80
60 60
60 60
1000/Q (Z)
1000/Q (Z)
30 30
40 40
20 20
0 0
20 40 60 80 20 40 60 80
1000/Q (X) 1000/Q (X)
(a) (b)
qP Waves qSV Waves
90 1. Brine saturated medium with fractures 90 1. Brine saturated medium with fractures
4. Fractal porosity−permeability medium 4. Fractal porosity−permeability medium
with fractures with fractures
4.0 4.0
60 60
3.0 3.0
Vez (km/s)
Vez (km/s)
30 30
2.0 2.0
1.0 1.0
0 0
1.0 2.0 3.0 4.0 1.0 2.0 3.0 4.0
Vex (km/s) Vex (km/s)
(c) (d)
Fig. 8.12 Dissipation factors for qP (a) and qSV (b) waves at 50 Hz for Cases 1 and 4. Figures (c)
and (d) correspond to the energy velocities for the respective wave propagation modes.
White’s mesoscopic attenuation theory of interlayer flow [White et al., 1975] and
[Carcione and Picotti, 2006] describes the equivalent viscoelastic medium of a stack
of two thin alternating porous layers of thickness d1 and d2 , such that the period of
the stratification is d = d1 + d2 . The theory gives the complex and frequency depen-
dent stiffness p33 . White model has been generalized in [Krzikalla and Müller, 2011]
to anisotropic media, i.e., they have obtained the five stiffnesses of the equivalent
TIV medium, denoted by pIJ . The stress-strain relations is given by equations (8.3)-
8 The meso-scale. Fractures as thin layers in Biot media and induced anisotropy 209
(8.8) and
cIJ − crIJ
pIJ (ω ) = cIJ + [p33 (ω ) − c33 ],
c33 − cr33
where crIJ and cIJ are the relaxed and unrelaxed stiffnesses.
According to [Gelinsky and Shapiro, 1997] (their eq. (14)), the quasi-static or
relaxed effective constants of a stack of poroelastic layers are
cr66 = B∗1 = μ ,
6 7 6 72 6 7−1
λm μ λm 1 (B∗ )2
cr11 − 2cr66 = cr12 = B∗2 = 2 + + 6∗ ,
E Em Em B8
6 7 6 7−1 m ∗ ∗
λ 1 B B
cr13 = B∗3 =
m
+ 6∗7,
Em Em B8
6 7−1 ∗2
1 B
cr33 = B∗4 = + 7∗
Em B
26 7 6 8 7 6 7 3−1
1 α 2 EG −1
= − ,
Em Em MEm
cr55 = B∗5 = μ −1 −1 ,
6 7 6 7 6 7 6 7−1
α μ α λm 1
B∗6 = −B∗8 2 + ,
Em Em Em Em
6 7 6 7−1
α 1
B∗7 = −B∗8 ,
Em Em
26 7 6 7 6 7 6 7 3−1
1 α2 α 2 1 −1
B∗8 = + − ,
M Em Em Em
where
2 4
λm = Km − μ and Em = Km + μ
3 3
and we have also reported the notation of that paper for clarity. In the case of no
interlayer flow, i.e., the unrelaxed regime, the stiffnesses are
c66 = cr66 ,
6 7 6 72 6 7−1
(EG − 2μ )μ EG − 2μ 1
c11 − 2c66 = c12 = 2 + ,
E EG EG
6 7 6 7−1G
EG − 2μ 1
c13 = ,
EG EG
6 7−1
1
c33 = ,
EG
c55 = c55
r
EG = Em + α 2 M,
210 8.9 Appendix 2. Wave velocities and quality factors
where
αM
r=
EG
and
η ad iωη EG
I= coth , a= ,
κa 2 κ MEm
for each single layer.
The main assumption by [Krzikalla and Müller, 2011] is that the fluid-flow di-
rection is perpendicular to the fracture layering and that the relaxation behaviour is
described by a single relaxation function or stiffness, i.e., p33 (ω ). Thus the theory
is valid for plane layers and can not be used when 2-D or 3-D heterogeneities are
present.
where ρ is the average density, l1 = sin θ and l3 = cos θ are the directions cosines,
θ is the propagation angle between the wave-number vector and the symmetry axis,
and the three velocities correspond to the qP, qSV and SH waves, respectively. The
phase velocity is given by
−1
1
v p = Re ,
v
while the quality factor is given by
Re(v2 )
Q= ,
Im(v2 )
8 The meso-scale. Fractures as thin layers in Biot media and induced anisotropy 211
where v represents either vqP , vqSV or vSH . The energy-velocity vector of the qP and
qSV waves is given by
ve
= (l1 + l3 cot ψ )−1 ê1 + (l1 tan ψ + l3 )−1 ê3 .
vp
Re(β ∗ X + ξ ∗W )
tan ψ = ,
Re(β ∗W + ξ ∗ Z)
defines the angle between the energy-velocity vector and the z-axis,
√
β = A± √B,
ξ = ±pv A ∓ B,
B = p11 l12 − p33 l32 + p55 cos 2θ .
and
c66
tan ψ = tan θ
c55
([Carcione, 2014] ( eq. 1.148)), since p55 and p66 are real quantities.
In general, the phase velocity is related to the energy velocity by
v p = ve cos(ψ − θ ),
where ve = |ve |.
Chapter 9
Fractures modeled as boundary conditions in
Biot media and induced anisotropy
9.1 Introduction
The acoustic response of fractures within a Biot medium can be modeled using a set
of boundary conditions proposed in [Nakagawa and Schoenberg, 2007] to represent
fluid-solid interaction within a fracture and the effect of its permeability on seismic
wave scattering.
These boundary conditions impose: continuity of the total stress components,
discontinuities of pressure (or pressures) -proportional to averaged fluid velocities
across the fracture- and discontinuities of displacements-proportional to stress com-
ponents and averaged fluid pressures along the fracture. They allow to represent
wave-induced fluid flow, by which the fast waves are converted to slow (diffusive)
Biot waves when travelling across fractures (mesoscopic-loss) and heterogeneities
[Carcione and Picotti, 2006, Carcione, 2014].
As explained in Chapter 8, dense set of horizontal fractures in a fluid-saturated
poroelastic medium behaves as a TIV medium when the average fracture distance is
much smaller than the predominant wavelength of the travelling waves. This leads
to frequency and angular variations of velocity and attenuation of seismic waves.
u f = φ (θ ) (
u f − us ),
be the average relative fluid displacement per unit volume of bulk material, with
φ (θ ) denoting the effective porosity. Set u = (us , u f ) and note that
ξ = −∇ · u f ,
The coefficient μ (θ ) is equal to the shear modulus of the bulk θ -material, considered
to be equal to the shear modulus of the dry θ -material. Also
(θ ) (θ ) 2
λu = Ku − μ (θ ) ,
3
9 Fractures modeled as boundary conditions in Biot media and induced anisotropy 215
(θ )
with Ku being the bulk modulus of the saturated θ -material. The coefficients in
(9.1)-(9.2) can be determined as indicated in Chapter 1.
Biot’s equations in the diffusive range (ignoring acceleration terms) are (see
Chapter 1, equations (1.79)-(1.80))
∇ · τ (u) = 0, (9.3)
iωη κ −1 u f + ∇p f (u) = 0, (9.4)
where ω is the angular frequency, η is the fluid viscosity and κ is the frame perme-
ability.
where u(s,l) |Γ ( f ,l) denotes the trace of us as seen from R(l) , with similar significance
for u(s,l+1) |Γ ( f ,l) .
The following boundary conditions on Γ ( f ,l) , representing the approximate acous-
tic response of a fracture as a very thin, compliant and highly permeable layer, are
derived in [Nakagawa and Schoenberg, 2007] (their equation (52)):
[us · νl,l+1 ] = ηN (1 − α ( f ) B( f ) (1 − Π ))τ (u)νl,l+1 · νl,l+1 (9.5)
1 (l+1) (l)
−α ( f ) (−p f ) + (−p f ) Π , Γ ( f ,l) ,
2
s
u · χl,l+1 = ηT τ (u)νl,l+1 · χl,l+1 , Γ ( f ,l) , (9.6)
f
u · νl,l+1 = α ( f ) ηN −τ (u)νl,l+1 · νl,l+1 (9.7)
216 9.2 A fractured Biot’s medium
1 1
(l+1) (l)
+ (−p f ) + (−p f ) Π , Γ ( f ,l) ,
(
B 2f )
In (9.5)-(9.6) ηN and ηT are the normal and tangential fracture compliances, respec-
tively and
(f)
Km
α( f ) = 1 − ( f ) .
Ks
(f) (f) 4
The fracture dry plane wave modulus Hm = Km + μ ( f ) and the dry fracture
3
shear modulus μ ( f ) are defined in terms of the fracture aperture h( f ) and the fracture
compliances by the relations
h( f ) h( f )
ηN = (f)
, ηT = .
Hm μ( f)
κ( f)
κ( f ) = .
h( f )
Besides,
1/2
(1 + i) ω η ( f ) α ( f ) ηN tanh ε
ε= , Π (ε ) = ,
2 2 B( f ) κ( f ) ε
α ( f )M( f ) (f) (f) 4
B( f ) = (f)
, Hu = Ku + μ ( f ) .
Hu 3
(l+1) (l)
(−p f ) = (−p f ), Γ ( f ,l) , (9.14)
The analysis will be performed for the more general fracture boundary condi-
tions (9.5)-(9.10), with the analysis for the boundary conditions (9.11)- (9.16) being
a particular case.
Set
a11 = ηN 1 − α ( f ) B̃( f ) (1 − Π ) , a12 = α ( f ) ηN Π , (9.17)
α ( f ) ηN Π
a22 = , Θ = a11 a22 − a212 .
B̃( f )
Then from (9.5) and (9.7) we obtain
a22 a12
τ (u)νl,l+1 · νl,l+1 = us · νl,l+1 + u f · νl,l+1 , Γ ( f ,l) , (9.18)
Θ Θ
(l+1) (l) a12 a11
(−p f ) + (−p f ) = 2 us · νl,l+1 + 2 u f · νl,l+1 , Γ ( f ,l) . (9.19)
Θ Θ
Adding (9.8) and (9.19)
(l+1) a12 a11
(−p f )= us · νl,l+1 + u f · νl,l+1 (9.20)
Θ Θ
iω η ( f ) Π 1 (l+1) (l)
+ u + u · νl,l+1 , Γ ( f ,l) .
κ( f ) 4 f f
In the analysis that follows we will employ the more convenient form of the
boundary conditions given by equations (9.6), (9.9), (9.10), (9.18) (9.20) and (9.21).
The experiments to determine the stiffnesses pIJ defining the equivalent TIV medium
at the macro-scale are those used in Chapter 8 when fractures are modeled as thin
highly permeable and compliant layers. In the next section we describe the varia-
218 9.3 A variational formulation
tional formulations that include the fracture boundary conditions (9.6), (9.9), (9.10),
(9.18) (9.20) and (9.21).
Also, set
H01 (div; ∪l R(l) )= {v ∈ [L2 (Ω )]2 : v|R(l) ∈ [H 1 (R(l) ]2 : ∇· v ∈ H 1 (R(l) ),v · ν = 0 on Γ}.
Next, for (I, J) = (1, 1), (3, 3), (1, 3), (5, 5), (6, 6) let
To obtain the variational formulation associated with p33 , multiply equation (9.3)
by vs and equation (9.4) by v f with v = vs , v f ∈ Z33 (Ω ), integrate over Ω and
add the resulting equations. Next, use integration by parts on each rectangle R(l)
applying the fracture boundary conditions (9.6), (9.9), (9.10), (9.18), (9.20), and
(9.21) and the additional boundary conditions (8.10),(8.11) (8.12) and (8.13) to get
the weak form: find u(33) = (u(s,33) , u( f ,33) ) ∈ Z33 (Ω ) such that:
Λ (u(33) , v) ≡ iω μ (b) (κ (b) )−1 u( f ,33) , v f + ∑ τst (u(33) ), εst (vs ) (l)
l R
− p f (u(33) ), ∇ · v f ) (l)
+ # R
$ # $ # $
(s,33)
+∑ F u (s,33)
· νl,l+1 , u(s,33) · χl,l+1 , u f · νl,l+1 , (9.22)
l
5
vs · νl,l+1 , vs · χl,l+1 , v f · νl,l+1 Γ ( f ,l)
9 Fractures modeled as boundary conditions in Biot media and induced anisotropy 219
8 9
iω η ( f ) Π 1 ( f ,33,l+1)
+∑ u + u( f ,33,l)
· ν l,l+1 , v( f ,l+1)
+ v( f ,l)
· νl,l+1
κ( f ) 4
l Γ ( f ,l)
= − Δ P, v · ν Γ T ,s
∀ v = (v , v ) ∈ Z33 (Ω ).
s f
Then we can state the variational formulation (9.22) in the equivalent form: find
(33) (33)
u(33) = (us , u f ) ∈ Z33 (Ω ) such that:
Λ (u(33) , v) ≡ iω μ (b) (κ (b) )−1 u( f ,33) , v f + ∑ D(b)
ε (u(33) ), ε̃ (v) (9.25)
l R(l)
+ # $ # $ # $
+ ∑ F u(s,33) · νl,l+1 , u(s,33) · χl,l+1 , u( f ,33) · νl,l+1 ,
l
5
vs · νl,l+1 , vs · χl,l+1 , v f · νl,l+1 Γ ( f ,l)
8 9
iω η ( f ) Π 1 ( f ,33,l+1)
+∑ u +u( f ,33,l)
· νl,l+1 , v( f ,l+1)
+v( f ,l)
· νl,l+1
κ( f ) 4
l Γ ( f ,l)
220 9.4 The finite element method
= − Δ P, vs · ν Γ T , ∀ v = (vs , v f ) ∈ Z33 (Ω ).
In a similar fashion we can obtain a weak formulation for the problems associated
with the other pIJ ’s applying the corresponding boundary conditions and the fracture
boundary conditions (9.6), (9.9), (9.10), (9.18) (9.20) and (9.21) to get:
Remark.The functions g and g2 in (9.28) and (9.29) are defined in (8.21) and (8.30),
respectively.
Uniqueness of the solution of the variational problems (9.25), (9.26), (9.27),
(9.28) and (9.29) is demonstrated in Appendix 9.7.
I I
Ω f = ∪ j=1
f
Ω j, Ω N f = Ω \ Ω f = ∪ j=1
Nf
Ω j. (9.30)
In (9.30) I f is the number of Ω j ’s having one top or bottom side contained in some
fracture Γ ( f ,l) for some l in the range 1 ≤ l ≤ J ( f ) , while IN f is the number of all
Ω j ’s such that ∂ Ω j ∩ Γ f ,l = 0/ ∀ l.
Let
Also set
222 9.4 The finite element method
Next, for (I, J) = (1, 1), (3, 3), (1, 3), (5, 5), (6, 6) to determine the stiffness pIJ we
will employ the finite element space
h,N
ZIJh (Ω ) = VIJ f (Ω N f ) ∪ VIJh, f (Ω f ) × W h,N f (Ω N f ) ∪ W h, f (Ω f ) . (9.31)
Then, for (I, J) = (1, 1), (3, 3), (1, 3), (5, 5), (6, 6) let
h,N f
ΠIJh : [H 3/2 (∪l R(l) )]2 → VIJ (Ω N f ) ∪ VIJh, f (Ω f )
be the bilinear interpolant operators defined locally on each rectangle R(l) and asso-
ciated with the spaces VIJh .
Also, let
J ( f ) +1
ϕ − ΠIJ
h
ϕ 0 + h ∑ ϕ − ΠIJ
h
ϕ 1,R(l) ≤ Chs ϕ s , 1 ≤ s ≤ 3/2,
l=1
J ( f ) +1
ψ − Q ψ 0 ≤ Ch
h
∑ ψ 1,R(l) ,
l=1
J ( f ) +1
∇ · (ψ − Qh ψ )0 ≤ Ch ∑ ψ 1,R(l) + ∇ · ψ 1,R(l) .
l=1
Λ (u(h,13) , v) = − Δ P, vs · ν Γ R ∪Γ T , ∀ v ∈ Z13
h
(Ω ). (9.34)
Uniqueness for the FE procedures (9.32)-(9.36) can be shown with the same argu-
ment used for the continuous case. Existence follows from finite dimensionality.
Here we state the a priori error estimates for the FE procedures (9.32)-(9.36).
J ( f ) +1
≤C h ∑ u(s,IJ,l) 3/2,R(l) + u( f ,IJ,l) 1,R(l) + ∇ · u( f ,IJ,l) 1,R(l) .
l=1
Proof: The proof follows the arguments given in [Santos and Carcione, 2015],
modified to take into account the presence of fractures.
224 9.6 Numerical experiments
In this section we first validate the procedure to determine the stiffnesses pIJ at the
¨ 2011]
macro-scale by comparison with the analytical solution in [Krzikalla and Muller,
included in Appendix 8.8. For this purpose the corresponding energy velocities
and dissipation coefficients are determined as indicated in Appendix 8.9 (see also
[Carcione, 2014]).
Next, the complex stiffnesses pIJ (ω ), energy velocities and dissipation coeffi-
cients are determined for the case of patchy brine-gas saturation for which no ana-
lytical solutions are avalaible.
The material properties of background and fractures are given in Table 9.1. The
saturant fluid is brine, with properties given in Table 9.2.
The numerical samples were discretized with a 100 × 100 uniform mesh.
The first experiment (Figures 9.1 9.2 and 9.3) validate the FE procedure against
the analytical solution given in Appendix 8.8.
Figures 9.1 and 9.2 show polar plots of the energy velocity vector and dissipation
factors [(1000/Q)(sinθ , cosθ )] for qP and qSV waves as functions of the propa-
gation angle, while Figure 9.3 display a polar plot of the energy velocity vector for
9 Fractures modeled as boundary conditions in Biot media and induced anisotropy 225
90 Theory 90 Theory
FE FE
4 300
60 60
3
Vez (km/s)
1000/Q(Z)
200
30 30
2
100
1
0 0
1 2 3 4 100 200 300
Vex (km/s) 1000/Q(X)
(a) (b)
Fig. 9.1 Polar representation of the energy velocity (a) and dissipation factor
[(1000/Q)(sinθ , cosθ )] (b) of qP waves as function of the propagation angle. Frequency
is 30 Hz. The symbols correspond to the analytical values, while solid lines indicate the FE
experiments.
The second experiment considers the same sample but for full brine saturation,
full gas saturation and 10 % and 50 % patchy brine-gas saturation. Brine and gas
have properties given in Table 9.2. Frequency is 30 Hz and a 100 × 100 mesh was
employed.
Patchy gas-brine distributions were generated using e Von Karman spectrum den-
sity formula (7.21) and the procedure explained in Subsection 7.7.1. In this example
the parameter values in (7.21) were chosen to be Ne = 2, D = 2.2 and correlation
length CL equal to one percent of the domain size.
Following this procedure two different patchy gas-brine distributions of overall
10 % and 50 % gas saturations were generated.
Figures 9.4 and 9.5 display polar plots of energy velocity vectors and quality
factor of qP and qSV waves, respectively, as a function of the propagation angle
for 0 %, 10 %, 50 % and 100 % global gas saturations. Frequency is 30 Hz. Figure
226 9.6 Numerical experiments
90 Theory 90 Theory
FE FE
2.5 300
60 60
2
Vez (km/s)
1000/Q(Z)
200
1.5
30 30
1
100
0.5
0 0
0.5 1 1.5 2 2.5 100 200 300
Vex (km/s) 1000/Q(X)
(a) (b)
Fig. 9.2 Polar representation of the energy velocity (a) and dissipation factor
[(1000/Q)(sinθ , cosθ )] (b) of qSV waves as function of the propagation angle. Frequency is 30
Hz. The symbols correspond to the the analytical values, while solid lines indicate FE experiments.
90 Theory
FE
2
60
1.5
Vez (km/s)
30
1
0.5
0
0.5 1 1.5 2
Vex (km/s)
Fig. 9.3 Polar representation of the energy velocity of SH waves as function of the propagation
angle. Frequency is 30 Hz. The symbols correspond to the analytical values, while solid lines
indicate the FE experiments.
9.4 (a) indicates that velocity of qP waves decreases as gas saturation increases,
while Figure 9.4 (b) shows that qP anisotropy is enhanced by patchy saturation, is
maximum for qP waves arriving normally to the fracture layering and decreases as
gas saturation increases. Maximum attenuation occurs at 10 % gas saturation for all
angles.
For qSV waves, Figure 9.5 (a) shows that velocity decreases as gas saturation
increases, with different anisotropic behaviour depending on the value of gas satu-
9 Fractures modeled as boundary conditions in Biot media and induced anisotropy 227
90 0% Gas 90 0% Gas
10% Gas 10% Gas
4 50% Gas 600 50% Gas
60 60
100% Gas 100% Gas
3
Vez (km/s)
1000/Q(Z)
400
30 30
2
200
1
0 0
1 2 3 4 200 400 600
Vex (km/s) 1000/Q(X)
(a) (b)
Fig. 9.4 Polar representation of the energy velocity (a) and dissipation factor
[(1000/Q)(sinθ , cosθ )] (b) of qP waves as function of the propagation angle for full brine,
full gas, 10 % and 50 % patchy gas-brine saturation. Frequency is 30 Hz.
90 0% Gas 90 0% Gas
10% Gas 10% Gas
2.5 50% Gas 500 50% Gas
60 60
100% Gas 100% Gas
2 400
Vez (km/s)
1000/Q(Z)
1.5 300
30 30
1 200
0.5 100
0 0
0.5 1 1.5 2 2.5 100 200 300 400 500
Vex (km/s) 1000/Q(X)
(a) (b)
Fig. 9.5 Polar representation of the energy velocity (a) and dissipation factor
[(1000/Q)(sinθ , cosθ )] (b) of qSV waves as function of the propagation angle for full brine, full
gas, 10 % and 50 % patchy gas-brine saturation. Frequency is 30 Hz.
ration. Concerning the dissipation factor for qSV waves, Figure 9.5 (b) shows maxi-
mum attenuation at 10 % gas saturation, and decreasing anisotropy as gas saturation
increases. Besides, qSV anisotropy shows different behaviour for different values
of gas saturation. The patchy nature of the saturation breaks the symmetry of the
curves (see the cuspidal triangles), with the attenuation of the qSV wave having
higher values at different angles.
228 9.7 Appendix 1. Uniqueness of the solution of the variational problems
0 10 20 30 40 50 60 70 80 90 100
100 1.8
’Salida_presion’
90 1.6
80
1.4
70
1.2
Pf [Pa]
60
1
nz
50
0.8
40
0.6
30
0.4
20
0.2
10
0
10 20 30 40 50 60 70 80 90 100
nx
Fig. 9.6 Fluid pressure for normal compression to the fracture plane at 10 % patchy gas-brine
saturation. Frequency is 30 Hz.
On the other hand, the SH wave is lossless and energy velocity of SH waves is
unaffected by different values of gas saturation, the corresponding curves coincide
with that in Figure 9.3 and are ommited.
Figure 9.6 shows the fluid pressure distribution (in Pa) for compressions normal
to the fracture layering (p33 experiment) for 10 % patchy gas saturation, where the
higher pressure values occur at the fracture locations and the darker regions values
identify the gas patches. This figure illustrates the mesoscopic attenuation mecha-
nism affecting compressional and shear waves at the macro-scale and characterized
in the previous patchy-related experiments.
Remark. In the analysis that follows it will be assumed that for all l, u(s,33) ∈
(s,33)
[H 3/2(R(l) )]2 , u f ∈[H 1(R(l))]2, so that u(s,33) ·νl,l+1 ∈ H 1 (Γ ( f ,l))⊂L2 (Γ ( f ,l)),u( f ,33)·
νl,l+1 ∈ H 1/2 (Γ ( f ,l) ) ⊂ L2 (Γ ( f ,l) ). Thus, since vs ∈ H 1 (R(l) ), v f ∈ H0 (div; ∪l R(l) ),
vs · νl,l+1 ∈ H (1/2) (Γ ( f ,l) ) ⊂ L2 (Γ ( f ,l) ), v f · νl,l+1 ∈ H −(1/2) (Γ ( f ,l) ). Then all bound-
ary terms in (9.22) or (9.25) make sense either in L2 (Γ ( f ,l) ) × L2 (Γ ( f ,l) ) or in
9 Fractures modeled as boundary conditions in Biot media and induced anisotropy 229
(IJ)
H (1/2) (Γ ( f ,l) ) × H −(1/2) (Γ ( f ,l) ). Identical assumptions will be made on us and
(IJ)
u f for (I, J) = (1, 1), (1, 3), (5, 5), (6, 6).
Let us analyze the uniqueness of the solution of (9.25). Set Δ P = 0 and choose
v = u(33) in (9.25) to obtain the equation
(iω μ (b) (κ (b) )−1 u( f ,33) , u( f ,33) + ∑ D(b)
ε (u(33) ),
ε (u(33) )
l R(l)
+ # $ # $ # $
+ ∑ (FR + iFI ) u(s,33) · νl,l+1 , u(s,33) · χl,l+1 , u( f ,33) · νl,l+1 ,
l
# $ # $ # $,
u(s,33) · νl,l+1 , u(s,33) · χl,l+1 , u( f ,33) · νl,l+1 (9.37)
Γ ( f ,l)
8 9
(iω η ( f )(ΠR −iΠI ) 1 ( f ,33,l+1) ( f ,33,l)
+∑ u +u ·νl,l+1 , u( f ,33,l+1)
+u( f ,33,l)
·νl,l+1
l κ( f ) 4 ( f ,l)
Γ
= 0.
Take imaginary part in (9.37) and use that FI is nonnegative and ΠR > 0 to see
that
where
⎛ ⎞ ⎛ ⎞
(b) (b)
λu + 2μ (b) λu 0 ε11 (u(s,33) )
(b) = ⎜
D ⎝ (b)
λu
(b)
λu + 2μ (b) 0 ⎠ ,
⎟
(33)
ε (us ) = ⎝ ε33 (u(s,33) ) ⎠ .
0 0 4μ (b) ε13 (u(s,33) )
(b)
Next, take real part in (9.39) and use that f11,R > 0, ηT > 0 and that the matrix D
(b)
is positive definite (since D is positive definite) to conclude that
ε11 (u(s,33) )0,R(l) = ε33 (u(s,33) )0,R(l) = ε13 (u(s,33) )0,R(l) = 0, ∀ l. (9.40)
Now use (9.40), the uniqueness argument given in Appendix (7.10) and the Sobolev
embedding [Adams, 1975]
(l)
H 3/2 (R(l) ) → CB0 (R )
(s,33)
u1 (x1 , x3 ) = C(l) x3 + B(l) ,
(s,33) (l)
u3 (x1 , x3 ) = −C(l) x1 + B(l) , ∀(x1 , x3 ) ∈ R , ∀ l.
Next using the boundary condition (8.12) we see that A(l) = B(l) = C(l) = 0, so that
(l)
u(s,33) = 0, (x1 , x3 ) ∈ R ∀ l. (9.41)
Combining (9.38) and (9.41) we conclude that uniqueness holds for the solution
of (9.25). Uniqueness for the solution of (9.26) and (9.27) follows with the same
argument.
Let us analyze uniqueness for the solution of (9.28). Repeating the argument
given to show uniqueness for u(33) we obtain
Next, consider the rectangle R(1) , which bottom boundary is Γ B . Let u(s,55,l) ≡
(u(s,55) )|R(l) . Thanks to the boundary condition (8.19),
1/2
|v|R(1) = ∑ (1)
|ε jk (v)| dx1 dx3
2
j,k R
defines a norm in R(1) equivalent to the H 1 -norm [Ciarlet, 1976]. Thus from (9.42)
we see that
u(s,55) 1,R(1) = 0,
(1)
and since u(s,55) ∈ [H 3/2 (R(1) )]2 → [CB0 (R )]2 , we conclude that
(s,55) (1)
u1 = 0, (x1 , x3 ) ∈ R . (9.43)
In particular,
(s,55) (s,55)
u1 = u3 = 0, on Γ1,2 . (9.44)
From (9.42) and (9.43) we see that (9.28) for the choice v = u(55) reduces to
J( f ) + # $ # $,
∑ ( f11,R + i f11,I ) u(s,55) · νl,l+1 , u(s,55) · νl,l+1
Γ ( f ,l)
l=2
6 $7
1 # (s,55) $ #
+ u · χl,l+1 , u(s,55) · χl,l+1 = 0. (9.45)
ηT Γ ( f ,l)
Take real part in (9.45), use that f11,R > 0, ηT > 0 to get
9 Fractures modeled as boundary conditions in Biot media and induced anisotropy 231
# $ # $
u(s,55) · νl,l+1 = u(s,55) · χl,l+1 = 0, L2 (Γ 1,2 ). (9.46)
(2)
Since u(s,55) ∈ [H 3/2 (R(2) )]2 → [CB0 (R )]2 using (9.44) and (9.46) we conclude that
(s,55,2) (s,55,2)
u1 = u3 = 0, on Γ1,2 . (9.47)
Now using (9.47) we can apply in R(2) the argument given for R(1) to conclude that
(55,2) (2)
us = 0, (x1 , x3 ) ∈ R .
so that uniqueness holds for the solution of (9.28). Uniqueness for the solution of
(9.29) follows with identical argument.
Chapter 10
The macro-scale. Seismic monitoring of CO2
sequestration
Abstract Storage of carbon dioxide (CO2 ) in deep saline aquifers and aging oil
reservoirs is a valid alternative approach for reducing the amount of greenhouse
gases in the atmosphere. This procedure can be modeled by the combined use
of multi-phase fluid flow and wave propagation. The flow simulator is used to
model the CO2 injection, and seismic monitoring is applied to determine the spatio-
temporal distribution of CO2 after several years of injection. The simultaneous flow
of brine and CO2 is modeled with the Black-Oil formulation for two-phase flow in
porous media, while wave propagation is formulated using an isotropic viscoelastic
model. The CO2 saturation and pressure data computed with the flow simulator is
used in a petrophysical model to determine the complex bulk and shear moduli of
the formation. Wave propagation is performed using a finite element domain de-
composition procedure. The simulation is used to model CO2 injection and flow
and compute time-lapse seismograms corresponding to the Utsira aquifer at Sleip-
ner field with the objective of identifying the spatio-temporal distribution of CO2
after injection.
10.1 Introduction
Capture and storage of carbon dioxide in deep saline aquifers and aging oil reser-
voirs is a valid alternative approach for reducing the amount of greenhouse gases
in the atmosphere [Arts et al., 2008]. Saline aquifers are suitable as storage sites
due to their large volume and their common occurrence in nature. The first indus-
trial scale CO2 injection project is the Sleipner gas field in the North Sea, where
CO2 separated from natural gas, is being injected in the Utsira formation, a highly
permeable porous sandstone, 800 m below the sea bottom. Within the formation,
there are several mudstone layers which act as barriers to the vertical flow of
the CO2 . Injection started in 1996 at a rate of about one million tonnes per year
[Arts et al., 2008, Chadwick et al., 2005].
Numerical modeling of CO2 injection and seismic monitoring are important tools
to understand the long term behaviour after injection and to test the effectiveness of
CO2 sequestration. In [Carcione and Picotti, 2006] and [Carcione et al., 2012] seis-
mic modeling is applied to monitor the spatio-temporal distribution of CO2 assum-
ing known CO2 saturation maps after injection. Instead, this Chapter describes a
methodology to model the CO2 flow and monitor the storage combining numeri-
cal simulations of CO2 -brine flow and seismic wave propagation. A petrophysical
model of the Utsira formation is built based on fractal porosity and clay content,
taking into account the variation of properties with pore pressure and saturation
[Carcione et al., 2003]. This model includes embedded mudstone layers of very low
permeability where CO2 is accumulated, although upward migration also occurs.
The simultaneous flow of brine and CO2 is modeled with the Black-Oil formulation
for two-phase flow in porous media [Aziz and Settari, 1985], which uses the PVT
data as a simplified thermodynamic model [Hassanzadeh et al., 2008]. The pressure
map before the injection is assumed to be hydrostatic for which a reference porosity
map is defined. The permeability is assumed to be anisotropic and is obtained from
first principles as a function of porosity and grain sizes [Carcione et al., 2003].
The wave propagation simulator is based on an isotropic viscoelastic model that
considers dispersion and attenuation effects. The complex P-wave and S-wave mod-
uli are determined as follows: in the brine saturated mudstone layers and the base
and top of the Utsira sand we use a Zener model to represent the viscoelastic be-
haviour of the material [Carcione, 2014]; outside the mudstone layers, we use the
time-harmonic compressibility tests described in Chapter 7 to model P-wave atten-
uation. The complex shear modulus is determined using another mechanism related
to the P-wave mechanism [Carcione et al., 2012].
The CO2 saturation and pressure data computed with the flow simulator were
used in the petrophysical (poro-viscoelastic) model to determine the complex bulk
and shear moduli of the formation in order to compute the synthetic seismograms.
The methodology was used to model CO2 injection and flow and compute time-
lapse seismograms corresponding to the Utsira aquifer at Sleipner field. It was
possible to identify the spatio-temporal distribution of CO2 after its injection over
long periods of time. Attenuation and dispersion effects are clearly observed in the
recorded traces. The synthetic seismograms show the progressive increase in CO2
accumulations below the mudstone layers and the pushdown effect observed in field
data [Chadwick et al., 2009]. Better results are obtained by updating the petrophys-
ical properties (mainly porosity, permeability and dry-rock moduli). Since the ef-
fectiveness of the time-lapse seismic method depends on the survey plan and the
properties of the storage site [Chadwick et al., 2010], these simulations may be used
to optimize the type of sources and the number and location of receivers to perform
the seismic surveys.
10 The macro-scale. Seismic monitoring of CO2 sequestration 235
The simultaneous flow of brine and CO2 in porous media is described by the
well-known Black-Oil formulation applied to two-phase, two component fluid flow
[Aziz and Settari, 1985]. We identify the CO2 saturated aqueous phase (CO2 and
brine components, subindex b) with the oil phase and the CO2 phase (subindex g)
with the gas phase. In this way, the CO2 component may dissolve in the aqueous
phase but the brine component is not allowed to vaporize into the CO2 phase. The
differential equations are obtained by combining the mass conservation equations
with Darcy’s empirical law.
The mass conservation equations are:
Krb
vb = −κ (∇pb − ρb g∇D), (10.5)
ηb
where D indicates depth, generally identified with the coordinate x3 , and g is the
gravity constant. Also, pg , pb are the fluid pressures and κ is the absolute permeabil-
ity tensor, assumed to be diagonal κ = diag(κx , κy , κz ). For β = g, b, the functions
Krβ and ηβ are the relative permeability and viscosity of the β -phase, respectively.
Replacing equations (10.3)-(10.5) into equations (10.1)-(10.2) and dividing by
ρgSC and ρbSC , the following nonlinear system of partial differential equations is ob-
tained,
Krg Rs Krb qg
∇ · (κ ( (∇pg − ρg g∇D) + (∇pb − ρb g∇D))) + SC (10.6)
Bg ηg Bb ηb ρg
# S
Rs Sb $
g
∂ φ +
Bg Bb
= ,
∂t
# S $
b
∂ φ
Krb qb Bb
∇ · (κ (∇pb − ρb g∇D)) + SC = . (10.7)
Bb ηb ρb ∂t
Two algebraic equations relating the saturations and pressures, complete the sys-
tem:
Sb + Sg = 1, pg − pb = PC (Sb ),
ρ = (1 − φ )ρs + φ ρ f
is the bulk density and B(ω ) is defined as a generalization of the absorbing bound-
ary condition for elastic solids derived in (5.37) to frequency dependent Lamé coef-
ficients λ (ω ) and μ (ω ) as
238 10.3 A viscoelastic model for wave propagation
⎛ ⎞
λ (ω ) + 2 μ (ω )
⎜ 0 ⎟
⎜ ρ ⎟ v p (ω ) 0
B(ω ) = ⎝
μ (ω ) ⎠ = 0 vs (ω )
0
ρ
where v p (ω ), vs (ω ) the phase velocities of the compressional and shear waves at the
frequency ω as defined in (7.10)-(7.12).
The stress tensor σ (u) is defined in the space-frequency domain by
where ejk (u) denotes the strain tensor and δ jk is the Kronecker delta.
The Lamé coefficients λu (ω ) and μ (ω ) are complex and frequency dependent
and are determined as follows.
Let Km and μm denote the bulk and shear moduli of the dry matrix. In the brine
saturated mudstone layers and in the base and top of the Utsira sand we use a Zener
model to make the Gassmann bulk modulus Ku = Km + α 2 M and the shear modulus
μm complex and frequency dependent. For this purpose, we define quality factors
(1) (2)
QKu associated with Ku and Qs associated with μm as
(2) μm (1)
Qs = Q (10.11)
Km Ku
and proceed as in Appendix 1.9; (see (1.50) and (1.51) for the definition of Ku , α
and M).
Within the Utsira sandstone and outside the mudstone layers P-wave attenuation
due to wave induced fluid flow at mesoscopic scale is modeled by computing the
complex plane wave modulus
(1) Re(Ku (ω )
QKu = minω
Im(Ku (ω ))
where
4
Ku (ω ) = Eu (ω ) − μm .
3
10 The macro-scale. Seismic monitoring of CO2 sequestration 239
The Zener model and the time-harmonic experiments to determine the P-wave
modulus require the knowledge of the bulk modulus Ks and density ρs of the solid
grains, the bulk and shear moduli Km and μm as well as the porosity φ and per-
meabilities κx , κz of the solid matrix. They also need the fluid bulk modulus and
viscosity. The determination of these parameters is explained in 10.5.1.
where ·, · denotes the approximation of ·, · on the boundary faces by the mid-
point quadrature rule
u, vΓ = ∑(uv)(ξ j )|Γj |
j
Theorem 10.1. Let u and uh be solutions of (10.13) and (10.15), respectively. Then,
for sufficiently small h > 0, we have
In (10.16) and (10.17) C(ω ) is a computable constant depending only on ω and the
domain Ω .
with β jk being a complex matrix function defined on the interior boundaries Γjk
such that its real and imaginary parts are positive semi-definite and positive definite
matrices, respectively.
A weak form of (10.18)–(10.19) and (10.22)–(10.23) is stated as follows: for all
j, find u j ∈ [H 1 (Ω j )]2 such that
4 5
−ω 2 (ρ u j , v) j + (σ (u j ), e(v )) j + iω Bu j , v Γ
j
Since the object of the domain decomposition procedure is to localize the calcu-
lations, we define the iterative procedure at the differential level in the following
fashion: given u0j ∈ [H 1 (Ω j )]2 for all j, find unj ∈ [H 1 (Ω j )]2 such that
4 5
−ω 2 (ρ unj , v) j + σ (unj ), e(v ) j + iω Bunj , v Γ
j
Here P0 (Γjk ) are constant functions on Γjk . Note that L jkh and Lkhj are considered to
be distinct.
Also the space M Ch is localized by removing the constrain imposing continuity
at the mid points ξ jk of Γjk as follows:
Here for the red-black type, the red and black parts of sub-domains are given al-
ternatively such that Ω = [∪ j∈IR Ω j ] ∪ [∪ j∈IB Ω j ]. If, for { j, k} ⊂ IR or { j, k} ⊂ IB ,
Ω j ∩ Ω k = 0,
/ then Ω j ∩ Ω k consists of a common vertex.
The proof of the convergence of the iterative domain decomposition procedure
(10.26)-(10.27) is based on the assumption that the imaginary part of the matrix D
and the real part of the matrix B are positive definite and is given in [Ha et al., 2002].
Finally, let us indicate the changes needed to treat the case of larger sub-domains
Ω j . For simplicity let us assume that the finite element partition T jh associated with
each sub-domain Ω j is a subset of the global finite element partition T h of Ω into
Lj l
rectangles. Let Ω j = ∪l=1 R j be a partition on each Ω j into rectangles Rlj of size
j
bounded by h and denote by ξlm the midpoint of the common interface between the
adjacent rectangles R j and R j . The non-conforming finite element space M Chj used
l m
where γ l,m l
jk denotes any of the common sides of rectangles R j and Rm
k in the partitions
l,m
T jh and Tkh such that γ jk ⊂ Γjk . Then, we define
# $2
L h = η : η |Γjk = η jk ∈ Πl,m P0 (γ l,m
jk ) ≡ L jkh , ∀{ j, k} .
With this new definitions of the spaces M Chj , and L h , the definition of the iterative
domain decomposition procedure (10.26)-(10.27) remains unchanged.
Note that the procedure (10.26)-(10.27) is an iterative method that converges
to the solution of the global non-conforming finite element procedure (10.15)
[Ha et al., 2002].
tion Multiple Data) in which all processors execute the same instruction stream on
different parts of the data. Note that the tasks are equally distributed among the pro-
cessors (R11 , R12 , R21 and R22 ) and the communications can be synchronized among
a small group of them. MPI (Message Passing Interface) communication structure
has been used in all of the parallel implementations.
Fig. 10.1 Domain decomposition scheme in a single problem for a two-dimensional grid with
technique red-black type (a). The dashed lines represent the artificial boundaries between regions
assigned to each processor. Exchange of information among shaded cells with the arrows indi-
cating the data streams (b). A single row and column of cells per processor are involved in this
communication process.
Before presenting the performance of the algorithm, some concepts and attributes
related to parallel computing are discussed.
Execution wall-clock time is the term during which a serial computer program is
executing, Ts (N), N being the program size. Arithmetic operations are several orders
of magnitude faster than Input/Output (read/write) operations, then
In parallel implementation, the execution wall-clock time, Tp (N, P), is also a func-
tion of the number of processors, P. Since our algorithm starts running on the master
processor 0 and ends at the same processor with the writing task of the results, we
can express
1 P−1 i
Tp (N, P) = ∑ (Tarithm + TI/O
i
+ Tcomm
i
+ Tidlet
i
).
P i=0
i
where Tcomm i
and Tidlet are the elapsed time in inter-processor communications and
idle time due to some type of synchronization between tasks prior to the commu-
10 The macro-scale. Seismic monitoring of CO2 sequestration 245
Ts (N)
S(N, P) = .
Tp (N, P)
If S(N, P)=P, the speedup is linear. In general, the communication implies overhead
and the parallel program is slower than serial program. Nevertheless, sometimes
a speedup greater than P can be obtained when using P processors, which is called
super-linear speedup. In this case, the cache size is large enough to contain all work-
ing set and the memory access time decreases severely.
It is worth remembering that Amdahl’s law states a maximum value of speedup
is given by 1/S, where S is the fraction of code that can not be parallelized. In other
words, for every program there is a number optimal processor. From these ideas
arise the concepts of efficiency and scalability. The efficiency gives an account of
the good use of the parallel computing and it is defined by
S(N, P)
E(N, P) = , E(N, P) ≤ 1,
P
where E(N, P) = 1 is maximum efficiency. With respect to scalability, it is called
strong scaling if the total problem size stays fixed as more processors are added and
weak scaling if the problem size per processor stays fixed as more processors are
added. These concepts are also associated with the concept of granularity that is a
qualitative measure of the ratio of computation to communication. The codes have
times of computation and times of communication that are separated by synchro-
nization events.
Note that the performance indicators are clearly dependent on hardware environ-
ment and software aspects. The algorithm was written in Fortran language. All the
tests were run on an cluster that comprises 6 nodes; each one with two-socket quad-
Core Intel Xeon E5620@2.40GHz CPUs with hyper-threading (16 threads) and 32
Gbit RAM, interconnected with 1 Gbit Ethernet, where the numerical calculations
are performed, plus a Front End, which is a single server with single-socket quad-
Core Intel Xeon X3430@2.40GHz CPUs with 8 Gbit RAM, 2.5 T hard disks, where
the programs are compiled and data produced during the runs is stored.
The parallel performance is illustrated using weak and strong scaling in Figure
10.2. The graphics were obtained for the same model and a relative error of 10−4
246 10.5 Petrophysical, fluid flow and seismic data
at f0 = 60 Hz. Figure 10.2a shows linear (ideal) speedup together with the results
for two different grids. The size of the problem is kept constant while increasing
the number of processors (weak scaling). The fine and coarse grids have 1024 ×
1024 elements (4,198,400 unknowns) and 512 × 512 sub-domains (1,050,624 un-
knowns), respectively. Both cases do not exhibit linear speedup due to dominant
communications among the processors and the coarse grid shows excessive over-
head. Therefore, the algorithm is specially suited to solve large scale problems. Fig-
ure 10.2b displays the performance when the problem size increases proportional
to the number of processors, in this example 256 × 256 elements for each proces-
sor does not change using more processors (strong scaling). It is observed an 70 %
efficiency using 64 processors.
70
1
Linear speedup
60 Grid 512 x 512
Grid 1024 x 1024
0,8
50
40 0,6
Efficiency
S(N,P)
30
0,4
20
0,2
10
0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60 70
Number of processors Number of processors
(a) (b)
Fig. 10.2 Speedup curves. Grid with 512 × 512 sub-domains and grid with 1024 × 1024 sub-
domains (a). Efficiency. For a constant size of 256 × 256 elements per processor, the efficiency
remains higher than 70 % (b).
Concerning robustness, it can be said that if the number of elements and the num-
ber of points per wavelength are kept fixed for different frequencies, necessarily the
mesh size will change and, in this case, the number of iterations remains constant
through the range 1 Hz-10 kHz, which is the one of interest in geophysical applica-
tions.
In this section, we describe the procedure used to determine the petrophysical and
fluid-flow parameters needed in the Black-Oil and wave propagation seismic simu-
lators.
10 The macro-scale. Seismic monitoring of CO2 sequestration 247
The shaly sandstone model used here is that in [Carcione et al., 2000], where the
topology of the sandy and shaly mixture has been defined to obtain the petrophysical
properties of the Utsira formation.
The pressure dependence of properties is based on the following relationship
between porosity and pore pressure:
(1 − φc ) φ (t)
(p(t) − pH ) = φ0 − φ (t) + φc ln , (10.28)
Ks φ0
where
p(t) = Sb pb (t) + Sg pg (t)
is the pore pressure, φc is a critical porosity, φ0 = φ0 (x, z) is the initial porosity
at hydrostatic pore pressure pH and Ks is the bulk modulus of the solid grains
[Carcione et al., 2003]. Porosity φ0 is assumed to have a fractal spatial distribution
around the average porosity φ0 , obtained from the neutron log by using standard
methods. The rock is formed with quartz (bulk modulus of 40 GPa) and clay (bulk
modulus of 15 GPa). Ks is computed as the arithmetic average of the Hashin Shtrik-
man upper and lower bounds [Hashin and Shtrikman, 1963].
The relationship among horizontal permeability (κx1 ), porosity and clay content
(C) is [Carcione et al., 2003],
1 45(1 − φ (t))2 (1 −C)2 C2
= + 2 , (10.29)
κx1 (t) φ (t)3 R2q Rc
where Rq and Rc are the average radii of the sand and clay grains, respectively.
Also, as permeability is anisotropic, we assume the following relationship be-
tween horizontal and vertical permeability κx3 [Carcione et al., 2003]
∗
Sg − Sgc ng
Krg (Sg ) = Krg , (10.33)
1 − Sgc − Sbc
∗
1 − Sg − Sbc nb
Krb (Sg ) = Krb , (10.34)
1 − Sgc − Sbc
∗
Sg − Sgc nc
Pca (Sg ) = Pca , (10.35)
1 − Sgc − Sbc
∗ and K ∗ are the maximum
where ng , nb , nc determine the curvature of functions, Krg rb
∗
values of Krg , Krb and Pca is a reference value of the capillary pressure Pca . Sgc and
Sbc are the saturations at which the CO2 and brine phases become mobile, respec-
tively. The influence of capillary pressure on CO2 injection, storage and monitoring
is analyzed in detail in [Santos et al., 2014b].
The PVT data, Rs and Bb , can be expressed in terms of the equilibrium properties
obtained from an equation of state as presented in [Hassanzadeh et al., 2008] and
[Spycher and Pruess, 2005], i.e.,
ρ̃bSC χg ρbSC
Rs = , Bb = ,
ρ̃gSC (1 − χg ) ρb (1 − ωg )
where ρ̃bSC and ρ̃gSC are the brine and CO2 molar densities at standard conditions,
respectively, χg and ωg are the CO2 mole and mass fractions in the brine phase.
The molar density is related to the mass density through the molecular weight.
While the CO2 molecular weight (Mg ) and mass density at standard conditions
(ρgSC ) are known, these properties for the brine phase must be estimated. The brine
molecular weight (Mb ) is simply computed from the mole fractions χsalt of NaCl
and χH2 O of H2 O. The brine mass density at standard conditions, ρbSC , is esti-
mated following the Rowe and Chou correlation as given in formulas (B6)-(B14)
of [Hassanzadeh et al., 2008] . Once ρbSC is computed, ρb at reservoir conditions is
obtained using the approach of Garcı́a, as stated in [Hassanzadeh et al., 2008] . This
approach and the estimation of the CO2 mole and mass fractions in the brine phase,
χg and ωg , are briefly explained in Appendix 10.9.
Finally, the viscosity, density and bulk modulus of CO2 are obtained from the
Peng-Robinson equations [Peng and Robinson, 1976] as a function of temperature
and pore pressure.
10 The macro-scale. Seismic monitoring of CO2 sequestration 249
In (10.36) φ0 denotes the spatial average of φ0 (x1 , x3 ) and f (x1 , x3 ) is a fractal
field representing the spatial fluctuation of φ0 (x1 , x3 ), for which the spectral density
is given in (7.21). For this application we use Euclidean dimension Ne = 2, fractal
dimension D = 2.2 and φ0 = 36.7 %. The correlation length CL in (7.21) was taken
to be 2 % of the domain size.
At the beginning of the simulation of the CO2 injection, the horizontal and verti-
cal permeabilities were determined by using equations (10.29) and (10.30), consid-
ering an anisotropy parameter a = 0.1 and a fixed clay content C = 6 %.
The minimum, average and maximum porosities obtained are 32.5 %, 36.7 % and
38.3 %. The corresponding minimum, average and maximum vertical permeabilities
are 0.1 D, 0.12 D and 0.145 D.
The mudstone layers are not completely sealed, having constant porosity 24 %
and vertical permeability 0.033 D. Besides, they have openings, that give a path
for the upward migration of CO2 . The top and bottom of the Utsira formation have
constant porosity 22 % and vertical permeability 0.02 D.
The initial porosity and vertical permeability fields can be observed in Figures
10.3 and 10.4, respectively. Since before CO2 injection begins Sb = 1, it follows
from (10.30) that initial horizontal permeability values are ten times those of the
initial vertical permeability.
250 10.6 Numerical simulations
0.77 0.4
0.38
0.36
0.87
0.34
Depth (km)
0.32
porosity
0.97 0.3
0.28
0.26
1.07
0.24
0.22
1.17 0.2
0 0.3 0.6 0.9 1.2
Horizontal Distance (km)
0.77 160
140
100
Depth (km)
0.97 80
60
1.07 40
20
1.17 0
0 0.3 0.6 0.9 1.2
Horizontal Distance (km)
CO2 is injected during seven years in the Utsira formation at a constant flow rate
of one million tons per year [Chadwick et al., 2010]. The injection is located at
the bottom of the formation: x1 = 0.6 km, x3 =1.082 km [Chadwick et al., 2005,
Chadwick et al., 2009]. The simulation uses a mesh with equally-spaced blocks
in each direction: nx1 = 300 in the x1 -direction, nx2 = 5 in the x2 -direction and
nx3 = 400 in the x3 -direction. Actually the model is 2.5D since the properties are
uniform along the x2 -direction, which has an extension of 10 km. It was verified
that taking larger extensions of the domain in the x2 -direction did not change signif-
icantly the CO2 saturation and pressure maps. Along the x2 -direction, the third grid
is used to locate the injection.
The flow parameter values in the relative permeability and capillary pressure
formulas (10.33), (10.34) and (10.35) were chosen as follows: Sgc = 0.02, Sbc =
0.12, ng = 1.15, nb = 2.5, nc = 4, Krb ∗ = 0, 95, K ∗ = 1, P∗ = 0.03 MPa.
rg ca
To satisfy the CFL stability condition due to IMPES formulation [Savioli and Bidner,
2005], the time step value is 0.125 d. With this choice of the mesh and time step, the
results of the BOAST simulator satisfy the mass conservation condition.
Recall that the petrophysical properties of the formation are time dependent due
to the CO2 injection and consequently pore pressure increase (cf. (10.28), (10.29),
(10.30)) but they change at a much slower rate than pressure and saturations. As a
consequence, we have two time scales, and we use a much larger time step to update
petrophysical properties than to run the flow simulator. After performing a sensibil-
ity analysis, in this numerical simulation we choose to update the petrophysical
properties every year.
Figures 10.5, 10.6, 10.7 and 10.8 show 2-D vertical slices (corresponding to
nx2 = 3) of the CO2 saturation fields after one, three, five and seven years of CO2 in-
jection, respectively. In all the cases, CO2 accumulations below the mudstone layers
can be observed. As injection proceeds, part of the injected fluid migrates upwards
due to the openings in the mudstone layers generating chimneys, and the vertical
fluid flow is ruled by the vertical permeability. As CO2 saturation increases, vertical
permeability updated with equation (10.30) also increases, in particular in the mud-
stone layers. This allows an increase in the CO2 upward motion across the layers
with the resulting low CO2 saturations levels observed between layers. As a conse-
quence, CO2 chimneys become less defined as injection time increases, as it can be
seen in these Figures.
A 2-D slice (at nx2 = 3) of the saturation dependent vertical permeability dis-
tribution after seven years of CO2 injection is shown in Figure 10.9. Porosity and
horizontal permeability depending only on pressure suffer little changes and are not
shown.
Capillary forces play an important role in multi-phase fluid flow, in particular
governing diffusion effects. This fact can be observed in Figures 10.10 and 10.11
displaying the difference between CO2 and brine pressure (capillary pressure) after
one and seven years of injection, respectively. As CO2 saturation increases and ac-
252 10.7 Seismic monitoring of CO 2 injection
cumulates below the mudstone layers, capillary pressure increases and flow is more
affected by capillary forces.
0.77 0.7
0.6
0.87
0.5
CO2 saturation
Depth (km)
0.4
0.97
0.3
0.2
1.07
0.1
1.17 0
0 0.3 0.6 0.9 1.2
Horizontal Distance (km)
Fig. 10.5 Spatial CO2 saturation distribution after one year of CO2 injection.
The viscoelastic model needs the complex plane-wave and shear moduli. In the top,
base and within the mudstone layers, the undrained bulk modulus Ku and the shear
modulus μ are computed using a Zener model as explained in Section 10.3 choosing
(1)
QKu = 100 at the central frequency of the source. Within the Utsira formation, the
CO2 -saturation dependent complex plane-wave and shear moduli, are determined
using the time-harmonic compressibility tests described in Chapter 7. For this pur-
10 The macro-scale. Seismic monitoring of CO2 sequestration 253
0.77 0.8
0.7
0.87 0.6
CO2 saturation
0.5
Depth (km)
0.97 0.4
0.3
1.07 0.2
0.1
1.17 0
0 0.3 0.6 0.9 1.2
Horizontal Distance (km)
Fig. 10.6 Spatial CO2 saturation distribution after three years of CO2 injection.
0.77 0.8
0.7
0.87 0.6
CO2 saturation
0.5
Depth (km)
0.97 0.4
0.3
1.07 0.2
0.1
1.17 0
0 0.3 0.6 0.9 1.2
Horizontal Distance (km)
Fig. 10.7 Spatial CO2 saturation distribution after five years of CO2 injection.
254 10.7 Seismic monitoring of CO 2 injection
0.77 0.8
0.7
0.87 0.6
CO2 saturation
0.5
Depth (km)
0.97 0.4
0.3
1.07 0.2
0.1
1.17 0
0 0.3 0.6 0.9 1.2
Horizontal Distance (km)
Fig. 10.8 Spatial CO2 saturation distribution after seven years of CO2 injection.
0.77 2500
2000
0.87 Vertical permeability (mD)
Depth (km)
1500
0.97
1000
1.07
500
1.17 0
0 0.3 0.6 0.9 1.2
Horizontal Distance (km)
Fig. 10.9 Vertical permeability distribution after seven years of CO2 injection.
10 The macro-scale. Seismic monitoring of CO2 sequestration 255
0.77 0.012
0.01
0.008
Depth (km)
0.97 0.006
0.004
1.07
0.002
1.17 0
0 0.3 0.6 0.9 1.2
Horizontal Distance (km)
Fig. 10.10 Capillary pressure distribution after one year of CO2 injection.
0.77 0.018
0.016
0.014
0.87 capillary pressure (MPa)
0.012
Depth (km)
0.01
0.97
0.008
0.006
1.07
0.004
0.002
1.17 0
0 0.3 0.6 0.9 1.2
Horizontal Distance (km)
Fig. 10.11 Capillary pressure distribution after seven years of CO2 injection.
256 10.7 Seismic monitoring of CO 2 injection
1/2
κe f f (Sb ) = κx1 κx3 (Sb ) .
Note that κx3 (Sb ) is fractal, since is defined in (10.30) in terms of κx1 .
The average properties of the fractal fields for porosity φ , horizontal permeability
κx1 , bulk modulus Km and shear modulus μm of the dry matrix, denoted by < ·, · >
are shown in Table 10.2.
These samples are assumed to be saturated with brine and CO2 , with the fluids
distribution in the form of irregular patches fully saturated with CO2 and zones fully
saturated with brine. No mixing forces are taken into account and the two fluids are
assumed to occupy different mesoscopic regions of the model. The generation of
these patchy brine-CO2 distributions was explained in Subsection 7.7.1. Table 10.3
shows the properties of the saturant fluids (brine and CO2 ) used in the harmonic
experiments.
In order to generate the CO2 patches, we need to fix both the sample size and
the correlation length, as well as the stochastic parameters. The correlation length
determines the maximum size for any CO2 bubble. Here we use square samples of
side length 80 cm and a correlation length of 20 cm; fractal dimension is D = 2.2.
Figure 10.12 show the CO2 patches generated for overall saturations of 5% (a)
and 30 % (b).
For all fluid-saturated poroelastic samples, we perform a time-harmonic com-
pressibility test. Each test allows to obtain the complex plane-wave modulus for
each saturation and for the 40 different frequencies used to generate the double-
entry table.
On the other hand, the complex shear modulus was determined as indicated in
Section 10.3.
Figure 10.13 displays fluid pressure maps at frequencies 2 Hz (a) and 60 Hz (b).
Both maps correspond to patchy brine-gas distributions for overall CO2 saturation
Sg = 30 %. It can be observed that pressure is lower in zones of 100 % CO2 satu-
ration, and pressure gradients at the boundaries of the CO2 bubbles are higher at 60
Hz than at 2 Hz.
10 The macro-scale. Seismic monitoring of CO2 sequestration 257
200 200
Vertical side length (cm)
0 0
0 100 200 0 100 200
Horizontal side length (cm) Horizontal side length (cm)
Fig. 10.12 Patchy brine-gas distributions for Sg = 5 % (a) and 30 % overall CO2 saturations (b).
White zones correspond to 100 % gas saturation and black zones to 100 % brine saturation.
80 0.8 80 0.8
0.7 0.7
Vertical side length (cm)
0.6 0.6
Fluid Pressure (Pa)
40 0.4 40 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0 0 0
Horizontal side length (cm) Horizontal side length (cm)
Fig. 10.13 Fluid pressure map at frequency 2 Hz (a) and 60 Hz (b), corresponding to the patchy
brine-gas distributions for overall CO2 saturation Sg = 30 %.
Table 10.2 Average physical properties of the solid matrix used in the time-harmonic experiments
Table 10.3 Physical properties of the saturant fluids used in the time-harmonic experiments
Kf ρf η
3
Brine 2.25 GPa 1040 Kg/m 0.0018 Pa · s
0.77 2600
2550
2500
2400
0.97 2350
2300
2250
1.07
2200
2150
1.17 2100
0 0.3 0.6 0.9 1.2
Horizontal Distance (km)
0.77 2600
2400
2200
Depth (km)
0.97 2000
1800
1.07
1600
1.17 1400
0 0.3 0.6 0.9 1.2
Horizontal Distance (km)
Fig. 10.15 P-wave phase velocity map at 50 Hz after seven years of CO2 injection.
0.77 1100
1050
1000
0.87 velocity (m/s)
950
900
Depth (km)
850
0.97
800
S-wave phase
750
700
1.07
650
600
1.17 550
0 0.3 0.6 0.9 1.2
Horizontal Distance (km)
0.77 1800
1600
1400
Depth (km)
0.97 1200
1000
1.07
800
1.17 600
0 0.3 0.6 0.9 1.2
Horizontal Distance (km)
Fig. 10.17 S-wave phase velocity map at 50 Hz after seven years of CO2 injection.
0.77 600
500
0.87
P-wave quality factor
400
Depth (km)
0.97 300
200
1.07
100
1.17 0
0 0.3 0.6 0.9 1.2
Horizontal Distance (km)
Fig. 10.18 Quality factor QP of compressional waves at 50 Hz after seven years of CO2 injection.
10 The macro-scale. Seismic monitoring of CO2 sequestration 261
0.77 120
110
100
0.87
80
0.97
70
60
1.07
50
40
1.17 30
0 0.3 0.6 0.9 1.2
Horizontal Distance (km)
Fig. 10.19 Quality factor QS of shear waves at 50 Hz after seven years of CO2 injection.
Distance (m)
200 400 600 800 1000
0
0.1
0.2
Time (s)
0.3
0.4
It is observed a noticeable decrease in phase velocity and quality factors for both
waves in zones of CO2 accumulation. These strong changes in velocities and atten-
uation induce delays and attenuation in the waves recorded at the top of the model,
to be observed in the next Figures. These Figures show the synthetic seismograms
(seismic sections) obtained measuring the vertical component of the displacement
vector on a line of receivers at the top of the model.
Figures 10.20, 10.21 10.22, 10.23 and 10.24 display migrated seismic sections
before CO2 injection and after one, three, five and seven years of CO2 injection
associated with the CO2 saturations shown in Figures 10.5, 10.6, 10.7 and 10.8
respectively.
A standard f -k filter is used to eliminate some spurious reflections from the artifi-
cial boundaries of the computational mesh, since the absorbing boundary conditions
used are only first order and can not eliminate all undesired boundary reflections.
The reflections seen in those seismograms show the progressive increment in
CO2 accumulations below the mudstone layers as injection proceeds. In particular,
10 The macro-scale. Seismic monitoring of CO2 sequestration 263
Distance (m)
200 400 600 800 1000
0
0.1
0.2
Time (s)
0.3
0.4
the pushdown effect observed in the real seismograms [Chadwick et al., 2009] due
to CO2 accumulations, as shown in the Figure 10.25, is clearly observed.
Summarizing, the combined use of multi-phase fluid-flow and viscoelastic wave
propagation simulators defines a methodology that becomes an important tool to
monitor the migration and dispersal of the CO2 plume, in order to analyze storage
integrity and to make long term predictions.
The IMPES technique combines the flow equations (10.6)- (10.7) to obtain a single
pressure equation. In this way, equation (10.6) multiplied by Bg and equation (10.7)
multiplied by (Bb − Rs Bg ) are added. The right-hand side of the combined equation
results,
264 10.8 Appendix 1. IMPES solution for Black-Oil formulation
Distance (m)
200 400 600 800 1000
0
0.1
0.2
Time (s)
0.3
0.4
# S
Rs Sb $ # S $
g
∂ φ + ∂ φ
b
Bg Bb Bb
Bg + (Bb − Rs Bg ) . (10.37)
∂t ∂t
Using the chain rule to expand the time derivatives, and after some algebraic
manipulations, the expression in 10.37 becomes:
# 1 dφ
1 dBg 1 dBb Bg dRs $ ∂ pb
φ + Sg − + Sb − + ,
φ d pb Bg d pb Bb d pb Bb d pb ∂t
Distance (m)
200 400 600 800 1000
0
0.1
0.2
Time (s)
0.3
0.4
1 dφ
Formation compressibility: c f = ,
φ d pb
1 dBg
Gas compressibility: cg = − ,
Bg d pb
1 dBb Bg dRs
Brine compressibility: cb = − + ,
Bb d pb Bb d pb
Total compressibility: ct = c f + Sg cg + Sb cb ,
the following simply expression is obtained for the right-hand side of the combined
equation,
∂ pb
φ ct .
∂t
266 10.9 Appendix 2. Estimation of brine density and CO 2 mole and mass fractions...
Distance (m)
200 400 600 800 1000
0
0.1
0.2
Time (s)
0.3
0.4
The brine mass density at reservoir conditions is estimate using the approach of
Garcı́a [Hassanzadeh et al., 2008] as follows,
Fig. 10.25 Real seismograms showing the pushdown effect (after [Chadwick et al., 2009]).
where Mg , Mb are the CO2 and brine molecular weights, respectively; χg is the CO2
mole fraction in the brine phase; ρbSC is the brine mass density at standard conditions
and Vm is a partial molar volume, computed as a function of temperature T (o C) as
in formula (21) of [Hassanzadeh et al., 2008]:
where mg is the molality of CO2 in saline water, ms is the salt molality and ν is the
stoichiometric number of ions in the dissolved salt. In order to compute mg we use
mog
mg = ,
γg∗
where mog is the molality of CO2 in pure water and γg∗ the activity coefficient.
To obtain mog for each pressure and temperature, we solve iteratively the thermo-
dynamic nonlinear equations (B1)-(B5) in [Hassanzadeh et al., 2008]. These equa-
268 10.9 Appendix 2. Estimation of brine density and CO 2 mole and mass fractions...
tions involve H2 O and CO2 molar volumes, fugacities and activity coefficients. The
CO2 molar volume is computed using the Peng-Robinson model [Peng and Robinson,
1976]. Also, γg∗ is obtained using the Rumpf model, as stated in equation (A6) in
Spycher & Pruess [Spycher and Pruess, 2005].
Once χg is computed, the CO2 mass fraction in the aqueous phase ωg is obtained
from the relation:
Mg
ωg = χg .
Maq
Here Maq is the aqueous molecular weight, given by
11.1 Introduction
a point source at ultrasonic frequencies. The snapshots show the generated wave
fields, where the events associated with the different types of waves can be clearly
observed.
Here β jk is a positive definite matrix function defined on the interior boundaries Γjk .
The Robin transmission conditions (11.2)–(11.3) impose the continuity of the solid
displacements, the normal component of the fluid displacements and the generalized
stresses at the interior interfaces Γjk .
ν 1,χ f 3,ν 3,χ
Next we introduce a set of Lagrange multipliers ηjk =(η 1,
jk , η jk , −η jk , η jk , η jk )
associated with the values of the generalized stresses G jk at the mid points ξ jk of Γjk
in the sense that η jk ∼ G jk (u j )(ξ jk ).
The Lagrange multipliers η jk belong to the following space of functions defined
on the interior interfaces Γjk :
L h = η : η |Γjk = η jk ∈ [P0 (Γjk )]5 = L jkh , ∀{ j, k} ,
v = (v1 , v2 , v3 ) ∈ M C hj × V jh × M C hj × L jkh ,
ν ,n ν ,n−1
η jk = ηk j − iωβ jk ΠΓjk (U j ) + ΠΓk j (Uk ) (ξ jk ).
n n−1
(11.5)
where
272 11.3 A numerical example in the ultrasonic range of frequencies
⎡ ⎤
m11 I m12 I m13 I
A = ⎣ m12 I m22 I m23 I ⎦ ,
m13 I m23 I m33 I
I denotes the identity matrix in R3×3 and the mass coefficients mi j = mi j (ω ) are
defined in (4.28).
In (11.4) u, vΓ denotes the approximation of the L2 (Γ ) inner product < ·, · >Γ
by the mid–point quadrature.
Equation (11.5), used to update the Lagrange multipliers, is obtained directly
from (11.2) evaluated at the mid point ξ jk . Equation (11.4) yields a 20 × 20 linear
system of equations for the degrees of freedom associated with the vector displace-
ments of the three phases on each sub-domain Ω j at the n−iteration level. The
iteration (11.4)–(11.5) is a Jacobi–type iteration. A twice as fast iteration may also
be defined by using a red–black type iteration (see Table 10.1).
The iteration matrix β jk is chosen to have the same form of the matrix B p in
(11.1). The space–time solution is obtained by solving (11.4)–(11.5) for a finite
number of frequencies and using an approximate inverse Fourier transform. The
definition of the iteration (11.4)–(11.5) can be extended to the case of larger sub-
domains Ω j , as explained in Chapter 10.
Ω = Ω1 ∪ Ω2 ∪ Ω3
with Ω1 , Ω2 and Ω3 denoting the Berea, water and ice portions of Ω . If Vi denotes
the volume of the i-portion of Ω , the ice content S3 in the pores is determined by
the relation
V3
S3 = .
V1 +V2
The sample is assumed to have an interior horizontal plane interface Γi at x3 = 1.5
cm defined by a change in ice content in the pores.
The material properties of the system, taken from [Carcione and Seriani, 1998,
Carcione and Tinivella, 2000a] are given in Table 11.1. In this case the indices 1
and 3 correspond to the sandstone and ice properties, respectively. The mesh is an
uniform partition of Ω into squares of side length h = L/nx1 , with nx1 = nx3 = 200.
The source function (f1 , f f , f3 ) is a compressional and shear point source located
at (xs , zs ) = (1.5 cm, 1.0 cm) applied to the solid and fluid phases, with a band limited
spectrum of central frequency equal to 500 kHz.
The absolute porosity of the sandstone is φ a = 0.3, with the water content φ chosen
such that the ice content S3 in the pores changes at the interface Γ from 66 percent
in the upper layer to 33 percent in the lower layer.
11 Wave propagation in partially frozen porous media 273
For the calculation of the elastic coefficients we need values for the bulk and
shear modulus of the two solid dry frames, denoted by Ks1,m , Ks3,m , μs1,m and
μs3,m , respectively. Following [Leclaire et al., 1994, Carcione and Tinivella, 2000a]
and [Santos et al., 2004a] it is assumed that Ks1,m = 14.4 GPa and that the modu-
lus μs1,m , μs3,m and Ks3,m can be computed using a percolation-type model us-
(max) (max) (max)
ing the relations (4.37) where μs1,m , μs3,m and Ks3,m are computed using the
Kuster and Toksöz’s model[Kuster and Toksöz, 1974], taking the known values of
Ks1 , μs1 , Ks3 , μs3 for the background medium with inclusions of air, with properties
Ka , μa . The moduli μs1,m 0 , μ0 0
s3,m and Ks3,m are appropriate reference values. Here we
take
μs1,m
0
= 13.3 GPa, Ks3,m
0
= μs3,m
0
= 0.
The viscoelastic behaviour of the material is modeled using the parameters t1,M =
10−1 ms, t2,M = 109 ms (see notation in Appendix 1.9) to make the moduli M =
KG1 , KG3 , μ1 , μ3 , μ13 , B3 complex and frequency dependent. The mean quality fac-
tors are taken to be Q K = Q K = Q μ = Q μ = 100 and Qμ = QB = 200. The
G1 G3 1 3 13 3
value of the Kozeny-Carman constant was taken to be 5 [Hovem and Ingram, 1979].
Table 11.2 displays values of the phase velocity and attenuation factors at 500
kHz for the five different types of waves for the two–layer model used in this exper-
iment.
The following Figures present snapshots of the wave fields for this experiment,
generated after solving (11.4)–(11.5) for 110 equally spaced temporal frequencies
in the interval (0, 1 MHz). To understand and identify more clearly the various kinds
of waves propagating in the partially frozen sandstone we also show snapshots of
the curl and divergence of the displacements of the three phases, i.e., the rock frame,
the ice and the fluid.
Figures 11.1, 11.2 and 11.3 show respectively snapshots of the vertical compo-
nent of the particle velocity of the three phases at t = 0.005 ms where we can observe
274 11.3 A numerical example in the ultrasonic range of frequencies
Table 11.2 Wave speeds and attenuation factors for all waves at frequency 500 kHz
Ice content 0.66 Ice content 0.33
Wave Phase velocity (km/s) Attenuation (dB) Phase velocity (Km/s) Attenuation (dB)
Fast P1 wave 4.08 0.35 3.90 0.47
Slow P2 wave 1.37 5.20 1.05 7.88
Slow P3 wave 0.22 30.0 0.22 6.70
Fast S1 wave 2.72 0.58 2.58 0.74
Slow S2 wave 0.57 2.45 0.22 1.04
that after arriving at the interface Γi , the direct P1 and P2 waves have generated re-
flected and transmitted fast and slow shear and compressional waves. In particular
the P2 wave generated at the source has arrived at Γi at t = 0.0035 and has generated
reflected and transmitted fast and slow compressional and shear waves. The short
wavelength located near the source corresponds to the slow shear wave generated
by the source.
Figures 11.4, 11.5 and 11.6 show respectively snapshots of the vertical compo-
nent of the particle velocity of the three phases at t = 0.01 ms. In Figure 11.4 can
be seen a fast transmitted P1 wavefront arriving at the top boundary and a slow P2
wavefront above Γi , that is clearly identified by observing the snapshot of the diver-
gence of the ice phase at at t = 0.01 ms shown in Figure 11.7. In the same Figure,
below Γi we see a reflected P2 wavefront which front is passing through the source
location, and a reflected P1 wavefront arriving at the bottom boundary. These last to
wavefronts are also clearly identified in Figure 11.7.
The other wavefronts seen in Figures 11.4, 11.5 and 11.6 correspond to shear
waves. The strong wavefront at the center in these 3 Figures is a slow shear wave
generated at the source that is arriving at the interface Γi generating fast and slow
compressional and shear waves. Figure 11.8 shows a slow wave shear wavefront
arriving at Γi and generating slow reflected and transmitted slow shear waves, iden-
tified as slow waves because of their short wavelengths.
Finally, Figures 11.9, 11.10 and 11.11 display snapshots of the three phases at t =
0.02 ms. There are fast and slow compressional and shear waves above and below
Γi . Looking at the divergence of the ice phase in Figure 11.12 and the corresponding
vertical displacement in Figure 11.10 we can identify a slow transmitted P2 wave
above Γi and a fast transmitted P1 wave of larger wavelength at the top of the Figure.
The strong wavefront below Γi is a fast transmitted P1 wave travelling towards the
bottom boundary. Concerning shear wavefronts, by observing the curl of the ice
phase in Figure 11.13, we can identify in Figure 11.10 a transmitted slow shear
wavefront above Γi and two reflected slow shear wavefronts below Γi . The center
of the wavefront below and closer to Γi is passing through the source location, and
the other slow shear wavefront below Γi is arriving at the bottom boundary. This last
slow shear wavefront is arriving at the bottom boundary before the fast reflected fast
P1 wave observed in Figure 11.12.
11 Wave propagation in partially frozen porous media 275
3.
Vertical side length (cm)
1.5
0
0 1.5 3.
Horizontal side length (cm)
Fig. 11.1 Ice content interface. Vertical component of the velocity of the rock frame at t = 0.005
ms.
3.
Vertical side length (cm)
1.5
0
0 1.5 3.
Horizontal side length (cm)
Fig. 11.2 Ice content interface. Vertical component of the velocity of the ice frame at t = 0.005
ms.
276 11.3 A numerical example in the ultrasonic range of frequencies
3.
Vertical side length (cm)
1.5
0
0 1.5 3.
Horizontal side length (cm)
Fig. 11.3 Ice content interface. Vertical component of the velocity of the fluid at t = 0.005 ms.
The relative amplitudes between the snapshots in Figures 11.1, 11.2 and 11.3 is 1/196.3/4.15.
3.
Vertical side length (cm)
1.5
0
0 1.5 3.
Horizontal side length (cm)
Fig. 11.4 Ice content interface. Vertical component of the velocity of the rock frame at t = 0.01
ms.
11 Wave propagation in partially frozen porous media 277
3.
Vertical side length (cm)
1.5
0
0 1.5 3.
Horizontal side length (cm)
Fig. 11.5 Ice content interface. Vertical component of the velocity of the ice frame at t = 0.01 ms.
3.
Vertical side length (cm)
1.5
0
0 1.5 3.
Horizontal side length (cm)
Fig. 11.6 Ice content interface. Vertical component of the velocity of the fluid at t = 0.01 ms. The
relative amplitudes between the snapshots in Figures 11.4, 11.5 and 11.6 is 1/154.2/3.26.
278 11.3 A numerical example in the ultrasonic range of frequencies
3.
Vertical side length (cm)
1.5
0
0 1.5 3.
Horizontal side length (cm)
Fig. 11.7 Ice content interface. Divergence of the displacement of the solid phase 3 at t = 0.01 ms.
3.
Vertical side length (cm)
1.5
0
0 1.5 3.
Horizontal side length (cm)
Fig. 11.8 Ice content interface. Curl of the displacement of the solid phase 3 at t = 0.01 ms.
11 Wave propagation in partially frozen porous media 279
3.
Vertical side length (cm)
1.5
0
0 1.5 3.
Horizontal side length (cm)
Fig. 11.9 Ice content interface. Vertical component of the velocity of the rock frame at t = 0.02
ms.
3.
Vertical side length (cm)
1.5
0
0 1.5 3.
Horizontal side length (cm)
Fig. 11.10 Ice content interface. Vertical component of the velocity of the ice frame at t = 0.02
ms.
280 11.3 A numerical example in the ultrasonic range of frequencies
3.
Vertical side length (cm)
1.5
0
0 1.5 3.
Horizontal side length (cm)
Fig. 11.11 Ice content interface. Vertical component of the velocity of the fluid at t = 0.02 ms.
The relative amplitudes between the snapshots in Figures 11.9, 11.10 and 11.11 is 1/157.2/3.4.
3.
Vertical side length (cm)
1.5
0
0 1.5 3.
Horizontal side length (cm)
Fig. 11.12 Ice content interface. Divergence of the displacement of the solid phase 3 at t = 0.02
ms.
11 Wave propagation in partially frozen porous media 281
3.
Vertical side length (cm)
1.5
0
0 1.5 3.
Horizontal side length (cm)
Fig. 11.13 Ice content interface. Curl of the displacement of the solid phase 3 at t = 0.02 ms.
Chapter 12
The macro-scale. Wave propagation in
transversely isotropic media
12.1 Introduction
Table 12.1 Material properties of the background, fractures and saturant fluids
Background bulk modulus, Ks 37. GPa
shear modulus, μs 44. GPa
grain density, ρs 2650 kg/m3
dry bulk modulus, Km 1.17 GPa
dry shear modulus, μ 1.4 GPa
porosity, φ 0.25
permeability, κ 2.5 10−13 m2
tortuosity, S 2.8
Fractures bulk modulus, Ks 37. GPa
dry bulk modulus, μm 4.55 GPa
dry shear modulus, μm 0.69 GPa
porosity, φ 0.5
permeability κ 2.5 10−10 m2
tortuosity, S 2.8
Brine bulk modulus, Kw 2.25 GPa
density, ρw 1040 kg/cm3
viscosity, ηw 0.0018 Pa · s
Oil bulk modulus, Ko 0.57 GPa
density, ρo 700 kg/cm3
viscosity, ηo 0.004 Pa · s
Gas bulk modulus, Kg 25. 106 Pa
density, ρg 500 kg/m3
viscosity, ηg 0.00002 Pa · s
90 qP 90 qP
4 qSV qSV
4
60 SH 60 SH
3 3
Vez (km/s)
Vez (km/s)
30 30
2 2
1 1
0 0
1 2 3 4 1 2 3 4
Vex (km/s) Vex (km/s)
(a) (b)
Fig. 12.1 Polar representation of the qP, qSV and SH energy velocity vectors at 30 Hz for gas-filled
fractures, (a) High fracture intensity, and (b) low fracture intensity.
286 12.2 Properties of the equivalent TIV medium
The influence of different fluids filling the fracture is illustrated in Figure 12.2.
In this case, the qP, qSV and SH energy velocities at 30 Hz are shown together with
their respective quality factors for oil saturated fractures. When comparing Figure
12.1a with Figure 12.2a, it is observed that qP velocities are higher for oil than for
gas saturated fractures. Concerning attenuation, energy losses for qSV waves are
much higher for angles between 30 and 60 degrees. Notice that for qP waves dis-
sipation anisotropy is much important for waves travelling in the direction incident
normal to the fracture planes and the quality factor,Q, is great on angles close to 45
degrees.
90 qP 90 qP
qSV qSV
4 3
60 SH 60
3
2
Vez (km/s)
1000/Q(Z)
30 30
2
1
1
0 0
1 2 3 4 1 2 3
Vex (km/s) 1000/Q(X)
(a) (b)
Fig. 12.2 Polar representation of the qP, qSV and SH energy velocity vectors (a), and the quality
factors (b) at 30 Hz for oil-filled fractures and high fracture intensity.
(−σ (us (x, ω ))ν · ν , −σ (us (x, ω ))ν · χ 1 , −σ (us (x, ω ))ν · χ 2 ) (12.2)
= iω B p (u (x, ω ) · ν , u (x, ω ) · χ , u (x, ω ) · χ ), x ∈ Γ .
s s 1 s 2
In (12.1), ρ is the density and σ and f represent the stress tensor of the equivalent
TIV medium and the external source vector, respectively. Also, in (12.2) B p is a
symmetric, positive definite matrix depending on the stiffness coefficients, ν is the
unit outer nornmal on Γ and χ 1 , χ 2 are two unit tangents on Γ so that (ν , χ 1 , χ 2 ) is
an orthonormal system on Γ . The boundary condition (12.2) is obtained by apply-
ing the momentum equations on Γ , so that Γ is transparent for normally incident
waves. The form of the matrix B p is obtained as an extension of the result given by
[Lovera and Santos, 1988] using the expression (5.38) in Chapter 5.
Next we formulate the variational form of (12.1)-(12.2): Find us (x, ω ) ∈ [H 1 (Ω )]3
such that
where e = ei j is the
;
strain tensor. ;
Here ( f , g) = Ω f gd(Ω) and f , g = Γ f gd(Γ ) indicate the complex [L2 (Ω )]3
and [L2 (Γ )]3 inner products, where g denotes the complex conjugate of g. H 1 (Ω )
denotes the usual Sobolev space of function in L2 (Ω ) with first derivatives in
L2 (Ω ) as defined in Chapter 6 (see also [Adams, 1975]). The arguments given in
[Ha et al., 2002] and [Douglas Jr. et al., 1994] can be used to show that existence
and uniqueness holds for the solution of (12.3).
Numerical dispersion is an important aspect to be taken into account when using
wave propagation algorithms. It is shown in [Zyserman et al., 2003] that using the
non-conforming FE space M C h described in Chapter 6 allows to use about half
288 12.3 The seismic modeling method
The degrees of freedom associated with S are the values at the mid points of the faces
of R. For example, if a1 = (−1, 0, 0), a2 = (0, 1, 0), a3 = (0, −1, 0), a4 = (0, 0, 1),
a5 = (0, 0, −1) and a6 = (1, 0, 0), the basis function
1 1 3 5 4 5 4 5 4
ψ1 (x1 , x2 , x3 ) = − x1 − (x1 − x1 ) − (x2 − x2 ) − (x3 − x3 )
2 2 2
4 2 8 3 3 3
It can be shown (see [Ha et al., 2002] for the isotropic case) that for h sufficiently
small the error associated with the global procedure (12.4) is of order h2 in the
L2 -norm and of order h in the broken H 1 -energy norm.
Note that (12.4) is a non-coercive elliptic Helmholtz-type problem, so that the
usual iterative procedures like preconditioned conjugate gradient iterative algo-
rithms can not be used. Consequently, to solve the algebraic problem associated
with the global non-conforming procedure (12.4), we will employ the iterative do-
main decomposition procedure described below.
In addition, to the low numerical dispersion properties of the space M C h , one of
the main advantages of using non-conforming elements to solve wave propagation
phenomena in parallel architectures is that the amount of information exchanged in
a domain decomposition iterative procedure is reduced by half as compared to the
case when conforming elements are employed.
Consider the decomposed problem over Ω j satisfying equation (12.1) in Ω j . The
boundary condition is
12 The macro-scale. Wave propagation in transversely isotropic media 289
(−σ (usj (x, ω ))ν · ν , −σ (usj (x, ω ))ν · χ 1 , −σ (usj (x, ω ))ν · χ 2 )
= iω B p (usj (x, ω ) · ν , usj (x, ω ) · χ 1 , usj (x, ω ) · χ 2 ), x ∈ Γ ,
where β jk are the components of a positive definite matrix function defined on the
interior boundaries Γjk . The iteration matrix β jk , defined on the interior interfaces
Γjk can be taken to be of the same form as the matrix B p using averaged properties
of the coefficients defining B p on the adjacent elements Ω j and Ωk .
Since the objective of the domain decomposition technique is to localize the cal-
culations, we define the iterative procedure at the differential level in the following
manner: Find unj ∈ [H 1 (Ω j )]3 such that
j , ϕ ) jk + ∑(e pq (u j ), e pq (ϕ )) jk +
−(ρω 2 uh,n h,n
pq
++ ,,
iω B p (uh,n
j (x, ω ) · ν , uh,n
j (x, ω ) · χ 1 h,n
, u j (x, ω ) · χ 2
), ϕ
Γj
290 12.4 Numerical experiments
++ ,,
+ ∑ η h,n
jk , ϕ = (f, ϕ ) j , ϕ ∈ M C hj (12.6)
k Γjk
and
η h,n h,n−1
jk = −ηk j + (12.7)
i β jk [(uh,n h,n 1 h,n
j (x, ω ) · ν , u j (x, ω ) · χ , u j (x, ω ) · χ )(ξ jk ) −
2
(uh,n−1
k (x, ω ) · ν , uh,n−1
k (x, ω ) · χ 1 , uh,n−1
k (x, ω ) · χ 2 )(ξ jk )], on Γjk .
In (12.6) ·, ·Γjk denote the approximation to the (complex) inner product ·, ·Γjk
in L2 (Γjk ) computed using the mid–point quadrature rule, and similarly for ·, ·Γj .
The argument given in [Ha et al., 2002] for isotropic viscoelastic solids can be
applied here with minor modifications to show that
uh,n − uh 0 → 0 as n → ∞,
The validation of the numerical simulations for 2-D and 3-D wave propagation was
performed by comparison with the analytical solution for transversely isotropic me-
dia given in [Carcione, 2014] (Chapter 1).
In the numerical experiments waves fields are modeled for the three-dimensional
viscoelastic anisotropic case. The domain Ω is a cube de side length 1500 m, with
a uniform partition T h of Ω into 200×200×200 cubic elements.
The external source is a dilatational perturbation in the center of the domain
whose waveform in the time domain is a Ricker wavelet with peak frequency, f0 =
30 Hz, i.e.,
g(t) = e− f0 (t−t0 ) /2 cos(π f0 (t − t0 )),
2 2
(12.8)
with t0 =70 ms. Therefore, if the source is located at the point (x1s , x2s , x3s ) and
δx1s ,x2s ,x3s indicates the Dirac distribution at this point, the external source is
where g(ω ) is the Fourier transform of the wavelet in (12.8). The solution was com-
puted for 160 frequencies in the range 0-80 Hz.
The VTIV medium corresponds to high fracture intensity with gas as the saturat-
ing fluid. The medium properties are those in Table 12.1.
Snapshots at 200 ms of the displacement magnitude on three orthogonal planes
(a) and two orthogonal planes (b) are shown in Figure 12.3. The VTIV medium
12 The macro-scale. Wave propagation in transversely isotropic media 291
x3
qSV
qP
z
P x2
x1
y
x
(a)
x3
qSV
qP
P x2
(b)
Fig. 12.3 Snapshots at 200 ms of the displacement magnitude on the three orthogonal planes (a)
and two orthogonal planes (b). The medium is VTIV. The dilatational point source is located at the
center of the domain.
corresponds to high fracture intensity with brine saturated background and gas as
the fluid saturating the fractures. Two coupled wave fronts labeled qP and qSV,
indicate quasi-longitudinal and quasi-transverse waves, respectively. The advance
of the wave fronts is consistent with the different velocities in Figure 12.1a.
For a better interpretation of the results, Figure 12.4 shows only the horizontal
(a) and vertical (b) planes. The coupled waves, qP and qSV, can be observed on
any arbitrary orientation of the vertical plane. Since the normal plane to the x3 -axes
is isotropic, the classical longitudinal (P) wave is seen to propagate in such plane
(Figure 12.4a). The S-wave is not generated due to the type of source used.
Wave propagation in the equivalent VTI medium whose velocities are character-
ized in Figures 12.1b and 12.2a are shown in Figure 12.5. We focus on and compare
travelling waves in the two principal planes at time 200 ms. For low fracture in-
tensity and gas-saturated fractures, the medium behaves essentially as an isotropic
qP
292 12.4 Numerical experiments
P x2
x1
(a)
x3
qSV
qP x2
(b)
Fig. 12.4 Snapshots at 200 ms of the displacement magnitude on the horizontal (a) and vertical
planes (b). The medium is VTIV as in Figure 12.3.
medium with the highest velocities (Figure 12.5a). High VTIV anisotropy can be
observed in Figure 12.5b, induced by high fracture intensity and oil-saturated frac-
tures. The curvature of the wavefronts are a measure of the degree of the anisotropy
of the medium.
The following simulations correspond to a TIV medium with horizontal sym-
metry axis (x1 -axis) (HTIV). Its stiffness matrix can be obtained by a matrix trans-
formation that rotates the x3 -axis clockwise with an angle π /2 about the x2 -axis.
Then, the new stiffness matrix is P = M P M t , where M is the Bond transforma-
tion matrix. The construction of this rotation transformation is given in Appendix
12.6.
Figure 12.6 shows a snapshot at 222 ms of the displacement magnitude of a
medium with HTIV anisotropy. The dilatational point source at the center of the
12 The macro-scale. Wave propagation in transversely isotropic media 293
x3
qSV qP
x2
P
(a)
x3
qSV qP
P x2
(b)
Fig. 12.5 Snapshots at 200 ms of the magnitude of displacement on two orthogonal planes. VTIV
anisotropy of very low-order for low fracture intensity and gas saturated fractures (a) and high-
order VTIV anisotropy for high fracture intensity and oil saturated fractures (b). The source is a
dilatational point perturbation at the center of the domain.
domain generate a qP-wave and a slow (qS2) shear-wave in the (x1 , x3 ) and (x1 , x2 )
planes. In this example, the (x2 , x3 ) plane is oriented in the direction of the vertical
fractures and behaves isotropically.
In order to highlight the S-waves propagation split, we consider a S-wave source
as a strain ε12 = 1 defined by the distribution ε12 δx1s ,x2s ,x3s located at the center of
the horizontal plane (x1 , x2 ) , i.e.
∂ δx1s ,x2s ,x3s ∂ δx1s ,x2s ,x3s
, . (12.9)
∂ x2 ∂ x1 )
Figure 12.7 illustrates fast and slow S-waves corresponding to directions parallel
and normal to the fractures, respectively. Consequently, 9-c multicomponent seismic
.
x3
qP
P
qSV
qP
x1
x2
Fig. 12.6 Snapshot at 222 ms of the displacement magnitude on three orthogonal planes. The
medium is HTI with high fracture intensity and gas saturated fractures. The source is a dilatational
point perturbation.
can be simulated using these techniques and shear sources similar to the one in
(12.9).
x3
S2 S1
P
qP
x2
x1
(a)
x3
S2
S1
qP
x1 x2
(b)
Fig. 12.7 Snapshot of the displacement magnitude for fast and slow (S1 and S2) split S-waves
on the three orthogonal planes at 200 ms (a) and 300 ms (b). The medium is HTIV with high
fracture intensity and gas-filled fractures. The perturbations are generated by a shear source at the
center of the (x1 , x2 ) plane.
Figures 12.9 and 12.10 display traces measured at a receiver located inmediately
below the source location.
Traces in Figure 12.9 show reflections from the top and bottom of the HTIV and
VTIV layers for the case of brine-saturated background and with fractures. It can
be seen that the reflection from the top of the VTIV layer has lower amplitude than
the corresponding one of the HTIV layer due to the lower reflectivity of the VTIV
layer. The reflection from the bottom of the VTIV layer is delayed and attenuated as
compare with the HTIV arrival due to lower velocity and high attenuation suffered
by qP-waves travelling within the VTIV layer, in accordance with Figure 12.2.
Figure 12.10 show traces displaying reflections from the top and bottom of the
HTIV layer for the cases of brine and patchy brine-gas saturated background and
fractures. The arrival times correspond to the speed of the media with a slight delay
and greater attenuation rates in the presence of the gas patches.
The synthetic seismograms in Figure 12.11 correspond the HTIV and VTIV lay-
ers with brine saturated background and fractures. The seismograms show the ar-
296 12 5 2-D seismic imaging of an anisotropic layer
(a) (b)
Fig. 12.8 2-D model with source and line of receivers. Isotropic background and the embedded
horizontal anisotropic layer. HTIV layer (a) and VTIV layer (b).
8e-06
6e-06 HTI
Reflection from bottom of VTI layer
VTI
x3z-component of particle velocity (m)
4e-06
2e-06
-2e-06
Reflection from top of
anisotropic layer
-4e-06
-6e-06
Reflection from bottom of HTI layer
-8e-06
0 200 400 600
Time (ms)
Fig. 12.9 Synthetic traces. HTIV and VTIV brine saturated fractures.
rivals of P and S waves reflected at the top of the HTIV (Figure 12.11a) and VTIV
(Figure 12.11b) layers, as well as later arrivals corresponding to conversions from
incident P to qP and qSV waves at the top and bottom of the layers. In particular, it
is clearly seen the late P-qP-P arrival of the VTIV case as compared with the corre-
sponding one in the HTIV case. The qP-velocities in the HTIV and VTIV layers are
about 3800 m/s and 3000m/s, respectively.
Finally, seismograms for the HTIV layer with brine and 10 % gas saturated back-
ground and fractures are shown in Figure 12.12. They display the arrivals of P and
12 The macro-scale. Wave propagation in transversely isotropic media 297
8e-06
Brine saturated fractures
Patchy brine-CO2 saturated fractures P-qP-qP-qP-qP-P multiple
6e-06
reflection
x3z-component of particle velocity (m)
4e-06
2e-06
-2e-06
Reflections from top
-4e-06 of HTI layer
-8e-06
0 200 400 600 800
Time (ms)
Fig. 12.10 Synthetic traces. HTIV medium with brine and patchy brine-gas saturated fractures.
Geophone Geophone
50 100 150 50 100 150
0 0
0.2 0.2
P−P
P−P
0.4 0.4
Time (s)
Time (s)
P−qP−S
P−S
P−qP−qP−qP−qP−P
P−qP−qP−qP−qP−P P−qP−qSV−S
0.8 0.8
P−qP−qSV−S
(a) (b)
Fig. 12.11 Synthetic seismograms. HTIV (a) and VTIV (b) brine saturated fractures.
S waves reflected at the top of the HTIV layer, and later arrivals corresponding to
conversions from incident P to qP and qSV waves at the top and bottom of the layer.
In particular, the P-qP-qP-qP-qP-P arrival in the brine-saturated case is not seen in
the patchy brine-gas case due to the strong attenuation of qP waves when gas is
present (Qp ≈ 10 at normal incidence). Instead qSV-waves are less attenuated when
gas is present and the P-qP-qSV-S-arrival is still observed.
298 12.6 Appendix 1. Rotation transformation in R3
Geophone Geophone
50 100 150 50 100 150
0 0
0.2 0.2
P−P P−P
0.4 0.4
Time (s)
Time (s)
P−S
P−S
P−qP−P
0.6 P−qP−P P−qP−qP−S 0.6
P−qP−qP−qP−qP−P P−qP−qP−S
0.8 0.8
P−qP−qSV−S P−qP−qSV−S
(a) (b)
Fig. 12.12 Synthetic seismograms. HTIV medium with brine (a) and 10 % gas (b) saturated frac-
tures.
Elemental rotation about one of the axes of a coordinate system can be expressed by
⎛ ⎞
a11 a12 a13
A = ⎝ a21 a22 a23 ⎠
a31 a32 a33
Then, the rotation applied to the stiffness matrix of the TI medium with vertical
symmetry axis is
⎛ ⎞ ⎛ ⎞
cosθ 0 −sinθ 0 0 −1
A = ⎝ 0 1 0 ⎠ = ⎝0 1 0 ⎠
sinθ 0 cosθ 10 0
12 The macro-scale. Wave propagation in transversely isotropic media 299
Finally, ⎛ ⎞
p33 p13 p13 0 0 0
⎜ p13 p11 p12 0 0 0 ⎟
⎜ ⎟
⎜ p13 p12 p11 0 0 0 ⎟
P = M P M t = ⎜
⎜ 0
⎟
⎜ 0 0 p66 0 0 ⎟⎟
⎝ 0 0 0 0 p55 0 ⎠
0 0 0 0 0 p55
with p66 = (p11 − p12 )/2.
Therefore, suitable rotations of the coordinate system allow to construct different
anisotropic medium.
Glossary
[Brown and Korriga, 1975] Brown, R.J.S., Korringa, J.: On the dependence of the elastic proper-
ties of a porous rock on the compressibility of a pore fluid. Geophysics, 40, 608–616 (1975)
[Burridge and Keller, 1981] Burridge, R., Keller, J.B.: Poroelasticity equations derived from mi-
crostructure. J. Acoust. Soc. Am., 70, 1140–1146 (1981)
[Cadoret et al., 1995] Cadoret, T., Marion, D., Zinszner, B.: Influence of frequency and fluid dis-
tribution on elastic wave velocities in partially saturated lime- stones. J. Geophys. Res., 100,
9789–9803 (1995)
[Carcione et al., 1991] Carcione, J.M., Kosloff, D., Behle, A.: Long wave anisotropy in stratified
media: a numerical test. Geophysics, 56, 245–254 (1991)
[Carcione, 1992] Carcione, J.M.: Anisotropic Q and velocity dispersion of finely layered media.
Geophys. Prosp., 40, 761–783 (1992)
[Carcione et al., 1994] Carcione, J.M., Böehm G., Marchetti, A.: Simulation of a CMP seismic
section. J. Seismic Expl., 3, 381–396 (1994)
[Carcione, 1996] Carcione, J.M.: Plane-layered models for the analysis of wave propagation in
reservoir environments. Geophys. Prosp., 44, 3–26 (1996)
[Carcione and Seriani, 1998] Carcione J.M., Seriani, G.: Seismic velocities in permafrost. Geo-
phys. Prosp., 46, 441–454 (1998)
[Carcione et al., 2000] Carcione, J.M., Gurevich B., Cavallini F.: A generalized Biot-Gassmann
model for the acoustic properties of shaley sandstones. Geophys. Prosp., 48, 539–557 (2000)
[Carcione and Tinivella, 2000a] Carcione, J.M., Tinivella U.: Bottom-simulating reflectors: Seis-
mic velocities and AVO effects. Geophysics, 65, 54–67 (2000)
[Carcione et al., 2003] Carcione, J.M., Helle, H.B., Pham, N.H.: Whites model for wave prop-
agation in partially saturated rocks: Comparison with po- roelastic numerical experiments.
Geophysics, 68, 1389–1398 (2003)
[Carcione et al., 2003] Carcione, J.M., Helbig, K., Helle, H.B.: Effects of pressure and saturating
fluid on wave velocity and attenuation of anisotropic rocks. Int. J. Rock Mech. Min. Sci., 40,
389–403 (2003)
[Carcione et al., 2004] Carcione, J.M., Cavallini, F., Santos, J.E., Ravazzoli, C.L., Gauzellino,
P.M.: Wave propagation in partially-saturated porous media: Simulation of a second slow
wave. Wave Motion, 39, 227–240 (2004)
[Carcione and Picotti, 2006] Carcione, J.M., Picotti, S.: P-wave seismic attenuation by slow-wave
diffusion, Effects of inhomogeneous rock properties. Geophysics, 71, 1–8 (2006)
[Carcione et al., 2012] Picotti, S., Carcione, J.M., Gei, D., Rossi, G., Santos, J.E.: Seismic model-
ing to monitor CO2 geological storage - 1 the Atzbach-Schwanenstadt gas field. J. Geophys.
Res., 117, 1–18 (2012)
[Carcione et al., 2012] Carcione, J.M., Santos, J.E., Picotti, S.: Fracture induced anisotropic at-
tenuation. Rock Mech. Rock Eng., 45, 929–942 (2012)
[Carcione, 2014] Carcione, J.M.: Wave fields in real media: wave propagation in anisotropic,
anelastic, porous and electromagnetic media. 3rd ed., Elsevier (2014)
[Carcione et al., 2013] Carcione, J.M., Gurevich, B., Santos, J.E., Picotti, S.: Angular and fre-
quency dependent wave velocity and attenuation in fractured porous media. Pure Appl. Geo-
phys., 170, 1673–1683 (2013)
[Chadwick et al., 2005] Chadwick, R.A., Arts, R., Eiken, O.: 4D seismic quantification of a grow-
ing CO2 plume at Sleipner, North Sea. In: Dore, A.G. and Vincent, B. (eds.) Petroleum Geol-
ogy: North West Europe and Global Perspectives - Proc. 6th Petroleum Geology Conference,
1385–1399 (2005)
[Chadwick et al., 2009] Chadwick, R.A., Noy, D., Arts, R., Eiken, O.: Latest time-lapse seismic
data from Sleipner yield new insights into CO2 plume development. Energy Procedia, 1,
2103–2110 (2009)
[Chadwick et al., 2010] Chadwick, R.A., Williams, G., Delepine, N., Clochard, V., Labat, V., Stur-
ton, S., Buddensiek, M., Dillen, M., Nickel, M., Lima, A., Arts, R., Neele, F., Rossi, G.: Quan-
titative analysis of time-lapse seismic monitoring data at the Sleipner CO2 storage operation.
The Leading Edge, 29, 170–77 (2010)
[Ciarlet, 1976] Ciarlet, P.: The finite elements method for elliptic problems. North Holland,
Amsterdam-New York-Oxford (1976)
References 305
[Collins, 1961] Collins, R.E.: Flow of fluids through porous materials. Reinhold Publishing Co.,
New York (1961)
[Chavent and Jaffré, 1986] Chavent, G. and Jaffré, J.: Mathematical models and finite elements
for reservoir simulation. Elsevier Science Publishers B.V., North Holland (1986)
[Corapcioglu, 1996] Corapcioglu, M.Y.: Advances in porous media, Volume 3. Elsevier Science
(1996)
[Deptuck et al., 1985] Deptuck, D., Harrison, J.P., Zawadzki, P.: Measurement of elasticity and
conductivity in a three-dimensional percolation system. Phys. Rev. Lett., 54, 913–916 (1985)
[Douglas Jr. et al., 1993] Douglas, J. Jr., Hensley, J.L., Paes Leme, P.J.: A study of the effect of in-
homogeneities on immiscible flow in naturally fractured reservoirs. Int. Series Numer. Math.
(Birkhauser Verlag Basel), 114, 59–74 (1993)
[Douglas Jr. et al., 1993a] Douglas, J. Jr., Santos, J.E., Sheen, D., Bennethum, L.S.: Frequency
domain treatment of one-dimensional scalar waves. Math. Models Meth. Appl. Sci., 3, 171–
194 (1993)
[Douglas Jr. et al., 1994] Douglas, J. Jr., Santos, J.E., Sheen, D.: Approximation of scalar waves
in the space-frequency domain, Math. Models Meth. Appl. Sci., 4, 509–531 (1994)
[Douglas Jr. et al., 1997] , Douglas, J. Jr, Furtado, F., Pereira, F.: The statistical behavior of insta-
bilities in immiscible displacement subject to fractal geology. In: Burgeaut, A., Carasso, C.,
Luckhaus, S., Mikelic, A. (eds.). Mathematical Modeling of Flow Through Porous Media,
pp. 115–137, World Scientific, Singapore (1997)
[Douglas Jr. et al., 1999] Douglas, J. Jr., Santos, J.E., Sheen, D., Ye, X.: Nonconforming Galerkin
methods based on quadrilateral elements for second order elliptic problems. Math. Modell.
Numer. Analysis, 33, 747–770 (1999)
[Douglas Jr. et al., 2001] Douglas, J. Jr., Santos, J.E., Sheen, D.: Nonconforming Galerkin meth-
ods for the Hemholtz equation. Numer. Meth. Part. Diff. Equat., 17, 475–494 (2001)
[Dutta and Odé, 1979] Dutta, N.C., Odé, H.: Attenuation and dispersion of compressional waves
in fluid-filled porous rocks with partial gas saturation (White model) - Part I: Biot theory.
Geophysics, 44, 1777–1788 (1979)
[Dutta and Odé, 1983] Dutta, N.C. and Odé, H.: Seismic reflections from a gas-water contact.
Geophysics, 48, 148–162 (1983)
[Dvorkin et al., 1995] Dvorkin, J., Mavko, G., Nur, A.: Squirt flow in fully saturated rocks. Geo-
physics, 60, 97–107 (1995)
[Fabrizio and Morro, 1992] Fabrizio, M., Morro, A.: Mathematical problems in linear viscoelas-
ticity. SIAM Studies in Applied Mathematics, 12, Philadelphia (1992)
[Fanchi, 1997] Fanchi, J.R.: Principles of Applied Reservoir Simulation. Gulf Professional Pub-
lishing Co., Houston, Texas (1997)
[Ferry, 1970] Ferry, J.D.: Viscoelastic properties of polymers. John Wiley (1970)
[Frankel and Clayton, 1986] Frankel, A., Clayton, R.W.: Finite difference simulation of seismic
wave scattering: implications for the propagation of short period seismic waves in the crust
and models of crustal heterogeneity. J. Geophys. Res., 91, 6465–6489 (1986)
[Fung, 1965] Fung, Y.C.: Foundations of solid mechanics. Prentice Hall, Englewood Cliffs, New
Jersey (1965)
[Gassmann, 1951] Gassmann, F.: Uber die Elastizitat poroser Mediem. Vierteljahrsschr Natur-
forsch. Ges. Zurich, 96, 1–23 (1951)
[Gauzellino et al., 2001] Gauzellino, P.M., Santos, J.E., Sheen D.: Frequency domain wave prop-
agation modelling in exploration seismology. J. Comp. Acoust., 12, 941–955 (2001)
[Gauzellino et al., 2014] Gauzellino, P.M., Santos, J.E., Carcione, J.M., Picotti, S.: A rheological
equation for anisotropic-anelastic media and simulation of field seismograms. Wave Motion,
51, 743–757 (2014)
[Gelinsky and Shapiro, 1997] Gelinsky, S., Shapiro, S.A.: Poroelastic Backus-averaging for
anisotropic, layered fluid and gas saturated sediments. Geophysics, 62, 1867–1878 (1997)
[Gurevich and Lopatnikov, 1995] Gurevich, B., Lopatnikov, S.L.: Velocity and attenuation of
elastic waves in finely layered porous rocks. Geophys. J. Int., 121, 933–947 (1995)
[Gurevich et al., 1997] Gurevich, B., Zyrianov, V.B., Lopatnikov, S.L.: Seismic attenuation in fin-
ley layered porous rocks: Effects of fluid flow and scattering. Geophysics, 62, 319–324 (1997)
306 References
[Gurevich, 2003] Gurevich, B.: Elastic properties of saturated porous rocks with aligned fractures.
J. Appl. Geophys., 54, 203–218 (2003)
[Gurevich et al., 2009] Gurevich, B., Brajanovski, M., Galvin, R.J., Müller, T.M., Toms-Stewart,
J.: P-wave dispersion and attenuation in fractured and porous reservoirs-poroelasticity ap-
proach. Geophys. Prosp., 57, 225–237 (2009)
[Ha et al., 2002] Ha, T., Santos, J.E., Sheen, D.: Non-conforming finite element methods for the
simulation of waves in viscoelastic solids. Comp. Meth. Appl. Mech. Eng., 191, 5647–5670
(2002)
[Han et al., 1986] Han, D.H., Nur, A., Morgan, D.: Effects of porosity and clay content on wave
velocities in sandstones. Geophysics, 51, 2093–2107 (1986)
[Hashin and Shtrikman, 1963] Hashin, Z., Shtrikman, S.: A variational approach to the elastic
behavior of multiphase minerals. J. Mech. and Phys. Solids, 11, 127–140 (1963)
[Hassanzadeh et al., 2008] Hassanzadeh, H., Pooladi-Darvish, M., Elsharkawy, A., Keith, D.,
Leonenko, Y.: Predicting PVT data for CO2 -brine mixtures for black-oil simulation of CO2
geological storage. Int. J. Greenhouse Gas Control, 2, 65–77 (2008)
[Helle et al., 2003] Helle, H.B., Pham, N.M., Carcione, J.M.: Velocity and attenuation in partially
saturated rocks: poroelastic numerical experiments. Geophys. Prosp., 51, 551–566 (2003)
[Hill, 1964] Hill, R.: Theory of mechanical properties of fiber-strengthened materials. J. Mech.
Phys. Solids, 11, 357–372 (1964)
[Hovem and Ingram, 1979] Hovem, J.M., Ingram, G.D.: Viscous attenuation of sound in saturated
sand. J. Acoust. Soc. Am., 66, 1807–1812 (1979)
[Ihlenburg and Babuska, 1997] Ihlenburg, F., Babushka, I.: Solution of Helmholtz problems by
knowledge-based FEM. Comp. Assisted Mech. Eng. Sci., 4, 397–415 (1997)
[Johnson et al., 1987] Johnson D.L., Koplik J., Dashen R.: Theory of dynamic permeability and
tortuosity in fluid-saturated porous media. J. Fluid Mech., 176, 379–402 (1987)
[Johnson, 2001] Johnson, D.L.: Theory of frequency dependent acoustics in patchy-saturated
porous media. J. Acoust. Soc. Am., 110, 682–694 (2001)
[Jones, 1986] Jones, T.D.: Pore-fluids and frequency dependent wave propagation rocks. Geo-
physics, 51, 1939–1953 (1986)
[Keller, 1989] Keller, J.D.: Acoustic wave propagation in composite fluid-saturated media. Geo-
physics, 54, 1554–1563 (1989)
[Kelder and Smeulders, 1997] Kelder, O., Smeulders, D.: Observation of the Biot slow wave in
water-saturated Nivelsteiner sandstone. Geophysics, 62, 1794–1796 (1997)
[Dahlberg et al., 1988] Dahlberg, B.E.J., Kenig, C.E., Verchota, G.C.: Boundary value problems
for the systems of elastostatics in Lipschitz domains. Duke Math. J., 57, 795–818 (1988)
[Kim, 1995] Kim, S.: Parallel multidomain iterative algorithms for the Helmholtz wave equation.
Appl. Numer. Math., 17, 411–429 (1995)
[Kolmogorov and Fomin, 1970] Kolmogorov, A.N., Fomin, S.V.: Introductory Real Analysis.
Dover Publications, Inc., New York (1970)
[Kolsky, 1963] Kolsky, H.: Stress Waves in Solids. Dover publications, New York (1963)
[Krief et al., 1990] Krief M., Garat J., Stellingwerff J., Ventre J.: A petrophysical interpretation
using the velocities of P and S waves (full waveform sonic). The Log Analyst, 31, 355–369
(1990)
[Krzikalla and Müller, 2011] Krzikalla, F., Müller, T. M.: Anisotropic P-SV-wave dispersion and
attenuation due to inter-layer flow in thinly layered porous rocks. Geophysics, 76, WA135–
WA145 (2011)
[Kuster and Toksöz, 1974] Kuster, G.T., Toksöz, M.N.: Velocity and attenuation of seismic waves
in two-phase media: Part 1. Theoretical formulations. Geophysics, 39, 587–606 (1974)
[Leclaire et al., 1994] Leclaire, P., Cohen-Tenoudji, F., Aguirre Puente, J.: Extension of Biots the-
ory to wave propagation in frozen porous media. J. Acoust. Soc. Am., 96, 3753–3767 (1994)
[Leclaire et al., 1985] Leclaire, P., Cohen-Tenoudji, F., Aguirre Puente, J.: Observation of two
longitudinal and two transverse waves in a frozen porous medium. J. Acoust. Soc. Am., 97,
2052–2055 (1995)
[Levy, 1979] Levy, T.: Propagation of waves in a fluid-saturated porous elastic solid. Int. J. Eng.
Sci., 17, 1005–1014 (1979)
References 307
[Liu et al., 1976] Liu, H.P., Anderson, D.L., Kanamori, H.: Velocity dispersion due to anelasticity;
implications for seismology and mantle composition. Geophys. J. R. Astr. Soc., 47, 41–58
(1976)
[Lo et al., 2005] Lo, W.C., Sposito, G., Majer, E.: Wave propagation through elastic porous media
containing two immiscible fluids. Water Resour. Res., 41, W02025, 1–20 (2005)
[Loeve, 1944] Loeve, A.E.: A Treatise on the Mathematical Theory of Elasticity. 4th ed., Dover,
New York (1944)
[Lovera and Santos, 1988] Lovera O.M., Santos J.E.: Numerical methods for a model for wave
propagation in composite anisotropic media. Math. Modell. Numer. Analysis, 22, 159–176
(1988)
[Lumley and Panofsky, 1964] Lumley, J., Panofsky, H: The structure of atmospheric turbulence.
Interscience Publishers, New York (1964)
[Masson and Pride, 2007] Masson, Y.J., Pride, S.R.: Poroelastic finite difference modelling of
seismic attenuation and dispersion due to mesoscopic-scale heterogeneity. J. Geophys. Res.,
112, B03204 (2007)
[Mavko et al., 1998] Mavko, G., Mukerji, T., Dvorkin, J.: The Rock Physics Handbook. Cam-
bridge University Press (1998)
[McCoy, 1983] McCoy, R.L.: Microcumputers programs for Petroleum Engineers, Volume 1,
Gulf Publishing Co. Houston, Texas (1983)
[Mochizuki, 1982] Mochizuki, S.: Attenuation in partially saturated rocks. J. Geophys. Res., 87,
8598–8604 (1982)
[Morack and Rogers, 1981] Morack, J.L., Rogers, J.C.: Seismic evidence of shallow permafrost
beneath the islands in the Beafort Sea. Artic, 3, 166–174 (1981)
[Müller and Gurevich, 2005] Müller, T., Gurevich, B.: Wave-induced fluid flow in random porous
media: Attenuation and dispersion of elastic waves. J. Acoust. Soc. Am., 117, 2732–2741
(2005)
[Müller and Toms, 2007] Müller, T., Toms, J.: Velocity-saturation relation for rocks with fractal
distribution of the pore fluids. In: 77th Ann. Internat. Mtg., Soc. Expl. Geophys., San Antonio,
USA, Expanded Abstracts, 1624–1628 (2007)
[Nakagawa and Schoenberg, 2007] Nakagawa, S., Schoenberg, M.A.: Poroelastic modeling of
seismic boundary conditions across a fracture. J. Acoust. Soc. Ame., 122, 831–847 (2007)
[Nedelec, 1980] Nedelec, J.C.: Mixed finite elements in R3. Numerische Mathematik, 35, 315–
341 (1980)
[Norris, 1993] Norris, A.N.: Low-frequency dispersion and attenuation in partially saturated
rocks. J. Acoust. Soc. Am., 94, 359–370 (1993)
[O’Connell and Budiansky, 1974] OConnell, R., Budiansky, B.: Seismic velocities in dry and sat-
urated cracked solids. J. Geophys. Res., 79, 5412–5426 (1974)
[Picotti et al., 2010] Picotti, S., Carcione, J.M., Santos, J.E., Gei, D.: Q-anisotropy in finely-
layered media. Geophys. Res. Lett., 37, L06302 (2010)
[Picotti et al., 2012] Picotti, S, Carcione, J.M., Gei, D., Rossi, G., Santos, J.E.: Seismic modeling
to monitoring CO2 geological storage. The Atzbach-Schwanenstadt gas field. J. Geophys.
Res., 117, B06103 (2012)
[Peaceman, 1977] Peaceman, D.W.: Fundamentals of numerical reservoir simulation. Elsevier
(1977)
[Peng and Robinson, 1976] Peng, D.Y., Robinson, K.I.: A new two-constant equation of state, Ind.
Eng. Chem. Fundam., 15, 59–64 (1976)
[Pride and Berryman, 2003] Pride, S.R., Berryman, J.G.: Linear dynamics of double-porosity and
dual-permeability materials. I. governing equations and acoustic attenuation. Phys. Rev. E,
68, 036603 (2003)
[Pride et al., 2004] Pride, S.R., Berryman, J.G., Harris, J.M.: Seismic attenuation due to wave-
induced flow. J. Geophys. Res., 109, B01201 (2004)
[Pyrak-Nolte et al., 1990] Pyrak-Nolte, L.J., Myer, L.R., Cook, N.G.: Anisotropy in seismic ve-
locities and amplitudes from multiple parallel fractures. J. Geophys. Res., 95, 11345–11358
(1990)
308 References
[Rasolofosaon, 1991] Rasolofosaon, P.: Plane acoustic waves in linear viscoelastic porous media:
energy, particle displacement and physical interpretation. J. Acoust. Soc. Am., 89, 1532–1550
(1991)
[Ravazzoli and Santos, 1995] Ravazzoli, C.L., Santos, J.E.: Consistency analysis for a model for
wave propagation in anelastic media. Latin Am. Appl. Res., 25, 141–152 (1995)
[Ravazzoli et al., 2003] Ravazzoli, C.L., Santos, J.E., Carcione, J.M.: Acoustic and mechanical
response of reservoir rocks under varaible saturation and effective pressure. J. Acoust. Soc.
Am., 113, 1801–1811 (2003)
[Raviart and Thomas, 1977] Raviart, P.A., Thomas, J.M.: A mixed finite element method for sec-
ond order elliptic problems. In: Galligani, I., Magenes, E. (eds.). Mathematical Aspects of
the Finite Element Method, Element Method, Lecture Notes in Mathematics, 606, 292–315
(1977)
[Rubino et al., 2007] Rubino, J.G., Santos, J.E., Picotti, S., Carcione, J.M.: Simulation of upscal-
ing effects due to wave-induced fluid flow in Biot media using the finite-element method. J.
Appl. Geophys., 62, 193–203 (2007)
[Rubino et al., 2008] Rubino, J.G., Ravazzoli, C.L., Santos, J.E.: Biot-type scattering effects in
gas-hydrate bearing sediments. J. Geophys. Res., Solid Earth, 113, B06102, 1–16 (2008)
[Russo and Bouton, 2007] Russo, D., Bouton, M.: Statistical analysis of spatial variability in un-
saturated flow parameters. Water Resour. Res., 28, 1911–1925 (1992)
[Russo et al., 1997] Russo, D., Russo, I., Laufer, A.: On the spatial variability of parameters of
the unsaturated hydraulic conductivity. Water Resour. Res., 33, 945–956 (1997)
[Saenger et al., 2007] Saenger E.H., Ciz, R., Krüger, O.S., Schmalholz, S.M., Gurevich, B.,
Shapiro, S.A.: Finite-difference modeling of wave propagation on microscale: A snapshot
of the work in progress. Geophysics 72, SM293–SM300 (2007)
[Sanchez Palencia, 1980] Sanchez-Palencia, E.: Non-homogeneous media and vibration theory.
In: Lecture Notes in Physics, Springer-Verlag, New York (1980)
[Santos et al., 1988] Santos, J.E., Douglas, J. Jr., Morley, M.E., Lovera., O.M.: Finite element
methods for a model for full waveform acoustic logging. IMA, J. Numer. Analysis, 8, 415–
433 (1988)
[Santos et al., 1990a] Santos, J.E., Corberó, J.M., Douglas, J. Jr.: Static dynamic behavior of a
porous solid saturated by a two-phase fluid. J. Acoust. Soc. Am., 87, 1428–1438 (1990)
[Santos et al., 1990b] Santos, J.E., Douglas, J. Jr., Corberó, J.M., Lovera, O.M.: A model for wave
propagation in a porous solid saturated by a two-phase fluid. J. Acoust. Soc. Am., 87, 1439–
1448 (1990)
[Santos et al., 1992] Santos, J.E., Corberó, J.M., Ravazzoli, C.L., Hensley, J.L.: Reflection and
transmission coefficients in fluid saturated porous media. J. Acoust. Soc. Am., 91, 1911–1923
(1992)
[Santos et al., 2004a] Santos, J.E., Ravazzoli, C.L., Carcione, J. M.: A model for wave propaga-
tion in a porous solid saturated by a two-phase fluid. J. Acoust. Soc. Am., 87, 1439–1448
(1990)
[Santos et al., 2004b] Santos, J.E., Ravazzoli, C.L., Gauzellino, P.M., Carcione, J.M., Cavallini,
F.: Simulation of Waves in Poro-viscoelastic Rocks Saturated by Immiscible Fluids. Numeri-
cal Evidence of a Second Slow Wave. J. Comp. Acoust., 12, 1–21 (2004)
[Santos et al., 2005a] Santos, J.E., Ravazzoli, C.L., Gauzellino, P.M., Carcione, J.M.: Numerical
simulation of ultrasonic waves in reservoir rocks with patchy saturation and fractal petrophys-
ical properties. Comp. Geosciences, 9, 1–27 (2005)
[Santos et al., 2005b] Santos, J.E., Ravazzoli, C.L., Geiser, J.: On the static and dynamic behavior
of fluid saturated composite porous solids; a homogenization approach. Int. J. Solids and
Structures, 43, 1224–1238 (2005)
[Santos et al., 2011] Santos, J.E., Carcione, J.M., Picotti, S.: Viscoelastic-stifness tensor of
anisotropic media from oscillatory numerical experiments. Comp. Meth. Appl. Mech. Eng.,
200, 896–904 (2011)
[Santos et al., 2012] Santos, J.E., Picotti, S., Carcione, J.M.: Evaluation of the stiffness tensor of
a fractured medium with harmonic experiments. Comp. Meth. Appl. Mech. Eng., 247-248,
130–145 (2012)
References 309
[Santos et al., 2014a] Santos, J.E., Martinez Corredor, R., Carcione, J.M.: Seismic velocity and Q
anisotropy in fractured poroelastic media. Int. J. Rock Mech. Min. Sci., 70, 212–218 (2014)
[Santos et al., 2014b] Santos, J.E., Savioli, G.B., Macias, L., Carcione, J.M., Gei, D.: Influence of
capillary pressure on CO2 storage and monitoring. In: Proceedings of the 84th Ann. Internat.
Mtg., Soc. Expl. Geophys., Denver, USA, Expanded Abstracts, 2212–2215 (2014)
[Santos and Carcione, 2015] Santos, J.E., Carcione, J.M.: Finite-element harmonic experiments
to model fractured induced anisotropy in poroelastic media. Comp. Meth. Appl. Mech. Eng.,
283, 1189–1213 (2015)
[Santos and Savioli, 2015] Santos, J.E., Savioli, G.B.: A parametric analysis of waves propagating
in a porous solid saturated by a three-phase fluid. J. Acoust. Soc. Am., 138, 3033–3042 (2015)
[Santos and Savioli, 2016] Santos, J.E., Savioli, G.B.: A model for wave propagation in a porous
solid saturated by a three-phase fluid. J. Acoust. Soc. Am., 139, 693 (2016)
[Savioli and Bidner, 2005] Savioli, G.B., Bidner, M.S.: Simulation of the oil and gas flow toward
a well - a stability analysis. J. Petroleum Sci. Eng., 48, 53–69 (2005)
[Scheidegger, 1974] Scheidegger, A.E.: The physics of flow through porous media. Toronto: Uni-
versity of Toronto (1974)
[Schoenberg, 1980] Schoemberg, M.: Elastic wave behavior across linear slip interfaces. J.
Acoust. Soc. Am., 68, 1516–1521 (1980)
[Sheen, 1993] Sheen, D.: Finite element methods for an acoustic well-logging problem associated
with a porous medium saturated by a two-phase immiscible fluid. Numer. Meth. Part. Diff.
Equat., 9, 155–174 (1993)
[Spycher and Pruess, 2005] Spycher, N., Pruess, K.: CO2 -H2 O mixtures in the geological seques-
tration of CO2 . II. Partitioning in chloride brines at 12-100 C and up to 600 bar. Geochim.
Cosmochim, Acta 69, 13, 3309–3320 (2005)
[Standing, 1977] Standing, M.B.: Volumetric and phase behavior of oil field hydrocarbon systems.
Soc. Petroleum Eng. AIME, Dallas (1977)
[Stoll and Bryan, 1970] Stoll R.D., Bryan, G.M.: Wave attenuation in saturated sediments. J.
Acoust. Soc. Am., 47, 1440–1447 (1970)
[Stoll, 1974] Stoll, R.D.: Acoustic waves in saturated sediments. In: Hampton L. (ed.). Physics of
Sound in Marine Sediments, pp 19–39, Plenum Press (1974)
[Toms et al., 2007] Toms, J., Müller, T., Gurevich, B.: Seismic attenuation in porous rocks with
random patchy saturation. Geophys. Prosp., 55, 671–678 (2007)
[Wenzlau et al., 2010] Wenzlau, F., Altmann, J.B., Müller, T.M.: Anisotropic dispersion and atten-
uation due to wave-induced flow: quasi-static finite element modeling in poroelastic solids. J.
Geophys. Res., 115, B07204 (2010)
[White et al., 1975] White, J.E., Mikhaylova, N.G., Lyakhovitskiy, F.M.: Low-frequency seismic
waves in fluid-saturated layered rocks. In: Izvestija Academy of Sciences USSR, Physics
Solid Earth, 10, 654–659 (1975)
[White, 1975] White, J.E.: Computed seismic speeds and attenuation in rocks with partial gas
saturation. Geophysics, 40, 224–232 (1975)
[Wood, 1955] Wood, A.W.: A Textbook of Sound. MacMillan Co. (1955)
[Yosida, 1980] Yosida, K.: Funcional Analysis. 6th ed., Springer-Verlag, Berlin Heidelberg-New
York (1980)
[Zener, 1948] Zener, C.: Elasticity of anelastic metals. The University of Chicago Press (1948)
[Zimmerman et al, 1986] Zimmerman, R.W., Somerton, W.H., King, M.S.: Compressibility of
porous rocks. J. Geophys. Res., 91 NOB12, 12765–12777 (1986)
[Zyserman et al., 2003] Zyserman, F.I., Gauzellino, P.M., Santos, J.E.: Dispersion analysis of a
nonconforming finite element method for the Hemholtz and elastodynamic equations. Int. J.
Numer. Meth. Eng., 58, 1381-1395 (2003)
[Zyserman and Gauzellino, 2005] Zyserman, F.I., Gauzellino, P.M.: Dispersion analysis of a non-
conforming finite element method for the three dimensional scalar and elastic wave equations.
Finite Element in Analysis and Design, 41, 1309–1326 (2005)