Reactive transport in damageable geomaterials Thermal-Hydrological-Mechanical-Chemical coupling of geological processes This thesis is presented to the School of Earth and Environment for the degree of Doctor of Philosophy By Thomas Poulet March 2012
Embed
Reactive transport in damageable geomaterials · Reactive transport in damageable geomaterials Thermal-Hydrological-Mechanical-Chemicalcouplingof geologicalprocesses This thesis is
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Reactive transport in damageablegeomaterials
Thermal-Hydrological-Mechanical-Chemical coupling of
In this dissertation, compact notation is used where X denotes a scalar value, X a
vector, and X a matrix or tensor. Here, the number of underlines shows the order
of the matrix/tensor. For instance, C represents a fourth order tensor, and the
1.5. NOTATIONS 9
notation is expended to the fully indexed tensorial form Cijkl when more clarity is
required. The Einstein convention of implicit summation of repeated indices is also
used unless specified otherwise, leading to the expression of tensorial products as
A.B = AijBjk and double contractions as A : B = AijBji.
A synoptic table of symbols is presented below, and all symbols are defined indi-
vidually in each chapter.
Ω0, Ω Physical domain in reference and current configurations
Γ Physical domain surface in current configuration
D(.)Dt
Material derivative attached to the mixture (material points related
to the solid or fluid phase)
d(.)dt
= ˙(.) Dot notation, material derivative attached to the skeleton
δij Kronecker delta
1 Identity tensor
α Thermal expansion coefficient [K−1]
β Compressibility [Pa−1]
Cv, Cp Isochoric and isobaric heat capacities [J.K−1]
D Damage parameter [-]
ǫ Strain tensor [-]
g Gravity [m.s−2]
G Gibbs free energy [J ]
H Enthalpy [J ]
K Kinetic energy [J ]
k Thermal conductivity [W.m−1.K−1]
K,G Bulk and shear moduli [Pa]
µ Viscosity [Pa.s]
N , b Biot modulus [Pa] and coefficient [-]
P Pore pressure [Pa]
φ Porosity [-]
ψ Helmholtz free energy [J ]
q Darcy flux [m.s−1]
ρ Density [kg.m−3]
σ Cauchy stress tensor [Pa]
S Entropy [J.K−1]
T Absolute temperature [K]
10 CHAPTER 1. INTRODUCTION
u Fluid velocity [m.s−1]
U Internal energy [J ]
Chapter 2
Consistent material properties
2.1 Introduction
Material properties are essential to realistic simulations of geodynamic and geolog-
ical processes. These properties generally derive from laboratory experiments and
geophysical observations (Bina, 1998a; Bina and Wood, 1987; Cammarano et al.,
2003; Deschamps and Trampert, 2004; Deuss et al., 2006; Irifune and Isshiki, 1998;
Irifune and Ringwood, 1987; Matas et al., 2007; Ringwood, 1991; Trampert et al.,
2001; Weidner, 1985). Additionally, numerical data calculated from thermody-
namic potentials are a complement to empirical data (Ita and Stixrude, 1992; Karki
et al., 2001; Stixrude and Lithgow-Bertelloni, 2005a; Vacher et al., 1998). There has
been considerable progress in geodynamic, thermodynamic and petrological mod-
elling (Connolly, 2005; Connolly and Petrini, 2002; Matas et al., 2007; Stixrude and
Lithgow-Bertelloni, 2005a,b). Here, a tool is presented for coupling thermochem-
istry with mechanics. The main purpose of this tool is to provide geodynamicists
and seismologists easy access to thermochemistry.
The thermodynamic equilibrium problem as solved by Gibbs energy minimisation
determines the basic mechanical properties needed for geophysics and geodynam-
ics. Reversible material properties such as the thermal expansion coefficient, specific
11
12 CHAPTER 2. CONSISTENT MATERIAL PROPERTIES
heat, elastic shear modulus, bulk modulus and density can thus be derived from
thermodynamics and thermodynamic potential functions can be used to model geo-
dynamical processes. Specifically, density differences driving for instance subduc-
tion do not need to be assigned but follow from chemical composition and tem-
perature. Moreover, reversible material property changes at phase transitions can
trigger events that may be of interest to the geodynamicists and seismologists (Bina,
1998b).
Previous works on thermodynamically derived material data have been presented in
the seismic tomographic community with the focus of solving the inverse problem
for chemistry from seismic velocities (Bina and Wood, 1987; Duffy and Ander-
son, 1989; Weidner, 1985). These models gather a collection of physical properties
from independent sources and are not thermodynamically self-consistent as pointed
out by several authors (Connolly, 2005; Connolly and Kerrick, 2002; Stixrude and
Lithgow-Bertelloni, 2005b). Stixrude and Lithgow-Bertelloni (2005b) have extended
the thermodynamic formulation by tensorial presentation of stress and the relation-
ship to entropy and temperature. Therefore, the elastic shear modulus is derived
without ad hoc assumptions. This formulation, among others, is implemented in
the software Perple X (Connolly, 2005) which permits extraction of any ther-
modynamic property as a generalised formulation of temperature, pressure and
composition.
Other codes have also been put forward (ThermoCalc, Domino, FreeGs) with differ-
ent thermodynamic solvers that focus on determining phase equilibria in petrolog-
ical systems regardless of the self consistency of material properties. These codes
use non-linear techniques for Gibbs energy minimisation. The strength of these
non-linear methods is their accuracy. However their weakness is that identification
of the stable mineral assemblage is probable but not certain. This lack of robust-
ness obviates the use of non-linear methods for embedded geodynamic calculations.
In contrast, Perple X utilises a linearised formulation of the minimisation prob-
lem which always converges (Connolly, 1990, 2005; Connolly and Petrini, 2002).
2.2. METHODS 13
For the derivation of self consistent material properties, this linearised algorithm is
convenient because it minimises data gaps. Another important aspect is that for
the purpose of deriving material data the best strategy is to reject mixed formu-
lations, where discrepancies with field observations are adjusted without changing
the underlying basic entropy model. Emergence of thermodynamic solvers for geo-
dynamic processes thus puts a strong constraint on basic consistency of the dataset.
All these reasons lead to use Perple X. A reference database is presented here for
the purpose of standardising material properties to be used particularly in geolog-
ical, geodynamic and geotechnical calculations. The numerical results have been
validated by comparing seismic velocities predicted for a pyrolitic composition to
the seismic models PREM (Dziewonski and Anderson, 1981) and ak135 (Kennett
et al., 1995; Montagner and Kennett, 1996). These models, constructed after travel
time data, give access to detailed information about the average Earth’s structure.
2.2 Methods
The purpose of PreMDB is to provide modellers with a complete and easy access
to fundamental material data for terrestrial rocks and minerals. Another goal is to
standardise material data in order to compare results from various numerical and
experimental techniques. In order to satisfy these requirements, thermodynami-
cally consistent data have been chosen as they are defined over a large range of
temperature and pressure and provide an important complement to experiments
and observations. Non-thermodynamic data such as transport or ad hoc properties
have also been added. Currently, PreMDB lists 20 material properties for each
rock and mineral (table 2). In isothermal-isobaric closed chemical system composed
of Π phases, the phase equilibria are determined minimising the Gibbs free energy of
the system (Gsys). This thermodynamic function is a function of temperature (T ),
pressure (P ) and the chemical composition of the system (Connolly, 2005; Connolly
14 CHAPTER 2. CONSISTENT MATERIAL PROPERTIES
and Kerrick, 2002). It is defined as:
Gsys(P, T, n) =p∑
i
niGmi (P, T, x1i , x2i , ..., xci) , (1)
where Gmi and ni are respectively the molar Gibbs energy and the number of moles
of phase i; xji is the composition of the ith phase with respect to the jth component
of the system.
Table 2: Physical properties available in PreMDB.
Properties calculated Symbol Units Thermodynamic-consistentin Perple X S&LB model* Other models
Enthalpy h J/kg
Specific enthalpy h× ρ J/m3
Entropy s J/K/kg
Specific entropy s× ρ J/K/m3
Isobaric heat capacity c J/K/kg
Specific heat c× ρ J/K/m3
Density ρ kg/m3
Thermal expansion α 1/K
Compressibility β 1/Pa
Bulk sound velocity Vφ km/s
P-wave velocity VP km/s
S-wave velocity VS km/s
Bulk modulus Ks GPa
Shear modulus G GPa
Elastic modulus E GPa
Poisson’s ratio ν
Gruneisen ratio γ
Additional transport propertiesThermal conductivity k W/K/mThermal diffusivity κ m2/sMelt viscosity η Pa.s
* The equations of state describing the solutions are from (Bina, 1998a; Cammaranoet al., 2003; Connolly and Kerrick, 2002; Duffy and Anderson, 1989). S&LB: Stixrudeand Lithgow-Bertelloni.
In Perple X, all reversible material properties are related to the thermodynamic
potentials G:
2.2. METHODS 15
• Entropy S
S = −(
∂G∂T
)
P
(2)
• Isochemical enthalpy H
H = G + TS (3)
• Internal Energy U
U = H − PV (4)
V being the volume.
• Heat capacity CP
CP = −T(
∂2G∂T 2
)
P
(5)
• Density ρ
ρ =N
V= N
∂G∂P
(6)
where N is the molar formula weight.
• Thermal expansion α
α = − 1
V
(
∂S
∂P
)
P
=1
V
(
∂2G∂P∂T
)
(7)
• Compressibility β
β = − 1
V
(
∂2G∂P 2
)
T
(8)
• Adiabatic bulk modulus Ks
Ks = − ∂G∂P
[
∂2G∂P 2
+
(
∂2G∂P∂T
)2
/∂2G∂T 2
]−1
(9)
• Gruneisen parameter γ
γ = V
(
∂P
∂U
)
V
(10)
16 CHAPTER 2. CONSISTENT MATERIAL PROPERTIES
Connolly and Kerrick (Connolly and Kerrick, 2002) use an ad hoc empirical
model (see equation 5 in (Connolly and Kerrick, 2002)) to calculate the shear
modulus. Stixrude and Lithgow-Bertelloni (2005b)’s formulation of Gibbs
energy for isotropic material permits to relate the shear and elastic moduli to
thermodynamic potentials:
• Elastic compliance tensor sijkl
sijkl = − 1
V
∂2G∂σij∂σkl
(11)
where σijkl is the stress tensor and the subscript on the derivative defining the
compliance means stress components except those involved in the derivative
are held constant. The elastic compliance tensor describes the general elastic
material behaviour in compression and shear. For an isotropic body there
are two independent quantities, e.g. bulk and shear moduli. For anisotropic
assemblages of phases, the shear modulus is essentially interpreted as that of
an isotropic polycrystalline aggregate.
• Shear modulus G, from one of its simplest forms
1
G=
1
V
n∑
ψ
(
xψVψ1
Gψ
)
(12)
For any equations of state other than those of Stixrude and Lithgow-Bertelloni
(2005a,b), the shear modulus calculated in Perple X is not computed self-
consistently from thermodynamic potentials (Connolly and Kerrick, 2002).
For isotropic elasticity the elastic modulus, as well as seismic properties are obtained
from the following equations:
• Elastic modulus E
E =9KsG
3Ks +G(13)
2.2. METHODS 17
• Poisson’s ratio ν
ν =3Ks − 2G
6Ks + 2G(14)
• Bulk sound velocity Vφ (Ita and Stixrude, 1992)
Vφ =
√
Ks
ρ(15)
• S-wave velocity VS (Ita and Stixrude, 1992)
VS =
√
G
ρ(16)
• P-wave velocity VP (Ita and Stixrude, 1992)
VP =
√
Ks + 4G/3
ρ(17)
Seismic wave velocities in a single crystal are thermodynamically self-consistent.
On the contrary, they are not in an aggregate of crystals and must be obtained by
an ad hoc averaging scheme such as the Voight-Reuss-Hill theory (Connolly and
Kerrick, 2002; Watt et al., 1976).
Transport properties are estimated from empirical models that are derived from
laboratory experiments and expressed as functions of the computed thermodynamic
properties:
• Thermal conductivity k Two equations have been implemented in PreMDB
to define k:
– As a function of the temperature T
k = A+B
350 + T(18)
18 CHAPTER 2. CONSISTENT MATERIAL PROPERTIES
where constants A and B are defined for different rock types. k is ex-
pressed in W/m/K and T in Celsius (Clauser and Huenges, 1995; Zoth
and Hanel, 1988).
– As a function of the P-wave velocity VP
k = 0.0681e0.0006VP (19)
with k in W/m/K, VP in m/s (Ozkahraman et al., 2004).
• Thermal diffusivity κ
κ =k
ρCP(20)
where k is defined according to eq. (19), ρ and CP derived from Perple X.
• Melt viscosity η
η = A+B
T+ exp
(
C +D
T
)
(21)
where A, B, C and D are linear functions of mole fractions of oxide compo-
nents, except for H2O (Hui and Zhang, 2007).
Depending on the complexity of the system and the precision required, the linearised
Gibbs energy minimisation problem can be time consuming. For this reason, com-
puted phase diagram sections and material properties are stored in the PreMDB
database. At present compositions of 48 major rock forming dry and wet miner-
als and 9 terrestrial rocks have been incorporated, representing a standard for the
sedimentary part of the crust (Plank and Langmuir, 1998), the upper and lower
continental crust (Rudnick and Fountain, 1995; Taylor and McLennan, 1985), the
oceanic crust (Staudigel et al., 1996) and the mantle (pyrolite and peridotite) (Hart
and Zindler, 1986; Ringwood, 1979). The rocks and minerals currently described in
PreMDB are listed in Table 3.
A graphical user interface has been developed to browse the database and plot all
2.2. METHODS 19
Table 3: Terrestrial rocks and minerals available in PreMDB
properties as 2D graphs. The visual representation of all properties is an impor-
tant role of PreMDB as it is generally cumbersome to interpret or validate data
in a tabulated format directly from text files as Perple X produces them. Re-
versible properties computed from Perple X can be visualised as a scale coloured
2D map function of temperature and pressure. Transport properties are either vi-
sualised the same way (thermal conductivity derived from eq. (19) and diffusivity
from eq. (20) or represented as functions of the temperature (thermal conductiv-
ity from eq. (18) and melt viscosity from eq. (21)). Source (Perple X/equation)
and references of each property are presented in the graphical interface. Python
(http://www.python.org) is used as a scripting language to post-process data from
Perple X and convert them to a specific internal data structure. The graphical
user interface is implemented with wxPython (http://www.wxpython.org) and the
plots are rendered with matplotlib (http://matplotlib.sourceforge.net/).
Figure 1: Example of PreMDB’s GUI showing a wet peridotite composition satu-rated in H2O. The major density jumps reflect dewatering reactions predicted fromthermodynamic modelling.
2.3. RESULTS 21
PreMDB’s main window lists all materials of the database regrouped by cate-
gories under different tabs. There are currently two categories: earth materials and
major rock forming minerals. Each category displays a table listing sequentially dif-
ferent rocks/minerals along with their name, composition, comments, Perple X
input files (when applicable) and resulting pseudo-sections. Double clicking on a
rock/mineral line opens a new window composed of two distinct parts. The first
one lists all material properties available and the second one displays extra infor-
mation and references on any property selected. Select a property to visualise it.
In the case of transport properties, an option is offered for each available equation.
The visualisation pops up in another window as the graphical representation of
the underlying equation or tabulated data (Fig. 1). Basic functionalities are then
available, for instance to magnify the picture, obtain data values under the mouse
in the status bar or save the pictures in different file formats. It is possible to open
up as many windows as needed to easily compare properties of one or several ma-
terials. The main window also displays a toolbar with a unit converter and search
functionality.
2.3 Results
Comparison of petrological data between Perple X and other thermodynamical
software has been presented elsewhere (Hoschek, 2004). The emphasis is here on
the validation of the simulated material properties comparing thermodynamic re-
sults to independent sources. The focus of this first presentation of PreMDB is
on bulk material properties not on individual mineral data, which will be done
in a future contribution. Standard rock compositions are assemblages of different
minerals and their properties here are derived from averaging of these constituents.
Moreover, simulations calculate ideal rock phase diagrams at thermodynamic equi-
librium. Therefore, because of the large variability of rock composition and nat-
ural inhomogeneities, it is impossible to perform an exact one to one comparison
22 CHAPTER 2. CONSISTENT MATERIAL PROPERTIES
with a real rock. Another limitation to the validation procedure is that material
property data are conventionally derived at ambient laboratory temperature and
pressure conditions while thermodynamic simulations cover an extended TP range.
Therefore, this study focuses here on a comparison with PREM (Dziewonski and
Anderson, 1981) and ak135 (Kennett et al., 1995; McKenzie and Bickle, 1988)
models for the Earth’s mantle, down to the core-mantle boundary. To assess the
influence of chemistry on seismic properties, three pyrolitic compositions have been
considered presenting variable Al2O3, CaO and FeO contents (Ringwood, 1979;
Saxena, 1996; Stixrude and Lithgow-Bertelloni, 2005a). These compositions are re-
spectively named “Sax”, “Rng” and “Stx”. Calculations have been performed using
the thermodynamic datafile developed by Stixrude and Lithgow-Bertelloni (2005a)
and augmented for the lower mantle as described by Khan et al. (Khan et al., 2006).
Both seismic models PREM and ak135 have been constructed from travel time
data. These 1D models provide a good description of the elastic moduli of the
mantle, but inferring average temperature and composition from them requires a
model of the equation of state (EoS) and accurate knowledge of the thermoelastic
properties of minerals. For the purpose of the comparison to seismic datasets,
the temperature in the mantle is calculated self-consistently following an approach
proposed by Ita and Stixrude (Ita and Stixrude, 1992). An adiabatic interior is
considered to be overlain by a lithosphere defined by the half-space conductive
cooling solution. A potential temperature of 1,600K for the 100 Ma geotherm is
considered, as this temperature lies between two estimates of potential temperature
required to produce oceanic crust of average thickness (Klein and Langmuir, 1987;
McKenzie and Bickle, 1988). The isentrope is computed self-consistently by finding
the PT path along which the total entropy of the assemblage is constant. While
the isentropic assumption follows the principle of self-consistency, the isentropic
character of the mantle is not argued.
The density, P- and S-wave velocity and Poisson’s ratio adopted in the seismic
models PREM and ak135 are compared to the ones calculated for pyrolite mantle
2.3. RESULTS 23
composition with varying Al2O3, CaO and FeO contents figs. 2 to 5. A global
agreement is obtained between seismological and thermochemical models (autocor-
relation functions ACF respectively varying between 0.996-0.998, 0.996-0.998,0.988-
0.995 and 0.748-0.779 for PreMDB/PREM and 0.992-0.996, 0.994-0.997, 0.987-
0.994 and 0.763-0.782 for PreMDB/ak135), except for the Poisson’s ratio. The
discrepancy for this parameter results from the square root relation between the P-
and S-wave velocity ratio (VP/VS) and the Poisson’s ratio (ν):
VPVS
=
√
2(1− ν)
1− 2ν(22)
Therefore, the deviation for this latter parameter is squared, making the Poisson’s
ratio more sensitive to chemistry.
The mixture of 60% of pyrolitic and 40% of chondritic composition proposed by
Matas et al. (2007) has also been tested but did not fit any better the seismic
models.
A robust result of the comparison presented is that the low-velocity layer underneath
the lithosphere emerges out of phase changes and does not necessitate partial melt
as pointed out by Stixrude and Lithgow-Bertelloni (2005a). The results presented
show the abrupt discontinuity around 410-420 km depth, with a better agreement
with ak135 than PREM (best fit obtained with the “Stx” composition). It cor-
responds to the formation of wadsleyite (“Stx” and “Rng” compositions) or the
disappearance of olivine (“Sax” composition). This result is in good agreement
with seismological studies (Bina, 1998a; Deuss et al., 2006; Irifune and Isshiki,
1998). The calculation also predicts two other seismic discontinuities around 536
and 565 km depth only in the case of a composition depleted in Al2O3 and CaO
but enriched in FeO (“Sax”). They correspond to the formation of majorite that
then transforms in akimotoite. This result fits seismic observations (Bina, 1991;
Irifune and Isshiki, 1998) and suggests that heterogeneities in the upper mantle
composition lead to the presence or absence of this seismic discontinuity. Another
24 CHAPTER 2. CONSISTENT MATERIAL PROPERTIES
significant result is that in none of the chemical models considered, the 660 km
discontinuity comes out as a sharp jump in physical properties. There is rather a
series of phase changes in the 650-890 km depth range depending on the compo-
sition considered. This result differs from PREM and ak135 seismological data
and may result from an incomplete description of the solid solutions considered
in the thermodynamic model. The chemical composition and temperature profile
considered, in particular the underlying equilibrium assumption, may also be re-
sponsible for this discrepancy. However, compositions depleted in Al2O3 and CaO
but enriched in FeO (Saxena, 1996) fit the jump more closely than the other com-
positions considered. The calculations predict the formation of perovskite at the
detriment of akimotoite at 649 km depth, followed by the appearance of periclase at
697 km depth and finally the transformation of ringwoodite in magnesiowustite at
758 km depth. This result supports the hypothesis that the 660 km discontinuity
is a transition zone as inferred by receiver function analyses (Deuss et al., 2006;
Vacher et al., 1998) and is thus in agreement with more recent observations than
ak135 and PREM.
A third deviation of the model is observed in the lower mantle increasing towards
the core-mantle boundary and particularly around 2,830-2,840 km depth that cor-
responds to the transformation of perovskite into post-perovskite. The deviation
to the seismic data suggests that the pyrolite composition model may not accu-
rately reflect the chemistry of the lowermost mantle. This observation supports
the hypothesis that the lowermost mantle has a different chemical composition re-
sulting possibly from an accumulation of subducted slabs (Grand, 2002). Other
parameters such as sub-adiabatic conditions in the lower mantle, anisotropy or iron
spin transition could also be responsible for the shift observed in the lower mantle
results.
2.3. RESULTS 25
Figure 2: Density-depth profile of PreMDB compared to the seismic models PREM
and ak135.
Figure 3: VP -depth profile of PreMDB compared to the seismic models PREM
and ak135.
26 CHAPTER 2. CONSISTENT MATERIAL PROPERTIES
Figure 4: VS-depth profile of PreMDB compared to the seismic models PREM
and ak135.
Figure 5: Poisson’s ratio profile of PreMDB compared to the seismic modelsPREM and ak135.
2.4. DISCUSSION 27
An interesting observation is the discrepancy between simulated and seismologi-
cal Poisson’s ratio (autocorrelation functions respectively of 0.748-0.779 and 0.763-
0.783 for PREM and ak135, depending on the pyrolitic composition considered-
fig. 5), especially around the 660 km discontinuity. This result shows that this
parameter is very sensitive to chemistry variation. Therefore it can be a useful
marker of the variation of mantle composition with depth and could be used for
future fine tuning of mantle chemistry.
2.4 Discussion
Recent studies (Stixrude and Lithgow-Bertelloni, 2005a,b) have shown that phys-
ical properties of terrestrial rocks and minerals derived from thermodynamic po-
tentials complement geophysical observations and experimental measurements. In
the benchmark study presented here, the model fits the PREM and ak135 seismic
models and may even record more precise discontinuities due to thermodynami-
cally predicted phase transitions. Discrepancies in the thermodynamic simulations
may result from an incomplete description of the solid solutions considered in the
thermodynamic model, as well as from the chemical composition and temperature
profile considered, in particular the underlying equilibrium assumption. Another
component of the uncertainties is that the PREM and ak135 seismic reference
data derive from experimental measurements.
The properties of terrestrial rocks and minerals previously determined from labora-
tory analyses can now be derived more accurately from thermodynamic modelling
for the entire temperature and pressure range of the Earth’s mantle. An advantage
of this approach is that it permits self-consistent extrapolation beyond the range of
the laboratory. The computational tool used for this purpose is Perple X, a Gibbs
energy minimisation algorithm that computes phase equilibria, maps phase relations
and extracts mineral physical properties of geodynamical interest. Perple X is ro-
bust and computes stable mineral assemblages. It computes the thermochemical,
28 CHAPTER 2. CONSISTENT MATERIAL PROPERTIES
thermal, seismic and other elastic properties from fundamental thermodynamic re-
lations. In particular, Perple X predicts phase transitions that are important for
seismic tomography and geodynamic modelling. In addition to input from Per-
ple X, complementary transport properties derived from laboratory experiments
were added to enhance the information for the user. The resulting reference mate-
rial database is a foundation for performing realistic simulations of rock behaviour
during geological and geodynamic processes.
2.5 Conclusion
PreMDB provides a database for material properties derived from thermodynam-
ics at a range of temperatures and pressures that are not always available from
laboratories, as well as for complementary transport properties. The influence of
the variation of those petrological properties with temperature and pressure was
demonstrated and validated against two accepted seismic models, focusing mainly
on the Earth mantle. These material properties variations however are also criti-
cal at shallower depths, even though the thermodynamic equilibrium assumption
is reaching its limit towards the surface. The graphical interface built around the
database also represents a valuable step to help scientists quickly visualise and
realise the effect of those changes. In the following Chapter 3 those important feed-
backs are incorporated in a numerical simulation package, escriptRT, which not
only accounts for material properties variations but also incorporate many more
chemical feedbacks by modelling directly the fluid-rock chemical interactions.
Chapter 3
Reactive transport
Symbols definition
This chapter uses the nomenclature defined in Table 4.
3.1 Introduction
This chapter presents escriptRT, a new reactive transport simulation code for
fully saturated porous media which is based on a finite element method (FEM)
combined with three other components: (i) a Gibbs minimisation solver for equi-
librium modelling of fluidrock interactions, (ii) an equation of state for pure water
to calculate fluid properties and (iii) the thermodynamically consistent material
database presented in chapter 2 to determine rocks’ material properties. Using de-
coupling of most of the standard governing equations, this code solves sequentially
for temperature, pressure, mass transport and chemical equilibrium. In contrast,
pressure and Darcy flow velocities are solved as a coupled system (Gross et al.,
2009).
Transport of heat and chemical species in porous media is a critical component
in understanding many geological processes such as convection, precipitation and
29
30 CHAPTER 3. REACTIVE TRANSPORT
Table 4: Nomenclature definition
Symbol Definition
ρ density [ kgm3 ]
φ porosity [-]Ω representative volume element [m3]q Darcy flux (in co-moving frame) [m
s]
κ intrinsic permeability [m2]λ thermal conductivity [ W
p fluid pressure [Pa]T temperature [K]Cp specific heat capacity [ J
kg K]
Q radiogenic heat source [ Wm3 ]
M masses [mol]S reactive masses [mol]L reactive mass ratios[-]V discrete volume of the mesh [m3]U molal concentrations [mol
kg]
P molal reaction rates [ molkg s
]
C molar concentrations [molm3 ]
R molar reaction rates [ molm3 s
]
D total mass dispersion [m2
s]
D diffusion coefficient of heat [m2
s]
Superscriptp component in porous spacel component in liquid phases component in solid phaseSubscriptf fluid componenti index of chemical elementz z-coordinate directionVectors and tensorsX vector XX tensor X
3.1. INTRODUCTION 31
dissolution of minerals, which can be applied to geothermal systems or the for-
mation of ore deposits. There already exists some good descriptions and robust
codes which address this class of problems such as those presented by Pruess et al.
(1999), Bartels et al. (2003), Diersch (2005), Geiger et al. (2006) or Ingebritsen
et al. (2006) for example. It is usually no trivial task, however, for researchers to
modify or extend these codes as those developments require a full set of different
skills, from a deep geo-scientific knowledge to a high proficiency in software de-
velopment and numerical analysis. The task can be even more daunting as the
code gets more sophisticated and the changes required need to be compliant with
all underlying software and hardware architecture choices. Object Oriented Pro-
gramming is fortunately increasingly used and facilitates the researcher’s work but
none of the mentioned codes satisfy for example the full separation of concerns
(SoC)1 paradigm (Hursch and Lopes, 1995) in order to let the expert user focus on
geosciences only.
The escript module (Gross et al., 2008) provides a generic environment for mod-
ellers to develop simulations solving partial differential equations (PDE) in python.
Its numerical solver can run in parallel and also offers the ability to work on un-
structured 2D and 3D meshes, which enables one to model complex and geologically
realistic structures for example. This modelling library is separated by design from
the underlying solver and provides with its modelframe approach an ideal modu-
larity which is key to escriptRT’s architecture. The development of escriptRT
was indeed motivated by the need of a powerful but also user-friendly, flexible, and
extensible platform. The purpose of this code is to allow geologists with different
levels of numerical skills to focus on the definition of their problem, with the pos-
sibility to interact with the code at different levels but no obligation to become
experts in applied mathematics and numerical programming.
This chapter is articulated in four parts. Section 3.2 presents the classical system of
1Separation of concerns in computer science is the process of building a clear architecture withseparate building blocks to minimise as much as possible the overlap of common functionalitiesbetween those blocks.
32 CHAPTER 3. REACTIVE TRANSPORT
equations describing all geological processes considered. Section 3.3 introduces all
solvers within the code dealing with those equations. The modular software archi-
tecture uses a conceptual-mathematical-numerical (CMN) pattern and is discussed
in Section 3.4. Modellers benefit from this abstraction process to describe their
problem at a conceptual level, independently from the corresponding mathematical
representation and in turn from the underlying numerical implementation. Finally,
Section 3.5 demonstrates the use of escriptRT by simulating the precipitation
of gold within a complex granite-greenstone sequence conceptualised from Archean
gold deposits.
3.2 Reactive transport formulation
The escriptRT code models the transport of heat and solutes in porous media,
as well as the chemical reactions between the fluid and the host rock. The physical
formulation of the problem is based on the assumption of a porous medium fully
saturated with a single phase fluid, for which Darcy’s law applies (see Bear, 1979).
The fluid and hosting rock are assumed to be always in thermodynamic and chemical
equilibrium, and capillary pressure effects are neglected (Bear, 1979). Following the
finite element methodology a representative volume element Ω of rock is considered
and decomposed as a sum of its solid and porous components
Ω = Ωs + Ωp . (23)
In this section the current status of escriptRT is presented to solve sequentially
a set of decoupled governing equations. The following subsections introduce con-
secutively all process models involved, which employ direct couplings and indirect
feedbacks (see chapter 5 or Poulet et al., 2010). A direct coupling is defined to
be the proper handling of the dynamical updates of all variables involved in the
solved system of equations. An indirect feedback refers to the sequential update of
selected variables in some of the equations, while their time evolution is neglected
3.2. REACTIVE TRANSPORT FORMULATION 33
in others.
3.2.1 Pressure equation
All processes simulated by escriptRT are based on a description of fluid flow
in saturated porous media and it is therefore natural to present firstly the pore
pressure equation. This equation follows from the fluid mass conservation which
can be written asd(φρf )
dt+∇ · (ρfq) = 0 , (24)
The flux q is defined using Darcy’s law under the assumption of quasi static fluid
flow with zero acceleration (γf ≈ 0) in absence of tortuosity
q = − k
µf(∇p− ρfg) . (25)
escriptRT proposes several constitutive models, supporting both compressible
and incompressible fluids. In the case of an incompressible fluid with no porosity
evolution, equations (24) and (25) provide directly the pressure equation
∇ · (ρfk
µf∇p) = ∇ · (ρ2f
k
µfg) (26)
When compressible fluid is considered, the fluid density (ρf ) is dependent on tem-
perature (T ) and pressure (p). A simplified mechanism is also introduced to modify
ρf as a function of salinity X (wrt CNaCl). The following fluid property definitions
are used:
• the thermal expansivity αf , defined as αf = − 1ρf
∂ρf∂T
;
• the compressibility βf , defined as βf =1ρf
∂ρf∂p
;
• the chemical expansivity γf , defined as γf =1ρf
∂ρf∂X
.
34 CHAPTER 3. REACTIVE TRANSPORT
Using the chain rule, the variation of fluid density becomes
dρfdt
= ρf
(
−αfdT
dt+ βf
dp
dt+ γf
dX
dt
)
. (27)
The porosity φ is also considered to be only dependent on pressure p (Geiger et al.,
2006)dφ
dt=∂φ
∂p
dp
dt. (28)
Equation 23 and the definition of porosity φ ≡ Ωp
Ωlead to
dΩt
Ωt
= (1− φ)dΩs
t
Ωst
+ φdΩp
t
Ωpt
. (29)
The compressibility of the rock βr ≡ − 1Ω∂Ω∂p
is then given (Bear, 1972) by
βr = −(1− φ)1
Ωst
dΩst
dp− φ
1
Ωpt
dΩpt
dp= (1− φ)βs + φβf . (30)
From the definition of the porosity and (23), a simple derivation leads to
∂φ
∂p= −φ(1− φ)(βf − βs) . (31)
Combining (27), (28) and (31) provides
d(φρf )
dt= φρf
[
βrdp
dt+ γf
dX
dt− αf
dT
dt
]
. (32)
Using expressions (24), (25) and (32), the pressure equation can finally be written
as
βrφρfdp
dt−∇ · (ρf
k
µf∇p) = −∇ · (ρ2f
k
µfb)
+αfφρfdT
dt− γfφρf
dX
dt. (33)
3.2. REACTIVE TRANSPORT FORMULATION 35
This equation can then be solved directly using escript as described in Sec-
tion 3.4.2.
3.2.2 Temperature transport
The second process model considered is heat transport, where thermal equilibrium
is assumed between the solid and liquid components of the medium. The effects
of mechanical deformation and the related thermal expansion of the solid part are
neglected. In respect to the liquid component, any heating due to viscous dissipation
is also neglected. The temperature is then transported via the standard advection
diffusion equation (Nield and Bejan, 2006)
ρ Cp∂T
∂t+ (ρ Cp)fv.∇T −∇.(D ∇T ) = Q , (34)
where Q is a radiogenic heat source, ρ Cp = φ(ρ Cp)f + (1 − φ)(ρ Cp)s and D =
φDf + (1− φ)Ds is the diffusion coefficient.
3.2.3 Transport of chemical elements
In the mass transport model, the fluid is regarded as being composed of a set of
chemical species in aqueous phase. Those species are transported with the fluid and
considered to be in chemical equilibrium with the solid minerals of the surrounding
rock matrix at the end of every discrete time step. These assumptions allow the
transport problem to be solved for a limited set of reaction invariants instead of the
full set of chemical species. This approach is similar to the concepts of “tenads” in-
troduced by Rubin (1983), as well as the tableaux defined by Morel (1983). We use
as dependent variables the masses of chemical elements constituting the chemical
species, as well as the total electrical charge of the species. The achieved reductions
of the size of the system and the computations in total are considerable. For exam-
ple, a typical simulation would involve tens of chemical elements but hundreds of
36 CHAPTER 3. REACTIVE TRANSPORT
chemical species, and therefore the codes transporting chemical species would solve
much larger systems of PDEs.
Another advantage of using the masses of chemical element as dependent variables is
the improved numerical stability of the system of transport equations. Indeed,when
the full list of chemical species is treated as dependent variables, intensive precip-
itation/dissolution of minerals can cause appearances and disappearances of some
species according to local conditions, resulting in pulses of zero/non-zero masses.
This can cause numerical difficulties to the solver and requires special attention,
see Guimaraes et al. (2007). This issue is generally not present when transporting
chemical elements.
The molal concentrations of these elements are defined at each mesh node of volume
V as
U l ≡ M l
Mf
, (35)
where M l represent the masses of chemical elements within all aqueous species at
the mesh node of interest and Mf is the corresponding mass of the fluid there
Mf = ρfVf . (36)
Similar ratios are defined for the solid phase of chemical elements
U s ≡ M s
Mf
, (37)
where M s are their masses within all mineral species.
Physically the solute is assumed to be pure water, which properties are computed
from available empirical data for the thermal equilibrium of its equation of state
(see Section 3.3.3). As a chemical species, the mas of water is updated according
to the current equilibrium state of the chemical system. In such a way the mass
conservation of the i-th chemical element is represented by the conservation of its
3.2. REACTIVE TRANSPORT FORMULATION 37
mole number, written in terms of molality as
∂(φρfUli )
∂t+ ∇ ·
[
φρfuUli − φD∇(ρfU
li )]
+ φρfPi = 0 , (38)
d(φρfUsi )
dt− φρfPi = 0 , (39)
where Pi is the molal production/consumption rate of this element in liquid phase.
The dispersion tensor D accounts for molecular diffusion, longitudinal and trans-
verse dispersions. The effects of tortuosity can also be considered in this term as
they are neglected in their full treatment. All chemical elements in liquid phase are
also assumed to disperse in the same way.
The conservation of fluid mass (24) allows (38) to be written as
φρf∂U l
i
∂t+ φρfu · ∇U l
i −∇ ·[
φD∇(ρU li )]
+ φρfPi = 0 . (40)
Since by Fick’s law the diffusive flux is driven by volumetric concentrations, mo-
lalities U are converted to molarities C = ρfU , and (39) and (40) are rewritten
as
φρf∂
∂t
(
C li
ρf
)
+ φρfu · ∇(
C li
ρf
)
− ∇ · (φD∇C li) + φRi = 0 , (41)
d(φCsi )
dt− φRi = 0 . (42)
In (41)–(42) C l and Cs are computed using the fluid volume Vf , similarly to
U l and U s in (35) and (37) respectively. The variables R represent the molar
production/consumption rates of chemical elements in liquid phase over the cor-
responding time step; they depend on the extent of the heterogeneous chemical
reactions.
38 CHAPTER 3. REACTIVE TRANSPORT
During transport, the temporal and spatial variations of the fluid density are as-
sumed to be small, keeping therefore ρf approximately constant. Equation (41) can
then be simplified to
φ∂C l
i
∂t+ φu · ∇C l
i −∇ · (φD∇C li) + φRi = 0 . (43)
The effects of chemical reactions are presented in (42)–(43) via the production/con-
sumption rates of chemical elements R , which are computed separately from the
transport by a chemical solver presented in Section 3.3.2. In such a way the
advection-reaction-diffusion (ARD) equation (43) is uncoupled and replaced with
an advection-diffusion equation and a system of mass updates for the chemical
elements in solid and liquid phase. Equations (42)–(43) then become
φ∂C l
i
∂t+ φu · ∇C l
i −∇ · (φD∇C li) = 0 , (44)
and
dC li
dt+ Ri = 0 , (45)
dCsi
dt− Ri = 0 . (46)
The total mass of each chemical element is preserved because the advection-diffusion
equation (44) is conservative and the production/consumption rates in (45)–(46)
have the same amplitude but opposite signs. The influence of the chemical reactions
is indirectly presented by the production/consumption mass rates in equations 45
and 46.
The basic assumptions behind the transition from the ARD equation (43) to the
system (44)–(45) are that all chemical species are always at equilibrium and the
chemical reactions between them are infinitely faster than the advective and diffu-
sive transport processes for the geological scenarios considered. These assumptions
allow the influence of the production/consumption term to be taken into account at
3.2. REACTIVE TRANSPORT FORMULATION 39
the beginning and at the end of every time step when solving (44). This translates
in this workflow by computing the initial mass of the chemical elements in liquid
phase with the chemical solver, and then updating these masses at every time step
by solving firstly (44) and then (45)–(46). Those equations use production/con-
sumption rates computed from the current reactive mass Sl in liquid phase
S =M leq −M l
tr , (47)
where
M ltr are the masses of chemical elements in liquid phase after solving (44);
M leq are the masses of chemical elements in liquid phase after chemical equilibrium
is obtained by the solver.
The aqueous masses in equilibrium Maeq vary within the range
0 ≤Maeq ≤M , (48)
whereM =Ma+M s are the total masses of chemical elements and the comparison
is component wise. In accordance with (47) the range of reactive masses is
−Matr ≤ S ≤M −Ma
tr . (49)
Then the production/consumption rates are
R =S
Vf∆t, (50)
where ∆t denotes the discrete time step.
To estimate the extent of the influence of reactive mass on the decoupling of (43)
40 CHAPTER 3. REACTIVE TRANSPORT
another parameter is introduced
L =S
Matr
=Ma
eq −Matr
Matr
=Caeq − Ca
tr
Catr
, (51)
where L is the reactive mass ratio and the division is component wise. It evaluates
the aqueous mass generated or consumed relative to the aqueous mass present at
the end of the transport. In accordance with (49) the range of reactive mass ratio
is
−1 ≤ L ≤ M
Matr
− 1 , (52)
and again the division is component wise. Since the current approach is using
transport of aqueous chemical elements, the relations (48)–(49) and (52) define
precise lower and upper limits of equilibrated and reactive masses. In contrast,
other approaches using transport of chemical reagents can only define the lower
limits of these parameters.
3.3 Reactive transport solvers
This section presents various solvers used by escriptRT to solve the constitutive
equations presented in Section 3.2. All PDE equations are solved directly using the
escript solver; the fluid-rock chemical reactions, as well as the equations of state
for the fluid and solid material properties are computed using separate external
packages.
3.3.1 The Finley PDE solver
escript can potentially use different FEM solvers and is linked in its current
implementation to the solver library Finley (Davies et al., 2004), which solves a
PDE given in weak form. For the case of a scalar unknown solution u the weak
3.3. REACTIVE TRANSPORT SOLVERS 41
form is
∫
Ω
(
Mv∂u
∂t+∇v · A · ∇u+∇v ·Bu+ vC · ∇u
+vDu) dΩ =
∫
Ω
(∇ ·X + vY ) dΩ , (53)
which needs to be fulfilled for all so-called test functions v. To simplify the presen-
tation, boundary conditions are ignored. It is assumed that this equation is solved
for a sufficiently small time interval such that it can be assumed that the PDE coef-
ficientsM , A, B, C, D, X and Y are constant in time and are independent from the
unknown solution u. Details on how the weak formulation of the PDE is applied
will be presented in Section 3.4.2. This equation is discretised using standard con-
form FEM in 2D and 3D . Finley supports first and second order approximations
of triangle, tetrahedron, quadrilateral, and hexahedron elements (see Zienkiewicz
et al., 2005). In order to support the usage of compute clusters and multi-core
architectures, Finley is parallelised for both shared and distributed memory using
element and node colouring on shared memory via OpenMP2 and domain decom-
position on distributed memory via the Message Passing Interface (MPI) library.
Finley offers therefore flexibility but also performance for our geological scenarios
as it includes optimised solvers for such transport problems (Gross et al., 2009),
including a coupled solver for the pore pressure and Darcy flow velocities.
In the case of a static problem (M = 0) Finley uses iterative techniques such
as preconditioned conjugate gradient method (PCG) (Weiss, 1996) to solve the
discretised version of the weak formulation (53) of the PDE. In the cases discussed
in this paper the coefficients may have large jumps across element boundaries,
typically at the conjunction of different geological structures. This situations leads
to slow convergence or even divergence when solving the discrete problem. To
accelerate the convergence, algebraic multi-grid is used as a preconditioner.
2http://www.openmp.org
42 CHAPTER 3. REACTIVE TRANSPORT
When solving the time dependent transport of chemical species (44) or tempera-
ture (34) the term Mv ∂u∂t
is present in equation (53). In this case Finley applies
the backward Euler or Crank-Nicolson scheme to discretise in time. In cases of an
advection dominated problem, i.e. large values for B, C in comparison to the diffu-
sion term A, unphysical oscillations in the solution can be observed, in particular in
the presence of a steep front in the solution as it is typically seen in reactive trans-
port models. To stabilise the solution artificial diffusion is introduced and limited
via a flux control scheme to restrict the application of artificial diffusion to locations
of large gradients in the solution. Using the framework introduced by Kuzmin et al.
(2004) the scheme is directly applied to the spatially discretised problem without
referring to the continuous problem or the underlying conform FEM mesh.
The incompressible fluid equation (26) can directly be solved within the Finley
framework (53); however, when using FEM the direct reconstruction of the flux
in (25) via gradient calculation can be unreliable and produce numerical artifacts.
This is caused by the fact that the flux constructed in this straight forward way
defines the defect of the FEM approximation of the pressure. In order to rectify the
problem one needs to solve for pressure and flux simultaneously. A least squares
approach is used, combining the incompressibility condition for the flux and the
definition of the flux and solve the optimisation
minq,p
∫
Ω
(
|µfkq −∇p+ ρfb|2 + λ · |∇ · ρfq|2
)
dΩ (54)
This optimisation problem can be reformulated as a weak PDE of the form (53)
for the solution u = (p, q). As in practice values of p and q are different by several
orders of magnitude, iterative schemes have problems to converge. Therefore, the
implementation successively solves for q and p starting from the pressure solution
of the incompressible fluid equation (26). In each step two PDE s need to be
solved, one for pressure and one for flux, using the Finley library. The iteration
is accelerated using the PCG method on the pressure corrections.
3.3. REACTIVE TRANSPORT SOLVERS 43
3.3.2 The GibbsLib solver for chemical equilibrium
Chemical equilibrium of the mass composition is solved using a Gibbs free energy
minimisation technique at prescribed temperature and pressure. It includes the
masses of the chemical elements and water (so called bulk composition), which were
computed during the transport step (see Section 3.2.3 and equation 44). The chem-
ical species existing at equilibrium are found using the GibbsLib solver (Shvarov,
1978) (via GibbsLib.dll) , which is the solver used within the HCh geochemical
modelling package (Shvarov and Bastrakov, 1999). The GibbsLib solver requires
access to an associated thermodynamic database, Unitherm, which is tuned for
hydrothermal ore systems and also forms part of the HCh software bundle. This
database specifies the thermodynamic properties of aqueous species, gases and min-
erals and is valid within the limits 0−5 kbar, 25−1000 C of the modified Helgeson-
Kirkham-Flowers (HKF) model (Helgeson et al., 1981; Shock et al., 1992; Tanger
and Helgeson, 1988). The user specifies a subset of the thermodynamic database
for use in the simulation (as a plane text and binary file) and this is done using
the CSIRO option in the UT2K software3. This option allows users to select
the underlying thermodynamic database of their choice according to the available
empirical data for the involved chemical species.
Several codes have already been developed to simulate reactive transport problems
and can be divided in two categories: Law of Mass Action (LMA) and Gibbs En-
ergy Minimisation (GEM). A presentation of their respective strengths and weak-
nesses can be found in (Shao et al., 2009), including a list of popular codes such
as TOUGHREACT (Xu and Pruess, 2001), PHT3D (Prommer, 2002) or SHE-
MAT(Clauser, 2003) for example. TheGEM solvers require more thermodynamical
data and more computation resources. They also have advantages however, includ-
ing the ability to work from fixed bulk compositions as a starting condition. In
this sense the starting mineral assemblages at equilibrium in the model vary across
3UT2K is available from Evgeniy Bastrakov at Geoscience Australia, www.ga.gov.au. Seealso (Cleverley and Bastrakov, 2005).
44 CHAPTER 3. REACTIVE TRANSPORT
the defined Pressure-Temperature gradient (at fixed total composition), when the
LMA approach requires close-to-equilibrium starting assemblage conditions. This
difference can be critical for the large-scale hydrothermal systems we are targeting.
A comparison of various geochemical modelling approaches and packages which in-
cludes the specific use of HCh in hydrothermal modelling was also demonstrated
in (Cleverley and Oliver, 2005).
There are a number of optional methods for passing chemical state information
to GibbsLib and in escriptRT the chemical composition are passed as moles
of elements at the specified pressure-temperature conditions. Users can choose
to include solute water in the total basis set of elements or pass this separately to
GibbsLib. The resultant chemical equilibrium state is returned as moles of aqueous
species and minerals. Although possible, gas phases or mineral solid solutions are
currently not dealt with explicitly.
3.3.3 The Equation of State (EOS) solver for fluid proper-
ties
escriptRT currently handles a single phase flow, and as a starting point of its fluid
EOS three different models for pure water were selected: the IAPWS-95 (Wagner
and Pruß, 2002), IAPWS-97 (Wagner et al., 2000) and revised HKF model (Helge-
son et al., 1981; Shock et al., 1992; Tanger and Helgeson, 1988). The HKF model
EOS provides thermodynamic consistency with GibbsLib, the chemical solver de-
scribed in Section 3.3.2. The fluid EOS module of escriptRT is critical to the
modelling accuracy obtained. For example fluid density is the main driver for con-
vection (Nield and Bejan, 2006). The water density EOS covers a wide range from
0 to 1000 C and 0 to 5,000 bar. Other water properties are defined on narrower
ranges, but nevertheless are sufficient to define the general conditions of many ge-
ological scenarios with a single liquid phase.
Note that the fluid density ρf (and corresponding chemical expansivity) can also
3.4. SOFTWARE ARCHITECTURE 45
be computed using other options:
• from a formula provided by the user to define ρf from chosen concentrations
ci through the definition of constants Ai, Bi and Ci as
ρf = ρ0
(
∑
i
Aici + exp
(
∑
i
Bici
))
+∑
i
Cici (55)
• as an interpolation table with temperature and pressure dependent values.
Users can then apply any pre-computed equation of state of their choice;
• using an equation of state for saline water proposed by Driesner (2007).
Note that the effect of salinity on fluid properties is an important research subject
and the thermodynamic treatment of salt solutions has not been considered in this
thesis beyond the use of the equations of state mentioned above.
3.3.4 The PreMDB database for rock properties
Solid material properties are calculated in a similar way to fluid properties and
are considered as dependent on temperature and pressure. PreMDB (Siret et al.,
2009) is used for that purpose, which is a thermodynamically consistent material
database based on thermodynamic potential functions to calculate all reversible
material properties, as presented in chapter 2. The code allows the solid density ρs,
specific heat Cp,s and thermal conductivity ks to be interpolated from tabulated
values for temperature and pressure. PreMDB also employs various published
empirical relationships.
3.4 Software architecture
This section presents the modular software architecture of escriptRT (see Fig-
ure 6) and the derived advantages of using such an approach. The underlying idea
46 CHAPTER 3. REACTIVE TRANSPORT
Figure 6: software blocks
is to apply the CMN pattern, which is achieved through the usage of five successive
layers presented in the following subsections.
3.4.1 python environment
python is a powerful and easy-to-use interpreted programming language (van
Rossum and Drake, 2009), ideally suited for rapid application development and
connecting existing components together. It benefits from a large development com-
munity and provides a wide range of additional packages, including very mature and
actively developed numerical extensions like numpy (Oliphant, 2006). The usage of
a high level language like python provides the full flexibility of an Object-Oriented
Programming language. It also enhances the clarity of the code and certainly adds
some user-friendliness when compared to lower level programming languages. It
is important to note that these advantages are not obtained at the detriment of
efficiency. python is used as a binding language to simplify the handling of other
optimised libraries implemented using lower level languages (Sanner, 1999). Within
escriptRT this is illustrated by the four following examples (see Figure 6):
• The escript python interface is used to expose the efficient underlying C++
and C implementation.
3.4. SOFTWARE ARCHITECTURE 47
• Some equations of state of water are implemented in fortran and compiled
into a python library using f2py.
• The initial implementation of the geochemistry module was implemented in
C and linked to python using the boost4 library.
• The current implementation of the geochemistry module uses python na-
tively and makes direct calls to the GibbsLib Windows dll.
3.4.2 escript framework
A cornerstone of escriptRT is the escript5 module (Gross et al., 2007), a high
level simulation environment written in python and providing access to underlying
linear PDE solver Finley as discussed in Section 3.3.1. This solver is available in
escript through the LinearPDE class object which defines a general second-order
linear PDE of the form
−∇ · (A · ∇u+ Bu) + C · ∇u+Du = −∇ ·X + Y (56)
where u represents an unknown scalar function and A, B, C, D, X and Y are
functions of their location in the domain. The boundary conditions are set in a
similar way (see Gross et al., 2007).
escript was designed to solve general, coupled, time-dependent, non-linear systems
of PDEs. Users can achieve that goal by appropriately setting the coefficients of
the LinearPDE. For example the compressible fluid equation (33) can be rewritten
doStep, terminateIteration, doStepPostprocessing, finalize and do-
Finalization (see Gross et al., 2008). This breakdown of such a generic workflow
provides a very convenient framework to implement all the couplings and feed-
backs most developers are interested in. Different Model derived classes can access
each other’s updated parameters through the use of Link objects (see Gross et al.,
2008) which provides the simplest mechanism to implement indirect feedbacks as
all variables from different classes get updated sequentially. Both the compressible
and incompressible pore pressure models were implemented in two different Model
classes. The power of the modelframe module is illustrated by the fact that in
the simulation script those two implementations can be seamlessly interchanged
without modifications to other parts of the code.
The modelframe module is a major component to ensure escriptRT’s extensibility
and re-usability (see Gross et al., 2008). Each Model derived class can be easily
extended with new features at the appropriate point in the workflow thanks to all
available methods. The code produced is also easily reusable as each Model derived
class can be used as is in a different simulation. Involving different process models
50 CHAPTER 3. REACTIVE TRANSPORT
then only becomes a matter of plugging together such classes to build simulations.
Coupling the existing code with other process models also becomes quite easy as
the only step required is to implement (possibly wrap) the new process models in
new Model derived classes. The modelframe module serves as the basis of the next
two higher-level modules escriptBaseRT and escriptFluidHeatChem.
3.4.4 EscriptBaseRT layer
escriptBaseRT module is a collection of utilities and basic classes which deals with
the numerical aspects of the solvers involved. It contains base Model classes to
map the top level classes to the corresponding numerical schemes. The advection-
diffusion equations dealing with the transport of heat and solutes can for example
use either the traditional backward Euler or Crank-Nicholson schemes. The calcula-
tion of material and fluid properties can be accelerated via the interpolateTable
functionality of escript. In short, escriptBaseRT encapsulates all external pack-
ages like GibbsLib to provide the required level of abstraction for escriptRT by
allowing the top level module to focus as much as possible on the geophysical and
geochemical concepts.
3.4.5 PmdPyGC
PmdPyGC is a generic package supporting a wide range of functionalities for geo-
chemical simulation. It plays four essential roles: (i) parsing chemical system def-
inition files with their associated thermodynamic database (see Section 3.3.2), (ii)
creating and updating chemical systems, (iii) updating their current state and (iv)
applying a prescribed external chemical solver to set the system into equilibrium.
The package is written in C++ and contains hierarchically ordered classes pertain-
ing to the building components of the chemical system, such as elements, charges
and reactants, and to quantitative aspects of the chemical state, such as masses,
3.4. SOFTWARE ARCHITECTURE 51
volumes, concentrations, phases and reactions, corresponding to the current temper-
ature and pressure. Various interfaces are supported to provide access to wide range
of applicable fluid EOS modules and chemical solvers. The current chemical solver
of usage is GibbsLib (see Section 3.3.2); via dynamical interaction of PmdPyGC
with its Windows dll, updates are provided for the chemical state at equilibrium,
including produced aqueous and mineral species and related phase transitions. The
access of escriptRT to PmdPyGC is done using the boost library.
The fluid EOS modules presented in Section 3.3.3 are implemented in escriptRT
in three different manners, as illustrated on Figure 6. For example they involve
fortran to C translation using f2c (Feldman et al., 1993), so they can be used
directly by escript. This is much more efficient for example than another pos-
sibility which would consist of translating fortran packages directly to python
using f2py (Peterson, 2009). As mentioned previously, python is mainly used as
a binding language to connect components while the numerics are more efficiently
implemented using lower-level languages (like C in escript). The final option
for access to the EOS of water is to save tabulated values in an ASCII file for a
given regular grid of temperatures and pressures. These file can then be loaded in
escript and used to interpolate the corresponding fluid properties for each mesh
point very efficiently.
3.4.6 EscriptFluidHeatChem structure
The escriptFluidHeatChem module is the top level interface to the whole pack-
age and represents the conceptual level of the CMN pattern. Its purpose is to
provide easy access to numerical modelling to geologists without much numerical
experience but also to expert users willing to develop new constitutive models.
escriptFluidHeatChem provides a clear interface for geologists to describe their geo-
physical/geochemical problem at a conceptual level and run numerical simulations
with or without any numerical knowledge of the underlying implementations.
52 CHAPTER 3. REACTIVE TRANSPORT
A simulation can be defined through this interface as a sequence of time slices, where
each time slice defines a period of time for a given set of parameters, including each
process model, constitutive model, stopping conditions, boundary conditions or
numerical parameters. Each time slice is defined using a large python dictionary
(mapping object) on multiple levels and allows users to set each variable using
values or predefined keywords to chose between all available methods. An extensive
check is run before launching simulations to ensure that all initial parameters have
been set up properly for all time slices. This avoids therefore any bad surprise
after several days of simulation. This top level module has also allowed an easy
integration of escriptRT within a graphical user interface, which allows geologists
and geochemists to use it without any specific programming knowledge.
3.5 escriptRT and hydrothermal gold systems
The escriptRT code was written to allow the simulation of complex coupled heat-
mass transport and chemical reaction problems associated with understanding pro-
cesses important for the formation of ore deposits. This section presents an example
of the application of the escriptRT code with model of transport and precipita-
tion of gold associated with oxidised granites in a granite-greenstone terrane. This
is a conceptual model developed from observations in some Archean hydrothermal
gold deposits in the Eastern Goldfields of the Yilgarn craton, Western Australia.
3.5.1 Physical Conditions
The model is constructed to represent one half of a symmetrical 2D vertical cross-
section of a mafic basin (top) underlain by gently folded mafic-ultramafic sequence
with granitic basement (bottom). Hot oxidised, gold-bearing granite is added into
the granitic basement (see Figure 8). A fault passes through all layers with a
permeability increase of one order of magnitude. Porosity in this fault is calculated
3.5. ESCRIPTRT AND HYDROTHERMAL GOLD SYSTEMS 53
accordingly with the classical Blake-Kozeny equation presented in McCune et al.
(1979). The model is 10 km deep and 12.5 km wide with the top buried to 3 km.
Temperature and pressure are fixed at the top to 100 C and 70 MPa while at
the bottom, pressure is free to vary dynamically throughout the model and fluid is
allowed to pass across this boundary. A fixed heat flux is applied to the bottom
boundary (47.95 mW/m2). This is derived from crustal thermal modelling for
a 35 km thickness Archean crust using material properties from PreMDB. The
model starts with a basal temperature of ∼ 350–370 C because of the variation
in thermal properties of the rocks related to temperature-pressure feedbacks. The
granite is initialised as a 650 C body superimposed on the steady state thermal
gradient, and is then allowed to cool due to advective-conductive heat transport to
the surrounding rocks.
3.5.2 Chemistry
The initial chemistry of the units in the model is defined as a bulk composition by
using volume fractions of the mineral phases and the chemistry of the fluid phase (if
different from pure water). As a starting condition HCh (Shvarov and Bastrakov,
1999) is used to help define the initial chemistry using the chemical composition of
standard Archean rocks. The initial fluid phase is defined as water — 0.5 molar
NaCl in most units except the granite which used H2O — 2MNaCl — 0.5MKCl
— 1MSO2 — 1.14e−4MAu. The Au concentration is set below saturation. The
major mineralogy of the various units are described in Table 5. The first step of
the simulation solves for chemical equilibrium between the model fluid and rock at
each node before any mass transport takes place. The simulation was run until all
transient effects on gold mineralisation were observed and the results are presented
for a total simulation time of 200 ky.
54 CHAPTER 3. REACTIVE TRANSPORT
Table 5: Initial major mineralogy of the various units in the model, from top tobottom. Faults are 1 order of magnitude more permeable than the surrounding rocks.
After 200 ky the model has developed a series of convective cells within the upper
basalt zone (Figures 8a and 8b) as well as a broad thermal anomaly above the
granite (which has cooled to ∼ 450 C ). If we look in detail at the mineralogical
development in the dome above the granite cupola we see the development of a C
and S anomaly shown by calcite, pyrite and pyrrhotite distribution (Figures 9a and
9b). The pyrrhotite, which was stable in the ultramafic unit at the start, forms as
a halo to the pyrite probably related to the fluids becoming more reduced as they
move away from the core of the upflow into the reducing rocks. Phlogopite (Mg-rich
biotite) is also developed in this zone related to the K-rich fluids coming from the
granite (Figure 9b).
Gold is developed within two distinct zones (Figure 10): the upper basalt top con-
tact in the upflow zone, and as a lithology perpendicular reaction front within the
ultramafics but away from the core upflow zone. The upper gold enrichment ap-
pears to be driven by a chemical contrast and the strong lateral flow within the
3.6. CONCLUSION 55
upper basalt unit as a function of enhanced permeability. The reaction front gold
zone is located at a thermal and redox gradient within the ultramafic, although
other mineralogical changes in this zone are subtle at best. As with all reactive
transport modelling, these results can be non intuitive as there is little gold pre-
served within the strong alteration above the granite. This model does not only
present interesting results about the nature of the deposition but it also shows that
for dynamic thermal-flows it is difficult to preserve concentrations of gold within
the core of the system. In addition, there are many subtle but important feedbacks
which could affect the gold trapping process, including porosity evolution.
(a) (b)
Figure 8: Results after 200 ky showing the strong control from the passage of thehot plume above the granite and the strong convection cell development in the upperbasalt unit: (a) Magnetite (total moles); and (b) Temperature distribution. Domainsare from top to bottom: 1) mafic basin, 2) upper basalt, 3) ultramafic, 4) lower basalt,5) granitic basement and 6) granite and 7) fault with enhanced permeability crossingall layers. Arrows represent the Darcy flux with velocities of about 1m/y in the leftpart of the upper basalt layer.
3.6 Conclusion
The motivation behind the development of escriptRT was to provide a powerful
but also user-friendly, flexible, and extensible platform. This allows numerical mod-
elling geologists to focus on the definition of their geophysical/geochemical problem
56 CHAPTER 3. REACTIVE TRANSPORT
(a) (b)
Figure 9: Distribution of minerals within the upflow zone inside the ultramafic after200 ky for a subset of the whole region: (a) Pyrrhotite and pyrite (contour, max 2.5m); and (b) Calcite and phlogopite (contour, max 0.6m).
at hand at the constitutive level without necessarily having to become experts at the
same time in applied mathematics and programming. The software architecture of
escriptRT presents some advantages in terms of flexibility (capacity to be adapted
or modified), extensibility (simplicity to add new features) and re-usability (possi-
bility to easily use some components or to couple with other codes), all derived from
the usage of carefully chosen components. A modular object-oriented architecture
was implemented using escript and its modelframe module. python was mainly
used as a high-level glue language to connect components while the numerics were
efficiently implemented using lower-level languages. This chapter also presented an
example that illustrates the complex coupling between thermal response and mass
transfer, as well as the complex mineral assemblages that develop and localise with
the stratigraphy. This simulation emphasises how critical process coupling is when
attempting to simulate large-scale hydrothermal systems.
3.6. CONCLUSION 57
Figure 10: Distribution of gold and fluid flow at the end of the model in a subsetof the whole region. There are two key areas of gold mineralisation : 1) reactionfront gold in the ultramafic and 2) boundary related gold at the upper mafic contact.Temperature contours (C) are added as dashed lines.
58 CHAPTER 3. REACTIVE TRANSPORT
Chapter 4
Continuum damage mechanics
4.1 Introduction
The strength of the lithosphere can be affected by various weakening mechanisms
that turn rigid tectonic plates into deforming plates whenever a critical energy
threshold is overcome for the onset of lithospheric failure. In previous contribu-
tions (Regenauer-Lieb et al., 2001, 2006) it was shown that weakening feedbacks
such as viscous dissipation and shear heating can have considerable influences on
the energy thresholds and thereby play key roles on stress and strain localisation
(Hobbs et al., 1990; Sengupta, 2010). Classical mechanical approaches which ignore
such energy balances are struggling to explain the level of forces required to drive
tectonics on our planet. They require forces that are at least four times as large
as those deemed available from slab pull or rigid push estimates (Regenauer-Lieb
et al., 2008). One possible solution to this problem is to consider the time-dependent
strength reduction caused by the feedback of deformation, shear-heating and ex-
ponential temperature dependence of flow laws. This feedback is very efficient for
materials with high activation energy such as olivine and it can lead to a substantial
reduction in lithospheric strength (Braeck and Podladchikov, 2007). However, the
59
60 CHAPTER 4. CONTINUUM DAMAGE MECHANICS
predicted level of forces is still an upper limit (Regenauer-Lieb et al., 2010). Addi-
tional weakening mechanisms through damage mechanics, temperature or fluid flow
must be taken into account.
Damage mechanics is commonly described by (i) intensive variables which ac-
count for the sensitivity of macroscopic elastic properties to distributed local weak-
nesses (ii) damage evolution under external loading. There is a general agreement
(Lemaitre, 1985; Chaboche, 1987; Cocks and Ashby, 1982; Lyakhovsky et al., 1997)
on the first point meaning that damage can be seen as distribution of cracks and
voids at the micro-scale. Consequently, the intensive variables of damage are in-
terpreted geometrically as the ratio of damaged over intact sections (Cocks and
Ashby, 1980) and used to describe the effective stresses in damageable materials
(Lemaitre and Dufailly, 1987). However, damage evolution is still a challenging
subject especially in case of geological materials. The available models in this con-
text dealt with several aspects which involve the thermodynamics of fluid materials
(Bercovici, 1998; Bercovici and Ricard, 2003; Landuyt and Bercovici, 2009; Ricard
et al., 2001), the phenomenological description of weakening (Regenauer-Lieb, 1998;
Hieronymous, 2004), the brittle damage of crustal materials (Lyakhovsky et al.,
1997; Hamiel et al., 2004a,b; Lyakhovsky et al., 2005; Nanjo et al., 2005). These
models cover a wide range of applications and can be used to study seismic events
or wave propagation in geological structures under small perturbations, the brittle
regime of crust deformation, the materials weakening due to thermal and chemi-
cal feedbacks etc. However, these models do not allow elasto-visco-plastic analyses
where reversible and permanent deformations both play crucial roles. Observations
of shear zones deformation at relatively high temperature show that visco-plastic
processes have considerable effects on void nucleation and are dominating the long
term response of geological materials (Fusseis et al., 2009).
This chapter proposes a new theory of continuum damage suitable for plate tec-
tonics which is characterised by large time and length scales. Classifying olivine
as a generalised standard material (Halphen and Nguyen, 1975a), a mathematical
4.2. THERMO-MECHANICAL BACKGROUND 61
model is formulated which describes constitutive behaviour of a damageable litho-
sphere. Although damage is interpreted as an intensive variable which accounts
for the sensitivity of the elastic properties to local weaknesses in accordance with
Lemaitre (1985) and Lyakhovsky et al. (1997), its evolution is described through a
new dissipation potential which is active in the inelastic regime.
The developed constitutive model is integrated using the user material subroutine
UMAT of ABAQUS/Standard (2008). The numerical approach developed in this
chapter is based on relatively new techniques of thermal and rate dependency sug-
gested by Ponthot (1995); Voyiadjis and Abed (2006) and Karrech et al. (2010).
The technique of return mapping developed by Simo and Taylor (1985); Karrech
et al. (2012) is used to guarantee a robust integration of the constitutive model.
The effectiveness of this prediction-correction method is also confirmed by Paulino
and Liu (2001); Kang (2004); Kumar and Nukala (2006). Based on the normal-
ity condition, a consistent tangent modulus is derived by taking into account the
CDM description with a combination of creep mechanisms, water content, rate, and
temperature dependency.
4.2 Thermo-mechanical background
Consider a representative volume element of material which is statistically homo-
geneous with its surrounding such that its averaged properties do not change if its
boundaries are expanded. Employing the notion of sequential equilibrium states,
an additive decomposition of strain increments is used to develop the constitutive
model:
dǫ = dǫe + dǫin (58)
The superscripts e and in denote respectively the elastic and inelastic strains. This
decomposition is valid in case of small deformation of standard materials (Karrech
and Seibi, 2010). The Helmholtz free energy ψ can be used to derive constitutive
relationships. It can be defined as a functional of observable variables (such as
62 CHAPTER 4. CONTINUUM DAMAGE MECHANICS
strain, ǫ, and Temperature, T) and the internal variables (such as damage D,
elastic strain, inelastic strain and other dissipation quantities, which are ignored in
this study). Hence, the Helmholtz energy can be expressed in terms of its variables
as follows: ψ(ǫe, T,D). It is worthwhile noting that other forms such as Gibb’s free
energy, enthalpy or internal energy can be deduced in terms of ψ using Legendre’s
transform. In particular, the Legendre transform of ψ with respect to temperature
results in the internal energy:
u(ǫe, s,D) = ψ(ǫe, T,D) + sT (59)
where s denotes the specific entropy. It is the dual of temperature in the sense of
the Legendre transform. Using the time derivative of equation (59) and equation
(176) in appendix A.1, one deduces that:
ρψ + ρT s+ ρsT = σ : ǫ+ r − div(q) (60)
where ρ is the material density, q is the heat flux vector, and σ is Cauchy’s stress
tensor. In accordance with Fourier’s law, the heat flux can be expressed as q =
−kgrad(T ), where k is the thermal conductivity. Equation (60) summarises the
first principle of thermodynamics in its local form. It shows that the local internal
energy is equal to the internal work augmented with the local heat production and
transfer. Combining equation (60) with equation (179) in appendix A.2 results in
the following inequality:
σ : ǫ− ρψ − ρsT −q
T.grad(T ) ≥ 0 (61)
The above equation contains the fundamental second principle of thermodynamics
stating that the rate of entropy production is higher or equal to the heat supply
over temperature. Equation (61) represents the local rate of irreversible entropy
production as discussed in details by Regenauer-Lieb et al. (2010). Using the addi-
tive decomposition of strain (58) as well as the derivative of Helmholtz free energy
4.2. THERMO-MECHANICAL BACKGROUND 63
with respect to time1: ψ = ∂ǫeψ : ǫ+ ∂TψT + ∂DψD, equation (61) reduces to:
σ : ǫin +
(
σ − ρ∂ψ
∂ǫe
)
: ǫe − ρ
(
s+∂ψ
∂T
)
T − ρ∂ψe
∂DD −
q
T.grad(T ) ≥ 0 (62)
This expression, which invokes the positivity of the irreversible entropy production
rate, is often called Clausius-Duhem inequality. In the above derivations, the stress
is considered to be the same across the elastic and viscoplastic regimes. Applying
the postulate of Coleman and Noll (1963) by considering that equation (62) must
hold for every admissible process, leads to the following relationships:
σ = ρ ∂ψ∂ǫe
(a) and s = − ∂ψ∂T
(b) (63)
Combining equations (62) and (63) results in the following local inequality, which
states that the rate of energy dissipation D is always positive:
D = σ : ˙ǫin − ρ∂ψe
∂DD −
q
T.grad(T ) ≥ 0 (64)
The thermodynamic force of damage, also known as triaxiality, can be defined as:
Y = −ρ∂Dψ, in analogy with the definitions of stress and entropy. The dissi-
pation D is a scalar product of duals involving the thermodynamic forces, σ, Y ,
T = −grad(T )/T , with their respective fluxes, ˙ǫin, D, q. With these notations,
Clausius-Duhem inequality (64) can be rewritten as:
D = σ : ˙ǫin + Y D + q.T ≥ 0 (65)
This expression includes intrinsic and thermal terms: D = Di + Dt. Since the two
dissipations can take place independently, the postulate of Coleman and Noll (1963)
requires that Di and DT are both positive. If the elastic behaviour is linear and
1∂xψ is the partial derivative of ψ with respect to the variable x
64 CHAPTER 4. CONTINUUM DAMAGE MECHANICS
isotropic then equation (63-a) reduces to:
σ = C : ǫe (66)
where C denotes the fourth order elasticity tensor of the damaged material. The
Cauchy stress σ is an homogenised quantity representing the average force on a
given surface. Therefore, the “effective” stress which applies on the actual un-
damaged skeleton can be tracked through the following definition suggested by
Kachanov (1986) and taken into account in the formulations of Lemaitre (1985)
and Lyakhovsky et al. (1997, 2005):
σ =σ
(1−D)(67)
The variable D can be seen as the ratio of damaged over undamaged sections.
Consequently, it can also be understood as an energy portioning variable. When
damage is isotropic, equations (66) and (67) can be used to deduce a relationship
between Cauchy stress, the damage variable D and the elastic properties of the
intact material:
σ = C : ǫe = (1−D)C0 : ǫe (68)
where Cijkl =(
K − 23G)
δijδkl+G (δikδjl + δilδjk) is the fourth order elasticity tensor
of the damaged material, C0ijkl is its equivalent in the undamaged configuration2, K
is the bulk modulus, G is a shear modulus, δij denote the Kronecker symbol, and
the indices (i, j, l, k) represent the directions of a Cartesian space. The selection
of the above form of elastic energy also implies that the thermodynamic force, Y ,
associated to the damage parameter, D, is given by:
Y = −ρ ∂ψ∂D
=1
2ǫe : C0 : ǫe (69)
2The term “undamaged” refers to the material at its original state when no degradation tookplace.
4.3. FROM DISSIPATION TO FLOW RULES 65
From the above definition (69) and following the derivation of Lemaıtre and Chaboche
(2001), it can be shown that Y is given by:
Y =σ2eq
2(1−D)
[
1
3G+
1
K
(
σHσeq
)2]
(70)
where σH = 13tr(σ) is the hydrostatic pressure, and σeq =
√
32s : s is the equivalent
stress. The deviatoric stress is expressed by s = σ − σH1 and 1 is the third order
unity matrix.
4.3 From dissipation to flow rules
In order to derive the flow rules, the existence of a regular potential of intrinsic
dissipation φ(σ, Y ) is assumed. As it is defined within a constant, the dissipation
potential could contain other variables which do not play any role in deriving the
flow rules. This section explains at first the different processes that contribute to
this potential and then uses them to deduce the flow rules.
4.3.1 Visco-plasticity
Plastic behaviour requires the definition of a yield function f(σ) (elasticity enve-
lope) and plastic potential g(σ). In case of associated materials, such as olivine,
these two functions coincide. This study uses the following expression:
f(σ) = g(σ) = σeq − (1−D)σ0 = σeq − σ0 (71)
where σ0 is the yield limit of the intact material which can be taken as a constant
if there is no plastic hardening or dependent on plastic deformation otherwise. σeq
in this case depends only on the second invariant for simplification, however, the
first and third invariants can be included depending on experimental evidence.
66 CHAPTER 4. CONTINUUM DAMAGE MECHANICS
As the lithospheric materials generally deform at high temperatures, the viscous
effects are considered in the inelastic regime. In its basic form, viscoplasticity
is commonly introduced through an overstress which was introduced by Perzyna
(1966). Unlike rate-independent plasticity, the equivalent stress, σeq, is no longer
constrained to remain less than the limit σ0. The exceeding quantity is known as
overstress, it is defined as follows:
O =< σeq − σ0 > (72)
where 2 < x >= x + ‖x‖. Therefore the magnitude of the permanent deformation
is proportional to the overstress can be written as:
ǫv =Oη
(73)
where ǫv =√
˙ǫin : ˙ǫin is an equivalent viscoplastic deformation and η is a fluidity
parameter. Notice that the original viscoplasticity introduced by Perzyna (1966) is
linear and athermal. It was later modified by Perzyna1966 himself to include non
linear effects of the form: ǫv = φ(O) =⟨
O
η
⟩m
. By inverting the latter relation-
ship, Ponthot (1995) noticed that φ[−1](ǫv) = ηǫv1/m
= O = σeq − σ0. Hence, he
introduced the new “continuous” condition which can be written as:
f(σ, ǫv) = σeq − σ0 − ηǫv1/m
= f(σ)− φ[−1](ǫv) = 0 (74)
This study considers combined dislocation and diffusion creep mechanisms. There
are well know temperature and water content dependent measurements describ-
ing them for olivine (Goetze, 1978; Mei and Kohlstedt, 2000a,b). In accordance
with Regenauer-Lieb et al. (2001) and Regenauer-Lieb (2006), their combination is
4.3. FROM DISSIPATION TO FLOW RULES 67
expressed as follows:
ǫv = φ(O) = AdO
g3d
exp(
−Qd+σH(Vd−rk∆VOH)
RT
)
+ ApOnexp(
−Qp+σH(Vp−rk∆VOH)
RT
)
(75)
where Aq (q=d,p) are the pre-factors of the diffusion and dislocation mechanisms
respectively, Qq are the activation energies, gd denotes the grain size, σH is the
hydrostatic pressure, ∆VOH is weakening term for increasing pressure (change in
molar volume associated with the incorporation of hydroxyl ions into forsterite), r
and k are fitting parameters of order 1, Vq is an activation volume, and the subscript
q refers to the corresponding creep mechanism. From equation (75), it can be seen
that the different creep mechanisms act in series in accordance with the model of
Regenauer-Lieb et al. (2001). By proceeding the same way as Ponthot (1995), one
obtains
f(σ, ǫv) = f(σ)− φ[−1](ǫv) = 0 (76)
where φ[−1] is the invert of equation (75). For this study the function φ defined
by (75) is more complex than the power law used by Ponthot (1995), therefore the
inversion is performed numerically using the dichotomy method. From equations
(71) and (76) one can obtain a potential of inelastic deformation:
φi(σ, ǫv) = σeq − σ0 − φ[−1](ǫv) (77)
4.3.2 Damage potential
In light of the formulations of Cocks and Ashby (1980, 1982); Chang et al. (1987),
the damage evolution is described by the following potential:
φDe = Y(
1(1−D)n+1 − 1
)
(78)
where n denotes the dislocation law exponent. Unlike the expressions postulated
Table 7: Simulation Parameters: Dissipation constants (*the subscripts q=p,d referto power law and diffusion respectively. The same order is valid for the following tworows. ** µm3 is included only if q=d)
4.5. NUMERICAL APPLICATION 75
Figure 11: Finite element modelling of a lithospheric layer containing a notch (seezoom)
Effect of temperature
In case of undamaged viscoplastic behaviour, the numerical results are similar to
those obtained by Regenauer-Lieb and Yuen (1998). Two shear bands take place
at the notch and propagate progressively in a 45o inclination with respect to the
horizontal, through the lithospheric cross section. Figure 12-a shows the equivalent
inelastic strain that takes place before t = 1 Myrs. It can be seen that regular peak
lines of high deformation with a maximum magnitude of ǫineq = 0.25 are concentrated
in the neighbourhood of the notch. The magnitude decreases sharply by about 75%,
4 km away from the centre of the notch and becomes insignificant in magnitude
outside this region. Figure 12-a also shows the initiation of propagating inelastic
strain along the shear bands. At t = 3Myrs, the propagation continues and a
high cumulative inelastic deformation reaches most of the shear bands as can be
shown in Figure 12-c. However, the peak of maximum equivalent deformation,
ǫineq = 0.5 is still in the neighbourhood of the notch centre. A sharp decrease of
about 75% in magnitude can still be noticed around 4 km away from the centre of
the notch. On the top of the propagating inelastic strain, Figure 12-d shows the
necking behaviour on both sides of the lithospheric layer. It can be seen that in
general the flow is towards the region of maximum deformation except at the notch
76 CHAPTER 4. CONTINUUM DAMAGE MECHANICS
itself where a high thermal expansion takes place and allows the material diffusion
to be accelerated. The model also shows a temperature increase at the shear bands
with contours of the same patterns as those obtained in case of inelastic strain.
Figure 14-a shows that at t = 1 Myrs the temperature increases up to 980o K with
a peak in the neighbourhood of the notch. This temperature variation represents
the conversion of dissipated energy into heat and the acceleration of dissipation due
to the increase of temperature. Figure 14-c shows the temperature distribution at
t = 3 Myrs. Note, as expected, that the temperature reaches a much higher peak
magnitude but with more uniform gradient along the shear bands.
(a) (b)
(c) (d)
Figure 12: Equivalent inelastic deformation of the rate-dependent undamaged (aand c) and damaged (b and d) lithospheric layer after 1 Myrs (a and b) and 3 Myrs(c and d)
Damage weakening
This section shows that continuum damage accelerates material weakening and
affects the magnitude of inelastic deformations in shear zones. Figures 13 show
the contours of damage distribution with respect to time. They also show that
damage initiates in the neighbourhood of the notch and diffuses along the shear
zones. It can be seen that damage varies from 0 to 0.85 as maximum critical value.
4.5. NUMERICAL APPLICATION 77
Theoretically, damage must always be smaller that unity (total degradation); in this
case, a critical value of the order 0.85 is considered as shear zones in geology can still
have a certain strength after material degradation. Comparison between Figures
12-a and 12-b shows that damage increases the equivalent inelastic deformation
by around 20% at the beginning of the loading process. At later stage and as
damage increases, this effect accelerates. Figures 12-c and 12-d shows that inelastic
deformation is three times higher when the geological structure is damaged. This
increase of inelastic strain is explained by the effect of damage on the strength of
materials which results in increase of the total deformation. In addition, the increase
of the damage parameter D results in a shrinkage of the yield surface, which in turn
increases the inelastic deformation. Figures 14-b and 14-d show the distributions
of temperature at t = 1 Myrs and t = 3 Myrs, respectively. The contours show
that the pattern is similar to the one obtained in the case of undamaged structure.
However, the overall differential magnitude in this case is much lower, as expected.
It varies from about 7 degrees down to 4 degrees when the structure is damageable.
This decrease of temperature is due to the decrease of heat generation related to
the higher thermal feedback of the inelastic dissipation.
(a) (b)
Figure 13: Damage distribution after (a) 1 Myrs (b) 3 Myrs
The former figures also show that the necking process is accelerated by damage since
the lithospheric shape undergoes higher material flow when it is subjected to severe
degradation than when it is continuously yielding. This result can be interpreted
qualitatively by comparing the undamaged and damaged deformed shapes at the
bottom of the plate in all the above mentioned figures. It is worthwhile noticing that
78 CHAPTER 4. CONTINUUM DAMAGE MECHANICS
(a) (b)
(c) (d)
Figure 14: Shear heating effect due to inelastic deformation of the undamaged (aand c) and damaged (b and d) lithospheric layer after 1 Myrs (a and b) and 3 Myrs(c and d)
all the deformed shapes are presented at a scale factor of 1 and are not enhanced.
It is also important to mention that the obtained shear bands width is independent
of the element size as shown in figure 15.
4.5.2 Energy partitioning
Figure 16 shows that the variation of elastic energy with respect to deformation is
quadratic before yielding, decreases continuously if damage takes place and main-
tains a plateau-like shape if no damage occurs. The decrease of the elastic energy is
attributed to the increase of the damage parameter which results in partial conver-
sion of the elastic energy into damage, whereas the plateau-like behaviour is due to
the non-hardening effect of the material. The figure also shows that the viscoplastic
energy increases linearly if the structure is undamaged. Again, this response can
be explained by the non-hardening of the material and the constancy of the applied
velocity. Figure 16 also shows that the viscoplastic energy is higher if the structure
is undamaged; the result is in accordance with the heat generation shown in figure
4.5. NUMERICAL APPLICATION 79
Figure 15: Independence of the shear band width on the size of elements
14. Therefore, it can be concluded that if damage is not taken into account the
necessary forces to deform lithospheric structures are overestimated.
(a) (b)
Figure 16: Elastic and viscoplastic energies for (a) undamaged and (b) damagedstructures.
4.5.3 Effect of loading rate on the structural integrity
Depending on the geological events lithospheric layers can undergo different levels
of loading rates. In this paragraph the above described model was slightly modified
as follows: (i) since the problem is symmetric, only the right hand side of it is
considered for simulation. (ii) The rectangular notch is replaced by a circular notch
80 CHAPTER 4. CONTINUUM DAMAGE MECHANICS
of radius 25 km to obtain more pronounced weakening. The rest of boundary and
initial conditions are all similar. Figure 17 shows the response of the new structure
under different loading velocities which are multiples of 0.63 mm/yr , in both cases
when damage is and is not taken into account. In the case of non-damaged material,
it can be seen that the responses are almost indiscernible for small loading rates;
however, the viscous effects are amplified gradually with larger loading rates. This
results in the main features of a rate-dependent behaviour which is characterised by
(1) the expansion of the yield envelope and (2) the increase of the viscous hardening,
as can be shown in Figure 17. These phenomena can be encountered in different
materials and represent the main feedback of viscoplastic behaviour. The first
phenomenon can be explained by the expansion of the yield function (increase of
radius due to the increase of φ[−1](ǫineq)) and the second is related to the increase of
hardening (derivative of φ[−1](ǫineq)).
Figure 17: Materials behaviour depending on viscoplasticity and damage: (a) re-sponse of the un-damaged structure (b) response of the damaged structure, underdifferent loading velocities (mm/yr).
A comparison between fig. 17-a and fig. 17-b also highlights the importance of
damage mechanics in the modelisation of brittle and ductile behaviours. The first
subfigure is characteristic of ductile behaviours for all rates of loading, whereas the
second one displays brittle responses for small loading rates and ductile responses
for higher loading rates. Loading rate provides a switching mechanism between
brittle and ductile mechanisms, allowing a smooth transition from one to the other.
4.6. CONCLUSION 81
4.6 Conclusion
A numerical model was suggested to study the geodynamics of a notched cross-
sectional lithospheric layer using finite element. It takes into account thermo-
mechanical coupling, continuum damage mechanics and viscoplastic rheology with
multiple creep mechanisms. The formulation was implemented using relatively new
integration techniques. The approach took into account persistent viscoplastic
yielding equality introduced by Ponthot (1995) and Carosio et al. (2000) as well
as the thermal and athermal effects on yielding as described by Voyiadjis and Abed
(2006) who noted that temperature and strain rates have small effects on harden-
ing curves, but mainly contribute to the change of yielding points. The numerical
approach also involved the predictor-corrector algorithm of return mapping and
Newton’s method which are characterised by high rates of convergence. The re-
sults in case of rate-dependent non damaging materials showed a good agreement
with those obtained by Regenauer-Lieb and Yuen (1998); they confirmed that shear
heating produces weakening. A comparative study showed that continuum dam-
age contributes significantly to the material softening and reduces the previously
estimated reaction forces in the lithosphere. Damage also plays the role of a loading-
rate-controlled switching mechanism between brittle and ductile responses. It was
also shown that continuum damage can reduce considerably the strength of the
lithosphere, especially at low loading rates. Plate tectonic deformation may indeed
be lubricated by damage mechanics.
82 CHAPTER 4. CONTINUUM DAMAGE MECHANICS
Chapter 5
Thermodynamic framework
5.1 Introduction
Thermodynamics provides a unified framework to couple mechanics and chemistry,
with practical numerical applications for realistic geodynamics problems. The work
presented in this chapter is developed within the framework of classical thermody-
namics, where the effect of chemical feedbacks are included by taking into account
the fluxes exchanged by a Representative Volume Element with its surrounding.
Earlier suggestions are followed using full explicit coupling, only to find that they
are numerically intractable for realistic geodynamics problems with a number of de-
grees of freedom potentially exceeding one hundred. As a method of reducing this
number a multi-scale approach is employed where the given scale of interest allows a
separation of direct and indirect feedbacks. Indirect feedbacks are not solved in the
system of equations but are incorporated through pre-calculated thermodynamic
databases. Direct feedbacks are calculated in the framework of thermodynamic
equations and solved explicitly to define the dissipative structures emerging out of
those feedbacks. Thus a framework is proposed that can be used to extend in a
computationally manageable manner the linear far-from-equilibrium theory from
Prigogine and co-workers into the non-linear regime for thermo-chemo-mechanical
83
84 CHAPTER 5. THERMODYNAMIC FRAMEWORK
problems.
The emergence of localisation phenomena is a key to understand problems rang-
ing from small scale mechanical behaviour to the large behaviour of lithospheric
plates. Those problems can be understood in a thermodynamical sense as dissi-
pative structures (Prigogine and Lefever, 1968). This term describes the patterns
which self-organise in far-from-equilibrium dissipative systems such as the chem-
ical oscillations shown in fig. 18. Application of this theory to mechanics is not
novel but since its introduction by Prigogine it has been frequently regarded as in-
herently too difficult in the geomechanical community because of two fundamental
problems. First, explicit calculation of thermodynamical fluxes requires very large
computational engines that were not commonly available until recently. Second,
and more fundamentally it was perceived that thermodynamics can not uniquely
define a mechanical problem. The reason for this concern is that knowing all
macroscopic variables one can only determine the material parameters in the con-
stitutive laws if there are fewer parameters than observable constitutive relations.
However it was pointed out by Kocks et al. (1975) that in crystal plasticity the
number of relevant material parameters may far exceed the number of macroscopi-
cally observable relations; the use of energy, volume, etc., as “state variables” then
becomes meaningless. The focus of this text is on the first problem by develop-
ing a thermodynamical framework that allows computational solutions for coupled
thermo-chemo-mechanical problems within reasonable computing time. Moreover
this framework is self-consistent as the heat and chemical feedbacks are tracked ex-
plicitly. The approach presented also addresses the second point of concern, where
the dimensionality of the mechanical problem is reduced into a numerically tractable
unique solution through a method of averaging variables (Rice, 1971). This allows
to extend far-from-equilibrium thermodynamic theories such as Prigogine’s theory
into the non-linear regime, with simplifications that make it possible for solutions
to be practical from the geological outcrop to the large geodynamics scale.
5.2. LITERATURE REVIEW 85
Figure 18: Liesegang patterns in a sandstone showing diffusion and chemical oscil-lations. Those patterns can be explained by auto-catalytic chemical reactions. Photocourtesy of Ron Vernon.
5.2 Literature review
Classical solid mechanics was initially formalised on mathematical concepts for de-
scribing the different relationships between actions and motion. The introduction of
thermodynamics offered a rigorous framework in which mechanical models can be
developed at least under the assumption of thermodynamic equilibrium. However,
this assumption is particularly deficient in describing dynamical processes that are
inherently time dependent as the heat is generated, conducted and dissipated in
the neighbouring domains (Baker, 2005). The notion of quasi-static deformation is
incompatible with the inherent time dependency introduced by the energy equation
and is an extreme end-member valid for a quasi-steady state. This incompatibil-
ity is mainly due to the difference between the primary length scales involved in
the equations of motion, heat transfer, mass transport and the secondary length
scales related to the different rheology mechanisms. Deviating slightly a system out
86 CHAPTER 5. THERMODYNAMIC FRAMEWORK
of thermodynamic equilibrium can result in coupled reactions which may involve
motion, heat transfer, matter transfer, etc. Within this framework of small ther-
modynamic perturbations, thermo-chemico-mechanical processes can be described
in a coupled manner where equations of balance and dissipation can be identified
and used in a closed loop. In the following a brief overview is given of some existing
thermodynamical approaches which tie the disciplines of chemistry and mechanics.
Prigogine and Glansdorff were the first ones to look into the stability of non-
equilibrium stationary thermodynamic states (Glansdorff et al., 1973). They used
a linear stability analysis to relate the rate equations of a chemical system to the
emergence of dissipative structures. The limitation of this approach was that far-
from-equilibrium systems were considered. However they were linear systems char-
acterised by linear partial differential equations. This allowed these authors to
obtain solution from a linear stability analysis and describe the system from the
stability of non-equilibrium states through the assessment of Lyapunov functions.
The predictive power of this method was limited for the particular case where the
ordinary differential equation is linear and temperature and pressure variations were
neglected. Since this assumption is a good first approach to chemical systems the
theory gained wide support for the description of non-equilibrium states that were
called dissipative structures.
The basic underlying principles for the pattern formation seen in fig. 18 go back
to the so-called Brusselator concept introduced by Prigogine and Lefever (1968).
The Brusselator shows how a non-equilibrium system can develop into oscillations
through a self-accelerating feedback process called an auto-catalytic reaction and
how non-linearity can lead to the spontaneous generation of ordered patterns. The
Liesegang rings shown in fig. 18 can then be seen as a natural example of auto-
catalytic chemical reactions in geology (Lebedeva et al., 2004). Many more such
oscillating chemical systems have been identified since then also in other fields such
as bio-chemistry, e.g. the life cycle of the cellular slime mould (Kondepudi and
5.2. LITERATURE REVIEW 87
Prigogine, 1998). The theory can readily be extended into the non linear far-from-
equilibrium regime through explicitly solving the non-linear differential equation
that is encountered in more realistic systems. Kuhl and Schmid (2007) for example
solved a fourth order equation and obtained results very similar to the Brusselator.
This is encouraging to pursue a general development for non-linear systems that
can be solved by numerical methods.
The above described processes did not deal explicitly with deformed solids under
large strain. In fact, deviatoric stresses were neglected in all approaches (Kocks
et al., 1975) . Most classical approaches also make the very strong assumption
of isothermal deformation which ignores the basic underlying physical theory of
thermodynamics and limits these approaches to the classical plasticity theory. The
consideration of thermodynamics in solid bodies that sustain shear and not only
volumetric deformation was pioneered by Ziegler (1977) who coined the term ther-
momechanics. His theory applies to the case of quasi-static deformation and pro-
vides such a thermodynamic consistency check in the strong form of the second
law of thermodynamics. The principal merit of the thermodynamic approach is
indeed to ensure that the mathematical framework suggested by plasticity theory
does not violate the second law of thermodynamics (Halphen and Nguyen, 1975b).
For frictional materials, an extension was suggested by (Collins and Houlsby, 1997)
who define self-consistently the yield criterion and the flow law from the postulate
of a dissipation function (see also (Houlsby and Puzrin, 2007)).
The incorporation of temperature dependency of the material properties into the
criteria for localisation has a long history but has not specifically been tied to
thermodynamics. Explicit calculation of a thermal feedback mechanism was found
by Gruntfest (1963) who analysed the behaviour of temperature sensitive fluids.
Shear heating feedback is based on the mechanics of conversion of deformational
work into heat for cases where a higher temperature lowers the viscosity; a small
fluctuation in temperature then accelerates the strain rate and in turn produces
even more shear heating, leading to a self propagating “thermal runaway”.
88 CHAPTER 5. THERMODYNAMIC FRAMEWORK
In a thermomechanical interpretation of infrared thermography deformation exper-
iments relaxation of the isothermal assumption was proposed by Chrysochoos and
Dupre (1991). A Portevin-Le Chatelier localisation shown in fig. 19 is an illustra-
tion of this non-isothermal localisation feedback. This feedback has been applied
to geoscience applications to explain deep earthquakes (Braeck and Podladchikov,
2007; Hobbs et al., 1986; Ogawa, 1987; Orowan, 1960) and localisation phenomena
in plate tectonics (Kaus et al., 2005; Regenauer-Lieb and Yuen, 1998). In geody-
namics a thermodynamics framework was put forward by Regenauer-Lieb and Yuen
(2003). An illustration of such localisation is shown in fig. 20.
Figure 19: Heat sources (in W.m−3) from infrared camera showing the evolution ofPortevin-Le Chatelier bands in a time series of AlMg alloy in extension. Picture fromLouche et al. (2005).
An extension of the approach to the chemical system with large shear deformation
was recently proposed (Regenauer-Lieb et al., 2009). It differs from Prigogine’s
5.2. LITERATURE REVIEW 89
Figure 20: A stiff and straight layer in a soft matrix is compressed (Hobbs etal., 2009). The lateral dimension after 54% shortening is 9.9 km. The localisationphenomenon from shear heating feedback causes the formation of shear bands andfolding of the layer. Localisation operates on the scale of hundreds of metres to km.The colour legend indicates the shear stress in Pa.
90 CHAPTER 5. THERMODYNAMIC FRAMEWORK
approach by considering shear deformation and non-linear thermodynamics. How-
ever it is still taking the extreme end-member view of isothermal deformation. In
this formulation the chemical diffusion process is formally equivalent to the thermal
diffusion process and the chemical reaction process to the thermal shear heating.
Consequently the shear heating theory can be mapped one to one over into a chem-
ical localisation theory. However an important difference between the two diffusion
processes is that the characteristic length scales are very different for the same
time scale. The thermal diffusion operates at kilometre scales while the granular
chemical diffusion takes place on the sub-centimetre scale (see fig. 21). Another
important difference to note is that geological velocities (with strain rates of the
order of 10−16 to 10−14 s− 1) are on the same order as the chemical diffusion rates,
leading to potential kinematic trapping of chemically reacting species.
A full thermo-chemo-mechanical framework was suggested recently (Rambert et al.,
2007). The approach suggested there is based on a gradient plasticity approach
which inherently considers all possible length scales in a single framework. This is
an elegant solution but is computationally expensive as those gradients are added
as extra degrees of freedom in the numerical solution. This approach has not been
applied to geodynamics to date.
A different approach is presented here that also relies on a non-linear far-from-
equilibrium thermodynamic context but using a multi-scale formulation. This ap-
proach is closer to the one presented in (Coussy, 2004) but presents a novel for-
mulation focusing on the thermodynamical fluxes exchanged by the system at its
surface.
5.3 Multiple scales approach
To describe a thermodynamic system one needs to define three different scales of
interest. The micro-scale can be thought of as the scale of atoms and molecules,
the meso-scale is the next scale up, at least at the level of grains or bigger, and the
5.3. MULTIPLE SCALES APPROACH 91
Figure 21: A stiff and straight layer in a soft matrix is compressed (Regenauer-Liebet al., 2009). The lateral dimension after 54% shortening is 3.7 cm. The localisationphenomenon resulting from chemical diffusion causes folding of the stiff layer. Thelocalisation mechanism operates on the mmcm scale. Contours are showing the shearstress in Pa.
92 CHAPTER 5. THERMODYNAMIC FRAMEWORK
macro-scale is at the level of geodynamic, geo-mechanical problems. A continuum
framework is considered where every material point can at least conceptually be
imagined as a continuum itself. This allows to simulate the essential features of the
true interactions between real physical entities at any given scale within the same
framework despite the extreme difference in nature of the processes considered at
different scales.
5.4 Representative Volume Element
A representative volume element (RVE) Ω is considered of a material specimen un-
dergoing non-uniform inelastic deformation from an arbitrary fixed reference con-
figuration. Its evolution is studied with an Eulerian approach, that is involving
only its current configuration. This RVE is defined as an open domain exchanging
mass, work, momentum, and heat with its surroundings through its surface δΩ. The
choice of a suitable RVE size for a given observation time is the crucial step for the
assumption of continuum RVE in thermodynamic equilibrium, RVE embedded in
a larger system that is not in equilibrium.
Consider two thermodynamics processes happening at two very different times and
length scales. Define t1 as the time to reach local equilibrium in a reference volume
much smaller than the size of the system under study and t2 as the time required to
reach the equilibrium in the entire system, t1 ≪ t2. Following Onsager’s regression
hypothesis (Onsager, 1931) the time evolution of the fluctuation of a given physical
value in an equilibrium system obeys the same laws on average as the change of
the corresponding macroscopic variable in a non-equilibrium system. t1 is chosen
as the smallest time step that will be used in the numerical simulation of the whole
model and derive a corresponding size that defines the arbitrary RVE in local
thermodynamic equilibrium. Note that the time and length scales of this RVE
are large enough for the ergodic hypothesis of thermodynamics to hold, as the
discrete nature of the underlying processes are not considered. The axiom of local
5.5. THERMODYNAMICS BACKGROUND 93
state is then adopted as in the classical theory of non-equilibrium thermodynamics
(de Groot and Mazur, 1962).
5.5 Thermodynamics background
This section defines all classical thermodynamic properties and equations involved
in the framework. T denotes the local absolute temperature and s the specific
entropy of the system (considered per unit mass). The specific internal energy e
can be decomposed as the sum of the internal heat energy Ts and the specific
Helmholtz free energy ψ
e = ψ + T s . (105)
ψ is a thermodynamic potential and must then be a function of a set of independent
state variables. It is a difficult problem to know which variables to use as state
variables and Collins and Houlsby (1997) showed that elastic and plastic strains
can only be taken as state variables when the elastic behaviour at the micro-level
is linear, which is a good approximation in geodynamics for example but not so
good for micro-polar materials (Collins, 2005). At this stage, the smallest scale is
selected in order to avoid micro-polar complexity. For geodynamic modelling the
stress tensor is therefore symmetric.
The local state of a material point is assumed to be characterised by the set
(T, ǫe, D, αk) of independent variables, where ǫe = (ǫeij)1≤i,j≤3 is the elastic part
of the local total strain tensor ǫ relative to a fixed reference configuration, D is a
damage parameter and αk1≤k≤n represent a set of n local internal state variables
considered in the reference configuration. The damage formulation chosen follows
the original thermodynamic framework from Lemaitre (1985) and Chaboche (1987).
An extension to geodynamic applications where multiple mechanism of void growth
and nucleation was suggested by Karrech et al. (2011c). At the arbitrary but macro-
scopic scale of the RVE, those state variables are averaging variables (Rice, 1971)
94 CHAPTER 5. THERMODYNAMIC FRAMEWORK
derived from the underlying micro-level for all processes considered. The identifica-
tion of those variables depends on the processes targeted and this chapter mainly
focuses on the diffusivity of chemical components and therefore denotes by αk the
number of moles of the kth chemical constituent. The notation for αk can however
be read in a more generic way and can also include other parameters such as those
related to damage mechanics, gravitational or electrical potentials, surface tension,
variation in fluid content, dislocation density, crystallographic preferred orientation,
etc. That choice of variables leads to
ψ = ψ(T, ǫe, D, αk) . (106)
To simplify the notation, the same symbol is used for a function and its value and
the set of independent variables is omitted whenever it is evident or has been defined
previously.
For this RVE the First Law of thermodynamics spells out the energy conservation.
The time rate of change of kinetic energy K and total internal energy E is equal to
the sum of the external power and heat supplied.
dE
dt+dK
dt= Pext +Q (107)
Using the principle of virtual power to relate the internal and external power with
the acceleration power
Pacc = Pext + Pint (108)
and the fact that dKdt
= Pacc the First Law can be rewritten as
dE
dt= −Pint +Q (109)
The heat term can be expressed through a volumetric source term r and an outgoing
5.5. THERMODYNAMICS BACKGROUND 95
heat flow vector q as
Q =
∫
Ω
r dV −∫
δΩ
q.n da =
∫
Ω
r − dev(q) dV (110)
where da is a material surface oriented by the outward unit normal n. The internal
power can be written as the sum of mechanical and chemical terms as defined in
(Nguyen et al., 2007):
Pint = −∫
Ω
σ : ǫ dV −∫
Ω
µkαk dV (111)
where µk represents the chemical potential of the kth chemical constituent. There-
fore the last term in eq. (111) is a sum over all constituents. The term “chemical
potential” must be taken here in a generalised sense as it is not necessarily restricted
to chemistry. Each variable µk is the dual variable of the associated state variable
αk and can also denote a gravitational, electrical potential or surface potential for
example as explained earlier in the definition of the set (αk) in its generalised sense.
In eq. (111) σ = (σij)1≤i,j≤3 represents the elastic stress and is work-conjugate to
the elastic strain ǫ. The usual dot notation is used for the material derivative ǫ =dǫ
dt.
The energy equation can then be written as
dE
dt=
∫
Ω
σ : ǫ dV +
∫
Ω
µkαk dV +
∫
Ω
[
r − div(q)]
dV (112)
It is important to note that the presence of the heat term r in equation eq. (112)
is a consequence of the choice of RVE which is the lowest scale considered for all
equations derived. The mechanisms (e.g. radioactive decay) that account for the
source term on the right hand side of eq. (112) are therefore hidden in the total
internal energy term E without decomposing E in different internal energies (e.g.
for all isotopes involved).
Some useful calculus formulas are mentioned now (see for example (Coussy, 2004))
96 CHAPTER 5. THERMODYNAMIC FRAMEWORK
which are valid for any arbitrary field G, using the velocity of a given particle:
d
dt
∫
Ω
GdV =
∫
Ω
∂G
∂t+ div(Gu)dV (113)
div(Gu) = u.grad(G) +G.div(u) (114)
G =G
dt=∂G
∂t+ u.grad(G) (115)
From the definitions of specific internal energy e, specific entropy s and density ρ,
the total internal energy E, entropy S and mass M get written as
E =
∫
Ω
ρe dV (116)
S =
∫
Ω
ρs dV (117)
M =
∫
Ω
ρ dV (118)
Using eqs. (113) and (115) to (118) and considering the thermodynamical fluxes
crossing the RVE, the balance equations of energy, entropy and mass are written
differently as
dE
dt=
∫
Ω
[ρ+ ρdiv(u)] e dV +
∫
Ω
ρedV +
∫
δΩ
ρke vk.nda (119)
dS
dt=
∫
Ω
[ρ+ ρdiv(u)] s dV +
∫
Ω
ρsdV +
∫
δΩ
ρks vk.nda (120)
dM
dt=
∫
Ω
[ρ+ ρdiv(u)] dV +
∫
δΩ
ρke vk.nda (121)
where ρk and vk represent the density and the entering velocity of the kth chemical
component in Ω through the surface da following an underlying process considered
which does not need to be detailed in this framework. This is a generic way to
consider any process and can conveniently model for example fluid flow in a porous
medium without having to consider a full framework as developed by (Coussy, 1995).
5.6. FROM THE SECOND LAW OF THERMODYNAMICS 97
Equations eqs. (112) and (119) provide the local form of the energy equation as
for undamaged material see (Boving andGrathwohl, 2001)
0.0011
Damage multiplying coefficient λ 100.
Many studies can be found in the literature on carbon dioxide diffusion in rocks
(Lai et al., 1976; Penman, 1940). Rock permeability and porosity are the major
parameters controlling the diffusion process and in (Boving and Grathwohl, 2001)
for instance the diffusion coefficient k was described by the following relationship
k = kaqφm (142)
where φ denotes the porosity, m a material dependent fitting parameter ranging
from 1.3 to 2 and kaq the aqueous diffusion coefficient in pure water. By taking
advantage of the implicit relationships between damage and both porosity and tor-
tuosity, a similar description of diffusion is proposed as:
k = k0 + λD (143)
The coefficient k0 represents the diffusion coefficient in an undamaged structure
5.9. RESULTS AND DISCUSSION 103
and λ is a multiplying factor which emphasises the accelerated diffusion process
in damaged rocks. Faults can allow high-flux flow (Sibson, 2000) and a constant
numerical value for λ is arbitrarily chosen in this conceptual example (see table 8).
5.9 Results and discussion
fig. 22 shows a sequence of values for the damage parameter in the model for up to
229,000 years and 4.5% extension. fig. 23 shows the corresponding time frames for
the normalised concentration of carbon dioxide. Damage accelerates the localisation
phenomenon and distinct shear bands are formed that cross the whole model across
its height, enabling therefore carbon dioxide degassing to reach the surface through
the more diffusive damaged areas. This generic model is not calibrated on any
particular geological example but shows clearly how damage can be considered as a
tool to break impermeable seals at depth and create channels for the carbon dioxide
to reach the surface.
This numerical example also demonstrates how the framework presented allows to
simplify the problem by considering thermodynamic fluxes only without having to
resolve the modelling of the underlying physical processes. The same example ap-
plies for both chemical diffusion and Darcian flow by changing the numerical values
for the diffusivity. It is important to note however that the framework described is
nonetheless very general and can be used as a starting point for modellers with a
more specific problem at hand. To study for example fluid flow in a porous medium
one can consider separately the fluid and solid part of the RVE and re-derive
Coussy’s framework (Coussy, 2004).
The goal of this chapter is to provide a framework for linking structural geology
observations to geodynamic modelling. In order to do so one needs to go through
vastly different length scales, beginning with micro structural observations, linking
them to field observations and interpreting the results by the largest hierarchical
driver which is plate tectonics. As in any thermodynamic problem these scales are
104 CHAPTER 5. THERMODYNAMIC FRAMEWORK
Figure 22: Damage parameter after (a) 77,000 years and 1.5% extension, (b)150,000and 3% extension, (c) 229,000 years and 4.5% extension. Damage zones havelocalised which connect the reservoir at the bottom of the model to the surface.
5.9. RESULTS AND DISCUSSION 105
Figure 23: Relative CO2 concentration after (a) 77,000 years and 1.5% extension, (b)150,000and 3% extension, (c) 229,000 years and 4.5% extension. Damage zones havelocalised which increased the CO2 diffusivity and allowed degassing at the surface.
106 CHAPTER 5. THERMODYNAMIC FRAMEWORK
defined by the sum of products of thermodynamic forces and their corresponding
fluxes which identifies the rate of change of entropy production. It is thus possible
to link that rate to the underlying basic physics. Likewise the internal energy can
also be linked to the underlying physics by solving either for the Gibbs free energy
or the Helmholtz free energy. This choice predicts thermodynamic length scales
and time scales. For instance those scales are linked in one-dimensional diffusive
processes by the equation
l ≡ 2√κt (144)
where l is the diffusion length, t the relaxation time and κ the diffusive constant
for the specific process modellers chose to consider. Note that eq. (144) only lists
diffusion length scales as examples and is not meant to be comprehensive. In a gen-
eralised thermodynamic system other metrics will be encountered such as reactive
and convective length scales. In order to illustrate this concept further, these scales
are related to physical processes in geodynamics and structural geology to give the
reader a better understanding of the mesh size and time steps involved. Here are
three examples for an arbitrarily time step of 3e10s ( 1000 years):
• Carbon dioxide diffusion is controlled by Darcy’s law. For the example se-
lected κ = 1.1e−3m2s−1 and then l = 11.5 km.
• Heat conduction in a solid is driven by Fourier’s law. For a granite buried at
10km depth κ = 7.5e−7m2s−1 and then l = 300m.
• Chemical diffusion in a crystal grain follows Fick’s law. For the case of oxygen
in quartz κ = 1.1e−19m2s−1 and then l = 115µm.
For a given geodynamic problem, the formulation can hence be anchored on the
basis of the real underlying physics and the problem can be solved with the full
thermodynamic framework.
The approach by Houlsby and Puzrin (2007) is more theoretical than practical
5.9. RESULTS AND DISCUSSION 107
because they are dealing with a large number of degrees of freedom. In their formu-
lation for porous continua for example they define their system with 57 equations
without considering any chemical reaction. Gradient theory approaches require
even more equations as they consider in addition to the above approach one or
more gradients of the total strain (Rambert et al., 2007), where the first order gra-
dient is a 3rd order tensor and the second order gradient is a fourth order tensor.
These approaches are arguably not practical for realistic simulations with the cur-
rent computing power available. Consequently, simplifications are introduced in
order to fulfil the postulate that present advances in computational power allow
the extension of far-from-equilibrium thermodynamic theories into the non-linear
regime. These simplifications are particularly suited to geodynamic systems with
vastly different length and time-scales.
The first simplification makes use of the inherent multi-scale nature of the prob-
lem. The complexity of the problem lies embedded in the multitude of physical
dissipation processes that control the dimensionality of the thermodynamic system
and hence the emergence of dissipative structures. A multi-scale problem lends
itself to reduction of dimensions of the underlying partial differential equations if a
given pre-set time and length scale are considered. For this pre-set scale, emergent
properties can be derived as spatial and temporal averages. Physical processes that
cannot be derived as average properties and hence need to be considered for the
explicit modelling of dissipative structures can be derived from the rate of change
of entropy production as discussed in eq. (144). A large plate tectonic length scale
of tens of kilometres and million of years time scale implies, for instance, a criti-
cal length scale governed by the thermal diffusion process. At this scale far from
equilibrium solutions to chemical or fluid problems need not be resolved. This is
because their inherent time and length scales are very much shorter (see time scales
for chemical diffusion or fluid flow discussed for eq. (144)). They are expected to
experience small perturbations from equilibrium according to Onsager’s regression
hypothesis. For instance, by assuming for the plate tectonic scale the physics of a
108 CHAPTER 5. THERMODYNAMIC FRAMEWORK
“creeping flow” and by using a flow potential that is normal to the second invariant
of the deviatoric stress tensor, only two critical quantities need to be considered
in the energy equation: the Peclet number describing the diffusion process and the
dissipation number describing the shear heating process (Regenauer-Lieb and Yuen,
2004). For general flows a third number would normally be considered (Cherukuri
and Shawki, 1995) which describes inertial processes.
The second simplification is to differentiate the different feedback processes in terms
of direct and indirect feedbacks. Direct feedbacks are defined as all couplings that
are solved explicitly by the numerical formulation of all partial differential equa-
tions. For example the energy equation allows to take into account shear heating
feedback (Regenauer-Lieb et al., 2006) or chemical processes (Regenauer-Lieb et al.,
2009). Indirect feedbacks are defined all couplings that are introduced by making
material properties dependent on other primary variables without solving for these
dependencies explicitly in the system of equations. However these material proper-
ties are calculated self-consistently from Gibbs minimisation techniques and stored
in a database for interpolation purposes (Siret et al., 2009), as presented in chap-
ter 2. The scale of observations provides the best mechanism for identification of
direct and indirect feedbacks.
Chapter 6
Reactive transport with damage
mechanics
6.1 Introduction
The natural complexity of geological systems has motivated researchers to consider
the processes of thermal transfer (T), hydraulic flow in porous media (H), mechan-
ical deformation (M) and fluid-rock chemical interactions (C) in a coupled manner.
These processes when considered individually rarely explain observed geological
features. Different combinations of those THMC processes have previously been
studied and are based on theoretical frameworks like the one developed by Coussy
(2004). Most THMC applications; however, occur in engineering domains such as
nuclear waste disposal, gas and oil recovery, hot-dry-rock geothermal systems, or
contaminant transport (Lanru and Xiating, 2003), and it is still rare to see THMC
analyses of larger scale geological systems such as those involved in ore body for-
mation (e.g. Lanru and Xiating, 2003; Tsang et al., 2004; Shao and Burlion, 2008).
This observation can be explained by two main reasons. The first one is that cou-
pled numerical simulation of all processes still represents a significant computational
challenge and cannot currently be solved within weeks, the time-scale of a mineral
109
110 CHAPTER 6. REACTIVE TRANSPORT WITH DAMAGE MECHANICS
exploration programmes (Potma et al., 2008). The second reason, which follows,
is that geo-scientists understand the importance of conceptualising their problems
(Andersson and Hudson, 2004) and are often able to identify a subset of relevant
processes which represent the first-order controls in their studies. These restrictions
often disregard the complex feedback interactions between processes, which might
not always be negligible.
There are many different potential feedbacks between all processes (see Lanru and
Xiating, 2003, for example) and the complexity of this set of checks and balances
justifies coupled THMC simulations for various geological scenarios. Numerical
simulations are specifically adapted to understand and isolate particular feedbacks
as they allow scenarios to be investigated methodically under changing conditions,
including the sets of processes considered. They often provide an indispensable
tool to analyse the relative importance of various feedback mechanisms and the
competition of rates of processes. Coupling mechanisms are critical factors for the
localisation of geological structures such as folds or shear zones, which are often
vital to the formation ore bodies. Shear heating in thermo-mechanical simulations,
for example, has helped to understand shear zone formations at the kilometre scale
(Regenauer-Lieb and Yuen, 2004). Additional coupling mechanisms can inhibit or
accentuate those behaviours, and damage mechanics is now a recognised means to
enhance localisation (Regenauer-Lieb, 1998; Bercovici and Ricard, 2003; Karrech
et al., 2011a), as shown in chapter 4.
This chapter presents a new THMC code designed to increase the understanding
of hydrothermal and geothermal geological systems. This coupled code is based
on escriptRT (Poulet et al., 2012a, see chapter 3) for the thermal, hydraulic
and chemical processes, as well as an Abaqus (ABAQUS/Standard, 2008) user
material implementation (Karrech et al., 2011a, see chapter 4) for the mechanical
deformation, including continuum damage mechanics. The chapter is composed of
three parts. Firstly, the constitutive models for the THMC processes are introduced,
including the important feedbacks considered (Section 6.2). These include a link
6.2. CONSTITUTIVE MODEL 111
between the damage parameter and the evolution of porosity, affecting in turn the
rock permeability. The numerical implementation is then presented which details
the coupling between the two codes used, escriptRT and Abaqus (Section 6.3).
Finally, Section 6.4 illustrates the importance of considering all THMC processes
for mineral exploration by simulating a generic unconformity-related albitisation
deposit scenario.
6.2 Constitutive model
The constitutive model presented in this chapter applies to a fully saturated porous
medium, where a fluid phase interacts with the host rock mechanically, energet-
ically and chemically. This model integrates several processes, that are coupled
sequentially. A Representative Volume Element (RVE) Ω of a material specimen is
considered, which contains a solid skeleton and a fluid saturating the porous space,
and undergoes non-uniform inelastic deformation from an arbitrary fixed reference
configuration.
6.2.1 Continuum damage mechanics
The mechanical model used for the skeleton is based on a damaged visco-plasticity
model for frictional geomaterials under the assumption of small deformation, as pre-
sented in chapter 4 (Karrech et al., 2011a). This model is described using a classical
Helmholtz free energy ψ(ǫij,αij, T,D) where ǫ represents the total strain tensor,
α the inelastic strain tensor, T the temperature and D a scalar damage parame-
ter. This free energy function along with the second principle of thermodynamic
results in a Clausius-Duhem inequality which allows to obtain some constitutive
relationships of the form:
σ = ρ∂ψ
∂ǫe, s = −∂ψ
∂T, and Y = ρ
∂ψ
∂D, (145)
112 CHAPTER 6. REACTIVE TRANSPORT WITH DAMAGE MECHANICS
where σ the Cauchy stress tensor, s is the entropy, ρ the density, and Y the ther-
modynamic force associated to the damage parameter D. In case of isotropic linear
behaviour, the involved thermodynamic forces can be expressed as (see chapter 4
or (Karrech et al., 2011a) for details):
σ = (1−D)(K − 3
2G)tr(ǫe)1+ 2G(1−D)ǫe , (146)
s = c(T − T0) + 3αK(1−D)tr(ǫe) , (147)
Y =−σ2
eq
2(1−D)2
[
1
3G+
1
K
(
σHσeq
)2]
, (148)
where K and G are respectively the bulk and shear moduli, c is the heat capacity,
α is the thermal expansion coefficient, σH = 13tr(σ) the hydrostatic pressure, and
σeq =√
32s : s the equivalent stress, with s = σ − σH1 representing the deviatoric
stress. This description is based on the definition of effective stress as introduced
by Kachanov (1958) and used extensively by Lemaıtre and Chaboche (2001): (1−D)σ = σ. A custom dissipation function is also postulated (see chapter 4 or Karrech
et al. (2011a) for details) to account for shear dissipation, volumetric change, rate
sensitivity, and damage. Combined with the principle of maximum dissipation
(Ziegler, 1963), this dissipation function is used to relate the thermodynamic forces
to their respective forces (Karrech et al., 2011b). The assumed damage potential
considered herein defines the evolution of the isotropic damage parameter with
time and takes into account void nucleation through a linear relationship involving
strain rate and damage. This potential gy is based on the theory of limit analysis.
It accounts for the nucleation and growth of voids and defects in the dissipative
regime and reads:
gy =
(
1
(1−D)n+1− 1
)
Y +H
κ+ 1(Y
H)κ+1 (149)
6.2. CONSTITUTIVE MODEL 113
where H and κ are material constants used to describe damage nucleation (Karrech
et al., 2011a), and n is the exponent of the dislocation power law1. This potential
is used to calculate the incremental variation of damage with respect to loading
parameters; deriving the potential with respect to the thermodynamics force of
damage Y results in a flow direction which orients the damage evolution:
D = λ
(
1
(1−D)n+1− 1 + (
Y
H)κ)
(150)
where λ is a Lagrange multiplier which is proportional to the equivalent inelastic
deformation. The usage of this recent model is an important component of this
study as (Karrech et al., 2011a, see chapter 4) show that the frictional behaviour
of geomaterials highly influences fault orientations.
6.2.2 Porosity
The evolution of the rock porosity φ is considered from its initial value φ0 due to
mechanical and chemical processes (Kuhl et al., 2004)
φ− φ0 = φem + φpm + φc , (151)
where φem and φpm represent the elastic and plastic partitions of the porosity varia-
tion due to mechanical deformation (Armero, 1999), and φc represents the porosity
evolution due to mineral dissolution and precipitation.
The calculation of φpm, the porosity update due to continuum damage, is based
on an elementary interpretation of damage as spherical void growth in rocks. The
scalar damage parameter D for a given RVE can be interpreted as the proportion of
void surface intersecting the RVE boundary surface over the whole surface (Cocks
and Ashby, 1980). Considering a RVE of characteristic length R containing voids
of radius r lead to the relationship D ∝ ( rR)2. The porosity φ is itself defined as
1The rheology used herein combines dislocation and diffusion mechanisms in series.
114 CHAPTER 6. REACTIVE TRANSPORT WITH DAMAGE MECHANICS
the volume ratio of the void compared to the whole RVE, hence φ ∝ ( rR)3. The
relationship φ ∝ D32 can then be deduced. Considering a maximum porosity value
of φmax for a totally damaged element, the following damage-porosity evolution can
then be postulated:
φpm = (φmax − φ0)D32 (152)
6.2.3 Effective stress
All pores are assumed to be fully saturated with an interstitial ideal fluid which
exerts a static pressure p on solid grains. Following Biot’s approach (Biot, 1941)
this pressure term is accounted for to calculate the effective or equivalent stress as
σijeq = σij + bpδij . (153)
b is the Biot coefficient defined in (Coussy, 2004) as
b = 1− K
ks, (154)
where K and ks are respectively the bulk moduli of the empty porous solid and of
the solid matrix forming the solid part of the porous solid2. The elastic part of the
evolution of porosity can be expressed (Coussy, 2004) as
dφem = bdǫ− αφdT +dp
N, (155)
where ǫ represents the volumetric strain, αφ is the thermal expansion coefficient of
the porosity, and N is the Biot modulus defined as
1
N=b− φ0
ks. (156)
2Note that if the solid matrix is incompressible ks becomes infinite and b = 1. Biot’s effectivestress σij
eq then reduces to the classical Terzaghi’s effective stress σij′ of soil mechanics defined by
σij′ = σij + pδij .
6.2. CONSTITUTIVE MODEL 115
6.2.4 Fluid transport
Fluid transport through the porous medium is described using the classical Darcy
law under the assumption of quasi–static fluid flow and neglecting the tortuosity
effect:
q = − k
µf(∇p− ρfb). (157)
where q denotes the Darcy flux vector, k the rock permeability, µf the fluid viscosity
and b the body force. The pore pressure equation is derived from the fluid mass
conservation law linking the fluid density ρf , porosity φ, Darcy flux q and the solid
velocity vs in the reference configuration. Its local form (see for example (Coussy,
2004)) can be written as
d(φρf )
dt+ φρf∇ · vs +∇ · (ρfq) = 0. (158)
The elastic part of the porosity φem is considered to be dependent on pore pressure
p, temperature T and volumetric strain ε, which can be expressed as
dφemdt
=∂φem∂p
dp
dt+∂φem∂T
dT
dt+∂φem∂ε
dε
dt(159)
This relation leads to
d(φρf )
dt= φρf
[
βrdp
dt+ γf
dX
dt− αr
dT
dt
]
+ bρfdε
dt
−φρf∇ · vs + ρf (dφpmdt
+dφcdt
) , (160)
where αr and βr are respectively the thermal expansivity and compressibility of
the porous rock, and γf is the fluid chemical expansivity defined as γf = 1ρf
∂ρf∂X
which depends on the salinity X (wrt CNaCl). The pressure equation can then be
116 CHAPTER 6. REACTIVE TRANSPORT WITH DAMAGE MECHANICS
rewritten as
βrφρfdp
dt−∇ · (ρf
k
µf∇p) = −∇ · (ρ2f
k
µfb)− φρf∇ · vs
+αrφρfdT
dt− γfφρf
dX
dt− bρf
dε
dt(161)
The fluid properties (density, viscosity, specific heat, compressibility and thermal
expansivity) are calculated from one of the three equations of state (EOS) for pure
water available in the escriptRT code (Poulet et al., 2012a, see chapter 3). This
choice allows to account for the important variation of fluid properties (especially
density) in a realistic way for low concentrations of chemical species, and with a
first order approximation for salinity.
6.2.5 Heat transport
The solid and liquid components of the medium are assumed to be in thermal
equilibrium. Heat is then transported via the standard advection diffusion equation
(Nield and Bejan, 2006) with an additional shear heating source term
where ρ Cp = φ(ρ Cp)f + (1− φ)(ρ Cp)s is an average specific heat Cp for the rock,
D = φDf +(1−φ)Ds is the averaged thermal diffusion coefficient, Q is a radiogenic
heat source, ǫdiss is the dissipative strain rate from creep and plastic deformations,
and χ is the nondimensional Taylor-Quinney heat conversion efficiency coefficient.
This study fixes the χ constant in time with a value of 0.9 in agreement with most
material values (Chrysochoos and Belmahjoub, 1992).
6.2. CONSTITUTIVE MODEL 117
6.2.6 Chemistry
In the mass transport, consider the set of chemical species in the aqueous phase
that are transported with the fluid and diffused using Fick’s law. Following (Poulet
et al., 2012a, see chapter 3) geological time scales are considered and those species
are assumed to be in chemical equilibrium with the solid minerals of the host rock
matrix at the end of every discrete time step. In other words, chemical reactions are
assumed to be infinitely faster than the advective and diffusive transport processes
for the geological scenarios considered. The transport problem is then solved firstly
for the chemical elements which constitute these aqueous species:
φ∂Ca
i
∂t+ φu · ∇Ca
i −∇ · (φD∇Cai ) = 0 , (163)
where Cai is the molar concentration of the ith chemical element in aqueous phase,
D is the mass dispersion coefficient tensor, u is the fluid velocity (q = φu) with
respect to the solid matrix.
The chemical equilibrium between the fluid and host rock is then computed sepa-
rately using theGibbsLib solver (viaGibbsLib.dll) from theHCh package (Shvarov
and Bastrakov, 1999). This solver uses a Gibbs free energy minimisation technique
at prescribed temperature and pore pressure.
6.2.7 Permeability evolution
Permeability is one of the most critical parameters in the calculation of transport
equations and there exist many empirical relationships linking it to porosity under
different assumptions (see Kuhn (2009) for example). The Blake-Kozeny equation
presented in McCune et al. (1979) was selected:
k = k0
(
φ
φ0
)(
1− φ0
1− φ
)2
, (164)
118 CHAPTER 6. REACTIVE TRANSPORT WITH DAMAGE MECHANICS
Figure 24: Comparison of permeability evolution due to increasing damage. Resultsshow an acceptable fit with a reference law presented by Pijaudier-Cabot et al. (2009),which matches two analytical solutions for very low and very high damage.
0.0 0.2 0.4 0.6 0.8 1.0Damage
0
2
4
6
8
10log(k/k0)
Reference
max=0.9
max=1
and validated its usage in combination with the damage-porosity relationship intro-
duced in eq (152) by comparing the porosity evolution as a function of the damage
parameter with a matching law presented in Pijaudier-Cabot et al. (2009). This
law was developed to model the interaction between material damage and transport
properties of concrete. However it was derived from two asymptotic cases where
theoretical modelling exists, for low and high values of damage, and can therefore be
applied to a wider range of geomaterials. Figure 24 shows good agreement between
this matching law and two results obtained with different values of the maximum
porosity φmax. φmax = 1 represents the case where the fully damaged rock (D = 1)
is completely dissolved. It reproduces very well the asymptotic behaviour of the
Poiseuille flow for large values of damage, where permeability is controlled by a
power function of the crack opening. This model however leads to excessive values
of permeability in that case (D → 1) as it can also exaggerate the value of porosity
for a fully damaged rock. The maximum value φmax = 0.9 was chosen to overcome
this problem as shown on Figure 24. The results obtained still match closely the
reference law visually, including for large values of damage. There is a loss of the
asymptotic behaviour for D → 1, which is non critical as the common practise of
capping the value of damage to a maximum of 0.9 is used.
6.3. NUMERICAL APPROACH 119
Figure 25: Simplified flow of information between escriptRT and Abaqus
6.3 Numerical approach
In order to solve the model described in Section 6.2 the capabilities of two simulation
codes, escriptRT and Abaqus, were coupled. The resulting code is based on the
modular architecture of escriptRT, which provides a suitable framework to link
to Abaqus easily.
6.3.1 escriptRT, a modular architecture
escriptRT was designed to provide a high level interface and a modular archi-
tecture to build simulations using various components implemented with different
programming languages (Poulet et al., 2012a, see chapter 3). For that purpose it
uses python, a high level object oriented language, as a binding tool for optimised
algorithms from different packages. The modularity of escriptRT mainly comes
from its usage of the escriptmodelling library and its modelframe approach (Gross
et al., 2008). This flexible approach provides an efficient mechanism to combine var-
ious processes sequentially, with different levels of coupling possible by recursively
120 CHAPTER 6. REACTIVE TRANSPORT WITH DAMAGE MECHANICS
using this framework when needed (Poulet et al., 2012a, see chapter 3). This re-
cursive method is useful for coupling two processes in a tight manner, where the
joint convergence of both codes at every time step is required. escriptRT is then
extended to account for mechanical processes using Abaqus as another external
component.
6.3.2 Mechanical modelling with Abaqus
The visco-elasto-plastic mechanical model was implemented numerically follow-
ing (Karrech et al., 2011a, see chapter 4) through the user material subroutine
UMAT of ABAQUS/Standard (2008). This implementation uses relatively new
techniques of thermal and rate dependency considerations. It also includes a predictor-
corrector algorithm of radial return mapping characterised by high convergence
rates. This subroutine implements a large deformation formulation and allows for
high strain simulations that can reproduce realistic geological structures (Zhang
et al., 2012). It also implements a tight thermo-mechanical coupling but in this
study only mechanical deformation part of the Abaqus implementation was used
while the temperature equation was solved using escript.
6.3.3 Connecting escriptRT and Abaqus
The high level interface from escriptRT is well suited to connect to other packages
as it uses python and its existing python threading and socket modules. The
master process of the coupling code is therefore escriptRT, which controls all
time steps and starts the Abaqus simulation on a thread during the initialisation
step, along with two other threads to monitor and communicate with Abaqus (see
Figure 25). The communication between both simulation codes is handled through
sockets and data files, for convenience in this initial version of the coupling code.
Communication on the Abaqus side is done through a C interface, which facilitates
the usage of threads and can exchange data easily with fortran, the language
6.4. APPLICATION TO ALBITISATION 121
Figure 26: Initial geometry and boundary conditions for the generic 2D modelused for the numerical application. The fault and lower sandstone layers are 50mthick. Boundary conditions involve the rock pressure P s, stress σ, pore pressure P ,temperature T and chemical composition c
used to implement the Abaqus subroutines. The temperature, pore pressure and
porosity values computed by escriptRT are used as input by Abaqus, which in
turn updates the damage, equivalent stress and dissipative strain rates that are used
by escriptRT to calculate the new porosity and shear heating source term. Both
simulation codes use specific mesh file formats and a converter was implemented to
convert from one file to another, keeping track of the mapping between indices for
the mesh nodes and elements.
6.4 Application to albitisation
The application of a numerical simulator which takes into account deformation,
fluid and heat transport as well as geochemical reactions is well suited to the study
of alteration and mineralisation. To demonstrate the capability of this simulator,
a simple geologic scenario is modelled which involves shear zone development and
albitisation. Albitisation is a common alteration process in which albite forms by
the replacement of primary feldspars, both K-feldspar and plagioclase. It can occur
122 CHAPTER 6. REACTIVE TRANSPORT WITH DAMAGE MECHANICS
Table 9: Initial material properties and mineralogy of the various units in the modelshown in Figure 26
over a wide range of pressure and temperature conditions, from a diagenetic envi-
ronment (Saigal et al., 1988) to a current geothermal environment (Cavarretta and
Puxeddu, 1998). Albitisation has also been observed to be associated with min-
eralising systems with extensive zones of albitisation associated with amphibolite
metamorphism and IOCG mineralisation in the Proterozoic Cloncurry district of
the Mount Isa Inlier in Australia (Oliver and Wall, 1987; Williams, 1994; Rube-
nach, 2005). An association between albitisation and uranium mineralisation has
also been recognised (da Silveira et al., 1991; Polito et al., 2009). Albitisation of
K-feldspar grains is strongly controlled by temperature and the K+/Na+ activity
ratio:
KAlSi3O8 +Na+ ↔ NaAlSi3O8 +K+ (165)
6.4.1 Problem description
In order to test the code presented in Section 6.3, a simple model was run which
simulates the process of albitisation. The initial model is a two-dimensional cross-
section (Figure 26) which is 1.5km wide and 1km deep. A 500m thick basal unit
6.4. APPLICATION TO ALBITISATION 123
comprising albite and quartz in equilibrium with an NaCl brine is overlain by a
unit comprising K-feldspar and quartz in equilibrium with a KCl brine. The two
units are separated by a 50 m thick impermeable quartz layer which ruptures during
fault development allowing fluid migration between the two units and subsequent
albitisation of the K-feldspar layer as fluids move up the fault zone. An initial fault
zone, 50 m thick with a dip of 60, is placed in the basement to localise a new shear
zone within the upper layers of the model and which deforms the quartz layer.
Mechanical, thermal and fluid properties used in the model are presented in Table 9.
Boundary conditions (see Figure 26) are applied on the top surface of the model to
simulate its burial 5km below ground, with a temperature of 150C, a pore pressure
of 49Mpa, an effective rock pressure of 76Mpa with a Biot coefficient of 0.85, and a
fixed chemical composition. A hydrostatic initial pore pressure gradient and an ini-
tial geothermal gradient of 30/km are applied through the section and the system
is equilibrated for those conditions. The model is then considered to be under com-
pression with horizontal velocity boundary conditions of 4mm/year applied on the
right hand side, no horizontal displacement on the left boundary, and no horizontal
fluid or heat fluxes on both sides. Free slip, fixed temperature and a fixed chem-
ical composition (after initial equilibrium) are applied on the bottom boundary.
Changes in porosity as a result of geochemical reactions via equation (151) were
excluded for this study, whose purpose is to demonstrate the first order importance
of porosity and permeability enhancements through damage. Full albitisation can
induce a molar volume decrease of approximately 8%. Since this simulation only
calculates the onset of albitisation, the chemical induced changes in porosity are
neglected as they are expected to be small compared to those induced by damage.
This assumption will be tested when the results are presented.
The model was first initialised without any compression to obtain an equilibrium
solution for the temperature (T ) and pore pressure (P ) fields, with fluid properties
dependent on T and P . The simulation was then run for more than 30,000 years
under compression.
124 CHAPTER 6. REACTIVE TRANSPORT WITH DAMAGE MECHANICS
(a) Without damage mechanics. (b) With damage included.
0 25 50 75 100 125 150True distance across shear zone
0.00
0.05
0.10
0.15
0.20
0.25
Equiv
ale
nt
inela
stic
str
ain
0 25 50 75 100 125 150True distance across shear zone
0.0
0.5
1.0
1.5
Equiv
ale
nt
inela
stic
str
ain
(c) Values across A-B in (a). (d) Values across A-B in (b).
0 25 50 75 100 125 150True distance across shear zone
4
3
2
1
0
log 1
0(D
)
(e) log10(Damage) across A-B in (b).
Figure 27: Effect of damage mechanics on equivalent plastic strain after30, 000 years, with the outline of the deformed quartz layer overlaid in white. Equiv-alent plastic strain values are plotted along a 150m long path (A-B) across the shearzone, showing higher and more localised strains when damage mechanics is consid-ered. A damage profile across A-B shows an exponential evolution of damage awayfrom the shear zone.
6.4. APPLICATION TO ALBITISATION 125
(a) After 15,000 years (3% strain). (b) After 30,000 years (6% strain).
Figure 28: Evolution of log10(permeability) over time, with temperature contourssuperimposed. Temperature contours vary between 150 at the top surface and 180
at the bottom boundary.
6.4.2 Results
Localisation and damage
Two mechanical models were tested: with and without the calculation of mechanical
damage. In both scenarios, plastic strain initiates at the intersection of the pre-
existing fault and the unconformity. This reactivates the original fault and then
propagates upwards into the upper layer. The fault is also reflected at the bottom
of the model towards the end of the simulation due to the choice of sliding boundary
condition, but this phenomenon occurs far from the zone of interest for our study at
the centre of the cross-section. The model without damage (Figure 27a) produces
a diffuse shear zone that appears to behave with more ductility than the discrete
more brittle fault formed in the model with damage mechanics (Figure 27). This is
illustrated by the basement offset at the quartz layer (overlaid in white on Figure 27)
which shows very angular block faulting when damage is considered. Two plots of
the equivalent plastic strain profile along a 150m long path across the propagated
126 CHAPTER 6. REACTIVE TRANSPORT WITH DAMAGE MECHANICS
Figure 29: Fluid flow and temperature map around the offset of the quartz layer atthe newly developed shear zone after 32,600 years, with boundaries of quartz layersuperimposed in white. Distances in metres from the bottom left corner of the modelare indicated on the axes.
6.4. APPLICATION TO ALBITISATION 127
(a) From damage (b) From poro-elasticity.
Figure 30: Decomposition of total porosity evolution from the initial configurationafter 32, 600 years.
shear zone (Figures 27c–d) allow a more quantitative comparison, with a more
localised shear zone and much larger equivalent plastic strain (more than 150%) in
the case when damage mechanics is considered, compared to less than 25% strain
in the other case for the same simulated time.
Damage evolution away from the new shear zone exhibits a logarithmic profile
with respect to the distance from the shear zone (Figure 27e). This behaviour is
consistent with some geological observations made by (Mitchell et al., 2011) around
the Arima-Takatsuki tectonic line in Japan, where the authors noted the same
evolution for the degree of pulverisation in the surrounding rock around the slip
zone. This observation reinforces the justification to use damage mechanics for
geological applications and link the theoretical concept of damage to porosity and
permeability evolution as done in Sections 6.2.7.
128 CHAPTER 6. REACTIVE TRANSPORT WITH DAMAGE MECHANICS
Figure 31: Subsection of model showing distribution of albite mineralisation andassociated alteration around fault zone after 32,600 years. a) K-feldspar is replacedby albite immediately above the fault zone. Number of moles are related to volumepercentages by comparison with Figure 31b. Unconformity and initial fault contoursin white dotted lines. Distances in metres from the bottom left corner of the modelare indicated on the axes; b) Plot of volume % albite as a proportion of total feldsparcontent shows the evolution of feldspar compositon over time for point C on Fig-ure 31a.
6.4. APPLICATION TO ALBITISATION 129
Reactive transport
The evolution of porosity and permeability with damage(see Sections 6.2.2 and 6.2.7)
along the created shear zone opens a fluid pathway with increases in permeabil-
ity by up to three orders of magnitude in the upper layer (see Figure 28). After
15,000 years the fault is fully reactivated and a zone of damage has begun prop-
agating into the upper layer (Figure 28a). Damage results in a porosity increase,
leading to a drop in pressure and fluid flow in the opposite direction of damage
propagation as a transient phenomenon at the tip of the propagating shear zone.
After 30,000 years the shear zone has reached the top boundary and opened a con-
duit through the entire model (Figure 28b), allowing hotter and more pressurised
fluid from the basement to flow through the upper layer, as illustrated by the tem-
perature contours in Figure 28. The simulation shows that after 30,000 years the
offset of the upper boundary has created an important topographical flow due to the
imposed pressure boundary condition. This causes fluids to be drawn downwards
as shown on the top of Figure 29 and therefore the simulation was terminated after
a simulated time of 32,600 years. At that time, the fluid flows downwards from
the top boundary across half the upper layer, and the upwards fluid across the rest
of the shear zones reaches Darcy velocities of 30cm/year in the lower part of the
upper layer.
Figure 30 shows the separate contributions of poro-elasticity and damage on the
total porosity evolution at the end of the simulation. This figure highlights the
major importance of damage in opening fluid pathways with porosity changes by
up to nearly 47 percentage points (Figure 30a). By comparison the contribution of
poro-elasticity is relatively minor with up to 0.3 percentage points porosity increase
in some dilation zones and 1.2 percentage points decrease in most of the shear
zone due to the compression boundary conditions. Those zones of dilation and
contraction do play a slight role on the transient evolution of fluid flow in the shear
zone, but their effect remains negligible compared to the effect of damage.
130 CHAPTER 6. REACTIVE TRANSPORT WITH DAMAGE MECHANICS
Due to the nature of the model and the initial chemical equilibrium fluid-rock
reactions are not initiated until the fault zone has been fully reactivated (after
10,000 years), allowing fluids to migrate from the albite-quartz basement into the
overlying K-feldspar-quartz unit. When the propagating fault reaches the bottom
of that top layer, the permeability at this location (point C on Figure 31b) jumps by
more than two orders of magnitudes. It reaches a maximum value of 10−13.83m2 after
11,800 years and stays at this level until the end of the simulation. The compression
boundary conditions also cause a slight reduction in permeability through the Biot
poro-elastic behaviour (see Section 6.2.3), but this effect is minor compared to the
damage impact and only reduces the permeability to 10−13.85m2 at the end of the
simulation. As the NaCl brine reacts with the K-feldspar progressive alteration to
albite is observed after 15,000 years (Figure 31b). Once started, the albitisation
process accelerates and reaches a constant rate after 28,000 years. At the end of the
simulation one can see a maximum of 12% albite formed (as a percentage of total
feldspar content), which would represent less than 1% porosity decrease based on
a 8% molar volume decrease for full albitisation. This result shows that chemical
porosity evolution, in this particular example, is indeed negligible compared to the
change of porosity attributed to damage. This validates the initial assumption not
to consider chemically induced porosity changes (equation (151)) in this application.
6.5 Conclusion
A new THMC simulation code is presented by linking escriptRT to Abaqus,
thus adding a thermodynamically consistent visco-elasto-plastic deformation im-
plementation to an existing reactive transport code. The mechanical model used
in this study includes a continuum damage mechanics formulation which is linked
to permeability through the evolution of porosity. This significant feedback allows
fluid pathways to be dynamically generated from the localisation of shear zones in
particular. The importance of this phenomenon is illustrated by a numerical study
6.5. CONCLUSION 131
of a generic albitisation scenario involving a pre-existing fault and an unconformity.
While a full geochemical study of this example is beyond the scope of this paper,
a significant conclusion from this study is that the coupling of THMC processes
and the evolution of permeability mechanism can allow to examine the progression
of geochemical reactions in comparison with the rates of other processes from the
fluid motion within the host rock. They can be used as well to predict the distribu-
tion of alteration assemblages which in term can provide vectors to mineralisation.
This information can also be used to recognise prospective geophysical responses in
mineral exploration (e.g., Chopping and Cleverley, 2008). The features and results
of the simulation presented may be applicable to a wide range of hydrothermal ore
deposits as they show that: (i) the inclusion of damage in the code produces narrow
discrete shear zone which match more closely those observed in nature, (ii) dam-
age mechanics reproduces the logarithmic pattern of rock pulverisation which can
be observed in nature on the sides of shear zones, (iii) permeability creation as a
function of damage greatly enhances flow rates and focusing, and (iv) elevated flow
rates can affect geochemical reactions by increasing the supply of reactants as well
as changing the temperature through heat advection. In this example the effects of
damage on permeability are shown to be significantly greater than those caused by
poro-elasticity. While the chemical/dissolution induced porosity changes were not
specifically addressed, the effects of damage are also interpreted to be significantly
greater in this particular example where the changes in mineralogy resulted in a
minor change in molar volume. Simulating scenarios where the fracturing of an im-
permeable seal promotes fluid flow from deep reservoirs into overlying sequences can
be used to examine processes in many mineral systems such as unconformity-related
uranium, Archean gold or others.
132 CHAPTER 6. REACTIVE TRANSPORT WITH DAMAGE MECHANICS
Chapter 7
Conclusions and perspectives
7.1 Conclusions
In this thesis, a novel numerical framework for THMC coupling has been formu-
lated, based fully on the thermodynamical potential functions. It distinguishes itself
from the many other THMC formulations in its emphasis on how mechanical de-
formation can create permeable pathways. Classical mechanics has previously been
used to close the system of equations to define the mechanical problem. This thesis
formulates an explicit fully coupled mechanical framework that considers entropy
production.
A particularly important aspect of the new approach presented consists of model-
ling explicitly the various feedbacks which couple all processes involved such as
mechanical deformation, heat transport, fluid flow, chemical transport, and fluid-
rock chemical reactions. Those feedbacks are indeed critical to explain geological
localisation phenomena such as folding or faulting, for example. Both the number of
processes considered as well as the quality of their coupling improve our geological
understanding. This thesis provides a unified framework based on thermodynamics
to tightly couple geological processes, as well as a scalable approach to combine
133
134 CHAPTER 7. CONCLUSIONS AND PERSPECTIVES
processes sequentially and include more feedbacks. Various couplings were investi-
gated successively and their importance was shown through rigorous benchmarks
and illustrative geological simulations. These couplings appear in two forms, di-
rect and indirect, as they can be directly explicited in the constitutive equations
or indirectly introduced through the material properties dependencies on the state
variables. For instance, the shear heating term in the heat equation represents a
direct feedback. An example of indirect coupling is the dependency of mechani-
cal properties on chemical compositions. The key to separate direct and indirect
coupling is the diffusion length and time scales of the feedback processes consid-
ered. For example, for geological problems at the kilometre scale and for times
of hundreds of thousands of years, chemical feedbacks can be incorporated to the
material properties through indirect coupling, while thermal feedbacks need to be
resolved explicitly. At those scales, most chemical processes can indeed be assumed
to have reached equilibrium. Therefore, temperature and pressure dependence of
these properties can be pre-calculated and considered as indirect couplings.
Chapter 2 demonstrates the importance of considering the link between thermo-
chemistry and mechanics, where thermodynamic potential functions can be used to
calculate reversible material properties such as thermal expansion coefficient, spe-