GOLEM

Solid Earth, 8, 921–941, 2017
https://doi.org/10.5194/se-8-921-2017
© Author(s) 2017. This work is distributed under
the Creative Commons Attribution 3.0 License.
Flexible parallel implicit modelling of coupled

thermal–hydraulic–mechanical processes in fractured rocks
Mauro Cacace1 and Antoine B. Jacquey1,2
1 Department 6: Geotechnologies, Section 6.1 Basin Modelling, GFZ – German Research Centre for Geosciences,
Potsdam, Germany
2 Department of Geology, Geochemistry of Petroleum and Coal, RWTH Aachen University, Aachen, Germany
Correspondence to: Mauro Cacace (mauro.cacace@gfz-potsdam.de)
Received: 23 March 2017 – Discussion started: 31 March 2017

Revised: 11 July 2017 – Accepted: 1 August 2017 – Published: 13 September 2017
Abstract. Theory and numerical implementation describing 1 Introduction

groundwater flow and the transport of heat and solute mass
in fully saturated fractured rocks with elasto-plastic mechan-
ical feedbacks are developed. In our formulation, fractures Reliable predictions of reservoirs’ performance, whether for
are considered as being of lower dimension than the hosting geothermal and fossil energy extraction, or water, CO2 , and
deformable porous rock and we consider their hydraulic and heat storage rely on an accurate representation of the physi-
mechanical apertures as scaling parameters to ensure contin- cal processes responsible for groundwater flow, heat and so-
uous exchange of fluid mass and energy within the fracture– lute mass transfer, mechanical deformation of the rock solid
solid matrix system. The coupled system of equations is skeleton, and, ultimately, chemical feedbacks from fluid–
implemented in a new simulator code that makes use of a rock interactions (Stephansson et al., 2004). All of these pro-
Galerkin finite-element technique. The code builds on a flex- cesses occur in a naturally complex geological setting, com-
ible, object-oriented numerical framework (MOOSE, Mul- prising discrete heterogeneities as faults and fractures, span
tiphysics Object Oriented Simulation Environment) which a relatively large spectrum of temporal and spatial scales and
provides an extensive scalable parallel and implicit coupling interact in a highly nonlinear fashion. In natural and engi-
to solve for the multiphysics problem. The governing equa- neered systems, onset conditions and the evolution in time
tions of groundwater flow, heat and mass transport, and rock and space of a particular process are affected by the initia-
deformation are solved in a weak sense (either by classi- tion and evolution of all the other processes. Therefore, mon-
cal Newton–Raphson or by free Jacobian inexact Newton– itored variations of rock masses to natural and anthropogenic
Krylow schemes) on an underlying unstructured mesh. Non- perturbations cannot be fully reconciled by considering the
linear feedbacks among the active processes are enforced by causative processes independently. This is particularly the
considering evolving fluid and rock properties depending on case for reservoir applications which require a complete un-
the thermo-hydro-mechanical state of the system and the lo- derstanding of the multicomponent (fractured) porous rock–
cal structure, i.e. degree of connectivity, of the fracture sys- fluid system and its multiphysics dynamics to predict the be-
tem. A suite of applications is presented to illustrate the flexi- haviour of a particular reservoir so as to improve its produc-
bility and capability of the new simulator to address problems tivity and sustainability.
of increasing complexity and occurring at different spatial To recognize that the majority of geological observations
(from centimetres to tens of kilometres) and temporal scales are the results of nonlinear interactions among classes of
(from minutes to hundreds of years). physical processes has justified the recent development and
massive use of so-called multiphysics numerical simula-
tors as complementary tools to classical experimental and
theoretical analyses. Numerical simulations are particularly
adapted to the study of a multicomponent physical system
Published by Copernicus Publications on behalf of the European Geosciences Union.

922 M. Cacace and A. B. Jacquey: Parallel modelling of THM processes
since they allow a systematic analysis of the dynamics of sin- problems even with relative large time step sizes (Kim et al.,
gle processes and their interactions (i) under proper time and 2015). In addition, recent advances in developing block-type
length scales, (ii) within complex geometries, and (iii) also preconditioners and “super-convergent” methods for systems
by considering changing loading conditions. In addition, of nonlinear equations at reduced memory consumption,
given their predictive capabilities, numerical modelling tech- e.g. preconditioned Jacobian free methods, make these ap-
niques can assist studies aiming at estimating performances proaches competitive for field-scale applications (Knoll and
and potential risks related to different operational scenarios Keyes, 2004).
(e.g. Blöcher et al., 2015). Despite all of this, only few attempts have been made
Numerical modelling of fractured reservoirs, whether so far in developing and implementing fully implicit nu-
petroleum or geothermal, has a long history dating back sev- merical solutions for thermal–hydraulic–mechanical (THM)
eral decades. Starting from the earlier works in the 1970s, problems.
a number of codes have been implemented under various li- The FRACON code of Nguyen and Selvadurai (1998)
cences by the modelling community (Jing, 2003; Stephans- considers a monolithic small strain and non-isothermal im-
son et al., 2004). The majority of these codes rely on mod- plementation, but thermal feedbacks effects on the skele-
elling techniques that were developed in the late 1980s and ton deformation and pore fluid pressure are only explicitly
early 1990s, and that can be grossly subdivided into two integrated. The open-source, object-oriented project Open-
classes if based on the approach followed to model coupled Geosys (Kolditz et al., 2012) provides a parallel platform for
processes. implicit solution of multiphysics problems (Blöcher et al.,
A first class of approaches, referred to as sequen- 2015) but so far, to the authors’ knowledge, has never been
tial/explicit coupling or operator splitting approaches, relies applied to 3-D THM problem in fractured reservoirs. More
on splitting the coupled physical problem into classes of pos- recently, Sun (2015) developed a monolithic framework for
sibly linear sub-problems and to numerically solve for each solving coupled THM processes at finite strain. However,
process sequentially. During each time step, coupling among such an approach has only been applied on generic and sim-
the processes is enforced by passing input/output data among plistic models and lacks a description of complex geological
the respective sub-problems. Usually, external iterations are geometries such as fractures and faults. More recent devel-
adopted for simulations characterized by a high degree of opments have been published focusing on quantifying chem-
nonlinearity (Kempka et al., 2016; Chabab and Kempka, ical feedbacks for tightly coupled problems though mainly
2016). In some cases, sequential coupling is achieved by targeting simplistic geometries (Poulet and Veveakis, 2016).
relying on different simulators, as for example by coupling The goal of this paper is not to summarize the state-of-
a flow simulator to a mechanical simulator for a thermo- the-art computational methods for problems that are relevant
hydro-mechanical problem (TOUGH-FLAC family of codes, for reservoir applications. Our interest herein is rather to-
e.g. Rutqvist, 2011). The main advantage of such approaches wards computational reliability and performances when sim-
stems from numerically integrating relatively complex prob- ulating the behaviour of a particular reservoir in a way that
lems within limited computer resources. However, sequential can be of help to improve scenario-oriented analysis of such
coupling schemes have important impacts on the efficiency, systems. In this context, we address issues related on how
stability, and accuracy of the numerical solutions. They usu- (i) to quantify the nonlinear feedbacks among the differ-
ally introduce a splitting error in the numerical approxima- ent physical processes, and (ii) to represent into a computa-
tion, which requires a careful monitoring of the nonlinear tional model the porous rock–fracture–fluid system by cap-
residuals within each single time step (Jha and Juanes, 2007; turing its discontinuous, anisotropic, inhomogeneous, and
Preisig and Prévost, 2011). This aspect limits sequential ap- non-elastic nature (Hudson and Harrison, 1997). For this
proaches to what are generally referred to as “loosely cou- purpose, we give an overview of the methods implemented
pled problems”, and they show a relatively slow, if at all, into a novel, yet robust and efficient multiphysics and mul-
convergence rate for tightly coupled problems. In addition, ticomponent porous media open-source modelling simula-
conditional stability of semi-implicit approaches impose se- tor called GOLEM which can deal with all these aspects.
vere time step restrictions thus increasing computation times. Our emphasis throughout this paper is to simulate THM pro-
As an alternative, it is possible to solve the full system of cesses of relevance for hydrothermal and geothermal sys-
coupled equations simultaneously, via a fully implicit cou- tems. Though GOLEM can also simulate the transport of
pling approach. This requires to solve for all the variables non-reactive chemical species, we do not discuss this aspect
of the problem simultaneously within an iterative approach, in the present paper. Consideration of additional chemical
either (in)exact Newton or simpler Picard methods, for the (fluid–solid) interactions is the subject of future work.
resulting algebraic system (Knoll and Keyes, 2004). Simulta- The remainder of the paper is structured as follows. In
neous solution schemes demand higher computer storage and Sect. 2, we introduce the constitutive model for THM pro-
processing times to compute (allocate the memory and fill) cesses for a two-phase system consisting of a deformable
the Jacobian of the problem than explicit approaches. How- solid skeleton and fully saturated pores in the presence of dis-
ever, they guarantee higher stability for strongly nonlinear crete heterogeneities as represented by faults and fractures.
Solid Earth, 8, 921–941, 2017 www.solid-earth.net/8/921/2017/

M. Cacace and A. B. Jacquey: Parallel modelling of THM processes 923
Attention is given to include the important nonlinear feed-

backs among the different processes as well as structure–
property relationships. These latter include links between
fluid properties (e.g. density and viscosity) and problem vari-
ables (e.g. pressure and temperature), as well as evolution of
material properties as a function of variations in the state of
the reservoirs (e.g. porosity and permeability relationships).
In Sect. 3, the numerical implementation of the derived con-
stitutive model is described. In Sect. 4, we apply the simula-
tor to a suite of applications of different level of complex-
ity (from synthetic benchmark cases to prototypical reser-
Figure 1. Schematic representation of the approach to model frac-
voir simulations) both in terms of the coupling among the ture rock masses as based on a superposition of lower- and higher-
processes considered as well as in terms of the geometry of dimensional geometric elements, after Cacace and Blöcher (2015):
the natural system. A field application related to modelling 0-D vertex (point-like sources/sinks), 1-D edge (well paths), 2-D
the hydromechanical response of a sandy reservoir during face (fractures an faults), and 3-D element (porous matrix).
a cyclic injection test will be presented in a separate paper
(Jacquey et al., 2017).
mass conservation equations for the fluid and solid phases.
Mass conservation for the liquid phase therefore requires
2 Mathematical formulation of the problem
∂ (nρf )
+ ∇ · (nρf v f ) = Qf , (1)
In this section, we present the constitutive model for a porous ∂t
medium consisting of a deformable hosting rock and a mo- where ρf is the density of the fluid phase, n is the porosity,
bile liquid saturating its pores. In the following we make use v f the fluid velocity, and Qf is a sink/source term considered
of subscripts f and s to refer to the solid and fluid phase null hereafter for the sake of simplicity. In a similar way, for
respectively. Deformation of the solid skeleton is consid- the solid phase we obtain
ered in terms of Biot consolidation theory (Biot, 1956; Biot
and Willis, 1957; Biot, 1973) under non-isothermal loading ∂ ((1 − n) ρs )
(Geertsma, 1957; McTigue, 1986) and its extension to take + ∇ · ((1 − n) ρs v s ) = Qs , (2)
∂t
into account time- and rate-dependent inelastic behaviour.
The model assumes the existence of a reference elemen- where ρs is the density of the solid phase, v s the solid ve-
tary volume (REV) for the porous medium in which the two locity, and Qs is a sink/source term (considered as null here-
components can interact thermo-hydro-mechanically (Bear, after). We use Darcy’s law to describe the conservation of
1988). The resulting physical system therefore incorporates momentum of the fluid phase, which can be expressed in
non-isothermal flow of the fluid phase (namely water) within terms of fluid velocity relative to the solid velocity as
a porous rock which is free to deform (in)elastically and
k
in the presence of discontinuities as represented by discrete q D = n (v f − v s ) = − · (∇pf − ρf g) , (3)
fractures (Boley and Weiner, 1960). µf
Fractures are considered as lower-dimensional elements where q D is a volumetric flow rate per unit of surface area
embedded in the porous matrix filled with fluid (see Fig. 1). (Darcy velocity), k is the permeability tensor of the porous
We consider the length of the discrete fracture as a repre- medium, µf the fluid viscosity, and g the gravity vector.
sentative measure of the REV of the porous matrix. This al- Substituting Eq. (3) into Eq. (1) yields
lows us to make use of a continuum approach to represent
the porous medium. We also take into account the effects of ∂ (nρf )
micro-defects (fissures and micro-cracks) on the thermal, hy- + ∇ · ρf q D + ∇ · (nρf v s ) = 0. (4)
∂t
draulic, and mechanical behaviour of the porous rock as they
mainly affect the evolution of the system material properties The equations of mass conservations can be rewritten by
(e.g. porosity; see Fig. 2). applying the concept of the Lagrangian (total) derivative with
s
In the following the main governing equations are derived. respect to a moving solid, e.g. DDt(•) ≡ ∂(•)
∂t + ∇(•) · v s , and
D f (•)
The problem variables considered are the pore fluid pressure
Dt ≡ ∂(•)
∂t + ∇(•) · v f for a moving fluid. By expanding the
pf , the temperature T , and the solid displacement vector u. fluid mass conservation equation and noting that ∇ ·[(•)v f ] =
Pore pressure is defined as compression positive for water, (•)∇ · v f + ∇(•) · v f , Eq. (4) can be rewritten as
while stress is defined as tension positive for the solid phase.
The mass balance equation for a deformable, saturated n D f ρf D s n
porous medium is described in terms of volumetric averaged + + n∇ · v s + ∇ · q D = 0. (5)
ρf Dt Dt
www.solid-earth.net/8/921/2017/ Solid Earth, 8, 921–941, 2017

Figure 2. Numerical simulations of tri-axial mechanical experiments on a Flechtinger sandstone. (a) Evolution of the mean effective stress
with respect to the volumetric strain. (b) Evolution of the normalized porosity with respect to the mean effective stress. The nonlinear
behaviour at low effective stress as observed in the experimental data (red line after Blöcher et al., 2014) is well captured by the modelling
results which take into account a nonlinear evolution of the elastic moduli resulting from closure of pre-existing micro-cracks and fissures
enhanced by compaction of the porous space followed by stiffening of the rock (increase Hertzian contact surface between solid grains).
The modelling results can also capture the compaction-induced reduction of porosity when compared to experimental data (in red) within
experimental errors (red shaded area). More details on the theory and results can be found in Jacquey et al. (2015).
In a similar way, it is possible to rework the solid mass The second term in the left-hand side of Eq. (8) can also
balance equation (Eq. 2) to obtain be cast in terms of the problem variables, e.g. pore pressure,
temperature, and solid skeleton displacements, by defining
(1 − n) D s ρs D s n
− + (1 − n) ∇ · v s = 0. (6) a proper constitutive mechanical model. The linear momen-
ρs Dt Dt tum balance equation of the mixture in terms of the effective
From Eq. (6), it can be noticed that even by considering Cauchy stress tensor σ 0 (x, t) takes the form
both the solid skeleton and the pore fluid to be incompress-
∇ · σ 0 − αpf 1 + ρb g = 0,

ible, the porous rock material will deform (contract or dilate) (10)
when fluid is expelled from or injected into the pore space.
Equation (6) can be used to express the evolution of the where 1 is the rank-two identity tensor, ρb is the bulk den-
porosity in terms of the Lagrangian derivative with respect to sity of the fluid–solid mixture (ρb = nρf + (1 − n)ρs ), and
the solid deformation velocity as α = 1 − KKs is the Biot coefficient, with K being the drain
bulk modulus and Ks the bulk modulus of the solid grains.
D s n (1 − n) D s ρs The geometrical compatibility condition gives the following
= + (1 − n) ∇ · v s . (7)
Dt ρs Dt strain–displacement relation:
Substituting Eq. (7) into Eq. (5), one obtains 1
= ∇u + ∇ T u = ∇ s u. (11)
n D f ρf (1 − n) D s ρs 2
+ + ∇ · v s + ∇ · q D = 0. (8)
ρf Dt ρs Dt Deformation of the solid skeleton is described in terms
The first term in the left-hand side of Eq. (8) can be ex- of thermo-poroelastic response (Biot’s consolidation theory)
pressed in terms of the fluid pore pressure and temperature and dissipative plastic behaviour. To simplify the presenta-
by thermodynamic differentiation as tion of the constitutive mechanical model, in the following
we will consider only small strain conditions, but the theory
D f ρf 1 D f pf DfT

has also been extended to account for finite deformation. In
=n − βf , (9)
Dt Kf Dt Dt addition, due to strain history dependence, we detail the for-
mulation in incremental form. Following Biot’s theory, (ef-
∂ρf
where K1f = ρ1f ∂p is the inverse of the fluid bulk modu- fective) stresses are related to elastic strains via the following
f T relationship:
lus and βf = − ρ1f ∂ρ
∂T
f
the fluid volumetric thermal expan-
pf
e
sion coefficient. σ̇ 0 = σ̇ij0 = Cij kl ˙kl = C : ˙ e , (12)

where C = Cij kl = λδij δkl +2Gδik δj l is the rank-four elastic To quantify any permanent (irreversible) deformation of
stiffness tensor, with λ and G being the first (volumetric) and the material due to inelastic processes we make use of the
second (shear) Lamé moduli respectively. concept of eigenstrain ( ∗ ) as derived from micromechanics
The stress–strain constitutive relation given by Eq. (12), (Mura, 1987). By assuming small strain approximation, the
can then be used to find an expression for the material deriva- total strain of the material can be decomposed as the sum
tive of the solid density in terms of the problem variables of the elastic strain ( e ) and eigenstrain components ( ∗ ) as
(second term in Eq. 8) as
= e + ∗. (17)
(1 − n) Dsρs (α − n) Dsp
f
= Therefore, Eq. (16) can be written as
ρs Dt Ks Dt
DsT 1 Dsσ 0 1 ∂pf ∂T ∗
− (1 − n)βs − , (13) − βb + α ˙kk + (1 − α) ˙kk + ∇ · qD
Dt Ks Dt Mb ∂t ∂t

1
where σ 0 indicates the mean effective stress and βs the volu- + qD · ∇pf − βf ∇T
metric thermal expansion coefficient of the solid grains. Kf

Substituting Eqs. (9) and (13) into Eq. (8), we obtain 1 1
+ vs · ∇pf − βb ∇T − ∇σ 0 = 0. (18)
Mb Ks
n Dfp f DfT (α − n) fDsp
− nβf + In the present study we focus on two major kinds of
Kf Dt Dt Ks Dt
DsT 1 Dsσ 0 residual deformation, that is, thermal expansion and plastic
∗p
− (1 − n)βs − + ˙kk + ∇ · q D = 0, (14) flow, i.e. ∗ = ij∗ = ij∗T +ij = ∗T + ∗p , though additional
Dt Ks Dt processes including for example swelling, fatigue, or phase
where we have expressed the gradient of the solid deforma- transformations can be relatively easily integrated in the cur-
tion velocity in terms of the volumetric component of the rent formulation.
total stain rate tensor, ∇ · v s = ∇ · u̇ = ˙kk . Thermal strains are related to deformation induced by tem-
By making use of the definition of the total derivative perature changes inside the material, and can be therefore ex-
and given the stress–strain constitutive equation (Eq. 12), pressed by
Eq. (14) can be recast as
1
˙ ∗T = βb Ṫ 1, (19)
1 ∂pf ∂T e 3
− βb − (1 − α)˙kk + ˙kk + ∇ · q D
Mb ∂t ∂t where Ṫ is the relative temperature rate.
n (α − n) We determine the plastic component of the strain tensor
+ ∇pf · v f + ∇pf · v s
Kf Ks (Eq. 17) by making use of the normality rule as
− (nβf ∇T · v f + (1 − n)βs ∇T · v s )
∂Q
1 ˙ ∗p = γ̇ , (20)
− ∇σ 0 · v s = 0, (15) ∂σ 0
Ks
where γ̇ = γ̇ (σ 0 , κ) is the plastic multiplier satisfying the
where M1b = Knf + (α−n)
Ks is the specific storage of the porous classical Kuhn–Tucker conditions (γ̇ ≥ 0, F ≤ 0, γ̇ F = 0)
medium (reciprocal of the Biot modulus Mb ) and βb = nβf + with F(σ 0 , κ) being the yield surface. Q = Q(σ 0 , κ) is the
(1 − n)βs is the bulk volumetric thermal expansion coeffi- plastic potential function giving the direction of the plastic
cient. strain increment, and κ represents the vector of internal vari-
It is possible to rewrite Eq. (15) in terms of the solid ve- ables affecting the evolution of the yield surface during load-
locity only, by integrating the momentum balance equation ing and unloading of the material.
(Eq. 3) as Therefore, Eq. (18) can be finally written for thermal and
plastic eigenstrains as
1 ∂pf ∂T e
− βb − (1 − α)˙kk + ˙kk + ∇ · q D
Mb ∂t ∂t 1 ∂pf ∂T ∗p
− αβb + α ˙kk + (1 − α) ˙kk + ∇ · q D
1 Mb ∂t ∂t
+ qD · ∇pf − βf ∇T
Kf 1
+ qD · ∇pf − βf ∇T
Kf

1 1 0
+ vs · ∇pf − βb ∇T − ∇σ = 0, (16)
Mb Ks 1 1 0
+ vs · ∇pf − βb ∇T − ∇σ = 0. (21)
Mb Ks
where the last two terms can be considered as second-order
correction terms taking into account nonlinear advective ef- Similarly, the evolution of porosity can therefore be ex-
fects. pressed using Eqs. (7) and (13) as well as the strain decom-

position as of interconnected issues must be taken into account. These

∂n (α − n) ∂pf ∂T ∗p
include for example the choice of the spatial discretization
= − βn + (α − n) ˙kk + (1 − α) ˙kk + v s adopted, the time stepping scheme and temporal integration,
∂t Ks ∂t ∂t
iterative solvers, and preconditioners. In what follows, we
(α − n) 1
· ∇pf − (1 − n) βs ∇T − ∇σ 0 − ∇n , (22) describe the methods adopted in GOLEM in order to tackle
Ks Ks these issues. In the next section, we also present some nu-
where βn = (1 − n) βs − (1 − α) βb is the volumetric thermal merical applications that serve as code benchmarking and il-
expansion coefficient of the pores and the last term can be lustrate the capability of the simulator to solve for problems
considered as being of second order. of different degree of difficulty at optimal cost.
The balance of energy for the fluid–rock mixture, assum- GOLEM is an open-source simulator specifically devel-
ing local thermal equilibrium between the two phases, and oped for thermo-hydro-mechanical coupled applications in
neglecting the dissipation of mechanical energy due to de- fractured geological systems, supporting 1-D, 2-D, and 3-
formation of the solid phase reads as D computations in a single code implementation. It builds
∂(ρc)b ∂T on the object-oriented numerical framework MOOSE devel-
T +(ρc)b +∇ · ρf cf q D T − λb ∇T − Ḣ = 0, (23) oped at the Idaho National Laboratory (Gaston et al., 2009).
∂t ∂t
MOOSE provides a flexible, massive parallel (including both
where (ρc)b = nρf cf + (1 − n)ρs cs is the bulk specific heat
MPI and multithreading) platform to solve for multiphysics
of the porous medium, λb = nλf + (1 − n)λs is the bulk ther-
and multicomponent problems in an implicit manner. It re-
mal conductivity, and Ḣ is a rate of energy production. The
lies on state-of-the-art and extensively tested libraries devel-
first term of the left-hand side of Eq. (23) takes into account
oped both at universities and national laboratories, as the
secondary, non-Boussinesq dissipative effects related to the
libMesh library (Kirk et al., 2006) developed at the Uni-
pressure temperature dependency of the bulk storage, which
versity of Austin in Texas for capabilities related to par-
are usually rewritten only considering variable fluid and solid
allel finite-element method, and a suite of scalable paral-
density as
lel nonlinear and linear solvers (PETSc: Balay et al., 2016;
∂(ρc)b ρs ∂pf ∂T Trilinos: Heroux et al., 2005; Hypre: Chow et al., 1998).
T = (1 − n) cs T − ρ s βs
∂t Ks ∂t ∂t The use of different open-source libraries provides a mod-

ρf ∂pf ∂T
ular structure to the framework, which allows the developer
+ ncf T − ρ f βf . (24) of an application to only concentrate on the high-level de-
Kf ∂t ∂t
scription of the multiphysics problem and to maintain the
The use of a conservative finite-element formulation in code relatively compact. Following the basic structure of
Eq. (23) ensures the balance of the fluid enthalpy (ρf cf T ) the MOOSE framework, GOLEM has also been developed
both at element and node levels. Therefore, it guarantees that as a modular application, where each module is responsi-
we avoid accumulating in time unbalances in intercell fluxes ble for the solution of a specific physical process. This as-
as is usually the case when relying on non-conservative pect makes easy any further modification, adjustment, and
formulations for convective-type problems (Giudice et al., improvement of the program with limited efforts from the
1992). user’s side. At the present stage, Golem is hosted and will
The energy conservation equation as given by Eq. (23) is be made available for download from a GitHub repository
valid for a porous medium in the absence of any thermoelas- (https://github.com/ajacquey/Golem) and it comes together
tic coupling. Following Biot consolidation theory, it is possi- with a suite of relatively simple benchmark problems and all
ble to consider the effects of the solid elastic deformation on input files for running the applications as presented in the
the temperature distribution by augmenting Eq. (23) with a paper.
thermoelastic dissipation rate term, as
∂(ρc)b ∂T e 3.1 Variational formulation and its numerical solution
T + (ρc)b + T0 βb ˙kk
∂t ∂t The finite-element discretization is based on the weak form,
+ ∇ · ρf cf q D T − λb ∇T − Ḣ = 0, (25)
in an integral sense, of the system of partial differential equa-
where T0 represents the absolute temperature of the porous tions as derived in the previous paragraph. For this purpose,
medium in a stress-free state. Equation (25) can be easily we consider the porous matrix to be described by a close do-
modified to take into account additional thermal effects from main of volume ⊂ <n bounded by a boundary 0 ⊂ <(n−1) .
fluid dilation and shear heating stresses. Given the length/width ratio typical of fractures, a discrete
fracture is represented by a lower-dimensional element of
volume f ⊂ <(n−1) and surface area 0f ⊂ <(n−2) . The cor-
3 Numerical implementation
responding weak form of the governing equations are then
In order to solve the system of coupled and nonlinear equa- derived by applying the method of the weighted residuals as
tions as described above (Eqs. 10, 21, and 25), a number

– Continuity of the total (diffusive plus advective) heat

Z
1 ∂pf
Z
∂T flow, i.e. ((ρc)f q D T −λb ∇T )·n0qT = qT on 0qT , where
ω d − ωαβb d qT is the rate of in/out heat flow per unit of area across
Mb ∂t ∂t
the boundary surface.
Z
∗p The resulting system of equations together with a proper
+ ω α ˙kk + (1 − α)˙kk d

choice of boundary and initial conditions is discretized spa-
Z Z tially by the finite-element method, while the temporal dis-
− ∇ω · q D d + ω q D · n0qH d0 cretization is done by traditional finite difference techniques.
0q H The nodal values of the primary variables of the problem,
Z temperature (T n ), pore pressure (pfn ), and the deformation
1
+ ωq D · ∇pf − βf ∇T d vector of the solid skeleton (un ) are approximated by linear
Kf Lagrangian interpolation polynomial functions as

Z
1 1 0 i=N
XT
+ ωv s · ∇pf − βb ∇T − ∇σ d = 0, (26)
Mb Ks Tn = Tin wiT (x)
i=1
i=Npf
p
X
pfn = pf ni wi f (x)
Z
∂T (29)
ω(ρc)b d
∂t i=1
i=N
Xu
un = uni w ui (x).
Z

− ∇ω · (ρc)f q D T − λb ∇T d
i=1

Though higher-order polynomials are also available
Z

+ ω (ρc)f q D T − λb ∇T · n0qT d0 through the libMesh library, we rely on linear finite-element
0q T approximation for all variables. Indeed, we have found that,
Z for the kinds of problems as those that will be presented in
+ ωḢ d = 0, (27) the next paragraph, higher-order approximations do not nec-
essarily guarantee a better convergence of the solution, being
subjected at the same time to severe under- or overshooting
Z Z numerical effects in the presence of sharp gradients.
∇ s ω : σ 0 − αpf 1 d −

ω · ρb gd Single-noded, zero-thickness finite elements are used to
explicitly represent individual fractures, the latter assumed
to be clean and fully saturated (i.e. n = 1 and Mb = Kf ). It
Z
ω · σ 0 − αpf 1 · n0qM d0 = 0,

− (28) follows that fractures can be parameterized as based on the
0 concept of their effective aperture, which provides a quanti-
tative measure of the geometric width of the fracture surface.
where in Eqs. (26), (27), and (28) we have omitted all supra-
The use of single-noded finite elements imposes continuity
(sub)scripting for easiness in the notation.
of gradients in both pore pressure and temperature across the
The equations derived above describe an initial and bound-
fracture width, as well as the absence of any shear and nor-
ary value problem, for which proper boundary and initial
mal strain acting on the fracture plane. While the latter as-
conditions need to be assigned. These can be set by prescrib-
sumption simplifies the problem formulation, it prevents us
ing either the value of the problem variables along or their
from including the effects of local fracture mechanics in the
flux across a portion of the boundary. More precisely, the
current formulation. Therefore, fractures are only considered
model discussed in the previous paragraph must satisfy the
as having a distinct hydraulic and thermal behaviour with re-
following set of boundary conditions:
spect to the porous domain. To extend the current formula-
– Prescribed displacement, pore pressure and tempera- tion to include mechanical feedback effects from localized
ture, i.e. u = u, pf = pf , T = T on 0u , 0pf , 0T respec- deformation, nucleation, and fracture propagation is part of
tively. ongoing activities.
– Equilibrium of boundary stresses and external loads
(last two terms in Eq. 28).
– Continuity of the fluid flux across the imposed bound-
ary, i.e. q D ·n0qH = qH on ∂qH , where qH is the rate of
in/out flow per unit of area across the boundary surface.

The mass balance equation for a discrete fracture then

reads as Z
b∇fT ω · cf ρf q D T − λf ∇f T df

b b −
Z Z 2 Z Z 2
1 ∂pf 0 f
ω dz df − ∇ω · q D dz0 df Z
Kf ∂t +

bω cf ρf q D T − λf ∇f T · n̂dSf = 0, (33)
f − b f− b
2 2
b ∂f
Z Z2
+ ωq D · n̂dz0 d0 + 3+ −
H + 3H = 0, (30) where we have made no distinction between the effective hy-
draulic aperture and the mechanical aperture (bh = bm = b =
0f − b R b2
2 0
b dz ).
−2
where 3± H are leakage terms (weak form) across each of the Derivatives of the test functions and of direction-
two sides of the fracture surface into the surrounding porous dependent material properties with respect to the system
domain and b is the aperture of the fracture. of global coordinates are computed by standard coordinate
In a similar manner, the energy balance equation for a frac- −1 ∂Nj
transformation, i.e. ∂N i
∂xi = Jij ∂x 0 , where Jij is the Jacobian
ture element reads as j
matrix of the mapping between global (xi ) and local (xi0 )
b
Z Z2 coordinates. Transformation in local coordinates for lower-
∂T dimensional elements
ω cf ρf + H dz0 df  is achieved by computing the rota-
cos x 0 , x cos x 0 , y cos x 0 , y

∂t
f − b
tional matrix, R =  cos y 0 , x cos y 0 , y cos y 0 , y ,
 
2
b
cos z0 , x cos z0 , y cos z0 , y

Z Z2
with cos xi0 , xi being the directional cosines. Coordinate

∇fT ω · cf ρf q D T − λf ∇f T dz0 df

−
f − b
transformation is applied to all direction-dependent (i.e. ten-
2 sorial) material properties as well as to the directional deriva-
b
Z Z2 tives as
ω cf ρf q D T − λf ∇f T · n̂dz0 df

+
N = RI f N0 I Tf RT , (34)
∂− b
2
+ 3+ − where N and N0 are direction-dependent material properties

T + 3T = 0, (31)
in global and local coordinates respectively, and I f is the unit
where 3± tangent vector of local coordinates.
T quantify the amount of heat leaking from the frac-
ture surfaces into the porous medium domain. The above system of coupled equation can be rewritten in
In Eqs. (30) and (31), integration is done in local coordi- a more concise form as
nates of the fracture tangential and normal directions respec- dx
tively (x 0 , y 0 , z0 ). This enables us to superimpose the dis- S + Kx − F = R(û), (35)
dt
cretized conservative equations for the porous medium and
the fractures at the nodal location shared by the two ele- where S is the nodal coefficient storage matrix, K denotes
ments, where the fluid mass and heat fluxes also cancelled the stiffness matrix of the problem, F is the load vec-
out. Therefore the weak form of the conservation equations tor, x is the vector of the problems variables, and R(û)
can be simplified as is the residual from the discrete approximation. We make
Z Z use of an unconditionally stable, backward Euler method
1 ∂pf
bω df − b∇f ωT · q D df to integrate Eq. (35) in time, and to arrive at the mono-
Kf ∂t lithic form of the coupled system. The monolithic form of
f f
Z Eq. (35) is then solved iteratively by the classical Newton
+ bωq D · n̂dSf = 0 (32) method, where different linear solvers (e.g. GMRES and its
∂f flexible variant, FGMRES) and preconditioners (including
also Newton–Krylov–Schwartz domain decomposition tech-
and niques) are available from the open source algebraic libraries.
Z
∂T 3.2 Nonlinearities and their stabilization
bω cf ρf + H df
∂t
f
The coefficient matrices in Eq. (35) contain off-diagonal
components, thus giving rise to a highly nonlinear problem

Figure 3. Example of a nonlinear convective flow problem after Elder (1967). A homogeneous, isotropic, and saturated porous medium is
heated from below with fluid density and viscosity being a function of temperature and pressure (IAPWS, 2008a, b). Due to the temporal
evolution in fluid property gradients, thermal buoyant forces develop thus giving rise to an unstable convective flow regime as illustrated by
the temperature isolines. H-mesh adaptive refinement with local error estimate based on the L-2 norm projection of the intercell thermal flux
vectors and a two-refinement cycle per time step has been enforced in order to guarantee numerical stability even in the presence of sharp
thermal gradients (see close-up view).
to be solved for. Nonlinearities arise due the nonlinear con- ω∗ , as

stitutive relationship between problem variables and material
ω∗ = ω + p, (36)
properties, as well as to the presence of internal feedback ef-
fects among the different processes. The constitutive relation where ω is the continuous weighting function and p = τ ∇ω ·
linking structure, properties, and primary variables consid- ∇qD kqD k is a discontinuous streamline upwind correction.
ered include equation of states for fluid density, thermal com- Note that by its definition, ω∗ is no longer continuous across
pressibility and expansivity, and fluid viscosity with respect inter-element boundaries. A critical question is related to the
to the pf − T state of the system (IAPWS, 2008a, b), as well choice of the upwind parameter τ , which might influence the
as porosity and permeability relations as a function of vari- stability and accuracy of the discrete solution. In all simu-
ation in the thermal, mechanical, and hydraulic state of the lations presented in the paper, when needed, we have made
system; see Figs. 3 and 2 respectively. used of the formula as presented in Galeão et al. (2004):
For problems involving heat transport, if the convective
h 1
term dominates over the diffusive term, the solution of the en- τ= coth(Pe ) − , (37)
ergy transport equation typically contains interior and bound- 2kqD kpp Pe
ary layers and its solution by the Galerkin finite-element where h is the diameter of the finite element adopted along
technique is usually globally unstable (Nield and Bejan, the direction of qD , pp is the order of approximation con-
2012). Numerical instabilities which take the form of spu- sidered in the interpolation, k · k is the Euclidean norm, and
rious oscillations around sharp gradients and boundary lay- Pe = kq2kp
D kh
is the local Peclet number. A major disadvan-
ers and likely lead to nonconvergent solutions have been ob- tage of the SUPG method is to add numerical diffusion to
served to occur independent of the level of mesh refinement all elements characterized by a high Peclet number, even in
(Diersch and Kolditz, 1998). In such situations, in order to the absence of sharp thermal gradients. We leave as a sub-
enhance the stability and accuracy of the finite-element so- ject of future studies the implementation of more recent and
lution, some sort of stabilization has to be added to the dis- more sophisticated damping methods, e.g. entropy viscosity
crete formulation. In GOLEM, we have opted for the stream- method as described in Guermond et al. (2011).
line upwind Petro–Galerkin (SUPG) method, which can be
conceived as adding an upwind perturbation along computed 3.3 Plasticity and return-map algorithm
streamlines to the standard Galerkin formulation (Brooks and
Hughes, 1982). The SUPG discretization of Eq. (27) is ob- We consider plastic deformation in terms of a Drucker–
tained by making use of an “enriched” weighting function, Prager frictional plastic yield criterion (Drucker, 1950) in
which the onset of yield is a function of the first and sec-
0 , J =
ond invariants of the effective stress tensor J1 = σkk 2

1
respectively, where τij is the deviatoric part of the
2 (τij τij ) where H is the hardening flow rule for the internal pa-
effective stress tensor (τij = σij0 − 13 J1 δij ): rameter κ, that is, κ̇ = −γ̇ H.
q
sin(ϕ) 2. This system of equations is then linearized as follows:
F= J2 + 02 + J1 − C cos(ϕ), (38)
3 
(k) (k) 0(k) (k) (k)
where ϕ and C are the Mohr–Coulomb friction angle and co-  R ,n+1 + ∂σ 0 R ,n+1 : 1σ n+1 + ∂κ R ,n+1 : 1κ n+1 =0


(k) (k) 0(k) (k) (k)
hesion respectively and 0 is a small non-hardening parame- R κ,n+1 + ∂σ 0 R κ,n+1 : 1σ n+1 + ∂κ R κ,n+1 : 1κ n+1 = 0,
 F (k) + ∂ 0 F (k) : 1σ 0(k) + ∂ F (k) : 1κ (k)

ter here introduced to relax the singularity at the cone’s tip of

n+1 σ n+1 n+1 κ n+1 n+1 =0
the Drucker–Prager yield envelope. An important aspect re-
lates to the form of the plastic potential function adopted. It which can be expressed as the following matrix system:
has been shown that associated flow rules might lead to over-
estimating the dilation of the rocks generally resulting in an
Jx = R,
overly weaker response of the rock to loading (Vermeer and
de Borst, 1984; Jiang and Xie, 2011). We avoid these issues where J is the Jacobian matrix
by implementing a non-associated form of Drucker–Prager,
in which the plastic potential function, G, is considered to 
(k) (k) (k)

k
C−1 + 1γn+1 ∂ 2 0 0 Gn+1 1γn+1 ∂ 2 0 Gn+1 (k) ∂σ 0 Gn+1
depend on the dilation angle, ψ as  σ σ σ κ 

J= (k) (k) (k) (k) (k)  ,
1γn+1 ∂σ 0 Hn+1 1 + 1γn+1 ∂κ Hn+1 Hn+1 
p sin(ψ)  
G= J2 + J1 (39)
 
(k) (k)
∂σ 0 Fn+1 ∂κ Fn+1 0
3
and, in consequence, the unnormalized flow directions can x is the vector of unknowns to compute,
be derived as
∂G 1 ∂J2 sin(ψ) 
0(k)

0 = √ 0 + δij . (40) 1σ n+1
∂σ ij 2 J2 ∂σ ij 3  
x =  1κ (k) ,
 
n+1
Note that in our formulation we take also into account  
(k)
possible degradation of the strength of the rock subjected to 12 γn+1
loading in terms of hardening (and softening) of the internal
parameters (friction angle, cohesion and dilation angle) as a and R is the residuals vector:
function of the accumulated plastic strain (internal variable
κ). 
(k)

−R ,n+1
The stress update procedure is conducted via a return-map  
algorithm based on the closest point projection on the yield R =  −R (k) .
 
κ,n+1 
surface (Simo and Hughes, 1998) within a Newton–Raphson 
(k)

procedure. This algorithm is presented in incremental form in −Fn+1
the following. Subscript n refers to a value at the time tn , that
3. The aforementioned matrix system of equation is solved
is, σn = σ (tn ) and superscript (k) refers to the kth iteration in
at each kth iteration using routines from the PETSc li-
the Newton–Raphson procedure. We use the following nota-
∂· brary (Balay et al., 2016) for the increment of stress,
tion for sake of simplicity: ∂σ 0 · = ∂σ 0. 0(k)
internal parameter, and plastic multiplier (1σ n+1 ,
trial > 0), then the increment
1. If plastic loading occurs (Fn+1 (k) (k)
1κ n+1 , and 12 γn+1 ).
of plastic multiplier is also positive according to the
Kuhn–Tucker conditions, 1γ > 0. Define the system 4. The variables are updated at the end of the k th iteration:
of equations with the residuals to minimize, the plastic 0(k+1) 0(k) 0(k)
σ n+1 = σ n+1 + 1σ n+1
flow residual R ,n+1 , the internal parameter residuals
∗p(k+1) ∗p(k) 0(k)
R κ,n+1 and the yield condition for this time step Fn+1 n+1 = n+1 − C−1 : 1σ n+1
as (k+1) (k) (k)
1γn+1 = 1γn+1 + 12 γn+1 .

R ,n+1
∗p ∗p
= − n+1 + n + 1γ ∂σ 0 G 5. Steps 1 to 4 are repeated until the residuals reached min-
imum threshold values.

 ∗p
= −1 n+1 + 1γ ∂σ 0 G




We have tested and validated the above-described return-

R κ,n+1 = κ n+1 + κ n − 1γ H
map algorithm to update the elasto-plastic deformation
= 1κ n+1 − 1γ H


against available algorithms in the MOOSE tensor mechan-




= F σ 0n+1 , κ n+1 ,

Fn+1 ics module.


4 Results Table 1. Fluid properties for the example of heat transport in a frac-
ture.
In this section we present five different applications of the nu-
merical simulator. These applications are intended to test the Property name Symbol Value SI unit
ability of the simulator to deal both with single processes and Fluid density ρf 1000 kg m−3
their coupling. By starting with simplistic benchmarks, for Fluid thermal conductivity λf 0.65 W m−1 K−1
which analytical solutions exist, we gradually increase the Fluid heat capacity cf 4000 J kg−1 K−1
complexity of the problem formulation in order to demon- Fluid viscosity µf 1.0 × 10−3 Pa s
strate the applicability of the approach to realistic operational Fluid permeability k 1.0 × 10−10 m2
cases.
4.1 Benchmarks for coupled processes

tained results, we can conclude that the TH numerical imple-
4.1.1 Thermal and hydraulic processes – heat mentation is accurate for practical applications.
transport in a fracture
4.1.2 Hydraulic processes in the presence of discrete
The first application deals with groundwater flow and (dif-
fractures – flow in a fractured porous medium
fusive and advective) heat transport in a fracture, i.e. a TH
application. We assume a fracture which is fully saturated
with water (n = 1), having homogeneous and isotropic prop- The purpose of this application is to test the numerical
erties. Groundwater flow in the fracture is assumed to be implementation of the formulation in the presence of dis-
unidirectional and the average velocity is considered concrete fractures. At this purpose, we refer to a relative com-
stant throughout the length of the flow field. The initial tem- mon benchmark dealing with uniform, steady-state flow in a
perature is set to zero. At time t = 0, a sudden increase porous medium locally disturbed by the presence of a frac-
in temperature is imposed along the inlet boundary of the ture (Strack, 1982; Watanabe, 2011). The original problem
medium (T = T0 ). The problem can be formulated as a semi- formulation considers a semi-infinite two-dimensional hor-
infinite medium with a point source at the inlet boundary izontal section with an embedded fracture located symmet-
(T = T0 H (x = 0, t), with H (t) being the Heaviside func- rically at its centre. Uniform flow is maintained by impos-
tion). Under these assumptions, Ogata and Banks (1961) ing a specific discharge (q0 ) from the left boundary inside
gave an analytical solution for the variation of the temper- the domain, the value of which is kept constant and equal to
ature as a function of the position and time as q0 = 1.0 × 10−4 m s−1 . The fracture is considered to extend
infinitely along the direction normal to the plane, while being
T0 x − vx t of finite along-plane length (L), with its middle point located
T (x, t) = erfc √
2 4kt exactly at the centre of the domain. It has a width which can
T0 v x
x

x + vx t
be varied along its length (though it is assumed to remain
+ exp erfc √ . (41) constant in the following), and it is inclined with respect to
2 k 4kt
the model boundary by a constant angle (α).
For the numerical simulation, we consider a fracture of We have extended the original formulation to a three-
length L = 100 m, which has been discretized into 1000 line dimensional case; see Fig. 5 for the model geometry and
elements of equal length and subjected to an imposed pres- boundary conditions. The setup of the problem comprises
sure gradient of ∇pf = 3 Pa m−1 . All material properties are a three-dimensional, quadrilateral box (10 × 10 × 1 m in the
listed in Table 1. x, y, and z direction respectively) representing the porous
Given the parameters considered, a constant fluid veloc- medium. The fracture is implemented as a two-dimensional
ity (vx = 3.0 × 10−7 m s−1 ) is obtained. Initial conditions for surface cutting entirely the model along the vertical and hav-
the pressure and temperature are pf = 0.1 MPa and T = 0 ◦ C ing a finite horizontal length of L = 2 m. It is inclined by
respectively. In this benchmark, we compare the evolution an angle α = 45◦ with respect to the model boundaries. In
of the temperature field (normalized by the inlet tempera- order to maintain a constant discharge from the left to the
ture value) at four different points along the length of the right of the model, constant pressure boundary conditions
fracture with the analytical solution as given by Eq. (41). are imposed along the same boundaries thus resulting in a
Figure 4 shows the comparison between model results (red constant pressure gradient along the x axis. The value of the
curves) and the corresponding analytical solutions (black cir- imposed pressure gradient (1p = 1 MPa) has been enforced
cles) during the entire simulation time. There is a general so to match the value of specific discharge of the original
good fit in terms of the temporal evolution of the advected problem as derived by Strack (1982), thus permitting a direct
front at all locations along the fracture plane, with the mod- comparison between the analytical and numerical solution.
elled thermal front moving slightly faster as visible from the No flow conditions are imposed along the other boundaries.
earlier stepping at the observation points. Based on the ob- We assume a laminar flow in the fracture plane, and pressure

Figure 4. Comparison of numerical results (red curves) and analytical solutions (black circles) for heat transport in a fracture. The figure
shows the temperature evolution at different positions along the fracture as a function of time.
Table 2. Material properties the example of flow in a fractured

porous medium.
Property name Symbol Value SI unit

Fracture Angle α 45 ◦
Length L 2 m
Aperture h 0.05 m
Permeability kf 1.0 × 10−10 m2
f
Viscosity µf 1.0 × 10−3 Pa s
Figure 5. Geometry and boundary conditions for the benchmark Porous Porosity n 0.15 –
case of groundwater flow in a fractured porous medium. medium Permeability k 1.0 × 10−12 m2
Viscosity µf 1.0 × 10−3 Pa s
variations across its width are neglected. All relevant param-

eters are summarized in Table 2.
Strack (1982) provided an analytical solution for the one
dimensional version of the problem derived in terms of a The presence of the discrete fracture disturbed the uniform
complex potential flow as horizontal flow in a close domain around its location. There,
the isolines of constant pressure are distorted, resulting in
p 1 a faster flow preferentially oriented parallel to the fracture
= −A (Z − 1)(Z + 1) + AZ − q0 Leiα Z, (42) plane. In order to test the reliability of the numerical solution,
2
we compare the computed pore pressure extracted along a
where q0 is the magnitude of the flow occurring within the diagonal cross section cutting the model domain along its
z− 12 (z1 +z2 ) bottom-left/top-right corners with the pore pressure derived
model domain (assumed uniform), Z = 1 is the
2 (z2 −z1 ) from the analytical solution. The comparison shows a perfect
fracture-related dimensionless variable (z1 and z2 being the fit between the two results, thus proving the applicability of
1
f Kh
endpoints of the fracture), and A = Km2L+K fh
q0 L cos α. the discrete fracture approach as implemented in the current
The numerical solution of the three-dimensional problem formulation.
has been obtained by solving a steady-state flow problem The results obtained from a variation of the above-
within the matrix–fracture domain. We plot the computed described problem by considering two self-intersecting frac-
pressure distribution together with the outline of the geome- tures embedded in a two-layered matrix are presented in the
try of the fracture in Fig. 6. Supplement.

Figure 6. Isolines of pressure computed from the 3-D numerical simulation extracted along a horizontal plane cutting the model domain.
Comparison between simulated (continuous red curve) and analytical derived (empty black circles) pressure distribution along a line through
the model.
Figure 7. Problem formulation and results of the oedometer benchmark. Panel (a) shows the problem formulation. Panel (b) illustrates the
results for different dilation angles with the stress-displacement curves.
4.1.3 Mechanical processes – 3-D elastoplastic yield envelope is a smoother version of the classical Mohr–
oedometer test Coulomb failure criterion. Under these conditions, an analyt-
ical solution for the stress state of the porous material can be
In the following test, we consider a cube of porous medium derived as described in Appendix A and serves here as ver-
with edges of 1 m. The cube is subjected to axial loading ification of the numerical implementation of the elastoplas-
with constant solid velocity vx under the conditions of an tic constitutive laws. The physical properties used for this
oedometer test (see setup in Fig. 7). The porous material un- benchmark are summarized in Table 3.
dergoes continuous loading, and it behaves elastically until Figure 7 shows the evolution of stress for an associative
the strength of the material is reached. From this time on, (red dots) and two non-associative (dilation angle different
the material undergoes plastic loading. The elastoplastic con- from friction angle) plastic potentials (blue and green dots).
stitutive laws adopted for this simple problem formulation The results exhibit a perfect agreement between the analyt-
is the Drucker–Prager plasticity model. The Drucker–Prager ical solution and the numerical prediction. In an attempt to

Figure 8. Results of the modified oedometer test to account for pore pressure coupling. Panel (a) shows the evolution of pore pressure (dia-
monds) and stress (circle) considering either a purely elastic case (red colours) or additional plastic contributions (green colours). Panel (b)
illustrates the evolution of the difference (elastic minus plastic).
better quantify the internal coupling among the elastic and Table 3. Mechanical properties for the oedometer benchmark.
inelastic component of the deformation, we have modified
the original elasto-plastic oedometer test to include in addi- Property name Symbol Value Unit
tion pore pressure coupling, therefore solving for a coupled
Bulk modulus K 2.0 × 103 MPa
hydraulic–mechanical (poroelasticity plus plasticity) prob-
Shear modulus G 2.0 × 103 MPa
lem. The aim of this test is to further validate the reliabil- Cohesion C 1 MPa
ity of the implementation of the plastic behaviour with re- Friction angle ϕ 20 ◦
spect to its coupling with the poroleastic response to external Dilation angle ψ 0, 10 or 20 ◦
loading of the porous medium. The geometry of the prob- Velocity vx 1.0 × 10−5 m s−1
lem has been kept the same as the one previously described, Edge of the cube L 1 m
as well as the solid material parameters (see Table 3). The
fluid parameters adopted comprise a constant fluid density
(ρf = 1000 kg m−3 ), fluid viscosity (µf = 1.0 × 10−3 Pa s), 4.2 Reservoir applications
porosity (n = 0.1) and permeability (k = 1.0 × 10−15 m 2 )
and a Biot coefficient of α = 0.6. Pore pressure is initialized In the following two subsections, we will describe two syn-
to zero at the beginning of the simulation inside the model thetic examples that deal with applications as considered of
domain and all lateral boundaries are considered to be close interest in the context of operational geothermal activities.
to pressure, thus simulating typical undrained experimental The examples presented do not address a specific field ap-
conditions. The boundary conditions in terms of mechani- plication, though they do bear similarities and links to real
cal displacement have been considered as for the reference cases. An application to an actual field study case based on an
case. Inelastic behaviour is introduced in the form of a non- injection test performed at the Groß Schönebeck geothermal
associated plasticity with a friction angle of φ = 20◦ and a site (north Germany) is the subject of a separate publication
dilation angle of ψ = 10◦ . Given the setup of the model, (Jacquey et al., 2017).
both the stress and the pore pressure increase with time as the
loading from the right lateral boundary continues. Indeed, the 4.2.1 Thermal and hydraulic processes – prototype of
pore pressure evolution is affected by variations in the total multi-fractured geothermal reservoir
volumetric strain, better quantified in terms of its rates (see
Eq. 27). The onset of plastic behaviour, indicated by a change In this example, we present a setup inspired by a typical
in the slope of the stress profile, affects the pore pressure geothermal reservoir application. The model aims at simulat-
distribution and evolution, with rates of pore pressure build- ing the thermal and hydraulic configuration of the reservoir
up decreasing due to accumulation of irreversible volumetric during operational activities (injection and production) span-
strains, thus validating the poroelastic and plastic coupling as ning a life time of the reservoir of approximately 100 years.
implemented in the simulator (Fig. 8). The model consists of four different geological formations,
two units representing the target reservoir plus an upper and
lower formation acting as cap rocks. The extent of the model
domain is 10 × 10 × 3 km in the x, y, and z directions. The
target reservoir is located at a depth of approximately 4.6 km

scribed by running a steady-state simulation (given the same

boundary conditions as described above) without considering
any operational activity. This was done in order to equilibrate
the fluid density and viscosity, here considered functional of
the primary variables according to IAPWS (2008a, b), thus
avoiding spurious oscillation at the beginning of the transient
simulation
Operational activity is simulated by injecting water at a
fixed temperature, Tin = 55 ◦ C, and at a constant rate qin =
30 L s−1 for the entire simulation period. Production rates are
kept equal to injection rates through the operation. Figure 9
illustrates the reservoir state at the end of the simulation (an
animation of the full evolution of the system is provided in
the Supplement). Figure 9 nicely illustrates the evolution of
the reservoir temperature as resulting from the dynamics of
interactions between the different components of the system
(i.e. reservoir matrix, fractures, and geothermal wells) and
shows how the simulator can be effectively applied to 3-D
modelling of complex reservoirs.
4.2.2 Thermal, hydraulic and mechanical processes –

thermo-poroelastic response of a synthetic
geothermal doublet
Figure 9. Problem formulation and results of the prototype of multi- This example considers operations of a synthetic geothermal
fractured geothermal reservoir. Panel (a) shows the geometry and doublet within a low permeability geological formation with
setup of the simulation and panel (b) the distribution of the temper- induced fractures located at the two operational wells. The
ature after approximately 100 years of production. model aims at describing the thermo-poroelastic response of
the reservoir due to geothermal operations. The extent of the
model domain is 500 × 500 × 200 m in the x, y, z directions.
below sea level, and is cut by a natural fault showing a slip The depth of the target reservoir is approximatively 4 km be-
of some hundreds metres at depths of relevance in the reser- low sea level. Pore pressure and temperature distributions are
voir. A doublet system is integrated in the model, consisting assumed to be homogeneous at the beginning of the simula-
of an injection and a production well. The open-hole section, and equal to 40 MPa and 150 ◦ C respectively. A regional
tion of the two wells is kept parallel and extends for approx- stress field is applied as background stress, to simulate a nor-
imately 1.5 km horizontally in the reservoir. The open-hole mal faulting regime with the following magnitudes:
section of the wells has been integrated as one-dimensional – Vertical stress S1 = 100 MPa in the z direction.
finite elements and homogenization of the resulting govern-
ing equations is done by considering the surface area of the – Maximum horizontal stress S2 = 90 MPa in the y direc-
well bore as scaling parameter. A system of hydraulically tion.
stimulated fractures is also considered to enhance the hy-
– Minimum horizontal stress S3 = 50 MPa in the x direc-
draulic connection between the two wells along their hori-
tion.
zontal sections. The multifrac system is intended to represent
a hydraulic stimulation campaign prior to reservoir exploita- Two hydraulic fractures are considered for this doublet
tion. There are a total of 10 fractures, equally spaced every system and represent the impacts of a prior hydraulic stimu-
100 m and all sharing the same geometry and material prop- lation campaign to enhance the productivity of the reservoir.
erties. A schematic representation of the main geometry is Given the in situ stress conditions, these two hydraulic frac-
illustrated in Fig. 9, and all properties are listed in Table 4. tures are orthogonal to the minimum horizontal stress. They
The reservoir boundary conditions adopted for the simu- are implemented as squares with edges of 100 m. The dis-
lation comprise a constant temperature fixed along the top tance between the fractures is 200 m.
of the reservoir (T = 137.5◦ , corresponding to a background In this simulation, we consider additional nonlinear ef-
thermal gradient of approximately 30◦ per kilometre), heat fects as related to imposed variations in the evolution of
input at the base of the model (qT = 72 mW m−2 ), and hy- the fluid (i.e. fluid density and viscosity) and rock proper-
drostatic pressure along all lateral borders of the domain. ties (i.e. porosity and permeability) as a function of the evo-
The natural system of the reservoir, before stimulation, is de- lution in the state of the reservoir during operational activ-

Table 4. Material properties for the multifrac reservoir application.
Unit Property name Symbol Value SI unit

Cap rock I Porosity n 0.01 –
Permeability kx 1.0 × 10−20 m2
ky 1.0 × 10−20 m2
kz 0.25 × 10−20 m2
Fluid modulus Kf 1.0 × 108 Pa
Fluid viscosity µf 3.0 × 10−4 Pa s
Rock density ρs 2650 kg m−3
Rock thermal conductivity λs 4 W m−1 K−1
Rock heat capacity cs 920 J kg−1 K−1
Cap rock II Porosity n 0.01 –
ky 1.0 × 10−20 m2
kz 1.0 × 10−20 m2
Rock thermal conductivity λs 2.31 W m−1 K−1
Reservoir unit I Porosity n 0.15 –
ky 1.28 × 10−15 m2
kz 3.2 × 10−16 m2
Reservoir unit II Porosity n 0.005 –
ky 9.87 × 10−17 m2
kz 2.4675 × 10−17 m2
Fault Porosity n 1 –
Aperture h 1.0 × 10−2 m
ky 1.0 × 10−15 m2
kz 1.0 × 10−15 m2
Fluid density ρs 1148 kg m−3
Fluid thermal conductivity λs 0.65 W m−1 K−1
Fluid heat capacity cs 4193.5 J kg−1 K−1
Fractures Porosity n 1 –
ky 4.33 × 10−9 m2
kz 4.33 × 10−9 m2
Fluid density ρs 1148 kg m−3
Fluid thermal conductivity λs 0.65 W m−1 K−1
Fluid heat capacity cs 4193.5 J kg−1 K−1

Table 5. Material properties for the thermo-poroelastic response of a geothermal doublet.
Unit Property name Symbol Value SI unit

Reservoir rock Porosity n 0.1 –
Permeability k 1.0 × 10−15 m2
Rock thermal conductivity λs 3 W m−1 K−1
Fractures Porosity n 1 –
Permeability k 8.333 × 10−10 m2
Fluid density ρf 1000 kg m−3
Fluid thermal conductivity λf 0.65 W m−1 K−1
Fluid heat capacity cf 4200 J kg−1 K−1
Figure 10. Problem formulation and results of the thermo-poroelastic response of a geothermal doublet. Panel (a) shows the problem
formulation. Panels (b) and (c) illustrates the pore pressure (with fluid velocity) and temperature distributions respectively after 50 years of
operations.
ities. Changes in porosity are controlled by Eq. (22), where where the coefficient A includes information about the pore
we neglected second-order terms for this specific application, and grain geometries and can be expressed via the initial
2
while the evolution in the rock permeability is governed by a value of porosity and permeability: A = k0 (1−n30 ) .
n0
classical Kozeny–Carman-like relation as
Figure 10 illustrates the model geometry.
Geothermal operations consist of injecting water at Tin =
n3
k=A , (43) 70 ◦ C and at a constant rate qin = 5 L s−1 and producing
(1 − n)2

Figure 11. Horizontal slices of the model illustrating the changes in transport properties. Panel (a) shows the distribution of the changes in
porosity and panel (b) the changes in permeability after 50 years of operations.
geothermal fluid at the same rate. Pore pressure and temper- erties, thus providing higher numerical stability in the con-
ature are kept constant on all sides of the model as bound- text of nonlinear problems. Geological heterogeneities, i.e.
ary conditions. All faces are considered as sliding faces for discrete fractures and fault zones, are taken into account in
boundary conditions in terms of displacements. The simula- our formulation. The latter are represented as finite element
tion covers a complete time of 50 years of geothermal op- of lower geometrical dimension, which allows us to model
erations. All physical properties used for this example are focused fluid and heat flows on fractures and faults planes
summarized in Table 5. or well paths. The capability and robustness of the simulator
Figure 10 shows the distribution of pore pressure and tem- has been illustrated by means of five numerical examples by
perature at the end of the simulation. It can be observed that increasing progressively the coupling and geometrical com-
colder temperature than the reservoir temperature are pro- plexity of the considered problem formulations.
duced at the right well after 50 years of operations. Improving the reliability of predictions made for geother-
Figure 11 shows horizontal slices at the centre of the do- mal operations at the field scale requires a better description
main illustrating the changes in porosity and permeability of the physical phenomena which can alter the reservoir pro-
due to the applied changes in strain, pore pressure, and tem- ductivity as well as the sustainability of the geothermal op-
perature. erations. In this respect, the current framework provides a
This example illustrates the simulator capability of solving powerful tool to analyse the dynamic behaviour of fractured
a full thermo-hydro-mechanical coupled problem as relevant reservoirs during geothermal operations.
for geothermal reservoir applications. At the same time, the Ongoing activities are towards integration of the details of
formulation adopted to simulate the evolution in time and the mechanical description of the geological discontinuities
space of the system properties enables us to quantify the im- (faults and fractures) either by means of a discrete (XFEM
pact of the mechanical alteration induced by geothermal op- approach) or continuous (such as phase field) approach. Such
erations (injection/production of fluid and applied injection a feature will help to better describe the dynamics defor-
temperature) on both fluid and heat flows within the reser- mation in heterogeneous rocks, including localization and
voir. evolution along fault zones, and will also permit to quanti-
tative integrate feedbacks on the hydraulic and thermal be-
haviours of such geological structures. The description of
5 Conclusions such processes would help at forecasting environmental im-
pacts of reservoir operations such as induced seismicity from
In this paper, we have presented a novel but robust sim-
dynamic reactivation of faults during operational activities.
ulator for modelling coupled THM processes within frac-
tured rocks, with specific focus on reservoir applications.
The code, GOLEM, relies on an open-source massive par-
Code and data availability. The source code and the input files of
allel finite-element-based numerical framework (MOOSE) to the five numerical examples presented in this paper, plus a suite
solve for the coupled problem. It makes use of a fully implicit of specific benchmark cases, are available for downloading from a
approach to treat the nonlinear coupling among the different GitHub repository (https://github.com/ajacquey/Golem).
processes and their feedback effects on fluid and rock prop-

Appendix A: Analytical solution for the oedometer Furthermore, the Kuhn–Tucker conditions imply that the
benchmark derivative over time of the yield function remains 0 during
plastic accumulation, that is, in incremental form
During elastic loading with constant velocity vx , the strain
evolves as −1
1F = 0 ⇔ √ 1σxx − 1σyy
vx t 3
xx = , yy = zz = 0 (A1) sin(ϕ)
L + 1σxx + 21σyy = 0. (A7)
3
because of the no-displacement boundary conditions in the y
and z directions. The stress during elastic loading therefore By injecting Eq. (A6) into Eq. (A7), one finally obtains the
reads value for the increment of plastic multiplier as
vx t vx t

σxx = (λ + 2G) , σyy = σzz = λ . (A2) √2 G − sin(ϕ) λ + 2 G
3 3 |vx |1t
L L 1γ = . (A8)
G + 32 (λ + G) sin(ϕ) sin(ψ) L
With these expressions of stress, the invariants of the stress
tensor, J1 and J2 can be written as The solution for plastic strain and stress can therefore be
2 integrated as
vx t 4 vx t
J1 = (3λ + 2G) , J2 = G . (A3) γ (t)

sin(ψ)

L 3 L ∗p
xx (t) = √ −1 + √
3 3
The onset of plastic strain accumulation is reached when
∗p γ (t) 1 sin(ψ)
the yield function reaches 0 at a time noted ty , that is, yy (t) = √ + √
F(ty ) = 0. By using the expressions of the stress invariants 3 2 3

in Eq. (A3), the time for onset of yielding can be expressed ∗p γ (t) 1 sin(ψ)
zz (t) = √ + √ (A9)
as 3 2 3
LC cos ϕ

vx t 2 sin(ψ)
ty = . (A4) σxx (t) = (λ + 2G) + γ (t) √ 1 − √ G − λ sin(ψ)
L 3 3
|vx | √2 G − sin(ϕ)(λ + 2 G)
3 3 vx t

2 1 sin(ψ)

σyy (t) = λ − γ (t) √ + √ G + λ sin(ψ)
The increment of plastic strain accumulation can be ex- L 3 2 3

pressed by reinjecting the expressions of the stress invariant vx t 2 1 sin(ψ)
σzz (t) = λ − γ (t) √ + √ G + λ sin(ψ) ,
and Eq. (40) into Eq. (20): L 3 2 3
(A10)

∗p 1γ sin(ψ)
1xx = √ −1 + √ with the plastic multiplier as
3 3

∗p 1γ 1 sin(ψ) √2 G − sin(ϕ) λ + 2 G
1yy = √ + √ 3 3 |vx |t
3 2 3 γ (t) = 2
. (A11)
G + 3 (λ + G) sin(ϕ) sin(ψ) L
∗p 1γ 1 sin(ψ)
1zz = √ + √ . (A5)
3 2 3
The increment of stress during plastic accumulation can
therefore be written:

vx 1t 2 sin(ψ)
1σxx = (λ + 2G) + 1γ √ 1 − √ G − λ sin(ψ)
L 3 3

vx 1t 2 1 sin(ψ)
1σyy = λ − 1γ √ + √ G + λ sin(ψ)
L 3 2 3

vx 1t 2 1 sin(ψ)
1σzz = λ − 1γ √ + √ G + λ sin(ψ) . (A6)
L 3 2 3

The Supplement related to this article is available online Boley, B. and Weiner, J. H.: Theory of Thermal Stresses, Dover
at https://doi.org/10.5194/se-8-921-2017-supplement. Civil and Mechanical Engineering, New York, NY, USA,
586 pp., 1960.
Brooks, A. N. and Hughes, T. J.: Streamline upwind/Petrov-
Galerkin formulations for convection dominated flows
with particular emphasis on the incompressible Navier-
Author contributions. MC and ABJ developed the model code, de- Stokes equations, Comput. Method. Appl. M., 32, 199–259,
signed the examples presented in the paper and performed the sim- https://doi.org/10.1016/0045-7825(82)90071-8, 1982.
ulations. Both authors also shared the preparation and writing of the Cacace, M. and Blöcher, G.: MeshIt – a software for three dimen-
paper. sional volumetric meshing of complex faulted reservoirs, Envi-
ron. Earth Sci., 74, 5191–5209, https://doi.org/10.1007/s12665-
015-4537-x, 2015.
Competing interests. The authors declare that they have no conflict Chabab, E. and Kempka, T.: Quantification of Fluid Mi-
of interest. gration Via faults Requires two-way Coupled Hydrome-
chanical Simulations, Energy Procedia, 97, 372–378,
https://doi.org/10.1016/j.egypro.2016.10.025, 2016.
Acknowledgements. The authors would like to thank the develop- Chow, E., Clearly, A. J., and Falgout, R. D.: Design of the hypre Pre-
ers team of the MOOSE framework as well as the members of the conditioner Library, in: Proc. of the SIAM Workshop on Object
MOOSE users community for their helpful support and guidance Priented Methods for Inter-operable Scientific and Engineering
for code development using the MOOSE framework. This work Computing, edited by: Henderson, M., Anderson, C., and Lyons,
was funded by the Helmholtz Association as part of the Helmholtz S., SIAM press, Yorktown Heights, New York, USA, 1998.
Portfolio Geo-Energy project. Diersch, H.-J. and Kolditz, O.: Coupled groundwater flow and
transport: 2. Thermohaline and 3D convection systems, Adv.
The article processing charges for this open-access Water Resour., 21, 401–425, https://doi.org/10.1016/S0309-
publication were covered by a Research 1708(97)00003-1, 1998.
Centre of the Helmholtz Association. Drucker, D.: Some implications of work-hardening and ideal plas-
ticity, Q. Appl. Math., 7, 411–418, 1950.
Edited by: Gwenn Peron-Pinvidic Elder, J. W.: Transient convection in a porous
Reviewed by: Joshua Taron and Manolis Veveakis medium, J. Fluid Mech., 27, 609–623,
https://doi.org/10.1017/S0022112067000576, 1967.
Galeão, A., Almeida, R., Malta, S., and Loula, A.: Fi-
nite element analysis of convection dominated reaction–
References diffusion problems, Appl. Numer. Math., 48, 205–222,
https://doi.org/10.1016/j.apnum.2003.10.002, 2004.
Balay, S., Abhyankhar, S., Adams, M. F., Brown, J., Brune, P., Gaston, D., Newman, C., Hansen, G., and Lebrun-Grandié, D.:
Buschelman, K., Dalcin, L., Eijkhout, V., Gropp, W. D., Kaushik, MOOSE: A parallel computational framework for coupled sys-
D., Knepley, M. G., Curfman McInnes, L., Rupp, K., Smith, tems of nonlinear equations, Nucl. Eng. Des., 239, 1768–1778,
B. F., Zampini, S., and Zhang, H.: PETSc Users Manual, Tech. https://doi.org/10.1016/j.nucengdes.2009.05.021, 2009.
rep., Argonne National Laboratory, available at: https://www. Geertsma, J.: A remark on the analogy between thermoelasticity and
mcs.anl.gov/petsc (last access: 11 September 2017), 2016. the elasticity of saturated porous media, J. Mech. Phys. Solids, 6,
Bear, J.: Dynamics of Fluids in Porous Media, Dover, Dover book 13–16, https://doi.org/10.1016/0022-5096(57)90042-X, 1957.
edn., New York, NY, USA, 1988. Giudice, S. D., Comini, G., and Nonino, C.: A physical interpreta-
Biot, M. A.: General Solutions of the Equations of Elasticity and tion of conservative and non-conservative finite element formu-
Consolidation for a Porous Material, J. Appl. Mech., 78, 91–96, lations of convection-type problems, Int. J. Numer. Meth. Eng.,
1956. 35, 709–727, https://doi.org/10.1002/nme.1620350406, 1992.
Biot, M. A.: Nonlinear and semilinear rheology of Guermond, J.-L., Pasquetti, R., and Popov, B.: Entropy viscosity
porous solids, J. Geophys. Res., 78, 4924–4937, method for nonlinear conservation laws, J. Comput. Phys., 230,
https://doi.org/10.1029/JB078i023p04924, 1973. 4248–4267, https://doi.org/10.1016/j.jcp.2010.11.043, 2011.
Biot, M. A. and Willis, D.: Theory of Consolidation, in: Theoretical Heroux, M. A., Phipps, E. T., Salinger, A. G., Thornquist, H. K.,
Soil Mechanics, 265–296, John Wiley & Sons, Inc., Hoboken, Tuminaro, R. S., Willenbring, J. M., Williams, A., Stanley, K. S.,
NJ, USA, https://doi.org/10.1002/9780470172766.ch13, 1957. Bartlett, R. A., Howle, V. E., Hoekstra, R. J., Hu, J. J., Kolda,
Blöcher, G., Reinsch, T., Hassanzadegan, A., Milsch, H., T. G., Lehoucq, R. B., Long, K. R., and Pawlowski, R. P.: An
and Zimmermann, G.: Direct and indirect laboratory overview of the Trilinos project, ACM T. Math. Software, 31,
measurements of poroelastic properties of two consoli- 397–423, https://doi.org/10.1145/1089014.1089021, 2005.
dated sandstones, Int. J. Rock Mech. Min., 67, 191–201, Hudson, J. A. and Harrison, J. P.: Engineering Rock Mechanics,
https://doi.org/10.1016/j.ijrmms.2013.08.033, 2014. Pergamon, Oxford, Uk, 1997.
Blöcher, G., Cacace, M., Reinsch, T., and Watanabe, N.: Evalua- IAPWS: Release on the IAPWS Formulation 2008 for the Viscosity
tion of three exploitation concepts for a deep geothermal sys- of Ordinary Water Substance, Tech. rep., IAPWS, available at:
tem in the North German Basin, Comput. Geosci., 82, 120–129, http://www.iapws.org (last access: 1 May 2017), 2008a.
https://doi.org/10.1016/j.cageo.2015.06.005, 2015.

IAPWS: Release on the IAPWS Formulation 2008 for the Thermo- McTigue, D. F.: Thermoelastic response of fluid-
dynamic Properties of Seawater, Tech. rep., IAPWS, available at: saturated porous rock, J. Geophys. Res., 91, 9533,
http://www.iapws.org (last access: 1 May 2017), 2008b. https://doi.org/10.1029/JB091iB09p09533, 1986.
Jacquey, A. B., Cacace, M., Blöcher, G., Watanabe, N., and Scheck- Mura, T.: Micromechanics of defects in solids, vol. 3 of Mechanics
Wenderoth, M.: Hydro-Mechanical Evolution of Transport Prop- of Elastic and Inelastic Solids, Springer Netherlands, Dordrecht,
erties in Porous Media: Constraints for Numerical Simula- https://doi.org/10.1007/978-94-009-3489-4, 1987.
tions Roman symbols, Transport Porous Med., 110, 409–428, Nguyen, T. S. and Selvadurai, A. P. S.: a Model for Coupled Me-
https://doi.org/10.1007/s11242-015-0564-z, 2015. chanical and Hydraulic Behaviour of a Rock Joint, Int. J. Numer.
Jacquey, A. B., Urpi, L., Cacace, M., Blöcher, G., Zimmermann, G., Anal. Met., 22, 29–48, 1998.
and Scheck-Wenderoth, M.: Poroelastic response of geothermal Nield, D. A. and Bejan, A.: Convection in Porous Media, Springer
reservoirs to hydraulic stimulation treatment: theory and appli- Verlag GmbH, New York, NY, USA, 2012.
cation to the Groß Schönebeck geothermal research facility, J. Ogata, A. and Banks, R. B.: A solution of the differential equation
Geophys. Res.-Sol. Ea., under review, 2017. of longitudinal dispersion in porous media, Tech. rep., US Ge-
Jha, B. and Juanes, R.: A locally conservative finite el- ological Survey, available at: http://pubs.er.usgs.gov/publication/
ement framework for the simulation of coupled flow pp411A (last access: 11 September 2017), 1961.
and reservoir geomechanics, Acta Geotech., 2, 139–153, Poulet T. and Veveakis, E.: A viscoplastic approach for
https://doi.org/10.1007/s11440-007-0033-0, 2007. pore collapse in saturated soft rocks using REDBACK: an
Jiang, H. and Xie, Y.: A note on the Mohr-Coulomb and Drucker- open-source parallel simulator for Rock mEchanics with
Prager strength criteria, Mech. Res. Commun., 38, 309–314, Dissipative feedBACKs, Comput. Geotech., 74, 211–221,
https://doi.org/10.1016/j.mechrescom.2011.04.001, 2011. https://doi.org/10.1016/j.compgeo.2015.12.015, 2016.
Jing, L.: A review of techniques, advances and outstand- Preisig, M. and Prévost, J. H.: Coupled multi-phase thermo-
ing issues in numerical modelling for rock mechanics and poromechanical effects. Case study: CO2 injection at In
rock engineering, Int. J. Rock Mech. Min., 40, 283–353, Salah, Algeria, Int. J. Greenh. Gas Con., 5, 1055–1064,
https://doi.org/10.1016/S1365-1609(03)00013-3, 2003. https://doi.org/10.1016/j.ijggc.2010.12.006, 2011.
Kempka, T., Nakaten, B., De Lucia, M., Nakaten, N., Otto, C., Rutqvist, J.: Status of the TOUGH-FLAC simulator and
Pohl, M., Tillner, E., and Kühn, M.: Flexible Simulation Frame- recent applications related to coupled fluid flow and
work to Couple Processes in Complex 3D Models for Sub- crustal deformations, Comput. Geosci., 37, 739–750,
surface Utilization Assessment, Energy Procedia, 97, 494–501, https://doi.org/10.1016/j.cageo.2010.08.006, 2011.
https://doi.org/10.1016/j.egypro.2016.10.058, 2016. Simo, J. C. and Hughes, T. J. R.: Computational Inelasticity, vol. 7
Kim, J., Wang, W., and Regueiro, R. A.: Hybrid time integra- of Interdisciplinary Applied Mathematics, Springer-Verlag, New
tion and coupled solution methods for nonlinear finite ele- York, https://doi.org/10.1007/b98904, 1998.
ment analysis of partially saturated deformable porous me- Stephansson, O., Hudson, J. A., and Jing, L.: Coupled Thermo-
dia at small strain, Int. J. Numer. Anal. Met., 39, 1073–1103, Hydro-Mechanical-Chemical processes in geo-systems. Funda-
https://doi.org/10.1002/nag.2350, 2015. mentals, ModeModel, Experiments and Applications, Elsevier
Kirk, B. S., Peterson, J. W., Stogner, R. H., and Carey, G. F.: Geo-Engineering book series volume 2, Amsterdam, the Nether-
libMesh: a C++ library for parallel adaptive mesh refine- lands, 2004.
ment/coarsening simulations, Eng. Comput., 22, 237–254, Strack, O.: Assessment of effectiveness of geologic isolation sys-
https://doi.org/10.1007/s00366-006-0049-3, 2006. tems. Analytic modeling of flow in a permeable fissured medium,
Knoll, D. and Keyes, D.: Jacobian-free Newton–Krylov methods: Tech. rep., Pacific Northwest National Laboratory (PNNL),
a survey of approaches and applications, J. Comput. Phys., 193, Richland, WA, USA, https://doi.org/10.2172/5399775, 1982.
357–397, https://doi.org/10.1016/j.jcp.2003.08.010, 2004. Sun, W.: A stabilized finite element formulation for
Kolditz, O., Bauer, S., Bilke, L., Böttcher, N., Delfs, J. O., Fis- monolithic thermo-hydro-mechanical simulations at fi-
cher, T., Görke, U. J., Kalbacher, T., Kosakowski, G., McDer- nite strain, Int. J. Numer. Meth. Eng., 103, 798–839,
mott, C. I., Park, C. H., Radu, F., Rink, K., Shao, H., Shao, https://doi.org/10.1002/nme.4910, 2015.
H. B., Sun, F., Sun, Y. Y., Singh, a. K., Taron, J., Walther, Vermeer, P. A. and de Borst, R.: Non-associated plasticity for soils,
M., Wang, W., Watanabe, N., Wu, Y., Xie, M., Xu, W., and concrete and rock, HERON, 29, 3–64, 1984.
Zehner, B.: OpenGeoSys: an open-source initiative for numeri- Watanabe, N.: Finite element method for coupled thermo-hydro-
cal simulation of thermo-hydro-mechanical/chemical (THM/C) mechanical processes in discretely fractured and non-fractured
processes in porous media, Environ. Earth Sci., 67, 589–599, porous media, PhD thesis, Dresden Technical University, Dres-
https://doi.org/10.1007/s12665-012-1546-x, 2012. den, Germany, 2011.

GOLEM

Uploaded by

Copyright:

Available Formats

GOLEM

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

GOLEM

Uploaded by

Copyright:

Available Formats

Solid Earth, 8, 921–941, 2017

Flexible parallel implicit modelling of coupled

Correspondence to: Mauro Cacace (mauro.cacace@gfz-potsdam.de)

Received: 23 March 2017 – Discussion started: 31 March 2017

Abstract. Theory and numerical implementation describing 1 Introduction

Published by Copernicus Publications on behalf of the European Geosciences Union.

Solid Earth, 8, 921–941, 2017 www.solid-earth.net/8/921/2017/

Attention is given to include the important nonlinear feed-

www.solid-earth.net/8/921/2017/ Solid Earth, 8, 921–941, 2017

Solid Earth, 8, 921–941, 2017 www.solid-earth.net/8/921/2017/

www.solid-earth.net/8/921/2017/ Solid Earth, 8, 921–941, 2017

position as of interconnected issues must be taken into account. These

Solid Earth, 8, 921–941, 2017 www.solid-earth.net/8/921/2017/

– Continuity of the total (diffusive plus advective) heat

www.solid-earth.net/8/921/2017/ Solid Earth, 8, 921–941, 2017

The mass balance equation for a discrete fracture then

+ 3+ − where N and N0 are direction-dependent material properties

Solid Earth, 8, 921–941, 2017 www.solid-earth.net/8/921/2017/

to be solved for. Nonlinearities arise due the nonlinear con- ω∗ , as

www.solid-earth.net/8/921/2017/ Solid Earth, 8, 921–941, 2017

Solid Earth, 8, 921–941, 2017 www.solid-earth.net/8/921/2017/

4.1 Benchmarks for coupled processes

www.solid-earth.net/8/921/2017/ Solid Earth, 8, 921–941, 2017

Table 2. Material properties the example of flow in a fractured

Property name Symbol Value SI unit

variations across its width are neglected. All relevant param-

Solid Earth, 8, 921–941, 2017 www.solid-earth.net/8/921/2017/

www.solid-earth.net/8/921/2017/ Solid Earth, 8, 921–941, 2017

Solid Earth, 8, 921–941, 2017 www.solid-earth.net/8/921/2017/

scribed by running a steady-state simulation (given the same

4.2.2 Thermal, hydraulic and mechanical processes –

www.solid-earth.net/8/921/2017/ Solid Earth, 8, 921–941, 2017

Table 4. Material properties for the multifrac reservoir application.

Unit Property name Symbol Value SI unit

Solid Earth, 8, 921–941, 2017 www.solid-earth.net/8/921/2017/

Table 5. Material properties for the thermo-poroelastic response of a geothermal doublet.

Unit Property name Symbol Value SI unit

www.solid-earth.net/8/921/2017/ Solid Earth, 8, 921–941, 2017

Solid Earth, 8, 921–941, 2017 www.solid-earth.net/8/921/2017/

www.solid-earth.net/8/921/2017/ Solid Earth, 8, 921–941, 2017

Solid Earth, 8, 921–941, 2017 www.solid-earth.net/8/921/2017/

www.solid-earth.net/8/921/2017/ Solid Earth, 8, 921–941, 2017

You might also like