Uploaded by Xuandung Vu

0.0 (0)

Editors:

J. Reitner, G¨ ottingen

M. H. Trauth, Potsdam

K. St ¨ uwe, Graz

D. Yuen, USA

Founding Editors:

G. M. Friedman, Brooklyn and Troy

A. Seilacher, T¨ ubingen and Yale

Chongbin Zhao · Bruce E. Hobbs · Alison Ord

Fundamentals

of Computational

Geoscience

Numerical Methods and Algorithms

123

Dr. Chongbin Zhao

Computational Geosciences

Research Centre

Central South University

Changsha, China

Chongbin.zhao@iinet.net.au

Dr. Alison Ord

School of Earth and Geographical Sciences

The University of Western Australia

Perth, Australia

and CSIRO Division of

Exploration and Mining

Perth, Australia

alison.ord@csiro.au

Dr. Bruce E. Hobbs

School of Earth and Geographical Sciences

The University of Western Australia

Perth, Australia

Bruce.Hobbs@csiro.au

ISBN 978-3-540-89742-2 e-ISBN 978-3-540-89743-9

DOI 10.1007/978-3-540-89743-9

Lecturer Notes in Earth Sciences ISSN 0930-0317

Library of Congress Control Number: 2008942068

c

Springer-Verlag Berlin Heidelberg 2009

This work is subject to copyright. All rights are reserved, whether the whole or part of the material is

concerned, speciﬁcally the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting,

reproduction on microﬁlm or in any other way, and storage in data banks. Duplication of this publication

or parts thereof is permitted only under the provisions of the German Copyright Law of September 9,

1965, in its current version, and permission for use must always be obtained from Springer. Violations

are liable to prosecution under the German Copyright Law.

The use of general descriptive names, registered names, trademarks, etc. in this publication does not

imply, even in the absence of a speciﬁc statement, that such names are exempt from the relevant protective

laws and regulations and therefore free for general use.

Cover design: deblik, Berlin

Printed on acid-free paper

9 8 7 6 5 4 3 2 1

springer.com

Acknowledgements

Dr Chongbin Zhao expresses his sincere thanks to his wife, Ms Peiying Xu, for

her persistent support and encouragement, without which it would be impossible

to write this monograph. We are very grateful to the Central South University for

ﬁnancial support during writing this monograph. The partial work of this monograph

is also ﬁnancially supported by the Natural Science Foundation of China (Grant

Nos: 10872219 and 10672190). We express our thanks to the anonymous referees

for their valuable reviews of this monograph.

v

Preamble

In recent years, numerical methods and computational simulations provide a new

way to deal with many geoscience problems, for which the traditionally-used the-

oretical and experimental methods may not be valid as a result of the large time

and length scales of the problems themselves. This enables many hitherto unsolv-

able geoscience problems to be solved using numerical methods and computational

simulations. In particular, through wide application of computational science to

geoscience problems, a new discipline, namely computational geoscience, has been

established. However, because of the extremely large length and time scales, the

numerical simulation of a real geological world also provides many challenging

problems for researchers involved in the ﬁeld of computational science. For this

reason, multidisciplinary knowledge and expertise frommathematicians, physicists,

chemists, computational scientists and geoscientists are required in the process of

establishing the research methodology of computational geoscience.

Since computational geoscience is an amalgamation of geoscience and com-

putational science, theoretical analysis and computational simulation are two of its

core members. On the theoretical analysis front, we need: (1) to measure and gather

data and information through traditional geoscience observations and measurements

such as those widely used in geology, geophysics, geochemistry and many other

scientiﬁc and engineering ﬁelds; (2) to conduct research to ﬁnd the key factors and

processes that control the geoscience problem under consideration; (3) to establish

the theoretical foundations of the geoscience problem through formulating a set

of partial differential equations on the basis of fundamental scientiﬁc principles;

(4) to investigate the solution characteristics of these partial differential equations

using rigorous mathematical treatments. On the computational simulation front, we

need: (1) to develop advanced numerical methods, procedures and algorithms for

simulating multi-scale and multi-process aspects of the geoscience problem on the

basis of contemporary computational science knowledge and expertise; (2) to verify

computational codes established on the basis of these advanced numerical meth-

ods, procedures and algorithms through comparing numerical solutions with bench-

mark solutions; (3) to produce and validate numerical solutions of real geoscience

problems.

Owing to the broad nature of geoscience problems, computational geoscience

is at a developing stage. Nevertheless, under the stimulus of ever-increasing demand

vii

viii Preamble

for natural mineral resources, computational geoscience has achieved much in the

past decade, driven from the need to understand controlling mechanisms behind

ore body formation and mineralization in hydrothermal systems within the upper

crust of the Earth. In order to disseminate widely the existing knowledge of com-

putational geoscience, to promote extensively and fastly further development of the

computational geoscience, and to facilitate efﬁciently the broad applications of com-

putational geoscience, it is high time to publish a monograph to report the current

knowledge in a systematic manner. This monograph aims to provide state-of-the-

art numerical methods, procedures and algorithms in the ﬁeld of computational

geoscience, based on the authors’ own work during the last decade. For this pur-

pose, although some theoretical results are provided to verify numerical ones, the

main focus of this monograph is on computational simulation aspects of this newly-

developed computational geoscience discipline. The advanced numerical methods,

procedures and algorithms contained in this monograph are also applicable to a

wide range of problems of other length-scales such as engineering length-scales.

To broaden the readership of this monograph, common mathematical notations are

used to describe the theoretical aspects of geoscience problems. This enables this

monograph to be used either as a useful textbook for postgraduate students or as

an indispensable reference book for computational geoscientists, mathematicians,

engineers and geoscientists. In addition, each chapter is written independently of

the remainder of the monograph so that readers may read the chapter of interest

separately.

In this monograph we use the ﬁnite element method, the ﬁnite difference

method and the particle simulation method as basic numerical methods for deal-

ing with geoscience problems. Not only have these three methods been well de-

veloped in the ﬁeld of computational science, but also they have been successfully

applied to a wide range of small-scale scientiﬁc and engineering problems. Based

on these three methods, we have developed advanced numerical procedures and

algorithms to tackle the large-scale aspects of geoscience problems. The speciﬁc

geoscience problem under consideration is the ore body formation and mineraliza-

tion problem in hydrothermal systems within the upper crust of the Earth. Towards

this end, we present the advanced procedures and algorithms in this monograph

as follows: (1) Due to the important role that convective pore-ﬂuid ﬂow plays in

the controlling processes of ore body formation and mineralization, a progressive

asymptotic approach procedure is proposed to solve steady-state convective pore-

ﬂuid ﬂow problems within the upper crust of the Earth. (2) To consider both the

thermoelastic effect and the double diffusion effect, a consistent point-searching

interpolation algorithmis proposed to develop a general interface between two com-

mercial computer codes, Fluid Dynamics Analysis Package (FIDAP) and Fast La-

grangian Analysis of Continua (FLAC). This general interface allows a combination

use of the two commercial codes for solving coupled problems between medium

deformation, pore-ﬂuid ﬂow, heat transfer and reactive mass transport processes

that can occur simultaneously in hydrothermal systems. (3) To simulate mineral

dissolution/precipitation and metamorphic processes, a term splitting algorithm is

developed for dealing with ﬂuid-rock interaction problems in ﬂuid-saturated hy-

Preamble ix

drothermal/sedimentary basins of subcritical Zhao numbers, in which the chemical

dissolution fronts are stable during their propagation. Note that the Zhao number

is a dimensionless number that can be used to represent the geometrical, hydro-

dynamic, thermodynamic and chemical kinetic characteristics of a reactive trans-

port system in a comprehensive manner. The condition, under which a chemical

dissolution front in the ﬂuid-saturated porous medium becomes unstable, can be

expressed by the critical value of this dimensionless number. (4) For a geochemical

system of critical and supercritical Zhao numbers, a segregated algorithm is pro-

posed for solving chemical-dissolution front instability problems in ﬂuid-saturated

porous rocks. Thus, the morphological evolution of chemical dissolution fronts in

ﬂuid-saturated porous media can be appropriately simulated. (5) To investigate the

effects of non-equilibrium redox chemical reactions on the mineralization patterns

in hydrothermal systems, a decoupling procedure is proposed for simulating ﬂu-

ids mixing, heat transfer and non-equilibrium redox chemical reactions in ﬂuid-

saturated porous rocks. (6) When thermal and chemical effects of intruded magma

are taken into account, an equivalent source algorithm is presented for simulating

thermal and chemical effects of intruded magma solidiﬁcation problems. This algo-

rithm enables the moving boundary problem associated with magma solidiﬁcation

to be effectively and efﬁciently solved using the ﬁxed ﬁnite element meshes. (7) To

simulate spontaneous crack generation in brittle rocks within the upper crust of the

Earth, the particle simulation method is extended to solve spontaneous crack gener-

ation problems associated with faulting and folding in large length-scale geological

systems. The resulting cracks may be connected to form ﬂow channels, which can

control ore body formation and mineralization patterns within the upper crust of the

Earth.

August 12, 2008 Chongbin Zhao

Bruce E. Hobbs

Alison Ord

Contents

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 Characteristics of Computational Geoscience . . . . . . . . . . . . . . . . . . . . 2

1.2 Basic Steps Associated with the Research Methodology

of Computational Geoscience . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.2.1 The Conceptual Model of a Geoscience Problem . . . . . . . . . . . 3

1.2.2 The Mathematical Model of a Geoscience Problem . . . . . . . . . 3

1.2.3 The Numerical Simulation Model of a Geoscience Problem . . 4

1.2.4 Graphical Display of the Numerical Simulation Results . . . . . 5

1.3 The Contextual Arrangements of this Monograph . . . . . . . . . . . . . . . . . 5

2 A Progressive Asymptotic Approach Procedure for Simulating

Steady-State Natural Convective Problems in Fluid-Saturated

Porous Media . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.1 Governing Equations of the Problem. . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.2 Finite Element Formulation of the Problem . . . . . . . . . . . . . . . . . . . . . . 11

2.3 The Progressive Asymptotic Approach Procedure

for Solving Steady-State Natural Convection Problems

in Fluid-Saturated Porous Media . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2.4 Derivation of Analytical Solution to a Benchmark Problem . . . . . . . . . 16

2.5 Veriﬁcation of the Proposed Progressive Asymptotic Approach

Procedure Associated with Finite Element Analysis . . . . . . . . . . . . . . . 19

2.6 Application of the Progressive Asymptotic Approach Procedure

Associated with Finite Element Analysis . . . . . . . . . . . . . . . . . . . . . . . . 22

2.6.1 Two-Dimensional Convective Pore-Fluid Flow Problems . . . . 22

2.6.2 Three-Dimensional Convective Pore-Fluid Flow Problems . . . 28

3 A Consistent Point-Searching Interpolation Algorithm for

Simulating Coupled Problems between Deformation, Pore-Fluid

Flow, Heat Transfer and Mass Transport Processes in Hydrothemal

Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

3.1 Statement of the Coupled Problem and Solution Method . . . . . . . . . . . 38

3.2 Mathematical Formulation of the Consistent Point-Searching

Interpolation Algorithm in Unstructured Meshes . . . . . . . . . . . . . . . . . 42

xi

xii Contents

3.2.1 Point Searching Step . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

3.2.2 Inverse Mapping Step . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

3.2.3 Consistent Interpolation Step . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

3.3 Veriﬁcation of the Proposed Consistent Point-Searching

Interpolation Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

3.4 Application Examples of the Proposed Consistent Point-Searching

Interpolation Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

3.4.1 Numerical Modelling of Coupled Problems Involving

Deformation, Pore-Fluid Flow and Heat Transfer

in Fluid-Saturated Porous Media . . . . . . . . . . . . . . . . . . . . . . . . 56

3.4.2 Numerical Modelling of Coupled Problems Involving

Deformation, Pore-Fluid Flow, Heat Transfer and Mass

Transport in Fluid-Saturated Porous Media . . . . . . . . . . . . . . . 59

4 A Term Splitting Algorithm for Simulating Fluid-Rock Interaction

Problems in Fluid-Saturated Hydrothermal Systems of Subcritical

Zhao Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

4.1 Key Issues Associated with the Numerical Modelling of Fluid-Rock

Interaction Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

4.2 Development of the Term Splitting Algorithm . . . . . . . . . . . . . . . . . . . . 77

4.3 Application Examples of the Term Splitting Algorithm . . . . . . . . . . . . 82

5 A Segregated Algorithm for Simulating Chemical Dissolution Front

Instabilities in Fluid-Saturated Porous Rocks . . . . . . . . . . . . . . . . . . . . . . . 95

5.1 Mathematical Background of Chemical Dissolution Front

Instability Problems in Fluid-Saturated Porous Rocks . . . . . . . . . . . . . 96

5.1.1 A General Case of Reactive Multi-Chemical-Species

Transport with Consideration of Porosity/Permeability

Feedback . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

5.1.2 A Particular Case of Reactive Single-Chemical-Species

Transport with Consideration of Porosity/Permeability

Feedack. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

5.2 Proposed Segregated Algorithm for Simulating the Morphological

Evolution of a Chemical Dissolution Front . . . . . . . . . . . . . . . . . . . . . . 109

5.2.1 Formulation of the Segregated Algorithm for Simulating

the Evolution of Chemical Dissolution Fronts . . . . . . . . . . . . . 109

5.2.2 Veriﬁcation of the Segregated Algorithm for Simulating

the Evolution of Chemical Dissolution Fronts . . . . . . . . . . . . . 111

5.3 Application of the Segregated Algorithm for Simulating

the Morphological Evolution of Chemical Dissolution Fronts . . . . . . . 115

Contents xiii

6 A Decoupling Procedure for Simulating Fluid Mixing, Heat Transfer

and Non-Equilibrium Redox Chemical Reactions in Fluid-Saturated

Porous Rocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

6.1 Statement of Coupled Problems between Fluids Mixing, Heat

Transfer and Redox Chemical Reactions . . . . . . . . . . . . . . . . . . . . . . . . 123

6.2 A Decoupling Procedure for Removing the Coupling between

Reactive Transport Equations of Redox Chemical Reactions . . . . . . . . 126

6.3 Veriﬁcation of the Decoupling Procedure . . . . . . . . . . . . . . . . . . . . . . . . 128

6.4 Applications of the Proposed Decoupling Procedure to Predict

Mineral Precipitation Patterns in a Focusing and Mixing System

Involving Two Reactive Fluids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

6.4.1 Key Factors Controlling Mineral Precipitation

Patterns in a Focusing and Mixing System Involving

Two Reactive Fluids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136

6.4.2 Theoretical Analysis of Mineral Precipitation

Patterns in a Focusing and Mixing System Involving

Two Reactive Fluids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138

6.4.3 Chemical Reaction Patterns due to Mixing and Focusing

of Two Reactive Fluids in Permeable Fault Zones . . . . . . . . . . 140

6.4.4 Numerical Illustration of Three Types of Chemical Reaction

Patterns Associated with Permeable Fault Zones . . . . . . . . . . . 145

7 An Equivalent Source Algorithm for Simulating Thermal and

Chemical Effects of Intruded Magma Solidiﬁcation Problems . . . . . . . . 153

7.1 An Equivalent Source Algorithm for Simulating

Thermal and Chemical Effects of Intruded Magma

Solidiﬁcation Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

7.2 Implementation of the Equivalent Source Algorithm in the Finite

Element Analysis with Fixed Meshes . . . . . . . . . . . . . . . . . . . . . . . . . . . 160

7.3 Veriﬁcation and Application of the Equivalent Source Algorithm . . . . 163

8 The Particle Simulation Method for Dealing with Spontaneous

Crack Generation Problems in Large-Scale Geological Systems . . . . . . 175

8.1 Basic Formulations of the Particle Simulation Method . . . . . . . . . . . . . 179

8.2 Some Numerical Simulation Issues Associated with the Particle

Simulation Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184

8.2.1 Numerical Simulation Issue Caused by the Difference

between an Element and a Particle . . . . . . . . . . . . . . . . . . . . . . 184

8.2.2 Numerical Simulation Issue Arising from Using the Explicit

Dynamic Relaxation Method to Solve a Quasi-Static Problem 186

8.2.3 Numerical Simulation Issue Stemming from the Loading

Procedure Used in the Particle Simulation Method . . . . . . . . . 189

8.3 An Upscale Theory of Particle Simulation for Two-Dimensional

Quasi-Static Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194

xiv Contents

8.4 Test and Application Examples of the Particle Simulation Method . . . 199

8.4.1 Comparison of the Proposed Loading Procedure

with the Conventional Loading Procedure . . . . . . . . . . . . . . . . 201

8.4.2 The Similarity Test of Two Particle Samples

of Different Length-Scales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204

8.4.3 Particle Simulation of the Folding Process Using Two

Similar Particle Models of Different Length-Scales . . . . . . . . 210

8.4.4 Particle Simulation of the Faulting Process Using

the Proposed Particle Method . . . . . . . . . . . . . . . . . . . . . . . . . . 216

Summary Statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239

Nomenclature

The following symbols are commonly used with the attached deﬁnitions, unless

otherwise speciﬁed in the monograph.

A area of a ﬁnite element

C species concentration

C species concentration vector

C

1

non-zero constant

C

2

arbitrary constant

c

p

speciﬁc heat of pore-ﬂuid

D mass diffusivity

g acceleration due to gravity

H reference length

K medium permeability

K

h

reference medium permeability in the horizontal direction

L length of a problem domain

Le Lewis number

P pressure

P pressure vector

P

0

hydrostatic pressure

q

c

mass ﬂux on the boundary of a ﬁnite element

q

T

heat ﬂux on the boundary of a ﬁnite element

Ra Rayleigh number

Ra

critical

critical Rayleigh number

S boundary length of a ﬁnite element

T temperature

T temperature vector

t temporal variable

u Darcy velocity in the x direction

U Darcy velocity vector

ν Darcy velocity in the y direction

w Darcy velocity in the z direction

x, y, z spatial coordinates

Zh Zhao number

xv

xvi Nomenclature

Zh

critical

critical Zhao number

λ thermal conductivity

λ

e0

reference thermal conductivity in the horizontal direction

φ porosity

ψ stream function

ψ shape function vector for the pressure of a ﬁnite element

ρ

0

reference density of pore-ﬂuid

μ dynamic viscosity of pore-ﬂuid

β thermal volume expansion coefﬁcient of pore-ﬂuid

σ stress on the boundary of a ﬁnite element

ϕ shape function vector for the temperature, species

concentration and Darcy velocity of a ﬁnite element

η permeability ratio of the underlying medium to its

overlying folded layer

thermal conductivity ratio of the underlying medium to its

overlying folded layer

ε penalty parameter associated with the penalty ﬁnite

element approach

Subscripts

f pertaining to pore-ﬂuid

0 pertaining to reference quantities

Superscripts

e pertaining to equivalent quantities of a porous medium

e pertaining to quantities in a ﬁnite element level

∗ pertaining to dimensionless quantities

s pertaining to solid matrix

Chapter 1

Introduction

Geoscience is a fundamental natural science discipline dealing with the origin,

evolutionary history and behaviour of the planet Earth. As a result of its complicated

and complex nature, the Earth system not only provides the necessary materials and

environment for mankind to live, but also brings many types of natural disasters,

such as earthquakes, volcanic eruptions, tsunamis, ﬂoods and tornadoes, to mention

just a few. With the ever-increasing demand for improving our living standards, it

has been recognized that the existing natural resources will be exhausted in the near

future and that our living environments are, in fact, deteriorating. To maintain the

sustainable development of our living standards and the further improvement of our

living environments, an inevitable and challenging task that geoscientists are now

confronting is how accurately to predict not only the occurrences of these natural

disasters, but also the locations of large concealed natural resources in the deep

Earth. For this reason, geoscientists must study the processes, rules and laws, by

which the Earth system operates, instead of simply describing and observing geo-

science phenomena. Speciﬁcally, geoscientists need to make greater efforts in the

following aspects relevant to solving contemporary geoscience problems: (1) the

complicated and complex interactions between multi-scales and multi-processes

occurring in the solid Earth; (2) gather, accumulate and analyze the large amount

of information and data that are essential to understand each of the controlling pro-

cesses within the interior of the Earth using modern observation equipment, mea-

surement tools, experimental instruments and information processing techniques;

(3) the intimate interplay between the solid Earth, biosphere, hydrosphere and atmo-

sphere. It is this intimate interplay that controls the global behaviour of the Earth

system. As a result, geoscientists must adopt scientiﬁc and predictive methods rel-

evant to conduction of contemporary geoscience research, instead of simply using

the traditional descriptive methods.

Computational science is a modern technological science discipline dealing

with the development and application of numerical methods, procedures, algo-

rithms and other numerical techniques for delivering numerical solutions for com-

plicated and complex scientiﬁc and engineering problems. With the rapid advances

and developments of modern computer technology, applications of computa-

tional science have penetrated almost all engineering ﬁelds: from the topological

C. Zhao et al., Fundamentals of Computational Geoscience,

Lecture Notes in Earth Sciences 122, DOI 10.1007/978-3-540-89743-9 1,

C

Springer-Verlag Berlin Heidelberg 2009

1

2 1 Introduction

optimization of a tooth brush to that of a giant aircraft; from the collapse simulation

of a concrete beam to that of a huge double curvature arch dam; from the optimal

layout design of a pipeline to that of a large-scale underground tunnel, to name just

a few. Since computational science is a comprehensive discipline bringing geology,

geophysics, geochemistry, mathematics, physics, chemistry, biology and numerical

techniques together, it can be used effectively and efﬁciently to simulate the pro-

cesses involved in complicated scientiﬁc and engineering problems in a scientiﬁc

and predictive manner. In this sense, computational science is a natural supplier to

meet the demands of geoscientists in solving contemporary geoscience problems.

It is this demand and supply relationship that has created a brand new discipline,

computational geoscience, in the past decade (Zhao et al. 2008a).

1.1 Characteristics of Computational Geoscience

Computational geoscience is a newly-developed discipline, which has been estab-

lished through applying the well-developed computational science discipline to

solve geoscience problems occurring in nature. This means that the computational

geoscience discipline is of multi-disciplinary nature crossing many ﬁelds of science.

The ultimate aim of computational geoscience is to deal with the origin, evolution

and behaviour of the Earth system in a predictive, scientiﬁc manner. Under the stim-

ulus of an ever-increasing demand for natural mineral resources, computational geo-

science has achieved, in the past decade, considerable development driven from the

need to understand the controlling mechanisms behind ore body formation and min-

eralization in hydrothermal and igneous systems within the upper crust of the Earth

(Garven and Freeze 1984, Raffensperger and Garven 1995, Doin et al. 1997, Jiang

et al. 1997, Zhao et al. 1997a, 1998a, Oliver et al. 1999, 2001, Zhao et al. 1999a,

2000a, Hobbs et al. 2000, Gow et al. 2002, Ord et al. 2002, Schaubs and Zhao

2002, Sorjonen-Ward et al. 2002, Zhao et al. 2002a, 2003a, McLellan et al. 2003,

Ord and Sorjonen-Ward 2003, Liu et al. 2005, Sheldon and Ord 2005, Zhao et al.

2005a, 2006a, b, 2007a, 2008a, Zhang et al. 2007, Murphy et al. 2008). As a result, a

fundamental and theoretical framework for the computational geoscience discipline

has been established. This enables many hitherto unsolvable geoscience problems

to be solved, both theoretically and practically, using the newly-developed research

methodology associated with computational geoscience. For instance, some typi-

cal examples of applying the newly-developed research methodology to deal with

geoscience problems are as follows: (1) the convective ﬂow of pore-ﬂuid within the

upper crust of the Earth (Phillips 1991, Nield and Bejan 1992, Zhao et al. 1997a,

1998b, 1999b, 2000b, 2001b, Lin et al. 2003), (2) ore body formation and mineral-

ization within hydrothermal systems (Zhao et al. 1998a, 1999c, 2000c, Gow et al.

2002, Ord et al. 2002, Zhao et al. 2002b, 2003b, 2006c), (3) pore-ﬂuid ﬂow focus-

ing within permeable faults (Obdam and Veling 1987, Zimmerman 1996, Zhao et

al. 1999d, 2006d, e, 2008b, c), (4) ﬂuid-rock/chemical interaction associated with

ore body formation processes (Steefel and Lasaga 1994, Zhao et al. 2001c, 2008d,

e, f) and (5) convective ﬂow of pore-ﬂuid within three-dimensional permeable faults

(Zhao et al. 2003c, d, 2004, 2005b, Yang 2006).

1.2 Basic Steps Associated with the Research Methodology of Computational Geoscience 3

1.2 Basic Steps Associated with the Research Methodology

of Computational Geoscience

Generally speaking, the research methodology of computational geoscience is a

comprehensive research methodology, which is formed by combining ﬁeld obser-

vation, theoretical analysis, numerical simulation and ﬁeld validation. The pri-

mary aim of using this research methodology is to investigate the dynamic pro-

cesses and mechanisms involved in an observed geological phenomenon, rather than

to describe the observed geological phenomenon itself. The appropriate research

methodology of computational geoscience is usually comprised of the following

four main steps: (1) the establishment of a conceptual model for a given geoscience

problem; (2) the establishment of a mathematical model for the given problem; (3)

the construction of a numerical simulation model for the given problem and (4) the

graphical display of the numerical results obtained from the numerical simulation.

1.2.1 The Conceptual Model of a Geoscience Problem

Based on extensive data and information obtained from ﬁeld and laboratory inves-

tigations of a geological phenomenon, a conceptual model is established, which

reﬂects the geometrical architecture and main processes associated with the phe-

nomenon. This is the key step in the process of using the research methodology of

computational geoscience to solve a problem. Due to the multiple processes and

multiple scales involved in a typical geoscience problem, only the major control-

ling dynamic processes and mechanisms associated with the problem need to be

considered during the initial establishment of the conceptual model. Since other

unimportant, or at least less critical, processes and factors are neglected, the initial

conceptual model of the problem is somewhat simpliﬁed. This conceptual model is

used to represent the main characteristics of the real geoscience problem. As under-

standing grows more detail may be added if necessary.

The fundamental principle involved in establishing the conceptual model for the

problemis that the details of the conceptual model should depend on both the length-

scale and time-scale of the problem. The conceptual model cannot be over simpli-

ﬁed, since then it cannot be used effectively to reﬂect the main dynamic processes

and mechanisms of the real problem. On the other hand, the conceptual model can-

not be over complicated, for then unnecessary problems for both the theoretical anal-

ysis and the numerical simulation may arise and it may become difﬁcult to unravel

which parts of the description of the problem are important.

1.2.2 The Mathematical Model of a Geoscience Problem

Using three fundamental principles, namely the conservation of mass, the conser-

vation of momentum and the conservation of energy, as well as the related physical

and chemical laws (Bear 1972, Bear and Bachmat 1990, Phillips 1991, Nield and

Bejan 1992, Zhao et al. 1997a), the conceptual model for the given problem can be

4 1 Introduction

translated into a mathematical model, which is usually comprised of a set of partial

differential equations. Due to the complex and complicated nature of these equa-

tions, it is very difﬁcult, if not impossible, to ﬁnd analytical solutions. Alternatively,

numerical simulation methods need to be used to ﬁnd approximate solutions for the

problem.

To ensure the accuracy and reliability of the numerical simulation solution, it is

necessary to investigate the solution characteristics of the partial differential equa-

tions through a theoretical analysis. For example, some theoretical methods can be

used to investigate the solution singularity and multiple solution characteristics of

the partial differential equations, as well as the conditions under which such char-

acteristics can occur. If possible, a benchmark model should be established for a

particular kind of geoscience problem. The geometrical nature and boundary con-

ditions of this benchmark model can be further simpliﬁed, so that the theoretical

solution, known as the benchmark solution, can be obtained. This benchmark solu-

tion is valuable and indispensable for the veriﬁcation of both the numerical algo-

rithm and the computer code, which are used to solve the problem that is generally

characterised by a complicated geometrical shape and complex material properties.

It must be pointed out that, due to the approximate nature of a numerical method,

the theoretical investigation of the solution characteristics associated with the partial

differential equations of a problem plays an important role in applying the research

methodology of computational geoscience to solve real problems. This is the key

step to ensure the accuracy and reliability of the numerical solution obtained from

the numerical simulation of the problem.

1.2.3 The Numerical Simulation Model of a Geoscience Problem

From a mathematical point of view, the numerical simulation model of a geoscience

problem can be also called the discretized type of mathematical model. Both the

ﬁnite element method and the ﬁnite difference method are commonly-used dis-

cretization methods for numerical simulation of geoscience problems (Zienkiewicz

1977, Zhao et al. 1998a, 2006a). The basic idea behind these numerical methods is

to translate the partial differential equations used to describe the geoscience problem

in a continuum system, into the corresponding algebraic equations in a discretized

system, which in turn is usually comprised of a large number of elements. Through

solving the resulting algebraic equations of the discretized system, a numerical solu-

tion can be obtained for the problem.

Compared with an engineering problem, a geoscience problem commonly has

both large length-scale and large time-scale characteristics. The length-scale of a

geoscience problem is commonly measured in either tens of kilometers or even

hundreds of kilometers, while the time-scale of a typical problem is often measured

in several million years or even several tens of million years. In addition, most geo-

science problems are coupled across both multiple processes and multiple scales.

Due to these signiﬁcant differences between engineering problems and geoscience

problems, commercial computer programs and related algorithms, which are mainly

1.3 The Contextual Arrangements of this Monograph 5

designed for solving engineering problems, cannot be directly used to solve geo-

science problems without modiﬁcation. For this reason, it is necessary either to

develop new computer programs for solving geoscience problems or to modify the

existing commercial computer programs, originally designed for solving engineer-

ing problems, so as to be suitable for these problems.

It is noted that the solution reliability of a geoscience problem is strongly depen-

dent on algorithm convergence, algorithm stability, mesh shape, time-step and other

factors. To ensure the accuracy and reliability of the computational simulation result

for a problem, the above-mentioned factors need to be carefully considered in the

process of establishing the computational simulation model. A newly-developed

computer program needs to be veriﬁed through the corresponding benchmark prob-

lem before it is used to solve any real geoscience problems. Otherwise, the reliabil-

ity of the numerical solution obtained from a newly-developed computer program

cannot be guaranteed.

1.2.4 Graphical Display of the Numerical Simulation Results

The numerical results obtained from the computer simulation of a geoscience prob-

lem are expressed as a large amount of data, which can be viewed using modern

technologies of computer graphical display. By comparing the numerical solution

with ﬁeld observations of the geological phenomenon, the correctness of the estab-

lished conceptual model for the geoscience problem can be tested. Thus the research

methodology of computational geoscience is ﬁrstly established on the basis of ﬁeld

observations, and then goes through theoretical analysis and computational simula-

tion. Finally the results must be tested through comparison with existing or new ﬁeld

observations. This fundamental research methodology requires that the recognition

of a natural phenomenon start from ﬁeld observations, and be completed through

further tests arising from ﬁeld observations, resulting in a circular iteration.

If the numerical solution is not compatible with the ﬁeld observations, then the

established conceptual model of the problem is questionable and therefore needs

to be modiﬁed through further reﬁnement of the natural data. On the contrary, if

the numerical solution is in accord with the ﬁeld observations, then the established

conceptual model of the geoscience problem is a reasonable interpretation of what

may have occurred in nature. In this case, the established conceptual model of the

geoscience problem can be further used to investigate the fundamental rules asso-

ciated with this kind of problem. In this regard, the research methodology of com-

putational geoscience can provide an effective scientiﬁc-judging method for solving

many controversial problems in the ﬁeld of geoscience.

1.3 The Contextual Arrangements of this Monograph

In this monograph we use the ﬁnite element method, the ﬁnite difference method

and the particle simulation method as basic numerical methods for dealing with

geoscience problems. Based on these three methods, we have developed advanced

6 1 Introduction

numerical procedures and algorithms to tackle the large-scale aspects of geoscience

problems. The geoscience problems to be considered are closely related to ore body

formation and mineralization in hydrothermal systems within the upper crust of the

Earth. The arrangements of the forthcoming parts of this monograph are as fol-

lows: In Chap. 2, a progressive asymptotic approach procedure is proposed to solve

steady-state convective pore-ﬂuid ﬂow problems within the upper crust of the Earth.

In combination with the ﬁnite element method, this procedure has been applied to

simulate convective pore-ﬂuid ﬂow that often plays an important role in ore body

formation and mineralization. In Chap. 3, a consistent point-searching interpola-

tion algorithm is proposed to develop a general interface between two commercial

computer codes, Fluid Dynamics Analysis Package (FIDAP, Fluid Dynamics Inter-

national, 1997) and Fast Lagrangian Analysis of Continua (FLAC, Itasca Consulting

Group, 1995). With this general interface, the two commercial codes have been used,

in an iterative and alternative manner, to solve coupled problems between medium

deformation, pore-ﬂuid ﬂow, heat transfer and reactive mass transport processes in

hydrothermal systems. In Chap. 4, a term splitting algorithm is developed for deal-

ing with ﬂuid-rock interaction problems that are closely associated with mineral

dissolution and precipitation as well as metamorphic processes in ﬂuid-saturated

hydrothermal/sedimentary basins of subcritical Zhao numbers. In this case, the

chemical dissolution fronts are stable during their propagation within the reactive

mass transport system. In contrast, a segregated algorithm is proposed, in Chap. 5,

for solving chemical-dissolution front instability problems in ﬂuid-saturated porous

rocks of critical and supercritical Zhao numbers. In this situation, the morphologi-

cal evolution of chemical dissolution fronts in ﬂuid-saturated porous media has been

appropriately simulated. In Chap. 6, a decoupling procedure is proposed for simu-

lating ﬂuids mixing, heat transfer and non-equilibrium redox chemical reactions in

ﬂuid-saturated porous rocks. The proposed procedure has been applied to investi-

gate the effects of non-equilibrium redox chemical reactions on the mineralization

patterns in hydrothermal systems. In Chap. 7, an equivalent source algorithm is pre-

sented for simulating thermal and chemical effects of intruded magma solidiﬁcation

problems. This algorithm has been used to simulate effectively and efﬁciently the

thermal and chemical effects of intruded magma in hydrothermal systems. In Chap.

8, the particle simulation method is extended to solve spontaneous crack generation

problems in brittle rocks within the upper crust of the Earth. The extended particle

method has been applied to simulate spontaneous crack generation associated with

faulting and folding in large length-scale geological systems. Finally, some conclu-

sions are given at the end of the monograph.

Chapter 2

A Progressive Asymptotic Approach Procedure

for Simulating Steady-State Natural Convective

Problems in Fluid-Saturated Porous Media

In a ﬂuid-saturated porous medium, a change in medium temperature may lead to a

change in the density of pore-ﬂuid within the medium. This change can be consid-

ered as a buoyancy force term in the momentum equation to determine pore-ﬂuid

ﬂow in the porous medium using the Oberbeck-Boussinesq approximation model.

The momentum equation used to describe pore-ﬂuid ﬂow in a porous medium is

usually established using Darcy’s law or its extensions. If a ﬂuid-saturated porous

medium has the geometry of a horizontal layer, and is heated uniformly from the

bottom of the layer, then there exists a temperature difference between the top

and bottom boundaries of the layer. Since the positive direction of the tempera-

ture gradient due to this temperature difference is opposite to that of the gravity

acceleration, there is no natural convection for a small temperature gradient in the

porous medium. In this case, heat energy is solely transferred from the high tem-

perature region (the bottom of the horizontal layer) to the low temperature region

(the top of the horizontal layer) by thermal conduction. However, if the temperature

difference is large enough, it may trigger natural convection in the ﬂuid-saturated

porous medium. This problem was ﬁrst treated analytically by Horton and Rogers

(1945) as well as Lapwood (1948), and is often called the Horton-Rogers-Lapwood

problem.

This kind of natural convection problem has been found in many geoscience

ﬁelds. For example, in geoenvironmental engineering, buried nuclear waste and

industrial waste in a ﬂuid-saturated porous medium may generate heat and result

in a temperature gradient in the vertical direction. If the Rayleigh number, which

is directly proportional to the temperature gradient, is equal to or greater than the

critical Rayleigh number, natural convection will take place in the porous medium,

so that the groundwater may be severely contaminated due to the pore-ﬂuid ﬂow cir-

culation caused by the natural convection. In geophysics, there exists a vertical tem-

perature gradient in the Earth’s crust. If this temperature gradient is large enough,

it will cause regional natural convection in the Earth’s crust. In this situation, the

pore-ﬂuid ﬂow circulation due to the natural convection can dissolve soluble min-

erals in some part of a region and carry them to another part of the region. This is

the mineralization problem closely associated with geophysics and geology. Since a

natural porous medium is often of a complicated geometry and composed of many

C. Zhao et al., Fundamentals of Computational Geoscience,

Lecture Notes in Earth Sciences 122, DOI 10.1007/978-3-540-89743-9 2,

C

Springer-Verlag Berlin Heidelberg 2009

7

8 2 Simulating Steady-State Natural Convective Problems

different materials, numerical methods are always needed to solve the aforemen-

tioned problems.

From the mathematical point of view, the Horton-Rogers-Lapwood problem pos-

sesses a bifurcation. The linear stability theory based on the ﬁrst-order perturbation

is commonly used to solve this problem analytically and numerically (Nield 1968,

Palm et al. 1972, Caltagirone 1975, 1976, Combarnous and Bories 1975, Buretta

and Berman 1976, McKibbin and O’Sullivan 1980, Kaviany 1984, Lebon and Cloot

1986, Pillatsis et al. 1987, Riley and Winters 1989, Islam and Nandakumar 1990,

Phillips 1991, Nield and Bejan 1992, Chevalier et al. 1999). However, Joly et al.

(1996) pointed out that: “The linear stability theory, in which the nonlinear term

of the heat disturbance equation has been neglected, does not describe the ampli-

tude of the resulting convection motion. The computed disturbances are correct

only for inﬁnitesimal amplitudes. Indeed, even if the form of convective motion

obtained for low supercritical conditions is often quite similar to the critical distur-

bance, the nonlinear term may produce manifest differences, especially when strong

constraints, such as impervious or adiabatic boundaries, are considered.” Since it is

the amplitude and the form of natural convective motion that signiﬁcantly affects or

dominates the contaminant transport and mineralization in a ﬂuid-saturated porous

medium, there is a deﬁnite need for including the full nonlinear term of the energy

equation in the ﬁnite element analysis.

From the ﬁnite element analysis point of view, the direct inclusion of the full

nonlinear term of the energy equation in the steady-state Horton-Rogers-Lapwood

problem would result in a formidable difﬁculty. The ﬁnite element method needs

to deal with a highly nonlinear problem and often suffers difﬁculties in establish-

ing the true non-zero velocity ﬁeld in a ﬂuid-saturated porous medium because the

Horton-Rogers-Lapwood problem always has a zero solution as one possible solu-

tion for the velocity ﬁeld of the pore-ﬂuid. If the velocity ﬁeld of the pore-ﬂuid

used at the beginning of an iteration method is not chosen appropriately, then the

resulting ﬁnite element solution always tends to zero for the velocity ﬁeld in a

ﬂuid-saturated porous medium. Although this difﬁculty can be circumvented by

turning a steady-state problem into a transient one (Trevisan and Bejan 1987), it

is often unnecessary and computationally inefﬁcient to obtain a steady-state solu-

tion from solving a transient problem. Therefore, it is highly desirable to develop

a numerical procedure to directly solve the steady-state Horton-Rogers-Lapwood

problem. For this reason, a progressive asymptotic approach procedure has been

developed in recent years (Zhao et al. 1997a, 1998a). The developed progressive

asymptotic approach procedure is based on the concept of an asymptotic approach,

which was previously and successfully applied to some other ﬁelds of the ﬁnite ele-

ment method. For instance, the h-adaptive mesh reﬁnement (Cook et al. 1989) is

based on the asymptotic approach concept and can produce a satisfactory solution

with the progressive reduction in the size of ﬁnite elements used in the analysis.

The same asymptotic approach concept was also employed to obtain asymptotic

solutions for natural frequencies of vibrating structures in a ﬁnite element analysis

(Zhao and Steven 1996a, b, c). To solve the steady-state Horton-Rogers-Lapwood

problem with the full nonlinear term of the energy equation included in the ﬁnite

2.1 Governing Equations of the Problem 9

element analysis, the asymptotic approach concept needs to be combined with the

ﬁnite element method in a different fashion (Zhao et al. 1997a).

2.1 Governing Equations of the Problem

For a two-dimensional ﬂuid-saturated porous medium, if Darcy’s law is used to

describe pore-ﬂuid ﬂow and the Oberbeck-Boussinesq approximation is employed

to describe a change in pore-ﬂuid density due to a change in pore-ﬂuid temperature,

the governing equations of a natural convection problem, known as the steady-state

Horton-Rogers-Lapwood problem (Nield and Bejan 1992, Zhao et al. 1997a), for

incompressible pore-ﬂuid can be expressed as

∂u

∂x

+

∂ν

∂y

= 0, (2.1)

u =

K

x

μ

−

∂ P

∂x

+ρ

f

g

x

, (2.2)

v =

K

y

μ

−

∂ P

∂y

+ρ

f

g

y

, (2.3)

ρ

f 0

c

p

u

∂T

∂x

+ν

∂T

∂y

= λ

ex

∂

2

T

∂x

2

+λ

ey

∂

2

T

∂y

2

, (2.4)

ρ

f

= ρ

f 0

[1 −β

T

(T − T

0

)], (2.5)

λ

ex

= φλ

fx

+(1 −φ)λ

sx

, λ

ey

= φλ

fy

+(1 −φ)λ

sy

, (2.6)

where u and v are the horizontal and vertical velocity components of the pore-ﬂuid in

the x and y directions respectively; P is the pore-ﬂuid pressure; T is the temperature

of the porous material; K

x

and K

y

are the permeabilities of the porous material in

the x and y directions respectively; μ is the dynamic viscosity of the pore-ﬂuid; ρ

f

is

the density of the pore-ﬂuid; ρ

f 0

and T

0

are the reference density and temperature;

λ

f x

and λ

sx

are the thermal conductivities of the pore-ﬂuid and rock mass in the x

direction; λ

f y

and λ

sy

are the thermal conductivities of the pore-ﬂuid and rock mass

in the y direction; c

p

is the speciﬁc heat of the pore-ﬂuid; g

x

and g

y

are the gravity

acceleration components in the x and y directions; φ and β

T

are the porosity of the

porous material and the thermal volume expansion coefﬁcient of the pore-ﬂuid.

It is noted that Eqs. (2.1), (2.2), (2.3) and (2.4) are derived under the assumption

that the porous medium considered is orthotropic, in which the y axis is upward in

the vertical direction and coincides with the principal direction of medium perme-

ability as well as that of medium conductivity.

10 2 Simulating Steady-State Natural Convective Problems

In order to simplify Eqs. (2.1), (2.2), (2.3) and (2.4), the following dimensionless

variables are deﬁned:

x

∗

=

x

H

, y

∗

=

y

H

, T

∗

=

T − T

0

ΔT

, (2.7)

u

∗

=

Hρ

f 0

c

p

λ

e0

u, ν

∗

=

Hρ

f 0

c

p

λ

e0

ν, P

∗

=

K

h

ρ

f 0

c

p

μλ

e0

(P − P

0

), (2.8)

K

∗

x

=

K

x

K

h

, K

∗

y

=

K

y

K

h

, λ

∗

ex

=

λ

ex

λ

e0

, λ

∗

ey

=

λ

ey

λ

e0

, (2.9)

where x

∗

and y

∗

are the dimensionless coordinates; u

∗

and v

∗

are the dimensionless

velocity components in the x and y directions respectively; P

∗

and T

∗

are the dimen-

sionless excess pressure and temperature; K

h

is a reference medium permeability

coefﬁcient in the horizontal direction; λ

e0

is a reference conductivity coefﬁcient of

the porous medium; ΔT = T

bottom

− T

0

is the temperature difference between the

bottom and top boundaries of the porous medium; H is a reference length and P

0

is

the static pore-ﬂuid pressure.

Substituting the above dimensionless variables into Eqs. (2.1), (2.2), (2.3) and

(2.4) yields the following dimensionless equations:

∂u

∗

∂x

∗

+

∂ν

∗

∂y

∗

= 0, (2.10)

u

∗

= K

∗

x

−

∂ P

∗

∂x

∗

+ RaT

∗

e

1

, (2.11)

v

∗

= K

∗

y

−

∂ P

∗

∂y

∗

+ RaT

∗

e

2

, (2.12)

u

∗

∂T

∗

∂x

∗

+ν

∗

∂T

∗

∂y

∗

= λ

∗

ex

∂

2

T

∗

∂x

∗2

+λ

∗

ey

∂

2

T

∗

∂y

∗2

, (2.13)

where e is a unit vector and e = e

1

i+e

2

j for a two-dimensional problem; Ra is the

Rayleigh number, deﬁned in this particular case as

Ra =

(ρ

f 0

c

p

)ρ

f 0

gβΔTK

h

H

μλ

e0

. (2.14)

2.2 Finite Element Formulation of the Problem 11

2.2 Finite Element Formulation of the Problem

By considering the dimensionless velocity, pressure and temperature as basic vari-

ables, Eqs. (2.10), (2.11), (2.12) and (2.13) can be discretized using the conventional

ﬁnite element method (Zienkiewicz 1977, Zhao et al. 1997a). For a typical 4-node

quadrilateral element, the velocity, pressure and temperature ﬁelds at the elemental

level can be expressed as

u

∗

(x

∗

, y

∗

) = ϕ

T

U

e

, (2.15)

v

∗

(x

∗

, y

∗

) = ϕ

T

V

e

, (2.16)

P

∗

(x

∗

, y

∗

) = Ψ

T

P

e

, (2.17)

T

∗

(x

∗

, y

∗

) = ϕ

T

T

e

, (2.18)

where U

e

, V

e

, P

e

and T

e

are the column vectors of the nodal velocity, excess pres-

sure and temperature of the element; ϕ is the column vector of the interpolation

functions for the dimensionless velocity and temperature ﬁelds within the element;

Ψ is the column vector of the interpolation functions for the excess pressure within

the element. For the 4-node quadrilateral element, it is assumed that ϕ is identical

to Ψ in the following numerical analysis.

The global coordinate components within the element can be deﬁned as

x

∗

= N

T

X, y

∗

= N

T

Y, (2.19)

where X and Y are the column vectors of nodal coordinate components in the x and

y directions of the global coordinate system respectively; N is the column vector of

the coordinate mapping function of the element. Based on the isoparametric element

concept, the following relationships exist:

N(ξ, η) = ϕ(ξ, η) = Ψ(ξ, η), (2.20)

where ξ and η are the local coordinate components of the element.

Using the Galerkin weighted-residual method, Eqs. (2.10), (2.11), (2.12) and

(2.13) can be expressed, with consideration of Eqs. (2.15), (2.16), (2.17) and (2.18),

as follows:

A

Ψ

∂ϕ

T

∂x

∗

U

e

dA +

A

Ψ

∂ϕ

T

∂y

∗

V

e

dA = 0, (2.21)

A

ϕϕ

T

U

e

dA +

A

ϕK

∗

x

∂Ψ

T

∂x

∗

P

e

dA +

A

ϕK

∗

x

Raϕ

T

T

e

e

1

dA = 0, (2.22)

12 2 Simulating Steady-State Natural Convective Problems

A

ϕϕ

T

V

e

dA +

A

ϕK

∗

y

∂Ψ

T

∂y

∗

P

e

dA +

A

ϕK

∗

y

Raϕ

T

T

e

e

2

dA = 0, (2.23)

A

ϕu

∗

∂ϕ

T

∂x

∗

T

e

dA+

A

ϕv

∗

∂ϕ

T

∂y

∗

T

e

dA−

A

ϕ λ

∗

ex

∂

2

ϕ

T

∂x

∗2

T

e

dA−

A

ϕ λ

∗

ey

∂

2

ϕ

T

∂y

∗2

T

e

dA = 0.

(2.24)

Using the Green-Gauss theorem and the technique of integration by parts, the

terms involving the second derivatives in Eq. (2.24) can be rewritten as

A

ϕ λ

∗

ex

∂

2

ϕ

T

∂x

∗2

T

e

dA = −

A

∂ϕ

∂x

∗

λ

∗

ex

∂ϕ

T

∂x

∗

T

e

dA +

S

ϕq

∗

x

n

x

dS = 0, (2.25)

A

ϕ λ

∗

ey

∂

2

ϕ

T

∂y

∗2

T

e

dA = −

A

∂ϕ

∂y

∗

λ

∗

ey

∂ϕ

T

∂y

∗

T

e

dA +

S

ϕq

∗

y

n

y

dS = 0, (2.26)

where q

∗

x

and q

∗

y

are the dimensionless heat ﬂuxes on the element boundary of a unit

normal vector, n; A and S are the area and boundary length of the element.

Note that Eqs. (2.21), (2.22), (2.23) and (2.24) can be expressed in a matrix form

as follows:

⎡

⎢

⎢

⎣

M

e

0 −B

e

x

−A

e

x

0 M

e

−B

e

y

−A

e

y

0 0 E

e

0

C

e

x

C

e

y

0 0

⎤

⎥

⎥

⎦

⎧

⎪

⎪

⎨

⎪

⎪

⎩

U

e

V

e

T

e

P

e

⎫

⎪

⎪

⎬

⎪

⎪

⎭

=

⎧

⎪

⎪

⎨

⎪

⎪

⎩

F

e

x

F

e

y

G

e

0

⎫

⎪

⎪

⎬

⎪

⎪

⎭

, (2.27)

where U

e

and V

e

are the nodal dimensionless velocity vectors of the element in the

x and y directions respectively; T

e

and P

e

are the nodal dimensionless temperature

and pressure vectors of the element; A

e

x

, A

e

y

, B

e

x

, B

e

y

C

e

x

, C

e

y

, E

e

and M

e

are the

property matrices of the element; F

e

x

, F

e

y

and G

e

are the dimensionless nodal load

vectors due to the dimensionless stress and heat ﬂux on the boundary of the element.

These matrices and vectors can be derived and expressed as follows:

A

e

x

=

A

∂ϕ

∂x

∗

K

∗

x

Ψ

T

dA, B

e

x

=

A

ϕK

∗

x

Raϕ

T

e

1

dA, C

e

x

=

A

Ψ

∂ϕ

T

∂x

∗

dA,

(2.28)

A

e

y

=

A

∂ϕ

∂y

∗

K

∗

y

Ψ

T

dA, B

e

y

=

A

ϕK

∗

y

Raϕ

T

e

2

dA, C

e

y

=

A

Ψ

∂ϕ

T

∂y

∗

dA,

(2.29)

D

e

x

(u

∗

) =

A

ϕu

∗

∂ϕ

T

∂x

∗

dA, L

e

x

=

A

∂ϕ

∂x

∗

λ

∗

ex

∂ϕ

∂x

∗

dA, F

e

x

=

S

σ

∗

x

ϕdS,

(2.30)

2.2 Finite Element Formulation of the Problem 13

D

e

y

(v

∗

) =

A

ϕν

∗

∂ϕ

T

∂x

∗

dA, L

e

y

=

A

∂ϕ

∂y

∗

λ

∗

ey

∂ϕ

∂y

∗

dA, F

e

y

=

S

σ

∗

y

ϕdS,

(2.31)

E

e

= D

e

x

(u

∗

) +D

e

y

(v

∗

) +L

e

x

+L

e

y

, G

e

= −

S

q

∗

ϕdS, (2.32)

M

e

=

A

ϕϕ

T

dA, q

∗

=

H

ΔTλ

e0

q, σ

∗

=

K

h

ρ

f 0

c

p

μλ

e0

σ, (2.33)

where ϕ is the shape function vector for the temperature and velocity components

of the element; Ψ is the shape function vector for the pressure of the element; σ and

q are the stress and heat ﬂux on the boundary of the element; A and S are the area

and boundary length of the element.

It is noted that since the full nonlinear term of the energy equation in the Horton-

Rogers-Lapwood problem is considered in the ﬁnite element analysis, matrix E

e

is

dependent on the velocity components of the element. Thus, a prediction for the ini-

tial velocities of an element is needed to have this matrix evaluated. This is the main

motivation for proposing a progressive asymptotic approach procedure in the next

section.

From the penalty ﬁnite element approach (Zienkiewicz 1977), the following

equation exists:

C

e

x

U

e

+C

e

y

V

e

= −εM

p

P

e

. (2.34)

Equation (2.34) can be rewritten as

P

e

= −

1

ε

M

−1

p

(C

e

x

U

e

+C

e

y

V

e

). (2.35)

Substituting Eq. (2.35) into Eq. (2.27) yields the following equation in the ele-

mental level:

¸

Q

e

−B

e

0 E

e

¸ ¸

U

e

F

T

e

¸

=

¸

F

e

G

e

¸

, (2.36)

where

Q

e

= M

e

+

1

ε

A

e

(M

e

p

)

−1

(C

e

)

T

, (2.37)

M

e

=

¸

M

e

0

0 M

e

¸

, U

e

F

=

¸

U

e

V

e

¸

, F

e

=

¸

F

e

x

F

e

y

¸

, (2.38)

B

e

=

¸

B

e

x

B

e

y

¸

, A

e

=

¸

A

e

x

A

e

y

¸

, C

e

=

¸

C

e

x

C

e

y

¸

, (2.39)

14 2 Simulating Steady-State Natural Convective Problems

M

e

p

=

A

ΨΨ

T

dA. (2.40)

It needs to be pointed out that ε is a penalty parameter in Eq. (2.37). For the pur-

pose of obtaining an accurate solution, this parameter must be chosen small enough

to approximate ﬂuid incompressibility well, but large enough to prevent the result-

ing matrix problem from becoming too ill-conditioned to solve.

By assembling all elements in a system, the ﬁnite element equation of the system

can be expressed in a matrix form as

¸

Q −B

0 E(U)

¸ ¸

U

F

T

¸

=

¸

F

G

¸

, (2.41)

where Q, B and E are global property matrices of the system; U

F

and T are global

nodal velocity and temperature vectors of the system; F and G are global nodal load

vectors of the system. Since Equation (2.41) is nonlinear, either the successive sub-

stitution method or the Newton-Raphson method can be used to solve this equation.

2.3 The Progressive Asymptotic Approach Procedure

for Solving Steady-State Natural Convection Problems

in Fluid-Saturated Porous Media

To solve the steady-state Horton-Rogers-Lapwood problem with the full nonlinear

term of the energy equation included in the ﬁnite element analysis, the asymptotic

approach concept (Cook et al. 1989, Zhao and Steven 1996a, b, c) needs to be used in

a progressive fashion (Zhao et al. 1997a). If the gravity acceleration is assumed to tilt

at a small angle, α, in the Horton -Rogers-Lapwood problem, then a non-zero veloc-

ity ﬁeld in a ﬂuid-saturated porous medium may be found using the ﬁnite element

method. The resulting non-zero velocity ﬁeld can be used as the initial velocity ﬁeld

of the pore-ﬂuid to solve the original Horton-Rogers-Lapwood problem with the

tilted small angle being zero. Thus, two kinds of problems need to be progressively

solved in the ﬁnite element analysis. One is the modiﬁed Horton-Rogers-Lapwood

problem, in which the gravity acceleration is tilted a small angle, and another is

the original Horton-Rogers-Lapwood problem. This forms two basic steps of the

progressive asymptotic approach procedure. Clearly, the basic idea behind the pro-

gressive asymptotic approach procedure is that when the small angle tilted by the

gravity acceleration approaches zero, the modiﬁed Horton-Rogers-Lapwood prob-

lem asymptotically approaches the original one and as a result, a solution to the

original Horton-Rogers-Lapwood problem can be obtained.

Based on the basic idea behind the progressive asymptotic approach procedure,

the key issue of obtaining a non-zero pore-ﬂuid ﬂow solution for the Horton-Rogers-

Lapwood problem is to choose the initial velocity ﬁeld of pore-ﬂuid correctly. If the

initial velocity ﬁeld is not correctly chosen, the ﬁnite element method will lead to

2.3 Solving Steady-State Natural Convection Problems in Fluid-Saturated Porous Media 15

a zero pore-ﬂuid ﬂow solution for natural convection of pore-ﬂuid, even though

the Rayleigh number is high enough to drive the occurrence of natural convection

in a ﬂuid-saturated porous medium. In order to overcome this difﬁculty, a modi-

ﬁed Horton-Rogers-Lapwood problem, in which the gravity acceleration is assumed

to tilt a small angle α, needs to be solved. Supposing the original Horton-Rogers-

Lapwood problem has a Rayleigh number (Ra) and that the non-zero solution for the

modiﬁed Horton-Rogers-Lapwood problem is S(Ra, α), it is possible to ﬁnd a non-

zero solution for the original Horton-Rogers-Lapwood problem by taking a limit of

S(Ra, α) when α approaches zero. This process can be mathematically expressed

as follows:

lim

α→0

S(Ra, α) = S(Ra, 0), (2.42)

where S(Ra, 0) is a solution for the original Horton-Rogers-Lapwood problem;

S(Ra, α) is the solution for the modiﬁed Horton-Rogers-Lapwood problem; S is

any variable to be solved in the original Horton-Rogers-Lapwood problem.

It is noted that in theory, if S(Ra, α) could be expressed as a function of α

explicitly, S(Ra, 0) would follow immediately. However, in practice, it is necessary

to ﬁnd out S(Ra, 0) numerically since it is very difﬁcult and often impossible to

express S(Ra, α) in an explicit manner. Thus, the question which must be answered

is how to choose α so as to obtain an accurate non-zero solution, S(Ra, 0). From

the theoretical point of view, it is desirable to choose α as small as possible. The

reason for this is that the smaller the value of α, the closer the characteristic of

S(Ra, α) to that of S(Ra, 0). This enables a more accurate solution S(Ra, 0) to

be obtained in the computation. From the ﬁnite element analysis point of view, α

cannot be chosen too small because the smaller the value of α, the more sensitive

the solution S(Ra, α) to the initial velocity ﬁeld of pore-ﬂuid. As a result, a very

small α usually leads to a zero velocity ﬁeld due to any inappropriate choice for the

initial velocity ﬁeld of pore-ﬂuid. To avoid this phenomenon, α should be chosen

big enough to eliminate the strong dependence of S(Ra, α) on the initial velocity

ﬁeld of pore-ﬂuid. For the purpose of using a big value of α and keeping the ﬁnal

solution S(Ra, 0) of good accuracy in the ﬁnite element analysis, S(Ra, α) needs to

approach S(Ra, 0) in a progressive asymptotic manner, as clearly shown in Fig. 2.1.

This leads to the following processes mathematically:

lim

α

i

→α

i +1

S(Ra, α

i

) = S(Ra, α

i +1

) (i = 1, 2, ......, n −1), (2.43)

lim

α

n

→0

S(Ra, α

n

) = S(Ra, 0), (2.44)

α

1

= α, α

i +1

=

1

R

α

i

, (2.45)

where n is the total step number for α approaching zero; R is the rate of α

i

approach-

ing α

i +1

. Generally, the values of α,n and R are dependent on the nature of a problem

to be analysed.

16 2 Simulating Steady-State Natural Convective Problems

Ra

critical

Ra

0

α is too small

α = α

1

α = α

2

α = α

n

S(Ra, α)

Fig. 2.1 The basic concept of the progressive asymptotic approach procedure

For solving the steady-state Horton-Rogers-Lapwood problem using the pro-

gressive asymptotic approach procedure associated with the ﬁnite element method,

numerical experience has shown that 1

◦

≤ α ≤ 5

◦

, 5 ≤ R ≤ 10 and 1 ≤ n ≤ 2

leads to acceptable solutions. Therefore, for α in the range of 1–5

◦

and R in the

range of 5–10, S(Ra, α) can asymptotically approach S(Ra, 0) in one step or two

steps. This indicates the efﬁciency of the present procedure.

2.4 Derivation of Analytical Solution to a Benchmark Problem

In order to verify the applicability of the progressive asymptotic approach procedure

for solving the Horton-Rogers-Lapwood convection problem, an analytical solution

is needed for a benchmark problem, the geometry and boundary conditions of which

can be exactly modelled by the ﬁnite element method. Although the existing solu-

tions (Phillips 1991, Nield and Bejan 1992) for a horizontal layer in porous media

can be used to check the accuracy of a ﬁnite element solution within a square box

with appropriate boundary conditions, it is highly desirable to examine the progres-

sive asymptotic approach procedure as extensively as possible. For this purpose,

a benchmark problem of any rectangular geometry is constructed and shown in

Fig. 2.2. Without losing generality, the dimensionless governing equations given

in Eqs. (2.10), (2.11), (2.12) and (2.13) are considered in this section. The boundary

conditions of the benchmark problem are expressed using the dimensionless vari-

ables as follows:

u

∗

= 0,

∂T

∗

∂x

∗

= 0 (at x

∗

= 0 and x

∗

= L

∗

), (2.46)

v

∗

= 0, T

∗

= 1 (at y

∗

= 0), (2.47)

v

∗

= 0, T

∗

= 0 (at y

∗

= 1), (2.48)

2.4 Derivation of Analytical Solution to a Benchmark Problem 17

1

L

*

y

*

0

x

*

Fig. 2.2 Geometry of a benchmark problem

where L

∗

is a dimensionless length in the horizontal direction and L

∗

= L/H, in

which L is the real length of the problem domain in the horizontal direction.

For ease of deriving an analytical solution to the benchmark problem, it

is assumed that the porous medium under consideration is ﬂuid-saturated and

isotropic. This means that K

x

= K

y

= K

h

and λ

ex

= λ

ey

= λ

e0

. As a result,

Eqs. (2.10), (2.11), (2.12) and (2.13) can be further simpliﬁed as follows:

∂u

∗

∂x

∗

+

∂ν

∗

∂y

∗

= 0, (2.49)

u

∗

= −

∂ P

∗

∂x

∗

+ RaT

∗

e

1

, (2.50)

v

∗

= −

∂ P

∗

∂y

∗

+ RaT

∗

e

2

, (2.51)

u

∗

∂T

∗

∂x

∗

+ν

∗

∂T

∗

∂y

∗

=

∂

2

T

∗

∂x

∗2

+

∂

2

T

∗

∂y

∗2

. (2.52)

Using the linearization procedure for temperature gradient and a dimensionless

stream function Ψ simultaneously, Eqs. (2.49), (2.50), (2.51) and (2.52) are reduced

to the following two equations:

∂

2

Ψ

∂x

∗2

+

∂

2

Ψ

∂y

∗2

= −Ra

∂T

∗

∂x

∗

, (2.53)

∂Ψ

∂x

∗

=

∂

2

T

∗

∂x

∗2

+

∂

2

T

∗

∂y

∗2

. (2.54)

Since Eqs. (2.53) and (2.54) are linear, solutions to Ψand T

∗

are of the following

forms:

Ψ = f (y

∗

) sin

q

x

∗

L

∗

(q = mπ, m = 1, 2, 3, ......), (2.55)

18 2 Simulating Steady-State Natural Convective Problems

T

∗

= θ(y

∗

) cos

q

x

∗

L

∗

+(1 − y

∗

) (q = mπ, m = 1, 2, 3, ......). (2.56)

Substituting Eqs. (2.55) and (2.56) into Eqs. (2.53) and (2.54) yields the follow-

ing equations:

f

(y

∗

) −

q

L

∗

2

f (y

∗

) =

q

L

∗

Raθ(y

∗

), (2.57)

q

L

∗

f (y

∗

) = −

q

L

∗

2

θ(y

∗

) +θ

(y

∗

). (2.58)

Combining Eqs. (2.57) and (2.58) leads to an equation containing f (y

∗

) only:

f

I V

(y

∗

) −2

q

L

∗

2

f

(y

∗

) −

q

L

∗

2

¸

Ra −

q

L

∗

2

¸

f (y

∗

) = 0. (2.59)

It is immediately noted that Equation (2.59) is a linear, homogeneous ordinary

differential equation so that it has a zero trivial solution. For the purpose of ﬁnd-

ing out a non-zero solution, it is noted that the non-zero solution satisfying both

Equation (2.59) and the boundary conditions in Eqs. (2.46), (2.47) and (2.48) can

be expressed as

f (y

∗

) = sin(r y

∗

) (r = nπ, n = 1, 2, 3, ......). (2.60)

Using this equation, the condition under which the non-zero solution exists for

Eq. (2.59) is derived and expressed as

Ra =

L

∗

q

r

2

+

q

L

∗

2

=

n

2

m

L

∗

+

m

L

∗

2

π

2

(m = 1, 2, 3, ......, n = 1, 2, 3, ......).

(2.61)

It can be observed from Eq. (2.61) that in the case of L

∗

being an integer, the

minimum Rayleigh number is 4π

2

, which occurs when n = 1 and m = L

∗

. How-

ever, if L

∗

is not an integer, the minimum Rayleigh number is (L

∗

+ 1

/

L

∗

)

2

π

2

,

which occurs when m = 1 and n = 1. Since the minimum Rayleigh number deter-

mines the onset of natural convection in a ﬂuid-saturated porous medium for the

Horton-Rogers-Lapwood problem, it is often labelled as the critical Rayleigh num-

ber, Ra

critical

.

For this benchmark problem, the mode shapes for the stream function and

related dimensionless variables corresponding to the critical Rayleigh number can

be derived and expressed as follows:

Ψ = C

1

sin

mπ

L

∗

x

∗

sin(nπy

∗

), (2.62)

2.5 Veriﬁcation of the Proposed Progressive Asymptotic Approach 19

u

∗

= nπC

1

sin

mπ

L

∗

x

∗

cos(nπy

∗

), (2.63)

v

∗

= −

mπ

L

∗

C

1

cos

mπ

L

∗

x

∗

sin(nπy

∗

), (2.64)

T

∗

= −

C

1

√

Ra

critical

cos

mπ

L

∗

x

∗

sin(nπy

∗

) +(1 − y

∗

), (2.65)

P

∗

=

nL

∗

m

C

1

cos

mπ

L

∗

x

∗

cos(nπy

∗

) −

Ra

critical

2

(1 − y

∗

)

2

+C

2

, (2.66)

where the values of m, n and Ra

critical

are dependent on whether L

∗

is an integer

or not; C

1

is a non-zero constant and C

2

is an arbitrary constant. It is interesting

to note that since Ra

critical

is a function of L

∗

, it can vary with a non-integer L

∗

.

This implies that if rectangular valleys are ﬁlled with porous media, they may have

different critical Rayleigh numbers when their ratios of length to height are different.

2.5 Veriﬁcation of the Proposed Progressive Asymptotic

Approach Procedure Associated with Finite

Element Analysis

Using the analytical solution derived for a benchmark problemin the last section, the

proposed progressive asymptotic approach procedure associated with the ﬁnite ele-

ment analysis for solving the Horton-Rogers-Lapwood problem in a ﬂuid-saturated

porous medium is veriﬁed in this section. A rectangular domain of L

∗

= 1.5 is con-

sidered in the calculation. The critical Rayleigh number for the test problem con-

sidered is 169π

2

/

36. As shown in Fig. 2.3, the problem domain is discretized into

v

*

=

0,T

*

=

1

v

*

=

0,T

*

=

0

0 =

L

*

=

1.5

0 =

∂n

*

∂T

∂n

*

∂T

u

*

= 0

u

*

= 0

1

*

y

*

x

0

Fig. 2.3 Finite element mesh for the benchmark problem

20 2 Simulating Steady-State Natural Convective Problems

864 nine-node quadrilateral elements of 3577 nodes in total. The mesh gradation

technique, which enables the region in the vicinity of problem boundaries to be

modelled using ﬁnite elements of small sizes, has been employed to increase the

solution accuracy in this region. The following parameters associated with the pro-

gressive asymptotic approach procedure are used in the calculation: α = 5

o

, n = 2

and R = 5.

Figures 2.4, 2.5, 2.6 and 2.7 show the comparison of numerical solutions with

analytical ones for dimensionless velocity, stream function, temperature and pres-

sure modes respectively. In these ﬁgures, the plots above are analytical solutions,

whereas the plots below are numerical solutions for the problem. It is observed from

these results that the numerical solutions from the progressive asymptotic approach

procedure associated with the ﬁnite element method are in good agreement with the

analytical solutions. Compared with the analytical solutions, the maximum error in

the numerical solutions is less than 2%. This demonstrates the usefulness of the

present progressive asymptotic approach procedure when it is used to solve the

steady-state Horton-Rogers-Lapwood problems.

At this point, there is a need to explain why both the analytical and the numer-

ical solutions for the pore-ﬂuid ﬂow are non-symmetric, although the geometry

and boundary conditions for the problem are symmetric. As stated previously, the

Horton-Rogers-Lapwood problembelongs mathematically to a bifurcation problem.

The trivial solution for the pore-ﬂuid ﬂow of the problem is zero. That is to say, if

the Rayleigh number of the problem is less than the critical Rayleigh number, the

solution resulting from any small disturbance or perturbation converges to the trivial

Fig. 2.4 Comparison of numerical solution with analytical solution (Dimensionless velocity)

2.5 Veriﬁcation of the Proposed Progressive Asymptotic Approach 21

Fig. 2.5 Comparison of numerical solution with analytical solution (Dimensionless stream

function)

solution. In this case, the solution for the pore-ﬂuid ﬂowis zero (and, of course, sym-

metric) and the system is in a stable state. However, if the Rayleigh number of the

problem is equal to or greater than the critical Rayleigh number, the solution result-

ing from any small disturbance or perturbation may lead to a non-trivial solution.

In this situation, the solution for the pore-ﬂuid ﬂow is non-zero and the system is in

an unstable state. Since the main purpose of this study is to ﬁnd out the non-trivial

solution for problems having a high Rayleigh number, Ra ≥ Ra

critical

, a small dis-

turbance or perturbation needs to be applied to the system at the beginning of a

computation. This is why gravity is ﬁrstly tilted a small angle away from vertical

and then gradually approaches and is ﬁnally restored to vertical in the proposed pro-

gressive asymptotic approach procedure. It is the small perturbation that makes the

non-trivial solution non-symmetric, even though the system considered is symmet-

ric. In addition, as addressed in Sect. 2.3, the solution dependence on the amplitude

of the initially-tilted small angle can be avoided by making this angle approach zero

22 2 Simulating Steady-State Natural Convective Problems

Fig. 2.6 Comparison of numerical solution with analytical solution (Dimensionless temperature)

in a progressive asymptotic manner. However, since there are two possible non-

symmetric solutions for convective pore-ﬂuid ﬂow, namely a clockwise convective

ﬂow and an anti-clockwise convective ﬂow, the solution dependence on the direction

of the initially-tilted small angle cannot be avoided.

2.6 Application of the Progressive Asymptotic Approach

Procedure Associated with Finite Element Analysis

2.6.1 Two-Dimensional Convective Pore-Fluid Flow Problems

The present progressive asymptotic approach procedure is employed to investi-

gate the effect of basin shapes on natural convection in a ﬂuid-saturated porous

medium when it is heated from below. Three different basin shapes having square,

2.6 Application of the Progressive Asymptotic Approach Procedure 23

Fig. 2.7 Comparison of numerical solution with analytical solution (Dimensionless pressure)

rectangular and trapezoidal geometries, which are ﬁlled with ﬂuid-saturated porous

media, are considered in the analysis. For the rectangular basin, the ratio of width to

height is 1.5. For the trapezoidal basin, the ratios of top width to height and bottom

width to height are 2 and 1 respectively. In order to reﬂect the anisotropic behaviour

of the porous media, the medium permeability in the horizontal direction is assumed

to be three times that in the vertical direction. As shown in Fig. 2.8, all three basins

are discretized into 484 nine-node quadrilateral elements of 2041 nodes in total.

The boundary conditions of the problems are also shown in Fig. 2.8, in which n is

the normal direction of a boundary. Two Rayleigh numbers, namely Ra = 80 and

Ra =400, are used to examine the effect of the Rayleigh number on natural convec-

tion in a ﬂuid-saturated porous medium. The same parameters as used in the above

model veriﬁcation examples have been used here for the progressive asymptotic

approach procedure.

Figure 2.9 shows the dimensionless velocity distribution for the three different

basins, whereas Figs. 2.10 and 2.11 show the dimensionless streamline contours

24 2 Simulating Steady-State Natural Convective Problems

0 =

∂n

∂T

*

u

*

=

0

v

*

=

0

1

1

1

0 =

∂n

∂T

*

u

*

=

0

v

*

=

0

0 =

∂n

∂T

*

u

*

=

0

L

*

=

1

L

*

=

1.5

L

*

=

2

u

*

=

0, v

*

=

0, T

*

=

1

u

*

=

0, v

*

=

0, T

*

=

1

L

*

=

1

u

*

=

0, v

*

=

0, T

*

=

1

v

*

=

0

0 =

∂n

∂T

*

u

*

=

0

v

*

=

0

0 =

∂n

∂T

*

u

*

=

0

v

*

=

0

0 =

∂n

∂T

*

u

*

=

0

v

*

=

0

T

*

=

0

T

*

=

0

T

*

=

0

Fig. 2.8 Finite element meshes for three different basins shapes

2.6 Application of the Progressive Asymptotic Approach Procedure 25

Fig. 2.9 Dimensionless velocity distribution for different basins

due to different basin shapes for Ra = 80 and Ra = 400 respectively. It is obvi-

ous that different basin shapes have a considerable effect on the patterns of con-

vective ﬂow in the ﬂuid-saturated porous medium, especially in the case of higher

Rayleigh numbers. Apart from notable differences in velocity distribution patterns,

maximum velocity amplitudes for three different basins are also signiﬁcantly differ-

ent. For instance, in the case of Ra =80, the maximum amplitudes of dimensionless

velocities are 5.29, 8.93 and 11.66 for square, rectangular and trapezoidal basins

respectively. This fact indicates that different basin shapes may affect the contami-

nant transport or mineralization processes in a ﬂuid-saturated porous medium once

natural convection is initiated in the medium.

26 2 Simulating Steady-State Natural Convective Problems

Fig. 2.10 Dimensionless streamline contours for different basins (Ra = 80)

2.6 Application of the Progressive Asymptotic Approach Procedure 27

Fig. 2.11 Dimensionless streamline contours for different basins (Ra = 400)

28 2 Simulating Steady-State Natural Convective Problems

2.6.2 Three-Dimensional Convective Pore-Fluid Flow Problems

The proposed progressive asymptotic approach procedure can be straightforwardly

extended to the simulation of three-dimensional convective pore-ﬂuid ﬂowproblems

in ﬂuid-saturated porous media (Zhao et al. 2001a, 2003a). Since the streamline

function is not available for three dimensional ﬂuid ﬂow problems, it is necessary

to use the particle tracking technique to show particle movements in three dimen-

sional ﬂuid ﬂow systems. In the particle tracking technique, a fundamental problem,

which needs to be solved effectively, is as follows. If the initial (known) location of

a particle is point A (x

A

, y

A

, z

A

), where x

A

, y

A

and z

A

are the coordinate compo-

nents of point A in the x, y and z directions of a global coordinate system, then we

need to determine where the new location (i.e. point A

**) of this particle is after a
**

given time interval, Δt . Clearly, if the velocity of the particle at point A is known,

then the coordinate components of point A

**in the global coordinate system can be
**

approximately determined for a small Δt as follows:

x

A

= x

A

+u

A

Δt, (2.67)

y

A

= y

A

+v

A

Δt, (2.68)

z

A

= z

A

+w

A

Δt, (2.69)

where x

A

, y

A

and z

A

are the coordinate components of point A

in the x, y and

z directions of the global coordinate system; u

A

, v

A

and w

A

are the velocity com-

ponents of point A in the x, y and z directions of the global coordinate system,

respectively.

In general cases, the location of point A is not coincident with the nodal points

in a ﬁnite element analysis so that the consistent interpolation of the ﬁnite element

solution is needed to determine the velocity components at this point. For this pur-

pose, it is essential to ﬁnd the coordinate components of point A in the local coordi-

nate system from the following equations for an isoparametric ﬁnite element.

x

A

=

n

¸

i =1

φ

i

(ξ

A

, η

A

, ζ

A

)x

i

, (2.70)

y

A

=

n

¸

i =1

φ

i

(ξ

A

, η

A

, ζ

A

)y

i

, (2.71)

z

A

=

n

¸

i =1

φ

i

(ξ

A

, η

A

, ζ

A

)z

i

, (2.72)

where x

A

, y

A

and z

A

are the coordinate components of point A in the x, y and z direc-

tions of the global coordinate system; ξ

A

, η

A

and ζ

A

are three coordinate components

2.6 Application of the Progressive Asymptotic Approach Procedure 29

of point A in the ξ, η and ζ directions of a local coordinate system; x

i

, y

i

and z

i

are

the coordinate components of nodal point i in the x, y and z directions of the global

coordinate system; n is the total nodal number of the element containing point A; φ

i

is the interpolation function of node i in the element containing point A.

In Eqs. (2.70), (2.71) and (2.72), the coordinate components of point A in the

global coordinate system are known, so that the coordinate components of this point

in the local system can be determined using any inverse mapping technique (Zhao

et al. 1999f). Once the coordinate components of point A in the global coordinate

system are determined, the velocity components of point A in the global system can

be straightforwardly calculated as follows:

u

A

=

n

¸

i =1

φ

i

(ξ

A

, η

A

, ζ

A

)u

i

(2.73)

v

A

=

n

¸

i =1

φ

i

(ξ

A

, η

A

, ζ

A

)v

i

(2.74)

w

A

=

n

¸

i =1

φ

i

(ξ

A

, η

A

, ζ

A

)w

i

, (2.75)

where u

A

, v

A

and w

A

are the velocity components of point A in the x, y and z direc-

tions of the global coordinate system; u

i

, v

i

and w

i

are the velocity components of

nodal point i in the x, y and z directions of the global coordinate system, respectively.

The above-mentioned process indicates that in the ﬁnite element analysis of ﬂuid

ﬂow problems, the trajectory of any given particle can be calculated using the nodal

coordinate and velocity components, which are fundamental quantities and therefore

available in the ﬁnite element analysis.

To demonstrate the applicability of the progressive asymptotic approach proce-

dure for simulating convective pore-ﬂuid ﬂow in three dimensional situations, the

example considered in this section is a cubic box of 10 ×10 ×10 km

3

in size. This

box is ﬁlled with pore-ﬂuid saturated porous rock, which is a part of the upper crust

of the Earth. In order to simulate geothermal conditions in geology, the bottomof the

box is assumed to be hotter than the top of the box. This means that the pore-ﬂuid

saturated porous rock is uniformly heated from below. For the system considered

here, the classical analysis (Phillips 1991, Nield and Bejan 1992, Zhao et al. 1997a)

indicates that the convective ﬂow is possible when the Rayleigh number of the sys-

tem is either critical or supercritical. For this reason, the parameters and properties

of the system are deliberately selected in such a way that the Rayleigh number of

the system is supercritical.

Figure 2.12 shows the ﬁnite element mesh of 8000 cubic elements for the three

dimensional convective ﬂow problem. For the purpose of investigating the pertur-

bation direction on the pattern of convective ﬂow, two cases are considered in the

following computations. In the ﬁrst case, the perturbation of gravity is applied in the

x-z plane only. This means that the problemis axisymmetrical about the y axis so that

30 2 Simulating Steady-State Natural Convective Problems

Fig. 2.12 Finite element mesh for the three-dimensional problem

the problem can be degenerated into a two dimensional problem, from the mathe-

matical and analytical points of view. Since the solutions for the axisymmetrical

convective ﬂow problem are available (Nield and Bejan 1992, Zhao et al. 1997a), the

numerical methods used in this study can be veriﬁed by comparing the related ana-

lytical solutions with the solutions obtained from this special three dimensional case

(i.e. axisymmetrical case). In the second case, the perturbation of gravity is equally

applied in both the x-z and y-z planes. This means that a true three dimensional con-

vective ﬂow problem is considered in this case. Table 2.1 shows the parameters used

in the computations for both cases. To reﬂect the three-dimensional features of con-

vective pore-ﬂuid ﬂow, the following boundary conditions are used. Temperatures at

Table 2. 1 Parameters used for the three-dimensional convective ﬂow problem

Material type Parameter Value

pore-ﬂuid dynamic viscosity 10

−3

N ×s/m

2

reference density 1000 kg/m

3

volumetric thermal expansion coefﬁcient 2.07 ×10

−4

1/

◦

C

speciﬁc heat 4185 J/(kg×

0

C)

thermal conductivity coefﬁcient 0.6 W/(m×

0

C)

porous matrix porosity 0.1

permeability 10

−14

m

2

Speciﬁc heat 815 J/(kg×

0

C)

thermal conductivity coefﬁcient 3.35 W/(m×

0

C)

2.6 Application of the Progressive Asymptotic Approach Procedure 31

the top and the bottom of the computational domain are 0

◦

C and 250

◦

C respectively.

Both the top and the bottom of the computational domain are impermeable in the

vertical direction, while all the four side boundaries of the computational domain

are assumed to be insulated and impermeable in the horizontal direction.

Figures 2.13 and 2.14 show the numerical and analytical solutions for the dis-

tributions of pore-ﬂuid velocity and temperature in the axisymmetrical case (i.e.

case 1) respectively. As expected, the numerical solutions for both pore-ﬂuid veloc-

ity and temperature in this case are exactly axisymmetrical about the y axis. This

implies that the three-dimensional problem considered in this particular case can be

reasonably treated as a two-dimensional one. It is also observed that the numerical

solutions in Fig. 2.13 compare very well with the previous solutions in Fig. 2.14

Fig. 2.13 Distributions of pore-ﬂuid velocity and temperature in the porous medium (Case 1,

numerical solutions)

32 2 Simulating Steady-State Natural Convective Problems

Fig. 2.14 Distributions of pore-ﬂuid velocity and temperature in the porous medium (Case 1,

Analytical solutions)

for the axisymmetrical convective pore-ﬂuid ﬂow problem (Zhao et al. 1997a). This

indicates that the progressive asymptotic approach procedure, although it was pre-

viously developed for the ﬁnite element modelling of two-dimensional convective

pore-ﬂuid ﬂow problems, is equally applicable to the ﬁnite element modelling of

three-dimensional convective pore-ﬂuid ﬂow in ﬂuid-saturated porous media when

they are heated from below.

2.6 Application of the Progressive Asymptotic Approach Procedure 33

Fig. 2.15 Distributions of pore-ﬂuid velocity and temperature in the porous medium (Case 2)

Figure 2.15 shows the numerical solutions for the distributions of pore-ﬂuid

velocity and temperature in the real three-dimensional case (i.e. case 2). By com-

paring the numerical solutions from case 1 (Fig. 2.13) with those from case 2

(Fig. 2.15), it is observed that the distribution patterns of both pore-ﬂuid velocity

and temperature are totally different for these two cases. This demonstrates that the

perturbation of gravity at different planes may have a signiﬁcant effect on the pat-

tern of convective pore-ﬂuid ﬂow in three-dimensional hydrothermal systems. Just

like two-dimensional convective ﬂow problems, the solution dependence of three-

dimensional convective ﬂow on the direction of the perturbation of gravity need to

be considered when these kinds of results are interpreted.

To further observe the movements of pore-ﬂuid particles, the particle tracking

technique introduced in this section is used during the numerical computation.

34 2 Simulating Steady-State Natural Convective Problems

(t = 1 × 10

13

s) (t = 0s)

(t = 3 × 10

13

s) (t = 2 × 10

13

s)

(t = 5 × 10

13

s) (t = 4 × 10

13

s)

Fig. 2.16 Particle trajectories in the porous medium (Case 1)

We selected 24 particles, 12 of which are in the front plane and the rest are

in the back plane of the computation domain, to view the trajectories of those

particles. Figure 2.16 shows the particle trajectories in the porous media for several

different time instants in case 1. It is clear that in this case, all the particles move

within the planes they are initially located within (at t =0 s), because the convective

pore-ﬂuid ﬂow (in case 1) is essentially axisymmetrical about the rotation axis of

these particles considered. Figures 2.17 and 2.18 show the particle trajectories in

the porous media for several different time instants in case 2. It is observed that the

particle trajectories shown in case 2 are totally different from those shown in case

1, although the initial locations (at t =0 s) of these 24 particles are exactly the same

for these two cases. In fact, the particle trajectories shown in case 2 are much more

complicated than those shown in case 1. Since pore-ﬂuid is often the sole agent to

carry the minerals from the lower crust to the upper crust of the Earth, the differ-

ent patterns of particle trajectories imply that the pattern of ore body formation and

mineralization may be totally different in those two three-dimensional hydrothermal

systems.

2.6 Application of the Progressive Asymptotic Approach Procedure 35

(t = 1 × 10

13

s)

(t = 0s)

(t = 3 × 10

13

s) (t = 2 × 10

13

s)

(t = 5 × 10

13

s) (t = 4 × 10

13

s)

Fig. 2.17 Particle trajectories in the porous medium (Case 2)

36 2 Simulating Steady-State Natural Convective Problems

(t = 6 × 10

13

s)

(t = 8 × 10

13

s)

(t = 1 × 10

14

s) (t = 1.1 × 10

14

s)

(t = 9 × 10

13

s)

(t = 7 × 10

13

s)

Fig. 2.18 Particle trajectories in the porous medium (Case 2)

Chapter 3

A Consistent Point-Searching Interpolation

Algorithm for Simulating Coupled Problems

between Deformation, Pore-Fluid Flow, Heat

Transfer and Mass Transport Processes

in Hydrothemal Systems

Over the past decade or so, many commercial computational codes have become

available for solving a great number of practical problems in both scientiﬁc and

engineering ﬁelds. Primary advantages of using commercial computational codes

are: (1) built-in pre-processing and post-processing tools make it very easy and

attractive to prepare, input and output data which are essential in a numerical analy-

sis; (2) provision of movie/animation functions enables numerical results, the treat-

ment of which is often a cumbersome and tedious task, to be visualised via clear and

colourful images; (3) detailed benchmark solutions and documentation as well as

many embedded robust solution algorithms allow the codes to be used more easily,

correctly, effectively and efﬁciently for solving a wide range of practical problems.

However, the main disadvantage of using commercial computational codes is that

each code is often designed, within a certain limit, for solving some particular kinds

of practical problems. This disadvantage becomes more and more obvious because

the ever-increasing competitiveness in the world economy requires us to deal with

more and more complicated and complex geoscience problems, which are encoun-

tered and not solved in the ﬁeld of contemporary computational geoscience. There

are three basic ways to overcome the above difﬁculties. The ﬁrst is to develop some

new commercial computational codes, which is time consuming and often not cost-

effective for numerical analysts and consultants. The second is to extend an existing

commercial computational code, which is usually impossible because the source

code is often not available for the code users. The third is to use several existing

commercial computational codes in combination. This requires development of a

data translation tool to transfer data necessary between each of the codes to be used.

Compared with the difﬁculties encountered in the ﬁrst two approaches, the third one

is more competitive for most numerical analysts and consultants.

Our ﬁrst successful example in the practice of using commercial computa-

tional codes in a combination manner was to optimize structural topologies under

either static or dynamic conditions using the commercial code STRAND6 (G+D

Computing 1991) and a home-made code GEMDYN. As a result, a generalized

evolutionary method for numerical topological optimization of structures has been

developed and many interesting numerical results have been produced (Zhao et al.

1996d, e, 1997b, c, d, 1998c, d). To extend further the idea of using commercial

C. Zhao et al., Fundamentals of Computational Geoscience,

Lecture Notes in Earth Sciences 122, DOI 10.1007/978-3-540-89743-9 3,

C

Springer-Verlag Berlin Heidelberg 2009

37

38 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems

computational codes in combination, we attempt to use a combination of FIDAP

(Fluid Dynamics International 1997) and FLAC (Itasca Consulting Group 1995)

for solving a fully coupled problem between medium deformation, pore-ﬂuid ﬂow,

heat transfer and reactive species transport in a porous medium under high Rayleigh

number situations. FIDAP is a well developed, ﬁnite element method based, com-

putational ﬂuid dynamics code, whereas FLAC, designed for civil engineering,

is based on a ﬁnite difference method, but can accommodate unstructured grids.

FIDAP can be used to model pore-ﬂuid ﬂow, heat transfer and reactive species trans-

port in a porous medium, but does not treat the medium deformation effects. On

the other hand, FLAC is very powerful in its modeling of geomechanical and geo-

logical deformation processes, especially for the simulation of large deformation

problems. However, the weakness of FLAC is that it cannot be used rigorously to

model steady-state pore-ﬂuid convection and the related reactive species transport

in a ﬂuid-saturated porous medium. Thus, it is very reasonable to envisage interac-

tively using FIDAP and FLAC for solving a fully coupled problem between medium

deformation, pore-ﬂuid ﬂow, heat transfer and reactive mass transport in a porous

medium under high Rayleigh number situations. In order to do this, it must be pos-

sible to relate accurately any point in the mesh for one code to the equivalent point

in the mesh for the other code.

To do this, we present a consistent point-searching interpolation algorithm, also

known as the consistent point-searching algorithm for solution interpolation in

unstructured meshes consisting of 4-node bilinear quadrilateral elements. The pro-

posed algorithm has the following signiﬁcant advantages: (1) the use of a point-

searching strategy allows a point in one mesh to be accurately related to an element

(containing this point) in another mesh. Thus, to translate/transfer the solution of

any particular point from mesh 2 to mesh 1, only one element in mesh 2 needs to

be inversely mapped. This certainly minimizes the number of elements to which the

inverse mapping is applied. In this regard, the proposed consistent algorithm is very

effective and efﬁcient. (2) Analytical solutions to the local coordinates of any point

in a four-node quadrilateral element, which are derived in a rigorous mathemati-

cal manner in the context of this chapter, make it possible to carry out an inverse

mapping process very effectively and efﬁciently. (3) The use of consistent interpo-

lation enables the interpolated solution to be compatible with an original solution

and therefore guarantees the interpolated solution of extremely high accuracy. Since

the algorithm is very general and robust, it makes it possible to translate and transfer

data between FIDAP and FLAC, and vice versa.

3.1 Statement of the Coupled Problem and Solution Method

In terms of simulating the physical and chemical processes associated with ore

body formation and mineralization in hydrothermal systems within the upper crust

of the Earth, the fully coupled problem between material deformation, pore-ﬂuid

ﬂow, heat transfer and mass transport/chemical reactions can be divided into two

sub-problems (Zhao et al. 1999c, 2000b). For the ﬁrst sub-problem, which is the

3.1 Statement of the Coupled Problem and Solution Method 39

problem describing the steady-state pore-ﬂuid ﬂow, heat transfer and mass trans-

port/chemical reactions in a porous medium, the corresponding governing equations

are expressed as follows:

∂u

∂x

+

∂v

∂y

= 0, (3.1)

u =

K

x

μ

−

∂P

∂x

, (3.2)

v =

K

y

μ

−

∂P

∂y

+ρ

f

g

, (3.3)

ρ

f 0

c

p

u

∂T

∂x

+v

∂T

∂y

= λ

ex

∂

2

T

∂x

2

+λ

ey

∂

2

T

∂y

2

, (3.4)

ρ

f 0

u

∂C

i

∂x

+v

∂C

i

∂y

= ρ

f 0

D

ex

∂

2

C

i

∂x

2

+ D

ey

∂

2

C

i

∂y

2

+φR

i

(i = 1, 2, . . . , N),

(3.5)

ρ

f

= ρ

f 0

¸

1 −β

T

(T − T

0

) −

N

¸

i =1

β

Ci

(C

i

−C

0

)

, (3.6)

λ

ex

= φλ

f x

+(1 −φ)λ

sx

, λ

ey

= φλ

f y

+(1 −φ)λ

sy

, (3.7)

D

ex

= φD

fx

, D

ey

= φD

fy

, (3.8)

where u and v are the horizontal and vertical velocity components of the pore-ﬂuid

in the x and y directions respectively; P is the pore-ﬂuid pressure; T is the temper-

ature of the porous medium; C

i

is the normalized concentration (in a mass fraction

form relative to the pore-ﬂuid density) of chemical species i; N is the total num-

ber of the active chemical species considered in the pore-ﬂuid; K

x

and K

y

are the

permeabilities of the porous medium in the x and y directions respectively; μ is

the dynamic viscosity of the pore-ﬂuid; ρ

f

is the density of the pore-ﬂuid and g is

the acceleration due to gravity; ρ

f 0

, T

0

and C

0

are the reference density, reference

temperature and reference normalized concentration of the chemical species used in

the analysis; λ

f x

and λ

sx

are the thermal conductivities of the pore-ﬂuid and solid

matrix in the x direction; λ

f y

and λ

sy

are the thermal conductivities of the pore-ﬂuid

and solid matrix in the y direction; c

p

is the speciﬁc heat of the pore-ﬂuid; D

f x

and

D

f x

are the diffusivities of the chemical species in the x and y directions respec-

tively; φ is the porosity of the porous medium; β

T

and β

Ci

are the thermal volume

expansion coefﬁcient of the pore-ﬂuid and the volumetric expansion coefﬁcient due

to chemical species i; R

i

is the chemical reaction term for the transport equation of

chemical species i.

40 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems

Equation (3.6) clearly indicates that the density of the pore-ﬂuid considered in

this study is a function of both temperature and chemical species concentrations.

This means that the double diffusion effect (Phillips 1991, Nield and Bejan 1992,

Alavyoon 1993, Gobin and Bennacer 1994, Nguyen et al. 1994, Goyeau et al. 1996,

Nithiarasu et al. 1996, Mamou et al. 1998, Zhao et al. 2000b, 2005b, 2006a) is taken

into account in the ﬁrst sub-problem.

Generally, the chemical reaction term involved in a chemical species transport

equation is strongly dependent on the speciﬁc chemical reaction considered in the

analysis. For a non-equilibrium chemical reaction consisting of aqueous chemical

species only, the following type of equation can be considered as an illustrative

example.

A + B

k

1

⇒F (3.9)

This equation states that species A and species B react chemically at a rate con-

stant of k

1

and the product of this reaction is species F. For this type of chemical

reaction, the reaction term involved in Equation (3.5) can be expressed as follows:

R

1

= −k

1

C

1

C

2

, (3.10)

R

2

= −k

1

C

1

C

2

, (3.11)

R

3

= k

1

C

1

C

2

, (3.12)

where C

1

and C

2

are the normalized concentrations (in a mass fraction relative to the

pore-ﬂuid density) of species A (i.e. species 1) and B (i.e. species 2) respectively; C

3

is the normalized concentration (in a mass fraction relative to the pore-ﬂuid density)

of species F (i.e. species 3).

Note that for most of the chemical reactions encountered in the ﬁeld of geoscience,

the rate of a chemical reaction is dependent on the temperature at which the chemical

reaction takes place. From a geochemical point of view (Nield and Bejan 1992), the

temperature dependent nature of the reactionrate for the chemical reactionconsidered

here can be taken into account using the Arrhenius law of the following form:

k

1

= k

A

exp

−E

a

RT

(3.13)

where E

a

is the activation energy; Ris the gas constant; T is the temperature in Kelvin

and k

A

is the pre-exponential chemical reaction constant.

The second sub-problem is a static deformation problem under plane strain con-

ditions. If the hydrothermal system is initially in a mechanically equilibrium state,

then body forces can be neglected in the corresponding force equilibrium equations.

This means that we assume that the material deformation of the hydrothermal sys-

tem due to gravity has completed before the system is heated by some thermal event

from below. Under this assumption, the governing equations for static deformation

(which is labeled as the second sub-problem) in the porous medium under plane

strain conditions are:

3.1 Statement of the Coupled Problem and Solution Method 41

∂σ

x

∂x

+

∂τ

yx

∂y

= 0, (3.14)

∂τ

xy

∂x

+

∂σ

y

∂y

= 0, (3.15)

σ

x

=

E(1 −v)

(1 −2v)(1 +v)

ε

x

+

v

1 −v

ε

y

−

EαT

1 −2v

−

¯

P, (3.16)

σ

y

=

E(1 −v)

(1 −2v)(1 +v)

v

1 −v

ε

x

+ε

y

−

EαT

1 −2v

−

¯

P, (3.17)

τ

xy

= τ

yx

= 2Gγ

xy

, (3.18)

ε

x

=

∂u

s

∂x

, ε

y

=

∂v

s

∂y

, γ

xy

=

1

2

∂u

s

∂y

+

∂v

s

∂x

, (3.19)

where σ

x

and σ

y

are normal stresses of the solid matrix in the x and y directions; ε

x

and ε

y

are the normal strains of the solid matrix in relation to σ

x

and σ

y

; τ

xy

and γ

xy

are shear stress and shear strain of the solid matrix; u

s

and v

s

are the horizontal and

vertical displacements of the solid matrix;

¯

P is the excess pore-ﬂuid pressure due to

the thermal effect; E and G are the elastic and shear modulus respectively; ν is the

Poisson ratio of the solid matrix and α is the linear thermal expansion coefﬁcient of

the solid matrix.

Note that Eqs. (3.14) and (3.15) represent the equilibrium equations, whereas

Eqs. (3.16), (3.17), (3.18) and (3.19) are the constitutive equations and strain-

displacement relationship equations, respectively.

To couple the ﬁrst sub-problem with the second sub-problem, we need to estab-

lish a relationship between the volumetric strain and the porosity of the porous

medium. For small strain problems, such a relationship can be expressed as (Itasca

Consulting Group, 1995):

φ = 1 −

1 −φ

0

1 +ε

v

, ε

v

= ε

x

+ε

y

, (3.20)

where φ and φ

0

are the porosity and initial porosity of the porous medium; ε

v

is the

volumetric strain of the solid matrix.

Using the Carman-Kozeny formula (Nield and Bejan 1992), the permeability of

an isotropic porous medium is expressed as a function of porosity as follows:

K

x

= K

y

=

K

0

(1 −φ

0

)

2

φ

3

φ

3

0

(1 −φ)

2

, (3.21)

where K

0

is the initial permeability corresponding to the initial porosity, φ

0

.

Obviously, the ﬁrst sub-problem is coupled with the second sub-problem through

the medium temperature, T, pore-ﬂuid pressure, p, and permeabilities, K

x

and K

y

.

42 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems

Within the ﬁrst sub-problem itself, the pore-ﬂuid ﬂow is coupled with the thermal

ﬂow and reactive ﬂow through the medium temperature, aqueous chemical species

concentrations and pore-ﬂuid velocity components, u and v. However, within the

second sub-problem itself, the medium deformation (displacement) is coupled with

the medium temperature, pore-ﬂuid pressure and permeability through the volumet-

ric strain, ε

v

, and the medium porosity, φ.

With regard to the general solution method used in this study, FIDAP (Fluid

Dynamics International 1997), which is one of the most powerful ﬁnite element

codes currently available for dealing with a very wide range of practical prob-

lems in ﬂuid dynamics, is used to solve the ﬁrst sub-problem, while FLAC (Itasca

Consulting Group 1995), which is a very useful ﬁnite difference code for dealing

with material deformation problems in the ﬁelds of both geotechnical engineering

and geoscience, is used to solve the second sub-problem. A general interface pro-

gram is developed to translate and transfer data between FIDAP and FLAC. To

obtain non-trivial (convective) solutions for the pore-ﬂuid ﬂow under supercritical

Rayleigh number conditions, the progressive asymptotic approach procedure dis-

cussed in Chap. 2 (Zhao et al. 1997a) is also employed in the computation.

3.2 Mathematical Formulation of the Consistent Point-Searching

Interpolation Algorithm in Unstructured Meshes

The motivation of developing a consistent point-searching interpolation algorithm

is described as follows. For two totally different meshes, namely mesh 1 (mesh in

FIDAP) and mesh 2 (mesh in FLAC) shown in Fig. 3.1, ﬁnite element solutions

to the temperature and pressure due to hydrothermal effects are available (from

FIDAP) at the nodal points of mesh 1 but not known at those of mesh 2. This

Fig. 3.1 Two totally different meshes used in FIDAP and FLAC

3.2 Mathematical Formulation of the Algorithm 43

indicates that to evaluate the thermal deformation using mesh 2 in FLAC, it is essen-

tial to translate/transfer the nodal solutions from mesh 1 into mesh 2. Since mesh 1

is totally different from mesh 2, it is only possible to interpolate the nodal solu-

tion of mesh 2, point by point, using the nodal solutions of mesh 1 and the related

elemental information. Thus, for any nodal point in mesh 2, it is necessary to ﬁnd

out the element, which contains this particular point, in mesh 1. This is the ﬁrst step,

termed point-searching, in the proposed algorithm. Once a point in mesh 2 is related

to an element (containing this point) in mesh 1, the local coordinates of this point

need to be evaluated using the related information of the element in mesh 1. This

requires an inverse mapping to be carried out for this particular element in mesh 1,

so that the nodal solution of the point in mesh 2 can be consistently interpolated

in the ﬁnite element sense. This is the main reason why the proposed algorithm is

called the consistent point-searching interpolation algorithm.

Clearly, for the purpose of developing the consistent point-searching interpola-

tion algorithm, one has to deal with the following two key issues. First, an efﬁcient

searching strategy needs to be developed to limit the number of elements, to which

the inverse mapping is applied, so that the algorithm growth rate can be reduced

to the minimum. Second, the issue of the parametric inverse mapping between the

real (global) and element (local) coordinates should be dealt with in an appropriate

manner. Generally, the inverse mapping problem is a nonlinear one, which requires

numerical solutions for higher order elements. However, for 4-node bilinear quadri-

lateral elements, it is possible to solve this problemanalytically. Since most practical

problems in ﬁnite element analysis can be modelled reasonably well using 4-node

bilinear quadrilateral elements, the analytical solution to the inverse mapping prob-

lemfor this kind of element may ﬁnd wide applications for many practical problems.

Given the importance of the above two key issues, they are addressed in great detail

in the following sections, respectively.

3.2.1 Point Searching Step

In this section, we deal with the ﬁrst issue, i.e., the development of an efﬁcient

searching strategy to limit the number of elements, which need to be inversely

mapped. To achieve this, we have to establish a strategy to relate a point in one

mesh to an element in another mesh accurately. Consider a four-node quadrilateral

element shown in Fig. 3.2. If any point (i.e., point A in Fig. 3.2) is located within

this element, then the following equations hold true:

→

12 ×

→

1A = λ

1

→

k

, (3.22)

→

23 ×

→

2A = λ

2

→

k

, (3.23)

→

34 ×

→

3A = λ

3

→

k

, (3.24)

→

41 ×

→

4A = λ

4

→

k

, (3.25)

λ

i

≥ 0 (i = 1, 2, 3, 4), (3.26)

44 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems

3

4

1

2

A

0

x

y

Fig. 3.2 Point A in a fournode quadrilateral element

where

→

12,

→

23,

→

34 and

→

41 are vectors of four sides of the element;

→

1A,

→

2A,

→

3A

and

→

4A are vectors of each node of the element to point A respectively;

→

k

is a

normal vector of the plane where the element is located; λ

i

(i = 1, 2, 3, 4) are four

different constants. Note that the left-hand sides of Eqs. (3.22), (3.23), (3.24) and

(3.25) represent the cross products of two vectors.

By using the global coordinates of point A and four nodal points of the element,

Eqs (3.22), (3.23), (3.24), (3.25) and (3.26) can be expressed in the following form:

(x

2

− x

1

)(y

A

− y

1

) −(x

A

− x

1

)(y

2

− y

1

) = λ

1

≥ 0, (3.27)

(x

3

− x

2

)(y

A

− y

2

) −(x

A

− x

2

)(y

3

− y

2

) = λ

2

≥ 0, (3.28)

(x

4

− x

3

)(y

A

− y

3

) −(x

A

− x

3

)(y

4

− y

3

) = λ

3

≥ 0, (3.29)

(x

1

− x

4

)(y

A

− y

4

) −(x

A

− x

4

)(y

1

− y

4

) = λ

4

≥ 0, (3.30)

where x

i

and y

i

are the global coordinates of nodal points of the element; x

A

and y

A

are the global coordinates of point A.

Note that when point A is located on the side of the element, two vectors related

to a node of the element are coincident. Consequently, the corresponding λ

i

to this

node must be equal to zero. Since Eqs. (3.27), (3.28), (3.29) and (3.30) are only

dependent on the global coordinates of ﬁve known points (i.e., point A and four

nodes of the element), they can be straightforwardly used to predict the element, in

which point A is located.

It must be pointed out that, using the point-searching strategy expressed by

Eqs. (3.27), (3.28), (3.29) and (3.30), it is very easy and accurate to relate a point

in mesh 2 to an element (containing this point) in mesh 1, since Eqs. (3.27), (3.28),

(3.29) and (3.30) only involve certain simple algebraic calculations which can be

carried out by computers at a very fast speed. Thus, to translate/transfer the solution

3.2 Mathematical Formulation of the Algorithm 45

of any point from mesh 1 to mesh 2, only one element in mesh 1 needs to be

inversely mapped. This certainly reduces the number of elements, to which the

inverse mapping is applied, to the minimum. From the computational point of view,

the present algorithm is both effective and efﬁcient.

3.2.2 Inverse Mapping Step

This section deals with the second key issue related to the development of the con-

sistent point-searching interpolation algorithm. In order to interpolate the solution

at point A consistently in the ﬁnite element sense, it is necessary to ﬁnd out the

local coordinates of this point by an inverse mapping. For a four-node quadrilateral

isoparametric element shown in Fig. 3.3, it is possible to implement the inverse

mapping analytically. The following gives the related mathematical equations.

0

3

4

1

2

0

x

y

ξ

η

2

1

4 3

η

ξ

0

(Physical element)

(Parent element)

Fig. 3.3 A four-node quadrilateral isoparametric element (Zienkiewicz 1977; Zhao et al. 1999f)

46 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems

The forward mapping of the physical element in the global system to the parent

element in the local system reads:

x =

4

¸

i =1

N

i

x

i

, (3.31)

y =

4

¸

i =1

N

i

y

i

, (3.32)

where

N

1

=

1

4

(1 −ξ)(1 −η), (3.33)

N

2

=

1

4

(1 +ξ)(1 −η), (3.34)

N

3

=

1

4

(1 +ξ)(1 +η), (3.35)

N

4

=

1

4

(1 −ξ)(1 +η), (3.36)

where N

1

, N

2

, N

3

and N

4

are shape functions of the element.

For a given point (i.e., point A), the corresponding local coordinates are ξ

A

and

η

A

. Substituting these local coordinates into Eqs. (3.31), (3.32), (3.33), (3.34), (3.35)

and (3.36) yields the following inverse mapping:

a

2

ξ

A

+a

3

η

A

+a

4

ξ

A

η

A

= 4x

A

−a

1

, (3.37)

b

2

ξ

A

+b

3

η

A

+b

4

ξ

A

η

A

= 4y

A

−b

1

, (3.38)

where

a

1

= x

1

+ x

2

+ x

3

+ x

4

, (3.39)

a

2

= −x

1

+ x

2

+ x

3

− x

4

, (3.40)

a

3

= −x

1

− x

2

+ x

3

+ x

4

, (3.41)

a

4

= x

1

− x

2

+ x

3

− x

4

, (3.42)

b

1

= y

1

+ y

2

+ y

3

+ y

4

, (3.43)

b

2

= −y

1

+ y

2

+ y

3

− y

4

, (3.44)

b

3

= −y

1

− y

2

+ y

3

+ y

4

, (3.45)

b

4

= y

1

− y

2

+ y

3

− y

4

. (3.46)

3.2 Mathematical Formulation of the Algorithm 47

Since Eqs. (3.37) and (3.38) are two second-order simultaneous equations, the

solution to the local coordinates, ξ

A

and η

A

, can be obtained in the following differ-

ent cases.

3.2.2.1 Case 1: a

4

= 0, b

4

= 0

In this case, Eqs. (3.37) and (3.38) can be written as

a

2

ξ

A

+a

3

η

A

= c

1

, (3.47)

b

2

ξ

A

+b

3

η

A

= c

2

, (3.48)

where

c

1

= 4x

A

−a

1

, (3.49)

c

2

= 4y

A

−b

1

. (3.50)

Clearly, Eqs. (3.47) and (3.48) are a set of standard linear simultaneous equations,

so that the corresponding solutions can be immediately obtained as

ξ

A

=

b

3

c

1

−a

3

c

2

a

2

b

3

−a

3

b

2

, (3.51)

η

A

=

−b

2

c

1

+a

2

c

2

a

2

b

3

−a

3

b

2

. (3.52)

3.2.2.2 Case 2: a

4

= 0, b

4

= 0

The corresponding equations in this case can be rewritten as

a

2

ξ

A

+a

3

η

A

= c

1

, (3.53)

b

2

ξ

A

+b

3

η

A

+b

4

ξ

A

η

A

= c

2

. (3.54)

The solutions to Eqs. (3.53) and (3.54) can be expressed for the following three

sub-cases:

(1) a

2

= 0, a

3

= 0

η

A

=

c

1

a

3

, (3.55)

ξ

A

=

c

2

−b

3

η

A

b

2

+b

4

η

A

. (3.56)

(2) a

2

= 0, a

3

= 0

ξ

A

=

c

1

a

2

, (3.57)

48 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems

η

A

=

c

2

−b

2

ξ

A

b

3

+b

4

ξ

A

. (3.58)

(3) a

2

= 0, a

3

= 0

η

A

=

−bb ±

bb

2

−(4aa)cc

2aa

, (3.59)

ξ

A

=

c

1

−a

3

η

A

a

2

, (3.60)

where

aa =

b

4

a

3

a

2

, (3.61)

bb =

b

2

a

3

−b

4

c

1

a

2

−b

3

, (3.62)

cc =

−b

2

c

1

a

2

. (3.63)

3.2.2.3 Case 3: a

4

= 0, b

4

= 0

This case is very similar to case 2. The corresponding equations in this case are as

follows:

a

2

ξ

A

+a

3

η

A

+a

4

ξ

A

η

A

= c

1

, (3.64)

b

2

ξ

A

+b

3

η

A

= c

2

. (3.65)

Similarly, the solution to Equations (3.64) and (3.65) can be expressed in the

following three sub-cases:

(1) b

2

= 0, b

3

= 0

η

A

=

c

2

b

3

, (3.66)

ξ

A

=

c

1

−a

3

η

A

a

2

+a

4

η

A

. (3.67)

(2) b

2

= 0, b

3

= 0

ξ

A

=

c

2

b

2

, (3.68)

η

A

=

c

1

−a

2

ξ

A

a

3

+a

4

ξ

A

. (3.69)

3.2 Mathematical Formulation of the Algorithm 49

(3) b

2

= 0, b

3

= 0

η

A

=

−bb ±

bb

2

−(4aa)cc

2aa

, (3.70)

ξ

A

=

c

2

−b

3

η

A

b

2

, (3.71)

where

aa =

a

4

b

3

b

2

, (3.72)

bb =

a

2

b

3

−a

4

c

2

b

2

−b

3

, (3.73)

cc =

−a

2

c

2

b

2

. (3.74)

3.2.2.4 Case 4: a

4

= 0, b

4

= 0

In this case, Eqs. (3.37) and (3.38) can be written in the following form:

a

∗

2

ξ

A

+a

∗

3

η

A

+ξ

A

η

A

= c

∗

1

, (3.75)

b

∗

2

ξ

A

+b

∗

3

η

A

+ξ

A

η

A

= c

∗

2

, (3.76)

where

a

∗

i

=

a

i

a

4

, b

∗

i

=

b

i

b

4

(i = 2, 3), (3.77)

c

∗

1

=

c

1

a

4

, c

∗

2

=

c

2

b

4

. (3.78)

Subtracting Eq. (3.78) from Eq. (3.77) yields the following equation:

(a

∗

2

−b

∗

2

)ξ

A

+(a

∗

3

−b

∗

3

)η

A

= c

∗

1

−c

∗

2

. (3.79)

Clearly, if a

∗

2

−b

∗

2

= 0, the solutions to Eqs. (3.77) and (3.78) can be straightfor-

wardly expressed as

η

A

=

c

∗

1

−c

∗

2

a

∗

3

−b

∗

3

, (3.80)

ξ

A

=

c

∗

1

−a

∗

3

η

A

a

∗

2

+η

A

. (3.81)

50 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems

Otherwise, Eq. (3.79) can be rewritten as

ξ

A

= α −βη

A

(3.82)

where

α =

c

∗

1

−c

∗

2

a

∗

2

−b

∗

2

, (3.83)

β =

a

∗

3

−b

∗

3

a

∗

2

−b

∗

2

. (3.84)

Therefore, the corresponding solutions can be expressed for the following sub-

cases.

(1) For β = 0

ξ

A

= α, (3.85)

η

A

=

c

∗

1

−a

∗

2

ξ

A

a

∗

3

+ξ

A

. (3.86)

(2) For β = 0

η

A

=

−(a

∗

2

β −a

∗

3

−α) ±

(a

∗

2

β −a

∗

3

−α)

2

−4β(c

∗

1

−a

∗

2

α)

2β

, (3.87)

ξ

A

= α −βη

A

. (3.88)

3.2.3 Consistent Interpolation Step

Based on the concept of isoparametric elements (Zienkiewicz 1977), any nodal solu-

tion at any point (i.e., point A) in mesh 1 can be consistently interpolated using the

following equation:

S

A

=

4

¸

i =1

N

i

(ξ

A

, η

A

)S

i

, (3.89)

where S

i

is the appropriate numerical solution at a nodal point in mesh 1.

Note that since the global coordinate systemused in mesh 1 is exactly the same as

in mesh 2, the nodal value of any concerned solution in mesh 2 can be interpolated

through mesh 1 and then directly transferred to mesh 2.

3.3 Veriﬁcation of the Proposed Consistent Point-Searching Interpolation Algorithm 51

3.3 Veriﬁcation of the Proposed Consistent Point-Searching

Interpolation Algorithm

In order to verify the proposed consistent point-searching interpolation algorithm in

two totally different meshes, we consider a coupled problembetween mediumdefor-

mation, pore-ﬂuid ﬂow and heat transfer processes in a ﬂuid-saturated porous elastic

medium. The square computational domain is 10 × 10 km

2

in size. This computa-

tional domain is discretized into mesh 1 of 1625 nodes and mesh 2 of 10000 nodes

respectively. Firstly, we use FIDAP with mesh 1 to simulate the high Rayleigh num-

ber convection cells, temperature and pressure distributions in the computational

domain. Then, we use the proposed algorithm to translate/transfer the temperature

and pressure solutions in mesh 1 into FLAC’s mesh, mesh 2.

To simulate the high Rayleigh number convection cells in the computational

domain of the test problem, the following parameters are used in the compu-

tation. For pore-ﬂuid, dynamic viscosity is 10

−3

N ×s/m

2

; reference density is

1000 kg/m

3

; volumetric thermal expansion coefﬁcient is 2.07×10

−4

(1/

0

C); speciﬁc

heat is 4185 J/(kg×

0

C); thermal conductivity coefﬁcient is 0.6 W/(m×

0

C) in both

the horizontal and vertical directions. For the porous matrix, initial porosity is 0.1;

initial permeability is 10

−14

m

2

; thermal conductivity coefﬁcient is 3.35 W/(m×

0

C)

in both the horizontal and vertical directions. The temperature is 25

◦

C at the top

of the domain, while it is 225

◦

C at the bottom of the domain. This means that

the computational domain is heated uniformly from below. The left and right lat-

eral boundaries of the computational domain are insulated and impermeable in the

horizontal direction, whereas the top and bottom are impermeable in the vertical

direction.

Figures 3.4 and 3.5 show the comparison of the original temperature and pres-

sure solutions in mesh 1 with the corresponding translated/ transferred solutions in

mesh 2. It is clear that although mesh 2 is totally different from mesh 1, the trans-

lated/transferred solutions in mesh 2 are essentially the same as in mesh 1. This

demonstrates the correctness and effectiveness of the proposed consistent point-

searching interpolation algorithm.

To further test the robustness of the proposed algorithm, we use the concept of a

transform in mathematics below. For instance, the robustness of a numerical Fourier

transform algorithm is often tested by the following procedure: (1) implementation

of a forward Fourier transform to an original function; and (2) implementation of

an inverse Fourier transform to the forwardly transformed function. If the inversely

transformed function is exactly the same as the original function, it demonstrates

that the numerical Fourier transform algorithm is robust. Clearly, the same proce-

dure as above can be followed to examine the robustness of the proposed consis-

tent point-searching algorithm. For this purpose, the forward transform is deﬁned as

translating/transferring solution data from mesh 1 to mesh 2, while the inverse trans-

form is deﬁned as translating/transferring the translated/transferred solution data

back from mesh 2 to mesh 1.

Figures 3.6 and 3.7 show the comparisons of the forwardly transformed temper-

ature and pressure solutions (from mesh 1 to mesh 2) with the inversely transformed

52 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems

Fig. 3.4 Comparison of the original solution with the translated/transferred solution (Temperature)

3.3 Veriﬁcation of the Proposed Consistent Point-Searching Interpolation Algorithm 53

Fig. 3.5 Comparison of the original solution with the translated/transferred solution (Pore-ﬂuid

pressure)

54 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems

Fig. 3.6 Comparison of the forwardly transformed solution with the inversely transformed solution

(Temperature)

3.3 Veriﬁcation of the Proposed Consistent Point-Searching Interpolation Algorithm 55

Fig. 3.7 Comparison of the forwardly transformed solution with the inversely transformed solution

(Pore-ﬂuid pressure)

56 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems

solutions (from mesh 2 to mesh 1). Clearly, the forwardly transformed solutions in

mesh 2 are exactly the same as the inversely transformed solutions in mesh 1. Fur-

thermore, the inversely transformed solutions in mesh 1 (see Figs. 3.6 and 3.7) also

compare very well with the original solutions in mesh 1 (see Figs. 3.4 and 3.5). This

means that after the original solutions are transformed from mesh 1 to mesh 2, they

can be transformed back exactly from mesh 2 to mesh 1. Such a reversible process

demonstrates the robustness of the proposed consistent point-searching interpola-

tion algorithm.

3.4 Application Examples of the Proposed Consistent

Point-Searching Interpolation Algorithm

3.4.1 Numerical Modelling of Coupled Problems Involving

Deformation, Pore-Fluid Flow and Heat Transfer

in Fluid-Saturated Porous Media

Since the veriﬁcation example considered in Sect. 3.3 is a coupled problem between

medium deformation, pore-ﬂuid ﬂow and heat transfer processes in a ﬂuid-saturated

porous elastic medium, it can be used as the ﬁrst application example of the pro-

posed consistent point-searching interpolation algorithm. Thus, we can continue the

simulation of the veriﬁcation example and use FLAC with the translated/transferred

temperature in mesh 2 to compute thermal deformation and stresses in the ﬂuid-

saturated porous elastic medium. Towards this end, it is assumed that: (1) the bot-

tom boundary of the computational domain is ﬁxed; (2) the top boundary is free; and

(3) the two lateral boundaries are horizontally ﬁxed but vertically free. Except for

the parameters used in Sect. 3.3, the following additional parameters are used in the

continued simulation: the elastic modulus of the porous medium is 1×10

10

Pa; Pois-

son’s ratio is 0.25; the volumetric thermal expansion coefﬁcient is 2.07×10

−4

(1/

o

C)

and the initial porosity is 0.1.

Figures 3.8 and 3.9 show the temperature induced deformation and stresses in

the porous elastic medium respectively. It is observed that the distribution pattern

of the volumetric strain is similar to that of the temperature. That is to say, higher

temperature results in larger volumetric strain, as expected from the physics point

of view. Owing to relatively larger volumetric strain in the left side of the compu-

tational domain, the maximum vertical displacement takes place at the upper left

corner of the domain. This is clearly evidenced in Fig. 3.8. As a direct consequence

of the thermal deformation, the temperature induced horizontal stress dominates in

the porous medium. The maximum horizontal compressive stress due to the thermal

effect takes place in the hottest region of the computational domain, while the max-

imum horizontal tensile stress occurs at the upper left corner of the domain, where

the vertical displacement reaches its maximum value. Apart from a small part of

the top region of the computational domain, the thermal induced vertical stress is

3.4 Examples of the Proposed Consistent Point-Searching Interpolation Algorithm 57

Fig. 3.8 Deformation of the porous medium due to temperature effect

58 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems

Fig. 3.9 Temperature induced stresses in the porous medium

3.4 Examples of the Proposed Consistent Point-Searching Interpolation Algorithm 59

well distributed anti-symmetrically. This indicates that the vertical forces are well

balanced in the horizontal cross-section of the computational domain.

3.4.2 Numerical Modelling of Coupled Problems Involving

Deformation, Pore-Fluid Flow, Heat Transfer and Mass

Transport in Fluid-Saturated Porous Media

Generally, the numerical algorithms and methods can be used to answer many what-

if questions related to ore body formation and mineralization that can be described

as coupled problems between medium deformation, pore-ﬂuid ﬂow, heat transfer

and reactive mass transport in ﬂuid-saturated porous media. However, due to the

approximate nature of a numerical solution, it is essential to evaluate the accuracy

of the numerical solution, at least qualitatively if the analytical solution to the prob-

lem is not available. This requires us to have a strong theoretical understanding of

the basic governing principles and processes behind the coupled problem. Speciﬁ-

cally, we must know, through some kind of theoretical analysis, what the pore-ﬂuid

can and cannot do, and what reaction patterns the reactive pore-ﬂuid can produce

in a hydrothermal system. For this reason, we have been making efforts, in recent

years, to develop theoretical solutions to verify the numerical methods developed

(Zhao and Valliappan 1993a, b, 1994a, b, Zhao et al. 1997a, 1998a, 1999b). On

the other hand, a good numerical solution can provide insights into the integrated

behaviour of different processes that occur in a hydrothermal system. Even in some

circumstances, a good numerical solution can provide some useful hints for deriving

analytical solutions to some aspects of the problem. This indicates that the numerical

and theoretical approaches are, indeed, complementary in the ﬁeld of computational

geoscience. Realizing this particular relationship between the analytical method and

the numerical method, Phillips (1991) stated that: “A conceptual framework eluci-

dating the relations among ﬂow characteristics, driving forces, structure, and reac-

tion patterns enables us not only to understand the results of numerical modelling

more clearly, but to check them. (Numerical calculations can converge to a grid-

dependent limit, and artifacts of a solution can be numerical rather than geolog-

ical.) Numerical modelling provides a quantitative description and synthesis of a

basin-wide ﬂow in far greater detail than would be feasible analytically. The com-

bination of the two techniques is a much more powerful research tool than either

alone.” Keeping this in mind, we have used the proposed consistent point-searching

interpolation algorithm to develop a general interface between the two commercial

computational codes, FIDAP and FLAC. This development enables us to investigate

the integrated behaviour of ore body formation and mineralization in hydrothermal

systems.

The ﬁrst application example, which is closely associated with the coupled prob-

lem between medium deformation, pore-ﬂuid ﬂow, heat transfer and reactive mass

transport in a ﬂuid-saturated porous medium, is to answer the question: What is

the pattern of pore-ﬂuid ﬂow, the distributions of temperature, reactant and product

60 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems

10km

10km

y

x

0

0 , 25 , 0

3 2 1

0

= = = = = C C C C T v

0 =

x

0 =

u = 0 u = 0

001 . 0

1

= C 001 . 0

2

= C

C T v

0

225 , 0 = =

∂

∂

T

x

∂

∂

T

Fig. 3.10 Geometry and boundary conditions of the coupled problem

chemical species, and the patterns of ﬁnal porosity and permeability, if a deformable

hydrothermal system has two reservoirs for two different reactant chemical species,

and is heated uniformly from below?

As shown in Fig. 3.10, the computational domain considered for this example

is a square box of 10 by 10 km in size. The temperature at the top of the domain

is 25

◦

C, while it is 225

◦

C at the bottom of the domain. The left and right lateral

boundaries are insulated and impermeable in the horizontal direction, whereas the

top and bottom are impermeable in the vertical direction. To consider the mixing of

reactant chemical species, the normalized concentration of species 1 is 0.001 at the

left quarter of the bottom and the normalized concentration of species 2 is 0.001 at

the right quarter of the bottom. This implies that species 1 and 2 are injected into

the computational domain through two different reservoirs at different locations.

The normalized concentrations of all three species involved in the chemical reaction

are assumed to be zero at the top surface of the computational domain. Table 3.1

shows the related parameters used in the computation. In addition, the computational

domain is discretized into 2704 quadrilateral elements with 2809 nodes in total.

Figure 3.11 shows the pore-ﬂuid velocity and temperature distributions in the

deformable porous medium. It is observed that a clockwise convection cell has

formed in the pore-ﬂuid ﬂow in the porous medium. This convective ﬂow results

in the localized distribution of temperature, as can be seen clearly from Fig. 3.11.

As we mentioned in the beginning of this section, we have to validate the numeri-

cal solution, at least from a qualitative point of view. For the hydrothermal system

3.4 Examples of the Proposed Consistent Point-Searching Interpolation Algorithm 61

Table 3.1 Parameters used for the ﬁrst coupled problem involving reactive mass transport

Material type Parameter Value

pore-ﬂuid dynamic viscosity 10

−3

N ×s/m

2

reference density 1000 kg/m

3

volumetric thermal expansion coefﬁcient 2.07 ×10

−4

(1/

0

C)

speciﬁc heat 4185 J/(kg×

0

C)

thermal conductivity coefﬁcient 0.6 W/(m×

0

C)

chemical species diffusivity coefﬁcient 3 ×10

−6

m

2

/s

pre-exponential reaction rate constant 10

−7

kg/(m

3

×s)

activate energy of reaction 5 ×10

4

J/mol

gas constant 8.315 J/(mol×

0

K)

porous matrix initial porosity 0.1

initial permeability 10

−14

m

2

elastic modulus 2 ×10

10

Pa

Poisson’s ratio 0.25

volumetric thermal expansion coefﬁcient 2.07 ×10

−5

(1/

0

C)

thermal conductivity coefﬁcient 3.35 W/(m×

0

C)

studied, the temperature gradient is the main driving force to initiate the convective

pore-ﬂuid ﬂow. There is a criterion available (Phillips 1991, Nield and Bejan 1992,

Zhao et al. 1997a) for judging whether or not the convective pore-ﬂuid ﬂow is pos-

sible in such a hydrothermal system as considered here. The criterion says that if the

Rayleigh number of the hydrothermal system, which has a ﬂat bottom and is heated

uniformly from below, is greater than the critical Rayleigh number, which has a the-

oretical value of 4π

2

, then the convective pore-ﬂuid ﬂow should take place in the

hydrothermal system considered. For the particular hydrothermal system consid-

ered in this application example, the Rayleigh number is 55.2, which is greater than

the critical Rayleigh number. Therefore, the convective pore-ﬂuid ﬂow should occur

in the hydrothermal system, from the theoretical point of view. This indicates that

the numerical solution obtained from this application example can be qualitatively

justiﬁed.

Figure 3.12 shows the normalized concentration distributions of reactant and

product chemical species, whereas Fig. 3.13 shows the ﬁnal distribution of porosity

and permeability in the computational domain of this application example, where

both the material deformation due to thermal effects and the reactive mass transport

are taken into account in the numerical computation. Obviously, the distributions

of the three chemical species are different. This indicates that the convective pore-

ﬂuid ﬂow may signiﬁcantly affect the chemical reactions (i.e. reaction ﬂow) in the

deformable porous medium. In particular, the distribution of chemical species 3,

which is the product of the chemical reaction considered in this application exam-

ple, demonstrates that the produced chemical species in the chemical reaction may

reach its equilibrium concentration in some regions, but may not necessarily reach

its equilibrium concentration in other regions in the computational domain. This

ﬁnding might be important for the further understanding of reactive ﬂow transport

62 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems

Fig. 3.11 Distribution of pore-ﬂuid velocity and temperature in the hydrothermal system

3.4 Examples of the Proposed Consistent Point-Searching Interpolation Algorithm 63

Fig. 3.12 Normalized concentration distribution of reactant and production chemical species in

the hydrothermal system

64 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems

Fig. 3.13 Distribution of ﬁnal porosity and permeability in the hydrothermal system

3.4 Examples of the Proposed Consistent Point-Searching Interpolation Algorithm 65

involved in ore body formation and mineralization in the upper crust of the Earth.

Since the feedback effect of the medium deformation on the porosity and perme-

ability has been considered in the numerical simulation, the distributions of the ﬁnal

porosity and permeability are totally different from that of their initial values, which

are uniformly distributed in the computational domain.

Again, we can justify the numerical solution for the ﬁnal porosity and perme-

ability in the hydrothermal system. From the physics point of view, we know that

the higher the relative temperature in a porous medium, the larger the deforma-

tion of the porous medium due to the thermal effect. On the other hand, since the

solid aggregates in the porous medium are relatively stiff, the larger the deformation

of the porous medium, the greater the porosity of the porous medium, indicating

that large deformation of a porous medium results in greater permeability of the

porous medium. This implies that for a porous medium, a region of relatively high

temperature favours the formation of ﬂow channels because of an increase in the

porosity of the region due to the thermal effect. This kind of phenomenon can be

clearly observed from the related numerical solutions for the temperature distribu-

tion (Fig. 3.11) and the ﬁnal porosity/permeability distributions (Fig. 3.13) in the

hydrothermal system. This further justiﬁes the numerical solutions obtained from

this application example, at least from a qualitative point of view.

The next application example is to investigate how the chemical reaction rate

affects the distribution of product chemical species (i.e. produced new minerals),

if all other parameters are kept unchanged in the hydrothermal system considered

above. Since the pre-exponential reaction rate constant, to a large extent, repre-

sents how fast the chemical reaction proceeds, three different values of the pre-

exponential reaction rate constant, namely 10

−5

kg/(m

3

×s), 10

−7

kg/(m

3

×s) and

10

−10

kg/(m

3

×s), have been used in the corresponding computations.

Figures 3.14, 3.15 and 3.16 showthe effects of chemical reaction rates on the nor-

malized concentration distributions of reactant and product chemical species in the

hydrothermal system. It is obvious that the chemical reaction rate has little inﬂuence

on either the distribution pattern or the magnitude of the reactant chemical species

concentration. Even though the chemical reaction rate may affect the magnitude

of the product chemical species concentration, it does not affect the overall distri-

bution pattern of the product chemical species concentration. This ﬁnding implies

that if the reactant minerals constitute only a small fraction of the whole matrix

in a porous medium, which is a commonly accepted assumption in geochemistry

(Phillips 1991), the distribution pattern of the normalized concentration of the new

mineral produced by chemical reactions is strongly dependent on the characteristics

of the convective pore-ﬂuid ﬂow, even though the magnitude of the normalized con-

centration of the product mineral may also strongly depend on the rates of chemical

reactions.

As mentioned previously, any numerical solutions have to be justiﬁed before they

can be safely accepted and used. For this purpose, we must answer the following

questions: Are the numerical results reported in Figs. 3.14, 3.15 and 3.16 correct?

Do they represent the physical and chemical characteristics of the hydrother-

mal system rather than something else (i.e. numerical rubbish)? From these

66 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems

Fig. 3.14 Effects of chemical reaction rates on the normalized concentration distribution of the

reactant chemical species (species 1)

3.4 Examples of the Proposed Consistent Point-Searching Interpolation Algorithm 67

Fig. 3.15 Effects of chemical reaction rates on the normalized concentration distribution of the

reactant chemical species (species 2)

68 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems

Fig. 3.16 Effects of chemical reaction rates on the normalized concentration distribution of the

product chemical species (species 3)

3.4 Examples of the Proposed Consistent Point-Searching Interpolation Algorithm 69

numerical results, can we ﬁnd something of more theoretical importance? In the fol-

lowing we will answer the above questions through some theoretical analysis. For

the hydrothermal system considered, the maximum change in the density of pore-

ﬂuid due to temperature is about 4%, while the maximum change in the density of

pore-ﬂuid due to the concentration of any reactant chemical species is only about

0.1%. This means that the buoyancy produced by temperature is much greater than

that produced by the concentration of the reactant chemical species. As a result,

the convective pore-ﬂuid ﬂow is predominantly driven by the temperature gradient,

rather than the concentration gradients of chemical species, even though the contri-

bution of chemical species to the buoyancy is taken into account. More speciﬁcally,

for the hydrothermal systemconsidered, the velocity components of pore-ﬂuid in the

system are strongly dependent on the temperature gradient, rather than the concen-

tration gradients of chemical species. This means that the concentrations of chemi-

cal species have minimal inﬂuence on the convective pore-ﬂuid ﬂow. Based on this

recognition, we can further explore why both the pattern and the magnitude of the

normalized concentration of reactant chemical species are very weakly dependent

on the rates of the chemical reaction. For the purpose of facilitating the theoretical

analysis, we need to rewrite the transport equation (see Eq. (3.5)) for one of the

reactant chemical species (i.e. species 1) as follows:

ρ

f 0

u

∂C

1

∂x

+v

∂C

1

∂y

= ρ

f 0

D

ex

∂

2

C

1

∂x

2

+ D

ey

∂

2

C

1

∂y

2

−φk

A

C

1

C

2

exp

−E

a

RT

.

(3.90)

Since the reactant chemical species constitutes only a small fraction of the whole

matrix in a porous medium, as is commonly assumed in geochemistry (Phillips

1991), we can view the normalized concentration of the reactant chemical species

as the ﬁrst order small quantity in the above equation, at least from the mathematical

point of view. Thus, the reaction term in this equation is at least the second order

small quantity. This implies that the distribution of the reactant chemical species

is controlled by the pore-ﬂuid velocity and the dispersivity of the porous medium,

rather than by the chemical reaction term unless the rate of chemical reaction is very

fast. This is the reason why both the pattern and the magnitude of the normalized

concentration of the reactant chemical species are very weakly dependent on the

rates of the chemical reaction in the related numerical results (Figs. 3.14 and 3.15).

Next, we will examine, analytically, why the distribution pattern of the normal-

ized concentration of the product chemical species is almost independent of the

rates of the chemical reaction, but the magnitude of the product chemical species is

strongly dependent on the rates of the chemical reaction (Fig. 3.16). In this case, we

need to rewrite the transport equation for chemical species 3 as follows:

ρ

f 0

u

∂C

3

∂x

+v

∂C

3

∂y

= ρ

f 0

D

ex

∂

2

C

3

∂x

2

+ D

ey

∂

2

C

3

∂y

2

+φk

A

C

1

C

2

exp

−E

a

RT

.

(3.91)

70 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems

Since the magnitude of the normalized concentration of the reactant chemical

species is very weakly dependent on the rates of the chemical reaction, the product

of C

1

and C

2

is almost independent of the chemical reaction rate in Eq. (3.91). As

mentioned before, the product of C

1

and C

2

is at least the second order small quan-

tity in the theoretical analysis. As a result, the normalized concentration of chemical

species 3 is at least the second (or higher) order small quantity. This means that the

magnitude of the normalized concentration of chemical species 3 is determined by

the reaction term in Eq. (3.91). More speciﬁcally, the magnitude of the normalized

concentration of chemical species 3 is strongly dependent on the rate of the chemi-

cal reaction because the product of φ, C

1

and C

2

are eventually independent of the

chemical reaction rate in Eq. (3.91). If C

3A

is a solution corresponding to the reac-

tion rate constant k

AA

for Eq. (3.91), we can, mathematically, deduce the solution

for another reaction rate constant k

AB

below.

Since C

3A

is a solution for Equation (3.91), we have the following equation:

ρ

f 0

u

∂C

3A

∂x

+v

∂C

3A

∂y

= ρ

f 0

D

ex

∂

2

C

3A

∂x

2

+ D

ey

∂

2

C

3A

∂y

2

+φk

AA

C

1

C

2

exp

−E

a

RT

.

(3.92)

Multiplying Eq. (3.92) by k

AB

/k

AA

yields the following equation:

ρ

f 0

k

AB

k

AA

u

∂C

3A

∂x

+v

∂C

3A

∂y

= ρ

f 0

k

AB

k

AA

D

ex

∂

2

C

3A

∂x

2

+ D

ey

∂

2

C

3A

∂y

2

+φk

AB

C

1

C

2

exp

−E

a

RT

.

(3.93)

Therefore, the solution corresponding to the reaction rate constant k

AB

can be

straightforwardly expressed as

C

3B

=

k

AB

k

AA

C

3A

, (3.94)

where C

3B

is the solution corresponding to the reaction rate constant k

AB

for

Eq. (3.91).

Equation (3.94) states that for any particular point in the hydrothermal system

considered, the normalized concentration of the product chemical species varies

linearly with the reaction rate constant. This is the reason why the distribution pat-

tern of concentration of the product chemical species is almost independent of the

rates of the chemical reaction, but the magnitude of the product chemical species

is strongly dependent on the rates of the chemical reaction (Fig. 3.16). If we fur-

ther observe the numerical solutions in Fig. 3.16, then we ﬁnd that the normalized

concentration of the product chemical species, indeed, varies linearly with the reac-

tion rate constant. Therefore, the numerical analysis carried out for this application

example is further validated by the related theoretical analysis.

It needs to be pointed out that the proposed consistent point-searching interpola-

tion and the related solution methodology have been successfully applied to more

3.4 Examples of the Proposed Consistent Point-Searching Interpolation Algorithm 71

realistic geological situations (Zhao et al. 1998b, 1999c, Hobbs et al 2000, Gow

et al. 2002, Ord et al. 2002, Schaubs and Zhao 2002, Sorjonen-Ward et al. 2002,

Zhang et al. 2003), where both complicated geometry and material nonhomogene-

ity are considered. Although the consistent point-searching interpolation algorithm

is developed for the unstructured meshes consisting of 4-node bilinear quadrilateral

elements, it can equally be extended to the unstructured meshes consisting of higher

order elements. In the latter case, one has to solve the inverse mapping problem

numerically, instead of analytically.

Chapter 4

A Term Splitting Algorithm for Simulating

Fluid-Rock Interaction Problems

in Fluid-Saturated Hydrothermal Systems

of Subcritical Zhao Numbers

In recent years, we have been making efforts to develop a practical and predictive

tool to explore for giant ore deposits in the upper crust of the Earth. Towards this

goal, signiﬁcant progress has been made towards a better understanding of the basic

physical and chemical processes behind ore body formation and mineralization in

hydrothermal systems. On the scientiﬁc development side, we have developed ana-

lytical solutions to answer the following scientiﬁc questions (Zhao et al. 1998e,

1999b): (1) Can the pore-ﬂuid pressure gradient be maintained at the value of the

lithostatic pressure gradient in the upper crust of the Earth? and, (2) Can convec-

tive pore-ﬂuid ﬂow take place in the upper crust of the Earth if there is a ﬂuid/mass

leakage from the mantle to the upper crust of the Earth? On the modelling develop-

ment side, we have developed numerical methods to model the following problems:

(1) convective pore-ﬂuid ﬂow in hydrothermal systems (Zhao et al. 1997a, 1998b);

(2) coupled reactive pore-ﬂuid ﬂow and species transport in porous media (Zhao

et al. 1999a); (3) precipitation and dissolution of minerals in the upper crust of

the Earth (Zhao et al. 1998a, 2000a); (4) double diffusion driven pore-ﬂuid ﬂow in

hydrothermal systems (Zhao et al. 2000b); (5) pore-ﬂuid ﬂow patterns near geologi-

cal lenses in hydrodynamic and hydrothermal systems (Zhao et al. 1999d); (6) vari-

ous aspects of the fully coupled problem involving material deformation, pore-ﬂuid

ﬂow, heat transfer and species transport/chemical reactions in pore-ﬂuid saturated

porous rock masses (Zhao et al. 1999c, 1999f, 1999g). The above-mentioned work

has signiﬁcantly enriched our knowledge about the physical and chemical processes

related to ore body formation and mineralization in the upper crust of the Earth.

Since ﬂuid-rock interaction is another potential mechanism of ore body formation

and mineralization in the upper crust of the Earth, it is necessary to extend further the

developed numerical tools to solve ﬂuid-rock interaction problems. This requires us

to deal with coupled reactive species transport phenomenon, which is the direct con-

sequence of the chemical reactions that take place between aqueous reactive species

in pore-ﬂuid and solid minerals in pore-ﬂuid saturated porous rock masses.

For a natural ﬂuid-rock interaction system, the reactant chemical species consti-

tutes only a small fraction of the whole matrix in a porous rock (Phillips 1991). In

this case, the chemical dissolution front is stable if the Zhao number of the system

is subcritical, while it becomes unstable otherwise. Based on the concept of the

C. Zhao et al., Fundamentals of Computational Geoscience,

Lecture Notes in Earth Sciences 122, DOI 10.1007/978-3-540-89743-9 4,

C

Springer-Verlag Berlin Heidelberg 2009

73

74 4 Algorithm for Simulating Fluid-Rock Interaction Problems

generalized dimensionless pore-ﬂuid pressure gradient (Zhao et al. 2008e), the cor-

responding dimensionless Zhao number of a single mineral dissolution system can

be deﬁned as follows:

Zh =

ν

ﬂow

φ

f

D(φ

f

)

V

p

k

chemical

A

p

C

eq

, (4.1)

where ν

f low

is the Darcy velocity of the pore-ﬂuid ﬂow;

φ

f

and D

φ

f

are the

ﬁnal porosity of the porous medium and the corresponding diffusivity of chemical

species after the completion of soluble mineral dissolution; C

eq

is the equilibrium

concentration of the chemical species; V

p

is the average volume of the soluble grain;

A

p

is the averaged surface area of the soluble grain; k

chemi cal

is the rate constant of

the chemical reaction.

To understand the physical meanings of each term in the Zhao number, Eq. (4.1)

can be rewritten in the following form:

Zh = F

Advection

F

Diffusion

F

Chemical

F

Shape

, (4.2)

where F

Advection

is a term to represent the solute advection; F

Diffusion

is a term to

represent the solute diffusion/dispersion; F

Chemical

is a term to represent the chem-

ical kinetics of the dissolution reaction; F

Shape

is a term to represent the shape fac-

tor of the soluble mineral in the ﬂuid-rock interaction system. These terms can be

expressed as follows:

F

Advection

= ν

ﬂow

, (4.3)

F

Diffusion

=

1

φ

f

D(φ

f

)

, (4.4)

F

Chemical

=

1

k

chemical

C

eq

, (4.5)

F

Shape

=

V

p

A

p

. (4.6)

Equations (4.2), (4.3), (4.4), (4.5) and (4.6) clearly indicate that the Zhao number

is a dimensionless number that can be used to represent the geometrical, hydrody-

namic, thermodynamic and chemical kinetic characteristics of a ﬂuid-rock system in

a comprehensive manner. The condition under which a chemical dissolution front in

a two-dimensional ﬂuid-saturated porous rock becomes unstable can be expressed

by the critical value of this dimensionless number as follows:

4 Algorithm for Simulating Fluid-Rock Interaction Problems 75

Zh

critical

=

(3 −β)(1 +β)

2(1 −β)

, β =

k(φ

0

)

k(φ

f

)

, (4.7)

where Zh

critical

is the critical Zhao number of the ﬂuid-rock interaction system;

k(φ

0

) is the initial permeability corresponding to the initial porosity of the porous

rock; k(φ

f

) is the ﬁnal permeability corresponding to the ﬁnal porosity, φ

f

, of the

porous rock.

Using the concepts of both the Zhao number and the corresponding critical one,

the instability of a chemical dissolution front in a ﬂuid-rock interaction system can

be determined. The focus of this chapter is to deal with ﬂuid-rock interaction sys-

tems of subcritical Zhao numbers, while the focus of the next chapter is to deal with

ﬂuid-rock interaction systems of critical and supercritical Zhao numbers.

In terms of numerical modelling of coupled reactive species transport phenomena

in pore-ﬂuid saturated porous rock masses, we have divided the reactive transport

problems into the following three categories (Zhao et al. 1998a). In the ﬁrst category

of reactive species transport problem, the time scale of the convective/advective ﬂow

is much smaller than that of the relevant chemical reaction in porous rock masses so

that the rate of the chemical reaction can be essentially taken to be zero in the numer-

ical analysis. For this reason, the ﬁrst category of species transport problem is often

called the non-reactive mass transport problem. In contrast, for the second category

of reactive species transport problem, the time scale of the convective/advective

ﬂow is much larger than that of the relevant chemical reaction in pore-ﬂuid satu-

rated porous rock masses so that the rate of the chemical reaction can be essen-

tially taken to be inﬁnite, at least from the mathematical point of view. This means

that the equilibrium state of the chemical reaction involved is always attained in

this category of reactive species transport problem. As a result, the second category

of reactive species transport problem is called the quasi-instantaneous equilibrium

reaction transport problem. The intermediate case between the ﬁrst and the sec-

ond category of reactive species transport problem belongs to the third category

of reactive species transport problem, in which the rate of the relevant chemical

reaction is a positive real number of ﬁnite value. Another signiﬁcant characteris-

tic of the third category of reactive species transport problem is that the detailed

chemical kinetics of chemical reactions must be taken into account. It is the chem-

ical kinetics of a chemical reaction that describes the reaction term in a reactive

species transport equation. Due to different regimes in which a chemical reaction

proceeds, there are two fundamental reactions, namely homogeneous and heteroge-

neous reactions, in geochemistry. For homogeneous reactions, the chemical reaction

takes place solely between reactive aqueous species. However, for heterogeneous

reactions, the chemical reaction takes place at the surfaces between reactive aque-

ous species and solid minerals. This implies that both solid and ﬂuid phases need

to be considered in the numerical modelling of reactive species transport problems

with heterogeneous reactions. Although signiﬁcant achievements have been made

for the numerical modelling of non-reactive species and quasi-instantaneous equi-

librium reaction transport problems, research on the numerical modelling of the

third category of reactive species transport problem is rather limited (Steefel and

76 4 Algorithm for Simulating Fluid-Rock Interaction Problems

Lasaga 1994, Raffensperger and Garven 1995). Considering this fact, we have suc-

cessfully used the ﬁnite element method to model coupled reactive multi-species

transport problems with homogeneous reactions (Zhao et al. 1999a). Here we will

extend the numerical method developed to model coupled reactive multi-species

transport problems with heterogeneous reactions.

4.1 Key Issues Associated with the Numerical Modelling

of Fluid-Rock Interaction Problems

Of central importance to the numerical modelling of ﬂuid-rock interaction problems

in pore-ﬂuid saturated hydrothermal/sedimentary basins is the appropriate consid-

eration of the heterogeneous chemical reaction which takes place slowly between

aqueous reactive species in the pore-ﬂuid and solid minerals in ﬂuid-saturated

porous rock masses. From the rock alteration point of view, the pore-ﬂuid ﬂow,

which carries reactive aqueous species, is the main driving force causing the hetero-

geneous chemical reactions at the interface between the pore-ﬂuid and solid miner-

als so that the rock can be changed from one type into another. The reason for this

is that these reactions between the aqueous species and solid minerals may result

in dissolution of one kind of mineral and precipitation of another and therefore, the

reactive minerals may be changed from one type into another type. Due to the dis-

solution and precipitation of minerals, the porosity of the porous rock mass evolves

during the rock alteration. Since a change in porosity can result in a change in the

pore-ﬂuid ﬂow path, a feed-back loop is formed between the pore-ﬂuid ﬂow and

the transport of reactive chemical species involved in heterogeneous chemical reac-

tions in ﬂuid-rock interaction systems and this porosity change in the pore-ﬂuid

ﬂow needs to be considered in the numerical modelling of ﬂuid-rock interaction

problems. This implies that the average linear velocity of pore-ﬂuid ﬂow varies with

time due to this porosity evolution. Since both the mesh Peclet number and Courant

number are dependent on the average linear velocity, we have to overcome a dif-

ﬁculty in dealing with the problem of variable Peclet and Courant numbers in the

transient analysis of ﬂuid-rock interaction problems. Generally, there are two ways

to overcome this difﬁculty, from the computational point of view. The ﬁrst one is to

use a very ﬁne mesh and very small time step of integration so that the requirements

for both the mesh Peclet number and Courant number can be satisﬁed at every time

step of the computation. The second one is to regenerate the mesh and re-determine

the time step of integration at every time step of computation so that the ﬁnite ele-

ment method can be used effectively. Since either the use of a very ﬁne mesh in the

whole process of computation or the regeneration of the mesh at every time step of

computation is computationally inefﬁcient, there is a deﬁnite need for developing

new numerical algorithms to deal with this kind of problem.

Another important issue related to the numerical modelling of ﬂuid-rock interac-

tion problems is that the dissolution rates of minerals are dependent on the existence

of the dissolving minerals. Once a dissolving mineral is exhausted in the rock mass,

4.2 Development of the Term Splitting Algorithm 77

the dissolution rate of this particular mineral must be identically equal to zero. This

indicates that the variation in the amount of the dissolving mineral should be con-

sidered in the dissolution rates of the minerals. Otherwise, the numerical modelling

may violate the real mechanism of the related chemical kinetics and produce incor-

rect numerical results.

Thus, in order to effectively and efﬁciently use the ﬁnite element method

for solving ﬂuid-rock interaction problems in pore-ﬂuid saturated hydrother-

mal/sedimentary basins, new concepts and numerical algorithms need to be devel-

oped for dealing with the following fundamental issues: (1) Since the ﬂuid-rock

interaction problem involves heterogeneous chemical reactions between reactive

aqueous chemical species in the pore-ﬂuid and solid minerals in the rock masses,

it is necessary to develop a new concept involving the generalized concentration of

a mineral, so that two types of reactive mass transport equations, namely the con-

ventional mass transport equation for the aqueous chemical species in the pore-ﬂuid

and the degenerated mass transport equation for the solid minerals in the rock mass,

can be solved simultaneously in computation. (2) Since the reaction area between

the pore-ﬂuid and mineral surfaces is basically a function of the generalized concen-

tration of the solid mineral, there is a need to consider the appropriate dependence

of the dissolution rate of a dissolving mineral on its generalized concentration in the

numerical analysis. (3) Considering the direct consequence of the porosity evolution

with time in the transient analysis of ﬂuid-rock interaction problems, the problem of

variable mesh Peclet number and Courant number needs to be converted into a prob-

lem involving constant mesh Peclet and Courant numbers, so that the conventional

ﬁnite element method can be directly used to solve ﬂuid-rock interaction problems.

Taking the above-mentioned factors into account, we focus this study on the

numerical modelling of mixed solid and aqueous species transport equations with

consideration of reaction terms from heterogeneous, isothermal chemical reactions.

This means that we will concentrate on the development of new concepts and algo-

rithms so as to solve the ﬂuid-rock interaction problems effectively and efﬁciently,

using the ﬁnite element method. For this purpose, a ﬂuid-rock interaction problem,

in which K-feldspar (KAlSi

3

O

8

) and/or muscovite (KAl

3

Si

3

O

10

(OH)

2

) are dissolved

and muscovite and/or pyrophyllite (Al

2

Si

4

O

10

(OH)

2

) are precipitated, is considered

as a representative example in this chapter.

4.2 Development of the Term Splitting Algorithm

For ﬂuid-rock interaction problems in pore-ﬂuid saturated hydrothermal/

sedimentary basins, heterogeneous chemical reactions take place at the interface

between the reactive aqueous species in the pore-ﬂuid and solid minerals in the

rock mass. This means that we need to deal with two types of transport equa-

tions in the numerical modelling of ﬂuid-rock interaction problems. The ﬁrst is

the conventional transport equation with the advection/convection term and the

diffusion/dispersion term for reactive aqueous species in the pore-ﬂuid. The second

is the degenerated transport equation, in which the advection/convection term and

78 4 Algorithm for Simulating Fluid-Rock Interaction Problems

the diffusion/dispersion term are identically equal to zero, for solid minerals in the

rock mass. In order to solve these two types of transport equations simultaneously,

we need to develop the new concept of the generalized concentration for the solid

mineral. The generalized concentration of a solid mineral is deﬁned as the moles of

the solid mineral per unit total volume (i.e. the volume of void plus the volume of

solid particles) of the porous rock mass. Using this new concept, the general form

of the second type of transport equation to be solved in this study can be expressed

as follows:

∂C

Gi

∂t

= φR

i

(i = 1, 2, . . . , m), (4.8)

where C

Gi

is the generalized concentration of solid mineral/species i; φ is the poros-

ity of the porous rock mass; R

i

is the reaction rate of solid mineral/species i and m

is the number of the solid minerals to be considered in the system.

For the reactive aqueous species, the general form of the ﬁrst type of trans-

port equation in a two dimensional pore-ﬂuid saturated, isotropic and homogeneous

porous medium reads

φ

∂C

Ci

∂t

+u

x

∂C

Ci

∂x

+u

y

∂C

Ci

∂y

= D

e

i

∇

2

C

Ci

+φR

i

(i = m +1, m +2, . . . , n),

(4.9)

where C

Ci

is the conventional concentration of aqueous species i; u

x

and u

y

are the

Darcy velocities of pore-ﬂuid in the x and y directions respectively; D

e

i

is of the

following form:

D

e

i

= φD

i

, (4.10)

where D

i

is the dispersivity of aqueous species i.

It needs to be pointed out that under some circumstances, where the reactant

chemical species constitute only a small fraction of the whole matrix in a porous

medium (Phillips 1991), the total ﬂux of pore-ﬂuid ﬂow in a horizontal aquifer may

be approximately considered as a constant. For example, in a groundwater supply

system, groundwater can be pumped out from a horizontal aquifer at a constant ﬂow

rate. In a geological system, topographically induced pore-ﬂuid ﬂow can also ﬂow

through a horizontal aquifer at a constant ﬂow rate. This indicates that in order to

satisfy the mass conservation requirement of the pore-ﬂuid in the above-mentioned

aquifers, the total ﬂux of the pore-ﬂuid ﬂow should be constant through all vertical

cross-sections, which are perpendicular to the direction of pore-ﬂuid ﬂow. There-

fore, the Darcy velocity in the ﬂow direction is constant if the horizontal aquifer has

a constant thickness. The reason for this is that the Darcy velocity is the velocity

averaged over the total area of a representative elementary area (Zhao et al. 1998e,

Phillips 1991, Nield and Bejan 1992, Zhao et al. 1994c), rather than over the pore

area of the representative elementary area. In other words, the Darcy velocity in

the above-mentioned aquifers can be maintained at a constant value, even though

4.2 Development of the Term Splitting Algorithm 79

the vertical cross-sections, which are perpendicular to the direction of the pore-ﬂuid

ﬂow, may have different values of the areal porosity (Zhao 1998e, Phillips 1991,

Nield and Bejan 1992, Zhao et al. 1994c). However, the linear averaged velocity,

which is the velocity averaged over the pore area of a representative elementary

area, varies if the vertical cross-sections have different values of the areal porosity.

It is also noted that Eq. (4.9) is a typical transport equation for species i. This

equation describes the mass conservation of species i in the pore-ﬂuid saturated

porous medium for both constant and variable porosity cases (Zhao et al. 1998e,

1994c).

In order to consider the feedback effect of porosity due to heterogeneous chemi-

cal reactions, we need a porosity evolution equation in the numerical analysis. Using

the generalized concentration of solid minerals, the evolution equation for the poros-

ity of the rock mass can be expressed as follows:

φ = 1 −

m

¸

i =1

W

i

ρ

s

C

Gi

, (4.11)

where W

i

is the molecular weight of mineral i; ρ

s

is the unit weight of solid minerals.

For the transient analysis, C

Gi

are the functions of space and time variables, so that

the porosity of the rock mass varies with both space and time. This is the reason

why Eq. (4.11) is called the evolution equation of porosity with time.

From the chemical kinetics point of view, the reaction rates involved in Eqs. (4.8)

and (4.9) can be expressed as follows (Lasaga 1984):

R

i

= −

N

¸

j =1

τ

i j

r

j

, (4.12)

where τ

i j

is the contribution factor of mineral/species j to the reaction rate of min-

eral/species i; N is the number of the heterogeneous chemical reactions to be con-

sidered in the rock mass; r

j

is the overall dissolution rate of mineral j.

r

j

=

A

j

V

f

k

j

1 −

Q

j

K

j

, (4.13)

where A

j

is the surface area of mineral j; k

j

and K

j

are the reaction constant and the

equilibrium constant of the jth heterogeneous chemical reaction; V

f

is the volume of

the solution; Q

j

is the chemical afﬁnity of the jth heterogeneous chemical reaction.

It needs to be pointed out that the surface area of a dissolving mineral varies in

the process of mineral dissolution. Once the dissolving mineral is depleted in the

rock mass, the surface area of the mineral must be equal to zero and therefore, the

chemical reaction dissolving this particular mineral should stop. If this area is sim-

ply assumed to be a constant in the numerical modelling, then the resulting numer-

ical solutions are incorrect because the real mechanism of chemical kinetics cannot

be simulated during the consumption of the dissolving mineral in the rock mass.

80 4 Algorithm for Simulating Fluid-Rock Interaction Problems

In other words, the use of a constant surface area cannot simulate the dependent

nature of the chemical reaction rate on a change in the amount of the dissolving

mineral during the heterogeneous chemical reaction process. In addition, the use of

a constant surface area cannot automatically terminate the dissolution reaction, even

though the dissolving mineral is absolutely exhausted in the rock mass. Since the

surface area of a dissolving mineral is strongly dependent on its generalized con-

centration, we can establish the following relation between the surface area and the

generalized concentration of the dissolving mineral:

A

j

V

f

= α

j

C

q

Gj

, (4.14)

where α

j

and q are positive real numbers.

Like the surface area of the dissolving minerals, the values of both α

j

and q are

dependent on the constituents, packing form, grain size and so forth of the minerals.

For simple packing of loose uniform particles of the minerals, these values can be

determined analytically. For example, in the case of packing uniform circles in a

two dimensional domain, q = 1

/

2, whereas in the case of packing uniform spheres

in a three dimensional domain, q = 2

/

3. However, for a real rock mass in pore-ﬂuid

saturated hydrothermal/sedimentary basins, the values of both α

j

and q need to be

determined by ﬁeld measurements and laboratory tests.

Inserting Eq. (4.14) into Eq. (4.13) yields the following equation:

r

j

= α

j

C

q

Gj

k

j

1 −

Q

j

K

j

. (4.15)

Clearly, this equation states that once the dissolving mineral is gradually con-

sumed in the rock mass, the generalized concentration of this mineral continuously

evolves to zero and therefore, its dissolution rate is automatically set to be zero

when it is completely depleted in the numerical analysis. This is the ﬁrst advantage

of introducing the new concept of the generalized concentration for solid minerals.

Another advantage of introducing the concept of the generalized concentration

for solid minerals is that both the ﬁrst type and second type of transport equations

(i.e. Eqs. (4.8) and (4.9)) can be solved simultaneously. Since several heterogeneous

chemical reactions take place simultaneously in dissolution problems involving

multiple minerals, it is important to solve simultaneously all the transport equations

with heterogeneous reaction terms in ﬂuid-rock interaction problems, if we want to

simulate the chemical kinetics of these heterogeneous chemical reactions correctly.

If the transient process of a heterogeneous chemical reaction is of interest,

then the dissolution and precipitation of minerals can result in the variation of

porosity with time. This indicates that the linear average velocity of pore-ﬂuid,

which is involved in the advection/convection term of a reactive aqueous species

transport equation, varies with time during the transient analysis. Even for a hori-

zontal aquifer, in which the horizontal Darcy velocity may be constant, the related

average linear velocity still varies with time, because it is inversely proportional

4.2 Development of the Term Splitting Algorithm 81

to the porosity of the aquifer. In this regard, we have to deal with the problem of

variable Peclet and Courant numbers in the transient analysis of ﬂuid-rock inter-

action problems. To solve this problem effectively and efﬁciently, we develop the

term splitting algorithm in this section. The basic idea behind this algorithm is that

through some mathematical manipulations, we invert the problem of variable Peclet

and Courant numbers into that of constant Peclet and Courant numbers so that a

ﬁxed ﬁnite element mesh and ﬁxed time step of integration can be safely used in the

numerical modelling of ﬂuid-rock interaction problems.

If the pore-ﬂuid saturated porous medium has a non-zero initial porosity, then

we can use this initial porosity as a reference porosity and change Eq. (4.9) into the

following form:

φ

φ

0

φ

0

∂C

Ci

∂t

+u

x

∂C

Ci

∂x

+u

y

∂C

Ci

∂y

= D

e

i

∇

2

C

Ci

+

φ

φ

0

φ

0

R

i

(i = m+1, m+2, . . . , n).

(4.16)

Dividing both the left and right hand sides of Eq. (4.16) by φ

/

φ

0

yields the fol-

lowing equation:

φ

0

∂C

Ci

∂t

+

φ

0

φ

u

x

∂C

Ci

∂x

+

φ

0

φ

u

y

∂C

Ci

∂y

=

φ

0

φ

D

e

i

∇

2

C

Ci

+φ

0

R

i

(i = m+1, m+2, . . . , n).

(4.17)

Using the initial Darcy velocities, u

x0

and u

y0

, as reference velocities in the x and

y directions respectively, Eq. (4.17) can be rewritten as follows:

φ

0

∂C

Ci

∂t

+

¸

u

x0

+

φ

0

φ

u

x

−u

x0

¸

∂C

Ci

∂x

+

¸

u

y0

+

φ

0

φ

u

y

−u

y0

¸

∂C

Ci

∂y

=

φ

0

φ

D

e

i

∇

2

C

Ci

+φ

0

R

i

(i = m +1, m +2, . . . , n). (4.18)

It is noted that in Eq. (4.18), the coefﬁcient in front of ∂C

Ci/

∂x has been split

into a constant coefﬁcient, u

x0

, and a variable term, (u

x

φ

0

)

/

φ − u

x0

. Similarly, the

coefﬁcient in front of ∂C

Ci/

∂y has also been split into a constant coefﬁcient, u

y0

,

and a variable term, (u

y

φ

0

)/φ − u

y0

. This term splitting process, although it is car-

ried out mathematically, forms a very important step in the proposed term splitting

algorithm.

If the ﬁrst derivative terms with constant coefﬁcients are kept in the left hand side

of Eq. (4.18) and the rest terms are moved to the right hand side, Eq. (4.18) can be

rewritten as follows:

φ

0

∂C

Ci

∂t

+u

x0

∂C

Ci

∂x

+u

y0

∂C

Ci

∂y

= D

e

0i

∇

2

C

Ci

+φ

0

R

i

+R

e

i

(i = m+1, m+2, . . . , n),

(4.19)

82 4 Algorithm for Simulating Fluid-Rock Interaction Problems

R

e

i

=

u

x0

−

φ

0

φ

u

x

∂C

Ci

∂x

+

u

yo

−

φ

0

φ

u

y

∂C

Ci

∂y

(i = m +1, m +2, . . . , n),

(4.20)

D

e

0i

= φ

0

D

i

(i = m +1, m +2, . . . , n), (4.21)

where φ

0

is the initial porosity of the porous rock mass; u

x0

and u

y0

are the reference

velocities, which are the initial Darcy velocities, in the x andy directions; R

e

i

is the

equivalent source/sink term due to the variation in both the velocity of pore-ﬂuid

and the porosity of the rock mass.

It is clear that the proposed term splitting algorithm consists of the following

two main steps. First, by means of some rigorous mathematical manipulations,

Eq. (4.9) with variable coefﬁcients in front of the ﬁrst and second derivatives has

been changed into Eq. (4.19), which has constant coefﬁcients in front of the ﬁrst

and second derivatives. From the mathematical point of view, this means that we

have changed a partial differential equation with variable coefﬁcients (Eq. (4.9))

into another one with constant coefﬁcients in front of the ﬁrst and second deriva-

tives. Second, Eq. (4.19) is directly used to obtain a numerical solution for Equation

(4.9) in the ﬁnite element analysis, because both Eqs. (4.19) and (4.9) are mathemat-

ically equivalent. However, from the computational point of view, Eq. (4.19) can be

solved much easier than Eq. (4.9) in the ﬁnite element analysis. The reason for this

is that in order to solve Eq. (4.9), we need to deal with a problem with variable mesh

Peclet number and Courant number. But in order to solve Eq. (4.19), we need only

to deal with a problem with constant mesh Peclet number and Courant number. On

the other hand, the numerical solvers presently available are more stable and robust

when they are used to solve Eq. (4.19), instead of Eq. (4.9).

Note that the equivalent source/sink term presented here has a very clear physical

meaning. For a representative elementary volume/area, a change either in the poros-

ity or in the pore-ﬂuid velocity is equivalent to the addition of a source/sink term

into the representative elementary volume/area. Since the constants involved in both

the advection and dispersion terms are basically independent of time, it is possible to

use the initial mesh and time step, which are determined using the initial conditions

at the beginning of computation, throughout the whole process of the transient anal-

ysis of ﬂuid-rock interaction problems. As a result, the ﬁnite element method with

the proposed term splitting algorithm can be efﬁciently used to solve both the ﬁrst

and second type of transient transport equations (i.e. Eqs. (4.8) and (4.19)) for ﬂuid-

rock interaction problems in pore-ﬂuid saturated hydrothermal/sedimentary basins.

4.3 Application Examples of the Term Splitting Algorithm

In order to illustrate the usefulness and applicability of the newly proposed concepts

and numerical algorithms, we have built them into a ﬁnite element code so that ﬂuid-

rock interaction problems in pore-ﬂuid saturated hydrothermal/sedimentary basins

can be solved effectively and efﬁciently. As shown in Fig. 4.1, the application exam-

4.3 Application Examples of the Term Splitting Algorithm 83

2

CO

3000 m

1000 m

s m u

x

/ 10 , 1 . 0

8

0 0

−

= =

99%

φ

Quartz + 1% K-feldspar

Fig. 4.1 Geometry and

initial conditions for the

ﬂuid-rock interaction

problem in a pore-ﬂuid

saturated aquifer

ple considered in this section is an isothermal ﬂuid-rock interaction problem in a

pore-ﬂuid saturated horizontal aquifer within a sedimentary basin. The topographi-

cally induced pore-ﬂuid ﬂow is horizontally from the left to the right of the aquifer.

This means that the horizontal Darcy velocity is constant within the aquifer. The

rock of the aquifer is initially composed of K-feldspar (KAlSi

3

O

8

) and quartz (SiO

2

).

There is an injection of carbon dioxide gas (CO

2

) at the left inlet of the aquifer. The

injected carbon dioxide ﬂow in the aquifer may dissolve K-feldspar and precipitate

muscovite (KAl

3

Si

3

O

10

(OH)

2

). If the injected carbon dioxide ﬂow is strong enough,

the precipitated muscovite may be dissolved and pyrophyllite ( Al

2

Si

4

O

10

(OH)

2

)

will be precipitated. Basically, the following three overall chemical reactions may

take place in the aquifer.

CO

2

+ H

2

O

fast

=⇒HCO

−

3

+ H

+

, (4.22)

3KAlSi

3

O

8

+2H

+

k

1

=⇒2K

+

+KAl

3

Si

3

O

10

(OH)

2

+6SiO

2

, (4.23)

2KAl

3

Si

3

O

10

(OH)

2

+2H

+

+6SiO

2

k

2

=⇒2K

+

+3Al

2

Si

4

O

10

(OH)

2

. (4.24)

The ﬁrst reaction (i.e. Eq. (4.22)) states that when the injected carbon dioxide

(CO

2

) gas enters the ﬂuid-rock system, it reacts very fast with water (H

2

O), so that

the chemical equilibrium can be reached quasi-instantaneously. This means that the

injection of carbon dioxide (CO

2

) gas is equivalent to the direct injection of H

+

into

the system. As will be demonstrated later, the use of this equivalence may result in

a reduction in the number of primary reactive chemical species and therefore, a

considerable reduction in the degrees of freedom of the whole system. As a direct

consequence, the number of reactive species transport equations can be reduced to

the minimum in computation. This can lead to a signiﬁcant reduction in require-

ments of both the computer storage and CPU time in the numerical modelling of

ﬂuid-rock interaction problems.

It needs to be pointed out that Eq. (4.23) describes a heterogeneous chem-

ical reaction, in which K-feldspar (KAlSi

3

O

8

) is dissolved and muscovite

(KAl

3

Si

3

O

10

(OH)

2

) is precipitated, while Eq. (4.24) describes a heterogeneous

chemical reaction, in which muscovite (KAl

3

Si

3

O

10

(OH)

2

) is dissolved and pyro-

phyllite (Al

2

Si

4

O

10

(OH)

2

) is precipitated in the ﬂuid-rock system. Chemically,

84 4 Algorithm for Simulating Fluid-Rock Interaction Problems

Eq. (4.23) states that the dissolution of one mole of K-feldspar needs to consume

2

/

3 moles of H

+

and then produces 1

/

3 moles of muscovite, 2

/

3 moles of K

+

and

two moles of quartz. Similarly, Eq. (4.24) states that the dissolution of one mole

of muscovite needs to consume one mole of H

+

and two moles of quartz and then

produces 1.5 moles of pyrophyllite and one mole of K

+

. This indicates that we

only need to consider six primary chemical reactive species, namely two aqueous

species (K

+

and H

+

) in the pore-ﬂuid and four solid minerals/species (K-feldspar,

muscovite, pyrophyllite and quartz) in the rock mass, in the computation.

As shown in Fig. 4.1, the computational domain to be used in the numerical anal-

ysis is a rectangle of 3000 m by 1000 m in size. This computational domain is dis-

cretized into 2700 4-node quadrilateral elements. Since the problemto be considered

here is essentially an initial value problem, the following initial conditions are used

in the computation. The initial porosity of the porous medium is 0.1. The initial val-

ues of the generalized concentrations of quartz and K-feldspar are 36.82 kmol/m

3

and 0.87 kmol/m

3

respectively. It is assumed that all the aqueous reactive species

involved in the chemical reactions are in chemical equilibrium at the beginning of

the computation. Under this assumption, the initial values of the conventional con-

centrations of K

+

and H

+

are 0.1 kmol/m

3

and 1.6 × 10

−5

kmol/m

3

. The hori-

zontal Darcy velocity of pore-ﬂuid ﬂow is 10

−8

m/s in the horizontal aquifer. The

dispersion coefﬁcient of K

+

is 2 × 10

−6

m

2

/s. Since the injected carbon dioxide

(CO

2

) gas diffuses much faster than aqueous species, the dispersion coefﬁcient of

the injected H

+

, which is the equivalence of the injected carbon dioxide (CO

2

) gas,

is 100 times that of K

+

in the computation. The concentration of the injected H

+

is 6.4 × 10

−3

kmol/m

3

at the left vertical boundary of the computational domain.

For the purpose of simulating the chemical kinetics of heterogeneous reactions,

the following thermodynamic data are used. The chemical equilibrium constants

are 3.89 × 10

7

and 8318.0 for the K-feldspar and muscovite dissolution reactions,

whereas the chemical reaction rate constants for these two dissolution reactions are

5.03 ×10

−12

kmol/(m

2

s) and 4.48 ×10

−12

kmol/(m

2

s) respectively. α

j

( j = 1, 2)

and q are assumed to be unity and 0.5 in the numerical analysis. In addition, the

integration time step is set to be 3×10

8

s, which is approximately equal to 10 years,

in the computation.

Figures 4.2, 4.3 and 4.4 show the generalized concentration distributions of

K-feldspar, muscovite and pyrophyllite in the ﬂuid-rock interaction system at four

different time instants, namely t = 3 × 10

9

s, t = 1.5 × 10

10

s, t = 6 × 10

10

s and

t = 1.5×10

11

s respectively. It is observed the dissolution front (i.e. the region from

red to blue in Fig. 4.2) of K-feldspar propagates from the left side to the right side of

the computational domain. Since muscovite can be precipitated and dissolved at the

same time, there are two propagation fronts, the precipitation front and dissolution

front, in Fig. 4.3. The precipitation front of muscovite is the region from blue to

red in the right half of the computational domain, whereas the dissolution front is

the region from red to blue in the left half of the computational domain. Both the

precipitation and dissolution fronts propagate from the left side to the right side of

the computational domain. In the case of the generalized concentration distribution

of pyrophyllite (see Fig. 4.4), there only exists the precipitation front (i.e. the region

4.3 Application Examples of the Term Splitting Algorithm 85

0.865E–01

0.860E–01

0.855E–01

0.850E–01

0.845E–01

0.840E–01

0.835E–01

0.830E–01

(t = 3 × 10

9

s)

0.847E–01

0.824E–01

0.800E–01

0.777E–01

0.754E–01

0.791E–01

0.707E–01

0.684E–01

(t = 1.5 × 10

10

s)

0.796E–01

0.729E–01

0.649E–01

0.576E–01

0.502E–01

0.428E–01

0.355E–01

0.281E–01

(t = 6 × 10

10

s)

0.207E–01

0.181E–01

0.155E–01

0.129E–01

0.103E–01

0.775E–02

0.517E–02

0.258E–02

(t = 1.5 × 10

11

s)

Fig. 4.2 Distribution of K-feldspar in the ﬂuid-rock interaction system

86 4 Algorithm for Simulating Fluid-Rock Interaction Problems

0.118E–02

0.109E–02

0.886E–03

0.739E–03

0.591E–03

0.449E–03

0.295E–03

0.148E–03

(t = 3 × 10

9

s)

0.594E–02

0.519E–02

0.445E–02

0.371E–02

0.297E–02

0.229E–02

0.148E–02

0.742E–03

(t = 1.5 × 10

10

s)

0.195E–01

0.170E–01

0.146E–01

0.122E–01

0.974E–02

0.791E–02

0.487E–02

0.244E–02

(t = 6 × 10

10

s)

(t = 1.5 × 10

11

s)

0.247E–01

0.218E–01

0.185E–01

0.154E–01

0.123E–01

0.928E–02

0.617E–02

0.309E–02

Fig. 4.3 Distribution of muscovite in the ﬂuid-rock interaction system

4.3 Application Examples of the Term Splitting Algorithm 87

0.427E–03

0.379E–03

0.320E–03

0.267E–03

0.213E–03

0.160E–03

0.107E–03

0.534E–04

(t = 3 × 10

9

s)

0.402E–02

0.352E–02

0.302E–02

0.251E–02

0.201E–02

0.151E–02

0.101E–02

0.503E–03

(t = 1.5 × 10

10

s)

0.225E–01

0.197E–01

0.169E–01

0.141E–01

0.113E–01

0.844E–02

0.563E–02

0.281E–02

(t = 6 × 10

10

s)

(t = 1.5 × 10

11

s)

0.387E–01

0.339E–01

0.290E–01

0.242E–01

0.194E–01

0.145E–01

0.968E–02

0.484E–02

Fig. 4.4 Distribution of pyrophyllite in the ﬂuid-rock interaction system

88 4 Algorithm for Simulating Fluid-Rock Interaction Problems

from blue to red in Fig. 4.4) of pyrophyllite in the computational domain. Clearly,

all the dissolution/precipitation propagation fronts propagate from the left side to

the right side of the computational domain, which is exactly in the same direction

as the pore-ﬂuid ﬂow in the aquifer. This indicates that when the injected carbon

dioxide (CO

2

) gas enters the ﬂuid-rock interaction system, it produces H

+

quasi-

instantaneously at the left entrance of the system. The produced H

+

propagates

from the left side to the right side of the aquifer so that K-feldspar gradually dis-

solves and muscovite precipitates along the same direction as the dissolution front

propagation of K-feldspar. Since the dissolution reaction constants of K-feldspar

and muscovite are of the same order in magnitude, the dissolution process of mus-

covite is in parallel with that of K-feldspar. This is to say, the precipitated muscovite

from the dissolution of K-feldspar can be dissolved and therefore, pyrophyllite can

be precipitated at an early stage, as clearly exhibited in Fig. 4.4. This fact indicates

that in order to model the chemical kinetics of the involved heterogeneous reactions

correctly, all the reactive species transport equations should be solved simultane-

ously in the numerical analysis.

It is noted that from the mathematical point of view, the problem solved here is

an initial value problem, rather than a boundary value problem. For an initial value

problem in a homogeneous, isotropic porous medium (as we considered here), the

chemical reaction/propagation front of chemically reactive species is stable before

the system reaches a steady state, from both the physical and chemical points of

view. However, if the numerical algorithm is not robust enough and the mesh/time

step used is not appropriate in a numerical analysis, numerical error may result in an

unstable/oscillatory chemical reaction/propagation front (Zienkiewicz 1977). Just

as Phillips (1991) stated, “Numerical calculations can converge to a grid-dependent

limit, and artifacts of a solution can be numerical rather than geological”. This indi-

cates that any numerical solution must be validated, at least qualitatively if the ana-

lytical solution to the problem is not available. We emphasise the importance of this

issue here because it is often overlooked by some purely numerical modellers. The

simplest way to evaluate a numerical solution in a qualitative manner is to check

whether or not the solution violates the common knowledge related to the problem

studied. Since all the propagation fronts of chemically reactive species in Figs. 4.2,

4.3 and 4.4 are comprised of vertically parallel lines, they agree very well with com-

mon knowledge, as the chemical dissolution system considered here is subcritical.

This demonstrates that the proposed numerical algorithms are robust enough for

solving ﬂuid-rock interaction problems, at least from a qualitative point of view.

Figures 4.5, 4.6 and 4.7 show the conventional/generalized concentration distri-

butions of K

+

, H

+

and quartz in the ﬂuid-rock interaction system at four different

time instants, namely t = 3×10

9

s, t = 1.5×10

10

s, t = 6×10

10

s and t = 1.5×10

11

s

respectively. Generally, K

+

is gradually produced and accumulated with time due

to both the K-feldspar and muscovite dissolution reactions, whereas H

+

is con-

tinuously injected at the left entrance of the aquifer and consumed with time due

to both the K-feldspar and muscovite dissolution reactions. Quartz is produced by

the K-feldspar dissolution reaction but is consumed by the muscovite dissolution

reaction. Since the generalized concentration of K-feldspar is greater than that of

4.3 Application Examples of the Term Splitting Algorithm 89

(t = 3 × 10

9

s)

0.103E+00

0.102E+00

0.102E+00

0.102E+00

0.101E+00

0.101E+00

0.101E+00

0.100E+00

(t = 1.5 × 10

10

s)

0.115E+00

0.114E+00

0.113E+00

0.111E+00

0.110E+00

0.108E+00

0.107E+00

0.106E+00

(t = 6 × 10

10

s)

0.160E+00

0.159E+00

0.158E+00

0.157E+00

0.156E+00

0.155E+00

0.154E+00

0.153E+00

(t = 1.5 × 10

11

s)

0.193E+00

0.193E+00

0.192E+00

0.191E+00

0.190E+00

0.189E+00

0.189E+00

0.188E+00

Fig. 4.5 Distribution of K

+

in the ﬂuid-rock interaction system

90 4 Algorithm for Simulating Fluid-Rock Interaction Problems

0.569E–02

0.498E–02

0.427E–02

0.358E–02

0.284E–02

0.219E–02

0.142E–02

0.711E–03

(t = 3 × 10

9

s)

0.569E–02

0.498E–02

0.427E–02

0.356E–02

0.284E–02

0.213E–02

0.142E–02

0.711E–03

(t = 1.5 × 10

10

s)

0.569E–02

0.498E–02

0.427E–02

0.356E–02

0.284E–02

0.219E–02

0.142E–02

0.711E–03

(t = 6 × 10

10

s)

0.590E–02

0.540E–02

0.490E–02

0.440E–02

0.389E–02

0.339E–02

0.289E–02

0.239E–02

(t = 1.5 × 10

11

s)

Fig. 4.6 Distribution of H

+

in the ﬂuid-rock interaction system

4.3 Application Examples of the Term Splitting Algorithm 91

0.405E+02

(t = 3 × 10

9

s)

0.405E+02

0.405E+02

0.405E+02

0.405E+02

0.405E+02

0.405E+02

0.405E+02

(t = 1.5 × 10

10

s)

0.405E+02

0.405E+02

0.405E+02

0.405E+02

0.405E+02

0.405E+02

0.405E+02

0.405E+02

(t = 6 × 10

10

s)

0.405E+02

0.405E+02

0.405E+02

0.406E+02

0.406E+02

0.406E+02

0.406E+02

0.406E+02

0.407E+02

(t = 1.5 × 10

11

s)

0.406E+02

0.406E+02

0.406E+02

0.406E+02

0.406E+02

0.406E+02

0.407E+02

Fig. 4.7 Distribution of quartz in the ﬂuid-rock interaction system

92 4 Algorithm for Simulating Fluid-Rock Interaction Problems

0.773E–03

0.714E–03

0.656E–03

0.597E–03

0.538E–03

0.479E–03

0.421E–03

0.362E–03

(t = 3 × 10

9

s)

0.262E–02

0.299E–02

0.204E–02

0.175E–02

0.146E–02

0.117E–02

0.882E–03

0.592E–03

(t = 1.5 × 10

10

s)

0.829E–02

0.729E–02

0.629E–02

0.529E–02

0.430E–02

0.330E–02

0.230E–02

0.130E–02

(t = 6 × 10

10

s)

0.122E–01

0.118E–01

0.110E–01

0.104E–01

0.976E–02

0.915E–02

0.855E–02

0.794E–02

(t = 1.5 × 10

11

s)

Fig. 4.8 Distribution of porosity variation in the ﬂuid-rock interaction system

4.3 Application Examples of the Term Splitting Algorithm 93

muscovite within the time frame considered, the accumulation of quartz exceeds

the consumption of quartz and therefore, the generalized concentration of quartz

increases with time due to both the K-feldspar and muscovite dissolution reac-

tions. In the case of the conventional concentration of K

+

, its maximum values

are 0.1032 kmol/m

3

and 0.1942 kmol/m

3

at t = 3 × 10

9

s and t = 1.5 × 10

11

s

respectively. This indicates a signiﬁcant increase in the conventional concentration

of K

+

with the increase of time within the scope of this study. However, in the

case of the conventional concentration of H

+

, its maximum value is a constant of

6.4 ×10

−3

kmol/m

3

at both t = 3 ×10

9

s and t = 1.5 ×10

11

s. The reason for this

is due to a continuous injection of H

+

at the left entrance of the aquifer. In the case

of the generalized concentration of quartz, its maximum values are 0.4051 kmol/m

3

and 0.4067 kmol/m

3

at t = 3×10

9

s and t = 1.5×10

11

s respectively. This indicates

a slight increase in the generalized concentration of quartz with the increase of time

within the scope of this study.

Figure 4.8 shows the porosity variation, (φ −φ

0

)

/

φ

0

, in the ﬂuid-rock interac-

tion system at four different time instants. It is clear that the porosity of the porous

medium evolves with time in the process of ﬂuid-rock interactions. The evolution of

porosity mainly depends on the evolution of the K-feldspar dissolution, muscovite

precipitation and dissolution and pyrophyllite precipitation in the ﬂuid-rock interac-

tion system. In addition, the porosity variation front propagates from the left side to

the right side of the aquifer, which is identical to the direction of pore-ﬂuid ﬂow in

the aquifer. The propagation of the porosity variation front can be clearly observed

from the numerical results shown in Fig. 4.8.

In summary, the related numerical solutions from an application example, which

is a K-feldspar dissolution problem in a pore-ﬂuid saturated, isothermal and homo-

geneous aquifer, have demonstrated that: (1) There exist only a dissolution propaga-

tion front for K-feldspar and a precipitation propagation front for pyrophyllite, but

there exist a precipitation propagation front and a dissolution propagation front for

muscovite during the heterogeneous chemical reactions in the aquifer. (2) The dis-

solution of K-feldspar and muscovite may take place simultaneously in the aquifer

so that pyrophyllite can be precipitated at the early stage of the heterogeneous chem-

ical reactions. (3) All the propagation fronts of chemically reactive species are com-

prised of vertically parallel lines, the propagation directions of which are exactly

the same as that of the pore-ﬂuid ﬂow in the aquifer. (4) The evolution of porosity

mainly depends on the evolution of K-feldspar dissolution, muscovite precipitation

and dissolution and pyrophyllite precipitation in the ﬂuid-rock interaction system.

Chapter 5

A Segregated Algorithm for Simulating

Chemical Dissolution Front Instabilities

in Fluid-Saturated Porous Rocks

When fresh pore-ﬂuid ﬂow enters a solute-saturated porous medium, where the

concentration of the solute (i.e. aqueous mineral) reaches its equilibrium concen-

tration, the concentration of the aqueous mineral is diluted so that the solid part of

the solute (i.e. solid mineral) is dissolved to maintain the equilibrium state of the

solution. This chemical dissolution process can result in the propagation of a disso-

lution front within the ﬂuid-saturated porous medium. Due to the dissolution of the

solid mineral, the porosity of the porous medium is increased behind the dissolution

front. Since a change in porosity can cause a remarkable change in permeability,

there is a feedback effect of the porosity change on the pore-ﬂuid ﬂow, according to

Darcy’s law. It is well known that because pore-ﬂuid ﬂow plays an important role in

the process of reactive chemical-species transport, a change in pore-ﬂuid ﬂow can

cause a considerable change in the chemical-species concentration within the porous

medium (Steefel and Lasaga 1990, 1994, Yeh and Tripathi 1991, Raffensperger and

Garven 1995, Shafter et al. 1998a, b, Xu et al. 1999, 2004, Ormond and Ortoleva

2000, Chen and Liu 2002, Zhao et al. 2005a, 2006c). This means that the problem

associated with the propagation of a dissolution front is a fully coupled nonlinear

problem between porosity, pore-ﬂuid pressure and reactive chemical-species trans-

port within the ﬂuid-saturated porous medium. If the fresh pore-ﬂuid ﬂow is slow,

the feedback effect of the porosity change is weak so that the dissolution front is

stable. However, if the fresh pore-ﬂuid ﬂow is fast enough, the feedback effect of

the porosity change becomes strong so that the dissolution front becomes unstable.

In this case, a new morphology (i.e. dissipative structure) of the dissolution front can

emerge due to the self-organization of this coupled nonlinear system. This leads to

an important scientiﬁc problem, known as the reactive inﬁltration instability prob-

lem (Chadam et al. 1986, 1988, Ortoleva et al. 1987), which is closely associated

with mineral dissolution in a ﬂuid-saturated porous medium.

This kind of chemical-dissolution-front instability problem exists ubiquitously

in many scientiﬁc and engineering ﬁelds. For example, in geo-environmental engi-

neering, the rehabilitation of contaminated sites using fresh water to wash the

sites involves the propagation problem of the removed contaminant material front

in water-saturated porous medium. In mineral mining engineering, the extraction

of minerals in the deep Earth using the in-situ leaching technique may result in

C. Zhao et al., Fundamentals of Computational Geoscience,

Lecture Notes in Earth Sciences 122, DOI 10.1007/978-3-540-89743-9 5,

C

Springer-Verlag Berlin Heidelberg 2009

95

96 5 Simulating Chemical Dissolution Front Instabilities in Fluid-Saturated Porous Rocks

the propagation problem of the dissolved mineral front in ﬂuid-saturated porous

medium. In the petroleum industry, the secondary recovery of oil by acidifying the

oil ﬁeld to uniformly increase porosity and hence the yield of oil is associated with

the propagation of the acid-dissolved material front in porous rocks. More impor-

tantly, due to the ever-increasing demand for mineral resources and the likelihood

of the exhaust of the existing ore deposits, it is imperative to develop advanced tech-

niques to explore for new ore deposits. Towards this goal, there is a deﬁnite need

to understand the important physical and chemical processes that control ore body

formation and mineralization in the deep Earth (Raffensperger and Garven 1995,

Zhao et al. 1997a, 1998a, 1999b, d, 2000b, 2001b, d, Gow et al. 2002, Ord et al.

2002, Schaubs and Zhao 2002, Zhao et al. 2002c, 2003a). According to modern

mineralization theory, ore body formation and mineralization is mainly controlled

by pore-ﬂuid ﬂow focusing and the equilibrium concentration gradient of the con-

cerned minerals (Phillips 1991, Zhao et al. 1998a). Since the chemical dissolution

front can create porosity and therefore can locally enhance the pore-ﬂuid ﬂow, it

becomes a potentially powerful mechanism to control ore body formation and min-

eralization in the deep Earth.

Although analytical solutions can be obtained for some reactive transport prob-

lems with simple geometry, it is very difﬁcult, if not impossible, to predict analyti-

cally the complicated morphological evolution of a chemical dissolution front in the

case of the chemical dissolution system becoming supercritical. As an alternative,

numerical methods are suitable to overcome this difﬁculty. Since numerical meth-

ods are approximate solution methods, they must be veriﬁed before they are used

to solve any new type of scientiﬁc and engineering problem. For this reason, it is

necessary to derive the analytical solution for the propagation of a planar dissolution

front within a benchmark problem, the geometry of which can be accurately simu-

lated using numerical methods such as the ﬁnite element method (Zienkiewicz 1977,

Lewis and Schreﬂer 1998) and the ﬁnite difference method. This makes it possible

to compare the numerical solution obtained from the benchmark problem with the

derived analytical solution so that the proposed numerical procedure can be veriﬁed

for simulating chemical-dissolution-front propagation problem in the ﬂuid-saturated

porous medium.

5.1 Mathematical Background of Chemical Dissolution Front

Instability Problems in Fluid-Saturated Porous Rocks

5.1.1 A General Case of Reactive Multi-Chemical-Species

Transport with Consideration of Porosity/Permeability

Feedback

For a pore-ﬂuid-saturated porous medium, Darcy’s law can be used to describe

pore-ﬂuid ﬂow and Fick’s law can be used to describe mass transport phenom-

ena respectively. If both the porosity change of the porous medium is caused by

5.1 Mathematical Background of Chemical Dissolution Front Instability Problems 97

chemical dissolution of soluble solid minerals within the porous medium and the

feedback effect of such a change on the variation of permeability and diffusivity

are taken into account, the governing equations of the coupled nonlinear problem

between porosity, pore-ﬂuid ﬂow and reactive multi-chemical-species transport in

the pore-ﬂuid-saturated porous medium can be expressed as follows:

∂

∂t

(ρ

f

φ) +∇ • (ρ

f

φ u

linear

) = 0, (5.1)

u = φ u

linear

= −

k(φ)

μ

∇p, (5.2)

∂

∂t

(φC

i

) +∇ • (φC

i

u

linear

) = ∇ • [φD

i

(φ)∇C

i

] + R

i

(i = 1, 2, . . . N),

(5.3)

where u

linear

is the averaged linear velocity vector within the pore space of the

porous medium; u is the Darcy velocity vector within the porous medium; p and C

i

are pressure and the concentration (moles/pore-ﬂuid volume) of chemical species i;

μ is the dynamic viscosity of the pore-ﬂuid; φ is the porosity of the porous medium;

D

i

(φ) is the diffusivity of chemical species i; ρ

f

is the density of the pore-ﬂuid; N

is the total number of all the chemical species to be considered in the system; R

i

is the source/sink term of chemical species i due to the dissolution/precipitation of

solid minerals within the system; k(φ) is the permeability of the porous medium.

It is noted that in Eqs. (5.1), (5.2) and (5.3), the chemical species concentration,

the ﬂuid density and averaged linear velocity of the pore-ﬂuid are deﬁned in the

pore space, while the source/sink term and the Darcy velocity of the pore ﬂuid are

deﬁned in the whole medium space (Phillips 1991, Nield and Bejan 1992, Zhao

et al. 1994c).

Since the diffusivity of each chemical species is considered as a function of

porosity, a common phenomenological relation can be used for describing this func-

tion (Bear 1972, Chadam et al. 1986).

D

i

(φ) = D

0i

φ

q

3

2

≤ q ≤

5

2

, (5.4)

where D

0i

is the diffusivity of chemical species i in pure water.

To consider the permeability change caused by a change in porosity, an equa-

tion is needed to express the relationship between permeability and porosity. In

this regard, Detournay and Cheng (1993) state that “The intrinsic permeability k

is generally a function of the pore geometry. In particular, it is strongly dependent

on porosity φ. According to the Carman-Kozeny law (Scheidegger 1974) which

is based on the conceptual model of packing of spheres, a power law relation of

k ∝ φ

3

(1 −φ)

2

exists. Other models based on different pore geometry give sim-

ilar power laws. Actual measurements on rocks, however, often yield power law

relations with exponents for φ signiﬁcantly larger than 3.” Also, Nield and Bejan

98 5 Simulating Chemical Dissolution Front Instabilities in Fluid-Saturated Porous Rocks

(1992) state that “The Carman-Kozeny law is widely used since it seems to be the

best simple expression available.” For these reasons, the Carman-Kozeny law will

be used to calculate permeability k, for a given porosity φ.

k(φ) =

k

0

(1 −φ

0

)

2

φ

3

φ

3

0

(1 −φ)

2

, (5.5)

where φ

0

and k

0

are the initial reference porosity and permeability of the porous

medium respectively.

The source/sink term of chemical species i due to the dissolution/precipitation

of solid minerals within the system can be determined in the following manner

(Chadam et al. 1986). At the particle level, it is assumed that the average volume of

soluble grains is V

p

and that the density of the soluble grains is D

p

, which is deﬁned

as the number of the soluble gains per unit medium volume. If the volume fraction

of insoluble gains is denoted by φ

insoluble

, then the ﬁnal (i.e. maximum) porosity of

the porous medium can be denoted by φ

f

= 1 − φ

insoluble

. In this case, the average

volume of soluble grains can be expressed as follows:

V

p

=

φ

f

−φ

D

p

, (5.6)

At the particle level, the rate of grain-volume change due to a chemical reaction

is denoted by R

p

, so that the rate of porosity change can be expressed as:

∂φ

∂t

= −D

p

R

p

. (5.7)

Without loss of generality, it is assumed that the solid grains are dissolved accord-

ing to the following formula:

Soli d ⇒

N

¸

i =1

χ

i

X

i

, (5.8)

where χ

i

is the stoichiometric coefﬁcient of the ith chemical species; X

i

represents

chemical species i in the pore-ﬂuid.

It is commonly assumed that the rate of grain-volume change due to a chemical

reaction can be expressed as follows (Chadam et al. 1986):

R

p

= k

chemical

A

p

N

¸

i =1

C

χ

i

i

− K

eq

(5.9)

where A

p

is the averaged surface area of soluble grains; k

chemical

and K

eq

are the

rate constant and equilibrium constant of the chemical reaction respectively.

5.1 Mathematical Background of Chemical Dissolution Front Instability Problems 99

If the molar density (i.e. moles per volume) of the soluble grains is denoted by ρ

s

,

then the source/sink term of chemical species i due to the dissolution/precipitation

of solid minerals within the system can be expressed as follows:

R

i

= −χ

i

ρ

s

k

chemical

D

p

A

p

N

¸

i =1

C

χ

i

i

− K

eq

= −χ

i

ρ

s

k

chemical

A

p

V

p

(φ

f

−φ)

N

¸

i =1

C

χ

i

i

− K

eq

.

(5.10)

5.1.2 A Particular Case of Reactive Single-Chemical-Species

Transport with Consideration of Porosity/Permeability

Feedack

If the pore-ﬂuid is incompressible, the governing equations of a reactive single-

chemical-species transport problem in a ﬂuid-saturated porous medium can be writ-

ten as follows:

∂φ

∂t

−∇ • [ψ(φ)∇p] = 0, (5.11)

∂

∂t

(φC) −∇ • [φD(φ)∇C +Cψ(φ)∇p] +ρ

s

k

chemical

A

p

V

p

(φ

f

−φ)(C −C

eq

) = 0,

(5.12)

∂φ

∂t

+k

chemical

A

p

V

p

(φ

f

−φ)(C −C

eq

) = 0, (5.13)

ψ(φ) =

k(φ)

μ

, (5.14)

where C and C

eq

are the concentration and equilibrium concentration of the single

chemical species. Other quantities in Eqs. (5.11), (5.12), (5.13) and (5.14) are of the

same meanings as those deﬁned in Eqs. (5.1), (5.2) (5.3), and (5.9).

Note that Eqs. (5.11) and (5.12) can be derived by substituting the linear average

velocity into Eqs. (5.1) and (5.3) with consideration of a single-chemical species.

It needs to be pointed out that for this single-chemical-species system, it is

very difﬁcult, even if not impossible, to obtain a complete set of analytical solu-

tions for the pore-ﬂuid pressure, chemical species concentration and porosity within

the ﬂuid-saturated porous medium. However, in some special cases, it is possible

to obtain analytical solutions for some variables involved in this single-chemical-

species system. The ﬁrst special case to be considered is a problem, in which a

100 5 Simulating Chemical Dissolution Front Instabilities in Fluid-Saturated Porous Rocks

planar dissolution front propagates in the full space. Since the dissolution front is a

plane, the problemdescribed in Eqs. (5.11), (5.12) and (5.13) degenerates into a one-

dimensional problem. For this particular case, analytical solutions can be obtained

for both the propagation speed of the dissolution front and the downstream pressure

gradient of the pore-ﬂuid. The second special case to be considered is an asymptotic

problem, in which the solid molar density greatly exceeds the equilibrium concen-

tration of the chemical species, implying that the region of a considerable porosity

change propagates very slowly within the ﬂuid-saturated porous medium. In this

particular case, it is possible to derive a complete set of analytical solutions for the

pore-ﬂuid pressure, chemical species concentration and porosity within the ﬂuid-

saturated porous medium. In addition, it is also possible to investigate the reactive

inﬁltration instability associated with the dissolution front propagation in this par-

ticular case (Chadam et al. 1986).

5.1.2.1 The First Special Case

In this special case, the planar dissolution front is assumed to propagate in the pos-

itive x direction, so that all quantities are independent of the transverse coordinates

y and z. For this reason, Eqs. (5.11), (5.12) and (5.13) can be rewritten as follows:

∂φ

∂t

−

∂

∂x

¸

ψ(φ)

∂p

∂x

¸

= 0, (5.15)

∂

∂t

(φC) −

∂

∂x

¸

φD(φ)

∂C

∂x

+Cψ(φ)

∂p

∂x

¸

+ρ

s

k

chemical

A

p

V

p

(φ

f

−φ)(C −C

eq

) = 0,

(5.16)

∂φ

∂t

+k

chemical

A

p

V

p

(φ

f

−φ)(C −C

eq

) = 0. (5.17)

If the chemical species is initially in an equilibrium state and fresh pore-ﬂuid is

injected at the location of x approaching negative inﬁnite, then the boundary condi-

tions of this special problem are expressed as

lim

x→−∞

C = 0, lim

x→−∞

φ = φ

f

, lim

x→−∞

∂p

∂x

= p

fx

(Upstream boundary),

(5.18)

lim

x→∞

C = C

eq

, lim

x→∞

φ = φ

0

, lim

x→∞

∂p

∂x

= p

0x

(Downstream boundary),

(5.19)

where φ

0

is the initial porosity of the porous medium; p

fx

is the pore-ﬂuid pressure

gradient as x approaching negative inﬁnite in the upstreamof the pore-ﬂuid ﬂow; p

0x

is the unknown pore-ﬂuid pressure gradient as x approaching positive inﬁnite in the

downstream of the pore-ﬂuid ﬂow. Since p

fx

drives the pore-ﬂuid ﬂow continuously

5.1 Mathematical Background of Chemical Dissolution Front Instability Problems 101

along the positive x direction, it has a negative algebraic value (i.e. p

fx

< 0) in this

analysis.

If the propagation speed of the planar dissolution front is denoted by v

front

, then

it is possible to transform a moving boundary problem of the dissolution front (in an

x-t coordinate system) into a steady-state boundary problem of the dissolution front

(in an ξ −t coordinate system) using the following coordinate mapping:

ξ = x −v

front

t. (5.20)

It is necessary to relate partial derivatives with respect to ξ and t to those with

respect to x and t (Turcotte and Schubert 1982).

∂

∂t

ξ

=

∂

∂t

x

+

∂

∂x

∂x

∂t

=

∂

∂t

x

+v

front

∂

∂x

, (5.21)

∂

∂ξ

t

=

∂

∂x

t

, (5.22)

where derivatives are taken with x or t held constant as appropriate.

Since the transformed system in the ξ −t coordinate system is in a steady state,

the following equations can be derived from Eqs. (5.21) and (5.22).

∂

∂t

x

= −v

front

∂

∂ξ

, (5.23)

∂

∂ξ

t

=

∂

∂x

t

. (5.24)

Substituting Eqs. (5.23) and (5.24) into Eqs. (5.15), (5.16) and (5.17) yields the

following equations:

∂

∂ξ

¸

ψ(φ)

∂p

∂ξ

+v

front

φ

¸

= 0, (5.25)

∂

∂ξ

¸

φD(φ)

∂C

∂ξ

+Cψ(φ)

∂p

∂ξ

+v

front

(C −ρ

s

)φ

¸

= 0, (5.26)

v

front

∂φ

∂ξ

−k

chemical

A

p

V

p

(φ

f

−φ)(C −C

eq

) = 0. (5.27)

Integrating Eqs. (5.25) and (5.26) from negative inﬁnite to positive inﬁnite and

using the boundary conditions (i.e. Eqs. (5.18) and (5.19)) yields the following

equations:

102 5 Simulating Chemical Dissolution Front Instabilities in Fluid-Saturated Porous Rocks

C

eq

ψ(φ

0

) p

ox

+v

front

φ

0

(C

eq

−ρ

s

) +v

front

φ

f

ρ

s

= 0, (5.28)

ψ(φ

0

) p

ox

+v

front

φ

0

−ψ(φ

f

) p

fx

−v

front

φ

f

= 0. (5.29)

Solving Eqs. (5.28) and (5.29) simultaneously results in the following analytical

solutions:

v

front

=

−ψ(φ

0

) p

0x

C

eq

φ

0

C

eq

+(φ

f

−φ

0

)ρ

s

=

u

0x

C

eq

φ

0

C

eq

+(φ

f

−φ

0

)ρ

s

, (5.30)

p

0x

=

ψ(φ

f

)[φ

0

C

eq

+(φ

f

−φ

0

)ρ

s

]

ψ(φ

0

)[φ

0

C

eq

+(φ

f

−φ

0

)(ρ

s

+C

eq

)]

p

fx

, (5.31)

where u

0x

is the Darcy velocity in the far downstream of the ﬂow as x approaches

positive inﬁnite. Using Darcy’s law, u

0x

can be expressed as

u

0x

=

φ

0

C

eq

+(φ

f

−φ

0

)ρ

s

φ

0

C

eq

+(φ

f

−φ

0

)(ρ

s

+C

eq

)

u

fx

, (5.32)

where u

fx

is the Darcy velocity in the far upstream of the ﬂow as x approaches

negative inﬁnite.

If the ﬁnite element method is used to solve this special problem, the accuracy

of the ﬁnite element simulation can be conveniently evaluated by comparing the

numerical solutions with the analytical ones for both the propagation speed of the

planar dissolution front (i.e. v

front

) and the Darcy velocity in the far downstream of

the ﬂow as x approaches positive inﬁnite (i.e. u

0x

).

5.1.2.2 The Second Special Case (Base Solutions for a Stable State)

Since the solid molar density greatly exceeds the equilibrium concentration of the

chemical species, a small parameter can be deﬁned as follows:

ε =

C

eq

ρ

s

<< 1. (5.33)

To facilitate the theoretical analysis in the limit case of ε approaching zero, the

following dimensionless parameters and variables can be deﬁned.

x =

x

L

∗

, y =

y

L

∗

, z =

z

L

∗

, (5.34)

C =

C

C

eq

, p =

p

p

∗

, u =

u

u

∗

, (5.35)

5.1 Mathematical Background of Chemical Dissolution Front Instability Problems 103

τ =

t

t

∗

ε, (5.36)

where τ is a slow dimensionless time to describe the slowness of the chemical

dissolution that takes place in the system. Other characteristic parameters used in

Eqs. (5.34), (5.35) and (5.36) can be expressed as follows:

t

∗

=

V

p

k

chemical

A

p

C

eq

, L

∗

=

φ

f

D(φ

f

)t

∗

, (5.37)

p

∗

=

φ

f

D(φ

f

)

ψ(φ

f

)

, u

∗

=

φ

f

D(φ

f

)

L

∗

, (5.38)

D

∗

(φ) =

φD(φ)

φ

f

D(φ

f

)

, ψ

∗

(φ) =

ψ(φ)

ψ(φ

f

)

, (5.39)

Substituting Eqs. (5.34), (5.35), (5.36), (5.37), (5.38) and (5.39) into Eqs. (5.11),

(5.12) and (5.13) yields the following dimensionless equations:

ε

∂φ

∂τ

−∇ • [ψ

∗

(φ)∇p ] = 0, (5.40)

ε

∂

∂τ

(φC) −∇ • [D

∗

(φ)∇C +Cψ

∗

(φ)∇p ] −

∂φ

∂τ

= 0, (5.41)

ε

∂φ

∂τ

+(φ

f

−φ)(C −1) = 0. (5.42)

Similarly, the boundary conditions for this special case can be expressed in a

dimensionless form as follows:

lim

x→∞

C = 1, lim

x→∞

φ = φ

0

, lim

x→∞

∂ p

∂x

= p

0x

(downstream boundary),

(5.43)

lim

x→−∞

C = 0, lim

x→−∞

φ = φ

f

, lim

x→−∞

∂ p

∂x

= p

fx

(upstream boundary).

(5.44)

It is noted that the propagation front due to chemical dissolution divides the prob-

lem domain into two regions, an upstream region and a downstream region, relative

to the propagation front. Across this propagation front, the porosity undergoes a

jump from its initial value into its ﬁnal value. Thus, this dissolution-front propa-

gation problem can be considered as a Stefan moving boundary problem (Chadam

et al. 1986). In the limit case of ε approaching zero, the corresponding governing

104 5 Simulating Chemical Dissolution Front Instabilities in Fluid-Saturated Porous Rocks

equations for the dimensionless variables of the problem in both the downstream

region and the upstream region can be expressed below:

C = 1, ∇

2

p = 0, φ = φ

0

(in the downstream region), (5.45)

∇ • (∇C +C∇p) = 0, ∇

2

p = 0, φ = φ

f

(in the upstream region).

(5.46)

If the chemical dissolution front is denoted by S(x, τ) = 0, then the dimension-

less pressure, chemical species concentration and mass ﬂuxes of both the chemical

species and the pore-ﬂuid should be continuous on S(x, τ) = 0. This leads to the

following interface conditions for this moving-front problem:

lim

S→0

−

C = lim

S→0

+

C, lim

S→0

−

p = lim

S→0

+

p, (5.47)

lim

S→0

−

∂C

∂n

= ν

front

(φ

f

−φ

0

), lim

S→0

−

∂ p

∂n

=

ψ(φ

0

)

ψ(φ

f

)

lim

S→0

+

∂ p

∂n

, (5.48)

where n is the unit normal vector of the moving dissolution front.

When the planar dissolution front is under stable conditions, the base solutions

for this special problem can be derived from Eqs. (5.45) and (5.46) with the related

boundary and interface conditions (i.e. Eqs. (5.43), (5.44), (5.47) and (5.48)). The

resulting base solutions are expressed as follows:

C(ξ) = 1, p(ξ) = p

0x

ξ + p

C1

, φ = φ

0

(in the downstream region), (5.49)

C(ξ) = exp(−p

fx

ξ), p(ξ) = p

fx

ξ + p

C2

, φ = φ

f

(in the upstream region),

(5.50)

where p

C1

and p

C2

are two constants to be determined. For example, p

C1

can be

determined by setting the dimensionless pressure p(ξ) to be a constant at a pre-

scribed location of the downstream region, while p

C2

can be determined using the

pressure continuity condition at the interface between the upstream and downstream

regions. Other parameters are deﬁned below:

ξ = x −ν

front

τ, p

0x

=

ψ(φ

f

)

ψ(φ

0

)

p

fx

, ν

front

= −

p

fx

φ

f

−φ

0

. (5.51)

Therefore, if the ﬁnite element method is used to solve the second special prob-

lem, the accuracy of the ﬁnite element simulation can be conveniently evaluated

by comparing the numerical solutions with a complete set of analytical solutions

including porosity, the location of the chemical dissolution front, the dimensionless

chemical-species concentration and the dimensionless pressure.

5.1 Mathematical Background of Chemical Dissolution Front Instability Problems 105

5.1.2.3 The Second Special Case (Perturbation Solutions

for an Unstable State)

When a reactive transport system represented by the above-mentioned second spe-

cial problem is stable, the planar dissolution front remains planar, even though

both small perturbations of the dissolution front and the feedback effect of

porosity/permeability change are simultaneously considered in the analysis. How-

ever, when the reactive transport system is unstable, the planar dissolution front

can change from a planar shape into a complicated one. The instability of the

above-mentioned second special problem can be determined using a linear stabil-

ity analysis (Chadam et al. 1986, 1988, Ortoleva et al. 1987). The main purpose

of conducting such a linear stability analysis is to determine the critical condition

under which the chemical dissolution front of the reactive transport system becomes

unstable.

If a small time-dependent perturbation is added to the planar dissolution front,

then the total solution of the system is equal to the summation of the base solution

and the perturbed solution of the system.

S(ξ, y, τ) = ξ −δ exp(ωτ) cos(my), (5.52)

p

total

(ξ, y, τ) = p(ξ, τ) +δ ˆ p(ξ) exp(ωτ) cos(my), (5.53)

C

total

(ξ, y, τ) = C(ξ, τ) +δ

ˆ

C(ξ) exp(ωτ) cos(my), (5.54)

where ω is the growth rate of the perturbation; m is the wavenumber of the perturba-

tion; δ is the amplitude of the perturbation and δ << 1 by the deﬁnition of a linear

stability analysis.

Since S(ξ, y, τ) is a function of coordinates ξ and y, the following derivatives

exist mathematically:

∂

∂ξ

ξ

=

∂S

∂ξ

∂

∂S

=

∂

∂ξ

S

, (5.55)

∂

∂ y

ξ

=

∂S

∂ y

∂

∂S

+

∂

∂ y

S

=

∂S

∂ y

∂

∂ξ

S

+

∂

∂ y

S

, (5.56)

∂

2

∂ξ

2

ξ

=

∂

2

∂ξ

2

S

, (5.57)

∂

2

∂ y

2

ξ

=

∂

2

S

∂ y

2

∂

∂ξ

+

∂S

∂ y

2

∂

2

∂ξ

2

+2

∂S

∂ y

∂

2

∂ξ∂ y

+

∂

2

∂ y

2

S

. (5.58)

106 5 Simulating Chemical Dissolution Front Instabilities in Fluid-Saturated Porous Rocks

It is noted that the total solutions expressed in Eqs. (5.53) and (5.54) must satisfy

the governing equations that are expressed in Eqs. (5.45) and (5.46). With consid-

eration of Eq. (5.58), the ﬁrst-order perturbation equations of this system can be

expressed as

ˆ

C = 0,

∂

2

ˆ p

∂ξ

2

−m

2

ˆ p +m

2

p

0x

= 0 (in the downstream region), (5.59)

∂

2

ˆ

C

∂ξ

2

+ p

fx

∂

ˆ

C

∂ξ

−m

2

ˆ

C −m

2

p

fx

exp(−p

fx

ξ) − p

fx

exp(−p

fx

ξ)

∂ ˆ p

∂ξ

= 0,

∂

2

ˆ p

∂ξ

2

−m

2

ˆ p +m

2

p

fx

= 0 (in the upstream region). (5.60)

The corresponding boundary conditions of the ﬁrst-order perturbation problem

are:

ˆ

C = 0, lim

x→∞

∂ ˆ p

∂ξ

= 0 (downstream boundary), (5.61)

lim

x→−∞

ˆ

C = 0, lim

x→−∞

∂ ˆ p

∂ξ

= 0 (upstream boundary). (5.62)

Similarly, the interface conditions for this ﬁrst-order perturbation problem can be

expressed as follows:

ˆ

C = 0, lim

S→0

−

ˆ p = lim

S→0

+

ˆ p, (5.63)

lim

S→0

−

∂

ˆ

C

∂n

= ω(φ

f

−φ

0

), lim

S→0

−

∂ ˆ p

∂n

=

ψ(φ

0

)

ψ(φ

f

)

lim

S→0

+

∂ ˆ p

∂n

. (5.64)

Solving Eqs. (5.59) and (5.60) with the boundary and interface conditions (i.e.

Eqs. (5.61) and (5.62)) yields the following analytical results:

ˆ

C = 0, ˆ p(ξ) = p

0x

¸

1 −

1 −β

1 +β

exp(−|m| ξ)

¸

(in the downstream region),

(5.65)

ˆ

C(ξ) = −p

fx

¸

exp(−p

fx

ξ) −

2

1 +β

exp(σξ) +

1 −β

1 +β

exp[(|m| − p

fx

)ξ]

¸

,

ˆ p(ξ) = p

fx

¸

1 +

1 −β

1 +β

exp(|m| ξ)

¸

(in the upstream region),

(5.66)

5.1 Mathematical Background of Chemical Dissolution Front Instability Problems 107

where

β =

ψ(φ

0

)

ψ(φ

f

)

=

k(φ

0

)

k(φ

f

)

, (5.67)

σ =

( p

fx

)

2

+4m

2

− p

fx

2

. (5.68)

Substituting Eq. (5.66) into Eq. (5.64) yields the following equation for the

growth rate of the small perturbation:

ω(m) =

−p

fx

(1 +β)(φ

f

−φ

0

)

[−p

fx

−

( p

fx

)

2

+4m

2

+(1 −β) |m|). (5.69)

Equation (5.69) clearly indicates that the planar dissolution front of the reactive

transport system, which is described by the above-mentioned second special prob-

lem, is stable to short wavelength (i.e. large wavenumber m) perturbations but it is

unstable to long wavelength (i.e. small wavenumber m) perturbations.

Letting ω(m) = 0 yields the following critical condition, under which the reac-

tive transport system can become unstable.

p

fx

critical

= −

(3 −β)(1 +β)

2(1 −β)

, (5.70)

where p

fx

critical

is the critical value of the generalized dimensionless pressure gra-

dient in the far upstream direction as x approaching negative inﬁnite (Zhao et al.

2008e). Since p

fx

critical

is usually of a negative value, the following critical Zhao

number is deﬁned to judge the instability of the reactive transport system:

Zh

critical

= −p

fx

critical

=

(3 −β)(1 +β)

2(1 −β)

. (5.71)

Thus, the Zhao number of the reactive transport system can be deﬁned as follows:

Zh = −p

fx

= −

p

fx

L

∗

p

∗

= −

k(φ

f

)L

∗

p

fx

φ

f

μD(φ

f

)

=

v

ﬂow

φ

f

D(φ

f

)

V

p

k

chemical

A

p

C

eq

. (5.72)

Using Eqs. (5.71) and (5.72), a criterion can be established to judge the instability

of a chemical dissolution front associated with the particular chemical system in this

investigation. If Zh > Zh

critical

, then the chemical dissolution front of the reactive

transport system becomes unstable, while if Zh < Zh

critical

, then the chemical dis-

solution front of the reactive transport system is stable. The case of Zh = Zh

critical

represents a situation where the chemical dissolution front of the reactive transport

108 5 Simulating Chemical Dissolution Front Instabilities in Fluid-Saturated Porous Rocks

system is neutrally unstable, implying that the introduced small perturbation can be

maintained but it does not grow in the corresponding reactive transport system.

Clearly, Eq. (5.72) indicates that for the reactive chemical-species transport con-

sidered in this investigation, the dissolution-enhanced permeability destabilizes the

instability of the chemical dissolution front, while the dissolution-enhanced diffu-

sivity stabilizes the instability of the chemical dissolution front. If the shape factor

of soluble grains is represented by θ = V

p

A

p

, then an increase in the shape fac-

tor of soluble grains can destabilize the instability of the chemical dissolution front,

indicating that the instability likelihood of a porous medium comprised of irregu-

lar grains, is higher than that of a porous medium comprised of regular spherical

grains. Similarly, an increase in either the equilibrium concentration of the chemi-

cal species or the chemical reaction constant of the dissolution reaction can cause

the stabilization of the chemical dissolution front, for the reactive chemical-species

transport considered in this investigation.

To understand the physical meanings of each term in the Zhao number, Equation

(5.72) can be rewritten in the following form:

Zh = F

Advection

F

Diffusion

F

chemical

F

Shape

, (5.73)

where F

Advection

is a term to represent the solute advection; F

Diffusion

is a term to

represent the solute diffusion/dispersion; F

chemical

is a term to represent the chemi-

cal kinetics of the dissolution reaction; F

Shape

is a term to represent the shape fac-

tor of the soluble mineral in the ﬂuid-rock interaction system. These terms can be

expressed as follows:

F

Advection

= v

ﬂow

, (5.74)

F

Diffusion

=

1

φ

f

D(φ

f

)

, (5.75)

F

chemical

=

1

k

chemical

C

eq

, (5.76)

F

Shape

=

V

p

A

p

. (5.77)

Equations (5.73), (5.74), (5.75), (5.75) and (5.77) clearly indicate that the Zhao

number is a dimensionless number that can be used to represent the geometrical,

hydrodynamic, thermodynamic and chemical kinetic characteristics of a ﬂuid-rock

system in a comprehensive manner. This dimensionless number reveals the intimate

interaction between solute advection, solution diffusion/dispersion, chemical kinet-

ics and mineral geometry in a reactive transport system.

5.2 Segregated Algorithm for Simulating the Morphological Evolution 109

5.2 Proposed Segregated Algorithm for Simulating

the Morphological Evolution of a Chemical Dissolution Front

Although analytical solutions can be obtained for the above-mentioned special

cases, it is very difﬁcult, if not impossible, to predict analytically the compli-

cated morphological evolution process of a planar dissolution front in the case of

the chemical dissolution system becoming supercritical. As an alternative, numer-

ical methods are suitable to overcome this difﬁculty. Since numerical methods are

approximate solution methods, they must be veriﬁed before they are used to solve

any new type of scientiﬁc and engineering problem. For this reason, the main pur-

pose of this section is to propose a numerical procedure for simulating how a planar

dissolution front evolves into a complicated morphological front. To verify the accu-

racy of the numerical solution, a benchmark problem is constructed from the theo-

retical analysis in the previous section. As a result, the numerical solution obtained

from the benchmark problem can be compared with the corresponding analytical

solution. After the proposed numerical procedure is veriﬁed, it will be used to sim-

ulate the complicated morphological evolution process of a planar dissolution front

in the case of the chemical dissolution system becoming supercritical.

5.2.1 Formulation of the Segregated Algorithm for Simulating

the Evolution of Chemical Dissolution Fronts

In this section, Eqs. (5.40), (5.41) and (5.42) are solved using the proposed numer-

ical procedure, which is a combination of both the ﬁnite element method and the

ﬁnite difference method. The ﬁnite element method is used to discretize the geo-

metrical shape of the problem domain, while the ﬁnite difference method is used to

discretize the dimensionless time. Since the system described by these equations is

highly nonlinear, the segregated algorithm, in which Eqs. (5.40), (5.41) and (5.42)

are solved separately in a sequential manner, is used to derive the formulation of the

proposed numerical procedure.

For a given dimensionless time-step, τ + Δτ, the porosity can be denoted by

φ

τ+Δτ

= φ

τ

+ Δφ

τ+Δτ

, where φ

τ

is the porosity at the previous time-step and

Δφ

τ+Δτ

is the porosity increment at the current time-step. Using the backward dif-

ference scheme, Eq. (5.42) can be written as follows:

¸

ε

Δτ

+(1 −C

τ+Δτ

)

¸

Δφ

τ+Δτ

= (φ

f

−φ

τ

)(1 −C

τ+Δτ

), (5.78)

where C

τ+Δτ

is the dimensionless concentration at the current time-step; Δτ is the

dimensionless time increment at the current time-step.

Mathematically, there exist the following relationships in the ﬁnite difference

sense:

110 5 Simulating Chemical Dissolution Front Instabilities in Fluid-Saturated Porous Rocks

ε

∂

φC

∂τ

= ε

Δ

φ

τ+Δτ

C

τ+Δτ

Δτ

= εC

τ+Δτ

Δφ

τ+Δτ

Δτ

+εφ

τ+Δτ

Δ

C

τ+Δτ

Δτ

,

(5.79)

ε

∂φ

∂τ

= ε

Δ(φ

τ+Δτ

)

Δτ

=

1 −C

τ+Δτ

φ

f

−φ

τ+Δτ

, (5.80)

∇ •

¸

D

∗

(φ)∇C

¸

= ∇ •

¸

D

∗

(φ

τ+Δτ

)∇C

τ+Δτ

¸

, (5.81)

∇ •

¸

Cψ

∗

(φ)∇p

¸

= C∇ •

¸

ψ

∗

(φ)∇p

¸

+∇p •

¸

ψ

∗

(φ)∇C

¸

= C

τ+Δτ

∇ •

¸

ψ

∗

(φ

τ+Δτ

)∇p

τ+Δτ

¸

+∇p

τ+Δτ

•

¸

ψ

∗

(φ

τ+Δτ

)∇C

τ+Δτ

¸

(5.82)

Substituting Eqs. (5.79), (5.80), (5.81) and (5.82) into Eq. (5.41) yields the fol-

lowing ﬁnite difference equation:

¸

ε

Δτ

φ

τ+Δτ

+

1

ε

(φ

f

−φ

τ+Δτ

)

¸

C

τ+Δτ

−∇ •

¸

D

∗

(φ

τ+Δτ

)∇C

τ+Δτ

¸

−∇p

τ+Δτ

•

¸

ψ

∗

(φ

τ+Δτ

)∇C

τ+Δτ

¸

=

ε

Δτ

φ

τ+Δτ

C

τ

+

1

ε

φ

f

−φ

τ+Δτ

(5.83)

Similarly, Eq. (5.40) can be rewritten in the following discretized form:

∇ •

¸

ψ

∗

(φ)∇p

¸

= ∇ •

¸

ψ

∗

(φ

τ+Δτ

)∇p

τ+Δτ

¸

= (1 −C

τ+Δτ

)(φ

f

−φ

τ+Δτ

). (5.84)

Using the proposed segregated scheme and ﬁnite element method, Eqs. (5.78),

(5.83) and (5.84) are solved separately and sequentially for the porosity, dimension-

less concentration and dimensionless pressure at the current time-step. Note that

when Eq. (5.78) is solved using the ﬁnite element method, the dimensionless con-

centration at the current time-step is not known. Similarly, when Eq. (5.83) is solved

using the ﬁnite element method, the dimensionless pressure at the current time-step

remains unknown. This indicates that these three equations are fully coupled so that

an iteration scheme needs to be used to solve them sequentially. At the ﬁrst iter-

ation step, the dimensionless concentration at the previous time-step is used as a

reasonable guess of the dimensionless concentration at the current time-step when

Eq. (5.78) is solved for the porosity. In a similar way, the dimensionless pressure at

the previous time-step is used as a reasonable guess for the current time-step when

Eq. (5.83) is solved for the dimensionless concentration. The resulting approximate

porosity and dimensionless concentration can be used when Eq. (5.84) is solved for

the dimensionless pressure. At the second iteration step, the same procedure as used

in the ﬁrst iteration step is followed, so that the following convergence criterion can

be established after the second iteration step.

5.2 Segregated Algorithm for Simulating the Morphological Evolution 111

E = Max

⎛

⎝

N

φ

¸

i =1

φ

k

i,τ+Δτ

−φ

k−1

i,τ+Δτ

2

,

N

C

¸

i =1

C

k

i,τ+Δτ

−C

k−1

i,τ+Δτ

2

,

N

p

¸

i =1

p

k

i,τ+Δτ

− p

k−1

i,τ+Δτ

2

⎞

⎟

⎠

< E,

(5.85)

where E and E are the maximum error at the k-th iteration step and the allowable

error limit; N

φ

, N

C

and N

p

are the total numbers of the degrees-of-freedom for the

porosity, dimensionless concentration and dimensionless pressure respectively; k is

the index number at the current iteration step and k − 1 is the index number at the

previous iteration step; φ

k

i,τ+Δτ

, C

k

i,τ+Δτ

and p

k

i,τ+Δτ

are the porosity, dimensionless

concentration and dimensionless pressure of node i at both the current time-step and

the current iteration step; φ

k−1

i,τ+Δτ

, C

k−1

i,τ+Δτ

and p

k−1

i,τ+Δτ

are the porosity, dimension-

less concentration and dimensionless pressure of node i at the current time-step but

at the previous iteration step. It is noted that k ≥ 2 in Eq. (5.85).

The convergence criterion is checked after the second iteration step. If the con-

vergence criterion is not met, then the iteration is repeated at the current time-step.

Otherwise, the convergence solution is obtained at the current time step and the

solution procedure goes to the next time-step until the ﬁnal time-step is reached.

5.2.2 Veriﬁcation of the Segregated Algorithm for Simulating

the Evolution of Chemical Dissolution Fronts

The main and ultimate purpose of a numerical simulation is to provide numeri-

cal solutions for practical problems in a real world. These practical problems are

impossible and impractical to solve analytically. Since numerical methods are the

basic foundation of a numerical simulation, only an approximate solution can be

obtained from a computational model, which is the discretized description of a con-

tinuum mathematical model. Due to inevitable round-off errors in computation and

discretized errors in temporal and spatial variables, it is necessary to verify the pro-

posed numerical procedure so that meaningful numerical results can be obtained

from a discretized computational model. For this reason, a benchmark problem, for

which the analytical solutions are available, is considered in this section.

Figure 5.1 shows the geometry and boundary conditions of the coupled problem

between porosity, pore-ﬂuid pressure and reactive chemical-species transport within

a ﬂuid-saturated porous medium. For this benchmark problem, the dimensionless-

pressure gradient (i.e. p

fx

= −1) is applied on the left boundary, implying that there

is a horizontal throughﬂow from the left to the right of the computational model. In

this case, the Zhao number of the reactive transport system is unity. The dimension-

less height and width of the computational model are 5 and 10 respectively. Except

for the left boundary, the initial porosity of the porous medium is 0.1, while the

112 5 Simulating Chemical Dissolution Front Instabilities in Fluid-Saturated Porous Rocks

0

y

10 =

x

L

fx

f

p

x

p

C C

′ =

∂

∂

=

=

τ) (

0 , 0 =

∂

∂

=

∂

∂

y

p

y

C

0 , 0 =

∂

∂

=

∂

∂

y

p

y

C

5 =

y

L

1 =

=

C

o

Inflow

x

φ

φ φ

φ

Fig. 5.1 Geometry and boundary conditions of the reactive inﬁltration problem

initial dimensionless-concentration is one within the computational domain. The

ﬁnal porosity after depletion of the soluble mineral is 0.2. This ﬁnal porosity is

applied on the left boundary as a boundary condition of the computational domain.

The permeability of the porous medium is calculated using the Carman-Kozeny for-

mula, which has the power of 3 in the power law. The diffusivity of chemical species

is calculated using the power law, which has the power of 2. Both the top and the

bottom boundaries are assumed to be impermeable for the pore-ﬂuid and chemical

species. The ratio of the equilibrium concentration to the solid molar density of the

chemical species is assumed to be 0.01, while the dimensionless time-step is set

to be 0.005 in the computation. Since the computational domain of the benchmark

problem is of ﬁnite size, a time-dependent-dimensionless-concentration boundary

condition (i.e. C(τ) = exp( p

fx

ν

front

τ)) needs to be applied on the left boundary so

that the numerical solutions can be compared with the analytical solutions derived

in the previous section. Using the above-mentioned parameters, the critical Zhao

number of the system is approximately equal to 1.77. Since the Zhao number of

the system is smaller than its critical value, the coupled system considered in this

section is sub-critical so that a planar dissolution front remains planar during its

propagation within the system. The dimensionless speed of the dissolution front

propagation is equal to 10, which is determined using Eq. (5.51). To simulate appro-

priately the propagation of the dissolution front, the whole computational domain

is simulated by 19701 four-node rectangular elements of 20000 nodal points in

total.

Figures 5.2, 5.3 and 5.4 show the comparison of numerical solutions with analyt-

ical ones for the porosity, dimensionless concentration and dimensionless pressure

distributions within the computational domain at three different time instants. In

these ﬁgures, the thick line shows the numerical results, while the thin line shows

the corresponding analytical solutions, which can be determined from Eqs. (5.49)

5.2 Segregated Algorithm for Simulating the Morphological Evolution 113

(τ = 0.25)

0.2

0.1

0 10

P

o

r

o

s

i

t

y

x

(τ = 0.625)

0.2

0.1

P

o

r

o

s

i

t

y

0 10

x

0.2

0.1

P

o

r

o

s

i

t

y

(τ = 0.8)

0 10

x

Fig. 5.2 Comparison of numerical solutions with analytical ones at different time instants (Poros-

ity): the thick line shows the numerical results, while the thin line shows the corresponding analyt-

ical solutions

and (5.50) with the boundary condition of p(L

x

, τ) = 100 at the right boundary

of the computational model. The resulting analytical solutions are expressed as

follows:

C(x, τ) = 1, φ(x, τ) = φ

0

(x > ν

front

τ), (5.86)

p(x, τ) = −p

0x

(L

x

− x) +100 (x > ν

front

τ), (5.87)

C(x, τ) = exp[−p

fx

(x −ν

front

τ)], φ(x, τ) = φ

f

(x ≤ ν

front

τ), (5.88)

p(x, τ) = p

fx

(x −ν

front

τ) − p

0x

(L

x

−ν

front

τ) +100 (x ≤ ν

front

τ). (5.89)

From these results, it can be observed that the numerical solutions agree very

well with the analytical solutions, indicating that the proposed numerical procedure

is capable of simulating the planar dissolution-front propagation within the ﬂuid-

saturated porous medium. As expected, the porosity propagation front is the sharpest

114 5 Simulating Chemical Dissolution Front Instabilities in Fluid-Saturated Porous Rocks

0 10

0

D

i

m

e

n

s

i

o

n

l

e

s

s

c

o

n

c

e

n

t

r

a

t

i

o

n

1

x

(τ = 0.25)

0

D

i

m

e

n

s

i

o

n

l

e

s

s

c

o

n

c

e

n

t

r

a

t

i

o

n

1

0 10

x

(τ = 0.625)

0

D

i

m

e

n

s

i

o

n

l

e

s

s

c

o

n

c

e

n

t

r

a

t

i

o

n

1

0 10

x

(τ = 0.8)

Fig. 5.3 Comparison of numerical solutions with analytical ones at different time instants (Dimen-

sionless concentration): the thick line shows the numerical results, while the thin line shows the

corresponding analytical solutions

(τ = 0.8)

180

140

D

i

m

e

n

s

i

o

n

l

e

s

s

p

r

e

s

s

u

r

e

100

0 10

x

(τ = 0.625)

180

150

D

i

m

e

n

s

i

o

n

l

e

s

s

p

r

e

s

s

u

r

e

100

0 10

x

(τ = 0.25)

0 10

100

D

i

m

e

n

s

i

o

n

l

e

s

s

p

r

e

s

s

u

r

e

180

x

Fig. 5.4 Comparison of numerical solutions with analytical ones at different time instants (Dimen-

sionless pressure): the thick line shows the numerical results, while the thin line shows the corre-

sponding analytical solutions

5.3 Application of the Segregated Algorithm 115

one among the three propagation fronts, namely a porosity propagation front, a

dimensionless-concentration propagation front and a dimensionless-pressure propa-

gation front, in the computational model. Clearly, the dimensionless-pressure prop-

agation front has the widest bandwidth, implying that it is the least sharp front

in the computational model. Although there are some smoothing effects on the

numerically-simulated propagation fronts as a result of numerical dispersion, the

propagation speed of the numerically-simulated propagation front is in good coinci-

dence with that of the analytically-predicted propagation front. For this benchmark

problem, the overall accuracy of the numerical results is indicated by the dimension-

less pressure. The maximum relative error of the numerically-simulated dimension-

less pressure is 2.2%, 4.6% and 5.8% for dimensionless times of 0.25, 0.625 and 0.8

respectively. If both a small mesh size and a small time step are used, then the max-

imum relative error can be further reduced in the numerical simulation. This quan-

titatively demonstrates that the proposed numerical procedure can produce accurate

numerical solutions for the planar dissolution-front propagation problem within a

ﬂuid-saturated porous medium.

5.3 Application of the Segregated Algorithm for Simulating

the Morphological Evolution of Chemical Dissolution Fronts

In this section, the proposed numerical procedure is used to simulate the morpho-

logical evolution of a chemical dissolution front in a supercritical system. For this

purpose, a dimensionless-pressure gradient (i.e. p

fx

= −10) is applied on the left

boundary of the computational domain so that the dimensionless speed of the dis-

solution front propagation is equal to 100. This means that the dissolution front

propagates much faster than it does within the system considered in the previous

section. Due to this change, the ratio of the equilibrium concentration to the solid

molar density of the chemical species is assumed to be 0.001, while the dimension-

less time-step is also assumed to be 0.001 in the computation. The Zhao number of

the system is increased to 10, which is greater than the critical Zhao number (i.e.

approximately 1.77) of the system. The values of other parameters are exactly the

same as those used in the previous section. Since the Zhao number of the system is

greater than its critical value, the coupled system considered in this section is super-

critical so that a planar dissolution front evolves into a complicated morphology

during its propagation within the system. In order to simulate the instability of the

chemical dissolution front, a small perturbation of 1% initial porosity is randomly

added to the initial porosity ﬁeld in the computational domain.

Figure 5.5 shows the porosity distributions due to the morphological evolution of

the chemical dissolution front in the ﬂuid-saturated porous medium, while Fig. 5.6

shows the dimensionless concentration distributions due to the morphological evo-

lution of the chemical dissolution front within the computational domain. It is

observed that for the values of the dimensionless time greater than 0.03, the ini-

tial planar dissolution front gradually changes into an irregular one. With a fur-

ther increase of the dimensionless time, the amplitude of the resulting irregular

116 5 Simulating Chemical Dissolution Front Instabilities in Fluid-Saturated Porous Rocks

(Numerical, 03 . 0 = (Numerical, )

)

)

04 . 0 = )

(Numerical, 05 . 0 = (Numerical, 06) . 0 =

(Numerical, 07 . 0 = (Numerical, 08) . 0 = τ τ

τ τ

τ τ

Fig. 5.5 Porosity distributions due to morphological evolution of the chemical dissolution front in

the ﬂuid-saturated porous medium

dissolution front increases signiﬁcantly, indicating that the chemical dissolution

front is morphologically unstable during its propagation within the computational

model. Although both the porosity and the dimensionless concentration have a simi-

lar propagation front, the distribution of their maximum values along the dissolution

front is clearly different. The peak value of the porosity is in good correspondence

with the trough value of the dimensionless concentration due to the chemical dis-

solution in the system. This demonstrates that the proposed numerical procedure

is capable of simulating the morphological instability of the chemical dissolution

front in a ﬂuid-saturated porous medium in the case of the coupled system being

supercritical.

5.3 Application of the Segregated Algorithm 117

(Numerical, 03) . 0 = (Numerical, 04) . 0 =

(Numerical, 05) . 0 = (Numerical, 06) . 0 =

τ τ

τ τ

(Numerical, 07 . 0 = (Numerical, 08) . 0 = τ τ )

Fig. 5.6 Dimensionless concentration distributions due to morphological evolution of the chemical

dissolution front in the ﬂuid-saturated porous medium

It is interesting to investigate how the dimensionless pressure and pore-ﬂuid ﬂow

evolve with time during propagation of the unstable dissolution front in the com-

putational model. Figure 5.7 shows the dimensionless pressure distributions dur-

ing the morphological evolution of the chemical dissolution front. It is noted that

although the dimensionless pressure is continuous, there exists a clear transition

for the dimensionless pressure-gradient distribution in the computational model.

This phenomenon can be clearly seen at the late stages of the numerical simula-

tion such as when the dimensionless time is equal to 0.06 and 0.07. The ﬂuid-ﬂow

pattern evolution during the propagation of the unstable dissolution front is exhib-

ited by the streamline evolution in the computational model. Figure 5.8 shows the

118 5 Simulating Chemical Dissolution Front Instabilities in Fluid-Saturated Porous Rocks

(Numerical, 03 . 0 = ) (Numerical, 04) . 0 =

(Numerical, 05) . 0 = (Numerical, 06) . 0 =

(Numerical, 07 . 0 = ) (Numerical, 08) . 0 = τ τ

τ τ

τ τ

Fig. 5.7 Dimensionless pressure distributions due to morphological evolution of the chemical

dissolution front in the ﬂuid-saturated porous medium

streamline distributions during the morphological evolution of the chemical dissolu-

tion front within the coupled system between porosity, pore-ﬂuid pressure and reac-

tive chemical-species transport. Due to the growth of the amplitude of the irregular

dissolution front, pore-ﬂuid ﬂow focusing takes place in the peak range of the poros-

ity, which can be observed from the streamline density (in Fig. 5.8). It is noted that

the width of the ﬂow focusing zone is closely associated with the peak and trough

values of the irregular dissolution front in the computational model. Since both the

porosity generation and the pore-ﬂuid ﬂow focusing play an important role in ore

body formation and mineralization, the proposed numerical procedure can provide

a useful tool for simulating the related physical and chemical processes associated

with the generation of giant ore deposits within the upper crust of the Earth.

5.3 Application of the Segregated Algorithm 119

(τ = 0.06) (τ = 0.07)

(τ = 0.08)

Fig. 5.8 Streamline distributions due to morphological evolution of the chemical dissolution front

in the ﬂuid-saturated porous medium

Chapter 6

A Decoupling Procedure for Simulating Fluid

Mixing, Heat Transfer and Non-Equilibrium

Redox Chemical Reactions in Fluid-Saturated

Porous Rocks

Non-equilibrium redox chemical reactions of high orders are ubiquitous in

ﬂuid-saturated porous rocks within the crust of the Earth. They play a very impor-

tant role in ore body formation and alteration closely associated with a miner-

alizing system. Since pore-ﬂuid is a major carrier transporting chemical species

from one part of the crust into another, the chemical process is coupled with

the pore-ﬂuid ﬂow process in ﬂuid-saturated porous rocks. In addition, if the

rate of a chemical reaction is dependent on temperature, the chemical process is

also coupled with the heat transfer process. When a pore-ﬂuid carrying one type

of chemical species meets with that carrying another type of chemical species,

these two types of pore-ﬂuids can mix together to allow the related chemical

reaction to take place due to solute molecular diffusion/dispersion and advec-

tion. For these reasons, the resulting patterns of mineral dissolution, transporta-

tion, precipitation and rock alteration are a direct consequence of coupled processes

between ﬂuids mixing, heat transfer and chemical reactions in ﬂuid-saturated porous

rocks.

Due to ever-increasing demands for minerals and possible exhaustion of the

existing mineral deposits in the foreseeable future, understanding the controlling

mechanisms behind ore body formation and mineralization within the upper crust

of the Earth becomes a very important research ﬁeld. There is no doubt that an

improved understanding of ore forming processes can signiﬁcantly promote mineral

exploration for new ore deposits within the upper crust of the Earth. Although exten-

sive studies have been carried out to understand the possible physical and chemical

processes associated with ore body formation and mineralization (Phillips 1991, Yeh

and Tripathi 1991, Nield and Bejan 1992, Steefel and Lasaga 1994, Raffensperger

and Garven 1995, Schafer et al. 1998a, b, Zhao et al. 1997a, 1998a, 1999b, Xu

et al. 1999, Zhao et al. 2000b, 2001b, 2001d, 2002c, Schaubs and Zhao 2002, Ord

et al. 2002, Gow et al. 2002, Zhao et al. 2003a), the kinetics of a redox chemi-

cal reaction and its interaction with physical processes are often overlooked in the

numerical modelling of ore forming systems. In most chemical reactions associated

with an ore forming system, the chemical reaction rate is ﬁnite so that an interaction

between the solute molecular diffusion/dispersion, advection and chemical kinetics

must be considered.

C. Zhao et al., Fundamentals of Computational Geoscience,

Lecture Notes in Earth Sciences 122, DOI 10.1007/978-3-540-89743-9 6,

C

Springer-Verlag Berlin Heidelberg 2009

121

122 6 Fluid Mixing, Heat Transfer and Non-Equilibrium Redox Chemical Reactions

In terms of numerical modelling of coupled problems between ﬂuids mixing,

heat transfer and chemical reactions in ﬂuid-saturated porous rocks, it is possible to

divide the coupled problems into the following three categories (Zhao et al. 1998a).

In the ﬁrst category of coupled problem, the time scale of the advective ﬂow is much

smaller than that of the relevant chemical reaction in porous rock masses so that the

rate of the chemical reaction can be essentially taken to be zero in the numerical

analysis. For this reason, the ﬁrst category of coupled problem is often called the

non-reactive mass transport problem. In contrast, for the second category of cou-

pled problem, the time scale of the advective ﬂow is much larger than that of the

relevant chemical reaction in pore-ﬂuid saturated porous rocks so that the rate of the

chemical reaction can be essentially taken to be inﬁnite, at least from the mathemat-

ical point of view. This means that the equilibrium state of the chemical reaction

involved is always attained in this category of coupled problem. As a result, the

second category of coupled problem is called the quasi-instantaneous equilibrium

problem. The intermediate case between the ﬁrst and the second category belongs

to the third category of coupled problem, in which the rate of the relevant chemical

reaction is a positive real number of ﬁnite value. Another signiﬁcant characteristic

of the third category of coupled problem is that the detailed kinetics of the chem-

ical reactions must be taken into account. It is the kinetics of a chemical reaction

that describes the reaction term in a reactive species transport equation. If a redox

chemical reaction is considered, both the forward reaction rate and the backward

one need to be included in the reaction term of a reactive species transport equation.

Although signiﬁcant achievements have been made for the numerical modelling of

non-reactive species and quasi-instantaneous equilibrium reaction transport prob-

lems, research on the numerical modelling of the third category of coupled prob-

lem with redox chemical reactions is rather limited. Considering this fact, we will

develop a numerical procedure to solve coupled problems between ﬂuids mixing,

heat transfer and redox chemical reactions in ﬂuid-saturated porous rocks.

Large geological faults and cracks are favorable locations for ﬂuids carrying dif-

ferent chemical species to focus and mix. For this reason, ore body formation and

mineralization are often associated with geological faults and cracks. When the per-

meability of a fault/crack is much bigger than that of the surrounding rock, the

pore-ﬂuid ﬂow in the fault/crack is much faster than that in the surrounding rock.

This implies that an interaction between the solute diffusion, advection and chemi-

cal kinetics is very strong within and around a fault/crack. Although it is well known

that ore body formation and mineralization are associated with geological faults and

cracks, the major factors controlling the reaction patterns within and around large

faults and cracks remains unclear.

Keeping the above-mentioned considerations in mind, a numerical approach

based on the ﬁnite element method is used to solve coupled problems between

ﬂuids mixing, heat transfer and redox chemical reactions in ﬂuid-saturated porous

rocks. In order to improve the efﬁciency of numerical modelling, the concept of the

chemical reaction rate invariant is used to convert the conventional reactive trans-

port equations with strong chemical reaction terms into the following two different

6.1 Problems between Fluids Mixing, Heat Transfer and Redox Chemical Reactions 123

kinds of equations: One is the same as the ﬁrst category of mass transport equation

without any reaction term; while another remains the same as the third category of

reactive transport equation with a strong reaction term. Since the solution of a reac-

tive transport equation with a chemical reaction term is computationally much more

expensive than that of a mass transport equation without a chemical reaction term,

any reduction in the total number of reactive transport equations can signiﬁcantly

save computer time in a numerical computation. Based on this idea, a decoupling

numerical procedure, which can be used to remove the coupling between redox

types of reactive transport equations, is proposed to solve coupled problems between

ﬂuids mixing, heat transfer and redox chemical reactions in ﬂuid-saturated porous

rocks. This allows the interaction between the solute molecular diffusion/dispersion,

advection and chemical kinetics to be investigated within and around faults and

cracks in the upper crust of the Earth.

6.1 Statement of Coupled Problems between Fluids Mixing, Heat

Transfer and Redox Chemical Reactions

For pore-ﬂuid saturated porous rocks, Darcy’s law can be used to describe pore-

ﬂuid ﬂow and the Oberbeck-Boussinesq approximation is employed to describe a

change in pore-ﬂuid density due to a change in the pore-ﬂuid temperature. Fourier’s

law and Fick’s law can be used to describe the heat transfer and mass transport

phenomena respectively. If the pore-ﬂuid is assumed to be incompressible, the

governing equations of the coupled problem between ﬂuids mixing, heat transfer

and redox chemical reactions in ﬂuid-saturated porous rocks can be expressed as

follows:

∂u

∂x

+

∂ν

∂y

= 0, (6.1)

u =

K

x

μ

−

∂ P

∂x

, (6.2)

ν =

K

y

μ

−

∂ P

∂y

+ρ

f

g

, (6.3)

¸

φρ

f

c

pf

+(1 −φ)ρ

s

c

ps

¸

∂T

∂t

+ρ

f

c

pf

u

∂T

∂x

+ν

∂T

∂y

= λ

ex

∂

2

T

∂x

2

+λ

ey

∂

2

T

∂y

2

,

(6.4)

φ

∂C

i

∂t

+

u

∂C

i

∂x

+ν

∂C

i

∂y

=

D

ex

∂

2

C

i

∂x

2

+ D

ey

∂

2

C

i

∂y

2

+φR

i

(i = 1, 2, . . . ,N),

(6.5)

ρ

f

= ρ

f 0

[1 −β

T

(T − T

0

)], (6.6)

124 6 Fluid Mixing, Heat Transfer and Non-Equilibrium Redox Chemical Reactions

λ

ex

= φλ

fx

+(1 −φ)λ

sx

, λ

ey

= φλ

fy

+(1 −φ)λ

sy

, (6.7)

D

ex

= φD

fx

, D

ey

= φD

fy

, (6.8)

where u and v are the horizontal and vertical velocity components of the pore-ﬂuid in

the x and y directions respectively; P is the pore-ﬂuid pressure; T is the temperature

of the porous medium; C

i

is the concentration of chemical species i; N is the total

number of the active chemical species considered in the pore-ﬂuid; K

x

and K

y

are

the permeabilities of the porous medium in the x and y directions respectively; μ is

the dynamic viscosity of the pore-ﬂuid; ρ

f

and ρ

s

are the densities of the pore-ﬂuid

and solid matrix; g is the acceleration due to gravity; ρ

f 0

and T

0

are the reference

density and reference temperature used in the analysis; λ

fx

and λ

sx

are the thermal

conductivities of the pore-ﬂuid and solid matrix in the x direction; λ

fy

and λ

sy

are the

thermal conductivities of the pore-ﬂuid and solid matrix in the y direction; c

pf

and

c

ps

are the speciﬁc heat of the pore-ﬂuid and solid matrix respectively; D

fx

and D

fx

are the diffusivities of the chemical species in the x and y directions respectively; φ is

the porosity of the porous medium; β

T

is the thermal volume expansion coefﬁcient

of the pore-ﬂuid; R

i

is the source/sink term for the reactive transport equation of

chemical species i.

It is noted that if the aqueous mineral concentrations associated with ore body

formation and mineralization are relatively small, their contributions to the density

of the pore-ﬂuid are negligible so that the mass transport process can be decoupled

from the pore-ﬂuid ﬂow and heat transfer processes. This means that the whole cou-

pled problem between ﬂuids mixing, heat transfer and redox chemical reactions in

ﬂuid-saturated porous rocks can be divided into two new problems. The ﬁrst is a

coupled problem between the pore-ﬂuid ﬂow and the heat transfer process, while

the second is a coupled problem between the mass transport process and the redox

chemical reaction process. Since the ﬁrst coupled problem, which is described by

Eqs. (6.1), (6.2), (6.3) and (6.4), can be solved using the existing ﬁnite element

method (Lewis and Schreﬂer 1998, Zienkiewicz 1977), the main purpose of this

study is to develop a new decoupling procedure to effectively and efﬁciently solve

the second coupled problem, which is described by Eq. (6.5) and the related chemi-

cal reaction equations.

If the reaction term in Eq. (6.5) can be determined and is linearly dependent

on the chemical species concentration, then the coupled problem deﬁned between

ﬂuids mixing, heat transfer and redox chemical reactions in ﬂuid-saturated porous

rocks above is solvable using the numerical methods available (Zhao et al. 1998a,

2003a). This requires that the chemical reaction be of the ﬁrst order. Since many

chemical reactions of different orders are associated with ore body formation and

mineralization in ﬂuid-saturated porous rocks, both the second order and the high

order chemical reactions are very common in nature. Without loss of generality, the

second order redox chemical reaction is considered in order to develop a concept

resulting in a new decoupling procedure for removing the coupling between reac-

tive transport equations of redox chemical reactions. In principle, the new concept

6.1 Problems between Fluids Mixing, Heat Transfer and Redox Chemical Reactions 125

and decoupling procedure can be extended to deal with the high order redox chem-

ical reactions. For this reason, a redox chemical reaction of the second order is

considered as follows:

A + B

k

f

⇔

k

b

AB, (6.9)

where A and B are two chemical reactants; AB is the chemical product due to this

redox chemical reaction; k

f

and k

b

are the forward and backward reaction rates of

this redox chemical reaction. It needs to be pointed out that Eq. (6.9) represents a

class of redox chemical reactions such as H

+

+OH

−

k

f

⇔

k

b

H

2

O, Na

+

+Cl

−

k

f

⇔

k

b

NaCl,

Ca

2+

+CO

2−

3

k

f

⇔

k

b

CaCO

3

and so forth in geochemical systems. It is clear that in the

ﬁrst chemical reaction example, chemical reactants A and B are H

+

and OH

−

, while

chemical product AB is H

2

O. In the second chemical reaction example, chemical

reactants A and B are Na

+

and Cl

−

, while chemical product AB is NaCl. Similarly,

in the third chemical reaction example, chemical reactants A and B are Ca

2+

and

CO

2−

3

, while chemical product AB is CaCO

3

.

From the chemical reaction point of view, the general chemical reaction

source/sink terms due to the redox chemical reaction expressed by Eq. (6.9) can

be written as follows:

R

A

= −k

f

r

n

f

−1

C

A

C

B

+k

b

r

n

b

−1

C

AB

, (6.10)

R

B

= −k

f

r

n

f

−1

C

A

C

B

+k

b

r

n

b

−1

C

AB

, (6.11)

R

AB

= k

f

r

n

f

−1

C

A

C

B

−k

b

r

n

b

−1

C

AB

, (6.12)

where C

A

, C

B

and C

AB

are the concentrations of chemical species A, B and AB; R

A

,

R

B

and R

AB

are the chemical reaction source/sink terms associated with chemical

species A, B and AB; n

f

and n

b

are the orders of the forward and backward reac-

tions respectively; r is a quantity of unity value to balance the unit of the reaction

source/sink terms due to different orders of chemical reactions so that it has a recip-

rocal unit of the chemical species concentration. For the redox reaction expressed by

Eq. (6.9), the forward reaction is of the second order, while the backward reaction

is of the ﬁrst order. Since a redox system allows chemical reactions to be proceeded

toward both the product and the reactant directions, the orders of the forward reac-

tion (i.e. the chemical reaction proceeds toward the product direction) and backward

reaction (i.e. the chemical reaction proceeds toward the reactant direction) can be

determined from the related chemical kinetics.

It is noted that the accumulation or diffusion of chemical species in the rock

matrix may lead to some change in porosity, which in turn affects permeabil-

ity and ﬂuid ﬂow in the rock matrix (Zhao et al. 2001d, Xu et al. 2004). This

126 6 Fluid Mixing, Heat Transfer and Non-Equilibrium Redox Chemical Reactions

inﬂuence can be straightforwardly considered using variable permeability within

the computational model. The permeability change induced by a chemical reaction

can be determined from the porosity variation induced by the chemical reaction. For

example, the Carman-Kozeny law can be used to establish a relationship between

the chemically induced porosity change and the chemically induced permeability

change in the rock matrix.

6.2 A Decoupling Procedure for Removing the Coupling between

Reactive Transport Equations of Redox Chemical Reactions

Substituting Eqs. (6.10), (6.11) and (6.12) into Eq. (6.5) yields the following equa-

tions:

φ

∂C

A

∂t

+

u

∂C

A

∂x

+ν

∂C

A

∂y

=

D

ex

∂

2

C

A

∂x

2

+ D

ey

∂

2

C

A

∂y

2

+φR

A

, (6.13)

φ

∂C

B

∂t

+

u

∂C

B

∂x

+ν

∂C

B

∂y

=

D

ex

∂

2

C

B

∂x

2

+ D

ey

∂

2

C

B

∂y

2

+φR

B

, (6.14)

φ

∂C

AB

∂t

+

u

∂C

AB

∂x

+ν

∂C

AB

∂y

=

D

ex

∂

2

C

AB

∂x

2

+ D

ey

∂

2

C

AB

∂y

2

+φR

AB

. (6.15)

Since the total number of linearly-independent reaction rates is identical to the

total number of chemical reactions involved, there is only one linearly-independent

reaction rate for this redox chemical reaction. From this point of view, the total num-

ber of reactive transport equations with source/sink terms due to chemical reactions

can be reduced into one, for this particular redox chemical system. This is the basic

idea behind the proposed decoupling procedure for removing the coupling between

reactive transport equations of redox chemical reactions.

Through some algebraic manipulations, Eqs. (6.13), (6.14) and (6.15) can be

rewritten as follows:

φ

∂(C

A

+C

AB

)

∂t

+

¸

u

∂(C

A

+C

AB

)

∂x

+ν

∂(C

A

+C

AB

)

∂y

¸

−

¸

D

ex

∂

2

(C

A

+C

AB

)

∂x

2

+ D

ey

∂

2

(C

A

+C

AB

)

∂y

2

¸

= 0,

(6.16)

φ

∂(C

B

+C

AB

)

∂t

+

¸

u

∂(C

B

+C

AB

)

∂x

+ν

∂(C

B

+C

AB

)

∂y

¸

−

¸

D

ex

∂

2

(C

B

+C

AB

)

∂x

2

+ D

ey

∂

2

(C

B

+C

AB

)

∂y

2

¸

= 0,

(6.17)

6.2 A Decoupling Procedure for Removing the Coupling 127

φ

∂C

AB

∂t

+

¸

u

∂C

AB

∂x

+ν

∂C

AB

∂y

¸

−

¸

D

ex

∂

2

C

AB

∂x

2

+ D

ey

∂

2

C

AB

∂y

2

¸

= φk

b

k

f

k

b

r

n

f

−1

C

A

C

B

−r

n

b

−1

C

AB

.

(6.18)

Equations (6.16) and (6.17) are two conventional mass transport equations with-

out any source/sink terms due to the redox chemical reaction so that they can be

solved by the well-developed numerical methods available. Since the two new vari-

ables, namely C

I

= C

A

+ C

AB

and C

II

= C

A

+ C

AB

, are independent of chemical

reaction rates, they can be referred to as chemical reaction rate invariants, which are

the analogues of the stress and strain invariants in the ﬁeld of solid mechanics.

If the redox chemical reaction is an equilibrium one, then both the forward and

the backward reaction rates are theoretically inﬁnite so that the chemical reaction

becomes predominant in the reactive transport process. In this case, Eq. (6.18) can

be written as

K

e

r

n

f

−1

C

A

C

B

−r

n

b

−1

C

AB

= 0, (6.19)

where K

e

= k

f

/k

b

is the chemical equilibrium constant.

Inserting the two chemical reaction rate invariants, C

I

= C

A

+ C

AB

and C

II

=

C

B

+C

AB

, into Eq. (6.19) yields the following equation:

K

e

r(C

I

−C

AB

)(C

II

−C

AB

) −C

AB

= 0. (6.20)

It is noted that n

f

= 2 and n

b

= 1 are substituted into Eq. (6.19) so as to

obtain Eq. (6.20). Clearly, Eq. (6.20) has the following mathematical solution for

the chemical product of the redox chemical reaction:

C

AB

=

K

e

r(C

I

+C

II

) +1

2K

e

r

−

[K

e

r(C

I

+C

II

) +1]

2

−4K

2

e

r

2

C

I

C

II

4K

2

e

r

2

. (6.21)

This indicates that for an equilibrium chemical reaction, we only need to solve

the mass transport equations of chemical reaction rate invariants (i.e. C

I

= C

A

+C

AB

and C

II

= C

B

+ C

AB

in this particular example) using the conventional numerical

methods. Once the distributions of the chemical reaction rate invariants are obtained

in a computational domain, the chemical product distribution due to the chemical

reaction can be calculated analytically. As a result, the distributions of the chemical

reactants can be calculated using the distributions of both the chemical product and

the chemical reaction rate invariants.

However, for non-equilibrium chemical reactions, the chemical reaction rates

are ﬁnite so that we need to solve at least one reactive transport equation with the

source/sink term for each of the chemical reactions in the geochemical system. This

means that for the general form of the redox chemical reaction considered in this

study, we need to solve Eq. (6.18) numerically if the reaction rates of this redox

128 6 Fluid Mixing, Heat Transfer and Non-Equilibrium Redox Chemical Reactions

chemical reaction are ﬁnite. For this reason, Equation (6.18) can be rewritten into

the following form:

φ

∂C

AB

∂t

+

¸

u

∂C

AB

∂x

+ν

∂C

AB

∂y

¸

−

¸

D

ex

∂

2

C

AB

∂x

2

+ D

ey

∂

2

C

AB

∂y

2

¸

−φk

b

¸

K

e

rC

2

AB

−[K

e

r(C

I

+C

II

) +1] C

AB

+ K

e

rC

I

C

II

¸

= 0.

(6.22)

Since Eq. (6.22) is strongly nonlinear with the nonlinear term, φk

b

K

e

rC

2

AB

, the

Newton-Raphson algorithm is suitable for solving this equation.

Using the proposed decoupling procedure, the coupled problem between ﬂuids

mixing, heat transfer and redox chemical reactions in ﬂuid-saturated porous rocks

can be solved in the following ﬁve main steps: (1) For a given time step, the coupled

problem described by Eqs. (6.1), (6.2), (6.3), (6.4) and (6.5) with the related bound-

ary and initial conditions are solved using the conventional ﬁnite element method;

(2) After the pore-ﬂuid velocities are obtained from the ﬁrst step, mass transport

equations of the chemical reaction rate invariants (i.e. Eqs. (6.16) and (6.17)) with

the related boundary and initial conditions are then solved using the existing ﬁnite

element method; (3) The chemical reaction source/sink terms involved in Eq. (6.22)

is determined from the related redox chemical reaction so that Eq. (6.22) can be

solved using the Newton-Raphson algorithm; (4) According to the deﬁnitions of the

chemical reaction rate invariants, C

I

= C

A

+C

AB

and C

II

= C

B

+C

AB

, the chemical

reactant concentrations (i.e. C

A

and C

B

) can be determined from simple algebraic

operations; (5) Steps (1–4) are repeated until the desired time step is reached. These

solution steps have been programmed into our research code.

6.3 Veriﬁcation of the Decoupling Procedure

The main and ultimate purpose of a numerical simulation is to provide numerical

solutions for practical problems in a real world. Since numerical methods are the

basic foundation of a numerical simulation, only can an approximate solution be

obtained from a computational model, which is the discretized description of a con-

tinuum mathematical model. Due to inevitable round-off errors in computation and

discretized errors in temporal and spatial variables, it is necessary to verify, at least

from the qualitative point of view, the proposed numerical procedure so that mean-

ingful numerical results can be obtained from a discretized computational model.

For this reason, a testing coupled problem, for some aspects of which the analytical

solutions are available, is considered in this section.

Figure 6.1 shows the geometry of the coupled problem between pore-ﬂuids mix-

ing, heat transfer and redox chemical reactions around a vertical geological fault

within the crust of the Earth. For this problem, the pore-ﬂuid pressure is assumed

to be lithostatic, implying that there is an upward throughﬂow at the bottom of the

computational model. The height and width of the computational model are 10 km

6.3 Veriﬁcation of the Decoupling Procedure 129

/ 1kmol C

B

=

/ 1 m

3

m

3

kmol C

A

=

T = 25

o

C, p = 0, C

A

= 0, C

B

= 0

10km

20km

T = 325

o

C, p = α

1

p

Hydrostatic

Fig. 6.1 Geometry and

boundary conditions of the

ﬂuid focusing and mixing

problem

and 20 km respectively. The length of the fault is 5 km, with an aspect ratio of 20.

The porosities of the fault and its surrounding rock are 0.35 and 0.1. The surround-

ing rock is assumed to have a permeability of 10

–16

m

2

in both the x and y directions,

while the permeability of the fault is calculated using the Carman-Kozeny formula,

which gives rise to a permeability of about 43 times that of the surrounding rock.

The top temperature is 25

◦

C and the geothermal gradient is 30

◦

C/km, meaning that

the temperature at the bottom is ﬁxed at 325

◦

C. The two chemical reactants with a

concentration of 1 kmol/m

3

are injected at the left half and right half bottom bound-

ary respectively, while the concentrations of both the two reactants and the prod-

uct are assumed to be zero at the top boundary of the computational model. The

dispersion/diffusivity of the chemical species is 3 ×10

–10

m

2

/s. For the pore-ﬂuid,

dynamic viscosity is 10

−3

N ×s/m

2

; reference density is 1000 kg/m

3

; volumetric

thermal expansion coefﬁcient is 2.1 ×10

–4

(1/

◦

C); speciﬁc heat is 4184 J/(kg ×

o

C);

thermal conductivity coefﬁcient is 0.59 W/(m×

o

C) in both the x and y directions.

For the porous matrix, the thermal conductivity coefﬁcient is 2.9 W/(m×

o

C) in

both the x and y directions; the speciﬁc heat is 878 J/(kg ×

o

C); reference rock den-

sity is 2600 kg/m

3

. In order to simulate the ﬂuids focusing and chemical reactions

within the fault in an appropriate manner, a mesh of small element sizes is used

to simulate the fault zone, while a mesh gradation scheme is used to simulate the

surrounding rock by gradually changing the mesh size from the outline of the fault

within the computational model. As a result, the whole computational domain is

simulated by 306,417 three-node triangle elements.

Figure 6.2 shows the streamline distribution of the system with the vertical fault

in the computational model. Since the pore-ﬂuids carrying two different chemical

reactants are uniformly and vertically injected into the computational model at the

left and right parts of the bottom, the pore-ﬂuid ﬂow converges into the vertical fault

at the inlet (i.e. the lower end) of the fault, but diverges out of the vertical fault at

the outlet (i.e. the upper end) of the fault. This phenomenon can be clearly observed

from Fig. 6.3, where the velocity distributions are displayed at both the inlet and

130 6 Fluid Mixing, Heat Transfer and Non-Equilibrium Redox Chemical Reactions

(Whole system view)

(Zoom-in view at the inlet of the fault)

(Zoom-in view at the outlet of the fault)

Fig. 6.2 Streamline distributions due to the ﬂuid focusing within a vertical fault in the crust

outlet of the vertical fault. For an elongate elliptic fault of large aspect ratios in an

inﬁnite medium, the existing analytical solution indicates that the streamlines of the

pore-ﬂuid ﬂow are parallel each other inside of the fault (Zhao et al. 1999d). As

expected, the computed streamlines concentrate vertically within the vertical fault,

indicating that the numerical result obtained fromthe computational model has good

agreement with the existing analytical result. This demonstrates that the proposed

numerical procedure can produce useful numerical solutions for ﬂuid focusing and

mixing, at least from the qualitative point of view. In order quantitatively to validate

the numerical solutions, the analytical solution for the ﬂow-focusing factor of an

elongate elliptic fault of large aspect ratios can be employed. Since the elongate fault

within the computational model is basically of a rectangular shape, the analytical

solution for its ﬂow-focusing factor can be evaluated using the following modiﬁed

formula:

6.3 Veriﬁcation of the Decoupling Procedure 131

(Zoom-in view at the inlet of the fault)

(Zoom-in view at the outlet of the fault)

Fig. 6.3 Velocity distributions due to the ﬂuids focusing within a vertical fault in the crust

λ =

4

π

α(β +1)

β +α

, (6.23)

where λ is the pore-ﬂuid ﬂow focusing factor of the rectangular fault of a large

aspect ratio; β is the aspect ratio of the rectangular fault; α is the permeability ratio

of the fault to its surrounding rock.

Since the vertical velocity of the injected ﬂuids due to the lithostatic pore-ﬂuid

pressure is equal to 1.6 ×10

–9

m/s and the numerical solution for the maximum

132 6 Fluid Mixing, Heat Transfer and Non-Equilibrium Redox Chemical Reactions

vertical velocity within the fault is equal to 3.02 ×10

–8

m/s, the corresponding

ﬂow-focusing factor of the fault is equal to the ratio of the maximum velocity

within the fault to that of the injected ﬂuids at the bottom of the computational

model. This results in a ﬂow-focusing factor of 18.88 for the rectangular fault

within the computational model. The analytical value of the ﬂow-focusing fac-

tor for the rectangular fault can be calculated from Eq. (6.23). Substituting the

related parameters into Eq. (6.23) yields the analytical ﬂow-focusing factor of 18.25.

Since the relative error of the ﬂow-focusing factor from the numerical simulation

is within 3.5%, it quantitatively demonstrates that the proposed numerical proce-

dure used in the computational model can produce accurate numerical solutions

for ﬂuid focusing and mixing within the fault. These processes are very impor-

tant for accurately simulating the chemical species transport and reaction within the

fault.

(Reactant A, t = 1000 years)

(Reactant A, t = 5000 years)

(Reactant A, t = 8000 years)

(Reactant B, t = 8000 years)

(Reactant B, t = 1000 years)

(Reactant B, t = 5000 years)

Fig. 6.4 Concentration distributions of the chemical reactants at different time instants (Equilib-

rium reaction)

6.3 Veriﬁcation of the Decoupling Procedure 133

In order to verify the proposed decoupling procedure for solving reactive trans-

port equations with strong nonlinear reaction source/sink terms, the redox chemi-

cal reaction due to an equilibrium reaction is considered and solved using the pro-

posed numerical procedure. The equilibrium constant is assumed to be 10 and the

time step used in the simulation is 100 years. Figure 6.4 shows the concentration

distributions of the two chemical reactants at three different time instants, while

Fig. 6.5 shows the comparison of the numerical solutions (which are obtained from

the proposed decoupling procedure) with the analytical solutions (which are derived

mathematically and expressed by Eq. (6.21)) for the chemical product. It can be

observed that with the increase of time, both chemical reactants are transported

into the computational domain from the left half and right half of the bottom. Due

to the ﬂuid ﬂow focusing, both chemical reactants are transported much faster in

the fault zone than in the surrounding rock. As expected, these chemical reactants

(Analytical, t = 1000 years) (Numerical, t = 1000 years)

(Analytical, t = 5000 years) (Numerical, t = 5000 years)

(Analytical, t = 8000 years) (Numerical, t = 8000 years)

Fig. 6.5 Comparison of numerical solutions with analytical ones for the chemical product (Equi-

librium reaction)

134 6 Fluid Mixing, Heat Transfer and Non-Equilibrium Redox Chemical Reactions

are divergent around the exit region of the fault. The comparison of the numer-

ical solutions with the analytical ones for the chemical product clearly demon-

strates that the proposed decoupling procedure can produce accurate numerical

solutions for simulating the equilibrium chemical reaction, in which the chemi-

cal reaction rate approaches inﬁnite. It is interesting to note that there is a strong

interaction between solute advection, diffusion and chemical reaction rate in the

considered equilibrium chemical system. Although two reactants are well trans-

ported into the fault zone, the mixing of the two ﬂuids carrying them is controlled

by the solute diffusion. Since the chemical reaction rate is inﬁnite for the equilib-

rium reaction, the corresponding chemical equilibrium length is identical to zero

due to the solute diffusion. This implies that the chemical reaction rate is too fast

to allow both the reactants to diffuse across the common boundary between them

so that ﬂuid mixing cannot effectively take place within the fault zone. This is the

reason why both chemical reactants are abundant but no chemical product is pro-

duced within the fault zone in the computational model. However, around the exit

region of the fault zone, the ﬂow of the ﬂuids is slowed and divergent so that the

ﬂuids carrying two different reactants can be mixed. Consequently, a high con-

centration of the chemical product is produced around the exit region of the fault

zone.

6.4 Applications of the Proposed Decoupling Procedure

to Predict Mineral Precipitation Patterns in a Focusing

and Mixing System Involving Two Reactive Fluids

Mixing of two or more ﬂuids is commonly suggested as a mechanism for precipi-

tating minerals from solution in porous rocks. Examples include uranium deposits

(Wilde and Wall 1987), MVT deposits (Appold and Garven 2000, Garven et al.

1999), Irish Pb-Zn deposits (Hitzmann, 1995, Everett et al. 1999, Murphy et al.

2008), vein gold deposits (Matthai et al. 1995, Cox et al. 1995) and Carlin gold

deposits (Cline and Hofstra 2000). The mixing process is attractive because it

enables two ﬂuids of contrasting Eh-pH conditions to mix and hence generate chem-

ical conditions conducive to mineral precipitation. Clearly rock-ﬂuid(s) interactions

must be involved as well as the ﬂuid mixing process but in depth exploration of such

multi-ﬂuid-rock reaction processes awaits robust and computationally fast ways of

handling realistic kinetic-reaction-transport phenomena.

There are three end member geometries that promote the mixing of two miscible

ﬂuids (Fig. 6.1). The ﬁrst we refer to as parallel-ﬂow geometries of the ﬁrst kind

(Figs. 6.6a, b) where two contrasting ﬂuids are brought alongside each other by

convection or focusing in a highly permeable fault or sedimentary lens (Phillips

1991). The second involves the production of a new ﬂuid within an existing ﬂuid

ﬂow system by thermal or chemical reactions with a mineral assemblage within the

ﬂow stream (Fig. 6.6c). An example is the production of CO

2

or of hydrocarbons

from carbonates or carbonaceous material within a ﬂuid stream that is hot, acid and

6.4 Applications of the Proposed Decoupling Procedure 135

oxidised (Matthai et al. 1995, Cox et al. 1995). This is, in fact, a special example of

parallel-ﬂow geometry except that now the new ﬂuid source is embedded within the

primary ﬂow stream. We refer to this process as parallel ﬂow geometry of the second

kind. The third end member involves injection of one ﬂuid perpendicular to the ﬂow

ﬁeld of a second ﬂuid as shown in Figs. 6.6d, e. This mixing geometry produces

a mixing plume that expands outwards mainly in the downstream direction of the

Fluid A Fluid B

Fluid B Fluid A

High Permeability

Lens

(a)

(c) (d)

(e)

(b)

Fluid A Fluid A

Fluid B

Fluid A

Fluid B

Fluid A

Fluid B

Fig. 6.6 Mixing geometries for two ﬂuids: (a) and (b) are parallel mixing ﬂows of the type

discussed in this paper; (c) is a parallel mixing ﬂow but the second ﬂuid is generated within the

ﬂow ﬁeld of the ﬁrst ﬂuid; (d) and (e) are perpendicular mixing ﬂows where one ﬂuid is injected

normal to the ﬂow ﬁeld of the other ﬂuid

136 6 Fluid Mixing, Heat Transfer and Non-Equilibrium Redox Chemical Reactions

second ﬂow. This situation has been considered by Phillips (1991) and analytical

solutions for the composition of the resulting mixing plume presented.

6.4.1 Key Factors Controlling Mineral Precipitation Patterns

in a Focusing and Mixing System Involving

Two Reactive Fluids

We consider the ﬁrst of these parallel ﬂow geometries in some detail. The classical

approach (see Phillips 1991) is to assume that chemical equilibrium has been estab-

lished within the mixing plume so that the equilibrium concentration of a particular

chemical species is expressed as a function of the environmental parameters, namely

temperature, ﬂuid pressure and the concentration of other chemical species. The fun-

damental point we explore here is that the overall pattern of mineralization in these

mixing systems results from intimate interactions between solute advection, solute

diffusion and/or dispersion and chemical kinetics. Thus, chemical equilibrium in

some cases may never be attained in these mixing systems.

In this section we concentrate solely on parallel ﬂows of the ﬁrst kind and show

that the patterns of chemical precipitation for such ﬂows result from intimate rela-

tions between ﬂuid ﬂow (which is a ﬁrst order control on solute advection), chem-

ical reaction rates, and solute diffusion and dispersion. We show, for instance, that

even for a situation where two ﬂuids are brought together by ﬂuid focusing within

a permeable fault zone, chemical reaction between the two ﬂuids may never occur

even if chemical equilibrium prevails so long as the advection of solute is large with

respect to solute diffusion/dispersion. In order to focus on the principles involved for

parallel ﬂows of the ﬁrst kind we idealise the situation by considering a vertical per-

meable fault within a less permeable rock mass. The whole model is fully saturated

with ﬂuids of different chemical compositions at different points on the system. We

consider only systems where the ﬂuids are miscible whilst noting for future inves-

tigation that multi-phase systems, that is, ﬂuid systems in which two or more ﬂuids

are present that are immiscible, comprise yet another hydraulic-chemical system

where intimate mixing of contrasting chemistries is possible at the pore scale. We

need to emphasise here that although we are considering only vertical permeable

faults the discussion is just as relevant to permeable sedimentary lenses or to any

other geometry that brings two chemically contrasted ﬂuids together. The vertical

geometry here allows simple boundary conditions to be imposed for the hydraulic

potential driving ﬂuid ﬂow. The discussion is equally relevant for any other orienta-

tion of the focusing “lens”.

Ore formation in the Earth’s crust is commonly considered to be a consequence

of the reaction of large ﬂuxes of one or more ﬂuids with rock. In many situations,

ﬂuid ﬂow is focused within fault zones and the important issue becomes the spatial

control exerted by the fault on the resultant mineralization. It is commonly assumed

that either ﬂuid mixing or ﬂuid/rock reactions or a combination of both are impor-

tant processes associated with permeable fault zones, due to the large ﬂuid ﬂuxes

typically inferred for such zones. However, recent extensive studies (Phillips 1991,

6.4 Applications of the Proposed Decoupling Procedure 137

Zhao et al. 1998a, 2007a) have demonstrated that ore formation is not only depen-

dent on ﬂuid ﬂow, but also upon solute diffusion/dispersion and chemical kinetics.

Mineralisation is commonly associated with the spatial distribution of chemical

reactions in permeable rocks (Steefel and Lasaga 1990, Phillips 1991, Raffensperger

and Garven 1995, Zhao et al. 1998a). These chemical reaction patterns are strongly

inﬂuenced by ﬂuid ﬂow, heat transfer and the transport of reactive chemical species.

Since ﬂuid ﬂow is an important agent for transporting aqueous chemical species

from one location into another (Garven and Freeze 1984, Peacock 1989, Ord and

Oliver 1997, Connolly 1997, Zhao et al. 1999d, Oliver 2001, Braun et al. 2003),

ﬂuid ﬂow patterns can also inﬂuence the positions where different chemical species

may meet and mix and hence where chemical reactions may take place (Yardley and

Bottrell 1992, Yardley and Lloyd 1995). In addition, heat transfer processes inﬂu-

ence the thermal structure of the Earth’s crust and hence the spatial distribution of

chemical reaction patterns through the temperature-dependence of the equilibrium

constant and of the reaction rate (Steefel and Lasaga 1994, Xu et al. 1999, Zhao

et al. 2000b).

For aqueous species, whether chemical equilibrium is or is not attained controls

the possibility of precipitation of mineral assemblages in permeable rocks. If an

aqueous species, produced by a chemical reaction, reaches equilibrium, then that

species is saturated within the ﬂuid and hence its concentration reaches a maxi-

mum value for a given mineralizing environment. In most cases, the solubility of

an aqueous species is directly proportional to the ambient temperature. It follows

that in cooling hydrothermal systems the saturated aqueous species becomes over-

saturated and is therefore precipitated in the porous rock. Another process caus-

ing mineral precipitation is transport within the pore-ﬂuid of the saturated aqueous

species from a high temperature region to a low temperature region. The aqueous

species can also become oversaturated and is precipitated in the porous rock. For a

given ore formation environment, ﬂow rates can control the total process of solute

advection. In such a case, a redox controlled chemical reaction can only reach equi-

librium through optimal coupling of solute advection, solute diffusion/dispersion

and chemical kinetics. With parallel ﬂows taken as an example, if the solute dif-

fusion/dispersion in the direction perpendicular to the interface between the ﬂows

is much slower than the chemical reaction rate, the chemical product cannot attain

equilibrium, even though the two ﬂuids are perfectly miscible. On the other hand,

if the ﬂow rate is much faster than the chemical reaction rate, the chemical product

can only attain equilibrium at a distance large with respect to a chemical equilib-

rium length-scale, which we introduce below, in the ﬂow direction. This implies

that for a permeable fault zone with a given chemical reaction, there exists an opti-

mal ﬂow rate resulting in chemical equilibrium being attained between two ﬂuids

that mix and focus within the fault zone. However, for parallel ﬂows, such as those

resulting from vertical super-hydrostatic pressure gradients, chemical equilibrium

may or may not be attained when two ﬂuids of different origins are brought together

through focusing within a permeable vertical fault. In this case, strong interactions

occur between solute advection, diffusion/dispersion and chemical kinetics. It is

this interaction that controls the equilibrium distribution of the resulting chemical

138 6 Fluid Mixing, Heat Transfer and Non-Equilibrium Redox Chemical Reactions

product. A conceptual model is presented here to investigate this interaction and the

corresponding inﬂuence on chemical reaction patterns.

6.4.2 Theoretical Analysis of Mineral Precipitation Patterns

in a Focusing and Mixing System Involving

Two Reactive Fluids

Although it is difﬁcult, if not impossible, to obtain analytical solutions for the cou-

pled problem expressed by Eqs. (6.1), (6.2), (6.3), (6.4) and (6.5) in general cases, it

is possible to gain theoretical understanding through analytical solutions for the cou-

pled problem in some limiting cases (Zhao et al. 2007a). This understanding can be

achieved by converting the reactive transport equation (i.e. Eq. (6.5)) into a dimen-

sionless one so that the controlling processes associated with the reactive chemical

species transport can be identiﬁed. This means that it is possible to investigate the

relationship between the controlling processes associated with reactive chemical

species transport so that the overall structure of their solutions can be understood.

Since there are three major controlling processes, namely solute advection, solute

diffusion/dispersion and chemical kinetics, that may play dominant roles in deter-

mining chemical reaction patterns, we need to consider the relationships between

the time scales for these three controlling processes. For this purpose, we ﬁrst con-

sider a dimensionless parameter known as the Damk¨ ohler number, Da, (Steefel and

Lasaga 1990, Ormond and Ortoleva 2000) to express the relative time scales of

solute advection and reaction kinetics:

Da =

φk

R

l

V

= Time Scale for Solute Advection/

Time Scale for Chemical Reaction,

(6.24)

where V is the characteristic Darcy velocity of the system; l is the characteris-

tic length of the controlling process in the system; k

R

is the controlling chemical

reaction rate with units of [s

-1

]; φ is the porosity of the porous medium. When the

time scale for solute advection is equal to the time scale for chemical kinetics, the

Damk¨ ohler number is equal to one. In this case, the chemical equilibrium length

scale of the system can be expressed as follows:

l

chemical

advection

=

V

φk

R

, (6.25)

where l

chemi cal

advection

is the chemical equilibrium length due to solute advection for a given

chemical reaction. Below we refer to l

chemi cal

advection

as the advection chemical equilibrium

length. It is clear that if a chemical reaction rate is given, there exists an optimal

ﬂow rate such that the chemical reaction can reach equilibrium beyond the advection

chemical equilibrium length determined from Eq. (6.25). Thus, for a given l

chemi cal

advection

,

the corresponding optimal ﬂow rate, V

optimal

, for which the chemical reaction can

reach equilibrium beyond the given l

chemi cal

advection

, is as follows:

6.4 Applications of the Proposed Decoupling Procedure 139

V

optimal

= φk

R

l

chemical

advection

. (6.26)

It should be noted that both l

chemi cal

advection

and the optimal ﬂow-rate have clear physi-

cal meanings: The physical meaning of l

chemi cal

advection

is that for a given ﬂuid ﬂow rate, a

chemical reaction with a given reaction rate can reach equilibrium once this distance

is traversed by the ﬂuid in the ﬂow direction, within the time scale of chemical equi-

librium. Since l

chemi cal

advection

is directly proportional to the ﬂuid ﬂow rate, the greater the

ﬂuid ﬂow rate, the larger l

chemi cal

advection

. This means that fast ﬂows require relatively long

distances in the ﬂow direction, beyond which a chemical reaction with a given reac-

tion rate can reach equilibrium. In contrast, the physical meaning of the optimal ﬂow

rate is that for a given l

chemi cal

advection

, a chemical reaction with a given reaction rate can

reach equilibrium if the ﬂuid ﬂow rate is within the time scale of chemical equilib-

rium. Since the optimal ﬂow rate is also directly proportional to l

chemi cal

advection

, the larger

l

chemi cal

advection

in the ﬂow direction, the greater the optimal ﬂuid ﬂow rate. This implies

that a large l

chemi cal

advection

requires a relatively fast optimal ﬂow rate, so that chemical

reaction for a given reaction rate can reach equilibrium beyond this large l

chemi cal

advection

in

the ﬂow direction.

If the ﬂow paths of two ﬂuids are parallel to each other in a ﬂuid mixing system,

solute diffusion/dispersion normal to the ﬂow direction plays a fundamental role

in promoting chemical reactions between different reactive chemical species. In this

case, a second dimensionless parameter, Z, needs to be deﬁned to express the relative

time scale between the solute diffusion/dispersion process and the chemical reaction

process. Notice that Z is independent of the ﬂuid velocity and so still has meaning

for zero ﬂuid ﬂow.

Z =

k

R

l

2

D

= (Time Scale for Solute Dispersion/Diffusion)/

Time Scale for Chemical Reaction,

(6.27)

where D is the solute diffusion/dispersion coefﬁcient; l is the characteristic length

of the controlling process in the system; k

R

is the reaction rate. Since this dimen-

sionless number expresses the ratio of the solute diffusion/dispersion time scale to

the chemical kinetic time scale, it is unity when the two time scales are equal. In this

situation, the chemical equilibrium length, l

chemical

diffusion

of the system can be expressed

as follows:

l

chemical

diffusion

=

D

k

R

, (6.28)

where l

chemical

diffusion

is the chemical equilibrium length due to solute diffusion/dispersion

for a given chemical reaction. For a given solute diffusion/dispersion coefﬁcient,

there exists an optimal reaction rate such that the chemical reaction can reach equi-

librium within l

chemical

diffusion

determined from Eq. (6.28). Thus, for a given l

chemical

diffusion

, the

corresponding optimal chemical reaction rate, k

optimal

R

, for which the chemical reac-

tion can reach equilibrium within the given l

chemical

diffusion

, is as follows:

140 6 Fluid Mixing, Heat Transfer and Non-Equilibrium Redox Chemical Reactions

k

optimal

R

=

D

l

chemical

diffusion

2

. (6.29)

Both l

chemical

diffusion

and the optimal chemical reaction-rate also have clear physical

meanings: The physical meaning of l

chemical

diffusion

is that for parallel ﬂow with a given

solute diffusion/dispersion coefﬁcient, a chemical reaction with a given reaction rate

can only reach equilibrium within the distance diffused by the solute in the direction

perpendicular to the ﬂuid ﬂow within the time scale of chemical equilibrium. Since

l

chemical

diffusion

is directly proportional to the square root of the solute diffusion/dispersion

coefﬁcient, the greater the solute diffusion/dispersion coefﬁcient, the larger l

chemical

diffusion

in the direction perpendicular to the ﬂuid ﬂow. This implies that a large solute

diffusion/dispersion coefﬁcient can result in a chemical reaction, with a given

reaction rate, reaching equilibrium over a large distance normal to the direction of

ﬂuid ﬂow. By contrast, the physical meaning of the optimal chemical reaction rate

is that for a given l

chemical

diffusion

, a chemical reaction with a given reaction rate can reach

equilibrium if the chemical reaction proceeds at this chemical reaction rate within

the time scale of chemical equilibrium. Since the optimal chemical reaction rate is

inversely proportional to the square of l

chemical

diffusion

, the larger l

chemical

diffusion

in the direction

perpendicular to the ﬂuid ﬂow, the smaller the optimal chemical reaction rate. This

implies that a large chemical equilibrium length due to solute diffusion/dispersion

in the direction perpendicular to the parallel ﬂuid ﬂow requires a relatively slow

optimal chemical reaction rate, so that for a given solute diffusion/dispersion

coefﬁcient, the chemical reaction can reach equilibrium within this long chemical

equilibrium length in the direction perpendicular to the parallel ﬂuid ﬂow.

The combined use of these two numbers, Da and Z, can express the interaction

between solute advection, diffusion/dispersion and chemical kinetics. Note that for

a given fault zone involving pore-ﬂuid ﬂow focusing and mixing, it is possible to

deﬁne three different types of mineral precipitation patterns in Z-Da number space.

For this purpose, the thickness of the fault zone is chosen as the characteristic length

of the Z Number, while the length of the fault zone is chosen as the characteristic

length of the Da Number. The relationships between Da, Z and mineral precipitation

types are shown in Fig. 6.7.

6.4.3 Chemical Reaction Patterns due to Mixing and Focusing

of Two Reactive Fluids in Permeable Fault Zones

The theoretical understanding of the interaction between solute advection, solute

diffusion/dispersion and chemical kinetics presented in the previous section is, in

principle, useful for investigating chemical reaction patterns resulting from chem-

ical equilibrium associated with ﬂuid ﬂow in all kinds of porous rocks. Chemical

reaction patterns arising from ﬂow focusing and mixing of two reactive ﬂuids within

permeable vertical fault zones are the subject of this section.

6.4 Applications of the Proposed Decoupling Procedure 141

Z = Time scale for Solute Diffusion/ Time

Scale for Chemical Reaction

Da = Time Scale

for Solute

Advection/ Time

Scale for

Chemical

Reaction

1.0

1.0

Chemical

Reaction

Dominates

Solute

Diffusion

Dominates

Chemical

Reaction

Dominates

Solute

Advection

Dominates

(A) Relationship between solution advection, diffusion and chemical kinetics

Z = Time scale for Solute Diffusion/ Time

Scale for Chemical Reaction

Da = Time Scale

for Solute

Advection/ Time

Scale for

Chemical

Reaction

1.0

1.0

TYPE 2 CHEMICAL

REACTION PATTERN

TYPE 1 CHEMICAL

REACTION PATTERN

TYPE 3

CHEMICAL

REACTION

PATTERN

(B) Three types of mineral precipitation patterns

Fig. 6.7 Three fundamental types of mineral precipitation patterns in the Z-Da Number space

The geometry and problem description of two ﬂuids mixing and focusing within

a permeable vertical fault zone are shown in Fig. 6.8. In this ﬁgure, L

fault

and W

fault

are the length and width of the fault zone; L

mp

represents the starting position of

mineral precipitation within the fault zone; W

mp

represents the thickness of mineral

142 6 Fluid Mixing, Heat Transfer and Non-Equilibrium Redox Chemical Reactions

0.5W

mp

L

mp

Fluid A

y

0

Fluid B

x

L

fault

W

fault

Fig. 6.8 The conceptual model of two ﬂuids mixing and focusing in a fault

precipitation within the fault zone. Two ﬂuids carrying two different solutes, namely

Fluid A and Fluid B, are injected from the left and right sides of the fault axis and are

focused into the fault zone according to the principles discussed by Phillips (1991)

and Zhao et al. (1999d). Since mineral precipitation is dependent on both chemical

equilibrium and the saturation concentration of a mineral, it is assumed that the

starting position of mineral precipitation is coincident with chemical equilibrium

being attained in the following theoretical analysis. This means that for a given

chemical reaction rate, the starting position of mineral precipitation is controlled by

solute advection and is measured from the lower tip of the fault. On the other hand,

since Fluid A ﬂows parallel to Fluid B, the mixing of these two ﬂuids is due solely

to solute diffusion/dispersion in the direction normal to the fault zone. Hence, the

thickness of mineral precipitation is controlled solely by solute diffusion/dispersion.

The starting position of mineral precipitation within the fault zone can be expressed

as follows:

L

mp

= l

chemical

advection

=

V

φk

R

. (6.30)

Similarly, the thickness of mineral precipitation within the fault zone can be

expressed as

W

mp

= 2l

chemical

diffusion

= 2

D

k

R

. (6.31)

6.4 Applications of the Proposed Decoupling Procedure 143

Here the factor 2 arises because of the symmetry of geometry; l

chemical

diffusion

is mea-

sured from the centre of the fault.

As indicated earlier in this chapter, there are three possible cases in which chemi-

cal equilibrium can be attained in the conceptual model shown in Fig. 6.8. The three

types of interest are: Type 1: chemical equilibrium is attained just at the lower tip of

the fault; Type 2: chemical equilibrium is attained at the upper tip of the fault; and

Type 3: chemical equilibriumis attained somewhere between the lower tip and upper

tip of the fault. As mentioned above, analytical solutions to Eqs. (6.1), (6.2), (6.3),

(6.4) and (6.5) are difﬁcult but we can gain some insight into the types of chem-

ical reaction patterns that are possible by considering some limiting cases below.

In the forthcoming section, we numerically examine more realistic and applicable

situations.

6.4.3.1 Type 1: Chemical Equilibrium is Attained at the Lower Tip

of the Fault

In this limiting case, the ﬂuid velocity is zero in the whole system. Substituting a

zero velocity into Eq. (6.30) yields a zero value of the starting position of mineral

precipitation within the fault zone. In this situation, it is possible for chemical equi-

librium to be attained throughout the whole length of the fault, depending on the

actual residence time of the two chemical reactants within the system. To enable

chemical equilibrium to be reached within the whole fault length, L

fault

, the resi-

dence time for which the two chemical reactants should exist in the system is equal

to the time interval, t

diffuse

, for the solute to diffuse from the lower tip to the upper

tip of the fault:

t

diffuse

=

L

2

fault

D

. (6.32)

Consider a speciﬁc example: Eq. (6.32) indicates that for a solute diffu-

sion/dispersion coefﬁcient of 10

–10

m

2

/s and a vertical fault 1 km long, the two

chemical reactants need to be present for 10

16

s (i.e. about 3 × 10

8

years) at the

bottom of the conceptual model in order for chemical equilibrium to be reached

within the whole length of the fault. Since the required duration for the existence of

these two chemical reactants is quadratically proportional to the length of a fault, it

is increased to about 25 ×10

16

s (i.e. about 7.5 × 10

9

years) to enable the chemi-

cal reaction to reach equilibrium within the whole length of a 5 km long fault. In

this example, where the ﬂuid ﬂow is negligible, the time required to reach chemical

equilibrium within the whole length of a fault 5 km long is greater than the age of

the Earth.

As indicated in Eq. (6.31), the thickness of mineral precipitation within the

fault zone is dependent on both the chemical reaction rate and the solute diffu-

sion/dispersion coefﬁcient. For a given solute diffusion/dispersion coefﬁcient, the

optimal chemical reaction rate k

optimal

R

, which will enable chemical reactions to attain

equilibrium across the whole width of the fault zone, is as follows:

144 6 Fluid Mixing, Heat Transfer and Non-Equilibrium Redox Chemical Reactions

k

optimal

R

=

4D

W

2

fault

. (6.33)

6.4.3.2 Type 2: Chemical Equilibrium is Attained at the Upper Tip

of the Fault

This is another limiting case, in which the ﬂuid rate (i.e. velocity) has reached a

critical value in the fault zone. The fundamental characteristic of this limiting case

is that if the ﬂuid rate is equal to or greater than this critical value, the starting point

of chemical equilibrium is at or beyond the upper tip of the fault, indicating that the

chemical equilibrium of the chemical product and therefore mineral precipitation

cannot be attained within the fault zone. For a given chemical reaction rate, this

critical ﬂow rate within the fault zone, V

critical

, can be expressed as follows:

V

critical

= φk

R

L

fault

. (6.34)

Equation (6.34) is useful for estimating the critical ﬂow rate for a vertical fault.

For example, if the porosity of a vertical fault is 0.3 and the chemical reaction

rate is 10

–11

(1/s), then the corresponding critical ﬂow rate is 3 ×10

–9

m/s for a

vertical fault 1 km long. If both the fault porosity and the chemical reaction rate

remain unchanged, the corresponding critical ﬂow rate of the fault is increased to

about 15 ×10

–9

m/s for a vertical fault 5 km long. Suppose the ﬂow-focusing factor

(Phillips, 1991) of the vertical fault is of the order of 15, the corresponding critical

ﬂow rate within the surrounding rock for a vertical fault is 10

–9

m/s. This critical

ﬂow rate can be easily exceeded when the ﬂuid within the surrounding rock of the

fault zone is under a lithostatic pressure gradient. If the permeability of the sur-

rounding rock is 10

–16

m

2

and the dynamic viscosity of the ﬂuid is 10

–3

N s/m

2

,

then the ﬂow rate induced by the lithostatic pressure gradient within the surround-

ing rock is 1.7 ×10

–9

m/s. This implies that the ﬂow rate induced by a lithostatic

pressure gradient within the surrounding rock is too high to enable minerals to be

precipitated within the vertical fault zone.

6.4.3.3 Type 3: Chemical Equilibrium is Attained Somewhere

between the Lower Tip and Upper Tip of the Fault

In this case, the equilibrium of the reaction product is attained within the fault zone.

The starting position of the chemical equilibrium is measured from the lower tip of

the fault and can be expressed by the following equation:

L

mp

= α

mp

L

fault

, (6.35)

where L

mp

is the starting position of chemical equilibrium within the fault zone in

the ﬂow direction; α

mp

is a coefﬁcient to express the relative relationship between

the length of the fault and the starting position of chemical equilibrium within the

fault zone; that is α

mp

= L

mp

L

fault

. It is clear that in this particular case, the value

6.4 Applications of the Proposed Decoupling Procedure 145

of α

mp

is greater than zero but less than one. A value of α

mp

equal to zero represents

the ﬁrst limiting case (Type 1), while a value of α

mp

equal to one represents the

second limiting case (Type 2), as discussed above.

Similarly, the width of chemical equilibrium within the fault can be expressed as

follows:

W

mp

= β

mp

W

fault

, (6.36)

where W

mp

is the width of chemical equilibrium within the fault zone; W

fault

is

the width of the fault; β

mp

is a coefﬁcient expressing the ratio between the width

of the fault and the width of chemical equilibrium within the fault zone; that is,

β

mp

= W

mp

W

fault

. In this case, the value of β

mp

is greater than zero but less than

one. A value of β

mp

equal to zero represents the ﬁrst limiting case (Type 2), whilst

a value of β

mp

equal to one represents another limiting case (i.e. the limiting case

of Type 3), in which case the width of chemical equilibrium within the fault zone is

just equal to that of the fault.

If both the solute diffusion/dispersion coefﬁcient and the width of chemical equi-

librium within the fault zone are known, the corresponding optimal chemical reac-

tion rate in this case can be estimated from the following formula:

k

optimal

R

=

4D

W

2

mp

=

4D

β

2

mp

W

2

fault

. (6.37)

Thus, the corresponding optimal ﬂow rate in this particular case is:

V

optimal

= φk

optimal

R

L

mp

=

4φDα

mp

L

fault

β

2

mp

W

2

fault

. (6.38)

Equation (6.38) indicates that for a particular pattern of mineral precipitation

within the fault zone, there exists an optimal ﬂow rate such that chemical equilib-

rium is attained within the region associated with this particular mineral precipita-

tion pattern.

6.4.4 Numerical Illustration of Three Types of Chemical Reaction

Patterns Associated with Permeable Fault Zones

The theoretical analysis carried out in the previous section predicts that there exist

three fundamental types of chemical reaction patterns associated with permeable

vertical fault zones due to two ﬂuids mixing and focusing. In order to illustrate

these different types of chemical reaction patterns, the ﬁnite element method is used

to solve the coupled problem numerically involving ﬂuid mixing, heat transfer and

chemical reactions expressed by Eqs. (6.1), (6.2), (6.3), (6.4) and (6.5). In theory,

it is possible to predict exactly the three fundamental types of chemical reaction

patterns. However, in numerical practice, it is difﬁcult to control the ﬂow rate of

146 6 Fluid Mixing, Heat Transfer and Non-Equilibrium Redox Chemical Reactions

a speciﬁc value due to numerical round off and cutoff errors. For this reason, the

numerical simulated mineral precipitation patterns here are approximate represen-

tations of the three fundamental types of mineral precipitation patterns.

The computational model (shown in Fig. 6.1) and related parameters for the ver-

iﬁcation example in the previous section (i.e. Sect. 6.3) are used to illustrate three

different types of mineral precipitation in a focusing and mixing system of two reac-

tive ﬂuids. Two different ﬂuid pressure gradients are considered to control the ﬂow

rate (i.e. vertical ﬂuid velocity) within the surrounding rock during the numerical

simulation. To simulate the fast ﬂow rate associated with type 2 (as stated in the

previous section), the ﬂuid pressure gradient is assumed to be a lithostatic pressure

gradient, which results in a vertical ﬂow rate of 1.7 ×10

–9

m/s within the surround-

ing rock. This gradient is used because it is considered to be near the highest reason-

able ﬂuid pressure gradient likely to be encountered in nature. On the other hand,

in order to simulate the slow ﬂow rate associated with types 1 and 3 (see the pre-

vious section), the excess ﬂuid pressure gradient is assumed to be one percent of

a lithostatic pressure gradient minus a hydrostatic one, which results in a vertical

ﬂow rate of 1.7 ×10

–11

m/s within the surrounding rock. It is known that for ﬂuid

pressure gradient dominated ﬂow, the ﬂow pattern around a permeable fault zone is

dependent only on the contrast in permeability between the fault and the surround-

ing rocks, the geometry of the fault zone and the inﬂow direction relative to the axis

of the fault (Phillips 1991, Zhao et al. 1999d) and is independent of the pressure

gradient along the fault. This means that although the ﬂuid pressure gradient is dif-

ferent in the three simulations, the streamline pattern within and around the fault

zone must be identical for all three cases, as demonstrated by the related numerical

result shown in Fig. 6.2. In all cases, ﬂuid ﬂow converges into the fault zone at the

lower end and diverges out of the fault zone at the upper end (see also Phillips 1991,

Zhao et al. 1999d).

6.4.4.1 The First Type of Chemical Reaction Pattern

The ﬁrst type of chemical reaction pattern results from an approximate representa-

tion of Type 1 in Sect. 6.4.3; the background ﬂow velocity within the surrounding

rock is not exactly equal to zero in the numerical simulation but instead is one per-

cent of the lithostatic pressure gradient minus the hydrostatic gradient. Here two

controlling chemical reaction rates, namely 10

–7

(1/s) and 10

–11

(1/s) are considered

to investigate the effect on chemical reaction patterns. Due to ﬂow focusing, the

maximum vertical ﬂow velocity within the fault zone is about 3.16 ×10

–10

m/s.

Figure 6.9 shows the concentration distributions of the two chemical reactants

and the corresponding chemical product at two time instants of t =500,000 and

t =800,000 years. The distribution of the concentration of the chemical product

comprises a lenticular shape within the fault zone. This coincides with what is

expected from the previous theoretical analysis. For a fault zone of width 250 m,

the optimal chemical reaction rate calculated from Eq. (6.33) is 4.8 ×10

–15

(1/s) so

that chemical equilibrium can be attained within the whole width of the fault zone.

Since the optimal chemical reaction rate is much smaller than the two chemical

6.4 Applications of the Proposed Decoupling Procedure 147

Reactant A (t = 500,000 years)

Reactant B (t = 500,000 years) Reactant B (t = 800,000 years)

Reactant A (t = 800,000 years)

Product (k

R

=

10

–7

(1/s), t = 500,000 years)

Product (k

R

=

10

–11

(1/s), t = 500,000 years) Product (k

R

=

10

–11

(1/s), t = 800,000 years)

Product (k

R

=

10

–7

(1/s), t = 800,000 years)

Fig. 6.9 Concentration distributions of the chemical reactants and product at two different time

instants (Type 1)

148 6 Fluid Mixing, Heat Transfer and Non-Equilibrium Redox Chemical Reactions

reaction rates used in the simulation, the thickness of the chemical equilibrium

within the fault zone is much smaller than the width of the fault zone itself. The

theoretical estimate of the thickness of chemical equilibrium within the fault zone

is 10.95 m for a controlling chemical reaction rate of 10

–11

(1/s), and 0.11 m for a

controlling chemical reaction rate of 10

–7

(1/s). Hence, the maximum concentration

distribution of the chemical product comprises a very thin membrane; that is, min-

eral precipitation comprises a thin lenticular shape within the fault zone and starting

at the lower tip of the fault. This is the fundamental characteristic of the ﬁrst type of

chemical reaction pattern.

6.4.4.2 The Second Type of Chemical Reaction Pattern

The second type of chemical reaction pattern is deﬁned as an approximate

representation of Type 2 in the previous section. In this example, the background

ﬂuid pressure gradient within the surrounding rock is set equal to a lithostatic

pressure gradient. Two controlling chemical reaction rates, namely k

R

= ∞ and

k

R

= 10

−11

(1/s), are considered to investigate the effects of different chemical

kinetics on chemical reaction patterns. From a chemical kinetics point of view,

k

R

= ∞ represents an equilibrium chemical reaction. Due to ﬂow focusing, the

maximum vertical ﬂow velocity within the fault zone is about 3.02 ×10

–8

m/s.

Figure 6.10 shows the concentration distributions for the two chemical reac-

tants and the corresponding chemical product at two time instants of t =5000

and t =8000 years. Both chemical reactants are transported into the computational

domain from the left half and right half of the base of the model. Due to ﬂuid ﬂow

focusing, both chemical reactants are transported much faster within the fault zone

than in the surrounding rock. There is a strong interaction between solute advec-

tion, diffusion/dispersion and chemical reaction rate (i.e. k

R

= ∞). Although both

reactants are transported into the fault zone, the mixing of the two ﬂuids carrying

them is controlled by solute diffusion and dispersion. Since the chemical reaction

rate is inﬁnite, the corresponding chemical equilibrium length due to solute diffu-

sion/dispersion is identical to zero normal to the fault zone. This implies that the

controlling chemical reaction rate is too fast to allow both the reactants to diffuse

across their common boundary so that ﬂuid mixing cannot effectively take place

within the fault zone. However, around the exit region of the fault zone, the ﬂuid

ﬂow decreases and, more importantly, diverges so that a high concentration of the

chemical product is produced around the exit region of the fault zone.

In the case of a non-equilibriumchemical reaction characterized by a slowchemi-

cal reaction rate (e.g. k

R

= 10

−11

(1/s)), the theoretical chemical equilibrium length

calculated from Eq. (6.28) is 5.48 m, while the theoretical chemical equilibrium

length calculated from Equation (6.25) is 8630 m in the direction of the fault axis.

Thus the chemical equilibrium length due to solute advection is greater than the

length of the fault itself, so that the chemical product distribution within the fault

zone is controlled by solute advection. Because the chemical equilibrium length due

to the solute advection is greater than the total length of the fault zone plus its exit

region (i.e. 5000 m plus 2500 m), chemical equilibrium cannot be reached within the

6.4 Applications of the Proposed Decoupling Procedure 149

Reactant A (t = 500,000 years)

Reactant B (t = 500,000 years) Reactant B (t = 800,000 years)

Reactant A (t = 800,000 years)

Product (k

R

=

∞, t = 500,000 years)

Product (k

R

=

10

–11

(1/s), t = 500,000 years) Product (k

R

=

10

–11

(1/s), t = 800,000 years)

Product (k

R

=

∞, t = 800,000 years)

Fig. 6.10 Concentration distributions of the chemical reactants and product at two different time

instants (Type 2)

150 6 Fluid Mixing, Heat Transfer and Non-Equilibrium Redox Chemical Reactions

height of the model, indicating that two ﬂuids cannot mix to produce an extensive

mixing region due to the fast advection of the ﬂuids within the fault.

Thus, for the two end-members of chemical reaction rates, namely a very fast

equilibrium reaction and a slow reaction with a controlling chemical reaction rate of

10

–11

(1/s), chemical equilibrium cannot be attained within the fault, implying that

mineral precipitation cannot take place within a permeable vertical fault zone for

the chemical reaction considered here. This is the fundamental characteristic of the

second type of chemical reaction pattern considered in this investigation.

6.4.4.3 The Third Type of Chemical Reaction Pattern

The third type of chemical reaction pattern is a representation of Type 3 in the

previous section. The fundamental characteristic of this type of chemical reaction

pattern is that chemical equilibrium, which results in a considerable equilibrium

thickness in the direction normal to the fault zone, can be achieved within the fault

zone. In order to enable the starting position of chemical equilibrium to be close to

the lower tip of the vertical fault, the background ﬂuid pressure gradient within the

surrounding rock is one percent of the lithostatic pressure gradient minus the hydro-

static gradient. Due to ﬂow focusing, the maximum vertical ﬂow velocity within the

fault zone is about 4.68 ×10

–10

m/s in the numerical simulation. Since the optimal

reaction rate is directly proportional to the solute diffusion/dispersion coefﬁcient, it

is desirable to select the value of a solute diffusion/dispersion coefﬁcient as large as

possible, so that the total CPU time in the simulation can be signiﬁcantly reduced.

For this reason, the solute diffusion/dispersion coefﬁcient is assumed to be 3 ×10

–8

m

2

/s in the numerical simulation. If the thickness of chemical equilibrium in the lat-

eral direction of the fault zone is 80 m, the optimal chemical reaction rate is about

1.9 ×10

–11

(1/s), as can be calculated from Eq. (6.37).

Figure 6.11 shows concentration distributions of the two chemical reactants

and the corresponding chemical product at two time instants of t =500,000 and

t =800,000 years. The maximum concentration distribution of the chemical prod-

uct generates considerable thickness within the fault zone. This agrees well with

what is expected from the previous theoretical analysis given in Sect. 6.4.3. Due

to the low ﬂuid velocity, a chemical equilibrium zone is also generated in the ﬂow

convergent region just in front of the fault zone. It is noted that this chemical equilib-

rium zone is almost separated from the chemical equilibrium zone within the fault.

This phenomenon results from the distribution of ﬂuid velocity vectors arising from

ﬂuid ﬂow focusing just outside and within the fault zone.

The geological implication of the third type of chemical reaction pattern is that if

a mineral precipitation pattern of a certain thickness and length within a permeable

vertical fault zone is observed, then it is possible to estimate both the optimal ﬂow

rate and the optimal reaction rate during the formation of this precipitation pattern.

On the other hand, if the ﬂow rate and reaction rate are known, then it is possible to

estimate the width of the potential mineral precipitation pattern within a permeable

vertical fault zone.

6.4 Applications of the Proposed Decoupling Procedure 151

Reactant A (t = 500,000 years)

Reactant B (t = 500,000 years)

Reactant B (t = 800,000 years)

Reactant A (t = 800,000 years)

Product (k

R

=

2

×

10

–11

(1/s), t = 500,000 years) Product (k

R

=

2

×

10

–11

(1/s), t = 800,000 years)

Fig. 6.11 Concentration distributions of the chemical reactants and product at two different time

instants (Type 3)

Chapter 7

An Equivalent Source Algorithm for Simulating

Thermal and Chemical Effects of Intruded

Magma Solidiﬁcation Problems

Consideration of the effects of magma ascending and solidiﬁcation is important to

the further understanding of ore body formation and mineralization in the crust of

the Earth. Although various possible fundamental mechanisms of magma ascending

in the crust are proposed (Johnson and Pollard, 1973, Marsh 1982, Lister and Kerr

1991, Rubin 1995, Weinberg 1996, Bons et al. 2001), the development of numeri-

cal algorithms for simulating the proposed magma ascending mechanisms is still

under-developed. For example, continuum-mechanics-based numerical methods

have encountered serious difﬁculties in simulating the random generation and prop-

agation of hydro-fractured cracks, the magma ﬂow within these cracks, the solidiﬁ-

cation of the ascending magma due to heat losses to the surrounding rocks, and so

forth. In order to overcome these difﬁculties, particle-based numerical simulations

have been developed rapidly in recent years (Zhao et al. 2006f, 2007b, c, d, 2008g).

However, due to the different time and length scales involved in ore body formation

and mineralization problems, it is also very difﬁcult, even if not impossible, to use

the present particle-based numerical methods to simulate all the important processes

associated with ore body formation and mineralization problems in the crust of the

Earth. As a long-term development strategy, we need to develop multiple time and

length scale modelling techniques and algorithms so that particle simulation meth-

ods, combined with newly-developed techniques and algorithms, can be used to

solve such large scale geological problems. As an expedient strategy, although it is

impossible to use the continuum-mechanics-based numerical methods to simulate

directly the magma ascent processes, we can develop some useful algorithms, in

combination with continuum-mechanics-based numerical methods, to simulate the

effects of the magma ascent processes. Thus, the main motivation of carrying out

this study is to develop a useful algorithm to consider the dynamic consequences

involved in magma ascent processes using continuum-mechanics-based numerical

methods.

In terms of the magma intrusion mechanism, a large amount of theoretical work

has been carried out previously, even though it is based on simple conceptual mod-

els (Johnson and Pollard, 1973, Marsh 1982, Lister and Kerr 1991, Rubin 1995,

Weinberg 1996, Bons et al. 2001). Although the previous theoretical work needs to

be quantitatively reﬁned, it can be used to estimate the total volume of the intruded

C. Zhao et al., Fundamentals of Computational Geoscience,

Lecture Notes in Earth Sciences 122, DOI 10.1007/978-3-540-89743-9 7,

C

Springer-Verlag Berlin Heidelberg 2009

153

154 7 Simulating Thermal and Chemical Effects of Intruded Magma Solidiﬁcation Problems

magma. If an analytical estimation of the volume of the intruded magma is not

available for most complicated geological situations, the particle numerical method

may be useful to simulate the magma intrusion process alone so that the volume of

the intruded magma can also be estimated. Given the total volume of the intruded

magma, we can use continuum-mechanics-based numerical methods, such as the

ﬁnite element method and ﬁnite difference method, to simulate its thermal effects

by considering the heat release during the solidiﬁcation of the intruded magma. This

means that we need to develop an equivalent algorithm to transform the original

magma intrusion problem into a heat transfer problem with the internal heat gener-

ation of the intruded magma. Clearly, the key issue associated with the developed

equivalent algorithm is to determine the heat release rate of the intruded magma dur-

ing its solidiﬁcation. Once the time history of the heat release rate of the intruded

magma during solidiﬁcation is obtained, the ﬁnite element method can be used to

simulate the thermal effects of the intruded magma in the crust of the Earth. Using

the proposed equivalent algorithm, the moving boundary problem (Crank 1984,

Alexiades and Solomon 1993) associated with the original problem during magma

solidiﬁcation can be avoided. As a direct result, the efﬁciency of the ﬁnite element

method can be much improved. This may be considered as one of the major advan-

tages of the proposed equivalent algorithm in dealing with the thermal effects of

magma intrusion problems in the crust of the Earth.

Geological problems may involve different time and length scales in the descrip-

tions of their different physical and chemical processes. With magma intrusion into

the Earth’s crust taken as an example, the time scale of the magma intrusion pro-

cess, which includes both the creation of the magma chamber and ascent process for

the intruded magma, is much smaller than that of the magma solidiﬁcation process,

which includes both the release of volatile ﬂuids from the magma and chemical

reaction processes within the crust due to the release of the volatile ﬂuids. On the

other hand, the volume of the intruded magma is usually much smaller than that of

the Earth’s crust of interest. This means that the whole magma intrusion problem

is, in essence, a problem of multiple time and length scales. Nevertheless, due to

the signiﬁcant time and length scale differences between the magma intrusion and

solidiﬁcation processes, it is possible to simulate these two different processes using

different analytical models. This will allow the detailed mechanisms associated with

each of the two processes to be modelled using totally different methodologies. For

example, if the magma intrusion process itself is of particular interest, then particle-

based numerical methods can be used to simulate the initiation and propagation of

random cracks during the ascending of the intruded magma. However, if the thermal

and chemical effects/consequences of the intruded magma are of particular inter-

est, then continuum-mechanics-based numerical methods can be used to simulate

heat transfer and mass (i.e. chemical species) transport within the Earth’s crust. To

the best knowledge of the authors, these methods have not been used to simulate

the release of volatile ﬂuids from the magma and the chemical reaction processes

within the crust due to the release of the volatile ﬂuids. Since both the release of

volatile ﬂuids from the magma and the transport of the released volatile ﬂuids may

have signiﬁcant effects on ore body formation and mineralization in the upper crust

7.1 An Equivalent Source Algorithm for Simulating Thermal and Chemical Effects 155

of the Earth, a numerical method needs to be developed to simulate the chemical

consequences of the magma solidiﬁcation in porous rocks.

A large amount of geochemical research has indicated that intruded magma

in the Earth’s crust has different chemical compositions and, therefore, different

types of rocks can be formed during the solidiﬁcation and crystallization of the

intruded magma. Both temperature and pressure conditions during solidiﬁcation of

the magma can also affect the resulting rock types dramatically. Rhyolite and basalt,

instead of granite and gabbro, are formed if the solidiﬁcation conditions of the fel-

sic and maﬁc magmas are changed from intrusive into extrusive. It is well known

that the solidus of the intruded magma is mainly dependent on the magma composi-

tion, the contents of water and other volatile ﬂuids. For a particular kind of intruded

magma, the abovementioned information is obtainable from geochemical and iso-

topic analyses. Therefore, given a particular kind of intruded magma, it is possible

to determine the contents of water and other volatile ﬂuids, which should be released

when the intruded magma becomes solidiﬁed. This implies that the volatile ﬂuids

released from the intruded magma can be quantitatively simulated in the numerical

analysis, which is another important issue to be addressed in this study.

Due to the complex nature of the magma intrusion problem within the crust

of the Earth, it is useful to conduct numerical simulation of this kind of problem

progressively from a simplistic stage into more complicated stages. This research

methodology is rational because of the multiple time and length scales of the prob-

lem itself. Thus, we plan to solve the magma intrusion problem using the following

three level models. For the ﬁrst level model, we primarily consider the effects of

the post-solidiﬁcation magma on pore-ﬂuid ﬂow, heat transfer and ore forming pat-

terns within the upper crust of the Earth. For the second level model, we consider

the thermal and chemical effects of the post-intrusion but pre-solidiﬁcation magma

on pore-ﬂuid ﬂow, heat transfer and ore forming patterns within the upper crust

of the Earth. In this model, we must develop some useful and efﬁcient computer

algorithms to simulate the magma solidiﬁcation problem. For the third level model,

we will consider the intrusion process itself at a much smaller scale using parti-

cle mechanics-based computer algorithms. Once efﬁcient numerical algorithms for

dealing with above three level models are developed, it is possible to integrate them

to simulate the whole process of the magma intrusion problem. So far, we have

completed some work for the ﬁrst level model (Zhao et al. 2003e). In this study, we

will develop some useful and efﬁcient computer algorithms to simulate the magma

solidiﬁcation problem associated with the second level model.

7.1 An Equivalent Source Algorithm for Simulating

Thermal and Chemical Effects of Intruded Magma

Solidiﬁcation Problems

Owing to the importance for many scientiﬁc issues and technical applications, theo-

retical and numerical analyses of heat transfer with phase change have been carried

out for more than half a century. There is an extensive literature which reports and

156 7 Simulating Thermal and Chemical Effects of Intruded Magma Solidiﬁcation Problems

reviews the development of this subject (Carslaw and Jaeger 1959, Crank 1984,

Alexiades and Solomon 1993, Zhao and Heinrich 2002). The problem associated

with numerical modelling of the thermal effect of solidiﬁcation of intruded magma

is that the characteristic dimension of the whole geological system under considera-

tion is on the scale of tens and hundreds of kilometers, but the characteristic dimen-

sion of the intruded magma, such as a sill or dike (Johnson and Pollard 1973, Lister

and Kerr 1991, Rubin 1995, Weinberg 1996) is on the scale of meters and tens of

meters. As a result, the detailed solidiﬁcation process of the intruded magma might

not be important, but the thermal effect caused by the heat release during the solidiﬁ-

cation of the intruded magma is important, at least from the ore body formation and

mineralization point of view. Since heat release during solidiﬁcation of the intruded

magma can be represented by a physically equivalent heat source, it is possible to

transform the original heat transfer problem with phase change due to solidiﬁcation

of the intruded magma into a physically equivalent heat transfer problem without

phase change but with the equivalent heat source. The above physical understand-

ing means that from the numerical modelling point of view, we can remove the

moving boundary problem associated with the detailed solidiﬁcation process of the

intruded magma, so that we can use a ﬁxed ﬁnite element mesh to consider the ther-

mal effect of the intruded magma by solving the physically equivalent heat transfer

problem with an equivalent heat source. This is the basic idea behind the proposed

equivalent algorithm for simulating the thermal effect of the intruded magma solid-

iﬁcation in this study. The similar, even though not identical, approaches, such as

the immersed boundary method (Beyer and LeVeque 1992), the level-set method

(Osher and Sethian 1998, Sethian 1999), the segment projection method (Tornberg

and Engquist 2003a) and so forth, have been used to tackle other different prob-

lems with moving interfaces or fronts (Smooke et al. 1999, Mazouchi and Homsy

2000). In terms of solving partial differential equations with delta function source

terms numerically, Walden (1999), and Tornberg and Engquist (2003b) have inves-

tigated the convergence and accuracy of the numerical method. Recently, the ﬁxed

grid approach has been successfully used to solve the phase change problem associ-

ated with moisture transport in high-temperature concrete materials (Schreﬂer 2004,

Schreﬂer et al. 2002, Gawin et al. 2003). In this chapter, it will be extended to the

solution of the phase change problem associated with intruded magma solidiﬁcation

in geological systems. The proposed algorithm is only valid when the characteristic

length scale of the system is much larger than that of the intruded magma, which is

true for most geological systems (Johnson and Pollard 1973, Lister and Kerr 1991,

Rubin 1995, Weinberg 1996).

In what follows, we will present the proposed equivalent algorithm for simulating

the chemical effect of solidiﬁcation of the intruded magma in geological systems.

Since the volatile ﬂuids released during solidiﬁcation of the intruded magma may

have many different chemical components/species, it is ideal to describe the gov-

erning equation of each chemical component separately. However, if the diffusivity

of each chemical component is assumed to be identical, then the concept of the

total concentration, which is the summation of the concentration of all the chemical

components, can be used to describe the governing equation of the released volatile

7.1 An Equivalent Source Algorithm for Simulating Thermal and Chemical Effects 157

ﬂuids. Thus, we can use one governing equation to describe variations of the total

concentration of the released volatile ﬂuids in the system. In this way, computer

efforts can be reduced signiﬁcantly in dealing with the intruded magma solidiﬁ-

cation problems. Because the volumetric amount of the released volatile ﬂuids is

relatively small to the intruded magma, their effects on the heat transfer process

can be neglected in the analysis of the intruded magma solidiﬁcation problem. This

allows us to assume that thermal equilibrium between the released volatile ﬂuids

and the intruded magma/rock has been achieved. Thus, the governing equations of

the original heat transfer and mass transport problem considering the phase change

during the intruded magma solidiﬁcation (Carslaw and Jaeger 1959, Crank 1984,

Alexiades and Solomon 1993, Zhao and Heinrich 2002) can be described, for a

two-dimensional problem, as

(ρ

R

c

pR

)

∂T

R

∂t

= λ

R

∂

2

T

R

∂x

2

+

∂

2

T

R

∂y

2

(x, y) ∈ V

R

, (7.1)

(ρ

M

c

pM

)

∂T

M

∂t

= λ

M

∂

2

T

M

∂x

2

+

∂

2

T

M

∂y

2

(x, y) ∈ V

M

, (7.2)

∂C

T

∂t

= D

∂

2

C

T

∂x

2

+

∂

2

C

T

∂y

2

+δ(x − x

I

, y − y

I

)Q(x, y, t ) (x, y) ∈ V

R

+ V

M

,

(7.3)

where T

R

and T

M

are the temperature of the rock and intruded magma; ρ

R

, c

pR

and

λ

R

are the density, speciﬁc heat and thermal conductivity of the rock; ρ

M

, c

pM

and

λ

M

are the density, speciﬁc heat and thermal conductivity of the intruded magma;

C

T

is the total concentration of the released volatile ﬂuids; D is the diffusivity of the

released volatile ﬂuids; Q is the mass source of the released volatile ﬂuids during

the intruded magma solidiﬁcation; x

I

and y

I

are the x and y coordinate components

of the interface position; δ is the delta function of values of unity and zero; V

R

and

V

M

are the spaces occupied by the rock and intruded magma.

It is noted that, although the released volatile ﬂuids during the intruded magma

solidiﬁcation are comprised of H

2

O, CO

2

, H

2

S, HCl, HF, SO

2

and other substances

(Burnham 1979, Barns 1997), H

2

O is the most abundant magmatic volatile and CO

2

is the second most abundant magmatic volatile in the intruded magma. For this rea-

son, the solubility of H

2

O in silicate melts has been investigated for many years.

These extensive studies (Burnham 1979, Barns 1997) have demonstrated that if the

constraints of the solution model for the NaAlSi

3

O

8

−H

2

O system are imposed, H

2

O

solubilities in the igneous-rock melts are essentially identical to those in NaAlSi

3

O

8

melts. Therefore, the solubility of H

2

O in the NaAlSi

3

O

8

melt can be used to approx-

imately determine the mass source of the released volatile ﬂuids during the intruded

magma solidiﬁcation.

Since the original magma solidiﬁcation problem belongs to a moving inter-

face problem, the temperature and heat ﬂux continuity conditions on the interface

between the rock and intruded magma are as follows:

158 7 Simulating Thermal and Chemical Effects of Intruded Magma Solidiﬁcation Problems

T

R

= T

M

(x

I

, y

I

) ∈ Γ

RM

, (7.4)

λ

R

n

x

∂T

R

∂x

+n

y

∂T

R

∂y

−λ

M

n

x

∂T

M

∂x

+n

y

∂T

M

∂y

=ρ

M

[L +c

pM

(T

I M

− T

m

)]

n

x

∂x

I

∂t

+n

y

∂y

I

∂t

(x

I

, y

I

) ∈ Γ

RM

,

(7.5)

where x

I

and y

I

are the x and y coordinate components of the interface position;

n

x

and n

y

are the x and y components of the unit normal to the interface between

the rock and intruded magma; T

IM

is the temperature of the intruded magma; T

m

is the solidiﬁcation temperature of the intruded magma; L is the latent heat of

fusion of the intruded magma; Γ

RM

is the interface between the rock and intruded

magma.

Except for the boundary conditions on the interface between the rock and

intruded magma, the boundary conditions on the other boundaries of the rock

domain and intruded magma domain can be either of the Dirichlet, Neumann or

mixed type (Carslaw and Jaeger 1959, Crank 1984, Alexiades and Solomon 1993,

Zhao and Heinrich 2002). Since the boundary conditions on the other boundaries of

the rock domain and intruded magma domain are trivial, it is not necessary to repeat

them here.

In order to develop an efﬁcient and effective numerical algorithm for dealing with

solidiﬁcation of intruded magma in sills and dikes, it is necessary to introduce some

related concepts. As shown in Fig. 7.1, supposing the initial position of the interface

between the magma and rock is at position 1, the interface moves to position 2

due to magma solidiﬁcation during a time period, Δt

M

. As a result, the magma

solidiﬁcation thickness is expressed by ΔL

M

. If the length of the interface front is

ΔL

R

, then the magma solidiﬁcation area is the product of ΔL

M

and ΔL

R

during

the time period Δt

M

. Since heat and volatile ﬂuids are only released during magma

solidiﬁcation, we may consider the released heat and volatile ﬂuids either as surface

heat and mass sources in a two-dimensional problem or as volumetric heat and mass

sources in a three-dimensional one. This means that it is physically possible to use

Position 2

Position 1

Interface movement

direction

R

L Δ

M

L Δ

Rock

Magma

Fig. 7.1 Basic concepts related to the magma solidiﬁcation problem

7.1 An Equivalent Source Algorithm for Simulating Thermal and Chemical Effects 159

a ﬁxed ﬁnite element mesh to consider the released heat and volatile ﬂuids during

the magma solidiﬁcation process. The key issue associated with the ﬁnite element

analysis using a ﬁxed mesh is that, for a given magma solidiﬁcation thickness ΔL

M

,

which can be easily simulated by the ﬁxed mesh, we need to determine the time

period Δt

M

so that the interface between the magma and rock just moves a distance

being equal to ΔL

M

in the interface movement direction. Since the magma is fully

solidiﬁed within the region of the solidiﬁcation thickness ΔL

M

, the delta function

used in Eq. (7.3) should have a value of unity. Except in this solidiﬁcation region,

the delta function should have a value of zero, meaning that there is no magma

solidiﬁcation taking place in other regions. This is the reason why the delta function

only has values of unity and zero in the present numerical algorithm. Generally, for

the same magma solidiﬁcation thickness ΔL

M

, the solidiﬁcation time period can

vary signiﬁcantly as the magma solidiﬁcation proceeds. This requires us to develop

a numerical algorithm, in which the ﬁnite element mesh is ﬁxed but the integration

time-step being equivalent to the solidiﬁcation time period must be variable.

Based on the above considerations, Eqs. (7.1), (7.2), (7.3), (7.4) and (7.5) can be

represented by the following equations:

(ρ

R

c

pR

)

∂T

R

∂t

= λ

R

∂

2

T

R

∂x

2

+

∂

2

T

R

∂y

2

+δ(x−x

I

, y−y

I

) f (x, y, t ) (x, y) ∈ V

R

−V

M

,

(7.6)

∂C

T

∂t

= D

∂

2

C

T

∂x

2

+

∂

2

C

T

∂y

2

+δ(x −x

I

, y −y

I

)Q(x, y, t ) (x, y) ∈ V

R

−V

M

,

(7.7)

where δ is the delta function with values equal to unity and zero; f (x, y, t ) is the

physically equivalent heat source due to the solidiﬁcation of the intruded magma.

f (x, y, t ) =

ρ

M

L +c

pM

(T

I M

− T

m

)

n

x

∂x

I

∂t

+n

y

∂y

I

∂t

ΔL

Mk

, (7.8)

where ΔL

Mk

is the solidiﬁcation thickness of the intruded magma during a time

period Δt

Mk

; k is the time-step index of integration in the ﬁnite element analysis.

The mass source of the released volatile ﬂuids in Eq. (7.7) during the intruded

magma solidiﬁcation can be determined using the solubility of H

2

O in the

NaAlSi

3

O

8

melt (Burnham 1979, Barns 1997). In this regard, the mole fraction of

H

2

O in the NaAlSi

3

O

8

melt is expressed as follows:

X

m

w

=

1

k

mf

w

(X

m

w

≤ 0.5), (7.9)

160 7 Simulating Thermal and Chemical Effects of Intruded Magma Solidiﬁcation Problems

X

m

w

= 0.5 +

1

6.25 −

2667

T

ln

4

k

mf

w

(X

m

w

> 0.5) (7.10)

where X

m

w

is the mole fraction of H

2

O in the NaAlSi

3

O

8

melt; T is in Kelvin; k

mf

w

is

the equilibrium constant for H

2

O in melts of feldspar composition.

ln k

mf

w

= 5 +(ln P)(4.481 ×10

−8

T

2

−1.51 ×10

−4

T −1.137)

+(ln P)

2

(1.831 ×10

−8

T

2

−4.882 ×10

−5

T +4.656 ×10

−2

)

+7.8 ×10

−3

(ln P)

3

−5.012 ×10

−4

(ln P)

4

+4.754 ×10

−3

T

−1.621 ×10

−6

T

2

,

(7.11)

where P is the pressure of the intruded magma; P and T are in bars and Kelvin,

respectively.

Using the concept of molar mass, the mass source of the volatile ﬂuids released

during solidiﬁcation from the intruded magma can be expressed as

Q(x, y, t ) =

X

m

w

W

m

w

X

m

w

W

m

w

ρ

m

w

+

(1 − X

m

w

)W

m

albi t e

ρ

m

albi t e

Δt

Mk

, (7.12)

where Δt

Mk

is the time period required to complete the magma solidiﬁcation within

a given solidiﬁcation thickness ΔL

Mk

; W

m

w

and ρ

m

w

are the molar mass and den-

sity of the volatile ﬂuids in the magma; W

m

albi t e

and ρ

m

albi t e

are the molecular mass

and density of the albite (NaAlSi

3

O

8

) melt. It is noted that using the deﬁnition in

Eq. (7.12), the mass source of the released volatile ﬂuids has units of the density of

the albite (NaAlSi

3

O

8

) melt divided by time.

In the case of the intruded magma temperature being equal to the solidiﬁcation

magma temperature, Eq. (7.8) can be rewritten as

f (x, y, t ) =

ρ

M

L

n

x

∂x

I

∂t

+n

y

∂y

I

∂t

ΔL

Mk

. (7.13)

7.2 Implementation of the Equivalent Source Algorithm

in the Finite Element Analysis with Fixed Meshes

If the physically equivalent heat source term, f (x, y, t ), is determined either ana-

lytically or experimentally, Eqs. (7.6) and (7.7) can be directly solved using the

conventional ﬁnite element method (Zienkiewicz 1977). For dike-like and sill-like

intruded magmas, the physically equivalent heat source due to the solidiﬁcation can

7.2 Implementation of the Equivalent Source Algorithm 161

R

L

M

L

R

L

R

L

y

x

0

Fig. 7.2 A representative model for the magma solidiﬁcation problem

be determined analytically through the following ideal experiment. As shown in

Fig. 7.2, the sample in the ideal experiment is a rectangular domain ﬁlled with the

intruding magma in the middle and the intruded rocks on the either side. All the

external boundaries of the sample are insulated and the temperatures of the rock and

intruded magma are T

R

and T

m

, respectively. The characteristic length of the rock

is L

R

, while the characteristic length of the intruded magma is L

M

2. It is assumed

that the characteristic length of the rock is much larger than that of the intruded

magma, which is true for sills and dikes in geology (Johnson and Pollard, 1973,

Lister and Kerr 1991, Rubin 1995, Weinberg 1996). Under this assumption, the

solidiﬁcation problem in the ideal experiment can be treated as a one-dimensional

Stefan solidiﬁcation problem, for which the analytical solution is already available

(Carslaw and Jaeger 1959).

If the characteristic length of the intruded magma is divided into K equal parts,

which are modelled by K ﬁnite element meshes of equal length in the magma solid-

iﬁcation direction, then the length of the ﬁnite element mesh in this direction is

Δx = L

M

**(2K). In order to use the ﬁxed ﬁnite element mesh, it is useful to keep
**

Δx constant. For this purpose, we deﬁne that the solidiﬁcation thickness of the

intruded magma during a time period Δt

Mk

is equal to the length of the ﬁnite ele-

ment mesh in the magma solidiﬁcation direction and, therefore, ΔL

Mk

= Δx =

L

M

**(2K). This implies that for this ideal experiment, the scalar product of (n
**

x

, n

y

)

and (∂x

I

∂t , ∂y

I

**∂t ) can be expressed as follows:
**

n

x

∂x

I

∂t

+n

y

∂y

I

∂t

= n

x

∂x

I

∂t

≈

Δx

Δt

Mk

=

ΔL

Mk

Δt

Mk

, (7.14)

where Δt

Mk

is the time period when the solidiﬁcation boundary moves from one

side of the kth ﬁnite element in the magma solidiﬁcation direction at time t = t

Mk−1

to another side of the kth ﬁnite element in the magma solidiﬁcation direction at time

t = t

Mk

.

Δt

Mk

= t

Mk

−t

Mk−1

(k = 1, 2, 3, . . . , K). (7.15)

162 7 Simulating Thermal and Chemical Effects of Intruded Magma Solidiﬁcation Problems

The time, at which the intruded magma solidiﬁcation boundary reaches both

sides of the kth ﬁnite element boundary in the magma solidiﬁcation direction, can

be determined from the analytical solution (Carslaw and Jaeger 1959).

t

Mk

=

k

2

(Δx)

2

4αβ

2

, t

Mk−1

=

(k −1)

2

(Δx)

2

4αβ

2

(k = 1, 2, 3, . . . , K),

(7.16)

where α = λ

R

(ρ

R

c

pR

) is the thermal diffusivity of the rock; β can be determined

from the following transcendental equation:

e

−β

2

β[1 +erf(β)]

=

L

√

π

c

pR

(T

m

− T

R0

)

, (7.17)

where T

R0

is the initial temperature of the rock; erf(β) is the error function of vari-

able β.

erf(β) =

2

√

π

β

0

e

−r

2

dr. (7.18)

Substituting Eqs. (7.14) into Eq. (7.13) yields the following equation:

f (x, y, t

Mk

) =

ρ

M

L

Δt

Mk

(k = 1, 2, 3, . . . , K). (7.19)

Finally, substituting Eq. (7.16) into Eq. (7.19) yields the following equation:

f (x, y, t

Mk

) =

4ρ

M

Lαβ

2

(2 k −1)(Δx)

2

(k = 1, 2, 3, . . . , K), (7.20)

where Δx = ΔL

Mk

= L

M

(2K) and ΔL

Mk

is the constant solidiﬁcation thickness

of the intruded magma during the variable time period Δt

Mk

.

It needs to be pointed out that since the magma solidiﬁcation boundary passes

only through the kth ﬁnite element from the initial intruded interface between the

rock and magma during the time period Δt

Mk

, both the mass source of the released

volatile ﬂuids and the physically equivalent heat source expressed in Eqs. (7.12) and

(7.20) only need to be applied in the kth ﬁnite element, which is numbered from the

initial intruded interface between the rock and magma, in the ﬁnite element analysis.

This means that if a ﬁxed ﬁnite element mesh is employed, the variable time step

expressed by Eqs. (7.15) and (7.16) needs to be used during the magma solidiﬁcation

period in the ﬁnite element analysis. From the numerical analysis point of view, a

change in the time step is much easier to implement than a change in the ﬁnite

element mesh in the ﬁnite element analysis. This is the main advantage in using the

proposed equivalent source algorithm to simulate the chemical effect of the intruded

magma solidiﬁcation in geological systems.

7.3 Veriﬁcation and Application of the Equivalent Source Algorithm 163

Due to the geometrical symmetry of the ideal experiment, analytical solutions

for the temperature distribution during the solidiﬁcation of the intruded magma can

be expressed as follows.

T

R

= T

m

0 ≤ x ≤

L

M

2

−2β

√

αt , t ≤ t

MMax

, (7.21)

T

R

= T

R0

+

(T

m

− T

R0

)erfc

2x + L

M

−4β

√

αt

4

√

αt

1 +erf(β)

L

M

2

−2β

√

αt ≤ x ≤ L

R

, t ≤ t

MMax

,

(7.22)

T

R

= T

R0

+

ρ

M

L

M

c

pM

(T

m

− T

R0

) + L

2ρ

R

c

pR

√

απt

e

−

x

2

4αt

( t > t

MMax

), (7.23)

where erfc(x) = 1 − erf(x) is the complementary error function of variable x;

t

MMax

= L

2

M

(16αβ

2

) is the maximum time instant to complete the solidiﬁcation

of the intruded magma.

In summary, the proposed equivalent algorithmfor simulating the chemical effect

of the intruded magma solidiﬁcation includes the following ﬁve main steps: (1) For

the given values of the initial temperature of the rock, T

R0

, and the solidiﬁcation

temperature of the intruded magma, T

m

, Eq. (7.17) is solved to determine the value

of β; (2) Substituting the value of β into Eq. (7.20) yields the value of the phys-

ically equivalent heat source due to the solidiﬁcation of the dike-like and sill-like

intruded magma; (3) Eqs. (7.12) and (7.15) are used to calculate the mass source of

the released volatile ﬂuids during the intruded magma solidiﬁcation; (4) The inte-

gration time step is determined using Eqs. (7.15) and (7.16); (5) The conventional

ﬁnite element method is used to solve Eqs. (7.6) and (7.7) for the temperature and

concentration distribution in the whole domain during and after solidiﬁcation of the

intruded magma. Note that both the mass source and the physically equivalent heat

source are only applied to some elements in the ﬁnite element analysis during the

solidiﬁcation process of the intruded magma. The automatic transition from solid-

iﬁcation to post-solidiﬁcation of the intruded magma is another advantage in using

the proposed equivalent source algorithm for simulating the chemical effect of the

intruded magma solidiﬁcation in the conventional ﬁnite element analysis.

7.3 Veriﬁcation and Application of the Equivalent

Source Algorithm

Since the key part of the proposed algorithmfor simulating the chemical effect of the

intruded magma during solidiﬁcation is to transform the original moving interface

(i.e. the solidiﬁcation interface between the rock and intruded magma) problem into

164 7 Simulating Thermal and Chemical Effects of Intruded Magma Solidiﬁcation Problems

a physical equivalent problem which can be solved using the conventional ﬁxed-

mesh ﬁnite element method, it is necessary to use a benchmark magma solidiﬁca-

tion problem, for which the analytical solution is available, to verify the proposed

algorithm. As mentioned in the previous section, the effect of the released volatile

ﬂuids on the heat transfer and magma solidiﬁcation is negligible due to the rela-

tively small amount of their volumetric mass sources. On the other hand, transport

of the released volatile ﬂuids can be treated as a conventional mass transport prob-

lem, for which the conventional ﬁnite element method has been used for many years

(Zienkiewicz 1977). This indicates that we only need to verify the proposed algo-

rithm for simulating the heat transfer process associated with the intruded magma

solidiﬁcation. For the above reasons, only the heat transfer process associated with

the benchmark magma solidiﬁcation problem is considered to verify the proposed

algorithm. Figure 7.3 shows the ﬁnite element mesh of the benchmark magma solid-

iﬁcation problem. In this ﬁgure, the original magma intruded region, which is indi-

cated in black in the ﬁnite element mesh of the whole computational domain, is

(Finite element mesh of the whole computational domain)

(Detailed mesh for the intruded magma solidification domain)

x 0

y

Fig. 7.3 Finite element mesh of the benchmark magma solidiﬁcation problem

7.3 Veriﬁcation and Application of the Equivalent Source Algorithm 165

modeled using 240 four-node quadrilateral ﬁnite elements, while the whole com-

putational domain is modeled with 640 four-node quadrilateral ﬁnite elements. The

length and width of the whole computational domain are 42 m and 20 m, respec-

tively. The width of the intruded dike-like magma is assumed to be 2 m in this

benchmark problem. For the purpose of comparing the numerical solutions with

the corresponding analytical ones, the thermal properties of the intruded magma are

assumed to be the same as those of the surrounding rocks. The following parameters

are used in the ﬁnite element analysis: the densities of both the magma and surround-

ing rocks are 2900 kg/m

3

; speciﬁc heat is 1200 J/(kg×

◦

C); thermal conductivity is

1.74 W/(m×

◦

C); the latent heat of fusion of the intruded magma is 3.2 ×10

5

J/kg.

Since the temperature difference between the intruded magma and the surround-

ing rocks is an important indicator of this benchmark problem, it is assumed to be

1000

◦

C in the numerical computation.

Using the above thermal properties and Eq. (7.17), the value of β is determined

to be 0.73 approximately. In order to examine the effect of time-steps of the intruded

magma solidiﬁcation on the overall accuracy of the numerical solution, two compu-

tational models, namely a one-step solidiﬁcation model and a three-step solidiﬁca-

tion model, are considered in the transient ﬁnite element analysis. For the one-step

solidiﬁcation model, the intruded magma is solidiﬁed just in one time step, which

is 10.86 days from Eq. (7.16). The corresponding physically-equivalent heat source

determined from Eq. (7.20) is 989.06 W/m

3

for this one-step solidiﬁcation model.

However, for the three-step solidiﬁcation model, the intruded magma is solidiﬁed in

three variable time steps so that the intruded magma solidiﬁed equal distance from

the beginning interface between the magma and surrounding/solidiﬁed rocks. Using

Eq. (7.16), the three variable time steps are 10.86

9, 10.86

3 and (5 ×10.86)

9

days and the corresponding physically-equivalent heat sources are 8901.54 W/m

3

,

2967.18 W/m

3

and 1780.31 W/m

3

respectively. After completion of the intruded

magma solidiﬁcation, the constant time step of 10.86 days is used throughout the

rest of the transient ﬁnite element analysis.

Figures 7.4 and 7.5 show the comparisons of the analytical solutions with the

corresponding numerical solutions from the one-step solidiﬁcation model and the

three-step solidiﬁcation model respectively. It is obvious that the numerical solu-

tions of the temperature difference between the intruded magma and the surround-

ing rocks, which are obtained from both the one-step solidiﬁcation model and the

three-step solidiﬁcation model, have very good agreement with the correspond-

ing analytical ones. For example, at the time instant of 21.73 days (i.e. t =21.73

days), the numerical solutions of the maximum temperature difference between

the intruded magma and the surrounding rocks are 732.1

◦

C and 733.5

◦

C for the

one-step solidiﬁcation model and the three-step solidiﬁcation model respectively,

while the corresponding analytical solution of the maximum temperature difference

between the intruded magma and the surrounding rocks is 737.8

◦

C. The relative

numerical error between the numerical and analytical solutions for the maximum

temperature difference between the intruded magma and the surrounding rocks is

0.78% and 0.58% for the one-step solidiﬁcation model and the three-step solidi-

ﬁcation model respectively. This demonstrates that the numerical model based on

166 7 Simulating Thermal and Chemical Effects of Intruded Magma Solidiﬁcation Problems

(Analytical solution)

x (m)

1000

800

600

400

200

0

0 8.4 16.8 25.2 33.6 42.0

) ( C T

o

Δ

(Numerical solution, one-step solidification)

o t=21.73 days

+ t=108.6 days

* t=217.2 days

x t=434.4 days

x (m)

1000

800

600

400

200

0

0 8.4 16.8 25.2 33.6 42.0

) ( C T

o

Δ

o t=21.73 days

+ t=108.6 days

* t=217.2 days

x t=434.4 days

Fig. 7.4 Comparison of analytical solutions with numerical solutions (One-step solidiﬁcation)

the proposed equivalent algorithm for simulating solidiﬁcation effects of magma

intrusion problems in porous rocks can produce highly accurate numerical solu-

tions. Since the numerical solution from the one-step solidiﬁcation model has sim-

ilar accuracy to that from the three-step solidiﬁcation model, the one-step solidiﬁ-

cation model is useful for simulating thermal effects of dike-like magma intrusion

problems in porous rocks. For the purpose of examining the numerical solution sen-

sitivity to the mesh density in the magma solidiﬁcation direction, we have doubled

and tripled the mesh density in the magma solidiﬁcation region. Since the solidi-

ﬁcation interface between the rock and intruded magma is precisely considered in

the magma solidiﬁcation direction, all the three meshes of the original mesh density

(i.e. 240 ﬁnite elements to simulate the intruded magma in the solidiﬁcation direc-

tion), the doubled mesh density (i.e. 480 ﬁnite elements to simulate the intruded

magma in the solidiﬁcation direction) and the tripled mesh density (i.e. 720 ﬁnite

7.3 Veriﬁcation and Application of the Equivalent Source Algorithm 167

(Numerical solution, three-step solidification)

1000

800

600

400

200

0

x (m)

0 8.4 16.8 25.2 33.6 42.0

) ( C T

o

Δ

(Analytical solution)

1000

800

600

400

200

0

0 8.4 16.8 25.2 33.6 42.0

) ( C T

o

Δ

x (m)

o t=21.73 days

+ t=108.6 days

* t=217.2 days

x t=434.4 days

o t=21.73 days

+ t=108.6 days

* t=217.2 days

x t=434.4 days

Fig. 7.5 Comparison of analytical solutions with numerical solutions (Three-step solidiﬁcation)

elements to simulate the intruded magma in the solidiﬁcation direction) produce

similar numerical results. This indicates that, due to the precise consideration of the

magma solidiﬁcation interface, the present numerical algorithm is not sensitive to

the mesh density in the solidiﬁcation direction of the magma solidiﬁcation region.

It is noted that, for intruded magma dikes and sills of large aspect ratios, one

could have expected on dimensional considerations that the solidiﬁcation of the

intruded magma could be treated by means of a one-dimensional model in the solid-

iﬁcation direction and that such an approximation is asymptotically valid to the

leading order in the aspect ratio of the dike and sill. Such a one-dimensional (1D)

solidiﬁcation model could be interactively used with two-dimensional (2D) models

for simulating the thermal ﬁeld in the surrounding rocks. The use of the resulting

168 7 Simulating Thermal and Chemical Effects of Intruded Magma Solidiﬁcation Problems

hybrid 1D-2Dmodel may avoid the need to introduce equivalent volumetric heat and

mass source terms. However, since the 1D and 2D models are considered separately

and interactively, one must use the 1D model to determine the temperature and ther-

mal ﬂux boundary conditions on the solidiﬁcation interface between the rock and

intruded magma, and then apply such boundary conditions to the 2D model for sim-

ulating the thermal ﬁeld in the surrounding rocks. Since the magma solidiﬁcation is

time-dependent, the above-mentioned boundary conditions need to be determined

and applied in the hybrid 1D-2D model repeatedly for each time step during the

intruded magma solidiﬁcation. This certainly adds numerical modelling complexity

during the magma solidiﬁcation, because the previous boundary conditions must be

modiﬁed and the current ones must be applied to the 2D model. This complexity

has been avoided by using the proposed numerical algorithm in this study, since

the magma solidiﬁcation in the intruded magma region and heat transfer in the sur-

rounding rocks are considered simultaneously, rather than separately, in the ﬁnite

element analysis. Due to this obvious advantage in using the proposed model, there

is no need to develop the hybrid 1D-2D model for simulating the solidiﬁcation prob-

lems associated with intruded magma dikes and sills.

After the veriﬁcation of the proposed equivalent source algorithm for simulat-

ing solidiﬁcation effects of dike-like magma intrusion problems in porous rocks,

it has been applied to investigate the chemical effects of a dike-like magma intru-

sion/solidiﬁcation problemin the upper crust of the Earth. Figure 7.6 shows the ﬁnite

element mesh of the dike-like magma intrusion problem. The length and width of

the whole computational domain are 40.2 km and 10 km, the region of the intruded

dike-like magma (as indicated by black colour) is 0.2 km and 6 km in the horizontal

and vertical directions, respectively. The intruded magma region is modelled with

144 four-node quadrilateral ﬁnite elements, while the whole computational domain

is modelled with 1040 four-node quadrilateral ﬁnite elements in the transient ﬁnite

element analysis. The thermal properties of this application problem are exactly the

same as those of the previous benchmark magma solidiﬁcation problem. The diffu-

sivity of the released volatile ﬂuids is 2 ×10

–6

m

2

/s. The pressure of the intruded

magma is assumed to be 3000 bars. Since the width of the intruded magma for this

application problem is 100 times that for the previous benchmark magma solidiﬁca-

tion problem, the solidiﬁcation time of the intruded magma for this application prob-

lem is 10,000 times that for the previous benchmark magma solidiﬁcation problem,

x

0

y

Fig. 7.6 Finite element mesh of the dike-like magma intrusion problem

7.3 Veriﬁcation and Application of the Equivalent Source Algorithm 169

as indicated from Eq. (7.16). Therefore, the time step of 9.3826 ×10

9

s (which is

approximately equal to 300 years) is used in the transient ﬁnite element analysis

of the dike-like magma intrusion problem in the upper crust of the Earth. For this

application problem, the boundary condition is that the temperature at the top of

the computational model is 20

◦

C throughout the transient ﬁnite element analysis.

The initial conditions are as follows: the initial temperature of the intruded dike-like

magma is 1230

◦

C and the initial temperature at the rest (i.e., except for the intruded

magma) of the bottom of the computational model is 320

◦

C. The initial average

temperature of the rocks surrounding the intruded dike-like magma is 230

◦

C. The

initial concentration of the released volatile ﬂuids is assumed to be zero in the whole

( s t

11

10 8765 . 1 × = )

( s t

11

10 6673 . 4 × = )

( s t

11

10 3826 . 9 × = )

( s t

12

10 8765 . 1 × = )

Fig. 7.7 Temperature distributions of the dike-like magma intrusion problem at different time

instants

170 7 Simulating Thermal and Chemical Effects of Intruded Magma Solidiﬁcation Problems

computational domain. Under the above temperature and pressure conditions of the

intruded magma, the density of the released volatile ﬂuids is 374.7 kg/m

3

(Haar

et al. 1984), while the density of the albite melt is 2700 kg/m

3

. In addition, the mole

fraction of the released volatile ﬂuids from the albite melt can be determined as

0.5306 from Eqs. (7.10) and (7.11), which results in a mass source of 1.433 ×10

–8

kg/(m

3

× s) for the released volatile ﬂuids in the computation.

Figure 7.7 shows the temperature distributions of the dike-like magma intrusion

problem at four different time instants. It is clear that with increasing time, the total

temperature localization area generated by the intruded magma becomes larger and

(Along the cross-section of y = 3000 m)

x (km)

500

440

380

320

260

200

0 8.4 16.8 25.2 33.6 42.0

) ( C T

o

o s t

11

10 8765 . 1 × =

+ s t

11

10 6673 . 4 × =

*

s t

11

10 3826 . 9 × =

s t

12

10 8765 . 1 × = x

(Along the cross-section of x = 20,100 m)

y (km)

620

500

380

260

140

20

0 2.0 4.0 6.0 8.0 10.0

) ( C T

o

o s t

11

10 8765 . 1 × =

+ s t

11

10 6673 . 4 × =

*

s t

11

10 3826 . 9 × =

s t

12

10 8765 . 1 × = x

Fig. 7.8 Temperature distributions of the dike-like magma intrusion problem along two different

cross-sections

7.3 Veriﬁcation and Application of the Equivalent Source Algorithm 171

larger, but the maximum temperature generated by the intruded magma becomes

smaller and smaller. These phenomena can be clearly seen from Fig. 7.8, where the

temperature distributions of the dike-like magma intrusion problem are displayed

along two typical cross-sections of the computational model. For instance, the max-

imum temperature along the cross-section of x = 20,100 m is 514.8

◦

C, 444.4

◦

C,

394.4

◦

C and 354.9

◦

C for the time instants of 1.8765 ×10

11

s, 4.6673 ×10

11

s,

9.3826 ×10

11

s and 1.8765 ×10

12

s respectively. Although the total temperature

localization area generated by the intruded magma is limited in the top part of the

( s t

9

10 3826 . 9 × = )

( s t

10

10 3826 . 9 × = )

( s t

11

10 8765 . 1 × = )

( s t

11

10 6673 . 4 × = )

LEGEND

A 0.5000E–03 –

0.1500E–02 – B

0.2500E–02 – C

0.3500E–02 – D

0.4500E–02 – E

0.5500E–02 – F

0.6500E–02 – G

0.7500E–02 – H

0.8500E–02 – I

0.9500E–02 – J

A

A

A

A

B

C

E

D

C

B A

A B B C

LEGEND

A 0.5000E–03 –

0.1500E–02 – B

0.2500E–02 – C

0.3500E–02 – D

0.4500E–02 – E

0.5500E–02 – F

0.6500E–02 – G

0.7500E–02 – H

0.8500E–02 – I

0.9500E–02 – J

LEGEND

A 0.5000E–03 –

0.1500E–02 – B

0.2500E–02 – C

0.3500E–02 – D

0.4500E–02 – E

0.5500E–02 – F

0.6500E–02 – G

0.7500E–02 – H

0.8500E–02 – I

0.9500E–02 – J

LEGEND

A 0.5000E–03 –

0.1500E–02 – B

0.2500E–02 – C

0.3500E–02 – D

0.4500E–02 – E

0.5500E–02 – F

0.6500E–02 – G

0.7500E–02 – H

0.8500E–02 – I

0.9500E–02 – J

Fig. 7.9 Concentration distributions of volatile ﬂuids for the dike-like magma intrusion problem

at different time instants (Whole system view)

172 7 Simulating Thermal and Chemical Effects of Intruded Magma Solidiﬁcation Problems

computational model, it is large enough to cause convective pore-ﬂuid ﬂow within

the whole system (Zhao et al. 1997a), if the porous rocks are saturated by the pore-

ﬂuid in the upper crust of the Earth. Once this convective pore-ﬂuid ﬂow takes place,

temperature localization in the top part of the computational model can be signif-

icantly enhanced so that a favourable region for ore body formation and mineral-

ization can be created just above the intruded dike-like magma (Zhao et al. 1998a).

Figure 7.9 shows the concentration distributions of the volatile ﬂuids for the

whole system of the dike-like magma intrusion problem at four different times,

while Figs. 7.10 and 7.11 show the detailed concentration distributions of the

volatile ﬂuids for a zoomed-in part of the dike-like magma intrusion problem at the

same four different times. It is clear that with an increase in time, the total concen-

tration area of the volatile ﬂuids generated during the intruded magma solidiﬁcation

becomes larger and larger, but the maximum concentration of the volatile ﬂuids gen-

erated by the intruded magma becomes smaller and smaller. These phenomena can

be clearly seen from Figs. 7.10 and 7.11, where the concentration distributions of

( s t

9

10 3826 . 9 × = )

( s t

10

10 3826 . 9 × = )

LEGEND

A 0.5000E–03 –

0.1500E–02 – B

0.2500E–02 – C

0.3500E–02 – D

0.4500E–02 – E

0.5500E–02 – F

0.6500E–02 – G

0.7500E–02 – H

0.8500E–02 – I

0.9500E–02 – J

LEGEND

A 0.5000E–03 –

0.1500E–02 – B

0.2500E–02 – C

0.3500E–02 – D

0.4500E–02 – E

0.5500E–02 – F

0.6500E–02 – G

0.7500E–02 – H

0.8500E–02 – I

0.9500E–02 – J

A

A

B

D

E

G

F

E

C

C

B

A

C

Fig. 7.10 Concentration distributions of volatile ﬂuids for the dike-like magma intrusion problem

at different time instants (Zoomed-in view)

7.3 Veriﬁcation and Application of the Equivalent Source Algorithm 173

( s t

11

10 8765 . 1 × = )

( s t

11

10 6673 . 4 × = )

LEGEND

A 0.5000E–03 –

0.1500E–02 – B

0.2500E–02 – C

0.3500E–02 – D

0.4500E–02 – E

0.5500E–02 – F

0.6500E–02 – G

0.7500E–02 – H

0.8500E–02 – I

0.9500E–02 – J

LEGEND

A 0.5000E–03 –

0.1500E–02 – B

0.2500E–02 – C

0.3500E–02 – D

0.4500E–02 – E

0.5500E–02 – F

0.6500E–02 – G

0.7500E–02 – H

0.8500E–02 – I

0.9500E–02 – J

Fig. 7.11 Concentration distributions of volatile ﬂuids for the dike-like magma intrusion problem

at different time instants (Zoomed-in view)

the volatile ﬂuids during the dike-like magma solidiﬁcation are displayed within the

zoomed-in part of the computational model. Since the released volatile ﬂuids can

react chemically with the surrounding rocks (Zhao et al. 2001d), an aureole, which

can be clearly observed in these ﬁgures, is formed during the intruded magma solid-

iﬁcation. Although the total area of the aureole generated by the intruded magma is

limited in the computational model, it is large enough to cause ore body formation

and mineralization to take place within the system. This indicates that the released

volatile ﬂuids can create a favourable region for ore body formation and mineral-

ization to take place through some chemical reactions around the intruded dike-like

magma.

Chapter 8

The Particle Simulation Method for Dealing

with Spontaneous Crack Generation Problems

in Large-Scale Geological Systems

Cracking and fracturing are one class of major failure mechanisms in brittle and

semi-brittle materials. Crustal materials of the Earth can be largely considered as

brittle rocks, and so cracking and fracturing phenomena are ubiquitous. Cracks cre-

ated within the Earth’s crust often provide a very useful channel for mineral-bearing

ﬂuids to ﬂow, particularly from the deep crust into the shallow crust of the Earth.

If other conditions such as ﬂuid chemistry, mineralogy, temperature and pressure

are appropriate, ore body formation and mineralization can take place as a result of

such ﬂuid ﬂow. Because of the ever-increasing demand for mineral resources in the

contemporary world, exploration for new mineral resources has become one of the

highest priorities for many industrial countries. For this reason, extensive studies

(Garven and Freeze 1984, Yeh and Tripathi 1989, 1991, Steefel and Lasaga 1994,

Raffensperger and Garven 1995, Zhao et al. 1997a, Schafer et al. 1998a, b, Zhao et

al. 1998a, Xu et al. 1999, Zhao et al. 2000b, Schaubs and Zhao 2002, Zhao et al.

2002a, 2003e, 2005a) have been conducted to understand the detailed physical and

chemical processes that control ore body formation and mineralization within the

upper crust of the Earth. Thus, the numerical simulation of spontaneous crack gen-

eration in brittle rocks within the upper crust of the Earth has become an important

research topic in the ﬁeld of computational geoscience.

The numerical simulation of crack initiation has existed from the inception

and development of numerical fracture mechanics. This was the direct outcome

of combining conventional fracture mechanics with numerical methods such as

the ﬁnite element and the boundary element methods. Due to the increased capa-

bility of numerical fracture mechanics for considering complicated geometry and

boundary conditions, it signiﬁcantly extends the applicability of conventional frac-

ture mechanics to a wide range of practical problems in civil and geotechnical

engineering ﬁelds. In the study of numerical simulation of crack initiation and

evolution, it is usually assumed that a crack initiates when the value of a local

principal stress attains the tensile strength of the brittle material. Once a crack is

initiated, propagation of the crack, including the propagation direction and incre-

mental growing length of the crack, can be determined by crack propagation the-

ories established in fracture mechanics. Although there are at least three kinds

of mixed-mode crack propagation theories available, namely a theory based on

C. Zhao et al., Fundamentals of Computational Geoscience,

Lecture Notes in Earth Sciences 122, DOI 10.1007/978-3-540-89743-9 8,

C

Springer-Verlag Berlin Heidelberg 2009

175

176 8 Spontaneous Crack Generation Problems in Large-Scale Geological Systems

propagation driven by the maximum circumferential tensile stress (Erdogan and

Sih 1963), a theory based on propagation driven by strain energy density near

the crack tip (Sih and Macdonald 1974) and a theory based on crack propaga-

tion driven by maximum energy release rate (Rice 1968, Hellen 1975, Lorenzi

1985, Li et al. 1985), the maximum circumferential tensile propagation theory has

been widely used in the ﬁnite element analysis of crack initiation and propaga-

tion problems because it is not only physically meaningful, but it is also easily

implemented in a ﬁnite element code. It has been widely demonstrated that, for

the ﬁnite element simulation of a two-dimensional fracture mechanics problem,

the triangular quarter-point element, which is formed by collapsing one side of a

quadrilateral 8-noded isoparametric element, results in improved numerical results

for stress intensity factors near a crack tip (Barsoum 1976, 1977). Similarly, for

the ﬁnite element modelling of a three-dimensional fracture mechanics problem,

the prismatic quarter-point element, which is formed by collapsing one face of

a cubic 20-noded isoparametric element, can also lead to much better numerical

results for stress intensity factors near a crack tip (Barsoum 1976, 1977, Ingraf-

fea and Manu 1980). This kind of quarter-point element can be used to simulate

the 1

/

√

r singularity in elastic fracture mechanics by simply assigning the same

displacement at the nodes located on the collapsed side or face of the element,

while it can be also used to simulate the 1

/

r singularity in perfect plasticity by sim-

ply allowing different displacements at the nodes located on the collapsed side or

face of the element. In order to simulate appropriately displacement discontinuities

around a crack, automatic meshing and re-meshing algorithms (Zienkiewicz and

Zhu 1991, Lee and Bathe 1994, Khoei and Lewis 1999, Kwak et al. 2002, Bouchard

et al. 2003) have been developed in recent years. However, such ﬁnite element

methods with automatic meshing and re-meshing algorithms have been applied

for solving crack generation and propagation problems in systems of only a few

cracks.

For the purpose of removing this limitation associated with the conventional

numerical method (which is usually based on continuum mechanics) in simulat-

ing a large number of spontaneously generated cracks, particle simulation meth-

ods, such as the distinct element method developed as a particle ﬂow code (Cundall

and Strack 1979, Cundall 2001, Itasca Consulting Group, inc. 1999, Potyondy and

Cundall 2004), provides a very useful tool to deal with this particular kind of prob-

lem. Since displacement discontinuities at a contact between two particles can be

readily considered in these methods, the formulation based on discrete particle sim-

ulation is conceptually simpler than that based on continuum mechanics, because

crack generation at a contact between two particles is a natural part of the particle

simulation process.

Even though the particle simulation method was initially developed for solv-

ing soil/rock mechanics, geotechnical and other engineering problems (Cundall and

Strack 1979, Bardet and Proubet 1992, Thomton et al. 1999, Tomas et al. 1999,

Salman and Gorham 2000, Klerck et al. 2004, Owen et al. 2004, McBride et al.

2004 and Schubert et al. 2005, Zhao et al. 2006f), it has been used to deal with a

large number of geological and geophysical problems in both two and three dimen-

8 Spontaneous Crack Generation Problems in Large-Scale Geological Systems 177

sions (Saltzer and Pollard 1992, Antonellini and Pollard 1995, Donze et al. 1996,

Scott 1996, Strayer and Huddleston 1997, Camborde et al. 2000, Iwashita and Oda

2000, Burbidge and Braun 2002, Strayer and Suppe 2002, Finch et al. 2003, 2004,

Imber et al. 2004, Zhao et al. 2007b, c, d, Zhao et al. 2008g). Although the particle

simulation method has been successfully used to solve these large-scale geologi-

cal problems, little work, if any, has been reported on using the particle simulation

method to deal with spontaneous crack generation problems in the upper crust of

the Earth. Nevertheless, the particle simulation method has been developed to sim-

ulate microscopic crack generation in small-scale laboratory specimens and mining

sites (Itasca Consulting Group, inc. 1999, Potyondy and Cundall 2004). Since both

the time-scale and the length-scale are quite different between laboratory speci-

mens and geological systems, it is necessary to deal with an upscale issue when the

particle simulation method is applied to solve spontaneous crack generation prob-

lems in the upper crust of the Earth. Due to the relative slowness of some geolog-

ical processes, many geological systems can be treated as quasi-static ones (that

is, inertia is neglected), at least from the mathematical point of view. Because the

mechanical response of a quasi-static system is theoretically independent of time,

the time-scale issue can be eliminated in the particle simulation of a quasi-static

system. For this reason, this chapter is restricted to deal with the particle simulation

of spontaneous crack generation problems within large-scale quasi-static geological

systems.

It is however computationally prohibitive to simulate a whole geological system

using real physical particles of a microscopic length-scale, even though with modern

computer capability. To overcome this difﬁculty, the following three approaches are

often used in dealing with the particle simulation of geological length-scale prob-

lems. In the ﬁrst approach, a geological length-scale problem, which is usually of

a kilometer-scale, is scaled down to a similar problem of a small length-scale (i.e.

a meter-scale or a centimeter-scale) and then particles of small length-scale (e.g. a

microscopic length-scale) are used to simulate this small length-scale geological

problem (e.g. Antonellini and Pollard 1995, Imber et al. 2004, Schopfer et al. 2006).

In the second approach, the geological length-scale problem is directly simulated

using particles with a relative large length-scale (e.g. Strayer and Huddleston 1997,

Burbidge and Braun 2002, Strayer and Suppe 2002, Finch et al. 2003, 2004). In

this case, the particles used in the simulation can be considered as the representa-

tion of a large rock block. Although detailed microscopic deformation cannot be

simulated using the second approach, the macroscopic deformation pattern can be

reasonably simulated in this approach (e.g. Strayer and Huddleston 1997, Burbidge

and Braun 2002, Strayer and Suppe 2002, Finch et al. 2003, 2004). This implies

that if the macroscopic deformation process of a geological length-scale system is

of interest, then the second approach can produce useful simulation results. This

is particularly true for understanding the controlling process of ore body forma-

tion and mineralization, in which the macroscopic length-scale geological struc-

tures are always of special interest. In the third approach, the combined use of both

a block-mechanics-based particle simulation method and a continuum-mechanics-

based method, such as the ﬁnite element method (Zienkiewicz 1977, Lewis and

178 8 Spontaneous Crack Generation Problems in Large-Scale Geological Systems

Schreﬂer 1998) or the ﬁnite difference method, are used to simulate the whole

geological system (Potyondy and Cundall 2004, Suiker and Fleck 2004, Fleck and

Willis 2004). In this approach, the continuous deformation range of a system is

simulated using the continuum-mechanics-based method, and the discontinuous

deformation range of the system is simulated using the block-mechanics-based par-

ticle simulation method. Since large element sizes can be used to simulate the con-

tinuous deformation range for a quasi-static geological system, the third approach

is computationally more efﬁcient than the ﬁrst and second approaches in dealing

with large systems. However, the computational difﬁculty associated with the third

approach is that an adaptive interface between the continuous deformation range

and the discontinuous deformation range must be appropriately developed. This

means that if the discontinuous deformation range is obvious a priori for some kind

of geological problem, then the third approach is computationally very efﬁcient.

Otherwise, the efﬁciency of using the third approach might be greatly reduced due

to the ambiguity in deﬁning interfaces between the continuous and the discontinu-

ous deformation ranges. For this reason, the ﬁrst and second approaches are com-

monly used to simulate large-scale geological systems using the particle simulation

method.

Because of the wide use of both the ﬁrst and the second approaches, the fol-

lowing scientiﬁc questions are inevitably posed: Are two particle simulation results

obtained from both the ﬁrst and the second approaches consistent with each other

for the same geological problem? If the two particle simulation results are consis-

tent, then what is the intrinsic relationship between the two similar particle mod-

els used in the ﬁrst approach and the second approach? To the best knowledge

of the authors, these questions remain unanswered. Therefore, we will develop an

upscale theory associated with the particle simulation of two-dimensional quasi-

static geological systems at different length-scales to clearly answer these two sci-

entiﬁc questions. The present upscale theory is of signiﬁcant theoretical value in

the particle simulation of two-dimensional systems, at least from the following

two points of view. (1) If the mechanical response of a particle model of a small

length-scale is used to investigate indirectly that of a large length-scale, then the

present upscale theory provides the necessary conditions that the particle model

of the small length-scale needs to satisfy so that similarity between the mechani-

cal responses of the two different length-scale particle models can be maintained.

(2) If a particle model of a large length-scale is used to investigate directly the

mechanical response of the model, then the present upscale theory can be used

to determine the necessary particle-scale mechanical properties from the macro-

scopic mechanical properties that are obtained from either a laboratory test or an

in-situ measurement. Because the particle simulation method has been used to solve

many kinds of scientiﬁc and engineering problems, the present upscale theory can

be directly used to extend the application range of the particle simulation results

for geometrically-similar problems without a need to conduct another particle sim-

ulation. This means that the present upscale theory compliments existing particle

simulation method, and is applicable for practical applications in many scientiﬁc

and engineering ﬁelds.

8.1 Basic Formulations of the Particle Simulation Method 179

8.1 Basic Formulations of the Particle Simulation Method

The basic idea behind the particle simulation method is that either a granular mate-

rial or a solid material can be simulated using an assembly of particles. In the case

of simulating granular materials, there is no cementation between two adjacent par-

ticles so that the particles are not bonded together in the assembly. However, in the

case of simulating solid materials, particles are bonded together through cementation.

Although particles may have different shapes and sizes, they are assumed to be rigid.

Thus, the motion of a particle can be described using the motion of its mass center. A

small overlapping between two particles is allowed so that deformation of the particle

assembly can be simulated. The magnitude of the overlapping depends on both the

contact force and the stiffness of a contact between any two particles. This means

that a relationship between contact force and displacement needs to be used to cal-

culate the contact force at a contact between two particles. In the case of simulating

solid materials such as brittle rocks, the strength of the material can be simulated by

using the strength of the bond at a contact between two particles. Once the contact

force reaches or exceeds the strength of the bond at a contact between two particles,

the bond is broken so that a microscopic crack is created to represent the failure of

the material at this particular bond. This consideration is very useful for explicitly

simulating the initiation of spontaneous and random microscopic cracks in a brittle

material, because there is no need to artiﬁcially describe any microscopic ﬂaws and

cracks in the beginning of a numerical simulation. Although the above idea was ini-

tially proposed in the discrete element method (Cundall and Strack 1979, Cundall

2001, Itasca Consulting Group, inc. 1999), it has been recently enhanced and built

into a two-dimensional Particle Flow Code. In this regard, the particle simulation

method can be viewed as a particular kind of discrete element method.

Considering a particle (i.e. particle α) shown in Fig. 8.1, the position of the par-

ticle is described using its x and y coordinates in the coordinate system shown.

According to Newton’s second law, the motion of the particle can be represented

using the motion of its mass center as follows:

αz

M

αx

F

αy

F

R

α

o

x

y

α

Fig. 8.1 Deﬁnition of the motion of a particle

180 8 Spontaneous Crack Generation Problems in Large-Scale Geological Systems

F

αx

= m

α

d

2

x

α

dt

2

, (8.1)

F

αy

= m

α

d

2

y

α

dt

2

, (8.2)

M

αz

= I

α

d

2

θ

αz

dt

2

, (8.3)

where F

αx

, F

αy

and M

αz

are the total translational force components and rotational

moment exerted on the mass center of particle α; m

α

and I

α

are the mass and princi-

pal moment of inertia with respect to the z axis that is perpendicular to the x-y plane;

x

α

and y

α

are the horizontal and vertical coordinates expressing the position of par-

ticle α; θ

αz

is the rotation angle of particle α with respect to the principal rotational

axis of the particle.

For a two-dimensional disk-shaped particle, the corresponding principal moment

of inertia is expressed as follows:

I

α

=

1

2

m

α

R

2

α

, (8.4)

where R

α

is the radius of particle α.

The translational forces and rotational moment expressed in the above equations

can be calculated by adding all the forces and moments exerted on the particle. It

is noted that a particle can contact several particles at the same time so that there

are several contact forces exerted on the particle. For a particular contact (i.e. con-

tact C) between two particles (i.e. particles A and B) shown in Fig. 8.2, the normal

component of the contact force can be calculated using the following formula:

q

o

x

y

n

→

Contact

plane

R

B

R

A

C

B

A

→

Fig. 8.2 Deﬁnition of the

contact between two

particles

8.1 Basic Formulations of the Particle Simulation Method 181

F

n

=

¸

k

n

u

n

(u

n

≥ u

n

)

0 (u

n

< u

n

)

(Before the normal contact bond is broken),

(8.5)

F

n

=

¸

k

n

u

n

(u

n

≥ 0)

0 (u

n

< 0)

(After the normal contact bond is broken),

(8.6)

where k

n

is the stiffness of contact C; u

n

is the normal displacement at contact C; u

n

is the critical normal displacement in correspondence with the normal contact bond

breakage at contact C; F

n

is the normal contact force at contact C. The contact force

at contact C is assumed to be exerted from particle A on particle B.

u

n

= R

A

+ R

B

−

(x

B

− x

A

)

2

+(y

B

− y

A

)

2

, (8.7)

where R

A

and R

B

are the radii of particles A and B; x

A

and y

A

are the horizontal

and vertical coordinates expressing the position of particle A; x

B

and y

B

are the

horizontal and vertical coordinates expressing the position of particle B.

Using the deﬁnition expressed in Fig. 8.2, the position of contact C can be

described as

x

C

= x

A

+

R

A

−

1

2

u

n

n

x

, (8.8)

y

C

= y

A

+

R

A

−

1

2

u

n

n

y

, (8.9)

where x

C

and y

C

are the horizontal and vertical coordinates expressing the position

of contact C; n

x

and n

y

are the direction cosines of the normal vector with respect

to the horizontal and vertical axes respectively.

The tangential component of the contact force at contact C can be calculated in

an incremental manner as follows:

ΔF

s

=

¸

−k

s

Δu

s

= −k

s

(V

S

Δt ) (u

n

≥ u

n

) (before the normal

0 (u

n

< u

n

) contact bond is broken),

(8.10)

ΔF

s

=

¸

−k

s

Δu

s

= −k

s

(V

S

Δt ) (u

n

≥ 0) (after the normal

0 (u

n

< 0) contact bond is broken),

(8.11)

where k

s

is the tangential stiffness of contact C; Δu

s

is the incremental tangential

displacement at contact C; ΔF

s

is the corresponding tangential component of the

contact force; V

S

is the tangential shear velocity at contact C; Δt is the time step in

the numerical simulation.

182 8 Spontaneous Crack Generation Problems in Large-Scale Geological Systems

q

→

S

V

n

→

B

A

o

Az

M

Ax

F

Ay

F

A

R

x

y

Bz

M

Bx

F

By

F

B

R

Fig. 8.3 Deﬁnition of the

shear velocity at a contact

between two particles

As shown in Fig. 8.3, the magnitude of the tangential shear velocity, V

S

, at con-

tact C can be determined using the relative motion of particles A and B.

V

S

=−

dx

B

dt

−

dx

A

dt

n

y

+

dy

B

dt

−

dy

A

dt

n

x

−ω

Bz

(x

B

− x

C

)

2

+(y

B

− y

C

)

2

−ω

Az

(x

A

− x

C

)

2

+(y

A

− y

C

)

2

,

(8.12)

where ω

Az

and ω

Bz

are the rotational angular speeds of particles A and B with respect

to their rotational axes, which are parallel to the z axis and passing through the

corresponding mass centers of each of the two particles respectively.

It is noted that the ﬁrst two terms in Eq. (8.12) represent the contributions of the

relative translational motion to the relative shear velocity between the two particles,

whereas the last two terms denote the contributions of the relative rotational motion

to the relative shear velocity between the two particles.

For any given time instant, t, the tangential component of the contact force at

contact C can be calculated by adding the contact force increment expressed in

Eq. (8.11) into the tangential component at the previous time, t −Δt .

F

s

t

= F

s

t −Δt

−k

s

(V

S

Δt ) ≤ μF

n

, (8.13)

where F

s

t

and F

s

t −Δt

are the tangential components of the contact force at t and

t −Δt respectively; μ is the friction coefﬁcient at contact C. It needs to be pointed

out that Eq. (8.13) holds true only when the normal component of the contact force

is greater than zero.

The normal and tangential components of the contact force at contact C can

be straightforwardly decomposed into the horizontal and vertical components as

follows:

8.1 Basic Formulations of the Particle Simulation Method 183

F

Cx

= F

n

n

x

+ F

s

t

q

x

, (8.14)

F

Cy

= F

n

n

y

+ F

s

t

q

y

, (8.15)

where F

Cx

and F

Cy

are the horizontal and vertical components of the contact force

at contact C; q

x

and q

y

are the direction cosines of the tangential vector at contact

C with respect to the horizontal and vertical axes respectively.

Consequently, the translational and rotational forces exerted on particles A and B

due to their contact at point C can be calculated using the following formulas:

F

Ax

= −F

Cx

, F

Ay

= −F

Cy

, (8.16)

F

Bx

= F

Cx

, F

By

= F

Cy

(8.17)

M

Az

= −[(x

C

−x

A

)F

Cx

+(x

C

−x

A

)F

Cy

+(y

C

−y

A

)F

Cx

+(y

C

−y

A

)F

Cy

], (8.18)

M

Bz

= (x

C

− x

B

)F

Cx

+(x

C

− x

B

)F

Cy

+(y

C

− y

B

)F

Cx

+(y

C

− y

B

)F

Cy

, (8.19)

where F

Ax

, F

Ay

and M

Az

are the translational force components and rotational

moment exerted on the mass center of particle A; F

Bx

, F

By

and M

Bz

are the trans-

lational force components and rotational moment exerted on the mass center of

particle B.

Since a particle may have contacts with several particles, it is necessary to search

the number of contacts for the particular particle under consideration. Therefore, the

total translational forces and rotational moment exerted on a particle can be calcu-

lated by adding the contributions of all the contacts to the translational forces and

rotational moments exerted on the particle. After the total translational forces and

rotational moment are calculated in a particle by particle manner, the central ﬁnite

difference method is used to solve the equations of motion expressed by Eqs. (8.1),

(8.2), and (8.3) for each of the particles in the simulation system, so that new dis-

placements can be determined and the position of each particle can be updated. As a

result, a solution loop is formed for each time step in the particle simulation method.

This solution loop is comprised of the following four sub-steps: (1) From the posi-

tion of particles at the beginning of a calculation time step, Eqs. (8.16), (8.17) and

(8.19) are used to calculate the contributions of a contact between a particle and

each of its surrounding particles to the translational force components and rota-

tional moment exerted on the mass center of both the particle and the surrounding

particle; (2) By adding the contributions of all the contacts of a particle to the trans-

lational forces and rotational moments exerted on the particle, the total translational

forces and rotational moment exerted on the particle are calculated. (3) Sub-steps

(1) and (2) are repeated for all the particles so that the total translational forces

184 8 Spontaneous Crack Generation Problems in Large-Scale Geological Systems

and rotational moment exerted on every particle in the simulation are calculated.

(4) Using the central ﬁnite difference method, the equations of motion expressed

by Eqs. (8.1), (8.2) and (8.3) are solved for each of the particles in the simulation

system, so that new displacements can be determined and the position of every par-

ticle can be updated at the end of the calculation time step. The above-mentioned

solution loop is repeated for each calculation time step until the ﬁnal stage of the

simulation is reached.

8.2 Some Numerical Simulation Issues Associated

with the Particle Simulation Method

Although the particle simulation method such as the distinct element method was

developed more than two decades ago (Cundall and Strack 1979), some numeri-

cal issues associated with it may need to be further addressed. These include: (1) an

issue caused by the difference between an element used in the ﬁnite element method

and a particle used in the distinct element method; (2) an issue resulting from using

the explicit dynamic relaxation method to solve a quasi-static problem; and (3) an

issue stemming from an inappropriate loading procedure used in the particle simula-

tion method. Although some aspects of these numerical simulation issues have been

brieﬂy discussed (Cundall and Strack 1979, Cundall 2001, Itasca Consulting Group,

inc. 1999, Potyondy and Cundall 2004), we will discuss them in greater detail so

that their impact on the particle simulation results of spontaneous crack generation

problems within large-scale quasi-static systems can be thoroughly understood.

8.2.1 Numerical Simulation Issue Caused by the Difference

between an Element and a Particle

In order to investigate the numerical simulation issue associated with the differ-

ence between an element used in the ﬁnite element method and a particle used in

the distinct element method, we need to understand how an element and a par-

ticle interact with their neighbors. Figure 8.4 shows the comparison of a typical

four-node element used in the ﬁnite element method with a typical particle used in

the distinct element method. In the ﬁnite element method, the degree-of-freedom

is represented by the nodal points of the element, while in the distinct element

one, the degree-of-freedom is represented by the mass center of the particle. In this

regard, the particle may be viewed as a rigid element of only one node. With a two-

dimensional material deformation problem taken as an example, the displacement

along the common side between two elements is continuous, implying that there is

no overlap between any two elements in the conventional ﬁnite element method.

However, particle overlap is allowed in the distinct element method. Since the ele-

ment is deformable, the mechanical properties calculated at the nodal point of an ele-

ment are dependent on the macroscopic mechanical properties of the element so that

8.2 Some Numerical Simulation Issues Associated with the Particle Simulation Method 185

(A) A four-node element in the mesh (B) A particle with its neighbors

e

p

Fig. 8.4 Comparison of a four-node element with a particle

the nodal mechanical property is called the consistent mechanical property of the

element. On the other hand, a particle used in the distinct element method is assumed

to be rigid and therefore, the mechanical property at a contact between the particle

and its neighboring particle is called the lumped mechanical property. Due to this

difference, the consistent mechanical properties (i.e. stiffness matrix) of an element

are directly calculated from the macroscopic mechanical properties and constitutive

law of the element material. This means that once the macroscopic mechanical prop-

erties and constitutive law of the element material are available from a laboratory

test or a ﬁeld measurement, the ﬁnite element analysis of a deformation problem

can be straightforwardly carried out using these macroscopic mechanical proper-

ties and constitutive laws of the material. On the contrary, because the particle-scale

mechanical properties of materials, such as particle stiffness and bond strength, are

used in the distinct element method but are not known a priori, it is important

to deduce these particle-scale mechanical properties of materials from the related

macroscopic ones measured from both laboratory and ﬁeld experiments. This indi-

cates that an inverse problem needs to be solved through the numerical simulation

of a particle system. Thus, the numerical simulation question introduced by the dif-

ference between an element used in the continuum-mechanics-based ﬁnite element

method and a particle used in the discrete-block-mechanics-based distinct element

method is how to determine particle-scale mechanical properties from macroscopic

mechanical properties available from both laboratory and ﬁeld experiments. Clearly,

the problem associated with this numerical simulation issue cannot be effectively

solved unless the required particle-scale mechanical properties of particles can be

directly determined from laboratory tests in the future.

As an expedient measure, primitive trial-and-error approaches can be used to

solve any inverse problems. For the above-mentioned inverse problem, input param-

eters are the macroscopic mechanical properties of materials, while the particle-

scale mechanical properties of materials, such as the particle stiffness and bond

strength, are unknown variables and therefore, need to be determined. Due to the

difﬁculty in directly solving this inverse problem, it is solved indirectly using a trial-

and-error approach, in which a set of particle-scale mechanical properties of mate-

rials are assumed so that the resulting macroscopic mechanical properties can be

186 8 Spontaneous Crack Generation Problems in Large-Scale Geological Systems

determined from the mechanical response of the particle model having this particu-

lar set of particle-scale mechanical properties. If the resulting macroscopic mechani-

cal properties are different from what we expected, then another set of particle-scale

mechanical properties of materials are used in the particle model. This trial-and-

error process needs to be continued until a set of particle-scale mechanical prop-

erties of materials can produce the expected macroscopic mechanical properties.

In geological practice, a kilometer-length-scale specimen is often used to conduct a

biaxial compression test and to measure the related macroscopic mechanical proper-

ties, such as the elastic modulus and material strength, fromthe mechanical response

of the particle model with an assumed set of particle-scale mechanical properties of

rocks. However, if some mechanical properties are independent of particle size or

other size-dependent mechanical properties can be determined from an appropriate

upscale rule, then the expected particle-scale mechanical properties of materials to

be used in a particle model can be determined without a need to conduct the afore-

mentioned trial-and-error exercise.

8.2.2 Numerical Simulation Issue Arising from Using the Explicit

Dynamic Relaxation Method to Solve a Quasi-Static Problem

For the purpose of demonstrating the numerical simulation issue resulting from

using the explicit dynamic relaxation method to solve a quasi-static problem, it is

helpful to explain brieﬂy how the ﬁnite element method is used to solve the same

kind of problem. For the sake of simplicity, a quasi-static elastic problem is used

to demonstrate the issue. Since the ﬁnite element method is based on continuum

mechanics, the governing equations of a two-dimensional quasi-static elastic prob-

lem in an isotropic and homogeneous material can be expressed as follows:

∂σ

x

∂x

+

∂τ

yx

∂y

= f

x

, (8.20)

∂τ

xy

∂x

+

∂σ

y

∂y

= f

y

, (8.21)

σ

x

=

E(1 −ν)

(1 −2ν)(1 +ν)

ε

x

+

ν

1 −ν

ε

y

, (8.22)

σ

y

=

E(1 −ν)

(1 −2ν)(1 +ν)

ν

1 −ν

ε

x

+ε

y

, (8.23)

τ

xy

= τ

yx

= 2Gγ

xy

, (8.24)

8.2 Some Numerical Simulation Issues Associated with the Particle Simulation Method 187

ε

x

=

∂u

s

∂x

, ε

y

=

∂v

s

∂y

, γ

xy

=

1

2

∂u

s

∂y

+

∂v

s

∂x

, (8.25)

where σ

x

and σ

y

are normal stresses of the solid matrix in the x and y directions; ε

x

and ε

y

are the normal strains of the solid matrix in relation to σ

x

and σ

y

; τ

xy

and γ

xy

are the shear stress and shear strain of the solid matrix; u

s

and v

s

are the horizontal

and vertical displacements of the solid matrix; E and G are the elastic and shear

modulus respectively; v is the Poisson ratio of the solid matrix; f

x

and f

y

are the

body forces in the x and y directions;.

Note that Eqs. (8.20) and (8.21) represent the equilibrium equations, whereas

Eqs. (8.22), (8.23), (8.24) and (8.25) are the constitutive equations and strain-

displacement relationship equations, respectively.

By using either the variational principle or the Galerkin method (Zienkiewicz

1977, Lewis and Schreﬂer 1998), the above-mentioned equations can be expressed

in the ﬁnite element form:

[K]

e

{u}

e

= {F}

e

, (8.26)

where [K]

e

is the stiffness matrix of an element; {u}

e

and {F}

e

are the displacement

and force vectors of the element.

In the ﬁnite element method, the quasi-static equilibrium problem is solved in

a global (i.e. system) manner. This means that the matrices and vectors of all the

elements in the system need to be assembled together to result in the following

global equation:

[K]{u} = {F}, (8.27)

where [K] is the global stiffness matrix of the system; {u} and {F} are the global

displacement and force vectors of the system.

Similarly, in the distinct element method, the quasi-static equilibrium equation

of a particle is of the following form:

[K]

p

{u}

p

= {F}

p

, (8.28)

where [K]

p

is the stiffness matrix of a particle; {u}

p

and {F}

p

are the relative dis-

placement and force vectors of the particle.

In order to reduce signiﬁcantly the requirement for computer storage and mem-

ory, the distinct element method solves the quasi-static equilibrium equation at

the particle level, rather than at the system level. This requires that a quasi-static

problem be turned into a ﬁctitious dynamic problem so that the explicit dynamic

relaxation method can be used to obtain the quasi-static solution from solving the

following ﬁctitious dynamic equation:

[M]

p

{¨ u}

p

+[K]

p

{u}

p

= {F}

p

, (8.29)

188 8 Spontaneous Crack Generation Problems in Large-Scale Geological Systems

where [M]

p

is the lumped mass matrix of the particle and {¨ u}

p

is the acceleration

vector of the particle.

A major difference between solving the global quasi-static equilibrium problem

(i.e. Eq. (4)) in the ﬁnite element analysis and solving the ﬁctitious dynamic problem

(i.e. Eq. (8.29)) in the particle simulation is that the numerical solution to Eq. (8.26)

is unconditionally stable so that it can be solved using the Gaussian elimination

method and the like, while the numerical solution to Eq. (8.29) is conditionally

stable if an explicit solver is used. For this reason, the critical time step, which is

required to result in a stable solution for the ﬁctitious dynamic problem, can be

expressed as follows (Itasca Consulting Group, inc. 1999):

Δt

critical

=

m

k

, (8.30)

where m is the mass of a particle and k is the stiffness between two particles.

It is immediately noted that since the value of the mass of a particle is usually

much smaller than that of the stiffness of the particle, the critical time step deter-

mined from Eq. (8.30) is considerably smaller than one. This indicates that for a

slow geological process of more than a few years, it may take too long to obtain

a particle simulation solution. To overcome this difﬁculty, the scaled mass is often

used in the distinct element method (Itasca Consulting Group, inc. 1999) so that the

critical time step can be increased to unity or any large number if needed. How-

ever, since the scaled mass, namely the ﬁctitious mass, is used, the time used in the

distinct element method is ﬁctitious, rather than physical. In order to remove pos-

sible chaotic behavior that may be caused by the use of arbitrarily-scaled masses,

ﬁctitious damping is also added to the particles used in a distinct element simulation

(Itasca Consulting Group, inc. 1999). Thus, the numerical simulation issue resulting

from using the explicit dynamic relaxation method to solve a quasi-static problem is

that the time used in the distinct element method is ﬁctitious, rather than physical.

If the mechanical response of a quasi-static system is elastic, then the solution to

the corresponding quasi-static problem is unique and independent of the deforma-

tion path of the system. In this case, the explicit dynamic relaxation method is valid

so that the elastic equilibrium solution can be obtained from the particle simulation

using the distinct element method. However, if any failure takes place in a quasi-

static system, then the quasi-static system behaves nonlinearly so that the solution

to the post-failure quasi-static problem is not unique and therefore, becomes depen-

dent on the deformation path of the system. Due to the use of both the ﬁctitious

time and the ﬁctitious scaled mass, the physical deformation path of a system can-

not be simulated using the explicit dynamic relaxation method. This implies that

the post-failure particle simulation result obtained from using the explicit dynamic

relaxation method may be problematic, at least from the rigorously scientiﬁc point

of view. Nevertheless, if one is interested in the phenomenological simulation of the

post-failure behavior of a quasi-static system, a combination of the distinct element

method and the explicit dynamic relaxation method may be used to produce some

useful simulation results in the engineering and geology ﬁelds (Cundall and Strack

8.2 Some Numerical Simulation Issues Associated with the Particle Simulation Method 189

1979, Bardet and Proubet 1992, Saltzer and Pollard 1992, Antonellini and Pollard

1995, Donze et al. 1996, Scott 1996, Strayer and Huddleston 1997, Camborde et al.

2000, Iwashita and Oda 2000, Burbidge and Braun 2002, Strayer and Suppe 2002,

Finch et al. 2003, 2004, Imber et al. 2004). In this case, it is strongly recommended

that a particle-size sensitivity analysis of at least two different models, which have

the same geometry but different smallest particle sizes, be carried out to conﬁrm the

particle simulation result of a large-scale quasi-static geological system.

8.2.3 Numerical Simulation Issue Stemming from the Loading

Procedure Used in the Particle Simulation Method

The distinct element method is based on the idea that the time step used in the

simulation is chosen so small that force, displacement, velocity and acceleration

cannot propagate from any particle farther than its immediate neighbors during a

single time step (Potyondy and Cundall 2004). The servo-control technique (Itasca

Consulting Group, inc. 1999) is often used to apply the equivalent velocity or dis-

placement to the loading boundary of the particle model. This will pose an important

scientiﬁc question: Is the mechanical response of a particle model dependent on the

loading procedure that is used to apply “loads” at the loading boundary of the par-

ticle model? If the mechanical response of a particle model is independent of the

loading procedure, then this issue can be neglected when we extend the applica-

tion range of the particle simulation method from a small-scale laboratory test to a

large-scale geological problem.

In order to answer this scientiﬁc question associated with the distinct element

method that is used in the PFC2D (i.e. Particle Flow Code in Two Dimensions), it

is necessary to investigate how a “load” is propagated within a particle system. For

the purpose of illustrating the “load” propagation mechanism, a one-dimensional

idealized model of ten particles of the same mass is considered in Fig. 8.5. The

“load” can be either a directly-applied force or an indirectly-applied force due to

a constant velocity in this idealized conceptual model. The question that needs to

be highlighted here is that when a “load” is applied to a particle system, what is

the appropriate time to record the correct response of the whole system due to this

“load”? This issue is important due to the fact that particle-scale material properties

of a particle are employed in a particle model and that a biaxial compression test is

often used to determine the macroscopic material properties of the particle model.

If the normal stiffness coefﬁcient between any two particles has the same value (i.e.

k

i

= k (i = 1, 2, 3, . . . , 10)) and the time step is equal to the critical time step

(Itasca Consulting Group, inc. 1999) in the particle model, then the displacement of

particle 1 (i.e. the particle with “load” P) is P

/

k at the end of the ﬁrst time step (i.e.

t = 1Δt , where Δt is the time step). The reason for this is that in the distinct element

method, the “load” can only propagate from a particle to its immediate neighboring

particles within a time step. Thus, during the ﬁrst time step, the other nine particles

in the right part of the model are still kept at rest. This is equivalent to applying a

190 8 Spontaneous Crack Generation Problems in Large-Scale Geological Systems

P

P

P

P

P

P

P

P

P

k

1

k

1

k

2

k

1

k

2

k

3

k

1

k

2

k

3

k

4

k

4

k

5

k

6

k

7

k

8

k

9

k

1

k

2

k

3

k

4

k

5

k

1

k

2

k

3

k

6

k

4

k

5

k

1

k

2

k

3

k

7

k

6

k

4

k

5

k

1

k

2

k

3

k

8

k

7

k

6

k

4

k

5

k

1

k

2

k

3

k

9

k

10

k

8

k

7

k

6

k

4

k

5

k

1

k

2

k

3

P

( t) t Δ =1

( t) t Δ = 2

( t) t Δ = 3

( t) t Δ = 4

( t) t Δ = 5

( t) t Δ = 6

( t) t Δ = 7

( t) t Δ = 8

( t) t Δ = 9

( t) t Δ =10

Fig. 8.5 Force propagation in a ten-particle system

ﬁxed boundary condition at the right end of spring 1, as shown in Fig. 8.5. Similar

considerations can be made for the consecutive time steps (see Fig. 8.5). Ideally,

the “load” propagates through the whole system at the end of t = 10Δt , resulting

in a displacement of 10P

/

k for particle 1. In general, if this one-dimensional ide-

alized model is comprised of n particles of equal normal stiffness and mass, then

the displacement of particle 1 (i.e. the particle with “load” P) is nP

/

k at the end of

t = nΔt . Clearly, if one uses the record of the “load” and displacement at the end of

the immediate loading step to determine the elastic modulus of this one-dimensional

idealized particle system, then the determined elastic modulus will be exaggerated

8.2 Some Numerical Simulation Issues Associated with the Particle Simulation Method 191

by n times. Although the one-dimensional idealized particle system is a highly sim-

pliﬁed representation of particle models, it illuminates the basic force propagation

mechanism, which is valid for two- and three-dimensional particle models.

The conventional loading procedure used in the distinct element method is

shown in Fig. 8.6. In order to reduce inertial forces exerted on the loading-

boundary particles due to a suddenly-applied velocity at the ﬁrst loading step,

an improved-conventional loading procedure is also used in the distinct element

method (see Fig. 8.6). Since both the conventional loading procedure and the

improved-conventional loading procedure are continuous loading procedures, it is

impossible to take the correct record of the “displacement”, just at the end of a

“load” increment. In other words, when a “load” increment is applied to the particle

system, it takes a large number of time steps for the system to reach a quasi-static

equilibrium state. It is the displacement associated with the quasi-static equilibrium

state that represents the correct displacement of the system due to this particular

“load” increment. For this reason, a new discontinuous loading procedure is pro-

posed in this section. As shown in Fig. 8.6, the proposed loading procedure com-

prises two main types of periods, a loading period and a frozen period. Note that

the proposed loading procedure shown in this ﬁgure is illustrative. In real numer-

ical practice, a loading period is only comprised of a few time steps to avoid the

(Conventional loading procedure)

V=V

wall

t

0

V

(Improved conventional loading procedure)

t

0

V

V=V

wall

t=t

full

(Proposed new loading procedure)

t

0

V

V=V

wall

t=t

full

Fig. 8.6 Illustration of

different loading procedures

for the loading of a particle

model

192 8 Spontaneous Crack Generation Problems in Large-Scale Geological Systems

occurrence of any unphysical damage/crack in the particle model, whereas a frozen

period can be comprised of thousands of time steps. In the loading period, a veloc-

ity increment is applied to the loading boundary of the system, while in the frozen

period the loading boundary is ﬁxed to allow the system to reach a quasi-static equi-

librium state after a large number of time steps. It needs to be pointed out that a pair

of load and displacement (or stress and strain) is correctly recorded at the end of a

frozen period.

Based on the above-mentioned conceptual understanding, the theoretical expres-

sion of the proposed loading algorithm can be deduced as follows. An applied force

in the quasi-static system can be divided into M loading increments.

P =

M

¸

i =1

ΔP

i

, (8.31)

where P is the applied force; ΔP

i

is the ith loading force increment and M is the

total number of loading steps.

For each loading force increment, ΔP

i

, it is possible to ﬁnd a solution, ΔS

i

,

which satisﬁes the following condition:

lim

j →N

i

(ΔS

i j

−ΔS

i j −1

) = 0, (8.32)

where ΔS

i j

is the solution at the jth time step due to the ith loading force incre-

ment; ΔS

i j −1

is the solution at the (j-1)th time step due to the ith loading force

increment; N

i

is the total number of time steps for the particle system to reach a

quasi-static equilibrium state after the application of the ith loading force increment

to the system.

From the numerical analysis point of view, Equation (8.32) can be straightfor-

wardly replaced by the following equation:

max

ΔS

i j

−ΔS

i j −1

ΔS

i j

< δ (at j = N

i

), (8.33)

where δ is the tolerance of the solution accuracy.

It needs to be pointed out that the value of N

i

can be determined using the con-

vergent condition expressed by Eq. (8.33), so that different values of N

i

may be

obtained in the numerical simulation. It is the convergent condition that approxi-

mately warrants the solution of the quasi-static nature, if the value of δ is not strictly

equal to zero.

Thus, the total solution S corresponding to the applied force P can be expressed as

S =

M

¸

i =1

ΔS

i

. (8.34)

8.2 Some Numerical Simulation Issues Associated with the Particle Simulation Method 193

It is noted that the total number of time steps to produce the total solution S is

calculated using the following formula:

N

t ot al

=

M

¸

i =1

N

i

. (8.35)

Similarly, at the end of each loading step, there exist the following relationships:

P

j

=

j

¸

i =1

ΔP

i

( j = 1, 2, . . . , M), (8.36)

S

j

=

j

¸

i =1

ΔS

i

( j = 1, 2, . . . , M), (8.37)

where P

j

is the total applied loading force at the end of loading step j; S

j

is the total

solution at the end of loading step j. This indicates that at the end of a loading step,

a point (P

j

, S

j

) has been obtained in the load-solution space. Therefore, at the end

of loading step M, we have obtained M points so that it is possible to link all these M

points together to obtain a load-solution path in the load-solution space. Clearly, if

P and S represent the applied loading force and resulting displacement, then a force-

displacement curve is obtained in the force-displacement space. Alternatively, if P

and S stand for the applied loading stress and resulting strain respectively, then a

stress-strain curve is obtained in the stress-strain space.

With the biaxial compression test of a particle simulation model taken as an

example, the general steps of using the proposed loading algorithm can be sum-

marized as follows. (1) For a given load increment, which needs to be applied to

the boundary of a particle model, the servo-control technique (Itasca Consulting

Group, inc. 1999) is used to apply the equivalent velocity to the boundary of the

particle model. This means that an equivalent velocity is applied to the boundary of

the particle model for a few time steps during the particle simulation. The number

of time steps, during which the equivalent velocity needs to be applied to the bound-

ary of the particle model, depends on the desired load increment that is applied to

the boundary of the particle model. This step is called the loading step and the cor-

responding loading time or the number of related time steps is called the loading

period. It is noted that in theory, a loading period should be as small as possible, so

that any unphysical damage/crack can be prevented from occurring during the load-

ing period. (2) After a loading period, the applied equivalent velocity is set to be zero

so that the loading boundary of the particle model becomes frozen (i.e. ﬁxed). This

step is called the frozen step. During a frozen step, the particle simulation model

is kept running until an equilibrium state is reached. Therefore, the duration of a

frozen period depends on how quickly the particle simulation model can reach its

corresponding equilibrium state. (3) When the particle simulation model reaches an

equilibrium state, an applied load increment such as a stress increment on the load-

194 8 Spontaneous Crack Generation Problems in Large-Scale Geological Systems

ing boundary of the particle model can be calculated so that the value of the desired

load increment is determined. At the same time, other quantities of interest such

as a displacement increment, strain increment and so forth can be also determined.

Thus, a point relating stress to strain in a stress-strain curve is obtained at the end

of this step. For this reason, this step is called the result acquisition step. (4) Steps

1–3 are repeated for every desired load increment until the ﬁnal stage of the particle

simulation is reached. As a result, many points relating stress to strain in a stress-

strain curve have been obtained at the end of the particle simulation. (5) Finally,

all the obtained points relating stress to strain are connected to generate a stress-

strain curve, which should represent the true quasi-static behavior for the particle

simulation model.

8.3 An Upscale Theory of Particle Simulation

for Two-Dimensional Quasi-Static Problems

For the purpose of establishing an upscale theory associated with the particle sim-

ulation of two-dimensional quasi-static systems, it is necessary to understand the

particle-scale mechanical properties and their relations to macroscopic mechanical

properties, which are available from either a laboratory test or an in-situ measure-

ment. If circular particles of unit thickness are used in the simulation of a particle

assembly, the particle-scale mechanical properties such as the stiffness and bond

strength of the particle are needed for a contact-bond model. Since it is very difﬁ-

cult, if not impossible, to directly measure particle-scale mechanical properties from

laboratory tests, it is common practice to determine these particle-scale mechanical

properties from the macroscopic mechanical properties such as the elastic modulus,

tensile and shear strength of particle materials. From the analog of a two-circle con-

tact with an elastic beam (Itasca Consulting Group, inc. 1999), it has been demon-

strated that there may exist an upscale rule, which states that the contact stiffness of

a circular particle is only dependent on the macroscopic elastic modulus and inde-

pendent of the diameter of the circular particle. The value of the contact stiffness

of a circular particle is equal to twice that of the macroscopic elastic modulus of

the material. On the other hand, the contact bond strength of a circular particle is

directly proportional to both the tensile/shear strength of the particle material and

the diameter of the circular particle.

In order to facilitate the derivation of the corresponding similarity criteria, it is

assumed that the problem domain is comprised of a homogeneous medium. Since

a heterogeneous medium can be divided into many sub-domains of homogeneous

materials, the derived similarity criteria in this investigation is, as demonstrated later

by the test and application example in this chapter, valid and applicable for any two

similar particle models of heterogeneous media, as long as each homogeneous sub-

domain of the two similar particle models satisﬁes the required geometrical similar-

ity criterion. For this reason, the proposed upscale theory associated with the particle

simulation is also applicable for the particle simulation of a geometrically-similar

8.3 An Upscale Theory of Particle Simulation for Two-Dimensional Quasi-Static Problems 195

geological medium, which may be non-homogeneous due to the presence of faults,

cracks and large geological structures, provided that the particle model of the geo-

logical medium satisﬁes the required geometrical similarity criterion.

For the analog of a two-circle contact with an elastic beam (Itasca Consulting

Group, inc. 1999), it is assumed that the behaviour of the contact between two parti-

cles is equivalent to that of an elastic beam with its ends at the two particle centres.

The beam is loaded at its ends by the force acting at the centre of each particle.

Under this assumption, the stress of the beam can be expressed as follows:

σ =

P

D

= Eε = E

Δ

D

, (8.38)

where σ and ε are the stress and strain of the equivalent elastic beam; P is the force

acting at the centre of each particle; E is the elastic modulus of the particle material;

Δ is the deformation of the equivalent elastic beam; D is the diameter of the particle.

Since the two particles are connected in series, the deformation of the equivalent

elastic beam can be also expressed as follows:

Δ =

2P

k

n

, (8.39)

where k

n

is the normal stiffness of a particle.

Substituting Eq. (8.39) into Eq. (8.38) yields the following equation:

k

n

= 2E. (8.40)

Since the contact-bond strength is expressed in the unit of force, the following

relationships exist mathematically:

b

n

= ασ

n

D, (8.41)

b

s

= ατ

s

D, (8.42)

where b

n

and b

s

are the normal and tangential bond strengths at a contact between

two particles; D is the diameter of the particle; α is a constant; σ

n

and τ

s

are the

unit normal and tangential contact bond strengths, which are deﬁned as the normal

and tangential contact bond strengths per unit length of the particle diameter (Zhao

et al. 2007b). If α is assumed to be equal to one, then the values and units of the

unit normal and tangential contact bond strengths are exactly the same as those of

the macroscopic tensile and shear strengths of the particle material, while the values

of the normal and tangential contact bond strengths of a particle are equal to the

product of the unit normal/tangential contact bond strengths and the diameter value

of the particle. Thus, the variation of the normal/tangential contact bond strength

196 8 Spontaneous Crack Generation Problems in Large-Scale Geological Systems

of a particle with its diameter can be straightforwardly considered in the particle

simulation.

Note that in the particle simulation method such as the distinct element method,

the contact force exerted on a particle is calculated using the following formula:

F

n

= k

n

u

n

, (8.43)

where F

n

is the normal contact force at a contact between two particles; u

n

is the

normal displacement at the contact. It needs to be pointed out that in the linear

elastic range of the particle material, a similar relationship to Eq. (8.43) is also valid

for the shear contact force and tangential displacement at a contact between two

particles.

For a quasi-static system, two particle models of different length-scales can be

considered to establish the upscale theory. The ﬁrst particle model (i.e. model one)

is of a small length-scale such as a laboratory length-scale, while the second parti-

cle model (i.e. model two) is of a large length-scale such as a regional geological

length-scale. From elasticity theory, the necessary condition, under which these two

models are similar, is that the relative displacements (i.e. strain) of the two mod-

els are identical. This results in the following similarity criterion for two particle

models of the same number of particles:

u

m1

n

u

m2

n

=

D

m1

D

m2

=

L

m1

L

m2

, (8.44)

where u

m1

n

and u

m2

n

are the displacements of models one and two; D

m1

and D

m2

are

the diameters of the particles of models one and two respectively; L

m1

and L

m2

are

the geometrical lengths of the two models.

Equation (8.44) indicates that if two particle models of different length-scales are

similar, then both the displacement ratio and the diameter ratio of the two models

are equal to their geometrical length ratio. For this reason, Eq. (8.44) is called the

ﬁrst similarity criterion between two particle models of different length-scales.

Consideration of Eqs. (8.40), (8.43) and (8.44) yields the following similarity

criterion for the two particle models:

F

m1

n

F

m2

n

=

u

m1

n

u

m2

n

=

L

m1

L

m2

, (8.45)

where F

m1

n

and F

m2

n

are the contact forces between two particles of models one and

two respectively.

Equation (8.45) states that the similarity ratio of the contact forces of the two

models is equal to both their similarity ratio of displacements and their similarity

ratio of geometrical lengths. This equation is called the second similarity criterion

between two particle models of different length-scales.

8.3 An Upscale Theory of Particle Simulation for Two-Dimensional Quasi-Static Problems 197

Similarly, consideration of Eqs. (8.41) and (8.42) results in the following simi-

larity criterion for the two particle models:

b

m1

n

b

m2

n

=

b

m1

s

b

m2

s

=

D

m1

D

m2

=

L

m1

L

m2

, (8.46)

where b

m1

n

and b

m2

n

are the normal bond strengths of the particles of models one and

two; b

m1

s

and b

m2

s

are the tangential bond strengths of the particles of models one

and two respectively.

Equation (8.46) indicates that the similarity ratios of the (normal and tangential)

bond strengths of the two particle models are equal to both their similarity ratio of

particle diameters and their similarity ratio of geometrical lengths. This equation is

called the third similarity criterion between two particle models of different length-

scales. Since the similarity ratios of the particle bond strengths are equal to the

similarity ratio of the particle contact forces, the occurrence of the ﬁrst failure should

be similar for two similar particle models. This indicates that if the ﬁrst failure

occurs at a particle in model one, then the ﬁrst failure should occur at a similar

counterpart in model two.

For a quasi-static geological system, it is important to consider the gravity effect

in two particle models of different length-scales. At the particle level, the gravity

force exerted on a circular particle can be expressed as follows:

G

p

=

π

4

ρ

p

gD

2

T, (8.47)

where G

p

is the gravity force exerted on a particle; ρ

p

is the density of the particle

material; g is the gravity acceleration; T is the thickness of the circular particle.

Note that T ≡ 1 in this investigation.

From Eq. (8.47), the similarity ratio of gravity forces for the two particle models

of different length-scales can be derived and expressed as follows:

G

m1

p

G

m2

p

=

ρ

m1

p

ρ

m2

p

g

m1

g

m2

(D

m1

)

2

(D

m2

)

2

, (8.48)

where G

m1

p

and G

m2

p

are the gravity forces exerted on the particles of models one

and two respectively; ρ

m1

p

and ρ

m2

p

are the densities of the particle materials of the

two models; g

m1

and g

m2

are the gravity accelerations of the particles of models one

and two respectively.

In order to implement this similarity criterion in particle simulation models, it

is desirable to keep the similarity ratio of the gravity forces equal to that of the

geometrical lengths of the two similar particle models. Since the explicit dynamic

relaxation method is used to solve the equation of motion in a particle simulation

(Itasca Consulting Group, inc. 1999), the similarity ratio of the particle densities

needs to be equal to one so that it does not affect the time-step similarity of the

198 8 Spontaneous Crack Generation Problems in Large-Scale Geological Systems

two models, as discussed below. For this reason, it is necessary to use the following

alternative relationships:

g

m1

g

m2

=

D

m2

D

m1

,

ρ

m1

p

ρ

m2

p

= 1, (8.49)

G

m1

p

G

m2

p

=

ρ

m1

p

ρ

m2

p

g

m1

g

m2

(D

m1

)

2

(D

m2

)

2

=

D

m1

D

m2

. (8.50)

Clearly, Eq. (8.49) states that in order to maintain the similarity of two particle

models of different length-scales, the similarity ratio of the gravity accelerations

should be equal to the inverse of the similarity ratio of the geometrical lengths for the

two similar particle models. In this regard, Eq. (8.49) is called the fourth similarity

criterion between two particle models of different length-scales.

For the particle simulation based on the distinct element method, the numeri-

cal solution to the equation of motion is conditionally stable because the explicit

dynamic relaxation solver is used. The critical time-step, which is required to result

in a stable solution, can be expressed as follows (Itasca Consulting Group, inc.

1999):

Δt

critical

=

m

k

. (8.51)

where Δt

critical

is the critical time-step; m is the mass of a particle and k is the

stiffness between two particles.

It is immediately noted from Eq. (8.51) that the critical time-steps, which are

used in two similar particle models of different length-scales, satisfy the following

similarity criterion:

Δt

m1

critical

Δt

m2

critical

=

ρ

m1

p

ρ

m2

p

D

m1

D

m2

=

D

m1

D

m2

, (8.52)

where Δt

m1

critical

and Δt

m2

critical

are the corresponding critical time-steps used in models

one and two respectively.

Equation (8.52) provides an auxiliary similarity criterion, which is a direct result

from the above-mentioned fourth similarity criterion, for the two similar particle

models of different length-scales.

It needs to be pointed out that the proposed upscale theory associated with par-

ticle simulation methods is strictly valid when the mechanical response of a two-

dimensional particle assembly is within the elastic range. If the loading increment

is small enough to prevent any unphysical damage/crack from occurring within a

two-dimensional particle assembly and the number of time-steps is large enough to

enable the particle assembly to reach a quasi-static state during this loading period,

which can be achieved using the newly-proposed loading procedure associated with

8.4 Test and Application Examples of the Particle Simulation Method 199

the distinct element method (Zhao et al. 2007b), the proposed upscale theory for the

particle simulation is also approximately valid for the post-failure response of the

particle assembly, as demonstrated in the next section.

8.4 Test and Application Examples of the Particle

Simulation Method

As shown in Fig. 8.7, two samples of different sizes are considered in the particle

simulation tests. The ﬁrst test sample is of a small size (1 by 2 m) and is simu-

lated using 1000 particles. It was noted that if a regular hexagonal lattice, in which

a particle is in contact with its six neighboring particles, is used in the particle

simulation (Donze et al. 1994), an unphysical, ﬁrst-order geometrical control may

result in well-deﬁned 60

◦

planes of weakness in the lattice. These planes can fur-

ther control the geometry of the resulting failure (Donze et al. 1994). To prevent the

above-mentioned effect from taking place, we generate a particle model in which

the particles are distributed randomly so that the likelihood for the occurrence of

the preferred planes of weakness can be eliminated. The maximum and minimum

radii of particles are approximately 0.0172 m and 0.0115 m, resulting in an average

radius of 0.0144 m. On the other hand, the second test sample is of large size (1 by

2 km) and is also simulated using 1000 particles. The maximum and minimum radii

of particles are approximately 17.24 m and 11.49 m, resulting in an average radius

of 14.37 m. The initial porosity of both the small and the large test samples is set to

be 0.17 in the particle simulation. The density of the particle material is 2500 kg/m

3

and the friction coefﬁcient of the particle material is 0.5, while the conﬁning stress is

assumed to be 10 MPa in the following numerical experiments. Due to a signiﬁcant

(A) 2 1× m sample 2 1× km sample (B)

Fig. 8.7 Two samples of the

same number of particles but

different sizes

200 8 Spontaneous Crack Generation Problems in Large-Scale Geological Systems

size difference between the small and large test samples, the size effect of the test

sample can be investigated through the particle simulation.

The stiffness and bond strength of particles in a test sample can be predicted

using the macroscopic mechanical properties such as the elastic modulus, tensile

and shear strength of particle materials. From the analog of a two-circle contact

with an elastic beam (Itasca Consulting Group, inc. 1999), it has been demonstrated

that there may exist an upscale rule, which states that the contact stiffness of a circu-

lar particle is only dependent on the macroscopic elastic modulus and independent

of the diameter of the circular particle. The value of the contact stiffness of a circu-

lar particle is equal to twice that of the macroscopic elastic modulus of the material.

On the other hand, the contact bond strength of a circular particle is directly pro-

portional to both the tensile/shear strength of the particle material and the diameter

of the circular particle. To reﬂect this fact, the concept of the unit normal/tangential

contact bond strength, which is deﬁned as normal/tangential contact bond strength

per unit length of the particle diameter, is used in this study. Using this deﬁnition,

the value of the unit normal/tangential contact bond strength is equal to that of

the macroscopic tensile/shear strength of the particle material, while the value of the

normal/tangential contact bond strength of a particle is equal to the product of the

unit normal/tangential contact bond strength and the diameter of the particle. In this

way, the variation of the normal/tangential contact bond strength of a particle with

its diameter is considered in the particle simulation. Keeping the above considera-

tions in mind, the following macroscopic mechanical properties of rock masses are

used to determine the contact mechanical properties of the particle material used in

the simulation. The macroscopic elastic modulus of the particle material is 0.5 GPa,

resulting in a contact stiffness (in both the normal and the tangential directions) of

1.0 GN/m for both the small and the large test samples. The macroscopic tensile

strength of the particle material is 10 MPa, while the macroscopic shear strength of

the particle material is 100 MPa for both the small and the large test samples. The

loading period is 10 time-steps, while the frozen period is 9990 time-steps in the

numerical biaxial compression tests.

For the particle simulation associated with the distinct element method (Itasca

Consulting Group, inc. 1999), crack initiation and generation are determined using

the following crack criteria:

F

Cn

≥ b

n

, (8.53)

|F

Cs

| ≥ b

s

, (8.54)

where F

Cn

and F

Cs

are the normal and shear contact forces at a contact between any

two particles; b

n

and b

s

are the normal and tangential bond strengths at the contact

between the two particles.

Obviously, Eq. (8.53) indicates that if a normal contact force exceeds the corre-

sponding normal tensile bond strength at a contact between two particles, the nor-

mal tensile bond is broken and therefore, a tensile crack is generated at the contact.

8.4 Test and Application Examples of the Particle Simulation Method 201

Similarly, Eq. (8.54) indicates that if the absolute value of a shear contact force

exceeds the corresponding tangential bond strength at a contact between two parti-

cles, the tangential shear bond is broken and therefore, a shear crack is generated

at the contact. The above-mentioned crack criteria are checked for each time step

during the particle simulation of a computational model.

8.4.1 Comparison of the Proposed Loading Procedure

with the Conventional Loading Procedure

Figure 8.8 shows the effect of the loading rate and sample size on the deviatoric

stress versus axial strain curve for both the small and the large samples of 1000

particles. The deviatoric stress is deﬁned as the axial stress minus the conﬁning

stress in this investigation. As usual, the servo-control technique (Itasca Consulting

Group, inc. 1999) is used to apply the equivalent velocity to the loading boundary of

the particle model. The equivalent velocity is called the loading rate hereafter. It is

obvious that the simulated stress-strain curve is independent of the two loading rates

(i.e. LR = 1.0 m/s and LR = 10 m/s in this ﬁgure) and sample sizes in the elastic

response range, where there is no occurrence of any failure in the test material. It

can be found, from the stress-strain curve, that the simulated elastic modulus of the

particle material is equal to 0.5 GPa, which is identical to the desired value of the

expected macroscopic elastic modulus of the particle material. This indicates that

the upscale rule established from the analog of a two-circle contact with an elastic

beam (Itasca Consulting Group, inc. 1999) is appropriate for predicting the elastic

modulus when the proposed loading procedure is used in the simulation of a two-

dimensional particle model.

It is worth pointing out that since the equivalent velocity is simultaneously

applied to both the upper and the lower boundaries of a test sample, the strain rate of

the sample is equal to the ratio of the loading rate to the half-length of the sample. In

the case of the small sample of 1000 particles, the corresponding strain rates of the

sample are 10 (1/s) and 1.0 (1/s) for loading rates of 10 (m/s) and 1.0 (m/s) respec-

tively. However, in the case of the large sample of 1000 particles, the corresponding

strain rates of the sample are 0.01 (1/s) and 0.001 (1/s) for the same loading rates

(i.e. 10 (m/s) and 1.0 (m/s)) as those used in the small sample. It can be observed that

due to the solution uniqueness of the samples in the elastic range, all the stress-strain

curves of both the small and the large samples are identical before the ﬁrst failure

takes place in these two samples. Although the post-failure stress-strain curves of

the two samples are very similar in shape, they are not identical for both the small

and the large samples, indicating that the post-failure response of a sample is also

dependent on the strain rate of the sample. This issue needs to be considered when

a simulation result is obtained from a particle model.

Next, we compare the particle simulation results obtained from using the pro-

posed loading procedure with those obtained from using the improved conventional

loading procedure. For this purpose, both the small and the large test samples of

202 8 Spontaneous Crack Generation Problems in Large-Scale Geological Systems

0.00E+00

1.00E+07

2.00E+07

3.00E+07

4.00E+07

5.00E+07

Axial Strain

D

e

v

i

a

t

o

r

i

c

S

t

r

e

s

s

(

P

a

)

LR= 1.0m/s

LR= 10m/s

(A) 2 1× m sample

0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20

0.00E+00

1.00E+07

2.00E+07

3.00E+07

4.00E+07

5.00E+07

0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20

Axial Strain

D

e

v

i

a

t

o

r

i

c

S

t

r

e

s

s

(

P

a

)

LR= 1.0m/s

LR= 10m/s

(B) 2 1× km sample

Fig. 8.8 Effects of loading rate and sample size on the deviatoric stress versus axial strain curve

using the proposed loading procedure

1000 particles are used to conduct biaxial compression tests using the improved

conventional loading procedure. Figure 8.9 shows the effect of the loading rate on

the curve of deviatoric stress versus axial strain for both the small and the large test

samples of 1000 particles using the improved conventional loading procedure. As

mentioned previously, the strain rates of both the small and the large test samples are

8.4 Test and Application Examples of the Particle Simulation Method 203

0.00E+00

2.00E+07

4.00E+07

6.00E+07

8.00E+07

1.00E+08

Axial Strain

D

e

v

i

a

t

o

r

i

c

S

t

r

e

s

s

(

P

a

)

LR= 1.0m/s

LR= 10m/s

0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20

1× m sample (A) 2

0.00E+00

2.00E+07

4.00E+07

6.00E+07

8.00E+07

1.00E+08

Axial Strain

D

e

v

i

a

t

o

r

i

c

S

t

r

e

s

s

(

P

a

)

LR= 1.0m/s

LR= 10m/s

0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20

(B) 2 1× km sample

Fig. 8.9 Effects of loading rate and sample size on the deviatoric stress versus axial strain curve

using the improved conventional loading procedure

different, even though the loading rates of these two samples are identical. It is

obvious that the general solution pattern for both the small and the large test sam-

ples of 1000 particles is, even though not identical, very similar. This indicates that

the sample size of a particle model has little inﬂuence on the mechanical response

of the model in the elastic range of the particle material, even if the improved

204 8 Spontaneous Crack Generation Problems in Large-Scale Geological Systems

conventional loading procedure is used to produce the simulation results. However,

the mechanical responses of both the small and the large test samples of 1000 par-

ticles are clearly dependent on the loading rate in the elastic range of the particle

material. In the case of a loading rate of 10 m/s, there is an oscillatory behaviour

in the stress-strain curve. Such an oscillatory behaviour does not occur in Fig. 8.8,

where the mechanical responses of both the small and the large test samples of

1000 particles are obtained from using the proposed loading procedure. Neverthe-

less, the oscillatory behaviour of the mechanical response obtained from using the

improved conventional loading procedure is greatly reduced when the smaller load-

ing rate (i.e. LR = 1.0 m/s) is used in the particle simulation, indicating that the

use of the improved conventional loading procedure in a particle simulation may

produce some useful results as long as the loading rate is kept very small in the

simulation. In the case of a loading rate of 10 m/s, the maximum yielding strength

obtained from using the improved conventional loading procedure is almost twice

that obtained from using the proposed loading procedure, implying that the maxi-

mum yielding strength can be overestimated when using the improved conventional

loading procedure. It is interesting to note that the mechanical response obtained

from using the improved conventional loading procedure exhibits stronger ductile

behavior (Fig. 8.9), while the mechanical response obtained from using the pro-

posed loading procedure exhibits stronger brittle behavior (Fig. 8.8) for exactly the

same test sample. This demonstrates that in addition to conceptual soundness, the

proposed loading procedure is more appropriate than the improved conventional

loading procedure in dealing with the numerical simulation of the brittle behavior

of crustal rocks.

8.4.2 The Similarity Test of Two Particle Samples

of Different Length-Scales

The same two samples of different length-scales are considered here. The ﬁrst test

sample is of small size (1 by 2 m) and is simulated using 1000 randomly-distributed

particles. The maximum and minimum radii of the particles used in the particle sam-

ple are approximately 0.01724 m and 0.01149 m, resulting in an average radius of

0.01437 m. On the other hand, the second test sample is of large size (1 by 2 km)

and is also simulated using 1000 randomly-distributed particles. The second test

sample is artiﬁcially designed to validate the proposed upscale theory in this study.

The maximum and minimum radii of the particles used in the particle sample are

approximately 17.24 m and 11.49 m, resulting in an average radius of 14.37 m.

Since the similarity ratio (i.e. 1/1000) of particle diameters is equal to that of geo-

metrical lengths for the two samples, the ﬁrst similarity criterion is satisﬁed between

these two test samples. The initial porosity of both the small and the large test sam-

ples is set to be 0.17 in the particle simulation. The density of the particle mate-

rial is 2500 kg/m

3

and the friction coefﬁcient of the particle material is 0.5, while

the conﬁning stress is taken as 10 MPa in the following numerical experiments.

The macroscopic elastic modulus of the particle material is 0.5 GPa, resulting in a

8.4 Test and Application Examples of the Particle Simulation Method 205

contact stiffness (in both the normal and the tangential directions) of 1.0 GN/m for

each particle in both the meter-scale and the kilometer-scale test samples. Since the

same value of the contact stiffness is used in both the meter-scale and kilometer-

scale samples, the second similarity criterion is satisﬁed between these two test

samples.

To satisfy the third similarity criterion, it is assumed that the macroscopic tensile

strength of the particle material is 10 MPa, while the macroscopic shear strength

of the particle material is 100 MPa for both the meter-scale and the kilometer-scale

test samples. In the case of α = 1, the values of the unit normal and tangential

contact bond strengths are equal to the macroscopic tensile and shear strengths of

the particle material (Zhao et al. 2007b). Since the normal and tangential contact

bond strengths are directly proportional to the particle diameter, the third similarity

criterion is satisﬁed between these two test samples. Thus, all the necessary simi-

larity criteria are satisﬁed for these two samples of different length-scales. For the

purpose of testing the upscale theory in a wide parameter space, effects of three

important parameters, such as the conﬁning stress, the normal bond strength and

the shear bond strength, are investigated using the two similar samples of different

length-scales.

Figure 8.10 shows the effect of the conﬁning stress on the deviatoric stress versus

axial strain curve for both the meter-scale and the kilometer-scale samples of 1000

particles. Keeping other parameters unchanged, three different values of the conﬁn-

ing stress, namely CS =0.1, 1 and 10 MPa shown in Fig. 8.10, are considered in the

particle simulation of the two similar test samples. Note that the deviatoric stress is

deﬁned as axial stress minus conﬁning stress in this investigation. Because it is difﬁ-

cult to directly apply a stress boundary condition to the boundary of a particle model,

the servo-control technique (Itasca Consulting Group, inc. 1999) is used, as an alter-

native, to apply the equivalent velocity of the applied stress to the loading boundary

of the particle model. Using the newly-proposed loading procedure associated with

the distinct element method (Zhao et al. 2007b), the equivalent velocity of 1 m/s is

applied to both the upper and the lower boundaries of the two similar test samples. It

is obvious that the simulated stress-strain curve is dependent on the conﬁning stress.

In particular, the maximum values of the failure stresses of the particle samples are

signiﬁcantly different for three different conﬁning stresses. The general trend of the

conﬁning stress effect is that the higher the conﬁning stress, the greater the max-

imum value of the failure stress of the particle sample. Since both the meter-scale

sample and the kilometer-scale sample are similar, the simulated stress-strain curves

are similar for these two samples of different length-scales, especially in the elas-

tic response range of the particle assemblies. This demonstrates that the proposed

upscale theory is appropriate and useful for establishing an intrinsic relationship

between two similar particle systems of different length-scales.

To examine whether or not the similar particle samples can reproduce the rock

dilation phenomenon observed from laboratory experiments, the effect of the con-

ﬁning stress on the dilation of a particle model is also considered in the parti-

cle simulation of the test sample. Figure 8.11 shows the effect of the conﬁning

stress on the volumetric strain versus axial strain curve for both the meter-scale

206 8 Spontaneous Crack Generation Problems in Large-Scale Geological Systems

0.00E+00

1.00E+07

2.00E+07

3.00E+07

4.00E+07

5.00E+07

0.12 0.08 0.04 0.00

Axial Strain

D

e

v

i

a

t

o

r

i

c

S

t

r

e

s

s

(

P

a

)

CS= 10MPa

CS= 1MPa

CS= 0.1MPa

(A) 2 1× m sample

0.12 0.08 0.04 0.00

Axial Strain

(B) 2 1× km sample

0.00E+00

1.00E+07

2.00E+07

3.00E+07

4.00E+07

5.00E+07

D

e

v

i

a

t

o

r

i

c

S

t

r

e

s

s

(

P

a

)

CS= 10MPa

CS= 1MPa

CS= 0.1MPa

Fig. 8.10 Effects of conﬁning stress on the curve of deviatoric stress versus axial strain

and the kilometer-scale samples. In this ﬁgure, the compressive axial strain is posi-

tive, while the expansive volumetric strain is considered to be positive. This means

that a positive dilation stands for a volumetric expansion, whereas a negative dila-

tion represents a volumetric compression of a particle sample. It can be observed

that the conﬁning stress has a signiﬁcant inﬂuence on the dilation of both the meter-

scale and the kilometer-scale particle samples. At the early stage of the particle

8.4 Test and Application Examples of the Particle Simulation Method 207

0.16 0.12 0.08 0.04 0.00

–8.00E–02

–4.00E–02

0.00E+00

4.00E–02

8.00E–02

1.20E–01

1.60E–01

2.00E–01

2.40E–01

2.80E–01

Axial Strain

V

o

l

u

m

e

t

r

i

c

S

t

r

a

i

n

CS= 10MPa

CS= 1MPa

CS= 0.1MPa

(B) 2 1× km sample

–8.00E–02

–4.00E–02

0.00E+00

4.00E–02

8.00E–02

1.20E–01

1.60E–01

0.16 0.12 0.08 0.04 0.00

Axial Strain

V

o

l

u

m

e

t

r

i

c

S

t

r

a

i

n

(A) 1×2 m sample

CS= 10MPa

CS= 1MPa

CS= 0.1MPa

Fig. 8.11 Effects of conﬁning stress on the curve of volumetric strain versus axial strain

simulation, the dilation of the particle sample is negative, implying that the total

volume of the particle sample decreases with an increase of the axial strain. This

phenomenon continues until the mechanical response of the particle sample reaches

a critical stage, at which a major failure takes place within the particle sample so

that there is a remarkable increase in the volumetric strain of the particle sample.

After this major failure, the volumetric strain of the particle sample becomes posi-

tive, implying that the volume expansion takes place within the particle sample. The

dilation phenomenon of the particle sample is consistent with what was observed in

208 8 Spontaneous Crack Generation Problems in Large-Scale Geological Systems

the laboratory experiments of rocks (Jaeger and Cook 1976). Since higher conﬁning

stress can effectively prevent the lateral expansive axial stain of a test sample from

occurring at the critical stage, the value of the volumetric strain in the case of the

conﬁning stress being 10 MPa is almost double the value of the volumetric strain in

the case of the conﬁning stress being 0.1 MPa. This indicates that an increase in the

conﬁning stress can result in an increase in the compressive volumetric strain at the

critical stage of the test sample. Clearly, the curves of volumetric strain versus axial

strain (as shown in Fig. 8.11) are identical in the elastic response ranges of both the

meter-scale and the kilometer-scale samples, while they are very similar in the post-

failure response ranges of the two samples of different length-scales. It is noted that

in the case of the conﬁning stress being 0.1 MPa, there is a considerable discrepancy

between the maximum values of the volumetric strain in the post-failure response

ranges of the two samples of different length-scales. Nevertheless, this discrepancy

is signiﬁcantly reduced in the case of the conﬁning stress being increased to 10 MPa.

This further demonstrates that the proposed upscale theory is appropriate and use-

ful for establishing an intrinsic relationship between two similar particle systems of

different length-scales.

Next, we investigate the effect of the normal bond strength of particles on the

mechanical responses of both the meter-scale and the kilometer-scale samples. In

this case, the conﬁning stress is taken as 10 MPa, while the unit shear bond strength

of particles is 100 MPa for both the test samples. Three different values of the unit

normal bond strengths of particles, namely NB = 0.1 MPa, 1 MPa and 10 MPa (as

shown in Figs. 8.12 and 8.13), are used in the particle simulation of the two similar

test samples.

Figure 8.12 shows the effect of the normal bond strength on the curves of devia-

toric stress versus axial strain, while Fig. 8.13 shows the effect of the normal bond

strength on the curves of volumetric strain (i.e. the dilation) versus axial strain for

both the meter-scale and the kilometer-scale samples. Due to the geometrical simi-

larity between these two samples, the simulation results fromthe meter-scale sample

are very similar to those from the kilometer-scale sample, especially in the elastic

response ranges of the two similar samples. It is also noted that the normal bond

strength of particles has a signiﬁcant effect on both the stress-strain and the dilation-

strain curves of the two similar samples. The general trend is that the maximum fail-

ure stress of a particle sample increases with an increase in the normal bond strength

of the particles used in the particle sample.

Similarly, the effect of the shear bond strength on the mechanical response of the

two similar samples is examined by considering three different values of the unit

shear bond strengths, namely SB = 1, 10 and 100 MPa (as shown in Figs. 8.14 and

8.15). In this situation, both the conﬁning stress and the unit normal bond strength

are kept as two different constants, which are equal to 10 and 1 MPa in the particle

simulation. Figures 8.14 and 8.15 show the effect of the shear bond strength of

particles on the stress-strain and dilation-strain curves for both the meter-scale and

the kilometer-scale samples respectively. In addition to a clear similarity between

the simulation results obtained from the two similar particle samples of different

length-scales, it is interesting to note that in the case of SB = 100 MPa, both the

8.4 Test and Application Examples of the Particle Simulation Method 209

0.00E+00

1.00E+07

2.00E+07

3.00E+07

4.00E+07

5.00E+07

0.12 0.08 0.04 0.00

Axial Strain

D

e

v

i

a

t

o

r

i

c

S

t

r

e

s

s

(

P

a

)

NB= 10MPa

NB= 1MPa

NB= 0.1MPa

(A) 2 1× m sample

0.12 0.08 0.04 0.00

0.00E+00

1.00E+07

2.00E+07

3.00E+07

4.00E+07

5.00E+07

Axial Strain

D

e

v

i

a

t

o

r

i

c

S

t

r

e

s

s

(

P

a

)

NB= 10MPa

NB= 1MPa

NB= 0.1MPa

(B) 2 1× km sample

Fig. 8.12 Effects of normal bond strength on the curve of deviatoric stress versus axial strain

two samples exhibit a strong brittle behaviour, while in the case of SB =1 MPa, the

same two samples exhibit a strong ductile behaviour, indicating that the shear bond

strength of particles has a signiﬁcant effect on the brittle and ductile behaviour of a

particle assembly. This recognition can provide a basic guidance for the selection of

the shear bond strength of the particle material.

210 8 Spontaneous Crack Generation Problems in Large-Scale Geological Systems

–1.00E–01

–6.00E–02

–2.00E–02

2.00E–02

6.00E–02

0.16 0.12 0.08 0.04 0.00

0.16 0.12 0.08 0.04 0.00

Axial Strain

V

o

l

u

m

e

t

r

i

c

S

t

r

a

i

n

NB= 10MPa

NB= 1MPa

NB= 0.1MPa

(A) 2 1× m sample

–1.00E–01

–6.00E–02

–2.00E–02

2.00E–02

6.00E–02

Axial Strain

V

o

l

u

m

e

t

r

i

c

S

t

r

a

i

n

NB= 10MPa

NB= 1MPa

NB= 0.1MPa

(B) 2 1× km sample

Fig. 8.13 Effects of normal bond strength on the curve of volumetric strain versus axial strain

8.4.3 Particle Simulation of the Folding Process Using Two

Similar Particle Models of Different Length-Scales

As the further test and application example, two particle models of different length-

scales, as shown in Fig. 8.16, are used to simulate the folding process that occurs

within the upper crust of the Earth. In the ﬁrst model, the length and thickness of

the crust are 20 m and 2.5 m, while in the second model, the length and thickness

8.4 Test and Application Examples of the Particle Simulation Method 211

0.12

0.08

0.04 0.00

0.00E+00

5.00E+06

1.00E+07

1.50E+07

2.00E+07

2.50E+07

3.00E+07

3.50E+07

Axial Strain

D

e

v

i

a

t

o

r

i

c

S

t

r

e

s

s

(

P

a

)

SB= 100MPa

SB= 10MPa

SB= 1MPa

(B) 2 1× km sample

0.00E+00

5.00E+06

1.00E+07

1.50E+07

2.00E+07

2.50E+07

3.00E+07

3.50E+07

0.12

0.08

0.04 0.00

Axial Strain

D

e

v

i

a

t

o

r

i

c

S

t

r

e

s

s

(

P

a

)

SB= 100MPa

SB= 10MPa

SB= 1MPa

(A) 2 1× m sample

Fig. 8.14 Effects of shear bond strength on the curve of deviatoric stress versus axial strain

of the crust are 20 km and 2.5 km respectively. The ﬁrst particle model is artiﬁcially

designed to validate the proposed upscale theory in this investigation. Each particle

model has 8000 randomly-distributed particles. The minimum and maximum radii

of the particles used in the meter-scale model are 0.020318 m and 0.030477 m,

resulting in an average radius of 0.025397 m. On the other hand, the minimum and

maximum radii of the particles used in the kilometer-scale model are 20.318 m

212 8 Spontaneous Crack Generation Problems in Large-Scale Geological Systems

–6.00E–02

–5.00E–02

–4.00E–02

–3.00E–02

–2.00E–02

–1.00E–02

0.00E+00

1.00E–02

2.00E–02

0.12 0.08 0.04 0.00

Axial Strain

V

o

l

u

m

e

t

r

i

c

S

t

r

a

i

n

SB= 100MPa

SB= 10MPa

SB= 1MPa

(A) 2 1× m sample

0.12 0.08 0.04 0.00

–6.00E–02

–4.00E–02

–2.00E–02

0.00E+00

2.00E–02

4.00E–02

Axial Strain

V

o

l

u

m

e

t

r

i

c

S

t

r

a

i

n

SB= 100MPa

SB= 10MPa

SB= 1MPa

(B) 2 1× km sample

Fig. 8.15 Effects of shear bond strength on the curve of volumetric strain versus axial strain

and 30.477 m, resulting in an average radius of 25.397 m. The porosities of both

the meter-scale and the kilometer-scale particle models are 0.17, while the density

and friction coefﬁcient of the particle material are 2500 kg/m

3

and 0.5 respectively.

Because the geometrical similarity ratio (i.e. 1/1000) of the two particle models

is equal to the corresponding particle diameter similarity ratio, the ﬁrst similarity

criterion of the upscale theory is satisﬁed between these two similar particle mod-

els. Since the same value of the contact stiffness is used in both the meter-scale

8.4 Test and Application Examples of the Particle Simulation Method 213

(The meter-scale model)

(The kilometer-scale model)

20 km

2.5 m

2.5 km

20 m

Fig. 8.16 Geometries of three-layer models of different length-scales: the middle layer thickness

is 0.5 m for the meter-scale model, while it is 500 m for the kilometer-scale model

and kilometer-scale models, the second similarity criterion is obviously satisﬁed

between these two models. The macroscopic tensile strength of the particle material

is 80 MPa, while the macroscopic shear strength of the particle material is 800 MPa

for both the meter-scale and the kilometer-scale models. As mentioned previously,

if α = 1, the values of the unit normal and tangential contact bond strengths are

equal to those of the macroscopic tensile and shear strengths of the particle material

(Zhao et al. 2007b). Because the normal and tangential contact bond strengths are

directly proportional to the particle diameter, the third similarity criterion is also

satisﬁed between these two models. To satisfy the fourth similarity criterion of the

proposed upscale theory, the gravity acceleration of the kilometer-scale model is

equal to 9.81 m/s

2

, while the gravity acceleration of the meter-scale model is equal

to 9810 m/s

2

, implying that a gravity-acceleration similarity ratio of the meter-scale

model to the kilometer-scale one is equal to 1000, which is the inverse of the geo-

metrical similarity of these two similar particle models. Thus, the fourth similar-

ity criterion as indicated by Eq. (8.49) is satisﬁed between the meter-scale and the

kilometer-scale particle models.

Figure 8.17 shows the evolution of the folding process of the kilometer-scale

model, in which the stiffer middle layer is embedded between softer upper and

lower layers, while Fig. 8.18 shows a similar evolution of the folding process of

the meter-scale model for several different deformation stages of the model. Note

that brown segments are used to show crack patterns in these two ﬁgures. Since

crack initiation and generation can be simulated in the particle simulation of both

the kilometer-scale and meter-scale models, the corresponding crack patterns are

also shown by brown segments in these two ﬁgures. It is clear that because the mid-

dle layer is 10 times stiffer than both the upper and the lower layers, the generated

cracks are almost entirely located within this stiffer middle layer. In the case of hor-

izontal shortening deformation equal to 10.8%, the ﬁrst major crack occurs at the

214 8 Spontaneous Crack Generation Problems in Large-Scale Geological Systems

(10.8% shortening)

(18.0% shortening)

(25.2% shortening)

(32.4% shortening)

(39.6% shortening)

Fig. 8.17 Evolution of the folding process of a stiffer middle layer embedded by softer upper and

lower layers (The kilometer-scale model)

left-hand side of both the kilometer-scale and the meter-scale models, indicating that

the mechanical responses of both the models are identical within the elastic range

of the particle material. Although there are some discrepancies between the simula-

tion results of the two models at post-failure stages, the overall deformation patterns

are very similar between the two similar models of different length-scales, indicat-

ing that the proposed upscale theory is correct and useful for revealing the intrinsic

relationship between the two similar particle models of different length-scales, even

though the two similar particle models are comprised of heterogeneous material

8.4 Test and Application Examples of the Particle Simulation Method 215

(10.8% shortening)

(18.0% shortening)

(25.2% shortening)

(32.4% shortening)

(39.6% shortening)

Fig. 8.18 Evolution of the folding process of a stiffer middle layer embedded by softer upper and

lower layers (The meter-scale model)

regions. This implies that if the four similar criteria of the proposed upscale theory

are satisﬁed, it is possible to use the mechanical response of a small length-scale

model such as a laboratory length-scale model to investigate the potential mechani-

cal response of a large length-scale model such as a geological length-scale model.

From this point of view, the proposed upscale theory of the particle simulation pro-

vides a useful bridge between the simulation results obtained from any two similar

particle models of different length-scales.

216 8 Spontaneous Crack Generation Problems in Large-Scale Geological Systems

8.4.4 Particle Simulation of the Faulting Process Using

the Proposed Particle Method

The problem associated with the crustal fault-propagation folding above rigid base-

ment blocks is considered to illustrate the particle simulation of spontaneous crack

generation problems in large-scale quasi-static geological systems. Figure 8.19

shows the geometry of the computational model, in which the length and height

are 10 km and 2.5 km respectively. The dip angle of an underlying fault in the

rigid basement is 60 degrees (i.e. θ = 60

◦

). The model is simulated by 4000 parti-

cles. The maximum and minimum radii of particles are approximately 30.48 m and

20.32 m, resulting in an average radius of 25.4 m. Although the model is mechani-

cally comprised of one homogeneous layer, we use 10 approximately ﬂat-lying and

constant-thickness marker beds to monitor the deformation patterns. The macro-

scopic elastic modulus of the particle material used in the model is 5 GPa, resulting

in a contact stiffness (in both the normal and the tangential directions) of 10 GN/m

for the computational model. The normal/tangential contact bond strength of a parti-

cle is considered in the exactly same manner as in the previous section. This means

that the value of the unit normal/tangential contact bond strength is equal to that

of the macroscopic tensile/shear strength of the particle material, while the value

of the normal/tangential contact bond strength of a particle is equal to the prod-

uct of the unit normal/tangential contact bond strength and the diameter value of the

particle. In this way, the variation of the normal/tangential contact bond strength of a

particle with its diameter value is considered in the particle simulation. The macro-

scopic tensile strength of the particle material is 20 MPa, while the macroscopic

shear strength of the particle material is 200 MPa for the computational model. The

friction coefﬁcient of the particle material is 0.5 and the density of the particle mate-

rial is assumed to be 2500 kg/m

3

in the particle simulation. The loading period is

10 time-steps, while the frozen period is 1000 time-steps in the numerical simula-

tion. Particles in contact with both the lateral boundaries are not allowed to move in

the horizontal direction but are allowed to move in the vertical direction. Since the

top of the computational model is a free surface, a stress-free boundary condition is

applied to this boundary. The model is run to reach an initial equilibrium state due

2.5 km

θ

5 km 5 km

Fig. 8.19 Geometry of the computational model

8.4 Test and Application Examples of the Particle Simulation Method 217

to gravity. In order to simulate the slip of the underlying fault, the right half of the

bottom is ﬁxed, while the left half of the bottom is allowed to move in the direction

that is parallel to the underlying fault plane in the rigid basement.

As we mentioned in the previous section, the second numerical simulation issue

associated with the distinct element method is an inherent issue, which is caused

by using the explicit dynamic relaxation method to solve a quasi-static problem.

Although the problem related to this issue cannot be completely solved at this stage,

an expedient measure is strongly recommended to carry out a particle-size sensitiv-

ity analysis of at least two different models, which have the same initial geometry

but different total numbers of particles, to conﬁrm the particle simulation result of

a large-scale quasi-static system. For this purpose, the same problem as considered

above is simulated using 8000 particles, so that the total number of particles used in

this simulation is twice that used in the previous simulation. For ease of discussion,

the previous model of 4000 particles is called the 4000-particle model, while the

model of 8000 particles is deﬁned as the 8000-particle model. For the 8000-particle

model, the maximum and minimum radii of particles are approximately 21.55 m

and 14.37 m, resulting in an average radius of 17.96 m. Note that the average radius

of particles used in the 4000-particle model is 25.4 m.

It needs to be pointed out that in theory, the smallest particle size of a particle

model is related directly to the material fracture toughness (Potyondy and Cundall

2004), especially under mixed compressive-extensile conditions. In the case of mod-

eling damage processes for which macroscopic cracks form, the smallest particle

size and model properties should be chosen to match the material fracture toughness

as well as the unconﬁned compressive strength. However, it was also found that the

formation of a failure plane and secondary macro-cracks may be independent of par-

ticle size under mixed compressive-shear conditions (Potyondy and Cundall 2004),

which are those that we consider in this study. Nevertheless, in order to test whether

or not the formation of macroscopic cracks is dependent on the smallest particle

size, it is recommended that a particle-size sensitivity analysis of at least two dif-

ferent models, which have the same geometry but different smallest particle sizes,

be carried out to conﬁrm the particle simulation result of a large-scale quasi-static

system.

Figure 8.20 shows a comparison of crack patterns within the 4000-particle and

8000-particle models respectively. Note that brown segments are used to show crack

patterns in this ﬁgure and the forthcoming ﬁgure (i.e. Fig. 8.21). It is observed that

in terms of the two major macroscopic cracks, both the 4000-particle model and

the 8000-particle model produce the identical results, although the simulation result

of the 8000-particle model is of higher resolution. This conﬁrms that the particle

simulation results obtained from the 4000-particle model is appropriate for show-

ing the major macroscopic cracks in the computational model. It is also noted that

the deformation pattern displayed in Fig. 8.20 is very similar to that reported in a

previous publication (Finch et al. 2003). This demonstrates that in addition to the

conceptual soundness, the proposed loading procedure is correct and useful for deal-

ing with the numerical simulation of the brittle behavior of crustal rocks. For the

above-mentioned reasons, the 8000-particle model is used hereafter to investigate

218 8 Spontaneous Crack Generation Problems in Large-Scale Geological Systems

(4000-particle model with showing particles)

(4000-particle model without showing particles)

(8000-particle model with showing particles)

(8000-particle model without showing particles)

Fig. 8.20 Effect of the total

number of particles on crack

patterns in two

computational models of the

same initial geometry

the effects of the dip angle of the underlying fault on the spontaneous crack gen-

eration patterns due to the crustal fault-propagation folding above rigid basement

blocks.

Figure 8.20 also shows that spontaneous cracks are only generated in the cen-

tral part of the computational model. Since most parts of the computational model

are still in the elastic response state, they can be simulated more efﬁciently using

continuum-mechanics-based numerical methods such as the ﬁnite element and ﬁnite

difference methods (Zhao et al. 1999f). In this regard, it is not the most efﬁ-

cient process to use particles to simulate the whole computational model. To over-

come this disadvantage, the combined use of both the continuum-mechanics-based

8.4 Test and Application Examples of the Particle Simulation Method 219

(

o

30 = θ )

(

o

45 = θ )

(

o

60 = θ )

Fig. 8.21 Effect of the dip angle of the underlying fault on crack generation within the computa-

tional model (Crack pattern with particles)

numerical method and the particle simulation method has been proposed in recent

years (Potyondy and Cundall 2004, Suiker and Fleck 2004, Fleck and Willis 2004).

Since the continuum-mechanics-based numerical method and particle simulation

method are used to simulate the elastic region and cracking region of a computa-

tional model respectively, the efﬁciency of the numerical simulation, as a whole,

can be greatly improved. Since the main purpose of this investigation is to extend

the application range of the particle simulation method from a laboratory scale into

a geological scale, the particular issue of combining the particle simulation method

with the continuum-mechanics-based numerical method is not discussed in detail

here, for the sake of saving space.

Keeping the initial geometry and material properties of the computational model

unchanged, three different dip angles, namely θ = 30

◦

, θ = 45

◦

and θ = 60

◦

, of

220 8 Spontaneous Crack Generation Problems in Large-Scale Geological Systems

the underlying fault in rigid basement blocks are considered to investigate the effect

of the dip angle of the underlying fault on spontaneous crack generation patterns

in crustal brittle rocks. Figure 8.21 shows the related numerical simulation results

of crack generation and deformation patterns in the computational model due to

three different dip angles of the underlying fault. The results shown in this ﬁg-

ure are obtained when the vertical displacement of the left-hand-side basement is

about 30% of the whole thickness of the computational model. It is observed from

these simulation results that there are some remarkable differences in the crack gen-

eration and deformation patterns due to these three different dip angles, implying

that the dip angle of the underlying fault has a signiﬁcant inﬂuence on spontaneous

crack generation patterns in crustal brittle rocks. The major reason for causing such

remarkable differences in the simulation results is that for the same amount of ver-

tical displacement of the left-hand-side basement, the corresponding horizontal dis-

placement of the left-hand-side basement is signiﬁcant different for the three differ-

ent dip angles of the underlying fault. For example, when the vertical displacement

of the left-hand-side basement is about 30% of the whole thickness of the com-

putational model, the corresponding horizontal displacement of the left-hand-side

basement is about 51.97, 30 and 17.32% of the whole thickness of the computa-

tional model in the case of the dip angle being 30

◦

, 45

◦

and 60

◦

respectively. Due to

the signiﬁcant difference in the horizontal displacement of the left-hand-side base-

ment, the crack pattern in the computational model of θ = 30

◦

is more diffuse than

that in the computational model of θ = 60

◦

. In the case of θ = 60

◦

, a macroscopic

crack (i.e. a new fault) is formed in the slip direction of the underlying fault and also

passes the tip point of the left-hand-side basement. However, in the cases of θ =30

◦

and θ = 45

◦

, the dip angle of the resulting new fault in the computational model is

steeper than that of the underlying fault in the rigid basement blocks. In addition,

it is noted that in the case of θ = 45

◦

, there is a large crack within the hanging-

wall of the computational model. This crack penetrates into about 40% of the whole

thickness of the computational model. This phenomenon further demonstrates that

the dip angle of the underlying fault has a signiﬁcant effect on spontaneous crack

generation patterns in crustal brittle rocks.

Summary Statements

As an amalgamation of traditional geoscience and contemporary computational

science, computational geoscience has become an emerging discipline in the past

decade. On the one hand, computational geoscience provides geoscientists with

useful modern scientiﬁc tools for revealing dynamic processes and mechanisms

behind the ﬁeld observations of complicated and complex geoscience phenomena,

but on the other hand, it provides computational scientists with challenging

problems of large scales and multiple processes. In addition to the well-known

experimental and theoretical analysis methods, computational simulation has

become a third dominating and indispensable scientiﬁc method for solving a broad

range of scientiﬁc and engineering problems. Without computational simulation,

it is impossible to solve geoscience problems at a high lever of understanding the

intrinsic dynamic processes and mechanisms that lead to the oberserved geoscience

phenomena in nature.

Owing to the broad and diversity characteristics of geoscience problems, a typ-

ical kind of geoscience problem, known as ore body formation and mineralization

in hydrothermal systems within the upper crust of the Earth, has been considered

to deal with the computational aspects of recent developments of computational

geoscience. Toward this end, advanced numerical methods, procedures and algo-

rithms have been systematically presented in this monograph. Through applying

these numerical methods, procedures and algorithms to some speciﬁc aspects of

ore body formation and mineralization, the following conclusions have been drawn

from the research work reported in the monograph.

(1) The newly-developed computational geoscience discipline is obviously of

multi-disciplinary nature crossing many ﬁelds of science such as mathematics,

physics, chemistry, computational science, geoscience and so forth. The ulti-

mate aim of computational geoscience is to deal with the origin, history and

behaviours of the Earth system in a scientiﬁc and predictive manner. The

research methodology of computational geoscience is a comprehensive research

methodology, which is formed by combining ﬁeld observation, theoretical anal-

ysis, numerical simulation and ﬁeld validation. The primary aim of using this

research methodology is to investigate the dynamic processes and mechanisms

221

222 Summary Statements

of observed geoscience phenomena, rather than to describe the observed phe-

nomena themselves.

(2) Convective pore-ﬂuid ﬂow, known as the steady-state Horton-Rogers-Lapwood

problem, is an important mechanism to control ore body formation and miner-

alization in hydrothermal systems within the upper crust of the Earth. This kind

of problem belongs to a kind of bifurcation problem, from a nonlinear math-

ematics point of view. A progressive asymptotic approach procedure has been

presented for solving the steady-state Horton-Rogers-Lapwood problem in a

ﬂuid-saturated porous medium. This problem possesses a bifurcation and there-

fore makes the direct use of conventional ﬁnite element methods difﬁcult. Even

if the Rayleigh number is high enough to drive the occurrence of natural con-

vection in a ﬂuid-saturated porous medium, the conventional methods often pro-

duce a trivial non-convective solution. This difﬁculty can be overcome using the

progressive asymptotic approach procedure associated with the ﬁnite element

method. The method considers a series of modiﬁed Horton-Rogers-Lapwood

problems in which gravity is assumed to tilt a small angle away from ver-

tical. The main idea behind the progressive asymptotic approach procedure

is that through solving a sequence of such modiﬁed problems with decreas-

ing tilt, an accurate non-zero velocity solution to the Horton-Rogers-Lapwood

problem can be obtained. This solution provides a very good initial prediction

for the solution to the original Horton-Rogers-Lapwood problem so that the

non-zero velocity solution can be successfully obtained when the tilted angle

is set to zero. Comparison of numerical solutions with analytical ones to a

benchmark problem of any rectangular geometry has demonstrated the use-

fulness of the proposed progressive asymptotic approach procedure for deal-

ing with convective pore-ﬂuid ﬂow problems within the upper crust of the

Earth.

(3) To deal with coupled problem between material deformation, pore-ﬂuid ﬂow,

heat transfer, mass transport and chemical reactions in hydrothermal sys-

tems within the upper crust of the Earth, the combined use of two or more

commercially available computer codes is a favorable choice. A consistent

point-searching algorithm for solution interpolation in unstructured meshes

consisting of 4-node bilinear quadrilateral elements has been presented to trans-

late and transfer solution data between two totally different meshes that are used

in two different computer codes, both commercially available. The proposed

algorithm has the following signiﬁcant advantages: ﬁrst, the use of a point-

searching strategy allows a point in one mesh to be accurately related to an

element (containing this point) in another mesh. Thus, to translate/transfer the

solution of any particular point from the mesh used in one computer code to

that in another computer code, only one element needs to be inversely mapped.

This certainly minimizes the number of elements, to which the inverse map-

ping is applied, so that the present algorithm is very effective and efﬁcient.

Second, analytical solutions to the local coordinates of any point in a four-node

quadrilateral element, which are derived in a rigorous mathematical manner,

make it possible to carry out an inverse mapping process very effectively and

Summary Statements 223

efﬁciently. Third, the use of consistent interpolation enables the interpolated

solution to be compatible with an original solution and therefore guarantees

the interpolated solution of extremely high accuracy. The related results from

the test problem have demonstrated the generality, accuracy, effectiveness, efﬁ-

ciency and robustness of the proposed consistent point-searching integration

algorithm.

(4) To effectively and efﬁciently use the ﬁnite element method for solving ﬂuid-

rock interaction problems of subcritical Zhao numbers in pore-ﬂuid saturated

hydrothermal/sedimentary basins, A term splitting algorithm on the basis of a

new concept of the generalized concentration of a solid mineral has been pre-

sented to deal with the following three fundamental issues associated with the

ﬂuid-rock interaction problems. First, since the ﬂuid-rock interaction problem

involves heterogeneous chemical reactions between reactive aqueous chemical

species in the pore-ﬂuid and solid minerals in the rock masses, it is necessary

to develop a new concept of the generalized concentration of a solid mineral,

so that two types of reactive mass transport equations, namely the conven-

tional mass transport equation for the aqueous chemical species in the pore-

ﬂuid and the degenerated mass transport equation for the solid minerals in the

rock mass, can be solved simultaneously in computation. Second, because the

reaction area between the pore-ﬂuid and mineral surfaces is basically a function

of the generalized concentration of the solid mineral, there is a deﬁnite need

to appropriately consider the dependence of the dissolution rate of a dissolv-

ing mineral on its generalized concentration in the numerical analysis. Third,

to consider porosity evolution with time in the transient analysis of ﬂuid-rock

interaction problems, the concept of the equivalent source/sink terms in mass

transport equations needs to be developed to convert the problem of variable

mesh Peclet number and Courant number into the problem of constant mesh

Peclet and Courant numbers. The related numerical results have demonstrated

the usefulness and robustness of the proposed term splitting algorithm for solv-

ing ﬂuid-rock interaction problems of subcritical Zhao numbers in pore-ﬂuid

saturated hydrothermal and sedimentary basins.

(5) The chemical-dissolution-front propagation problem exists ubiquitously not

only in ore forming systems within the upper crust of the Earth, but also in

many other scientiﬁc and engineering ﬁelds. To solve this problem, it is nec-

essary to deal with a coupled system between porosity, pore-ﬂuid pressure and

reactive chemical-species transport in ﬂuid-saturated porous media. Due to the

morphological instability of a chemical dissolution front, this problem needs

to be solved numerically. A segregated algorithm on the basis of a combi-

nation of the ﬁnite element and ﬁnite difference methods has been proposed

for simulating the morphological evolution of chemical dissolution fronts in

reactive transport systems of critical and supercritical Zhao numbers. A set of

analytical solutions have been derived for a benchmark problem to verify the

proposed numerical procedure. Not only can the derived analytical solutions

be used to verify any numerical method before it is used to solve this kind

of chemical-dissolution-front propagation problem, but also they can be used

224 Summary Statements

to understand the fundamental mechanisms behind the morphological instabil-

ity of a chemical dissolution front during its propagation within ﬂuid-saturated

porous media of critical and supercritical Zhao numbers. The related numeri-

cal results have demonstrated that the proposed segregated algorithm and the

related numerical procedure are useful for and capable of simulating the mor-

phological instability of a chemical dissolution front within the ﬂuid-saturated

porous medium.

(6) Non-equilibrium redox chemical reactions of high orders are ubiquitous in

ﬂuid-saturated porous rocks within the crust of the Earth. The numerical mod-

elling of such high-order chemical reactions becomes a challenging prob-

lem because these chemical reactions are not only produced strong nonlin-

ear source/sink terms for reactive transport equations, but also often coupled

with the ﬂuids mixing, heat transfer and reactive mass transport processes.

In order to solve this problem effectively and efﬁciently, it is desirable to

reduce the total number of reactive transport equations with strong nonlinear

source/sink terms to a minimum in a computational model. For this purpose,

a decoupling procedure on the basis of the concept of the chemical reaction

rate invariant has been developed for dealing with ﬂuids mixing, heat transfer

and non-equilibrium redox chemical reactions in ﬂuid-saturated porous rocks.

Using the proposed decoupling procedure, only one reactive transport equa-

tion, which is used to describe the distribution of the chemical product and has

a strong nonlinear source/sink term, needs to be solved for each of the non-

equilibrium redox chemical reactions. The original reactive transport equations

of the chemical reactants with strong nonlinear source/sink terms are turned

into the conventional mass transport equations of the chemical reaction rate

invariants without any nonlinear source/sink terms. A testing example, for some

aspects of which the analytical solutions are available, is used to verify the

proposed numerical procedure. The proposed decoupling procedure associated

with the ﬁnite element method has been used to investigate mineral precipita-

tion patterns due to two reactive ﬂuids focusing and mixing within permeable

faults within the upper crust of the Earth. The related numerical solutions have

demonstrated that the proposed numerical procedure is useful and applicable for

dealing with the coupled problem between ﬂuids mixing, heat transfer and non-

equilibrium redox chemical reactions of high orders in ﬂuid-saturated porous

rocks.

(7) The solidiﬁcation of intruded magma in porous rocks can result in the follow-

ing two consequences: (1) heat release due to the solidiﬁcation of the interface

between the rock and intruded magma and (2) mass release of the volatile ﬂuids

in the region where the intruded magma is solidiﬁed into the rock. Traditionally,

the intruded magma solidiﬁcation problem is treated as a moving interface (i.e.,

the solidiﬁcation interface between the rock and intruded magma) problem to

consider these consequences in conventional numerical methods. An equivalent

source algorithm has been presented to simulate thermal and chemical conse-

quences/effects of magma intrusion in geological systems, which are composed

of porous rocks. Using the proposed equivalent source algorithm, an original

Summary Statements 225

magma solidiﬁcation problem with a moving boundary between the rock and

intruded magma has been transformed into a new problem without the moving

boundary but with the proposed mass sources and physically equivalent heat

sources. The major advantage in using the proposed equivalent source algo-

rithm is that a ﬁxed mesh of ﬁnite elements with a variable integration time-

step can be employed to simulate the consequences and effects of the intruded

magma solidiﬁcation using the conventional ﬁnite element method. The related

results from a benchmark magma solidiﬁcation problem have demonstrated the

correctness and usefulness of the proposed equivalent source algorithm.

(8) To extend the application range of the particle simulation method from a labo-

ratory scale to a large scale such as a geological scale, we need to deal with an

upscale issue associated with simulating spontaneous crack generation prob-

lems in large-scale quasi-static systems. Toward this direction, three impor-

tant simulation issues, which may affect the quality of the particle simulation

results of a quasi-static system, have been addressed. The ﬁrst simulation issue

is how to determine the particle-scale mechanical properties of a particle from

the measured macroscopic mechanical properties of rocks. The second sim-

ulation issue is that ﬁctitious time, rather than physical time, is used in the

particle simulation of a quasi-static problem. The third simulation issue is that

the conventional loading procedure used in the particle simulation method is

conceptually inaccurate, at least from the force propagation point of view. A

new loading procedure and an upscale theory have been presented to solve the

conceptual problems arising from the ﬁrst and third simulation issues. The pro-

posed loading procedure is comprised of two main types of periods, a loading

period and a frozen period. Using the proposed loading procedure and upscale

theory, the parameter selection problem stemming from the ﬁrst issue can be

solved. Since the second issue is an inherent one, it is strongly recommended

that a particle-size sensitivity analysis of at least two different models, which

have the same geometry but different smallest particle sizes, be carried out

to conﬁrm the particle simulation result of a large-scale quasi-static system.

The related simulation results have demonstrated the usefulness and correct-

ness of the proposed loading procedure and upscale theory for dealing with

spontaneous crack generation problems in large-scale quasi-static geological

systems.

(9) This is the end of this monograph, but just the beginning of the computational

geoscience world in the sense that more and more complicated and complex

geoscience problems need to be solved from now on. During writing this mono-

graph, there was the most disastrous earthquake occurring in Wenchuan, China.

Unfortunately, both the time and location for the occurrence of this earthquake

cannot be predicted by the present day’s knowledge of geoscientists. While we

express our deep condolence to those who lost their precious lives during this

earthquake, we really hope that with the further development of computational

geoscience, we can understand all the dynamic processes and mechanisms that

control the occurrence of earthquakes better, so that we could accurately predict

it in such a way as predicting weather today.

References

Alavyoon, F., 1993. On natural convection in vertical porous enclosures due to prescribed ﬂuxes

of heat and mass at the vertical boundaries, International Journal of Heat and Mass Transfer,

36, 2479–2498.

Alexiades, V. and Solomon, A. D., 1993. Mathematical Modelling of Melting and Freezing Pro-

cesses, Hemisphere Publishing Co., Washington, D. C.

Antonellini, M. A. and Pollard, D. D., 1995. Distinct element modeling of deformation bands in

sandstone, Journal of Structural Goelogy, 17, 1165–1182.

Appold, M. S. and Garven, G., 2000. Reactive ﬂow models of ore formation in the Southeast

Missouri district, Economic Geology and the Bulletin of the Society of Economic Geologists,

95, 1605–1626.

Bardet, J. P. and Proubet, J., 1992. The structure of shear bands in idealized granular materials,

Applied Mechamics Review, 45, 118–122.

Barns, H. L., 1997., Geochemistry of Hydrothermal Ore Deposits, John Wiley and Sons Inc.,

New York.

Barsoum, R. S., 1976. On the use of isoparametric ﬁnite elements in linear fracture mechanics,

International Journal for Numerical Methods in Engineering, 10, 25–37.

Barsoum, R. S., 1977. Triangular quarter point elements elastic and perfectly plastic crack tip

elements, International Journal for Numerical Methods in Engineering, 11, 85–98.

Bear, J., 1972. Dynamics of Fluids in Porous Media, American Elsevier Publishing Company,

New York.

Bear, J. and Bachmat, Y., 1990. Introduction to Modelling of Transport Phenomena in Porous

Fractured Media. Kluwer Academic Press.

Beyer, R. P. and LeVeque, R. J., 1992. Analysis of a one-dimensional model for the immersed

boundary method, SIAM Journal on Numerical Analysis, 29, 332–364.

Bons, P. D., Dougherty-Page, J. and Elburg, M. A., 2001. Stepwise accumulation and ascent of

magmas, Journal of Metamorphic Geology, 19, 627–633.

Bouchard, P. O., Bay, Y. and Chastel Y., 2003. Numerical modelling of crack propagation: auto-

matic remeshing and comparison of different criteria, Computer Methods in Applied Mechanics

and Engineering, 192, 3887–3908.

Braun, J., Munroe, S. M. and Cox, S. F., 2003. Transient ﬂuid ﬂow in and around a fault, Geoﬂuids,

3, 81–87.

Buck, W. R. and Parmentier, E. M., 1986. Convection between young oceanic lithosphere: Impli-

cation for thermal structure and gravity, Journal of Geophysical Research, 91, 1961–1974.

Burbidge, D. R. and Braun, J., 2002. Numerical models of the evolution of accretionary wedges

and fold-and-thrust belts using the distinct-element method, Geophysical Journal International,

148, 542–561.

Buretta, R. J. and Berman, A. S., 1976. Convective heat transfer in a liquid saturated porous layer,

ASME Journal of Applied Mechanics, 43, 249–253.

227

228 References

Burnham, C. W., 1979. The importance of volatile constituents, in The Evolution of the Igneous

Rocks: Fiftieth Anniversary Perspectives, Princeton University Press, New York.

Caltagirone, J. P., 1975. Thermoconvective instabilities in a horizontal layer, Journal of Fluid

Mechanics, 72, 269–287.

Caltagirone, J. P., 1976. Thermoconvective instabilities in a porous medium bounded by two con-

centric horizontal cylinders, Journal of Fluid Mechanics, 76, 337–362.

Camborde, F., Mariotti, C. and Donze, F. V., 2000. Numerical study of rock and concrete behavior

by distinct element modeling, Computers and Geotechnics, 27, 225–247.

Carslaw, H. and Jaeger, J., 1959. Conduction of Heat in Solids, Clarendon Press, Oxford.

Chadam, J., Hoff, D., Merino, E., Ortoleva, P. and Sen, A., 1986. Reactive inﬁltration instabilities,

IMA Journal of Applied Mathematics, 36, 207–221.

Chadam, J., Ortoleva, P. and Sen, A., 1988. A weekly nonlinear stability analysis of the reactive

inﬁltration interface, IMA Journal of Applied Mathematics, 48, 1362–1378.

Chen, J. S. and Liu, C. W., 2002. Numerical simulation of the evolution of aquifer poros-

ity and species concentrations during reactive transport, Computers and Geosciences, 28,

485–499.

Chevalier, S., Bernard, D. and Joly, N., 1999. Natural convection in a porous layer bounded by

impervious domains: from numerical approaches to experimental realization, International

Journal of Heat and Mass Transfer, 42, 581–597.

Cline, J. S. and Hofstra, A. A., 2000. Ore-ﬂuid evolution at the Getchell Carline-type gold deposit,

Nevada, USA, European Journal of Mineralogy, 12, 195–212.

Combarnous, M. A. and Bories, S. A., 1975. Hydrothermal convection in saturated porous media,

Adv. Hydroscience, 10, 231–307.

Connolly, J. A. D., 1997. Mid-crustal focused ﬂuid movements: Thermal consequences and silica

transport. In: Jamtveit, B. and Yardley, B. W. D. (Eds.), Fluid Flow and Transport in Rocks:

Mechanics and Effects, Chapman and Hall, London, pp. 235–250.

Cook, R. D., Malkus, D. S. and Plesha, M. E., 1989. Concepts and Applications of Finite Element

Analysis, Wiley, New York.

Cox, S. F., Sun, S. S., Etheridge, M. A., Wall, V. J. and Potter, T. F., 1995. Structural and geo-

chemical controls on the development of turbidite-hosted gold quartz vein deposits, Wattle

Gully mine, central Victoria, Australia, Economic Geology and the Bulletin of the Society of

Economic Geologists, 90, 1722–1746.

Crank, J., 1984. Free and Moving Boundary Problems, Oxford University Press, London.

Cundall, P. A. and Strack, O. D. L., 1979. A discrete numerical model for granular assemblies,

Geotechnique, 29, 47–65.

Cundall, P. A., 2001. A discontinuous future for numerical modelling in geomechanics?, Proceed-

ings of the Institution of Civil Engineers: Geotechnical Engineering, 149, 41–47.

Detournay, E. and Cheng, A. H. D., 1993. Fundamentals of poroelasticity. In: J. A. Hudson and

C. Fairhurst (Eds.), Comprehensive Rock Engineering, Vol.2: Analysis and Design Methods,

Pergamon Press, New York.

Doin, M. P., Fleitout, L. and Christensen, U., 1997. Mantle convection and stability of

depleted and undepleted continental lithosphere, Journal of Geophysical Research, 102,

2771–2787.

Donze, F., Mora, P. and Magnier, S. A., 1994. Numerical simulation of faults and shear zones,

Geophysical Journal International, 116, 46–52.

Donze, F., Magnier, S. A. and Bouchez, J., 1996. Numerical modeling of a highly explosive source

in an elastic-brittle rock mass, Journal of Geophysical Research, 101, 3103–3112.

Erdogan, F. and Sih, G. C., 1963. On the crack extension in plane loading and transverse shear,

Journal Basic Engineering, 85, 519–527.

Everett, C. E., Wilkinson, J. J. and Rye, D. M., 1999. Fracture-controlled ﬂuid ﬂow in the Lower

Palaeozoic basement rocks of Ireland: implications for the genesis of Irish-type Zn-Pb deposits.

In: McCaffrey, K. J. W., Lonergan. L. and Wilkinson, J. J. (Eds), Fractures, Fluid Flow and

Mineralization, Geological Society, London, Special Publications 155, 247–276.

References 229

Finch, E., Hardy, S. and Gawthorpe, R., 2003. Discrete element modeling of contractional fault-

propagation folding above rigid basement fault blocks, Journal of Strctural Geology, 25,

515–528.

Finch, E., Hardy, S. and Gawthorpe, R., 2004. Discrete element modeling of extensional fault-

propagation folding above rigid basement fault blocks, Basin Research, 16, 489–506.

Fleck, N. A. and Willis, J. R., 2004. Bounds and estimates for the effects of strain gradients upon

the effective plastic properties of an isotropic two-phase composite, Journal of the Mechanics

and Physics of Solids, 52, 1855–1888.

Fluid Dynamics International, 1997. Fluid Dynamics Analysis Package: FIDAP, Fluid Dynamics

International Inc., Illinois.

G+D Computing, 1991. Strand6 Reference Manual and User Guide, G+D Computing, Sydney.

Garven, G. and Freeze, R. A., 1984. Theoretical analysis of the role of groundwater ﬂow in the

genesis of stratabound ore deposits: Mathematical and numerical model, American Journal of

Science, 284, 1085–1124.

Garven, G., Appold, M. S., Topygina, V. I. and Hazlett, T. J., 1999. Hydrologic modeling of the

genesis of carbonate-hosted lead-zinc ores. Hydrogeology Journal, 7, 108–126.

Gawin, D., Pesavento, F. and Schreﬂer, B. A., 2003. Modelling of hygro-thermal behaviour of con-

crete at high temperature with thermal-chemical and mechanical material degradation, Com-

puter Methods in Applied Mechanics and Engineering, 192, 1731–1771.

Gobin, D. and Bennacer, R., 1994. Double diffusion convection in a vertical ﬂuid layer: onset of

the convection regime, Physical Fluids, 6, 59–67.

Gow, P., Upton, P., Zhao, C. and Hill, K., 2002. Copper-gold mineralization in the New Guinea:

Numerical modeling of collision, ﬂuid ﬂow and intrusion-related hydrothermal systems,

Australian Journal of Earth Sciences, 49, 753–771.

Goyeau, B., Songbe, J. P. and Gobin, D., 1996. Numerical study of double-diffusive convection in

a porous cavity using Darcy-Brinkman formulation, International Journal of Heat and Mass

Transfer, 39, 1363–1378.

Haar L., Gallagher, J. S. and Kell, G. S., 1984. NBS/NRC Steam Tables: Thermodynamic and

Transport Properties and Computer Programs for Vapor and Liquid States of Water in SI Units,

Taylor & Francis, New York.

Hellen, T. K., 1975. On the method of virtural crack extensions, International Journal for Numer-

ical Methods in Engineering, 9, 187–207.

Hitzmann, M. W., 1995. Mineralisation in the Irish Zn-Pb-(Ba-Ag) Oreﬁeld. In: Anderson, K.,

Ashton, J., Earls, G., Hitzman, M. and Tear, S. (Eds.), Irish Carbonate-Hosted Deposits, Soci-

ety of Economic Geologists, Guidebook Series, 21, 25–61.

Hobbs, B. E., Zhang, Y., Ord, A. and Zhao, C., 2000. Application of coupled deformation, ﬂuid

ﬂow, thermal and chemical modelling to predictive mineral exploration, Journal of Geochemi-

cal Exploration, 69, 505–509.

Horton, C. W. and Rogers, F. T., 1945. Convection currents in a porous medium, Journal of Applied

Physics, 16, 367–370.

Imber, J., Tuckwell, G. W., Childs, C., Walsh, J. J., Manzocchi, T., Heath, A. E., Bonson, C. G. and

Strand, J., 2004. Three-dimensional distinct element modeling of relay growth and breaching

along normal faults, Journal of Structural Geology, 26, 1897–1911.

Ingraffea, A. R. and Manu, C., 1980. Stress-intensity factor computation in three dimensions

with quarter-point elements, International Journal for Numerical Methods in Engineering, 15,

1427–1445.

Islam, M. R. and Nandakumar, K., 1990. Transient convection in saturated porous layers, Interna-

tional Journal of Heat Mass Transfer, 33, 151–161.

Itasca Consulting Group, 1995. Fast Lagrangian Analysis of Continua (FLAC), Itasca Consulting

Group, Inc., Minnesota.

Itasca Consulting Group, Inc., 1999. Particle Flow Code in Two Dimensions (PFC2D), Minneapo-

lis, Minnesota, USA.

Iwashita, K. and Oda, M., 2000. Micro-deformation mechanism of shear banding process based on

modiﬁed distinct element method, Powder Technology, 109, 192–205.

230 References

Jaeger, J. C. and Cook, N. G. W., 1976. Fundamentals of rock mechanics, Chapman & Hall,

London.

Jiang, Z., Oliver, N. H. S., Barr, T. D., Power, W. L. and Ord, A., 1997. Numerical modelling of

fault-controlled ﬂuid ﬂow in the genesis of tin deposits of the Malage ore ﬁeld, Gejiu mining

district, China, Economic Geology, 92, 228–247.

Johnson, A. M. and Pollard, D. D., 1973. Mechanics of growth of some laccolithic intrusions in

the Henry Mountains, Utah, I, Tectonophysics, 18, 261–309.

Joly, N., Bernard, D. and Menegazzi, P., 1996. ST2D3D: A ﬁnite element program to compute

stability criteria for natural convection in complex porous structures, Numerical Heat Transfer,

Part B, 29, 91–112.

Kaviany, M., 1984. Thermal convective instabilities in a porous medium, ASME Journal of Heat

Transfer, 106, 137–142.

Khoei, A. R. and Lewis, R. W., 1999. Adaptive ﬁnite element remeshing in a large deformation

analysis of metal power forming, International Journal for Numerical Methods in Engineering,

45, 801–820.

Klerck, P. A., Sellers, E. J. and Owen, D. R. J., 2004. Discrete fracture in quasi-brittle materi-

als under compressive and tensile stress states, Computer Methods in Applied Mechanics and

Engineering, 193, 3035–3056.

Kwak, D. Y., Cheon, J. S. and Im, Y. T., 2002. Remeshing for metal forming simulations: part I:

Two-dimensional quadrilateral remeshing, International Journal for Numerical Methods in

Engineering, 53, 2436–2500.

Lapwood, E. R., 1948. Convection of a ﬂuid in a porous medium, Proceedings of the Cambridge

Philosophical Society, 44, 508–521.

Lasaga, A. C., 1984. Chemical kinetics of water-rock interactions, Journal of Geophysical

Research, 89, 4009–4025.

Lebon, G. and Cloot, A., 1986. A thermodynamical modeling of ﬂuid ﬂows through porous

media: Application to natural convection, International Journal of Heat Mass Transfer, 29,

381–390.

Lee, N. S. and Bathe, K. J., 1994. Error indicators and adaptive remeshing in large deformation

ﬁnite element analysis, Finite Elements in Analysis and Design, 16, 99–139.

Lewis, R. W. and Schreﬂer, B. A., 1998. The Finite Element Method in the Static and Dynamic

Deformation and Consolidation of Porous Media, John Wiley & Sons, New York.

Li, F. Z., Shih, C. F. and Needleman, A., 1985. A comparison of methods for calculating energy

release rate, Engineering Fracture Mechanics, 21, 405–421.

Lin, G., Zhao, C., Hobbs, B. E., Ord, A. and Muhlhaus, H. B., 2003. Theoretical and numeri-

cal analyses of convective instability in porous media with temperature–dependent viscosity,

Communications in Numerical Methods in Engineering, 19, 787–799.

Lister, J. R. and Kerr, R. C., 1991. Fluid-mechanics models of crack propagation and

their application to magma transport in dykes, Journal of Geophysical Research, 96,

10049–10077.

Liu, L., Yang, G., Peng, S. and Zhao, C., 2005. Numerical modelling of coupled geodynamical

processes and its role in facilitating predictive ore discovery: An example fromTongling, China,

Resource Geology, 55, 21–31.

Lorenzi, H. G., 1985, Energy release rate calculations by the ﬁnite element method, Engineering

Fracture Mechanics, 21, 129–143.

Mamou, M., Vasseur, P. and Bilgen, E., 1998. Double-diffusive convection instability in a vertical

porous enclosure, Journal of Fluid Mechanics, 368, 263–289.

Marsh, B. D., 1982. On the mechanics of igneous diapirism, stoping, and zone melting, American

Journal of Science, 282, 808–855.

Matthai, S. K., Henley, R. W. and Heinrich, C. A., 1995. Gold precipitation by ﬂuid mixing in

bedding-parallel fractures near carbonaceous slates at the Cosmopolitan Howley gold deposit,

northern Australia, Economic Geology and the Bulletin of the Society of Economic Geologists,

90, 2123–2142.

Mazouchi, A. and Homsy, G. M., 2000. Thermocapillary migration of long bubbles in cylindrical

capillary tubes, Physics of Fluids, 12, 542–549.

References 231

McBride, A. Govender, I., Powell, M. and Cloete, T., 2004. Contributions to the experimental val-

idation of the discrete element method applied to tumbling mills, Engineering Computations,

21, 119–136.

McKibbin, R. and O’Sullivan, M. J., 1980. Onset of convection in a layered porous medium heated

from below, Journal of Fluid Mechanics, 96, 375–393.

McLellan, J, G., Oliver, N. H. S., Ord, A., Zhang, Y. and Schaubs, P. M., 2003. A numer-

ical modelling approach to ﬂuid ﬂow in extensional environments: implications for gen-

esis of large microplaty hermatite ores, Journal of Geochemical Exploration, 78–79,

675–679.

Murphy, F. C., Ord, A., Hobbs, B. E., Willetts, G. and Barnicoat, A. C., 2008. Targeting strat-

iform Zn-Pb-Ag massive sulﬁde deposits in Ireland through numerical modeling of coupled

deformation, thermal transport and ﬂuid ﬂow, Economic Geology, 103, 1437–1458.

Nguyen, H. D., Pack, S. and Douglass, R. W., 1994. Study of double diffusive convection in a

layered anisotropic porous medium, Numerical Heat Transfer, 26, 489–505.

Nield, D. A., 1968. Onset of thermohaline convection in a porous medium, Water Resources

Research, 11, 553–560.

Nield, D. A. and Bejan, A., 1992. Convection in Porous Media, Springer-Verlag, New York.

Nithiarasu, P., Seetharamu, K. N. and Sundararajan, T., 1996. Double-diffusive natural convection

in an enclosure ﬁlled with ﬂuid-saturated porous medium: a generalized non-Darcy approach,

Numerical Heat Transfer, 30, 430–436.

Obdam, A. N. M. and Veling, E. J. M., 1987. Elliptical inhomogeneities in groundwater ﬂow: An

analytical description, Journal of Hydrology, 95, 87–96.

Oliver, N. H. S., Pearson, P. J., Holcombe, R. J. and Ord, A., 1999. Mary Kathleen metamorphic-

hydrothermal uranium-rare-earth element deposit: ore genesis and numerical model of coupled

deformation and ﬂuid ﬂow, Australian Journal of Earth Sciences, 46, 467–484.

Oliver N. H. S., 2001. Linking of regional and local hydrothermal systems in the mid-crust by

shearing and faulting, Tectonophysics, 335, 147–161.

Oliver, N. H. S., Ord, A., Valenta, R. and Upton, P., 2001. Deformation, ﬂuid ﬂow, and ore genesis

in heterogeneous rocks, with examples and numerical models from the Mount Isa District,

Australia, Reviews in Economic Geology, 14, 51–74.

Ord, A. and Oliver, N. H. S., 1997. Mechanical controls on ﬂuid ﬂow during regional metamor-

phism: some numerical models, Journal of Metamorphic Geology, 15, 345–359.

Ord, A., Hobbs, B. E., Zhang, Y., Broadbent, G. C., Brown, M., Willetts, G., Sorjonen-Ward, P.,

Walshe, J. and Zhao, C., 2002. Geodynamic modelling of the Century deposit, Mt Isa Province,

Queensland, Australian Journal of Earth Sciences, 49, 1011–1039.

Ord, A. and Sorjonen-Ward, P., 2003. Simulating the Outokumpu mineralising system, The

AusIMM Bulletin, 5, 46–47.

Ormond, A. and Ortoleva, P., 2000. Numerical modeling of reaction-induced cavities in a porous

rock, Journal of Geophysical Research, 105, 16737–16747.

Ortoleva, P., Chadam, J., Merino, E. and Sen, A., 1987. Geochemical self-organization II: The

reactive-inﬁltration instability, American Journal of Science, 287, 1008–1040.

Osher, S. and Sethian, J. A., 1998. Fronts propagating with curvature dependent speed: algorithms

based on Hamilton-Jacobi formulations, Journal of Computational Physics, 79, 12–49.

Owen, D. R. J., Feng, Y. T., de Souza Neto, E. A., Cottrell, M. G., Wang, F., Andrade Pires, F. M.

and Yu, J., 2004. The modelling of multi-fracturing solids and particular media, International

Journal for Numerical Methods in Engineering, 60, 317–339.

Palm, E., Weber, J. E. and Kvernvold, O., 1972. On steady convection in a porous medium, Journal

of Fluid Mechanics, 54, 153–161.

Peacock, S. M., 1989. Numerical constraints on rates of metamorphism, ﬂuid production and

ﬂuid ﬂux during regional metamorphism, Geological Society of America Bulletin, 101,

476–485.

Phillips, O. M., 1991. Flow and Reactions in Permeable Rocks, Cambridge University Press,

Cambridge.

232 References

Pillatsis, G., Taslim, M. E. and Narusawa, U., 1987. Thermal instability of a ﬂuid-saturated porous

medium bounded by thin ﬂuid layers, ASME Journal of Heat Transfer, 109, 677–682.

Potyondy, D. O. and Cundall, P. A., 2004. A bonded-particle model for rock, International Journal

of Rock Mechanics and Mining Science, 41, 1329–1364.

Raffensperger, J. P. and Garven, G., 1995, The formation of unconformity-type uranium ore

deposits: Coupled hydrochemical modelling, American Journal of Science, 295, 639–696.

Rice, J. R., 1968. A path independent integral and the approximate analysis of strain concentrations

by notches and cracks, Journal of Applied Mechanics, 35, 379–386.

Riley, D. S. and Winters, K. H., 1989. Modal exchange mechanisms in Lapwood convection, Jour-

nal of Fluid Mechanics, 204, 325–358.

Rubin, A. M., 1995. Propagation of magma-ﬁlled cracks, Annual Reviews of Earth Planetary Sci-

ence, 23, 287–336.

Saltzer, S. D. and Pollard, D. D., 1992. Distinct element modeling of structures formed in sedimen-

tary overburden by extensional reactivation of basement normal faults, Tectonics, 11, 165–174.

Salman, A. D. and Gorham, D. A., 2000. The fracture of glass spheres, Powder Technology, 107,

179–185.

Schafer, D., Schafer, W. and Kinzelbach, W., 1998a. Simulation of reactive processes related to

biodegradation in aquifers: 1. Structure of the three-dimensional reactive transport model, Jour-

nal of Contaminant Hydrology, 31, 167–186.

Schafer, D., Schafer, W. and Kinzelbach, W., 1998b. Simulation of reactive processes related to

biodegradation in aquifers: 2. Model application to a column study on organic carbon degrada-

tion, Journal of Contaminant Hydrology, 31, 187–209.

Schaubs, P. and Zhao, C., 2002. Numerical modelling of gold-deposit formation in the Bendigo-

Ballarat zone, Victoria, Australian Journal of Earth Sciences, 49, 1077–1096.

Scheidegger, A. E., 1974. The Physics of Flow through Porous Media, University of Toronto Press,

Toronto.

Schopfer, M. P. J., Childs, C. and Walsh, J. J., 2006. Localization of normal faults in multilayer

sequences, Journal of Structural Geology, 28, 816–833.

Schreﬂer, B. A., 2004. Multiphase ﬂow in deforming porous material, International Journal for

Numerical Methods in Engineering, 60, 27–50.

Schreﬂer, B. A., Khoury, G. A., Gawin, D. and Majorana C. E., 2002. Thermo-hydro-mechanical

modelling of high performance concrete at high temperatures, Engineering Computations, 19,

787–819.

Schubert, W., Khanal, M. and Tomas, J., 2005. Impact crushing of particle-particle compounds:

experiment and simulation, International Journal of Mineral Processing, 75, 41–52.

Scott, D. R., 1996. Seismicity and stress rotation in a granular model of the brittle crust, Nature,

381, 592–595.

Sethian, J. A., 1999. Level Set Methods and Fast Marching Methods, Cambridge University,

Cambridge.

Sheldon, H. A. and Ord, A., 2005. Evolution of porosity, permeability and ﬂuid pressure

in dilatant faults post-failure: Implications for ﬂuid ﬂow and mineralization, Geoﬂuids, 5,

272–288.

Sih, G. C. and Macdonald, B., 1974. Fracture mechanics applied to engineering problems: strain

energy density fracture criterion, Engineering Fracture Mechanics, 6, 361–386.

Smooke M. D., Mcenally, C. S., Pfefferle, L. D., Hall, R. J. and Colket, M. B., 1999. Computational

and experimental study of soot formation in a coﬂow, laminar diffusion ﬂame, Combustion and

Flame, 117, 117–139.

Sorjonen-Ward, P., Zhang, Y. and Zhao, C., 2002. Numerical modelling of orogenic processes and

mineralization in the south eastern part of the Yilgarn Craton, Western Australia, Australian

Journal of Earth Sciences, 49, 935–964.

Steefel, C. I. and Lasaga, A. C., 1990. Evolution of dissolution patterns: Permeability change due

to coupled ﬂow and reaction. In: Melchior, D. C. and Basset, R. L. (Eds.), Chemical Modeling

in Aqueous Systems II, American Chemistry Society Symposium Series, 416, 213–225.

References 233

Steefel, C. I. and Lasaga, A. C., 1994. A coupled model for transport of multiple chemical species

and kinetic precipitation/dissolution reactions with application to reactive ﬂow in single phase

hydrothermal systems. American Journal of Science, 294, 529–592.

Strayer, L. M. and Huddleston, P. J., 1997. Numerical modeling of fold initiation at thrust ramps,

Journal of Structural Geology, 19, 551–566.

Strayer, L. M. and Suppe, J., 2002. Out-of-plane motion of a thrust sheet during along-strike

propagation of a thrust ramp: a distinct-element approach, Journal of Structural Geology, 24,

637–650.

Suiker, A. S. J. and Fleck, N. A., 2004. Frictional collapse of granular assemblies, Journal of

Applied Mechanics-Transactions of the ASME, 71, 350–358.

Thomton, C., Ciomocos, M. T. and Adams, M. J., 1999. Numerical simulations of agglomerate

impact breakage, Powder Technology, 105, 74–82.

Tomas, J., Schreier, M., Groger, T. and Ehlers, S., 1999. Impact crushing of concrete for liberation

and recycling, Powder Technology, 105, 39–51.

Tornberg, A. K. and Engquist, B., 2003a. The segment projection method for interface tracking,

Communications on Pure and Applied Mathematics, 56, 47–79.

Tornberg, A. K. and Engquist, B., 2003b. Regularization techniques for numerical approximation

of PDEs with singularities, Journal of Scientiﬁc Computing, 19, 527–552.

Trevisan, O. V. and Bejan, A., 1987. Mass and heat transfer by high Rayleigh number convec-

tion in a porous medium heated from below, International Journal of Mass Heat Transfer, 30,

2341–2356.

Turcotte, D. L. and Schubert, G., 1982. Geodynamics: Applications of Continuum Physics to Geo-

logical Problems, John Wiley & Sons, New York.

Walden, J., 1999. On the approximation of singular source terms in differential equations, Numer-

ical Methods for Partial Differential Equations, 15, 503–520.

Weinberg, R. F., 1996. Ascent mechanism of felsic magmas: News and views, Transactions of the

Royal Society of Edinburgh: Earth Sciences, 87, 95–103.

Wilde, A. R. and Wall, V. J., 1987. Geology of the Nabarlek Uranium deposit, Northern-territory,

Australia, Economic Geology, 82, 1152–1168.

Xu, T. F., Samper, J. Ayora, C., Manzano, M. and Custodio, E., 1999. Modelling of non-isothermal

multi-component reactive transport in ﬁeld scale porous media ﬂowsystems, Journal of Hydrol-

ogy, 214, 144–164.

Xu, T. F., Apps, J. A. and Pruess, K., 2004. Numerical simulation of CO2 disposal by mineral

trapping in deep aquifers, Applied Geochemistry, 19, 917–936.

Yang, J. W., 2006. Full 3-D numerical simulation of hydrothermal ﬂuid ﬂow in faulted sedimentary

basins: Example of the McArthur Basin, Northern Australia. Journal of Geochemical Explo-

ration, 89, 440–444.

Yardley, B. W. D. and Bottrell, S. H., 1992. Silica mobility and ﬂuid movement during

metamorphism of the Connemara schists, Ireland, Journal of Metamorphic Geology, 10,

453–464.

Yardley, B. W. D. and Lloyd, G. E., 1995. Why metasomatic fronts are really sides, Geology, 23,

53–56.

Yeh, G. T. and Tripathi, V. S., 1989. A critical evaluation of recent developments in hydrogeo-

chemical transport models of reactive multichemical components, Water Resources Research,

25, 93–108.

Yeh, G. T. and Tripathi, V. S., 1991. A model for simulating transport of reactive multi-

species components: Model development and demonstration, Water Resources Research, 27,

3075–3094.

Zhang, Y., Hobbs, B. E., Ord, A. Barnicoat, A., Zhao, C. and Lin, G., 2003. The inﬂuence of

faulting on host-rock permeability, ﬂuid ﬂow and ore genesis of gold deposits: a theoretical 2D

numerical model, Journal of Geochemical Exploration, 78–79, 279–284.

Zhang, Y., Lin, G., Wang, Y. J., Roberts, P. A. and Ord, A., 2007. Numerical modelling of defor-

mation and ﬂuid ﬂow in the Shui-Kou-Shan Mineralisation District, Hunan Province, China,

Ore Geology Review, 31, 261–278.

234 References

Zhao, C. and Valliappan S., 1993a. Transient inﬁnite elements for seepage problems in inﬁnite

media, International Journal for Numerical and Analytical Methods in Geomechanics, 17,

324–341.

Zhao, C. and Valliappan S., 1993b. Mapped transient inﬁnite elements for heat transfer problems

in inﬁnite media, Computer Methods in Applied Mechanics and Engineering, 108, 119–131.

Zhao, C. and Valliappan S., 1994a. Transient inﬁnite elements for contaminant transport prob-

lems in inﬁnite media, International Journal for Numerical Methods in Engineering, 37,

1143–1158.

Zhao, C. and Valliappan S., 1994b. Numerical modelling of transient contaminant migration prob-

lems in inﬁnite porous fractured media using ﬁnite/inﬁnite element technique: Theory and Para-

metric study, International Journal for Numerical and Analytical Methods in Geomechanics,

18, 523–564.

Zhao, C., Xu, T. P. and Valliappan, S., 1994c. Numerical modelling of mass transport problems in

porous media: A review, Computers and Structures, 53, 849–860.

Zhao, C. and Steven, G. P., 1996a. Asymptotic solutions for predicted natural frequencies of two-

dimensional elastic solid vibration problems in ﬁnite element analysis, International Journal

for Numerical Methods in Engineering, 39, 2821–2835.

Zhao, C. and Steven, G. P., 1996b. An asymptotic formula for correcting ﬁnite element predicted

natural frequencies of membrane vibration problems, Communications in Numerical Methods

in Engineering, 11, 63–73.

Zhao, C. and Steven, G. P., 1996c. A practical error estimator for predicted natural frequencies of

two-dimensional elastodynamic problems, Engineering Computations, 13, 19–37.

Zhao, C., Steven, G. P. and Xie, Y. M., 1996d. Evolutionary natural frequency optimization of thin

plate bending vibration problems, Structural Optimization, 11, 244–251.

Zhao, C., Steven, G. P. and Xie, Y. M., 1996e. General evolutionary path for fundamental natural

frequencies of structural vibration problems: towards optimum from below, Structural Engi-

neering and Mechanics, 4, 513–527.

Zhao C., M¨ uhlhaus, H. B. and Hobbs, B. E., 1997a. Finite element analysis of steady-state natural

convection problems in ﬂuid-saturated porous media heated from below, International Journal

for Numerical and Analytical Methods in Geomechanics, 21, 863–881.

Zhao, C., Steven, G. P. and Xie, Y. M., 1997b. Effect of initial nondesign domain on optimal

topologies of structures during natural frequency optimization, Computers and Structures, 62,

119–131.

Zhao, C., Steven, G. P. and Xie, Y. M., 1997c. Evolutionary natural frequency optimization of

2D structures with additional nonstructural lumped masses, Engineering Computations, 14,

233–251.

Zhao, C., Steven, G. P. and Xie, Y. M., 1997d. Evolutionary optimization of maximizing the dif-

ference between two natural frequencies of a vibrating structure, Structural Optimization, 13,

148–154.

Zhao C., Hobbs, B. E. and M¨ uhlhaus, H. B., 1998a. Finite element modelling of temperature gra-

dient driven rock alteration and mineralization in porous rock masses, Computer Methods in

Applied Mechanics and Engineering, 165, 175–187.

Zhao C., Muhlhaus, H. B. and Hobbs, B. E., 1998b. Effects of geological inhomogeneity on high

Rayleigh number steady-state heat and mass transfer in ﬂuid-saturated porous media heated

from below, Numerical Heat Transfer, 33, 415–431.

Zhao, C., Steven, G. P. and Xie, Y. M., 1998c. A generalized evolutionary method for natural

frequency optimization of membrane vibration problems in ﬁnite element analysis, Computers

and Structures, 66, 353–364.

Zhao, C., Hornby, P., Steven, G. P. and Xie, Y. M., 1998d. A generalized evolutionary method

for numerical topology optimization of structures under static loading conditions, Structural

Optimization, 15, 251–260.

Zhao C., Hobbs, B. E. and M¨ uhlhaus, H. B., 1998e. Analysis of pore-ﬂuid pressure gradient and

effective vertical-stress gradient distribution in layered hydrodynamic systems, Geophysical

Journal International, 134, 519–526.

References 235

Zhao, C., Hobbs, B. E. and M¨ uhlhaus, H. B., 1999a. Finite element modelling of reactive mass

transport problems in ﬂuid-saturated porous media, Communications in Numerical Methods in

Engineering, 15, 501–513.

Zhao C., Hobbs, B. E. and M¨ uhlhaus, H. B., 1999b. Theoretical and numerical analyses of convec-

tive instability in porous media with upward throughﬂow, International Journal for Numerical

and Analytical Methods in Geomechanics, 23, 629–646.

Zhao, C., Hobbs, B. E. and M¨ uhlhaus, H. B., 1999c. Effects of medium thermoelasticity on high

Rayleigh number steady-state heat transfer and mineralization in deformable ﬂuid-saturated

porous media heated from below, Computer Methods in Applied Mechanics and Engineering,

173, 41–54.

Zhao C., Hobbs, B. E., M¨ uhlhaus, H. B. and Ord, A., 1999d. Finite element analysis of ﬂow pat-

terns near geological lenses in hydrodynamic and hydrothermal systems, Geophysical Journal

International, 138, 146–158.

Zhao, C., Hobbs, B. E., M¨ uhlhaus, H. B. and Ord, A., 1999f. A consistent point-searching

algorithm for solution interpolation in unstructured meshes consisting of 4-node bilinear

quadrilateral elements, International Journal for Numerical Methods in Engineering, 45,

1509–1526.

Zhao, C., Hobbs B. E., Baxter, K., M¨ uhlhaus H. B. and Ord, A., 1999g. A numerical study of pore-

ﬂuid, thermal and mass ﬂow in ﬂuid-saturated porous rock basins, Engineering Computations,

16, 202–214.

Zhao, C., Hobbs, B. E. and M¨ uhlhaus, H. B., 2000a. Finite element analysis of heat transfer and

mineralization in layered hydrothermal systems with upward throughﬂow, Computer Methods

in Applied Mechanics and Engineering, 186, 49–64.

Zhao, C., Hobbs, B. E., M¨ uhlhaus, H. B., Ord, A. and Lin, G., 2000b. Numerical modelling of

double diffusion driven reactive ﬂow transport in deformable ﬂuid-saturated porous media with

particular consideration of temperature-dependent chemical reaction rates, Engineering Com-

putations, 17, 367–385.

Zhao, C., Hobbs, B. E., M¨ uhlhaus, H. B. and Ord, A., 2000c. Finite element modelling of dissi-

pative structures for nonequilibrium chemical reactions in ﬂuid-saturated porous media, Com-

puter Methods in Applied Mechanics and Engineering, 184, 1–14.

Zhao, C., Hobbs, B. E., M¨ uhlhaus, H. B., Ord, A. and Lin, G., 2001a. Finite element modelling

of three-dimensional convection problems in pore-ﬂuid saturated porous media heated from

below, Communications in Numerical Methods in Engineering, 17, 101–114.

Zhao C., Lin, G., Hobbs, B. E., M¨ uhlhaus, H. B., Ord, A. and Wang, Y., 2001b. Finite element

modelling of heat transfer through permeable cracks in hydrothermal systems with upward

throughﬂow, Engineering Computations, 18, 996–1011.

Zhao, C., Hobbs, B. E., Walshe, J. L., M¨ uhlhaus, H. B. and Ord, A., 2001c. Finite

element modeling of ﬂuid-rock interaction problems in pore-ﬂuid saturated hydrother-

mal/sedimentary basins, Computer Methods in Applied Mechanics and Engineering, 190,

2277–2293.

Zhao, C., Hobbs, B. E., M¨ uhlhaus, H. B. and Ord, A., 2001d., Finite element modelling of rock

alteration and metamorphic process in hydrothermal systems, Communications in Numerical

Methods in Engineering, 17, 833–843.

Zhao, C., Lin, G., Hobbs, B. E., Wang, Y., M¨ uhlhaus, H. B. and Ord, A., 2002a. Finite

element modelling of reactive ﬂuids mixing and mineralization in pore-ﬂuid saturated

hydrothermal/sedimentary basins, Engineering Computations, 19, 364–387.

Zhao, C., Hobbs, B. E., M¨ uhlhaus, H. B., Ord, A. and Lin, G., 2002b. Computer simulations

of coupled problems in geological and geochemical systems, Computer Methods in Applied

Mechanics and Engineering, 191, 3137–3152.

Zhao C., Hobbs, B. E., M¨ uhlhaus, H. B., Ord, A. and Lin, G., 2002c. Analysis of steady-state

heat transfer through mid-crustal vertical cracks with upward throughﬂow in hydrothermal

systems, International Journal for Numerical and Analytical Methods in Geomechanics, 26,

1477–1491.

236 References

Zhao, C., Hobbs, B. E., M¨ uhlhaus, H. B., Ord, A. and Lin, G., 2003a. Finite element modeling of

three-dimensional steady-state convection and lead/zinc mineralization in ﬂuid-saturated rocks,

Journal of Computational Methods in Science and Engineering, 3, 73–89.

Zhao, C., Hobbs, B. E., Ord, A., Lin, G. and M¨ uhlhaus, H. B., 2003b. An equivalent algorithm for

simulating thermal effects of magma intrusion problems in porous rocks, Computer Methods

in Applied Mechanics and Engineering, 192, 3397–3408.

Zhao, C., Hobbs, B. E., Ord, A., M¨ uhlhaus, H. B. and Lin, G., 2003c. Effect of material anisotropy

on the onset of convective ﬂow in three-dimensional ﬂuid-saturated faults, Mathematical

Geology, 35, 141–154.

Zhao, C., Hobbs, B. E., M¨ uhlhaus, H. B., Ord, A. and Lin, G., 2003d. Convective instability of

three-dimensional ﬂuid-saturated geological fault zones heated from below, Geophysical Jour-

nal International, 155, 213–220.

Zhao, C., Lin, G., Hobbs, B. E., Ord, A., Wang, Y. and M¨ uhlhaus, H. B., 2003e. Effects of hot

intrusions on pore-ﬂuid ﬂow and heat transfer in ﬂuid-saturated rocks, Computer Methods in

Applied Mechanics and Engineering, 192, 2007–2030.

Zhao, C., Hobbs, B. E., Ord, A., Peng, S., M¨ uhlhaus, H. B. and Liu, L., 2004. Theoretical inves-

tigation of convective instability in inclined and ﬂuid-saturated three-dimensional fault zones,

Tectonophysics, 387, 47–64.

Zhao, C., Hobbs, B. E., Ord, A., Peng, S., M¨ uhlhaus, H. B. and Liu, L., 2005a. Numerical modeling

of chemical effects of magma solidiﬁcation problems in porous rocks, International Journal for

Numerical Methods in Engineering, 64, 709–728.

Zhao, C., Hobbs, B. E., Ord, A., Peng, S., M¨ uhlhaus, H. B. and Liu, L., 2005b. Double diffusion-

driven convective instability of three-dimensional ﬂuid-saturated geological fault zones heated

from below, Mathematical Geology, 37, 373–391.

Zhao, C., Hobbs, B. E., Ord, A., Kuhn, M., M¨ uhlhaus, H. B. and Peng, S., 2006a. Numerical

simulation of double-diffusion driven convective ﬂow and rock alteration in three-dimensional

ﬂuid-saturated geological fault zones, Computer Methods in Applied Mechanics and Engineer-

ing, 195, 2816–2840.

Zhao, C., Hobbs, B. E., Ord, A. and Hornby, P., 2006b. Chemical reaction patterns due to ﬂuids

mixing and focusing around faults in ﬂuid-saturated porous rocks, Journal of Geochemical

Exploration, 89, 470–473.

Zhao, C., Hobbs, B. E., Hornby, P., Ord, A. and Peng, S., 2006c. Numerical modelling of ﬂuids

mixing, heat transfer and non-equilibrium redox chemical reactions in ﬂuid-saturated porous

rocks, International Journal for Numerical Methods in Engineering, 66, 1061–1078.

Zhao, C., Hobbs, B. E., Ord, A., Hornby, P., Peng, S. and Liu, L., 2006d. Theoretical and numer-

ical analyses of pore-ﬂuid ﬂow patterns around and within inclined large cracks and faults,

Geophysical Journal International, 166, 970–988.

Zhao, C., Hobbs, B. E., Ord, A., Peng, S., Liu, L. and M¨ uhlhaus, H. B., 2006e. Analytical solu-

tions for pore-ﬂuid ﬂow focusing within inclined elliptical inclusions in pore-ﬂuid-saturated

porous rocks: Solutions derived in an elliptical coordinate system, Mathematical Geology, 38,

987–1010.

Zhao, C., Nishiyama, T. and Murakami, A., 2006 f. Numerical modelling of spontaneous crack

generation in brittle materials using the particle simulation method, Engineering Computations,

23, 566–584.

Zhao, C., Hobbs, B. E., Ord, A., Hornby, P., Peng, S. and Liu, L., 2007a. Mineral precipitation

associated with vertical fault zones: The interaction of solute advection, diffusion and chemical

kinetics, Geoﬂuids, 7, 3–18.

Zhao, C., Hobbs, B. E., Ord, A., Hornby, P., Peng, S. and Liu, L., 2007b. Particle simulation

of spontaneous crack generation problems in large-scale quasi-static systems, International

Journal for Numerical Methods in Engineering, 69, 2302–2329.

Zhao, C., Hobbs, B. E., Ord, A., Peng, S. and Liu, L., 2007c. An upscale theory of particle simula-

tion for two-dimensional quasi-static problems, International Journal for Numerical Methods

in Engineering, 72, 397–421.

References 237

Zhao, C., Hobbs, B. E., Ord, A., Robert, P. A., Hornby, P. and Peng, S., 2007d. Phenomenologi-

cal modeling of crack generation in brittle crustal rocks using the particle simulation method,

Journal of Structural Geology, 29, 1034–1048.

Zhao, C., Hobbs, B. E. and Ord, A., 2008a. Investigating dynamic mechanisms of geological phe-

nomena using methodology of computational geosciences: an example of equal-distant miner-

alization in a fault, Science in China D-Series: Earth Sciences, 51, 947–954.

Zhao, C., Hobbs, B. E., Ord, A., Peng, S. and Liu, L., 2008b. Inversely-mapped analytical solu-

tions for ﬂow patterns around and within inclined elliptic inclusions in ﬂuid-saturated rocks,

Mathematical Geosciences, 40, 179–197.

Zhao, C., Hobbs, B. E., Ord, A., Hornby, P., M¨ uhlhaus, H. B. and Peng, S., 2008c. Theoretical and

numerical analyses of pore-ﬂuid-ﬂow focused heat transfer around geological faults and large

cracks, Computers and Geotechnics, 35, 357–371.

Zhao, C., Hobbs, B. E., Ord, A., Hornby, P. and Peng, S., 2008d. Effect of reactive surface

areas associated with different particle shapes on chemical-dissolution front instability in ﬂuid-

saturated porous rocks, Transport in Porous Media, 73, 75–94.

Zhao, C., Hobbs, B. E., Ord, A., Hornby, P. and Peng, S., 2008e. Morphological evolution of three-

dimensional chemical dissolution front in ﬂuid-saturated porous media: Anumerical simulation

approach, Geoﬂuids, 8, 113–127.

Zhao, C., Hobbs, B. E., Hornby, P., Ord, A., Peng, S. and Liu, L., 2008 f. Theoretical and numerical

analyses of chemical-dissolution front instability in ﬂuid-saturated porous rocks, International

Journal for Numerical and Analytical Methods in Geomechanics, 32, 1107–1130.

Zhao, C., Hobbs, B. E., Ord, A., Peng, S., 2008 g. Particle simulation of spontaneous crack gener-

ation associated with the laccolithic type of magma intrusion processes, International Journal

for Numerical Methods in Engineering, 75, 1172–1193.

Zhao, P. H. and Heinrich, J. C., 2002. Approximation to the interface velocity in phase change

front tracking, Communications in Numerical Methods in Engineering, 18, 77–88.

Zienkiewicz, O. C., 1977. The Finite Element Method, McGraw-Hill, London.

Zienkiewicz, O. C. and Zhu, J. Z., 1991. Adaptivity and mesh generation, Int. J. Num. Meth. Eng.,

32, 783–810.

Zimmerman, R. W., 1996. Effective conductivity of a two-dimensional medium containing ellipti-

cal inhomogeneities, Proceedings of the Royal Society of London A, 452, 1713–1727.

Index

A

Advection, 74, 77, 80, 82, 121–123, 134,

136–142, 148, 150

Algorithm, 4–6, 37–38, 42–43, 45, 51, 56, 59,

71, 73, 76–77, 81–82, 88, 95, 109, 111,

115, 128, 153–156, 158–160, 162–164,

166–168, 176, 192–193, 221–225

Application, 1–2, 22, 43, 56, 59, 61, 65, 70, 82,

93, 115, 134, 155, 163, 168–169, 178, 192,

194, 199, 210, 219, 225

Approach, 6, 8–9, 13–16, 19–23, 28–29, 32,

37, 42, 59, 100, 102–103, 107, 122, 134,

136, 156, 177–178, 185, 222

Approximate, 4, 14, 28, 59, 78, 84, 96, 109,

110–112, 115, 128, 146, 148, 165, 169,

199, 204, 216–217

Assumption, 9, 40, 65, 84, 161, 195

Asymptotic approach, 6–9, 13–16, 19–23,

28–29, 32, 42, 100, 167, 222

B

Benchmark, 4–5, 16–19, 37, 96, 109, 111–112,

115, 164–165, 168, 222–223, 225

Bifurcation, 8, 20, 222

Boundary condition, 16, 18, 20, 23, 30, 60,

101, 103, 106, 111–113, 129, 136, 158,

168–169, 175, 190, 205, 216

Buoyancy, 7, 69

C

Concentration, 39–40, 42, 60–61, 63, 65–70,

74, 77–80, 84, 88, 93, 95–97, 99–100,

102, 104, 108–112, 114–117, 124–125,

128–129, 132–133, 136–137, 142,

146–151, 156–157, 163, 169, 171–173, 223

Conduction, 1, 7

Conductivity, 9–10, 30, 51, 61, 129, 157, 165

Conservation, 3, 78–79

Continuity, 104, 157

Convection, 7–9, 14–16, 18, 22, 25, 38, 51, 60,

77, 80, 134

Convergence, 5, 110–111, 156

Coordinate, 10–11, 28–29, 38, 43–47, 50,

100–101, 105, 157–158, 179–181, 222

Crack, 6, 122–123, 153–154, 175–179, 184,

192–195, 198, 200–201, 213, 216–220, 225

Crust, 2, 6–7, 29, 34, 38, 65, 73, 118, 121,

123, 128, 130–131, 136–137, 153–155,

168–169, 172, 175, 177, 204, 210–211,

216–218, 220–224

D

Darcy’s law, 7, 9, 95–96, 123

Darcy velocity, 74, 78, 80, 83–84, 97, 102, 138

Decoupling, 6, 121, 123–126, 128, 133–134,

224

Density, 7, 9, 30, 39–40, 51, 61, 69, 97–100,

102, 112, 114, 118, 123–124, 129, 157,

160, 166–167, 170, 176, 197, 199, 204,

212, 216

Diffusivity, 61, 74, 97, 112, 129, 156–157, 162

Dispersion, 74, 77–78, 82, 84, 108, 115, 121,

123, 129, 136–140, 142–145, 148, 150

Dispersivity, 69, 78

Dissolution, 6, 73–77, 79–80, 84, 88, 93,

95–105, 107–109, 111–113, 115–119, 121,

223–224

Double diffusion, 40, 73

E

Energy, 3, 7–8, 13–14, 40, 61, 176

Equation, 4, 7–10, 13–14, 16–18, 28, 39–41,

43, 45, 47–50, 69–70, 74–75, 77–83,

88, 97, 99, 101, 103–104, 106–110,

122–124, 126–128, 133, 138, 144–145,

148, 156–157, 159, 162, 180, 183–184,

186–187, 192, 195–198, 223–224

239

240 Index

Equilibrium, 6, 40–41, 61, 74–75, 79,

83–84, 95–96, 98–100, 102, 108, 112,

115, 121–122, 124, 126–128, 133–134,

136–140, 142–146, 148, 150, 157, 160,

187–188, 191–193, 216, 224

Expansion coefﬁcient, 9, 30, 39, 41, 51, 56, 61,

124, 129

F

Fault, 2, 6, 122–123, 128–134, 136–137,

140–146, 148, 150, 195, 216–220, 224

Faulting process, 216

Fick’s law, 96, 123

Finite difference, 4–5, 38, 42, 96, 109–110,

154, 178, 183–184, 218, 223

Finite element, 11, 13, 19, 22, 24, 30, 160,

164, 168

Flow, 22, 28, 37, 56, 59, 74, 112, 179, 189

Fluids mixing, 123

Focusing, 96, 118, 129–134, 136–138,

140–142, 144–146, 148, 150, 224

Folding process, 210, 213–215

Fourier’s law, 123

G

Gravity, 7, 9, 14–15, 21, 29–30, 33, 39–40,

124, 197–198, 213, 217, 222

Green-Gauss theorem, 12

H

Heat, 6–9, 12–13, 22, 29–30, 32, 37–40, 51,

56, 59–61, 73, 121–124, 128–129, 137,

145, 153–160, 162–165, 168, 222, 224–225

Heat transfer, 6, 38–39, 51, 56, 59, 73,

121–124, 128, 137, 145, 154–157, 164, 168

Hydrothermal system, 2, 6, 33, 38, 40, 59–65,

69–70, 73, 137, 221–222

I

Initial condition, 82–84, 128, 169

Instability, 6, 75, 95–96, 100, 105, 107–108,

115–116, 223–224

Intrusion, 153–155, 166, 168–173

Inverse problem, 185

K

Kilometer, 4, 156, 177, 186, 205–206, 208,

211–214

Knowledge, 73, 88, 154, 178, 225

L

Length scale, 153–156

Linear problem, 8, 97

M

Magma intrusion, 153–155, 165, 168–173, 224

Mineral, 2, 6–8, 25, 34, 38, 59, 65, 73–80,

84, 95–99, 108, 112, 118, 121–122, 124,

134, 136–138, 141–146, 150, 153–156,

172–173, 175, 177, 221, 223–224

Mineralization, 6–8, 25, 34, 38, 59, 65, 73, 96,

118, 121–122, 124, 136, 153–154, 156,

173, 175, 177, 221

N

Non-equilibrium, 6, 40, 121, 127, 148, 224

Numerical method, 1, 4–5, 8, 30, 59, 73, 76,

96, 109, 111, 124, 127–128, 153–156,

175–176, 218–219, 221–222, 224

Numerical solution, 1, 4–5, 20–23, 31, 33, 43,

50, 59, 61, 65, 70, 82, 88, 93, 96, 102, 104,

109, 112–115, 130–133, 165–167, 188,

222, 224

O

Oberbeck-Boussinesq approximation, 7, 9, 123

Ore body formation, 2, 34, 59, 65, 73, 96,

121–122, 124, 153–154, 156, 172–173,

175, 221–222

Ore deposits, 73, 96, 118, 121

P

Particle simulation, 5–6, 175–179, 183–184,

188–189, 193–194, 196–201, 204–205,

208, 210, 213, 215–217, 219, 225

Peclet number, 76–77, 81–82, 223

Permeability, 10, 23, 30, 41–42, 51, 60–61,

64–65, 75, 95–99, 105, 108, 112, 126, 129,

131, 135, 144, 146

Pore-ﬂuid, 22, 28, 37, 56, 59

Porosity, 9, 30, 39–42, 51, 56, 60–61,

64–65, 74–77, 79–82, 84, 92–93, 95–100,

103–105, 109–113, 115–116, 118,

124–126, 138, 144, 199, 204, 223

Porous medium, 7–10, 14–15, 17–19, 23,

25, 31–36, 38–41, 56–61, 65, 69, 74,

78–79, 81, 84, 88, 95–100, 108, 111–113,

115–119, 124, 138, 222, 224

Precipitation, 6, 73, 76, 80, 84, 88, 93, 97–99,

121, 134, 136–138, 140–146, 148, 150

Pressure gradient, 73–74, 100, 111, 115, 137,

144, 146, 148, 150

Procedure, 1, 6–8, 13–14, 16–17, 19–23, 28,

32, 42, 51, 96, 109–111, 113, 115–116,

118, 121–126, 128, 130, 133–134, 184,

189, 191, 198, 201–205, 217, 221–225

Progressive asymptotis approach, 6–8, 13–14,

16, 19–20, 22, 28–29, 32, 42, 222

Index 241

R

Rayleigh number, 7, 10, 15, 18–21, 23, 25, 29,

38, 42, 51, 61, 222

Response, 177–178, 186, 188–189, 198–199,

201, 203–205, 207–208, 214–215, 218

Rock, 2, 6, 9, 29, 73–93, 95–97, 108, 121–126,

128–129, 131, 133–134, 136–137, 140,

144, 146, 148, 150, 153, 155, 157–159,

161–163, 165–169, 172–173, 175–177,

179, 186, 200, 204–205, 208, 217, 220,

223–225

S

Shape function, 13, 46

Solution, 1, 4–6, 8, 14–23, 28, 30–33, 37–38,

42–45, 47–56, 59–61, 65, 70, 73, 79, 82,

88, 93, 95–96, 99–100, 102, 104–106,

108–115, 123, 127–128, 130–134, 138,

141, 143, 156–157, 161–167, 183–184,

187–188, 192–193, 198, 201, 203, 217,

222–224

Source, 1, 6, 37, 82, 96–99, 124–128, 133, 135,

153, 155–160, 162–165, 168

Steady-state, 6–9, 14, 16, 20, 38–39, 101, 222

Streamline, 23, 26–28, 117–119, 129–130, 146

T

Temperature, 7, 9–14, 17, 20, 22, 30–33,

39–42, 51–52, 54, 56–62, 65, 69, 121,

123–124, 129, 136–137, 155–158,

160–163, 165, 168–172, 175

Term splitting, 6, 73, 77, 81–82, 223

Thermal effect, 41, 61, 65, 154, 156, 166

Transport, 6, 8, 25, 37–40, 59, 61, 69,

73, 75–80, 82–83, 88, 95–97, 99, 105,

107–108, 111, 118, 121–124, 126–128,

133–134, 137–138, 148, 154–157, 164,

222–224

U

Upper crust, 2, 6, 29, 34, 38, 65, 73, 118, 121,

123, 154–155, 168–169, 172, 175, 177,

210, 221–224

Upscale theory, 178, 194, 196, 198–199,

204–205, 208, 211–215, 225

Upward throughﬂow, 128

V

Validation, 3, 221

Velocity, 8–15, 20, 23, 25, 28–29, 31–33, 39,

42, 60, 62, 69, 74, 76, 78–80, 82–84, 97,

99, 102, 124, 129, 131–132, 138–139,

143–144, 146, 148, 150, 181–182, 189,

191, 193, 201, 205, 222

Veriﬁcation, 4, 19, 23, 51, 56, 111, 128, 146,

163, 168

Vertical stress, 56

Viscosity, 9, 30, 39, 51, 61, 97, 124, 129, 144

Z

Zhao number, 6, 73–75, 107–108, 111–112,

115, 223–224

Zone, 118, 129, 133–134, 136–137, 140–146,

148, 150

**Lecture Notes in Earth Sciences
**

Editors: J. Reitner, G¨ ttingen o M. H. Trauth, Potsdam K. St¨ we, Graz u D. Yuen, USA Founding Editors: G. M. Friedman, Brooklyn and Troy A. Seilacher, T¨ bingen and Yale u

122

Chongbin Zhao · Bruce E. Hobbs · Alison Ord

**Fundamentals of Computational Geoscience
**

Numerical Methods and Algorithms

123

Dr. Chongbin Zhao Computational Geosciences Research Centre Central South University Changsha, China Chongbin.zhao@iinet.net.au Dr. Alison Ord School of Earth and Geographical Sciences The University of Western Australia Perth, Australia and CSIRO Division of Exploration and Mining Perth, Australia alison.ord@csiro.au

Dr. Bruce E. Hobbs School of Earth and Geographical Sciences The University of Western Australia Perth, Australia Bruce.Hobbs@csiro.au

ISBN 978-3-540-89742-2 DOI 10.1007/978-3-540-89743-9 Lecturer Notes in Earth Sciences ISSN 0930-0317

Library of Congress Control Number: 2008942068 c Springer-Verlag Berlin Heidelberg 2009

e-ISBN 978-3-540-89743-9

This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, speciﬁcally the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microﬁlm or in any other way, and storage in data banks. Duplication of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version, and permission for use must always be obtained from Springer. Violations are liable to prosecution under the German Copyright Law. The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply, even in the absence of a speciﬁc statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. Cover design: deblik, Berlin Printed on acid-free paper 9 8 7 6 5 4 3 2 1 springer.com

for her persistent support and encouragement. v . We are very grateful to the Central South University for ﬁnancial support during writing this monograph.Acknowledgements Dr Chongbin Zhao expresses his sincere thanks to his wife. The partial work of this monograph is also ﬁnancially supported by the Natural Science Foundation of China (Grant Nos: 10872219 and 10672190). without which it would be impossible to write this monograph. We express our thanks to the anonymous referees for their valuable reviews of this monograph. Ms Peiying Xu.

However.Preamble In recent years. Nevertheless. procedures and algorithms through comparing numerical solutions with benchmark solutions. On the theoretical analysis front. theoretical analysis and computational simulation are two of its core members. (4) to investigate the solution characteristics of these partial differential equations using rigorous mathematical treatments. geophysics. For this reason. under the stimulus of ever-increasing demand vii . has been established. On the computational simulation front. through wide application of computational science to geoscience problems. Since computational geoscience is an amalgamation of geoscience and computational science. numerical methods and computational simulations provide a new way to deal with many geoscience problems. computational scientists and geoscientists are required in the process of establishing the research methodology of computational geoscience. for which the traditionally-used theoretical and experimental methods may not be valid as a result of the large time and length scales of the problems themselves. (2) to verify computational codes established on the basis of these advanced numerical methods. (3) to produce and validate numerical solutions of real geoscience problems. procedures and algorithms for simulating multi-scale and multi-process aspects of the geoscience problem on the basis of contemporary computational science knowledge and expertise. multidisciplinary knowledge and expertise from mathematicians. (3) to establish the theoretical foundations of the geoscience problem through formulating a set of partial differential equations on the basis of fundamental scientiﬁc principles. geochemistry and many other scientiﬁc and engineering ﬁelds. a new discipline. the numerical simulation of a real geological world also provides many challenging problems for researchers involved in the ﬁeld of computational science. (2) to conduct research to ﬁnd the key factors and processes that control the geoscience problem under consideration. chemists. physicists. we need: (1) to develop advanced numerical methods. namely computational geoscience. we need: (1) to measure and gather data and information through traditional geoscience observations and measurements such as those widely used in geology. In particular. computational geoscience is at a developing stage. because of the extremely large length and time scales. This enables many hitherto unsolvable geoscience problems to be solved using numerical methods and computational simulations. Owing to the broad nature of geoscience problems.

driven from the need to understand controlling mechanisms behind ore body formation and mineralization in hydrothermal systems within the upper crust of the Earth. it is high time to publish a monograph to report the current knowledge in a systematic manner. common mathematical notations are used to describe the theoretical aspects of geoscience problems. This enables this monograph to be used either as a useful textbook for postgraduate students or as an indispensable reference book for computational geoscientists. mathematicians. heat transfer and reactive mass transport processes that can occur simultaneously in hydrothermal systems. (3) To simulate mineral dissolution/precipitation and metamorphic processes. In order to disseminate widely the existing knowledge of computational geoscience. procedures and algorithms in the ﬁeld of computational geoscience. but also they have been successfully applied to a wide range of small-scale scientiﬁc and engineering problems. a progressive asymptotic approach procedure is proposed to solve steady-state convective poreﬂuid ﬂow problems within the upper crust of the Earth. each chapter is written independently of the remainder of the monograph so that readers may read the chapter of interest separately. Not only have these three methods been well developed in the ﬁeld of computational science. This monograph aims to provide state-of-theart numerical methods. Fluid Dynamics Analysis Package (FIDAP) and Fast Lagrangian Analysis of Continua (FLAC). the main focus of this monograph is on computational simulation aspects of this newlydeveloped computational geoscience discipline. we have developed advanced numerical procedures and algorithms to tackle the large-scale aspects of geoscience problems. we present the advanced procedures and algorithms in this monograph as follows: (1) Due to the important role that convective pore-ﬂuid ﬂow plays in the controlling processes of ore body formation and mineralization.viii Preamble for natural mineral resources. In this monograph we use the ﬁnite element method. a consistent point-searching interpolation algorithm is proposed to develop a general interface between two commercial computer codes. computational geoscience has achieved much in the past decade. The advanced numerical methods. although some theoretical results are provided to verify numerical ones. the ﬁnite difference method and the particle simulation method as basic numerical methods for dealing with geoscience problems. procedures and algorithms contained in this monograph are also applicable to a wide range of problems of other length-scales such as engineering length-scales. engineers and geoscientists. and to facilitate efﬁciently the broad applications of computational geoscience. Towards this end. (2) To consider both the thermoelastic effect and the double diffusion effect. based on the authors’ own work during the last decade. In addition. to promote extensively and fastly further development of the computational geoscience. a term splitting algorithm is developed for dealing with ﬂuid-rock interaction problems in ﬂuid-saturated hy- . This general interface allows a combination use of the two commercial codes for solving coupled problems between medium deformation. pore-ﬂuid ﬂow. To broaden the readership of this monograph. Based on these three methods. The speciﬁc geoscience problem under consideration is the ore body formation and mineralization problem in hydrothermal systems within the upper crust of the Earth. For this purpose.

(4) For a geochemical system of critical and supercritical Zhao numbers. the particle simulation method is extended to solve spontaneous crack generation problems associated with faulting and folding in large length-scale geological systems. August 12. Note that the Zhao number is a dimensionless number that can be used to represent the geometrical. thermodynamic and chemical kinetic characteristics of a reactive transport system in a comprehensive manner. a decoupling procedure is proposed for simulating ﬂuids mixing.Preamble ix drothermal/sedimentary basins of subcritical Zhao numbers. which can control ore body formation and mineralization patterns within the upper crust of the Earth. (5) To investigate the effects of non-equilibrium redox chemical reactions on the mineralization patterns in hydrothermal systems. (7) To simulate spontaneous crack generation in brittle rocks within the upper crust of the Earth. can be expressed by the critical value of this dimensionless number. Thus. the morphological evolution of chemical dissolution fronts in ﬂuid-saturated porous media can be appropriately simulated. The condition. heat transfer and non-equilibrium redox chemical reactions in ﬂuidsaturated porous rocks. hydrodynamic. under which a chemical dissolution front in the ﬂuid-saturated porous medium becomes unstable. a segregated algorithm is proposed for solving chemical-dissolution front instability problems in ﬂuid-saturated porous rocks. in which the chemical dissolution fronts are stable during their propagation. This algorithm enables the moving boundary problem associated with magma solidiﬁcation to be effectively and efﬁciently solved using the ﬁxed ﬁnite element meshes. Hobbs Alison Ord . The resulting cracks may be connected to form ﬂow channels. 2008 Chongbin Zhao Bruce E. (6) When thermal and chemical effects of intruded magma are taken into account. an equivalent source algorithm is presented for simulating thermal and chemical effects of intruded magma solidiﬁcation problems.

. . . . . .1 Statement of the Coupled Problem and Solution Method . . . . . . . . Heat Transfer and Mass Transport Processes in Hydrothemal Systems . . . . . . . . . . . . . . . . . . . 2. . . . . . . . . . . . . . 1. . . . . . . . . . . . . . . . . . . . . . . . .6. . 38 3. .1 Governing Equations of the Problem .2. . . . . . . . . . . . . . . . .3 The Numerical Simulation Model of a Geoscience Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . .1 Two-Dimensional Convective Pore-Fluid Flow Problems . . . 2. . . . . . 2. . . . . . . . . Pore-Fluid Flow. . . . . . . . . . 1. . . . . . . . . . . . . . . . . . . . . . . . . 1. . . . . . . . . . . . . . . . .3 The Progressive Asymptotic Approach Procedure for Solving Steady-State Natural Convection Problems in Fluid-Saturated Porous Media .2 The Mathematical Model of a Geoscience Problem . . . . 1. . . . . . . . .5 Veriﬁcation of the Proposed Progressive Asymptotic Approach Procedure Associated with Finite Element Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 2. . . . .2 Three-Dimensional Convective Pore-Fluid Flow Problems . . . . . . . . . . . . . . . . . 1. . . . .4 Derivation of Analytical Solution to a Benchmark Problem . . . . . . . . . . . . . . . . . .6 Application of the Progressive Asymptotic Approach Procedure Associated with Finite Element Analysis . .3 The Contextual Arrangements of this Monograph . . . . . . . . . . . . 2 A Progressive Asymptotic Approach Procedure for Simulating Steady-State Natural Convective Problems in Fluid-Saturated Porous Media . . . . . .Contents 1 Introduction . . . . . . . . . . .6. . . . 1. . . . . . . . . . . . . . . . . . . . . . . . .2 Finite Element Formulation of the Problem . . . . . . .4 Graphical Display of the Numerical Simulation Results . . . .2. . . 42 xi . 2. . . . .2 Basic Steps Associated with the Research Methodology of Computational Geoscience . . . . . . . . . . . . . . . . 37 3. . . . 2. . . . . . . . . . . . . . .2 Mathematical Formulation of the Consistent Point-Searching Interpolation Algorithm in Unstructured Meshes . . . . . . . . . . .1 Characteristics of Computational Geoscience . . . . . . . . 1 2 3 3 3 4 5 5 7 9 11 14 16 19 22 22 28 3 A Consistent Point-Searching Interpolation Algorithm for Simulating Coupled Problems between Deformation. . 2.2. . . . . 2. . 1. . . . . . . . . . .1 The Conceptual Model of a Geoscience Problem . . . . . .

. . . . . . . . . .3 Application of the Segregated Algorithm for Simulating the Morphological Evolution of Chemical Dissolution Fronts . . . . . . . . . . . . . . . . . . .3 Application Examples of the Term Splitting Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . . . . . . . . . . . . . . . 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . . . . . . .1 Formulation of the Segregated Algorithm for Simulating the Evolution of Chemical Dissolution Fronts . . . . . . . . . . . . . 96 5. . . 3. . 73 76 77 82 5 A Segregated Algorithm for Simulating Chemical Dissolution Front Instabilities in Fluid-Saturated Porous Rocks . . . . . . . . . . . .3 Veriﬁcation of the Proposed Consistent Point-Searching Interpolation Algorithm . . . 3. . . . . . Pore-Fluid Flow. . . . . . . . . . . . . . . . . . . . . . . . .1 Key Issues Associated with the Numerical Modelling of Fluid-Rock Interaction Problems . . 115 . . . . . . . 3. . . Heat Transfer and Mass Transport in Fluid-Saturated Porous Media . . .4 Application Examples of the Proposed Consistent Point-Searching Interpolation Algorithm . . .2 Numerical Modelling of Coupled Problems Involving Deformation. . . . . . . . . . . . . . . 96 5. . . . . . . .1 Numerical Modelling of Coupled Problems Involving Deformation. . . . . . .3 Consistent Interpolation Step . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 45 50 51 56 56 59 4 A Term Splitting Algorithm for Simulating Fluid-Rock Interaction Problems in Fluid-Saturated Hydrothermal Systems of Subcritical Zhao Numbers .4. . . 3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 5. . . . . .1 Mathematical Background of Chemical Dissolution Front Instability Problems in Fluid-Saturated Porous Rocks . . . . . . . . . . . . . Pore-Fluid Flow and Heat Transfer in Fluid-Saturated Porous Media . . . . . . . . . . . . . . . . . . . . . . 3. . . . . . . . .2 A Particular Case of Reactive Single-Chemical-Species Transport with Consideration of Porosity/Permeability Feedack . . . . . . . . . . . . . . . . . . .4. 109 5. . . . . . . . . . . . . . . . .2 Development of the Term Splitting Algorithm . .2 Inverse Mapping Step . . . . . . . . . . . . . . . . . . . . . . . . . . . 4. . . . . . .1. . . . . . . . . . .1 A General Case of Reactive Multi-Chemical-Species Transport with Consideration of Porosity/Permeability Feedback . . . . . . 3. . . . . . . . . . . . .2. . . . . . . .2 Proposed Segregated Algorithm for Simulating the Morphological Evolution of a Chemical Dissolution Front . . . . .1. . 99 5. .2. . . . . . . . . . . . . . .2. . . . . . . 4.xii Contents 3. .2 Veriﬁcation of the Segregated Algorithm for Simulating the Evolution of Chemical Dissolution Fronts . . . . . .1 Point Searching Step . . 95 5. 111 5. . . .

. . . . . .2. . . . . . . . . . . . . . . . . 184 8. . . . .1 An Equivalent Source Algorithm for Simulating Thermal and Chemical Effects of Intruded Magma Solidiﬁcation Problems . . . . . . .4 Applications of the Proposed Decoupling Procedure to Predict Mineral Precipitation Patterns in a Focusing and Mixing System Involving Two Reactive Fluids . . . . . . . . . . . . . . . . . . . 140 6. . . . 128 6. . . . . . . . . . . . . . . . Heat Transfer and Redox Chemical Reactions . . .Contents xiii 6 A Decoupling Procedure for Simulating Fluid Mixing. . . . . . . . . . . . . . . . 175 8. . . . . . . . . . . . . 194 . . . . . . . . . . . . . 160 7.2 Theoretical Analysis of Mineral Precipitation Patterns in a Focusing and Mixing System Involving Two Reactive Fluids . . . . . . . . . . .4. . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . 134 6. . . . . . . . . . . . . . . . . . . . . . . . 184 8. .4. . . . . . . . . . . . . . . . . . . . 153 7. . . . . .1 Basic Formulations of the Particle Simulation Method . .1 Statement of Coupled Problems between Fluids Mixing. . . . . . . . . .2 Implementation of the Equivalent Source Algorithm in the Finite Element Analysis with Fixed Meshes . . . . .3 Chemical Reaction Patterns due to Mixing and Focusing of Two Reactive Fluids in Permeable Fault Zones . . . . . . . . . . . . . . . . . . . . . . . . .3 Veriﬁcation of the Decoupling Procedure . . . . . . . . . . . . . . . . . . . . . . . . 163 8 The Particle Simulation Method for Dealing with Spontaneous Crack Generation Problems in Large-Scale Geological Systems . . . . . Heat Transfer and Non-Equilibrium Redox Chemical Reactions in Fluid-Saturated Porous Rocks . . . . . . . . . . . 136 6. . . . . . . . . . 179 8. .4 Numerical Illustration of Three Types of Chemical Reaction Patterns Associated with Permeable Fault Zones . . . .2 Numerical Simulation Issue Arising from Using the Explicit Dynamic Relaxation Method to Solve a Quasi-Static Problem 186 8. . . . . .3 An Upscale Theory of Particle Simulation for Two-Dimensional Quasi-Static Problems . .2 Some Numerical Simulation Issues Associated with the Particle Simulation Method . . . . . 138 6. . .3 Numerical Simulation Issue Stemming from the Loading Procedure Used in the Particle Simulation Method . . . 126 6. . . . .2 A Decoupling Procedure for Removing the Coupling between Reactive Transport Equations of Redox Chemical Reactions . . . .4. . . . . . .1 Numerical Simulation Issue Caused by the Difference between an Element and a Particle . . . . . . . . . . . . . . . . . . . . . . . . . 121 6. . . . .1 Key Factors Controlling Mineral Precipitation Patterns in a Focusing and Mixing System Involving Two Reactive Fluids . . . . . . . . . . . . . . . . . . . .3 Veriﬁcation and Application of the Equivalent Source Algorithm . 189 8. . 145 7 An Equivalent Source Algorithm for Simulating Thermal and Chemical Effects of Intruded Magma Solidiﬁcation Problems . . . . . . 123 6. . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . . . . . . . .4. . 155 7. . . . . . . . . . . . . .

. . . .1 Comparison of the Proposed Loading Procedure with the Conventional Loading Procedure . . . . . . . . .4. . . . . . . . . . . .4 Particle Simulation of the Faulting Process Using the Proposed Particle Method . . . . . .4. . . . . . . . . . . . . . . . . . . . . . . 221 References . . . . . . . . . . . . . . . . 210 8. . . . . . . . . . . . . . . . . . . . . . . . . . . . 239 . . . . .xiv Contents 8. . . . . . . . . . . . . . . . . 204 8. . . . . . . . . 216 Summary Statements . . . . . . . . . . . . . . . . . . . . . .4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Particle Simulation of the Folding Process Using Two Similar Particle Models of Different Length-Scales . 227 Index . . . . . . . . . . . . . . . 199 8. . . . . . . . . . . . . . . . . . . . . . . .2 The Similarity Test of Two Particle Samples of Different Length-Scales . . . . . . 201 8. . . . . . . .4. . .4 Test and Application Examples of the Particle Simulation Method . . . . . . . . . . . .

z Zh area of a ﬁnite element species concentration species concentration vector non-zero constant arbitrary constant speciﬁc heat of pore-ﬂuid mass diffusivity acceleration due to gravity reference length medium permeability reference medium permeability in the horizontal direction length of a problem domain Lewis number pressure pressure vector hydrostatic pressure mass ﬂux on the boundary of a ﬁnite element heat ﬂux on the boundary of a ﬁnite element Rayleigh number critical Rayleigh number boundary length of a ﬁnite element temperature temperature vector temporal variable Darcy velocity in the x direction Darcy velocity vector Darcy velocity in the y direction Darcy velocity in the z direction spatial coordinates Zhao number xv . unless otherwise speciﬁed in the monograph. A C C C1 C2 cp D g H K Kh L Le P P P0 qc qT Ra Racritical S T T t u U ν w x.Nomenclature The following symbols are commonly used with the attached deﬁnitions. y.

xvi Nomenclature Z h critical λ λe0 φ ψ ψ ρ0 μ β σ ϕ η ε Subscripts f 0 Superscripts e e ∗ s critical Zhao number thermal conductivity reference thermal conductivity in the horizontal direction porosity stream function shape function vector for the pressure of a ﬁnite element reference density of pore-ﬂuid dynamic viscosity of pore-ﬂuid thermal volume expansion coefﬁcient of pore-ﬂuid stress on the boundary of a ﬁnite element shape function vector for the temperature. species concentration and Darcy velocity of a ﬁnite element permeability ratio of the underlying medium to its overlying folded layer thermal conductivity ratio of the underlying medium to its overlying folded layer penalty parameter associated with the penalty ﬁnite element approach pertaining to pore-ﬂuid pertaining to reference quantities pertaining to equivalent quantities of a porous medium pertaining to quantities in a ﬁnite element level pertaining to dimensionless quantities pertaining to solid matrix .

. With the ever-increasing demand for improving our living standards. As a result of its complicated and complex nature. rules and laws.Chapter 1 Introduction Geoscience is a fundamental natural science discipline dealing with the origin. DOI 10. such as earthquakes. deteriorating. volcanic eruptions. applications of computational science have penetrated almost all engineering ﬁelds: from the topological C. With the rapid advances and developments of modern computer technology. ﬂoods and tornadoes. but also the locations of large concealed natural resources in the deep Earth. Fundamentals of Computational Geoscience. accumulate and analyze the large amount of information and data that are essential to understand each of the controlling processes within the interior of the Earth using modern observation equipment. by which the Earth system operates. (2) gather. it has been recognized that the existing natural resources will be exhausted in the near future and that our living environments are. Zhao et al. instead of simply using the traditional descriptive methods. in fact. biosphere. to mention just a few. algorithms and other numerical techniques for delivering numerical solutions for complicated and complex scientiﬁc and engineering problems. geoscientists need to make greater efforts in the following aspects relevant to solving contemporary geoscience problems: (1) the complicated and complex interactions between multi-scales and multi-processes occurring in the solid Earth. Lecture Notes in Earth Sciences 122. To maintain the sustainable development of our living standards and the further improvement of our living environments. the Earth system not only provides the necessary materials and environment for mankind to live. As a result.1007/978-3-540-89743-9 1. (3) the intimate interplay between the solid Earth. tsunamis. geoscientists must adopt scientiﬁc and predictive methods relevant to conduction of contemporary geoscience research. evolutionary history and behaviour of the planet Earth. C Springer-Verlag Berlin Heidelberg 2009 1 . an inevitable and challenging task that geoscientists are now confronting is how accurately to predict not only the occurrences of these natural disasters. experimental instruments and information processing techniques. instead of simply describing and observing geoscience phenomena. For this reason. It is this intimate interplay that controls the global behaviour of the Earth system. geoscientists must study the processes. hydrosphere and atmosphere. Computational science is a modern technological science discipline dealing with the development and application of numerical methods. procedures. measurement tools. but also brings many types of natural disasters. Speciﬁcally.

b. using the newly-developed research methodology associated with computational geoscience. Zhao et al. (4) ﬂuid-rock/chemical interaction associated with ore body formation processes (Steefel and Lasaga 1994. both theoretically and practically. Ord and Sorjonen-Ward 2003. e. Jiang et al. a fundamental and theoretical framework for the computational geoscience discipline has been established. Zhao et al. This enables many hitherto unsolvable geoscience problems to be solved. 1999c. Gow et al. Liu et al. 2004. 2003a. Murphy et al. scientiﬁc manner. 2000b. 2002a. 2006a. it can be used effectively and efﬁciently to simulate the processes involved in complicated scientiﬁc and engineering problems in a scientiﬁc and predictive manner. 2002b. from the collapse simulation of a concrete beam to that of a huge double curvature arch dam. 2002. c). 1999d. Zhao et al. chemistry. 2008b. mathematics. considerable development driven from the need to understand the controlling mechanisms behind ore body formation and mineralization in hydrothermal and igneous systems within the upper crust of the Earth (Garven and Freeze 1984. Schaubs and Zhao 2002. 2005a. from the optimal layout design of a pipeline to that of a large-scale underground tunnel. which has been established through applying the well-developed computational science discipline to solve geoscience problems occurring in nature. 2000a. 2007. in the past decade (Zhao et al. Zhao et al. 2002.1 Characteristics of Computational Geoscience Computational geoscience is a newly-developed discipline. Lin et al. e. (3) pore-ﬂuid ﬂow focusing within permeable faults (Obdam and Veling 1987. evolution and behaviour of the Earth system in a predictive. Oliver et al. 1. 2008a. 2005. computational science is a natural supplier to meet the demands of geoscientists in solving contemporary geoscience problems. 2008a). 1998a. to name just a few. The ultimate aim of computational geoscience is to deal with the origin. 2002. 2003c. 2005b. 1997. 1997a. 2007a. geochemistry. 1997. Ord et al. 2006d. 2003b. f) and (5) convective ﬂow of pore-ﬂuid within three-dimensional permeable faults (Zhao et al. 2000c. 1999a. Hobbs et al. Sheldon and Ord 2005. computational geoscience has achieved. 2001c. This means that the computational geoscience discipline is of multi-disciplinary nature crossing many ﬁelds of science. 2001b. (2) ore body formation and mineralization within hydrothermal systems (Zhao et al. 2008d. Raffensperger and Garven 1995. 2003). Gow et al. 2008). some typical examples of applying the newly-developed research methodology to deal with geoscience problems are as follows: (1) the convective ﬂow of pore-ﬂuid within the upper crust of the Earth (Phillips 1991. 1999. 2003. 1999b. computational geoscience. It is this demand and supply relationship that has created a brand new discipline. d. 1998b. Nield and Bejan 1992. physics. Sorjonen-Ward et al.2 1 Introduction optimization of a tooth brush to that of a giant aircraft. . biology and numerical techniques together. 2000. 2001. As a result. Zhao et al. Since computational science is a comprehensive discipline bringing geology. In this sense. Ord et al. Yang 2006). 1998a. 1997a. Zhao et al. 2006c). Zhang et al. Under the stimulus of an ever-increasing demand for natural mineral resources. geophysics. For instance. in the past decade. Zhao et al. 2002. 2002. Doin et al. Zhao et al. McLellan et al. Zimmerman 1996.

Since other unimportant. This conceptual model is used to represent the main characteristics of the real geoscience problem. The conceptual model cannot be over simpliﬁed. Bear and Bachmat 1990. Zhao et al.2 Basic Steps Associated with the Research Methodology of Computational Geoscience 3 1.1. since then it cannot be used effectively to reﬂect the main dynamic processes and mechanisms of the real problem. the conceptual model for the given problem can be . As understanding grows more detail may be added if necessary.2 The Mathematical Model of a Geoscience Problem Using three fundamental principles. This is the key step in the process of using the research methodology of computational geoscience to solve a problem. 1. The appropriate research methodology of computational geoscience is usually comprised of the following four main steps: (1) the establishment of a conceptual model for a given geoscience problem. Phillips 1991. On the other hand. processes and factors are neglected.2 Basic Steps Associated with the Research Methodology of Computational Geoscience Generally speaking. rather than to describe the observed geological phenomenon itself. Nield and Bejan 1992. (2) the establishment of a mathematical model for the given problem. Due to the multiple processes and multiple scales involved in a typical geoscience problem. (3) the construction of a numerical simulation model for the given problem and (4) the graphical display of the numerical results obtained from the numerical simulation. The fundamental principle involved in establishing the conceptual model for the problem is that the details of the conceptual model should depend on both the lengthscale and time-scale of the problem. the research methodology of computational geoscience is a comprehensive research methodology. the initial conceptual model of the problem is somewhat simpliﬁed. for then unnecessary problems for both the theoretical analysis and the numerical simulation may arise and it may become difﬁcult to unravel which parts of the description of the problem are important. only the major controlling dynamic processes and mechanisms associated with the problem need to be considered during the initial establishment of the conceptual model. 1997a).1 The Conceptual Model of a Geoscience Problem Based on extensive data and information obtained from ﬁeld and laboratory investigations of a geological phenomenon. or at least less critical. the conservation of momentum and the conservation of energy.2. the conceptual model cannot be over complicated. which is formed by combining ﬁeld observation. numerical simulation and ﬁeld validation. as well as the related physical and chemical laws (Bear 1972. theoretical analysis. 1. namely the conservation of mass. a conceptual model is established. The primary aim of using this research methodology is to investigate the dynamic processes and mechanisms involved in an observed geological phenomenon.2. which reﬂects the geometrical architecture and main processes associated with the phenomenon.

Zhao et al. If possible. due to the approximate nature of a numerical method. a numerical solution can be obtained for the problem. most geoscience problems are coupled across both multiple processes and multiple scales. the numerical simulation model of a geoscience problem can be also called the discretized type of mathematical model. it is necessary to investigate the solution characteristics of the partial differential equations through a theoretical analysis. as well as the conditions under which such characteristics can occur. This benchmark solution is valuable and indispensable for the veriﬁcation of both the numerical algorithm and the computer code.3 The Numerical Simulation Model of a Geoscience Problem From a mathematical point of view. which is usually comprised of a set of partial differential equations. Due to these signiﬁcant differences between engineering problems and geoscience problems.2. which are mainly . Both the ﬁnite element method and the ﬁnite difference method are commonly-used discretization methods for numerical simulation of geoscience problems (Zienkiewicz 1977. a geoscience problem commonly has both large length-scale and large time-scale characteristics. if not impossible. For example. so that the theoretical solution. numerical simulation methods need to be used to ﬁnd approximate solutions for the problem. some theoretical methods can be used to investigate the solution singularity and multiple solution characteristics of the partial differential equations. Through solving the resulting algebraic equations of the discretized system. which in turn is usually comprised of a large number of elements. The geometrical nature and boundary conditions of this benchmark model can be further simpliﬁed. The length-scale of a geoscience problem is commonly measured in either tens of kilometers or even hundreds of kilometers. the theoretical investigation of the solution characteristics associated with the partial differential equations of a problem plays an important role in applying the research methodology of computational geoscience to solve real problems. The basic idea behind these numerical methods is to translate the partial differential equations used to describe the geoscience problem in a continuum system. to ﬁnd analytical solutions. can be obtained. commercial computer programs and related algorithms. 1. known as the benchmark solution.4 1 Introduction translated into a mathematical model. into the corresponding algebraic equations in a discretized system. a benchmark model should be established for a particular kind of geoscience problem. which are used to solve the problem that is generally characterised by a complicated geometrical shape and complex material properties. 1998a. It must be pointed out that. In addition. Due to the complex and complicated nature of these equations. Alternatively. This is the key step to ensure the accuracy and reliability of the numerical solution obtained from the numerical simulation of the problem. Compared with an engineering problem. 2006a). it is very difﬁcult. while the time-scale of a typical problem is often measured in several million years or even several tens of million years. To ensure the accuracy and reliability of the numerical simulation solution.

Finally the results must be tested through comparison with existing or new ﬁeld observations. mesh shape. then the established conceptual model of the problem is questionable and therefore needs to be modiﬁed through further reﬁnement of the natural data. algorithm stability. On the contrary. Thus the research methodology of computational geoscience is ﬁrstly established on the basis of ﬁeld observations. Based on these three methods. If the numerical solution is not compatible with the ﬁeld observations.1.2. the reliability of the numerical solution obtained from a newly-developed computer program cannot be guaranteed. the established conceptual model of the geoscience problem can be further used to investigate the fundamental rules associated with this kind of problem. By comparing the numerical solution with ﬁeld observations of the geological phenomenon. if the numerical solution is in accord with the ﬁeld observations. and then goes through theoretical analysis and computational simulation. the correctness of the established conceptual model for the geoscience problem can be tested.3 The Contextual Arrangements of this Monograph In this monograph we use the ﬁnite element method. For this reason. In this regard. so as to be suitable for these problems. To ensure the accuracy and reliability of the computational simulation result for a problem.4 Graphical Display of the Numerical Simulation Results The numerical results obtained from the computer simulation of a geoscience problem are expressed as a large amount of data. cannot be directly used to solve geoscience problems without modiﬁcation. the above-mentioned factors need to be carefully considered in the process of establishing the computational simulation model. we have developed advanced . 1. then the established conceptual model of the geoscience problem is a reasonable interpretation of what may have occurred in nature. the ﬁnite difference method and the particle simulation method as basic numerical methods for dealing with geoscience problems. it is necessary either to develop new computer programs for solving geoscience problems or to modify the existing commercial computer programs. time-step and other factors. Otherwise. This fundamental research methodology requires that the recognition of a natural phenomenon start from ﬁeld observations. the research methodology of computational geoscience can provide an effective scientiﬁc-judging method for solving many controversial problems in the ﬁeld of geoscience. resulting in a circular iteration. In this case. which can be viewed using modern technologies of computer graphical display.3 The Contextual Arrangements of this Monograph 5 designed for solving engineering problems. 1. and be completed through further tests arising from ﬁeld observations. originally designed for solving engineering problems. A newly-developed computer program needs to be veriﬁed through the corresponding benchmark problem before it is used to solve any real geoscience problems. It is noted that the solution reliability of a geoscience problem is strongly dependent on algorithm convergence.

the morphological evolution of chemical dissolution fronts in ﬂuid-saturated porous media has been appropriately simulated. in Chap. 7. . to solve coupled problems between medium deformation. 1995). heat transfer and non-equilibrium redox chemical reactions in ﬂuid-saturated porous rocks. Finally. Fluid Dynamics International. The geoscience problems to be considered are closely related to ore body formation and mineralization in hydrothermal systems within the upper crust of the Earth. 1997) and Fast Lagrangian Analysis of Continua (FLAC.6 1 Introduction numerical procedures and algorithms to tackle the large-scale aspects of geoscience problems. the particle simulation method is extended to solve spontaneous crack generation problems in brittle rocks within the upper crust of the Earth. This algorithm has been used to simulate effectively and efﬁciently the thermal and chemical effects of intruded magma in hydrothermal systems. In this case. in an iterative and alternative manner. 3. 5. The extended particle method has been applied to simulate spontaneous crack generation associated with faulting and folding in large length-scale geological systems. 6. a progressive asymptotic approach procedure is proposed to solve steady-state convective pore-ﬂuid ﬂow problems within the upper crust of the Earth. With this general interface. 8. a term splitting algorithm is developed for dealing with ﬂuid-rock interaction problems that are closely associated with mineral dissolution and precipitation as well as metamorphic processes in ﬂuid-saturated hydrothermal/sedimentary basins of subcritical Zhao numbers. In combination with the ﬁnite element method. an equivalent source algorithm is presented for simulating thermal and chemical effects of intruded magma solidiﬁcation problems. In Chap. In Chap. a segregated algorithm is proposed. for solving chemical-dissolution front instability problems in ﬂuid-saturated porous rocks of critical and supercritical Zhao numbers. In Chap. a decoupling procedure is proposed for simulating ﬂuids mixing. this procedure has been applied to simulate convective pore-ﬂuid ﬂow that often plays an important role in ore body formation and mineralization. the chemical dissolution fronts are stable during their propagation within the reactive mass transport system. In contrast. pore-ﬂuid ﬂow. the two commercial codes have been used. 2. In Chap. In Chap. Itasca Consulting Group. The proposed procedure has been applied to investigate the effects of non-equilibrium redox chemical reactions on the mineralization patterns in hydrothermal systems. Fluid Dynamics Analysis Package (FIDAP. some conclusions are given at the end of the monograph. The arrangements of the forthcoming parts of this monograph are as follows: In Chap. In this situation. 4. heat transfer and reactive mass transport processes in hydrothermal systems. a consistent point-searching interpolation algorithm is proposed to develop a general interface between two commercial computer codes.

.1007/978-3-540-89743-9 2. Fundamentals of Computational Geoscience. if the temperature difference is large enough. This is the mineralization problem closely associated with geophysics and geology. it may trigger natural convection in the ﬂuid-saturated porous medium. then there exists a temperature difference between the top and bottom boundaries of the layer. For example. In geophysics. there exists a vertical temperature gradient in the Earth’s crust. is equal to or greater than the critical Rayleigh number. This kind of natural convection problem has been found in many geoscience ﬁelds. a change in medium temperature may lead to a change in the density of pore-ﬂuid within the medium. there is no natural convection for a small temperature gradient in the porous medium. In this case. Since a natural porous medium is often of a complicated geometry and composed of many C. If the Rayleigh number. heat energy is solely transferred from the high temperature region (the bottom of the horizontal layer) to the low temperature region (the top of the horizontal layer) by thermal conduction. buried nuclear waste and industrial waste in a ﬂuid-saturated porous medium may generate heat and result in a temperature gradient in the vertical direction.Chapter 2 A Progressive Asymptotic Approach Procedure for Simulating Steady-State Natural Convective Problems in Fluid-Saturated Porous Media In a ﬂuid-saturated porous medium. which is directly proportional to the temperature gradient. If a ﬂuid-saturated porous medium has the geometry of a horizontal layer. Lecture Notes in Earth Sciences 122. DOI 10. C Springer-Verlag Berlin Heidelberg 2009 7 . However. and is often called the Horton-Rogers-Lapwood problem. and is heated uniformly from the bottom of the layer. This change can be considered as a buoyancy force term in the momentum equation to determine pore-ﬂuid ﬂow in the porous medium using the Oberbeck-Boussinesq approximation model. the pore-ﬂuid ﬂow circulation due to the natural convection can dissolve soluble minerals in some part of a region and carry them to another part of the region. In this situation. Zhao et al. so that the groundwater may be severely contaminated due to the pore-ﬂuid ﬂow circulation caused by the natural convection. The momentum equation used to describe pore-ﬂuid ﬂow in a porous medium is usually established using Darcy’s law or its extensions. This problem was ﬁrst treated analytically by Horton and Rogers (1945) as well as Lapwood (1948). If this temperature gradient is large enough. natural convection will take place in the porous medium. Since the positive direction of the temperature gradient due to this temperature difference is opposite to that of the gravity acceleration. in geoenvironmental engineering. it will cause regional natural convection in the Earth’s crust.

does not describe the amplitude of the resulting convection motion. Although this difﬁculty can be circumvented by turning a steady-state problem into a transient one (Trevisan and Bejan 1987). The ﬁnite element method needs to deal with a highly nonlinear problem and often suffers difﬁculties in establishing the true non-zero velocity ﬁeld in a ﬂuid-saturated porous medium because the Horton-Rogers-Lapwood problem always has a zero solution as one possible solution for the velocity ﬁeld of the pore-ﬂuid. From the ﬁnite element analysis point of view. For instance. especially when strong constraints. Indeed. b. If the velocity ﬁeld of the pore-ﬂuid used at the beginning of an iteration method is not chosen appropriately. are considered. Riley and Winters 1989. Pillatsis et al. such as impervious or adiabatic boundaries. Therefore. McKibbin and O’Sullivan 1980. Lebon and Cloot 1986. the nonlinear term may produce manifest differences. For this reason. Buretta and Berman 1976. The linear stability theory based on the ﬁrst-order perturbation is commonly used to solve this problem analytically and numerically (Nield 1968. 1999). Kaviany 1984. the Horton-Rogers-Lapwood problem possesses a bifurcation. even if the form of convective motion obtained for low supercritical conditions is often quite similar to the critical disturbance. From the mathematical point of view. (1996) pointed out that: “The linear stability theory. The computed disturbances are correct only for inﬁnitesimal amplitudes.” Since it is the amplitude and the form of natural convective motion that signiﬁcantly affects or dominates the contaminant transport and mineralization in a ﬂuid-saturated porous medium. 1997a. To solve the steady-state Horton-Rogers-Lapwood problem with the full nonlinear term of the energy equation included in the ﬁnite . c). the h-adaptive mesh reﬁnement (Cook et al. a progressive asymptotic approach procedure has been developed in recent years (Zhao et al. 1989) is based on the asymptotic approach concept and can produce a satisfactory solution with the progressive reduction in the size of ﬁnite elements used in the analysis. Chevalier et al. which was previously and successfully applied to some other ﬁelds of the ﬁnite element method. However.8 2 Simulating Steady-State Natural Convective Problems different materials. there is a deﬁnite need for including the full nonlinear term of the energy equation in the ﬁnite element analysis. Combarnous and Bories 1975. 1972. 1987. then the resulting ﬁnite element solution always tends to zero for the velocity ﬁeld in a ﬂuid-saturated porous medium. The same asymptotic approach concept was also employed to obtain asymptotic solutions for natural frequencies of vibrating structures in a ﬁnite element analysis (Zhao and Steven 1996a. Joly et al. the direct inclusion of the full nonlinear term of the energy equation in the steady-state Horton-Rogers-Lapwood problem would result in a formidable difﬁculty. Nield and Bejan 1992. it is highly desirable to develop a numerical procedure to directly solve the steady-state Horton-Rogers-Lapwood problem. 1976. The developed progressive asymptotic approach procedure is based on the concept of an asymptotic approach. numerical methods are always needed to solve the aforementioned problems. it is often unnecessary and computationally inefﬁcient to obtain a steady-state solution from solving a transient problem. in which the nonlinear term of the heat disturbance equation has been neglected. 1998a). Palm et al. Caltagirone 1975. Islam and Nandakumar 1990. Phillips 1991.

1 Governing Equations of the Problem 9 element analysis. μ is the dynamic viscosity of the pore-ﬂuid. ∂x ∂y u= Kx μ Ky μ − ∂P + ρ f gx .6) where u and v are the horizontal and vertical velocity components of the pore-ﬂuid in the x and y directions respectively.2. . T is the temperature of the porous material.3) and (2. λex = φλfx + (1 − φ)λsx . in which the y axis is upward in the vertical direction and coincides with the principal direction of medium permeability as well as that of medium conductivity.1 Governing Equations of the Problem For a two-dimensional ﬂuid-saturated porous medium. the asymptotic approach concept needs to be combined with the ﬁnite element method in a different fashion (Zhao et al. λey = φλfy + (1 − φ)λsy . the governing equations of a natural convection problem. 1997a).2) v= − (2. (2. K x and K y are the permeabilities of the porous material in the x and y directions respectively. Zhao et al.5) (2. known as the steady-state Horton-Rogers-Lapwood problem (Nield and Bejan 1992.1) (2. c p is the speciﬁc heat of the pore-ﬂuid. P is the pore-ﬂuid pressure. ρ f is the density of the pore-ﬂuid. ∂y = λex ∂2T ∂2T + λey 2 . (2. φ and βT are the porosity of the porous material and the thermal volume expansion coefﬁcient of the pore-ﬂuid. for incompressible pore-ﬂuid can be expressed as ∂ν ∂u + = 0. λ f x and λsx are the thermal conductivities of the pore-ﬂuid and rock mass in the x direction. 1997a). 2.1). 2 ∂x ∂y (2.4) ρ f = ρ f 0 [1 − βT (T − T0 )]. gx and g y are the gravity acceleration components in the x and y directions. if Darcy’s law is used to describe pore-ﬂuid ﬂow and the Oberbeck-Boussinesq approximation is employed to describe a change in pore-ﬂuid density due to a change in pore-ﬂuid temperature.4) are derived under the assumption that the porous medium considered is orthotropic. λ f y and λsy are the thermal conductivities of the pore-ﬂuid and rock mass in the y direction. (2.2).3) ρ f 0c p u ∂T ∂T +ν ∂x ∂y (2. ∂x ∂P + ρ f gy . ρ f 0 and T0 are the reference density and temperature. It is noted that Eqs. (2.

λe0 λ∗ = ey (2. H is a reference length and P0 is the static pore-ﬂuid pressure. Kh λ∗ = ex λex . Ra is the Rayleigh number.11) ∗ v∗ = K y − (2. deﬁned in this particular case as Ra = (ρ f 0 c p )ρ f 0 gβΔTKh H . ∂x∗ ∂ P∗ + RaT ∗ e2 . P ∗ and T ∗ are the dimensionless excess pressure and temperature. λe0 Kx . (2. (2. ∂ y∗ (2. λe0 P∗ = Kh ρ f 0c p (P − P0 ). Kh ν∗ = Hρ f 0 c p ν.2). (2.3) and (2. (2.1). ΔT = Tbottom − T0 is the temperature difference between the bottom and top boundaries of the porous medium.4).9) where x ∗ and y ∗ are the dimensionless coordinates.1). Substituting the above dimensionless variables into Eqs. (2.7) u∗ = Hρ f 0 c p u. ex ey ∂x∗ ∂y ∂ x ∗2 ∂ y ∗2 (2. the following dimensionless variables are deﬁned: x∗ = x .3) and (2.4) yields the following dimensionless equations: ∂u ∗ ∂ν ∗ + ∗ = 0. λe0 is a reference conductivity coefﬁcient of the porous medium. ΔT (2. H y∗ = y .13) where e is a unit vector and e = e1 i+e2 j for a two-dimensional problem. H T∗ = T − T0 . K h is a reference medium permeability coefﬁcient in the horizontal direction. λe0 (2.10 2 Simulating Steady-State Natural Convective Problems In order to simplify Eqs.12) u∗ ∂T ∗ ∂T ∗ ∂2T ∗ ∂2T ∗ + ν ∗ ∗ = λ∗ + λ∗ .10) ∗ u∗ = K x − (2.8) ∗ Kx = ∗ Ky = Ky . ∗ ∂x ∂y ∂ P∗ + RaT ∗ e1 . u ∗ and v ∗ are the dimensionless velocity components in the x and y directions respectively. (2. μλe0 λey .2).14) . μλe0 (2.

y ∗ ) = ϕ T Ve .17) (2. (2. (2.12) and (2.19) where X and Y are the column vectors of nodal coordinate components in the x and y directions of the global coordinate system respectively. T ∗ (x ∗ . (2. η) = ϕ(ξ.16) (2. y ∗ ) = Ψ T Pe . (2.22) A .2.18) where Ue . η). (2. ϕ is the column vector of the interpolation functions for the dimensionless velocity and temperature ﬁelds within the element. excess pressure and temperature of the element. pressure and temperature as basic variables.15). it is assumed that ϕ is identical to Ψ in the following numerical analysis. y ∗ ) = ϕ T Te .13) can be expressed. η) = Ψ(ξ. N is the column vector of the coordinate mapping function of the element. (2.21) ϕϕ T Ue dA + A A ∗ ϕ Kx ∂Ψ T e P dA + ∂x∗ (2. Ψ is the column vector of the interpolation functions for the excess pressure within the element. The global coordinate components within the element can be deﬁned as x ∗ = NT X. Ve .16). Eqs. Eqs.10). as follows: Ψ A ∂ϕ T e U dA + ∂x∗ Ψ A ∂ϕ T e V dA = 0.18).20) where ξ and η are the local coordinate components of the element. v∗ (x ∗ . Using the Galerkin weighted-residual method. pressure and temperature ﬁelds at the elemental level can be expressed as u ∗ (x ∗ . For a typical 4-node quadrilateral element.11). (2. ∂ y∗ ∗ ϕ K x Raϕ T Te e1 dA = 0. (2. (2. (2. the velocity. y ∗ ) = ϕ T Ue .11). the following relationships exist: N(ξ.12) and (2. (2. (2. P ∗ (x ∗ . Pe and Te are the column vectors of the nodal velocity.13) can be discretized using the conventional ﬁnite element method (Zienkiewicz 1977.10). For the 4-node quadrilateral element.17) and (2.15) (2. (2. Based on the isoparametric element concept. y ∗ = NT Y. 1997a). with consideration of Eqs.2 Finite Element Formulation of the Problem By considering the dimensionless velocity.2 Finite Element Formulation of the Problem 11 2. Zhao et al.

24) Using the Green-Gauss theorem and the technique of integration by parts. Ce . ∂ x ∗ ex ∂ x ∗ Fe = x S σx∗ ϕd S. (2.23) and (2. Be . ∂x∗ Be = x ∗ ϕ K x Raϕ T e1 dA. Be Ce .27) where Ue and Ve are the nodal dimensionless velocity vectors of the element in the x and y directions respectively.23) A ϕu ∗ A ∂ϕ T e T dA+ ∂x∗ ϕv ∗ A ∂ϕ T e T dA− ∂ y∗ A ∂ 2ϕT e T dA− ∂ x ∗2 A ϕ λ∗ ey ∂ 2ϕT e T dA = 0.24) can be expressed in a matrix form as follows: ⎤⎧ Me 0 −Be −Ae ⎪ Ue x x ⎪ ⎢ 0 Me −Be −Ae ⎥ ⎨ V e y y ⎥ ⎢ ⎣ 0 0 Ee 0 ⎦ ⎪ Te ⎪ e ⎩ e e P 0 0 Cx Cy ⎡ ⎧ e ⎫ ⎪ Fx ⎪ ⎪ e ⎪ ⎨ ⎬ Fy = . (2. (2.25) A A S A ϕ λ∗ ey A S ∗ ϕq y n y d S = 0.30) .26) ∗ ∗ where qx and q y are the dimensionless heat ﬂuxes on the element boundary of a unit normal vector. A and S are the area and boundary length of the element. e ⎪ ⎪G ⎪ ⎪ ⎪ ⎪ ⎭ ⎩ ⎭ 0 ⎫ ⎪ ⎪ ⎬ (2. These matrices and vectors can be derived and expressed as follows: Ae = x ∂ϕ ∗ T K Ψ dA. ϕϕ T Ve dA + A A ∗ ϕKy ∂Ψ T e P dA + ∂ y∗ ϕ λ∗ ex (2. the terms involving the second derivatives in Eq.29) Ae = y A Be = y A ∗ ϕ K y Raϕ T e2 dA. Fe . A A Ce = x Ψ A ∂ϕ T dA. ∂ y∗ (2. ∂x∗ x ∂ϕ ∗ T K Ψ dA. (2.24) can be rewritten as ϕ λ∗ ex ∂ 2ϕ T e T dA = − ∂ x ∗2 ∂ 2ϕ T e T dA = − ∂ y ∗2 ∂ϕ ∗ ∂ϕ T e λ T dA + ∂ x ∗ ex ∂ x ∗ ∂ϕ ∗ ∂ϕ T e λ T dA + ∂ y ∗ ey ∂ y ∗ ∗ ϕqx n x d S = 0. ∂x∗ (2. (2. (2. Te and Pe are the nodal dimensionless temperature and pressure vectors of the element. Ae .28) ∂ϕ T dA.12 2 Simulating Steady-State Natural Convective Problems ∗ ϕ K y Raϕ T Te e2 dA = 0. ∂ y ∗2 (2.21). Note that Eqs. Ee and Me are the x y x y x y property matrices of the element. ∂ y∗ y ∂ϕ T dA.22). Ce = y Ψ A De (u ∗ ) = x ϕu ∗ A Le = x A ∂ϕ ∗ ∂ϕ λ dA. Ae . n. (2. Fe and Ge are the dimensionless nodal load x y vectors due to the dimensionless stress and heat ﬂux on the boundary of the element.

(2. Ge = − x y x y Me = A q ∗ ϕd S.31) (2.38) Be = Ae = Ce = (2.35) (2. the following equation exists: Ce Ue + Ce Ve = −εMp Pe . x y ε p (2. (2. ΔT λe0 σ∗ = Kh ρ f 0c p σ. q∗ = H q. Me 0 . a prediction for the initial velocities of an element is needed to have this matrix evaluated. A and S are the area and boundary length of the element.34) can be rewritten as 1 Pe = − M−1 (Ce Ue + Ce Ve ). ∂ y ∗ ey ∂ y ∗ Fe = y S ∗ σ y ϕd S. Ve . This is the main motivation for proposing a progressive asymptotic approach procedure in the next section. p ε M = e Ue F Te = Fe .34) Substituting Eq. Fe = Ce x Ce y . Thus. matrix Ee is dependent on the velocity components of the element. From the penalty ﬁnite element approach (Zienkiewicz 1977). μλe0 (2. Ue = F Ae x Ae y Ue .32) Ee = De (u ∗ ) + De (v ∗ ) + Le + Le . Ψ is the shape function vector for the pressure of the element. σ and q are the stress and heat ﬂux on the boundary of the element. x y Equation (2.27) yields the following equation in the elemental level: Qe −Be 0 Ee where 1 e Qe = M + Ae (Me )−1 (Ce )T . S ϕϕ T dA. 0 Me Be x Be y .35) into Eq.2 Finite Element Formulation of the Problem 13 De (v ∗ ) = y ϕν ∗ A ∂ϕ T dA. (2. Ge (2. ∂x∗ Le = y A ∂ϕ ∗ ∂ϕ λ dA. (2. It is noted that since the full nonlinear term of the energy equation in the HortonRogers-Lapwood problem is considered in the ﬁnite element analysis.37) Fe x Fe y .39) .33) where ϕ is the shape function vector for the temperature and velocity components of the element.2.36) (2.

but large enough to prevent the resulting matrix problem from becoming too ill-conditioned to solve. the ﬁnite element method will lead to . the key issue of obtaining a non-zero pore-ﬂuid ﬂow solution for the Horton-RogersLapwood problem is to choose the initial velocity ﬁeld of pore-ﬂuid correctly. F and G are global nodal load vectors of the system. the asymptotic approach concept (Cook et al. this parameter must be chosen small enough to approximate ﬂuid incompressibility well. Thus. Clearly. in the Horton -Rogers-Lapwood problem. Since Equation (2.3 The Progressive Asymptotic Approach Procedure for Solving Steady-State Natural Convection Problems in Fluid-Saturated Porous Media To solve the steady-state Horton-Rogers-Lapwood problem with the full nonlinear term of the energy equation included in the ﬁnite element analysis. α. If the gravity acceleration is assumed to tilt at a small angle. (2. a solution to the original Horton-Rogers-Lapwood problem can be obtained. the modiﬁed Horton-Rogers-Lapwood problem asymptotically approaches the original one and as a result. then a non-zero velocity ﬁeld in a ﬂuid-saturated porous medium may be found using the ﬁnite element method. 1989. For the purpose of obtaining an accurate solution. The resulting non-zero velocity ﬁeld can be used as the initial velocity ﬁeld of the pore-ﬂuid to solve the original Horton-Rogers-Lapwood problem with the tilted small angle being zero. (2. By assembling all elements in a system. the ﬁnite element equation of the system can be expressed in a matrix form as Q 0 −B E(U) UF T = F G . either the successive substitution method or the Newton-Raphson method can be used to solve this equation. B and E are global property matrices of the system. UF and T are global nodal velocity and temperature vectors of the system.40) It needs to be pointed out that ε is a penalty parameter in Eq. This forms two basic steps of the progressive asymptotic approach procedure. two kinds of problems need to be progressively solved in the ﬁnite element analysis. A (2. Based on the basic idea behind the progressive asymptotic approach procedure. in which the gravity acceleration is tilted a small angle. One is the modiﬁed Horton-Rogers-Lapwood problem. Zhao and Steven 1996a.41) where Q.41) is nonlinear. 1997a). If the initial velocity ﬁeld is not correctly chosen.14 2 Simulating Steady-State Natural Convective Problems Me = p Ψ Ψ T dA. b. and another is the original Horton-Rogers-Lapwood problem. 2.37). c) needs to be used in a progressive fashion (Zhao et al. the basic idea behind the progressive asymptotic approach procedure is that when the small angle tilted by the gravity acceleration approaches zero.

45) lim S(Ra. α) = S(Ra.42) where S(Ra. needs to be solved. In order to overcome this difﬁculty. αi+1 ) αn →0 (i = 1. S(Ra. α) is the solution for the modiﬁed Horton-Rogers-Lapwood problem.1.. α). a modiﬁed Horton-Rogers-Lapwood problem. n − 1). S is any variable to be solved in the original Horton-Rogers-Lapwood problem. 0) of good accuracy in the ﬁnite element analysis. From the theoretical point of view. This leads to the following processes mathematically: αi →αi+1 lim S(Ra. S(Ra. α) in an explicit manner. 0) is a solution for the original Horton-Rogers-Lapwood problem. 2. 2. αi+1 = 1 αi . α) to that of S(Ra. α should be chosen big enough to eliminate the strong dependence of S(Ra.. .. in which the gravity acceleration is assumed to tilt a small angle α. The reason for this is that the smaller the value of α. a very small α usually leads to a zero velocity ﬁeld due to any inappropriate choice for the initial velocity ﬁeld of pore-ﬂuid. α) on the initial velocity ﬁeld of pore-ﬂuid.43) (2. where n is the total step number for α approaching zero.n and R are dependent on the nature of a problem to be analysed. Supposing the original Horton-RogersLapwood problem has a Rayleigh number (Ra) and that the non-zero solution for the modiﬁed Horton-Rogers-Lapwood problem is S(Ra. αi ) = S(Ra.3 Solving Steady-State Natural Convection Problems in Fluid-Saturated Porous Media 15 a zero pore-ﬂuid ﬂow solution for natural convection of pore-ﬂuid. the more sensitive the solution S(Ra. 0) numerically since it is very difﬁcult and often impossible to express S(Ra. For the purpose of using a big value of α and keeping the ﬁnal solution S(Ra. the question which must be answered is how to choose α so as to obtain an accurate non-zero solution. . it is necessary to ﬁnd out S(Ra.. R α1 = α.2. R is the rate of αi approaching αi+1 . 0). it is desirable to choose α as small as possible. α cannot be chosen too small because the smaller the value of α. S(Ra. 0) in a progressive asymptotic manner.. if S(Ra. However. α) could be expressed as a function of α explicitly. Generally. α) needs to approach S(Ra. even though the Rayleigh number is high enough to drive the occurrence of natural convection in a ﬂuid-saturated porous medium.44) (2. αn ) = S(Ra. (2. To avoid this phenomenon. (2. the closer the characteristic of S(Ra. From the ﬁnite element analysis point of view.. the values of α. in practice. 0). Thus. This enables a more accurate solution S(Ra. it is possible to ﬁnd a nonzero solution for the original Horton-Rogers-Lapwood problem by taking a limit of S(Ra. 0) to be obtained in the computation. This process can be mathematically expressed as follows: α→0 lim S(Ra. α) to the initial velocity ﬁeld of pore-ﬂuid. 0) would follow immediately. 0). As a result. It is noted that in theory. S(Ra. 0). α) when α approaches zero. as clearly shown in Fig.

(at y ∗ = 0). Although the existing solutions (Phillips 1991.13) are considered in this section.16 2 Simulating Steady-State Natural Convective Problems S(Ra. The boundary conditions of the benchmark problem are expressed using the dimensionless variables as follows: u ∗ = 0. Without losing generality. for α in the range of 1–5◦ and R in the range of 5–10. an analytical solution is needed for a benchmark problem. (2. the geometry and boundary conditions of which can be exactly modelled by the ﬁnite element method.46) (2.10). 2.11). α) can asymptotically approach S(Ra. (2. 0) in one step or two steps. 2. 2.47) (2. numerical experience has shown that 1◦ ≤ α ≤ 5◦ . (2. ∂T ∗ =0 ∂x∗ T∗ = 1 T∗ = 0 (at x ∗ = 0 and x ∗ = L ∗ ). a benchmark problem of any rectangular geometry is constructed and shown in Fig.4 Derivation of Analytical Solution to a Benchmark Problem In order to verify the applicability of the progressive asymptotic approach procedure for solving the Horton-Rogers-Lapwood convection problem.48) . For this purpose. (at y ∗ = 1). Nield and Bejan 1992) for a horizontal layer in porous media can be used to check the accuracy of a ﬁnite element solution within a square box with appropriate boundary conditions.1 The basic concept of the progressive asymptotic approach procedure For solving the steady-state Horton-Rogers-Lapwood problem using the progressive asymptotic approach procedure associated with the ﬁnite element method. (2.2. This indicates the efﬁciency of the present procedure. it is highly desirable to examine the progressive asymptotic approach procedure as extensively as possible. Therefore. v∗ = 0. S(Ra. v∗ = 0. the dimensionless governing equations given in Eqs.12) and (2. 5 ≤ R ≤ 10 and 1 ≤ n ≤ 2 leads to acceptable solutions. α) α = α1 α = α2 α = αn α is too small 0 Ra critical Ra Fig.

For ease of deriving an analytical solution to the benchmark problem.52) ∂T ∗ ∂T ∗ ∂2T ∗ ∂2T ∗ + ν∗ ∗ = + .51) (2. in which L is the real length of the problem domain in the horizontal direction.. (2.54) are linear.. 2. (2..49) (2.50).. (2. ∂x∗ ∂y u∗ = − v∗ = − u∗ ∂ P∗ + RaT ∗ e1 . . (2. m = 1. 3.2.12) and (2. (2.52) are reduced to the following two equations: ∂ 2Ψ ∂ 2Ψ ∂T ∗ + ∗2 = −Ra ∗ . 2. ∂ y∗ (2. Eqs. ∂x∗ ∂y ∂ x ∗2 ∂ y ∗2 Using the linearization procedure for temperature gradient and a dimensionless stream function Ψ simultaneously. ∂x∗ ∂ P∗ + RaT ∗ e2 . (2.). ∂ x ∗2 ∂y ∂x ∂Ψ ∂2T ∗ ∂2T ∗ = + .49). (2. solutions to Ψ and T ∗ are of the following forms: Ψ = f (y ∗ ) sin q x∗ L∗ (q = mπ.53) and (2.. (2. As a result.54) Since Eqs. ∗ ∗2 ∂x ∂x ∂ y ∗2 (2.4 Derivation of Analytical Solution to a Benchmark Problem 17 1 y* x* 0 L* Fig.10).53) (2.2 Geometry of a benchmark problem where L ∗ is a dimensionless length in the horizontal direction and L ∗ = L/H . Eqs. This means that K x = K y = K h and λex = λey = λe0 . it is assumed that the porous medium under consideration is ﬂuid-saturated and isotropic.11).50) (2.13) can be further simpliﬁed as follows: ∂ν ∗ ∂u ∗ + ∗ = 0.55) .51) and (2.

However. the minimum Rayleigh number is 4π 2 . which occurs when m = 1 and n = 1.. which occurs when n = 1 and m = L ∗ .59) is a linear.48) can be expressed as f (y ∗ ) = sin(r y ∗ ) (r = nπ. n = 1. it is noted that the non-zero solution satisfying both Equation (2.. (2. For this benchmark problem. .53) and (2.56) into Eqs.46).. Racritical .18 2 Simulating Steady-State Natural Convective Problems T ∗ = θ (y ∗ ) cos q x∗ L∗ + (1 − y ∗ ) (q = mπ. the mode shapes for the stream function and related dimensionless variables corresponding to the critical Rayleigh number can be derived and expressed as follows: Ψ = C1 sin mπ ∗ x sin(nπ y ∗ ).57) and (2.59) It is immediately noted that Equation (2.59) is derived and expressed as L∗ 2 q 2 n2 ∗ m Ra = r + ∗ L + ∗ π2 = q L m L (m = 1.. 2.58) Combining Eqs.. . L∗ (2. (2. (2. For the purpose of ﬁnding out a non-zero solution.55) and (2.. (2.). n = 1. 2.47) and (2.58) leads to an equation containing f (y ∗ ) only: f I V (y ∗ ) − 2 q L∗ 2 f (y ∗ ) − q L∗ 2 Ra − q L∗ 2 f (y ∗ ) = 0..61) It can be observed from Eq.. if L ∗ is not an integer. (2.56) Substituting Eqs. the minimum Rayleigh number is (L ∗ + 1 / L ∗ )2 π 2 . 3. (2..59) and the boundary conditions in Eqs.54) yields the following equations: f (y ∗ ) − q L∗ 2 f (y ∗ ) = 2 q Raθ (y ∗ ).). 3....57) q q f (y ∗ ) = − ∗ L L∗ θ (y ∗ ) + θ (y ∗ ). 3. (2.. 2. 3.60) Using this equation. (2. . (2. Since the minimum Rayleigh number determines the onset of natural convection in a ﬂuid-saturated porous medium for the Horton-Rogers-Lapwood problem. the condition under which the non-zero solution exists for Eq.62) . (2. it is often labelled as the critical Rayleigh number..61) that in the case of L ∗ being an integer. 2.. (2. ... L∗ (2. 2 (2... m = 1.). homogeneous ordinary differential equation so that it has a zero trivial solution...

L∗ (2. A rectangular domain of L ∗ = 1. L∗ L∗ (2.T * = 1 L* = 1.5 Veriﬁcation of the Proposed Progressive Asymptotic Approach Procedure Associated with Finite Element Analysis Using the analytical solution derived for a benchmark problem in the last section.65) nL∗ mπ ∗ Racritical x cos(nπ y ∗ ) − C1 cos (1 − y ∗ )2 + C2 . 2.2. C1 is a non-zero constant and C2 is an arbitrary constant. This implies that if rectangular valleys are ﬁlled with porous media. they may have different critical Rayleigh numbers when their ratios of length to height are different. L∗ Racritical (2. m L∗ 2 (2.5 Fig. It is interesting to note that since Racritical is a function of L ∗ . the proposed progressive asymptotic approach procedure associated with the ﬁnite element analysis for solving the Horton-Rogers-Lapwood problem in a ﬂuid-saturated porous medium is veriﬁed in this section.3.T * = 0 u* = 0 ∂T =0 ∂n * u* = 0 ∂T * =0 ∂n 1 y* x* 0 v* = 0. it can vary with a non-integer L ∗ . The critical Rayleigh number for the test problem considered is 169π 2 / 36. n and Racritical are dependent on whether L ∗ is an integer or not. 2.5 is considered in the calculation.63) mπ mπ ∗ C1 cos x sin(nπ y ∗ ). As shown in Fig. the problem domain is discretized into v* = 0.3 Finite element mesh for the benchmark problem .64) C1 mπ ∗ cos x sin(nπ y ∗ ) + (1 − y ∗ ).5 Veriﬁcation of the Proposed Progressive Asymptotic Approach 19 u ∗ = nπC1 sin v∗ = − T∗ = −√ P∗ = mπ ∗ x cos(nπ y ∗ ).66) where the values of m. 2.

temperature and pressure modes respectively. which enables the region in the vicinity of problem boundaries to be modelled using ﬁnite elements of small sizes. 2. That is to say. the maximum error in the numerical solutions is less than 2%. Figures 2. As stated previously. the Horton-Rogers-Lapwood problem belongs mathematically to a bifurcation problem. Compared with the analytical solutions. there is a need to explain why both the analytical and the numerical solutions for the pore-ﬂuid ﬂow are non-symmetric. whereas the plots below are numerical solutions for the problem. At this point.4 Comparison of numerical solution with analytical solution (Dimensionless velocity) . 2. The mesh gradation technique. It is observed from these results that the numerical solutions from the progressive asymptotic approach procedure associated with the ﬁnite element method are in good agreement with the analytical solutions. although the geometry and boundary conditions for the problem are symmetric.4. This demonstrates the usefulness of the present progressive asymptotic approach procedure when it is used to solve the steady-state Horton-Rogers-Lapwood problems. n = 2 and R = 5. if the Rayleigh number of the problem is less than the critical Rayleigh number. the solution resulting from any small disturbance or perturbation converges to the trivial Fig. has been employed to increase the solution accuracy in this region. 2.20 2 Simulating Steady-State Natural Convective Problems 864 nine-node quadrilateral elements of 3577 nodes in total. In these ﬁgures. stream function.7 show the comparison of numerical solutions with analytical ones for dimensionless velocity.6 and 2. the plots above are analytical solutions.5. The trivial solution for the pore-ﬂuid ﬂow of the problem is zero. The following parameters associated with the progressive asymptotic approach procedure are used in the calculation: α = 5o .

the solution for the pore-ﬂuid ﬂow is non-zero and the system is in an unstable state. However. Since the main purpose of this study is to ﬁnd out the non-trivial solution for problems having a high Rayleigh number. symmetric) and the system is in a stable state. In this situation. the solution dependence on the amplitude of the initially-tilted small angle can be avoided by making this angle approach zero . In this case.2.5 Veriﬁcation of the Proposed Progressive Asymptotic Approach 21 Fig. if the Rayleigh number of the problem is equal to or greater than the critical Rayleigh number. 2. of course. even though the system considered is symmetric. 2. the solution resulting from any small disturbance or perturbation may lead to a non-trivial solution. This is why gravity is ﬁrstly tilted a small angle away from vertical and then gradually approaches and is ﬁnally restored to vertical in the proposed progressive asymptotic approach procedure. It is the small perturbation that makes the non-trivial solution non-symmetric.3. the solution for the pore-ﬂuid ﬂow is zero (and. as addressed in Sect.5 function) Comparison of numerical solution with analytical solution (Dimensionless stream solution. a small disturbance or perturbation needs to be applied to the system at the beginning of a computation. Ra ≥ Racritical . In addition.

22

2

Simulating Steady-State Natural Convective Problems

Fig. 2.6 Comparison of numerical solution with analytical solution (Dimensionless temperature)

in a progressive asymptotic manner. However, since there are two possible nonsymmetric solutions for convective pore-ﬂuid ﬂow, namely a clockwise convective ﬂow and an anti-clockwise convective ﬂow, the solution dependence on the direction of the initially-tilted small angle cannot be avoided.

2.6 Application of the Progressive Asymptotic Approach Procedure Associated with Finite Element Analysis 2.6.1 Two-Dimensional Convective Pore-Fluid Flow Problems

The present progressive asymptotic approach procedure is employed to investigate the effect of basin shapes on natural convection in a ﬂuid-saturated porous medium when it is heated from below. Three different basin shapes having square,

2.6

Application of the Progressive Asymptotic Approach Procedure

23

Fig. 2.7 Comparison of numerical solution with analytical solution (Dimensionless pressure)

rectangular and trapezoidal geometries, which are ﬁlled with ﬂuid-saturated porous media, are considered in the analysis. For the rectangular basin, the ratio of width to height is 1.5. For the trapezoidal basin, the ratios of top width to height and bottom width to height are 2 and 1 respectively. In order to reﬂect the anisotropic behaviour of the porous media, the medium permeability in the horizontal direction is assumed to be three times that in the vertical direction. As shown in Fig. 2.8, all three basins are discretized into 484 nine-node quadrilateral elements of 2041 nodes in total. The boundary conditions of the problems are also shown in Fig. 2.8, in which n is the normal direction of a boundary. Two Rayleigh numbers, namely Ra = 80 and Ra = 400, are used to examine the effect of the Rayleigh number on natural convection in a ﬂuid-saturated porous medium. The same parameters as used in the above model veriﬁcation examples have been used here for the progressive asymptotic approach procedure. Figure 2.9 shows the dimensionless velocity distribution for the three different basins, whereas Figs. 2.10 and 2.11 show the dimensionless streamline contours

24

2

Simulating Steady-State Natural Convective Problems

T* = 0

u* = 0 v* = 0

∂T * =0 ∂n

u* = 0 v* = 0

∂T * =0 ∂n

1

u* = 0, v* = 0, T * = 1 L* = 1 T* = 0

u* = 0 v* = 0

∂T * =0 ∂n

u* = 0 v* = 0

∂T * =0 ∂n

1

u* = 0, v* = 0, T * = 1 L* = 1.5

L* = 2 T* = 0

u* = 0 v* = 0

∂T * =0 ∂n

u* = 0 v* = 0

∂T * =0 ∂n

1

u* = 0, v* = 0, T * = 1 L* = 1

Fig. 2.8 Finite element meshes for three different basins shapes

2.6

Application of the Progressive Asymptotic Approach Procedure

25

Fig. 2.9 Dimensionless velocity distribution for different basins

due to different basin shapes for Ra = 80 and Ra = 400 respectively. It is obvious that different basin shapes have a considerable effect on the patterns of convective ﬂow in the ﬂuid-saturated porous medium, especially in the case of higher Rayleigh numbers. Apart from notable differences in velocity distribution patterns, maximum velocity amplitudes for three different basins are also signiﬁcantly different. For instance, in the case of Ra = 80, the maximum amplitudes of dimensionless velocities are 5.29, 8.93 and 11.66 for square, rectangular and trapezoidal basins respectively. This fact indicates that different basin shapes may affect the contaminant transport or mineralization processes in a ﬂuid-saturated porous medium once natural convection is initiated in the medium.

2.10 Dimensionless streamline contours for different basins (Ra = 80) .26 2 Simulating Steady-State Natural Convective Problems Fig.

6 Application of the Progressive Asymptotic Approach Procedure 27 Fig. 2.11 Dimensionless streamline contours for different basins (Ra = 400) .2.

η A .e. y A = y A + v A Δt. it is necessary to use the particle tracking technique to show particle movements in three dimensional ﬂuid ﬂow systems. (2. (2. if the velocity of the particle at point A is known. a fundamental problem. is as follows. (2.6.67) (2. z A = z A + w A Δt. y and z directions of the global coordinate system. then the coordinate components of point A in the global coordinate system can be approximately determined for a small Δt as follows: x A = x A + u A Δt. ξ A . y A and z A are the coordinate components of point A in the x. which needs to be solved effectively. For this purpose. In the particle tracking technique. y A and z A are the coordinate components of point A in the x. ζ A )yi . it is essential to ﬁnd the coordinate components of point A in the local coordinate system from the following equations for an isoparametric ﬁnite element. y A and z A are the coordinate components of point A in the x. y A .70) yA = i=1 n φi (ξ A . where x A .72) where x A . In general cases. ζ A )z i . Clearly.2 Three-Dimensional Convective Pore-Fluid Flow Problems The proposed progressive asymptotic approach procedure can be straightforwardly extended to the simulation of three-dimensional convective pore-ﬂuid ﬂow problems in ﬂuid-saturated porous media (Zhao et al.69) where x A . the location of point A is not coincident with the nodal points in a ﬁnite element analysis so that the consistent interpolation of the ﬁnite element solution is needed to determine the velocity components at this point. u A . z A ). respectively. 2001a.71) zA = i=1 φi (ξ A . If the initial (known) location of a particle is point A (x A . y and z directions of the global coordinate system.68) (2. then we need to determine where the new location (i. Δt. η A . Since the streamline function is not available for three dimensional ﬂuid ﬂow problems. (2. v A and w A are the velocity components of point A in the x. y and z directions of the global coordinate system. point A ) of this particle is after a given time interval. η A . 2003a). η A and ζ A are three coordinate components . ζ A )xi .28 2 Simulating Steady-State Natural Convective Problems 2. n xA = i=1 n φi (ξ A . y and z directions of a global coordinate system.

The above-mentioned process indicates that in the ﬁnite element analysis of ﬂuid ﬂow problems. In order to simulate geothermal conditions in geology.74) wA = i=1 φi (ξ A .71) and (2. ζ A )u i (2. η A . η A .12 shows the ﬁnite element mesh of 8000 cubic elements for the three dimensional convective ﬂow problem. the example considered in this section is a cubic box of 10 × 10 × 10 km 3 in size.6 Application of the Progressive Asymptotic Approach Procedure 29 of point A in the ξ . Once the coordinate components of point A in the global coordinate system are determined. (2. vi and wi are the velocity components of nodal point i in the x. the velocity components of point A in the global system can be straightforwardly calculated as follows: n uA = i=1 n φi (ξ A . ζ A )wi . y and z directions of the global coordinate system. n is the total nodal number of the element containing point A. ζ A )vi (2. two cases are considered in the following computations. y and z directions of the global coordinate system. η and ζ directions of a local coordinate system.72). This means that the pore-ﬂuid saturated porous rock is uniformly heated from below. Figure 2. Zhao et al. This means that the problem is axisymmetrical about the y axis so that . yi and z i are the coordinate components of nodal point i in the x. For the system considered here.70). In Eqs. xi . 1999f).73) vA = i=1 n φi (ξ A . Nield and Bejan 1992. the perturbation of gravity is applied in the x-z plane only. u i . (2.75) where u A . the parameters and properties of the system are deliberately selected in such a way that the Rayleigh number of the system is supercritical. This box is ﬁlled with pore-ﬂuid saturated porous rock. which are fundamental quantities and therefore available in the ﬁnite element analysis.2. η A . φi is the interpolation function of node i in the element containing point A. y and z directions of the global coordinate system. v A and w A are the velocity components of point A in the x. For this reason. the trajectory of any given particle can be calculated using the nodal coordinate and velocity components. 1997a) indicates that the convective ﬂow is possible when the Rayleigh number of the system is either critical or supercritical. (2. which is a part of the upper crust of the Earth. the classical analysis (Phillips 1991. the coordinate components of point A in the global coordinate system are known. In the ﬁrst case. For the purpose of investigating the perturbation direction on the pattern of convective ﬂow. so that the coordinate components of this point in the local system can be determined using any inverse mapping technique (Zhao et al. To demonstrate the applicability of the progressive asymptotic approach procedure for simulating convective pore-ﬂuid ﬂow in three dimensional situations. respectively. the bottom of the box is assumed to be hotter than the top of the box.

the perturbation of gravity is equally applied in both the x-z and y-z planes.07 × 10−4 1/◦ C 4185 J/(kg×0 C) 0. Temperatures at Table 2. To reﬂect the three-dimensional features of convective pore-ﬂuid ﬂow.35 W/(m×0 C) porous matrix .6 W/(m×0 C) 0. Since the solutions for the axisymmetrical convective ﬂow problem are available (Nield and Bejan 1992.1 shows the parameters used in the computations for both cases. the following boundary conditions are used.e. Table 2. axisymmetrical case).12 Finite element mesh for the three-dimensional problem the problem can be degenerated into a two dimensional problem. the numerical methods used in this study can be veriﬁed by comparing the related analytical solutions with the solutions obtained from this special three dimensional case (i. 2. In the second case. from the mathematical and analytical points of view. This means that a true three dimensional convective ﬂow problem is considered in this case. 1997a). 1 Parameters used for the three-dimensional convective ﬂow problem Material type pore-ﬂuid Parameter dynamic viscosity reference density volumetric thermal expansion coefﬁcient speciﬁc heat thermal conductivity coefﬁcient porosity permeability Speciﬁc heat thermal conductivity coefﬁcient Value 10−3 N × s/m2 1000 kg/m3 2.30 2 Simulating Steady-State Natural Convective Problems Fig.1 10−14 m2 815 J/(kg×0 C) 3. Zhao et al.

numerical solutions) . 2.13 Distributions of pore-ﬂuid velocity and temperature in the porous medium (Case 1. the numerical solutions for both pore-ﬂuid velocity and temperature in this case are exactly axisymmetrical about the y axis.13 and 2.14 Fig. Figures 2.13 compare very well with the previous solutions in Fig. while all the four side boundaries of the computational domain are assumed to be insulated and impermeable in the horizontal direction. 2. Both the top and the bottom of the computational domain are impermeable in the vertical direction.6 Application of the Progressive Asymptotic Approach Procedure 31 the top and the bottom of the computational domain are 0◦ C and 250◦ C respectively. This implies that the three-dimensional problem considered in this particular case can be reasonably treated as a two-dimensional one. As expected.e. 2. It is also observed that the numerical solutions in Fig.14 show the numerical and analytical solutions for the distributions of pore-ﬂuid velocity and temperature in the axisymmetrical case (i.2. case 1) respectively.

is equally applicable to the ﬁnite element modelling of three-dimensional convective pore-ﬂuid ﬂow in ﬂuid-saturated porous media when they are heated from below. 1997a). Analytical solutions) for the axisymmetrical convective pore-ﬂuid ﬂow problem (Zhao et al. . This indicates that the progressive asymptotic approach procedure. 2.32 2 Simulating Steady-State Natural Convective Problems Fig.14 Distributions of pore-ﬂuid velocity and temperature in the porous medium (Case 1. although it was previously developed for the ﬁnite element modelling of two-dimensional convective pore-ﬂuid ﬂow problems.

2.15). Just like two-dimensional convective ﬂow problems. This demonstrates that the perturbation of gravity at different planes may have a signiﬁcant effect on the pattern of convective pore-ﬂuid ﬂow in three-dimensional hydrothermal systems. 2. the particle tracking technique introduced in this section is used during the numerical computation. .e. the solution dependence of threedimensional convective ﬂow on the direction of the perturbation of gravity need to be considered when these kinds of results are interpreted. 2. it is observed that the distribution patterns of both pore-ﬂuid velocity and temperature are totally different for these two cases. To further observe the movements of pore-ﬂuid particles.15 shows the numerical solutions for the distributions of pore-ﬂuid velocity and temperature in the real three-dimensional case (i. case 2).2. By comparing the numerical solutions from case 1 (Fig.6 Application of the Progressive Asymptotic Approach Procedure 33 Fig.15 Distributions of pore-ﬂuid velocity and temperature in the porous medium (Case 2) Figure 2.13) with those from case 2 (Fig.

It is clear that in this case. Figure 2.18 show the particle trajectories in the porous media for several different time instants in case 2. although the initial locations (at t = 0 s) of these 24 particles are exactly the same for these two cases.16 Particle trajectories in the porous medium (Case 1) We selected 24 particles. the different patterns of particle trajectories imply that the pattern of ore body formation and mineralization may be totally different in those two three-dimensional hydrothermal systems. 12 of which are in the front plane and the rest are in the back plane of the computation domain.16 shows the particle trajectories in the porous media for several different time instants in case 1. . the particle trajectories shown in case 2 are much more complicated than those shown in case 1. Since pore-ﬂuid is often the sole agent to carry the minerals from the lower crust to the upper crust of the Earth. In fact. Figures 2. all the particles move within the planes they are initially located within (at t = 0 s).17 and 2. It is observed that the particle trajectories shown in case 2 are totally different from those shown in case 1. to view the trajectories of those particles. because the convective pore-ﬂuid ﬂow (in case 1) is essentially axisymmetrical about the rotation axis of these particles considered.34 2 Simulating Steady-State Natural Convective Problems (t = 0s) (t = 1 × 1013 s) (t = 2 × 1013 s) (t = 3 × 1013 s) (t = 4 × 1013 s) (t = 5 × 1013 s) Fig. 2.

2.2.6 Application of the Progressive Asymptotic Approach Procedure 35 (t = 0s) (t = 1 × 1013 s) (t = 2 × 1013 s) (t = 3 × 1013 s) (t = 4 × 1013 s) Fig.17 Particle trajectories in the porous medium (Case 2) (t = 5 × 1013 s) .

2.36 2 Simulating Steady-State Natural Convective Problems (t = 6 × 1013 s) (t = 7 × 1013 s) (t = 8 × 1013 s) (t = 9 × 1013 s) (t = 1 × 1014 s) Fig.18 Particle trajectories in the porous medium (Case 2) (t = 1.1 × 1014 s) .

The second is to extend an existing commercial computational code. Primary advantages of using commercial computational codes are: (1) built-in pre-processing and post-processing tools make it very easy and attractive to prepare. correctly. Heat Transfer and Mass Transport Processes in Hydrothemal Systems Over the past decade or so.Chapter 3 A Consistent Point-Searching Interpolation Algorithm for Simulating Coupled Problems between Deformation. the main disadvantage of using commercial computational codes is that each code is often designed. many commercial computational codes have become available for solving a great number of practical problems in both scientiﬁc and engineering ﬁelds. 1996d. which are encountered and not solved in the ﬁeld of contemporary computational geoscience. (2) provision of movie/animation functions enables numerical results. effectively and efﬁciently for solving a wide range of practical problems. This disadvantage becomes more and more obvious because the ever-increasing competitiveness in the world economy requires us to deal with more and more complicated and complex geoscience problems. Compared with the difﬁculties encountered in the ﬁrst two approaches. The ﬁrst is to develop some new commercial computational codes. c. e. (3) detailed benchmark solutions and documentation as well as many embedded robust solution algorithms allow the codes to be used more easily. within a certain limit. for solving some particular kinds of practical problems. Pore-Fluid Flow. Zhao et al. Our ﬁrst successful example in the practice of using commercial computational codes in a combination manner was to optimize structural topologies under either static or dynamic conditions using the commercial code STRAND6 (G+D Computing 1991) and a home-made code GEMDYN. 1997b. d). which is time consuming and often not costeffective for numerical analysts and consultants. DOI 10. This requires development of a data translation tool to transfer data necessary between each of the codes to be used. a generalized evolutionary method for numerical topological optimization of structures has been developed and many interesting numerical results have been produced (Zhao et al. input and output data which are essential in a numerical analysis.. C Springer-Verlag Berlin Heidelberg 2009 37 . To extend further the idea of using commercial C.1007/978-3-540-89743-9 3. As a result. 1998c. to be visualised via clear and colourful images. There are three basic ways to overcome the above difﬁculties. However. The third is to use several existing commercial computational codes in combination. Fundamentals of Computational Geoscience. d. which is usually impossible because the source code is often not available for the code users. the third one is more competitive for most numerical analysts and consultants. Lecture Notes in Earth Sciences 122. the treatment of which is often a cumbersome and tedious task.

it must be possible to relate accurately any point in the mesh for one code to the equivalent point in the mesh for the other code. especially for the simulation of large deformation problems. which is the .1 Statement of the Coupled Problem and Solution Method In terms of simulating the physical and chemical processes associated with ore body formation and mineralization in hydrothermal systems within the upper crust of the Earth. FIDAP is a well developed. and vice versa. to translate/transfer the solution of any particular point from mesh 2 to mesh 1. it makes it possible to translate and transfer data between FIDAP and FLAC. 1999c. which are derived in a rigorous mathematical manner in the context of this chapter. the proposed consistent algorithm is very effective and efﬁcient. heat transfer and mass transport/chemical reactions can be divided into two sub-problems (Zhao et al. This certainly minimizes the number of elements to which the inverse mapping is applied. also known as the consistent point-searching algorithm for solution interpolation in unstructured meshes consisting of 4-node bilinear quadrilateral elements. The proposed algorithm has the following signiﬁcant advantages: (1) the use of a pointsearching strategy allows a point in one mesh to be accurately related to an element (containing this point) in another mesh. On the other hand. it is very reasonable to envisage interactively using FIDAP and FLAC for solving a fully coupled problem between medium deformation. whereas FLAC. Thus. the fully coupled problem between material deformation. only one element in mesh 2 needs to be inversely mapped. To do this. we present a consistent point-searching interpolation algorithm. Thus. pore-ﬂuid ﬂow. heat transfer and reactive species transport in a porous medium under high Rayleigh number situations. we attempt to use a combination of FIDAP (Fluid Dynamics International 1997) and FLAC (Itasca Consulting Group 1995) for solving a fully coupled problem between medium deformation. heat transfer and reactive species transport in a porous medium. make it possible to carry out an inverse mapping process very effectively and efﬁciently. 2000b). but does not treat the medium deformation effects. computational ﬂuid dynamics code. heat transfer and reactive mass transport in a porous medium under high Rayleigh number situations. However. the weakness of FLAC is that it cannot be used rigorously to model steady-state pore-ﬂuid convection and the related reactive species transport in a ﬂuid-saturated porous medium. In this regard. pore-ﬂuid ﬂow. In order to do this. is based on a ﬁnite difference method. designed for civil engineering. but can accommodate unstructured grids. (3) The use of consistent interpolation enables the interpolated solution to be compatible with an original solution and therefore guarantees the interpolated solution of extremely high accuracy. 3. ﬁnite element method based. Since the algorithm is very general and robust. FLAC is very powerful in its modeling of geomechanical and geological deformation processes. For the ﬁrst sub-problem. FIDAP can be used to model pore-ﬂuid ﬂow. (2) Analytical solutions to the local coordinates of any point in a four-node quadrilateral element.38 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems computational codes in combination. pore-ﬂuid ﬂow.

3) ∂T ∂T +v ∂x ∂y (3. . K x and K y are the permeabilities of the porous medium in the x and y directions respectively. P is the pore-ﬂuid pressure. Dex = φ Dfx . reference temperature and reference normalized concentration of the chemical species used in the analysis. . λey = φλ f y + (1 − φ)λsy . . (3.5) ρ f = ρ f 0 1 − βT (T − T0 ) − i=1 βCi (C i − C 0 ) . Dey = φ Dfy . where u and v are the horizontal and vertical velocity components of the pore-ﬂuid in the x and y directions respectively.4) = ρ f 0 Dex ∂ 2C i ∂ 2C i + Dey ∂x2 ∂ y2 N (i = 1.3. φ is the porosity of the porous medium. Ri is the chemical reaction term for the transport equation of chemical species i. N ).1 Statement of the Coupled Problem and Solution Method 39 problem describing the steady-state pore-ﬂuid ﬂow. . c p is the speciﬁc heat of the pore-ﬂuid. 2. N is the total number of the active chemical species considered in the pore-ﬂuid. ∂y = λex ∂2T ∂2T + λey 2 .6) (3. the corresponding governing equations are expressed as follows: ∂v ∂u + = 0. λ f x and λsx are the thermal conductivities of the pore-ﬂuid and solid matrix in the x direction. ∂x (3. (3. D f x and D f x are the diffusivities of the chemical species in the x and y directions respectively. C i is the normalized concentration (in a mass fraction form relative to the pore-ﬂuid density) of chemical species i. ∂x ∂y u= v= ρ f 0c p u ρf0 u ∂C i ∂C i +v ∂x ∂y Ky μ Kx μ − − ∂P . T is the temperature of the porous medium.8) λex = φλ f x + (1 − φ)λsx . ρ f is the density of the pore-ﬂuid and g is the acceleration due to gravity. T0 and C 0 are the reference density. βT and βCi are the thermal volume expansion coefﬁcient of the pore-ﬂuid and the volumetric expansion coefﬁcient due to chemical species i.2) ∂P + ρf g .1) (3. heat transfer and mass transport/chemical reactions in a porous medium. ρ f 0 . λ f y and λsy are the thermal conductivities of the pore-ﬂuid and solid matrix in the y direction. μ is the dynamic viscosity of the pore-ﬂuid.7) (3. . 2 ∂x ∂y + φ Ri (3.

1998. R2 = −k1 C 1 C 2 . Nithiarasu et al. 1 A+B⇒F k (3. 1996. This means that we assume that the material deformation of the hydrothermal system due to gravity has completed before the system is heated by some thermal event from below. species 3). the reaction term involved in Equation (3. Generally.13) where E a is the activation energy. Gobin and Bennacer 1994. Nield and Bejan 1992. 2006a) is taken into account in the ﬁrst sub-problem. R3 = k 1 C 1 C 2 . Alavyoon 1993. species 2) respectively. Zhao et al. R is the gas constant. The second sub-problem is a static deformation problem under plane strain conditions.e.12) where C 1 and C 2 are the normalized concentrations (in a mass fraction relative to the pore-ﬂuid density) of species A (i. Note that for most of the chemical reactions encountered in the ﬁeld of geoscience.9) This equation states that species A and species B react chemically at a rate constant of k1 and the product of this reaction is species F. (3.e. the governing equations for static deformation (which is labeled as the second sub-problem) in the porous medium under plane strain conditions are: . 1994.5) can be expressed as follows: R1 = −k1 C 1 C 2 . This means that the double diffusion effect (Phillips 1991.6) clearly indicates that the density of the pore-ﬂuid considered in this study is a function of both temperature and chemical species concentrations. C 3 is the normalized concentration (in a mass fraction relative to the pore-ﬂuid density) of species F (i. the chemical reaction term involved in a chemical species transport equation is strongly dependent on the speciﬁc chemical reaction considered in the analysis.11) (3. 1996.e. the rate of a chemical reaction is dependent on the temperature at which the chemical reaction takes place. Goyeau et al. then body forces can be neglected in the corresponding force equilibrium equations. If the hydrothermal system is initially in a mechanically equilibrium state. the following type of equation can be considered as an illustrative example. 2000b.40 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems Equation (3. the temperature dependent nature of the reaction rate for the chemical reaction considered here can be taken into account using the Arrhenius law of the following form: k1 = k A exp −E a RT (3. Under this assumption. For this type of chemical reaction. T is the temperature in Kelvin and k A is the pre-exponential chemical reaction constant. From a geochemical point of view (Nield and Bejan 1992).10) (3. Nguyen et al. 2005b. Mamou et al. species 1) and B (i. For a non-equilibrium chemical reaction consisting of aqueous chemical species only.

21) where K 0 is the initial permeability corresponding to the initial porosity. (3. (1 − 2v)(1 + v) 1−v 1 − 2v E(1 − v) (1 − 2v)(1 + v) EαT v εx + ε y − − P. 1995): φ =1− 1 − φ0 . we need to establish a relationship between the volumetric strain and the porosity of the porous medium. whereas Eqs. respectively. T. τx y and γx y are shear stress and shear strain of the solid matrix. pore-ﬂuid pressure. φ0 .18) and (3.1 Statement of the Coupled Problem and Solution Method 41 ∂τ yx ∂σx + = 0.3.17) (3. ∂x εy = ∂vs . To couple the ﬁrst sub-problem with the second sub-problem.20) where φ and φ0 are the porosity and initial porosity of the porous medium. such a relationship can be expressed as (Itasca Consulting Group. (3. (3. E and G are the elastic and shear modulus respectively.14) (3.16). (3. 3 φ0 (1 − φ)2 (3. u s and vs are the horizontal and vertical displacements of the solid matrix. p. ∂y γx y = 1 2 where σx and σ y are normal stresses of the solid matrix in the x and y directions.15) (3. and permeabilities. Obviously.19) τx y = τ yx = 2Gγx y .19) are the constitutive equations and straindisplacement relationship equations. ∂x ∂y σx = σy = EαT E(1 − v) v εx + εy − − P. K x and K y . the ﬁrst sub-problem is coupled with the second sub-problem through the medium temperature.16) (3. 1 + εv εv = εx + ε y . Note that Eqs. . (3. ∂x ∂y ∂σ y ∂τx y + = 0. Using the Carman-Kozeny formula (Nield and Bejan 1992). 1−v 1 − 2v (3.15) represent the equilibrium equations. For small strain problems. εv is the volumetric strain of the solid matrix. ν is the Poisson ratio of the solid matrix and α is the linear thermal expansion coefﬁcient of the solid matrix. (3.18) ∂u s ∂vs + ∂y ∂x .14) and (3. the permeability of an isotropic porous medium is expressed as a function of porosity as follows: Kx = K y = K 0 (1 − φ0 )2 φ 3 .17). εx and ε y are the normal strains of the solid matrix in relation to σx and σ y . P is the excess pore-ﬂuid pressure due to the thermal effect. εx = ∂u s .

within the second sub-problem itself. To obtain non-trivial (convective) solutions for the pore-ﬂuid ﬂow under supercritical Rayleigh number conditions. u and v. 3.1.42 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems Within the ﬁrst sub-problem itself. aqueous chemical species concentrations and pore-ﬂuid velocity components. the progressive asymptotic approach procedure discussed in Chap. while FLAC (Itasca Consulting Group 1995). This Fig. εv . the pore-ﬂuid ﬂow is coupled with the thermal ﬂow and reactive ﬂow through the medium temperature. 3. which is one of the most powerful ﬁnite element codes currently available for dealing with a very wide range of practical problems in ﬂuid dynamics.1 Two totally different meshes used in FIDAP and FLAC . is used to solve the second sub-problem. With regard to the general solution method used in this study. the medium deformation (displacement) is coupled with the medium temperature. ﬁnite element solutions to the temperature and pressure due to hydrothermal effects are available (from FIDAP) at the nodal points of mesh 1 but not known at those of mesh 2. 3. For two totally different meshes. However. 1997a) is also employed in the computation. FIDAP (Fluid Dynamics International 1997). A general interface program is developed to translate and transfer data between FIDAP and FLAC. pore-ﬂuid pressure and permeability through the volumetric strain. is used to solve the ﬁrst sub-problem. which is a very useful ﬁnite difference code for dealing with material deformation problems in the ﬁelds of both geotechnical engineering and geoscience. φ. 2 (Zhao et al.2 Mathematical Formulation of the Consistent Point-Searching Interpolation Algorithm in Unstructured Meshes The motivation of developing a consistent point-searching interpolation algorithm is described as follows. and the medium porosity. namely mesh 1 (mesh in FIDAP) and mesh 2 (mesh in FLAC) shown in Fig.

the development of an efﬁcient searching strategy to limit the number of elements. If any point (i. 2.. using the nodal solutions of mesh 1 and the related elemental information. so that the algorithm growth rate can be reduced to the minimum. in the proposed algorithm. Clearly. To achieve this. 41 × 4A = λ4 k . then the following equations hold true: 12 × 1A = λ1 k . one has to deal with the following two key issues. the local coordinates of this point need to be evaluated using the related information of the element in mesh 1.25) (3. 3. Since mesh 1 is totally different from mesh 2.26) . 4). respectively.22) (3. to which the inverse mapping is applied. point A in Fig. Second.2. in mesh 1.3. for 4-node bilinear quadrilateral elements. the analytical solution to the inverse mapping problem for this kind of element may ﬁnd wide applications for many practical problems. the inverse mapping problem is a nonlinear one. Consider a four-node quadrilateral element shown in Fig. they are addressed in great detail in the following sections. which requires numerical solutions for higher order elements. for the purpose of developing the consistent point-searching interpolation algorithm.2) is located within this element. 34 × 3A = λ3 k . which need to be inversely mapped. it is possible to solve this problem analytically. Given the importance of the above two key issues. First. This is the main reason why the proposed algorithm is called the consistent point-searching interpolation algorithm. 23 × 2A = λ2 k .e. 3. the issue of the parametric inverse mapping between the real (global) and element (local) coordinates should be dealt with in an appropriate manner. we have to establish a strategy to relate a point in one mesh to an element in another mesh accurately.1 Point Searching Step In this section. Thus. 3. for any nodal point in mesh 2. This requires an inverse mapping to be carried out for this particular element in mesh 1.2 Mathematical Formulation of the Algorithm 43 indicates that to evaluate the thermal deformation using mesh 2 in FLAC.e. which contains this particular point. Since most practical problems in ﬁnite element analysis can be modelled reasonably well using 4-node bilinear quadrilateral elements.24) (3. Once a point in mesh 2 is related to an element (containing this point) in mesh 1. → → → → → → → → → → → → (3.2. λi ≥ 0 (i = 1. so that the nodal solution of the point in mesh 2 can be consistently interpolated in the ﬁnite element sense.23) (3. it is necessary to ﬁnd out the element. However. 3. it is essential to translate/transfer the nodal solutions from mesh 1 into mesh 2. it is only possible to interpolate the nodal solution of mesh 2.. i. point by point. This is the ﬁrst step. an efﬁcient searching strategy needs to be developed to limit the number of elements. we deal with the ﬁrst issue. termed point-searching. Generally.

λi (i = 1. (3.27). (3.28). k is a normal vector of the plane where the element is located.44 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems 3 4 y A 1 2 0 x Fig. (x1 − x4 )(y A − y4 ) − (x A − x4 )(y1 − y4 ) = λ4 ≥ 0. 3A and 4A are vectors of each node of the element to point A respectively. (3. 4) are four different constants. (3. point A and four nodes of the element). 2. (3.23). in which point A is located. 2A. 3. Note that when point A is located on the side of the element. it is very easy and accurate to relate a point in mesh 2 to an element (containing this point) in mesh 1.27).25) represent the cross products of two vectors. using the point-searching strategy expressed by Eqs. (3.29) and (3. they can be straightforwardly used to predict the element. (3. 23.29) and (3.24).27) (3.24) and (3.28). (3.23)..28). Consequently. Thus.2 Point A in a fournode quadrilateral element → → → → → → → where 12. (3. By using the global coordinates of point A and four nodal points of the element.e. (3. (3.26) can be expressed in the following form: (x2 − x1 )(y A − y1 ) − (x A − x1 )(y2 − y1 ) = λ1 ≥ 0.28) (3.30) only involve certain simple algebraic calculations which can be carried out by computers at a very fast speed.30) → → where xi and yi are the global coordinates of nodal points of the element. two vectors related to a node of the element are coincident. (3. Note that the left-hand sides of Eqs. (x3 − x2 )(y A − y2 ) − (x A − x2 )(y3 − y2 ) = λ2 ≥ 0.27).25) and (3. 3. since Eqs. Since Eqs. (3. (3. Eqs (3. 1A.22).29) and (3.29) (3. the corresponding λi to this node must be equal to zero.30). It must be pointed out that. (x4 − x3 )(y A − y3 ) − (x A − x3 )(y4 − y3 ) = λ3 ≥ 0.22). to translate/transfer the solution . (3.30) are only dependent on the global coordinates of ﬁve known points (i. 34 and 41 are vectors of four sides of the element. x A and y A are the global coordinates of point A. (3.

In order to interpolate the solution at point A consistently in the ﬁnite element sense. to which the inverse mapping is applied. only one element in mesh 1 needs to be inversely mapped. The following gives the related mathematical equations.2 Mathematical Formulation of the Algorithm 45 of any point from mesh 1 to mesh 2. For a four-node quadrilateral isoparametric element shown in Fig. it is possible to implement the inverse mapping analytically.3. 1999f) . η 3 4 ξ y 0 1 0 x (Physical element) η 4 3 2 0 ξ 1 (Parent element) 2 Fig. 3. it is necessary to ﬁnd out the local coordinates of this point by an inverse mapping. Zhao et al. 3.3 A four-node quadrilateral isoparametric element (Zienkiewicz 1977. 3. the present algorithm is both effective and efﬁcient. to the minimum. From the computational point of view.2. This certainly reduces the number of elements.2 Inverse Mapping Step This section deals with the second key issue related to the development of the consistent point-searching interpolation algorithm.3.

43) (3. (3.35) (3. N2 .34). b1 = y1 + y2 + y3 + y4 .35) and (3. b2 = −y1 + y2 + y3 − y4 .44) (3.37) (3. (3. the corresponding local coordinates are ξ A and η A . where a1 = x 1 + x 2 + x 3 + x 4 .e. point A). 4 N4 = 1 (1 − ξ )(1 + η). a2 = −x1 + x2 + x3 − x4 .36) where N1 . b3 = −y1 − y2 + y3 + y4 . a3 = −x1 − x2 + x3 + x4 .39) (3.42) (3. b4 = y1 − y2 + y3 − y4 .46 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems The forward mapping of the physical element in the global system to the parent element in the local system reads: 4 x= i=1 4 Ni x i .33) (3.32). (3. N3 and N4 are shape functions of the element. (3.36) yields the following inverse mapping: a2 ξ A + a3 η A + a4 ξ A η A = 4x A − a1 . For a given point (i.45) (3.34) (3. (3.31). Substituting these local coordinates into Eqs.46) (3. (3. 4 1 N3 = (1 + ξ )(1 + η).40) (3.31) y= where N1 = (3. i=1 (3. 4 1 N2 = (1 + ξ )(1 − η). Ni yi . 4 (3.41) (3. a4 = x 1 − x 2 + x 3 − x 4 ..38) .33).32) 1 (1 − ξ )(1 − η). b2 ξ A + b3 η A + b4 ξ A η A = 4y A − b1 .

Eqs. a2 b3 − a3 b2 (3.2 Case 2: a4 = 0.3. (3.1 Case 1: a4 = 0.53) and (3.38) can be written as a2 ξ A + a3 η A = c1 .48) Clearly. (3.54) b3 c1 − a3 c2 .49) (3. where c1 = 4x A − a1 . so that the corresponding solutions can be immediately obtained as ξA = ηA = 3.52) The solutions to Eqs.54) can be expressed for the following three sub-cases: (1) a2 = 0.53) (3. a3 = 0 ηA = ξA = (2) a2 = 0.51) (3.2 Mathematical Formulation of the Algorithm 47 Since Eqs. b2 ξ A + b3 η A + b4 ξ A η A = c2 .50) (3. (3.56) c2 − b3 η A . b4 = 0 In this case.47) and (3. ξ A and η A .37) and (3. c2 = 4y A − b1 .2.55) (3.48) are a set of standard linear simultaneous equations.37) and (3. a3 (3. a3 = 0 ξA = c1 . (3.47) (3. (3. b2 + b4 η A c1 . b4 = 0 The corresponding equations in this case can be rewritten as a2 ξ A + a3 η A = c1 .38) are two second-order simultaneous equations. Eqs. 3. (3.57) . a2 b3 − a3 b2 −b2 c1 + a2 c2 . can be obtained in the following different cases. b2 ξ A + b3 η A = c2 . the solution to the local coordinates.2.2. a2 (3.2.

b3 (3.59) (3. a2 b4 a3 .58) bb2 − (4aa)cc .3 Case 3: a4 = 0.65) Similarly.65) can be expressed in the following three sub-cases: (1) b2 = 0.68) (3.64) and (3.66) (3. a3 = 0 ηA = −bb ± c2 − b2 ξ A .64) (3. b3 = 0 ξA = ηA = c2 . a2 (3. b4 = 0 This case is very similar to case 2. b3 = 0 ηA = ξA = (2) b2 = 0. bb = a2 aa = cc = −b2 c1 . The corresponding equations in this case are as follows: a2 ξ A + a3 η A + a4 ξ A η A = c1 . the solution to Equations (3.69) c2 .48 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems ηA = (3) a2 = 0. b3 + b4 ξ A (3. (3. a3 + a4 ξ A .61) (3. b2 (3.67) c1 − a3 η A .2.63) 3.60) ξA = where c1 − a3 η A .62) (3. a2 + a4 η A c1 − a2 ξ A . a2 b2 a3 − b4 c1 − b3 .2. b2 ξ A + b3 η A = c2 . 2aa (3.

the solutions to Eqs.2. b4 (3.72) (3. ∗ a2 + η A . a3 − b3 ηA = ξA = (3. a4 ∗ c1 = bi∗ = c1 .70) (3. if a2 − b2 = 0.76) where ai∗ = ai . b2 cc = −a2 c2 .77) (3.77) yields the following equation: ∗ ∗ ∗ ∗ ∗ ∗ (a2 − b2 )ξ A + (a3 − b3 )η A = c1 − c2 . b3 = 0 ηA = −bb ± bb2 − (4aa)cc .4 Case 4: a4 = 0.73) (3. (3.77) and (3.37) and (3.75) (3. b4 = 0 In this case.3.2 Mathematical Formulation of the Algorithm 49 (3) b2 = 0.80) (3.78) from Eq. (3.2. b2 3. Eqs. (3.38) can be written in the following form: ∗ ∗ ∗ a2 ξ A + a3 η A + ξ A η A = c1 .71) ξA = where c2 − b3 η A . b2 (3. ∗ ∗ ∗ b2 ξ A + b3 η A + ξ A η A = c2 . 2aa (3. (3. a4 bi b4 ∗ c2 = (i = 2. (3.78) Subtracting Eq. 3).79) ∗ ∗ Clearly. b2 aa = bb = a4 b3 .81) ∗ ∗ c1 − a3 η A .78) can be straightforwardly expressed as ∗ ∗ c1 − c2 ∗ ∗.74) a2 b3 − a4 c2 − b3 . (3. c2 .

2.e.84) Therefore.83) (3.85) (3.89) where Si is the appropriate numerical solution at a nodal point in mesh 1.87) (3. the nodal value of any concerned solution in mesh 2 can be interpolated through mesh 1 and then directly transferred to mesh 2. a2 − b2 ∗ ∗ a3 − b3 ∗ ∗.50 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems Otherwise. point A) in mesh 1 can be consistently interpolated using the following equation: 4 SA = i=1 Ni (ξ A . η A )Si . 3. ∗ a3 + ξ A (3. any nodal solution at any point (i.3 Consistent Interpolation Step Based on the concept of isoparametric elements (Zienkiewicz 1977). Note that since the global coordinate system used in mesh 1 is exactly the same as in mesh 2.82) (3.. a2 − b2 (3. Eq.79) can be rewritten as ξ A = α − βη A where α= β= ∗ ∗ c1 − c2 ∗ ∗. 2β ∗ ∗ c1 − a2 ξ A . ηA = (2) For β = 0 ηA = ∗ ∗ −(a2 β − a3 − α) ± ∗ ∗ ∗ ∗ (a2 β − a3 − α)2 − 4β(c1 − a2 α) . . (3. (1) For β = 0 ξ A = α.86) (3. the corresponding solutions can be expressed for the following subcases.88) ξ A = α − βη A . (3.

5 show the comparison of the original temperature and pressure solutions in mesh 1 with the corresponding translated/ transferred solutions in mesh 2.6 W/(m×0 C) in both the horizontal and vertical directions. thermal conductivity coefﬁcient is 3. For the porous matrix. we consider a coupled problem between medium deformation. The left and right lateral boundaries of the computational domain are insulated and impermeable in the horizontal direction. pore-ﬂuid ﬂow and heat transfer processes in a ﬂuid-saturated porous elastic medium. Firstly. For this purpose.3 Veriﬁcation of the Proposed Consistent Point-Searching Interpolation Algorithm 51 3. the translated/transferred solutions in mesh 2 are essentially the same as in mesh 1. Then. temperature and pressure distributions in the computational domain. the robustness of a numerical Fourier transform algorithm is often tested by the following procedure: (1) implementation of a forward Fourier transform to an original function. Clearly.3.6 and 3. It is clear that although mesh 2 is totally different from mesh 1. we use the proposed algorithm to translate/transfer the temperature and pressure solutions in mesh 1 into FLAC’s mesh. volumetric thermal expansion coefﬁcient is 2.35 W/(m×0 C) in both the horizontal and vertical directions. For instance. it demonstrates that the numerical Fourier transform algorithm is robust. Figures 3. initial porosity is 0. while the inverse transform is deﬁned as translating/transferring the translated/transferred solution data back from mesh 2 to mesh 1.3 Veriﬁcation of the Proposed Consistent Point-Searching Interpolation Algorithm In order to verify the proposed consistent point-searching interpolation algorithm in two totally different meshes. dynamic viscosity is 10−3 N × s/m2 . This demonstrates the correctness and effectiveness of the proposed consistent pointsearching interpolation algorithm. the forward transform is deﬁned as translating/transferring solution data from mesh 1 to mesh 2. reference density is 1000 kg/m3 . For pore-ﬂuid. To simulate the high Rayleigh number convection cells in the computational domain of the test problem. the same procedure as above can be followed to examine the robustness of the proposed consistent point-searching algorithm. the following parameters are used in the computation. thermal conductivity coefﬁcient is 0.7 show the comparisons of the forwardly transformed temperature and pressure solutions (from mesh 1 to mesh 2) with the inversely transformed . If the inversely transformed function is exactly the same as the original function. whereas the top and bottom are impermeable in the vertical direction. we use FIDAP with mesh 1 to simulate the high Rayleigh number convection cells. Figures 3. To further test the robustness of the proposed algorithm. The temperature is 25◦ C at the top of the domain. The square computational domain is 10 × 10 km2 in size.4 and 3. initial permeability is 10−14 m2 .1. mesh 2. we use the concept of a transform in mathematics below. This means that the computational domain is heated uniformly from below. This computational domain is discretized into mesh 1 of 1625 nodes and mesh 2 of 10000 nodes respectively. and (2) implementation of an inverse Fourier transform to the forwardly transformed function.07×10−4 (1/0 C). while it is 225◦ C at the bottom of the domain. speciﬁc heat is 4185 J/(kg×0 C).

52 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems Fig.4 Comparison of the original solution with the translated/transferred solution (Temperature) . 3.

5 Comparison of the original solution with the translated/transferred solution (Pore-ﬂuid pressure) .3 Veriﬁcation of the Proposed Consistent Point-Searching Interpolation Algorithm 53 Fig.3. 3.

54 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems Fig.6 Comparison of the forwardly transformed solution with the inversely transformed solution (Temperature) . 3.

7 Comparison of the forwardly transformed solution with the inversely transformed solution (Pore-ﬂuid pressure) .3. 3.3 Veriﬁcation of the Proposed Consistent Point-Searching Interpolation Algorithm 55 Fig.

8 and 3. 3.5). 3.3.7) also compare very well with the original solutions in mesh 1 (see Figs.56 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems solutions (from mesh 2 to mesh 1). Pore-Fluid Flow and Heat Transfer in Fluid-Saturated Porous Media Since the veriﬁcation example considered in Sect. Towards this end. 3.9 show the temperature induced deformation and stresses in the porous elastic medium respectively. the volumetric thermal expansion coefﬁcient is 2. Figures 3. Poisson’s ratio is 0. Furthermore. they can be transformed back exactly from mesh 2 to mesh 1. (2) the top boundary is free. 3. it is assumed that: (1) the bottom boundary of the computational domain is ﬁxed. Clearly.1. Owing to relatively larger volumetric strain in the left side of the computational domain. the forwardly transformed solutions in mesh 2 are exactly the same as the inversely transformed solutions in mesh 1. The maximum horizontal compressive stress due to the thermal effect takes place in the hottest region of the computational domain.6 and 3. That is to say. the temperature induced horizontal stress dominates in the porous medium. higher temperature results in larger volumetric strain.8. Thus.4 and 3. It is observed that the distribution pattern of the volumetric strain is similar to that of the temperature. This is clearly evidenced in Fig. we can continue the simulation of the veriﬁcation example and use FLAC with the translated/transferred temperature in mesh 2 to compute thermal deformation and stresses in the ﬂuidsaturated porous elastic medium. As a direct consequence of the thermal deformation. Except for the parameters used in Sect.25. the inversely transformed solutions in mesh 1 (see Figs. the maximum vertical displacement takes place at the upper left corner of the domain. Apart from a small part of the top region of the computational domain. 3.4.3 is a coupled problem between medium deformation. the thermal induced vertical stress is . Such a reversible process demonstrates the robustness of the proposed consistent point-searching interpolation algorithm. while the maximum horizontal tensile stress occurs at the upper left corner of the domain. pore-ﬂuid ﬂow and heat transfer processes in a ﬂuid-saturated porous elastic medium. the following additional parameters are used in the continued simulation: the elastic modulus of the porous medium is 1×1010 Pa. it can be used as the ﬁrst application example of the proposed consistent point-searching interpolation algorithm.07×10−4 (1/o C) and the initial porosity is 0.4 Application Examples of the Proposed Consistent Point-Searching Interpolation Algorithm 3. and (3) the two lateral boundaries are horizontally ﬁxed but vertically free. where the vertical displacement reaches its maximum value. This means that after the original solutions are transformed from mesh 1 to mesh 2.1 Numerical Modelling of Coupled Problems Involving Deformation. 3. as expected from the physics point of view.

8 Deformation of the porous medium due to temperature effect .3.4 Examples of the Proposed Consistent Point-Searching Interpolation Algorithm 57 Fig. 3.

3.9 Temperature induced stresses in the porous medium .58 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems Fig.

we have used the proposed consistent point-searching interpolation algorithm to develop a general interface between the two commercial computational codes. structure. This requires us to have a strong theoretical understanding of the basic governing principles and processes behind the coupled problem. Pore-Fluid Flow. Phillips (1991) stated that: “A conceptual framework elucidating the relations among ﬂow characteristics. Speciﬁcally. However. the numerical algorithms and methods can be used to answer many whatif questions related to ore body formation and mineralization that can be described as coupled problems between medium deformation. b.” Keeping this in mind. Even in some circumstances. indeed. For this reason. and artifacts of a solution can be numerical rather than geological. but to check them. The ﬁrst application example. to develop theoretical solutions to verify the numerical methods developed (Zhao and Valliappan 1993a. and what reaction patterns the reactive pore-ﬂuid can produce in a hydrothermal system. This indicates that the vertical forces are well balanced in the horizontal cross-section of the computational domain. Heat Transfer and Mass Transport in Fluid-Saturated Porous Media Generally. FIDAP and FLAC. we must know. in recent years.3. 1997a. due to the approximate nature of a numerical solution. what the pore-ﬂuid can and cannot do. through some kind of theoretical analysis. 1999b). which is closely associated with the coupled problem between medium deformation. driving forces. the distributions of temperature. pore-ﬂuid ﬂow. complementary in the ﬁeld of computational geoscience. at least qualitatively if the analytical solution to the problem is not available. (Numerical calculations can converge to a griddependent limit. and reaction patterns enables us not only to understand the results of numerical modelling more clearly. This indicates that the numerical and theoretical approaches are. Zhao et al. a good numerical solution can provide some useful hints for deriving analytical solutions to some aspects of the problem. The combination of the two techniques is a much more powerful research tool than either alone.2 Numerical Modelling of Coupled Problems Involving Deformation. 1998a.) Numerical modelling provides a quantitative description and synthesis of a basin-wide ﬂow in far greater detail than would be feasible analytically.4 Examples of the Proposed Consistent Point-Searching Interpolation Algorithm 59 well distributed anti-symmetrically. On the other hand. heat transfer and reactive mass transport in a ﬂuid-saturated porous medium. pore-ﬂuid ﬂow. This development enables us to investigate the integrated behaviour of ore body formation and mineralization in hydrothermal systems. it is essential to evaluate the accuracy of the numerical solution.4. 3. is to answer the question: What is the pattern of pore-ﬂuid ﬂow. Realizing this particular relationship between the analytical method and the numerical method. b. reactant and product . a good numerical solution can provide insights into the integrated behaviour of different processes that occur in a hydrothermal system. heat transfer and reactive mass transport in ﬂuid-saturated porous media. 1994a. we have been making efforts.

the computational domain considered for this example is a square box of 10 by 10 km in size. whereas the top and bottom are impermeable in the vertical direction. the computational domain is discretized into 2704 quadrilateral elements with 2809 nodes in total.11. Figure 3. if a deformable hydrothermal system has two reservoirs for two different reactant chemical species. In addition.001 v = 0. the normalized concentration of species 1 is 0. To consider the mixing of reactant chemical species. we have to validate the numerical solution.60 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems v = 0. and is heated uniformly from below? As shown in Fig. The left and right lateral boundaries are insulated and impermeable in the horizontal direction. The normalized concentrations of all three species involved in the chemical reaction are assumed to be zero at the top surface of the computational domain. 3. T = 25 0 C .001 at the left quarter of the bottom and the normalized concentration of species 2 is 0. at least from a qualitative point of view.10 Geometry and boundary conditions of the coupled problem chemical species. For the hydrothermal system . while it is 225◦ C at the bottom of the domain.11 shows the pore-ﬂuid velocity and temperature distributions in the deformable porous medium.1 shows the related parameters used in the computation.10.001 Fig. As we mentioned in the beginning of this section. The temperature at the top of the domain is 25◦ C.001 at the right quarter of the bottom. 3. T = 225 0 C 10km C 2 = 0. and the patterns of ﬁnal porosity and permeability. This convective ﬂow results in the localized distribution of temperature. Table 3. It is observed that a clockwise convection cell has formed in the pore-ﬂuid ﬂow in the porous medium. as can be seen clearly from Fig. This implies that species 1 and 2 are injected into the computational domain through two different reservoirs at different locations. C1 = C 2 = C 3 = 0 u=0 u=0 ∂T = 0 ∂x ∂T = 0 ∂x 10km y x 0 C1 = 0. 3.

315 J/(mol×0 K) 0.07 × 10−4 (1/0 C) 4185 J/(kg×0 C) 0. the Rayleigh number is 55. from the theoretical point of view. but may not necessarily reach its equilibrium concentration in other regions in the computational domain.1 Parameters used for the ﬁrst coupled problem involving reactive mass transport 61 Material type pore-ﬂuid Parameter dynamic viscosity reference density volumetric thermal expansion coefﬁcient speciﬁc heat thermal conductivity coefﬁcient chemical species diffusivity coefﬁcient pre-exponential reaction rate constant activate energy of reaction gas constant initial porosity initial permeability elastic modulus Poisson’s ratio volumetric thermal expansion coefﬁcient thermal conductivity coefﬁcient Value 10−3 N × s/m2 1000 kg/m3 2.e. is greater than the critical Rayleigh number. Obviously. the distribution of chemical species 3.6 W/(m×0 C) 3 × 10−6 m2 /s 10−7 kg/(m3 ×s) 5 × 104 J/mol 8. The criterion says that if the Rayleigh number of the hydrothermal system. This ﬁnding might be important for the further understanding of reactive ﬂow transport . 3. Figure 3. which has a theoretical value of 4π 2 . whereas Fig.3. For the particular hydrothermal system considered in this application example. demonstrates that the produced chemical species in the chemical reaction may reach its equilibrium concentration in some regions. 1997a) for judging whether or not the convective pore-ﬂuid ﬂow is possible in such a hydrothermal system as considered here. Therefore. where both the material deformation due to thermal effects and the reactive mass transport are taken into account in the numerical computation. which is greater than the critical Rayleigh number.35 W/(m×0 C) porous matrix studied. This indicates that the convective poreﬂuid ﬂow may signiﬁcantly affect the chemical reactions (i. then the convective pore-ﬂuid ﬂow should take place in the hydrothermal system considered. which has a ﬂat bottom and is heated uniformly from below.07 × 10−5 (1/0 C) 3. which is the product of the chemical reaction considered in this application example.1 10−14 m2 2 × 1010 Pa 0.12 shows the normalized concentration distributions of reactant and product chemical species.4 Examples of the Proposed Consistent Point-Searching Interpolation Algorithm Table 3.13 shows the ﬁnal distribution of porosity and permeability in the computational domain of this application example. Nield and Bejan 1992.25 2. the temperature gradient is the main driving force to initiate the convective pore-ﬂuid ﬂow. the distributions of the three chemical species are different. reaction ﬂow) in the deformable porous medium.2. the convective pore-ﬂuid ﬂow should occur in the hydrothermal system. In particular. Zhao et al. There is a criterion available (Phillips 1991. This indicates that the numerical solution obtained from this application example can be qualitatively justiﬁed.

62 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems Fig.11 Distribution of pore-ﬂuid velocity and temperature in the hydrothermal system . 3.

3.3.12 Normalized concentration distribution of reactant and production chemical species in the hydrothermal system .4 Examples of the Proposed Consistent Point-Searching Interpolation Algorithm 63 Fig.

3.13 Distribution of ﬁnal porosity and permeability in the hydrothermal system .64 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems Fig.

10−7 kg/(m3 ×s) and 10−10 kg/(m3 ×s). 3. we can justify the numerical solution for the ﬁnal porosity and permeability in the hydrothermal system. the greater the porosity of the porous medium. Since the pre-exponential reaction rate constant. From the physics point of view.16 show the effects of chemical reaction rates on the normalized concentration distributions of reactant and product chemical species in the hydrothermal system. Even though the chemical reaction rate may affect the magnitude of the product chemical species concentration. it does not affect the overall distribution pattern of the product chemical species concentration. This further justiﬁes the numerical solutions obtained from this application example. represents how fast the chemical reaction proceeds. which are uniformly distributed in the computational domain. It is obvious that the chemical reaction rate has little inﬂuence on either the distribution pattern or the magnitude of the reactant chemical species concentration. Since the feedback effect of the medium deformation on the porosity and permeability has been considered in the numerical simulation.3. any numerical solutions have to be justiﬁed before they can be safely accepted and used. 3. This kind of phenomenon can be clearly observed from the related numerical solutions for the temperature distribution (Fig. Again. For this purpose. numerical rubbish)? From these . at least from a qualitative point of view.15 and 3. 3. 3. if all other parameters are kept unchanged in the hydrothermal system considered above.16 correct? Do they represent the physical and chemical characteristics of the hydrothermal system rather than something else (i.15 and 3. Figures 3. we must answer the following questions: Are the numerical results reported in Figs. As mentioned previously. have been used in the corresponding computations.e.14. produced new minerals).13) in the hydrothermal system. namely 10−5 kg/(m3 ×s). This ﬁnding implies that if the reactant minerals constitute only a small fraction of the whole matrix in a porous medium. 3. which is a commonly accepted assumption in geochemistry (Phillips 1991). three different values of the preexponential reaction rate constant.e. The next application example is to investigate how the chemical reaction rate affects the distribution of product chemical species (i. to a large extent. the larger the deformation of the porous medium due to the thermal effect. even though the magnitude of the normalized concentration of the product mineral may also strongly depend on the rates of chemical reactions.4 Examples of the Proposed Consistent Point-Searching Interpolation Algorithm 65 involved in ore body formation and mineralization in the upper crust of the Earth. This implies that for a porous medium. the distributions of the ﬁnal porosity and permeability are totally different from that of their initial values.11) and the ﬁnal porosity/permeability distributions (Fig. indicating that large deformation of a porous medium results in greater permeability of the porous medium. a region of relatively high temperature favours the formation of ﬂow channels because of an increase in the porosity of the region due to the thermal effect. since the solid aggregates in the porous medium are relatively stiff. the larger the deformation of the porous medium. the distribution pattern of the normalized concentration of the new mineral produced by chemical reactions is strongly dependent on the characteristics of the convective pore-ﬂuid ﬂow.14. On the other hand. we know that the higher the relative temperature in a porous medium.

66 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems Fig.14 Effects of chemical reaction rates on the normalized concentration distribution of the reactant chemical species (species 1) . 3.

3.15 Effects of chemical reaction rates on the normalized concentration distribution of the reactant chemical species (species 2) .4 Examples of the Proposed Consistent Point-Searching Interpolation Algorithm 67 Fig.3.

68 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems Fig. 3.16 Effects of chemical reaction rates on the normalized concentration distribution of the product chemical species (species 3) .

90) Since the reactant chemical species constitutes only a small fraction of the whole matrix in a porous medium. as is commonly assumed in geochemistry (Phillips 1991). for the hydrothermal system considered. rather than the concentration gradients of chemical species. but the magnitude of the product chemical species is strongly dependent on the rates of the chemical reaction (Fig. RT (3. RT (3. even though the contribution of chemical species to the buoyancy is taken into account.14 and 3. rather than the concentration gradients of chemical species. analytically. we need to rewrite the transport equation (see Eq.15). For the hydrothermal system considered. For the purpose of facilitating the theoretical analysis. This means that the buoyancy produced by temperature is much greater than that produced by the concentration of the reactant chemical species. Next. the convective pore-ﬂuid ﬂow is predominantly driven by the temperature gradient.e. 3. we will examine.5)) for one of the reactant chemical species (i. can we ﬁnd something of more theoretical importance? In the following we will answer the above questions through some theoretical analysis. while the maximum change in the density of pore-ﬂuid due to the concentration of any reactant chemical species is only about 0. the maximum change in the density of poreﬂuid due to temperature is about 4%. This means that the concentrations of chemical species have minimal inﬂuence on the convective pore-ﬂuid ﬂow. More speciﬁcally. In this case. This implies that the distribution of the reactant chemical species is controlled by the pore-ﬂuid velocity and the dispersivity of the porous medium.91) ρf0 u = ρ f 0 Dex + φk A C 1 C 2 exp . the reaction term in this equation is at least the second order small quantity. This is the reason why both the pattern and the magnitude of the normalized concentration of the reactant chemical species are very weakly dependent on the rates of the chemical reaction in the related numerical results (Figs.1%.4 Examples of the Proposed Consistent Point-Searching Interpolation Algorithm 69 numerical results. (3. we can view the normalized concentration of the reactant chemical species as the ﬁrst order small quantity in the above equation. 3. Based on this recognition. why the distribution pattern of the normalized concentration of the product chemical species is almost independent of the rates of the chemical reaction. the velocity components of pore-ﬂuid in the system are strongly dependent on the temperature gradient. Thus.16). at least from the mathematical point of view. As a result. rather than by the chemical reaction term unless the rate of chemical reaction is very fast. we need to rewrite the transport equation for chemical species 3 as follows: ∂C 3 ∂C 3 +v ∂x ∂y ∂ 2C 3 ∂ 2C 3 + Dey ∂x2 ∂ y2 −E a . we can further explore why both the pattern and the magnitude of the normalized concentration of reactant chemical species are very weakly dependent on the rates of the chemical reaction. species 1) as follows: ρf0 u ∂C 1 ∂C 1 +v ∂x ∂y = ρ f 0 Dex ∂ 2C 1 ∂ 2C 1 + Dey 2 ∂x ∂ y2 − φk A C 1 C 2 exp −E a .3.

91). As mentioned before.16.91).92) by k AB /k A A yields the following equation: ρf0 k AB kAA u ∂C 3A ∂C 3A +v ∂x ∂y = ρf0 k AB kAA Dex ∂ 2 C 3A ∂ 2 C 3A −E a . the product of C 1 and C 2 is almost independent of the chemical reaction rate in Eq.91). then we ﬁnd that the normalized concentration of the product chemical species. the solution corresponding to the reaction rate constant k AB can be straightforwardly expressed as C 3B = kAB C 3A . we have the following equation: ρf0 u ∂C 3A ∂C 3A +v ∂x ∂y = ρ f 0 Dex ∂ 2 C 3A ∂ 2 C 3A + Dey 2 ∂x ∂ y2 + φk A A C 1 C 2 exp −E a . (3. kAA (3. varies linearly with the reaction rate constant.94) where C 3B is the solution corresponding to the reaction rate constant k AB for Eq. the product of C 1 and C 2 is at least the second order small quantity in the theoretical analysis. we can. 3. C 1 and C 2 are eventually independent of the chemical reaction rate in Eq. the numerical analysis carried out for this application example is further validated by the related theoretical analysis.70 3 Algorithm for Simulating Coupled Problems in Hydrothermal Systems Since the magnitude of the normalized concentration of the reactant chemical species is very weakly dependent on the rates of the chemical reaction. (3. + Dey +φk AB C 1 C 2 exp ∂x2 ∂ y2 RT (3. (3.91). It needs to be pointed out that the proposed consistent point-searching interpolation and the related solution methodology have been successfully applied to more . the normalized concentration of chemical species 3 is at least the second (or higher) order small quantity.92) Multiplying Eq. (3. This is the reason why the distribution pattern of concentration of the product chemical species is almost independent of the rates of the chemical reaction. (3. deduce the solution for another reaction rate constant k AB below.91).91). Equation (3.16). 3. If C 3A is a solution corresponding to the reaction rate constant k A A for Eq. If we further observe the numerical solutions in Fig. (3. More speciﬁcally. the magnitude of the normalized concentration of chemical species 3 is strongly dependent on the rate of the chemical reaction because the product of φ. indeed. but the magnitude of the product chemical species is strongly dependent on the rates of the chemical reaction (Fig.94) states that for any particular point in the hydrothermal system considered. This means that the magnitude of the normalized concentration of chemical species 3 is determined by the reaction term in Eq. Therefore.93) Therefore. the normalized concentration of the product chemical species varies linearly with the reaction rate constant. mathematically. Since C 3A is a solution for Equation (3. As a result. RT (3.

Ord et al. Hobbs et al 2000. Gow et al. it can equally be extended to the unstructured meshes consisting of higher order elements. Sorjonen-Ward et al. where both complicated geometry and material nonhomogeneity are considered. 1998b. Schaubs and Zhao 2002. . In the latter case.3. Zhang et al. 2002. instead of analytically. 2002. one has to solve the inverse mapping problem numerically. Although the consistent point-searching interpolation algorithm is developed for the unstructured meshes consisting of 4-node bilinear quadrilateral elements. 2002. 1999c.4 Examples of the Proposed Consistent Point-Searching Interpolation Algorithm 71 realistic geological situations (Zhao et al. 2003).

heat transfer and species transport/chemical reactions in pore-ﬂuid saturated porous rock masses (Zhao et al. (6) various aspects of the fully coupled problem involving material deformation. 1999d). 1997a. For a natural ﬂuid-rock interaction system. pore-ﬂuid ﬂow. C Springer-Verlag Berlin Heidelberg 2009 73 . The above-mentioned work has signiﬁcantly enriched our knowledge about the physical and chemical processes related to ore body formation and mineralization in the upper crust of the Earth. we have been making efforts to develop a practical and predictive tool to explore for giant ore deposits in the upper crust of the Earth. DOI 10.Chapter 4 A Term Splitting Algorithm for Simulating Fluid-Rock Interaction Problems in Fluid-Saturated Hydrothermal Systems of Subcritical Zhao Numbers In recent years. (2) Can convective pore-ﬂuid ﬂow take place in the upper crust of the Earth if there is a ﬂuid/mass leakage from the mantle to the upper crust of the Earth? On the modelling development side. Zhao et al. 1999f. (4) double diffusion driven pore-ﬂuid ﬂow in hydrothermal systems (Zhao et al.1007/978-3-540-89743-9 4. it is necessary to extend further the developed numerical tools to solve ﬂuid-rock interaction problems. the chemical dissolution front is stable if the Zhao number of the system is subcritical. while it becomes unstable otherwise. 1998e. Towards this goal. 1999b): (1) Can the pore-ﬂuid pressure gradient be maintained at the value of the lithostatic pressure gradient in the upper crust of the Earth? and. 1999a). 2000b). Fundamentals of Computational Geoscience. signiﬁcant progress has been made towards a better understanding of the basic physical and chemical processes behind ore body formation and mineralization in hydrothermal systems. (3) precipitation and dissolution of minerals in the upper crust of the Earth (Zhao et al. 1998a. This requires us to deal with coupled reactive species transport phenomenon. the reactant chemical species constitutes only a small fraction of the whole matrix in a porous rock (Phillips 1991). (2) coupled reactive pore-ﬂuid ﬂow and species transport in porous media (Zhao et al. 1999g). On the scientiﬁc development side. Lecture Notes in Earth Sciences 122. we have developed numerical methods to model the following problems: (1) convective pore-ﬂuid ﬂow in hydrothermal systems (Zhao et al. In this case. Based on the concept of the C. (5) pore-ﬂuid ﬂow patterns near geological lenses in hydrodynamic and hydrothermal systems (Zhao et al. which is the direct consequence of the chemical reactions that take place between aqueous reactive species in pore-ﬂuid and solid minerals in pore-ﬂuid saturated porous rock masses. 1999c. 2000a). we have developed analytical solutions to answer the following scientiﬁc questions (Zhao et al. 1998b). Since ﬂuid-rock interaction is another potential mechanism of ore body formation and mineralization in the upper crust of the Earth..

φ f D(φ f ) 1 kchemical Ceq (4.6) Equations (4.3).1) where ν f low is the Darcy velocity of the pore-ﬂuid ﬂow. φ f and D φ f are the ﬁnal porosity of the porous medium and the corresponding diffusivity of chemical species after the completion of soluble mineral dissolution. hydrodynamic. (4. (4. Ap (4.5) FShape = Vp . kchemical is the rate constant of the chemical reaction. V p is the average volume of the soluble grain. (4. kchemical A p Ceq (4.5) and (4.74 4 Algorithm for Simulating Fluid-Rock Interaction Problems generalized dimensionless pore-ﬂuid pressure gradient (Zhao et al. FDiffusion is a term to represent the solute diffusion/dispersion.2). (4. FChemical is a term to represent the chemical kinetics of the dissolution reaction. A p is the averaged surface area of the soluble grain. thermodynamic and chemical kinetic characteristics of a ﬂuid-rock system in a comprehensive manner. Ceq is the equilibrium concentration of the chemical species. 2008e). These terms can be expressed as follows: FAdvection = νﬂow . (4. the corresponding dimensionless Zhao number of a single mineral dissolution system can be deﬁned as follows: Zh = νﬂow φ f D(φ f ) Vp .4) FChemical = .3) (4. Eq. To understand the physical meanings of each term in the Zhao number.2) where FAdvection is a term to represent the solute advection. The condition under which a chemical dissolution front in a two-dimensional ﬂuid-saturated porous rock becomes unstable can be expressed by the critical value of this dimensionless number as follows: .4).6) clearly indicate that the Zhao number is a dimensionless number that can be used to represent the geometrical.1) can be rewritten in the following form: Z h = FAdvection FDiffusion FChemical FShape . (4. FDiffusion = 1 . FShape is a term to represent the shape factor of the soluble mineral in the ﬂuid-rock interaction system.

This means that the equilibrium state of the chemical reaction involved is always attained in this category of reactive species transport problem. For this reason. at least from the mathematical point of view. In contrast. k(φ f ) (4. The focus of this chapter is to deal with ﬂuid-rock interaction systems of subcritical Zhao numbers. namely homogeneous and heterogeneous reactions. in which the rate of the relevant chemical reaction is a positive real number of ﬁnite value. The intermediate case between the ﬁrst and the second category of reactive species transport problem belongs to the third category of reactive species transport problem. Due to different regimes in which a chemical reaction proceeds. of the porous rock. 1998a).4 Algorithm for Simulating Fluid-Rock Interaction Problems 75 Z h critical = (3 − β)(1 + β) . for the second category of reactive species transport problem. φ f . the chemical reaction takes place at the surfaces between reactive aqueous species and solid minerals. However. k(φ0 ) is the initial permeability corresponding to the initial porosity of the porous rock. the time scale of the convective/advective ﬂow is much smaller than that of the relevant chemical reaction in porous rock masses so that the rate of the chemical reaction can be essentially taken to be zero in the numerical analysis. the chemical reaction takes place solely between reactive aqueous species. This implies that both solid and ﬂuid phases need to be considered in the numerical modelling of reactive species transport problems with heterogeneous reactions. while the focus of the next chapter is to deal with ﬂuid-rock interaction systems of critical and supercritical Zhao numbers. Using the concepts of both the Zhao number and the corresponding critical one. In terms of numerical modelling of coupled reactive species transport phenomena in pore-ﬂuid saturated porous rock masses. In the ﬁrst category of reactive species transport problem. the ﬁrst category of species transport problem is often called the non-reactive mass transport problem. 2(1 − β) β= k(φ0 ) . For homogeneous reactions.7) where Z h critical is the critical Zhao number of the ﬂuid-rock interaction system. k(φ f ) is the ﬁnal permeability corresponding to the ﬁnal porosity. we have divided the reactive transport problems into the following three categories (Zhao et al. It is the chemical kinetics of a chemical reaction that describes the reaction term in a reactive species transport equation. Another signiﬁcant characteristic of the third category of reactive species transport problem is that the detailed chemical kinetics of chemical reactions must be taken into account. Although signiﬁcant achievements have been made for the numerical modelling of non-reactive species and quasi-instantaneous equilibrium reaction transport problems. As a result. there are two fundamental reactions. the instability of a chemical dissolution front in a ﬂuid-rock interaction system can be determined. research on the numerical modelling of the third category of reactive species transport problem is rather limited (Steefel and . the time scale of the convective/advective ﬂow is much larger than that of the relevant chemical reaction in pore-ﬂuid saturated porous rock masses so that the rate of the chemical reaction can be essentially taken to be inﬁnite. in geochemistry. the second category of reactive species transport problem is called the quasi-instantaneous equilibrium reaction transport problem. for heterogeneous reactions.

1999a). there are two ways to overcome this difﬁculty. Generally. is the main driving force causing the heterogeneous chemical reactions at the interface between the pore-ﬂuid and solid minerals so that the rock can be changed from one type into another. Due to the dissolution and precipitation of minerals. Raffensperger and Garven 1995). The reason for this is that these reactions between the aqueous species and solid minerals may result in dissolution of one kind of mineral and precipitation of another and therefore. the pore-ﬂuid ﬂow. 4. the porosity of the porous rock mass evolves during the rock alteration.1 Key Issues Associated with the Numerical Modelling of Fluid-Rock Interaction Problems Of central importance to the numerical modelling of ﬂuid-rock interaction problems in pore-ﬂuid saturated hydrothermal/sedimentary basins is the appropriate consideration of the heterogeneous chemical reaction which takes place slowly between aqueous reactive species in the pore-ﬂuid and solid minerals in ﬂuid-saturated porous rock masses. Here we will extend the numerical method developed to model coupled reactive multi-species transport problems with heterogeneous reactions. we have successfully used the ﬁnite element method to model coupled reactive multi-species transport problems with homogeneous reactions (Zhao et al. Since a change in porosity can result in a change in the pore-ﬂuid ﬂow path. from the computational point of view. Since both the mesh Peclet number and Courant number are dependent on the average linear velocity. The second one is to regenerate the mesh and re-determine the time step of integration at every time step of computation so that the ﬁnite element method can be used effectively. Once a dissolving mineral is exhausted in the rock mass. a feed-back loop is formed between the pore-ﬂuid ﬂow and the transport of reactive chemical species involved in heterogeneous chemical reactions in ﬂuid-rock interaction systems and this porosity change in the pore-ﬂuid ﬂow needs to be considered in the numerical modelling of ﬂuid-rock interaction problems. which carries reactive aqueous species. Considering this fact. the reactive minerals may be changed from one type into another type.76 4 Algorithm for Simulating Fluid-Rock Interaction Problems Lasaga 1994. . This implies that the average linear velocity of pore-ﬂuid ﬂow varies with time due to this porosity evolution. there is a deﬁnite need for developing new numerical algorithms to deal with this kind of problem. Another important issue related to the numerical modelling of ﬂuid-rock interaction problems is that the dissolution rates of minerals are dependent on the existence of the dissolving minerals. From the rock alteration point of view. we have to overcome a difﬁculty in dealing with the problem of variable Peclet and Courant numbers in the transient analysis of ﬂuid-rock interaction problems. Since either the use of a very ﬁne mesh in the whole process of computation or the regeneration of the mesh at every time step of computation is computationally inefﬁcient. The ﬁrst one is to use a very ﬁne mesh and very small time step of integration so that the requirements for both the mesh Peclet number and Courant number can be satisﬁed at every time step of the computation.

Taking the above-mentioned factors into account. is considered as a representative example in this chapter. there is a need to consider the appropriate dependence of the dissolution rate of a dissolving mineral on its generalized concentration in the numerical analysis. using the ﬁnite element method. in which K-feldspar (KAlSi3 O8 ) and/or muscovite (KAl3 Si 3 O10 (OH)2 ) are dissolved and muscovite and/or pyrophyllite (Al2 Si 4 O10 (OH)2 ) are precipitated. we focus this study on the numerical modelling of mixed solid and aqueous species transport equations with consideration of reaction terms from heterogeneous. the numerical modelling may violate the real mechanism of the related chemical kinetics and produce incorrect numerical results.2 Development of the Term Splitting Algorithm For ﬂuid-rock interaction problems in pore-ﬂuid saturated hydrothermal/ sedimentary basins.2 Development of the Term Splitting Algorithm 77 the dissolution rate of this particular mineral must be identically equal to zero. heterogeneous chemical reactions take place at the interface between the reactive aqueous species in the pore-ﬂuid and solid minerals in the rock mass. namely the conventional mass transport equation for the aqueous chemical species in the pore-ﬂuid and the degenerated mass transport equation for the solid minerals in the rock mass. This means that we need to deal with two types of transport equations in the numerical modelling of ﬂuid-rock interaction problems. can be solved simultaneously in computation. This means that we will concentrate on the development of new concepts and algorithms so as to solve the ﬂuid-rock interaction problems effectively and efﬁciently. 4. For this purpose. The second is the degenerated transport equation. the problem of variable mesh Peclet number and Courant number needs to be converted into a problem involving constant mesh Peclet and Courant numbers. in order to effectively and efﬁciently use the ﬁnite element method for solving ﬂuid-rock interaction problems in pore-ﬂuid saturated hydrothermal/sedimentary basins. This indicates that the variation in the amount of the dissolving mineral should be considered in the dissolution rates of the minerals. isothermal chemical reactions. Otherwise. Thus. (3) Considering the direct consequence of the porosity evolution with time in the transient analysis of ﬂuid-rock interaction problems. in which the advection/convection term and . so that the conventional ﬁnite element method can be directly used to solve ﬂuid-rock interaction problems.4. new concepts and numerical algorithms need to be developed for dealing with the following fundamental issues: (1) Since the ﬂuid-rock interaction problem involves heterogeneous chemical reactions between reactive aqueous chemical species in the pore-ﬂuid and solid minerals in the rock masses. so that two types of reactive mass transport equations. a ﬂuid-rock interaction problem. it is necessary to develop a new concept involving the generalized concentration of a mineral. (2) Since the reaction area between the pore-ﬂuid and mineral surfaces is basically a function of the generalized concentration of the solid mineral. The ﬁrst is the conventional transport equation with the advection/convection term and the diffusion/dispersion term for reactive aqueous species in the pore-ﬂuid.

(4. the general form of the ﬁrst type of transport equation in a two dimensional pore-ﬂuid saturated. Therefore.9) where CCi is the conventional concentration of aqueous species i. Nield and Bejan 1992. For the reactive aqueous species. . even though . In a geological system. n). .10) where Di is the dispersivity of aqueous species i. . . Die is of the following form: Die = φDi . the volume of void plus the volume of solid particles) of the porous rock mass. where the reactant chemical species constitute only a small fraction of the whole matrix in a porous medium (Phillips 1991). 1998e. In other words. groundwater can be pumped out from a horizontal aquifer at a constant ﬂow rate. (4. The reason for this is that the Darcy velocity is the velocity averaged over the total area of a representative elementary area (Zhao et al. which are perpendicular to the direction of pore-ﬂuid ﬂow. . Phillips 1991. u x and u y are the Darcy velocities of pore-ﬂuid in the x and y directions respectively. It needs to be pointed out that under some circumstances. 2. The generalized concentration of a solid mineral is deﬁned as the moles of the solid mineral per unit total volume (i. m). we need to develop the new concept of the generalized concentration for the solid mineral.e. φ is the porosity of the porous rock mass. the Darcy velocity in the above-mentioned aquifers can be maintained at a constant value. the Darcy velocity in the ﬂow direction is constant if the horizontal aquifer has a constant thickness. 1994c). for solid minerals in the rock mass. Zhao et al. topographically induced pore-ﬂuid ﬂow can also ﬂow through a horizontal aquifer at a constant ﬂow rate.8) where C Gi is the generalized concentration of solid mineral/species i. the total ﬂux of the pore-ﬂuid ﬂow should be constant through all vertical cross-sections. This indicates that in order to satisfy the mass conservation requirement of the pore-ﬂuid in the above-mentioned aquifers. the general form of the second type of transport equation to be solved in this study can be expressed as follows: ∂C Gi = φRi ∂t (i = 1. in a groundwater supply system. m + 2. Using this new concept. . In order to solve these two types of transport equations simultaneously. Ri is the reaction rate of solid mineral/species i and m is the number of the solid minerals to be considered in the system. .78 4 Algorithm for Simulating Fluid-Rock Interaction Problems the diffusion/dispersion term are identically equal to zero. For example. isotropic and homogeneous porous medium reads φ ∂CCi ∂CCi ∂CCi + ux + uy = Die ∇ 2 CCi + φRi ∂t ∂x ∂y (i = m + 1. rather than over the pore area of the representative elementary area. the total ﬂux of pore-ﬂuid ﬂow in a horizontal aquifer may be approximately considered as a constant. . (4.

11) where Wi is the molecular weight of mineral i. r j is the overall dissolution rate of mineral j. (4. varies if the vertical cross-sections have different values of the areal porosity.8) and (4. ρs is the unit weight of solid minerals.11) is called the evolution equation of porosity with time.4. k j and K j are the reaction constant and the equilibrium constant of the jth heterogeneous chemical reaction. Using the generalized concentration of solid minerals. ρs (4. the linear averaged velocity. This equation describes the mass conservation of species i in the pore-ﬂuid saturated porous medium for both constant and variable porosity cases (Zhao et al. It needs to be pointed out that the surface area of a dissolving mineral varies in the process of mineral dissolution. rj = Qj Aj kj 1 − Vf Kj . the surface area of the mineral must be equal to zero and therefore. 1994c). From the chemical kinetics point of view. V f is the volume of the solution. 1998e. the reaction rates involved in Eqs. Phillips 1991. the evolution equation for the porosity of the rock mass can be expressed as follows: m φ =1− i=1 Wi C Gi . Nield and Bejan 1992. which is the velocity averaged over the pore area of a representative elementary area.12) where τij is the contribution factor of mineral/species j to the reaction rate of mineral/species i. may have different values of the areal porosity (Zhao 1998e. (4. which are perpendicular to the direction of the pore-ﬂuid ﬂow. It is also noted that Eq.9) can be expressed as follows (Lasaga 1984): N Ri = − j=1 τij r j . the chemical reaction dissolving this particular mineral should stop. In order to consider the feedback effect of porosity due to heterogeneous chemical reactions. (4. (4. Once the dissolving mineral is depleted in the rock mass. If this area is simply assumed to be a constant in the numerical modelling. C Gi are the functions of space and time variables. For the transient analysis. 1994c). N is the number of the heterogeneous chemical reactions to be considered in the rock mass. so that the porosity of the rock mass varies with both space and time. Q j is the chemical afﬁnity of the jth heterogeneous chemical reaction. (4. . we need a porosity evolution equation in the numerical analysis.13) where A j is the surface area of mineral j. Zhao et al. then the resulting numerical solutions are incorrect because the real mechanism of chemical kinetics cannot be simulated during the consumption of the dissolving mineral in the rock mass.2 Development of the Term Splitting Algorithm 79 the vertical cross-sections. However. This is the reason why Eq.9) is a typical transport equation for species i.

13) yields the following equation: r j = α j CGj k j 1 − q Qj Kj . This is the ﬁrst advantage of introducing the new concept of the generalized concentration for solid minerals.9)) can be solved simultaneously. the generalized concentration of this mineral continuously evolves to zero and therefore. This indicates that the linear average velocity of pore-ﬂuid.8) and (4. the values of both α j and q are dependent on the constituents. the use of a constant surface area cannot automatically terminate the dissolution reaction. we can establish the following relation between the surface area and the generalized concentration of the dissolving mineral: Aj q = α j CGj .14) where α j and q are positive real numbers. Since several heterogeneous chemical reactions take place simultaneously in dissolution problems involving multiple minerals. Inserting Eq. the values of both α j and q need to be determined by ﬁeld measurements and laboratory tests. whereas in the case of packing uniform spheres in a three dimensional domain. in the case of packing uniform circles in a two dimensional domain. In addition. Since the surface area of a dissolving mineral is strongly dependent on its generalized concentration. even though the dissolving mineral is absolutely exhausted in the rock mass. which is involved in the advection/convection term of a reactive aqueous species transport equation. (4. in which the horizontal Darcy velocity may be constant. For example. because it is inversely proportional . these values can be determined analytically. varies with time during the transient analysis. (4. If the transient process of a heterogeneous chemical reaction is of interest. it is important to solve simultaneously all the transport equations with heterogeneous reaction terms in ﬂuid-rock interaction problems. For simple packing of loose uniform particles of the minerals. then the dissolution and precipitation of minerals can result in the variation of porosity with time. q = 2/ 3. the related average linear velocity still varies with time. However. if we want to simulate the chemical kinetics of these heterogeneous chemical reactions correctly. Even for a horizontal aquifer. the use of a constant surface area cannot simulate the dependent nature of the chemical reaction rate on a change in the amount of the dissolving mineral during the heterogeneous chemical reaction process. its dissolution rate is automatically set to be zero when it is completely depleted in the numerical analysis. Vf (4. for a real rock mass in pore-ﬂuid saturated hydrothermal/sedimentary basins. q = 1/ 2. Like the surface area of the dissolving minerals. Another advantage of introducing the concept of the generalized concentration for solid minerals is that both the ﬁrst type and second type of transport equations (i. Eqs.80 4 Algorithm for Simulating Fluid-Rock Interaction Problems In other words. this equation states that once the dissolving mineral is gradually consumed in the rock mass.14) into Eq.15) Clearly. packing form. (4.e. (4. grain size and so forth of the minerals.

m +2. .16) Dividing both the left and right hand sides of Eq. and a variable term. The basic idea behind this algorithm is that through some mathematical manipulations. This term splitting process. In this regard.18) and the rest terms are moved to the right hand side. . . n). If the ﬁrst derivative terms with constant coefﬁcients are kept in the left hand side of Eq. Eq. we develop the term splitting algorithm in this section.19) . and a variable term. . u y0 . Eq. m +2. (4. n). (u y φ0 )/φ − u y0 . .18) It is noted that in Eq.2 Development of the Term Splitting Algorithm 81 to the porosity of the aquifer.9) into the following form: ∂CCi ∂CCi φ φ ∂CCi +u x +u y = Die ∇ 2 CCi + φ0 Ri (i = m +1.17) can be rewritten as follows: φ0 ∂CCi ∂CCi φ0 + u x0 + u x − u x0 + u y0 + ∂t φ ∂x φ0 e 2 D ∇ CCi + φ0 Ri = φ i (i = m + 1.4. m + 2. we invert the problem of variable Peclet and Courant numbers into that of constant Peclet and Courant numbers so that a ﬁxed ﬁnite element mesh and ﬁxed time step of integration can be safely used in the numerical modelling of ﬂuid-rock interaction problems. the coefﬁcient in front of ∂CCi /∂ x has been split into a constant coefﬁcient. we have to deal with the problem of variable Peclet and Courant numbers in the transient analysis of ﬂuid-rock interaction problems. Similarly. u x0 . (4. . (4. . (4. . m + 2. . . To solve this problem effectively and efﬁciently. n). φ0 u y − u y0 φ ∂CCi ∂y (4. (4. (u x φ0 )/φ − u x0 .18) can be rewritten as follows: φ0 ∂CCi ∂CCi ∂CCi e + u x0 + u y0 = D0i ∇ 2 CCi + φ0 Ri + Rie ∂t ∂x ∂y (i = m + 1. u x0 and u y0 . . (4. forms a very important step in the proposed term splitting algorithm. the coefﬁcient in front of ∂CCi /∂ y has also been split into a constant coefﬁcient. φ0 φ0 ∂t ∂x ∂y φ0 (4. (4. as reference velocities in the x and y directions respectively. (4.17) Using the initial Darcy velocities. n). If the pore-ﬂuid saturated porous medium has a non-zero initial porosity. . . . . then we can use this initial porosity as a reference porosity and change Eq. .18). although it is carried out mathematically.16) by φ / φ0 yields the following equation: φ0 ∂CCi φ0 ∂CCi φ0 ∂CCi φ0 e 2 + ux + uy = D ∇ CCi +φ0 Ri ∂t φ ∂x φ ∂y φ i (i = m +1.

m + 2. we need only to deal with a problem with constant mesh Peclet number and Courant number.21) (i = m + 1. which has constant coefﬁcients in front of the ﬁrst and second derivatives. a change either in the porosity or in the pore-ﬂuid velocity is equivalent to the addition of a source/sink term into the representative elementary volume/area. instead of Eq. As shown in Fig.9). n).19). Second. from the computational point of view. (4. It is clear that the proposed term splitting algorithm consists of the following two main steps.1. (4.20) (4. in the x andy directions.9) are mathematically equivalent. 4. .19)) for ﬂuidrock interaction problems in pore-ﬂuid saturated hydrothermal/sedimentary basins.19).19) can be solved much easier than Eq. we need to deal with a problem with variable mesh Peclet number and Courant number. (4. which are the initial Darcy velocities. The reason for this is that in order to solve Eq.3 Application Examples of the Term Splitting Algorithm In order to illustrate the usefulness and applicability of the newly proposed concepts and numerical algorithms. (4. Eq. (4. .19) is directly used to obtain a numerical solution for Equation (4. (4. n).9) with variable coefﬁcients in front of the ﬁrst and second derivatives has been changed into Eq.19).19) and (4. First. (4.9). However.9) in the ﬁnite element analysis. Eq. by means of some rigorous mathematical manipulations. u x0 and u y0 are the reference velocities. Since the constants involved in both the advection and dispersion terms are basically independent of time. . the numerical solvers presently available are more stable and robust when they are used to solve Eq. . (4. the ﬁnite element method with the proposed term splitting algorithm can be efﬁciently used to solve both the ﬁrst and second type of transient transport equations (i. Eq. .9) in the ﬁnite element analysis. which are determined using the initial conditions at the beginning of computation.82 4 Algorithm for Simulating Fluid-Rock Interaction Problems Rie = u x0 − ∂CCi ∂CCi φ0 φ0 ux + u yo − u y φ ∂x φ ∂y e D0i = φ0 Di (i = m + 1. m + 2. (4. . From the mathematical point of view. On the other hand. (4. Rie is the equivalent source/sink term due to the variation in both the velocity of pore-ﬂuid and the porosity of the rock mass. we have built them into a ﬁnite element code so that ﬂuidrock interaction problems in pore-ﬂuid saturated hydrothermal/sedimentary basins can be solved effectively and efﬁciently. this means that we have changed a partial differential equation with variable coefﬁcients (Eq.e. because both Eqs. (4. it is possible to use the initial mesh and time step. Eqs. As a result. (4.9)) into another one with constant coefﬁcients in front of the ﬁrst and second derivatives. the application exam- . Note that the equivalent source/sink term presented here has a very clear physical meaning.8) and (4. throughout the whole process of the transient analysis of ﬂuid-rock interaction problems. But in order to solve Eq. . where φ0 is the initial porosity of the porous rock mass. 4. (4. For a representative elementary volume/area. .

Basically. 4.24) The ﬁrst reaction (i. If the injected carbon dioxide ﬂow is strong enough.23) describes a heterogeneous chemical reaction. fast (4. As a direct consequence. Chemically.23) 2 2KAl3 Si 3 O10 (OH)2 + 2H + + 6SiO2 =⇒ 2K + + 3Al2 Si 4 O10 (OH)2 . it reacts very fast with water (H2 O). (4. while Eq. the use of this equivalence may result in a reduction in the number of primary reactive chemical species and therefore. The rock of the aquifer is initially composed of K-feldspar (KAlSi3 O8 ) and quartz (SiO2 ). the following three overall chemical reactions may take place in the aquifer. Eq. As will be demonstrated later.e.3 Application Examples of the Term Splitting Algorithm 3000 m 83 Fig.22)) states that when the injected carbon dioxide (CO2 ) gas enters the ﬂuid-rock system. It needs to be pointed out that Eq. a considerable reduction in the degrees of freedom of the whole system. in which muscovite (KAl3 Si 3 O10 (OH)2 ) is dissolved and pyrophyllite (Al2 Si 4 O10 (OH)2 ) is precipitated in the ﬂuid-rock system. (4. so that the chemical equilibrium can be reached quasi-instantaneously.24) describes a heterogeneous chemical reaction. 3 1 3KAlSi3 O8 + 2H + =⇒ 2K + + KAl3 Si 3 O10 (OH)2 + 6SiO2 .22) k (4. There is an injection of carbon dioxide gas (CO2 ) at the left inlet of the aquifer. This means that the injection of carbon dioxide (CO2 ) gas is equivalent to the direct injection of H + into the system.1 Geometry and initial conditions for the ﬂuid-rock interaction problem in a pore-ﬂuid saturated aquifer CO 2 99% Quartz + 1% K-feldspar φ0 = 0. ux 0 = 10−8 m/s 1000 m ple considered in this section is an isothermal ﬂuid-rock interaction problem in a pore-ﬂuid saturated horizontal aquifer within a sedimentary basin.1. in which K-feldspar (KAlSi3 O8 ) is dissolved and muscovite (KAl3 Si 3 O10 (OH)2 ) is precipitated.4. . CO2 + H2 O =⇒ HCO− + H + . The topographically induced pore-ﬂuid ﬂow is horizontally from the left to the right of the aquifer. This can lead to a signiﬁcant reduction in requirements of both the computer storage and CPU time in the numerical modelling of ﬂuid-rock interaction problems. The injected carbon dioxide ﬂow in the aquifer may dissolve K-feldspar and precipitate muscovite (KAl3 Si 3 O10 (OH)2 ). the number of reactive species transport equations can be reduced to the minimum in computation. k (4. (4. the precipitated muscovite may be dissolved and pyrophyllite ( Al2 Si 4 O10 (OH)2 ) will be precipitated. This means that the horizontal Darcy velocity is constant within the aquifer.

4 × 10−3 kmol/m3 at the left vertical boundary of the computational domain. Figures 4.5 in the numerical analysis. Eq. the integration time step is set to be 3 × 108 s. which is approximately equal to 10 years. the dispersion coefﬁcient of the injected H + . The concentration of the injected H + is 6.2. For the purpose of simulating the chemical kinetics of heterogeneous reactions. there only exists the precipitation front (i. The precipitation front of muscovite is the region from blue to red in the right half of the computational domain.3 and 4.84 4 Algorithm for Simulating Fluid-Rock Interaction Problems Eq. 4.0 for the K-feldspar and muscovite dissolution reactions. is 100 times that of K + in the computation. It is observed the dissolution front (i. Similarly. Since muscovite can be precipitated and dissolved at the same time. (4. in the computation. Since the injected carbon dioxide (CO2 ) gas diffuses much faster than aqueous species. α j ( j = 1.5 × 1010 s. the region from red to blue in Fig.87 kmol/m3 respectively.3. muscovite and pyrophyllite in the ﬂuid-rock interaction system at four different time instants.24) states that the dissolution of one mole of muscovite needs to consume one mole of H + and two moles of quartz and then produces 1. Since the problem to be considered here is essentially an initial value problem. muscovite.89 × 107 and 8318. (4.1. 2/ 3 moles of K + and two moles of quartz.1 kmol/m3 and 1. in the computation. namely two aqueous species (K + and H + ) in the pore-ﬂuid and four solid minerals/species (K-feldspar.82 kmol/m3 and 0. This indicates that we only need to consider six primary chemical reactive species. As shown in Fig.4). This computational domain is discretized into 2700 4-node quadrilateral elements. t = 1.23) states that the dissolution of one mole of K-feldspar needs to consume 2/ 3 moles of H + and then produces 1/ 3 moles of muscovite. whereas the dissolution front is the region from red to blue in the left half of the computational domain. there are two propagation fronts. the initial values of the conventional concentrations of K + and H + are 0. 4.5 × 1011 s respectively.5 moles of pyrophyllite and one mole of K + . It is assumed that all the aqueous reactive species involved in the chemical reactions are in chemical equilibrium at the beginning of the computation. the following thermodynamic data are used.2) of K-feldspar propagates from the left side to the right side of the computational domain.4 show the generalized concentration distributions of K-feldspar. In addition. In the case of the generalized concentration distribution of pyrophyllite (see Fig. namely t = 3 × 109 s.6 × 10−5 kmol/m3 . the precipitation front and dissolution front. pyrophyllite and quartz) in the rock mass. which is the equivalence of the injected carbon dioxide (CO2 ) gas. Under this assumption.1. 2) and q are assumed to be unity and 0. The horizontal Darcy velocity of pore-ﬂuid ﬂow is 10−8 m/s in the horizontal aquifer. t = 6 × 1010 s and t = 1. the following initial conditions are used in the computation. the region . The initial values of the generalized concentrations of quartz and K-feldspar are 36. whereas the chemical reaction rate constants for these two dissolution reactions are 5. Both the precipitation and dissolution fronts propagate from the left side to the right side of the computational domain.48 × 10−12 kmol/(m2 s) respectively.e. The dispersion coefﬁcient of K + is 2 × 10−6 m2 /s. in Fig. The initial porosity of the porous medium is 0.e. The chemical equilibrium constants are 3. the computational domain to be used in the numerical analysis is a rectangle of 3000 m by 1000 m in size. 4. 4.03 × 10−12 kmol/(m2 s) and 4. 4.

517E–02 0.850E–01 0.428E–01 0.355E–01 0.777E–01 0.729E–01 0.181E–01 0.4.5 × 1010 s) 0.258E–02 (t = 1.155E–01 0.845E–01 0.855E–01 0.707E–01 0.860E–01 0.281E–01 (t = 6 × 1010 s) 0.791E–01 0.865E–01 0.824E–01 0.103E–01 0.775E–02 0.840E–01 0.576E–01 0.649E–01 0.502E–01 0.754E–01 0.835E–01 0.847E–01 0.129E–01 0.5 × 1011 s) Fig.684E–01 (t = 1. 4.2 Distribution of K-feldspar in the ﬂuid-rock interaction system .3 Application Examples of the Term Splitting Algorithm 85 0.800E–01 0.207E–01 0.796E–01 0.830E–01 (t = 3 × 109 s) 0.

5 × 1011 s) Fig.118E–02 0.148E–03 (t = 3 × 109 s) 0.591E–03 0.122E–01 0.449E–03 0.86 4 Algorithm for Simulating Fluid-Rock Interaction Problems 0.244E–02 (t = 6 × 1010 s) 0. 4.445E–02 0.170E–01 0.247E–01 0.487E–02 0.519E–02 0.974E–02 0.229E–02 0.218E–01 0.371E–02 0.886E–03 0.617E–02 0.739E–03 0.791E–02 0.742E–03 (t = 1.154E–01 0.3 Distribution of muscovite in the ﬂuid-rock interaction system .123E–01 0.109E–02 0.146E–01 0.309E–02 (t = 1.594E–02 0.5 × 1010 s) 0.928E–02 0.295E–03 0.195E–01 0.185E–01 0.297E–02 0.148E–02 0.

339E–01 0.4.352E–02 0.503E–03 (t = 1.169E–01 0.251E–02 0.3 Application Examples of the Term Splitting Algorithm 87 0.4 Distribution of pyrophyllite in the ﬂuid-rock interaction system .302E–02 0.101E–02 0.242E–01 0.379E–03 0.290E–01 0.320E–03 0.160E–03 0.151E–02 0.968E–02 0.201E–02 0.194E–01 0.563E–02 0.484E–02 (t = 1.213E–03 0.281E–02 (t = 6 × 1010 s) 0.402E–02 0.844E–02 0.5 × 1010 s) 0.267E–03 0.141E–01 0.107E–03 0.197E–01 0.113E–01 0.427E–03 0.387E–01 0.534E–04 (t = 3 × 109 s) 0.5 × 1011 s) Fig.225E–01 0. 4.145E–01 0.

We emphasise the importance of this issue here because it is often overlooked by some purely numerical modellers. Since the dissolution reaction constants of K-feldspar and muscovite are of the same order in magnitude. as the chemical dissolution system considered here is subcritical. This is to say. H + and quartz in the ﬂuid-rock interaction system at four different time instants. However. which is exactly in the same direction as the pore-ﬂuid ﬂow in the aquifer. Figures 4.3 and 4.7 show the conventional/generalized concentration distributions of K + . For an initial value problem in a homogeneous. This indicates that any numerical solution must be validated.6 and 4. the precipitated muscovite from the dissolution of K-feldspar can be dissolved and therefore. from both the physical and chemical points of view. This fact indicates that in order to model the chemical kinetics of the involved heterogeneous reactions correctly.5×1011 s respectively. The simplest way to evaluate a numerical solution in a qualitative manner is to check whether or not the solution violates the common knowledge related to the problem studied. “Numerical calculations can converge to a grid-dependent limit. 4.88 4 Algorithm for Simulating Fluid-Rock Interaction Problems from blue to red in Fig. rather than a boundary value problem. and artifacts of a solution can be numerical rather than geological”. whereas H + is continuously injected at the left entrance of the aquifer and consumed with time due to both the K-feldspar and muscovite dissolution reactions. if the numerical algorithm is not robust enough and the mesh/time step used is not appropriate in a numerical analysis. the dissolution process of muscovite is in parallel with that of K-feldspar. This indicates that when the injected carbon dioxide (CO2 ) gas enters the ﬂuid-rock interaction system. It is noted that from the mathematical point of view. pyrophyllite can be precipitated at an early stage. The produced H + propagates from the left side to the right side of the aquifer so that K-feldspar gradually dissolves and muscovite precipitates along the same direction as the dissolution front propagation of K-feldspar. 4. they agree very well with common knowledge. all the dissolution/precipitation propagation fronts propagate from the left side to the right side of the computational domain. Just as Phillips (1991) stated. Generally.2.4. isotropic porous medium (as we considered here). at least qualitatively if the analytical solution to the problem is not available. numerical error may result in an unstable/oscillatory chemical reaction/propagation front (Zienkiewicz 1977).4 are comprised of vertically parallel lines. Since all the propagation fronts of chemically reactive species in Figs. This demonstrates that the proposed numerical algorithms are robust enough for solving ﬂuid-rock interaction problems. t = 6×1010 s and t = 1. 4. t = 1. namely t = 3×109 s. 4. 4. the chemical reaction/propagation front of chemically reactive species is stable before the system reaches a steady state.5.5×1010 s.4) of pyrophyllite in the computational domain. Clearly. the problem solved here is an initial value problem. Since the generalized concentration of K-feldspar is greater than that of . it produces H + quasiinstantaneously at the left entrance of the system. all the reactive species transport equations should be solved simultaneously in the numerical analysis. at least from a qualitative point of view. K + is gradually produced and accumulated with time due to both the K-feldspar and muscovite dissolution reactions. as clearly exhibited in Fig. Quartz is produced by the K-feldspar dissolution reaction but is consumed by the muscovite dissolution reaction.

188E+00 (t = 1.3 Application Examples of the Term Splitting Algorithm 89 0.5 × 1011 s) Fig.192E+00 0.102E+00 0.107E+00 0.102E+00 0.189E+00 0.114E+00 0.190E+00 0.160E+00 0.101E+00 0.106E+00 (t = 1.158E+00 0.101E+00 0.102E+00 0. 4.157E+00 0.103E+00 0.5 Distribution of K + in the ﬂuid-rock interaction system .155E+00 0.113E+00 0.153E+00 (t = 6 × 1010 s) 0.5 × 1010 s) 0.101E+00 0.156E+00 0.191E+00 0.193E+00 0.115E+00 0.4.110E+00 0.193E+00 0.100E+00 (t = 3 × 109 s) 0.108E+00 0.189E+00 0.111E+00 0.154E+00 0.159E+00 0.

5 × 1010 s) 0.90 4 Algorithm for Simulating Fluid-Rock Interaction Problems 0.711E–03 (t = 3 × 109 s) 0.440E–02 0.284E–02 0.711E–03 (t = 1.569E–02 0.213E–02 0.358E–02 0.219E–02 0.142E–02 0.284E–02 0.284E–02 0.569E–02 0.219E–02 0.289E–02 0.389E–02 0.356E–02 0.498E–02 0.498E–02 0.427E–02 0.427E–02 0.6 Distribution of H + in the ﬂuid-rock interaction system .427E–02 0. 4.498E–02 0.590E–02 0.339E–02 0.142E–02 0.239E–02 (t = 1.490E–02 0.540E–02 0.142E–02 0.711E–03 (t = 6 × 1010 s) 0.356E–02 0.5 × 1011 s) Fig.569E–02 0.

405E+02 0.405E+02 0.406E+02 0.407E+02 0.405E+02 0.405E+02 0.405E+02 0.406E+02 (t = 1.406E+02 0.405E+02 0.405E+02 0.406E+02 0.406E+02 0.405E+02 0.5 × 1010 s) 0.405E+02 0.5 × 1011 s) Fig.405E+02 (t = 1.3 Application Examples of the Term Splitting Algorithm 91 0.405E+02 0.407E+02 0.406E+02 0.405E+02 0.406E+02 0.406E+02 0.405E+02 0.406E+02 0.7 Distribution of quartz in the ﬂuid-rock interaction system . 4.405E+02 (t = 3 × 109 s) 0.405E+02 0.406E+02 0.405E+02 (t = 6 × 1010 s) 0.405E+02 0.405E+02 0.405E+02 0.406E+02 0.4.

592E–03 (t = 1.299E–02 0.597E–03 0.729E–02 0.794E–02 (t = 1.915E–02 0.538E–03 0.629E–02 0.430E–02 0.175E–02 0.5 × 1011 s) Fig.330E–02 0.204E–02 0.479E–03 0.110E–01 0.882E–03 0.714E–03 0.529E–02 0.262E–02 0.362E–03 (t = 3 × 109 s) 0. 4.92 4 Algorithm for Simulating Fluid-Rock Interaction Problems 0.656E–03 0.146E–02 0.130E–02 (t = 6 × 1010 s) 0.5 × 1010 s) 0.122E–01 0.773E–03 0.421E–03 0.230E–02 0.118E–01 0.829E–02 0.855E–02 0.117E–02 0.976E–02 0.8 Distribution of porosity variation in the ﬂuid-rock interaction system .104E–01 0.

which is a K-feldspar dissolution problem in a pore-ﬂuid saturated. (φ − φ0 )/ φ0 . but there exist a precipitation propagation front and a dissolution propagation front for muscovite during the heterogeneous chemical reactions in the aquifer.3 Application Examples of the Term Splitting Algorithm 93 muscovite within the time frame considered. However. the accumulation of quartz exceeds the consumption of quartz and therefore. in the case of the conventional concentration of H + . Figure 4. isothermal and homogeneous aquifer. In the case of the generalized concentration of quartz. The propagation of the porosity variation front can be clearly observed from the numerical results shown in Fig. have demonstrated that: (1) There exist only a dissolution propagation front for K-feldspar and a precipitation propagation front for pyrophyllite. (4) The evolution of porosity mainly depends on the evolution of K-feldspar dissolution. its maximum values are 0. 4.4. the related numerical solutions from an application example.1032 kmol/m3 and 0.1942 kmol/m3 at t = 3 × 109 s and t = 1. . in the ﬂuid-rock interaction system at four different time instants.8 shows the porosity variation. In summary. which is identical to the direction of pore-ﬂuid ﬂow in the aquifer.8. (2) The dissolution of K-feldspar and muscovite may take place simultaneously in the aquifer so that pyrophyllite can be precipitated at the early stage of the heterogeneous chemical reactions. The evolution of porosity mainly depends on the evolution of the K-feldspar dissolution. (3) All the propagation fronts of chemically reactive species are comprised of vertically parallel lines. This indicates a slight increase in the generalized concentration of quartz with the increase of time within the scope of this study.5 × 1011 s respectively. the propagation directions of which are exactly the same as that of the pore-ﬂuid ﬂow in the aquifer. muscovite precipitation and dissolution and pyrophyllite precipitation in the ﬂuid-rock interaction system. the generalized concentration of quartz increases with time due to both the K-feldspar and muscovite dissolution reactions. In addition. The reason for this is due to a continuous injection of H + at the left entrance of the aquifer.4067 kmol/m3 at t = 3×109 s and t = 1. This indicates a signiﬁcant increase in the conventional concentration of K + with the increase of time within the scope of this study. In the case of the conventional concentration of K + .5 × 1011 s. its maximum value is a constant of 6. its maximum values are 0.5×1011 s respectively.4 × 10−3 kmol/m3 at both t = 3 × 109 s and t = 1. muscovite precipitation and dissolution and pyrophyllite precipitation in the ﬂuid-rock interaction system.4051 kmol/m3 and 0. the porosity variation front propagates from the left side to the right side of the aquifer. It is clear that the porosity of the porous medium evolves with time in the process of ﬂuid-rock interactions.

the concentration of the aqueous mineral is diluted so that the solid part of the solute (i.Chapter 5 A Segregated Algorithm for Simulating Chemical Dissolution Front Instabilities in Fluid-Saturated Porous Rocks When fresh pore-ﬂuid ﬂow enters a solute-saturated porous medium. 2006c). However. This means that the problem associated with the propagation of a dissolution front is a fully coupled nonlinear problem between porosity. Fundamentals of Computational Geoscience. 1998a. b. It is well known that because pore-ﬂuid ﬂow plays an important role in the process of reactive chemical-species transport. Lecture Notes in Earth Sciences 122. the feedback effect of the porosity change is weak so that the dissolution front is stable. 2004. 1988.e. For example. This chemical dissolution process can result in the propagation of a dissolution front within the ﬂuid-saturated porous medium. where the concentration of the solute (i. If the fresh pore-ﬂuid ﬂow is slow. 1986. in geo-environmental engineering. dissipative structure) of the dissolution front can emerge due to the self-organization of this coupled nonlinear system. the feedback effect of the porosity change becomes strong so that the dissolution front becomes unstable. Xu et al. there is a feedback effect of the porosity change on the pore-ﬂuid ﬂow. This leads to an important scientiﬁc problem. 1987). Since a change in porosity can cause a remarkable change in permeability. Yeh and Tripathi 1991. Zhao et al. Due to the dissolution of the solid mineral. 1999. Ormond and Ortoleva 2000. solid mineral) is dissolved to maintain the equilibrium state of the solution. In this case.1007/978-3-540-89743-9 5. a change in pore-ﬂuid ﬂow can cause a considerable change in the chemical-species concentration within the porous medium (Steefel and Lasaga 1990.. a new morphology (i. if the fresh pore-ﬂuid ﬂow is fast enough. the porosity of the porous medium is increased behind the dissolution front.e. which is closely associated with mineral dissolution in a ﬂuid-saturated porous medium. DOI 10. Chen and Liu 2002. Ortoleva et al. C Springer-Verlag Berlin Heidelberg 2009 95 . This kind of chemical-dissolution-front instability problem exists ubiquitously in many scientiﬁc and engineering ﬁelds. the rehabilitation of contaminated sites using fresh water to wash the sites involves the propagation problem of the removed contaminant material front in water-saturated porous medium. known as the reactive inﬁltration instability problem (Chadam et al. 2005a. Shafter et al. Zhao et al. the extraction of minerals in the deep Earth using the in-situ leaching technique may result in C.e. pore-ﬂuid pressure and reactive chemical-species transport within the ﬂuid-saturated porous medium. aqueous mineral) reaches its equilibrium concentration. Raffensperger and Garven 1995. 1994. according to Darcy’s law. In mineral mining engineering.

1997a. More importantly. Zhao et al. the secondary recovery of oil by acidifying the oil ﬁeld to uniformly increase porosity and hence the yield of oil is associated with the propagation of the acid-dissolved material front in porous rocks. Zhao et al. 2000b. Lewis and Schreﬂer 1998) and the ﬁnite difference method. to predict analytically the complicated morphological evolution of a chemical dissolution front in the case of the chemical dissolution system becoming supercritical. 2002. Darcy’s law can be used to describe pore-ﬂuid ﬂow and Fick’s law can be used to describe mass transport phenomena respectively. 2002. d. This makes it possible to compare the numerical solution obtained from the benchmark problem with the derived analytical solution so that the proposed numerical procedure can be veriﬁed for simulating chemical-dissolution-front propagation problem in the ﬂuid-saturated porous medium. As an alternative. For this reason. it becomes a potentially powerful mechanism to control ore body formation and mineralization in the deep Earth. 2003a). they must be veriﬁed before they are used to solve any new type of scientiﬁc and engineering problem. If both the porosity change of the porous medium is caused by . In the petroleum industry. numerical methods are suitable to overcome this difﬁculty. Since the chemical dissolution front can create porosity and therefore can locally enhance the pore-ﬂuid ﬂow.1. According to modern mineralization theory. it is necessary to derive the analytical solution for the propagation of a planar dissolution front within a benchmark problem. ore body formation and mineralization is mainly controlled by pore-ﬂuid ﬂow focusing and the equilibrium concentration gradient of the concerned minerals (Phillips 1991. it is imperative to develop advanced techniques to explore for new ore deposits. the geometry of which can be accurately simulated using numerical methods such as the ﬁnite element method (Zienkiewicz 1977. Although analytical solutions can be obtained for some reactive transport problems with simple geometry.1 Mathematical Background of Chemical Dissolution Front Instability Problems in Fluid-Saturated Porous Rocks 5. 2002c. Schaubs and Zhao 2002. Towards this goal. Since numerical methods are approximate solution methods. 5. 2001b. Zhao et al. there is a deﬁnite need to understand the important physical and chemical processes that control ore body formation and mineralization in the deep Earth (Raffensperger and Garven 1995. 1998a). if not impossible. d. 1998a. due to the ever-increasing demand for mineral resources and the likelihood of the exhaust of the existing ore deposits. Ord et al. 1999b. Gow et al.96 5 Simulating Chemical Dissolution Front Instabilities in Fluid-Saturated Porous Rocks the propagation problem of the dissolved mineral front in ﬂuid-saturated porous medium.1 A General Case of Reactive Multi-Chemical-Species Transport with Consideration of Porosity/Permeability Feedback For a pore-ﬂuid-saturated porous medium. it is very difﬁcult.

pore-ﬂuid ﬂow and reactive multi-chemical-species transport in the pore-ﬂuid-saturated porous medium can be expressed as follows: ∂ (ρ f φ) + ∇ • (ρ f φ u linear ) = 0. it is strongly dependent on porosity φ. . φ is the porosity of the porous medium. In this regard. Actual measurements on rocks.2) (i = 1. Ri is the source/sink term of chemical species i due to the dissolution/precipitation of solid minerals within the system.2) and (5. Chadam et al. Nield and Bejan . 2. To consider the permeability change caused by a change in porosity. .1 Mathematical Background of Chemical Dissolution Front Instability Problems 97 chemical dissolution of soluble solid minerals within the porous medium and the feedback effect of such a change on the variation of permeability and diffusivity are taken into account. an equation is needed to express the relationship between permeability and porosity. μ (5. the ﬂuid density and averaged linear velocity of the pore-ﬂuid are deﬁned in the pore space. however.1). Nield and Bejan 1992. According to the Carman-Kozeny law (Scheidegger 1974) which is based on the conceptual model of packing of spheres. 1986). Since the diffusivity of each chemical species is considered as a function of porosity.3) ∂ (φCi ) + ∇ • (φCi u linear ) = ∇ • [φ Di (φ)∇Ci ] + Ri ∂t where u linear is the averaged linear velocity vector within the pore space of the porous medium. It is noted that in Eqs. often yield power law relations with exponents for φ signiﬁcantly larger than 3. a power law relation of k ∝ φ 3 (1 − φ)2 exists. 1994c). (5. while the source/sink term and the Darcy velocity of the pore ﬂuid are deﬁned in the whole medium space (Phillips 1991. Di (φ) is the diffusivity of chemical species i. the chemical species concentration. Zhao et al. In particular. ∂t u = φ u linear = − k(φ) ∇ p.5. Di (φ) = D0i φ q 3 5 ≤q≤ . ρ f is the density of the pore-ﬂuid. (5. the governing equations of the coupled nonlinear problem between porosity. Detournay and Cheng (1993) state that “The intrinsic permeability k is generally a function of the pore geometry. .1) (5. a common phenomenological relation can be used for describing this function (Bear 1972. μ is the dynamic viscosity of the pore-ﬂuid.” Also.3). u is the Darcy velocity vector within the porous medium. k(φ) is the permeability of the porous medium. 2 2 (5. N is the total number of all the chemical species to be considered in the system. Other models based on different pore geometry give similar power laws. (5. N ).4) where D0i is the diffusivity of chemical species i in pure water. p and Ci are pressure and the concentration (moles/pore-ﬂuid volume) of chemical species i.

so that the rate of porosity change can be expressed as: ∂φ = −D p R p . The source/sink term of chemical species i due to the dissolution/precipitation of solid minerals within the system can be determined in the following manner (Chadam et al. then the ﬁnal (i. (5. for a given porosity φ.8) where χi is the stoichiometric coefﬁcient of the ith chemical species.5) where φ0 and k0 are the initial reference porosity and permeability of the porous medium respectively. which is deﬁned as the number of the soluble gains per unit medium volume. 1986). it is assumed that the average volume of soluble grains is V p and that the density of the soluble grains is D p .98 5 Simulating Chemical Dissolution Front Instabilities in Fluid-Saturated Porous Rocks (1992) state that “The Carman-Kozeny law is widely used since it seems to be the best simple expression available. If the volume fraction of insoluble gains is denoted by φinsoluble .” For these reasons.6) At the particle level. it is assumed that the solid grains are dissolved according to the following formula: N Solid ⇒ i=1 χi X i . ∂t (5. 1986): N R p = kchemical A p i=1 Ci i − K eq χ (5.9) where A p is the averaged surface area of soluble grains. the rate of grain-volume change due to a chemical reaction is denoted by R p . kchemical and K eq are the rate constant and equilibrium constant of the chemical reaction respectively. the average volume of soluble grains can be expressed as follows: Vp = φf − φ . It is commonly assumed that the rate of grain-volume change due to a chemical reaction can be expressed as follows (Chadam et al. maximum) porosity of the porous medium can be denoted by φ f = 1 − φinsoluble . X i represents chemical species i in the pore-ﬂuid.7) Without loss of generality. At the particle level. Dp (5. the Carman-Kozeny law will be used to calculate permeability k. k(φ) = k0 (1 − φ0 )2 φ 3 .e. . 3 φ0 (1 − φ)2 (5. In this case.

μ (5. However. in which a .12) can be derived by substituting the linear average velocity into Eqs. χ 5.9).e. moles per volume) of the soluble grains is denoted by ρs .2) (5. then the source/sink term of chemical species i due to the dissolution/precipitation of solid minerals within the system can be expressed as follows: Ri = −χi ρs kchemical D p A p A N i=1 Ci i − K eq (5.1). (5. The ﬁrst special case to be considered is a problem. (5.1) and (5.12). ∂t Vp ψ(φ) = k(φ) .5. (φC) − ∇ • [φ D(φ)∇C + Cψ(φ)∇ p] + ρs kchemical ∂t Vp (5. and (5. the governing equations of a reactive singlechemical-species transport problem in a ﬂuid-saturated porous medium can be written as follows: ∂φ − ∇ • [ψ(φ)∇ p] = 0.2 A Particular Case of Reactive Single-Chemical-Species Transport with Consideration of Porosity/Permeability Feedack If the pore-ﬂuid is incompressible. in some special cases. it is possible to obtain analytical solutions for some variables involved in this single-chemicalspecies system. (5. even if not impossible. (5.12) Ap ∂φ + kchemical (φ f − φ)(C − Ceq ) = 0.3). to obtain a complete set of analytical solutions for the pore-ﬂuid pressure. (5. (5. it is very difﬁcult. chemical species concentration and porosity within the ﬂuid-saturated porous medium.14) are of the same meanings as those deﬁned in Eqs. Other quantities in Eqs. (5.11). Note that Eqs.3) with consideration of a single-chemical species.10) N i=1 χ = −χi ρs kchemical V p (φ f − φ) p Ci i − K eq .11) Ap ∂ (φ f − φ)(C − Ceq ) = 0. ∂t (5. It needs to be pointed out that for this single-chemical-species system.1 Mathematical Background of Chemical Dissolution Front Instability Problems 99 If the molar density (i.14) where C and Ceq are the concentration and equilibrium concentration of the single chemical species.13) and (5.13) (5.11) and (5.1.

in which the solid molar density greatly exceeds the equilibrium concentration of the chemical species. ∂t ∂x ∂x ∂x Vp (5. it is possible to derive a complete set of analytical solutions for the pore-ﬂuid pressure. p0x is the unknown pore-ﬂuid pressure gradient as x approaching positive inﬁnite in the downstream of the pore-ﬂuid ﬂow. (5. then the boundary conditions of this special problem are expressed as lim C = 0. 1986). The second special case to be considered is an asymptotic problem. lim φ = φ f . ∂p = p0x x→∞ ∂ x lim where φ0 is the initial porosity of the porous medium.11).12) and (5. implying that the region of a considerable porosity change propagates very slowly within the ﬂuid-saturated porous medium. In addition.12) and (5.17) If the chemical species is initially in an equilibrium state and fresh pore-ﬂuid is injected at the location of x approaching negative inﬁnite.1.15) Ap ∂ ∂C ∂p ∂ (φC) − φ D(φ) + Cψ(φ) + ρs kchemical (φ f − φ)(C − Ceq ) = 0. For this particular case. the planar dissolution front is assumed to propagate in the positive x direction. In this particular case.11). x→∞ lim φ = φ0 . analytical solutions can be obtained for both the propagation speed of the dissolution front and the downstream pressure gradient of the pore-ﬂuid. (5.19) x→−∞ x→−∞ x→−∞ x→∞ lim C = Ceq .1 The First Special Case In this special case. Since pfx drives the pore-ﬂuid ﬂow continuously .13) degenerates into a onedimensional problem. ∂t Vp (5. 5. For this reason.16) Ap ∂φ + kchemical (φ f − φ)(C − Ceq ) = 0. the problem described in Eqs.100 5 Simulating Chemical Dissolution Front Instabilities in Fluid-Saturated Porous Rocks planar dissolution front propagates in the full space. it is also possible to investigate the reactive inﬁltration instability associated with the dissolution front propagation in this particular case (Chadam et al. (5. (5.13) can be rewritten as follows: ∂ ∂p ∂φ − ψ(φ) = 0. lim ∂p = pfx ∂x (Upstream boundary).2. ∂t ∂x ∂x (5. Since the dissolution front is a plane.18) (Downstream boundary). (5. pfx is the pore-ﬂuid pressure gradient as x approaching negative inﬁnite in the upstream of the pore-ﬂuid ﬂow. so that all quantities are independent of the transverse coordinates y and z. chemical species concentration and porosity within the ﬂuidsaturated porous medium. (5. Eqs.

e. ∂ ∂t = ∂ ∂t + x ξ ∂ ∂x = ∂ x ∂t = t ∂ ∂t .18) and (5.23) and (5. (5.24) into Eqs.22). ∂ξ Vp (5.5. the following equations can be derived from Eqs. it has a negative algebraic value (i. ∂ξ ∂ξ ∂C ∂p ∂ φ D(φ) + Cψ(φ) + vfront (C − ρs )φ = 0.23) ∂ ∂ξ = t ∂ ∂x (5.25) (5.e.26) from negative inﬁnite to positive inﬁnite and using the boundary conditions (i. pfx < 0) in this analysis.17) yields the following equations: ∂ ∂p ψ(φ) + vfront φ = 0. (5. ∂x (5. then it is possible to transform a moving boundary problem of the dissolution front (in an x-t coordinate system) into a steady-state boundary problem of the dissolution front (in an ξ − t coordinate system) using the following coordinate mapping: ξ = x − vfront t. If the propagation speed of the planar dissolution front is denoted by vfront .25) and (5.21) ∂ ∂ξ ∂ ∂x (5. t + vfront x ∂ . (5.20) It is necessary to relate partial derivatives with respect to ξ and t to those with respect to x and t (Turcotte and Schubert 1982).16) and (5.27) Integrating Eqs. (5. Eqs.1 Mathematical Background of Chemical Dissolution Front Instability Problems 101 along the positive x direction. (5. t (5.15). ∂ξ ∂ξ ∂ξ vfront Ap ∂φ − kchemical (φ f − φ)(C − Ceq ) = 0. ∂ξ . (5. Since the transformed system in the ξ − t coordinate system is in a steady state.24) Substituting Eqs.19)) yields the following equations: .21) and (5.26) (5.22) where derivatives are taken with x or t held constant as appropriate. ∂ ∂t = −vfront x ∂ . (5.

a small parameter can be deﬁned as follows: ε= Ceq << 1. If the ﬁnite element method is used to solve this special problem. φ0 Ceq + (φ f − φ0 )ρs φ0 Ceq + (φ f − φ0 )ρs ψ(φ f )[φ0 Ceq + (φ f − φ0 )ρs ] p .31) where u 0x is the Darcy velocity in the far downstream of the ﬂow as x approaches positive inﬁnite.1. x= x .32) where u fx is the Darcy velocity in the far upstream of the ﬂow as x approaches negative inﬁnite.29) simultaneously results in the following analytical solutions: vfront = −ψ(φ0 ) p0x Ceq u 0x Ceq = . u 0x can be expressed as u 0x = φ0 Ceq + (φ f − φ0 )ρs u fx . Using Darcy’s law.34) C= p= u= (5. (5. u∗ (5.28) and (5. L∗ C .2 The Second Special Case (Base Solutions for a Stable State) Since the solid molar density greatly exceeds the equilibrium concentration of the chemical species. vfront ) and the Darcy velocity in the far downstream of the ﬂow as x approaches positive inﬁnite (i.35) . u 0x ). the following dimensionless parameters and variables can be deﬁned.e. ψ(φ0 )[φ0 Ceq + (φ f − φ0 )(ρs + Ceq )] fx (5. the accuracy of the ﬁnite element simulation can be conveniently evaluated by comparing the numerical solutions with the analytical ones for both the propagation speed of the planar dissolution front (i.2. p∗ z= z . (5.102 5 Simulating Chemical Dissolution Front Instabilities in Fluid-Saturated Porous Rocks Ceq ψ(φ0 ) pox + vfront φ0 (Ceq − ρs ) + vfront φ f ρs = 0. L∗ p . ρs (5.28) (5.30) p0x = (5. 5. φ0 Ceq + (φ f − φ0 )(ρs + Ceq ) (5.29) Solving Eqs.e.33) To facilitate the theoretical analysis in the limit case of ε approaching zero. ψ(φ0 ) pox + vfront φ0 − ψ(φ f ) pfx − vfront φ f = 0. L∗ u . Ceq y= y .

relative to the propagation front.39) into Eqs. (5. Thus.36) where τ is a slow dimensionless time to describe the slowness of the chemical dissolution that takes place in the system.37). Other characteristic parameters used in Eqs. lim φ = φ0 . (5.40) ε ∂ ∂φ (φC) − ∇ • [D ∗ (φ)∇C + Cψ ∗ (φ)∇ p ] − = 0. (5.12) and (5.36) can be expressed as follows: t∗ = Vp .39) Substituting Eqs.38) and (5.5. t∗ (5. the porosity undergoes a jump from its initial value into its ﬁnal value.35).38) D ∗ (φ) = ψ ∗ (φ) = (5. ∂τ ∂τ ε ∂φ + (φ f − φ)(C − 1) = 0.44) x→∞ x→∞ x→−∞ lim C = 0. It is noted that the propagation front due to chemical dissolution divides the problem domain into two regions.41) (5. φ f D(φ f ) . this dissolution-front propagation problem can be considered as a Stefan moving boundary problem (Chadam et al.36). ψ(φ f ) (5. (5. the boundary conditions for this special case can be expressed in a dimensionless form as follows: lim C = 1. (5.34). (5. (5.35) and (5. an upstream region and a downstream region.37) p∗ = u∗ = (5. ∂p = p 0x x→∞ ∂ x lim ∂p = p fx x→−∞ ∂ x lim (downstream boundary).43) (upstream boundary). ∂τ (5. L∗ ψ(φ) .1 Mathematical Background of Chemical Dissolution Front Instability Problems 103 τ= t ε. kchemical A p Ceq φ f D(φ f ) . ψ(φ f ) φ D(φ) . In the limit case of ε approaching zero. x→−∞ lim φ = φ f . (5.11). ∂τ (5. (5. Across this propagation front.42) Similarly.34). 1986). (5. the corresponding governing .13) yields the following dimensionless equations: ε ∂φ − ∇ • [ψ ∗ (φ)∇ p ] = 0. (5. φ f D(φ f ) L∗ = φ f D(φ f )t ∗ .

(5. (5. p(ξ ) = p fx ξ + pC2 . ψ(φ0 ) fx ν front = − p fx φ f − φ0 . (in the upstream region). φ = φ0 (in the downstream region).45) ∇ • (∇C + C∇ p) = 0.43). The resulting base solutions are expressed as follows: C(ξ ) = 1.45) and (5. (5. τ ) = 0. (5. (5.48) where n is the unit normal vector of the moving dissolution front. ∇ 2 p = 0. This leads to the following interface conditions for this moving-front problem: S→0− lim C = lim+ C. p 0x = ψ(φ f ) p . chemical species concentration and mass ﬂuxes of both the chemical species and the pore-ﬂuid should be continuous on S(x. (in the downstream region). When the planar dissolution front is under stable conditions. the location of the chemical dissolution front. while pC2 can be determined using the pressure continuity condition at the interface between the upstream and downstream regions. S→0 (5. Eqs. ∂n S→0 lim− ψ(φ0 ) ∂p ∂p = lim+ . if the ﬁnite element method is used to solve the second special problem. pC1 can be determined by setting the dimensionless pressure p(ξ ) to be a constant at a prescribed location of the downstream region.49) φ = φf (in the upstream region). φ = φf (5. the base solutions for this special problem can be derived from Eqs. τ ) = 0.104 5 Simulating Chemical Dissolution Front Instabilities in Fluid-Saturated Porous Rocks equations for the dimensionless variables of the problem in both the downstream region and the upstream region can be expressed below: C = 1. (5.44). (5. then the dimensionless pressure.51) Therefore. .47) and (5. (5.47) S→0 lim− ∂C = ν front (φ f − φ0 ). the accuracy of the ﬁnite element simulation can be conveniently evaluated by comparing the numerical solutions with a complete set of analytical solutions including porosity. Other parameters are deﬁned below: ξ = x − ν front τ.46) If the chemical dissolution front is denoted by S(x. ∂n ψ(φ f ) S→0 ∂n (5.46) with the related boundary and interface conditions (i. the dimensionless chemical-species concentration and the dimensionless pressure. φ = φ0 ∇ 2 p = 0.50) C(ξ ) = exp(− p fx ξ ). For example.e.48)). S→0 S→0− lim p = lim+ p. where pC1 and pC2 are two constants to be determined. p(ξ ) = p 0x ξ + pC1 .

τ ) + δ p (ξ ) exp(ωτ ) cos(m y). ˆ p total (ξ. y. even though both small perturbations of the dissolution front and the feedback effect of porosity/permeability change are simultaneously considered in the analysis.55) ξ ∂S ∂ + ∂y ∂S ∂2 ∂ξ 2 ∂S ∂y ∂ ∂ξ + S ∂ ∂y . ˆ C total (ξ. S (5. (5. δ is the amplitude of the perturbation and δ << 1 by the deﬁnition of a linear stability analysis. Since S(ξ. y.2. S(ξ. If a small time-dependent perturbation is added to the planar dissolution front. then the total solution of the system is equal to the summation of the base solution and the perturbed solution of the system. 1986. 1988. 1987). Ortoleva et al. However. the following derivatives exist mathematically: ∂ ∂ξ ∂ ∂y = = ∂S ∂ = ∂ξ ∂ S ∂ ∂y = S ξ ∂ ∂ξ .53) (5. τ ) = p(ξ. S (5.57) ∂2 ∂ y2 ξ = ∂2S ∂ + ∂ y 2 ∂ξ ∂S ∂y 2 ∂2 ∂ S ∂2 + +2 ∂ξ 2 ∂ y ∂ξ ∂ y ∂2 ∂ y2 S . the planar dissolution front can change from a planar shape into a complicated one. the planar dissolution front remains planar. (5. τ ) + δ C(ξ ) exp(ωτ ) cos(m y). when the reactive transport system is unstable. The main purpose of conducting such a linear stability analysis is to determine the critical condition under which the chemical dissolution front of the reactive transport system becomes unstable. τ ) = C(ξ.58) .56) ξ = ∂2 ∂ξ 2 . y. S (5. y. m is the wavenumber of the perturbation. τ ) = ξ − δ exp(ωτ ) cos(m y). τ ) is a function of coordinates ξ and y.3 The Second Special Case (Perturbation Solutions for an Unstable State) When a reactive transport system represented by the above-mentioned second special problem is stable.52) (5.1 Mathematical Background of Chemical Dissolution Front Instability Problems 105 5. The instability of the above-mentioned second special problem can be determined using a linear stability analysis (Chadam et al.54) where ω is the growth rate of the perturbation.5.1.

Eqs. ˆ p (ξ ) = p 0x 1 − 1−β exp(− |m| ξ ) 1+β (in the downstream region).e. (5. (5. 1+β 1+β ˆ p (ξ ) = p fx 1 + 1−β exp(|m| ξ ) 1+β (in the upstream region). (5. the interface conditions for this ﬁrst-order perturbation problem can be expressed as follows: ˆ C = 0.66) .62)) yields the following analytical results: ˆ C = 0.65) ˆ C(ξ ) = − p fx exp(− p fx ξ ) − 1−β 2 exp(σ ξ ) + exp[(|m| − p fx )ξ ] . (5. (5. (5.59) ˆ ˆ ˆ ∂ 2C ∂C ∂p ˆ + p fx − m 2 C − m 2 p fx exp(− p fx ξ ) − p fx exp(− p fx ξ ) = 0. (5. (5. ∂n S→0− ˆ ˆ lim p = lim+ p . ˆ ∂2 p ˆ − m 2 p + m 2 p 0x = 0 ∂ξ 2 (in the downstream region).58). the ﬁrst-order perturbation equations of this system can be expressed as ˆ C = 0.59) and (5.45) and (5.63) S→0 lim− S→0− lim ˆ ˆ ψ(φ0 ) ∂p ∂p = lim . (5.53) and (5. x→−∞ lim (upstream boundary).61) x→∞ x→−∞ ˆ lim C = 0.61) and (5. With consideration of Eq.64) Solving Eqs.46).62) Similarly. ∂n ψ(φ f ) S→0+ ∂n (5.106 5 Simulating Chemical Dissolution Front Instabilities in Fluid-Saturated Porous Rocks It is noted that the total solutions expressed in Eqs. ˆ ∂C = ω(φ f − φ0 ). (5.54) must satisfy the governing equations that are expressed in Eqs. S→0 (5. (5. 2 ∂ξ ∂ξ ∂ξ ˆ ∂2 p ˆ − m 2 p + m 2 p fx = 0 ∂ξ 2 (in the upstream region).60) The corresponding boundary conditions of the ﬁrst-order perturbation problem are: ˆ C = 0.60) with the boundary and interface conditions (i. lim ˆ ∂p =0 ∂ξ ˆ ∂p =0 ∂ξ (downstream boundary).

which is described by the above-mentioned second special problem.e.5.67) σ = . (5.e.64) yields the following equation for the growth rate of the small perturbation: ω(m) = − p fx (1 + β)(φ f − φ0 ) [− p fx − ( p fx )2 + 4m 2 + (1 − β) |m|).1 Mathematical Background of Chemical Dissolution Front Instability Problems 107 where β= ψ(φ0 ) k(φ0 ) = .72). (5. while if Z h < Z h critical . p fx =− (3 − β)(1 + β) . ψ(φ f ) k(φ f ) ( p fx )2 + 4m 2 − p fx 2 (5. If Z h > Z h critical . the Zhao number of the reactive transport system can be deﬁned as follows: Z h = − p fx = − pfx L ∗ p∗ =− k(φ f )L ∗ pfx φ f μD(φ f ) = vﬂow φ f D(φ f ) Vp . 2(1 − β) (5. (5. Since p fx critical is usually of a negative value. the following critical Zhao number is deﬁned to judge the instability of the reactive transport system: Z h critical = − p fx = (3 − β)(1 + β) . small wavenumber m) perturbations.68) Substituting Eq.70) critical where p fx critical is the critical value of the generalized dimensionless pressure gradient in the far upstream direction as x approaching negative inﬁnite (Zhao et al. 2008e).66) into Eq. large wavenumber m) perturbations but it is unstable to long wavelength (i.69) Equation (5.71) critical Thus. then the chemical dissolution front of the reactive transport system becomes unstable. Letting ω(m) = 0 yields the following critical condition. (5.72) kchemical A p Ceq Using Eqs. 2(1 − β) (5. then the chemical dissolution front of the reactive transport system is stable.69) clearly indicates that the planar dissolution front of the reactive transport system.71) and (5. is stable to short wavelength (i. The case of Z h = Z h critical represents a situation where the chemical dissolution front of the reactive transport . (5. under which the reactive transport system can become unstable. (5. a criterion can be established to judge the instability of a chemical dissolution front associated with the particular chemical system in this investigation.

thermodynamic and chemical kinetic characteristics of a ﬂuid-rock system in a comprehensive manner. for the reactive chemical-species transport considered in this investigation. then an increase in the shape factor of soluble grains can destabilize the instability of the chemical dissolution front. while the dissolution-enhanced diffusivity stabilizes the instability of the chemical dissolution front.72) indicates that for the reactive chemical-species transport considered in this investigation.77) clearly indicate that the Zhao number is a dimensionless number that can be used to represent the geometrical.76) FShape = (5.75). Fchemical is a term to represent the chemical kinetics of the dissolution reaction.108 5 Simulating Chemical Dissolution Front Instabilities in Fluid-Saturated Porous Rocks system is neutrally unstable. FDiffusion is a term to represent the solute diffusion/dispersion.73) where FAdvection is a term to represent the solute advection. Clearly. These terms can be expressed as follows: FAdvection = vﬂow . 1 . is higher than that of a porous medium comprised of regular spherical grains. This dimensionless number reveals the intimate interaction between solute advection. φ f D(φ f ) 1 kchemical Ceq Vp . Eq.77) Equations (5.74) FDiffusion = (5. (5. . If the shape factor of soluble grains is represented by θ = V p A p . (5. Ap . (5. hydrodynamic. chemical kinetics and mineral geometry in a reactive transport system. (5.75) and (5. an increase in either the equilibrium concentration of the chemical species or the chemical reaction constant of the dissolution reaction can cause the stabilization of the chemical dissolution front. Similarly. (5. implying that the introduced small perturbation can be maintained but it does not grow in the corresponding reactive transport system. the dissolution-enhanced permeability destabilizes the instability of the chemical dissolution front. solution diffusion/dispersion.73).74).75) Fchemical = (5. To understand the physical meanings of each term in the Zhao number. FShape is a term to represent the shape factor of the soluble mineral in the ﬂuid-rock interaction system. (5. indicating that the instability likelihood of a porous medium comprised of irregular grains.72) can be rewritten in the following form: Z h = FAdvection FDiffusion Fchemical FShape . Equation (5.

As a result. Since numerical methods are approximate solution methods. (5. which is a combination of both the ﬁnite element method and the ﬁnite difference method.41) and (5. Δτ is the dimensionless time increment at the current time-step.2 Proposed Segregated Algorithm for Simulating the Morphological Evolution of a Chemical Dissolution Front Although analytical solutions can be obtained for the above-mentioned special cases. (5.78) where C τ +Δτ is the dimensionless concentration at the current time-step.2. (5. the segregated algorithm.41) and (5. in which Eqs.42) are solved separately in a sequential manner. τ + Δτ . there exist the following relationships in the ﬁnite difference sense: . As an alternative. the numerical solution obtained from the benchmark problem can be compared with the corresponding analytical solution. 5.1 Formulation of the Segregated Algorithm for Simulating the Evolution of Chemical Dissolution Fronts In this section. Eqs. Using the backward difference scheme. The ﬁnite element method is used to discretize the geometrical shape of the problem domain. Δτ (5. (5. numerical methods are suitable to overcome this difﬁculty. they must be veriﬁed before they are used to solve any new type of scientiﬁc and engineering problem. a benchmark problem is constructed from the theoretical analysis in the previous section. Mathematically. To verify the accuracy of the numerical solution. it will be used to simulate the complicated morphological evolution process of a planar dissolution front in the case of the chemical dissolution system becoming supercritical. For a given dimensionless time-step. For this reason. After the proposed numerical procedure is veriﬁed. Eq.40). (5. to predict analytically the complicated morphological evolution process of a planar dissolution front in the case of the chemical dissolution system becoming supercritical. is used to derive the formulation of the proposed numerical procedure.2 Segregated Algorithm for Simulating the Morphological Evolution 109 5. the porosity can be denoted by φτ +Δτ = φτ + Δφτ +Δτ .42) can be written as follows: ε + (1 − C τ +Δτ ) Δφτ +Δτ = (φ f − φτ )(1 − C τ +Δτ ). the main purpose of this section is to propose a numerical procedure for simulating how a planar dissolution front evolves into a complicated morphological front. it is very difﬁcult.40). if not impossible.42) are solved using the proposed numerical procedure. while the ﬁnite difference method is used to discretize the dimensionless time. where φτ is the porosity at the previous time-step and Δφτ +Δτ is the porosity increment at the current time-step. Since the system described by these equations is highly nonlinear.5.

(5. (5. the dimensionless concentration at the current time-step is not known. (5.83) .78).81) and (5. (5.40) can be rewritten in the following discretized form: ∇ • ψ ∗ (φ)∇ p = ∇ • ψ ∗ (φτ +Δτ )∇ p τ +Δτ = (1 − C τ +Δτ )(φ f − φτ +Δτ ).83) is solved using the ﬁnite element method. so that the following convergence criterion can be established after the second iteration step.78) is solved using the ﬁnite element method. Δτ Δτ Δτ (5.41) yields the following ﬁnite difference equation: ε 1 φτ +Δτ + (φ f − φτ +Δτ ) C τ +Δτ − ∇ • D ∗ (φτ +Δτ )∇C τ +Δτ Δτ ε −∇ p τ +Δτ • ψ ∗ (φτ +Δτ )∇C τ +Δτ ε 1 φτ +Δτ C τ + φ f − φτ +Δτ = Δτ ε Similarly.81) (5.84) is solved for the dimensionless pressure.82) into Eq. Similarly. (5. (5. (5. ∇ • Cψ ∗ (φ)∇ p = C∇ • ψ ∗ (φ)∇ p + ∇ p • ψ ∗ (φ)∇C = C τ +Δτ ∇ • ψ ∗ (φτ +Δτ )∇ p τ +Δτ + ∇ p τ +Δτ • ψ (φτ +Δτ )∇C τ +Δτ ∗ (5.78) is solved for the porosity.79).110 5 Simulating Chemical Dissolution Front Instabilities in Fluid-Saturated Porous Rocks ε ∂ φC ∂τ =ε Δ C τ +Δτ Δ φτ +Δτ C τ +Δτ Δφτ +Δτ = εC τ +Δτ + εφτ +Δτ .80). dimensionless concentration and dimensionless pressure at the current time-step. Note that when Eq.83) is solved for the dimensionless concentration.80) ε Δ (φτ +Δτ ) ∂φ =ε = 1 − C τ +Δτ ∂τ Δτ ∇ • D ∗ (φ)∇C = ∇ • D ∗ (φτ +Δτ )∇C τ +Δτ . This indicates that these three equations are fully coupled so that an iteration scheme needs to be used to solve them sequentially. the same procedure as used in the ﬁrst iteration step is followed. (5.84) Using the proposed segregated scheme and ﬁnite element method. Eqs.83) and (5. when Eq. The resulting approximate porosity and dimensionless concentration can be used when Eq.82) Substituting Eqs. Eq. the dimensionless concentration at the previous time-step is used as a reasonable guess of the dimensionless concentration at the current time-step when Eq.79) φ f − φτ +Δτ . (5. the dimensionless pressure at the previous time-step is used as a reasonable guess for the current time-step when Eq. (5. (5. In a similar way. At the ﬁrst iteration step. (5.84) are solved separately and sequentially for the porosity. (5. the dimensionless pressure at the current time-step remains unknown. (5. At the second iteration step. (5.

(5.1.5. which is the discretized description of a continuum mathematical model.2 Segregated Algorithm for Simulating the Morphological Evolution 111 ⎛ E = Max ⎝ Nφ k φi.τ +Δτ and pi. If the convergence criterion is not met. Except for the left boundary. ⎞ ⎟ ⎠ < E.τ +Δτ − C i. then the iteration is repeated at the current time-step. the initial porosity of the porous medium is 0. Figure 5. Nφ . Due to inevitable round-off errors in computation and discretized errors in temporal and spatial variables.2 Veriﬁcation of the Segregated Algorithm for Simulating the Evolution of Chemical Dissolution Fronts The main and ultimate purpose of a numerical simulation is to provide numerical solutions for practical problems in a real world.τ +Δτ are the porosity. For this benchmark problem. It is noted that k ≥ 2 in Eq. The dimensionless height and width of the computational model are 5 and 10 respectively.τ +Δτ . The convergence criterion is checked after the second iteration step.e. Otherwise. k is the index number at the current iteration step and k − 1 is the index number at the k k k previous iteration step. the convergence solution is obtained at the current time step and the solution procedure goes to the next time-step until the ﬁnal time-step is reached. These practical problems are impossible and impractical to solve analytically. NC C i.τ +Δτ and pi.85) 2 where E and E are the maximum error at the k-th iteration step and the allowable error limit.τ +Δτ .τ +Δτ i=1 − 2 k−1 φi. the Zhao number of the reactive transport system is unity. dimensionless concentration and dimensionless pressure of node i at both the current time-step and k−1 k−1 k−1 the current iteration step. implying that there is a horizontal throughﬂow from the left to the right of the computational model. φi. (5. For this reason.τ +Δτ . while the . dimensionless concentration and dimensionless pressure respectively. 5. Since numerical methods are the basic foundation of a numerical simulation. only an approximate solution can be obtained from a computational model. a benchmark problem. pore-ﬂuid pressure and reactive chemical-species transport within a ﬂuid-saturated porous medium.τ +Δτ are the porosity. NC and N p are the total numbers of the degrees-of-freedom for the porosity. φi.τ +Δτ i=1 k k−1 2 .2. dimensionless concentration and dimensionless pressure of node i at the current time-step but at the previous iteration step. C i. C i.85).τ +Δτ i=1 Np k−1 k pi. is considered in this section. the dimensionlesspressure gradient (i.τ +Δτ − pi. p fx = −1) is applied on the left boundary.1 shows the geometry and boundary conditions of the coupled problem between porosity. for which the analytical solutions are available. In this case. it is necessary to verify the proposed numerical procedure so that meaningful numerical results can be obtained from a discretized computational model.

Since the computational domain of the benchmark problem is of ﬁnite size. while the thin line shows the corresponding analytical solutions. This ﬁnal porosity is applied on the left boundary as a boundary condition of the computational domain. 5. the critical Zhao number of the system is approximately equal to 1. In these ﬁgures.01. The ﬁnal porosity after depletion of the soluble mineral is 0. The diffusivity of chemical species is calculated using the power law.2. dimensionless concentration and dimensionless pressure distributions within the computational domain at three different time instants.1 Geometry and boundary conditions of the reactive inﬁltration problem initial dimensionless-concentration is one within the computational domain.49) . while the dimensionless time-step is set to be 0. which can be determined from Eqs.51). Since the Zhao number of the system is smaller than its critical value. Figures 5. Using the above-mentioned parameters. 5. (5. The dimensionless speed of the dissolution front propagation is equal to 10. which has the power of 2. To simulate appropriately the propagation of the dissolution front.005 in the computation.2.77.e. the whole computational domain is simulated by 19701 four-node rectangular elements of 20000 nodal points in total.4 show the comparison of numerical solutions with analytical ones for the porosity.3 and 5. =0 ∂y ∂y φ =φ f C = C (τ) ∂p ′ = pfx ∂x Inflow φ =φo C =1 Ly = 5 y x 0 ∂p ∂C = 0. the thick line shows the numerical results. which is determined using Eq. (5. the coupled system considered in this section is sub-critical so that a planar dissolution front remains planar during its propagation within the system.112 5 Simulating Chemical Dissolution Front Instabilities in Fluid-Saturated Porous Rocks ∂C ∂p = 0. C(τ ) = exp( p fx ν front τ )) needs to be applied on the left boundary so that the numerical solutions can be compared with the analytical solutions derived in the previous section. The permeability of the porous medium is calculated using the Carman-Kozeny formula. =0 ∂y ∂y Lx = 10 Fig. a time-dependent-dimensionless-concentration boundary condition (i. The ratio of the equilibrium concentration to the solid molar density of the chemical species is assumed to be 0. Both the top and the bottom boundaries are assumed to be impermeable for the pore-ﬂuid and chemical species. which has the power of 3 in the power law.

1 0 10 0 0.25) x (τ = 0. τ ) = p fx (x − ν front τ ) − p 0x (L x − ν front τ ) + 100 (5.1 10 x (τ = 0. (x > ν front τ ). (x ≤ ν front τ ).86) p(x. τ ) = 100 at the right boundary of the computational model. it can be observed that the numerical solutions agree very well with the analytical solutions. As expected.1 0 10 x (τ = 0. (5.88) p(x. τ ) = φ0 (x > ν front τ ). φ(x. the porosity propagation front is the sharpest . indicating that the proposed numerical procedure is capable of simulating the planar dissolution-front propagation within the ﬂuidsaturated porous medium. τ ) = exp[− p fx (x − ν front τ )]. τ ) = φ f (5. τ ) = 1.2 Comparison of numerical solutions with analytical ones at different time instants (Porosity): the thick line shows the numerical results.2 Porosity 0. The resulting analytical solutions are expressed as follows: C(x.5.87) (5.625) 0. (x ≤ ν front τ ).2 Segregated Algorithm for Simulating the Morphological Evolution 113 0.89) From these results. while the thin line shows the corresponding analytical solutions and (5.2 Porosity 0. 5.8) Fig.2 Porosity 0. τ ) = − p 0x (L x − x) + 100 C(x.50) with the boundary condition of p(L x . φ(x.

while the thin line shows the corresponding analytical solutions Dimensionless pressure 180 Dimensionless pressure 0 x (τ = 0. while the thin line shows the corresponding analytical solutions .8) 10 Fig. 5.3 Comparison of numerical solutions with analytical ones at different time instants (Dimensionless concentration): the thick line shows the numerical results.25) 10 180 150 100 100 0 x (τ = 0.625) 10 Dimensionless concentration 1 0 0 x (τ = 0.8) 10 Fig.625) 10 Dimensionless pressure 180 140 100 0 x (τ = 0.4 Comparison of numerical solutions with analytical ones at different time instants (Dimensionless pressure): the thick line shows the numerical results.25) 10 0 0 x (τ = 0. 5.114 Dimensionless concentration 5 Simulating Chemical Dissolution Front Instabilities in Fluid-Saturated Porous Rocks Dimensionless concentration 1 1 0 0 x (τ = 0.

3 Application of the Segregated Algorithm 115 one among the three propagation fronts. p fx = −10) is applied on the left boundary of the computational domain so that the dimensionless speed of the dissolution front propagation is equal to 100. If both a small mesh size and a small time step are used. Since the Zhao number of the system is greater than its critical value. 5.001 in the computation. For this benchmark problem. It is observed that for the values of the dimensionless time greater than 0.5. namely a porosity propagation front. the coupled system considered in this section is supercritical so that a planar dissolution front evolves into a complicated morphology during its propagation within the system. The values of other parameters are exactly the same as those used in the previous section. In order to simulate the instability of the chemical dissolution front. Clearly.8% for dimensionless times of 0.e.25. a dimensionless-pressure gradient (i. a dimensionless-concentration propagation front and a dimensionless-pressure propagation front.3 Application of the Segregated Algorithm for Simulating the Morphological Evolution of Chemical Dissolution Fronts In this section. the propagation speed of the numerically-simulated propagation front is in good coincidence with that of the analytically-predicted propagation front. The Zhao number of the system is increased to 10. 4. the initial planar dissolution front gradually changes into an irregular one. This means that the dissolution front propagates much faster than it does within the system considered in the previous section. the ratio of the equilibrium concentration to the solid molar density of the chemical species is assumed to be 0.001. then the maximum relative error can be further reduced in the numerical simulation.5 shows the porosity distributions due to the morphological evolution of the chemical dissolution front in the ﬂuid-saturated porous medium.6% and 5. the proposed numerical procedure is used to simulate the morphological evolution of a chemical dissolution front in a supercritical system.77) of the system.8 respectively. while Fig. Figure 5. This quantitatively demonstrates that the proposed numerical procedure can produce accurate numerical solutions for the planar dissolution-front propagation problem within a ﬂuid-saturated porous medium. approximately 1. With a further increase of the dimensionless time. which is greater than the critical Zhao number (i.e. the overall accuracy of the numerical results is indicated by the dimensionless pressure. while the dimensionless time-step is also assumed to be 0.6 shows the dimensionless concentration distributions due to the morphological evolution of the chemical dissolution front within the computational domain. a small perturbation of 1% initial porosity is randomly added to the initial porosity ﬁeld in the computational domain. the amplitude of the resulting irregular . 0. implying that it is the least sharp front in the computational model.2%. the dimensionless-pressure propagation front has the widest bandwidth.625 and 0. 5. The maximum relative error of the numerically-simulated dimensionless pressure is 2. For this purpose. in the computational model. Although there are some smoothing effects on the numerically-simulated propagation fronts as a result of numerical dispersion.03. Due to this change.

5 Porosity distributions due to morphological evolution of the chemical dissolution front in the ﬂuid-saturated porous medium dissolution front increases signiﬁcantly.08) Fig. τ = 0.03) (Numerical. τ = 0. τ = 0. The peak value of the porosity is in good correspondence with the trough value of the dimensionless concentration due to the chemical dissolution in the system.07) (Numerical. τ = 0.116 5 Simulating Chemical Dissolution Front Instabilities in Fluid-Saturated Porous Rocks (Numerical.04) (Numerical. 5. . This demonstrates that the proposed numerical procedure is capable of simulating the morphological instability of the chemical dissolution front in a ﬂuid-saturated porous medium in the case of the coupled system being supercritical. indicating that the chemical dissolution front is morphologically unstable during its propagation within the computational model.05) (Numerical. the distribution of their maximum values along the dissolution front is clearly different. τ = 0. τ = 0. Although both the porosity and the dimensionless concentration have a similar propagation front.06) (Numerical.

τ = 0. there exists a clear transition for the dimensionless pressure-gradient distribution in the computational model.7 shows the dimensionless pressure distributions during the morphological evolution of the chemical dissolution front. τ = 0.07.06) (Numerical.8 shows the .04) (Numerical. This phenomenon can be clearly seen at the late stages of the numerical simulation such as when the dimensionless time is equal to 0.07) (Numerical. It is noted that although the dimensionless pressure is continuous. τ = 0.03) (Numerical. 5.06 and 0.6 Dimensionless concentration distributions due to morphological evolution of the chemical dissolution front in the ﬂuid-saturated porous medium It is interesting to investigate how the dimensionless pressure and pore-ﬂuid ﬂow evolve with time during propagation of the unstable dissolution front in the computational model. τ = 0. The ﬂuid-ﬂow pattern evolution during the propagation of the unstable dissolution front is exhibited by the streamline evolution in the computational model. τ = 0.3 Application of the Segregated Algorithm 117 (Numerical. Figure 5. Figure 5.08) Fig.5. τ = 0.05) (Numerical.

5. 5. which can be observed from the streamline density (in Fig. the proposed numerical procedure can provide a useful tool for simulating the related physical and chemical processes associated with the generation of giant ore deposits within the upper crust of the Earth. pore-ﬂuid pressure and reactive chemical-species transport. pore-ﬂuid ﬂow focusing takes place in the peak range of the porosity. τ = 0. It is noted that the width of the ﬂow focusing zone is closely associated with the peak and trough values of the irregular dissolution front in the computational model.08) Fig.03) (Numerical. Since both the porosity generation and the pore-ﬂuid ﬂow focusing play an important role in ore body formation and mineralization. . τ = 0. τ = 0.7 Dimensionless pressure distributions due to morphological evolution of the chemical dissolution front in the ﬂuid-saturated porous medium streamline distributions during the morphological evolution of the chemical dissolution front within the coupled system between porosity.8).07) (Numerical.04) (Numerical. τ = 0. Due to the growth of the amplitude of the irregular dissolution front. τ = 0. τ = 0.05) (Numerical.06) (Numerical.118 5 Simulating Chemical Dissolution Front Instabilities in Fluid-Saturated Porous Rocks (Numerical.

5.8 Streamline distributions due to morphological evolution of the chemical dissolution front in the ﬂuid-saturated porous medium .08) Fig.3 Application of the Segregated Algorithm 119 (τ = 0.06) (τ = 0.07) (τ = 0.5.

the kinetics of a redox chemical reaction and its interaction with physical processes are often overlooked in the numerical modelling of ore forming systems. 1997a. Since pore-ﬂuid is a major carrier transporting chemical species from one part of the crust into another. Xu et al. 2000b. Nield and Bejan 1992. 1999b. 2002c. precipitation and rock alteration are a direct consequence of coupled processes between ﬂuids mixing.. Zhao et al. Although extensive studies have been carried out to understand the possible physical and chemical processes associated with ore body formation and mineralization (Phillips 1991. the chemical process is also coupled with the heat transfer process. In most chemical reactions associated with an ore forming system. understanding the controlling mechanisms behind ore body formation and mineralization within the upper crust of the Earth becomes a very important research ﬁeld. C Springer-Verlag Berlin Heidelberg 2009 121 . Steefel and Lasaga 1994. Schaubs and Zhao 2002. heat transfer and chemical reactions in ﬂuid-saturated porous rocks. 2002. if the rate of a chemical reaction is dependent on temperature. Fundamentals of Computational Geoscience. Lecture Notes in Earth Sciences 122. They play a very important role in ore body formation and alteration closely associated with a mineralizing system. the resulting patterns of mineral dissolution. Yeh and Tripathi 1991. Schafer et al. advection and chemical kinetics must be considered. Zhao et al.1007/978-3-540-89743-9 6. Zhao et al. b. 2001d. DOI 10. the chemical process is coupled with the pore-ﬂuid ﬂow process in ﬂuid-saturated porous rocks. 2002. Heat Transfer and Non-Equilibrium Redox Chemical Reactions in Fluid-Saturated Porous Rocks Non-equilibrium redox chemical reactions of high orders are ubiquitous in ﬂuid-saturated porous rocks within the crust of the Earth. 2001b. Raffensperger and Garven 1995. Ord et al. Due to ever-increasing demands for minerals and possible exhaustion of the existing mineral deposits in the foreseeable future.Chapter 6 A Decoupling Procedure for Simulating Fluid Mixing. Zhao et al. 1998a. Gow et al. When a pore-ﬂuid carrying one type of chemical species meets with that carrying another type of chemical species. 1998a. For these reasons. these two types of pore-ﬂuids can mix together to allow the related chemical reaction to take place due to solute molecular diffusion/dispersion and advection. There is no doubt that an improved understanding of ore forming processes can signiﬁcantly promote mineral exploration for new ore deposits within the upper crust of the Earth. 1999. the chemical reaction rate is ﬁnite so that an interaction between the solute molecular diffusion/dispersion. C. 2003a). In addition. transportation.

a numerical approach based on the ﬁnite element method is used to solve coupled problems between ﬂuids mixing. The intermediate case between the ﬁrst and the second category belongs to the third category of coupled problem. the ﬁrst category of coupled problem is often called the non-reactive mass transport problem. 1998a). For this reason. the second category of coupled problem is called the quasi-instantaneous equilibrium problem. heat transfer and redox chemical reactions in ﬂuid-saturated porous rocks. As a result. heat transfer and redox chemical reactions in ﬂuid-saturated porous rocks. at least from the mathematical point of view. we will develop a numerical procedure to solve coupled problems between ﬂuids mixing. Keeping the above-mentioned considerations in mind. Although signiﬁcant achievements have been made for the numerical modelling of non-reactive species and quasi-instantaneous equilibrium reaction transport problems. advection and chemical kinetics is very strong within and around a fault/crack. When the permeability of a fault/crack is much bigger than that of the surrounding rock. Although it is well known that ore body formation and mineralization are associated with geological faults and cracks. the time scale of the advective ﬂow is much smaller than that of the relevant chemical reaction in porous rock masses so that the rate of the chemical reaction can be essentially taken to be zero in the numerical analysis. in which the rate of the relevant chemical reaction is a positive real number of ﬁnite value. This implies that an interaction between the solute diffusion. In the ﬁrst category of coupled problem. it is possible to divide the coupled problems into the following three categories (Zhao et al. the concept of the chemical reaction rate invariant is used to convert the conventional reactive transport equations with strong chemical reaction terms into the following two different . heat transfer and chemical reactions in ﬂuid-saturated porous rocks. In order to improve the efﬁciency of numerical modelling. For this reason. research on the numerical modelling of the third category of coupled problem with redox chemical reactions is rather limited. for the second category of coupled problem. In contrast. Another signiﬁcant characteristic of the third category of coupled problem is that the detailed kinetics of the chemical reactions must be taken into account.122 6 Fluid Mixing. both the forward reaction rate and the backward one need to be included in the reaction term of a reactive species transport equation. Large geological faults and cracks are favorable locations for ﬂuids carrying different chemical species to focus and mix. This means that the equilibrium state of the chemical reaction involved is always attained in this category of coupled problem. the pore-ﬂuid ﬂow in the fault/crack is much faster than that in the surrounding rock. If a redox chemical reaction is considered. Considering this fact. the time scale of the advective ﬂow is much larger than that of the relevant chemical reaction in pore-ﬂuid saturated porous rocks so that the rate of the chemical reaction can be essentially taken to be inﬁnite. Heat Transfer and Non-Equilibrium Redox Chemical Reactions In terms of numerical modelling of coupled problems between ﬂuids mixing. ore body formation and mineralization are often associated with geological faults and cracks. the major factors controlling the reaction patterns within and around large faults and cracks remains unclear. It is the kinetics of a chemical reaction that describes the reaction term in a reactive species transport equation.

advection and chemical kinetics to be investigated within and around faults and cracks in the upper crust of the Earth. This allows the interaction between the solute molecular diffusion/dispersion. ∂x2 ∂y (6. the governing equations of the coupled problem between ﬂuids mixing. while another remains the same as the third category of reactive transport equation with a strong reaction term. . which can be used to remove the coupling between redox types of reactive transport equations. 6. ∂y u ∂T ∂T +ν ∂x ∂y ∂T + ρ f cpf ∂t = Dex φ ∂ 2 Ci ∂ 2 Ci +φ Ri + Dey ∂x2 ∂ y2 ρ f = ρ f 0 [1 − βT (T − T0 )]. heat transfer and redox chemical reactions in ﬂuid-saturated porous rocks.5) (6. . Since the solution of a reactive transport equation with a chemical reaction term is computationally much more expensive than that of a mass transport equation without a chemical reaction term. . .3) = λex ∂2T ∂2T + λey 2 . . Fourier’s law and Fick’s law can be used to describe the heat transfer and mass transport phenomena respectively.1 Statement of Coupled Problems between Fluids Mixing.N ). a decoupling numerical procedure.6) ∂P + ρf g . Darcy’s law can be used to describe poreﬂuid ﬂow and the Oberbeck-Boussinesq approximation is employed to describe a change in pore-ﬂuid density due to a change in the pore-ﬂuid temperature. Heat Transfer and Redox Chemical Reactions 123 kinds of equations: One is the same as the ﬁrst category of mass transport equation without any reaction term. ∂x ∂y u= ν= φρ f c p f + (1 − φ)ρs c ps ∂Ci ∂Ci ∂Ci + u +ν ∂t ∂x ∂y Ky μ Kx μ − − ∂P ∂x . (6. Based on this idea. 2.2) (6. heat transfer and redox chemical reactions in ﬂuid-saturated porous rocks can be expressed as follows: ∂ν ∂u + = 0. any reduction in the total number of reactive transport equations can signiﬁcantly save computer time in a numerical computation.1 Problems between Fluids Mixing. Heat Transfer and Redox Chemical Reactions For pore-ﬂuid saturated porous rocks.1) (6. (6.4) (i = 1.6. If the pore-ﬂuid is assumed to be incompressible. is proposed to solve coupled problems between ﬂuids mixing.

(6. λfy and λsy are the thermal conductivities of the pore-ﬂuid and solid matrix in the y direction.4).7) (6. In principle.3) and (6. φ is the porosity of the porous medium. The ﬁrst is a coupled problem between the pore-ﬂuid ﬂow and the heat transfer process.5) and the related chemical reaction equations. μ is the dynamic viscosity of the pore-ﬂuid. heat transfer and redox chemical reactions in ﬂuid-saturated porous rocks can be divided into two new problems. their contributions to the density of the pore-ﬂuid are negligible so that the mass transport process can be decoupled from the pore-ﬂuid ﬂow and heat transfer processes. This requires that the chemical reaction be of the ﬁrst order. g is the acceleration due to gravity. Ri is the source/sink term for the reactive transport equation of chemical species i. It is noted that if the aqueous mineral concentrations associated with ore body formation and mineralization are relatively small. Ci is the concentration of chemical species i. the new concept . 2003a). c p f and c ps are the speciﬁc heat of the pore-ﬂuid and solid matrix respectively. then the coupled problem deﬁned between ﬂuids mixing. both the second order and the high order chemical reactions are very common in nature. Since many chemical reactions of different orders are associated with ore body formation and mineralization in ﬂuid-saturated porous rocks. P is the pore-ﬂuid pressure. can be solved using the existing ﬁnite element method (Lewis and Schreﬂer 1998. N is the total number of the active chemical species considered in the pore-ﬂuid. Zienkiewicz 1977). K x and K y are the permeabilities of the porous medium in the x and y directions respectively. (6. λey = φλfy + (1 − φ)λsy . the second order redox chemical reaction is considered in order to develop a concept resulting in a new decoupling procedure for removing the coupling between reactive transport equations of redox chemical reactions. Heat Transfer and Non-Equilibrium Redox Chemical Reactions λex = φλfx + (1 − φ)λsx . Dey = φ Dfy . (6. (6.2). ρ f and ρs are the densities of the pore-ﬂuid and solid matrix. Without loss of generality. 1998a. which is described by Eqs. T is the temperature of the porous medium. (6. Since the ﬁrst coupled problem.8) where u and v are the horizontal and vertical velocity components of the pore-ﬂuid in the x and y directions respectively. heat transfer and redox chemical reactions in ﬂuid-saturated porous rocks above is solvable using the numerical methods available (Zhao et al.5) can be determined and is linearly dependent on the chemical species concentration. Dex = φ Dfx . βT is the thermal volume expansion coefﬁcient of the pore-ﬂuid. If the reaction term in Eq.124 6 Fluid Mixing. λfx and λsx are the thermal conductivities of the pore-ﬂuid and solid matrix in the x direction. This means that the whole coupled problem between ﬂuids mixing. Dfx and Dfx are the diffusivities of the chemical species in the x and y directions respectively. ρ f 0 and T0 are the reference density and reference temperature used in the analysis. which is described by Eq. the main purpose of this study is to develop a new decoupling procedure to effectively and efﬁciently solve the second coupled problem. (6. while the second is a coupled problem between the mass transport process and the redox chemical reaction process.1).

It is noted that the accumulation or diffusion of chemical species in the rock matrix may lead to some change in porosity. r is a quantity of unity value to balance the unit of the reaction source/sink terms due to different orders of chemical reactions so that it has a reciprocal unit of the chemical species concentration. kb kb kf kf kf CO2− ⇔ CaCO3 3 kb Ca 2+ + and so forth in geochemical systems.e. kb kf (6. This . (6. chemical reactants A and B are Ca2+ and CO2− .6.11) (6. the forward reaction is of the second order. while chemical product AB is NaCl. B and AB. while chemical product AB is H2 O. which in turn affects permeability and ﬂuid ﬂow in the rock matrix (Zhao et al. B and AB. in the third chemical reaction example. Similarly. R B and RAB are the chemical reaction source/sink terms associated with chemical species A. It is clear that in the ﬁrst chemical reaction example. 2001d. 2004). chemical reactants A and B are H + and OH− .9) represents a class of redox chemical reactions such as H + + OH− ⇔ H2 O. R A .9) where A and B are two chemical reactants. In the second chemical reaction example. AB is the chemical product due to this redox chemical reaction.12) where C A . chemical reactants A and B are Na+ and Cl − . n f and n b are the orders of the forward and backward reactions respectively. the general chemical reaction source/sink terms due to the redox chemical reaction expressed by Eq. Since a redox system allows chemical reactions to be proceeded toward both the product and the reactant directions.e. the orders of the forward reaction (i. (6.9) can be written as follows: R A = −k f r n f −1 C A C B + kb r n b −1 CAB . For the redox reaction expressed by Eq.1 Problems between Fluids Mixing. C B and CAB are the concentrations of chemical species A. R B = −k f r n f −1 C A C B + kb r n b −1 CAB .10) (6. k f and kb are the forward and backward reaction rates of this redox chemical reaction.9). while the backward reaction is of the ﬁrst order. 3 From the chemical reaction point of view. while chemical product AB is CaCO3 . (6. (6. Xu et al. RAB = k f r n f −1 C A C B − kb r n b −1 CAB . a redox chemical reaction of the second order is considered as follows: A + B ⇔ AB. Heat Transfer and Redox Chemical Reactions 125 and decoupling procedure can be extended to deal with the high order redox chemical reactions. the chemical reaction proceeds toward the reactant direction) can be determined from the related chemical kinetics. It needs to be pointed out that Eq. Na+ + Cl − ⇔ NaCl. the chemical reaction proceeds toward the product direction) and backward reaction (i. For this reason.

Through some algebraic manipulations. From this point of view. (6.11) and (6.15) can be rewritten as follows: φ ∂(C A + CAB ) ∂(C A + CAB ) ∂(C A + CAB ) + u +ν ∂t ∂x ∂y 2 ∂ (C A + CAB ) ∂ 2 (C A + CAB ) = 0. + Dey ∂x2 ∂ y2 (6. (6. (6.14) and (6.16) φ (6. For example.15) Since the total number of linearly-independent reaction rates is identical to the total number of chemical reactions involved. the total number of reactive transport equations with source/sink terms due to chemical reactions can be reduced into one. for this particular redox chemical system. the Carman-Kozeny law can be used to establish a relationship between the chemically induced porosity change and the chemically induced permeability change in the rock matrix. − Dex + Dey 2 ∂x ∂ y2 ∂(C B + CAB ) ∂(C B + CAB ) ∂(C B + CAB ) + u +ν ∂t ∂x ∂y ∂ 2 (C B + CAB ) ∂ 2 (C B + CAB ) − Dex = 0. (6.10).14) φ ∂CAB ∂CAB ∂CAB + u +ν ∂t ∂x ∂y = Dex + φ RAB . Eqs. The permeability change induced by a chemical reaction can be determined from the porosity variation induced by the chemical reaction. (6. (6. (6.13) φ = Dex + φ RB .5) yields the following equations: φ ∂C A ∂C A ∂C A + u +ν ∂t ∂x ∂y ∂C B ∂C B ∂C B + u +ν ∂t ∂x ∂y = Dex ∂ 2C A ∂ 2C A + Dey ∂x2 ∂ y2 ∂ 2C B ∂ 2C B + Dey ∂x2 ∂ y2 ∂ 2 CAB ∂ 2 CAB + Dey ∂x2 ∂ y2 + φ RA. Heat Transfer and Non-Equilibrium Redox Chemical Reactions inﬂuence can be straightforwardly considered using variable permeability within the computational model. (6.2 A Decoupling Procedure for Removing the Coupling between Reactive Transport Equations of Redox Chemical Reactions Substituting Eqs.17) .126 6 Fluid Mixing.12) into Eq. 6. there is only one linearly-independent reaction rate for this redox chemical reaction.13). This is the basic idea behind the proposed decoupling procedure for removing the coupling between reactive transport equations of redox chemical reactions.

the chemical product distribution due to the chemical reaction can be calculated analytically. In this case. (6. they can be referred to as chemical reaction rate invariants. into Eq. Eq. Clearly.e. (6. Inserting the two chemical reaction rate invariants. the chemical reaction rates are ﬁnite so that we need to solve at least one reactive transport equation with the source/sink term for each of the chemical reactions in the geochemical system. (6. (6. 2r 2 4K e (6. are independent of chemical reaction rates. Once the distributions of the chemical reaction rate invariants are obtained in a computational domain. Eq. If the redox chemical reaction is an equilibrium one.21) This indicates that for an equilibrium chemical reaction. As a result.18) can be written as K e r n f −1 C A C B − r n b −1 CAB = 0.20) has the following mathematical solution for the chemical product of the redox chemical reaction: CAB = K e r (C I + CII ) + 1 − 2K e r 2 [K e r (C I + CII ) + 1]2 − 4K e r 2 C I CII .16) and (6. which are the analogues of the stress and strain invariants in the ﬁeld of solid mechanics. namely C I = C A + CAB and CII = C A + CAB . for non-equilibrium chemical reactions.6.20) It is noted that n f = 2 and n b = 1 are substituted into Eq. (6. (6. Since the two new variables.19) where K e = k f /kb is the chemical equilibrium constant.20).18) Equations (6. C I = C A +CAB and CII = C B + CAB in this particular example) using the conventional numerical methods. we need to solve Eq. kb (6.19) yields the following equation: K e r (C I − CAB )(CII − CAB ) − CAB = 0. the distributions of the chemical reactants can be calculated using the distributions of both the chemical product and the chemical reaction rate invariants. we only need to solve the mass transport equations of chemical reaction rate invariants (i. C I = C A + CAB and CII = C B + CAB .17) are two conventional mass transport equations without any source/sink terms due to the redox chemical reaction so that they can be solved by the well-developed numerical methods available.18) numerically if the reaction rates of this redox .2 A Decoupling Procedure for Removing the Coupling 127 φ ∂CAB ∂CAB ∂ 2 CAB ∂ 2 CAB ∂CAB + u +ν − Dex + Dey ∂t ∂x ∂y ∂x2 ∂ y2 k f n f −1 = φkb r C A C B − r n b −1 CAB . (6.19) so as to obtain Eq. However. then both the forward and the backward reaction rates are theoretically inﬁnite so that the chemical reaction becomes predominant in the reactive transport process. (6. This means that for the general form of the redox chemical reaction considered in this study.

mass transport equations of the chemical reaction rate invariants (i. Using the proposed decoupling procedure. only can an approximate solution be obtained from a computational model. These solution steps have been programmed into our research code. (6.e. (6.1 shows the geometry of the coupled problem between pore-ﬂuids mixing.3 Veriﬁcation of the Decoupling Procedure The main and ultimate purpose of a numerical simulation is to provide numerical solutions for practical problems in a real world.e. at least from the qualitative point of view. which is the discretized description of a continuum mathematical model.22) can be solved using the Newton-Raphson algorithm. The height and width of the computational model are 10 km . Eqs.1). the chemical reactant concentrations (i. For this reason.17)) with the related boundary and initial conditions are then solved using the existing ﬁnite element method.128 6 Fluid Mixing. it is necessary to verify. For this reason. (6. (6. (5) Steps (1–4) are repeated until the desired time step is reached.3). is considered in this section.22) is strongly nonlinear with the nonlinear term. the coupled problem described by Eqs. Equation (6. heat transfer and redox chemical reactions around a vertical geological fault within the crust of the Earth.18) can be rewritten into the following form: φ ∂CAB ∂CAB ∂ 2 CAB ∂ 2 CAB ∂CAB + u +ν + Dey − Dex 2 ∂t ∂x ∂y ∂x ∂ y2 − φkb 2 K e rCAB (6. Heat Transfer and Non-Equilibrium Redox Chemical Reactions chemical reaction are ﬁnite.16) and (6. (6. for some aspects of which the analytical solutions are available. (2) After the pore-ﬂuid velocities are obtained from the ﬁrst step. (6.22) − [K e r (C I + CII ) + 1] CAB + K e rC I CII = 0.2). 2 Since Eq. implying that there is an upward throughﬂow at the bottom of the computational model. the pore-ﬂuid pressure is assumed to be lithostatic. (6. C I = C A +CAB and CII = C B +CAB . (6. Due to inevitable round-off errors in computation and discretized errors in temporal and spatial variables. a testing coupled problem. Figure 6. φkb K e rCAB . the Newton-Raphson algorithm is suitable for solving this equation. the coupled problem between ﬂuids mixing. (3) The chemical reaction source/sink terms involved in Eq. Since numerical methods are the basic foundation of a numerical simulation. For this problem. heat transfer and redox chemical reactions in ﬂuid-saturated porous rocks can be solved in the following ﬁve main steps: (1) For a given time step. 6.5) with the related boundary and initial conditions are solved using the conventional ﬁnite element method. the proposed numerical procedure so that meaningful numerical results can be obtained from a discretized computational model. C A and C B ) can be determined from simple algebraic operations.22) is determined from the related redox chemical reaction so that Eq.4) and (6. (4) According to the deﬁnitions of the chemical reaction rate invariants.

while the concentrations of both the two reactants and the product are assumed to be zero at the top boundary of the computational model. The surrounding rock is assumed to have a permeability of 10–16 m2 in both the x and y directions.e.1 Geometry and boundary conditions of the ﬂuid focusing and mixing problem 10km T = 325oC.9 W/(m × o C) in both the x and y directions. which gives rise to a permeability of about 43 times that of the surrounding rock. In order to simulate the ﬂuids focusing and chemical reactions within the fault in an appropriate manner. The porosities of the fault and its surrounding rock are 0. dynamic viscosity is 10−3 N × s/m2 . while the permeability of the fault is calculated using the Carman-Kozeny formula. Figure 6. reference density is 1000 kg/m3 . The dispersion/diffusivity of the chemical species is 3 × 10–10 m2 /s. but diverges out of the vertical fault at the outlet (i.6.35 and 0.2 shows the streamline distribution of the system with the vertical fault in the computational model. the whole computational domain is simulated by 306. CA = 0. the upper end) of the fault.e. a mesh of small element sizes is used to simulate the fault zone. CB = 0 129 Fig.417 three-node triangle elements. where the velocity distributions are displayed at both the inlet and . meaning that the temperature at the bottom is ﬁxed at 325◦ C. the pore-ﬂuid ﬂow converges into the vertical fault at the inlet (i. As a result. The two chemical reactants with a concentration of 1 kmol/m3 are injected at the left half and right half bottom boundary respectively. The top temperature is 25◦ C and the geothermal gradient is 30◦ C/km. p = α1 pHydrostatic C A = 1kmol / m3 20km C B = 1kmol / m3 and 20 km respectively. For the pore-ﬂuid. reference rock density is 2600 kg/m3 .3. 6.59 W/(m × o C) in both the x and y directions.1 × 10–4 (1/◦ C).3 Veriﬁcation of the Decoupling Procedure T = 25oC. the lower end) of the fault. with an aspect ratio of 20. thermal conductivity coefﬁcient is 0. the speciﬁc heat is 878 J/(kg × o C). This phenomenon can be clearly observed from Fig. The length of the fault is 5 km. p = 0. Since the pore-ﬂuids carrying two different chemical reactants are uniformly and vertically injected into the computational model at the left and right parts of the bottom. while a mesh gradation scheme is used to simulate the surrounding rock by gradually changing the mesh size from the outline of the fault within the computational model. 6. the thermal conductivity coefﬁcient is 2.1. volumetric thermal expansion coefﬁcient is 2. For the porous matrix. speciﬁc heat is 4184 J/(kg × o C).

In order quantitatively to validate the numerical solutions. the analytical solution for the ﬂow-focusing factor of an elongate elliptic fault of large aspect ratios can be employed. the computed streamlines concentrate vertically within the vertical fault. As expected. 1999d). Heat Transfer and Non-Equilibrium Redox Chemical Reactions (Whole system view) (Zoom-in view at the inlet of the fault) (Zoom-in view at the outlet of the fault) Fig. This demonstrates that the proposed numerical procedure can produce useful numerical solutions for ﬂuid focusing and mixing. the analytical solution for its ﬂow-focusing factor can be evaluated using the following modiﬁed formula: . For an elongate elliptic fault of large aspect ratios in an inﬁnite medium. 6. indicating that the numerical result obtained from the computational model has good agreement with the existing analytical result. the existing analytical solution indicates that the streamlines of the pore-ﬂuid ﬂow are parallel each other inside of the fault (Zhao et al. Since the elongate fault within the computational model is basically of a rectangular shape.130 6 Fluid Mixing.2 Streamline distributions due to the ﬂuid focusing within a vertical fault in the crust outlet of the vertical fault. at least from the qualitative point of view.

23) where λ is the pore-ﬂuid ﬂow focusing factor of the rectangular fault of a large aspect ratio. 6. Since the vertical velocity of the injected ﬂuids due to the lithostatic pore-ﬂuid pressure is equal to 1. α is the permeability ratio of the fault to its surrounding rock.3 Veriﬁcation of the Decoupling Procedure 131 (Zoom-in view at the inlet of the fault) (Zoom-in view at the outlet of the fault) Fig.3 Velocity distributions due to the ﬂuids focusing within a vertical fault in the crust λ= 4 π α(β + 1) . β is the aspect ratio of the rectangular fault. β +α (6.6.6 × 10–9 m/s and the numerical solution for the maximum .

t = 1000 years) (Reactant B. The analytical value of the ﬂow-focusing factor for the rectangular fault can be calculated from Eq. Since the relative error of the ﬂow-focusing factor from the numerical simulation is within 3. (6.25. t = 5000 years) (Reactant A. t = 1000 years) (Reactant A.88 for the rectangular fault within the computational model.5%. (Reactant A. the corresponding ﬂow-focusing factor of the fault is equal to the ratio of the maximum velocity within the fault to that of the injected ﬂuids at the bottom of the computational model. Substituting the related parameters into Eq. it quantitatively demonstrates that the proposed numerical procedure used in the computational model can produce accurate numerical solutions for ﬂuid focusing and mixing within the fault.23). t = 8000 years) Fig.02 × 10–8 m/s.132 6 Fluid Mixing.4 Concentration distributions of the chemical reactants at different time instants (Equilibrium reaction) . This results in a ﬂow-focusing factor of 18. t = 5000 years) (Reactant B.23) yields the analytical ﬂow-focusing factor of 18. 6. (6. Heat Transfer and Non-Equilibrium Redox Chemical Reactions vertical velocity within the fault is equal to 3. t = 8000 years) (Reactant B. These processes are very important for accurately simulating the chemical species transport and reaction within the fault.

t = 8000 years) (Numerical.3 Veriﬁcation of the Decoupling Procedure 133 In order to verify the proposed decoupling procedure for solving reactive transport equations with strong nonlinear reaction source/sink terms. t = 5000 years) (Analytical. t = 8000 years) Fig. Figure 6.21)) for the chemical product. t = 1000 years) (Numerical. while Fig.5 Comparison of numerical solutions with analytical ones for the chemical product (Equilibrium reaction) .5 shows the comparison of the numerical solutions (which are obtained from the proposed decoupling procedure) with the analytical solutions (which are derived mathematically and expressed by Eq. these chemical reactants (Analytical.6. 6. 6. Due to the ﬂuid ﬂow focusing. t = 1000 years) (Analytical. both chemical reactants are transported much faster in the fault zone than in the surrounding rock. the redox chemical reaction due to an equilibrium reaction is considered and solved using the proposed numerical procedure. The equilibrium constant is assumed to be 10 and the time step used in the simulation is 100 years.4 shows the concentration distributions of the two chemical reactants at three different time instants. It can be observed that with the increase of time. (6. t = 5000 years) (Numerical. both chemical reactants are transported into the computational domain from the left half and right half of the bottom. As expected.

Consequently. An example is the production of CO2 or of hydrocarbons from carbonates or carbonaceous material within a ﬂuid stream that is hot. Clearly rock-ﬂuid(s) interactions must be involved as well as the ﬂuid mixing process but in depth exploration of such multi-ﬂuid-rock reaction processes awaits robust and computationally fast ways of handling realistic kinetic-reaction-transport phenomena. the mixing of the two ﬂuids carrying them is controlled by the solute diffusion. Heat Transfer and Non-Equilibrium Redox Chemical Reactions are divergent around the exit region of the fault. 1995. 1995. 6. 1999). The ﬁrst we refer to as parallel-ﬂow geometries of the ﬁrst kind (Figs. b) where two contrasting ﬂuids are brought alongside each other by convection or focusing in a highly permeable fault or sedimentary lens (Phillips 1991). This is the reason why both chemical reactants are abundant but no chemical product is produced within the fault zone in the computational model. Although two reactants are well transported into the fault zone.1). 1995) and Carlin gold deposits (Cline and Hofstra 2000). Garven et al. a high concentration of the chemical product is produced around the exit region of the fault zone.6c). 6. The mixing process is attractive because it enables two ﬂuids of contrasting Eh-pH conditions to mix and hence generate chemical conditions conducive to mineral precipitation. It is interesting to note that there is a strong interaction between solute advection.4 Applications of the Proposed Decoupling Procedure to Predict Mineral Precipitation Patterns in a Focusing and Mixing System Involving Two Reactive Fluids Mixing of two or more ﬂuids is commonly suggested as a mechanism for precipitating minerals from solution in porous rocks. Cox et al. the corresponding chemical equilibrium length is identical to zero due to the solute diffusion. diffusion and chemical reaction rate in the considered equilibrium chemical system. The second involves the production of a new ﬂuid within an existing ﬂuid ﬂow system by thermal or chemical reactions with a mineral assemblage within the ﬂow stream (Fig. The comparison of the numerical solutions with the analytical ones for the chemical product clearly demonstrates that the proposed decoupling procedure can produce accurate numerical solutions for simulating the equilibrium chemical reaction. 6. This implies that the chemical reaction rate is too fast to allow both the reactants to diffuse across the common boundary between them so that ﬂuid mixing cannot effectively take place within the fault zone. 6. Examples include uranium deposits (Wilde and Wall 1987). around the exit region of the fault zone. Irish Pb-Zn deposits (Hitzmann. However. MVT deposits (Appold and Garven 2000. There are three end member geometries that promote the mixing of two miscible ﬂuids (Fig. acid and . Everett et al. 1999.134 6 Fluid Mixing. in which the chemical reaction rate approaches inﬁnite. Since the chemical reaction rate is inﬁnite for the equilibrium reaction.6a. Murphy et al. the ﬂow of the ﬂuids is slowed and divergent so that the ﬂuids carrying two different reactants can be mixed. 2008). vein gold deposits (Matthai et al.

6.6d. We refer to this process as parallel ﬂow geometry of the second kind.6 Mixing geometries for two ﬂuids: (a) and (b) are parallel mixing ﬂows of the type discussed in this paper. in fact. 6. The third end member involves injection of one ﬂuid perpendicular to the ﬂow ﬁeld of a second ﬂuid as shown in Figs. (c) is a parallel mixing ﬂow but the second ﬂuid is generated within the ﬂow ﬁeld of the ﬁrst ﬂuid. 1995). This is. 1995. a special example of parallel-ﬂow geometry except that now the new ﬂuid source is embedded within the primary ﬂow stream. This mixing geometry produces a mixing plume that expands outwards mainly in the downstream direction of the Fluid A Fluid B High Permeability Lens Fluid A Fluid B (a) (b) Fluid B Fluid B Fluid A Fluid A Fluid A (c) (d) Fluid B Fluid A (e) Fig.4 Applications of the Proposed Decoupling Procedure 135 oxidised (Matthai et al. 6. e. (d) and (e) are perpendicular mixing ﬂows where one ﬂuid is injected normal to the ﬂow ﬁeld of the other ﬂuid . Cox et al.

ﬂuid pressure and the concentration of other chemical species. In many situations. chemical reaction rates. that is. 6. Ore formation in the Earth’s crust is commonly considered to be a consequence of the reaction of large ﬂuxes of one or more ﬂuids with rock. However. ﬂuid systems in which two or more ﬂuids are present that are immiscible. . solute diffusion and/or dispersion and chemical kinetics. The discussion is equally relevant for any other orientation of the focusing “lens”. The fundamental point we explore here is that the overall pattern of mineralization in these mixing systems results from intimate interactions between solute advection. We show. due to the large ﬂuid ﬂuxes typically inferred for such zones. It is commonly assumed that either ﬂuid mixing or ﬂuid/rock reactions or a combination of both are important processes associated with permeable fault zones.1 Key Factors Controlling Mineral Precipitation Patterns in a Focusing and Mixing System Involving Two Reactive Fluids We consider the ﬁrst of these parallel ﬂow geometries in some detail. comprise yet another hydraulic-chemical system where intimate mixing of contrasting chemistries is possible at the pore scale. for instance. chemical equilibrium in some cases may never be attained in these mixing systems. recent extensive studies (Phillips 1991. chemical reaction between the two ﬂuids may never occur even if chemical equilibrium prevails so long as the advection of solute is large with respect to solute diffusion/dispersion. In order to focus on the principles involved for parallel ﬂows of the ﬁrst kind we idealise the situation by considering a vertical permeable fault within a less permeable rock mass. The whole model is fully saturated with ﬂuids of different chemical compositions at different points on the system. We need to emphasise here that although we are considering only vertical permeable faults the discussion is just as relevant to permeable sedimentary lenses or to any other geometry that brings two chemically contrasted ﬂuids together. This situation has been considered by Phillips (1991) and analytical solutions for the composition of the resulting mixing plume presented. The classical approach (see Phillips 1991) is to assume that chemical equilibrium has been established within the mixing plume so that the equilibrium concentration of a particular chemical species is expressed as a function of the environmental parameters. that even for a situation where two ﬂuids are brought together by ﬂuid focusing within a permeable fault zone. ﬂuid ﬂow is focused within fault zones and the important issue becomes the spatial control exerted by the fault on the resultant mineralization. The vertical geometry here allows simple boundary conditions to be imposed for the hydraulic potential driving ﬂuid ﬂow. Thus.136 6 Fluid Mixing. Heat Transfer and Non-Equilibrium Redox Chemical Reactions second ﬂow.4. We consider only systems where the ﬂuids are miscible whilst noting for future investigation that multi-phase systems. namely temperature. In this section we concentrate solely on parallel ﬂows of the ﬁrst kind and show that the patterns of chemical precipitation for such ﬂows result from intimate relations between ﬂuid ﬂow (which is a ﬁrst order control on solute advection). and solute diffusion and dispersion.

In addition. a redox controlled chemical reaction can only reach equilibrium through optimal coupling of solute advection. the chemical product cannot attain equilibrium. produced by a chemical reaction. solute diffusion/dispersion and chemical kinetics. In this case. 2007a) have demonstrated that ore formation is not only dependent on ﬂuid ﬂow. Xu et al. which we introduce below.4 Applications of the Proposed Decoupling Procedure 137 Zhao et al. Braun et al. Zhao et al. Zhao et al. These chemical reaction patterns are strongly inﬂuenced by ﬂuid ﬂow. the solubility of an aqueous species is directly proportional to the ambient temperature. if the solute diffusion/dispersion in the direction perpendicular to the interface between the ﬂows is much slower than the chemical reaction rate. Phillips 1991. In such a case. there exists an optimal ﬂow rate resulting in chemical equilibrium being attained between two ﬂuids that mix and focus within the fault zone. strong interactions occur between solute advection. Since ﬂuid ﬂow is an important agent for transporting aqueous chemical species from one location into another (Garven and Freeze 1984. chemical equilibrium may or may not be attained when two ﬂuids of different origins are brought together through focusing within a permeable vertical fault. Yardley and Lloyd 1995). if the ﬂow rate is much faster than the chemical reaction rate. The aqueous species can also become oversaturated and is precipitated in the porous rock. Mineralisation is commonly associated with the spatial distribution of chemical reactions in permeable rocks (Steefel and Lasaga 1990. whether chemical equilibrium is or is not attained controls the possibility of precipitation of mineral assemblages in permeable rocks. On the other hand. However. for parallel ﬂows. diffusion/dispersion and chemical kinetics. Ord and Oliver 1997. such as those resulting from vertical super-hydrostatic pressure gradients. In most cases. For a given ore formation environment. heat transfer processes inﬂuence the thermal structure of the Earth’s crust and hence the spatial distribution of chemical reaction patterns through the temperature-dependence of the equilibrium constant and of the reaction rate (Steefel and Lasaga 1994. For aqueous species. the chemical product can only attain equilibrium at a distance large with respect to a chemical equilibrium length-scale. ﬂow rates can control the total process of solute advection. but also upon solute diffusion/dispersion and chemical kinetics. Another process causing mineral precipitation is transport within the pore-ﬂuid of the saturated aqueous species from a high temperature region to a low temperature region. It is this interaction that controls the equilibrium distribution of the resulting chemical . Zhao et al. Connolly 1997. in the ﬂow direction. Oliver 2001. 1999. even though the two ﬂuids are perfectly miscible. 1998a).6. Raffensperger and Garven 1995. This implies that for a permeable fault zone with a given chemical reaction. then that species is saturated within the ﬂuid and hence its concentration reaches a maximum value for a given mineralizing environment. With parallel ﬂows taken as an example. If an aqueous species. 1998a. It follows that in cooling hydrothermal systems the saturated aqueous species becomes oversaturated and is therefore precipitated in the porous rock. 1999d. heat transfer and the transport of reactive chemical species. 2003). ﬂuid ﬂow patterns can also inﬂuence the positions where different chemical species may meet and mix and hence where chemical reactions may take place (Yardley and Bottrell 1992. 2000b). reaches equilibrium. Peacock 1989.

5) in general cases. namely solute advection. the Damk¨ hler number is equal to one. for which the chemical reaction can chemical reach equilibrium beyond the given ladvection .4) and (6. This means that it is possible to investigate the relationship between the controlling processes associated with reactive chemical species transport so that the overall structure of their solutions can be understood. 2007a). Da.25) chemical where ladvection is the chemical equilibrium length due to solute advection for a given chemical chemical reaction. if not impossible.5)) into a dimensionless one so that the controlling processes associated with the reactive chemical species transport can be identiﬁed.2).3). the chemical equilibrium length o scale of the system can be expressed as follows: chemical ladvection = V . (6. Eq. (6. It is clear that if a chemical reaction rate is given.1). When the time scale for solute advection is equal to the time scale for chemical kinetics.24) where V is the characteristic Darcy velocity of the system.138 6 Fluid Mixing. For this purpose. for a given ladvection . we ﬁrst consider a dimensionless parameter known as the Damk¨ hler number. This understanding can be achieved by converting the reactive transport equation (i. solute diffusion/dispersion and chemical kinetics. (6. Below we refer to ladvection as the advection chemical equilibrium length. (6.e. to obtain analytical solutions for the coupled problem expressed by Eqs. (Steefel and o Lasaga 1990. (6. Voptimal . is as follows: . φk R (6. (6. Ormond and Ortoleva 2000) to express the relative time scales of solute advection and reaction kinetics: Da = φk R l = Time Scale for Solute Advection/ V Time Scale for Chemical Reaction.4. φ is the porosity of the porous medium. the corresponding optimal ﬂow rate. In this case.25). l is the characteristic length of the controlling process in the system. Thus. (6. there exists an optimal ﬂow rate such that the chemical reaction can reach equilibrium beyond the advection chemical chemical equilibrium length determined from Eq. Heat Transfer and Non-Equilibrium Redox Chemical Reactions product.2 Theoretical Analysis of Mineral Precipitation Patterns in a Focusing and Mixing System Involving Two Reactive Fluids Although it is difﬁcult. we need to consider the relationships between the time scales for these three controlling processes. k R is the controlling chemical reaction rate with units of [s-1 ]. it is possible to gain theoretical understanding through analytical solutions for the coupled problem in some limiting cases (Zhao et al. 6. that may play dominant roles in determining chemical reaction patterns. A conceptual model is presented here to investigate this interaction and the corresponding inﬂuence on chemical reaction patterns. Since there are three major controlling processes.

for a given ldiffusion . Since this dimensionless number expresses the ratio of the solute diffusion/dispersion time scale to the chemical kinetic time scale. k R is the reaction rate. solute diffusion/dispersion normal to the ﬂow direction plays a fundamental role in promoting chemical reactions between different reactive chemical species. Since the optimal ﬂow rate is also directly proportional to ladvection . If the ﬂow paths of two ﬂuids are parallel to each other in a ﬂuid mixing system. so that chemical chemical reaction for a given reaction rate can reach equilibrium beyond this large ladvection in the ﬂow direction. the corresponding optimal chemical reaction rate. This implies chemical that a large ladvection requires a relatively fast optimal ﬂow rate. a chemical reaction with a given reaction rate can reach equilibrium if the ﬂuid ﬂow rate is within the time scale of chemical equilibchemical rium. 139 (6. (6. Thus. the greater the optimal ﬂuid ﬂow rate. the chemical equilibrium length. In this chemical situation. Z= kRl2 = (Time Scale for Solute Dispersion/Diffusion)/ D Time Scale for Chemical Reaction. l is the characteristic length of the controlling process in the system. Since ladvection is directly proportional to the ﬂuid ﬂow rate. needs to be deﬁned to express the relative time scale between the solute diffusion/dispersion process and the chemical reaction process. for which the chemical reacchemical tion can reach equilibrium within the given ldiffusion . the greater the chemical ﬂuid ﬂow rate. In this case. the larger ladvection . k R .6. the physical meaning of the optimal ﬂow chemical rate is that for a given ladvection . (6. the larger chemical ladvection in the ﬂow direction. ldiffusion of the system can be expressed as follows: chemical ldiffusion = D .26) chemical It should be noted that both ladvection and the optimal ﬂow-rate have clear physichemical cal meanings: The physical meaning of ladvection is that for a given ﬂuid ﬂow rate.28) chemical where ldiffusion is the chemical equilibrium length due to solute diffusion/dispersion for a given chemical reaction. a chemical reaction with a given reaction rate can reach equilibrium once this distance is traversed by the ﬂuid in the ﬂow direction.28). within the time scale of chemical equichemical librium. This means that fast ﬂows require relatively long distances in the ﬂow direction. For a given solute diffusion/dispersion coefﬁcient.4 Applications of the Proposed Decoupling Procedure chemical Voptimal = φk R ladvection . kR (6. it is unity when the two time scales are equal. In contrast. there exists an optimal reaction rate such that the chemical reaction can reach equichemical chemical librium within ldiffusion determined from Eq. a second dimensionless parameter.27) where D is the solute diffusion/dispersion coefﬁcient. is as follows: optimal . Z. beyond which a chemical reaction with a given reaction rate can reach equilibrium. Notice that Z is independent of the ﬂuid velocity and so still has meaning for zero ﬂuid ﬂow.

reaching equilibrium over a large distance normal to the direction of ﬂuid ﬂow. the thickness of the fault zone is chosen as the characteristic length of the Z Number. the chemical reaction can reach equilibrium within this long chemical equilibrium length in the direction perpendicular to the parallel ﬂuid ﬂow. (6.140 6 Fluid Mixing.3 Chemical Reaction Patterns due to Mixing and Focusing of Two Reactive Fluids in Permeable Fault Zones The theoretical understanding of the interaction between solute advection. By contrast. the larger ldiffusion in the direction perpendicular to the ﬂuid ﬂow. 6. in principle. This implies that a large chemical equilibrium length due to solute diffusion/dispersion in the direction perpendicular to the parallel ﬂuid ﬂow requires a relatively slow optimal chemical reaction rate. the physical meaning of the optimal chemical reaction rate chemical is that for a given ldiffusion . with a given reaction rate. Since the optimal chemical reaction rate is chemical chemical inversely proportional to the square of ldiffusion . diffusion/dispersion and chemical kinetics. a chemical reaction with a given reaction rate can reach equilibrium if the chemical reaction proceeds at this chemical reaction rate within the time scale of chemical equilibrium. the larger ldiffusion in the direction perpendicular to the ﬂuid ﬂow.29) chemical Both ldiffusion and the optimal chemical reaction-rate also have clear physical chemical meanings: The physical meaning of ldiffusion is that for parallel ﬂow with a given solute diffusion/dispersion coefﬁcient. the greater the solute diffusion/dispersion coefﬁcient. This implies that a large solute diffusion/dispersion coefﬁcient can result in a chemical reaction. the smaller the optimal chemical reaction rate. The relationships between Da. 6. Heat Transfer and Non-Equilibrium Redox Chemical Reactions kR optimal = D chemical ldiffusion 2 . it is possible to deﬁne three different types of mineral precipitation patterns in Z-Da number space. Note that for a given fault zone involving pore-ﬂuid ﬂow focusing and mixing. a chemical reaction with a given reaction rate can only reach equilibrium within the distance diffused by the solute in the direction perpendicular to the ﬂuid ﬂow within the time scale of chemical equilibrium. solute diffusion/dispersion and chemical kinetics presented in the previous section is. . Since chemical ldiffusion is directly proportional to the square root of the solute diffusion/dispersion chemical coefﬁcient. For this purpose. Z and mineral precipitation types are shown in Fig. Da and Z. Chemical reaction patterns arising from ﬂow focusing and mixing of two reactive ﬂuids within permeable vertical fault zones are the subject of this section.7. useful for investigating chemical reaction patterns resulting from chemical equilibrium associated with ﬂuid ﬂow in all kinds of porous rocks.4. can express the interaction between solute advection. so that for a given solute diffusion/dispersion coefﬁcient. while the length of the fault zone is chosen as the characteristic length of the Da Number. The combined use of these two numbers.

0 Solute Advection Dominates 1. L fault and Wfault are the length and width of the fault zone.8. L mp represents the starting position of mineral precipitation within the fault zone.0 TYPE 2 CHEMICAL REACTION PATTERN 1. diffusion and chemical kinetics Da = Time Scale for Solute Advection/Time Scale for Chemical Reaction TYPE 3 CHEMICAL REACTION PATTERN TYPE 1 CHEMICAL REACTION PATTERN 1. 6.0 Z = Time scale for Solute Diffusion/Time Scale for Chemical Reaction (A) Relationship between solution advection.7 Three fundamental types of mineral precipitation patterns in the Z-Da Number space The geometry and problem description of two ﬂuids mixing and focusing within a permeable vertical fault zone are shown in Fig. In this ﬁgure.0 Z = Time scale for Solute Diffusion/Time Scale for Chemical Reaction (B) Three types of mineral precipitation patterns Fig. Wmp represents the thickness of mineral .4 Applications of the Proposed Decoupling Procedure 141 Da = Time Scale for Solute Advection/Time Scale for Chemical Reaction Solute Diffusion Dominates Chemical Reaction Dominates Chemical Reaction Dominates 1.6. 6.

(1999d). Two ﬂuids carrying two different solutes.30) L mp = ladvection = φk R Similarly. namely Fluid A and Fluid B. Heat Transfer and Non-Equilibrium Redox Chemical Reactions Wfault 0.5Wmp y Lfault 0 x Lmp Fluid A Fluid B Fig. since Fluid A ﬂows parallel to Fluid B. 6.142 6 Fluid Mixing. kR (6. the mixing of these two ﬂuids is due solely to solute diffusion/dispersion in the direction normal to the fault zone. The starting position of mineral precipitation within the fault zone can be expressed as follows: V chemical . the starting position of mineral precipitation is controlled by solute advection and is measured from the lower tip of the fault. the thickness of mineral precipitation is controlled solely by solute diffusion/dispersion. This means that for a given chemical reaction rate. are injected from the left and right sides of the fault axis and are focused into the fault zone according to the principles discussed by Phillips (1991) and Zhao et al. (6. On the other hand. Hence.8 The conceptual model of two ﬂuids mixing and focusing in a fault precipitation within the fault zone. Since mineral precipitation is dependent on both chemical equilibrium and the saturation concentration of a mineral. the thickness of mineral precipitation within the fault zone can be expressed as chemical Wmp = 2ldiffusion = 2 D .31) . it is assumed that the starting position of mineral precipitation is coincident with chemical equilibrium being attained in the following theoretical analysis.

the thickness of mineral precipitation within the fault zone is dependent on both the chemical reaction rate and the solute diffusion/dispersion coefﬁcient.e. (6. As indicated in Eq. (6. In the forthcoming section. To enable chemical equilibrium to be reached within the whole fault length. In this situation.4 Applications of the Proposed Decoupling Procedure 143 chemical Here the factor 2 arises because of the symmetry of geometry. the time required to reach chemical equilibrium within the whole length of a fault 5 km long is greater than the age of the Earth. ldiffusion is measured from the centre of the fault. Type 2: chemical equilibrium is attained at the upper tip of the fault.1). (6. which will enable chemical reactions to attain optimal chemical reaction rate k R equilibrium across the whole width of the fault zone. the two chemical reactants need to be present for 1016 s (i. we numerically examine more realistic and applicable situations.2). In this example.4) and (6. (6. 6.3.32) indicates that for a solute diffusion/dispersion coefﬁcient of 10–10 m2 /s and a vertical fault 1 km long. (6.4.30) yields a zero value of the starting position of mineral precipitation within the fault zone. depending on the actual residence time of the two chemical reactants within the system. the residence time for which the two chemical reactants should exist in the system is equal to the time interval. the optimal . (6. L fault .5) are difﬁcult but we can gain some insight into the types of chemical reaction patterns that are possible by considering some limiting cases below. tdiffuse .31). As mentioned above. about 7.e. there are three possible cases in which chemical equilibrium can be attained in the conceptual model shown in Fig. where the ﬂuid ﬂow is negligible. Substituting a zero velocity into Eq. the ﬂuid velocity is zero in the whole system.32) Consider a speciﬁc example: Eq. analytical solutions to Eqs. (6.8.1 Type 1: Chemical Equilibrium is Attained at the Lower Tip of the Fault In this limiting case. it is possible for chemical equilibrium to be attained throughout the whole length of the fault.3). 6.6. The three types of interest are: Type 1: chemical equilibrium is attained just at the lower tip of the fault. is as follows: . (6. As indicated earlier in this chapter.5 × 109 years) to enable the chemical reaction to reach equilibrium within the whole length of a 5 km long fault. about 3 × 108 years) at the bottom of the conceptual model in order for chemical equilibrium to be reached within the whole length of the fault. and Type 3: chemical equilibrium is attained somewhere between the lower tip and upper tip of the fault. for the solute to diffuse from the lower tip to the upper tip of the fault: tdiffuse = L2 fault D . Since the required duration for the existence of these two chemical reactants is quadratically proportional to the length of a fault. For a given solute diffusion/dispersion coefﬁcient. it is increased to about 25 × 1016 s (i.

144

6 Fluid Mixing, Heat Transfer and Non-Equilibrium Redox Chemical Reactions

kR

optimal

=

4D . 2 Wfault

(6.33)

6.4.3.2 Type 2: Chemical Equilibrium is Attained at the Upper Tip of the Fault This is another limiting case, in which the ﬂuid rate (i.e. velocity) has reached a critical value in the fault zone. The fundamental characteristic of this limiting case is that if the ﬂuid rate is equal to or greater than this critical value, the starting point of chemical equilibrium is at or beyond the upper tip of the fault, indicating that the chemical equilibrium of the chemical product and therefore mineral precipitation cannot be attained within the fault zone. For a given chemical reaction rate, this critical ﬂow rate within the fault zone, Vcritical , can be expressed as follows: Vcritical = φk R L fault . (6.34)

Equation (6.34) is useful for estimating the critical ﬂow rate for a vertical fault. For example, if the porosity of a vertical fault is 0.3 and the chemical reaction rate is 10–11 (1/s), then the corresponding critical ﬂow rate is 3 × 10–9 m/s for a vertical fault 1 km long. If both the fault porosity and the chemical reaction rate remain unchanged, the corresponding critical ﬂow rate of the fault is increased to about 15 × 10–9 m/s for a vertical fault 5 km long. Suppose the ﬂow-focusing factor (Phillips, 1991) of the vertical fault is of the order of 15, the corresponding critical ﬂow rate within the surrounding rock for a vertical fault is 10–9 m/s. This critical ﬂow rate can be easily exceeded when the ﬂuid within the surrounding rock of the fault zone is under a lithostatic pressure gradient. If the permeability of the surrounding rock is 10–16 m2 and the dynamic viscosity of the ﬂuid is 10–3 N s/m2 , then the ﬂow rate induced by the lithostatic pressure gradient within the surrounding rock is 1.7 × 10–9 m/s. This implies that the ﬂow rate induced by a lithostatic pressure gradient within the surrounding rock is too high to enable minerals to be precipitated within the vertical fault zone. 6.4.3.3 Type 3: Chemical Equilibrium is Attained Somewhere between the Lower Tip and Upper Tip of the Fault In this case, the equilibrium of the reaction product is attained within the fault zone. The starting position of the chemical equilibrium is measured from the lower tip of the fault and can be expressed by the following equation: L mp = αmp L fault , (6.35)

where L mp is the starting position of chemical equilibrium within the fault zone in the ﬂow direction; αmp is a coefﬁcient to express the relative relationship between the length of the fault and the starting position of chemical equilibrium within the fault zone; that is αmp = L mp L fault . It is clear that in this particular case, the value

6.4

Applications of the Proposed Decoupling Procedure

145

of αmp is greater than zero but less than one. A value of αmp equal to zero represents the ﬁrst limiting case (Type 1), while a value of αmp equal to one represents the second limiting case (Type 2), as discussed above. Similarly, the width of chemical equilibrium within the fault can be expressed as follows: Wmp = βmp Wfault , (6.36)

where Wmp is the width of chemical equilibrium within the fault zone; Wfault is the width of the fault; βmp is a coefﬁcient expressing the ratio between the width of the fault and the width of chemical equilibrium within the fault zone; that is, βmp = Wmp Wfault . In this case, the value of βmp is greater than zero but less than one. A value of βmp equal to zero represents the ﬁrst limiting case (Type 2), whilst a value of βmp equal to one represents another limiting case (i.e. the limiting case of Type 3), in which case the width of chemical equilibrium within the fault zone is just equal to that of the fault. If both the solute diffusion/dispersion coefﬁcient and the width of chemical equilibrium within the fault zone are known, the corresponding optimal chemical reaction rate in this case can be estimated from the following formula: kR

optimal

=

4D 4D = 2 2 . 2 Wmp βmp Wfault

(6.37)

**Thus, the corresponding optimal ﬂow rate in this particular case is: Voptimal = φk R
**

optimal

L mp =

4φ Dαmp L fault . 2 2 βmp Wfault

(6.38)

Equation (6.38) indicates that for a particular pattern of mineral precipitation within the fault zone, there exists an optimal ﬂow rate such that chemical equilibrium is attained within the region associated with this particular mineral precipitation pattern.

**6.4.4 Numerical Illustration of Three Types of Chemical Reaction Patterns Associated with Permeable Fault Zones
**

The theoretical analysis carried out in the previous section predicts that there exist three fundamental types of chemical reaction patterns associated with permeable vertical fault zones due to two ﬂuids mixing and focusing. In order to illustrate these different types of chemical reaction patterns, the ﬁnite element method is used to solve the coupled problem numerically involving ﬂuid mixing, heat transfer and chemical reactions expressed by Eqs. (6.1), (6.2), (6.3), (6.4) and (6.5). In theory, it is possible to predict exactly the three fundamental types of chemical reaction patterns. However, in numerical practice, it is difﬁcult to control the ﬂow rate of

146

6 Fluid Mixing, Heat Transfer and Non-Equilibrium Redox Chemical Reactions

a speciﬁc value due to numerical round off and cutoff errors. For this reason, the numerical simulated mineral precipitation patterns here are approximate representations of the three fundamental types of mineral precipitation patterns. The computational model (shown in Fig. 6.1) and related parameters for the veriﬁcation example in the previous section (i.e. Sect. 6.3) are used to illustrate three different types of mineral precipitation in a focusing and mixing system of two reactive ﬂuids. Two different ﬂuid pressure gradients are considered to control the ﬂow rate (i.e. vertical ﬂuid velocity) within the surrounding rock during the numerical simulation. To simulate the fast ﬂow rate associated with type 2 (as stated in the previous section), the ﬂuid pressure gradient is assumed to be a lithostatic pressure gradient, which results in a vertical ﬂow rate of 1.7 × 10–9 m/s within the surrounding rock. This gradient is used because it is considered to be near the highest reasonable ﬂuid pressure gradient likely to be encountered in nature. On the other hand, in order to simulate the slow ﬂow rate associated with types 1 and 3 (see the previous section), the excess ﬂuid pressure gradient is assumed to be one percent of a lithostatic pressure gradient minus a hydrostatic one, which results in a vertical ﬂow rate of 1.7 × 10–11 m/s within the surrounding rock. It is known that for ﬂuid pressure gradient dominated ﬂow, the ﬂow pattern around a permeable fault zone is dependent only on the contrast in permeability between the fault and the surrounding rocks, the geometry of the fault zone and the inﬂow direction relative to the axis of the fault (Phillips 1991, Zhao et al. 1999d) and is independent of the pressure gradient along the fault. This means that although the ﬂuid pressure gradient is different in the three simulations, the streamline pattern within and around the fault zone must be identical for all three cases, as demonstrated by the related numerical result shown in Fig. 6.2. In all cases, ﬂuid ﬂow converges into the fault zone at the lower end and diverges out of the fault zone at the upper end (see also Phillips 1991, Zhao et al. 1999d). 6.4.4.1 The First Type of Chemical Reaction Pattern The ﬁrst type of chemical reaction pattern results from an approximate representation of Type 1 in Sect. 6.4.3; the background ﬂow velocity within the surrounding rock is not exactly equal to zero in the numerical simulation but instead is one percent of the lithostatic pressure gradient minus the hydrostatic gradient. Here two controlling chemical reaction rates, namely 10–7 (1/s) and 10–11 (1/s) are considered to investigate the effect on chemical reaction patterns. Due to ﬂow focusing, the maximum vertical ﬂow velocity within the fault zone is about 3.16 × 10–10 m/s. Figure 6.9 shows the concentration distributions of the two chemical reactants and the corresponding chemical product at two time instants of t = 500,000 and t = 800,000 years. The distribution of the concentration of the chemical product comprises a lenticular shape within the fault zone. This coincides with what is expected from the previous theoretical analysis. For a fault zone of width 250 m, the optimal chemical reaction rate calculated from Eq. (6.33) is 4.8 × 10–15 (1/s) so that chemical equilibrium can be attained within the whole width of the fault zone. Since the optimal chemical reaction rate is much smaller than the two chemical

6.4

Applications of the Proposed Decoupling Procedure

147

Reactant A (t = 500,000 years)

Reactant A (t = 800,000 years)

Reactant B (t = 500,000 years)

Reactant B (t = 800,000 years)

Product (kR = 10–7(1/s), t = 500,000 years)

Product (kR = 10–7(1/s), t = 800,000 years)

Product (kR = 10–11(1/s), t = 500,000 years)

Product (kR = 10–11(1/s), t = 800,000 years)

Fig. 6.9 Concentration distributions of the chemical reactants and product at two different time instants (Type 1)

10 shows the concentration distributions for the two chemical reactants and the corresponding chemical product at two time instants of t = 5000 and t = 8000 years.g. the theoretical chemical equilibrium length calculated from Eq. are considered to investigate the effects of different chemical kinetics on chemical reaction patterns. There is a strong interaction between solute advection. chemical equilibrium cannot be reached within the .48 m.4.148 6 Fluid Mixing. mineral precipitation comprises a thin lenticular shape within the fault zone and starting at the lower tip of the fault. From a chemical kinetics point of view.02 × 10–8 m/s. Figure 6. In this example. This implies that the controlling chemical reaction rate is too fast to allow both the reactants to diffuse across their common boundary so that ﬂuid mixing cannot effectively take place within the fault zone. the maximum concentration distribution of the chemical product comprises a very thin membrane. The theoretical estimate of the thickness of chemical equilibrium within the fault zone is 10. Thus the chemical equilibrium length due to solute advection is greater than the length of the fault itself. Heat Transfer and Non-Equilibrium Redox Chemical Reactions reaction rates used in the simulation. Due to ﬂuid ﬂow focusing. the ﬂuid ﬂow decreases and. (6.11 m for a controlling chemical reaction rate of 10–7 (1/s). In the case of a non-equilibrium chemical reaction characterized by a slow chemical reaction rate (e. k R = ∞).25) is 8630 m in the direction of the fault axis. namely k R = ∞ and k R = 10−11 (1/s). diffusion/dispersion and chemical reaction rate (i. Although both reactants are transported into the fault zone.4. k R = ∞ represents an equilibrium chemical reaction. the thickness of the chemical equilibrium within the fault zone is much smaller than the width of the fault zone itself.e. that is. Because the chemical equilibrium length due to the solute advection is greater than the total length of the fault zone plus its exit region (i. Due to ﬂow focusing. the maximum vertical ﬂow velocity within the fault zone is about 3.e. around the exit region of the fault zone. k R = 10−11 (1/s)).28) is 5. the mixing of the two ﬂuids carrying them is controlled by solute diffusion and dispersion. Since the chemical reaction rate is inﬁnite. so that the chemical product distribution within the fault zone is controlled by solute advection. the corresponding chemical equilibrium length due to solute diffusion/dispersion is identical to zero normal to the fault zone. Two controlling chemical reaction rates. 6. diverges so that a high concentration of the chemical product is produced around the exit region of the fault zone. However. Both chemical reactants are transported into the computational domain from the left half and right half of the base of the model. both chemical reactants are transported much faster within the fault zone than in the surrounding rock.95 m for a controlling chemical reaction rate of 10–11 (1/s). more importantly.2 The Second Type of Chemical Reaction Pattern The second type of chemical reaction pattern is deﬁned as an approximate representation of Type 2 in the previous section. the background ﬂuid pressure gradient within the surrounding rock is set equal to a lithostatic pressure gradient. Hence. 5000 m plus 2500 m). while the theoretical chemical equilibrium length calculated from Equation (6. and 0. This is the fundamental characteristic of the ﬁrst type of chemical reaction pattern.

000 years) Reactant B (t = 500. t = 800. t = 500.000 years) Fig.000 years) Product (kR = ∞.6. t = 800. t = 500.10 Concentration distributions of the chemical reactants and product at two different time instants (Type 2) .000 years) Product (kR = 10–11(1/s).000 years) Reactant A (t = 800. 6.000 years) Product (kR = 10–11(1/s).4 Applications of the Proposed Decoupling Procedure 149 Reactant A (t = 500.000 years) Product (kR = ∞.000 years) Reactant B (t = 800.

Heat Transfer and Non-Equilibrium Redox Chemical Reactions height of the model. which results in a considerable equilibrium thickness in the direction normal to the fault zone. Thus. The fundamental characteristic of this type of chemical reaction pattern is that chemical equilibrium.3 The Third Type of Chemical Reaction Pattern The third type of chemical reaction pattern is a representation of Type 3 in the previous section. Due to the low ﬂuid velocity.000 years. 6.11 shows concentration distributions of the two chemical reactants and the corresponding chemical product at two time instants of t = 500. This is the fundamental characteristic of the second type of chemical reaction pattern considered in this investigation. a chemical equilibrium zone is also generated in the ﬂow convergent region just in front of the fault zone.4. If the thickness of chemical equilibrium in the lateral direction of the fault zone is 80 m. (6. indicating that two ﬂuids cannot mix to produce an extensive mixing region due to the fast advection of the ﬂuids within the fault. Due to ﬂow focusing. Since the optimal reaction rate is directly proportional to the solute diffusion/dispersion coefﬁcient. This phenomenon results from the distribution of ﬂuid velocity vectors arising from ﬂuid ﬂow focusing just outside and within the fault zone.3. implying that mineral precipitation cannot take place within a permeable vertical fault zone for the chemical reaction considered here. chemical equilibrium cannot be attained within the fault. then it is possible to estimate the width of the potential mineral precipitation pattern within a permeable vertical fault zone. if the ﬂow rate and reaction rate are known.68 × 10–10 m/s in the numerical simulation. Figure 6. for the two end-members of chemical reaction rates. . the maximum vertical ﬂow velocity within the fault zone is about 4. In order to enable the starting position of chemical equilibrium to be close to the lower tip of the vertical fault. This agrees well with what is expected from the previous theoretical analysis given in Sect. 6. the solute diffusion/dispersion coefﬁcient is assumed to be 3 × 10–8 m2 /s in the numerical simulation. The geological implication of the third type of chemical reaction pattern is that if a mineral precipitation pattern of a certain thickness and length within a permeable vertical fault zone is observed. It is noted that this chemical equilibrium zone is almost separated from the chemical equilibrium zone within the fault. the optimal chemical reaction rate is about 1.150 6 Fluid Mixing.4. then it is possible to estimate both the optimal ﬂow rate and the optimal reaction rate during the formation of this precipitation pattern. so that the total CPU time in the simulation can be signiﬁcantly reduced.4.9 × 10–11 (1/s). can be achieved within the fault zone.000 and t = 800.37). it is desirable to select the value of a solute diffusion/dispersion coefﬁcient as large as possible. On the other hand. as can be calculated from Eq. The maximum concentration distribution of the chemical product generates considerable thickness within the fault zone. the background ﬂuid pressure gradient within the surrounding rock is one percent of the lithostatic pressure gradient minus the hydrostatic gradient. For this reason. namely a very fast equilibrium reaction and a slow reaction with a controlling chemical reaction rate of 10–11 (1/s).

t = 800.6. 6.4 Applications of the Proposed Decoupling Procedure 151 Reactant A (t = 500.000 years) Reactant B (t = 500. t = 500.000 years) Reactant B (t = 800.000 years) Reactant A (t = 800.11 Concentration distributions of the chemical reactants and product at two different time instants (Type 3) .000 years) Fig.000 years) Product (kR = 2 × 10–11(1/s).000 years) Product (kR = 2 × 10–11(1/s).

to simulate the effects of the magma ascent processes.Chapter 7 An Equivalent Source Algorithm for Simulating Thermal and Chemical Effects of Intruded Magma Solidiﬁcation Problems Consideration of the effects of magma ascending and solidiﬁcation is important to the further understanding of ore body formation and mineralization in the crust of the Earth. Rubin 1995. Although various possible fundamental mechanisms of magma ascending in the crust are proposed (Johnson and Pollard. 2001). can be used to solve such large scale geological problems. In order to overcome these difﬁculties. 2008g). we can develop some useful algorithms.. in combination with continuum-mechanics-based numerical methods. Fundamentals of Computational Geoscience. even if not impossible.1007/978-3-540-89743-9 7. 2007b. 1973. DOI 10. although it is impossible to use the continuum-mechanics-based numerical methods to simulate directly the magma ascent processes. particle-based numerical simulations have been developed rapidly in recent years (Zhao et al. Rubin 1995. due to the different time and length scales involved in ore body formation and mineralization problems. Bons et al. it is also very difﬁcult. For example. and so forth. c. 2006f. Lecture Notes in Earth Sciences 122. In terms of the magma intrusion mechanism. the solidiﬁcation of the ascending magma due to heat losses to the surrounding rocks. even though it is based on simple conceptual models (Johnson and Pollard. However. d. Marsh 1982. As a long-term development strategy. continuum-mechanics-based numerical methods have encountered serious difﬁculties in simulating the random generation and propagation of hydro-fractured cracks. Zhao et al. C Springer-Verlag Berlin Heidelberg 2009 153 . we need to develop multiple time and length scale modelling techniques and algorithms so that particle simulation methods. Marsh 1982. Although the previous theoretical work needs to be quantitatively reﬁned. it can be used to estimate the total volume of the intruded C. to use the present particle-based numerical methods to simulate all the important processes associated with ore body formation and mineralization problems in the crust of the Earth. Weinberg 1996. the magma ﬂow within these cracks. the development of numerical algorithms for simulating the proposed magma ascending mechanisms is still under-developed. 2001). Lister and Kerr 1991. Thus. As an expedient strategy. a large amount of theoretical work has been carried out previously. the main motivation of carrying out this study is to develop a useful algorithm to consider the dynamic consequences involved in magma ascent processes using continuum-mechanics-based numerical methods. 1973. combined with newly-developed techniques and algorithms. Lister and Kerr 1991. Weinberg 1996. Bons et al.

the moving boundary problem (Crank 1984. With magma intrusion into the Earth’s crust taken as an example. To the best knowledge of the authors. Given the total volume of the intruded magma. the efﬁciency of the ﬁnite element method can be much improved. the particle numerical method may be useful to simulate the magma intrusion process alone so that the volume of the intruded magma can also be estimated. However. the time scale of the magma intrusion process. On the other hand. which includes both the release of volatile ﬂuids from the magma and chemical reaction processes within the crust due to the release of the volatile ﬂuids. This will allow the detailed mechanisms associated with each of the two processes to be modelled using totally different methodologies. Once the time history of the heat release rate of the intruded magma during solidiﬁcation is obtained. This may be considered as one of the major advantages of the proposed equivalent algorithm in dealing with the thermal effects of magma intrusion problems in the crust of the Earth. Using the proposed equivalent algorithm. then continuum-mechanics-based numerical methods can be used to simulate heat transfer and mass (i. For example. chemical species) transport within the Earth’s crust.154 7 Simulating Thermal and Chemical Effects of Intruded Magma Solidiﬁcation Problems magma. to simulate its thermal effects by considering the heat release during the solidiﬁcation of the intruded magma. if the thermal and chemical effects/consequences of the intruded magma are of particular interest. This means that the whole magma intrusion problem is. it is possible to simulate these two different processes using different analytical models. such as the ﬁnite element method and ﬁnite difference method. which includes both the creation of the magma chamber and ascent process for the intruded magma. in essence. is much smaller than that of the magma solidiﬁcation process. due to the signiﬁcant time and length scale differences between the magma intrusion and solidiﬁcation processes. the ﬁnite element method can be used to simulate the thermal effects of the intruded magma in the crust of the Earth. Nevertheless. these methods have not been used to simulate the release of volatile ﬂuids from the magma and the chemical reaction processes within the crust due to the release of the volatile ﬂuids. the volume of the intruded magma is usually much smaller than that of the Earth’s crust of interest. As a direct result. we can use continuum-mechanics-based numerical methods. This means that we need to develop an equivalent algorithm to transform the original magma intrusion problem into a heat transfer problem with the internal heat generation of the intruded magma. then particlebased numerical methods can be used to simulate the initiation and propagation of random cracks during the ascending of the intruded magma. if the magma intrusion process itself is of particular interest. Alexiades and Solomon 1993) associated with the original problem during magma solidiﬁcation can be avoided. Clearly. If an analytical estimation of the volume of the intruded magma is not available for most complicated geological situations. Geological problems may involve different time and length scales in the descriptions of their different physical and chemical processes. Since both the release of volatile ﬂuids from the magma and the transport of the released volatile ﬂuids may have signiﬁcant effects on ore body formation and mineralization in the upper crust . a problem of multiple time and length scales. the key issue associated with the developed equivalent algorithm is to determine the heat release rate of the intruded magma during its solidiﬁcation.e.

it is useful to conduct numerical simulation of this kind of problem progressively from a simplistic stage into more complicated stages.1 An Equivalent Source Algorithm for Simulating Thermal and Chemical Effects 155 of the Earth. we plan to solve the magma intrusion problem using the following three level models. it is possible to integrate them to simulate the whole process of the magma intrusion problem. For a particular kind of intruded magma. 2003e). instead of granite and gabbro.1 An Equivalent Source Algorithm for Simulating Thermal and Chemical Effects of Intruded Magma Solidiﬁcation Problems Owing to the importance for many scientiﬁc issues and technical applications. Therefore. A large amount of geochemical research has indicated that intruded magma in the Earth’s crust has different chemical compositions and. For the second level model. heat transfer and ore forming patterns within the upper crust of the Earth. therefore. we will develop some useful and efﬁcient computer algorithms to simulate the magma solidiﬁcation problem associated with the second level model. we consider the thermal and chemical effects of the post-intrusion but pre-solidiﬁcation magma on pore-ﬂuid ﬂow. given a particular kind of intruded magma. heat transfer and ore forming patterns within the upper crust of the Earth. the abovementioned information is obtainable from geochemical and isotopic analyses. it is possible to determine the contents of water and other volatile ﬂuids. theoretical and numerical analyses of heat transfer with phase change have been carried out for more than half a century. It is well known that the solidus of the intruded magma is mainly dependent on the magma composition. different types of rocks can be formed during the solidiﬁcation and crystallization of the intruded magma. are formed if the solidiﬁcation conditions of the felsic and maﬁc magmas are changed from intrusive into extrusive.7. Rhyolite and basalt. For the ﬁrst level model. we will consider the intrusion process itself at a much smaller scale using particle mechanics-based computer algorithms. In this model. For the third level model. which should be released when the intruded magma becomes solidiﬁed. we have completed some work for the ﬁrst level model (Zhao et al. we must develop some useful and efﬁcient computer algorithms to simulate the magma solidiﬁcation problem. This research methodology is rational because of the multiple time and length scales of the problem itself. This implies that the volatile ﬂuids released from the intruded magma can be quantitatively simulated in the numerical analysis. So far. Both temperature and pressure conditions during solidiﬁcation of the magma can also affect the resulting rock types dramatically. In this study. the contents of water and other volatile ﬂuids. Thus. a numerical method needs to be developed to simulate the chemical consequences of the magma solidiﬁcation in porous rocks. Once efﬁcient numerical algorithms for dealing with above three level models are developed. which is another important issue to be addressed in this study. Due to the complex nature of the magma intrusion problem within the crust of the Earth. 7. There is an extensive literature which reports and . we primarily consider the effects of the post-solidiﬁcation magma on pore-ﬂuid ﬂow.

Weinberg 1996) is on the scale of meters and tens of meters. Since heat release during solidiﬁcation of the intruded magma can be represented by a physically equivalent heat source. the detailed solidiﬁcation process of the intruded magma might not be important. but the characteristic dimension of the intruded magma. This is the basic idea behind the proposed equivalent algorithm for simulating the thermal effect of the intruded magma solidiﬁcation in this study. The problem associated with numerical modelling of the thermal effect of solidiﬁcation of intruded magma is that the characteristic dimension of the whole geological system under consideration is on the scale of tens and hundreds of kilometers. 2003). Mazouchi and Homsy 2000).156 7 Simulating Thermal and Chemical Effects of Intruded Magma Solidiﬁcation Problems reviews the development of this subject (Carslaw and Jaeger 1959. it is possible to transform the original heat transfer problem with phase change due to solidiﬁcation of the intruded magma into a physically equivalent heat transfer problem without phase change but with the equivalent heat source. it is ideal to describe the governing equation of each chemical component separately. even though not identical. Sethian 1999). and Tornberg and Engquist (2003b) have investigated the convergence and accuracy of the numerical method. Gawin et al. such as the immersed boundary method (Beyer and LeVeque 1992). Zhao and Heinrich 2002). Schreﬂer et al. the level-set method (Osher and Sethian 1998. 1999. so that we can use a ﬁxed ﬁnite element mesh to consider the thermal effect of the intruded magma by solving the physically equivalent heat transfer problem with an equivalent heat source. 2002. Recently. if the diffusivity of each chemical component is assumed to be identical. can be used to describe the governing equation of the released volatile . it will be extended to the solution of the phase change problem associated with intruded magma solidiﬁcation in geological systems. but the thermal effect caused by the heat release during the solidiﬁcation of the intruded magma is important. such as a sill or dike (Johnson and Pollard 1973. In terms of solving partial differential equations with delta function source terms numerically. Rubin 1995. at least from the ore body formation and mineralization point of view. which is true for most geological systems (Johnson and Pollard 1973. we can remove the moving boundary problem associated with the detailed solidiﬁcation process of the intruded magma. the ﬁxed grid approach has been successfully used to solve the phase change problem associated with moisture transport in high-temperature concrete materials (Schreﬂer 2004. Alexiades and Solomon 1993. In this chapter. we will present the proposed equivalent algorithm for simulating the chemical effect of solidiﬁcation of the intruded magma in geological systems. Weinberg 1996). The proposed algorithm is only valid when the characteristic length scale of the system is much larger than that of the intruded magma. which is the summation of the concentration of all the chemical components. However. approaches. then the concept of the total concentration. Lister and Kerr 1991. In what follows. The above physical understanding means that from the numerical modelling point of view. Since the volatile ﬂuids released during solidiﬁcation of the intruded magma may have many different chemical components/species. the segment projection method (Tornberg and Engquist 2003a) and so forth. Lister and Kerr 1991. Rubin 1995. have been used to tackle other different problems with moving interfaces or fronts (Smooke et al. The similar. Walden (1999). As a result. Crank 1984.

t) (x.3) where TR and TM are the temperature of the rock and intruded magma. ρ R . Barns 1997) have demonstrated that if the constraints of the solution model for the NaAlSi3 O8 −H2O system are imposed. H2O is the most abundant magmatic volatile and CO2 is the second most abundant magmatic volatile in the intruded magma. ρ M .7. HCl.1) (ρ M c pM ) ∂C T =D ∂t ∂ TM = λM ∂t (x. D is the diffusivity of the released volatile ﬂuids. In this way. δ is the delta function of values of unity and zero. H2O solubilities in the igneous-rock melts are essentially identical to those in NaAlSi3 O8 melts. speciﬁc heat and thermal conductivity of the intruded magma. (7. It is noted that. Zhao and Heinrich 2002) can be described. their effects on the heat transfer process can be neglected in the analysis of the intruded magma solidiﬁcation problem. Because the volumetric amount of the released volatile ﬂuids is relatively small to the intruded magma. as (ρ R c p R ) ∂ TR = λR ∂t ∂ 2 TR ∂ 2 TR + ∂x2 ∂ y2 ∂ 2 TM ∂ 2 TM + ∂x2 ∂ y2 (x.1 An Equivalent Source Algorithm for Simulating Thermal and Chemical Effects 157 ﬂuids. speciﬁc heat and thermal conductivity of the rock. y) ∈ VR . Thus. These extensive studies (Burnham 1979. for a two-dimensional problem. (7. x I and y I are the x and y coordinate components of the interface position. y − y I )Q(x. HF. Therefore. computer efforts can be reduced signiﬁcantly in dealing with the intruded magma solidiﬁcation problems. Thus. c p R and λ R are the density. Crank 1984. Q is the mass source of the released volatile ﬂuids during the intruded magma solidiﬁcation. Since the original magma solidiﬁcation problem belongs to a moving interface problem.2) ∂ 2CT ∂ 2CT + 2 ∂x ∂ y2 + δ(x − x I . For this reason. y) ∈ VM . H2 S. Alexiades and Solomon 1993. y) ∈ VR + VM . we can use one governing equation to describe variations of the total concentration of the released volatile ﬂuids in the system. SO2 and other substances (Burnham 1979. the solubility of H2O in silicate melts has been investigated for many years. C T is the total concentration of the released volatile ﬂuids. although the released volatile ﬂuids during the intruded magma solidiﬁcation are comprised of H2 O. y. VR and VM are the spaces occupied by the rock and intruded magma. c pM and λ M are the density. the solubility of H2O in the NaAlSi3 O8 melt can be used to approximately determine the mass source of the released volatile ﬂuids during the intruded magma solidiﬁcation. CO2 . the governing equations of the original heat transfer and mass transport problem considering the phase change during the intruded magma solidiﬁcation (Carslaw and Jaeger 1959. Barns 1997). This allows us to assume that thermal equilibrium between the released volatile ﬂuids and the intruded magma/rock has been achieved. the temperature and heat ﬂux continuity conditions on the interface between the rock and intruded magma are as follows: . (7.

7. where x I and y I are the x and y coordinate components of the interface position. the magma solidiﬁcation thickness is expressed by ΔL M . it is necessary to introduce some related concepts. Zhao and Heinrich 2002). (7. it is not necessary to repeat them here. Except for the boundary conditions on the interface between the rock and intruded magma. n x and n y are the x and y components of the unit normal to the interface between the rock and intruded magma. L is the latent heat of fusion of the intruded magma. Crank 1984. Neumann or mixed type (Carslaw and Jaeger 1959. Alexiades and Solomon 1993. Since heat and volatile ﬂuids are only released during magma solidiﬁcation.4) = ρ M [L + c pM (TI M ∂ TM ∂ TM + ny ∂x ∂y ∂ xI ∂ yI − Tm )] n x + ny ∂t ∂t (7. y I ) ∈ Γ R M .1 Basic concepts related to the magma solidiﬁcation problem . If the length of the interface front is ΔL R . As shown in Fig. Δt M . we may consider the released heat and volatile ﬂuids either as surface heat and mass sources in a two-dimensional problem or as volumetric heat and mass sources in a three-dimensional one. the boundary conditions on the other boundaries of the rock domain and intruded magma domain can be either of the Dirichlet.5) (x I . As a result. TIM is the temperature of the intruded magma. Γ R M is the interface between the rock and intruded magma. In order to develop an efﬁcient and effective numerical algorithm for dealing with solidiﬁcation of intruded magma in sills and dikes. Tm is the solidiﬁcation temperature of the intruded magma. Since the boundary conditions on the other boundaries of the rock domain and intruded magma domain are trivial. then the magma solidiﬁcation area is the product of ΔL M and ΔL R during the time period Δt M .1. y I ) ∈ Γ R M .158 7 Simulating Thermal and Chemical Effects of Intruded Magma Solidiﬁcation Problems T R = TM λR nx ∂ TR ∂ TR + ny ∂x ∂y − λM n x (x I . the interface moves to position 2 due to magma solidiﬁcation during a time period. 7. This means that it is physically possible to use Position 1 Rock Position 2 Magma ΔLR Interface movement direction ΔLM Fig. supposing the initial position of the interface between the magma and rock is at position 1.

(7. y. In this regard. the solidiﬁcation time period can vary signiﬁcantly as the magma solidiﬁcation proceeds. Since the magma is fully solidiﬁed within the region of the solidiﬁcation thickness ΔL M . y) ∈ VR −VM .6) (ρ R c p R ) ∂C T =D ∂t ∂ 2CT ∂ 2CT + ∂x2 ∂ y2 + δ(x − x I . (7. y−y I ) f (x. Based on the above considerations. k is the time-step index of integration in the ﬁnite element analysis.3) should have a value of unity. This is the reason why the delta function only has values of unity and zero in the present numerical algorithm.1 An Equivalent Source Algorithm for Simulating Thermal and Chemical Effects 159 a ﬁxed ﬁnite element mesh to consider the released heat and volatile ﬂuids during the magma solidiﬁcation process. in which the ﬁnite element mesh is ﬁxed but the integration time-step being equivalent to the solidiﬁcation time period must be variable. y − y I )Q(x. the mole fraction of H2O in the NaAlSi3 O8 melt is expressed as follows: m Xw = 1 mf kw m (X w ≤ 0. for a given magma solidiﬁcation thickness ΔL M . which can be easily simulated by the ﬁxed mesh.5) can be represented by the following equations: ∂ TR = λR ∂t ∂ 2 TR ∂ 2 TR + +δ(x−x I . (7. Generally. Eqs. t) ∂x2 ∂ y2 (x. The key issue associated with the ﬁnite element analysis using a ﬁxed mesh is that.2).7) where δ is the delta function with values equal to unity and zero. (7.3). y) ∈ VR − VM .7. we need to determine the time period Δt M so that the interface between the magma and rock just moves a distance being equal to ΔL M in the interface movement direction.9) . y. ρ M L + c pM (TI M − Tm ) f (x. y.5). This requires us to develop a numerical algorithm. Barns 1997). f (x.8) where ΔL Mk is the solidiﬁcation thickness of the intruded magma during a time period Δt Mk . The mass source of the released volatile ﬂuids in Eq. meaning that there is no magma solidiﬁcation taking place in other regions.7) during the intruded magma solidiﬁcation can be determined using the solubility of H2O in the NaAlSi3 O8 melt (Burnham 1979. (7. the delta function should have a value of zero. t) (x. (7.4) and (7. t) = ΔL Mk ∂ xI ∂ yI + ny ∂t ∂t nx . (7. the delta function used in Eq. (7.1). t) is the physically equivalent heat source due to the solidiﬁcation of the intruded magma. Except in this solidiﬁcation region. (7. y. for the same magma solidiﬁcation thickness ΔL M . (7.

12). Eqs. is determined either analytically or experimentally. Ww and ρw are the molar mass and denm m sity of the volatile ﬂuids in the magma.8) can be rewritten as ρM L n x f (x.012 × 10−4 (ln P)4 + 4.11) where P is the pressure of the intruded magma. the physically equivalent heat source due to the solidiﬁcation can . the mass source of the volatile ﬂuids released during solidiﬁcation from the intruded magma can be expressed as Q(x. y. t).6) and (7.831 × 10−8 T 2 − 4.5) (7. Using the concept of molar mass.10) m where X w is the mole fraction of H2O in the NaAlSi3 O8 melt.7) can be directly solved using the conventional ﬁnite element method (Zienkiewicz 1977). respectively.882 × 10−5 T + 4. In the case of the intruded magma temperature being equal to the solidiﬁcation magma temperature. f (x.51 × 10−4 T − 1.754 × 10−3 T − 1. m ln kw f = 5 + (ln P)(4. t) = ∂x I ∂y + ny I ∂t ∂t ΔL Mk . (7.481 × 10−8 T 2 − 1. the mass source of the released volatile ﬂuids has units of the density of the albite (NaAlSi3 O8 ) melt divided by time.621 × 10−6 T 2 . (7. For dike-like and sill-like intruded magmas.13) 7. t) = m m X w Ww m m (1 − X w )Walbite + m ρalbite m m X w Ww m ρw .12) where Δt Mk is the time period required to complete the magma solidiﬁcation within m m a given solidiﬁcation thickness ΔL Mk . kw is the equilibrium constant for H2O in melts of feldspar composition. It is noted that using the deﬁnition in Eq. Eq.160 7 Simulating Thermal and Chemical Effects of Intruded Magma Solidiﬁcation Problems m X w = 0. P and T are in bars and Kelvin.25 − T ln 4 mf kw m (X w > 0. Δt Mk (7. Walbite and ρalbite are the molecular mass and density of the albite (NaAlSi3 O8 ) melt.137) mf + (ln P)2 (1.5 + 1 2667 6.8 × 10−3 (ln P)3 − 5. T is in Kelvin.2 Implementation of the Equivalent Source Algorithm in the Finite Element Analysis with Fixed Meshes If the physically equivalent heat source term. (7. (7. y.656 × 10−2 ) + 7. y. (7.

7. while the characteristic length of the intruded magma is L M 2. All the external boundaries of the sample are insulated and the temperatures of the rock and intruded magma are TR and Tm . Lister and Kerr 1991.7. which is true for sills and dikes in geology (Johnson and Pollard. If the characteristic length of the intruded magma is divided into K equal parts. therefore. 2. K ). for which the analytical solution is already available (Carslaw and Jaeger 1959). Rubin 1995.15) . As shown in Fig. The characteristic length of the rock is L R . 7. . . 3. respectively. For this purpose. the scalar product of (n x . (7. the solidiﬁcation problem in the ideal experiment can be treated as a one-dimensional Stefan solidiﬁcation problem. Weinberg 1996). which are modelled by K ﬁnite element meshes of equal length in the magma solidiﬁcation direction. It is assumed that the characteristic length of the rock is much larger than that of the intruded magma. we deﬁne that the solidiﬁcation thickness of the intruded magma during a time period Δt Mk is equal to the length of the ﬁnite element mesh in the magma solidiﬁcation direction and. In order to use the ﬁxed ﬁnite element mesh. . it is useful to keep Δx constant.2 Implementation of the Equivalent Source Algorithm 161 y LR x 0 LM LR LR Fig. ∂t ∂t ∂t Δt Mk Δt Mk (7. n y ) and (∂ x I ∂t. 1973.2 A representative model for the magma solidiﬁcation problem be determined analytically through the following ideal experiment. This implies that for this ideal experiment. Δt Mk = t Mk − t Mk−1 (k = 1.2. ΔL Mk = Δx = L M (2K ). the sample in the ideal experiment is a rectangular domain ﬁlled with the intruding magma in the middle and the intruded rocks on the either side. then the length of the ﬁnite element mesh in this direction is Δx = L M (2K ). ∂ y I ∂t) can be expressed as follows: nx Δx ∂ xI ∂ yI ∂ xI ΔL Mk + ny = nx ≈ = . .14) where Δt Mk is the time period when the solidiﬁcation boundary moves from one side of the kth ﬁnite element in the magma solidiﬁcation direction at time t = t Mk−1 to another side of the kth ﬁnite element in the magma solidiﬁcation direction at time t = t Mk . Under this assumption.

t Mk = k 2 (Δx)2 . . (7. t Mk ) = 4ρ M Lαβ 2 (2 k − 1)(Δx)2 (k = 1. 2. 3. which is numbered from the initial intruded interface between the rock and magma. 3. K ). . (7. 2 (7. substituting Eq. This is the main advantage in using the proposed equivalent source algorithm to simulate the chemical effect of the intruded magma solidiﬁcation in geological systems. . 2. can be determined from the analytical solution (Carslaw and Jaeger 1959). .14) into Eq. the variable time step expressed by Eqs. erf(β) is the error function of variable β. (7.162 7 Simulating Thermal and Chemical Effects of Intruded Magma Solidiﬁcation Problems The time. .20) only need to be applied in the kth ﬁnite element. y. . 2. K ). . (7. This means that if a ﬁxed ﬁnite element mesh is employed. . in the ﬁnite element analysis. at which the intruded magma solidiﬁcation boundary reaches both sides of the kth ﬁnite element boundary in the magma solidiﬁcation direction. . . (7.20) where Δx = ΔL Mk = L M (2K ) and ΔL Mk is the constant solidiﬁcation thickness of the intruded magma during the variable time period Δt Mk . 2 erf(β) = √ π β 0 e−r dr. (7. both the mass source of the released volatile ﬂuids and the physically equivalent heat source expressed in Eqs. . β can be determined from the following transcendental equation: √ 2 e−β L π = .19) yields the following equation: f (x. (7. (7. t Mk ) = ρM L Δt Mk (k = 1. From the numerical analysis point of view. . y.15) and (7.16) into Eq. 3.13) yields the following equation: f (x.16) where α = λ R (ρ R c p R ) is the thermal diffusivity of the rock. (7. It needs to be pointed out that since the magma solidiﬁcation boundary passes only through the kth ﬁnite element from the initial intruded interface between the rock and magma during the time period Δt Mk . a change in the time step is much easier to implement than a change in the ﬁnite element mesh in the ﬁnite element analysis. K ).16) needs to be used during the magma solidiﬁcation period in the ﬁnite element analysis. 4αβ 2 t Mk−1 = (k − 1)2 (Δx)2 4αβ 2 (k = 1. .17) where TR0 is the initial temperature of the rock.18) Substituting Eqs.19) Finally.12) and (7. β[1 + erf(β)] c p R (Tm − TR0 ) (7.

3 Veriﬁcation and Application of the Equivalent Source Algorithm Since the key part of the proposed algorithm for simulating the chemical effect of the intruded magma during solidiﬁcation is to transform the original moving interface (i. Note that both the mass source and the physically equivalent heat source are only applied to some elements in the ﬁnite element analysis during the solidiﬁcation process of the intruded magma. tMMax = L 2 (16αβ 2 ) is the maximum time instant to complete the solidiﬁcation M of the intruded magma.22) (7.7.16).15) are used to calculate the mass source of the released volatile ﬂuids during the intruded magma solidiﬁcation.17) is solved to determine the value of β. 2 (7. and the solidiﬁcation temperature of the intruded magma. the proposed equivalent algorithm for simulating the chemical effect of the intruded magma solidiﬁcation includes the following ﬁve main steps: (1) For the given values of the initial temperature of the rock.3 Veriﬁcation and Application of the Equivalent Source Algorithm 163 Due to the geometrical symmetry of the ideal experiment. TR0 . TR = Tm 0≤x ≤ √ LM − 2β αt.15) and (7.7) for the temperature and concentration distribution in the whole domain during and after solidiﬁcation of the intruded magma. t ≤ tMMax .21) √ 2x + L M − 4β αt (Tm − TR0 )erfc √ 4 αt TR = TR0 + 1 + erf(β) √ LM − 2β αt ≤ x ≤ L R . the solidiﬁcation interface between the rock and intruded magma) problem into . Eq. (7. (2) Substituting the value of β into Eq. In summary.6) and (7. Tm . analytical solutions for the temperature distribution during the solidiﬁcation of the intruded magma can be expressed as follows. (7. (7. The automatic transition from solidiﬁcation to post-solidiﬁcation of the intruded magma is another advantage in using the proposed equivalent source algorithm for simulating the chemical effect of the intruded magma solidiﬁcation in the conventional ﬁnite element analysis.23) where erfc(x) = 1 − erf(x) is the complementary error function of variable x. (4) The integration time step is determined using Eqs.12) and (7. 2 TR = TR0 + ρ M L M c pM (Tm − TR0 ) + L − x 2 e 4αt √ 2ρ R c p R απ t ( t > tMMax ). t ≤ tMMax . (5) The conventional ﬁnite element method is used to solve Eqs. (7. (3) Eqs.20) yields the value of the physically equivalent heat source due to the solidiﬁcation of the dike-like and sill-like intruded magma. 7.e. (7. (7.

This indicates that we only need to verify the proposed algorithm for simulating the heat transfer process associated with the intruded magma solidiﬁcation. only the heat transfer process associated with the benchmark magma solidiﬁcation problem is considered to verify the proposed algorithm. On the other hand. for which the conventional ﬁnite element method has been used for many years (Zienkiewicz 1977). the effect of the released volatile ﬂuids on the heat transfer and magma solidiﬁcation is negligible due to the relatively small amount of their volumetric mass sources. is (Finite element mesh of the whole computational domain) y x 0 (Detailed mesh for the intruded magma solidification domain) Fig. For the above reasons. In this ﬁgure. As mentioned in the previous section. transport of the released volatile ﬂuids can be treated as a conventional mass transport problem. Figure 7. 7.164 7 Simulating Thermal and Chemical Effects of Intruded Magma Solidiﬁcation Problems a physical equivalent problem which can be solved using the conventional ﬁxedmesh ﬁnite element method. for which the analytical solution is available. the original magma intruded region. it is necessary to use a benchmark magma solidiﬁcation problem.3 shows the ﬁnite element mesh of the benchmark magma solidiﬁcation problem.3 Finite element mesh of the benchmark magma solidiﬁcation problem . which is indicated in black in the ﬁnite element mesh of the whole computational domain. to verify the proposed algorithm.

thermal conductivity is 1.06 W/m3 for this one-step solidiﬁcation model. for the three-step solidiﬁcation model. the thermal properties of the intruded magma are assumed to be the same as those of the surrounding rocks.58% for the one-step solidiﬁcation model and the three-step solidiﬁcation model respectively.31 W/m3 respectively. respectively.16). Using Eq. Since the temperature difference between the intruded magma and the surrounding rocks is an important indicator of this benchmark problem.16).78% and 0.7. namely a one-step solidiﬁcation model and a three-step solidiﬁcation model. This demonstrates that the numerical model based on . It is obvious that the numerical solutions of the temperature difference between the intruded magma and the surrounding rocks. For example. The relative numerical error between the numerical and analytical solutions for the maximum temperature difference between the intruded magma and the surrounding rocks is 0. it is assumed to be 1000◦ C in the numerical computation.2 × 105 J/kg. (7.74 W/(m×◦ C). For the one-step solidiﬁcation model. The corresponding physically-equivalent heat source determined from Eq. which is 10.86) 9 days and the corresponding physically-equivalent heat sources are 8901. the numerical solutions of the maximum temperature difference between the intruded magma and the surrounding rocks are 732. the constant time step of 10.5 show the comparisons of the analytical solutions with the corresponding numerical solutions from the one-step solidiﬁcation model and the three-step solidiﬁcation model respectively.86 9. the value of β is determined to be 0.5◦ C for the one-step solidiﬁcation model and the three-step solidiﬁcation model respectively.17).86 3 and (5 × 10.73 days). while the whole computational domain is modeled with 640 four-node quadrilateral ﬁnite elements. The following parameters are used in the ﬁnite element analysis: the densities of both the magma and surrounding rocks are 2900 kg/m3 .86 days is used throughout the rest of the transient ﬁnite element analysis.73 days (i. (7.e.54 W/m3 . the intruded magma is solidiﬁed just in one time step. The width of the intruded dike-like magma is assumed to be 2 m in this benchmark problem.8◦ C.86 days from Eq. After completion of the intruded magma solidiﬁcation.20) is 989. Using the above thermal properties and Eq. the latent heat of fusion of the intruded magma is 3.73 approximately. (7. Figures 7.3 Veriﬁcation and Application of the Equivalent Source Algorithm 165 modeled using 240 four-node quadrilateral ﬁnite elements. are considered in the transient ﬁnite element analysis. the three variable time steps are 10. have very good agreement with the corresponding analytical ones. while the corresponding analytical solution of the maximum temperature difference between the intruded magma and the surrounding rocks is 737. which are obtained from both the one-step solidiﬁcation model and the three-step solidiﬁcation model.1◦ C and 733.4 and 7. at the time instant of 21. For the purpose of comparing the numerical solutions with the corresponding analytical ones.18 W/m3 and 1780. However. In order to examine the effect of time-steps of the intruded magma solidiﬁcation on the overall accuracy of the numerical solution. t = 21. the intruded magma is solidiﬁed in three variable time steps so that the intruded magma solidiﬁed equal distance from the beginning interface between the magma and surrounding/solidiﬁed rocks. (7. two computational models. 10. speciﬁc heat is 1200 J/(kg×◦ C). 2967. The length and width of the whole computational domain are 42 m and 20 m.

8 33.4 days 0 8.4 Comparison of analytical solutions with numerical solutions (One-step solidiﬁcation) the proposed equivalent algorithm for simulating solidiﬁcation effects of magma intrusion problems in porous rocks can produce highly accurate numerical solutions.2 days t=434. 7.6 days t=217. one-step solidification) ΔT (oC ) ΔT (oC ) o + * x t=21. Since the solidiﬁcation interface between the rock and intruded magma is precisely considered in the magma solidiﬁcation direction. Since the numerical solution from the one-step solidiﬁcation model has similar accuracy to that from the three-step solidiﬁcation model.6 42.e.4 25. we have doubled and tripled the mesh density in the magma solidiﬁcation region.e. 240 ﬁnite elements to simulate the intruded magma in the solidiﬁcation direction).73 days t=108.2 33. the doubled mesh density (i. all the three meshes of the original mesh density (i.e.73 days t=108. 480 ﬁnite elements to simulate the intruded magma in the solidiﬁcation direction) and the tripled mesh density (i. For the purpose of examining the numerical solution sensitivity to the mesh density in the magma solidiﬁcation direction.6 x (m) (Numerical solution.0 Fig.2 x (m) (Analytical solution) 16.6 days t=217.8 42.2 days t=434.166 7 Simulating Thermal and Chemical Effects of Intruded Magma Solidiﬁcation Problems 1000 800 600 400 200 0 0 25. the one-step solidiﬁcation model is useful for simulating thermal effects of dike-like magma intrusion problems in porous rocks.4 16.0 1000 800 600 400 200 0 o + * x t=21. 720 ﬁnite .4 days 8.

This indicates that.6 42. 7.2 days t=434.0 Fig.4 days 0 8.6 days t=217. for intruded magma dikes and sills of large aspect ratios.4 16.4 days 1000 800 600 400 200 0 ΔT (oC ) o + * x t=21. It is noted that.7.5 Comparison of analytical solutions with numerical solutions (Three-step solidiﬁcation) elements to simulate the intruded magma in the solidiﬁcation direction) produce similar numerical results.2 days t=434.6 days t=217. The use of the resulting .8 25. due to the precise consideration of the magma solidiﬁcation interface.4 42.2 x (m) (Analytical solution) 33.73 days t=108. three-step solidification) 8.3 Veriﬁcation and Application of the Equivalent Source Algorithm 1000 800 600 400 200 0 0 16.0 ΔT (oC ) 167 o + * x t=21.6 x (m) (Numerical solution.8 25. Such a one-dimensional (1D) solidiﬁcation model could be interactively used with two-dimensional (2D) models for simulating the thermal ﬁeld in the surrounding rocks. the present numerical algorithm is not sensitive to the mesh density in the solidiﬁcation direction of the magma solidiﬁcation region. one could have expected on dimensional considerations that the solidiﬁcation of the intruded magma could be treated by means of a one-dimensional model in the solidiﬁcation direction and that such an approximation is asymptotically valid to the leading order in the aspect ratio of the dike and sill.73 days t=108.2 33.

6 Finite element mesh of the dike-like magma intrusion problem . After the veriﬁcation of the proposed equivalent source algorithm for simulating solidiﬁcation effects of dike-like magma intrusion problems in porous rocks. because the previous boundary conditions must be modiﬁed and the current ones must be applied to the 2D model. Due to this obvious advantage in using the proposed model.2 km and 10 km. there is no need to develop the hybrid 1D-2D model for simulating the solidiﬁcation problems associated with intruded magma dikes and sills. The diffusivity of the released volatile ﬂuids is 2 × 10–6 m2 /s.6 shows the ﬁnite element mesh of the dike-like magma intrusion problem. 7. one must use the 1D model to determine the temperature and thermal ﬂux boundary conditions on the solidiﬁcation interface between the rock and intruded magma. since the magma solidiﬁcation in the intruded magma region and heat transfer in the surrounding rocks are considered simultaneously. The intruded magma region is modelled with 144 four-node quadrilateral ﬁnite elements. Since the width of the intruded magma for this application problem is 100 times that for the previous benchmark magma solidiﬁcation problem. the above-mentioned boundary conditions need to be determined and applied in the hybrid 1D-2D model repeatedly for each time step during the intruded magma solidiﬁcation. respectively. since the 1D and 2D models are considered separately and interactively.168 7 Simulating Thermal and Chemical Effects of Intruded Magma Solidiﬁcation Problems hybrid 1D-2D model may avoid the need to introduce equivalent volumetric heat and mass source terms. The length and width of the whole computational domain are 40. This complexity has been avoided by using the proposed numerical algorithm in this study.000 times that for the previous benchmark magma solidiﬁcation problem. This certainly adds numerical modelling complexity during the magma solidiﬁcation. Since the magma solidiﬁcation is time-dependent. However. The pressure of the intruded magma is assumed to be 3000 bars. The thermal properties of this application problem are exactly the same as those of the previous benchmark magma solidiﬁcation problem. and then apply such boundary conditions to the 2D model for simulating the thermal ﬁeld in the surrounding rocks. the solidiﬁcation time of the intruded magma for this application problem is 10. y 0 x Fig. it has been applied to investigate the chemical effects of a dike-like magma intrusion/solidiﬁcation problem in the upper crust of the Earth. the region of the intruded dike-like magma (as indicated by black colour) is 0. Figure 7. rather than separately.2 km and 6 km in the horizontal and vertical directions. in the ﬁnite element analysis. while the whole computational domain is modelled with 1040 four-node quadrilateral ﬁnite elements in the transient ﬁnite element analysis.

8765 × 1012 s ) Fig. For this application problem. 7.3826 × 1011 s ) ( t = 1. The initial average temperature of the rocks surrounding the intruded dike-like magma is 230◦ C. the boundary condition is that the temperature at the top of the computational model is 20◦ C throughout the transient ﬁnite element analysis. (7.16). The initial concentration of the released volatile ﬂuids is assumed to be zero in the whole ( t = 1.e. Therefore..6673 × 1011 s ) ( t = 9. The initial conditions are as follows: the initial temperature of the intruded dike-like magma is 1230◦ C and the initial temperature at the rest (i.8765 × 1011 s ) ( t = 4. the time step of 9.3 Veriﬁcation and Application of the Equivalent Source Algorithm 169 as indicated from Eq.7.3826 × 109 s (which is approximately equal to 300 years) is used in the transient ﬁnite element analysis of the dike-like magma intrusion problem in the upper crust of the Earth.7 Temperature distributions of the dike-like magma intrusion problem at different time instants . except for the intruded magma) of the bottom of the computational model is 320◦ C.

In addition.5306 from Eqs. the mole fraction of the released volatile ﬂuids from the albite melt can be determined as 0.6 x (km) (Along the cross-section of y = 3000 m) 8.7 kg/m3 (Haar et al.0 10.3826 × 10 s 12 x t = 1.433 × 10–8 kg/(m3 × s) for the released volatile ﬂuids in the computation.8765 × 1012 s 42.0 8.7 shows the temperature distributions of the dike-like magma intrusion problem at four different time instants.11).2 33.3826 × 10 s x t = 1.8 25. It is clear that with increasing time.8765 × 1011 s + t = 4.0 y (km) (Along the cross-section of x = 20.0 620 500 380 260 140 20 0 6. while the density of the albite melt is 2700 kg/m3 .6673 × 1011 s 11 * t = 9.8 Temperature distributions of the dike-like magma intrusion problem along two different cross-sections . 7.6673 × 1011 s 11 * t = 9. (7.10) and (7.0 Fig. the density of the released volatile ﬂuids is 374. which results in a mass source of 1.0 4.8765 × 1011 s + t = 4. 1984). the total temperature localization area generated by the intruded magma becomes larger and 500 440 380 320 260 200 T ( oC ) o t = 1.170 7 Simulating Thermal and Chemical Effects of Intruded Magma Solidiﬁcation Problems computational domain.100 m) 2. Under the above temperature and pressure conditions of the intruded magma.4 T ( oC ) o t = 1.8765 × 10 s 0 16. Figure 7.

3 Veriﬁcation and Application of the Equivalent Source Algorithm 171 larger. Although the total temperature localization area generated by the intruded magma is limited in the top part of the A A B C D E F G H I J LEGEND – 0. but the maximum temperature generated by the intruded magma becomes smaller and smaller.3826 × 1011 s and 1.5000E–03 – 0.3500E–02 – 0.7500E–02 – 0.2500E–02 – 0.6500E–02 – 0.9500E–02 A A BC B A B C D E F G H I J ( t = 4.5500E–02 – 0.8765 × 1011 s.9500E–02 A D B AB E C C A B C D E F G H I J ( t = 1.9◦ C for the time instants of 1.100 m is 514. For instance.7500E–02 – 0.6673 × 1011 s. These phenomena can be clearly seen from Fig.8◦ C.5000E–03 – 0.8500E–02 – 0.8500E–02 – 0.5000E–03 – 0.9500E–02 A ( t = 9.6500E–02 – 0.4◦ C and 354.6673 × 1011 s ) Fig. the maximum temperature along the cross-section of x = 20. where the temperature distributions of the dike-like magma intrusion problem are displayed along two typical cross-sections of the computational model.4500E–02 – 0. 9.3500E–02 – 0.2500E–02 – 0.9 Concentration distributions of volatile ﬂuids for the dike-like magma intrusion problem at different time instants (Whole system view) .9500E–02 ( t = 9.4500E–02 – 0.2500E–02 – 0.8. 4.7500E–02 – 0.3826 × 10 9 s ) A B C D E F G H I J LEGEND – 0. 7.3500E–02 – 0.8765 × 1012 s respectively.8500E–02 – 0.1500E–02 – 0.1500E–02 – 0.5000E–03 – 0.6500E–02 – 0.4500E–02 – 0.5500E–02 – 0.2500E–02 – 0.7500E–02 – 0.1500E–02 – 0.4◦ C. 444.6500E–02 – 0.5500E–02 – 0.8500E–02 – 0. 394.3826 × 1010 s ) LEGEND – 0.8765 × 1011 s ) LEGEND – 0.1500E–02 – 0.3500E–02 – 0.5500E–02 – 0. 7.7.4500E–02 – 0.

5500E–02 – 0.3500E–02 – 0. 7.5500E–02 – 0.9500E–02 ( t = 9.5000E–03 – 0. the total concentration area of the volatile ﬂuids generated during the intruded magma solidiﬁcation becomes larger and larger. It is clear that with an increase in time.1500E–02 – 0.6500E–02 – 0. 7.3826 × 10 9 s ) A CB C A B EF C G ED A B C D E F G H I J LEGEND – 0.10 and 7. while Figs.3500E–02 – 0.7500E–02 – 0.3826 × 1010 s ) Fig.9 shows the concentration distributions of the volatile ﬂuids for the whole system of the dike-like magma intrusion problem at four different times.7500E–02 – 0.9500E–02 ( t = 9.6500E–02 – 0.11 show the detailed concentration distributions of the volatile ﬂuids for a zoomed-in part of the dike-like magma intrusion problem at the same four different times.10 Concentration distributions of volatile ﬂuids for the dike-like magma intrusion problem at different time instants (Zoomed-in view) .4500E–02 – 0.2500E–02 – 0.4500E–02 – 0. 1998a).2500E–02 – 0.172 7 Simulating Thermal and Chemical Effects of Intruded Magma Solidiﬁcation Problems computational model.10 and 7. temperature localization in the top part of the computational model can be significantly enhanced so that a favourable region for ore body formation and mineralization can be created just above the intruded dike-like magma (Zhao et al. if the porous rocks are saturated by the poreﬂuid in the upper crust of the Earth.1500E–02 – 0.8500E–02 – 0.11. where the concentration distributions of A A B C D E F G H I J LEGEND – 0.5000E–03 – 0.8500E–02 – 0. 7. but the maximum concentration of the volatile ﬂuids generated by the intruded magma becomes smaller and smaller. Once this convective pore-ﬂuid ﬂow takes place. Figure 7. 1997a). it is large enough to cause convective pore-ﬂuid ﬂow within the whole system (Zhao et al. These phenomena can be clearly seen from Figs.

2500E–02 – 0.7500E–02 – 0. is formed during the intruded magma solidiﬁcation. 2001d). This indicates that the released volatile ﬂuids can create a favourable region for ore body formation and mineralization to take place through some chemical reactions around the intruded dike-like magma.9500E–02 ( t = 1.6500E–02 – 0.5000E–03 – 0.4500E–02 – 0. it is large enough to cause ore body formation and mineralization to take place within the system.7.11 Concentration distributions of volatile ﬂuids for the dike-like magma intrusion problem at different time instants (Zoomed-in view) the volatile ﬂuids during the dike-like magma solidiﬁcation are displayed within the zoomed-in part of the computational model.9500E–02 ( t = 4. Since the released volatile ﬂuids can react chemically with the surrounding rocks (Zhao et al.7500E–02 – 0. Although the total area of the aureole generated by the intruded magma is limited in the computational model.1500E–02 – 0.6500E–02 – 0.8500E–02 – 0.5000E–03 – 0.5500E–02 – 0. which can be clearly observed in these ﬁgures.3 Veriﬁcation and Application of the Equivalent Source Algorithm 173 A B C D E F G H I J LEGEND – 0. an aureole.8765 × 1011 s ) A B C D E F G H I J LEGEND – 0. .8500E–02 – 0. 7.6673 × 1011 s ) Fig.4500E–02 – 0.3500E–02 – 0.5500E–02 – 0.2500E–02 – 0.3500E–02 – 0.1500E–02 – 0.

Although there are at least three kinds of mixed-mode crack propagation theories available. and so cracking and fracturing phenomena are ubiquitous. particularly from the deep crust into the shallow crust of the Earth. exploration for new mineral resources has become one of the highest priorities for many industrial countries. Zhao et al.. Thus. including the propagation direction and incremental growing length of the crack. DOI 10. can be determined by crack propagation theories established in fracture mechanics. Schaubs and Zhao 2002. Yeh and Tripathi 1989. Schafer et al. Lecture Notes in Earth Sciences 122. 1997a. it is usually assumed that a crack initiates when the value of a local principal stress attains the tensile strength of the brittle material. Crustal materials of the Earth can be largely considered as brittle rocks. Zhao et al. b. In the study of numerical simulation of crack initiation and evolution. Zhao et al. C Springer-Verlag Berlin Heidelberg 2009 175 . namely a theory based on C. Raffensperger and Garven 1995. Once a crack is initiated. 2005a) have been conducted to understand the detailed physical and chemical processes that control ore body formation and mineralization within the upper crust of the Earth. Zhao et al. temperature and pressure are appropriate. extensive studies (Garven and Freeze 1984. ore body formation and mineralization can take place as a result of such ﬂuid ﬂow. 1998a.1007/978-3-540-89743-9 8. 1998a. Xu et al. Fundamentals of Computational Geoscience. Cracks created within the Earth’s crust often provide a very useful channel for mineral-bearing ﬂuids to ﬂow. 1991. Steefel and Lasaga 1994. 1999. Due to the increased capability of numerical fracture mechanics for considering complicated geometry and boundary conditions. the numerical simulation of spontaneous crack generation in brittle rocks within the upper crust of the Earth has become an important research topic in the ﬁeld of computational geoscience. mineralogy. If other conditions such as ﬂuid chemistry. propagation of the crack. Zhao et al. For this reason. Because of the ever-increasing demand for mineral resources in the contemporary world. 2000b. This was the direct outcome of combining conventional fracture mechanics with numerical methods such as the ﬁnite element and the boundary element methods. 2003e.Chapter 8 The Particle Simulation Method for Dealing with Spontaneous Crack Generation Problems in Large-Scale Geological Systems Cracking and fracturing are one class of major failure mechanisms in brittle and semi-brittle materials. it signiﬁcantly extends the applicability of conventional fracture mechanics to a wide range of practical problems in civil and geotechnical engineering ﬁelds. The numerical simulation of crack initiation has existed from the inception and development of numerical fracture mechanics. 2002a.

Itasca Consulting Group. Lee and Bathe 1994. 1977). 1999. such as the distinct element method developed as a particle ﬂow code (Cundall and Strack 1979. Bardet and Proubet 1992. Similarly. geotechnical and other engineering problems (Cundall and Strack 1979. Potyondy and Cundall 2004). inc. the triangular quarter-point element. 2005. Klerck et al. automatic meshing and re-meshing algorithms (Zienkiewicz and Zhu 1991. particle simulation methods. Khoei and Lewis 1999. for the ﬁnite element simulation of a two-dimensional fracture mechanics problem. Zhao et al. Kwak et al. 2006f). results in improved numerical results for stress intensity factors near a crack tip (Barsoum 1976. Lorenzi 1985. For the purpose of removing this limitation associated with the conventional numerical method (which is usually based on continuum mechanics) in simulating a large number of spontaneously generated cracks. the maximum circumferential tensile propagation theory has been widely used in the ﬁnite element analysis of crack initiation and propagation problems because it is not only physically meaningful. a theory based on propagation driven by strain energy density near the crack tip (Sih and Macdonald 1974) and a theory based on crack propagation driven by maximum energy release rate (Rice 1968. 1985). the formulation based on discrete particle simulation is conceptually simpler than that based on continuum mechanics. 1999. 1977. 2004 and Schubert et al. which is formed by collapsing one side of a quadrilateral 8-noded isoparametric element. McBride et al. Salman and Gorham 2000.176 8 Spontaneous Crack Generation Problems in Large-Scale Geological Systems propagation driven by the maximum circumferential tensile stress (Erdogan and Sih 1963). 2003) have been developed in recent years. provides a very useful tool to deal with this particular kind of problem. such ﬁnite element methods with automatic meshing and re-meshing algorithms have been applied for solving crack generation and propagation problems in systems of only a few cracks. for the ﬁnite element modelling of a three-dimensional fracture mechanics problem. In order to simulate appropriately displacement discontinuities around a crack. while it can be also used to simulate the 1/ r singularity in perfect plasticity by simply allowing different displacements at the nodes located on the collapsed side or face of the element. but it is also easily implemented in a ﬁnite element code. it has been used to deal with a large number of geological and geophysical problems in both two and three dimen- . It has been widely demonstrated that. Even though the particle simulation method was initially developed for solving soil/rock mechanics. Owen et al. However. 2004. Since displacement discontinuities at a contact between two particles can be readily considered in these methods. Ingraffea and Manu 1980). 2004. Thomton et al. Cundall 2001. because crack generation at a contact between two particles is a natural part of the particle simulation process. Tomas et al. Bouchard et al. 2002. Hellen 1975. the prismatic quarter-point element. This kind of quarter-point element can be used to simulate √ the 1/ r singularity in elastic fracture mechanics by simply assigning the same displacement at the nodes located on the collapsed side or face of the element. can also lead to much better numerical results for stress intensity factors near a crack tip (Barsoum 1976. which is formed by collapsing one face of a cubic 20-noded isoparametric element. Li et al. 1999.

little work. 2000. in which the macroscopic length-scale geological structures are always of special interest. this chapter is restricted to deal with the particle simulation of spontaneous crack generation problems within large-scale quasi-static geological systems. Strayer and Suppe 2002. Donze et al. Antonellini and Pollard 1995. Zhao et al. 2008g). a geological length-scale problem. Strayer and Huddleston 1997. has been reported on using the particle simulation method to deal with spontaneous crack generation problems in the upper crust of the Earth. For this reason. In the ﬁrst approach. Antonellini and Pollard 1995. Nevertheless. In the second approach. such as the ﬁnite element method (Zienkiewicz 1977. the macroscopic deformation pattern can be reasonably simulated in this approach (e. Burbidge and Braun 2002. the following three approaches are often used in dealing with the particle simulation of geological length-scale problems. To overcome this difﬁculty. Finch et al. Scott 1996. It is however computationally prohibitive to simulate a whole geological system using real physical particles of a microscopic length-scale. Strayer and Suppe 2002. 2004. a meter-scale or