Computers and Geotechnics 63 (2015) 199–214

Contents lists available at ScienceDirect

Computers and Geotechnics

journal homepage: www.elsevier.com/locate/compgeo

Coupling of soil deformation and pore fluid flow using material point
method
Samila Bandara ⇑, Kenichi Soga
Cambridge University, Department of Engineering, Trumpington Street, Cambridge CB2 1PZ, UK

a r t i c l e i n f o a b s t r a c t

Article history: This paper presents the formulation and implementation of a numerical procedure based on material
Received 21 March 2014 point method (MPM) to solve fully coupled dynamic problems that undergo large deformations in satu-
Received in revised form 21 August 2014 rated soils. The key aspect of this formulation is that it considers two sets of Lagrangian material points to
Accepted 27 September 2014
represent soil skeleton and pore water layers. The accuracy of the method is tested by comparing the
results to some analytical solutions of consolidation theory. The developed method has been applied
to model progressive failure of river levees to illustrate the practical applications. The numerical results
Keywords:
show the robustness of the proposed method with regard to large deformations that undergo rapid fail-
Material point method
Saturated soil
ure mechanisms.
Large deformations Ó 2014 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
Dynamic coupled analysis (http://creativecommons.org/licenses/by-nc-nd/3.0/).
Meshfree methods
River levees

1. Introduction be difficult to determine these hydrodynamics parameters from

Large deformation problems in fluid-saturated granular med-
ium are of great interest in areas such as geophysics, engineering
applications and industrial processes. In particular, geophysical
and gravity driven flows such as fast catastrophic landslides, flow-
slides, avalanches, and debris flows cause much damages in many
parts of the world due to large travel distances. Similarly failures of
man-made structures (i.e. levees, dikes, and embankments) that
occur mainly due to intense rainfall and flooding can also results
in severe damages due to seepage flows.
In problems such as landslides and flowslides, it is important to
predict the travel distances and to identify the post-failure mech-
anism. Generally it is difficult to use classical mesh based methods
such as the finite element method (FEM) to model the behaviour
that involves large deformations due to severe mesh distortion
errors. Most of the continuum techniques that have been used to
model large deformations in granular soils (for instance rapid
landslides with long travel distances) often consider soil as a
visco-plastic fluid (i.e. Bingham fluid) and fluid-mechanics based
hydrodynamics equations have been used [1]. However, it may
⇑ Corresponding author at: Laboratory of Soil Mechanics, Swiss Federal Institute
of Technology (EPFL), EPFL-ENAC-LMS Station 18, CH-1015 Lausanne, Switzerland.
Tel.: +41 21 693 2353.
E-mail addresses: samila.bandara@cantab.net (S. Bandara), ks207@cam.ac.uk
(K. Soga).
typical geotechnical investigation.
The recent advances in particle methods (or meshfree methods)
that can be derived in continuum mechanics framework allow us
to model large deformation behaviour using conventional geotech-
nical constitutive models. Wide range of meshfree methods are
available in the literature and these include smoothed particle
hydrodynamics (SPH) method [2,3], material point method
(MPM) [4], finite point method [5], element free Galerkin method
[6], particle in cell method [7]. Many of these meshfree methods
have been mainly applied to fluid dynamics problems. Only
recently they have been applied to study solid mechanics prob-
lems, and SPH and MPM have been popular choices.
SPH method is a fully Lagrangian, particle method, in which a
continuum material is divided into a set of particles with constant
mass and each particle is assigned a spatial distance called a
smoothing length in which the physical properties of each particle
is smoothed using a kernel function (i.e. cubic spline kernel func-
tion) [8]. Although SPH method is widely used in the research com-
munity it has certain disadvantages for geotechnical problems.
Many geotechnical problems involve boundary interfaces (soil-
structure interaction) and its accuracy near the boundaries is less
due to insufficient neighbouring particles and thus resulting in loss
of consistency, which is a widely investigated topic in the SPH
research community. Furthermore, numerical oscillations of parti-
cles occur since SPH does not require the velocity field to be single
valued (i.e. allows particle penetration). Different stabilisation

http://dx.doi.org/10.1016/j.compgeo.2014.09.009
0266-352X/Ó 2014 The Authors. Published by Elsevier Ltd.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/).
200 S. Bandara, K. Soga / Computers and Geotechnics 63 (2015) 199–214
techniques are being used to minimise this issue (i.e. artificial vis-
cosity criterion, average velocity criterion [9,10]).
This study adopts MPM [4,11], in which Lagrangian point
masses, or material points, are moved through an Eulerian back-
ground mesh (usually a rectangular mesh). Although there is a
mesh, it is only used to solve governing equations and purely cho-
sen for computational convenience. All the properties of the con-
tinuum are assigned to the material points and all the
information is carried out by these material points while the mesh
does not carry permanent information. The major advantage of this
method compared to other methods is that application of bound-
ary conditions are straightforward due to the presence of back-
ground grid, since the boundary conditions can be directly
applied to grid nodes as in the FEM. Also particle penetrations
are avoided since the particles move in a single valued velocity
field (i.e. particle velocities are interpolated from the nodal veloc-
ities). MPM has many similarities to FEM and therefore has the
advantage of using advanced features that are well established in
the FEM.
The single phase MPM and its extension generalised interpola-
tion material point (GIMP) method [12] has been applied to several
geotechnical problems that involve large deformations [13–15].
MPM has also been applied to study problems in saturated soil
by several researchers [16–20]. It is common in geotechnical engi-
neering to use multi-phase numerical techniques that couples soil
deformation and pore fluid flow so that the entire time dependent
process of initial undrained behaviour followed by a consolidating
behaviour can be modelled. Since MPM is capable of modelling
large deformations, it is only required to propose a coupled formu-
lation to model large deformations.
A fully coupled MPM model still requires further development
for refinement. Zhang et al. [19], Jassim et al. [17] and Alonso
and Zabala [16] use a single layer of material points to model
two phase soil-pore fluid behaviour, whereas Shin [20],
Mackenzie-Helnwein et al. [21] and Abe et al. [18] use two
layers of material points (one for solid skeleton phase and the
other for pore fluid phase).
Zhang et al. [19] proposed a coupling MPM based on the u–p
form governing equations of saturated soil. In this study, a simple
contact algorithm was developed using Coulomb friction law and
the results were compared against FEM results. The authors solved
problems with dynamic responses of saturated soil under compact/
impact with solid bodies and did not apply it to model large defor-
mations. Alonso and Zabala [16] used MPM to simulate progressive
failure of Aznalcollar dam in saturated soil condition. The brittle
foundation clay was modelled using strain-softening Mohr–
Coulomb elasto-plastic model. In this study, each material point
was represented by saturated soil with constant mass (i.e. porosity
is constant). Internal forces and pressure increments at nodes were
calculated using the shape function gradients evaluated at cell cen-
tres to avoid issues due to incompressibility (due to the usage of
very low permeability), and this approach would be difficult to
use with large deformations since it will lead to zero energy modes
as observed in FEM with low order elements [22]. Additionally, the
Rayleigh damping method was used to reduce numerical oscilla-
tions. Similarly, Jassim et al. [17] proposed a coupled dynamic,
two-phase MPM formulation via velocity formulation (i.e. consider
both solid and fluid phase accelerations) with a single layer of
material points and used the method to model the pore pressure
development under wave attack on sea dike that does not undergo
large deformations. The velocities of both soil and water were
stored at single material point and updated using the computed
nodal accelerations of soil and water at each step. They used
enhanced (averaged) volumetric strains to mitigate spurious pres-
sure fields and locking as observed in FEM, although this method
do not guarantee oscillation-free results for saturated soils that
undergo incompressibility [23]. Also, the internal force at nodes
were calculated using Gauss integration considering the total vol-
ume of cells when a cell volume is filled 90% by material points and
using averaged stress in cells that is determined by volume averag-
ing of particle stresses (i.e. same stress for Gauss points inside
same cell). However, this approach may not conserve energy (i.e.
mass conservation is violated due to the use of cell volumes that
does not represent true material point volumes).
In reality pore water can move relative to solid skeleton, thus
resulting soil material point to have different water velocity than
the velocity computed using the above method. When using the
inertia terms, the coupled problem is of hyperbolic type and it is
necessary consider the motion of the water phase and this can
be achieved from considering either the true water velocity field
or relative water velocity with respect to the solid skeleton [24].
Shin [20] and Mackenzie-Helnwein et al. [21] modelled solid–fluid
mixtures using MPM by representing solid and fluid particles as
Lagrangian particles (i.e. two layers of material points). Mixture
theory approach was considered in this study and the momentum
balance equation for the mixture was solved using Eulerian
approach. Solid particles were considered as individual particles
(i.e. not considering as a solid skeleton as in soil mechanics) with-
out considering the effective stress concept that is available in soil
mechanics. Volume fractions and equilibrium pressures have to be
calculated at each time step while using mass balance equation for
the mixture for each Eulerian mesh. This formulation faces difficul-
ties since the mixture material would not exactly fill the cell. That
is, mass would not properly conserve in these methods due to the
calculation method that is used when evaluating volume fraction
and pressure of each phase.
Abe et al. [18] were the first to model coupled hydromechanical
problems in saturated soil using two layers of material points to
represent solid skeleton layer and pore fluid layer based on Biot’s
mixture theory. In this formulation, the velocity of water material
points were evaluated at each time step using the generalised
Darcy’s law while neglecting the relative acceleration of water
with respect to the solid skeleton. This may limit the applicability
to extremely rapid motions due to the numerical limitations [25]
and also the direct calculation of water velocity in each time step
may lead to errors when dealing with large deformations. They
have applied this method to simulate the deformation of a river
embankment due to seepage failure while using Mohr–Coulomb
soil model to represent effective stress behaviour and compara-
tively good results were achieved with respect to the results from
a large scale experiment.
The main object of this study was to develop a MPM methodol-
ogy that addresses some of the issues described above so that it
becomes possible to study large deformation problems in both
dry and saturated granular soils. A fully coupled MPM formulation
was derived using a mixture theory based approach (i.e. consider-
ing the recent versions of Theory of Porous Media that use the
Lagrangian description for the solid skeleton and where the theory
is based on the concept of superimposed continua with internal
interactions and individual states of motion [26]) and then adapt-
ing the MPM formulation proposed by [11]. The key aspect of this
formulation is that it considers two sets of Lagrangian material
points to represent soil and water. Use of Lagrangian particles con-
serves mass and allows history dependent material models to be
used. Also, the discrete equations for the momentum balances
are obtained on the background grid similar to the finite element
method with an updated Lagrangian formulation. The proposed
coupled MPM formulation was implemented into a numerical code
and the performance was validated by comparing the results to
some analytical solutions of consolidation theory. It was also
used to simulate progressive failure of a river levees to illustrate
a practical application of the code.
 S. Bandara, K. Soga / Computers and Geotechnics 63 (2015) 199–214
2. Governing equations
Saturated soil is considered to be comprised of two constitu-
ents: solid skeleton and water phase, and the kinematics of each
constituent is described with Lagrangian coordinates. Similar to
mixture theory based approaches (i.e. theory of porous media), it
is assumed that the two constituents can occupy the same position
at the same time. However, the main difference between the pro-
posed MPM approach and traditional FE approaches that are based
on theory of porous media or Biot’s theory [27,26,25] is such that
the water phase is also described using Lagrangian coordinates
instead of using Eulerian coordinates. This allows us to: (a) con-
serve mass of water in the system which is very important for
applications such as modelling of advective chemical transport in
nuclear waste systems; (b) model erosion problems where the
water layer can carry the eroded soil mass content which is impor-
tant in applications such as sand production and erosion in water
retaining structures; and (c) model interaction problems of satu-
rated soil and water in an easier way. Some of these applications
will be published in future papers.
In this section, first, we list the basic assumptions in the pro-
posed approach. Second, we include the basic balance laws and
constitutive equations that describe the mechanics of saturated
soil. Then we formulate the full governing equations that are to
be used with MPM.
2.1. Basic assumptions
The balance laws are applied for each constituent separately
(i.e. solid skeleton, water phase). Soil is assumed as a multi-phase
medium and the governing equations are implemented in the
current configuration. Each material point (i.e. solid skeleton
and water) has their individual motion which is followed in time
due to the Lagrangian description. The solid skeleton and water
material points that are located at the current configuration
(Xk ) might have located at different locations in the reference
configuration (X0 ). This is described in Fig. 1. Since an Updated
Lagrangian formulation is used in this study, the last equilibrium
configuration at time t can be considered as the reference
configuration at the beginning of each time step. The solid
skeleton-water medium can be viewed as the superposition of
two continuous media (i.e. a solid skeleton and a fluid phase)
in the current configuration.
Fig. 1. Movement of solid skeleton and water phases.
201
In this formulation the momentum balance equations are
solved to obtain the accelerations of the solid skeleton and the fluid
phase. Momentum exchange term (or interaction force) is consid-
ered in fluid momentum balance equation. This approach is suit-
able for rapid failure problems.
The following assumptions are made for the coupled MPM
formulation:
(i) Solid skeleton and water phases are described in a Lagrang-
ian formulation assuming each constituents can be repre-
sented as a continuous medium (Fig. 2).
(ii) Solid grains are incompressible.
(iii) Terzaghi’s effective stress concept is assumed to be valid.
Solid skeleton is modelled using an effective stress model.
(iv) Isothermal conditions.
(v) No mass exchange between solid and water constituents.
(vi) Variation of water density (qw ) is negligible.
(vii) Variation of porosity, permeability and volume of solid skel-
eton with time is considered.
2.2. Kinematics of porous media
The initial reference (material) positions of the solid skeleton
phase and the water phase are denoted by X0s ¼ Xs ð0Þ  R3 and
X0w ¼ Xw ð0Þ  R3 , and their material point coordinates are labelled
as Xs and Xw . The spatial (current) positions of the solid skeleton
phase and the water phase are denoted by Xks ¼ Xs ðtÞ  R3 and
Xkw ¼ Xw ðtÞ  R3 , and their material point coordinates are labelled
as xs and xw . The current positions of solid and water material
points (or particles) are written as xs ¼ us ðXs ; tÞ and xw ¼
uw ðXw ; tÞ, with us ðXs ; 0Þ ¼ Xs and us ðXw ; 0Þ ¼ Xw , where u
denotes the motion of a material point.
The velocity of material points of each phase (i.e. v a where
a ¼ s; w represent soil and water) is written as:
a
D ua ðXa ; tÞ
va ¼ : ð1Þ
Dt
Then the acceleration of each phase (aa ) is written using the
a
material time derivative D =Dt as:
a
D va
aa ¼ : ð2Þ
Dt
202 S. Bandara, K. Soga / Computers and Geotechnics 63 (2015) 199–214

Fig. 2. Continuum assumption of porous medium.

2.3. Balance laws and the formulation of governing equations and water continua) and the water phase are considered. The
momentum balance equation for the mixture is considered
Using the mixture theory, the momentum balance equation for because it uses the total stress, which is a measurable quantity
the solid skeleton can be written as: in the laboratory. It is also possible to consider the momentum bal-
ance equations for the solid skeleton to calculate the acceleration
ð1  nÞqs as ¼ r  r ^ s þ ð1  nÞqs b
s þ p ð3Þ
of the solid skeleton, although it is not considered in this research.
where qs is the density of solid grains, n is the porosity, and b is the The mass balance equation for the solid skeleton can be used to
body force. r
 s is the partial stress tensor based on the mixture the- update the porosity as follows:
ory which can be written as: s s
Dðms Þ Dðð1  nÞqs V s Þ
r s ¼ r0  ð1  nÞmpw ð4Þ ¼
Dt Dt
where r0 is Terzaghi’s effective stress, pw is the water pressure, and s s s
Dð1  nÞ D qs D Vs
m is the Kronecker delta vector (½1 1 1 0 0 0T ). p^ s is the fluid-cou- ¼ qs V s þ ð1  nÞV s þ ð1  nÞqs ¼0
pled momentum interaction term which is defined as:
Dt Dt Dt
where ms is the mass of the solid skeleton, and V s is the volume of
n2 qw g
^s ¼
p  ðv w  v s Þ  pw rnw ð5Þ the solid skeleton. Dividing by V s :
k
s s
where k is the soil permeability. The second term in (5) is called the Dð1  nÞ D qs
qs þ ð1  nÞ þ ð1  nÞqs r  v s ¼ 0
‘buoyancy term’ in mixture theories and this term must be consid- Dt Dt
ered for immiscible mixtures such as fluid saturated porous solid
and neglecting the second term due to the assumption of incom-
[28].
pressible solid grains, the material time derivative of the porosity
Using (3)–(5), the final version of the momentum balance equa-
can be written as:
tion for the solid skeleton can be written as:
s
n2 qw g Dn
ð1  nÞqs as ¼ r  r0  ð1  nÞrpw þ  ðv w  v s Þ ¼ ð1  nÞr  v s : ð12Þ
k Dt
þ ð1  nÞqs b: ð6Þ
This equation can be used to obtain the porosity of the soil at the
The momentum balance equation for the water phase can be current configuration.
written as: The mass balance equation for the water phase can be used to
derive the equation of state as follows:
nqw aw ¼ r  r ^ w þ nqw b
w þ p ð7Þ
w w w w
where qw is the density of water, p ^ w is equal and opposite to p
^ s , and Dðmw Þ Dn D qw D Vw
¼ qw V w þ nV w þ nqw ¼0 ð13Þ
r w is the partial stress tensor of water, which is written as: Dt Dt Dt Dt

r w ¼ nmpw : ð8Þ where mw is the mass of water, and V w represents the phase aver-
aged volume of the water. Dividing by V w and assuming the water
The final version of the momentum balance equation for the flow is barotropic:
water phase can be written as:
w w
n2 qw g 1 D qw 1 D pw
nqw aw ¼ nrpw   ðv w  v s Þ þ nqw b: ð9Þ ¼ ð14Þ
k qw Dt K w Dt
The momentum balance equation for the entire mixture can be (13) can be written as:
written as: w w
Dn n D pw
ð1  nÞqs as þ nqw aw ¼ r  r þ ð1  nÞqs b þ nqw b: ð10Þ þ þ nr  v w ¼ 0:
Dt K w Dt
where r is the total stress of the mixture which can be defined as:
Applying (12) into the above equation (this is possible due to
r ¼ r0  mp ð11Þ the fact that the term r  v s is calculated in the same cell where
the material point of water is located), the equation of state for
where p is the effective water pressure. This is equal to pw for fully
the water phase can be written as:
saturated condition and this equation is the classical effective stress
based momentum equation used in soil mechanics. w
D pw Kw
For the coupled MPM formulation developed in this study, the ¼ ½nr  v w þ ð1  nÞr  v s : ð15Þ
momentum balance equations for the mixture (this is possible in
Dt n
the coupled MPM since the equation is solved in the background It should be noted that this equation is applied to the water
grid nodes and the components can be obtained from both soil material point and not to the mixture.
 S. Bandara, K. Soga / Computers and Geotechnics 63 (2015) 199–214 203

3. Coupled MPM formulation volume associated with the solid skeleton material point (Xsp ðtÞ)
and the current phase averaged density at the solid skeleton mate-
In the proposed formulation, the solid skeleton layer and the rial point location can be obtained using the Jacobian of the defor-
seepage water layer are represented by two layers of material mation gradient as follows:
points (i.e. solid skeleton phase and water phase) that are followed
Xsp ðtÞ ¼ JðXsp ; tÞXsp ð0Þ ð19aÞ
in a Lagrangian sense throughout the computation. Fig. 3 shows a
basic description of the coupled MPM formulation.
q s ðXsp ; 0Þ
q s ðXsp ; tÞ ¼ : ð19bÞ
The initial regions X0s and X0w for solid and water are divided JðXsp ; tÞ
SNsp
into N sp and N wp disjoint subdomains, with X0s ¼ sp¼1 Xsp ð0Þ and The current soil porosity at the solid particle location can be
SNwp written as:
X0w ¼ wp¼1 Xwp ð0Þ. The centroids of Xsp ð0Þ and Xwp ð0Þ are repre-
sented by Xsp and Xwp . The position, mass, and volume of each ½1  nðXsp ; 0Þ
nðXsp ; tÞ ¼ 1  : ð20Þ
material point initially located at the centroid of each subdomain JðXsp ; tÞ
are tracked in time. The initial mass of solid skeleton and water
material points can be written as: A similar approach can be used to calculate the current volume
Z Z (Xwp ðtÞ) and the phase averaged density (q  w ðXwp ; tÞ) associated
with the water particle using the Jacobian of the deformation gra-
m0sp ¼ q 0s ðXÞdV  q 0s ðXsp ÞXsp ð0Þ
Xsp ð0Þ Xsp ð0Þ dient of the water phase. It is also possible to calculate Xwp ðtÞ and
Z
q
 w ðXwp ; tÞ while using the current soil porosity value at the water
¼ ½1  n0 ðXsp Þq0s ðXsp ÞXsp ð0Þ ð16aÞ particle location as follows:
Xsp ð0Þ
Z Z
m0wp ¼ q 0w ðXÞdV  q 0w ðXwp ÞXwp ð0Þ Xwp ðtÞ ¼ Xwp ð0ÞnðXwp ; 0Þ=nðXwp ; tÞ ð21aÞ
Xwp ð0Þ Xwp ð0Þ
Z q w ðXwp ; tÞ ¼ nðXwp ; tÞqw ðXwp ; 0Þ ð21bÞ
¼ n0 ðXwp Þq0w ðXwp ÞXwp ð0Þ ð16bÞ
Xwp ð0Þ
and the latter method is considered in this study.
where q 0s ðXÞ and q
 0w ðXÞ represent the original phase averaged mass
densities of the solid skeleton phase and the water phase, and are
related to the intrinsic densities of each phase (i.e. q0s ðXÞ & q0w ðXÞ) 3.1. Weak form of the governing equations
as:
The main component in the MPM is the spatial discretisation of
q 0s ðXÞ ¼ ½1  n0 ðXÞq0s ðXÞ ð17aÞ momentum equations. For coupled MPM formulation, the finite
q 0w ðXÞ ¼ n0 ðXÞq0w ðXÞ ð17bÞ element method with an updated Lagrangian formulation is
adopted to obtain discrete equations for the momentum balance
where n0 ðXÞ is the initial soil porosity. The intrinsic density of the on the background grid. A single background grid is used for the
solid phase represents the density of solid grains. The total mass coupled MPM formulation. This grid is subdivided into isoparamet-
of each phase is conserved during the computations, since the mass ric quadrilateral elements Xe ; e ¼ 1; 2; . . . ; N e in two-dimensions
of the material points are constant in time. The density and volume with mesh nodes xI ; I ¼ 1; 2; 3; . . . ; N n . Following the finite element
of each material point can change during the computation. method, the functions (i.e. displacement, velocity, and accelera-
The deformation gradient for the solid skeleton phase is defined tion) are approximated by interpolating from the nodal values
as the derivative of the current position with respect to the original using shape functions (i.e. tensor product of piecewise linear basis
position as: functions). The locations of the solid (xs ¼ us ðXs ; tÞ) and the water
(xw ¼ uw ðXw ; tÞ) phases can be approximated using the global
@ us ðX; tÞ shape functions (N I ðxs Þ; N I ðxw Þ) as:
Fs ðX; tÞ ¼ ¼ Grad us ð18Þ
@X
X
Nn
and the Jacobian of the transformation from the original to current xha ¼ xaI NI ðxa Þ ða ¼ s; wÞ ð22Þ
coordinates can be written as Js ðX; tÞ ¼ detðFs ðX; tÞÞ > 0. The current I¼1

Fig. 3. Schematic diagram of the computational grid and material points for the coupled MPM: (a) at initial reference configuration; (b) at current configuration.
204 S. Bandara, K. Soga / Computers and Geotechnics 63 (2015) 199–214

where the superscript h is used to denote finite element approxima- The components of the test functions, wsI and wwI are arbitrary
tions to the corresponding continuum quantity. except at constrained nodes on the boundaries where the compo-
Then the discrete displacement field can be defined based on nents of displacement are prescribed. The spatial discretisation is
the nodal positions as: further carried out considering material points as quadrature
X
Nn points and the integrals become sums over material points. The
uha ðxa ; tÞ ¼ xha  Xa ¼ ðxaI ðtÞ  XaI ÞNI ðxa Þ inertial terms can be discretised using the consistent mass
I¼1 matrix:
X
Nn
Z
¼ uaI ðtÞNI ðxa Þ ða ¼ s; wÞ: ð23Þ
I¼1
MaIJ ðtÞ ¼ q a ðxa ; tÞNI ðxa ÞNJ ðxa Þ dv
Xa ðtÞ
The velocity field can be approximated as: Nap
X
X
Nn X
Nn
 q a ðxap ðtÞ; tÞNI ðxap ðtÞÞNJ ðxap ðtÞÞXap ðtÞ
v ha ðxa ; tÞ ¼ u_ ha ðxa ; tÞ ¼ u_ aI ðtÞNI ðxa Þ ¼ vaI ðtÞNI ðxa Þ ða ¼ s; wÞ ap¼1
I¼1 I¼1
Nap
X
ð24Þ ¼ map ðtÞNI ðxap ÞNJ ðxap Þ ða ¼ s; wÞ ð29Þ
and the acceleration field can be approximated as: ap¼1

X
Nn X
Nn
aha ðxa ; tÞ ¼ v_ ha ðxa ; tÞ ¼ v_ aI ðtÞNI ðxa Þ ¼ aaI ðtÞNI ðxa Þ ða ¼ s; wÞ: where the current material point mass of each phase is the same as
I¼1 I¼1 its initial mass (i.e. msp ðtÞ ¼ msp ð0Þ  msp and mwp ðtÞ ¼ mwp ð0Þ 
ð25Þ mwp ).
The nodal values of the internal forces can be written as:
The weak form of the momentum balance equation for the
N sp
X N wp
X
solid-water mixture is derived by multiplying (10) by a test func-
tion wh and then integrating over the current configurations Xks & Fint
sI ðtÞ   GIsp  r0sp ðtÞXsp ðtÞ þ GIwp  mpwp ðtÞXwp ðtÞ ð30Þ
sp¼1 wp¼1
Xkw . The term involving the stress is integrated by parts and the
divergence theorem is also applied. where simpler notations have been introduced for the gradient of
Z Z
the shape function (i.e. GIsp ¼ rNI ðxs Þjxs ¼xsp and GIwp ¼ rNI ðxw Þ
q s w h
ahs dv þ q w w  h
ahw dv
Xs ðtÞ Xw ðtÞ jxw ¼xwp ), for the effective stress (r0sp ðtÞ ¼ r0 ðxsp ðtÞ; tÞ), and for the pore
Z Z
water pressure (pwp ðtÞ ¼ pw ðxwp ðtÞ; tÞ), respectively.
¼ rwh : r0 dv þ wh  ss ds
Xs ðtÞ @ Xs ðtÞ The nodal values of the external forces (i.e. body forces and
Z Z
applied traction) can be written as:
þ rwh : pw dv  wh  p
 w ds
Xw ðtÞ @ Xp ðtÞ Nsp
X Nwp
X
Z Z
Fext
sI ðtÞ  msp ðtÞbsp ðtÞN I ðxsp Þ þ mwp ðtÞbwp ðtÞNI ðxwp Þ
þ q s wh  b dv þ q w wh  b dv ð26Þ sp¼1 wp¼1
Xs ðtÞ Xw ðtÞ Z Z
where ss ¼ r0  n is the prescribed traction that acts on the part of þ ss ðxs ; tÞNI ðxs Þ ds   w ðxw ; tÞNI ðxw Þ ds
p ð31Þ
@ Xs ðtÞ @ Xp ðtÞ
the solid skeleton phase boundary @ Xs ðtÞ; n is the unit normal to
the boundary, ds represents surface elements, and p  w ¼ mpw  n is where simpler notations have been introduced for the body force
the prescribed pore pressure that acts on the part of the water (i.e. bsp ðtÞ ¼ bðxsp ðtÞ; tÞ).
phase boundary @ Xp ðtÞ. It is assumed that the test function is zero The semidiscrete momentum balance equation for the mixture
where the displacement boundary conditions for solid and water can be obtained as:
phases (@ Xus and @ Xuw ) are prescribed.
X
Nn X
Nn
The test function is defined similar to the other fields as: MsIJ ðtÞasJ ðtÞ þ MwIJ ðtÞawJ ðtÞ ¼ Fint ext
sI ðtÞ þ FsI ðtÞ: ð32Þ
X
Nn J¼1 J¼1
wh ðxa Þ ¼ waI NI ðxa Þ ða ¼ s; wÞ: ð27Þ
I¼1 This equation is solved for the solid skeleton phase
acceleration at unconstrained nodes while using the water
Substituting (24), (25), and (27) into each component of (26):
phase acceleration, which can be obtained by solving the
X
Nn X
Nn X
Nn X
Nn
wsI  MsIJ ðtÞasJ ðtÞ þ wwI  MwIJ ðtÞawJ ðtÞ momentum balance equation for the water phase. This is
I¼1 J¼1 I¼1 J¼1 shown below.
X
Nn Z The weak form of the momentum balance equation for the
¼  wsI  rNI ðxs Þ  r0 ðxs ; tÞ dv water phase is derived by multiplying (9) by a test function wh
I¼1 Xs ðtÞ and then integrating over the current configuration Xkw as:
X
Nn Z
þ wwI  rNI ðxw Þ  mpw ðxw ; tÞ dv Z Z Z
I¼1 Xw ðtÞ q w wh  ahw dv ¼ nrwh : pw dv  wh  p
 w ds
X
Nn Z Xw ðtÞ Xw ðtÞ @ Xp ðtÞ
Z
þ wsI  ss ðxs ; tÞN I ðxs Þ ds n2 qw gwh
@ Xs ðtÞ   ðv hw  v hs Þ dv
I¼1
Xw ðtÞ k
X
Nn Z Z
 wwI   w ðxw ; tÞNI ðxw Þ ds
p þ q w wh  b dv : ð33Þ
I¼1 @ Xp ðtÞ Xw ðtÞ

X
Nn Z
The semidiscrete momentum balance equation for the water
þ wsI  q s ðxs ; tÞbðxs ; tÞNI ðxs Þ dv
I¼1 Xs ðtÞ phase can be obtained by considering the components of the test
X
Nn Z function, wwI are arbitrary except where the components of
þ wwI  q w ðxw ; tÞbðxw ; tÞNI ðxw Þ dv ð28Þ displacement and pressure are prescribed:
I¼1 Xw ðtÞ
 S. Bandara, K. Soga / Computers and Geotechnics 63 (2015) 199–214 205

X
Nn Nwp
X where Q kwIJ can be obtained from:
MwIJ ðtÞawJ ðtÞ ¼ nwp ðtÞGIwp  mpwp ðtÞXwp ðtÞ
Nwp
X
J¼1 wp¼1 mkwp nkwp g
Z Q kwIJ ¼ k
NI ðxkwp ÞNJ ðxkwp Þ: ð40Þ
  w ðxw ; tÞNI ðxw Þ ds
p wp¼1 kwp
@ Xp ðtÞ
N wp
X
þ mwp ðtÞbwp ðtÞNI ðxwp Þ 3.2.2. Solving for solid skeleton phase acceleration as
wp¼1 The momentum balance equation for the solid-water mixture
X
Nn can then be solved for aksI with using the solution akwI . Therefore
 Q wIJ ðtÞðv wJ ðtÞ  v sJ ðtÞÞ ð34Þ Eq. (32) can be written for an explicit scheme as:
J¼1
X
Nn
where the matrix Q wIJ ðtÞ is derived by considering the material MksI aksI ¼ Fk;int
I þ Fk;ext
I  MkwIJ akwJ ð41Þ
points as quadrature points, which can be written as: J¼1

Z where MksI is computed using:

nðxw ; tÞ2 qw ðxw ; tÞg
Q wIJ ðtÞ ¼ NI ðxw ÞNJ ðxw Þ dv
Xw ðtÞ kðxw ; tÞ N sp
X
N wp MksI ¼ msp NI ðxksp Þ: ð42Þ
X nðxwp ðtÞ; tÞq
 w ðxwp ðtÞ; tÞg sp¼1
 NI ðxwp ÞNJ ðxwp ÞXwp ðtÞ
wp¼1
kðx wp ðtÞ; tÞ
It should be noted that the consistent matrices should be con-
XN wp
mwp ðtÞnwp ðtÞg sidered (i.e. Q kwIJ and MkwIJ ) in the right hand side of Eqs. (39) and
¼ NI ðxwp ÞNJ ðxwp Þ: ð35Þ (43).
wp¼1
kwp ðtÞ
Eq. (41) can be expanded as:
Nsp
X N wp
X
3.2. Time integration MksI aksI ¼  GkIsp  r0k k
sp Xsp þ GkIwp  mpkwp Xkwp
sp¼1 wp¼1
The numerical solution of each momentum balance equation is N sp
X N wp
X
k k
obtained at a discrete set of times, t k ; k ¼ 1; . . . ; K. A discrete þ mksp bsp NI ðxksp Þ þ mkwp bwp NI ðxkwp Þ
approximation at time tk is indicated by a superscript k. In the sp¼1 wp¼1
Z Z
MPM formulation derived in this study, the background computa-
þ ss ðxs ; tÞNI ðxs Þ ds   w ðxw ; tÞNI ðxw Þ ds
p
tional grid is used as a finite element mesh with updated Lagrang- @ Xs ðtÞ @ Xp ðtÞ
ian frame with explicit time integration. A superscript L is used to X
Nn
represent the solution on the nodes at the end of the Lagrangian  MkwIJ akwJ : ð43Þ
step. Due to the usage of isoparametric elements in the updated J¼1
Lagrangian frame, it is not necessary to recompute the shape func-
tions during the timestep (i.e. N I ðxksp Þ ¼ N I ðxLsp Þ).
3.2.3. Solving for water and solid velocities and displacements
The nodal velocities at the beginning of the time step (i.e. v ksI &
3.2.1. Solving for water phase acceleration aw
The momentum balance equation for the water phase is solved
vkwI ) are needed for an explicit time integration of the nodal veloc-
ity. These velocities can be obtained by considering the momen-
for akwJ while considering a lumped mass matrix which is generally
tum at a node on the current background mesh from the
used in the explicit finite element methods. The computational
material points as:
time can be significantly reduced due to the use of lumped mass
matrix since it avoids matrix inversion at each time step. The N ap
X
lumped mass matrix is diagonal whose entries are the correspond- MkaI v kaI ¼ mkap v kap NI ðxkap Þ ða ¼ s; wÞ: ð44Þ
ing row sum of the consistent mass matrix. The lumped mass ap¼1

matrix for each phase can be derived as: The nodal velocities at the end of the Lagrangian step can be
Z N ap
X obtained as:
MaI ðtÞ ¼ q a ðxa ;tÞNaI ðxa Þdv  map NI ðxap Þ ða ¼ s;wÞ: ð36Þ
Xa ðtÞ ap¼1
vLaI ¼ v kaI þ Dt akaI ða ¼ s; wÞ ð45Þ

Using the above definition, Eq. (34) becomes: where Dt is the current time increment, Dt ¼ t kþ1  t k . Due to the
assumption of the Lagrangian grid, it is possible to consider that
MkwI akwI ¼ Fk;int k;ext
wI þ FwI þ Fk;mom
wI ð37Þ the grid nodes move in the computed velocity field and the material
points also move in this flow. Hence the movement of the material
where the lumped mass matrix for the water phase at a node at
points can be written as:
time t k can be computed using:
N wp X
Nn
X xkþ1 k
v LaI NI ðxkap Þ ða ¼ s; wÞ;
MkwI ¼ mwp NI ðxkwp Þ: ð38Þ ap ¼ xap þ Dt ð46aÞ
I¼1
wp¼1
X
Nn

Eq. (37) can be expanded as: vkþ1

ap ¼ v ap þ Dt
k
akaI NI ðxkap Þ ða ¼ s; wÞ: ð46bÞ
I¼1
Nwp
X Z
MkwI akwI ¼ nkwp GkIwp  mpkwp Xkwp   w ðxw ; tÞNI ðxw Þ ds
p
wp¼1 @ Xp ðtÞ 3.2.4. Computing solid skeleton’s porosity and permeability
Nwp
X X
Nn In the case of large deformations, the volume of solid skele-
k
þ mkwp bwp NI ðxkwp Þ  Q kwIJ ðv kwJ  v ksJ Þ; ð39Þ ton material points can be updated using the determinant (J)
kþ1
wp¼1 J¼1 of the deformation gradient Fsp (see Eq. (19a)) and can be
shown as:
206 S. Bandara, K. Soga / Computers and Geotechnics 63 (2015) 199–214

kþ1
Xsp ¼ J X0sp : ð47Þ 3.2.6. Effective stress calculation
The effective stress at solid skeleton material point in the case
The deformation gradient can be calculated using: of small deformations is updated using:
!
X
Nn
r0kþ1 ¼ r0k
kþ1
Fsp I þ Dt GkIsp v LsI Fksp ¼ DFkþ1 k sp þ T : Desp ð57Þ
¼ sp Fsp ð48Þ sp

I¼1
where T is the tangent modulus tensor. Any type of history depen-
where DFkþ1
sp is the increment of deformation gradient. v LsI is re- dent soil model can be used in the MPM formulation due to its
calculated using v kþ1
ap in a similar method as shown in (44) for Lagrangian nature. In the case of finite deformations T : Desp does
enhanced stability, without directly using the value obtained from not represent an objective stress rate. In order to achieve stress
(45). r r
kþ1 objectivity (i.e. r0 sp ¼ T : Desp where r0 is an objective stress rate),
The porosity of solid skeleton material point nsp can be sp

updated using J as: the Jaumann rate can be used to update the effective stress of solid
kþ1
skeleton material points as:
nsp ¼ 1  ð1  n0sp Þ=J: ð49Þ  
k k
r0kþ1
sp ¼ r0k 0k 0k
sp þ Dt rsp  Wsp  Wsp  rsp þ T : Desp ð58Þ
It is also necessary to compute the soil porosity at the location of
water material point (see Eq. (51b)). The volume of water material where Wksp is the vorticity which is defined as:
point at current time step (Xkwp ) can be obtained using the current
Nn h
X i
soil porosity value at the water material point location (nkwp ) as: T
Wksp ¼ GkIsp v LsI  ðGkIsp v LsI Þ : ð59Þ
Xkwp ¼ ðn0wp X0wp Þ=nkwp ð50Þ I¼1

where nkwp can be evaluated by adopting a weighted least squares

3.3. Soil model
approach as follows:
N sp
X Mohr–Coulomb model with strain hardening/softening is used
nksI ¼ mksp nksp NI ðxksp Þ=MksI ð51aÞ in this study. Following yield and potential functions are
sp¼1
considered:
X
Nn
nkwp ¼ nksI NI ðxkwp Þ: ð51bÞ 1 0 1
Fs ¼ ðr  r03 Þ  ðr01 þ r03 Þ sin /0 þ c0 cos /0 ð60aÞ
I¼1 2 1 2
The coefficient of soil permeability at water particle location s 1 0 1
G ¼ ðr1  r3 Þ  ðr01 þ r03 Þ sin w þ C
0
ð60bÞ
k k
(kwp ) can be calculated using ksp , similar to the nkwp calculation. 2 2
s s
The coefficient of soil permeability is generally dependent on sev- where F s is the yield function, G is the potential function (G corre-
eral factors: grain size distribution, grain shape, void ratio, and sponds to an non-associated flow rule), /0 is the internal friction
roughness of grains. In large deformation analysis, the variation angle of soil, w is the dilation angle of soil, c0 is the cohesion, and
of void ratio results in changes in the soil permeability. The C is a constant. r01 ; r02 ; r03 are the principal stresses where
Kozeny–Carman formula [29] can be used to modify the soil r01 6 r02 6 r03 relation is considered since compressive stresses are
permeability. The soil permeability (nksp ) can be written in terms assumed as negative.
of the current porosity as follows: In reality friction, dilation, and cohesion are dependent on den-
3 sity and mean pressure of soil (e.g. [30]). This study considers
k ðnksp Þ strain hardening/softening by varying the mobilised values of
ksp ¼ C 1 2
ð52Þ
ð1  nksp Þ /0 ; w, and c0 with respect to the total plastic deviatoric strain after
the onset of plastic yield. These values are varied in a piecewise-
where C 1 is a parameter that is determined experimentally. linear manner as shown in Fig. 4 with respect to the total plastic
deviatoric strain, which can be shown as follows for plane-strain
3.2.5. Pore pressure calculation conditions:
The pore pressure increment at water particle location is rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
derived using (15) and can be shown as: 2 3
cdev ;pl ¼ e2x;pl þ e2y;pl  ex;pl ey;pl þ e2xy;pl : ð61Þ
Kw h i 3 4
Dpwp ¼ Dt k nkwp e_ vs;wp þ ð1  nkwp Þe_ vw;wp ð53Þ
nwp In Fig. 4, a represents /0 ; w, or c0 . The values /0peak ; wpeak , and c0peak
represent peak values of friction angle, dilation angle and cohesion.
where e_ vs;wp is the volumetric strain rate of solid skeleton phase eval- /0crit , wcrit , and c0crit represent critical state values of friction angle,
uated at the water particle location and is calculated using:
! dilation angle and cohesion. cpeak
dev ;pl represents the plastic deviatoric
Nn h
X i
T strain at peak, and ccrit represents the plastic deviatoric strain
e_ vs;wp ¼ trðe_ s;wp Þ ¼ tr GkIwp v LsI þ ðGkIwp v LsI Þ ð54Þ dev ;pl

I¼1 when the softening or hardening is completed.

and e_ vw;wp is the volumetric strain rate of the water phase evaluated
3.4. Coupled MPM algorithm
at the water particle location and can be defined as:
!
Nn h
X i The coupled MPM algorithm is summarised in Fig. 5.
T
e_ vw;wp ¼ trðe_ w;wp Þ ¼ tr GkIwp v L
wI þ ðGkIwp v L
wI Þ : ð55Þ
I¼1
(i) The Information carried by the material points is projected
The pore water pressure is updated using (53) at the water on to the background mesh. The nodal mass for each phase
material point location as: is calculated using (42) and (38).
kþ1
(ii) The soil porosity at the water material point location is
pwp ¼ pkwp þ Dpwp : ð56Þ obtained using (51b). A similar set of equations as in
the nkwp calculation can be adopted to obtain the soil
 S. Bandara, K. Soga / Computers and Geotechnics 63 (2015) 199–214 207

(viii) At the end of the step, material points carry all the informa-
tion about the solution. The computational grid is initialised
for the next step.

3.5. Boundary conditions

The solid skeleton phase boundary is divided into two disjoint

pieces @ Xs ¼ @ Xus [ @ Xs where @ Xus \ @ Xs – 0 holds and the water
phase boundary is divided into two disjoint pieces @ Xw ¼ @ Xuw [
@ Xp where @ Xuw \ @ Xp – 0 holds.

3.5.1. Displacement boundary condition

The prescribed displacement boundaries @ Xus and @ Xuw , can be
applied using prescribed velocities at background nodes since the
displacements are related to the velocities (i.e. u_ s ¼ v sI for soil
and u_ w ¼ v wI for water). Hence the prescribed displacement
Fig. 4. Variation of mobilised friction angle, dilation angle, and cohesion with
(or velocity) boundaries can be defined for the solid skeleton phase
plastic deviatoric shear strain.
as:

vksI ¼ v sI ð62Þ
permeability at water material point location. The volume of and for the water phase as:
water material point is calculated using (50). The momen-
tum balance equation for the water phase is solved in the vkwI ¼ v wI ð63Þ
background mesh using (39). The boundary conditions for
In the case of complex boundary geometries, it is advisable to use a
akwI are applied if it is necessary.
background mesh with triangular elements, which is generally eas-
(iii) The momentum balance equation for the solid skeleton
ier to align with nodal boundaries. A mesh that moves with the
phase is solved using (43). The boundary conditions for aksI
boundary would be useful to represent complex moving boundaries
are applied if necessary.
as discussed in Jassim et al. [31].
(iv) The nodal velocities at the end of Lagrangian step are calcu-
A friction boundary condition is adopted in this study [32] for
lated using (45). The boundary conditions for the nodal
planar boundaries using Coulomb’s friction criterion. The friction
velocities are applied if it is necessary. The material point
boundary condition is applied on the mesh nodes by controlling
velocities and the material point positions are updated for
the nodal acceleration tangential to the planar boundary. Both sta-
both phases using (46a) and (46b).
tic and kinetic friction conditions are considered, and the friction is
(v) The strain increment Desp is calculated for the solid skeleton
applied only when the material points are in contact with the
phase and then the effective stress is updated using either
boundary (i.e. force or acceleration that is perpendicular to the
(57) or (58).
boundary surface should be negative).
(vi) The volumetric strain rates e_ vs;wp and e_ vw;wp are calculated at
the water material point locations and then the pore water
3.5.2. Prescribed traction and pressure boundary conditions
pressure increment is calculated using (53). The pore water
The prescribed traction on a moving boundary @ Xs is enforced
pressure of water material points is updated using (56).
in the solid skeleton phase using the concept of boundary layer
(vii) Solid skeleton material point volumes are updated using
that was proposed by [33]. The term related to the traction in
(47). The soil porosity and soil permeability are updated
(43) is therefore re-written as:
using (49) and (52) respectively.

Fig. 5. Coupled MPM algorithm.

208 S. Bandara, K. Soga / Computers and Geotechnics 63 (2015) 199–214
Z N sp
X is required when using a coarser mesh (in order to achieve better
ss ðxs ; tÞNI ðxs Þ ds  sksp h1 NI ðxksp ÞXksp
@ Xs ðtÞ sp¼1
where h is the thickness of the boundary layer. The boundary layer
is comprised of a layer of material points along the specific bound-
ary section where the traction has to be applied. These material
points are assigned with sksp at each time step while using constant
h that is equal to initial material point thickness. However, for com-
plex traction boundary geometries, more accurate results could be
obtained by using a thin element layer along the boundary surface
of the mesh that aligns with the boundary material points. This was
not considered in this study due to mesh complexities.
Application of the prescribed pressure boundary condition
requires special procedure since it is not possible to directly
impose pore pressure on @ Xp as p  w . This is due to the fact that
the pore pressure increment and pressure are computed at mate-
rial point locations at each time step and hence the most straight-
forward method of prescribing pore pressure at nodes cannot be
applied. In this study, a boundary layer is considered to prescribe
pore pressure on the boundary. This boundary layer is comprised
of boundary cells that are located in the boundary of the contin-
uum configuration of the water phase, and the pore pressure of
each material point inside those cells is assigned with correspond-
) in each time step. During
ing boundary pressures (i.e. pwp ¼ p
large deformations, different material points and different cells
can be identified as boundary points and cells. Application of this
boundary for horizontal and vertical surfaces are straightforward.
If the geometry of the water phase boundary changes with time,
an additional algorithm is considered to identify the boundary cells
and boundary material points that are located in the boundary of
the continuum configuration of the water phase in each time step.
3.5.3. Inflow and outflow boundary condition
Special boundary treatments are needed for inflow and outflow
boundary conditions (i.e. when water flows into and out from the
solid skeleton) due to the use of separate water material point
layer. As an example for inflow boundary, if there is a static pressure
head applied to the boundary (i.e. seepage boundary), the water
material points will move away from that boundary due to the pres-
sure gradient, leaving the cells with soil material points only. Hence
it is required to feed water material points for boundaries where
there can be any water inflow. In this study, additional water mate-
rial points have been fed into corresponding cells while assigning
same properties as the remaining material points in each cell.
An outflow boundary is needed to remove the water material
points that comes out from the soil body. This boundary is applied
when the water flow outside the soil body is not important in the
analysis. Additional algorithm is needed to specially treat the inter-
face between the pore water and free water [32] and this is not
included in this paper.
3.6. Initial conditions
The initial conditions for stresses (i.e. effective stress and pore
water pressure) for solid and water phases has to be applied at
the start of the computational cycle. It is advisable to start the sim-
ulation with correct initial values for stresses since it can take addi-
tional simulation time to reach stable values if the initial values are
incompatible or sometimes it can lead to erroneous results.
The initial number of material points (for soil and water) per
cell has to be selected according to the problem type (i.e. presence
of large deformations or not), required accuracy [18], and the com-
puting time. Based on authors’ experiences [18], 4 material points
per cell for each phase was selected in the numerical examples in
this paper. However, a higher number material points (i.e. 9 or 16)
ð64Þ results with less computing time) or when the problem involves
large deformations that occur very rapidly.
4. Stability of the numerical solution
In order to obtain a stable solution from the coupled MPM algo-
rithm, it is important to consider techniques to improve the
numerical stability. Following sections discuss some of the impor-
tant concepts.
4.1. Critical time step criterion
The Courant-Friedrichs-Levy condition (CFL) [34] is considered to
obtain the critical time interval in order to achieve a stable solution
in the explicit time integration of the couple MPM formulation. The
time increment is chosen to be less than the critical time increment
(Dtcrit ) which is obtained using:
Dcrit ¼ L=c ð65Þ
where L is the background cell length, and c is the pressure wave
velocity of the soil–water mixture which is defined as:
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
u 1
u n þ 1n þ3G 4
t Kw Ks
c¼ ð66Þ
ð1  nÞqs þ nqw
where K w is the bulk modulus of water, K s is the bulk modulus of
solid grains, and G is the shear modulus of solid skeleton.
4.2. Incompressibility constraint
In the case of constant volume conditions, the above mentioned
coupled MPM formulation can exhibit numerical instability due to
pressure oscillations. This type of pressure instability is widely
reported in the FE literature and the methods to overcome these
instabilities in the FEM can be also applied to MPM due to their
similarities. Incompressibility of material can occur: (a) in satu-
rated soil with low water compressibility and low soil permeabil-
ity, (b) when the loading rate is fast such that it creates an
undrained behaviour, and (c) in soils that undergo plastic deforma-
tions under constant volume or with quasi-incompressible elasticity.
Numerical stability in saturated soil is generally tested using the so-
called patch test for mixed formulations by Zienkiewicz et al. [35,36]
and the stability condition Ladyzhenskaya–Babuska–Brezzi (LBB or
inf-sup test) [37]. As suggested by Mira et al. [38], the satisfaction
of the first condition is a necessary but not a sufficient condition
for stability, but the second condition is a sufficient condition for sta-
bility. These conditions can be fulfilled by selecting displacement
interpolation to be one order higher than the pressure interpolation
(i.e. Taylor–hood elements). Numerical stability under plastic soil
deformations can be achieved by methods that remove volumetric
locking (i.e. selective reduced integration for low order elements
with elastic material, B-bar approach [39,22] for elasto-plastic mate-
rial, Simo-Rifai enhanced strain element [40]).
In this study, a simple B-bar approach is applied to the solid
skeleton phase in simulations that involve plastic deformations.
The strain increments at material point locations and the internal
forces at nodes are calculated by replacing the general B-matrix
(67) with the B-bar matrix (68) (plane strain configuration is
selected for simplicity).
2 @NI ðxsp Þ 3
@x
0
6 @N I ðxsp Þ 7
BIsp ¼6
4 0 @y
7
5 ð67Þ
@N I ðxsp Þ @N I ðxsp Þ
@y @x
 S. Bandara, K. Soga / Computers and Geotechnics 63 (2015) 199–214
2   
1 @N ðx Þ
I sp
þ @NI@xðxsc Þ 1 @NI ðxsc Þ

@N I ðxsp Þ
6 2  @x 
2

@y @y
7
6 7
 Isp
B ¼6
62
1 @N I ðxsc Þ
@x
@N ðx Þ
 I@x sp 1
2
@NI ðxsp Þ
@y
þ @NI ðxsc Þ 7
@y 7 ð68Þ
4 5
@N I ðxsp Þ @N I ðxsp Þ
@y @x
In Eq. (68), @N I ðxsc Þ=@x is the gradient of shape function evalu-
ated at the centre of the cell for x direction. Reduced integration
method is not considered in this study since it creates a similar
situation as in the FEM when all the gradients are calculated at
element centre that represents single Gauss point approach in
the FEM, and this mechanism leads to hourglassing or zero energy
mode in low order elements. For elastic material, selective reduced
integration method can be used to avoid volumetric locking.
When considering the water phase, the interpolation order for
pressure is reduced by replacing GIwp with gradients of shape func-
tions evaluated at the centre of the cell (i.e. GIwp;c ¼ rN I ðxw Þjxw ¼xc ,
where subscript c represents the centre of cell in which the water
material point is located). Hence the strain increments and pore
water pressure increment are calculated using GIwp;c . This approach
is quite similar to the use of Q 1  P 0 quadrilateral element for the
Stokes equations, where a continuous piecewise bilinear approxi-
mation for the velocity and discontinuous piecewise constant
approximation for the pressure are assumed. Although this ele-
ment is widely used in practise, it is an unstable element that
sometimes undergoes spurious pressure modes that are mesh
dependent [41,42]. However, in the current MPM formulation this
approach has given stable pressure distributions for large deforma-
tion problems, apart from the cases of completely undrained con-
ditions [32]. For fully incompressible problems other stabilisation
methods that are generally used in FEM can be adopted with an
additional computational cost. Most of these techniques have been
proposed for equal order interpolation for both displacement and
pressure for low order elements [43,23]. As future study, some of
the important formulations such as fractional step algorithms
[44,45] and characteristic-based split algorithms [46] can be
extended to the present MPM formulations considering water
material point locations as integration points.
4.3. Numerical damping
In the case of dynamic problems in soils, the damping intro-
duced by the plastic behaviour of the material and the viscous
effects of the fluid flow are sufficient to damp out any non-physical
or numerical oscillation [25]. However, if the problem consists of
very small plastic behaviour of soil in the low-strain region, or if
the material behaviour is assumed to be purely elastic when there
is a sudden load application, damping is helpful to remove the ini-
tial oscillations. In these circumstances, a damping term is consid-
ered in the momentum equilibrium equation of the mixture as
follows:
qas þ Cvs ¼ r  r þ qb
where C is the damping constant.
4.4. Treatment of pore pressure gradient
Pore pressure oscillations have been observed during this study
when modelling large deformation problems if there is a spatial
variation of soil porosity and when there is an imbalance of volume
of water in adjacent cells (when water material points cross cell
boundaries).
This problem mainly occurs when solving the momentum bal-
ance equation of the water phase. The method of calculation of
the pressure gradient has a certain effect on giving erroneous
results in MPM formulation when soil porosities vary spatially. In
209
3
the momentum balance equation of the water phase (see Eq. (9)),
the pressure gradient term is written as nrpw where n is the soil
porosity. This term should be used for immiscible mixtures and
to avoid fluid flow for homogeneous fluid pressure (in contrary
the term rðnpw Þ results in fluid flow for homogeneous fluid pres-
sure which is not valid for pore water) [28].
When considering the weak form of the momentum balance
equation in the MPM formulation, the term nrpw is written as:
Nwp
X
Fint
wI ¼ GIwp  mpwp ðtÞX00
wp ð70Þ
wp¼1
where X00wp is the intrinsic volume of a water material point. If there
is an imbalance of total intrinsic water volumes in adjacent cells, Eq.
(70) will result in water flow for homogeneous fluid pressure. This
represents a similar situation that occurs due to the term rðnpw Þ
and it is purely due to the nature of the discretised form.
In order to avoid errors that occur due to volume imbalances,
the term rpw should be calculated without considering the effect
of different water volumes in cells. In this study, a simple approach
is implemented by first calculating the term rpw considering the
total cell volume as:
N cell
X
Fint
cellI ¼ GIcell;c  mpw;cell ðtÞX0cell ð71Þ
cell¼1
where Ncell represents the number of cells, GIcell;c represents the gra-
dients of shape functions evaluated at the centre of the cell, pw;cell ðtÞ
represents the averaged water pressure at each cell (i.e. dividing the
total pore water pressure from all the water material points by the
total number of water material points in each cell), and X0cell repre-
sents the volume of a cell. Fint
cellI avoids the errors associated with dif-
ferent volumes of water in adjacent cells. Eq. (71) is then multiplied
by a factor (awI ) that is defined at nodes to consider the term n,
since Fint
cellI only provides the term rpw . awI is computed as the ratio
of nodal interpolated intrinsic water volume and the nodal interpo-
lated components of the cell volumes that contain water material
points, which is shown as:
PNwp
NI ðxkwp ÞX00
wp
awI ¼ PNcell wp¼1 ð72Þ
k
cell¼1 N I ðxcell;c Þð X0cell =4Þ
where NI ðxkcell;c Þ represents the shape function of cell centre. Finally
the corrected Fint wI is obtained as:
Fint int
wI ¼ awI FcellI : ð73Þ
5. Numerical examples
In this section, the developed coupled MPM code is evaluated
using a number of examples. First, the formulation is verified using
ð69Þ
analytical solutions that exist in the literature: one-dimensional
consolidation test to check the Terzaghi’s theory and two-dimen-
sional consolidation with Cryer’s sphere to check Mandel–Cryer
theory. Finally, the developed method is applied to model levee
failure simulation to demonstrate the practical application of the
method.
5.1. One dimensional consolidation
Terzaghi’s one dimensional consolidation theory [47] is used to
validate the coupled code to check the behaviour under consolida-
tion. The theory is valid for a soil layer composed of an isotropic,
homogeneous and fully saturated elastic material under small
strain conditions.
210 S. Bandara, K. Soga / Computers and Geotechnics 63 (2015) 199–214
The one dimensional consolidation test was numerically simu-
lated using the coupled MPM for a plane strain model which is
shown in Fig. 6(a). A 1.0 m high and 0.06 m width soil column, fully
saturated with water is considered with 600 soil material points
and 600 water material points. The background grid consists of
cells with 0.02 m in length and 4 material points per each cell. Soil
behaviour is modelled using isotropic linear elastic material and
the following material properties are considered: Young’s modu-
lus, E ¼ 1:0  107 Pa; Poisson’s ratio, m ¼ 0:3; solid grain density,
qs ¼ 2650 kg/m3; initial porosity n0 ¼ 0:3; initial permeability
0
k ¼ 1:0  103 m/s; water density, qw ¼ 1000 kg/m3; bulk modu-
lus of water K w ¼ 2:2 GPa. The simulation is carried out using a
time increment, Dt ¼ 1:0  107 s for a total time duration of
ttot ¼ 2:5 s. Zero initial pore pressure and zero effective stresses
are assigned as the initial values and the simulation is carried
out considering zero gravitational acceleration. The consolidation
test is simulated by applying a 10 kPa traction to the top most
material point layer (i.e. traction boundary layer) while allowing
drainage to occur from the top surface (i.e. set zero pore pressure
for water material points in the top layer of the column).
The numerical results from the consolidation test are compared
with the analytical solution in Fig. 6(b) in which the pore pressure
variation with height at different time factors (i.e. T v ¼ cv t=d
where cv is the coefficient of consolidation and d is the soil layer
thickness) are plotted. The results from the coupled MPM matches
very well with the analytical solution.
5.2. Cryer’s problem
The Mandel–Cryer problem [48,49] is used to verify the accu-
racy of the coupled MPM code in two-dimension. The Mandel–
Cryer effect captures the non-monotonic pore pressure behaviour
that can be only achieved using a coupled solution such as three-
dimensional Biot’s theory, and not from an uncoupled solution
such as Terzaghi’s one-dimensional consolidation theory. The the-
ory was first described by Mandel [48] for a fully saturated, isotro-
pic poroelastic specimen of infinitely long rectangular plate
sandwiched between two rigid, impermeable plates. Later, Cryer
[49] presented the theory for an isotropic poroelastic sphere con-
solidating under hydrostatic pressure. Both formulations show that
the induced pore pressure at the centre of the specimen becomes
higher than the applied pressure in the saturated sample. The
Cryer’s sphere problem is selected to check the accuracy of the cou-
pled MPM formulation in this section. An axisymmetric coupled
MPM code was developed to simulate the sphere problem. The
axisymmetric MPM formulation was derived to solve coupled
Fig. 6. One-dimensional consolidation test: (a) schematic diagram of the test; (b)
comparison of pore pressure profiles from coupled MPM with Terzaghi’s solution.
analysis based on the axisymmetric solid phase MPM formulation
that was proposed by Sulsky and Schreyer [50].
In the Cryer’s sphere problem, a fully saturated, isotropic poro-
elastic sphere is subjected to an uniform normal traction P0 at the
surface. The pore fluid can drain freely from the surface of the
sphere. The pore pressure at the centre of the sphere first attains
P0 , and then continues to rise to a peak before gradually decaying
to zero. When the fluid drains out of the surface of the sphere, a lar-
ger portion of the applied load is transferred towards effectively
stiffer central region of the sphere by the contraction around the
surface causing pore pressure rise in the centre of the sphere
[51]. Cryer [49] presented the solution for the pore pressure for dif-
ferent Poisson’s ratios. Mason et al. [52] further derived the analyt-
ical solution for the strain tensor, total stress tensor, and the
effective stress tensor.
The numerical solution to the Cryer’s sphere problem is
obtained using the coupled axisymmetric MPM formulation for
the model shown in Fig. 7(a). Using symmetry, only half of the
sphere is modelled in the cylindrical coordinate system with coor-
dinates (r; h; z), while considering h ¼ 1 radian and radius,
R ¼ 0:5 m. The initial pore pressure of the sphere is set to zero
and then a sudden normal traction of 10 kPa is applied to the sur-
face material points of the sphere, while allowing drainage to occur
from the outer surface. An impermeable boundary condition is pro-
vided along the symmetry line of the sphere. Four case studies
were performed for Poisson’s ratios = 0.1, 0.2, 0.3, and 0.45 while
assuming zero gravity. The simulation was carried out using 988
soil material points and 988 water material points with a
background grid of 4 material points per cell and 0.04 m cell
length. Isotropic linear elastic material is considered for soil behav-
iour. Material properties of the simulation are as follows:
0
E ¼ 1:0  107 Pa, qs ¼ 2143 kg=m3 ; n0 ¼ 0:3; k ¼ 1:0  103 m/s,
3
qw ¼ 1000 kg/m , and K w ¼ 2:2 GPa. The time increment of the
simulation is 1:0  107 s and the total time is 0.2 s.
Results from the coupled MPM analysis were obtained using
selective reduced integration for the solid skeleton phase and using
cell centre integration for water phase. Some numerical damping
was also considered to avoid pressure oscillations in the beginning
of the simulations due to the existence of purely elastic behaviour
in the small strain region and the sudden application of the load.
Fig. 7(b) shows the variation of pore pressure inside the sphere
when t ¼ 0:01 s (or dimensionless time T ¼ 0:234, where
T ¼ cv t=R2 ) during which the pore pressure in the centre of the
sphere rises above the applied pressure of 10 kPa. Fig. 7(c) com-
pares the pore pressure variation from the coupled MPM simula-
tions with Cryer’s analytical solution for different Poisson’s ratios
and the results are in good agreement with the analytical results.
However, some discrepancies in the pore pressure were noticed
closer to the surface of the sphere during the initial stage of the
simulation and this may be due to the close proximity to the drain-
age boundary and the loading surface. This behaviour may be over-
come by locally refining the mesh specially near the draining
boundary [23] or by using a sophisticated stabilisation technique
as discussed in Section 4.2.
5.3. Failure of a levee
In this example, two simulations of levee failure were per-
formed to demonstrate the robustness of the proposed numerical
method for practical geotechnical problems. The schematic dia-
gram of the model is shown in Fig. 8. The simulations were per-
formed using the Mohr–Coulomb material model (Section 3.3). In
the first simulation, the soil is assumed as dry above the phreatic
surface since the proposed formulation can represent only the
dry and the fully saturated regions using material point layers.
However, in reality there exists an unsaturated region above the
 S. Bandara, K. Soga / Computers and Geotechnics 63 (2015) 199–214
Fig. 7. Cryer’s sphere problem: (a) numerical model details; (b) pressure variation inside the sphere at T = 0.234 for
pressure time histories at centre of sphere for different Poisson’s ratios.
phreatic surface and this region can have a higher strength due to
matric suction. Additionally, this unsaturated region can consist of
a capillary cohesion which exists due to the suction induced hard-
ening occurring at the microscopic particle contact level and its
sudden removal due to wetting (sudden reduction of void ratio
due to wetting that occurs as a result of reduction of suction or
removal of capillary bonds) can results in collapsible behaviour
in soil. Different types of techniques are available in the literature
to capture the collapsible behaviour in the constitutive equations
[53,54]. Although, this behaviour cannot be directly modelled
using the present formulation, a simple approach is considered to
roughly capture this behaviour and to show this method has the
capability of modelling different mechanisms related large defor-
mation that are generally difficult to model with FEM. Hence an
additional simulation was carried out considering the soil material
points that are located above the phreatic surface are unsaturated
and with some capillary cohesion. This will be discussed further in
the remaining part of this section.
Both simulations were performed using 7200 soil material
points and 4974 water material points with a background grid of
4 material points per cell and 0.1 m cell length. The time increment
of both simulations is 1:0  105 s. Material properties for water
are qw ¼ 1000 kg/m3, and K w ¼ 1:0 GPa. Mohr–Coulomb material
model with following common material properties is considered
for both simulations: E ¼ 1:0  107 Pa, m ¼ 0:3; qs ¼ 2700 kg=m3 ;
0
n0 ¼ 0:46, k ¼ 5:0  104 m/s, and classical cohesion c0 ¼ 0 Pa.
For soil layer, vertical movement is constrained at bottom bound-
ary and horizontal movement is constrained at left hand side
boundary where there is an inflow boundary. For water layer,
bottom boundary is an impermeable boundary (i.e. water velocity
normal to the bottom boundary is zero) and left hand side bound-
ary is an inflow boundary (i.e. feeding water material points) with
a constant static pressure head. A Zero pressure boundary layer is
Fig. 8. Schematic of the MPM model for levee failure problem.
211
m ¼ 0:3; and (c) comparison of dimensionless pore
considered along the predefined phreatic surface for the water
layer. However, it is necessary to consider the unsaturated perme-
ability and matric suction variations above the phreatic surface in
order to capture the unsaturated seepage characteristics including
the location of the phreatic surface, and this was not considered in
this study. The friction coefficient of the bottom boundary is equal
to 0.3.
The first simulation was performed using the following con-
stant values for mobilised internal friction angle, dilation angle
and cohesion: /0mob ¼ 30:8 , w0mob ¼ 0 ; c0mob ¼ 0 Pa. The failure pat-
tern and the profiles of pore water pressures and vertical effective
stresses for the first simulation with dry soil above the phreatic
surface are shown in Fig. 9. When considering the failure mecha-
nisms, a shear band develops at shallow location from the slope
surface with a circular shape and then propagates into the levee.
The levee undergoes a smoother soil movement with smoother
changes of pore water pressures and effective stresses and this
shows the capability of the method to simulate moderate
deformations.
The second simulation was carried out to demonstrate the capa-
bility of the proposed method to handle complicated mechanisms.
In this simulation, an unsaturated region is considered above the
phreatic surface with a saturation ratio (Sw ) of 0.82 with a matric
suction (s) value of 2.7 kPa. A simple methodology is adopted in
the coupled MPM formulation to model the unsaturated soil
behaviour using the Bishop’s effective stress concept. The material
points that lie above the phreatic surface are considered as par-
tially saturated instead of dry by considering: (1) the total stress
in the momentum balance equation of the mixture (only soil mate-
rial points are contributing above the phreatic surface) as
r ¼ r0 þ Sw sm, where each soil material point knows their r0 ; Sw ,
and s values; (2) an increased mass due to water (msp;u ¼ msp þ
n0 Sw qw X0sp ). When a partially saturated soil material point
becomes fully saturated due to seepage flow (i.e. cell is occupied
with water material points), s and Sw are set to zero and the dry
mass is considered afterwords. The initial effective stresses in the
unsaturated region are obtained considering the effect due to mat-
ric suction. In the unsaturated region, capillary cohesion (c0u ) is
considered in addition to c0 (considered as zero) to represent the
effect of suction-induced hardening occurring at the microscopic
particle contact level [55] with the following values: peak value
of capillary cohesion c0peak u ¼ 2 kPa when total plastic deviatoric
strain (cdev ;pl ) is less than cpeak
dev ;pl ¼ 0:01, linear variation of capillary
cohesion from c0peak u to critical state capillary cohesion c0crit
u ¼ 0 kPa
212 S. Bandara, K. Soga / Computers and Geotechnics 63 (2015) 199–214

Fig. 9. Failure of initially dry levee: (a) deviatoric shear strain variation; (b) pore water pressure variation (Pa); and (c) vertical effective stress variation (Pa).

Fig. 10. Levee failure with very large deformations: (a) deviatoric shear strain variation; (b) pore water pressure variation (Pa); and (c) vertical effective stress variation (Pa).

when cdev ;pl varies between cpeak crit

dev ;pl and cdev ;pl ¼ 0:2. cu is set to zero
0
in soil material points when its cell is occupied with water material
points. A higher value for c0peak
u is considered to create discontinu-
ities and to check the effectiveness of the method with discontinu-
ities (tension cracks). Additionally, a negative dilation angle (w) is
considered to model contractile soil since this will allow us to
model very large deformations because the associated reduction
of volumetric strains of soil in the saturated region will generate
very high excess pore pressures (i.e. if the soil is unable to drain
when its volume attempts to contract, excess pore pressure devel-
ops) and this will result in reduction of mean effective stress (soil
state reaches an instability where the shear strength can suddenly
reduces to zero), which is quite similar to the behaviour as in the
cases of static liquefaction and instability in loose granular soil that
tend to contract upon shearing and this behaviour can happen
during the movement [56,57]. As similar to the variation of c0u ,
the values for w are selected as follows: wpeak ¼ 5 and wcrit ¼ 0 .
Fig. 10 shows the shear strain, pore pressure and vertical effec-
tive stress profiles for the second simulation. Initial shear failure
has resulted in generating very high excess pore water pressures
along the shear bands, thus resulting in nearly zero mean effective
stresses in the saturated region. Initial shear band develops at a
deep location from the slope surface and more shear bands develop
with time and spread towards the levee thus resulting in a rapid
progressive failure. The unsaturated soil that lies above the satu-
rated region undergoes discontinuous failure mechanisms and
the resulting soil blocks tend to flow above the saturated region
thus resulting in higher excess pore pressures and more instabili-
ties. After the rapid movement finished, the excess pore pressures
tend to dissipate and the effective stresses slowly increase as
shown in Fig. 10 at t ¼ 5 s, which represents the consolidation
stage. This simulation has shown the capability of the current for-
mulation to capture the failure, post-failure, and deposition stages
when modelling large deformations that includes discontinuities.
 S. Bandara, K. Soga / Computers and Geotechnics 63 (2015) 199–214
Better results for the soil behaviour could be obtained by using
more advanced soil constitutive models. Although the inclusion of
advanced soil models into MPM is straightforward, it will reduce
the computational efficiency due to high computing time that
requires when solving advanced soil models at each time step
when using explicit time integration of MPM. Additionally, the
results could be improved by using a basis function that is C 1 con-
tinuous at cell boundaries as in GIMP method [12], in order to
reduce the integration errors that occur with piecewise linear basis
functions when material points cross cell boundaries. However,
this approach needs higher computing time due to extended sup-
port region of the basis function and requires ghost nodes beyond
the boundary to sustain partition of unity property.
6. Conclusions
A numerical formulation that is based on the material point
method to analyse dynamic, coupled hydro-mechanical problems
in geomechanics that involve large deformations, has been pre-
sented in this paper. Unique feature of the developed method is
the consideration of two sets of Lagrangian material point layers
to represent solid skeleton layer and pore-water layer in satu-
rated deforming porous media, and this allows us to: (a) conserve
mass of water in the system that is important in some applica-
tions; (b) consider the acceleration of water that is useful when
applying the method for both low and high frequency problems;
and (c) extend the method to model soil–water interaction prob-
lems (i.e. submarine landslides). The derivation of the formulation
and the ways to achieve stable numerical solution have been
described in the paper. The method was first validated by com-
paring results with those predicted by analytical solutions. The
method was then applied to model levee failure problem that
involve large deformations while using a strain-softening Mohr–
Coulomb model. From this study, the following conclusions can
be drawn:
The presented coupled MPM method is able to model large
deformation problems in geomechanics that involve coupling
of soil deformations with pore water pressures and can be
applied to low and high frequency problems.
The results from levee failure analysis showed that the method
was able to model large deformation behaviour, which is diffi-
cult to achieve in conventional finite element analysis. The large
deformation analysis of the levee suggests that the presented
method has much potential to understand the complex hydro-
mechanical behaviour in large deformations problems.
The reduced order pressure interpolation technique presented
in this paper has shown stable for large deformations. An
extensive parametric study of elasto-plastic cases is needed to
identify the limitation of the current MPM formulation on
fully undrained problems and to find a better stabilisation
method.
Additional modifications are needed when modelling problems
that involve unsaturated soil (i.e. unsaturated soil mechanics
plays an important role on the failure initiation) and a simple
modification has been applied in this study to consider this
region. However, it is important to consider another material
point layer for gas phase or a variation of saturation ratio of soil
material points to represent unsaturated region with a suitable
soil constitutive model.
The presented formulation needs additional modifications/
improvements when modelling post-failure behaviour in the
cases of increased water content after the failure (i.e. mixing
with water bodies in the flowing paths as in debris flows) that
results in highly fluidised flows.
213
Conflict of interest
There is no conflict of interest.
Acknowledgement
The first author would like to thank the Cambridge Common-
wealth Trust for the financial support to pursue this research.
References
[1] Hungr O. A model for the runout analysis of rapid flow slides, debris flows, and
avalanches. Can Geotech J 1995;32:610–23.
[2] Gingold RA, Monaghan JJ. Smoothed particle hydrodynamics: theory and
application to non-spherical stars. Mon Not R Astron Soc 1977;181:399–406.
[3] Monaghan JJ. Simulating free surface flows with SPH. J Comput Phys
1994;110(2):399–406.
[4] Sulsky DL, Zhou SJ, Schreyer HL. Application of a particle-in-cell method to
solid mechanics. Comput Phys Commun 1995;87:236–52.
[5] Onate E, Idelsohn SR, Zienkiewicz OC, Taylor RL, Sacco C. A stabilized finite
point method for analysis of fluid mechanics problems. Comput Methods Appl
Mech Eng 1996;139(1–4):315–46.
[6] Belytschko T, Lu YY. Element free Galerkin method. Int J Numer Methods
1994;37:229–56.
[7] Brackbill JU, Kothe DB, Ruppel HM, low-dissipation FLIP: a. particle-in-cell
method for fluid flow. Comput Phys Commun 1988;48(1):25–38.
[8] Liu GR, Liu MB. Smoothed particle hydrodynamics. New Jersey: World
Scientific; 2003.
[9] Monaghan JJ, Gingold RA. Shock simulation by the particle method sph. J
Comput Phys 1983;52:374–89.
[10] Monaghan JJ. On the problem of penetration in particle methods. J Comput
Phys 1983;82:1–15.
[11] Sulsky DL, Schreyer HL, Peterson K, Kwok R, Coon M. Using the material-point
method to model sea ice dynamics. J Geophys Res 2007;112(C2):C02S90.
[12] Bardenhagen SG, Kober EM. The generalized interpolation material point
method. Comput Model Eng Sci 2004;5(6):477–96.
[13] Andersen SM. Material-point analysis of large-strain problems: modelling of
landslides, Ph.D. thesis. Aalborg University; 2009.
[14] Beuth L, Benz T, Vermeer PA, Wieckowski Z. Large deformation analysis using a
quasi-static material point method. J Theor Appl Mech 2007;38:45–60.
[15] Coetzee CJ, Vermeer PA, Basson AH. The modelling of anchors using the
material point method. Int J Numer Anal Meth Geomech 2005;29(9):879–95.
[16] Alonso E, Zabala F. Progressive failure of Aznalcóllar dam using the material
point method. Geotechnique 2011;61(9):795–808.
[17] Jassim I, Stolle D, Vermeer P. Two-phase dynamic analysis by material point
method. Int J Numer Anal Meth Geomech 2013;37:2502–22.
[18] Abe K, Soga K, Bandara S. Material point method for coupled hydromechanical
problems. J Geotech Geoenviron Eng 2014;140(3):04013033.
[19] Zhang HW, Wang KP, Chen Z. Material point method for dynamic analysis of
saturated porous media under external contact/impact of solid bodies. Comput
Methods Appl Mech Eng 2009;198(17-20):1456–72.
[20] Shin W. Numerical simulation of landslides and debris flows using an enhanced
material point method, Ph.D. thesis. University of Washington; 2009.
[21] Mackenzie-Helnwein P, Arduino P, Shin W, Mooney MA, Miller GR. Modeling
strategies for multiphase drag interactions using the material point method.
Int J Numer Meth Eng 2010;83(3):295–322.
[22] Hughes TJR. The finite element method: linear static and dynamic finite
element analysis. New Jersey: Prentice Hall; 1987.
[23] Preisig M, Prevost JH. Stabilization procedures in coupled poromechanics
problems: a critical assessment. Int J Numer Anal Methods Geomech
2010;35:1207–25.
[24] Li C, Borja RI, Regueiro RA. Dynamics of porous media at finite strain. Comput
Methods Appl Mech Eng 2004;193(36-38):3837–70.
[25] Zienkiewicz O, Chan A, Pastor M, Schrefler B, Shiomi T. Computational
geomechanics with special reference to earthquake engineering. New
York: John Wiley & Sons; 1999.
[26] Ehlers W, Graf T, Ammann M. Deformation and localisation analysis of
partially saturated soil. Comput Methods Appl Mech Eng 2004;193:2885–910.
[27] Lewis RW, Schrefler BA. The finite element method in the static and dynamic
deformation and consolidation of porous media. 2nd ed. New York: John Wiley
& Sons; 1998.
[28] Drumheller DS. On theories for reacting immiscible mixtures. Int J Eng Sci
2000;38(3):347–82.
[29] Carrier WD. Goodbye, Hazen; Hello, Kozeny–Carman. J Geotech Geoenviron
Eng 2003;129(11):1054–6.
[30] Bolton MD. The strength and dilatancy of sands. Geotechnique
1986;36(1):65–78.
[31] Jassim I, Coetzee CJ, Vermeer PA. A dynamic material point method for
geomechanics. In: Installation effects in geotechnical engineering; 2013. p. 15–23.
[32] Bandara SS. Material point method to simulate large deformation problems in
fluid-saturated granular medium, Ph.D. thesis. University of Cambridge; 2013.
[33] Chen Z, Brannon R. An evaluation of the material point method, Tech. rep.
Albuquerque, New Mexico: Sandia National Laboratories; 2002.
214 S. Bandara, K. Soga / Computers and Geotechnics 63 (2015) 199–214
[34] Courant R, Friedrichs K, Lewy H. On the partial difference equations of
mathematical physics. IBM J Res Dev 1967;11:215–34.
[35] Zienkiewicz O, Qu S, Taylor RL, Nakazawa S. The patch test for mixed
formulations. Int J Numer Methods Eng 1986;23:1873–83.
[36] Zienkiewicz OC, Taylor RL. The finite element method. 1st ed. McGraw-Hill;
1989.
[37] Brezzi F, Fortin M. Mixed and hybrid finite element methods. New
York: Springer Verlag; 1987.
[38] Mira P, Pastor M, Li T, Liu X. A new stabilized enhanced strain element with
equal order of interpolation for soil consolidation problems. Comput Methods
Appl Mech Eng 2003;192(37-38):4257–77.
[39] Hughes TJR, Winget J. Finite rotation effects in numerical integration of rate
constitutive equations arriving in large deformation analysis. Int J Numer Anal
Meth Geomech 1980;15:1862–7.
[40] Simo JC, Rifai MS. A class of mixed assumed strain methods and the method of
incompatible modes. Int J Numer Meth Eng 1990;29:1595–638.
[41] Brezzi F, Fortin M. Mixed and hybrid finite element methods. New
York: Springer Verlag; 1991.
[42] Boffi D, Brezzi F, Fortin M. Mixed finite element methods and
applications. Berlin: Springer Verlag; 2013.
[43] Huang M, Yue ZQ, Tham LG, Zienkiewicz OC. On the stable finite element
procedures for dynamic problems of saturated porous media. Int J Numer
Meth Eng 2004;61(9):1421–50.
[44] Li X, Han X, Pastor M. An iterative stabilized fractional step algorithm for finite
element analysis in saturated soil dynamics. Comput Methods Appl Mech Eng
2003;192(35-36):3845–59.
[45] Blanc T, Pastor M. A stabilized Fractional Step, RungeKutta Taylor SPH
algorithm for coupled problems in geomechanics. Comput Methods Appl
Mech Eng 2012;221–222:41–53.
[46] Salomoni VA, Schrefler BA. A CBS-type stabilizing algorithm for the
consolidation of saturated porous media. Int J Numer Meth Eng 2005;63(4):
502–27.
[47] Terzaghi K. Theoretical soil mechanics. New York: John Wiley; 1943.
[48] Mandel J. Consolidation des sols (étude mathématique). Geotechnique
1953;3:287–99.
[49] Cryer CW. A Comparison of the three-dimensional consolidation theories of
Biot and Terzaghi. Q J Mech Appl Math 1963;16(4):401–12.
[50] Sulsky DL, Schreyer HL. Axisymmetric form of the material point method with
applications to upsetting and Taylor impact problems. Comput Methods Appl
Mech Eng 1996;139(1-4):409–29.
[51] Cheng A, Abousleiman YN, Detournay E, Cui L, Roegiers J. Mandel’s problem
revisited. Geotechnique 1996;46(2):187–95.
[52] Mason D, Solomon A, Nicolaysen L. Evolution of stress and strain during the
consolidation of a fluid-saturated porous elastic sphere. J Appl Phys
1991;70:4724.
[53] Alonso EE, Gens A, Josa A. A constitutive model for partially saturated soils.
Geotechnique 1990;40(3):405–30.
[54] Nuth M, Laloui L. Effective stress concept in unsaturated soils: clarification and
validation of a unified framework. Int J Numer Anal Meth Geomech
2008;32(7):771–801.
[55] Robert D, Soga K. Soil pipeline interaction in unsaturated soils. In: Laloui L,
editor. Mechanics of Unsaturated Geomaterials. Hoboken, NJ USA: John Wiley
& Sons Inc.; 2013. p. 303–25 [chapter 13].
[56] Eckersley D. Instrumented laboratory flowslides. Geotechnique 1990;40(3):
489–502.
[57] Olson SM, Stark TD, Walton WH, Castro G. 1907 static liquefaction flow failure
of the north dike of Wachusett dam. J Geotech Geoenviron Eng
2000;126(12):1184–93.