Computers and Geotechnics 172 (2024) 106424

Contents lists available at ScienceDirect

Computers and Geotechnics

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

Research Paper

A nonlocal material point method for the simulation of large deformation

problems in brittle soils
José L. González Acosta a ,∗, Miguel A. Mánica b , Philip J. Vardon c , Michael A. Hicks c ,
Antonio Gens d
a
Geosciences and Material Technology, Energy and Materials Transition, TNO, Utrecht, The Netherlands
b Institute of Engineering, National Autonomous University of Mexico, Mexico City, Mexico
c Faculty of Civil Engineering and Geosciences, Delft University of Technology, Delft, The Netherlands
d Department of Civil and Environmental Engineering, Universitat Politècnica de Catalunya, Barcelona Tech - CIMNE, Barcelona, Spain

ARTICLE INFO ABSTRACT

Keywords: This paper investigates the implementation of a nonlocal regularisation of the material point method to
Brittle soil mitigate mesh-dependency issues for the simulation of large deformation problems in brittle soils. The adopted
Material point method constitutive description corresponds to a simple elastoplastic model with nonlinear strain softening. A number
Mesh dependency
of benchmark simulations, assuming static and dynamic conditions, were performed to show the importance
Nonlocal regularisation
of regularisation, as well as to assess the performance and robustness of the implemented nonlocal approach.
Stress oscillations
The relevance of addressing stress oscillation issues, due to material points crossing element boundaries,
is also demonstrated. The obtained results provide relevant insights into brittle materials undergoing large
deformations within the MPM framework.

1. Introduction advantages, it also shares many of its limitations, such as volumetric

locking, which often occurs during the simulation of materials with
In the field of geotechnical engineering, particularly in the study low compressibility, or the hourglass effect, which is linked to the
of large deformation problems, the material point method (MPM) has utilisation of low-order elements. Similarly, it is well-known that the
emerged as a reliable technique for generating accurate and objective simulation of the softening response in brittle materials leads to non-
results. Established from the same principles as the finite element objective FEM results with a pathological dependence on the adopted
method (FEM), MPM inherits many of the mathematical features of mesh (Sluys and de Borst, 1992; de Borst et al., 1993; Mánica et al.,
FEM, and facilitates the computation of large deformations by adding 2018) and numerical convergence issues (Summersgill et al., 2017;
an additional step in the solution process. The standard MPM com- Mánica et al., 2022a; Cui et al., 2023). Deformation in simulations of
putational steps are depicted in Fig. 1. While Fig. 1a and b show brittle materials localises in shear bands with a thickness of a single
the standard FEM solution steps (that is, nodal solution and mesh element and, therefore, vanishing energy dissipation is attained as the
deformation), Fig. 1c illustrates the fundamental MPM step, in which size of the elements is reduced (Bažant and Pijaudier-Cabot, 1988),
the numerical mesh is reset to its original position while the material
which is physically inadmissible. The source of these numerical diffi-
points remain in their last location. Subsequently, an allocation pro-
culties is the absence of an internal length scale in classical continuum
cedure is implemented to establish the new location of each material
formulations representing the microstructure of the material, which in
point relative to the undeformed mesh (indicated with bold dashed
reality controls the size of the localisation region (Desrues and Viggiani,
lines). Extensive research has substantiated the versatility of MPM in
2004).
simulating a diverse array of problems, including, but not limited to,
Solutions have been developed to address issues related to volu-
soil–structure interaction (González Acosta et al., 2021a; Lian et al.,
metric locking (Coombs et al., 2018; González Acosta et al., 2019) or
2011; Phuong et al., 2016), the triggering and post-failure behaviour of
landslides (González Acosta et al., 2021b; Troncone et al., 2022; Yerro the hourglass effect (Zhang et al., 2017) within the MPM framework.
et al., 2019), earthquake phenomena (Kohler et al., 2022), CPT pene- Regarding mesh dependency issues, different approaches have been
tration (Martinelli and Galavi, 2021), and unsaturated behaviour (Yerro proposed and implemented in conventional FEM formulations, usually
et al., 2022). However, just as MPM possesses many of the FEM known as regularisation techniques, such as enriched continua with

∗ Corresponding author.
E-mail address: leon.gonzalezacosta@tno.nl (J.L. González Acosta).

https://doi.org/10.1016/j.compgeo.2024.106424
Received 21 February 2024; Received in revised form 20 April 2024; Accepted 10 May 2024
0266-352X/© 2024 Elsevier Ltd. All rights are reserved, including those for text and data mining, AI training, and similar technologies.
J.L. González Acosta et al. Computers and Geotechnics 172 (2024) 106424

Fig. 1. Solution steps in the MPM.

microstructure (Tejchman and Bauer, 2005; Khoei and Karimi, 2008), functions interpolation, such as δ𝐮 = 𝐍δ𝐮 and δ𝛜 = 𝐁δ𝐮, where 𝐍
gradient models (Anand et al., 2012; Frédéric et al., 2021), or nonlocal and 𝐁 are the shape functions (SFs) and strain–displacement matrices,
models (Galavi and Schweiger, 2009; Summersgill et al., 2017; Mánica respectively, and an incremental formulation (i.e. 𝝈 = 𝝈 0 + 𝐃𝛜), the
et al., 2018; Cui et al., 2023). However, their implementation in MPM following element equation is obtained:
formulations remains limited. Some authors have demonstrated that [𝑛mp ]
∑ ( ) ( )
nonlocal regularisation in MPM can have a significant impact on the 𝐁T 𝐱𝑝 𝐃𝑝 𝐁 𝐱𝑝 |𝐉|W𝑝 𝛥𝐮 =
obtained results (Burghardt et al., 2012; Goodarzi and Rouainia, 2017). 𝑖=1
(3)
Nevertheless, a comprehensive study on the use of nonlocal models in 𝑛
∑
mp
( ) ∑
mp 𝑏
( ) ∑
mp
( )
𝑛
MPM is still missing in the literature. 𝜌𝑝 𝐍T 𝐱𝑝 𝐛|𝐉|W𝑝 + 𝐍T 𝐱𝑝 τ𝛤 − 𝐁T 𝐱𝑝 𝝈|𝐉|W𝑝
In this paper, an investigation is conducted regarding the imple- 𝑖=1 𝑚𝑝=1 𝑖=1

mentation of a nonlocal regularisation into the MPM framework, for which can be expressed in matrix form as
the simulation of large deformation problems in quasi-brittle materials. |
𝐊𝛥𝐮||elem = 𝐅ext ||elem − 𝐅int | (4)
Initially, the mathematical framework of the adopted MPM formulation |elem
is described. Subsequently, the considered strain-softening elastoplastic where 𝐱𝑝 is the position of the material point 𝑝, 𝑛mp and 𝑏mp are the
model for the constitutive description of quasi-brittle geomaterials is number of material points and boundary material points 𝑚𝑝 in the
presented, as well as its nonlocal extension within the MPM formu- computational domain, |𝐉| is the Jacobian determinant, 𝐛 are the body
lation. Finally, a series of simple boundary value problems (BVP) are forces, W𝑝 is the material point weight, |elem indicates integration of all
analysed to demonstrate the effect and importance of the implemented elements, and 𝛥𝐮 is the vector of incremental displacements as a func-
nonlocal regularisation for the simulation of softening materials in the tion of the initial and final displacement (i.e. 𝛥𝐮 = 𝐮new − 𝐮old , where
MPM. 𝐮new and 𝐮old are the initial and final displacements, respectively). Note
that, at the start of each iteration, 𝐮old is equal to zero, thus 𝛥𝐮 = 𝐮new .
2. MPM formulation Then, to derive the dynamic time-dependent formulation, D’Alembert’s
principle and Newmark (1959) time-scheme methods are assumed.
2.1. Conventional MPM Through D’Alembert’s principle, inertial forces are incorporated into
#»
Eq. (2) as part of the body forces, that is 𝜌( 𝐛 − #» 𝐚 ) where 𝐚 is the
The exploration of the advantages of using a nonlocal formulation vector of accelerations. Newmark’s time-dependent formulation, on the
is performed by examining a range of static and dynamic problems. other hand, captures the progression of displacements, velocities, and
Although both formulations have been broadly elaborated by numerous accelerations from time 𝑡 to 𝑡 + 𝛥𝑡 as
authors, such as Beuth et al. (2011), Guilkey and Weiss (2003) and [ ]
𝑡+𝛥𝑡 𝑡 𝑡 𝑡+𝛥𝑡
Wang et al. (2016), they are delineated here for completeness. For the 𝐯 = 𝐯 + 0.5 𝐚 + 𝐚 𝛥𝑡 (5)
dynamic MPM formulation, the implicit framework is used. Based on [ ]
𝑡+𝛥𝑡 𝑡 𝑡 𝑡 𝑡+𝛥𝑡
the principle of virtual work (i.e. 𝛱 = Wint − Wext ), the equation of 𝐮 = 𝐮 + 𝐯 𝛥𝑡 + 0.25 𝐚 + 𝐚 𝛥𝑡2 (6)
implicit equilibrium is 𝑡+𝛥𝑡 𝑡+𝛥𝑡 𝑡+𝛥𝑡
where 𝐮 , 𝐯 and 𝐚 denote the vectors of nodal displacements,
#»
𝐮T #»
1
𝛱= 𝛜T 𝐃𝛜dV − 𝐮T 𝜌 𝐛 dV − τ d𝛤 = 0 (1) velocities, and accelerations at time 𝑡 + 𝛥𝑡 respectively. From Eqs. (5)
2 ∫V ∫V ∫𝛤
and (6), the velocity and acceleration at time 𝑡 + 𝛥𝑡 are
where 𝐃 is the stiffness matrix, 𝐛 is the vector of body forces, 𝐮 and 𝑡
𝑡+𝛥𝑡 2𝐮 𝑡
𝛜 are the actual displacement and strain fields describing the state of 𝐯 = −𝐯 (7)
𝛥𝑡
the system, 𝜌 is the material density, τ is the vector of boundary loads,
𝑡 𝑡
and V and 𝛤 are the body volume and boundary, respectively. Then, 𝑡+𝛥𝑡4𝐮 4𝐯 𝑡
𝐚 = − −𝐚 . (8)
by following standard finite element discretisation procedures (Bathe, 𝛥𝑡2 𝛥𝑡
1995), and by evaluating δ𝛱 = 0 (i.e. the first variation of the potential Finally, by substituting Eq. (8) into Eq. (2) and employing both
energy is equal to zero for any admissible virtual displacement) with D’Alembert’s principle and Newton–Raphson iteration procedure, the
respect to the displacements, the equation of equilibrium can be written equilibrium equation at the computational step 𝑘 is written as
as 𝑘 ( )𝑡+𝛥𝑡
#» 𝐊 𝛥𝐮 =(𝑘−1) 𝐅ext − 𝐅kin − 𝐅int (9)
δ𝛜T 𝐃𝛜dV = + δ𝐮T 𝜌 𝐛 dV + δ𝐮T #»
τ d𝛤 = 0 (2)
∫V ∫V ∫𝛤 where
where δ𝛜 and δ𝐮 represent the virtual displacement and virtual strain, ( 𝑡)
4𝐦
respectively, and are used as test functions. Then, by employing shape 𝐊 = 𝐊𝑡 + (10)
𝛥𝑡2

2
J.L. González Acosta et al. Computers and Geotechnics 172 (2024) 106424

( ) where 𝛺 is the problem domain, 𝛺𝑝 is the material point support

𝑡 𝑡
𝑡 4(𝑘−1) 𝐮 4𝐯 𝑡 domain, and 𝜒 is the function delimiting the area of influence of the
𝐅kin,𝑡+𝛥𝑡 = 𝐦 − −𝐚 (11)
𝛥𝑡2 𝛥𝑡 material point, presented originally as
{
∑
𝑚𝑝
( ) ( ) 1 if x ∉ 𝛺𝑝
𝐦= 𝜌𝑝 𝐍T 𝐱𝑝 𝐍 𝐱𝑝 |𝐉|W𝑝 (12) 𝜒𝑝 (x𝑝 ) = (18)
𝑖=1 0 otherwise
The local version of the original GIMP SFs, on the other hand, is
2.2. Enhanced MPM created by computing the convolution of the original 𝜒 function and
single-element SFs. This can be observed in Fig. 2a where two elements
Whilst the standard MPM framework can analyse a range of large (E1 and E2 ) share a central node and the linear SFs N𝑖 . When the
deformation problems, it is well known that issues, such as stress convolution of the 𝜒 function, of length 2lp, and the SFs are computed,
oscillation owing to element crossings and the use of linear-based the standard GIMP SF (S𝑖𝑝 ) are obtained (Fig. 2b). In contrast, when the
SFs, can substantially compromise the accuracy of the simulation. This convolution is computed with a single SF (Fig. 2c), the local GIMP SFs
is particularly problematic when advanced constitutive descriptions (S𝑖𝑝∗ ) are obtained (Fig. 2d). Therefore, by following similar steps as in
are employed, in which the inaccuracies in stresses might lead to the deduction of Eq. (15), the same equation can be derived but using
convergence problems and the failure of the simulation. To account instead the local GIMP SF, thus resulting in
( )
for stress oscillation problems, an enhanced version of the MPM is ∑𝑛𝑛 ∑
𝑐𝑚𝑝
∗
considered, namely DM-GC MPM (González Acosta et al., 2020). DM- 𝐃𝑔 = 𝐍𝑖 (x𝑔 ) 𝐒𝑖𝑝∗ (x𝑝 )𝐃𝑝 W . (19)
𝑖=1 𝑝=1
GC MPM is a methodology in which a double mapping (DM) procedure
is combined with two stress recovery techniques: the generalised in- Stiffness integration via Eq. (19) has shown to be far superior in
terpolation material point (GIMP) method and the composite MPM contrast to the use of linear SFs or by introducing a mapping step
(González Acosta et al., 2017). In the DM procedure, the material using linear SFs (e.g. as proposed in Eq. (15)). Evidence of the va-
point stiffness contribution is mapped to the Gauss point locations lidity of this approach can be found in González Acosta et al. (2017)
using shape functions (via the nodes) for the subsequent global stiffness and González Acosta (2020).
integration. This is achieved using standard finite element SFs as
3. Nonlocal constitutive model for quasi-brittle materials
∑
𝑐𝑚𝑝
𝐃𝑖 = 𝐍𝑖 (x𝑝 )𝐃𝑝 W∗ (13)
The nonlocal elastoplastic constitutive model described in Mánica
𝑝=1
(2018) was considered here for its implementation in the described
where 𝐃𝑖 is the elastic matrix at the node 𝑖, 𝑐𝑚𝑝 is the current number MPM formulation. The model is intended for the simulation of argilla-
of material points 𝑝 in the element, and W∗ is the (dimensionless) ceous hard soils/weak rocks. Therefore, it incorporates many of the
𝑜𝑚𝑝
modified material point weight (W∗ = W𝑝 𝑐𝑚𝑝 , where 𝑜𝑚𝑝 is the original behavioural features of these materials, such as a nonlinear yield
number of material points in the element). After the total stiffness criterion, a non-associated flow rule, stiffness and strength anisotropy,
contribution of the material points is accumulated at the nodes, it is rate-dependency, strain-softening, creep, and the ability to simulate
then redistributed to the original Gauss positions as objectively localised deformations by means of a nonlocal regularisa-
∑
𝑛𝑛 tion. The model has been previously implemented in the finite element
𝐃𝑔 = 𝐍𝑖 (x𝑔 )𝐃𝑖 (14) code Plaxis (Bentley Systems, 2022), and its capacity to produce mesh-
𝑖=1 independent results has been demonstrated previously (Mánica et al.,
where 𝐃𝑔 is the elastic matrix at the Gauss point 𝑔, 𝐍𝑖 (xg ) is the nodal 2018, 2020, 2022a,b). However, for the sake of simplicity, most of
SF evaluated at the Gauss points, and 𝑛𝑛 is the number of nodes of the features of the model were not employed here. Therefore, as
the element. Note that the location of the Gauss points is the same as done in Mánica et al. (2020) and Romero et al. (2024), the adopted
in the original FEM for a 2 × 2 integration, which is (±𝜚, ±𝜂) where constitutive description is reduced to a simple nonlocal elastoplastic
model with isotropic linear elasticity and nonlinear strain-softening.
𝜚 = 𝜂 = √1 . Finally, by substituting Eq. (13) into Eq. (14), the stiffness
3 The yield function corresponds to the following hyperbolic approxi-
at the Gauss integration point is
mation of the Mohr–Coulomb (MC) criterion (Gens et al., 1990), which
( )
∑
𝑛𝑛 ∑
𝑐𝑚𝑝 allows the tensile strength to be limited, that is usually overestimated
𝐃𝑔 = 𝐍𝑖 (x𝑔 ) 𝐍𝑖 (x𝑝 )𝐃𝑝 W∗ (15) by the classical MC criterion:
𝑖=1 𝑝=1 √
𝐽2 ( )2 ( )
which can be integrated into Eq. (3) by replacing 𝐃𝑝 , thus guaranteeing 𝑓= + 𝑐 ∗ + 𝑝𝑡 tan 𝜙∗ − 𝑐 ∗ + 𝑝 tan 𝜙∗ (20)
a smoother distribution of the stiffness within the group of elements 𝑓𝑑 (𝜃)
activated by the material points. Although the incorporation of Eq. (15) where 𝑐 ∗ and 𝜙∗ are the asymptotic cohesion and friction angle, re-
into Eq. (3) improves significantly the stiffness distribution within the spectively, 𝑝𝑡 is the isotropic tensile strength, 𝑓𝑑 (𝜃) is a given function
activated elements, the consequences of using standard FE SFs (which defining the shape of the criterion in the deviatoric plane, and 𝑝, 𝐽 , and
extend only over a single element) hinder the effectiveness of this 𝜃 are stress invariants with their usual definitions:
technique. To overcome this limitation, local GIMP SFs (Charlton et al., 1
𝑝 = tr(𝝈) (21)
2017) can be incorporated into the equation. Local GIMP SFs are a 3
variation of the original GIMP SFs proposed by Bardenhagen and Kober 1 ( )
(2004) to reduce oscillations derived from material points crossing 𝐽2 = tr 𝐬2 (22)
2
element boundaries. These functions are constructed by performing a ( √ )
convolution of standard FE SFs and a material point support domain, 1 3 3 det 𝐬
𝜃 = − sin−1 (23)
3 3∕2
leading to the following set of equations: 2𝐽 2

1 where 𝐬 is the deviatoric stress tensor, defined as 𝐬 = 𝝈 − 𝑝𝐈, and 𝐈 is the

𝐒𝑖𝑝 = 𝜒 (x )𝐍 (x ) dx (16)
V𝑝 ∫𝛺𝑝 ∩𝛺 𝑝 𝑝 𝑖 𝑝 identity tensor. The shape of the yield function in the deviatoric plane
is defined as van Eekelen (1980):
1
∇𝐒𝑖𝑝 = 𝜒 (x )∇𝐍𝑖 (x𝑝 ) dx (17)
V𝑝 ∫𝛺𝑝 ∩𝛺 𝑝 𝑝 𝑓𝑑 (𝜃) = 𝛼 (1 + 𝛽 sin 3𝜃)𝑛 (24)

3
J.L. González Acosta et al. Computers and Geotechnics 172 (2024) 106424

Fig. 2. (a) Nodal FE SF and convolution with the material point support domain, (b) original GIMP SF (S𝑖𝑝 ), (c) nodal FE SF and convolution with the material point support
domain in a single element, and (d) local GIMP SF S𝑖𝑝∗ .

where 𝛽 = 0.85𝛼 1∕2 , 𝑛 = −0.229, and 𝛼 = 0.972 were employed. distance between them ‖𝐱 − 𝝃‖ is assumed, and 𝑤 is defined in the
Strain-softening is considered, driven by the evolution of the following normalised form to prevent modifying a uniform field:
strength parameters, and it is characterised by the following exponen- 𝑤0 (‖𝐱 − 𝝃‖)
tial decay functions: 𝑤 (𝐱, 𝝃) = 𝑑𝜻 . (31)
∫V 𝑤0 (‖𝐱 − 𝜻‖)
( )[ ( p )]
tan 𝜙∗ = tan 𝜙∗peak − tan 𝜙∗peak − tan 𝜙∗res 1 − exp −𝑏res ϵeq (25) Different nonlocal models are obtained depending on the variable,
( ) or variables, that are assumed nonlocal. In the context of nonlocal
( p ) ( p )
𝑐 ∗ = 𝑐peak
∗ ∗
− 𝑐post ∗
exp −𝑏post ϵeq + 𝑐post exp −𝑏res ϵeq (26) plasticity models, different alternatives have been studied. For instance,
stress or elastic strains (Eringen, 1981), total strains (Eringen, 1983),
( ) ( p ) ( p ) or plastic strains (Bažant and Lin, 1988) have been considered as the
𝑝t = 𝑝t peak − 𝑝t post exp −𝑏post ϵeq + 𝑝t post exp −𝑏res ϵeq (27)
nonlocal fields. However, under certain conditions, these formulations
where the subscripts peak, post, and res refer to peak, post-rupture, might exhibit unwanted effects such as stress looking, vanishing energy
and residual conditions, respectively, 𝑏post and 𝑏res are parameters dissipation, or localisation into a zone of vanishing volume (Bažant
p
controlling the rate of softening, and ϵeq is a state variable defined here and Jirásek, 2002). Improved results have been obtained by assuming
as as nonlocal the scalar state variable controlling the softening pro-
p ( )1∕2 cess (Galavi and Schweiger, 2010; Summersgill et al., 2017; Mánica
ϵeq = 𝛜p ∶ 𝛜p (28) et al., 2018; Singh et al., 2021). Therefore, the nonlocal extension of
p
where 𝛜p is the plastic strain tensor. A distinction is made in Eqs. (25) the described elastoplastic model (Section 3) is obtained replacing ϵeq
to (27) between two stages of the softening process. The first one is by the following nonlocal counterpart:
related to a rapid degradation and breakage of interparticle bonds, p p
ϵ̄ eq (𝐱) = 𝑤 (𝐱, 𝝃) ϵeq (𝝃) 𝑑𝝃 . (32)
i.e. a reduction of cohesion and tensile strength, up to post-rupture ∫𝑉
conditions (Burland, 1990), whose rate is defined by 𝑏post . In the second Furthermore, a Gaussian function has been historically employed to
one, with a slower rate 𝑏res , the remaining cohesion and tensile strength characterise 𝑤0 (see e.g. Bažant and Pijaudier-Cabot, 1988). However,
vanish and the friction angle decreases as a result of the polishing and as has been extensively demonstrated (Mánica et al., 2018; Summersgill
reorientation of particles in the formed failure surface (Gens, 2013). et al., 2017; Monforte et al., 2019; Singh et al., 2021; Gao et al., 2022),
Regarding the direction of plastic flow, a non-associated flow rule an enhanced performance is obtained with the following function
is considered, which is obtained by scaling the volumetric component proposed by Galavi and Schweiger (2010):
of the yield function in the following way: [ ( )2 ]
‖𝐱 − 𝝃‖ ‖𝐱 − 𝝃‖
𝜕𝑔 𝜕𝑓 𝜕𝑝 𝜕𝑓 𝜕𝐽2 𝜕𝑓 𝜕𝜃 𝑤0 = exp − (33)
=𝜔 + + (29) 𝑙s 𝑙s
𝜕𝝈 𝜕𝑝 𝜕𝝈 𝜕𝐽2 𝜕𝝈 𝜕𝜃 𝜕𝝈
where 𝑔 is the plastic potential function and 𝜔 is a parameter control- where 𝑙s is a parameter controlling the spread of the function with
ling the amount of plastic volumetric deformation. respect to ‖𝐱−𝝃‖, which in turn introduces a length scale to the material
behaviour controlling the size of the localisation region. The particular
3.1. Nonlocal extension shape resulting from Eq. (33) (Fig. 3) prevents the concentration of
plastic deformations along the forming shear band, similarly to the
In general, a nonlocal model is one where the behaviour of a given over-nonlocal approach (Vermeer and Brinkgreve, 1994).
point in the material depends not only on its state, but it also depends The described nonlocal regularisation is implemented within the
on the state of neighbouring points. If 𝐹 (𝐱) is some given local field MPM framework and, therefore, Eq. (31) and (32) are replaced by the
within a body of volume V, its nonlocal version 𝐹̄ (𝐱) can be written as following discrete versions:
𝑛mpl
p
∑ p
ϵ̄ eq 𝑝 = 𝑤𝑝𝑗 ϵeq 𝑗 (34)
𝐹̄ (𝐱) = 𝑤 (𝐱, 𝝃) 𝐹 (𝝃) 𝑑𝝃 (30) 𝑗=1
∫V ( )
𝑤0 ‖𝐱𝑝 − 𝐱𝑗 ‖
where 𝑤 is a weighting factor controlling the relative importance 𝑤𝑝𝑗 = (35)
𝑛mpl
of neighbouring points as a function of its position (𝝃) relative to ∑ ( )
𝑤0 ‖𝐱𝑝 − 𝐱𝑘 ‖
the position of the point under consideration (𝐱). Usually, the radial
𝑘=1

4
J.L. González Acosta et al. Computers and Geotechnics 172 (2024) 106424

Fig. 3. Weighting function in the nonlocal approach.

where 𝑛mpl is the number of material points considered in the com-

Fig. 4. Search area for the identification of neighbouring points for the computation
putation of the nonlocal state variable. Theoretically, 𝑛mpl should be
of the nonlocal state variable.
equal to the total number of material points in the simulation, since
the weights from Eq. (33) only vanish at an infinite radial distance.
However, the latter is generally computationally prohibitive and the
effect of neighbouring points at distances greater than 2𝑙s is negligible. constitutive description; they are presented in the following sections.
Therefore, as suggested by Galavi and Schweiger (2010), only material All simulations were performed using squared four-noded plane strain
points within an interaction radius of 2𝑙s are considered. However, elements, with four material points evenly distributed inside the el-
unlike conventional FEM formulations, the material points, where the ements at the beginning of the analysis. It is important to mention
constitutive equations are integrated, do not have a definite position, that although the conventional MPM formulation (Section 2.1) was
and they can move significantly and change elements during the sim- also tested, it was incapable of completing any analysis when softening
ulation. Therefore, for each material point, neighbouring points within was considered in the constitutive description. Numerical convergence
the corresponding interaction radius must be continually identified for issues are well-known in conventional FEM simulations when dealing
the computation of the nonlocal state variable (Eq. (34)). with softening materials (Summersgill et al., 2017; Mánica et al.,
The task of searching for neighbouring points, which is closely 2022a; Cui et al., 2023). Mathematically, this occurs because the
related to the nearest-neighbour search problem (Knuth, 1997), is governing partial differential equation changes locally from elliptic
very computationally intensive if linear search is performed, i.e. by to hyperbolic, causing an ill-posed BVP (Read and Hegemier, 1984;
computing the radial distance from the query point to every other point Benallal and Marigo, 2007; Lu et al., 2012). These issues are accen-
in the simulation. However, advantage can be taken from two facts tuated in the standard MPM due to the well-known stress oscillations
inherent to the MPM: (1) that the background mesh is constructed as a resulting from the movement of material points from one element
regular grid and (2) that tracking of the material points within a given to another (Tielen et al., 2017; González Acosta, 2020), preventing
element is already performed. Therefore, searching for neighbouring convergence to a steady condition as soon as softening has occurred,
points can be limited to the square area shown in Fig. 4, which always at the onset of plastic deformations. Therefore, in the following, re-
contains the interaction radius. For a given material point, within an sults labelled as ‘‘regularised MPM’’ and ‘‘standard MPM’’ refer to the
element E [𝑖, 𝑗], all elements possibly containing material points within enhanced MPM formulation (Section 2.2), with and without nonlocal
the interaction radius can be found in E [𝛼, 𝛽], where 𝛼 = 𝑖 − 𝑛ele , … , 𝑖 + regularisation, respectively.
𝑛ele , 𝛽 = 𝑗 − 𝑛ele , … , 𝑗 + 𝑛ele , and 𝑛ele is the number of elements in each
direction of the search area (Fig. 4), computed as 4.1. Biaxial test
( )
2𝑙s
𝑛ele = f loor +1 (36)
𝑙ele The first benchmark corresponds to the simulation of the simple
where 𝑙ele is the length of elements in the background mesh and f loor() biaxial test shown in Fig. 5. This analysis assumes static conditions
is a function accounting for the integer part. (Eq. (4)) and, therefore, inertial forces are neglected. Due to symmetry,
The search for neighbouring points and the computation of the only a quarter of the sample was represented in the analysis, with
nonlocal state variable is performed at the beginning of each global a height ℎ = 0.05 m and a width 𝑤 = 0.03 m. Regarding boundary
step, and the nonlocal variable is assumed to remain constant over the conditions, at the left and bottom boundaries displacements were fixed
step (Rolshoven, 2003). This results in an efficient and simple algorithm only in the normal direction, representing the symmetry axes. At the
rendering reasonable computation times. right and top boundaries, a constant confinement pressure 𝜎3 = 100 kPa
was applied. This value was also considered as the initial isotropic stress
4. Benchmark simulations state assigned to all material points. Loading was applied by means
of a prescribed downward vertical displacement at the top boundary,
A number of two-dimensional (2D) plane strain simulations were applied in fixed increments of 𝛥𝛿𝑦 = 1.0 × 10−5 m up to a total dis-
performed to assess the effectiveness of the nonlocal regularisation placement 𝛿𝑦 = 0.00226 m. Horizontal displacements were fixed at the
implemented in the MPM, for both the static (Eq. (4)) and dynamic top boundary in order to induce a non-homogeneous stress/strain field
(Eq. (9)) formulations, when softening behaviour is considered in the and, therefore, favour the onset of localisation. Table 1 summarises the

5
J.L. González Acosta et al. Computers and Geotechnics 172 (2024) 106424

Table 1
Parameters used in the benchmark simulations performed.
Parameter Symbol Units Biaxial test Column collapse Foundation
Young’s modulus 𝐸 kPa 20,000 10,000 50,000
Poisson’s ratio 𝜈 – 0.2 0.35 0.45
Initial asymptotic friction angle 𝜙∗ini ◦
20 20 20
Peak asymptotic friction angle 𝜙∗peak ◦ 20 20 20
Residual friction angle 𝜙∗res ◦ 15 15 15
∗
Initial asymptotic cohesion 𝑐ini kPa 200 100 100
∗
Post-rupture asymptotic cohesion 𝑐post kPa 0 0 0
Initial tensile strength 𝑝𝑡 ini kPa 0 0 0
Post-rupture tensile strength 𝑝𝑡 post kPa 0 0 0
Post-rupture softening rate 𝑏post – 10 5 5
Residual softening rate 𝑏res – 2 2 2
Non-associativity constant 𝜔 – 1 1 1
Density 𝜌 g/cm3 – 1.6 –
Length scale parameter 𝑙s m 0.01a 0.4a 0.8a
a
Only applies for the regularised MPM.

Fig. 5. Geometry and boundary conditions of the simulated biaxial test.

Fig. 7. Contours of deviatoric strains and position of material points from the biaxial
test simulation obtained with the standard MPM using element sizes 𝑙ele of (a) 0.0033,
Fig. 6. Load vs. displacement curves from the biaxial test simulation obtained with (b) 0.002, (c) 0.0014, and (d) 0.001 m.
the standard (dashed lines) and regularised MPM (solid lines) and for different element
sizes 𝑙ele .

the size of elements is reduced. For the regularised MPM simulations,

a length scale parameter 𝑙s = 0.01 m was employed. Therefore, for
adopted parameters, which are similar to those employed in Mánica all meshes, the recommendation of 𝑙s ≥ 𝑙ele is fulfilled (Galavi and
et al. (2018). Schweiger, 2010) so that a sufficient number of material points are con-
Analyses of the biaxial test were performed with the standard sidered in the nonlocal averaging. Unlike the standard MPM, practically
and regularised MPM and for different element sizes of the back- a single curve is obtained regardless of the size of elements employed.
ground mesh. Specifically, element sizes 𝑙ele of 0.0033, 0.002, 0.0014, Only the curve with 𝑙ele = 0.0033 m lies slightly above the rest. In this
and 0.001 m were considered. Since a non-homogeneous stress/strain regard, it is important to mention that the recommendation from Galavi
field is generated, results are assessed in terms of global measures. and Schweiger (2010) about the minimum element size for a given
Fig. 6 shows the obtained deviator load vs. prescribed vertical dis- 𝑙s was derived for 15-noded triangular elements with fourth-order
placement curves. The results from the standard MPM show typical interpolation and 12 integration points. In the case of the four-noded
mesh-dependent behaviour, with an increasingly brittle response as linear elements used here, with four material points per element, the

6
J.L. González Acosta et al. Computers and Geotechnics 172 (2024) 106424

Fig. 10. Geometry and boundary conditions of the vertical cut simulation.

is reduced. On the other hand, contours from the regularised MPM

(Fig. 8) show quite similar magnitudes of deviatoric strains and a
similar thickness of the formed shear band, which is approximately
equal to 𝑙s , as previously reported for regularised FEM formulations
using the same weighting function as Eq. (33) (Galavi and Schweiger,
2010; Mánica et al., 2018).
The parameter 𝑙s controls the size of the localisation region, in-
troducing a length scale to the material behaviour. If 𝑙s is reduced,
a smaller thickness of the shear band is obtained, in turn resulting
in a more brittle response of the BVP, as shown in Fig. 9. Ideally,
𝑙s should be selected to obtain a width of the localisation process
similar to that occurring in the real material, which depends on its mi-
crostructure (Desrues and Viggiani, 2004). However, in most cases, this
Fig. 8. Contours of deviatoric strains and position of material points from the biaxial shear zone can be very small, resulting in an excessively refined mesh,
test simulation obtained with the regularised MPM using element sizes 𝑙ele of (a) 0.0033, exceeding conventional computational capacities, when solving practi-
(b) 0.002, (c) 0.0014, and (d) 0.001 m and a length scale parameter 𝑙s = 0.01 m. cal BVPs. To overcome this issue, the softening scaling technique can
be applied (Pietruszczak and Mroz, 1981; Brinkgreve, 1994; Marcher,
2003; Galavi and Schweiger, 2010; Mánica et al., 2018). It assumes
that the effect of the real shear zone can be merged into a numerical
shear band of a larger size, in accordance with our computational
resources. Since the global post-localisation behaviour will depend on
both the length scale parameter and the constitutive softening rate,
they must be chosen jointly to represent a given material. However,
as demonstrated by Romero et al. (2024), the latter is not a trivial
task, and the derivation of parameters for a regularised simulation from
conventional laboratory test results is still an open problem.

4.2. Vertical cut

The second benchmark seeks to assess the performance of the

implemented nonlocal regularisation in a dynamic large deformation
problem. Therefore, the formulation described by Eq. (9) is employed
here. The analysis corresponds to a vertical cut where gravity is grad-
Fig. 9. Load vs. displacement curves from the biaxial test simulation obtained with ually increased until failure is induced. The square domain, shown in
the regularised MPM for different values of the length scale parameter 𝑙s .
Fig. 10, is adopted, with dimensions ℎ = 𝑤 = 3 m. At the left boundary,
null normal displacements were prescribed, while at the bottom bound-
ary both vertical and horizontal displacement components were fixed.
required minimum element size for the proper performance of the The employed parameters are shown in Table 1; they are similar to
nonlocal regularisation appears to be smaller, in the order of 0.2𝑙s . those used in the biaxial test, although some modifications were made
Figs. 7 and 8 show the results in terms of contours of deviatoric to obtain a well-defined failure mechanism for this BVP. A null initial
strains ϵ𝑞 for the standard and regularised MPM, respectively, where stress state was considered throughout the domain, and loading was
√ [ ] applied by increasing gravity in the 𝑦 direction in increments of 𝛥𝑔
2
ϵ𝑞 = 9
(ϵ𝑥𝑥 − ϵ𝑦𝑦 )2 + ϵ2𝑦𝑦 + ϵ2𝑥𝑥 + 13 𝛾𝑥𝑦
2 . (37) = 0.1 g every 0.001 s, reaching a maximum value of 𝑔max = 60 g at 𝑡
= 0.6 s. From this point onward, gravity was maintained constant and
The position of the material points is also depicted in the figures. In the the simulation was continued up to 𝑡 = 0.75 s to allow the evolution
case of the standard MPM (Fig. 7), mesh-dependent behaviour can also of the formed failure mechanism. For an in-depth analysis, two points,
be identified, with the width of the formed shear band decreasing and labelled A and B, are identified in the domain (Fig. 10). Point A, located
the magnitude of deviatoric strains increasing as the size of elements at the top-right corner, is used to monitor displacements over time.

7
J.L. González Acosta et al. Computers and Geotechnics 172 (2024) 106424

Fig. 11. Contours of normalised deviatoric strains and position of material points from the vertical cut simulation obtained with the standard MPM, for times 𝑡 of 0.6 and 0.7 s,
using element sizes 𝑙ele of (a,e) 0.1, (b,f) 0.083, (c,g) 0.071, and (d,h) 0.062 m.

mechanism, when displacements grow rapidly, takes place earlier as

the element size is reduced, i.e. it is triggered by a lower 𝑔 value.
Displacements at the end of the simulation also increase significantly as
the mesh is refined. Furthermore, for the two finer meshes, with 𝑙ele of
0.071 m and 0.062 m, numerical convergence issues were encountered
and the simulations were not completed. This is a typical outcome when
simulating softening materials in non-regularised continuum-based nu-
merical analyses (Mánica et al., 2022a). The fact that accelerations,
and in turn final displacements, depend on the adopted mesh is par-
ticularly relevant in the context of the MPM. These kinds of numerical
techniques are generally adopted to address deformations beyond the
capabilities of standard FEM formulations, for instance, to study not
only the trigger of a failure but the consequences that the failure will
have in the surrounding area. Therefore, run-out distances should not
depend on the resolution of the background mesh adopted.
Fig. 13 shows contours of deviatoric strains and the position of
materials points obtained with the regularised MPM. Here, a length
scale parameter 𝑙s = 0.4 m was adopted. Although some differences
Fig. 12. Evolution of displacement at point A from the vertical cut simulation obtained can still be identified for different meshes, they are much smaller than
with the standard (dashed lines) and regularised (solid lines) MPM and for different for the standard MPM. This is also evident in Fig. 12, which also
element sizes 𝑙ele . shows the evolution of displacement at point A for the regularised
simulations. Larger differences are obtained for the analysis with 𝑙ele
= 0.1 m. However, it is important to notice that this element size
Point B, located at a distance from the bottom-right corner of ℎB = 0.34 is somewhat larger than the maximum identified in the biaxial test
and 𝑤B = 0.22 m, is used to track the evolution of deviatoric stresses simulations for the proper performance of the nonlocal regularisation,
close to the region where the failure mechanism initiates. of 0.2𝑙s .
Fig. 11 shows the progression of failure, in terms of contours To further stress the importance of regularisation, Fig. 14 shows
of deviatoric strains (normalised with respect to the corresponding curves of deviatoric stress 𝑞 vs. displacement magnitude at point B,
maximum value) and the position of the material points, for 𝑡 = 0.6 s obtained with both the standard and regularised MPM, where 𝑞 is
(Fig. 11a to d) and 𝑡 = 0.7 s (Fig. 11e to h) and for different element defined as
√
sizes of the background mesh. Again, the mesh-dependent behaviour is
𝑞 = √1 (𝜎1 − 𝜎2 )2 + (𝜎2 − 𝜎3 )2 + (𝜎1 − 𝜎3 )2 . (38)
evident. Although the change in the width of the formed shear bands 2
is less noticeable than in the biaxial test (Fig. 7), the magnitude of The standard MPM resulted in a more brittle response and with very
deviatoric strains increases as the size of the elements is reduced. The different paths followed between the different analyses. Furthermore,
material points experience higher accelerations for the finer meshes, a point was reached where displacements no longer accumulate and
resulting in significantly larger displacements. The latter can be further the deviatoric stress increases again. This is attributed to the formed
identified in Fig. 12, where the evolution of the vertical displacement shear band slightly shifting its position, causing point B to be located
at point A for the standard MPM is shown. The activation of the failure outside, to the left of the shear band and, thus, remaining relatively

8
J.L. González Acosta et al. Computers and Geotechnics 172 (2024) 106424

Fig. 13. Contours of normalised deviatoric strains and position of material points from the vertical cut simulation obtained with the regularised MPM, for times 𝑡 of 0.6 and 0.7 s,
using element sizes 𝑙ele of (a,e) 0.1, (b,f) 0.083, (c,g) 0.071, and (d,h) 0.062 m.

Fig. 14. Evolution of deviatoric stress at point B from the vertical cut simulation Fig. 15. Geometry and boundary conditions of the shallow foundation simulation.
obtained with the standard (dashed lines) and regularised (solid line) MPM and for
different element sizes 𝑙ele .

ground surface. The foundation has dimensions of 𝑤𝐹 = 1.5 and ℎ𝐹 =

stationary during failure. In contrast, the regularised MPM resulted in 3 m. The adopted properties are listed in Table 1. Again, modifications
with respect to the other benchmarks were made to obtain a marked
similar curves regardless of the element size, showing a more or less
failure mechanism for this BVP. For the regularised simulations, a
monotonic reduction of 𝑞 throughout the simulation.
length scale 𝑙s = 0.8 m was employed.
Fig. 16 shows results in terms of normalised contours of deviatoric
4.3. Bearing capacity strains and the position of the material points, obtained with both
the standard and regularised MPM and for different element sizes,
The last benchmark, shown schematically in Fig. 15, corresponds after a foundation penetration of 0.3 m. All simulations exhibit similar
to a stiff foundation that is pushed into the ground at a rate of 𝛥𝛿 = results, with the formation of the classical Prandtl’s failure mechanism.
2.5 × 10−3 m per loading step. The interaction between the foundation However, two notable differences can be identified for the standard
and the ground is considered infinitely rough, hence no slippage occurs, MPM: (1) significantly larger deviatoric strains were developed and
ensuring that the foundation’s movement into the soil is perfectly (2) an additional (vertical) shear path emerged below the bottom right
constrained against lateral displacement. Here, the static formulation corner of the foundation, with larger deviatoric strains as the element
is employed and, therefore, inertial forces are neglected. The adopted size is reduced. In the regularised simulations, this vertical shear band
domain has a width and height of 𝑤 = 14 and ℎ = 8 m, respectively. At does not develop.
the bottom and lateral boundaries the displacements were fixed in the Differences between the standard and regularised simulations can
normal direction, while a null stress condition was considered at the also be identified in Fig. 17, which shows curves of the total deviatoric

9
J.L. González Acosta et al. Computers and Geotechnics 172 (2024) 106424

Fig. 16. Contours of normalised deviatoric strains and position of material points from the shallow foundation simulation obtained with the standard (top row) and regularised
(bottom row) MPM using element sizes 𝑙ele of (a, d) 0.25, (b, e) 0.13, and (c,f) 0.1 m.

computation of the nonlocal state variable, is performed at each global

increment to account for the significant movement of material points.
This identification was efficiently performed by taking advantage of
the structured background mesh and a pre-established neighbourhood
patch computed for each element. A number of benchmark simulations
were carried out to demonstrate the importance of regularisation in
MPM. However, an enhanced MPM formulation, namely the DM-GC
MPM, was necessary since the well-known stress oscillation problem
hindered a satisfactory analysis of the BVPs when dealing with soft-
ening materials. In general, the same numerical pathologies found in
conventional FEM formulations were observed here. A more brittle
response and larger strains and displacements are obtained for the
non-regularised MPM as the size of elements is reduced. Furthermore,
dynamic simulations showed that the run-out distance also depends
on the adopted mesh, which is particularly relevant when using the
MPM to study the consequences of a given collapse. On the other hand,
regularised simulations showed consistent behaviour, with a global
Fig. 17. Load vs. displacement curves from the shallow foundation simulation obtained
with the standard (dashed lines) and regularised (solid lines) MPM and for different
response and a configuration of localised deformations that are ap-
element sizes 𝑙ele . proximately independent of the employed mesh. However, in the case
of the four-noded plane strain elements adopted here, the maximum
element size with respect to the selected length scale parameter, for
load vs. the prescribed vertical displacement of the foundation. All the proper performance of the nonlocal averaging, seems to be smaller
simulations reached a similar peak load. However, results from the than previous recommendations, in the order of 0.2𝑙s .
standard MPM showed a more brittle response with a softening rate that
increases as the element size is reduced. On the other hand, regularised
CRediT authorship contribution statement
simulations showed quite similar results, more or less independent of
the adopted mesh. This is again significant for analyses of the whole
failure process undertaken by MPM. José L. González Acosta: Writing – review & editing, Writing –
original draft, Visualization, Validation, Software, Project administra-
5. Conclusions tion, Methodology, Investigation, Formal analysis, Conceptualization.
Miguel A. Mánica: Writing – review & editing, Writing – original draft,
In this paper, a nonlocal regularisation was implemented into the Visualization, Validation, Software, Project administration, Methodol-
MPM framework for the simulation of large deformation problems in ogy, Investigation, Formal analysis, Conceptualization. Philip J. Var-
brittle geomaterials. Unlike other nonlocal implementations in conven- don: Writing – review & editing. Michael A. Hicks: Writing – review
tional FEM formulations, identification of neighbouring points, for the & editing. Antonio Gens: Writing – review & editing.

10
J.L. González Acosta et al. Computers and Geotechnics 172 (2024) 106424

Declaration of competing interest González Acosta, J.L., Vardon, P.J., Hicks, M.A., 2021b. Study of landslides and soil-
structure interaction problems using the implicit material point method. Eng. Geol.
285, 106043.
The authors declare that they have no known competing finan-
González Acosta, J.L., Vardon, P.J., Remmerswaal, G., Hicks, M.A., 2020. An investi-
cial interests or personal relationships that could have appeared to gation of stress inaccuracies and proposed solution in the material point method.
influence the work reported in this paper. Comput. Mech. 65 (2), 555–581.
González Acosta, J.L., Zheng, X., Vardon, P.J., Hicks, M.A., Pisano, F., 2019. On stress
Data availability oscillation in MPM simulations involving one or two phases. In: Proceedings of
the Second International Conference on the Material Point Method for Modelling
Soil-Water-Structure Interaction. Cambridge, pp. 135–139.
No data was used for the research described in the article. Goodarzi, M., Rouainia, M., 2017. Modelling slope failure using a quasi-static MPM
with a non-local strain softening approach. Procedia Eng. 175, 220–225.
References Guilkey, J.E., Weiss, J.A., 2003. Implicit time integration for the material point method:
Quantitative and algorithmic comparisons with the finite element method. Internat.
J. Numer. Methods Engrg. 57 (9), 1323–1338.
Anand, L., Aslan, O., Chester, S.A., 2012. A large-deformation gradient theory for elastic
Khoei, A.R., Karimi, K., 2008. An enriched-FEM model for simulation of localization
– plastic materials : Strain softening and regularization of shear bands. Int. J. Plast.
phenomenon in cosserat continuum theory. Comput. Mater. Sci. 44 (2), 733–749.
30–31, 116–143.
Knuth, D.E., 1997. The Art of Computer Programming, third ed. Addison Wesley,
Bardenhagen, S.G., Kober, E.M., 2004. The generalized interpolation material point
Boston, MA.
method. Comput. Model. Eng. Sci. 5 (6), 477–496.
Kohler, M., Stoecklin, A., Puzrin, A.M., 2022. A MPM framework for large-deformation
Bathe, K.J., 1995. Finite Element Procedures, first ed. Prentice-Hall, Englewood Cliffs,
seismic response analysis. Can. Geotech. J. 59 (6), 1046–1060.
NJ.
Lian, Y., Zhang, X., Liu, Y., 2011. Coupling of finite element method with material
Bažant, Z.P., Jirásek, M., 2002. Nonlocal integral formulations of plasticity and damage:
point method by local multi-mesh contact method. Comput. Methods Appl. Mech.
survey of progress. J. Eng. Mech. 128 (11), 1119–1149.
Engrg. 200 (47–48), 3482–3494.
Bažant, Z.P., Lin, F.-B., 1988. Nonlocal yield limit degradation. Internat. J. Numer.
Lu, X., Bardet, J.-P., Huand, M., 2012. Spectral analysis of nonlocal regularization in
Methods Engrg. 26 (8), 1805–1823.
two-dimensional finite element models. Int. J. Numer. Anal. Methods Geomech.
Bažant, Z.P., Pijaudier-Cabot, G., 1988. Nonlocal continuum damage, localization
36, 219–235, URL http://onlinelibrary.wiley.com/doi/10.1002/nag.527/abstract%
instability and convergence. J. Appl. Mech. 55 (2), 287–293.
5Cnhttps://web.natur.cuni.cz/uhigug/masin/download/MTVC_IJNAMG06-pp.pdf.
Benallal, A., Marigo, J.J., 2007. Bifurcation and stability issues in gradient theories
Mánica, M., 2018. Analysis of Underground Excavations in Argillaceous Hard Soils -
with softening. Modelling Simul. Mater. Sci. Eng. 15 (1), S283–S295.
Weak Rocks (Ph.D. thesis). Technical University of Catalonia.
Bentley Systems, 2022. Plaxis CONNECT edition V22.01.
Mánica, M.A., Ciantia, M.O., Gens, A., 2020. On the stability of underground caves
Beuth, L., Więckowski, Z., Vermeer, P.A., 2011. Solution of quasi-static large-strain
in calcareous rocks due to long-term weathering. Rock Mech. Rock Eng. 53,
problems by the material point method. Int. J. Numer. Anal. Methods Geomech.
3885–3901.
35 (13), 1451–1465.
Mánica, M.A., Gens, A., Vaunat, J., Armand, G., Vu, M.N., 2022a. Numerical simulation
Brinkgreve, R.B.J., 1994. Geomaterials Models and Numerical Analysis of Softening
of underground excavations in an indurated clay using non-local regularisation. Part
(Ph.D. thesis). Delft University of Technology, p. 156.
1: formulation and base case. Geotechnique 72 (12), 1092–1112.
Burghardt, J., Brannon, R., Guilkey, J., 2012. A nonlocal plasticity formulation for the Mánica, M.A., Gens, A., Vaunat, J., Armand, G., Vu, M.N., 2022b. Numerical simulation
material point method. Comput. Methods Appl. Mech. Engrg. 225, 55–64. of underground excavations in an indurated clay using non-local regularisation. Part
Burland, J.B., 1990. On the compressibility and shear strength of natural clays. 2: sensitivity analysis. Géotechnique 72 (12), 1113–1128.
Géotechnique 40 (3), 329–378. Mánica, M.A., Gens, A., Vaunat, J., Ruiz, D.F., 2018. Nonlocal plasticity modelling of
Charlton, T., Coombs, W., Augarde, C., 2017. iGIMP: An implicit generalised in- strain localisation in stiff clays. Comput. Geotech. 103, 138–150.
terpolation material point method for large deformations. Comput. Struct. 190, Marcher, T., 2003. Nichtlokale Modellierung der Entfestigung Dichter Sande und Steifer
108–125. Tone (Ph.D. thesis). University of Stuttgart, p. 151.
Coombs, W.M., Charlton, T.J., Cortis, M., Augarde, C.E., 2018. Overcoming volumetric Martinelli, M., Galavi, V., 2021. Investigation of the material point method in the
locking in material point methods. Comput. Methods Appl. Mech. Engrg. 333, 1–21. simulation of cone penetration tests in dry sand. Comput. Geotech. 130, 103923.
Cui, W., Wu, X., Potts, D.M., Zdravković, L., 2023. Nonlocal strain regularisation for Monforte, L., Ciantia, M.O., Carbonell, J.M., Arroyo, M., Gens, A., 2019. A stable mesh-
critical state models with volumetric hardening. Comput. Geotech. 157 (February). independent approach for numerical modelling of structured soils at large strains.
de Borst, R., Sluys, L.J., Mühlhaus, H.-B., Pamin, J., 1993. Fundamental issues in finite Comput. Geotech. 116.
element analyses of localization of deformation. Eng. Comput. 10 (2), 99–121, URL Newmark, N.M., 1959. A method of computation for structural dynamics. J. Eng. Mech.
http://www.emeraldinsight.com/doi/10.1108/eb023897. Div. 85 (3), 67–94.
Desrues, J., Viggiani, G., 2004. Strain localization in sand: an overview of the Phuong, N.T.V., Van Tol, A.F., Elkadi, A.S.K., Rohe, A., 2016. Numerical investigation
experimental results obtained in grenoble using stereophotogrammetry. Int. J. of pile installation effects in sand using material point method. Comput. Geotech.
Numer. Anal. Methods Geomech. 28 (4), 279–321. 73, 58–71.
Eringen, A.C., 1981. On nonlocal plasticity. Internat. J. Engrg. Sci. 19 (12), 1461–1474. Pietruszczak, S., Mroz, Z., 1981. Finite element analysis of deformation of
Eringen, A.C., 1983. Theories of nonlocal plasticity. Internat. J. Engrg. Sci. 21 (7), strain-softening materials. Internat. J. Numer. Methods Engrg. 17 (3), 327–334.
741–751. Read, H.E., Hegemier, G.A., 1984. Strain softening of rock, soil and concrete - a review
Frédéric, C., Panagiotis, K., Benoît, P., 2021. Numerical modeling of multiphysics article. Mech. Mater. 3 (4), 271–294.
couplings and strain localization. In: Instabilities Modeling in Geomechanics. John Rolshoven, S., 2003. Nonlocal Plasticity Models for Localized Failure (Ph.D. thesis).
Wiley & Sons, Ltd, pp. 203–251, chapter 7. URL https://onlinelibrary.wiley.com/ École Polytechnique Fédérale de Lausanne.
doi/abs/10.1002/9781119755203.ch7. Romero, T., Mánica, M.A., Ovando-Shelley, E., Rodríguez-Rebolledo, J.F., Buritica, J.A.,
Galavi, V., Schweiger, H.F., 2009. A multilaminate model with destructuration. Soils 2024. On the determination of softening parameters for nonlocal constitutive
Found. 49 (3), 341–353. models. J. Geotech. Geoenviron. Eng. submitted for publication.
Galavi, V., Schweiger, H.F., 2010. Nonlocal multilaminate model for strain softening Singh, V., Stanier, S., Bienen, B., Randolph, M.F., 2021. Modelling the behaviour
analysis. Int. J. Geomech. 10 (1), 30–44. of sensitive clays experiencing large deformations using non-local regularisation
Gao, Z., Li, X., Lu, D., 2022. Nonlocal regularization of an anisotropic critical state techniques. Comput. Geotech. 133, 104025.
model for sand. Acta Geotech. 17 (2), 427–439. Sluys, L.J., de Borst, R., 1992. Wave propagation and localization in a rate-dependent
Gens, A., 2013. On the hydromechanical behaviour of argillaceous hard soils-weak cracked medium-model formulation and one-dimensional examples. Int. J. Solids
rocks. In: Anagnostopoulos, A., Pachakis, M., Tsatsanifos, C. (Eds.), Proceedings of Struct. 29 (23), 2945–2958.
the 15th European Conference on Soil Mechanics and Geotechnical Engineering - Summersgill, F.C., Kontoe, S., Potts, D.M., 2017. Critical assessment of nonlocal
Geotechnics of Hard Soils - Weak Rocks. Vol. 4, IOS Press, Athens, pp. 71–118, strain-softening methods in biaxial compression. Int. J. Geomech. 17 (7), 1–14.
URL https://doi.org/10.3233/978-1-61499-199-1-71. Tejchman, J., Bauer, E., 2005. FE-simulations of a direct and a true simple shear test
Gens, A., Carol, I., Alonso, E.E., 1990. A constitutive model for rock joints formulation within a polar hypoplasticity. Comput. Geotech. 32 (1), 1–16.
and numerical implementation. Comput. Geotech. 9 (1–2), 3–20. Tielen, R., Wobbes, E., Möller, M., Beuth, L., 2017. A high order material point method.
González Acosta, J.L., 2020. Investigation of MPM Inaccuracies, Contact Simulation and Procedia Eng. 175, 265–272.
Robust Implementation for Geotechnical Problems (Ph.D. thesis). Delft University Troncone, A., Pugliese, L., Conte, E., 2022. Analysis of an excavation-induced landslide
of Technology. in stiff clay using the material point method. Eng. Geol. 296, 106479.
González Acosta, J.L., Vardon, P.J., Hicks, M.A., 2017. Composite material point van Eekelen, H.A.M., 1980. Isotropic yield surfaces in three dimensions for use in soil
method (CMPM) to improve stress recovery for quasi-static problems. Procedia Eng. mechanics. Int. J. Numer. Anal. Methods Geomech. 4 (1), 89–101.
175, 324–331. Vermeer, P.A., Brinkgreve, R.B.J., 1994. A new effective non-local strain measure for
González Acosta, J.L., Vardon, P.J., Hicks, M.A., 2021a. Development of an implicit softening plasticity. In: Chambon, R., Desrues, J., Vardoulakis, I. (Eds.), Localisation
contact technique for the material point method. Comput. Geotech. 130, 103859. and Bifurcation Theory for Soils and Rock. Balkema, Rotterdam, pp. 89–100.

11
J.L. González Acosta et al. Computers and Geotechnics 172 (2024) 106424

Wang, B., Vardon, P.J., Hicks, M.A., Chen, Z., 2016. Development of an im- Yerro, A., Soga, K., Bray, J., 2019. Runout evaluation of oso landslide with the material
plicit material point method for geotechnical applications. Comput. Geotech. 71, point method. Can. Geotech. J. 56 (9), 1304–1317.
159–167. Zhang, F., Zhang, X., Sze, K.Y., Lian, Y., Liu, Y., 2017. Incompressible material point
Yerro, A., Girardi, V., Martinelli, M., Ceccato, F., 2022. Modelling unsaturated soils with method for free surface flow. J. Comput. Phys. 330, 92–110.
the material point method. A discussion of the state-of-the-art. Geomech. Energy
Environ. 32, 100343.

12