+Model

MATCOM 2939 1–48 ARTICLE IN PRESS

Available online at www.sciencedirect.com

Mathematics and Computers in Simulation xxx (2008) xxx–xxx

1 Meshless methods: A review and computer

F
2 implementation aspects

OO
3 Nguyen Vinh Phu a , Timon Rabczuk b , Stéphane Bordas c,∗ , Marc Duflot d
4
a Ecole Nationale d’Ingénieur de Saint Etienne (ENISE), Laboratoire de Tribologie et Dynamique des Systèmes (LTDS), France
5
b University of Canterbury, Department of Mechanical Engineering, 4800 Private Bag, Christchurch, New Zealand

Q1 c University of Glasgow, Civil Engineering, Rankine Building, Glasgow G12 8LT, UK

6
7
d CENAERO, Belgium

PR
Received 30 April 2007; received in revised form 17 September 2007; accepted 8 January 2008

9 Abstract
The aim of this manuscript is to give a practical overview of meshless methods (for solid mechanics) based on global weak forms
ED
10

11 through a simple and well-structured MATLAB code, to illustrate our discourse. The source code is available for download on our
12 website and should help students and researchers get started with some of the basic meshless methods; it includes intrinsic and
13 extrinsic enrichment, point collocation methods, several boundary condition enforcement schemes and corresponding test cases.
14 Several one and two-dimensional examples in elastostatics are given including weak and strong discontinuities and testing different
15 ways of enforcing essential boundary conditions.
CT

16 © 2007 IMACS. Published by Elsevier B.V. All rights reserved.

17 Q3 Keywords: Meshless methods; Partition of unity; Intrinsic enrichment; Extrinsic discontinuities; Computer implementation; MATLAB
18

18 1. Introduction
E

19 The finite element method has been used with great success in many fields with both academic and industrial
RR

20 applications. It is however not without limitations. Due to mesh-based interpolation, distorted or low quality meshes
21 lead to higher errors, necessitate remeshing, a time and human labour consuming task, which is not guaranteed to be
22 feasible in finite time for complex three-dimensional geometries.
23 Additionally, due to the underlying structure of the classical mesh-based methods, they are not well suited to treat
24 problems with discontinuities that do not align with element edges. One strategy for dealing with moving discontinuities
in mesh-based methods is remeshing or discontinuous enrichment. However, remeshing is costly, still difficult in three
CO

25

26 dimensions and requires projection of quantities between successive meshes and consequential degradation of accuracy.
27 The extended finite element method (XFEM) [6,79,24,45,23,22,24] enriches the approximation space so that weak
28 and strong discontinuities can be captured.
29 Meshless methods (MMs) were born with the objective of eliminating part of the difficulties associated with reliance
30 on a mesh to construct the approximation. In MMs, the approximation is built from nodes only. One of the first meshless
UN

∗ Corresponding author. Tel.: +44 1413304075.

Q2
E-mail addresses: [email protected] (S. Bordas), [email protected] (M. Duflot).
URL: http://www.civil.gla.ac.uk/∼bordas (S. Bordas).

0378-4754/$32.00 © 2007 IMACS. Published by Elsevier B.V. All rights reserved.

doi:10.1016/j.matcom.2008.01.003

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
2 N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx

31 methods is the smooth particle hydrodynamics (SPH) method by Lucy [77] and Gingold and Monaghan [54]. It was
32 born to solve problems in astrophysics and, later on, in fluid dynamics [20,81,80]. Libersky et al. [71] were the first
33 to employ SPH in solid mechanics (impact). Since the original SPH version suffered from spurious instabilities and
34 inconsistencies [97,9,101], many improvements were incorporated into SPH [12,88,20,21,61,62,35,36,100]. While
35 SPH and their corrected versions were based on a strong form, other methods were developed in the 1990s, based on a
36 weak form. Major applications of these methods are in solid mechanics. The element-free Galerkin (EFG) method [14]
37 was developed in 1994 and was one of the first meshless methods based on a global weak form. The reproducing kernel
38 particle method (RKPM) [73] was developed 1 year later. Though the final equations are very similar to the equations

F
39 of the EFG method, RKPM has its origin in wavelets. In contrast to RKPM and the EFG method, that use a so-called
40 intrinsic basis, other methods were developed that use an extrinsic basis and the partition of unity concept. This extrinsic

OO
41 basis was initially used to increase the approximation order similar to a p-refinement as, e.g. in the hp-cloud method
42 [40,72]. Melenk and Babuška [78] pointed out the similarities of meshless and finite element methods and developed
43 the so-called partition of unity finite element method (PUFEM). The method is very similar to the hp-cloud method.
44 Generally, PUFEM shape functions are based on Lagrange polynomials, while the general form of the hp-cloud method
45 also includes the MLS-approximation. Strouboulis et al. [91] pointed out in their generalized finite element method
46 (GFEM) that different partition of unities can be used for the usual approximation and the so-called enrichment. In

PR
47 the XFEM [6,79,94], the extrinsic enrichment was modified such that it can handle strong discontinuities without
48 remeshing. Moreover, XFEM is based on a local PU concept. Another class of meshless methods are methods that are
49 based on local weak forms. The most popular method is the meshless local Petrov–Galerkin (MLPG) method [2–4].
50 The main difference of the MLPG method to methods such as EFG or RKPM is that local weak forms are generated
51 on overlapping subdomains rather than using global weak forms. The integration of the weak form is then carried out
52 in these local subdomains. Atluri [1] introduced the notion “truly” meshless since no construction of a background
ED
53 mesh is needed for integration purposes. Another well known method that was mainly applied in fluid mechanics is
54 the moving point method [83,82,75].
55 Some major advantages of MMs are (i) h-adaptivity is simpler to incorporate in MMs than in mesh-based methods,
56 (ii) problems with moving discontinuities such as crack propagation, shear bands and phase transformation can be
57 treated with ease, (iii) large deformation can be handled more robustly, [30,29], (iv) higher-order continuous shape
CT

58 functions, (v) non-local interpolation character and (vi) no mesh alignment sensitivity. Beside these advantages, MMs
59 are not without disadvantages. The MMs shape functions are rational functions which requires high-order integration
60 scheme to be correctly computed. The treatment of essential boundary conditions is not as straightforward as in mesh-
61 based methods since the MMs shape functions are not interpolants. They do not satisfy the Kronecker delta property.
62 In general, the computational cost of MMs is higher than one of FEM.
E

63 To avoid some difficulties inherent in MMs, MMs were coupled successfully to finite element methods, see, e.g.
64 [18,56,47,58,48,49]. Meanwhile, hybrid methods are available that exploit the advantages of meshfree methods and
RR

65 finite elements [55,74,60], e.g. the shape functions fulfill the Kronecker delta property while simultaneously exploiting
66 the smoothness and higher-order continuity of meshfree shape functions.
67 The purpose of this manuscript is to give a practical overview of meshless methods, especially with respect to
68 their computer implementation. Common issues in MMs are approximation, integration of the weak form, imposing
69 essential boundary conditions, how to efficiently compute shape functions and how to incorporate strong and weak
discontinuities. In addition, the weighted residual methods such as collocation and Galerkin procedures are also stated
CO

70

71 with examples. Advanced issues in application of MMs to fracture mechanics, coupling MMs with finite elements are
72 reviewed.
73 Computer implementation aspects of the EFG and enriched EFG are given in detail through a MATLAB code.1 In
74 particular, the source code of the program includes intrinsic and extrinsic enrichment for cracks and material interfaces.
75 The paper is organized as follows. Section 2 gives a detailed description of MMs including their approximations,
UN

76 imposition of essential boundary conditions, numerical integration of the weak form. Some of the typical MMs such
77 as the element-free Galerkin method are introduced. The computer implementation aspects are introduced in Section
78 3. Section 4 presents some numerical examples on linear elasticity.

1 Which is available at http://www.civil.gla.ac.uk/∼bordas/codes/efgMatlab/EFGMatlabCode.rar.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx 3

F
OO
Fig. 1. Discretization using meshless methods: nodes, domains of influence (circular shape).

PR
79 2. Meshless methods

80 2.1. Basic approximations

ED
81 Meshless approximations for a scalar function u in terms of the material (Lagrangian) coordinates can be written as

82 u(x, t) = ΦI (x)uI (t) (1)
I ∈S

83 where ΦI : Ω → R are the shape functions and the uI ’s are the nodal values at particle I located at position xI and S is
CT

84 the set of nodes I for which ΦI (x) = 0. Note, that the above form is identical to an FEM approximation. However, in
85 contrast to FEM, the shape functions in Eq. (1) are only approximants and not interpolants, since uI = u(xI ). There-
86 fore special techniques are needed to treat displacement boundary conditions, that will be discussed in a subsequent
87 section.
E

88 2.2. Kernel (weight) function

RR

89 The shape functions ΦI are obtained from the kernel functions, often called window or weighting functions, which
90 are denoted by wI : Ω → R. The kernel functions have compact support. The support size is defined by the so called
91 dilatation parameter or smoothing length. It is critical to solution accuracy, stability and plays the role of the element
92 Q4 size in the finite element method (Fig. 1).
93 The final characteristic of weight functions is its functional forms. The weight function should be continuous and
CO

94 positive in its support. For all the meshless methods that we will review in this paper, the continuity of the shape
95 function will be determined solely by the continuity of the kernel function, for details see, e.g. [57]. For example, if
96 the kernel function is C2 , then the corresponding shape function is also C2 .
97 Some commonly used weight functions are
UN

98 • the cubic spline weight function:

⎧
⎪
⎪
2
− 4r 2 + 4r 3 , r≤
1
⎪
⎪
⎨3 2
w(r) = 4 4 3 1 (2)
− 4r + 4r − r ,
2
<r≤1
99
⎪
⎪
⎪
⎪ 3 3 2
⎩
0, r>1

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
4 N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx

100 • the quartic spline weight function:


1 − 6r 2 + 8r 3 − 3r 4 , r ≤ 1
101 w(r) = (3)
0, r>1

102 with
||xI − x||

F
103 r= (4)
dI

OO
104 where dI is the support size of node I.
105 In two dimensions, circular and rectangular supports are used.

106 • Circular support:


||xI − x||

PR
107 w(x − xI ) = w (5)
dI

108 • Rectangular support:

 
|xI − x| |yI − y|
109 w(x − xI ) = w w y (6)
dIx dI
ED
110 The derivatives of the weight functions can be compute using the chain rule. For example, for circular supports, we
111 have
xk − xIk
CT

112 wk (r) = wr (r)rk = wr (7)

rdI2

113 2.3. Completeness

E

114 Completeness, often referred to as reproducibility, in Galerkin methods plays the same role as consistency in finite
115 difference methods. Completeness means the ability of an approximation to reproduce a polynomial of a certain order.
RR

116 An approximation is called zero-order complete if it reproduces constant functions exactly. It is called linear (first
117 order) complete if it reproduces linear functions exactly, and so on for higher orders of completeness.

118 2.4. Partition of unity

CO

119 A partition of unity (PU) is a paradigm where a domain is divided into overlapping subdomains ΩI , each of which
120 is associated with a function ΦI (x) which is nonzero only in ΩI and has the following property:


N
121 ΦI (x) = 1 in Ω (8)
I=1
UN

122 Let us recall Eq. (1). There are basically two ways to increase the order of completeness of that approximation.
123 The first opportunity is to increase the completeness of the shape function intrinsically, i.e. by increasing the order of
124 completeness of the shape functions directly. Alternatively, the order of completeness may be increased by modifying
125 Eq. (1) using the partition of unity (PU) concept. In this case, a low-order approximation space (low-order shape
126 functions ΦI ) is enriched with additional functions, which increases the order of completeness. These two concepts
will be explained subsequently.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx 5

127 2.5. Intrinsic meshless methods

128 2.5.1. Smooth particle hydrodynamics

129 One of the oldest MMs is the smoothed particle hydrodynamics (SPH) [77]:

130 u (x) =
h
w(x − y, h)u(y) dΩy (9)
Ω
131 While the continuous form of SPH is second-order complete, it can easily be shown that the discrete SPH form:

F

N
132 uh (x) = w(x − xI )uI VI (10)

OO
I
133 cannot even reproduce constant fields, and hence is not a partition of unity. In Eq. (10)VI is some measure of the
134 domain surrounding node I.

135 2.5.2. Reproducing kernel particle method (RKPM)

The reproducing kernel particle method (RKPM) [73] is an improvement of the continuous SPH approximation. In

PR
136

137 order to increase the order of completeness of the approximation, a correction function C(x, y) is introduced into the
138 approximation:

139 uh (x) = C(x, y)w(x − y)u(y) dΩy (11)

Ωy

where K(x, y) = C(x, y)w(x − y) with C(x, y) is defined such that the approximation is n th order consistent:
ED
140

141 u(x) = pT (x)a (12)

142 p(x)u(x) = p(x)p (x)a
T
(13)

143 p(y)w(x − y)u(y) dΩy = p(y)pT (y)w(x − y) dΩy a (14)

CT

Ωy Ωy

144 This is a system of equations for a, which can then be substituted into the approximation uh (x) = pT (x)a, it yields:

−1

145 uh (x) = pT (x) p(y)pT (y)w(x − p) dΩy p(y)w(x − y)u(y) dΩy (15)
E

Ωy Ωy

146 with the correction function:


−1
RR

147 C(x, y) = p (x)

T
p(y)p (y)w(x − y) dΩy
T
p(y) = pT (x)[M(x)]−1 p(y) (16)
Ωy

148 To evaluate this continuous expression, numerical integration must be employed. This step leads from the reproducing
149 kernel method to its discrete version, the reproducing kernel particle method [73]:
CO

N
150 uh (x) = C(x, y)w(x − y)u(y)dΩy = C(x, xI )w(x − xI )uI VI
Ωy I=1


N
151 = pT (x)[M(x)]−1 p(xI )w(x − xI )uI VI (17)
UN

I=1
152

153 The moment matrix M(x) is also computed by numerical integration:

N
154 M(x) = p(y)pT (y)w(x − y)dΩy = p(xI )pT (xI )w(x − xI ) VI (18)
Ωy I=1
An interesting remark is observed if we choose VI = 1: the RKPM and MLS are the same.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
6 N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx

155 2.5.3. Moving least squares (MLS) approximation

156 This method was introduced by Shepard [90] in the late 1960s for constructing smooth approximations to fit a
157 specified cloud of points. It was then extended in [65] for general surface generation problems. The most famous
158 application of MLS approximation is probably within the element-free Galerkin (EFG) method, [17,16,7,19].
159 The approximation uh : Ω → R of the function u: Ω → R is posed as a polynomial of order m but with non-constant
160 coefficients. The local approximation around a point x̄ ∈ Ω, evaluated at a point x ∈ Ω is given by
161 uhL (x, x̄) = pT (x)a(x̄) (19)

F
162 where p(x) is a complete polynomial of order m:
pT (x) = [1 x x2 , . . . , xm ] (20)

OO
163

164 and a(x) contains non constant coefficients that depend on x (hence the name “moving”):
165 aT (x) = [a0 (x) a1 (x) a2 (x), . . . am (x)] (21)
166 The unknown parameters aj (x) are determined at any point x, by minimizing a functional J(x) defined by a
weighted2 average over all nodes I ∈ {1, . . . , n} where the parameters uI are specified, of the difference between the

PR
167

168 local approximation uhL (xI , x) and the value uI , at node I, of the function u to be approximated:

n
2 
n
2
169 J(x) = w(x − xI )[uhL (xI , x) − uI ] = w(x − xI )[pT (xI )a(x) − uI ] (22)
I=1 I=1

170 where n is the number of nodes in the neighborhood of x where the weight function w(x − xI ) = 0.
ED
171 An extremum of J in Eq. (22) with respect to the coefficients a(x) can be obtained by setting the derivative of J
172 with respect to a(x) equal to zero. The following equations result:

n
w(x − xI )2p1 (xI )[pT (xI )a(x) − uI ] = 0
I=1
CT


n
w(x − xI )2p2 (xI )[pT (xI )a(x) − uI ] = 0
173 I=1 (23)
..
.
E

 n
w(x − xI )2pm (xI )[pT (xI )a(x) − uI ] = 0
I=1
RR

174 After rearrangements, the above becomes:


n 
n
175 w(x − xI )p(xI )pT (xI )a(x) = w(x − xI )p(xI )uI (24)
I=1 I=1
CO

176 Or more compact as

177 A(x)a(x) = B(x)u (25)
178 where

n
A(x) = w(x − xI )p(xI )pT (xI ) (26)
UN

179

I=1

180 and
181 B(x) = [w(x − x1 )p(x1 ) w(x − x2 )p(x2 ) . . . w(x − xn )p(xn )] (27)

2 Weight function w defined in Section 2.2.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx 7

F
OO
Fig. 2. (a) Particle arrangement for a regular moment matrix for a linear complete MLS basis. (b) Particle arrangement for a singular moment matrix
for a linear complete MLS basis.

182 Solving a(x) from Eq. (25) and substituting it into Eq. (19), the MLS approximants can be defined as
uh (x) = pT (x)[A(x)]−1 B(x)u

PR
183 (28)
184 Recalling the form of the approximation defined in Eq. (1):


N
185 uh (x) = ΦI (x)uI = T (x)u (29)
I=1
ED
186 we can immediately write the MLS shape functions as
187 T (x) = pT (x)[A(x)]−1 B(x) (30)
188 or, for the shape function ΦI associated with node I at a point x:
CT

189 ΦI (x) = pT (x)[A(x)]−1 w(x − xI )p(xI ) (31)

190 The matrix A(x) is often called moment matrix, it is of size m × m. This matrix must be inverted wherever the MLS
191 shape functions are to be evaluated. Obviously, this fact is one drawback of MLS-based MMs because of computational
192 cost and the possibility that this moment matrix may be singular.
E

193 Consider a linear basis in one dimension, the moment matrix then becomes:
     
1 x1 1 x2 1 xn
RR

194 A(x) = w(x − x1 ) + w(x − x2 ) + · · · + w(x − xn ) (32)

x1 x12 x2 x22 xn xn2

195 It is clear from this equation that if n = 1, i.e. point x is covered by only one nodal support while the basis is linear
196 (m = 2), then the matrix is singular and cannot be inverted. Therefore, a necessary condition for the moment matrix to
197 be invertible is that n ≥ m. Note also that if n = m, the nodes have to be arranged in different coordinate directions,
CO

198 otherwise the matrix will be singular as well, see Fig. 2.

199 It can be seen graphically that the MLS shape functions are indeed a partition of unity in Fig. 3 in 1D. Consider an
200 interval 0 ≤ x ≤ 4 divided into four (4) equal domains. The weight and shape functions of all five (5) nodes are plotted
201 in Fig. 3. The functions associated with the centre node are represented by a heavy line. In this example, the weight
202 function is the quartic spline, the size of all five (5) domains of influence is 2.5. Remark that the MLS shape functions
do not satisfy the Kronecker delta property.3 The derivatives of the MLS shape functions are given in Fig. 4. To get
UN

203

204 the smooth curves, we computed these derivatives at 150 sampling points on the interval 0 ≤ x ≤ 1. An important
205 property of the first derivatives can be observed from this figure: the first derivative of node I vanishes at this node.
This makes MLS-collocation–MMs unstable.

3 The shape function associated with a node is not exactly equal to one at this node (in the present case, it is about 0.7, for the centre node), and
this shape function is not exactly zero at the other nodes in the domain.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
+Model
MATCOM 2939 1–48 ARTICLE IN PRESS
8 N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx

F
OO
PR
Fig. 3. Weight and shape function of the central node: (a) weight function and (b) shape function.

ED
E CT
RR
CO
UN

Fig. 4. MLS shape functions and derivatives with quadratic basis.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx 9

F
OO
Fig. 5. Weight and MLS shape function: (a) weight function and (b) shape function.

PR
206 For two and three dimensions, x becomes vector x and the basis p(x) is given by (only for two dimensions):

207 • Linear basis:

208 pT (x) = [1 x y] (33)

ED
209 • Quadratic basis:

210 pT (x) = [1 x y x2 y2 xy] (34)

CT

211 If p(x) is chosen to be a zeroth order basis, i.e. p(x) = 1, then the resulting MLS shape function is given by
w(x − xI )
212 Φ0I (x) = n (35)
w(x − xI )
I
E

213 which is known as the Shepard function, the lowest order form of MLS shape functions. Note that the basis p is often
214 shifted by (x − xJ )/dI since the shifting improves the conditioning of the moment matrix. A two-dimensional graphical
representation of the quartic spline with the corresponding linear complete MLS shape function is shown in Fig. 5.
RR

215

216 The first spatial derivatives of the shape functions in the x- and y- directions are depicted in Fig. 6.
CO
UN

Fig. 6. MLS shape function first derivatives: (a) Φx and (b) Φy .

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
10 N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx

217 2.6. Extrinsic meshless methods

218 2.6.1. The partition of unity finite element method

219 In [78], a method called partition of unity finite element method (PUFEM) was developed. The approximation in
220 the PUFEM is given by

N 
l 
N
221 uh (x) = φI0 (x) pj (x)vjI = φI0 (x)pT (x)vI (36)
I

F
I=1 j=1

222 where φ0 (x) are usually shape functions based on Lagrange polynomials. The coefficients vjI are nodal unknowns. The

OO
223 attractive property of the approximation is that it is the number of terms xk which dictates the order of completeness
224 of the approximation. Another useful property of this approximation is that, special enhancement functions, usually a
225 known feature of the sought solution, are easily incorporated into the approximation through this extrinsic basis.
226 By examining Eq. (19), we see that in MLS approximations, the basis p(x), and hence the order of consistency, cannot
227 be varied from node to node without introducing a discontinuity in the approximation. This means that p-adaptivity is
228 not naturally obtained by intrinsic enrichment. Regions with different order of consistency may be obtained, but need

PR
229 to be blended together, to acertain continuity between the regions.

230 2.6.2. hp-clouds

231 The approximation in the hp clouds method [40] writes, at any point x ∈ Ω:
⎛ ⎞

N  l
uh (x) = φI (x) ⎝uI + pj (x)vjI ⎠
ED
232 (37)
I j

233 where the pj form the so-called extrinsic basis since it contains both high-order monomials and enhancement functions
234 as well. Enhancement functions or enrichment functions are usually introduced into the approximation space to capture
235 special properties such as discontinuities, singularities, boundary layers, or other relevant features of a solution.
CT

236 Different partitions of unity can be used for the standard and enhanced/enriched parts of the approximation [91]:

N 
M 
l
237 u (x) =
h
φIk (x)uI + φIm (x) pj (x)vjI (38)
I I j
E

238 where φIk (x) and φIm (x) are meshless shape functions of order k and m, respectively.
RR

239 2.6.3. A simple example with extrinsic global enrichment

240 To show the power of enrichment, consider a one-dimensional problem featuring large localized gradients, as shown
241 below:
242 uxx (x) + b(x) = 0, x ∈ [0, 1]; u(0) = 0, u(1) = 1 (39)
CO

243 with
 2
{2α2 − 4[α2 (x − 0.5)] } exp{−[α(x − 0.5)]2 }, x ∈ [0.42, 0.58]
244 b(x) = (40)
0, otherwise
245 The exact solution of this problem is
UN

246 u(x) = x + exp{−[α(x − 0.5)]2 }x ∈ [0, 1] (41)

247 At first, the standard EFG method with linear-complete shape functions, a discretization of 30 evenly-spaced particles
248 (29 intervals) and four Gauss points in each interval is employed. The numerical displacement is compared to the exact
249 displacement in Fig. 7(a). It is worth noting that the numerical solution cannot capture the local character of the
250 exact solution (around x = 0.5). In order to obtain acceptable results, the discretization must be heavily refined in the
251 neighbourhood of the large gradients, otherwise, spurious oscillations appear (Fig. 7(b)).

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx 11

F
OO
PR
ED
E CT
RR

Fig. 7. One-dimensional problem with localized solution: comparison between EFG and enriched EFG solutions.

252 In order to capture the local character, the exact solution can be incorporated into the meshless approximation. For
253 such a simple one-dimensional problem, one can choose global enrichment:

CO

254 uh (x) = ΦI (x)uI + bΨ (x) with Ψ (x) = exp{−[α(x − 0.5)]2 } (42)

I

255 The result obtained with this approximation (30 uniform nodes and 4 G points for each of 29 subcells) is given in
256 Fig. 7(c) with excellent agreement between numerical and exact solution.
257 The global enrichment strategy has the advantage that only one additional unknown is added for each special
UN

258 function to be added. It has the drawbacks that (1) the enrichment function must have local character, i.e. have a
259 compact support “small” relative to the domain size, to ensure that the left hand side matrix remains banded; (2) the
260 discrete equations are modified, which complicated the implementation into existing codes. The local (extrinsic) PU
261 enriched formulation is given by
 
262 uh (x) = ΦI (x)uI + ΦJ (x)Ψ (x)aJ (43)
I ∈S J ∈ Sc

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
12 N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx

F
OO
PR
Fig. 8. Comparison of enrichment strategies and effects of nodal support size.

263 where Sc is the set of nodes whose supports contain the point x = 0.5. The displacements obtained with global
264 enrichment, PU-enrichment and the exact solution are plotted in Fig. 8(a). In all computations, the cubic spline with
265 circular support and radius r̄ = s x with s = 2.5, x is the nodal spacing, is employed.
ED
266 It is obvious that, the number of enriched nodes changes when the size of nodal supports varies. Precisely, when
267 s increases, the number of enriched nodes increases, hence increase the number of problem unknowns. Therefore,
268 choosing a proper value for the support size is necessary in both computational cost and accuracy. Fig. 8(b) shows the
269 results obtained with various support sizes.
CT

270 2.7. Weighted residual methods

271 Considering a partial differential equation on a domain Ω with boundary Γ , defined by the differential operator L:
272 u → Lu and linear form f : Ω → R:
E

273 Lu(x) = f (x) in Ω (44a)

274 u = ū on Γ (44b)
RR

275 One of the most general techniques to solve such an equation numerically is the weighted residual method. In this
276 method, the unknown field u is approximated by trial functions  and nodal parameters u in the form u ≈ uh = T u.
277 Replacing u with uh in the PDE gives:
278 ∀x ∈ Ω, εh (x) = Luh (x) − f (x) (45)
CO

279 where εh is the residual error, which is non-zero, since an approximation function, living in a function space of finite
280 size, cannot fulfill the original equation exactly everywhere in Ω.
281 A set of test functions  are chosen and the system of equations is determined by setting εh orthogonal4 to this set
282 of test functions:

ε dΩ = 0 or (Luh (x) − f (x)) dΩ = 0 (46)

UN

283
Ω Ω

  N 

284  L ΦI (x)uI − f (x) dΩ = 0 (47)
Ω I=1


4 In the sense of the inner product u, v = Ω
uv dΩ.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx 13

285 In the above equations, it was implicitly assumed that integrals are capable of being evaluated. This places certain
286 restrictions on the families to which functions Ψ and Φ must belong. In general, if n th order derivatives occur in the
287 operator L, then the trial and test functions must be Cn−1 (n − 1 continuous derivatives). Usually, integration by parts
288 is applied in Eq. (47) to lower the order of derivation, decreasing the order of continuity required for the test and trial
289 spaces. The form of the partial differential equation is called the weak form associated with the strong form given in
290 Eq. (44).
291 In order to obtain the discrete equations, the unknown function u(x) and the test function  are approximated by

 

F
N N
292 u (x) =
h
ΦI (x)uI and (x) = ΨI (x)δuI (48)
I=1 I=1

OO
293 where δuI are arbitrary coefficients, and uI are unknowns of the problem.
294 The choice of the functions ΨI (x) leading to different methods such as collocation and Galerkin methods which are
295 described in the next section.

2.7.1. Collocation method

PR
296

297 Assume the xI to denote the set of points in the computational domain, in the collocation method, the test functions
298  are chosen to be Dirac delta distributions δ(x − xI ). Because of the shifting property of the Dirac delta distributions,
299 the weak form, Eq. (47), reduces to the strong form, evaluated at all the nodes in the domain. The discrete equation
300 can be written as

301 Luh (xI ) = f (xI ), I ∈Ω − Γ (49a)

ED
302 u(xI ) = ū(xI ), I ∈Γ (49b)

303 The above is a set of algebraic equations whose unknowns are uI .

304 The collocation method has two major advantages, namely (i) efficiency in constructing the final system of equations
since no integration is required and (ii) shape functions are only evaluated at nodes rather than at integration points
CT

305

306 as in other methods. The price to pay is that, one must evaluate high-order derivatives of MMs shape functions which
307 is quite burdensome. In addition, two other drawbacks are difficulties in imposing natural boundary conditions and
308 non-symmetric stiffness matrix.
309 To better illustrate the method, consider the problem of a string on an elastic foundation with the governing equations:
E

d2 u
310 −a (x) + cu(x) + f = 0, 0 < x < 1; u(0) = u(1) = 0 (50)
dx2
RR

311 with specific parameters for solution are chosen a = 0.01, c = 1 and f = −1. The domain is divided into an equally
312 spaced set of nodes located at xJ , J = 1, . . . , N where the boundary points are nodes x1 and xN . By imposing the
313 equations given in Eq. (50) at the N nodes, we obtain the following equations:
 N   N 
d2  
−a 2 ΦI (xJ )uI + c ΦI (xJ )uI + f = 0, J = 2, . . . , N − 1
CO

314 (51a)
dx
I=1 I=1


N 
N
315 ΦI (x1 )uI = 0, ΦI (xN )uI = 0 (51b)
I=1 I=1
UN

316 The Eq. (51a) is rewritten in the familiar form:

 
 N N
317 −a ΦI,xx (xJ ) + c ΦI (xJ ) uI + f = 0, J = 2, . . . , N − 1 (52)
I=1 I=1

318 which is of the familiar form Ku = f where the assembly procedure is performed by looping on separate sets of nodes
319 (herein, there are interior and essential boundary nodes).

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
14 N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx

F
OO
PR
Fig. 9. String on elastic foundation: point collocation solutions.

320 It is worth noting that, using the point collocation method, one must deal with high-order derivatives (here second
321 order).5 Hence the meshless shape functions must have at least continuous second-order derivatives, which is the case
322 if the kernel (weight) function is C2 continuous. The numerical solution obtained with the point collocation method is
ED
323 given in Fig. 9.

324 2.7.2. Galerkin methods

325 The trial and test functions in Galerkin methods are given by
CT


N 
N
326 uh (x) = ΦI (x)uI , δuh (x) = ΨI (x)δuI (53)
I=1 I=1

327 If different shape functions are used for the approximation of the test and trial functions, a Petrov–Galerkin method
is obtained, otherwise we have a Bubnov–Galerkin method.6 We will assume now that ΨI = ΦI though all derivations
E

328

329 apply also for a Petrov–Galerkin method.

330 As an example, the problem of a string on an elastic foundation is solved again, but now with a Galerkin-based
RR

331 meshless methods. The weak form of this problem is

1
1
1
332 a vx ux dx + c vu dx + f v dx = 0 (54)
0 0 0

where v is the test function. The discrete equations are obtained by substituting the approximations of u and v into the
CO

333

334 above:


1 
1
1
335 a ΦI,x ΦJ,x dx + c ΦI ΦJ dx uJ + f ΦI dx = 0 (55)
0 0 0

The above has the familiar matrix form Ku = f where

UN

336

  1
337 KIJ = aΦI,x ΦJ,x + cΦI ΦJ dx, fI = −f ΦI dx (56)
0 0

5 The second derivatives of MLS shape functions are given in Section 3.2.
6 Often called Galerkin method.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx 15

338 The exact solution of this problem is given by

sinh(mx)  c 1/2
339 u(x) = 1 − cosh(mx) − (1 − cosh(m)) , m= (57)
sinh(m) a
340 The numerical solutions obtained with the element free Galerkin method is given in Fig. 10.

F
OO
PR
ED
E CT
RR
CO
UN

Fig. 10. String on elastic foundation: EFG solutions.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
16 N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx

341 2.8. Discrete equations for elastostatics

342 Consider a domain Ω, bounded by Γ . The boundary is partitioned into two sets: Γu and Γt . Displacements are
343 prescribed on Γu whereas tractions are prescribed on Γt . The weak form of linear elastostatics problems is to find u in
344 the trial space,7 such that for all test functions δu in the test space8 :

345 ε(u) : C : ε(δu) dΩ = t̄ · δu dΓ + b · v dΩ (58)

Ω Γ Ω

F
346 Substitution of approximations for u and δu into the above gives the discrete equations:

OO
347 Ku = f (59)

348 with

349 KIJ = BTI CBJ dΩ, fI = ΦI t̄dΓ + ΦI b dΩ (60)

Ω Γt Ω

PR
350 In two dimensions, the B matrix is given by
⎡ ⎤
ΦI,x 0
⎢ ⎥
351 BI = ⎣ 0 ΦI,y ⎦ (61)
ΦI,y ΦI,x
ED
352 Note that we have omitted Dirichlet boundary conditions in our formulations. The incorporation of Dirichlet bound-
353 ary conditions will be discussed in the next section. Note also that if an extrinsic basis is used the nodal vector u
354 will contain additional unknowns, see Section 2.6. Different methods can now be constructed by using different shape
355 functions. If we choose Dirac delta functions for the test function, we have a collocation method. Otherwise we obtain
356 a Galerkin method.
CT

357 2.8.1. Integration

358 The major disadvantage of MMs using Galerkin method is the numerical integration of the weak form. This is due
359 to the non-polynomial (rational) form of most meshless shape functions (MLS for instance). So, exact integration is
E

360 difficult to impossible for most meshfree methods. The most frequently used techniques include:
361 Direct nodal integration. The integrals are evaluated only at the nodes that also serve as integration points:


RR

362 f (X) dΩ = f (XJ )VJ (62)

Ω J ∈S

363 The quadrature weights VJ are usually volume associated with the node. The volume is obtained from a Voronoi
364 diagram that is constructed at the beginning of the computation. This approach is more efficient than using full
CO

365 integration. However, nodal integration leads to instabilities due to rank deficiency similar to reduced integrated finite
366 elements. We would also like to remark that nodal integrated meshless methods are very similar to meshless collocation
367 methods [13,5,10].
368 Stabilized nodal integration. Chen et al. [31] proposed the stabilized confirming nodal integration using strain
369 smoothing. They recognized that the vanishing derivatives of the meshfree shape functions at the particles cause of the
370 instabilities. In their strain smoothing procedure, the nodal strains are computed as the divergence of a spatial average
UN

371 of the strain field. The strain smoothing avoids evaluating derivatives of the shape functions at the nodes and hence
372 eliminates defective modes. An excellent overview of different methods to stabilize nodal integration is given by Puso
373 et al. [85].

7 Contain C0 functions.
8 Contain C0 functions but vanishes on Γu .

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx 17

374 Stress point integration. Adding additional stress points to the nodes is another possibility to avoid instabilities due
375 to rank deficiency:

 
376 f (X) dΩ = f (XJ )VJN + f (XJ )VJS (63)
Ω
J ∈ SN J ∈ SS

377 where the superimposed N denote nodes and the superimposed S denote stress points. Note that all kinematic values
378 are obtained via the nodes and only stresses are evaluated at these stress points. This concept of stress points was

F
379 first introduced in an SPH setting in one dimension by Dyka and Ingel [46] and later on extended into higher-order
380 dimensions by Randles and Libersky [89] and Belytschko et al. [9]. Note that there is a subtle difference between
the stress point integration of Randles and Libersky [89] and Belytschko et al. [9]. While Randles and Libersky [89]

OO
381

382 evaluate stresses only at the stress point, Belytschko et al. [9] evaluate stresses also at the nodes. A slightly different
383 approach was proposed by Cueto-Felgueroso et al. [34]. For large deformations, rules have to be found to move the
384 stress points.
385 Support-based integration. In the method of finite spheres, the integration is performed on every intersections of
386 overlapping supports. A truly meshfree method for integrating the weak form over overlapping supports, related to the

PR
387 supports of the meshfree approximation was developed independently by Duflot and Nguyen-Dang [43](called moving
388 least square quadrature) and Carpinteri et al. [27](called partition of unity quadrature). This integration technique is
389 improved in Carpinteri et al. [28] and Zhang et al. [103] to take cracks into account.
390 Background mesh or cell structure. The domain is divided into integration cells over which Gaussian quadrature is
391 performed:


ED
392 f (X) dΩ = f (ξ J )wJ det J ξ (ξ) (64)
Ω J

393 where ξ are local coordinates and det J ξ (ξ) is the determinant of the Jacobian, i.e. the mapping from the parent into
394 the physical domain. If a background mesh is present, nodes and the vertices of the integration usually coincide
(as in conventional FEM meshes, Fig. 11). When cell structures are utilized, a regular array of domains is created,
CT

395

396 independently of the particle position [38].

397 MMs which are based on local weak form such as MLPG adopts the integration over the shape function supports
398 or intersection of supports. Interested readers should refer to [2–4,70] and references therein for details.
399 Methods based on nodal and stress point integration are frequently employed in dynamics and where large defor-
E

400 mations are expected. We will consider only methods that employ Gauss quadrature and utilize a background mesh.
401 These methods are more accurate and they are ideally applicable to small and moderate deformation.
RR
CO
UN

Fig. 11. Integration in Galerkin-based MMs: background mesh (left) and background structure cells (right).

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
18 N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx

402 2.8.2. Essential boundary conditions

403 Due to the lack of the Kronecker delta property of MMs shape functions, the essential boundary conditions cannot
404 be imposed as easily as in FEM. Several techniques have been proposed, namely (i) methods based on the modification
405 of the weak form and (ii) methods using the modified shape functions. This section gives a brief description of these
406 methods, for more details, one should refer to [57].
407 Methods based on the modification of the weak form includes the Lagrange multiplier method, the penalty method
408 and Nitsche’s method. In order to understand these methods, the so-called variational principle should be first presented.
409 A variational principle specifies a scalar quantity, named functional Π, which is defined by an integral form:

410 Π= F (u, ux , . . .) dΩ + E(u, ux , . . .) dΓ (65)

OO
Ω Γ
411 where u is the unknown function, F and E are differential operators. The solution to the continuum problem is a function
412 u which makes Π stationary with any arbitrary variations δu:
413 δΠ = 0 with any δu (66)

PR
414 2.8.2.1. Lagrange multiplier. Let us consider a general problem of making a functional Π stationary with constraints:
415 C(u) = 0 on Γ (67)
416 To satisfy the above constraint, we build the following functional:

417 Π̄(u, λ) = Π(u) + λT C(u) dΓ (68)

ED
Γ
418 The variation of this new functional is given by

419 δΠ̄ = δΠ + δλ C(u) dΓ +

T
λT δC(u) dΓ (69)
Γ Γ
In order to derive the discrete equations, the Lagrange multipliers must be approximated:
CT

420


l
421 λ= NIL (x)λI (70)
I=1

There are several choices for the approximation space for the Lagrange multipliers, i.e. choices of NIL (x), namely,
E

422

423 (i) finite element interpolation on the boundary Γ , (ii) meshless approximations on this boundary and (iii) the point
424 collocation method which uses the Dirac delta function:
RR

425 NIL (x) = δ(x − xLI ) (71)

426 where xLI
is a set of points locating along the boundary Γ . Using this method, the system of equations of elastostatics
427 problem is given by
    
CO

K G u f
428 = (72)
G T
0 λ q
429 with

 
ΦI (xK ) 0
430 GIK = − ΦI NK dΓ = − (73)
UN

Γu 0 ΦI (xK )

431 qK = − NK ū dΓ = −ū(xK ) (74)

Γu

432 It is obvious that one drawback of the Lagrange multiplier method is the introduction of additional unknowns to the
433 problem. In addition, from Eq. (72), there are now zero terms on the diagonal of the matrix which makes the matrix
no longer positive definite.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx 19

434 2.8.2.2. Penalty function. We have the functional for the problem given in previous section:

α
435 Π̄(u, α) = Π(u) + C(u)T C(u) dΓ (75)
2 Γ
436 Applying the penalty method to elastostatics, we obtain the following weak form:

437 εT (u)Cε(v) dΩ = t̄ · v dΓ + b · v dΩ + α u · v dΓ − α ū · v dΓ (76)

Ω Γ Ω Γu Γu

F
438 which gives the equation Ku = f, where

OO
439 KIJ = BTI CBJ dΩ − α ΦI ΦJ dΓ (77)
Ω Γu

440 fI = ΦI t̄ dΓ + ΦI b dΩ − α ΦI ū dΓ (78)
Γt Ω Γu

441 The main advantage of the penalty method compared with the Lagrange multiplier approach is that no additional

PR
442 unknowns required. However, the conditioning of the matrix much depends on the choice of the penalty parameter.
443 Recently, the augmented Lagrangian method has been proposed by Ventura [98] to handle essential boundary
444 conditions in meshfree methods. This method has been shown to be stable and effective, particularly in contact
445 problems where it has replaced the penalty and Lagrangian multipliers methods.

446 2.9. Discontinuities

ED
447 There are mainly four approaches to model discontinuities in meshless methods, namely (i) modification of weight
448 function such as the visibility method, diffraction method and transparency method [11,63,84], (ii) modification of
449 intrinsic basis [50] to incorporate special functions, (iii) methods based on an extrinsic MLS enrichment [50] and (iv)
450 methods based on the extrinsic PUM enrichment [99]. More recently, the augmented Lagrangian method has been used
CT

451 to model strong discontinuity (crack problems) in Carpinteri [25] and weak discontinuity (material discontinuity) in
452 Carpinteri [26].

453 2.9.1. Modification of weight function

The visibility method [8,15] was the earliest method that incorporates strong discontinuities into meshless methods.
E

454

455 In the visibility method, the crack boundary is considered to be opaque. Nodes that are on the opposite site of the
456 crack are excluded in the approximation of the displacement field. Difficulties arise for particles close to the crack tip
RR
CO
UN

Fig. 12. Undesired introduced discontinuities by the visibility method.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
20 N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx

F
OO
Fig. 13. (a) Scheme of the visibility method and (b) scheme of the diffraction/transparency method.

since undesired interior discontinuities occur, see Fig. 12. Non-convex boundaries cannot be treated by the visibility

PR
457

458 criterion correctly either.

459 The diffraction method [84] is an improvement of the visibility method. It removes the undesired interior discon-
460 tinuities as shown in Fig. 13. The diffraction method is also suitable for non-convex crack boundaries. The method is
461 motivated by the way light diffracts around a sharp corner but the equations used in constructing the domain of influence
462 and the weight function bear almost no relationship to the equation of diffraction. The method is only applicable to
radial basis kernel functions with a single parameter. The idea of the diffraction method is to treat the crack as opaque
ED
463

464 but to evaluate the length of the ray by a path which passes around the corner of the discontinuity, see Fig. 13. It
465 should be noted that the shape function of the diffraction method is quite complex with several areas of rapidly varying
466 derivatives that complicates quadrature of the discrete Galerkin form. Moreover, the extension of the diffraction method
467 into three dimensions is complex.
468 The transparency method was developed as an alternative to the diffraction method by Organ et al. [84]. The
CT

469 transparency method is easier to extend into three dimensions than the diffraction method. In the transparency method,
470 the crack is made transparent near the crack tip. An additional requirement is usually imposed for particles close to
471 the crack. Since the angle between the crack and the ray from the node to the crack tip is small, a sharp gradient in
472 the weight function across the line ahead of the crack is introduced. In order to reduce this effect, Organ et al. [84]
imposed that all nodes have a minimum distance from the crack surface.
E

473

474 2.9.2. Modification of intrinsic basis

RR

475 In methods that use an intrinsic basis such as the EFGM, the intrinsic basis can be modified according to the
476 crack kinematics [50]. In LEFM, generally the asymptotic near-tip displacement field of the Westergaard solution is
477 introduced into the basis p:
 √ √ √ √
478 pT (X) = 1, X, Y, r sin(θ/2), r cos(θ/2), r sin(θ/2) sin(θ), r cos(θ/2) sin(θ) (79)
CO

479 where r is the radial distance to the crack tip and θ the angle to the crack. One drawback of intrinsic enrichment is that
480 it has to be used in the entire domain. Otherwise, undesired discontinuities are introduced. To reduce computational
481 cost, a blending domain is often introduced where the higher-order basis is decreased to a basis of lower-order (in our
482 case linear complete basis) continuously.
483 [44] suggested an alternative intrinsic enrichment by enriched kernel functions:
UN

√ θ
wc (X) = α r cos w4 (X)
 2
√ θ
484 wp (X) = α r 1 + sin w4 (X) (80)
2

√ θ
wp (X) = α r 1 − sin w4 (X)
2

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx 21

485 where w4 (X) is the quartic spline and the factor α controls the amplitude of the enriched kernel function compared
486 with the amplitude of the regular nodes. The value of α is usually set to 1. The indices c, m and p stand for cos, minus
487 sin and plus sin, respectively. An advantage of this method is that no blending domain needs to be introduced.

488 2.9.3. Methods based on an extrinsic MLS enrichment

489 Another opportunity to model cracks in meshless methods is to introduce the analytical solution extrinsically [50]:

 
nc
 

F
490 uh (X, t) = p(XJ )T a(X, t) + I (XI ) + kI QII (XI )
kIK QK K K
(81)
J ∈S K=1

OO
491 where nc is the number of cracks in the model, uh is the approximation of u, p is the usual polynomial basis and kI
492 and kII are additional degrees of freedom associated with mode-I fracture and mode-II fracture. The functions QiI and
493 QiII , i = 1, 2 describe the near-tip displacement field and are given by
!
1 r
QI (X) =
1
cos(0.5θ)(κ − 1 + 2 sin2 (0.5θ)) (82)

PR
494
2G 2π
!
1 r
495 QI (X) =
2
sin(0.5θ)(κ + 1 − 2 cos2 (0.5θ)) (83)
2G 2π
!
1 r
496 Q1II (X) = sin(0.5θ)(κ + 1 + 2 cos2 (0.5θ)) (84)
ED
2G 2π
!
1 r
497 Q2II (X) = − cos(0.5θ)(κ − 1 − 2 sin2 (0.5θ)) (85)
2G 2π
where G is the shear modulus and κ is the Kolosov constant defined as κ = 3 − 4ν for plane strain and κ = (3 −
CT

498

499 ν)/(1 + ν) for plane stress conditions where ν is the Poisson ratio.
500 One advantage of the MLS extrinsic enrichment is that the stress intensity factors can be directly obtained without
501 considering the J-integral. Therefore, the enrichment has to be introduced globally, which comes with additional
502 computational cost.
E

503 2.9.4. Methods based on an extrinsic PUM

RR

504 Motivated by the XFEM [79], an extrinsic PU enrichment for meshless methods was presented in [99]:

   
4
505 uh (x) = ΦI (x)uI + ΦJ (x)H(x)aJ + ΦK (x) bαK (86)
I ∈S J ∈ Sc K ∈ Sf α=1
CO

506 where ΦI are MLS shape functions. The Heaviside function and the branch functions are given by

+1 if (x − x∗ ) · n ≥ 0
507 H(x) = (87)
−1 otherwise

where x∗ is the projection of point x on the crack:

UN

508

" #
√ θ √ θ √ θ √ θ
509 B ≡ [B1 , B2 , B3 , B4 ] = r sin , r cos , r sin cos θ, r cos cos θ (88)
2 2 2 2

510 where r and θ are polar coordinates in the local crack front coordinate system. A two-dimensional plot of the branch
511 functions is shown in Fig. 14 The set Sc includes the nodes whose support contains point x and is cut by the crack, see
512 Fig. 15 whereas the set Sf are nodes whose support contains point x and the crack tip xtip , see Fig. 16.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
+Model
MATCOM 2939 1–48 ARTICLE IN PRESS
22 N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx

F
Fig. 14. Two-dimensional plot of branch functions. It is clear that the first function is discontinuous through crack face.

OO
PR
Fig. 15. The elements of set Nc are nodes whose support contains point x and cut by the crack.
ED
E CT
RR
CO
UN

Fig. 16. The elements of set Nf are nodes whose support contains point x and the crack tip xtip .

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx 23

F
OO
PR
Fig. 17. Enrichment function Ψ with discontinuous derivative.

513 Using the Galerkin procedure as described in previous sections, the usual discrete equations are obtained with only
514 one difference in the B 9 matrix which is now larger:
515 B = [Bstd |Benr ] (89)
ED
516 where Bstd is the standard B and Benr is the enriched B matrix:
⎡ ⎤
(ΦI )x ΨI + ΦI (ΨI )x 0
⎢ ⎥
517
I =⎣
Benr 0 (ΦI )y ΨI + ΦI (ΨI )y ⎦ (90)
(ΦI )y ΨI + ΦI (ΨI )y (ΦI )x ΨI + ΦI (ΨI )x
CT

518 where ΦI (x) can be either the Heaviside function H(x), or the branch functions Bα (x). This enriched EFG can be
519 implemented within an available EFG code with little modification.

520 2.9.5. Discontinuous derivatives

E

521 For PDEs with discontinuous coefficients, the solutions usually have discontinuous derivatives along the disconti-
522 nuity. While it is trivial to treat discontinuous derivatives such as material interfaces in FEM by meshing the domain
such that the element edges are aligned with the interface, it is not so simple in MMs. There are different approaches
RR

523

524 in treating discontinuous derivatives such as the Lagrange multiplier method, the global enrichment approach [13],
525 the local or PUM-enrichment strategy [92]. In the global enrichment method, a special function Ψ whose derivative is
526 discontinuous through the line of discontinuity (material interface for instance) is added into the approximation space:

527 uh (x) = ΦI (x)uI + bΨ (x − xa ) (91)
CO

I
528 where Ψ (x) is the enrichment function having the form (see Fig. 17 to see its discontinuous derivative) and b is
529 additional unknown of problem:

530 Ψ (x) = x − xa − φI (x)xI − xa (92)
I
UN

531 with

0 if x < 0
532 x = (93)
x if x ≥ 0

9 With the assumption that nodes on essential and natural boundaries are not enriched. For more details, refer to [93].

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
24 N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx

533 As an example, consider the following problem:

534 (E(x)ux )x + x = 0, 0 ≤ x ≤ 10; u(0) = u(10) = 0 (94a)

1 0≤x<5
535 E(x) = (94b)
0.5 5 ≤ x ≤ 10

536 The weak form of this problem is given by

1
1

F
537 − vx (x)E(x)ux (x) dx + v(x)x dx = 0 (95)
0 0

OO
538 Trial and test functions are constructed by Eq. (91), it results in (using the arbitrariness of δui and δb):

1
1
1
539 ΦI,x (x)E(x)ΦJ,x (x)uJ + ΦI,x (x)E(x)Ψx (x)b dx − ΦI (x)x dx = 0,
0 0 0

1
1
1
Ψx (x)E(x)ΦJ,x (x)uJ dx + Ψx (x)Ψx (x)b dx − Ψ (x)x dx = 0 (96)

PR
540
0 0 0
541

542 In matrix form:

    
K B u f
543 = (97)
BT g1 b g
ED
544 with

1
1
545 KIJ = ΦI,x (x)E(x)ΦJ,x (x) dx, BI1 = ΦI,x (x)E(x)Ψx (x) dx (98)
0 0

1
1
1
CT

546 fI = ΦI (x)x dx, g= Ψ (x)x dx, g1 = Ψx (x)Ψx (x) dx (99)

0 0 0
547 The results obtained with this enrichment are plotted in Fig. 18. It is clear that without enrichment, the discontinuity
548 in strain cannot be captured.
549 It is clear that with this global enrichment method, one must choose smartly the enrichment function, namely
E

550 this function must have local character (discontinuous derivative through a material interface, for instance) and zero
551 elsewhere. The choice of this function is not trivial in two dimensions, especially for complex interfaces. For these cases,
RR
CO
UN

Fig. 18. Strain computed with and without enrichment.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx 25

F
OO
PR
Fig. 19. Numerical solution with the PUM enrichment strategy.

552 the PUM-enrichment strategy best. To model a discontinuity in the derivative (or weak discontinuity), the following
553 approximation is used:
 
ED
554 uh (x) = ΦI (x)uI + ΦJ (x)|f (x)|aJ (100)
I ∈S J ∈ Sc

555 where f is the signed distance to the discontinuity line and Sc is the set of nodes whose support is cut by this line.
556 The problem of a one-dimensional bar with two materials is now solved again with this so-called local enrichment
CT

557 strategy. In the computation, 21 equally spaced particles are used, in each of the 20 intervals, 3 Gauss points are
558 employed. The size of the circular nodal support is 2 x where x is the nodal spacing. The solution is given in Fig. 19
559 and a comparison with the solution obtained by 21 linear finite elements is also given.
E

560 2.10. Error estimation and adaptivity

561 Due to the absence of a mesh, h-adaptivity is easier to incorporate in MMs than in mesh-based methods. Also
RR

562 p-adaptivity and r-adaptivity is conceptually easier to implement in a meshfree framework. To drive the adaptivity,
563 a local error estimator is necessary. The most significant works on error estimation in the frame of MMs are as
564 follows. [39] present an a posteriori error estimator and use it in an hp-adaptive method. [64] developed an a posteriori
565 approximation error in order to adaptively refine corrected derivative meshfree methods. [32] suggest a residual-based
566 error estimator based on the difference between a recovered stress field and a raw EFG field, like in the well-known
CO

567 ZZ error estimator in the FEM [104]. This estimator is used in an adaptive method for static cracks in [33] and for
568 propagating cracks in [69]. This estimator is also found in [67,68]. Other estimators and adaptive methods are proposed
569 in [59,66,87,86,102,52,51,53]. Global strict bounds on the energy are obtained by a dual meshfree method in [42]. An
570 excellent overview on adaptive Galerkin meshfree methods is given in [76].
UN

571 3. Computer implementation aspects

572 There are considerable differences between the finite element methods and meshless methods, which leads to
573 different computer implementation of MMs compared to FEM. We could cite (i) computation of shape functions and
574 their derivatives, (ii) assembly procedure, (iii) imposing essential boundary conditions and (iv) post-processing step.
575 This section gives details on how to write an EFG code. In addition, the PUM-enriched EFG is also presented. The
Matlab language is chosen.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
26 N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx

576 3.1. General meshless procedure

577 1. Node generation including node coordinates and associated weight functions. At each node, one must specify (i)
578 the shape of the domain of influence (for example, circular), (ii) size of this support (radius for circular support)
579 and (iii) the functional form (for instance the quartic spline function).
580 2. Insert integration points (coordinates and weights) in the domain.
581 3. Insert integration points along traction and essential boundaries.
582 4. Integrate on the domain. For each Gauss point xg :

F
583 • Find nodes within the support of xg . For each of these nodes, compute.
584 • weight function, shape function and shape function derivatives.

OO
585 • Compute Bmatrix.
586 • Compute and assemble K matrix.
587 5. Integrate on the boundaries. Integrate forces along the traction boundary to form the nodal force vector f and also
588 on the essential boundary to impose essential boundary conditions.
589 6. Solve the resulting system of equations (obtain the fictitious displacement field, if the approximation does not have
590 the Kronecker delta property).

PR
591 7. Reconstruct the true nodal displacement from the fictitious displacements.

592 3.2. Efficient shape function computation

593 The computation of the MLS shape functions as well as its derivatives involves the inverse of the moment matrix
594 which becomes burdensome in two and three dimensions. An efficient approach, presented in [11,41] is reproduced
ED
595 here for completeness.
596 In order to avoid the direct computation of the inverse of the moment matrix, the MLS shape function is usually
597 written in the form:
598 ΦI (x) = cT (x)wI (x)p(xI ), where A(x)c(x) = p(x) (101)
CT

599 with

n
600 A(x) = wI (x)p(xI )pT (xI ) (102)
I=1
E

601 To efficiently compute c(x), the LU factorization of A is performed together with backward substitution:
LUc(x) = p(x), Uc(x) = L−1 p(x), c(x) = U−1 L−1 p(x) (103)
RR

602

603 The first derivatives of shape functions are given by

604 ΦI,k (x) = cTk (x)p(xI )wI (x) + cT (x)p(xI )wI,k (x) (104)
605 with
CO

606 ck (x) = A−1 −1 −1 −1 −1

k (x)p(x) + A (x)pk (x) = −A (x)Ak (x)A (x)p(x) + A (x)pk (x)

607 = A−1 (x)[−Ak (x)c(x) + pk (x)] = A−1 (x)bk (105)

608 and
UN


n
609 Ak (x) = wi,k (x)p(xi )pT (xi ) (106)
i=1

610 The second derivatives (usually necessary in point collocation method, in plate and shell modeling or for stabilization
611 schemes based on finite increment calculus) are computed in the same manner:
612 ΦI,kl (x) = cTkl (x)p(xI )wI (x) + cTk (x)p(xI )wI,l (x) + cTl (x)p(xI )wI,k (x) + cT (x)p(xI )wI,kl (x) (107)

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx 27

613 with
 
614 ckl (x) = A−1 (x) pkl (x) − Al (x)ck (x) − Ak (x)cl (x) − Akl (x)c(x) (108)

615 where


n
Akl (x) = wI,kl (x)p(xI )pT (xI ) (109)

F
616

I=1

OO
617 Listing 1. Stiffness matrix assembly.

PR
ED
E CT
RR
CO
UN

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
28 N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx

F
618 3.3. Gauss point generation

OO
619 Assuming that the integration is performed with background integration cells. In two dimensions, each integra-
620 tion cell is a four node quadrilateral element with shape functions NI and nodal coordinates x0 (I = 1, . . . , 4).
621 For each Gauss point (ξ gp , wgp ) of a given cell, the isoparametric mapping is used to get its global coordinates
622 xgp :

PR
4
623 xgp = NI (ξ gp )xI (110)
I=1

624 and its global weight is given by

ED
625 w = wgp × detJ (111)

626 where J is the Jacobian of the physical-parent transformation.

CT

627 3.4. Assembly procedure

628 The assembly procedure in MMs is performed on the domain of influence of the point under consideration (often a
629 Gauss point). If we store the nodal unknowns uI as follows:
E

630 uT = [u1 v1 u2 v2 . . . un vn ] (112)

RR

631 where n is the number of nodes. Then a node I will contribute to the (2I − 1) th row and the (2I) th column. If we
632 denote the variable index containing the number of nodes within the support of a given Gauss point, then the assembly
633 procedure at this Gauss point is given by.
634 Listing 2. Stiffness matrix assembly.
CO
UN

In collocation methods, assembly of the stiffness matrix is done row by row.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx 29

635 3.5. Integration on the essential boundaries

636 3.5.1. Point collocation method

637 Recall the formulas for matrix G and vector q:

638 GIK = −ΦI (xK )S (113)

F
639 qK = −Sū(xK ) (114)

OO
640 where S is a diagonal matrix of size 2 × 2 in two dimensions, and Sii = 1 if the displacement is imposed on xi and
641 Sjj = 0 otherwise. xK are collocation points.
642 Assume that, along the essential boundary Γu , m collocation points are used. Then we have m × 2 con-
643 straint equations (in two dimensions). Hence, the dimension of G is 2n × 2m with n the number of nodes in the
644 domain.

PR
645 Listing 3. Point collocation for imposing essential BCs.

ED
E CT
RR
CO

646 3.5.2. Finite element interpolation for Lagrange multiplier

647 For ease of reading, the equations are recalled:

648 GIK = − ΦI NK S dΓ, qK = − NK Sū dΓ (115)

Γu Γu
UN

649 Let us discretize the essential boundary with (m − 1) two-noded finite elements. For each element, ngp Gauss points
650 are used. The shape functions for a two-node element are given by

x
651 N1 (x) = 1 − , N2 (x) = 1 − N1 (x) (116)
le

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
30 N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx

652
Listing 4. Finite element Lagrange multipliers.

F
OO
PR
ED
E CT

653 3.6. Enriched EFG

RR

654 In comparison to the EFG, the enriched EFG has the following differences:

655 • Detection of non-enriched and enriched particles.

CO
UN

Fig. 20. Selection of enriched particles (filled particles): (a) discontinuous enriched particles; (b) near tip enriched particles.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx 31

656 • Treatment of enriched (additional) degrees of freedom.

657 • Computation of stiffness matrices.

658 The selection of enriched particles with circular support is illustrated in Fig. 20. It suffices to compute the signed
659 distances from particles to the crack line and the distances from particles to the crack tip and compare these distances
660 to the radius of the domains of influence. This procedure, implemented in Matlab, is given in the following listing.
661 Listing 5. Selection of enriched particles.

F
OO
PR
ED
662 Due to the presence of additional doffs, the assembly procedure has to be revised. We use fictitious nodes to handle
663 these additional doffs. At a H-enriched node (discontinuous enrichment), we add one fantom node and, at a tip enriched
664 node, we add four fantom nodes. The numbering of these fictitious nodes start from the total number of true nodes plus
one. For example, if there are five nodes numbered from one to five where the third node and fifth is enriched with the
CT

665

666 Heaviside function, the fourth node is a near tip enriched one. Then, we have 5 + 2 × 1 + 1 × 4 = 11nodes. Then, at
667 the third node, we add a phantom node numbered 6, at the fourth node, we add four fantom nodes numbered 7, 8, 9,
668 10 and at the fifth node, a fantom node numbered 11 is added. An array named pos is built to contain the number of
669 these fantom nodes. It is an array of dimension numnode × 1 where numnode is the number of true nodes. For this
example, pos is pos = [0 0 6 7 11].
E

670

671 Listing 6. Selection of enriched particles.

RR
CO
UN

672 In two dimensions, at a certain node numbered i there are always two unknowns associated with equation num-
673 bers 2i − 1 and 2i in the global matrix. If this node is a discontinuous-enriched node, then it has two additional
674 unknowns associated with equation numbers at 2 × pos(i) − 1 and 2 × pos(i) in the global matrix. If it is a near tip
675 enriched node, then it has eight additional unknowns with equation numbers (2 × pos(i) − 1, 2 × pos(i)),(2 × (pos(i) +
676 1) − 1, 2 × (pos(i) + 1)),(2 × (pos(i) + 2) − 1, 2 × (pos(i) + 2)) and (2 × (pos(i) + 3) − 1, 2 × (pos(i) + 3)), where
677 each pair corresponds to each added fantom node. Listing 7 gives the implementation in Matlab of this assembly
678 procedure.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
32 N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx

679 Listing 7. Assembly procedure.

F
OO
PR
ED
E CT
RR
CO
UN

680 The B matrix at a Gauss point gp is composed of two parts: the standard and the enriched part, where the standard
681 part is always computed and the enriched part is only computed if in the nodes whose supports cover gp, there exist
enriched nodes. This is implemented in Matlab and given in the following listing.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx 33

Listing 8. Stiffness matrix computation.

682

F
OO
PR
ED
E CT
RR
CO
UN

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
34 N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx

F
OO
683 4. Numerical examples PR
ED
684 In this section, numerical examples in linear elasticity are presented with the purpose to verify the Matlab
685 code.

686 4.1. The Timoshenko beam

CT

687 Consider a beam of dimensions L × D, subjected to a parabolic traction at the free end as shown in Fig. 21. The
688 beam is considered to be of unit depth and is in plane stress state. This problem was solved previously by Dolbow and
689 Belytschko [37].
690 The parabolic traction is given by
E


P D2
691 ty (y) = − −y 2
(117)
2I 4
RR

692 where I = D3 /12 is the moment of inertia. The exact displacement of this problem is
"  #
Py D2
693 ux (x, y) = − (6L − 3x)x + (2 + ν) y2 − ,
6EI 4
" #
CO

P D2 x
694 uy (x, y) = 3νy2 (L − x) + (4 + 5ν) + (3L − x)x2 (118)
6EI 4
UN

Fig. 21. The Timoshenko beam.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx 35

Table 1
Comparison of vertical displacement at end of beam
Nodes uy exact uy EFG Error (%)

7×5 −0.0089 −0.0083 −6.74

11 × 5 −0.0089 −0.0087 −2.24
15 × 9 −0.0089 −0.0088 −1.12
20 × 9 −0.0089 −0.0088 −1.12

F
695 and the exact stresses are


OO
P(L − x)y P D2
696 σx (x, y) = − , σxy (x, y) = − − y2 , σy (x, y) = 0 (119)
I 2I 4
697 In the computations, material properties are taken as E = 3.0 × 107 , ν = 0.3 and the beam dimensions are D = 12
698 and L = 48. The shear force is P = 1000. The regular node distribution together with the background mesh used for
699 numerical integration of the weak form are shown in Fig. 22. In each integration cell, 4 × 4 Gauss quadrature is used.

PR
700 The linear basis and cubic spline weight function is used in the MLS approximation. Circular nodal support of radius
701 3.5 times the nodal spacing is employed.
702 We check the error in the energy and displacement norm. The energy norm is given by
"
#1/2
1  
703 eenergy = εnum − εexact D (εnum − εexact ) dΩ
T
(120)
2 Ω
ED
704 and the displacement norm is given by

705 edisplacement = sqrt [(unum − uexact ) · (unum − uexact ) dΩ] (121)

Ω
706 where εnum and εexact are the numerical strain vector and exact strain vector, respectively. The same notation applies
CT

707 to the displacement vector unum and uexact . The calculation has been done with the same Gauss quadrature as given
708 above and the node distributions are 17 × 5, 33 × 9, 65 × 17 and 133 × 34.
709 In Table 1, the vertical displacement at point (L, 0) calculated by EFG is compared with the exact solution. This
710 table shows excellent agreement between EFG and the analytical solution.
The stresses at the center of the beam (x = L/2, y ∈ [−D/2, D/2]) are computed and compared with the exact
E

711

712 solution. Eleven Gauss points along the vertical line x = L/2 are used in the computation. Fig. 23 shows very good
713 agreement between the EFG result and the exact stresses. In addition, the distribution of the normal stress in the beam
RR

714 is plotted in Fig. 24. It is of particular interest that very smooth stresses were obtained without any additional treatment
715 as is necessary in FEM (stress extrapolation or stress recovery).
CO
UN

Fig. 22. 18 × 7 regular nodes distribution for the Timoshenko beam.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
36 N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx

F
OO
Fig. 23. Stresses comparison: (a) shear stress; (b) normal stress.

PR
716 In Fig. 25, we show the error in the energy and in the displacement norm. h is the horizontal spacing between the
717 nodes and e is the error. The convergence rate in the energy norm is 1.43 and the error in the displacement norm is still
718 missing.

719 4.2. Plate with hole

ED
720 Consider an infinite plate with a centered circular hole under unidirectional tension along the x-direction. The plate
721 Q5 dimension is taken to be L × L and the circle of radius a(Fig. 26).
722 The exact stress in the plate is given by

a2 3 3 a4
CT

723 σx (r, θ) = 1 − 2 cos 2θ + cos 4θ + cos 4θ (122a)

r 2 2 r4

a2 1 3 a4
724 σy (r, θ) = − 2 cos 2θ − cos 4θ − cos 4θ (122b)
r 2 2 r4

E

a2 1 3 a4
725 σxy (r, θ) = − 2 sin 2θ + sin 4θ + sin 4θ (122c)
r 2 2 r4
RR
CO
UN

Fig. 24. Distribution of normal stress in the beam.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx 37

F
OO
PR
ED
E CT
RR

Fig. 25. Convergence in the energy (top) and displacement (bottom) norm. The optimal rates are 1.0 and 2.0, respectively.
CO

726 where r, θ are the usual polar coordinates centered at the center of the hole.
727 The boundary conditions include (i) essential boundary conditions on the bottom uy (x, y = 0) and left edges
728 (ux (x = 0, x = L, y) = 0); (ii) natural boundary conditions on the right and top edges on which traction t̄ computed
729 from the exact stress given in Eq. (122) are applied. More precisely:
UN

730 t̄i = σij nj (123)

731 On the right edge, n = (1, 0), hence t̄T = (σx , σxy ). Similarly, on the top edge, the imposed traction is t̄T = (σxy , σy ).
732 In the computation, the material properties are taken as a Young modulus of 103 , a Poisson’s ratio equal to 0.3,
733 and the geometry is such that L = 10 and a = 1. The background mesh constructed for the numerical integration is
734 given in Fig. 27(a). Another possibility is the structured cells as shown in Fig. 27(b). However, the background mesh
735 is chosen for simplicity. For each background cell, a 4 × 4 Gauss quadrature is employed (Fig. 27).

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
38 N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx

F
OO
Fig. 26. Plate with a hole: (a) whole domain; (b) 1/4 model with irregular nodes.

PR
ED
E CT

Fig. 27. Gauss points: (a) background mesh; (b) structured cells.
RR

736 The cubic spline function serves as a weight function. The domain of influence for all nodes is a circle with varying
737 radius. They are chosen such that the support is small for nodes near the hole and bigger for nodes near the edges.
The essential boundary conditions are imposed with the boundary point collocation method where collocation points
CO

738

739 are coincident with nodes along the bottom and left edges.
740 The stress σx computed at nodes are plotted and compared to the exact solution (Fig. 28). With a coarse discretization
741 of 99 nodes, quite accurate results are obtained.
742 It would be great to show the error in the energy and displacement norm (or just one) as well.
UN

743 4.3. Infinite plate with a center crack

744 Consider an infinite plate containing a straight crack of length 2a and loaded by a remote uniform stress field σ.
745 Along ABCD the closed form solution in terms of polar coordinates in a reference frame (r, θ) centered at the crack
746 tip is

2(1 + υ) KI √ θ 2 θ
747 ux (r, θ) = √ r cos 2 − 2υ − cos ,
2π E 2 2

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx 39

F
OO
Fig. 28. Stress plot: (a) EFG σx ; (b) exact σx .

PR
2(1 + υ) KI √ θ 2 θ
748 uy (r, θ) = √ r sin 2 − 2υ − cos (124)
2π E 2 2
√
749 where KI = σ πa is the stress intensity factor, υ is Poisson’s ratio and E is Young’s modulus. ABCD is a square of
750 10 mm × 10 mm, a = 100 mm; E = 107 N/mm2 , υ = 0.3 and σ = 104 N/mm2 .
751 The geometry of the computational domain ABCD is shown in Fig. 29. Displacement of nodes on the bottom, right
ED
752 and top edges are prescribed by Eq. (124).
753 The crack is modeled by different techniques described in Section 2.9. The deformed configuration is plotted in
754 Fig. 30 for the enriched partition of unity method (see Section 2.9.4) where the level set method is used to represent
755 the geometry of the crack.
E CT
RR
CO
UN

Fig. 29. Infinite cracked plate under remote tension: geometry and loads.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
40 N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx

F
OO
PR
Fig. 30. Computed deformed configuration (scaled).

Numerical integration is performed on a background mesh of 20 × 20 rectangular elements. On each element, a

ED
756

757 6 × 6 Gauss quadrature is adopted.

758 The error in the energy norm is illustrated in Fig. 31(a) for the visibility, diffraction and transparency method. Note
759 that we did not use any crack tip enrichment. The convergence rate of these methods are similar though the absolute
760 error is smaller for the diffraction and the transparency method. Fig. 31(b) shows the normalized SIF KI . Also for
761 local convergence, the diffraction and transparency methods perform better. There is barely a difference in the results
CT

762 for the transparency and diffraction method.

763 The error in the energy norm is illustrated in Fig. 32(a) for the some enriched methods (with crack tip enrichment),
764 Sections 2.9.2 and 2.9.4. We have also included the results obtained with the visibility criterion in this figure. The
765 methods that include the crack tip enrichment give more accurate results and a much better convergence rate which is
expected of course. The most accurate results and the highest convergence rate of 0.94 are obtained with the extrinsic
E

766

767 MLS enrichment. However, the computational cost is higher since the enrichment is applied in the entire domain. The
768 extrinsic PU enrichment gives a convergence rate of 0.86 and is only slightly less accurate than the extrinsic MLS
RR

769 enrichment. The intrinsic PU enrichment lies in between these two results. The same observation can be made for local
770 convergence. The fact that the SIFs can be directly obtained is a major advantage of the extrinsic MLS enrichment and
CO
UN

Fig. 31. (a) Error in the energy for the mode I problem using the visibility, diffraction and transparency criterion; (b) normalized stress intensity
factor vs. h.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx 41

F
OO
Fig. 32. (a) Error in the energy for the mode I problem using the intrinsic and extrinsic PU enrichment, respectively, and the extrinsic MLS
enrichment; (b) normalized stress intensity factor vs. h.

PR
771 probably leads to more accurate results with respect to local convergence. Nevertheless, also the PU enrichment based
772 method give excellent local convergence, see the scale of the y-axis in Figs. 31(b) and 32(b).

773 4.4. Infinite plate with a center inclusion

ED
774 Consider an infinite plate with a center inclusion. For the computation, the square domain has dimensions 10 × 10
775 with a circular inclusion of radius 1 as shown in Fig. 33. The matrix properties are E1 = 1000, ν1 = 0.3 whereas
776 material characteristics of the inclusion are taken as E2 = 1, ν2 = 0.3. The traction along the vertical direction is
777 applied on the top edge while nodes along the bottom edge are constrained along the vertical direction. One more
778 constraint is imposed to avoid rigid body modes.
CT

779 This problem is solved with both PUM-enriched EFG and the extended finite element method XFEM (this XFEM
780 code written in Matlab and available from the website http://www.civil.gla.ac.uk/∼bordas). The particle arrangement
781 is uniform(20 × 20) as well as finite element mesh (20 × 20 four noded quadrilateral elements) are given in Fig. 34. It
782 is emphasized that the particles and the finite element mesh are independent of the shape and position of the inclusion.
783 For numerical integration of the EFG weak form, a background mesh is built with 6 × 6 Gauss quadrature for
E

784 each cell, whereas, for the XFEM weak form, usual Gauss quadrature is adopted for elements which are not cut
785 by the inclusion’s boundary, while, for elements that are cut by the inclusion boundary, element partitioning is
RR

786 used.
787 Fig. 35 gives a comparison of vertical displacement fields computed by both methods.
CO
UN

Fig. 33. Infinite plate with circular inclusion.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
42 N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx

F
OO
Fig. 34. Domain discretization with enriched nodes: (a) meshless particles; (b) finite element mesh.

788 4.5. Quasi-static crack propagation

789 We study the cracked speciment of Fig. 36, also known as double cantilever beam (DCB). The dimensions are as
follows: length L = 300 mm, height h = 100, initial crack length a = 138 mm, the load P = 100 N and mechanical

PR
790

791 properties E = 200 GPa and ν = 0, 3, and we assume plane strain conditions.
792 A perturbation of the crack direction is introduced at the crack tip: angle 4.8◦ and length dx = 12 mm, which,
793 physically, could be associated with the presence of a defect (void, inclusion) in the vicinity of the crack tip. The crack
ED
E CT
RR

Fig. 35. Vertical displacement field: (a) EFG solution and (b) XFEM solution.
CO
UN

Fig. 36. Geometry of the double cantilever beam.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx 43

F
OO
PR
Fig. 37. Stress intensity factors for a crack growth increment a = 5 mm at each step.
ED
794 propagation phenomenon is studied experimentally by Sumi et al. [96] who show the unstable nature of the crack path,
795 which either curves downwards or upwards. The crack path is represented by a dashed line in Fig. 36. Sumi [95] solve
796 this problem by the finite element method and by Fleming et al. [50] by meshfree methods (either with refined point
797 distributions in the vicinity of the crack tip or with intrinsic enrichment of the MLS basis). We solve the problem using
CT

798 both a standard EFG method (node distribution 49 × 17) and an enriched EFG method.
E
RR
CO
UN

Fig. 38. Crack path (a = 8 mm).

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
44 N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx

F
OO
PR
Fig. 39. Crack path (a = 5 mm).
ED
799 The evolution of the stress intensity factors during propagation are shown in Fig. 37 where we see that the crack
800 is predominantly loaded in mode I. The slightly negative SIF is sufficient to curve the crack path in the direction
801 of the top edge of the beam. We compare the experimental results to our numerical results in Figs. 38–40 for crack
802 propagation increments, at each step of 8, 5 and 3 mm, respectively. We note that enrichment significantly improves the
results.
CT

803

804 Fig. 41 shows the different crack paths obtained for three values of the crack growth increment. We note that the latter
805 should not be “too small”, otherwise, the nodal supports and J integral domains also become small which decreases
E
RR
CO
UN

Fig. 40. Crack path (a = 3 mm).

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx 45

F
OO
PR
Fig. 41. Crack paths for various crack growth increments.
ED
806 the role of near-tip enrichment and poor accuracy in the stress intensity factors calculation, which exhibit spurious
807 oscillations.

808 5. Conclusions
CT

809 We have given an overview of meshless methods, that are based on a global weak form, with emphasize on imple-
810 mentation aspects. The meshless methods described in this manuscript are especially well-suited for Solid Mechanics
811 applications and we have applied them to linear elastic material problems.
812 The meshless methods were classified into two categories, methods that are based on an intrinsic basis and methods
813 based on an extrinsic basis. The latter class of methods were found especially useful for problems where information
E

814 of the solution can be built into the approximation though it is also possible to include such information into certain
815 intrinsic meshless methods. We have discussed properties, advantages and drawbacks of meshless methods compared
RR

816 to standard finite elements for simple examples. A summary is already given in the introduction of this manuscript.
817 Since standard meshless methods do not fulfill the so-called Kronecker–Delta property, essential boundary condi-
818 tions cannot be applied as straightforward as in finite element methods. We have summarized different opportunities
819 how to incorporate essential boundary conditions and discussed more detailed two of them, the penalty method and
820 the Lagrange multiplier method. Both methods are also described with respect to their implementation.
Different integration techniques used in meshless methods based on a global weak form were summarized: nodal
CO

821

822 integration, stress-point integration and background integration. We have decided to use background integration since
823 background integration is most suited for the range of problems we studied here, i.e. small deformations. Background
824 integration is the most accurate way of obtaining the discrete equations.
825 We have also given an overview of how to incorporate strong discontinuities, i.e. cracks, and weak discontinuities,
826 i.e. material interfaces, into meshless methods. There are basically six ways how to handle cracks in MMs. The earliest
UN

827 methods use the visibility method that consider the crack as opaque. The diffraction and transparency is an extension
828 of the visibility method that removes certain inaccuracies, i.e. undesired discontinuities. Newer approaches are based
829 on the PU concept. In these methods, the information of the solution is incorporated in the approximation. That can
830 be done intrinsically or extrinsically. We have also proposed an extrinsic enrichment based on an MLS technique.
831 With respect to LEFM, the main advantage of an MLS extrinsic enrichment is that SIFs are obtained directly from
832 the analysis. The drawback of the method is that it uses global enrichment in contrast to local enrichment typically
833 employed in PU based methods.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
46 N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx

834 This paper is addressed to students or researcher who would like to have a quick start with meshfree meth-
835 ods, especially with respect to Solid Mechanics application and Fracture Mechanics. Therefore, we have illustrated
836 the capabilities of some methods for very simple examples. For the EFG-method, we have given and discussed
837 a Matlab code for several problems discussed in Section 3. This includes the imposition of essential boundary
838 conditions and different types of enrichment. The Matlab code can be downloaded from the following website:
839 http://www.civil.gla.ac.uk/∼bordas/codes/efgMatlab/EFGMatlabCode.rar

840 Acknowledgements

F
841 The author would like to thank the input from Dr Thomas Zimmermann during his stay at the Laboratory of Structural

OO
842 and Continuum Mechanics, Swiss Institute of Technology, Lausanne, Switzerland, from 2003 to 2006.

843 References

844 [1] S.N. Atluri, The Meshless Local Petrov–Galerkin (MLPG) Method, Tech Science Press, 2002.
[2] S.N. Atluri, S. Shen, The meshless local Petrov–Galerkin (MLPG) method: a simple and less-costly alternative to the finite element and

PR
845
846 boundary element methods, Comput. Model Eng. Sci. 3 (2002) 11–51.
847 [3] S.N. Atluri, T. Zhu, A new meshless local Petrov–Galerkin (MLPG) approach in computational mechanics, Comput. Mech. 22 (1998)
848 117–127.
849 [4] S.N. Atluri, T. Zhu, The meshless local Petrov–Galerkin (MLPG) approach for solving problems in elasto-statics, Comput. Mech. 25 (2000)
850 169–179.
851 [5] S. Beissel, T. Belytschko, Nodal integration of the element-free Galerkin method, Comput. Methods Appl. Mech. Eng. 139 (1996) 49–74.
852 [6] T. Belytschko, T. Black, Elastic crack growth in finite elements with minimal remeshing, Int. J. Numer. Methods Eng. 45 (1999) 601–
ED
853 620.
854 [7] T. Belytschko, M. Fleming, D. Organ, Y. Krongauz, W.K. Liu, Smoothing and accelerated computations in the element free Galerkin method,
855 J. Comput. Appl. Math. 74 (1996) 111–126.
856 [8] T. Belytschko, L. Gu, Y.Y. Lu, Fracture and crack growth by element-free Galerkin methods, Model. Simul. Mater. Sci. Eng. 2 (1994) 519–534.
857 [9] T. Belytschko, Y. Guo, W.K. Liu, S.P. Xiao, A unified stability analysis of meshfree particle methods, Int. J. Numer. Methods Eng. 48 (2000)
858 1359–1400.
CT

859 [10] T. Belytschko, Y. Krongauz, J. Dolbow, C. Gerlach, On the completeness of meshfree particle methods, Int. J. Numer. Methods Eng. 43
860 (1998) 785–819.
861 [11] T. Belytschko, Y. Krongauz, M. Fleming, D. Organ, W.K. Liu, Smoothing and accelerated computations in the element-free Galerkin method,
862 J. Comput. Appl. Math. 74 (1996) 111–126.
863 [12] T. Belytschko, Y. Krongauz, D. Organ, M. Fleming, P. Krysl, Meshless methods: an overview and recent developments, Comput. Methods
864 Appl. Mech. Eng. 139 (1996) 3–47.
E

865 [13] T. Belytschko, Y. Krongauz, D. Organ, M. Fleming, P. Krysl, Meshless methods: an overview and recent developments, Comput. Methods
866 Appl. Mech. Eng. 139 (1996) 3–47.
867 [14] T. Belytschko, Y.Y. Lu, L. Gu, Element-free Galerkin methods, Int. J. Numer. Methods Eng. 37 (1994) 229–256.
RR

868 [15] T. Belytschko, Y.Y. Lu, L. Gu, Crack propagation by element-free Galerkin methods, Eng. Fract. Mech. 51 (2) (1995) 295–315.
869 [16] T. Belytschko, Y.Y. Lu, Element-free Galerkin methods for static and dynamic fracture, Int. J. Solids Struct. 32 (1995) 2547–2570.
870 [17] T. Belytschko, Y.Y. Lu, L. Gu, Element-free Galerkin methods, Int. J. Numer. Methods Eng. 37 (1994) 229–256.
871 [18] T. Belytschko, D. Organ, Y. Krongauz, A coupled finite element–element-free Galerkin method, Comput. Mech. 17 (3) (1995) 186–195.
872 [19] T. Belytschko, M. Tabbara, Dynamic fracture using element-free Galerkin methods, Int. J. Numer. Methods Eng. 39 (6) (1996) 923–938.
873 [20] J. Bonet, S. Kulasegaram, Correction and stabilization of smooth particle hydrodynamics methods with application in metal forming
CO

874 simulations, Int. J. Numer. Methods Eng. 47 (6) (2000) 1189–1214.

875 [21] J. Bonet, T. Lok, Variational and momentum preservation aspects of smooth particle hydrodynamic formulations., Comput. Methods Appl.
876 Mech. Eng. 180 (1–2) (1999) 97–115.
877 [22] S. Bordas, J.G. Conley, B. Moran, J. Gray, E. Nichols, A simulation-based design paradigm for complex cast components, Eng. Comput. 23
878 (1) (2007) 25–37.
879 [23] S. Bordas, B. Moran, Enriched finite elements and level sets for damage tolerance assessment of complex structures, Eng. Fract. Mech. 73
(2006) 1176–1201.
UN

880
881 [24] S. Bordas, V.P. Nguyen, C. Dunant, H. Nguyen-Dang, A. Guidoum, An extended finite element library, Int. J. Numer. Methods Eng. (2007),
882 doi:10.1002/nme.1966.
883 [25] A. Carpinteri, Post-peak and post-bifurcation analysis of cohesive crack propagation, Eng. Fract. Mech. 32 (1989) 265–278.
884 [26] A. Carpinteri, A scale-invariant cohesive crack model for quasi-brittle materials, Eng. Fract. Mech. 69 (2002) 207–217.
885 [27] A. Carpinteri, G. Ferro, G. Ventura, The partition of unity quadrature in meshless methods, Int. J. Numer. Methods Eng. 54 (2002) 987–1006.
886 [28] A. Carpinteri, G. Ferro, G. Ventura, The partition of unity quadrature in element-free crack modelling, Comput. Struct. 81 (2003) 1783–1794.
887 [29] J.S. Chen, C. Pan, C.M.O.L. Rogue, H.P. Wang, A Lagrangian reproducing kernel particle method for metal forming analysis, Comput. Mech.
888 22 (1998) 289–307.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx 47

889 [30] J.S. Chen, C. Pan, C.T. Wu, W.K. Liu, Reproducing kernel particle methods for large deformation analysis of nonlinear structures, Comput.
890 Methods Appl. Mech. Eng. 139 (1996) 195–227.
891 [31] J.S. Chen, C.T. Wu, S. Yoon, Y. You, A stabilized conforming nodal integration for Galerkin meshfree-methods, Int. J. Numer. Methods Eng.
892 50 (2001) 435–466.
893 [32] H.J. Chung, T. Belytschko, An error estimate in the EFG method, Comput. Mech. 21 (1998) 91–100.
894 [33] H.J. Chung, G.H. Lee, C.K. Choi, Adaptive nodal generation with the element-free Galerkin method, Struct. Eng. Mech. 10 (2000) 635–650.
895 [34] L. Cueto-Felgueroso, I. Colominas, G. Mosqueira, F. Navarrina, M. Casteleiro, On the Galerkin formulation of smoothed particle hydrody-
896 namics, Int. J. Numer. Methods Eng. 60 (2004) 1475–1512.
897 [35] G.A. Dilts, Moving least square particle hydrodynamics. I. Consistency and stability, Int. J. Numer. Methods Eng. 44 (2000) 1115–1155.

F
898 [36] G.A. Dilts, Moving least square particle hydrodynamics. II. Conservation and boundaries, Int. J. Numer. Methods Eng. (2000) 1503–1524.
899 [37] J. Dolbow, T. Belytschko, An introduction to programming the meshless element-free Galerkin method, Arch. Comput. Mech. 5 (3) (1998)
900 207–241.

OO
901 [38] J. Dolbow, T. Belytschko, Numerical integration of the Galerkin weak form in meshfree methods, Comput. Mech. 23 (1999) 219–230.
902 [39] C.A. Duarte, J.T. Oden, An hp adaptive method using clouds, Comput. Methods Appl. Mech. Eng. 139 (1996) 237–262.
903 [40] C.A. Duarte, J.T. Oden, Hp clouds–an hp meshless method, Numerical Methods for Partial Differential Equations 12 (1996) 673–705.
904 [41] M. Duflot, A meshless method with enriched weight functions for three-dimensional crack propagation, Int. J. Numer. Methods Eng. 65
905 (2006) 1970–2006.
906 [42] M. Duflot, H. Nguyen-Dang, Dual analysis by a meshless method, Commun. Numer. Methods Eng. 18 (2002) 621–631.
907 [43] M. Duflot, H. Nguyen-Dang, A truly meshless method based on a moving least squares quadrature, Commun. Numer. Methods Eng. 18

PR
908 (2002) 441–449.
909 [44] M. Duflot, H. Nguyen-Dang, A meshless method with enriched weight functions for fatigue crack growth, Int. J. Numer. Methods Eng. 59
910 (2004) 1945–1961.
911 [45] C. Dunant, P. Nguyen, M. Belgasmia, S. Bordas, A. Guidoum, H. Nguyen-Dang, Architecture trade-offs of including a mesher in an object-
912 oriented extended finite element code, Eur. J. Comput. Mech. (16) (2007) 237–258, Special issue on the extended finite element method
913 (XFEM).
914 [46] C.T. Dyka, R.P. Ingel, An approach for tensile instability in smoothed particle hydrodynamics, Comput. Struct. 57 (1995) 573–580.
ED
915 [47] S. Fernández-Méndez, P. Dı́ez, A. Huerta, Convergence of finite elements enriched with meshless methods, Numer. Math. (2003),
916 doi:10.1007/s00211–003-0465-x.
917 Q6 [48] S. Fernández-Méndez, A. Huerta, Imposing essential boundary conditions in mesh-free methods, Comput. Methods Appl. Mech. Eng., in
918 press.
919 [49] S. Fernández-Méndez, A. Huerta, Coupling finite elements and particles for adaptivity: an application to consistently stabilized
920 convection–diffusion, in: M. Griebel, M.A. Schweitzer (Eds.), Meshfree Methods for Partial Differential Equations, Lecture Notes in Compu-
CT

921 Q7 tational Science and Engineering, vol. 26, Springer-Verlag, Berlin, 2002, pp. 117–129 (Papers from the International Workshop, Universität
922 Bonn, Germany, September 11–14, 2001).
923 [50] M. Fleming, Y.A. Chu, B. Moran, T. Belytschko, Enriched element-free Galerkin methods for crack tip fields, Int. J. Numer. Methods Eng.
924 40 (1997) 1483–1504.
925 [51] L. Gavete, J.L. Cuesta, A. Ruiz, A procedure for approximation of the error in the EFG method, Int. J. Numer. Methods Eng. 53 (2002)
926 677–690.
E

927 [52] L. Gavete, S. Falcën, A. Ruiz, An error indicator for the element-free Galerkin method, Eur. J. Mech. A/Solids 20 (2001) 327–341.
928 [53] L. Gavete, M.L. Gavete, B. Alonso, A.J. Martën, A posteriori error approximation in EFG method, Int. J. Numer. Methods Eng. 58 (2003)
929 2239–2263.
RR

930 [54] R.A. Gingold, J.J. Monaghan, Smoothed particle hydrodynamics: theory and application to non-spherical stars, Monthly Notices R. Astron.
931 Soc. 181 (1977) 375–389.
932 Q8 [55] S. Hao, W.K. Liu, Moving particle finite element method with superconvergence: nodal integration formulation and applications, Comput.
933 Methods Appl. Mech. Eng., in press.
934 [56] A. Huerta, S. Fernández-Méndez, Enrichment and coupling of the finite element and meshless methods, Int. J. Numer. Methods Eng. 48 (11)
935 (2000) 1615–1636.
CO

936 [57] A. Huerta, T. Belytschko, S. Fernandez-Mendez, T. Rabczuk, Encyclopedia of Comput. Mech., John Wiley and Sons, 2004 (Chapter Meshfree
937 Methods).
938 Q9 [58] A. Huerta, S. Fernández-Méndez, W.K. Liu, A comparison of two formulations to blend finite elements and mesh-free methods, Comput.
939 Methods Appl. Mech. Eng., in press.
940 [59] U. Hëussler-Combe, C. Korn, An adaptive approach with the element-free Galerkin method, Comput. Methods Appl. Mech. Eng. 162 (1998)
941 203–222.
942 [60] S.R. Idelsohn, E. Onate, F. Del Pin, The particle finite element method: a powerful tool to solve incompressible flows with free surfaces and
UN

943 breaking waves, Int. J. Numer. Methods Eng. 61 (2004) 964–989.

944 [61] G.R. Johnson, S.R. Beissel, Normalized smoothing functions for sph impact computations, Int. J. Numer. Methods Eng. 39 (1996) 2725–
945 2741.
946 [62] G.R. Johnson, S.R. Beissel, R.A. Stryk, A generalized particle algorithm for high velocity impact computations, Comput. Mech. 25 (2000)
947 245–256.
948 [63] Y. Krongauz, T. Belytschko, EFG approximation with discontinuous derivatives, Int. J. Numer. Methods Eng. 41 (1998) 1215–1233.
949 [64] Y. Krongauz, T. Belytschko, EFG approximation with discontinuous derivatives, Int. J. Numer. Methods Eng. 41 (7) (1998) 1215–1233.
950 [65] P. Lancaster, K. Salkauskas, Surfaces generated by moving least squares methods, Math. Comput. 37 (1981) 141–158.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003
 +Model
MATCOM 2939 1–48 ARTICLE IN PRESS
48 N.V. Phu et al. / Mathematics and Computers in Simulation xxx (2008) xxx–xxx

951 [66] T. Laouar, P. Villon, Adaptive analysis for the diffuse element method, in: S. Idelsohn, E. Oëate, E. Dvorkin (Eds.), Comput. Mech., CIMNE,
952 Barcelona, 1998.
953 [67] C.K. Lee, C.E. Zhou, On error estimation and adaptive refinement for element free Galerkin method. Part I. Stress recovery and a posteriori
954 error estimation, Comput. Struct. 82 (2004) 413–428.
955 [68] C.K. Lee, C.E. Zhou, On error estimation and adaptive refinement for element free Galerkin method. Part II. Adaptive refinement, Comput.
956 Struct. 82 (2004) 429–443.
957 [69] G.H. Lee, H.J. Chung, C.K. Choi, Adaptive crack propagation analysis with the element-free Galerkin method, Int. J. Numer. Methods Eng.
958 56 (2003) 331–350.
959 Q10 [70] Q. Li, J. Soric, T. Jarak, S.N. Atluri, A locking-free meshless local Petrov–Galerkin formulation for thick and thin plates, J. Comput. Phys.

F
960 (2005).
961 [71] L.D. Libersky, A.G. Petscheck, T.C. Carney, J.R. Hipp, F.A. Allahdadi, High strain Lagrangian hydrodynamics, J. Comput. Phys. 109 (1993)
962 67–75.

OO
963 [72] T.J. Liszka, C.A. Duarte, W.W. Tworzydlo, Hp-meshless cloud method, Comput. Methods Appl. Mech. Eng. 139 (1996) 263–288.
964 [73] W.K. Liu, S. Jun, Y.F. Zhang, Reproducing kernel particle methods, Int. J. Numer. Methods Eng. 20 (1995) 1081–1106.
965 [74] W.K. Liu, S. Li, W. Han, H. Lu, J. Cao, Reproducing kernel element method. Part I. Theoretical formulation, Comput. Methods Appl. Mech.
966 Eng. 193 (2004) 933–951.
967 [75] R. Loehner, C. Sacco, E. Onate, A finite point method for compressible flow, Int. J. Numer. Methods Eng. 53 (2002) 1765–1779.
968 [76] H. Lu, J.S. Chen, Adaptive Galerkin particle method, Lect. Notes Comput. Sci. Eng. 26 (2002) 251–267.
969 [77] L.B. Lucy, A numerical approach to the testing of the fission hypothesis, Astron. J. 82 (1977) 1013–1024.

PR
970 [78] J.M. Melenk, I. Babuška, The partition of unity finite element method: Basic theory and applications, Comput. Methods Appl. Mech. Eng.
971 139 (1996) 289–314.
972 [79] N. Moes, J. Dolbow, T. Belytschko, A finite element method for crack growth without remeshing, Int. J. Numer. Methods Eng. 46 (1999)
973 131–150.
974 [80] J.J. Monaghan, Why particle methods work, SIAM J. Sci. Stat. Comput. 3 (3) (1982) 422–433.
975 [81] J.J. Monaghan, An introduction to SPH, Comput. Phys. Commun. 48 (1) (1988) 89–96.
976 [82] E. Oñate, S. Idelsohn, A mesh-free finite point method for advective-diffusive transport and fluid flow problems, Comput. Mech. 21 (4/5)
ED
977 (1998) 283–292.
978 [83] E. Onate, S. Idelsohn, O.C. Zienkievicz, R.L. Taylor, A finite point method in computational mechanics: applications to convective transport
979 and fluid flow, Int. J. Numer. Methods Eng. 39 (1996) 3839–3866.
980 [84] D. Organ, M. Fleming, T. Terry, T. Belytschko, Continuous meshless approximations for nonconvex bodies by diffraction and transparency,
981 Comput. Mech. 18 (1996) 225–235.
982 Q11 [85] M.A. Puso, J.S. Chen, E. Zywicz, W. Elmer, Meshfree and finite element nodal integration methods, Int. J. Numer. Methods Eng. (early view).
CT

983 [86] T. Rabczuk, T. Belytschko, An adaptive continuum/discrete crack approach for meshfree particle methods, Latin Am. J. Solids Struct. 1
984 (2003) 141–166.
985 [87] T. Rabczuk, T. Belytschko, Adaptivity for structured meshfree particle methods in 2D and 3D, Int. J. Numer. Methods Eng. 63 (11) (2005)
986 1559–1582.
987 [88] P.W. Randles, L.D. Libersky, Recent improvements in sph modeling of hypervelocity impact, Int. J. Impact Eng. 20 (1997) 525–532.
988 [89] P.W. Randles, L.D. Libersky, Normalized sph with stress points, Int. J. Numer. Methods Eng. 48 (2000) 1445–1462.
E

989 [90] D. Shepard, A two-dimensional function for irregularly spaced points, in: Proceedings of the 23rd ACM National Conference, 1968, pp.
990 517–524.
991 [91] T. Strouboulis, K. Copps, I. Babuška, The generalized finite element method: an example of its implementation and illustration of its
RR

992 performance, Int. J. Numer. Methods Eng. 47 (8) (2000) 1401–1417.

993 [92] N. Sukumar, D.L. Chopp, N. Moës, T. Belytschko, Modelling holes and inclusions by level sets in the extended finite element method,
994 Comput. Methods Appl. Mech. Eng. 190 (2000) 6183–6200.
995 [93] N. Sukumar, D.L. Chopp, B. Moran, Extended finite element method and fast marching method for three-dimensional fatigue crack propa-
996 gation, Eng. Fract. Mech. 70 (2003) 29–48.
997 [94] N. Sukumar, N. Moës, B. Moran, T. Belytschko, Extended finite element method for three-dimensional crack modelling, Int. J. Numer.
CO

998 Methods Eng. 48 (2000) 1549–1570.

999 [95] Y. Sumi, Computational crack path prediction, Theor. Appl. Fract. Mech. 4 (1985) 149–156.
1000 [96] Y. Sumi, S. Nemat-Nasser, L.M. Keer, On crack path stability in a finite body, Eng. Fract. Mech. 22 (1985) 759–771.
1001 [97] L.W. Swegle, D.L. Hicks, S.W. Attaway, Smooth particle hydrodynamics stability analysis, J. Comput. Phys. 116 (1995) 123–134.
1002 [98] G. Ventura, An augmented Lagrangian approach to essential boundary conditions in meshless methods, Int. J. Numer. Methods Eng. 53
1003 (2002) 825–842.
1004 [99] G. Ventura, J.X. Xu, T. Belytschko, A vector level set method and new discontinuity approximations for crack growth by EFG, Int. J. Numer.
UN

1005 Methods Eng. 54 (2002) 923–944.

1006 [100] L.P. Vila, On particle weighted methods and smooth particle hydrodynamics, Math. Models Methods Appl. Sci. 9 (2) (1999) 161–209.
1007 Q12 [101] S.P. Xiao, T. Belytschko, Material stability analysis of particle methods, Adv. Comput. Math. (2005).
1008 [102] Y. You, J.S. Chend, H. Lu, Filters, reproducing kernel and adaptive meshfree method, Comput. Mech. 31 (2003) 316–326.
1009 [103] Z.Q. Zhang, J.X. Zhou, X.M. Wang, Y.F. Zhang, L. Zhang, Investigations on reproducing kernel particle method enriched by partition of
1010 unity and visibility criterion, Comput. Mech. (2004), http://www.springerlink.com/link.asp?id=pxqfcbkr3ggfaf2b.
1011 [104] O.C. Zienkiewicz, J.Z. Zhu, A simple error estimator and adaptive procedure for practical engineering analysis, Int. J. Numer. Methods Eng.
1012 24 (1987) 337–357.

Please cite this article in press as: N.V. Phu, et al., Meshless methods: A review and computer implementation aspects, Math.
Comput. Simul. (2008), doi:10.1016/j.matcom.2008.01.003