Extended Meshfree Method for Elastic and

Inelastic Media
Jiun-Shyan Chen

and Dongdong Wang

Department of Civil and Environmental Engineering, University of California, Los

Angeles, CA 90095-1593, USA.
Abstract In this paper, an extended meshfree method [9] for solving elastic bound-
ary value problems is summarized, and its extension to the elasto-plasticity problem
is presented. In extended meshfree method, the solution is decomposed into partic-
ular solution and homogeneous solution. The particular solution without satisfying
boundary conditions is obtained analytically, while a homogeneous problem with
auxiliary boundary conditions is then solved under a Galerkin framework with mov-
ing least-squares reproducing kernel approximation. The proposed method for lin-
ear dierential operator in Poisson, elasticity, and Mindlin-Reissner problems is rst
summarized. The extension to dierential equations with nonlinear self-adjoint dif-
ferential operator is then introduced, and the application to elasto-plasticity problem
is presented. Numerical results of an elasto-plasticity problem demonstrate a signif-
icant accuracy gain in the solution of extended meshfree method compared to that
of the conventional Galerkin meshfree approach.
1 Introduction
The naturally conforming properties of Galerkin meshfree methods oer
tremendous exibility for approximation of solution with arbitrary locality
and smoothness [1, 4, 6, 11, 18, 19]. For example, extended nite element
method with enriched local basis functions was used to model crack tip char-
acteristics [3, 10]. Hierarchical enrichment [16, 17, 22], coupling between FEM
and meshfree methods [13, 15, 17], and interface enrichment [14, 24] were also
developed for enhancement of local solutions. Partition of unity and general-
ized nite element method [1, 2, 11, 12, 19, 20, 21] provide a general frame-
work for enrichment of solution with special basis functions to simultaneously
achieve local and global solution accuracy.
Alternative to the aforementioned methods where the degrees of freedom
associated with the global and enriched solutions are obtained concurrently,

jschen@seas.ucla.edu

ddwang@seas.ucla.edu
18 J.-S. Chen, D. Wang
an extended meshfree method [9] has been proposed in which the solution
is solved in 2 sequential steps. In this approach, the solution is decomposed
into particular solution and homogeneous solution. The particular solution is
solved by satisfying the dierential equation containing the source term in an
innite domain without the imposition of boundary conditions. This particu-
lar solution can often be obtained analytically. When such particular solution
in an innite domain is available, the original problem can be reduced to a
homogeneous problem with auxiliary boundary conditions (corrected by the
particular solution) that is then solved numerically. It has been shown that if
a linearly complete approximation function in conjunction with a stabilized
conforming nodal integration for domain integration of the weak form are em-
ployed in the Galerkin approximation of PDEs, linear exactness and bending
exactness [7, 8, 23] can be achieved in the numerical solution of potential
and plate problems, respectively. These properties are particularly useful for
solving homogeneous solution in solid and structural problems, where substan-
tially higher solution accuracy can be obtained compared to the conventional
Galerkin meshfree formulation.
In this paper, we rst review the basic concept and solution procedures
of extended meshfree method in Section 2. In this section, a general and
straightforward approach is presented for constructing the particular solution
by employing a fundamental solution of the dierential operator without sat-
isfying the boundary conditions. Consequently, only a homogeneous governing
equation with auxiliary boundary conditions is solved by a Galerkin meshfree
method with stabilized conforming nodal integration. The specic extended
meshfree formulations for problems with linear dierential operator such as
Poisson, elasticity, and Mindlin-Reissner problems are discussed in Section 3.
In section 4, we rst discuss extended meshfree method for PDEs with nonlin-
ear self-adjoint dierential operator. The application to elasto-plasticity prob-
lem and some comments on numerical procedures are then presented. Several
numerical examples are demonstrated in Sections 3 and 4. Concluding remarks
are given in Section 5.
2 Review of Extended Meshfree Method
2.1 Moving Least-Square Reproducing Kernel (MLS/RK)
Approximation
The MLS/RK approximation [18] has been widely used in the meshfree ap-
proximation of the unknown variables in dierential equations. The problem
domain is rst discretized into a set of poins {x
1
, x
2
, , x
NP
}, where x
I
is the location of node I and NP denotes the total number of points. The
unknown variable u(x) of a dierential equation is approximated by:
u
h
(x) =
NP

I=1

I
(x)d
I
(2.1)
Extended Meshfree Method for Elastic and Inelastic Media 19
where u
h
(x) is the approximation of u(x),
I
and d
I
are the shape functions
and their associated coecients, respectively. In MLS/RK approximation [18],
a shape function
I
(x) takes the form:

I
(x) =

i+j=0
(x
1
x
1I
)
i
(x
2
x
2I
)
j

b
ij
(x)

a
(x x
I
) (2.2)
where
a
(xx
I
) is a kernel function that denes the smoothness and locality
of the approximation with a compact support
I
= supp(
a
(xx
I
)), where a
is the radius of
I
, and
NP
I=1

I


. The coecient vector

b(x) in Eq. (2.2)
is obtained by satisfying the following n-th order reproducing conditions:
NP

I=1

I
(x)x

I1
x

I2
= x

1
x

2
, + = 0, 1, 2, . . . , n (2.3)
Upon solving

b
ij
(x) from Eq. (2.3), the MLS/RK shape functions is obtained:

I
(x) = {H}
T
(0){M}
1
(x){H}(x x
I
)
a
(x x
I
) (2.4)
where
{H}
T
(x x
I
) =
_
1, x
1
x
I1
, x
2
x
I2
, (x
1
x
I1
)
2
, ... ..., (x
2
x
I2
)
n
_
(2.5)
{M}(x) =
NP

I=1
{H} (x x
I
){H}
T
(x x
I
)
a
I
(x x
I
) (2.6)
The reproducing kernel particle method (RKPM) is formulated by introduc-
ing the above approximation of unknown into the weak form of the dierential
equation L(u) + b = 0 where L is the dierential operator and b is the source
term. The order of the solution is mainly dependent upon the form of source
term and the dierential operator. The following problem introduced in [9]
is solved to illustrate the property of the RKPM using a stabilized conform-
ing nodal integration for domain integration [7], denoted as SC-RKPM, in
representing the behavior of a localized high-gradient source term.
u
,xx
+b(x) = 0 in (0, 1)
b(x) =
__
2

2
4[
(x0.5)

]
2
_
exp
_
[
(x0.5)

]
2
_
for 0.42 x 0.58
0 otherwise
u(0) = 0, u(1) = 1
(2.7)
A comparison of the SC-RKPM solution using linear basis function with a
20-node discretization and exact solutions in Fig. 1 shows large dierences in
the vicinity of the localized source term. Figures 2 (a) and (b) demonstrate
the eects of discretization with nodal distance (h= 0.0625, 0.0313, 0.0156)
and the source term localization ( = 0.5, 0.3, 0.1) on the error of solution
20 J.-S. Chen, D. Wang
0 0.2 0.4 0.6 0.8 1
0
0.5
1
1.5
x
u
Exact
SCRKPM
Figure 2.1. Solution comparison for problem (7)
accuracy using SC-RKPM. The results show that the L
2
error norm of u in-
creases as the source term becomes more localized in all three discretizations.
It is also shown that the level of localization in the source term aects the
constant of error in the numerical solution.
0.1 0.2 0.3 0.4 0.5
5
4
3
2
1

l
o
g
(
L
2
(
u
h

u
e
x
a
c
t
)
)
h=0.0625
h=0.0313
h=0.0156
1.8 1.7 1.6 1.5 1.4 1.3 1.2
5
4
3
2
log(h)
l
o
g
(
L
2
(
u
h

u
e
x
a
c
t
)
)
=0.5
=0.3
=0.1
2.09
2.07
2.04
(a) (b)
Figure 2.2. Eects of discretization and source term localization on L
2
error norm
of u in problem (2.7) using SC-RKPM
2.2 Extended Meshfree Method for Boundary Value Problems
with Linear Dierential Operator
Consider the following multi-dimensional boundary value problem (BVP):
L(u(x)) +b(x) = 0 in
B
u
(u(x)) = 0 on
u

B
h
(u(x)) = 0 on
h

(2.8)
where L is the linear dierential tensor operator, b(x) is the source term
tensor, B
u
and B
h
are the tensor operators of essential and natural boundary
Extended Meshfree Method for Elastic and Inelastic Media 21
conditions, respectively. Let g(x, x) be the fundamental solution tensor that
satises the PDE in an innite domain (i.e., without considering boundary
conditions):
L(g(x, x)) +(x x) = 0 in R
3
(2.9)
or in a matrix form
[L]

g
11
g
21
g
31

0
0

= 0, [L]

g
12
g
22
g
32

= 0, [L]

g
13
g
23
g
33

0
0

= 0
(2.10)
where = (x x), [L] is the 33 matrix corresponding to matrix expression
of L(g(x, x)), and (x x) is the delta function. Then a particular solution
is obtained by the integral:
u
p
(x) =
_

g(x, y) b(y)dy (2.11)

where is a inner product. From Eqns. (2.9) and (2.11), it can be easily
shown that the particular solution u
p
satises
L(u
p
(x)) +b(x) = 0 in (2.12)
Note that in general u
p
(x) does not satisfy boundary conditions, i.e.,
B
u
(u
p
(x)) = 0, B
h
(u
p
(x)) = 0. The expressions of fundamental solutions
and particular solutions for Poisson equation, elasticity, and Mindlin-Reissner
plate problems are listed in Table 1. Next, the solution of the given problem
is decomposed into
u(x) = u
p
(x) +u
0
(x) (2.13)
If L is the linear dierential operator, we have the following condition:
L(u(x)) +b(x) = L(u
p
(x)) +L
_
u
0
(x)
_
+b(x) = L
_
u
0
(x)
_
= 0 (2.14)
This leads to the following BVP for the homogeneous solution u
0
(x) (assuming
B
u
and B
h
are linear operators):
L
_
u
0
(x)
_
= 0 in
B
u
_
u
0
(x)
_
= B
u
(u
p
(x) ) on
u

B
h
_
u
0
(x)
_
= B
h
(u
p
(x) ) on
h

(2.15)
Here the homogeneous solution u
0
(x) is obtained from the above homoge-
neous equation with boundary conditions modied by u
p
(x). A few examples
of fundamental solutions [5, 25] for linear dierential operator are given in
Table 1.
22 J.-S. Chen, D. Wang
Table 2.1. Particular Solutions of Several Boundary Value Problems
Poisson Equation Elasticity Mindlin-Reissner Plate
Tensor
Form
L(u) +b
= 0

2
u + b = 0
(C :
s
u) +b = 0
(
s
u) =
(u +u ) /2
P
1
( ) + P
2
( )
+Q(w ) = 0
Q( w ) + q = 0
Matrix
Form
[L] {u} +
{b} = 0
[L] =
2
{u} = u
{b} = b
[L] =
h

i
T
[D]
h

i
{u} =

u
1
, u
2
, u
3

T
{b} =

b
1
, b
2
, b
3

T
[L] =
(

T
[G] + [D
s
] [N]

T
[D
s
] [N]
)
{u} =

1
,
2
, w

T
{b} =

0, 0, q

T
Fund.
Sol.
g(x, x)
g(x, x) =
1
2
ln r
r =
s
n
P
i=1
(x
i
x
i
)
2
g
ij
(x, x)
=
1
16(1)r

[(3 4)
ij
+ r
,i
r
,j
]
r
,j
=
r
x
j
g
13
(x, x)
=
1
8D
r(2 ln r + 1) cos (r, x
1
)
g
23
(x, x)
=
1
8D
r(2 ln r + 1) sin (r, x
1
)
g
33
(x, x) =
1
8D
r
2
ln r

1
2Q
ln r
g
ij
= 0 for j = 3
Part.Sol.
u
p
(x)
u
p
(x) =
R

g(x, y)b(y)dy
u
p
i
(x) =
R

g
ij
(x, y)b
j
(y)dy
u
p
i
(x) =
R

g
ij
(x, y)b
j
(y)dy
Note: P
1
, P
2
, Q, D,
h

i
,

, [D
s
], [G], [N] in Table 1 are dened in Appendix.
2.3 A Stabilized Conforming Nodal Integration for Homogeneous
Solution
The variable u
0
(x) satisfying Eq. (2.15) without a source term involves a lower
order behavior compared with the original problem and can be eectively
solved numerically. It has been studied by Chen et al. [7, 8] that the neces-
sary conditions for achieving linear exactness in the Galerkin approximation
of second order dierential equations are: (1) shape functions
I
for approxi-
mation of the unknown u
h
i
(x) =

NP
I=1

I
(x)d
iI
are linearly complete, and (2)
integration of the weak form meets the following integration constraint:
Int

(
I
) =
_

I
n d for {I : supp(
I
)
u
= } (2.16)
where Int

() denotes numerical integration over domain and n is the surface

normal on
h
. A stabilized conforming nodal integration (SCNI) has been
proposed to stabilize the nodal integration of the weak form and fulll linear
exactness in the Galerkin approximation of second order dierential equation
Extended Meshfree Method for Elastic and Inelastic Media 23
[7]. In this approach, a smoothed nodal gradient of u at point x
L
is computed
as:

u
h
(x
L
) =
1
A
L
_

L
u
h
dA =
1
A
L
_

L
nu
h
d =

I
(x
L
)d
I
(2.17)
where

I
(x
L
) =
1
A
L
_

I
(x)n(x)d (2.18)
Here
L
and
L
are the representative domain and boundary associated with
the node L as shown in Fig. 3, and A
L
is the area (or volume) of
L
. It can
be easily shown that this smoothed gradient meets the integration constraints
for nodal integration:
Int

I
) =
NP

L=1

I
(x
L
)A
L
=
_

I
n d for {I : supp(
I
)
u
= }
(2.19)
This SCNI method will be employed in the integration of weak form of the
homogeneous solution for Poisson, elasticity, and Mindlin-Reissner problems
in the next Section.

X
L
L
L
n
Figure 2.3. Nodal representative domain obtained by Voronoi diagram
3 Extended Meshfree Method for Elastic Boundary
Value Problems
In this section we discuss the extended meshfree solution procedure when the
particular solutions are obtained in Section 2. The construction of discrete
equations for the homogeneous solution of Poisson, elasticity, and Mindlin-
Reissner plate problems is also summarized. To distinguish from the tensor
notation, matrix and vector are denoted by [] and {} symbols, respec-
tively, in the following discussion.
24 J.-S. Chen, D. Wang
3.1 Poisson Problem
Consider a Poisson problem in Table 1 with boundary conditions u = g on

u
and u/n = h on
h
. The weak form of the homogeneous solution of
Poisson problem is:
_

u
0
u
0
d =
_

u
0
_
h
u
p
n
_
d
u
0
= g u
p
on
u

(3.1)
Introducing a nodal integration of weak form with assumed gradient eld
(2.17) into Eq. (3.1) yields:
NP

L=1

u
0
(x
L
)

u
0
(x
L
)A
L
=
NB
int

K=1
u
0
( x
K
)
_
h( x
K
)
u
p
n
( x
K
)
_

K
(3.2)
where NP and NB
int
are the numbers of nodal points and boundary in-
tegration points, respectively, and x
L
and
K
are the integration points
and weights for boundary integration, respectively, which are consistent
with the boundary integration for gradient smoothing term
_

I
(x)n(x)d
in Eq. (2.18). By substituting the meshfree approximation for u
0
h
(x) =

NP
I=1

I
(x)d
0
I
into Eq. (3.2), we obtain the following discrete equation for
homogeneous solution:
[K]
_
d
0
_
=
_
f
0
_
(3.3)
K
IJ
=
NP

L=1
_

b
I
(x
L
)
_
T
_

b
J
(x
L
)
_
A
L
(3.4)
_

b
I
(x
L
)
_
=
_

b
1I
(x
L
)

b
2I
(x
L
)
_
(3.5)

b
iI
(x
L
) =
1
A
L
_

I
(x)n
i
(x)d (3.6)
f
0
I
=
NBint

K=1

I
( x
K
)
_
h( x
K
)
u
p
n
( x
K
)
_

K
(3.7)
Problem (2.7) is revisited using the aforementioned Extended-RKPM. The
particular solution of this problem in Table 1 is computed using three integra-
tion zones with 6-point quadrature rule in the small domain where b(x) = 0,
and the homogeneous solution is obtained by RKPM with SCNI. The com-
parison of the results in Fig. 4 shows a signicant improvement in solution
accuracy using the Extended-RKPM compared to SC-RKPM in Fig. 3. Note
that since the particular solution is obtained analytically, the solution error is
independent to the source term localization parameter .
Extended Meshfree Method for Elastic and Inelastic Media 25
0.1 0.2 0.3 0.4 0.5
5
4
3
2
1

l
o
g
(
L
2
(
u
h

u
e
x
a
c
t
)
)
h=0.0625
h=0.0313
h=0.0156
1.8 1.7 1.6 1.5 1.4 1.3 1.2
5
4
3
2
log(h)
l
o
g
(
L
2
(
u
h

u
e
x
a
c
t
)
)
=0.5
=0.3
=0.1
2.07
(a) (b)
Figure 3.4. Eects of discretization and source term localization on L
2
error norm
of u in problem (2.7) using Extend-RKPM
3.2 Elasticity
Consider an elasticity problem in Table 1 with boundary conditions u
i
= g
i
on
u
and (C : (
s
u)) n = h on
h
. The weak form of the homogeneous
solution of elasticity problem is:
_

s
u
0
_
: C :
_

s
u
0
_
d =
_

u
0
(h (C : (
s
u
p
)) n) d
u
0
i
= g
i
u
p
i
on
u

(3.8)
Similar to the smoothing of u
0
in Poisson problem, the strain corre-
sponding to u
0
is smoothed in elasticity. Let

0
ij
=
_

s
u
0
_
ij
=
1
2
_
u
0
i,j
+u
0
j,i
_
(3.9)
The smoothed strain of u
0
at point x
L
is obtained by

0
ij
(x
L
) =
1
A
L
_

0
ij
dA =
1
2A
L
_

L
_
u
0
i,j
+u
0
j,i
_
dA
=
1
2A
L
_

L
_
u
0
i
n
j
+u
0
j
n
i
_
d
(3.10)
Dening a smoothed strain vector
_

0
_
=
_

0
11
,
0
22
, 2
0
12
_
T
and intro-
ducing MLS/RK approximation for u
0
h
i
(x) =

NP
I=1

I
(x)d
0
iI
in Eq. (3.10),
we have
_

0
h
(x
L
)
_
=
NP

I=1
_

B
I
(x
L
)
_
_
d
0
I
_
(3.11)
where
_

B
I
(x
L
)
_
=

b
I1
(x
L
) 0
0

b
I2
(x
L
)

b
I2
(x
L
)

b
I2
(x
L
)

,
_
d
0
I
_
=
_
d
0
1I
d
0
2I
_
(3.12)
26 J.-S. Chen, D. Wang
and

b
iI
(x
L
) is dened in Eq. (3.6). Introducing a nodal integration to the
weak form of elasticity in Eq. (3.8) with smoothed strain eld dened in Eq.
(3.10) yields the following stiness matrix and force vector for solving the
homogeneous solution [K]
_
d
0
_
=
_
f
0
_
:
[K
IJ
] =
NP

L=1
_

B
I
(x
L
)
_
T
[D]
_

B
J
(x
L
)
_
A
L
(3.13)
_
f
0
I
_
=
NB
int

K=1

I
( x
K
)
_
h
1
( x
K
)
p
1j
( x
K
)n
j
( x
K
)
h
2
( x
K
)
p
2j
( x
K
)n
j
( x
K
)
_

K
(3.14)
where
p
= C :
s
u
p
, and [D] is the matrix form of the elasticity tensor C.
A cantilever beam subjected to a tip shear force shown in Fig. 5(a) is ana-
lyzed using SC-RKPM (RKPM with stabilized conforming nodal integration),
G-RKPM (RKPM with Gauss integration) and Extended-RKPM. The geom-
etry and material properties are; L=10, D=2, E=21MPa, =0.3. The total
shear load is 1, which is distributed in a parabolic function over the depth
of the free end. Three discretizations as shown in Fig. 5(b) are employed to
compute error in the numerical solution. The SCNI is employed for the total
solution in SC-RKPM and the homogeneous solution of Extended-RKMP as
discussed above. In Extended-RKPM, the shear force is treated as a source
term with distributed load, and the fundamental solution in Table 1 is used to
calculate the particular solution. Figure 5(c) shows that the Extended-RKPM
greatly reduces the L
2
error norm of u compared to those of SC-RKPM and
G-RKPM.
3.3 Mindlin-Reissner Plate Problem
Consider a Mindlin-Reissner plate problem in Table 1 with boundary condi-
tions

( = 1, 2), w = w on
u
and
_
C
b
: (
s
)
_
n = m on
h
.
The weak form of the homogeneous part of Mindlin-Reissner problem given
in Table 1 can be expressed as
_

0
: C
b
:
s

0
d +
_

_
w
0

0
_
: C
s
:
_
w
0

0
_
d
+
_

_
m
_
C
b
: (
s

p
)
_
n
_
d = 0

, w = w, on
u

(3.15)
where C
b
and C
s
are the elasticity moduli for bending and shear, respectively.
To meet integration constraint and provide stability to the nodally integrated
weak form of Eq. (3.15), a curvature smoothing at the nodal point is intro-
duced [24]. Let curvature of the homogeneous solution be denoted as
k
0

=
_

0
_

=
1
2
_

0
,
+
0
,
_
(3.16)
Extended Meshfree Method for Elastic and Inelastic Media 27
(a) (b)
0.9 0.8 0.7 0.6 0.5 0.4 0.3
5
4
3
2
1
log(h)
l
o
g
(
L
2
(
u
h

u
e
x
a
c
t
)
)
GRKPM
SCRKPM
ExtendedRKPM
(c)
Figure 3.5. Beam problem solved by various methods: (a) problem description; (b)
discretizations; (c) L
2
error norm of u
The smoothed curvature at point x
L
is obtained by

0

(x
L
) =
1
A
L
_

(x)dA =
1
2A
L
_

L
(
0
,
+
0
,
)dA
=
1
2A
L
_

L
(
0

+
0

)d
(3.17)
Introducing MLS/RK to the approximation of
0
h

(x) =

NP
I=1

I
(x)
0
I
and w
0
h
(x) =

NP
I=1

I
(x)w
0
I
in smoothed curvature
_

0
_
=
_

0
11
,
0
22
, 2
0
12
_
T
and shear strain
_

0
_
=
_
w
0
,1

0
1
, w
0
,2

0
2
_
T
, we have
_

0
h
_
=
NP

I=1
_

B
b
I
_
_
d
0
I
_
,
_

0
h
_
=
NP

I=1
[B
s
I
]
_
d
0
I
_
(3.18)
_

B
b
I
_
=

0

b
I1
0
0 0

b
I2
0

b
I2

b
I1

, [B
s
I
] =
_

I,x

I
0

I,y
0
I
_
,
_
d
0
I
_
=

w
0
I

0
1I

0
2I

(3.19)
Introducing a nodal integration to weak form in Eq. (3.15) with smoothed
curvature in Eq. (3.18) yields the following stiness matrix and force vector
for solving the homogeneous solution [K]
_
d
0
_
=
_
f
0
_
:
28 J.-S. Chen, D. Wang
[K] =
_
K
b

+ [K
s
] (3.20)
_
K
b
IJ

=
NP

L=1
_

B
b
I
(x
L
)
_
T
_
D
b

B
b
J
(x
L
)
_
A
L
[K
s
IJ
] =
NP

L=1
[B
s
I
(x
L
)]
T
[D
s
] [B
s
J
(x
L
)] A
L
(3.21)
_
f
0
I
_
=
NB
int

K=1

I
( x
K
)

0
m
p
1
( x
K
)n

( x
K
) m
1
( x
K
)
m
p
2
( x
K
)n

( x
K
) m
2
( x
K
)

K
(3.22)
where m
p
= C
b
:
s

p
,
_
D
b

and [D
s
] are the matrix forms of C
b
and C
s
tensors, respectively.
A clamped circular plate subjected to a unit concentrated load P at the
plate centroid shown in Fig. 6(a) is analyzed. Since the source term is due to
a point load, the particular solution is the fundamental solution. RKPM with
SCNI (SC-RKPM), RKPM with Gauss integration (G-RKPM) and Extended-
RKPM discussed above using three discretizations shown in Fig. 6(b) are
compared in Fig. 6 (c). A substantial improvement of the solution accuracy
in Extended-RKPM over G-RKPM and SC-RKPM is achieved.
4 Extended Meshfree Method for Inelastic Boundary
Value Problems
4.1 Self-Adjoint Operator
Here we consider BVP in Eq. (2.8), with L the dierential operator for inelastic
media which, in general, is a nonlinear operator, i.e., L
_
u
0
+u
p
_
= L
_
u
0
_
+
L(u
p
). To start, express the weak form of BVP as
_

v L(u)d +
_

v b d = 0 (4.1)
where v is the weight function and v = 0 on essential boundary
u
. If L is
self-adjoint, then we have
_

v L(u)d =
_

u L(v) d +
_

i
B
i
(u, v)d (4.2)
where the 2
nd
term on the right hand side of Eq. (4.2) is a boundary term
resulting from integration by parts, and B
i
are the corresponding boundary
operators. Thus Eq. (4.1) can be expressed as
Extended Meshfree Method for Elastic and Inelastic Media 29
(a) (b)
0.3 0.4 0.5 0.6
3
2
1
0
log(h)
l
o
g
(
L
2
(
u
h

u
e
x
a
c
t
)
)
GRKPM
SCRKPM
ExtendedRKPM
(c)
Figure 3.6. Mindlin-Reissner plate solved by various methods:(a) problem descrip-
tion; (b) discretizations; (c) L
2
error norm of u
_

u L(v)d +
_

v bd +
_

i
B
i
(u, v)d = 0 (4.3)
By deposition of unknown u = u
p
+u
0
, we have
_

u
0
L(v)d+
_

u
p
L(v)d+
_

v b d+
_

i
B
i
(u, v)d = 0 (4.4)
Since L is a self-adjoint dierential operator, the rst 2 terms on the left hand
side of Eq. (4.4) can be transformed according to Eq. (4.2) to yield
_

v L
_
u
0
_
d +
_

v L(u
p
)d +
_

v bd
+
_

i
(B
i
(v, u
0
) +B
i
(v, u
p
) +B
i
(u, v))d = 0
(4.5)
If a particular solution for L(u
p
) + b = 0 exists, Eq. (4.5) is reduced to
the following problem:
30 J.-S. Chen, D. Wang
_

v L(u
0
)d +
_

i
(B
i
(u, v) +B
i
(v, u
0
) +B
i
(v, u
p
))d = 0 (4.6)
Equation (4.6) leads to the following strong form:
L(u
0
) = 0 (4.7)
with natural boundary conditions:

i
(B
i
(u
0
+u
p
, v) +B
i
(v, u
0
) +B
i
(v, u
p
)) = 0
for arbitrary v with v = 0 on
u

(4.8)
For a given u
p
, Eq. (4.8) gives the natural boundary conditions for the ho-
mogeneous solution u
0
. Further, Eq. (4.5) provides the weak form for solving
u
0
.
4.2 Elasto-plasticity Problem
Consider the following elasto-plasticity problem:
+

b = 0 in
x
= C
e
:
e
= C

:
=
s
u =
e
+

u = u on
u

x
,
n =

h on
h

x
(4.9)
where
x
is the deformed conguration, C
e
and C

are elastic and elasto-

plastic moduli, respectively,
e
and

are elastic and plastic strain rates,

respectively,
s
is the symmetric gradient operator, i.e., (
s
u)
ij
= u
(i,j)
=
( u
i,j
+ u
j,i
) /2, is the Cauchy stress rate, n is the normal unit vector of
traction surface, and

h is the surface traction rate. For elasto-plastic materials
that obey associative ow rule, we have:

=
f

(4.10)
where f is the yield function and denotes the plastic consistency parameter.
To employ the extended meshfree method discussed earlier, Eq. (4.9) is
recast into the following form:
(C
e
:
s
u)
. .
L( u)
+(

b (C
e
:

)
. .
source
= 0 (4.11)
Here the nonlinearity is included in the source term. The solution is solved as
follows:
Extended Meshfree Method for Elastic and Inelastic Media 31
1. Particular solution
u
p
i
(x) =
_

x
g
ij
(x, y)
_

b
j
(y)
(C
e
jmkl

kl
(y))
y
m
_
dy (4.12)
where g
ij
is the fundamental solution associated with the dierential op-
erator L( u) = (C
e
:
s
u), and the particular solution u
p
i
is computed
at every time step. The integration of Eq. (4.12) is performed over the
deformed conguration
x
.
2. Homogeneous problem:
(C
e
:
s
u
0
) = 0 in
u
0
= u u
p
on
u
,

0
n =

h (C

:
s
u
p
) n on
h

(4.13)
where
0
= (C

:
s
u
0
) n is the Cauchy stress rate associated with the
homogeneous solution u
0
.
Note that the homogeneous solution in Eq. (4.13) is almost identical to the
homogeneous solution of elasticity, except for the corrected natural boundary
condition involving elasto-plastic moduli. Thus the meshfree approximation
and SCNI for domain integration used in elasticity are directly applicable to
this case.
Remark 4.1. An alternative approach for solving problem (4.9) is to consider
the following strong form:
(C

:
s
u)
. .
L( u)
+

b = 0 (4.14)
Here, the elasto-plastic moduli C

reside in the dierential operator. We

consider the weak form of Eq. (4.14)
_

v ( (C

:
s
u)) d +
_

v

b d = 0 (4.15)
where the weight function v = 0 on
u
. Via integration by parts and diver-
gence theorem, we have
_

v

hd
_

u (
s
v : C

n)d +
_

u ( (C

:
s
v))d
+
_

v

bd = 0
(4.16)
Note that we have used the properties C

ijkl
= C

klij
= C

jikl
= C

ijlk
. Next, by
the decomposition of u = u
p
+ u
0
, we have
32 J.-S. Chen, D. Wang
_

u ( (C

:
s
v))d
=
_

u
p
( (C

:
s
v))d +
_

u
0
( (C

:
s
v))d
(4.17)
Further applying integration by parts, the following equation can be obtained
_

v ( (C

:
s
u
p
))d +
_

v ( (C

:
s
u
0
))d +
_

v

hd

v ((C

:
s
u
p
) n)d +
_

v

bd = 0
(4.18)
In the typical incremental solution procedure, C

is formed with solution of the

previous time step. Thus if the fundamental solution of dierential operator
in Eq. (4.14) with an explicit C

exists, Eq. (4.18) can be reduced to

_

v (

h (C

:
s
u
p
) n)d +
_

v ( (C

:
s
u
0
))d = 0 (4.19)
This leads to the following strong form and natural boundary conditions for
the homogeneous solution:
(C

:
s
u
0
) = 0 in
(C

:
s
u
0
) n =

h (C

:
s
u
p
) n on
h

(4.20)
The corresponding weak form of Eq. (4.20) is
_

s
v : C

:
s
u
0
d =
_

v (

h (C

:
s
u
p
) n)d (4.21)
4.3 Some Remarks on Consideration of Large Deformation
In the case where the materials undergo large deformation, the undeformed
conguration
X
and deformed conguration
x
of the material domain need
to be distinguished. Considering a material point X
X
under deformation
is moved to x
x
, where
X
is the undeformed conguration and
x
is the
corresponding deformed conguration, and x = X + u, where u is the dis-
placement. For a general nonlinear problem, incremental solution procedures
are required, and the incremental equation of Eq. (4.11) is solved:
(C
e
:
s
u)
. .
L(u)
+(b (C
e
:

)
. .
source
= r (4.22)
where denote a nite increment quantity, and r is the residual. The
increment of particular solution u
p
i
is obtained by
Extended Meshfree Method for Elastic and Inelastic Media 33
u
p
i
(x) =
_

x
g
ij
(x, y)
_
b
j

_
C
e
jmkl

kl
_
,m
r
j
_
(y)dy (4.23)
where g
ij
(x, y) denotes the fundamental solution of the dierential operator
L(u) = (C
e
:
s
u). With the consideration of large deformation, the
Galerkin weak form of the homogeneous problem of Eq. (4.22) is transformed
to the undeformed conguration
X
, and a Lagrangian kernel [6] is used to
approximate the homogeneous solution following 6:
u
0
i
(X) =
NP

I=1

I
(X)d
0
iI
(4.24)
where
I
(X) is the MLS/RK shape function using Lagrangian kernel. Note
that the solution of homogeneous solution in Eq. (4.22) requires the particular
solution in Eq. (4.23). Since the transformation of weak form of the homo-
geneous solution in Eq. (4.22) to the undeformed conguration requires the
particular solution to be evaluated in the unformed conguration, a mapping
from x to X is needed. A stabilized conforming nodal integration (SCNI)
that satises integration constraints in the Lagrangian Galerkin approxima-
tion of the homogeneous solution is employed in the domain integration of
weak form in the undeformed conguration [8]. In SCNI, to meet integration
constraints using nodal integration in Lagrangian Gakerkin formulation, the
displacement spatial gradient u
i,j
= u
i
/x
j
at a nodal point X
L
in the
undeformed conguration is smoothed as:
u
0
i,j
(X
L
) =

F
0
ik
(X
L
)

F
1
kj
(X
L
) (4.25)
where

F
0
ij
(X
L
) =
1
A
X
L
_

X
L
F
0
ij
d =
1
A
X
L
_

X
L
u
0
i
X
j
d =
1
A
X
L
_

X
L
(u
0
i
N
j
)d
(4.26)

F
ij
(X
L
) =
1
A
X
L
_

X
L
F
ij
d =
1
A
X
L
_

X
L
u
i
X
j
d +
ij
=
1
A
X
L
_

X
L
(u
i
N
j
)d +
ij
(4.27)
Here N
i
is the surface normal in the undeformed conguration,
X
L
and

X
L
are the nodal representative domain and boundary of the Voronoi cell
associatd with node L in the undeformed conguration, respectively, and A
X
L
is the area (volume) of
X
L
. The other details of Lagrangian meshfree method
using SCNI for nonlinear problems can be found in [8]. Due to the absence of
source term in the homogeneous problem, the Lagrangian meshfree formula-
tion for solving the homogeneous solution can achieve much greater solution
accuracy.
34 J.-S. Chen, D. Wang
Figure 4.7. One dimensional rod subjected body force
4.4 Numerical Example
A one dimensional elasto-plastic bar subjected to a body force as shown in
Fig. 7 is analyzed. In this example, the following isotropic hardening elasto-
plasticity model is employed, where the yield function is expressed as:
f = || (
y
+K
p
) (4.28)
Here,
y
is the initial yield stress and K is hardening parameter. The material
0 0.2 0.4 0.6 0.8 1
0.015
0.01
0.005
0
0.005
0.01
0.015
x
T
o
t
a
l

S
t
r
a
i
n
Reference
SCRKPM
ExtendedRKPM
Figure 4.8. Total strain comparison of
elasto-plasticity problem
0 0.2 0.4 0.6 0.8 1
0
2
4
6
8
x 10
3
x
P
l
a
s
t
i
c

s
t
r
a
i
n
Reference
SCRKPM
ExtendedRKPM
Figure 4.9. Plastic strain comparison
of elasto-plasticity problem
0 0.2 0.4 0.6 0.8 1
3
2
1
0
1
2
x 10
5
x
S
t
r
e
s
s
Reference
SCRKPM
ExtendedRKPM
Figure 4.10. Stress comparison of elasto-plasticity problem
Extended Meshfree Method for Elastic and Inelastic Media 35
properties are: Youngs modulus E = 2.010
7
, K = 0.2E and
y
= 8.010
4
.
The body force function given in Eq. (2.7) is used, and the rod is xed at the
left end. Both SC-RKPM and the proposed Extended-RKPM are employed.
The rod is discretized by only 8 nodes using both methods. For comparison
purpose, a rened model with 51-node discretization is used to compute a
reference solution. The numerical results for total strain, plastic strain and
stress distributions are shown in Figs. 8, 9, and 10, respectively. A much
enhanced solution is observed in the proposed new approach.
5 Conclusions
An extended meshfree method for elastic and inelastic media has been pre-
sented. In this approach, the total solution has been expressed by a combina-
tion of particular and homogeneous solutions. The particular solution is any
analytical (or numerical) expression satisfying the governing dierential equa-
tion containing the source term but not necessarily the boundary conditions.
Thus, the problem is reduced to a homogeneous equation where the original
boundary conditions are modied by the particular solution. In this paper, a
general method has been presented for constructing the particular solution.
The homogeneous solution has been solved numerically using Galerkin approx-
imation. By employing moving least-square reproducing kernel (MLS/RK)
approximation with linear completeness, as well as a stabilized conforming
nodal integration (SCNI) for domain integration of the weak form, a linear
exactness in solid continuum and a bending exactness in plate bending can be
achieved in the homogeneous counterpart of these problems. Compared to the
conventional Galerkin meshfree method, this new approach signicantly re-
duces the constant in the error norms, and examples have been demonstrated
in Poisson, elasticity, and Mindlin-Reissner plate problems.
The proposed method for dierential equations with nonlinear dierential
operators has also been presented. It has been shown that if the dierential
operator is self-adjoint, a homogeneous problem can be obtained with stan-
dard integration by parts procedures. A Lagrangian MLS/RK approximation
and the corresponding SCNI for Lagrangian weak form of large deformation
problems have also been discussed. Using elasto-plasticity as a model problem,
a simplied approach to obtain particular solution and to construct homoge-
neous problem has been presented. In this approach, the nonlinearity resides
in the source term, and the dierential operator takes the form of linear elas-
ticity. Thus the fundamental solution is identical to that of linear elasticity,
and the particular solution is obtained by integration of fundamental solution
and the nonlinear source term at every incremental step in nonlinear calcu-
lation. The corresponding homogeneous problem has been constructed with
boundary conditions corrected by the particular solution in every incremental
step, and it has been solved using MLS/RK approximation with SCNI domain
integration of the incremental weak form. The numerical results demonstrate a
36 J.-S. Chen, D. Wang
substantially higher solution accuracy in extended meshfree method compared
to that of the conventional Galerkin meshfree method.
6 Appendix
The parameters and matrices used Table 1 are dened as follows:
P
1
=
Et
3
24(1 +)
, P
1
=
Et
3
24(1 )
, Q =
5Et
12 (1 +)
, D =
Et
3
12(1
2
)
(6.1)
_

_
=

x
1
0 0
0

x
2
0
0 0

x
3

x
2

x
1
0

x
3
0

x
1
0

x
3

x
2

,
_

x
1
0
0

x
2

x
2

x
1

, =
_

x
1

x
2
_
(6.2)
[N] =

1 0

x
1
0 1

x
2

, [G] =
_
D
b
_

{0}

(6.3)
_
D
b

=
Et
3
12(1
2
)

1 0
1 0
0 0
(1)
2

, [D
s
] =
5
6
t
_
1 0
0 1
_
(6.4)
Acknowledgement. The support of this work by NSF grant CMS-0296112 to UCLA
is greatly acknowledged.
References
[1] I. Babuska, U. Banerjee and J.E. Osborn: Survey of meshless and generalized
nite element methods: a unied approach. Acta Numerica, 1125(2003).
[2] I. Babuska and J. M. Melenk: The partition of unity nite element method. Int.
J. Numer. Meth. Eng., 40,727758(1997).
[3] T. Belytschko, N. Moes, S. Usui and C. Parimi: Arbitrary discontinuities in
nite elements.Int. J. Numer. Meth. Eng., 50, 9931013(2001).
[4] T. Belytschko, Y.Y. Lu and L. Gu:Element-free Galerkin methods.Int. J. Numer.
Meth. Eng., 37,229256(1994).
[5] C.A. Brebbia, J. Dominguez:Boundary elements: an introductory course. Com-
putational mechanics publications, McGraw-Hill book company (1989).
Extended Meshfree Method for Elastic and Inelastic Media 37
[6] J.S. Chen, C. Pan, C.T. Wu and W.K. Liu: Reproducing kernel particle meth-
ods for large deformation analysis of nonlinear structures. Comput Meth. Appl.
Mech. Engng., 139, 195227(1996).
[7] J.S. Chen, C.T. Wu, S. Yoon, and Y. You: A stabilized conforming nodal in-
tegration for Galerkin meshfree methods. Int. J. Numer. Meth. Eng., 50, 435
466(2001).
[8] J.S. Chen, C.T. Wu and S. Yoon: Nonlinear version of stabilized conforming
nodal integration for Galerkin meshfree methods. Int. J. Numer. Meth. Eng.,
53,25872615(2002).
[9] J.S. Chen, D. Wang and S.B. Dong: An extended meshfree method for boundary
valu problems.Comput. Meth. Appl. Mech. Engng., 193, 10851103(2004).
[10] J. Dolbow, N. Moes and T. Belytschko: Discontinuous enrichment in nite
elements with a partition of unity method. Finite Elements Anal. Design,36,235
260(2000).
[11] C.A.M. Duarte and J.T. Oden: A h-p adaptive method using clouds. Com-
put.Meth. Appl. Mech. Engng., 139,237262(1996).
[12] C. A. M. Duarte, I. Babuska, and J. T. Oden: Generalized nite element method
for three dimensional structural problems. Computer and Structures, 77, 219
232(2000).
[13] A. Huerta and S. Fernandez-Mendez: Enrichment and coupling of the nite
element and meshless methods. Int. J. Numer. Meth. Eng., 48, 16151636(2000).
[14] Y. Krongauz and T. Belytschko: EFG Approximation with discontinuous
derivatives, Int. J. Numer. Meth. Eng., 41, 12151233(1998).
[15] Y. Krongauz and T. Belytschko: Enforcement of essential boundary conditions
in meshless approximations using nite elements. Comput. Meth. Appl. Mech.
Engng., 131, 133145(1996).
[16] S. Li and W. K. Liu: Reproducing kernel hierarchical partition of unity part I:
formulation and theory. Int. J. Numer. Meth., Eng., 45,251288 (1999).
[17] W.K. Liu, R.A. Uras and Y. Chen: Enrichment of the nite element method
with the reproducing kernel particle method. Journal of Applied Mechanics, 64,
861870(1997).
[18] W.K. Liu, S. Jun, Y.F. Zhang: Reproducing kernel particle methods. Int. J.
Numer. Meth. Fluids, 20 10811106(1995).
[19] J.M. Melenk and I. Babuska: The partition of unity nite element method:
basic theory and applications. Comput. Meth. Appl. Mech. Engng.,139, 289
314(1996).
[20] T. Strouboulis, I. Babuska, and K. Copps: The design and analysis of the gen-
eralized nite element method, Comput. Meth. Appl. Mech. Engng.,181, 43
69(2001).
[21] T. Strouboulis, K. Copps, and I. Babuska: The generalized nite element
method. Comput. Meth. Appl. Mech. Engng., 190, 40814193(2001).
[22] G.J. Wagner and W.K. Liu: Hierarchical enrichment for bridging scales and
meshfree boundary conditions. Int. J. Numer. Meth. Eng., 50, 507524(2001).
[23] D. Wang and J.S. Chen: Locking-free stabilized conforming nodal integration
for meshfree Mindlin-Reissner plate formulation. Comput. Meth. Appl. Mech.
Engng., 193, 10651083(2004).
[24] D. Wang, J.S. Chen and L. Sun: Homogenization of magnetostrictive particle-
lled elastomers using an interface-enriched reproducing kernel particle method.
Finite Element Anal. Design, 39, 765782(2003).
38 J.-S. Chen, D. Wang
[25] Y. Wang: Boundary elements in engineering. China Hydraulic and Hydroelec-
tric Press, Beijing (1995).