Moving Least Square:

In this Appendix Moving Least Square (MLS) approximation is presented. MLS

approximation is widely used for generation of shape function in meshfree methods. In this
method function u(x) is approximated by u
h
(x), as follows:
1
( ) ( ) ( ) ( ) ( )
m
h
j j
j
u p a
=
=

T
x = x x p x a x (A.1)
where m is number of terms in the basis, p
i
(x) are the monomial basis functions, and a
i
(x)
are unknown coefficients which depend on location x. Commonly used bases are the linear basis
and quadratic basis functions. Quadratic basis in two-dimensional domain has the following
form:
T 2 2
[1, , , , , ] x y x xy y = P (A.2)
The unknown coefficients a
j
(x) in Eq. (A.1) can be determined by minimizing weighted
square of residual function as follows:
T 2
1
( )[ ( ) ( ) ]
n
I I I
I
J w u
=
=

x x p x a x (A.3)
where u
I
are nodal parameters of field variable at node I and n is the number of nodes in
the neighborhood of x which is called domain of influence. The unknown coefficient a(x) can be
obtained by minimizing the functional of the weighted residual as follows:
0
J c
=
ca
(A.4)
which yields following system of linear equations:
s
A(x)a(x)=B(x)U (A.5)
where,
T
1
( ) ( ) ( ) ( )
n
I I I
I
w
=
=

A x x x p x p x (A.6)
( ) [ ( ) ( ),. . . , ( ) ( )]
I I n n
w w = B x x x p x x x p x (A.7)
T
1 2
{ . . . }
s n
u u u = U (A.8)
By substituting Eq. (A.5) into Eq. (A.1), MLS approximant can be written as:
1
( ) ( ) ( )
n
h
I I
I
u u |
=
=
s
x x = x U (A.9)
where (x) is matrix of shape functions and defined as follows:
T 1
1 2
( ) [ ( ) ( ) . . . ( )] ( ) ( ) ( )
n
| | |

= = x x x x p x A x B x (A.10)
In order to calculate partial derivatives of (x), Eq. (A.10) is rewritten as follows [43]:
( ) ( ) ( ) =
T
x x B x (A.11)
where
1
( ) ( ) (

= x A x p x) (A.12)
or,
( ) ( ) ( = A x x p x) (A.13)
The partial derivatives of (x) can be expressed as follows:
, , , x x x
= A p A (A.14)
, , , y y y
= A p A (A.15)
, , , ,
( 2 )
xx xx xx x x
= + A p A A (A.16)
, , , , , , ,
( )
xy xy xy x y y x
= + + A p A A A (A.17)
, , , , ,
( 2 )
yy yy yy y y
= + A p A A (A.18)
The partial derivatives of (x) would be as follows:
T T
, , , x x x
= + B B (A.19)
T T
, , , y y y
= + B B (A.20)
T T T
, , , , ,
2
xx xx x x xx
= + + B B B (A.21)
T T T T
, , , , , , , xy xy x y y x xy
= + + + B B B B (A.22)
T T T
, , , , ,
2
yy yy y y yy
= + + B B B (A.23)
Weight function plays an important role in the formulation of MLS method. This function
should be non-zero in the domain of influence and zero outside of the domain. The precise
character of this function seems to be unimportant although it is almost mandatory that it be
positive and increase monotonically as
I
x x decreases, furthermore it is desirable that weight
function be smooth [44]. Several weight functions have been proposed by researchers. In this
work quartic spline weight function is chosen,
2 3 4
1 6 8 3 , 1
( ) ( )
0, 1
I
r r r r
w w r
r

+ s
=

>

x x (A.24)
where,
I
I
r
d

=
x x
(A.25)
where d
I
determines the size of the influence domain at node I. The most commonly used
domains are circles and rectangles (Fig.A.1). For circular domain, d
I
is the radius of circle and
for rectangular domain, d
I
is equal to length of rectangle in x, and y direction which are denoted
by d
Ix
and d
Iy
, respectively. For the latter case, weight function can be written as follows:
( ) ( ). ( )
I x y
w w r w r = x x (A.26)
where
x
r
and
y
r
are given by
I
x
Ix
x x
r
d

= (A.27)
I
y
Iy
y y
r
d

=
(A.28)
where
max
.
Ix Ix
d d c = (A.29)
max
.
Iy Iy
d d c = (A.30)
where d
max
is scaling parameter, c
Ix
and c
Iy
are distance around node I which are
determined in such a way that enough nodes be in the domain in order that matrix A in Eq.(A.13)
to be invertible at every point in the domain [44].