Professional Documents
Culture Documents
A Galerkin/least-Square Finite Element Formulation For Nearly Incompressible Elasticity/stokes Ow
A Galerkin/least-Square Finite Element Formulation For Nearly Incompressible Elasticity/stokes Ow
_
_
_; (25)
where the rst term represents the general Galerkin form and the appended term represents the residual-based
term including an element-dependent stabilized matrix s [4], which is an element-dependent parameter and has
to be selected for good performance. dA(u)
T
is the corresponding weighting part regarded in GLS method.
A(u) represents the residual part of the strong form corresponding to the high-order equation.
3.2. Decomposition of least-square part
As required for our current mixed displacement-pressure formulation, the corresponding least-square part
can be modied as follows:
_
X
dA(u)
T
sA(u) dX =
_
X
(\ r(w
u
; w
p
))
T
s(\ r(u; p) b) dX. (26)
From Eqs. (5), (12) and (13), we can obtain that weighting term
(\ r(w
u
; w
p
))
T
= (\ (D
uu
~e 1~p))
T
; (27)
where
~e = (dd
u
)
T
(\
s
N
u
)
T
; (28)
~p = (dP
p
)
T
(N
p
)
T
. (29)
Thus, Eq. (27) can be rewritten as follows:
(\ r(w
u
; w
p
))
T
= (dd
u
)
T
(\ (D
uu
\
s
N
u
))
T
(dP
p
)
T
(\ (1N
p
))
T
= (dd
u
)
T
L
T
uu
(dP
p
)
T
L
T
up
. (30)
Also, the residual of the strong form can be rewritten as
\ r(u; p) b = \ (D
uu
\
s
N
u
u) \ (1N
p
P) b = L
uu
u L
up
P b; (31)
where
L
uu
= L(D
uu
\
s
N
u
); (32)
L
up
= L(D
up
N
p
). (33)
When using the GLS scheme, some dicult points should be mentioned. The derivation of matrix L
uu
and L
up
seems very complicated. Detailed derivations are provided below.
For isotropic elasticity, deviatoric modulus is a symmetric matrix and can be given as follows:
516 K. Xia, H. Yao / Applied Mathematical Modelling 31 (2007) 513529
D
uu
= 2lI
dev
=
2l
3
2 1 1 0 0 0
1 2 1 0 0 0
1 1 2 0 0 0
0 0 0 3 0 0
0 0 0 0 3 0
0 0 0 0 0 3
_
_
_
_
=
c
11
c
12
c
13
0 0 0
c
12
c
22
c
23
0 0 0
c
13
c
23
c
33
0 0 0
0 0 0 c
44
0 0
0 0 0 0 c
55
0
0 0 0 0 0 c
66
_
_
_
_
. (34)
Then we can obtain matrix L
uu
as follows:
L
uu
= L(D
uu
\
s
N
u
) =
l
1;3i1
l
1;3i2
l
1;3i3
l
2;3i1
l
2;3i2
l
2;3i3
l
3;3i1
l
3;3i2
l
3;3i3
_
_
_
_
33nel
; (35)
where
l
1;3i1
= c
11
o
2
N
i
ox
2
c
44
o
2
N
i
oy
2
c
66
o
2
N
i
oz
2
; (36)
l
1;3i2
= l
23i1
= (c
12
c
44
)
o
2
N
i
oxoy
; (37)
l
1;3i3
= l
3;3i1
= (c
13
c
66
)
o
2
N
i
oxoz
; (38)
l
2;3i2
= c
44
o
2
N
i
ox
2
c
22
o
2
N
i
oy
2
c
55
o
2
N
i
oz
2
; (39)
l
2;3i3
= l
3;3i2
= (c
23
c
55
)
o
2
N
i
oyoz
; (40)
l
3;3i3
= c
66
o
2
N
i
ox
2
c
55
o
2
N
i
oy
2
c
33
o
2
N
i
oz
2
. (41)
For elastoplasticity, consistent tangent modulus will vary from point to point and be updated for each incre-
mental step. The gradients of components of the consistent tangent modulus will be required to accurately
compute L
up
. This will bring additional diculty to the nonlinear computational solid mechanics. This might
be the intrinsic diculty of GLS method.
L
up
= L D
up
N
p
( ) =
oN
1
ox
oN
nel
ox
;
oN
1
oy
oN
nel
oy
;
oN
1
oz
oN
nel
oz
_
_
_
_
3nel
. (42)
3.3. Mixed displacement-pressure formulations for GLS
Then Eq. (26) can be further decomposed into two parts as follows:
_
X
dA(u)
T
sA(u) dX =
_
X
(dd
u
)
T
L
T
uu
(dP
p
)
T
L
T
up
_ _
s(L
uu
u L
up
P b) dX
=
_
X
(dd
u
)
T
L
T
uu
s(L
uu
u L
up
P b) dX
_
X
(dP
p
)
T
L
T
up
s(L
uu
u L
up
P b) dX. (43)
By substituting Eq. (12) into Eq. (10), we can obtain the general weak form (Galerkin form) for the equilib-
rium equation as follows:
K. Xia, H. Yao / Applied Mathematical Modelling 31 (2007) 513529 517
_
X
(dd
u
)
T
\
T
s
N
u
: (D
uu
\
s
Nu D
up
N
p
P) dX =
_
X
(dd
u
)
T
(N
u
)
T
bdX
_
C
(dd
u
)
T
(N
u
)
T
t dC. (44)
Then, by appending the rst part of Eq. (43) to Eq. (44), we can obtain the Galerkin/least-square form for our
mixed displacement-pressure formulation
_
X
B
T
D
uu
BdXu
_
X
B
T
D
up
N
p
dXP
_
X
L
T
uu
sL
uu
dXu
_
X
L
T
uu
sL
up
dXP
=
_
X
(N
u
)
T
bdX
_
C
(N
u
)
T
t dC
_
X
L
T
uu
sbdX. (45)
Eq. (45) can be rewritten as follows:
K
uu
K
G
uu
_ _
u K
up
K
G
up
_ _
P = R
u
R
G
; (46)
where
K
G
uu
=
_
X
L
T
uu
sL
uu
dX; (47)
K
G
up
=
_
X
L
T
uu
sL
up
dX; (48)
R
G
u
=
_
X
L
T
uu
sbdX. (49)
By taking the same procedure and substitute Eq. (13) into Eq. (11), one will get
_
X
(dP
p
)
T
(N
p
)
T
\ u
p
K
_ _
dX = 0. (50)
Next, append the second part of Eq. (43) to Eq. (50), and we can obtain the weak form for the volumetric
governing equation corresponding to Galerkin/least-square approach
_
X
(dP
p
)
T
(N
p
)
T
\ u
p
K
_ _
dX
_
X
(dP
p
)
T
L
T
up
s(L
uu
u L
up
P b) dX = 0. (51)
Removing dP
p
on both sides of Eq. (51), and insert Eq. (15), thus we obtain
K
up
K
G
up
_ _
u K
pp
K
G
pp
_ _
P = R
G
p
; (52)
K
G
pu
=
_
X
L
T
up
sL
uu
dX; (53)
K
G
up
=
_
X
L
T
up
sL
up
dX; (54)
R
G
p
=
_
X
L
T
up
sbdX. (55)
Combining Eq. (46) and Eq. (52) together, we rewrite the mixed displacement-pressure formulation in matrix
form as follows:
K
uu
K
G
uu
K
up
K
G
up
K
up
K
G
up
_ _
T
K
pp
K
G
pp
_
_
_
_
u
P
_ _
=
R
u
R
G
u
R
G
p
_ _
. (56)
The above Eq. (56) is the Galerkin/least-square formulation for the mixed form of nearly incompressible elas-
ticity/stokes ow.
518 K. Xia, H. Yao / Applied Mathematical Modelling 31 (2007) 513529
3.4. Second derivatives of shape functions
As we described earlier, the important part to completely implementing the Galerkin/least-square approach
is to use the second derivatives of shape functions. The key formulations for calculation of second derivatives
of shape functions are given as follows:
o
2
N
I
ox
2
o
2
N
I
oy
2
o
2
N
I
oz
2
o
2
N
I
oxoy
o
2
N
I
oyoz
o
2
N
I
ozox
_
_
_
_
= [M
2
[[M
1
[J
1
oN
I
on
oN
I
og
oN
I
of
_
_
_
_
[M
2
[
o
2
N
I
on
2
o
2
N
I
og
2
o
2
N
I
of
2
o
2
N
I
onog
o
2
N
I
ogof
o
2
N
I
ofon
_
_
_
_
; (57)
where matrix [M
2
] and [M
1
] are given by
[M
2
[ =
j
2
11
j
2
12
j
2
13
2j
11
j
12
2j
12
j
13
2j
13
j
11
;
j
2
21
j
2
22
j
2
23
2j
21
j
22
2j
22
j
23
2j
23
j
21
;
j
2
31
j
2
32
j
2
33
2j
31
j
32
2j
32
j
33
2j
33
j
31
;
j
11
j
21
j
12
j
22
j
13
j
23
j
11
j
22
j
13
j
22
j
12
j
23
j
13
j
22
j
11
j
23
j
13
j
21
;
j
21
j
31
j
22
j
32
j
23
j
33
j
21
j
32
j
22
j
31
j
22
j
33
j
23
j
32
j
21
j
33
j
23
j
31
;
j
31
j
31
j
32
j
32
j
33
j
33
j
11
j
32
j
12
j
31
j
12
j
33
j
13
j
32
j
11
j
33
j
13
j
31
_
_
_
_
(58)
and
[M
1
[ =
o
2
N
I
on
2
x
I
o
2
N
I
on
2
y
I
o
2
N
I
on
2
z
I
o
2
N
I
og
2
x
I
o
2
N
I
og
2
y
I
o
2
N
I
og
2
z
I
o
2
N
I
of
2
x
I
o
2
N
I
of
2
y
I
o
2
N
I
of
2
z
I
o
2
N
I
onog
x
I
o
2
N
I
onog
y
I
o
2
N
I
onog
z
I
o
2
N
I
ogof
x
I
o
2
N
I
ogof
y
I
o
2
N
I
ogof
z
I
o
2
N
I
ofon
x
I
o
2
N
I
ofon
y
I
o
2
N
I
ofon
z
I
_
_
_
_
; (59)
K. Xia, H. Yao / Applied Mathematical Modelling 31 (2007) 513529 519
J =
ox
on
oy
on
oz
on
ox
og
oy
og
oz
og
ox
of
oy
of
oz
of
_
_
_
_
=
oN
I
on
x
I
oN
i
on
y
I
oN
i
on
z
I
oN
i
og
x
I
oN
i
og
y
I
oN
i
og
z
I
oN
i
of
x
I
oN
i
of
y
I
oN
i
of
z
I
_
_
_
_
=
J
11
J
12
J
13
J
21
J
22
J
23
J
31
J
32
J
33
_
_
_
_; (60)
J
1
=
1
det[J[
J
22
J
33
J
32
J
23
J
13
J
32
J
12
J
33
J
12
J
23
J
13
J
22
J
31
J
23
J
21
J
33
J
11
J
33
J
13
J
31
J
21
J
13
J
23
J
11
J
21
J
32
J
31
J
22
J
12
J
31
J
32
J
11
J
11
J
22
J
12
J
21
_
_
_
_
=
j
11
j
12
j
13
j
21
j
22
j
23
j
31
j
32
j
33
_
_
_
_. (61)
4. Numerical examples
4.1. Convergence rate study
For convergence study, the developed method is applied to study a widely used cantilever beam with ana-
lytical solutions [11]. The beam is of length 10 m, height 2m, and thickness 1 m and subjected to a parabolic
shear traction at the free end as shown in Fig. 1. The material properties for Fig. 1 are Youngs modulus
E = 7.5E+07 N/m
2
and Poissons ratio m = 0.4999. The load is P = 2560 N.
Figs. 24 shows the results of the convergence rate study for the 3 node triangle and the 4 node quadrilat-
eral, which was performed using uniform meshes of 10 2, 20 4, 40 8, and 80 8. Totally four dierent
Fig. 1. Diagram of parabolic shear-loaded beam.
-6
-5
-4
-3
-2
-1
-0.2 0.0 0.2 0.4 0.6 0.8 1.0
Log (1/h)
L
o
g
(
e
r
r
o
r
n
o
r
m
o
f
d
i
s
p
l
a
c
e
m
e
n
t
)
GLS, k=1.78 for T3
GLS, k=1.80 for Q4
Fig. 2. Convergence rate for the L
2
norm of the displacement.
520 K. Xia, H. Yao / Applied Mathematical Modelling 31 (2007) 513529
0.5
0.6
0.7
0.8
0.9
1.0
1.1
10 100 1000 10000
Number of nodes
N
o
r
m
a
l
i
z
e
d
d
i
s
p
l
a
c
e
m
e
n
t
GLS for T3
GLS for Q4
Fig. 5. Tip deection convergence for plane stress.
-0.2 0.0 0.2 0.4 0.6 0.8 1.0
Log (1/h)
0
1
2
3
4
5
L
o
g
(
e
r
r
o
r
n
o
r
m
o
f
p
r
e
s
s
u
r
e
)
GLS, k=1.81 for T3
GLS, k=1.44 for Q4
Fig. 4. Convergence rate for the L
2
norm of the pressure eld.
-1.0
-0.5
0.0
0.5
1.0
-0.2 0.0 0.2 0.4 0.6 0.8 1.0
Log (1/h)
L
o
g
(
e
r
r
o
r
n
o
r
m
o
f
e
n
e
r
g
y
)
GLS, k=0.98 for T3
GLS, k=1.0 for Q4
Fig. 3. Convergence rate for the energy norm.
K. Xia, H. Yao / Applied Mathematical Modelling 31 (2007) 513529 521
10 100 1000 10000
Number of nodes
0.5
0.6
0.7
0.8
0.9
1.0
1.1
N
o
r
m
a
l
i
z
e
d
s
t
r
e
s
s
GLS for T3
GLS for Q4
Fig. 6. Stress convergence for plane stress.
0.6
0.7
0.8
0.9
1.0
1.1
10 100 1000 10000
Number of nodes
N
o
r
m
a
l
i
z
e
d
d
i
s
p
l
a
c
e
m
e
n
t
GLS for T3
GLS for Q4
Fig. 7. Tip deection convergence for plane strain.
10 100 1000 10000
Number of nodes
0.5
0.6
0.7
0.8
0.9
1.0
1.1
N
o
r
m
a
l
i
z
e
d
s
t
r
e
s
s
GLS for T3
GLS for Q4
Fig. 8. Stress convergence for plane strain.
522 K. Xia, H. Yao / Applied Mathematical Modelling 31 (2007) 513529
mesh sizes h were used here. This paper presents the mathematical rate of convergence study for displacements
and pressure in the L
2
(X) norm and the energy norm. As described above, GLS method requires numeric tests
to obtain the optimal stabilized parameter. In this paper, we provide all the results corresponding to the
parameter a
0
in Eq. (25) a
0
= 0.05 for 4-node quadrilateral element. For the linear elements of the 3-node tri-
angle and 4-node quadrilateral, the theoretical rate of convergence for the displacement and pressure in the
L
2
(X) norm is 2 and the energy norm is 1. The results of the convergence rate study were carried out for
the plane stress problem and shown in Figs. 24.
The engineering convergence study was presented for both plane stress and plane strain problems, the nor-
malized tip deection convergence study is given at point (10, 0) while normalized stress convergence is selected
at point (0, 10). The engineering convergence studies can be seen in Figs. 58, which show the numerical
solutions will converge to exact solutions with rened meshes and convergent elements were obtained.
4.2. Plane strain Cooks membrane problem
The Cooks membrane beamproblemhas been widely used as a benchmarking test to check the performance
of developed nite element formulations. Here it was used to demonstrate the performance of the stabilized
Fig. 9. Plane strain Cooks membrane problem.
0
4
8
12
16
20
0 4 8 12 16 20
Elements/side
T
o
p
c
o
r
n
e
r
d
i
s
p
l
a
c
e
m
e
n
t
GLS method
Displacement-based
Fig. 10. Plane strain Cooks membrane problem: convergence for incompressible elasticity.
K. Xia, H. Yao / Applied Mathematical Modelling 31 (2007) 513529 523
nite element formulation on the alleviation of volumetric locking issue. We consider a tapered panel, clamped
on one end and subjected to a shearing load at the free end. The geometry of plane strain Cooks membrane
beam problem is shown in Fig. 9. The plane strain problem can be considered as a three-dimensional problem
with the xed displacement boundary conditions on the front and back surfaces, which provides the most
highly constrained problem and has volumetric locking issue in solid mechanics. In order to test the conver-
gence behavior of the GLS formulation, the problem has been discretized into 2 2, 4 4, 8 8, 16 16 nite
element meshes. Fig. 10 shows that the displacement will converge quickly to the exact solution for nearly
incompressible elasticity (Young modulus E = 250, Poisson ratio m = 0.4999) while element size decreases.
Fig. 11 shows that the GLS formulations can eectively remove locking phenomena while the standard dis-
placement-based formulation will exhibit locking eect. Fig. 12 shows that spurious unstablized pressure eld
will be obtained if Galerkin method was used. Fig. 13 shows the pressure eld can be stabilized if Galerkin/
least-square formulation was used with distorted mesh. Also, Fig. 14 shows the pressure eld can be stabilized
0
2
4
6
8
10
0 0.1 0.2 0.3 0.4 0.5 0.6
Poisson ratio
T
o
p
c
o
r
n
e
r
v
e
r
t
i
c
a
l
d
i
s
p
l
a
c
e
m
e
n
t
Displacement-based method
GLS method
Fig. 11. Plane strain Cooks membrane problem: displacement versus Poisson ratio.
Fig. 12. Unstabilized pressure eld with 4-node Q4 for Galerkin method.
524 K. Xia, H. Yao / Applied Mathematical Modelling 31 (2007) 513529
with distorted composite mesh (3-node triangle and 4-node quadrilateral). Therefore this example demonstrates
the stabilization method is very eective in suppressing the oscillation of pressure eld.
4.3. Stokes ow analogy
Since the equations of Stokes ow are similar to the equations of isotropic nearly incompressible elasticity.
The only dierence is in the interpretation of the variables. For Stokes ow, u will be regarded as the velocity
of the uid. Stokes ow governs highly viscous phenomena. Simulations of the incompressible Stokes ow
with the classical Galerkin method may suer from spurious oscillations arising from the source, which has
Fig. 13. Stabilized pressure feld with 4-node Q4 for GLS method.
Fig. 14. Stabilized pressure eld with composite mesh for GLS method.
K. Xia, H. Yao / Applied Mathematical Modelling 31 (2007) 513529 525
to do with the mixed formulation character of the equations and is limited by the choice of equal linear order
nite element interpolations used to approximate the velocity and pressure elds. A two-dimensional case is
Fig. 15. Lid-driven cavity ow analogy: geometry and boundary conditions.
Fig. 16. Oscillated pressure eld by Galerkin method.
Fig. 17. Oscillated pressure eld by GLS (a = 0.01).
526 K. Xia, H. Yao / Applied Mathematical Modelling 31 (2007) 513529
Fig. 18. Stabilized pressure eld by GLS (a = 0.1).
Fig. 19. Stabilized pressure eld by GLS (a = 0.5).
-15
-10
-5
0
5
10
15
0.0 0.2 0.4 0.6 0.8 1.0 1.2
X-coordinate
P
r
e
s
s
u
r
e
=0.01
=0.1
=0.5
Fig. 20. Driven cavity ow problem: pressure distribution at y = 0.35, 20 20 mesh.
K. Xia, H. Yao / Applied Mathematical Modelling 31 (2007) 513529 527
considered for a square domain with unit side lengths. The boundary conditions are as specied in Fig. 15. The
material properties are the same as those of incompressible elasticity used for the convergence study.
Fig. 16 shows that the pressure eld by the Galerkin method will be highly oscillated. Figs. 1719 shows the
pressure eld can be gradually stabilized by using an appropriate stabilization parameter a, which is included
in the stabilization matrix s . If a is too small, oscillations remain in the pressure eld. In the other extreme, if a
is too large, the stabilization will be too strong and the pressure eld will turn out to be too smooth and might
fail to capture the correct solution in the corners. Therefore an optimum stabilization parameter has to be
obtained when using Galerkin/least-square method as well as PetrovGalerkin method [8,9]. Fig. 20 shows
the pressure distribution along the horizontal line (y = 0.25).
5. Concluding remarks
In this paper, we have derived the second derivatives of shape functions for the stabilized formulation,
which is important for the stabilized Galerkin/least-square method and PetrovGalerkin method. The second
derivatives of shape functions were used to revisit the GLS method. The numerical results conrm that the
Galerkin/least-square method can eectively stabilize the pressure eld and the volumetric locking can auto-
matically removed. The convergence studies show that GLS method promises convergent elements and allows
the use of equal low-order interpolations for both displacement and pressure elds. However, GLS method
still has an intrinsic diculty to study material nonlinearity, which has not been veried before. In the past,
the residual-based terms involved with second derivatives were always neglected without using the second
derivatives of shape functions. For the inelastic problem, accurately accounting for the residual-based term
will require the calculations of the derivatives of the tangent modulus at each integral point, which might pose
a diculty even though GLS and PetrovGalerkin method have been used in solid mechanics. Hopefully this
paper with the second derivatives of shape functions can provide some further insight for the applications of
GLS and PetrovGalerkin method in solid mechanics.
Appendix A
For a three-dimensional problem, one can directly calculate the four modulus D
uu
, D
up
, D
pu
and D
pp
based
on the formulations shown above. For plane stress and plane strain problems, the calculation should be mod-
ied based on the assumptions of plane stress and plane strain. The corresponding tangent modulus are
derived and provided below.
A.1. Plane stress
For plane stress problems, based on the assumption that the stresses in third direction are zero and uniform
strain along the thickness, the tangent modulus is as follows:
D
uu
=
E
1 m
2
2 m
3
2 m
3
0
2 m
3
2 m
3
0
0 0
1 m
2
_
_
_
_
; (62)
D
pp
=
1
K
1 m
1 2m
. (63)
A.2. Plane strain
For plane strain problems, the strains in the third direction are zero, thus we have the corresponding tan-
gent modulus as follows:
528 K. Xia, H. Yao / Applied Mathematical Modelling 31 (2007) 513529
D
uu
=
E
1 m ( ) 1 2m ( )
2 4m
3
2m 1
3
0
2m 1
3
2 4m
3
0
0 0
1 2m
2
_
_
_
_
; (64)
D
pp
=
1
K
. (65)
References
[1] I. Babuska, The nite element method with lagrange multipliers, Numer. Math. 20 (1973) 179192.
[2] T.J.R. Hughes, L.P. Franca, M.A. Balestra, A new nite element formulation for computational uid dynamics: V. Circumventing
the BabuskaBrezzi condition: a stable PetrovGalerkin formulation of the stokes problem accommodating equal-order
interpolations, Comput. Meth. Appl. Mech. Eng. 59 (1986) 8599.
[3] O.C. Ozienkiewicz, R.L. Taylor, The Finite Element Method, vol. 1, fourth ed., McGraw Hill, New York, 1994.
[4] T.J.R. Hughes, L.P. Franca, M.A. Balestra, A new nite element formulation for computational uid dynamics: VIII. The Galerkin/
least square methods for advective-diusive equations, Comput. Meth. Appl. Mech. Eng. 73 (1989) 173189.
[5] F. Brezzi, M.O. Bristeau, L. Franca, M. Mallet, G. Roge, A relationship between stabilized nite element methods and the Galerkin
method with bubble functions, Comput. Meth. Appl. Mech. Eng. 96 (1992) 117129.
[6] C. Baiocchi, F. Brezzi, L. Franca, Virtual bubbles and Galerkin-least-squares type methods (Ga.L.S), Comput. Meth. Appl. Mech.
Eng. 105 (1993) 125141.
[7] E. Onate, Derivation of stabilized equations for numerical solution of advective diusive transport and uid ow problems, Comput.
Meth. Appl. Mech. Eng. 151 (1998) 233265.
[8] O. Klaas, A. Maniatty, M. Shephard, A stabilized mixed nite element method for nite elasticity. Formulation for linear
displacement and pressure interpolation, Comput. Meth. Appl. Mech. Eng. 180 (1999) 6579.
[9] A. Maniatty, Y. Liu, O. Klass, M.S. Shephard, Higher order stabilized nite element method for hyperelastic nite deformation,
Comput. Meth. Appl. Mech. Eng. 191 (2002) 14911503.
[10] S. Commend, A. Truty, T. Zimmermann, Stabilized nite elements applied to elastoplasticity: I. Mixed displacement-pressure
formulation, Comput. Meth. Appl. Mech. Eng. 193 (2003) 35593586.
[11] T.J.R. Hughes, The Finite Element Method: Linear Static and Dynamic Finite Element Analysis, Prentice-Hall, Englewoods Clis,
NJ, 1987.
K. Xia, H. Yao / Applied Mathematical Modelling 31 (2007) 513529 529