Journal of Sound and Vibration (1998) 218(4), 573585

Article No. sv981833

DIFFERENTIAL QUADRATURE METHOD:
APPLICATION TO INITIAL-BOUNDARY-VALUE
PROBLEMS
S. Tox:siriio
DiSGG, Faculty of Engineering, University of Basilicata, Via della Tecnica 3,
85100 Potenza, Italy
(Received 11 March 1998, and in nal form 22 June 1998)
Two non-linear dynamical systems have been considered. In both cases, the
governing equation of motion is reduced to two second-order non-linear
non-autonomous ordinary dierential equations using the dierential quadrature
method with a careful distribution of sampling points. To check the numerical
results, a comparison with those obtained using the Galerkin approach is
proposed.
7 1998 Academic Press
1. INTRODUCTION
The reduction of continuous non-linear systems to single-degree-of-freedom
oscillators can be performed by using various equivalent approaches, as for
example the averaging methods or the RitzGalerkin techniques [1]. The resulting
system can be investigated deeply in the non-linear regime, and harmonic and
non-harmonic response can be obtained. Nevertheless, the drastic reduction to a
single degree of freedom leads to neglect of some qualitative phenomena, as for
example the internal resonances, which can signicantly change the system
behaviour. Consequently, a number of papers have been devoted to the inuence
of higher modes on both regular and chaotic responses [2].
In this paper, the reduction of non-linear boundary-initial-value problems to a
system of coupled non-linear ordinary dierential equations is carried out by using
the generalized dierential quadrature (GDQ) method [3]. The method can be
conveniently applied to the purpose, by substituting the four boundary conditions
into the governing equations [4], so that the number of sampling points reduced
by four represents the degrees of freedom of the discretized system.
Because of the complexities involved in the analysis of multi-degree-of-freedom
systems, attention will be restricted to two-degree-of-freedom systems. So, caution
is required in the choice of the sampling points, since their number must be limited
to six. Actually, the GDQ method uses the Lagrange interpolated polynomial as
test functions and the roots of shifted Legendre polynomial as grid co-ordinates
(a brief overview of the method is given in the next section), and this choice is
0022460X/98/490573 +13 $30.00/0 7 1998 Academic Press



s. 1ox:siriio 574
shown to produce inaccurate results, even in the linear regime, at least if the
number of sampling points is a priori limited. Consequently, a new, more
satisfactory distribution is proposed, which is able to reduce the discretization
errors.
Finally, the strongly non-linear oscillations of the two-degree-of-freedom
systems obtained are studied with the modied LindstedtPoincare` (MLP)
method, using an automatic code [5].
This perturbation approach, which was recently developed by Cheung and
coworkers [6], seems to be the most natural choice for a strongly non-linear,
undamped system where no restriction on the non-linearity can be introduced.
As a guideline, two prototype structural models are used, i.e., the simply
supported beam resting on a non-linear Winkler soil and the slender
clampedhinged beam in which the axis shortening is taken into account. In both
cases, the numerical results have been compared with the results obtained by using
the usual GDQ grid co-ordinates, and with the results given by the classical
Galerkin approach.
For a dierent application of the dierential quadrature method to the vibration
analysis of a geometrically non-linear beam, see reference [7].
2. METHODS OF ANALYSIS
The basis idea of the dierential quadrature method is that the derivative of a
function with respect to a space variable at a given point is approximated as a
weighted linear sum of the function values at all discrete points in the domain of
that variable. In terms of dimensionless variables, one has that, at a point z =z
i
,
the rth-order derivative of a function w(z), dened in the domain (0, 1) with N
discrete grid points, is given by:
$
d
r
w
dz
r
%
z =z
i
= s
N
j =1
A
(r)
ij
w
j
i =1,2, . . . , N, (1)
where A
(r)
ij
are the weighting coefcients of the rth-order derivative.
The weighting coefcients are determined by substituting approximating
functions to the originary function w(z) in equation (1). In the GDQ method [3, 4],
these test functions are assumed to be the Lagrange interpolation polynomials.
The o-diagonal terms of the weighting coefcient matrix of the rst-order
derivative turns out to be:
A
(1)
ij
=
t
N
n =1
n $i
(z
i
z
n
)
(z
i
z
j
) t
N
n =1
n $j
(z
j
z
n
)
i, j =1, 2, . . . , N j $i. (2)



iNi1i:i-noiNb:r.-v:iir ironirxs 575
The o-diagonal terms of the weighting coefcient matrix of the higher-order
derivative are obtained through the recurrence relationship:
A
(r)
ij
=r
$
A
(r 1)
ii
A
(1)
ij

A
(r 1)
ij
(z
i
z
j
)
%
i, j =1, 2, . . . , N j $i, (3)
where 2 Er E(N1).
The diagonal terms of the weighting coefcient matrix are given by:
A
(r)
ii
= s
N
n =1
n $i
A
(r)
in
i =1, 2, . . . , N, (4)
where 1 Er E(N1).
Assuming the Lagrange interpolation polynomials as test functions, there is no
restriction in the choice of the grid co-ordinates. So, in order to have more accurate
solutions, it is possible to generate the sampling points as follows:
z
i
=
1
2
$
1 cos
(i 1)
(N1)
p
%
i =1, 2, . . . ,N. (5)
In order to overcome the problem of the d-points [8], Shu and Du [4] support the
GDQ method with a direct substitution of the boundary conditions into the
governing equation. In the present work, the Shu and Du approach is used, but
the grid co-ordinates are not given by equation (5).
It is well-known that the distribution of sampling points must be symmetric and
it must result in z
1
=0 and z
N
=1. So, it will be sufcient to determine (N2)/2
(two, for N=6) points after the rst, since the others are automatically xed by
the relation that denes the symmetry of the distribution:
z
i
+z
(N+1 i)
=1.
By induction, the following rule has been deduced:
z
i
=
0
i 1
N1
1
Nb
i
/izi
, (6)
where b
i
are unknown coefcients to be xed.
As it has been said, for the symmetry, only b
2
and b
3
need to be xed.
This distribution has been checked with a linear analysis: the comparison term
between numerical and exact results has been given by the critical value of the axial
force. After many numerical simulations, it has been possible to see that the third
co-ordinate plays an important role in obtaining good results. In fact, for a certain
value of b
3
, the non-linear results are not inuenced very much by varying b
2
,
whereas the linear results are the same. This is true even if the second point is near
to the third. Finally, it has been noted that the results are in good agreement for



s. 1ox:siriio 576
values of b
3
close to N/5 =12. For b
3
=12 and b
2
=1, for example, the following
distribution of sampling points is obtained:
{0, 0033, 0281, 0719, 0966, 1} (7)
This is the distribution which is used for the nal non-linear results.
3. THE FIRST MODEL: A SIMPLY SUPPORTED BEAM ON A NON-LINEAR
WINKLER SOIL
Consider a simply supported beam with span L, Young modulus E, moment
of inertia I, mass per unit length m and cross-sectional area A, which rests on an
hardening non-linear elastic foundation and which is subjected to a compressive
load P and to an exciting transverse force F(z, t) =F(z) cos v t. The foundation
is supposed to be dened by the following loaddisplacement relationship:
q(z) =k
1
v(z) +k
3
v(z)
3
, where q(z) is the force per unit length, k
1
is the linear
Winkler foundation stiness and k
3
q0 is the hardening non-linear elastic
foundation stiness.
If the beam is considered to be slender, the equation of motion can be written
as:
m
1
2
v
1t
2
+EI
1
4
v
1z
4
+P
1
2
v
1z
2
+k
1
v +k
3
v
3
=F(z) cos v t. (8)
Equation (8) can be conveniently written in terms of dimensionless variables as:
1
2
w
1t
2
+
1
4
w
1z
4
+s
1
2
w
1z
2
+u
1
w+u
3
w
3
=f(z) cos vt, (9)
where
w=
v
L
, z =
z
L
, t =
X
EI
m
t
L
2
, v=
X
m
EI
L
2
v ,
s =
PL
2
EI
, u
1
=
k
1
L
4
EI
, u
3
=
k
3
L
6
EI
, f(z) =
F(z)L
3
EI
.
In the Galerkin method, the deection is approximated by:
w(z, t) = s
M
m=1
u
m
(t)f
m
(z), (10)
where u
m
are generalized co-ordinates and f
m
=sin mpz are the characteristic
modal functions of a simply supported beam subjected to a compressive load.
If two co-ordinates are employed, equation (10) can be inserted into equation
(9), the result multiplied by f
n
and integrated over the span. Then, the
orthogonality property of the modal functions f
n
, allows one to write the
following equations:
u
i
+v
2
i0
u
i
+u
3
(
3
4
u
3
i
+
1
2
u
i
u
2
k
) =f
i
cos vt i, k =1, 2 k $i, (11)



iNi1i:i-noiNb:r.-v:iir ironirxs 577
where the dot denotes dierentiation with respect to t, and
v
2
i0
=i
2
p
2
(i
2
p
2
s) +u
1
, f
i
=2
g
1
0
f(z)f
i
dz
are the natural frequencies squared and the excitation amplitudes, respectively.
The dierential quadrature analog of equation (9) may be written, using the
quadrature rules in the z co-ordinate only, as
w
i
+ s
N
j =1
L
ij
w
j
+u
1
w
i
+u
3
w
3
i
=f
i
cos vt i =1, 2, . . . , N, (12)
where
L
ij
=A
(4)
ij
+sA
(2)
ij
, f
i
=f(z
i
)
and N is the number of the sampling points.
It may be noted that the derivative is total, since w
i
=w
i
(t) at a sampling point
z =z
i
. The boundary conditions are:
w
1
=0 w0
1
=0,
w
N
=0 w0
N
=0,
which can be immediately written as:
s
N1
j =2
A
(2)
1j
w
j
=0, s
N1
j =2
A
(2)
Nj
w
j
=0. (13, 14)
The rst two (geometric) boundary conditions, in z =0 and in z =1, have already
been imposed by changing the summation limits. The other two (natural)
boundary conditions, now expressed by equations (13) and (14), can be used in
order to obtain w
2
and w
(N1)
:
w
2
=
1
D
s
N2
j =3
E
j
w
j
, w
(N1)
=
1
G
s
N2
j =3
H
j
w
j
, (15, 16)
where
D=A
(2)
N2

A
(2)
N(N1)
A
(2)
1(N1)
A
(2)
12
, E
j
=A
(2)
Nj

A
(2)
N(N1)
A
(2)
1(N1)
A
(2)
1j
,
G=A
(2)
N(N1)

A
(2)
N2
A
(2)
12
A
(2)
1(N1)
, H
j
=A
(2)
Nj

A
(2)
N2
A
(2)
12
A
(2)
1j
.
Finally, w
2
and w
(N1)
can be substituted into the equations system (12), giving:
w
i
+ s
N2
j =3
R
ij
w
j
+u
3
w
3
i
=f
i
cos vt i =3, . . . . , (N2), (17)



s. 1ox:siriio 578
where
R
ij
=L
ij

E
j
D
L
i2

H
j
G
L
i(N1)
+u
1
d
ij
(18)
and d
ij
is the Kronecker operator.
Choosing N=6, one obtains a set of two non-linear ordinary dierential
equations, coupled in the linear part. As already stated, this sytem will be studied
by means of the MLP perturbation method, so that it is necessary to reduce it to
the following normalized form
u
i
+v
2
i0
u
i
+(k
i1
u
3
i
+k
i2
u
2
i
u
j
+k
i3
u
i
u
2
j
+k
i4
u
3
j
) =f
i
cos vt i, j =1, 2 j $i (19)
and this is possible through a change of reference.
The direction cosines of the new rotated axes are the eigenvectors of the matrix
R, whose elements are dened by equation (18).
Let V be the matrix which has columns that are the eigenvectors of R, u the
new co-ordinates vector, w the old co-ordinates vector.
From the relation u =Vw, one obtains:
0
w
1
w
21
=
1
C
0
V
22
V
21
V
12
V
1110
u
1
u
21
, (20)
where C= det (V) and w
i
=w
(i +2)
.
After some algebra, it is possible to write the non-linearity coefcients as
follows:
k
11
=
u
3
C
3
V
3
22
, k
12
=3
u
3
C
3
V
2
22
V
12
, k
13
=3
u
3
C
3
V
22
V
2
12
,
k
14
=
u
3
C
3
V
3
12
,
k
21
=
u
3
C
3
V
3
11
, k
22
=3
u
3
C
3
V
2
11
V
21
, k
23
=3
u
3
C
3
V
11
V
2
21
,
k
24
=
u
3
C
3
V
3
21
.
It is obvious that the eigenvalues of R are the squares of the natural frequencies
v
i0
of the normalized system.
Now, let s
crit
be the critical value of the dimensionless compressive axial load
for the associated linear system
The exact value of s
crit
is:
s
ex
crit
=p
2
0
1 +
u
1
p
4
1
.
If u
1
=1, then s
ex
crit
=997092.



iNi1i:i-noiNb:r.-v:iir ironirxs 579
Using the quadrature method with six grid points (N=6) and with the classical
grid distribution [cf. equation (5)], the non-dimensional critical load is found to
be equal to s
DQM
crit
=894676, which is more than 10% smaller than the exact value.
This error implies that the subsequent non-linear analysis will be almost
meaningless.
If seven points are employed (N=7), then the non-dimensional critical load
becomes s
DQM
crit
=100691, which is less than 1% larger than the exact value.
Nevertheless, as already pointed out, the introduction of other grid points greatly
complicates the non-linear analysis, and should be avoided, as far as possible.
If the proposed grid distribution [cf. equation (6)] is employed, the
non-dimensional critical loads reported in Table 1 are obtained, from which it is
possible to realize the good improvement of the linear results.
4. THE SECOND MODEL: CLAMPEDHINGED BEAM
Now consider a slender clampedhinged beam, having the same physical and
mechanical characteristics of the preceding beam and subjected to the same
harmonic exciting force. In this case, however, the beam is considered to be so
slender that the shortening of the beam axis cannot be neglected.
The equation of motion is:
m
1
2
v
1t
2
+EI
1
4
v
1z
4

EA
2L
0g
L
0
0
1v
1z
1
2
dz
1
1
2
v
1z
2
=F(z) cos v t. (21)
This equation in terms of dimensionless variables becomes:
1
2
w
1t
2
+
1
4
w
1z
4
k
0g
1
0
0
1w
1z
1
2
dz
1
1
2
w
1z
2
=F(z) cos vt, (22)
where k =AL
2
/2I.
In this second case, the f
m
appearing in equation (10) are given by:
f
m
=A
m
(sin l
m
z B
m
sinhl
m
z) m=1, 2,
where
A
m
=
$
1
2
(1 B
2
m
) +
1
4l
m
(B
2
m
sinh 2l
m
sin 2l
m
)
%
1/2
, B
m
=
sin l
m
sinh l
m
and the l
m
=zv
m0
are the roots of the transcendental equation tan l
m
=tanh l
m
.
T:nir 1
Non-dimensional critical load of a simply supported beam on non-linear soil
b
3
1056 115 118 119 1194 1195 12
s
crit
91896 96984 98740 99358 99568 99778 99993



s. 1ox:siriio 580
The two Galerkins equations are:
u
n
+v
2
n0
u
n
s
2
m,p,q =1
a
nmpq
u
m
u
p
u
q
=f
n
cos vt n =1, 2, (23)
where
a
nmpq
=2k
g
1
0
1u
m
1z
1u
p
1z
dz
g
1
0
1
2
u
q
1z
2
f
n
dz, f
n
=2
g
1
0
f(z)f
n
dz
and
a
1121
=a
1211
, a
1122
=a
1212
,
a
2121
=a
2211
, a
2122
=a
2212
.
The system (23) is in normal form with:
k
11
=a
1111
, k
12
=(a
1112
+a
1121
+a
1211
),
k
13
=(a
1122
+a
1212
+a
1221
), k
14
=a
1222
,
k
21
=a
2222
, k
22
=(a
2122
+a
2212
+a
2221
),
k
23
=(a
2112
+a
2121
+a
2211
), k
24
=a
2111
.
Before applying the quadrature method to equation (21), it is convenient to
rewrite the integral at the rst member of this equation as:
$
1w
1z
w
%
1
0

g
1
0
1
2
w
1z
2
w dz.
The bracket is null, so using the quadrature rules in the space co-ordinate, the
following equations are obtained:
w
i
+ s
N1
j =2
A
(4)
ij
w
j
+k s
N1
k,l,m=2
C
k
A
(2)
kl
A
(2)
im
w
k
w
l
w
m
=f
i
cos vt i =2, . . . , (N1),
(24)
where C
k
are the weighting coefcients of the integral.
It is also worth noting that the boundary conditions
w
1
=0, w
N
=0
have been already introduced.
The remaining boundary conditions are written as follows:
s
N1
j =2
A
(1)
1j
w
j
=0, s
N1
j =2
A
(2)
Nj
w
j
=0. (25, 26)



iNi1i:i-noiNb:r.-v:iir ironirxs 581
T:nir 2
Non-dimensional critical load of a clampedhinged beam
b
3
1194 12 122 1225 123
s
crit
195703 197401 20094 201624 203009
From these two equations, one can obtain w
2
and w
(N1)
, which have the same form
as (15) and (16), with:
D=A
(2)
N2

A
(2)
N(N1)
A
(1)
1(N1)
A
(1)
12
, E
j
=A
(2)
Nj

A
(2)
N(N1)
A
(1)
1(N1)
A
(1)
1j
,
G=A
(2)
N(N1)

A
(2)
N2
A
(1)
12
A
(1)
1(N1)
, H
j
=A
(2)
Nj

A
(2)
N2
A
(1)
12
A
(1)
1j
.
Finally, w
2
and w
(N1)
have to be substituted into the equations system (24). The
weighting coefcients C
k
are derived by the NewtonCotes integration formulas:
C
k
=
g
1
0
t
N
i =1
i $k
z z
i
z
k
z
i
dz,
using unequally spaced sampling points.
After all the substitutions, for N=6, the nal form of the system (24) may be
written as:
w
i
+ s
4
j =3
S
ij
w
j
+k s
4
k,l,m=3
T
iklm
w
k
w
l
w
m
=f
i
cos vt i =3, 4, (27)
where
S
ij
=A
(4)
ij

E
j
D
A
(4)
i2

H
j
G
A
(4)
i(N1)
. (28)
The T
iklm
coefcients have been calculated with a simple Mathematica code.
The normalization of the system (27) requires the calculation of the eigenvalues
and the eigenvectors of the matrix S whose elements are dened by equation (28).
The new k
ij
coefcients are more complex than the preceding ones and they have
been calculated with the quoted Mathematica code.
Imagine now that the beam examined is subjected to an axial load P.
Let s be the dimensionless compressive axial load and s
crit
the critical value of
it for the associated linear system.
The exact value of s
crit
is s
ex
crit
=201907.
Using the quadrature method with the distribution (5), for N=6 results
s
DQM
crit
=165861.
Using the quadrature method with the distribution (6), one obtain the results
reported in Table 2.



8
6
4
2
0
2
4
6
8
9.0 10.5 10.0 9.5 11.5 11.0 12.0
A
1
0
s. 1ox:siriio 582
5. NUMERICAL RESULTS
The solutions obtained by using the sampling points generated by the relation
(5) are compared with the results obtained by using the distribution given by (7).
The cases f =0, f =1, f =100 have been studied. Since f is constant, operating
with the quadrature method results in f
i
=f, whereas with the Galerkins
procedure it results in f
i
=2f f
1
0
f
i
dz.
Fundamental resonance with v1v
10
has been considered.
Figures 1 and 2 show the relationship frequencyinitial amplitude of the rst
mode (A
10
) for the examined cases and f =0.
The curves obtained for f q0 are not reported, because only for f =100 and
for small initial amplitudes A
10
, the discrepancies become noticeable.
5.1. rirs1 xobri
Assume s =01 and u
1
=u
3
=1. The matrix R reects the system symmetry with
the peculiarity of having equal diagonal elements. It will be:
R=
0
5162777
4296952
4296952
5162777
1
if the distribution (5) is adopted and:
R=
0
6017479
5039223
5039223
6017479
1
if the distribution (7) is chosen.
Figure 1. Free vibration response vA10 for the simply supported beam examined. ,
Distribution (5); distribution (7); ---- Galerkin method.



8
6
4
2
0
2
4
6
8
10 160 110 60 260 210 310 360 410 460
A
1
0
iNi1i:i-noiNb:r.-v:iir ironirxs 583
Figure 2. Free vibration response vA10 for the clampedhinged beam examined. ,
Distribution (5); distribution (7); ---- Galerkin method.
The natural frequencies are: for the distribution (5), v
10
=9305 and
v
20
=307568; for the distribution (7), v
10
=99411 and v
20
=332666. As one can
see, v
20
13v
10
, which is the condition of internal resonance.
The non-linear coefcients are, for both cases:
k
11
=035355, k
12
=106066, k
13
=k
12
, k
14
=k
11
,
k
21
=k
11
, k
22
= k
12
, k
23
=k
12
, k
24
= k
11
,
There is a s value (s
0
) close to 228 for which R
ii
=0.
If s Qs
0
, the results are R
ii
q0 and R
ij
Q0; if s qs
0
, then R
ii
Q0 and R
ij
Q0,
but only if s Es
crit
the frequencies are real. In particular, if s =s
crit
all the R
elements are equal in absolute value so that one frequency is null. If s qs
crit
one
frequency becomes imaginary.
Finally, as expected, as s is increased, the frequency values reduce since all the
R elements in absolute value reduce.
The application of Galerkins method produces the following results:
v
10
=98703, v
20
=394411.
5.2. srcoNb xobri
It is assumed that k =12.
The quadrature method with the distribution (5) leads to the following results:
S=
0
13095985
4788968
7521546
5369786
1



s. 1ox:siriio 584
v
10
=144754, v
20
=404603,
k
11
=384714, k
12
=4473144, k
13
=276444, k
14
=8228388,
k
21
=56352, k
22
= 1398684, k
23
= 1365696, k
24
= 3582516.
The quadrature method with the distribution (7) gives:
S=
0
16040068
7474094
766903
6674618
1
v
10
=156693, v
20
=450105,
k
11
=508389, k
12
=590325, k
13
=344923, k
14
=115799,
k
21
= 220258, k
22
= 231226, k
23
= 353211, k
24
= 573842.
The results obtained by the application of the Galerkins method are:
v
10
=154182, v
20
=498238,
k
11
=3172632, k
12
=35394696, k
13
=12 6846096, k
14
=4390776,
k
21
=43 94088, k
22
=13 172328, k
23
=k
13
, k
24
=11798232.
6. CONCLUSIONS
In this paper a convenient approach is proposed, in order to study the non-linear
dynamics of continuous systems taking into account the internal resonances and
other phenomena which are neglected in the usual reduction to a single-degree-of-
freedom oscillator. In particular, attention is drawn to two-degree-of-freedom
systems. So, the initial-boundary value problem is rst discretized by using the
generalized dierential quadrature method, with an optimized choice of the
sampling points. In this way, the non-linear partial dierential equation reduces
to a set of two non-linear ordinary dierential equations. The problem, so
governed by two coupled equations, is then investigated in detail by using a
modied version of the LindstedtPoincare` method, in which the non-linearity
needs not to be small.
As an example, two simple structural models have been investigated by using
three dierent methods, and the numerical results show that the proposed
approach behaves quite satisfactorily.
ACKNOWLEDGMENT
A sincere thanks to the referee who has brought to the attention of the author
reference [7].



iNi1i:i-noiNb:r.-v:iir ironirxs 585
REFERENCES
1. W. SzrxiiiN si:-S1iiNici: 1990 The Behaviour of Nonlinear Vibrating Systems,
Volume I. Dordrecht: Kluwer Academic Publishers.
2. T. Sxriov:-Rr.Noibs and E. H. Dovrii 1996 International Journal of Non-linear
Mechanics 31, 931939. The role of higher modes in the chaotic motion of the buckled
beamI.
3. R. M. LiN, M. K. Lix and H. Di 1994 Computer and Structures 53, 993999.
Deection of plates with nonlinear boundary supports using generalized dierential
quadrature.
4. C. Sni and H. Di 1997 International Journal of Solids and Structures 34, 819835.
Implementation of clamped and simply supported boundary conditions in the GDQ
free vibration analysis of beams and plates.
5. C. Fr:Nciosi and S. Tox:siriio 1998 Journal of Sound and Vibration 211, 145156.
The use of Mathematica for the analysis of a strongly nonlinear two-degree-of-free-
dom system by means of the modied LindstedtPoincare` method.
6. S. H. CnrN and Y. K. CnriNc 1996 Journal of Sound and Vibration 193, 751762.
A modied LindstedtPoincare` method for a strongly non-linear two-degree-of-free-
dom system.
7. Y. FrNc and C. W. Brr1 1992 Nonlinear Dynamics 3, 1318. Application of the
quadrature method to exural vibration analysis of a geometrically nonlinear beam.
8. C. W. Brr1 and M. M:iii 1996 Applied Mechanic Review 49, 128. Dierential
quadrature method in computational mechanics: a review.