157

15

Differential Quadrature Method

15.1 Introduction
We have already discussed Akbari–Ganji’s method (AGM), exp-function
method, Adomian decomposition method (ADM), homotopy perturbation
method (HPM), variational iteration method (VIM), and homotopy analysis
method (HAM) in Chapters 9, 10, 11, 12, 13, and 14, respectively. This chapter
presents another effective numerical method that approximates the solution
of the PDEs by functional values at certain discretized points. This method can
be applied with considerably less number of grid points, whereas the methods
like finite difference method (FDM), finite element method (FEM), and finite
volume method (FVM) as given in Chapters 5, 6, and 7, respectively, may need
more number of grid points to obtain the solution. However, FDM, FEM, and
FVM are versatile methods which may certainly be used to handle regular as
well as irregular domains.
Approximating partial derivatives by means of weighted sum of function
values is known as Differential Quadrature (DQ). The differential quadrature
method (DQM) was proposed by Bellman et al. [1–4]. Determination of
weighted coefficients plays an important and crucial role in DQM. As such, an
effective procedure for finding weighted coefficients suggested by Bellman is
to use a simple algebraic formulation of weighted coefficient with the help of
coordinates of grid points. Here, the coordinate of the grid points are the roots
of the base functions like Legendre polynomials, Chebyshev polynomials, etc.
Interested readers may see Refs. [5–10] and references therein for more details
and application of DQM.

15.2 DQM Procedure

In this section, DQM has been illustrated [1, 2] by considering a second-order
nonlinear partial differential equation (PDE) of the form
ut (x, t) = f (x, t, u(x, t), ux (x, t) , uxx (x, t)), (15.1)

Advanced Numerical and Semi-Analytical Methods for Differential Equations, First Edition.
Snehashish Chakraverty, Nisha Rani Mahato, Perumandla Karunakar, and Tharasi Dilleswar Rao.
© 2019 John Wiley & Sons, Inc. Published 2019 by John Wiley & Sons, Inc.
158 15 Differential Quadrature Method

subject to initial condition

u(x, 0) = g(x). (15.2)
The DQ of the first- and second-order spatial derivatives at the grid points xi
are given by

∑
N
ux (xi , t) = aij u(xj , t), i = 1, 2, 3, … , N, (15.3)
j=1

∑
N
uxx (xi , t) = bij u(xj , t), i = 1, 2, 3, … , N, (15.4)
j=1

By plugging Eqs. (15.3) and (15.4) in Eq. (15.1), we obtain a system of N ordi-
nary differential equations (ODEs) as follows:
( )
∑
N
∑
N
ut (xi , t) = f xi , t, u(xi , t), aij u(xj , t), bij u(xj , t) , (15.5)
j=1 j=1

subject to initial conditions

u(xi , 0) = g(xi ). (15.6)
The weighted coefficients aij and bij are to be found using N grid points xi and
these grid points depend on the considered polynomials like Legendre, Cheby-
shev, etc. In this chapter, shifted Legendre polynomials PN∗ (xi ) in [0, 1] have
been used to find weighted coefficients.
The shifted Legendre polynomials [11, 12] of order N are given as

∑
N
(N + k)!
PN∗ (xi ) = (−1)N+k xk
k=0
(N − k)!(k!)2 i

First, five shifted Legendre polynomials are

P0∗ (xi ) = 1
P1∗ (xi ) = 2xi − 1
P2∗ (xi ) = 6x2i − 6xi + 1
P3∗ (xi ) = 20x3i − 30x2i + 12xi − 1
P4∗ (xi ) = 70x4i − 140x3i + 90x2i − 20xi + 1
The weighted coefficients in terms of shifted Legendre polynomial PN∗ (xi ) [2]
are taken as follows:
P′N∗ (xi )
aij = , i≠j (15.7)
(xi − xj )P′N∗ (xj )
 15.3 Numerical Examples 159

and
1 − 2xi
aii = , i=j (15.8)
2xi (xi − 1)
Once all aii are obtained, it is then easy to find bii following the below proce-
dure.
Let us denote Eq. (15.3) as
𝜕u
= Au. (15.9)
𝜕x
where A = [aij ]N × N .
The second-order derivative can be approximated as
( )
𝜕2u 𝜕 𝜕u 𝜕 𝜕u
= = (Au) = A = A(Au) = A2 u, (15.10)
𝜕x2 𝜕x 𝜕x 𝜕x 𝜕x
where [bij ] = A2 .
Hence, weighted coefficients bij can be obtained by squaring A as
[bij ] = [aij ]N×N × [aij ]N×N .
Substituting the obtained weighted coefficients aij and bij in Eq. (15.5) and
then using a suitable method to solve the system of ODEs, the solution of the
model problem may be found at selected grid points.
Next, we solve two test problems to demonstrate the DQM.

15.3 Numerical Examples

In this section, two nonlinear PDEs are solved by considering different N for
better understanding of the DQM.

Example 15.1 Let us consider the following PDE:

1
ut (x, t) = x2 + u2x (x, t), (15.11)
4
subject to the initial condition u(x, 0) = 0.

Solution First, we need to write derivative term as approximating sum except

the derivatives with respect to t.
∑N
Hence, replacing ux (x, t) by j=1 aij u(xj , t) in Eq. (15.11), we get a set of ODEs
as
[N ]2
1 ∑
ut (xi , t) = xi +
2
a u(x , t) , (15.12)
4 j=1 ij j
160 15 Differential Quadrature Method

subject to the initial conditions

u(xi , 0) = 0, i = 1, 2, … , N (15.13)
Here, we demonstrate the procedure for N = 3. Then, Eq. (15.12) may be
written as
1
ut (xi , t) = x2i + [ai1 u(x1 , t) + ai2 u(x2 , t) + ai3 u(x3 , t)]2 , i = 1, 2, 3
4
(15.14)
subject to the conditions u(xi , 0) = 0, i = 1, 2, 3.
The matrix form of Eq. (15.14) may be written as
2
⎡ut (x1 , t)⎤ ⎧⎡a11 a12 a13 ⎤ ⎡u(x1 , t)⎤⎫ ⎡x21 ⎤
⎢ut (x2 , t)⎥ = ⎨⎢a21 a22 a23 ⎥ ⎢u(x2 , t)⎥⎬ + ⎢⎢x2 ⎥⎥
1 ⎪ ⎢ ⎥ ⎢ ⎥ ⎪
(15.15)
⎢ ⎥ 4 ⎪⎢ ⎥⎢ ⎥⎪
2
⎢x2 ⎥
⎣ut (x3 , t)⎦ ⎩⎣a31 a32 a33 ⎦ ⎣u(x3 , t)⎦⎭ ⎣ 3⎦
Now we find the weighted coefficient aij using shifted Legendre polynomial
P3∗ (xi ).
The grid points xi , i = 1, 2, 3 are the roots of P3∗ (xi ) = 20x3i − 30x2i + 12xi − 1 =
0 and they may be found as
x1 = 0.112 701 665 4
x2 = 0.5
and
x3 = 0.887 298 334 6
Here, the weighted coefficients of the derivative that are found using
Eqs. (15.7) and (15.8) are
⎡−3.872 983 35 −2.581 988 90 −1.290 994 45⎤
[aij ] = ⎢−2.581 988 90 0 −2.581 988 90⎥
⎢ ⎥
⎣ 1.290 994 45 −2.581 988 90 −3.872 983 35⎦
and
⎡0.012 701 67⎤
[x2i ] = ⎢0.250 000 00⎥ , i = 1, 2, 3
⎢ ⎥
⎣0.787 298 33⎦
The system of nonlinear ODEs (15.12) along with initial conditions has
been solved using the fourth-order Runge–Kutta method and the results are
reported in Table 15.1.
Here, the exact solution [13] of Eq. (15.11) is
u(x, t) = x2 tan(t). (15.16)
 15.3 Numerical Examples 161

Table 15.1 Comparison of solutions of PDE (15.11) by

DQM for N = 3 with the exact solution.

t x DQM solution Exact solution

0.01 x1 0.00012726 0.00012702

x2 0.00250033 0.00250008
x3 0.00787413 0.00787325
0.1 x1 0.00151955 0.00127442
x2 0.02533779 0.02508367
x3 0.07989748 0.07899332

From the results presented in Table 15.1 it may be confirmed that solutions
at particular nodes obtained by the DQM are in good agreement with that of
the exact solution. It may be worth mentioning that we have considered here a
small N value (N = 3) so that readers can understand the method well. However,
better results may be obtained by taking a large value of N.
This effect of N is shown in Example 15.2 by considering two different N
values for the same problem.

Example 15.2 Let us consider now a nonhomogeneous advection equation

from Ref. [10] as
ut + uux = x (15.17)
subject to the initial condition: u(x, 0) = 2.
Solution The exact solution of Eq. (15.17) is u = 2 sech t + x tanh t [10].
Now we solve PDE (15.17), with the given initial condition using the DQM
for N = 4 and N = 7.
∑N
By replacing ux (x, t) by j=1 aij u(xj , t) in Eq. (15.17), we get a set of nonlinear
system of ODEs as
∑
N
ut (xi , t) = −u(xj , t) aij u(xj , t) + xi (15.18)
j=1

The grid points of PN∗ (x) for N = 4 may be obtained by following the previously
mentioned procedure as
x1 = 0.069 431 844 202 974;
x2 = 0.330 094 782 075 720;
x3 = 0.669 990 521 792 431;
x4 = 0.930 568 155 790 230;
162 15 Differential Quadrature Method

The weighted coefficients using Eqs. (15.7) and (15.8) are found as

⎡−6.664 000 47 −3.836 372 01 −1.665 116 23 −1.161 256 34⎤

⎢ 3.836 372 02 −0.768 342 22 −2.942 078 65 −1.665 352 78⎥
[aij ] = ⎢
1.665 116 23 2.942 078 65 0.768 828 78 −3.837 627 91⎥
⎢ ⎥
⎣ 1.161 256 34 1.665 352 78 3.837 627 91 6.664 000 47 ⎦
By plugging weighted coefficients aij and nodes xi in Eq. (15.18) and solving
nonlinear system of differential equations using the Runge–Kutta method of
order four, we may get the solution of the nonlinear PDE (15.17). The results
obtained for selected nodes xi are given in Table 15.2 for two t values t = 0.01
and 0.1 along with the exact solution [10].
Next, we take N = 7 to find the solution of the PDE (15.7) using the DQM.
Here, we get seven grid points {xi , i = 1, 2, 3, 4, 5, 6, 7}. These are the roots of
Legendre polynomial equation of order 7
P7∗ (x) = 3 432x7 − 12 012x6 + 166 632x5 − 11 550x4 + 4 200x3 − 756x2
+ 56x − 1 = 0.
So, the nodes xi are
x1 = 0.025 446 04; x2 = 0.129 234 40; x3 = 0.297 077 42; x4 = 0.5;
x5 = 0.702 922 57; x6 = 0.800 765 59; x7 = 0.974 553 95
Proceeding as above using Eqs. (15.7) and (15.8), we may find weighted coef-
ficients [aij ] for i = 1, 2, …, 7. Substituting 49 weighted coefficients [aij ] and 7
grid points xi in Eq. (15.18) gives a nonlinear system of ODEs with constant
coefficients. Then, the results obtained are presented in Table 15.3 after solving
the system of nonlinear ODEs using the fourth-order Runge–Kutta method.

Table 15.2 Comparison of solutions of PDE (15.17) by

the DQM for N = 4 with the exact solution.

t x DQM solution Exact solution

0.01 x1 2.65887716 2.00059430

x2 2.03204598 2.00320084
x3 1.91820603 2.00659969
x4 1.56030917 2.00920538
0.1 x1 2.74196598 1.99696163
x2 2.03117563 2.02294138
x3 1.90734032 2.05681811
x4 1.52403620 2.08278936
 15.3 Numerical Examples 163

Table 15.3 Comparison of solutions of PDE (15.17) by

the DQM for N = 7 with the exact solution.

t x DQM solution Exact solution

0.01 x1 1.99986243 2.00015446

x2 2.00148324 2.00119231
x3 2.00387433 2.00287068
x4 2.00688137 2.00489984
x5 2.01417620 2.00692900
x6 2.01936977 2.00790739
x7 2.03492958 2.00964522
0.1 x1 1.99859172 1.99257765
x2 2.01483750 2.00292203
x3 2.03886500 2.01965061
x4 2.06911125 2.03987550
x5 2.14410838 2.06010038
x6 2.19849403 2.06985220
x7 2.36865789 2.08717334

× 10–3
1

0.9
N=3
N=7
0.8

0.7
Absolute error

0.6

0.5

0.4

0.3

0.2

0.1

0
0 0.2 0.4 0.6 0.8 1
x

Figure 15.1 Absolute error plots for N = 3, N = 7 at t = 0.01.

164 15 Differential Quadrature Method

0.7
N=4
N=7
0.6

0.5
Absolute error

0.4

0.3

0.2

0.1

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x

Figure 15.2 Absolute error plots for N = 4, N = 7 at t = 0.01.

From Tables 15.2 and 15.3, one may confirm that the results obtained are
in good agreement with the exact solution. It may be observed that error in
solution for N = 4 is more than that for N = 7.
Finally, error analysis has been done for both the problems discussed above
in Examples 15.1 and 15.2. In the case of Example 15.1, if we consider N = 7,
the error between the solutions by the DQM and exact is very less. The corre-
sponding error plots for N = 3 and N = 7 at t = 0.01 are depicted in Figure 15.1.
The similar type of behavior may be observed in the case of Example 15.2 also.
Figure 15.2 represents the error plots for N = 4 and N = 7 at t = 0.01 for Example
15.2.
From Figures 15.1 and 15.2, one may conclude that the error is drastically
decreasing as the N value increases from 3 and 4 to 7 as shown in Examples 15.1
and 15.2, respectively.

Exercise
1 Find the approximate solution of the solution of PDE ut (x, t) =
x2 − 14 u2x (x, t), subject to the initial condition u(x, 0) = 0 using the
DQM with N = 3, 5, and 7. Compare the approximate DQM solution in
each case with the exact solution u(x, t) = x2 tanh(t).
 References 165

References
1 Bellman, R. and Casti, J. (1971). Differential quadrature and long-term inte-
gration. Journal of Mathematical Analysis and Applications 34: 235–238.
2 Bellman, R., Kashef, B.G., and Casti, J. (1972). Differential quadrature: a
technique for the rapid solution of nonlinear partial differential equations.
Journal of Computational Physics 10: 40–52.
3 Bellman, R., Kashef, B.G., Lee, E.S., and Vasudevan, R. (1975a). Solving hard
problems by easy methods: differential and integral quadrature. Computers
and Mathematics with Applications 1: 133–143.
4 Bellman, R., Kashef, B.G., Lee, E.S., and Vasudevan, R. (1975b). Differential
quadrature and splines. Computers and Mathematics with Applications 1:
371–376.
5 Bert, C.W., Jang, S.K., and Striz, A.G. (1989). Nonlinear bending analysis
of orthotropic rectangular plates by the method of differential quadrature.
Computational Mechanics 5: 217–226.
6 Bert, C.W. and Kang, K.J. (1996). Stress analysis of closely-coiled helical
springs using differential quadrature. Mechanics Research Communications
23 (6): 589–598.
7 Bert, C.W. and Malik, M. (1996). Differential quadrature method in compu-
tational mechanics. Applied Mechanics Reviews 49 (1): 1–28.
8 Shu, C. and Chen, W. (1999). On optimal selection of interior points for
applying discretized boundary conditions in DQ vibration analysis of beams
and plates. Journal of Sound and Vibration 222 (2): 239–257.
9 Shu, C. (1999). Application of differential quadrature method to simu-
late natural convection in a concentric annulus. International Journal for
Numerical Methods in Fluids 30: 977–993.
10 Wazwaz, A.M. (2010). Partial Differential Equations and Solitary Waves
Theory. Beijing: Springer Science & Business Media, Higher Education
Press.
11 Ezz-Eldien, S.S., Doha, E.H., Baleanu, D., and Bhrawy, A.H. (2017). A
numerical approach based on Legendre orthonormal polynomials for
numerical solutions of fractional optimal control problems. Journal of
Vibration and Control 23 (1): 16–30.
12 Bhrawy, A.H., Alofi, A.S., and Ezz-Eldien, S.S. (2011). A quadrature tau
method for fractional differential equations with variable coefficients.
Applied Mathematics Letters 24 (12): 2146–2152.
13 Debnath, L. (2012). Nonlinear Partial Differential Equations for Scientists
and Engineers. New York: Springer Science & Business Media.