applying multiquadric quasi-interpolation to solve burgers’ equation
TRANSCRIPT
Applied Mathematics and Computation 172 (2006) 472–484
www.elsevier.com/locate/amc
Applying multiquadric quasi-interpolationto solve Burgers� equationq
Ronghua Chen a,b,*, Zongmin Wu c
a Institute of Mathematics, Fudan University, Shanghai 200433, PR Chinab School of Mathematics and Computational Science, Hunan University of Science and
Technology, Xiangtan, Hunan 411201, PR Chinac Department of Mathematics, Fudan University, Shanghai 200433, PR China
Abstract
In this paper, we develop a kind of univariate multiquadric (MQ) quasi-interpolation
and use it to solve Burgers� equation (with viscosity). At first we construct the MQ
quasi-interpolation, which possesses the properties of linear reproducing and preserving
monotonicity. Next we obtain the numerical scheme, by using the derivative of the
quasi-interpolation to approximate the spatial derivative of the dependent variable
and a low order forward difference to approximate the time derivative of the dependent
variable. Then, we verify our method for two examples with distinguishing initial value
condition. One example is tested for three Reynolds number, that is, R = 10, R = 100,
and R = 10,000. From the numerical experiments, we see that the presented method
in this paper is valid although the accuracy of the technique is not higher than Hon
and Mao�s one. Another example is used to examine the travelling of the shock. We
can improve the accuracy by selecting an appropriate shape parameter and using a
0096-3003/$ - see front matter � 2005 Elsevier Inc. All rights reserved.
doi:10.1016/j.amc.2005.02.027
q Supported by NSFC 19971017 and NOYS 10125102.* Corresponding author. Address: Institute of Mathematics, Fudan University, Shanghai 200433,
PR China.
E-mail addresses: [email protected] (R. Chen), [email protected] (Z. Wu).
R. Chen, Z. Wu / Appl. Math. Comput. 172 (2006) 472–484 473
higher accurate MQ quasi-interpolation. The advantage of the resulting scheme is that
the algorithm is very simple, so it is very easy to implement.
� 2005 Elsevier Inc. All rights reserved.
Keywords: Multiquadric quasi-interpolation; Burgers� equation; Shape parameter; Radial basis
function
1. Introduction
Since Hardy proposed in 1968, the multiquadric (MQ) which is a kind of
radial basis function (RBF) has been investigated thoroughly. Hardy [22] sum-
marized the achievement of study of MQ from 1968 to 1988 and showed that
MQ can be applied in hydrology, geodesy, photogrammetry, surveying and
mapping, geophysics and crustal movement, geology and mining, and so on.In Franke�s review paper [14], the MQ was rated as one of the best methods
among 29 scattered data interpolation schemes based on their accuracy, stabil-
ity, efficiency, memory requirement, and ease to implementation. As an inter-
polation method, the MQ always produces a minimal semi-norm error as
proven by Madych and Nelson [29]. Thanks to the investigation of Beatson,
Buhmann, Carlson, Chen, Cheng, Chui, Dyn, Fasshauer, Foley, Fornberg,
Jetter, Kansa, Larsson, Li, Madych, Micchelli, Narcowich, Powell, Schaback,
Wendland, Ward, Wright and Wu et al. for the theory of the radial basis func-tions (see [1–4,8,10,13,25,28,30–35,37,39,42,43] and so on), the usages of the
RBFs become more and more wider. Since Kansa [23,24] successfully modified
MQ for solving partial differential equation (PDE), more and more researchers
have been attracted by this meshless, scattered data approximation scheme
(see, e.g., [5,11,12,15–21,26,27,36,38,40,41]). In most of the known methods
of solving differential equations using multiquadric, one must resolve a linear
system of equation at each time step. Hon and Wu [21], Wu [41], and others
have provided some successful examples using MQ quasi-interpolation to solv-ing differential equations. In this paper, we still use MQ quasi-interpolation so
that we do not require to solve any linear system of equation that we do not
meet the question of the ill-condition of the matrix. Therefore, we can save the
computational time and decrease the numerical error.
Hon and Mao [20] developed an efficient numerical scheme for Burgers�equation (with viscosity). They apply the MQ as a spatial approximation
scheme and a low order explicit finite difference approximation to the time
derivative. The method requires to solve a linear system, by using Gaussianelimination with partial pivoting, in order to obtain the coefficients of the inter-
polation function. Then get the value of the given points at the given time by
using the interpolation function, and the interpolation function is the linear
combinations with the MQ and the linear function. The method is valid for
474 R. Chen, Z. Wu / Appl. Math. Comput. 172 (2006) 472–484
the various Reynolds number R whose scope from 0.1 to 10,000, namely, the
method, has very broad applicability. They find that the method offers better
accuracy than other numerical methods [20]. Again, the results of the method
are very close to the analytical solution obtained by Cole [9] and the accurate
solution given by Christie and Mitchell [7].
In our methods, we use the derivative of the MQ quasi-interpolation toapproximate the spatial derivative of the differential equations and employ a
first order accurate forward difference for the approach of the temporal deriv-
ative as Hon and Mao do [20]. In our methods, we follow the Wu and Scha-
back�s [43] idea to construct the MQ quasi-interpolation.
The organization of the rest of this paper is as follows: In Section 2, we con-
struct the univariate MQ quasi-interpolation respectively. In Section 3, we de-
velop the numerical techniques using MQ to solve Burgers� equation. In
Section 4, we give the numerical results for R = 10, R = 100, and R = 10,000firstly. The accuracy of the method is not higher than that of Hon and Mao�sone [20], but we can improve the accuracy by selecting the shape parameter and
employing the high-order quasi-interpolation as shown by Beatson and Dyn in
[1] or by Zhang and Wu in [44]. Then, we use the method for another initial
value condition. The results are also acceptable. So the technique is valid
too. In Section 5, we derive conclusion and give remarks for the resulting
scheme and the further work.
2. Multiquadric quasi-interpolation
Beatson and Powell [2] proposed three univariate multiquadric quasi-inter-
polations, namely, LA, LB, and LC, to approximate a function {f(x),x0 6
x 6 xm} from the space that is spanned by the multiquadrics f/jðxÞ ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiðx� xjÞ2 þ c2
q; x 2 R; j ¼ 0; . . . ;mg and linear function, where c is a positive
constant and the centers {xj: j = 0, . . .,m} being given distinct points in the
interval [x0,xm]. Afterward, Beatson and Dyn [1] studied the properties of
the W-splines, the combination of the MQs, and obtained the error estimates
for quasi-interpolation schemes involving MQ based on a finite number of cen-
ters. Wu and Schaback [43] have proposed the univariate multiquadric quasi-
interpolation LD on [a,b] and proven that the scheme is shape preserving and
convergent. In this section, we construct a kind of special MQ quasi-interpola-
tion, which generalized LD.Given that fðxj; fjÞgmj¼0, where x0 < x1 < � � � < xm, we construct the univari-
ate quasi-interpolation in the form of
f �ðxÞ ¼Xmj¼0
fjWjðxÞ; ð2:1Þ
R. Chen, Z. Wu / Appl. Math. Comput. 172 (2006) 472–484 475
where
WjðxÞ ¼/jþ1ðxÞ � /jðxÞ2ðxjþ1 � xjÞ
�/jðxÞ � /j�1ðxÞ2ðxj � xj�1Þ
; 0 6 j 6 m. ð2:2Þ
We give the definition of /j(x) in the posterior definition.
Now, for the sake of reading easily, we introduce some definitions related to
quasi-interpolation.
Definition 2.1. If the quasi-interpolation f �(x) possesses the property
f �ðxÞ � C if f 0 ¼ f1 ¼ � � � ¼ fm ¼ C; ð2:3Þ
where C is any real constant, we say that the quasi-interpolation is constant
reproducing on [x0,xm].
Definition 2.2. We say that the quasi-interpolation f �(x) possesses linear repro-
ducing property on [x0,xm], if f�(x) = px + q as fj = pxj + q, j = 0, . . .,m, for all
p; q 2 R.
Remark 2.1. It is obvious that if a quasi-interpolation f �(x) possesses linear
reproducing property on [x0,xm], then it must be constant reproducing.
Definition 2.3. If the quasi-interpolation f �(x) is monotone increasing
(decreasing) for monotone increasing (decreasing) data fj, j = 0, . . .,m, then we
say that it possesses preserving monotonicity on [x0,xm].
In this paper, we define the quasi-interpolation as follows:
Definition 2.4. For the initial data fðxj; fjÞgmj¼0, fj = f(xj), the univariate quasi-
interpolation on [x0,xm], f�(x), is defined by (2.1) and (2.2),
/mðxÞ ¼ /0ðxÞ � 2xþ xm þ x0; ð2:4Þ
/�1ðxÞ ¼ /0ðxÞ þ x0 � x�1;
/mþ1ðxÞ ¼ /mðxÞ þ xmþ1 � xm
�ð2:5Þ
and
/jðxÞ ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiðx� xjÞ2 þ k2
q; 0 6 j 6 m� 1; ð2:6Þ
where k 2 R.
We can prove the following theorem (see [6] for detail).
476 R. Chen, Z. Wu / Appl. Math. Comput. 172 (2006) 472–484
Theorem 2.1. The quasi-interpolation f �(x), defined by Definition 2.4, possesses
linear reproducing property and preserving monotonicity on [x0,xm]. Meantime,
on [x0,xm], f�(x) can be rewritten as follows:
f �ðxÞ ¼ 1
2
Xm�1
j¼1
/jþ1ðxÞ � /jðxÞxjþ1 � xj
�/jðxÞ � /j�1ðxÞ
xj � xj�1
� �fj
þ 1
21þ /1ðxÞ � /0ðxÞ
x1 � x0
� �f0 þ
1
21� /mðxÞ � /m�1ðxÞ
xm � xm�1
� �fm; ð2:7Þ
or
f �ðxÞ ¼ f0 þ fm2
þ 1
2
Xm�1
j¼0
/jðxÞ � /jþ1ðxÞxjþ1 � xj
ðfjþ1 � fjÞ; ð2:8Þ
or
f �ðxÞ ¼ 1
2
Xm�1
j¼1
fjþ1 � fjxjþ1 � xj
� fj � fj�1
xj � xj�1
� �/jðxÞ þ
f0 þ fm2
þ f1 � f02ðx1 � x0Þ
/0ðxÞ �fm � fm�1
2ðxm � xm�1Þ/mðxÞ. ð2:9Þ
Moreover, on [x0,xm], we have
ðf �ðxÞÞ0 ¼ 1
2
Xm�1
j¼0
/0jðxÞ � /0
jþ1ðxÞxjþ1 � xj
ðfjþ1 � fjÞ ð2:10Þ
and
ðf �ðxÞÞ00 ¼ 1
2
Xm�1
j¼0
/00j ðxÞ � /00
jþ1ðxÞxjþ1 � xj
ðfjþ1 � fjÞ. ð2:11Þ
Remark 2.2. We note that the formulae (2.7)–(2.11) and the linear reproduc-
ing property of the quasi-interpolation f �(x) have no relation to the definition
of /j(x), j = 0, . . .,m � 1 i.e. (2.6). In other words, all quasi-interpolation f �(x)
defined by (2.1), (2.2), (2.4) and (2.5) satisfy with (2.7)–(2.11) and possess the
linear reproducing property.
Theorem 2.2. Denote h = max16i6m{xi � xi�1}. f�(x) is the univariate multiqua-
dric quasi-interpolation defined by Definition 2.4. For k > 0, and f(x) 2C2(x0,xm), we have,
kf �ðxÞ � f ðxÞk1 6 K0Ch þ K1h2 þ K2khþ K3k
2 log h; ð2:12Þ
R. Chen, Z. Wu / Appl. Math. Comput. 172 (2006) 472–484 477
where
Ch ¼ min k;k2
h
� �; ð2:13Þ
K0, K1, K2, and K3 are the positive constants independent of h and k.
Remark 2.3. As k = 0, f �(x) changes into L(x), and now kf �(x) � f(x)k1 6
Kh2, where K is a constant which is independent of h.
3. Numerical scheme using MQ quasi-interpolation
In this section, we present the numerical scheme for solving Burgers� equa-tion (with viscosity) [20] by using the multiquadric (MQ) quasi-interpolation.
Discretizing Burgers� equation
ut þ uux ¼1
Ruxx; ð3:1Þ
in time, we get
unþ1j ¼ unj � s � unj � ðuxÞ
nj þ s
1
RðuxxÞnj . ð3:2Þ
Then, we use the derivatives of the MQ quasi-interpolation to approximate uxand uxx. To dump the dispersion of the scheme, we define a switch function
g(x, t), whose values are 0 or 1 at discrete points (xj, tn), as follows:
gnj ¼ maxf0; 1þminf0; signððuxÞnj � ðuxÞnkÞgg; ð3:3Þ
where k ¼ j� signðunj Þ. Thus, the resulting numerical scheme is
unþ1j ¼ unj � s � unj � ðuxÞ
nj � gnj þ s
1
RðuxxÞnj ; ð3:4Þ
where
ðuxÞnj ¼1
2
Xm�1
k¼0
/0kðxjÞ � /0
kþ1ðxjÞxkþ1 � xk
ðunkþ1 � unkÞ; ð3:5Þ
ðuxxÞnj ¼1
2
Xm�1
k¼0
/00kðxjÞ � /00
kþ1ðxjÞxkþ1 � xk
ðunkþ1 � unkÞ. ð3:6Þ
/j(x), j = 0, . . .,m is defined in (2.6) and (2.4). unj is the approximation of the
value of u(x, t) at point (xj, tn), tn = ns; s is time step.
478 R. Chen, Z. Wu / Appl. Math. Comput. 172 (2006) 472–484
4. Numerical examples
We test the algorithm by two examples. The first example is comparing the
computational results to that of analytic solution (for R = 10, and R = 100) or
Christie accurate solution (R = 10,000) and Hon and Mao�s technique. The ini-tial conditions are
uðx; 0Þ ¼ u0ðxÞ ¼ sin px; 0 6 x 6 1 ð4:1Þand the boundary conditions are
uð0; tÞ ¼ 0 ¼ uð1; tÞ. ð4:2ÞWe denote this scheme by MQQI scheme. For the sake of simplification, we set
hi ¼ h ¼ 1
m.
Thus, xj = jh, j = 0, . . .,m. The comparison of the results is shown in Tables
1–3. In Tables 1 and 2, we also give the relative errors with respect to the
Table 1
Comparison of results for R = 10 at t = 1
x Analytic
solution
Hon and Mao�smethod s = 10�3
MQQI
scheme s = 10�3
Relative
error · 10�2
0.10 0.0663 0.0664 0.071238 7.448
0.20 0.1312 0.1313 0.13431 2.3679
0.30 0.1928 0.1928 0.19339 0.30347
0.40 0.2480 0.2481 0.24538 1.0572
0.50 0.2919 0.2919 0.28517 2.3052
0.60 0.3161 0.3159 0.30473 3.5977
0.70 0.3081 0.3079 0.29288 4.9416
0.80 0.2537 0.2534 0.23784 6.2509
0.90 0.1461 0.1459 0.13542 7.3088
Table 2
Comparison of results for R = 100 at t = 1
x Analytic
solution
Hon and Mao�smethod s = 10�3
MQQI
scheme s = 10�3
Relative
error · 10�2
0.10 0.0754 0.0755 0.078675 4.3439
0.20 0.1506 0.1507 0.15202 0.96341
0.30 0.2257 0.2257 0.22554 0.072147
0.40 0.3003 0.3003 0.29904 0.41805
0.50 0.3744 0.3744 0.37226 0.57245
0.60 0.4478 0.4478 0.44484 0.66202
0.70 0.5203 0.5202 0.51643 0.74341
0.80 0.5915 0.5913 0.58622 0.89348
0.90 0.6600 0.6607 0.62956 4.6118
Table 3
Comparison of results for R = 10,000 at t = 1
x Christie accurate
solution
Hon and Mao�smethod s = 10�3
MQQI scheme
s = 10�3
0.056(0.05) 0.0422 0.0424 0.042395
0.111(0.11) 0.0843 0.0843 0.084337
0.167(0.16) 0.1263 0.1263 0.12619
0.222(0.22) 0.1684 0.1684 0.16796
0.278(0.27) 0.2103 0.2103 0.20957
0.333(0.33) 0.2522 0.2522 0.25129
0.389(0.38) 0.2939 0.2939 0.29281
0.444(0.44) 0.3355 0.3355 0.3342
0.500(0.50) 0.3769 0.3769 0.37544
0.556(0.55) 0.4182 0.4182 0.41468
0.611(0.61) 0.4592 0.4592 0.45729
0.667(0.66) 0.5000 0.4999 0.49783
0.722(0.72) 0.5404 0.5404 0.53806
0.778(0.77) 0.5806 0.5805 0.57794
0.833(0.83) 0.6203 0.6201 0.61739
0.889(0.88) 0.6596 0.6600 0.65635
0.944(0.94) 0.6983 0.6957 0.69475
R. Chen, Z. Wu / Appl. Math. Comput. 172 (2006) 472–484 479
analytical solutions given by Cole [9], where s = 0.001, h = 0.01, and k = 0.0072
for R = 10, k = 0.0029 for R = 100, respectively. The numbers in the parenthe-
ses of the first column, the second column, and the third column in Table 3 arequoted from Ref. [20], where the Christie accurate solution is computed by
Christie and Mitchell [7] using the Galerkin method with fully upwind cubic
functions and a particularly small value of spatial step h, for MQQI scheme,
we set R = 10,000, s = 10�3, h ¼ 172, and k = 1.43 · 10�4. By the way, we give
the results with respect to k = 10�4 in Ref. [6]. The numerical results of the
MQQI scheme, for t = 0.2, 0.4, 0.6, 0.8, and 1.0 together with the initial data,
are given in Figs. 1–3, which correspond to R = 10, R = 100, and R = 10,000,
respectively.From the tables and figures above, we can say that the scheme is feasible
although the accuracy is not higher than that of Hon and Mao�s method.
We know that, at each time step, the complexity of our techniques is
only OðmÞ. Furthermore, the implementation of the present methods is very
easy.
For another example, the boundary conditions are same as above, while the
initial value conditions are uðx; 0Þ ¼ sin 2pxþ 0.5 sin px. The numerical results
of this example are shown in Fig. 4.From Fig. 4, we believe the results are acceptable. It means that this scheme
is valid.
Fig. 1. The results of the Burgers� equation for R = 10 by using MQQI scheme, where the spatial
step h ¼ 1100, the temporal step s = 10�3, and the shape parameter k = 0.0072.
Fig. 2. The results of the Burgers� equation for R = 100 by using MQQI scheme, where the spatial
step h ¼ 1100, the temporal step s = 10�3, and the shape parameter k = 0.0029.
480 R. Chen, Z. Wu / Appl. Math. Comput. 172 (2006) 472–484
Fig. 3. The results of the Burgers� equation for R = 10,000 by using MQQI scheme, where the
spatial step h ¼ 172, the temporal step s = 10�3, and the shape parameter k = 1.43 · 10�4.
Fig. 4. The results of the Burgers� equation for R = 10,000 by using MQQI scheme, where the
spatial step h ¼ 180, the temporal step s = 10�3, the shape parameter k = 1.2 · 10�4, and the initial
value conditions are uðx; 0Þ ¼ sin 2pxþ 0.5 sin px.
R. Chen, Z. Wu / Appl. Math. Comput. 172 (2006) 472–484 481
482 R. Chen, Z. Wu / Appl. Math. Comput. 172 (2006) 472–484
5. Conclusion
From the above figures and tables, all in all, we conclude that the methods
are feasible and valid. Moreover, we can improve the accuracy by selecting the
appropriate shape parameter and using higher accurate MQ quasi-
interpolation.The techniques can be used for the non-equidistant grids, for comparison,
although we are using equidistant grids in our numerical experiments. We
see that the present techniques require to calculate
/0jðxÞ � /0
jþ1ðxÞxjþ1 � xj
;/00
j ðxÞ � /00jþ1ðxÞ
xjþ1 � xj; j ¼ 0; . . . ;m� 1
once for all.
The results have very close relation to the shape parameter k. In fact, the
choice of the shape parameter is still a pendent question.
In Tables 1–3, we see that the accuracy of the present scheme is higher as the
value of R is larger. Unfortunately, from our numerical experiment whose re-
sults are not given here, we see that the effects are very bad for R = 0.1 andR = 1. So the method must be rebuilt for R = 0.1 and R = 1.
References
[1] R.K. Beatson, N. Dyn, Multiquadric B-splines, J. Approx. Theory 87 (1996) 1–24.
[2] R.K. Beatson, M.J.D. Powell, Univariate multiquadric approximation: quasi-interpolation to
scattered data, Constr. Approx. 8 (1992) 275–288.
[3] M.D. Buhmann, C.A. Micchelli, Multiquadric interpolation improved, Comput. Math. Appl.
24 (1992) 21–25.
[4] R.E. Carlson, T.A. Foley, Interpolation of track data with radial basis methods, Comput.
Math. Appl. 24 (12) (1992) 27–34.
[5] C.S. Chen, G. Kuhn, J.C. Li, G. Mishuris, Radial basis functions for solving near singular
Poisson problems, Commun. Numer. Meth. En. 19 (2003) 333–347.
[6] R.H. Chen, Z.M. Wu, Solving hyperbolic conservation laws using multiquadric quasi-
interpolation, Numer. Method PDE, submitted for publication.
[7] I. Christie, A.R. Mitchell, Upwinding of high order Galerkin methods in conduction-
convection problems, Int. J. Numer. Methods Eng. 12 (1978) 1764–1771.
[8] C.K. Chui, J.D. Ward, K. Jetter, Cardinal interpolation with differences of tempered
functions, Comput. Math. Appl. 24 (12) (1992) 35–48.
[9] J.D. Cole, On a quasi-linear parabolic equation occurring in aerodynamics, Quart. Appl.
Math. 9 (1951) 225–236.
[10] G.E. Fasshauer, Solving differential equations with radial basis functions: Multilevel methods
and smoothing, Adv. Comput. Math. 11 (1999) 139–159.
[11] G.E. Fasshauer, Newton iteration with multiquadrics for the solution of nonlinear PDEs,
Comput. Math. Appl. 43 (2002) 423–438.
[12] A.I. Fedoseyev, M.J. Friedman, E.J. Kansa, Continuation for nonlinear elliptic partial
differential equations discretized by the multiquadric method, Int. J. Bifurcat. Chaos 10 (2000)
481–492.
R. Chen, Z. Wu / Appl. Math. Comput. 172 (2006) 472–484 483
[13] B. Fornberg, G. Wright, E. Larsson, Some observations regarding interpolants in the limit of
flat radial basis functions, Comput. Math. Appl. 47 (2004) 37–55.
[14] R. Franke, Scattered data interpolation: test of some methods, Math. Comput. 38 (1982) 181–
200.
[15] M.J. Friedman, Improved detection of bifurcations in large nonlinear system via the
continuation of invariant subspace algorithm, Int. J. Bifurcat. Chaos 11 (2000) 2277–2285.
[16] M.A. Golberg, C.S. Chen, Improved multiquadric approximation for partial differential
equations, Eng. Anal. Bound. Elem. 18 (1996) 9–17.
[17] Y.C. Hon, Multiquadric collocation method with adaptive technique for problem with
boundary layer, Int. J. Appl. Sci. Comput. 6 (1999) 173–184.
[18] Y.C. Hon, M.W. Lu, W.M. Xue, Y.M. Zhu, Multiquadric method for the numerical solution
of a biphasic model, Appl. Math. Comput. 88 (1997) 153–175.
[19] Y.C. Hon, X.Z. Mao, A multiquadric interpolation method for solving initial value problems,
J. Sci. Comput. 12 (1997) 51–55.
[20] Y.C. Hon, X.Z. Mao, An efficient numerical scheme for Burgers� equation, Appl. Math.
Comput. 95 (1998) 37–50.
[21] Y.C. Hon, Z.M. Wu, A quasi-interpolation method for solving ordinary differential equations,
Int. J. Numer. Methods Eng. 48 (2000) 1187–1197.
[22] R.L. Hardy, Theory and applications of the multiquadric–biharmonic method, 20 years of
discovery 1968–1988, Comput. Math. Appl. 19 (1990) 163–208.
[23] E.J. Kansa, Multiquadrics—a scattered data approximation scheme with applications to
computational fluid dynamics I, Comput. Math. Appl. 19 (1990) 127–145.
[24] E.J. Kansa, Multiquadrics—a scattered data approximation scheme with applications to
computational fluid dynamics II, Comput. Math. Appl. 19 (1990) 147–161.
[25] E.J. Kansa, R.E. Carlson, Improved accuracy of multiquadric interpolation using variable
shape parameters, Comput. Math. Appl. 24 (12) (1992) 99–120.
[26] E.J. Kansa, Y.C. Hon, Circumventing the ill-conditioning problem with multiquadric radial
basis functions: applications to ellipitic partial differential equations, Comput. Math. Appl. 39
(2000) 123–137.
[27] J.C. Li, C.S. Chen, Some observations on unsymmetric radial basis function collocation
methods for convection-diffusion problems, Int. J. Numer. Methods Eng. 57 (2003) 1085–1094.
[28] J.C. Li, A.H.D. Cheng, C.S. Chen, A comparison of efficiency and error convergence of
multiquadric collocation method and finite element method, Eng. Anal. Bound. Elem. 27
(2003) 251–257.
[29] W.R. Madych, S.A. Nelson, Multivariate interpolation and conditionally positive definite
functions I, J. Approx. Theory Appl. 4 (1988) 77–89.
[30] W.R. Madych, Miscellaneous error bounds for multiquadric and related interpolators,
Comput. Math. Appl. 24 (12) (1992) 121–138.
[31] C.A. Micchelli, Interpolations of scattered data: distance matrices and conditionally positive
definite functions, Constr. Approx. 2 (1986) 11–22.
[32] F.J. Narcowich, J.D. Ward, Norm estimates for the inverses of a general class of scattered-
data radial-function interpolation matrices, J. Approx. Theory 69 (1992) 84–109.
[33] F.J. Narcowich, R. Schaback, J.D. Ward, Multilevel interpolation and approximation, Appl.
Comput. Harmonic Anal. 7 (1999) 243–261.
[34] F.J. Narcowich, J.D. Ward, H. Wendland, Refined error estimates for radial basis function
interpolation, Constr. Approx. 19 (2003) 541–564.
[35] R. Schaback, Z.M. Wu, Operators on radial basis functions, J. Comput. Appl. Math. 73
(1996) 257–270.
[36] M. Sharan, E.J. Kansa, S. Gupta, Application of the multiquadric method for numerical
solution of elliptic partial differential equations, Appl. Math. Comput. 84 (1997) 275–
302.
484 R. Chen, Z. Wu / Appl. Math. Comput. 172 (2006) 472–484
[37] H. Wendland, Piecewise polynomial positive definite and compactly supported radial basis
functions of minimal degree, Adv. Comput. Math. 4 (1995) 389–396.
[38] A.S.M. Wong, Y.C. Hon, T.S. Li, S.L. Chung, E.J. Kansa, Multizone decomposition for
simulation of time-dependent problem using the multiquadric scheme, Comput. Math. Appl.
37 (1999) 23–43.
[39] Z.M. Wu, Compactly supported radial functions and the Strang-Fix condition, Appl. Math.
Comput. 84 (1997) 115–124.
[40] Z.M. Wu, Solving differential equation with radial basis function, in: Advances in
Computational Mathematics, Lecture Note in Pure and Applied Mathematics, vol. 202,
Dekker, 1999, pp. 537–544.
[41] Z.M. Wu, Dynamically knots setting in meshless method for solving time dependent
propagations equation, Comput. Methods Appl. Mech. Eng. 193 (2004) 1221–1229.
[42] Z.M. Wu, R. Schaback, Local error estimates for radial basis function interpolation of
scattered data, IMA J. Numer. Anal. 13 (1993) 13–27.
[43] Z.M. Wu, R. Schaback, Shape preserving properties and convergence of univariate
multiquadric quasi-interpolation, ACTA Math. Appl. Sinica 10 (1994) 441–446.
[44] W.X. Zhang, Z.M. Wu, Some shape-preserving quasi-interpolations to non-uniformly
distributed data constructed by MQ-B-splines, preprint Fudan University, 2003.