International Journal of Computer Mathematics

Vol. 83, No. 1, January 2006, 143157

A three-step wavelet Galerkin method for parabolic and
hyperbolic partial differential equations
B. V. RATHISH KUMAR* and MANI MEHRA
Department of Mathematics, Indian Institute of Technology, Kanpur, India
(Received 11 November 2004; in nal form 2 February 2005)
Athree-stepwavelet GalerkinmethodbasedonTaylor series expansionintime is proposed. The scheme
is third-order accurate in time and O(2
jp
) accurate in space. Unlike TaylorGalerkin methods, the
present scheme does not contain any new higher-order derivatives which makes it suitable for solving
non-linear problems. The compactly supported orthogonal wavelet bases D6 developed by Daubechies
are used in the Galerkin scheme. The proposed scheme is tested with both parabolic and hyperbolic
partial differential equations. The numerical results indicate the versatility and effectiveness of the
proposed scheme.
Keywords: Three-step method; Wavelets; Parabolic equation; Hyperbolic equation
C.R. Category: G.1.8
1. Introduction
Wavelet analysis is a new numerical concept which allows one to represent a function in terms
of a set of basis functions, called wavelets, which are localized in space. As we have noted
earlier, spectral bases are innitely differentiable but have global support. On the other hand,
basis functions used in nite-element methods have small compact support but poor continu-
ity properties. As a result, spectral methods have good spectral localization but poor spatial
localization, while nite element methods have good spatial localization, but poor spectral
localization. Wavelet bases appear to combine the advantages of both spectral and nite ele-
ment bases. We can expect numerical methods based on wavelet bases to be able to attain
good spatial and spectral resolution.
Recently much effort has been put into developing schemes based on the properties of
orthonormal wavelet bases introduced in the 1980s by Stromberg and Meyer. These bases are
formed by the dilation and translation of a single function (x),
j,k
(x) = 2
j/2
(2
j
x k),
where j, k Z. Then for certain (x), the sequence of functions (
j,k
(x))
(j,k)
Z
2
forms
an orthonormal basis in L
2
(R). In wavelet applications to the solution of partial differential
equations (PDEs) the most frequently used wavelets are those with compact support introduced
*Corresponding author. Email: bvrk@iitk.ac.in
International Journal of Computer Mathematics
ISSN 0020-7160 print/ISSN 1029-0265 online 2006 Taylor & Francis
http://www.tandf.co.uk/journals
DOI: 10.1080/00207160500112985
144 B. V. Rathish Kumar and M. Mehra
by Daubechies [1]. Exploration of the usage of Daubechies wavelets to solve PDEs has been
undertaken by a number of investigators [25] . In another useful approach wavelets are used to
drive the adaptive nite-difference method, as advocated by Jameson [6], and this application
has been established in a series of papers [79].
In the conventional numerical approach to transient problems the accuracy gained by using
the high-order spatial discretization is partially lost because of the use of low-order time
discretization schemes. Here, spatial approximation usually precedes temporal discretization.
However, the reversed order of discretization can lead to better time accurate schemes with
improved stability properties. The fundamental idea behind the TaylorGalerkin approach [10]
is the substitution of space derivatives for the time derivatives in the Taylor series, as used in the
derivation of the LaxWendroff method [11], the only modication being that the procedure
is carried out to third order. However, its applications are mainly to hyperbolic problems and
some convection diffusion equations because too many terms are introduced in the third-order
time derivative term, especially for non-linear multidimensional equations, and treatment of
the boundary integrations arising from high-order time derivative terms is too complicated.
A three-step nite-element method based on a Taylor series expansion in time is proposed
in [12]. This scheme involves neither complicated expression nor higher-order derivatives
as in the wavelet TaylorGalerkin method [13] and the wavelet multilayer TaylorGalerkin
method [14]. In [13] we investigated the feasibility of applying the new wavelet Taylor
Galerkin to an important application problem of the Kortewegde Vires (KdV) equation. A
combination of such a time marching scheme with wavelet approximation in space can lead
to simple higher-order space and time accurate numerical methods. In this paper we develop
a three-step wavelet Galerkin method (T-WGM) for a class of linear problems such as the
heat equation, the convective transport problem and the non-linear Burgers equation in one
dimension and the heat equation in two dimensions. This scheme retains the good accuracy and
stability properties of the TaylorGalerkin method without needing to calculate higher-order
derivative terms.
The outline of paper is as follows. Insection2we summarize some basics of wavelet analysis.
In section 3 we consider the application of T-WGM to the heat equation, the convection
equation and the non-linear Burgers equation. In section 4 we study stability of the algorithm
in the context of ODEs. Numerical results are given in section 5. In section 6 we extend
T-WGM to the heat equation in two dimensions. Finally, in section 7 we draw a number of
conclusions based on our results.
2. Wavelet preliminaries
2.1 Compactly supported wavelets
Daubechies [1] has dened the class of compactly supported wavelets. Briey, let be a
solution of the scaling relation
(x) =

k
a
k
(2x k)
where the a
k
are a collection of coefcients that categorize the specic wavelet basis. The
expression is called the scaling function. The associated wavelet function is dened by
the equation
(x) =

k
(1)
k
a
1k
(2x k).
Three-step wavelet method 145
Normalization
_
dx = 1 of the scaling function leads to the condition

k
a
k
= 2.
The translates of are required to be orthonormal, i.e.
_
(x k)(x m)dx =
k,m
.
The scaling relation implies the condition
N1

k=0
a
k
a
k2m
=
0,m
where N is the order of wavelet. For coefcients verifying the above two conditions, the
functions consisting of translates and dilations of the wavelet function (2
j
x k) form a
complete orthogonal basis for square integrable functions on the real line L
2
(R).
If only a nite number of the a
k
are non-zero, will have compact support. Since
_
(x)(x m)dx =

k
(1)
k
a
1k
a
k2m
= 0
the translates of the scaling function and wavelet dene orthogonal subspaces
V
j
= {2
j/2
(2
j
x k); m = . . . , 1, 0, 1, . . . }
W
j
= {2
j/2
(2
j
x k); m = . . . , 1, 0, 1, . . . }.
The relation
V
j+1
= V
j
+W
j
implies the Mallat transform [1]
V
0
V
1
V
j+1
V
j+1
= V
0
W
0
W
1
W
j
.
Smooth scaling functions arise as a consequence of the degree of approximation of the trans-
lates. The result that the polynomials 1, x, . . . , x
p1
are expressed as linear combinations of
the translates of (x k) is implied by the conditions

k
(1)
k
k
m
a
k
= 0
for m = 0, 1, . . . , p 1. The following are equivalent results:
{1, x, . . . , x
p1
} are linear combinations of (x k)
f

c
j
k
(2
j
x k) C2
jp
f
p
, where c
j
k
=
_
f (x)(2
j
x k)dx

_
x
m
(x)dx = 0 for m = 0, 1, . . . , p 1

_
f (x)(2
j
x)dx c2
jp
.
For the Daubechies scaling/wavelet function DN, we have p = N/2 where N is the order
of the wavelet. Figure 1 shows an example of a compactly supported scaling function and
its associated fundamental wavelet function. For arbitrarily large even N there is Daubechies
146 B. V. Rathish Kumar and M. Mehra
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
1.5
1
0.5
0
0.5
1
1.5
2
Scaling function
Wavelet function
Figure 1. Daubechies scaling and wavelet functions for N = 6 with support on [0, 5].
example of a fundamental scaling function dening a wavelet family with support in the
interval [0, N 1] [1]. For any j, l Z we dene the 1-periodic scaling function

j,l
(x) =

n=

j,l
(x +n) = 2
j/2

n=
(2
j
(x +n) l), x R
and the 1-periodic wavelet

j,l
(x) =

n=

j,l
(x +n) = 2
j/2

n=
(2
j
(x +n) l), x R.
For a PDE of the form
F(u, u
t
, . . . , u
x
, u
xx
, . . . ) = 0 (1)
the wavelet approximation is of the form
u
j
(x, t ) =
2
j
1

k,l=0
c
j,k
(t )

j,k
(x) (2)
where c
j,k
is an unknown coefcient of the scaling function expansion. Since we assume
periodic boundary conditions, there is a periodic wrap around in (x, y) and we let the period
scale with the number of terms in the expansion. To determine the coefcient of expansion (2)
we substitute (2) into equation (1) and again project the resulting expression onto subspace V
j
.
Three-step wavelet method 147
The projection requires c
j,k
to satisfy the equation
_

j,m
(x)F(u
j
, u
j
t
, u
j
x
, u
j
xx
, . . . )dx = 0.
To evaluate this expression we must know the coefcients of the form
(l
1
, l
2
, . . . , l
n
, d
1
, d
2
, . . . , d
n
) =
d
1
d
2
d
n
l
1
l
2
l
n
=
_

i=1

d
i
l
i
(y)dy =
_

i=1

d
i
l
i
(y)dy.
Since the scaling function used to dene compact wavelets has a limited number of deriva-
tives, the numerical evaluation of these expressions is often unstable or inaccurate. A special
algorithm to evaluate the connection coefcients has been devised by Latto et al. [15].
2.2 Multivariate wavelets
The simplest way to obtain multivariate wavelets is to employ anisotropic or isotropic tensor
products.
MRA-d Here, the multivariate wavelets are dened by

j,l
(x) :=
j
1
,l
1
(x
1
), . . . ,
j
d
,l
d
(x
d
), j := (j
1
, . . . , j
d
) x, l analogous.
MRA Here, anisotropy is avoided. The scaling functions are simply the tensor products of the
univariate scaling functions. A two-dimensional MRA can be constructed from the following
decomposition:
V
j
= V
j
V
j
= (V
j1
W
j1
) (V
j1
W
j1
)
= (W
j1
W
j1
) (W
j1
V
j1
) (V
j1
W
j1
) V
j1
V
j1
=W
j1
V
j1
.
Then we have V
J
=W
J1
W
0
V
0
and the wavelet basis is given by
{
j,k

j,l
,
j,k

j,l
,
j,k

j,l
}
k,lZ,0jJ1
{
0,k

0,l
}
k,lZ
.
We have used this MRA approach in our two dimensional problem.
3. Three-step wavelet Galerkin method
Before introducing the three-step wavelet Galerkin method (T-WGM), it is necessary to give a
brief statement of the two-step LaxWendroff wavelet Galerkin method (L-WGM). Consider
the equation
u
t
= Lu +Nf (u) (3)
148 B. V. Rathish Kumar and M. Mehra
with the initial condition
u(x, 0) = u
0
(x), 0 x 1 (4)
and with suitable boundary conditions. We explicitly separate equation (3) into a linear part Lu
and non-linear part Nf (u), where the operations Land N are constant-coefcient differential
operation that do not depend upon time t . The function f (u) is non-linear. Performing a Taylor
series expansion in time, we have
u(t +t ) = u(t ) +t
u(t )
t
+
t
2
2

2
u(t )
t
2
+
t
3
6

3
u(t )
t
3
+O(t
4
). (5)
By approximating equation (5) to second-order accuracy, the formulation of the two-step
method can be derived as follows:
u
_
t +
t
2
_
= u(t ) +
t
2
u(t )
t
u(t +t ) = u(t ) +t
u(t +(t /2))
t
.
(6)
Spatial discretization of equations (6) can be performed using the WGM. We will call this
method L-WGM.
Now we introduce the three-step wavelet Galerkin method. By approximating equation (5)
up to third-order accuracy, the formulation of the three-step method can be written as follows:
u
_
t +
t
3
_
= u(t ) +
t
3
u(t )
t
u
_
t +
t
2
_
= u(t ) +
t
2
u(t +(t /3))
t
u(t +t ) = u(t ) +t
u(t +(t /2))
t
.
(7)
Spatial discretization of equations (7) can be performed using the WGM. In contrast with the
TaylorGalerkinmethod, theT-WGMdoes not containanynewhigher-order spatial derivatives
and thus can easily be used to solve non-linear multidimensional ows.
3.1 T-WGM for the heat equation
For the heat equation L = (
2
/
2
x
) and therefore
u
t
= u
xx
+f (x) (8)
where is a positive constant. Let us rst keep the spatial variable x continuous and discretize
the time by the three-step method:
u
_
t +
t
3
_
= u(t ) +
t
3
_
u
n
xx
+f (x)
_
u
_
t +
t
2
_
= u(t ) +
t
2
_
u
n+1/3
xx
+f (x)
_
u(t +t ) = u(t ) +t
_
u
n+1/2
xx
+f (x)
_
.
(9)
Three-step wavelet method 149
In equation (9) space variable is continuous. Further we are discretizing variable x by wavelet
Galerkin method (WEM). The Galerkin discretization scheme gives
d
n+1/3
u
= d
n
u
+
t
3
_
D
(2)
d
n
u
+d
f
_
d
n+1/2
u
= d
n
u
+
t
2
_
D
(2)
d
n+1/3
u
+d
f
_
d
n+1
u
= d
n
u
+t
_
D
(2)
d
n+1/2
u
+d
f
_
(10)
where d
u
denotes the vector of the scaling function coefcients corresponding to u and d
f
denotes the vector of the scaling function coefcients corresponding to f . We will refer to
the matrix D
(d)
as the differentiation matrix of order d. This matrix is derived in [16]. If the
function to be differentiated is periodic with period L, then we have D
(d)
1
= D
(d)
/L
d
.
3.2 T-WGM for the convection equation
For the convection equation L = a(/x) and therefore
u
t
= au
x
(11)
where a is a positive constant. Then the time discretization by the three-step method is
u
_
t +
t
3
_
= u(t ) +
t
3
_
u
n
x
_
u
_
t +
t
2
_
= u(t ) +
t
2
_
u
n+1/3
x
_
u(t +t ) = u(t ) +t
_
u
n+1/2
x
_
.
(12)
Spatial discretization by WGM gives
d
n+1/3
u
= d
n
u
+
t
3
_
D
(1)
1
d
n
u
_
d
n+1/2
u
= d
n
u
+
t
2
_
D
(1)
1
d
n+1/3
u
_
d
n+1
u
= d
n
u
+t
_
D
(1)
1
d
n+1/2
u
_
.
(13)
3.3 T-WGM for the Burgers equation
Here L = u(/x) and therefore
u
t
+uu
x
= u
xx
(14)
Then time discretization by the three-step method is:
u
_
t +
t
3
_
= u(t ) +
t
3
_
u
n
u
n
x
+u
n
xx
_
u
_
t +
t
2
_
= u(t ) +
t
2
_
u
n+1/3
u
n+1/3
x
+u
n+1/3
xx
_
u(t +t ) = u(t ) +t
_
u
n+1/2
u
n+1/2
x
+u
n+1/2
xx
_
.
(15)
150 B. V. Rathish Kumar and M. Mehra
Spatial discretization by WGM gives
d
n+1/3
u
= d
n
u
+
t
3
_
2
3j/2

k

m
(c
j
k
)
n
(c
j
m
)
n

001
lkm
+(2
j
)
2

k
(c
j
k
)
n

02
lk
_
d
n+1/2
u
= d
n
u
+
t
2
_
2
3j/2

k

m
(c
j
k
)
n+1/3
(c
j
m
)
n+1/3

001
lkm
+(2
j
)
2

k
(c
j
k
)
n+1/3

02
lk
_
d
n+1
u
= d
n
u
+t
_
2
3j/2

k

m
(c
j
k
)
n+1/2
(c
j
m
)
n+1/2

001
lkm
+(2
j
)
2

k
(c
j
k
)
n+1/2

02
lk
_
.
(16)
4. Theoretical stability of the linearized schemes
We use the concept of asymptotic stability of a numerical method as dened in [17] for a
discrete problem of the form
dU
dt
= LU
where L is assumed to be a diagonalizable matrix. The most crucial property of L is its
spectrum, for this will determine the stability of time discretization. If the spatial discretization
is presumed to be xed, then we use the term stability in its ODE context.
DEFINITION The region of absolute stability of a numerical method is dened for the scalar
model problem
dU
dt
= U
to be the set of all t such that ||U
n
|| is bounded as t .
Finally we say that a numerical method is asymptotically stable for a particular problem
if, for sufciently small t > 0, the product of t and every eigenvalue of L lies within the
region of absolute stability.
5. Results of numerical experiments
In this section we solve PDEs with periodic boundary conditions and initial condition u
0
(x). A
heat equation, a linear convection equation and a non-linear Burgers equation are considered
in one dimension. All the results are presented using Daubechies D6 scaling function. The
error produced by the numerical schemes was measured against the values of the analytical
solution u
e
by the L

norm calculated as
u
L

= max
k=0,1,...,2
j
1

u
_
k
2
j
_

.
Three-step wavelet method 151
Table 1. Accuracy of results given by the L

norm.
j t LaxWendroff WGM L-WGM T-WGM
4 0.0005 (0.01) 3.178 10
6
3.138 10
7
2.2884 10
11
1.6209 10
11
6 0.0005 (0.01) 3.142 10
6
3.138 10
7
6.5721 10
12
1.0299 10
13
7 0.0005 (0.01) 3.142 10
6
3.138 10
7
6.5721 10
12
1.0299 10
13
5.1 Heat equation
We have tested T-WGM on a heat equation with u
0
(x) = 0 and f (x) = sin(2x). Errors in
the L

norm of the solution obtained by LaxWendroff, WGM using Euler time stepping,
L-WGM and T-WGM are compared in table 1. In the table the notation x(y) means that the
target time y is reached by time marching with a step size of x.
5.2 Convection equation
The accuracy of the proposed T-WGM for hyperbolic problems has been veried numerically
on the classical test problemof convection with a Gaussian prole. We assume that the solution
is periodic with some large period, say 4. The solutions obtained by WGM using Euler time
stepping, L-WGM and T-WGM are compared in table 2. For stability analysis of L-WGM,
L
j
is the matrix dened by
L
j
= aD
(1)
+
a
2
t
2
D
(2)
,
and for T-WGM
L
j
= aD
(1)
+
a
2
t
2
D
(2)
+
a
3
t
2
6
D
(3)
.
For j = 4, we have found by stability analysis that t 0.00835. Regions of absolute stability
for T-WGM are plotted in gure 2.
5.3 Burgers equation
The analytical solution to equation (14) with the initial condition u
0
(x) after using the Cole
Hopf transformation is given by
u(x, t ) =
_

(x )/t exp
_
(x )
2
/4t
_
exp
_
(2)
1
_

0
u
o
()d
_
d
_

exp
_
(x )
2
/4t
_
exp
_
(2)
1
_

0
u
0
()d
_
d
. (17)
We have integrated equation (17) numerically using GaussHermite quadrature in order to
determine the accuracy of our method.
Table 2. Accuracy of results given by the L

norm.
j t WGM L-WGM T-WGM
4 0.01 (0.5) 0.0861 0.0014 0.0012
6 0.001 (0.5) 0.0021 2.042 10
5
3.94 10
7
6 0.0001 (0.5) 0.0002 3.7881 10
7
3.70 10
7
152 B. V. Rathish Kumar and M. Mehra
Figure 2. Absolute stability region for T-WGMand the product of t and the eigenvalues of L
j
, where t = 0.001:
(a) j = 4; (b) j = 6.
Periodic boundary conditions are imposed together with a full sine wave initial condition
[u
0
(x) = sin(2x), x (0, 1)]. The resultant solution is a stationary wave with a steep front
built up at the midpoint of the domain. Figure 3 shows the exact solution at various times for
= 10
2
/.
Figure 4 shows the solution for the scales 4 j 7, with the value = 10
2
/ for the
viscosity. As time evolves, a shock develops and the solution presents Gibbs-like oscillations
for coarse scales. However, the oscillations are conned to the neighbourhood of the shock.
Apart from the shock area, the solution is correctly represented. This ability of wavelets
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
Figure 3. Exact solution of Burgers equation plotted at time intervals of 0.1 between t = 0 and t = 1.2.
Three-step wavelet method 153
Figure 4. Numerical solution of Burgers equation at various times and scales. The time step is t = 10
4
.
to conne oscillations only in the vicinity of the shock has already been observed for the
Galerkin method [18, 19]. It is a consequence of the good localization of the basis functions.
In the Fourier method, for instance, where the basis functions have non-compact support,
it is well known that oscillations spread through the whole domain unless the number of
points considered is sufcient to resolve correctly the sharp gradient in the shock layer. The
oscillations occur because the number of points in the shock layer, which is very thin, is
not sufcient for correct resolution of the large gradients which occur. In order to improve
representation of the solution in the area of shocks it is necessary to add more points. One
way of doing this is to decrease the resolution. As the scale becomes smaller, the shock is
progressivelyresolvedas canbe seeningure 4. Withscale j = 7(i.e. 128points), for instance,
the numerical solution is smooth almost everywhere except at the points immediately before
and after the shocks.
The L

errors obtained using L-WGM and T-WGM are compared in table 3.

For the stability analysis we consider the linearized Burgers equation
u
t
=

2
u
x
2

u
x
where the linearization coefcient stands for the value of u. Because at the initial time
u
0
(x) = sin(2x) and because the amplitude of u decreases with time, we assume throughout
that || 1. The region of absolute stability for this scheme is plotted in gure 5.
154 B. V. Rathish Kumar and M. Mehra
Table 3. Accuracy of results given by the L

norm.
j t Time t L-WGM T-WGM
4 10
3
0.02 0.0112 0.0112
0.14 0.1557 0.1557
0.26 0.6828 0.6828
0.38 0.6953 0.6953
5 10
3
0.02 0.0046 0.0046
0.14 0.0476 0.0475
0.26 0.4724 0.4725
0.38 0.3674 0.3674
6 10
3
0.02 0.0016 0.0016
0.14 0.0114 0.0114
0.26 0.2120 0.2120
0.38 0.1546 0.1546
7 10
3
0.02 0.0005 0.0005
0.14 0.0049 0.0050
0.26 0.0500 0.0500
0.38 0.0418 0.0418
Figure 5. Absolute stability region for (a) forward Euler and the product of t and the eigenvalues of L
7
and (b)
the product of t and the eigenvalues of L
7
, for D = 6, where t = 10
3
, = 1 and = 10
2
/.
6. Extension to multidimensional problems
The schemes developed in the preceding sections for a one-dimensional problems have demon-
strated the value of T-WGM compared with WGM and L-WGM. However, to be of practical
interest T-WGM should also be applicable to multidimensional problems. To show that this is
indeed the case, we consider the problem of heat conduction and the standard test problems
of hill translation in two dimensions.
Problem 1 Heat conduction problem
u
t
= F(x, y) (18)
Three-step wavelet method 155
where
F(x, y) = u +f (x, y)
=

2
x
2
+

2
y
2
is a two-dimensional Laplacian operator and u is an unknown scalar function dened on
a bounded square domain with a periodic boundary condition. Time discretization by the
three-step method is given by
u
_
t +
t
3
_
= u(t ) +
t
3
_
u
n
xx
+u
n
yy
+f (x, y)
_
u
_
t +
t
2
_
= u(t ) +
t
2
_
u
n+1/3
xx
+u
n+1/3
yy
+f (x, y)
_
u(t +t ) = u(t ) +t
_
u
n+1/2
xx
+u
n+1/2
yy
+f (x, y)
_
.
(19)
Spatial discretization of equations (19) is performed by WGM. The Galerkin discretizaton
scheme gives
(c
j
p,q
)
n+1/3
= (c
j
p,q
)
n
+
(2
j
)
2
t
3
_

k
(c
j
k,q
)
n

02
pk
+

l
(c
j
p,l
)
n

02
ql
+f
j
p,q
_
(c
j
p,q
)
n+1/2
= (c
j
p,q
)
n
+
(2
j
)
2
t
2
_

k
(c
j
k,q
)
n+1/3

02
pk
+

l
(c
j
p,l
)
n+1/3

02
ql
+f
j
p,q
_
(20)
(c
j
p,q
)
n+1
= (c
j
p,q
)
n
+(2
j
)
2
t
_

k
(c
j
k,q
)
n+1/2

02
pk
+

l
(c
j
p,l
)
n+1/2

02
ql
+f
j
p,q
_
where c
j
k,l
is an unknown coefcient of scaling function expansion, f
j
p,q
is the scaling func-
tion coefcient of wavelet expansion corresponding to f in two dimensions and p, q =
0, 1, . . . , 2
j
1. Let us consider equation (18) on a doubly periodic square, with vanishing
(u(x, y, 0) = 0) initial condition and the following forcing term:
f (x, y) = sin(2x) cos(2y).
Then the exact analytical solution is given by
u(x, y, t ) =
1
8
2

_
1 e
8
2
t
_
sin(2x) cos(2y).
Figure 6(a) illustrates the exact solution and gure 6(b) shows the wavelet solution using
Daubechies D6 scaling function on a mesh with n = 2
5
unknown in each direction at time
t = 0.005.
Problem 2 Consider a Gaussian hill translating with a uniform velocity a and spreading
isotropically with a diffusivity :
u
t
= a.u +
2
u (21)
156 B. V. Rathish Kumar and M. Mehra
Figure 6. (a) Analytical solution. (b) Wavelet solution using the three-step method.
Figure 7. (a) Initial distribution of the hill; (b) solution at t = 0.5 using T-WGM.
Time discretization will be same as for the one-dimensional case in problem1 for the T-WGM
scheme. The initial distribution is given in gure 7(a) and the equations are integrated until
time t = 0.5 is reached.
7. Conclusion
In the three-step wavelet Galerkin method the precedence of time discretization over space dis-
cretization in conjunction with wavelet bases for expressing spatial terms renders the proposed
schemes robust and makes them space and time accurate. The three-step wavelet Galerkin
method retains the third-order accuracy and stability properties of the TaylorGalerkin
method. Since no new higher-order derivative term occurs in the numerical formulations,
the present method is suitable for non-linear problems. Further, the method can be directly
extended to three-dimensional problems. The numerical results show that the present method
is computationally efcient.
References
[1] Daubechies, I., 1988, Orthonormal basis of compactly supported wavelets. Communications on Pure andApplied
Mathematics, 41, 906966.
[2] Beylkin, G., Coifman R. and Rokhlin, V., 1991, Fast wavelet transforms and numerical algorithms. Communi-
cations on Pure and Applied Mathematics, 44, 141183.
Three-step wavelet method 157
[3] Glowinski, R., Lawton, W., Ravachol, M. and Tenenbaum, E., 1990, Wavelet solutions of linear and non-linear
elliptic, parabolic and hyperbolic problems in ID. In: R. Glowinski (Ed.) Computing Methods inApplied Sciences
and Engineering, pp. 55120 (Philadelphia, PA: SIAM).
[4] Latto, A. and Tenenbaum, E., 1990, Compactly supported wavelets and the numerical solution of Burgers
equation. Comptes Rendus, French Academy of Science, Series I, 311, 903909.
[5] Qian, S. and Weiss, J., 1992, Wavelets and the numerical solution of partial differential equations. Journal of
Computational Physics, 106, 155175.
[6] Jameson, L., 1998, A wavelet-optimized very high order adaptive grid and order numerical method. SIAM
Journal on Scientic Computing, 19, 19802013.
[7] Holmstrom, M., 1999, Solving hyperbolic PDEs using interpolating wavelets. SIAM Journal on Scientic
Computing, 21, 405420.
[8] Holmstrom, M., and Walden, J., 1998, Adaptive wavelets methods for hyperbolic PDEs. Journal of Scientic
Computing, 13, 1949.
[9] Fatkulin, I. andHesthaven, J.S., 2001, Adaptive high-order nite-difference methodfor nonlinear wave problems.
Journal of Scientic Computing, 16, 4767.
[10] Donea, J., 1984, A TaylorGalerkin method for convective transport problems. International Journal for
Numerical Methods in Engineering, 20, 101119.
[11] Lax, P.D. and Wendrof, B., 1960, Systems of conservation laws. Communications on Pure and Applied
Mathematics, 13, 217.
[12] Jiang, C.B. and Kawahara, M., 1993, The analysis of unsteady incompressible ows by a three-step nite element
method. International Journal for Numerical Methods in Fluids, 16, 793811.
[13] Rathish Kumar, B.V. and Mehra, M., 2004, Time accurate solution of Kortewegde Vries equation using wavelet
Galerkin method. Applied Mathematics and Computation, 162, 447460.
[14] Rathish Kumar, B.V. and Mehra, M., 2005, Wavelet multilayer Taylor Galerkin schemes for hyperbolic and
parabolic problems. Accepted for publication in Applied Mathematics and Computation. Available online at:
http://authors.elsevier.com/sd/article/S0096300304004394 (accessed 4 May 2005).
[15] Latto, A., Resnikoff, H.L. and Tenenbaum, E., 1991, The evaluation of connection coefcients of compactly
supported wavelets. In: Proceedings of the FrenchUSAWorkshop onWavelets and Turbulence, Princeton, 1991
(NewYork: Springer-Verlag).
[16] Nilsen, O.M., 1998, Wavelets in scientic computing. PhD thesis, Technical University of Denmark, Lyngby.
[17] Canute, C., Hussaini, M.Y., Quarteroni, A. and Zang, T.A., 1988, Spectral Methods in Fluid Dynamics (New
York: Springer-Verlag)
[18] Liandrat, J. and Tchamitchian, Ph., 1990, Resolution of the 1-D regularised Burgers equation using spatial
wavelet approximation algorithm and numerical results. Technical Report 90-83. ICASE, NASA Langley
Research Center.
[19] Avudainayagam, A. and Vani, C., 1999, Wavelet-Galerkin solutions of quasilinear hyperbolic conservation
equations. Communications in Numerical Methods in Engineering, 15, 589601.