research article the higher accuracy fourth-order iade...
TRANSCRIPT
Hindawi Publishing CorporationJournal of Applied MathematicsVolume 2013 Article ID 236548 13 pageshttpdxdoiorg1011552013236548
Research ArticleThe Higher Accuracy Fourth-Order IADE Algorithm
N Abu Mansor1 A K Zulkifle1 N Alias2 M K Hasan3 and M J N Boyce1
1 College of Engineering Universiti Tenaga Nasional Jalan Ikram-UNITEN 43000 Kajang Selangor Malaysia2 Ibnu Sina Institute of Fundamental Science Studies Universiti Teknologi Malaysia 81310 Skudai Johor Malaysia3 Faculty of Information Science and Technology Universiti Kebangsaan Malaysia 43600 Bangi Selangor Malaysia
Correspondence should be addressed to N Abu Mansor norelizaunitenedumy
Received 21 May 2013 Accepted 26 July 2013
Academic Editor Juan Manuel Pena
Copyright copy 2013 N Abu Mansor et alThis is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited
This study develops the novel fourth-order iterative alternating decomposition explicit (IADE)method ofMitchell and Fairweather(IADEMF4) algorithm for the solution of the one-dimensional linear heat equationwithDirichlet boundary conditionsThehigher-order finite difference scheme is developed by representing the spatial derivative in the heat equation with the fourth-order finitedifference Crank-Nicolson approximationThis leads to the formation of pentadiagonal matrices in the systems of linear equationsThe algorithm also employs the higher accuracy of theMitchell and Fairweather variant Despite the schemersquos higher computationalcomplexity experimental results show that it is not only capable of enhancing the accuracy of the original correspondingmethod ofsecond-order (IADEMF2) but its solutions are also in very much agreement with the exact solutions Besides it is unconditionallystable and has proven to be convergent The IADEMF4 is also found to be more accurate more efficient and has better rate ofconvergence than the benchmarked fourth-order classical iterative methods namely the Jacobi (JAC4) the Gauss-Seidel (GS4)and the successive over-relaxation (SOR4) methods
1 Introduction
Numerical methods with accuracy of the order 119874(Δℎ)119899 119899 gt
2 are referred to as higher-order methods (ℎ = mesh size)Recent developments seem to desire methods of higher-order for achieving higher accuracy numerical solutionsfor problems involving partial differential equations Whenhigher-order methods are evaluated factors such as rate ofconvergence stability and boundary conditions have to beconsidered too There are mainly two categories of finitedifference higher-order schemes
The first category is the noncompact stencils that util-ize any grid point surrounding the grid points of interestwhere the difference schemes are implemented The methodinvolves larger matrix bandwidth but in many cases theincrease is not large [1] There will be a probable increasein execution time as more grid points are used Howeverincreasing the number of grid points provides advantagesin terms of enhancement in accuracy and improvement inresolution [2] To enhance accuracy the approach shouldconsider proper treatment of boundary conditions of com-parable accuracy
The second category is the compact stencils that useLax-Wendroff rsquos idea [3] proposed by MacKinnon and Carey[4] The method uses smaller number of stencils makingit computationally efficient and highly accurate However itrequires inversion of matrix to obtain spatial derivative ateach point Also the boundary stencil has a large effect onthe stability and accuracy of the scheme [5] Usually compactschemes of order higher than four require formation ofauxiliary equations due to boundary conditions This resultsin large bandwidth matrices which are not symmetric [6]Furthermore the schemes can get fairly complicated formorecomplex equations
Some of the current findings on higher-order methodsinclude the work by Sulaiman et al [7] who suggesteda fourth-order quarter sweep modified successive over-relaxation (QSMSOR) iterative method for solving a one-dimensional parabolic equation It is found to be superior interms of rate of convergence and execution time as comparedto other SOR methods Jha [8] formulated the six-orderaccurate quarter sweep alternating group explicit (QSAGE)iterative finite difference method for solving nonlinear sin-gular two-point boundary value problems The method can
2 Journal of Applied Mathematics
be implemented in parallel and is found to be superiorto the corresponding full sweep alternating group explicit(AGE) and SOR methods Jin et al [9] proposed an AGEiteration method of the fourth-order accuracy by integratingthe grouping explicit method with numerical boundary con-ditionsThemethod was used to solve initial-boundary valueproblem of convection equations Fu and Tan [10] showedthat an unconditionally stable split-step FDTD method withhigher-order spatial accuracy is more accurate than thelower-order methods The dispersion error of the proposedmethod is comparable with the higher-order ADI-FDTDmethod
Higher-order methods are also studied by Mohebbi andDeghan [11] who applied a compact finite difference approx-imation of fourth-order and the cubic C1 spline colloca-tion method to some one-dimensional heat and advection-diffusion equations The scheme has fourth-order accuracyin both space and time and unconditionally stable Liao[12] proposed an efficient and high-order accuracy of thefourth-order compact finite difference method to solve one-dimensional Burgersrsquo equation Tian and Ge [13] studied astable fourth-order compact ADI method for solving two-dimensional unsteady convection diffusion problems Themethod is temporally second-order and spatially fourth-order accurate which requires only a regular five-point 2Dstencil similar to that in the standard second-order methodsChun [14] solved some nonlinear equations by applying thefourth-order iterative methods containing the Kingrsquos fourth-order family It was observed that the proposed methodhas at least equal performance compared to other methodsof the same order Zhu et al [15] presented a high-orderparallel finite difference algorithm based on the domaindecomposition strategy The study used the classical explicitscheme to calculate the interface values between subdomainsand the fourth-order compact schemes for the interior valuesThe method has high accuracy and is convergent and stableGao and Xie [16] devised a fourth-order alternating directionimplicit compact finite difference schemes for the solutionof two-dimensional Schrodinger equations The method ishighly competitive as compared to other existing methodsand it achieves the expected convergence rate
This study develops a higher-order finite difference algo-rithm with noncompact stencils that is capable of deliveringhighly accurate solutions for a one-dimensional heat equa-tion with Dirichlet boundary conditions The study focuseson modifying the unconditionally stable and convergentsecond-order iterative alternating decomposition explicit(IADE) method of Mitchell and Fairweather (IADEMF2)The IADEMF2 which was originally proposed by Sahimiet al [17] employs the fractional splitting of the Mitchell andFairweather (MF) variant that has an accuracy of the order119874((Δ119905)
2
+ (Δ119909)4
) The scheme executes a two-stage processinvolving the solution of sets of tridiagonal equations alonglines parallel to the first and second time steps respectivelyIt is found to be unconditionally stable convergent andmoreaccurate than the classical AGE class of methods namely theAGE method based on the Peaceman and Rachford variant(AGE-PR) whose spatial accuracy is of second order and
the AGEmethod based on the Douglas and Rachford variant(AGE-DR) whose temporal accuracy is only of first orderThe detailed derivation of the IADEMF2 can be obtainedfrom [17]
In this paper a fourth-order Crank-Nicolson (CN) dif-ference approximation is applied to the spatial derivative inthe heat equation and the MF variant is employed leadingto the formation of the fourth-order IADEMF (IADEMF4)numerical algorithm The convergence of the IADEMF4is analyzed and proven Numerical experiments verify thepotential of the IADEMF4 in enhancing the accuracy ofthe IADEMF2 The results of the proposed higher-orderscheme are also compared with the benchmarked fourth-order classical iterative methods such as the fourth-orderGauss-Seidel (GS4) the fourth-order Jacobi (JAC4) and thefourth-order successive over-relaxation (SOR4) methods
This paper is organized as follows Section 2 discusses theimplementation and stability of the fourth-order CN approx-imation on the heat equation In Section 3 the IADEMF4algorithm is formulated Section 4 analyses the convergenceof the IADEMF4 Section 5 provides the equations for thebenchmarked fourth-order classical iterative methods Thecomputational complexity of the methods considered in thispaper is given in Section 6 while the pseudocode for theIADEMF4 sequential algorithm is presented in Section 7Theexperiments conducted are given in Section 8 Sections 9and 10 provide some discussion and conclusion based on theobtained numerical results
2 Fourth-Order Crank-NicolsonApproximation to the Spatial Derivativein the Heat Equation
Consider the following one-dimensional parabolic heat equa-tion (1) that has been suitably assumed to be nondimensiona-lised It models the flow of heat in a homogeneous unchang-ing medium of finite extent in the absence of heat source
120597119880
120597119905=1205972
119880
1205971199092(1)
subject to given initial and Dirichlet boundary conditions
119880 (119909 0) = 119891 (119909) 0 le 119909 le 1
119880 (0 119905) = 119892 (119905) 0 lt 119905 le 119879
119880 (1 119905) = ℎ (119905) 0 lt 119905 le 119879
(2)
For the problem in (1) the finite difference approach dis-cretizes the time-space domain by placing a rectangular gridover the domain with grid spacing ofΔ119905 andΔ119909 in the 119905- and119909-directions respectively The grid consists of the set of linesparallel to the 119905-axis given by 119909
119894= 119894Δ119909 119894 = 0 1 119898119898 + 1
and a set of lines parallel to the 119909-axis given by 119905119896= 119896Δ119905 119896 =
0 1 119899 119899+1 For simplicity the grid spacing is taken to beuniform so thatΔ119909 = 1(119898+1) andΔ119905 = 119879(119899+1) At a grid-point119875(119909
119894 119905119896) in the solution domain the dependent variable
119880(119909 119905) which represents the nondimensional temperature attime 119905 and at position 119909 is approximated by 119906119896
119894
Journal of Applied Mathematics 3
At the grid-point 119875(119909119894 119905119896+12
) the IADEMF4 schemereplaces the spatial derivative in the heat equation with ahigher-order particularly the fourth-order Crank-Nicolson(CN) difference approximation [18] This is shown in theexpression given in (3) with the central difference operatorsdefined as 1205752
119909119906119896
119894= 119906119896
119894minus1minus 2119906119896
119894+ 119906119896
119894+1and 120575
2
119909119906119896+1
119894= 119906119896+1
119894minus1minus
2119906119896+1
119894+ 119906119896+1
119894+1 The approach gives the fourth-order scheme a
spatial truncation error of the order 119874(Δ119909)4 It enhances theaccuracy of its second-order counterpart which has a largererror of the order 119874(Δ119909)2
1
Δ119905(119906119896+1
119894minus 119906119896
119894) =
1
2(Δ119909)2(1205752
119909minus
1
121205754
119909) (119906119896+1
119894+ 119906119896
119894) (3)
To determine the stability of (3) the von Neumann sta-bility analysis can be applied Let 120582 = Δ119905(Δ119909)
2 and since1205754
119909119906119896
119894= 1205752
119909(1205752
119909119906119896
119894) the discretization of (3) becomes
119906119896+1
119894minus 119906119896
119894=120582
2(minus
1
12119906119896+1
119894minus2+4
3119906119896+1
119894minus1
minus5
2119906119896+1
119894+4
3119906119896+1
119894+1minus
1
12119906119896+1
119894+2
minus1
12119906119896
119894minus2+4
3119906119896
119894minus1minus5
2119906119896
119894
+4
3119906119896
119894+1minus
1
12119906119896
119894+2)
(4)
The discretized equation in (4) is assumed to have a solu-tion in the form of a Fourier harmonic function that is 119906119896
119894=
120588119896
119890119904120573119894Δ119909 where 120588 is referred to as the amplification factor 120573 is
an arbitrary constant and 119904 = radicminus1 The amplification factorrepresents the time dependence of the solution If the Fourierfunction is substituted into (4) and then solve for 120588 the resultwill be
120588 =
1 minus (1205826) ((cos120573 minus 4)2
minus 9)
1 + (1205826) ((cos120573 minus 4)2
minus 9)
with 120573 = 120573Δ119909 (5)
Since |120588| le 1 then the fourth-order CN approximation isunconditionally stable for any choice of 120573 Δ119905 and Δ119909
3 The Formulation of the IADEMF4
The IADEMF4 is firstly developed based on the execution ofthe unconditionally stable fourth-order CN approximation(3) on the heat equation
Equation (4) can also be expressed as in (6) using thedefinitions of the constants given in (7)
119886119906119896+1
119894minus2+ 119887119906119896+1
119894minus1+ 119888119906119896+1
119894+ 119889119906119896+1
119894+1+ 119890119906119896+1
119894+2
= minus119886119906119896
119894minus2minus 119887119906119896
119894minus1+ 119888119906119896
119894minus 119889119906119896
119894+1minus 119890119906119896
119894+2
119894 = 2 3 119898 minus 1
(6)
119886 =120582
24 119887 = minus
2120582
3 119888 =
4 + 5120582
4
119889 = minus2120582
3 119890 =
120582
24 119888 =
4 minus 5120582
4
(7)
The approximation in (6) can be displayed in a matrix formsuch as 119860u = f (8) where 119860 is a sparse pentadiagonalcoefficient matrix and u = (119906
2 1199063 119906
119898minus2 119906119898minus1
)119879 is the
columnvector containing the unknownvalues of119906 at the timelevel 119896 + 1 The column vector f = (119891
2 1198913 119891
119898minus2 119891119898minus1
)119879
consists of boundary values and known 119906 values at theprevious time level 119896 The definitions for every entry in f aregiven in (9)
119860u = f
[[[[[[[[[
[
119888 119889 119890
119887 119888 119889 119890 O119886 119887 119888 119889 119890
d d d d119886 119887 119888 119889 119890
O 119886 119887 119888 119889
119886 119887 119888
]]]]]]]]]
](119898minus2)times(119898minus2)
[[[[[[
[
1199062
1199063
119906119898minus2
119906119898minus1
]]]]]]
]119896+1
=
[[[[[[
[
1198912
1198913
119891119898minus2
119891119898minus1
]]]]]]
]
(8)
1198912= minus119887 (119906
119896
1+ 119906119896+1
1) + 119888119906
119896
2minus 119889119906119896
3minus 119890119906119896
4
1198913= minus119886 (119906
119896
1+ 119906119896+1
1) minus 119887119906
119896
2+ 119888119906119896
3minus 119889119906119896
4minus 119890119906119896
5
119891119894= minus119886119906
119896
119894minus2minus 119887119906119896
119894minus1+ 119888119906119896
119894minus 119889119906119896
119894+1minus 119890119906119896
119894+2
for 119894 = 4 5 119898 minus 3
119891119898minus2
= minus119886119906119896
119898minus4minus 119887119906119896
119898minus3+ 119888119906119896
119898minus2
minus 119889119906119896
119898minus1minus 119890 (119906
119896
119898+ 119906119896+1
119898)
119891119898minus1
= minus119886119906119896
119898minus3minus 119887119906119896
119898minus2+ 119888119906119896
119898minus1minus 119889 (119906
119896
119898+ 119906119896+1
119898)
(9)
The evaluations of 1198912 1198913 119891119898minus2
and 119891119898minus1
require thevalues of 119906 at the boundaries 119894 = 1 and 119894 = 119898 Howeverthese values cannot be obtained numerically because theircomputations involve nodes at 119894 = minus1 and 119894 = 119898 + 2which are exterior to the considered solution domain Ifthe exact solutions of 119906 are available at 119894 = 1 and 119894 =
119898 then it is appropriate to consider them as the requiredboundary values Otherwise the boundary conditions haveto be formulated bearing in mind that they should be ofcomparable accuracy [1]
The IADEMF4 scheme secondly employs the higher-order accuracy formula of MF [19] The variant whose accu-racy is of the order 119874((Δ119905)2 + (Δ119909)
4
) is as given in (10) and(11)
(119903119868 + 1198661) u(119901+12) = (119903119868 minus 119892119866
2) u(119901) + f (10)
(119903119868 + 1198662) u(119901+1) = (119903119868 minus 119892119866
1) u(119901+12) + 119892f (11)
where 119903 119901 and 119868 represent an acceleration parameter theiteration index and an identity matrix respectively 119866
1and
1198662are two constituent matrices The vectors u(119901+1) and
u(119901+12) represent the required solution at the iteration level(119901 + 1) and at some intermediate level (119901 + 12) respectivelyThe relation of 119892 and 119903 is given by 119892 = (6 + 119903)6 119903 gt 0
4 Journal of Applied Mathematics
Substitute the following expression obtained from (10)into (11)
u(119901+12) = (119903119868 + 1198661)minus1
[(119903119868 minus 1198921198662) u(119901) + f] (12)
and then simplify to obtain
(119903119868 + 1198662) u(119901+1) = (119903119868 minus 119892119866
1) (119903119868 + 119866
1)minus1
(119903119868 minus 1198921198662) u(119901)
+ [ (119903119868 minus 1198921198661) (119903119868 minus 119892119866
1)minus1
+ 119892] f (13)
As 119901 in (13) becomes sufficiently large the temperaturesolution reaches a steady state that is as 119901 rarr infin thenu(119901+1) rarr u and u(119901) rarr u Simplify the equation and thenmultiply by (119903119868+119866
1) After some algebraicmanipulations the
following expression is obtained
[119903 (1198661+ 1198662) (1 + 119892) + (1 minus 119892
2
)11986611198662] u = 119903 (1 + 119892) f (14)
Multiply (14) by 1119903(1 + 119892) and use the definition of 119892 tofinally obtain the following form
[ (1198661+ 1198662) minus
1
611986611198662] u = f (15)
The comparison between the expression 119860u = f in (8)and the form given in (15) suggests that the coefficient matrix119860 for the IADEMF4 can be decomposed into
119860 = 1198661+ 1198662minus1
611986611198662 (16)
The IADEMF4 requires the constituentmatrices1198661and119866
2in
(16) to be in the form of lower and upper tridiagonalmatricesrespectively in order to retain the pentadiagonal structure of119860 Thus
1198661=
[[[[[[[
[
1
1198971
1 119874
1
1198972
d2d dd 119897119898minus4
1
119874 119898minus4
119897119898minus3
1
]]]]]]]
](119898minus2)times(119898minus2)
1198662=
[[[[[[[
[
11989011
V1
1198902
2
V2
119874
d d d119874 119890
119898minus4119898minus4
V119898minus4
119890119898minus3
119898minus3
119890119898minus2
]]]]]]]
](119898minus2)times(119898minus2)
(17)
By substituting 1198661and 119866
2in (17) into the formula for the
decomposition of 119860 the entries of the resultant matrix arecompared with the entries of 119860 in (8) yielding the followingdefinitions
1198901=6 (119888 minus 1)
5
1=6119889
5 119897
1=
6119887
6 minus 1198901
1198902=6 (119888 minus 1) + 119897
11
5 V
119894=6119890
5for 119894 = 1 2 119898 minus 4
(18)
And for 119894 = 2 3 119898 minus 3
119894=6119889 + 119897
119894minus1V119894minus1
5
119894minus1=
6119886
6 minus 119890119894minus1
119897119894=6119887 +
119894minus1119894minus1
6 minus 119890119894
119890119894+1
=6 (119888 minus 1) + 119897
119894119894+ 119894minus1
V119894minus1
5
(19)
Since1198661and119866
2are three bandedmatrices then (119903119868+119866
1)
and (119903119868+1198662) can be inverted easilyThe equations in (10) and
(11) are rearranged as in (20) and (21) respectively
u(119901+12) = (119903119868 + 1198661)minus1
(119903119868 minus 1198921198662) u(119901)
+ (119903119868 + 1198661)minus1f
(20)
u(119901+1) = (119903119868 + 1198662)minus1
(119903119868 minus 1198921198661) u(119901+12)
+ 119892(119903119868 + 1198662)minus1f
(21)
The above two equations are computed and simplifiedleading to the computational formulae at each of the halfiteration levels as given in (22) and (23)
(i) At the (119901 + 12) iteration level
119906(119901+12)
119894=
1
119877(119864119894minus1
119906(119901)
119894+119882119894minus1
119906(119901)
119894+1+ 119881119894minus1
119906(119901)
119894+2
minus119894minus3
119906(119901+12)
119894minus2minus 119897119894minus2
119906(119901+12)
119894minus1+ 119891119894)
119894 = 2 3 119898 minus 2119898 minus 1
(22)
(ii) At the (119901 + 1) iteration level
119906(119901+1)
119894=
1
119885119894minus1
(119878119894minus3
119906(119901+12)
119894minus2+ 119876119894minus2
119906(119901+12)
119894minus1+ 119875119906(119901+12)
119894
minus119894minus1
119906(119901+1)
119894+1minus V119894minus1
119906(119901+1)
119894+2+ 119892119891119894)
119894 = 119898 minus 1 119898 minus 2 3 2
(23)
with
minus1
= 0= 1198970= 119881119898minus2
= 119881119898minus3
= 119882119898minus2
= 119898minus2
= V119898minus2
= V119898minus3
= 1198760= 119878minus1
= 1198780= 0
119877 = 1 + 119903 119875 = 119903 minus 119892
119864119894= 119903 minus 119892119890
119894 119885119894= 119903 + 119890
119894 119894 = 1 2 119898 minus 2
119882119894= minus119892
119894 119876119894= minus119892119897
119894 119894 = 1 2 119898 minus 3
119881119894=minus119892V119894 119878119894= minus119892
119894 119894 = 1 2 119898 minus 4
(24)
The IADEMF4 algorithm is regarded as a two-stageprocess involving two iteration levels (119901 + 12) and (119901 + 1)It is completed explicitly by using the required equations atthe two levels in alternate sweeps along all the grid pointsin the interval (0 1) until convergence is reached In (22)the calculation to determine the unknown 119906
(119901+12)
119894begins at
the left boundary and then moves to the right In a similar
Journal of Applied Mathematics 5
manner 119906(119901+1)119894
in (23) is calculated by proceeding from theright boundary towards the left (Figure 1)
The computational molecules are depicted in Figures 2and 3
At each level of iteration the computational moleculesinvolve two known grid points at the new level and anotherthree known ones at the old level Clearly the method isexplicit
4 Convergence Analysis of the IADEMF4
This section proves the convergence of the IADEMF4 Since120582 gt 1will not guarantee an accurate approximation for 120597119880120597119905[18] the values of 120582 that are considered appropriate for theproof are 0 lt 120582 le 1
From the definitions in ((18) and (19)) the followingresults are obtained
1198901=6 (119888 minus 1)
5=3120582
2 implying that 0 lt 119890
1le3
2
1=6119889
5lt0 119897
1=
6119887
6 minus 1198901
lt0 V119894=6119890
5gt 0
for 119894 = 1 2 119898 minus 4
2=6119889 + 119897
1V1
5= 1+1198971V1
5lt 1 since 119897
1V1lt 0
1198902=6 (119888 minus 1) + 119897
11
5= 1198901+11989711
5gt 1198901 since 119897
11gt 0
(25)
Simple computation shows that 119897115 lt 1 Clearly 119890
2lt 6
Lemma 1 119890119894in ((18) and (19)) is such that 0 lt 119890
119894lt 6 for all
119894 = 1 2 3 119898 minus 2
Proof The results in (25) show that 6 gt 1198902gt 1198901gt 0 Assume
it is true that 6 gt 119890119896gt 119890119896minus1
gt 0 for all 119896 = 3 4 5 119898 minus 3Since the assumption implies that 6 gt 119890
119896minus1gt 119890119896minus2
gt 0 then119896minus1
= 6119886(6 minus 119890119896minus1
) gt 119896minus2
= 6119886(6 minus 119890119896minus2
) gt 0 Therefore119896minus1
V119896minus1
gt 119896minus2
V119896minus2
It has also been shown that
2lt 1lt 0 Assume it is true
that 119895lt 119895minus1
lt 0 for all 119895 = 3 4 119898 minus 4 The assumptionimplies that
119895minus1lt 119895minus2
lt 0 Thus 119895minus1
119895minus1
lt 119895minus2
119895minus2
lt 0
119895+1
=6119889 + 119897
119895V119895
5= 1+119897119895V119895
5= 1+
(6119887 + 119895minus1
119895minus1
) V119895
5 (6 minus 119890119895)
lt 1+
(6119887 + 119895minus2
119895minus2
) V119895minus1
5 (6 minus 119890119895minus1
)
= 1+119897119895minus1
V119895minus1
5= 119895lt 0
(26)
Since it is true that for 119895 + 1 119895+1
lt 119895lt 0 then by
induction the assumption 119895lt 119895minus1
lt 0 is true for all 119895 An
0
u(p+12)
i
u(p+1)
i
i iminus 1 i + 1 m + 1
Figure 1 The two-stage IADEMF4 algorithmThe directions of thesweeps at the (119901 + 12) and (119901 + 1) iteration levels
i iminus 1i minus 2 i + 1 i + 2
(p + 12)
(p)
Figure 2 Computational molecule of the IADEMF4 at the (119901+12)iteration level
i iminus 1i minus 2 i + 1 i + 2
(p + 12)
(p + 1)
Figure 3 Computational molecule of the IADEMF4 at the (119901 + 1)
iteration level
equivalent to the preceding statement would be 119896lt 119896minus1
lt 0
for all 119896 = 3 4 5 119898 minus 3 It follows that
119897119896119896=(6119887 +
119896minus1119896minus1
) 119896
6 minus 119890119896
gt(6119887 +
119896minus2119896minus2
) 119896minus1
6 minus 119890119896minus1
= 119897119896minus1
119896minus1
119890119896+1
=6 (119888 minus 1) + 119897
119896119896+ 119896minus1
V119896minus1
5= 1198901+119897119896119896+ 119896minus1
V119896minus1
5
gt 1198901+119897119896minus1
119896minus1
+ 119896minus2
V119896minus2
5= 119890119896gt 0
(27)
Since it is true that for 119896+1 119890119896+1
gt 119890119896gt 0 then by induction
119890119896gt 119890119896minus1
gt 0 is also true for all 119896Suppose there is a 119896 such that 119890
119896gt 6 Then
119890119896+1
= 1198901+119897119896119896+ 119896minus1
V119896minus1
5
= 1198901+
6119887 + 119896
5 (6 minus 119890119896)+119896minus1
5(119896minus1
119896
6 minus 119890119896
+ V119896minus1
) lt 1198901
(28)
This is a contradiction since 119890119896+1
gt 1198901 This verifies that 6 gt
119890119896gt 119890119896minus1
gt 0 119896 = 3 4 5 119898 minus 3 Let 119894 = 2 3 4 119898 minus 2and then 6 gt 119890
119894gt 119890119894minus1
gt 0 Therefore 0 lt 119890119894lt 6 119894 =
1 2 3 119898 minus 2 Lemma 1 is proved
6 Journal of Applied Mathematics
Lemma 2 If 119903 gt 0 and 119892 = (6+ 119903)6 thenmax |(119903 minus119892119890119894)(119903 +
119890119894)| lt 1 i = 1 2 3 119898 minus 2
Proof Let max |(119903 minus 119892119890119894)(119903 + 119890
119894)| = |(119903 minus 119892119890
119895)(119903 + 119890
119895)| for
some 119895Assume |(119903 minus 119892119890
119895)(119903 + 119890
119895)| ge 1 Since 119903 gt 0 and 119890
119895gt 0
then 119903 + 119890119895gt 0
If (119903 minus 119892119890119895)(119903 + 119890
119895) ge 1 then minus((6 + 119903)6)119890
119895ge 119890119895 which
implies that 119903 le minus12 This contradicts the fact that 119903 gt 0If (119903 minus 119892119890
119895)(119903 + 119890
119895) le minus1 then 2119903 minus ((6 + 119903)6)119890
119895le minus119890119895
which implies that 119890119895ge 12 This contradicts Lemma 1
So the assumption that |(119903 minus 119892119890119895)(119903 + 119890
119895)| ge 1 is false
Hence max |(119903 minus 119892119890119894)(119903 + 119890
119894)| = |(119903 minus 119892119890
119895)(119903 + 119890
119895)| lt 1
Proposition 3 (119903119868 minus 1198921198661)(119903119868 + 119866
1)minus1
2lt 1
Proof Let 119865 = (119903119868 minus 1198921198661)(119903119868 + 119866
1)minus1 that is
119865 =
[[[[[[[[[[
[
119903 minus 119892
119889
d119903 minus 119892
119889119874
d d d
d d
sdot sdot sdot sdot sdot sdot sdot sdot sdot sdot sdot sdot119903 minus 119892
119889
]]]]]]]]]]
]
(29)
119865 is a lower triangular matrix with all the diagonal entries(whence the eigenvalues of 119865) equal to (119903 minus 119892)119889 where 119889 =
119903 + 1 Denote all the eigenvalues of 119865 by 120582119865 If 120588[119865] is defined
as the spectral radius of 119865 then
120588 [119865] = max 10038161003816100381610038161205821198651003816100381610038161003816 =
10038161003816100381610038161205821198651003816100381610038161003816 =
1003816100381610038161003816100381610038161003816
119903 minus 119892
119903 + 1
1003816100381610038161003816100381610038161003816le 119865
2 (30)
But by definition of 2-norm
1198652= maxx2 = 0
119865x2
x2
= maxy2=1
1003817100381710038171003817119865y10038171003817100381710038172
1003817100381710038171003817y10038171003817100381710038172
= maxy2=1
1003817100381710038171003817119865y10038171003817100381710038172
le maxw2=1
119865w2= maxw2=1
1003817100381710038171003817120582119865w10038171003817100381710038172
(31)
Since all the eigenvalues of 119865 are equal then
1198652le1003816100381610038161003816120582119865
1003816100381610038161003816 maxw2=1
w2=1003816100381610038161003816120582119865
1003816100381610038161003816 = 120588 [119865] (32)
Thus from (30) and (32) 1198652= |120582119865| = |(119903 minus 119892)(119903 + 1)|
By Lemma 2 with 119890119895replaced by 1 |(119903 minus 119892)(119903 + 1)| lt 1
is obtained leading to the result of Proposition 3 which is1198652= (119903119868 minus 119892119866
1)(119903119868 + 119866
1)minus1
2lt 1
Proposition 4 (119903119868 minus 1198921198662)(119903119868 + 119866
2)minus1
2lt 1
Proof Let 119870 = (119903119868 minus 1198921198662)(119903119868 + 119866
2)minus1 that is
119870 =
[[[[[[[[[[[[[
[
119903 minus 1198921198901
1199111
d sdot sdot sdot sdot sdot sdot sdot sdot sdot
119903 minus 1198921198902
1199112
d sdot sdot sdot
d
119874 d d
119903 minus 119892119890119898
119911119898
]]]]]]]]]]]]]
]
(33)
119870 is an upper triangular matrix with all the diagonal entriesequal to (119903 minus 119892119890
119894)119911119894 119894 = 1 2 119898 minus 2 where 119911
119894= 119903 + 119890
119894
Since all the eigenvalues of119870 are distinct then119870 is similar toa diagonal matrix119863
By the Schur triangularization theorem [20] there is anorthogonal matrix 119874 such that 119874119879119870119874 = 119863 The diagonalentries of119863 are the eigenvalues of119870
1198702=10038171003817100381710038171003817119874119879
119870119874100381710038171003817100381710038172
= 1198632
= radicmaximum eigenvalue of 119863119879119863
= radicmaximum eigenvalue of 1198632
= radic120588 (1198632) = radic120588 (1198702) = 120588 (119870)
(34)
By Lemma 2
1198702= 120588 (119870) = max
10038161003816100381610038161003816100381610038161003816
119903 minus 119892119890119894
119903 + 119890119894
10038161003816100381610038161003816100381610038161003816
=
1003816100381610038161003816100381610038161003816100381610038161003816
119903 minus 119892119890119895
119903 + 119890119895
1003816100381610038161003816100381610038161003816100381610038161003816
lt 1 (35)
for some 119895 Hence this proves Proposition 4
From (20) and (21)
u(119901+1) = (119903119868 + 1198662)minus1
(119903119868 minus 1198921198661)
times [(119903119868 + 1198661)minus1
(119903119868 minus 1198921198662) u(119901) + (119903119868 + 119866
1)minus1f]
+ 119892 (119903119868 + 1198662)minus1f
(36)
Let
119872(119903) = (119903119868 + 1198662)minus1
(119903119868 minus 1198921198661) (119903119868 + 119866
1)minus1
(119903119868 minus 1198921198662)
(37)
And let
119902 (119903) = [(119903119868 + 1198662)minus1
(119903119868 minus 1198921198661) (119903119868 + 119866
1)minus1
+119892(119903119868 + 1198662)minus1
] f (38)
Then u(119901+1) = 119872(119903)u(119901) + 119902(119903)
Theorem 5 The IADEMF4 is convergent if 120588[119872(119903)] lt 1 for119903 gt 0
Journal of Applied Mathematics 7
Proof Define (119903) = (119903119868 + 1198662)119872(119903)(119903119868 + 119866
2)minus1 then
(119903) = (119903119868 minus 1198921198661) (119903119868 + 119866
1)minus1
(119903119868 minus 1198921198662) (119903119868 + 119866
2)minus1
(39)
Thus by similarity119872(119903) and (119903) have the same set of eigen-values Therefore
120588 [119872 (119903)]
= 120588 [ (119903)] le10038171003817100381710038171003817 (119903)
10038171003817100381710038171003817 2
=10038171003817100381710038171003817(119903119868 minus 119892119866
1) (119903119868 + 119866
1)minus1
(119903119868 minus 1198921198662) (119903119868 + 119866
2)minus110038171003817100381710038171003817 2
le10038171003817100381710038171003817(119903119868 minus 119892119866
1) (119903119868 + 119866
1)minus1100381710038171003817100381710038172
10038171003817100381710038171003817(119903119868 minus 119892119866
2) (119903119868 + 119866
2)minus110038171003817100381710038171003817 2
lt 1
(40)
The last inequality is due to Propositions 3 and 4The provenTheorem 5 assures the convergence of the IADEMF4
5 The Fourth-Order ClassicalIterative Methods
The system of linear equations in (8) may also be solvedby using the classical iterative methods of the fourth orderThey include the JAC4 the GS4 and the SOR4 methodsThese iterative methods are capable of exploiting the sparsestructure of the pentadiagonal matrix
The JAC4 algorithm can be represented by
119906(119901+1)
119894=
(119891119894minus119886119906(119901)
119894minus2minus 119887119906(119901)
119894minus1minus 119889119906(119901)
119894+1minus 119890119906(119901)
119894+2)
119888
119894 = 2 3 119898 minus 1
(41)
The approximation of 119906(119901+1)119894
at the (119901 + 1)th iteration level iscomputed using the relevant values in the 119901th iteration
The GS4 method uses the most recent values of 119906(119901+1)119894minus2
and 119906(119901+1)
119894minus1to update the approximation value of 119906(119901+1)
119894 The
algorithm for the GS4 is as expressed in (42)
119906(119901+1)
119894=
(119891119894minus119886119906(119901+1)
119894minus2minus 119887119906(119901+1)
119894minus1minus 119889119906(119901)
119894+1minus 119890119906(119901)
119894+2)
119888
119894 = 2 3 119898 minus 1
(42)
The SOR4 iterative method accelerates the convergencerate of the GS4 If 120596 is a relaxation parameter then for any120596 = 0 (42) can be rewritten as
119906(119901+1)
119894= (1 minus 120596) 119906
(119901)
119894
+
120596 (119891119894minus 119886119906(119901+1)
119894minus2minus 119887119906(119901+1)
119894minus1minus 119889119906(119901)
119894+1minus 119890119906(119901)
119894+2)
119888
119894 = 2 3 119898 minus 1
(43)
i iminus 1i minus 2 i + 1 i + 2
(p + 1)
(p)
Figure 4 Computational molecule of the JAC4
i iminus 1i minus 2 i + 1 i + 2
(p + 1)
(p)
Figure 5 Computational molecule of the GS4SOR4
Table 1 Sequential arithmetic operations per iteration (119898 problemsize)
Method Number ofadditions
Number ofmultiplications
Total operationcount
IADEMF4 10(119898 minus 2) 13(119898 minus 2) 23(119898 minus 2)
IADEMF2 6119898 9119898 15119898
JAC4GS4 4(119898 minus 2) 5(119898 minus 2) 9(119898 minus 2)
SOR4 5(119898 minus 2) 7(119898 minus 2) 12(119898 minus 2)
The SOR4 algorithm reduces to the GS4 if 120596 = 1 Exceptfor very special cases it is difficult to obtain the analyticexpression for 120596 According to Young [21] the precise deter-mination of the optimal 120596 in the SOR is only known for asmall class of matrices The value of 120596 generally lies in therange of 1 lt 120596 lt 2
The computational molecules for the JAC4 and the GS4are as illustrated in Figures 4 and 5 respectively The SOR4has the same form as the GS4
6 Computational Complexity
The cost of implementing an algorithm can be assessedby examining its computational complexity The number ofarithmetic operations such as additions (including subtrac-tions) andmultiplications (including divisions) that is neededto perform by the algorithm can be straightforwardly count-ed Table 1 gives the number of sequential arithmetic opera-tions per iteration that is required to evaluate the algorithms
For the higher-order schemes trade-offbetween accuracyand speed usually happens The computational complexityhas an effect on the efficiency of a particular scheme Thusthis factor will be taken into account when discussing theresults of the numerical experiments conducted in this study
7 The Pseudocode of the IADEMF4Sequential Algorithm
Algorithm 1 illustrates the pseudocode of the IADEMF4sequential algorithm This algorithm can also be generalizedand applied to other methods discussed in this paper
8 Journal of Applied Mathematics
begindetermine the parameters119898 119903 Δ119905 Δ119909 120582 120576 119879
determine initial conditions 119906(119901)119894
determine exact solutions 119880119894at time level 119879
while (time level lt 119879)determine boundary conditionsfor 119894 = 2 to 119894 = 119898 minus 1
compute 119891119894(refer to (9))
end-forset iteration = 0while (convergence conditions are not satisfied)
for 119894 = 2 to 119894 = 119898 minus 1
compute 119906(119901+12)119894
(refer to (22))end-forfor 119894 = 2 to 119894 = 119898 minus 1
compute 119906(119901+1)119894
(refer to (23))end-fortest for convergencecompute 119890
119894larr
10038161003816100381610038161003816119906(119901+1)
119894minus 119906(119901)
119894
10038161003816100381610038161003816if (max1003816100381610038161003816119890119894
1003816100381610038161003816 lt 120576)then 119906
(119901)
119894larr 119906(119901+1)
119894
add 1 to iteration (if necessary)end-while
end-whilefor 119894 = 2 to 119894 = 119898 minus 1
determine average absolute error root meansquare error and maximum error
end-forend
Algorithm 1 The IADEMF4 sequential algorithm
8 Numerical Experiments
Two experiments were conducted to test the sequentialnumerical performance of the proposed IADEMF4 methodagainst those of the benchmarked classical iterative methodsnamely the JAC4 theGS4 and the SOR4 Comparison is alsomade with the corresponding IADE method of the secondorder
Experiment 1 This problem was taken from Saulev [22]
120597119880
120597119905=1205972
119880
1205971199092 0 le 119909 le 1 (44)
subject to the initial condition
119880 (119909 0) = 4119909 (1 minus 119909) 0 le 119909 le 1 (45)
and the boundary conditions
119880 (0 119905) = 119880 (1 119905) = 0 119905 ge 0 (46)
The exact solution to the given problem is given by
119880 (119909 119905) =32
1205873
infin
sum
119896=1(2)
1
1198963119890minus12058721198962119905 sin (119896120587119909) (47)
Experiment 2 This problem was taken from Johnson andReiss [23]
120597119880
120597119905=1205972
119880
1205971199092 0 le 119909 le 1 (48)
subject to the initial condition
119880 (119909 0) = sin (120587119909) (1 + 6 cos (120587119909)) 0 le 119909 le 1 (49)
and the boundary conditions
119880 (0 119905) = 119880 (1 119905) = 0 119905 ge 0 (50)
The exact solution to the given problem is given by
119880 (119909 119905) = sin (120587119909) 119890minus1205872119905
+ 3 sin (2120587119909) 119890minus41205872119905
(51)
Journal of Applied Mathematics 9
9 Results and Discussion
In each experiment the exact solutions at 119894 = 1 and 119894 = 119898
were taken as boundary values for the IADEMF4 and theother fourth-order classical iterative methods The conver-gence criterion used in the testing of each method was takenas 119906(119901+1) minus 119906
(119901)
infin
le 120576 where 120576 is the convergence toleranceThe selections of the optimum 119903 or 120596 were determined byexperiments As for the IADEMF2 the CN scheme with 120579 =
12 was employedFigures 6 7 8 and 9 visualize the behavior of the one-
dimensional parabolic heat solutions for both experimentsBy using119898 = 10 and a tolerance requirement of 120576 = 10
minus4 twodifferent mesh sizes 120582 = 05 and 120582 = 10 were consideredin each experiment The exact solutions are compared withthe IADEMF4 and the IADEMF2 numerical solutions Theoptimum value of 119903 chosen for each method as well as theircorresponding outcome of the number of iterations 119899 isstated in the legend of each figure Every figure reveals thatthe IADEMF4 is more accurate than the IADEMF2 andthe former converges with fewer numbers of iterations incomparison to the latter The numerical solutions of theIADEMF4 seem to be in very good agreement with theexact solutions For example in Figure 7 at 119909 = 05 thedifference between the exact solution and the numericalsolution using the IADEMF2 and using the IADEMF4 isabout 36 and 0 respectively These results imply thatthe accuracy and convergence rate of the second-order IADEmethod are enhanced by the implementation of the fourth-order CN approximation that leads to the formation of thecorresponding fourth-order IADE scheme
Figure 10 displays the graph of log (RMSE) versus log(Δ119909)for decreasing values of Δ119909 implemented on Experiment 2By considering 120576 = 10
minus4 and fixing Δ119905 = 00001 the value ofΔ119909 was initially taken as Δ119909 = 0125 It was then successivelyhalved into Δ1199092 Δ1199094 and Δ1199098
The figure shows that amongst the tested methods theroot mean square error (RMSE) of the IADEMF4 and theIADEMF2 decreases linearly as the values of Δ119909 decreasesThe slope of the IADEMF4 is approximately equal to 4which corresponds to its fourth-order spatial accuracy TheIADEMF2 with a second-order spatial accuracy has a slopethat is approximately equal to 2 It is clear that the IADEMF4is always more accurate than the IADEMF2 for the differentconsidered mesh sizes The graphs of the SOR4 (120596 = 105)the GS4 and the JAC4 show that their accuracies of fourthorder tend to lack as Δ119909 decreases largely due to the effect ofincreasing round-off errors as the value of119898 increases
Tables 2 3 4 5 6 7 8 and 9 provide numerical results interms of the average absolute error (AAE) root mean squareerror (RMSE) maximum error (ME) number of iterations(119899) and execution time (ET) measured in seconds (s) Theresults are obtained from both experiments for two differentvalues of 119898 and mesh size 120582 It is generally observed thatwhen 119898 = 700 120582 = 05 and 120576 = 10
minus6 the IADEMF4has the least average absolute error root mean square errorand maximum error in comparison with the other methodsunder consideration (Tables 2ndash5) When the size of 119898 was
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 08 n = 172)IADEMF2 (r = 03 n = 185)
00E + 00
10E minus 02
20E minus 02
30E minus 02
40E minus 02
50E minus 02
60E minus 02
70E minus 02
80E minus 02
90E minus 02
10E minus 01
U(xt)
Figure 6 Numerical and exact solutions for Experiment 1 120582 = 05Δ119909 = 01 Δ119905 = 0005 119905 = 025 and 120576 = 10
minus4
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 14 n = 159)IADEMF2 (r = 05 n = 162)
00E + 00
10E minus 03
20E minus 03
30E minus 03
40E minus 03
50E minus 03
60E minus 03
70E minus 03
80E minus 03
90E minus 03
U(xt)
Figure 7 Numerical and exact solutions for Experiment 1 120582 = 10Δ119909 = 01 Δ119905 = 001 119905 = 05 and 120576 = 10
minus4
ten times bigger (119898 = 7000) and a more stringent tolerancecriterion was set (120576 = 10
minus10
) the accuracy of the IADEMF4clearly outperforms the other methods for a mesh size of120582 = 05 (Tables 6 and 8) However the difference in the errorsamongst the tested methods is not so obvious for the case of120582 = 10 (Tables 7 and 9)The achievement of the fourth-orderIADEmethod in Experiments 1 and 2 can be clearly seen fromthe results in Tables 2 and 8 respectively where it has caused ahuge 94 reduction of RMSE from its corresponding second-order IADE method
The IADEMF4 has the advantage of featuring higheraccuracy due to the execution of the fourth-order CNapprox-imation coupled with the fourth-order accurate MF variant
10 Journal of Applied Mathematics
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 10 n = 169)IADEMF2 (r = 03 n = 213)
00E + 00
10E minus 02
20E minus 02
30E minus 02
40E minus 02
50E minus 02
60E minus 02
70E minus 02
80E minus 02
90E minus 02
10E minus 01
U(xt)
Figure 8 Numerical and exact solutions for Experiment 2 120582 = 05Δ119909 = 01 Δ119905 = 0005 119905 = 025 and 120576 = 10
minus4
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 10 n = 175)IADEMF2 (r = 05 n = 182)
00E + 00
10E minus 03
20E minus 03
30E minus 03
40E minus 03
50E minus 03
60E minus 03
70E minus 03
80E minus 03
U(xt)
Figure 9 Numerical and exact solutions for Experiment 2 120582 = 10Δ119909 = 01 Δ119905 = 001 119905 = 05 and 120576 = 10
minus4
The IADEMF2 is only derived from the second-order CNapproximation but its combination with its correspondingfourth-order MF variant managed to produce errors whichare better than theGS4 JAC4 and the SOR4 Even though theclassical iterative methods are also derived from the fourth-order accurate CN type approximation they are lackingin accuracy due to the round-off errors that have beenaccumulated from the time the execution starts till it ends
With regards to the rate of convergence the resultsfrom each table demonstrate that the number of iterationsproduced by the IADEMF2 and the fourth-order classicaliterative methods is greater than or at least equal to that of
0
minus2
minus208 minus277 minus347 minus416
minus4
minus8
minus6
minus10
minus12
minus14
minus16
minus18
GS4JAC4SOR4
log
(RM
SE)
log(Δx)
IADEMF4
IADEMF2
Figure 10 Log(RMSE) versus Log(Δ119909) Experiment 2 Δ119905 = 00001119905 = 0002 and 120576 = 10
minus4
Table 2 Experiment 1 119898 = 700 120582 = 05 Δ119909 = 143 times 10minus3 Δ119905 =
102 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 07) 267119890 minus 9 330119890 minus 9 212119890 minus 8 100 440119890 minus 2
IADEMF2(119903 = 08) 188119890 minus 8 495119890 minus 8 323119890 minus 7 100 635119890 minus 3
SOR4(120596 = 12) 540119890 minus 7 544119890 minus 7 552119890 minus 7 108 460119890 minus 2
GS4 540119890 minus 6 543119890 minus 6 332119890 minus 6 150 472119890 minus 2
JAC4 228119890 minus 5 229119890 minus 5 232119890 minus 5 150 473119890 minus 2
Table 3 Experiment 1 119898 = 700 120582 = 10 Δ119909 = 143 times 10minus3 Δ119905 =
204 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 297119890 minus 7 301119890 minus 7 301119890 minus 7 50 201119890 minus 2
IADEMF2(119903 = 09) 557119890 minus 7 559119890 minus 7 848119890 minus 7 50 317119890 minus 3
SOR4(120596 = 12) 329119890 minus 6 331119890 minus 6 335119890 minus 6 76 225119890 minus 2
GS4 895119890 minus 6 900119890 minus 6 900119890 minus 6 100 238119890 minus 2
JAC4 214119890 minus 5 215119890 minus 5 217119890 minus 5 125 250119890 minus 2
the IADEMF4 The convergence rate of the latter surpassesthe others in the case of 119898 = 7000 and 120576 = 10
minus10 (Tables6ndash8) Even though the operational count of the IADEMF4is relatively quite large (Table 1) due to its higher level ofaccuracy its increasing number of correct digits at eachiteration causes it to converge at a faster rate The IADEMF2has the advantage of having less mathematical operationsthus it is competitive in terms of convergence rate but atthe expense of accuracy The application of the fourth-orderCN approximation on the heat equation proves that thethree classical iterative methods also converge with JAC4appearing to be the slowest and the least accurate amongst all
Journal of Applied Mathematics 11
Table 4 Experiment 2 119898 = 700 120582 = 05 Δ119909 = 143 times 10minus3 Δ119905 =
102 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME n ET (s)IADEMF4(119903 = 07) 207119890 minus 8 231119890 minus 8 345119890 minus 8 100 903119890 minus 3
IADEMF2(119903 = 08) 106119890 minus 7 175119890 minus 7 175119890 minus 7 100 647119890 minus 3
SOR4(120596 = 11) 154119890 minus 6 171119890 minus 6 255119890 minus 6 200 110119890 minus 2
GS4 294119890 minus 6 327119890 minus 6 487119890 minus 6 250 118119890 minus 2
JAC4 124119890 minus 5 138119890 minus 5 206119890 minus 5 300 127119890 minus 2
Table 5 Experiment 2 119898 = 700 120582 = 10 Δ119909 = 143 times 10minus3 Δ119905 =
204 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 09) 138119890 minus 6 154119890 minus 6 229119890 minus 6 75 650119890 minus 3
IADEMF2(119903 = 05) 187119890 minus 6 207119890 minus 6 309119890 minus 6 75 472119890 minus 3
SOR4(120596 = 12) 312119890 minus 6 347119890 minus 6 517119890 minus 6 125 906119890 minus 3
GS4 667119890 minus 6 742119890 minus 6 111119890 minus 5 175 932119890 minus 3
JAC4 126119890 minus 5 141119890 minus 5 210119890 minus 5 250 105119890 minus 2
Table 6 Experiment 1 119898 = 7000 120582 = 05 Δ119909 = 143 times 10minus4 Δ119905 =
102 times 10minus8 119905 = 510 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME n ET (s)IADEMF4(119903 = 08) 367119890 minus 10 433119890 minus 10 366119890 minus 9 173 146119890 minus 1
IADEMF2(119903 = 05) 375119890 minus 10 435119890 minus 10 376119890 minus 9 200 124119890 minus 1
SOR4(120596 = 112) 455119890 minus 10 507119890 minus 10 367119890 minus 9 257 147119890 minus 1
GS4 110119890 minus 9 111119890 minus 9 376119890 minus 9 300 160119890 minus 1
JAC4 230119890 minus 9 231119890 minus 9 378119890 minus 9 400 163119890 minus 1
Table 7 Experiment 1 119898 = 7000 120582 = 10 Δ119909 = 143 times 10minus4 Δ119905 =
204 times 10minus8 119905 = 51 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 163119890 minus 7 163119890 minus 7 163119890 minus 7 112 960119890 minus 2
IADEMF2(119903 = 12) 163119890 minus 7 163119890 minus 7 163119890 minus 7 120 736119890 minus 2
SOR4(120596 = 115) 163119890 minus 7 163119890 minus 7 164119890 minus 7 171 999119890 minus 2
GS4 164119890 minus 7 164119890 minus 7 164119890 minus 7 216 107119890 minus 1
JAC4 165119890 minus 7 165119890 minus 7 165119890 minus 7 312 111119890 minus 1
It is found that the convergence rate of the fourth-orderand second-order methods improves with the application oflarger mesh size that is from 120582 = 05 to 120582 = 10 Thecoarser meshes cause a reduction in the computational oper-ations thus giving better rate of convergence However a
Table 8 Experiment 2 119898 = 7000 120582 = 05 Δ119909 = 143 times 10minus4 Δ119905 =
102 times 10minus8 119905 = 510 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 07) 567119890 minus 13 321119890 minus 12 130119890 minus 10 150 131119890 minus 1
IADEMF2(119903 = 07) 199119890 minus 11 460119890 minus 11 182119890 minus 9 200 127119890 minus 1
SOR4(120596 = 115) 245119890 minus 10 273119890 minus 10 399119890 minus 10 260 151119890 minus 1
GS4 398119890 minus 10 443119890 minus 10 661119890 minus 10 400 171119890 minus 1
JAC4 105119890 minus 9 117119890 minus 9 174119890 minus 9 550 176119890 minus 1
Table 9 Experiment 2 119898 = 7000 120582 = 10 Δ119909 = 143 times 10minus4 Δ119905 =
204 times 10minus8 119905 = 51 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 154119890 minus 6 171119890 minus 6 255119890 minus 6 96 912119890 minus 2
IADEMF2(119903 = 10) 154119890 minus 6 171119890 minus 6 255119890 minus 6 96 607119890 minus 2
SOR4(120596 = 12) 154119890 minus 6 171119890 minus 6 255119890 minus 6 192 997119890 minus 2
GS4 154119890 minus 6 171119890 minus 6 255119890 minus 6 288 128119890 minus 1
JAC4 154119890 minus 6 172119890 minus 6 256119890 minus 6 408 146119890 minus 1
more accurate solution is obtained by using finer mesh Forexample Table 6 shows that for 120582 = 05 the IADEMF4 hasRMSE = 433119890 minus 10 and 119899 = 173 whereas Table 7 shows thatfor 120582 = 10 its RMSE = 163119890 minus 7 and 119899 = 112 In generalamongst all the testedmethods the IADEMF4 still maintainsits greater accuracy characteristic even with coarser meshes
In terms of execution time the results from every tabledisplay shorter execution time for the IADEMF2 in compari-son to the IADEMF4 This is expected since the IADEMF2has lower computational complexity (Table 1) Despite theachievement in accuracy the IADEMF4 has to performmorecomputational work than the IADEMF2 since the former hasto utilize values of 119906 at two grid points on either side of thepoint (119894Δ119909 119896Δ119905) along the 119896th time level Thus if accuracyis desired then the preferred sequential numerical algorithmwould be the IADEMF4 On the other hand if execution timematters then the choice would be the IADEMF2
Amongst the fourth-order methods the IADEMF4 exe-cutes in the least amount of time Even though its computa-tional complexity is relatively large it operates with the leastnumber of iterations thus enabling it to be the most efficient
10 Conclusion
This paper proposes the development of the novel fourth-order IADEMF4 finite difference scheme The higher-orderscheme is developed by representing the spatial derivative inthe heat equation with the fourth-order finite difference CNapproximation This leads to the formation of pentadiagonalmatrices in the systems of linear equations with larger com-putational stencils The algorithm also employs the higheraccuracy of the Mitchell and Fairweather variant Despite
12 Journal of Applied Mathematics
the fourth-order IADE schemersquos higher computational com-plexity this technique is proved to be valuable becauseit enhances the accuracy of its second-order counterpartnamely the IADEMF2 In addition the higher accuracy ofIADEMF4 is verified as a convergent and unconditionallystable scheme and is superior in terms of rate of convergenceIt has also been proven to execute more efficiently in compar-ison to the other benchmarked fourth-order classical iterativemethods such as the GS4 the SOR4 and the JAC4 The in-creasing number of correct digits at each iteration serves as anadvantage for the IADEMF4 thus yielding faster rate of con-vergence with higher level of accuracy
In conclusion the proposed IADEMF4 scheme affordsusers many advantages with respect to higher accuracy sta-bility and rate of convergence and it serves as an alternativeefficient technique for the solution of a one-dimensional heatequation with Dirichlet boundary conditions
The proposed fourth-order scheme can be modified andadapted to more general multidimensional linear and non-linear parabolic elliptic and hyperbolic partial differentialequations using different types of boundary conditionsBesides it can also be considered for applications in problemsthat require higher-order accuracy with high resolution suchas problems in nanocomputing that require the solution ofvery large sparse systems of equation
The explicit and high accuracy feature of the IADEMF4can be exploited in two- and three-dimensional heat prob-lems by applying the scheme as pentadiagonal solvers forthe system of linear equations arising in the sweeps of ahigher-order alternating direction implicit (ADI) schemeThe lower-order ADI scheme was initially proposed byPeaceman and Rachford [24] The higher space dimensionsare expected not to cause unsatisfactory performance for theADI-IADEMF4 as long as the proposed scheme is stableconvergent and efficient as the process of its implementa-tion continues to advance from one-time step to anotherPathirana et al [25] implemented the ADI in developing atwo-dimensional model for incorporating flood damage inurban drainage planningThe proposed model is found to bestable numerically accurate and computationally efficientMirzavand et al [26] proposed the ADI-FDTD methodfor the physical modeling of high-frequency semiconductordevices The approach is able to reduce significantly the full-wave simulation time
It is possible to parallelize the IADEMF4 algorithm sincethe calculation of the new iteration only depends on theknown values from the last iteration (Figures 2 and 3) Futurework is to exploit the explicit computational properties andefficiency of the IADEMF4 for parallelization and executionon distributed parallel computing systems The idea is tospeed up the execution time without compromising its accu-racy especially on problems involving very large linear sys-tems of equations
References
[1] J Noye Computational Techniques for Differential Equationsvol 83 Elsevier Science North-Holland Publishing Amster-dam The Netherlands 1984
[2] G W Recktenwald ldquoFinite difference approximations to theheat equationrdquoMechanical Engineering vol 10 pp 1ndash27 2004
[3] P Lax and B Wendroff ldquoSystems of conservation lawsrdquo Com-munications on Pure and Applied Mathematics vol 13 pp 217ndash237 1960
[4] R J MacKinnon and G F Carey ldquoAnalysis of material interfacediscontinuities and superconvergent fluxes in finite differencetheoryrdquo Journal of Computational Physics vol 75 no 1 pp 151ndash167 1988
[5] R Hixon and E Turkel High Accuracy Compact MacCormack-Type Schemes For Computational Aerocoustics NASACentre forAerospace Information Lewis Research Centre 1998
[6] Y Lin Higher-order finite difference methods for solving heatequations [PhD thesis] Southern Illinois University 2008
[7] J Sulaiman M K Hasan M Othman M and S A AbdulKarim ldquoFourth-order QSMOR iterative method for the solu-tion of one-dimensional parabolic PDErsquosrdquo in Proceedings of theInternational Conference on Industrial and AppliedMathematics(CIAM rsquo10) pp 34ndash39 ITB Bandung Indonesia 2010
[8] N Jha ldquoThe application of sixth order accurate parallel quar-ter sweep alternating group explicit algorithm for nonlinearboundary value problems with singularityrdquo in Proceedings ofthe 2nd International Conference on Methods and Models inComputer Science (ICM2CS rsquo10) pp 76ndash80 December 2010
[9] Y Jin G Jin and J Li ldquoA class of high-precision finite differenceparallel algorithms for convection equationsrdquo Journal of Conver-gence Information Technology vol 6 no 1 pp 79ndash82 2011
[10] W Fu and E L Tan ldquoDevelopment of split-step FDTDmethodwith higher-order spatial accuracyrdquo Electronics Letters vol 40no 20 pp 1252ndash1253 2004
[11] A Mohebbi andM Dehghan ldquoHigh-order compact solution ofthe one-dimensional heat and advection-diffusion equationsrdquoApplied Mathematical Modelling vol 34 no 10 pp 3071ndash30842010
[12] W Liao ldquoAn implicit fourth-order compact finite differencescheme for one-dimensional Burgersrsquo equationrdquo Applied Math-ematics and Computation vol 206 no 2 pp 755ndash764 2008
[13] Z F Tian and Y B Ge ldquoA fourth-order compact ADI methodfor solving two-dimensional unsteady convection-diffusionproblemsrdquo Journal of Computational and Applied Mathematicsvol 198 no 1 pp 268ndash286 2007
[14] C Chun ldquoSome fourth-order iterativemethods for solving non-linear equationsrdquo Applied Mathematics and Computation vol195 no 2 pp 454ndash459 2008
[15] S Zhu Z Yu and J Zhao ldquoAhigh-order parallel finite differencealgorithmrdquo Applied Mathematics and Computation vol 183 no1 pp 365ndash372 2006
[16] Z Gao and S Xie ldquoFourth-order alternating direction implicitcompact finite difference schemes for two-dimensionalSchrodinger equationsrdquo Applied Numerical Mathematics vol61 no 4 pp 593ndash614 2011
[17] M S Sahimi A Ahmad and A A Bakar ldquoThe Iterative Alter-nating Decomposition Explicit (IADE) method to solve theheat conduction equationrdquo International Journal of ComputerMathematics vol 47 pp 219ndash229 1993
[18] G D Smith Numerical Solution of Partial Differential Equa-tions Finite Difference Methods Oxford University Press 2ndedition 1978
[19] A RMitchell andG Fairweather ldquoImproved forms of the alter-nating direction methods of Douglas Peaceman and Rachfordfor solving parabolic and elliptic equationsrdquo Numerische Math-ematik vol 6 no 1 pp 285ndash292 1964
Journal of Applied Mathematics 13
[20] B N DattaNumerical Linear Algebra and Applications BrooksCole Publishing Pacific Grove Calif USA 1995
[21] D M Young Iterative Solution of Large Linear Systems Aca-demic Press New York NY USA 1971
[22] V K Saulev Integration of Equations of Parabolic Type by theMethod of Nets Pergamon Press Oxford UK 1964
[23] L W Johnson and R D Riess Numerical Analysis Addison-Wesley Reading Mass USA 2nd edition 1982
[24] D W Peaceman and H H Rachford Jr ldquoThe numerical solu-tion of parabolic and elliptic differential equationsrdquo Journal ofthe Society for Industrial and Applied Mathematics vol 3 pp28ndash41 1955
[25] A Pathirana S Tsegaye B Gersonius and K VairavamoorthyldquoA simple 2-D inundation model for incorporating flood dam-age in urban drainage planningrdquo Hydrology and Earth SystemSciences vol 15 no 8 pp 2747ndash2761 2011
[26] R Mirzavand A Abdipour G Moradi and M MovahhedildquoFull-wave semiconductor devices simulation using ADI-FDTD methodrdquo Progress in Electromagnetics Research vol 11pp 191ndash202 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
2 Journal of Applied Mathematics
be implemented in parallel and is found to be superiorto the corresponding full sweep alternating group explicit(AGE) and SOR methods Jin et al [9] proposed an AGEiteration method of the fourth-order accuracy by integratingthe grouping explicit method with numerical boundary con-ditionsThemethod was used to solve initial-boundary valueproblem of convection equations Fu and Tan [10] showedthat an unconditionally stable split-step FDTD method withhigher-order spatial accuracy is more accurate than thelower-order methods The dispersion error of the proposedmethod is comparable with the higher-order ADI-FDTDmethod
Higher-order methods are also studied by Mohebbi andDeghan [11] who applied a compact finite difference approx-imation of fourth-order and the cubic C1 spline colloca-tion method to some one-dimensional heat and advection-diffusion equations The scheme has fourth-order accuracyin both space and time and unconditionally stable Liao[12] proposed an efficient and high-order accuracy of thefourth-order compact finite difference method to solve one-dimensional Burgersrsquo equation Tian and Ge [13] studied astable fourth-order compact ADI method for solving two-dimensional unsteady convection diffusion problems Themethod is temporally second-order and spatially fourth-order accurate which requires only a regular five-point 2Dstencil similar to that in the standard second-order methodsChun [14] solved some nonlinear equations by applying thefourth-order iterative methods containing the Kingrsquos fourth-order family It was observed that the proposed methodhas at least equal performance compared to other methodsof the same order Zhu et al [15] presented a high-orderparallel finite difference algorithm based on the domaindecomposition strategy The study used the classical explicitscheme to calculate the interface values between subdomainsand the fourth-order compact schemes for the interior valuesThe method has high accuracy and is convergent and stableGao and Xie [16] devised a fourth-order alternating directionimplicit compact finite difference schemes for the solutionof two-dimensional Schrodinger equations The method ishighly competitive as compared to other existing methodsand it achieves the expected convergence rate
This study develops a higher-order finite difference algo-rithm with noncompact stencils that is capable of deliveringhighly accurate solutions for a one-dimensional heat equa-tion with Dirichlet boundary conditions The study focuseson modifying the unconditionally stable and convergentsecond-order iterative alternating decomposition explicit(IADE) method of Mitchell and Fairweather (IADEMF2)The IADEMF2 which was originally proposed by Sahimiet al [17] employs the fractional splitting of the Mitchell andFairweather (MF) variant that has an accuracy of the order119874((Δ119905)
2
+ (Δ119909)4
) The scheme executes a two-stage processinvolving the solution of sets of tridiagonal equations alonglines parallel to the first and second time steps respectivelyIt is found to be unconditionally stable convergent andmoreaccurate than the classical AGE class of methods namely theAGE method based on the Peaceman and Rachford variant(AGE-PR) whose spatial accuracy is of second order and
the AGEmethod based on the Douglas and Rachford variant(AGE-DR) whose temporal accuracy is only of first orderThe detailed derivation of the IADEMF2 can be obtainedfrom [17]
In this paper a fourth-order Crank-Nicolson (CN) dif-ference approximation is applied to the spatial derivative inthe heat equation and the MF variant is employed leadingto the formation of the fourth-order IADEMF (IADEMF4)numerical algorithm The convergence of the IADEMF4is analyzed and proven Numerical experiments verify thepotential of the IADEMF4 in enhancing the accuracy ofthe IADEMF2 The results of the proposed higher-orderscheme are also compared with the benchmarked fourth-order classical iterative methods such as the fourth-orderGauss-Seidel (GS4) the fourth-order Jacobi (JAC4) and thefourth-order successive over-relaxation (SOR4) methods
This paper is organized as follows Section 2 discusses theimplementation and stability of the fourth-order CN approx-imation on the heat equation In Section 3 the IADEMF4algorithm is formulated Section 4 analyses the convergenceof the IADEMF4 Section 5 provides the equations for thebenchmarked fourth-order classical iterative methods Thecomputational complexity of the methods considered in thispaper is given in Section 6 while the pseudocode for theIADEMF4 sequential algorithm is presented in Section 7Theexperiments conducted are given in Section 8 Sections 9and 10 provide some discussion and conclusion based on theobtained numerical results
2 Fourth-Order Crank-NicolsonApproximation to the Spatial Derivativein the Heat Equation
Consider the following one-dimensional parabolic heat equa-tion (1) that has been suitably assumed to be nondimensiona-lised It models the flow of heat in a homogeneous unchang-ing medium of finite extent in the absence of heat source
120597119880
120597119905=1205972
119880
1205971199092(1)
subject to given initial and Dirichlet boundary conditions
119880 (119909 0) = 119891 (119909) 0 le 119909 le 1
119880 (0 119905) = 119892 (119905) 0 lt 119905 le 119879
119880 (1 119905) = ℎ (119905) 0 lt 119905 le 119879
(2)
For the problem in (1) the finite difference approach dis-cretizes the time-space domain by placing a rectangular gridover the domain with grid spacing ofΔ119905 andΔ119909 in the 119905- and119909-directions respectively The grid consists of the set of linesparallel to the 119905-axis given by 119909
119894= 119894Δ119909 119894 = 0 1 119898119898 + 1
and a set of lines parallel to the 119909-axis given by 119905119896= 119896Δ119905 119896 =
0 1 119899 119899+1 For simplicity the grid spacing is taken to beuniform so thatΔ119909 = 1(119898+1) andΔ119905 = 119879(119899+1) At a grid-point119875(119909
119894 119905119896) in the solution domain the dependent variable
119880(119909 119905) which represents the nondimensional temperature attime 119905 and at position 119909 is approximated by 119906119896
119894
Journal of Applied Mathematics 3
At the grid-point 119875(119909119894 119905119896+12
) the IADEMF4 schemereplaces the spatial derivative in the heat equation with ahigher-order particularly the fourth-order Crank-Nicolson(CN) difference approximation [18] This is shown in theexpression given in (3) with the central difference operatorsdefined as 1205752
119909119906119896
119894= 119906119896
119894minus1minus 2119906119896
119894+ 119906119896
119894+1and 120575
2
119909119906119896+1
119894= 119906119896+1
119894minus1minus
2119906119896+1
119894+ 119906119896+1
119894+1 The approach gives the fourth-order scheme a
spatial truncation error of the order 119874(Δ119909)4 It enhances theaccuracy of its second-order counterpart which has a largererror of the order 119874(Δ119909)2
1
Δ119905(119906119896+1
119894minus 119906119896
119894) =
1
2(Δ119909)2(1205752
119909minus
1
121205754
119909) (119906119896+1
119894+ 119906119896
119894) (3)
To determine the stability of (3) the von Neumann sta-bility analysis can be applied Let 120582 = Δ119905(Δ119909)
2 and since1205754
119909119906119896
119894= 1205752
119909(1205752
119909119906119896
119894) the discretization of (3) becomes
119906119896+1
119894minus 119906119896
119894=120582
2(minus
1
12119906119896+1
119894minus2+4
3119906119896+1
119894minus1
minus5
2119906119896+1
119894+4
3119906119896+1
119894+1minus
1
12119906119896+1
119894+2
minus1
12119906119896
119894minus2+4
3119906119896
119894minus1minus5
2119906119896
119894
+4
3119906119896
119894+1minus
1
12119906119896
119894+2)
(4)
The discretized equation in (4) is assumed to have a solu-tion in the form of a Fourier harmonic function that is 119906119896
119894=
120588119896
119890119904120573119894Δ119909 where 120588 is referred to as the amplification factor 120573 is
an arbitrary constant and 119904 = radicminus1 The amplification factorrepresents the time dependence of the solution If the Fourierfunction is substituted into (4) and then solve for 120588 the resultwill be
120588 =
1 minus (1205826) ((cos120573 minus 4)2
minus 9)
1 + (1205826) ((cos120573 minus 4)2
minus 9)
with 120573 = 120573Δ119909 (5)
Since |120588| le 1 then the fourth-order CN approximation isunconditionally stable for any choice of 120573 Δ119905 and Δ119909
3 The Formulation of the IADEMF4
The IADEMF4 is firstly developed based on the execution ofthe unconditionally stable fourth-order CN approximation(3) on the heat equation
Equation (4) can also be expressed as in (6) using thedefinitions of the constants given in (7)
119886119906119896+1
119894minus2+ 119887119906119896+1
119894minus1+ 119888119906119896+1
119894+ 119889119906119896+1
119894+1+ 119890119906119896+1
119894+2
= minus119886119906119896
119894minus2minus 119887119906119896
119894minus1+ 119888119906119896
119894minus 119889119906119896
119894+1minus 119890119906119896
119894+2
119894 = 2 3 119898 minus 1
(6)
119886 =120582
24 119887 = minus
2120582
3 119888 =
4 + 5120582
4
119889 = minus2120582
3 119890 =
120582
24 119888 =
4 minus 5120582
4
(7)
The approximation in (6) can be displayed in a matrix formsuch as 119860u = f (8) where 119860 is a sparse pentadiagonalcoefficient matrix and u = (119906
2 1199063 119906
119898minus2 119906119898minus1
)119879 is the
columnvector containing the unknownvalues of119906 at the timelevel 119896 + 1 The column vector f = (119891
2 1198913 119891
119898minus2 119891119898minus1
)119879
consists of boundary values and known 119906 values at theprevious time level 119896 The definitions for every entry in f aregiven in (9)
119860u = f
[[[[[[[[[
[
119888 119889 119890
119887 119888 119889 119890 O119886 119887 119888 119889 119890
d d d d119886 119887 119888 119889 119890
O 119886 119887 119888 119889
119886 119887 119888
]]]]]]]]]
](119898minus2)times(119898minus2)
[[[[[[
[
1199062
1199063
119906119898minus2
119906119898minus1
]]]]]]
]119896+1
=
[[[[[[
[
1198912
1198913
119891119898minus2
119891119898minus1
]]]]]]
]
(8)
1198912= minus119887 (119906
119896
1+ 119906119896+1
1) + 119888119906
119896
2minus 119889119906119896
3minus 119890119906119896
4
1198913= minus119886 (119906
119896
1+ 119906119896+1
1) minus 119887119906
119896
2+ 119888119906119896
3minus 119889119906119896
4minus 119890119906119896
5
119891119894= minus119886119906
119896
119894minus2minus 119887119906119896
119894minus1+ 119888119906119896
119894minus 119889119906119896
119894+1minus 119890119906119896
119894+2
for 119894 = 4 5 119898 minus 3
119891119898minus2
= minus119886119906119896
119898minus4minus 119887119906119896
119898minus3+ 119888119906119896
119898minus2
minus 119889119906119896
119898minus1minus 119890 (119906
119896
119898+ 119906119896+1
119898)
119891119898minus1
= minus119886119906119896
119898minus3minus 119887119906119896
119898minus2+ 119888119906119896
119898minus1minus 119889 (119906
119896
119898+ 119906119896+1
119898)
(9)
The evaluations of 1198912 1198913 119891119898minus2
and 119891119898minus1
require thevalues of 119906 at the boundaries 119894 = 1 and 119894 = 119898 Howeverthese values cannot be obtained numerically because theircomputations involve nodes at 119894 = minus1 and 119894 = 119898 + 2which are exterior to the considered solution domain Ifthe exact solutions of 119906 are available at 119894 = 1 and 119894 =
119898 then it is appropriate to consider them as the requiredboundary values Otherwise the boundary conditions haveto be formulated bearing in mind that they should be ofcomparable accuracy [1]
The IADEMF4 scheme secondly employs the higher-order accuracy formula of MF [19] The variant whose accu-racy is of the order 119874((Δ119905)2 + (Δ119909)
4
) is as given in (10) and(11)
(119903119868 + 1198661) u(119901+12) = (119903119868 minus 119892119866
2) u(119901) + f (10)
(119903119868 + 1198662) u(119901+1) = (119903119868 minus 119892119866
1) u(119901+12) + 119892f (11)
where 119903 119901 and 119868 represent an acceleration parameter theiteration index and an identity matrix respectively 119866
1and
1198662are two constituent matrices The vectors u(119901+1) and
u(119901+12) represent the required solution at the iteration level(119901 + 1) and at some intermediate level (119901 + 12) respectivelyThe relation of 119892 and 119903 is given by 119892 = (6 + 119903)6 119903 gt 0
4 Journal of Applied Mathematics
Substitute the following expression obtained from (10)into (11)
u(119901+12) = (119903119868 + 1198661)minus1
[(119903119868 minus 1198921198662) u(119901) + f] (12)
and then simplify to obtain
(119903119868 + 1198662) u(119901+1) = (119903119868 minus 119892119866
1) (119903119868 + 119866
1)minus1
(119903119868 minus 1198921198662) u(119901)
+ [ (119903119868 minus 1198921198661) (119903119868 minus 119892119866
1)minus1
+ 119892] f (13)
As 119901 in (13) becomes sufficiently large the temperaturesolution reaches a steady state that is as 119901 rarr infin thenu(119901+1) rarr u and u(119901) rarr u Simplify the equation and thenmultiply by (119903119868+119866
1) After some algebraicmanipulations the
following expression is obtained
[119903 (1198661+ 1198662) (1 + 119892) + (1 minus 119892
2
)11986611198662] u = 119903 (1 + 119892) f (14)
Multiply (14) by 1119903(1 + 119892) and use the definition of 119892 tofinally obtain the following form
[ (1198661+ 1198662) minus
1
611986611198662] u = f (15)
The comparison between the expression 119860u = f in (8)and the form given in (15) suggests that the coefficient matrix119860 for the IADEMF4 can be decomposed into
119860 = 1198661+ 1198662minus1
611986611198662 (16)
The IADEMF4 requires the constituentmatrices1198661and119866
2in
(16) to be in the form of lower and upper tridiagonalmatricesrespectively in order to retain the pentadiagonal structure of119860 Thus
1198661=
[[[[[[[
[
1
1198971
1 119874
1
1198972
d2d dd 119897119898minus4
1
119874 119898minus4
119897119898minus3
1
]]]]]]]
](119898minus2)times(119898minus2)
1198662=
[[[[[[[
[
11989011
V1
1198902
2
V2
119874
d d d119874 119890
119898minus4119898minus4
V119898minus4
119890119898minus3
119898minus3
119890119898minus2
]]]]]]]
](119898minus2)times(119898minus2)
(17)
By substituting 1198661and 119866
2in (17) into the formula for the
decomposition of 119860 the entries of the resultant matrix arecompared with the entries of 119860 in (8) yielding the followingdefinitions
1198901=6 (119888 minus 1)
5
1=6119889
5 119897
1=
6119887
6 minus 1198901
1198902=6 (119888 minus 1) + 119897
11
5 V
119894=6119890
5for 119894 = 1 2 119898 minus 4
(18)
And for 119894 = 2 3 119898 minus 3
119894=6119889 + 119897
119894minus1V119894minus1
5
119894minus1=
6119886
6 minus 119890119894minus1
119897119894=6119887 +
119894minus1119894minus1
6 minus 119890119894
119890119894+1
=6 (119888 minus 1) + 119897
119894119894+ 119894minus1
V119894minus1
5
(19)
Since1198661and119866
2are three bandedmatrices then (119903119868+119866
1)
and (119903119868+1198662) can be inverted easilyThe equations in (10) and
(11) are rearranged as in (20) and (21) respectively
u(119901+12) = (119903119868 + 1198661)minus1
(119903119868 minus 1198921198662) u(119901)
+ (119903119868 + 1198661)minus1f
(20)
u(119901+1) = (119903119868 + 1198662)minus1
(119903119868 minus 1198921198661) u(119901+12)
+ 119892(119903119868 + 1198662)minus1f
(21)
The above two equations are computed and simplifiedleading to the computational formulae at each of the halfiteration levels as given in (22) and (23)
(i) At the (119901 + 12) iteration level
119906(119901+12)
119894=
1
119877(119864119894minus1
119906(119901)
119894+119882119894minus1
119906(119901)
119894+1+ 119881119894minus1
119906(119901)
119894+2
minus119894minus3
119906(119901+12)
119894minus2minus 119897119894minus2
119906(119901+12)
119894minus1+ 119891119894)
119894 = 2 3 119898 minus 2119898 minus 1
(22)
(ii) At the (119901 + 1) iteration level
119906(119901+1)
119894=
1
119885119894minus1
(119878119894minus3
119906(119901+12)
119894minus2+ 119876119894minus2
119906(119901+12)
119894minus1+ 119875119906(119901+12)
119894
minus119894minus1
119906(119901+1)
119894+1minus V119894minus1
119906(119901+1)
119894+2+ 119892119891119894)
119894 = 119898 minus 1 119898 minus 2 3 2
(23)
with
minus1
= 0= 1198970= 119881119898minus2
= 119881119898minus3
= 119882119898minus2
= 119898minus2
= V119898minus2
= V119898minus3
= 1198760= 119878minus1
= 1198780= 0
119877 = 1 + 119903 119875 = 119903 minus 119892
119864119894= 119903 minus 119892119890
119894 119885119894= 119903 + 119890
119894 119894 = 1 2 119898 minus 2
119882119894= minus119892
119894 119876119894= minus119892119897
119894 119894 = 1 2 119898 minus 3
119881119894=minus119892V119894 119878119894= minus119892
119894 119894 = 1 2 119898 minus 4
(24)
The IADEMF4 algorithm is regarded as a two-stageprocess involving two iteration levels (119901 + 12) and (119901 + 1)It is completed explicitly by using the required equations atthe two levels in alternate sweeps along all the grid pointsin the interval (0 1) until convergence is reached In (22)the calculation to determine the unknown 119906
(119901+12)
119894begins at
the left boundary and then moves to the right In a similar
Journal of Applied Mathematics 5
manner 119906(119901+1)119894
in (23) is calculated by proceeding from theright boundary towards the left (Figure 1)
The computational molecules are depicted in Figures 2and 3
At each level of iteration the computational moleculesinvolve two known grid points at the new level and anotherthree known ones at the old level Clearly the method isexplicit
4 Convergence Analysis of the IADEMF4
This section proves the convergence of the IADEMF4 Since120582 gt 1will not guarantee an accurate approximation for 120597119880120597119905[18] the values of 120582 that are considered appropriate for theproof are 0 lt 120582 le 1
From the definitions in ((18) and (19)) the followingresults are obtained
1198901=6 (119888 minus 1)
5=3120582
2 implying that 0 lt 119890
1le3
2
1=6119889
5lt0 119897
1=
6119887
6 minus 1198901
lt0 V119894=6119890
5gt 0
for 119894 = 1 2 119898 minus 4
2=6119889 + 119897
1V1
5= 1+1198971V1
5lt 1 since 119897
1V1lt 0
1198902=6 (119888 minus 1) + 119897
11
5= 1198901+11989711
5gt 1198901 since 119897
11gt 0
(25)
Simple computation shows that 119897115 lt 1 Clearly 119890
2lt 6
Lemma 1 119890119894in ((18) and (19)) is such that 0 lt 119890
119894lt 6 for all
119894 = 1 2 3 119898 minus 2
Proof The results in (25) show that 6 gt 1198902gt 1198901gt 0 Assume
it is true that 6 gt 119890119896gt 119890119896minus1
gt 0 for all 119896 = 3 4 5 119898 minus 3Since the assumption implies that 6 gt 119890
119896minus1gt 119890119896minus2
gt 0 then119896minus1
= 6119886(6 minus 119890119896minus1
) gt 119896minus2
= 6119886(6 minus 119890119896minus2
) gt 0 Therefore119896minus1
V119896minus1
gt 119896minus2
V119896minus2
It has also been shown that
2lt 1lt 0 Assume it is true
that 119895lt 119895minus1
lt 0 for all 119895 = 3 4 119898 minus 4 The assumptionimplies that
119895minus1lt 119895minus2
lt 0 Thus 119895minus1
119895minus1
lt 119895minus2
119895minus2
lt 0
119895+1
=6119889 + 119897
119895V119895
5= 1+119897119895V119895
5= 1+
(6119887 + 119895minus1
119895minus1
) V119895
5 (6 minus 119890119895)
lt 1+
(6119887 + 119895minus2
119895minus2
) V119895minus1
5 (6 minus 119890119895minus1
)
= 1+119897119895minus1
V119895minus1
5= 119895lt 0
(26)
Since it is true that for 119895 + 1 119895+1
lt 119895lt 0 then by
induction the assumption 119895lt 119895minus1
lt 0 is true for all 119895 An
0
u(p+12)
i
u(p+1)
i
i iminus 1 i + 1 m + 1
Figure 1 The two-stage IADEMF4 algorithmThe directions of thesweeps at the (119901 + 12) and (119901 + 1) iteration levels
i iminus 1i minus 2 i + 1 i + 2
(p + 12)
(p)
Figure 2 Computational molecule of the IADEMF4 at the (119901+12)iteration level
i iminus 1i minus 2 i + 1 i + 2
(p + 12)
(p + 1)
Figure 3 Computational molecule of the IADEMF4 at the (119901 + 1)
iteration level
equivalent to the preceding statement would be 119896lt 119896minus1
lt 0
for all 119896 = 3 4 5 119898 minus 3 It follows that
119897119896119896=(6119887 +
119896minus1119896minus1
) 119896
6 minus 119890119896
gt(6119887 +
119896minus2119896minus2
) 119896minus1
6 minus 119890119896minus1
= 119897119896minus1
119896minus1
119890119896+1
=6 (119888 minus 1) + 119897
119896119896+ 119896minus1
V119896minus1
5= 1198901+119897119896119896+ 119896minus1
V119896minus1
5
gt 1198901+119897119896minus1
119896minus1
+ 119896minus2
V119896minus2
5= 119890119896gt 0
(27)
Since it is true that for 119896+1 119890119896+1
gt 119890119896gt 0 then by induction
119890119896gt 119890119896minus1
gt 0 is also true for all 119896Suppose there is a 119896 such that 119890
119896gt 6 Then
119890119896+1
= 1198901+119897119896119896+ 119896minus1
V119896minus1
5
= 1198901+
6119887 + 119896
5 (6 minus 119890119896)+119896minus1
5(119896minus1
119896
6 minus 119890119896
+ V119896minus1
) lt 1198901
(28)
This is a contradiction since 119890119896+1
gt 1198901 This verifies that 6 gt
119890119896gt 119890119896minus1
gt 0 119896 = 3 4 5 119898 minus 3 Let 119894 = 2 3 4 119898 minus 2and then 6 gt 119890
119894gt 119890119894minus1
gt 0 Therefore 0 lt 119890119894lt 6 119894 =
1 2 3 119898 minus 2 Lemma 1 is proved
6 Journal of Applied Mathematics
Lemma 2 If 119903 gt 0 and 119892 = (6+ 119903)6 thenmax |(119903 minus119892119890119894)(119903 +
119890119894)| lt 1 i = 1 2 3 119898 minus 2
Proof Let max |(119903 minus 119892119890119894)(119903 + 119890
119894)| = |(119903 minus 119892119890
119895)(119903 + 119890
119895)| for
some 119895Assume |(119903 minus 119892119890
119895)(119903 + 119890
119895)| ge 1 Since 119903 gt 0 and 119890
119895gt 0
then 119903 + 119890119895gt 0
If (119903 minus 119892119890119895)(119903 + 119890
119895) ge 1 then minus((6 + 119903)6)119890
119895ge 119890119895 which
implies that 119903 le minus12 This contradicts the fact that 119903 gt 0If (119903 minus 119892119890
119895)(119903 + 119890
119895) le minus1 then 2119903 minus ((6 + 119903)6)119890
119895le minus119890119895
which implies that 119890119895ge 12 This contradicts Lemma 1
So the assumption that |(119903 minus 119892119890119895)(119903 + 119890
119895)| ge 1 is false
Hence max |(119903 minus 119892119890119894)(119903 + 119890
119894)| = |(119903 minus 119892119890
119895)(119903 + 119890
119895)| lt 1
Proposition 3 (119903119868 minus 1198921198661)(119903119868 + 119866
1)minus1
2lt 1
Proof Let 119865 = (119903119868 minus 1198921198661)(119903119868 + 119866
1)minus1 that is
119865 =
[[[[[[[[[[
[
119903 minus 119892
119889
d119903 minus 119892
119889119874
d d d
d d
sdot sdot sdot sdot sdot sdot sdot sdot sdot sdot sdot sdot119903 minus 119892
119889
]]]]]]]]]]
]
(29)
119865 is a lower triangular matrix with all the diagonal entries(whence the eigenvalues of 119865) equal to (119903 minus 119892)119889 where 119889 =
119903 + 1 Denote all the eigenvalues of 119865 by 120582119865 If 120588[119865] is defined
as the spectral radius of 119865 then
120588 [119865] = max 10038161003816100381610038161205821198651003816100381610038161003816 =
10038161003816100381610038161205821198651003816100381610038161003816 =
1003816100381610038161003816100381610038161003816
119903 minus 119892
119903 + 1
1003816100381610038161003816100381610038161003816le 119865
2 (30)
But by definition of 2-norm
1198652= maxx2 = 0
119865x2
x2
= maxy2=1
1003817100381710038171003817119865y10038171003817100381710038172
1003817100381710038171003817y10038171003817100381710038172
= maxy2=1
1003817100381710038171003817119865y10038171003817100381710038172
le maxw2=1
119865w2= maxw2=1
1003817100381710038171003817120582119865w10038171003817100381710038172
(31)
Since all the eigenvalues of 119865 are equal then
1198652le1003816100381610038161003816120582119865
1003816100381610038161003816 maxw2=1
w2=1003816100381610038161003816120582119865
1003816100381610038161003816 = 120588 [119865] (32)
Thus from (30) and (32) 1198652= |120582119865| = |(119903 minus 119892)(119903 + 1)|
By Lemma 2 with 119890119895replaced by 1 |(119903 minus 119892)(119903 + 1)| lt 1
is obtained leading to the result of Proposition 3 which is1198652= (119903119868 minus 119892119866
1)(119903119868 + 119866
1)minus1
2lt 1
Proposition 4 (119903119868 minus 1198921198662)(119903119868 + 119866
2)minus1
2lt 1
Proof Let 119870 = (119903119868 minus 1198921198662)(119903119868 + 119866
2)minus1 that is
119870 =
[[[[[[[[[[[[[
[
119903 minus 1198921198901
1199111
d sdot sdot sdot sdot sdot sdot sdot sdot sdot
119903 minus 1198921198902
1199112
d sdot sdot sdot
d
119874 d d
119903 minus 119892119890119898
119911119898
]]]]]]]]]]]]]
]
(33)
119870 is an upper triangular matrix with all the diagonal entriesequal to (119903 minus 119892119890
119894)119911119894 119894 = 1 2 119898 minus 2 where 119911
119894= 119903 + 119890
119894
Since all the eigenvalues of119870 are distinct then119870 is similar toa diagonal matrix119863
By the Schur triangularization theorem [20] there is anorthogonal matrix 119874 such that 119874119879119870119874 = 119863 The diagonalentries of119863 are the eigenvalues of119870
1198702=10038171003817100381710038171003817119874119879
119870119874100381710038171003817100381710038172
= 1198632
= radicmaximum eigenvalue of 119863119879119863
= radicmaximum eigenvalue of 1198632
= radic120588 (1198632) = radic120588 (1198702) = 120588 (119870)
(34)
By Lemma 2
1198702= 120588 (119870) = max
10038161003816100381610038161003816100381610038161003816
119903 minus 119892119890119894
119903 + 119890119894
10038161003816100381610038161003816100381610038161003816
=
1003816100381610038161003816100381610038161003816100381610038161003816
119903 minus 119892119890119895
119903 + 119890119895
1003816100381610038161003816100381610038161003816100381610038161003816
lt 1 (35)
for some 119895 Hence this proves Proposition 4
From (20) and (21)
u(119901+1) = (119903119868 + 1198662)minus1
(119903119868 minus 1198921198661)
times [(119903119868 + 1198661)minus1
(119903119868 minus 1198921198662) u(119901) + (119903119868 + 119866
1)minus1f]
+ 119892 (119903119868 + 1198662)minus1f
(36)
Let
119872(119903) = (119903119868 + 1198662)minus1
(119903119868 minus 1198921198661) (119903119868 + 119866
1)minus1
(119903119868 minus 1198921198662)
(37)
And let
119902 (119903) = [(119903119868 + 1198662)minus1
(119903119868 minus 1198921198661) (119903119868 + 119866
1)minus1
+119892(119903119868 + 1198662)minus1
] f (38)
Then u(119901+1) = 119872(119903)u(119901) + 119902(119903)
Theorem 5 The IADEMF4 is convergent if 120588[119872(119903)] lt 1 for119903 gt 0
Journal of Applied Mathematics 7
Proof Define (119903) = (119903119868 + 1198662)119872(119903)(119903119868 + 119866
2)minus1 then
(119903) = (119903119868 minus 1198921198661) (119903119868 + 119866
1)minus1
(119903119868 minus 1198921198662) (119903119868 + 119866
2)minus1
(39)
Thus by similarity119872(119903) and (119903) have the same set of eigen-values Therefore
120588 [119872 (119903)]
= 120588 [ (119903)] le10038171003817100381710038171003817 (119903)
10038171003817100381710038171003817 2
=10038171003817100381710038171003817(119903119868 minus 119892119866
1) (119903119868 + 119866
1)minus1
(119903119868 minus 1198921198662) (119903119868 + 119866
2)minus110038171003817100381710038171003817 2
le10038171003817100381710038171003817(119903119868 minus 119892119866
1) (119903119868 + 119866
1)minus1100381710038171003817100381710038172
10038171003817100381710038171003817(119903119868 minus 119892119866
2) (119903119868 + 119866
2)minus110038171003817100381710038171003817 2
lt 1
(40)
The last inequality is due to Propositions 3 and 4The provenTheorem 5 assures the convergence of the IADEMF4
5 The Fourth-Order ClassicalIterative Methods
The system of linear equations in (8) may also be solvedby using the classical iterative methods of the fourth orderThey include the JAC4 the GS4 and the SOR4 methodsThese iterative methods are capable of exploiting the sparsestructure of the pentadiagonal matrix
The JAC4 algorithm can be represented by
119906(119901+1)
119894=
(119891119894minus119886119906(119901)
119894minus2minus 119887119906(119901)
119894minus1minus 119889119906(119901)
119894+1minus 119890119906(119901)
119894+2)
119888
119894 = 2 3 119898 minus 1
(41)
The approximation of 119906(119901+1)119894
at the (119901 + 1)th iteration level iscomputed using the relevant values in the 119901th iteration
The GS4 method uses the most recent values of 119906(119901+1)119894minus2
and 119906(119901+1)
119894minus1to update the approximation value of 119906(119901+1)
119894 The
algorithm for the GS4 is as expressed in (42)
119906(119901+1)
119894=
(119891119894minus119886119906(119901+1)
119894minus2minus 119887119906(119901+1)
119894minus1minus 119889119906(119901)
119894+1minus 119890119906(119901)
119894+2)
119888
119894 = 2 3 119898 minus 1
(42)
The SOR4 iterative method accelerates the convergencerate of the GS4 If 120596 is a relaxation parameter then for any120596 = 0 (42) can be rewritten as
119906(119901+1)
119894= (1 minus 120596) 119906
(119901)
119894
+
120596 (119891119894minus 119886119906(119901+1)
119894minus2minus 119887119906(119901+1)
119894minus1minus 119889119906(119901)
119894+1minus 119890119906(119901)
119894+2)
119888
119894 = 2 3 119898 minus 1
(43)
i iminus 1i minus 2 i + 1 i + 2
(p + 1)
(p)
Figure 4 Computational molecule of the JAC4
i iminus 1i minus 2 i + 1 i + 2
(p + 1)
(p)
Figure 5 Computational molecule of the GS4SOR4
Table 1 Sequential arithmetic operations per iteration (119898 problemsize)
Method Number ofadditions
Number ofmultiplications
Total operationcount
IADEMF4 10(119898 minus 2) 13(119898 minus 2) 23(119898 minus 2)
IADEMF2 6119898 9119898 15119898
JAC4GS4 4(119898 minus 2) 5(119898 minus 2) 9(119898 minus 2)
SOR4 5(119898 minus 2) 7(119898 minus 2) 12(119898 minus 2)
The SOR4 algorithm reduces to the GS4 if 120596 = 1 Exceptfor very special cases it is difficult to obtain the analyticexpression for 120596 According to Young [21] the precise deter-mination of the optimal 120596 in the SOR is only known for asmall class of matrices The value of 120596 generally lies in therange of 1 lt 120596 lt 2
The computational molecules for the JAC4 and the GS4are as illustrated in Figures 4 and 5 respectively The SOR4has the same form as the GS4
6 Computational Complexity
The cost of implementing an algorithm can be assessedby examining its computational complexity The number ofarithmetic operations such as additions (including subtrac-tions) andmultiplications (including divisions) that is neededto perform by the algorithm can be straightforwardly count-ed Table 1 gives the number of sequential arithmetic opera-tions per iteration that is required to evaluate the algorithms
For the higher-order schemes trade-offbetween accuracyand speed usually happens The computational complexityhas an effect on the efficiency of a particular scheme Thusthis factor will be taken into account when discussing theresults of the numerical experiments conducted in this study
7 The Pseudocode of the IADEMF4Sequential Algorithm
Algorithm 1 illustrates the pseudocode of the IADEMF4sequential algorithm This algorithm can also be generalizedand applied to other methods discussed in this paper
8 Journal of Applied Mathematics
begindetermine the parameters119898 119903 Δ119905 Δ119909 120582 120576 119879
determine initial conditions 119906(119901)119894
determine exact solutions 119880119894at time level 119879
while (time level lt 119879)determine boundary conditionsfor 119894 = 2 to 119894 = 119898 minus 1
compute 119891119894(refer to (9))
end-forset iteration = 0while (convergence conditions are not satisfied)
for 119894 = 2 to 119894 = 119898 minus 1
compute 119906(119901+12)119894
(refer to (22))end-forfor 119894 = 2 to 119894 = 119898 minus 1
compute 119906(119901+1)119894
(refer to (23))end-fortest for convergencecompute 119890
119894larr
10038161003816100381610038161003816119906(119901+1)
119894minus 119906(119901)
119894
10038161003816100381610038161003816if (max1003816100381610038161003816119890119894
1003816100381610038161003816 lt 120576)then 119906
(119901)
119894larr 119906(119901+1)
119894
add 1 to iteration (if necessary)end-while
end-whilefor 119894 = 2 to 119894 = 119898 minus 1
determine average absolute error root meansquare error and maximum error
end-forend
Algorithm 1 The IADEMF4 sequential algorithm
8 Numerical Experiments
Two experiments were conducted to test the sequentialnumerical performance of the proposed IADEMF4 methodagainst those of the benchmarked classical iterative methodsnamely the JAC4 theGS4 and the SOR4 Comparison is alsomade with the corresponding IADE method of the secondorder
Experiment 1 This problem was taken from Saulev [22]
120597119880
120597119905=1205972
119880
1205971199092 0 le 119909 le 1 (44)
subject to the initial condition
119880 (119909 0) = 4119909 (1 minus 119909) 0 le 119909 le 1 (45)
and the boundary conditions
119880 (0 119905) = 119880 (1 119905) = 0 119905 ge 0 (46)
The exact solution to the given problem is given by
119880 (119909 119905) =32
1205873
infin
sum
119896=1(2)
1
1198963119890minus12058721198962119905 sin (119896120587119909) (47)
Experiment 2 This problem was taken from Johnson andReiss [23]
120597119880
120597119905=1205972
119880
1205971199092 0 le 119909 le 1 (48)
subject to the initial condition
119880 (119909 0) = sin (120587119909) (1 + 6 cos (120587119909)) 0 le 119909 le 1 (49)
and the boundary conditions
119880 (0 119905) = 119880 (1 119905) = 0 119905 ge 0 (50)
The exact solution to the given problem is given by
119880 (119909 119905) = sin (120587119909) 119890minus1205872119905
+ 3 sin (2120587119909) 119890minus41205872119905
(51)
Journal of Applied Mathematics 9
9 Results and Discussion
In each experiment the exact solutions at 119894 = 1 and 119894 = 119898
were taken as boundary values for the IADEMF4 and theother fourth-order classical iterative methods The conver-gence criterion used in the testing of each method was takenas 119906(119901+1) minus 119906
(119901)
infin
le 120576 where 120576 is the convergence toleranceThe selections of the optimum 119903 or 120596 were determined byexperiments As for the IADEMF2 the CN scheme with 120579 =
12 was employedFigures 6 7 8 and 9 visualize the behavior of the one-
dimensional parabolic heat solutions for both experimentsBy using119898 = 10 and a tolerance requirement of 120576 = 10
minus4 twodifferent mesh sizes 120582 = 05 and 120582 = 10 were consideredin each experiment The exact solutions are compared withthe IADEMF4 and the IADEMF2 numerical solutions Theoptimum value of 119903 chosen for each method as well as theircorresponding outcome of the number of iterations 119899 isstated in the legend of each figure Every figure reveals thatthe IADEMF4 is more accurate than the IADEMF2 andthe former converges with fewer numbers of iterations incomparison to the latter The numerical solutions of theIADEMF4 seem to be in very good agreement with theexact solutions For example in Figure 7 at 119909 = 05 thedifference between the exact solution and the numericalsolution using the IADEMF2 and using the IADEMF4 isabout 36 and 0 respectively These results imply thatthe accuracy and convergence rate of the second-order IADEmethod are enhanced by the implementation of the fourth-order CN approximation that leads to the formation of thecorresponding fourth-order IADE scheme
Figure 10 displays the graph of log (RMSE) versus log(Δ119909)for decreasing values of Δ119909 implemented on Experiment 2By considering 120576 = 10
minus4 and fixing Δ119905 = 00001 the value ofΔ119909 was initially taken as Δ119909 = 0125 It was then successivelyhalved into Δ1199092 Δ1199094 and Δ1199098
The figure shows that amongst the tested methods theroot mean square error (RMSE) of the IADEMF4 and theIADEMF2 decreases linearly as the values of Δ119909 decreasesThe slope of the IADEMF4 is approximately equal to 4which corresponds to its fourth-order spatial accuracy TheIADEMF2 with a second-order spatial accuracy has a slopethat is approximately equal to 2 It is clear that the IADEMF4is always more accurate than the IADEMF2 for the differentconsidered mesh sizes The graphs of the SOR4 (120596 = 105)the GS4 and the JAC4 show that their accuracies of fourthorder tend to lack as Δ119909 decreases largely due to the effect ofincreasing round-off errors as the value of119898 increases
Tables 2 3 4 5 6 7 8 and 9 provide numerical results interms of the average absolute error (AAE) root mean squareerror (RMSE) maximum error (ME) number of iterations(119899) and execution time (ET) measured in seconds (s) Theresults are obtained from both experiments for two differentvalues of 119898 and mesh size 120582 It is generally observed thatwhen 119898 = 700 120582 = 05 and 120576 = 10
minus6 the IADEMF4has the least average absolute error root mean square errorand maximum error in comparison with the other methodsunder consideration (Tables 2ndash5) When the size of 119898 was
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 08 n = 172)IADEMF2 (r = 03 n = 185)
00E + 00
10E minus 02
20E minus 02
30E minus 02
40E minus 02
50E minus 02
60E minus 02
70E minus 02
80E minus 02
90E minus 02
10E minus 01
U(xt)
Figure 6 Numerical and exact solutions for Experiment 1 120582 = 05Δ119909 = 01 Δ119905 = 0005 119905 = 025 and 120576 = 10
minus4
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 14 n = 159)IADEMF2 (r = 05 n = 162)
00E + 00
10E minus 03
20E minus 03
30E minus 03
40E minus 03
50E minus 03
60E minus 03
70E minus 03
80E minus 03
90E minus 03
U(xt)
Figure 7 Numerical and exact solutions for Experiment 1 120582 = 10Δ119909 = 01 Δ119905 = 001 119905 = 05 and 120576 = 10
minus4
ten times bigger (119898 = 7000) and a more stringent tolerancecriterion was set (120576 = 10
minus10
) the accuracy of the IADEMF4clearly outperforms the other methods for a mesh size of120582 = 05 (Tables 6 and 8) However the difference in the errorsamongst the tested methods is not so obvious for the case of120582 = 10 (Tables 7 and 9)The achievement of the fourth-orderIADEmethod in Experiments 1 and 2 can be clearly seen fromthe results in Tables 2 and 8 respectively where it has caused ahuge 94 reduction of RMSE from its corresponding second-order IADE method
The IADEMF4 has the advantage of featuring higheraccuracy due to the execution of the fourth-order CNapprox-imation coupled with the fourth-order accurate MF variant
10 Journal of Applied Mathematics
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 10 n = 169)IADEMF2 (r = 03 n = 213)
00E + 00
10E minus 02
20E minus 02
30E minus 02
40E minus 02
50E minus 02
60E minus 02
70E minus 02
80E minus 02
90E minus 02
10E minus 01
U(xt)
Figure 8 Numerical and exact solutions for Experiment 2 120582 = 05Δ119909 = 01 Δ119905 = 0005 119905 = 025 and 120576 = 10
minus4
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 10 n = 175)IADEMF2 (r = 05 n = 182)
00E + 00
10E minus 03
20E minus 03
30E minus 03
40E minus 03
50E minus 03
60E minus 03
70E minus 03
80E minus 03
U(xt)
Figure 9 Numerical and exact solutions for Experiment 2 120582 = 10Δ119909 = 01 Δ119905 = 001 119905 = 05 and 120576 = 10
minus4
The IADEMF2 is only derived from the second-order CNapproximation but its combination with its correspondingfourth-order MF variant managed to produce errors whichare better than theGS4 JAC4 and the SOR4 Even though theclassical iterative methods are also derived from the fourth-order accurate CN type approximation they are lackingin accuracy due to the round-off errors that have beenaccumulated from the time the execution starts till it ends
With regards to the rate of convergence the resultsfrom each table demonstrate that the number of iterationsproduced by the IADEMF2 and the fourth-order classicaliterative methods is greater than or at least equal to that of
0
minus2
minus208 minus277 minus347 minus416
minus4
minus8
minus6
minus10
minus12
minus14
minus16
minus18
GS4JAC4SOR4
log
(RM
SE)
log(Δx)
IADEMF4
IADEMF2
Figure 10 Log(RMSE) versus Log(Δ119909) Experiment 2 Δ119905 = 00001119905 = 0002 and 120576 = 10
minus4
Table 2 Experiment 1 119898 = 700 120582 = 05 Δ119909 = 143 times 10minus3 Δ119905 =
102 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 07) 267119890 minus 9 330119890 minus 9 212119890 minus 8 100 440119890 minus 2
IADEMF2(119903 = 08) 188119890 minus 8 495119890 minus 8 323119890 minus 7 100 635119890 minus 3
SOR4(120596 = 12) 540119890 minus 7 544119890 minus 7 552119890 minus 7 108 460119890 minus 2
GS4 540119890 minus 6 543119890 minus 6 332119890 minus 6 150 472119890 minus 2
JAC4 228119890 minus 5 229119890 minus 5 232119890 minus 5 150 473119890 minus 2
Table 3 Experiment 1 119898 = 700 120582 = 10 Δ119909 = 143 times 10minus3 Δ119905 =
204 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 297119890 minus 7 301119890 minus 7 301119890 minus 7 50 201119890 minus 2
IADEMF2(119903 = 09) 557119890 minus 7 559119890 minus 7 848119890 minus 7 50 317119890 minus 3
SOR4(120596 = 12) 329119890 minus 6 331119890 minus 6 335119890 minus 6 76 225119890 minus 2
GS4 895119890 minus 6 900119890 minus 6 900119890 minus 6 100 238119890 minus 2
JAC4 214119890 minus 5 215119890 minus 5 217119890 minus 5 125 250119890 minus 2
the IADEMF4 The convergence rate of the latter surpassesthe others in the case of 119898 = 7000 and 120576 = 10
minus10 (Tables6ndash8) Even though the operational count of the IADEMF4is relatively quite large (Table 1) due to its higher level ofaccuracy its increasing number of correct digits at eachiteration causes it to converge at a faster rate The IADEMF2has the advantage of having less mathematical operationsthus it is competitive in terms of convergence rate but atthe expense of accuracy The application of the fourth-orderCN approximation on the heat equation proves that thethree classical iterative methods also converge with JAC4appearing to be the slowest and the least accurate amongst all
Journal of Applied Mathematics 11
Table 4 Experiment 2 119898 = 700 120582 = 05 Δ119909 = 143 times 10minus3 Δ119905 =
102 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME n ET (s)IADEMF4(119903 = 07) 207119890 minus 8 231119890 minus 8 345119890 minus 8 100 903119890 minus 3
IADEMF2(119903 = 08) 106119890 minus 7 175119890 minus 7 175119890 minus 7 100 647119890 minus 3
SOR4(120596 = 11) 154119890 minus 6 171119890 minus 6 255119890 minus 6 200 110119890 minus 2
GS4 294119890 minus 6 327119890 minus 6 487119890 minus 6 250 118119890 minus 2
JAC4 124119890 minus 5 138119890 minus 5 206119890 minus 5 300 127119890 minus 2
Table 5 Experiment 2 119898 = 700 120582 = 10 Δ119909 = 143 times 10minus3 Δ119905 =
204 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 09) 138119890 minus 6 154119890 minus 6 229119890 minus 6 75 650119890 minus 3
IADEMF2(119903 = 05) 187119890 minus 6 207119890 minus 6 309119890 minus 6 75 472119890 minus 3
SOR4(120596 = 12) 312119890 minus 6 347119890 minus 6 517119890 minus 6 125 906119890 minus 3
GS4 667119890 minus 6 742119890 minus 6 111119890 minus 5 175 932119890 minus 3
JAC4 126119890 minus 5 141119890 minus 5 210119890 minus 5 250 105119890 minus 2
Table 6 Experiment 1 119898 = 7000 120582 = 05 Δ119909 = 143 times 10minus4 Δ119905 =
102 times 10minus8 119905 = 510 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME n ET (s)IADEMF4(119903 = 08) 367119890 minus 10 433119890 minus 10 366119890 minus 9 173 146119890 minus 1
IADEMF2(119903 = 05) 375119890 minus 10 435119890 minus 10 376119890 minus 9 200 124119890 minus 1
SOR4(120596 = 112) 455119890 minus 10 507119890 minus 10 367119890 minus 9 257 147119890 minus 1
GS4 110119890 minus 9 111119890 minus 9 376119890 minus 9 300 160119890 minus 1
JAC4 230119890 minus 9 231119890 minus 9 378119890 minus 9 400 163119890 minus 1
Table 7 Experiment 1 119898 = 7000 120582 = 10 Δ119909 = 143 times 10minus4 Δ119905 =
204 times 10minus8 119905 = 51 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 163119890 minus 7 163119890 minus 7 163119890 minus 7 112 960119890 minus 2
IADEMF2(119903 = 12) 163119890 minus 7 163119890 minus 7 163119890 minus 7 120 736119890 minus 2
SOR4(120596 = 115) 163119890 minus 7 163119890 minus 7 164119890 minus 7 171 999119890 minus 2
GS4 164119890 minus 7 164119890 minus 7 164119890 minus 7 216 107119890 minus 1
JAC4 165119890 minus 7 165119890 minus 7 165119890 minus 7 312 111119890 minus 1
It is found that the convergence rate of the fourth-orderand second-order methods improves with the application oflarger mesh size that is from 120582 = 05 to 120582 = 10 Thecoarser meshes cause a reduction in the computational oper-ations thus giving better rate of convergence However a
Table 8 Experiment 2 119898 = 7000 120582 = 05 Δ119909 = 143 times 10minus4 Δ119905 =
102 times 10minus8 119905 = 510 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 07) 567119890 minus 13 321119890 minus 12 130119890 minus 10 150 131119890 minus 1
IADEMF2(119903 = 07) 199119890 minus 11 460119890 minus 11 182119890 minus 9 200 127119890 minus 1
SOR4(120596 = 115) 245119890 minus 10 273119890 minus 10 399119890 minus 10 260 151119890 minus 1
GS4 398119890 minus 10 443119890 minus 10 661119890 minus 10 400 171119890 minus 1
JAC4 105119890 minus 9 117119890 minus 9 174119890 minus 9 550 176119890 minus 1
Table 9 Experiment 2 119898 = 7000 120582 = 10 Δ119909 = 143 times 10minus4 Δ119905 =
204 times 10minus8 119905 = 51 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 154119890 minus 6 171119890 minus 6 255119890 minus 6 96 912119890 minus 2
IADEMF2(119903 = 10) 154119890 minus 6 171119890 minus 6 255119890 minus 6 96 607119890 minus 2
SOR4(120596 = 12) 154119890 minus 6 171119890 minus 6 255119890 minus 6 192 997119890 minus 2
GS4 154119890 minus 6 171119890 minus 6 255119890 minus 6 288 128119890 minus 1
JAC4 154119890 minus 6 172119890 minus 6 256119890 minus 6 408 146119890 minus 1
more accurate solution is obtained by using finer mesh Forexample Table 6 shows that for 120582 = 05 the IADEMF4 hasRMSE = 433119890 minus 10 and 119899 = 173 whereas Table 7 shows thatfor 120582 = 10 its RMSE = 163119890 minus 7 and 119899 = 112 In generalamongst all the testedmethods the IADEMF4 still maintainsits greater accuracy characteristic even with coarser meshes
In terms of execution time the results from every tabledisplay shorter execution time for the IADEMF2 in compari-son to the IADEMF4 This is expected since the IADEMF2has lower computational complexity (Table 1) Despite theachievement in accuracy the IADEMF4 has to performmorecomputational work than the IADEMF2 since the former hasto utilize values of 119906 at two grid points on either side of thepoint (119894Δ119909 119896Δ119905) along the 119896th time level Thus if accuracyis desired then the preferred sequential numerical algorithmwould be the IADEMF4 On the other hand if execution timematters then the choice would be the IADEMF2
Amongst the fourth-order methods the IADEMF4 exe-cutes in the least amount of time Even though its computa-tional complexity is relatively large it operates with the leastnumber of iterations thus enabling it to be the most efficient
10 Conclusion
This paper proposes the development of the novel fourth-order IADEMF4 finite difference scheme The higher-orderscheme is developed by representing the spatial derivative inthe heat equation with the fourth-order finite difference CNapproximation This leads to the formation of pentadiagonalmatrices in the systems of linear equations with larger com-putational stencils The algorithm also employs the higheraccuracy of the Mitchell and Fairweather variant Despite
12 Journal of Applied Mathematics
the fourth-order IADE schemersquos higher computational com-plexity this technique is proved to be valuable becauseit enhances the accuracy of its second-order counterpartnamely the IADEMF2 In addition the higher accuracy ofIADEMF4 is verified as a convergent and unconditionallystable scheme and is superior in terms of rate of convergenceIt has also been proven to execute more efficiently in compar-ison to the other benchmarked fourth-order classical iterativemethods such as the GS4 the SOR4 and the JAC4 The in-creasing number of correct digits at each iteration serves as anadvantage for the IADEMF4 thus yielding faster rate of con-vergence with higher level of accuracy
In conclusion the proposed IADEMF4 scheme affordsusers many advantages with respect to higher accuracy sta-bility and rate of convergence and it serves as an alternativeefficient technique for the solution of a one-dimensional heatequation with Dirichlet boundary conditions
The proposed fourth-order scheme can be modified andadapted to more general multidimensional linear and non-linear parabolic elliptic and hyperbolic partial differentialequations using different types of boundary conditionsBesides it can also be considered for applications in problemsthat require higher-order accuracy with high resolution suchas problems in nanocomputing that require the solution ofvery large sparse systems of equation
The explicit and high accuracy feature of the IADEMF4can be exploited in two- and three-dimensional heat prob-lems by applying the scheme as pentadiagonal solvers forthe system of linear equations arising in the sweeps of ahigher-order alternating direction implicit (ADI) schemeThe lower-order ADI scheme was initially proposed byPeaceman and Rachford [24] The higher space dimensionsare expected not to cause unsatisfactory performance for theADI-IADEMF4 as long as the proposed scheme is stableconvergent and efficient as the process of its implementa-tion continues to advance from one-time step to anotherPathirana et al [25] implemented the ADI in developing atwo-dimensional model for incorporating flood damage inurban drainage planningThe proposed model is found to bestable numerically accurate and computationally efficientMirzavand et al [26] proposed the ADI-FDTD methodfor the physical modeling of high-frequency semiconductordevices The approach is able to reduce significantly the full-wave simulation time
It is possible to parallelize the IADEMF4 algorithm sincethe calculation of the new iteration only depends on theknown values from the last iteration (Figures 2 and 3) Futurework is to exploit the explicit computational properties andefficiency of the IADEMF4 for parallelization and executionon distributed parallel computing systems The idea is tospeed up the execution time without compromising its accu-racy especially on problems involving very large linear sys-tems of equations
References
[1] J Noye Computational Techniques for Differential Equationsvol 83 Elsevier Science North-Holland Publishing Amster-dam The Netherlands 1984
[2] G W Recktenwald ldquoFinite difference approximations to theheat equationrdquoMechanical Engineering vol 10 pp 1ndash27 2004
[3] P Lax and B Wendroff ldquoSystems of conservation lawsrdquo Com-munications on Pure and Applied Mathematics vol 13 pp 217ndash237 1960
[4] R J MacKinnon and G F Carey ldquoAnalysis of material interfacediscontinuities and superconvergent fluxes in finite differencetheoryrdquo Journal of Computational Physics vol 75 no 1 pp 151ndash167 1988
[5] R Hixon and E Turkel High Accuracy Compact MacCormack-Type Schemes For Computational Aerocoustics NASACentre forAerospace Information Lewis Research Centre 1998
[6] Y Lin Higher-order finite difference methods for solving heatequations [PhD thesis] Southern Illinois University 2008
[7] J Sulaiman M K Hasan M Othman M and S A AbdulKarim ldquoFourth-order QSMOR iterative method for the solu-tion of one-dimensional parabolic PDErsquosrdquo in Proceedings of theInternational Conference on Industrial and AppliedMathematics(CIAM rsquo10) pp 34ndash39 ITB Bandung Indonesia 2010
[8] N Jha ldquoThe application of sixth order accurate parallel quar-ter sweep alternating group explicit algorithm for nonlinearboundary value problems with singularityrdquo in Proceedings ofthe 2nd International Conference on Methods and Models inComputer Science (ICM2CS rsquo10) pp 76ndash80 December 2010
[9] Y Jin G Jin and J Li ldquoA class of high-precision finite differenceparallel algorithms for convection equationsrdquo Journal of Conver-gence Information Technology vol 6 no 1 pp 79ndash82 2011
[10] W Fu and E L Tan ldquoDevelopment of split-step FDTDmethodwith higher-order spatial accuracyrdquo Electronics Letters vol 40no 20 pp 1252ndash1253 2004
[11] A Mohebbi andM Dehghan ldquoHigh-order compact solution ofthe one-dimensional heat and advection-diffusion equationsrdquoApplied Mathematical Modelling vol 34 no 10 pp 3071ndash30842010
[12] W Liao ldquoAn implicit fourth-order compact finite differencescheme for one-dimensional Burgersrsquo equationrdquo Applied Math-ematics and Computation vol 206 no 2 pp 755ndash764 2008
[13] Z F Tian and Y B Ge ldquoA fourth-order compact ADI methodfor solving two-dimensional unsteady convection-diffusionproblemsrdquo Journal of Computational and Applied Mathematicsvol 198 no 1 pp 268ndash286 2007
[14] C Chun ldquoSome fourth-order iterativemethods for solving non-linear equationsrdquo Applied Mathematics and Computation vol195 no 2 pp 454ndash459 2008
[15] S Zhu Z Yu and J Zhao ldquoAhigh-order parallel finite differencealgorithmrdquo Applied Mathematics and Computation vol 183 no1 pp 365ndash372 2006
[16] Z Gao and S Xie ldquoFourth-order alternating direction implicitcompact finite difference schemes for two-dimensionalSchrodinger equationsrdquo Applied Numerical Mathematics vol61 no 4 pp 593ndash614 2011
[17] M S Sahimi A Ahmad and A A Bakar ldquoThe Iterative Alter-nating Decomposition Explicit (IADE) method to solve theheat conduction equationrdquo International Journal of ComputerMathematics vol 47 pp 219ndash229 1993
[18] G D Smith Numerical Solution of Partial Differential Equa-tions Finite Difference Methods Oxford University Press 2ndedition 1978
[19] A RMitchell andG Fairweather ldquoImproved forms of the alter-nating direction methods of Douglas Peaceman and Rachfordfor solving parabolic and elliptic equationsrdquo Numerische Math-ematik vol 6 no 1 pp 285ndash292 1964
Journal of Applied Mathematics 13
[20] B N DattaNumerical Linear Algebra and Applications BrooksCole Publishing Pacific Grove Calif USA 1995
[21] D M Young Iterative Solution of Large Linear Systems Aca-demic Press New York NY USA 1971
[22] V K Saulev Integration of Equations of Parabolic Type by theMethod of Nets Pergamon Press Oxford UK 1964
[23] L W Johnson and R D Riess Numerical Analysis Addison-Wesley Reading Mass USA 2nd edition 1982
[24] D W Peaceman and H H Rachford Jr ldquoThe numerical solu-tion of parabolic and elliptic differential equationsrdquo Journal ofthe Society for Industrial and Applied Mathematics vol 3 pp28ndash41 1955
[25] A Pathirana S Tsegaye B Gersonius and K VairavamoorthyldquoA simple 2-D inundation model for incorporating flood dam-age in urban drainage planningrdquo Hydrology and Earth SystemSciences vol 15 no 8 pp 2747ndash2761 2011
[26] R Mirzavand A Abdipour G Moradi and M MovahhedildquoFull-wave semiconductor devices simulation using ADI-FDTD methodrdquo Progress in Electromagnetics Research vol 11pp 191ndash202 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Journal of Applied Mathematics 3
At the grid-point 119875(119909119894 119905119896+12
) the IADEMF4 schemereplaces the spatial derivative in the heat equation with ahigher-order particularly the fourth-order Crank-Nicolson(CN) difference approximation [18] This is shown in theexpression given in (3) with the central difference operatorsdefined as 1205752
119909119906119896
119894= 119906119896
119894minus1minus 2119906119896
119894+ 119906119896
119894+1and 120575
2
119909119906119896+1
119894= 119906119896+1
119894minus1minus
2119906119896+1
119894+ 119906119896+1
119894+1 The approach gives the fourth-order scheme a
spatial truncation error of the order 119874(Δ119909)4 It enhances theaccuracy of its second-order counterpart which has a largererror of the order 119874(Δ119909)2
1
Δ119905(119906119896+1
119894minus 119906119896
119894) =
1
2(Δ119909)2(1205752
119909minus
1
121205754
119909) (119906119896+1
119894+ 119906119896
119894) (3)
To determine the stability of (3) the von Neumann sta-bility analysis can be applied Let 120582 = Δ119905(Δ119909)
2 and since1205754
119909119906119896
119894= 1205752
119909(1205752
119909119906119896
119894) the discretization of (3) becomes
119906119896+1
119894minus 119906119896
119894=120582
2(minus
1
12119906119896+1
119894minus2+4
3119906119896+1
119894minus1
minus5
2119906119896+1
119894+4
3119906119896+1
119894+1minus
1
12119906119896+1
119894+2
minus1
12119906119896
119894minus2+4
3119906119896
119894minus1minus5
2119906119896
119894
+4
3119906119896
119894+1minus
1
12119906119896
119894+2)
(4)
The discretized equation in (4) is assumed to have a solu-tion in the form of a Fourier harmonic function that is 119906119896
119894=
120588119896
119890119904120573119894Δ119909 where 120588 is referred to as the amplification factor 120573 is
an arbitrary constant and 119904 = radicminus1 The amplification factorrepresents the time dependence of the solution If the Fourierfunction is substituted into (4) and then solve for 120588 the resultwill be
120588 =
1 minus (1205826) ((cos120573 minus 4)2
minus 9)
1 + (1205826) ((cos120573 minus 4)2
minus 9)
with 120573 = 120573Δ119909 (5)
Since |120588| le 1 then the fourth-order CN approximation isunconditionally stable for any choice of 120573 Δ119905 and Δ119909
3 The Formulation of the IADEMF4
The IADEMF4 is firstly developed based on the execution ofthe unconditionally stable fourth-order CN approximation(3) on the heat equation
Equation (4) can also be expressed as in (6) using thedefinitions of the constants given in (7)
119886119906119896+1
119894minus2+ 119887119906119896+1
119894minus1+ 119888119906119896+1
119894+ 119889119906119896+1
119894+1+ 119890119906119896+1
119894+2
= minus119886119906119896
119894minus2minus 119887119906119896
119894minus1+ 119888119906119896
119894minus 119889119906119896
119894+1minus 119890119906119896
119894+2
119894 = 2 3 119898 minus 1
(6)
119886 =120582
24 119887 = minus
2120582
3 119888 =
4 + 5120582
4
119889 = minus2120582
3 119890 =
120582
24 119888 =
4 minus 5120582
4
(7)
The approximation in (6) can be displayed in a matrix formsuch as 119860u = f (8) where 119860 is a sparse pentadiagonalcoefficient matrix and u = (119906
2 1199063 119906
119898minus2 119906119898minus1
)119879 is the
columnvector containing the unknownvalues of119906 at the timelevel 119896 + 1 The column vector f = (119891
2 1198913 119891
119898minus2 119891119898minus1
)119879
consists of boundary values and known 119906 values at theprevious time level 119896 The definitions for every entry in f aregiven in (9)
119860u = f
[[[[[[[[[
[
119888 119889 119890
119887 119888 119889 119890 O119886 119887 119888 119889 119890
d d d d119886 119887 119888 119889 119890
O 119886 119887 119888 119889
119886 119887 119888
]]]]]]]]]
](119898minus2)times(119898minus2)
[[[[[[
[
1199062
1199063
119906119898minus2
119906119898minus1
]]]]]]
]119896+1
=
[[[[[[
[
1198912
1198913
119891119898minus2
119891119898minus1
]]]]]]
]
(8)
1198912= minus119887 (119906
119896
1+ 119906119896+1
1) + 119888119906
119896
2minus 119889119906119896
3minus 119890119906119896
4
1198913= minus119886 (119906
119896
1+ 119906119896+1
1) minus 119887119906
119896
2+ 119888119906119896
3minus 119889119906119896
4minus 119890119906119896
5
119891119894= minus119886119906
119896
119894minus2minus 119887119906119896
119894minus1+ 119888119906119896
119894minus 119889119906119896
119894+1minus 119890119906119896
119894+2
for 119894 = 4 5 119898 minus 3
119891119898minus2
= minus119886119906119896
119898minus4minus 119887119906119896
119898minus3+ 119888119906119896
119898minus2
minus 119889119906119896
119898minus1minus 119890 (119906
119896
119898+ 119906119896+1
119898)
119891119898minus1
= minus119886119906119896
119898minus3minus 119887119906119896
119898minus2+ 119888119906119896
119898minus1minus 119889 (119906
119896
119898+ 119906119896+1
119898)
(9)
The evaluations of 1198912 1198913 119891119898minus2
and 119891119898minus1
require thevalues of 119906 at the boundaries 119894 = 1 and 119894 = 119898 Howeverthese values cannot be obtained numerically because theircomputations involve nodes at 119894 = minus1 and 119894 = 119898 + 2which are exterior to the considered solution domain Ifthe exact solutions of 119906 are available at 119894 = 1 and 119894 =
119898 then it is appropriate to consider them as the requiredboundary values Otherwise the boundary conditions haveto be formulated bearing in mind that they should be ofcomparable accuracy [1]
The IADEMF4 scheme secondly employs the higher-order accuracy formula of MF [19] The variant whose accu-racy is of the order 119874((Δ119905)2 + (Δ119909)
4
) is as given in (10) and(11)
(119903119868 + 1198661) u(119901+12) = (119903119868 minus 119892119866
2) u(119901) + f (10)
(119903119868 + 1198662) u(119901+1) = (119903119868 minus 119892119866
1) u(119901+12) + 119892f (11)
where 119903 119901 and 119868 represent an acceleration parameter theiteration index and an identity matrix respectively 119866
1and
1198662are two constituent matrices The vectors u(119901+1) and
u(119901+12) represent the required solution at the iteration level(119901 + 1) and at some intermediate level (119901 + 12) respectivelyThe relation of 119892 and 119903 is given by 119892 = (6 + 119903)6 119903 gt 0
4 Journal of Applied Mathematics
Substitute the following expression obtained from (10)into (11)
u(119901+12) = (119903119868 + 1198661)minus1
[(119903119868 minus 1198921198662) u(119901) + f] (12)
and then simplify to obtain
(119903119868 + 1198662) u(119901+1) = (119903119868 minus 119892119866
1) (119903119868 + 119866
1)minus1
(119903119868 minus 1198921198662) u(119901)
+ [ (119903119868 minus 1198921198661) (119903119868 minus 119892119866
1)minus1
+ 119892] f (13)
As 119901 in (13) becomes sufficiently large the temperaturesolution reaches a steady state that is as 119901 rarr infin thenu(119901+1) rarr u and u(119901) rarr u Simplify the equation and thenmultiply by (119903119868+119866
1) After some algebraicmanipulations the
following expression is obtained
[119903 (1198661+ 1198662) (1 + 119892) + (1 minus 119892
2
)11986611198662] u = 119903 (1 + 119892) f (14)
Multiply (14) by 1119903(1 + 119892) and use the definition of 119892 tofinally obtain the following form
[ (1198661+ 1198662) minus
1
611986611198662] u = f (15)
The comparison between the expression 119860u = f in (8)and the form given in (15) suggests that the coefficient matrix119860 for the IADEMF4 can be decomposed into
119860 = 1198661+ 1198662minus1
611986611198662 (16)
The IADEMF4 requires the constituentmatrices1198661and119866
2in
(16) to be in the form of lower and upper tridiagonalmatricesrespectively in order to retain the pentadiagonal structure of119860 Thus
1198661=
[[[[[[[
[
1
1198971
1 119874
1
1198972
d2d dd 119897119898minus4
1
119874 119898minus4
119897119898minus3
1
]]]]]]]
](119898minus2)times(119898minus2)
1198662=
[[[[[[[
[
11989011
V1
1198902
2
V2
119874
d d d119874 119890
119898minus4119898minus4
V119898minus4
119890119898minus3
119898minus3
119890119898minus2
]]]]]]]
](119898minus2)times(119898minus2)
(17)
By substituting 1198661and 119866
2in (17) into the formula for the
decomposition of 119860 the entries of the resultant matrix arecompared with the entries of 119860 in (8) yielding the followingdefinitions
1198901=6 (119888 minus 1)
5
1=6119889
5 119897
1=
6119887
6 minus 1198901
1198902=6 (119888 minus 1) + 119897
11
5 V
119894=6119890
5for 119894 = 1 2 119898 minus 4
(18)
And for 119894 = 2 3 119898 minus 3
119894=6119889 + 119897
119894minus1V119894minus1
5
119894minus1=
6119886
6 minus 119890119894minus1
119897119894=6119887 +
119894minus1119894minus1
6 minus 119890119894
119890119894+1
=6 (119888 minus 1) + 119897
119894119894+ 119894minus1
V119894minus1
5
(19)
Since1198661and119866
2are three bandedmatrices then (119903119868+119866
1)
and (119903119868+1198662) can be inverted easilyThe equations in (10) and
(11) are rearranged as in (20) and (21) respectively
u(119901+12) = (119903119868 + 1198661)minus1
(119903119868 minus 1198921198662) u(119901)
+ (119903119868 + 1198661)minus1f
(20)
u(119901+1) = (119903119868 + 1198662)minus1
(119903119868 minus 1198921198661) u(119901+12)
+ 119892(119903119868 + 1198662)minus1f
(21)
The above two equations are computed and simplifiedleading to the computational formulae at each of the halfiteration levels as given in (22) and (23)
(i) At the (119901 + 12) iteration level
119906(119901+12)
119894=
1
119877(119864119894minus1
119906(119901)
119894+119882119894minus1
119906(119901)
119894+1+ 119881119894minus1
119906(119901)
119894+2
minus119894minus3
119906(119901+12)
119894minus2minus 119897119894minus2
119906(119901+12)
119894minus1+ 119891119894)
119894 = 2 3 119898 minus 2119898 minus 1
(22)
(ii) At the (119901 + 1) iteration level
119906(119901+1)
119894=
1
119885119894minus1
(119878119894minus3
119906(119901+12)
119894minus2+ 119876119894minus2
119906(119901+12)
119894minus1+ 119875119906(119901+12)
119894
minus119894minus1
119906(119901+1)
119894+1minus V119894minus1
119906(119901+1)
119894+2+ 119892119891119894)
119894 = 119898 minus 1 119898 minus 2 3 2
(23)
with
minus1
= 0= 1198970= 119881119898minus2
= 119881119898minus3
= 119882119898minus2
= 119898minus2
= V119898minus2
= V119898minus3
= 1198760= 119878minus1
= 1198780= 0
119877 = 1 + 119903 119875 = 119903 minus 119892
119864119894= 119903 minus 119892119890
119894 119885119894= 119903 + 119890
119894 119894 = 1 2 119898 minus 2
119882119894= minus119892
119894 119876119894= minus119892119897
119894 119894 = 1 2 119898 minus 3
119881119894=minus119892V119894 119878119894= minus119892
119894 119894 = 1 2 119898 minus 4
(24)
The IADEMF4 algorithm is regarded as a two-stageprocess involving two iteration levels (119901 + 12) and (119901 + 1)It is completed explicitly by using the required equations atthe two levels in alternate sweeps along all the grid pointsin the interval (0 1) until convergence is reached In (22)the calculation to determine the unknown 119906
(119901+12)
119894begins at
the left boundary and then moves to the right In a similar
Journal of Applied Mathematics 5
manner 119906(119901+1)119894
in (23) is calculated by proceeding from theright boundary towards the left (Figure 1)
The computational molecules are depicted in Figures 2and 3
At each level of iteration the computational moleculesinvolve two known grid points at the new level and anotherthree known ones at the old level Clearly the method isexplicit
4 Convergence Analysis of the IADEMF4
This section proves the convergence of the IADEMF4 Since120582 gt 1will not guarantee an accurate approximation for 120597119880120597119905[18] the values of 120582 that are considered appropriate for theproof are 0 lt 120582 le 1
From the definitions in ((18) and (19)) the followingresults are obtained
1198901=6 (119888 minus 1)
5=3120582
2 implying that 0 lt 119890
1le3
2
1=6119889
5lt0 119897
1=
6119887
6 minus 1198901
lt0 V119894=6119890
5gt 0
for 119894 = 1 2 119898 minus 4
2=6119889 + 119897
1V1
5= 1+1198971V1
5lt 1 since 119897
1V1lt 0
1198902=6 (119888 minus 1) + 119897
11
5= 1198901+11989711
5gt 1198901 since 119897
11gt 0
(25)
Simple computation shows that 119897115 lt 1 Clearly 119890
2lt 6
Lemma 1 119890119894in ((18) and (19)) is such that 0 lt 119890
119894lt 6 for all
119894 = 1 2 3 119898 minus 2
Proof The results in (25) show that 6 gt 1198902gt 1198901gt 0 Assume
it is true that 6 gt 119890119896gt 119890119896minus1
gt 0 for all 119896 = 3 4 5 119898 minus 3Since the assumption implies that 6 gt 119890
119896minus1gt 119890119896minus2
gt 0 then119896minus1
= 6119886(6 minus 119890119896minus1
) gt 119896minus2
= 6119886(6 minus 119890119896minus2
) gt 0 Therefore119896minus1
V119896minus1
gt 119896minus2
V119896minus2
It has also been shown that
2lt 1lt 0 Assume it is true
that 119895lt 119895minus1
lt 0 for all 119895 = 3 4 119898 minus 4 The assumptionimplies that
119895minus1lt 119895minus2
lt 0 Thus 119895minus1
119895minus1
lt 119895minus2
119895minus2
lt 0
119895+1
=6119889 + 119897
119895V119895
5= 1+119897119895V119895
5= 1+
(6119887 + 119895minus1
119895minus1
) V119895
5 (6 minus 119890119895)
lt 1+
(6119887 + 119895minus2
119895minus2
) V119895minus1
5 (6 minus 119890119895minus1
)
= 1+119897119895minus1
V119895minus1
5= 119895lt 0
(26)
Since it is true that for 119895 + 1 119895+1
lt 119895lt 0 then by
induction the assumption 119895lt 119895minus1
lt 0 is true for all 119895 An
0
u(p+12)
i
u(p+1)
i
i iminus 1 i + 1 m + 1
Figure 1 The two-stage IADEMF4 algorithmThe directions of thesweeps at the (119901 + 12) and (119901 + 1) iteration levels
i iminus 1i minus 2 i + 1 i + 2
(p + 12)
(p)
Figure 2 Computational molecule of the IADEMF4 at the (119901+12)iteration level
i iminus 1i minus 2 i + 1 i + 2
(p + 12)
(p + 1)
Figure 3 Computational molecule of the IADEMF4 at the (119901 + 1)
iteration level
equivalent to the preceding statement would be 119896lt 119896minus1
lt 0
for all 119896 = 3 4 5 119898 minus 3 It follows that
119897119896119896=(6119887 +
119896minus1119896minus1
) 119896
6 minus 119890119896
gt(6119887 +
119896minus2119896minus2
) 119896minus1
6 minus 119890119896minus1
= 119897119896minus1
119896minus1
119890119896+1
=6 (119888 minus 1) + 119897
119896119896+ 119896minus1
V119896minus1
5= 1198901+119897119896119896+ 119896minus1
V119896minus1
5
gt 1198901+119897119896minus1
119896minus1
+ 119896minus2
V119896minus2
5= 119890119896gt 0
(27)
Since it is true that for 119896+1 119890119896+1
gt 119890119896gt 0 then by induction
119890119896gt 119890119896minus1
gt 0 is also true for all 119896Suppose there is a 119896 such that 119890
119896gt 6 Then
119890119896+1
= 1198901+119897119896119896+ 119896minus1
V119896minus1
5
= 1198901+
6119887 + 119896
5 (6 minus 119890119896)+119896minus1
5(119896minus1
119896
6 minus 119890119896
+ V119896minus1
) lt 1198901
(28)
This is a contradiction since 119890119896+1
gt 1198901 This verifies that 6 gt
119890119896gt 119890119896minus1
gt 0 119896 = 3 4 5 119898 minus 3 Let 119894 = 2 3 4 119898 minus 2and then 6 gt 119890
119894gt 119890119894minus1
gt 0 Therefore 0 lt 119890119894lt 6 119894 =
1 2 3 119898 minus 2 Lemma 1 is proved
6 Journal of Applied Mathematics
Lemma 2 If 119903 gt 0 and 119892 = (6+ 119903)6 thenmax |(119903 minus119892119890119894)(119903 +
119890119894)| lt 1 i = 1 2 3 119898 minus 2
Proof Let max |(119903 minus 119892119890119894)(119903 + 119890
119894)| = |(119903 minus 119892119890
119895)(119903 + 119890
119895)| for
some 119895Assume |(119903 minus 119892119890
119895)(119903 + 119890
119895)| ge 1 Since 119903 gt 0 and 119890
119895gt 0
then 119903 + 119890119895gt 0
If (119903 minus 119892119890119895)(119903 + 119890
119895) ge 1 then minus((6 + 119903)6)119890
119895ge 119890119895 which
implies that 119903 le minus12 This contradicts the fact that 119903 gt 0If (119903 minus 119892119890
119895)(119903 + 119890
119895) le minus1 then 2119903 minus ((6 + 119903)6)119890
119895le minus119890119895
which implies that 119890119895ge 12 This contradicts Lemma 1
So the assumption that |(119903 minus 119892119890119895)(119903 + 119890
119895)| ge 1 is false
Hence max |(119903 minus 119892119890119894)(119903 + 119890
119894)| = |(119903 minus 119892119890
119895)(119903 + 119890
119895)| lt 1
Proposition 3 (119903119868 minus 1198921198661)(119903119868 + 119866
1)minus1
2lt 1
Proof Let 119865 = (119903119868 minus 1198921198661)(119903119868 + 119866
1)minus1 that is
119865 =
[[[[[[[[[[
[
119903 minus 119892
119889
d119903 minus 119892
119889119874
d d d
d d
sdot sdot sdot sdot sdot sdot sdot sdot sdot sdot sdot sdot119903 minus 119892
119889
]]]]]]]]]]
]
(29)
119865 is a lower triangular matrix with all the diagonal entries(whence the eigenvalues of 119865) equal to (119903 minus 119892)119889 where 119889 =
119903 + 1 Denote all the eigenvalues of 119865 by 120582119865 If 120588[119865] is defined
as the spectral radius of 119865 then
120588 [119865] = max 10038161003816100381610038161205821198651003816100381610038161003816 =
10038161003816100381610038161205821198651003816100381610038161003816 =
1003816100381610038161003816100381610038161003816
119903 minus 119892
119903 + 1
1003816100381610038161003816100381610038161003816le 119865
2 (30)
But by definition of 2-norm
1198652= maxx2 = 0
119865x2
x2
= maxy2=1
1003817100381710038171003817119865y10038171003817100381710038172
1003817100381710038171003817y10038171003817100381710038172
= maxy2=1
1003817100381710038171003817119865y10038171003817100381710038172
le maxw2=1
119865w2= maxw2=1
1003817100381710038171003817120582119865w10038171003817100381710038172
(31)
Since all the eigenvalues of 119865 are equal then
1198652le1003816100381610038161003816120582119865
1003816100381610038161003816 maxw2=1
w2=1003816100381610038161003816120582119865
1003816100381610038161003816 = 120588 [119865] (32)
Thus from (30) and (32) 1198652= |120582119865| = |(119903 minus 119892)(119903 + 1)|
By Lemma 2 with 119890119895replaced by 1 |(119903 minus 119892)(119903 + 1)| lt 1
is obtained leading to the result of Proposition 3 which is1198652= (119903119868 minus 119892119866
1)(119903119868 + 119866
1)minus1
2lt 1
Proposition 4 (119903119868 minus 1198921198662)(119903119868 + 119866
2)minus1
2lt 1
Proof Let 119870 = (119903119868 minus 1198921198662)(119903119868 + 119866
2)minus1 that is
119870 =
[[[[[[[[[[[[[
[
119903 minus 1198921198901
1199111
d sdot sdot sdot sdot sdot sdot sdot sdot sdot
119903 minus 1198921198902
1199112
d sdot sdot sdot
d
119874 d d
119903 minus 119892119890119898
119911119898
]]]]]]]]]]]]]
]
(33)
119870 is an upper triangular matrix with all the diagonal entriesequal to (119903 minus 119892119890
119894)119911119894 119894 = 1 2 119898 minus 2 where 119911
119894= 119903 + 119890
119894
Since all the eigenvalues of119870 are distinct then119870 is similar toa diagonal matrix119863
By the Schur triangularization theorem [20] there is anorthogonal matrix 119874 such that 119874119879119870119874 = 119863 The diagonalentries of119863 are the eigenvalues of119870
1198702=10038171003817100381710038171003817119874119879
119870119874100381710038171003817100381710038172
= 1198632
= radicmaximum eigenvalue of 119863119879119863
= radicmaximum eigenvalue of 1198632
= radic120588 (1198632) = radic120588 (1198702) = 120588 (119870)
(34)
By Lemma 2
1198702= 120588 (119870) = max
10038161003816100381610038161003816100381610038161003816
119903 minus 119892119890119894
119903 + 119890119894
10038161003816100381610038161003816100381610038161003816
=
1003816100381610038161003816100381610038161003816100381610038161003816
119903 minus 119892119890119895
119903 + 119890119895
1003816100381610038161003816100381610038161003816100381610038161003816
lt 1 (35)
for some 119895 Hence this proves Proposition 4
From (20) and (21)
u(119901+1) = (119903119868 + 1198662)minus1
(119903119868 minus 1198921198661)
times [(119903119868 + 1198661)minus1
(119903119868 minus 1198921198662) u(119901) + (119903119868 + 119866
1)minus1f]
+ 119892 (119903119868 + 1198662)minus1f
(36)
Let
119872(119903) = (119903119868 + 1198662)minus1
(119903119868 minus 1198921198661) (119903119868 + 119866
1)minus1
(119903119868 minus 1198921198662)
(37)
And let
119902 (119903) = [(119903119868 + 1198662)minus1
(119903119868 minus 1198921198661) (119903119868 + 119866
1)minus1
+119892(119903119868 + 1198662)minus1
] f (38)
Then u(119901+1) = 119872(119903)u(119901) + 119902(119903)
Theorem 5 The IADEMF4 is convergent if 120588[119872(119903)] lt 1 for119903 gt 0
Journal of Applied Mathematics 7
Proof Define (119903) = (119903119868 + 1198662)119872(119903)(119903119868 + 119866
2)minus1 then
(119903) = (119903119868 minus 1198921198661) (119903119868 + 119866
1)minus1
(119903119868 minus 1198921198662) (119903119868 + 119866
2)minus1
(39)
Thus by similarity119872(119903) and (119903) have the same set of eigen-values Therefore
120588 [119872 (119903)]
= 120588 [ (119903)] le10038171003817100381710038171003817 (119903)
10038171003817100381710038171003817 2
=10038171003817100381710038171003817(119903119868 minus 119892119866
1) (119903119868 + 119866
1)minus1
(119903119868 minus 1198921198662) (119903119868 + 119866
2)minus110038171003817100381710038171003817 2
le10038171003817100381710038171003817(119903119868 minus 119892119866
1) (119903119868 + 119866
1)minus1100381710038171003817100381710038172
10038171003817100381710038171003817(119903119868 minus 119892119866
2) (119903119868 + 119866
2)minus110038171003817100381710038171003817 2
lt 1
(40)
The last inequality is due to Propositions 3 and 4The provenTheorem 5 assures the convergence of the IADEMF4
5 The Fourth-Order ClassicalIterative Methods
The system of linear equations in (8) may also be solvedby using the classical iterative methods of the fourth orderThey include the JAC4 the GS4 and the SOR4 methodsThese iterative methods are capable of exploiting the sparsestructure of the pentadiagonal matrix
The JAC4 algorithm can be represented by
119906(119901+1)
119894=
(119891119894minus119886119906(119901)
119894minus2minus 119887119906(119901)
119894minus1minus 119889119906(119901)
119894+1minus 119890119906(119901)
119894+2)
119888
119894 = 2 3 119898 minus 1
(41)
The approximation of 119906(119901+1)119894
at the (119901 + 1)th iteration level iscomputed using the relevant values in the 119901th iteration
The GS4 method uses the most recent values of 119906(119901+1)119894minus2
and 119906(119901+1)
119894minus1to update the approximation value of 119906(119901+1)
119894 The
algorithm for the GS4 is as expressed in (42)
119906(119901+1)
119894=
(119891119894minus119886119906(119901+1)
119894minus2minus 119887119906(119901+1)
119894minus1minus 119889119906(119901)
119894+1minus 119890119906(119901)
119894+2)
119888
119894 = 2 3 119898 minus 1
(42)
The SOR4 iterative method accelerates the convergencerate of the GS4 If 120596 is a relaxation parameter then for any120596 = 0 (42) can be rewritten as
119906(119901+1)
119894= (1 minus 120596) 119906
(119901)
119894
+
120596 (119891119894minus 119886119906(119901+1)
119894minus2minus 119887119906(119901+1)
119894minus1minus 119889119906(119901)
119894+1minus 119890119906(119901)
119894+2)
119888
119894 = 2 3 119898 minus 1
(43)
i iminus 1i minus 2 i + 1 i + 2
(p + 1)
(p)
Figure 4 Computational molecule of the JAC4
i iminus 1i minus 2 i + 1 i + 2
(p + 1)
(p)
Figure 5 Computational molecule of the GS4SOR4
Table 1 Sequential arithmetic operations per iteration (119898 problemsize)
Method Number ofadditions
Number ofmultiplications
Total operationcount
IADEMF4 10(119898 minus 2) 13(119898 minus 2) 23(119898 minus 2)
IADEMF2 6119898 9119898 15119898
JAC4GS4 4(119898 minus 2) 5(119898 minus 2) 9(119898 minus 2)
SOR4 5(119898 minus 2) 7(119898 minus 2) 12(119898 minus 2)
The SOR4 algorithm reduces to the GS4 if 120596 = 1 Exceptfor very special cases it is difficult to obtain the analyticexpression for 120596 According to Young [21] the precise deter-mination of the optimal 120596 in the SOR is only known for asmall class of matrices The value of 120596 generally lies in therange of 1 lt 120596 lt 2
The computational molecules for the JAC4 and the GS4are as illustrated in Figures 4 and 5 respectively The SOR4has the same form as the GS4
6 Computational Complexity
The cost of implementing an algorithm can be assessedby examining its computational complexity The number ofarithmetic operations such as additions (including subtrac-tions) andmultiplications (including divisions) that is neededto perform by the algorithm can be straightforwardly count-ed Table 1 gives the number of sequential arithmetic opera-tions per iteration that is required to evaluate the algorithms
For the higher-order schemes trade-offbetween accuracyand speed usually happens The computational complexityhas an effect on the efficiency of a particular scheme Thusthis factor will be taken into account when discussing theresults of the numerical experiments conducted in this study
7 The Pseudocode of the IADEMF4Sequential Algorithm
Algorithm 1 illustrates the pseudocode of the IADEMF4sequential algorithm This algorithm can also be generalizedand applied to other methods discussed in this paper
8 Journal of Applied Mathematics
begindetermine the parameters119898 119903 Δ119905 Δ119909 120582 120576 119879
determine initial conditions 119906(119901)119894
determine exact solutions 119880119894at time level 119879
while (time level lt 119879)determine boundary conditionsfor 119894 = 2 to 119894 = 119898 minus 1
compute 119891119894(refer to (9))
end-forset iteration = 0while (convergence conditions are not satisfied)
for 119894 = 2 to 119894 = 119898 minus 1
compute 119906(119901+12)119894
(refer to (22))end-forfor 119894 = 2 to 119894 = 119898 minus 1
compute 119906(119901+1)119894
(refer to (23))end-fortest for convergencecompute 119890
119894larr
10038161003816100381610038161003816119906(119901+1)
119894minus 119906(119901)
119894
10038161003816100381610038161003816if (max1003816100381610038161003816119890119894
1003816100381610038161003816 lt 120576)then 119906
(119901)
119894larr 119906(119901+1)
119894
add 1 to iteration (if necessary)end-while
end-whilefor 119894 = 2 to 119894 = 119898 minus 1
determine average absolute error root meansquare error and maximum error
end-forend
Algorithm 1 The IADEMF4 sequential algorithm
8 Numerical Experiments
Two experiments were conducted to test the sequentialnumerical performance of the proposed IADEMF4 methodagainst those of the benchmarked classical iterative methodsnamely the JAC4 theGS4 and the SOR4 Comparison is alsomade with the corresponding IADE method of the secondorder
Experiment 1 This problem was taken from Saulev [22]
120597119880
120597119905=1205972
119880
1205971199092 0 le 119909 le 1 (44)
subject to the initial condition
119880 (119909 0) = 4119909 (1 minus 119909) 0 le 119909 le 1 (45)
and the boundary conditions
119880 (0 119905) = 119880 (1 119905) = 0 119905 ge 0 (46)
The exact solution to the given problem is given by
119880 (119909 119905) =32
1205873
infin
sum
119896=1(2)
1
1198963119890minus12058721198962119905 sin (119896120587119909) (47)
Experiment 2 This problem was taken from Johnson andReiss [23]
120597119880
120597119905=1205972
119880
1205971199092 0 le 119909 le 1 (48)
subject to the initial condition
119880 (119909 0) = sin (120587119909) (1 + 6 cos (120587119909)) 0 le 119909 le 1 (49)
and the boundary conditions
119880 (0 119905) = 119880 (1 119905) = 0 119905 ge 0 (50)
The exact solution to the given problem is given by
119880 (119909 119905) = sin (120587119909) 119890minus1205872119905
+ 3 sin (2120587119909) 119890minus41205872119905
(51)
Journal of Applied Mathematics 9
9 Results and Discussion
In each experiment the exact solutions at 119894 = 1 and 119894 = 119898
were taken as boundary values for the IADEMF4 and theother fourth-order classical iterative methods The conver-gence criterion used in the testing of each method was takenas 119906(119901+1) minus 119906
(119901)
infin
le 120576 where 120576 is the convergence toleranceThe selections of the optimum 119903 or 120596 were determined byexperiments As for the IADEMF2 the CN scheme with 120579 =
12 was employedFigures 6 7 8 and 9 visualize the behavior of the one-
dimensional parabolic heat solutions for both experimentsBy using119898 = 10 and a tolerance requirement of 120576 = 10
minus4 twodifferent mesh sizes 120582 = 05 and 120582 = 10 were consideredin each experiment The exact solutions are compared withthe IADEMF4 and the IADEMF2 numerical solutions Theoptimum value of 119903 chosen for each method as well as theircorresponding outcome of the number of iterations 119899 isstated in the legend of each figure Every figure reveals thatthe IADEMF4 is more accurate than the IADEMF2 andthe former converges with fewer numbers of iterations incomparison to the latter The numerical solutions of theIADEMF4 seem to be in very good agreement with theexact solutions For example in Figure 7 at 119909 = 05 thedifference between the exact solution and the numericalsolution using the IADEMF2 and using the IADEMF4 isabout 36 and 0 respectively These results imply thatthe accuracy and convergence rate of the second-order IADEmethod are enhanced by the implementation of the fourth-order CN approximation that leads to the formation of thecorresponding fourth-order IADE scheme
Figure 10 displays the graph of log (RMSE) versus log(Δ119909)for decreasing values of Δ119909 implemented on Experiment 2By considering 120576 = 10
minus4 and fixing Δ119905 = 00001 the value ofΔ119909 was initially taken as Δ119909 = 0125 It was then successivelyhalved into Δ1199092 Δ1199094 and Δ1199098
The figure shows that amongst the tested methods theroot mean square error (RMSE) of the IADEMF4 and theIADEMF2 decreases linearly as the values of Δ119909 decreasesThe slope of the IADEMF4 is approximately equal to 4which corresponds to its fourth-order spatial accuracy TheIADEMF2 with a second-order spatial accuracy has a slopethat is approximately equal to 2 It is clear that the IADEMF4is always more accurate than the IADEMF2 for the differentconsidered mesh sizes The graphs of the SOR4 (120596 = 105)the GS4 and the JAC4 show that their accuracies of fourthorder tend to lack as Δ119909 decreases largely due to the effect ofincreasing round-off errors as the value of119898 increases
Tables 2 3 4 5 6 7 8 and 9 provide numerical results interms of the average absolute error (AAE) root mean squareerror (RMSE) maximum error (ME) number of iterations(119899) and execution time (ET) measured in seconds (s) Theresults are obtained from both experiments for two differentvalues of 119898 and mesh size 120582 It is generally observed thatwhen 119898 = 700 120582 = 05 and 120576 = 10
minus6 the IADEMF4has the least average absolute error root mean square errorand maximum error in comparison with the other methodsunder consideration (Tables 2ndash5) When the size of 119898 was
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 08 n = 172)IADEMF2 (r = 03 n = 185)
00E + 00
10E minus 02
20E minus 02
30E minus 02
40E minus 02
50E minus 02
60E minus 02
70E minus 02
80E minus 02
90E minus 02
10E minus 01
U(xt)
Figure 6 Numerical and exact solutions for Experiment 1 120582 = 05Δ119909 = 01 Δ119905 = 0005 119905 = 025 and 120576 = 10
minus4
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 14 n = 159)IADEMF2 (r = 05 n = 162)
00E + 00
10E minus 03
20E minus 03
30E minus 03
40E minus 03
50E minus 03
60E minus 03
70E minus 03
80E minus 03
90E minus 03
U(xt)
Figure 7 Numerical and exact solutions for Experiment 1 120582 = 10Δ119909 = 01 Δ119905 = 001 119905 = 05 and 120576 = 10
minus4
ten times bigger (119898 = 7000) and a more stringent tolerancecriterion was set (120576 = 10
minus10
) the accuracy of the IADEMF4clearly outperforms the other methods for a mesh size of120582 = 05 (Tables 6 and 8) However the difference in the errorsamongst the tested methods is not so obvious for the case of120582 = 10 (Tables 7 and 9)The achievement of the fourth-orderIADEmethod in Experiments 1 and 2 can be clearly seen fromthe results in Tables 2 and 8 respectively where it has caused ahuge 94 reduction of RMSE from its corresponding second-order IADE method
The IADEMF4 has the advantage of featuring higheraccuracy due to the execution of the fourth-order CNapprox-imation coupled with the fourth-order accurate MF variant
10 Journal of Applied Mathematics
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 10 n = 169)IADEMF2 (r = 03 n = 213)
00E + 00
10E minus 02
20E minus 02
30E minus 02
40E minus 02
50E minus 02
60E minus 02
70E minus 02
80E minus 02
90E minus 02
10E minus 01
U(xt)
Figure 8 Numerical and exact solutions for Experiment 2 120582 = 05Δ119909 = 01 Δ119905 = 0005 119905 = 025 and 120576 = 10
minus4
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 10 n = 175)IADEMF2 (r = 05 n = 182)
00E + 00
10E minus 03
20E minus 03
30E minus 03
40E minus 03
50E minus 03
60E minus 03
70E minus 03
80E minus 03
U(xt)
Figure 9 Numerical and exact solutions for Experiment 2 120582 = 10Δ119909 = 01 Δ119905 = 001 119905 = 05 and 120576 = 10
minus4
The IADEMF2 is only derived from the second-order CNapproximation but its combination with its correspondingfourth-order MF variant managed to produce errors whichare better than theGS4 JAC4 and the SOR4 Even though theclassical iterative methods are also derived from the fourth-order accurate CN type approximation they are lackingin accuracy due to the round-off errors that have beenaccumulated from the time the execution starts till it ends
With regards to the rate of convergence the resultsfrom each table demonstrate that the number of iterationsproduced by the IADEMF2 and the fourth-order classicaliterative methods is greater than or at least equal to that of
0
minus2
minus208 minus277 minus347 minus416
minus4
minus8
minus6
minus10
minus12
minus14
minus16
minus18
GS4JAC4SOR4
log
(RM
SE)
log(Δx)
IADEMF4
IADEMF2
Figure 10 Log(RMSE) versus Log(Δ119909) Experiment 2 Δ119905 = 00001119905 = 0002 and 120576 = 10
minus4
Table 2 Experiment 1 119898 = 700 120582 = 05 Δ119909 = 143 times 10minus3 Δ119905 =
102 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 07) 267119890 minus 9 330119890 minus 9 212119890 minus 8 100 440119890 minus 2
IADEMF2(119903 = 08) 188119890 minus 8 495119890 minus 8 323119890 minus 7 100 635119890 minus 3
SOR4(120596 = 12) 540119890 minus 7 544119890 minus 7 552119890 minus 7 108 460119890 minus 2
GS4 540119890 minus 6 543119890 minus 6 332119890 minus 6 150 472119890 minus 2
JAC4 228119890 minus 5 229119890 minus 5 232119890 minus 5 150 473119890 minus 2
Table 3 Experiment 1 119898 = 700 120582 = 10 Δ119909 = 143 times 10minus3 Δ119905 =
204 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 297119890 minus 7 301119890 minus 7 301119890 minus 7 50 201119890 minus 2
IADEMF2(119903 = 09) 557119890 minus 7 559119890 minus 7 848119890 minus 7 50 317119890 minus 3
SOR4(120596 = 12) 329119890 minus 6 331119890 minus 6 335119890 minus 6 76 225119890 minus 2
GS4 895119890 minus 6 900119890 minus 6 900119890 minus 6 100 238119890 minus 2
JAC4 214119890 minus 5 215119890 minus 5 217119890 minus 5 125 250119890 minus 2
the IADEMF4 The convergence rate of the latter surpassesthe others in the case of 119898 = 7000 and 120576 = 10
minus10 (Tables6ndash8) Even though the operational count of the IADEMF4is relatively quite large (Table 1) due to its higher level ofaccuracy its increasing number of correct digits at eachiteration causes it to converge at a faster rate The IADEMF2has the advantage of having less mathematical operationsthus it is competitive in terms of convergence rate but atthe expense of accuracy The application of the fourth-orderCN approximation on the heat equation proves that thethree classical iterative methods also converge with JAC4appearing to be the slowest and the least accurate amongst all
Journal of Applied Mathematics 11
Table 4 Experiment 2 119898 = 700 120582 = 05 Δ119909 = 143 times 10minus3 Δ119905 =
102 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME n ET (s)IADEMF4(119903 = 07) 207119890 minus 8 231119890 minus 8 345119890 minus 8 100 903119890 minus 3
IADEMF2(119903 = 08) 106119890 minus 7 175119890 minus 7 175119890 minus 7 100 647119890 minus 3
SOR4(120596 = 11) 154119890 minus 6 171119890 minus 6 255119890 minus 6 200 110119890 minus 2
GS4 294119890 minus 6 327119890 minus 6 487119890 minus 6 250 118119890 minus 2
JAC4 124119890 minus 5 138119890 minus 5 206119890 minus 5 300 127119890 minus 2
Table 5 Experiment 2 119898 = 700 120582 = 10 Δ119909 = 143 times 10minus3 Δ119905 =
204 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 09) 138119890 minus 6 154119890 minus 6 229119890 minus 6 75 650119890 minus 3
IADEMF2(119903 = 05) 187119890 minus 6 207119890 minus 6 309119890 minus 6 75 472119890 minus 3
SOR4(120596 = 12) 312119890 minus 6 347119890 minus 6 517119890 minus 6 125 906119890 minus 3
GS4 667119890 minus 6 742119890 minus 6 111119890 minus 5 175 932119890 minus 3
JAC4 126119890 minus 5 141119890 minus 5 210119890 minus 5 250 105119890 minus 2
Table 6 Experiment 1 119898 = 7000 120582 = 05 Δ119909 = 143 times 10minus4 Δ119905 =
102 times 10minus8 119905 = 510 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME n ET (s)IADEMF4(119903 = 08) 367119890 minus 10 433119890 minus 10 366119890 minus 9 173 146119890 minus 1
IADEMF2(119903 = 05) 375119890 minus 10 435119890 minus 10 376119890 minus 9 200 124119890 minus 1
SOR4(120596 = 112) 455119890 minus 10 507119890 minus 10 367119890 minus 9 257 147119890 minus 1
GS4 110119890 minus 9 111119890 minus 9 376119890 minus 9 300 160119890 minus 1
JAC4 230119890 minus 9 231119890 minus 9 378119890 minus 9 400 163119890 minus 1
Table 7 Experiment 1 119898 = 7000 120582 = 10 Δ119909 = 143 times 10minus4 Δ119905 =
204 times 10minus8 119905 = 51 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 163119890 minus 7 163119890 minus 7 163119890 minus 7 112 960119890 minus 2
IADEMF2(119903 = 12) 163119890 minus 7 163119890 minus 7 163119890 minus 7 120 736119890 minus 2
SOR4(120596 = 115) 163119890 minus 7 163119890 minus 7 164119890 minus 7 171 999119890 minus 2
GS4 164119890 minus 7 164119890 minus 7 164119890 minus 7 216 107119890 minus 1
JAC4 165119890 minus 7 165119890 minus 7 165119890 minus 7 312 111119890 minus 1
It is found that the convergence rate of the fourth-orderand second-order methods improves with the application oflarger mesh size that is from 120582 = 05 to 120582 = 10 Thecoarser meshes cause a reduction in the computational oper-ations thus giving better rate of convergence However a
Table 8 Experiment 2 119898 = 7000 120582 = 05 Δ119909 = 143 times 10minus4 Δ119905 =
102 times 10minus8 119905 = 510 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 07) 567119890 minus 13 321119890 minus 12 130119890 minus 10 150 131119890 minus 1
IADEMF2(119903 = 07) 199119890 minus 11 460119890 minus 11 182119890 minus 9 200 127119890 minus 1
SOR4(120596 = 115) 245119890 minus 10 273119890 minus 10 399119890 minus 10 260 151119890 minus 1
GS4 398119890 minus 10 443119890 minus 10 661119890 minus 10 400 171119890 minus 1
JAC4 105119890 minus 9 117119890 minus 9 174119890 minus 9 550 176119890 minus 1
Table 9 Experiment 2 119898 = 7000 120582 = 10 Δ119909 = 143 times 10minus4 Δ119905 =
204 times 10minus8 119905 = 51 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 154119890 minus 6 171119890 minus 6 255119890 minus 6 96 912119890 minus 2
IADEMF2(119903 = 10) 154119890 minus 6 171119890 minus 6 255119890 minus 6 96 607119890 minus 2
SOR4(120596 = 12) 154119890 minus 6 171119890 minus 6 255119890 minus 6 192 997119890 minus 2
GS4 154119890 minus 6 171119890 minus 6 255119890 minus 6 288 128119890 minus 1
JAC4 154119890 minus 6 172119890 minus 6 256119890 minus 6 408 146119890 minus 1
more accurate solution is obtained by using finer mesh Forexample Table 6 shows that for 120582 = 05 the IADEMF4 hasRMSE = 433119890 minus 10 and 119899 = 173 whereas Table 7 shows thatfor 120582 = 10 its RMSE = 163119890 minus 7 and 119899 = 112 In generalamongst all the testedmethods the IADEMF4 still maintainsits greater accuracy characteristic even with coarser meshes
In terms of execution time the results from every tabledisplay shorter execution time for the IADEMF2 in compari-son to the IADEMF4 This is expected since the IADEMF2has lower computational complexity (Table 1) Despite theachievement in accuracy the IADEMF4 has to performmorecomputational work than the IADEMF2 since the former hasto utilize values of 119906 at two grid points on either side of thepoint (119894Δ119909 119896Δ119905) along the 119896th time level Thus if accuracyis desired then the preferred sequential numerical algorithmwould be the IADEMF4 On the other hand if execution timematters then the choice would be the IADEMF2
Amongst the fourth-order methods the IADEMF4 exe-cutes in the least amount of time Even though its computa-tional complexity is relatively large it operates with the leastnumber of iterations thus enabling it to be the most efficient
10 Conclusion
This paper proposes the development of the novel fourth-order IADEMF4 finite difference scheme The higher-orderscheme is developed by representing the spatial derivative inthe heat equation with the fourth-order finite difference CNapproximation This leads to the formation of pentadiagonalmatrices in the systems of linear equations with larger com-putational stencils The algorithm also employs the higheraccuracy of the Mitchell and Fairweather variant Despite
12 Journal of Applied Mathematics
the fourth-order IADE schemersquos higher computational com-plexity this technique is proved to be valuable becauseit enhances the accuracy of its second-order counterpartnamely the IADEMF2 In addition the higher accuracy ofIADEMF4 is verified as a convergent and unconditionallystable scheme and is superior in terms of rate of convergenceIt has also been proven to execute more efficiently in compar-ison to the other benchmarked fourth-order classical iterativemethods such as the GS4 the SOR4 and the JAC4 The in-creasing number of correct digits at each iteration serves as anadvantage for the IADEMF4 thus yielding faster rate of con-vergence with higher level of accuracy
In conclusion the proposed IADEMF4 scheme affordsusers many advantages with respect to higher accuracy sta-bility and rate of convergence and it serves as an alternativeefficient technique for the solution of a one-dimensional heatequation with Dirichlet boundary conditions
The proposed fourth-order scheme can be modified andadapted to more general multidimensional linear and non-linear parabolic elliptic and hyperbolic partial differentialequations using different types of boundary conditionsBesides it can also be considered for applications in problemsthat require higher-order accuracy with high resolution suchas problems in nanocomputing that require the solution ofvery large sparse systems of equation
The explicit and high accuracy feature of the IADEMF4can be exploited in two- and three-dimensional heat prob-lems by applying the scheme as pentadiagonal solvers forthe system of linear equations arising in the sweeps of ahigher-order alternating direction implicit (ADI) schemeThe lower-order ADI scheme was initially proposed byPeaceman and Rachford [24] The higher space dimensionsare expected not to cause unsatisfactory performance for theADI-IADEMF4 as long as the proposed scheme is stableconvergent and efficient as the process of its implementa-tion continues to advance from one-time step to anotherPathirana et al [25] implemented the ADI in developing atwo-dimensional model for incorporating flood damage inurban drainage planningThe proposed model is found to bestable numerically accurate and computationally efficientMirzavand et al [26] proposed the ADI-FDTD methodfor the physical modeling of high-frequency semiconductordevices The approach is able to reduce significantly the full-wave simulation time
It is possible to parallelize the IADEMF4 algorithm sincethe calculation of the new iteration only depends on theknown values from the last iteration (Figures 2 and 3) Futurework is to exploit the explicit computational properties andefficiency of the IADEMF4 for parallelization and executionon distributed parallel computing systems The idea is tospeed up the execution time without compromising its accu-racy especially on problems involving very large linear sys-tems of equations
References
[1] J Noye Computational Techniques for Differential Equationsvol 83 Elsevier Science North-Holland Publishing Amster-dam The Netherlands 1984
[2] G W Recktenwald ldquoFinite difference approximations to theheat equationrdquoMechanical Engineering vol 10 pp 1ndash27 2004
[3] P Lax and B Wendroff ldquoSystems of conservation lawsrdquo Com-munications on Pure and Applied Mathematics vol 13 pp 217ndash237 1960
[4] R J MacKinnon and G F Carey ldquoAnalysis of material interfacediscontinuities and superconvergent fluxes in finite differencetheoryrdquo Journal of Computational Physics vol 75 no 1 pp 151ndash167 1988
[5] R Hixon and E Turkel High Accuracy Compact MacCormack-Type Schemes For Computational Aerocoustics NASACentre forAerospace Information Lewis Research Centre 1998
[6] Y Lin Higher-order finite difference methods for solving heatequations [PhD thesis] Southern Illinois University 2008
[7] J Sulaiman M K Hasan M Othman M and S A AbdulKarim ldquoFourth-order QSMOR iterative method for the solu-tion of one-dimensional parabolic PDErsquosrdquo in Proceedings of theInternational Conference on Industrial and AppliedMathematics(CIAM rsquo10) pp 34ndash39 ITB Bandung Indonesia 2010
[8] N Jha ldquoThe application of sixth order accurate parallel quar-ter sweep alternating group explicit algorithm for nonlinearboundary value problems with singularityrdquo in Proceedings ofthe 2nd International Conference on Methods and Models inComputer Science (ICM2CS rsquo10) pp 76ndash80 December 2010
[9] Y Jin G Jin and J Li ldquoA class of high-precision finite differenceparallel algorithms for convection equationsrdquo Journal of Conver-gence Information Technology vol 6 no 1 pp 79ndash82 2011
[10] W Fu and E L Tan ldquoDevelopment of split-step FDTDmethodwith higher-order spatial accuracyrdquo Electronics Letters vol 40no 20 pp 1252ndash1253 2004
[11] A Mohebbi andM Dehghan ldquoHigh-order compact solution ofthe one-dimensional heat and advection-diffusion equationsrdquoApplied Mathematical Modelling vol 34 no 10 pp 3071ndash30842010
[12] W Liao ldquoAn implicit fourth-order compact finite differencescheme for one-dimensional Burgersrsquo equationrdquo Applied Math-ematics and Computation vol 206 no 2 pp 755ndash764 2008
[13] Z F Tian and Y B Ge ldquoA fourth-order compact ADI methodfor solving two-dimensional unsteady convection-diffusionproblemsrdquo Journal of Computational and Applied Mathematicsvol 198 no 1 pp 268ndash286 2007
[14] C Chun ldquoSome fourth-order iterativemethods for solving non-linear equationsrdquo Applied Mathematics and Computation vol195 no 2 pp 454ndash459 2008
[15] S Zhu Z Yu and J Zhao ldquoAhigh-order parallel finite differencealgorithmrdquo Applied Mathematics and Computation vol 183 no1 pp 365ndash372 2006
[16] Z Gao and S Xie ldquoFourth-order alternating direction implicitcompact finite difference schemes for two-dimensionalSchrodinger equationsrdquo Applied Numerical Mathematics vol61 no 4 pp 593ndash614 2011
[17] M S Sahimi A Ahmad and A A Bakar ldquoThe Iterative Alter-nating Decomposition Explicit (IADE) method to solve theheat conduction equationrdquo International Journal of ComputerMathematics vol 47 pp 219ndash229 1993
[18] G D Smith Numerical Solution of Partial Differential Equa-tions Finite Difference Methods Oxford University Press 2ndedition 1978
[19] A RMitchell andG Fairweather ldquoImproved forms of the alter-nating direction methods of Douglas Peaceman and Rachfordfor solving parabolic and elliptic equationsrdquo Numerische Math-ematik vol 6 no 1 pp 285ndash292 1964
Journal of Applied Mathematics 13
[20] B N DattaNumerical Linear Algebra and Applications BrooksCole Publishing Pacific Grove Calif USA 1995
[21] D M Young Iterative Solution of Large Linear Systems Aca-demic Press New York NY USA 1971
[22] V K Saulev Integration of Equations of Parabolic Type by theMethod of Nets Pergamon Press Oxford UK 1964
[23] L W Johnson and R D Riess Numerical Analysis Addison-Wesley Reading Mass USA 2nd edition 1982
[24] D W Peaceman and H H Rachford Jr ldquoThe numerical solu-tion of parabolic and elliptic differential equationsrdquo Journal ofthe Society for Industrial and Applied Mathematics vol 3 pp28ndash41 1955
[25] A Pathirana S Tsegaye B Gersonius and K VairavamoorthyldquoA simple 2-D inundation model for incorporating flood dam-age in urban drainage planningrdquo Hydrology and Earth SystemSciences vol 15 no 8 pp 2747ndash2761 2011
[26] R Mirzavand A Abdipour G Moradi and M MovahhedildquoFull-wave semiconductor devices simulation using ADI-FDTD methodrdquo Progress in Electromagnetics Research vol 11pp 191ndash202 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
4 Journal of Applied Mathematics
Substitute the following expression obtained from (10)into (11)
u(119901+12) = (119903119868 + 1198661)minus1
[(119903119868 minus 1198921198662) u(119901) + f] (12)
and then simplify to obtain
(119903119868 + 1198662) u(119901+1) = (119903119868 minus 119892119866
1) (119903119868 + 119866
1)minus1
(119903119868 minus 1198921198662) u(119901)
+ [ (119903119868 minus 1198921198661) (119903119868 minus 119892119866
1)minus1
+ 119892] f (13)
As 119901 in (13) becomes sufficiently large the temperaturesolution reaches a steady state that is as 119901 rarr infin thenu(119901+1) rarr u and u(119901) rarr u Simplify the equation and thenmultiply by (119903119868+119866
1) After some algebraicmanipulations the
following expression is obtained
[119903 (1198661+ 1198662) (1 + 119892) + (1 minus 119892
2
)11986611198662] u = 119903 (1 + 119892) f (14)
Multiply (14) by 1119903(1 + 119892) and use the definition of 119892 tofinally obtain the following form
[ (1198661+ 1198662) minus
1
611986611198662] u = f (15)
The comparison between the expression 119860u = f in (8)and the form given in (15) suggests that the coefficient matrix119860 for the IADEMF4 can be decomposed into
119860 = 1198661+ 1198662minus1
611986611198662 (16)
The IADEMF4 requires the constituentmatrices1198661and119866
2in
(16) to be in the form of lower and upper tridiagonalmatricesrespectively in order to retain the pentadiagonal structure of119860 Thus
1198661=
[[[[[[[
[
1
1198971
1 119874
1
1198972
d2d dd 119897119898minus4
1
119874 119898minus4
119897119898minus3
1
]]]]]]]
](119898minus2)times(119898minus2)
1198662=
[[[[[[[
[
11989011
V1
1198902
2
V2
119874
d d d119874 119890
119898minus4119898minus4
V119898minus4
119890119898minus3
119898minus3
119890119898minus2
]]]]]]]
](119898minus2)times(119898minus2)
(17)
By substituting 1198661and 119866
2in (17) into the formula for the
decomposition of 119860 the entries of the resultant matrix arecompared with the entries of 119860 in (8) yielding the followingdefinitions
1198901=6 (119888 minus 1)
5
1=6119889
5 119897
1=
6119887
6 minus 1198901
1198902=6 (119888 minus 1) + 119897
11
5 V
119894=6119890
5for 119894 = 1 2 119898 minus 4
(18)
And for 119894 = 2 3 119898 minus 3
119894=6119889 + 119897
119894minus1V119894minus1
5
119894minus1=
6119886
6 minus 119890119894minus1
119897119894=6119887 +
119894minus1119894minus1
6 minus 119890119894
119890119894+1
=6 (119888 minus 1) + 119897
119894119894+ 119894minus1
V119894minus1
5
(19)
Since1198661and119866
2are three bandedmatrices then (119903119868+119866
1)
and (119903119868+1198662) can be inverted easilyThe equations in (10) and
(11) are rearranged as in (20) and (21) respectively
u(119901+12) = (119903119868 + 1198661)minus1
(119903119868 minus 1198921198662) u(119901)
+ (119903119868 + 1198661)minus1f
(20)
u(119901+1) = (119903119868 + 1198662)minus1
(119903119868 minus 1198921198661) u(119901+12)
+ 119892(119903119868 + 1198662)minus1f
(21)
The above two equations are computed and simplifiedleading to the computational formulae at each of the halfiteration levels as given in (22) and (23)
(i) At the (119901 + 12) iteration level
119906(119901+12)
119894=
1
119877(119864119894minus1
119906(119901)
119894+119882119894minus1
119906(119901)
119894+1+ 119881119894minus1
119906(119901)
119894+2
minus119894minus3
119906(119901+12)
119894minus2minus 119897119894minus2
119906(119901+12)
119894minus1+ 119891119894)
119894 = 2 3 119898 minus 2119898 minus 1
(22)
(ii) At the (119901 + 1) iteration level
119906(119901+1)
119894=
1
119885119894minus1
(119878119894minus3
119906(119901+12)
119894minus2+ 119876119894minus2
119906(119901+12)
119894minus1+ 119875119906(119901+12)
119894
minus119894minus1
119906(119901+1)
119894+1minus V119894minus1
119906(119901+1)
119894+2+ 119892119891119894)
119894 = 119898 minus 1 119898 minus 2 3 2
(23)
with
minus1
= 0= 1198970= 119881119898minus2
= 119881119898minus3
= 119882119898minus2
= 119898minus2
= V119898minus2
= V119898minus3
= 1198760= 119878minus1
= 1198780= 0
119877 = 1 + 119903 119875 = 119903 minus 119892
119864119894= 119903 minus 119892119890
119894 119885119894= 119903 + 119890
119894 119894 = 1 2 119898 minus 2
119882119894= minus119892
119894 119876119894= minus119892119897
119894 119894 = 1 2 119898 minus 3
119881119894=minus119892V119894 119878119894= minus119892
119894 119894 = 1 2 119898 minus 4
(24)
The IADEMF4 algorithm is regarded as a two-stageprocess involving two iteration levels (119901 + 12) and (119901 + 1)It is completed explicitly by using the required equations atthe two levels in alternate sweeps along all the grid pointsin the interval (0 1) until convergence is reached In (22)the calculation to determine the unknown 119906
(119901+12)
119894begins at
the left boundary and then moves to the right In a similar
Journal of Applied Mathematics 5
manner 119906(119901+1)119894
in (23) is calculated by proceeding from theright boundary towards the left (Figure 1)
The computational molecules are depicted in Figures 2and 3
At each level of iteration the computational moleculesinvolve two known grid points at the new level and anotherthree known ones at the old level Clearly the method isexplicit
4 Convergence Analysis of the IADEMF4
This section proves the convergence of the IADEMF4 Since120582 gt 1will not guarantee an accurate approximation for 120597119880120597119905[18] the values of 120582 that are considered appropriate for theproof are 0 lt 120582 le 1
From the definitions in ((18) and (19)) the followingresults are obtained
1198901=6 (119888 minus 1)
5=3120582
2 implying that 0 lt 119890
1le3
2
1=6119889
5lt0 119897
1=
6119887
6 minus 1198901
lt0 V119894=6119890
5gt 0
for 119894 = 1 2 119898 minus 4
2=6119889 + 119897
1V1
5= 1+1198971V1
5lt 1 since 119897
1V1lt 0
1198902=6 (119888 minus 1) + 119897
11
5= 1198901+11989711
5gt 1198901 since 119897
11gt 0
(25)
Simple computation shows that 119897115 lt 1 Clearly 119890
2lt 6
Lemma 1 119890119894in ((18) and (19)) is such that 0 lt 119890
119894lt 6 for all
119894 = 1 2 3 119898 minus 2
Proof The results in (25) show that 6 gt 1198902gt 1198901gt 0 Assume
it is true that 6 gt 119890119896gt 119890119896minus1
gt 0 for all 119896 = 3 4 5 119898 minus 3Since the assumption implies that 6 gt 119890
119896minus1gt 119890119896minus2
gt 0 then119896minus1
= 6119886(6 minus 119890119896minus1
) gt 119896minus2
= 6119886(6 minus 119890119896minus2
) gt 0 Therefore119896minus1
V119896minus1
gt 119896minus2
V119896minus2
It has also been shown that
2lt 1lt 0 Assume it is true
that 119895lt 119895minus1
lt 0 for all 119895 = 3 4 119898 minus 4 The assumptionimplies that
119895minus1lt 119895minus2
lt 0 Thus 119895minus1
119895minus1
lt 119895minus2
119895minus2
lt 0
119895+1
=6119889 + 119897
119895V119895
5= 1+119897119895V119895
5= 1+
(6119887 + 119895minus1
119895minus1
) V119895
5 (6 minus 119890119895)
lt 1+
(6119887 + 119895minus2
119895minus2
) V119895minus1
5 (6 minus 119890119895minus1
)
= 1+119897119895minus1
V119895minus1
5= 119895lt 0
(26)
Since it is true that for 119895 + 1 119895+1
lt 119895lt 0 then by
induction the assumption 119895lt 119895minus1
lt 0 is true for all 119895 An
0
u(p+12)
i
u(p+1)
i
i iminus 1 i + 1 m + 1
Figure 1 The two-stage IADEMF4 algorithmThe directions of thesweeps at the (119901 + 12) and (119901 + 1) iteration levels
i iminus 1i minus 2 i + 1 i + 2
(p + 12)
(p)
Figure 2 Computational molecule of the IADEMF4 at the (119901+12)iteration level
i iminus 1i minus 2 i + 1 i + 2
(p + 12)
(p + 1)
Figure 3 Computational molecule of the IADEMF4 at the (119901 + 1)
iteration level
equivalent to the preceding statement would be 119896lt 119896minus1
lt 0
for all 119896 = 3 4 5 119898 minus 3 It follows that
119897119896119896=(6119887 +
119896minus1119896minus1
) 119896
6 minus 119890119896
gt(6119887 +
119896minus2119896minus2
) 119896minus1
6 minus 119890119896minus1
= 119897119896minus1
119896minus1
119890119896+1
=6 (119888 minus 1) + 119897
119896119896+ 119896minus1
V119896minus1
5= 1198901+119897119896119896+ 119896minus1
V119896minus1
5
gt 1198901+119897119896minus1
119896minus1
+ 119896minus2
V119896minus2
5= 119890119896gt 0
(27)
Since it is true that for 119896+1 119890119896+1
gt 119890119896gt 0 then by induction
119890119896gt 119890119896minus1
gt 0 is also true for all 119896Suppose there is a 119896 such that 119890
119896gt 6 Then
119890119896+1
= 1198901+119897119896119896+ 119896minus1
V119896minus1
5
= 1198901+
6119887 + 119896
5 (6 minus 119890119896)+119896minus1
5(119896minus1
119896
6 minus 119890119896
+ V119896minus1
) lt 1198901
(28)
This is a contradiction since 119890119896+1
gt 1198901 This verifies that 6 gt
119890119896gt 119890119896minus1
gt 0 119896 = 3 4 5 119898 minus 3 Let 119894 = 2 3 4 119898 minus 2and then 6 gt 119890
119894gt 119890119894minus1
gt 0 Therefore 0 lt 119890119894lt 6 119894 =
1 2 3 119898 minus 2 Lemma 1 is proved
6 Journal of Applied Mathematics
Lemma 2 If 119903 gt 0 and 119892 = (6+ 119903)6 thenmax |(119903 minus119892119890119894)(119903 +
119890119894)| lt 1 i = 1 2 3 119898 minus 2
Proof Let max |(119903 minus 119892119890119894)(119903 + 119890
119894)| = |(119903 minus 119892119890
119895)(119903 + 119890
119895)| for
some 119895Assume |(119903 minus 119892119890
119895)(119903 + 119890
119895)| ge 1 Since 119903 gt 0 and 119890
119895gt 0
then 119903 + 119890119895gt 0
If (119903 minus 119892119890119895)(119903 + 119890
119895) ge 1 then minus((6 + 119903)6)119890
119895ge 119890119895 which
implies that 119903 le minus12 This contradicts the fact that 119903 gt 0If (119903 minus 119892119890
119895)(119903 + 119890
119895) le minus1 then 2119903 minus ((6 + 119903)6)119890
119895le minus119890119895
which implies that 119890119895ge 12 This contradicts Lemma 1
So the assumption that |(119903 minus 119892119890119895)(119903 + 119890
119895)| ge 1 is false
Hence max |(119903 minus 119892119890119894)(119903 + 119890
119894)| = |(119903 minus 119892119890
119895)(119903 + 119890
119895)| lt 1
Proposition 3 (119903119868 minus 1198921198661)(119903119868 + 119866
1)minus1
2lt 1
Proof Let 119865 = (119903119868 minus 1198921198661)(119903119868 + 119866
1)minus1 that is
119865 =
[[[[[[[[[[
[
119903 minus 119892
119889
d119903 minus 119892
119889119874
d d d
d d
sdot sdot sdot sdot sdot sdot sdot sdot sdot sdot sdot sdot119903 minus 119892
119889
]]]]]]]]]]
]
(29)
119865 is a lower triangular matrix with all the diagonal entries(whence the eigenvalues of 119865) equal to (119903 minus 119892)119889 where 119889 =
119903 + 1 Denote all the eigenvalues of 119865 by 120582119865 If 120588[119865] is defined
as the spectral radius of 119865 then
120588 [119865] = max 10038161003816100381610038161205821198651003816100381610038161003816 =
10038161003816100381610038161205821198651003816100381610038161003816 =
1003816100381610038161003816100381610038161003816
119903 minus 119892
119903 + 1
1003816100381610038161003816100381610038161003816le 119865
2 (30)
But by definition of 2-norm
1198652= maxx2 = 0
119865x2
x2
= maxy2=1
1003817100381710038171003817119865y10038171003817100381710038172
1003817100381710038171003817y10038171003817100381710038172
= maxy2=1
1003817100381710038171003817119865y10038171003817100381710038172
le maxw2=1
119865w2= maxw2=1
1003817100381710038171003817120582119865w10038171003817100381710038172
(31)
Since all the eigenvalues of 119865 are equal then
1198652le1003816100381610038161003816120582119865
1003816100381610038161003816 maxw2=1
w2=1003816100381610038161003816120582119865
1003816100381610038161003816 = 120588 [119865] (32)
Thus from (30) and (32) 1198652= |120582119865| = |(119903 minus 119892)(119903 + 1)|
By Lemma 2 with 119890119895replaced by 1 |(119903 minus 119892)(119903 + 1)| lt 1
is obtained leading to the result of Proposition 3 which is1198652= (119903119868 minus 119892119866
1)(119903119868 + 119866
1)minus1
2lt 1
Proposition 4 (119903119868 minus 1198921198662)(119903119868 + 119866
2)minus1
2lt 1
Proof Let 119870 = (119903119868 minus 1198921198662)(119903119868 + 119866
2)minus1 that is
119870 =
[[[[[[[[[[[[[
[
119903 minus 1198921198901
1199111
d sdot sdot sdot sdot sdot sdot sdot sdot sdot
119903 minus 1198921198902
1199112
d sdot sdot sdot
d
119874 d d
119903 minus 119892119890119898
119911119898
]]]]]]]]]]]]]
]
(33)
119870 is an upper triangular matrix with all the diagonal entriesequal to (119903 minus 119892119890
119894)119911119894 119894 = 1 2 119898 minus 2 where 119911
119894= 119903 + 119890
119894
Since all the eigenvalues of119870 are distinct then119870 is similar toa diagonal matrix119863
By the Schur triangularization theorem [20] there is anorthogonal matrix 119874 such that 119874119879119870119874 = 119863 The diagonalentries of119863 are the eigenvalues of119870
1198702=10038171003817100381710038171003817119874119879
119870119874100381710038171003817100381710038172
= 1198632
= radicmaximum eigenvalue of 119863119879119863
= radicmaximum eigenvalue of 1198632
= radic120588 (1198632) = radic120588 (1198702) = 120588 (119870)
(34)
By Lemma 2
1198702= 120588 (119870) = max
10038161003816100381610038161003816100381610038161003816
119903 minus 119892119890119894
119903 + 119890119894
10038161003816100381610038161003816100381610038161003816
=
1003816100381610038161003816100381610038161003816100381610038161003816
119903 minus 119892119890119895
119903 + 119890119895
1003816100381610038161003816100381610038161003816100381610038161003816
lt 1 (35)
for some 119895 Hence this proves Proposition 4
From (20) and (21)
u(119901+1) = (119903119868 + 1198662)minus1
(119903119868 minus 1198921198661)
times [(119903119868 + 1198661)minus1
(119903119868 minus 1198921198662) u(119901) + (119903119868 + 119866
1)minus1f]
+ 119892 (119903119868 + 1198662)minus1f
(36)
Let
119872(119903) = (119903119868 + 1198662)minus1
(119903119868 minus 1198921198661) (119903119868 + 119866
1)minus1
(119903119868 minus 1198921198662)
(37)
And let
119902 (119903) = [(119903119868 + 1198662)minus1
(119903119868 minus 1198921198661) (119903119868 + 119866
1)minus1
+119892(119903119868 + 1198662)minus1
] f (38)
Then u(119901+1) = 119872(119903)u(119901) + 119902(119903)
Theorem 5 The IADEMF4 is convergent if 120588[119872(119903)] lt 1 for119903 gt 0
Journal of Applied Mathematics 7
Proof Define (119903) = (119903119868 + 1198662)119872(119903)(119903119868 + 119866
2)minus1 then
(119903) = (119903119868 minus 1198921198661) (119903119868 + 119866
1)minus1
(119903119868 minus 1198921198662) (119903119868 + 119866
2)minus1
(39)
Thus by similarity119872(119903) and (119903) have the same set of eigen-values Therefore
120588 [119872 (119903)]
= 120588 [ (119903)] le10038171003817100381710038171003817 (119903)
10038171003817100381710038171003817 2
=10038171003817100381710038171003817(119903119868 minus 119892119866
1) (119903119868 + 119866
1)minus1
(119903119868 minus 1198921198662) (119903119868 + 119866
2)minus110038171003817100381710038171003817 2
le10038171003817100381710038171003817(119903119868 minus 119892119866
1) (119903119868 + 119866
1)minus1100381710038171003817100381710038172
10038171003817100381710038171003817(119903119868 minus 119892119866
2) (119903119868 + 119866
2)minus110038171003817100381710038171003817 2
lt 1
(40)
The last inequality is due to Propositions 3 and 4The provenTheorem 5 assures the convergence of the IADEMF4
5 The Fourth-Order ClassicalIterative Methods
The system of linear equations in (8) may also be solvedby using the classical iterative methods of the fourth orderThey include the JAC4 the GS4 and the SOR4 methodsThese iterative methods are capable of exploiting the sparsestructure of the pentadiagonal matrix
The JAC4 algorithm can be represented by
119906(119901+1)
119894=
(119891119894minus119886119906(119901)
119894minus2minus 119887119906(119901)
119894minus1minus 119889119906(119901)
119894+1minus 119890119906(119901)
119894+2)
119888
119894 = 2 3 119898 minus 1
(41)
The approximation of 119906(119901+1)119894
at the (119901 + 1)th iteration level iscomputed using the relevant values in the 119901th iteration
The GS4 method uses the most recent values of 119906(119901+1)119894minus2
and 119906(119901+1)
119894minus1to update the approximation value of 119906(119901+1)
119894 The
algorithm for the GS4 is as expressed in (42)
119906(119901+1)
119894=
(119891119894minus119886119906(119901+1)
119894minus2minus 119887119906(119901+1)
119894minus1minus 119889119906(119901)
119894+1minus 119890119906(119901)
119894+2)
119888
119894 = 2 3 119898 minus 1
(42)
The SOR4 iterative method accelerates the convergencerate of the GS4 If 120596 is a relaxation parameter then for any120596 = 0 (42) can be rewritten as
119906(119901+1)
119894= (1 minus 120596) 119906
(119901)
119894
+
120596 (119891119894minus 119886119906(119901+1)
119894minus2minus 119887119906(119901+1)
119894minus1minus 119889119906(119901)
119894+1minus 119890119906(119901)
119894+2)
119888
119894 = 2 3 119898 minus 1
(43)
i iminus 1i minus 2 i + 1 i + 2
(p + 1)
(p)
Figure 4 Computational molecule of the JAC4
i iminus 1i minus 2 i + 1 i + 2
(p + 1)
(p)
Figure 5 Computational molecule of the GS4SOR4
Table 1 Sequential arithmetic operations per iteration (119898 problemsize)
Method Number ofadditions
Number ofmultiplications
Total operationcount
IADEMF4 10(119898 minus 2) 13(119898 minus 2) 23(119898 minus 2)
IADEMF2 6119898 9119898 15119898
JAC4GS4 4(119898 minus 2) 5(119898 minus 2) 9(119898 minus 2)
SOR4 5(119898 minus 2) 7(119898 minus 2) 12(119898 minus 2)
The SOR4 algorithm reduces to the GS4 if 120596 = 1 Exceptfor very special cases it is difficult to obtain the analyticexpression for 120596 According to Young [21] the precise deter-mination of the optimal 120596 in the SOR is only known for asmall class of matrices The value of 120596 generally lies in therange of 1 lt 120596 lt 2
The computational molecules for the JAC4 and the GS4are as illustrated in Figures 4 and 5 respectively The SOR4has the same form as the GS4
6 Computational Complexity
The cost of implementing an algorithm can be assessedby examining its computational complexity The number ofarithmetic operations such as additions (including subtrac-tions) andmultiplications (including divisions) that is neededto perform by the algorithm can be straightforwardly count-ed Table 1 gives the number of sequential arithmetic opera-tions per iteration that is required to evaluate the algorithms
For the higher-order schemes trade-offbetween accuracyand speed usually happens The computational complexityhas an effect on the efficiency of a particular scheme Thusthis factor will be taken into account when discussing theresults of the numerical experiments conducted in this study
7 The Pseudocode of the IADEMF4Sequential Algorithm
Algorithm 1 illustrates the pseudocode of the IADEMF4sequential algorithm This algorithm can also be generalizedand applied to other methods discussed in this paper
8 Journal of Applied Mathematics
begindetermine the parameters119898 119903 Δ119905 Δ119909 120582 120576 119879
determine initial conditions 119906(119901)119894
determine exact solutions 119880119894at time level 119879
while (time level lt 119879)determine boundary conditionsfor 119894 = 2 to 119894 = 119898 minus 1
compute 119891119894(refer to (9))
end-forset iteration = 0while (convergence conditions are not satisfied)
for 119894 = 2 to 119894 = 119898 minus 1
compute 119906(119901+12)119894
(refer to (22))end-forfor 119894 = 2 to 119894 = 119898 minus 1
compute 119906(119901+1)119894
(refer to (23))end-fortest for convergencecompute 119890
119894larr
10038161003816100381610038161003816119906(119901+1)
119894minus 119906(119901)
119894
10038161003816100381610038161003816if (max1003816100381610038161003816119890119894
1003816100381610038161003816 lt 120576)then 119906
(119901)
119894larr 119906(119901+1)
119894
add 1 to iteration (if necessary)end-while
end-whilefor 119894 = 2 to 119894 = 119898 minus 1
determine average absolute error root meansquare error and maximum error
end-forend
Algorithm 1 The IADEMF4 sequential algorithm
8 Numerical Experiments
Two experiments were conducted to test the sequentialnumerical performance of the proposed IADEMF4 methodagainst those of the benchmarked classical iterative methodsnamely the JAC4 theGS4 and the SOR4 Comparison is alsomade with the corresponding IADE method of the secondorder
Experiment 1 This problem was taken from Saulev [22]
120597119880
120597119905=1205972
119880
1205971199092 0 le 119909 le 1 (44)
subject to the initial condition
119880 (119909 0) = 4119909 (1 minus 119909) 0 le 119909 le 1 (45)
and the boundary conditions
119880 (0 119905) = 119880 (1 119905) = 0 119905 ge 0 (46)
The exact solution to the given problem is given by
119880 (119909 119905) =32
1205873
infin
sum
119896=1(2)
1
1198963119890minus12058721198962119905 sin (119896120587119909) (47)
Experiment 2 This problem was taken from Johnson andReiss [23]
120597119880
120597119905=1205972
119880
1205971199092 0 le 119909 le 1 (48)
subject to the initial condition
119880 (119909 0) = sin (120587119909) (1 + 6 cos (120587119909)) 0 le 119909 le 1 (49)
and the boundary conditions
119880 (0 119905) = 119880 (1 119905) = 0 119905 ge 0 (50)
The exact solution to the given problem is given by
119880 (119909 119905) = sin (120587119909) 119890minus1205872119905
+ 3 sin (2120587119909) 119890minus41205872119905
(51)
Journal of Applied Mathematics 9
9 Results and Discussion
In each experiment the exact solutions at 119894 = 1 and 119894 = 119898
were taken as boundary values for the IADEMF4 and theother fourth-order classical iterative methods The conver-gence criterion used in the testing of each method was takenas 119906(119901+1) minus 119906
(119901)
infin
le 120576 where 120576 is the convergence toleranceThe selections of the optimum 119903 or 120596 were determined byexperiments As for the IADEMF2 the CN scheme with 120579 =
12 was employedFigures 6 7 8 and 9 visualize the behavior of the one-
dimensional parabolic heat solutions for both experimentsBy using119898 = 10 and a tolerance requirement of 120576 = 10
minus4 twodifferent mesh sizes 120582 = 05 and 120582 = 10 were consideredin each experiment The exact solutions are compared withthe IADEMF4 and the IADEMF2 numerical solutions Theoptimum value of 119903 chosen for each method as well as theircorresponding outcome of the number of iterations 119899 isstated in the legend of each figure Every figure reveals thatthe IADEMF4 is more accurate than the IADEMF2 andthe former converges with fewer numbers of iterations incomparison to the latter The numerical solutions of theIADEMF4 seem to be in very good agreement with theexact solutions For example in Figure 7 at 119909 = 05 thedifference between the exact solution and the numericalsolution using the IADEMF2 and using the IADEMF4 isabout 36 and 0 respectively These results imply thatthe accuracy and convergence rate of the second-order IADEmethod are enhanced by the implementation of the fourth-order CN approximation that leads to the formation of thecorresponding fourth-order IADE scheme
Figure 10 displays the graph of log (RMSE) versus log(Δ119909)for decreasing values of Δ119909 implemented on Experiment 2By considering 120576 = 10
minus4 and fixing Δ119905 = 00001 the value ofΔ119909 was initially taken as Δ119909 = 0125 It was then successivelyhalved into Δ1199092 Δ1199094 and Δ1199098
The figure shows that amongst the tested methods theroot mean square error (RMSE) of the IADEMF4 and theIADEMF2 decreases linearly as the values of Δ119909 decreasesThe slope of the IADEMF4 is approximately equal to 4which corresponds to its fourth-order spatial accuracy TheIADEMF2 with a second-order spatial accuracy has a slopethat is approximately equal to 2 It is clear that the IADEMF4is always more accurate than the IADEMF2 for the differentconsidered mesh sizes The graphs of the SOR4 (120596 = 105)the GS4 and the JAC4 show that their accuracies of fourthorder tend to lack as Δ119909 decreases largely due to the effect ofincreasing round-off errors as the value of119898 increases
Tables 2 3 4 5 6 7 8 and 9 provide numerical results interms of the average absolute error (AAE) root mean squareerror (RMSE) maximum error (ME) number of iterations(119899) and execution time (ET) measured in seconds (s) Theresults are obtained from both experiments for two differentvalues of 119898 and mesh size 120582 It is generally observed thatwhen 119898 = 700 120582 = 05 and 120576 = 10
minus6 the IADEMF4has the least average absolute error root mean square errorand maximum error in comparison with the other methodsunder consideration (Tables 2ndash5) When the size of 119898 was
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 08 n = 172)IADEMF2 (r = 03 n = 185)
00E + 00
10E minus 02
20E minus 02
30E minus 02
40E minus 02
50E minus 02
60E minus 02
70E minus 02
80E minus 02
90E minus 02
10E minus 01
U(xt)
Figure 6 Numerical and exact solutions for Experiment 1 120582 = 05Δ119909 = 01 Δ119905 = 0005 119905 = 025 and 120576 = 10
minus4
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 14 n = 159)IADEMF2 (r = 05 n = 162)
00E + 00
10E minus 03
20E minus 03
30E minus 03
40E minus 03
50E minus 03
60E minus 03
70E minus 03
80E minus 03
90E minus 03
U(xt)
Figure 7 Numerical and exact solutions for Experiment 1 120582 = 10Δ119909 = 01 Δ119905 = 001 119905 = 05 and 120576 = 10
minus4
ten times bigger (119898 = 7000) and a more stringent tolerancecriterion was set (120576 = 10
minus10
) the accuracy of the IADEMF4clearly outperforms the other methods for a mesh size of120582 = 05 (Tables 6 and 8) However the difference in the errorsamongst the tested methods is not so obvious for the case of120582 = 10 (Tables 7 and 9)The achievement of the fourth-orderIADEmethod in Experiments 1 and 2 can be clearly seen fromthe results in Tables 2 and 8 respectively where it has caused ahuge 94 reduction of RMSE from its corresponding second-order IADE method
The IADEMF4 has the advantage of featuring higheraccuracy due to the execution of the fourth-order CNapprox-imation coupled with the fourth-order accurate MF variant
10 Journal of Applied Mathematics
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 10 n = 169)IADEMF2 (r = 03 n = 213)
00E + 00
10E minus 02
20E minus 02
30E minus 02
40E minus 02
50E minus 02
60E minus 02
70E minus 02
80E minus 02
90E minus 02
10E minus 01
U(xt)
Figure 8 Numerical and exact solutions for Experiment 2 120582 = 05Δ119909 = 01 Δ119905 = 0005 119905 = 025 and 120576 = 10
minus4
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 10 n = 175)IADEMF2 (r = 05 n = 182)
00E + 00
10E minus 03
20E minus 03
30E minus 03
40E minus 03
50E minus 03
60E minus 03
70E minus 03
80E minus 03
U(xt)
Figure 9 Numerical and exact solutions for Experiment 2 120582 = 10Δ119909 = 01 Δ119905 = 001 119905 = 05 and 120576 = 10
minus4
The IADEMF2 is only derived from the second-order CNapproximation but its combination with its correspondingfourth-order MF variant managed to produce errors whichare better than theGS4 JAC4 and the SOR4 Even though theclassical iterative methods are also derived from the fourth-order accurate CN type approximation they are lackingin accuracy due to the round-off errors that have beenaccumulated from the time the execution starts till it ends
With regards to the rate of convergence the resultsfrom each table demonstrate that the number of iterationsproduced by the IADEMF2 and the fourth-order classicaliterative methods is greater than or at least equal to that of
0
minus2
minus208 minus277 minus347 minus416
minus4
minus8
minus6
minus10
minus12
minus14
minus16
minus18
GS4JAC4SOR4
log
(RM
SE)
log(Δx)
IADEMF4
IADEMF2
Figure 10 Log(RMSE) versus Log(Δ119909) Experiment 2 Δ119905 = 00001119905 = 0002 and 120576 = 10
minus4
Table 2 Experiment 1 119898 = 700 120582 = 05 Δ119909 = 143 times 10minus3 Δ119905 =
102 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 07) 267119890 minus 9 330119890 minus 9 212119890 minus 8 100 440119890 minus 2
IADEMF2(119903 = 08) 188119890 minus 8 495119890 minus 8 323119890 minus 7 100 635119890 minus 3
SOR4(120596 = 12) 540119890 minus 7 544119890 minus 7 552119890 minus 7 108 460119890 minus 2
GS4 540119890 minus 6 543119890 minus 6 332119890 minus 6 150 472119890 minus 2
JAC4 228119890 minus 5 229119890 minus 5 232119890 minus 5 150 473119890 minus 2
Table 3 Experiment 1 119898 = 700 120582 = 10 Δ119909 = 143 times 10minus3 Δ119905 =
204 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 297119890 minus 7 301119890 minus 7 301119890 minus 7 50 201119890 minus 2
IADEMF2(119903 = 09) 557119890 minus 7 559119890 minus 7 848119890 minus 7 50 317119890 minus 3
SOR4(120596 = 12) 329119890 minus 6 331119890 minus 6 335119890 minus 6 76 225119890 minus 2
GS4 895119890 minus 6 900119890 minus 6 900119890 minus 6 100 238119890 minus 2
JAC4 214119890 minus 5 215119890 minus 5 217119890 minus 5 125 250119890 minus 2
the IADEMF4 The convergence rate of the latter surpassesthe others in the case of 119898 = 7000 and 120576 = 10
minus10 (Tables6ndash8) Even though the operational count of the IADEMF4is relatively quite large (Table 1) due to its higher level ofaccuracy its increasing number of correct digits at eachiteration causes it to converge at a faster rate The IADEMF2has the advantage of having less mathematical operationsthus it is competitive in terms of convergence rate but atthe expense of accuracy The application of the fourth-orderCN approximation on the heat equation proves that thethree classical iterative methods also converge with JAC4appearing to be the slowest and the least accurate amongst all
Journal of Applied Mathematics 11
Table 4 Experiment 2 119898 = 700 120582 = 05 Δ119909 = 143 times 10minus3 Δ119905 =
102 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME n ET (s)IADEMF4(119903 = 07) 207119890 minus 8 231119890 minus 8 345119890 minus 8 100 903119890 minus 3
IADEMF2(119903 = 08) 106119890 minus 7 175119890 minus 7 175119890 minus 7 100 647119890 minus 3
SOR4(120596 = 11) 154119890 minus 6 171119890 minus 6 255119890 minus 6 200 110119890 minus 2
GS4 294119890 minus 6 327119890 minus 6 487119890 minus 6 250 118119890 minus 2
JAC4 124119890 minus 5 138119890 minus 5 206119890 minus 5 300 127119890 minus 2
Table 5 Experiment 2 119898 = 700 120582 = 10 Δ119909 = 143 times 10minus3 Δ119905 =
204 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 09) 138119890 minus 6 154119890 minus 6 229119890 minus 6 75 650119890 minus 3
IADEMF2(119903 = 05) 187119890 minus 6 207119890 minus 6 309119890 minus 6 75 472119890 minus 3
SOR4(120596 = 12) 312119890 minus 6 347119890 minus 6 517119890 minus 6 125 906119890 minus 3
GS4 667119890 minus 6 742119890 minus 6 111119890 minus 5 175 932119890 minus 3
JAC4 126119890 minus 5 141119890 minus 5 210119890 minus 5 250 105119890 minus 2
Table 6 Experiment 1 119898 = 7000 120582 = 05 Δ119909 = 143 times 10minus4 Δ119905 =
102 times 10minus8 119905 = 510 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME n ET (s)IADEMF4(119903 = 08) 367119890 minus 10 433119890 minus 10 366119890 minus 9 173 146119890 minus 1
IADEMF2(119903 = 05) 375119890 minus 10 435119890 minus 10 376119890 minus 9 200 124119890 minus 1
SOR4(120596 = 112) 455119890 minus 10 507119890 minus 10 367119890 minus 9 257 147119890 minus 1
GS4 110119890 minus 9 111119890 minus 9 376119890 minus 9 300 160119890 minus 1
JAC4 230119890 minus 9 231119890 minus 9 378119890 minus 9 400 163119890 minus 1
Table 7 Experiment 1 119898 = 7000 120582 = 10 Δ119909 = 143 times 10minus4 Δ119905 =
204 times 10minus8 119905 = 51 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 163119890 minus 7 163119890 minus 7 163119890 minus 7 112 960119890 minus 2
IADEMF2(119903 = 12) 163119890 minus 7 163119890 minus 7 163119890 minus 7 120 736119890 minus 2
SOR4(120596 = 115) 163119890 minus 7 163119890 minus 7 164119890 minus 7 171 999119890 minus 2
GS4 164119890 minus 7 164119890 minus 7 164119890 minus 7 216 107119890 minus 1
JAC4 165119890 minus 7 165119890 minus 7 165119890 minus 7 312 111119890 minus 1
It is found that the convergence rate of the fourth-orderand second-order methods improves with the application oflarger mesh size that is from 120582 = 05 to 120582 = 10 Thecoarser meshes cause a reduction in the computational oper-ations thus giving better rate of convergence However a
Table 8 Experiment 2 119898 = 7000 120582 = 05 Δ119909 = 143 times 10minus4 Δ119905 =
102 times 10minus8 119905 = 510 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 07) 567119890 minus 13 321119890 minus 12 130119890 minus 10 150 131119890 minus 1
IADEMF2(119903 = 07) 199119890 minus 11 460119890 minus 11 182119890 minus 9 200 127119890 minus 1
SOR4(120596 = 115) 245119890 minus 10 273119890 minus 10 399119890 minus 10 260 151119890 minus 1
GS4 398119890 minus 10 443119890 minus 10 661119890 minus 10 400 171119890 minus 1
JAC4 105119890 minus 9 117119890 minus 9 174119890 minus 9 550 176119890 minus 1
Table 9 Experiment 2 119898 = 7000 120582 = 10 Δ119909 = 143 times 10minus4 Δ119905 =
204 times 10minus8 119905 = 51 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 154119890 minus 6 171119890 minus 6 255119890 minus 6 96 912119890 minus 2
IADEMF2(119903 = 10) 154119890 minus 6 171119890 minus 6 255119890 minus 6 96 607119890 minus 2
SOR4(120596 = 12) 154119890 minus 6 171119890 minus 6 255119890 minus 6 192 997119890 minus 2
GS4 154119890 minus 6 171119890 minus 6 255119890 minus 6 288 128119890 minus 1
JAC4 154119890 minus 6 172119890 minus 6 256119890 minus 6 408 146119890 minus 1
more accurate solution is obtained by using finer mesh Forexample Table 6 shows that for 120582 = 05 the IADEMF4 hasRMSE = 433119890 minus 10 and 119899 = 173 whereas Table 7 shows thatfor 120582 = 10 its RMSE = 163119890 minus 7 and 119899 = 112 In generalamongst all the testedmethods the IADEMF4 still maintainsits greater accuracy characteristic even with coarser meshes
In terms of execution time the results from every tabledisplay shorter execution time for the IADEMF2 in compari-son to the IADEMF4 This is expected since the IADEMF2has lower computational complexity (Table 1) Despite theachievement in accuracy the IADEMF4 has to performmorecomputational work than the IADEMF2 since the former hasto utilize values of 119906 at two grid points on either side of thepoint (119894Δ119909 119896Δ119905) along the 119896th time level Thus if accuracyis desired then the preferred sequential numerical algorithmwould be the IADEMF4 On the other hand if execution timematters then the choice would be the IADEMF2
Amongst the fourth-order methods the IADEMF4 exe-cutes in the least amount of time Even though its computa-tional complexity is relatively large it operates with the leastnumber of iterations thus enabling it to be the most efficient
10 Conclusion
This paper proposes the development of the novel fourth-order IADEMF4 finite difference scheme The higher-orderscheme is developed by representing the spatial derivative inthe heat equation with the fourth-order finite difference CNapproximation This leads to the formation of pentadiagonalmatrices in the systems of linear equations with larger com-putational stencils The algorithm also employs the higheraccuracy of the Mitchell and Fairweather variant Despite
12 Journal of Applied Mathematics
the fourth-order IADE schemersquos higher computational com-plexity this technique is proved to be valuable becauseit enhances the accuracy of its second-order counterpartnamely the IADEMF2 In addition the higher accuracy ofIADEMF4 is verified as a convergent and unconditionallystable scheme and is superior in terms of rate of convergenceIt has also been proven to execute more efficiently in compar-ison to the other benchmarked fourth-order classical iterativemethods such as the GS4 the SOR4 and the JAC4 The in-creasing number of correct digits at each iteration serves as anadvantage for the IADEMF4 thus yielding faster rate of con-vergence with higher level of accuracy
In conclusion the proposed IADEMF4 scheme affordsusers many advantages with respect to higher accuracy sta-bility and rate of convergence and it serves as an alternativeefficient technique for the solution of a one-dimensional heatequation with Dirichlet boundary conditions
The proposed fourth-order scheme can be modified andadapted to more general multidimensional linear and non-linear parabolic elliptic and hyperbolic partial differentialequations using different types of boundary conditionsBesides it can also be considered for applications in problemsthat require higher-order accuracy with high resolution suchas problems in nanocomputing that require the solution ofvery large sparse systems of equation
The explicit and high accuracy feature of the IADEMF4can be exploited in two- and three-dimensional heat prob-lems by applying the scheme as pentadiagonal solvers forthe system of linear equations arising in the sweeps of ahigher-order alternating direction implicit (ADI) schemeThe lower-order ADI scheme was initially proposed byPeaceman and Rachford [24] The higher space dimensionsare expected not to cause unsatisfactory performance for theADI-IADEMF4 as long as the proposed scheme is stableconvergent and efficient as the process of its implementa-tion continues to advance from one-time step to anotherPathirana et al [25] implemented the ADI in developing atwo-dimensional model for incorporating flood damage inurban drainage planningThe proposed model is found to bestable numerically accurate and computationally efficientMirzavand et al [26] proposed the ADI-FDTD methodfor the physical modeling of high-frequency semiconductordevices The approach is able to reduce significantly the full-wave simulation time
It is possible to parallelize the IADEMF4 algorithm sincethe calculation of the new iteration only depends on theknown values from the last iteration (Figures 2 and 3) Futurework is to exploit the explicit computational properties andefficiency of the IADEMF4 for parallelization and executionon distributed parallel computing systems The idea is tospeed up the execution time without compromising its accu-racy especially on problems involving very large linear sys-tems of equations
References
[1] J Noye Computational Techniques for Differential Equationsvol 83 Elsevier Science North-Holland Publishing Amster-dam The Netherlands 1984
[2] G W Recktenwald ldquoFinite difference approximations to theheat equationrdquoMechanical Engineering vol 10 pp 1ndash27 2004
[3] P Lax and B Wendroff ldquoSystems of conservation lawsrdquo Com-munications on Pure and Applied Mathematics vol 13 pp 217ndash237 1960
[4] R J MacKinnon and G F Carey ldquoAnalysis of material interfacediscontinuities and superconvergent fluxes in finite differencetheoryrdquo Journal of Computational Physics vol 75 no 1 pp 151ndash167 1988
[5] R Hixon and E Turkel High Accuracy Compact MacCormack-Type Schemes For Computational Aerocoustics NASACentre forAerospace Information Lewis Research Centre 1998
[6] Y Lin Higher-order finite difference methods for solving heatequations [PhD thesis] Southern Illinois University 2008
[7] J Sulaiman M K Hasan M Othman M and S A AbdulKarim ldquoFourth-order QSMOR iterative method for the solu-tion of one-dimensional parabolic PDErsquosrdquo in Proceedings of theInternational Conference on Industrial and AppliedMathematics(CIAM rsquo10) pp 34ndash39 ITB Bandung Indonesia 2010
[8] N Jha ldquoThe application of sixth order accurate parallel quar-ter sweep alternating group explicit algorithm for nonlinearboundary value problems with singularityrdquo in Proceedings ofthe 2nd International Conference on Methods and Models inComputer Science (ICM2CS rsquo10) pp 76ndash80 December 2010
[9] Y Jin G Jin and J Li ldquoA class of high-precision finite differenceparallel algorithms for convection equationsrdquo Journal of Conver-gence Information Technology vol 6 no 1 pp 79ndash82 2011
[10] W Fu and E L Tan ldquoDevelopment of split-step FDTDmethodwith higher-order spatial accuracyrdquo Electronics Letters vol 40no 20 pp 1252ndash1253 2004
[11] A Mohebbi andM Dehghan ldquoHigh-order compact solution ofthe one-dimensional heat and advection-diffusion equationsrdquoApplied Mathematical Modelling vol 34 no 10 pp 3071ndash30842010
[12] W Liao ldquoAn implicit fourth-order compact finite differencescheme for one-dimensional Burgersrsquo equationrdquo Applied Math-ematics and Computation vol 206 no 2 pp 755ndash764 2008
[13] Z F Tian and Y B Ge ldquoA fourth-order compact ADI methodfor solving two-dimensional unsteady convection-diffusionproblemsrdquo Journal of Computational and Applied Mathematicsvol 198 no 1 pp 268ndash286 2007
[14] C Chun ldquoSome fourth-order iterativemethods for solving non-linear equationsrdquo Applied Mathematics and Computation vol195 no 2 pp 454ndash459 2008
[15] S Zhu Z Yu and J Zhao ldquoAhigh-order parallel finite differencealgorithmrdquo Applied Mathematics and Computation vol 183 no1 pp 365ndash372 2006
[16] Z Gao and S Xie ldquoFourth-order alternating direction implicitcompact finite difference schemes for two-dimensionalSchrodinger equationsrdquo Applied Numerical Mathematics vol61 no 4 pp 593ndash614 2011
[17] M S Sahimi A Ahmad and A A Bakar ldquoThe Iterative Alter-nating Decomposition Explicit (IADE) method to solve theheat conduction equationrdquo International Journal of ComputerMathematics vol 47 pp 219ndash229 1993
[18] G D Smith Numerical Solution of Partial Differential Equa-tions Finite Difference Methods Oxford University Press 2ndedition 1978
[19] A RMitchell andG Fairweather ldquoImproved forms of the alter-nating direction methods of Douglas Peaceman and Rachfordfor solving parabolic and elliptic equationsrdquo Numerische Math-ematik vol 6 no 1 pp 285ndash292 1964
Journal of Applied Mathematics 13
[20] B N DattaNumerical Linear Algebra and Applications BrooksCole Publishing Pacific Grove Calif USA 1995
[21] D M Young Iterative Solution of Large Linear Systems Aca-demic Press New York NY USA 1971
[22] V K Saulev Integration of Equations of Parabolic Type by theMethod of Nets Pergamon Press Oxford UK 1964
[23] L W Johnson and R D Riess Numerical Analysis Addison-Wesley Reading Mass USA 2nd edition 1982
[24] D W Peaceman and H H Rachford Jr ldquoThe numerical solu-tion of parabolic and elliptic differential equationsrdquo Journal ofthe Society for Industrial and Applied Mathematics vol 3 pp28ndash41 1955
[25] A Pathirana S Tsegaye B Gersonius and K VairavamoorthyldquoA simple 2-D inundation model for incorporating flood dam-age in urban drainage planningrdquo Hydrology and Earth SystemSciences vol 15 no 8 pp 2747ndash2761 2011
[26] R Mirzavand A Abdipour G Moradi and M MovahhedildquoFull-wave semiconductor devices simulation using ADI-FDTD methodrdquo Progress in Electromagnetics Research vol 11pp 191ndash202 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Journal of Applied Mathematics 5
manner 119906(119901+1)119894
in (23) is calculated by proceeding from theright boundary towards the left (Figure 1)
The computational molecules are depicted in Figures 2and 3
At each level of iteration the computational moleculesinvolve two known grid points at the new level and anotherthree known ones at the old level Clearly the method isexplicit
4 Convergence Analysis of the IADEMF4
This section proves the convergence of the IADEMF4 Since120582 gt 1will not guarantee an accurate approximation for 120597119880120597119905[18] the values of 120582 that are considered appropriate for theproof are 0 lt 120582 le 1
From the definitions in ((18) and (19)) the followingresults are obtained
1198901=6 (119888 minus 1)
5=3120582
2 implying that 0 lt 119890
1le3
2
1=6119889
5lt0 119897
1=
6119887
6 minus 1198901
lt0 V119894=6119890
5gt 0
for 119894 = 1 2 119898 minus 4
2=6119889 + 119897
1V1
5= 1+1198971V1
5lt 1 since 119897
1V1lt 0
1198902=6 (119888 minus 1) + 119897
11
5= 1198901+11989711
5gt 1198901 since 119897
11gt 0
(25)
Simple computation shows that 119897115 lt 1 Clearly 119890
2lt 6
Lemma 1 119890119894in ((18) and (19)) is such that 0 lt 119890
119894lt 6 for all
119894 = 1 2 3 119898 minus 2
Proof The results in (25) show that 6 gt 1198902gt 1198901gt 0 Assume
it is true that 6 gt 119890119896gt 119890119896minus1
gt 0 for all 119896 = 3 4 5 119898 minus 3Since the assumption implies that 6 gt 119890
119896minus1gt 119890119896minus2
gt 0 then119896minus1
= 6119886(6 minus 119890119896minus1
) gt 119896minus2
= 6119886(6 minus 119890119896minus2
) gt 0 Therefore119896minus1
V119896minus1
gt 119896minus2
V119896minus2
It has also been shown that
2lt 1lt 0 Assume it is true
that 119895lt 119895minus1
lt 0 for all 119895 = 3 4 119898 minus 4 The assumptionimplies that
119895minus1lt 119895minus2
lt 0 Thus 119895minus1
119895minus1
lt 119895minus2
119895minus2
lt 0
119895+1
=6119889 + 119897
119895V119895
5= 1+119897119895V119895
5= 1+
(6119887 + 119895minus1
119895minus1
) V119895
5 (6 minus 119890119895)
lt 1+
(6119887 + 119895minus2
119895minus2
) V119895minus1
5 (6 minus 119890119895minus1
)
= 1+119897119895minus1
V119895minus1
5= 119895lt 0
(26)
Since it is true that for 119895 + 1 119895+1
lt 119895lt 0 then by
induction the assumption 119895lt 119895minus1
lt 0 is true for all 119895 An
0
u(p+12)
i
u(p+1)
i
i iminus 1 i + 1 m + 1
Figure 1 The two-stage IADEMF4 algorithmThe directions of thesweeps at the (119901 + 12) and (119901 + 1) iteration levels
i iminus 1i minus 2 i + 1 i + 2
(p + 12)
(p)
Figure 2 Computational molecule of the IADEMF4 at the (119901+12)iteration level
i iminus 1i minus 2 i + 1 i + 2
(p + 12)
(p + 1)
Figure 3 Computational molecule of the IADEMF4 at the (119901 + 1)
iteration level
equivalent to the preceding statement would be 119896lt 119896minus1
lt 0
for all 119896 = 3 4 5 119898 minus 3 It follows that
119897119896119896=(6119887 +
119896minus1119896minus1
) 119896
6 minus 119890119896
gt(6119887 +
119896minus2119896minus2
) 119896minus1
6 minus 119890119896minus1
= 119897119896minus1
119896minus1
119890119896+1
=6 (119888 minus 1) + 119897
119896119896+ 119896minus1
V119896minus1
5= 1198901+119897119896119896+ 119896minus1
V119896minus1
5
gt 1198901+119897119896minus1
119896minus1
+ 119896minus2
V119896minus2
5= 119890119896gt 0
(27)
Since it is true that for 119896+1 119890119896+1
gt 119890119896gt 0 then by induction
119890119896gt 119890119896minus1
gt 0 is also true for all 119896Suppose there is a 119896 such that 119890
119896gt 6 Then
119890119896+1
= 1198901+119897119896119896+ 119896minus1
V119896minus1
5
= 1198901+
6119887 + 119896
5 (6 minus 119890119896)+119896minus1
5(119896minus1
119896
6 minus 119890119896
+ V119896minus1
) lt 1198901
(28)
This is a contradiction since 119890119896+1
gt 1198901 This verifies that 6 gt
119890119896gt 119890119896minus1
gt 0 119896 = 3 4 5 119898 minus 3 Let 119894 = 2 3 4 119898 minus 2and then 6 gt 119890
119894gt 119890119894minus1
gt 0 Therefore 0 lt 119890119894lt 6 119894 =
1 2 3 119898 minus 2 Lemma 1 is proved
6 Journal of Applied Mathematics
Lemma 2 If 119903 gt 0 and 119892 = (6+ 119903)6 thenmax |(119903 minus119892119890119894)(119903 +
119890119894)| lt 1 i = 1 2 3 119898 minus 2
Proof Let max |(119903 minus 119892119890119894)(119903 + 119890
119894)| = |(119903 minus 119892119890
119895)(119903 + 119890
119895)| for
some 119895Assume |(119903 minus 119892119890
119895)(119903 + 119890
119895)| ge 1 Since 119903 gt 0 and 119890
119895gt 0
then 119903 + 119890119895gt 0
If (119903 minus 119892119890119895)(119903 + 119890
119895) ge 1 then minus((6 + 119903)6)119890
119895ge 119890119895 which
implies that 119903 le minus12 This contradicts the fact that 119903 gt 0If (119903 minus 119892119890
119895)(119903 + 119890
119895) le minus1 then 2119903 minus ((6 + 119903)6)119890
119895le minus119890119895
which implies that 119890119895ge 12 This contradicts Lemma 1
So the assumption that |(119903 minus 119892119890119895)(119903 + 119890
119895)| ge 1 is false
Hence max |(119903 minus 119892119890119894)(119903 + 119890
119894)| = |(119903 minus 119892119890
119895)(119903 + 119890
119895)| lt 1
Proposition 3 (119903119868 minus 1198921198661)(119903119868 + 119866
1)minus1
2lt 1
Proof Let 119865 = (119903119868 minus 1198921198661)(119903119868 + 119866
1)minus1 that is
119865 =
[[[[[[[[[[
[
119903 minus 119892
119889
d119903 minus 119892
119889119874
d d d
d d
sdot sdot sdot sdot sdot sdot sdot sdot sdot sdot sdot sdot119903 minus 119892
119889
]]]]]]]]]]
]
(29)
119865 is a lower triangular matrix with all the diagonal entries(whence the eigenvalues of 119865) equal to (119903 minus 119892)119889 where 119889 =
119903 + 1 Denote all the eigenvalues of 119865 by 120582119865 If 120588[119865] is defined
as the spectral radius of 119865 then
120588 [119865] = max 10038161003816100381610038161205821198651003816100381610038161003816 =
10038161003816100381610038161205821198651003816100381610038161003816 =
1003816100381610038161003816100381610038161003816
119903 minus 119892
119903 + 1
1003816100381610038161003816100381610038161003816le 119865
2 (30)
But by definition of 2-norm
1198652= maxx2 = 0
119865x2
x2
= maxy2=1
1003817100381710038171003817119865y10038171003817100381710038172
1003817100381710038171003817y10038171003817100381710038172
= maxy2=1
1003817100381710038171003817119865y10038171003817100381710038172
le maxw2=1
119865w2= maxw2=1
1003817100381710038171003817120582119865w10038171003817100381710038172
(31)
Since all the eigenvalues of 119865 are equal then
1198652le1003816100381610038161003816120582119865
1003816100381610038161003816 maxw2=1
w2=1003816100381610038161003816120582119865
1003816100381610038161003816 = 120588 [119865] (32)
Thus from (30) and (32) 1198652= |120582119865| = |(119903 minus 119892)(119903 + 1)|
By Lemma 2 with 119890119895replaced by 1 |(119903 minus 119892)(119903 + 1)| lt 1
is obtained leading to the result of Proposition 3 which is1198652= (119903119868 minus 119892119866
1)(119903119868 + 119866
1)minus1
2lt 1
Proposition 4 (119903119868 minus 1198921198662)(119903119868 + 119866
2)minus1
2lt 1
Proof Let 119870 = (119903119868 minus 1198921198662)(119903119868 + 119866
2)minus1 that is
119870 =
[[[[[[[[[[[[[
[
119903 minus 1198921198901
1199111
d sdot sdot sdot sdot sdot sdot sdot sdot sdot
119903 minus 1198921198902
1199112
d sdot sdot sdot
d
119874 d d
119903 minus 119892119890119898
119911119898
]]]]]]]]]]]]]
]
(33)
119870 is an upper triangular matrix with all the diagonal entriesequal to (119903 minus 119892119890
119894)119911119894 119894 = 1 2 119898 minus 2 where 119911
119894= 119903 + 119890
119894
Since all the eigenvalues of119870 are distinct then119870 is similar toa diagonal matrix119863
By the Schur triangularization theorem [20] there is anorthogonal matrix 119874 such that 119874119879119870119874 = 119863 The diagonalentries of119863 are the eigenvalues of119870
1198702=10038171003817100381710038171003817119874119879
119870119874100381710038171003817100381710038172
= 1198632
= radicmaximum eigenvalue of 119863119879119863
= radicmaximum eigenvalue of 1198632
= radic120588 (1198632) = radic120588 (1198702) = 120588 (119870)
(34)
By Lemma 2
1198702= 120588 (119870) = max
10038161003816100381610038161003816100381610038161003816
119903 minus 119892119890119894
119903 + 119890119894
10038161003816100381610038161003816100381610038161003816
=
1003816100381610038161003816100381610038161003816100381610038161003816
119903 minus 119892119890119895
119903 + 119890119895
1003816100381610038161003816100381610038161003816100381610038161003816
lt 1 (35)
for some 119895 Hence this proves Proposition 4
From (20) and (21)
u(119901+1) = (119903119868 + 1198662)minus1
(119903119868 minus 1198921198661)
times [(119903119868 + 1198661)minus1
(119903119868 minus 1198921198662) u(119901) + (119903119868 + 119866
1)minus1f]
+ 119892 (119903119868 + 1198662)minus1f
(36)
Let
119872(119903) = (119903119868 + 1198662)minus1
(119903119868 minus 1198921198661) (119903119868 + 119866
1)minus1
(119903119868 minus 1198921198662)
(37)
And let
119902 (119903) = [(119903119868 + 1198662)minus1
(119903119868 minus 1198921198661) (119903119868 + 119866
1)minus1
+119892(119903119868 + 1198662)minus1
] f (38)
Then u(119901+1) = 119872(119903)u(119901) + 119902(119903)
Theorem 5 The IADEMF4 is convergent if 120588[119872(119903)] lt 1 for119903 gt 0
Journal of Applied Mathematics 7
Proof Define (119903) = (119903119868 + 1198662)119872(119903)(119903119868 + 119866
2)minus1 then
(119903) = (119903119868 minus 1198921198661) (119903119868 + 119866
1)minus1
(119903119868 minus 1198921198662) (119903119868 + 119866
2)minus1
(39)
Thus by similarity119872(119903) and (119903) have the same set of eigen-values Therefore
120588 [119872 (119903)]
= 120588 [ (119903)] le10038171003817100381710038171003817 (119903)
10038171003817100381710038171003817 2
=10038171003817100381710038171003817(119903119868 minus 119892119866
1) (119903119868 + 119866
1)minus1
(119903119868 minus 1198921198662) (119903119868 + 119866
2)minus110038171003817100381710038171003817 2
le10038171003817100381710038171003817(119903119868 minus 119892119866
1) (119903119868 + 119866
1)minus1100381710038171003817100381710038172
10038171003817100381710038171003817(119903119868 minus 119892119866
2) (119903119868 + 119866
2)minus110038171003817100381710038171003817 2
lt 1
(40)
The last inequality is due to Propositions 3 and 4The provenTheorem 5 assures the convergence of the IADEMF4
5 The Fourth-Order ClassicalIterative Methods
The system of linear equations in (8) may also be solvedby using the classical iterative methods of the fourth orderThey include the JAC4 the GS4 and the SOR4 methodsThese iterative methods are capable of exploiting the sparsestructure of the pentadiagonal matrix
The JAC4 algorithm can be represented by
119906(119901+1)
119894=
(119891119894minus119886119906(119901)
119894minus2minus 119887119906(119901)
119894minus1minus 119889119906(119901)
119894+1minus 119890119906(119901)
119894+2)
119888
119894 = 2 3 119898 minus 1
(41)
The approximation of 119906(119901+1)119894
at the (119901 + 1)th iteration level iscomputed using the relevant values in the 119901th iteration
The GS4 method uses the most recent values of 119906(119901+1)119894minus2
and 119906(119901+1)
119894minus1to update the approximation value of 119906(119901+1)
119894 The
algorithm for the GS4 is as expressed in (42)
119906(119901+1)
119894=
(119891119894minus119886119906(119901+1)
119894minus2minus 119887119906(119901+1)
119894minus1minus 119889119906(119901)
119894+1minus 119890119906(119901)
119894+2)
119888
119894 = 2 3 119898 minus 1
(42)
The SOR4 iterative method accelerates the convergencerate of the GS4 If 120596 is a relaxation parameter then for any120596 = 0 (42) can be rewritten as
119906(119901+1)
119894= (1 minus 120596) 119906
(119901)
119894
+
120596 (119891119894minus 119886119906(119901+1)
119894minus2minus 119887119906(119901+1)
119894minus1minus 119889119906(119901)
119894+1minus 119890119906(119901)
119894+2)
119888
119894 = 2 3 119898 minus 1
(43)
i iminus 1i minus 2 i + 1 i + 2
(p + 1)
(p)
Figure 4 Computational molecule of the JAC4
i iminus 1i minus 2 i + 1 i + 2
(p + 1)
(p)
Figure 5 Computational molecule of the GS4SOR4
Table 1 Sequential arithmetic operations per iteration (119898 problemsize)
Method Number ofadditions
Number ofmultiplications
Total operationcount
IADEMF4 10(119898 minus 2) 13(119898 minus 2) 23(119898 minus 2)
IADEMF2 6119898 9119898 15119898
JAC4GS4 4(119898 minus 2) 5(119898 minus 2) 9(119898 minus 2)
SOR4 5(119898 minus 2) 7(119898 minus 2) 12(119898 minus 2)
The SOR4 algorithm reduces to the GS4 if 120596 = 1 Exceptfor very special cases it is difficult to obtain the analyticexpression for 120596 According to Young [21] the precise deter-mination of the optimal 120596 in the SOR is only known for asmall class of matrices The value of 120596 generally lies in therange of 1 lt 120596 lt 2
The computational molecules for the JAC4 and the GS4are as illustrated in Figures 4 and 5 respectively The SOR4has the same form as the GS4
6 Computational Complexity
The cost of implementing an algorithm can be assessedby examining its computational complexity The number ofarithmetic operations such as additions (including subtrac-tions) andmultiplications (including divisions) that is neededto perform by the algorithm can be straightforwardly count-ed Table 1 gives the number of sequential arithmetic opera-tions per iteration that is required to evaluate the algorithms
For the higher-order schemes trade-offbetween accuracyand speed usually happens The computational complexityhas an effect on the efficiency of a particular scheme Thusthis factor will be taken into account when discussing theresults of the numerical experiments conducted in this study
7 The Pseudocode of the IADEMF4Sequential Algorithm
Algorithm 1 illustrates the pseudocode of the IADEMF4sequential algorithm This algorithm can also be generalizedand applied to other methods discussed in this paper
8 Journal of Applied Mathematics
begindetermine the parameters119898 119903 Δ119905 Δ119909 120582 120576 119879
determine initial conditions 119906(119901)119894
determine exact solutions 119880119894at time level 119879
while (time level lt 119879)determine boundary conditionsfor 119894 = 2 to 119894 = 119898 minus 1
compute 119891119894(refer to (9))
end-forset iteration = 0while (convergence conditions are not satisfied)
for 119894 = 2 to 119894 = 119898 minus 1
compute 119906(119901+12)119894
(refer to (22))end-forfor 119894 = 2 to 119894 = 119898 minus 1
compute 119906(119901+1)119894
(refer to (23))end-fortest for convergencecompute 119890
119894larr
10038161003816100381610038161003816119906(119901+1)
119894minus 119906(119901)
119894
10038161003816100381610038161003816if (max1003816100381610038161003816119890119894
1003816100381610038161003816 lt 120576)then 119906
(119901)
119894larr 119906(119901+1)
119894
add 1 to iteration (if necessary)end-while
end-whilefor 119894 = 2 to 119894 = 119898 minus 1
determine average absolute error root meansquare error and maximum error
end-forend
Algorithm 1 The IADEMF4 sequential algorithm
8 Numerical Experiments
Two experiments were conducted to test the sequentialnumerical performance of the proposed IADEMF4 methodagainst those of the benchmarked classical iterative methodsnamely the JAC4 theGS4 and the SOR4 Comparison is alsomade with the corresponding IADE method of the secondorder
Experiment 1 This problem was taken from Saulev [22]
120597119880
120597119905=1205972
119880
1205971199092 0 le 119909 le 1 (44)
subject to the initial condition
119880 (119909 0) = 4119909 (1 minus 119909) 0 le 119909 le 1 (45)
and the boundary conditions
119880 (0 119905) = 119880 (1 119905) = 0 119905 ge 0 (46)
The exact solution to the given problem is given by
119880 (119909 119905) =32
1205873
infin
sum
119896=1(2)
1
1198963119890minus12058721198962119905 sin (119896120587119909) (47)
Experiment 2 This problem was taken from Johnson andReiss [23]
120597119880
120597119905=1205972
119880
1205971199092 0 le 119909 le 1 (48)
subject to the initial condition
119880 (119909 0) = sin (120587119909) (1 + 6 cos (120587119909)) 0 le 119909 le 1 (49)
and the boundary conditions
119880 (0 119905) = 119880 (1 119905) = 0 119905 ge 0 (50)
The exact solution to the given problem is given by
119880 (119909 119905) = sin (120587119909) 119890minus1205872119905
+ 3 sin (2120587119909) 119890minus41205872119905
(51)
Journal of Applied Mathematics 9
9 Results and Discussion
In each experiment the exact solutions at 119894 = 1 and 119894 = 119898
were taken as boundary values for the IADEMF4 and theother fourth-order classical iterative methods The conver-gence criterion used in the testing of each method was takenas 119906(119901+1) minus 119906
(119901)
infin
le 120576 where 120576 is the convergence toleranceThe selections of the optimum 119903 or 120596 were determined byexperiments As for the IADEMF2 the CN scheme with 120579 =
12 was employedFigures 6 7 8 and 9 visualize the behavior of the one-
dimensional parabolic heat solutions for both experimentsBy using119898 = 10 and a tolerance requirement of 120576 = 10
minus4 twodifferent mesh sizes 120582 = 05 and 120582 = 10 were consideredin each experiment The exact solutions are compared withthe IADEMF4 and the IADEMF2 numerical solutions Theoptimum value of 119903 chosen for each method as well as theircorresponding outcome of the number of iterations 119899 isstated in the legend of each figure Every figure reveals thatthe IADEMF4 is more accurate than the IADEMF2 andthe former converges with fewer numbers of iterations incomparison to the latter The numerical solutions of theIADEMF4 seem to be in very good agreement with theexact solutions For example in Figure 7 at 119909 = 05 thedifference between the exact solution and the numericalsolution using the IADEMF2 and using the IADEMF4 isabout 36 and 0 respectively These results imply thatthe accuracy and convergence rate of the second-order IADEmethod are enhanced by the implementation of the fourth-order CN approximation that leads to the formation of thecorresponding fourth-order IADE scheme
Figure 10 displays the graph of log (RMSE) versus log(Δ119909)for decreasing values of Δ119909 implemented on Experiment 2By considering 120576 = 10
minus4 and fixing Δ119905 = 00001 the value ofΔ119909 was initially taken as Δ119909 = 0125 It was then successivelyhalved into Δ1199092 Δ1199094 and Δ1199098
The figure shows that amongst the tested methods theroot mean square error (RMSE) of the IADEMF4 and theIADEMF2 decreases linearly as the values of Δ119909 decreasesThe slope of the IADEMF4 is approximately equal to 4which corresponds to its fourth-order spatial accuracy TheIADEMF2 with a second-order spatial accuracy has a slopethat is approximately equal to 2 It is clear that the IADEMF4is always more accurate than the IADEMF2 for the differentconsidered mesh sizes The graphs of the SOR4 (120596 = 105)the GS4 and the JAC4 show that their accuracies of fourthorder tend to lack as Δ119909 decreases largely due to the effect ofincreasing round-off errors as the value of119898 increases
Tables 2 3 4 5 6 7 8 and 9 provide numerical results interms of the average absolute error (AAE) root mean squareerror (RMSE) maximum error (ME) number of iterations(119899) and execution time (ET) measured in seconds (s) Theresults are obtained from both experiments for two differentvalues of 119898 and mesh size 120582 It is generally observed thatwhen 119898 = 700 120582 = 05 and 120576 = 10
minus6 the IADEMF4has the least average absolute error root mean square errorand maximum error in comparison with the other methodsunder consideration (Tables 2ndash5) When the size of 119898 was
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 08 n = 172)IADEMF2 (r = 03 n = 185)
00E + 00
10E minus 02
20E minus 02
30E minus 02
40E minus 02
50E minus 02
60E minus 02
70E minus 02
80E minus 02
90E minus 02
10E minus 01
U(xt)
Figure 6 Numerical and exact solutions for Experiment 1 120582 = 05Δ119909 = 01 Δ119905 = 0005 119905 = 025 and 120576 = 10
minus4
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 14 n = 159)IADEMF2 (r = 05 n = 162)
00E + 00
10E minus 03
20E minus 03
30E minus 03
40E minus 03
50E minus 03
60E minus 03
70E minus 03
80E minus 03
90E minus 03
U(xt)
Figure 7 Numerical and exact solutions for Experiment 1 120582 = 10Δ119909 = 01 Δ119905 = 001 119905 = 05 and 120576 = 10
minus4
ten times bigger (119898 = 7000) and a more stringent tolerancecriterion was set (120576 = 10
minus10
) the accuracy of the IADEMF4clearly outperforms the other methods for a mesh size of120582 = 05 (Tables 6 and 8) However the difference in the errorsamongst the tested methods is not so obvious for the case of120582 = 10 (Tables 7 and 9)The achievement of the fourth-orderIADEmethod in Experiments 1 and 2 can be clearly seen fromthe results in Tables 2 and 8 respectively where it has caused ahuge 94 reduction of RMSE from its corresponding second-order IADE method
The IADEMF4 has the advantage of featuring higheraccuracy due to the execution of the fourth-order CNapprox-imation coupled with the fourth-order accurate MF variant
10 Journal of Applied Mathematics
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 10 n = 169)IADEMF2 (r = 03 n = 213)
00E + 00
10E minus 02
20E minus 02
30E minus 02
40E minus 02
50E minus 02
60E minus 02
70E minus 02
80E minus 02
90E minus 02
10E minus 01
U(xt)
Figure 8 Numerical and exact solutions for Experiment 2 120582 = 05Δ119909 = 01 Δ119905 = 0005 119905 = 025 and 120576 = 10
minus4
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 10 n = 175)IADEMF2 (r = 05 n = 182)
00E + 00
10E minus 03
20E minus 03
30E minus 03
40E minus 03
50E minus 03
60E minus 03
70E minus 03
80E minus 03
U(xt)
Figure 9 Numerical and exact solutions for Experiment 2 120582 = 10Δ119909 = 01 Δ119905 = 001 119905 = 05 and 120576 = 10
minus4
The IADEMF2 is only derived from the second-order CNapproximation but its combination with its correspondingfourth-order MF variant managed to produce errors whichare better than theGS4 JAC4 and the SOR4 Even though theclassical iterative methods are also derived from the fourth-order accurate CN type approximation they are lackingin accuracy due to the round-off errors that have beenaccumulated from the time the execution starts till it ends
With regards to the rate of convergence the resultsfrom each table demonstrate that the number of iterationsproduced by the IADEMF2 and the fourth-order classicaliterative methods is greater than or at least equal to that of
0
minus2
minus208 minus277 minus347 minus416
minus4
minus8
minus6
minus10
minus12
minus14
minus16
minus18
GS4JAC4SOR4
log
(RM
SE)
log(Δx)
IADEMF4
IADEMF2
Figure 10 Log(RMSE) versus Log(Δ119909) Experiment 2 Δ119905 = 00001119905 = 0002 and 120576 = 10
minus4
Table 2 Experiment 1 119898 = 700 120582 = 05 Δ119909 = 143 times 10minus3 Δ119905 =
102 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 07) 267119890 minus 9 330119890 minus 9 212119890 minus 8 100 440119890 minus 2
IADEMF2(119903 = 08) 188119890 minus 8 495119890 minus 8 323119890 minus 7 100 635119890 minus 3
SOR4(120596 = 12) 540119890 minus 7 544119890 minus 7 552119890 minus 7 108 460119890 minus 2
GS4 540119890 minus 6 543119890 minus 6 332119890 minus 6 150 472119890 minus 2
JAC4 228119890 minus 5 229119890 minus 5 232119890 minus 5 150 473119890 minus 2
Table 3 Experiment 1 119898 = 700 120582 = 10 Δ119909 = 143 times 10minus3 Δ119905 =
204 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 297119890 minus 7 301119890 minus 7 301119890 minus 7 50 201119890 minus 2
IADEMF2(119903 = 09) 557119890 minus 7 559119890 minus 7 848119890 minus 7 50 317119890 minus 3
SOR4(120596 = 12) 329119890 minus 6 331119890 minus 6 335119890 minus 6 76 225119890 minus 2
GS4 895119890 minus 6 900119890 minus 6 900119890 minus 6 100 238119890 minus 2
JAC4 214119890 minus 5 215119890 minus 5 217119890 minus 5 125 250119890 minus 2
the IADEMF4 The convergence rate of the latter surpassesthe others in the case of 119898 = 7000 and 120576 = 10
minus10 (Tables6ndash8) Even though the operational count of the IADEMF4is relatively quite large (Table 1) due to its higher level ofaccuracy its increasing number of correct digits at eachiteration causes it to converge at a faster rate The IADEMF2has the advantage of having less mathematical operationsthus it is competitive in terms of convergence rate but atthe expense of accuracy The application of the fourth-orderCN approximation on the heat equation proves that thethree classical iterative methods also converge with JAC4appearing to be the slowest and the least accurate amongst all
Journal of Applied Mathematics 11
Table 4 Experiment 2 119898 = 700 120582 = 05 Δ119909 = 143 times 10minus3 Δ119905 =
102 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME n ET (s)IADEMF4(119903 = 07) 207119890 minus 8 231119890 minus 8 345119890 minus 8 100 903119890 minus 3
IADEMF2(119903 = 08) 106119890 minus 7 175119890 minus 7 175119890 minus 7 100 647119890 minus 3
SOR4(120596 = 11) 154119890 minus 6 171119890 minus 6 255119890 minus 6 200 110119890 minus 2
GS4 294119890 minus 6 327119890 minus 6 487119890 minus 6 250 118119890 minus 2
JAC4 124119890 minus 5 138119890 minus 5 206119890 minus 5 300 127119890 minus 2
Table 5 Experiment 2 119898 = 700 120582 = 10 Δ119909 = 143 times 10minus3 Δ119905 =
204 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 09) 138119890 minus 6 154119890 minus 6 229119890 minus 6 75 650119890 minus 3
IADEMF2(119903 = 05) 187119890 minus 6 207119890 minus 6 309119890 minus 6 75 472119890 minus 3
SOR4(120596 = 12) 312119890 minus 6 347119890 minus 6 517119890 minus 6 125 906119890 minus 3
GS4 667119890 minus 6 742119890 minus 6 111119890 minus 5 175 932119890 minus 3
JAC4 126119890 minus 5 141119890 minus 5 210119890 minus 5 250 105119890 minus 2
Table 6 Experiment 1 119898 = 7000 120582 = 05 Δ119909 = 143 times 10minus4 Δ119905 =
102 times 10minus8 119905 = 510 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME n ET (s)IADEMF4(119903 = 08) 367119890 minus 10 433119890 minus 10 366119890 minus 9 173 146119890 minus 1
IADEMF2(119903 = 05) 375119890 minus 10 435119890 minus 10 376119890 minus 9 200 124119890 minus 1
SOR4(120596 = 112) 455119890 minus 10 507119890 minus 10 367119890 minus 9 257 147119890 minus 1
GS4 110119890 minus 9 111119890 minus 9 376119890 minus 9 300 160119890 minus 1
JAC4 230119890 minus 9 231119890 minus 9 378119890 minus 9 400 163119890 minus 1
Table 7 Experiment 1 119898 = 7000 120582 = 10 Δ119909 = 143 times 10minus4 Δ119905 =
204 times 10minus8 119905 = 51 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 163119890 minus 7 163119890 minus 7 163119890 minus 7 112 960119890 minus 2
IADEMF2(119903 = 12) 163119890 minus 7 163119890 minus 7 163119890 minus 7 120 736119890 minus 2
SOR4(120596 = 115) 163119890 minus 7 163119890 minus 7 164119890 minus 7 171 999119890 minus 2
GS4 164119890 minus 7 164119890 minus 7 164119890 minus 7 216 107119890 minus 1
JAC4 165119890 minus 7 165119890 minus 7 165119890 minus 7 312 111119890 minus 1
It is found that the convergence rate of the fourth-orderand second-order methods improves with the application oflarger mesh size that is from 120582 = 05 to 120582 = 10 Thecoarser meshes cause a reduction in the computational oper-ations thus giving better rate of convergence However a
Table 8 Experiment 2 119898 = 7000 120582 = 05 Δ119909 = 143 times 10minus4 Δ119905 =
102 times 10minus8 119905 = 510 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 07) 567119890 minus 13 321119890 minus 12 130119890 minus 10 150 131119890 minus 1
IADEMF2(119903 = 07) 199119890 minus 11 460119890 minus 11 182119890 minus 9 200 127119890 minus 1
SOR4(120596 = 115) 245119890 minus 10 273119890 minus 10 399119890 minus 10 260 151119890 minus 1
GS4 398119890 minus 10 443119890 minus 10 661119890 minus 10 400 171119890 minus 1
JAC4 105119890 minus 9 117119890 minus 9 174119890 minus 9 550 176119890 minus 1
Table 9 Experiment 2 119898 = 7000 120582 = 10 Δ119909 = 143 times 10minus4 Δ119905 =
204 times 10minus8 119905 = 51 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 154119890 minus 6 171119890 minus 6 255119890 minus 6 96 912119890 minus 2
IADEMF2(119903 = 10) 154119890 minus 6 171119890 minus 6 255119890 minus 6 96 607119890 minus 2
SOR4(120596 = 12) 154119890 minus 6 171119890 minus 6 255119890 minus 6 192 997119890 minus 2
GS4 154119890 minus 6 171119890 minus 6 255119890 minus 6 288 128119890 minus 1
JAC4 154119890 minus 6 172119890 minus 6 256119890 minus 6 408 146119890 minus 1
more accurate solution is obtained by using finer mesh Forexample Table 6 shows that for 120582 = 05 the IADEMF4 hasRMSE = 433119890 minus 10 and 119899 = 173 whereas Table 7 shows thatfor 120582 = 10 its RMSE = 163119890 minus 7 and 119899 = 112 In generalamongst all the testedmethods the IADEMF4 still maintainsits greater accuracy characteristic even with coarser meshes
In terms of execution time the results from every tabledisplay shorter execution time for the IADEMF2 in compari-son to the IADEMF4 This is expected since the IADEMF2has lower computational complexity (Table 1) Despite theachievement in accuracy the IADEMF4 has to performmorecomputational work than the IADEMF2 since the former hasto utilize values of 119906 at two grid points on either side of thepoint (119894Δ119909 119896Δ119905) along the 119896th time level Thus if accuracyis desired then the preferred sequential numerical algorithmwould be the IADEMF4 On the other hand if execution timematters then the choice would be the IADEMF2
Amongst the fourth-order methods the IADEMF4 exe-cutes in the least amount of time Even though its computa-tional complexity is relatively large it operates with the leastnumber of iterations thus enabling it to be the most efficient
10 Conclusion
This paper proposes the development of the novel fourth-order IADEMF4 finite difference scheme The higher-orderscheme is developed by representing the spatial derivative inthe heat equation with the fourth-order finite difference CNapproximation This leads to the formation of pentadiagonalmatrices in the systems of linear equations with larger com-putational stencils The algorithm also employs the higheraccuracy of the Mitchell and Fairweather variant Despite
12 Journal of Applied Mathematics
the fourth-order IADE schemersquos higher computational com-plexity this technique is proved to be valuable becauseit enhances the accuracy of its second-order counterpartnamely the IADEMF2 In addition the higher accuracy ofIADEMF4 is verified as a convergent and unconditionallystable scheme and is superior in terms of rate of convergenceIt has also been proven to execute more efficiently in compar-ison to the other benchmarked fourth-order classical iterativemethods such as the GS4 the SOR4 and the JAC4 The in-creasing number of correct digits at each iteration serves as anadvantage for the IADEMF4 thus yielding faster rate of con-vergence with higher level of accuracy
In conclusion the proposed IADEMF4 scheme affordsusers many advantages with respect to higher accuracy sta-bility and rate of convergence and it serves as an alternativeefficient technique for the solution of a one-dimensional heatequation with Dirichlet boundary conditions
The proposed fourth-order scheme can be modified andadapted to more general multidimensional linear and non-linear parabolic elliptic and hyperbolic partial differentialequations using different types of boundary conditionsBesides it can also be considered for applications in problemsthat require higher-order accuracy with high resolution suchas problems in nanocomputing that require the solution ofvery large sparse systems of equation
The explicit and high accuracy feature of the IADEMF4can be exploited in two- and three-dimensional heat prob-lems by applying the scheme as pentadiagonal solvers forthe system of linear equations arising in the sweeps of ahigher-order alternating direction implicit (ADI) schemeThe lower-order ADI scheme was initially proposed byPeaceman and Rachford [24] The higher space dimensionsare expected not to cause unsatisfactory performance for theADI-IADEMF4 as long as the proposed scheme is stableconvergent and efficient as the process of its implementa-tion continues to advance from one-time step to anotherPathirana et al [25] implemented the ADI in developing atwo-dimensional model for incorporating flood damage inurban drainage planningThe proposed model is found to bestable numerically accurate and computationally efficientMirzavand et al [26] proposed the ADI-FDTD methodfor the physical modeling of high-frequency semiconductordevices The approach is able to reduce significantly the full-wave simulation time
It is possible to parallelize the IADEMF4 algorithm sincethe calculation of the new iteration only depends on theknown values from the last iteration (Figures 2 and 3) Futurework is to exploit the explicit computational properties andefficiency of the IADEMF4 for parallelization and executionon distributed parallel computing systems The idea is tospeed up the execution time without compromising its accu-racy especially on problems involving very large linear sys-tems of equations
References
[1] J Noye Computational Techniques for Differential Equationsvol 83 Elsevier Science North-Holland Publishing Amster-dam The Netherlands 1984
[2] G W Recktenwald ldquoFinite difference approximations to theheat equationrdquoMechanical Engineering vol 10 pp 1ndash27 2004
[3] P Lax and B Wendroff ldquoSystems of conservation lawsrdquo Com-munications on Pure and Applied Mathematics vol 13 pp 217ndash237 1960
[4] R J MacKinnon and G F Carey ldquoAnalysis of material interfacediscontinuities and superconvergent fluxes in finite differencetheoryrdquo Journal of Computational Physics vol 75 no 1 pp 151ndash167 1988
[5] R Hixon and E Turkel High Accuracy Compact MacCormack-Type Schemes For Computational Aerocoustics NASACentre forAerospace Information Lewis Research Centre 1998
[6] Y Lin Higher-order finite difference methods for solving heatequations [PhD thesis] Southern Illinois University 2008
[7] J Sulaiman M K Hasan M Othman M and S A AbdulKarim ldquoFourth-order QSMOR iterative method for the solu-tion of one-dimensional parabolic PDErsquosrdquo in Proceedings of theInternational Conference on Industrial and AppliedMathematics(CIAM rsquo10) pp 34ndash39 ITB Bandung Indonesia 2010
[8] N Jha ldquoThe application of sixth order accurate parallel quar-ter sweep alternating group explicit algorithm for nonlinearboundary value problems with singularityrdquo in Proceedings ofthe 2nd International Conference on Methods and Models inComputer Science (ICM2CS rsquo10) pp 76ndash80 December 2010
[9] Y Jin G Jin and J Li ldquoA class of high-precision finite differenceparallel algorithms for convection equationsrdquo Journal of Conver-gence Information Technology vol 6 no 1 pp 79ndash82 2011
[10] W Fu and E L Tan ldquoDevelopment of split-step FDTDmethodwith higher-order spatial accuracyrdquo Electronics Letters vol 40no 20 pp 1252ndash1253 2004
[11] A Mohebbi andM Dehghan ldquoHigh-order compact solution ofthe one-dimensional heat and advection-diffusion equationsrdquoApplied Mathematical Modelling vol 34 no 10 pp 3071ndash30842010
[12] W Liao ldquoAn implicit fourth-order compact finite differencescheme for one-dimensional Burgersrsquo equationrdquo Applied Math-ematics and Computation vol 206 no 2 pp 755ndash764 2008
[13] Z F Tian and Y B Ge ldquoA fourth-order compact ADI methodfor solving two-dimensional unsteady convection-diffusionproblemsrdquo Journal of Computational and Applied Mathematicsvol 198 no 1 pp 268ndash286 2007
[14] C Chun ldquoSome fourth-order iterativemethods for solving non-linear equationsrdquo Applied Mathematics and Computation vol195 no 2 pp 454ndash459 2008
[15] S Zhu Z Yu and J Zhao ldquoAhigh-order parallel finite differencealgorithmrdquo Applied Mathematics and Computation vol 183 no1 pp 365ndash372 2006
[16] Z Gao and S Xie ldquoFourth-order alternating direction implicitcompact finite difference schemes for two-dimensionalSchrodinger equationsrdquo Applied Numerical Mathematics vol61 no 4 pp 593ndash614 2011
[17] M S Sahimi A Ahmad and A A Bakar ldquoThe Iterative Alter-nating Decomposition Explicit (IADE) method to solve theheat conduction equationrdquo International Journal of ComputerMathematics vol 47 pp 219ndash229 1993
[18] G D Smith Numerical Solution of Partial Differential Equa-tions Finite Difference Methods Oxford University Press 2ndedition 1978
[19] A RMitchell andG Fairweather ldquoImproved forms of the alter-nating direction methods of Douglas Peaceman and Rachfordfor solving parabolic and elliptic equationsrdquo Numerische Math-ematik vol 6 no 1 pp 285ndash292 1964
Journal of Applied Mathematics 13
[20] B N DattaNumerical Linear Algebra and Applications BrooksCole Publishing Pacific Grove Calif USA 1995
[21] D M Young Iterative Solution of Large Linear Systems Aca-demic Press New York NY USA 1971
[22] V K Saulev Integration of Equations of Parabolic Type by theMethod of Nets Pergamon Press Oxford UK 1964
[23] L W Johnson and R D Riess Numerical Analysis Addison-Wesley Reading Mass USA 2nd edition 1982
[24] D W Peaceman and H H Rachford Jr ldquoThe numerical solu-tion of parabolic and elliptic differential equationsrdquo Journal ofthe Society for Industrial and Applied Mathematics vol 3 pp28ndash41 1955
[25] A Pathirana S Tsegaye B Gersonius and K VairavamoorthyldquoA simple 2-D inundation model for incorporating flood dam-age in urban drainage planningrdquo Hydrology and Earth SystemSciences vol 15 no 8 pp 2747ndash2761 2011
[26] R Mirzavand A Abdipour G Moradi and M MovahhedildquoFull-wave semiconductor devices simulation using ADI-FDTD methodrdquo Progress in Electromagnetics Research vol 11pp 191ndash202 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
6 Journal of Applied Mathematics
Lemma 2 If 119903 gt 0 and 119892 = (6+ 119903)6 thenmax |(119903 minus119892119890119894)(119903 +
119890119894)| lt 1 i = 1 2 3 119898 minus 2
Proof Let max |(119903 minus 119892119890119894)(119903 + 119890
119894)| = |(119903 minus 119892119890
119895)(119903 + 119890
119895)| for
some 119895Assume |(119903 minus 119892119890
119895)(119903 + 119890
119895)| ge 1 Since 119903 gt 0 and 119890
119895gt 0
then 119903 + 119890119895gt 0
If (119903 minus 119892119890119895)(119903 + 119890
119895) ge 1 then minus((6 + 119903)6)119890
119895ge 119890119895 which
implies that 119903 le minus12 This contradicts the fact that 119903 gt 0If (119903 minus 119892119890
119895)(119903 + 119890
119895) le minus1 then 2119903 minus ((6 + 119903)6)119890
119895le minus119890119895
which implies that 119890119895ge 12 This contradicts Lemma 1
So the assumption that |(119903 minus 119892119890119895)(119903 + 119890
119895)| ge 1 is false
Hence max |(119903 minus 119892119890119894)(119903 + 119890
119894)| = |(119903 minus 119892119890
119895)(119903 + 119890
119895)| lt 1
Proposition 3 (119903119868 minus 1198921198661)(119903119868 + 119866
1)minus1
2lt 1
Proof Let 119865 = (119903119868 minus 1198921198661)(119903119868 + 119866
1)minus1 that is
119865 =
[[[[[[[[[[
[
119903 minus 119892
119889
d119903 minus 119892
119889119874
d d d
d d
sdot sdot sdot sdot sdot sdot sdot sdot sdot sdot sdot sdot119903 minus 119892
119889
]]]]]]]]]]
]
(29)
119865 is a lower triangular matrix with all the diagonal entries(whence the eigenvalues of 119865) equal to (119903 minus 119892)119889 where 119889 =
119903 + 1 Denote all the eigenvalues of 119865 by 120582119865 If 120588[119865] is defined
as the spectral radius of 119865 then
120588 [119865] = max 10038161003816100381610038161205821198651003816100381610038161003816 =
10038161003816100381610038161205821198651003816100381610038161003816 =
1003816100381610038161003816100381610038161003816
119903 minus 119892
119903 + 1
1003816100381610038161003816100381610038161003816le 119865
2 (30)
But by definition of 2-norm
1198652= maxx2 = 0
119865x2
x2
= maxy2=1
1003817100381710038171003817119865y10038171003817100381710038172
1003817100381710038171003817y10038171003817100381710038172
= maxy2=1
1003817100381710038171003817119865y10038171003817100381710038172
le maxw2=1
119865w2= maxw2=1
1003817100381710038171003817120582119865w10038171003817100381710038172
(31)
Since all the eigenvalues of 119865 are equal then
1198652le1003816100381610038161003816120582119865
1003816100381610038161003816 maxw2=1
w2=1003816100381610038161003816120582119865
1003816100381610038161003816 = 120588 [119865] (32)
Thus from (30) and (32) 1198652= |120582119865| = |(119903 minus 119892)(119903 + 1)|
By Lemma 2 with 119890119895replaced by 1 |(119903 minus 119892)(119903 + 1)| lt 1
is obtained leading to the result of Proposition 3 which is1198652= (119903119868 minus 119892119866
1)(119903119868 + 119866
1)minus1
2lt 1
Proposition 4 (119903119868 minus 1198921198662)(119903119868 + 119866
2)minus1
2lt 1
Proof Let 119870 = (119903119868 minus 1198921198662)(119903119868 + 119866
2)minus1 that is
119870 =
[[[[[[[[[[[[[
[
119903 minus 1198921198901
1199111
d sdot sdot sdot sdot sdot sdot sdot sdot sdot
119903 minus 1198921198902
1199112
d sdot sdot sdot
d
119874 d d
119903 minus 119892119890119898
119911119898
]]]]]]]]]]]]]
]
(33)
119870 is an upper triangular matrix with all the diagonal entriesequal to (119903 minus 119892119890
119894)119911119894 119894 = 1 2 119898 minus 2 where 119911
119894= 119903 + 119890
119894
Since all the eigenvalues of119870 are distinct then119870 is similar toa diagonal matrix119863
By the Schur triangularization theorem [20] there is anorthogonal matrix 119874 such that 119874119879119870119874 = 119863 The diagonalentries of119863 are the eigenvalues of119870
1198702=10038171003817100381710038171003817119874119879
119870119874100381710038171003817100381710038172
= 1198632
= radicmaximum eigenvalue of 119863119879119863
= radicmaximum eigenvalue of 1198632
= radic120588 (1198632) = radic120588 (1198702) = 120588 (119870)
(34)
By Lemma 2
1198702= 120588 (119870) = max
10038161003816100381610038161003816100381610038161003816
119903 minus 119892119890119894
119903 + 119890119894
10038161003816100381610038161003816100381610038161003816
=
1003816100381610038161003816100381610038161003816100381610038161003816
119903 minus 119892119890119895
119903 + 119890119895
1003816100381610038161003816100381610038161003816100381610038161003816
lt 1 (35)
for some 119895 Hence this proves Proposition 4
From (20) and (21)
u(119901+1) = (119903119868 + 1198662)minus1
(119903119868 minus 1198921198661)
times [(119903119868 + 1198661)minus1
(119903119868 minus 1198921198662) u(119901) + (119903119868 + 119866
1)minus1f]
+ 119892 (119903119868 + 1198662)minus1f
(36)
Let
119872(119903) = (119903119868 + 1198662)minus1
(119903119868 minus 1198921198661) (119903119868 + 119866
1)minus1
(119903119868 minus 1198921198662)
(37)
And let
119902 (119903) = [(119903119868 + 1198662)minus1
(119903119868 minus 1198921198661) (119903119868 + 119866
1)minus1
+119892(119903119868 + 1198662)minus1
] f (38)
Then u(119901+1) = 119872(119903)u(119901) + 119902(119903)
Theorem 5 The IADEMF4 is convergent if 120588[119872(119903)] lt 1 for119903 gt 0
Journal of Applied Mathematics 7
Proof Define (119903) = (119903119868 + 1198662)119872(119903)(119903119868 + 119866
2)minus1 then
(119903) = (119903119868 minus 1198921198661) (119903119868 + 119866
1)minus1
(119903119868 minus 1198921198662) (119903119868 + 119866
2)minus1
(39)
Thus by similarity119872(119903) and (119903) have the same set of eigen-values Therefore
120588 [119872 (119903)]
= 120588 [ (119903)] le10038171003817100381710038171003817 (119903)
10038171003817100381710038171003817 2
=10038171003817100381710038171003817(119903119868 minus 119892119866
1) (119903119868 + 119866
1)minus1
(119903119868 minus 1198921198662) (119903119868 + 119866
2)minus110038171003817100381710038171003817 2
le10038171003817100381710038171003817(119903119868 minus 119892119866
1) (119903119868 + 119866
1)minus1100381710038171003817100381710038172
10038171003817100381710038171003817(119903119868 minus 119892119866
2) (119903119868 + 119866
2)minus110038171003817100381710038171003817 2
lt 1
(40)
The last inequality is due to Propositions 3 and 4The provenTheorem 5 assures the convergence of the IADEMF4
5 The Fourth-Order ClassicalIterative Methods
The system of linear equations in (8) may also be solvedby using the classical iterative methods of the fourth orderThey include the JAC4 the GS4 and the SOR4 methodsThese iterative methods are capable of exploiting the sparsestructure of the pentadiagonal matrix
The JAC4 algorithm can be represented by
119906(119901+1)
119894=
(119891119894minus119886119906(119901)
119894minus2minus 119887119906(119901)
119894minus1minus 119889119906(119901)
119894+1minus 119890119906(119901)
119894+2)
119888
119894 = 2 3 119898 minus 1
(41)
The approximation of 119906(119901+1)119894
at the (119901 + 1)th iteration level iscomputed using the relevant values in the 119901th iteration
The GS4 method uses the most recent values of 119906(119901+1)119894minus2
and 119906(119901+1)
119894minus1to update the approximation value of 119906(119901+1)
119894 The
algorithm for the GS4 is as expressed in (42)
119906(119901+1)
119894=
(119891119894minus119886119906(119901+1)
119894minus2minus 119887119906(119901+1)
119894minus1minus 119889119906(119901)
119894+1minus 119890119906(119901)
119894+2)
119888
119894 = 2 3 119898 minus 1
(42)
The SOR4 iterative method accelerates the convergencerate of the GS4 If 120596 is a relaxation parameter then for any120596 = 0 (42) can be rewritten as
119906(119901+1)
119894= (1 minus 120596) 119906
(119901)
119894
+
120596 (119891119894minus 119886119906(119901+1)
119894minus2minus 119887119906(119901+1)
119894minus1minus 119889119906(119901)
119894+1minus 119890119906(119901)
119894+2)
119888
119894 = 2 3 119898 minus 1
(43)
i iminus 1i minus 2 i + 1 i + 2
(p + 1)
(p)
Figure 4 Computational molecule of the JAC4
i iminus 1i minus 2 i + 1 i + 2
(p + 1)
(p)
Figure 5 Computational molecule of the GS4SOR4
Table 1 Sequential arithmetic operations per iteration (119898 problemsize)
Method Number ofadditions
Number ofmultiplications
Total operationcount
IADEMF4 10(119898 minus 2) 13(119898 minus 2) 23(119898 minus 2)
IADEMF2 6119898 9119898 15119898
JAC4GS4 4(119898 minus 2) 5(119898 minus 2) 9(119898 minus 2)
SOR4 5(119898 minus 2) 7(119898 minus 2) 12(119898 minus 2)
The SOR4 algorithm reduces to the GS4 if 120596 = 1 Exceptfor very special cases it is difficult to obtain the analyticexpression for 120596 According to Young [21] the precise deter-mination of the optimal 120596 in the SOR is only known for asmall class of matrices The value of 120596 generally lies in therange of 1 lt 120596 lt 2
The computational molecules for the JAC4 and the GS4are as illustrated in Figures 4 and 5 respectively The SOR4has the same form as the GS4
6 Computational Complexity
The cost of implementing an algorithm can be assessedby examining its computational complexity The number ofarithmetic operations such as additions (including subtrac-tions) andmultiplications (including divisions) that is neededto perform by the algorithm can be straightforwardly count-ed Table 1 gives the number of sequential arithmetic opera-tions per iteration that is required to evaluate the algorithms
For the higher-order schemes trade-offbetween accuracyand speed usually happens The computational complexityhas an effect on the efficiency of a particular scheme Thusthis factor will be taken into account when discussing theresults of the numerical experiments conducted in this study
7 The Pseudocode of the IADEMF4Sequential Algorithm
Algorithm 1 illustrates the pseudocode of the IADEMF4sequential algorithm This algorithm can also be generalizedand applied to other methods discussed in this paper
8 Journal of Applied Mathematics
begindetermine the parameters119898 119903 Δ119905 Δ119909 120582 120576 119879
determine initial conditions 119906(119901)119894
determine exact solutions 119880119894at time level 119879
while (time level lt 119879)determine boundary conditionsfor 119894 = 2 to 119894 = 119898 minus 1
compute 119891119894(refer to (9))
end-forset iteration = 0while (convergence conditions are not satisfied)
for 119894 = 2 to 119894 = 119898 minus 1
compute 119906(119901+12)119894
(refer to (22))end-forfor 119894 = 2 to 119894 = 119898 minus 1
compute 119906(119901+1)119894
(refer to (23))end-fortest for convergencecompute 119890
119894larr
10038161003816100381610038161003816119906(119901+1)
119894minus 119906(119901)
119894
10038161003816100381610038161003816if (max1003816100381610038161003816119890119894
1003816100381610038161003816 lt 120576)then 119906
(119901)
119894larr 119906(119901+1)
119894
add 1 to iteration (if necessary)end-while
end-whilefor 119894 = 2 to 119894 = 119898 minus 1
determine average absolute error root meansquare error and maximum error
end-forend
Algorithm 1 The IADEMF4 sequential algorithm
8 Numerical Experiments
Two experiments were conducted to test the sequentialnumerical performance of the proposed IADEMF4 methodagainst those of the benchmarked classical iterative methodsnamely the JAC4 theGS4 and the SOR4 Comparison is alsomade with the corresponding IADE method of the secondorder
Experiment 1 This problem was taken from Saulev [22]
120597119880
120597119905=1205972
119880
1205971199092 0 le 119909 le 1 (44)
subject to the initial condition
119880 (119909 0) = 4119909 (1 minus 119909) 0 le 119909 le 1 (45)
and the boundary conditions
119880 (0 119905) = 119880 (1 119905) = 0 119905 ge 0 (46)
The exact solution to the given problem is given by
119880 (119909 119905) =32
1205873
infin
sum
119896=1(2)
1
1198963119890minus12058721198962119905 sin (119896120587119909) (47)
Experiment 2 This problem was taken from Johnson andReiss [23]
120597119880
120597119905=1205972
119880
1205971199092 0 le 119909 le 1 (48)
subject to the initial condition
119880 (119909 0) = sin (120587119909) (1 + 6 cos (120587119909)) 0 le 119909 le 1 (49)
and the boundary conditions
119880 (0 119905) = 119880 (1 119905) = 0 119905 ge 0 (50)
The exact solution to the given problem is given by
119880 (119909 119905) = sin (120587119909) 119890minus1205872119905
+ 3 sin (2120587119909) 119890minus41205872119905
(51)
Journal of Applied Mathematics 9
9 Results and Discussion
In each experiment the exact solutions at 119894 = 1 and 119894 = 119898
were taken as boundary values for the IADEMF4 and theother fourth-order classical iterative methods The conver-gence criterion used in the testing of each method was takenas 119906(119901+1) minus 119906
(119901)
infin
le 120576 where 120576 is the convergence toleranceThe selections of the optimum 119903 or 120596 were determined byexperiments As for the IADEMF2 the CN scheme with 120579 =
12 was employedFigures 6 7 8 and 9 visualize the behavior of the one-
dimensional parabolic heat solutions for both experimentsBy using119898 = 10 and a tolerance requirement of 120576 = 10
minus4 twodifferent mesh sizes 120582 = 05 and 120582 = 10 were consideredin each experiment The exact solutions are compared withthe IADEMF4 and the IADEMF2 numerical solutions Theoptimum value of 119903 chosen for each method as well as theircorresponding outcome of the number of iterations 119899 isstated in the legend of each figure Every figure reveals thatthe IADEMF4 is more accurate than the IADEMF2 andthe former converges with fewer numbers of iterations incomparison to the latter The numerical solutions of theIADEMF4 seem to be in very good agreement with theexact solutions For example in Figure 7 at 119909 = 05 thedifference between the exact solution and the numericalsolution using the IADEMF2 and using the IADEMF4 isabout 36 and 0 respectively These results imply thatthe accuracy and convergence rate of the second-order IADEmethod are enhanced by the implementation of the fourth-order CN approximation that leads to the formation of thecorresponding fourth-order IADE scheme
Figure 10 displays the graph of log (RMSE) versus log(Δ119909)for decreasing values of Δ119909 implemented on Experiment 2By considering 120576 = 10
minus4 and fixing Δ119905 = 00001 the value ofΔ119909 was initially taken as Δ119909 = 0125 It was then successivelyhalved into Δ1199092 Δ1199094 and Δ1199098
The figure shows that amongst the tested methods theroot mean square error (RMSE) of the IADEMF4 and theIADEMF2 decreases linearly as the values of Δ119909 decreasesThe slope of the IADEMF4 is approximately equal to 4which corresponds to its fourth-order spatial accuracy TheIADEMF2 with a second-order spatial accuracy has a slopethat is approximately equal to 2 It is clear that the IADEMF4is always more accurate than the IADEMF2 for the differentconsidered mesh sizes The graphs of the SOR4 (120596 = 105)the GS4 and the JAC4 show that their accuracies of fourthorder tend to lack as Δ119909 decreases largely due to the effect ofincreasing round-off errors as the value of119898 increases
Tables 2 3 4 5 6 7 8 and 9 provide numerical results interms of the average absolute error (AAE) root mean squareerror (RMSE) maximum error (ME) number of iterations(119899) and execution time (ET) measured in seconds (s) Theresults are obtained from both experiments for two differentvalues of 119898 and mesh size 120582 It is generally observed thatwhen 119898 = 700 120582 = 05 and 120576 = 10
minus6 the IADEMF4has the least average absolute error root mean square errorand maximum error in comparison with the other methodsunder consideration (Tables 2ndash5) When the size of 119898 was
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 08 n = 172)IADEMF2 (r = 03 n = 185)
00E + 00
10E minus 02
20E minus 02
30E minus 02
40E minus 02
50E minus 02
60E minus 02
70E minus 02
80E minus 02
90E minus 02
10E minus 01
U(xt)
Figure 6 Numerical and exact solutions for Experiment 1 120582 = 05Δ119909 = 01 Δ119905 = 0005 119905 = 025 and 120576 = 10
minus4
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 14 n = 159)IADEMF2 (r = 05 n = 162)
00E + 00
10E minus 03
20E minus 03
30E minus 03
40E minus 03
50E minus 03
60E minus 03
70E minus 03
80E minus 03
90E minus 03
U(xt)
Figure 7 Numerical and exact solutions for Experiment 1 120582 = 10Δ119909 = 01 Δ119905 = 001 119905 = 05 and 120576 = 10
minus4
ten times bigger (119898 = 7000) and a more stringent tolerancecriterion was set (120576 = 10
minus10
) the accuracy of the IADEMF4clearly outperforms the other methods for a mesh size of120582 = 05 (Tables 6 and 8) However the difference in the errorsamongst the tested methods is not so obvious for the case of120582 = 10 (Tables 7 and 9)The achievement of the fourth-orderIADEmethod in Experiments 1 and 2 can be clearly seen fromthe results in Tables 2 and 8 respectively where it has caused ahuge 94 reduction of RMSE from its corresponding second-order IADE method
The IADEMF4 has the advantage of featuring higheraccuracy due to the execution of the fourth-order CNapprox-imation coupled with the fourth-order accurate MF variant
10 Journal of Applied Mathematics
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 10 n = 169)IADEMF2 (r = 03 n = 213)
00E + 00
10E minus 02
20E minus 02
30E minus 02
40E minus 02
50E minus 02
60E minus 02
70E minus 02
80E minus 02
90E minus 02
10E minus 01
U(xt)
Figure 8 Numerical and exact solutions for Experiment 2 120582 = 05Δ119909 = 01 Δ119905 = 0005 119905 = 025 and 120576 = 10
minus4
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 10 n = 175)IADEMF2 (r = 05 n = 182)
00E + 00
10E minus 03
20E minus 03
30E minus 03
40E minus 03
50E minus 03
60E minus 03
70E minus 03
80E minus 03
U(xt)
Figure 9 Numerical and exact solutions for Experiment 2 120582 = 10Δ119909 = 01 Δ119905 = 001 119905 = 05 and 120576 = 10
minus4
The IADEMF2 is only derived from the second-order CNapproximation but its combination with its correspondingfourth-order MF variant managed to produce errors whichare better than theGS4 JAC4 and the SOR4 Even though theclassical iterative methods are also derived from the fourth-order accurate CN type approximation they are lackingin accuracy due to the round-off errors that have beenaccumulated from the time the execution starts till it ends
With regards to the rate of convergence the resultsfrom each table demonstrate that the number of iterationsproduced by the IADEMF2 and the fourth-order classicaliterative methods is greater than or at least equal to that of
0
minus2
minus208 minus277 minus347 minus416
minus4
minus8
minus6
minus10
minus12
minus14
minus16
minus18
GS4JAC4SOR4
log
(RM
SE)
log(Δx)
IADEMF4
IADEMF2
Figure 10 Log(RMSE) versus Log(Δ119909) Experiment 2 Δ119905 = 00001119905 = 0002 and 120576 = 10
minus4
Table 2 Experiment 1 119898 = 700 120582 = 05 Δ119909 = 143 times 10minus3 Δ119905 =
102 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 07) 267119890 minus 9 330119890 minus 9 212119890 minus 8 100 440119890 minus 2
IADEMF2(119903 = 08) 188119890 minus 8 495119890 minus 8 323119890 minus 7 100 635119890 minus 3
SOR4(120596 = 12) 540119890 minus 7 544119890 minus 7 552119890 minus 7 108 460119890 minus 2
GS4 540119890 minus 6 543119890 minus 6 332119890 minus 6 150 472119890 minus 2
JAC4 228119890 minus 5 229119890 minus 5 232119890 minus 5 150 473119890 minus 2
Table 3 Experiment 1 119898 = 700 120582 = 10 Δ119909 = 143 times 10minus3 Δ119905 =
204 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 297119890 minus 7 301119890 minus 7 301119890 minus 7 50 201119890 minus 2
IADEMF2(119903 = 09) 557119890 minus 7 559119890 minus 7 848119890 minus 7 50 317119890 minus 3
SOR4(120596 = 12) 329119890 minus 6 331119890 minus 6 335119890 minus 6 76 225119890 minus 2
GS4 895119890 minus 6 900119890 minus 6 900119890 minus 6 100 238119890 minus 2
JAC4 214119890 minus 5 215119890 minus 5 217119890 minus 5 125 250119890 minus 2
the IADEMF4 The convergence rate of the latter surpassesthe others in the case of 119898 = 7000 and 120576 = 10
minus10 (Tables6ndash8) Even though the operational count of the IADEMF4is relatively quite large (Table 1) due to its higher level ofaccuracy its increasing number of correct digits at eachiteration causes it to converge at a faster rate The IADEMF2has the advantage of having less mathematical operationsthus it is competitive in terms of convergence rate but atthe expense of accuracy The application of the fourth-orderCN approximation on the heat equation proves that thethree classical iterative methods also converge with JAC4appearing to be the slowest and the least accurate amongst all
Journal of Applied Mathematics 11
Table 4 Experiment 2 119898 = 700 120582 = 05 Δ119909 = 143 times 10minus3 Δ119905 =
102 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME n ET (s)IADEMF4(119903 = 07) 207119890 minus 8 231119890 minus 8 345119890 minus 8 100 903119890 minus 3
IADEMF2(119903 = 08) 106119890 minus 7 175119890 minus 7 175119890 minus 7 100 647119890 minus 3
SOR4(120596 = 11) 154119890 minus 6 171119890 minus 6 255119890 minus 6 200 110119890 minus 2
GS4 294119890 minus 6 327119890 minus 6 487119890 minus 6 250 118119890 minus 2
JAC4 124119890 minus 5 138119890 minus 5 206119890 minus 5 300 127119890 minus 2
Table 5 Experiment 2 119898 = 700 120582 = 10 Δ119909 = 143 times 10minus3 Δ119905 =
204 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 09) 138119890 minus 6 154119890 minus 6 229119890 minus 6 75 650119890 minus 3
IADEMF2(119903 = 05) 187119890 minus 6 207119890 minus 6 309119890 minus 6 75 472119890 minus 3
SOR4(120596 = 12) 312119890 minus 6 347119890 minus 6 517119890 minus 6 125 906119890 minus 3
GS4 667119890 minus 6 742119890 minus 6 111119890 minus 5 175 932119890 minus 3
JAC4 126119890 minus 5 141119890 minus 5 210119890 minus 5 250 105119890 minus 2
Table 6 Experiment 1 119898 = 7000 120582 = 05 Δ119909 = 143 times 10minus4 Δ119905 =
102 times 10minus8 119905 = 510 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME n ET (s)IADEMF4(119903 = 08) 367119890 minus 10 433119890 minus 10 366119890 minus 9 173 146119890 minus 1
IADEMF2(119903 = 05) 375119890 minus 10 435119890 minus 10 376119890 minus 9 200 124119890 minus 1
SOR4(120596 = 112) 455119890 minus 10 507119890 minus 10 367119890 minus 9 257 147119890 minus 1
GS4 110119890 minus 9 111119890 minus 9 376119890 minus 9 300 160119890 minus 1
JAC4 230119890 minus 9 231119890 minus 9 378119890 minus 9 400 163119890 minus 1
Table 7 Experiment 1 119898 = 7000 120582 = 10 Δ119909 = 143 times 10minus4 Δ119905 =
204 times 10minus8 119905 = 51 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 163119890 minus 7 163119890 minus 7 163119890 minus 7 112 960119890 minus 2
IADEMF2(119903 = 12) 163119890 minus 7 163119890 minus 7 163119890 minus 7 120 736119890 minus 2
SOR4(120596 = 115) 163119890 minus 7 163119890 minus 7 164119890 minus 7 171 999119890 minus 2
GS4 164119890 minus 7 164119890 minus 7 164119890 minus 7 216 107119890 minus 1
JAC4 165119890 minus 7 165119890 minus 7 165119890 minus 7 312 111119890 minus 1
It is found that the convergence rate of the fourth-orderand second-order methods improves with the application oflarger mesh size that is from 120582 = 05 to 120582 = 10 Thecoarser meshes cause a reduction in the computational oper-ations thus giving better rate of convergence However a
Table 8 Experiment 2 119898 = 7000 120582 = 05 Δ119909 = 143 times 10minus4 Δ119905 =
102 times 10minus8 119905 = 510 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 07) 567119890 minus 13 321119890 minus 12 130119890 minus 10 150 131119890 minus 1
IADEMF2(119903 = 07) 199119890 minus 11 460119890 minus 11 182119890 minus 9 200 127119890 minus 1
SOR4(120596 = 115) 245119890 minus 10 273119890 minus 10 399119890 minus 10 260 151119890 minus 1
GS4 398119890 minus 10 443119890 minus 10 661119890 minus 10 400 171119890 minus 1
JAC4 105119890 minus 9 117119890 minus 9 174119890 minus 9 550 176119890 minus 1
Table 9 Experiment 2 119898 = 7000 120582 = 10 Δ119909 = 143 times 10minus4 Δ119905 =
204 times 10minus8 119905 = 51 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 154119890 minus 6 171119890 minus 6 255119890 minus 6 96 912119890 minus 2
IADEMF2(119903 = 10) 154119890 minus 6 171119890 minus 6 255119890 minus 6 96 607119890 minus 2
SOR4(120596 = 12) 154119890 minus 6 171119890 minus 6 255119890 minus 6 192 997119890 minus 2
GS4 154119890 minus 6 171119890 minus 6 255119890 minus 6 288 128119890 minus 1
JAC4 154119890 minus 6 172119890 minus 6 256119890 minus 6 408 146119890 minus 1
more accurate solution is obtained by using finer mesh Forexample Table 6 shows that for 120582 = 05 the IADEMF4 hasRMSE = 433119890 minus 10 and 119899 = 173 whereas Table 7 shows thatfor 120582 = 10 its RMSE = 163119890 minus 7 and 119899 = 112 In generalamongst all the testedmethods the IADEMF4 still maintainsits greater accuracy characteristic even with coarser meshes
In terms of execution time the results from every tabledisplay shorter execution time for the IADEMF2 in compari-son to the IADEMF4 This is expected since the IADEMF2has lower computational complexity (Table 1) Despite theachievement in accuracy the IADEMF4 has to performmorecomputational work than the IADEMF2 since the former hasto utilize values of 119906 at two grid points on either side of thepoint (119894Δ119909 119896Δ119905) along the 119896th time level Thus if accuracyis desired then the preferred sequential numerical algorithmwould be the IADEMF4 On the other hand if execution timematters then the choice would be the IADEMF2
Amongst the fourth-order methods the IADEMF4 exe-cutes in the least amount of time Even though its computa-tional complexity is relatively large it operates with the leastnumber of iterations thus enabling it to be the most efficient
10 Conclusion
This paper proposes the development of the novel fourth-order IADEMF4 finite difference scheme The higher-orderscheme is developed by representing the spatial derivative inthe heat equation with the fourth-order finite difference CNapproximation This leads to the formation of pentadiagonalmatrices in the systems of linear equations with larger com-putational stencils The algorithm also employs the higheraccuracy of the Mitchell and Fairweather variant Despite
12 Journal of Applied Mathematics
the fourth-order IADE schemersquos higher computational com-plexity this technique is proved to be valuable becauseit enhances the accuracy of its second-order counterpartnamely the IADEMF2 In addition the higher accuracy ofIADEMF4 is verified as a convergent and unconditionallystable scheme and is superior in terms of rate of convergenceIt has also been proven to execute more efficiently in compar-ison to the other benchmarked fourth-order classical iterativemethods such as the GS4 the SOR4 and the JAC4 The in-creasing number of correct digits at each iteration serves as anadvantage for the IADEMF4 thus yielding faster rate of con-vergence with higher level of accuracy
In conclusion the proposed IADEMF4 scheme affordsusers many advantages with respect to higher accuracy sta-bility and rate of convergence and it serves as an alternativeefficient technique for the solution of a one-dimensional heatequation with Dirichlet boundary conditions
The proposed fourth-order scheme can be modified andadapted to more general multidimensional linear and non-linear parabolic elliptic and hyperbolic partial differentialequations using different types of boundary conditionsBesides it can also be considered for applications in problemsthat require higher-order accuracy with high resolution suchas problems in nanocomputing that require the solution ofvery large sparse systems of equation
The explicit and high accuracy feature of the IADEMF4can be exploited in two- and three-dimensional heat prob-lems by applying the scheme as pentadiagonal solvers forthe system of linear equations arising in the sweeps of ahigher-order alternating direction implicit (ADI) schemeThe lower-order ADI scheme was initially proposed byPeaceman and Rachford [24] The higher space dimensionsare expected not to cause unsatisfactory performance for theADI-IADEMF4 as long as the proposed scheme is stableconvergent and efficient as the process of its implementa-tion continues to advance from one-time step to anotherPathirana et al [25] implemented the ADI in developing atwo-dimensional model for incorporating flood damage inurban drainage planningThe proposed model is found to bestable numerically accurate and computationally efficientMirzavand et al [26] proposed the ADI-FDTD methodfor the physical modeling of high-frequency semiconductordevices The approach is able to reduce significantly the full-wave simulation time
It is possible to parallelize the IADEMF4 algorithm sincethe calculation of the new iteration only depends on theknown values from the last iteration (Figures 2 and 3) Futurework is to exploit the explicit computational properties andefficiency of the IADEMF4 for parallelization and executionon distributed parallel computing systems The idea is tospeed up the execution time without compromising its accu-racy especially on problems involving very large linear sys-tems of equations
References
[1] J Noye Computational Techniques for Differential Equationsvol 83 Elsevier Science North-Holland Publishing Amster-dam The Netherlands 1984
[2] G W Recktenwald ldquoFinite difference approximations to theheat equationrdquoMechanical Engineering vol 10 pp 1ndash27 2004
[3] P Lax and B Wendroff ldquoSystems of conservation lawsrdquo Com-munications on Pure and Applied Mathematics vol 13 pp 217ndash237 1960
[4] R J MacKinnon and G F Carey ldquoAnalysis of material interfacediscontinuities and superconvergent fluxes in finite differencetheoryrdquo Journal of Computational Physics vol 75 no 1 pp 151ndash167 1988
[5] R Hixon and E Turkel High Accuracy Compact MacCormack-Type Schemes For Computational Aerocoustics NASACentre forAerospace Information Lewis Research Centre 1998
[6] Y Lin Higher-order finite difference methods for solving heatequations [PhD thesis] Southern Illinois University 2008
[7] J Sulaiman M K Hasan M Othman M and S A AbdulKarim ldquoFourth-order QSMOR iterative method for the solu-tion of one-dimensional parabolic PDErsquosrdquo in Proceedings of theInternational Conference on Industrial and AppliedMathematics(CIAM rsquo10) pp 34ndash39 ITB Bandung Indonesia 2010
[8] N Jha ldquoThe application of sixth order accurate parallel quar-ter sweep alternating group explicit algorithm for nonlinearboundary value problems with singularityrdquo in Proceedings ofthe 2nd International Conference on Methods and Models inComputer Science (ICM2CS rsquo10) pp 76ndash80 December 2010
[9] Y Jin G Jin and J Li ldquoA class of high-precision finite differenceparallel algorithms for convection equationsrdquo Journal of Conver-gence Information Technology vol 6 no 1 pp 79ndash82 2011
[10] W Fu and E L Tan ldquoDevelopment of split-step FDTDmethodwith higher-order spatial accuracyrdquo Electronics Letters vol 40no 20 pp 1252ndash1253 2004
[11] A Mohebbi andM Dehghan ldquoHigh-order compact solution ofthe one-dimensional heat and advection-diffusion equationsrdquoApplied Mathematical Modelling vol 34 no 10 pp 3071ndash30842010
[12] W Liao ldquoAn implicit fourth-order compact finite differencescheme for one-dimensional Burgersrsquo equationrdquo Applied Math-ematics and Computation vol 206 no 2 pp 755ndash764 2008
[13] Z F Tian and Y B Ge ldquoA fourth-order compact ADI methodfor solving two-dimensional unsteady convection-diffusionproblemsrdquo Journal of Computational and Applied Mathematicsvol 198 no 1 pp 268ndash286 2007
[14] C Chun ldquoSome fourth-order iterativemethods for solving non-linear equationsrdquo Applied Mathematics and Computation vol195 no 2 pp 454ndash459 2008
[15] S Zhu Z Yu and J Zhao ldquoAhigh-order parallel finite differencealgorithmrdquo Applied Mathematics and Computation vol 183 no1 pp 365ndash372 2006
[16] Z Gao and S Xie ldquoFourth-order alternating direction implicitcompact finite difference schemes for two-dimensionalSchrodinger equationsrdquo Applied Numerical Mathematics vol61 no 4 pp 593ndash614 2011
[17] M S Sahimi A Ahmad and A A Bakar ldquoThe Iterative Alter-nating Decomposition Explicit (IADE) method to solve theheat conduction equationrdquo International Journal of ComputerMathematics vol 47 pp 219ndash229 1993
[18] G D Smith Numerical Solution of Partial Differential Equa-tions Finite Difference Methods Oxford University Press 2ndedition 1978
[19] A RMitchell andG Fairweather ldquoImproved forms of the alter-nating direction methods of Douglas Peaceman and Rachfordfor solving parabolic and elliptic equationsrdquo Numerische Math-ematik vol 6 no 1 pp 285ndash292 1964
Journal of Applied Mathematics 13
[20] B N DattaNumerical Linear Algebra and Applications BrooksCole Publishing Pacific Grove Calif USA 1995
[21] D M Young Iterative Solution of Large Linear Systems Aca-demic Press New York NY USA 1971
[22] V K Saulev Integration of Equations of Parabolic Type by theMethod of Nets Pergamon Press Oxford UK 1964
[23] L W Johnson and R D Riess Numerical Analysis Addison-Wesley Reading Mass USA 2nd edition 1982
[24] D W Peaceman and H H Rachford Jr ldquoThe numerical solu-tion of parabolic and elliptic differential equationsrdquo Journal ofthe Society for Industrial and Applied Mathematics vol 3 pp28ndash41 1955
[25] A Pathirana S Tsegaye B Gersonius and K VairavamoorthyldquoA simple 2-D inundation model for incorporating flood dam-age in urban drainage planningrdquo Hydrology and Earth SystemSciences vol 15 no 8 pp 2747ndash2761 2011
[26] R Mirzavand A Abdipour G Moradi and M MovahhedildquoFull-wave semiconductor devices simulation using ADI-FDTD methodrdquo Progress in Electromagnetics Research vol 11pp 191ndash202 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Journal of Applied Mathematics 7
Proof Define (119903) = (119903119868 + 1198662)119872(119903)(119903119868 + 119866
2)minus1 then
(119903) = (119903119868 minus 1198921198661) (119903119868 + 119866
1)minus1
(119903119868 minus 1198921198662) (119903119868 + 119866
2)minus1
(39)
Thus by similarity119872(119903) and (119903) have the same set of eigen-values Therefore
120588 [119872 (119903)]
= 120588 [ (119903)] le10038171003817100381710038171003817 (119903)
10038171003817100381710038171003817 2
=10038171003817100381710038171003817(119903119868 minus 119892119866
1) (119903119868 + 119866
1)minus1
(119903119868 minus 1198921198662) (119903119868 + 119866
2)minus110038171003817100381710038171003817 2
le10038171003817100381710038171003817(119903119868 minus 119892119866
1) (119903119868 + 119866
1)minus1100381710038171003817100381710038172
10038171003817100381710038171003817(119903119868 minus 119892119866
2) (119903119868 + 119866
2)minus110038171003817100381710038171003817 2
lt 1
(40)
The last inequality is due to Propositions 3 and 4The provenTheorem 5 assures the convergence of the IADEMF4
5 The Fourth-Order ClassicalIterative Methods
The system of linear equations in (8) may also be solvedby using the classical iterative methods of the fourth orderThey include the JAC4 the GS4 and the SOR4 methodsThese iterative methods are capable of exploiting the sparsestructure of the pentadiagonal matrix
The JAC4 algorithm can be represented by
119906(119901+1)
119894=
(119891119894minus119886119906(119901)
119894minus2minus 119887119906(119901)
119894minus1minus 119889119906(119901)
119894+1minus 119890119906(119901)
119894+2)
119888
119894 = 2 3 119898 minus 1
(41)
The approximation of 119906(119901+1)119894
at the (119901 + 1)th iteration level iscomputed using the relevant values in the 119901th iteration
The GS4 method uses the most recent values of 119906(119901+1)119894minus2
and 119906(119901+1)
119894minus1to update the approximation value of 119906(119901+1)
119894 The
algorithm for the GS4 is as expressed in (42)
119906(119901+1)
119894=
(119891119894minus119886119906(119901+1)
119894minus2minus 119887119906(119901+1)
119894minus1minus 119889119906(119901)
119894+1minus 119890119906(119901)
119894+2)
119888
119894 = 2 3 119898 minus 1
(42)
The SOR4 iterative method accelerates the convergencerate of the GS4 If 120596 is a relaxation parameter then for any120596 = 0 (42) can be rewritten as
119906(119901+1)
119894= (1 minus 120596) 119906
(119901)
119894
+
120596 (119891119894minus 119886119906(119901+1)
119894minus2minus 119887119906(119901+1)
119894minus1minus 119889119906(119901)
119894+1minus 119890119906(119901)
119894+2)
119888
119894 = 2 3 119898 minus 1
(43)
i iminus 1i minus 2 i + 1 i + 2
(p + 1)
(p)
Figure 4 Computational molecule of the JAC4
i iminus 1i minus 2 i + 1 i + 2
(p + 1)
(p)
Figure 5 Computational molecule of the GS4SOR4
Table 1 Sequential arithmetic operations per iteration (119898 problemsize)
Method Number ofadditions
Number ofmultiplications
Total operationcount
IADEMF4 10(119898 minus 2) 13(119898 minus 2) 23(119898 minus 2)
IADEMF2 6119898 9119898 15119898
JAC4GS4 4(119898 minus 2) 5(119898 minus 2) 9(119898 minus 2)
SOR4 5(119898 minus 2) 7(119898 minus 2) 12(119898 minus 2)
The SOR4 algorithm reduces to the GS4 if 120596 = 1 Exceptfor very special cases it is difficult to obtain the analyticexpression for 120596 According to Young [21] the precise deter-mination of the optimal 120596 in the SOR is only known for asmall class of matrices The value of 120596 generally lies in therange of 1 lt 120596 lt 2
The computational molecules for the JAC4 and the GS4are as illustrated in Figures 4 and 5 respectively The SOR4has the same form as the GS4
6 Computational Complexity
The cost of implementing an algorithm can be assessedby examining its computational complexity The number ofarithmetic operations such as additions (including subtrac-tions) andmultiplications (including divisions) that is neededto perform by the algorithm can be straightforwardly count-ed Table 1 gives the number of sequential arithmetic opera-tions per iteration that is required to evaluate the algorithms
For the higher-order schemes trade-offbetween accuracyand speed usually happens The computational complexityhas an effect on the efficiency of a particular scheme Thusthis factor will be taken into account when discussing theresults of the numerical experiments conducted in this study
7 The Pseudocode of the IADEMF4Sequential Algorithm
Algorithm 1 illustrates the pseudocode of the IADEMF4sequential algorithm This algorithm can also be generalizedand applied to other methods discussed in this paper
8 Journal of Applied Mathematics
begindetermine the parameters119898 119903 Δ119905 Δ119909 120582 120576 119879
determine initial conditions 119906(119901)119894
determine exact solutions 119880119894at time level 119879
while (time level lt 119879)determine boundary conditionsfor 119894 = 2 to 119894 = 119898 minus 1
compute 119891119894(refer to (9))
end-forset iteration = 0while (convergence conditions are not satisfied)
for 119894 = 2 to 119894 = 119898 minus 1
compute 119906(119901+12)119894
(refer to (22))end-forfor 119894 = 2 to 119894 = 119898 minus 1
compute 119906(119901+1)119894
(refer to (23))end-fortest for convergencecompute 119890
119894larr
10038161003816100381610038161003816119906(119901+1)
119894minus 119906(119901)
119894
10038161003816100381610038161003816if (max1003816100381610038161003816119890119894
1003816100381610038161003816 lt 120576)then 119906
(119901)
119894larr 119906(119901+1)
119894
add 1 to iteration (if necessary)end-while
end-whilefor 119894 = 2 to 119894 = 119898 minus 1
determine average absolute error root meansquare error and maximum error
end-forend
Algorithm 1 The IADEMF4 sequential algorithm
8 Numerical Experiments
Two experiments were conducted to test the sequentialnumerical performance of the proposed IADEMF4 methodagainst those of the benchmarked classical iterative methodsnamely the JAC4 theGS4 and the SOR4 Comparison is alsomade with the corresponding IADE method of the secondorder
Experiment 1 This problem was taken from Saulev [22]
120597119880
120597119905=1205972
119880
1205971199092 0 le 119909 le 1 (44)
subject to the initial condition
119880 (119909 0) = 4119909 (1 minus 119909) 0 le 119909 le 1 (45)
and the boundary conditions
119880 (0 119905) = 119880 (1 119905) = 0 119905 ge 0 (46)
The exact solution to the given problem is given by
119880 (119909 119905) =32
1205873
infin
sum
119896=1(2)
1
1198963119890minus12058721198962119905 sin (119896120587119909) (47)
Experiment 2 This problem was taken from Johnson andReiss [23]
120597119880
120597119905=1205972
119880
1205971199092 0 le 119909 le 1 (48)
subject to the initial condition
119880 (119909 0) = sin (120587119909) (1 + 6 cos (120587119909)) 0 le 119909 le 1 (49)
and the boundary conditions
119880 (0 119905) = 119880 (1 119905) = 0 119905 ge 0 (50)
The exact solution to the given problem is given by
119880 (119909 119905) = sin (120587119909) 119890minus1205872119905
+ 3 sin (2120587119909) 119890minus41205872119905
(51)
Journal of Applied Mathematics 9
9 Results and Discussion
In each experiment the exact solutions at 119894 = 1 and 119894 = 119898
were taken as boundary values for the IADEMF4 and theother fourth-order classical iterative methods The conver-gence criterion used in the testing of each method was takenas 119906(119901+1) minus 119906
(119901)
infin
le 120576 where 120576 is the convergence toleranceThe selections of the optimum 119903 or 120596 were determined byexperiments As for the IADEMF2 the CN scheme with 120579 =
12 was employedFigures 6 7 8 and 9 visualize the behavior of the one-
dimensional parabolic heat solutions for both experimentsBy using119898 = 10 and a tolerance requirement of 120576 = 10
minus4 twodifferent mesh sizes 120582 = 05 and 120582 = 10 were consideredin each experiment The exact solutions are compared withthe IADEMF4 and the IADEMF2 numerical solutions Theoptimum value of 119903 chosen for each method as well as theircorresponding outcome of the number of iterations 119899 isstated in the legend of each figure Every figure reveals thatthe IADEMF4 is more accurate than the IADEMF2 andthe former converges with fewer numbers of iterations incomparison to the latter The numerical solutions of theIADEMF4 seem to be in very good agreement with theexact solutions For example in Figure 7 at 119909 = 05 thedifference between the exact solution and the numericalsolution using the IADEMF2 and using the IADEMF4 isabout 36 and 0 respectively These results imply thatthe accuracy and convergence rate of the second-order IADEmethod are enhanced by the implementation of the fourth-order CN approximation that leads to the formation of thecorresponding fourth-order IADE scheme
Figure 10 displays the graph of log (RMSE) versus log(Δ119909)for decreasing values of Δ119909 implemented on Experiment 2By considering 120576 = 10
minus4 and fixing Δ119905 = 00001 the value ofΔ119909 was initially taken as Δ119909 = 0125 It was then successivelyhalved into Δ1199092 Δ1199094 and Δ1199098
The figure shows that amongst the tested methods theroot mean square error (RMSE) of the IADEMF4 and theIADEMF2 decreases linearly as the values of Δ119909 decreasesThe slope of the IADEMF4 is approximately equal to 4which corresponds to its fourth-order spatial accuracy TheIADEMF2 with a second-order spatial accuracy has a slopethat is approximately equal to 2 It is clear that the IADEMF4is always more accurate than the IADEMF2 for the differentconsidered mesh sizes The graphs of the SOR4 (120596 = 105)the GS4 and the JAC4 show that their accuracies of fourthorder tend to lack as Δ119909 decreases largely due to the effect ofincreasing round-off errors as the value of119898 increases
Tables 2 3 4 5 6 7 8 and 9 provide numerical results interms of the average absolute error (AAE) root mean squareerror (RMSE) maximum error (ME) number of iterations(119899) and execution time (ET) measured in seconds (s) Theresults are obtained from both experiments for two differentvalues of 119898 and mesh size 120582 It is generally observed thatwhen 119898 = 700 120582 = 05 and 120576 = 10
minus6 the IADEMF4has the least average absolute error root mean square errorand maximum error in comparison with the other methodsunder consideration (Tables 2ndash5) When the size of 119898 was
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 08 n = 172)IADEMF2 (r = 03 n = 185)
00E + 00
10E minus 02
20E minus 02
30E minus 02
40E minus 02
50E minus 02
60E minus 02
70E minus 02
80E minus 02
90E minus 02
10E minus 01
U(xt)
Figure 6 Numerical and exact solutions for Experiment 1 120582 = 05Δ119909 = 01 Δ119905 = 0005 119905 = 025 and 120576 = 10
minus4
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 14 n = 159)IADEMF2 (r = 05 n = 162)
00E + 00
10E minus 03
20E minus 03
30E minus 03
40E minus 03
50E minus 03
60E minus 03
70E minus 03
80E minus 03
90E minus 03
U(xt)
Figure 7 Numerical and exact solutions for Experiment 1 120582 = 10Δ119909 = 01 Δ119905 = 001 119905 = 05 and 120576 = 10
minus4
ten times bigger (119898 = 7000) and a more stringent tolerancecriterion was set (120576 = 10
minus10
) the accuracy of the IADEMF4clearly outperforms the other methods for a mesh size of120582 = 05 (Tables 6 and 8) However the difference in the errorsamongst the tested methods is not so obvious for the case of120582 = 10 (Tables 7 and 9)The achievement of the fourth-orderIADEmethod in Experiments 1 and 2 can be clearly seen fromthe results in Tables 2 and 8 respectively where it has caused ahuge 94 reduction of RMSE from its corresponding second-order IADE method
The IADEMF4 has the advantage of featuring higheraccuracy due to the execution of the fourth-order CNapprox-imation coupled with the fourth-order accurate MF variant
10 Journal of Applied Mathematics
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 10 n = 169)IADEMF2 (r = 03 n = 213)
00E + 00
10E minus 02
20E minus 02
30E minus 02
40E minus 02
50E minus 02
60E minus 02
70E minus 02
80E minus 02
90E minus 02
10E minus 01
U(xt)
Figure 8 Numerical and exact solutions for Experiment 2 120582 = 05Δ119909 = 01 Δ119905 = 0005 119905 = 025 and 120576 = 10
minus4
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 10 n = 175)IADEMF2 (r = 05 n = 182)
00E + 00
10E minus 03
20E minus 03
30E minus 03
40E minus 03
50E minus 03
60E minus 03
70E minus 03
80E minus 03
U(xt)
Figure 9 Numerical and exact solutions for Experiment 2 120582 = 10Δ119909 = 01 Δ119905 = 001 119905 = 05 and 120576 = 10
minus4
The IADEMF2 is only derived from the second-order CNapproximation but its combination with its correspondingfourth-order MF variant managed to produce errors whichare better than theGS4 JAC4 and the SOR4 Even though theclassical iterative methods are also derived from the fourth-order accurate CN type approximation they are lackingin accuracy due to the round-off errors that have beenaccumulated from the time the execution starts till it ends
With regards to the rate of convergence the resultsfrom each table demonstrate that the number of iterationsproduced by the IADEMF2 and the fourth-order classicaliterative methods is greater than or at least equal to that of
0
minus2
minus208 minus277 minus347 minus416
minus4
minus8
minus6
minus10
minus12
minus14
minus16
minus18
GS4JAC4SOR4
log
(RM
SE)
log(Δx)
IADEMF4
IADEMF2
Figure 10 Log(RMSE) versus Log(Δ119909) Experiment 2 Δ119905 = 00001119905 = 0002 and 120576 = 10
minus4
Table 2 Experiment 1 119898 = 700 120582 = 05 Δ119909 = 143 times 10minus3 Δ119905 =
102 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 07) 267119890 minus 9 330119890 minus 9 212119890 minus 8 100 440119890 minus 2
IADEMF2(119903 = 08) 188119890 minus 8 495119890 minus 8 323119890 minus 7 100 635119890 minus 3
SOR4(120596 = 12) 540119890 minus 7 544119890 minus 7 552119890 minus 7 108 460119890 minus 2
GS4 540119890 minus 6 543119890 minus 6 332119890 minus 6 150 472119890 minus 2
JAC4 228119890 minus 5 229119890 minus 5 232119890 minus 5 150 473119890 minus 2
Table 3 Experiment 1 119898 = 700 120582 = 10 Δ119909 = 143 times 10minus3 Δ119905 =
204 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 297119890 minus 7 301119890 minus 7 301119890 minus 7 50 201119890 minus 2
IADEMF2(119903 = 09) 557119890 minus 7 559119890 minus 7 848119890 minus 7 50 317119890 minus 3
SOR4(120596 = 12) 329119890 minus 6 331119890 minus 6 335119890 minus 6 76 225119890 minus 2
GS4 895119890 minus 6 900119890 minus 6 900119890 minus 6 100 238119890 minus 2
JAC4 214119890 minus 5 215119890 minus 5 217119890 minus 5 125 250119890 minus 2
the IADEMF4 The convergence rate of the latter surpassesthe others in the case of 119898 = 7000 and 120576 = 10
minus10 (Tables6ndash8) Even though the operational count of the IADEMF4is relatively quite large (Table 1) due to its higher level ofaccuracy its increasing number of correct digits at eachiteration causes it to converge at a faster rate The IADEMF2has the advantage of having less mathematical operationsthus it is competitive in terms of convergence rate but atthe expense of accuracy The application of the fourth-orderCN approximation on the heat equation proves that thethree classical iterative methods also converge with JAC4appearing to be the slowest and the least accurate amongst all
Journal of Applied Mathematics 11
Table 4 Experiment 2 119898 = 700 120582 = 05 Δ119909 = 143 times 10minus3 Δ119905 =
102 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME n ET (s)IADEMF4(119903 = 07) 207119890 minus 8 231119890 minus 8 345119890 minus 8 100 903119890 minus 3
IADEMF2(119903 = 08) 106119890 minus 7 175119890 minus 7 175119890 minus 7 100 647119890 minus 3
SOR4(120596 = 11) 154119890 minus 6 171119890 minus 6 255119890 minus 6 200 110119890 minus 2
GS4 294119890 minus 6 327119890 minus 6 487119890 minus 6 250 118119890 minus 2
JAC4 124119890 minus 5 138119890 minus 5 206119890 minus 5 300 127119890 minus 2
Table 5 Experiment 2 119898 = 700 120582 = 10 Δ119909 = 143 times 10minus3 Δ119905 =
204 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 09) 138119890 minus 6 154119890 minus 6 229119890 minus 6 75 650119890 minus 3
IADEMF2(119903 = 05) 187119890 minus 6 207119890 minus 6 309119890 minus 6 75 472119890 minus 3
SOR4(120596 = 12) 312119890 minus 6 347119890 minus 6 517119890 minus 6 125 906119890 minus 3
GS4 667119890 minus 6 742119890 minus 6 111119890 minus 5 175 932119890 minus 3
JAC4 126119890 minus 5 141119890 minus 5 210119890 minus 5 250 105119890 minus 2
Table 6 Experiment 1 119898 = 7000 120582 = 05 Δ119909 = 143 times 10minus4 Δ119905 =
102 times 10minus8 119905 = 510 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME n ET (s)IADEMF4(119903 = 08) 367119890 minus 10 433119890 minus 10 366119890 minus 9 173 146119890 minus 1
IADEMF2(119903 = 05) 375119890 minus 10 435119890 minus 10 376119890 minus 9 200 124119890 minus 1
SOR4(120596 = 112) 455119890 minus 10 507119890 minus 10 367119890 minus 9 257 147119890 minus 1
GS4 110119890 minus 9 111119890 minus 9 376119890 minus 9 300 160119890 minus 1
JAC4 230119890 minus 9 231119890 minus 9 378119890 minus 9 400 163119890 minus 1
Table 7 Experiment 1 119898 = 7000 120582 = 10 Δ119909 = 143 times 10minus4 Δ119905 =
204 times 10minus8 119905 = 51 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 163119890 minus 7 163119890 minus 7 163119890 minus 7 112 960119890 minus 2
IADEMF2(119903 = 12) 163119890 minus 7 163119890 minus 7 163119890 minus 7 120 736119890 minus 2
SOR4(120596 = 115) 163119890 minus 7 163119890 minus 7 164119890 minus 7 171 999119890 minus 2
GS4 164119890 minus 7 164119890 minus 7 164119890 minus 7 216 107119890 minus 1
JAC4 165119890 minus 7 165119890 minus 7 165119890 minus 7 312 111119890 minus 1
It is found that the convergence rate of the fourth-orderand second-order methods improves with the application oflarger mesh size that is from 120582 = 05 to 120582 = 10 Thecoarser meshes cause a reduction in the computational oper-ations thus giving better rate of convergence However a
Table 8 Experiment 2 119898 = 7000 120582 = 05 Δ119909 = 143 times 10minus4 Δ119905 =
102 times 10minus8 119905 = 510 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 07) 567119890 minus 13 321119890 minus 12 130119890 minus 10 150 131119890 minus 1
IADEMF2(119903 = 07) 199119890 minus 11 460119890 minus 11 182119890 minus 9 200 127119890 minus 1
SOR4(120596 = 115) 245119890 minus 10 273119890 minus 10 399119890 minus 10 260 151119890 minus 1
GS4 398119890 minus 10 443119890 minus 10 661119890 minus 10 400 171119890 minus 1
JAC4 105119890 minus 9 117119890 minus 9 174119890 minus 9 550 176119890 minus 1
Table 9 Experiment 2 119898 = 7000 120582 = 10 Δ119909 = 143 times 10minus4 Δ119905 =
204 times 10minus8 119905 = 51 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 154119890 minus 6 171119890 minus 6 255119890 minus 6 96 912119890 minus 2
IADEMF2(119903 = 10) 154119890 minus 6 171119890 minus 6 255119890 minus 6 96 607119890 minus 2
SOR4(120596 = 12) 154119890 minus 6 171119890 minus 6 255119890 minus 6 192 997119890 minus 2
GS4 154119890 minus 6 171119890 minus 6 255119890 minus 6 288 128119890 minus 1
JAC4 154119890 minus 6 172119890 minus 6 256119890 minus 6 408 146119890 minus 1
more accurate solution is obtained by using finer mesh Forexample Table 6 shows that for 120582 = 05 the IADEMF4 hasRMSE = 433119890 minus 10 and 119899 = 173 whereas Table 7 shows thatfor 120582 = 10 its RMSE = 163119890 minus 7 and 119899 = 112 In generalamongst all the testedmethods the IADEMF4 still maintainsits greater accuracy characteristic even with coarser meshes
In terms of execution time the results from every tabledisplay shorter execution time for the IADEMF2 in compari-son to the IADEMF4 This is expected since the IADEMF2has lower computational complexity (Table 1) Despite theachievement in accuracy the IADEMF4 has to performmorecomputational work than the IADEMF2 since the former hasto utilize values of 119906 at two grid points on either side of thepoint (119894Δ119909 119896Δ119905) along the 119896th time level Thus if accuracyis desired then the preferred sequential numerical algorithmwould be the IADEMF4 On the other hand if execution timematters then the choice would be the IADEMF2
Amongst the fourth-order methods the IADEMF4 exe-cutes in the least amount of time Even though its computa-tional complexity is relatively large it operates with the leastnumber of iterations thus enabling it to be the most efficient
10 Conclusion
This paper proposes the development of the novel fourth-order IADEMF4 finite difference scheme The higher-orderscheme is developed by representing the spatial derivative inthe heat equation with the fourth-order finite difference CNapproximation This leads to the formation of pentadiagonalmatrices in the systems of linear equations with larger com-putational stencils The algorithm also employs the higheraccuracy of the Mitchell and Fairweather variant Despite
12 Journal of Applied Mathematics
the fourth-order IADE schemersquos higher computational com-plexity this technique is proved to be valuable becauseit enhances the accuracy of its second-order counterpartnamely the IADEMF2 In addition the higher accuracy ofIADEMF4 is verified as a convergent and unconditionallystable scheme and is superior in terms of rate of convergenceIt has also been proven to execute more efficiently in compar-ison to the other benchmarked fourth-order classical iterativemethods such as the GS4 the SOR4 and the JAC4 The in-creasing number of correct digits at each iteration serves as anadvantage for the IADEMF4 thus yielding faster rate of con-vergence with higher level of accuracy
In conclusion the proposed IADEMF4 scheme affordsusers many advantages with respect to higher accuracy sta-bility and rate of convergence and it serves as an alternativeefficient technique for the solution of a one-dimensional heatequation with Dirichlet boundary conditions
The proposed fourth-order scheme can be modified andadapted to more general multidimensional linear and non-linear parabolic elliptic and hyperbolic partial differentialequations using different types of boundary conditionsBesides it can also be considered for applications in problemsthat require higher-order accuracy with high resolution suchas problems in nanocomputing that require the solution ofvery large sparse systems of equation
The explicit and high accuracy feature of the IADEMF4can be exploited in two- and three-dimensional heat prob-lems by applying the scheme as pentadiagonal solvers forthe system of linear equations arising in the sweeps of ahigher-order alternating direction implicit (ADI) schemeThe lower-order ADI scheme was initially proposed byPeaceman and Rachford [24] The higher space dimensionsare expected not to cause unsatisfactory performance for theADI-IADEMF4 as long as the proposed scheme is stableconvergent and efficient as the process of its implementa-tion continues to advance from one-time step to anotherPathirana et al [25] implemented the ADI in developing atwo-dimensional model for incorporating flood damage inurban drainage planningThe proposed model is found to bestable numerically accurate and computationally efficientMirzavand et al [26] proposed the ADI-FDTD methodfor the physical modeling of high-frequency semiconductordevices The approach is able to reduce significantly the full-wave simulation time
It is possible to parallelize the IADEMF4 algorithm sincethe calculation of the new iteration only depends on theknown values from the last iteration (Figures 2 and 3) Futurework is to exploit the explicit computational properties andefficiency of the IADEMF4 for parallelization and executionon distributed parallel computing systems The idea is tospeed up the execution time without compromising its accu-racy especially on problems involving very large linear sys-tems of equations
References
[1] J Noye Computational Techniques for Differential Equationsvol 83 Elsevier Science North-Holland Publishing Amster-dam The Netherlands 1984
[2] G W Recktenwald ldquoFinite difference approximations to theheat equationrdquoMechanical Engineering vol 10 pp 1ndash27 2004
[3] P Lax and B Wendroff ldquoSystems of conservation lawsrdquo Com-munications on Pure and Applied Mathematics vol 13 pp 217ndash237 1960
[4] R J MacKinnon and G F Carey ldquoAnalysis of material interfacediscontinuities and superconvergent fluxes in finite differencetheoryrdquo Journal of Computational Physics vol 75 no 1 pp 151ndash167 1988
[5] R Hixon and E Turkel High Accuracy Compact MacCormack-Type Schemes For Computational Aerocoustics NASACentre forAerospace Information Lewis Research Centre 1998
[6] Y Lin Higher-order finite difference methods for solving heatequations [PhD thesis] Southern Illinois University 2008
[7] J Sulaiman M K Hasan M Othman M and S A AbdulKarim ldquoFourth-order QSMOR iterative method for the solu-tion of one-dimensional parabolic PDErsquosrdquo in Proceedings of theInternational Conference on Industrial and AppliedMathematics(CIAM rsquo10) pp 34ndash39 ITB Bandung Indonesia 2010
[8] N Jha ldquoThe application of sixth order accurate parallel quar-ter sweep alternating group explicit algorithm for nonlinearboundary value problems with singularityrdquo in Proceedings ofthe 2nd International Conference on Methods and Models inComputer Science (ICM2CS rsquo10) pp 76ndash80 December 2010
[9] Y Jin G Jin and J Li ldquoA class of high-precision finite differenceparallel algorithms for convection equationsrdquo Journal of Conver-gence Information Technology vol 6 no 1 pp 79ndash82 2011
[10] W Fu and E L Tan ldquoDevelopment of split-step FDTDmethodwith higher-order spatial accuracyrdquo Electronics Letters vol 40no 20 pp 1252ndash1253 2004
[11] A Mohebbi andM Dehghan ldquoHigh-order compact solution ofthe one-dimensional heat and advection-diffusion equationsrdquoApplied Mathematical Modelling vol 34 no 10 pp 3071ndash30842010
[12] W Liao ldquoAn implicit fourth-order compact finite differencescheme for one-dimensional Burgersrsquo equationrdquo Applied Math-ematics and Computation vol 206 no 2 pp 755ndash764 2008
[13] Z F Tian and Y B Ge ldquoA fourth-order compact ADI methodfor solving two-dimensional unsteady convection-diffusionproblemsrdquo Journal of Computational and Applied Mathematicsvol 198 no 1 pp 268ndash286 2007
[14] C Chun ldquoSome fourth-order iterativemethods for solving non-linear equationsrdquo Applied Mathematics and Computation vol195 no 2 pp 454ndash459 2008
[15] S Zhu Z Yu and J Zhao ldquoAhigh-order parallel finite differencealgorithmrdquo Applied Mathematics and Computation vol 183 no1 pp 365ndash372 2006
[16] Z Gao and S Xie ldquoFourth-order alternating direction implicitcompact finite difference schemes for two-dimensionalSchrodinger equationsrdquo Applied Numerical Mathematics vol61 no 4 pp 593ndash614 2011
[17] M S Sahimi A Ahmad and A A Bakar ldquoThe Iterative Alter-nating Decomposition Explicit (IADE) method to solve theheat conduction equationrdquo International Journal of ComputerMathematics vol 47 pp 219ndash229 1993
[18] G D Smith Numerical Solution of Partial Differential Equa-tions Finite Difference Methods Oxford University Press 2ndedition 1978
[19] A RMitchell andG Fairweather ldquoImproved forms of the alter-nating direction methods of Douglas Peaceman and Rachfordfor solving parabolic and elliptic equationsrdquo Numerische Math-ematik vol 6 no 1 pp 285ndash292 1964
Journal of Applied Mathematics 13
[20] B N DattaNumerical Linear Algebra and Applications BrooksCole Publishing Pacific Grove Calif USA 1995
[21] D M Young Iterative Solution of Large Linear Systems Aca-demic Press New York NY USA 1971
[22] V K Saulev Integration of Equations of Parabolic Type by theMethod of Nets Pergamon Press Oxford UK 1964
[23] L W Johnson and R D Riess Numerical Analysis Addison-Wesley Reading Mass USA 2nd edition 1982
[24] D W Peaceman and H H Rachford Jr ldquoThe numerical solu-tion of parabolic and elliptic differential equationsrdquo Journal ofthe Society for Industrial and Applied Mathematics vol 3 pp28ndash41 1955
[25] A Pathirana S Tsegaye B Gersonius and K VairavamoorthyldquoA simple 2-D inundation model for incorporating flood dam-age in urban drainage planningrdquo Hydrology and Earth SystemSciences vol 15 no 8 pp 2747ndash2761 2011
[26] R Mirzavand A Abdipour G Moradi and M MovahhedildquoFull-wave semiconductor devices simulation using ADI-FDTD methodrdquo Progress in Electromagnetics Research vol 11pp 191ndash202 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
8 Journal of Applied Mathematics
begindetermine the parameters119898 119903 Δ119905 Δ119909 120582 120576 119879
determine initial conditions 119906(119901)119894
determine exact solutions 119880119894at time level 119879
while (time level lt 119879)determine boundary conditionsfor 119894 = 2 to 119894 = 119898 minus 1
compute 119891119894(refer to (9))
end-forset iteration = 0while (convergence conditions are not satisfied)
for 119894 = 2 to 119894 = 119898 minus 1
compute 119906(119901+12)119894
(refer to (22))end-forfor 119894 = 2 to 119894 = 119898 minus 1
compute 119906(119901+1)119894
(refer to (23))end-fortest for convergencecompute 119890
119894larr
10038161003816100381610038161003816119906(119901+1)
119894minus 119906(119901)
119894
10038161003816100381610038161003816if (max1003816100381610038161003816119890119894
1003816100381610038161003816 lt 120576)then 119906
(119901)
119894larr 119906(119901+1)
119894
add 1 to iteration (if necessary)end-while
end-whilefor 119894 = 2 to 119894 = 119898 minus 1
determine average absolute error root meansquare error and maximum error
end-forend
Algorithm 1 The IADEMF4 sequential algorithm
8 Numerical Experiments
Two experiments were conducted to test the sequentialnumerical performance of the proposed IADEMF4 methodagainst those of the benchmarked classical iterative methodsnamely the JAC4 theGS4 and the SOR4 Comparison is alsomade with the corresponding IADE method of the secondorder
Experiment 1 This problem was taken from Saulev [22]
120597119880
120597119905=1205972
119880
1205971199092 0 le 119909 le 1 (44)
subject to the initial condition
119880 (119909 0) = 4119909 (1 minus 119909) 0 le 119909 le 1 (45)
and the boundary conditions
119880 (0 119905) = 119880 (1 119905) = 0 119905 ge 0 (46)
The exact solution to the given problem is given by
119880 (119909 119905) =32
1205873
infin
sum
119896=1(2)
1
1198963119890minus12058721198962119905 sin (119896120587119909) (47)
Experiment 2 This problem was taken from Johnson andReiss [23]
120597119880
120597119905=1205972
119880
1205971199092 0 le 119909 le 1 (48)
subject to the initial condition
119880 (119909 0) = sin (120587119909) (1 + 6 cos (120587119909)) 0 le 119909 le 1 (49)
and the boundary conditions
119880 (0 119905) = 119880 (1 119905) = 0 119905 ge 0 (50)
The exact solution to the given problem is given by
119880 (119909 119905) = sin (120587119909) 119890minus1205872119905
+ 3 sin (2120587119909) 119890minus41205872119905
(51)
Journal of Applied Mathematics 9
9 Results and Discussion
In each experiment the exact solutions at 119894 = 1 and 119894 = 119898
were taken as boundary values for the IADEMF4 and theother fourth-order classical iterative methods The conver-gence criterion used in the testing of each method was takenas 119906(119901+1) minus 119906
(119901)
infin
le 120576 where 120576 is the convergence toleranceThe selections of the optimum 119903 or 120596 were determined byexperiments As for the IADEMF2 the CN scheme with 120579 =
12 was employedFigures 6 7 8 and 9 visualize the behavior of the one-
dimensional parabolic heat solutions for both experimentsBy using119898 = 10 and a tolerance requirement of 120576 = 10
minus4 twodifferent mesh sizes 120582 = 05 and 120582 = 10 were consideredin each experiment The exact solutions are compared withthe IADEMF4 and the IADEMF2 numerical solutions Theoptimum value of 119903 chosen for each method as well as theircorresponding outcome of the number of iterations 119899 isstated in the legend of each figure Every figure reveals thatthe IADEMF4 is more accurate than the IADEMF2 andthe former converges with fewer numbers of iterations incomparison to the latter The numerical solutions of theIADEMF4 seem to be in very good agreement with theexact solutions For example in Figure 7 at 119909 = 05 thedifference between the exact solution and the numericalsolution using the IADEMF2 and using the IADEMF4 isabout 36 and 0 respectively These results imply thatthe accuracy and convergence rate of the second-order IADEmethod are enhanced by the implementation of the fourth-order CN approximation that leads to the formation of thecorresponding fourth-order IADE scheme
Figure 10 displays the graph of log (RMSE) versus log(Δ119909)for decreasing values of Δ119909 implemented on Experiment 2By considering 120576 = 10
minus4 and fixing Δ119905 = 00001 the value ofΔ119909 was initially taken as Δ119909 = 0125 It was then successivelyhalved into Δ1199092 Δ1199094 and Δ1199098
The figure shows that amongst the tested methods theroot mean square error (RMSE) of the IADEMF4 and theIADEMF2 decreases linearly as the values of Δ119909 decreasesThe slope of the IADEMF4 is approximately equal to 4which corresponds to its fourth-order spatial accuracy TheIADEMF2 with a second-order spatial accuracy has a slopethat is approximately equal to 2 It is clear that the IADEMF4is always more accurate than the IADEMF2 for the differentconsidered mesh sizes The graphs of the SOR4 (120596 = 105)the GS4 and the JAC4 show that their accuracies of fourthorder tend to lack as Δ119909 decreases largely due to the effect ofincreasing round-off errors as the value of119898 increases
Tables 2 3 4 5 6 7 8 and 9 provide numerical results interms of the average absolute error (AAE) root mean squareerror (RMSE) maximum error (ME) number of iterations(119899) and execution time (ET) measured in seconds (s) Theresults are obtained from both experiments for two differentvalues of 119898 and mesh size 120582 It is generally observed thatwhen 119898 = 700 120582 = 05 and 120576 = 10
minus6 the IADEMF4has the least average absolute error root mean square errorand maximum error in comparison with the other methodsunder consideration (Tables 2ndash5) When the size of 119898 was
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 08 n = 172)IADEMF2 (r = 03 n = 185)
00E + 00
10E minus 02
20E minus 02
30E minus 02
40E minus 02
50E minus 02
60E minus 02
70E minus 02
80E minus 02
90E minus 02
10E minus 01
U(xt)
Figure 6 Numerical and exact solutions for Experiment 1 120582 = 05Δ119909 = 01 Δ119905 = 0005 119905 = 025 and 120576 = 10
minus4
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 14 n = 159)IADEMF2 (r = 05 n = 162)
00E + 00
10E minus 03
20E minus 03
30E minus 03
40E minus 03
50E minus 03
60E minus 03
70E minus 03
80E minus 03
90E minus 03
U(xt)
Figure 7 Numerical and exact solutions for Experiment 1 120582 = 10Δ119909 = 01 Δ119905 = 001 119905 = 05 and 120576 = 10
minus4
ten times bigger (119898 = 7000) and a more stringent tolerancecriterion was set (120576 = 10
minus10
) the accuracy of the IADEMF4clearly outperforms the other methods for a mesh size of120582 = 05 (Tables 6 and 8) However the difference in the errorsamongst the tested methods is not so obvious for the case of120582 = 10 (Tables 7 and 9)The achievement of the fourth-orderIADEmethod in Experiments 1 and 2 can be clearly seen fromthe results in Tables 2 and 8 respectively where it has caused ahuge 94 reduction of RMSE from its corresponding second-order IADE method
The IADEMF4 has the advantage of featuring higheraccuracy due to the execution of the fourth-order CNapprox-imation coupled with the fourth-order accurate MF variant
10 Journal of Applied Mathematics
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 10 n = 169)IADEMF2 (r = 03 n = 213)
00E + 00
10E minus 02
20E minus 02
30E minus 02
40E minus 02
50E minus 02
60E minus 02
70E minus 02
80E minus 02
90E minus 02
10E minus 01
U(xt)
Figure 8 Numerical and exact solutions for Experiment 2 120582 = 05Δ119909 = 01 Δ119905 = 0005 119905 = 025 and 120576 = 10
minus4
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 10 n = 175)IADEMF2 (r = 05 n = 182)
00E + 00
10E minus 03
20E minus 03
30E minus 03
40E minus 03
50E minus 03
60E minus 03
70E minus 03
80E minus 03
U(xt)
Figure 9 Numerical and exact solutions for Experiment 2 120582 = 10Δ119909 = 01 Δ119905 = 001 119905 = 05 and 120576 = 10
minus4
The IADEMF2 is only derived from the second-order CNapproximation but its combination with its correspondingfourth-order MF variant managed to produce errors whichare better than theGS4 JAC4 and the SOR4 Even though theclassical iterative methods are also derived from the fourth-order accurate CN type approximation they are lackingin accuracy due to the round-off errors that have beenaccumulated from the time the execution starts till it ends
With regards to the rate of convergence the resultsfrom each table demonstrate that the number of iterationsproduced by the IADEMF2 and the fourth-order classicaliterative methods is greater than or at least equal to that of
0
minus2
minus208 minus277 minus347 minus416
minus4
minus8
minus6
minus10
minus12
minus14
minus16
minus18
GS4JAC4SOR4
log
(RM
SE)
log(Δx)
IADEMF4
IADEMF2
Figure 10 Log(RMSE) versus Log(Δ119909) Experiment 2 Δ119905 = 00001119905 = 0002 and 120576 = 10
minus4
Table 2 Experiment 1 119898 = 700 120582 = 05 Δ119909 = 143 times 10minus3 Δ119905 =
102 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 07) 267119890 minus 9 330119890 minus 9 212119890 minus 8 100 440119890 minus 2
IADEMF2(119903 = 08) 188119890 minus 8 495119890 minus 8 323119890 minus 7 100 635119890 minus 3
SOR4(120596 = 12) 540119890 minus 7 544119890 minus 7 552119890 minus 7 108 460119890 minus 2
GS4 540119890 minus 6 543119890 minus 6 332119890 minus 6 150 472119890 minus 2
JAC4 228119890 minus 5 229119890 minus 5 232119890 minus 5 150 473119890 minus 2
Table 3 Experiment 1 119898 = 700 120582 = 10 Δ119909 = 143 times 10minus3 Δ119905 =
204 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 297119890 minus 7 301119890 minus 7 301119890 minus 7 50 201119890 minus 2
IADEMF2(119903 = 09) 557119890 minus 7 559119890 minus 7 848119890 minus 7 50 317119890 minus 3
SOR4(120596 = 12) 329119890 minus 6 331119890 minus 6 335119890 minus 6 76 225119890 minus 2
GS4 895119890 minus 6 900119890 minus 6 900119890 minus 6 100 238119890 minus 2
JAC4 214119890 minus 5 215119890 minus 5 217119890 minus 5 125 250119890 minus 2
the IADEMF4 The convergence rate of the latter surpassesthe others in the case of 119898 = 7000 and 120576 = 10
minus10 (Tables6ndash8) Even though the operational count of the IADEMF4is relatively quite large (Table 1) due to its higher level ofaccuracy its increasing number of correct digits at eachiteration causes it to converge at a faster rate The IADEMF2has the advantage of having less mathematical operationsthus it is competitive in terms of convergence rate but atthe expense of accuracy The application of the fourth-orderCN approximation on the heat equation proves that thethree classical iterative methods also converge with JAC4appearing to be the slowest and the least accurate amongst all
Journal of Applied Mathematics 11
Table 4 Experiment 2 119898 = 700 120582 = 05 Δ119909 = 143 times 10minus3 Δ119905 =
102 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME n ET (s)IADEMF4(119903 = 07) 207119890 minus 8 231119890 minus 8 345119890 minus 8 100 903119890 minus 3
IADEMF2(119903 = 08) 106119890 minus 7 175119890 minus 7 175119890 minus 7 100 647119890 minus 3
SOR4(120596 = 11) 154119890 minus 6 171119890 minus 6 255119890 minus 6 200 110119890 minus 2
GS4 294119890 minus 6 327119890 minus 6 487119890 minus 6 250 118119890 minus 2
JAC4 124119890 minus 5 138119890 minus 5 206119890 minus 5 300 127119890 minus 2
Table 5 Experiment 2 119898 = 700 120582 = 10 Δ119909 = 143 times 10minus3 Δ119905 =
204 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 09) 138119890 minus 6 154119890 minus 6 229119890 minus 6 75 650119890 minus 3
IADEMF2(119903 = 05) 187119890 minus 6 207119890 minus 6 309119890 minus 6 75 472119890 minus 3
SOR4(120596 = 12) 312119890 minus 6 347119890 minus 6 517119890 minus 6 125 906119890 minus 3
GS4 667119890 minus 6 742119890 minus 6 111119890 minus 5 175 932119890 minus 3
JAC4 126119890 minus 5 141119890 minus 5 210119890 minus 5 250 105119890 minus 2
Table 6 Experiment 1 119898 = 7000 120582 = 05 Δ119909 = 143 times 10minus4 Δ119905 =
102 times 10minus8 119905 = 510 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME n ET (s)IADEMF4(119903 = 08) 367119890 minus 10 433119890 minus 10 366119890 minus 9 173 146119890 minus 1
IADEMF2(119903 = 05) 375119890 minus 10 435119890 minus 10 376119890 minus 9 200 124119890 minus 1
SOR4(120596 = 112) 455119890 minus 10 507119890 minus 10 367119890 minus 9 257 147119890 minus 1
GS4 110119890 minus 9 111119890 minus 9 376119890 minus 9 300 160119890 minus 1
JAC4 230119890 minus 9 231119890 minus 9 378119890 minus 9 400 163119890 minus 1
Table 7 Experiment 1 119898 = 7000 120582 = 10 Δ119909 = 143 times 10minus4 Δ119905 =
204 times 10minus8 119905 = 51 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 163119890 minus 7 163119890 minus 7 163119890 minus 7 112 960119890 minus 2
IADEMF2(119903 = 12) 163119890 minus 7 163119890 minus 7 163119890 minus 7 120 736119890 minus 2
SOR4(120596 = 115) 163119890 minus 7 163119890 minus 7 164119890 minus 7 171 999119890 minus 2
GS4 164119890 minus 7 164119890 minus 7 164119890 minus 7 216 107119890 minus 1
JAC4 165119890 minus 7 165119890 minus 7 165119890 minus 7 312 111119890 minus 1
It is found that the convergence rate of the fourth-orderand second-order methods improves with the application oflarger mesh size that is from 120582 = 05 to 120582 = 10 Thecoarser meshes cause a reduction in the computational oper-ations thus giving better rate of convergence However a
Table 8 Experiment 2 119898 = 7000 120582 = 05 Δ119909 = 143 times 10minus4 Δ119905 =
102 times 10minus8 119905 = 510 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 07) 567119890 minus 13 321119890 minus 12 130119890 minus 10 150 131119890 minus 1
IADEMF2(119903 = 07) 199119890 minus 11 460119890 minus 11 182119890 minus 9 200 127119890 minus 1
SOR4(120596 = 115) 245119890 minus 10 273119890 minus 10 399119890 minus 10 260 151119890 minus 1
GS4 398119890 minus 10 443119890 minus 10 661119890 minus 10 400 171119890 minus 1
JAC4 105119890 minus 9 117119890 minus 9 174119890 minus 9 550 176119890 minus 1
Table 9 Experiment 2 119898 = 7000 120582 = 10 Δ119909 = 143 times 10minus4 Δ119905 =
204 times 10minus8 119905 = 51 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 154119890 minus 6 171119890 minus 6 255119890 minus 6 96 912119890 minus 2
IADEMF2(119903 = 10) 154119890 minus 6 171119890 minus 6 255119890 minus 6 96 607119890 minus 2
SOR4(120596 = 12) 154119890 minus 6 171119890 minus 6 255119890 minus 6 192 997119890 minus 2
GS4 154119890 minus 6 171119890 minus 6 255119890 minus 6 288 128119890 minus 1
JAC4 154119890 minus 6 172119890 minus 6 256119890 minus 6 408 146119890 minus 1
more accurate solution is obtained by using finer mesh Forexample Table 6 shows that for 120582 = 05 the IADEMF4 hasRMSE = 433119890 minus 10 and 119899 = 173 whereas Table 7 shows thatfor 120582 = 10 its RMSE = 163119890 minus 7 and 119899 = 112 In generalamongst all the testedmethods the IADEMF4 still maintainsits greater accuracy characteristic even with coarser meshes
In terms of execution time the results from every tabledisplay shorter execution time for the IADEMF2 in compari-son to the IADEMF4 This is expected since the IADEMF2has lower computational complexity (Table 1) Despite theachievement in accuracy the IADEMF4 has to performmorecomputational work than the IADEMF2 since the former hasto utilize values of 119906 at two grid points on either side of thepoint (119894Δ119909 119896Δ119905) along the 119896th time level Thus if accuracyis desired then the preferred sequential numerical algorithmwould be the IADEMF4 On the other hand if execution timematters then the choice would be the IADEMF2
Amongst the fourth-order methods the IADEMF4 exe-cutes in the least amount of time Even though its computa-tional complexity is relatively large it operates with the leastnumber of iterations thus enabling it to be the most efficient
10 Conclusion
This paper proposes the development of the novel fourth-order IADEMF4 finite difference scheme The higher-orderscheme is developed by representing the spatial derivative inthe heat equation with the fourth-order finite difference CNapproximation This leads to the formation of pentadiagonalmatrices in the systems of linear equations with larger com-putational stencils The algorithm also employs the higheraccuracy of the Mitchell and Fairweather variant Despite
12 Journal of Applied Mathematics
the fourth-order IADE schemersquos higher computational com-plexity this technique is proved to be valuable becauseit enhances the accuracy of its second-order counterpartnamely the IADEMF2 In addition the higher accuracy ofIADEMF4 is verified as a convergent and unconditionallystable scheme and is superior in terms of rate of convergenceIt has also been proven to execute more efficiently in compar-ison to the other benchmarked fourth-order classical iterativemethods such as the GS4 the SOR4 and the JAC4 The in-creasing number of correct digits at each iteration serves as anadvantage for the IADEMF4 thus yielding faster rate of con-vergence with higher level of accuracy
In conclusion the proposed IADEMF4 scheme affordsusers many advantages with respect to higher accuracy sta-bility and rate of convergence and it serves as an alternativeefficient technique for the solution of a one-dimensional heatequation with Dirichlet boundary conditions
The proposed fourth-order scheme can be modified andadapted to more general multidimensional linear and non-linear parabolic elliptic and hyperbolic partial differentialequations using different types of boundary conditionsBesides it can also be considered for applications in problemsthat require higher-order accuracy with high resolution suchas problems in nanocomputing that require the solution ofvery large sparse systems of equation
The explicit and high accuracy feature of the IADEMF4can be exploited in two- and three-dimensional heat prob-lems by applying the scheme as pentadiagonal solvers forthe system of linear equations arising in the sweeps of ahigher-order alternating direction implicit (ADI) schemeThe lower-order ADI scheme was initially proposed byPeaceman and Rachford [24] The higher space dimensionsare expected not to cause unsatisfactory performance for theADI-IADEMF4 as long as the proposed scheme is stableconvergent and efficient as the process of its implementa-tion continues to advance from one-time step to anotherPathirana et al [25] implemented the ADI in developing atwo-dimensional model for incorporating flood damage inurban drainage planningThe proposed model is found to bestable numerically accurate and computationally efficientMirzavand et al [26] proposed the ADI-FDTD methodfor the physical modeling of high-frequency semiconductordevices The approach is able to reduce significantly the full-wave simulation time
It is possible to parallelize the IADEMF4 algorithm sincethe calculation of the new iteration only depends on theknown values from the last iteration (Figures 2 and 3) Futurework is to exploit the explicit computational properties andefficiency of the IADEMF4 for parallelization and executionon distributed parallel computing systems The idea is tospeed up the execution time without compromising its accu-racy especially on problems involving very large linear sys-tems of equations
References
[1] J Noye Computational Techniques for Differential Equationsvol 83 Elsevier Science North-Holland Publishing Amster-dam The Netherlands 1984
[2] G W Recktenwald ldquoFinite difference approximations to theheat equationrdquoMechanical Engineering vol 10 pp 1ndash27 2004
[3] P Lax and B Wendroff ldquoSystems of conservation lawsrdquo Com-munications on Pure and Applied Mathematics vol 13 pp 217ndash237 1960
[4] R J MacKinnon and G F Carey ldquoAnalysis of material interfacediscontinuities and superconvergent fluxes in finite differencetheoryrdquo Journal of Computational Physics vol 75 no 1 pp 151ndash167 1988
[5] R Hixon and E Turkel High Accuracy Compact MacCormack-Type Schemes For Computational Aerocoustics NASACentre forAerospace Information Lewis Research Centre 1998
[6] Y Lin Higher-order finite difference methods for solving heatequations [PhD thesis] Southern Illinois University 2008
[7] J Sulaiman M K Hasan M Othman M and S A AbdulKarim ldquoFourth-order QSMOR iterative method for the solu-tion of one-dimensional parabolic PDErsquosrdquo in Proceedings of theInternational Conference on Industrial and AppliedMathematics(CIAM rsquo10) pp 34ndash39 ITB Bandung Indonesia 2010
[8] N Jha ldquoThe application of sixth order accurate parallel quar-ter sweep alternating group explicit algorithm for nonlinearboundary value problems with singularityrdquo in Proceedings ofthe 2nd International Conference on Methods and Models inComputer Science (ICM2CS rsquo10) pp 76ndash80 December 2010
[9] Y Jin G Jin and J Li ldquoA class of high-precision finite differenceparallel algorithms for convection equationsrdquo Journal of Conver-gence Information Technology vol 6 no 1 pp 79ndash82 2011
[10] W Fu and E L Tan ldquoDevelopment of split-step FDTDmethodwith higher-order spatial accuracyrdquo Electronics Letters vol 40no 20 pp 1252ndash1253 2004
[11] A Mohebbi andM Dehghan ldquoHigh-order compact solution ofthe one-dimensional heat and advection-diffusion equationsrdquoApplied Mathematical Modelling vol 34 no 10 pp 3071ndash30842010
[12] W Liao ldquoAn implicit fourth-order compact finite differencescheme for one-dimensional Burgersrsquo equationrdquo Applied Math-ematics and Computation vol 206 no 2 pp 755ndash764 2008
[13] Z F Tian and Y B Ge ldquoA fourth-order compact ADI methodfor solving two-dimensional unsteady convection-diffusionproblemsrdquo Journal of Computational and Applied Mathematicsvol 198 no 1 pp 268ndash286 2007
[14] C Chun ldquoSome fourth-order iterativemethods for solving non-linear equationsrdquo Applied Mathematics and Computation vol195 no 2 pp 454ndash459 2008
[15] S Zhu Z Yu and J Zhao ldquoAhigh-order parallel finite differencealgorithmrdquo Applied Mathematics and Computation vol 183 no1 pp 365ndash372 2006
[16] Z Gao and S Xie ldquoFourth-order alternating direction implicitcompact finite difference schemes for two-dimensionalSchrodinger equationsrdquo Applied Numerical Mathematics vol61 no 4 pp 593ndash614 2011
[17] M S Sahimi A Ahmad and A A Bakar ldquoThe Iterative Alter-nating Decomposition Explicit (IADE) method to solve theheat conduction equationrdquo International Journal of ComputerMathematics vol 47 pp 219ndash229 1993
[18] G D Smith Numerical Solution of Partial Differential Equa-tions Finite Difference Methods Oxford University Press 2ndedition 1978
[19] A RMitchell andG Fairweather ldquoImproved forms of the alter-nating direction methods of Douglas Peaceman and Rachfordfor solving parabolic and elliptic equationsrdquo Numerische Math-ematik vol 6 no 1 pp 285ndash292 1964
Journal of Applied Mathematics 13
[20] B N DattaNumerical Linear Algebra and Applications BrooksCole Publishing Pacific Grove Calif USA 1995
[21] D M Young Iterative Solution of Large Linear Systems Aca-demic Press New York NY USA 1971
[22] V K Saulev Integration of Equations of Parabolic Type by theMethod of Nets Pergamon Press Oxford UK 1964
[23] L W Johnson and R D Riess Numerical Analysis Addison-Wesley Reading Mass USA 2nd edition 1982
[24] D W Peaceman and H H Rachford Jr ldquoThe numerical solu-tion of parabolic and elliptic differential equationsrdquo Journal ofthe Society for Industrial and Applied Mathematics vol 3 pp28ndash41 1955
[25] A Pathirana S Tsegaye B Gersonius and K VairavamoorthyldquoA simple 2-D inundation model for incorporating flood dam-age in urban drainage planningrdquo Hydrology and Earth SystemSciences vol 15 no 8 pp 2747ndash2761 2011
[26] R Mirzavand A Abdipour G Moradi and M MovahhedildquoFull-wave semiconductor devices simulation using ADI-FDTD methodrdquo Progress in Electromagnetics Research vol 11pp 191ndash202 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Journal of Applied Mathematics 9
9 Results and Discussion
In each experiment the exact solutions at 119894 = 1 and 119894 = 119898
were taken as boundary values for the IADEMF4 and theother fourth-order classical iterative methods The conver-gence criterion used in the testing of each method was takenas 119906(119901+1) minus 119906
(119901)
infin
le 120576 where 120576 is the convergence toleranceThe selections of the optimum 119903 or 120596 were determined byexperiments As for the IADEMF2 the CN scheme with 120579 =
12 was employedFigures 6 7 8 and 9 visualize the behavior of the one-
dimensional parabolic heat solutions for both experimentsBy using119898 = 10 and a tolerance requirement of 120576 = 10
minus4 twodifferent mesh sizes 120582 = 05 and 120582 = 10 were consideredin each experiment The exact solutions are compared withthe IADEMF4 and the IADEMF2 numerical solutions Theoptimum value of 119903 chosen for each method as well as theircorresponding outcome of the number of iterations 119899 isstated in the legend of each figure Every figure reveals thatthe IADEMF4 is more accurate than the IADEMF2 andthe former converges with fewer numbers of iterations incomparison to the latter The numerical solutions of theIADEMF4 seem to be in very good agreement with theexact solutions For example in Figure 7 at 119909 = 05 thedifference between the exact solution and the numericalsolution using the IADEMF2 and using the IADEMF4 isabout 36 and 0 respectively These results imply thatthe accuracy and convergence rate of the second-order IADEmethod are enhanced by the implementation of the fourth-order CN approximation that leads to the formation of thecorresponding fourth-order IADE scheme
Figure 10 displays the graph of log (RMSE) versus log(Δ119909)for decreasing values of Δ119909 implemented on Experiment 2By considering 120576 = 10
minus4 and fixing Δ119905 = 00001 the value ofΔ119909 was initially taken as Δ119909 = 0125 It was then successivelyhalved into Δ1199092 Δ1199094 and Δ1199098
The figure shows that amongst the tested methods theroot mean square error (RMSE) of the IADEMF4 and theIADEMF2 decreases linearly as the values of Δ119909 decreasesThe slope of the IADEMF4 is approximately equal to 4which corresponds to its fourth-order spatial accuracy TheIADEMF2 with a second-order spatial accuracy has a slopethat is approximately equal to 2 It is clear that the IADEMF4is always more accurate than the IADEMF2 for the differentconsidered mesh sizes The graphs of the SOR4 (120596 = 105)the GS4 and the JAC4 show that their accuracies of fourthorder tend to lack as Δ119909 decreases largely due to the effect ofincreasing round-off errors as the value of119898 increases
Tables 2 3 4 5 6 7 8 and 9 provide numerical results interms of the average absolute error (AAE) root mean squareerror (RMSE) maximum error (ME) number of iterations(119899) and execution time (ET) measured in seconds (s) Theresults are obtained from both experiments for two differentvalues of 119898 and mesh size 120582 It is generally observed thatwhen 119898 = 700 120582 = 05 and 120576 = 10
minus6 the IADEMF4has the least average absolute error root mean square errorand maximum error in comparison with the other methodsunder consideration (Tables 2ndash5) When the size of 119898 was
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 08 n = 172)IADEMF2 (r = 03 n = 185)
00E + 00
10E minus 02
20E minus 02
30E minus 02
40E minus 02
50E minus 02
60E minus 02
70E minus 02
80E minus 02
90E minus 02
10E minus 01
U(xt)
Figure 6 Numerical and exact solutions for Experiment 1 120582 = 05Δ119909 = 01 Δ119905 = 0005 119905 = 025 and 120576 = 10
minus4
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 14 n = 159)IADEMF2 (r = 05 n = 162)
00E + 00
10E minus 03
20E minus 03
30E minus 03
40E minus 03
50E minus 03
60E minus 03
70E minus 03
80E minus 03
90E minus 03
U(xt)
Figure 7 Numerical and exact solutions for Experiment 1 120582 = 10Δ119909 = 01 Δ119905 = 001 119905 = 05 and 120576 = 10
minus4
ten times bigger (119898 = 7000) and a more stringent tolerancecriterion was set (120576 = 10
minus10
) the accuracy of the IADEMF4clearly outperforms the other methods for a mesh size of120582 = 05 (Tables 6 and 8) However the difference in the errorsamongst the tested methods is not so obvious for the case of120582 = 10 (Tables 7 and 9)The achievement of the fourth-orderIADEmethod in Experiments 1 and 2 can be clearly seen fromthe results in Tables 2 and 8 respectively where it has caused ahuge 94 reduction of RMSE from its corresponding second-order IADE method
The IADEMF4 has the advantage of featuring higheraccuracy due to the execution of the fourth-order CNapprox-imation coupled with the fourth-order accurate MF variant
10 Journal of Applied Mathematics
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 10 n = 169)IADEMF2 (r = 03 n = 213)
00E + 00
10E minus 02
20E minus 02
30E minus 02
40E minus 02
50E minus 02
60E minus 02
70E minus 02
80E minus 02
90E minus 02
10E minus 01
U(xt)
Figure 8 Numerical and exact solutions for Experiment 2 120582 = 05Δ119909 = 01 Δ119905 = 0005 119905 = 025 and 120576 = 10
minus4
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 10 n = 175)IADEMF2 (r = 05 n = 182)
00E + 00
10E minus 03
20E minus 03
30E minus 03
40E minus 03
50E minus 03
60E minus 03
70E minus 03
80E minus 03
U(xt)
Figure 9 Numerical and exact solutions for Experiment 2 120582 = 10Δ119909 = 01 Δ119905 = 001 119905 = 05 and 120576 = 10
minus4
The IADEMF2 is only derived from the second-order CNapproximation but its combination with its correspondingfourth-order MF variant managed to produce errors whichare better than theGS4 JAC4 and the SOR4 Even though theclassical iterative methods are also derived from the fourth-order accurate CN type approximation they are lackingin accuracy due to the round-off errors that have beenaccumulated from the time the execution starts till it ends
With regards to the rate of convergence the resultsfrom each table demonstrate that the number of iterationsproduced by the IADEMF2 and the fourth-order classicaliterative methods is greater than or at least equal to that of
0
minus2
minus208 minus277 minus347 minus416
minus4
minus8
minus6
minus10
minus12
minus14
minus16
minus18
GS4JAC4SOR4
log
(RM
SE)
log(Δx)
IADEMF4
IADEMF2
Figure 10 Log(RMSE) versus Log(Δ119909) Experiment 2 Δ119905 = 00001119905 = 0002 and 120576 = 10
minus4
Table 2 Experiment 1 119898 = 700 120582 = 05 Δ119909 = 143 times 10minus3 Δ119905 =
102 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 07) 267119890 minus 9 330119890 minus 9 212119890 minus 8 100 440119890 minus 2
IADEMF2(119903 = 08) 188119890 minus 8 495119890 minus 8 323119890 minus 7 100 635119890 minus 3
SOR4(120596 = 12) 540119890 minus 7 544119890 minus 7 552119890 minus 7 108 460119890 minus 2
GS4 540119890 minus 6 543119890 minus 6 332119890 minus 6 150 472119890 minus 2
JAC4 228119890 minus 5 229119890 minus 5 232119890 minus 5 150 473119890 minus 2
Table 3 Experiment 1 119898 = 700 120582 = 10 Δ119909 = 143 times 10minus3 Δ119905 =
204 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 297119890 minus 7 301119890 minus 7 301119890 minus 7 50 201119890 minus 2
IADEMF2(119903 = 09) 557119890 minus 7 559119890 minus 7 848119890 minus 7 50 317119890 minus 3
SOR4(120596 = 12) 329119890 minus 6 331119890 minus 6 335119890 minus 6 76 225119890 minus 2
GS4 895119890 minus 6 900119890 minus 6 900119890 minus 6 100 238119890 minus 2
JAC4 214119890 minus 5 215119890 minus 5 217119890 minus 5 125 250119890 minus 2
the IADEMF4 The convergence rate of the latter surpassesthe others in the case of 119898 = 7000 and 120576 = 10
minus10 (Tables6ndash8) Even though the operational count of the IADEMF4is relatively quite large (Table 1) due to its higher level ofaccuracy its increasing number of correct digits at eachiteration causes it to converge at a faster rate The IADEMF2has the advantage of having less mathematical operationsthus it is competitive in terms of convergence rate but atthe expense of accuracy The application of the fourth-orderCN approximation on the heat equation proves that thethree classical iterative methods also converge with JAC4appearing to be the slowest and the least accurate amongst all
Journal of Applied Mathematics 11
Table 4 Experiment 2 119898 = 700 120582 = 05 Δ119909 = 143 times 10minus3 Δ119905 =
102 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME n ET (s)IADEMF4(119903 = 07) 207119890 minus 8 231119890 minus 8 345119890 minus 8 100 903119890 minus 3
IADEMF2(119903 = 08) 106119890 minus 7 175119890 minus 7 175119890 minus 7 100 647119890 minus 3
SOR4(120596 = 11) 154119890 minus 6 171119890 minus 6 255119890 minus 6 200 110119890 minus 2
GS4 294119890 minus 6 327119890 minus 6 487119890 minus 6 250 118119890 minus 2
JAC4 124119890 minus 5 138119890 minus 5 206119890 minus 5 300 127119890 minus 2
Table 5 Experiment 2 119898 = 700 120582 = 10 Δ119909 = 143 times 10minus3 Δ119905 =
204 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 09) 138119890 minus 6 154119890 minus 6 229119890 minus 6 75 650119890 minus 3
IADEMF2(119903 = 05) 187119890 minus 6 207119890 minus 6 309119890 minus 6 75 472119890 minus 3
SOR4(120596 = 12) 312119890 minus 6 347119890 minus 6 517119890 minus 6 125 906119890 minus 3
GS4 667119890 minus 6 742119890 minus 6 111119890 minus 5 175 932119890 minus 3
JAC4 126119890 minus 5 141119890 minus 5 210119890 minus 5 250 105119890 minus 2
Table 6 Experiment 1 119898 = 7000 120582 = 05 Δ119909 = 143 times 10minus4 Δ119905 =
102 times 10minus8 119905 = 510 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME n ET (s)IADEMF4(119903 = 08) 367119890 minus 10 433119890 minus 10 366119890 minus 9 173 146119890 minus 1
IADEMF2(119903 = 05) 375119890 minus 10 435119890 minus 10 376119890 minus 9 200 124119890 minus 1
SOR4(120596 = 112) 455119890 minus 10 507119890 minus 10 367119890 minus 9 257 147119890 minus 1
GS4 110119890 minus 9 111119890 minus 9 376119890 minus 9 300 160119890 minus 1
JAC4 230119890 minus 9 231119890 minus 9 378119890 minus 9 400 163119890 minus 1
Table 7 Experiment 1 119898 = 7000 120582 = 10 Δ119909 = 143 times 10minus4 Δ119905 =
204 times 10minus8 119905 = 51 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 163119890 minus 7 163119890 minus 7 163119890 minus 7 112 960119890 minus 2
IADEMF2(119903 = 12) 163119890 minus 7 163119890 minus 7 163119890 minus 7 120 736119890 minus 2
SOR4(120596 = 115) 163119890 minus 7 163119890 minus 7 164119890 minus 7 171 999119890 minus 2
GS4 164119890 minus 7 164119890 minus 7 164119890 minus 7 216 107119890 minus 1
JAC4 165119890 minus 7 165119890 minus 7 165119890 minus 7 312 111119890 minus 1
It is found that the convergence rate of the fourth-orderand second-order methods improves with the application oflarger mesh size that is from 120582 = 05 to 120582 = 10 Thecoarser meshes cause a reduction in the computational oper-ations thus giving better rate of convergence However a
Table 8 Experiment 2 119898 = 7000 120582 = 05 Δ119909 = 143 times 10minus4 Δ119905 =
102 times 10minus8 119905 = 510 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 07) 567119890 minus 13 321119890 minus 12 130119890 minus 10 150 131119890 minus 1
IADEMF2(119903 = 07) 199119890 minus 11 460119890 minus 11 182119890 minus 9 200 127119890 minus 1
SOR4(120596 = 115) 245119890 minus 10 273119890 minus 10 399119890 minus 10 260 151119890 minus 1
GS4 398119890 minus 10 443119890 minus 10 661119890 minus 10 400 171119890 minus 1
JAC4 105119890 minus 9 117119890 minus 9 174119890 minus 9 550 176119890 minus 1
Table 9 Experiment 2 119898 = 7000 120582 = 10 Δ119909 = 143 times 10minus4 Δ119905 =
204 times 10minus8 119905 = 51 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 154119890 minus 6 171119890 minus 6 255119890 minus 6 96 912119890 minus 2
IADEMF2(119903 = 10) 154119890 minus 6 171119890 minus 6 255119890 minus 6 96 607119890 minus 2
SOR4(120596 = 12) 154119890 minus 6 171119890 minus 6 255119890 minus 6 192 997119890 minus 2
GS4 154119890 minus 6 171119890 minus 6 255119890 minus 6 288 128119890 minus 1
JAC4 154119890 minus 6 172119890 minus 6 256119890 minus 6 408 146119890 minus 1
more accurate solution is obtained by using finer mesh Forexample Table 6 shows that for 120582 = 05 the IADEMF4 hasRMSE = 433119890 minus 10 and 119899 = 173 whereas Table 7 shows thatfor 120582 = 10 its RMSE = 163119890 minus 7 and 119899 = 112 In generalamongst all the testedmethods the IADEMF4 still maintainsits greater accuracy characteristic even with coarser meshes
In terms of execution time the results from every tabledisplay shorter execution time for the IADEMF2 in compari-son to the IADEMF4 This is expected since the IADEMF2has lower computational complexity (Table 1) Despite theachievement in accuracy the IADEMF4 has to performmorecomputational work than the IADEMF2 since the former hasto utilize values of 119906 at two grid points on either side of thepoint (119894Δ119909 119896Δ119905) along the 119896th time level Thus if accuracyis desired then the preferred sequential numerical algorithmwould be the IADEMF4 On the other hand if execution timematters then the choice would be the IADEMF2
Amongst the fourth-order methods the IADEMF4 exe-cutes in the least amount of time Even though its computa-tional complexity is relatively large it operates with the leastnumber of iterations thus enabling it to be the most efficient
10 Conclusion
This paper proposes the development of the novel fourth-order IADEMF4 finite difference scheme The higher-orderscheme is developed by representing the spatial derivative inthe heat equation with the fourth-order finite difference CNapproximation This leads to the formation of pentadiagonalmatrices in the systems of linear equations with larger com-putational stencils The algorithm also employs the higheraccuracy of the Mitchell and Fairweather variant Despite
12 Journal of Applied Mathematics
the fourth-order IADE schemersquos higher computational com-plexity this technique is proved to be valuable becauseit enhances the accuracy of its second-order counterpartnamely the IADEMF2 In addition the higher accuracy ofIADEMF4 is verified as a convergent and unconditionallystable scheme and is superior in terms of rate of convergenceIt has also been proven to execute more efficiently in compar-ison to the other benchmarked fourth-order classical iterativemethods such as the GS4 the SOR4 and the JAC4 The in-creasing number of correct digits at each iteration serves as anadvantage for the IADEMF4 thus yielding faster rate of con-vergence with higher level of accuracy
In conclusion the proposed IADEMF4 scheme affordsusers many advantages with respect to higher accuracy sta-bility and rate of convergence and it serves as an alternativeefficient technique for the solution of a one-dimensional heatequation with Dirichlet boundary conditions
The proposed fourth-order scheme can be modified andadapted to more general multidimensional linear and non-linear parabolic elliptic and hyperbolic partial differentialequations using different types of boundary conditionsBesides it can also be considered for applications in problemsthat require higher-order accuracy with high resolution suchas problems in nanocomputing that require the solution ofvery large sparse systems of equation
The explicit and high accuracy feature of the IADEMF4can be exploited in two- and three-dimensional heat prob-lems by applying the scheme as pentadiagonal solvers forthe system of linear equations arising in the sweeps of ahigher-order alternating direction implicit (ADI) schemeThe lower-order ADI scheme was initially proposed byPeaceman and Rachford [24] The higher space dimensionsare expected not to cause unsatisfactory performance for theADI-IADEMF4 as long as the proposed scheme is stableconvergent and efficient as the process of its implementa-tion continues to advance from one-time step to anotherPathirana et al [25] implemented the ADI in developing atwo-dimensional model for incorporating flood damage inurban drainage planningThe proposed model is found to bestable numerically accurate and computationally efficientMirzavand et al [26] proposed the ADI-FDTD methodfor the physical modeling of high-frequency semiconductordevices The approach is able to reduce significantly the full-wave simulation time
It is possible to parallelize the IADEMF4 algorithm sincethe calculation of the new iteration only depends on theknown values from the last iteration (Figures 2 and 3) Futurework is to exploit the explicit computational properties andefficiency of the IADEMF4 for parallelization and executionon distributed parallel computing systems The idea is tospeed up the execution time without compromising its accu-racy especially on problems involving very large linear sys-tems of equations
References
[1] J Noye Computational Techniques for Differential Equationsvol 83 Elsevier Science North-Holland Publishing Amster-dam The Netherlands 1984
[2] G W Recktenwald ldquoFinite difference approximations to theheat equationrdquoMechanical Engineering vol 10 pp 1ndash27 2004
[3] P Lax and B Wendroff ldquoSystems of conservation lawsrdquo Com-munications on Pure and Applied Mathematics vol 13 pp 217ndash237 1960
[4] R J MacKinnon and G F Carey ldquoAnalysis of material interfacediscontinuities and superconvergent fluxes in finite differencetheoryrdquo Journal of Computational Physics vol 75 no 1 pp 151ndash167 1988
[5] R Hixon and E Turkel High Accuracy Compact MacCormack-Type Schemes For Computational Aerocoustics NASACentre forAerospace Information Lewis Research Centre 1998
[6] Y Lin Higher-order finite difference methods for solving heatequations [PhD thesis] Southern Illinois University 2008
[7] J Sulaiman M K Hasan M Othman M and S A AbdulKarim ldquoFourth-order QSMOR iterative method for the solu-tion of one-dimensional parabolic PDErsquosrdquo in Proceedings of theInternational Conference on Industrial and AppliedMathematics(CIAM rsquo10) pp 34ndash39 ITB Bandung Indonesia 2010
[8] N Jha ldquoThe application of sixth order accurate parallel quar-ter sweep alternating group explicit algorithm for nonlinearboundary value problems with singularityrdquo in Proceedings ofthe 2nd International Conference on Methods and Models inComputer Science (ICM2CS rsquo10) pp 76ndash80 December 2010
[9] Y Jin G Jin and J Li ldquoA class of high-precision finite differenceparallel algorithms for convection equationsrdquo Journal of Conver-gence Information Technology vol 6 no 1 pp 79ndash82 2011
[10] W Fu and E L Tan ldquoDevelopment of split-step FDTDmethodwith higher-order spatial accuracyrdquo Electronics Letters vol 40no 20 pp 1252ndash1253 2004
[11] A Mohebbi andM Dehghan ldquoHigh-order compact solution ofthe one-dimensional heat and advection-diffusion equationsrdquoApplied Mathematical Modelling vol 34 no 10 pp 3071ndash30842010
[12] W Liao ldquoAn implicit fourth-order compact finite differencescheme for one-dimensional Burgersrsquo equationrdquo Applied Math-ematics and Computation vol 206 no 2 pp 755ndash764 2008
[13] Z F Tian and Y B Ge ldquoA fourth-order compact ADI methodfor solving two-dimensional unsteady convection-diffusionproblemsrdquo Journal of Computational and Applied Mathematicsvol 198 no 1 pp 268ndash286 2007
[14] C Chun ldquoSome fourth-order iterativemethods for solving non-linear equationsrdquo Applied Mathematics and Computation vol195 no 2 pp 454ndash459 2008
[15] S Zhu Z Yu and J Zhao ldquoAhigh-order parallel finite differencealgorithmrdquo Applied Mathematics and Computation vol 183 no1 pp 365ndash372 2006
[16] Z Gao and S Xie ldquoFourth-order alternating direction implicitcompact finite difference schemes for two-dimensionalSchrodinger equationsrdquo Applied Numerical Mathematics vol61 no 4 pp 593ndash614 2011
[17] M S Sahimi A Ahmad and A A Bakar ldquoThe Iterative Alter-nating Decomposition Explicit (IADE) method to solve theheat conduction equationrdquo International Journal of ComputerMathematics vol 47 pp 219ndash229 1993
[18] G D Smith Numerical Solution of Partial Differential Equa-tions Finite Difference Methods Oxford University Press 2ndedition 1978
[19] A RMitchell andG Fairweather ldquoImproved forms of the alter-nating direction methods of Douglas Peaceman and Rachfordfor solving parabolic and elliptic equationsrdquo Numerische Math-ematik vol 6 no 1 pp 285ndash292 1964
Journal of Applied Mathematics 13
[20] B N DattaNumerical Linear Algebra and Applications BrooksCole Publishing Pacific Grove Calif USA 1995
[21] D M Young Iterative Solution of Large Linear Systems Aca-demic Press New York NY USA 1971
[22] V K Saulev Integration of Equations of Parabolic Type by theMethod of Nets Pergamon Press Oxford UK 1964
[23] L W Johnson and R D Riess Numerical Analysis Addison-Wesley Reading Mass USA 2nd edition 1982
[24] D W Peaceman and H H Rachford Jr ldquoThe numerical solu-tion of parabolic and elliptic differential equationsrdquo Journal ofthe Society for Industrial and Applied Mathematics vol 3 pp28ndash41 1955
[25] A Pathirana S Tsegaye B Gersonius and K VairavamoorthyldquoA simple 2-D inundation model for incorporating flood dam-age in urban drainage planningrdquo Hydrology and Earth SystemSciences vol 15 no 8 pp 2747ndash2761 2011
[26] R Mirzavand A Abdipour G Moradi and M MovahhedildquoFull-wave semiconductor devices simulation using ADI-FDTD methodrdquo Progress in Electromagnetics Research vol 11pp 191ndash202 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
10 Journal of Applied Mathematics
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 10 n = 169)IADEMF2 (r = 03 n = 213)
00E + 00
10E minus 02
20E minus 02
30E minus 02
40E minus 02
50E minus 02
60E minus 02
70E minus 02
80E minus 02
90E minus 02
10E minus 01
U(xt)
Figure 8 Numerical and exact solutions for Experiment 2 120582 = 05Δ119909 = 01 Δ119905 = 0005 119905 = 025 and 120576 = 10
minus4
0 01 02 03 04 05 06 07 08 09 1x
Exact solutionIADEMF4 (r = 10 n = 175)IADEMF2 (r = 05 n = 182)
00E + 00
10E minus 03
20E minus 03
30E minus 03
40E minus 03
50E minus 03
60E minus 03
70E minus 03
80E minus 03
U(xt)
Figure 9 Numerical and exact solutions for Experiment 2 120582 = 10Δ119909 = 01 Δ119905 = 001 119905 = 05 and 120576 = 10
minus4
The IADEMF2 is only derived from the second-order CNapproximation but its combination with its correspondingfourth-order MF variant managed to produce errors whichare better than theGS4 JAC4 and the SOR4 Even though theclassical iterative methods are also derived from the fourth-order accurate CN type approximation they are lackingin accuracy due to the round-off errors that have beenaccumulated from the time the execution starts till it ends
With regards to the rate of convergence the resultsfrom each table demonstrate that the number of iterationsproduced by the IADEMF2 and the fourth-order classicaliterative methods is greater than or at least equal to that of
0
minus2
minus208 minus277 minus347 minus416
minus4
minus8
minus6
minus10
minus12
minus14
minus16
minus18
GS4JAC4SOR4
log
(RM
SE)
log(Δx)
IADEMF4
IADEMF2
Figure 10 Log(RMSE) versus Log(Δ119909) Experiment 2 Δ119905 = 00001119905 = 0002 and 120576 = 10
minus4
Table 2 Experiment 1 119898 = 700 120582 = 05 Δ119909 = 143 times 10minus3 Δ119905 =
102 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 07) 267119890 minus 9 330119890 minus 9 212119890 minus 8 100 440119890 minus 2
IADEMF2(119903 = 08) 188119890 minus 8 495119890 minus 8 323119890 minus 7 100 635119890 minus 3
SOR4(120596 = 12) 540119890 minus 7 544119890 minus 7 552119890 minus 7 108 460119890 minus 2
GS4 540119890 minus 6 543119890 minus 6 332119890 minus 6 150 472119890 minus 2
JAC4 228119890 minus 5 229119890 minus 5 232119890 minus 5 150 473119890 minus 2
Table 3 Experiment 1 119898 = 700 120582 = 10 Δ119909 = 143 times 10minus3 Δ119905 =
204 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 297119890 minus 7 301119890 minus 7 301119890 minus 7 50 201119890 minus 2
IADEMF2(119903 = 09) 557119890 minus 7 559119890 minus 7 848119890 minus 7 50 317119890 minus 3
SOR4(120596 = 12) 329119890 minus 6 331119890 minus 6 335119890 minus 6 76 225119890 minus 2
GS4 895119890 minus 6 900119890 minus 6 900119890 minus 6 100 238119890 minus 2
JAC4 214119890 minus 5 215119890 minus 5 217119890 minus 5 125 250119890 minus 2
the IADEMF4 The convergence rate of the latter surpassesthe others in the case of 119898 = 7000 and 120576 = 10
minus10 (Tables6ndash8) Even though the operational count of the IADEMF4is relatively quite large (Table 1) due to its higher level ofaccuracy its increasing number of correct digits at eachiteration causes it to converge at a faster rate The IADEMF2has the advantage of having less mathematical operationsthus it is competitive in terms of convergence rate but atthe expense of accuracy The application of the fourth-orderCN approximation on the heat equation proves that thethree classical iterative methods also converge with JAC4appearing to be the slowest and the least accurate amongst all
Journal of Applied Mathematics 11
Table 4 Experiment 2 119898 = 700 120582 = 05 Δ119909 = 143 times 10minus3 Δ119905 =
102 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME n ET (s)IADEMF4(119903 = 07) 207119890 minus 8 231119890 minus 8 345119890 minus 8 100 903119890 minus 3
IADEMF2(119903 = 08) 106119890 minus 7 175119890 minus 7 175119890 minus 7 100 647119890 minus 3
SOR4(120596 = 11) 154119890 minus 6 171119890 minus 6 255119890 minus 6 200 110119890 minus 2
GS4 294119890 minus 6 327119890 minus 6 487119890 minus 6 250 118119890 minus 2
JAC4 124119890 minus 5 138119890 minus 5 206119890 minus 5 300 127119890 minus 2
Table 5 Experiment 2 119898 = 700 120582 = 10 Δ119909 = 143 times 10minus3 Δ119905 =
204 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 09) 138119890 minus 6 154119890 minus 6 229119890 minus 6 75 650119890 minus 3
IADEMF2(119903 = 05) 187119890 minus 6 207119890 minus 6 309119890 minus 6 75 472119890 minus 3
SOR4(120596 = 12) 312119890 minus 6 347119890 minus 6 517119890 minus 6 125 906119890 minus 3
GS4 667119890 minus 6 742119890 minus 6 111119890 minus 5 175 932119890 minus 3
JAC4 126119890 minus 5 141119890 minus 5 210119890 minus 5 250 105119890 minus 2
Table 6 Experiment 1 119898 = 7000 120582 = 05 Δ119909 = 143 times 10minus4 Δ119905 =
102 times 10minus8 119905 = 510 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME n ET (s)IADEMF4(119903 = 08) 367119890 minus 10 433119890 minus 10 366119890 minus 9 173 146119890 minus 1
IADEMF2(119903 = 05) 375119890 minus 10 435119890 minus 10 376119890 minus 9 200 124119890 minus 1
SOR4(120596 = 112) 455119890 minus 10 507119890 minus 10 367119890 minus 9 257 147119890 minus 1
GS4 110119890 minus 9 111119890 minus 9 376119890 minus 9 300 160119890 minus 1
JAC4 230119890 minus 9 231119890 minus 9 378119890 minus 9 400 163119890 minus 1
Table 7 Experiment 1 119898 = 7000 120582 = 10 Δ119909 = 143 times 10minus4 Δ119905 =
204 times 10minus8 119905 = 51 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 163119890 minus 7 163119890 minus 7 163119890 minus 7 112 960119890 minus 2
IADEMF2(119903 = 12) 163119890 minus 7 163119890 minus 7 163119890 minus 7 120 736119890 minus 2
SOR4(120596 = 115) 163119890 minus 7 163119890 minus 7 164119890 minus 7 171 999119890 minus 2
GS4 164119890 minus 7 164119890 minus 7 164119890 minus 7 216 107119890 minus 1
JAC4 165119890 minus 7 165119890 minus 7 165119890 minus 7 312 111119890 minus 1
It is found that the convergence rate of the fourth-orderand second-order methods improves with the application oflarger mesh size that is from 120582 = 05 to 120582 = 10 Thecoarser meshes cause a reduction in the computational oper-ations thus giving better rate of convergence However a
Table 8 Experiment 2 119898 = 7000 120582 = 05 Δ119909 = 143 times 10minus4 Δ119905 =
102 times 10minus8 119905 = 510 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 07) 567119890 minus 13 321119890 minus 12 130119890 minus 10 150 131119890 minus 1
IADEMF2(119903 = 07) 199119890 minus 11 460119890 minus 11 182119890 minus 9 200 127119890 minus 1
SOR4(120596 = 115) 245119890 minus 10 273119890 minus 10 399119890 minus 10 260 151119890 minus 1
GS4 398119890 minus 10 443119890 minus 10 661119890 minus 10 400 171119890 minus 1
JAC4 105119890 minus 9 117119890 minus 9 174119890 minus 9 550 176119890 minus 1
Table 9 Experiment 2 119898 = 7000 120582 = 10 Δ119909 = 143 times 10minus4 Δ119905 =
204 times 10minus8 119905 = 51 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 154119890 minus 6 171119890 minus 6 255119890 minus 6 96 912119890 minus 2
IADEMF2(119903 = 10) 154119890 minus 6 171119890 minus 6 255119890 minus 6 96 607119890 minus 2
SOR4(120596 = 12) 154119890 minus 6 171119890 minus 6 255119890 minus 6 192 997119890 minus 2
GS4 154119890 minus 6 171119890 minus 6 255119890 minus 6 288 128119890 minus 1
JAC4 154119890 minus 6 172119890 minus 6 256119890 minus 6 408 146119890 minus 1
more accurate solution is obtained by using finer mesh Forexample Table 6 shows that for 120582 = 05 the IADEMF4 hasRMSE = 433119890 minus 10 and 119899 = 173 whereas Table 7 shows thatfor 120582 = 10 its RMSE = 163119890 minus 7 and 119899 = 112 In generalamongst all the testedmethods the IADEMF4 still maintainsits greater accuracy characteristic even with coarser meshes
In terms of execution time the results from every tabledisplay shorter execution time for the IADEMF2 in compari-son to the IADEMF4 This is expected since the IADEMF2has lower computational complexity (Table 1) Despite theachievement in accuracy the IADEMF4 has to performmorecomputational work than the IADEMF2 since the former hasto utilize values of 119906 at two grid points on either side of thepoint (119894Δ119909 119896Δ119905) along the 119896th time level Thus if accuracyis desired then the preferred sequential numerical algorithmwould be the IADEMF4 On the other hand if execution timematters then the choice would be the IADEMF2
Amongst the fourth-order methods the IADEMF4 exe-cutes in the least amount of time Even though its computa-tional complexity is relatively large it operates with the leastnumber of iterations thus enabling it to be the most efficient
10 Conclusion
This paper proposes the development of the novel fourth-order IADEMF4 finite difference scheme The higher-orderscheme is developed by representing the spatial derivative inthe heat equation with the fourth-order finite difference CNapproximation This leads to the formation of pentadiagonalmatrices in the systems of linear equations with larger com-putational stencils The algorithm also employs the higheraccuracy of the Mitchell and Fairweather variant Despite
12 Journal of Applied Mathematics
the fourth-order IADE schemersquos higher computational com-plexity this technique is proved to be valuable becauseit enhances the accuracy of its second-order counterpartnamely the IADEMF2 In addition the higher accuracy ofIADEMF4 is verified as a convergent and unconditionallystable scheme and is superior in terms of rate of convergenceIt has also been proven to execute more efficiently in compar-ison to the other benchmarked fourth-order classical iterativemethods such as the GS4 the SOR4 and the JAC4 The in-creasing number of correct digits at each iteration serves as anadvantage for the IADEMF4 thus yielding faster rate of con-vergence with higher level of accuracy
In conclusion the proposed IADEMF4 scheme affordsusers many advantages with respect to higher accuracy sta-bility and rate of convergence and it serves as an alternativeefficient technique for the solution of a one-dimensional heatequation with Dirichlet boundary conditions
The proposed fourth-order scheme can be modified andadapted to more general multidimensional linear and non-linear parabolic elliptic and hyperbolic partial differentialequations using different types of boundary conditionsBesides it can also be considered for applications in problemsthat require higher-order accuracy with high resolution suchas problems in nanocomputing that require the solution ofvery large sparse systems of equation
The explicit and high accuracy feature of the IADEMF4can be exploited in two- and three-dimensional heat prob-lems by applying the scheme as pentadiagonal solvers forthe system of linear equations arising in the sweeps of ahigher-order alternating direction implicit (ADI) schemeThe lower-order ADI scheme was initially proposed byPeaceman and Rachford [24] The higher space dimensionsare expected not to cause unsatisfactory performance for theADI-IADEMF4 as long as the proposed scheme is stableconvergent and efficient as the process of its implementa-tion continues to advance from one-time step to anotherPathirana et al [25] implemented the ADI in developing atwo-dimensional model for incorporating flood damage inurban drainage planningThe proposed model is found to bestable numerically accurate and computationally efficientMirzavand et al [26] proposed the ADI-FDTD methodfor the physical modeling of high-frequency semiconductordevices The approach is able to reduce significantly the full-wave simulation time
It is possible to parallelize the IADEMF4 algorithm sincethe calculation of the new iteration only depends on theknown values from the last iteration (Figures 2 and 3) Futurework is to exploit the explicit computational properties andefficiency of the IADEMF4 for parallelization and executionon distributed parallel computing systems The idea is tospeed up the execution time without compromising its accu-racy especially on problems involving very large linear sys-tems of equations
References
[1] J Noye Computational Techniques for Differential Equationsvol 83 Elsevier Science North-Holland Publishing Amster-dam The Netherlands 1984
[2] G W Recktenwald ldquoFinite difference approximations to theheat equationrdquoMechanical Engineering vol 10 pp 1ndash27 2004
[3] P Lax and B Wendroff ldquoSystems of conservation lawsrdquo Com-munications on Pure and Applied Mathematics vol 13 pp 217ndash237 1960
[4] R J MacKinnon and G F Carey ldquoAnalysis of material interfacediscontinuities and superconvergent fluxes in finite differencetheoryrdquo Journal of Computational Physics vol 75 no 1 pp 151ndash167 1988
[5] R Hixon and E Turkel High Accuracy Compact MacCormack-Type Schemes For Computational Aerocoustics NASACentre forAerospace Information Lewis Research Centre 1998
[6] Y Lin Higher-order finite difference methods for solving heatequations [PhD thesis] Southern Illinois University 2008
[7] J Sulaiman M K Hasan M Othman M and S A AbdulKarim ldquoFourth-order QSMOR iterative method for the solu-tion of one-dimensional parabolic PDErsquosrdquo in Proceedings of theInternational Conference on Industrial and AppliedMathematics(CIAM rsquo10) pp 34ndash39 ITB Bandung Indonesia 2010
[8] N Jha ldquoThe application of sixth order accurate parallel quar-ter sweep alternating group explicit algorithm for nonlinearboundary value problems with singularityrdquo in Proceedings ofthe 2nd International Conference on Methods and Models inComputer Science (ICM2CS rsquo10) pp 76ndash80 December 2010
[9] Y Jin G Jin and J Li ldquoA class of high-precision finite differenceparallel algorithms for convection equationsrdquo Journal of Conver-gence Information Technology vol 6 no 1 pp 79ndash82 2011
[10] W Fu and E L Tan ldquoDevelopment of split-step FDTDmethodwith higher-order spatial accuracyrdquo Electronics Letters vol 40no 20 pp 1252ndash1253 2004
[11] A Mohebbi andM Dehghan ldquoHigh-order compact solution ofthe one-dimensional heat and advection-diffusion equationsrdquoApplied Mathematical Modelling vol 34 no 10 pp 3071ndash30842010
[12] W Liao ldquoAn implicit fourth-order compact finite differencescheme for one-dimensional Burgersrsquo equationrdquo Applied Math-ematics and Computation vol 206 no 2 pp 755ndash764 2008
[13] Z F Tian and Y B Ge ldquoA fourth-order compact ADI methodfor solving two-dimensional unsteady convection-diffusionproblemsrdquo Journal of Computational and Applied Mathematicsvol 198 no 1 pp 268ndash286 2007
[14] C Chun ldquoSome fourth-order iterativemethods for solving non-linear equationsrdquo Applied Mathematics and Computation vol195 no 2 pp 454ndash459 2008
[15] S Zhu Z Yu and J Zhao ldquoAhigh-order parallel finite differencealgorithmrdquo Applied Mathematics and Computation vol 183 no1 pp 365ndash372 2006
[16] Z Gao and S Xie ldquoFourth-order alternating direction implicitcompact finite difference schemes for two-dimensionalSchrodinger equationsrdquo Applied Numerical Mathematics vol61 no 4 pp 593ndash614 2011
[17] M S Sahimi A Ahmad and A A Bakar ldquoThe Iterative Alter-nating Decomposition Explicit (IADE) method to solve theheat conduction equationrdquo International Journal of ComputerMathematics vol 47 pp 219ndash229 1993
[18] G D Smith Numerical Solution of Partial Differential Equa-tions Finite Difference Methods Oxford University Press 2ndedition 1978
[19] A RMitchell andG Fairweather ldquoImproved forms of the alter-nating direction methods of Douglas Peaceman and Rachfordfor solving parabolic and elliptic equationsrdquo Numerische Math-ematik vol 6 no 1 pp 285ndash292 1964
Journal of Applied Mathematics 13
[20] B N DattaNumerical Linear Algebra and Applications BrooksCole Publishing Pacific Grove Calif USA 1995
[21] D M Young Iterative Solution of Large Linear Systems Aca-demic Press New York NY USA 1971
[22] V K Saulev Integration of Equations of Parabolic Type by theMethod of Nets Pergamon Press Oxford UK 1964
[23] L W Johnson and R D Riess Numerical Analysis Addison-Wesley Reading Mass USA 2nd edition 1982
[24] D W Peaceman and H H Rachford Jr ldquoThe numerical solu-tion of parabolic and elliptic differential equationsrdquo Journal ofthe Society for Industrial and Applied Mathematics vol 3 pp28ndash41 1955
[25] A Pathirana S Tsegaye B Gersonius and K VairavamoorthyldquoA simple 2-D inundation model for incorporating flood dam-age in urban drainage planningrdquo Hydrology and Earth SystemSciences vol 15 no 8 pp 2747ndash2761 2011
[26] R Mirzavand A Abdipour G Moradi and M MovahhedildquoFull-wave semiconductor devices simulation using ADI-FDTD methodrdquo Progress in Electromagnetics Research vol 11pp 191ndash202 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Journal of Applied Mathematics 11
Table 4 Experiment 2 119898 = 700 120582 = 05 Δ119909 = 143 times 10minus3 Δ119905 =
102 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME n ET (s)IADEMF4(119903 = 07) 207119890 minus 8 231119890 minus 8 345119890 minus 8 100 903119890 minus 3
IADEMF2(119903 = 08) 106119890 minus 7 175119890 minus 7 175119890 minus 7 100 647119890 minus 3
SOR4(120596 = 11) 154119890 minus 6 171119890 minus 6 255119890 minus 6 200 110119890 minus 2
GS4 294119890 minus 6 327119890 minus 6 487119890 minus 6 250 118119890 minus 2
JAC4 124119890 minus 5 138119890 minus 5 206119890 minus 5 300 127119890 minus 2
Table 5 Experiment 2 119898 = 700 120582 = 10 Δ119909 = 143 times 10minus3 Δ119905 =
204 times 10minus6 119905 = 509 times 10
minus5 and 120576 = 10minus6
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 09) 138119890 minus 6 154119890 minus 6 229119890 minus 6 75 650119890 minus 3
IADEMF2(119903 = 05) 187119890 minus 6 207119890 minus 6 309119890 minus 6 75 472119890 minus 3
SOR4(120596 = 12) 312119890 minus 6 347119890 minus 6 517119890 minus 6 125 906119890 minus 3
GS4 667119890 minus 6 742119890 minus 6 111119890 minus 5 175 932119890 minus 3
JAC4 126119890 minus 5 141119890 minus 5 210119890 minus 5 250 105119890 minus 2
Table 6 Experiment 1 119898 = 7000 120582 = 05 Δ119909 = 143 times 10minus4 Δ119905 =
102 times 10minus8 119905 = 510 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME n ET (s)IADEMF4(119903 = 08) 367119890 minus 10 433119890 minus 10 366119890 minus 9 173 146119890 minus 1
IADEMF2(119903 = 05) 375119890 minus 10 435119890 minus 10 376119890 minus 9 200 124119890 minus 1
SOR4(120596 = 112) 455119890 minus 10 507119890 minus 10 367119890 minus 9 257 147119890 minus 1
GS4 110119890 minus 9 111119890 minus 9 376119890 minus 9 300 160119890 minus 1
JAC4 230119890 minus 9 231119890 minus 9 378119890 minus 9 400 163119890 minus 1
Table 7 Experiment 1 119898 = 7000 120582 = 10 Δ119909 = 143 times 10minus4 Δ119905 =
204 times 10minus8 119905 = 51 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 163119890 minus 7 163119890 minus 7 163119890 minus 7 112 960119890 minus 2
IADEMF2(119903 = 12) 163119890 minus 7 163119890 minus 7 163119890 minus 7 120 736119890 minus 2
SOR4(120596 = 115) 163119890 minus 7 163119890 minus 7 164119890 minus 7 171 999119890 minus 2
GS4 164119890 minus 7 164119890 minus 7 164119890 minus 7 216 107119890 minus 1
JAC4 165119890 minus 7 165119890 minus 7 165119890 minus 7 312 111119890 minus 1
It is found that the convergence rate of the fourth-orderand second-order methods improves with the application oflarger mesh size that is from 120582 = 05 to 120582 = 10 Thecoarser meshes cause a reduction in the computational oper-ations thus giving better rate of convergence However a
Table 8 Experiment 2 119898 = 7000 120582 = 05 Δ119909 = 143 times 10minus4 Δ119905 =
102 times 10minus8 119905 = 510 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 07) 567119890 minus 13 321119890 minus 12 130119890 minus 10 150 131119890 minus 1
IADEMF2(119903 = 07) 199119890 minus 11 460119890 minus 11 182119890 minus 9 200 127119890 minus 1
SOR4(120596 = 115) 245119890 minus 10 273119890 minus 10 399119890 minus 10 260 151119890 minus 1
GS4 398119890 minus 10 443119890 minus 10 661119890 minus 10 400 171119890 minus 1
JAC4 105119890 minus 9 117119890 minus 9 174119890 minus 9 550 176119890 minus 1
Table 9 Experiment 2 119898 = 7000 120582 = 10 Δ119909 = 143 times 10minus4 Δ119905 =
204 times 10minus8 119905 = 51 times 10
minus7 and 120576 = 10minus10
Method AAE RMSE ME 119899 ET (s)IADEMF4(119903 = 10) 154119890 minus 6 171119890 minus 6 255119890 minus 6 96 912119890 minus 2
IADEMF2(119903 = 10) 154119890 minus 6 171119890 minus 6 255119890 minus 6 96 607119890 minus 2
SOR4(120596 = 12) 154119890 minus 6 171119890 minus 6 255119890 minus 6 192 997119890 minus 2
GS4 154119890 minus 6 171119890 minus 6 255119890 minus 6 288 128119890 minus 1
JAC4 154119890 minus 6 172119890 minus 6 256119890 minus 6 408 146119890 minus 1
more accurate solution is obtained by using finer mesh Forexample Table 6 shows that for 120582 = 05 the IADEMF4 hasRMSE = 433119890 minus 10 and 119899 = 173 whereas Table 7 shows thatfor 120582 = 10 its RMSE = 163119890 minus 7 and 119899 = 112 In generalamongst all the testedmethods the IADEMF4 still maintainsits greater accuracy characteristic even with coarser meshes
In terms of execution time the results from every tabledisplay shorter execution time for the IADEMF2 in compari-son to the IADEMF4 This is expected since the IADEMF2has lower computational complexity (Table 1) Despite theachievement in accuracy the IADEMF4 has to performmorecomputational work than the IADEMF2 since the former hasto utilize values of 119906 at two grid points on either side of thepoint (119894Δ119909 119896Δ119905) along the 119896th time level Thus if accuracyis desired then the preferred sequential numerical algorithmwould be the IADEMF4 On the other hand if execution timematters then the choice would be the IADEMF2
Amongst the fourth-order methods the IADEMF4 exe-cutes in the least amount of time Even though its computa-tional complexity is relatively large it operates with the leastnumber of iterations thus enabling it to be the most efficient
10 Conclusion
This paper proposes the development of the novel fourth-order IADEMF4 finite difference scheme The higher-orderscheme is developed by representing the spatial derivative inthe heat equation with the fourth-order finite difference CNapproximation This leads to the formation of pentadiagonalmatrices in the systems of linear equations with larger com-putational stencils The algorithm also employs the higheraccuracy of the Mitchell and Fairweather variant Despite
12 Journal of Applied Mathematics
the fourth-order IADE schemersquos higher computational com-plexity this technique is proved to be valuable becauseit enhances the accuracy of its second-order counterpartnamely the IADEMF2 In addition the higher accuracy ofIADEMF4 is verified as a convergent and unconditionallystable scheme and is superior in terms of rate of convergenceIt has also been proven to execute more efficiently in compar-ison to the other benchmarked fourth-order classical iterativemethods such as the GS4 the SOR4 and the JAC4 The in-creasing number of correct digits at each iteration serves as anadvantage for the IADEMF4 thus yielding faster rate of con-vergence with higher level of accuracy
In conclusion the proposed IADEMF4 scheme affordsusers many advantages with respect to higher accuracy sta-bility and rate of convergence and it serves as an alternativeefficient technique for the solution of a one-dimensional heatequation with Dirichlet boundary conditions
The proposed fourth-order scheme can be modified andadapted to more general multidimensional linear and non-linear parabolic elliptic and hyperbolic partial differentialequations using different types of boundary conditionsBesides it can also be considered for applications in problemsthat require higher-order accuracy with high resolution suchas problems in nanocomputing that require the solution ofvery large sparse systems of equation
The explicit and high accuracy feature of the IADEMF4can be exploited in two- and three-dimensional heat prob-lems by applying the scheme as pentadiagonal solvers forthe system of linear equations arising in the sweeps of ahigher-order alternating direction implicit (ADI) schemeThe lower-order ADI scheme was initially proposed byPeaceman and Rachford [24] The higher space dimensionsare expected not to cause unsatisfactory performance for theADI-IADEMF4 as long as the proposed scheme is stableconvergent and efficient as the process of its implementa-tion continues to advance from one-time step to anotherPathirana et al [25] implemented the ADI in developing atwo-dimensional model for incorporating flood damage inurban drainage planningThe proposed model is found to bestable numerically accurate and computationally efficientMirzavand et al [26] proposed the ADI-FDTD methodfor the physical modeling of high-frequency semiconductordevices The approach is able to reduce significantly the full-wave simulation time
It is possible to parallelize the IADEMF4 algorithm sincethe calculation of the new iteration only depends on theknown values from the last iteration (Figures 2 and 3) Futurework is to exploit the explicit computational properties andefficiency of the IADEMF4 for parallelization and executionon distributed parallel computing systems The idea is tospeed up the execution time without compromising its accu-racy especially on problems involving very large linear sys-tems of equations
References
[1] J Noye Computational Techniques for Differential Equationsvol 83 Elsevier Science North-Holland Publishing Amster-dam The Netherlands 1984
[2] G W Recktenwald ldquoFinite difference approximations to theheat equationrdquoMechanical Engineering vol 10 pp 1ndash27 2004
[3] P Lax and B Wendroff ldquoSystems of conservation lawsrdquo Com-munications on Pure and Applied Mathematics vol 13 pp 217ndash237 1960
[4] R J MacKinnon and G F Carey ldquoAnalysis of material interfacediscontinuities and superconvergent fluxes in finite differencetheoryrdquo Journal of Computational Physics vol 75 no 1 pp 151ndash167 1988
[5] R Hixon and E Turkel High Accuracy Compact MacCormack-Type Schemes For Computational Aerocoustics NASACentre forAerospace Information Lewis Research Centre 1998
[6] Y Lin Higher-order finite difference methods for solving heatequations [PhD thesis] Southern Illinois University 2008
[7] J Sulaiman M K Hasan M Othman M and S A AbdulKarim ldquoFourth-order QSMOR iterative method for the solu-tion of one-dimensional parabolic PDErsquosrdquo in Proceedings of theInternational Conference on Industrial and AppliedMathematics(CIAM rsquo10) pp 34ndash39 ITB Bandung Indonesia 2010
[8] N Jha ldquoThe application of sixth order accurate parallel quar-ter sweep alternating group explicit algorithm for nonlinearboundary value problems with singularityrdquo in Proceedings ofthe 2nd International Conference on Methods and Models inComputer Science (ICM2CS rsquo10) pp 76ndash80 December 2010
[9] Y Jin G Jin and J Li ldquoA class of high-precision finite differenceparallel algorithms for convection equationsrdquo Journal of Conver-gence Information Technology vol 6 no 1 pp 79ndash82 2011
[10] W Fu and E L Tan ldquoDevelopment of split-step FDTDmethodwith higher-order spatial accuracyrdquo Electronics Letters vol 40no 20 pp 1252ndash1253 2004
[11] A Mohebbi andM Dehghan ldquoHigh-order compact solution ofthe one-dimensional heat and advection-diffusion equationsrdquoApplied Mathematical Modelling vol 34 no 10 pp 3071ndash30842010
[12] W Liao ldquoAn implicit fourth-order compact finite differencescheme for one-dimensional Burgersrsquo equationrdquo Applied Math-ematics and Computation vol 206 no 2 pp 755ndash764 2008
[13] Z F Tian and Y B Ge ldquoA fourth-order compact ADI methodfor solving two-dimensional unsteady convection-diffusionproblemsrdquo Journal of Computational and Applied Mathematicsvol 198 no 1 pp 268ndash286 2007
[14] C Chun ldquoSome fourth-order iterativemethods for solving non-linear equationsrdquo Applied Mathematics and Computation vol195 no 2 pp 454ndash459 2008
[15] S Zhu Z Yu and J Zhao ldquoAhigh-order parallel finite differencealgorithmrdquo Applied Mathematics and Computation vol 183 no1 pp 365ndash372 2006
[16] Z Gao and S Xie ldquoFourth-order alternating direction implicitcompact finite difference schemes for two-dimensionalSchrodinger equationsrdquo Applied Numerical Mathematics vol61 no 4 pp 593ndash614 2011
[17] M S Sahimi A Ahmad and A A Bakar ldquoThe Iterative Alter-nating Decomposition Explicit (IADE) method to solve theheat conduction equationrdquo International Journal of ComputerMathematics vol 47 pp 219ndash229 1993
[18] G D Smith Numerical Solution of Partial Differential Equa-tions Finite Difference Methods Oxford University Press 2ndedition 1978
[19] A RMitchell andG Fairweather ldquoImproved forms of the alter-nating direction methods of Douglas Peaceman and Rachfordfor solving parabolic and elliptic equationsrdquo Numerische Math-ematik vol 6 no 1 pp 285ndash292 1964
Journal of Applied Mathematics 13
[20] B N DattaNumerical Linear Algebra and Applications BrooksCole Publishing Pacific Grove Calif USA 1995
[21] D M Young Iterative Solution of Large Linear Systems Aca-demic Press New York NY USA 1971
[22] V K Saulev Integration of Equations of Parabolic Type by theMethod of Nets Pergamon Press Oxford UK 1964
[23] L W Johnson and R D Riess Numerical Analysis Addison-Wesley Reading Mass USA 2nd edition 1982
[24] D W Peaceman and H H Rachford Jr ldquoThe numerical solu-tion of parabolic and elliptic differential equationsrdquo Journal ofthe Society for Industrial and Applied Mathematics vol 3 pp28ndash41 1955
[25] A Pathirana S Tsegaye B Gersonius and K VairavamoorthyldquoA simple 2-D inundation model for incorporating flood dam-age in urban drainage planningrdquo Hydrology and Earth SystemSciences vol 15 no 8 pp 2747ndash2761 2011
[26] R Mirzavand A Abdipour G Moradi and M MovahhedildquoFull-wave semiconductor devices simulation using ADI-FDTD methodrdquo Progress in Electromagnetics Research vol 11pp 191ndash202 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
12 Journal of Applied Mathematics
the fourth-order IADE schemersquos higher computational com-plexity this technique is proved to be valuable becauseit enhances the accuracy of its second-order counterpartnamely the IADEMF2 In addition the higher accuracy ofIADEMF4 is verified as a convergent and unconditionallystable scheme and is superior in terms of rate of convergenceIt has also been proven to execute more efficiently in compar-ison to the other benchmarked fourth-order classical iterativemethods such as the GS4 the SOR4 and the JAC4 The in-creasing number of correct digits at each iteration serves as anadvantage for the IADEMF4 thus yielding faster rate of con-vergence with higher level of accuracy
In conclusion the proposed IADEMF4 scheme affordsusers many advantages with respect to higher accuracy sta-bility and rate of convergence and it serves as an alternativeefficient technique for the solution of a one-dimensional heatequation with Dirichlet boundary conditions
The proposed fourth-order scheme can be modified andadapted to more general multidimensional linear and non-linear parabolic elliptic and hyperbolic partial differentialequations using different types of boundary conditionsBesides it can also be considered for applications in problemsthat require higher-order accuracy with high resolution suchas problems in nanocomputing that require the solution ofvery large sparse systems of equation
The explicit and high accuracy feature of the IADEMF4can be exploited in two- and three-dimensional heat prob-lems by applying the scheme as pentadiagonal solvers forthe system of linear equations arising in the sweeps of ahigher-order alternating direction implicit (ADI) schemeThe lower-order ADI scheme was initially proposed byPeaceman and Rachford [24] The higher space dimensionsare expected not to cause unsatisfactory performance for theADI-IADEMF4 as long as the proposed scheme is stableconvergent and efficient as the process of its implementa-tion continues to advance from one-time step to anotherPathirana et al [25] implemented the ADI in developing atwo-dimensional model for incorporating flood damage inurban drainage planningThe proposed model is found to bestable numerically accurate and computationally efficientMirzavand et al [26] proposed the ADI-FDTD methodfor the physical modeling of high-frequency semiconductordevices The approach is able to reduce significantly the full-wave simulation time
It is possible to parallelize the IADEMF4 algorithm sincethe calculation of the new iteration only depends on theknown values from the last iteration (Figures 2 and 3) Futurework is to exploit the explicit computational properties andefficiency of the IADEMF4 for parallelization and executionon distributed parallel computing systems The idea is tospeed up the execution time without compromising its accu-racy especially on problems involving very large linear sys-tems of equations
References
[1] J Noye Computational Techniques for Differential Equationsvol 83 Elsevier Science North-Holland Publishing Amster-dam The Netherlands 1984
[2] G W Recktenwald ldquoFinite difference approximations to theheat equationrdquoMechanical Engineering vol 10 pp 1ndash27 2004
[3] P Lax and B Wendroff ldquoSystems of conservation lawsrdquo Com-munications on Pure and Applied Mathematics vol 13 pp 217ndash237 1960
[4] R J MacKinnon and G F Carey ldquoAnalysis of material interfacediscontinuities and superconvergent fluxes in finite differencetheoryrdquo Journal of Computational Physics vol 75 no 1 pp 151ndash167 1988
[5] R Hixon and E Turkel High Accuracy Compact MacCormack-Type Schemes For Computational Aerocoustics NASACentre forAerospace Information Lewis Research Centre 1998
[6] Y Lin Higher-order finite difference methods for solving heatequations [PhD thesis] Southern Illinois University 2008
[7] J Sulaiman M K Hasan M Othman M and S A AbdulKarim ldquoFourth-order QSMOR iterative method for the solu-tion of one-dimensional parabolic PDErsquosrdquo in Proceedings of theInternational Conference on Industrial and AppliedMathematics(CIAM rsquo10) pp 34ndash39 ITB Bandung Indonesia 2010
[8] N Jha ldquoThe application of sixth order accurate parallel quar-ter sweep alternating group explicit algorithm for nonlinearboundary value problems with singularityrdquo in Proceedings ofthe 2nd International Conference on Methods and Models inComputer Science (ICM2CS rsquo10) pp 76ndash80 December 2010
[9] Y Jin G Jin and J Li ldquoA class of high-precision finite differenceparallel algorithms for convection equationsrdquo Journal of Conver-gence Information Technology vol 6 no 1 pp 79ndash82 2011
[10] W Fu and E L Tan ldquoDevelopment of split-step FDTDmethodwith higher-order spatial accuracyrdquo Electronics Letters vol 40no 20 pp 1252ndash1253 2004
[11] A Mohebbi andM Dehghan ldquoHigh-order compact solution ofthe one-dimensional heat and advection-diffusion equationsrdquoApplied Mathematical Modelling vol 34 no 10 pp 3071ndash30842010
[12] W Liao ldquoAn implicit fourth-order compact finite differencescheme for one-dimensional Burgersrsquo equationrdquo Applied Math-ematics and Computation vol 206 no 2 pp 755ndash764 2008
[13] Z F Tian and Y B Ge ldquoA fourth-order compact ADI methodfor solving two-dimensional unsteady convection-diffusionproblemsrdquo Journal of Computational and Applied Mathematicsvol 198 no 1 pp 268ndash286 2007
[14] C Chun ldquoSome fourth-order iterativemethods for solving non-linear equationsrdquo Applied Mathematics and Computation vol195 no 2 pp 454ndash459 2008
[15] S Zhu Z Yu and J Zhao ldquoAhigh-order parallel finite differencealgorithmrdquo Applied Mathematics and Computation vol 183 no1 pp 365ndash372 2006
[16] Z Gao and S Xie ldquoFourth-order alternating direction implicitcompact finite difference schemes for two-dimensionalSchrodinger equationsrdquo Applied Numerical Mathematics vol61 no 4 pp 593ndash614 2011
[17] M S Sahimi A Ahmad and A A Bakar ldquoThe Iterative Alter-nating Decomposition Explicit (IADE) method to solve theheat conduction equationrdquo International Journal of ComputerMathematics vol 47 pp 219ndash229 1993
[18] G D Smith Numerical Solution of Partial Differential Equa-tions Finite Difference Methods Oxford University Press 2ndedition 1978
[19] A RMitchell andG Fairweather ldquoImproved forms of the alter-nating direction methods of Douglas Peaceman and Rachfordfor solving parabolic and elliptic equationsrdquo Numerische Math-ematik vol 6 no 1 pp 285ndash292 1964
Journal of Applied Mathematics 13
[20] B N DattaNumerical Linear Algebra and Applications BrooksCole Publishing Pacific Grove Calif USA 1995
[21] D M Young Iterative Solution of Large Linear Systems Aca-demic Press New York NY USA 1971
[22] V K Saulev Integration of Equations of Parabolic Type by theMethod of Nets Pergamon Press Oxford UK 1964
[23] L W Johnson and R D Riess Numerical Analysis Addison-Wesley Reading Mass USA 2nd edition 1982
[24] D W Peaceman and H H Rachford Jr ldquoThe numerical solu-tion of parabolic and elliptic differential equationsrdquo Journal ofthe Society for Industrial and Applied Mathematics vol 3 pp28ndash41 1955
[25] A Pathirana S Tsegaye B Gersonius and K VairavamoorthyldquoA simple 2-D inundation model for incorporating flood dam-age in urban drainage planningrdquo Hydrology and Earth SystemSciences vol 15 no 8 pp 2747ndash2761 2011
[26] R Mirzavand A Abdipour G Moradi and M MovahhedildquoFull-wave semiconductor devices simulation using ADI-FDTD methodrdquo Progress in Electromagnetics Research vol 11pp 191ndash202 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Journal of Applied Mathematics 13
[20] B N DattaNumerical Linear Algebra and Applications BrooksCole Publishing Pacific Grove Calif USA 1995
[21] D M Young Iterative Solution of Large Linear Systems Aca-demic Press New York NY USA 1971
[22] V K Saulev Integration of Equations of Parabolic Type by theMethod of Nets Pergamon Press Oxford UK 1964
[23] L W Johnson and R D Riess Numerical Analysis Addison-Wesley Reading Mass USA 2nd edition 1982
[24] D W Peaceman and H H Rachford Jr ldquoThe numerical solu-tion of parabolic and elliptic differential equationsrdquo Journal ofthe Society for Industrial and Applied Mathematics vol 3 pp28ndash41 1955
[25] A Pathirana S Tsegaye B Gersonius and K VairavamoorthyldquoA simple 2-D inundation model for incorporating flood dam-age in urban drainage planningrdquo Hydrology and Earth SystemSciences vol 15 no 8 pp 2747ndash2761 2011
[26] R Mirzavand A Abdipour G Moradi and M MovahhedildquoFull-wave semiconductor devices simulation using ADI-FDTD methodrdquo Progress in Electromagnetics Research vol 11pp 191ndash202 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of