Open Access Baghdad Science Journal P-ISSN: 2078-8665

2020, 17(2 Special Issue)NICST:675-681 E-ISSN: 2411-7986

DOI: http://dx.doi.org/10.21123/bsj.2020.17.2(SI).0675

Splitting the One-Dimensional Wave Equation.

Part I: Solving by Finite-Difference Method and Separation Variables
S.O. Hussein

Received 18/9/2019, Accepted 23/4/2020, Published 23/6/2020

This work is licensed under a Creative Commons Attribution 4.0 International License.

Abstract:
In this study, an unknown force function dependent on the space in the wave equation is
investigated. Numerically wave equation splitting in two parts, part one using the finite-difference method
(FDM). Part two using separating variables method. This is the continuation and changing technique for
solving inverse problem part in (1,2). Instead, the boundary element method (BEM) in (1,2), the finite-
difference method (FDM) has applied. Boundary data are in the role of overdetermination data. The second
part of the problem is inverse and ill-posed, since small errors in the extra boundary data cause errors in the
force solution. Zeroth order of Tikhonov regularization, and several parameters of regularization are
employed to decrease errors for output force solution. It is obvious from figures how error affects the results
and zeroth order stables the solution.

Key words: Finite difference method, Inverse force problem, Regularization, Separation variables method,
Wave equation.

Introduction:
In physics, the wave equation uses 2
𝑢𝑡𝑡(𝑥, 𝑡) = 𝛻 𝑢(𝑥, 𝑡) + 𝑓(𝑥), (𝑥, 𝑡) ∈ 𝛺 ×
vibrations of a spring or membrane, acoustic
(0, 𝑇), 𝑇 > 0, 𝛺 ∈ 𝑅 (1)
scattering, etc. (1-3). The aim of this study: Firstly,
where (𝑢(𝑥, 𝑡), 𝑓(𝑥)) represents the displacement
to investigate and use finite-difference method
and force, respectively.
(FDM) rather than boundary element method
(BEM) in (1,2). Secondly, to show numerical results 𝑢(𝑥, 0) = 𝑢0(𝑥), 𝑢𝑡(𝑥, 0) = 𝑣0(𝑥),
and compare to (3-6), in addition presented 𝑥 ∈ 𝛺. (2)
numerical results were not shown in (4). Equation (2) is the initial condition, when 𝑢0
Furthermore, numerical solving for inverse represent the initial displacement and 𝑣0 velocity.
problems part by finite-difference method (FDM), For solving equation (1)-(2) need to have boundary
and apply the Tikhonov regularization to stable the condition, first let consider Dirichlet boundary
ill-posed inverse problem with various conditions
regularization parameters (2,3). In this research the 𝑢(𝑥, 𝑡) = 𝑃(𝑥, 𝑡), (𝑥, 𝑡) ∈ 𝜕𝛺 × (0, 𝑇). (3)
condition and the theory that insure uniqueness The equation above calls direct well-posed problem
solution is not provided because it is the same in (1- if the 𝑓(𝑥) is given. Otherwise, for solving
5). Inverse problem is linear source inverse problem equations (1)-(3) need to have extra data usually if
(1-3). The design of the paper is as follows: The Dirichlet boundary conditions taken as boundary
mathematical formulation is given in Section 2. condition then Neumann condition will be
Sections 3, illustrates numerical results and additional data or posit, i.e. the same condition will
discussed, finally, conclusions are offered in not be existed in original problem
Section 4. 𝜕𝑢
(𝑥, 𝑡) = 𝑞(𝑥, 𝑡), (𝑥, 𝑡) ∈ 𝛤 × (0, 𝑇),
𝜕𝜈
𝛤 ⊂ 𝜕𝛺 (4)
Mathematical Formulations:
where 𝜈 is the outward unit normal to 𝜕𝛺, (2,3).
The require equation for nonhomogeneous
After that the inverse problem (1)-(4) split into
and one-dimension wave equation (1-3)
the form 𝑢 = 𝑣 + 𝑤 (1,2,4). Part one: 𝑣 satisfies
Department of Mathematics, College of Science,
University of Sulaimani, Sulaimani, Iraq.
the well-posed direct problem (1,2,4)
E-mails: [email protected] 𝑣𝑡𝑡(𝑥, 𝑡) = 𝑣𝑥𝑥(𝑥, 𝑡),
*
ORCID ID: https://orcid.org/0000-0002-9652-9261 (𝑥, 𝑡) ∈ 𝛺 × (0, 𝑇), (5)

675
Open Access Baghdad Science Journal P-ISSN: 2078-8665
2020, 17(2 Special Issue)NICST:675-681 E-ISSN: 2411-7986

𝑣(𝑥, 0) = 𝑢0(𝑥), 𝑣𝑡(𝑥, 0) = 𝑣0(𝑥), 𝜕𝑣 4𝑣1,𝑗 − 𝑣2,𝑗 − 3𝑣0,𝑗 𝜕𝑣

− (0, 𝑡𝑗 ) = − , (𝐿, 𝑡𝑗 )
𝑥 ∈ 𝛺, (6) 𝜕𝑥 2∆𝑥 𝜕𝑥
𝑣(0, 𝑡) = 𝑝0(𝑡), 𝑣(𝐿, 𝑡) = 𝑝𝐿(𝑡), 3𝑣𝑀,𝑗 − 4𝑣𝑀−1,𝑗 + 𝑣𝑀−2,𝑗
= , 𝑗
(𝑥, 𝑡) ∈ 𝜕𝛺 × (0, 𝑇). (7) 2∆𝑥
Or using mixed boundary condition instead = ̅̅̅̅̅
1, 𝑁. (18)
equation (7)
𝑣𝑥 (0, 𝑡) = 𝑞0(𝑡), 𝑣(𝐿, 𝑡) = 𝑝𝐿(𝑡), The method for solving the second part for
(𝑥, 𝑡) ∈ 𝜕𝛺 × (0, 𝑇). (8) slit wave equation (9)-(12) is built on the separation
Part two: (𝑤, 𝑓) satisfies the ill-posed inverse of variables, (see equations (53)-(55) in (1)). This
problem (1,2,4) results (see (1,2))
𝑤𝑡𝑡(𝑥, 𝑡) = 𝑤𝑥𝑥(𝑥, 𝑡) + 𝑓(𝑥), 𝜕𝑤𝑘
𝑞0(𝑡) − 𝑣𝑥 (0, 𝑡) =: 𝑔(𝑡) = (0, 𝑡; 𝑏)
(𝑥, 𝑡) ∈ 𝛺 × (0, 𝑇), (9) 𝜕𝑥
𝐾
𝑤(𝑥, 0) = 𝑤𝑡(𝑥, 0) = 0, 𝑥 ∈ 𝛺, (10) √2 𝑏𝑘
𝑤(0, 𝑡) = 0, 𝑤(𝐿, 𝑡) = 0, = 2 ∑ (1 − cos(𝑐𝜆𝑘𝑡)), 𝑡
𝑐 𝜆𝑘
(𝑥, 𝑡) ∈ 𝜕𝛺 × (0, 𝑇). (11) 𝑘=1
𝑤𝑥(0, 𝑡) = 𝑞0(𝑡) − 𝑣𝑥(0, 𝑡), ∈ [0, 𝑇]. (19)
(𝑥, 𝑡) ∈ 𝛤 × (0, 𝑇), (12) And for (8) and (13) (see equations (78) and (79) in
And for using mixed boundary condition (8), (1))
equation (12) changed to (see (1)) 𝑝0(𝑡) − 𝑣(0, 𝑡) =: ℎ(𝑡)
𝐾
𝑤(0, 𝑡) = 𝑝0(𝑡) − 𝑣(0, 𝑡), √2 𝑏𝑘
(𝑥, 𝑡) ∈ 𝛤 × (0, 𝑇). (13) = 2 ∑ 2 (1 − cos(𝑐𝜆𝑘𝑡)), 𝑡
𝑐 𝜆
For the numerical solution of the part 𝑣 direct 𝑘=1 𝑘
problem using FDM (2,3,6). Split wave equation in ∈ [0, 𝑇]. (20)
two part, solving part one by FDM which is a new Consequently, replacing the exact data
work in this research. Divide the solution domain 𝑞0 (𝑡)|or 𝑝0 (𝑡) by the noisy data (see (1,2))
(0, 𝐿) × (0, 𝑇 ) into 𝑀 and 𝑁 subintervals of equal 𝑞0𝜖 (𝑡𝑛)|𝑝0𝜖 (𝑡𝑛) = 𝑞0 (𝑡)| 𝑝0 (𝑡) + 𝜖,
space length ∆𝑥 and time step ∆𝑡, respectively, 𝑛 = ̅̅̅̅̅
1, 𝑁, (21)
where ∆𝑥 = 𝐿/𝑀 and ∆𝑡 = 𝑇/𝑁. 𝑣𝑖,𝑗 ∶= 𝑣(𝑥𝑖 , 𝑡𝑗 ) where 𝜖 are 𝑁 (1-3) random noisy variables
is denoted, where 𝑥𝑖 = 𝑖∆𝑥, 𝑡𝑗 = 𝑗∆𝑡, for generated (using the Fortran NAG routine
𝑖 = ̅̅̅̅̅̅
0, 𝑀, 𝑗 = ̅̅̅̅̅
0, 𝑁 (2,3). Then, a central-difference G05DDF) from a Gaussian normal distribution with
approximation to equations (5)-(7) at the mesh mean zero and standard deviation 𝜎 given by (1-3)
points (𝑥𝑖 , 𝑡𝑗 ) = (𝑖∆𝑥, 𝑗∆𝑡) of the rectangular mesh 𝜎 = 𝑝% × 𝑚𝑎𝑥𝑡∈[0,𝑇] |𝑞0(𝑡)|or 𝑝0(𝑡)| , (22)
covering the solution domain (0, 𝐿) × (0, 𝑇) is 𝑝% is the percentage of noise. The noisy data (21)
(2,3), also makes noise in 𝑔|or ℎ (1-3) as:
𝑣𝑖,𝑗+1 = 𝑟 2 𝑣𝑖+1,𝑗 + 2(1 – 𝑟 2 )𝑣𝑖,𝑗 + 𝑟 2 𝑣𝑖−1,𝑗 𝑔𝜖 (𝑡𝑛)|ℎ𝜖 (𝑡𝑛) = 𝑞0𝜖 (𝑡𝑛)|𝑝0𝜖 (𝑡𝑛)
− 𝑣𝑥 (0, 𝑡𝑛 )|𝑣(0, 𝑡𝑛 ) = 𝑔(𝑡𝑛 )|ℎ(𝑡𝑛 )
− 𝑣𝑖,𝑗−1 , 𝑖 = ̅̅̅̅̅̅̅̅̅̅̅̅̅
1, (𝑀 − 1), 𝑗
+ 𝜖, 𝑛 = ̅̅̅̅̅
1, 𝑁. (23)
= ̅̅̅̅̅̅̅̅̅̅̅̅̅
1, (𝑁 − 1), (14) Then, apply the condition (19)|(20) with 𝑔|or ℎ
𝑣𝑖,1 − 𝑣𝑖,−1
𝑣𝑖,0 = 𝑢0 (𝑥𝑖 ), 𝑖 = ̅̅̅̅̅̅
0, 𝑀 , replaced by 𝑔𝜖 |or ℎ𝜖 then regularized by zero
2∆𝑡
= 𝑣0 (𝑥𝑖 ), 𝑖 order Tikhonov functional
𝑁 𝐾
= ̅̅̅̅̅̅̅̅̅̅̅̅̅
1, (𝑀 − 1), (15) √2 𝑏𝑘
𝑣0,𝑗 = 𝑃0 (𝑡𝑗 ), 𝑣𝑀,𝑗 = 𝑃𝐿 (𝑡𝑗 ), 𝑗 𝒥(𝑏) ∶= ∑[ 2 ∑ (1 − cos(𝑐𝜆𝑘𝑡𝑛 ))
𝑐 𝜆𝑘
𝑛=1 𝑘=1
= ̅̅̅̅̅
0, 𝑁, (16) 𝐾
where 𝑟 = ∆𝑡/∆𝑥. Putting 𝑗 = 0 in equation (14) −𝑔 𝜖 (𝑡 )]2 + 𝜆 ∑ 𝑏𝑘2 (24)
𝑛
and using (15), this will obtain (see (2,3)) 𝑘=1
1 Or,
𝑣𝑖,1 = 𝑟 2 𝑢0 (𝑥𝑖+1 ) + (1 – 𝑟 2 )𝑢0 (𝑥𝑖 ) 𝑁 𝐾
2 √2 𝑏𝑘
1 𝒥(𝑏) ∶= ∑[ 2 ∑ 2 (1 − cos(𝑐𝜆𝑘𝑡𝑛 ))
+ 𝑟 2 𝑢0 (𝑥𝑖−1 ) + (∆𝑡)𝑣0 (𝑥𝑖 ), 𝑖 𝑐 𝜆𝑘
2 𝑛=1 𝑘=1
= ̅̅̅̅̅̅̅̅̅̅̅̅̅
1, (𝑀 − 1). (17) 𝐾
∂v ∂v
The normal derivatives ∂x (0, t) and ∂x (L, t) are −ℎ 𝜖 (𝑡
𝑛 )]2 + 𝜆 ∑ 𝑏𝑘2 (25)
calculated using the finite-difference 𝑘=1
where 𝜆 ≥ 0 is a regularization parameter. (1-3)
approximations (see (2,3))
Note that: From equations (21)-(23), "|" means

676
Open Access Baghdad Science Journal P-ISSN: 2078-8665
2020, 17(2 Special Issue)NICST:675-681 E-ISSN: 2411-7986

"or". Second, mixed condition

𝑡2
𝑢(1, 𝑡) = 𝑝𝐿(𝑡) = 𝑡 + , 𝑢𝑥(0, 𝑡) = 𝑞0(𝑡) = 𝜋,
Numerical Results and Discussion: 2
This study, illustrates same numerical 𝑡 ∈ [0,1], (30)
example of (1,2), in order to see difference between and additional overdetermination
BEM and FDM. Take the one-dimensional case, 𝑡2
𝛺 = (0, 𝐿) with 𝐿 = 1, 𝑇 = 1 (2,3), consider an 𝑢(0, 𝑡) = 𝑝0(𝑡) = 𝑡 + , 𝑡 ∈ [0,1]. (31)
2
analytical solution with the input data given by (see The numerical results for (a) 𝑣𝑥 (0, 𝑡) (b) 𝑣(0, 𝑡),
(1,3)) presenting in Fig. 1, achieved by the FDM with
𝑡2 various 𝑀 = 𝑁 ∈ {20, 40, 80}. However, the
𝑢(𝑥, 𝑡) = sin(𝜋𝑥) + 𝑡 + , exact solution is not given, but it can see convergent
2
𝑓(𝑥) = 1 + 𝜋 2 sin(𝜋𝑥) , with various 𝑀and 𝑁. And compare this figure with
𝑥 ∈ [0,1], (26) Fig. 2 in (1) when boundary element method had
𝑢(𝑥, 0) = 𝑢0(𝑥) = sin(𝜋𝑥) , been used, which can be seen in Table 1, there is
𝑢𝑡(𝑥, 0) = 𝑣0(𝑥) = 1, difference when choosing Dirichlet boundary
𝑥 ∈ [0, 1], (27) condition as extra data. Then input the numerical
(a) First, Dirichlet condition solution (a) 𝑣𝑥(0, 𝑡) (b) 𝑣(0, 𝑡) at the points
𝑡2 (𝑡)𝑛=1,𝑁
̅̅̅̅̅ into equation (a) (19) to determine the
𝑢(0, 𝑡) = 𝑝0(𝑡) = 𝑡 + , values for 𝑔(𝑡𝑛) and its noisy counterpart 𝑔𝜖 (𝑡𝑛)
2
𝑡2 given by (23), (b) (20) to determine the values for
𝑢(1, 𝑡) = 𝑝𝐿(𝑡) = 𝑡 + , ℎ(𝑡𝑛) and its noisy counterpart ℎ𝜖 (𝑡𝑛) given by
2
𝑡 ∈ (0,1], (28) (23) for 𝑛 = ̅̅̅̅̅
1, 𝑁, respectively.
and additional condition
𝑢𝑥(0, 𝑡) = 𝑞0(𝑡) = 𝜋, 𝑡 ∈ [0,1], (29)

(a) (b)

Figure 1. (a) 𝐯𝐱(𝟎, 𝐭), (b) 𝐯(𝟎, 𝐭) come by using the FDM with 𝐌 = 𝐍 ∈ {𝟐𝟎, 𝟒𝟎, 𝟖𝟎}.
Table 1. (a) 𝐯𝐱(𝟎, 𝐭), (b) 𝐯(𝟎, 𝐭) come by using the BEM and FDM with 𝐌 = 𝐍 ∈ {𝟐𝟎, 𝟒𝟎, 𝟖𝟎}.
(a) 𝑣𝑥 (0, 𝑡)
𝑡𝑗 𝑀 = 𝑁 = 20 𝑀 = 𝑁 = 40 𝑀=𝑁 = 80
BEM FDM BEM FDM BEM FDM
0.2 2.50089 2.36232 2.42396 2.34681 2.38345 2.34291
0.4 0.82599 0.57872 0.69957 0.57279 0.63546 0.57130
0.6 -1.30763 -1.57872 -1.44003 -1.57279 -1.50564 -1.571304
0.8 -3.16143 -3.36232 -3.25401 -3.34681 -3.29844 -3.342907
(b) 𝑣(0, 𝑡)
0.2 0.1595 0.15959 0.1595 0.15948 0.1595 0.15946
0.4 0.0944 0.09407 0.0944 0.094376 0.0944 0.09441
0.6 -0.3339 -0.33483 -0.3339 -0.33401 -0.3339 -0.33391
0.8 -1.1255 -1.12690 -1.1255 -1.12566 -1.1255 -1.12551

From Fig. 2a one can note that the good approached. Also there is different shape
numerical solution 𝑏 = (𝑏𝑘 )𝑘=1,𝐾 ̅̅̅̅̅ get from (19) where boundary element method (BEM) had been
compare with analytical values for the sine Fourier used (see Figure 4 in (1)). From absolute error
series coefficients (see equation (75) in (1)) have a between exact and numerical in Table 2a shows

677
Open Access Baghdad Science Journal P-ISSN: 2078-8665
2020, 17(2 Special Issue)NICST:675-681 E-ISSN: 2411-7986

(FDM) is better than (BEM) where k = 5,10, but it the exact and numerical one is decried. Nearly the
is opposite in k = 15,20. same figure has been obtained when boundary is
Furthermore, Fig. 2b shows the 𝑏 = (𝑏𝑘 )𝑘=1,𝐾
̅̅̅̅̅ used element method (BEM) (1). But the values are
obtained from (20), in comparison with cosine closer to the exact one, when FDM has used one
Fourier series coefficients exact one when given by can see in Table 2b.
(equation (83) in (1)) the amount of error between
(a) (b)

Figure 2. The numerical solution (…) for (𝒃𝒌 )𝒌=𝟏,𝑲 ̅̅̅̅̅ for 𝑲 = 𝟐𝟎, 𝑵 = 𝟖𝟎 obtained, in comparison
with the exact solution (a) equation ((75) in (1)) (b) equation ((83) in (1)) (— ).

Table 2. The numerical solution for (𝒃𝒌 )𝒌=𝟏,𝑲

̅̅̅̅̅ for 𝑲 = 𝟐𝟎, 𝑵 = 𝟖𝟎 obtained using BEM and FDM,
in comparison with the exact solution (a) equation ((75) in (1)) (b) equation ((83) in (1)).
(a) 𝑏𝑘 In case 𝑣𝑥 (0, 𝑡)
𝑘 Exact BEM Error FDM Error
5 0.180063263 0.183231939 0.00316 0.1868537 0.00679045
10 0 -1.77E-02 0.0177 0.0053252 0.00532524
15 0.060021088 6.96E-02 0.00957 0.0796422 0.01962113
20 0 -8.95E-03 0.00895 0.0106504 0.01065049
(b) 𝑏𝑘 In case 𝑣(0, 𝑡)
5 -0.13076 -0.1317 0.0009360 -0.1310896 0.00032572
10 -9.72E-02 -0.0962 0.0009652 -0.0976859 0.00052070
15 0.009812 0.0056 0.0042129 0.00695022 0.00286274
20 -3.48E-02 -0.0346 0.0001999 -0.0368118 0.00201184
Give the numerical solutions 𝑓(𝑥) from obtained for both FDM and BEM (see Figure 14
equations ((54) and (79) in (1)). Note that: They are (1)), and they are clear in Table (3). Moreover, the
the same formula but different method has been figure of the numerical and exact solution for
used for finding 𝑏𝑘 , it is FDM. Then 𝑓(𝑥) are 𝑢(𝑥, 𝑡) and condition number of a matrix see (1) are
plotted in Fig. 3(a) and 3(b), respectively. It is clear not showed in this study, the reason for that: nearly
from those figures that good approaches are I have gotten the same shape for u(x, t) and used
achieved between numerical and exact solutions. In the same matrix (Q (3)). In this study the method
3(a) more accuracy has been obtained in boundary has been changed for solving v instead applied
element method (BEM) rather than FDM (see BEM apply FEM.
Figure 5 (1)), and 3(b) nearly same figures are
(a) (b)

Figure 3. The exact solution (26) for 𝒇(𝒙) in comparison with the numerical solution, (a)
equation ((54) in (1)) and (b) equation ((79) in (1)) for 𝑲 ∈ {𝟓, 𝟏𝟎, 𝟐𝟎} respectively.

678
Open Access Baghdad Science Journal P-ISSN: 2078-8665
2020, 17(2 Special Issue)NICST:675-681 E-ISSN: 2411-7986

Table 3. The exact solution (26) for 𝒇(𝒙) in comparison with the numerical solution BEM and FDM,
(a) equation ((54) in (1)) and (b) equation ((79) in (1)) for 𝑲 = 𝟐𝟎.
𝑓𝑘=20 In case 𝑣𝑥 (0, 𝑡)
𝑀 = 𝑁 = 80 𝑓(𝑥),exact 𝑓(𝑥), BEM Error 𝑓(𝑥),FDM Error
0.2 6.801207913 6.582675008 0.218532905 6.709727966 0.091479947
0.4 10.38655158 10.28346389 0.103087691 10.33671287 0.04983871
0.6 10.38655158 10.3991275 0.012575923 10.34160683 0.04494475
0.8 6.801207913 6.879010004 0.077802091 6.730459123 0.07074879
𝑓𝑘=20 In case 𝑣(0, 𝑡)
0.2 6.801207913 6.7895 0.011707913 6.788235668 0.012972245
0.4 10.38655158 10.3705 0.01605158 10.36905097 0.01750061
0.6 10.38655158 10.3639 0.02265158 10.36180593 0.02474565
0.8 6.801207913 6.7581 0.043107913 6.754137179 0.047070734

Figure 4 and 5 have plotted the numerical solutions instabilities as 𝑝% = 1%. This means, in this case,
for 𝑓(𝑥) for various values of 𝐾 ∈ {5, 10, 20} and FDM is a better solution rather than BEM. And
contain some (𝑝% = {1, 3, 5}%) noise in the data when the amount of noise increase 𝑝% = {3, 5}%
(21). First, in case (a) Dirichlet condition: From Fig. it can be clearly seen that oscillations start to appear
4(a) it can be observed that there is little difference see figures 4(b) and 4(c). Fig. 4(d) shows
between the exact solution (26) and the results for instabilities as 𝑝 increases and fixed 𝐾 = 20.
𝑓(𝑥) obtained with and without noise (where Meanwhile, in case (b) mixed condition: Fig. 5
𝑝% = 1%). In contrast, in (Figure 7 (1)) the oscillations start to appear as 𝐾 increases to 10 or
unregularized numerical solution for 𝑓(𝑥) 20 with noise (𝑝% = 1%.).

(a) (b)

(c) (d)

Figure 4. Comparison between exact solution (26) for 𝐟(𝐱) and numerical solution (54) in (1) for
various 𝐊 ∈ {𝟓, 𝟏𝟎, 𝟐𝟎}, no regularization, for (a) 𝐩% = 𝟏% (b) 𝐩% = 𝟑% (c) 𝐩% = 𝟓% (d)
𝐩% = {𝟏, 𝟑, 𝟓}% and 𝐊 = 𝟐𝟎 noisy data.

679
Open Access Baghdad Science Journal P-ISSN: 2078-8665
2020, 17(2 Special Issue)NICST:675-681 E-ISSN: 2411-7986

In order to deal with this instability, the

zero order Tikhonov regularization has been
employed, see (1-3). Fix 𝐾 = 20 and to stabilise of
the numerical solution for 𝑓(𝑥) shown by (—△— )
in Fig. 4(d) and 5 obtained with no regularization,
i.e. 𝜆 = 0, for 𝑝% = 5% and 𝑝% = 1% noisy
data, respectively. First, to choose suitable
regularization parameter 𝜆, using error between
exact and numerical solution for 𝑓(𝑥), from Fig.
6(a) and 6(b) can be seen that the minimum of the
error occurs around 𝜆 = 10−1 (the same amount in
using BEM (1) and difference shape compare to
Figure 5. Comparison between exact solution Fig. 9 in (1) and 𝜆 = 5 × 10−4 (difference
(26) for 𝐟(𝐱) numerical solution (79) in (1) for amount in using BEM (1)), respectively. From Fig.
various 𝐊 ∈ {𝟓, 𝟏𝟎, 𝟐𝟎}, no regularization, for 7 those amounts of 𝜆 are optimal one.
𝐩% = 𝟏% noisy data.

(a) (b)

Figure 6. The accuracy error ||𝒇numerical – 𝒇exact||, as a function of 𝝀, for 𝑲 = 𝟐𝟎 and (a) 𝒑% =
𝟓% (b) 𝒑% = 𝟏% noise.
(a) (b)

Figure 7. Comparison between exact solution (26) for 𝒇(𝒙) and numerical solution, for 𝑲 = 𝟐𝟎, (a)
𝒑% = 𝟓% noise, and regularization parameters 𝝀 ∈ {𝟏𝟎−𝟏 , 𝟏𝟎𝟎 } (b) 𝒑% = 𝟏% noise, and
regularization parameters 𝝀 ∈ {𝟏𝟎−𝟒 , 𝟓 × 𝟏𝟎−𝟒 , 𝟏𝟎−𝟑 }.

Conclusions: causes large errors in the output solving force. The

In this study, the solving wave equation by problem was divided into two parts, the part one
using FDM and separation variable has been discretised using the FDM, and next part using
applied. Meanwhile, some comparison between separation variable (1,2), in addition for
FDM and BEM (1) has been mentioned in Section regularization and stabilize, the zero order is
3. The inverse problem is ill-posed since small employed Tikhonov regularization method (1,2).
errors in the input addition boundary condition The choice of the regularization parameter was

680
Open Access Baghdad Science Journal P-ISSN: 2078-8665
2020, 17(2 Special Issue)NICST:675-681 E-ISSN: 2411-7986

found by the minimum error method, it is clear and References:

from plots as well. The corresponding using FDM 1. Hussein SO, Lesnic D. Determination of a space-
and separation variable with different boundary dependent source functions in the one-dimensional
condition will be investigated in Part II, (1-3). wave equation. EJBE. 2014; 12: 1-26.
2. Hussein SO. Inverse force problems for the wave
equation. PhD Thesis, The University of Leeds
Author's declaration: Department of Applied Mathematics; 2016
- Conflicts of Interest: None. 3. Hussein SO, Lesnic D. Determination of forcing
- I hereby confirm that all the Figures and Tables functions in the wave equation. Part I: the space-
in the manuscript are mine. Besides, the Figures dependent case. J. Eng. Math. 2016; 96: 115-133.
and images, which are not mine, have been given 4. Cannon JR, Dunninger DR. Determination of an
the permission for re-publication attached with unknown forcing function in a hyperbolic equation
the manuscript. from overspecified data. ANN MAT PUR APPL.
- Ethical Clearance: The project was approved by 1970; 1: 49-62.
the local ethical committee in University of 5. Colton D, Kress R. Inverse Acoustic and
Electromagnetic Scattering Theory. 3rd ed. New
Sulaimani. York: Springer Verlag; 2013.
6. Morse PM, Feshbach H. Methods of Theoretical
Physics. New York: McGraw-Hill; 1953.

‫تقسيم معادلة الموجة واحدة االبعاد‬

‫ استخدام طريقة الفروق المحدودة مع فصل المتغيرات‬:‫جزء االول‬
‫شيالن عثمان حسين‬

.‫ العراق‬،‫ السلمانية‬،‫ الجامعة السليمانية‬،‫ كلية العلوم‬،‫قسم الرياضيات‬

:‫الخالصة‬
‫ تم التحقيق فى دالة قوة مجهولة تعتمد على الفضاء في معادلة الموجة حيث تم تقسيم معادلة الموجة عدديا الى‬،‫في هذه الدراسة‬
‫ ) فى القسم االول واستخدام طريقة فصل المتغيرات فى القسم الثاني وقد اجرى تغييرا فى‬MDF( ‫قسمين وتم استخدام طريقة الفروق المحدودة‬
‫) بدالً من طريقة‬MDF( ‫) حيث تم تطبيق طريقة الفروق المحدودة‬2,1( ‫تقنية الحل للقسم المتعلق بالمسآلة العكسية المذكورة فى‬
‫ وتم توظيف البيانات الحدودية لخدمة البيانات ما فوق المحدودة وان القسم الثاني للمسألة‬.) 2,1( ‫( الواردة فى‬BEM) ‫العناصرالحدودية‬
‫هوعكسى وفى وضع معتل الن االخطاء الصغيرة فى البيانات الحدودية االضافية تسبب اخطاءأ فى حل المسألة حيث وظفت الرتبة الصفرية‬
‫لنظام تيكونوف وكذلك باراميترات ( معلمات) عديدة للنظام لتقليل االخطاء من اجل الحصول على حل القوة ومن الواضح من الرسوم البيانية‬
.‫كيف ان الخطأ يوء ثر على النتيجة وكيف ان الرتبة الصفرية تعمل على استقرار الحل‬

.‫ معادلة الموجة‬،‫ مسألة القوة العكسية‬،‫ طريقة فصل المتغيرات‬،‫ طريقة الفروقات المحدودة‬،‫ التنظيم‬:‫الكلمات المفتاحية‬

681