c c
=
c I c
}
(9)
2
2 1
0
( , ) 1 ( , )
.
(2 ) ( )
t
y x t y t
d
x x x
|
| |
|
c c
=
c I c
}
(10)
Suppose that Eqs. (6)-(8) has a unique and smooth
enough solution. / T n t = is time step, / x h L m A = = is
space step, ( 0,1, 2, , ),
k
t k k n t = = L
( 0,1, 2, ,
i
x ih i = = L ). m For the time fractional
derivative, we usually adopt following finite difference
approximation
1
( , )
i k
y x t
t
o
o
+
c
c
( 1)
1
0
1
( , ) ( , )
1
(1 ) ( )
k
j
i j i j
j
j
k
y x t y x t
d
t
t
o
t
t
o t
+
+
=
+
+
I
=
}
1
1 1
0
( , ) ( , )
[( 1)
(2 )
k
i k j i k j
j
y x t y x t
j
o
o
t
o t
+
=
= +
I
1
] ( ), j O
o
t
+
(11)
1 1 1 1
( , ) ( , ) ( , )
( ).
i k i k i k
y x t y x t y x t
O h
x h
+ + +
c
= +
c
(12)
For
( , ) y x t
x
|
|
c
c
, by using Grnwalds improved formula
[4], we have that
1
1
1
0
( , ) 1
( ( 1) , ) ( ),
i
i k
j i k
j
y x t
a y x j h t O h
x h
|
| |
t
+
+
+
=
c
= + +
c
(13)
where
( )
0 1
1, , ( 1) ,
j
j j
a a a
|
| = = =
( )
, ( 1) ( 1) / ( !)
j
j j
|
| | | = + L 1, 2, 3, . j = L
Substituting Eqs. (11)-(13) into Eqs. (6)-(8), we obtain
1
1 1 1
0
( , ) ( , )
[( 1) ]
(2 )
k
i k j i k j
j
y x t y x t
j j
o
o o
t
o t
+
=
+
I
1
1 1
0
( )
( ( 1) , ) - ( ) ( , )
i
i
j i k i i k
j
u x
a y x j h t c x y x t
h
|
+
+ +
=
=
1 -1 1
1
( , ) - ( , )
- ( ) ( , ) ( ).
i k i k
i i k
y x t y x t
v x b x t O h
h
t
+ +
+
+ + +
Let
%
( ), ( ), ( ), (2 ) 0,
i i i i i i i
u u x v v x c c x c
o
t o = = = = I >
( , ),
k
i i k
b b x t =
1 1
( 1) ,
j
j j
o o
o
= + 0,1, 2, , ; j n = L
i
r =
1 1 2
(2 ) 0, (2 ) 0, ( ),
k k i i
i k
u v
p g g t g
h h
o o
|
t t
o o I > = I > =
2
( ),
k
g t =
k
i
y represents numerical solution of Eqs. (6)-(8),
then we have
1
1 1 1 1
1 1
0 0
( ) ( )
k i
k j k j k k k
j i i i i i i j i j
j j
y y p y y r a y o
+
+ + + +
+
= =
+ =
% %
1 1
.
k k
i i i i
c y c b
+ +
+
(14)
Since local truncation error is ( ), O h t + thus the
difference scheme is consistent[10]. Eqs. (14) will be
repalced by
When 0, k =
%
1
1 1 1 1
1 1 2 1 1
3
(1 ) ( )
i
i i i i i i i i i i j i j
j
r y p ra c y p ra y r a y
+
+ +
=
+ + + +
%
0 1
,
i i i
y c b = +
(15)
when 0, k >
%
1
1 1 1 1
1 1 2 1 1
3
(1 ) ( )
i
k k k k
i i i i i i i i i i j i j
j
r y p ra c y p ra y r a y
+
+ + + +
+ +
=
+ + + +
%
1 1 0 1
0
( ) (2 2 )
k
k k i k j k k
i j i i i k i i i
j
y y y y y c b
o
o o
+ +
=
= = + +
1
1 1 1
1
[2( 1) ( 2) ]
k
k j
i
j
y j j j
o o o
=
+ +
%
1
0 1
1 1
1
,
k
k k j k
i i j k i i i
j
d y y d y c b o
+
+
=
= + +
(16)
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 2, No. 8, 2011
78 | P a g e
www.ijacsa.thesai.org
where
1
,
j j j
d o o
= 1, 2, , 1; 1, 2, , 1. i m k n = = L L
III. STABILITY AND CONVERGENCE OF THE DIFFERENCE
SCHEME
Lemma 2.1 For arbitrary real number , , , a b c d , we have
| | | | | | | | | | . a b c d a b c d + s +
Proof. From | | | | b a b c d a c d = + + + +
| | | | | | | |, a b c d a c d s + + + +
we obtain | | | | | | | | | | . a b c d a b c d + s +
Lemma 2.2 (1) 0( 2).
j
a j > > (2) For any positive
integer N we have
0
0.
N
j
j
a
=
<
Proof. (1) Note that
( )
( 1)
j
j j
a
|
= and 1 2, | < < we
know that 0( 2).
j
a j > >
(2) Since
( )
0
(1 ) , [ 1,1],
j
j
j
x x x
| |
=
+ = e
let 1, x =
then
0
0.
j
j
a
=
=
= = +
< =
Lemma2.3
1
(1) 1 ;
n
j n
j
d o
=
=
1
(2) 0, .
j j j
d o o
> >
Proof. (1)From
1 1
1
, ( 1) ,
j j j j
d j j
o o
o o o
= = +
we easily know that
1
1 .
n
j n
j
d o
=
=
(2) Let
1 1
( ) ( 1) ( 1), h x x x x
o o
= + > then '( ) h x =
(1 )[( 1) ] 0, x x
o o
o
+ < so ( ) h x is decreasing
function,
1
( 1) ( ) 0.
j j j
d h j h j o o
= = > Therefore,
we have that
1
0, .
j j j
d o o
> >
A. Stability of the difference scheme
Theorem 2.1 Implicit difference schemes defined by Eqs.
(15)-(16) are unconditionally steady for initial value[10].
Proof. Suppose that ,
k
k
i i
y y
%
represent solutions of Eqs.
(15)-(16) for initial value
1 2
( ), ( ) f x f x respectively, and
k
i
b
is accurate vale, then error
k
k k
i i i
y y c =
%
satisfies
When 0, k =
%
1 1 1
1 1 2 1
(1 ) ( )
i i i i i i i i i
r p ra c p ra c c c
+
+ + + +
1
1 0
1
3
,
i
i j i j i
j
r a c c
+
+
=
=
when 0, k >
%
1 1 1
1 1 2 1
(1 ) ( )
k k k
i i i i i i i i i
r p ra c p ra c c c
+ + +
+
+ + + +
1 1
1 0
1 1 1
3 1
.
i k
k k k j
i j i j i j i k i
j j
r a d d c c c o c
+
+
+ +
= =
= + +
Let
'
1 2 1
( , , , ) ,
k k k k
m
E c c c
= L we prove
0
|| || || ||
k
E E
s with mathematical induction in the
following.
When 1, k = let
1 1
1 1
| | max | | .
l i
i m
c c
s s
= Note that 0,
l
r >
0,
l
p >
0 1
1, 0, a a | = = < we have from Lemma 2.2 that
1
1 1 1 1 1 1
1
0
|| || | | | | (| | | |) | |
l
l l l l l l j l
j
E p r a c c c c c
+
=
= s +
%
1 1
1 1
| | (1 ) | |
l l l l i l
r p ra c c c
+
s + + +
1
1 1
2 1 1
3
( ) | | | |.
l
l l l l j l j
j
p r a r a c c
+
+
=
+
Note that
%
1
2 1
3
0, 0,1 0,
l
l l l j l l i
j
p r a r a p r a c
+
=
+ > > + + >
and from Lemma 2.1, we further obtain that
%
1 1 1 1
1 1 2 1
| | | (1 ) ( )
l l l l l i l l l l
r p ra c p ra c c c c
+
s + + + +
1
1 0 0
1
3
| | | || || .
l
l j l j l
j
r a E c c
+
+
=
= s
Thus
1 1 0
|| || | | || || .
l
E E c
= s
Assume that we always have
0
|| || || ||
k
E E
s when
, k s s let
1 1
1 1
| | max | |,
s s
l i
i m
c c
+ +
s s
= then when 1, k s = + we
have
%
1 1 1
1 1
| | | | (1 ) | |
s s s
l l l l l i l
r p ra c c c c
+ + +
+
s + + +
1
1 1
2 1 1
3
( ) | | | |
l
s s
l l l l j l j
j
p r a r a c c
+
+ +
+
=
+
%
1 1
1 1
| (1 )
s s
l l l l i l
r p ra c c c
+ +
+
s + + +
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 2, No. 8, 2011
79 | P a g e
www.ijacsa.thesai.org
1
1 1
2 1 1
3
( ) |
l
s s
l l l l j l j
j
p r a r a c c
+
+ +
+
=
+
1
0
1 1
1
| |
s
s s j
l j l s l
j
d d c c o c
+
=
= + +
1
0
1 1
1
|| || || || || ||
s
s s j
j s
j
d E d E E o
+
=
s + +
1
0
1 1
1
( ) || ||
s
j s
j
d d E o
+
=
s + +
0 0
(1 ) || || || || .
s s
E E o o
= + =
Consequently, the desired result follows.
B. Convergence of the difference scheme
Suppose that ( , )
i k
y x t is exact solution of the
differential equation at ( , ).
i k
x t Let ( , ) ,
k k
i i k i
e y x t y =
1 2
( , , ,
k k k
e e e = L
'
1
) ,
k
m
e
then ( , ) ,
k k
i i k i
y y x t e =
substituting it into Eqs. (15)-(16), and note
0
0, e = we have
that
When 0, k =
%
1 1 1
1 1 2 1
(1 ) ( )
i i i i i i i i i
re p ra c e p ra e
+
+ + + +
1
1 1
1
3
,
i
i j i j i
j
r a e R
+
+
=
=
when 0, k >
%
1 1 1
1 1 2 1
(1 ) ( )
k k k
i i i i i i i i i
re p ra c e p ra e
+ + +
+
+ + + +
1 1
1 1
1 1 1
3 1
,
i k
k k k j k
i j i j i j i i
j j
r a e d e d e R
+
+ +
+ +
= =
= + +
where
1
| | ( ),
k
i
R h
o o
t t
+
s + is a positive constant and it
has nothing to do with , . h t
Theorem 2.2 There is a constant 0 > such that
1 1
1
|| || ( ), 1, 2, , ,
k
k
e h k n
o o
o t t
+
s + = L
where has nothing to do with , , h t and || ||
k
e
=
1 1
max | | .
k
i
i m
e
s s
Proof. When 1, k = let
1 1
1 1
| | max | |,
l i
i m
e e
s s
= then
1
|| || e
=
1
| |,
l
e we have from Lemma 2.1 that
%
1 1 1
1 1
| | | | (1 ) | |
l l l l l i l
e r e p ra c e
+
s + + +
1
1 1
2 1 1
3
( ) | | | |
l
l l l l j l j
j
p r a e r a e
+
+
=
+
%
1 1 1
1 1 2 1
| (1 ) ( )
l l l l i l l l l
re p ra c e p ra e
+
s + + + +
1
1 1
1
3
| | |
l
l j l j l
j
r a e R
+
+
=
=
1
( ) h
o o
t t
+
s +
1 1
0
( ). h
o o
o t t
+
= +
Assume that
1 1
1
|| || ( )
s
s
e h
o o
o t t
+
s + when , k s s
let
1
| |
s
l
e
+
=
1
1 1
max | |,
s
i
i m
e
+
s s
then when 1, k s = + we have that
1 1
|| || | |
s s
l
e e
+ +
=
1
1
1 1
1
|| || || || ( )
s
s s j
j
j
d e d e h
o o
t t
+
+
=
s + + +
1 1 1 1
1 1 2 2 0
(1 ) ( ),
s s s
d d d h
o o
o o o t t
+
s + + + + + L
as in Lemma 2.3, we have found that
1 1
( ),
j s
j s o o
s s so
we futher obtain
1
|| ||
s
e
+
1
1 1
1
0
( ) ( )
s
s i s
i
d h
o o
o o t t
+
+
=
s + +
1 1
( ).
s
h
o o
o t t
+
= +
The desired result follows.
Since
1
1
1 1
lim lim lim
[(1 1/ ) 1]
k
k k k
k
k k k k
o o o
o
o
= =
+
=
1
,
1 o
thus there is a constant 0 > such that
1
|| || ( ) ( ) ( ).
k
E k h k h
o o o o
t t t t
+
s + = +
Consequently, we can obtain the following result when
k T t s .
Theorem 2.3 Suppose that ( , )
i k
y x t denotes exact
solution at ( , ),
i k
x t
k
i
y is numerical solution of implicit
difference scheme, then there exists a constant 0 T
o
= >
%
such that
| ( , ) | ( ),
k
i k i
y x t y h t s +
%
1 ,1 . i m k n s s s s
IV. SOURCE TERM INVERSION
The inverse problem, which is composed of Eqs. (2)-(5),
is to solve nonlinear operator equation
[ ( )] ( ). A b x x =
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 2, No. 8, 2011
80 | P a g e
www.ijacsa.thesai.org
Suppose that ( ( ); , ) y b x x t denotes solution of Eqs. (2)-
(4) for ( ), b x
0
( ) b x denotes a function near
*
( ), b x where
0
( ) b x =
0 T
0
1
( ) ( ),
n
i i
i
k x K x
=
= +
and
*
( ) b x denotes exact
solution of Eqs. (2)-(4), ( ) , b x K e
1 2
( ), ( ), , x x L are a
group of primary functions on K , then a tiny disturbing
quantity of
0
( ) b x is
0 T
0 0
1
( ) ( ) ( )
n
i i
i
b x k x K x o o o
=
= = +
(17)
where
1 2
( ) ( ( ), ( ), , ( )) ,
T
n
x x x x + = L
1 2
( , , ,
T
K k k = L ) .
n
n
k R e
Assume that
1
( ( ); , ) y b x x t denotes the solution of initial
boundary value problem for
1
( ), b x where
1 0
( ) ( ) b x b x =
0
( ), b x o + using Taylor formula, then we have[6]
0 0
( ( ) ( ); , ) y b x b x x t o +
0
0 0 0 0
( ( ); , ) ( ( ); , ) (|| ( ) ||),
T
K
y b x x t y b x x t K o b x o o = +V +
by using the Tikhonov regularization method, solving ( ) b x
is converted into
0
K o , and
0
K o can be determined by local
minimum of the following function[9]
2 '
2
0 0 0
( [0, ])
[ ]=|| ( ( ) ( ); , ) ( , ) ||
L T
G K y b x b x x t y x T o o
cO
+
0
[ ] S K o o +
0
=|| ( ( ); , ) ( , ) y b x x t y x T +
2 '
0
2
0 0 0
( [0, ])
( ( ); , ) || [ ],
T
K
L T
y b x x t K S K o o o
cO
V +
(18)
where
'
, xecO ccO o denotes regularization coefficient,
0
[ ] S K o denotes steady functional of
0
. K o
Assume that there are discrete points ( 1, 2, ,
m
x m = L
) M on
'
, cO
0 0 0
[ ]= ,
T
S K K K o o o then
0 0 0 0
[ ]= 2 ( )
T T T T
G K K A A K K A P Q o o o o +
0 0
( ) ( ) .
T T
P Q P Q K K oo o + +
(19)
It is easy to prove that solving the local minimum values
of Eqs. (19) is equivalent to slove
0
( )
T T
A A I K A o o + =
( ), Q P so we have
1
0
( ) ( ),
T T
K A A I A Q P o o
= +
(20)
where
0 1
0 2
0
( ( ); , )
( ( ); , )
,
( ( ); , )
M
y b x x T
y b x x T
P
y b x x T
=
(
(
(
(
(
M
1
2
( )
( )
,
( )
T
T
T M
y x
y x
Q
y x
=
(
(
(
(
(
M
( ) ,
ij M N
A a
=
0
( ( ); , ).
ij i
j
a y b x x T
k
c
=
c
Substituting Eqs. (20) into Eqs. (17), we can obtain
0
( ) b x o and a new approximation of the exact solution,
1 0 0
( ) ( ) ( ). b x b x b x o = +
Repeating the above, until satisfies the precision
requirement.
V. NUMERICAL EXPERIMENTS
In order to verify the effectiveness of the algorithm in the
source term identification, we do the following numerical
experiment[7]. For simplicity, we set part variables as
follows
( ) 0.05, ( ) 292, ( ) 365, 4000, 11, c x u x v x L T = = = = =
20, 11, [0.01, 0.01, 0.01]. m n = = A =
Where A is the increment of K when computing matrix A
from Eqs. (20). Moreover, we always take polynomial
function space as primary function space in the following
computation, and setting initial boundary condition as
follows
( ) 0.057 45.6, 0 4000,
T
y x x x = + s s
2
1
( ) 0.724 45.6, 0 11, g t t t = + s s
2
2
( ) 2.2 273.4, 0 11. g t t t = + s s
Let ( ) 1 b x x = in the Eqs. (6)-(8), and substituting
initial boundary condition, we can obtain ( , ) y X T by
sloving forward problem. And as the additional data, we can
do inversion calculation by using the above algorithm. Let
initial iteration vector
0
[1,1,1], K = and iterative termination
condition ( ) 1 4, b x e o < then we obtain inversion results
under different regularization coefficient(see TABLE), let
1 3, theta e = we get inversion results under different initial
value(see TABLE ).
TABLE . THE INVERSION RESULTS UNDER DIFFERENT REGULARIZATION
COEFFICIENT
u
Iteration times Results
(IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 2, No. 8, 2011
81 | P a g e
www.ijacsa.thesai.org
1e-1
1e-2
1e-3
1e-4
4
3
3
3
1.0000 -1.0000 0
1.0000 -1.0000 0
1.0000 -1.0000 0
1.0000 -1.0000 0
TABLE . THE INVERSION RESULTS UNDER DIFFERENT INITIAL VALUE
0
K
Iteration
times
Results
-100 -100 -100
-10 -10 -10
1 1 1
10 10 10
100 100 100
4
3
3
3
4
1.0000 -1.0000 0
1.0000 -1.0000 0
1.0000 -1.0000 0
1.0000 -1.0000 0
1.0000 -1.0000 0
To better simulate the errors generated by actual data,
and verify the effectiveness of the algorithm, we choose the
disturbance error (1 ), V V