Beruflich Dokumente
Kultur Dokumente
3.1 Introduction
An extension of the procedure IRSS1 (see Monsi (1988)) is the interval repeated
midpoint symmetric single-step procedure IRMSS1. This procedure is considered as a
new procedure which is based on the interval midpoint symmetric single-step procedure
IMSS1 (see Rusli (2011)) and it can be called as the repeated version of IMSS1. The
idea is to introduce inner iteration to the procedure IMSS1. It also can be achieved by
using the updated midpoint in every step in the IRSS1 procedure.
the sequences
(k)
X i (i=1, , n) from
X (1,0)
= X(0)
i
i (3.2 .1 a)
For ( l=1, , m )( k 1 ) ;
x(ki ,2 l2) =m ( X (ik , 2l 2 ) ) ( i=1, ,n ) ,(3.2 .1 b)
( k ,2 l1)
i
= x
( k ,2 l2 )
i
i1
p ( x (ik , 2l 2 ))
n
j=i+1
(k ,2 l)
i
= x
( k , 2l1 )
i
i1
p ( x(ik ,2 l1) )
n
j=i+1
,2 l1)
X (k
( i=n , ,1 ) ,(3.2 .1e )
i
The above procedure consists of two steps of calculating the new intervals in every
inner iteration. The value of inner iteration depends on the value of
where we can
set the value of at the beginning of the procedure. In this research, for IRMSS1
procedure, three values of inner iteration m are set initially and to be tested numerically.
The values are m=2, m=3 and m=4 . Among these three values of m , the best
value of m that gives us the most outstanding result is chosen. In addition, in this
procedure, each time before we compute new intervals in the step (3.2.1e), the new
midpoints
( k ,2 l1)
Xi
35
Theorem 3.3.1
If (i)
x i X (i0) (i=1, , n)
( k)
sequences (X i ) ( i=1, , n )
and
are
generated
from
(3.2.1),
then
0 Di I ( R ) is such that
di
1
( k +1)
w ( X (ik ) ) and
p' ( x ) D i ( x X (i0) ) ( i=1, , n ) , then ( b ) w ( X i ) 1
2
di
( )
I
( k)
Finally
(c)
for
( i=1, , n ) ,O R ( IRMSS 1, xi ) 4 m.
Proof
The proof of (a) and (b) are almost identical with the corresponding proofs in Alefeld
and Herzberger (1983). The proofs are as follow:
36
p
Let
be defined by
p ( x )= (xx j )
j=1
(1,2l 1 )
i
=x
( 1,2l 2 )
i
p( x (i1,2l2) )
i1
j=1
j =1 +1
( X (i1,2l2) ) (3.3.1)
( i=1, n ) ,
l1)
x(1,2
=midpoint ( X (i1 ,2 l1) )
i
p (x (i1,2l1))
)
X (1,2l
= x(i1,2l1) i1
i
j=1
j=1+1
( i=n , , 1 ) .
( 1)
Let X i ( i=1, , n ) be defined by;
For l=1 ,
x(1,0)
=midpoint ( X (i1,0 ) )
i
X (1,1)
= x (i1,0 ) i1
i
p (xi(1,0) )
j=1
j=1+1
( i=1, n ) ,
37
x(1,1)
=midpoint ( X (i1 , 1) ) ( i=1, ,n )
i
(1,2)
i
= x
( 1,1)
i
p(x (i1,1))
i 1
j=1
j=1+1
( i=n , , 1 ) .
( 1)
( 1,1)
x i X i ( i=1, , n ) , because x i X i (i=1, , n ) .
By (3.3.4),
X
so,
(1,2)
n
= x
( 1,1)
n
n1
p ( x (n1,1 ) )
} {
X
( x (n1,1 )X (j1,1) )
j=1
( 1,1)
n
( 1,1)
n
n 1
p ( x (n1,1 ) )
=x n ,
( x (i1,1 )x j )
j=1
( 1,2 )
(1,2 )
x n X n . Suppose that for some i^ n , xi X i ( i=n , , i^ ) . Then by (3.3.4),
(1,2)
^
i1
= x
( 1,1)
^i1
i2
^
1,1 )
p ( x (i1
^
)
1,1 )
1,1 )
)
X (j1,1) ) ( x (^i1
X (1,2
( x (i1
^
j )
^
j=1
j= i
} {
X
( 1,1)
^i1
( 1,1)
i^ 1
i2
^
( 1,2)
x i X i (i=n , , 1).
x(1,2)
=midpoint ( X (i1,2 ) )
i
= x
( 1,2 )
i
i1
p (xi(1,2) )
n
j=1
j=i+1
38
1,1 )
1,1 )
x j ) ( x (^i1
x j )
( x (^i1
^
j=1
(1,3)
i
1,1 )
p ( x (i1
^
)
( X (i1,2 )) (3.3.5)
j=i
=x ^i1 .
( i=1, n ) ,
x(1,3)
=midpoint ( X (i1 , 3) ) ( i=1, , n )
i
p(x (i1,3 ))
X (1,4)
= x (i1,3 ) i1
i
j=1
j=i +1
( i=n , , 1 ) .
( 1,3)
xi Xi
By (3.3.6),
so,
(1,4)
n
= x
( 1,3 )
n
n1
p ( x (n1,3 ))
} {
X
( x(n1,3) X (j1,3 ) )
j=1
( 1,3)
n
( 1,3)
n
n1
p ( x (n1,3 ) )
( x (i1,3 )x j )
j=1
=x n ,
(1,4 )
( 1,4 )
x n X n . Suppose that for some i^ n , xi X i ( i=n , , i^ ) . Then by (3.3.6),
(1,4)
^
i1
= x
( 1,3 )
^i1
1,3 )
p ( x (^i1
)
i2
^
1,3)
1,3)
X (j1,3 ) ) ( x(i1
X (j1,4 ) )
( x(i1
^
^
^
j=1
j= i
} {
X
( 1,3 )
^i1
( 1,3 )
^i1
^i2
( 1,4 )
x i X i (i=n , ,1).
l)
x(1,2
=midpoint ( X (i1,2 l ))
i
= x
( 1,2 l )
i
i1
p(x (i1,2l ))
j=1
j=i+1
39
1,3 )
1,3)
x j) ( x (i1
x j)
( x (i1
^
^
^
j =1
(1,2l +1)
i
1,3 )
p ( x (^i1
)
j=i
= xi1
.
^
( i=1, n ) ,
l+1)
x(1,2
=midpoint ( X (i1 , 2l +1) )
i
p ( x (i1,2l+1 ))
+2)
X (1,2l
= x (i1,2 l+ 2) i1
i
l +1)
(x(i1,2l+1 )X (1,2
) (x (i1,2l +1)X (j1,2 l+2 ))
j
j=1
j =i +1
( i=n , , 1 ) .
( 1,2l+1)
xi X i
( 1,2l )
xi Xi
( i=1, ,n ) because
(i=1, , n ) .
By (3.3.8),
X
so,
(1,2l +2)
n
= x
( 1,2 l+1 )
n
j=1
xn X n
} {
X
n1
p ( x (n1,2l +1) )
( 1,2 l+1)
n
( 1,2l+1 )
n
n1
p ( x(n1,2l+1) )
( x (i1,2 l +1)x j )
j=1
l +2)
( i=n , , i^ ) .
i^ n , xi X (1,2
i
=x n ,
Then by
(3.3.8),
X
(1,2l +2)
^
i1
= x
( 1,2 l+1 )
^i1
1,2 l+1)
p ( x (i1
^
)
^i2
(x
j=1
( 1,2 l +1)
j
) ( x
j= ^i
40
} {
X
( 1,2l+1)
^
i1
( 1,2 l+1 )
^i1
( 1,2l +2)
j
( 1,2l+1 )
^i1
( 1,2 l +1)
^
i1
i2
^
1,2 l+1)
p ( x(i1
^
n
1,2l+1 )
xj ) (
( x(i1
^
j=1
j= ^i
by
( 1,2 l+2)
xi X i
finite
( 1,2l+2)
xi Xi
induction
(i=n , ,1).
l where
on
( 1,2r )
l=1, , r , x i X i
because
(i=n , ,1) .
(k)
Let the sequence X i (i=1, , n) be generated from (3.3.1) and (3.3.2). Suppose that
(k)
for some k 1, x i X i ( i=1, , n ) .
(1,2l)
(k ,2 l)
( i=1, , n ) (l 1)
been used to show that x i X i
, it follows that x i X i
(k+1)
(k , 2 l)
where X i = X i
.
Therefore
by
finite
induction
on k ,
x i X (k)
i ( i=1, ,n )( k 0 ) .
Also by (3.3.8),
,2 l)
X (k
X (ki , 2 l1 ) X (ik ) ( i=1, , n ) (l 1) ( k 1 ) .
i
As in the proof of Theorem 2.3.1 in Alefeld and Herzberger (1983), it may be shown
that >0 such that ( k 0) ,
41
i1
j=1
j=i+1
w(kj ,2 l1)
and
w(ki ,2 l) w(ki , 2l1)
i1
w(kj ,2 l1)
j=1
j=i+1
where
w(ki ,r ) ( n1 ) w ( X (ik , r )) ( i=1, , n ) ( r=1, , 2 m) ,(3.3 .11)
and
=
1
.(3.3.12)
n1
4 l2(i=1, , n1)
,(3.3 .13)
4 l1 ( i=n )
4 l +2 ( i=1 )
)
u(1,2l
=
4 l ( i=2, , n1 ) ,(3.3 .14)
i
4 l +1 ( i=n )
and for r=1, , 2 m , let
,r )
( 4 m)u(k
+ 2 ( i=1 )
i
(k +1,r )
(k ,r )
ui
= (4 m) ui ( i=2, , n1 ) ,(3.3 .15)
,r )
( 4 m)u(k
+1 ( i=n )
i
42
4 l16 ml +8 m4
2
(4 m)k1 +
(i=1 )
14 m
14 m
(k , 2 l1)
ui
=
,(3.3 . 16)
( 4 l2)( 4 m) k1 ( i=2, , n1 )
k1
4 l16 ml +4 m2
1
(4 m) +
( i=n )
14 m
14 m
and
4 l16 ml8 m
2
(4 m)k1+
( i=1 )
14 m
14 m
u(ki , 2 l)=
,(3.3 . 17)
(4 l)(4 m)k1 ( i=2, , n1 )
k1
4 l16 ml4 m
1
(4 m) +
( i=n )
14 m
14 m
Suppose, without any loss of generality, that
w(0,0)
h<1 (i=1, , n ) (3.3. 18)
i
Then by a lengthy inductive argument it follows from (3.3.9) - (3.3.18) that for
l=1, , m, i=1, , n , k 0 ,
w(ki ,1) hu
(k+ 1,2l1)
i
and
w(ki ,1) hu
(k+ 1,2l)
i
(k+1)
Xi
(k , 2 l)
= Xi
( k+1)
)
w(k+1
h4 m ( i=1, , n ) .
i
43
, ( k 0) ,
X
w ( i(k) )
( ) h
( 4 m)
Let,
X
w( i (k))
(k)
w =max
1 i n
Then, by (3.3.19)
w(k)
(4 m )
h
( k 0) .
()
So,
(k )
1
k
(k ) ( 4 m )
R4 m ( w )=lim ( w )
k
lim
k
{( ) }
1
k
( 4 m)
1.
Therefore, it is proven that according to Alefeld and Herzberger (1983), the order of
convergence of IRMSS1 is at least 4 m or
OR ( IRMSS 1, x i ) 4 m ( i=1, , n ) ( m 1 ) .
44
numerical results below represent all ten test polynomials (see Appendix A) for
comparisons of effectiveness of the IRMSS1 procedure compared to previous ISS1 and
IMSS1 procedure.
Table 3.1 to 3.4 show the numerical results when ten test polynomials are tested
numerically. The abbreviations used in the tables are defined as follows:
TP
n
m
k
t
i
r
w(k)
i
X (k ,r )
Procedures
ISS1
IMSS1
IRMSS1
Test polynomial.
Degree of polynomial.
Inner iteration.
Iteration.
CPU times.
Component.
Step number in algorithm.
Interval width.
Interval of iteration k, step r.
Procedure employed.
Interval symmetric single-step.
Interval midpoint symmetric single-step
Interval repeated midpoint symmetric single-step
Note that the same abbreviations will be used in the tabulated results in every chapter.
45
Table 3.1 Comparison of number of iterations k and CPU times t for every inner
iteration m for all test polynomials
TP
Procedures
n
ISS1
IMSS1
m=1
m =1
IRMSS1
m =2
m =3
m=4
0.202801
0.171601
0.121680
0.124800
0.134160
0.221521
0.184081
0.137280
0.121680
0.131040
0.243361
0.218401
0.177841
0.168481
0.180961
0.321362
0.265201
0.243361
0.243361
0.243361
0.546003
0.489843
0.483603
0.564723
0.620883
0.271441
0.237121
0.184081
0.199681
0.212161
0.377522
0.315122
0.249601
0.290161
0.280801
0.202801
0.171601
0.090480
0.102960
0.112320
0.312002
0.258961
0.224641
0.224641
0.230881
10
10
0.577203
0.461762
0.486723
0.511683
0.539763
46
0.6
0.5
0.4
CPUtimes
0.3
ISS1
0.2
IMSS1
IRMSS1
0.1
0
1
10
Test Polynomials
Figure 3.1 Comparison of CPU times for all test polynomials for the best value of
m
Number of iterations
ISS1
IMSS1
IRMSS1
0
1
10
Test Polynomials
Table 3.2 Values of intervals for each component i for test polynomial 8
Table 3.2a Component i=1
i k m Step
s
X
1 1 1
IMSS1
IRMSS1
Intervals
Interval
Widths
Intervals
Interval
Widths
[-20.08301193755741,
0.162960
[-20.08301193755741,
0.162960
(k ,r )
X (1,1)
1
-19.92005108556831]
X (1,2)
1
[-20.00000000427950,
-19.92005108556831]
8.701054e-09
-19.99999999557843]
2
[-20.00000000427950,
8.701054e-09
-19.99999999557843]
[-20.00000000001158,
(1,3)
X1
(Not applicable*)
2.354028e-11
-19.99999999998802]
[-20.00000000000001,
X (1,4)
1
3.552713e-15
-20.00000000000000]
2 1
X (2,1)
1
[-20.00000000000000,
3.552713e-15
-19.99999999999999]
X (2,2)
1
[-20.00000000000000,
3.552713e-15
(Already converged)
-19.99999999999999]
2
(2,3)
1
X (2,4)
1
(Not applicable*)
*These steps are not applicable because m=2 does not apply in the algorithm of IMSS1
procedure
48
IRMSS1
Intervals
w(k)
i
Intervals
w(k)
i
X1
(1,1)
[3.70009242751634,
3.89999215566156]
0.199899
[3.70009242751634,
3.89999215566156]
0.199899
X (1,2)
1
[3.81959192477592,
3.81972623564522]
1.343108e-04
[3.81959192477592,
3.81972623564522]
1.343108e-04
[3.81965999690657,
3.81966028366770]
2.867611e-07
[3.81966011250105,
3.81966011250106]
4.440892e-16
X (k ,r )
2 1 1
IMSS1
X (1,3)
1
(1,4)
(Not applicable*)
X1
2 1
X (2,1)
1
[3.81966011250101,
3.81966011250111]
9.237055e-14
X (2,2)
1
[3.81966011250101,
3.81966011250111]
9.237055e-14
(Already converged)
(2,3)
X1
X (2,4)
1
(Not applicable*)
*These steps are not applicable because m=2 does not apply in the algorithm of IMSS1
procedure.
49
X
3 1 1
(k ,r )
IMSS1
Intervals
IRMSS1
(k)
wi
(k)
[26.17904982613198,
26.18131053217847]
0.002260
[26.17904982613198,
26.18131053217847]
0.002260
X (1,2)
1
[26.18033875299099,
26.18034074130472]
1.988313e-06
[26.18033875299099,
26.18034074130472]
1.988313e-06
[26.18033988749894,
26.18033988749896]
7.105427e-15
[26.18033988749894,
26.18033988749896]
7.105427e-15
(1,3)
X1
X
(1,4)
1
(2,1)
[26.18033988749894,
26.18033988749896]
3.552713e-15
(2,2)
[26.18033988749894,
26.18033988749896]
3.552713e-15
X1
X1
wi
X (1,1)
1
(Not applicable*)
2 1
Intervals
X (2,3)
1
(Already converged)
(Not applicable*)
X (2,4)
1
*These steps are not applicable because m=2 does not apply in the algorithm of IMSS1
procedure.
50
(k)
i
Steps
Step 1
(3.2.1e)
Interval Widths
Step 2
(3.2.1f)
X (k)
i
w(1)
1
IMSS1
0.222813
w(1)
2
0.125019
w(1,1)
2
0.125019
w(1)
3
1.508622
w(1,1)
3
1.508622
w(1)
4
0.103066
w(1,1)
4
0.103066
w5
(1)
0.330997
w5
(1,1)
0.330997
w(1)
6
0.049919
w(1,1)
6
0.049919
w1
(1,2)
9.989220
(1)
w 1 9.989220e-05
w(k)
i
of
IRMSS1
0.222813
w(1,1)
1
w(1)
2
0.001317
w(1,2)
2
0.001317
w3
(1)
0.010475
w3
(1,2)
0.010475
w(1)
4
0.010496
w(1,2)
4
0.010496
w(1)
5
0.006637
w(1,2)
5
0.006637
w(1)
6
0.001611
w(1,2)
6
0.001611
2.909768e-07
w(1,3)
1
(1,3)
w2
Step 3
51
8.765317e-04
(3.2.1e)
(Not applicable*)
w(1,3)
3
0.008574
w4
(1,3)
0.002953
w(1,3)
5
8.033651
w(1,3)
6
2.665297
1.421085e-14
w(1,4)
1
Step 4
(3.2.1f)
(Not applicable*)
6.739497e-11
w(1,4)
2
(1,4)
w3
4.188939e-07
1.235115e-06
w(1,4)
4
(1,4)
w5
7.824946e-08
2.252065e-11
w(1,4)
6
*These steps are not applicable because m=2 does not apply in the algorithm of IMSS1
procedure.
(k)
i
Steps
Interval Widths
w(2)
1
Step 1
(3.2.1e)
IMSS1
2.897238e-10
w(k)
i
of
X (k)
i
IRMSS1
1.421085e-14
w(2,1)
1
(2)
w 2 9.513039e-07
w2
7.533329e-06
w(2)
3
4.263256e-14
w(2,1)
3
(2)
w 4 2.548911e-06
w4
4.334879e-07
w(2)
5
1.776356e-15
w(2,1)
5
2.187787e-10
w(2)
6
8.881784e-16
w(2,1)
6
52
(2,1)
(2,1)
3.552713e-15
5.329070e-15
w(2)
7.105427e-15
1
(2)
w 2 3.552713e-15
Step 2
(3.2.1f)
(2)
3
(Already converged)
2.380318e-13
w(2)
5.258016e-13
4
w(2)
7.460698e-14
5
w(2)
8.881784e-16
6
Step 3
(3.2.1e)
(Not applicable*)
(Already converged)
Step 4
(3.2.1f)
(Not applicable*)
(Already converged)
*These steps are not applicable because m=2 does not apply in the algorithm of IMSS1
procedure.
(k)
wi
IMSS1
IRMSS1
w(1)
1
6.560552e-04
1.158184e-12
w(1)
2
0.021827
2.780242e-08
53
w(1)
3
0.006104
3.522296e-07
w4
7.709550e-05
9.846008e-10
w(1)
5
0.003085
5.420934e-08
w6
(1)
0.005258
3.497880e-08
w(1)
7
6.222428e-05
5.036993e-10
w8
(1)
4.793474e-04
7.522427e-10
w(1)
9
0.009712
1.190734e-08
w 10
(1)
4.723636e-04
3.552713e-13
w(2)
1
7.105427e-15
1.158184e-12
w2
(2)
2.131628e-14
3.552713e-15
w(2)
3
2.664535e-14
1.776356e-15
w4
(2)
8.881784e-16
8.881784e-16
w(2)
5
4.884981e-15
4.440892e-16
w6
(2)
1.554312e-15
2.220446e-16
w(2)
7
8.881784e-16
8.881784e-16
w8
(2)
1.776356e-15
1.776356e-15
w(2)
9
3.552713e-15
3.552713e-15
(2)
5.854872e-12
3.552713e-13
(1)
w 10
54
that is
m=1
because both of
these procedures do not have the process of inner iteration. While for IRMSS1
procedure, there are three values of
m=2, m=3
and
m=4 .
The highlighted cells in IRMSS1 column show the best results selected for
corresponding value of m .From this table, clearly we can see that for all 10 test
polynomials, IRMSS1 procedure gives us less CPU times compared to both ISS1 and
IMSS1 procedures except for test polynomial 10. For test polynomial 10, IRMSS1
shows slightly longer CPU times. In terms of number of iterations, IRMSS1 shows
good results where it always has the same or less number of iterations compared to both
previous procedures. Figure 3.1 and Figure 3.2 show the comparison clearer.
Table 3.2 above shows the values of intervals for each component i=1 ,
i=3
for
,r )
X (k
i
i=2
and
polynomial 8 was used. Number of iteration needed for IRMSS1 procedure for the
55
intervals to converge is 1, compared to IMSS1 which needs two iterations. The values
of intervals that had converged to the zeros were bolded and underlined so that we can
see the convergences of the intervals clearer. From the first step until the final step of
the procedure, it can be seen how the intervals converge to the zeros. For IRMSS1
procedure, the best value of
each iteration k
iteration k .
For Table 3.2a, the results show the intervals generated for component i=1 . For this
component, four steps were computed for IRMSS1 procedure while three steps were
computed for IMSS1 procedure for the intervals to converge. The widths of the final
intervals for both procedures for this component are similar.
procedure requires four steps while IMSS1 requires three steps for the intervals to
converge, the final intervals widths are smaller for IRMSS1 procedure.
steps compared to IMSS1 procedure which needs four steps for the intervals to
56
converge. The widths of the final intervals for both procedures are of not much
difference.
From Table 3.3 above, we can see that both procedures need two iterations for the
intervals to converge. However, for IRMSS1 procedure, the intervals converge earlier at
the first step of iteration 2. The widths of the final intervals for IRMSS1 procedure also
are smaller.
Finally, even though the results in Table 3.4 show that the procedure IRMSS1 requires
the same number of iterations compared to the procedure IMSS1, the final widths of the
intervals for IRMSS1 are always smaller than or the same as the IMSS1 procedure. In
addition, there are two components in IRMSS1 that have already converged in iteration
1 which are components 1 and 10.
57