On The Semilocal Convergence of A Fast Two-Step Newton Method
On The Semilocal Convergence of A Fast Two-Step Newton Method
On The Semilocal Convergence of A Fast Two-Step Newton Method
Ioannis K. Argyros
1. Introduction
In this study we are concerned with the problem of approximating a locally
unique solution x of the nonlinear equation
F (x) = 0, (1)
15
16 IOANNIS K. ARGYROS
for all n 0.
It follows by the definition of sequences {sn }, {tn } that if there exists
[0, R] such that
sn tn+1 < w01 (1) for all n 0, (10)
then both sequences are monotonically increasing, bounded above by , and
as such they converge to a common limit t such that
tn sn tn+1 (n 0), (11)
and
t . (12)
We can show the following semilocal convergence theorem for Newton-like
method (2) using majorizing sequences {tn } and {sn }.
Theorem 2.1. Under conditions (3), (4) and (8) for F (x0 )1 F (x0 ) ,
kF (z0 )1 F (x0 )k t1 sequence {xn } (n 0) generated by Newton-like method
(2) is well defined, remains in U(x0 , t ) for all n 0, and converges to a unique
solution x of equation F (x) = 0 in U(x0 , t ).
Moreover the following estimates hold for all n 0:
kyn xn k s n tn , (13)
kxn+1 xn k tn+1 tn , (14)
kyn x k t s n , (15)
and
kxn x k t tn . (16)
It follows from (30) and the Banach Lemma on invertible operators that
linear operator L given by
Z 1
L= F (y + t (x y )) dt (34)
0
is invertible.
Using the identity
0 = F (x ) F (y ) = L(x y ), (35)
we deduce x = y .
Finally to show uniqueness in U (x0 , R0 ), again as in (30) we obtain
Z 1
F (x0 )1 (L F (x0 ))
w0 (tt + (1 t)R0 ) dt < 1, (36)
0
which again together with (33) yields to x = y . That completes the proof of
the theorem. X
Remark 2.1. Although stronger but easier to verify conditions implying crucial
hypothesis (8) have already been given in [2], when w0 (r) = w(r) for all r
[0, R], and us [3], [4], when functions w0 and w are not necessarily the same,
we decided to leave condition (8) as uncluttered as possible. In order for us
to find conditions other than (8), let us assume there exists a monotonically
increasing function w satisfying (5) and for all t s, with s, t [0, R]:
Z ts Z t
w(t)dt [w(t) w(s)] dt. (37)
0 s
Such an estimate can follow e.g. from
w(r) = sup{w(u) + w(v) : u + v = r}. (38)
This function may be calculated explicitly in some cases. For example, in the
Holder case
w(r) = r (0 < 1) (39)
we have
w(r) = 21 r . (40)
r
In general, if w is a concave function on [0, R], we have w(r) = 2w 2 . Clearly
w is always increasing, concave, and
w(r) w(r) for all r [0, R]. (41)
Conditions of the form (35) - (36) were first given in [2]. More information on
the motivation for the introduction of function w can be found in [2] - [4].
It is convenient n
for us
o to define scalar functions f , g on [0, R], and sequences
{sn }, tn , sn , tn (n 0) for all n 0 by
Z r
f (r) = r + w(t)dt, (42)
0
Z 1
g(r) = r+ w(t)dt, (43)
0
s0
t0 = 0, s0 = , t1 = s0 + ,
1 w s0 +t2
0
R1
0
w t tn+1 tn sn tn dt
sn+1 = tn+1 + , (44)
1 w(tn+1 )
R 1 1
0
w 2 sn tn + t tn+1 tn tn+1 tn dt
tn+2 = tn+1 + (45)
1 w tn+1 +s
2
n+1
t0 = t0 , s 0 = s 0 , t1 = t1 ,
f1 tn , sn , tn+1
sn+1 = tn+1 , (46)
g tn+1
f2 tn , sn , tn+1
tn+2 = tn+1 , (47)
tn+1 +sn+1
g
2
where,
Z 1
f1 (a, b, c) = w[a + t(c a)](b a)dt w(a)(b a),
0
and
1
a+b
Z
f2 (a, b, c) = w[b + t(c a)](c a)dt w (c a).
0 2
In view of (3) and (4) it follows that
w0 (r) w(r) for all r [0, R], (48)
w(r)
and can be arbitrarily large [3], [4]. By comparing sequences {sn }, {tn }
w0 (r)
with {sn } and tn and using induction on n 0 we deduce
sn sn , (49)
tn tn , (50)
s n tn sn tn , (51)
tn+1 tn tn+1 tn , (52)
t sn t sn , t = lim tn , (53)
n
t tn+1 t tn+1 , (54)
and
t t . (55)
Note also that strict inequality holds in (47) - (50) if (44) also holds as a strict
inequality.
Moreover if (35) or (36) hold then
sn sn , (56)
tn tn , (57)
s n tn sn tn , (58)
tn+1 tn tn+1 tn , (59)
t sn t sn , t = lim tn , (60)
n
t tn+1 t tn+1 , (61)
and
t t . (62)
n o
Clearly, if conditions for the convergence of sequences sn , tn are imposed,
the same conditions
will imply the convergence of the finer sequences {sn },
{tn }, {sn }, and tn (n 0). Such a condition is:
(C) Equation
f (r) = 0 (63)
has a unique solution [0, R].
The proof is omitted since it has essentially been given in Theorem 2 in [2,
p. 5].
Remark 2.2. Concerning related method (7), let us consider the corresponding
scalar majorizing sequences {pn }, {qn }, {pn }, {q n }, pn , q n , (n 0)
defined as the stsequences, respectively.
For example, sequences {pn }, {qn } as defined as {sn }, {tn } given in (6) and
tn + s n
(7) but sn , tn , tn+1 , are now pn , qn , pn+1 , pn , respectively, etc.
2
Clearly, method (7) also converges under condition (C).
Note that a similar proof as in Theorem 2.1 can be given for method (7).
We do not known if the stsequences are finer than the pqsequences. In
practice, we will use both to see which ones provide the more precise estimates
on the distances k yn xn k, k xn+1 xn k, k yn x k (n 0).
Finally note that the results obtained here can be extended to the more
general method (2) where zn = (1 )xn + yn , 0 1. However here we
decided to examine (2) only in the case = 12 which although seems to be the
most popular [7], [8], [13] we do not know yet if it is always the best choice.
References
[1] Amat, S., Busquier, S., and Salanova, M. A. A fast Chebyshevs method for quadratic
equations. Appl. Math. Comput. 148 (2004), 461474.
[2] Appel, J., DePascale, E., Lysenko, J. V., and Zabrejko, P. P. New result on
NewtonKantorovich approximations with applications to nonlinear integral equations.
Numer. Funct. Anal. and Optimiz. 18, 12 (1997), 117.
[3] Argyros, I. K. A unifying local-semilocal convergence analysis and applications for
two-point Newton-like methods in Banach space. J. Math. Anal. Applic. 298 (2004),
374397.
[4] Argyros, I. K. Convergence and applications of Newtontype iterations. Springer Ver-
lag, New York, 2008.
[5] Argyros, I. K., and Chen, D. On the midpoint method for solving nonlinear oper-
ator equations and applications to the solution of integral equations. Revue dAnalyse
Numerique et de Theorie de lApproximation 23 (1994), 139152.
[6] Gutierrez, J. M., and Hernandez, M. A. An acceleration of Newtons method: Super-
Halley method. Appl. Math. Comp. 117 (2001), 223239.
[7] Hernandez, M. A., and Salanova, M. A. Modification of the Kantorovich assump-
tions for semilocal convergence of the Chebyshev methods. J. Comput. Appl. Math. 126
(2000), 131143.
[8] Homeir, H. A modified method for root finding with cubic convergence. J. Comput.
Appl. Math. 157 (2003), 227230.
[9] Homeir, H. A modified Newton method with cubic convergence. J. Comput. Appl. Math.
169 (2004), 161169.
[10] Kantorovich, L. V., and Akilov, G. P. Functional Analysis in Normed Spaces. Perg-
amon Press, Oxford, 1982.
[11] K.Argyros, I., and Chen, D. The midpoint method for solving equations in Banach
spaces. Appl. Math. Letters 5 (1992), 79.
[12] Ortega, J. M., and Rheinboldt, W. C. Iterative Solution of Nonlinear Equations in
Several Variables. Academic Press, New York, 1970.
[13] Ozban, A. Y. Some new variants of Newtons method. Appl. Math. Letters 17 (2004),
677682.
[14] Yamamoto, T. A convergence theorem for Newton-like methods in Banach spaces. Nu-
mer. Math. 51 (1987), 545557.
Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.
Alternative Proxies: