Tome
XXVtr,
Nol,
199g,pp.
23-31ON TTTE MONOTONE CONVERGENCE
OF AN EULER-CTIEBYSIIEFF.TYPE METHOD
IN PARTIALLY ORDERED TOPOLOGICAL SPACES
IOANNIS K. ARGYROS
1,INTRODUCTION
- In this study we
areconcemed with the problem of approximating a solution x. ofthe nonlinear operator equation
(l) F(x) =
0in
alinear
space 81,where F is defined on
aconvex
subsetD of D1 with values in
a
linear
space82.
We
haverecently
shownthât if .El
aridEzare
Bànach spaces, thenunder
stan- dardNewton-Kantorovich
hypotheses the Euler-Chebysheff-type methodof
thefonn
(2) ln=xn-lxn,xn]-'¡'(r,)
(3) xn+1= yn -f)ín,xnf-'(fx,,!,f-l_x,,x,l)(yn-xn) xo eD (n20)
converges
with order almost three to a locally unique solution x*
çD of equation (1). Here [x,y]
denotes adivided difference of
order one,which
is alinear
operator.we introduce
and studythe monotone
convergenceof
theiterations {a,}
and{x,l
@> 0) given by
(4) F(a,)+fxn,xnl(wn -u,¡=g
(5) F(x,) +fxn, xnl(yn - x^) =
0(6) (fx,,!nl-[x,, x,f)(un-u,)+fx,,xnl(u,*,-w,)=0
and
(7) (lxn,l,f-lx,,x,l)(y, -xn)+lxn,x,f(x,*t-U)=0
to approximate
asolution
x*of equation (l).
(1991) AMS (MOS) Subject Classif¡cation Codes. 65H10, 6SJl5, 49D15, 47H17
rvhere
Rro*r.',=
9',0*'h('
- x).
Choosing
s='2k
+2' in (6)' from the boundness of h, Lemma 2 md the Cauchy inequality it results that there exists a constant y which
dependson
fr and d; suchthat
QlÏ)Rru.r.") (x)
< Ym-(k+r)'
lTheorem
3follows from (3)
andTheorem 2' REFERENCES
l.
B. Della vecchia,on
the approxìmation of functions by means of the operators or D' D' stancu' Studia Univ. Babeç-Bolyai, Mathematica, 37,I
(1992)'3-36'?..A.
Di
Lorenzo andM.
R.-Occorsio,
Polinomi
tli
Stancu, Rapp' Tecnico, no'l2ll95' I'A'M''
Napoli, 1995.
3.
ti. H.
Gonska andJ.
Meier, Quantitative theoremsoi
approximationby
Bernstein-slancu operators,Calcolo 21 (1984), fasc' IV,3l'l-j35'
4. G. G. Lorentz, Bernstein Polynomials,University of Toronto Press' Toronto' 1953'
5
G.MastroianniandM.R.Occorsio,Sullederívatedeipolinomídistuncu,Rend'Accad'Sci'Fis' Mat. NaPoli, 45,4 (1978),n3-281'6. G. MasUoianni and M. R. Occorsi o, I)na generalizutione (lell'operatore tli Stancu' Rend' Accad' Sci, Fis. Mat' Napoli, 45, 4 (1978), 495-511'
'7.C.
p.
May, Saturation and inverse theoremsfor
combinationsoJ'a
classof
exponential-rype oPerators,Canad' J. Math., 2E (1976), 1224-1250:8. G. Múhlbach, verallgemeínerungen der Bernstein- und lttg,range-Polynome. Bemerkungen zu einer Klasse linearer Polynomoperatoren von D. D. Stancu, Rev. Roumaine Math' Pures Appl. 15, S (1970), 1235-1252.
9. R. K. S. Rathore, Linear Combinøtions of Linear Posítive Operalors and Generating Relations in
S p e c iul l' u nctions, Thesis, I.l.T., Delhi, I 973'
10. D. D. Stancu, Approxirw.ttion of functions by a new class of linear polynomial operators,Rev' Roumaine Math, Pures Appl., 13,8 (1968), 1173-1194'
11. D. D. Stancu, IJse of prohabílisric methocls ìn the theory of
uniþrn
appruximation of continuotts .functions,Rev. Roumaine Math. Pures Appl', 14 (1969)'673491'
12. D. D. Stancu, Recurrence relations
for
the central moments of some discrete prohability laws, Stuclia Univ. Babeç-Bolyai, Cluj, Ser. Math.-Mech', 15,I
(1970),5542'13.
D. D.
Stancu, Approximaiion propertiesttJ'a
class of linear positive oper\tors, Studia Univ' Babeç-Bolyai, Cluj, Ser. Math'-Me¿h., 15, 2 ( 1970)' 33-38'14.
D. D,
Stancu,on
the remainderof
approximation of functktns by means oJ'a
pararneter- dependenr linear polynomial operator, Studia Univ. Babeç-Bolyai, Cluj, Ser' Math'-Mech',16,2
(1971),5946.
Received November 15, 1997 " Bab e S - B o lYai
"
U nive rsitY, Faculty oJ Mathematics and Computer Sciencel,
M.Kogölniceanu St., 3400 Cluj'NaPocø,Romania
L
Ioannis K, ArgYros
The
Euler-ChebYsheff method(or
otra [7]. Ul'm used divided divided diffe- orem
[7,p'91]'The
orderofconvergence
of our iterations is almost
s
oforder
oneonlY'
2. MONOTONE CONVERGENCE
On the Monotone Convergence 25
Moreover, if
theoperators Ar=fx,,x,f are inverse nonnegative, then any solution
uof
theequøtion F(x) = 0 in (uo, xr) belong to
(u*, x')
'Proof. Let
usdefine
the oPerator.ff
:(0,.t0 -un)-+ 81, PtG)
=x - Bo(F(u,r) + Ao@)'
This operator
isisotone
andcontinuous. We
canhave in turn 4 (o)
=-aoF(uo) 2
o,4 (xn -t o) =Ío -uu- BuF(xo)+4(F(xo)- F(uo)-lo(xo -an)) 1
xtt-
uo + Bo ([-xo,uo]-
[x6 , xe]) (xo -
uo) þy (9))
1xo-uo,
since
[-xn,urlS
[xu ,t¡ ] by (12).
By Kantorovich's theorem [4], operator Pt has a fixed point z,
e(0,
x o-
u o):
Pr(zr)
= zr. Set
u o = u 0 + zr, thenwe
have estimatesF(uu)+
An(w,',-ro) =
0,F(uo)
=F(wo) - F(uo
)- Ao(wo- on)
< oand
uo
<llo < xo, We define the
operatorPr:(O,x0 -ut¡) + E', Pr(x) =
tt- Bo(F(ro)+ ln(x)).
This operator
isisotone
andcontinuous. We
can havein turn
Pz(0)=Bu¡(-ru)>0,
P,
(*
u-
to¡) =
x o-
u or
B oF (u.tu)+
Bo(F(xu
)-
F (w n)- A,
(x u-
uo)) <
lxo - wo+ Bo(fxr,wr,]-lxo,xo]) (¡o -Ø,,) < (by (9))
1 Xç _ lJ)ç¡
r
since
[.ru,ur]3
[xu , xs]
bY(12)'
By Kantorovich's theorem there exists z, e(0,
x,¡- uo) such that
Pt(rr)
=zz.
Setlo
=ro - zr, thenwe have the
estimatesF(¡u)+Ao(yo-r,,)=0,
F(Yu) = F(lo) - F(xo) - AuQu-rr,) )
o2 3
24
Weshallassumethatthereaderisfamiliæwiththemqaningofadivided difference of order on" un¿ the notion of a partially ordered topological
space(POTL) ([1], [2]' tZl, iql)'Moreovet' from now on we shall
assume thatù
and E2are POTL-sPaces.
We .átt no*
state the mainresult'
Tlæongvt|,LetFbeanonlinearoperatordeJìnedonaconvexsubsetDofa regularPoTl-spacenlw¡thvaluesin-qPoTL.spaceE2'LetU0andxgbetwo
SupposethatFhasadivided'dffirenceoforderoneonDn=(uo'ro)=
= {x eE1
luo <¡ I xo} ç D
satisJYing(10)
Ao=fxo, xol
has acontinuous nonnegative left
subinverse Bs' points of D such that
(8) (e)
(1 1)
(r2)
and
(1 3)
(4]'l.7), (14)
(1s)
uu3xo
F(u,,)<0<F(ro)
u0
<wo <ut < 3ll), 1un*t 3
xr+t 3ln 3 "' ( rl (
Yo3 xs' limr)-=t)' , lim
,-+* "
n-+úand u'
,xo
eD, with u* (x'
[xo,Y]20 for all tto3YSxs' [x,u]-[x,Y]<0 iÍ u<Y
lz,wl+lw,ql-lz,zl-lu,zl20 if uSw<z for some q e(u'z)
'Then there exist
twosequences {un\'{x'\ n20 satisfuing approximations
xn=x
and
Lo(yo -"u)<0.
From approximation (6)
wenow have
Ut
- uo =
tÐ0* Brlr(wu -
uo)-
wr,= BnLu(wo -ro)
2 0.Hence, we obtain
u)o1uo.
Moreover, from approximation (5) we have
x't - lo= l, i Bulu(y, -rr) - lo= BoLo(yo -xo)10.
That is, we get
)ctI lo
Furthermore, we
canobtain in turn
u
t - x'¡
=:',:7,'lï:r:,;,' -l:...7 !;jí=
- "' =
- uo BuLoF(uu) - (xo - BoF(x,,)) +
+
BuLo(u, - BuF(un - BnF(uu)) -
BnZo (uu )+ +BrLoQo) - BrLo(xo - BrF(xo)) =
.
= u0-
x,0- Bo(F(uo) - J¡(¡o
))-
B0L0E0(F(¿o )- F("r,, )) =
=
(I - Bslus, xof - BoLrBofuo, xrl)
(ao- ro
).But using
hypotheses(12)
and(13) we
haveB0L,B,lt)r, xul+ Bofun, xo)S
B0L0B0A0+
Bofuo,ro ]l
<
BrLu +
Bn[uu, 'xo ]I Bo(Lo
+ [uo, xu])
<3 Bolyo, qlS BrAo 3,L
Wenow obtain ut 1xt.
From all the
abovewe
haveuoSuo1ut1x,1yo3xo.
By hypothesis (12),.it follows that the operator An has a continuous nonnegative left subinverse 4 for all n> 0. Proceeding by induction, we
canshow that there exist two sequences {u,), {xn} (n
>0) satisffing (4) - (7)
and(l ) is
aregular
space E1 and as suchthey converge to some u' ,x' eDo. That is,
we
have
)y_r, =u' 3x.
=]*x,.
l
and
urr
Swrr 3
Yo< xo.
We now define the
oPeratorP¡:(0, xo -uo)-+ Et, Pr(x) = x - Bo(LgBrE(uo)+'4n(x))' where I, = lxo
, r,r ]-
lxo,.Yn]'
'Ihis operator is isotone
andcontinuous' We
havein turn Pr(0) = -BoLoBoF(un) 2 0 bv (9)
P,
(xo -
Ito)
= xo-
u o- BrrLrBoF(x
o)+ +Br(LoBo(F(x,,
)- F(uu)) -[¡n, x'
l(ro -'o ))' But, try (11)
and(12), we
canliave
L.B,,F(x
o)- F(ao)) - [xo, xo] (ro -
/u )=
..
= (Lo/,o[xo,
uo]- [x,,,
xo D(xo -
uo )<
<(Io -lxo,xo]) (x, -uo) I -[xt,
J¡o](xo -uo) I 0'
Therefore, we
haveP,(to -uo)1xo-u,¡'
By Kantorovich's theorem there exists z, e(0,
xo- uo) such that
Pr(rr)
=z¡. Set ut = uo
+zr, thenwe have
estimates- Lr(wr-ro)
+Ar(uo -Øo)
= 0and
Lo(wr-au)20.
Furthermore,
\rye candefine
theoperator
&:(0, xu -
0o)-+
Er, Pr(x) - x
+Br(LoB,rF(x
u)- AuQÐ'
This operator is isotone
andcontinuous. We
havein turn P4(0) = BoIo-BoF(xo)> 0 bY(9),
Po(xo -uo)=
xo-uo l BoLoBoF(ur)+
+
Bo(LoBo(F(xo
)- F(uo)) - Ar(xo -
uo))
< xo-uo
(by using the same approach as for P¡). tsy Kantorovich's theorem there exists
z o e
(0,
x o-
Uo) such that
Po(, ì
= zo. Set
Xt
=!
o-
zo, thenwe have
estimates-Lu(yo - ro) + Ar(x, -./o)
= 0On the Monotone Convergence 29
Hence, we
get0> 9-t F1u,))
un-u),,
0< 9-I F1x,)3
xn- yn.
Since E¡ is notmal and 1im
(un-w,)
=H (x, - y)
=0, we have ,lim Q-t Flun¡=
=,lg Q-t F(*,)
=0.
ftrence,by continuity, we
getF(u' )= F(r. ) =
0.(c) As
above,we
get02 F(a,)2 R(u, -w,),
0 <F(x,)
<R(x, -y,).
Using
thenorrnality
of Ez and thecontinuity of F
and R, weget F(u.
)= F(x* ) =
0.(d) From the equicontinuity of the operator
Anwe have
,lim A,
(u n-
un) =
=,lg An(x, - yn\ =0.
Hence, bV(a)
and(6)
F(u")=F(x')=0.
(e) Using
hypotheses(10) - (14), we
getin turn
0
( F(Y,) = F(Y,)- F(x,) - A,(Y, - x,)
==
(An -1.!n,
xn))(x, - y,)
<([xo, xof-l**, x' l) G, - y,).
Since Ez is normal and lim (x, -!,)=0, we get lim F(x,)=Q. Moreover,
from hypothesis (12)
[x*, ¡*] ("n -
*n )I [,r-, xn)(x, - x'
)=
= F(xn) -_F(¡.) ( [ro,
xn ](x, -x')
and by the normality of E2,F(x-)
=j_T F(x,).Hence, we g:t F(x-) = 0. The result F(u- )
=0
can beobtained similarly.
The proof of
thetheorem is now complete.
As in Theorems I
and2, we can prove the following result (see also [7, Theorem
6.21:Trmonuvt
3. Assumethat hypotheses of Theorem I are true.
Thenthe appro- ximations
lr=xn-BnF(xn),
xn+t
=yn+ BnLr(yn- xn), Ln=fx*xrl-fxn,!rl
,:
( ,i , un =un - B,F(un)
7
lf
u,, < u1 xtt aîd F(u) =0, then we
canobtain
An(Y o
- u)
=Ao(x
o-
-BoF(xn )) -
Asu=
=
Ao()(o-
u)- AnBr(F(xo
)- F(z)) =
=Ao(I_Bo[¡n,u])(xo_u)20,sinceBu[x9,uf<BoA,,<1.
Similarly, we show Au(w, -
a) <0'
Iftheoperator,4gisinversenonnegative,thenitfollowsfromtheúï.:
*o 3,3 yo. i'ro...Airrg by induction, weãeduce that
w n3
u <l" from which it follows that wnlunlun*t31t3!n+t3xn1!n' fot all n>0' That is' we
haveu, 1u3 x, for all n>0'
Hence,we get u' 1tt< x'
'That
completes theproof
of thetheorem'
ln what follows, we shall give some natural conditions under which
thepoints u*
andx* aresolutionsof equation F(x)=Q'
TITEOREM
2. Under
hypothesesof Theorem I
supposeF is continuoîts at
t)*and x'
. IJ oneof
thefolowing conditions
issatisfied (a) N' =y" ' terator
Q:E', -+ Er' (Q(0)=0) which
(b),El
¡snormal and there
existsan
ophas an isotone inverse continuous at the origin and such that An<T for sfficientlY large
n,(c)EzisnormalandthereexistsanoperatorR:8,-+82(R(0)=0)conti-
nuous
at
theorigin
and suchthat A, 3 R for stfficiently larg
n,(d) operators
Anare
equicontinuousJbr all n20'
qnd
(e) Ez
ß normal and lu,u)<Íx, yl if
u <x ønd
u <y'
thenwe
haveF(u.)=F(x.)=0'
Proof. (a) Using the continutty of F and F(u')30<F(x')' we
getF
(u. )<
0< F(u. ). That
is,we obtain F(x
-) =
F(u"
) --0' (b) By (a)
and(6)
0
) F(u,) = An(u, - u,)>- Q@, - w,)
0
I
F(r,
)= An(x
n- Y,)> QG, - Y,)'
Ioannis K. Argyros 6 28
6.
M.
T. Necepurenko,on
chebysheffs method for functional equatiors (in Russian), usephi Mat' Nauk., 9 (1954). 163-170.7. F.
A.
Pota, on an ileralive algorithm of order L839 ..' Jor solving nonlinear operqlor equations' Numer. Funct. Anal' Optimiz'' 7(/),
(1934-1985)' 75-106'g. s. ul,m, Iteration methods with divided diferences of the second order
(n
Russian), Dokl' Akad' Nauk SSSR, 158 (1964),55-58' Soviet Math' Dokl' 5' ll87-1190'g. J.
A.
Vandergraft, Newton's methodfor
convett operators in partially ordered spaces,sIAM
J'Numer. Anal.
4
(19ó7), 406-
432'Received August 10, 199ó Deparlmen! of Mathem atics' Cameron (lniversity, Lawton, OK 7i505'
U,S,A.
30 Ioærnis K. ArgYros
and
un*t
=ll)n
+BnL'(wn -l)')
\)here the operators
Bnare nonnegative
'ubinverses ofAr, generate two
sequences{u,) and {x,\(n'0¡' satisfying"approximations (4)-(7) and(14)' Moreover'for iny solution u e(un, xr) of
theequation F(x)
=0 we
høveue(ur,x,) (,¿>0).
Furthermore,ctssuruethatthefollov)ingaretrue:
(a)
E2is
aPOTL-space
andE¡ is
anormøl POTL-space;
(t ) ,l1x x, = x* and lim Ü' =Ü*;
(c) f'rs continuous at u* and x'
;and
(d)thereexistsacontinuousnonsingularnonnegdtiveoperatorTsuchthat ß,,2T for sfficientlY large n'
ThenF(u.)=F(x*)=0'
Remarlc.(a) our conditions coinci
tewith (44) and(5Ð in [7, p. 98].In
caseEl=åz=[R,ourconditions(12)and(13)aresatisfiedifandonlyifFis
ditferentiable
on Ds, aîd F,F'are convex
on Do '(b) It followu ao- ui tlr" above that otlr method
usesthe same
01simpler
conditionsthanthoseusedinallpreviousresults(t4]_t9])buttheorderof
convergence is faster [3]' (c) Similar ,.roi,.
"* immediately follow if the divided difference
[x0 'xs'l is replaced by lxo ,xttfuttlzo3x" in (10)' lx"xn] is replaced by [x" ln-tf
(n>l) in(a)-(7).
REFERENCES
1.I.K.Argy'os,andF.Szidarovszþ'OnthemonotoneconvergenceofgeneralNewton-like
m e t h od s,B ull. Austral' Math' Soc'' 45' (1992)' 489-5 02'
2.LK.AtglTos,andF.Szida¡ovszky,TheTheoryandApplicationsoflterationMethods'C'R'C'
Press, [nc. Boca Raton, Florida, U'S'A' (1993)'
3.
I.
K. Argyro s, On lhe,oní"'g'n'"
o¡ ChebysheffHatley-type nethod under Newton-Kantorovichh yp o t h e s e s,Appl. Math' Letter, 6, 5' (1993)' 7 l-7 4'
4.
L .v.
Kantorovich, The methodof
successive approximatiott for functional equotions' Acta Math''T
(1939),63-97.5.