J. Math. Anal. Appl.

324 (2006) 344357

www.elsevier.com/locate/jmaa
Existence and quasilinearization methods
in Hilbert spaces
Mohamed El-Gebeily
a,,1
, Donal ORegan
b
a
Department of Mathematical Sciences, King Fahd University of Petroleum and Minerals,
Dhahran 31261, Saudi Arabia
b
Department of Mathematics, National University of Ireland, Galway, Ireland
Received 14 November 2005
Available online 9 January 2006
Submitted by Steven G. Krantz
Abstract
In this paper we discuss some existence results and the application of quasilinearization methods, devel-
oped so far for differential equations, to the solution of the abstract problem

Lu =Fu in a Hilbert space H.
Under fairly general assumptions on

L, F and H, we show that this problem has a solution that can be
obtained as the limit of a quadratically convergent nondecreasing sequence of approximate solutions. If
the assumptions are strengthened, we show that the abstract problem has a solution which is quadratically
bracketed between two monotone sequences of approximate solutions of certain related linear equations.
2005 Elsevier Inc. All rights reserved.
Keywords: Self adjoint operators; Nonlinear operators; Existence; Resonance; Quasilinearization methods
1. Introduction
In this paper we discuss the quasilinearization method for the abstract problem

Lu =Fu, (1)
*
Corresponding author. Fax: +966 3 8602340.
E-mail address: mgebeily@kfupm.edu.sa (M. El-Gebeily).
1
Research of the author has been funded by King Fahd University of Petroleum and Minerals under Project number
MS/Singular ODE/274.
0022-247X/$ see front matter 2005 Elsevier Inc. All rights reserved.
doi:10.1016/j.jmaa.2005.11.077
M. El-Gebeily, D. ORegan / J. Math. Anal. Appl. 324 (2006) 344357 345
where

L is a (self adjoint) operator and F is a nonlinear operator in a Hilbert space H. Such
methods take advantage of the existence of upper and lower solutions to obtain monotonically
convergent sequences of approximations of nonlinear differential equations (see [6]). The orig-
inal method was developed in connection with convex functions and was later generalized with
the goal of relaxing the requirement of convexity [1,7]. By generalizing the method to Hilbert
spaces, the scope of its applicability should be substantially widened.
It turns out that the existence of upper and lower solutions, by itself, provides a power-
ful way of showing the existence of solutions for (1) under very general conditions on

L and
F (namely,

L is bounded below and F is weakly continuous), irrespective, for instance, of
whether or not (1) is at resonance. Under a stronger assumption on F (namely, F is weakly
continuous and monotone decreasing), one can obtain a monotone nondecreasing sequence of
approximate solutions which converges quadratically in the norm of H to a solution of (1). Still
under stronger assumptions on

L and F (

L strongly positive denite and F possessing two
Frchet derivatives, the rst derivative being negative), one can obtain two sequences of so-
lutions of linear equations which quadratically bracket a solution of (1). Of course, the space
H itself should have some special structure to allow partial orderings and a sense of positiv-
ity.
In the next section we discuss some abstract existence and xed point theorems related to
our problem. The main tool in Section 2 is the Galerkin method, and we obtain sequences of
solutions which converge weakly to a solution of (1) and other related problems. We also dis-
cuss a uniqueness result. In Section 3 we discuss the notions of upper and lower solutions in
abstract Hilbert spaces with cones. The main result there is to show that if problem (1) has a
lower solution and an upper solution , then it has a solution in between. In Section 4
we discuss two generalizations of quasilinearization methods for approximating solutions of
(1). A fair amount of effort is spent in Sections 3 and 4 on dening and proving the proper-
ties of the Hilbert space H that are needed for the existence and the quasilinearization methods
to work.
2. Abstract existence and weak continuity
Let H be a separable Hilbert space with inner product , and norm . Let F : HH be
a nonlinear weakly continuous function and

L: D(

L) HHbe a densely dened self adjoint

operator and b H. In this section we investigate the abstract problems
Fu =b, (P)
Fu =u (Q)
and

Lu =Fu. (R)
It is possible to consider the above problems with the function F dened on a closed convex set
with nonempty interior, but we will not do this here. Although our main concern in the sequel will
be with problem(R), we include the other two problems for the sake of completeness and because
problem (R) can either be put in one of the two forms (P) or (Q). Our main tool for establishing
the existence of solutions to these problems is the Galerkin method. In anticipation of the set up
of the method, we let {w
i
}

i=1
be a complete set in H, which is also in the domain of

L in case
we are considering problem (R). Each problem will have its own appropriate assumptions. For
problem (P) we assume:
346 M. El-Gebeily, D. ORegan / J. Math. Anal. Appl. 324 (2006) 344357
(P1) There exists >0 such that the following coercivity condition holds:
Fu b, u >0 (<0)
for all u span{w
i
}

i=1
such that u =.
Theorem 1. Under assumption (P1), problem (P) has at least one solution.
Proof. We rst consider the case when the inequality in (P1) is greater than zero. Let
V
m
= span{w
1
, w
2
, . . . , w
m
}.
Consider the problem: Find u
m
V
m
such that
Fu
m
, v = b, v v V
m
. (P
m
)
We show that (P
m
) has a solution u
m
V
m
. Dene the operator T
m
: V
m
V
m
by
T
m
u, v = Fu, v b, v v V
m
.
The weak continuity of F and the nite dimensionality of V
m
imply that T
m
is continuous. Fur-
thermore, T
m
u, u >0 for all u B
m
(0, ). A consequence of Brouwers xed point theorem
(see, e.g., [4]) then gives us that T
m
has a zero u
m


B
m
(0, ).
Hence, we have a uniformly bounded sequence {u
m
}

m=1
such that
Fu
m
, w
i
= b, w
i
i m.
Since {u
m
}

m=1
is weakly compact, we get a subsequence (denoted {u
m
k
}

k=1
) and an element
u H such that u
m
k
u. By the weak continuity of F we get Fu
m
k
Fu. Fixing i and taking
the limit as k gives
Fu, w
i
= b, w
i
.
Since i is arbitrary and {w
i
}

i=1
is complete, Fu =b.
Finally note that if the inequality in (P1) is less than zero then we let

F = F and

b = b. In
this case

Fu

bu, u = Fu b, u >0. 2
The solvability of problem (Q) can be proven in a similar fashion under the following assump-
tion:
(Q1) There exists >0 such that the following coercivity condition holds:
Fu, u >
2
_
<
2
_
for all u span{w
i
}

i=1
such that u =.
Note that condition (Q1) implies that Fu u, u > 0 (< 0) for all u span{w
i
}

i=1
such
that u =, which is exactly what we need for the proof of Theorem 1 to work in this case. We
thus have:
Theorem 2. Under assumption (Q1), problem (Q) has at least one solution.
For problem (R) we make the following assumption:
M. El-Gebeily, D. ORegan / J. Math. Anal. Appl. 324 (2006) 344357 347
(R1) There exists >0 such that the following coercivity condition holds:

Lu Fu, u >0
for all u span{w
i
}

i=1
such that u =.
Theorem 3. Under assumption (R1), problem (R) has at least one solution.
Proof. Recall that in this case {w
i
}

i=1
D(

L). Following the same steps as in the proof of

Theorem 1 we obtain a sequence {u
m
}

m=1
D(

L)

B(0, ) such that
u
m
,

Lw
i
=

Lu
m
, w
i
= Fu
m
, w
i
i m. (4)
Since {u
m
}

m=1
is weakly compact, we get a subsequence (denoted {u
m
k
}

k=1
) and an element
u H such that u
m
k
u. By the weak continuity of F we get Fu
m
k
Fu. Since u
m
k
also
satises (4), xing i and taking the limit as k in (4) gives
u,

Lw
i
= Fu, w
i
. (5)
Since i is arbitrary, Eq. (5) holds also for all v

D
1
= span{w
i
}

i=1
. Dene an antilinear func-
tional f on

D
1
by
f (v) = u,

Lv.
Equation (5), applied to elements v of

D
1
, then gives

f (v)

u,

Lv

Fu, v

Fuv.
Since

D
1
is dense in H, it follows that f extends to a continuous antilinear functional on all of H,
and, in particular on D(

L). Thus the functional v u,

Lv is continuous on D(

L). Therefore,
u D(

L). Consequently,

Lu, v = Fu, v
for all v H. Therefore,

Lu =Fu. 2
We also have strong convergence of the sequence {u
m
k
} to u in the following special case.
Proposition 4. Let {u
m
} be the sequence dened in the proof of Theorem 3. If, in addition to (R1),

L has a completely continuous inverse and the sequence {w

i
}

i=1
is taken to be the eigenvectors
of

L, then there is a subsequence {u
m
k
} that converges in the norm of H to a solution of prob-
lem (R).
Proof. If {w
i
}

i=1
is the sequence of eigenvectors of

L, then the spaces V
m
are all invariant
under

L. This is equivalent to the commutativity of

Lwith the projection operators P
m
: HV
m
.
Therefore, equality

Lu
m
, v = Fu
m
, v v V
m
(6)
can be restated as
u
m
, v =
_

L
1
Fu
m
, v
_
v V
m
. (7)
348 M. El-Gebeily, D. ORegan / J. Math. Anal. Appl. 324 (2006) 344357
Indeed, Eq. (6) can be written in the form
P
m

Lu
m
=P
m
Fu
m
.
Since

L commutes with P
m
, we have
u
m
=P
m
u
m
=P
m

L
1
Fu
m
,
which is equivalent to (7).
Let {u
m
k
} be a subsequence of {u
m
} converging weakly to u H. By the proof of Theorem 3,
u is a solution of (R). We want to establish that {u
m
k
} converges strongly to u. The weak continu-
ity of F implies the weak convergence of Fu
m
k
to Fu. Therefore {Fu
m
k
} is norm bounded. The
complete continuity of

L
1
implies that, for a subsequence (still denoted {Fu
m
k
}), {

L
1
Fu
m
k
}
converges strongly to some y H. Now
u
m
k
y =
_
_
P
m
k

L
1
Fu
m
k
y
_
_

_
_
P
m
k
_

L
1
Fu
m
k
y
__
_
+
_
_
(E P
m
k
)y
_
_

_
_
L
1
Fu
m
k
y
_
_
+
_
_
(E P
m
k
)y
_
_
,
where E is the identity operator. Since P
m
k
y y, u
m
k
y. Hence, u
m
k
y. By the uniqueness
of the weak limit, y =u. 2
It is to be observed here that the conditions we assumed about

L and F include the possibility
of resonance. Next, we give a uniqueness result.
Theorem 5. Assume that

L is strictly positive (

Lu, u > 0 for u = 0) and F is monotone de-
creasing (Fu Fv, u v 0); then problem (R) can have at most one solution.
Proof. Suppose problem (R) has two solutions u, v such that u =v. Then, on the one hand
_

L(u v), u v
_
>0,
and on the other hand,
Fu Fv, u v 0.
Since

L(u v) =Fu Fv, we have the contradiction 0 <0. 2
3. Upper and lower solutions and existence
In this section we discuss the notion of upper and lower solutions for abstract operators in a
Hilbert space H with a cone. We show that if problem (R) has a lower solution and an upper
solution , then it has a solution u in between.
Let H be a separable Hilbert space. Suppose also that the norm in H is monotone and that
it is partially ordered by a regular minihedral cone K (see [3] or [5]). The partial order relation
induced by K is denoted in the usual way: u v means v u K. We denote by K

the dual
cone corresponding to K. The order in K

will be denoted by : u v means v u K

. We
will also need the following special notation.
Denition 6. For u, v H, we will say that u v if u, z = v, z for all z K.
M. El-Gebeily, D. ORegan / J. Math. Anal. Appl. 324 (2006) 344357 349
If F
1
and F
2
are operators on H, we will say that F
1
F
2
if F
1
u F
2
u for every u in their
common domain. Analogous denitions are given to F
1
F
2
and F
1
F
2
. The following lemma
shows that given any u
0
H we can always nd a majorant u H such that u u
0
. The proof
is straightforward.
Lemma 7. Given u H, let M
u
= {v H: v u}. Then M
u
= u +K

.
For H dene the lower -trimmer operator T

: HH by
T

u = sup{u, }
and dene the upper -trimmer operator T

: HH by
T

u = inf{u, }.
The following are some elementary properties of the trimmer operators.
Proposition 8. For H, we have the following properties:
(1) T

u u, T

u , T

u u and T

u for all u H.
(2) T

u =u for all u H, such that u and T

u =u for all u H, such that u .

(3) For u, v H with u v, T

u T

v and T

u T

v.
Proof. We prove property (3) only. By property (1), T

u u v and T

u . Therefore,
T

u is a lower bound for {v, }. Since K is minihedral, T

u inf{v, } = T

v. Similarly,
T

u T

v. 2
Assume the cone K has the property
_
T
0
u, T
0
u
_
= 0
for all u H. It can easily be checked that
u, T
0
u = T
0
u, T
0
u.
Proposition 9. Suppose , H such that . Then, for any u H, precisely one of the
following statements is true:
(1) T

u >;
(2) u ; or
(3) T

u <.
Proof. Let u H. Since T

u , either T

u > or T

u =. In the rst case, there is nothing

to prove. In the second case, u T

u =. In this case we consider T

u. Since T

u , either
T

u < or T

u = . In the rst subcase we are done; in the second subcase u T

u = .
Therefore, u . 2
Given , H such that , we dene the operator Q by
Qu =
_
if T

u >,
u if u ,
if T

u <.
350 M. El-Gebeily, D. ORegan / J. Math. Anal. Appl. 324 (2006) 344357
Proposition 10. Q is continuous and QH= [, ]. Consequently, QH is bounded.
Proof. Suppose u H and {u
n
} is a sequence converging to u. We want to show Qu
n
Qu. If
T

u > , then by the continuity of T

, T

u
n
> for all n sufciently large. Therefore, Qu
n
=
b = Qu. A similar argument holds in T

u < . If u , we claim that T

u
n
> for at
most nitely many values on n. If not then suppose T

u
n
k
> , k = 1, 2, . . . . Since u
n
k
u as
k , T

u > , a contradiction. Similarly, T

u
n
< for at most nitely many values of n.
It follows that
n
u
n
for all sufciently large n. Therefore, Qu
n
= u
n
u = Qu. It is
obvious that QH= [, ]. In order to show that QH is bounded, we need to show that [, ] is
bounded. For u [, ], we have u . Hence, 0 u . Since the norm in H is
monotonic, u . Therefore, u +. Thus, [, ] is bounded. 2
Denition 11. A function D(

L) is called a lower solution of (R) if

L F. A function
D(

L) is called an upper solution of (R) if

L F.
Our assumptions on

L are now modied to:
(R1

) The operator

L is bounded below: there exists
0
R such that

Lu, u
0
u
2
for all
u D(

L).
(R2

)

L is T
0
positive:

Lu, T
0
u 0 for all u D(

L).
Theorem 12. Assume (R1

) and (R2

) hold (recall that F is assumed to be weakly continuous).

Suppose that problem (R) has a lower solution and an upper solution such that . Then
(R) has a solution u such that u .
Proof. Choose >
0
and consider the modied problem

Lu +u =F

u, (8)
where F

:= (F +)Q. Note that QH is weakly compact since it is bounded and closed in H.

Since operator F + is weakly continuous, F

is bounded (say by M) on H. Furthermore,

Lu +u F

u, u (
0
+)u
2
Mu >0
for u = := 2M/(
0
+ ). With V
m
dened as in Theorem 1, consider the problem: Find
u
m
V
m
such that
_
(

L+)u
m
, v
_
= F

u
m
, v v V
m
. (9)
We show that (P
m
) has a solution u
m
V
m
such that u
m
. With T
m
: V
m
V
m
dened
by
T
m
u, v =
_
(

L+)u F

u, v
_
v V
m
,
the same reasoning in Theorem 1 gives us that (P
m
) has a solution u
m
V
m
. We claim that
u
m
. If not, then either T

u
m
> or T

u
m
< . If T

u
m
> let z = u
m
. Since
T
0
z = 0, T
0
z >0. Also,
0

Lz, T
0
z = F

u
m
u
m


L, T
0
z = F z

L, T
0
z z, T
0
z
= T
0
z
2
<0,
M. El-Gebeily, D. ORegan / J. Math. Anal. Appl. 324 (2006) 344357 351
a contradiction. We get a similar contradiction if T

u
m
< holds. Again, as in Theorem 1 we
get a solution u of problem (R) which is the weak limit of a subsequence of {u
m
}. Since the cone
K is weakly closed and u
m
, it follows that u . 2
4. The abstract quasilinearization method
In this section we consider two generalized quasilinearization methods [1,6] for problem (R).
We retain the assumptions made in Section 3 about the Hilbert space H.
For the next theorem we need the following denition of w-convex operators.
Denition 13. Suppose C H is a convex set and H : C H. We will say that H is w-convex
if
H
_
(1 )u +v) (1
_
Hu +Hv (10)
for all [0, 1], u, v C.
Theorem 14. Let C H be convex and suppose that H : C H has a Gateaux derivative H

on C. The following are equivalent:

(1) H is convex;
(2) Hu Hv +H

u(v u) for all u, v C.

Proof. The proof is an adaptation of [2, Theorem 4.3.16].
(1) (2). Let u, v C, (0, 1). Inequality (10) may be rewritten as
1

_
H
_
u +(v u)
_
Hu
_
Hv Hu.
Taking the limit as 0
+
and noting the continuity of the inner product, we get
H

u(v u) Hv Hu.
(2) (1). Let u, v C and [0, 1]. Then
Hv H
_
u +(v u)
_
+(1 )H

_
u +(v u)
_
(v u)
and
Hu H
_
u +(v u)
_
+H

_
u +(v u)
_
(u v).
Multiplying the rst inequality by and the second by (1 ) and adding we get
Hv +(1 )Hu H
_
u +(v u)
_
. 2
Theorem 15. Assume (R1

) and (R2

) hold (recall that F is assumed to be weakly continuous).

Assume that
0
,
0
D(

L) are lower and upper solutions of (R), respectively, such that

0

0
.
Then there exists a monotone nondecreasing sequence {
n
} which converges in H to a solution
of (R). Moreover, if F is monotone decreasing and
0
>0, the convergence is quadratic.
Proof. Let C = [
0
,
0
] and let H : H H be convex with two weakly continuous Frchet
derivatives H

: HL(H), H

: HL(H) L(H). Since C is weakly compact, H

, H

are
uniformly bounded on C. That is
H

u, H

u M
352 M. El-Gebeily, D. ORegan / J. Math. Anal. Appl. 324 (2006) 344357
for all u C. Dene the operators : HH and G: HHH by
u =Hu Fu and G(u, v) =Fv +H

v(u v) [u v].
It is easy to check that and G are weakly continuous. Furthermore, by Theorem 14,
Fu G(u, v) for all u, v C.
Let k = 0 and consider the equation

Lu =G(u,
k
). (11)
Since

L
0
F
0
= G(
0
,
0
) and

L
0
F
0
G(
0
,
0
),
0
,
0
are lower and upper solu-
tions for (11). By Theorem 12, (11) has a solution
1
C. Furthermore,

L
1
=G(
1
,
0
) F
1
.
Therefore,
1
,
0
are lower and upper solutions of (R). Repeating the same steps with k =
1, 2, . . . , we obtain a sequence {
n
} in D(

L) C such that

L
n
=G(
n
,
n1
),
0

1

n

0
.
The regularity of the cone K implies that there is an C such that
n
in the norm
of H. Hence, G(
n
;
n1
) F. Now, since

L
n
=G(
n
;
n1
),
for any v D(

L),

n
,

Lv =
_
G(
n
;
n1
), v
_
.
Thus, taking the limit on both sides as n , we get
,

Lv = F, v.
By an argument similar to that in the proof of Theorem 3, we can show that D(

L) and

L =F.
To show that the convergence is quadratic observe rst that H has the representation
Hu =Hv +H

v(u v) +
1
_
0
(1 )
2
2
H

_
v +(u v)
_
(u v)
2
d
for all u, v H. Using this representation for H, the operator G can be written as
G(u, v) =Fu
1
_
0
(1 )
2
2
H

_
v +(u v)
_
(u v)
2
d.
Let e
n
=
n
. Then

Le
n
=F G(
n
,
n1
) =F F
n
+
1
_
0
(1 )
2
2
H

n1
+(
n

n1
)
_
(
n

n1
)
2
d.
Therefore, using the monotonicity assumption on F, we get
M. El-Gebeily, D. ORegan / J. Math. Anal. Appl. 324 (2006) 344357 353

Le
n
, e
n

1
_
0
(1 )
2
2
_
_
H

n1
+(
n

n1
)
__
_

n

n1

2
e
n
d

1
6
max
uC
_
_
H

(u)
_
_

n

n1

2
e
n
.
From this inequality and the strong positivity of

L we conclude that
e
n
M
n

n1

2
.
Since
n1

n
, we have 0
n

n1

n1
. The monotonicity of the norm in H
gives
n

n1
e
n1
and so
e
n
Me
n1

2
. 2
In the above theorem, the sequence {
n
} is generated as solutions of nonlinear equations.
However, under stronger conditions on the function F, one can work with linear equations and
still retain quadratic convergence. This is important, e.g., for constructing numerical approxima-
tions. We discuss this theory now. We begin with the following denition.
Denition 16. Let K be a subset of H and assume that F : K H.
(1) F is called w-increasing if, for every u, v K such that u v, Fu Fv.
(2) F is called w-decreasing if, for every u, v K such that u v, Fu Fv.
(3) F is called w-positive if Fu 0 for all u K. In this case we will write F 0.
(4) F is called w-negative if Fu 0 for all u K. In this case we will write F 0.
The following two lemmas can be proven by standard techniques.
Lemma 17. Suppose F : K H is Frchet differentiable. Then F has the w-mean value
property: Given u, v K, there exists (0, 1) such that Fu Fv F

w(u v), where

w =u +(1 )v.
Lemma 18. Suppose F : K H has a w-negative (w-positive) Frchet derivative F

. Then F is
w-decreasing (w-increasing).
We now strengthen the assumptions on

L and F (the right-hand side of problem (R)) to:
(R1

)

L is strongly positive: (
0
>0 in (R1

));
(R3

) F has two weakly continuous Frchet derivatives on C and F

is w-negative.
In the course of the proof of the next theorem we will need to be able to choose certain
positive majorant operators. For such choices to be possible, we will need to assume the following
property about the cone K

.
(R4) For every u H,
M
u
K

=, (12)
where M
u
is the majorant set of u (see Lemma 7).
354 M. El-Gebeily, D. ORegan / J. Math. Anal. Appl. 324 (2006) 344357
One condition under which (12) holds: K

is also minihedral with respect to the partial or-

der . In this case, we can take u
+
= sup

{u, 0}. It also holds if K

has an interior point. This is

because if u
0
is an interior point of K

, then there is an r > 0 such that B(u

0
, r) K

. Hence,
for some >0, u
0
+u B(u
0
, r) K

. It follows that u +
1
u
0
M
u
K

.
Lemma 19. Suppose (12) is satised.
(1) Let K H be weakly compact. Then K has a positive majorant.
(2) Let {T

be a uniformly bounded family of linear operators on H. Then there is a w-

positive bounded linear operator T
0
such that T
0
T

, .
(3) Let {F

be a uniformly bounded family of bilinear operators on H. Then there is a

w-positive bounded bilinear operator F
0
such that F
0
F

, . (This means that

F
0
uv, w F

uv, w for all u, v, w K.)

Proof. (1) We will show rst that every chain in K has a majorant. Let u
1
u
2
be a chain
in K. Since K is weakly compact, we have u
k
i
u. We claim that u is a majorant of the chain.
To see this, let z K. Then
u u
k
i
, z = lim
j
u
k
j
u
k
i
, z 0.
It follows that u u
k
i
for all i. Consequently, u u
k
for all k. It follows from Zorns lemma that
K has a majorant u
0
. Since (12) is satised, we can nd a majorant u
+
of u
0
in K

.
(2) We rst show that there is a bounded linear operator T such that T T

, . We use
again Zorns lemma. Let T

1
T

2
be a chain in M := {T

. Dene the operator T

1
by
T
1
u = lim
n
T

n
u.
Then T
1
is linear and bounded. Furthermore, for u, v K,
T
1
u T

k
u, v = lim
n
T

n
u T

k
u, v 0.
Thus T
1
T

k
for all k. It follows from Zorns lemma that there is a bounded linear operator T
such that T T

, .
Next, we construct a w-positive majorant of T . For this we consider the set
B =
_
u K: u = 1
_
.
Since B is weakly compact, it has a positive majorant u
0
. Observe that, for any v B,
u
0
, v v, v = 1. Dene the operator T
0
by T
0
u = T u
0
, uu
0
. Then T
0
is linear and
bounded. Furthermore, for u, v B,
T
0
u, v = T u
0
, uu
0
, v T T u T u, v.
(3) The proof is exactly the same as that of (2) except for the choice of F
0
. In this case we
dene it by
F
0
uv =Mu, u
0
v, u
0
u
0
,
with M being an upper bound of the norms of the members of {F

. 2
M. El-Gebeily, D. ORegan / J. Math. Anal. Appl. 324 (2006) 344357 355
Theorem 20. Assume (R1

), (R2

), (R3

) and (R4) hold and problem (R) has a lower and an up-
per solution
0
,
0
D(

L), respectively such that

0

0
. Then there exist monotone sequences
{
n
}, {
n
} which converge in H to a solution of (R). Moreover, the convergence is quadratic.
Proof. Observe rst that F and F

are continuous since they have Frchet derivatives. However,

all we need here is the weak continuity of both. Let C = [
0
,
0
] and choose H : HHwith two
weakly continuous Frchet derivatives such that, u C, H

u is a bounded w-positive bilinear

operator and H

u F

u. This choice of H is possible because {F

u: u C} is a uniformly
bounded set of bilinear operators (since C is closed and bounded, and thus weakly compact and
F

is weakly continuous). In fact, by Lemma 19, H can be chosen such that H

is independent
of u.
Set
=F H.
Then, for any u, v C,

u 0, (13)
Hu Hv +H

v(u v), (14)

u v +

u(u v), (15)

Fu Fv +H

v(u v) +[u v] (16)

Fv +[H

v +

u](u v). (17)

Let us show, for example, (15). We have, by the w-mean value property
v u +

u(v u) +
1
2

w(v u)
2
,
where w = u + (1 )v for some (0, 1). By the choice of H, we have, for any u, v C
and z in K,
_

w(uv), z
_
=
_
F

w(uv), z
_

_
H

w(uv), z
_
0.
Hence,

w 0. It follows that
v u +

u(v u).
We get (15) upon rearranging the terms in the above inequality.
Let k = 0 and consider the problems

Lu =G(u;
k
,
k
) (18)
and

Lu =D(u;
k
,
k
), (19)
where
G(u; , ) =F +[H

](u )
and
D(u; , ) =F +[H

](u ).
356 M. El-Gebeily, D. ORegan / J. Math. Anal. Appl. 324 (2006) 344357
Then

L
0
F
0
=G(
0
;
0
,
0
)
and, by inequality (17),

L
0
F
0
G(
0
;
0
,
0
).
Therefore, by Theorem 12, problem (18) has a solution
1
such that
0

0
. On the other
hand, we have

L
0
F
0
=D(
0
;
0
,
0
)
and

L
1
=G(
1
;
0
,
0
) =F
0
+[H

0
+

0
](
1

0
)
F
0
[H

0
+

0
](
0

0
) +[H

0
+

0
](
1

0
)
=F
0
+[H

0
+

0
](
1

0
) =D(
1
;
0
,
0
).
Therefore,
1
and
0
are lower and upper solutions for (19). Thus, by Theorem 12, (19) has a
solution
1

1
. Furthermore,

L
1
=D(
1
;
0
,
0
) =F
0
+[H

0
+

0
](
1

0
)
F
0
[H

1
+

0
](
0

1
) F
1
and

L
1
=G(
1
;
0
,
0
) =F
0
+[H

0
+

0
](
1

0
)
F
0
+[H

0
+

1
](
1

0
) F
1
.
Hence,
1
,
1
are lower and upper solutions for (R). Repeating the same arguments with k =
1, 2, . . . , we obtain a sequence of solutions {
n
} of (18) and a sequence of solutions {
n
} of (19)
such that, for all n,

1

n

n

1

0
.
The monotonicity of the norm in Hgives that
n
and
n
in Hwith . Further-
more, since G is weakly continuous, G(
n
;
n1
,
n1
) F. Now, since

L
n
=G(
n
;
n1
,
n1
),
for any v D(

L),

n
,

Lv =
_
G(
n
;
n1
,
n1
), v
_
.
Thus, taking the limit on both sides as n , we get
,

Lv = F, v.
By an argument similar to that in the proof of Theorem 3, we can show that D(

L) and

L =F. Similarly,

L =F. We claim that =. Indeed, ( ) K and

2

1
0
_

L( ),
_
=
1
0
F F,
=
1
0
_
F

w( ),
_
0.
To show the quadratic rate of convergence, dene the error functions
M. El-Gebeily, D. ORegan / J. Math. Anal. Appl. 324 (2006) 344357 357
e
n
=
n
, r
n
=
n
.
Then
F G(
n
;
n1
,
n1
)
=F
_
F
n1
+[H

n1
+

n1
](
n

n1
)
_
=
_
H H
n1
H

n1
(
n1
)
_
+
_

n1

n1
(
n1
)
_
+[H

n1
+

n1
](
n
)

1
2
H

w
1
e
2
n1
{

n1

n1
}e
n1
+[H

n1
+

n1
]e
n

1
2
H

w
1
e
2
n1

w
2
(
n1

n1
)e
n1
+F

n1
e
n

1
2
H

w
1
e
2
n1

w
2
(e
n1
+r
n1
)e
n1
,
where w
1
=
1
+(1
1
)
n1
and w
2
=
2

n1
+(1
2
)
n1
,
1
,
2
(0, 1).
Since e
n
K, using the boundedness of H

on C, we have

Le
n
, e
n

_
1
2
H

w
1
e
2
n1

w
2
(e
n1
+r
n1
)e
n1
, e
n
_

_
_
_
_
1
2
H

w
1
e
2
n1

w
2
(e
n1
+r
n1
)e
n1
_
_
_
_
e
n

M
_
e
n1

2
+e
n1
r
n1

_
e
n
M
_
e
n1

2
+r
n1

2
_
e
n
.
From the strong positivity of

L, we get
e
n

M

0
_
e
n1

2
+r
n1

2
_
.
Similarly,
r
n

M

0
_
e
n1

2
+r
n1

2
_
.
This establishes the quadratic convergence of the iterates. 2
References
[1] B. Ahmad, J. Nieto, N. Shahzad, Generalized quasilinearization method for mixed boundary value problems, Appl.
Math. Comput. 133 (2002) 423429.
[2] K. Atkinson, W. Han, Theoretical Numerical Analysis: AFunctional Analysis Framework, Texts Appl. Math., vol. 39,
Springer, New York, 2001.
[3] K. Deimling, Nonlinear Functional Analysis, Springer, Berlin, 1985.
[4] A. Granas, J. Dugundji, Fixed Point Theory, Springer, New York, 2003.
[5] M.A. Krasnoselskij, J.A. Lifshits, A.V. Sobolev, Positive Linear SystemsThe Method of Positive Operators, Sigma
Ser. Appl. Math., vol. 5, Heldermann, Berlin, 1989.
[6] V. Lakshmikantham, A. Vatsala, Generalized Quasilinearization for Nonlinear Problems, Math. Appl., Kluwer, Dor-
drecht, 1998.
[7] J.J. Nieto, Generalized quasilinearization method for a second order ordinary differential equation with Dirichlet
boundary conditions, Proc. Amer. Math. Soc. 125 (1997) 25992604.