Professional Documents
Culture Documents
George A. Anastassiou, Ioannis K. Argyros-Functional Numerical Methods. Applications To Abstract Fractional Calculus-Springer (2018)
George A. Anastassiou, Ioannis K. Argyros-Functional Numerical Methods. Applications To Abstract Fractional Calculus-Springer (2018)
George A. Anastassiou, Ioannis K. Argyros-Functional Numerical Methods. Applications To Abstract Fractional Calculus-Springer (2018)
Argyros
Functional Numerical
Methods: Applications
to Abstract Fractional
Calculus
123
George A. Anastassiou Ioannis K. Argyros
Department of Mathematical Sciences Department of Mathematical Sciences
University of Memphis Cameron University
Memphis, TN Lawton, OK
USA USA
This is the abstract sequel booklet monograph to the recently published mono-
graphs, by the same authors, titled: Intelligent Numerical Methods: Applications to
Fractional Calculus, Studies in Computational Intelligence 624, and Intelligent
Numerical Methods II: Applications to Multivariate Fractional Calculus, Studies in
Computational Intelligence 649, both in Springer Heidelberg New York, 2016. It is
regarding applications of Newton-like and other similar methods for solving
abstract functional equations, which involve abstract Caputo and Canavati type
fractional derivatives. The functions we are dealing with are Banach space valued
of a real domain. These are studied for the rst time in the literature, and chapters
are self-contained and can be read independently. In each chapter, the rst sections
are prerequisites for the nal section of abstract fractional calculus applications.
This short monograph is suitable to be used in related graduate classes and research
projects. We exhibit the maximum of these numerical methods at the abstract
fractional level.
The motivation to write this monograph came by the following: Various issues
related to the modeling and analysis of fractional order systems have gained an
increased popularity, as witnessed by many books and volumes in Springers
program:
http://www.springer.com/gp/search?query=fractional&submit=Prze%C5%9Blij
and the purpose of our book is to provide a deeper formal analysis on some issues
that are relevant to many areas for instance: decision making, complex processes,
systems modeling and control, and related areas. The above are deeply embedded in
the elds of engineering, computer science, physics, economics, social and life
sciences.
The list of covered topics here follows:
explicitimplicit methods with applications to Banach space valued functions in
abstract fractional calculus,
convergence of iterative methods in abstract fractional calculus,
equations for Banach space valued functions in fractional vector calculi,
vii
viii Preface
ix
x Contents
Explicit iterative methods have been used extensively to generate a sequence approx-
imating a solution of an equation on a Banach space setting. However, little attention
has been given to the study of implicit iterative methods. We present a semi-local con-
vergence analysis for a some general implicit and explicit iterative methods. Some
applications are suggested including Banach space valued functions of fractional
calculus, where all integrals are of Bochner-type. It follows [5].
1.1 Introduction
Secant method: 1
xn+1 = xn xn1 , xn ; F F (xn ) , (1.1.3)
where E n = E (F) (xn ) and E : L (B1 , B2 ) the space of bounded linear oper-
ators from B1 into B2 . Other explicit methods can be found in [7, 11, 14, 15] and
the references there in.
Implicit Methods [6, 9, 11, 15]:
xn+1 = xn A1
n F (x n ) , (1.1.6)
Pz (x) = x, (1.1.7)
where
Pz (x) = x + F (z) + A (x, z) (x z) (1.1.8)
or
Pz (x) = z A (x, z)1 F (z) (1.1.9)
and
xn+1 = xn A (xn , xn1 )1 F (xn ) . (1.1.14)
In this chapter in Sect. 1.2, we study the semi-local convergence of method (1.1.5)
and method (1.1.6). Section 1.3 contains the semi-local convergence of method
(1.1.11), method (1.1.12), method (1.1.13) and method (1.1.14). Some applications
to Abstract Fractional Calculus are suggested in Sect. 1.4 on a certain Banach space
valued functions, where all the integrals are of Bochner-type [8].
The semi-local convergence analysis of method (1.1.6) that follows is based on the
conditions (H ):
(h 1 ) F : B1 B2 is continuous and A (F) (x, y) L (B1 , B2 ) for each
(x, y) .
(h 2 ) There exist l > 0 and 0 B1 such that A (F) (x, y)1 L (B2 , B1 ) for
each (x, y) 0 0 and
A (F) (x, y)1 l 1 .
Set 1 = 0 .
(h 3 ) There exist real numbers 1 , 2 , 3 satisfying
0 2 1 and 0 3 < 1
(h 6 ) h := 1 21 (1 3 )2 .
and
(h 7 ) U (x0 , t ) 0 , where
4 1 Explicit-Implicit Methods with Applications to Banach Space
13 (13 )2 2h
t = 1
, 1
= 0
1
13
, 1 = 0.
Then, we can show the following semi-local convergence result for method (1.1.6)
under the preceding notation and conditions (H ).
Theorem 1.1 Suppose that the conditions (H ) are satisfied. Then, sequence {xn }
generated by method (1.1.6) starting at x0 is well defined in U (x0 , t ), remains
in U (x0 , t ) for each n = 0, 1, 2, ... and converges to a solution x U (x0 , t )
of equation F (x) = 0. Moreover, provided that (h3 ) holds with A(F)(z,y) replacing
A(F)(x,y) for each z 1 , if 1
= 0, the equation F (x) = 0 has a unique solution
, where
x in U
= U (x0 , t ) 0 , if h = 21(1 3 ) 2
1 2
U
U (x0 , t ) 0 , if h < 2 (1 3 )
and, if 1 = 0, the solution x is unique in U x0 , 1 , where t =
3
13 + (13 )2 2h
1
.
1 2
Proof Case 1
= 0. Let us define scalar function g on R by g (t) = 2
t
(1 3 ) t + and majorizing sequence {tn } by
It follows from (h 6 ) that function g has two positive roots t and t , t t , and
tk tk+1 . That is, sequence {tk } converges to t .
(a) Using mathematical induction on k, we shall show that
xk+1 xk = A1
k (F (x k ) F (x k1 ) Ak1 (x k x k1 )) . (1.2.4)
and
xk x0 tk t . (1.2.8)
It follows by (1.2.7) and (1.2.8) that {xk } is a complete sequence in a Banach space
B1 and as such it converges to some x U (x0 , t ) (since U (x0 , t ) is a closed
set). By letting k +, using (h 1 ) and (h 2 ), we get l 1 lim F (xk ) = 0, so
k+
F (x ) = 0.
be such that F (x ) = 0. We shall show by induction that
Let x U
x xk t tk for each k = 0, 1, 2, ... . (1.2.9)
. Suppose that
Estimate (1.2.9) holds for k = 0 by the definition of x and U
x xk t tk . Then, as in (1.2.5), we obtain in turn that
x xk+1 = x xk + A1 F (xk ) A1 F x =
k k
1
A Ak x xk + F (xk ) F x
k
1
A F x F (xk ) Ak x xk
k
1
x xk + 2 xk x0 + 3 x xk
2
t tk + 2 tk + 3 t tk =
1
2
6 1 Explicit-Implicit Methods with Applications to Banach Space
1 2 1
t + (tk )2 1 tk t + 2 t tk tk + 3 t tk =
2 2
1 2
+ (1 3 ) t + t 1 tk t + 2 tk t 2 tk2 + 3 t 3 tk
2 k
= t tk+1, (1.2.10)
which completes the induction for estimate (1.2.9). Hence, lim xk = x . But we
k+
showed that lim xk = x , so x = x .
k+
Case 1 = 0. Then, we have by (h 3 ) that 2 = 0 and estimate (1.2.5) gives
and
xk+1 x0 xk+1 xk + xk xk1 + + x1 x0
1 3k+1
< . (1.2.12)
1 3 1 3
Then, as in the previous case it follows from (1.2.11) and (1.2.12) that
1 i3 k
xk+i xk , (1.2.13)
1 3 3
so sequence {xk } is complete and x solves equation F (x) = 0. Finally, the unique-
ness part follows from (1.2.10) for 1 = 2 = 0, since
x xk+1 3 x xk k+1 x x0 k+1 , (1.2.14)
3 3
1 3
(2) Suppose that there exist l0 > 0, 4 > 0 and L L (B1 , B2 ) with L 1
L (B2 , B1 ) such that L 1 l01
and
5 := l01 4 < 1.
Then, it follows from the Banach lemma on invertible operators [7, 9, 11, 14, 15]
and 1
L A (F) (x, y) L l 1 4 = 5 < 1
0
l 1
that A (F) (x, y)1 L (B2 , B1 ). Set l 1 = 1 0
5
. Then, under these replacements,
condition (h 2 ) is implied, so it can be dropped from the conditions (H ).
(3) Clearly method (1.1.5) converges under the conditions (H ), since (1.1.6)
implies (1.1.5).
(4) Let R > 0 and define R0 = sup {t [0, R) : U (x0 , R0 ) D}. Set 0 =
U (x0 , R0 ). Condition (h 3 ) can be extended, if the additional term a2 x x0
is inserted inside the paranthesis at the right hand side for some a2 0. Then,
the conclusions of Theorem 1.1 hold in this more general setting, provided that
a3 = a2 R0 + 3 replaces 3 in conditions (h 6 ) and (h 7 ).
(5) Concerning the solvability of Eq. (1.1.6) (or (1.1.5)), we wanted to leave condi-
tion (h 4 ) as uncluttered as possible in conditions (H ). We did this because in practice
these equations may be solvable in a way other than using the contraction mapping
principle already mentioned earlier.
Next, we show the solvability of method (1.1.5) using a stronger version of the
contraction mapping principle and based on the conditions (C) :
(c1 ) = (h 1 ) .
(c2 ) There exist 0 [0, 1), 1 [0, +), 2 [0, 1), x0 such that for each
x, y, z
I + A (x, z) A (y, z) 0 ,
(c3 )
0 + 1 x0 + 2 1 for 2
= 0,
(1 (0 + 1 x0 ))2
F (x0 ) for 1
= 0,
1
8 1 Explicit-Implicit Methods with Applications to Banach Space
0 < 1 for 1 = 0
and
(c4 ) U (x0 , r ) , where
F (x0 ) 1 (0 + 1 x0 )
r < for 1
= 0,
1 (0 + 1 x0 ) 1
F (x0 )
r for 1 = 0,
1 0
1 (0 + 1 x0 )
r< for z = x0 , 1
= 0.
1
Theorem 1.3 Suppose that the conditions (C) are satisfied. Then, for each n =
0, 1, 2, ... Eq. (1.1.5) is unique solvable. Moreover, if A1 n L (B2 , B1 ), then
Eq. (1.1.6) is also uniquely solvable for each n = 0, 1, 2, ...
Proof We base the proof on the contraction mapping principle. Let x, y U (x0 , r ).
Then, using (1.1.8) we have in turn by (c2 ) that
Pz (x) Pz (y) = (I + A (x, z) A (y, z)) (x y) (A (x, z) A (y, z)) z
0 x y + 1 (z x0 + x0 ) x y
(x x0 ) x y , (1.2.15)
where
0 + 1 (t + x0 ) for z
= x0
(t) = (1.2.16)
0 + 1 x0 for z = x0 .
Moreover, if z
= x0 , the last condition in (c3 ), (c3 ) , (c4 ) and (1.2.17) give instead
of (1.2.18) that
(0 + 1 x0 + 2 ) r r. (1.2.19)
Then, again by (1.2.15), (1.2.19) and the contraction mapping principle, we guarantee
the unique solvability of Eq. (1.1.5) and the existence of a unique sequence {xn }
for each n = 0, 1, 2, ... Finally, equation (1.1.6) is also uniquely solvable by the
preceding proof and the condition A1 n L (B2 , B1 ).
Remark 1.4 (a) The gamma conditions can be weakened, if i are replaced by func-
tions i (t), i = 0, 1, 2, 3. Then, i will appear as i (x x0 ) and i (r ) in the
conditions (C) .
(b) Sect. 1.2 has an interest independent of Sect. 1.4. However, the results espe-
cially of Theorem 1.1 can apply in Abstract Fractional Calculus as suggested in
Sect. 1.4. As an example crucial condition (h 3 ) is satisfied in (1.4.8), if we choose
2 = 3 = 0 and l1 = 2c , where c is defined in (1.4.8). Similar choices can be given
for the rest of the special cases of (h 3 ) appearing in Sect. 1.4.
Theorem 1.1 is general enough so it can be used to study the semi-local convergence of
method (1.1.11), method (1.1.12), method (1.1.13) and method (1.1.14). In particular,
for the study of method (1.1.12) (and consequently method (1.1.11)), we use the
conditions
H :
h 1 F : B1 B2 is continuous and A (F) (x, x) L (B1 , B2 ) for each
x .
(h 2 ) There exist l > 0 and 0 B1 such that A (F) (x, x)1 L (B2 , B1 ) and
A (F) (x, x)1 l 1 .
Set 1 = 0 .
(h 3 ) There exist real numbers 1 , 2 , 3 satisfying
0 2 1 and 0 3
or
A (x, y) A (y, y) l5 x y .
(h 6 ) = (h 6 )
and
(h 7 ) = (h 7 ).
Then, we can show the following semi-local convergence of method (1.1.12) using
the conditions H and the preceding notation.
Proposition 1.5 Suppose that the conditions (H ) are satisfied. Then, sequence {xn }
generated by method (1.1.12) starting at x0 is well defined in U (x0 , t ), remains
in U (x0 , t ) for each n = 0, 1, 2, ... and converges to a solution x U (x0 , t )
of equation F (x) = 0. Moreover, if 1
= 0, the equation F (x) = 0 has a unique
solution x in U
, where
= U (x0 , t ) 0 , if h = 21 (1 3 )2
U
U (x0 , t ) 0 , if h < 21 (1 3 )2
and, if 1 = 0, the solution x is unique in U x0 , 1 3
, where t and t are given
in Theorem 1.1.
Proof Use in the proof of Theorem 1.1 instead of estimate (1.2.5) the analogous
estimate
0 2 1 and 0 3
(h 6 ) = (h 6 )
and
(h 7 ) = (h 7 ).
Then, we can present
the following semi-local convergence of method (1.1.14)
using the conditions H and the preceding notation.
Proposition 1.7 Suppose that the conditions (H ) are satisfied. Then, sequence {xn }
generated by method (1.1.14) starting at x0 is well defined in U (x0 , t ), remains
in U (x0 , t ) for each n = 0, 1, 2, ... and converges to a solution x U (x0 , t )
of equation F (x) = 0. Moreover, if 1
= 0, the equation F (x) = 0 has a unique
solution x in U
, where
U (x0 , t ) 0 , if h = 21 (1 3 )2
U=
U (x0 , t ) 0 , if h < 21 (1 3 )2
12 1 Explicit-Implicit Methods with Applications to Banach Space
and, if 1 = 0, the solution x is unique in U x0 , 1 3
, where t and t are given
in Theorem 1.1.
Proof As in Proposition 1.5, use in the proof of Theorem 1.1 instead of estimate
(1.2.5) the analogous estimate
F (xk ) =
Remark 1.8 Comments similar to Remark 1.2 (1)(3) can follow but for method
(1.1.13) and method (1.1.14) instead of method (1.1.5) and method (1.1.6), respec-
tively.
Here we deal with Banach space (X, ) valued functions f of real domain [0, a],
a > 0. All integrals here are of Bochner-type, see [8, 13]. The derivatives of f are
defined similarly to numerical ones, see [16], pp. 8386 and p. 93.
In this section we apply our Newton like numerical methods to X -valued fractional
calculus.
We want to solve
f (x) = 0. (1.4.1)
(I) Let 1 < < 2, i.e. = 2 ( ceiling of number); x, y [0, a], a > 0, and
f C 2 ([0, a] , X ).
We define the following left X -valued Caputo fractional derivatives (see [4]),
1 x
Dy f (x) := (x t)1 f (t) dt, (1.4.2)
(2 ) y
1.4 Applications to X-valued Fractional Calculus 13
when x y, and
1 y
Dx f (y) := (y t)1 f (t) dt, (1.4.3)
(2 ) x
By X -valued left fractional Caputo Taylors formula (see [4]) we get that
x
1
f (x) f (y) = f (y) (x y) + (x t)1 Dy
f (t) dt, for x > y,
() y
(1.4.5)
and
y
1
f (y) f (x) = f (x) (y x) + (y t)1 Dx
f (t) dt, for x < y,
() x
(1.4.6)
equivalently, it holds
y
1
f (x) f (y) = f (x) (x y) (y t)1 Dx
f (t) dt, for x < y.
() x
(1.4.7)
We would like to prove that
(x y)2
f (x) f (y) (A0 ( f )) (x, y) (x y) c , (1.4.8)
2
for any x, y [0, a], 0 < c < 1.
When x = y the last condition (1.4.8) is trivial.
We assume x
= y. We distinguish the cases:
(1) x > y : We observe that
1 x 1 x
(x t) 1
f (t) dt (x t) 1
f (x) dt
() Dy
()
Dy =
y y
(1.4.11)
(by [1], p. 426, Theorem 11.43)
x
1
(x t)1
Dy f (t) Dy f (x) dt
() y
(by [8])
1 x
(x t)1 Dy f (t) Dy f (x) dt =: () , (1.4.12)
() y
(assume that
D f (t) D f (x) 1 |t x|2 , (1.4.13)
y y
1
for any t, x, y [0, a] : x t y, where 1 < (), i.e. 1 := ()
< 1).
Therefore x
1
() (x t)1 (x t)2 dt (1.4.14)
() y
1 x
1 (x y)2 (x y)2
= (x t) dt = = 1 . (1.4.15)
() y () 2 2
(x y)2
f (x) f (y) (A0 ( f )) (x, y) (x y) 1 , (1.4.16)
2
where 0 < 1 < 1, and x > y.
(2) x < y : We observe that
(y x)1
f (x) +
Dx f (y) (x y)
=
( + 1)
1.4 Applications to X-valued Fractional Calculus 15
y (y x)
1 (y t)1 Dx
f (t) dt + Dx f (y) = (1.4.18)
() ( + 1)
x
1 y (y x)
(y t) 1
Dx f (t) dt
Dx f (y) = (1.4.19)
() ( + 1)
x
y y
1 1
(y t) 1
Dx f (t) dt (y t)1
Dx f (y) dt
=
() x () x
y
1
(y t) 1
Dx f (t) Dx f (y) dt
(1.4.20)
() x
1 y
(y t)1 Dx
f (t) Dx f (y) dt
() x
(by assumption,
D f (t) D f (y) 2 |t y|2 , (1.4.21)
x x
2 y
2 (x y)2
= (y t) dt = .
() x () 2
2
Assuming also 2 := ()
< 1 (i.e. 2 < ()), we have proved that
(x y)2
f (x) f (y) (A0 ( f )) (x, y) (x y) 2 , for x < y. (1.4.23)
2
Conclusion: choosing := max (1 , 2 ) and := ()
< 1, we have proved that
(x y)2
f (x) f (y) (A0 ( f )) (x, y) (x y) , for any x, y [0, a] .
2
(1.4.24)
This is a condition needed to solve numerically f (x) = 0.
II) Let n 1 < < n, n N {1}, i.e. = n; x, y [0, a], a > 0, and f
C n ([0, a] , X ).
We define the following X -valued right Caputo fractional derivatives (see [3]),
16 1 Explicit-Implicit Methods with Applications to Banach Space
(1)n x
Dx f (y) = (z y)n1 f (n) (z) dz, for y x, (1.4.25)
(n ) y
and
(1)n y
D y f (x) = (z x)n1 f (n) (z) dz, for x y. (1.4.26)
(n ) x
y
n1
f (k) (y) 1
f (x) f (y) = (x y)k + (z x)1 D y f (z) dz,
k=1
k! () x
(1.4.27)
when x y, and
x
n1
f (k) (x) 1
f (y) f (x) = (y x) +
k
(z y)1 Dx f (z) dz,
k=1
k! () y
(1.4.28)
when x y.
We define also the fractional linear operator
n1 f (k) (x) (xy)1
k=1 k! (y x) Dx f (y) (+1) , x > y,
k
n1 f (k) (y) 1
(A0 ( f )) (x, y) := (x y)k D y f (x) (yx) , y > x,
k=1 k! (+1)
0, x = y.
(1.4.29)
We would like to prove that
|x y|n
f (x) f (y) (A0 ( f )) (x, y) (x y) c , (1.4.30)
n
for any x, y [0, a], 0 < c < 1.
When x = y the last condition (1.4.30) is trivial.
We assume x
= y. We distinguish the cases:
(1) x > y : We observe that
1 x (x y)1
(z y) 1
f (z) dz +
(y x)
() Dx f (y)
( + 1) = Dx
y
(1.4.32)
1
x
1 (x y) =
() (z y) Dx f (z) dz Dx f (y)
y ( + 1)
x x
1
(z y) 1
Dx f (z) dz (z y) 1
Dx f (y) dz
=
() y y
x
1 1
(z y) D f (z) D f (y) dz (1.4.33)
() y x x
1 x
(z y)1 Dx f (z) Dx f (y) dz
() y
1 x
1 (x y)n
= (z y)n1 dz =
() y () n
1
(assume 1 < (), i.e. 1 := ()
< 1)
(x y)n
= 1 .
n
We have proved, when x > y, that
(x y)n
f (x) f (y) (A0 ( f )) (x, y) (x y) 1 . (1.4.36)
n
(2) y > x : We observe that
n1
f (k) (y) (y x)1
(x y) k1
D y f (x) (x y) = (1.4.37)
k! ( + 1)
k=1
1 y (y x)
(z x) 1
D f (z) dz D f (x) = (1.4.38)
() y y
( + 1)
x
1 y 1 y
(z x) 1
D y f (z) dz (z x) 1
D y f (x) dz
() () =
x x
y
1
(z x) 1
D y f (z) D y f (x) dz
(1.4.39)
() x
1 y
(z x)1 D y f (z) D y f (x) dz
() x
(y x)n
f (x) f (y) (A0 ( f )) (x, y) (x y) 2 . (1.4.42)
n
0 < := < 1. (1.4.43)
()
|x y|n
f (x) f (y) (A0 ( f )) (x, y) (x y) , for any x, y [0, a] .
n
(1.4.44)
In the special case of 1 < < 2, we obtain that
(x y)2
f (x) f (y) (A0 ( f )) (x, y) (x y) , (1.4.45)
2
for any x, y [0, a], 0 < < 1.
This is a condition needed to solve numerically f (x) = 0.
References
1. C.D. Aliprantis, K.C. Border, Infinite Dimensional Analysis (Springer, New York, 2006)
2. S. Amat, S. Busquier, S. Plaza, Chaotic dynamics of a third-order Newton-type method. J.
Math. Anal. Appl. 366(1), 164174 (2010)
3. G.A. Anastassiou, Strong right fractional calculus for banach space valued functions. Revis.
Proyecc. 36(1), 149186 (2017)
4. G.A. Anastassiou, A strong fractional calculus theory for banach space valued functions, in
Nonlinear Functional Analysis and Applications (Korea) (2017). accepted for publication
5. G.A. Anastassiou, I.K. Argyros, Iterative methods and their applications to Banach space
valued functions in abstract fractional calculus, in Progress in Fractional Differentiation and
Applications (2017). accepted
6. I.K. Argyros, A unifying local-semilocal convergence analysis and applications for two-point
Newton-like methods in Banach space. J. Math. Anal. Appl. 298, 374397 (2004)
7. I.K. Argyros, A. Magran, Iterative Methods and their Dynamics with Applications (CRC
Press, New York, 2017)
8. Bochner integral, Encyclopedia of Mathematics, http://www.encyclopediaofmath.org/index.
php?title=Bochner_integral&oldid=38659
9. M. Edelstein, On fixed and periodic points under contractive mappings. J. Lond. Math. Soc.
37, 7479 (1962)
10. J.A. Ezquerro, J.M. Gutierrez, M.A. Hernandez, N. Romero, M.J. Rubio, The Newton method:
from Newton to Kantorovich (Spanish). Gac. R. Soc. Mat. Esp. 13, 5376 (2010)
11. L.V. Kantorovich, G.P. Akilov, Functional Analysis in Normed Spaces (Pergamon Press, New
York, 1982)
12. A. Magran, A new tool to study real dynamics: the convergence plane. Appl. Math. Comput.
248, 215224 (2014)
13. J. Mikusinski, The Bochner integral (Academic Press, New York, 1978)
14. F.A. Potra, V. Ptak, Nondiscrete Induction and Iterative Processes (Pitman Publ, London, 1984)
15. P.D. Proinov, New general convergence theory for iterative processes and its applications to
Newton-Kantorovich type theorems. J. Complex. 26, 342 (2010)
16. G.E. Shilov, Elementary Functional Analysis (Dover Publications Inc, New York, 1996)
Chapter 2
Convergence of Iterative Methods
in Abstract Fractional Calculus
2.1 Introduction
Secant method: 1
xn+1 = xn xn1 , xn ; F F (xn ) , (2.1.3)
where E n = E (F) (xn ) and E : L (B1 , B2 ) the space of bounded linear oper-
ators from B1 into B2 . Other explicit methods can be found in [8, 12, 16, 17] and
the references there in.
Implicit Methods [7, 10, 12, 17]:
xn+1 = xn A1
n F (x n ) , (2.1.6)
Pz (x) = x, (2.1.7)
where
Pz (x) = x + F (z) + A (x, z) (x z) (2.1.8)
or
Pz (x) = z A (x, z)1 F (z) (2.1.9)
and
xn+1 = xn A (xn , xn1 )1 F (xn ) . (2.1.14)
2.1 Introduction 23
Set 1 = 0 .
(s3 ) There exists a continuous and nondecreasing function : [0, +)3
[0, +) such that for each x, y 1
(x y , x x0 , y x0 ) x y .
t (1 q (t)) = 0
Next, we present the semi-local convergence analysis for method (2.1.6) using
the conditions (S) and the preceding notation.
Theorem 2.1 Assume that the conditions (S) hold. Then, sequence {xn } generated by
method (2.1.6) starting at x0 is well defined in U (x0 , s), remains in U (x0 , s) for
each n = 0, 1, 2, ... and converges to a solution x U (x0 , s) of equation F (x) =
0. Moreover, suppose that there exists a continuous and nondecreasing function
1 : [0, +)4 [0, +) such that for each x, y, z 1
1 (x y , x x0 , y x0 , z x0 ) x y
and q1 = 1 (, s, s, s) < 1.
Then, x is the unique solution of equation F (x) = 0 in U (x0 , s) .
Proof By the definition of s and (s5 ), we have x1 U (x0 , s). The proof is based on
mathematical induction on k. Suppose that xk xk1 q0k1 and xk x0 s.
We get by (2.1.6), (s2 ) (s5 ) in turn that
1
xk+1 xk = A1
k F (x k ) = Ak (F (x k ) F (x k1 ) Ak1 (x k x k1 ))
A1
k
F (xk ) F (xk1 ) Ak1 (xk xk1 )
and
xk+1 x0 xk+1 xk + ... + x1 x0
1 q0k+1
q0k + ... + = < = s.
1 q0 1 q0
1 q0m k
xk+m xk q .
1 q0 0
It follows from the preceding inequation that sequence {xk } is complete in a Banach
space B1 and as such it converges to some x U (x0 , s) (since U (x0 , s) is a closed
ball). By letting k + in (2.2.1) we get F (x ) = 0. To show the uniqueness
part, let x U (x0 , s) be a solution of equation F (x) = 0. By using (2.1.6) and
the hypothesis on 1 , we obtain in turn that
x xk+1 = x xk + A1 F (xk ) A1 F x
k k
2.2 Semi-local Convergence for Implicit Methods 25
1
A F x F (xk ) Ak x xk
k
1 1 x xk , xk1 x0 , xk x0 , x x0 x xk
q1 x xk q1k+1 x x0 ,
Remark 2.2 (1) The equation in (s6 ) is used to determine the smallness of . It can
be replaced by a stronger condition as follows. Choose (0, 1). Denote by s0 the
smallest positive solution of equation q (t) = . Notice that if function q is strictly
increasing,
we can set s0 = q 1 (). Then, we can suppose instead of (s6 ) :
s6 (1 ) s0
which is a stronger condition than (s6 ).
However, we wanted to leave the equation in (s6 ) as uncluttered and as weak as
possible.
(2) Condition (s2 ) can become part of condition (s3 ) by considering
(s3 ) There exists a continuous and nondecreasing function : [0, +)3
[0, +) such that for each x, y 1
A (x, y)1 [F (x) F (y) A (x, y) (x, y)]
(x y , x x0 , y x0 ) x y .
Notice that
(u 1 , u 2 , u 3 ) (u 1 , u 2 , u 3 )
A (x, y) L 1
and
2 := 1 1 < 1.
Then, it follows from the Banach lemma on invertible operators [12], and
1
L A (x, y) L 1 1 = 2 < 1
26 2 Convergence of Iterative Methods in Abstract Fractional Calculus
1
that A (x, y)1 L (B2 , B1 ). Let = 1 2
. Then, under these replacements, condi-
tion (s2 ) is implied, therefore it can be dropped from the conditions (S).
(4) Clearly method (2.1.5) converges under the conditions (S), since (2.1.6)
implies (2.1.5).
(5) We wanted to leave condition (s4 ) as uncluttered as possible, since in practice
Eqs. (2.1.6) (or (2.1.5)) may be solvable in a way avoiding the already mentioned con-
ditions of the contraction mapping principle. However, in what follows we examine
the solvability of method (2.1.5) under a stronger version of the contraction mapping
principle using the conditions (V ) :
(v1 ) = (s1 ) .
(v2 ) There exist functions w1 : [0, +)4 [0, +), w2 : [0, +)4
[0, +) continuous and nondecreasing such that for each x, y, z
and
w1 (0, 0, 0, 0) = w2 (0, 0, 0, 0) = 0.
Set
w1 (2t, t, t, t) + w2 (2t, t, t, t) (t + x0 ) , z = x0
h (t, t, t, t) =
w1 (2t, t, t, 0) + w2 (2t, t, t, 0) x0 , z = x0 .
h (t, t, t, t) < 1
and
h (t, t, 0, t) t + F (x0 ) t
(v4 ) U (x0 , ) D.
Theorem 2.3 Suppose that the conditions (V ) are satisfied. Then, Eq. (2.1.5) is
uniquely solvable for each n = 0, 1, 2, .... Moreover, if A1 n L (B2 , B1 ), the
Eq. (2.1.6) is also uniquely solvable for each n = 0, 1, 2, ...
Proof The result is an application of the contraction mapping principle. Let x, y, z
U (x0 , ). By the definition of operator Pz , (v2 ) and (v3 ), we get in turn that
Pz (x) Pz (y) = (I + A (x, z) A (y, z)) (x y) (A (x, z) A (y, z)) z
[w1 (x y , x x0 , y x0 , z x0 ) +
2.2 Semi-local Convergence for Implicit Methods 27
h ( , , , ) x y
and
Pz (x) x0 Pz (x) Pz (x0 ) + Pz (x0 ) x0
h (x x0 , x x0 , 0, z x0 ) x x0 + F (x0 )
h ( , , 0, ) + F (x0 ) .
Remark 2.4 Sections 2.2 and 2.3 have an interest independent of Sect. 2.4. It is worth
noticing that the results especially of Theorem 2.1 can apply in Abstract Fractional
Calculus as illustrated in Sect. 2.4. By specializing function , we can apply the
results of say Theorem 2.1 in the examples suggested in Sect. 2.4. In particular for
p
c1 u 1
(2.4.1), we choose (u 1 , u 2 , u 3 ) = ( p+1) for u 1 0, u 2 0, u 3 0 and c1 , p are
given in Sect. 2.4. Similar choices for the other examples of Sect. 2.4. It is also worth
noticing that estimate (2.4.2) derived in Sect. 2.4 is of independent interest but not
needed in Theorem 2.1.
Set 1 = 0 .
(s3 ) There exist continuous and nondecreasing functions 0 : [0, +)3
[0, +), 2 : [0, +)3 [0, +) with 0 (0, 0, 0) = 2 (0, 0, 0) = 0 such that
for each x, y 1
0 (x y , x x0 , y x0 ) x y
28 2 Convergence of Iterative Methods in Abstract Fractional Calculus
and
A (x, y) A (y, y) 2 (x y , x x0 , y x0 ) .
Set = 0 + 2 .
(s4 ) There exist x0 0 and 0 such that A (x0 , x0 )1 L (B2 , B1 ) and
A (x0 , x0 )1 F (x0 ) .
(s5 ) = (s6 )
(s6 ) = (s7 ).
Next, we present
the following semi-local convergence analysis of method (2.1.12)
using the S conditions and the preceding notation.
Proposition 2.5 Suppose that the conditions (S ) are satisfied. Then, sequence {xn }
generated by method (2.1.12) starting at x0 is well defined in U (x0 , s), remains
in U (x0 , s) for each n = 0, 1, 2, ... and converges to a unique solution x U (x0 , s)
of equation F (x) = 0.
Proof We follow the proof of Theorem 2.1 but use instead the analogous estimate
The rest of the proof is identical to the one in Theorem 2.1 until the uniqueness part
for which we have the corresponding estimate
x xk+1 = x xk + A1 F (xk ) A1 F x
k k
1
A F x F (xk ) Ak x xk
k
1 0 x xk , xk1 x0 , xk x0
q x xk q k+1 x x0 .
2.3 Semi-local Convergence for Explicit Methods 29
Remark 2.6 Comments similar to the ones given in Sect. 2.2 can follows but for
method (2.1.13) and method (2.1.14) instead of method (2.1.5) and method (2.1.6),
respectively.
Here we deal with Banach space (X, ) valued functions f of real domain [a, b].
All integrals here are of Bochner-type, see [15]. The derivatives of f are defined
similarly to numerical ones, see [18], pp. 8386 and p. 93.
In this section we apply the earlier numerical methods to X -valued fractional
calculus for solving f (x) = 0.
Here we would like to establish for [a, b] R, a < b, f C p ([a, b] , X ), p N,
that
|x y| p+1
f (y) f (x) A (x, y) (y x) c1 , (2.4.1)
p+1
So let f C y+ ([a, b] , X ), we define the X -valued generalized fractional
derivative of f over [y, b] as
30 2 Convergence of Iterative Methods in Abstract Fractional Calculus
y
D y f = J1 f ( p) , (2.4.5)
that is
1 d x
D y f (x) = (x t) f ( p) (t) dt, (2.4.6)
(1 ) d x y
which exists for f C y+ ([a, b] , X ), for a y x b.
Here we consider f C p ([a, b] , X ) such that f C y+ ([a, b] , X ), for every
y [a, b], which means also that f C x+ ([a, b] , X ), for every x [a, b] (i.e.
exchange roles of x and y), we write that as f C+ ([a, b] , X ) .
That is y
1 d
Dx f (y) = (y t) f ( p) (t) dt (2.4.7)
(1 ) dy x
exists for f C x+ ([a, b] , X ), for a x y b.
We mention the following left generalized X -valued fractional Taylor formula
( f C y+ ([a, b] , X ), > 1), see [5].
It holds
p1 (k) x
f (y) 1
f (x) f (y) = (x y)k + (x t)1 D y f (t) dt,
k=1
k! () y
(2.4.8)
all x, y [a, b] with x y.
Similarly for f C x+ ([a, b] , X ) we have
p1 (k) y
f (x) 1
f (y) f (x) = (y x)k + (y t)1 Dx f (t) dt,
k=1
k! () x
(2.4.9)
all x, y [a, b] with y x.
So here we work with f C p ([a, b] , X ), such that f C+ ([a, b] , X ) .
We define the X -valued left linear fractional operator
1
p1 f (k) (y)
k=1 k! (x y)
k1
+ D y f (x) (xy)
(+1)
, x > y,
p1 f (k) (x) (yx)1
(A+ ( f )) (x, y) := (y x)k1 + Dx f (y) (+1) , y > x,
( k=1 k!
f p1) (x) , x = y.
(2.4.10)
Notice that (see [13], p. 3)
(A+ ( f )) (x, x) (A+ ( f )) (y, y) = f ( p1) (x) f ( p1) (y) (2.4.11)
f ( p) |x y| , x, y [a, b] ,
p1 (k)
f (y) (x y)
(x y)k D y f (x) =
k! ( + 1)
k=1
1 x 1 x
(x t) 1
D y f (t) dt (x t) 1
D y f (x) dt
() ()
y y
(by [9])
1 x
(x t)1 D y f (t) D y f (x) dt
() y
for all x, y, t [a, b] with x t y, with 1 (y) > 0 and sup 1 (y) =: 1 < ,
y[a,b]
also it is 0 < p + 1 < 1)
1 x
(x t)1 (x t) p+1 dt = (2.4.15)
() y
1 x
1 (x y) p+1
(x t) p dt = .
() y () ( p + 1)
32 2 Convergence of Iterative Methods in Abstract Fractional Calculus
1 (x y) p+1
f (y) f (x) A+ ( f ) (x, y) (y x) , for x > y.
() ( p + 1)
(2.4.16)
(2) x < y : We observe that
(y x)
p1 (k)
f (x)
(y x) Dx f (y)
k
=
k! ( + 1)
k=1
1 y (y x)
(y t) 1
Dx f (t) dt Dx f (y) =
() ( + 1)
x
1 y 1 y
(y t) 1
Dx f (y) dt
f (t) dt
(y t) 1
Dx
() () =
x x
y (2.4.18)
1
(y t)1
Dx f (t) Dx f (y) dt
() x
1 y
(y t)1 Dx f (t) Dx f (y) dt
() x
for all x, y, t [a, b] with y t x, with 2 (x) > 0 and sup 2 (x) =: 2 < )
x[a,b]
2 y
(y t)1 (y t) p+1 dt = (2.4.20)
() x
2 y
2 (y x) p+1
(y t) p dt = .
() x () ( p + 1)
2.4 Applications to Abstract Fractional Calculus 33
2 (y x) p+1
f (y) f (x) (A+ ( f )) (x, y) (y x) , (2.4.21)
() ( p + 1)
|x y| p+1
f (y) f (x) (A+ ( f )) (x, y) (y x) , x, y [a, b] .
() ( p + 1)
(2.4.22)
(II) Here see [3] and [5].
Let x, y [a, b] such that x y, > 0,
/ N, such that p = [v], = p
(0 < < 1).
Let f C p ([a, b] , X ) and define
1 y
Jy f (x) := (z x)1 f (z) dz, a x y, (2.4.23)
() x
Notice that
y
1
1 ( p)
Jy f (x) = (z x) f ( p) (z) dz, (2.4.26)
(1 ) x
exists for f C y ([a, b] , X ), and
(1) p1 d y
D y f (x) = (z x) f ( p) (z) dz. (2.4.27)
(1 ) d x x
I.e.
(1) p1 d y
D y f (x) = (z x) p f ( p) (z) dz. (2.4.28)
( p + 1) d x x
34 2 Convergence of Iterative Methods in Abstract Fractional Calculus
Here we consider f C p ([a, b] , X ) such that f C y ([a, b] , X ), for every
y [a, b], which means also that f C x ([a, b] , X ), for every x [a, b] (i.e.
exchange roles of x and y), we write that as f C ([a, b] , X ) .
That is
(1) p1 d x
Dx f (y) = (z y) p f ( p) (z) dz (2.4.29)
( p + 1) dy y
exists for f C x ([a, b] , X ), for a y x b.
We mention the following X -valued right generalized fractional Taylor formula
( f C y ([a, b] , X ), > 1), see [5].
It holds
p1 (k) y
f (y) 1
f (x) f (y) = (x y) +k
(z x)1 D y f (z) dz,
k=1
k! () x
(2.4.30)
all x, y [a, b] with x y.
Similarly for f C x ([a, b] , X ) we have
p1 (k) x
f (x) 1
f (y) f (x) = (y x) +k
(z y)1 Dx f (z) dz,
k=1
k! () y
(2.4.31)
all x, y [a, b] with x y.
So here we work with f C p ([a, b] , X ), such that f C ([a, b] , X ) .
We define the X -valued right linear fractional operator
1
p1 f (k) (x)
k=1 k! (y x)
k1
Dx f (y) (xy)
(+1)
, x > y,
p1 f (k) (y) 1
A ( f ) (x, y) := (x y) k1 (yx)
D y f (x) (+1) , y > x,
( k=1 k!
f p1) (x) , x = y.
(2.4.32)
Condition (2.4.2) is fulfilled, the same as in (2.4.11), now for A ( f ) (x, x) .
We would like to prove that
|x y| p+1
f (x) f (y) (A ( f )) (x, y) (x y) c , (2.4.33)
p+1
p1
f (k) (x) (x 1
y)
(y x)k1 Dx f (y) (y x) = (2.4.35)
k! ( + 1)
k=1
1 x (x y)1
(z y) 1
f (z) dz +
(y x)
() Dx Dx f (y)
( + 1) =
y
1 x (x y)
(z y)1 Dx f (z) dz Dx f (y) =
() ( + 1)
y
1
x x
(z y) 1
Dx f (z) dz (z y) 1
Dx f (y) dz
=
() y y
x
1 1
(z y) D f (z) D f (y) dz (2.4.36)
() y x x
1 x
(z y)1 Dx f (z) Dx f (y) dz
() y
1 x
1 (x y) p+1 (x y) p+1
(z y) p dz = = 1 ,
() y () p + 1 p+1
1
where 1 := () > 0.
We have proved, when x > y, that
(x y) p+1
f (x) f (y) (A ( f )) (x, y) (x y) 1 . (2.4.39)
p+1
p1
f (k) (y) (y 1
x)
(x y)k1 D y f (x) (x y) = (2.4.40)
k! ( + 1)
k=1
1 y (y x)
(z x) 1
D y f (z) dz D y f (x) =
() ( + 1)
x
1 y 1 y
(z x) 1
D y f (z) dz (z x) 1
D y f (x) dz
() () =
x x
y (2.4.41)
1
(z x) 1
D y f (z) D y f (x) dz
(2.4.42)
() x
1 y
(z x)1 D y f (z) D y f (x) dz
() x
2 y
2 (y x) p+1
(z x) p dz = .
() x () p + 1
(y x) p+1
f (x) f (y) (A ( f )) (x, y) (x y) 2 , (2.4.45)
p+1
2
where 2 := () > 0.
Set := max (1 , 2 ) and := () > 0.
Conclusion We have proved (2.4.1) that
|x y| p+1
f (x) f (y) (A ( f )) (x, y) (x y) , (2.4.46)
p+1
p x
f (k) (y) 1
f (x) f (y) = (x y)k + (x t) p1 f ( p) (t) f ( p) (y) dt,
k! ( p 1)! y
k=1
(2.4.47)
x, y [a, b] .
We define the X -valued function
p f (k) (y)
(x y)k1 , x = y,
(A0 ( f )) (x, y) := k=1
( p1)
k! (2.4.48)
f (x) , x = y.
1 x
(x t) ( p)
(t) f ( p)
(y) dt
=: () .
p1
f (2.4.51)
( p 1)! y
x
c
(x t) p1 (t y)21 dt = (2.4.53)
( p 1)! y
38 2 Convergence of Iterative Methods in Abstract Fractional Calculus
( p) (2) ( p 1)!
c (x y) p+1 = c (x y) p+1
( p 1)! ( p + 2) ( p 1)! ( p + 1)!
c (x y) p+1
= .
( p + 1)!
Hence
(x y) p+1
() c , x > y. (2.4.54)
( p + 1)!
c (2) ( p) c ( p 1)!
(y x) p+1 = (y x) p+1
( p 1)! ( p + 2) ( p 1)! ( p + 1)!
(y x) p+1
=c .
( p + 1)!
That is
(y x) p+1
() c , y > x. (2.4.57)
( p + 1)!
Therefore it holds
|x y| p+1
() c , all x, y [a, b] such that x = y. (2.4.58)
( p + 1)!
|x y| p+1
f (x) f (y) (A0 ( f )) (x, y) (x y) c , c > 0, (2.4.59)
( p + 1)!
for all x = y.
2.4 Applications to Abstract Fractional Calculus 39
References
1. C.D. Aliprantis, K.C. Border, Infinite Dimensional Analysis (Springer, New York, 2006)
2. S. Amat, S. Busquier, S. Plaza, Chaotic dynamics of a third-order Newton-type method. J.
Math. Anal. Applic. 366(1), 164174 (2010)
3. G.A. Anastassiou, Strong right fractional calculus for banach space valued functions. Rev.
Proyecc. 36(1), 149186 (2017)
4. G.A. Anastassiou, A strong Fractional Calculus Theory for Banach Space Valued Functions,
Nonlinear Functional Analysis and Applications (Korea) (accepted for publication, 2017)
5. G.A. Anastassiou, Strong Mixed and Generalized Fractional Calculus for Banach Space Valued
Functions, Mat. Vesnik (2017)
6. G.A. Anastassiou, I.K. Argyros, Semi-local Convergence of Iterative Methods and Banach
Space Valued Functions in Abstract Fractional Calculus (submitted, 2017)
7. I.K. Argyros, A unifying local-semilocal convergence analysis and applications for two-point
Newton-like methods in Banach space. J. Math. Anal. Appl. 298, 374397 (2004)
8. I.K. Argyros, A. Magran, Iterative Methods and their Dynamics with Applications (CRC
Press, New York, 2017)
9. Bochner integral, Encyclopedia of Mathematics, http://www.encyclopediaofmath.org/index.
php?title=Bochner_integral&oldid=38659
10. M. Edelstein, On fixed and periodic points under contractive mappings. J. Lond. Math. Soc.
37, 7479 (1962)
11. J.A. Ezquerro, J.M. Gutierrez, M.A. Hernandez, N. Romero, M.J. Rubio, The Newton method:
from Newton to Kantorovich (Spanish). Gac. R. Soc. Mat. Esp. 13, 5376 (2010)
12. L.V. Kantorovich, G.P. Akilov, Functional Analysis in Normed Spaces (Pergamon Press, New
York, 1982)
13. G.E. Ladas, V. Lakshmikantham, Differential Equations in Abstract Spaces (Academic Press,
New York, 1972)
14. A. Magran, A new tool to study real dynamics: the convergence plane. Appl. Math. Comput.
248, 215224 (2014)
15. J. Mikusinski, The Bochner Integral (Academic Press, New York, 1978)
16. F.A. Potra, V. Ptak, Nondiscrete Induction and Iterative Processes (Pitman Publ., London,
1984)
17. P.D. Proinov, New general convergence theory for iterative processes and its applications to
Newton-Kantorovich type theorems. J. Complex. 26, 342 (2010)
18. G.E. Shilov, Elementary Functional Analysis (Dover Publications Inc., New York, 1996)
Chapter 3
Equations for Banach Space Valued
Functions in Fractional Vector Calculi
The aim of this chapter is to solve equations on Banach space using iterative methods
under generalized conditions. The differentiability of the operator involved is not
assumed and its domain is not necessarily convex. Several applications are suggested
including Banach space valued functions of abstract fractional calculus, where all
integrals are of Bochner-type. It follows [5].
3.1 Introduction
Secant method: 1
xn+1 = xn xn1 , xn ; F F (xn ) , (3.1.3)
where E n = E (F) (xn ) and E : L (B1 , B2 ) the space of bounded linear oper-
ators from B1 into B2 . Other explicit methods can be found in [8, 12, 16, 17] and
the references there in.
Implicit Methods: [7, 10, 12, 17]:
xn+1 = xn A1
n F (x n ) , (3.1.6)
Pz (x) = x, (3.1.7)
where
Pz (x) = x + F (z) + A (x, z) (x z) (3.1.8)
or
Pz (x) = z A (x, z)1 F (z) (3.1.9)
and
xn+1 = xn A (xn , xn1 )1 F (xn ) . (3.1.14)
3.1 Introduction 43
Set 1 = 0 .
(s3 ) There exists a continuous and nondecreasing function : [0, +)3
[0, +) such that for each x, y 1
(x y , x x0 , y x0 ) x y .
t (1 q (t)) = 0
Next, we present the semi-local convergence analysis for method (3.1.6) using
the conditions (S) and the preceding notation.
Theorem 3.1 Assume that the conditions (S) hold. Then, sequence {xn } generated by
method (3.1.6) starting at x0 is well defined in U (x0 , s), remains in U (x0 , s) for
each n = 0, 1, 2, ... and converges to a solution x U (x0 , s) of equation F (x) =
0. Moreover, suppose that there exists a continuous and nondecreasing function
1 : [0, +)4 [0, +) such that for each x, y, z 1
1 (x y , x x0 , y x0 , z x0 ) x y
and q1 = 1 (, s, s, s) < 1.
Then, x is the unique solution of equation F (x) = 0 in U (x0 , s) .
Proof By the definition of s and (s5 ), we have x1 U (x0 , s). The proof is based on
mathematical induction on k. Suppose that xk xk1 q0k1 and xk x0 s.
We get by (3.1.6), (s2 ) (s5 ) in turn that
1
xk+1 xk = A1
k F (x k ) = Ak (F (x k ) F (x k1 ) Ak1 (x k x k1 ))
A1
k
F (xk ) F (xk1 ) Ak1 (xk xk1 )
and
xk+1 x0 xk+1 xk + ... + x1 x0
1 q0k+1
q0k + ... + = < = s.
1 q0 1 q0
1 q0m k
xk+m xk q .
1 q0 0
It follows from the preceding inequation that sequence {xk } is complete in a Banach
space B1 and as such it converges to some x U (x0 , s) (since U (x0 , s) is a closed
ball). By letting k + in (3.2.1) we get F (x ) = 0. To show the uniqueness
part, let x U (x0 , s) be a solution of equation F (x) = 0. By using (3.1.6) and
the hypothesis on 1 , we obtain in turn that
x xk+1 = x xk + A1 F (xk ) A1 F x
k k
3.2 Semi-local Convergence for Implicit Methods 45
1
A F x F (xk ) Ak x xk
k
1 1 x xk , xk1 x0 , xk x0 , x x0 x xk
q1 x xk q1k+1 x x0 ,
Remark 3.2 (1) The equation in (s6 ) is used to determine the smallness of . It can
be replaced by a stronger condition as follows. Choose (0, 1). Denote by s0 the
smallest positive solution of equation q (t) = . Notice that if function q is strictly
increasing,
we can set s0 = q 1 (). Then, we can suppose instead of (s6 ) :
s6 (1 ) s0
which is a stronger condition than (s6 ).
However, we wanted to leave the equation in (s6 ) as uncluttered and as weak as
possible.
(2) Condition (s2 ) can become part of condition (s3 ) by considering
(s3 ) There exists a continuous and nondecreasing function : [0, +)3
[0, +) such that for each x, y 1
A (x, y)1 [F (x) F (y) A (x, y) (x, y)]
(x y , x x0 , y x0 ) x y .
Notice that
(u 1 , u 2 , u 3 ) (u 1 , u 2 , u 3 )
A (x, y) L 1
and
2 := 1 1 < 1.
Then, it follows from the Banach lemma on invertible operators [12], and
1
L A (x, y) L 1 1 = 2 < 1
46 3 Equations for Banach Space Valued Functions in Fractional Vector Calculi
1
that A (x, y)1 L (B2 , B1 ). Let = 1 2
. Then, under these replacements, condi-
tion (s2 ) is implied, therefore it can be dropped from the conditions (S).
(4) Clearly method (3.1.5) converges under the conditions (S), since (3.1.6)
implies (3.1.5).
(5) We wanted to leave condition (s4 ) as uncluttered as possible, since in practice
Eqs. (3.1.6) or (3.1.5) may be solvable in a way avoiding the already mentioned con-
ditions of the contraction mapping principle. However, in what follows we examine
the solvability of method (3.1.5) under a stronger version of the contraction mapping
principle using the conditions (V ):
(v1 ) = (s1 ) .
(v2 ) There exist functions w1 : [0, +)4 [0, +), w2 : [0, +)4
[0, +) continuous and nondecreasing such that for each x, y, z
and
w1 (0, 0, 0, 0) = w2 (0, 0, 0, 0) = 0.
Set
w1 (2t, t, t, t) + w2 (2t, t, t, t) (t + x0 ) , z = x0
h (t, t, t, t) =
w1 (2t, t, t, 0) + w2 (2t, t, t, 0) x0 , z = x0 .
h (t, t, t, t) < 1
and
h (t, t, 0, t) t + F (x0 ) t
(v4 ) U (x0 , ) D.
Theorem 3.3 Suppose that the conditions (V ) are satisfied. Then, Eq. (3.1.5) is
uniquely solvable for each n = 0, 1, 2, .... Moreover, if A1 n L (B2 , B1 ), the
Eq. (3.1.6) is also uniquely solvable for each n = 0, 1, 2, ...
Proof The result is an application of the contraction mapping principle. Let x, y, z
U (x0 , ). By the definition of operator Pz , (v2 ) and (v3 ), we get in turn that
Pz (x) Pz (y) = (I + A (x, z) A (y, z)) (x y) (A (x, z) A (y, z)) z
[w1 (x y , x x0 , y x0 , z x0 ) +
3.2 Semi-local Convergence for Implicit Methods 47
h ( , , , ) x y
and
Pz (x) x0 Pz (x) Pz (x0 ) + Pz (x0 ) x0
h (x x0 , x x0 , 0, z x0 ) x x0 + F (x0 )
h ( , , 0, ) + F (x0 ) .
Remark 3.4 Sections 3.2 and 3.3 have an interest independent of Sect. 3.4. It is worth
noticing that the results especially of Theorem 3.1 can apply in Abstract Fractional
Calculus as illustrated in Sect. 3.4. By specializing function , we can apply the
results of say Theorem 3.1 in the examples suggested in Sect. 3.4. In particular for
p1
cu
(3.4.8), we choose (u 1 , u 2 , u 3 ) = 1p for u 1 0, u 2 0, u 3 0 and c, p are
given in Sect. 3.4. Similar choices for the other examples of Sect. 3.4.
Set 1 = 0 .
(s3 ) There exist continuous and nondecreasing functions 0 : [0, +)3
[0, +), 2 : [0, +)3 [0, +) with 0 (0, 0, 0) = 2 (0, 0, 0) = 0 such that
for each x, y 1
0 (x y , x x0 , y x0 ) x y
and
A (x, y) A (y, y) 2 (x y , x x0 , y x0 ) .
48 3 Equations for Banach Space Valued Functions in Fractional Vector Calculi
Set = 0 + 2 .
(s4 ) There exist x0 0 and 0 such that A (x0 , x0 )1 L (B2 , B1 ) and
A (x0 , x0 )1 F (x0 ) .
(s5 ) = (s6 )
(s6 ) = (s7 ).
Next, we present
the following semi-local convergence analysis of method (3.1.12)
using the S conditions and the preceding notation.
Proposition 3.5 Suppose that the conditions (S ) are satisfied. Then, sequence {xn }
generated by method (3.1.12) starting at x0 is well defined in U (x0 , s), remains
in U (x0 , s) for each n = 0, 1, 2, ... and converges to a unique solution x U (x0 , s)
of equation F (x) = 0.
Proof We follow the proof of Theorem 3.1 but use instead the analogous estimate
The rest of the proof is identical to the one in Theorem 3.1 until the uniqueness part
for which we have the corresponding estimate
x xk+1 = x xk + A1 F (xk ) A1 F x
k k
1
A F x F (xk ) Ak x xk
k
1 0 x xk , xk1 x0 , xk x0
q x xk q k+1 x x0 .
Remark 3.6 Comments similar to the ones given in Sect. 3.2 can follows but for
method (3.1.13) and method (3.1.14) instead of method (3.1.5) and method (3.1.6),
respectively.
3.4 Applications to X -valued Fractional and Vector Calculi 49
Here we deal with Banach space (X, ) valued functions f of real domain [a, b].
All integrals here are of Bochner-type, see [14]. The derivatives of f are defined
similarly to numerical ones, see [17], pp. 8386 and p. 93.
We want to solve numerically
f (x) = 0. (3.4.1)
when x y, and
1 y
Dx f (y) := (y t) p1 f ( p) (t) dt, (3.4.3)
( p ) x
p1 (k) x
f (y) 1
f (x) f (y) = (x y)k + (x t)1 Dy
f (t) dt, for x > y,
k! () y
k=1
(3.4.5)
and
p1 (k) y
f (x) 1
f (y) f (x) = (y x) +
k
(y t)1 Dx
f (t) dt, for x < y.
k! () x
k=1
(3.4.6)
Immediately, we observe that (by [12], p. 3)
(A1 ( f )) (x, x) (A1 ( f )) (y, y) = f ( p1) (x) f ( p1) (y) (3.4.7)
50 3 Equations for Banach Space Valued Functions in Fractional Vector Calculi
f ( p) |x y| , x, y [a, b] ,
|x y| p
f (x) f (y) (A1 ( f )) (x, y) (x y) c , (3.4.8)
p
(x y)
p1 (k)
f (y)
(x y) Dy f (x)
k
=
k! ( + 1)
k=1
1 x 1 x
(x t)1 Dy f (t) dt (x t)1 Dy f (x) dt
() ()
y y
(by [8])
1 x
(x t)1 Dy f (t) Dy f (x) dt (3.4.10)
() y
(assume that
D f (t) D f (x) 1 |t x| p , (3.4.11)
y y
1
for any t, x, y [a, b] : x t y, where 1 < (), i.e. 1 := ()
< 1)
1 x
(x t)1 (x t) p dt = (3.4.12)
() y
1 x
1 (x y) p (x y) p
(x t) p1 dt = = 1 . (3.4.13)
() y () p p
3.4 Applications to X -valued Fractional and Vector Calculi 51
(x y) p
f (x) f (y) (A1 ( f )) (x, y) (x y) 1 , (3.4.14)
p
(y x)
p1 (k)
f (x)
(y x) Dx f (y)
k
=
k! ( + 1)
k=1
1 y 1 y
(y t)1 Dx f (t) dt (y t)1 Dx f (y) dt
() () =
x x
(3.4.16)
1
y
(y t) 1
Dx f (t)
Dx f (y) dt
() x
1 y
(y t)1 Dx f (t) Dx f (y) dt (3.4.17)
() x
Assuming also
2
2 := <1 (3.4.20)
()
52 3 Equations for Banach Space Valued Functions in Fractional Vector Calculi
(y x) p
f (x) f (y) (A1 ( f )) (x, y) (x y) 2 , for x < y. (3.4.21)
p
Conclusion: Choosing := max (1 , 2 ) and := ()
< 1, we have proved that
|x y| p
f (x) f (y) (A1 ( f )) (x, y) (x y) , for any x, y [a, b] .
p
(3.4.22)
(II) Application from Banach space Mathematical Analysis
In [4], we proved the following general X -valued Taylors formula:
Theorem 3.7 Let p N and f C p ([A, B] , X ), where [A, B] R and (X, )
is a Banach space. Let g C 1 ([A, B]), strictly increasing, such that, g 1 C p
([g (A) g (B)]). Let any a, b [A, B]. Then
p1 (k)
f g 1 (g (a))
f (b) = f (a) + (g (b) g (a))k + R p (a, b) , (3.4.23)
k=1
k!
where
1 b ( p)
R p (a, b) = (g (b) g (s)) p1 f g 1 (g (s)) g (s) ds (3.4.24)
( p 1)! a
g(b)
1 ( p)
= (g (b) t) p1 f g 1 (t) dt.
( p 1)! g(a)
Theorem 3.7 will be applied next for g (x) = e x . One can give similar applications
for g = sin, cos, tan, etc., over suitable intervals.
p1
( f ln)(k) (ea ) b k
f (b) = f (a) + e ea + R p (a, b) , (3.4.25)
k=1
k!
where
1 eb b p1
R p (a, b) = e t ( f ln)( p) (t) dt
( p 1)! ea
1 b b p1
= e ea ( f ln)( p) es es ds, a, b [A, B] .
( p 1)! a
where
R p (, ) =
1 ( p) ( p)
(g () g (s)) p1 f g 1 (g (s)) f g 1 (g ()) g (s) ds
( p 1)!
(3.4.27)
g()
1 ( p) ( p)
= (g () t) p1 f g 1 (t) f g 1 (g ()) dt,
( p 1)! g()
, [A, B] .
Proof Easy.
Remark 3.10 Call l = f g 1 . Then l, l , ..., l ( p) are continuous from [g (A) , g (B)]
into f ([A, B]).
g()
1 ( p) ( p)
(g () t) p1 f g 1 (t) f g 1 (g ()) dt
( p 1)! g()
g()
K
(g () t) p1 (t g ())21 dt = (3.4.29)
( p 1)! g()
K ( p) (2)
(g () g ()) p+1 =
( p 1)! ( p + 2)
R (, ) K (g () g ())
p+1
, (3.4.31)
p
( p + 1)!
when g () > g () .
(ii) if g () > g (), then
R (, ) =
p
g()
1
1 ( p)
1 ( p)
g g g dt
(t ()) (t) (g ())
p1
f f
( p 1)! g()
g()
1 ( p) ( p)
(t g ()) p1 f g 1 (t) f g 1 (g ()) dt
( p 1)! g()
(3.4.32)
g()
K
(g () t)21 (t g ()) p1 dt =
( p 1)! g()
K (2) ( p)
(g () g ()) p+1 = (3.4.33)
( p 1)! ( p + 2)
R (, ) K (g () g ())
p+1
, (3.4.35)
p
( p + 1)!
whenever g () > g () .
Conclusion: It holds
R (, ) K |g () g ()|
p+1
, (3.4.36)
p
( p + 1)!
, [A, B].
Both sides of (3.4.36) equal zero when = .
We define the following X -valued linear operator:
p ( f g1 )(k) (g(y)) (g (x) g (y))k1 , when g (x) = g (y) ,
(A3 ( f )) (x, y) := k=1
( p1)
k!
f (x) , x = y,
(3.4.37)
for any x, y [A, B] .
3.4 Applications to X -valued Fractional and Vector Calculi 55
p (k)
f g 1 (g (y))
(g (x) g (y)) (g (x) g (y)) =
k1
k!
k=1
(3.4.36)
R (y, x) K |g (x) g (y)|
p+1
, (3.4.40)
p
( p + 1)!
f (x) f (y) (A31 ( f )) (x, y) e x e y K 1 |e e |
x y p+1
, (3.4.43)
( p + 1)!
We assume that
( p) ( p)
f tan1 (t) f tan1 (tan y) K 4 |t tan y| , (3.4.52)
t, tan y tan 2 + , tan 2 , where K 4 > 0.
It holds that
References
1. C.D. Aliprantis, K.C. Border, Infinite Dimensional Analysis (Springer, New York, 2006)
2. S. Amat, S. Busquier, S. Plaza, Chaotic dynamics of a third-order Newton-type method. J.
Math. Anal. Applic. 366(1), 164174 (2010)
3. G.A. Anastassiou, A strong fractional calculus theory for banach space valued functions, in
Nonlinear Functional Analysis and Applications (Korea) (2017). accepted for publication
4. G.A. Anastassiou, Principles of general fractional analysis for Banach space valued functions
(2017). submitted for publication
5. G.A. Anastassiou, I.K. Argyros, On the solution of equations and applications on Banach space
valued functions and fractional vector calculi (2017). submitted
6. I.K. Argyros, A unifying local-semilocal convergence analysis and applications for two-point
Newton-like methods in Banach space. J. Math. Anal. Appl. 298, 374397 (2004)
7. I.K. Argyros, A. Magran, Iterative Methods and their Dynamics with Applications (CRC
Press, New York, 2017)
8. Bochner integral, Encyclopedia of Mathematics, http://www.encyclopediaofmath.org/index.
php?title=Bochner_integral&oldid=38659
9. M. Edelstein, On fixed and periodic points under contractive mappings. J. Lond. Math. Soc.
37, 7479 (1962)
10. J.A. Ezquerro, J.M. Gutierrez, M.A. Hernandez, N. Romero, M.J. Rubio, The Newton method:
from Newton to Kantorovich (Spanish). Gac. R. Soc. Mat. Esp. 13, 5376 (2010)
11. L.V. Kantorovich, G.P. Akilov, Functional Analysis in Normed Spaces (Pergamon Press, New
York, 1982)
12. G.E. Ladas, V. Lakshmikantham, Differential Equations in Abstract Spaces (Academic Press,
New York, London, 1972)
13. A. Magran, A new tool to study real dynamics: the convergence plane. Appl. Math. Comput.
248, 215224 (2014)
14. J. Mikusinski, The Bochner Integral (Academic Press, New York, 1978)
15. F.A. Potra, V. Ptak, Nondiscrete Induction and Iterative Processes (Pitman Publ, London, 1984)
16. P.D. Proinov, New general convergence theory for iterative processes and its applications to
Newton-Kantorovich type theorems. J. Complex. 26, 342 (2010)
17. G.E. Shilov, Elementary Functional Analysis (Dover Publications Inc, New York, 1996)
Chapter 4
Iterative Methods in Abstract Fractional
Calculus
The goal of this chapter is to present a semi-local convergence analysis for some
iterative methods under generalized conditions. The operator is only assumed to
be continuous and its domain is open. Applications are suggested including Banach
space valued functions of fractional calculus, where all integrals are of Bochner-type.
It follows [5].
4.1 Introduction
Secant method: 1
xn+1 = xn xn1 , xn ; F F (xn ) , (4.1.3)
Springer International Publishing AG 2018 59
G. A. Anastassiou and I. K. Argyros, Functional Numerical Methods:
Applications to Abstract Fractional Calculus, Studies in Systems,
Decision and Control 130, https://doi.org/10.1007/978-3-319-69526-6_4
60 4 Iterative Methods in Abstract Fractional Calculus
where E n = E (F) (xn ) and E : L (B1 , B2 ) the space of bounded linear oper-
ators from B1 into B2 . Other explicit methods can be found in [8, 12, 16, 17] and
the references there in.
Implicit Methods [7, 10, 12, 17]:
xn+1 = xn A1
n F (x n ) , (4.1.6)
Pz (x) = x, (4.1.7)
where
Pz (x) = x + F (z) + A (x, z) (x z) (4.1.8)
or
Pz (x) = z A (x, z)1 F (z) (4.1.9)
and
xn+1 = xn A (xn , xn1 )1 F (xn ) . (4.1.14)
4.1 Introduction 61
Set 1 = 0 .
(s3 ) There exists a continuous and nondecreasing function : [0, +)3
[0, +) such that for each x, y 1
(x y , x x0 , y x0 ) x y .
t (1 q (t)) = 0
Next, we present the semi-local convergence analysis for method (4.1.6) using
the conditions (S) and the preceding notation.
Theorem 4.1 Assume that the conditions (S) hold. Then, sequence {xn } generated by
method (4.1.6) starting at x0 is well defined in U (x0 , s), remains in U (x0 , s) for
each n = 0, 1, 2, ... and converges to a solution x U (x0 , s) of equation F (x) =
0. Moreover, suppose that there exists a continuous and nondecreasing function
1 : [0, +)4 [0, +) such that for each x, y, z 1
1 (x y , x x0 , y x0 , z x0 ) x y
and q1 = 1 (, s, s, s) < 1.
Then, x is the unique solution of equation F (x) = 0 in U (x0 , s) .
Proof By the definition of s and (s5 ), we have x1 U (x0 , s). The proof is based on
mathematical induction on k. Suppose that xk xk1 q0k1 and xk x0 s.
We get by (4.1.6), (s2 ) (s5 ) in turn that
1
xk+1 xk = A1
k F (x k ) = Ak (F (x k ) F (x k1 ) Ak1 (x k x k1 ))
A1
k
F (xk ) F (xk1 ) Ak1 (xk xk1 )
and
xk+1 x0 xk+1 xk + ... + x1 x0
1 q0k+1
q0k + ... + = < = s.
1 q0 1 q0
1 q0m k
xk+m xk q .
1 q0 0
It follows from the preceding inequation that sequence {xk } is complete in a Banach
space B1 and as such it converges to some x U (x0 , s) (since U (x0 , s) is a closed
ball). By letting k + in (4.2.1) we get F (x ) = 0. To show the uniqueness
part, let x U (x0 , s) be a solution of equation F (x) = 0. By using (4.1.6) and
the hypothesis on 1 , we obtain in turn that
x xk+1 = x xk + A1 F (xk ) A1 F x
k k
4.2 Semi-local Convergence for Implicit Methods 63
1
A F x F (xk ) Ak x xk
k
1 1 x xk , xk1 x0 , xk x0 , x x0 x xk
q1 x xk q1k+1 x x0 ,
Remark 4.2 (1) The equation in (s6 ) is used to determine the smallness of . It can
be replaced by a stronger condition as follows. Choose (0, 1). Denote by s0 the
smallest positive solution of equation q (t) = . Notice that if function q is strictly
increasing,
we can set s0 = q 1 (). Then, we can suppose instead of (s6 ) :
s6 (1 ) s0
which is a stronger condition than (s6 ).
However, we wanted to leave the equation in (s6 ) as uncluttered and as weak as
possible.
(2) Condition (s2 ) can become part of condition (s3 ) by considering
(s3 ) There exists a continuous and nondecreasing function : [0, +)3
[0, +) such that for each x, y 1
A (x, y)1 [F (x) F (y) A (x, y) (x, y)]
(x y , x x0 , y x0 ) x y .
Notice that
(u 1 , u 2 , u 3 ) (u 1 , u 2 , u 3 )
A (x, y) L 1
and
2 := 1 1 < 1.
Then, it follows from the Banach lemma on invertible operators [12], and
1
L A (x, y) L 1 1 = 2 < 1
64 4 Iterative Methods in Abstract Fractional Calculus
1
that A (x, y)1 L (B2 , B1 ). Let = 1 2
. Then, under these replacements, condi-
tion (s2 ) is implied, therefore it can be dropped from the conditions (S).
(4) Clearly method (4.1.5) converges under the conditions (S), since (4.1.6)
implies (4.1.5).
(5) We wanted to leave condition (s4 ) as uncluttered as possible, since in practice
Eqs. (4.1.6) or (4.1.5) may be solvable in a way avoiding the already mentioned con-
ditions of the contraction mapping principle. However, in what follows we examine
the solvability of method (4.1.5) under a stronger version of the contraction mapping
principle using the conditions (V ):
(v1 ) = (s1 ).
(v2 ) There exist functions w1 : [0, +)4 [0, +), w2 : [0, +)4
[0, +) continuous and nondecreasing such that for each x, y, z
and
w1 (0, 0, 0, 0) = w2 (0, 0, 0, 0) = 0.
Set
w1 (2t, t, t, t) + w2 (2t, t, t, t) (t + x0 ) , z = x0
h (t, t, t, t) =
w1 (2t, t, t, 0) + w2 (2t, t, t, 0) x0 , z = x0 .
h (t, t, t, t) < 1
and
h (t, t, 0, t) t + F (x0 ) t
(v4 ) U (x0 , ) D.
Theorem 4.3 Suppose that the conditions (V ) are satisfied. Then, Eq. (4.1.5) is
uniquely solvable for each n = 0, 1, 2, .... Moreover, if A1 n L (B2 , B1 ), the
Eq. (4.1.6) is also uniquely solvable for each n = 0, 1, 2, ...
Proof The result is an application of the contraction mapping principle. Let x, y, z
U (x0 , ). By the definition of operator Pz , (v2 ) and (v3 ), we get in turn that
Pz (x) Pz (y) = (I + A (x, z) A (y, z)) (x y) (A (x, z) A (y, z)) z
[w1 (x y , x x0 , y x0 , z x0 ) +
4.2 Semi-local Convergence for Implicit Methods 65
h ( , , , ) x y
and
Pz (x) x0 Pz (x) Pz (x0 ) + Pz (x0 ) x0
h (x x0 , x x0 , 0, z x0 ) x x0 + F (x0 )
h ( , , 0, ) + F (x0 ) .
Remark 4.4 Sections 4.2 and 4.3 have an interest independent of Sect. 4.4. It is worth
noticing that the results especially of Theorem 4.1 can apply in Abstract Fractional
Calculus as illustrated in Sect. 4.4. By specializing function , we can apply the
results of say Theorem 4.1 in the examples suggested in Sect. 4.4. In particular for
u (n+1)
(4.4.21), we choose (u 1 , u 2 , u 3 ) = ((n+1))((n+1)+1)
1
for u 1 0, u 2 0, u 3 0
and , are given in Sect. 4.4. Similar choices for the other examples of Sect. 4.4.
It is also worth noticing that estimate (4.4.2) derived in Sect. 4.4 is of independent
interest but not needed in Theorem 4.1.
Set 1 = 0 .
(s3 ) There exist continuous and nondecreasing functions 0 : [0, +)3
[0, +), 2 : [0, +)3 [0, +) with 0 (0, 0, 0) = 2 (0, 0, 0) = 0 such that
for each x, y 1
0 (x y , x x0 , y x0 ) x y
66 4 Iterative Methods in Abstract Fractional Calculus
and
A (x, y) A (y, y) 2 (x y , x x0 , y x0 ) .
Set = 0 + 2 .
(s4 ) There exist x0 0 and 0 such that A (x0 , x0 )1 L (B2 , B1 ) and
A (x0 , x0 )1 F (x0 ) .
(s5 ) = (s6 )
(s6 ) = (s7 ).
Next, we present
the following semi-local convergence analysis of method (4.1.12)
using the S conditions and the preceding notation.
Proposition 4.5 Suppose that the conditions (S ) are satisfied. Then, sequence {xn }
generated by method (4.1.12) starting at x0 is well defined in U (x0 , s), remains
in U (x0 , s) for each n = 0, 1, 2, ... and converges to a unique solution x U (x0 , s)
of equation F (x) = 0.
Proof We follow the proof of Theorem 4.1 but use instead the analogous estimate
The rest of the proof is identical to the one in Theorem 4.1 until the uniqueness part
for which we have the corresponding estimate
x xk+1 = x xk + A1 F (xk ) A1 F x
k k
1
A F x F (xk ) Ak x xk
k
1 0 x xk , xk1 x0 , xk x0
q x xk q k+1 x x0 .
Remark 4.6 Comments similar to the ones given in Sect. 4.2 can follows but for
method (4.1.13) and method (4.1.14) instead of method (4.1.5) and method (4.1.6),
respectively.
4.4 Applications to X -valued Fractional Calculus 67
Here we deal with Banach space (X, ) valued functions f of real domain [a, b].
All integrals are of Bochner-type, see [14]. The derivatives of f are defined similarly
to numerical ones, see [17], pp. 8386 and p. 93.
Let f : [a, b] X such that f (m) L ([a, b] , X ), the X -valued left Caputo
fractional derivative of order / N, > 0, m = ( ceiling) is defined as
follows (see [3]):
1 x
Da f (x) = (x t)m1 f (m) (t) dt, (4.4.1)
(m ) a
By [4], we have
n
(x a)i i
f (x) = D f (a) + (4.4.5)
i=0
(i + 1) a
1 x
(x t)(n+1)1 Da(n+1) f (t) dt, x [a, b].
((n + 1) ) a
68 4 Iterative Methods in Abstract Fractional Calculus
Under our assumption and conclusion, see (4.4.4), Taylors formula (4.4.5) becomes
n
(x a)i i
f (x) f (a) = D f (a) +
i=2
(i + 1) a
1 x
(x t)(n+1)1 Da(n+1) f (t) dt, x [a, b], for 0 < < 1.
((n + 1) ) a
(4.4.6)
Here we are going to operate more generally. Again we assume 0 < 1, and
f : [a, b] X , such that f C ([a, b] , X ). We define the following X -valued left
Caputo fractional derivatives:
1 x
D y f (x) = (x t) f (t) dt, (4.4.7)
(1 ) y
n
(x y)i i
f (x) f (y) = D y f (y) +
i=2
(i + 1)
1 x
(x t)(n+1)1 D (n+1) f (t) dt, (4.4.9)
((n + 1) ) y
y
n
(y x)i i
f (y) f (x) = Dx f (x) +
i=2
(i + 1)
1 y
(y t)(n+1)1 Dx(n+1) f (t) dt, (4.4.10)
((n + 1) ) x
4.4 Applications to X -valued Fractional Calculus 69
(A ( f )) (x, y) =
(xy)i1 i (n+1) (n+1)1
n
i=2 (i+1) D y f (y) + D y f (x) (xy)
((n+1)+1)
, x > y,
n (yx)i1 i (n+1) (yx)(n+1)1
(i+1)
D f (x) + D f (y) ((n+1)+1)
, y > x, (4.4.11)
i=2 x x
f (x) , when x = y,
(x y)(n+1)
D (n+1) f (x)
y
((n + 1) + 1)
(by [8])
1 x
(x t)(n+1)1 D (n+1) f (t) D (n+1) f (x) dt
((n + 1) ) y
y y
1 x
1 (x y)(n+1)+1
(x t)(n+1) dt = . (4.4.15)
((n + 1) ) y ((n + 1) ) ((n + 1) + 1)
1 (x y)(n+1)+1
f (x) f (y) (A ( f )) (x, y) (x y) ,
((n + 1) ) ((n + 1) + 1)
(4.4.16)
for any x, y [a, b] : x > y, 0 < < 1.
(ii) case of x < y :
2 y
2 (y x)(n+1)+1
(y t)(n+1) dt = . (4.4.19)
((n + 1) ) x ((n + 1) ) ((n + 1) + 1)
4.4 Applications to X -valued Fractional Calculus 71
2 (y x)(n+1)+1
f (x) f (y) A ( f ) (x, y) (x y) ,
((n + 1) ) ((n + 1) + 1)
(4.4.20)
x, y [a, b] : y > x, 0 < < 1.
Conclusion Let := max (1 , 2 ) . It holds
|x y|(n+1)+1
f (x) f (y) (A ( f )) (x, y) (x y) ,
((n + 1) ) ((n + 1) + 1)
(4.4.21)
x, y [a, b], where 0 < < 1, n N.
One may assume that ((n+1)) < 1.
(Above notice that (4.4.21) is trivial when x = y.)
Now based on (4.4.12) and (4.4.21), we can apply our numerical methods pre-
sented in this chapter, to solve f (x) = 0.
To have (n + 1) + 1 2, we need to take 1 > n+1
1
, where n N.
References
1. C.D. Aliprantis, K.C. Border, Infinite Dimensional Analysis (Springer, New York, 2006)
2. S. Amat, S. Busquier, S. Plaza, Chaotic dynamics of a third-order Newton-type method. J.
Math. Anal. Applic. 366(1), 164174 (2010)
3. G.A. Anastassiou, A strong Fractional Calculus theory for Banach space valued functions,
nonlinear functional analysis and applications (Korea) (2017). Accepted for publication
4. G.A. Anastassiou, Principles of general fractional analysis for Banach space valued functions
(2017). submitted for publication
5. G.A. Anastassiou, I.K. Argyros, Iterative convergence with Banach space valued functions in
abstract fractional calculus (2017). submitted
6. I.K. Argyros, A unifying local-semilocal convergence analysis and applications for two-point
Newton-like methods in Banach space. J. Math. Anal. Appl. 298, 374397 (2004)
7. I.K. Argyros, A. Magran, Iterative methods and their dynamics with applications (CRC
Press, New York, 2017)
8. Bochner integral. Encyclopedia of Mathematics, http://www.encyclopediaofmath.org/index.
php?title=Bochner_integral&oldid=38659
9. M. Edelstein, On fixed and periodic points under contractive mappings. J. Lond. Math. Soc.
37, 7479 (1962)
10. J.A. Ezquerro, J.M. Gutierrez, M.A. Hernandez, N. Romero, M.J. Rubio, The Newton method:
from Newton to Kantorovich (Spanish). Gac. R. Soc. Mat. Esp. 13, 5376 (2010)
11. L.V. Kantorovich, G.P. Akilov, Functional Analysis in Normed Spaces (Pergamon Press, New
York, 1982)
12. G.E. Ladas, V. Lakshmikantham, Differential equations in abstract spaces (Academic Press,
New York, London, 1972)
13. S. Maruster, Local convergence of Ezquerro-Hernandez method. Ann. West. Univ. Timisoara
Math. Comput. Sci. 54(1), 19 (2016)
14. J. Mikusinski, The Bochner integral (Academic Press, New York, 1978)
72 4 Iterative Methods in Abstract Fractional Calculus
15. F.A. Potra, V. Ptak, Nondiscrete induction and iterative processes (Pitman Publication, London,
1984)
16. P.D. Proinov, New general convergence theory for iterative processes and its applications to
Newton-Kantorovich type theorems. J. Complexity 26, 342 (2010)
17. G.E. Shilov, Elementary Functional Analysis (Dover Publications Inc, New York, 1996)
Chapter 5
Semi-local Convergence in Right Abstract
Fractional Calculus
5.1 Introduction
Secant method: 1
xn+1 = xn xn1 , xn ; F F (xn ) , (5.1.3)
where E n = E (F) (xn ) and E : L (B1 , B2 ) the space of bounded linear oper-
ators from B1 into B2 . Other explicit methods can be found in [8, 12, 16, 17] and
the references there in.
Implicit Methods [7, 10, 12, 17]:
xn+1 = xn A1
n F (x n ) , (5.1.6)
Pz (x) = x, (5.1.7)
where
Pz (x) = x + F (z) + A (x, z) (x z) (5.1.8)
or
Pz (x) = z A (x, z)1 F (z) (5.1.9)
and
xn+1 = xn A (xn , xn1 )1 F (xn ) . (5.1.14)
5.1 Introduction 75
Set 1 = 0 .
(s3 ) There exists a continuous and nondecreasing function : [0, +)3
[0, +) such that for each x, y 1
(x y , x x0 , y x0 ) x y .
t (1 q (t)) = 0
Next, we present the semi-local convergence analysis for method (5.1.6) using
the conditions (S) and the preceding notation.
Theorem 5.1 Assume that the conditions (S) hold. Then, sequence {xn } generated by
method (5.1.6) starting at x0 is well defined in U (x0 , s), remains in U (x0 , s) for
each n = 0, 1, 2, ... and converges to a solution x U (x0 , s) of equation F (x) =
0. Moreover, suppose that there exists a continuous and nondecreasing function
1 : [0, +)4 [0, +) such that for each x, y, z 1
1 (x y , x x0 , y x0 , z x0 ) x y
and q1 = 1 (, s, s, s) < 1.
Then, x is the unique solution of equation F (x) = 0 in U (x0 , s) .
Proof By the definition of s and (s5 ), we have x1 U (x0 , s). The proof is based on
mathematical induction on k. Suppose that xk xk1 q0k1 and xk x0 s.
We get by (5.1.6), (s2 ) (s5 ) in turn that
1
xk+1 xk = A1
k F (x k ) = Ak (F (x k ) F (x k1 ) Ak1 (x k x k1 ))
A1
k
F (xk ) F (xk1 ) Ak1 (xk xk1 )
and
xk+1 x0 xk+1 xk + + x1 x0
1 q0k+1
q0k + + = < = s.
1 q0 1 q0
1 q0m k
xk+m xk q .
1 q0 0
It follows from the preceding inequation that sequence {xk } is complete in a Banach
space B1 and as such it converges to some x U (x0 , s) (since U (x0 , s) is a closed
ball). By letting k + in (5.2.1) we get F (x ) = 0. To show the uniqueness
part, let x U (x0 , s) be a solution of equation F (x) = 0. By using (5.1.6) and
the hypothesis on 1 , we obtain in turn that
x xk+1 = x xk + A1 F (xk ) A1 F x
k k
5.2 Semi-local Convergence for Implicit Methods 77
1
A F x F (xk ) Ak x xk
k
1 1 x xk , xk1 x0 , xk x0 , x x0 x xk
q1 x xk q1k+1 x x0 ,
Remark 5.2 (1) The equation in (s6 ) is used to determine the smallness of . It can
be replaced by a stronger condition as follows. Choose (0, 1). Denote by s0 the
smallest positive solution of equation q (t) = . Notice that if function q is strictly
increasing,
we can set s0 = q 1 (). Then, we can suppose instead of (s6 ) :
s6 (1 ) s0
which is a stronger condition than (s6 ).
However, we wanted to leave the equation in (s6 ) as uncluttered and as weak as
possible.
(2) Condition (s2 ) can become part of condition (s3 ) by considering
(s3 ) There exists a continuous and nondecreasing function : [0, +)3
[0, +) such that for each x, y 1
A (x, y)1 [F (x) F (y) A (x, y) (x, y)]
(x y , x x0 , y x0 ) x y .
Notice that
(u 1 , u 2 , u 3 ) (u 1 , u 2 , u 3 )
A (x, y) L 1
and
2 := 1 1 < 1.
Then, it follows from the Banach lemma on invertible operators [12], and
1
L A (x, y) L 1 1 = 2 < 1
78 5 Semi-local Convergence in Right Abstract Fractional Calculus
1
that A (x, y)1 L (B2 , B1 ). Let = 1 2
. Then, under these replacements, condi-
tion (s2 ) is implied, therefore it can be dropped from the conditions (S).
(4) Clearly method (5.1.5) converges under the conditions (S), since (5.1.6)
implies (5.1.5).
(5) We wanted to leave condition (s4 ) as uncluttered as possible, since in practice
equations (5.1.6) (or (5.1.5)) may be solvable in a way avoiding the already men-
tioned conditions of the contraction mapping principle. However, in what follows we
examine the solvability of method (5.1.5) under a stronger version of the contraction
mapping principle using the conditions (V ) :
(v1 ) = (s1 ) .
(v2 ) There exist functions w1 : [0, +)4 [0, +), w2 : [0, +)4
[0, +) continuous and nondecreasing such that for each x, y, z
and
w1 (0, 0, 0, 0) = w2 (0, 0, 0, 0) = 0.
Set
w1 (2t, t, t, t) + w2 (2t, t, t, t) (t + x0 ) , z = x0
h (t, t, t, t) =
w1 (2t, t, t, 0) + w2 (2t, t, t, 0) x0 , z = x0 .
h (t, t, t, t) < 1
and
h (t, t, 0, t) t + F (x0 ) t
(v4 ) U (x0 , ) D.
Theorem 5.3 Suppose that the conditions (V ) are satisfied. Then, equation (5.1.5)
is uniquely solvable for each n = 0, 1, 2, .... Moreover, if A1n L (B2 , B1 ), the
Eq. (5.1.6) is also uniquely solvable for each n = 0, 1, 2, ...
Proof The result is an application of the contraction mapping principle. Let x, y, z
U (x0 , ). By the definition of operator Pz , (v2 ) and (v3 ), we get in turn that
Pz (x) Pz (y) = (I + A (x, z) A (y, z)) (x y) (A (x, z) A (y, z)) z
[w1 (x y , x x0 , y x0 , z x0 ) +
5.2 Semi-local Convergence for Implicit Methods 79
h ( , , , ) x y
and
Pz (x) x0 Pz (x) Pz (x0 ) + Pz (x0 ) x0
h (x x0 , x x0 , 0, z x0 ) x x0 + F (x0 )
h ( , , 0, ) + F (x0 ) .
Remark 5.4 Sections 5.2 and 5.3 have an interest independent of Sect. 5.4. It is worth
noticing that the results especially of Theorem 5.1 can apply in Abstract Fractional
Calculus as illustrated in Sect. 5.4. By specializing function , we can apply the
results of say Theorem 5.1 in the examples suggested in Sect. 5.4. In particular for
u (n+1)
(5.4.28), we choose (u 1 , u 2 , u 3 ) = ((n+1))((n+1)+1)
1
for u 1 0, u 2 0, u 3 0
and , are given in Sect. 5.4. Similar choices for the other examples of Sect. 5.4.
It is also worth noticing that estimate (5.4.2) derived in Sect. 5.4 is of independent
interest but not needed in Theorem 5.1.
Set 1 = 0 .
(s3 ) There exist continuous and nondecreasing functions 0 : [0, +)3
[0, +), 2 : [0, +)3 [0, +) with 0 (0, 0, 0) = 2 (0, 0, 0) = 0 such that
for each x, y 1
0 (x y , x x0 , y x0 ) x y
and
A (x, y) A (y, y) 2 (x y , x x0 , y x0 ) .
Set = 0 + 2 .
80 5 Semi-local Convergence in Right Abstract Fractional Calculus
(s5 ) = (s6 )
(s6 ) = (s7 ).
Next, we present
the following semi-local convergence analysis of method (5.1.12)
using the S conditions and the preceding notation.
Proposition 5.5 Suppose that the conditions (S ) are satisfied. Then, sequence {xn }
generated by method (5.1.12) starting at x0 is well defined in U (x0 , s), remains
in U (x0 , s) for each n = 0, 1, 2, ... and converges to a unique solution x U (x0 , s)
of equation F (x) = 0.
Proof We follow the proof of Theorem 5.1 but use instead the analogous estimate
The rest of the proof is identical to the one in Theorem 5.1 until the uniqueness part
for which we have the corresponding estimate
x xk+1 = x xk + A1 F (xk ) A1 F x
k k
1
A F x F (xk ) Ak x xk
k
1 0 x xk , xk1 x0 , xk x0
q x xk q k+1 x x0 .
Remark 5.6 Comments similar to the ones given in Sect. 5.2 can follows but for
method (5.1.13) and method (5.1.14) instead of method (5.1.5) and method (5.1.6),
respectively.
5.4 Applications to X -valued Right Fractional Calculus 81
Here we deal with Banach space (X, ) valued functions f of real domain [a, b].
All integrals here are of Bochner-type, see [15]. The derivatives of f are defined
similarly to numerical ones, see [18], pp. 8386 and p. 93.
Let f : [a, b] X such that f (m) L ([a, b] , X ). The X -valued right Caputo
fractional derivative of order / N, > 0, m = ( ceiling), is defined as
follows (see [3]):
(1)m b
Db f (x) := (z x)m1 f (m) (z) dz, (5.4.1)
(m ) x
x [a, b], Ib
0
:= I (the identity operator).
We denote also
n
Ib := Ib Ib ...Ib (n - times). (5.4.6)
82 5 Semi-local Convergence in Right Abstract Fractional Calculus
n
(b x)i i
f (x) = D f (b) + (5.4.7)
i =0
(i + 1) b
b
1 (n+1)
(z x)(n+1)1 Db f (z) dz, x [a, b] .
((n + 1) ) x
We make
1 when f L ([a, b] , X ), 0 < < 1, we have that
Remark 5.8 In particular,
Db f (b) = 0, also Db f (x) = f (x), and
1 b
Db f (x) = (z x) f (z) dz, x [a, b] . (5.4.8)
(1 ) x
n
(b x)i i
f (x) f (b) = D f (b) + (5.4.9)
i=2
(i + 1) b
b
1 (n+1)
(z x)(n+1)1 Db f (z) dz, x [a, b] ; 0 < < 1.
((n + 1) ) x
Here we are going to operate more generally. Again we assume 0 < 1, and
f : [a, b] X , such that f C ([a, b] , X ). We define the following X -valued
right Caputo fractional derivatives:
1 y
D y f (x) := (t x) f (t) dt, (5.4.10)
(1 ) x
Clearly here D y f , Dx
f are continuous functions over [a, b], see [3]. We also
make the convention that D y f (x) = 0, for x > y, and Dx f (y) = 0, for y >
x.
Here we assume that
y f, D x f C ([a, b] , X ) ,
D k k 1
(5.4.12)
n
(y x)i i
f (x) f (y) = D y f (y) +
i=2
(i + 1)
y
1
(z x)(n+1)1 D (n+1)
y f (z) dz, (5.4.13)
((n + 1) ) x
n
(x y)i i
f (y) f (x) = Dx f (x) +
i=2
(i + 1)
x
1 (n+1)
(z y)(n+1)1 Dx f (z) dz, (5.4.14)
((n + 1) ) y
(A ( f )) (x, y) :=
n
(yx)i1 i (n+1) (yx)(n+1)1
D f (y) D f (x) , x < y,
i=2 (i+1) y y ((n+1)+1)
(xy)i1 i
n
(n+1) (n+1)1 (5.4.15)
(i+1)
Dx f (x) Dx f (y) (xy)
((n+1)+1)
, x > y,
i=2
f (x) , when x = y,
z, x, y [a, b] : y z x; 1 > 0)
1 y
(z x)(n+1)1 (z x) dz =
((n + 1) ) x
1 y
1 (y x)(n+1)+1
(z x)(n+1) dz = . (5.4.20)
((n + 1) ) x ((n + 1) ) ((n + 1) + 1)
1 (y x)(n+1)+1
f (x) f (y) (A ( f )) (x, y) (x y) ,
((n + 1) ) ((n + 1) + 1)
(5.4.21)
for any x, y [a, b] : x < y; 0 < < 1.
(ii) Case of x > y : We have
(n+1) (x y)(n+1)
Dx f (y) =
((n + 1) + 1)
x
1
(n+1)1 (n+1)
(n+1)
(y) dz
((n + 1) ) y
(z y) D x f (z) D x f
x (5.4.23)
1 (n+1)1 (n+1) (n+1)
(z y) Dx f (z) Dx f (y) dz
((n + 1) ) y
z, y, x [a, b] : x z y; 2 > 0)
2 x
(z y)(n+1)1 (z y) dz =
((n + 1) ) y
2 x
2 (x y)(n+1)+1
(z y)(n+1) dz = . (5.4.25)
((n + 1) ) y ((n + 1) ) ((n + 1) + 1)
2 (x y)(n+1)+1
f (x) f (y) (A ( f )) (x, y) (x y) ,
((n + 1) ) ((n + 1) + 1)
(5.4.26)
for any x, y [a, b] : x > y; 0 < < 1.
|x y|(n+1)+1
f (x) f (y) (A ( f )) (x, y) (x y) ,
((n + 1) ) ((n + 1) + 1)
(5.4.27)
x, y [a, b] ; where 0 < < 1, n N.
Now based on (5.4.16) and (5.4.27), we can apply our numerical methods pre-
sented in this chapter to solve f (x) = 0.
To have (n + 1) + 1 2, we need to take 1 > n+1
1
, where n N.
References
1. C.D. Aliprantis, K.C. Border, Infinite Dimensional Analysis (Springer, New York, 2006)
2. S. Amat, S. Busquier, S. Plaza, Chaotic dynamics of a third-order Newton-type method. J.
Math. Anal. Appl. 366(1), 164174 (2010)
3. G.A. Anastassiou, Strong right fractional calculus for banach space valued functions. Revis.
Proyecc. 36(1), 149186 (2017)
4. G.A. Anastassiou, Principles of general fractional analysis for Banach space valued functions
(submitted for publication, 2017)
5. G.A. Anastassiou, I.K. Argyros, Iterative convergence and applications on Banach space val-
ued functions in right abstract fractional calculus (submitted, 2017)
6. I.K. Argyros, F. Szidarovszky, The Theory and Applications of Iteration Methods (CRC Press,
Boca Raton, FL, USA, 1993)
7. I.K. Argyros, A unifying local-semilocal convergence analysis and applications for two-point
Newton-like methods in Banach space. J. Math. Anal. Appl. 298, 374397 (2004)
8. I.K. Argyros, A. Magran, Iterative Methods and their Dynamics with Applications (CRC
Press, New York, 2017)
9. Bochner integral, Encyclopedia of Mathematics, http://www.encyclopediaofmath.org/index.
php?title=Bochner_integral&oldid=38659
10. M. Edelstein, On fixed and periodic points under contractive mappings. J. Lond. Math. Soc.
37, 7479 (1962)
11. J.A. Ezquerro, J.M. Gutierrez, M.A. Hernandez, N. Romero, M.J. Rubio, The Newton method:
from Newton to Kantorovich (Spanish). Gac. R. Soc. Mat. Esp. 13, 5376 (2010)
12. L.V. Kantorovich, G.P. Akilov, Functional Analysis in Normed Spaces (Pergamon Press, New
York, 1982)
13. G.E. Ladas, V. Lakshmikantham, Differential Equations in Abstract Spaces (Academic Press,
New York, 1972)
14. A. Magran, A new tool to study real dynamics: the convergence plane. Appl. Math. Comput.
248, 215224 (2014)
15. J. Mikusinski, The Bochner Integral (Academic Press, New York, 1978)
16. F.A. Potra, V. Ptak, Nondiscrete Induction and Iterative Processes (Pitman Publication, London,
1984)
17. P.D. Proinov, New general convergence theory for iterative processes and its applications to
Newton-Kantorovich type theorems. J. Complex. 26, 342 (2010)
18. G.E. Shilov, Elementary Functional Analysis (Dover Publications Inc., New York, 1996)
Chapter 6
Algorithmic Convergence in Abstract
g-Fractional Calculus
The novelty of this chapter is the design of suitable algorithms for solving equations
on Banach spaces. Some applications of the semi-local convergence are suggested
including Banach space valued functions of fractional calculus, where all integrals
are of Bochner-type. It follows [6].
6.1 Introduction
Secant method: 1
xn+1 = xn xn1 , xn ; F F (xn ) , (6.1.3)
G (xn+1 ) = G (xn ) A1
n F (x n ) , (6.1.5)
Set 1 = 0 .
(m 3 ) There exists a continuous and nondecreasing function : [0, +)3
[0, +) such that for each x, y 1
(, s, s) < 1,
0 (s) < 1
and
G (x0 ) x0 s ,
1 q0
where q0 = (, s, s) .
(m 7 ) U (x0 , s) .
Next, we present the semi-local convergence analysis for method (6.1.5) using
the conditions (M) and the preceding notation.
Theorem 6.1 Assume that the conditions (M) hold. Then, sequence {xn } generated
by method (6.1.5) starting at x0 is well defined in U (x0 , s), remains in U (x0 , s)
for each n = 0, 1, 2, ... and converges to a solution x U (x0 , s) of equation
F (x) = 0. The limit point x is the unique solution of equation F (x) = 0 in
U (x0 , s) .
Proof By the definition of s and (m 5 ), we have x1 U (x0 , s). The proof is based on
mathematical induction on k. Suppose that xk xk1 q0k1 and xk x0 s.
We get by (6.1.5), (m 2 ) (m 5 ) in turn that
G (xk+1 ) G (xk ) = A1
k F (x k ) =
1
A (F (xk ) F (xk1 ) Ak1 (G (xk ) G (xk1 )))
k
A1
k
F (xk ) F (xk1 ) Ak1 (G (xk ) G (xk1 ))
and by (m 6 )
0 (s) s + G (x0 ) x0 s.
1 q0m k
xk+m xk q .
1 q0 0
1
A F x F (xk ) Ak G x G (xk )
k
1 0 x xk , xk+1 x0 , xk x0 G x G (xk )
q0 G x G (xk ) q0k+1 x x0 ,
Notice that
(u 1 , u 2 , u 3 ) (u 1 , u 2 , u 3 )
(2) Suppose that there exist > 0, 1 > 0 and L L (B1 , B2 ) with L 1
L (B2 , B1 ) such that 1
L 1
A (x, y) L 1
and
2 := 1 1 < 1.
Then, it follows from the Banach lemma on invertible operators [12], and
1
L A (x, y) L 1 1 = 2 < 1
1
that A (x, y)1 L (B2 , B1 ). Let = 1 2
. Then, under these replacements, condi-
tion (m 2 ) is implied, therefore it can be dropped from the conditions (M).
Remark 6.3 Section 6.2 has an interest independent of Sect. 6.3. It is worth noticing
that the results especially of Theorem 6.1 can apply in Abstract g-Fractional Calculus
as illustrated in Sect. 6.3. By specializing function , we can apply the results of say
Theorem 6.1 in the examples suggested in Sect. 6.3. In particular for (6.3.4), we
choose for u 1 0, u 2 0, u 3 0
(n+1)
(u 1 , u 2 , u 3 ) = 1
,
((n + 1) ) ((n + 1) + 1)
(n+1)
(u 1 , u 2 , u 3 ) = 2
,
((n + 1) ) ((n + 1) + 1)
(n+1)
(u 1 , u 2 , u 3 ) = 3
,
((n + 1) ) ((n + 1) + 1)
Here we deal with Banach space (X, ) valued functions f of real domain [a, b].
All integrals here are of Bochner-type, see [15]. The derivatives of f are defined
similarly to numerical ones, see [18], pp. 8386 and p. 93.
Let 0 < 1, m =
= 1 (
ceiling of number), g is strictly increasing and
g C 1 ([a, b]) , g 1 C ([g (a) , g (b)]) . Assume that f C 1 ([a, b] , X ). In both
backgrounds here we follow [5].
(I) The X -valued right generalized g-fractional derivative of f of order is defined
as follows:
b
1
Db;g f (x) := (g (t) g (x)) g (t) f g 1 (g (t)) dt,
(1 ) x
(6.3.1)
a x b.
If 0 < < 1, then Db;g f C ([a, b] , X ), see [4].
Also we define
1
Db;g f (x) := f g 1 g (x) , (6.3.2)
0
Db;g f (x) := f (x) , x [a, b] .
Also denote by
n
Ib;g := Ib;g Ib;g ...Ib;g (n times). (6.3.6)
We will be using the following X -valued modified g-right generalized Taylors for-
mula
n
(g (b) g (x))i
f (x) f (b) = i
Db;g f (b) + (6.3.7)
i=1
(i + 1)
b
1 (n+1)
(g (t) g (x))(n+1)1 g (t) Db;g f (t) dt,
((n + 1) ) x
x [a, b] .
Similarly we define:
1 x
Dx;g f (y) := (g (t) g (y)) g (t) f g 1 (g (t)) dt,
(1 ) y
(6.3.10)
all a y x; x [a, b] ,
1
Dx;g f (y) := f g 1 g (y) , y [a, b] . (6.3.11)
Denote by
y
Fk := D k
y;g f, Fk := D x;g f, x, y [a, b] .
x k
(6.3.13)
We assume that
y y
Fk , Fkx C 1 ([a, b] , X ) , and Fn+1 , Fn+1
x
C ([a, b] , X ) , (6.3.14)
In particular it holds
Dx;g f (x) = D y;g f (y) = 0, x, y [a, b] ; 0 < < 1. (6.3.18)
By (6.3.7) we derive
n
(g (y) g (x))i
f (x) f (y) = y;g f (y) +
D i (6.3.19)
i=2
(i + 1)
y
1
(g (t) g (x))(n+1)1 g (t) D (n+1)
y;g f (t) dt,
((n + 1) ) x
n
(g (x) g (y))i
f (y) f (x) = i
Dx;g f (x) + (6.3.20)
i=2
(i + 1)
x
1 (n+1)
(g (t) g (y))(n+1)1 g (t) Dx;g f (t) dt,
((n + 1) ) y
(A1 ( f )) (x, y) :=
n
(g(y)g(x))i1 (n+1) (g(y)g(x))(n+1)1
y;g f (y) D y;g f (x)
D i , x < y,
i=2 (i+1) ((n+1)+1)
n (n+1)1
(g(x)g(y))i1 (n+1)
(i+1)
i
Dx;g f (x) Dx;g f (y) (g(x)g(y))
((n+1)+1)
, x > y,
i=2
f (x) , when x = y,
(6.3.21)
x, y [a, b]; 0 < < 1.
We may assume that (see [13], p. 3)
(A1 ( f )) (x, x) (A1 ( f )) (y, y) = f (x) f (y)
= f g 1 (g (x)) f g 1 (g (y)) |g (x) g (y)| , (6.3.22)
1
=
((n + 1) )
y
(g (t) g (x)) (n+1)1
g (t) D (n+1) f (t) D (n+1) f (x) dt
y;g y;g
x
96 6 Algorithmic Convergence in Abstract g-Fractional Calculus
(by [9])
1
((n + 1) )
y
(g (t) g (x))(n+1)1 g (t) D (n+1)
y;g f (t) D (n+1)
y;g f (x) dt (6.3.24)
x
t, x, y [a, b] : y t x; 1 > 0)
1 y
(g (t) g (x))(n+1)1 g (t) (g (t) g (x)) dt = (6.3.26)
((n + 1) ) x
1 y
(g (t) g (x))(n+1) g (t) dt =
((n + 1) ) x
1 (g (y) g (x))(n+1)+1
. (6.3.27)
((n + 1) ) ((n + 1) + 1)
1 (g (y) g (x))(n+1)+1
, (6.3.28)
((n + 1) ) ((n + 1) + 1)
(n+1)
(n+1) (g (x) g (y))
Dx;g f (y) (6.3.30)
((n + 1) + 1)
1
=
((n + 1) )
6.3 Applications to X -valued Modified g-Fractional Calculus 97
x
(g (t) g (y))(n+1)1 g (t) (n+1) (n+1)
f (t) Dx;g f (y) dt
Dx;g
y
x
1 (n+1) (n+1)
(g (t) g (y))(n+1)1 g (t) Dx;g f (t) Dx;g f (y) dt
((n + 1) ) y
t, y, x [a, b] : x t y; 2 > 0)
2 x
(g (t) g (y))(n+1)1 g (t) (g (t) g (y)) dt = (6.3.32)
((n + 1) ) y
2 x
(g (t) g (y))(n+1) g (t) dt =
((n + 1) ) y
2 (g (x) g (y))(n+1)+1
.
((n + 1) ) ((n + 1) + 1)
2 (g (x) g (y))(n+1)+1
, (6.3.33)
((n + 1) ) ((n + 1) + 1)
|g (x) g (y)|(n+1)+1
, (6.3.34)
((n + 1) ) ((n + 1) + 1)
Now based on (6.3.22) and (6.3.34), we can apply our numerical methods presented
in this chapter to solve f (x) = 0.
To have (n + 1) + 1 2, we need to take 1 > n+1
1
, where n N.
Some examples of g follow:
g (x) = e x , x [a, b] R,
g (x) = sin x,
(6.3.36)
g (x) = tanx,
where x 2 + , 2 , > 0 small.
0
Da+;g f (x) = f (x) , x [a, b] .
Also denote by
n
Ia+;g := Ia+;g Ia+;g ...Ia+;g (n times). (6.3.41)
We will be using the following X -valued modified g-left generalized Taylors for-
mula.
Theorem 6.6 ([5]) Let 0 < 1, n N, f C 1 ([a, b] , X ), g C 1 ([a, b]),
strictly increasing, such that g 1 C 1 ([g (a) , g (b)]). Let Fk := Da+;g k
f, k =
1, ..., n, that fulfill Fk C ([a, b] , X ), and Fn+1 C ([a, b] , X ). Then
1
6.3 Applications to X -valued Modified g-Fractional Calculus 99
n
(g (x) g (a))i
f (x) f (a) = i
Da+;g f (a) + (6.3.42)
i=1
(i + 1)
x
1 (n+1)
(g (x) g (t))(n+1)1 g (t) Da+;g f (t) dt,
((n + 1) ) a
x [a, b] .
Similarly, we define
1 y
Dx+;g f (y) = (g (y) g (t)) g (t) f g 1 (g (t)) dt,
(1 ) x
(6.3.45)
for any x y b; x, y [a, b] ,
1
Dx+;g f (y) = f g 1 (g (y)) , y [a, b] . (6.3.46)
Denote by
y
G k := D k
y+;g f, G k := D x+;g f, x, y [a, b] .
x k
(6.3.48)
We assume that
y y
G k , G kx C 1 ([a, b] , X ) , and G n+1 , G n+1
x
C ([a, b] , X ) , (6.3.49)
In particular it holds
Da+;g f (a) = 0, 0 < < 1, (6.3.52)
and
D y+;g f (y) = Dx+;g f (x) = 0, x, y [a, b] ; 0 < < 1. (6.3.53)
By (6.3.42) we derive
n
(g (x) g (y))i
f (x) f (y) = y+;g f (y) +
D i
i=2
(i + 1)
x
1
(g (x) g (t))(n+1)1 g (t) D (n+1)
y+;g f (t) dt, (6.3.54)
((n + 1) ) y
n
(g (y) g (x))i
f (y) f (x) = i
Dx+;g f (x) + (6.3.55)
i=2
(i + 1)
y
1 (n+1)
(g (y) g (t))(n+1)1 g (t) Dx+;g f (t) dt,
((n + 1) ) x
(A2 ( f )) (x, y) :=
n
(g(x)g(y))i1 (n+1) (g(x)g(y))(n+1)1
y+;g f (y) + D y+;g f (x)
D i , x > y,
i=2 (i+1) ((n+1)+1)
n (n+1)1
(g(y)g(x))i1 (n+1)
(i+1)
i
Dx+;g f (x) + Dx+;g f (y) (g(y)g(x))
((n+1)+1)
, y > x,
i=2
f (x) , when x = y,
(6.3.56)
x, y [a, b]; 0 < < 1.
We may assume that (see [13], p. 3)
(A2 ( f )) (x, x) (A2 ( f )) (y, y) = f (x) f (y) (6.3.57)
with > 0.
We estimate and have
(i) case of x > y :
(g (x) g (y))(n+1)
D (n+1) f (x)
y+;g
((n + 1) + 1)
1
=
((n + 1) )
x
(g (x) g (t x))(n+1)1 g (t) D (n+1) f (t) D (n+1)
f (x) dt
y+;g y+;g
y
(6.3.59)
102 6 Algorithmic Convergence in Abstract g-Fractional Calculus
(by [9])
1
((n + 1) )
x
(g (x) g (t))(n+1)1 g (t) D (n+1)
y+;g f (t) D (n+1)
y+;g f (x) dt
y
t, x, y [a, b] : x t y; 1 > 0)
1 x
(g (x) g (t))(n+1)1 g (t) (g (x) g (t)) dt =
((n + 1) ) y
1 x
(g (x) g (t))(n+1) g (t) dt =
((n + 1) ) y
1 (g (x) g (y))(n+1)+1
. (6.3.61)
((n + 1) ) ((n + 1) + 1)
1 (g (x) g (y))(n+1)+1
, (6.3.62)
((n + 1) ) ((n + 1) + 1)
1
=
((n + 1) )
6.3 Applications to X -valued Modified g-Fractional Calculus 103
y
(g (y) g (t))(n+1)1 g (t) (n+1)
Dx+;g (n+1)
f (t) Dx+;g f (y) dt
x
1
((n + 1) )
y
(n+1) (n+1)
(g (y) g (t))(n+1)1 g (t) Dx+;g f (t) Dx+;g f (y) dt (6.3.64)
x
t, y, x [a, b] : y t x; 2 > 0)
2 y
(g (y) g (t))(n+1)1 g (t) (g (y) g (t)) dt =
((n + 1) ) x
2 (g (y) g (x))(n+1)+1
. (6.3.66)
((n + 1) ) ((n + 1) + 1)
2 (g (y) g (x))(n+1)+1
,
((n + 1) ) ((n + 1) + 1)
|g (x) g (y)|(n+1)+1
, (6.3.67)
((n + 1) ) ((n + 1) + 1)
Now based on (6.3.57) and (6.3.67), we can apply our numerical methods presented
in this chapter to solve f (x) = 0.
104 6 Algorithmic Convergence in Abstract g-Fractional Calculus
References
1. C.D. Aliprantis, K.C. Border, Infinite Dimensional Analysis (Springer, New York, 2006)
2. S. Amat, S. Busquier, S. Plaza, Chaotic dynamics of a third-order Newton-type method. J.
Math. Anal. Applic. 366(1), 164174 (2010)
3. G.A. Anastassiou, Strong right fractional calculus for banach space valued functions. Rev.
Proyecc. 36(1), 149186 (2017)
4. G.A. Anastassiou, A strong Fractional Calculus Theory for Banach space valued functions,
Nonlinear Functional Analysis and Applications (Korea) (2017). accepted for publication
5. G.A. Anastassiou, Principles of general fractional analysis for Banach space valued functions
(2017). submitted for publication
6. G.A. Anastassiou, I.K. Argyros, Algorithmic convergence on Banach space valued functions
in abstract g-fractional calculus. Prog. Fract. Differ. Appl. (2017). accepted
7. I.K. Argyros, A unifying local-semilocal convergence analysis and applications for two-point
Newton-like methods in Banach space. J. Math. Anal. Appl. 298, 374397 (2004)
8. I.K. Argyros, A. Magran, Iterative methods and their dynamics with applications (CRC
Press, New York, 2017)
9. Bochner integral, Encyclopedia of Mathematics, http://www.encyclopediaofmath.org/index.
php?title=Bochner_integral&oldid=38659
10. M. Edelstein, On fixed and periodic points under contractive mappings. J. London Math. Soc.
37, 7479 (1962)
11. J.A. Ezquerro, J.M. Gutierrez, M.A. Hernandez, N. Romero, M.J. Rubio, The Newton method:
From Newton to Kantorovich (Spanish). Gac. R. Soc. Mat. Esp. 13, 5376 (2010)
12. L.V. Kantorovich, G.P. Akilov, Functional Analysis in Normed Spaces (Pergamon Press, New
York, 1982)
13. G.E. Ladas, V. Lakshmikantham, Differential equations in abstract spaces (Academic Press,
New York, London, 1972)
14. A. Magran, A new tool to study real dynamics: the convergence plane. Appl. Math. Comput.
248, 215224 (2014)
15. J. Mikusinski, The Bochner Integral (Academic Press, New York, 1978)
16. F.A. Potra, V. Ptak, Nondiscrete Induction and Iterative Processes (Pitman Publishing, London,
1984)
17. P.D. Proinov, New general convergence theory for iterative processes and its applications to
Newton-Kantorovich type theorems. J. Complex. 26, 342 (2010)
18. G.E. Shilov, Elementary Functional Analysis (Dover Publications Inc, New York, 1996)
Chapter 7
Iterative Procedures for Solving Equations
in Abstract Fractional Calculus
7.1 Introduction
Secant method: 1
xn+1 = xn xn1 , xn ; F F (xn ) , (7.1.3)
Springer International Publishing AG 2018 105
G. A. Anastassiou and I. K. Argyros, Functional Numerical Methods:
Applications to Abstract Fractional Calculus, Studies in Systems,
Decision and Control 130, https://doi.org/10.1007/978-3-319-69526-6_7
106 7 Iterative Procedures for Solving Equations in Abstract Fractional Calculus
xn+1 = xn A1
n F (x n ) , (7.1.6)
Pz (x) = x, (7.1.7)
where
Pz (x) = x + F (z) + A (x, z) (x z) (7.1.8)
or
Pz (x) = z A (x, z)1 F (z) (7.1.9)
and
xn+1 = xn A (xn , xn1 )1 F (xn ) . (7.1.14)
7.1 Introduction 107
In this chapter in Sect. 7.2, we study the semi-local convergence of method (7.1.5)
and method (7.1.6). Section 7.3 contains the semi-local convergence of method
(7.1.11), method (7.1.12), method (7.1.13) and method (7.1.14). Some applications
to Abstract Fractional Calculus are suggested in Sect. 7.4 on a certain Banach space
valued functions, where all the integrals are of Bochner-type [8, 14].
The semi-local convergence analysis of method (7.1.6) that follows is based on the
conditions (H ):
(h 1 ) F : B1 B2 is continuous and A (F) (x, y) L (B1 , B2 ) for each
(x, y) .
(h 2 ) There exist l > 0 and 0 B1 such that A (F) (x, y)1 L (B2 , B1 ) for
each (x, y) 0 0 and
A (F) (x, y)1 l 1 .
Set 1 = 0 .
(h 3 ) There exist real numbers 1 , 2 , 3 satisfying
0 2 1 and 0 3 < 1
(h 6 ) h := 1 1
2 (1 3 )2 .
108 7 Iterative Procedures for Solving Equations in Abstract Fractional Calculus
and
(h 7 ) U (x0 , t ) 0 , where
13 (13 )2 2h
t = 1
, 1
= 0
1
13
, 1 = 0.
Then, we can show the following semi-local convergence result for method (7.1.6)
under the preceding notation and conditions (H ).
Theorem 7.1 Suppose that the conditions (H ) are satisfied. Then, sequence {xn }
generated by method (7.1.6) starting at x0 is well defined in U (x0 , t ), remains
in U (x0 , t ) for each n = 0, 1, 2, ... and converges to a solution x U (x0 , t )
of equation F (x) = 0. Moreover, provided that (h3 ) holds with A(F)(z,y) replacing
A(F)(x,y) for each z 1 , if 1
= 0, the equation F (x) = 0 has a unique solution
, where
x in U
U (x0 , t ) 0 , if h = 21 (1 3 )2
U=
U (x0 , t ) 0 , if h < 21 (1 3 )2
and, if 1 = 0, the solution x is unique in U x0 , 1 , where t =
3
13 + (13 )2 2h
1
.
1 2
Proof Case 1
= 0. Let us define scalar function g on R by g (t) = 2
t
(1 3 ) t + and majorizing sequence {tn } by
It follows from (h 6 ) that function g has two positive roots t and t , t t , and
tk tk+1 . That is, sequence {tk } converges to t .
(a) Using mathematical induction on k, we shall show that
xk+1 xk = A1
k (F (x k ) F (x k1 ) Ak1 (x k x k1 )) . (7.2.4)
and
xk x0 tk t . (7.2.8)
It follows by (7.2.7) and (7.2.8) that {xk } is a complete sequence in a Banach space
B1 and as such it converges to some x U (x0 , t ) (since U (x0 , t ) is a closed
set). By letting k +, using (h 1 ) and (h 2 ), we get l 1 lim F (xk ) = 0, so
k+
F (x ) = 0.
be such that F (x ) = 0. We shall show by induction that
Let x U
x xk t tk for each k = 0, 1, 2, ... . (7.2.9)
. Suppose that
Estimate (7.2.9) holds for k = 0 by the definition of x and U
x xk t tk . Then, as in (7.2.5), we obtain in turn that
x xk+1 = x xk + A1 F (xk ) A1 F x =
k k
1
A Ak x xk + F (xk ) F x
k
1
A F x F (xk ) Ak x xk
k
1
x xk + 2 xk x0 + 3 x xk
2
110 7 Iterative Procedures for Solving Equations in Abstract Fractional Calculus
t tk + 2 tk + 3 t tk =
1
2
1 2 1
t + (tk )2 1 tk t + 2 t tk tk + 3 t tk =
2 2
1 2
+ (1 3 ) t + t 1 tk t + 2 tk t 2 tk2 + 3 t 3 tk
2 k
= t tk+1, (7.2.10)
which completes the induction for estimate (7.2.9). Hence, lim xk = x . But we
k+
showed that lim xk = x , so x = x .
k+
Case 1 = 0. Then, we have by (h 3 ) that 2 = 0 and estimate (7.2.5) gives
and
xk+1 x0 xk+1 xk + xk xk1 + ... + x1 x0
1 3k+1
< . (7.2.12)
1 3 1 3
Then, as in the previous case it follows from (7.2.11) and (7.2.12) that
1 i3 k
xk+i xk , (7.2.13)
1 3 3
so sequence {xk } is complete and x solves equation F (x) = 0. Finally, the unique-
ness part follows from (7.2.10) for 1 = 2 = 0, since
x xk+1 3 x xk k+1 x x0 k+1 , (7.2.14)
3 3
1 3
and
5 := l01 4 < 1.
Then, it follows from the Banach lemma on invertible operators [6, 10, 11, 15, 16]
and 1
L A (F) (x, y) L l 1 4 = 5 < 1
0
l 1
that A (F) (x, y)1 L (B2 , B1 ). Set l 1 = 1
0
5
. Then, under these replacements,
condition (h 2 ) is implied, so it can be dropped from the conditions (H ).
(3) Clearly method (7.1.5) converges under the conditions (H ), since (7.1.6)
implies (7.1.5).
(4) Let R > 0 and define R0 = sup {t [0, R) : U (x0 , R0 ) D}. Set 0 =
U (x0 , R0 ). Condition (h 3 ) can be extended, if the additional term a2 x x0 is
inserted inside the parenthesis at the right hand side for some a2 0 . Then, the
conclusions of Theorem 7.1 hold in this more general setting, provided that a3 =
a2 R0 + 3 replaces 3 in conditions (h 6 ) and (h 7 ).
(5) Concerning the solvability of Eqs. (7.1.6) (or (7.1.5)), we wanted to leave
condition (h 4 ) as uncluttered as possible in conditions (H ). We did this because in
practice these equations may be solvable in a way other than using the contraction
mapping principle already mentioned earlier.
Next, we show the solvability of method (7.1.5) using a stronger version of the
contraction mapping principle and based on the conditions (C) :
(c1 ) = (h 1 ) .
(c2 ) There exist 0 [0, 1), 1 [0, +), 2 [0, 1), x0 such that for each
x, y, z
I + A (x, z) A (y, z) 0 ,
(c3 )
0 + 1 x0 + 2 1 for 2
= 0,
112 7 Iterative Procedures for Solving Equations in Abstract Fractional Calculus
(1 (0 + 1 x0 ))2
F (x0 ) for 1
= 0,
1
0 < 1 for 1 = 0
and
(c4 ) U (x0 , r ) , where
F (x0 ) 1 (0 + 1 x0 )
r < for 1
= 0,
1 (0 + 1 x0 ) 1
F (x0 )
r for 1 = 0,
1 0
1 (0 + 1 x0 )
r< for z = x0 , 1
= 0.
1
Theorem 7.3 Suppose that the conditions (C) are satisfied. Then, for each n =
0, 1, 2, ... Eq. (7.1.5) is unique solvable. Moreover, if A1 n L (B2 , B1 ), then
Eq. (7.1.6) is also uniquely solvable for each n = 0, 1, 2, ...
Proof We base the proof on the contraction mapping principle. Let x, y U (x0 , r ).
Then, using (7.1.8) we have in turn by (c2 ) that
Pz (x) Pz (y) = (I + A (x, z) A (y, z)) (x y) (A (x, z) A (y, z)) z
0 x y + 1 (z x0 + x0 ) x y
(x x0 ) x y , (7.2.15)
where
0 + 1 (t + x0 ) for z
= x0
(t) = (7.2.16)
0 + 1 x0 for z = x0 .
(0 + 1 x0 + 2 ) r r. (7.2.19)
Then, again by (7.2.15), (7.2.19) and the contraction mapping principle, we guarantee
the unique solvability of Eq. (7.1.5) and the existence of a unique sequence {xn } for
each n = 0, 1, 2, . . . . Finally, Eq. (7.1.6) is also uniquely solvable by the preceding
proof and the condition A1 n L (B2 , B1 ).
Remark 7.4 (a) The gamma conditions can be weakened, if i are replaced by func-
tions i (t), i = 0, 1, 2, 3. Then, i will appear as i (x x0 ) and i (r ) in the
conditions (C) .
(b) Sections 7.2 and 7.3 have an interest independent of Sect. 7.4. However, the
results especially of Theorem 7.1 can apply in Abstract Fractional Calculus as sug-
gested in Sect. 7.4.
Theorem 7.1 is general enough so it can be used to study the semi-local convergence of
method (7.1.11), method (7.1.12), method (7.1.13) and method (7.1.14). In particular,
for the study of method (7.1.12) (and consequently method (7.1.11)), we use the
conditions
H :
h 1 F : B1 B2 is continuous and A (F) (x, y) L (B1 , B2 ) for each
x .
(h 2 ) There exist l > 0 and 0 B1 such that A (F) (x, x)1 L (B2 , B1 ) and
A (F) (x, x)1 l 1 .
Set 1 = 0 .
(h 3 ) There exist real numbers 1 , 2 , 3 satisfying
0 2 1 and 0 3
114 7 Iterative Procedures for Solving Equations in Abstract Fractional Calculus
or
A (x, y) A (y, y) l5 x y .
(h 6 ) = (h 6 )
and
(h 7 ) = (h 7 ).
Then, we can show the following semi-local convergence of method (7.1.12) using
the conditions H and the preceding notation.
Proposition 7.5 Suppose that the conditions (H ) are satisfied. Then, sequence {xn }
generated by method (7.1.12) starting at x0 is well defined in U (x0 , t ), remains
in U (x0 , t ) for each n = 0, 1, 2, ... and converges to a solution x U (x0 , t )
of equation F (x) = 0. Moreover, if 1
= 0, the equation F (x) = 0 has a unique
solution x in U
, where
= U (x0 , t ) 0 , if h = 21 (1 3 )2
U
U (x0 , t ) 0 , if h < 21 (1 3 )2
and, if 1 = 0, the solution x is unique in U x0 , 1 3
, where t and t are given
in Theorem 7.1.
Proof Use in the proof of Theorem 7.1 instead of estimate (7.2.5) the analogous
estimate
0 2 1 and 0 3
(h 6 ) = (h 6 )
and
(h 7 ) = (h 7 ).
Then, we can present
the following semi-local convergence of method (7.1.14)
using the conditions H and the preceding notation.
Proposition 7.7 Suppose that the conditions (H ) are satisfied. Then, sequence {xn }
generated by method (7.1.14) starting at x0 is well defined in U (x0 , t ), remains
116 7 Iterative Procedures for Solving Equations in Abstract Fractional Calculus
Here we deal with Banach space (X, ) valued functions f of real domain [c, d].
All integrals here are of Bochner-type, see [14]. The derivatives of f are defined
similarly to numerical ones, see [17], pp. 8386 and p. 93.
We want to solve numerically f (x) = 0.
Let 0 < < 1, hence = 1, where is the ceiling of the number. Let also
c < a < b < d, and f C 2 ([c, d] , X ), with f
= 0. Clearly we have (see [12], p.
3)
f (x) f (y) f |x y| , x, y [c, d] . (7.4.1)
(I) The X -valued left Caputo fractional derivative (see [4]) of f of order (0, 1),
anchored at a, is defined as follows:
x
1
Da f (x) = (x t) f (t) dt, x [a, d] , (7.4.2)
(1 ) a
while Da f (x) = 0, for c x a, where is the gamma function.
Next we consider a < a < b, and x [a , b], also x0 (c, a).
We define the function
(2 )
A1 (x) := Da f (x) , x a , b . (7.4.3)
(x a) 1
(2 ) 1 x
(2 ) (x a)1
(x t) f (t) dt f (x) =
(x a)1 (1 ) (x a) 1 (2 )
a
x x
(2 )
1 1
(x t) f (t) dt (x t)
f (x)
dt
(x a)1 (1 ) a (1 ) a
118 7 Iterative Procedures for Solving Equations in Abstract Fractional Calculus
(by [8])
(1 ) x (7.4.1)
(x t) f (t) f (x) dt
(x a)1 a
(1 ) f x
(1 ) f x
(x t) (x t) dt = (x t)1 dt =
(x a)1 a (x a)1 a
(1 ) f (x a)2 (1 )
= f (x a) . (7.4.6)
(2 )
(x a) 1 2
b
(2 ) (2 ) (b x)1
1
(t x) f (t) dt f (x)=
(b x)1 (1 ) x (b x)1 (2 )
(2 )
1 b 1 b
(t x)
f (t) dt (t x)
f (x) dt
=
(b x)1 (1 ) x (1 ) x
(7.4.12)
(2 ) 1 b
(t x)
f (t) f (x) dt
(b x) 1 (1 )
x
(1 ) b (7.4.1)
(t x) f (t) f (x) dt
(b x)1 x
(1 ) f
b
(1 ) f b
(t x) (t x) dt = (t x)1 dt =
(b x) 1
x (b x) 1
x
(7.4.13)
(1 ) f (b x)2 (1 ) f
= (b x) .
(b x)1 2 (2 )
References
1. C.D. Aliprantis, K.C. Border, Infinite Dimensional Analysis (Springer, New York, 2006)
2. S. Amat, S. Busquier, S. Plaza, Chaotic dynamics of a third-order Newton-type method. J.
Math. Anal. Appl. 366(1), 164174 (2010)
3. G.A. Anastassiou, Strong right fractional calculus for banach space valued functions. Rev.
Proyecc. 36(1), 149186 (2017)
120 7 Iterative Procedures for Solving Equations in Abstract Fractional Calculus
4. G.A. Anastassiou, A strong fractional calculus theory for Banach space valued functions.
Nonlinear Funct. Anal. Appl. (Korea), (accepted for publication, 2017)
5. G.A. Anastassiou, I.K. Argyros, Generalized Iterative procedures and their applications to
Banach space valued functions in abstract fractional calculus (submitted, 2017)
6. I.K. Argyros, A unifying local-semilocal convergence analysis and applications for two-point
Newton-like methods in Banach space. J. Math. Anal. Appl. 298, 374397 (2004)
7. I.K. Argyros, A. Magran, Iterative Methods and their Dynamics with Applications (CRC
Press, New York, 2017)
8. Bochner integral, Encyclopedia of Mathematics, http://www.encyclopediaofmath.org/index.
php?title=Bochner_integral&oldid=38659
9. M. Edelstein, On fixed and periodic points under contractive mappings. J. London Math. Soc.
37, 7479 (1962)
10. J.A. Ezquerro, J.M. Gutierrez, M.A. Hernandez, N. Romero, M.J. Rubio, The Newton method:
from Newton to Kantorovich (Spanish). Gac. R. Soc. Mat. Esp. 13, 5376 (2010)
11. L.V. Kantorovich, G.P. Akilov, Functional Analysis in Normed Spaces (Pergamon Press, New
York, 1982)
12. G.E. Ladas, V. Lakshmikantham, Differential Equations in Abstract Spaces (Academic Press,
New York, 1972)
13. A. Magran, A new tool to study real dynamics: the convergence plane. Appl. Math. Comput.
248, 215224 (2014)
14. J. Mikusinski, The Bochner Integral (Academic Press, New York, 1978)
15. F.A. Potra, V. Ptak, Nondiscrete Induction and Iterative Processes (Pitman Publication, London,
1984)
16. P.D. Proinov, New general convergence theory for iterative processes and its applications to
Newton-Kantorovich type theorems. J. Complex. 26, 342 (2010)
17. G.E. Shilov, Elementary Functional Analysis (Dover Publications Inc., New York, 1996)
Chapter 8
Approximate Solutions of Equations in
Abstract g-Fractional Calculus
The novelty of this chapter is the design of suitable iterative methods for generating
a sequence approximating solutions of equations on Banach spaces. Applications of
the semi-local convergence are suggested including Banach space valued functions
of fractional calculus, where all integrals are of Bochner-type. It follows [6].
8.1 Introduction
Secant method: 1
xn+1 = xn xn1 , xn ; F F (xn ) , (8.1.3)
G (xn+1 ) = G (xn ) A1
n F (x n ) , (8.1.5)
Set 1 = 0 .
(m 3 ) There exists a continuous and nondecreasing function : [0, +)3
[0, +) such that for each x, y 1
(, s, s) < 1,
0 (s) < 1
and
G (x0 ) x0 s ,
1 q0
where q0 = (, s, s) .
(m 7 ) U (x0 , s) .
Next, we present the semi-local convergence analysis for method (8.1.5) using
the conditions (M) and the preceding notation.
Theorem 8.1 Assume that the conditions (M) hold. Then, sequence {xn } generated
by method (8.1.5) starting at x0 is well defined in U (x0 , s), remains in U (x0 , s)
for each n = 0, 1, 2, ... and converges to a solution x U (x0 , s) of equation
F (x) = 0. The limit point x is the unique solution of equation F (x) = 0 in
U (x0 , s) .
Proof By the definition of s and (m 5 ), we have x1 U (x0 , s). The proof is based on
mathematical induction on k. Suppose that xk xk1 q0k1 and xk x0 s.
We get by (8.1.5), (m 2 ) (m 5 ) in turn that
G (xk+1 ) G (xk ) = A1
k F (x k ) =
1
A (F (xk ) F (xk1 ) Ak1 (G (xk ) G (xk1 )))
k
A1
k
F (xk ) F (xk1 ) Ak1 (G (xk ) G (xk1 ))
and by (m 6 )
0 (s) s + G (x0 ) x0 s.
1 q0m k
xk+m xk q .
1 q0 0
1
A F x F (xk ) Ak G x G (xk )
k
1 0 x xk , xk+1 x0 , xk x0 G x G (xk )
q0 G x G (xk ) q0k+1 x x0 ,
Notice that
(u 1 , u 2 , u 3 ) (u 1 , u 2 , u 3 )
(2) Suppose that there exist > 0, 1 > 0 and L L (B1 , B2 ) with L 1
L (B2 , B1 ) such that 1
L 1
A (x, y) L 1
and
2 := 1 1 < 1.
Then, it follows from the Banach lemma on invertible operators [12], and
1
L A (x, y) L 1 1 = 2 < 1
1
that A (x, y)1 L (B2 , B1 ). Let = 1 2
. Then, under these replacements, condi-
tion (m 2 ) is implied, therefore it can be dropped from the conditions (M).
Remark 8.3 Section 8.2 has an interest independent of Sect. 8.3. It is worth noticing
that the results especially of Theorem 8.1 can apply in Abstract g-Fractional Calculus
as illustrated in Sect. 8.3. By specializing function , we can apply the results of say
Theorem 8.1 in the examples suggested in Sect. 8.3. In particular for (8.3.33), we
choose for u 1 0, u 2 0, u 3 0
1
(u 1 , u 2 , u 3 ) = ,
() ( + 1)
2
(u 1 , u 2 , u 3 ) = ,
() ( + 1)
3
(u 1 , u 2 , u 3 ) = ,
() ( + 1)
Here we deal with Banach space (X, ) valued functions f of real domain [a, b].
All integrals here are of Bochner-type, see [15]. The derivatives of f are defined
similarly to numerical ones, see [18], pp. 8386 and p. 93.
Here both backgrounds needed come from [5].
(I) We need
Definition 8.4 ([5]) Let > 0,
= n,
the ceiling of the number. Let f
C n ([a, b] , X ), where [a, b] R, and (X, ) is a Banach space. Let g C 1 ([a, b]),
strictly increasing, such that g 1 C n ([g (a) , g (b)]) .
We define the left generalized g-fractional derivative X -valued of f of order as
follows:
x
1 (n)
Da+;g f (x) := (g (x) g (t))n1 g (t) f g 1 (g (t)) dt,
(n ) a
(8.3.1)
x [a, b]. The last integral is of Bochner
type.
If
/ N, by [5], we have that Da+;g f C ([a, b] , X ).
We see that
(n)
n
Ja;g f g 1 g (x) = Da+;g f (x) ,x [a, b] . (8.3.2)
We set
n
n
Da+;g f (x) := f g 1 g (x) C ([a, b] , X ) ,n N, (8.3.3)
0
Da+;g f (x) = f (x) , x [a, b] .
n1
(g (x) g (a))i (i)
f (x) = f (a) + f g 1 (g (a)) +
i=1
i!
1 x
(g (x) g (t))1 g (t) Da+;g f (t) dt =
() a
8.3 Applications to X -valued g-Fractional Calculus 127
n1
(g (x) g (a))i (i)
f (a) + f g 1 (g (a)) + (8.3.4)
i=1
i!
g(x)
1
(g (x) z)1
Da+;g f g 1 (z) dz.
() g(a)
and
D 0y+;g f (x) = f (x) , x [a, b] . (8.3.7)
For > 0,
/ N, by convention we set that
D y+;g f (x) = 0, forx < y, x, y [a, b] . (8.3.8)
Similarly, we define
1 y (n)
Dx+;g f (y) := (g (y) g (t))n1 g (t) f g 1 (g (t)) dt,
(n ) x
(8.3.9)
for any a x y b;
(n)
n
Dx+;g f (y) = f g 1 g (y) , x, y [a, b] , (8.3.10)
and
0
Dx+;g f (y) = f (y) , y [a, b] . (8.3.11)
For > 0,
/ N, by convention we set that
Dx+;g f (y) = 0, fory < x, x, y [a, b] . (8.3.12)
128 8 Approximate Solutions of Equations in Abstract g-Fractional Calculus
That is
Da+;g f (a) = 0, (8.3.15)
and
D y+;g f (y) = Dx+;g f (x) = 0, x, y [a, b] . (8.3.16)
n1
(k)
f g 1 (g (y))
f (x) f (y) = (g (x) g (y))k +
k=1
k!
1 x
(g (x) g (t))1 g (t) D y+;g f (t) dt, x [y, b] , (8.3.17)
() y
and
n1
(k)
f g 1 (g (x))
f (y) f (x) = (g (y) g (x))k +
k=1
k!
1 y
(g (y) g (t))1 g (t) Dx+;g f (t) dt, y [x, b] , (8.3.18)
() x
8.3 Applications to X -valued g-Fractional Calculus 129
(A1 ( f )) (x, y) :=
n1
( f g1 )(k) (g(y)) 1
(g (x) g (y)) k1
+ D
f (x) (g(x)g(y)) , for x > y,
k=1 k! y+;g (+1)
( f g1 )(k) (g(x))
n1 1
(g (y) g (x))k1 + Dx+;g
f (y) (g(y)g(x)) , for x < y,
k! (+1)
k=1
f (n) (x) , when x = y,
(8.3.19)
x, y [a, b]; > 0, n =
.
We may assume that (see [13], p. 3)
(A1 ( f )) (x, x) (A1 ( f )) (y, y) = f (n) (x) f (n) (y) (8.3.20)
(n)
f g 1 (g (x)) f (n) g 1 (g (y)) |g (x) g (y)| , x, y [a, b] ;
where > 0.
We estimate and have
(i) case of x > y :
(g (x) g (y))
D y+;g f (x) (8.3.21)
( + 1)
(by [9])
1 x
(g (x) g (t))1 g (t) D y+;g f (t) D y+;g f (x) dt (8.3.22)
() y
t, x, y [a, b] : x t y; 1 > 0)
130 8 Approximate Solutions of Equations in Abstract g-Fractional Calculus
1 x
(g (x) g (t))1 g (t) (g (x) g (t)) dt =
() y
1 x
1 (g (x) g (y))+1
(g (x) g (t)) g (t) dt = . (8.3.24)
() y () ( + 1)
1 (g (x) g (y))+1
, (8.3.25)
() ( + 1)
x, y [a, b] : x > y.
(ii) case of y > x : We have that
t, y, x [a, b] : y t x; 2 > 0)
2 y
(g (y) g (t))1 g (t) (g (y) g (t)) dt = (8.3.30)
() x
2 y
2 (g (y) g (x))+1
(g (y) g (t)) g (t) dt = . (8.3.31)
() x () ( + 1)
2 (g (y) g (x))+1
, x, y [a, b] : y > x.
() ( + 1)
|g (x) g (y)|+1
, x, y [a, b] . (8.3.33)
() ( + 1)
Now based on (8.3.20) and (8.3.33), we can apply our numerical methods presented
in this chapter to solve f (x) = 0.
(II) In the next background again we use [5].
We need
Definition 8.7 ([5]) Let > 0,
= n,
the ceiling of the number. Let
f C n ([a, b] , X ), where [a, b] R, and (X, ) is a Banach space. Let
g C 1 ([a, b]) , strictly increasing, such that g 1 C n ([g (a) , g (b)]) .
We define the right generalized g-fractional derivative X -valued of f of order
as follows:
b
(1)n (n)
Db;g f (x) := (g (t) g (x))n1 g (t) f g 1 (g (t)) dt,
(n ) x
(8.3.35)
x [a, b]. The last integral is of Bochner type.
If / N, by [5], we have that Db;g f C ([a, b] , X ).
We see that
(n)
n
Jb;g (1)n f g 1 g (x) = Db;g f (x) , a x b. (8.3.36)
We set
n
n
Db;g f (x) := (1)n f g 1 g (x) C ([a, b] , X ) ,n N, (8.3.37)
0
Db;g f (x) := f (x) , x [a, b] .
n1
(g (x) g (b))i (i)
f (x) = f (b) + f g 1 (g (b)) +
i=1
i!
1 b
(g (t) g (x))1 g (t) Db;g f (t) dt =
() x
n1
(g (x) g (b))i (i)
f (b) + f g 1 (g (b)) + (8.3.39)
i=1
i!
g(b)
1
(z g (x))1
Db;g f g 1 (z) dz.
() g(x)
D 0y;g f (x) = f (x) , x [a, b] . (8.3.42)
For > 0,
/ N, by convention we set that
D y;g f (x) = 0, for x > y, x, y [a, b] . (8.3.43)
Similarly, we define
(1)n x (n)
Dx;g f (y) = (g (t) g (y))n1 g (t) f g 1 (g (t)) dt,
(n ) y
(8.3.44)
8.3 Applications to X -valued g-Fractional Calculus 133
0
Dx;g f (y) = f (y) , y [a, b] . (8.3.46)
For > 0,
/ N, by convention we set that
Dx;g f (y) = 0, for y > x, x, y [a, b] . (8.3.47)
We get that
1 (n)
f g
g
D ,[a,b]
b;g f (x) (g (b) g (x))n (8.3.48)
(n + 1)
(n)
f g 1 g
,[a,b]
(g (b) g (a))n , x [a, b] .
(n + 1)
That is
Db;g f (b) = 0, (8.3.49)
and
D y;g f (y) = Dx;g f (x) = 0, x, y [a, b] . (8.3.50)
n1
(k)
f g 1 (g (y))
f (x) f (y) = (g (x) g (y))k +
k=1
k!
1 y
(g (t) g (x))1 g (t) D y;g f (t) dt, all a x y b. (8.3.51)
() x
Also, we have
n1
(k)
f g 1 (g (x))
f (y) f (x) = (g (y) g (x))k +
k=1
k!
1 x
(g (t) g (y))1 g (t) Dx;g f (t) dt, all a y x b. (8.3.52)
() y
134 8 Approximate Solutions of Equations in Abstract g-Fractional Calculus
(A2 ( f )) (x, y) :=
n1
( f g1 )(k) (g(y)) 1
(g (x) g (y)) k1
D
f (x) (g(y)g(x)) , for x < y,
k=1 k! y;g (+1)
( f g1 )(k) (g(x))
n1 1
(g (y) g (x))k1 Dx;g
f (y) (g(x)g(y)) , for x > y,
k! (+1)
k=1
f (n) (x) , when x = y,
(8.3.53)
x, y [a, b]; > 0, n =
.
We may assume that ([13], p. 3)
(A2 ( f )) (x, x) (A2 ( f )) (y, y) = f (n) (x) f (n) (y) (8.3.54)
where > 0.
We estimate and have
(i) case of x < y :
(by [9])
1 y
(g (t) g (x))1 g (t) D y;g f (t) D y;g f (x) dt (8.3.56)
() x
t, x, y [a, b] : y t x; 1 > 0)
8.3 Applications to X -valued g-Fractional Calculus 135
1 y
(g (t) g (x))1 g (t) (g (t) g (x)) dt =
() x
1 y
1 (g (y) g (x))+1
(g (t) g (x)) g (t) dt = . (8.3.58)
() x () ( + 1)
1 (g (y) g (x))+1
, (8.3.59)
() ( + 1)
x, y [a, b] : x < y.
(ii) case of x > y :
(g (x) g (y))
=
Dx;g f (y)
( + 1)
x
1 1
(g (t) g (y)) g (t) D f (t) D f (y) dt
() y x;g x;g
x (8.3.61)
1
(g (t) g (y))1
g (t) Dx;g f (t) Dx;g f (y) dt (8.3.62)
() y
t, y, x [a, b] : x t y; 2 > 0)
2 x
(g (t) g (y))1 g (t) (g (t) g (y)) dt =
() y
2 x
(g (t) g (y)) g (t) dt = (8.3.64)
() y
136 8 Approximate Solutions of Equations in Abstract g-Fractional Calculus
2 (g (x) g (y))+1
. (8.3.65)
() ( + 1)
2 (g (x) g (y))+1
, x, y [a, b] : x > y. (8.3.66)
() ( + 1)
|g (x) g (y)|+1
, x, y [a, b] . (8.3.67)
() ( + 1)
Now based on (8.3.54) and (8.3.67), we can apply our numerical methods presented
in this chapter to solve f (x) = 0.
In both fractional applications + 1 2, iff 1.
Also some examples for g follow:
g (x) = e x , x [a, b] R,
g (x) = sin x,
(8.3.69)
g (x) = tanx,
where x 2 + , 2 , where > 0 small.
References
1. C.D. Aliprantis, K.C. Border, Infinite Dimensional Analysis (Springer, New York, 2006)
2. S. Amat, S. Busquier, S. Plaza, Chaotic dynamics of a third-order Newton-type method. J.
Math. Anal. Appl. 366(1), 164174 (2010)
3. G.A. Anastassiou, Strong right fractional calculus for Banach space valued functions. Rev.
Proyecc. 36(1), 149186 (2017)
4. G.A. Anastassiou, A strong fractional calculus theory for Banach space valued functions.
Nonlinear Funct. Anal. Appl. (Korea) (accepted for publication, 2017)
5. G.A. Anastassiou, Principles of general fractional analysis for Banach space valued functions
(submitted for publication, 2017)
6. G.A. Anastassiou, I.K. Argyros, Iterated convergence on Banach space valued functions of
abstract g-fractional calculus (submitted, 2017)
References 137
7. I.K. Argyros, A unifying local-semilocal convergence analysis and applications for two-point
Newton-like methods in Banach space. J. Math. Anal. Appl. 298, 374397 (2004)
8. I.K. Argyros, A. Magran, Iterative Methods and their Dynamics with Applications (CRC
Press, New York, 2017)
9. Bochner integral, Encyclopedia of Mathematics, http://www.encyclopediaofmath.org/index.
php?title=Bochner_integral&oldid=38659
10. M. Edelstein, On fixed and periodic points under contractive mappings. J. London Math. Soc.
37, 7479 (1962)
11. J.A. Ezquerro, J.M. Gutierrez, M.A. Hernandez, N. Romero, M.J. Rubio, The Newton method:
from Newton to Kantorovich (Spanish). Gac. R. Soc. Mat. Esp. 13, 5376 (2010)
12. L.V. Kantorovich, G.P. Akilov, Functional Analysis in Normed Spaces (Pergamon Press, New
York, 1982)
13. G.E. Ladas, V. Lakshmikantham, Differential Equations in Abstract Spaces (Academic Press,
New York, 1972)
14. A. Magran, A new tool to study real dynamics: the convergence plane. Appl. Math. Comput.
248, 215224 (2014)
15. J. Mikusinski, The Bochner Integral. (Academic Press, New York, 1978)
16. F.A. Potra, V. Ptak, Nondiscrete Induction and Iterative Processes (Pitman Publication, London,
1984)
17. P.D. Proinov, New general convergence theory for iterative processes and its applications to
Newton-Kantorovich type theorems. J. Complex. 26, 342 (2010)
18. G.E. Shilov, Elementary Functional Analysis (Dover Publications Inc., New York, 1996)
Chapter 9
Generating Sequences for Solving in Abstract
g-Fractional Calculus
The aim of this chapter is to utilize proper iterative methods for solving equations on
Banach spaces. The differentiability of the operator involved is not assumed neither
the convexity of its domain. Applications of the semi-local convergence are suggested
including Banach space valued functions of fractional calculus, where all integrals
are of Bochner-type. It follows [6].
9.1 Introduction
Secant method: 1
xn+1 = xn xn1 , xn ; F F (xn ) , (9.1.3)
G (xn+1 ) = G (xn ) A1
n F (x n ) , (9.1.5)
Set 1 = 0 .
(m 3 ) There exists a continuous and nondecreasing function : [0, +)3
[0, +) such that for each x, y 1
(, s, s) < 1,
0 (s) < 1
and
G (x0 ) x0 s ,
1 q0
where q0 = (, s, s) .
(m 7 ) U (x0 , s) .
Next, we present the semi-local convergence analysis for method (9.1.5) using
the conditions (M) and the preceding notation.
Theorem 9.1 Assume that the conditions (M) hold. Then, sequence {xn } generated
by method (9.1.5) starting at x0 is well defined in U (x0 , s), remains in U (x0 , s)
for each n = 0, 1, 2, ... and converges to a solution x U (x0 , s) of equation
F (x) = 0. The limit point x is the unique solution of equation F (x) = 0 in
U (x0 , s) .
Proof By the definition of s and (m 5 ), we have x1 U (x0 , s). The proof is based on
mathematical induction on k. Suppose that xk xk1 q0k1 and xk x0 s.
We get by (9.1.5), (m 2 ) (m 5 ) in turn that
G (xk+1 ) G (xk ) = A1
k F (x k ) =
1
A (F (xk ) F (xk1 ) Ak1 (G (xk ) G (xk1 )))
k
A1
k
F (xk ) F (xk1 ) Ak1 (G (xk ) G (xk1 ))
and by (m 6 )
0 (s) s + G (x0 ) x0 s.
1 q0m k
xk+m xk q .
1 q0 0
1
A F x F (xk ) Ak G x G (xk )
k
1 0 x xk , xk+1 x0 , xk x0 G x G (xk )
q0 G x G (xk ) q0k+1 x x0 ,
Notice that
(u 1 , u 2 , u 3 ) (u 1 , u 2 , u 3 )
[7, 12, 16] and influence the weaking of the convergence criterion in (m 6 ), error
bounds and the precision of s.
(2) Suppose that there exist > 0, 1 > 0 and L L (B1 , B2 ) with L 1
L (B2 , B1 ) such that 1
L 1
A (x, y) L 1
and
2 := 1 1 < 1.
Then, it follows from the Banach lemma on invertible operators [12], and
1
L A (x, y) L 1 1 = 2 < 1
1
that A (x, y)1 L (B2 , B1 ). Let = 1 2
. Then, under these replacements, condi-
tion (m 2 ) is implied, therefore it can be dropped from the conditions (M).
Remark 9.3 Sect. 9.2 has an interest independent of Sect. 9.3. It is worth noticing that
the results especially of Theorem 9.1 can apply in Abstract g-Fractional Calculus as
illustrated in Sect. 9.3. By specializing function , we can apply the results of say
Theorem 9.1 in the examples suggested in Sect. 9.3. In particular for (9.3.21), we
choose for u 1 0, u 2 0, u 3 0
1
(u 1 , u 2 , u 3 ) = ,
() ( + 1)
2
(u 1 , u 2 , u 3 ) = ,
() ( + 1)
3
(u 1 , u 2 , u 3 ) = ,
() ( + 1)
Here we deal with Banach space (X, ) valued functions f of real domain [a, b].
All integrals here are of Bochner-type, see [15]. The derivatives of f are defined
similarly to numerical ones, see [18], pp. 8386 and p. 93.
Here both needed backgrounds come from [5].
Let > 1, / N, with integral part [] = n N. Let g : [a, b] R be a
1
g 1
strictly increasing function, such that g C 1 ([a, b]), C n ([g (a) , g (b)]), and
let f C ([a, b] , X ). It clear then we obtain that f g
n
C n ([g (a) , g (b)] , X ). Let := [] = n (0 < < 1).
(I) See [5]. Let h C ([g (a) , g (b)] , X ), we define the X -valued left Riemann-
Liouville fractional integral as
1 z
Jz0 h (z) := (z t)1 h (t) dt, (9.3.1)
() z0
So let h Cg(x) ([g (a) , g (b)] , X ); we define the X -valued left g -generalized
fractional derivative of h of order , of Canavati type, over [g (x) , g (b)] as
g(x)
Dg(x) h := J1 h (n) . (9.3.3)
Clearly, for h Cg(x) ([g (a) , g (b)] , X ), there exists
1 d z
Dg(x) h (z) = (z t) h (n) (t) dt, (9.3.4)
(1 ) dz g(x)
n1
(k)
f g 1 (g (x))
f (y) f (x) = (g (y) g (x))k + (9.3.6)
k=1
k!
g(y)
1
(g (y) t)1 Dg(x) f g 1 (t) dt, for all y [a, b] : y x.
() g(x)
Alternatively, for f g 1 Cg(y) ([g (a) , g (b)] , X ), where y [a, b], we can
write (again X -valued left fractional Taylors formula) that:
n1
(k)
f g 1 (g (y))
f (x) f (y) = (g (x) g (y))k + (9.3.7)
k=1
k!
g(x)
1
(g (x) t)1 Dg(y) f g 1 (t) dt, for all x [a, b] : x y.
() g(y)
Here we consider f C n ([a, b] , X ), such that f g 1 Cg(x) ([g (a) ,
1
g (b)], X ), for every x [a, b]; which is the same as f g Cg(y) ([g (a) ,
g (b)], X ), for every y [a, b] (i.e. exchange roles of x and y); we write that as
f g 1 Cg+
([g (a) , g (b)] , X ).
We have that
z
1
1 d (n)
Dg(y) f g (z) = (z t) f g 1 (t) dt, (9.3.8)
(1 ) dz g(y)
k=1
k!
Dg(x) f g 1 (g (y)) (g(y)g(x)) , y > x,
(+1)
(A1 ( f )) (x, y) := n1 ( f g1 ) (g(y)) (g (x) g (y))k1 +
(k)
(9.3.9)
D f g 1 (g (x)) (g(x)g(y))1 , x > y,
k=1
k!
g(y) (+1)
(n)
f (x) , x = y.
146 9 Generating Sequences for Solving in Abstract g-Fractional Calculus
(g (y) g (x))
Dg(x) f g 1 (g (y))
( + 1)
1
= (g (y) t)1 Dg(x)
f g 1 (t) Dg(x)
f g 1 (g (y)) dt
() g(x)
(9.3.11)
(by [9])
1 g(y)
(g (y) t)1 Dg(x) f g 1 (t) Dg(x)
f g 1 (g (y)) dt
() g(x)
for every t, g (y) , g (x) [g (a) , g (b)] such that g (y) t g (x) ; 1 > 0)
g(y)
1
(g (y) t)1 (g (y) t) dt = (9.3.13)
() g(x)
1 g(y)
1 (g (y) g (x))+1
(g (y) t) dt = . (9.3.14)
() g(x) () ( + 1)
1 (g (y) g (x))+1
f (y) f (x) (A1 ( f )) (x, y) (g (y) g (x)) ,
() ( + 1)
(9.3.15)
9.3 Applications to X -valued g-Fractional Calculus of Canavati Type 147
(g (x) g (y))
=
Dg(y) f g 1 (g (x)) (9.3.16)
( + 1)
1
g(x)
1
1
1
(g (x) t) Dg(y) f g (t) Dg(y) f g (g (x)) dt
() g(y)
1 g(x)
(g (x) t)1 Dg(y) f g 1 (t) Dg(y)
f g 1 (g (x)) dt
() g(y)
(9.3.17)
(we assume that
D f g 1 (t) Dg(y) f g 1 (g (x)) 2 |t g (x)| , (9.3.18)
g(y)
for all t, g (x) , g (y) [g (a) , g (b)] such that g (x) t g (y) ; 2 > 0)
g(x)
2
(g (x) t)1 (g (x) t) dt = (9.3.19)
() g(y)
2 g(x)
2 (g (x) g (y))+1
(g (x) t) dt = .
() g(y) () ( + 1)
2 (g (x) g (y))+1
f (y) f (x) (A1 ( f )) (x, y) (g (y) g (x)) ,
() ( + 1)
(9.3.20)
for any x, y [a, b] : x > y.
|g (y) g (x)|+1
f (y) f (x) (A1 ( f )) (x, y) (g (y) g (x)) ,
() ( + 1)
(9.3.21)
x, y [a, b] (the case of x = y is trivially true).
148 9 Generating Sequences for Solving in Abstract g-Fractional Calculus
We may choose that () < 1.
Also we notice here that + 1 > 2.
(II) See [5] again. Let h C ([g (a) , g (b)] , X ), we define the X -valued right
Riemann-Liouville fractional integral as
1 z0
Jz0 h (z) := (t z)1 h (t) dt, (9.3.22)
() z
So let h Cg(x) ([g (a) , g (b)] , X ); we define the X -valued right g -generalized
fractional derivative of h of order , of Canavati type, over [g (a) , g (x)] as
1 (n)
Dg(x) h := (1)n1 Jg(x) h . (9.3.24)
Clearly, for h Cg(x) ([g (a) , g (b)] , X ), there exists
g(x)
(1)n1 d
Dg(x) h (z) = (t z) h (n) (t) dt, (9.3.25)
(1 ) dz z
and
0
Dg(x) f g 1 (z) = f g 1 (z) , (9.3.28)
n1
(k)
f g 1 (g (x))
f (y) f (x) = (g (y) g (x))k +
k=1
k!
g(x)
1
(t g (y))1 Dg(x) f g 1 (t) dt, all a y x. (9.3.29)
() g(y)
Alternatively, for f g 1 Cg(y) ([g (a) , g (b)] , X ), where y [a, b], 1
(again X -valued right fractional Taylors formula) that:
n1
(k)
f g 1 (g (y))
f (x) f (y) = (g (x) g (y))k +
k=1
k!
g(y)
1
(t g (x))1 Dg(y) f g 1 (t) dt, all a x y. (9.3.30)
() g(x)
Here we consider f C n ([a, b] , X ), such that f g 1 Cg(x) ([g (a) ,
1
g (b)], X ), for every x [a, b]; which is the same as f g Cg(y) ([g (a) ,
g (b)], X ), for every y [a, b] ; (i.e. exchange roles of x and y) we write that as
f g 1 Cg
([g (a) , g (b)] , X ).
We have that
g(y)
1
(1)n1 d (n)
Dg(y) f g (z) = (t z) f g 1 (t) dt,
(1 ) dz z
(9.3.31)
for all g (a) z g (y) g (b) .
So here we work with f C n ([a, b] , X ), such that f g 1 Cg
([g(a) ,
g (b)], X ).
We define the X -valued right linear fractional operator
(k)
n1 ( f g 1 ) (g(x))
(g (y) g (x)) 1
k1
k=1
k!
Dg(x) f g 1 (g (y)) (g(x)g(y)) , x > y,
(+1)
(A2 ( f )) (x, y) := n1 ( f g1 ) (g(y)) (g (x) g (y))k1
(k)
(9.3.32)
k=1 k!
D 1
g(y) f g
1
(g (x)) (g(y)g(x))
(+1)
, y > x,
(n)
f (x) , x = y.
(g (x) g (y))
Dg(x) f g 1 (g (y)) (9.3.35)
( + 1)
(g (y)) dt
(by [9])
1 g(x)
(t g (y))1 Dg(x) f g 1 (t) Dg(x)
f g 1 (g (y)) dt
() g(y)
(9.3.36)
(we assume here that
D f g 1 (t) Dg(x) f g 1 (g (y)) 1 |t g (y)| , (9.3.37)
g(x)
for every t, g (y) , g (x) [g (a) , g (b)] such that g (x) t g (y) ; 1 > 0)
g(x)
1
(t g (y))1 (t g (y)) dt =
() g(y)
1 g(x)
1 (g (x) g (y))+1
(t g (y)) dt = . (9.3.38)
() g(y) () ( + 1)
9.3 Applications to X -valued g-Fractional Calculus of Canavati Type 151
1 (g (x) g (y))+1
f (x) f (y) (A2 ( f )) (x, y) (g (x) g (y)) ,
() ( + 1)
(9.3.39)
x, y [a, b] : x > y.
(ii) Case of x < y : We have that
1
(g (y) g (x))
=
Dg(y) f g (g (x))
( + 1)
g(y)
1 1 1 1
(t g (x)) D g(y) f g (t) D g(y) f g (g (x)) dt
() g(x)
1 g(y)
(t g (x))1 Dg(y) f g 1 (t) Dg(y)
f g 1 (g (x)) dt
() g(x)
(9.3.41)
(we assume that
D f g 1 (t) Dg(y) f g 1 (g (x)) 2 |t g (x)| , (9.3.42)
g(y)
g(y)
2
(t g (x)) dt = (9.3.43)
() g(x)
2 (g (y) g (x))+1
. (9.3.44)
() ( + 1)
2 (g (y) g (x))+1
f (x) f (y) (A2 ( f )) (x, y) (g (x) g (y)) ,
() ( + 1)
(9.3.45)
x, y [a, b] : x < y.
152 9 Generating Sequences for Solving in Abstract g-Fractional Calculus
|g (x) g (y)|+1
f (x) f (y) (A2 ( f )) (x, y) (g (x) g (y)) ,
() ( + 1)
(9.3.46)
x, y [a, b] ((9.3.46) is trivially true when x = y).
One may choose () < 1.
Here again + 1 > 2.
Conclusion 9.6 Based on (9.3.10) and (9.3.21) of (I), and based on (9.3.33) and
(9.3.46) of (II), using our numerical results presented earlier, we can solve numeri-
cally f (x) = 0.
g (x) = e x , x [a, b] R,
g (x) = sin x,
g (x) = tanx,
where x 2 + , 2 , with > 0 small.
References
1. C.D. Aliprantis, K.C. Border, Infinite Dimensional Analysis (Springer, New York, 2006)
2. S. Amat, S. Busquier, S. Plaza, Chaotic dynamics of a third-order Newton-type method.
J. Math. Anal. Appl. 366(1), 164174 (2010)
3. G.A. Anastassiou, Strong right fractional calculus for banach space valued functions. Rev
Proyecc. 36(1), 149186 (2017)
4. G.A. Anastassiou, A strong Fractional Calculus Theory for Banach space valued functions,
Nonlinear Functional Analysis and Applications (Korea) (2017). accepted for publication
5. G.A. Anastassiou, Principles of general fractional analysis for Banach space valued functions
(2017). submitted for publication
6. G.A. Anastassiou, I.K. Argyros, Equations on Banach space valued functions of abstract
g-fractional calculus. J. Comput. Anal. Appl. 25, 15471560 (2018)
7. I.K. Argyros, A unifying local-semilocal convergence analysis and applications for two-point
Newton-like methods in Banach space. J. Math. Anal. Appl. 298, 374397 (2004)
8. I.K. Argyros, A. Magran, Iterative Methods and Their Dynamics with Applications (CRC
Press, New York, 2017)
9. Bochner integral, Encyclopedia of Mathematics, http://www.encyclopediaofmath.org/index.
php?title=Bochner_integral&oldid=38659
10. M. Edelstein, On fixed and periodic points under contractive mappings. J. London Math. Soc.
37, 7479 (1962)
11. J.A. Ezquerro, J.M. Gutierrez, M.A. Hernandez, N. Romero, M.J. Rubio, The Newton method:
from Newton to Kantorovich (Spanish). Gac. R. Soc. Mat. Esp. 13, 5376 (2010)
12. L.V. Kantorovich, G.P. Akilov, Functional Analysis in Normed Spaces (Pergamon Press, New
York, 1982)
13. G.E. Ladas, V. Lakshmikantham, Differential Equations in Abstract Spaces (Academic Press,
New York, London, 1972)
References 153
14. A. Magran, A new tool to study real dynamics: the convergence plane. Appl. Math. Comput.
248, 215224 (2014)
15. J. Mikusinski, The Bochner Integral (Academic Press, New York, 1978)
16. F.A. Potra, V. Ptak, Nondiscrete Induction and Iterative Processes (Pitman Publishing, London,
1984)
17. P.D. Proinov, New general convergence theory for iterative processes and its applications to
Newton-Kantorovich type theorems. J. Complex. 26, 342 (2010)
18. G.E. Shilov, Elementary Functional Analysis (Dover Publications Inc, New York, 1996)
Chapter 10
Numerical Optimization and Fractional
Invexity
We present some proximal methods with invexity results involving fractional calcu-
lus. It follows [3, 4].
10.1 Introduction
2
is strongly convex [9]. That is there exist
L 2 , r2 > 0 such that for each u, v U (x , r2 )
1 2
d (u, x ) 1 d 2 (v, x ) L 2 d(u, v). (10.2.2)
2 2
Let
r = min{r1 , r2 } and L 0 = L 1 + L 2 . (10.2.3)
F(x ) x L 3 , (10.2.8)
L := L 1 + 2 L 2 < 1. (10.2.9)
10.2 Convergence of Method (10.1.2) 157
L3
r := . (10.2.10)
1L
and
G(u) x G(u) G(x ) + G(x ) x
Ld(u, x ) + F(x ) x
Ld(u, x ) + L 3 r . (10.2.13)
The result now follows from (10.2.9), (10.2.12), (10.2.13) and the contraction map-
ping principle [1, 58].
n
Let X = [ai , bi ] .
i=1
1. Let 0 < < 1, we consider the left Caputo fractional partial derivatives of f
of order :
xi
f (x) 1 f (x1 , x2 , . . . , ti , . . . , xn ))
= (xi ti ) dti , (10.3.1)
xi (1 ) ai xi
158 10 Numerical Optimization and Fractional Invexity
xi
f (x) 1
(x t )
dt
x (1 ) ai
i i i
i
f (x1 , x2 , . . . , xi1 , ., xi+1 , . . . , xn ))
dt
xi
i
,(ai ,bi )
(xi ti )1 f (x , x , . . . , x , ., x , . . . , xn ))
dti
1 2 i1 i+1
=
(2 ) xi
,(ai ,bi )
< ,
for all i = 1, 2, . . . , n. Therefore, xf (x)
exist for all i = 1, 2, . . . n.
i
Now we consider the left fractional Gradient of F of order , 0 < < 1 :
f (x ) f (x )
+ f (x ) = , . . . , . (10.3.2)
x1 xn
f ((x1 ,...,.,...,xn ))
where x = (x1 , . . . , xn ) X, i = 1, 2, . . . n and xi
L (ai , bi ),
i = 1, 2, . . . n. Note that
bi
f (x)
1
(xi ti ) dti
xi (1 ) xi
f (x1 , x2 , . . . , xi1 , ., xi+1 , . . . , xn ))
dt
xi
i
,(ai ,bi )
(bi xi )1 f (x , x , . . . , x , ., x , . . . , x n ))
dti
1 2 i1 i+1
=
(2 ) x i ,(ai ,bi )
< , (10.3.4)
for all i = 1, 2, . . . , n. Therefore, xf (x)
exist for all i = 1, 2, . . . n.
i
Now we consider the right fractional Gradient of F of order , 0 < < 1 :
10.3 Multivariate Fractional Derivatives and Invexity 159
f (x ) f (x )
f (x ) =
,..., . (10.3.5)
x1 xn
k f (x)
where xi
= xi
. . . x f, ktimes composition of left partial fractional
i
i
4. Define for k N : k
f = . . . f, k times composition of right frac-
tional gradient, i.e.,
k f (x ) k f (x )
k
f = ,..., , (10.3.7)
x1 xn
k f (x)
where xi
= xi
. . . x f, ktimes composition of right partial fractional
i
k
derivative, i = 1, 2, . . . n. We assume that x f exist for all i = 1, 2, . . . n.
i
5. Let 1, we consider the left Caputo fractional partial derivatives of f of
order (
= m N,
. ceiling of the number [2]):
f (x) 1 xi
m f (x1 , x2 , . . . , ti , . . . , xn ))
= (xi ti )m1 dti ,
xi (m ) ai xim
(10.3.8)
i = 1, 2, . . . n. We set xf m(x) equal to the ordinary partial derivative xf m(x) . We
m m
i i
assume that
f
m
(x1 , . . . , ., . . . , xn ) L (ai , bi )
xim
i.e., m
f
x m (x1 , . . . , ., . . . , xn ) <
i ,(ai ,bi )
We can replace in Definition 10.3.1 of Verma [9] (or similar invexity formula),
f (x ) by ++ f (x ).
6. Let 1, we consider the right Caputo fractional partial derivatives of f of
order (
= m):
bi
f (x) (1)m m f (x1 , x2 , . . . , ti , . . . , xn ))
= (xi ti )m1 dti ,
xi (m ) xi xim
(10.3.11)
m
i = 1, 2, . . . n. We set x mf = (1)m x mf (where x mf is the ordinary partial). We
m m
i i i
assume that
m f
(x1 , . . . , ., . . . , xn ) L (ai , bi )
xim
for all i = 1, 2, . . . n. Therefore, xf (x)
exist for all i = 1, 2, . . . n. Now we
i
consider the right fractional gradient of f of order , 1 :
f (x ) f (x )
f (x ) = ,..., . (10.3.12)
x1 xn
We can replace in Definition 10.3.1 of Verma [9] (or similar invexity formula),
f (x ) by f (x ).
f i (x)
(P) Minimize max , (10.3.13)
1i p gi (x)
where
f (x ) f (x )
+ f x = , ..., . (10.3.17)
x1 xn
References