Fractional integration and differentiation

arXiv:2309.03207v1 [math.GM] 18 Aug 2023

Yachmenev Andrey Igorevich

Moscow State University of Technology ”Stankin”
Vadkovsky Lane, 1, Moscow, 127055, Russia,
Undergraduate student
Yaremko Oleg Emanuilovich
Moscow State University of Technology ”Stankin”
Vadkovsky Lane, 1, Moscow, 127055, Russia,
Doctor of Physics and Mathematics, Professor, Docent
September 8, 2023

Abstract
In this paper, we introduce a new method for calculating fractional integrals
and differentials. The method involves an equation that we have obtained from
infinite applied integration by parts. The equation works for special class of
functions and provides a series representation of integration. This representa-
tion will be useful for working with smooth functions and for approximation
due to the potential reduction of the value of the sequential elements.
Keywords: Fractional integral, fractional differential, approximation, differen-
tial equation.

1 Introduction
Fractional integrals and differentials were first studied in the early 19th
century by mathematicians Liouville and Riemann. They were later expanded
upon by several other scientists, including Cauchy, Laplace and Lebesgue [1].
The use of fractional differentials is now seen in various fields such as diffu-
sion, thermal conductivity, fractal structures, and much more [1]. In modern
representation of fractional calculus, there are multiple definitions, but two
main ones stand out [2].
Riemann–Liouville integral:
Zx
1
I α f (x) = f (t)(x − t)α−1 dt (1)
Γ(α) α

1
 d ⌈α⌉ ⌈α⌉−α
 

 I
dx⌈α⌉ x
f(x) α > 0 

Dxα f (x) = 
 
(2)
 
 f (x) α = 0 

Ix−α f (x) α < 0

 


Grunwald-Letnikov derivative:
left-sided derivative:
⌊n⌋
GL 1 X Γ(α + 1)f (x − kh)
Daα+ f (x) = lim α (−1)k , nh = x − a (3)
h→0 h Γ(k + 1)Γ(α − k + 1)
k=0

right-sided derivative:
⌊n⌋
GL 1 X Γ(α + 1)f (x + kh)
Dbα− f (x) = lim α
(−1)k , nh = b − x (4)
h→0 h Γ(k + 1)Γ(α − k + 1)
k=0

Other definitions attempt to solve certain specific cases, but they are con-
ceptually similar to either the first or the second. The peculiarity of calculating
these approaches is similar in that it requires to calculate the sum of all ele-
ments in some approximation, since each of them can have an equal influence
on the result.

2 Derivation of a new equation

The basis for the derivation of a new representation of the fractional derivative
will be one of the basic properties of integration - integration by parts. For our
objectives, we define the boundaries of integration from the constant (zero) to
the variable x [3].
Zx Zx
f (t) = u dt = dv
f (t)dt = = f (x)x − f ′ (t)tdt (5)
0 f ′ (t) = du t=v 0

After applying this method, we obtain two components: an expression that

does not contain an integral, and an expression similar to the original integral.
Next, we will use integration by parts sequentially on the integral that appears
at each step.

f ′ (x)x2 f ′′ (t)t 2
Rx Rx Rx
f (t)dt = f (x)x − 0 f ′ (t)tdt = f (x)x − 2 + 0
0 2 dt = ...
(−1)k f (k) (x)xk+1
R x f (n+1) (t)t n+1 (6)
= nk=0 + (−1)n+1 0
P
(k+1)! (n+1)!
dt

After that, we turn to the limiting transition of the two resulting parts with
n → ∞. To obtain a new expression, it is worth highlighting a class of functions
for which the integral will tend to zero at the limit transition.
f (n) (x)
A suitable class of functions can be defined asf (x) ∈ C ∞ ∨ lim =0
n→∞ n!

2
 For this class of functions, the integral can be represented as an infinite
sum of their derivatives.
Zx n Z x (n+1) n+1 ∞
X (−1)k f (k) (x)xk+1 f (t)t X (−1)k f (k) (x)xk+1
f (t)dt = lim ( +(−1)n+1 dt) =
0 n→∞ (k + 1)! 0 (n + 1)! (k + 1)!
k=0 k=0
(7)
To utilize a different constant for integration, it is necessary to make a slight
adjustment to the equation.
Rx Rx Rα
α
f (x)dx = 0
f (x)dx − 0
f (x)dx
P∞ (−1)k k+1 d k f dk f (8)
= k=0 (k+1)! (x dxk
(x) − α k+1 dxk (a))

After getting Eq. (7) and Eq. (8) we present some illustrative examples.
Example 1.
Finding power rule for integration.
x
xm+1
Z
t m dt =
0 m+1
∞ m m
X (−1)k xk+1 d k m X (−1)k xk+1 m!xm−k m+1
X (−1)k m! xm+1
k
x = = x =
(k + 1)! dx (k + 1)!(m − k)! (k + 1)!(m − k)! m + 1
k=0 k=0 k=0
Example 2.
Finding integration of exponent.
Zx
e t dt = e t − 1
0

∞ ∞ ∞
X (−1)k xk+1 d k x X (−1)k xk+1 x X (−1)k xk+1
e = e = ex = e x (−e −x + 1) = e x − 1
(k + 1)! dxk (k + 1)! (k + 1)!
k=0 k=0 k=0
Example 3.
Finding integration of natural logarithm.
Zx
ln(x)dx = x ln(x) − x + 1
1

∞ ∞
X (−1)k k+1 d k ln k
k+1 d ln
X (−1)k+1 k+2 d k x−1 k+2 d k x−1
(x (x)−α (α)) = x ln(x)−0+ (x −1 (1))
(k + 1)! dxk dxk (k + 2)! dxk dxk
k=0 k=0
∞ ∞
X (−1)k+1 k+2 X 1
= x ln(x)+ (x (−1)k k!x−1−k −(−1)k k!) = x ln(x)− (x−1)
(k + 2)! (k + 1)(k + 2)
k=0 k=0
∞
X 1 1
= x ln(x) − ( − )(x − 1) = x ln(x) − x + 1
k +1 k+2
k=0

3
3 Generalization
The equation for double
R x integration can be derived from the Eq. 1.7.
Let’s suppose F(x) = 0 f (x)dx

(−1)k xk+1 d k
RxRx Rx
f (x)dxdx = 0 F(x)dx = ∞
P
0 0 k=0 (k+1)! dxk F(x) =
(−1)k xk+1 d k−1
Rx
= x 0 f (x)dx + ∞
P
k=1 (k+1)! dxk−1 f (x)
(−1)k xk+2 d k P∞ (−1)k+1 xk+2 d k
= ∞ (9)
P
k=0 (k+1)! dxk f (x) + k=0 (k+2)! dxk
f (x)
P∞ (−1)k (−1) d k
= k=0 ( (k+1)! (1 + (k+2) )xk+2 dxk f (x) =
(−1)k (k+1) k+2 d k
= ∞
P
k=0 ( (k+2)! x dxk
f (x)

Using the same technique, it is possible to obtain an equation for a larger

number of integrals applied in a row. In general, it is possible to derive a new
equation for n consecutive integrals applied.
∞
1 X (−1)k xk+n d k
f (−n) (x) = f (x) (10)
(n − 1)! k!(k + n) dxk
k=0

Eq. (10) can be proved by induction.

1.Base case we already get.
∞
X (−1)k xk+1 d k
f (−1) (x) = f (x)
k!(k + 1) dxk
k=0

2. Induction step Rx
Let’s suppose F(x) = 0 f (x)dx
k k+n F(x)xn k k+n
1 P∞ (−1) x dk 1 P∞ (−1) x dk
f (−n−1) (x) = F (−n) (x) = (n−1)! k=0 k!(k+n) dxk F(x) = n! + (n−1)! k=1 k!(k+n) dxk F(x) =
k k+1 k k+n P∞ (−1)k xk+n+1 d k
xn P ∞ (−1) x dk 1 P∞ (−1) x dk
n! k=0 (k+1)! dxk f (x) + (n−1)! k=1 k!(k+n) dxk F(x) = k=0 n!(k+1)! dxk f (x)+
k+1 k+n+1 (−1)k xk+n+1 1
1 P ∞ (−1) x dk dk
f (x) = ∞ −1
P
(n−1)! k=0 (k+1)!(k+n+1) dxk k=0 (k+1)!(n−1)! ( n + (n+k+1) ) dxk f (x) =
P∞ (−1)k xk+n+1 k+1 d k 1 P∞ (−1)k xk+(n+1) d k (−n−1) (x)
k=0 (k+1)!(n−1)! n(n+k+1) dxk f (x) = ((n+1)−1)! k=0 (k+1)!(k+(n+1)) dxk f (x) = f

Q.E.D.
Now, we can go into the space of fractional integration, since all the el-
ements adhere to the analytic continuation, and derive an expression of the
following form:
∞
α 1 X (−1)k xk+α d k
0 Ix f (x) = f (x) (11)
Γ(α) k!(k + α) dxk
k=0

Further, using the definition (2), we can obtain the fractional differential
equation.

4
Let’s supposen = ⌈α⌉ for α > 0

α dn n−α f (x) = d n 1 P∞ (−1)k xk+n−α d k

0 Dx f (x) = dx n 0 Ix dxn Γ(n−α) k=0 k!(k+n−α) dxk f (x) =
1 P∞ (−1)k dn k+n−α d k f (x)) =
= Γ(n−α) k=0 k!(k+n−α) dxn (x !dxk
1 P∞ (−1)k Pn n d n−b k+n−α d k+b
= Γ(n−α) k=0 k!(k+n−α) b=0 (x ) dxk+b f (x) =
b ! dxn−b
1 P∞ (−1)k Pn n Γ(k+n−α+1) k+b−α d k+b
= Γ(n−α) k=0 k!(k+n−α) b=0 b Γ(k+b−α+1)
x dxk+b
f (x) =
! (12)
1 Pn n P∞ (−1)k Γ(k+n−α) k+b−α d k+b
= Γ(n−α) b=0 b k=0 k! Γ(k+b−α+1) x dxk+b
f (x) =
! !
n P∞ k+n−α−1 d k+b
= nb=0 (−1)k 1
xk+b−α dx
P
k+b f (x) =
b ! k=0 ! k Γ(k+b−α+1)

n P∞ α−n xk+b−α d k+b

= nb=0
P
k=0 Γ(k+b−α+1) dxk+b
f (x)
b k

If we equalize equation (12) according to the degree of differentiation, we

get an equation similar to equation (11)
∞
xk−α dk
!
α
X α
0 Dx f (x) = f (x) (13)
k Γ(k − α + 1) dxk
k=0

This equation work for any complex α.

4 Operator scope
It is a complex task to investigate the convergence of this series. The basis
of convergence of the series should be expanded in comparison to the previ-
ous definition.
! Now it is worth defining a class of functions as f (x) ∈ C ∞ ∨
α f (n) (x)
lim Γ(n+1−α)
=0
n→∞ n
It is also important to introduce a restriction on the growth of the function.
Using Fabry’s theorem and Field’s theorem, we can limit the equation (13) to
the sum with gaps [4][5]. !
α f (n) (z)
Let’s suppose Nn = {ωn ∈ R|∃h ∈ R : Γ(n+1−α)
≤ hzωn }
n
So, if for every n Nn , ∅ and sequence min (Nn ) grows linearly, sum con-
verge.
The radius of convergence can be determined by standard methods after
function substitution.

5
5 Properties
Now it is possible to define the properties of the new equation. To describe
some properties, it is useful to define a falling factorial [6].
n
Y
(x)n = (x − (k − 1)) (14)
k=1

Derivative of the sum:

!
α P∞ α x k−α d k
0 Dx [f (x) + g(x)] = k=0 Γ(k−α+1) dxk
[f (x) + g(x)]
! k !
α xk−α dk α xk−α dk (15)
= ∞
P P∞
k=0 k f (x) + k=0 g(x)
k Γ(k−α+1) dx k Γ(k−α+1) dxk
= 0 Dxα f (x) + 0 Dxα g(x)

Derivative of the product:

!
α (f (x)g(x)) = P∞ α xk−α dk
D
0 x k=0 Γ(k−α+1) dxk
[f (x)g(x)]
! k !
α xk−α k dh d k−h
= ∞
P Pk
k=0 f (x) dx k−h g(x) =
k Γ(k−α+1) h=0 h dxh (16)
k
(α) xk+α Pk 1 dh d k−h
= ∞ 1
P
k=0 Γ(k−α+1) h=0 h! dxh f (x) (k−h)! dxk−h g(x) =
Γ(α+1) P ∞ sin(πα−πk)xk+α P k 1 dh 1 d k−h
= π k=0 (α−k) h=0 h! dxh f (x) (k−h)! dxk−h g(x)

The sum of differentials of different degrees of one function

β
a0 Dxα f (x) + b!0 Dx f (x) !
P∞ α xk−α dk P∞ β xk−β dk
= k=0 f (x) + k=0 f (x) = (17)
k Γ(k−α+1) dxk k Γ(k−β+1) dxk
k k
xk (α) xα (β) xβ dk
= ∞
P
k=0 k! ( Γ(k−α+1) + Γ(k−β+1) ) dxk f (x)

6 Application
The new equation possesses several distinguishing features that make it practi-
cal for application compared to equations (1), (2) and (3), (4). Firstly, its struc-
ture is similar to the Taylor series, leading to a decrease of each subsequent
element decreases inversely proportionally to the factorial during numerical
calculation. Secondly, it does not require special algorithms for extending into
a set of complex numbers. By utilizing equation (13), it is possible to devise an
algorithm for approximate calculation. The first version of the algorithm will
be presented in the Wolfram Mathematica language.
FractDiff[pow , f , t , noo ] := Sum[Binomial[pow, k] ∗
t ∧ (k − pow)/Gamma[k + 1 − pow] ∗ D[f , {t, k}], {k, 0, noo}]

6
 The function takes 4 arguments as input:
1. pow (power) - Power of the differentiation operator.
2. f (function) - The function to which the operator is applied.
3. t - The variable by which calculations are performed.
4. noo (Number of Operations) - The number of operations that determine the
accuracy of calculations.
This version will be useful for scientific calculations. However, for use in
large projects, a C++ implementation may be more practical. Equation (4) will
be used to calculate the integer power of the derivative of function in C++.
By combining these approaches, though with slightly increased time costs, the
accuracy of calculations increases greatly [7].
We will need auxiliary functions that are not contained in standard li-
braries:
1. Binomiald - a function that calculates the binomial coefficient, capable of
calculating non-integer values
double binomiald(double k, int a) noexcept
{
double b = 1.0;
double k2 = k + 1.0;
double p = 1.0;
for (int c = 1; c <= a; ++c) {
b *= k2 / p - 1.0;
p += 1.0;
}
return b;
}

2. Grunwald-Letnikov derivative
double GL righthand derivative(std::function<double(double)> f, dou-
ble pow, double point, unsigned long N) noexcept
{
const double h = point / N;
double sign = 1.0;
double sum = 0;
const double correction = pow * h / 2.0;
double pointshift = point - correction - h;
for (size t k = 0; k < N; k++)//gofor(k, range(0, N))
{
sum += sign * binomiald(pow, k) * f(pointshift += h);
sign *= -1.0;
}
return sum * std::pow(-1.0/h,pow);
}

3. Approximate calculation of the ”pow” order differential of function

7
 double N derivative(std::function<double(double)> f, double pow, dou-
ble point, unsigned long N) noexcept
{
double sum = 0;
for (size t k = 0; k < N; k++)//gofor(k, range(0, N))
{
sum += binomiald(pow, k) / std::exp(std::lgamma(k - pow + 1)) *
std::pow(point, k - pow) * GL righthand derivative(f, k, point, N);
}
return sum;
}
Example 4.
For example, investigate the transformationD 1/3 (x ∧ 2). The exact answer
Γ(3)
will be Γ(8/3) x5/3
Table 1 - Comparison of the accuracy of fractional derivative calculation
algorithms
Point Exact answer GL with 9 iteration Eq (13) with 3 itera-
tion
0.2 0.0909213773601394 0.0909397004356575 0.0909213773601394
0.4 0.288657380135766 0.2887155522744711 0.288657380135766
0.6 0.5673722586492711 0.5674865992481914 0.5673722586492709
0.8 0.9164300577717989 0.9166147428001502 0.9164300577717989
1.0 1.329278600918967 1.329546485908133 1.329278600918967
1.2 1.801294640474853 1.801657649248871 1.801294640474852
1.4 2.328963807359781 2.329433155503858 2.328963807359781
1.6 2.909484075524463 2.910070413941038 2.909484075524462
1.8 3.540545570594309 3.541259084684707 3.540545570594309
Approximation of the solution of a specific differential equation.

dk
D α u(x) + bxk−α u(x) = g(x) (18)
dxk
Equation (18) can be solved by using n first members of the sequence (13)
n
xk−α dk k
!
α k−α d
X
u(x) + bx u(x) = g(x) (19)
k Γ(k − α + 1) dxk dxk
k=0

The resulting differential equation can be solved using standard methods.

The resulting answer will contain an approximate solution and a set of func-
tions that depend on constants. The solution will be closest when the constants
are defined as zeros.
Example 5.
Consider the differential equation.
d ln(4x) 1
D 0.5 u(x) + x0.5 u(x) = √ + √
dx πx x

8
 First, replace the fractional derivative with the first 3 elements of the sum
of the equation (13).
2
xk−0.5 d k
!
X 0.5 d ln(4x) 1
u(x) + x0.5 u(x) = √ + √
k Γ(k + 0.5) dx k dx πx x
k=0

Solve the equation with the help of software, for example, wolframalpha.
c1
u(x) = 0.170874 + c2 x17.9686 + + ln(x)
x0.333915
The exact answer will be u(x) = ln(x) which means that we have gotten
a solution that differs only by a constant when zeroing the constants in the
answer.

7 Discussion
This work creates a number of issues that require future consideration. One of
the interesting questions arises because
! of the high similarity of equality (13)
α P∞ α k−α
in the form of D f (x) = k=0 I (1)D k f (x) with the Newton binomial
k
formula. Other formats for applying equality (13) can also be considered [8].

9
References
1. Zenyuk D.A., Orlov Yu.N. About application fractional Riemann-Liouville
calculus for describing probability distributions // IPM preprints im.
M.V. Keldysh. 2014. No. 18. 21 p. URL: http://library.keldysh.ru/preprint.asp?id=2014-18
2. Edmundo Capelas de Oliveira, José António Tenreiro Machado, ”A Re-
view of Definitions for Fractional Derivatives and Integral”, Mathemati-
cal Problems in Engineering, vol. 2014, Article ID 238459, 6 pages, 2014.
https://doi.org/10.1155/2014/238459
3. Chii-Huei Yu, ” Study of Fractional Analytic Functions and Local Frac-
tional Calculus, International Journal of Scientific Research in Science,
Engineering and Technology(IJSRSET), Print ISSN : 2395-1990, Online
ISSN : 2394-4099, Volume 8, Issue 5, pp.39-46, September-October-2021.
Available at doi : https://doi.org/10.32628/IJSRSET218482
4. Madan, S. Some G Pólya gems from complex analysis. Reson 19, 323–337
(2014). https://doi.org/10.1007/s12045-014-0038-6
5. Puchta, Jan-Christoph. ”On Fabry’s gap theorem.” Archivum Mathe-
maticum 038.4 (2002): 307-309. <http://eudml.org/doc/248932>.
6. Osgood, Brad & Wu, William. (2008). Falling Factorials, Generating
Functions, and Conjoint Ranking Tables. 12.
7. Persechino, André. (2020). An introduction to fractional calculus Nu-
merical methods and application to HF dielectric response. Smart Mate-
rials and Structures. 9. 20. 10.7716/aem.v9i1.1192.
8. Ghorbani, H., Mahmoudi, Y., Saei, F.D. (2020). Numerical study of frac-
tional Mathieu differential equation using radial basis functions. Mathe-
matical Modelling of Engineering Problems, Vol. 7, No. 4, pp. 568-576.
https://doi.org/10.18280/mmep.070409

10