October 15, 2015 12:18 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 1

Chapter 1

Discrete-variable real functions

by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.
Discrete Fractional Calculus Downloaded from www.worldscientific.com

In this chapter a notion of one discrete-variable real-valued functions is introduced.

Fundamental operations are defined. A great emphasis is put on the discrete con-
volution of two functions [Ostalczyk (2001); O’Flynn and Moriarty (1987); Proakis
(2007); Weeks (2011)]. Two special discrete-variable functions denoted as a(ν)(k)
and c(ν)(k) are defined and discussed. The mentioned functions play an important
role in fractional-order differences and sums. The link between two discrete-variable
function and a digital image finishes this chapter.

1.1 Notation

Firstly one defines some sets:

R – the set of real numbers,

R+ = [0 + ∞) – set of real non-negative numbers,
R− = [0 + ∞) – set of real negative numbers,
Rt0 = [t0 , +∞) – subset of real nonnegative numbers,
Rt0 ,t = [t0 , t] – finite interval,
C – set of complex numbers,
Z – set of integers,
ZN1 ,N2 = [N1 , N2 ] – finite set of integers,
Zk0 = [k0 , ∞) – subset of integers greater or equal to k0 > 0,
Z+ = Z0 – set of non-negative integers,
Zk0 ,k = [k0 , k] – finite set of integers greater or equal to k0 > 0 and less or equal
to k,
Tk0 ,k,h = [k0 h, (k0 + 1)h, (k0 + 2) . . . , kh] – finite set of equally spaced real numbers
ih, i = k0 , k0 + 1, . . . , k0 < k. (T0,k,1 = Z+ ),
Q+ – set of rational numbers,

Re{.} – real part of a complex number,

Im{.} – imaginary part of a complex number,
z̄ – complex conjugate to z,

1
September 7, 2015 12:4 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 2

2 Discrete Fractional Calculus

h > 0 – constant (sampling period),

t – continuous time,
kh – discrete time (k for h = 1),

Ak – real (k + 1) × (k + 1) matrix,
1k – real (k + 1) × (k + 1) unit matrix,
0k – real (k + 1) × (k + 1) zero matrix,
ak , bk , ck – real (k + 1) × 1 column vectors,

T – matrix (vector) transposition,

by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

* – discrete convolution operation,

tr – trase of the matrix,
Discrete Fractional Calculus Downloaded from www.worldscientific.com

|.| – absolute value,

[rad] – radian,
[dB] – decibel,
[dec] – decade,
[V ] – volt,
[s] – second.

1.2 Class of one discrete-variable functions

Different signals encountered in science and engineering are described by real-valued

functions of a single independent discrete real variable (discrete time instants, dis-
crete distance), k ∈ Tk0 ,k,h is denoted as
f : Tk0 ,k,h → R (1.1)
where Tk0 ,k,h is the function domain and R its image (codomain). There are several
alternative representations of the discrete-variable functions. The most popular is
the so-called functional representation. Below an example is given.
f (kh) = eah(k−k0 ) sin[bh(k − k0 )] (1.2)
where h > 0 is the so-called sampling period (the time, or space between successive
samples) and a, b ∈ R are function constants. The graphical representation of the
function (1.2) for h = 0.1, k0 = 7 and a = −0.03, b = 0.125π is presented in Fig. 1.1.
In many practical applications the discrete-variable functions are represented by an
ordered set of real numbers. For every number there is a uniquely assigned corre-
sponding index. Such a tabular representation collects usually a set of measured
data. In Table 1.1 the function is tabularized. Very often the initial independent
variable is set k0 = 0. Such an assumption does not reduce generality of the con-
siderations. Moreover, for known and constant sampling time h one can omit it
considering function (1.2)
0
f (k) = ea (k−k0 ) sin[b0 (k − k0 )] (1.3)
October 15, 2015 12:18 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 3

Discrete-variable real functions 3

0.8

0.6

0.4

0.2
f(kh)

−0.2
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

−0.4

−0.6
Discrete Fractional Calculus Downloaded from www.worldscientific.com

−0.8
0 1 2 3 4 5
kh

Fig. 1.1 Graphical representation of a function f (k).

Table 1.1 Tabular representation

of the discrete-variable function.
k f (k)
0 0
1 3.713734277017140e-001
2 6.659280879784083e-001
3 8.443623161958684e-001
4 8.869204367171575e-001
5 7.951904828883181e-001
6 5.906252306120317e-001
7 3.101971614689395e-001
8 9.633412921752834e-017
9 -2.921326850899535e-001
10 -5.238375874705950e-001

with a0 = ah and b0 = bh. The plot of (1.3) evaluated over interval [0, k] (k0 = 0) is
presented in Fig. 1.2. Notice that this time a horizontal scale is changed.
In many technical applications the function range is also quantized. Here one
can mention a 2D digital signal (image) where Tgmin ,gmax ,hg , i.e.

g : Tk0 ,k,h → Tgmin ,gmax ,hg . (1.4)

The function (1.4) defined by formula (1.3) is plotted in Fig. 1.3 for hg = 0.2.
Here one should mention that the immense progress in electronics causes that h
in modern devices rapidly decreases achieving in 16bit A/C and C/A converters
values hg = 2−16 = 1.52587890625e − 005[V ]. Applying this value for h one gets
September 4, 2015 16:22 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 4

4 Discrete Fractional Calculus

0.8

0.6

0.4

0.2
f(k)

−0.2
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

−0.4

−0.6
Discrete Fractional Calculus Downloaded from www.worldscientific.com

−0.8
0 10 20 30 40 50
k

Fig. 1.2 Graphical representation of a function (1.3).

0.8

0.6

0.4

0.2
f(k)

−0.2

−0.4

−0.6

−0.8

−1
0 10 20 30 40 50
k

Fig. 1.3 Graphical representation of function (1.4).

the plot presented in Fig. 1.4. The difference between the plots 1.2 and 1.4 is
almost negligible. The difference values are really very small. These are indicated
in Fig. 1.5.
The discrete-variable function may be also represented by a discrete image char-
acterized by the function value f (k) ∈ Rymin ,ymax , whose codomain is represented
by an appropriate element g(k1 ) ∈ Rgmin ,gmax where Rgmin ,gmax is a set collecting
the image gray levels. Hence, a one dimensional image (vector of pixels) described
October 15, 2015 12:18 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 5

Discrete-variable real functions 5

1 1

0.8 0.8

0.6
0.6

0.4
0.4
0.2
0.2

f(kh)
g(k)

0
0
−0.2
−0.2
−0.4

−0.4
−0.6

−0.8 −0.6

−1 −0.8
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

0 10 20 30 40 50 0 1 2 3 4 5
k kh
Discrete Fractional Calculus Downloaded from www.worldscientific.com

Fig. 1.4 Plot of the function (1.4). Fig. 1.5 Plot of a difference between func-
tions f (k) (1.3) and g(k) (1.4).

shortly as 1D image is described by a concatenation of two functions

f : Tk0 ,k,h → Tfmin ,fmax
(1.5)
g : Tfmin ,fmax → Tgmin ,gmax .
Images related to the discrete-variable functions (1.3) and (1.4) are given in
Figs. 1.6 and 1.7, respectively. In fact, for a better visibility of the vectors of pixels
every vector pixel is magnified to a square box containing l × l identical pixels.

5 5

4.5 4.5

4 4

3.5 3.5
l[px]

l[px]

3 3

2.5 2.5

2 2

1.5 1.5

1 1
50 100 150 200 250 50 100 150 200 250
kl[px] kl[px]

Fig. 1.6 Image of the function (1.3). Fig. 1.7 1D image of the function (1.4).

1.3 Selected discrete-variable functions

1.3.1 Fundamental discrete-variable functions

The study of discrete-variable functions starts with the presentation of a number
of basic functions that appear very often in analysis. These functions are defined
below.
September 4, 2015 11:36 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 6

6 Discrete Fractional Calculus

Definition 1.1 (Discrete Dirac pulse). The discrete Dirac pulse is defined as

1 for k = 0
δ(k) = . (1.6)
0 for k 6= 0

1.2 2

1.9
1
1.8

0.8 1.7
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

1.6
0.6

1[px]
δ(k)

1.5
0.4
Discrete Fractional Calculus Downloaded from www.worldscientific.com

1.4

0.2 1.3

1.2
0
1.1

−0.2 1
−10 −5 0 5 10 15 20 25 2 4 6 8 10 12 14 16 18 20
k k[px]
(a) (b)

Fig. 1.8 Plot (a) and 1D image (b) of the discrete Dirac pulse (1.6).

The next function of a great importance in the discrete fractional calculus is the
so-called unit step function. Its definition is given below, and its plot and image
are presented in Figs. 1.9(a) and 1.9(b), respectively.

Definition 1.2 (Discrete unit step function). The discrete unit step function
is defined as

0 for k < 0
1(k) = . (1.7)
1 for k > 0

1.2 2

1.9
1
1.8

0.8 1.7

1.6
0.6
1[px]
1(k)

1.5
0.4
1.4

0.2 1.3

1.2
0
1.1

−0.2 1
−10 −5 0 5 10 15 20 25 2 4 6 8 10 12 14 16 18 20
k k[px]
(a) (b)

Fig. 1.9 Plot (a) and 1D image (b) of the discrete unit step function (1.7).
October 15, 2015 12:18 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 7

Discrete-variable real functions 7

1.3.2 Elementary discrete-variable functions

Definition 1.3. The general exponential function

f (k) = abk 1(k) where a, b, ∈ R. (1.8)

In linear discrete-time systems, an important role is played by the exponential

function, which is a special case of the general exponential function (1.8) where
a = e (e = 2.718281828459046 is the Euler number). The plot of the exponential
function is presented in Fig. 1.10.
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

1.2 2

1.9
Discrete Fractional Calculus Downloaded from www.worldscientific.com

1
1.8

0.8 1.7

1.6
eak 1(k)

0.6
1.5
0.4
1.4

0.2 1.3

1.2
0
1.1

−0.2 1
−10 −5 0 5 10 15 20 25 30 5 10 15 20 25
k k[px]
(a) (b)

Fig. 1.10 Plot (a) and 1D image (b) of the exponential function evaluated for b = −0.05.

On the basis of the exponential function one defines discrete-variable trigono-

metric functions: sin and cos. Below their definitions are provided.

Definition 1.4. The discrete sin function

f (k) = sin(ak + b)1(k) where a, b ∈ R. (1.9)

A typical shape of the discrete sin function is given in Fig. 1.11(a). Its image
is given beside in Fig. 1.11(b). Here it is worth noting that in a discrete-variable
function appropriate choice of its parameters may essentially change its plot. Such
a situation occurs when in the sin function one puts a = π2 or a = π2 . Appropriate
plots and related 1D images are presented in Figs. 1.12(a) and Fig. 1.12(b) as well as
Figs. 1.13(a) and Fig. 1.13(b), respectively. It must be admitted that the function
graphs do not resemble the graphs of the sin function.
Even more surprising is the assumption that a = 2π, b = π2 or a = 2π, b = − π2 .
For such parameters one gets sin(2πk + 0.5π) = 1(k) and sin(2πk − 0.5π) = −1(k).
The discrete cos function may be derived from formula (1.9) by the assumption
that b = π2 . The discrete ramp function is defined as follows.

Definition 1.5. The discrete ramp function defines a formula

f (k) = k1(k). (1.10)
September 21, 2015 11:43 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 8

8 Discrete Fractional Calculus

1 5

0.8
4.5
0.6
4
0.4
3.5
0.2

l[px]
f(k)

0 3

−0.2
2.5
−0.4
2
−0.6
1.5
−0.8

−1 1
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

−5 0 5 10 15 20 25 30 20 40 60 80 100 120
k kl[px]

(a) (b)
Discrete Fractional Calculus Downloaded from www.worldscientific.com

π
Fig. 1.11 Plot (a) and 1D image (b) of the sin function evaluated for a = 2
,b = 0.

1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2
f(k)

f(k)

0 0

−0.2 −0.2

−0.4 −0.4

−0.6 −0.6

−0.8 −0.8

−1 −1
−5 0 5 10 15 20 25 30 −5 0 5 10 15 20 25 30
k k
(a) (b)
π
Fig. 1.12 Plot of the discrete sin function evaluated for special parameters. (a) a = 2
,b = 0.
(b) a = π3 , b = 0.

This function is a special case of so-called discrete-variable power functions

f (k) = k s 1(k) where s ∈ Z+ . Realize that for s = 0 it is the unit step function.
The plot and image of the ramp function are given in Figs. 1.14(a) and 1.14(b).

1.3.3 Function a(ν) (k)

Now one defines a discrete-variable function, which plays a crucial role in discrete
fractional calculus. This discrete-variable function is characterized by one parameter
denoted by ν. The function may be expressed by several equivalent formulas.

Definition 1.6. The discrete function a(ν) (k) is defined as



 0 for k < 0
(ν)
a (k) = 1 for k = 0 . (1.11)

 (−1)k ν(ν−1)···(ν−k+1)
k! for k = 1, 2, 3, . . .
October 15, 2015 12:18 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 9

Discrete-variable real functions 9

5 5

4.5 4.5

4 4

3.5 3.5
l[px]

l[px]
3 3

2.5 2.5

2 2

1.5 1.5

1 1
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

20 40 60 80 100 120 20 40 60 80 100 120

kl[px] kl[px]

(a) (b)
Discrete Fractional Calculus Downloaded from www.worldscientific.com

π
Fig. 1.13 1D image of the discrete sin function evaluated for special parameters. (a) a = 2
,b = 0.
(b) a = π3 , b = 0.

30 5

25 4.5

4
20

3.5
15
l[px]
f(k)

3
10
2.5

5
2

0 1.5

−5 1
−5 0 5 10 15 20 25 30 0 20 40 60 80 100 120 140
k kl[px]

(a) (b)

Fig. 1.14 Plot (a) and 1D image (b) of the discrete ramp function.

By direct calculations one may verify that the function a(ν) (k) may be evaluated in
a recursive way

(ν) 1 for k = 0
a (k) = . (1.12)
(1 − ν+1
k )a (ν)
(k − 1) for k = 1, 2, 3, . . .
This can be verified by direct calculation
ν(ν − 1) · · · (ν − k + 1)
a(ν) (k) = (−1)k
k!
ν(ν − 1) · · · (ν − k + 2) (ν − k + 1)
×(−1)k−1 (−1)
(k − 1)! k
 
(ν − k + 1) ν+1
= a(ν) (k − 1)(−1) = a(ν) (k − 1) 1 − .
k k
September 7, 2015 12:4 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 10

10 Discrete Fractional Calculus

where a(ν) (k−1) is the same function as (1.11) but subjected to a forward horizontal
shift operation. This operation is examined more fully in Subsection 1.4.2. The next
equivalent form of the considered function introduces the following theorem.

Theorem 1.1. For ν ∈ R,

Γ(k − ν)
a(ν) (k) = . (1.13)
Γ(−ν)Γ(k + 1)
Proof. By fundamental property of the Euler Gamma function defined by formula
(A.3) (see Appendix A) for z = −ν and n = k
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

Γ(−ν + k)
Γ(−ν) = . (1.14)
Discrete Fractional Calculus Downloaded from www.worldscientific.com

(−ν)(−ν + 1) · · · (−ν + k − 1)

After a simple rearrangements one gets

Γ(−ν + k)
(−1)k (ν)(ν − 1) · · · (ν − k + 1) = . (1.15)
Γ(−ν)
Moreover, by (A.3)
Γ(k + 1) = k!. (1.16)
A substitution of (1.15) and (1.16) into (1.11) yields (1.12).
One should also note that function a(ν) (k) values appear in the expansion of the
function (1 − x)ν into Maclaurin infinite series
+∞
X
(1 − x)ν = aν (i)xi for all |x| 6 1. (1.17)
i=0

The plot and 1D image of the function a(ν) (k) are given in Figs. 1.15(a) and
1.15(b). In order to fully present the nature of the analyzed function a(ν) (k) in
Figs. 1.16(a) and 1.16(b) plots of a(ν) (k) related to different parameter ν values
from a set {0.10.2 · · · 0.9} are given.
In Figs. 1.15(a) and 1.15(b) the plots of function a(ν) (k) evaluated for coefficients
ν from a set {0.1, 0.2, . . . , 1.0} and two different ranges k = 0, 1, 2, . . . , 9, 10 and
k = 2, . . . , 9, 10 are presented, respectively. Every plot is uniquely identified by its
value related to k = 1, where a(ν) (1) = −ν. For better clarity of plots, its distinct
points are connected by straight line segments. The second plot is a fragment of
the first one. It shows that the individual segments can be crossed. In the next
four figures, the plots of the a(ν) (k) function, for further parameter ν ranges are
shown. The plots reveal that function a(ν) (k) sharply tends to zero for all k > ν + 1.
This property will be discussed in details in the next subsection. The plots 1.16
and 1.17 also contain integer values of parameter ν. The following Figs. 1.18(a) and
1.18(b) present the function a(ν) (k) plots evaluated only for integer parameters
ν ∈ {1, 2, . . . , 9} (a) and {10, 11, . . . , 19} (b), respectively.
October 15, 2015 12:18 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 11

Discrete-variable real functions 11

1.2 2

1 1.9

1.8
0.8

1.7
0.6
1.6
a(0.5) (k)

0.4

1[px]
1.5
0.2
1.4
0
1.3

−0.2
1.2

−0.4 1.1

−0.6 1
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

−10 −5 0 5 10 15 20 25 2 4 6 8 10 12 14 16 18 20
k k[px]
(a) (b)
Discrete Fractional Calculus Downloaded from www.worldscientific.com

Fig. 1.15 Plot (a) and an image (b) of the function a(ν) (k).

1 0

0.8
−0.02
0.6

0.4 −0.04

0.2
−0.06
a(ν) (k)
a(ν) (k)

0
−0.08
−0.2

−0.4 −0.1

−0.6
−0.12
−0.8

−1 −0.14
0 2 4 6 8 10 2 3 4 5 6 7 8 9 10
k k
(a) (b)

Fig. 1.16 Plots of the functions a(ν) (k) for ν ∈ {0.1, 0.2, . . . , 1.0} and two ranges: k ∈
{0, 1, . . . , 10} and k ∈ {2, 3, . . . , 10}.

Now, the plots of function a(ν) (k) related to negative parameter ν values are
given. Similarly to the positive values of parameter ν four figures related to four
negative parameter ranges are presented.
In contrast to the previous plots, which tend to zero for growing k all plots
related to negative parameter ν values increase due to lowering ν. Properties of
this important function are discussed in the next sub-section. In the following two
examples one compares the results of the function a(ν) (k) evaluation due to three
equivalent formulas (1.11)–(1.13).

Example 1.1. For ν = 0.5 one calculates the function a(ν) (k) values according
to formulas (1.11)–(1.13). In order to distinguish between the results calculated
according to three mathematically equivalent formulas, the function a(ν) (k) will be
(ν) (ν) (ν)
temporarily denoted as a11 (k), a12 (k) and a13 (k), respectively. Error functions
September 4, 2015 11:36 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 12

12 Discrete Fractional Calculus

1 3

0.5 2
a(ν) (k), ν ∈ {1.1, 1.2, · · · , 2}

a(ν) (k), ν ∈ {2.1, 2.2, · · · , 3}

0 1

−0.5 0

−1 −1

−1.5 −2

−2 −3
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

0 1 2 3 4 5 0 1 2 3 4 5
k k
(a) (b)
Discrete Fractional Calculus Downloaded from www.worldscientific.com

Fig. 1.17 Plots of the function (a) a(ν) (k) for ν ∈ {1.1, 1.2, . . . , 2.0} and (b) {2.1, 2.2, . . . , 3.0}.

6 10

5 8
a(ν) (k), ν ∈ {3.1, 3.2, · · · , 4}

a(ν) (k), ν ∈ {4.1, 4.2, · · · , 5}

4 6

3 4

2 2

1 0

0 −2

−1 −4

−2 −6

−3 −8

−4 −10
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
k k
(a) (b)

Fig. 1.18 Plots of the function (a) a(ν) (k) for ν ∈ {3.1, 3.2, . . . , 4.0} and (b) {4.1, 4.2, . . . , 5.0}.

defined below will serve as tools to three considered functions shapes comparison

(ν) (ν) (ν)

e11,12 (k) = |a11 (k) − a12 (k)|
(ν) (ν) (ν)
e11,13 (k) = |a11 (k) − a13 (k)| (1.18)
(ν) (ν) (ν)
e12,13 (k) = |a12 (k) − a13 (k)|.

In Figs. 1.22–1.24 three pairs of the error functions evaluated for ν = 0.5 and
ν = −0.5 are given, respectively.
(ν) (ν)
The error plots reveal very similar results of e11,13 (k) and e12,13 (k) for posi-
tive and negative parameter ν. For negative value of ν, errors are two orders of
magnitude greater then for positive values. The evaluation of a(ν) (k) by formula
October 15, 2015 12:18 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 13

Discrete-variable real functions 13

5
x 10
150 1

0.8
100

a(ν) (k), ν ∈ {11, 12, · · · , 20}

0.6
a(ν) (k), ν ∈ {1, 2, · · · , 9}

0.4
50
0.2

0 0

−0.2
−50
−0.4

−0.6
−100
−0.8
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

−150 −1
0 2 4 6 8 10 12 0 5 10 15 20
k k
(a) (b)
Discrete Fractional Calculus Downloaded from www.worldscientific.com

Fig. 1.19 Plots of the functions (a) a(ν) (k) for ν ∈ {1, 2, . . . , 9} and (b) {10, 11, . . . , 20}.

1.2 12
a(ν) (k), ν ∈ {−1.1, −1.2, · · · , −2.0}
a(ν) (k), ν ∈ {−0.1, −0.2, · · · , −1.0}

1 10

0.8 8

0.6 6

0.4 4

0.2 2

0 0

−0.2 −2
0 2 4 6 8 10 0 2 4 6 8 10
k k
(a) (b)

Fig. 1.20 Plots of the functions (a) a(ν) (k) for ν ∈ {−0.1, −0.2, . . . , −1.0} and (b)
{−1.1, −1.2, . . . , −2.0}.

1.13 disqualifies an explosion’ of factors in this formula. This is caused by rapidly

increasing values of gamma functions. (In the Matlab R
program one can evaluate
(ν)
a13 (k) only for k < 170).

1.3.3.1 Function a(ν) (k) properties

Next some properties of the discrete-variable function a(ν) (k)indexfunction a(ν) (k)
with one parameter ν will be analyzed. They are formulated in the form of a few
theorems. The most important range of parameter ν are (0, 1) and (−1, 0), thus the
analysis will mainly concentrate on the mentioned ranges. Here one assumes that

0 < ν < 1. (1.19)

September 4, 2015 11:36 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 14

14 Discrete Fractional Calculus

70 300
a(ν) (k), ν ∈ {−2.1, −2.2, · · · , −3.0}

a(ν) (k), ν ∈ {−3.1, −3.2, · · · , −4.0}

60 250

50
200

40
150
30
100
20

50
10

0 0
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

−10 −50
0 2 4 6 8 10 0 2 4 6 8 10
k k
(a) (b)
Discrete Fractional Calculus Downloaded from www.worldscientific.com

Fig. 1.21 Plots of the functions (a) a(ν) (k) for ν ∈ {−2.1, −2.2, . . . , −3.0} and (b)
{−3.1, −3.2, . . . , −4.0}.

−19 −16
x 10 x 10
4.5 1

4 0.9

0.8
3.5
0.7
3
e11,12 (k)

0.6
e11,12(k)

2.5
(−0.5)

0.5
(0.5)

2
0.4
1.5
0.3

1
0.2

0.5 0.1

0 0
0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 180
k k

(a) (b)
(ν)
Fig. 1.22 Plots of the error functions (a) e11,12 (k) for ν = 0.5 and (b) ν = −0.5.

Theorem 1.2. For i = 1, 2, . . . and 0 < ν < 1, the following inequalities hold

a(ν) (k) < 0

(1.20)
a(−ν) (k) > 0.

Proof. To prove inequalities 1.20 one uses a recurrent formula (1.12)

 
±ν + 1
a(±ν) (k) = 1 − a(±ν) (k) for k = 1, 2, 3, . . . (1.21)
k
with
a(±ν) (1) = ∓ν. (1.22)
October 15, 2015 12:18 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 15

Discrete-variable real functions 15

−16 −14
x 10 x 10
1.2 1.2

1 1

0.8 0.8

e11,13 (k)
e11,13(k)

(−0.5)
0.6 0.6
(0.5)

0.4 0.4

0.2 0.2
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

0 0
0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 180
k k

(a) (b)
Discrete Fractional Calculus Downloaded from www.worldscientific.com

(ν)
Fig. 1.23 Plots of the error functions (a) e11,13 (k) for ν = 0.5 and (b) ν = −0.5.

−16 −14
x 10 x 10
1.2 1.2

1 1

0.8 0.8
e12,13 (k)
e12,13(k)

(−0.5)

0.6 0.6
(0.5)

0.4 0.4

0.2 0.2

0 0
0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 180
k k

(a) (b)
(ν)
Fig. 1.24 Plots of the error functions (a) e12,13 (k) for ν = 0.5 and (b) ν = −0.5.

Now one assumes that a(ν) (k − 1) < 0 and a(−ν) (k − 1) > 0 for k − 1 > 0. One may
then verify that
 
ν+1 a(ν) (k)
0< 1− = (ν) for k = 2, 3, . . . (1.23)
k a (k − 1)
 
−ν + 1 a(−ν) (k)
0> 1− = (−ν) for k = 2, 3, . . . (1.24)
k a (k − 1)
because
   
−ν + 1 ν+1
1− > 1− >0 for k = 2, 3, . . . . (1.25)
k k
From (1.23) and (1.24) one deduces (1.20).
September 4, 2015 11:36 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 16

16 Discrete Fractional Calculus

A slightly different proof may be found in [Kaczorek (2011)].

Theorem 1.3. For 0 < ν < 2

lim a(ν) (k) = a(ν) (+∞) = 0. (1.26)

k→+∞

Proof. For the discrete-variable function a(ν) (k), consecutive values may be treated
as a sequence of real numbers whose limit can be evaluated by the Cauchy formula
theorem. A finite limit exists if and only if for every  > 0 there is an L such that
|a(ν) (k1 ) − a(ν) (k2 )| <  for every k1 , k2 > L. In fact for any  > 0 one can find L
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

such that
Discrete Fractional Calculus Downloaded from www.worldscientific.com

|a(ν) (L)| < . (1.27)

Subsequently, any k1 , k2 > L can be expressed as k1 = L + L1 , k2 = L + L1 + L2

where L, L1 , L − 2 ∈ Z and

|a(ν) (k1 ) − a(ν) (k2 )| = |a(ν) (L + L1 ) − a(ν) (L + L1 + L2 )|. (1.28)

By definition from formula 1.12

 
 L2
Y ν − L − L − i + 1 
 1 
|a(ν) (L + L1 + L2 )| = a(ν) (L + L1 )  (1.29)
 L + L1 + i 
i=1

 
 L1
Y ν − L − i + 1 
 
|a(ν) (L + L1 )| = a(ν) (L) . (1.30)
 L+i 
i=1

A substitution of (1.29) and (1.30) into (1.28) yields

|a(ν) (L + L1 ) − a(ν) (L + L1 + L2 )|
 
 L2
Y ν − L − L1 − i + 1 
 (ν) (ν)
= a (L + L1 ) − a (L + L1 ) 
 L + L1 + i 
i=1
 !
 L1
Y L2
ν−L−i+1 Y ν − L − L1 − i + 1 
 (ν)
= a (L) 1 − 
 L+i L + L1 + i 
i=1 i=1
 
 L1
Y L2
ν−L−i+1 Y ν − L − L1 − i + 1 
(ν) 
= |a (L)| 1 − . (1.31)
 L+i L + L1 + i 
i=1 i=1

Now one examines a second factor in the above expression. One can see that for
k = 2, 3, . . .
 
ν − k + 1
  = 1 − ν + 1 < 1. (1.32)
 k  k
October 15, 2015 12:18 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 17

Discrete-variable real functions 17

Hence
 
 L1
Y ν − L − i + 1
L2
Y ν − L − L − i + 1 
 1 
|a(ν) (L)| 1 −  < |a(ν) (L)| <  (1.33)
 L+i L + L1 + i 
i=1 i=1

and

|a(ν) (k1 ) − a(ν) (k2 )| < . (1.34)

The last inequality is also valid for a(ν) (k2 ) = a(ν) (+∞) = 0. This ends the proof.
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

The next function a(ν) (k) property is stated by the following theorem.
Discrete Fractional Calculus Downloaded from www.worldscientific.com

Theorem 1.4. For the coefficient ν satisfying (1.19) a following equality holds
+∞
X
a(ν) (i) = 0. (1.35)
i=0

Proof. To prove equality (1.35) one refers to equality (1.17). For x = 1, one
immediately obtains the result
+∞
X +∞
X

(1 − x)ν x=1 = (1 − 1)ν = 0 = aν (i)1i = aν (i). (1.36)
i=0 i=0

As a consequence of Theorem 1.4 and the function a(ν) (k) definition formula (1.11)
one has
+∞
X
1=− a(ν) (i). (1.37)
i=1

Moreover, for (1.19) by (1.12) and (1.22)

a(ν) (i) ν+1

=1− >0 (1.38)
a(ν) (i− 1) i
which means that for k = 2, 3, . . .

a(ν) (i − 1) < a(ν) (i) < 0. (1.39)

Hence, (1.37) can be rewritten to the form

+∞
X
1= |a(ν) (i)|. (1.40)
i=1
September 4, 2015 11:36 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 18

18 Discrete Fractional Calculus

Inspection of formula (1.38) reveals that a(ν) (k) → 0 for k → +∞. This result
is of practical use in real-time microprocessor calculation of the function a(ν) (k)
values. For any L > 1 formula (1.40) can be rewritten in the form
L
X +∞
X
1= |a(ν) (i)| + |a(ν) (i)| (1.41)
i=1 i=L+1

with
+∞
X
lim |a(ν) (i)| → 0. (1.42)
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

L→+∞
i=L+1
Discrete Fractional Calculus Downloaded from www.worldscientific.com

(ν)
Hence in practical applications one may define two parameter function aL (k).
(ν)
Definition 1.7. The discrete function aL (k) is defined as

 0 for k <0


(ν) 1 for k =0
aL (k) = k ν(ν−1)···(ν−k+1) . (1.43)

 (−1) for k = 1, 2, . . . , L
 k!
0 for k = L + 1, L + 2, . . .
(ν)
Evidently for this simplified form of the function aL (k)
+∞
X L
X +∞
X L
X
(ν) (ν) (ν) (ν)
|aL (i)| = |aL (i)| + |aL (i)| = |aL (i)| ≈ 1. (1.44)
i=1 i=1 i=L+1 i=1

To evaluate the impact of the neglected term one may use a two variables error
function which may help in an approximation of the function a(ν) (k) by function
(1.43)
+∞
X L
X
(ν) (ν)
e(ν, L) = |aL (i)| = 1 − |aL (i)|. (1.45)
i=L+1 i=1

Plot of the error function e(ν, L) for ν ∈ (01) and L ∈ (0, 200) is given in
Fig. 1.25(a).
The error function may be treated as a useful tool in establishing a relation
between the coefficient ν and function (1.43) length L. Assuming that function
(1.43) takes a constant value eν,L from a range (0, 1] one may derive a relation
between ν and L
L
X (ν)
|aL (i)| = 1 − eν,L < 1. (1.46)
i=1

Curves of a constant error function on a Lν plain are presented in Fig. 1.26.

Establishing a horizontal line (dashed line in the above figure) one may find a
relation between a coefficient ν and a length L for different constants.
August 20, 2015 10:54 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 19

October 15, 2015 12:18 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 19

Discrete-Variable Functions 19
Discrete-variable real functions 19

0.8

0.6
e(ν,L)

0.4

0.2

0
0 1
50 0.8
0.6
100
0.4
150
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

0.2
200 0 ν
L

(a)
(a) (b) (b)
Discrete Fractional Calculus Downloaded from www.worldscientific.com

Fig.
Fig. 1.25
1.25 Plot
Plotand
and2D2D
image of the
image error
of the function
error e(ν, L).
function e(ν, L).

0.1

0.2 0.1
0.3
0.4 0.2
0.5
0.6 0.3
0.7
0.8 0.4
0.9
1 0.5
0.6
0.7
0.8
0.9
1

Fig. 1.26 Curves of constant values eν,L (1.45).

Fig. 1.26 Curves of constant values eν,L (1.45).

Theorem 1.5. For k = 1, 2, 3, . . . and a coefficient ν ∈ (0, 1] the following inequality
is valid
Theorem 1.5. For k = 1, 2, 3,|a.(ν)
. .(k)|
and6a |a (k)|. ν ∈ (0, 1] the following
coefficient
(−ν) inequality
(1.47)
is valid  (ν)   (−ν)(1) 
Proof. For k = 1 evidently a (1) = |−ν| = ν = a  = ν. By induction, if
(ν) (−ν)
|a (k)|  |a (k)|.
(1.47) holds for some k − 1 > 1 then in view of the inequalities (1.20) (1.47)
  (−ν)  (−ν)(1) 
|a(−ν)
Proof. For = −1 1)|
k (k − |a(ν) (ka−(ν)
evidently 1)| = a=
(1) (k −=1)ν +=a(ν)
|−ν| a (k − 1) >
= 0ν. By induction,
(1.48) if
(1.47) holds for some k − 1 > 1 then in view of the inequalities (1.20)
and evidently
|a(−ν) (k − 1)| − |aa(ν)
(−ν)(k(k−−1)|
1) =
(−ν)
− aa(ν) (k (k −>
− 1) + a(ν) (k − 1)  0
1) 0. (1.49)(1.48)

and evidently

a(−ν) (k − 1) − a(ν) (k − 1) > 0. (1.49)

September 21, 2015 11:43 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 20

20 Discrete Fractional Calculus

Hence,
|a(−ν) (k)| − |a(ν) (k)| = a(−ν) (k) + a(ν) (k)
   
(−ν) −ν + 1 (ν) ν+1
=a (k − 1) 1 − + a (k − 1) 1 −
k k
  1

= a(−ν) (k − 1) + a(ν) (k − 1) 1−
k
 ν
+ a(−ν) (k − 1) − a(ν) (k − 1) > 0. (1.50)
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

k
Discrete Fractional Calculus Downloaded from www.worldscientific.com

   
For the sake of comparison, in Fig. 1.27 plots of functions a(ν) (k) and a(−ν) (k)
for ν = 0.5 are presented.

0.5

0.45

0.4
|a(ν)(k)|(.), |a(−ν)|(k)(×), ν = 0.5

0.35

0.3

0.25

0.2

0.15

0.1

0.05

0
0 5 10 15 20 25 30
k

Plots of functions a(ν) (k) and a(−ν) (k) for ν = 0.5.

   
Fig. 1.27

1.3.3.2 Function a(n) (k) properties

Now the function a(ν) (k) for parameter ν ∈ Z0 will be examined. To distinguish
this case from the one considered before one introduces a notation ν = n ∈ Z0 . The
function is defined as follows

Definition 1.8. The discrete a(n) (k) is defined as


 0 for k < 0


(n) 1 for k = 0
a (k) = k n(n−1)···(n−k+1) . (1.51)

 (−1) for k = 1, 2, . . . , n − 1, n
 k!
0 for k = n + 1, n + 2, . . .
October 15, 2015 12:18 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 21

Discrete-variable real functions 21

Table 1.2 Function a(n) (k) values.

n a(n)(k) , k = 0, . . . , n
0 1
1 11
2 121
3 1331
4 14641
5 1 5 10 10 5 1
6 1 6 15 20 15 6 1
7 1 7 21 35 35 21 7 1
8 1 8 28 56 70 56 28 8 1
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

9 1 9 36 84 126 126 84 36 9 1
Discrete Fractional Calculus Downloaded from www.worldscientific.com

Table 1.3 Function a(−n) (k) values.

n a(−n)(k) , k = 0, . . . , n
0 1000000000
1 1111111111
2 1 2 3 4 5 6 7 8 9 10
3 1 3 6 10 15 21 28 36 45 55
4 1 4 10 20 35 56 84 120 165 220
5 1 5 15 30 70 126 210 330 495 715
6 1 6 21 56 126 252 462 792 1287 2002
7 1 7 28 84 210 462 924 1716 3003 5005
8 1 8 36 120 330 792 1716 3432 6435 11440
9 1 9 45 165 495 1287 3003 6435 12870 24310

The function has only n + 1 non-zero values. Its absolute values for n = 0, 1, . . . , 9
are given below. The integers create a so-called Pascal triangle.
From Table 1.1, one can see that the maximal value is reached for b n2 c, i.e.
 j n k
 
max |a(n) (k)| = a(n) . (1.52)
k∈Z0 2

For negative integers n the function a(n) (k) always takes non-zero values. To em-
phasize this case the condition n > 0 will be preserved but the function will be
denoted as a(−n) (k). Values of a(−n) (k) for n = 0, 1, . . . , 9 are collected in Table 1.3

1.3.4 Function c(ν) (k)

Another discrete-variable function of great importance is the function c(ν) (k). It is
defined as

 0 for k < 0
c(ν) (k) = 1 for k = 0 . (1.53)

1 − ν+1
k for k = 1, 2, . . .
September 4, 2015 11:36 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 22

22 Discrete Fractional Calculus

Simple transformations show that

a(ν) (k)
c(ν) (k) = (1.54)
a(ν) (k − 1)

and

ν+1
c(ν) (k) = c(ν) (k − 1) + . (1.55)
k(k − 1)
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

Similarly to the analysis of function a(ν) (k) plots of function c(ν) (k) due to
different parameter positive values from four sets are shown in Figs. 1.28(a), 1.28(b),
Discrete Fractional Calculus Downloaded from www.worldscientific.com

1.29(a) and 1.29(b).

1.5 1.5

1
1
c(ν) (k), ν ∈ {1.0, 1.1, · · · , 2.0}
c(ν) (k), ν ∈ {0.0, 0.1, · · · , 1.0}

0.5

0.5
0

0 −0.5

−1
−0.5

−1.5

−1
=1 −2
=2

−1.5 −2.5
0 2 4 6 8 10 0 2 4 6 8 10
k k
(a) (b)

Fig. 1.28 Plots of function (a) c(ν) (k) for ν ∈ {0, 0.1, . . . , 1.0} and (b) ν ∈ {1.0, 1.1, . . . , 2.0}.

1.5 2

1
1
c(ν) (k), ν ∈ {3.0, 3.1, · · · , 4.0}
c(ν) (k), ν ∈ {2.0, 2.1, · · · , 3.0}

0.5

0 0

−0.5
−1
−1
−2
−1.5

−2 −3

−2.5
−4 =4
−3
=3

−3.5 −5
0 2 4 6 8 10 0 2 4 6 8 10
k k
(a) (b)

Fig. 1.29 Plots of function (a) c(ν) (k) for ν ∈ {2.0, 2.1, . . . , 3.0} and (b) ν ∈ {3.0, 3.1, . . . , 4.0}.
 October
August 20, 15,
20152015 12:18
10:54 BC:BC:
98339833 - Discrete
- Discrete Fractional
Fractional Calculus
Calculus ws-book975x65
ws-book975x65 pagepage
23 23

Discrete-variable Functions
Discrete-Variable real functions 23 23

1.2 1.2 2.5

c(ν) (k), ν ∈ {−1.0, −1.1, · · · , −2.0}

c(ν) (k), ν ∈ {0.0, −0.1, · · · , −1.0}
c(ν) (k), ν ∈ {0.0, −0.1, · · · , −1.0}

1 1
2

0.8 0.8
1.5

0.6 0.6
1
ν=−1
0.4 0.4

0.5
0.2 0.2

0
0 0
ν=0 =0
−0.2 −0.2 −0.5
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

0 0 2 2 4 4 6 6 8 8 10 10 0 2 4 6 8 10
k k k
(a)(a) (b)(b)
Discrete Fractional Calculus Downloaded from www.worldscientific.com

Fig.
Fig.1.30
1.30Plots function(a)(a)c(ν)c(ν)
Plotsof offunction (k)(k)forforν ν∈ ∈{0,{0,
−0.1, . . . ., .−1.0}
−0.1, andand
. , −1.0} ν ν∈ ∈
(b) (b)
{−1.0, −1.1,
{−1.0, . . ..,.−2.0}.
−1.1, . , −2.0}.

3.5 3.5 4.5 4.5

ν=−4 =−4
ν=−3=−3
c(ν) (k), ν ∈ {−3.0, −3.1, · · · , −4.0}
c(ν) (k), ν ∈ {−2.0, −2.1, · · · , −3.0}

4 4
c(ν) (k), ν ∈ {−3.0, −3.1, · · · , −4.0}
c(ν) (k), ν ∈ {−2.0, −2.1, · · · , −3.0}

3 3
3.5 3.5
2.5 2.5
3 3
2 2 2.5 2.5

1.5 1.5 2 2

1.5 1.5
1 1
1 1
0.5 0.5
0.5 0.5
0 0 0 0

−0.5 −0.5
0−0.5 2 4 6 8 10 0 −0.5 2 4 6 8 10
0 2 4 6 8 10 0 2 4 6 8 10
k k k k
(a)(a) (b)(b)
(ν) (k) for ν
Fig.
Fig.1.31
1.31Plots
Plotsof offunction
function(a)(a)c c(ν) (k) for ν∈ ∈{−2.0, −2.1,
{−2.0, . . . ., .−3.0}
−2.1, andand
. , −3.0} ν ν∈ ∈
(b) (b)
{−3.0, −3.1, . . . , −4.0}.
{−3.0, −3.1, . . . , −4.0}.

1.3.4.1
1.3.4.1 Function c(ν)
Function (k)(k)
c(ν) properties
properties
InIn
this Section
this sectionseveral properties
several properties of of
thethe
function c(ν)c(ν)
function (k)(k)
willwill
be be
proved. Immediate
proved. Immediate
calculation reveals that
calculation reveals that

ν +ν 1+ 1 
limlimc(ν) (k)
(ν) =
c (k) = lim 1 −
lim 1 − k = 1.
= 1. (1.56)
(1.56)
k→∞ k→∞
k→∞ k→∞ k
ByBy
(1.54) from
(1.54) (1.56)
from one
(1.56) states
one that
states that
a(ν)(ν)
(k)
limlim (ν) a (k) = 1= 1 (1.57)
(1.57)
k→∞ a (ν)
k→∞ a
(k − 1)
(k − 1)
September 7, 2015 12:4 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 24

24 Discrete Fractional Calculus

and
ν+1
c(ν) (k) 1− k
lim = lim ν+1 = 1. (1.58)
k→∞ c(ν) (k − 1) k→∞ 1 −
k−1

The function c(ν) (k) is a spline function of degree 3 (i.e. piecewise defined over three
intervals). The first interval is degenerated to one point k = 0, and the second
for k > 0. By formulas (1.53) and (1.56), c(ν) (0) = c(ν) (+∞) = 1. Assuming
that c(ν) (k) is non-constant one concludes that there must be an extremum. The
following Theorems determine this property.
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

Theorem 1.6. Function c(ν) (k) is monotonically decreasing for ν ∈ (−∞, −1) and
Discrete Fractional Calculus Downloaded from www.worldscientific.com

k = 1, 2, . . . and monotonically increasing for ν ∈ (−1, +∞) for k = 1, 2, . . . .

Proof. To prove this theorem one should check inequality for k = 2, 3, . . .

c(ν) (k) − c(ν) (k − 1) < 0 for ν ∈ (−∞, −1). (1.59)

A substitution of the second non-zero spline in (1.53) into (1.59) yields

ν+1 ν+1 (ν + 1)
1− −1+ = <0 for ν ∈ (−∞, −1). (1.60)
k k−1 k(k − 1)

Hence, for ν +1 < 0 the considered spline is monotonically decreasing with c(ν) (1) =
−ν > 0 as its maximal element. Next,

c(ν) (k) − c(ν) (k − 1) > 0 for ν ∈ (−1, +∞). (1.61)

Similar substitution gives the opposite result for ν + 1 > 0 proving that this
time a spline is monotonically increasing with c(ν) (1) = −ν < 0 as its minimal
element.

Theorem 1.7. Extreme values of the function c(ν) (k) equal to −ν

max c(ν) (k) = c(ν) (1) = −ν for ν ∈ (−∞, −1) (1.62)

k∈Z0

c(ν) (k) = const = c(ν) (1) = 1, k ∈ Z0 for ν = −1 (1.63)

min c(ν) (k) = c(ν) (1) = −ν for ν ∈ (−1, +∞). (1.64)

k∈Z0

Proof. For ν ∈ (−∞, −1) the inequality 1 = c(ν) (0) < c(ν) (1) = −ν holds. In view
of the result stated by Theorem 1.6 c(ν) (1) > c(ν) (k) for k = 2, 3, . . . and formula
(1.62) is proved. By analogy, for ν ∈ (−1, −∞) there is 1 = c(ν) (0) > c(ν) (1) = −ν
and the function c(ν) (k) increases for k = 2, 3, . . . . This shows the correctness of
(1.64). Finally, for ν = −1 the function c(−1) (k) = 1 (is constant). This ends the
proof.
October 15, 2015 12:18 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 25

Discrete-variable real functions 25

The next theorem is of great practical importance especially in a real-time

micro-processor functions a(ν) (k) and c(ν) (k) evaluations. Information about func-
tions convergence provides, inter alia, differences and ration between its consecutive
values.

Theorem 1.8. For 0 < ν < 1 and k = 1, 2, . . . the following inequality holds
a(ν) (k) − a(ν) (k − 1) > c(ν) (k) − c(ν) (k − 1). (1.65)

Proof. First inequality (1.65) will be proved for k = 3, 4, . . . . By definition formula

(1.11)
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

 
(ν) (ν) (ν) ν+1
L = a (k) − a (k − 1) = a (k − 1) 1 − − a(ν) (k − 1)
Discrete Fractional Calculus Downloaded from www.worldscientific.com

k
ν + 1 (ν)
=− a (k − 1) > 0. (1.66)
k
Simple calculations reveal that
ν + 1 (ν)
L=− a (k − 1) = −a(ν+1) (k) > 0. (1.67)
k
Inequality (1.66) results from the first formula (1.20) and is confirmed by inequality
(1.67) which is valid on grounds of the second formula (1.20). Similarly, by (1.53)
   
(ν) (ν) ν+1 ν+1
R = c (k) − c (k − 1) = 1 − − 1−
k k−1
ν+1
= > 0. (1.68)
k(k − 1)
Up to now it was proved that L, R > 0. What remains to be determined is the
relationship between L and R. The assumption that L < R yields
(ν + 1)ν(ν − 1) · · · (ν − k + 2) ν+1
(−1)k+1 < . (1.69)
k! k(k − 1)
ν+1
Dividing both sides by positive factor k(k−1) one obtains
ν(ν − 1) · · · (ν − k + 2)
−(−1)k−2
(k − 2)!
     
ν ν ν
= −ν 1 − 1− ··· 1 − < 1. (1.70)
1 2 k−2
This is because by assumption, ν > 0 and all terms in brackets are positive. For
k = 2 there is
ν(ν + 1) ν+1
a(ν) (2) − a(ν) (1) = < c(ν) (2) − c(ν) (1) = (1.71)
2 2
bearing in mind that 0 < ν < 1. Finally for k = 1
a(ν) (1) − a(ν) (0) = −1 + ν = c(ν) (1) − c(ν) (0) = 1 − ν (1.72)
which ends the proof.
September 4, 2015 11:36 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 26

26 Discrete Fractional Calculus

The next Theorem estimates the functions a(ν) (k) and c(ν) (k) convergence.

Theorem 1.9. For 0 < ν < 1 and k = 3, 4, . . . the following inequality holds
   
 a(ν) (k)   c(ν) (k) 
 < 
 a(ν) (k − 1)   c(ν) (k − 1)  . (1.73)

Proof. By definition formula (1.53) one has

 (ν)       
 c (k)   1 − ν+1  
 ν + 1  1 

 = k 1 − 
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

 =
 c(ν) (k − 1)   1 − ν+1     1 − ν+1  . (1.74)
k−1
k k−1
Discrete Fractional Calculus Downloaded from www.worldscientific.com

From (1.12) one obtains

   
 a(ν) (k)   
  = 1 − ν + 1  . (1.75)
 a(ν) (k − 1)   k 

Hence
 (ν)    
 c (k)   a(ν) (k)   1 

   
 c(ν) (k − 1)  =  a(ν) (k − 1)   1 − ν+1  . (1.76)
k−1

Noting that
 
 1 
 
 ν+1  > 1 for k = 3, 4, . . .
 1 − k−1
(1.77)

one obtains the result (1.73).

1.4 Fundamental operations on the discrete-variable functions

In this section fundamental operations that can be performed on discrete-variable

functions will be introduced. On the basis of the numerous references one can state
that the operations follow the well-known ones applied to the continuous-variable
functions case. In both cases the fundamental assumption is the same function
domain. Hence one assumes that these functions are defined over the same domain
Tk0 ,k,h .

1.4.1 Discrete-variable functions addition and multiplication

Assuming that f, g : Tk0 ,k,h → R one defines two operations: function addition and
multiplication executed in a classical way and denoted by f (k)+g(k) and f (k)·g(k),
respectively.
October 15, 2015 12:18 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 27

Discrete-variable real functions 27

1.4.2 Discrete-variable function forward and backward shift

In the discrete-variable function analysis it is very useful to define a function shift
understood as a backward (left) and forward (right) shift performed along the hor-
izontal axis. Using this operation one can easily define new non-typical discrete-
variable functions. First one defines so-called shifted forward Dirac pulse. It is
defined on the basis of function (1.6).

Definition 1.9. The shifted forward discrete Dirac pulse is defined as


1 for k = k0
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

δ(k − k0 ) = . (1.78)
0 for k 6= k0
Discrete Fractional Calculus Downloaded from www.worldscientific.com

The plot and image of the shifted forward Dirac pulse are presented in Figs. 1.15(a)
and 1.15(b), respectively.

1.2 2

1.9
1
1.8

0.8 1.7

1.6
0.6
1[px]
δ(k)

1.5
0.4
1.4

0.2 1.3

1.2
0
1.1

−0.2 1
−10 −5 0 5 10 15 20 25 2 4 6 8 10 12 14 16 18 20
k k[px]
(a) (b)

Fig. 1.32 Plot (a) and 1D image (b) of the shifted forward Dirac pulse for k0 = 5.

Similarly, one defines the shifted forward unit step function. Its definition is
given below.

Definition 1.10. The shifted forward discrete unit step function defines the follow-
ing formula

0 for k < k0
1(k − k0 ) = . (1.79)
1 for k > k0

The plot and image of the shifted forward discrete unit step function are presented
in Figs. 1.33(a) and 1.33(b), respectively. Here it is worth pointing that the discrete
unit step function can be equivalently defined by a sum of shifted forward discrete
Dirac pulses.
k
X
1(k − k0 ) = δ(k − i). (1.80)
i=k0
September 4, 2015 11:36 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 28

28 Discrete Fractional Calculus

1.2 2

1.9
1
1.8

0.8 1.7

1.6
δ(k − k0 )

0.6

1[px]
1.5
0.4
1.4

0.2 1.3

1.2
0
1.1

−0.2 1
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

−5 0 5 10 15 20 25 30 2 4 6 8 10 12 14 16 18 20
k k[px]
(a) (b)
Discrete Fractional Calculus Downloaded from www.worldscientific.com

Fig. 1.33 Plot (a) and 1D image (b) of the shifted forward unit step function for k0 = 5.

Definition 1.11. The shifted forward general exponential function

f (k) = ab(k−k0 ) 1(k − k0 ) where a, b, ∈ R. (1.81)

One should note that formula (1.13) is defined, in fact, by a product of two equally
shifted forward functions: ab(k−k0 ) and 1(k − k0 ). This is not obligatory. One can
define two different shifts leading to different function shapes. In Figs. 1.16(a) and
1.16(b) plots of functions eb(k−k0 ) 1(k − k0 ) and ebk 1(k − k0 ) with b = −0.05 and
k0 = 5 are presented. One can easily observe the difference between the plots. The
following figures show the images of the considered functions

1.2 1.2

1 1

0.8 0.8

0.6 0.6
f(k)

f(k)

0.4 0.4

0.2 0.2

0 0

−0.2 −0.2
−10 0 10 20 30 40 50 60 −10 0 10 20 30 40 50 60
k k
(a) (b)

Fig. 1.34 Plots of functions (a) eb(k−k0 ) 1(k − k0 ) and (b) ebk 1(k − k0 ) for k0 = 7.

As was mentioned earlier the shift operations may be very useful in description
of several discrete-variable signals. Next some examples are presented.
October 15, 2015 12:18 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 29

Discrete-variable real functions 29

5 5

4.5 4.5

4 4

3.5 3.5
l[px]

l[px]
3 3

2.5 2.5

2 2

1.5 1.5

1 1
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

0 50 100 150 200 250 300 0 50 100 150 200 250 300
kl[px] kl[px]

(a) (b)
Discrete Fractional Calculus Downloaded from www.worldscientific.com

Fig. 1.35 Images of functions (a) eb(k−k0 ) 1(k − k0 ) and (b) ebk 1(k − k0 ) for k0 = 7.

Example 1.2. A function defined as


0 for k < 0
f (k) = 1 for 0 6 k 6 5 (1.82)

0 for k > 5
can be equivalently described as

f (k) = 1(k) − 1(k − 6). (1.83)

Example 1.3. Consider a function defined by this formula


 0 for k < 0



 k for 0 6 k 6 5
f (k) = 5 for 6 6 k 6 10 . (1.84)



 −k + 15 for 11 6 k 6 14

0 for k > 15
It may be also described as a sum of shifted functions

f (k) = k1(k) − (k − 5)1(k − 5) − (k − 10)1(k − 10) + (k − 15)1(k − 15). (1.85)

A plot and a related image are given in Figs. 1.36(a) and 1.36(b).

1.4.3 Convolution of two discrete-variable functions

The most important operation one can perform on two discrete-variable functions
is called a real convolution. It plays an important role in transient analysis of linear
time-invariant dynamical systems.

Definition 1.12. For two scalar functions f1 (k), f2 (k) of a discrete-variable k sat-
isfying the condition f1 (k), f2 (k) = 0 for k < 0 one defines the third one f (k) in the
September 4, 2015 11:36 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 30

30 Discrete Fractional Calculus

6 5

5 4.5

4
4

3.5
3

l[px]
f(k)

3
2
2.5

1
2

0 1.5

−1 1
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

−5 0 5 10 15 20 25 30 0 20 40 60 80 100 120 140

k kl[px]

(a) (b)
Discrete Fractional Calculus Downloaded from www.worldscientific.com

Fig. 1.36 Plot and an image of function (1.85).

following way
k
X
f (k) = f1 (i)f2 (k − i). (1.86)
i=0

The asterisk(∗) denotes the discrete convolution operation. Hence,

f (k) = f1 (k) ∗ f2 (k). (1.87)

The real discrete convolution satisfies three fundamental laws:

(I) Commutative law

f1 (k) ∗ f2 (k) = f2 (k) ∗ f1 (k). (1.88)

(II) Distributive law

f1 (k) ∗ [f2 (k) ± f3 (k)] = f1 (k) ∗ f2 (k) ± ∗f1 (k) ∗ f3 (k). (1.89)

(III) Associative law

f1 (k) ∗ [f2 (k) ∗ f3 (k)] = [f1 (k) ∗ f2 (k)] ∗ f3 (k). (1.90)

The proofs of the laws mentioned above can be found in numerous positions. For-
mula (1.87) can be also expressed in a vector-matrix product form. Using vectors
and (k + 1) × (k + 1) anti-diagonal matrix Nk defined in (A.19)
   
f1 (0) f2 (0)
f1 (1) f2 (1)
   
f1 (k) =  .  , f2 (k) =  .  (1.91)
 . .  . .
f1 (k) f2 (k)
October 15, 2015 12:18 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 31

Discrete-variable real functions 31

the discrete convolution is expressed as a product

f (k) = f1 (k) ∗ f2 (k) = [f1 (k)]T Nk f2 (k). (1.92)

The convolution sum is valid for all non-negative values of k. In evaluating all such
sums for k, k − 1, . . . , 1, 0 one obtains a set of equalities. Collecting all of them in
one matrix-vector formula yields:
   
f (k) f1 (k) ∗ f2 (k)
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

   
f (k − 1) f1 (k − 1) ∗ f2 (k − 1)
   
   
Discrete Fractional Calculus Downloaded from www.worldscientific.com

f(k) =  .. = ..
.   . 
   
   
 f (1)   f1 (1) ∗ f2 (1) 
f (0) f1 (0) ∗ f2 (0)
 
f1 (0) f1 (1) . . . f1 (k − 1) f1 (k)
 
 0 f1 (0) . . . f1 (k − 2) f1 (k − 1)
 
 
=  ... ..
.
..
.
..
.  Nk f2 (k)
 
 
 0 0 . . . f1 (0)) f1 (2) 
0 0 ... 0 f1 (0)
  
f1 (0) f1 (1) . . . f1 (k − 1) f1 (k) f2 (k)
  
 0 f1 (0) . . . f1 (k − 2) f1 (k − 1) f2 (k − 1)
  
  
=  ... ..
.
..
.
..
. 
..
. . (1.93)
  
  
 0 0 . . . f1 (0)) f1 (2)   f2 (1) 
0 0 ... 0 f1 (0) f2 (0)

Formula (1.93) is very convenient in computer calculations giving the discrete con-
volution values ordered in a vector form. The numerical examples presented above
explain the calculation procedure.

Example 1.4. Consider two simple functions

f1 (k) = 1(k) − 1(k − 3) (1.94)

f2 (k) = 1(k − 2) − 1(k − 5). (1.95)

Plots of functions f1 (k) and f2 (k) are presented in Figs. 1.20(a) and Fig. 1.20(b),
respectively. One should evaluate the discrete convolution of these functions due to
formula (1.93).
September 4, 2015 11:36 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 32

32 Discrete Fractional Calculus

1.2 1.2

1 1

0.8 0.8

0.6 0.6
f1 (k)

f2 (k)
0.4 0.4

0.2 0.2

0 0

−0.2 −0.2
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

−6 −4 −2 0 2 4 6 8 10 12 −6 −4 −2 0 2 4 6 8 10 12
k k
(a) (b)
Discrete Fractional Calculus Downloaded from www.worldscientific.com

Fig. 1.37 Plot of functions (1.94) and (1.95).

    
.. .. .. .. .. .. .. .. . .
 . . . . . . . .   ..   .. 
. . . 1 1 1 0 0 0 0 0    
  0 0
. . . 0 1 1 1 0 0 0 0 
   
 0 1
. . . 0 0 1 1 1 0 0 
0 0
  
  2
   =  . (1.96)
. . . 0 0 0 1 1 1 0 0 1 3
    
. . . 0 0 0 0 1 1 1 0 1 2
    
. . . 0 0 0 0 0 1 1 1    
  1 1
. . . 0 0 0 0 0 0 1 1 0 0

... 0 0 0 0 0 0 0 1 0 0

From (1.96) follows that

f (k) = f1 (k) ∗ f2 (k)

= 1(k − 2) + 1(k − 3) + 1(k − 4) − 1(k − 5) − 1(k − 6) − 1(k − 7)
= (k − 1)1(k − 1) − 2(k − 4)1(k − 4) + (k − 7)1(k − 7). (1.97)

Plot of the convolution function f (k) and its image are presented in Figs. 1.38(a)
and Fig. 1.38(b), respectively.
In the next example one calculates a convolution sum of two trigonometric
functions.

Example 1.5. For the two trigonomeric functions below, evaluate its convolution
sum.

f1 (k) = sin(ak)1(k), f2 (k) = cos(ak)1(k). (1.98)

October 15, 2015 12:18 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 33

Discrete-variable real functions 33

3.5 2

1.9
3
1.8
2.5
1.7
2
1.6
f(k)

1[px]
1.5 1.5

1.4
1
1.3
0.5
1.2
0
1.1

−0.5 1
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

−6 −4 −2 0 2 4 6 8 10 12 14 2 4 6 8 10 12
k k[px]
(a) (b)
Discrete Fractional Calculus Downloaded from www.worldscientific.com

Fig. 1.38 Plot (a) and an image (b) of function (1.97).

By definition (1.12)

f (k) = f1 (k) ∗ f2 (k) = sin(ak)1(k) ∗ cos(ak)1(k)

k
X
= sin(ai)1(i) cos[a(k − i)]1(k − i). (1.99)
i=0

Now one considers two cases where: k − odd and k − even. In the first case (k −
odd, k > 0), the right-hand terms in expression (1.99) can be re-grouped in the
following way:
k−1
X
2

f (k) = {sin(ai) cos[a(k − i)] + sin[a(k − i)] cos(ai)}1(k − i). (1.100)

i=0

Note that in the considered case k−1 2 − even and 1(k)1(k − i) = 1(k − i). From
trigonometric identity for i = 0, 1, . . . one obtains:

sin(ai) cos[a(k − i)] + sin[a(k − i)] cos(ai) = sin(ak). (1.101)

Substitution of (1.101) into (1.105) yields

k−1
X
2

f (k) = sin(ak) 1(k − i). (1.102)

i=0

From Chapter 2, it will be clear that

k−1
X
2
k+1
1(k − i) = 1(k). (1.103)
i=0
2
September 4, 2015 11:36 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 34

34 Discrete Fractional Calculus

Hence for k − odd one finally gets

k+1
f (k) = sin(ak)1(k) ∗ cos(ak)1(k) = sin(ak)1(k). (1.104)
2
For k − even, k > 0 the formula slightly changes.
k
2 −1
X
f (k) = {sin(ai) cos[a(k − i)] + sin[a(k − i)] cos(ai)}1(k − i)}
i=0
     
k k k
+ sin a cos a 1 . (1.105)
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

2 2 2
Analogous substitution of (1.101) into (1.105) leads to the final form:
Discrete Fractional Calculus Downloaded from www.worldscientific.com

k
2 −1
X      
k k k
f (k) = sin(ak) 1(k − i) + sin a cos a 1 . (1.106)
i=0
2 2 2

By formula (1.103) evaluated over interval [0, k2 ] one finally obtains for k − even,
k > 0:
k 1
f (k) = sin(ak)1(k) ∗ cos(ak)1(k) = sin(ak)1(k) + sin(ak)1(k)
2 2
k+1
=
sin(ak)1(k). (1.107)
2
Finally, by direct calculation of the convolution sum for k = 0 one obtains:
f (k)|k=0 = [sin(ak)1(k) ∗ cos(ak)1(k)]|k=0 = 0. (1.108)
Summing up the results shown above, the final form of evaluated convolution sum
equals
k+1
sin(ak)1(k).
f (k) = (1.109)
2
The plot of the convolution function f (k) and its 1D image are presented in
Figs. 1.39(a) and 1.39(b), respectively.
Here it is worth commenting upon result (1.109). Though both functions f1 (k) and
f2 (k) defined in (1.104) are bounded, i.e.: −1 < sin(ak), cos(ak) < 1 for all k its
convolution sum values tend to −∞ and +∞. Such a transient behaviour can be
explained on the basis of the discrete linear systems stability analysis.

1.5 Two discrete-variables functions

Now one considers two independent discrete-variable real function. The variables
will be denoted as k1 and k2 . Hence,
f :Z×Z→R (1.110)
October 15, 2015 12:18 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 35

Discrete-variable real functions 35

50 5

40
4.5
30
4
20
3.5
10
f(k)

l[px]
0 3

−10
2.5
−20
2
−30
1.5
−40

−50 1
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

−20 0 20 40 60 80 100 120 50 100 150 200 250 300 350 400 450 500
k kl[px]

(a) (b)
Discrete Fractional Calculus Downloaded from www.worldscientific.com

Fig. 1.39 Plot (a) and an image (b) of function (1.109).

In practice, considerations will be limited to

f : ZN1,min ,N1,max ,h × ZN2,min ,N2,max ,h → Ttmin ,tmax . (1.111)

Similarly to the one discrete-variable real function there are several representations
of two discrete-variable functions. Here will be mentioned only two which are the
most popular. The first one is a functional representation. The following numerical
example gives such a representation.

Example 1.6. Consider a slightly modified “sombrero” function described by the

formula
p 
sin 0.5k12 + k22
f (k1 , k2 ) = p . (1.112)
0.5k12 + k22

Its 3D plot for k1 ∈ [−N1,min , N1,max ] = [−10, 10] and k2 ∈ [−N2,min , N2,max ] =
[−10, 10] is given in Fig. 1.40.

Another very useful definition of a two discrete-variable function is similar to

a tabularization of the one discrete-variable function. In the following example
instead of the two column table one has an array. Such an array defining the
function f (k1 , k2 ) is presented in Fig. 1.41.
A classical matrix with real elements is an example of such an array. In matrix
algebra it is generally accepted that all matrix element indices are positive integers
starting at the left upper corner. Appropriate shift and rotation of the coordinate
axes enables a representation of any function f (k1 , k2 ) as a matrix Fk1,max,k2,max
with k1,max = N1,min + N1,max , k2,max = N2,min + N2,max .
The referenced operations are depicted in in Figs. 1.42(a) and Fig. 1.42(b),
respectively.
September 4, 2015 11:36 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 36

36 Discrete Fractional Calculus

by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.
Discrete Fractional Calculus Downloaded from www.worldscientific.com

Fig. 1.40 3D plot of function (1.112).

Fig. 1.41 Array containing data representing function f (k1 , k2 ) values.

1.5.1 Fundamental two discrete-variable functions

There is a two dimensional analogue of fundamental discrete-variable functions an-
alyzed in Subsection 1.3.1 These are the two dimensional (2D) discrete Dirac pulse
and 2D unit step function.

Definition 1.13 (2D discrete Dirac pulse). The 2D discrete Dirac pulse is de-
fined as:

1 for k1 = k2 = 0
δ(k1 , k2 ) = . (1.113)
0 for Z \ (0, 0)
October 15, 2015 12:18 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 37

Discrete-variable real functions 37

by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.
Discrete Fractional Calculus Downloaded from www.worldscientific.com

(a) (b)

Fig. 1.42 Shift and rotation of an array.

In Figs. 1.43(a) and 1.43(b) the 3D plot and image of the discrete 2D Dirac pulse
are presented.

10

9
1
8
0.8
7
δ(k1 , k2 )

0.6
6
k2

0.4
5
0.2
4
0
10 3
10
8 2
5 6
4
2 1
k2 0 0 0 2 4 6 8 10
k1
k1

(a) (b)

Fig. 1.43 Plot (a) and the image (b) of the 2D discrete Dirac pulse (1.113).

The function is the so-called 2D unit step function. Its definition is given below,
and its plot and image are depicted in Figs. 1.44(a) and 1.44(b), respectively.

Definition 1.14 (2D discrete unit step function). The 2D discrete unit step
function is defined as

0 for k1 , k2 < 0
1(k1 , k2 ) = . (1.114)
1 for k1 , k2 > 0
September 4, 2015 11:36 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 38

38 Discrete Fractional Calculus

10

9
1
8
0.8
7
1(k1 , k2 )

0.6
6

k2
0.4
5
0.2
4
0
10 3
10
8 2
5 6
4
2 1
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

k2 0 0 0 2 4 6 8 10
k1
k1

(a) (b)
Discrete Fractional Calculus Downloaded from www.worldscientific.com

Fig. 1.44 Plot (a) and the image (b) of the discrete unit step function (1.7).

1.5.2 Fundamental operations on the two discrete-variable

functions
Regarding the basic operations on the two discrete-variable functions they are al-
most the same as those mentioned in the previous section. The only condition is
the same domain. Thus, for functions f, g : ZN1,min ,N1,max ,h × ZN2,min ,N2,max ,h → R
one can evaluate:

(I) weighted sum of two (or more) functions h(k1 , k2 ) = af (k1 , k2 ) + bg(k1 , k2 ),
where a, b, ∈ R

(II) product of two functions h(k1 , k2 ) = f (k1 , k2 )g(k1 , k2 )

f (k1 ,k2 )
(III) quotient of two functions h(k1 , k2 ) = g(k1 ,k2 ) where g(k1 , k2 ) 6= 0 for k1 ∈
ZN1,min ,N1,max ,h , k2 ∈ ZN2,min ,N2,max ,h

(IV) concatenation of two functions: g(f ) : ZN1,min ,N1,max ,h × ZN2,min ,N2,max ,h → R

where f : ZN1,min ,N1,max ,h ×ZN2,min ,N2,max ,h → ZN3,min ,N3,max ,h ×ZN4,min ,N4,max ,h
and g : ZN3,min ,N3,max ,h × ZN4,min ,N4,max ,h → R.

The next numerical example shows the result of a two functions’ product.

Example 1.7. Consider slightly special two discrete-variables functions

 
πk1
f (k1 , k2 ) = sin (1.115)
6
 
πk2
g(k1 , k2 ) = cos . (1.116)
6

Plot its product.

October 15, 2015 12:18 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 39

Discrete-variable real functions 39

   
πk1 πk2
h(k1 , k2 ) = f (k1 , k2 )g(k1 , k2 ) = sin cos . (1.117)
6 6

Solution. First in Figs. 1.45(a) and 1.45(b) the plots of functions (1.113) and
(1.114) are given, respectively. Next, its product is plotted in Fig. 1.46.
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.
Discrete Fractional Calculus Downloaded from www.worldscientific.com

(a) (b)

Fig. 1.45 3D plot of function (a) (1.113) and (b) (1.114).

Fig. 1.46 3D plot of function (1.115).

September 4, 2015 11:36 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 40

40 Discrete Fractional Calculus

1.5.3 Two dimensional discrete convolution

A discrete two dimensional convolution plays a very important role in the two
dimensional signal and image processing. It is mainly defined as
∞
X ∞
X
g(k1 , k2 ) = f1 (k1 , k2 ) ∗ f1 (k1 , k2 ) = f1 (k1 , k2 )f2 (k1 − i1 , k2 − i2 ).
i1 =−∞ i2 =−∞
(1.118)
However, for f1 (k1 , k2 ) = f2 (k1 , k2 ) = 0 for k1 , k2 < 0 the formula simplifies to the
form
by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

k1 X
X k2
g(k1 , k2 ) = f1 (k1 , k2 ) ∗ f1 (k1 , k2 ) = f1 (k1 , k2 )f2 (k1 − i1 , k2 − i2 ). (1.119)
Discrete Fractional Calculus Downloaded from www.worldscientific.com

i1 =0 i2 =0

Assuming that function values fi (k1 , k2 ) are treated as a matrix Fi,k1,max, k2,max for
i = 1, 2 formula (1.117) may be expressed as

g(k1 , k2 ) = f1 (k1 , k2 ) ∗ f1 (k1 , k2 ) = tr (F1,k1 ,k2 Nk2 F2,k2 ,k1 Nk1 ) (1.120)

where a matrix Nk is defined by formula (11.19). The next example shows an

application of the formula (1.118).

Example 1.8. Consider two functions

 
πk1
f1 (k1 , k2 ) = f2 (k1 , k2 ) = sin . (1.121)
24
Plot the discrete convolutions f1 (k1 , k2 ) ∗ f2 (k1 , k2 ),f1 (k1 , k2 ) ∗ δ(k1 , k2 ) and
f1 (k1 , k2 ) ∗ 1(k1 , k2 ).

Solution. Appropriate plots and related images are given in Figs. 1.47(a)–1.47(b),
1.48(a)–1.48(b), 1.49(a) and 1.49(b).

(a) (b)

Fig. 1.47 3D plot and image of 2D convolution of f1 (k1 , k2 ) ∗ f1 (k1 , k2 ).

October 15, 2015 12:18 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 41

Discrete-variable real functions 41

by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.

(a) (b)
Discrete Fractional Calculus Downloaded from www.worldscientific.com

Fig. 1.48 3D plot and image of 2D convolution of f1 (k1 , k2 ) ∗ δ(k1 , k2 ).

(a) (b)

Fig. 1.49 3D plot and image of 2D convolution of f1 (k1 , k2 ) ∗ 1(k1 , k2 ).

1.5.4 Discrete-image as the two discrete-variables function

Up to now there have been considered explicit functions of one or two discrete-
variables. Such functions can be represented as discrete images when their scaled
values are represented by levels of a specified gray scale. Such an image is considered
as a real two discrete-variable function F (k1 , k2 ) for k1 ∈ [1k1,max ], k2 ∈ [1k2,max ]
with elements coding the gray scale (or in general color). An example of a real
digital image is given in Fig. 1.50, [MF (2015)].
Chosen fragments of the function F (k1 , k2 ) can be plotted as 1D plots. So, for
k2 = const ∈ [103, 113] one gets a sub-image and a set of plots.
The plots in Fig. 1.51(b) clearly show the “Small Dogs” legs. The “Small Dog”
ears are more clearly indicated in Figs. 1.52(a)–1.52(b) which present a sub-image
and related plots.
September 4, 2015 11:36 BC: 9833 - Discrete Fractional Calculus ws-book975x65 page 42

42 Discrete Fractional Calculus

by UNIVERSITY OF CALIFORNIA @ SAN DIEGO on 04/01/16. For personal use only.
Discrete Fractional Calculus Downloaded from www.worldscientific.com

Fig. 1.50 Discrete image of the “Small Dog”.

220

200 k1 ∈ [1, 619], k2 ∈ [103, 113]

180

160

140
F (k1 , k2 )

120

100

80

60

40
Rear legs
20 F ront legs
0
20 40 60 80 100 120
k1

(a) (b)

Fig. 1.51 Sub-image of Fig. 1.50 and related plot of F (k1 , k2 ) for k2 = const ∈ [103, 113].

220

200 k1 ∈ [1, 619], k2 ∈ [14, 24]

180

160

140
F (k1 , k2 )

120

100

80

60

40

20
Right ear
Lef t ear
0
20 40 60 80 100 120
k1

(a) (b)

Fig. 1.52 Sub-image of Fig. 1.50 and related plots of F (k1 , k2 ) for k2 = const ∈ [14, 24].