Professional Documents
Culture Documents
PII: S2090-1232(20)30004-7
DOI: https://doi.org/10.1016/j.jare.2020.01.004
Reference: JARE 749
Please cite this article as: Tenreiro Machado, J.A., Lopes, A.M., Multidimensional Scaling Locus of Memristor
and Fractional Order Elements, Journal of Advanced Research (2020), doi: https://doi.org/10.1016/j.jare.
2020.01.004
This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover
page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version
will undergo additional copyediting, typesetting and review before it is published in its final form, but we are
providing this version to give early visibility of the article. Please note that, during the production process, errors
may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Abstract
This paper combines the synergies of three mathematical and computational generalizations. The concepts of frac-
tional calculus, memristor and information visualization extend the classical ideas of integro-dierential calculus,
electrical elements and data representation, respectively. The study embeds these notions in a common framework,
with the objective of organizing and describing the “continuum” of fractional order elements (FOE). Each FOE is
characterized by its behavior, either in the time or in the frequency domains, and the dierences between the FOE
are captured by a variety of distinct indices, such as the Arccosine, Canberra, Jaccard and Sørensen distances. The
dissimilarity information is processed by the multidimensional scaling (MDS) computational algorithm to unravel
possible clusters and to allow a direct pattern visualization. The MDS yields 3-dimensional loci organized according
to the FOE characteristics both for linear and nonlinear elements. The new representation generalizes the standard
Cartesian 2-dimensional periodic table of elements.
Keywords: Fractional calculus, Memristor, Information visualization, Multidimensional scaling, Procrustes analysis
1. Introduction
Leibniz (1646-1716) extended the dierential calculus to the paradigm known as “Fractional Calculus” (FC)
[1, 2]. However, the FC remained an abstract tool restricted to the area of mathematics. The first application of FC
is usually credited to Abel (1802-1829) with the so-called tautochrone curve problem [3, 4]. Later we find the work
of Heaviside (1850-1925), who fist applied such ideas in the scope of the operational calculus and electromagnetism
[5, 6]. Nonetheless, it was during the last two decades that FC was recognized as a good tool to characterize complex
phenomena, due to the ability of describing adequately non-locality and long-range memory eects [7, 8, 9, 10, 11,
12].
Paynter (1923-2002) formulated one systematic approach to modeling and invented the so-called bond graphs
[13]. He considered four generalized variables, namely the eort, flow, momentum and displacement fe; f ; p; qg so
R R
that p = e(t)dt and q = f (t)dt. In page 136 of his class notes [13] he designed a diagram including the 4
state variables with vertex of a “tetrahedron of state”. Paynter characterized the functional relationship between the
Corresponding author
2
the mathematical and computational methods. Sections 3 characterizes the FOE by distinct methods, namely in the
time and frequency domains. Additionally the FOE are compared with four distances and the information is processed
by means of the MDS technique. Section 4 compares the eect of nonlinearities by means of Procrustes analysis.
Finally, Section 5 draws the most important conclusions.
FC generalizes the concept of dierentiation and integration to non integer and complex orders [25, 26]. We find
a variety of applications of FC, such as in control, physics, anomalous diusion, and many others [27, 28, 29, 30].
Fractional derivatives and integrals are non-local operators that capture the history dynamics, contrary to what happens
with integer derivatives. Fractional systems have a memory of the dynamical evolution and many natural and artificial
phenomena revealed these characteristics [31, 32, 7, 8, 9, 10, 11, 33].
The most used definitions of fractional derivative are the Riemann-Liouville, Grünwald-Letnikov and Caputo
formulations [34, 35, 36]. For certain functions, the fractional derivative follows closely their integer order version.
For example, at steady state, a sinusoidal function with amplitude A and phase has the derivative of order 2R
given by [37]:
d
cos !t + + ;
[A cos( ! t + )] = A! (1)
dt 2
d
where dt denotes the fractional derivative or order , t represents time, and f and ! = 2 f are the frequency and
angular frequency, respectively.
In the frequency domain, for zero initial conditions and the function x(t), we can write:
( )
d
L x(t) = s Lf x(t)g; (2)
dt
( )
d
F x(t) = ( j!) F f x(t)g; (3)
dt
p
where Lfg and F fg represent the Fourier and Laplace operators, s stands for the Laplace variable and j = 1.
The frequency dependent negative conductance and frequency dependent negative resistance (FDNC and FDNR)
were introduced in 1969 and 1971 by Bruton [38] and Antoniou [39]. The electronic implementation of these elements
have been under progress during the last decades [40, 41, 42, 43]. The FDNC and FDNR are denoted by D- and N-
elements and are usually considered with linear behavior, having admittance and impedance Y (s) = Z 1
(s) = Ds2
3
and Z (s) = N s2 , respectively. These devices are often adopted in ladder filters without inductors [38] and chaotic
oscillators [44].
The FDNC and FDNR require an implementation using active devices and, although not passive, demonstrate that
they are feasible and useful. Moreover, for circuit branch impedances Zi (s) = ki sni ; ki 2 R, ni 2 N, i = 1; 2; , of
integer order we obtain also an integer order input impedance Z (s). On the other hand, if we use fractional impedances
Zi (s) = ki si ; i 2 R, then we can obtain Z (s) both integer and fractional [45].
The magnetic flux and the electrical charge, (t) and q(t), are related to the voltage and current, v (t) and i (t), by:
Zt Zt
(t) = v () d; q (t) = i () d: (4)
1 1
In linear circuits, the resistor, inductor and capacitor, R, L and C, follow the relations:
The “memristor” M is the element verifying the relation [46, 47, 48, 49, 50, 51]:
If we have a linear relationship between and q, then M(q) = M similarly to a resistance, since ddt = M dq
dt ,v=
Mi.
The generalization of the memristor concept to a larger class, the so-called “memristive systems”, is also pos-
sible [52, 53, 54, 16]. The charge-controlled memristor and flux-controlled memconductance are modeled by the
expressions:
= ˆ (q) ; q = q̂ () ; (7)
Zt Zt
(t) = q () d; (t) = () d; (9)
1 1
then we have [55]:
4
= ˆ () ; = ˆ (q) ; (10)
Similarly to what occurs with Mi (q) and Mv (), the elements C M () and L M (q) “remember” the flux and charge
previously applied.
These ideas support the so-called one-port higher order element, establishing a relation between v (t) and i (t), such
that: !
dm dn
v (t) = i (t) ; m; n 2 Z: (12)
dtm dtn
Linearizing expression (12) around some operating point Q = dm
;dn
dtm vQ dtn iQ on the element characteristic (12), we
obtain:
dn dm
v vQ = mQ i iQ ; (13)
dtn dtm
where mQ denotes the slope of the line tangent to the characteristic dm
dtm v
(t) = dn
dtn i
(t) at point Q.
In the frequency domain, expression (13) yields:
where
Z( j!) = ( j!)n m
mQ ; (15)
5
n
3
0 D L R m
-1 LM M C
-2 CM
-3
-3 -2 -1 0 1 2 3
Figure 1: Simplified Cartesian representation of the PTE of two-terminal IOE. The acronyms fR; L; D; C; M; L M ; C M g stand for fresistor, inductor,
frequency dependent negative conductance, capacitor, memristor, meminductor, memcapacitorg.
The classical PTE represents only IOE and, therefore, we are restricted to
2 Z. However, other non-planar
and non-Cartesian arrangements are possible for representing the IOE. If we apply the coordinate transformation
(m; n) ! ('; ; z), such that
' = (m n) ; = m + n; z = '; (16)
2
then we obtain the 3-dimensional PTE illustrated in Figure 2. With this representation, R is located at the center of
the spiral-like locus and the elements in the diagonals are represented at horizontal lines.
In another point of view, a closer look to the standard PTE reveals that the space in the middle of the grid lines is, in
fact, the locus for the FOE. Indeed, it is known the existence of fractional inductors and capacitors [57, 58, 59, 60, 61]
and, therefore, the generalization of the PTE to a “continuum” of FOE is the logical step to follow [62].
We adopt a set of 4 distances, fA ; C ; J ; S g, to measure the dissimilarity between pairs (Pn ; P p ) of objects with
real and imaginary components [63]. Therefore, the items are characterized by K
h
2 dimensional matrices Pn =
h i iT
[RefPn1 g ; RefPnK g]T ; [ImfPn1 g; ; ImfPnK g]T and P p = RefP p1 g; ; RefP pK g ; [ImfP p1 g; ; ImfP pK g]T ,
where Refg and Imfg stand for the real and imaginary parts. The distances are given by the expressions:
0 1
B
B X K K
X C
C
B
B C
C
B
B
B
B
B
f
Re Pnk gRefP pk g + ImfPnk gImfP pk g C
C
C
C
C
B
B C
C
B = k=1 C
A (Pn ; P p ) = arccos BBBB u
v
t
k 1
v
u
t
C
C
C
C
; (17)
B
B C
C
B
B XK XK C
C
B C
B
B
B
@ f
Re Pnk 2 g + ImfPnk g2 RefP pk g2 + Imf g
P pk CCCA
2
k =1 k=1
6
6
2
C
0 R
CM
LM
M
L
-2
D
-4
-6
4
2
2
0 0
-2 -2
-4
Figure 2: The 3-dimensional representation of the PTE of two-terminal IOE using the coordinate transformation (16). The acronyms
fR; L; D; C; M; LM ; C M g stand for fresistor, inductor, frequency dependent negative conductance, capacitor, memristor, meminductor,
memcapacitorg.
K
X jRefPnk g RefP pk gj X K
jImfPnk g ImfP pk gj
C (Pn ; P p ) = + ; (18)
k=1
jRefPnk gj + jRefP pk gj k=1 jImfPnk gj + jImfP pk gj
K
X 2
RefPnk g RefP pk g
k=1
J (Pn ; P p ) = K K K
+
X X X
RefPnk g2 + RefP pk g2 RefPnk gRefP pk g
k=1 k=1 k=1
(19)
K
X 2
ImfPnk g ImfP pk g
k =1
K K K
;
X X X
ImfPnk g2 + ImfP pk g2 ImfPnk gImfPnk g
k =1 k=1 k=1
K
X K
X
jRefPnk g RefP pk gj + jImfPnk g ImfP pk gj
k=1 k=1
S (Pn ; P p ) = K K
: (20)
X X
jRefPnk gj + jRefP pk gj + jImfPnk gj + jImfP pk gj
k=1 k=1
If the objects to be compared have no imaginary part, then the vectors Pn and P p are K 1 dimensional, and the
set fA ; C ; J ; S g corresponds to the standard fArccosine, Canberra, Jaccard, Sørenseng distances [64].
7
We must note that other distances are possible [64] and that several of them were also tested. By other words, we
are not restricted to the standard Cartesian concepts, neither for in the chart nor for the dierence measurements. It
is well known that the Cartesian perspective is a particular case of the Minkowski formulation and that this is just a
family of distances within a plethora of generalized expressions [64, 63]. However, further distances are not included
herein for sake of parsimony, since fA ; C ; J ; S g illustrate adequately the proposed concepts.
The MDS is a computational recursive method that provides dimensionality reduction and envisages to produce
a locus with clusters and, possibly, some data organization capable of being visualized and interpreted [65, 66, 67].
Given a set of N objects K-dimensional and a dissimilarity index, we calculate a N N matrix, = [np ], n; p =
1; : : : ; N, of object-to-object dissimilarities, such that np = pn and nn = 0. This information represents the input of
the visualization algorithm. The MDS represents the N objects by points in a W-dimensional space, with W < K, and
tries to reproduce the measured dissimilarities. The MDS iterates the estimate of point configuration for optimizing a
given fitness, achieving a matrix of distances ˆ = [ˆ np ], n; p = 1; : : : ; N, that approximates the original one = [np ].
A common fitness is the raw stress:
X n 1h
N X i2
S= ˆ np (np ) ; (21)
n=2 p=1
In this Section we generate several MDS representations both of linear and nonlinear FOE. Firstly, we describe the
FOE by their behavior either in the time or in the frequency domains. This information will represent the objects P,
that is, the FOE. Secondly, we use the resulting data for comparing the FOE and calculate the dissimilarity matrix
measured by means of a given distance function . Finally, we feed the data into the MDS for constructing the matrix
ˆ and the W-dimensional loci of FOE.
8
Let us consider N FOE of orders = f
n :
min
n
max ; n = 1; : : : ; N g. To each FOE we apply a collection
of sinusoidal signals xd;e (t) = Ae cos(!d t) with frequencies
= f!d : !min !d !max ; d = 1; : : : ; N! g. For
nonlinear systems the set of testing amplitudes is given by A = fAe : Amin Ae Amax ; e = 1; : : : ; NA g, but, obviously,
for the linear case we can use just one value (NA = 1). Then we compute the N! NA system outputs zd;e (t) =
h i
Ae !
dn cos !d t +
n 2 , where represents some kind of linear/nonlinear function, t = l td , l = 0; 1; : : : ; Nt 1
and td = N p !d (N 2
t 1)
, with td denoting the sampling period, Nt standing for the number of time samples, and N p
representing the number of periods of the signals (Figure 3).
Figure 3: Block diagram of a FOE where () is some linear/nonlinear function: input i(t) and output v(t) given by xd;e (t) = Ae cos (!d t) and
h i
zd;e (t) = Ae !dn cos !d t +
n 2 , respectively.
During the experiments some eect of truncating the series of
n values, that is, of limiting to
min and
max was
observed on the produced loci. Therefore, to reduce that eect, all experiments adopted some extra values at both
extremes that are not represented.
3.1. Time Domain Analysis and Visualization of Linear Fractional Order Elements
In this case we compare linear FOE in the time domain, meaning that we consider NA = 1. Therefore, after
collecting the N! outputs of the n-th FOE, we construct the K = Nt N! dimensional real-valued vectors Pn (t) =
z1;1 (t); : : : ; zN! ;1 (t) and calculate the distance matrices = [(Pn (t); P p (t))], n; p = 1; : : : ; N, where (Pn (t); P p (t))
denotes one distance of the set fA ; C ; J ; S g between the vectors Pn (t) and P p (t) given by expressions (17)-(20).
Finally, we process each matrix by means of the MDS for constructing the FOE loci.
Figure 4 depicts the 3-dimensional MDS loci of N = 721 linear FOE with order values spaced linearly in the
interval 10
n 10, when adopting N! = 40 test frequencies spaced logarithmically in the interval 10 1
!d 10 , Nt = 1000 time samples and N p = 5 periods. Several distinct amplitudes were tested numerically, but, as
1
expected, the FOE loci do not depend on this parameter. All other parameters were adjusted by successively increasing
their values until the loci are insensitive to changes. The markers represent the FOE and the colors vary with the FOE
order,
n , to enhance the visualization. In the loci (a) and (b) the IOE of the same category are connected by dashed
lines. Such lines are not included in (c) and (d), since for their good visualization we need to rotate the charts. For
all distances, we verify that the FOE form smooth patterns, exhibiting regularities that depend on the FOE categories.
Moreover, the representations do not follow the standard Cartesian arrangement and use eciently the 3-dimensional
visualization space.
Variations to the previous loci are possible to highlight specific aspects of the organization of the FOE and to
capture distinct information provided by the MDS computational scheme. These possibilities are illustrated in Figure
9
A C
0.3 7
8
1.5
0.2 4 3
59 -9
1 -5
z-MDS coordinate
8
1 6
z-MDS coordinate
4 -1 0.1 5
0.5 0 10
-4 9
-8 0 0 -1
0 2
-10 1
-6 -0.1
-2 -4 -9
-0.5 62
10 -8 -5
-3 -0.2 -2
-1 3
-7 -3
7 -10
-0.3
-1.5 0.4 -6
2 2 -7
0.2 0.4
1 1
0 0.2
0 0 0
-1 -1 -0.2
-0.2
y-MDS coordinate -2 -2 x-MDS coordinate y-MDS coordinate -0.4 -0.4 x-MDS coordinate
(a) (b)
J S
0.3 9
8
0.2
-1 -8 0.3
z-MDS coordinate
-2 5 -7
0.1 2 1
4 0.2 -9 3
-8 -10
z-MDS coordinate
0 2 0
0.1 -7 4
3
-0.1 -4
-5 0 -6 -1
5
-6 -3
-0.2 -0.1
-10 0 -5 6
-9 -2
10 1
-0.3 -0.2 -4 -3
0.4 6 7
0.6
10
7 -0.3
0.2 8 0.4
0.4 9
0 0.3 0.2 0.2
0.2
-0.2 0.1 0
0 0
-0.1 -0.2
y-MDS coordinate -0.4 -0.2 x-MDS coordinate
-0.3 y-MDS coordinate -0.4 -0.2
x-MDS coordinate
(c) (d)
Figure 4: The 3-dimensional loci of N = 721 linear FOE, characterized in the time domain by means of the distances: (a) A ; (b) C ; (c) J ; (d)
S . The markers represent the FOE and their color varies with the FOE order
n 2[ 10; 10]. The other parameters are N! = 40, !d 2 [10 1 ; 101 ],
NA = 1, Nt = 1000 and N p = 5. In the loci (a) and (b) the IOE of the same category are connected with dashed lines.
5, where we consider two MDS dimensions for the (x; y) coordinates, while the z coordinate is calculated by means
of radial basis interpolation (RBI) [68] of the FOE order
n . The thin-plate spline RBI function, ( ) = 2 log , is
considered, where the variable denotes the Euclidean distance between the points generated by the 2-dimensional
MDS. Nonetheless, we believe that the 3-dimensional visualization is more advantageous than the 2+1-dimensional
portrait. Therefore, for reducing length, in the follow-up we restrict to the richer visualization method.
3.2. Frequency Domain Analysis and Visualization of Linear Fractional Order Elements
For the n-th FOE, n = 1; : : : ; N, we convert the sinusoidal outputs zd;1 (t), d = 1; : : : ; N! , to the Fourier domain,
yielding F fzd;1 (t)g = Zd;1 ( j!), where ! = !d . We generate the N! 2 dimensional complex-valued matrix Pn (!) =
10
(a) (b)
(c) (d)
Figure 5: The 2+1-dimensional loci of N = 721 linear FOE, characterized in the time domain by means of the distances: (a) A ; (b) C ; (c) J ;
(d) S . The z coordinate of the loci is calculated by means of RBI based on the value of
n 2[ 10; 10] at each MDS (x; y) coordinate. The other
parameters are N! = 40, !d 2 [10 1 ; 101 ], NA = 1, Nt = 1000 and N p = 5
RefZd;1 ( j!)g; ImfZd;1 ( j!)g and calculate the dissimilarity matrices that feed the MDS algorithm and generate the
FOE loci.
Figure 6 depicts the 3-dimensional loci of the N = 721 linear FOE, characterized in the frequency domain by
means of the distances fA ; C ; J ; S g. The values of the parameters are identical to those adopted in the Subsection
3.1. For all distances we verify that the linear FOE loci do not depend on the amplitude of the sinusoidal inputs
and that the loci form smooth patterns with regularities that depend on the FOE categories. The charts are dierent
from those obtained in the time domain, but follow an identical logic, namely using a non-Cartesian arrangement in a
3-dimensional space.
11
A C
9 5 0.5
1 -5
-9
z-MDS coordinate
-8 1 -1
-4 8 3
4 -1
-5
0
0.5 7
z-MDS coordinate
0 0
-9
-2 4
-4 8
2
0 -6
2 6
-0.5 10
-10
6 -2 -8
10 -10
1
-3
-0.5
-7 -6 -3 1
5 9
3 0.5 -7
1
7
-1 0.5
1 0 0 0.5
0.5
0 -0.5 0
-0.5 y-MDS coordinate
-1 -1 x-MDS coordinate -0.5
y-MDS coordinate -0.5 x-MDS coordinate
(a) (b)
J S
0.5 -4
z-MDS coordinate
4 -8
8
0.3
0 9
7 10 8
6 -3
-4 0.2
5 -5 -3
-0.5 -2 1 -7
4 -6
z-MDS coordinate
-7 5
3 -10 -9 -8 -1 0.1
9
-9
-1 -5
2 -1
-0.5 1
0 3
7
-10
0 -0.1
-6
0 10
-0.2 -2 0.4
6 2
0.2
0.5 -0.3
0.4 0
x-MDS coordinate 0.6 0.8 0.2 -0.2
0.2 0.4 0
-0.2 0 -0.2
1 -0.6 -0.4 -0.4 x-MDS coordinate
-0.8 -0.4
y-MDS coordinate y-MDS coordinate
(c) (d)
Figure 6: The 3-dimensional loci of N = 721 linear FOE, characterized in the frequency domain by means of the distances: (a) A ; (b) C ; (c) J ; (d)
S . The markers represent the FOE and their color varies with the FOE order
n 2[ 10; 10]. The other parameters are N! = 40, !d 2 [10 1 ; 101 ],
NA = 1, Nt = 1000 and N p = 5. In the loci (a) and (b) the IOE of the same category are connected with dashed lines.
3.3. Time Domain Analysis and Visualization of Nonlinear Fractional Order Elements
In this case we compare nonlinear FOE in the time domain adopting for a cubic nonlinearity. We must note that
other nonlinearities [69] are possible and that several were tested. However, they are not included herein for sake of
parsimony, since this one illustrates well the proposed ideas.
For the n-th nonlinear FOE, n = 1; : : : ; N, we collect N! NA outputs, zd;e (t), where d = 1; : : : ; N! and e =
1; : : : ; NA . Then, for comparing the FOE, we construct the K = Nt N! NA dimensional real-valued vectors Pn (t) =
z1;1 (t); : : : ; z1;NA (t); : : : ; zN! ;1 (t); : : : ; zN! ;NA (t) . Finally, we calculate the distance matrices = [(Pn ; P p )], n; p =
1; : : : ; N, and apply the MDS numerical algorithm. Figure 7 depicts the 3-dimensional MDS loci of the N = 721
nonlinear FOE, characterized in the time domain by means of fA ; C ; J ; S g. The values of the parameters are
12
identical to those adopted in the previous Subsections, but for the nonlinear case we consider NA = 10 amplitudes of
the input signal spaced logarithmically in the interval 10 2
Ne 102 . Comparing Figures 7 and 4, we verify that
the loci generated with the distances fA ; S g do not vary, while those generated with fC ; J g vary considerably with
the presence of the nonlinearity. We verify again that we can adjust the characteristics of the loci, in this case the
sensitivity to the nonlinearity , by a judicious choice of the proper distance.
A C
1.5
0.3
1 4 -3
9 -2 -4
5
8 0.2
-1 -5
z-MDS coordinate
z-MDS coordinate
1 -1
0.5 -4 -5
-9 0.1
-8 0 -6
0
0 0 10
1 -7
-2 -6 -0.1 9
-0.5 -8
-3 -10 2 8
-7 -0.2 7
6 3
-1 7 3 -9
2 6
10
-0.3 4 5 -10
-1.5 0.4
2 0.4
1 2 0.2
1 0.2
0 0
0 0
-1 -1 -0.2
y-MDS coordinate -2 -2 y-MDS coordinate -0.2 -0.4
x-MDS coordinate x-MDS coordinate
(a) (b)
S
J
8 0.3
0.3
0.2
-8 -8
0.2
z-MDS coordinate
-2 -7
4 -9
0.1
-6
z-MDS coordinate
9 -10
1
0 2 -1 5 -7 0.1 -5 2
3
-3 -4
-0.1 -5 0
-9 -4
1 0 3
-0.2 7
-6 -10 -3
-0.3 0 4
-1
10 -0.1
0.4 -2
5
0.2 10
6 6
0.6
0.3 -0.2 9
0 7 0.4
0.2 0.3
0.1 0.2 8 0.2
0.1
-0.2 0 0 0
-0.1 -0.1
-0.2 -0.2
y-MDS coordinate -0.4 -0.2 -0.3 x-MDS coordinate
-0.3 x-MDS coordinate y-MDS coordinate
(c) (d)
Figure 7: The 3-dimensional loci of N = 721 nonlinear ( of degree 3) FOE, characterized in the time domain by means of the distances: (a) A ;
(b) C ; (c) J ; (d) S . The markers represent the FOE and their color varies with the FOE order
n 2[ 10; 10]. The other parameters are N! = 40,
!d 2 [10 1 ; 101 ], NA = 10, Ne 2 [10 2 ; 102 ], Nt = 1000 and N p = 5. In the loci (a) and (c) the IOE of the same category are connected with dashed
lines.
3.4. Frequency Domain Analysis and Visualization of Nonlinear Fractional Order Elements
We compare N = 721 nonlinear FOE in the frequency domain adopting for the cubic nonlinearity.
13
In a first phase, for the n-th FOE, n = 1; : : : ; N, we convert the N! NA outputs zd;e (t) to the Fourier domain,
yielding F fzd;e (t)g = Zd;e ( j!), noting that for a cubic nonlinearity zd;e (t) has the first and third harmonics. In a
second phase, for comparing the FOE, we generate the (2 N! NA ) 2 dimensional complex-valued array Pn (!) =
RefZd;e ( j!)g; ImfZd;e ( j!)g . Finally, we calculate the dissimilarity matrices , and generate the MDS FOE loci.
Figure 8 depicts the 3-dimensional MDS loci of the N = 721 nonlinear FOE, characterized in the frequency
domain. All values of the parameters are kept unchanged from the previous Subsections.
A
C
1.5
-4 0.5
1 8 4 -8
-1
3
z-MDS coordinate
-5
z-MDS coordinate
7
0.5 5
9
1 0
-2
-5 -9 0
-92 4 8
0 -4
-1 -6
0 6 10
-7
-10
3 -3 1-8
-3 9
7 5
-0.5
-7
2
-2
-0.5
-10 6
0.5
-1 10
-1 -6 0.5
0 1.5 0
1 0.5 1 0
-0.5 0
2 -1.5 -1
x-MDS coordinate y-MDS coordinate y-MDS coordinate -0.5 -0.5 x-MDS coordinate
(a) (b)
S
J 0.3
0.2
-1
1 3
z-MDS coordinate
0.1 -5
0.5 7
z-MDS coordinate
0
-9
-4 0
-2
0 -0.1 2 4
2 -58 0 -6
10 65 -10 8
-93 7 1
-1 4
-6 -8 -0.2 -8 -10 6 10
-0.5 9 -3 -7
-4
-2
-3
-0.3
0.4 -7 1
-1
5 9
1
0.2
0.5 0.8
0.6 0
0 0.4
-0.5 0.2 y-MDS coordinate 0.2 0.3
-0.2 0 0.1
0 -0.2 -0.1
y-MDS coordinate -0.3
-1 -0.2 x-MDS coordinate x-MDS coordinate
(c) (d)
Figure 8: The 3-dimensional loci of N = 721 nonlinear ( of degree 3) FOE, characterized in the frequency domain by means of the distances:
(a) A ; (b) C ; (c) J ; (d) S . The markers represent the FOE and their color varies with the FOE order
2[ 10; 10]. The other parameters are
N! = 40, !d 2 [10 1 ; 101 ], NA = 10, Ne 2 [10 2 ; 102 ], Nt = 1000 and N p = 5. In the loci (a) and (d) the IOE of the same category are connected
with dashed lines.
14
4. Procrustes Analysis and Visualization of Nonlinear Fractional Order Elements
In this Section, we compare the loci obtained with dierent nonlinearities by means of Procrustes analysis [70, 71,
72, 73]. The Procrustes analysis takes a collection of loci and transforms them for obtaining the “best” superposition.
The algorithm performs four iterative numerical steps: (i) the user chooses a reference locus (by selecting one of the
available instances); (ii) superimposes all other loci into the current reference by means of linear transformations,
namely translation, reflection, orthogonal rotation and scaling; (iii) computes the mean form of the current set of
superimposed loci; (iv) compares the distance between the mean and the reference instances to a given threshold
value and, if above, sets the reference to the mean form and continues to step (ii). The result is a global representation
of all loci that best conforms them.
Figures 9 and 10 depict three superimposed 3-dimensional MDS loci of N = 721 FOE (using Procrustes), char-
acterized in the time and frequency domains, respectively. Besides the linear case we adopt power law nonlinearities
of degree 3 and 5. The values of all parameters are identical to those used in the previous Section. We verify an
evolution of the loci with n, demonstrating the sensitivity of the technique to the nonlinearity. For the distances A
and S this evolution is smooth, while for C and J we obtain a sharp transition between the linear and the nonlinear
cases, when the FOE are characterized in the time and the frequency domains, respectively. Therefore, we verify that
we can extend the construction of the MDS loci and their comparison to other types of nonlinearites.
5. Conclusions
This paper used clustering and information visualization techniques to organize and map FOE accordingly to their
characteristics. The new representation generalizes the concept of PTE, revealing that the integer order cases are
just a limited number of cases in the FOE “continuum”. The use of MDS allows exploring the 3-dimensional space
for representation and the adoption of distinct measures, so that users can choose the one fitting better their needs.
The technique is eective both in the time and frequency domains and can be extended from linear to nonlinear
elements. Moreover, the study provides a complementary perspective in the on-going discussion about the properties
of the memristor and fractional-order elements. Indeed, a new form of representation, based in distinct domains and
distances, may shed further light into possible similarities or dissimilarities between elements.
In summary, this paper did not intend to give responses to a variety of possible questions such as if there are
finite boundaries, or not, to the Chua’s PTE, or what is the physical meaning of fractional elements. The study shows
that we are often conditioned by representations methods that can be bettered by modern computer-based information
visualization algorithms. Furthermore, in the scope of the new visualization methods, the use of Cartesian concepts,
namely for graphical representations and for distance (or dierence) assessment, can be outperformed by a careful
selection of the formulation that fits better a specific application.
15
A C
Linear Linear
Nonlinear ( of degree 3) Nonlinear ( of degree 3)
Nonlinear ( of degree 5) Nonlinear ( of degree 5)
1.5
0.4
1
z-MDS coordinate
z-MDS coordinate
0.2
0.5
0 0
-0.5
-0.2
-1
-0.4
-1.5
2 -0.4 -0.4
2
1 1 -0.2 -0.2
0 0 0 0
-1 -1 0.2 0.2
y-MDS coordinate -2 -2 x-MDS coordinate y-MDS coordinate 0.4 0.4 x-MDS coordinate
(a) (b)
J S
Linear Linear
Nonlinear ( of degree 3) Nonlinear ( of degree 3)
Nonlinear ( of degree 5) Nonlinear ( of degree 5)
0.4 0.4
z-MDS coordinate
z-MDS coordinate
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4
0.5 0.4
0.4 1
0.2
0 0.2 0.5
0
0
-0.2 -0.2 0
y-MDS coordinate -0.5 -0.4 x-MDS coordinate y-MDS coordinate -0.4 -0.5 x-MDS coordinate
(c) (d)
Figure 9: Three superimposed 3-dimensional loci of N = 721 FOE (using Procrustes), characterized in the time domain by means of the distances:
(a) A ; (b) C ; (c) J ; (d) S . The functions of degree 1 (linear case), 3 and 5 are adopted.
References
16
A C
Linear Linear
Nonlinear ( of degree 3) 0.6 Nonlinear ( of degree 3)
1.5 Nonlinear ( of degree 5) Nonlinear ( of degree 5)
0.4
1
z-MDS coordinate
z-MDS coordinate
0.2
0.5
0
0
-0.2
-0.5
-0.4
-1
-0.6
-1.5 1
2
0.5
2 0.6
0 1 0 0.4
0 0 0.2
-1 -0.4 -0.2
-2 -2 y-MDS coordinate -0.5 -0.6
y-MDS coordinate x-MDS coordinate x-MDS coordinate
(a) (b)
J S
Linear
Linear
1 Nonlinear ( of degree 3)
Nonlinear ( of degree 3)
Nonlinear ( of degree 5)
Nonlinear ( of degree 5)
0.4
z-MDS coordinate
0.5
z-MDS coordinate
0.2
0
0
-0.5
-0.2
-1
1 -0.4
0.5 0.5
0 0.4
1 0 0.2
0.8
-0.5 0.6 0
0.4
0.2 -0.2
y-MDS coordinate -1 0
-0.2 y-MDS coordinate -0.5 -0.4
x-MDS coordinate x-MDS coordinate
(c) (d)
Figure 10: Three superimposed 3-dimensional loci of N = 721 FOE (using Procrustes), characterized in the frequency domain by means of the
distances: (a) A ; (b) C ; (c) J ; (d) S . The functions of degree 1 (linear case), 3 and 5 are adopted.
[8] J. T. Machado, A. M. Lopes, Fractional electromagnetics, in: V. E. Tarasov (Ed.), Handbook of Fractional Calculus with Applications:
Applications in Physics, Part B, Vol. 5, De Gruyter, Berlin, pp. 1–23.
[9] D. Valério, M. Ortigueira, J. T. Machado, A. M. Lopes, Continuous-time fractional linear systems: Steady-state behaviour, in: I. Petráš (Ed.),
Handbook of Fractional Calculus with Applications: Applications in Engineering, Life and Social Sciences, Part A, Vol. 6, De Gruyter,
Berlin, pp. 149–174.
[10] A. M. Lopes, J. A. T. Machado, Fractional-order modeling of electro-impedance spectroscopy information, in: D. Baleanu, A. M. Lopes
(Eds.), Handbook of Fractional Calculus with Applications: Applications in Engineering, Life and Social Sciences, Part A, Vol. 7, De
Gruyter, Berlin, pp. 21–41.
[11] B. Parsa, A. Dabiri, J. A. T. Machado, Application of variable order fractional calculus in solid mechanics, in: D. Baleanu, A. M. Lopes (Eds.),
Handbook of Fractional Calculus with Applications: Applications in Engineering, Life and Social Sciences, Part A, Vol. 7, De Gruyter, Berlin,
pp. 207–224.
[12] J. T. Machado, A. M. Lopes, Relative fractional dynamics of stock markets, Nonlinear Dynamics 86 (3) (2016) 1613–1619.
17
[13] H. Paynter, Analysis and design of engineering systems: Class notes for M.I.T. course 2.751, M.I.T. Press, Boston, 1961.
[14] L. O. Chua, Memristor - the missing circuit element, IEEE Transactions on Circuit Theory 18 (2) (1971) 507–519.
[15] D. B. Dmitri B. Strukov, G. S. Snider, D. R. Stewart, R. S. Williams, The missing memristor found, Nature 97 (2008) 80–83.
[16] M. Di Ventra, Y. V. Pershin, L. O. Chua, Circuits elements with memory: Memristors, memcapacitors and meminductors, Proceedings of the
IEEE 97 (10) (2009) 1717–1724.
[17] M.-S. Abdelouahab, R. Lozi, L. Chua, Memfractance: A mathematical paradigm for circuit elements with memory, International Journal of
Bifurcation and Chaos 24 (09) (2014) 1430023.
[18] F. Z. Wang, L. Shi, H. Wu, N. Helian, L. O. Chua, Fractional memristor, Applied Physics Letters 111 (24) (2017) 243502.
[19] I. Abraham, The case for rejecting the memristor as a fundamental circuit element, Scientific Reports 8 (1) (2018) 10972.
[20] F. Z. Wang, A triangular periodic table of elementary circuit elements, IEEE Transactions on Circuits and Systems I: Regular Papers 60 (3)
(2013) 616–623.
[21] C. Ware, Information visualization: Perception for design, Elsevier, Waltham, 2012.
[22] R. Spence, Information visualization: An introduction, Vol. 1, Springer, Cham, 2001.
[23] L. Van Der Maaten, E. Postma, J. Van den Herik, Dimensionality reduction: A comparative review, J Mach Learn Res 10 (2009) 66–71.
[24] A. K. Jain, M. N. Murty, P. J. Flynn, Data clustering: A review, ACM computing surveys (CSUR) 31 (3) (1999) 264–323.
[25] I. Podlubny, Fractional dierential equations, Volume 198: An Introduction to Fractional Derivatives, Fractional Dierential Equations, to
Methods of Their Solution, Mathematics in Science and Engineering, Academic Press, San Diego, 1998.
[26] A. Kilbas, H. Srivastava, J. Trujillo, Theory and applications of fractional dierential equations, Vol. 204, North-Holland mathematics studies,
Elsevier, Amsterdam, 2006.
[27] A. Oustaloup, Systèmes asservis linéaires d’ordre fractionnaire: Théorie et pratique, Masson, Paris, 1983.
[28] J. A. T. Machado, Analysis and design of fractional-order digital control systems, Systems Analysis, Modelling, Simulation 27 (2-3) (1997)
107–122.
[29] V. E. Tarasov, Fractional dynamics: Applications of fractional calculus to dynamics of particles, fields and media, Springer, Beijing, Heidel-
berg, 2010.
[30] F. Mainardi, Fractional Calculus and Waves in Linear Viscoelasticity: An Introduction to Mathematical Models, Imperial College Press,
London, 2010.
[31] J. Gómez-Aguilar, H. Yépez-Martı́nez, R. Escobar-Jiménez, C. Astorga-Zaragoza, J. Reyes-Reyes, Analytical and numerical solutions of
electrical circuits described by fractional derivatives, Applied Mathematical Modelling 40 (21-22) (2016) 9079–9094.
[32] J. F. Gómez-Aguilar, B. Dumitru, Fractional transmission line with losses, Zeitschrift für Naturforschung A 69 (10-11) (2014) 539–546.
[33] C. Ionescu, A. Lopes, D. Copot, J. T. Machado, J. Bates, The role of fractional calculus in modeling biological phenomena: A review,
Communications in Nonlinear Science and Numerical Simulation 51 (2017) 141–159.
[34] M. D. Ortigueira, J. T. Machado, F. J. V. Coito, G. Bengochea, Discrete-time fractional signals and systems, in: D. Baleanu, A. M. Lopes
(Eds.), Handbook of Fractional Calculus with Applications: Applications in Engineering, Life and Social Sciences, Part B, Vol. 8, De Gruyter,
Berlin, pp. 207–224.
[35] M. D. Ortigueira, J. T. Machado, What is a fractional derivative?, Journal of computational Physics 293 (2015) 4–13.
[36] G. S. Teodoro, J. T. Machado, E. C. De Oliveira, A review of definitions of fractional derivatives and other operators, Journal of Computational
Physics 388 (2019) 195–208.
[37] D. Valério, J. J. Trujillo, M. Rivero, J. T. Machado, D. Baleanu, Fractional calculus: A survey of useful formulas, The European Physical
Journal Special Topics 222 (8) (2013) 1827–1846.
[38] L. T. Bruton, Network transfer functions using the concept of frequency dependent negative resistance, IEEE Transactions on Circuit Theory
16 (3) (1969) 406–408.
[39] A. Antoniou, Novel RC-active-network synthesis using generalized immittance converters, IEEE Transactions on Circuit Theory 17 (2)
(1970) 212–217.
18
[40] A. M. Soliman, R. A. Saad, Two new families of floating fdnr circuits, Journal of Electrical and Computer Engineering 2010 (Article ID
563761) (2010) 7 pages.
[41] C. Psychalinos, K. Pal, S. Vlassis, A floating generalized impedance converter with current feedback operational amplifiers, AEU-
International Journal of Electronics and Communications 62 (2) (2008) 81–85.
[42] R. Senani, D. Bhaskar, Versatile voltage-controlled impedance configuration, IEE Proceedings-Circuits, Devices and Systems 141 (5) (1994)
414–416.
[43] K. Pal, Novel FDNC simulation using current conveyors, Electronics Letters 16 (16) (1980) 639–641.
[44] A. S. Elwakil, M. P. Kennedy, Chaotic oscillator configuration using a frequency dependent negative resistor, International Journal of Circuit
Theory and Applications 28 (1) (2000) 69–76.
[45] S. Kapoulea, G. Tsirimokou, C. Psychalinos, A. S. Elwakil, Generalized fully adjustable structure for emulating fractional-order capacitors
and inductors of orders less than two, Circuits, Systems, and Signal Processing (2019) 1–18.
[46] V. Mladenov, Advanced Memristor Modeling: Memristor Circuits and Networks, MDPI, 2019.
[47] D. Hajtó, Á. Rák, G. Cserey, Robust Memristor Networks for Neuromorphic Computation Applications, Materials 12 (21) (2019) 3573.
[48] T. V. Nguyen, K. V. Pham, K.-S. Min, Hybrid Circuit of Memristor and Complementary Metal-Oxide-Semiconductor for Defect-Tolerant
Spatial Pooling with Boost-Factor Adjustment, Materials 12 (13) (2019) 2122.
[49] J. Ordonez-Miranda, Y. Ezzahri, J. A. Tiburcio-Moreno, K. Joulain, J. Drevillon, Radiative Thermal Memristor, Physical Review Letters
123 (2) (2019) 025901.
[50] M. S. Semary, M. E. Fouda, H. N. Hassan, A. G. Radwan, Realization of fractional-order capacitor based on passive symmetric network,
Journal of Advanced Research 18 (2019) 147–159.
[51] J. Wu, G. Wang, H. H.-C. Iu, Y. Shen, W. Zhou, A Nonvolatile Fractional Order Memristor Model and Its Complex Dynamics, Entropy
21 (10) (2019) 955.
[52] L. O. Chua, S. M. Kang, Memristive devices and systems, Proceedings of the IEEE 64 (2) (1976) 209–223.
[53] L. O. Chua, S. M. Kang, Device modeling via basic nonlinear circuits elements, IEEE Transactions on Circuits and Systems 27 (11) (1980)
1014–1044.
[54] L. O. Chua, Nonlinear circuit foundations for nanodevices, Part I: The four-element torus, Proceedings of the IEEE 91 (11) (2003) 1830–1859.
[55] D. Jeltsema, A. Dòria-Cerezo, Port-Hamiltonian formulation of systems with memory, Proceedings of the IEEE 100 (6) (2012) 1928–1937.
[56] Z. Biolek, D. Biolek, V. Biolkova, Lagrangian for Circuits with Higher-Order Elements, Entropy 21 (11) (2019) 1059.
[57] S. Westerlund, Dead Matter has Memory, Causal Consulting, Kalmar, Sweden, 2002.
[58] G. W. Bohannan, Analog realization of a fractional control element - revisited, in: Proc. of the 41st IEEE Int. Conf. on Decision and Control,
Tutorial Workshop 2: Fractional Calculus Applications in Automatic Control and Robotics, Las Vegas, USA, 2002.
[59] K. Biswas, S. Sen, P. K. Dutta, Realization of a constant phase element and its performance study in a dierentiator circuit, IEEE Transactions
on Circuits and Systems II: Express Briefs 53 (9) (2006) 802–806.
[60] J. A. Tenreiro Machado, A. M. Galhano, Fractional order inductive phenomena based on the skin eect, Nonlinear Dynamics 68 (1-2) (2012)
107–115.
[61] A. G. Radwan, A. Soliman, A. S. Elwakil, A. Sedeek, On the stability of linear systems with fractional-order elements, Chaos, Solitons &
Fractals 40 (5) (2009) 2317–2328.
[62] J. T. Machado, Fractional generalization of memristor and higher order elements, Communications in Nonlinear Science and Numerical
Simulation 18 (2) (2013) 264–275.
[63] S.-H. Cha, Comprehensive survey on distance/similarity measures between probability density functions, International Journal of Mathemat-
ical Models and Methods in Applied Sciences 1 (4) (2007) 300–307.
[64] M. M. Deza, E. Deza, Encyclopedia of Distances, Springer-Verlag, Berlin Heidelberg, 2009.
[65] N. Saeed, H. Nam, M. I. Ul Haq, D. B. Muhammad Saqib, A survey on multidimensional scaling, ACM Computing Surveys (CSUR) 51 (3)
(2018) 47.
19
[66] J. T. Machado, A. M. Lopes, Multidimensional scaling analysis of soccer dynamics, Applied Mathematical Modelling 45 (2017) 642–652.
[67] J. T. Machado, A. M. Lopes, A fractional perspective on the trajectory control of redundant and hyper-redundant robot manipulators, Applied
Mathematical Modelling 46 (2017) 716–726.
[68] J. C. Carr, W. R. Fright, R. K. Beatson, Surface interpolation with radial basis functions for medical imaging, IEEE Transactions on Medical
Imaging 16 (1) (1997) 96–107.
[69] J. Machado, Visualizing non-linear control system performance by means of multidimensional scaling, Journal of Computational and Non-
linear Dynamics 8 (4) (2013) 041017.
[70] F. L. Bookstein, Landmark methods for forms without landmarks: Morphometrics of group dierences in outline shape, Medical Image
Analysis 1 (3) (1997) 225–243.
[71] J. C. Gower, G. B. Dijksterhuis, Procrustes problems, Vol. 3, Oxford University Press, Oxford, 2004.
[72] A. M. Lopes, J. Tenreiro Machado, A. M. Galhano, Multidimensional scaling visualization using parametric entropy, International Journal of
Bifurcation and Chaos 25 (14) (2015) 1540017.
[73] J. Tenreiro Machado, A. M. Lopes, A. M. Galhano, Multidimensional scaling visualization using parametric similarity indices, Entropy 17 (4)
(2015) 1775–1794.
20
Compliance with Ethics Requirements
This article does not contain any studies with human or animal subjects
Corresponding author