Applied Analysis Optimization and Soft Computing ICNAAO 2021 Varanasi India December 21 23
Applied Analysis Optimization and Soft Computing ICNAAO 2021 Varanasi India December 21 23
Applied Analysis Optimization and Soft Computing ICNAAO 2021 Varanasi India December 21 23
Applied Analysis,
Optimization
and Soft
Computing
ICNAAO-2021, Varanasi, India,
December 21–23
Springer Proceedings in Mathematics &
Statistics
Volume 419
This book series features volumes composed of selected contributions from
workshops and conferences in all areas of current research in mathematics and
statistics, including data science, operations research and optimization. In addition
to an overall evaluation of the interest, scientific quality, and timeliness of each
proposal at the hands of the publisher, individual contributions are all refereed to the
high quality standards of leading journals in the field. Thus, this series provides the
research community with well-edited, authoritative reports on developments in the
most exciting areas of mathematical and statistical research today.
Tanmoy Som · Debdas Ghosh · Oscar Castillo ·
Adrian Petrusel · Dayaram Sahu
Editors
Applied Analysis,
Optimization and Soft
Computing
ICNAAO-2021, Varanasi, India,
December 21–23
Editors
Tanmoy Som Debdas Ghosh
Department of Mathematical Sciences Department of Mathematical Sciences
Indian Institute of Technology (BHU) Indian Institute of Technology (BHU)
Varanasi, Uttar Pradesh, India Varanasi, Uttar Pradesh, India
© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature
Singapore Pte Ltd. 2023
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse
of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721,
Singapore
Contents
Fractals
Clifford-Valued Fractal Interpolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
Peter R. Massopust
Optimal Quantizers for a Nonuniform Distribution on a Sierpiński
Carpet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
Mrinal Kanti Roychowdhury
Fractal Dimension for a Class of Complex-Valued Fractal
Interpolation Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
Manuj Verma, Amit Priyadarshi, and Saurabh Verma
A Note on Complex-Valued Fractal Functions on the Sierpiński
Gasket . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
V. Agrawal and T. Som
Dimensional Analysis of Mixed Riemann–Liouville Fractional
Integral of Vector-Valued Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
Megha Pandey, Tanmoy Som, and Saurabh Verma
v
vi Contents
Mathematical Modeling
A Multi-strain Model for COVID-19 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
Samiran Ghosh and Malay Banerjee
Effect of Nonlinear Prey Refuge on Predator–Prey Dynamics . . . . . . . . . . 143
Shilpa Samaddar, Mausumi Dhar, and Paritosh Bhattacharya
Effects of Magnetic Field and Thermal Conductivity Variance
on Thermal Excitation Developed by Laser Pulses and Thermal
Shock . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
Rakhi Tiwari
ix
x About the Editors
been awarded the Outstanding Potential for Excellence in Research and Academics
Award (2014) from BITS-Pilani (Hyderabad Campus), where he worked as Assistant
Professor of mathematics, for the period June 2014–June 2016.
His broad research interest includes optimization theory, fuzzy geometry, compu-
tational multiobjective optimization, and robust optimization. With more than
40 papers and 12 conference papers, he has published five papers on fuzzy
geometrical ideas on plane and space. He has authored/edited several books: An
Introduction to Analytical Fuzzy Plane Geometry, Mathematics and Computing
(Springer), and A Primer on Interval Optimization (all with Springer). Dr Ghosh
has completed a research project entitled “On Characterizing and Obtaining the
Complete Efficient Solution Set of an Interval Optimization Problem under a D-
Dominance and a Variable Dominance Structure”, funded by Science and Engi-
neering Research Board, India. He is now handling a research project entitled
“On Developing Polynomial-time Interior-Point Methods for Robust Multiobjec-
tive Convex Optimization Problems”, funded by Science and Engineering Research
Board, India.
ICFPTA-2019 in China (July 2019), and ICTP-2002 in Italy. He has visited for
academic works IISc Bangalore (2002), India; South Korea (2008), Taiwan (2018);
Romania (2018); Guru Ghasidas University, Chhattisgarh, India (2018); and Jamia
Millia Islamia, New Delhi (2019).
Fixed Point Theory
Large Contractions and Surjectivity
in Banach Spaces
Fi x(t) := {x ∈ X |x = t (x)}
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 3
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_1
4 M. Moga and A. Petruşel
Definition 2 (see, e.g., Rus [10]) Let (X, · ) be a normed space. An operator
t : X → X is called quasi-bounded if there exist two numbers a, b ∈]0, ∞[ such
that
t (x) ≤ ax + b, for all x ∈ X. (1)
|t| = inf {a > 0| there exists b > 0 such that the relation (1) holds} .
Definition 3 (see, e.g., Rus [10]) Let (X, · ) be a normed space and t : X → X
be a quasi-bounded operator. If the quasi-norm of t is strictly less than one, then t is
called norm-contraction.
for the field 1 X − T generated by the multivalued operator T are given. The results
of this paper extend and complement fixed-point and surjectivity theorems (see [8,
9, 11]), as well as other related results (see lemma in [1]), in the recent literature.
In this section, we will prove the property that the field 1 X − t is surjective, where
t is a single-valued large contraction. The proof is based on T.A. Burton’s large
contraction principle [1] and a fixed-point theorem of Granas [4].
Theorem 1 (Granas [4]) Let (X, · ) be a Banach space and g : X → X be a com-
plete continuous operator. If, additionally, g is a norm-contraction, then Fi x(g) = ∅.
The following notion was introduced by T.A. Burton in 1996, in the frame of a
metric space. We recall the definition in the context of a Banach space.
Definition 4 (Burton [1]) Let (X, · ) be a normed space. An operator t : X → X
is said to be a large contraction if
1. t is contractive, i.e.,
Jachymski in [7] noted that the contractive condition in the above definition can
be avoided and, as a consequence, an operator t : X → X is a large contraction if,
for each ε > 0, there exists δ(ε) ∈]0, 1[ such that
The following example is related to some notions and results from the paper of
Burton and Purnaras [2].
Let K be a closed interval of the real axis and SK be the set of continuous functions
defined on K . Then (SK , · ) is a normed space, where · is the sup-norm. Let
g : [a, b] → R be a given function. We consider the subspace M ⊂ SK defined by
M := {x ∈ SK : a ≤ x(s) ≤ b, s ∈ K } . (2)
t (x) − t (y)
0< < 2, x, y ∈ I, x = y. (4)
x−y
e x (x 2 − x + 1)
t (x) := , x ∈ [0, 1] := I.
2e
Large Contractions and Surjectivity in Banach Spaces 7
It follows that t satisfies the relation (4). From Theorem 3, we get that the function
e x (x 2 − x + 1)
h(x) := x − , x ∈ [0, 1]
2e
is a large contraction on I . We observe that h is not a contraction on I since
xe x (1 + x)
h (x) = 1 − and h (0) = 1.
2e
We will present now another relevant example of large contraction, using the
concept of Meir–Keeler operator. Recall that if (X, · ) is a normed space, then
t : X → X is called a Meir–Keeler operator if for every ε > 0 there is δ > 0, such
that
x, y ∈ X, ε ≤ x − y < ε + δ ⇒ t (x) − t (y) < ε.
It is easy to see that any Meir–Keeler operator is contractive. Moreover, the following
result of Suzuki is well known.
By the above considerations, one can conclude that any self-Meir–Keeler operator
on a nonempty and convex subset of a Banach space is a large contraction.
The first main result of this section is the following surjectivity theorem.
t (x) ≤ t (
B(u; r )).
(b) Let x ∈ X \
B(u; r ). Then, we have that
Thus, we have
for each x ∈ X .
Thus, t is a norm-contraction.
To prove that 1 X − t is surjective we have to show that for any y ∈ X exists x ∈ X
such that (1 X − t)(x) = y. Thus, we should prove that, for every y ∈ X , the equation
x = t (x) + y has at least one solution x ∈ X . For the above conclusion, it is sufficient
to prove that for each y ∈ X , the set Fi x(g y ) is nonempty, where g y : X → X is
given by g y (x) = t (x) + y.
Let y ∈ X . Then, because t is complete continuous, it follows that g y is complete
continuous. Moreover, because t is norm-contraction, we immediately obtain that g y
is a norm-contraction, too. Thus, from Theorem 1, we have that Fi x(g y ) = ∅. Thus,
the operator 1 X − t is surjective and the proof is complete.
Large Contractions and Surjectivity in Banach Spaces 9
Let (X, · ) be a normed space. We recall first some necessary notations and notions,
which will help us prove the main result of the section (see, e.g., [9, 11]).
(1) The distance from a point a ∈ X to a set B ∈ P(X )
|T | = inf {m > 0| there exists M > 0 such that the relation (5) holds} .
If the quasi-norm of T is less than one (i.e., |T | < 1), then T is called a multivalued
norm-contraction. In the above setting, we will denote
Theorem 6 (Iannacci [6]) Let (X, · ) be a Banach space and let T : X → Pcv (X )
be a multivalued a completely continuous operator. Suppose that T is multivalued
norm-contraction. Then, the field 1 X − T generated by T is surjective.
Recall also that, in the same setting, a multivalued operator T : X → P(X ) is
called contractive (see, e.g., [13]) if
T (x) ≤ T (
B(u; r )).
Now, we can take into account the fact that T is a multivalued large contraction.
Then, for ε > 0, there exists δ ∈ (0, 1) such that
As a consequence, we get
Concerning the above open problem, we can prove the following partial answer.
(iii) there exist K > 0, x ∈ X and a sequence {xn }n∈N of Picard type iterates for
T starting from x0 := x (i.e., xn+1 ∈ T (xn ), for every n ∈ N), such that d(x, xn ) ≤ K ,
for every n ∈ N∗ .
Then:
(a) T has at least one fixed point in X ;
(b) if, instead of (i), we suppose that
Proof (a) For x ∈ X , we consider the sequence {xn }n∈N of Picard-type iterates
for T starting from x0 := x. Suppose that this sequence is not Cauchy. Then,
there exist ε > 0, 0 < Nk ∞ and there exists m k , n k > Nk . m k > n k such
that d(xm k , xn k ) ≥ ε. Then, by (i), we get
Thus, for these distances, using (ii) and then (iii), we obtain
(b) The uniqueness of the fixed point follows by (i)’. Indeed, if x ∗ and x̃ are two
distinct fixed points of T , then we can write
a contradiction.
References
1. Burton, T.-A.: Integral equations, implicit functions, and fixed points. Proc. Am. Math. Soc.
124(8), 2383–2390 (1996)
2. Burton, T.-A., Purnaras, I.-K.: Necessary and sufficient conditions for large contractions in
fixed point theory, Electron. J. Qual. Theory Differ. Equ. 94, 1–24 (2019)
3. Chen, Y.-Z.: Inhomogeneous iterates of contraction mappings and nonlinear ergodic theorems.
Nonlinear Anal. 39(1), 1–10 (2000)
4. Granas, A.: On a certain class of nonlinear mappings in Banach space. Bull. Acad. Pol. Sci. 9,
867–871 (1957)
5. Granas, A., Dugundji, J.: Fixed Point Theory. Springer, New York (2003)
6. Iannacci, R.: The spectrum for nonlinear multi-valued maps via approximations, Boll. Un. Mat.
Ital. 15-B, 527–545 (1978)
7. Jachymski, J, Jóźwik, I.: Nonlinear contractive conditions: a comparison and related problems.
In: Fixed Point Theory and its Applications, Banach Center Publ. vol. 77, pp. 123–146 (2007)
8. Moga, M.: Meir-Keeler operators and applications to surjectivity theorems. J. Nonlinear and
Convex Anal. 23(3), 625–634 (2022)
9. Petruşel, G.: Generalized multivalued contractions which are quasi-bounded. Demonstratio
Math. 40, 639–648 (2007)
10. Rus, I.-A.: Normcontraction mappings outside a bounded set, Itinerant Seminar on Functional
Equations Approximation and Convexity, Cluj-Napoca, pp. 257–260 (1986)
11. Rus, I.-A., Petruşel, A., Petruşel, G.: Fixed point theorems for set-valued Y-contractions, In:
Fixed Point Theory and its Applications, Banach Center Publ. vol. 77, pp. 227–237 (2007)
12. Suzuki, T.: Moudafi’s viscosity approximations with Meir-Keeler contractions. J. Math. Anal.
Appl. 325, 342–352 (2007)
13. Xu, H.K.: Metric fixed point theory for multivalued mappings. Diss. Math. vol. 389 39 pp.
(2000)
On Hick’s Contraction Using a Control
Function
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 13
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_2
14 V. Tiwari et al.
different from the form considered by Sehgal et al. [2]. This contraction is called
Hick’s contraction or C-contraction which along with its several modifications and
generalizations have been introduces in various research papers.
In probabilistic fixed point theory, a new domain of study started with the intro-
duction of control functions by Choudhury et al. [12]. This parallels a corresponding
development in metric spaces which was initiated by Khan et al. [12] and elaborated
through several works [5–7]. Other types of control functions have been used by
several authors like Ciric [8], Fang [12], etc. Particularly, the result of Fang [12] is a
culmination of a trend of development in this life.
In the present work, we use another control function for obtaining a generalized
C-contraction result in probabilistic metric spaces. The control function is different
from that of Fang [12] and its use has warranted a new method of proof of the
corresponding fixed point result. In the following, we dwell upon some aspects of
the probabilistic metric space and some other concepts which are required for further
discussion.
Definition 4 ([1, 3]) Let be a t-norm and X be a nonempty set, then the triplet
(X, F, ) is defined as a Menger space. Here F : X × X → D + satisfies the fol-
lowing ( F (x, y) is denoted by Fx,y for x, y ∈ X ):
On Hick’s Contraction Using a Control Function 15
1+t
, if 0 ≤ t < 1.
Lemma 1 If ϕ ∈ . then there exists r ≥ t such that ϕ(r ) < t for each t > 0.
2 Main Results
Fx,y (t) > 1 − t ⇒ FT x,T y (ϕ(t)) > 1 − kt, for all t > 0 and x, y ∈ X, (1)
where ϕ ∈ and 0 < k < 1. Then x∗ ∈ X is a unique fixed point of T and also
{T n (x0 )} converges to x∗ for any arbitrary x0 ∈ X .
Proof For any x0 ∈ X,. we write xn = T n x0 = T xn−1 for all n ≥ 1. Let 0 < η < 1.
For Fx0 ,x1 (t) → 1 as t → ∞, there exists r > 0, such that
that is,
Fx1 ,x2 (ϕ(r )) > 1 − kη.
In general, ∀ n ∈ N, we have
Again, Fx0 ,x1 (t) → 1 as t → ∞, for any ε ∈ (0, 1]. Hence, there exists t1 > 0
such that Fx0 ,x1 (t1 ) > 1 − ε. Here ϕ ∈ . Therefore, there exists t0 ≥ t1 such that
ϕn (t0 ) → 0 as n → ∞.
ϕn (t0 ) → 0 as n → ∞. (2)
Now,
Fx0 ,x1 (t0 ) ≥ Fx0 ,x1 (t1 ) > 1 − ε.
that is,
Fx1 ,x2 (ϕ(t0 )) > 1 − kε > 1 − ε,
Now, from (2) and (3), by L-convergence criteria, xn becomes a convergent sequence.
Let
xn → x as n → ∞. (4)
Fxn−1 ,x (r ) → 1 .
On Hick’s Contraction Using a Control Function 17
Thus, for arbitrary 0 < λ < 1, we obtain N1 such that for all n > N1 ,
Fxn−1 ,x (r ) > 1 − λ,
Again, as {xn } converges to x, it is possible to find N2 such that for all n > N2 , we
obtain
Fxn ,x (ε − ϕ(r )) > 1 − λ. (7)
Here, is continuous t-norm and λ is arbitrary, hence for any ε > 0 we find,
Fx,T x (ε) = 1,
thus, x = T x.
Next, to establish the uniqueness of the fixed point, let x and y be any two fixed
points of T, that is, y = T y and x = T x.
There exists t0 > 0 such that Fx,y (t0 ) > 1 − ε because Fx,y (t) → 1 as t → ∞,
for any ε ∈ (0, 1]. Since ϕ ∈ , therefore, as n → ∞, there exists t1 ≥ t0 , such that
ϕn (t1 ) → 0. Let t > 0 be arbitrary. Then we can obtain n 0 ∈ N, such that ϕn (t1 ) < t
for all n ≥ n 0 . Thus, by monotonicity of the distribution function, we get
Now,
Fx,y (t1 ) ≥ Fx,y (t0 ) > 1 − ε.
that is,
Fx,y (ϕ(t1 )) > 1 − kε,
3 Conclusion
Incidentally, we may also mention that there is no unique way of defining proba-
bilistic metrics. This flexibility makes it possible to suitably orient the definition to
fulfil specific purposes as, for instance, in [20], the definition has been tailored to
describe a nuclear fusion-related problem. In particular, fixed point-related studies
have required considerations of t-norm, the different choices of which radically alter
the characteristics of the space. It may be seen in future work how our result presented
here is varied with different choices of t-norms.
Acknowledgements The third author’s research is supported by the University Grant Commission
(No. 19-06/2011(i)EU-IV), India. The research work of the second author is supported by DST-WB,
India (624(sane)/ST/P/S & T/MISC-5/2012/Dated 27.08.2013).
References
1. Meneger, K.: Statistical metrics. Proc. Nat. Acad. Sci. USA 28, 535–537 (1942)
2. Sehgal, V.M., Bharucha-Reid, A.T.: Fixed points of contraction mappings on PM-spaces. Math.
Syst. Theory 6, 97–102 (1972)
3. Schweizer, B., Sklar, A.: Probabilistic Metric Spaces. Elsevier, New York (1983)
4. Hicks, T.L.: Fixed point theory in probabilistic metric spaces, Zb. Rad. Prir. Mat. Fak. Univ.
Novom Sadu 13, 63–72 (1983)
5. Hicks, T.L.: Fixed point theory in probabilistic metric spaces II. Math. Japon- ica 44(3), 487–
493 (1996)
6. Mihet, D.: Generalized Hicks contractions: an extension of a result of Žikić. Fuzzy Sets Syst.
157, 2384–2393 (2006)
7. Miheţ, D.: Weak-Hicks contractions, 6, 71–78 (2005)
8. Ciric, L.: Solving the Banach fixed point principle for nonlinear contractions in probabilistic
metric spaces,. Nonlinear Anal. 72, 2009–2018 (2010)
On Hick’s Contraction Using a Control Function 19
9. Jachymski, J.: On probabilistic ϕ-contractions on Menger spaces. Nonlinear Anal. 73, 2199–
2203 (2010)
10. Schweizer, B., Sklar, A., Thorp, E.: The metrization of statistical metric spaces. Pac. J. Math.
10, 673–675 (1960)
11. O. Hadzic, Some theorems on the fixed points in probabilistic metric and random normed
spaces. Boll. Unione Mat. Ital. Sez. B 1(6), 381–391 (1982),
12. Fang, J.-X.: On ϕ-contractions in probabilistic and fuzzy metric spaces. Fuzzy Sets Syst. 267,
86–99 (2015)
13. Xiao, J.Z., Zhu, X.H., Cao, Y.F.: Common coupled fixed point results for probabilistic ϕ-
contractions in Menger spaces. Nonlinear Anal. Theory Methods Appl. 74, 4589–4600 (2011)
14. Chauhan, S., Pant, B.D.: Fixed point theorems for compatible and subsequentially continuous
mappings in Menger spaces. J. Nonlinear Sci. Appl. 7, 78–89 (2014)
15. Choudhury, B.S., Das, K.: A new contraction principle in Menger spaces. Acta Math. Sin.
Engl. Ser 24(8), 1379–1386 (2008)
16. Dutta, P.N., Choudhury, B.S., Das, K.: Some fixed point results in Menger spaces using a
control function. Surv. Math. Appl. 4, 41–52 (2009)
17. Kutbi, M.A., Gopal, D., Vetro, C., Sintunavarat, W.: Further generalization of fixed point
theorems in Menger PM-spaces. Fixed Point Theory Appl. 2015(1), 32 (2015)
18. Hadzic, O., Pap, E.: Fixed Point Theory in Probabilistic Metric Spaces. Springer (2001)
19. Samet, B.: Coupled fixed point theorems for a generalized Meir-Keeler contraction in partially
ordered metric spaces. Nonlinear Anal. 72, 4508–4517 (2010)
20. Verdoolaege, G., Karagounis, G., Murari, A., Vega, J., Van Oost, G.: JET-EFDA contribu-
tors, modeling fusion data in probabilistic metric spaces: applications to the identification of
confinement regimes and plasma disruptions. Fusion Sci. Technol. 62, 356–365 (2012)
Coupled Fixed Points for Multivalued
Feng–Liu-Type Contractions
with Application to Nonlinear Integral
Equation
Abstract In this paper, we establish existence of the coupled fixed point for set-
valued Feng–Liu-type contractions in complete metric spaces under two different
sets of conditions. Some consequences are obtained and an application to a nonlinear
integral equation is included.
Multivalued nonlinear contractions appeared first in fixed point theory in the work of
Nadler [14]. This work was followed by a development of the branch of fixed point
theory in the domain of set-valued analysis through works like [8, 11, 12, 15, 17,
21]. Our aim in this paper is to establish the existence of fixed points of a coupled
multivalued mapping satisfying a generalized Feng–Liu-type contraction and new
admissibility condition which we defined in [3]. Further we apply our coupled fixed
point result for solving a system of nonlinear integral equations.
B. S. Choudhury · P. Maity
Department of Mathematics, Indian Institute of Engineering Science and Technology,
Shibpur, Howrah 711103, West Bengal, India
e-mail: [email protected]
N. Metiya (B)
Department of Mathematics, Sovarani Memorial College, Jagatballavpur,
Howrah 711408, West Bengal, India
e-mail: [email protected]; [email protected]
S. Kundu
Department of Mathematics, Government General Degree College, Salboni,
Paschim Medinipur 721516, West Bengal, India
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 21
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_3
22 B. S. Choudhury et al.
Suppose CL(X ) denotes the collection of all nonempty closed subsets of a metric
space (X, ρ). We use following notations and definitions when P, Q ∈ CL(X ):
and
max {sup D( p, Q), sup D(q, P)}, if the infimum exists,
H(P, Q) = p∈P q∈Q
∞, otherwise.
The pair (CL(X ), H) is a generalized metric space and H is called the generalized
Hausdorff distance [6].
The idea of a coupled fixed point was introduced by Guo and Lakshmikantham
[10] in 1987. But only after the publication of the work of Bhaskar and Lakshmikan-
tham [9], a large number of papers have been written on this topic and on topics
related to it [2, 3, 5, 17–19].
A coupled fixed point of a mapping S : X × X → X is a point (u, v) ∈ X ×
X such that u = S(u, v) and v = S(v, u). For a multivalued mapping S : X ×
X → CL(X ), a coupled fixed point is an element (u, v) ∈ X × X satisfying u ∈
S(u, v) and v ∈ S(v, u)
Various admissibility criteria have been introduced in the study of fixed points of
mappings. In particular, we refer the reader to [1, 3, 4, 7, 13, 20].
Example 1.1 Take the usual metric space X = [0, 1]. Define T : X × X → CL(X )
as T (x, y) = [0, x+y
16
], for x, y ∈ X and α, β : X → [0, ∞) as
e x , if x ∈ [0, 21 ], cosh x, if x ∈ [0, 21 ],
α(x) = and β(x) =
0, otherwise 0, otherwise.
Suppose that (x, y) ∈ X × X such that α(x) ≥ 1 and β(y) ≥ 1. Then x, y ∈ [0, 21 ]
and T (x, y) ⊆ [0, 21 ]. It follows that β(u) ≥ 1 for all u ∈ T (x, y). Similarly, one
can show α(v) ≥ 1 for all v ∈ T (x, y) whenever (x, y) ∈ X × X with β(x) ≥ 1 and
α(y) ≥ 1. The mapping T is here cyclic (α, β)-admissible.
Coupled Fixed Points for Multivalued Feng–Liu-Type Contractions … 23
By we denote the class of all functions φb : [0, ∞) → [0, b), 0 < b < 1,
satisfying lim supr −→t + φb (r ) < b, for each t ∈ [0, ∞). We use this class of functions
in our theorems.
2 Main Results
f (u, v) ≤ φb (max {d(x, u), d(y, v)}) max {d(x, u), d(y, v)}, (2.1)
where f (s, t) = max{D(s, T (s, t)), D(t, T (t, s))} for s, t ∈ X and (x, y) =
α(x) β(y) or (x, y) = α(y) β(x). If there exist x0 , y0 ∈ X such that α(x0 ) ≥ 1
and β(y0 ) ≥ 1, then T has a coupled fixed point.
Proof If possible, suppose that f (x, y) = max {D(x, T (x, y)), D(y, T (y, x))} = 0
for some (x, y) ∈ X × X . Then D(x, T (x, y)) = 0 and D(y, T (y, x)) = 0, which
imply that x ∈ T (x, y) and y ∈ T (y, x), that is, (x, y) is a coupled fixed point of T .
Hence we shall assume that f (x, y) = 0 for every (x, y) ∈ X × X . Since b ∈ (0, 1),
there exist u ∈ T (x, y) and v ∈ T (y, x) such that
b max {d(x, u), d(y, v)} ≤ max {D(x, T (x, y)), D(y, T (y, x))} = f (x, y).
(2.2)
Continuing this process, we construct two sequences {xn } and {yn } in X such that
xn+1 ∈ T (xn , yn ) and yn+1 ∈ T (yn , xn ) with (xn , yn ) ≥ 1, for all n ≥ 0. (2.3)
Also,
⎫
b max {d(xn , xn+1 ), d(yn , yn+1 )} ≤ f (xn , yn ) ⎬
and
⎭
f (xn+1 , yn+1 ) ≤ φb (max{d(xn , xn+1 ), d(yn , yn+1 )}) max{d(xn , xn+1 ), d(yn , yn+1 )}.
(2.4)
Let rn = max {d(xn , xn+1 ), d(yn , yn+1 )}, for all n ≥ 0. (2.5)
Therefore,
⎫
f (xn+1 , yn+1 ) φb (rn ) ⎪
rn+1 ≤ ≤ rn ⎬
b b (2.6)
and ⎪
⎭
f (xn , yn ) − f (xn+1 , yn+1 ) ≥ b rn − φb (rn ) rn = [b − φb (rn )] rn .
It follows that {rn } and { f (xn , yn )} are strictly decreasing sequences of non-negative
real numbers. From the property of φb there exists a q ∈ [0, b) such that
φb (rn )
f (xn+1 , yn+1 ) ≤ φb (rn ) rn ≤ f (xn , yn )
b
φb (rn ) φb (rn−1 )
≤ f (xn−1 , yn−1 )
b2
φb (rn ) φb (rn−1 ) φb (rn−2 )
≤ f (xn−2 , yn−2 )
b3
≤ ...
φb (rn ) φb (rn−1 ) φb (rn−2 )...φb (r2 ) φb (r1 )
≤ f (x1 , y1 )
bn
φb (rn ) φb (rn−1 ) ...φb (rn 0 +1 ) φb (rn 0 ) φb (rn 0 −1 ) ...φb (r1 )
= f (x1 , y1 )
bn−n 0 bn 0
b0 n−n 0 φb (rn 0 ) φb (rn 0 −1 ) ...φb (r1 )
< f (x1 , y1 ) , for all n > n 0 . (2.10)
b bn 0
b0 b0
As b0 < b, < 1 and hence limn→∞ ( )n−n 0 = 0. Then it follows that
b b
1 f (xn , yn )
= [ f (xn , yn ) − f (xm , ym )] ≤ .
γ γ
xn → x and yn → y, as n → ∞. (2.12)
3 Some Consequences
f (u, v) ≤ φb (max {d(x, u), d(y, v)}) max {d(x, u), d(y, v)},
where f (s, t) is as defined in Theorem 2.1. Then T has a coupled fixed point if either
T is continuous or f is lower semi-continuous.
Proof Define α, β : X → [0, ∞) as α(x) = β(x) = 1, for all x ∈ X . Then the proof
follows from that of Theorem 2.1 if T is continuous and from that of Theorem 2.2 if
f is lower semi-continuous.
H(T (x, y), T (u, v)) ≤ φb (max {d(x, u), d(y, v)}) max {d(x, u), d(y, v)}.
Proof By the condition of the theorem for any u ∈ T (x, y) and v ∈ T (y, x), we
have
D(u, T (u, v)) ≤ H(T (x, y), T (u, v)) ≤ φb (max {d(x, u), d(y, v)}) max {d(x, u), d(y, v)}
and
D(v, T (v, u)) ≤ H(T (y, x), T (v, u)) ≤ φb (max {d(x, u), d(y, v)}) max {d(x, u), d(y, v)}.
f (u, v) = max{D(u, T (u, v)), D(v, T (v, u))} ≤ φb (max{d(x, u), d(y, v)})max {d(x, u), d(y, v)}.
b
H(T (x, y), T (u, v)) ≤ max {d(x, u), d(y, v)}.
2
If either T is continuous or f (x, y) = max{D(x, T (x, y)), D(y, T (y, x))} is lower
semi-continuous then T has a coupled fixed point.
b
Proof Take b ∈ (0, 1) and φb ∈ , where φb (t) = , for all t ∈ [0, ∞). Then the
2
proof follows from that of Theorem 3.2.
28 B. S. Choudhury et al.
In this section, we present an application of our coupled fixed point results derived
in Sect. 2 to solve a nonlinear integral equation.
Every single-valued mapping T : X × X → X can be treated as a multival-
ued mapping T : X × X → CL(X ) in which case T (x, y) is a singleton set for
(x, y) ∈ X × X . Taking α(x) = β(x) = 1, for all x ∈ X in Theorem 2.1, we have
the following result which is a special case of Theorem 2.1.
max {d(T (x, y), T (T (x, y), T (y, x))), d(T (y, x), T (T (y, x), T (x, y)))}
≤ φb (max {d(x, T (x, y)), d(y, T (y, x))}) max {d(x, T (x, y)), d(y, T (y, x))}.
Fixed point theorems for operators in metric spaces have found applications in
differential and integral equations (see [16] and references therein). We consider here
a system of nonlinear integral equation as follows:
b
x(t) = f (t) + a h(t, s, x(s), y(s))ds and
b
y(t) = f (t) + a h(t, s, y(s), x(s))ds, where t, s ∈ [a, b]
(4.1)
and the unknown functions x(t) and y(t) are real valued.
Let X = C([a, b]), where b > a be the space of all real-valued continuous func-
tions defined on [a, b]. It is well known that C([a, b]) endowed with the metric
Theorem 4.2 Let (X, d) = (C([a, b]), d), T, h(t, s, x, y) satisfy the assumptions
A1 , A2 and A3 . Then the system of integral equations (4.1) has a unique solution in
C([a, b]) × C([a, b]).
Proof Consider the mapping T : X × X → X defined by (4.3). Take c = 2 b −
a and φc (t) = b − a, for all t ∈ [0, ∞). Then lim supr −→t + φc (r ) < 1, for all t ∈
[0, ∞). By A3 , φc ∈ .
By assumptions A1 and A2 , for all (x, y) ∈ C([a, b]) × C([a, b]) and for u =
T (x, y), v = T (y, x) with t, s ∈ [a, b], we have
b
| u(t) − F(u, v)(t) |=| F(x, y)(t) − F(u, v)(t) |=| [h(t, s, x(s), y(s)) − h(t, s, u(s), v(s))]ds |
a
b
= | [h(t, s, x(s), y(s)) − h(t, s, u(s), v(s))] | ds
a
b
1
≤ [ [ | x(s) − u(s) | + | y(s) − v(s) | ]ds
a 2
d(x, u) + d(y, v) b (b − a) [d(x, u) + d(y, v)]
= 1 ds = .
2 a 2
Similarly, we have
max {d(T (x, y), T (T (x, y), T (y, x))), d(T (y, x), T (T (y, x), T (x, y)))}
= max {d(u, T (u, v)), d(v, T (v, u))} = max {| u(t) − F(u, v)(t) |, | v(t) − F(v, u)(t) |}
≤ φc (max {d(x, T (x, y)), d(y, T (y, x))}) max {d(x, T (x, y)), d(y, T (y, x))}.
30 B. S. Choudhury et al.
Therefore, all the conditions of Theorem 4.1 are satisfied. Then there exists a point
(x, y) in X × X such that x = T (x, y) and y = T (y, x), that is, (x, y) is a solution
of the system of nonlinear integral equations (4.1).
References
1. Alizadeh, S., Moradlou, F., Salimi, P.: Some fixed point results for (α − β) - (ψ − ϕ) - con-
tractive mappings. Filomat 28(3), 635–647 (2014)
2. Choudhury, B.S., Metiya, N., Kundu, S.: Existence, data-dependence and stability of coupled
fixed point sets of some multivalued operators. Chaos, Soliton Fractals 133, 109678 (2020)
3. Choudhury, B.S., Gnana Bhaskar, T., Metiya, N., Kundu, S.: Existence and stability of coupled
fixed point sets of multi-valued mappings. Fixed Point Theory 22(2), 571–586 (2021)
4. Choudhury, B.S., Metiya N., Kundu, S.: A multivalued fixed point result with associated data
dependence and stability study. Prob. Analiza-Issue Anal. 11(1)(29), 45–57 (2022)
5. Ćirić, L.B., Lakshmikantham, V.: Coupled random fixed point theorems for nonlinear contrac-
tions in partially ordered metric spaces. Stoch. Anal. Appl. 27(6), 1246–1259 (2009)
6. Covitz, H., Nadler, S.B.: Multi-valued contraction mappings in generalized metric spaces 8(1),
5–11 (1970)
7. Felhi, A., Aydi, H., Zhang, D.: Fixed points for α - admissible contractive mappings via
simulation functions. J. Nonlinear Sci. Appl. 9, 5544–5560 (2016)
8. Feng, Y., Liu, S.: Fixed point theorems for multi-valued contractive mappings and multi-valued
Caristi type mappings. J. Math. Anal. Appl. 317, 103–112 (2006)
9. Gnana Bhaskar, T., Lakshmikantham, V.: Fixed point theorems in partially ordered metric
spaces and applications. Nonlinear Anal. 65, 1379–1393 (2006)
10. Guo, D., Lakshmikantham, V.: Coupled fixed points of nonlinear operators with applications.
Nonlinear Anal. 11, 623–632 (1987)
11. Klim, D., Wardowski, D.: Fixed point theorems for set-valued contractions in complete metric
spaces. J. Math. Anal. Appl. 334, 132–139 (2007)
12. Laha, A.K., Saha, M.: Fixed point for a class of set valued mappings on a metric space endowed
with a graph. ROMAI J. 11(1), 115–129 (2015)
13. Laha, A.K., Saha, M.: Fixed point on α − ψ multivalued contractive mappings in cone metric
space. Acta Comment. Univ. Tartu. Math. 20(1), 35–43 (2016)
14. Nadler, S.B., Jr.: Multivalued contraction mappings. Pac. J. Math. 30, 475–488 (1969)
15. Nicolae, A.: Fixed point theorems for multi-valued mappings of Feng-Liu type. Fixed Point
Theory 12(1), 145–154 (2011)
16. Nieto, J.J., Rodrguez-López, R.: Contractive mapping theorems in partially ordered sets and
applications to ordinary differential equations. Order 22, 223–239 (2005)
17. Petrusel, A., Petrusel, G., Samet, B., Yao, J.C.: Scalar and vectorial approaches for multi-valued
fixed point and multi-valued coupled fixed point problems in b-metric spaces. J. Nonlinear
Convex Anal. 10(17), 2049–2061 (2016)
18. Li, J., Petrusel, A.: Extended coupled fixed point problems for set-valued mappings on partially
ordered Banach spaces and their applications to systems of Hammerstein integral equations. J.
Nonlinear Convex Anal. 11(20), 2321–2333 (2019)
19. Samet, B., Vetro, C.: Coupled fixed point theorems for multi-valued nonlinear contraction
mappings in partially ordered metric spaces. Nonlinear Anal. 74(12), 4260–4268 (2011)
20. Samet, B., Vetro, C., Vetro, P.: Fixed point theorems for α − ψ-contractive type mappings.
Nonlinear Anal. 75, 2154–2165 (2012)
21. Zhang, X.: Fixed point theorems of multivalued monotone mappings in ordered metric spaces.
Appl. Math. Lett. 23(3), 235–240 (2010)
Fractals
Clifford-Valued Fractal Interpolation
Peter R. Massopust
Abstract In this short note, we merge the areas of hypercomplex algebras with that
of fractal interpolation and approximation. The outcome is a new holistic method-
ology that allows the modeling of phenomena exhibiting a complex self-referential
geometry and which require for their description an underlying algebraic structure.
1 Introduction
In this short note, we merge two areas of mathematics: the theory of hypercomplex
algebras as exemplified by Clifford algebras and the theory of fractal approximation
or interpolation.
In recent years, hypercomplex methodologies have found their way into many
applications one of which is digital signal processing. See, for instance, [1, 36] and
the references given therein. The main idea is to use the multidimensionality of hyper-
complex algebras to model signals with multiple channels or images with multiple
color values and to use the underlying algebraic structure of such algebras to oper-
ate on these signals or images. The results of these algebraic or analytic operations
produce again elements of the hypercomplex algebra. This holistic approach cannot
be performed in finite dimensional vector spaces as these do not possess an intrinsic
algebraic structure.
On the other hand, the concept of fractal interpolation has been employed suc-
cessfully in numerous applied situations over the last decades. The main purpose
of fractal interpolation or approximation is to take into account complex geometric
https://www-m15.ma.tum.de/Allgemeines/PeterMassopust.
P. R. Massopust (B)
Department of Mathematics, Technical University of Munich, Boltzmannstr. 3, 85748 Garching b.
Munich, Germany
e-mail: [email protected]
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 33
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_4
34 P. R. Massopust
self-referential structures and to employ approximants that are well suited to model
these types of structures. These approximants or interpolants are elements of vector
spaces and cannot be operated on in an algebraic way to produce the same type of
object. Hence, the need for an extension of fractal interpolation to the hypercom-
plex setting. An initial investigation into the novel concept of hypercomplex iterated
function system was already undertaken in [32] albeit along a different direction.
The structure of this paper is as follows. In Sect. 2, we give a brief introduction of
Clifford algebras and mention a few items from Clifford analysis. In the third section,
we review some techniques and state relevant results from the theory of fractal
interpolation in Banach spaces. These techniques are then employed in Sect. 4 to a
Clifford algebraic setting. The next section briefly mentions a special case of Clifford-
valued fractal interpolation, namely that based on paravector-valued functions. In the
last section, we provide a brief summary and mention future research directions.
F(X, Rn ) := F(X ) ⊗R Rn .
This linear space becomes a Banach space when endowed with the norm
⎛ ⎞1/2
f := ⎝ fA 2
F (X )
⎠ .
A⊆Nn
It is known [20, Remark 2.2. and Proposition 2.3.] that f ∈ F(X, Rn ) iff
f = f AeA (3)
A⊆Nn
To this end, let X be a nonempty bounded subset of the Banach space Rm . Suppose
we are given a finite family {L i }i=1
N
of injective contractions X → X generating a
partition of X in the sense that
∀ i, j ∈ N N , i = j : L i (X ) ∩ L j (X ) = ∅; (4)
N
X= L i (X ). (5)
i=1
N
ψ:X= Xi → R
i=1
on X i , i ∈ N N , or, equivalently, by
N
N
Tf = qi ◦ L i−1 χXi + si ◦ L i−1 · f ◦ L i−1 χ X i
i=1 i=1
N
= T (0) + si ◦ L i−1 · f ◦ L i−1 χ X i , x ∈ X,
i=1
N
T f − Tg F (X ) = si ◦ L i−1 · ( f − g) ◦ L i−1 χ X i
i=1 F (X )
≤ γF (X ) f − g F (X )
As above, we write X i := L i (X ), i ∈ N N .
On the spaces F(X, Rn ), we define an RB operator T as follows. Let f ∈
F(X, Rn ) with f = f A e A , where f A ∈ F(X ), for all A ⊆ Nn . Let T : F(X ) →
A⊆Nn
38 P. R. Massopust
provided that T ( f A ) ∈ F(X ) for all A ⊆ Nn . Under the latter assumption and the
supposition that T is contractive on F(X ) with Lipschitz constant γF (X ) , we obtain
for f, g ∈ F(X, Rn )
2
T f − Tg = T f A − T gA 2
F (X )
A⊆Nn
≤ γF
2
(X ) f A − gA 2
F (X )
A⊆Nn
= γF
2
(X ) f − g .
2
Hence, T is also contractive on F(X, Rn ) and with the same Lipschitz constant
γF (X ) .
The following diagram illustrates the above approach.
T
F(X ) −−−−→ F(X )
⏐ ⏐
⏐⊗ R ⏐⊗ R (9)
R n R n
T
F(X, Rn ) −−−−→ F(X, Rn )
Proof The validity of these statements follows directly from the above elaborations.
For the sake of completeness, we now list the Lipschitz constants γF (X ) for the
functions spaces listed in Sect. 2 in the case m = 1. The conditions are γF (X ) < 1.
Note that the expressions are different for the case m > 1.
1. C k (X ): γC k (X ) = max{Lip(L i )−(k+1) si L ∞ : i ∈ N N }.
−(k+α)
2. C k,α (X ): γC k,α (X ) =
max{Lip(L i ) p si L ∞ : i ∈ N N }.
3. L (X ): γ L p (X ) =
p
Lip(L i ) si L ∞ .
i∈N N
p
4. W s, p (X ): γW s, p (X ) = Lip(L i )1−sp si L∞ .
i∈N N
Lip(L i )(1/ p−s)q si
q
5. B sp,q (X ): γ B sp,q (X ) = L∞ .
i∈N N
p
s
6. F p,q (X ): γ Fp,q
s (X ) = Lip(L i )1−sp si L∞ .
i∈N N
5 Paravector-Valued Functions
[20, 35].
Prominent examples of paravector-valued functions are, for instance, the expo-
nential and sine functions [35] for x ∈ An+1 :
n
f = f k ek .
k=0
Clifford-Valued Fractal Interpolation 41
In this short note, we have initiated the investigation of fractal interpolation into
a hypercomplex setting. The main idea was to define fractal interpolants along the
different directions defined by a Clifford algebra Rn and use the underlying algebraic
structure to manipulate the hypercomplex fractal object to yield another hypercom-
plex fractal object.
There are several extensions of this first initial approach:
1. Define—under suitable conditions—RB operators acting directly on appropri-
ately defined function spaces F(X, Rn ) instead of resorting to the “component”
RB operators.
2. Provide a local version of the defined hypercomplex fractal interpolation in the
sense first defined in [7] and further investigated in, i.e., [5, 29].
3. Construct nonstationary approaches to Clifford-valued fractal interpolation in the
spirit of [31].
4. Extend the notion of hypercomplex fractal interpolation to systems of function
systems as described in [14, 25].
References
1. Alsmann, D.: On families of 2 N -dimensional hypercomplex algebras suitable for digital image
processing. In: 14th European Signal Processing Conference (EUSIPCO 2006), pp. 1–4 (2006)
2. Barnsley, M.F.: Fractals Everywhere. Dover Publications Inc. (2012)
3. Barnsley, M.F.: Fractal functions and interpolation. Constr. Approx. 2, 303–329 (1986)
4. Barnsley, M.F., Harding, B., Vince, C., Viswanathan, P.: Approximation of rough functions. J.
Approx. Th. 209, 23–43 (2016)
5. Barnsley, M.F., Hegland, M., Massopust, P.R.: Numerics and fractals. Bull. Inst. Math. Acad.
Sin. (N.S.) 9(3), 389–430 (2014)
6. Barnsley, M.F., Hegland, M., Massopust, P.R.: Self-referential functions. arXiv: 1610.01369
7. Barnsley, M.F., Hurd, L.P.: Fractal Image Compression. AK Peters Ltd., Wellesly (1993)
8. Bedford, T., Dekking, M., Keane, M.: Fractal image coding techniques and contraction oper-
ators. Delft University of Technology Report, pp. 92–93 (1992)
9. Brackx, F., Delanghe, R., Sommen, F.: Clifford Analysis. Pitman Books (1982)
10. Abłamowicz, R., Sobczyk, G.: Lectures on Clifford (Geometric) Algebras and Applications.
Birkhäuser, New York (2004)
42 P. R. Massopust
11. Colombo, F., Sabadini, I., Struppa, D.C.: Noncommutative Functional Calculus: Theory and
Applications of Slice Hyperholomorphic Functions. Birkhäuser Verlag (2010)
12. Colombo, F., Sabadini, I., Struppa, D.C.: Entire Slice Regular Functions. Springer (2016)
13. Delanghe, R., Sommen, F., Souček, V.: Clifford Algebra and Spinor-Valued Functions. A
Function Theory for the Dirac Operator. Springer, Dordrecht (1992)
14. Dira, N., Levin, D., Massopust, P.: Attractors of trees of maps and of sequences of maps between
spaces and applications to subdivision. J. Fixed Point Theory Appl. 22(14), 1–24 (2020)
15. Dubuc, S.: Interpolation through an iterative scheme. J. Math. Anal. Appl. 114(1), 185–204
(1986)
16. Dubuc, S.: Interpolation fractale. In: Bélais , J., Dubuc, S. (eds.) Fractal Geomety and Analysis.
Kluwer Academic Publishers, Dordrecht, The Netherlands (1989)
17. Falconer, K.: Fractal Geometry: Mathematical Foundations and Applications. Wiley (2014)
18. Gürlebeck, K., Habetha, K., Sprößig, W.: Holomorphic Functions in the Plane and n-
Dimensional Space. Birkhäuser Verlag (2000)
19. Gürlebeck, K., Habetha, K., Sprößig, W.: Application of Holomorphic Functions in Two and
Higher Dimensions. Birkhäuser Verlag (2016)
20. Gürlebeck, K., Sprößig, W.: Quaternionic and Clifford Calculus for Physicists and Engineers.
Wiley (1997)
21. Hutchinson, J.E.: Fractals and self-similarity. Indiana Univ. Math. J. 30, 713–747 (1981)
22. Huang, S., Qiao, Y.Y., Wen, G.C.: Real and Complex Clifford Analysis. Springer (2006)
23. Kantor, I.L., Solodovnik, A.S.: Hypercomplex Numbers. An Elementary Introduction to Alge-
bras. Springer, New York (1989)
24. Kravchenko, V.: Applied Quaternionic Analysis. Heldermann Verlag, Lemgo, Germany (2003)
25. Levin, D., Dyn, N., Viswanathan, P.: Non-stationary versions of fixed-point theory, with appli-
cations to fractals and subdivision. J. Fixed Point Theory Appl. 21, 1–25 (2019)
26. Massopust, P.R.: Fractal functions and their applications. Chaos, Solitons Fractals 8(2), 171–
190 (1997)
27. Massopust, P.R.: Interpolation and Approximation with Splines and Fractals. Oxford University
Press, Oxford, USA (2010)
28. Massopust, P.R.: Fractal Functions, Fractal Surfaces, and Wavelets, 2nd edn. Academic, San
Diego, USA (2016)
29. Massopust, P.R.: Local fractal functions and function spaces. Springer Proc. Math. Stat.: Frac-
tals, Wavelets and their Appl. 92, 245–270 (2014)
30. Massopust, P.R.: Local fractal functions in Besov and Triebel-Lizorkin spaces. J. Math. Anal.
Appl. 436, 393–407 (2016)
31. Massopust, P.R.: Non-stationary fractal interpolation. Mathematics 7(8), 1–14 (2019)
32. Massopust, P.R.: Hypercomplex iterated function systems. In: Cereijeras, P., Reissig, M., Saba-
dini, I., Toft, J. (eds.) Current Trends in Analysis, Its Applications and Computation. Proceed-
ings of the 12th ISAAC Congress, Aveiro, Portugal. Birkhäuser (2019)
33. Serpa, C., Buescu, J.: Constructive solutions for systems of iterative functional equations.
Constr. Approx. 45(2), 273–299 (2017)
34. Serpa, C., Buescu, J.: Compatibility conditions for systems of iterative functional equations
with non-trivial contact sets. Results Math. 2, 1–19 (2021)
35. Sprössig, W.: On operators and elementary functions in clifford analysis. Zeitschrift für Anal-
ysis und ihre Anwendung. 19(2), 349–366 (1999)
36. Schutte, H.-D., Wenzel, J.: Hypercomplex numbers in digital signal processing. In: IEEE
International Symposium on Circuits and Systems (1990). 10.1109ISCAS.1990.112431
Optimal Quantizers for a Nonuniform
Distribution on a Sierpiński Carpet
1 Introduction
If x2 d P(x) < ∞, then there is some set α for which the infimum is achieved
(see [1, 5–7]). Such a set α for which the infimum occurs and contains no more than
n points is called an optimal set of n-means, or optimal set of n-quantizers. The
M. K. Roychowdhury (B)
School of Mathematical and Statistical Sciences, University of Texas Rio Grande Valley,
1201 West University Drive, Edinburg 78539-2999, USA
e-mail: [email protected]
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 43
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_5
44 M. K. Roychowdhury
i.e., the Voronoi region generated by a ∈ α is the set of all points in Rd which are
closest to a ∈ α, and the set {M(a|α) : a ∈ α} is called the Voronoi diagram or
Voronoi tessellation of Rd with respect to α. A Borel measurable partition {Aa : a ∈
α} of Rd is called a Voronoi partition of Rd with respect to α (and P) if P-almost
surely Aa ⊂ M(a|α) for every a ∈ α. Given a Voronoi tessellation {Mi }i=1 k
generated
by a set of points {z i }i=1 (called sites or generators), the mass centroid ci of Mi with
k
which implies that a is the centroid of the Voronoi region M(a|α) associated with
the probability measure P (see also [3, 11]).
A transformation f : X → X on a metric space (X, d) is called contractive or
a contraction mapping if there is a constant 0 < c < 1 such that d( f (x), f (y)) ≤
cd(x, y) for all x, y ∈ X . On the other hand, f is called a similarity mapping or a
similitude if there exists a constant s > 0 such that d( f (x), f (y)) = sd(x, y) for
all x, y ∈ X . Here s is called the similarity ratio of the similarity mapping f . Let
C be the Cantor set generated by the two contractive similarity mappings S1 and S2
on R given by S1 (x) = r1 x and S2 (x) = r2 x + (1 − r2 ) where 0 < r1 , r2 < 1 and
r1 + r2 < 21 . Let P = p1 P ◦ S1−1 + p2 P ◦ S2−1 , where P ◦ Si−1 denotes the image
measure of P with respect to Si for i = 1, 2 and ( p1 , p2 ) is a probability vector
with 0 < p1 , p2 < 1. Then, P is a singular continuous probability measure on R
with support the Cantor set C (see [10]). For r1 = r2 = 13 and p1 = p2 = 21 , Graf
Optimal Quantizers for a Nonuniform Distribution … 45
and Luschgy gave a closed formula to determine the optimal sets of n-means for
the probability distribution P for any n ≥ 2 (see [8]). For r1 = 41 , r2 = 21 , p1 = 14 ,
and p2 = 43 , L. Roychowdhury gave an induction formula to determine the optimal
sets of n-means and the nth quantization error for the probability distribution P for
any n ≥ 2 (see [12]). Let P be a Borel probability measure on R2 supported by
the Cantor dusts generated by a set of 4u , u ≥ 1, contractive similarity mappings
satisfying the strong separation condition. For this probability measure, Cömez and
Roychowdhury determined the optimal sets of n-means and the nth quantization
errors for all n ≥ 2 (see [2]). In addition, they showed that though the quantization
dimension of the measure P is known, the quantization coefficient for P does not
exist.
In this paper, we have considered the probability distribution P given by P =
1
8
P ◦ S1−1 + 18 P ◦ S2−1 + 38 P ◦ S3−1 + 38 P ◦ S4−1 which has support on the Sierpiński
carpet generated by the four contractive similarity mappings given by S1 (x1 , x2 ) =
1
(x , x ), S2 (x1 , x2 ) = 13 (x1 , x2 ) + ( 23 , 0), S3 (x1 , x2 ) = 13 (x1 , x2 ) + (0, 23 ), and
3 1 2
S4 (x1 , x2 ) = 13 (x1 , x2 ) + ( 23 , 23 ) for all (x1 , x2 ) ∈ R2 . The probability distribution
P considered in this paper is called ‘nonuniform’ to mean that all the basic squares
at a given level that generate the Sierpiński carpet do not have the same probability.
For this probability distribution in Propositions 3.1–3.3, first we have determined the
optimal sets of n-means and the nth quantization errors for n = 2, 3, and 4. Then,
in Theorem 1 we state and prove an induction formula to determine the optimal sets
of n-means for all n ≥ 2. We also give some figures to illustrate the locations of the
optimal points (see Fig. 1). In addition, using the induction formula, we obtain some
results and observations about the optimal sets of n-means which are given in Sect. 4;
a tree diagram of the optimal sets of n-means for a certain range of n is also given
(see Fig. 2).
2 Preliminaries
In this section, we give the basic definitions and lemmas that will be instrumental in
our analysis. For k ≥ 1, by a word ω of length k over the alphabet I := {1, 2, 3, 4}
it is meant that ω := ω1 ω2 · · · ωk , i.e., ω is a finite sequence of symbols over the
alphabet I . Here k is called the length of the word ω. If k = 0, i.e., if ω is a word
of length zero, we call it the empty word and is denoted by ∅. Length of a word ω
is denoted by |ω|. I ∗ denotes the set of all words over the alphabet I including the
empty word ∅. By ωτ := ω1 · · · ωk τ1 · · · τ it is meant that the word obtained from
the concatenations of the words ω := ω1 ω2 · · · ωk and τ := τ1 τ2 · · · τ for k, ≥ 0.
The maps Si : R2 → R2 , 1 ≤ i ≤ 4, will be the generating maps of the Sierpiński
carpet defined as before. For ω = ω1 ω2 · · · ωk ∈ I k , set Sω = Sω1 ◦ · · · ◦ Sωk and
Jω = Sω ([0, 1] × [0, 1]). For the empty word ∅, by S∅ we mean the identity mapping
on R2 , and write J = J∅ = S∅ ([0, 1] × [0, 1]) = [0, 1] × [0, 1]. The sets {Jω : ω ∈
{1, 2, 3, 4}k } are just the 4k squares in the kth level in the construction of the Sierpiński
46 M. K. Roychowdhury
carpet. The squares Jω1 , Jω2 , Jω3 and Jω4 into which Jω is split up at the (k + 1)th level
are called the basic squares of Jω . The set S = ∩k∈N ∪ω∈{1,2,3,4}k Jω is the Sierpiński
carpet and equals the support of the probability measure P given by P = 18 P ◦
S1−1 + 18 P ◦ S2−1 + 38 P ◦ S3−1 + 38 P ◦ S4−1 . Set s1 = s2 = s3 = s4 = 13 , p1 = p2 =
1
8
and p3 = p4 = 38 , and for ω = ω1 ω2 · · · ωk ∈ I k , write c(ω) := card({i : ωi =
3 or 4, 1 ≤ i ≤ k}), where card(A) of a set A represents the number of elements in
the set A. Then, for ω = ω1 ω2 · · · ωk ∈ I k , k ≥ 1, we have
1 3c(ω)
sω = and p ω = p ω 1
p ω 2
· · · p ω k
= .
3k 8k
Let us now give the following lemma.
Optimal Quantizers for a Nonuniform Distribution … 47
Let us now state the following lemma. The proof is similar to Lemma 2.2 in [2].
Lemma 2 Let P1 and P2 be the marginal distributions of the probability measure
P. Then,
−1 −1 −1 −1
• P1 = 18 P1 ◦ S(11) + 18 P1 ◦ S(21) + 38 P1 ◦ S(31) + 38 P1 ◦ S(41) and
−1 −1 −1 −1
• P2 = 8 P2 ◦ S(12) + 8 P2 ◦ S(22) + 8 P2 ◦ S(32) + 8 P2 ◦ S(42) .
1 1 3 3
Proof We have
1 −1 1 −1
E(X 1 ) = x d P1 = x d P1 ◦ S(11) +
x d P1 ◦ S(21)
8 8
3 −1 3 −1
+ x d P1 ◦ S(31) + x d P1 ◦ S(41)
8 8
1 1 1 1 2 3 1 3 1 2
= x d P1 + x+ d P1 + x d P1 + x+ d P1 ,
8 3 8 3 3 8 3 8 3 3
2 2
1 3 2 1 3 7
EX − , =E X1 − +E X2 − = V (X 1 ) + V (X 2 ) = .
2 4 2 4 32
Optimal Quantizers for a Nonuniform Distribution … 49
which implies
x − (a, b)2 d P = pω sω2 V + a(ω) − (a, b)2 . (2)
Jω
The expressions (1) and (2) are useful to obtain the optimal sets and the corresponding
quantization errors with respect to the probability distribution P. The Sierpiński
carpet has the maximum symmetry with respect to the vertical line x1 = 21 , i.e., with
respect to the line x1 = 21 the Sierpiński carpet is geometrically symmetric as well
as symmetric with respect to the probability distribution: if the two basic rectangles
of similar geometrical shape lie in the opposite sides of the line x1 = 21 , and are
equidistant from the line x1 = 21 , then they have the same probability.
In this section we determine the optimal sets of n-means for all n ≥ 2. First, prove
the following proposition.
Proposition 3.1 The set α = {a(1, 3), a(2, 4)}, where a(1, 3) = ( 61 , 43 ) and
a(2, 4) = ( 56 , 34 ), is an optimal set of two-means with quantization error V2 = 288
31
=
0.107639.
50 M. K. Roychowdhury
Proof Since the Sierpiński carpet has the maximum symmetry with respect to the
vertical line x1 = 21 , among all the pairs of two points which have the boundaries of
the Voronoi regions oblique lines passing through the point ( 21 , 34 ), the two points
which have the boundary of the Voronoi regions the line x1 = 21 will give the smallest
distortion error. Again, we know that the two points which give the smallest distortion
error are the centroids of their own Voronoi regions. Let (a1 , b1 ) and (a2 , b2 ) be the
centroids of the left half and the right half of the Sierpiński carpet with respect to the
line x1 = 21 , respectively. Then using (1), we have
1 1 3
(a1 , b1 ) = E(X : X ∈ J1 ∪ J3 ) = xd P = ,
P(J1 ∪ J3 ) J1 ∪J3 6 4
and
1 5 3
(a2 , b2 ) = E(X : X ∈ J2 ∪ J4 ) = xd P = , .
P(J2 ∪ J4 ) J2 ∪J4 6 4
Write α := { 16 , 34 , ,
5 3
6 4
}. Then, the distortion error is obtained as
1 3 5 3 31
min x − c2 d P = x − ( , )2 d P + x − ( , )2 d P = = 0.107639.
c∈α 6 4 6 4 288
J1 ∪J3 J2 ∪J4
1 5 1 35 515
min x − c2 d P ≥ x − ( , )2 d P + x − ( , )2 d P = = 0.111762,
c∈α 2 12 2 36 4608
J1 ∪J2 J33 ∪J34 ∪J43 ∪J44
Optimal Quantizers for a Nonuniform Distribution … 51
J33 ∪ J34 ∪ J43 ∪ J44 ∪ J313 ∪ J314 ∪ J323 ∪ J324 ∪ J413 ∪ J414 ∪ J423 ∪ J424 . Using
(1), we have E(X : X ∈ B) = ( 21 , 540
503
) which implies that b ≤ 540
503
. Now if a ≥ 13 ,
we have
1 1 1 503 106847
min x − c2 d P ≥ x − , 2 d P + x − , 2 d P = = 0.128818 > V2 ,
c∈α 2 3 2 540 829440
J1 ∪J2 B
which is a contradiction. So, we can assume that a < 13 . Then, J1 ∪ J1 ⊂ M(( 21 , a)|α)
and J3 ∪ J4 ⊂ M(( 21 , b)|α), and so ( 21 , a) = E(X : X ∈ J1 ∪ J2 ) = ( 21 , 41 ) and
( 21 , b) = E(X : X ∈ J3 ∪ J4 ) = ( 21 , 12
11
), and
1 1 1 11 13
min x − c2 d P = x − , 2 d P + x − , 2 d P = = 0.135417 > V2 ,
c∈α 2 4 2 12 96
J1 ∪J2 J3 ∪J4
which leads to another contradiction. Therefore, we can assume that the points in an
optimal set of two-means cannot lie on a vertical line. Hence, α = {( 16 , 34 ), ( 56 , 43 )}
forms an optimal set of two-means with quantization error V2 = 288
31
= 0.107639.
Remark 1 The set α in Proposition 3.1 forms a unique optimal set of two-means.
Proposition 3.2 The set α = {a(1, 2), a(3), a(4)}, where a(1, 2) = E(X : X ∈
J1 ∪ J2 ) = ( 21 , 41 ), a(3) = E(X : X ∈ J3 ) = ( 16 , 12
11
) and a(4) = E(X : X ∈ J4 ) =
( 6 , 12 ), forms an optimal set of three-means with quantization error V3 = 96
5 11 5
=
0.0520833.
Proof Let us first consider the three-point set β given by β = {a(1, 2), a(3), a(4)}.
Then, the distortion error is obtained as
min x − c2 d P
c∈α
= x − a(1, 2)2 d P + x − a(3)2 d P + x − a(4)2 d P = 0.0520833.
J1 ∪J2 J3 J4
1 3
(ai , bi )P(M((ai , bi )|α)) = , ,
(ai ,bi )∈α
2 4
52 M. K. Roychowdhury
which implies (ai ,bi )∈α ai P(M((ai , bi )|α)) = 21 and (ai ,bi )∈α bi P(M((ai , bi )|α))
= 34 . Thus, we conclude that all the points in an optimal set cannot lie in one side of
the vertical line x1 = 21 or in one side of the horizontal line x2 = 43 . Without any loss
of generality, due to symmetry we can assume that one of the optimal points, say
(a1 , b1 ), lies on the vertical line x1 = 21 , i.e., a1 = 21 , and the optimal points (a2 , b2 )
and (a3 , b3 ) lie on a horizontal line and are equidistant from the vertical line x1 = 21 .
Further, due to symmetry we can assume that (a2 , b2 ) and (a3 , b3 ) lie on the vertical
lines x1 = 16 and x1 = 56 respectively, i.e., a2 = 16 and a3 = 56 .
Suppose that ( 21 , b1 ) lies on or above the horizontal line x2 = 34 , and so ( 16 , b2 )
and ( 56 , b3 ) lie on or below the line x2 = 43 . Then, if 23 ≤ b2 , b3 ≤ 43 , we have
1
min x − c d P ≥ 2 2
min x − , b 2 d P = 0.0820313 > V3 ,
c∈α
3 ≤b≤ 4
2 3 6
J1 ∪J31 ∪J33
which is a contradiction. If 1
2
≤ b2 , b3 ≤ 23 ,
min x − c2 d P
c∈α
1 1 2 1
≥2 min x − , b 2 d P + x − , 2 d P + min x − , b 2 d P
1 2
2 ≤b≤ 3
6 6 3 3
4 ≤b≤1
2
J1 ∪J31 ∪J321 J33 J342 ∪J344
6521 281 277
=2 + + = 0.0649821 > V3 ,
442368 18432 110592
b3 ≤ 84
79
. Suppose that 21 ≤ b1 ≤ 43 . Then, writing A = J133 ∪ J321 ∪ J324 and B =
J11 ∪ J12 ∪ J14 ∪ J132 , we have
min x − c2 d P
c∈α
1 1 3 1 1
≥2 min x − , b 2 d P + x − , 2 d P + x − , 2 d P
3 79
4 ≤b≤ 84
6 6 4 2 2
J31 ∪J33 ∪J34 ∪J323 A B
588517 5347 6601
=2 + + = 0.0529346 > V3 ,
78299136 1327104 442368
which is a contradiction. So, we can assume that b1 < 21 . Suppose that 13 ≤ b1 < 21 .
Then, as 43 ≤ b2 ≤ 84
79
, we see that J31 ∪ J33 ∪ J34 ∪ J321 ∪ J323 ∪ J324 ⊂ M(( 61 , b2 )
|α). Then, writing A1 := J31 ∪ J33 ∪ J34 ∪ J321 ∪ J323 ∪ J324 and A2 := J322 ∪ J1331
∪ J1333 ∪ J1334 ∪ J13323 ∪ J13324 and A3 := J11 ∪ J12 ∪ J14 ∪ J131 ∪ J132 ∪ J134 ∪
J13322 , we have
min x − c2 d P
c∈α
1 1 3 2 1 1 2
≥2 min x − , b 2 d P + x − , d P + x − , dP
3
4 ≤b≤ 79
84
6 6 4 2 3
A1 A2 A3
242191 4135547 31584803
=2 + + = 0.0521401 > V3 ,
27869184 1146617856 2293235712
which gives a contradiction. So, we can assume that b1 ≤ 13 . Then, notice that J11 ∪
J12 ∪ J132 ∪ J141 ∪ J142 ∪ J144 ∪ J21 ∪ J22 ∪ J241 ∪ J231 ∪ J232 ∪ J233 ⊂ M(( 21 , b1 )
|α) which implies that b1 ≥ 13 68
. Thus, we have 1368
≤ b1 ≤ 13 . Suppose that 43 ≤
b2 , b3 ≤ 6 . Then,
5
min x − c2 d P
c∈α
1 1
≥2 min x − , b 2 d P + min x − , b 2 d P
3 ≤b≤ 5 6 13 ≤b≤ 1 2
J3 4 6 J11 ∪J12 ∪J14 ∪J131 ∪J132 68 3
1 3 2 1 1 2
+ x − , dP + x − , dP
6 4 2 3
J1331 ∪J1333 ∪J1334 J134
3 147359 32969 3881
=2 + + + = 0.054808 > V3 ,
256 15261696 10616832 1327104
which leads to a contradiction. So, we can assume that 56 < b2 , b3 ≤ 1. Then, we have
J1 ∪ J2 ⊂ M(( 21 , b1 )|α), J3 ⊂ M(( 61 , b2 )|α) and J4 ⊂ M(( 65 , b3 )|α) which yield
that ( 21 , b1 ) = a(1, 2), ( 16 , b2 ) = a(3) and ( 56 , b3 ) = a(4), and the quantization error
is V3 = 96 5
= 0.0520833. Thus, the proof of the proposition is complete.
Proposition 3.3 The set α = {a(1), a(2), a(3), a(4)} forms an optimal set of four-
means with quantization error V4 = 288
7
= 0.0243056.
54 M. K. Roychowdhury
Proof Let us consider the four-point set β given by β := {a(1), a(2), a(3), a(4)}.
Then, the distortion error is given by
4
7
min x − c d P =
2
x − a(i)2 d P = = 0.0243056.
c∈β
i=1 Ji 288
If all the points of α are below the line x2 = 43 , i.e., if b < 43 for all (a, b) ∈ α, then by
(3), we see that 43 = (a,b)∈α b P(M((a, b)|α)) < (a,b)∈α 43 P(M((a, b)|α)) = 34 ,
which is a contradiction. Similarly, it follows that if all the points of α are above the
line x2 = 43 , or left of the line x1 = 21 , or right of the line x1 = 21 , a contradiction will
arise. Suppose that all the points of α are on the line x2 = 43 . Then, for (x1 , x2 ) ∈
∪i,4 j=3 Ji j , we have minc∈α (x1 , x2 ) − c ≥ 365
, and for (x1 , x2 ) ∈ ∪i,2 j=1 Ji j , we have
minc∈α (x1 , x2 ) − c ≥ 36 , which implies that
23
min x − c d P ≥ 4
2
min (x1 , x2 ) − c d P + 4
2
min (x1 , x2 ) − c2 d P
c∈α c∈α c∈α
J33 J11
5 2 23 2 377
=4 P(J33 ) + 4 P(J11 ) = = 0.0363619 > V4 ,
36 36 10368
which is a contradiction. Thus, we see that all the points of α cannot lie on x2 = 34 .
Similarly, all the points of α cannot lie on x1 = 21 . Recall that the Sierpiński carpet
has maximum symmetry with respect to the line x1 = 21 . As all the points of α cannot
lie on the line x1 = 21 , due to symmetry we can assume that the points of α lie either
on the three lines x1 = 16 , x1 = 56 , and x1 = 21 , or on the two lines x1 = 16 and x1 = 56 .
Suppose α contains points from the line x1 = 21 . As α cannot contain all the points
from x1 = 21 , we can assume that α contains two points, say ( 21 , b1 ) and ( 21 , b2 ) with
b1 < b2 , from the line x1 = 21 which are in the opposite sides of the centroid ( 21 , 34 ),
and the other two points, say ( 61 , a1 ) and ( 56 , a2 ), from the lines x1 = 16 and x1 = 56 .
Then, if α does not contain any point from J3 ∪ J4 , we have
1 2 2 25
min x − c2 d P ≥ 2 x − , dP = = 0.0325521 > V4 ,
c∈α 6 3 768
J31 ∪J33
Optimal Quantizers for a Nonuniform Distribution … 55
which leads to a contradiction. So, we can assume that 56 < a1 , a2 ≤ 1. Then, we see
that J1 ∪ J2 ⊂ M(( 21 , b1 )|α) for b1 = 21 , and so the distortion error is
1 13
min x − c2 d P ≥ 2 min x − , b 2 d P = = 0.0338542 > V4
c∈α 3
J1 0≤b≤ 4 2 384
which is a contradiction. All these contradictions arise due to our assumption that
α contains points from the line x1 = 21 . So, we can assume that α cannot contain
any point from the line x1 = 21 , i.e., we can assume that α contains two points
from the line x1 = 16 and two points from the line x1 = 56 . Thus, we can take α :=
{( 61 , a1 ), ( 16 , b1 ), ( 56 , a2 ), ( 56 , b2 )} where a1 ≤ 43 ≤ b1 and a2 ≤ 43 ≤ b2 . Notice that
the Voronoi region of ( 16 , a1 ) contains J1 and the Voronoi region of ( 56 , a2 ) contains J2 .
If the Voronoi region of ( 16 , a1 ) contains points from J3 , we must have 21 (a1 + b1 ) ≥
2
3
which yields a1 ≥ 43 − b1 ≥ 43 − 43 = 12 7
, and similarly if the Voronoi region of
( 6 , a2 ) contains points from J4 , we must have a2 ≥ 12
5 7
. But, then
1 7
min x − c2 d P ≥ 2 x − , 2 d P + 2 x − a(33, 34)2 d P
c∈α 6 12
J1 J33 ∪J34
65
= = 0.0423177 > V4 ,
1536
which is a contradiction. So, we can assume that the Voronoi regions of ( 16 , a1 ) and
( 65 , a2 ) do not contain any point from J3 ∪ J4 . Thus, we have ( 16 , a1 ) = a(1) = ( 16 , 14 ),
( 65 , a2 ) = a(2) = ( 65 , 41 ), ( 16 , b1 ) = a(3) = ( 16 , 12
11
), and ( 65 , b2 ) = a(4) = ( 56 , 12
11
),
and the quantization error is V4 = 288 = 0.0243056. Thus, the proof of the proposi-
7
tion is complete.
= pω1 (sω1
2
V + a(ω1) − a(ω1, ω3)2 ) + pω3 (sω3
2
V + a(ω3) − a(ω1, ω3)2 ).
Notice that
1 1 3 1 3
a(ω1, ω3) = pω1 Sω1 , + pω3 Sω3 ,
pω1 + pω3 2 4 2 4
1 1 1 3 3 1 3
= 1 3 Sω1 , + Sω3 , ,
8
+8 8 2 4 8 2 4
Optimal Quantizers for a Nonuniform Distribution … 57
and similarly, we can prove the rest of the lemma. Thus, the proof of the lemma is
complete.
Remark 2 From the above lemma it follows that E(ω1, ω3) = E(ω2, ω4) >
E(ω1, ω2) > E(ω3) = E(ω4) > E(ω1) = E(ω2).
The following lemma gives some important properties about the distortion error.
Lemma 5 Let ω, τ ∈ I ∗ . Then
(i) E(ω) > E(τ ) if and only if E(ω1, ω3) + E(ω2, ω4) + E(τ ) < E(ω) +
E(τ 1, τ 3) + E(τ 2, τ 4);
(ii) E(ω) > E(τ 1, τ 3)(= E(τ 2, τ 4)) if and only if E(ω1, ω3) + E(ω2, ω4)
+ E(τ 1, τ 3) + E(τ 2, τ 4) < E(ω) + E(τ 1, τ 2) + E(τ 3) + E(τ 4);
(iii) E(ω1, ω3)(= E(ω2, ω4)) > E(τ 1, τ 3)(= E(τ 2, τ 4)) if and only
if E(ω1, ω2) + E(ω3) + E(ω4) + E(τ 1, τ 3) + E(τ 2, τ 4) < E(ω1, ω3) +
E(ω2, ω4) + E(τ 1, τ 2) + E(τ 3) + E(τ 4);
(iv) E(ω1, ω3)(= E(ω2, ω4)) > E(τ ) if and only if E(ω1, ω2) + E(ω3) +
E(ω4) + E(τ ) < E(ω1, ω3) + E(ω2, ω4) + E(τ 1, τ 3) + E(τ 2, τ 4);
(v) E(ω1, ω2) > E(τ ) if and only if E(ω1) + E(ω2) + E(τ ) < E(ω1, ω2) +
E(τ 1, τ 3) + E(τ 2, τ 4);
(vi) E(ω1, ω2) > E(τ 1, τ 3)(= E(τ 2, τ 4)) if and only if E(ω1) + E(ω2) +
E(τ 1, τ 3) + E(τ 2, τ 4) < E(ω1, ω2) + E(τ 1, τ 2) + E(τ 3) + E(τ 4);
(vii) E(ω1, ω2) > E(τ 1, τ 2) if and only if E(ω1) + E(ω2) + E(τ 1, τ 2) <
E(ω1, ω2) + E(τ 1) + E(τ 2);
(viii) E(ω) > E(τ 1, τ 2) if and only if E(ω1, ω3) + E(ω2, ω4) + E(τ 1, τ 2) <
E(ω) + E(τ 1) + E(τ 2).
58 M. K. Roychowdhury
5 31
L H S = E(ω1, ω2) + E(ω3) + E(ω4) + E(τ 1, τ 3) + E(τ 2, τ 4) = E(ω) + E(τ ),
21 63
31 5
R H S = E(ω1, ω3) + E(ω2, ω4) + E(τ 1, τ 2) + E(τ 3) + E(τ 4) = E(ω) + E(τ ).
63 21
In the following theorem, we give the induction formula to determine the optimal
sets of n-means for any n ≥ 2.
Theorem 1 For any n ≥ 2, let αn := {a(i) : 1 ≤ i ≤ n} be an optimal set of n-
means, i.e., αn ∈ Cn := Cn (P). For ω ∈ I ∗ , let E(ω), E(ω1, ω3) and E(ω2, ω4) be
defined by (4). Set
E(ω) if a(i) = a(ω) for some ω ∈ I ∗ ,
Ẽ(a(i)) :=
E(ωk, ω) if a(i) = a(ωk, ω) for some ω ∈ I ∗ ,
Then αn+1 (a( j)) is an optimal set of (n + 1)-means, and the number of such sets is
given by
card {αn+1 (a( j)) : a( j) ∈ W (αn )} .
αn ∈Cn
Ẽ(a(i)) + E(ω1, ω2) + E(ω3) + E(ω4) if a( j) = a(ω1, ω3) or a(ω2, ω4),
a(i)∈(αn \{a(ω1,ω3), a(ω2,ω4)})
Ẽ(a(i)) + E(ω1) + E(ω2) if a( j) = a(ω1, ω2),
a(i)∈(αn \{a( j)})
obtained in this case is strictly greater than the corresponding error obtained in the
case when a( j) ∈ W (αn ). Hence for any a( j) ∈ W (αn ), the set αn+1 (a( j)), where
⎧
⎪
⎪ (αn \ {a( j)}) ∪ {a(ω1, ω3), a(ω2, ω4)} if a( j) = a(ω),
⎨
(αn \ {a(ω1, ω3), a(ω2, ω4)}) ∪ {a(ω1, ω2), a(ω3), a(ω4)}
αn+1 (a( j)) :=
⎪
⎪ if a( j) = a(ω1, ω3) or a(ω2, ω4),
⎩
(αn \ {a( j)}) ∪ {a(ω1), a(ω2)} if a( j) = a(ω1, ω2),
Thus the proof of the theorem is complete (also see Note 3).
Remark 3 Once an optimal set of n-means is known, by using (2), the corresponding
quantization error can easily be calculated.
First, we explain about some notations that we are going to use in this section.
Recall that the optimal set of one-mean consists of the expected value of the
random variable X , and the corresponding quantization error is its variance. Let
αn be an optimal set of n-means, i.e., αn ∈ Cn , and then for any a ∈ αn , we
have a = a(ω), or a = a(ωi, ωj) for some ω ∈ I ∗ , where (i = 1, j = 3), (i =
2, j = 4), or (i = 1, j = 2). For ω = ω1 ω2 · · · ωk ∈ I k , k ≥ 1, in the sequel, we
will identify the elements a(ω) and a(ωi, ωj) by the sets {{ω1 , ω2 , . . . , ωk }} and
{{ω1 , ω2 , . . . , ωk , i}, {ω1 , ω2 , . . . , ωk , j}}, respectively. Thus, we can write
{{3, 1}, {3, 2}}, {{4, 1}, {4, 3}}, {{4, 2}, {4, 4}} ,
α9,2 = {{4, 3}}, {{4, 4}}, {{1, 1}, {1, 3}}, {{1, 2}, {1, 4}}, {{2, 1}, {2, 3}}, {{2, 2}, {2, 4}},
25
{{3, 1}, {3, 3}}, {{3, 2}, {3, 4}}, {{4, 1}, {4, 2}} with V9 = = 0.00964506;
2592
α10 = {{3, 3}}, {{3, 4}}, {{4, 3}}, {{4, 4}}, {{1, 1}, {1, 3}}, {{1, 2}, {1, 4}}, {{2, 1}, {2, 3}},
19
{{2, 2}, {2, 4}}, {{3, 1}, {3, 2}}, {{4, 1}, {4, 2}} with V10 = = 0.00733025,
2592
and so on.
Note 3 Notice that there is only one optimal set of n-means for n = 72. By the
notations used in Theorem 1, we can write α72 = {a(i) : 1 ≤ i ≤ 72}. Then,
W (α72 ) = {{{1, 3, 3}}, {{1, 3, 4}}, {{1, 4, 3}}, {{1, 4, 4}}, {{2, 3, 3}}, {{2, 3, 4}}, {{2, 4, 3}},
{{2, 4, 4}}, {{3, 1, 3}}, {{3, 1, 4}}, {{3, 2, 3}}, {{3, 2, 4}}, {{3, 3, 1}}, {{3, 3, 2}},
{{3, 4, 1}}, {{3, 4, 2}}, {{4, 1, 3}}, {{4, 1, 4}}, {{4, 2, 3}}, {{4, 2, 4}}, {{4, 3, 1}}, {{4, 3, 2}},
{{4, 4, 1}}, {{4, 4, 2}}}.
carpet generated by S(i, j) for all 1 ≤ i, j ≤ 2. Let P be the Borel probability mea-
−1 −1 −1 −1
sure on R2 such that P = 16 1
P ◦ S(1,1) + 16
3
P ◦ S(2,1) + 16
3
P ◦ S(1,2) + 16
9
P ◦ S(2,2) .
Then, P has support the Siepiński carpet S. For this probability measure, the optimal
sets of n-means and the nth quantization errors are not known yet for all n ≥ 2.
Acknowledgements The research of the author was supported by U.S. National Security Agency
(NSA) Grant H98230-14-1-0320
References
1. Abaya, E.F., Wise, G.L.: Some remarks on the existence of optimal quantizers. Stat. Prob. Lett.
2(6), 349–351 (1984)
2. Çömez, D., Roychowdhury, M.K.: Quantization for uniform distributions of Cantor dusts on
R2 . Topol. Proc. 56, 195–218 (2020)
3. Du, Q., Faber, V., Gunzburger, M.: Centroidal Voronoi tessellations: applications and algo-
rithms. SIAM Rev. 41(4), 637–676 (1999)
4. Gersho, A., Gray, R.M.: Vector Quantization and Signal Compression. Kluwer Academy pub-
lishers, Boston (1992)
5. Gray, R.M., Kieffer, J.C., Linde, Y.: Locally optimal block quantizer design. Inf. Control 45,
178–198 (1980)
6. György, A., Linder, T.: On the structure of optimal entropy-constrained scalar quantizers. IEEE
Trans. Inf. Theory 48(2) (2002)
7. Graf, S., Luschgy, H.: Foundations of Quantization for Probability Distributions. Lecture Notes
in Mathematics, vol. 1730. Springer, Berlin (2000)
8. Graf, S., Luschgy, H.: The Quantization of the Cantor Distribution. Math. Nachr. 183, 113–133
(1997)
9. Gray, R.M., Neuhoff, D.L.: Quantization. IEEE Trans. Inf. Theory 44(6), 2325–2383 (1998)
10. Hutchinson, J.: Fractals and self-similarity. Indiana Univ. J. 30, 713–747 (1981)
11. Roychowdhury, M.K.: Quantization and centroidal Voronoi tessellations for probability mea-
sures on dyadic Cantor sets. J. Fractal Geom. 4, 127–146 (2017)
12. Roychowdhury, L.: Optimal quantization for nonuniform Cantor distributions. J. Interdiscip.
Math. 22(8), 1325–1348 (2019)
13. Zam, R.: Lattice Coding for Signals and Networks: A Structured Coding Approach to Quanti-
zation, Modulation, and Multiuser Information Theory. Cambridge University Press (2014)
Fractal Dimension for a Class
of Complex-Valued Fractal Interpolation
Functions
Abstract There are many research papers dealing with fractal dimension of real-
valued fractal functions in the recent literature. The main focus of our paper is to
study the fractal dimension of complex-valued functions. This paper also highlights
the difference between dimensional results of the complex-valued and real-valued
fractal functions. We study the fractal dimension of the graph of complex-valued
function g(x) + i h(x), compare its fractal dimension with the graphs of functions
g(x) + h(x) and (g(x), h(x)) and also obtain some bounds. Moreover, we study
the fractal dimension of the graph of complex-valued fractal interpolation function
associated with a germ function f , base function b, and scaling functions αk .
1 Introduction
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 63
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_6
64 M. Verma et al.
calculated the fractal dimension of the graphs of FIFs. In 1991, Massopust [16]
determined BD of the graphs of vector-valued FIFs. Later, Hardin and Massopust
[12] described the FIFs from Rn to Rm and determined the BD of its graph. The
reader can see some recent works on fractal dimension of fractal functions defined
on different domains such as Sierpinski gasket [1, 20], rectangular domain [8], and
interval [13, 22]. To the best of our knowledge, we may say that there is no work
available for the dimension of complex-valued fractal functions. Here, we give some
basic results for complex-valued FIF and provide some results to convince the reader
that there is some difference between the dimensional result of the complex-valued
and real-valued fractal functions.
In 1986, Mauldin and Williams [17] were the pioneers who studied the problem
of decomposition of the continuous functions in terms of the Hausdorff dimension
(HD). They proved the existence of decomposition of any continuous function on
[0, 1] into a sum of two continuous functions, where each has HD one. Later in 2000,
Wingren [23] gave a technique to construct the above decomposition of Mauldin and
Williams. Moreover, he proved the same type of result as Mauldin and Williams for
the lower BD. Bayart and Heurteaux [7] also proved the similar result for HD β = 2,
and raised the question for β ∈ [1, 2]. Recently, in 2013, Jia Liu and Jun Wu [15]
solved the question which was raised by Bayart and Haurteaux. More precisely, they
showed that, for any given β ∈ [1, 2], we can decompose any continuous function
on [0, 1] into a sum of two continuous functions, where each has HD β. Falconer
and Fraser [10] proved that the upper BD of the graph of the sum of two continuous
functions depends on BD of both graphs.
In [10, 15], the authors determined that HD of the graph of g + h does not depend
on HD of the graph of g and h, whereas the upper BD depends on both. Motivated by
this, we think about the behavior of HD of the graph of g + i h, whether it depends on
HD of graphs of g and h or not. We obtained an affirmative answer to this question.
Also, the upper BD of g + i h depends on the upper BD of g and h which is quite
different from the upper BD of g + h. Finally, we studied some relations between
fractal dimensions of the graphs of g(x) + i h(x), g(x) + h(x), and (g(x), h(x)).
The article is arranged as follows. In the upcoming Sect. 1.1, we provide some
preliminary results and the required definition for the next section. Section 2 contains
some results related to the dimension of the complex-valued continuous functions and
the FIFs. In this section, first, we establish some results to form a connection between
the fractal dimension of complex-valued and real-valued continuous functions. After
that, we determine the fractal dimension of FIFs under some assumptions. We also
obtain some conditions under which α-fractal function becomes Hölder continuous
function and bounded variation function, and calculate its fractal dimension.
1.1 Preliminaries
log Nδ (F)
dim B F = lim ,
δ→0 − log δ
where Nδ (F) is the minimum number of sets of diameter δ > 0 that can cover F. If
this limit does not exist, then liminf and limsup are known as the lower and upper
BDs, respectively.
Definition 3 For any δ > 0 and l ≥ 0, let Pδl (F) := sup n |On | , where {On } is
l
a set of the pairwise disjoint balls of the diameter less than or equal to 2δ with centers
in F ⊆ Y . One can observe that Pδs decreases as δ decreases. Thus, limδ→0 Pδl (F) =
P0l (F) exists. We define l-dimensional packing measure as
∞
P (F) = inf
l
P0l (Fn ) :F⊂ Fn .
n n=1
With the help of packing measure, we define the packing dimension (PD) as follows:
Note The graph of function f will be denoted by G( f ) throughout this paper. For
σ ∈ (0, 1], the Hölder space Hσ ([a, b], R),
Hσ ([a, b], R) := { f : [a, b] → R : | f (t1 ) − f (t2 )| ≤ C f |t1 − t2 |σ , ∀ t1 , t2 ∈ [a, b], for some C f > 0}.
Let (Y, d) be a complete metric space. The class of all non-empty compact subsets
of Y is denoted by H (Y ). Let A1 , A2 ∈ H (Y ). The Hausdorff metric D on H (Y ) is
given by
k (x1 , y1 ) = yk , k (x N , y N ) = yk+1 ,
where 0 ≤ sk < 1 and (t, ξ1 ), (t, ξ2 ) ∈ L × C. For every k ∈ T , we can take partic-
ular choices of Pk and k as
Since Pk satisfy Eq. (1), we can obtain the unique constants ak and dk . The constant
multiplier αk is said to be a scaling factor and |αk | < 1. The map qk : L → C is a
Fractal Dimension for a Class of Complex-Valued Fractal … 67
We can adapt the idea of the construction of FIF. The set of complex-valued continu-
ous functions defined on L = [x1 , x N ] ⊂ R is denoted by C(L , C), with the sup norm.
Let f be a given function in C(L , C), known as the germ function. For constructing
the IFS, we consider the following assumptions:
1. Let := {(x1 , x2 , . . . , x N ) : x1 < x2 < · · · < x N } be a partition of L = [x1 , x N ].
2. Let αk ∈ C(L , C) with αk ∞ = max{|αk (t)| : t ∈ L} < 1, for all k ∈ T . These
αk are called the scaling functions.
3. Let b ∈ C(L , C) with b = f and b(xi ) = f (xi ) for i ∈ {1, N } and, named as
the base function.
Motivated by [2, 3], Navascués [18] considered the following set of functions:
Pk (t) = ak t + dk ,
(2)
k (t, ξ) = αk (t)ξ + f Pk (t) − αk (t)b(t).
α
has a unique attractor and this attractor is the graph of a continuous function f ,b :
α α α
L → C with f ,b (xk ) = f (xk ), k ∈ T . For simplicity, we write f ,b by f . The real
valued f α is widely known as α-fractal function, see, for instance, [1, 8, 13, 22].
Moreover, f α satisfies
2 Main Theorems
Proof Using the same idea as in part (1) of Lemma 1, one can easily prove this.
and
dim B (G( f )) = dim B (G(g)).
Proof Using the idea of part (2) in Lemma 1, one can obtain the required result.
dim H (G(g + i h)) = dim H (G(g + h)) = dim H (G(g, h)) = dim H (G(g)),
dim B (G(g + i h)) = dim B (G(g + h)) = dim B (G(g, h)) = dim B (G(g)),
dim P (G(g + i h)) = dim P (G(g + h)) = dim P (G(g, h)) = dim P (G(g)).
is a bi-Lipschitz map, see part(1) of Lemma 1. Now, from [9, Corollary 2.4], we get
70 M. Verma et al.
is a bi-Lipschitz map, see part (2) of Lemma 1. Thus, from [9, Corollary 2.4], we
obtain
dim H (G(g, h)) = dim H (G(g)). (5)
dim H (G(g + i h)) = dim H (G(g + h)) = dim H (G(g, h)) = dim H (G(g)).
Since upper BD, lower BD, and PD satisfy bi-Lipschitz invariance property, the rest
follows.
Since upper BD, lower BD, and PD also fulfill the bi-Lipschitz invariance property,
we complete the proof.
Remark 2 The Peano space filling curve g : [0, 1] → [0, 1] × [0, 1] is a function,
which is 21 -Hölder, see details [14]. The component functions satisfy dim H G(g1 ) =
dim H G(g2 ) = 1.5. Since g is a space filling curve, we have dim H G(g ≥ 2. Now,
consider a complex-valued mapping f (x) = g1 (x) + ig2 (x), and using Lemma 3,
dim H G( f ) ≥ 2. Now, we conclude this remark. If f ∈ Hσ ([0, 1], R) for σ ∈ (0, 1),
Fractal Dimension for a Class of Complex-Valued Fractal … 71
then dim H G( f ) ≤ 2 − σ. But if f ∈ Hσ ([0, 1], C) for σ ∈ (0, 1). Is 2 − σ again the
upper bound of dim H G( f )? From the above, we may not get a positive answer.
From Remark 2, it is clear that, in general, the dimensional results for the complex-
valued and real-valued functions are not same. Now, we are ready to give some
dimensional results for the complex-valued FIFs.
We define a metric D0 on L × C by
Proof For upper bound of dim H (G( f α )), one may follow Proposition 9.6 in [9].
For the lower bound of dim H (G( f α )), we shall proceed in the following way.
Let V = (x1 , x N ) × C. Thus, we have
Wi (V ) ∩ W j (V ) = ∅,
Pi (x1 , x N ) ∩ P j (x1 , x N ) = ∅, ∀ i = j ∈ T.
We can observe that V ∩ G( f α ) = ∅, thus the IFS I satisfies the SOSC. Then, there
is an index i ∈ T ∗ with Wi (G( f α )) ⊂ V, where T ∗ := ∪n∈N {1, 2, . . . , N − 1}n .
We denote Wi (G( f α )) by (G( f α ))i for any i ∈ T ∗ . Now, one can observe that
for each n ∈ N, the sets {(G( f α )) ji : j ∈ T n } is disjoint. Then, for each n ∈
N, IFS Ln = {W ji : j ∈ T n } satisfies all the assumptions of Proposition 9.7 in
[9]. Hence, by Proposition 9.7 in [9], if A∗n is an attractor of Ln , then rn ≤
∗ rn ∗ α
dim H (An ), where rn is given by j∈T n c ji = 1. Since An ⊂ G( f ), we have
∗ α α
rn ≤ dim H (An ) ≤ dim H (G( f )). Suppose that dim H (G( f )) < r. Thus, rn < r .
Let cmax = max{c1 , c2 , . . . , c N −1 }. We have
dim H (G( f α ))−r
ci−rn =
α
n(dim H (G( f ))−r )
crjn ≥ crj c j ≥ crj cmax .
j∈T n j∈T n j∈T n
72 M. Verma et al.
Hσ (L , C) := {h : L → C : h
is a Hölder function with exponent σ}. We know that (Hσ (L , C), .H ) is a complete
norm linear space, where hH := h∞ + [h]σ and
|h(t1 ) − h(t2 )|
[h]σ = sup .
t1 ,t2 ∈L ,t1 =t2 |t1 − t2 |σ
αH
Theorem 3 Let f, b, α ∈ Hσ (L , C). Set c := min{ak : k ∈ T }. If cσ
< 1
N
, then
f α ∈ Hσ (L , C).
Since (Hσ (L , C), .H ) is a complete norm linear space, Hσf (L , C) is a com-
plete metric space under the metric endowed by .H . Now, we consider a map
S : Hσf (L , C) → Hσf (L , C) as follows
Our claim is that the map S is well defined and a contraction on Hσf (L , C).
Fractal Dimension for a Class of Complex-Valued Fractal … 73
|Sh(t1 ) − Sh(t2 )|
σ ≤ N = max sup
k∈T t1 ,t2 ∈L k ,t1 =t2 |t1 − t2 |σ
| f (t1 ) − f (t2 )|
≤N max sup
k∈T t1 ,t2 ∈L k ,t1 =t2 |t1 − t2 |σ
|αk (Pk−1 (t1 ))|(h − b)(Pk−1 (t1 )) − (h − b)(Pk−1 (t2 ))
+ sup
t1 ,t2 ∈L k ,t1 =t2 |t1 − t2 |σ
|(h − b)(Pk−1 (t2 ))|αk (Pk−1 (t1 )) − αk (Pk−1 (t2 ))
+ sup
t1 ,t2 ∈L k ,t1 =t2 |t1 − t2 |σ
α∞ h − b∞
≤ N [ f ]σ + N σ [h]σ + [b]σ + N [α]σ ,
c cσ
|αk (t1 )−αk (t2 )|
where [α]σ = max sup |t1 −t2 |σ
. Thus, S is well-defined map. For g1 , g2 ∈
k∈T t1 =t2 ,t1 ,t2 ∈L
Hσf (L , C), we have
Since α
cσ
H
< N1 , S is contraction. Using Banach fixed point theory, there is a unique
function f ∈ Hσf (L , C) with S( f α ) = f α . Hence, we are done.
α
for some constant Cα, f,b > 0 and for i = 1, 2. Firstly, we try to obtain the upper
bound of the upper BD of G( f iα ) ∀ i ∈ {1, 2} as follows: For δ ∈ (0, 1), assume
m = 1δ , where . denotes the ceiling function and Nδ (G( f iα )) is the least number
of δ-squares that covers G( f iα ), we have
m−1
R fiα [(r δ, (r + 1)δ]
Nδ (G( f iα )) ≤ 2m +
r =0
δ
m−1
R f α [r δ, (r + 1)δ]
1
≤2 +1 + i
(7)
δ r =0
δ
m−1
1
≤2 +1 + Cα, f,b δ σ−1 .
δ r =0
log Nδ (G( f iα ))
dim B G( f iα ) = lim ≤2−σ ∀ i = 1, 2.
δ→0 − log δ
w
Nδ (G( f 1α )) ≥ max 1, c−n R f1α [r δ, (r + 1)δ]
r =0
w
≥ c−n R f1α [r δ, (r + 1)δ]
r =0
w
≥ M0 c−n cnσ
r =0
= M0 cn(σ−2) .
Thus, we have
Similarly, we get
dim B G( f 2α ) ≥ 2 − σ,
The space BV(L , C) is Banach under the norm hBV := |h(x0 )| + V (h, L , C).
Proof Following Theorem 3.11 and [13, Theorem 3.24 ], one may complete the
proof.
Remark 5 The above theorem will reduce to [13, Theorem 3.24] when all functions
f, b and αk are real-valued.
References
1. Agrawal, V., Som, T.: Fractal dimension of α-fractal function on the Sierpiński Gasket. Eur.
Phys. J. Spec. Top. 230(21), 3781–3787 (2021)
2. Barnsley, M.F.: Fractal functions and interpolation. Constr. Approx. 2(1), 303–329 (1986)
3. Barnsley, M.F.: Fractal Everywhere. Academic, Orlando, Florida (1988)
4. Barnsley, M.F., Elton, J., Hardin, D.P., Massopust, P.R.: Hidden variable fractal interpolation
functions. SIAM J. Math. Anal. 20(5), 1218–1242 (1989)
5. Barnsley, M.F., Harrington, A.N.: The calculus of fractal interpolation functions. J. Approx.
Theory 57(1), 14–34 (1989)
6. Barnsley, M.F., Massopust, P.R.: Bilinear fractal interpolation and box dimension. J. Approx.
Theory 192, 362–378 (2015)
7. Bayart, F., Heurteaux, Y.: On the Hausdorff dimension of graphs of prevalent continuous
functions on compact sets. Further Developments in Fractals and Related Fields, pp. 25–34
(2013)
8. Chandra, S., Abbas, S.: The calculus of bivariate fractal interpolation surfaces. Fractals 29(3),
2150066 (2021)
9. Falconer, K.J.: Fractal Geometry: Mathematical Foundations and Applications. Wiley, New
York (1999)
10. Falconer, K.J., Fraser, J.M.: The horizon problem for prevalent surfaces. Math. Proc. Camb.
Philos. Soc. 151(2), 355–372 (2011)
11. Hardin, D.P., Massopust, P.R.: The capacity for a class of fractal functions. Commun. Math.
Phys. 105(3), 455–460 (1986)
12. Hardin, D.P., Massopust, P.R.: Fractal interpolation functions from Rn to Rm and their projec-
tions. Zeitschrift für Analysis und ihre Anwendungen 12(3), 535–548 (1993)
13. Jha, S., Verma, S.: Dimensional analysis of α-fractal functions. RM 76(4), 1–24 (2021)
Fractal Dimension for a Class of Complex-Valued Fractal … 77
14. Kono, N.: On self-affine functions. Jpn. J. Appl. Math. 3(2), 259–269 (1986)
15. Liu, J., Wu, J.: A remark on decomposition of continuous functions. J. Math. Anal. Appl.
401(1), 404–406 (2013)
16. Massopust, P.R.: Vector-valued fractal interpolation functions and their box dimension. Aequa-
tiones Math. 42(1), 1–22 (1991)
17. Mauldin, R.D., Williams, S.C.: On the Hausddorff dimension of some graphs. Trans. Am.
Math. Soc. 298(2), 793–803 (1986)
18. Navascués, M.A.: Fractal polynomial interpolation. Zeitschrift für Analysis und ihre Anwen-
dungen 24(2), 401–418 (2005)
19. Roychowdhury, M.K.: Hausdorff and upper box dimension estimate of hyperbolic recurrent
sets. Israel J. Math. 201(2), 507–523 (2014)
20. Sahu, A., Priyadarshi, A.: On the box-counting dimension of graphs of harmonic functions on
the Sierpiński gasket. J. Math. Anal. Appl. 487(2), 124036 (2020)
21. Schief, A.: Separation properties for self-similar sets. Proc. Am. Math. Soc. 122(1), 111–115
(1994)
22. Verma, S., Viswanathan, P.: A revisit to α-fractal function and box dimension of its graph.
Fractals 27(6), 1950090 (2019)
23. Wingren, P.: Dimensions of graphs of functions and lacunary decompositions of spline approx-
imations. Real Analysis Exchange, pp. 17–26 (2000)
A Note on Complex-Valued Fractal
Functions on the Sierpiński Gasket
1 Introduction
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 79
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_7
80 V. Agrawal and T. Som
functions that exhibit self-similarity under magnification, Barnsley [5] proposed the
idea of FIFs by using IFS on a compact interval of R, a number of his followers
further explored the field of fractal interpolation and approximation. Over the past two
decades, this field of study has experienced a tremendous expansion, and intriguing
new possibilities have emerged. Celik et al. [7] have extended Barnsley’s work by
introducing FIFs on the SG. Furthermore, Ruan [17] generalized the notion of FIFs
on a fractal domain. We suggest the reader to [11] for a comprehensive development
of the theory of IFS and fractal functions, and fractal surfaces. Massopust [11] has
discussed local IFS as well as novel fractal interpolation. One of the most prominent
example of post-critically finite fractal in fractal theory is SG, which was invented
by Polish mathematician W. Sierpiński in 1915. Kigami [10] has explored fractal
analysis on a self-similar set. FIFs and linear FIFs have been presented by Ruan
[17] on a self-similar set. He furthermore showed that linear FIFs exhibit finite
energy. On SG, Ri and Ruan [16] have examined some fundamental features of
uniform FIFs, a specific family of FIFs. Verma et al. [20] have studied the fractal
operator associated with the bivariate FIFs. In [23], through fractal dimensions, the
authors have developed a novel concept of constrained approximation. On the SG,
the dimensions of graphs of FIFs have been investigated by Sahu and Priyadarshi
in [18]. Fractal dimension is a crucial aspect of fractal geometry, since it provides
details about the geometric structure of the objects it examines. There are several
notions of fractal dimension, with the Hausdorff dimension and the box dimension
being the most prevalent and these dimensions of the graphs of FIFs have been
analyzed in detail. In the oscillation spaces, Verma and Sahu [24] have investigated
the fractal dimensions of various functions. They have also developed a certain
bounded variation concepts for the SG, from which they deduced several dimensional
conclusions. T. Bagby [4] has explored mean approximation in the plane C using
complex analytic functions. Recently, FIFs have been produced by Prasad and Verma
[14] on the products of two SG. Furthermore, they have gathered certain observations
about the smoothness of the produced FIFs. The author of [15] has demonstrated that
the graphs of FIFs formed on the SG are attractors of some IFS and provide the new
nonlinear FIFs. Furthermore, Navascués et al. [13] have studied the vector-valued
interpolation functions on the SG through a certain family of fractal functions. In [8,
9], dimensions of FIFs are investigated more rigorously by Jha and her collaborators.
In [20]–[25], Verma and his collaborators have discussed the dimension of α-fractal
functions in more detail. They have discussed the class of univariate and bivariate
FIFs and constrained approximation in their research. In [1], Agrawal and Som
have studied the fractal dimension of α-fractal function on the SG and the same
authors in [2] have investigated the L p approximation using fractal functions on the
SG. Agrawal et al. [3] have further introduced the concept of dimension-preserving
approximation for bivariate continuous functions. In [12], Navascués has defined an
α-fractal function associated with the square-integrable complex-valued function on
the real compact interval. Furthermore, she has explored significant properties of the
associated fractal operator. Motivated by her work, we define the α-fractal function
associated with the square-integrable complex-valued function on SG. Furthermore,
we study the properties of the associated fractal operator.
A Note on Complex-Valued Fractal Functions on the Sierpiński Gasket 81
This paper becomes more intriguing in many ways because it will use the definition
of self-similar measures on SG to get most of the important results. The paper is
organized as follows: In Sect. 2, we give a few preliminaries required for the paper.
We denote the space of all the real-valued continuous functions defined on SG by
C(SG) and space of all the complex-valued continuous functions defined on SG by
C(SG, C). Let L2 (SG) = {h | h : SG → R and h2 < ∞} and L2 (SG, C) = {h|h :
SG → C and h2 < ∞}. In Sect. 3, we determine the bounds of the real-valued
fractal operator by imposing certain conditions. In Sect. 4, we establish some bounds
of the complex-valued fractal operator FCα : C(SG, C) → C(SG, C) that maps to a
complex-valued continuous function f to its fractal version f α .
2 Technical Introduction
{H ; W j , j = 1, 2, . . . , k},
The symbol D(H ) stands for the class of all non-empty compact subsets of H .
The map W acting on D(H ) endowed with Hausdorff metric h d is a contraction
mapping. The contraction ratio α of W is equal to max{α j : 1 ≤ j ≤ k}, where α j
is the contraction ratio of W j .Then, by the Banach contraction principle, we get a
unique non-empty compact subset F∗ , which satisfies F∗ = ∪kj=1 W j (F∗ ), the set F∗
is referred to be an attractor of the IFS. One can refer to [6] for more information.
We use a self-similar measure to prove all the results in this paper and this measure
arises from the IFS with probability vectors, which is the fixed point of the Markov
operator, i.e., the invariant measure of the IFS with probability vectors. To understand
the self-similar measure, we refer the reader to [6].
We begin with a brief review of the relevant definitions and preliminary information
on the SG. The reader can find further information at [7, 10, 19].
SG is constructed with the help of a very important technique, which is known
as the IFS. Here, we generate this system via three contraction mappings defined on
the R2 plane. Let V0 = { p j : 1 ≤ j ≤ 3} is the collection of the equilateral triangle’s
82 V. Agrawal and T. Som
three vertices. Corresponding to these three points, contraction mappings are defined
as follows:
1
ψ j (t) = (t + p j ).
2
The following IFS provides the SG as an attractor,
For n, N ∈ N, let us represent the set of all words having a length of n by {1, 2, 3}n ,
that is, if i ∈ {1, 2, 3}n , then i = i 1 , i 2 , . . . , i n , where i j ∈ {1, 2, 3}. We define it for
i ∈ {1, 2, 3} N and we further write S instead of {1, 2, 3} N .
ψi = ψi1 ◦ ψi2 ◦ · · · ◦ ψi N .
1
3
μs = μs ◦ ψi−1 ,
3 i=1
(1)
1
3
dμs = d(μs ◦ ψi−1 ).
3 i=1
graph( f α ). The set graph( f α ) = {(x, f α (x)) : x ∈ SG} is the graph of a square-
integrable function f α : SG → R which satisfies f α |VN = f |VN . If α∞ < 1N ,
32
then f α satisfies the following functional equation:
f α (t) = f (t) + αi (ψi−1 (t))( f α − b) ψi−1 (t) ∀ t ∈ ψi (SG), i ∈ S. (4)
Proof Let L2f (SG) = {g ∈ L2 (SG) : g|V0 = f |V0 }. One can derive directly that set
L2f (SG) is a closed subset of L2 (SG) by ordinary calculations. Since (L2 (SG), .2 )
is a Banach space, we get L2f (SG) is a CMS endowed with the norm .2 . Consider
the map T defined by T : L2f (SG) → L2f (SG) by
for every t ∈ ψi (SG), where i ∈ S. It is easy to derive that T is well defined. Now,
consider g, h ∈ L2f (SG) to obtain the following:
(T g)(t) − (T h)(t) = αi (ψi−1 (t)) (g − b)(ψi−1 (t)) − αi (ψi−1 (t)) (h − b)(ψi−1 (t))
N
Therefore, we obtain that T g − T h2 ≤ 3 2 α∞ g − h2 . Using α∞ = maxS
N
αi ∞ and 3 2 α∞ < 1, this implies that T is a contraction map on L2f (SG). Banach
contraction principle is used to get a unique fixed point of T , namely f α ∈ L2f (SG).
Hence, T ( f α ) = f α , it immediately follows that f α (t) = E i ψi−1 (t), f α (ψi−1 (t))
∀ t ∈ ψi (SG), i ∈ S. This is further represented as f α (ψi (t)) = E i (t, f α (t)) for
i ∈ S. It can be verified that the graph( f α ) is an attractor of the IFS and hence
N
3 2 α∞
b2 . Hence, one gets f α − f 2 ≤ N f − b2 .
1−3 2 α∞
Finally, we get the following:
N
3 2 α∞
f α 2 − f 2 ≤ f α − f 2 ≤ N f − b2 .
1 − 3 2 α∞
Here, I d represents the identity operator, and .2 is the square-integrable operator
norm and = max{I d − L 1 2 }, I d − L 2 2 , . . . , I d − L N0 2 }, > 0.
Note that
f nα − f f nα (t) − f (t) dμs (t)
2 2
2
≤
SG
2
f nα − f 22 ≤ αi 2∞ ( f nα − L n ( f )) t˜ dμs (t˜)
i∈S SG
2
≤ α2∞ ( f nα − L n ( f )) t˜ dμs (t˜)
i∈S SG
≤3 N
α2∞ f nα − L n ( f )22 .
Consequently, we get
f nα − f ≤ 3 2 α∞ f nα − L n ( f )2 .
N
2
(7)
f nα − f ≤ 3 2 α∞ f nα − f 2 + f − L n ( f )2 .
N
Hence, from the part above being bounded for the perturbation error, we have
N
3 2 α∞
f nα − f 2
≤ N f − L n ( f )2 . (8)
1 − 3 2 α∞
N
3 2 α∞
f nα − f 2
≤ N f 2 I d − L n 2 . (9)
1 − 3 2 α∞
Therefore, we obtain
I d − L n 2 < ∀ n ∈ N. (11)
Finally, applying the definition of a fractal operator, we have the following inequality:
N
3 2 α∞
Fnα 2 − I d2 ≤ Fnα − I d2 ≤ N . (13)
1 − 3 2 α∞
Therefore,
N
3 2 α∞
Fnα 2 ≤ I d2 + N .
1 − 3 2 α∞
Since
N
I d2 ≤ 3 2 ,
hence, we have
N
3 2 α∞
Fnα 2
N
≤3 +
2
N .
1 − 3 2 α∞
Moreover, figures, (Figs. 1 and 2) below give us a better understanding of how param-
eter changes affect graph( f α ).
In this section, we obtain the bound for the operator norm of FCα . Furthermore, we
find the bounds of the perturbation error.
Theorem 3 Consider a real-valued square-integrable fractal operator F α on
C(SG). The complex-valued fractal operator FCα : C(SG, C) → C(SG, C) defined
as
Hence,
Finally, we get
FCα ≤ F α .
FCα (h) = F α (h r e ) + i F α (h im )
We now use Theorem 2 and a few inequalities to obtain the desired result
FCα (h) − h22 = |FCα (h) − h|2 dμs
SG
= |FCα (h r e ) − h r e |2 + |FCα (h im ) − h im |2 dμs
SG
= F α (h r e ) − h r e 22 + F α (h im ) − h im 22
≤ 3 N α2∞ F α (h r e ) − L(h r e )22 + F α (h im ) − L(h im )22
= 3 N α2∞ |F α (h r e ) − L(h r e )|2 dμs + |F α (h im ) − L(h im )|2 dμs
SG SG
= 3 N α2∞ |FCα (h) − L C (h)|2 dμs
SG
= 3 N α2∞ FCα (h) − L C (h)22 .
(14)
Finally, we get
That is, FCα (h) − h2 ≤ 3 2 α∞ FCα (h) − L C (h)2 . Thanks to the triangle
N
inequality,
N
3 2 α∞
FCα (h) − h2 ≤ N h − L C (h)2 .
1 − 3 2 α∞
5 Conclusion
We have explored some properties of the real-valued fractal operator and complex-
valued fractal operator defined on SG. The bound for the perturbation error on SG
has also been calculated. Under certain conditions, the fractal operator’s bounds are
established. Furthermore, we have established relations between the complex-valued
fractal operator and the real-valued fractal operator.
References
1. Agrawal, V., Som, T.: Fractal dimension of α-fractal function on the Sierpiński Gasket. Eur.
Phys. J. Spec. Top. (2021). https://doi.org/10.1140/epjs/s11734-021-00304-9
2. Agrawal, V., Som, T.: L p approximation using fractal functions on the Sierpiński gasket. Results
Math. 77(2), 1–17 (2021)
3. Agrawal, V., Som, T., Verma, S.: On bivariate fractal approximation. J. Anal. (2022). https://
doi.org/10.1007/s41478-022-00430-0
4. Bagby, T.: L p approximation by analytic functions. J. Approx. Theory 5, 401–404 (1972)
5. Barnsley, M.F.: Fractal functions and interpolation. Constr. Approx. 2, 301–329 (1986)
6. Barnsley, M.F.: Fractals Everywhere. Academic, Orlando, Florida (1988)
7. Celik, D., Kocak, S., Özdemir, Y.: Fractal interpolation on the Sierpiński Gasket. J. Math. Anal.
Appl. 337, 343–347 (2008)
8. Jha, S., Verma, S., Chand, A. K. B.: Non-stationary zipper α-fractal functions and associated
fractal operator. Fractional Calculus and Applied Analysis (2022). https://doi.org/10.1007/
s13540-022-00067-7
9. Jha, S., Verma, S.: Dimensional analysis of α-fractal functions. Results Math 76(4), 1–24
(2021)
10. Kigami, J.: Analysis on Fractals. Cambridge University Press, Cambridge, UK (2001)
11. Massopust, P.R.: Fractal Functions, Fractal Surfaces, and Wavelets, 2nd edn. Academic (2016)
12. Navascués, M.A.: Fractal approximation. Complex Anal. Oper. Theory 4, 953–974 (2010).
https://doi.org/10.1007/s11785-009-0033-1
13. Navascués, M.A., Verma, S., Viswanathan, P.: Concerning the Vector-Valued Fractal Interpo-
lation Functions on the Sierpiński Gasket. Mediterr. J. Math. 18(5), 1–26 (2021)
14. Prasad, S. A., Verma S.: Fractal Interpolation Function On Products of the Sierpiński Gaskets
(2022). arXiv:2206.01920v1
15. Ri, S.: Fractal Functions on the Sierpiński Gasket. Chaos, Solitons Fractals 138, 110142 (2020)
16. Ri, S.G., Ruan, H.J.: Some properties of fractal interpolation functions on Sierpiński gasket. J.
Math. Anal. Appl. 380, 313–322 (2011)
17. Ruan, H.J.: Fractal interpolation functions on post critically finite self-similar sets. Fractals 18,
119–125 (2010)
18. Sahu, A., Priyadarshi, A.: On the box-counting dimension of graphs of harmonic functions on
the Sierpiński gasket. J. Math. Anal. Appl. 487, 124036 (2020)
92 V. Agrawal and T. Som
19. Strichartz, R.S.: Differential Equations on Fractals. Princeton University Press, Princeton, NJ
(2006)
20. Verma, S., Viswanathan, P.: A fractal operator associated with bivariate fractal interpolation
functions on rectangular grids. Result Math. 75, 25 (2020)
21. Verma, S., Viswanathan, P.: Katugampola fractional integral and fractal dimension of bivariate
functions. Results Math. 76, 165 (2021). https://doi.org/10.1007/s00025-021-01475-6
22. Verma, S., Viswanathan, P.: A revisit to α-fractal function and box dimension of its graph.
Fractals 27(06), 1950090 (2020)
23. Verma, S., Massopust, P. R.: Dimension preserving approximation, To appear in Aequationes
Mathematicae, https://doi.org/10.48550/arXiv.2002.05061
24. Verma, S., Sahu, A.: Bounded variation on the Sierpiński Gasket. Fractals (2022). https://doi.
org/10.1142/S0218348X2250147X
25. Verma, S., Viswanathan, P.: Parameter identification for a class of bivariate fractal interpola-
tion functions and constrained approximation. Numer. Funct. Anal. Optim. 41(9), 1109–1148
(2020)
Dimensional Analysis of Mixed
Riemann–Liouville Fractional Integral
of Vector-Valued Functions
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 93
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_8
94 M. Pandey et al.
1 Introduction
Jordan [10] was the one who first introduced bounded variation for the function,
g : J → R, where J is a compact interval of R. The present paper is focused on those
functions which are bivariate vector-valued bounded variation. Unlike univariate
functions, there are several definitions which have been introduced in the literature on
a bivariate function known to be of bounded variation, for example, the definitions by
Vitali, Hardy, Arzelá, Pierpoint, Fréchet, Tonelli, and Hahn [1, 2, 6]. In this article, we
have given the results for a function, which satisfies the bounded variation definition
in Arzelá sense.
Integration and differentiation have always played a significant role in mathemat-
ics. Fractional Calculus (FC), dealing with the basic as well as advanced theories on
integration and differentiation of non-integer (fractional) order and their properties,
is one of the wide subjects existed in the literature from the centuries. Several papers
and books have been published so far, interested reader can follow: [4, 14–16, 19–22]
to know more about FC and its applications in different areas. To define Fractional
Integral (FI), many formulae have been given in the theory, for example, Hadamard
FI [26], Riemann–Liouville Fractional Integral (RLFI) [4, 23], Katugampola FI [5,
25], etc.
Fractal Dimension (FD) is one of the most entertained topics in fractal geometry.
It has always gained the attention of researchers with its extensive applications in
different areas. Although estimating the FD of the graph of elementary functions is
not simple. However, many theories have been developed for finding FD of the graph
of special classes of functions (see, for instance, [3, 7, 9, 13]).
1.1 Motivation
1.2 Delineation
The proposed work is assembled as follows. The next section is reserved for back-
ground and preliminaries required for the study. In Sect. 3, we have proved some
basic results for the bivariate vector-valued functions. We have concluded our paper
in Sect. 4, where we have determined that FD of the graph of bivariate vector-valued
bounded variation function defined on rectangular domain is 1 and so is the FD of
the graph of RLFI of such functions. Also, we have given an upper bound for the
graphs of coordinate functions of bivariate vector-valued Hölder continuous function
and ended the section with a question that “ whether we can get an upper bound for
bivariate vector-valued Hölder continuous function?”
The following are the notations which we shall be using in the paper:
• R: collection of all real numbers.
• N: collection of natural numbers.
• [c, d] × [ p, q]: Rectangular domain in R2 .
• σ-HC: Hölder continuous function having exponent σ.
In what follows, we have collected the preliminary materials required for our study.
To know more in detail, one can follow [7, 8].
i=1
and s-dimensional Hausdorff measure of B is given by H s (B) = lim Hδs (B). Then,
δ→0
Hausdorff dimension of B is defined as
96 M. Pandey et al.
dim H (B) = sup{s : H s (B) = ∞ = inf s : H s (B) = 0}.
log Nδ (B)
dim B (B) = lim inf
δ→0 − log δ
is known as upper box dimension of B. If both lower and upper box dimensions are
same, then that quantity is known as box dimension of B and defined as
log Nδ (B)
dim B (B) = lim .
δ→0 − log δ
Note
The set of all continuous maps defined on [c, d] × [ p, q] is denoted by
C [c, d] × [ p, q], Rn .
g(u, w) − g(v, y)2 ≤ K(u, w) − (v, y)σ2 for all (u, w), (v, y) ∈ [c, d] × [ p, q], (1)
c = u 0 ≤ u 1 ≤ u 2 ≤ · · · ≤ u m = d;
p = w0 ≤ w1 ≤ w2 ≤ · · · ≤ wm = q,
the sum
m−1
|Δf (u i , wi )|, where Δf (u i , wi ) = f (u i+1 , wi+1 ) − f (u i , wi )
i=0
is bounded.
Theorem 1 ([1]) The below statements are equivalent:
(i) g : [c, d] × [ p, q] → R is a real-valued bivariate function of bounded varia-
tion satisfying the Arzelá condition.
(ii) There exist two bounded variation functions, φ1 : [c, d] × [ p, q] → R and φ2 :
[c, d] × [ p, q] → R such that g can be written as difference of φ1 and φ2 , where
φ1 and φ2 satisfy the relation
u w
(ρ,μ) 1
(c, p) J g (u, w) = (u − t)ρ−1 (w − s)μ−1 g(t, s) dt ds,
Γ (ρ)Γ (μ) c p
where ρ, μ > 0 for all (u, w) ∈ [c, d] × [ p, q], and 0 ≤ c < d < ∞, 0 ≤ p < q <
∞.
Definition 9 Consider the vector-valued integrable function g defined on [c, d] ×
[ p, q], then RLFI of g is determined as
(ρ,μ) g(u, w) = (ρ,μ) g (u, w), (ρ,μ) g (u, w), . . . , (ρ,μ) g (u, w) ,
(c, p) J (c, p) J 1 (c, p) J 2 (c, p) J n
where
u w
(ρ,μ) 1
(c, p) J gi (u, w) = (u − t)ρ−1 (u − s)μ−1 gi (t, s) dt ds,
Γ (ρ)Γ (μ) c p
Remark 3 Note that in [11] the Peano space filling curve g : [c, d] → [0, 1] ×
[0, 1] is 21 -HC and coordinate functions g1 and g2 satisfy dim H (Gr (g1 )) =
dim H (Gr (g2 )) = 1.5.
Now define h : [c, d] × [ p, q] → Rn such that
Proof Since g is σ-HC, using Proposition 1 we have gi is also σ-HC for each i ∈ 1, n.
Then, using Lemma 3 we get
dim H Gr (gi ) ≤ 3 − σ.
We present our findings in this section. We back up our claims with straightforward
arguments.
Throughout this section, we consider 0 ≤ c < d < ∞ and 0 ≤ p < q < ∞.
(ρ,μ)
Theorem 4 Let g : [c, d] × [ p, q] → Rn be a bounded function, then (c, p) J g
is also bounded.
√
(ρ,μ) g(x , x )
(c, p) J 1 2 ≤ n max (c, p) J(ρ,μ) gi (x1 , x2 )
2 1≤i≤n
√ 1 x1 x2
= n max (x1 − s)ρ−1 (x2 − t)μ−1 gi (s, t)ds dt
1≤i≤n Γ (ρ)Γ (μ) c p
√ x1 x2
1
≤ n max (x1 − s)ρ−1 (x2 − t)μ−1 gi (s, t) ds dt
1≤i≤n Γ (ρ)Γ (μ) c p
√ x1 x2
K
≤ n (x1 − s)ρ−1 (x2 − t)μ−1 ds dt
Γ (ρ)Γ (μ) c p
√ K (x1 − a)ρ (x2 − c)μ
≤ n .
Γ (ρ)Γ (μ) ρμ
That is,
(ρ,μ) √ K
(c, p) J g(x1 , x2 ) ≤ n (d − c)ρ (q − p)μ for all (x1 , x2 ) ∈ [c, d] × [ p, q].
2 Γ (ρ + 1)Γ (μ + 1)
(ρ,μ)
(c, p) J + h, x2 + k) −(c, p) J(ρ,μ) g(x1 , x2 ) 2
g(x1
√
≤ n max1≤i≤n (c, p) J(ρ,μ) gi (x1 + h, x2 + k) −(c, p) J(ρ,μ) gi (x1 , x2 )
√ x1 +h x2 +k
1
= n max1≤i≤n Γ (ρ)Γ (μ) c p (x1 + h − s)ρ−1 (x2 + k − t)μ−1 gi (s, t) ds dt
x1 x2
ρ−1 μ−1
− Γ (ρ)Γ 1
(μ) c p (x 1 − s) (x 2 − t) gi (s, t) ds dt
√ c+h p+k
1
= n max1≤i≤n Γ (ρ)Γ (μ) c p (x1 + h − s)ρ−1 (x2 + k − t)μ−1 gi (s, t) ds dt
x1 +h p+k
+ Γ (ρ)Γ1
(μ) c+h p (x1 + h − s)ρ−1 (x2 + k − t)μ−1 gi (s, t) ds dt
c+h x2 +k ρ−1
+ Γ (ρ)Γ1
(μ) c p+k (x 1 + h − s) (x2 + k − t)μ−1 gi (s, t) ds dt
x1 +h x2 +k ρ−1
+ Γ (ρ)Γ
1
(μ) c+h p+k (x 1 + h − s) (x2 + k − t)μ−1 gi (s, t) ds dt
x1 x2
ρ−1 μ−1
− Γ (ρ)Γ 1
(μ) c p (x 1 − s) (x 2 − t) gi (s, t) ds dt
√
= n max1≤i≤n |Ii1 + Ii2 + Ii3 + Ii4 − Ii5 |,
where
Dimensional Analysis of Mixed Riemann–Liouville Fractional … 103
c+h p+k
Ii1 = 1
Γ (ρ)Γ (μ) c p (x1 + h − s)ρ−1 (x2 + k − t)μ−1 gi (s, t) ds dt,
x1 +h p+k
Ii2 = 1
Γ (ρ)Γ (μ) c+h p (x1 + h − s)ρ−1 (x2 + k − t)μ−1 gi (s, t) ds dt,
c+h x2 +k ρ−1
Ii3 = 1
Γ (ρ)Γ (μ) c p+k (x 1 + h − s) (x2 + k − t)μ−1 gi (s, t) ds dt,
x1 +h x2 +k ρ−1
Ii4 = 1
Γ (ρ)Γ (μ) c+h p+k (x 1 + h − s) (x2 + k − t)μ−1 gi (s, t) ds dt,
x1 x2
Ii5 = 1
Γ (ρ)Γ (μ) c p (x1 − s)ρ−1 (x2 − t)μ−1 gi (s, t) ds dt.
where
1 x1 x2
Ii 6 = (x1 − s)ρ−1 (x2 − t)μ−1 gi (s + h, t + k) − gi (s, t) ds dt.
Γ (ρ)Γ (μ) c p
Thus,
(c, p) J(ρ,μ) gi (x1 + h, x2 + k) −(c, p) J(ρ,μ) gi (x1 , x2 ) ≤ M2 k + M2 k + M3 h + .
4
Hence, proof follows.
The next result can be observed in [24] for bivariate real-valued bounded variation
maps. But for completeness, we demonstrate the result here.
Lemma 4 For a bounded variation function, g : [c, d] × [ p, q] → Rn the following
will hold:
(i) If gi (c, p) ≥ 0, then there will exist monotone functions h i and f i such that
gi = f i − h i with f i (c, p) ≥ 0 and h i (c, p) = 0.
(ii) If gi (c, p) < 0, then there will exist monotone functions h i and f i such that
gi = f i − h i with f i (c, p) = 0 and h i (c, p) > 0.
104 M. Pandey et al.
Next theorem is a well-known result for univariate real-valued function. One can refer
[7] for detail study. In the interest of completeness, we include the demonstration of
the theorem.
Theorem 6 If g is of bounded variation on [c, d] × [ p, q], then dim H (Gr (g)) = 2.
Proof To prove the theorem it will be sufficient to prove (c, p) J(ρ,μ) g ∈ BV([c, d] ×
[ p, q], Rn ). Since g = (g1 , . . . , gn ) is of bounded variation, gi is also be of bounded
variation for each i ∈ 1, n. Without loss of generality if gi (c, p) ≥ 0, then using
Lemma 4, there exist two monotone functions ηi : [c, d] × [ p, q] → R and ξi :
[c, d] × [ p, q] → R such that gi = ηi − ξi with ηi (c, p) ≥ 0 and ξi (c, p) = 0.
Define functions Φ, θ : [c, d] × [ p, q] → Rn such that
Φ(x, y) =(c, p) J(ρ,μ) η(x, y) and θ(x, y) =(c, p) J(ρ,μ) ξ(x, y),
(ρ,μ)
(c, p) J g(x, y) = Φ(x, y) − θ(x, y).
1 x2 y
Φi (x2 , y) − Φi (x1 , y) = (x2 − s)ρ−1 (y − t)μ−1 ηi (s, t) ds dt
Γ (ρ)Γ (μ) c p
1 x1 y
− (x2 − s)ρ−1 (y − t)μ−1 ηi (s, t) ds dt
Γ (ρ)Γ (μ) c p
1 c+x2 −x1 y
= (x2 − s)ρ−1 (y − t)μ−1 ηi (s, t) ds dt
Γ (ρ)Γ (μ) c p
1 x2 y
+ (x2 − s)ρ−1 (y − t)μ−1 ηi (s, t) ds dt
Γ (ρ)Γ (μ) c+x2 −x1 p
1 x1 y
− (x2 − s)ρ−1 (y − t)μ−1 ηi (s, t) ds dt.
Γ (ρ)Γ (μ) c p
1 c+x2 −x1 y
Φi (x2 , y) − Φi (x1 , y) = (x2 − s)ρ−1 (y − t)μ−1 ηi (s, t) ds dt
Γ (ρ)Γ (μ) c p
1 x1 y
+ (x2 − s)ρ−1 (y − t)μ−1 [ηi (s + x2 − x1 , t) − ηi (s, t)] ds dt.
Γ (ρ)Γ (μ) c p
(ρ,μ)
Then, g is a bounded variation function (see [6]). Note that (0,0) J gi (u, w) =0
for each i ∈ 1, n.
Hence, RLFI of g is
(ρ,μ) g(u, w) = (ρ,μ) g (u, w), (ρ,μ) g (u, w), . . . , (ρ,μ) g (u, w)
(0,0) J (0,0) J 1 (0,0) J 2 (0,0) J n
=(0, 0, . . . , 0).
In the following theorem, an upper bound for the upper box dimension of RLFI
of the coordinate functions of a vector-valued Hölder continuous function has been
provided.
(ρ,μ)
Theorem 8 Let g ∈ C([c, d] × [ p, q], Rn ), then (c, p) J g will be σ-
HC, where σ = min{ρ, μ} provided that 0 < ρ, μ < 1. Moreover, we get
dim B (Gr ( (c, p) J(ρ,μ) gi )) ≤ 3 − σ for every i ∈ 1, n.
where
u y
Ii7 = ((u + h) − s)ρ−1 (y + k − t)μ−1 − (u − s)ρ−1 (y − t)μ−1 gi (s, t) ds dt
c p
u+h y+k ρ−1
and Ii8 = (u + h) − s (y + k) − s gi (s, t) ds dt.
u y
The fact g ∈ C([c, d] × [ p, q], Rn ) yields that gi ∈ C([c, d] × [ p, q], R). There-
fore, there exists a γ > 0 such that max sup |gi (t)| ≤ γ. Further, Ii7 can be
1≤i≤n t∈[c,d]×[ p,q]
approximated as follows:
Dimensional Analysis of Mixed Riemann–Liouville Fractional … 107
u y
|Ii7 | ≤ γ (u − s)ρ−1 (y − t)μ−1 − ((u + h) − s)ρ−1 (y + k − t)μ−1 .
c p
That is, (c, p) J(ρ,μ) g is σ-HC. Therefore, using Lemma 1 and Proposition 1, we get
dim B (Gr ( (c, p) J(ρ,μ) gi )) ≤ 3 − σ, completing the proof.
Remark 4 Comparing Theorem 8 with Theorem 5, where it is demonstrated that for
0 < μ, ρ, the RL integral (c, p) J(ρ,μ) g is continuous whenever g is continuous, we
notice that (c, p) J(ρ,μ) g is also Hölder continuous on [c, d] × [ p, q] whenever g is a
continuous function on [c, d] × [ p, q] with 0 < c < d < ∞, 0 < p < q < ∞, and
ρ, μ > 0.
Further the semigroup property of the RL integral for a “sufficiently good” function
is there as a preamble to the next theorem.
(ρ1 ,μ1 ) (ρ2 ,μ2 )
(c, p) J (c, p) J g =(c, p) J(ρ1 ,μ1 )+(ρ2 ,μ2 ) g.
Now, the next theorem is the direct deduction of the previous theorem:
Theorem 9 Let g be continuous on [c, d] × [ p, q], 0 < c < d < ∞, 0 < p < q <
∞ and 0 < ρ, μ ≤ 1.
1. If 0 < σ < 1, then
2 ≤ dim H Gr (c, p) J(ρ,μ) gi ≤ dim B Gr (c, p) J(ρ,μ) gi ≤ 3 − σ.
2. If σ ≥ 1, then
dim H Gr (c, p) J(ρ,μ) gi = dim B Gr (c, p) J(ρ,μ) gi = 2.
Remark 5 In Remark 3, we notice that h is also a 21 -HC but dim H (Gr (h)) ≥ 3 −
1
2
= 2.5. It shows that even if a vector-valued function is σ-HC, we cannot provide
its upper bound like we do for a real-valued function.
108 M. Pandey et al.
We have given the concept of mixed RLFI of vector-valued functions (Definition 9),
thereafter we have given dimensional bounds for a bivariate vector-valued func-
tions (Lemma 2). In Sect. 4, we have explored the properties of the mixed RLFI
of the vector-valued functions (Theorems 4, 5, Lemma 4) and proved some dimen-
sional results for bivariate vector-valued functions (Theorem 6) and for mixed RLFI
(Theorems 7, 8, and 9)
In view of Remark 5, a question arises: can we find an upper bound for the dimen-
sion of the graph of (c, p) J(ρ,μ) g?
One may ask a more general question on the fractal dimension of RLFI of a
set-valued function. Recently, Pandey et al. [18] tried to initiate a study on fractal
dimension of set-valued functions. From [18], one can notice that to establish results
in terms of set-valued mappings, a different set of tools is required.
References
1. Adams, C.R., Clarkson, J.A.: Properties of functions f (x, y) of bounded variation. Trans. Amer.
Math. Soc. 36(4), 711–730 (1934)
2. Adams, C.R., Clarkson, J.A.: A correction to “properties of functions f (x, y) of bounded
variation.” Trans. Amer. Math. Soc. 46, 468 (1939)
3. Barnsley, M.: Fractals Everywhere, 2nd edn (1993)
4. Chandra, S., Abbas, S.: The calculus of bivariate fractal interpolation surfaces. Fractals 29(03),
2150066 (2021)
5. Chandra, S., Abbas, S.: Box dimension of mixed katugampola fractional integral of two-
dimensional continuous functions. Fract. Calcul. Appl. Anal. 1–15 (2022)
6. Clarkson, J.A., Adams, C.R.: On definitions of bounded variation for functions of two variables.
Trans. Amer. Math. Soc. 35(4), 824–854 (1933)
7. Falconer, K.: Fractal Geometry: Mathematical Foundations and Applications. Wiley, New York
(2004)
8. Gordon, R.A.: Real Analysis: A First Course. Pearson College Division (2002)
9. Jha, S., Verma, S.: Dimensional analysis of α-fractal functions. Results Math. 76(4), 1–24
(2021)
10. Jordan, C.: Sur la series de fourier. CR Acad. Sci. Paris 92, 228–230 (1881)
11. Kôno, N.: On self-affine functions. Jpn. J. Appl. Math. 3(2), 259–269 (1986)
12. Liang, Y.: Box dimensions of Riemann-Liouville fractional integrals of continuous functions
of bounded variation. Nonlinear Anal.: Theory, Methods & Appl. 72(11), 4304–4306 (2010)
13. Massopust, P.R.: Fractal Functions, Fractal Surfaces, and Wavelets. Academic, Cambridge
(2016)
14. Moshrefi-Torbati, M., Hammond, J.: Physical and geometrical interpretation of fractional oper-
ators. J. Frank. Inst. 335(6), 1077–1086 (1998)
15. Nigmatullin, R.: Fractional integral and its physical interpretation. Theor. Math. Phys. 90(3),
242–251 (1992)
16. Oldham, K., Spanier, J.: The Fractional Calculus Theory and Applications of Differentiation
and Integration to Arbitrary Order. Elsevier, Amsterdam (1974)
17. Pandey, M., Som, T., Verma, S.: Fractal dimension of Katugampola fractional integral of vector-
valued functions. Eur. Phys. J. Spec. Topics 1–8 (2021)
18. Pandey, M., Som, T., Verma, S.: Set-valued α-fractal functions (2022). arXiv:2207.02635
Dimensional Analysis of Mixed Riemann–Liouville Fractional … 109
19. Podlubny, I.: Fractional Differential Equations: An Introduction to Fractional Derivatives, Frac-
tional Differential Equations, to Methods of Their Solution and Some of Their Applications.
Elsevier, Amsterdam (1998)
20. Podlubny, I.: Geometric and Physical interpretation of fractional integration and fractional
differentiation (2001). arXiv:math/0110241
21. Ross, B.: Fractional Calculus and Its Applications: Proceedings of the International Conference
Held at the University of New Haven, June 1974, vol. 457. Springer, Berlin (2006)
22. Samko, S.G.: Fractional integrals and derivatives, theory and applications. Minsk, Nauka I
Tekhnika (1987)
23. Verma, S., Liang, Y.: Effect of the Riemann-Liouville fractional integral on unbounded variation
points. arXiv:2008.11113
24. Verma, S., Viswanathan, P.: Bivariate functions of bounded variation: fractal dimension and
fractional integral. Indag. Math. 31(2), 294–309 (2020)
25. Verma, S., Viswanathan, P.: A note on Katugampola fractional calculus and fractal dimensions.
Appl. Math. Comput. 339, 220–230 (2018)
26. Wu, X.E., Du, J.H.: Box dimension of Hadamard fractional integral of continuous functions
of bounded and unbounded variation. Fractals 25(03), 1750035 (2017)
Fractional Operator Associated with the
Fractal Integral of A-Fractal Function
Fractal calculus has been introduced as a new framework to study the fractal curves
and fractal like sets, which is different from fractional and classical calculus methods
(refer [1, 2]). The derivatives and integrals involved in fractal calculus are, respec-
tively, called as fractal derivatives and fractal integrals. As fractal functions are non-
differentiable, its fractional calculus have been investigated by many researchers
in [3–12]. Besides, recently fractal calculus of fractal functions has been discussed
in [13–15]. For more interesting results on types of fractal interpolation functions
(FIFs) and its developments, refer [16–21]. The fractal integral of hidden variable
FIF and α-fractal function has been investigated in [22]. So far, the fractal calculus
of A-fractal function, which is the blend of α-fractal function and hidden variable
FIF [23, 24] and [25, 26], has not been discussed. This literature gap instigated us to
work on the fractal integral of A-fractal function. In addition, a fractional operator is
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 111
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_9
112 T. M. C. Priyanka and A. Gowrisankar
proposed, which gives the fractal integral of A-fractal function for each vector-valued
continuous function.
In this section, the general material required for defining the fractal integral is
summarized. Section 2 discusses the construction of fractal interpolation function
and A-fractal function. The fractal integral of A-fractal function is explored with the
predefined initial conditions in Sect. 3. Fractional operator associated with the fractal
integral of A-fractal function is presented in Sect. 4.
This paper begins with the definition of mass function. A-fractal curve can be
described as the image of R2 -valued continuous functions f defined on R, which are
fractals. Let F ⊂ Rn be a fractal curve with sub-division D[a,b] . The fractal curve F
is parameterizable if there exists a bijective continuous function w : R → F.
n−1
|w(yk ) − w(yk+1 )|α
γ α (F, a, b) = lim inf ,
→0 {P[a,b] :|P|≤}
k=1
Γ (α + 1)
where Γ (x) is the gamma function, |.| is the Euclidean metric on Rn , 1 < α ≤ 2 and
|P| denotes the maximum of (yk+1 − yk ) for k = 0, 1, . . . , n − 1.
Definition 2 For the fractal curve F and for some arbitrary fixed point r0 ∈ [a, b],
the staircase function is defined by
γ α (F, r0 , y), y ≥ r0 ,
S Fα (t) =
−γ α (F, y, r0 ), otherwise.
Definition 3 For the function f , the upper and the lower f α -sum over the sub-
division D are provided by
n−1
U α [ f, F, D] = M[ f, C(yk , yk+1 )][S Fα (yk+1 ) − S Fα (yk )] (1)
k=0
n−1
L α [ f, F, D] = m[ f, C(yk , yk+1 )][S Fα (yk+1 ) − S Fα (yk )]. (2)
k=0
Define
f (x)d Fα x = sup L α [ f, F, D]
C(a,b) P[a,b]
Fractional Operator Associated with the Fractal Integral … 113
and
f (x)d Fα x = inf U α [ f, F, D].
C(a,b) P[a,b]
Definition 4 For the continuous bounded function f and for x ∈ F, the F α -integral
is given by
f (x)d Fα x = f (x)d Fα x = f (x)d Fα x.
C(a,b) C(a,b) C(a,b)
The following section describes the generation of fractal interpolation function and
A-fractal function via the iterated function system defined on I × R and I × R2 ,
respectively.
xk − xk−1 x N xk−1 − x0 xk
ak = , bk = ,
x N − x0 x N − x0
|L k (t) − L k (t )| ≤ m k |t − t |,
where rk is the contraction factor such that rk ∈ (0, 1), i ∈ {1, 2, . . . , N }. In addition,
Fk satisfy the join-up conditions
114 T. M. C. Priyanka and A. Gowrisankar
Now, using the two contraction maps L k and Fk , a new contraction map wk : X →
Ik × R, k = 1, 2, . . . , N is defined by
The complete metric space X with the finite collection of contraction maps wk
constitutes an Iterated Function System (IFS) denoted by
{X ; wk : k = 1, 2, . . . , N }. (6)
N
W (B ∗ ) = wk (B ∗ )
k=1
for any B ∗ ∈ K(X ). As W is contraction on the complete metric space K(X ) with
respect to Hausdorff metric, by the Banach contraction theorem, there is a unique
invariant set G ∗ for the map W satisfying G ∗ = W (G ∗ ). Therefore, the set G ∗ is
known as the attractor or deterministic fractal for the IFS (6).
Let C(I ) be the set of continuous functions g : I → R satisfying g(x0 ) =
y0 , g(x N ) = y N and ρ be the uniform metric defined on C(I ) as follows:
Then, (C(I ), ρ) becomes a complete metric space. To obtain the fractal interpolation
function, the Read–Bajraktarević (RB) operator T is defined on C(I ) by
T(h(t)) = Fk L −1 −1
k (t), h ◦ L k (t) , t ∈ Ik , k = 1, 2, . . . , N . (7)
By the Banach contraction principle, the contraction map T has a unique fixed point
f ∈ C(I ) satisfying the functional equation,
f (t) = T( f (t)) = Fk L −1 −1
k (t), f ◦ L k (t) , k = 1, 2, . . . , N , (8)
for all t ∈ I . This function is referred to as the Fractal Interpolation Function (FIF)
corresponding to the mappings wk for all k = 1, 2, . . . , N . The reader is encouraged
to refer [27–33].
Fractional Operator Associated with the Fractal Integral … 115
{I × R2 ; wk : k = 1, 2, . . . , N }, (9)
where b = (b1 , b2 ) ∈ C(I, R2 ) obeys b(x0 ) = f(x0 ) and b(x N ) = f(x N ). The fixed
point of the IFS (9) is the graph of the continuous vector-valued function f[A] =
( f 1 [A], f 2 [A]) which obeys the following fixed point equation:
This function f[A] is known as the hidden variable A-fractal function (or simply
A-fractal function) approximating the given continuous function f with respect to
x0 < x1 < · · · < x N and the base function b. Then, the two components of f[A],
namely, f 1 [A] and f 2 [A] satisfy
f 1 [A](x) = αk f 1 [A]L −1 −1 −1
k (x) + βk f 2 [A]L k (x) + pk (L k (x)),
(12)
f 2 [A](x) = γk f 2 [A]L −1 −1
k (x) + qk (L k (x)).
The fractal integral of A-fractal function defined at the initial point is investigated in
this section. The following is the definition of fractal integral of A-fractal function.
Definition 5 Let f[A] be the A-fractal function corresponding to the IFS (6). For
given ŷ0 and ẑ 0 , the fractal integral of order μ for f[A] is defined componentwise as
follows:
x
ˆ μ μ
f 1 [A](x) = ŷ0 + S F ( f 1 [A](t))d F t,
x
0x (13)
ˆ μ μ
f 2 [A](x) = ẑ 0 + S F ( f 2 [A](t))d F t.
x0
The following theorem examines the resultant function of the fractal integral of
A-fractal function.
Theorem 1 Suppose f[A] is the (hidden variable) A-fractal function generated by
the IFS {L k (x), Fk (x, y, z) : k = 1, 2, . . . , N }. The fractal integral of f[A], defined
in (13), is denoted as f̂[A]. Then, for given ŷ0 = 0 and ẑ 0 = 0, f̂[A] = ( fˆ1 [A], fˆ2 [A])
is again an A-fractal function determined by the IFS {(L k (x), F̂k (x, ŷ, ẑ)) : k ∈
1, 2, . . . , N }, where F̂1k (x, ŷ, ẑ) = ak αk ŷ + ak βk ẑ + p̂k (x), F̂2k (x, ẑ) = ak γk ẑ +
N N
q̂k (x) , with k=1 ak αk = 1, k=1 ak γk = 1,
x x x
μ μ μ μ μ μ
p̂k (x) = ŷk−1 + ak S F ( f 1 ◦ L k (t))d F t − ak αk S F (b1 (t))d F t − ak βk S F (b2 (t))d F t,
x0 x0 x0
k x x
N μ μ N μ μ
ŷk = an αn ŷ N + βn ẑ N + S F ( f 1 ◦ L n (t))d F t − αn S F (b1 (t))d F t
n=1 x0 x0
x
N μ μ
− βn S F (b2 (t))d F t ,
x0
N xN μ μ xN μ μ xN μ μ
n=1 an βn ẑ N + x0 S F ( f 1 ◦ L n (t))d F t − αn x0 S F (b1 (t))d F t − βn x0 S F (b2 (t))d F t
ŷ N = N
,
1− n=1 an αn
x x
μ μ μ μ
q̂k (x) = ẑ k−1 + ak S F ( f 2 ◦ L k (t))d F t − ak γk S F (b2 (t))d F t,
x0 x0
k xN xN
μ μ μ μ
ẑ k = an γn ẑ N + S F ( f 2 ◦ L n (t))d F t − γn S F (b2 (t))d F t ,
n=1 x0 x0
N xN μ μ xN μ μ
n=1 an x0 S F ( f 2 ◦ L n (t))d F t − γn x0 S F (b2 (t))d F t
ẑ N = N
,
1− n=1 an γn
for k = 1, 2, . . . , N .
Fractional Operator Associated with the Fractal Integral … 117
Proof Using the definition of fractal integral for the function fˆ1 [A] provides
L k (x)
μ μ
fˆ1 [A](L k (x)) = ŷ0 + S F ( f 1 [A](t))d F t
x0
xk−1 L k (x)
μ μ μ μ
= ŷ0 + S F ( f 1 [A](t))d F t + S F ( f 1 [A](t))d F t
x0 xk−1
x
μ μ
= ŷk−1 + ak S F ( f 1 [A]L k (t))d F t.
x0
x μ μ x μ μ
Denote p̂k (x) = ŷk−1 + ak x0 S F ( f 1 ◦ L k (t))d F t − ak αk x0 S F (b1 (t))d F t −
x μ μ
ak βk x0 S F (b2 (t))d F t. Now, applying the definition of fractal integral for the
function fˆ2 [A] provides
L k (x)
μ μ
fˆ2 [A](L k (x)) = ẑ 0 + S F ( f 2 [A](t))d F t
x0
xk−1 L k (x)
μ μ μ μ
= ẑ 0 + S F ( f 2 [A](t))d F t + S F ( f 2 [A](t))d F t
x0 xk−1
x
μ μ
= ẑ k−1 + ak S F ( f 2 [A]L k (t))d F t.
x0
x μ μ x μ μ
Denote q̂k (x) = ẑ k−1 + ak x0 S F ( f 2 ◦ L k (t))d F t − ak γk x0 S F (b2 (t))d F t. In order
to find the new data points ŷk and ẑ k , take x = x N and L k (x N ) = xk ,
xN
μ μ
ŷk = ŷk−1 + ak αk fˆ1 [A](x N ) + ak βk fˆ2 [A](x N ) + ak S F ( f 1 ◦ L k (t))d F t
x0
xN xN
μ μ μ μ
− ak αk S F (b1 (t))d F t − ak βk S F (b2 (t))d F t
x0 x0
xN
μ μ
ŷk − ŷk−1 = ak αk ŷ N + ak βk ẑ N + ak S F ( f 1 ◦ L k (t))d F t
x0
xN xN
μ μ μ μ
− ak αk S F (b1 (t))d F t − ak βk S F (b2 (t))d F t
x0 x0
xN xN
μ μ μ μ
ẑ k = ẑ k−1 + ak γk fˆ2 [A](x N ) + ak S F ( f 2 ◦ L k (t))d F t − ak γk S F (b2 (t))d F t
x0 x0
xN xN
μ μ μ μ
ẑ k − ẑ k−1 = ak γk ẑ N + ak S F ( f 2 ◦ L k (t))d F t − ak γk S F (b2 (t))d F t.
x0 x0
k
Using the system of equations, ŷk = ŷ0 + n=1 (yn+1 − yn ), the new data points can
be obtained as follows:
k xN xN xN
μ μ μ μ μ μ
ŷk = an αn ŷ N + βn ẑ N + S F ( f 1 ◦ L n (t))d F t − αn S F (b1 (t))d F t − βn S F (b2 (t))d F t
n=1 x0 x0 x0
k xN xN
μ μ μ μ
ẑ k = an γn ẑ N + S F ( f 2 ◦ L n (t))d F t − γn S F (b2 (t))d F t .
n=1 x0 x0
Since the continuous functions p̂k (x) and q̂k (x) obey the below endpoint conditions:
and
q̂k (x0 ) = ẑ k−1 , q̂k (x N ) = ẑ k − ak γk ẑ N
which implies F̂k (x0 , ŷ0 , ẑ 0 ) = ( ŷk−1 , ẑ k−1 ) and F̂k (x N , ŷ N , ẑ N ) = ( ŷk , ẑ k ). There-
fore, the function f̂[A] = ( fˆ1 [A], fˆ2 [A]) is again an A-fractal function corresponding
to the IFS {(L k (x), F̂k (x, ŷ, ẑ)) : k ∈ 1, 2, . . . , N }.
4 Fractional Operator
This section proposes a fractional operator associated with the fractal integral of
A-fractal function and verifies its linearity.
Let f ∈ C(I, R2 ) consider the base function b = f ◦ c, where c = (c1 , c2 ) ∈
C(I, R2 ) is not the identity function such that c(x0 ) = x0 and c(x N ) = x N . The frac-
tional operator of order 0 < m < 1 for the vector-valued function,
is defined by
x
μ μ
F m [A](f(x)) = ŷk−1 + ak αk fˆ1 [A](x) + ak βk fˆ2 [A](x) + ak S F ( f 1 ◦ L k (t))d F t
x0
x x
μ μ μ μ
− a k αk S F (b1 (t))d F t − ak βk S F (b2 (t))d F t
x0 x0
x x
μ μ μ μ
+ ẑ k−1 + ak γk fˆ2 [A](x) + ak S F ( f 2 ◦ L k (t))d F t − ak γk S F (b2 (t))d F t,
x0 x0
(14)
References
1. Parvate, A., Satin, S., Gangal, A.D.: Calculus on fractal curves in Rn . Fractals 19(01), 15–27
(2011). https://doi.org/10.1142/S0218348X1100518X
2. Parvate, A., Gangal, A.D.: Calculus on fractal subsets of real line - I: formulation. Fractals
17(01), 53–81 (2009). https://doi.org/10.1142/S0218348X09004181
3. Tatom, F.B.: The relationship between fractional calculus and fractal. Fractals 3(1), 217–229
(1995). https://doi.org/10.1142/S0218348X95000175
4. Yao, K., Su, W.Y., Zhou, S.P.: On the connection between the order of fractional calculus and
the dimensions of a fractal function. Chaos Solitons Fractals 23(2), 621–629 (2005). https://
www.sciencedirect.com/science/article/abs/pii/S0960077904002966
5. Liang, Y.-S., Zhang, Q.: A type of fractal interpolation functions and their fractional calculus.
Fractals 24(2), 1650026 (2016). https://doi.org/10.1142/S0218348X16500262
6. Gowrisankar, A., Uthayakumar, R.: Fractional calculus on fractal interpolation function for a
sequence of data with countable iterated function system. Mediterr. J. Math. 13(6), 3887–3906
(2016). https://doi.org/10.1007/s00009-016-0720-x
7. Gowrisankar, A., Prasad, M.G.P.: Riemann-Liouville calculus on quadratic fractal interpolation
function with variable scaling factors. J. Anal. 27(2), 347–363 (2019). https://doi.org/10.1007/
s41478-018-0133-2
8. Priyanka, T.M.C., Gowrisankar, A.: Analysis on Wely-Marchaud fractional derivative of types
of fractal interpolation function with fractal dimension. Fractals 29(07), 2150215 (2021).
https://doi.org/10.1142/S0218348X21502157
9. Chandra, S., Abbas, S.: Analysis of fractal dimension of mixed Riemann-Liouville fractional
integral (2021). arXiv:2105.06648
10. Chandra, S., Abbas, S.: Analysis of mixed Weyl-Marchaud fractional derivative
and box dimensions. Fractals 29(6), 2150145–2150386 (2021). https://doi.org/10.1142/
S0218348X21501450
11. Pandey, M., Som, T., Verma, S.: Fractal dimension of Katugampola fractional integral of vector-
valued functions. Eur. Phys. J. Spec. Top. 230, 3807–3814 (2021). https://doi.org/10.1140/epjs/
s11734-021-00327-2
12. Priyanka, T.M.C., Gowrisankar, A.: Riemann-Liouville fractional integral of non-affine fractal
interpolation function and its fractional operator. Eur. Phys. J. Spec. Top. 230, 3789–3805
(2021). https://doi.org/10.1140/epjs/s11734-021-00315-6
13. Golmankhaneh, A.K., Fernandez, A.: Fractal calculus of functions on Cantor Tartan spaces.
Fractal Fract. 2(4), 30 (2018). https://www.mdpi.com/2504-3110/2/4/30
Fractional Operator Associated with the Fractal Integral … 121
14. Alireza Khalili Golmankhaneh: A review on application of the local fractal calculus. Num.
Com. Meth. Sci. Eng. 1(2), 57–66 (2019)
15. Gowrisankar, A., Golmankhaneh, A.K., Serpa, C.: Fractal calculus on fractal interpolation
functions. Fractal Fract. 5(4), 157 (2021). https://www.mdpi.com/2504-3110/5/4/157
16. Easwaramoorthy, D., Uthayakumar, R.: Analysis on fractals in fuzzy metric spaces. Fractals
19(03), 379–386 (2011). https://doi.org/10.1142/S0218348X11005543
17. Agathiyan, A., Gowrisankar, A., Priyanka, T.M.C.: Construction of new fractal interpolation
functions through integration method. Results Math. 77, 122 (2022). https://doi.org/10.1007/
s00025-022-01666-9
18. Priyanka, T.M.C., Agathiyan, A., Gowrisankar, A.: Weyl-Marchaud fractional derivative of a
vector valued fractal interpolation function with function contractivity factors. J. Anal. (2022).
https://doi.org/10.1007/s41478-022-00474-2
19. Agrawal, V., Som, T.: Fractal dimension of α -fractal function on the Sierpiński gasket. Eur.
Phys. J. Spec. Top. 230(21), 3781–3787 (2021). https://doi.org/10.1140/epjs/s11734-021-
00304-9
20. Chandra, S., Abbas, S.: The calculus of bivariate fractal interpolation surfaces. Fractals 29(03),
2150066 (2021). https://doi.org/10.1142/S0218348X21500663
21. Agrawal, V., Som, T.: L p approximation using fractal functions on the Sierpinski gasket. Results
Math. 77, 74 (2022). https://doi.org/10.1007/s00025-021-01565-5
22. Banerjee, S., Gowrisankar, A.: Frontiers of Fractal Analysis: Recent Advances and Challenges.
CRC Press, Boca Raton (2022)
23. Chand, A.K.B., Kapoor, G.P.: Spline coalescence hidden variable fractal interpolation function.
J. Appl. Math. 1–17 (2006). https://doi.org/10.1155/JAM/2006/36829
24. Banerjee, S., Easwaramoorthy, D., Gowrisankar, A.: Fractal Functions, Dimensions and Signal
Analysis. Springer, Cham (2021)
25. Navascués, M.A.: Fractal polynomial interpolation. Z. Anal. Anwend. 25(2), 401–418 (2005).
https://doi.org/10.4171/ZAA/1248
26. Navascués, M.A.: Non-smooth polynomial. Int. J. Math. Anal. 1(4), 159–174 (2007)
27. Barnsley, M.F.: Fractals Everywhere. Academic, USA (1993)
28. Banerjee, S., Hassan, M.K., Mukherjee, S., Gowrisankar, A.: Fractal Patterns in Nonlinear
Dynamics and Applications. CRC Press, Baco Raton (2020)
29. Massopust, P.R.: Fractal Functions, Fractal Surfaces and Wavelets. Academic, Cambridge
(2017)
30. Barnsley, M.F.: Fractal functions and interpolation. Constr. Approx. 2(1), 303–329 (1986).
https://doi.org/10.1007/BF01893434
31. Kigami, J.: Analysis on Fractals. Cambridge University Press, Cambridge (2001)
32. Strichartz, R.S.: Analysis on fractals. Notices AMS 46(10), 1199–1208 (1999)
33. Barnsley, M.F., Elton, J., Hardin, D., Massopust, P.: Hidden variable fractal interpolation func-
tions. SIAM J. Math. Anal. 20(5), 1218–1242 (1989). https://doi.org/10.1137/0520080
34. Chand, A.K.B., Katiyar, S.K., Viswanathan, P.: Approximation using hidden variable fractal
interpolation function. J. Fractal Geom. 2, 81–114 (2015). https://doi.org/10.4171/JFG/17
35. Katiyar, S.K., Chand, A.K.B., Navascués, M.A.: Hidden variable A-fractal functions and their
monotonicity aspects. Rev. R. Acad. Cienc. Zaragoza 71, 7–30 (2016)
Mathematical Modeling
A Multi-strain Model for COVID-19
Abstract The main objective of this work is to propose and analyze a multi-
compartment ordinary differential equation model for multi-strain epidemic disease.
The proposed model mainly focuses on the epidemic disease spread due to SARS-
CoV-2, and the recurrent outbreaks are due to the emergence of a new strain. The
possibility of reinfection of the recovered individuals is considered in the model. The
multi-strain model is validated with the help of strain-specific daily infection data
from France and Italy.
1 Introduction
In description of the dynamics of disease progression over a short or long period, two
different types of compartmental epidemic models are used, namely, the model with
demography and model without demography [10, 23]. Researchers have considered
models with demography for established epidemic diseases which helps to perform
some preliminary stability analysis of the model under consideration around the
disease-free equilibrium (DFE) and the endemic equilibrium point. However, the
possibility of an endemic equilibrium in the context of COVID-19 remains in vein.
In some reported researches, the demographic terms are incorporated into the models
by considering certain rate of recruitment in the susceptible compartment from the
healthy compartment and the natural mortality within each of the compartments
involved with the model. In reality, one can determine the recruitment rate in the
susceptible compartment from healthy individuals only when we have a detailed
history of epidemic progression and relevant data for 10 years or more [6, 12, 16].
Further, the natural mortality rate of each compartment is based on the average life
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 125
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_10
126 S. Ghosh and M. Banerjee
2 Single-Strain Model
d S(t) β S(t)
=− (I (t) + α A(t)), (1a)
dt N
d E(t) β S(t)
= (I (t) + α A(t)) − σ E(t), (1b)
dt N
d I (t)
= r σ E(t) − ηI (t) − (δ I + μ I )I (t), (1c)
dt
d A(t)
= (1 − r )σ E(t) − δ A A(t), (1d)
dt
d H (t)
= ηI (t) − (δ H + μ H )H (t), (1e)
dt
d R(t)
= δ I I (t) + δ A A(t) + δ H H (t), (1f)
dt
subject to non-negative initial condition with empty hospitalized and recovered com-
partments. Here β is the disease transmission rate by symptomatic infected individ-
uals I (t); α is the ratio of the disease transmission rate by asymptomatic infected
individuals A(t) over symptomatic infected individuals I (t); σ is the reciprocal of
latent period; r is the rate at which exposed individuals become symptomatic; η is the
rate of hospitalization of symptomatic individuals; δ I , δ A , and δ H are recovery rate
from symptomatic, asymptomatic, and hospitalized compartments, respectively; μ I ,
μ H denote death rates in symptomatic and hospitalized compartments, respectively;
and N is the total population size.
First, we compute the controlled reproduction number of the system (1). The disease-
free equilibrium point is given by
(S, E, I, A, H, R) ≡ P1 (N , 0, 0, 0, 0, 0).
We use the next-generation matrix approach [13] to calculate the controlled reproduc-
tion number. We consider the compartments S, R as the non-infected compartments
and the compartments E, I , A, and H as the infected compartments. Then the matrix
F corresponding to the new infection and the matrix V corresponding to the outflow
from the infected compartments are given by
⎛ βS ⎞ ⎛ ⎞
N
(I + α A) σE
⎜ 0 ⎟ ⎜ ηI + (δ I + μ I )I − r σ E ⎟
F =⎜
⎝
⎟,
⎠ V=⎜
⎝ δ A A − (1 − r )σ E
⎟.
⎠
0
0 (δ H + μ H )H − ηI
A Multi-strain Model for COVID-19 129
The controlled reproduction number is the spectral radius of the matrix F V −1 and
is given by
βr αβ(1 − r )
Rc = + .
η + δI + μI δA
3 Two-Strain Model
To formulate the two-strain model, we assume that only one strain is responsible
for the initial outbreak of the epidemic until a new strain appears, and the second
strain appears at time t1 say. Then the model (1) is valid for t ≤ t1 and for t > t1
we have to define the model with two strains. Since there are many literature which
suggest that the re-infection may occur after 5–6 months after the earlier infection
[20], the re-infection of the individuals recovered from new strain (R2 compartment)
is ignored for the time being, and the same will be incorporated when another strain
will emerge. So, for the time t > t1 and till another strain emerges, we will use the
following model:
130 S. Ghosh and M. Banerjee
In this section, our focus is to compute the effective reproduction number at the time
t1 when a new strain emerges. We will find the effective reproduction number of the
system (2) around the point
where
⎛ βj αjβj ⎞
0 N
(S(t1 ) + ξ1 j R1 (t1 )) N
(S(t1 ) + ξ1 j R1 (t1 )) 0
⎜0 0 0 0⎟
Fj = ⎜
⎝0
⎟,
0 0 0⎠
0 0 0 0
⎛ ⎞
σj 0 0 0
⎜ −r j σ j (η + δ I + μjI) 0 0 ⎟
Vj = ⎜
⎝ −(1 − r j )σ j
j j ⎟,
⎠ j = 1, 2,
0 δjA 0
0 −η j 0 (δ j H + μjH)
and θ4×4 is the 4 × 4 null matrix. The spectral radius of F j V j−1 is given by
The corresponding effective reproduction number for the system (2) calculated at
time t1 is given by
R(t1 ) = max{R1 (t1 ), R2 (t1 )}.
Based on the magnitudes of the parameters and size of different compartments, one
can have either one or both of the quantities (R1 (t1 ) and R2 (t1 )) greater than one.
Here two parameters ξ and β play crucial role to determine R j (t1 ), j = 1, 2 greater
than one or not. Further outbreak of the epidemic is indicated by at least one of them
is greater than one.
The inflow in the exposed compartment corresponding to jth strain comes from
susceptible compartment and recovered from strain-1 to nth strain and hence their
growth rate can be described as
d E j (t) β j S(t)
= (I j (t) + α j A j (t)) − σ j E j (t)
dt N
n−1
βj
+ ξr j (I j (t) + α j A j (t))Rr (t), (4)
r =1
N
where η j is the rate of transfer from exposed to infected compartment once the incuba-
tion period is over, j = 1, 2, . . . , n. We can write the growth equations symptomatic
A Multi-strain Model for COVID-19 133
d I j (t)
= r j σ j E j (t) − η j I j (t) − (δ j I + μ j I )I j (t), (5)
dt
d A j (t)
= (1 − r j )σ j E j (t) − δ j A A j (t), (6)
dt
d H j (t)
= η j I j (t) − (δ j H + μ j H )H j (t), (7)
dt
where 1 ≤ j ≤ n, and the parameters have similar meaning as described earlier.
Once the nth strain arrives, the individuals of all the recovered compartments R j ,
1 ≤ j ≤ n − 1 can get re-infection from any one of the existing strains and hence
we can write the growth equation for R j as follows:
d R j (t)
= δ j I I j (t) + δ j A A j (t) + δ j H H j (t)
dt
n
βr
− ξ jr (Ir (t) + αr Ar (t))R j (t). (8)
r =1
N
d Rn (t)
= δn I In (t) + δn A An (t) + δn H Hn (t). (9)
dt
One can easily verify that the model (2) can be obtained from (3)–(9) with n = 2.
The proposed modeling approach takes care of the possibility of re-infection without
transferring the individuals from the recovered compartment to the susceptible com-
partment. One can calculate the effective reproduction number following the same
approach as described above for the model with two strains. Explicit expression will
be quite lengthy and hence we omit such calculations here.
Now we consider the numerical simulation results with hypothetical set of parameter
values before proceeding to the model validation with realistic dataset. Through-
out the numerical simulation, all the parameter values of β, σ , η, δ I , μ I , δ A , δ H ,
μ H in model (1) and the corresponding parameters in the model (3)–(9) are cho-
sen in the unit of day −1 [29]. α and r are proportionality constants and hence
they are dimensionless parameters. We consider the hypothetical set of parame-
134 S. Ghosh and M. Banerjee
Fig. 1 Simulation corresponding to model (2). Panel-a: The strain-1 appears at time t = 0 and no
new strain appears thereafter. The blue curve in this panel corresponds to only strain-1 with β1 = 0.6.
Panel-b: The strain-1 appears at time t = 0 and the strain-2 appears at time t = t1 = 1000. The
magenta curve extension and red curve extension in this panel correspond to β2 = 0.8 and β2 = 0.9,
respectively. Other parameter values are: N = 107 , r1 = 0.7, r2 = 0.8, ξ11 = 0.01, ξ12 = 0.05,
α j = 1, σ j = 1/5.2, η j = 0.1, δ j I = δ j A = 1/2.3, δ j H = 0.535, μ j I = μ j H = 0.03, for j = 1, 2.
Initial conditions: S(1) = N − 3, E(1) = I (1) = A(1) = 1, H (1) = R(1) = 0
and the peak is observed around 1,310th day. To understand the sensitivity of the
model parameters on the second outbreak, without any detailed mathematical insight,
simply we decrease the transmission rate of the second strain slightly by choosing
β2 = 0.8. Then we observe that the peak of the second outbreak (shown by the
magenta curve in Fig. 1b) is significantly low. Also the maximum number of daily
infected appears on 1,565th day approximately. For two different choices of β2 ,
the total number of infected on t = 2, 000 days are given by 4.2371 × 106 and
3.1721 × 106 corresponding to β2 = 0.9 (red curve in Fig. 1b) and β2 = 0.8 (magenta
curve in Fig. 1b), respectively. These simulation results indicate that the transmission
rate of the emerging strain affects the epidemic progression significantly.
It is important to mention here that the proposed model is capable of capturing
the multiple outbreaks of an epidemic disease due to the appearance of new strain.
This scenario is quite relevant in the context of COVID-19. For better illustration and
validation of our proposed model, we consider the model validation by estimating
the parameters from real dataset in the next section.
In this section, we consider the strain-specific data from France during the time
period March 1, 2020 to September 30, 2021, collected from [2]. During this span
of 608 days for COVID-19 epidemic spread and recurrent outbreaks, continuous
mutation of spike protein has led to many new variants of SARS-CoV-2 virus [3].
We have noticed that a new variant is responsible for behind each outbreak. Within
the said time period, we have observed four major outbreaks due to four different
strains which were identified as VOC. The first outbreak started roughly on March
1, 2020 followed by the successive outbreaks due to Alpha strain, Beta strain, and
Delta strain, respectively. It is important to mention here that during the beginning
of the COVID-19 epidemic spread, the strain-specific data source was not available,
that is why we consider all the strains before Alpha strain by a single strain which
is not identified with any specific name. As per information and data available at
[4], the VOC Alpha (20I (Alpha, V1)) appeared in France through the mutation
S:H69 around August 24, 2020. After that the VOC Beta (20H (Beta, V2)) appeared
approximately around December 31, 2020. The proportion of Gamma variant (20J
(Gamma, V3)) in France was negligible compared to some other countries. Further,
the VOC Delta (21J (Delta)) was reported around May 3, 2021.
For simplicity of numerical simulation and rapidity of convergence of the scheme,
we kept most of the parameters are same for all the variants except the disease
transmission rates (β j ) and the latency periods (1/r j ) for j = 1, 2, 3, 4. Here j = 1
corresponds to unidentified variant and then for France, Alpha, Beta, and Delta
strains correspond to j = 2, 3, 4, respectively. We fit the 7-day moving average
data of COVID-19 daily infection for France with the model (3)–(9) and estimate
136 S. Ghosh and M. Banerjee
Fig. 2 Plots of 7 days moving average of number of daily infected individuals in France starting
from March 1, 2020 to September 30, 2021 are plotted with the simulation results using the model
(3)–(9) with n = 4. Data points are marked in black and the simulation results are presented in
magenta color, a daily data and b cumulative data. The values of the parameters are given in the
text, and the strain-specific transmission rates β j are mentioned in Table 1
the parameters β j and r j (see Fig. 2). Other parameter values used for simulation
and fitting of the model with the data are chosen from the information available in
literature and mentioned below.
The estimated values of r j are r1 = 0.7, r2 = 0.8, r3 = 0.8, and r4 = 0.85. The
values of β j are estimated over different time intervals in order to take care of
change in the rate of disease progression due to several restrictions imposed and
relaxed successively. Estimated values of β j over different time intervals are listed
in Table 1.
The other parameter values are α j = 1, σ j = 0.1923, η j = 0.1, δ j I = δ j A =
0.4348, δ j H = 0.5347, μ j I = μ j H = 0.03, for j = 1, . . . , 4, (see [29] for details).
The value of N ≈ 6.74 × 107 and initial condition is S(0) ≈ N , E 1 (0) = 346,
I1 (0) = 6, A1 (0) = 6, H1 (0) = 12 and R1 (0) = 0. Initial values of other compart-
ments are considered as zero and here t = 0 corresponds to March 1, 2020. In case
of France, t1 = (August 24, 2020), t2 = (January 1, 2021), and t3 = (March 3, 2020)
as per the information available at [4]. The rates of re-infection with the same strain
are comparatively less and hence ξii = 0.007 and ξik = 0.009, for i = 1, 2, 3 and
k = 1, 2, 3, 4 and k = i. As ξik ’s are kept fixed throughout the simulation, the values
of βk ’s are estimated from time to time to validate with strain-specific daily infection
data.
The proposed model can be validated with strain-specific data from any other
country and the parameters can be estimated accordingly. To substantiate our claim,
the model is fitted with the strain-specific data from Italy. Estimated values of β j ’s
with associated time intervals are presented in Table 2. The other parameter values are
α j = 1, σ j = 0.1923, η j = 0.1, δ j I = δ j A = 0.4348, δ j H = 0.5347, μ j I = μ j H =
0.03, for j = 1, . . . , 4, (see [29] for details). The value of N ≈ 5.96 × 107 and
initial condition is S(0) ≈ N , E 1 (0) = 10, I1 (0) = 2, A1 (0) = 1, H1 (0) = 5, and
A Multi-strain Model for COVID-19 137
R1 (0) = 0. The rates of re-infection with the same strain are comparatively less
and hence ξii = 0.007 and ξik = 0.009, for i = 1, 2, 3 and k = 1, 2, 3, 4 and k = i.
Simulation results along with the data are presented in Fig. 3. In [29], one can find the
estimates of the parameters for four countries—France, Italy, Germany, and Spain.
It is worthy to mention here that the proposed modeling approach is capable of
capturing the recent outbreak due to Omicron strain but we refrain ourselves from
such attempt as the outbreak is not over yet.
138 S. Ghosh and M. Banerjee
Fig. 3 Number of daily infected and 7-day cumulative number of infected individuals in Italy
starting from February 1, 2020 to May 31, 2021 are plotted with the simulation results using the
model (3)–(9) with n = 4. The values of the parameters are given in the text, and the strain-specific
transmission rates β j are presented in Table 2
5 Discussion
estimated effectively once we find accurate re-infection data. We did not consider
any isolated compartment and quarantined compartment but the estimation of the
rate of infection over different time period implicitly takes care of social restrictions
imposed and withdrawn from time to time [27].
In France, the first outbreak roughly started on March 1, 2020, and then sev-
eral VOCs appeared afterward. From the model and estimated parameter values, we
find the controlled reproduction number at the epidemic’s beginning was 1.833. The
effective reproduction numbers at the time of emergence of Alpha, Beta, and Delta
strains were 1.68, 1.71, and 1.63, respectively. During the calculation of effective
reproduction numbers at different time points, it is observed that the corresponding
new emerging strain dominates the effective reproduction number. Hence, the emerg-
ing strain at a particular time point plays a crucial role in determining the growth or
decay of an epidemic. The disease transmission rate (β) and the latency period (1/r )
corresponding to a particular strain depend upon the within-body viral load dynam-
ics. It is essential to incorporate within-body viral dynamics of a particular strain
into the modeling approach to have a better result, like the immuno-epidemic model
recently considered in [15]. In the context of COVID-19, it is also essential to incor-
porate vaccine-induced immunological responses into the model to understand the
epidemic progression more accurately. The vaccinated compartment is overlooked
here as the strain-specific effectiveness of the vaccines is not clear yet. Including
a vaccinated compartment will only alter the estimates of infectivity, which can be
carried out similarly as outlined here.
In this work, we have fitted our model with the data from France and Italy for
three prominent VOCs and one initial outbreak for which the strain is not identified.
Before the appearance of Omicron strain, the infection is reported through some other
strains apart from Alpha, Beta, and Delta in these two countries. Other strains are
also responsible for the spread of the epidemic to a certain extent, but their numbers
are negligible compared to the number of infections due to the VOCs. Of course, one
can consider all these strains to develop a better model that can effectively capture
the number of hospitalizations and deaths more accurately. Our primary motivation
for this work is to propose and validate a relatively new model of epidemic spread
which can capture the successive outbreaks and validate the model with the accurate
strain-specific data for the COVID-19 epidemic. It is needless to say that the proposed
modeling approach can be applied to other epidemic diseases also where more than
one strain is responsible for successive outbreaks.
References
1. https://www.who.int/en/activities/tracking-SARS-CoV-2-variants/
2. https://covid.ourworldindata.org
3. https://nextstrain.org/ncov/gisaid/global/6m
4. https://ourworldindata.org/grapher/covid-cases-omicron?time=2022-01-24&
country=GBR~FRA~BEL~DEU~ITA~ESP~USA~ZAF~BWA~AUS
140 S. Ghosh and M. Banerjee
5. Abu-Raddad, L.J., Chemaitelly, H., Ayoub, H.H., Tang, P., Coyle, P., Hasan, M.R., Yassine,
H.M., Benslimane, F.M., Al Khatib, H.A., Al Kanaani, Z., et al.: Effect of vaccination and of
prior infection on infectiousness of vaccine breakthrough infections and reinfections. medRxiv
(2021)
6. Anderson, R.M., May, R.M.: Infectious Diseases of Humans: Dynamics and Control. Oxford
University Press, Oxford (1992)
7. Aniţa, S., Banerjee, M., Ghosh, S., Volpert, V.: Vaccination in a two-group epidemic model.
Appl. Math. Lett. 119, 107197 (2021)
8. Banerjee, M., Ghosh, S., Volpert, V.: Epidemic progression in a heterogeneously distributed
population (2021)
9. Banerjee, M., Lipniacki, T., d’Onofrio, A., Volpert, V.: Epidemic model with a strain-dependent
transmission rate. Commun. Nonlinear Sci. Num. Simul. 114, 106641 (2022)
10. Brauer, F., Castillo-Chavez, C., Feng, Z.: Mathematical Models in Epidemiology, vol. 32.
Springer, Berlin (2019)
11. Chintu, C., Athale, U.H., Patil, P.: Childhood cancers in zambia before and after the hiv epi-
demic. Arch. Dis. Childhood 73(2), 100–105 (1995)
12. Diekmann, O., Heesterbeek, J.A.P.: Mathematical Epidemiology of Infectious Diseases: Model
Building, Analysis and Interpretation, vol. 5. Wiley, New York (2000)
13. Van den Driessche, P., Watmough, J.: Reproduction numbers and sub-threshold endemic equi-
libria for compartmental models of disease transmission. Math. Biosci. 180(1–2), 29–48 (2002)
14. Fisher-Hoch, S., Hutwagner, L.: Opportunistic candidiasis: an epidemic of the 1980s. Clin.
Infecti. Dis. 21(4), 897–904 (1995)
15. Ghosh, S., Banerjee, M., Volpert, V.: Immuno-epidemiological model-based prediction of fur-
ther covid-19 epidemic outbreaks due to immunity waning. Math. Modell. Nat. Phenom. 17, 9
(2022)
16. Keeling, M.J., Rohani, P.: Modeling Infectious Diseases in Humans and Animals. Princeton
University Press, Princeton (2011)
17. Kermack, W.O., McKendrick, A.G.: A contribution to the mathematical theory of epidemics.
Proc. R. Soc. Lond. Ser. A Contain. Papers Math. Phys. Character 115(772), 700–721 (1927)
18. Kermack, W.O., McKendrick, A.G.: Contributions to the mathematical theory of epidemics.
ii.-the problem of endemicity. Proc. R. Soc. Lond. Ser. A Contain. Papers Math. Phys. Character
138(834), 55–83 (1932)
19. Kermack, W.O., McKendrick, A.G.: Contributions to the mathematical theory of epidemics.
iii.-further studies of the problem of endemicity. Proc. R. Soc. Lond. Ser. A Contain. Papers
Math. Phys. Character 141(843), 94–122 (1933)
20. Kim, P., Gordon, S.M., Sheehan, M.M., Rothberg, M.B.: Duration of sars-cov-2 natural immu-
nity and protection against the delta variant: a retrospective cohort study. Clin. Inf. Dis. (2021)
21. de León, U.A.P., Pérez, Á.G., Avila-Vales, E.: An seiard epidemic model for covid-19 in
mexico: mathematical analysis and state-level forecast. Chaos Solitons Fractals 140, 110165
(2020)
22. Mahase, E.: Covid-19: past infection provides 83% protection for five months but may not stop
transmission, study finds (2021)
23. Martcheva, M.: An Introduction to Mathematical Epidemiology, vol. 61. Springer, Berlin (2015)
24. Murray, J.: Mathematical Biology, vol. 1. Springer, Heidelberg (2002)
25. Parolini, N., Dede’, L., Antonietti, P.F., Ardenghi, G., Manzoni, A., Miglio, E., Pugliese, A.,
Verani, M., Quarteroni, A.: Suihter: a new mathematical model for covid-19. application to
the analysis of the second epidemic outbreak in Italy. Proc. R. Soc. A 477(2253), 20210027
(2021)
26. Patterson, K.D., Pyle, G.F.: The geography and mortality of the 1918 influenza pandemic. Bull.
Hist. Med. 65(1), 4–21 (1991)
A Multi-strain Model for COVID-19 141
27. Sharma, S., Volpert, V., Banerjee, M.: Extended seiqr type model for covid-19 epidemic and
data analysis. Math. Biosci. Eng. (2020)
28. Volpert, V., Banerjee, M., Sharma, S.: Epidemic progression and vaccination in a heterogeneous
population. application to the covid-19 epidemic. Ecolog. Compl. 47, 100940 (2021)
29. Wang, K., Ding, L., Yan, Y., Dai, C., Qu, M., Jiayi, D., Hao, X.: Modelling the initial epidemic
trends of covid-19 in Italy, Spain, Germany, and France. PLoS One 15(11), e0241743 (2020)
Effect of Nonlinear Prey Refuge
on Predator–Prey Dynamics
1 Introduction
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 143
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_11
144 S. Samaddar et al.
response first described the downtrend consumption rate. It elaborates that predator
growth due to prey consumption is limited by its food processing capability. Math-
bx
ematical expression for Holling type-II functional response is 1+hbx where b and h,
respectively, represent the prey attack rate and handling time of predators. By con-
b(1−m)x
sidering constant prey refuge, the function becomes 1+hb(1−m)x . Using this response
eminent works have been performed. Ji and Wu [4] have shown in their study how
prey refuge is capable of sustaining the prey density from extinction as well as the
dynamical stability behavior depending on refuge factor on harvesting efforts.
Although there are marvelous results on constant prey refuge, for natural ecolog-
ical scenario it is little unrealistic. In general, preys are seeking refuge due to the
presence of predator. It is quite obvious that the refuge count cannot be exactly con-
stant as refuge preys will have less foraging that can lead to starvation. It will force
them to be exposed again. As the refuge prey depends on the existence of predators,
it should depend on predator density also. Since prey refuge is a very important
factor to any dynamics, consideration of nonlinear refuge which depends on both
the species can solve this problem. In that aspect we have incorporated a nonlinear
refuge in the functional response as described in the study [5–7]. The nonlinear prey
refuge count can be expressed as the function g(x, y) = a+y mx y
where m is the prey
refuge coefficient, a is the half saturation constant of refuge prey, and y is the preda-
tor population. Then the density of prey available for hunting becomes 1 − a+y my
x
my
b 1− a+y x
and the functional response turns to .
my
1+hb 1− a+y x
The rest of the paper is organized as follows: in Sect. 2, we have first described the
mathematical model with constant prey refuge and find all the conditions for local
stability, instability, and Hopf bifurcation. Here we also numerically represent all the
analytical findings. Later, in Sect. 3, we formulated the model with nonlinear prey
refuge and elaborated all the stability and bifurcation conditions analytically and
numerically. To show the dynamical changes a system can achieve due to nonlinear
prey, we perform a comparison of these two models in Sect. 4. Finally, in Sect. 5, we
concluded all the findings of this study.
First we will discuss about the mathematical model which consists of constant prey
refuge in a Holling type-II functional response. The model can be represented as
dX X B(1 − m)X Y
= R X (1 − ) − ,
dT K 1 + H B(1 − m)X
dY E B(1 − m)X Y
= − DY,
dT 1 + H B(1 − m)X
X (0) > 0; Y (0) > 0.
Effect of Nonlinear Prey Refuge on Predator–Prey Dynamics 145
0.7 0.7
0.6 0.65
Predator
Species
y(t)
0.5 0.6
x(t)
0.4 0.55
(0.3846,0.5917)
0.3 0.5
0 50 100 150 200 250 300 0.3 0.35 0.4 0.45 0.5 0.55
Time Prey
( a) (b)
Here X, Y represent prey and predator density at any time T . The parameter used
R, K , B, H, E, D, respectively, represent prey intrinsic growth rate, prey carrying
capacity, predator attack rate, prey handing time of predators, conversion factor from
prey biomass to predator biomass, and death rate of predator. Here m denotes the
constant prey proportion and m ∈ [0, 1).
Nondimensionalization of the system using the transformations X = K x, Y =
E K y, and T = Rt provides
dx b(1 − m)x y
= x(1 − x) − , (1a)
dt 1 + hx(1 − m)
dy b(1 − m)x y
= − dy, (1b)
dt 1 + hx(1 − m)
x(0) > 0, y(0) > 0, (1c)
where b = B ER K , h = B H K , and d = D
R
. The system has three ecological equilib-
rium points which are
1. E 0 (0, 0) which is always unstable saddle.
2. E 1 (1, 0) which is stable if d > 1+h(1−m)
b(1−m)
else unstable.
3. E 2 (x∗ , y∗ ) where x∗ = (1−m)(b−dh) and y∗ = (1−x∗ )(1+h(1−m)x
d
b(1−m)
∗)
. The interior equi-
librium exists if b > dh. The system is locally asymptotically stable at E 2 if
(1 − m) < h(b−dh)
b+dh
else unstable.
We fix the parameter values as b = 2.5, h = 3, d = 0.4, m = 0.2. For these
parameter values, the system (1) gets stability at (0.3846, 0.5917) which is presented
in Fig. 1.
Theorem 1 The system encounters Hopf bifurcation at m [h] = 1 − b+dh
h(b−dh)
.
Proof The trace of the Jacobian matrix J of the system around E 2 is zero at m [h]
and the determinant is positive. Additionally,
d
dm
T race(J ) |m=m [h] = − b(1−m [h] )2 (b−dh) = 0.
d(b+dh)
0.6 0.7
0.5
0.6
Max(y)
Predator
0.4
Prey
0.3 Min(x)
0.4
0.2
0.1 0.3
0 0.05 0.1 0.15 0 0.05 0.1 0.15
m m
(a) (b)
0.38 0.6
0.34 0.56
Prey
0.32 0.54
0.3 0.52
0.28 0.5
700 750 800 850 900 950 1000 700 750 800 850 900 950 1000
Time Time
(a) (b)
0.551
0.55
0.549
Predator
0.548
0.547
(0.32433,0.54785)
0.546
0.545
0.322 0.324 0.326 0.328
Prey
(c)
Fig. 3 Occurrence of Limit cycle of the system (1) around E 2 = (0.32433, 0.54785) near m [h]
At m [h] = 0.0513, the system undergoes Hopf bifurcation and generates limit cycle
around (0.32433, 0.54785). Figure 2 denotes the bifurcation diagram of the system
with respect to m. Figure 3a and b, respectively, represents the periodic behavior of
prey and predator biomass and Fig. 3c denotes the limit cycle around the interior
equilibrium point.
Effect of Nonlinear Prey Refuge on Predator–Prey Dynamics 147
In natural scenario, it is very rare to get a species which has constant refuge. Prey
takes refuge seeking some protection from predator. It is pretty clear that prey refuge
count must depend on predator density. For this purpose, we have incorporated a
nonlinear prey refuge function g(x, y) = a+y mx y
, where m is the refuge constant and a
is the half saturation constant of prey refuge. The modified model becomes
dx b(1 − a+y
my
)x y
= x(1 − x) − my , (2a)
dt 1 + hx(1 − a+y )
dy b(1 − a+y )x y
my
= − dy, (2b)
dt my
1 + hx(1 − a+y )
x(0) > 0, y(0) > 0. (2c)
b{a + (1 − m)y}y
f1 ≡ 1−x − =0 (3a)
a + y + h{a + (1 − m)y}x
b{a + (1 − m)y}y
f2 ≡ − d = 0. (3b)
a + y + h{a + (1 − m)y}x
a[(b−dh)x∗ −d]
Solving the nullcline f 2 = 0 we get y∗ = d−(1−m)(b−dh)x ∗
and putting the value in
the nullcline f 1 = 0 we get x∗ is a solution of
Descartes’ rule of signs ensures that Eq. 4 has exactly one negative solution and at
most two positive solutions. Now the sum of the solutions is 1 + (1−m)(b−dh)
d
. Since
x∗ < min{1, (1−m)(b−dh) }, if positive solution exists then there will be exactly one
d
148 S. Samaddar et al.
where
bh{a + (1 − m)y}2 y
f x1 = −1 +
[a + y + h{a + (1 − m)y}x]2
b(1 − m)y 2 {1 + h(1 − m)x} + ab(1 + hx){a + 2(1 − m)y}
f y1 = − <0
[a + y + h{a + (1 − m)y}x]2
b(a + y){a + (1 − m)y}
f x2 = >0
[a + y + h{a + (1 − m)y}x]2
abmx
f y2 = − < 0.
[a + y + h{a + (1 − m)y}x]2
Using the implicit function theorem, the determinant of the Jacobian can be written as
( f 2)
dy ( f )
1
1 2 dy
Det (J ) = x y f y f y − .
dx dx
(x∗ ,y∗ )
2) 1
dy ( f dy ( f )
Now dx
is always positive and dx
can be positive or negative at E 2 . If
1 1
dy ( f ) dy ( f )
dx
< 0, then Det (J ) is positive. If d x > 0, then from Fig. 4 it is clear that the
( f 2) ( f 1)
slope of f 2 is greater than slope of f 1 . So dyd x is always greater than dyd x . Hence
Det (J ) is always positive at E 2 (x∗ , y∗ ). By Routh–Hurwitz criterion, the system is
stable at positive equilibrium if T race(J ) < 0, i.e., x f x1 + y f 2 y < 0 at E 2 .
For numerical simulation purpose, we have taken the value of a = 1.
Theorem 3 The interior equilibrium point E 2 of the system (2) loses its stability
when x f x1 + y f 2 y > 0. The equilibrium point may change its stability trough Hopf
bifurcation at m = m [h] .
Effect of Nonlinear Prey Refuge on Predator–Prey Dynamics 149
y
0.4
f2
0.2
0
0 0.1 0.2 0.3 0.4 0.5
x
Proof The explicit expression of the equilibrium points is not available. This makes
the analytical proof of Hopf bifurcation very difficult. But numerically we can prove
that the system encounters Hopf bifurcation for the preassigned parameter values at
m [h] = 0.11358 around the positive equilibrium point E 2 = (0.320526, 0.544473).
4 Model Comparison
The axial equilibrium point of the system (1) with constant prey refuge is stable
when d > 1+h(1−m)
b(1−m)
while in case of the system (2) with nonlinear prey refuge it is
stable if d > 1+hb
. This means the stability of system (2) does not depend on prey
refuge. Further, for the refuge count m = 0.7, we have chosen three initial values
and compare the dynamics for both the systems. It is noted that for the first system
the trajectories approach to (1, 0), which means predators are extinct but in the case
of the system (2), the trajectories approach to an interior equilibrium point, which
means predators survive in system (2).
0.8 0.8
0.7
0.7
0.6
Predator
Species
0.6
0.5 y(t)
0.5
0.4 x(t)
0.3 0.4
(0.3313,0.5539)
0.2 0.3
0 200 400 600 800 1000 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Time Prey
(a) (b)
0.6 0.7
0.5
0.6
Max(y)
Predator
0.4 Min(y)
Prey
Max(x)
0.5
Min(x)
0.3
0.4
0.2
0.1 0.3
0 0.05 0.1 0.15 0.2 0.25 0.3 0 0.05 0.1 0.15 0.2 0.25 0.3
m m
(a) (b)
models with constant prey refuge and nonlinear prey refuge. For both the models, we
have described ecologically possible equilibrium points and derive all the stability
and bifurcation conditions. Local stability nature of the positive equilibrium of the
system (1) can be understood from Figs. 1, 2 and 3. On the other hand, the stability
nature of the positive equilibrium of system (2) can be understood from Figs. 5, 6
and 7. Also a model comparison is performed (see Fig. 8) which confirms that at high
refuge, predators can survive in the nonlinear prey refuge environment. At m = 0.7,
predator extincts in the system (1) but survives in the system (2). Hence, the nonlinear
prey refuge has an significant impact on any system and our study drives the system
toward more natural scenario.
Effect of Nonlinear Prey Refuge on Predator–Prey Dynamics 151
0.33 0.555
0.325 0.55
Predator
Prey
0.32 0.545
0.315 0.54
0.31 0.535
3500 3550 3600 3650 3700 3750 3800 3500 3550 3600 3650 3700 3750 3800
Time Time
(a ) (b)
0.552
0.55
Predator
0.548
0.546
(0.3232,0.5469)
0.544
0.542
0.318 0.32 0.322 0.324 0.326 0.328
Prey
(c)
Fig. 7 Occurrence of Hopf bifurcation of the system (2) around E 2 at m [h] = 0.11358
constant
prey refuge
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1 1.2
Prey
References
4. Ji, L., Wu, C.: Qualitative analysis of a predator-prey model with constant-rate prey harvesting
incorporating a constant prey refuge. Nonlinear Anal.: Real World Appl. 11(4), 2285–2295
(2010)
5. Molla, H., Sarwardi, S., Sajid, M.: Predator-prey dynamics with allee effect on predator species
subject to intra-specific competition and nonlinear prey refuge. J. Math. Comput. Sci 25, 150–
165 (2021)
6. Mondal, S., Samanta, G.: Dynamics of a delayed predator-prey interaction incorporating non-
linear prey refuge under the influence of fear effect and additional food. J. Phys. A: Math.
Theor. 53(29), 295601 (2020)
7. Mondal, S., Samanta, G., Nieto, J.J.: Dynamics of a predator-prey population in the presence
of resource subsidy under the influence of nonlinear prey refuge and fear effect. Complexity
2021 (2021)
8. Samaddar, S., Dhar, M., Bhattacharya, P.: Effect of fear on prey–predator dynamics: Exploring
the role of prey refuge and additional food. Chaos: Interdiscip. J. Nonlinear Sci. 30(6), 063129
(2020)
9. Samaddar, S., Dhar, M., Bhattacharya, P.: Supplement of additional food: dynamics of self-
competitive prey-predator system incorporating prey refuge. Iran. J. Sci. Technol. Trans. A:
Sci. 44(1), 143–153 (2020)
10. Samaddar, S., Dhar, M., Bhattacharya, P.: Impact of refuge to the heterogeneous interaction of
species in food chain model: a holistic approach. Iran. J. Sci. Technol. Trans. A: Sci. 45(1),
221–233 (2021)
11. Wang, J., Cai, Y., Fu, S., Wang, W.: The effect of the fear factor on the dynamics of a predator-
prey model incorporating the prey refuge. Chaos: Interdiscip. J. Nonlinear Sci. 29(8), 083109
(2019)
Effects of Magnetic Field and Thermal
Conductivity Variance on Thermal
Excitation Developed by Laser Pulses
and Thermal Shock
Rakhi Tiwari
Abstract The current investigation is aimed to execute the influence of the magnetic
field on the transient outcomes inside a semi-infinite medium with dual-phase lag
thermoelasticity. Properties of the considered material are taken to be variable, i.e.
not constant. The boundary of the medium is exposed to a sudden heat input (thermic
shock). Moreover, the bounded surface is assumed to be affected by a non-Gaussian
laser beam-type heat source. Closed-form solutions are evaluated by adopting the
concept of Kirchhoff transformation and Laplace transform. Impacts of the magnetic
field as well as thermal conductivity variance are deduced on the important field quan-
tities such as dimensionless displacement, dimensionless conductive temperature, as
well as dimensionless stress through quantitative results. Auspicious outcomes are
achieved and the prominent role of the magnetic field and variations of the thermal
conductivity are observed on the field components. The author believes that the
current theoretical study will be helpful in designing the various structures affected
by the laser beam of a non-Gaussian pattern.
1 Introduction
R. Tiwari (B)
Department of Mathematics, Nitishwar College, a Constituent Unit of Babasaheb Bhimrao
Ambedkar Bihar University, Bihar, India
e-mail: [email protected]
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 153
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_12
154 R. Tiwari
I0 denotes the laser intensity, and t p is the characteristic time of the laser pulse.
Heat source Q(x, t) is stated as
x
1− R Ra I 0 t x t
Q(x, t) = exp − L(t) = exp − − . (2)
b b bt p2 b tp
The basic equations representing the generalized dual-phase lag theory of thermoe-
lasticity are presented below.
Equation of motion (body force is observed to be absent) [15]:
σi j, j + μ0 (( J × H)i = ρ ü i . (3)
μ0 (( J × H)i represents the Lorentz force as the half-space medium is suffering from
the magnetic field.
The strain–displacement relation is [15]
1
ei j = u i, j + u j,i . (4)
2
The constitutive equations are [15]
ρCv
ρη = γ ekk + θ. (6)
T0
The thermal conductivity equation in DPL theory (in the presence of a heat source)
[15]:
∂ ∂
K 1 + τT ∇ θ = 1 + τq
2
ρCv θ̇ + γ T0 ė − Q . (7)
∂t ∂t
∂u
ex x = . (9)
∂x
Maxwell equations:
The following Maxwell equations will take place in the mathematical modelling of
the problem
Effects of Magnetic Field and Thermal Conductivity Variance … 157
∂E ∂h ∂u
curl h = J + 0 , curl E = −μ0 , E = −μ0 × H , div h = 0.
∂t ∂t ∂t
(10)
Now the governing Eqs. (3), (5) and (7) obtain the following form:
∂ 2u 2∂ u
2
γ ∂θ
α = c − , (12)
∂t 2 0
∂x 2 ρ ∂x
∂u
σx x = (λ + 2μ)− γ θ, (13)
∂x
∂ ∂ 2θ ∂ ·
K 1 + τT = 1 + τq ρC v θ̇ + γ T0 ekk −Q . (14)
∂t ∂ x 2 ∂t
2
α0 μ0 (λ+2μ)
α = 1+ c 2 , α0 = H0 ρ
represents Alfven velocity, c0
2
= C 1
2
+α0
2
, C 1 = ρ
represents the propagation of longitudinal wave and c = 1
0 μ0
, where c denotes the
speed of light.
u(x, 0) = u̇(x, 0) = 0,
(15)
θ (x, 0) = θ̇ (x, 0) = 0.
σ (0, t) = 0,
(16)
θ (0, t) = θ0 H (t),
158 R. Tiwari
σ (x → ∞, t) = 0,
(17)
θ (x → ∞, t) = 0.
K = K (θ ) = K 0 (1 + K 1 · θ ), (18)
K0
N= . (19)
ρCv
θ
1
ψ= K (θ )dθ. (20)
K0
0
∂ψ ∂θ
K0 = K (θ ) . (21)
∂x ∂x
Again differentiating Eq. (21) with respect to x, we obtain
∂ 2ψ ∂ ∂θ
K0 = K (θ ) . (22)
∂x2 ∂x ∂x
∂ψ ∂θ
K0 = K (θ ) . (23)
∂t ∂t
With the help of Eqs. (20), (22), we obtain
1
ψ =θ+ K1 · θ 2. (24)
2
Effects of Magnetic Field and Thermal Conductivity Variance … 159
1
ψ(0, t) = θ (0, t) + K 1 · (θ (0, t))2 . (25)
2
With the help of the equation mentioned above, the value of θ can be determined
in terms of ψ in the following manner:
√
−1 + 2K 1 · ψ + 1
θ= . (26)
K1
Using Eqs. (20)–(26) in Eqs. (12)–(14), we find the following equations (we have
neglected second and higher order):
∂ 2u ∂ 2u γ ∂ψ
α
= c02 2 − , (27)
∂t 2 ∂x ρ ∂x
∂ ∂ ∂ ψ γ T0 Q
1 + τT ∇ θ = 1 + τq
2
+ ekk − , (28)
∂t ∂t ∂t N K0 K0
∂u
σx x = σ = (λ + 2μ) − γ ψ. (29)
∂x
Non-dimensionalization: For the purpose of simplifying Eqs. ((27)–(29)), we
convert these equations to non-dimensional form with the help of following non-
dimensional quantities:
x = c0 η0 x, u = c0 η0 u, b = c0 η0 b, t = c02 η0 t, t p = c02 η0 t p , τT = c02 η0 τT ,
τq = c02 η0 τq ,
η0 γθ γψ
L 0 = Ccv0ρT 0
L 0 , θ = λ+2μ , ψ = λ+2μ , σ = λ+2μσ
, c02 = λ+2μ ρ
, h = Hh0 ,
E = μ0EH0 v , J = H0Jvη .
∂ 2u ∂ 2u ∂ψ
α= β − , (30)
∂t 2 ∂x2 ∂x
x
∂ ∂ 2ψ ∂ ∂ t
1 + τT = 1 + τq (ψ + N 1 ekk ) − N 2 t exp − .ex p − ,
∂t ∂ x 2 ∂t ∂t b tp
(31)
∂u
σ = − ψ. (32)
∂x
T0 γ 2 N γρCv T0 N Ra L 0
N1 = , N2 = . (33)
K 0 (λ + 2μ) K 0 (λ + 2μ)bt p2
c02
For convenience, primes are suppressed from Eqs. (30)–(32). β = C12
.
160 R. Tiwari
(1 + sτT )D 2 ψ = 1 + sτq s ψ + N1 Du − G(x, s) , (35)
σ = Du − ψ. (36)
∂
x
Here, D = ∂x
and G(x, s) =
N2
2 exp −b .
s+ t1p
Initial and boundary conditions are obtained in Laplace transform domain in the
following way:
σ (0, s) = 0,
(38)
θ(0, s) = 1s θ0 .
σ (x → ∞, s) = 0,
(39)
θ(x → ∞, s) = 0.
1 1 1
ψ(0, s) = θ0 + K 1 · (θ0 )2 . (40)
s 2 s
Eliminating ψ from Eqs. (34), (35), we obtain the following decoupled differential
equation in terms of u:
x
D 4 − a1 D 2 + a2 u = N4 exp − . (41)
b
1 1 + sτq s 3 1 + sτq
a1 = (αs 2 + s (N1 + 1)), a2 = ,
β 1 + sτT β 1 + sτT
N2 1 + sτq
N4 = 2 .
bβ(1 + sτT ) s + t1p
The general solution of the differential equation with degree four (Eq. (41)) is
obtained as
x
u(x, s) = Ae−ξ1 x + Be−ξ2 x + Ce− b . (42)
Effects of Magnetic Field and Thermal Conductivity Variance … 161
ξi (i = 1, 2) denote the roots with positive real parts of Eq. (44) which are evaluated
as
a + a 2 − 4a a − a 2 − 4a
1 1 2 1 1 2 N4 b4
ξ1 = , ξ2 = and C = .
2 2 1 − a1 b 2 + a2 b 4
2 2
where ψ1 = 1
ξ1
αs − βξ12 A, ψ2 = 1
ξ2
αs − βξ22 B, ψ3 = αbs 2 − βb C,
−αs 2 + (β − 1)ξ12 A −αs 2 + (β − 1)ξ22 B (β − 1)C
σ1 = σ2 = , σ3 = − αbs 2 C.
ξ1 ξ2 b
(β − 1)C
M1 A + M2 B = αbs 2 C − , (45)
b
β θ0 K1 2
N1 A + N2 B = − αbs C +
2
+ θ . (46)
b s 2s 0
2 2
αs − βξ12 αs − βξ22 −αs 2 + (β − 1)ξ12
N1 = , N2 = , M1 = ,
ξ1 ξ2 ξ1
−αs 2 + (β − 1)ξ22
M2 = .
ξ2
θ0 M2 K 1 M2
A= + θ2
s(N1 M2 − N2 M1 ) 2s(N1 M2 − N2 M1 ) 0
C β (β + 1)
+ M2 −αbs 2 + + N2 − αbs 2 , (47)
(N1 M2 − N2 M1 ) b b
θ0 M1 K 1 M1
B= + θ 2
s(N2 M1 − N1 M2 ) 2s(N2 M1 − N1 M2 ) 0
C β (β + 1)
+ M1 −αbs 2 + + N1 − αbs 2 . (48)
(N2 M1 − N1 M2 ) b b
162 R. Tiwari
7 Quantitative Results
The current section derives the variations of the dimensionless conductance tempera-
ture ψ, dimensionless displacement u and non-dimensional stress σ̃ for time t = 0.35
against the non-dimensional distance x by sketching graphical results. The ‘Bellmen
Technique’ has been adopted in order to find the numerical inversion of the Laplace
transform. We have divided numerical results into two subsections. The first subsec-
tion exhibits the impact of magnetic field on the variations of all physical fields,
while the other subsection characterizes the role of changing thermal conductivity
on the nature of field quantities.
Copper material is selected for the purpose of the computational study. Physical
data [15] for the material is
383.1J
K 0 = 386 Wm−1 K−1 , cv = , T0 = 293 K,
kg K
L 0 = 1 × 102 JM−2 , Ra = 0.5, τT = 0.15, τq = 0.2,
b = 0.01 m, θ0 = 1α = 0.01442, β = 1.
The present subsection exhibits the impact of magnetic field intensity H on the vari-
ations in various dimensionless field components—conductive temperature. Here,
the value of K 1 = 0.2.
Figure 1 exhibits the variations of conductive temperature ψ with respect to the
distance x at time t = 0.35. The trend of variation of plots is observed to be almost
similar to the change in the magnetic field intensity. However, the values of the
temperature field are altered significantly as we change the values of the magnetic
field. Plots start from the same constant value and increase as the distance increases
but after providing maximum value, they go down and vanish. It is noticed that
the highest values of the temperature field are observed for the highest value of the
magnetic field, and values of the temperature profile enhance as the magnetic field
intensity increases. Further, it is notified that the peak point shifts far from the first
boundary when the magnetic field intensity becomes smaller.
Figures 2 and 3 demonstrate the disparity of the distributions of displacement
and stress field, respectively, at time t = 0.35. Similar to the temperature field, the
impact of the magnetic field is observed to be very prominent on the variations of
displacement and stress fields. The stress field is most influenced by the variations of
Effects of Magnetic Field and Thermal Conductivity Variance … 163
Fig. 1 Disparity of
conductive temperature ψ
versus distance x for time
t = 0.35
Fig. 2 Disparity of
displacement u versus
distance x for time t = 0.35
magnetic field intensity. Additionally, we observe that changing the magnetic field
affects the maximum value of the displacement field. Apart from this, it is noticed
that as the value of the magnetic field goes beyond 2000, changes in the measures of
the displacement fields become maximum compared to lower values of the magnetic
field.
For H = 3000, the nature of the stress curve is observed to be changed compared
to the lower intensity of the magnetic field.
It can be concluded from the above discussion that stress profiles suffer from the
discontinuities unlike the temperature and displacement fields.
164 R. Tiwari
In the current subsection, the act of the thermal conductivity variance is displayed on
the variations in different dimensionless physical fields—displacement, conductive
temperature and stress. The value of magnetic field intensity H is taken to be constant
such as H = 3500.
Figures 4, 5 and 6 demonstrate the changes in the dimensionless profiles of conduc-
tive temperature, displacement as well as stress with respect to the dimensionless
distance x, (0 ≤ x ≤ 5) at time t = 0.35, respectively.
Figure 4 states the behaviour of conductive temperature for the varied measures of
changing thermal conductivity parameter K 1 = 0, 0.2, 0.4, 0.6. The variance pattern
of the temperature line is found to be unaffected by the changes in the conductivity
Fig. 4 Disparity of
conductive temperature ψ
versus distance x for time
t = 0.35
Effects of Magnetic Field and Thermal Conductivity Variance … 165
Fig. 5 Disparity of
displacement u versus
distance x for time t = 0.35
8 Conclusion
On the basis of the present study, the following observations can be made:
(1) All field quantities attain non-zero values in the bounded region and fade
beyond this region. This nature of magneto-thermoelastic waves shows that
waves propagate with a finite speed.
(2) Alteration in the magnetic field intensity changes prominently the values of
the profiles of all field quantities, and these changes occur prominently near
the boundary of the half-space for stress and displacement fields; however, the
temperature field is highly affected at the peak values. Stress curves suffer major
changes at the boundary of the half-space. The high value of the magnetic field
intensity increases the values of physical fields. Prominent changes among the
values of physical fields are noticed for the range 2000 < H < 3000.
(3) The changing thermal conductivity affects the distributions of stress curves as
well as displacement curves, while the temperature field is observed to be less
effective with the alterations in the values of the thermal conductivity parameter
stress and conductive temperature significantly. Influences of changing thermal
conductivity parameters are noticed to be maximum at the jumps of the field
quantities.
(4) Stress curves are affected by the discontinuities, while conductive temperature
and displacement fields are observed to be smooth in nature.
The author strongly believes that the present work may help the scientific commu-
nity in designing the structures and for the study of waves inside the various materials
influenced by the heat source.
References
1. Lord, H.W., Shulman, Y.: A generalized dynamical theory of thermoelasticity. J. Mech. Phys.
Solids 15, 299–309 (1967)
2. Green, A.E., Lindsay, K.A.: Thermoelasticity. J. Elast. 2, 1–7 (1972)
3. Green, A.E., Naghdi, P.M.: A Re-Examination of the basic postulates of thermomechanics.
Proc. R. Soc. A: Math. Phys. Eng. Sci. 432, 171–194 (1991)
4. Green, A.E., Naghdi, P.M.: On undamped heat waves in an elastic solid. J. Therm. Stresses 15,
253–264 (1992)
5. Green, A.E., Naghdi, P.M.: Thermoelasticity without energy dissipation. J. Elast. 31, 189–208
(1993)
6. Tzou, D.Y.: A unified field approach for heat conduction from macro- to micro-scales. J. Heat
Trans. 117, 9–16 (1995)
7. Chandrasekharaiah, D.S.: Hyperbolic thermoelasticity: a review of recent literature. Appl.
Mech. Rev. 51, 705–529 (1998)
8. Ho, J.R., Kuo, C.P., Jiaung, W.S.: Study of heat transfer in multilayered structure within the
framework of dual-phase-lag heat conduction model using lattice Boltzmann method. Int. J.
Heat Mass Transf. 48, 55–69 (2003)
Effects of Magnetic Field and Thermal Conductivity Variance … 167
9. Kumar, R., Tiwari, R., Singhal, A., Mondal, S.: Characterization of thermal damage of
skin tissue subjected to moving heat source in the purview of dual phase lag theory with
memory-dependent derivative. Waves Random Compl. Media (2021). https://doi.org/10.1080/
17455030.2021.1979273
10. Youssef, H.M.: State-Space approach on generalized thermoelasticity for an infinite material
with a spherical cavity and variable thermal conductivity subjected to ramp-type heating. Can.
Appl. Math. Q. 13, 369–390 (2005)
11. Youssef, H.M.: Three-dimensional generalized thermoelasticity with variable thermal conduc-
tivity Int. J. Comput. Mater. Sci. Surf. Eng. 9(1) (2010)
12. Li, C., Guo, H., Tian, X., et al.: Transient response for a half-space with variable thermal
conductivity and diffusivity under thermal and chemical shock. J. Therm. Stresses 40, 389–401
(2017)
13. Li, X., Xue, Z., Tian, X.: A modified fractional order generalized bio-thermoelastic theory with
temperature-dependent thermal material properties. Int. J. Therm. Sci. 132, 249–256 (2018)
14. Li, X., Li, C., Xue, Z., Tian, X.: Analytical study of transient thermo-mechanical responses
of dual-layer skin tissue with variable thermal material properties. Int. J. Therm. Sci. 124,
459–466 (2018)
15. Tiwari, R., Kumar, R.: Investigation of thermal excitation induced by laser pulses and thermal
shock in the half space medium with variable thermal conductivity. Waves Random Compl.
Media (2020). https://doi.org/10.1080/17455030.2020.1851067
16. Sherief, H., Abd El-Latief, A.M.: Effect of variable thermal conductivity on a half-space under
the fractional order theory of thermoelasticity. Int. J. Mech. Sci. 74, 185–189 (2013)
17. Wang, Y., Liu, D., Wang, Q., et al.: Thermoelastic response of thin plate with variable material
properties under transient thermal shock. Int. J. Mech. Sci. 104, 200–206 (2015)
18. Tiwari, R., Abouelregal, A.E.: Memory response on magneto-thermoelastic vibrations on a
viscoelastic micro-beam exposed to a laser pulse heat source. Appl. Math. Model. 99, 328–245
(2021)
19. Abd-Elaziz, E.M., Othman, M.I.A:. Effect of Thomson and thermal loading due to laser pulse
in a magneto-thermo-elastic porous medium with energy dissipation. J. Appl. Math. Mech.
(2019)
Differential and Integral Equations
On Unique Positive Solution of
Hadamard Fractional Differential
Equation Involving p-Laplacian
Abstract In this paper, the authors have studied p-Laplacian Hadamard fractional
differential equation with integral boundary condition. The sufficient condition for
the existence and uniqueness of solution is developed using a new fixed point theorem
(Zhai and Wang [21]) of ϕ − (h, e)-concave operator. Further, an iterative method is
also given for approximating the solution corresponding to any arbitrary initial value
taken from an appropriate set.
1 Introduction
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 171
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_13
172 R. K. Vats et al.
tional calculus (Garra et al. [6]). For further insight of fractional derivatives, one can
go through (Diethelm [5], Kilbas et al. [9], Nain et al. [12, 13], Podlubny [16]) and
references therein.
Differential equations of fractional order with boundary conditions involving inte-
gral term have received much consideration of the researchers due to the applica-
tions in problems of thermal conduction, semiconductors, and hydrodynamic issues
which all include integral boundary conditions (Henderson and Luca [8]). Also,
equations consisting of p-Laplacian operator have been derived from the studies of
non-Newtonian fluid theory and nonlinear elastic mechanics. In 1945, Leibenson [10]
described the turbulent flow model in porous media using the following p-Laplacian
equation:
(φ p (ψ (x))) = Φ(x, ψ(x), ψ (x)),
Diaz and Tehlin [4] established a series of models involving the p-Laplacian opera-
tor, which emerges in the study of incompressible turbulent fluids flowing through
porous surfaces and gases moving through pipes with uniform cross-sectional areas.
Using the properties of green function and utilizing nonlinear alternative of Leray–
Schauder, Li and Lin [11] established the existence of positive solution for the
Hadamard fractional boundary value problem (FBVP):
β
H
D1 (φ p ( H D1α ψ(x))) = Φ(x, ψ(x)), 1 < x < e,
ψ(1) = ψ (1) = ψ (e) = 0, H
D1α ψ(1) = H D1α ψ(e) = 0,
where 2 < α ≤ 3, 1 < β ≤ 2 and H D1− is the Hadamard fractional derivative. Fol-
lowing the work of Li and Lin [11], Wang and Wang [18] derived the sufficient con-
dition for the existence of solution using Schaefer’s fixed point theorem for FBVP
with p-Laplacian operator with strip conditions. Recently, Wang and Zhai [19] stud-
ied the existence and uniqueness of solution for FBVP in partially ordered cone and
established an iterative method for obtaining the solution. Taking into account their
importance in analysis and applications, much more attention have been drawn to
the analysis of p-Laplacian FBVP. For more information on the applications of p-
Laplacian operator in differential equations, one can go through (Benedikt et al. [1],
Cheng and Wang [3], Wang et al. [17], Xue et al. [20]), and references therein. Also,
further development of some new fixed point theorems like α-type F-contractive
mappings, α-type F-contractions, etc. can be found in Gopal et al. [2, 7, 15] and
references therein.
Supposedly, there are not many articles that study the existence and uniqueness of
positive solution for p-Laplacian FBVP. Therefore, followed by the work discussed
above, the authors have considered the Hadamard p-Laplacian FBVP involving inte-
gral boundary conditions as follows:
On Unique Positive Solution of Hadamard Fractional Differential … 173
⎧ H d2
⎪ D1 (φ p ( H D1d1 ψ(x) − χ(x))) + Φ(x, ψ(x)) = 0, 1 < x < e,
⎨ e
ψ(1) = δψ(1) = 0, ψ(e) = 1 ρ(x)ψ(x) dx , (1)
⎪
⎩
x
D1 ψ(1) = 0,
H d1
provided the integral exists, where ln(·) = loge (·) (see Kilbas et al. [9]).
2 Preliminaries
This section presents the definitions, notations, and lemmas which supports the results
presented in Sect. 3. A subset C(= ∅) of a real Banach space (U, · ), which is closed
and convex is said to be a cone if it gratifies the following conditions:
(i) ψ ∈ C, ξ ≥ 0 implies ξψ ∈ C,
(ii) ψ ∈ C, −ψ ∈ C implies ψ = θ, where θ is zero element of U.
ψ ≤ ω ⇔ ω − ψ ∈ C.
A cone C ⊂ U is said to be normal if there exists a constant M > 0 such that for
all ψ, ω ∈ U, θ ≤ ψ ≤ ω implies ψ ≤ M ω . For ψ, ω ∈ U, the notation ψ ∼ ω
means that there exist μ, ν > 0 such that μψ ≤ ω ≤ νψ. For θ < η ∈ U, denote
Cη = {ψ ∈ U | ψ ∼ η}, with Cη ⊂ C. Let e ∈ C with θ ≤ e ≤ η, denote a set
Cη,e = {ψ ∈ U | ψ + e ∈ Cη }.
Then one can see that η ∈ Cη,e and for any ψ ∈ U there exists μ = μ(η, e, ψ) >
0 and ν = ν(η, e, ψ) > 0 such that μη ≤ ψ + e ≤ νη. An operator Υ : U → U is
increasing if ψ ≤ ω implies Υ (ψ) ≤ Υ (ω).
174 R. K. Vats et al.
Definition 1 (Zhai and Wang [21]) Let Υ : Cη,e → U be a given operator. For any
ψ ∈ Pη,e and ξ ∈ (0, 1), there exists ϕ(ξ) > ξ such that
n
D1 ψ(x)
H d1 H d1
I1 = ψ(x) + c j (ln x)d1 − j . (3)
j=1
Lemma 2 (Zhai and Wang [21]) Let C be normal and Υ be an increasing ϕ − (η, e)-
concave operator, Υ (η) ∈ Cη,e , then Υ has a unique fixed point ψ ∗ in Cη,e . Moreover,
for any ψ0 ∈ Cη,e , making the sequence ψn = Υ (ψn−1 ), n = 1, 2, . . . , then ψn −
ψ ∗ → 0 as n → ∞.
Lemma 3 (Nain et al. [14]) Let ϑ ∈ C[1, e]. Then the Hadamard FBVP
H
D1d1 ψ(x)) + ϑ(x) = 0,
e ρ(x)ψ(x) (4)
ψ(1) = δψ(1) = 0, ψ(e) = 1 x
dx
where
H (x, w) = H1 (x, w) + H2 (x, w), (6)
(ln x)d1 −1 e
dx
H2 (x, w) = H1 (x, w) p(x) , (8)
1−a 1 x
e
dx
a= (ln x)d1 −1 p(x) > 1. (9)
1 x
On Unique Positive Solution of Hadamard Fractional Differential … 175
Lemma 4 (Nain et al. [14]) The function H1 (x, w) is a continuous function along
with H1 (x, w) > 0 and
(ln x)d1 −1 (1 − ln x)(1 − ln w)d1 −1 ln w ≤ Γ (d1 )H1 (x, w) ≤ (d1 − 1)(1 − ln w)d1 −1 ln w.
Lemma 5 If χ ∈ C([1, e], R) with χ(1) = 0. Then (11) has the unique solution
given by
e e
dw dw
ψ(x) = H (x, w)φq ( H I1d2 ϑ(w)) − H (x, w)χ(w) . (12)
1 w 1 w
Proof Let Θ = H
D1d1 ψ, ν = φ p (Θ − χ). Then the initial value problem
H
D1d1 ψ(x) = φq (− H I1d2 ϑ(x))
e+ χ(x) (14)
ψ(1) = δψ(1) = 0, ψ(e) = 1 ρ(x)ψ(x)x
dx.
176 R. K. Vats et al.
Proof In order to apply Lemma 2, construct the set Cη,e with respect to function
η and e. First, we show that 0 ≤ e(x) ≤ η(x), x ∈ J. From, Lemma 4, H (x, w) ≥
0 and χ(x) ≥ 0, thus for x ∈ J,
e
dw
e(x) = H (x, w)χ(w) ≥ 0, (16)
1 w
d1 −1
≤ K (ln x) = η(x).
Therefore, concluding from (16) and (17), 0 ≤ e(x) ≤ η(x) and the set Cη,e = {ψ ∈
U, ψ + e ∈ Cη } is well defined.
The FBVP (1) has the following integral representation, which is due to Lemma 5
is as follows:
(q−1) e w d2 −1
1 w dt dw
ψ(x) = H (x, w)φq ln Φ(t, ψ(t))
Γ (d2 ) 1 1 t t w
e
dw
− H (x, w)χ(w)
1 w
q−1 e w d2 −1
1 w dt dw
= H (x, w)φq ln Φ(t, ψ(t)) − e(x),
Γ (d2 ) 1 1 t t w
where H (x, w) is given in (6). Using the above integral representation of solution,
set an operator Υ : Cη,e → U which is defined as follows:
(q−1) e w
1 w d2 −1 dt dw
(Υ ψ)(x) = H (x, w)φq ln Φ(t, ψ(t)) − e(x).
Γ (d2 ) 1 1 t t w
(18)
Obviously, ψ(x) is the solution of the problem (1) if and only if ψ(x) = (Υ ψ)(x),
i.e., Υ has a fixed point. The operator φq is monotone increasing in [0, +∞) and
suppose ψ ∈ Cη,e , ξ ∈ (0, 1)
Υ (ξψ + (ξ − 1)e)(x) =
(q−1) e w
1 w d2 −1 dt dw
H (x, w)φq ln Φ(t, ξψ(t) + (1 − ξ)e(t)) − e(x)
Γ (d2 ) 1 1 t t w
(q−1) e w d2 −1
1 w dt dw
≥ (υ(ξ))(q−1) H (x, w)φq ln Φ(t, ψ(t)) − e(x)
Γ (d2 ) 1 1 t t w
178 R. K. Vats et al.
(q−1) e w
1 w d2 −1 dt dw
= (υ(ξ))(q−1) H (x, w)φq ln Φ(t, ψ(t)) − e(x)
Γ (d2 ) 1 1 t t w
+ (υ(ξ))(q−1) e(x) − e(x)
= (υ(ξ))(q−1) Υ ψ(x) + [(υ(ξ))(q−1) − 1]e(x).
υ(ξ)
Set ϕ(ξ) = (υ(ξ))(q−1) for ξ ∈ (0, 1). From (H2), lnln(ξ) > (q−1)
1
for ξ ∈ (0, 1),
which implies
ln ϕ(ξ) = (q − 1) · ln(υ(ξ))
υ(ξ)
= (q − 1) · lnln(ξ) · ln(ξ)
> (q − 1) · (q−1) · ln(ξ) = ln(ξ),
1
which gives ϕ(ξ) > ξ, for ξ ∈ (0, 1).Hence, for ψ ∈ Cη,e , ξ ∈ (0, 1), we obtain
(Υ η)(x) + e(x)
(q−1) e w
1 w d2 −1 dt dw
= H (x, w)φq ln Φ(t, η(t))
Γ (d2 ) 1 1 t t w
(q−1) e w
1 w d2 −1 d1 −1 dt dw
= H (x, w)φq ln Φ(t, K (ln t) )
Γ (d2 ) 1 1 t t w
(q−1) e −1 −1 w d2 −1
1 (ln x) d1 (1 − ln w) d1 w dt dw
≤ φq ln Φ(t, K )
Γ (d2 ) 1 Γ (d1 )(1 − a) 1 t t w
(q−1) e w d2 −1
1 1 w dt dw
= (1 − ln w)d1 −1 φq ln Φ(t, K ) · (ln x)d1 −1
Γ (d2 ) Γ (d1 )(1 − a) 1 1 t t w
(q−1) w
1 1 e w d2 −1 dt dw
= (1 − ln w)d1 −1 φq ln Φ(t, K ) · η(x),
Γ (d2 ) K Γ (d1 )(1 − a) 1 1 t t w
and
(Υ η)(x) + e(x)
(q−1) e w
1 w d2 −1 dt dw
= H (x, w)φq ln Φ(t, K (ln t)d1 −1 )
Γ (d2 ) 1 1 t t w
(q−1) e w
1 (ln x)d1 −1 (ln w)(1 − ln w)d1 −1 a w d2 −1 dt dw
≥ φq ln Φ(t, 0)
Γ (d2 ) 1 Γ (d1 )(1 − a) 1 t t w
(q−1) w d2 −1
1 a e w dt dw
= (ln w)(1 − ln w)d1 −1 φq ln Φ(t, 0) · (ln x)d1 −1
Γ (d2 ) Γ (d1 )(1 − a) 1 1 t t w
(q−1) w
1 a e w d2 −1 dt dw
= (ln w)(1 − ln w)d1 −1 φq ln Φ(t, 0) · η(x).
Γ (d2 ) K Γ (d1 )(1 − a) 1 1 t t w
On Unique Positive Solution of Hadamard Fractional Differential … 179
Fix
(q−1) w
1 a e w d2 −1 dt dw
μ= (ln w)(1 − ln w)d1 −1 φq ln Φ(t, 0) ,
Γ (d2 ) K Γ (d1 )(1 − a) 1 1 t t w
(q−1)
1 1 e w w d2 −1 dt dw
ν= (1 − ln w)d1 −1 φq ln Φ(t, K ) ,
Γ (d2 ) K Γ (d1 )(1 − a) 1 1 t t w
and
(q−1) w
1 1 e w d2 −1 dt dw
ν= (1 − ln w)d1 −1 φq ln Φ(t, K )
Γ (d2 ) K Γ (d1 )(1 − a) 1 1 t t w
(q−1) e w d2 −1
1 a w dt dw
≥ (1 − ln w)d1 −1 φq ln Φ(t, 0)
Γ (d2 ) K Γ (d1 )(1 − a) 1 1 t t w
(q−1) −1
1 a e w w d2 dt dw
≥ (1 − ln w)d1 −1 φq ln Φ(t, 0)
Γ (d2 ) K Γ (d1 )(1 − a) 1 1 t t w
(q−1) w d2 −1
1 a e w dt dw
≥ (ln w)(1 − ln w)d1 −1 φq ln Φ(t, 0)
Γ (d2 ) K Γ (d1 )(1 − a) 1 1 t t w
=μ > 0.
Thus, there are two functions μ and ν such that μη ≤ Υ (η) + e ≤ νη which implies
Υ (η) + e ∈ Cη .
Therefore, Υ satisfies all the conditions of Lemma 2 and thus Υ has a unique fixed
point ψ ∗ ∈ Cη,e and thus
(q−1) e w
1 w d2 −1 dt dw
ψ ∗ (x) = H (x, w)φq ln Φ(t, ψ ∗ (t)) − e(x), x ∈ [1, e],
Γ (d2 ) 1 1 t t w
i.e., ψ ∗ (x) is the solution of the problem (1). Moreover, for any ψ0 ∈ Cη,e , the
sequence ψn = Υ ψn−1 , n = 1, 2, . . . , satisfies ψn → ψ ∗ as n → ∞.Namely,
(q−1) e w d2 −1
1 w dt dw
ψn (x) = H (x, w)φq ln Φ(t, ψn−1 (t))
Γ (d2 ) 1 1 t t w
e
dw
− H (x, w)χ(w) , n = 1, 2, . . .
1 w
Remark 1 For some FBVP, we can derive functions e(x), η(x) and construct func-
tions which satisfies the postulates of Theorem 1. For example, suppose Φ(x, ψ) =
13
e(x)
e∗
ψ + e(x) , where θ ≤ e(x) ≤ η(x). Then Φ is increasing w.r.t. second variable
and Φ ≡ 0. Set ϕ(ξ) = ξ 3 , then for ξ ∈ (0, 1), μ ∈ (−∞, ∞), ν ∈ [0, e∗ ],
1
180 R. K. Vats et al.
13
e(x)
Φ(x, ξμ + (ξ − 1)ν) = [ξμ + (ξ − 1)ν] + e(x)
e∗
1
1 e(x) 1 e(x) 3
=ξ 3 μ+ 1− ν +
e∗ ξ ξ
1
1 e(x) 1 e(x) e(x) 3
=ξ 3 μ+ 1− ν+
e∗ ξ e∗ ξ
13
1 e(x)
≥ξ 3 μ + e(x) = ϕ(ξ)Φ(x, μ).
e∗
4 Conclusion
The existence and uniqueness of solution together with an iterative method has been
established for a Hadamard FBVP. The result has been established using a fixed point
theorem for ϕ − (η, e)-concave operator in a partially ordered cone. The method for
constructing the solution presented in Theorem 1 starts with a simple function which
is useful for computational purpose.
References
1. Benedikt, J., Girg, P., Kotrla, L., Takac, P.: Origin of the p-laplacian and a. missbach. Electron.
J. Differ. Equ. 2018(16), 1–17 (2018)
2. Budhia, L., Hassen A., Arslan, H.A., Gopal, D.: Some new fixed point results in rectangular
metric spaces with an application to fractional-order functional differential equations. Nonlin-
ear Anal.-Model. Control 25(4), 580–597 (2020)
3. Cheng, Y.H., Wang, W.C.: Nodal properties for p-laplacian systems. Electr. J. Diff. Equ.
2017(87), 1–8 (2017)
4. Diaz, J.I., De Thelin, F.: On a nonlinear parabolic problem arising in some models related to
turbulent flows. SIAM J. Math. Anal. 25(4), 1085–1111 (1994)
5. Diethelm, K.: The Analysis of Fractional Differential Equations. Springer Science & Business
Media, Berlin (2010)
6. Garra, R., Mainardi, F., Spada, G.: A generalization of the lomnitz logarithmic creep law via
hadamard fractional calculus. Chaos, Solitons Fractals 102, 333–338 (2017)
7. Gopal, D., Mujahid, A., Patel, D.K., Vetro, C.: Fixed points of α-type F-contractive mappings
with an application to nonlinear fractional differential equation. Acta Math. Sci. 36(3), 957–970
(2016)
8. Henderson, J., Luca, R.: Boundary Value Problems for Systems of Differential, Difference and
Fractional Equations: Positive Solutions. Academic, Cambridge (2015)
9. Kilbas, A.A., Srivastava, H.M., Trujillo, J.J.: Theory and applications of fractional differential
equations, vol. 204. Elsevier, Amsterdam (2006)
10. Leibenson, L.: General problem of the movement of a compressible fluid in a porous medium.
Izv. Akad. Nauk Kirg. SSSR 9(1), 7–10 (1945)
11. Li, Y.l., Lin, S.y.: Positive solution for the nonlinear hadamard type fractional differential
equation with p-laplacian. J. Funct. Spaces 2013(951643), 1–10, (2013)
On Unique Positive Solution of Hadamard Fractional Differential … 181
12. Nain, A., Vats, R., Kumar, A.: Coupled fractional differential equations involving caputo-
hadamard derivative with nonlocal boundary conditions. Math. Methods Appl. Sci. 44(5),
4192–4204 (2021)
13. Nain, A.K., Vats, R.K., Kumar, A.: Caputo-hadamard fractional differential equation with
impulsive boundary conditions. J. Math. Model. 9(1), 93–106 (2021)
14. Nain, A.K., Vats, R.K., Verma, S.K.: Existence and uniqueness results for positive solutions of
hadamard type fractional BVP. J. Interdiscip. Math. 22(5), 697–710 (2019)
15. Padcharoen, A., Gopal, D., Chaipunya, P., Kumam, P.: Fixed point and periodic point results
for α-type F-contractions in modular metric spaces. J. Fixed Point Theory Appl. 2016(1), 1–12
(2016)
16. Podlubny, I.: Fractional Differential Equations: An Introduction to Fractional Derivatives, Frac-
tional Differential Equations, to Methods of Their Solution and Some of Their Applications.
Academic, Cambridge (1999)
17. Wang, G., Ren, X., Zhang, L., Ahmad, B.: Explicit iteration and unique positive solution for a
caputo-hadamard fractional turbulent flow model. IEEE Access 7, 109833–109839 (2019)
18. Wang, G., Wang, T.: On a nonlinear hadamard type fractional differential equation with p-
laplacian operator and strip condition. J. Nonlinear Sci. Appl. 9(7), 5073–5081 (2016)
19. Wang, L., Zhai, C.: Unique solutions for new fractional differential equations with p-laplacian
and infinite-point boundary conditions. Int. J. Dyn. Syst. Differ. Equ. 9(1), 1–13 (2019)
20. Xue, T., Liu, W., Shen, T.: Extremal solutions for p-laplacian boundary value problems with
the right-handed riemann-liouville fractional derivative. Math. Methods Appl. Sci. 42(12),
4394–4407 (2019)
21. Zhai, C., Wang, L.: ϕ-(h, e)-concave operators and applications. J. Math. Anal. Appl. 454(2),
571–584 (2017)
22. Zhou, Y., Wang, J., Zhang, L.: Basic Theory of Fractional Differential Equations. World Sci-
entific, Singapore (2016)
Eigenvalue Criteria for Existence
and Nonexistence of Positive Solutions
for α-Order Fractional Differential
Equations on the Half-Line (2 < α ≤ 3)
with Integral Condition
Abstract This article concerns the nonexistence and existence of positive solutions
to the fractional differential equations
⎧ α
⎨ D u(t) + f (t, u(t)) = 0 0 ≤ t < +∞
η
⎩ u(0) = D α−2 u(0) = 0 lim D α−1 u(t) = A u(s)dμ(s)
t→+∞ 0
where η, A ∈ (0, +∞), and 2 < α ≤ 3, μ(t) is the continuous nondecreasing func-
η
tion on (0, +∞), with μ(0) = 0 and u(s)dμ(s) denotes the Riemann–Stieljes
0
integrals of u with respect to μ; D α is standard Riemann–Liouville derivative,
f : R+ × R+ → R+ is a continuous function.
Since fractional differential equations are considered as alternative models for the
nonlinear differential equations, the study of the existence of positive solutions
to boundary value problems associated with fractional differential equations has
become a very important area of applied mathematics over the last few decades.
A. Benmezai
National Higher School of Mathematics, Algiers, Algeria
S. Chentout (B)
Faculty of Mathematics, USTHB, Algiers, Algeria
e-mail: [email protected]
W. Esserhan
ENSSEA, Pole Universitaire KOLEA, 42003 Tipaza, Algeria
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 183
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_14
184 A. Benmezai et al.
Such a subject has been discussed in many recent papers; see for examples,
[5, 6, 13], and references therein. We are concerned in this paper with the nonexis-
tence and existence of positive solutions of the fractional boundary value problem
(fbvp):
D α u(t) + f (t, u(t)) = 0, 0 ≤ t < +∞ η (1)
u(o) = D α−2 u(o) = 0, limt→+∞ D α−1 u(t) = A 0 u(s)dμ(s)
The existence result for positive solutions of the fbvp (1) needs the introduction
of the following additional notations. Set for a Dα -Caratheodory function h : R+ ×
R+ → R+ , q ∈ Dα , and ν = 0, +∞
h(t, (1 + t)α−1 u)
h+
ν (q) = lim sup max ,
u→v t≥0 (1 + t)α−1 q(t)u
h(t, (1 + t)α−1 u)
h−
ν (q) = lim inf min ,
u→v t≥0 (1 + t)α−1 q(t)u
f 0+ (q0 ) f − (q0 )
< 1 < +∞ . (6)
ρα (q0 ) ρα (q∞ )
2 Abstract Background
Let X be a real Banach space, the standard notations L(X ) and r (L) refer, respec-
tively, to the set of all linear bounded self-mapping defined on X .
Definition 1 A nonempty closed convex subset K of B is called a cone if
1. tx ∈ K for all x∈ K and t≥ 0.
2. K ∩ (−K ) = {0 B }
Given a cone K ⊂ B,we defined a partial order on B, denoted by “≤”, as follows
for all x, y ∈ B, x ≤ y if y − x ∈ K and y = x and x y if y − x ∈
/ K . Notations
≥, >, and denote, respectively, reverse situations.
Definition 2 Let L be a compact operator. L is said to be
1. positive, if L(K ) ⊂ K ;
2. Strongly positive, if int (K ) = ∅ and L(K \ {0 X }) ⊂ int (K );
3. lower bounded on the cone K , if
In all what follows, L K (X ) denotes the subset of all positive compact operators
in L(X ) and we set for all L ∈ L K (X )
or
r (L) < 1 and T u Lu for all u ∈ K , (8)
L 1 u − F1 u T u L 2 u + F2 u.
If either
F1 u = ◦ ( u ) as u → ∞ and F2 u = ◦ ( u ) as u → (9)
or
F1 u = ◦ ( u ) as u → 0 and F2 u = ◦ ( u ) as u → ∞, (10)
then T has a positive fixed point. For a delaited presentation on fixed point theory
see [8] and [14].
Eigenvalue Criteria for Existence and Nonexistence of Positive … 187
3 Riemann–Liouville Derivative
Now, let us recall some basic facts related to the theory of fractional differential
equations. Let β be a positive real number, the Riemann–Liouville fractional integral
of order β of a function f : (0, +∞) → R is defined by
t
β 1
I 0+ f (t) = (t − s)β−1 f (s)ds, (11)
Γ (β) 0
where Γ (β) is the gamma function, provided that the right side is pointwise
defined on (0, +∞). For example, we have where n = [β] + 1, [β] denotes the
integer part of the number β, provided that the right side is pointwise defined on
β
(0, ∞). As a basic example, we quote for σ > −1, D0+ t σ = Γ Γ(σ−β+1)
(σ+1) σ−β
t . Thus,
β
if u ∈ C (0, 1) ∩ L 1 (0, 1), then the fractional differential equation D0+ u(t) = 0 has
i=[β]+1 β−i
u(t) = i=1 ci t , ci ∈ R, as unique solution and if u has a fractional derivative
of order β in C (0, 1) ∩ L 1 (0, 1), then
β β
i=[β]+1
I0+ D0+ u(t) = u(t) + ci t β−i , ci ∈ R. (12)
i=1
For a detailed presentation on fractional differential calculus, see [10] and [11].
Now, we introduce some spaces and operators needed for the proof of the main
results. we let E be the linear space defined by
u(t)
E = u ∈ C R+ , R , lim α−1 exists
t→+∞ t
|u(t)|
u E = sup α−1
t∈[0,+∞[ (1 + t)
where
1
γ1 (t) = min(λ, λt α−1 ) and λ = η 1
1
γ(η)
+ 1
(1−σ)Γ (α) 0 γ(t)
dμ(t) (1 + η)α−1
is a cone in E.
η
Lemma 1 If σ = A
Γ (α) 0 t α−1 dμ(t) ≺ 1 and q(t) ∈ L 1 ([0, +∞)), then the fbvp
⎧
⎪ α
⎨ D u(t) + q(t) = 0 0 ≤ t < +∞
u(0) = D α−2 u(0) = 0,
(13)
⎪
⎩ lim D α−1 u(t) = A 0η u(s)dμ(s).
t→+∞
and α−1
1 t − (t − s)α−1 t ≥s
G 1 (t, s) = α−1
Γ (α) t t ≤s
and t
1
D α−2 u(t) = − (t − s)q(s)ds + C1 Γ (α)t + C2 Γ (α − 1)
Γ (2) 0
Dα−2 u(0) = 0 =⇒ C2 = 0
Eigenvalue Criteria for Existence and Nonexistence of Positive … 189
t
1
u(t) = C1 t α−1 − (t − s)α−1 q(s)ds
Γ (α) 0
and t
D α−1 u(t) = − q(s)ds + C1 Γ (α)
0
+∞ η
Lim t→+∞ D α−1 u(t) = − q(s)ds + C1 Γ (α) = A u(s)dμ(s)
0 0
hence +∞ η
1 A
C1 = q(s)ds + u(s)dμ(s)
Γ (α) 0 Γ (α) 0
We obtain
+∞ η
t α−1 At α−1 1 t
u(t) = q(s)ds + u(s)dμ(s) − (t − s)α−1 q(s)ds
Γ (α) 0 Γ (α) 0 Γ (α) 0
t +∞
1 α−1 α−1 1
u(t) = t − (t − s) q(s)ds + t α−1 q(s)ds
Γ (α) 0 Γ (α) t
At α−1 η
+ u(s)dμ(s)
Γ (α) 0
+∞
At α−1 η
= G 1 (t, s)q(s)ds + u(s)dμ(s)
0 Γ (α) 0
Which means
η η η +∞
A
1− t α−1 dμ(t) u(t)dμ(t) = G 1 (t, s)q(s)ds dμ(t)
Γ (α) 0 0 0 0
190 A. Benmezai et al.
η +∞
η G 1 (t, s)q(s)ds dμ(t)
0 0
u(t)dμ(t) = η
0 1 − Γ (α)
A
t α−1 dμ(s)
0
so
η +∞
+∞ G 1 (t, s)q(s)ds dμ(t)
0 0
u(t) = G 1 (t, s)q(s)ds +
η α−1
0 1 − Γ (α)
A
0 t dμ(s)
+∞ α−1 η
At
= G 1 (t, s) + (G 1 (t, s)dμ(t)) q(s)ds
0 (1 − σ) Γ (α) 0
+∞
u(t) = G(t, s)q(s)ds
0
where
η
At α−1
G(t, s) = G 1 (t, s) + (G 1 (t, s)dμ(t))
(1 − σ) Γ (α) 0
η
σ= A
Γ (α) 0 t α−1 dμ(s) and
α−1
1 t − (t − s)α−1 t ≥s
G 1 (t, s) = α−1
Γ (α) t t ≤s
3.
(t,s)
limt→0 Gt1α−1 = Γ (α)
1
≺∞
(t,s) (16)
limt→+∞ Gt1α−1 0
4.
G 1 (τ , s)
G 1 (t, s) ≥ γ(t) , for all t, τ , s ≥ 0 with γ(t) = min(1, t α−1 )
(1 + τ )α−1
(17)
Eigenvalue Criteria for Existence and Nonexistence of Positive … 191
5.
∂G 1
(t, s) > 0 for all t, s ∈ [0, +∞[. (18)
∂t
1. α−1
Aη α−1 t
G(t, s) ≤ 1 + μ(η) (19)
1−σ Γ (α)
2.
G(τ , s)
G(t, s) ≥ γ1 (t) for t, s, τ ≥ 0 (20)
(1 + τ )α−1
1
where γ1 (t) = λγ(t), and λ = η ,
1
γ(η)
+ 1
1−σ 0
1
γ(t)
dμ(t)
3.
η
G(t, s) 1 G(t, s) A
lim = ≺ ∞, lim = (G 1 (t, s)dμ(t)) > 0
t→0 t α−1 Γ (α) t→+∞ t α−1 (1 − σ) Γ (α) 0
(21)
∂G
(t, s) 0 for all t, s ∈ [0, +∞[ (22)
∂t
Proof Properties (19), (21), and (22) are easy to check. Let us prove property (20)
t α−1
η
G(t, s) G 1 (t, s) + (1−σ)Γ (α) 0 (G 1
(t, s)dμ(t))
= η (23)
G(τ , s) τ
G 1 (τ , s) + (1−σ)Γ
α−1
(α) 0 (G 1
(τ , s)dμ(τ ))
G 1 (η,s) t α−1
η G 1 (η,s)
G(t, s) γ(t) (1+η) α−1 + 1−σ 0 γ(k) (1+η) α−1 dμ(k) γ(t)
G(r, s)
≥
(1+τ )α−1 G 1 (η,s) τ α−1
η (1+η)α−1 G 1 (η,s) ≥λ
(1 + τ )α−1
(25)
γ(η)
+ 1−σ 0 γ(k)
dμ(k)
192 A. Benmezai et al.
with
1
λ= η 1
1
γ(η)
+ A
(1−σ)Γ (α) 0 γ(t)
dμ(t) (1 + η)α−1
1. N is bounded in E,
2. the function belonging to u : u(t)= (1+t)α−1 , x ∈ N
x(t)
are locally equicontinuous
on [0, +∞[, that is, equicontinuous
on every compact interval
of R+ and
3. the functions belonging to u : u(t) = (1+t) α−1 , x ∈ N
x(t)
are equiconvergent at
+∞, that is, given ε 0, there corresponds T (ε) 0 such that |x(t) − x(+∞)|
≺ ε, for any t T (ε) and x ∈ N .
T u(t2 ) T u(t1 )
α−1
−
(1 + t2 ) (1 + t1 )α−1
+∞
G(t2 , s) G(t1 , s)
= − f (s, u(s))ds
0 (1 + t2 )α−1 (1 + t1 )α−1
+∞
G(t2 , s) − G(t1 , s) 1 1
= + − G(t 1 , s) f (s, u(s))ds
0 (1 + t2 )α−1 (1 + t2 )α−1 (1 + t1 )α−1
1 1
− ≤ (α − 1) |t1 − t2 | (1 + c)α−2
(1 + t2 )α−1 (1 + t1 )α−1
T u(t2 ) T u(t1 )
α−1
− ≤
(1 + t2 ) (1 + t1 )α−1
+∞ |G(t2 , s) − G(t1 , s)| (1 + t1 )α−1 _ (1 + t2 )α−1 u(s)
+ G(t1 , s) f (s, (1 + s)α−1 ) )ds
0 (1 + t2 )α−1 (1 + t1 )α−1 (1 + t2 )α−1 (1 + s)α−1
+∞ |G(t2 , s) − G(t1 , s)| (α − 1) |t1 − t2 | (1 + c)α−2 u(s)
≤ + G(t1 , s) (1 + s)α−1 Ψ ( )ds
0 (1 + t2 )α−1 (1 + t1 )α−1 (1 + t2 )α−1 (1 + s)α−1
+∞ |G(t2 , s) − G(t1 , s)| (α − 1) |t1 − t2 | (1 + c)α−2
≤ + G(t1 , s) (1 + s)α−1 Ψ R (s)ds
0 (1 + t2 )α−1 (1 + t1 )α−1 (1 + t2 )α−1
since for s ∈ [0, +∞[, the function t → G(t, s) is continuous on the compact interval
[0, θ] then it is uniformly continuous on [0, θ] ,
Hence
so
T u(t2 ) T u(t1 )
α−1
− →0
(1 + t2 ) (1 + t1 )α−1
uniformly as |t1 − t2 | → 0
and we have
+∞
|T u| G(t, s) u(s)
≤ f (s, (1 + s)α−1 ds
(1 + t)α−1 0 (1 + t)α−1
(1 + s)α−1
+∞
G(t, s) u(s)
≤ α−1
(1 + s)α−1 Ψ ( )ds
0 (1 + t) (1 + s)α−1
+∞
G(t, s)
≤ (1 + s)α−1 Ψ R (s)ds ≺ ∞
0 (1 + t)α−1
the property (21) of the function G and the dominated convergence theorem lead to
T u(t) T u(t)
− lim t→+∞ → 0
(1 + t)α−1 (1 + t)α−1
and we have
leading to
+∞
G(τ , s)
T u(t) ≥ γ1 (t) sup f (s, u(s)ds ≥ γ1 (t) T u
τ ≥0 0 (1 + τ )α−1
The main result of this section is the proof need to introduce additional notations,
with a function q in Dα and T 0 are associated the linear operators L q in £(E),
L qF , L q,T
F F
in £(E) and K q,T in£(FT ) defined by
+∞
L q u(t) = G(t, s)q(s)u(s)ds for all u ∈ E
0
L qF u = L q u(t) for all u ∈ F
T
F
L q,T u(t) = G T (t, s)q(s)u(s)ds for all u ∈ F
0
T
F
K q,T u(t) = G T (t, s)q(s)u(s)ds for all u ∈ FT
0
|u(t)|
u FT = sup α−1
, for all u ∈ FT
t∈[0,T ] t
u(t)
u FT1 = u FT + sup , for all u ∈ FT1
t∈[0,T ] t α−1
Set for T 0
u(t)
ST = u ∈ FT : u(t) > 0, for all t ∈ [0, T ] and lim α−1 > 0
t→0 t
Lemma 6 See [2]. The set ST is open in the Banach space FT and S⊂ FT+ .
Lemma 7 For all functions q in Dα and all T > 0, the operator L q,T
F
has the SIJP
at r (L q,T ).
F
where
F
K q,T u(t) = K q,T
F
u(t), for all u ∈ FT1 and for t ∈ [0, T ] ,
Eigenvalue Criteria for Existence and Nonexistence of Positive … 197
F
is well defined and K q,T ∈ L(FT , FT1 ).
Since the embedding j of FT1 in FT is compact and K q,T
F
= jo K q,T
F
. We have
F
that K q,T is compact.
Now, since for all u ∈ FT+ with u = 0
T
F
K q,T u(t) = G T (t, s)q(s)u(s)ds > 0 for all t ∈ [0, T ]
0
we have
F
K q,T (FT+ \ {o}) ⊂ ST ⊂ FT+
F
and the operator K q,T is strongly positive. This ends the proof.
Theorem 3 For all functions q in Dα , the operator L q has the SIJP at r (L q ) and
is lower bounded on the cone P.
Proof First let us prove that L qF has the SIJP at r (L qF ). This will be obtained from
Theorem 2, whence we prove that L qF = lim T →+∞ L q,T F
in operator norm and T →
L q,T is increasing We have for all u ∈ F with u F = 1
F
L qF u(t) − L q,T
F u(t) +∞
G(t, s) G T (t, s) T
= q(s)u(s)ds − q(s)u(s)ds
t α−1 0 t α−1 t α−1 0
T ∞
G(t, s) − G T (t, s) G(t, s)
≤ α−1
q(s)u(s)ds + q(s)u(s)ds
0 t T t α−1
∞ +∞ η
G 1 (t, s) α−1 A
≤ q(s)s ds + (G 1 (t, s)dμ(t)) q(s)ds
T t α−1 (1 − σ) Γ (α) T 0
+∞ +∞ η
A
≤ q(s)s α−1 ds + (G 1 (t, s)dμ(t)) q(s)ds
T (1 − σ) Γ (α) T 0
L qF u(t) − L q,T
F
u(t) Aμ(η) +∞
≤ 1+ q(s)s α−1 ds, for all t ≤ T
t α−1 (1 − σ) Γ (α) T
∂
and since ∂t
G(t,s)
t α−1
< o, for s ∈ (0, t). We have in the case of t ≥ T,
198 A. Benmezai et al.
L qF u(t) − L q,T
F u(t) +∞
T G(t, s) T G T (t, s)
≤ q(s)s α−1 ds + q(s)s α−1 ds + q(s)s α−1 ds
t α−1 T 0 t α−1 0 t α−1
+∞ TT
G(t, s) G(T, s)
≤ q(s)s α−1 ds +
α−1
q(s)s α−1 ds + q(s)s α−1 ds
T 0 t 0 T α−1
+∞ T
Aμ(η) G(T, s)
≤ 1+ q(s)s α−1 ds + 2 q(s)s α−1 ds
(1 − σ) Γ (α) T 0 T α−1
L qF u(t) − L q,T
F u(t)
F F
L q − L q,T = sup sup
u F =1 t>1 t α−1
+∞ T
Aμ(η) G(T, s)
≤ 1+ q(s)s α−1 ds + 2 q(s)s α−1 ds → 0 as T → ∞
(1 − σ) Γ (α) T 0 T α−1
Then by means of the dominated convergence theorem, we conclude that
F
lim T →+∞ L qF − L q,T = 0 and L q,T
F
converge to L qF in operator norm.
For T1 < T2 and u ∈ F + , we have
T2 T1
F
L q,T2
u(t) − L q,T
F
1
u(t) = G T2 (t, s)q(s)u(s)ds − G T2 (t, s)q(s)u(s)ds
0 0
T1 T2
= G T2 (t, s) − G T1 (t, s) q(s)u(s)ds + G T2 (t, s)q(s)u(s)ds
0 T1
Because of
⎧
⎪
⎪ 0,if t ≤ T1
⎪
⎪ η α−1
⎨ A
(G 1 (t, s)dμ(t)) t α−1 − T1 , if T1 ≤ t ≤ T2 , ≥ 0
(1−σ)Γ (α) 0
G T2 (t, s) − G T1 (t, s) =
⎪
⎪ ϕT2 (s) − ϕ
T1 (s)+
⎪
⎪ η
⎩ A
(G 1 (t, s)dμ(t)) T2α−1 − T1
α−1
, if T2 ≤ t
(1−σ)Γ (α) 0
F
we have L q,T 2
≥ L q,T
F
1
.
At this stage, we are able to prove that L q has the SIJP at r (L q ). We have Λ L qF
⊂ Λ L q and Γ L qF ⊂ Γ L q . So, let us prove that Λ L qF = Λ L q and Γ L qF = Γ L q . To
this aim, let λ ≥ 0 and u∈ E + \ {0} be such that L q u ≥ λu. We have U = L q u ∈
F + \ {0} , L qF U = L q L q u ≥ λL q u = λU and λ ∈ Λ L qF .
This proves that Λ L qF = Λ L q . In a similar way, we also obtain that Γ L qF = Γ L q .
Thus, we have
r (L qF ) = sup(Λ L qF ) = sup(Λ L q ) = inf(Γ L qF ) = inf Γ L q
Eigenvalue Criteria for Existence and Nonexistence of Positive … 199
and the operator Lq has the SIJP at r (L qF ). Furthermore, the cone E + is total in E.
Claims that r (L qF ) is the unique positive eigenvalue of L q , we have r (L q ) = r (L qF )
and L q has the SIJP at r (L q ). It remains to show that the operator L q is lower bounded
on the cone P.
+∞ +∞
L q u(t) = G(t, s)q(s)u(s)ds ≥ G(t, s)q(s)γ1 (s)(1 + s)α−1 ds u E
0 0
leading to
+∞
L q u(t) G(t, s)
L q u = sup
α−1
≥ sup q(s)γ1 (s)(1 + s)α−1 ds u E
t≥0 (1 + t) t≥0 0 (1 + t)α−1
We have from Lemma 4 that ρ is a positive eigenvalue of the linear eigenvalue problem
(2) if and only if ρ−1 is a positive eigenvalue of the compact operator L q . Since
Proposition 3 claims that L q has the SIJP at r (L q ), we have from Propositions 3.14
and 3.15 in [1] that r (L q ) is the unique positive eigenvalue of L q .
Therefore, we have that ρ = r (L1 q ) is the unique positive eigenvalue of the linear
eigenvalue problem (2).
Assume that hypothesis (3) holds true, we have then from Proposition 3 the operator
L q has the SIJP at r (L q ) where
1
r (L q ) =
ρα (q)
thus hypothesis (7) holds and Propositions 4 claims that the operator T f has no fixed
point. At the end, we conclude by Lemma 5 that fbvp (1) has no positive solution.
200 A. Benmezai et al.
f (t, u) ≤ (ρα (q∞ ) − ε)q∞ (t)(1 + t)α−1 u + Ψ R (t)(1 + t)α−1 for all t,u ≥ 0 (29)
Also, we have from f 0− (q0 ) > ρα (q0 ) that for ε ∈ 0, f 0− (q0 ) − ρα (q∞ ) there
is r > 0 such that
f (t, u) ≥ (ρα (q0 ) + ε) q0 (t)(1 + t)α−1 u for all t ≥ 0 and u ∈ [0, r ] (30)
Thus, we have
where
f (t, u) = sup 0, (ρα (q∞ + ε)q0 (t)(1 + t)α−1 u − f (t, u) .
L q0 u − F0 u ≤ T f u ≤ L q∞ u + F∞ u, ∀u ∈ P,
where
+∞
F0 u(t) = ˜ u(s))ds
G(t, s) f (s,
0
+∞
F∞ u(t) = G(t, s)Ψ R (s)(1 + s)α−1 ds
0
ρα (q0 ) + ρα (q∞ ) +
r (L q0 ) = >1> = r (L q∞ )
ρα (q0 ) ρα (q∞)
6 Examples
6.1 Example 1
μ(t) = t η = 1 f (t, u) = u
3 e−t ≤ ue−t .
(1+t) 2
We find σ = 158√π < 1, q(t) = e−t .
Hence, all conditions of Proposition 1 hold, then the fbvp (30) admit unique eigen-
value μα (q) < 1 then fbvp (31) has not positive solution.
6.2 Example 2
where
|u|
f (t, u) = 3 e−t
(1+t) 2
√
f (t, (1 + t) 2 u(t)) = |u|e−t
3
Hence, all conditions of Theorem 1 hold, then fpvp (32) admit a positive solution.
References
1. Benmezai, A., Boucheneb, B., Henderson, J., Mecherouk, S.: The index jump property for
I-homogeneous positive maps and fixed point theorems in cones. J. Nonlinear Funct. Anal.
2017, Article ID 6 (2017)
2. Benmezai, A., Chentout, S.: Eigenvalue-criteria for existence and nonexistence of positive
solutions for α−order fractional differential equations(2<α<3), on the half-line.Differ. Equ.
Appl. 11(4), 463–480 (2019). https://doi.org/10.7153/dea-2019-11-22
202 A. Benmezai et al.
3. Benmezai, A., Graef, J.R., Kong, L.: Positive solutions for the abstract Hammerstein equations
and applications. Commune. Math. Anal. 16(1), 47–65 (2014)
4. Benmezai, A., Graef, J.R., Kong, L.: Positive solutions to a two point singular boundary value
problem. Differ. Equ. Appl. 3(3), 347–373 (2011). https://doi.org/10.7153/dea-03-22
5. Cabada, A., Wang, G.: positive solutions of nonlinear fractional differential equations with
integral boundary conditions. J. Math. Anal. Appl. 389, 403–411 (2012). https://doi.org/10.
1016/j.jmaa.2011.11.065
6. Caballero, J., Cabrera, I., Sadaranani, K.: Positive solutions of nonlinear fractional differential
equations with Integral boundary value conditions. Abstr. Appl. Anal. Article ID 303545, 11
(2012). https://doi.org/10.1155/2012/303545
7. Corduneanu, C.: Integral Equations and Stability of Feedback Systems. Academic, New York
(1973)
8. Deimling, K.: Nonlinear Functional Analysis. Springer, Berlin (1985)
9. Guo, D., Lakshmikantaham, V.: Nonlinear Problems in Abstract Cones. Academic, SanDiego
(1988)
10. Kilbas, A.A., Srivastava, H.M., Trujillo, J.J.: Theory and Applications of Fractional Differential
Equations, North-Holland Mathematics studies, 204. Elsevier science B. V, Amsterdam (2006)
11. Podlubny, I.: Fractional Differential Equations. Academic, San Diego (1999)
12. Webb, J.R.L.: A class of positive linear operators and applications to nonlinear boundary value
problems. Topolog. Methods Nonlinear Anal. 39, 221–242 (2012)
13. Xiao, J., Nieto, J.J., Luo, Z.: Multiple positive solutions of the singular boundary value problem
for second order impulsive differential equations on the half-line. Bound. Value Probl. (2010),
Article ID 281908, 13. https://doi.org/10.1155/2010/281908
14. Zeidler, E.: Nonlinear Functional Analysis and Its Applications, vol. I. Fixed Point Theorems.
Springer, New-York (1986)
A Collocation Method for Solving
Proportional Delay Riccati Differential
Equations of Fractional Order
1 Introduction
Over the last few years, fractional calculus has provided generalized methods to
describe the behaviors of several physical systems. Fractional order models have
attracted the attention of researchers because differential equations involving non-
integer derivatives demonstrate the dynamics of many systems more realistically.
In recent years, mathematicians have contributed a large literature on the analysis
and applications of fractional differential equations [9, 19]. Several authors proposed
different definitions of fractional derivatives. The Riemann–Liouville and Liouville–
Caputo define a fractional derivative which has importance in the field of fractional
calculus but has certain limitations due to singular local power kernel [19]. Caputo and
Fabrizio propose a new definition of fractional derivatives using exponential decay
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 203
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_15
204 B. Hussain and A. Afroz
kernel [7]. Atangana and Baleanu suggested another fractional derivative which is
based on the concept of Mittag–Leffler function [1]. Some authors suggested more
interesting definitions where time-dependent variable-order fractional derivative and
integral α(t) is involved [23].
Generally, researchers construct mathematical models which are governed by
differential equations in which the present state of the system depends only on the
current value of the dependent variable and/or its derivative. Sometimes, these models
show severe inconsistency with reality, especially in real-time modeling, economics
model, cell growth model, and analysis of stock marketing. In order to improve the
dynamics of such mathematical model researchers incorporated delay terms in the
governing differential equation which results in a delay differential equation. Delay
differential equations (DDEs) form a special class of differential equations in which,
the rate of solution, depends on the present as well as some previous values of the
dependent variable and/or their derivative. DDEs are widely used to model processes
in physical sciences, biosciences, engineering, electrodynamics, and economics. The
detailed analysis and applications of DDEs can be found in [5, 22, 26]. Depending
upon the nature of delay/lag τ , DDEs have various formats such as DDEs with
constant delay, DDEs with time-dependent delay τ (t), DDEs with state-dependent
delay τ (t, y), neutral DDEs, and proportional DDEs or pantograph equations. Herein,
the subject of our interest is the following fractional order proportional delay Riccati
differential equation (PDRDE):
tograph DDEs, Izadi and Srivastava [12] produced a numerical solution of Lane
Emeden pantograph using Bessels polynomials and collocation point.
Haar wavelet is compact support, box function which takes only three values
{0, ±1}. From the last two decades, it has been used to solve a wide variety of
differential equations. The detailed applications of Haar wavelet can be found in
monograph and references in [15]. In [28], Shah et al. proposed a numerical technique
using Haar wavelet for solving fractional differential equations. Recently, Akmal
and Arshad [20] solved neutral DDEs using Haar wavelet bases, Abdullah and Rafiq
[3] combined the backward Euler method and Haar wavelet collocation method to
obtain the approximate solution of the Chen–Lee–Liu equation. Motivated by the
above literature, we aim to apply Haar wavelet series method (HWSM) to solve
the proportional delay Riccati differential equation of fractional order with Caputo
derivative.
The present article is organized in the following outline: Definition of Haar wavelet
is given in Sect. 2. The method is described in Sect. 3. In Sect. 4, the method is applied
to some test problems to check its efficiency and correctness. Section 5 is devoted to
conclusion.
2 Preliminaries
ζ − 1 < α < ζ, ζ ∈ N.In particular, for 0 < α < 1, we have n = 1, and hence,
t
α 1
C Dc,t g(t) = (t − u)−α g (u)du. (6)
(1 − α) c
2.2 Wavelet
In 1982, Jean Morlet, a French geophysical engineer, first introduced the concept
of wavelets as a family of functions generated by shifting and stretching of a single
function known as the “mother wavelet”:ψ(t). When the stretch a and shift b varies
continuously, we get the family of continuous wavelet as
1 t −b
ψa,b (t) = √ ψ , a = 0, b ∈ R.
a a
−j
If we restrict a and b to discrete values as a = a0 , b = kb0 a − j , where a0 ≥ 1, b0 ≥
0 and j, k ∈ N. We have the following family of discrete wavelets as
1 j
ψ j,k (t) = ψ(a0 t − kb0 ),
−j
a0
where {ψ j,k (t)} j,k∈N forms a wavelet basis for L 2 (R) − space. In particular, the
choices a0 = 2 and b0 = 1 produced an orthonormal basis [17, 18].
∞
To construct the Haar wavelet system {hi (t)}i=1 on [μi , μ f ] two basic functions are
required, namely:
(a). The Haar scaling function (father wavelet):
where I[μi ,μ f ] (t) is characteristic/indicator function. Now for generating the Haar
wavelet series, let j be dilation and k be translation parameter.
Then i-th Haar Wavelet is defined as
⎧
⎪
⎨1 for t ∈ [ϑ1 (i), ϑ2 (i))
hi (t) = −1 for t ∈ [ϑ2 (i), ϑ3 (i)) (9)
⎪
⎩
0 other wise,
where
j j
ϑ1 (i) = μi + (μ f − μi )k/2 , ϑ2 (i) = μi + (μ f − μi )(k + 0.5)/2 , ϑ3 (i) = μi +
j j
(μ f − μi )(k + 1)/2 . The index i = 2 + k + 1, j = 0, 1, . . . , J where J is maxi-
j
mum level of wavelet and k = 0, 1, . . . , 2 − 1.
(c). Define scaling function space and wavelet space as follows:
2 −1 j
V j = span{2 j/2 h1 (2 j t − k), t ∈ [μi , μ f ]}k=0 , (10)
2 j −1
W j = span{2 j/2
h2 (2 t − k), t ∈
j
[μi , μ f ]}k=0 . (11)
V J = V J0 ⊗ W J0 ⊗ W J0 +1 · · · ⊗ W J −1 . (12)
2 J +1
f appr ox (t) = ai hi (t). (13)
i=1
φ1 (t) = (t − ϑ1 (i))α ,
φ2 (t) = [(t − ϑ1 (i))α − 2(t − ϑ2 (i))α ],
φ3 (t) = [(t − ϑ1 (i))α − 2(t − ϑ2 (i))α + (t − ϑ3 (i))α ],
where ϑ1 (i), ϑ2 (i), ϑ3 (i) are same as defined in Eq. (9). Haar wavelet is considered
an efficient tool in numerical analysis, image processing, signal processing, and has
numerous other applications in mathematics and engineering. For details, readers
may refer to [3, 15, 20, 21] and their further references.
3 Description of Method
In this section, we present the Haar wavelet series method to find the approximate
solution of the proportional delay Riccati differential equation of fractional order
represented in Eq. (1). For that, we approximate χα (t) present in Eq. (1) by truncated
Haar wavelet series as follows:
2 J +1
α
χ (t) = ai hi (t). (15)
i=1
2 J +1
χ(t) = ai Piα (t) + χ(0). (16)
i=1
2 J +1
χ(qt) = ai Piα (qt) + χ(0). (17)
i=1
2 J +1 2 J +1
ai hi (t) = (t) + c1 (t)( ai Piα (t) + χ(0))
i=1 i=1
2 J +1
+c2 (t)( ai Piα (qt) + χ(0))(c3 (t) (18)
i=1
2 J +1
−( ai Piα (qt) + χ(0))).
i=1
A Collocation Method for Solving Proportional … 209
(l−0.5)
Discretize the system (18) with the chosen collocation points tl = 2 J +1
, we get
2 J +1 2 J +1
ai hi (tl ) = (tl ) + c1 (tl )( ai Piα (tl ) + χ(0))
i=1 i=1
2 J +1
+c2 (tl )( ai Piα (qtl ) + χ(0))(c3 (tl ) (19)
i=1
2 J +1
−( ai Piα (qtl ) + χ(0))).
i=1
Solve the above system for Haar wavelet coefficients ai, s. Plugging these coefficients
into the Eq. (16) produces the approximate solution χ(tl ).
The combination of fractional calculus with the theory of delay differential equa-
tions has enhanced the mathematical description of a number of real-world phe-
nomena during the past few years. On the other hand, several numerical treatments
have been developed for solving fractional differential models. However, very few
researchers have thoroughly investigated fractional differential equations with delay.
In this section, we shall be concerned with numerical treatment of some fractional
order delay differential equations using the Haar wavelet series method (HWSM).
Before solving numerical examples, we shall state some real-world applications of
HWSM from existing literature. In December 2019, a threatful outbreak called the
novel corona virus-2019 disease brought the world to its knees and took daily life to
a grinding halt in much of the world. The researchers claim that the virus was initi-
ated in the Chinese city of Wuhan. Planet-wide research to identify the symptoms,
to control its spread, and to cure and eradicate the disease is still in full swing. In an
attempt, Shah et al. [24] studied the transmission dynamics of the novel coronavirus-
2019 and construct a fractional order differential mathematical model by considering
three compartments including the susceptible population, infected population, and
recovered population. Further, the solution of the model is computed using the Haar
wavelet collocation method. Hence, the method is proven an efficient tool in infec-
tious disease spread modeling. Several recent studies which have promoted Haar
wavelet as a favorable mathematical tool are [4, 14, 16, 25].
1 t t
y α (t) = y(t) + y( )(1 − y( )), 0 ≤ α ≤ 1, t ∈ (0, 1), (20)
4 2 2
210 B. Hussain and A. Afroz
1.25
Approximate solution
Exact solution
1.25 1.2
1.2
approxinate solution
1.15
1.15
1.1
1.1
1.05
1
1 1.05
0.8 1
0.6 0.9
0.4 0.8
0.7
0.2 1
0.6
t 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0 0.5
(a) (b)
Fig. 1 a Haar Solution at α = 0.5, 0.6, 0.7, 0.8, 0.9, 1. b Solution Comparison at α = 1
2.8
Approximate solution
2.6 Exact solution
5
2.4
4.5
4
Approximate solution
2.2
3.5
3 2
2.5
1.8
2
1.5 1.6
1
1 1.4
0.8 1
0.6 0.9 1.2
0.4 0.8
0.7
0.2
0.6 1
t
0 0.5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
(a ) (b )
Fig. 2 a Haar Solution at α = 0.5, 0.6, 0.7, 0.8, 0.9, 1. b Solution Comparison at α = 1
1 t t
y α (t) = − y(t) + y( )(1 − y( )), 0 ≤ α ≤ 1, t ∈ (0, 1), (22)
8 2 2
with initial condition
1
y(0) = ,
4
212 B. Hussain and A. Afroz
0.42
Approximate solution
0.4 Exact solution
0.45
0.38
0.4
Approximate solution
0.36
0.35 0.34
0.32
0.3
0.3
0.25
1 0.28
0.8 1
0.6 0.9 0.26
0.4 0.8
0.7
0.2
0.6 0.24
t
0 0.5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
(a) (b)
Fig. 3 a Haar Solution at α = 0.5, 0.6, 0.7, 0.8, 0.9, 1. b Solution Comparison at α = 1
Example 4 Now for comparison, we choose the following fractional Riccati differ-
ential equation from literature [27]:
Approximate solution
0.9 Exact solution
1.2
0.8
1
0.7
Approximate solution
0.8
0.6
0.6
0.5
0.4
0.4
0.2
0.3
0
1
0.2
0.8 1
0.6 0.9
0.1
0.4 0.8
0.7
0.2
0.6 0
t
0 0.5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
(a) (b)
Fig. 4 a Haar solution at α = 0.5, 0.6, 0.7, 0.8, 0.9, 1. b Solution comparison at α = 1
10-16
1.2
0.8
Absolute error
0.6
0.4
0.2
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
t
5 Conclusion
In this paper, the HWSM is employed to explore the solutions of fractional order
proportional delay Riccati differential equations. We illustrate applicability and util-
ity of the method by solving a few benchmark problems. The comparison against
existing methods is presented in Tables 5, 6, and 7. Numerical simulations presented
in the form of tables and graphs show that the obtained results are comparatively
more promising and the method is well accurate for computing the solutions.
References
1. Atangana, A., Baleanu, D.: New fractional derivatives with nonlocal and non-singular kernel:
theory and application to heat transfer model (2016). arXiv:1602.03408
2. Ali, K.K., El Salam, M.A.A., Mohamed, E.M.: Chebyshev operational matrix for solving
fractional order delay-differential equations using spectral collocation method. Arab J. Basic
Appl. Sci. 26(1), 342–353 (2019)
3. Abdullah, A., Rafiq, M.: A new numerical scheme based on Haar wavelets for the numerical
solution of the Chen-Lee-Liu equation. Optik 226, 165847 (2021)
4. Hussain, B., Afroz., Jahan, S.: Approximate solution for proportional-delay Riccati differential
equations by Haar wavelet method. Poincare J. Anal. Appl. 8(2):157–170(2021)
5. Bellen, A., Zennaro, M.: Numerical Methods for Delay Differential Equations. Oxford Uni-
versity Press (2013)
6. Bahşi, M.M., Çevik, M.: Numerical solution of pantograph-type delay differential equations
using perturbation-iteration algorithms. J. Appl. Math. Art. ID 139821, 10 pp (2015)
7. Caputo, M., Fabrizio, M.: A new definition of fractional derivative without singular kernel.
Progr. Fract. Differ. Appl. 1(2), 1–13 (2015)
8. Davaeifar, S., Rashidinia, J.: Solution of a system of delay differential equations of multi
pantograph type. J. Taibah Univ. Sci. 11(6), 1141–1157 (2017)
9. Diethelm, K.: An application-oriented exposition using differential operators of Caputo type.
In: The Analysis of Fractional Differential Equations. Springer, Berlin (2010)
10. Garrappa, R.: Numerical solution of fractional differential equations: a survey and a software
tutorial. Mathematics 6(2):16 (2018)
11. Ghomanjani, F., Shateyi, S.: Solving a Quadratic Riccati differential equation, multi-pantograph
delay differential equations, and optimal control systems with pantograph delays. Axioms
9(3):82 (2020)
12. Izadi, M., Srivastava, H.M.: An efficient approximation technique applied to a non-linear Lane-
Emden pantograph delay differential model. Appl. Math. Comput. 401:126123,10pp (2021)
13. Jafari, H., Mahmoudi, M., Noori Skandari, M.H.: A new numerical method to solve pantograph
delay differential equations with convergence analysis. Adv. Differ. Equ. 2021:129 (2021).
https://doi.org/10.1186/s13662-021-03293-0
14. Kumar, S., Kumar, R., Agarwal, R.P., Samet, B.: A study of fractional Lotka-Volterra population
model using Haar wavelet and Adams-Bashforth-Moulton methods. Math. Methods Appl. Sci.
43(8), 5564–5578 (2020)
15. Lepik Ü., Hein H.: Haar Wavelets. In: Haar Wavelets. Mathematical Engineering. Springer,
Cham (2014)
16. Meng, L., Kexin, M., Ruyi, X., Mei, S., Cattani, C.: Haar wavelet transform and variational
iteration method for fractional option pricing models. Math. Methods Appl. Sci. (2022)
17. Mehra, M., Mehra, W., Ahmad, M.: Wavelets Theory and Its Applications. Springer, Singapore
(2018)
A Collocation Method for Solving Proportional … 217
18. Nievergelt, Y.: Wavelets Made Easy. Birkhäuser Boston, Inc., Boston (1999)
19. Podlubny, I.: Fractional Differential Equations: An Introduction to Fractional Derivatives, Frac-
tional Differential Equations, to Methods of Their Solution and Some of Their Applications.
Academic Press Inc, San Diego, CA (1999)
20. Raza, A., Khan, A.: Haar wavelet series solution for solving neutral delay differential equations.
J. King Saud Univ.-Sci. 31(4), 1070–1076 (2019)
21. Ray, S.S., Gupta, A.K.: Wavelet methods for solving partial differential equations and fractional
differential equations. CRC Press, Boca Raton, FL (2018)
22. Rihan, F.A.: Delay Differential Equations and Applications to Biology. Springer, Singapore
(2021)
23. Samko, S.G., Ross, B.: Integration and differentiation to a variable fractional order. Integral
transforms and special functions 1(4), 277–300 (1993)
24. Shah, K., Khan, Z.A., Ali, A., Amin, R., Khan, H., Khan, A.: Haar wavelet collocation approach
for the solution of fractional order COVID-19 model using Caputo derivative. Alexandria
Engineering Journal 59(5), 3221–3231 (2020)
25. Srivastava, H.M., Shah, F.A., Irfan, M.: Generalized wavelet quasilinearization method for
solving population growth model of fractional order. Mathematical Methods in the Applied
Sciences 43(15), 8753–8762 (2020)
26. Smith, H.: An introduction to delay differential equations with applications to the life sciences.
Texts in Applied Mathematics 57,Springer, New York (2011)
27. Sakar, M.G., Akgül, A., Baleanu, D.: On solutions of fractional Riccati differential equations.
Advances in Difference Equations 2017(1), 1–10 (2017)
28. Shah, F.A., Abass, R., Debnath, L.: Numerical solution of fractional differential equations using
Haar wavelet operational matrix method. International Journal of Applied and Computational
Mathematics 3(3), 2423–2445 (2017)
29. Yüzbaşı, Ş, Sezer, M.: An exponential approximation for solutions of generalized pantograph-
delay differential equations. Applied Mathematical Modelling 37(22), 9160–9173 (2013)
On the Solution of Generalized
Proportional Hadamard Fractional
Integral Equations
1 Introduction
The MNC for the very first time was initiated by Kuratowski [1] in 1930. This
notion of MNC is generalized by Banas [2] for the convenience to solve functional
equations, which is applicable to numerous mathematical problems. Using the notion
of MNC, Darbo [3] ensures that the existence of fixed points, which is obtained by
the generalization of Schauder and Banach’s fixed point theory.
Fractional integral equations (FIE) play a very important role in different fields
of mathematical analysis and still continue to earned the attention of researchers
in various applications of functional calculus in science and technology. Fractional
calculus is a very powerful tool to achieve differentiation and integration with real
or complex number powers, which is adopted in the sixteenth century. For recent
research on fractional calculus, we refer reader to (see [4–6]). In the present time,
the fixed point theory (FPT) has applications in several fields of mathematics. Also,
FPT can be applied for the existence of solutions of FIE.
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 219
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_16
220 Rahul and N. K. Mahato
τ σ1 σn−1 τ
dσ1 dσ2 f (σn ) 1 τ α−1 dσ
... dσn = log f (σ) , α > 0, τ > a.
σ1 σ2 σn Γ (α) σ σ
a a a a
We will be using the following notations, definitions, and theorems throughout this
paper.
– . E : norm on the Banach space E ;
– N̄ : closure of N ;
– ConvN : convex closure of N ;
– ME : collection of all nonempty and bounded subsets of E;
– NE : collection of all relatively compact sets;
– R: (−∞, ∞);
– R+ = [0, ∞);
– N: the set of natural numbers.
Banas and Lecko [8] have given the following definition of MNC.
Definition 1 A MNC is a mapping η : ME → R+ , if it fulfills the following con-
straints, for all N , N1 , N2 ∈ ME .
(N1 ) The family ker η = {N ∈ ME : η (N ) = 0} = ∅ and ker η ⊂ NE
(N2 ) N1 ⊂ N2 =⇒ η (N1 ) ≤ η (N2 )
(N3 ) η N̄ = η (N )
(N4 ) η (Conv N ) = η (N )
On the Solution of Generalized Proportional … 221
η (Φ M) < k η(M),
Definition 2 [11] Let N be a bounded subset of metric space E. Then for bounded
set N , the Hausdorff MNC η is defined as
Let E = C(I ) is the space of continuous functions on I = [a, b] with the norm
ϕ = sup {|ϕ(t)| : t ∈ I } , ϕ ∈ E.
Let Υ (= Φ) ⊂ E, so for any ϕ ∈ Υ and > 0, the modulus of the continuity of ϕ
is denoted as ω(ϕ, ) and defined by
3 Proposed Method
and
|L(s, P1 ) − L(s, P2 )| ≤ μ3 |P1 − P2 | , P1 , P2 ∈ R.
Theorem 3 If the conditions (i) − (iv) are satisfied, then the Eq. (1) has a solution
in E = C(I ).
Step 1: First, we have to prove that Φ maps B f0 into B f0 . Let Φ ∈ B f0 , then, by using
the assumptions, we have
,ρ
|(Φz)(s)| = | s, L(s, z(s)), H I1 z (s) − (s, 0, 0)| + |(s, 0, 0)|
,ρ
≤ μ1 |L(s, z(s)) − 0| + μ2 | H I1 z − 0| + |(s, 0, 0)|.
Also,
H ,ρ
| z (s) − 0|
I1
s
1 (ρ − 1) (log(s) − log(t)) z(t)
= exp (log(s) − log(t))−1 dt
ρ Γ () 1 ρ t
f 0 exp (ρ−1)ρlog T s dt
≤ (log(s) − log(t))−1
ρ Γ () 1 t
On the Solution of Generalized Proportional … 223
(ρ−1) log T
f 0 exp ρ
≤ (log T )
ρ Γ ( + 1)
≤ f0 .
(ρ−1) log T
f 0 exp
(log T ) .
ρ
Hence z < f 0 , gives Φz < μ1 μ3 f 0 + μ2 ρ Γ (+1)
So by the assumption (iv), Φ maps B f0 into B f0 .
Step 2: Now, we prove that Φ is continuous on B f0 . Let > 0 and z, z̄ ∈ B f0 such
that z − z̄ < , we have
Also,
H ,ρ H ,ρ
I1 z (s) − I1 z̄ (s)
s
1 (ρ − 1) (log(s) − log(t)) dt
= exp (log(s) − log(t))−1 (z(s) − z̄(s))
ρ Γ () 1 ρ t
s
1 (ρ − 1) (log(s) − log(t)) dt
≤ exp (log(s) − log(t))−1 |z(t) − z̄(t)|
ρ Γ () 1 ρ t
(ρ−1) log T
exp ρ
< (log T ) .
ρ Γ ( + 1)
where
Also,
H
,ρ ,ρ
I1 z (s2 ) − H I1 z (s1 )
s2
1 (ρ − 1) (log(s2 ) − log(t)) z(t)
= exp (log(s2 ) − log(t))−1 dt
ρ Γ () 1 ρ t
s1
1 (ρ − 1) (log(s1 ) − log(t)) z(t)
− exp (log(s1 ) − log(t))−1 dt
ρ Γ () 1 ρ t
s2
1 (ρ − 1) (log(s2 ) − log(t)) z(t)
≤ exp (log(s2 ) − log(t))−1 dt
ρ Γ () 1 ρ t
s1
(ρ − 1) (log(s2 ) − log(t)) z(t)
− exp (log(s2 ) − log(t))−1 dt
1 ρ t
s1
1 (ρ − 1) (log(s2 ) − log(t)) z(t)
+ exp (log(s2 ) − log(t))−1 dt
ρ Γ () 1 ρ t
s1
(ρ − 1) (log(s1 ) − log(t)) z(t)
− exp (log(s1 ) − log(t))−1 dt
1 ρ t
s2
1 (ρ − 1) (log(s2 ) − log(t)) |z(t)|
≤ exp (log(s2 ) − log(t))−1 dt
ρ Γ () s1 ρ t
s1
1 (ρ − 1) (log(s2 ) − log(t))
+ exp (log(s2 ) − log(t))−1
ρ Γ () 1 ρ
(ρ − 1) (log(s2 ) − log(t)) z(t)
− exp (log(s1 ) − log(t))−1 dt
ρ t
exp (ρ−1)ρlog T
≤ z(log T )
ρ Γ ( + 1)
s1
1 (ρ − 1) (log(s2 ) − log(t))
+ z exp (log(s2 ) − log(t))−1
ρ Γ () 1 ρ
(ρ − 1) (log(s2 ) − log(t)) 1
− exp (log(s1 ) − log(t))−1 dt.
ρ t
H ,ρ ,ρ
As → 0, then s2 → s1 , and also I1 z (s2 ) − H I1 z (s1 ) → 0.
On the Solution of Generalized Proportional … 225
Therefore,
gives
H ,ρ ,ρ
ω(Φz, ) ≤ μ2 I1 z (s2 ) − H I1 z (s1 ) + μ1 μ3 ω(z, ) + ω (I, ).
ω0 (ΦΥ ) ≤ μ1 μ3 ω0 (Υ ).
Thus, by DFPT Φ has a FP in Υ ⊆ B f0 . Hence the Eq. (1) has a solution in C(I ).
for s ∈ [1, 2] = I.
Here,
1
H 5, 5 3125 s z(t)
I1 z (s) = exp −4 (log(s) − log(t)) (log(s) − log(t))4 dt.
Γ (5) 1 t
√
I1 3 z(s)
Also, (s, L, I1 ) = s 3 + L + 3500 and L(s, z) = 7+s 5 +s 2
. It can be seen that both
, L are the continuous functions satisfying
|P1 − P2 |
|L(s, P1 ) − L(s, P2 )| ≤
9
and
1
(s, L, I1 ) − (s, L̄, Ī1 ) ≤ L − L̄ + I1 − Ī1 .
3500
Therefore, μ1 = 1, μ2 = 3500
1
, μ3 = 19 and μ1 μ3 = 19 < 1. If z ≤ f 0 , then we
have
f0 55 f 0 exp[−4(log 2)](log 2)4
L= , I= .
9 Γ (6)
226 Rahul and N. K. Mahato
Further,
55 f 0 exp[−4(log 2)](log 2)4
|(s, L, I1 | ≤ ≤ f0 .
3500Γ (6)
≤ 5, μ1 μ3 < 1.
We see that all the assumptions from (i)−(iv) of Theorem 3 are fulfilled. Hence, by
the Theorem 3, we concluded that the Eq. (1) has a solution in C(I ).
4 Conclusion
In the current paper, we have defined a new class of fractional integral operators,
which can be reduced to other related operators by choosing suitable values. Then,
we established the endurance of solution of a GPHF integral equation, using DFPT.
Finally, the obtained result is illustrated by a suitable example.
References
1. Kuratowski, K.: Sur les espaces complets. Fundamenta Mathematicae 15, 301–309 (1930)
2. Banas, J., Goebel, K.: Measures of Noncompactness in Banach Spaces. Lecture Notes in Pure
and Applied Mathematics, Dekker, New York (1980)
3. Darbo, G.: Punti uniti in trasformazioni a codominio non compatto (Italian). Rendiconti Del
Seminario Matematico Della Universit di Padova 24, 84–92 (1955)
4. Agarwal, R., Almeida, R., Hristova, S., O‘Regan, D.: Caputo fractional differential equation
with state dependent delay and practical stability. Dyn. Syst. Appl. 28(3), 715–742 (2019)
5. Hristova, S., Ivanova, K.: Caputo fractional differential equations with non-instantaneous ran-
dom erlang distributed impulses. Fractal Fract. 3(2), 28–37 (2019)
6. Rahul, Mahato, N.K.: Existence solution of a system of differential equations, using generalized
Darbo’s fixed point theorem. AIMS Math. 6(12), 13358–13369 (2021)
7. Hadamard, J.: Essai sur l‘étude des fonctions données par leur développement de Taylor. J.
Pure Appl. Math. 4(8), 101–186 (1892)
8. Banas, J., Lecko, M.: Solvability of infinite systems of differential equations in Banach sequence
spaces. J. Comput. Appl. Math. 6(2), 363–375 (2001)
9. Aghajani, A., Banas, J., Sabzali, N.: Some generalizations of Darbo fixed point theorem and
applications. Bull. Belgian Math. Soc-Simon Stevin 2(2), 345–358 (2013)
10. Altun, I., Turkoglu, D.: A Fixed point theorem for mappings satisfying a general condition of
operator type. J. Comput. Anal. Appl. 22(1), 1–9 (2007)
11. Banas, J., Goebel, K.: Measure of Noncompactness in Banach Spaces. Lecture Notes Pure
Appl. Math. 60 (1980)
Optimization Theory and Applications
An Invitation to Optimality Conditions
Through Non-smooth Analysis
Joydeep Dutta
Abstract In this short article, we show the fundamental role that non-smooth anal-
ysis plays in devising optimality conditions. Written with the graduate students and
young researchers in mind, this article aims to bring to the fore how non-smooth
analysis lies at the core of modern optimization.
1 Intoduction
J. Dutta (B)
Department of Economic Sciences, Indian Institute of Technology, Kanpur 208016, India
e-mail: [email protected]
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 229
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_17
230 J. Dutta
∇ f (x ∗ ) + J H (x ∗ )T y ∗ = 0.
This is the celebrated Lagrange multiplier rule and y ∗ is the Lagrange multiplier
vector. What happens if we ignore the rank condition on the Jacobian, can we still
have something to say about the local minimizer of (P). The answer in fact is yes.
We can without any condition prove that there exists (y0∗ , y ∗ ) ∈ R × Rm such that,
(y0∗ , y ∗ ) = 0,
y0∗ ∇ f (x ∗ ) + J H (x ∗ )T y ∗ = 0.
This can be viewed as a primitive form of the Lagrange multiplier rule. However, in
the erstwhile in the Soviet union, this approach to the multiplier rule was fundamental.
In this situation one might have y ∗ = 0, and thus the objective function gets removed
from the process of computing a candidate minimizer of (P1). In fact in many prob-
lems from the hypothesis of the problem, one can deduce that y0∗ = 0. This issue has
been beautifully dealt with in the text by Brinkhuis and Tikhomirov [2]. Further it
goes without saying that if y ∗ = 0 then we can normalize to consider y0∗ = 1. Once
we set y0∗ = 1 in the above equation it reduces to the usual Lagrange multiplier rule.
Another approach to guarantee that y0∗ = 0 is to make certain assumptions on the
Jacobian of H at x ∗ . The most natural assumption is to assume that J H (x ∗ ) has full
row-rank m. Once this assumption is in place, once we have y0∗ = 0, we shall imme-
diately conclude that the vector y ∗ = 0 and this contradicts the fact (y0∗ , y ∗ ) = 0.
During the Second World War, new optimization problems arose. The hallmark
of these problems was that they were having inequalities as constraints. Consider the
problem (P2)
the issue see, for example, Bazaara, Sherali, and Shetty [1]. Let us now turn our
attention to the condition (iii) in the Fritz John necessary optimality conditions. If
we write the vector function G(x) = (g1 (x), . . . , gm (x)) and λ∗ = (λi∗ , . . . , λ∗m ),
then the condition (iii) implies that λi∗ gi (x ∗ ) = 0 for all i = 1, . . . , m. Observe that
if gi (x ∗ ) < 0 then λi∗ = 0 and if λi∗ > 0 then gi (x ∗ ) = 0. This condition is thus
called the complementary slackness condition which means that both λi∗ and gi (x ∗ )
cannot hold with strict inequalities at the same time. Further we can also guarantee
that λ∗0 > 0 by assuming certain additional conditions on the constraints. This was
independently achieved by Kuhn and Tucker [4] in 1951. I had the great privilege to
know how the now celebrated Karush-Kuhn-Tucker conditions or KKT conditions
were developed by Harold Kuhn himself at the sidelines of the EURO-OR conference
held in Lisbon in 2010. They approached the problem form a very different angle and
not through the lens of the Fritz John conditions. In fact Kuhn, Tucker, and Gale had
already established the necessary and sufficient condition for a linear programming
problem. These conditions for the linear programming problems were their guide to
prove a necessary and if possible sufficient condition for the problem (P2). Tucker
then wanted to device an optimality conditions for quadratic programming problems
David Gale however left the team to focus on game theory and it was Kuhn who
suggested that they focus on the case of differentiable functions and that allowed
them to provide the following results for (P2).
If we assume a suitable condition satisfied by the constraints at a local minimizer
x ∗ , then Kuhn and Tucker established the existence of λ∗ ∈ Rm + such that
(i) ∇ f (x ∗ ) + J G(x ∗ )T λ∗ = 0
(ii) λ∗ , G(x ∗ ) = 0.
Observe that the above conditions are the Fritz John conditions with λ∗0 = 1, and
hence the condition (λ∗0 , λ∗ ) = 0 is automatically met. The question that is of crucial
importance is that what is that condition which guarantees that λ∗0 = 0, i.e., λ∗0 > 0
in the case. In their famous paper of 1951, Kuhn and Tucker introduced a geometric
condition now called the Kuhn-Tucker constraint qualification. However, a more
natural condition can be provided as follows.
the non-differentiability of the function generically lay precisely at the point where
the function has a minimum value. Traditionally, f (x) = |x|, x ∈ R is usually men-
tioned as a prototype model of such a function. Observe that x = 0 is the (global)
minimizer over R and the function has no derivative there. A key notion was that
of a subdifferential, which is a set-valued map playing the role of a derivative. An
extensive calculus rule was developed for the subdifferential and it deviated from
the usual calculus since the subdifferential could be calculated for convex functions
which intrinsically has no derivative. The calculus of convex functions was extended
for locally Lipschitz function by Clarke [5]. However the calculus rules for the now
famous Clarke generalized gradient or Clarke subdifferential are weaker than those
of convex functions. Later on in the setting of a lower-semicontinuous function, the
limiting subdifferential played a key role. Boris Mordukhovich [7, 8] Alexander
Kruger [9], Jon. Borwein and Zhu [10], Rockafellar and Wets [11] played a crucial
role in the development of the calculus of limiting subdifferential. From the erstwhile
Soviet union came two other different approaches to subdifferential for non-smooth
functions. These are, namely, the tangential subdifferential of Pschenichny and the
quasi-differential by Demyanov and Rubinov [12].
Our focus in this article will be a brief survey of role played by the limiting subd-
ifferential as the uniting force in developing the necessary optimality conditions for
various classes of optimization problems. We list below the headings of the various
sections of the paper.
Section 2: A non-smooth Analysis Tool Box
Section 3: Basic Optimality Conditions
Section 4: The geometry of BCQ
In Sect. 2, we shall briefly present the main tools of non-smooth analysis, reflecting
both the geometric and the analytic aspects. While discussing non-smooth geometry,
our key focus would be to discuss the notion of the limiting normal cone and how
such a robust object can be built up from more elementary notions of normal cones
specially in a non-convex setting. In the analytical aspect, our key goal would be to
elucidate the notion of the limiting subdifferential and tie it up with robust calculus
rules associated with it. We shall also emphasize the role of the proximal normal
cone and the proximal subdifferential in providing a more clear view of the limiting
normal cone and the limiting subdifferential.
In Sect. 3, our focus is on developing optimality conditions for the basic problem
of minimizing a function f over the set C. We will show how a geometric condition
called the transversality condition plays a crucial role in deriving the optimality
conditions. In Sect. 4 we show that how non-smooth tools can be effectively used in
developing the KKT optimality conditions for smooth optimization problems with
both inequality and equality constraints. In this, we also show that the transversality
conditions are actually the geometric version of the Basic Constraint Qualification
(BCQ). In this section, we show an example of an optimistic bilevel programming
problem where the Basic Constraint Qualification fails at a local minimizer.
Our notations as we have already used some in this section are fairly standard.
For example, for the inner product of two vectors in Rn , we use the notation x, y ,
An Invitation to Optimality Conditions Through Non-smooth Analysis 233
where x, y ∈ Rn . For any set A, intA and Ā denote the interior and closure of A,
respectively. Other notations denoting sequences and convergence are also fairly
standard. A sequence of vectors in Rn , (n > 1) will be denoted by {v k }, where as a
sequence in Rn is given as {tk }, k ∈ N. The convex hull of a set A is denoted as co A.
We begin with the hope that the reader is fairly conversant with convex analysis and
basic optimization. Our aim here is only to open the entrance of huge edifices called
non-smooth analysis and optimization using the limiting subdifferential as the guide.
Let us note in the beginning that this article is more in the form of a research exposition
or survey rather than an original work. Even in this exposition we want to keep
ourselves at the simplest level. Our key tool will be the limiting normal cone and
the limiting subdifferential. We intend to study variational geometry or non-smooth
geometry first and then move on to subdifferentials and useful calculus rules. Our
key source would be Rockafellar and Wets [11], Mordukhovich [7] and Loewen [13].
One of the key tools of non-smooth geometry is the notion of a Bouligand tangent
cone introduced in 1938. Let S ⊆ Rn , and let x̄ ∈ Rn , then v is a tangent vector to
S at x̄ ∈ Rn if there exists a sequence {v k } in Rn , such that v k → v and tk ↓ 0, (i.e.,
tk > 0 & tk → 0) such that x̄ + tk v k ∈ S. The collection of all tangents vectors of S
at x̄ is called the Bouligand tangent cone and is denoted by TS (x̄).
Given a convex set C ⊆ Rn , the normal cone to C at x̄ ∈ C, is given as the set
NC (x̄) is a closed convex cone and when C is a convex set, TC (x̄) is also a closed
convex cone and both of these objects are connected by the polarity relation
o(x − x̄)
lim =0
x→x̄ x − x̄
Intuitively the right-hand side in (1) measures the possible derivation from con-
vexity. The set of all regular normals to C at x̄ forms a cone called the regular normal
cone to C at x̄ and is denoted as N̂C (x̄) = NC (x̄). The drawback of N̂C (x̄) is that it
can just reduce to the trivial set {0}. To avoid such unpleasant situations the notion
of the limiting normal cone was introduced. A vector v ∈ Rn is called a limiting
normal to C at x̄, if there exists a sequence of vectors {x k } in C, such that x k → x̄
and sequence {v k } in Rn such that v k → v, as k → ∞ with v k ∈ N̂C (x k ) for each
k ∈ N. The collection of all the limiting subdifferentials is called the limiting normal
cone and is denoted as NCL (x̄). Further it is clear that N̂C (x̄) ⊂ NCL (x̄), but unlike
N̂C (x̄), the limiting normal cone to C at x̄ need not be convex even though it is
closed. Further if C is convex we have
This has given rise to notion of regularity of sets. A set C ⊆ Rn is said to be regular
at x̄ if N̂C (x̄) = NCL (x̄). A convex set is thus regular at all its points. However if one
needs to visualize the limiting normal cone in a more effective way, we need to
develop the idea of the limiting normal cone form the lens of a proximal normal
cone. Assuming that the idea of the projection on to a closed convex set we define
the notion of projection map. Given S ⊆ Rn , the projection map Pr ojS is a set-valued
map form Rn to S, i.e., Pr ojS : Rn ⇒ S, given as
If S is a closed set then Pr ojS (x) = ∅ for each x. Else there can be x ∈ Rn for which
Pr ojS (x) = ∅.
We say a vector v ∈ Rn is a proximal normal to S at x̄, provided x̄ ∈ S, and there
exists y ∈ Rn such that x̄ ∈ Proj S (y) and v = λ(y − x̄) for some λ > 0. The set
of all proximal normals forms a cone called the proximal normal cone, denoted as
NCP (x̄). Thus
for all x ∈ C.
An Invitation to Optimality Conditions Through Non-smooth Analysis 235
This shows that v ∈ N̂C (x̄). This shows that NCP (x̄) ⊂ N̂C (x̄), though the reverse
inclusion may not hold.
In fact if v ∈ NCL (x̄), then there exists a sequence {x k } in C and {v k } in
R , with x k → x̄ and v k → v such that v k ∈ NCP (x k ). This view of limiting nor-
n
C = epi(−|x|), x ∈ R
where epi f denotes the epigraph of the function. In Figs. 2 and 3 below the green
coloured part denotes the epigraph of f (x) = −|x|.
y
4
1
x
-4 -3 -2 -1 1 2 3 4
-1
-2
-3 y=-|x|
-4
Observe that outside C, there is no point whose projection is the point (0, 0), i.e.,
2.2 Subdifferentials
If x̄ ∈
/ dom f , then define ∂ f (x̄) = ∅. It is simple to observe that ∂ f (x̄) is a convex
set. Further x̄ is a global minima of f over Rn , if and only if 0 ∈ ∂ f (x̄). An important
An Invitation to Optimality Conditions Through Non-smooth Analysis 237
example which we will need in what follows is the subdifferential of the indicator
function δC of a set C ⊆ Rn . We have
0, if x ∈ C
δC (x) =
∞ if x ∈/ C.
o(y − x̄)
lim = 0.
y→x̄ y − x̄
The set of all regular subgradients forms a set called the regular subdifferential ∂ˆ f (x̄).
If f is convex, then
∂ˆ f (x̄) = ∂ f (x̄).
• It is important to note that ∂ˆ f (x̄) need not be non-empty at each of dom f point
even if f is locally Lipschitz.
• Further there is an elegant geometrical representation of ∂ˆ f (x̄) given as
∇ f (x̄), v ≥ ξ, v
∇ f (x̄) − ξ, v ≥ 0, ∀v ∈ Rn .
238 J. Dutta
For more details on how to arrive at above form of the limiting subdifferential see,
for example, [7, 11, 14].
Example 2 Let f (x) = −|x|, then we shall compute ∂ L f (0). Thus
Observe that
previous subsection. Hence from (2.2) we conclude that ∂ L f (0) = {−1, +1} . It is
clear that ∂ L f (0) is closed though not convex.
However a more simpler way or rather a geometrical way of viewing the limiting
normal is through the notion of a proximal subdifferential rather than the regular
subdifferential. Looking at the geometrical representation of other subdifferentials it
is intuitive to define the proximal of a proper function f : Rn → R̄ at x̄ ∈ dom f as
The set ∂ P f (x̄) is convex but may not be non-empty at each x̄ ∈ dom f .
Observe that if f (x) = −|x|, then,
Observe that for f (x) = −|x|, we have ∂ P f (0) = ∅ and ∂ L f (0) = {−1, +1}.
Observe that if {x k } ⊆ R, x k ≥ 0, x k → 0, then ∂ P f (x k ) = {−1} while if we take
{x k } ⊆ R, x k ≤ 0, x k → 0, then ∂ P f (x k ) = {+1} . Thus from the above 3 it is clear
that ∂ L f (0) = {−1, +1}. Let us now provide an example of computing the limiting
subdifferential of a function f : R2 → R and demonstrate that even for a simple
function such a computation need not be very simple.
Our aim is to compute ∂ L φ(x, y), for any (x, y) ∈ R2 . Observe that along the x-axis
and y-axis, φ is not differentiable. So let us have a look at the structure of φ in detail.
Also note that if φ is continuously differentiable at (x̄, ȳ), then
then,
∂ L h(x, y) = ∂ L f (x) × ∂ L g(y).
One can obtain this using the definition of the limiting subgradients in terms of
proximal normal. We will provide the proof after we complete the example.
Observe that
Now let us compute ∂ L φ(x, y) along the y-axis, at all points except the origin. If
y>0
If y < 0
We will now compute along x-axis, except the origin. If x > 0, then
It is important to note that if (ξ, η) ∈ ∂ P φ(x̄, ȳ) then using Proposition 4A.3 from
Loewen [13], we conclude that ∃δ > 0 and μ > 0 such that
φ(x, y) − φ(x̄, ȳ) ≥ (ξ, η), (x, y) − (x̄, ȳ) − μ(x, y) − (x̄, ȳ)2 ,
for any (x, y) such that (x, y) − (x̄, ȳ) < δ, thus
for all (x, y) with (x, y) − (x̄, ȳ) < δ. Note that (x, ȳ) − (x̄, ȳ) < δ., whenever
x − x̄ < δ Hence,
Hence ξ ∈ ∂ P f (x̄) using Proposition 4.3A of Loewen [13]. We can in a similar way
prove that η ∈ ∂ P g( ȳ). Hence (ξ, η) ∈ ∂ P f (x̄) × ∂ P g( ȳ). Thus
Consider now ξ ∈ ∂ P f (x̄) and η ∈ ∂ P g( ȳ). Now again using Proposition 4A.3 in
Loewen [13], we have that ∃δ > 0 and μ > 0 such that
δ
f (x) − f (x̄) ≥ ξ, x − x̄ − μx − x̄2 , ∀x with x − x̄ < .
2
Further
δ
g(y) − g( ȳ) ≥ η, x − x̄ − μy − ȳ2 ∀y with y − ȳ < .
2
Thus
φ(x, y) − φ(x̄, ȳ) ≥ (ξ, η), (x, y) − (x̄, ȳ) − μ[(x − x̄, y − ȳ)2 ].
for all x, y with (x, y) − (x̄, ȳ) < δ. Hence (ξ, η) ∈ ∂ P φ(x̄, ȳ). This immediately
shows that
∂ P φ(x̄, ȳ) = ∂ P f (x̄) × ∂ P g( ȳ).
Now let (ξ, η) ∈ ∂ L φ(x̄, ȳ), then there exists (x k , y k ) → (x̄, ȳ) and (ξ k , η k ) ∈
∂ P φ(x̄ k , ȳ k ) such that (ξ k , η k ) → (ξ, η). Thus ξ k ∈ ∂ P f (x k ) and η k ∈ ∂ P g(y k ).
242 J. Dutta
Let (ξ, η) ∈ ∂ L f (x̄) × ∂ L g( ȳ). Then ξ ∈ ∂ L f (x̄) and η ∈ ∂ L g( ȳ). Hence there exists
x k → x̄, ξ k → ξ with ξ k ∈ ∂ P f (x k ) and also y k → ȳ, η k → η such that η k ∈
∂ P g(y k ). Hence (ξ k , η k ) ∈ ∂ P φ(x k , y k ) as (ξ k , η k ) → (ξ, η) and (x k , y k ) → (x̄, ȳ).
Hence (ξ, η) ∈ ∂ L φ(x̄, ȳ). Thus
Also observe in Example 2 the function φ(x, y) = |x| − |y| is locally Lipschitz and
also observe that ∂ L φ(x̄, ȳ) was closed and bounded for any (x, y) ∈ R2 . In fact it
can be seen, for example, from Vinter [14] that for any locally Lipschitz function on
Rn , the limiting subdifferential is always non-empty and compact.
In the world of non-smooth, non-convex optimization, subdifferential plays an
important role. However, from my own individual experience, they play more of an
explanatory role rather than playing the key role in the actual computation of a local
minimizer. From mid-1970s till the early half of the twenty-first century, the land-
scape of non-smooth analysis was dominated by the Clarke subdifferential or Clarke
generalized gradient. Unlike the limiting subdifferential the Clarke subdifferential is
built on a more fundamental object called the Clarke generalized derivative.
If f is a locally Lipschitz function in Rn , i.e., f : Rn → R and locally Lipschitz,
then the Clarke generalized directional derivative of f at x̄ in the direction of v ∈ Rn
f (y + λv) − f (y)
f 0 (x̄, v) = lim sup .
y→x̄ λ
λ↓0
The function f 0 (x̄, v) exists finitely, for each x̄ ∈ Rn and v ∈ Rn . The Clarke subd-
ifferential of f at x̄ is given by
Of course one has 0 ∈ ∂ 0 f (x̄) if x̄ is a local minimizer however the converse is not
true. The set ∂ 0 f (x̄) is always a non-empty convex and compact set. Further
Before we end this section we shall introduce the notion of a singular subdifferential,
which can be a measure of deviation of a function from being locally Lipschitz.
An Invitation to Optimality Conditions Through Non-smooth Analysis 243
In fact one show that v ∈ ∂ L∞ f (0) implies that v ∈ Ndom f (x̄), if f is convex. This
object will play a key role in calculus rules.
Our aim here would be to present two calculus rules. Namely, the sum rule and
the calculation of subdifferential of the max. In our exposition, the use of the term
subdifferential only refers to the limiting subdifferential. We begin with the sum rule.
Then
Another result which follows from Theorem 1 is a calculus of the limiting normal
cones, and will have a key role in understanding optimality conditions.
Theorem 2 Let C1 and C2 be two non-empty sets. If C1 ∩ C2 = ∅. Let x̄ ∈ C1 ∩ C2
and assume that
[NCL1 (x̄)] ∩ [−NCL2 (x̄)] = {0}.
Then
NCL1 ∩C2 (x̄) ⊂ NCL1 (x̄) + NCL2 (x̄).
244 J. Dutta
The proof of the result follows by a direct application of Theorem 1, with f 1 = δC1
and f 2 = δC2 and noting that for any C ⊆ Rn ,
Leaving the proof of the first equality to the reader, who can again use the power of
the proximal subdifferential. The reader can first establish the fact that
Let us just briefly show how to obtain the second inequality and thereby bringing
out some beauty of non-smooth analysis. Observe that epiδC = C × R+ . Therefore
We are now going to state how to estimate the limiting subdifferential of a composi-
tion of two mappings and the estimate of the subdifferential of a max function will
follow as a corollary. We shall state this result as given in Loewen [13].
If F : Rn → Rm , then we can write
m
∂ L g(x) ⊂ {∂ L (y F)(x) : y ∈ Rm
+, yi = 1 with yi = 0 if i ∈
/ J (x)}, (2)
i=1
min f (x)
x∈C
Hence we have 0 ∈ ∂ L ( f + δC )(x̄). To apply the sum rule to get a more easily
verifiable condition we need to assume the condition
Once this is satisfied the sum rule applies and we have the optimality condition
f (y) ≥ f (x),
for all y in such y − x < δ . This show that 0 ∈ ∂ p f (x) and hence 0 ∈ ∂ L f (x) as
∂ P f (x) ⊂ ∂ L f (x). Similar arguments will work for the regular normal cone. Note
that in the problem (P), if f is locally Lipschitz then ∂ L∞ f (x̄) = {0} and hence the
condition
∂ L∞ f (x̄) ∩ (−NC (x̄)) = {0} (3)
automatically holds.
The condition (3) is often referred to as the transversality condition and as we
will see later is deeply linked with the basic constraint qualification. Clarke subdif-
ferential of a locally Lipschitz finite valued function is the convex hull of its limiting
subdifferential. This convexification kills several key properties. Observe that for
f (x) = −|x|, x̄ = 0 is the global maximizer and not the minimizer, local or global.
Here the limiting subdifferential ∂ L f (0) = {−1, +1}, showing that 0 ∈ / ∂ L f (0) and
An Invitation to Optimality Conditions Through Non-smooth Analysis 247
hence zero cannot be a local minimizer. Thus the limiting subdifferential is a powerful
tool, to analyze local minimizers. While ∂ ◦ f (0) = co∂ L f (0) = [−1, +1], and we
have 0 ∈ ∂ ◦ f (0), though the construction of the Clarke subdifferential is such that
it is geared towards analyzing local minimizers. The Clarke subdifferential does not
provide more information about the nature of an optimizing point though the limiting
subdifferential does. Let us have a look again at the transversality condition and we
will observe that it is equivalent to, the following. If v1 ∈ ∂ L∞ f (x̄) and v2 ∈ NCL (x̄).
Then the transversality condition holds at x̄ iff
v1 + v2 = 0 =⇒ v1 = 0, v2 = 0.
One of the key focuses of this exposition is to explore the transversality condition
and its role in devising optimality conditions. Now consider the problem (P) with,
C = C1 ∩ C2 , where C1 , C2 ⊂ Rn
If x ∗ be a local minimizer of (P), then the necessary condition for optimality is given
as
0 ∈ ∂ L ( f + δC1 ∩C2 )(x̄). (4)
This motivates us to look into the extended form of the sum rule. Let f (x) = f 1 (x) +
f 2 (x) . . . + f m (x), where each f i : Rn → R̄ is proper and lower-semicontinuous. Let
x̄ ∈ ∩i=1m
dom f i . Further assume that whenever vi ∈ ∂ L∞ f i (x̄), i = 1, . . . , m and
v1 + . . . + vm = 0 we have vi = 0 for all i. Then
∇ f (x̄), d ≥ 0, ∀ d ∈ TC (x̄)
C (x̄).
f (x̄) ∈ N
Let us turn over attention to convex programming problems and more specifically
for a convex programming with linear constraints.
Consider the problem (CLP)
min f (x)
subject to Ax = b
x ≥ 0,
min f (x)
x ∈ C1 ∩ C2 ,
where
C1 = {x : Ax = b}
C2 = {x : x ≥ 0} = Rn+
NC1 (x) = Im A T ; ∀ x ∈ C1 .
So the natural question is as follows: Does the condition (5) holds automatically for
the problem (CLP)? The answer is surprisingly no.
Example 5 Consider
C1 = {(x1 , x2 ) ∈ R2 : x1 + x2 = 0}
C2 = R2+
Hence,
min f (x)
gi (x) ≤ 0, i = 1, . . . , m
h j (x) = 0, j = 1, . . . , k,
250 J. Dutta
m
k
0= λi ∇gi (x0 ) + μ j ∇h j (x0 )
i=1 j=1
Proof We shall essentially provide a scheme of the proof. The reader is requested to
fill up the gaps. Since x0 is a local minimizer of (MP) we observe that x0 is a local min-
imizer of F(x) over x ∈ X , where F(x) = max{ f (x) − f (x0 ), g1 (x), . . . , gm (x)}
and X = {x ∈ Rn : h j (x) = 0, j = 1, . . . , k}. Since F is locally Lipschitz and hence
0 ∈ ∂ F(x0 ) + N X (x0 ).
0 = λ0 ∇ f (x0 ) + λi ∇gi (x0 ) + N XL (x0 ).
i∈I (x0 )
Since BCQ holds at x0 , {∇h 1 (x0 ), . . . , ∇h j (x0 )} are linearly independent. Hence,
using Proposition 1.9 in Clarke et.al.[6], we have
⎧ ⎫
⎨k ⎬
N XP (x0 ) ⊂ μ j ∇h j (x0 ) : μ j ∈ R .
⎩ ⎭
j=1
k
ξn = λnj ∇h j (x k ) (6)
j=1
k
0= ω j ∇h j (x0 ),
j=1
k
0 = λ0 ∇ f (x0 ) + λi ∇gi (x0 ) + μ j ∇h j (x0 ).
i∈I (x0 ) j=1
If λ0 = 0, then
k
0= λi ∇gi (x0 ) + μ j ∇h j (x0 ),
i∈I (x0 ) j=1
as λi + λ0 = 1 it implies that there exists i ∈ I (x0 ) such that λi > 0 and thus
i∈I (x0 )
violating BCQ. Hence the result. The part (ii) is obtained by setting λi = 0, for
i ∈ I (x0 ).
Our next aim is to show that the satisfaction BCQ at a feasible point x0 , then the
transversality condition
holds with
C1 = {x : gi (x) ≤ 0, i = 1, . . . , m} (7)
C2 = {x : h j (x) = 0, j = 1, . . . , k} (8)
252 J. Dutta
Theorem 4 Let C1 and C2 be defined above in (7) and (8). Let x0 ∈ C1 ∩ C2 . Let
us assume that BCQ holds at x0 for the problem (MP). Then
and ⎧ ⎫
⎨k ⎬
NCL2 ⊂ μ j ∇h j (x0 ) : μ j ∈ R .
⎩ ⎭
j=1
Note the BCQ implies that {∇h 1 (x0 ), . . . , ∇h j (x0 )} forms a linearly independent set
of vectors. From the discussion in Theorem 3 we know that NCL2 can be estimated as
above. Note the BCQ at x0 also implies that
m
λi ≥ 0, λi = 0, if i ∈ I (x0 ) and λi ∇gi (x0 ) = 0 =⇒ λi = 0, ∀ i = 1, . . . , m.
i=1
Now arguing in the way we did in the proof of the last theorem (Theorem 3) we
conclude that
m
NC 1 ⊂
L
λi ∇gi (x0 ) : λi ≥ 0, λi gi (x0 ) = 0 .
i=1
v1 + v2 = 0.
m
Now ∃ λ̂i ≥ 0, λ̂i gi (x0 ) = 0 such that v1 = λ̂i ∇gi (x0 ) and also there exists
i=1
μ̂ j ∈ R, such that
m
v2 = μ̂ j ∇h j (x0 )
i=1
An Invitation to Optimality Conditions Through Non-smooth Analysis 253
since v1 + v2 = 0, we have
m
k
λ̂i ∇gi (x0 ) + μ̂ j ∇h j (x0 ) = 0.
i=1 j=1
Further, assume that {∇h 1 (x0 ), . . . , ∇h m (x0 )} are linearly independent and there
exists d ∈ R such that ∇gi (x0 ), d < 0 for all i ∈ I (x0 ). Then, BCQ holds at x0 for
(MP).
Our next and final discussion is of particular class of problems, called bilevel pro-
gramming problems for which the Basic Constraint qualification never holds at any
feasible point. This was discussed in Dutta [17], where the proof appears to be
incomplete. We supply the full proof here.
Consider the following optimistic bilevel programming problem (OBP)
min F(x, y)
x,y
subject to y ∈ S(x)
where S(x) = argmin{ f (x, y) : y ∈ K }
y
We assume that F(x, y) is jointly convex in (x, y), f (x, y) is also jointly convex
and y ∈ K is convex. Here F(x, y) is called the upper-level objective or the leader’s
objective and f (x, y) is called the lower-level objective or follower’s objective.
Even though the problem has fully convex data; the problem is intrinsically non-
smooth and non-convex. For more details on bilevel programming see, for example,
Dempe [18] and the references therein. For optimistic bilevel programming see, for
example, Dempe et al. [19] and the references there in. Here we assume the upper-
level objective function is smooth. In Dutta [17] it was shown that if (x̄, ȳ) is a local
minimizer, then there exists λ0 ≥ 0 and λ1 ≥ 0 such that
(i) 0 ∈ λ0 ∇ F(x̄, ȳ) + λ1 (∂ f (x̄, ȳ) − (∂v(x̄) × {0}) + {0} × N K ( ȳ))
(ii) 0 = (λ0 , λ1 ) ∈ R+ × R+
Here v(x) = inf { f (x, y) : y ∈ K }, the value function associated with the lower-
y
level problem of the bilevel problem (OBP).
254 J. Dutta
The above Fritz John type necessary condition was possible since the bilevel
problem is equivalent to the single-level problem
min F(x, y)
x,y
Note that v(x) is also a convex function on Rn . Since we have not made any assump-
tion on the differentiability of the convex functions and the single-level problem above
is a non-convex and non-smooth optimization problem. So we shall now describe
what should be the form of BCQ for the following problem (NSP),
m
0∈ λi ∂ L gi (x0 ) + N X (x0 )
1=1
implies λ1 = 0.
Using partial subgradients of the convex function, the Basic Constraint Qualifi-
cation can be written as
implies λ1 = 0.
We shall show that (9) and (10) can hold with λ1 > 0. Here ∂x f (x̄, ȳ) means
the subdifferential with respect to x for the convex function f (., ȳ). The partial
subdifferential with y can be defined similarly.
Let w ∈ ∂v(x̄). Hence, for any x ∈ Rn
But v(x̄) = f (x̄, ȳ), as ȳ ∈ argmin f (x̄, y). Further, v(x) ≤ f (x, ȳ); for any
y∈K
x ∈ Rn .
Hence, for any x ∈ Rn
Hence, we can take λ1 > 0 in (10). This shows us that BCQ never holds for the
problem (OBP).
The qualification condition BCQ has another deep link with the geometry of the
set of Lagrange multipliers or if the reader prefers KKT multipliers of the problem
at a local minimizer of (MP). The KKT multiplier set at a feasible point x0 of (MP)
is given as
K K T (x0 ) = (λ, μ) ∈ Rm
+ × R : ∇x L(x 0 , λ, μ) = 0, λ, g(x) = 0 .
k
where
m
k
L(x, λ, μ) = f (x) + λi gi (x) + μ j ∇h j (x),
i=1 j=1
denotes the Lagrangian function associated with (MP). We shall present below the
following result whose proof can be completed by the reader without much difficulty.
Theorem 6 Let x0 be a local minimizer of the problem (MP). Then BCQ holds at
x0 if and only if the set K K T (x0 ) is bounded.
256 J. Dutta
5 Conclusions
This expository article has been written keeping in mind the young researchers and
graduate students. The key aim was to show how non-smooth analysis lies at the
heart of modern optimization. Of course a lot of studies have been carried out seek-
ing connections between non-smooth analysis and optimality conditions (see, for
example, Borwein and Zhu [10]) our aim here is to bring out the central theme of the
issue like the constraint qualification BCQ and also it’s geometrical significance. We
also show that though non-smooth analysis provides a unifying theme in the study
of optimality conditions linear programming problems seem to be outside that uni-
fying framework. There had been debates over which is the most important class of
optimality conditions in modern optimization. One that has been given by Fritz John
or that by Karush, Kuhn, and Tucker. Pourciau [20], for example, considers that the
Fritz John conditions are more fundamental and thus of more value as the KKT con-
ditions can be derived from them by assuming simple conditions on the constraints.
However one needs to know that even for linear programming problems the Fritz
John conditions can hold at points which are just feasible and not optimal. For linear
programming problems as we know that KKT conditions always hold at optimal
points and not at non-optimal feasible points. Thus the KKT conditions are more
important from the practical point of view. The importance of the KKT conditions
can be gauged from the following special class of nonlinear programming problems
called mathematical programming problem with complementarity constraints called
MPCC problems in short. An MPCC problem is given as follows. We shall now
formally define the notion of BCQ for the problem (MP),
min f (x)
gi (x) ≤ 0, i = 1, . . . , m
h j (x) = 0, j = 1, . . . , k,
H (x) ≥ 0,
G(x) ≥ 0,
H (x), G(x) = 0
conditions can lead us to one of the two conclusions or both. The feasible point is
not a local minimizer or some underlying qualification condition like BCQ has failed
or both. We hope that this exposition will provide the reader with some interesting
insights into both the fundamentals of non-smooth analysis and its role in optimality
conditions.
Acknowledgements The author is grateful to the financial assistance from the project titled: “First
Order Methods in Scalar and Vector Optimization” funded by the Science and Engineering research
Board of the Government of India. The author is also grateful to the anonymous referees whose
comments have led to the improvement in the presentation.
References
1. Bazaraa, M.S., Sherali, H.D., Shetty C.M.: Nonlinear Programming: Theory and Algorithms.
Wiley, New York (1993)
2. Brinkhuis, J., Tikhomirov, V.: Optimization: Insights and Applications. Princeton University
Press (2005)
3. John, F.: Extremum problems with inequalities as subsidiary conditions, Studies and Essays
Presented to R. Courant on his 60th Birthday, 8 Jan. 1948, pp. 187-204 (1948)
4. Kuhn, H.W., Tucker, A.W.: Nonlinear programming. In: 2nd Berkeley Symposium, pp. 481–
492 (1951)
5. Clarke, F.H.: Optimization and Nonsmooth Analysis. Wiley-Interscience, New York (1983)
6. Clarke, F.H., Ledyaev, Y.S., Stern, R.J., Wolenski P.R.: Nonsmooth Analysis and Control
Theory, vol. 178. Springer Science and Business Media (2008)
7. Mordukhovich, B.S.: Variational analysis and generalized differentation, vol. I: Basic Theory,
vol. II: Applications (2006)
8. Chikrii, A.A.: A review of the book variational analysis and generalized differentiation (vol. 1,
basic theory and vol. 2, applications) by BS Mordukhovich. Cybernetics and Systems Analysis,
vol. 42, no. 5, p. 769 (2006)
9. Kruger, A.Y.: Properties of generalized differentials. Sib. Math. J. 26(6), 822–832 (1985)
10. Borwein, J.M., Qiji J.Z.: Variational techniques in convex analysis. Tech. Var. Anal. 111–163
(2005)
11. Rockafellar, R.T., Wets R.J.B. .: Variational Analysis, vol. 317. Springer Science and Business
Media (2009)
12. Demyanov, V.F., Rubinov A.M.: Constructive Nonsmooth Analysis, vol. 7. Peter Lang Pub
Incorporated (1995)
13. Loewen, P.D.: Optimal control via nonsmooth analysis, No. 2. American Mathematical Soc
(1993)
14. Vinter, R.B.: Optimal Control, Birkhauser (2001)
15. Hiriart-Urruty, J.B., Lemaréchal, C.: Fundamentals of Convex Analysis. Springer Science and
Business Media (2004)
16. Güler, O.: Foundations of Optimization, vol. 258. Springer Science and Business Media (2010)
17. Dutta, J.: Optimality conditions for bilevel programming: an approach through variational
analysis. In: Generalized Nash Equilibrium Problems. Bilevel Programming and MPEC, pp.
43–64. Springer, Singapore (2017)
18. Dempe, S.: Foundations of Bilevel Programming. Springer Science and Business Media (2002)
19. Dempe, S., Dutta, J., Mordukhovich, B.S.: New necessary optimality conditions in optimistic
bilevel programming. Optimization 56(5–6), 577–604 (2007)
20. Pourciau, B.H.: Modern multiplier rules. Am. Math. Month. 87(6), 433–452 (1980)
Solving Multiobjective Environmentally
Friendly and Economically Feasible
Electric Power Distribution Problem
by Primal-Dual Interior-Point Method
1 Introduction
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 259
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_18
260 Jauny et al.
distribution (EFEFD) of electric power, the generating units must provide outputs
that fulfill the demand, cause the least amount of pollution, and emit the least amount
of atmospheric emissions.
EFEFD problem of electric power systems aims to obtain generating unit out-
puts in which the operating system cost remains minimal. Furthermore, the system
constraints must be met while reducing pollution and emissions. This type of appli-
cation involves optimizing multiple conflicting nonlinear objectives simultaneously.
Multiobjective optimization problems (MOPs) consider to optimize several conflict-
ing objectives simultaneously. Therefore, most often, a single solution that performs
well for each objective function does not exist. In solving MOP problems, some-
times decision makers will come up with a compromise solution by analyzing a set
of points that are representative of the entire Pareto set. A feasible point is called
Pareto optimal (nondominated point [2]) if it is impossible to improve one objective
without sacrificing another. When solving a MOP, the goal is to identify all possible
Pareto optimal solutions.
MOPs have been solved through several scalarization techniques [7] over the last
few years. The reputed classical methods such as weighted sum [5, 6, 8], ε-constraint
[4], normal boundary intersection [1, 10], physical programming [9], cone method [3,
14], etc., are known to find the Pareto optimal solutions. However, these methods are
not able to yield a complete Pareto front. Recently, Pascoletti–Serafini [13] technique
has been established that can generate all Pareto solutions.
In this article, we propose a primal-dual interior-point method (PDIPM) combined
with the Pascoletti–Serafini [13] technique to find nondominated points of EFEFD
problem. The proposed method exploits the efficiencies of the Pascoletti–Serafini
[13] and PDIPM [11] in solving EFEFD problem. As a consequence, the proposed
method captures the discrete set of nondominated points.
This paper follows the following structure. In Sect. 2, we describe the EFEFD
problem in detail. In Sect. 3, a brief review of Pascoletti–Serafini technique is dis-
cussed. Then we give a formulation of interior-point method (IPM) for a nonlinear
problem, which is formulated into Sect. 2, and find the search direction formulas. In
Sect. 4, a merit function is presented. In Sect. 5, numerical results of EFEFD problem
are presented. Finally, Sect. 6 ends with a few concluding remarks.
2 EFEFD Problem
In EFEFD problem, the objective is to minimize two conflicting objectives fuel cost
and emissions subject to an equality and bound constraints. Let p1 , p2 , . . . , pn be
the power outputs of the generators G 1 , G 2 , . . . , G n , and f c (P) denotes the cost
function of the generators. Thus, the total fuel cost f c (P) is given by the following
expression:
n
f c (P) = Pi + Qi pi + Ri pi2 , (1)
i=1
Solving Multiobjective Environmentally Friendly and Economically … 261
n
f e (P) = 10−2 αi + βi pi + γi pi2 + ξi eλi pi , (2)
i=1
Also, the entire power generation will be equal to the sum of two quantities, PDemand
and Ploss , where PDemand is the total demand and Ploss is the power loss in the trans-
mission lines. Hence,
n
pi = PDemand + Ploss . (4)
i=1
In this paper, we have taken six generator data from [12]. The data of cost coef-
ficients and emission characteristic coefficients for six generators are provided in
sup
Tables 1 and 2. Also, piinf = 10, pi = 120, and PDemand + Ploss = 283. Therefore,
the mathematical EFEFD problem is
To obtain the weakly efficient and efficient solutions [13] of EFEFD problem (5), we
formulate it into a parametric scalar optimization problem with the help of Pascoletti
and Serafini scalarization technique [13]. The formulation of Pascoletti and Serafini
scalar optimization problem with parameters a, r ∈ R2 , with respect to the ordering
cone K = R2 :
minimize y,t t
subject to G (y) = 0
(7)
a + tr − F (y) ≥ 0
l ≤ y ≤ u, t ∈ R.
By solving the scalar optimization problem (6) for various values of a and r , one
can be obtained the Pareto front of EFEFD problem. To simplicity, we take a = F 0 ,
where F 0 = (F10 , F20 ) is the ideal point of EFEFD problem. For EFEFD problem,
the ideal point is (599.22, 0.19) .
Now, problem (6) can be rewritten as
minimize ȳ F( ȳ)
subject to G ( ȳ) = 0
(8)
H ( ȳ) ≥ 0,
In the following section, a PDIPM is discussed to solve (8). We formulate problem (8)
into barrier problem and then Karush–Kuhn–Tucker (KKT) conditions are derived.
Thereafter, PDIPM takes advantage of the Newton method to obtain the solution of
the KKT system. An overview of the method is described below.
We formulate the problem (8) by introducing slack variables vector s = (s1 ,
s2 , . . . , s5 ) as follows :
Solving Multiobjective Environmentally Friendly and Economically … 263
minimizex̄,s F( ȳ)
subject to G ( ȳ) = 0
(9)
H ( ȳ) − s = 0
s ≥ 0.
The followings are the first-order KKT conditions for barrier problem (10):
⎫
∇ ȳ F( ȳ) − ∇ ȳ G ( ȳ) y − ∇ ȳ (H ( ȳ) − s) z = 0 ⎪ ⎪
⎪
⎪
−μS −1 e + z = 0 ⎪ ⎪
⎬
G ( ȳ) = 0 ⎪ (11)
⎪
H ( ȳ) − s = 0 ⎪ ⎪
⎪
⎪
⎭
z ≥ 0,
For a fixed μ > 0, the step Δd = (Δ ȳ, Δs, Δx, Δz) at the point ( ȳ, s, x, z) is
obtained by applying Newton method to the system (11) and solving the following
primal-dual system:
⎡ 2 ⎤⎡ ⎤ ⎡ ⎤
∇ ȳ ȳ L 0 −(A( ȳ)) −(B( ȳ)) Δ ȳ ∇ ȳ F( ȳ) − (A( ȳ)) x − (B( ȳ)) z
⎢ 0 Z 0 S ⎥ ⎢ Δs ⎥ ⎢ Sz − μe ⎥
⎢ ⎥⎢ ⎥ = − ⎢ ⎥,
⎣ A( ȳ) 0 0 0 ⎦ ⎣Δx ⎦ ⎣ G ( ȳ) ⎦
B( ȳ) −I 0 0 Δz H ( ȳ) − s
(12)
A( ȳ) and B( ȳ) are the Jacobian matrix of the function g( ȳ) and h( ȳ) − s, respec-
tively.
264 Jauny et al.
The matrix on the left of (12) is not symmetric. However, it can be easily sym-
metrized by multiplying the first equation by −1 and the second equation by −S −1 .
Accordingly, we get the following to modify primal-dual system (12)
⎡ ⎤⎡ ⎤ ⎡ ⎤
−∇ ȳ2ȳ L 0 (A( ȳ)) (B( ȳ)) Δ ȳ ∇ ȳ F( ȳ) − (A( ȳ)) x − (B( ȳ)) z
⎢ 0 −S −1 Z −I ⎥ ⎢ ⎥ ⎢ z − μS −1 e ⎥
⎢ 0 ⎥ ⎢ Δs ⎥ = ⎢ ⎥.
⎣ A( ȳ) 0 0 0 ⎦ ⎣Δx ⎦ ⎣ − G ( ȳ) ⎦
B( ȳ) −I 0 0 Δz −H ( ȳ) + s
(14)
We note that second equation of (14) can be used to eliminate Δs without produc-
ing any off-diagonal fill-in in the remaining system with the help of the following
equation:
Δs = S Z −1 z − μS −1 e + Δz . (15)
In order to solve reduced KKT system (16) one can apply a Cholesky factorization.
However, due to the indefiniteness of the matrix ∇x̄2x̄ L , we cannot apply the Cholesky
factorization. In this case, symmetric indefinite factorization (see [15]) will be the
best strategy to use. Therefore, we applied symmetric indefinite factorization to solve
the system (16).
After solving (12), we obtain the step Δd and then calculate the new iterate
(x̄ + , s + , y + , z + ) as follows:
⎫
ȳ + = ȳ + ζs Δ ȳ ⎪
⎪
⎪
s + = s + ζs Δs ⎬
, (17)
x + = x + ζz Δx ⎪ ⎪
⎪
⎭
z + = z + ζz Δz
where
ζs = max{ζ ∈ (0, 1) : s + ζ Δs ≥ (1 − η)s}
, (18)
ζz = max{ζ ∈ (0, 1) : z + ζ Δz ≥ (1 − η)z}
with η ∈ (0, 1). The steplength calculated by (18) ensures that the variables s and
y remain positive at every iterate. But, there is no guarantee of the reduction in the
objective function and convergence of the generated sequence to a minimum point.
To evaluate the progress toward optimality of the algorithm, we take the advantage
of the merit function.
Solving Multiobjective Environmentally Friendly and Economically … 265
With merit functions, steps are shortened so that an appropriate reduction toward
optimality can be made along the search direction. IPM can be viewed as methods for
solving the barrier problem (10). Therefore, we define the following merit function
in terms of barrier functions:
φν ( ȳ, s) = B( ȳ, s; μ) + ν H ( ȳ − s)22 + G ( ȳ)22 , (19)
∇ F Δp + (σ/2)Δp ∇ ȳ2ȳ L Δp
ν≥ , (20)
(1 − ρ) (H ( ȳ) − s + G ( ȳ))
After computing the step Δd from (12), we compute the steplength αs and
αz with the help of (18) and reach the next iteration by using (17). Now, if the
new point ( ȳ + , s + , x + , z + ) is able to reduce the merit function φν ( ȳ + , s + ), then
( ȳ + , s + , x + , z + ) is accepted and the algorithm continues. If the merit function
φν ( ȳ + , s + ) does not reduce, then the new point ( ȳ + , s + , x + , z + ) is not accepted.
In this case, we choose α ∈ [0, α max ], where α max = min{αs , αz } so that the follow-
ing Armijo condition satisfied:
φν ( ȳ + , s + ) − φν ( ȳ, s) ≤ δα ∇φν ( ȳ + , s + ) Δp, (21)
In IPMs, the choice of the barrier parameter μ is critical since optimality is attained
when the barrier parameter approaches zero. In addition, if the value of μ is set to
decline slowly, many iterations will be needed to reach convergence. However, if it
is reduced rapidly, some slack variables or Lagrange multipliers will approach zero
very soon. The following technique of updating μ has demonstrated the effectiveness
in practice:
s y
μ=ς , (22)
6
266 Jauny et al.
Ξ ( ȳ, s, x, z) = max ∇ ȳ F( ȳ) − (A( ȳ)) y − (B( ȳ)) z, Sx − μe, H ( ȳ) − s, G ( ȳ) .
(24)
1: Initialization:
Set Pareto set D ← ∅
Provide an initial point w (0) = ( ȳ (0) , s (0) , x (0) , z (0) ) with ȳ (0) > 0, s (0) > 0, x (0) > 0, z (0) >
0
Give the accuracy precision ε > 0 and choose δ ∈ (0, 1)
Choose initial values for μ > 0
Set k ← 0
2: Main Part:
3: for i = 1 : N do
4: Choose randomly a ∈ R p , r ∈ R p \ {0}
5: while Ξ ( ȳ (k) , s (k) , x (k) , z (k) ) ≥ ε do
6: Compute Δd (k) by solving the system (12)
7: Compute steplength αs and αz with the help of (18)
8: Set ȳ (k+1) = ȳ (k) + ζs Δ ȳ (k) , s (k+1) = s (k) + ζs Δs (k) , x (k+1) = x (k) +
(k)
ζz Δx , z (k+1) (k)
= z + ζz Δz (k)
12: Set ȳ (k+1) = ȳ (k) + ζ Δ ȳ (k) , s (k+1) = s (k) + ζ Δs (k) , x (k+1) = x (k) + ζ Δx (k) ,
z (k+1) = z (k) + ζ Δz (k)
13: end if
14: Update barrier parameter μ by using (22)
15: k ←k+1
16: end while
17: Calculate F ( ȳ)
18: Update D ← D {F ( ȳ)}
19: end for
20: return The set D (a discrete approximation of the whole Pareto set)
Solving Multiobjective Environmentally Friendly and Economically … 267
5 Numerical Results
6 Conclusion
In this paper, EFEFD problem has been solved using IPM. To solve EFEFD problem,
we have used a Pascoletti and Serafini scalarization technique to transform EFEFD
problem into a parametric scalar optimization problem. Thereafter, the parametric
scalar optimization problem has been solved with the help of IPM by changing the
parameters a and r . The results in Sect. 5 have shown that the proposed algorithm
efficiently solves the EFEFD problem.
Acknowledgements Authors are truly thankful to the reviewers for their comments on the paper.
Jauny gratefully acknowledges a Senior Research Fellowship from the Council of Scientific and
Industrial Research, India (File No. 09/1217(0025)2017-EMR-I), to perform this research work.
Debdas Ghosh acknowledges the research grant MATRICS (MTR/2021/000696) from SERB, India,
to carry out this research work.
References
1. Das, I., Dennis, J.E.: Normal-boundary intersection: a new method for generating the Pareto
surface in nonlinear multicriteria optimization problems. SIAM J. Optim. 8(3), 631–657 (1998)
2. Ehrgott, M.: Multicriteria Optimization, 2nd edn. Springer, Berlin, Heidelberg (2005)
3. Ghosh, D., Chakraborty, D.: A new Pareto set generating method for multi-criteria optimization
problems. Oper. Res. Lett. 42(8), 514–521 (2014)
Solving Multiobjective Environmentally Friendly and Economically … 269
Debabrata Datta
D. Datta
Former Scientist, Bhabha Atomic Research Centre, & Head, RP&AD, Mumbai 400085, India
(B)
Department of Information Technology, Heritage Institute of Technology, Kolkata, WB 700017,
India
e-mail: [email protected]
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 271
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_19
272 D. Datta
1 Introduction
The BAT algorithm is a metaheuristics algorithm and has been developed by Yang
[6, 7]. The algorithm is based on the behaviour of bats like echolocation character
and this characteristics is used to sense the distance. Bats typically emit short, loud
sound impulses and listen to the bounced-back echo from an obstacle or prey while
hunting at night. A bat can use its special auditory mechanism to identify the size
and position of an object. The steps of the bat algorithm are as follows:
Step 1: Initialize the algorithm parameters.
Step 2: Update the best global position x ∗ , pulse frequency, velocity, and position
of the ith bat using the expression
vit+1 = vit + xit + x ∗ f i , (2)
Step 3: If β(randomnumber ) is > ri , then write down the new solution as.
xnew = xold + At , where At represents the average loudness of all bats at time t
and liesintherange[−1, 1].
Step 4: Accept the new solution, provided the random number is lower than Ai
and f (xi ) > f (x ∗ ). Failure to the condition as prescribed, update Ai and ri using
the expression
rit = ri0 1 − e−γ t (5)
where, rit andri0 are the pulse rate at time t and at the initial phase. The constant
γ [0, 1].
Step 5: On the basis of their fitness sort the bats and find the optimal solution, x ∗
of the present iteration.
Step 6: Return to step 2 till the maximum iteration is reached. Finally, output the
global optimal solution.
A case study of the bat algorithm for function optimization is presented in Sect. 4.
where x pr evious is the existing pitch or solution from HM and xnew is the new pitch
after adjustment. Random number generator, ε lies in the range of [−1, 1]. We can
say that pitch adjustment is equivalent to mutation operator in a genetic algorithm.
The PAR can be assigned for controlling the degree of adjustment. A lower value
of PAR with a narrow bandwidth can slow down the convergence of HS due to a
limitation in the exploration of a small subspace of the complete search space.
3.3 Randomization
In order to increase the diversity of the solution, we need randomization in HS. The
usage of randomization drives the system further to explore various diverse solutions
so as to find global optimality.
Algorithm:
1. Construct a Harmony Memory
2. Improvise a new Harmony with Experience (HM) or Randomness (rather than Gradient)
3. If the new Harmony is better, include it in Harmony Memory
4. Repeat Step 1 and Step 2 till goal is achieved
Pseudocode:
Begin
4 Case Study
N −1
2
f (x) = 100 xn+1 − xn2 + (1 − xn )2 (7)
n=1
10 -1
Best Cost
-2
10
10 -3
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Iteration
One of the possible ways to treat cancer is radiation therapy. Treatment of cancer
using radiation by Cobalt-60, proton, and X-ray is known as radiation oncology. In
the field of radiation oncology, we use CT/MRI machines to allow radiation to pass
through cancerous cells or affected organs of the patient. Typically, multiple beams
of different radiation doses are used from different sides and different angles. In
this kind of problem, the primary aim is to decide the strength of the radiation dose
coming from the particular beam to use to achieve sufficient damage to the target
tumour and limit the damage to healthy tissues. Figure 4 presents the orientation of
beam therapy to treat the cancerous tumour.
Our goal is to optimize the orientation of the beam to identify the target volume.
In practice, this concept is implemented in Bhabhatron, an indigenously designed
Cobalt-60 therapeutic machine, and the resulting hardware is known as a multileaf
collimator. The data used in radiotherapy is presented in Table 1.
Decision variables D1 and D2 represent the dose strength for beam 1 and beam 2,
respectively. Mathematical statement of the problem is as follows:
subject to
where
vi (t)—real-valued velocity vector of i-th bat,
xi (t)—real-valued position vector of i-th bat,
Q i —pulsation frequency of i-th bat,
α, γ , Q min , Q max —constant.
In this case study, we have used the BAT algorithm to optimize sphere and Rosen-
brock functions with comparative results using HS. Results are tabulated in Tables 2
and 3. d
We obtain the best estimate (optimal value) of sphere function f (x) = i=1 xi2
using HS as 1.2E + 1 with decision variables x(1), x(2), and x(3) are equal to 2.0
Results of optimization of Rosenbrock function using HS are good in agreement
with that using BAT algorithm. However, the BAT algorithm is computationally
intensive compared to the harmony search.
Optimization Methods Using Music-Inspired Algorithm and Its … 281
5 Conclusions
In this article, we have presented the harmony search algorithm and nature-inspired
BAT algorithm. Nature-inspired algorithms are based on metaheuristics. In music-
inspired optimization (harmony search), we have learned the importance of music
(harmony, pitch adjustment) for solving optimization problems. In nature-inspired
optimization, we have learned various social behaviour which can be converted into
an optimization algorithm (metaheuristics). Music-inspired-based Harmony search
innovates a new optimization algorithm. Several case studies are presented to illus-
trate the working methodology of the harmony search algorithm and bat algorithm.
In future, our work will be towards the usage of the harmony search algorithm in
data science rather than data optimization.
282 D. Datta
References
1. Spall, J.C.: Introduction to Stochastic Search and Optimization: Estimation. Wiley, Simulation
and Control (2003)
2. Cui, Z., Sun, B., Wang, G., Xue, Y., Chen, J.: A novel oriented cuckoo search algorithm to
improve DV-hop per-formance for cyber-physical systems. J. Parallel Distrib. Comput. 103,
42–52 (2016)
3. Yang, X.S., Deb, S.: Engineering optimization by cuckoo search. Int. J. Math. Model. Numer.
Optim. 1, 4, 330–343 (2010)
4. Yang, X.-S.: A new metaheuristic bat-inspired algorithm. Nat. Insp. Cooper. Strat. Optim. 284,
65–74 (2010)
5. Geem, Z.W., Kim, J.H., Loganathan, G.V.: A New Heuristic Optimization Algorithm: Harmony
Search. SIMULATION 76, 60–68 (2001)
6. Yang, X.S.: Nature-inspired Metaheuristic Algorithms. Luniver Press, Backington, UK (2008)
7. Yang, X.S., Gandomi, A.H.: Bat algorithm: a novel approach for global engineering optimiza-
tion. Eng. Comput. 29, 464–483 (2012)
8. Omran, M., Mahdavi: Global-best harmony search. Appl. Math. Comput. 198, 643–656 (2008)
9. Geem, Z.W.: Music-Inspired Harmony Search Algorithm. Springer, Heidelberg, Germany
(2009)
10. Wolpert, D.H., Macready, W.G.: No free lunch theorems for optimization. IEEE Trans. Evol.
Comput. 1:67–82 (1997)
11. Back, T., Fogel, D., Michalewicz Z.: Handbook of Evolutionary Computation. Oxford
University Press (1997)
12. Dervis, K., Bahriye, B.: A powerful and efficient algorithm for numerical function optimization:
Artificial Bee Colony (ABC) algorithm J. Glob. Optim. 39, 459–471 (2007)
On Mathematical Programs
with Equilibrium Constraints Under
Data Uncertainty
1 Introduction
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 283
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_20
284 V. Laha and L. Pandey
investigated several CQs for MPECs. Guo et al. [19] derived second-order optimal-
ity conditions for MPECs. Duality results for MPECs were produced by Pandey and
Mishra [39] along with Guo et al. [18].
A mathematical modeling technique called robust optimization (RO) is used to
address issues with the uncertain objectives or the uncertain data in the feasible
region (see, e.g. [1–4, 16]). Jeyakumar et al. [20] studied robust duality under data
uncertainty. To find efficient robust solutions, Kuroiwa and Lee [22] constructed the
optimality theorems. Strong duality minimax results were obtained by Jeyakumar et
al. [21] under robustness. Lee and Kim [24] developed Wolfe duality results for a RO
problem. Soleimanian and Jajaei [40] worked on robust nonlinear optimization with
conic representable uncertainty sets. Chuong [6] dealt with robust multiobjective
optimization problems with nonsmooth nonlinear data. Lee and Kim [25] provided
optimality and duality results for robust non-smooth multiobjective optimization.
Fakhar et al. [9] analysed robust portfolio optimization. Theorems for robust semiin-
finite optimization were found by Lee and Lee [26]. Chen et al. [5] achieved results
for situations involving robust non-smooth multiobjective optimization with restric-
tions. For nonsmooth robust multiobjective optimization problems, Fakhar et al. [10]
researched approximative solutions. Wang et al. [42] used image space analysis to
study general robust dual problems.
The objective is to solve MPECs with uncertainty using RO. The outline is: we
review definitions and outcomes from RO in Sect. 2. In Sect. 3, we deal with an opti-
mization problem with mixed assumptions of deterministic inequality and equality
constraints along with inequality constraints with uncertainties denoted by MUP.
We create robust FJ criteria for the given problem and utilize them to derive robust
Karush-Kuhn-Tucker (KKT) conditions under the premise that there is no nonzero
abnormal multiplier CQ (NNAMCQ). Additionally, we find that the MUP has suffi-
cient optimality criteria when convexity assumptions are made. The results of Sect. 3
are used in Sect. 4 to derive robust optimality criterion for the MPECs with mixed
assumptions of certain constraints and uncertain constraints in the feasible region
denoted by UMPEC. We define a suitable constraint qualification using a nonlinear
program connected to the UMPEC, from which we derive necessary and sufficient
weak stationary optimality requirements. The definitions and relationships of a num-
ber of other stationary conditions connected to the UMPEC, such as the C-, A-, M-,
and S-stationary conditions, are discussed. Section 5 wraps up the findings of this
paper and explores several potential directions for further research.
2 Preliminaries
where
p
g
∗
0 f(ζ ) + i 1 gi (ζ ∗ , ui ) = 0,
i=1
g
i gi (ζ ∗ , ui ) = 0, ∀i ∈ P.
where
F2 := {ζ ∈ Rn : gi (ζ, ui ) ≤ 0, ∀ui ∈ Ui , ∀i ∈ P,
φi (ζ ) ≤ 0, ∀i ∈ Q,
hi (ζ ) = 0, ∀i ∈ R}.
P1 (ζ ∗ ) := {i ∈ P : ∃ui ∈ Ui s. t. gi (ζ ∗ , ui ) = 0}, P2 := P \ P1 (ζ ∗ ),
Q1 (ζ ∗ ) := {i ∈ Q : φi (ζ ∗ ) = 0}, Q2 := Q \ Q1 (ζ ∗ ),
and
Ui0 = {ui ∈ Ui : gi (ζ ∗ , ui ) = 0}, ∀i ∈ P1 (ζ ∗ ).
The FJ condition by Mangasarian and Fromovitz [32] and by Lee and Son [29,
Theorem 2.4] lead to the following FJ condition for MRP.
Theorem 2 (FJ condition for MRP) Let ζ ∗ ∈ F2 be a local minimizer of MRP
and let gi (ζ, ·) be concave on Ui for every ζ ∈ Rn and for every i ∈ P. Then, there
g φ h
exists 0 ≥ 0, i ≥ 0 (i ∈ P), i ≥ 0 (i ∈ Q), i ∈ R (i ∈ R), not all zero, and
ui ∈ Ui (i ∈ P) such that
p
g
q
φ
r
h
0 f(ζ ∗ ) + i 1 gi (ζ ∗ , ui ) + i φi (ζ ∗ ) + i hi (ζ ∗ ) = 0,
i=1 i=1 i=1
g ∗
i gi (ζ , ui ) = 0, ∀i ∈ P,
φ
i φi (ζ ∗ ) = 0, ∀i ∈ Q.
On Mathematical Programs with Equilibrium Constraints Under Data Uncertainty 287
By the FJ condition for MRP, if 0 is never zero, then it can be taken as 1 which
will lead to KKT-type robust optimality condition. We introduce the following
extended NNAMCQ (ENNAMCQ) for MUP which will solve the purpose based
on the NNAMCQ introduced by Ye [43].
288 V. Laha and L. Pandey
p
g
q
φ
r
h
f(ζ ) + i 1 gi (ζ ∗ , ui ) + i φi (ζ ∗ ) + i hi (ζ ∗ ) = 0,
i=1 i=1 i=1
g ∗
i gi (ζ , ui ) = 0, ∀i ∈ P,
φ
i φi (ζ ∗ ) = 0, ∀i ∈ Q.
Proof According to the theorem’s premise, there will be scalars that satisfy the
FJ condition of Theorem 2 at ζ ∗ ∈ F2 . Since ENNAMCQ is satisfied at ζ ∗ , for
g g φ h
0 = 0, we will have i = 0 (i ∈ P), i = 0 (i ∈ Q), i = 0 (i ∈ R) which is
inconsistent with the observation that all scalar multipliers are not zero. Hence,
g
0 > 0 and this gives the required KKT condition at ζ ∗ .
We specify the upcoming index sets that will be utilized to derive sufficient opti-
mality condition for MU P :
g φ
P + := {i ∈ P : i > 0}, Q+ := {i ∈ Q : i > 0},
h h
R+ := {i ∈ R : i > 0}, R− := {i ∈ R : i < 0}.
Theorem 4 (Condition for robust sufficient optimality in MUP) Assume that for
g φ h
some ui ∈ Ui (i ∈ P), i ≥ 0 (i ∈ P), i ≥ 0 (i ∈ Q) and i ∈ R (i ∈ R) the
∗
KKT condition of Theorem 3 for MRP is fulfilled at ζ . If the functions f, gi (., ui )(i ∈
P + ), φi (i ∈ Q+ ), hi (i ∈ R+ ), −hi (i ∈ R− ) are convex at ζ ∗ ∈ F2 , then ζ ∗ is a
global minimizer of MRP.
Proof Let ζ ∈ F2 . Since f is convex, therefore
f(ζ ) − f(ζ ∗ )
T
p
g
q
φ
r
h
≥− i 1 gi (ζ ∗ , ui ) + i φi (ζ ∗ ) + i hi (ζ ∗ ) (ζ − ζ ∗ ),
i=1 i=1 i=1
∀ζ ∈ F2 .
1 gi (ζ ∗ , ui )T (ζ − ζ ∗ ) ≤ gi (ζ, ui ) − gi (ζ ∗ , ui ), ∀ζ ∈ F2 , ∀i ∈ P + .
Since gi (ζ ∗ , ui ) = 0 (i ∈ P + ), therefore
1 gi (ζ ∗ , ui )T (ζ − ζ ∗ ) ≤ 0, ∀ζ ∈ F2 , ∀i ∈ P + . (1)
Similarly
φi (ζ ∗ )T (ζ − ζ ∗ ) ≤ 0, ∀ζ ∈ F2 , ∀i ∈ Q+ , (2)
hi (ζ ∗ )T (ζ − ζ ∗ ) ≤ 0, ∀ζ ∈ F2 , ∀i ∈ R+ , (3)
hi (ζ ∗ )T (ζ − ζ ∗ ) ≥ 0, ∀ζ ∈ F2 , ∀i ∈ R− . (4)
g φ h
Multiplying (1)–(4) by i > 0 (i ∈ P + ), i > 0 (i ∈ Q+ ), i > 0(i ∈ R+ ),
h
and i < 0 (i ∈ R− ), respectively, and adding, we get
⎛ ⎞T
p q
r
g ∗ φ ∗ h
⎝ i 1 gi (ζ , ui ) + i φi (ζ ) + i hi (ζ )⎠ (ζ − ζ ∗ ) ≤ 0, , ∀ζ ∈ F2 ,
∗
where
Ω := {ζ ∈ Rn : gi (ζ, ui ) ≤ 0, ∀ui ∈ Ui , ∀ i ∈ P,
φi (ζ ) ≤ 0, ∀i ∈ Q,
hi (ζ ) = 0, ∀i ∈ R,
Gi (ζ ) ≥ 0, ∀i ∈ S,
Hi (ζ ) ≥ 0, ∀i ∈ S,
Gi (ζ )Hi (ζ ) = 0, ∀i ∈ S}.
.
Definition 4 A point ζ ∗ ∈ Ω is a robust global minimizer of U M P EC iff ζ ∗ is a
global minimizer of RMPEC, that is, f(ζ ) ≥ f(ζ ∗ ) for all ζ ∈ Ω. A point ζ ∗ is a
robust local minimizer of U M P EC iff there exist > 0 such that f(ζ ) ≥ f(ζ ∗ ) with
ζ ∈ Ω ∩ B(ζ ∗ , ).
In addition to the index sets defined earlier, we need the following index sets
related to the equilibrium constraints
P1 (ζ ∗ ) := {i ∈ P : ∃ui ∈ Ui , s. t. gi (ζ ∗ , ui ) = 0},
Q1 (ζ ∗ ) := {i ∈ Q : φi (ζ ∗ ) = 0},
I0+ (ζ ∗ ) := {i ∈ S : Gi (ζ ∗ ) = 0, Hi (ζ ∗ ) > 0},
I00 (ζ ∗ ) := {i ∈ S : Gi (ζ ∗ ) = 0, Hi (ζ ∗ ) = 0},
I+0 (ζ ∗ ) := {i ∈ S : Gi (ζ ∗ ) > 0, Hi (ζ ∗ ) = 0},
I++ (ζ ∗ ) := {i ∈ S : Gi (ζ ∗ ) > 0, Hi (ζ ∗ ) > 0},
Ui0 := {ui ∈ Ui : gi (ζ ∗ , ui ) = 0}, ∀i ∈ P1 (ζ ∗ ).
292 V. Laha and L. Pandey
Since the feasible set of RTNLP(ζ ∗ ) is a subset of the feasible set of RMPEC.
Therefore, if ζ ∗ is a local minimizer of RMPEC, then ζ ∗ is also a local minimizer of
corresponding RTNLP(ζ ∗ ) and RTNLP(ζ ∗ ) may be used to define a suitable variant
of ENNAMCQ for RMPEC.
Definition 5 The R M P EC satisfies RMPEC-ENNAMCQ at ζ ∗ iff RTNLP(ζ ∗ )
satisfies the ENNAMCQ at ζ ∗ , that is, for any ui ∈ Ui (i ∈ P), one has
⎧ q r
⎪ p g ∗ φ ∗ h ∗
⎪
⎪
⎪ i=1 i 1 gi (ζ , ui ) + i=1 i φi (ζ ) + i=1 i hi (ζ )
⎪
⎪ G ∗ G
− i∈I+0 ςi Gi (ζ ) − i∈I0+ ∪I00 ςi Gi (ζ ) ∗
⎪
⎪
⎪
⎪
⎪
⎪
⎪ − i∈I0+ ςiH Hi (ζ ∗ ) − i∈I+0 ∪I00 ςiH Hi (ζ ∗ ) = 0,
⎪
⎨ g φ
i ≥ 0 (i ∈ P1 (ζ ∗ )), i ≥ 0 (i ∈ Q1 (ζ ∗ )), ςiG ≥ 0 (i ∈ I+0 ), ςiH ≥ 0, (i ∈ I0+ )
⎪ =⇒
⎪
⎪
⎪
⎪
⎪ g φ h
⎪ i = 0 (i ∈ P1 (ζ ∗ )), i = 0 (i ∈ Q1 (ζ ∗ )), i = 0 (i ∈ R),
⎪
⎪
⎪
⎪
⎪
⎪ ςiG = 0 (i ∈ I+0 ), ςiG = 0 (i ∈ I0+ ∪ I00 ),
⎪
⎩ ς H = 0 (i ∈ I ), ς H = 0 (i ∈ I ∪ I ).
i 0+ i +0 00
(RMPEC-ENNAMCQ)
We can now establish the KKT criteria for RMPEC.
Based on the above theorem and following the notion of stationary points for
MPEC (see, e.g. [44]), we may now define weak stationary points for the RMPEC.
f(ζ ) − f(ζ ∗ )
T
p
g
q
φ
r
h
≥− i 1 gi (ζ ∗ , ui ) + i φi (ζ ∗ ) + i hi (ζ ∗ ) (ζ − ζ ∗ )
i=1 i=1 i=1
T
s
+ [ςiG Gi (ζ ∗ ) + ςiH Hi (ζ ∗ )] (ζ − ζ ∗ ),
i=1
∀ζ ∈ Ω.
(6)
By the convexity of gi (., ui ) at (ζ ∗ , ui ) for every i ∈ P + , one has
1 gi (ζ ∗ , ui )T (ζ − ζ ∗ ) ≤ gi (ζ, ui ) − gi (ζ ∗ , ui ), ∀ζ ∈ Ω ∀i ∈ P + .
Since gi (ζ ∗ , ui ) = 0 (i ∈ P + ), therefore
1 gi (ζ ∗ , ui )T (ζ − ζ ∗ ) ≤ 0, ∀ζ ∈ Ω ∀i ∈ P + . (7)
Similarly
φi (ζ ∗ )T (ζ − ζ ∗ ) ≤ 0, ∀ζ ∈ Ω ∀i ∈ Q+ , (8)
hi (ζ ∗ )T (ζ − ζ ∗ ) ≤ 0, ∀ζ ∈ Ω, ∀i ∈ R+ , (9)
hi (ζ ∗ )T (ζ − ζ ∗ ) ≥ 0, ∀ζ ∈ Ω, ∀i ∈ R− , (10)
+ +
Gi (ζ ∗ )T (ζ − ζ ∗ ) ≥ 0, ∀ζ ∈ Ω, ∀i ∈ I0+ ∪ I00 , (11)
+ +
Hi (ζ ∗ )T (ζ − ζ ∗ ) ≥ 0, ∀ζ ∈ Ω ∀i ∈ I+0 ∪ I00 . (12)
g φ h
Multiplying (7)–(12) by i > 0 (i ∈ P + ), i > 0 (i ∈ Q+ ), i > 0(i ∈
h + +
R+ ), −i > 0 (i ∈ R− ), ςiG > 0 (i ∈ I0+ ∪ I00 ), and ςiH > 0 (i ∈
+ +
I+0 ∪ I00 ), respectively, and summing, we obtain
T
p
g
q
φ
r
h
i 1 gi (ζ ∗ , ui ) + i φi (ζ ∗ ) + i hi (ζ ∗ ) (ζ − ζ ∗ )
i=1 i=1 i=1
T
s
− [ςiG Gi (ζ ∗ ) + ςiH Hi (ζ ∗ )] (ζ − ζ ∗ ) ≤ 0, ∀ζ ∈ Ω,
i=1
−
Gi (ζ ∗ )T (ζ − ζ ∗ ) ≤ 0, ∀i ∈ I0+ . (13)
On Mathematical Programs with Equilibrium Constraints Under Data Uncertainty 295
−
Similarly, for every ζ close enough to ζ ∗ , by the convexity of Hi (i ∈ I+0 ), one
has
−
Hi (ζ ∗ )T (ζ − ζ ∗ ) ≤ 0, ∀i ∈ I+0 . (14)
g φ h
Multiplying (7)–(14) by i > 0(i ∈ P + ), i > 0(i ∈ Q+ ), i > 0(i ∈
h + + + +
R+ ), −i > 0(i ∈ R− ), ςiG > 0(i ∈ I0+ ∪ I00 ), ςiH > 0(i ∈ I+0 ∪ I00 ),
− H −
−ςiG > 0(i ∈ I0+ ) and −ςi > 0(i ∈ I+0 ), respectively, we get
T
p
g
q
φ
r
h
i 1 gi (ζ ∗ , ui ) + i φi (ζ ∗ ) + i hi (ζ ∗ ) (ζ − ζ ∗ )
i=1 i=1 i=1
T
s
− [ςiG Gi (ζ ∗ ) + ςiH Hi (ζ ∗ )] (ζ − ζ ∗ ) ≤ 0, ∀ζ ∈ Ω,
i=1
which implies from (5) that f(ζ ) ≥ f(ζ ∗ ) for every ζ close enough to ζ ∗ . Hence,
ζ ∗ is a local minimizer of the RMPEC.
(c) Since ζ ∗ is an interior point wrt Ω ∩ {ζ : Gi (ζ ) = 0, Hi (ζ ) = 0} , for every ζ
− −
close enough to ζ ∗ , by the convexity of Gi (i ∈ I00 ) and Hi (i ∈ I00 ), one has
−
Gi (ζ ∗ )T (ζ − ζ ∗ ) ≤ 0, ∀i ∈ I00 , (15)
and
−
Hi (ζ ∗ )T (ζ − ζ ∗ ) ≤ 0, ∀i ∈ I00 . (16)
g φ h
Multiplying (7)–(16) by i > 0 (i ∈ P + ), i > 0 (i ∈ Q+ ), i > 0(i ∈
h + + + +
R+ ), −i > 0 (i ∈ R− ), ςiG > 0 (i ∈ I0+ ∪ I00 ), ςiH > 0 (i ∈ I+0 ∪ I00 ),
− − −
−ςiG > 0(i ∈ I0+ ), −ςiH > 0(i ∈ I+0 ), −ςiG > 0(i ∈ I00 ) and −ςiH > 0(i ∈
−
I00 ), respectively, and adding, we get
T
p
g
q
φ
r
h
i 1 gi (ζ ∗ , ui ) + i φi (ζ ∗ ) + i hi (ζ ∗ ) (ζ − ζ ∗ )
i=1 i=1 i=1
T
s
− [ςiG Gi (ζ ∗ ) + ςiH Hi (ζ ∗ )] (ζ − ζ ∗ ) ≤ 0, ∀ζ ∈ Ω.
i=1
which implies from (5) that f(ζ ) ≥ f(ζ ∗ ) for every ζ close enough to ζ ∗ . Hence,
ζ ∗ is a local minimizer of the RMPEC. This concludes the proof.
Now, consider a point ζ ∗ := (1, 0, 0) in the feasible region where the constraints
g φ
g1 (., 1), φ1 and H1 are active with I+0 (ζ ∗ ) = {1} . Now, for any 1 ≥ 0, 1 ≥ 0 and
H
ς1 ≥ 0 with
g φ
1 1 g1 (ζ ∗ , 1) + 1 φ1 (ζ ∗ ) − ς1H H1 (ζ ∗ ) = 0,
g φ
one has 1 = 1 = ς1H = 0, and hence RMPEC-ENNAMCQ is satisfied at ζ ∗ . Now
g φ
f(ζ ∗ ) + 1 1 g1 (ζ ∗ , v) + 1 φ1 (ζ ∗ ) − ς1H H1 (ζ ∗ ) = 0,
g φ
gives the existence of 1 = 0, ∀v ∈ [0.5, 1], 1 = 0, ς1H = 1, which implies that
+
the robust KKT conditions are fulfilled from Theorem 5 at ζ ∗ . Since I+0 (ζ ∗ ) is
nonempty, therefore to verify the sufficient optimality conditions of Theorem 6, we
have to verify the convexity of f and H1 at ζ ∗ . Indeed, f and −H1 are both convex at
ζ ∗ over the feasible region, which implies by Theorem 6 that ζ ∗ is a robust global
minimizer.
Similarly, we can show that ζ̄ := (0, 0, 1) is also a robust global minimizer of the
uncertain problem. In fact, any point of the type (t, 0, 0) or (0, 0, t) with t ∈ [0, 1]
minimizes the RMPEC globally. Further, if we add another equality constraint like
ζ1 + ζ2 + ζ3 = 0, then the origin will be an unique local minimizer.
We can now define some other stationary points for the RMPEC as follows:
(c) either ςiG > 0, ςiH > 0 or ςiG ςiH = 0 for any i ∈ I00 , then M-stationarity holds
at ζ ∗ ;
(d) ςiG ≥ 0, ςiH ≥ 0 for any i ∈ I00 , then S-stationarity holds at ζ ∗ .
Remark 1 If the uncertainty set Ui is a singleton set for every i ∈ P, then the sta-
tionary concepts given in Definitions 6 and 7 reduce to the corresponding stationary
concepts for MPEC (see, e.g. [44]). We can derive sufficient optimality conditions
for different stationary concepts for RMPEC similar to Theorem 6. Moreover, FJ and
KKT type C-, A-, M- and S-stationary conditions can be derived using the approaches
of Scheel and Scholtes [41], Flegel and Kanzow [11], Ye [44] along with Theorems
3 and 5. The relation among different stationary points are given in Fig. 1.
5 Conclusion
References
1. Ben-Tal, A., Nemirovski, A.: Robust optimization methodology and applications. Math. Pro-
gram. 92, 453–480 (2002)
2. Beyer, H.G., Sendhoff, B.: Robust optimization-a comprehensive survey. Comput. Methods
Appl. Mech. Eng. 196, 3190–3218 (2007)
3. Ben-Tal, A., El-Ghaoui, L., Nemirovski, A.: Robust Optimization. Princeton University Press
(2009)
4. Bertsimas, D., Brown, D.B., Caramanis, C.: Theory and applications of robust optimization.
SIAM Rev. 53, 464–501 (2011)
5. Chen, J., Köbis, E., Yao, J.C.: Optimality conditions and duality for robust nonsmooth multi-
objective optimization problems with constraints. J. Optim. Theory Appl. 181, 411–436 (2019)
6. Chuong, T.D.: Optimality and duality for robust multiobjective optimization problems. Non-
linear Anal. 134, 127–143 (2016)
7. Colson, B., Marcotte, P., Savard, G.: An overview of bilevel optimization. Ann. Oper. Res. 153,
235–256 (2007)
8. Dempe, S.: Annotated bibliography on bilevel programming and mathematical programs with
equilibrium constraints. Optimization 52(3), 333–359 (2003)
9. Fakhar, M., Mahyarinia, M.R., Zafarani, J.: On nonsmooth robust multiobjective optimization
under generalized convexity with applications to portfolio optimization. Eur. J. Oper. Res. 265,
39–48 (2017)
10. Fakhar, M., Mahyarinia, M.R., Zafarani, J.: On approximate solutions for nonsmooth robust
multiobjective optimization problems. Optimization 68(9), 1653–1683 (2019)
11. Flegel, M.L., Kanzow, C.: Optimality conditions for mathematical programs with equilibrium
constraints: Fritz John and Abadie-type approaches. Institute of Applied Mathematics and
Statistics (2002)
12. Flegel, M.L., Kanzow, C.: A Fritz John approach to first order optimality conditions for math-
ematical programs with equilibrium constraints. Optimization 52, 277–286 (2003)
13. Flegel, M.L., Kanzow, C.: Abadie-type constraint qualification for mathematical programs with
equilibrium constraints. J. Optim. Theory Appl. 124, 595–614 (2005)
14. Flegel, M.L., Kanzow, C.: On M-stationary points for mathematical programs with equilibrium
constraints. J. Math. Anal. Appl. 310(1), 286–302 (2005)
15. Flegel, M.L., Kanzow, C., Outrata, J.V.: Optimality conditions for disjunctive programs with
application to mathematical programs with equilibrium constraints. Set-Valued Anal. 15, 139–
162 (2007)
16. Gabrel, V., Murat, C., Thiele, A.: Recent advances in robust optimization: an overview. Eur. J.
Oper. Res. 235, 471–483 (2014)
17. Guo, L., Lin, G.H.: Notes on some constraint qualifications for mathematical programs with
equilibrium constraints. J. Optim. Theory Appl. 156, 600–616 (2013)
On Mathematical Programs with Equilibrium Constraints Under Data Uncertainty 299
18. Guo, L., Lin, G.H., Zhao, J.: Wolfe-type duality for mathematical programs with equilibrium
constraints. Acta Math. Appl. Sin. Engl. Ser. 35, 532–540 (2019)
19. Guu, S.M., Mishra, S.K., Pandey, Y.: Duality for nonsmooth mathematical programming prob-
lems with equilibrium constraints. J. Inequalities Appl. 28, 1–5 (2016)
20. Jeyakumar, V., Li, G., Lee, G.M.: Robust duality for generalized convex programming problems
under data uncertainty. Nonlinear Anal. 75, 1362–1373 (2012)
21. Jeyakumar, V., Li, G.Y., Srisatkunarajah, S.: Strong duality for robust minimax fractional
programming problems. Eur. J. Oper. Res. 228(2), 331–336 (2013)
22. Kuroiwa, D., Lee, G.M.: On robust multiobjective optimization. Vietnam J. Math. 40, 305–317
(2012)
23. Laha, V., Al-Shamary, B., Mishra, S.K.: On nonsmooth V-invexity and vector variational-like
inequalities in terms of the Michel-Penot subdifferentials. Optim. Lett. 8(5), 1675–1690 (2014)
24. Lee, G.M., Kim, M.H.: On duality theorems for robust optimization problems. J. Chungcheong
Math. Soc. 26, 723–734 (2013)
25. Lee, G.M., Kim, M.H.: Robust duality for nonsmooth multiobjective optimization problems.
J. Chungcheong Math. Soc. 30, 31–40 (2017)
26. Lee, J.H., Lee, G.M.: On optimality conditions and duality theorems for robust semi-infinite
multiobjective optimization problems. Ann. Oper. Res. 269, 419–438 (2018)
27. Liu, G., Ye, J.J., Zhu, J.: Partial exact penalty for mathematical programs with equilibrium
constraints. Set-Valued Anal. 16, 785–804 (2008)
28. Laha, V., Kumar, R., Singh, H.N., Mishra, S.K.: On minimax programming with vanishing
constraints. In: Laha, V., Maréchal, P., Mishra, S.K. (eds.) Optimization, Variational Analysis
and Applications. IFSOVAA 2020. Springer Proceedings in Mathematics & Statistics, vol 355.
Springer, Singapore (2021)
29. Lee, G.M., Son, P.T.: On nonsmooth optimality theorems for robust optimization problems.
Bull. Korean Math. Soc. 51(1), 287–301 (2014)
30. Laha, V., Singh, V., Pandey, Y., Mishra, S.K.: Nonsmooth mathematical programs with van-
ishing constraints in Banach spaces. In: Nikeghbali, A., Pardalos, P.M., Raigorodskii, A.M.,
Rassias, M.T. (eds.) High-Dimensional Optimization and Probability. Springer Optimization
and Its Applications, vol 191. Springer, Cham (2022)
31. Luo, Z.Q., Pang, J.S., Ralph, D.: Mathematical Programs with Equilibrium Constraints. Cam-
bridge University Press, Cambridge (1996)
32. Mangasarian, O.L., Fromovitz, S.: The Fritz John necessary optimality conditions in the pres-
ence of equality and inequality constraints. J. Math. Anal. Appl. 17(1), 37–47 (1967)
33. Mishra, S.K., Laha, V.: On V − r −invexity and vector variational-like inequalities. Filomat
26(5), 1065–1073 (2012)
34. Mishra, S.K., Laha, V.: On approximately star-shaped functions and approximate vector vari-
ational inequalities. J. Optim. Theory Appl. 156(2), 278–293 (2013)
35. Mishra, S.K., Laha, V.: On minty variational principle for nonsmooth vector optimization
problems with approximate convexity. Optim. Lett. 10(3), 577–589 (2016)
36. Movahedian, N., Nobakhtian, S.: Constraint qualifications for nonsmooth mathematical pro-
grams with equilibrium constraints. Set-Valued Var. Anal. 17, 63–95 (2009)
37. Mishra, S.K., Singh, V., Laha, V.: On duality for mathematical programs with vanishing con-
straints. Ann. Oper. Res 243(1), 249–272 (2016)
38. Mishra, S.K., Singh, V., Laha, V., Mohapatra, R.N.: On constraint qualifications for multiob-
jective optimization problems with vanishing constraints. In: Xu, H., Wang, S., Wu, S.Y. (eds.)
Optimization Methods, Theory and Applications. Springer, Berlin, Heidelberg (2015)
39. Pandey, Y., Mishra, S.K.: Duality of mathematical programming problems with equilibrium
constraints. Pac. J. Optim. 13, 105–122 (2017)
40. Soleimanian, A., Jajaei, G.S.: Robust nonlinear optimization with conic representable uncer-
tainty set. Eur. J. Oper. Res. 228(2), 337–344 (2013)
41. Scheel, H., Scholtes, S.: Mathematical programs with complementarity constraints: stationarity,
optimality and sensitivity. Math. Oper. Res. 25, 1–22 (2000)
300 V. Laha and L. Pandey
42. Wang, J., Li, S.J., Chen, C.R.: Generalized robust duality in constrained nonconvex optimiza-
tion. Optimization 70(3), 591–612 (2021)
43. Ye, J.J.: Multiplier rules under mixed assumptions of differentiability and Lipschitz continuity.
SIAM J. Control Optim. 39(5), 1441–1460 (2001)
44. Ye, J.J.: Necessary and sufficient optimality conditions for mathematical programs with equi-
librium constraints. J. Math. Anal. Appl. 307, 350–369 (2005)
A New Approach to Solve Fuzzy
Transportation Problem
1 Introduction
The theory of fuzzy set (FS) was first established by Zadeh [25], by defining single
membership degree. FS has been applied in several disciplines of mathematics, engi-
neering, and management. The application of FS is more popular in the domain of
optimization and in real-world decision-making problems after the groundbreaking
work carried out by Bellman and Zadeh [5].
Atanassov [3] developed the theory of intuitionistic fuzzy set (IFS) in terms of
generalization of FS, which extends the single membership degree of FS to two
more logical terms, the membership degree and the non-membership degree, so that
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 301
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_21
302 A. Choudhary and S. P. Yadav
their summation lies in [0, 1]. The IFS describes the vagueness of an element in the
set more comprehensively than the FS and has been very helpful in dealing with
uncertainty.
In the real-life TPs, there are several instances where transportation cost (TC)
may be unsure due to number of reasons (Dempe and Starostina [6]). Many authors
(Nagoorgani and Razak [17], Dinager and Palanivel [7], Pandian and Natrajan [18])
have studied fuzzy transportation problems (FTPs) to address the ambiguity that
arises in TP. Mohideen and Kumar [15] investigate a TP with all the parameters being
trapezoidal fuzzy numbers. By utilizing the ranking of fuzzy numbers, Basirzadeh
[4] suggested a simplistic but efficient parametric approach to solving the FTP. Kaur
and Kumar [12] introduced a novel approach for determining the fuzzy optimum
solution for an FTP where TC is expressed by generalized fuzzy numbers. Szmidt
and Kacprzyk [22] have introduced the notion of distance between two IFSs with
the assistance of membership and non-membership functions. Mukherjee and Basu
[16] investigated the solution of the assignment problem using similarity measures
under intuitionistic environment.
In various real-world TP, DM does not assure about TCs due to uncertain quanti-
ties arising from variation in fuel prices, heavy traffic, temperature fluctuations, etc.
In this circumstance, DM hesitates in prediction of TC. So, to cover hesitation factor,
IFS is more suitable than the FS. Hussain and Kumar [10], Singh and Yadav [21]
have used TFS in real-life TP. Antony et al. [1] have studied TP using TIFNs. Singh
and Yadav [20] solved TP using intuitionistic fuzzy cost but crisp availabilities and
demand. Kumar [13] suggested a novel approach to obtain a solution of intuitionistic
FTP of type-2 where TC is expressed by TIFN. Hunwisai et al. [9] proposed a novel
method to solving FTP using trapezoidal intuitionistic fuzzy numbers in generalized
form. Traneva and Tranev [23] extended the fuzzy zero point method (FZPM) to
intuitionistic FZPM and obtained the solution of intuitionistic FTP using the concept
of IFSs and index matrices. Josephine et al. [11] proposed an efficient algorithm
to find the optimal solution of a TP in which costs, supplies, and demands all are
trapezoidal fuzzy numbers (TrFNs). Mishra and kumar [14] proposed a new method
(named JMD method) to transform an unbalanced triangular intuitionistic fully IFTP
into a balanced fully IFTP and then obtained the intuitionistic fuzzy optimal solu-
tion of unbalanced fully IFTP. Singh and Garg [19] proposed a family of Hamming,
Euclidean, and utmost distance measures for type-2 IFSs. Further proposed a rank-
ing method based on these measures for solving group decision-making problems.
Anusha and Sireesha proposed [2], a new distance measure for type-2 IFSs and its
application to multi-criteria group decision-making. Xue and Deng [24] presented
the decision-making under measure-based granular uncertainty in intuitionistic fuzzy
environment. Garg and Singh [8] presented a new group decision-making approach
based on similarity measure between type-2 IFSs.
In the present study, we have modeled a TP in which TCs are TIFNs, availabilities
and requirements are real numbers. For ordering of TIFNs, we propose the accuracy
function for TIFNs. To determine the optimal solution of FTP, we develop a new
algorithm. The rest of the article is organized as follows: in Sect. 2, definitions and
mathematical operations on TIFNs from the existing literature (Singh and Yadav [21],
A New Approach to Solve Fuzzy Transportation Problem 303
Atanassov [3]) are defined. In Sect. 3, ranking and ordering of TIFNs are defined.
Section 4 deals with solution procedure of TIFTP. A numerical example is used in
Sect. 5 to show how to use the approach to obtain the optimal solution.
2 Some Definitions
Let the collection of real numbers be R. Then an IFS Ã I = {(x, μ A˜I (x), ν A˜I (x)) :
x ∈ R} is called an IFN if the following two necessary conditions hold:
2. μ A˜I , ν A˜I : R → [0, 1] are piece-wise continuous functions and 0 ≤ μ A˜I (x) +
ν A˜I (x) ≤ 1 ∀x ∈ R, where for l, r, l , r ∈ R with l ≤ l , r ≤ r , we have
⎧
⎪
⎪ g1 (x), k1 − l ≤ x < k1 ,
⎪
⎨ 1, x = k1 ,
μ A˜I (x) =
⎪
⎪ g (x), k 1 < x ≤ k1 + r,
⎪
⎩
2
0, other wise,
and
⎧
⎪
⎪ h 1 (x), k1 − l ≤ x < k1 ; 0 ≤ g1 (x) + h 1 (x) ≤ 1,
⎪
⎨0, x = k1 ,
ν A˜I (x) =
⎪
⎪ h 2 (x), k1 < x ≤ k1 + r ; 0 ≤ g2 (x) + h 2 (x) ≤ 1,
⎪
⎩
1, other wise.
304 A. Choudhary and S. P. Yadav
Here k1 is called the mean value of à I ; l and r are called the left and right spreads of
μ A˜I , respectively; l and r are called the left and right spreads of ν A˜I , respectively;
g1 and h 2 are piece-wise continuous and increasing functions in [k1 − l, k1 ) and
(k1 , k1 + r ], respectively; and g2 and h 1 are piece-wise continuous and decreasing
functions in (k1 , k1 + r ] and [k1 − l , k1 ), respectively. The IFN Ã I is denoted by
à I = (k1 ; l, r ; l , r ).
and ⎧
⎪ ξ2 − x
⎪
⎪ , ξ1 < x ≤ ξ2 ,
⎪
⎨ ξ2 − a1
ν A˜I (x) = x − ξ2
⎪
⎪ , ξ2 ≤ x < ξ3 ,
⎪
⎪ aξ3 − ξ2
⎩
1, other wise,
where ξ1 ≤ ξ1 < ξ2 < ξ3 ≤ ξ3 . The TIFN Ã I is denoted by
à I = (ξ1 , ξ2 , ξ3 ; ξ1 , ξ2 , ξ3 ) and is shown in Fig. 1.
B˜ I = (ξ1 − ζ3 , ξ2 − ζ2 , ξ3 − ζ1 ; ξ1 − ζ3 , ξ2 − ζ2 , ξ3 − ζ1 ),
Subtraction: Ã I
Multiplication: Ã I ⊗ B˜ I = (m 1 , m 2 , m 3 ; m 1 , m 2 , m 3 ),
A New Approach to Solve Fuzzy Transportation Problem 305
where
m 1 = min{ξ1 ζ1 , ξ1 ζ3 , ξ3 ζ1 , ξ3 ζ3 }, m 2 = ξ2 ζ2 , m 3 = max{ξ1 ζ1 , ξ1 ζ3 , ξ3 ζ1 , ξ3 ζ3 }
m 1 = min{ξ1 ζ1 , ξ1 ζ3 , ξ3 ζ1 , ξ3 ζ3 }, m 3 = max{ξ1 ζ1 , ξ1 ζ3 , ξ3 ζ1 , ξ3 ζ3 }.
Scalar multiplication
1. λ Ã I = (λξ1 , λξ2 , λξ3 ; λξ1 , λξ2 , λξ3 ) : λ ≥ 0.
2. λ Ã I = (λξ3 , λξ2 , λξ1 ; λξ3 , λξ2 , λξ1 ) : λ < 0.
(a) Ã I B˜ I ⇐⇒ r( Ã I ) ≤ r ( B˜ I ).
(b) Ã I ≈ B˜ I ⇐⇒ r( Ã I ) = r ( B˜ I ).
(c) Ã I ≺ B˜ I ⇐⇒ r( Ã I ) < r ( B˜ I ).
(d) min{ Ã I , B˜ I } = Ã I ⇐⇒ Ã I B˜ I .
306 A. Choudhary and S. P. Yadav
α, β ∈ R.
Case 1. Let α > 0, β > 0.
r (α A˜I + β B˜ I ) = r [(αξ1 , αξ2 , αξ3 ; αξ1 , αξ2 , αξ3 ) ⊕ (βζ1 , βζ2 , βζ3 ; βζ1 , βζ2 , βζ3 )]
= r [(αξ1 + βζ1 , αξ2 + βζ2 , αξ3 + βζ3 ; αξ1 + βζ1 , αξ2 +
βζ2 , , αξ3 βζ3 )]
αξ1 + βζ1 + 4(αξ2 + βζ2 ) + αξ3 + βζ3 + αξ1 + βζ1 + αξ3 + βζ3
=
8
αξ1 + 4αξ2 + αξ3 + αξ1 + αξ3 βζ1 + 4βζ2 + βζ3 + βζ1 + βζ3
= + +
8 8
ξ1 + 4ξ2 + ξ3 + ξ1 + ξ3 ζ1 + 4ζ2 + ζ3 + ζ1 + ζ3
=α +β
8 8
= α r ( Ã I ) + β r ( B˜ I ).
Case 2. Let α > 0, β < 0.
r (α A˜I + β B˜ I ) = r [(αξ1 , αξ2 , αξ3 ; αξ1 , αξ2 , αξ3 ) ⊕ (βζ3 , βζ2 , βζ1 ; βζ3 , βζ2 , βζ1 )]
= r [(αξ1 + βζ3 , αξ2 + βζ2 , αξ3 + βζ1 ; αξ1 + βζ3 , αξ2 +
βζ2 , , αξ3 βζ1 )]
αξ1 + βζ3 + 4(αξ2 + βζ2 ) + αξ3 + βζ1 + αξ1 + βζ3 + αξ3 + βζ1
=
8
αξ1 + 4αξ2 + αξ3 + αξ1 + αξ3 βζ3 + 4βζ2 + βζ1 + βζ3 + βζ1
= + +
8 8
ξ1 + 4ξ2 + ξ3 + ξ1 + ξ3 ζ1 + 4bζ2 + ζ3 + ζ1 + ζ3
=α +β
8 8
= α r ( Ã I ) + β r ( B˜ I ).
Case 3. Let α < 0, β < 0.
r (α A˜I + β B˜ I ) = r [(αξ3 , αξ2 , αξ1 ; αξ3 , αξ2 , αξ1 ) ⊕ (βζ3 , βζ2 , βζ1 ; βζ3 , βζ2 , βζ1 )]
= r [(αξ3 + βζ3 , αξ2 + βζ2 , αξ1 + βζ1 ; αξ3 + βζ3 , αξ2 +
βζ2 , , αξ3 βζ1 )]
αξ3 + βζ3 + 4(αξ2 + βζ2 ) + αξ1 + βζ1 + αξ3 + βζ3 + αξ1 + βζ1
=
8
A New Approach to Solve Fuzzy Transportation Problem 307
αξ3 + 4αξ2 + αξ1 + αξ3 + αξ1 βζ3 + 4βζ2 + βζ1 + βζ3 + βζ1
= + +
8 8
ξ1 + 4ξ2 + ξ3 + ξ1 + ξ3 ζ1 + 4ζ2 + ζ3 + ζ1 + ζ3
=α +β
8 8
I ˜
= α r ( Ã ) + β r ( B ).
I
Definition 3.2 A balanced IFTP (BIFTP) in which costs are TIFN but supplies and
demands are crisp is defined as
m
n
Min Z̃ I ≈ c̃iIj ⊗ xi j
i=1 j=1
n
subject to xi j ≈ ai i = 1, 2, 3, . . . , m,
j=1
m
xi j ≈ b j j = 1, 2, 3, . . . , n,
i=1
xi j ≥ 0 i = 1, 2, 3, . . . , m; j = 1, 2, 3, . . . , n,
where c̃iIj is the unit fuzzy transportation cost (IFTC) from the ith supply point to
the jth destination,
ai is the availability of the commodity at the ith supply point,
b j is the requirement of the commodity at the jth destination,
xi j is the number of units of the commodity transported from the ith supply point to
the jth destination.
4 Proposed Method
Step4: For each row (column) of the reduced IFTC matrix, find the difference between
the smallest and next higher cost. Write them by the side of the matrix against the
respective rows (column) and call them as row (column) IF penalties.
Step5: Choose the row or column with the largest IF penalty. Make the maximum
possible allocation to the IF zero cost cell and cross off the satisfied row or column.
If there is a tie for the largest IF penalty or the IF zero cost cell, go to Step 7. If not,
go to Step 6.
Step6: Now consider modified matrix obtained in Step 5. Repeat Steps 2–5 until all
the rows and columns are satisfied.
Step7 (Tie) 1. If the largest IF penalty is not unique, then break the tie among them
by choosing the row or column having the smallest IF cost, i.e., IF zero cost.
2. If in the chosen row or column, the smallest IF cost is not unique, then choose the
cell from tied ones to which more allocation can be made.
5 Numerical Example
Consider the following BIFTP. In this problem, there are four supply points S1 , S2 ,
S3 , S4 and four destinations D1 , D2 , D3 , D4 (Tables 1 and 2).
Applying Step 2, the resultant IFTC is given in Table 3.
Applying Step 3, the resultant IFTC is given in Table 4.
In Table 4, calculate the row and column IF penalties and write them against the
respective rows and columns. The resultant IFTC matrix is given in Table 5.
Applying steps 2 and 3 to Table 6, we get the reduced IFTC matrix as given in
Table 7. Calculate the new penalties.
(1, 2, 4; 1, 2, 5) is the largest penalty corresponding to row S4 . Allocate 1 to the
cell (4, 1) and cross off the fourth row. The resultant IFTC matrix is shown in Table 8.
Applying steps 2 and 3 to Table 8, we get the reduced IFTC matrix as given in
Table 9. Calculate the new penalties.
Now we verify that the solution obtained in Table 15 is optimal with the help of
MODI method [20].
In Table 16, we observed that r (d˜i j ) ≤ 0 where d˜i j = u˜i I ⊕ v˜j I C˜i j . Hence
I I I
6 Conclusion
In this article, a novel approach to obtaining the optimal solution to the intuitionistic
FTP is proposed, which is used for anticipating the transportation cost, which varies
due to weather condition, diesel cost, traffic conditions, etc. The proposed method
provides a simpler method compared to the solution available in the existing literature
[1, 21]. The advantage of the suggested approach is that it is very easy to figure out
the best way to solve the problem. The suggested approach provides solution which
is very near to optimal solution. There is almost no scope of improvement of the
solution and this can be seen when MODI method is applied in our problem. The TC
based on suggested approach is less than the previous work by [1] and equal to [21].
The suggested approach would facilitate the DMs in solving logistical problem that
arises in the actual world by helping them in their decision-making and providing
an optimal solution in a simple and efficient manner. The proposed method can be
extended for solving FTP by taking all the parameters of TIFNs in future.
References
1. Antony, R.J.P., Savarimuthu, S.J., Pathinathan, T.: Method for solving the transportation prob-
lem using triangular intuitionistic fuzzy number. Int. J. Comput. Algorithm 3(1), 590–605
(2014)
2. Anusha, V., Sireesha, V.: A new distance measure to rank type-2 intuitionistic fuzzy sets and
its application to multi-criteria group decision making. Int. J. Fuzzy Syst. Appl. (IJFSA) 11(1),
1–17 (2022)
3. Atanassov, K.T.: Intuitionistic fuzzy sets. Fuzzy Sets Syst. 20, 87–96 (1986)
4. Basirzadeh, H.: An approach for solving fuzzy transportation problem. Appl. Math. Sci. 5(32),
1549–1566 (2011)
5. Bellman, R.E., Zadeh, L.A.: Decision-making in a fuzzy environment. Manag. Sci. 17(4),
B-141 (1970)
314 A. Choudhary and S. P. Yadav
6. Dempe, S., Starostina, T.: Optimal toll charges in a fuzzy flow problem. In: Computational
Intelligence, Theory and Applications, pp. 405–413. Springer, Berlin, Heidelberg (2006)
7. Dinagar, D.S., Palanivel, K.: The transportation problem in fuzzy environment. Int. J. Algo-
rithms, Comput. Math. 2(3), 65–71 (2009)
8. Garg, H., Singh, S.: Algorithm for solving group decision-making problems based on the
similarity measures under type 2 intuitionistic fuzzy sets environment. Soft Comput. 24(10),
7361–7381 (2020)
9. Hunwisai, D., Kumam, P., Kumam, W.: A new method for optimal solution of intuitionistic
fuzzy transportation problems via generalized trapezoidal intuitionistic fuzzy numbers. Fuzzy
Inf. Eng. 11(1), 105–120 (2019)
10. Hussain, R.J., Kumar, P.S.: Algorithmic approach for solving intuitionistic fuzzy transportation
problem. Appl. Math. Sci. 6(80), 3981–3989 (2012)
11. Josephine, F.S., Saranya, A., Nishandhi, I.F.: A dynamic method for solving intuitionistic fuzzy
transportation problem. Eur. J. Mol. Clin. Med. 7(11), 2020 (2020)
12. Kaur, A., Kumar, A.: A new approach for solving fuzzy transportation problems using gener-
alized trapezoidal fuzzy numbers. Appl. Soft Comput. 12(3), 1201–1213 (2012)
13. Kumar, P.S.: A note on’a new approach for solving intuitionistic fuzzy transportation problem
of type-2’. Int. J. Logist. Syst. Manag. 29(1), 102–129 (2018)
14. Mishra, A., Kumar, A.: JMD method for transforming an unbalanced fully intuitionistic fuzzy
transportation problem into a balanced fully intuitionistic fuzzy transportation problem. Soft
Comput. 24(20), 15639–15654 (2020)
15. Mohideen, S.I., Kumar, P.S.: A comparative study on transportation problem in fuzzy environ-
ment. Int. J. Math. Res. 2(1), 151–158 (2010)
16. Mukherjee, S., Basu, K.: Solution of a class of intuitionistic fuzzy assignment problem by
using similarity measures. Knowl.-Based Syst. 27, 170–179 (2012)
17. Nagoorgani, A., Razak, K.A.: Two stage fuzzy transportation problem. J. Phys. Sci. 10, 63–69
(2006)
18. Pandian, P., Natarajan, G.: A new algorithm for finding a fuzzy optimal solution for fuzzy
transportation problems. Appl. Math. Sci. 4(2), 79–90 (2010)
19. Singh, S., Garg, H.: Distance measures between type-2 intuitionistic fuzzy sets and their appli-
cation to multi-criteria decision-making process. Appl. Intell. 46(4), 788–799 (2017)
20. Singh, S.K., Yadav, S.P.: Efficient approach for solving type-1 intuitionistic fuzzy transportation
problem. Int. J. Syst. Assur. Eng. Manag. 6(3), 259–267 (2015)
21. Singh, S.K., Yadav, S.P.: A new approach for solving intuitionistic fuzzy transportation problem
of type-2. Ann. Oper. Res. 243(1), 349–363 (2016)
22. Szmidt, E., Kacprzyk, J.: Distances between intuitionistic fuzzy sets. Fuzzy Sets Syst. 114(3),
505–518 (2000)
23. Traneva, V., Tranev, S.: Intuitionistic fuzzy transportation problem by zero point method. In:
2020 15th Conference on Computer Science and Information Systems (FedCSIS), pp. 349–358.
IEEE (2020)
24. Xue, Y., Deng, Y.: Decision making under measure-based granular uncertainty with intuition-
istic fuzzy sets. Appl. Intell. 51(8), 6224–6233 (2021)
25. Zadeh, L.A.: Fuzzy sets. Inf. Control 8(3), 338–353 (1965)
The Best State-Based Development
of Fuzzy DEA Model
Abstract This paper studies fuzzy data envelopment analysis (FDEA) to measure
the relative efficiencies of homogeneous decision-making units (DMUs) using the
α-cut approach. The proposed model uses the same set of constraints for all DMUs.
It gives a uniform environment for all DMUs. Furthermore, a ranking method based
on lower and upper bound FDEA performance efficiencies is proposed to rank the
DMUs. The proposed FDEA models and ranking approach are demonstrated using
an example with triangular fuzzy numbers (TFNs), and the results are presented in
Tables 2 and 3.
1 Introduction
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 315
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_22
316 A. Sonkariya and S. P. Yadav
ities with cone-ratio envelopment theory and assurance region DEA models. Cooper
et al. [4] applied these models to evaluate the efficiency of a telecommunication com-
pany’s branch offices in Korea. Kao and Liu [10] proposed a method for calculating
DMU efficiency using fuzzy observations. The core concept uses the α-cut strategy
to convert FDEA models into conventional DEA models with crisp data. The mem-
bership functions of the efficiency scores are constructed using a pair of parametric
algorithms that characterize the class of crisp DEA models. Guo and Tanaka [8]
developed the FDEA model to cope with the efficiency assessment problem using
fuzzy data. Furthermore, considering the interaction between DEA and Regression
analysis (RA), an augmentation of the FDEA model to a more generic form is pro-
vided. Despotis and Smirlis [5] used a basic formulation that differs from that used
in IDEA to convert the non-linear DEA model to an LP. Unlike IDEA, they used
variable transformations based on the original data set, and no scale modifications
were applied to the data. Jahanshahloo et al. [9] developed the slack-based mea-
sure (SBM)—DEA model in fuzzy environment for assessing the relative efficiency
scores and ranking technique using fuzzy input. Wen and Li [15] created a hybrid
technique that combines fuzzy simulation and genetic algorithm to encounter fuzzy
problems. The hybrid model can be changed to linear programming when all inputs
and outputs are trapezoidal or triangular fuzzy variables. Dotoli et al. [6] introduced
a new cross-efficiency FDEA approach for assessing distinct elements (DMUs) in an
environment of uncertainty. A cross-assessment derived from a compromise between
adequately specified objectives is used to assign a fuzzy triangular efficiency to each
DMU. The DMUs are then ranked once the results have been defuzzified. The sug-
gested technique is used to assess the performance of healthcare systems in a Southern
Italian area. Arya and Yadav [1] used an α-cut strategy to create FDEA models in
order to quantify the left-hand relative efficiency and right-hand relative efficiency
for DMUs and suggested mechanism for ranking DMUs accordingly their obtained
left-hand efficiency and right-hand efficiencies. Tavassoli et al. [12] examined four
different types of supply chain supplier selection models and offered a decision-
making framework for supplier selection with fuzzy, deterministic and stochastic
settings. Using the α-cut approach, the suggested stochastic FDEA (SFDEA) model
can be solved as a crisp programme. Wang et al. [13] measured the efficiency of
DMUs under fixed production frontier using interval input–output data.
In this study, FDEA models are developed to avert the usage of separate production
boundaries to estimate the efficiency of distinct DMUs. For measuring both the lower
bound efficiency and upper bound efficiency of each DMU, models employ interval
arithmetic using the same constraints, resulting in a unified and consistent production
frontier. The proposed FDEA model is applied to a numerical example with TFNs.
Further, a ranking approach is proposed for ranking DMUs. Results are presented in
Table 2.
The framework of this paper is as follows. Section 2 presents the preliminaries
of basic fuzzy set theory. The CCR efficiency model is stated in Sect. 3. In Sect. 4,
the proposed FDEA model is demonstrated. Section 5 discusses numerical examples,
and results (lower and upper bound efficiencies) are demonstrated in Table 2. The
The Best State-Based Development of Fuzzy DEA Model 317
proposed ranking approach and ranks of DMUs are presented in Sect. 6. The paper’s
last Sect. 7 ends with conclusions.
2 Preliminaries
Some key concepts and fuzzy arithmetic operations that are necessary for developing
the FDEA model are provided in this section.
Definition 1 (Fuzzy Set) [17] The definition of a fuzzy set (FS) in a universal set X
is given as
R̃ = {(x, μ R̃ (x)) : x ∈ X},
Definition 3 (Fuzzy number) [17] A fuzzy number (FN) R̃ is a CFS R̃ in R, the set
of real numbers, if:
(i) μ R̃ (xo ) = 1, for a unique xo ∈ R,
(ii) μ R̃ is a piecewise continuous function in R.
xo is said to be the mean value of R̃.
∀x ∈ R.
Rα = {x ∈ X : μ R̃ (x) ≥ α}, 0 ≤ α ≤ 1.
318 A. Sonkariya and S. P. Yadav
Remark 1 R0 = X.
Definition 6 Arithmetic operations on TFNs. Let R̃ = (r L , r M , r U ) and S̃ =
(s L , s M , s U ) be two positive TFNs, i.e., r L > 0, s L > 0. Then the arithmetic opera-
tions are defined below [14]:
(i) Addition: R̃ + S̃ = (r L + s L , r M + s M , r U + s U ),
(ii) Subtraction: R̃ − S̃ = (r L − s U , r M − s M , r U − s L ),
(iii) Multiplication: R̃ × S̃ ≈ (r L s L , r M s M , r U s U ),
(iv) Division: R̃/ S̃ ≈ (r L /s U , r M /s M , r U /s L ).
Suppose we have n DMUs and each DMU produces q outputs using p inputs. Let the
quantity of ith input utilized and rth outputs produced by jth DMU are xi j and yr j
respectively, where i = 1, 2, . . . , p and r = 1, 2, . . . q. Then DEA models developed
by Arya and Yadav [1] to calculate efficiency for DMU j is given as below
Input minimization lower bound DEA model
p
min E kL = u
xik u ik
i=1
q
subject to yrl k vr k = 1,
r =1
p
q
u
xik u ik − yrl k vr k ≥ 0,
i=1 r =1
p
q
xil j u ik − yruj vr k ≥ 0 ∀ j, where j = k,
i=1 r =1
u ik ≥ ε ∀i, vr k ≥ ε ∀ r.
where u ik and vr k are the weights corresponding to the xik and yr k , respectively, and
ε is non-archimedean infinitesimal.
p
min E kU = l
xik u ik
i=1
q
subject to yruk vr k = 1,
r =1
The Best State-Based Development of Fuzzy DEA Model 319
p
q
l
xik u ik − yruk vr k ≥ 0,
i=1 r =1
p
q
xiuj u ik − yrl j vr k ≥ 0 ∀ j, where j = k,
i=1 r =1
u ik ≥ ε ∀i, vr k ≥ ε ∀ r.
If we look closely at the lower and upper efficiency models, we can see that constraint
sets in the above models to assess DMUs’ efficiencies differ from one DMU to the
other. The most significant disadvantage of using multiple constraint sets to estimate
DMUs’ efficiencies is that the efficiencies are not comparable because different pro-
duction boundaries were used in the efficiency measurement process. So, we develop
the models in which we use the same set of constraints. The conventional CCR DEA
[2] model is presented as follows:
subject to
q
yr j vr k
r =1
≤ 1,
p
xi j u ik
i=1
u ik , vr k ≥ 0 ∀ i, r respectively.
q
max Ek = yr k vr k
r =1
subject to
p
xik u ik = 1,
i=1
q
p
yr j vr k − xi j u ik ≤ 0,
r =1 i=1
320 A. Sonkariya and S. P. Yadav
u ik , vr k ≥ 0 ∀ i, r respectively.
q
max E˜k = ỹr j vr k
r =1
subject to
p
x̃ik u ik = 1̃,
i=1
q
p
ỹr j vr k − x̃i j u ik ≤ 0̃,
r =1 i=1
u ik , vr k ≥ 0 ∀ i, r respectively.
In Model 3, input and output quantities are FNs, so resulted efficiency score will
also be a fuzzy quantity. Here, we take them as TFNs,
x̃i j = xiLj , xiMj , xiUj
ỹr j = yrLj , yrMj , yrUj
Model 4
q
max Ek = yrLk , yrMk , yrUk vr k
r =1
subject to
p
xikL , xikM , xik
U
= (1, 1, 1),
i=1
q
p
yrLj , yrMj , yrUj vr k − xiLj , xiMj , xiUj u ik ≤ (0, 0, 0),
r =1 i=1
u ik , vr k ≥ 0 ∀ i, r respectively.
Model 5
q
max (E˜k )α = [(Ek )αL , (Ek )Uα ] = αyrMk + (1 − α)yrLk , αyrMk + (1 − α)yrUk vr k
r =1
subject to
p
αxikM + (1 − α)xikL , αxikM + (1 − α)xik
U
u ik = [1, 1],
i=1
q
p
αyrMj + (1 − α)yrLj , αyrMj + (1 − α)yrUj vrk − αxiMj + (1 − α)xiLj , αxiMj + (1 − α)xiUj u ik ≤ [0, 0],
r=1 i=1
⎤
q
p
αyrMj + (1 − α)yrUj vr k − αxiMj + (1 − α)xiLj u ik ⎦ ≤ [0, 0],
r =1 i=1
u ik , vr k ≥ 0 ∀ i, r respectively.
Model 5’
q
max (E˜k )α = [(Ek )αL , (Ek )Uα ] = αyrMk + (1 − α)yrLk , αyrMk + (1 − α)yrUk vr k
r =1
subject to
p
αxikM + (1 − α)xikL , αxikM + (1 − α)xik
U
u ik = [1, 1],
i=1
q
M
p
M
αyr j + (1 − α)yrLj vr k − αxi j + (1 − α)xiUj u ik ,
r =1 i=1
⎤
q
p
αyrMj + (1 − α)yrUj vr k − αxiMj + (1 − α)xiLj u ik ⎦ ≤ [0, 0],
r =1 i=1
u ik , vr k ≥ 0 ∀ i, r respectively.
Model 6
q
max (E˜k )α = [(Ek )αL , (Ek )Uα ] = αyrMk + (1 − α)yrLk , αyrMk + (1 − α)yrUk vr k
r =1
subject to
p
αxikM + (1 − α)xikL , αxikM + (1 − α)xik
U
u ik = [1, 1],
i=1
The Best State-Based Development of Fuzzy DEA Model 323
q
M
p
M
αyr j + (1 − α)yrUj vr k − αxi j + (1 − α)xiLj u ik ≤ 0,
r =1 i=1
u ik , vr k ≥ 0 ∀ i, r respectively.
Model 6 is the FDEA model in the best state. By extracting lower and upper
bounds, we get two models (Models 7 and 8), the lower and upper bound efficiency,
respectively.
q
M
max (Ek )αL = αyr k + (1 − α)yrLk vr k
r =1
subject to
p
M
αxik + (1 − α)xik
U
u ik = 1,
i=1
q
M
p
M
αyr j + (1 − α)yrUj vr k − αxi j + (1 − α)xiLj u ik ≤ 0,
r =1 i=1
u ik , vr k ≥ 0 ∀ i, r respectively.
q
M
max (Ek )Uα = αyr k + (1 − α)yrUk vr k
r =1
subject to
p
αxikM + (1 − α)xikL u ik = 1,
i=1
q
M
p
M
αyr j + (1 − α)yr j vr k −
U
αxi j + (1 − α)xiLj u ik ≤ 0,
r =1 i=1
u ik , vr k ≥ 0 ∀ i, r respectively.
324 A. Sonkariya and S. P. Yadav
5 Numerical Example
In this numerical illustration, we have taken five DMUs into consideration. Here,
each DMU consumes 2 inputs and generates 2 outputs. Data is presented in fuzzy
environment, specifically taken as TFNs presented in Table 1. After applying Models
7 and 8, each DMU’s lower and upper bound efficiencies for α ∈ (0, 1] are presented
in Table 2.
The Best State-Based Development of Fuzzy DEA Model 325
In DEA models, we rank the DMUs according to their efficiency scores in decreasing
order. But in FDEA models, it’s not possible because α-cut gives two bounds (lower
and upper). Consequently, the resulted efficiency score is in interval form. So, we
propose the aggregated accuracy function (Rk , k = 1, 2, . . . , n) as follows:
1
Rk = (Ek )αL .(Ek )Uα , k = 1, 2, . . . , n.
α=0.1
Step 1: Find the lower and upper bound efficiencies of DMUs using models 7 and
8, respectively.
Step 2: From step 1, obtain the efficiency scores (E j )αL , (E j )Uα of DMUs
j = 1, 2, . . . , 5 for α = 0.1, 0.2, . . . , 1.0.
Step 3: With the help of efficiency scores, calculate the aggregated accuracy func-
tion Rk for all DMUs.
7 Conclusion
In this study, the FDEA model is developed to assess the relative efficiencies of DMUs
in a uncertain environment. We employ the same set of constraints for all DMUs in
the proposed model, resulting in a consistent production frontier for all DMUs. Using
α-cut, lower and upper bound efficiency models are developed. Further, to rank the
DMUs, a ranking technique based on lower and upper bound FDEA performance
efficiency is given. An example shown in Table 1, in which five DMUs each uses two
fuzzy inputs and produces two fuzzy outputs in the form of TFNs, is used to show
the proposed FDEA model. Lower and upper bound efficiency results are presented
in Table 2. Ranked DMUs, on the basis of the proposed ranking method, resulted in
DMU2 > DMU4 > DMU5 > DMU1 > DMU3 . DMU2 is the best performer in this
DMUs group and DMU3 is the worst performer.
The presented model is limited to TFNs. In the future, we can extend it to L-R type
fuzzy numbers, trapezoidal fuzzy numbers and intuitionistic fuzzy environments.
References
1. Arya, A., Yadav, S.P.: Development of FDEA models to measure the performance efficiencies
of DMUs. Int. J. Fuzzy Syst. 20(1), 163–173 (2018)
2. Charnes, A., Cooper, W.W., Rhodes, E.: Measuring the efficiency of decision making units.
Eur. J. Oper. Res. 2(6), 429–444 (1978)
3. Cooper, W.W., Park, K.S., Yu, G.: IDEA and AR-IDEA: models for dealing with imprecise
data in DEA. Manag. Sci. 45(4), 597–607 (1999)
4. Cooper, W.W., Park, K.S., Yu, G.: An illustrative application of idea (imprecise data envelop-
ment analysis) to a Korean mobile telecommunication company. Oper. Res. 49(6), 807–820
(2001)
5. Despotis, D.K., Smirlis, Y.G.: Data envelopment analysis with imprecise data. Eur. J. Oper.
Res. 140(1), 24–36 (2002)
6. Dotoli, M., Epicoco, N., Falagario, M., Sciancalepore, F.: A cross-efficiency fuzzy data envelop-
ment analysis technique for performance evaluation of decision making units under uncertainty.
Comput. Ind. Eng. 79, 103–114 (2015)
7. Ghasemi, M.-R., Ignatius, J., Lozano, S., Emrouznejad, A., Hatami-Marbini, A.: A fuzzy
expected value approach under generalized data envelopment analysis. Knowl.-Based Syst.
89, 148–159 (2015)
8. Guo, P., Tanaka, H.: Fuzzy DEA: a perceptual evaluation method. Fuzzy Sets Syst. 119(1),
149–160 (2001)
9. Jahanshahloo, G.R., Soleimani-Damaneh, M., Nasrabadi, E.: Measure of efficiency in DEA
with fuzzy input-output levels: a methodology for assessing, ranking and imposing of weights
restrictions. Appl. Math. Comput. 156(1), 175–187 (2004)
10. Kao, C., Liu, S.-T.: Fuzzy efficiency measures in data envelopment analysis. Fuzzy Sets Syst.
113(3), 427–437 (2000)
11. Sengupta, J.K.: A fuzzy systems approach in data envelopment analysis. Comput. Math. Appl.
24(8–9), 259–266 (1992)
12. Tavassoli, M., Saen, R.F., Zanjirani, D.M.: Assessing sustainability of suppliers: a novel
stochastic fuzzy DEA model. Sustain. Prod. Consum. 21, 78–91 (2020)
13. Wang, Y.-M., Greatbanks, R., Yang, J.-B.: Interval efficiency assessment using data envelop-
ment analysis. Fuzzy Sets Syst. 153(3), 347–370 (2005)
The Best State-Based Development of Fuzzy DEA Model 327
14. Wang, Y.-M., Luo, Y., Liang, L.: Fuzzy data envelopment analysis based upon fuzzy arithmetic
with an application to performance assessment of manufacturing enterprises. Expert Syst. Appl.
36(3), 5205–5211 (2009)
15. Wen, M., Li, H.: Fuzzy data envelopment analysis (DEA): model and ranking method. J.
Comput. Appl. Math. 223(2), 872–878 (2009)
16. Zadeh, L.A.: Fuzzy sets. Inf. Control 8(3), 338–353 (1965)
17. Zimmermann, H.-J.: Fuzzy Set Theory-and Its Applications. Springer Science & Business
Media (2011)
Performance Evaluation of DMUs Using
Hybrid Fuzzy Multi-objective Data
Envelopment Analysis
1 Introduction
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 329
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_23
330 A. P. Singh and S. P. Yadav
of 1, and called the optimistic non-efficient if they have an efficiency score of less
than 1.
The ratio of DMU’s efficiency to the smallest efficiency under consideration is
called the pessimistic efficiency or the worst relative efficiency. It is computed from a
pessimistic viewpoint. Its value is greater than or equal to l. The DMUs are called the
pessimistic inefficient if they have an efficiency score of 1 and the pessimistic non-
inefficient if they have a pessimistic efficiency score greater than 1. We must assess
both optimistic and pessimistic efficiencies simultaneously for the total performance
of DMUs because we have two types of efficiencies: optimistic and pessimistic. In
literature, Entani et al. [2], Azizi [3, 4] evaluated the efficiencies from both optimistic
and pessimistic viewpoints in a crisp environment while Arya and Yadav [5] in the
fuzzy environment. Gupta et al. [6] developed intuitionistic fuzzy optimistic and
pessimistic multi-period portfolio optimization models. In their investigation, Puri
and Yadav [7] created intuitionistic fuzzy optimistic and pessimistic DEA models.
The goal of these studies was to create an interval using optimistic and pessimistic
efficiencies. Optimistic efficiency is the lower end of the interval in all of these
studies, whereas pessimistic efficiency is the upper end.
In all the researches mentioned above, the lower bound of optimistic efficiency and
the upper bound of pessimistic efficiency are evaluated. The ranking is based on an
interval created by combining the optimistic and pessimistic efficiency DEA models’
lower bound and upper bound efficiency. We are not in favour of considering only one
bound for each optimistic and pessimistic efficiencies. To overcome this shortcoming,
we suggest considering both bounds of optimistic and pessimistic efficiency intervals
for performance assessment. Based on this idea, we propose fuzzy multi-objective
optimistic (FMOO) and fuzzy multi-objective pessimistic DEA models to rank the
DMUs. Awadh et al. [8] proposed a fuzzy multi-objective DEA model to evaluate the
performance of DMUs in a fuzzy environment. The advantage of the methodology
is that it provides the complete ranking of DMUs. There are several multi-objective
optimization techniques that exist in the literature for solving DEA models [9–12].
But to the best of our knowledge, this is the first study using the fuzzy multi-objective
technique for performance evaluation of DMUs in the optimistic and pessimistic
environment. In this study, we develop FMOO and FMOP DEA models and use
Wang et al’s [13] geometric average efficiency approach to rank the DMUs.
The rest of the paper is organized as follows. Section 2 presents the preliminaries
and some basic definitions. The proposed FMOO and FMOP DEA models along
with the solving techniques are described in Sect. 3. The complete hybrid fuzzy
multi-objective (FMO) DEA technique is explained in Sect. 4. Section 5 presents
the numerical illustration of the proposed methodology. Finally, Sect. 6 gives the
concluding remarks and future scope.
Performance Evaluation of DMUs Using Hybrid Fuzzy Multi-objective Data … 331
2 Preliminary
For developing fuzzy optimistic and pessimistic DEA models, some basic definitions
and operation are given in this section.
Definition 1 (Fuzzy Set (FS) [14]) Let X be a universal set. A FS Ã can be defined
by à = {(x, μ à (x)) : x ∈ X }, wher e μ à : X → [0, 1].
Definition 2 (Convex Fuzzy Set (CFS) [14]) A FS Ã is defined as a CFS if for all
x1 , x2 in X ,
min {μ Ã (x1 ), μ Ã (x2 )} ≤ μ Ã (λx1 + (1 − λ)x2 ), wher e λ ∈ [0, 1].
Definition 3 (Fuzzy number (FN) [14]) A CFS Ã is defined as a FN on the real
line R if
(i) ∃ an unique xo ∈ R with μ Ã (xo ) = 1;
(ii) μ Ã is a continuous function in piece-wise sense,
xo is called the mean value of Ã.
Definition 4 (Triangular Fuzzy Number (TFN) [14]) The TFN Ã = (a L , a M , a U )
is defined by the membership function μ Ã given by
⎧
x−a L
⎪
⎨ a M −a L , aL < x ≤ aM;
U
−x
μ Ã (x) = aaU −a M , a M ≤ x < aU ;
⎪
⎩
0 otherwise.
∀x ∈ R.
Definition 6 (Positive TFN [14]) A TFN Ã = (a L , a M , a U ) is positive if and only
if a L > 0.
(i) Addition: Ã + B̃ = (a L + b L , a M + b M , a U + bU ),
(ii) Subtraction: Ã − B̃ = (a L − bU , a M − b M , a U − b L ),
(iii) Multiplication: Ã × B̃ ≈ (a L b L , a M b M , a U bU ),
(iv) Division: Ã/ B̃ ≈ (a L /bU , a M /b M , a U /b L )
332 A. P. Singh and S. P. Yadav
FP models can be transformed into triangular fuzzy optimistic (TFO) and triangular
fuzzy pessimistic (TFP) DEA models as follows (see Table 3).
Now, we will propose a methodology to solve TFO and TFP DEA models given
in Table 3. In this methodology, we will use Awadh et. al’s FMODEA [8]. First,
let us try to develop FMOO DEA model for the efficiency evaluation of DMUs
in an optimistic sense. The TFO DEA model, described in Table 3 is re-written in
Model 7.
Model 7 For k = 1, 2, 3, . . . , n,
( s vr k yrLk , rs =1 vr k yrMk , rs =1 vr k yrUk )
Max (E kO,L , E kO,M , E kO,U ) = rm=1 m m (1)
( i=1 u ik xik L,
i=1 u ik x ik ,
M
i=1 u ik x ik )
U
( rs =1 vr k yrLj , rs =1 vr k yrMj , rs =1 vr k yrUj )
subject to m m m ≤ (1, 1, 1) ∀ j = 1, 2, 3, . . . , n;
( i=1 u ik xiLj , i=1 u ik xiMj , i=1 u ik xiUj )
(2)
u ik ≥ ε ∀i, vr k ≥ ε ∀r, ε > 0.
On applying division rule for two TFNs, Model 7 reduces to Model 8 as follows:
334 A. P. Singh and S. P. Yadav
Model 8 For k = 1, 2, 3, . . . , n,
s vr k y L s vr k y M s vr k y U
r =1 r =1 r =1
Max (E kO,L , E kO,M , E kO,U ) = m
rk , rk ,
U m u x M m u x L
rk (3)
i=1 u ik x ik i=1 ik ik i=1 ik ik
s M s
rs =1 vr k y L r =1 vr k yr j
U
r =1 vr k yr j
rj
subject to m ,
U m u x M m u x L
, ≤ (1, 1, 1) ∀ j = 1, 2, 3, . . . , n; (4)
i=1 u ik xi j i=1 ik i j i=1 ik i j
u ik ≥ ε ∀i, vr k ≥ ε ∀r, ε > 0.
Now, using the property stated in Eq. 5, Model 8 can be transformed into FMOO
DEA model as follows.
s m
vr k yrUj − u ik xiLj ≤ 0 ∀ j = 1, 2, 3, . . . , n; (12)
r =1 i=1
u ik ≥ ε ∀i, vr k ≥ ε ∀r, ε > 0.
This section presents an algorithm for solving the proposed FMOO and FMOP DEA
models. Algorithm is given as follows:
Step 1: First of all, for each DMU, convert MOOP to SOOP (single objective opti-
mization problem) with the help of the weighted sum method [17].
Step 2: In this step, random weights are generated. Suppose that there are d objectives
to be optimized with p variables. Then generate (100 × p) set of d weights.
There is no thumb rule for population selection. It depends upon the decision-
maker. For the computational purpose, we take a hundred times the number
of variables present in the problem.
Step 3: Solve the SOOP formulated in Step 1 with the help of the weights generated
in Step 2 for each DMU.
Step 4: 100 × p Pareto solutions are obtained from Step 3. Choose the most
favourable solution among 100 × p.
Step 5: The solution obtained in Step 4 is the efficiency of DMU.
Step 6: Use above steps to determine E kO∗ and E kP∗ .
After getting both optimistic and pessimistic efficiency scores we rank the DMUs
by considering both the efficiencies simultaneously. To rank the DMUs we will the
geometric average efficiency approach, proposed by Wang et al. [13]. According to
Wang et al. [13], if E kO∗ and E kP∗ are the optimistic and pessimistic efficiencies,
geometric
respectively, for D MUk , then the geometric average efficiency E k is defined
as follows:
= E k∗O × E k∗P
geometric
Ek (20)
Wang et al. [13] proposed geometric average efficiency approach for the DMUs
with crisp input and output data. We extend this idea to the DMUs with fuzzy input
and output data, particularly triangular fuzzy data.
5 Numerical Illustration
In this section, we use an example to demonstrate the efficacy of the proposed models.
This example is based on Guo and Tanaka’s study [18], which used five DMUs with
two fuzzy inputs and two fuzzy outputs. We try to validate the proposed models and
ranking approach using this scenario. A case study in the field of education is also
provided.
Table 4 presents the fuzzy input-output data for the problem considered by Guo and
Tanaka [18], which has five DMUs with 2 fuzzy inputs and 2 fuzzy outputs.
The optimistic and pessimistic efficiency scores are calculated by using the pro-
posed FMOO and FMOP DEA models, respectively. Table 5 presents the outcomes
(E k∗O and E k∗P ) of both the models.
Based on the efficiency scores (E k∗O , E k∗P ) obtained from FMOO and FMOP DEA
models; the geometric efficiency score is obtained by using Eq. (20). The ranking is
geometric
done based on the E k . The ranking obtained from the proposed FMODEA
model (see Table 7) is compared with Wang et al.’s [19] method. Since both the
methods provide different levels of efficiency scores; so it will be appropriate to
compare the ranks of DMUs with the help of Spearman’s rank correlation coefficient
338 A. P. Singh and S. P. Yadav
Data Collection
Recommendations
Fig. 1 Schematic view of the proposed hybrid FMODEA performance decision model
Performance Evaluation of DMUs Using Hybrid Fuzzy Multi-objective Data … 339
[20]. The correlation coefficient (ρ = 0.883) indicates that the efficiency obtained
from both models is highly correlated. This indicates that the proposed methodology
is quite efficient to rank the DMUs in fuzzy environment.
340 A. P. Singh and S. P. Yadav
Table 7 The proposed and geometric average efficiency scores and ranks of DMUs
E k∗O E k∗P
geometric
DMUs Ek Proposed Wang et al. Rank
rank
D MU A 0.6579 1.2611 0.9108 4 1.090 3
D MU B 0.7347 1.4335 1.0262 1 1.154 1
D MUC 0.6822 1.1158 0.8725 5 1.092 2
D MU D 0.8061 1.2151 0.9867 3 1.154 1
D MU E 0.8011 1.2675 1.0076 2 1.154 1
The ranking of DMUs depends upon the choice of the selection of input–output data.
The aim of this piece of work is to provide a novel approach for evaluating DMU’s
performance. The ranking is done based on the efficiencies obtained from Tables 9
and 10. The proposed ranking method is used and DMUs are ranked as presented in
Table 11.
342 A. P. Singh and S. P. Yadav
Table 11 The proposed geometric average efficiency scores and ranks of DMUs
E k∗O E k∗P
geometric
DMUs Ek Ranks
D1 0.7226 3.2013 1.5209 2
D2 0.6898 3.0711 1.4554 3
D3 0.5587 4.2401 1.5392 1
D4 0.4328 1.7888 0.8798 10
D5 0.4108 2.5246 1.1084 6
D6 0.3215 4.2338 1.1667 5
D7 0.5495 1.7457 0.9795 8
D8 0.4879 1.3323 0.8063 13
D9 0.3668 2.6345 0.9829 7
D10 0.3456 4.0468 1.1826 4
D11 0.3981 1.8937 0.8683 11
D12 0.3050 2.8733 0.9366 9
D13 0.3498 2.0613 0.8491 12
6 Conclusions
References
1. Charnes, A., Cooper, W.W., Rhodes, E.: Measuring the efficiency of decision making units.
Eur. J. Oper. Res. 2(6), 429–444 (1978)
2. Entani, T., Maeda, Y., Tanaka, H.: Dual models of interval DEA and its extension to interval
data. Eur. J. Oper. Res. 136(1), 32–45 (2002)
3. Azizi, H.: The interval efficiency based on the optimistic and pessimistic points of view. Appl.
Math. Model. 35(5), 2384–2393 (2011)
4. Azizi, H.: DEA efficiency analysis: a DEA approach with double frontiers. Int. J. Syst. Sci.
45(11), 2289–2300 (2014)
Performance Evaluation of DMUs Using Hybrid Fuzzy Multi-objective Data … 343
5. Arya, A., Yadav, S.P.: Development of optimistic and pessimistic models using FDEA to
measure performance efficiencies of DMUs. Int. J. Process. Manag. Benchmarking 9(3), 300–
322 (2019)
6. Gupta, P., Mehlawat, M.K., Yadav, S., Kumar, A.: Intuitionistic fuzzy optimistic and pessimistic
multi-period portfolio optimization models. Soft Comput. 1–26 (2020)
7. Puri, J., Yadav, S.P.: Intuitionistic fuzzy data envelopment analysis: an application to the banking
sector in India. Expert Syst. Appl. 42(11), 4982–4998 (2015)
8. Singh, A., Yadav, S., Singh, S.: A multi-objective optimization approach for DEA models in a
fuzzy environment. Soft Comput. (2022). https://doi.org/10.1007/s00500-021-06627-y
9. Wang, W.-K., Lu, W.-M., Liu, P.-Y.: A fuzzy multi-objective two-stage DEA model for eval-
uating the performance of us bank holding companies. Expert Syst. Appl. 41(9), 4290–4297
(2014)
10. Chen, W., Li, S.-S., Zhang, J., Mehlawat, M.K.: A comprehensive model for fuzzy multi-
objective portfolio selection based on DEA cross-efficiency model. Soft Comput. 24(4), 2515–
2526 (2020)
11. Boubaker, S., Do, D.T., Hammami, H., Ly, K.C.: The role of bank affiliation in bank efficiency:
a fuzzy multi-objective data envelopment analysis approach. Ann. Oper. Res. 1–29 (2020)
12. Zamani, P.: The position of multiobjective programming methods in fuzzy data envelopment
analysis. Int. J. Math. Model. Comput. 10(2) (SPRING), 95–101 (2020)
13. Wang, Y.-M., Chin, K.-S., Yang, J.-B.: Measuring the performances of decision-making units
using geometric average efficiency. J. Oper. Res. Soc. 58(7), 929–937 (2007)
14. Zimmermann, H.-J.: Fuzzy Set Theory-and Its Applications. Springer Science & Business
Media (2011)
15. Wang, J.-W., Cheng, C.-H., Huang, K.-C.: Fuzzy hierarchical TOPSIS for supplier selection.
Appl. Soft Comput. 9(1), 377–386 (2009)
16. Charnes, A., Cooper, W., Lewin, A.Y., Seiford, L.M.: Data envelopment analysis theory,
methodology and applications. J. Oper. Res. Soc. 48(3), 332–333 (1997)
17. Marler, R.T., Arora, J.S.: The weighted sum method for multi-objective optimization: new
insights. Struct. Multidiscip. Optim. 41(6), 853–862 (2010)
18. Guo, P., Tanaka, H.: Fuzzy DEA: a perceptual evaluation method. Fuzzy Sets Syst. 119(1),
149–160 (2001)
19. Wang, Y.-M., Yang, J.-B., Xu, D.-L., Chin, K.-S.: On the centroids of fuzzy numbers. Fuzzy
Sets Syst. 157(7), 919–926 (2006)
20. Zar, J.H.: Spearman rank correlation. Encycl. Biostat. 7 (2005)
Development of Intuitionistic Fuzzy Data
Envelopment Analysis Model Based
on Interval Data Envelopment Analysis
Model
1 Introduction
The traditional set theory originated from the crisp set theory, where either an element
belongs or does not belong to the set. There is no ambiguity regarding an elements’
belongingness/not belongingness in the set. For example, whether a student has
done the assignment or not. It is a yes or no answer to whether a particular element
(student) is contained in the set (of students that have done their assignments) or
not. The function that defines the containment of elements in the crisp set is called
the characteristic function. This concept of crisp set theory was extended by Zadeh
[27] when he gave the concept of fuzzy set theory. It was based on the fact that
there is always some ambiguity regarding the membership of elements in the set. For
example, consider the set of intelligent students in a class. The answer to whether a
student is intelligent could be a yes or no but it does not reveal all the information
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 345
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_24
346 M. Yadav and S. P. Yadav
regarding the question. However, a simple yes or no does not clarify the extent to
which a student is intelligent. So, Zadeh [27] introduced the concept of membership
function. For each element, the membership function defines a value between 0 to 1
to define its level of presence in the set. For example, based on the previous year’s
grades, the teacher concludes that the fuzzy set of intelligence of students could be
Ã, as given below. From the set, we get the information that Arti is intelligent, Rakhi
might be intelligent, Sunil is better than average, Arun is not really intelligent, and
Anil is not intelligent at all.
à = {(Aar ti, 1), (Rakhi, 0.5), (Sunil, 0.7), (Ar un, 0.2), (Anil, 0)}.
Now, based on the class performance of students, the teacher also knows the weak
points or what is lacking in the students, i.e., she can also estimate the level of
non-intelligence present in students. This led to the extension of fuzzy sets (FS) to
intuitionistic fuzzy sets (IFS) by introducing the concept of non-membership and
hesitation present in a variable by Atanassov [4]. Like membership function, he
defined the non-membership function to define the extent to which an element does
not belong to the set. For example, the above set can be rewritten as à I given below.
The set informs that Aarti is brilliant and does not lack anything, Rakhi might be
intelligent but she lacks a few qualities, Sunil is intelligent but lacks a few qualities;
Arun is not really intelligent and needs a lot of improvement and Anil is not intelligent
at all and needs improvement in everything. Here, the hesitation can be defined as
the lack of knowledge of teacher about each student in defining the membership and
non-membership grades. For Aarti, the hesitation is zero (1-1-0), for Rakhi, it is 0.2
(1-0.5-0.3), for Sunil, it is 0.1 (1-0.7-0.2), for Arun, it is 0.2 (1-0.2-0.6), and for Anil
it is zero (1-0-1).
à I = {(Aar ti, 1, 0), (Rakhi, 0.5, 0.3), (Sunil, 0.7, 0.2), (Ar un, 0.2, 0.6), (Anil, 0, 1)}
The IFS Ã I gives all the desired information about the set. There are well-defined
operations for addition, subtraction, multiplication, and division of IFSs [26]. Now,
the efficiency of a decision-making unit (DMU) under consideration is defined as
the ratio of the sum of weighted outputs to the sum of weighted inputs. One such
non-parametric technique to find the relative efficiency of homogeneous DMUs (with
crisp data) and to identify the best performer in the DMUs is the technique of data
envelopment analysis (DEA), proposed by Charnes et al. [8]. Wang et al. [24] pro-
posed an interval efficiency model for crisp data using DEA and extended it to fuzzy
sets. IF variables are preferred over fuzzy variables as they define linguistic vari-
ables mathematically in a more realistic way. IF variables convey the information of
membership, non-membership values, and the lack of knowledge of decision maker
in deciding these values. Fuzzy sets are unable to express this lack of knowledge.
Being a hot topic of research in Operations research, many studies have been done
to calculate the efficiency of DMUs with IF input and output variables [1, 11, 15,
16].
Development of Intuitionistic Fuzzy Data Envelopment Analysis Model … 347
Table 1 Example
Input Output
DMU A (2,0.8,0.1) (2,0.8,0.1)
DMU B (2,0.6,0.1) (2,0.6,0.1)
Most of the recent studies have ignored the effect of hesitation on efficiency.
Hesitation indirectly affects the efficiency of a DMU that utilizes IF inputs to produce
IF outputs. This effect can differ from small to large depending on the extent of
hesitation present in the DMU. For example, consider 2 DMUs A and B with single
input and output as given in Table 1.
Both the input and output of DMU A have a hesitation of 0.1. Hesitation of both the
input and output of DMU B is 0.3. Now though both DMUs are using the same amount
of input (2 units) to produce the same amount of output (2 units). The information
about DMU A is more precise than DMU B. In other words, the ambiguity regarding
the knowledge of input and output of DMU A is less than that of B. Lower hesitation
in variables means higher precision in membership and non-membership values.
Hence, DMU A should be more efficient than DMU B.
In the present study, we propose a new IF efficiency incorporated with hesitation
and identify the best performer among the set of homogeneous DMUs. We will also
rank the DMUs based on the proposed efficiency.
The rest of the paper is organized as follows. Section 2 deals with the recent
developments in IFS theory and IF efficiency. Section 3 defines the preliminaries of
IFSs and basic definitions of IFSs. Section 4 presents the methodology of the present
study, followed by the merits of the proposed model in Sect. 5, a numerical example
in Sect. 6, and conclusion in Sect. 7.
2 Literature Review
IFSs have a wide range of applications (in medical diagnosis [9], career determination
[12], pattern recognition [10], Clustering algorithm [6], multi-attribute information
classification [21], banking [19], risk assessment methodology [7], health sector [3],
transportation problem [22], and many more). The wide field of applications of IFSs
has also led to the development of various IF models ([13, 25], and many others).
Arya and Yadav [2] used the α − and β − cuts to calculate the efficiency of DMUs
with IF variables and proposed input and output targets. Santos Arteaga et al. [20]
developed an alphabetical approach to solve the IF efficiency model. Hajiagha et al.
[14] developed an IF efficiency model based on an aggregation operator. Moham-
madi Ardakani et al. [18] developed a model based on Stackelberg game theory and
efficiency decomposition using the best and worst return approach of Azizi and Wang
[5]. Szmidt et al. [23] gave a measure of the amount of knowledge conveyed by IFSs
348 M. Yadav and S. P. Yadav
3 Preliminaries
s
vrk .yrj
subject to rm=1 ≤ 1, 1 ≤ j ≤ n,
i=1 u ik .x ij
u ik ≥ 0, 1 ≤ i ≤ m,
vrk ≥ 0, 1 ≤ r ≤ s,
where u ik and vrk are unknown weights associated with the ith input and r th out-
put of DMUk .
The CCR model measures the relative efficiency of a homogeneous set of ’n’
DMUs in the interval (0,1]. DMUk is said to be efficient if the optimal value of
Model 1 is E k ∗ = 1 and the optimal weights u ik ∗ and vrk ∗ are not all zero.
3. Fuzzy Set [27]: A fuzzy subset à of a universal set X is defined by its mem-
bership function μ Ã : X → [0, 1], where the value of μ Ã (x) shows the degree of
membership of x in Ã.
4. Intuitionistic fuzzy set [4]: An IFS Ã I in the universal set X is defined by
where μ A˜I : X → [0, 1] is called the membership function and ν A˜I : X → [0, 1]
is called the non-membership function of à I . The function π A˜I : X → [0, 1]
defined by π A I (x) − ν
A I (x) = 1 − μ A I (x) denotes the degree of hesitation asso-
Development of Intuitionistic Fuzzy Data Envelopment Analysis Model … 349
ciated with the element x in à I . If the hesitation π A˜I (x) = 0 ∀ x ∈ X , then the
IFS à I is reduced to a fuzzy set à in X .
5. Triangular Intuitionistic Fuzzy Number (TIFN) [2]: A TIFN Ã I is written
as (a, b, c; a’, b, c’), is an IFS with the membership function μ A (x) and non-
membership function ν A (x) given by
x−a
b−a
, a < x ≤ b; b−x
b−a
, a < x ≤ b;
μ Ã I (x) = x−c
b−c
, b ≤ x < c; andν A˜I (x) = c −b , b ≤ x < c ;
x−b
6. Alpha cut [17] For α ∈ (0, 1], the α − cut of an IFS set à I , denoted by A I α , is
defined by A I α = {x ∈ X : μ A˜I (x) ≥ α}.
4 Methodology
1. Efficiency of TIFNs: Various studies are done to develop efficiency models for
DMUs with TIFN data. We propose to develop a new efficiency model to calcu-
late the efficiencies of DMUs using Wang’s efficiency model [24] with inputs and
outputs in the form of interval data. Let the interval input data be [xik l , xik u ] and
interval output data be [yrk l , yrk u ] for DMUk , k = 1, 2, . . . ,n. Then the interval
DEA model as presented by Wang [24] is given by
Model 2 s
v .[y l , y u ]
Max [θ k l , θk u ]= rm=1 rk rkl rk u
u ik .[xik , xik ]
s i=1
vrk .[yrj l , yrj u ]
u ≤ 1,
1 ≤ j ≤ n,
subject to rm=1
i=1 ik .[x ij , x ij ]
u l
u ik ≥ 0, 1 ≤ i ≤ m,
vrk ≥ 0, 1 ≤ r ≤ s.
Using interval arithmetic, we can split Model 2 into two linear models, namely,
lower bound and upper bound models, given by
[Model 3U] s
Max θku = i=1 vrk .yrk u
m
subject
s to i=1 u ik .xik l = 1
m
i=1 vrk .yrj − i=1 u ik .x ij ≤ 0, 1 ≤ j ≤ n,
u l
u ik ≥ 0, 1 ≤ i ≤ m,
vrk ≥ 0, 1 ≤ r ≤ s.
[Model 3L] s
Max θkl = i=1 vrk .yrk l
m
subject
s to i=1 u ik .xik u = 1
m
i=1 vrk .yrj − i=1 u ik .x ij ≤ 0, 1 ≤ j ≤ n,
u l
350 M. Yadav and S. P. Yadav
u ik ≥ 0, 1 ≤ i ≤ m,
vrk ≥ 0, 1 ≤ r ≤ s.
[θk l , θk u ] forms the efficiency interval of DMUk , k=1, 2, …, n.
Now, suppose that the inputs and outputs are TIFNs. Let the r th output of DMUk
be given by
ỹr k = (yrk l , yrk m , yrk u ; yrk l , yrk m , yrk u ) and the ith input of DMUk is given
by x̃ik = (xik , xik , xik ; xik , xik , xik u ).
l m u l m
For α ∈ (0, 1], the input α-cut is given by the closed interval (x˜ik )α = [αxik m +
(1 − α)xik l , αxik m + (1 − α)xik u ]. Similarly, the output α − cut is given by
(y˜rk )α = [α yrk m + (1 − α)yrk l , α yrk m + (1 − α)yrk u ].
So, the above interval efficiency model (Model 2) can be written as Model 4 given
below.
[Model 4] s
v .[αyrk m
+(1−α)yrk l
, αyrk m
+(1−α)yrk
u
]
Max [E k,α
l
, E k,α
u
]= mr =1 rk
i=1 u ik . [αx ik +(1−α)x ik , αx ik +(1−α)x ik ]
m l m u
s
v .[αyrk m
+(1−α)yrk l
, αyrk
m
+(1−α)yrk u
]
subject to mr =1 rk
. [αx +(1−α)x , αx
1 ≤ k ≤ n,
+(1−α)x
≤ [1, 1] ,
ik ]
m l m u
i=1 u ik ik ik ik
u rk ≥ 0, 1 ≤ r ≤ s,
vik ≥ 0, 1 ≤ i ≤ m.
Note that Model 4 is an interval DEA model which can be rewritten as two linear
programming problem given by Model 5U and Model 5L respectively. For com-
parability of different α − cuts and to keep the production possibility set same
for all α, we use α = 0 in the constraint [24].
[Model 5U]
Max E k,αu
= rs =1 vrk .(αyrk
m
+ (1 − α)yrk u
)
subject
m to m
s i=1 u ik . αx ik +
m (1 − α) x l
ik = 1,
v .y
r =1 rk rj
u
− u .x
i=1 ik ij
l
≤ 0, 1 ≤ j ≤ n,
u ik ≥ 0, 1 ≤ i ≤ m,
vrk ≥ 0, 1 ≤ r ≤ s.
[Model 5L]
Max E k,α l
= rs =1 vrk .(αyrkm
+ (1 − α)yrk
u
)
subject
m to
i=1 + (1 − α) xiku = 1,
u ik . αxikm
s m
r =1 vrk .yrj − i=1 u ik .x ij ≤ 0, 1 ≤ j ≤ n,
u l
u ik ≥ 0, 1 ≤ i ≤ m,
vr k ≥ 0, 1 ≤ r ≤ s.
Model 5U and Model 5L give the upper and lower bounds of the membership
function of each DMU for α ∈ (0, 1]. The interval [E k,α l
, E k,α
u
] forms the mem-
bership interval of efficiency for the kth DMU for respective α.
2. Hesitation: Consider the TIFN Ã I = a, b, c; a , b, c . The hesitation is
defined by π A˜I (x) = 1 − μ A˜I (x) − ν A˜I (x), where
Development of Intuitionistic Fuzzy Data Envelopment Analysis Model … 351
x−a
b−a
, a < x ≤ b; b−x
b−a
, a < x ≤ b;
μ A˜I (x) = c−x
c−b
, b ≤ x < c; and ν A˜I (x) = c −b , b ≤ x < c ;
x−b
l
E k,α + E k,α
u
E k,α =
2(1 + Hk )
The above defined efficiency incorporated with hesitation radially reduces the
average of efficiency interval of each DMU by the extent of hesitation present in
respective DMU. One can easily note that E k,α = 1.
352 M. Yadav and S. P. Yadav
4. Ranking of DMUs: Now, we can compare the DMUs on the basis of their effi-
ciency incorporated with hesitation, i.e., by taking E k,α as the ranking index. A
DMU with higher ranking index will perform better than the DMU with lower
ranking index. Now, rank the DMUs in decreasing order of their ranking index.
6 Numerical Example
Consider 5 firms that use 2 IF inputs to produce 2 IF outputs. The input/output data
for each firm is given in Table 2.
To check which firm is performing best, let us first find the efficiency interval for
each firm (or DMU) using Model 5. The calculated efficiency intervals for different
values of α = 0, 0.25, 0.5, 0.75 and 1 are given in Table 3. We calculate the
hesitation present in each firm using average hesitation. The hesitation in firm A, B,
C, D, and E come out to be 0.443, 0.291, 0.294, 0.473, and 0.383, respectively. Firm
D has the highest hesitation and firm B has the lowest hesitation. Now we calculate
the efficiency incorporated with hesitation in Table 4. At α =0.25, the averages of
efficiency interval for the firms A, B, C, D, and E are 0.762, 0.909, 0.768, 0.925,
and 0.767, respectively, and the efficiencies incorporated with hesitation are 0.528,
0.704, 0.593, 0.627, and 0.554, respectively. The reductions in efficiencies due to
incorporation of hesitation are 0.233, 0.204, 0.174, 0.297, and 0.212, respectively.
Firm D has faced the largest reduction as it faces bigger penalty. Firm B has faced the
smallest reduction as it faces comparably smaller penalty than other firms. Firm B
is the best performer and firm A is worst performer. The rankings obtained in Table
4 are the same for all values of α = 0, 0.25, 0.5, 0.75 and 1.
7 Conclusion
them on the basis of their efficiency scores. At last, we present an example to find
the efficiency incorporated with hesitation and rank a set of 5 hypothetical DMUs.
In the proposed model, we have used only membership function and hesitation
to calculate efficiency intervals of DMUs with inputs and outputs in form of TIFNs.
The model can be extended to DMUs with inputs and outputs as trapezoidal IFNs
with appropriate changes. We can also use non-membership function or β− cuts to
improve the efficiency calculated.
Acknowledgements The first author is deeply thankful to the funding agency UGC, Govt. of India,
for the financial support to carry out this research work.
References
1. Arya, A., Yadav, S.P.: Development of intuitionistic fuzzy super-efficiency slack based measure
with an application to health sector. Comput. Ind. Eng. 115, 368–380 (2018)
2. Arya, A., Yadav, S.P.: Development of intuitionistic fuzzy data envelopment analysis models
and intuitionistic fuzzy input–output targets. Soft Comput. 23(18), 8975–8993 (2019)
3. Arya, A., Yadav, S.P.: Performance efficiency of public health sector using intuitionistic fuzzy
DEA. Int. J. Uncertain., Fuzziness Knowl.-Based Syst. 28(02), 289–315 (2020)
4. Atanassov, K.: Review and new results on intuitionistic fuzzy sets. Preprint Im-MFAIS-1-88,
Sofia, 5, 1 (1988)
5. Azizi, H., Wang, Y.M.: . Improved DEA models for measuring interval efficiencies of decision-
making units. Measurement 46(3), 1325–1332 (2013)
6. Chaira, T.: A novel intuitionistic fuzzy C means clustering algorithm and its application to
medical images. Appl. Soft Comput. 11(2), 1711–1717 (2011)
7. Chang, K.H., Cheng, C.H.: A risk assessment methodology using intuitionistic fuzzy set in
FMEA. Int. J. Syst. Sci. 41(12), 1457–1471 (2010)
8. Charnes, A., Cooper, W.W., Rhodes, E.: Measuring the efficiency of decision making units.
Eur. J. Oper. Res. 2(6), 429–444 (1978)
9. De, S.K., Biswas, R., Roy, A.R.: An application of intuitionistic fuzzy sets in medical diagnosis.
Fuzzy Sets Syst. 117(2), 209–213 (2001)
10. Dengfeng, L., Chuntian, C.: New similarity measures of intuitionistic fuzzy sets and application
to pattern recognitions. Pattern Recognit. Lett. 23(1–3), 221–225 (2002)
11. Edalatpanah, S.A.: A data envelopment analysis model with triangular intuitionistic fuzzy
numbers. Int. J. Data Envel. Anal. 7(4), 47–58 (2019)
12. Ejegwa, P., Akubo, A., Joshua, O.: Intuitionistic fuzzy set and its application in career deter-
mination via normalized Euclidean distance method. Eur. Sci. J. 10(15) (2014)
13. Guha, D., Chakraborty, D.: A theoretical development of distance measure for intuitionistic
fuzzy numbers. Int. J. Math. Math. Sci. 2010 (2010)
14. Hajiagha, S.H.R., Akrami, H., Kazimieras Zavadskas, E., Hashemi, S.S.: An intuitionistic fuzzy
data envelopment analysis for efficiency evaluation under uncertainty: case of a finance and
credit institution. DSpace at University of West Bohemia (2013)
15. Hao, Z., Xu, Z., Zhao, H., Zhang, R.: Novel intuitionistic fuzzy decision making models in the
framework of decision field theory. Inf. Fusion 33, 57–70 (2017)
16. Liu, J., Song, J., Xu, Q., Tao, Z., Chen, H.: Group decision making based on DEA cross-
efficiency with intuitionistic fuzzy preference relations. Fuzzy Optim. Decis. Mak. 18(3), 345–
370 (2019)
17. Mahapatra, G., Roy, T.: Reliability evaluation using triangular intuitionistic fuzzy numbers
arithmetic operations. World Acad. Sci., Eng. Technol. 574–581 (2009)
Development of Intuitionistic Fuzzy Data Envelopment Analysis Model … 355
18. Mohammadi Ardakani, S., Babaei Meybodi, H., Sayyadi Tooranloo, H.: Development of a
bounded two-stage data envelopment analysis model in the intuitionistic fuzzy environment.
Adv. Oper. Res. 2022 (2022)
19. Puri, J., Yadav, S.P.: Intuitionistic fuzzy data envelopment analysis: an application to the banking
sector in India. Expert Syst. Appl. 42(11), 4982–4998 (2015)
20. Santos Arteaga, F.J., Ebrahimnejad, A., Zabihi, A.: A new approach for solving intuitionistic
fuzzy data envelopment analysis problems. Fuzzy Optim. Model. J. 2(2), 46–56 (2021)
21. Singh, P., Huang, Y.P., Wu, S.I.: An intuitionistic fuzzy set approach for multi-attribute infor-
mation classification and decision-making. Int. J. Fuzzy Syst. 22, 1506–1520 (2020)
22. Singh, S.K., Yadav, S.P.: Efficient approach for solving type-1 intuitionistic fuzzy transportation
problem. Int. J. Syst. Assur. Eng. Manag. 6(3), 259–267 (2015)
23. Szmidt, E., Kacprzyk, J., Bujnowski, P.: How to measure the amount of knowledge conveyed
by Atanassov’s intuitionistic fuzzy sets. Inf. Sci. 257, 276–285 (2014)
24. Wang, Y.M., Greatbanks, R., Yang, J.B.: Interval efficiency assessment using data envelopment
analysis. Fuzzy Sets Syst. 153(3), 347–370 (2005)
25. Xu, Z.: Models for multiple attribute decision making with intuitionistic fuzzy information.
Int. J. Uncertain., Fuzziness Knowl.-Based Syst. 15(03), 285–297 (2007)
26. Yadav, M., Yadav, S.P.: Performance efficiency measurement of MGNREGA 2018-19 in Indian
states and union territories based on DEA. Int. J. Syst. Assur. Eng. Manag. 1–8 (2022)
27. Zadeh, L.A.: Fuzzy sets. In: Fuzzy Sets, Fuzzy Logic, and Fuzzy Systems: Selected Papers by
Lotfi A Zadeh, pp. 394–432. World Scientific (1996)
Pricing Policy with the Effect of Fairness
Concern, Imprecise Greenness, and
Prices in Imprecise Market for a Dual
Channel
Keywords Game theory · Fuzzy theory · Fairness concern · Green supply chain
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 357
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_25
358 S. Ganguly et al.
1 Introduction
2 Literature Survey
Our findings are relevant to three areas of research: green supply chain, game in a
fuzzy environment, and fairness in supply chain.
Product greening in the supply chain has been studied in several research. Ghosh
and Shah [8] studied an apparel supply chain that addressed product greening. They
investigated the influence of greening expenses and customer sensitivity to green
clothes using game-theoretical models. Parsaeifar et al. [17] examined a supply chain
comprised of one manufacturer, many suppliers, and retailers, where the supplier
and retailer compete horizontally in Nash equilibrium, while everyone competes
vertically in Stackelberg equilibrium. Regarding green product decisions, they found
that their profit increases as retailer market competition expands but declines while
supplier market competition grows. Wang et al. [22] focuses on the selection and
coordination of green e-commerce supply chains under the fairness concerns of green
producers, taking into account the green degree of product and service provided by
the e-commerce platform. A decision-making problem with three-level green SC is
addressed by Das et al. [2] under various game structures. Das et al. [3] considered
interconnected three-stage forward and reverse SC consisting of green products.
They formulated and solved one centralized and decentralized models using a game.
The majority of past game theory research has focused on deterministic demands. In
recent works, several academics have employed fuzzy theory to express uncertain-
ties in supply chain models. In a fuzzy supply chain with two competitive producers
and a common retailer, Zhao et al. [27] examined the pricing policy of substitutable
products. Li [14] delves into a new set of ideas and methods for making decisions by
using intuitionistic fuzzy sets in real-world decisions and games. He researched the
methodologies and models of interval-valued cooperative games. Yiang and Xiao
[23] explored a green supply chain under several scenarios with government action.
Under demand uncertainty, the price strategies for a dual-channel supply chain con-
sidering the sales effort and green investment were studied by Wang and Song [21].
Liu et al. [16] proposed one centralized and three decentralized models for a closed-
loop SC with fuzzy demand and a variety of quality levels for second-hand products.
The optimal solutions are determined using the Stackelberg game and fuzzy cut-set
method. De et al. [4] developed a cost-minimization problem in a pollution-sensitive
production-transportation supply chain using binomial and Gaussian strategic fuzzy
Pricing Policy with the Effect of Fairness Concern, Imprecise Greenness, and Prices … 361
game approach. In our problem, the market demand, price elasticity, and product
greenness coefficient are all considered uncertain.
Social preferences and disparity result in fairness concerns among the supply chain
partners. Haitao Cui et al. [9] included fairness in a dyadic channel and investi-
gated the effect of fairness on channel coordination using a linear demand function,
claiming that a manufacturer can set the wholesale price higher than his marginal
cost for channel coordination to achieve maximum channel utility. Taking private
fairness concern as fuzzy, Liang and Qin [15] create an estimation model using
fuzzy theory. Sharma [19] examined the pricing decisions of a dyadic supply chain
with one fair-minded manufacturer and retailer under several gaming structures and
found that for the Stackelberg game, the manufacturer’s (retailer’s) profit decreases
(increases) with respect to her fairness and is uncertain for the Vertical Nash game.
Yoshihara and Matsubayashi [24] suggested a setup with a single manufacturer and
two competing fair-minded retailers. In a dual channel where manufacturers sell
items through online retailer, Du and Zhao [6] studied the combined effects of fair-
ness preference and channel preference on the firms’ operational strategies. Wang
et al. [20] investigated the pricing decisions of two competitive manufacturers under
horizontal and vertical fairness concerns.
We find from the above literature that none of them examined cognitive bias in
a green supply chain with online and offline channels in an uncertain environment.
Table 1 illustrates our work in context with the above literature survey. Through the
current investigation, we have attempted to fill this gap. In this study, we examine the
impact of retailer’s fairness concern on the optimal decisions of channel members
when the parameters are defined by fuzzy variables.
3 Model Formulation
Now, for avoiding the channel conflict, the selling prices of the green product are
considered the same in both offline and online channels, i.e., pr = pe = p.
Taking β̃ = b̃ − d̃, demand functions take the form:
All parameters which are characterized by fuzzy are independent and non-negative
by our assumption in this study. Also, E[b̃] > E[d̃] > 0. In the real world, customer
demand for both channels is non-negative. Therefore,
Pos({ãr − β̃ p + γ̃r θ } < 0) = 0 and Pos({ãe − β̃ p + γ̃e θ } < 0) = 0, where
Pos(A) is the possibility that event A will occur.
The profits of the manufacturer and retailer can be expressed as follows:
364 S. Ganguly et al.
1 2
πm = (w − c) D̃r + ( p − c) D̃e − Iθ (1)
2
πr = ( p − w) D̃r (2)
The greening cost is considered as 21 θ 2 in the profit function [12]. The total profit of
the supply chain becomes
1 2
πc = ( p − c)[(ãr + ãe ) − 2β̃ p + (γ̃r + γ̃e )θ ] − Iθ (3)
2
3.1 Preliminaries
Let (ϑ, P(ϑ), Pos) be a possibility space, where ϑ is a non-empty set, P(ϑ) is its
power set, and Pos is a possibility measure. Each element in P(ϑ) is referred to
as a fuzzy event, and for each event A, Pos(A) denotes the possibility that event A
occurs.
• The credit index of A is defined as Cr(A)= 21 (1 + Pos(A) − Pos(Ac )), where Ac
is the complement of A.
• A fuzzy variable ξ has its expected value as
+∞ 0
Cr ({ξ ≥ z} dz + Cr ({ξ ≤ z} dz
0 −∞
.
• If ξ = (x1 , x2 , x3 ) is a triangular fuzzy variable, then the expected value of ξ is
E[ξ ] = x1 +2x42 +x3 .
1
• If its expected value is finite, then E[ξ ] = 21 0 (ξαL + ξαU ) dα.
4 Formulation of Models
Both centralized and decentralized models are formulated for the two-echelon SC
(see Fig. 1).
Pricing Policy with the Effect of Fairness Concern, Imprecise Greenness, and Prices … 365
In this scenario, the manufacturer and the retailer act in an integrated way to maximize
the total expected SC profit, or E[πc ( p, θ )]. The centralized model is described as
follows:
1 2
max E[πc ( p, θ)] = max ( p − c)[E[ãr ] + E[ãe ] − 2 pE[β̃] + θ(E[γ̃r ] + E[γ̃e ])] − Iθ
( p,θ) ( p,θ) 2
subject to
Pos({ãr − β̃ p + γ̃r θ } < 0) = 0, Pos({ãe − β̃ p + γ̃e θ } < 0) = 0 (4)
p − c > 0, p > 0
hold, where Ar = E[ãr ] − E[β̃]c and Ae = E[ãe ] − E[β̃]c, then the optimal retail
price and greenness level are given respectively as p = c + ΔI (Ar + Ae ), θ =
(E[γ̃r ]+E[γ̃e ])
Δ
(Ar + Ae )
∂E[πc ]
= 0 =⇒ −4 pE[β̃] + θ (E[γ̃r ] + E[γ̃e ]) + 2E[β̃]c + E[ãr ] + E[ãe ] = 0
∂p
(5)
∂E[πc ]
= 0 =⇒ (E[γ̃r ] + E[γ̃e ]) p − I θ − (E[γ̃r ] + E[γ̃e ])c = 0 (6)
∂θ
−4E[β̃] E[γ̃r ] + E[γ̃e ]
The Hessian matrix obtained is H = = [4I E[β̃] −
E[γ̃r ] + E[γ̃e ] −I
(E[γ̃r ] + E[γ̃e ]) ] = Δ.
2
For maintaining the concavity of the retailer’s expected profit, the first principal
minor of the Hessian matrix should be negative, i.e., −4E[β̃] < 0 and the second
principal minor should be positive, i.e., 4I E[β̃] − (E[γ̃r ] + E[γ̃e ])2 > 0. For feasi-
bility of the optimal decisions, I > (E[γ̃r4E[
]+E[γ̃e ])2
β̃]
must be satisfied. By solving Eqs. (5)
and (6), p and θ are derived, provided the possibility conditions are satisfied and the
theorem is proven.
366 S. Ganguly et al.
πrf = πr − λ[πm,r − πr ]+
f 1
max E[πr ( p)] = max[(1 + λ) p − (1 + 2λ)w + λc][E[ãr ] − E[β̃] p + E[γ̃r ]θ] + λI θ 2
p p 2
(8)
subject to Pos({ãr − β̃ p + γ̃r θ } < 0) = 0.
Lemma 1 Using the backward induction method, the retailer’s optimal decision
(retail price) is obtained as
Pricing Policy with the Effect of Fairness Concern, Imprecise Greenness, and Prices … 367
1
pf = [E[β̃](1 + 2λ)(w − c) + (1 + λ)(E[ãr ] + E[γ̃r ]θ + E[β̃]c)]
2E[β̃](1 + λ)
(9)
provided 2E[β̃](1 + λ)(ãrU0+ + γ̃rU0+ θ ) > β̃0L+ [E[β̃](1 + 2λ)(w − c) + (1 + λ)(E[ãr ]
+ E[γ̃r ]θ + E[β̃]c)] holds.
Proof From the possibility constraint of the retailer’s problem, we get 2E[β̃](1 +
λ)(ãrU0+ + γ̃rU0+ θ ) > β̃0L+ [E[β̃](1 + 2λ)(w − c) + (1 + λ)(E[ãr ]+E[γ̃r ]θ +E[β̃]c)].
The retail price of the green product is obtained by solving the first-order condi-
tion of the retailer’s expected profit
f
∂E[πr ]
= 0 =⇒ −2(1 + λ)E[β̃] p + (1 + 2λ)E[β̃](w − c) + (1 + λ)(E[β̃]c + θ E[γ̃r ] + E[ãr )] = 0
∂p
(10)
f
∂ 2 E[πr ]
The second derivative of the expected profit function of the retailer is ∂ p2
=
−2E[β̃](1 + λ) < 0. Hence, the retailer’s expected profit function is concave in p.
Replacing this optimal price in the manufacturer’s problem, we evaluate the optimal
decisions (wholesale price and greenness).
β̃0L+ E[γ̃r ]) > β̃0L+ [(1 + 2λ)(G 3 Ar + G 4 Ae ) + Δ1 Ar ] (i = r, e) hold, then the opti-
mal wholesale price and greenness level are given as
1+λ 1
wf = c + [G 3 Ar + G 4 Ae ] and θ f = [G 1 Ar + (1 + 2λ)G 2 Ae ]
E[β̃]Δ1 Δ1
To maintain concavity of the profit function of the manufacturer, the first principal
minor should be less than 0, i.e., −E[β̃](1+2λ)(3+4λ)
1+λ
< 0 and second principal minor
should be positive, i.e., Δ1 > 0 should be maintained. By solving Eqs. (11) and (12),
w f and θ f are obtained, provided the possibility condition is satisfied and the theorem
is proven.
Particular case: When the retailer is not fairness sensitive about the offline channel,
then the above-decentralized model reduces to the general model with all rational
channel members where the previously mentioned parameters remain fuzzy in nature.
E[γ̃e ])γ̃i0+ − β̃0L+ B4 ] > 2E[β̃]Δ1 cβ̃0L+ (i = r, e) hold, then the optimal wholesale
U
and
1
θ= [3δγ Ar + (3E[γ̃r ] + E[γ̃e ])Ae ]
Δ1
5 Numerical Experiments
In this section, numerical results of the centralized and decentralized models are
provided against the same data. Experts’ expertise is frequently used to determine
the relationship between linguistic variables and fuzzy triangular numbers for base
market demand, self-price elasticity, cross-price elasticity, and coefficient of product
greenness. For the numerical example, Table 2 shows linguistic variables as well
as fuzzy triangular numbers. Based on Table 2, we have ãr = (1200, 1500, 1700),
ãe = (1100, 1250, 1400), b̃ = (40, 50, 65), d̃ = (30, 40, 55).
For these data, the possibility constraints of the models are satisfied. According to
preliminaries, triangular fuzzy variable ξ = (x1 , x2 , x3 ) has expected value E[ξ ] =
x1 +2x2 +x3
4
. Thus, the expected values of the fuzzy parameters in Table 2 are as follows:
E[ãr ] = 1475, E[ãe ] = 1250, E[β̃] = 10.
We present the output values for optimal decisions and channel members’ expected
profits for the decentralized model in Table 3.
From Table 4, it is noticed that when E[γr ] increases, the expected profit of the
manufacturer increases up to a certain limit, then decreases but for the fair-minded
retailer and overall SC, the profits escalates. The level of product greenness and retail
price increases when customers’ sensitivity to product greenness level grows in the
Pricing Policy with the Effect of Fairness Concern, Imprecise Greenness, and Prices … 369
offline channel but the wholesale price decreases. When customers’ sensitivity about
product greenness increases more in the online channel, all the optimal decisions and
corresponding expected profits of the channel members increase. Also, from Table
4, it is observed that the retailer’s fairness is harmful to the decisions of the channel
members but beneficial for her own profit.
370 S. Ganguly et al.
In this section, we compare fuzzy solutions and deterministic solutions of the cen-
tralized and decentralized models (with and without retailer’s fairness) numerically,
where we assumed the value of fuzzy parameters (expected value) near the determin-
istic values. These values are as per the expert’s opinions. The difference between the
optimal values in percentage is shown in Table 3. It is noticed that the highest differ-
ence between the centralized scenario and decentralized scenario with and without
the cognitive bias is in product greenness level. For the comparison, the determinis-
tic decentralized models with and without the cognitive bias (fairness concern) are
solved. The optimal decisions of the manufacturer for the deterministic model with
retailer’s fairness are as follows:
wf = c + 1+λ
E[β̃]Δ2
[G 3 Ar + G 4 Ae ] and θ f = 1
Δ2
[G 1 Ar + (1 + 2λ)G 2 Ae ]
where Ar = ar − βc, Ae = ae − βc and Δ2 > 0 should be satisfied and the opti-
mal retail price can be obtained similarly following previous section (Sect. 4).
Pricing Policy with the Effect of Fairness Concern, Imprecise Greenness, and Prices … 371
6 Sensitivity Analyses
To illustrate the effect of λ for the offline and online channels, on optimal retail
and wholesale prices of the green product, product greenness level as well as
expected profits of the channel members, the following numerical values of the model
parameters are considered: ãr = (1200, 1500, 1700), ãe = (1100, 1250, 1400), b̃ =
(40, 50, 65), d̃ = (30, 40, 55), γ̃r = (10, 15, 20), γ̃e = (21, 26, 31), and c = 20,
I = 100. We evaluate the optimal decisions presented in Fig. 2 w.r.t. λ.
When the retailer exhibits more and more fairness about the offline channel w.r.t.
the manufacturer from Fig. 2, it is observed that the wholesale price of the man-
ufacturer decreases. Consequently, it is also realized that an increase in retailer’s
fairness concern has an adverse impact on both channels’ retail prices as well as
on the greening level of the product. The power of bargaining increases when the
retailer’s fairness concern intensifies, so for the sake of a fair deal, she will force
the manufacturer to reduce her wholesale price to such an extent that their profits
are not jeopardized. Hence, the retailer is focused on her fairness more, and both
channel members’ optimal prices descend. Also, the fairness of retailer is harmful
to the environmental performance of the supply chain.
Effect of λ on channel members’ profits: The effect of the fairness concern coef-
ficient λ on the expected profits of the channel members is depicted in Fig. 3. From
Fig. 3, it is clear that when a retailer’s fairness w.r.t. the manufacturer increases, the
expected profit of the manufacturer diminishes but her own revenue gets benefited.
Now, when the retailer is fair-minded, the manufacturer sacrifices her wholesale price
to care about her concern for fairness. So, the expected profit of the manufacturer
is declined. It is inferred that the manufacturer cannot be aggressive while choosing
her strategy when the retailer is fair-minded. Hence, the retailer’s fairness concern
about the offline channel w.r.t. the manufacturer is favorable for her expected profit
but detrimental for the manufacturer.
Pricing Policy with the Effect of Fairness Concern, Imprecise Greenness, and Prices … 373
The effect of the fuzzy degree of product greenness coefficient for offline and online
channels on expected profits of the channel members (decentralized model) as well
as on expected total SC’s profit (centralized model) is depicted in Figs. 4 and 5,
respectively. From Fig. 4, it is noticed that with the increment of customer’s sensitivity
to product greenness level in offline channel, the expected profit of the manufacturer
increases first, until E[γ̃r ] exceeds a certain amount and when E[γ̃r ] > 11.4, her
profit decreases. Also, the total expected SC’s profit in the centralized model as
well as the retailer’s expected profit increases when E[γ̃r ] increases for our chosen
range of E[γ̃r ]. Hence, when customers prefer product greenness more, the retailer
(for decentralized channel) and overall supply chain profit (for centralized channel)
get an advantage, but it is beneficial for the manufacturer up to a certain limit,
then it is detrimental for her expected profit. It can also be seen from Fig. 5 that
when the product greenness coefficient increases in the online channel, the channel
members’ expected profits for the decentralized model as well as the total SC’s
profit for the centralized model increase. Therefore, if E[γ̃e ] increases in the online
channel, the channel members’ expected profits are enhanced. As the manufacturer’s
and retailer’s expected profit increase with the coefficient of product greenness for a
certain range, the total expected profit for SC (centralized model) also grows. In the
market, where consumers are highly sensitive to greenness, they frequently choose
products with high levels of greening. The development of new products lead to
higher cost, resulting in increase in sales, so the overall expected profit of the SC
increases.
When the cost for product greening increases more, the expected profits of the channel
members (decentralized model) when the retailer is fair-minded as well as the total
SC’s profit (centralized model) decrease. This follows naturally and is justified by
Fig. 6.
Pricing Policy with the Effect of Fairness Concern, Imprecise Greenness, and Prices … 375
7 Conclusion
(iii) Customer’s sensitivity toward product greening for the online channel has a
beneficial impact on all channel members’ profits in the decentralized model as well
as on total supply chain profit in the centralized model.
(iv) Increment in customer’s sensitivity toward product greening for the retail channel
is helpful for the total expected supply chain profit in the centralized model and the
retailer’s expected profit in the decentralized model but has a positive effect on the
manufacturer’s expected profit up to a certain limit.
(v) The manufacturer’s expected profit decreases with the retailer’s fairness, but the
retailer’s profit enhances.
(vi) All channel members’ profits, as well as total SC’s profit, reduce when the cost
of greening increases.
The present study has the following limitations: We have considered a two-level SC
with two members—one manufacturer and one retailer only. However, an SC with
several channel members can be considered. In this study, fairness concern is repre-
sented by crisp values, though this can be uncertain in nature.
Nomenclature
References
1. Chen, X., Wang, X., Jiang, X.: The impact of power structure on the retail service supply chain
with an o2o mixed channel. J. Oper. Res. Soc. 67(2), 294–301 (2016)
2. Das, M., Jana, D.K., Alam, S.: Comparative study of centralized and decentralized scenarios
of a three-tiered green supply chain in two-period using the game theoretical approach. Clean.
Logist. Supply Chain. 100054 (2022)
3. Das, M., Jana, D.K., Alam, S.: Game theoretic analysis of a three-stage interconnected forward
and reverse supply chain. Environ. Dev. Sustain. 24(6), 7976–8007 (2022)
4. De, S.K., Bhattacharya, K., Bhattacharya, P.P., Nayak, P.K., Joardar, S.: Solution of a pollution
sensitive supply chain model for novel strategic fuzzy game via bernoulli trial. Comput. &
Oper. Res. 144, 105846 (2022)
5. Du, B., Liu, Q., Li, G.: Coordinating leader-follower supply chain with sustainable green
technology innovation on their fairness concerns. Int. J. Environ. Res. Public Health. 14(11),
1357 (2017)
6. Du, X., Zhao, W.: Managing a dual-channel supply chain with fairness and channel preference.
Math. Probl. Eng. 2021 (2021)
7. Fehr, E., Schmidt, K.M.: A theory of fairness, competition, and cooperation. Q. J. Econ. 114(3),
817–868 (1999)
8. Ghosh, D., Shah, J.: A comparative analysis of greening policies across supply chain structures.
Int. J. Prod. Econ. 135(2), 568–583 (2012)
9. Haitao Cui, T., Raju, J.S., Zhang, Z.J.: Fairness and channel coordination. Manag. Sci. 53(8),
1303–1314 (2007)
10. He, N., Jiang, Z.Z., Sun, M., Sheng, Y.: Group buying of competing retailers with fairness
concerns. J. Syst. Sci. Syst. Eng. 1–20 (2020)
11. Jamali, M.B., Gorji, M.A., Iranpoor, M.: Pricing policy on a dual competitive channel for a
green product under fuzzy conditions. Neural Comput. Appl. 1–13 (2021)
12. Jamali, M.B., Gorji, M.A., Iranpoor, M.: Pricing policy on a dual competitive channel for a
green product under fuzzy conditions. Neural Comput. Appl. 33, 1–13 (2021). https://doi.org/
10.1007/s00521-020-05567-2
13. Ke, H., Wu, Y., Huang, H., Chen, Z.: Pricing decision in a two-echelon supply chain with
competing retailers under uncertain environment. J. Uncertain. Anal. Appl. 5(1), 1–21 (2017)
14. Li, D.F.: Decision and Game Theory in Management with Intuitionistic Fuzzy Sets, vol. 308.
Springer (2014)
378 S. Ganguly et al.
15. Liang, G.B., Qin, Y.H.: The game model of fuzzy fairness concern for supplier and retailer. J.
Interdiscip. Math. 20(6–7), 1537–1542 (2017)
16. Liu, W., Liu, W., Shen, N., Xu, Z., Xie, N., Chen, J., Zhou, H.: Pricing and collection decisions
of a closed-loop supply chain with fuzzy demand. Int. J. Prod. Econ. 245, 108409 (2022)
17. Parsaeifar, S., Bozorgi-Amiri, A., Naimi-Sadigh, A., Sangari, M.S.: A game theoretical for
coordination of pricing, recycling, and green product decisions in the supply chain. J. Clean.
Prod. 226, 37–49 (2019)
18. Ranjan, A., Jha, J.: Pricing and coordination strategies of a dual-channel supply chain consid-
ering green quality and sales effort. J. Clean. Prod. 218, 409–424 (2019)
19. Sharma, A.: Game-theoretic analysis of pricing models in a dyadic supply chain with fairness
concerns. Int. J. Strat. Decis. Sci. (IJSDS). 10(2), 1–24 (2019)
20. Wang, D., Ge, G., Zhou, Y., Zhu, M.: Pricing-decision analysis of green supply chain with two
competitive manufacturers considering horizontal and vertical fairness concerns. Environ. Sci.
Pollut. Res. 1–24 (2022)
21. Wang, L., Song, Q.: Pricing policies for dual-channel supply chain with green investment and
sales effort under uncertain demand. Math. Comput. Simul. 171, 79–93 (2020)
22. Wang, Y., Fan, R., Shen, L., Jin, M.: Decisions and coordination of green e-commerce supply
chain considering green manufacturer’s fairness concerns. Int. J. Prod. Res. 58(24), 7471–7489
(2020)
23. Yang, D., Xiao, T.: Pricing and green level decisions of a green supply chain with governmental
interventions under fuzzy uncertainties. J. Clean. Prod. 149, 1174–1187 (2017)
24. Yoshihara, R., Matsubayashi, N.: Channel coordination between manufacturers and competing
retailers with fairness concerns. Eur. J. Oper. Res. 290(2), 546–555 (2021)
25. Zadeh, L.A., Klir, G.J., Yuan, B.: Fuzzy Sets, Fuzzy Logic, and Fuzzy Systems: Selected
Papers, vol. 6. World Scientific (1996)
26. Zhang, R., Ma, W., Si, H., Liu, J., Liao, L.: Cooperative game analysis of coordination mech-
anisms under fairness concerns of a green retailer. Eur. J. Oper. Res. 59, 102361 (2021)
27. Zhao, J., Tang, W., Zhao, R., Wei, J.: Pricing decisions for substitutable products with a common
retailer in fuzzy environments. Eur. J. Oper. Res. 216(2), 409–419 (2012)
28. Zhao, Y.F., Wang, Y., Shi, G.Q.: The impact of consumers’ peer-induced fairness concerns on
mixed retail and e-tail channels. Discret. Dyn. Nat. Soc. 2020 (2020)
Fuzzy Set Theory
A Similarity Measure of Picture Fuzzy
Soft Sets and Its Application
Keywords Picture fuzzy sets · Soft sets · Picture fuzzy soft sets
1 Introduction
In 1965, Zadeh [4] invented the notion of fuzzy set (FS), an approach for dealing
with uncertainties. Researchers have given a lot of attention to the FS theory, and
numerous experts have applied it to a variety of situations. In fuzzy sets, there is a
function that returns the degree of membership of an object to a non-empty set from
a non-empty set to the closed unit interval [0, 1]. It is a framework for dealing with
ambiguity, exaggeration, and confusion, as well as assigning a degree of membership
to each member of the nature of discussion to a subset of it.
A soft set, which is a parameterized family of subsets of a crisp universal set,
could handle more data which contains uncertainty. Using the parametrization of
tools methodology, Molodtsov [1] offered the idea of soft sets in 1999. Molodtsov
has effectively employed soft set theory in a variety of regions including game theory,
operations research, and Reimann integration. Soft set theory offers a wide range of
applications, only a few of which Molodtsov demonstrated in his pioneering work
[1]. Yang et al. [5] defined picture fuzzy soft set (PFS f S) as an extended version of
soft sets.
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 381
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_26
382 V. Salsabeela and S. J. John
Similarity measures are a crucial metric for analyzing the degree to which two things
are alike. The idea of similarity measure is significant in practically every sector of
research and engineering. Many researchers have now developed many similarity
measures and applied them to MADM, pattern identification, mineral field recog-
nition, building material recognition, strategy decision-making, and other domains.
We frequently want to know if two patterns or images are alike or nearly alike, or
at the very least to what extent they are alike. Various academics have explored
the topic of measuring similarity between fuzzy sets, fuzzy numbers, and ambigu-
ous sets, including Chen, [6, 7] and [8]. Li and Xu, [9] Hong and Kim, [10] C. P.
Pappis, [11, 12], and others. However, in [13], Kharal provides counterexamples to
demonstrate various inaccuracies, as well as introduce some set operations on soft set
distances and similarity measurements based on distance. According to the Hamming
distance, Euclidean distance, and their normalized correspondents, Atanassov [14],
Szmidt, and Kacprzyk [15] suggested a similarity measure for intuitionistic fuzzy
sets. Guiwu Wei and Yu Wei [16] suggested ten similarity measures in accordance
with the cosine function for Pythagorean fuzzy sets. Guiwu Wei [17] has presented
eight cosine-based similarity measurements between picture fuzzy sets. Kifayat et al.
[18] recently discussed certain gray similarity measures, cosine similarity measures,
and set-theoretic similarity measures that can be used to compare spherical fuzzy
sets and T-spherical fuzzy sets. Cuong [19] recently suggested the picture fuzzy set
(PFS) and examined a few of its fundamental operations and features.
Here, put forward a similarity measure and a weighted similarity measure for
PFS f S and explain with an illustrative example related with our real-life situation.
The following is the format of the paper’s presentation. The concepts of soft sets,
fuzzy soft sets, picture fuzzy sets, and similarity measures between two fuzzy soft
sets are reviewed in segment 2. Part 3 introduces the concept of PFS f S similarity
measure and proposes an equation for finding the similarity measure and a weighted
similarity measure between two PFS f S. The mentioned similarity measure is applied
to a problem related with the medical field in Sect. 4. Section 5 wraps up the project
with some recommendations for the future.
2 Preliminaries
In the present section, we will review certain basic concepts for soft sets, fuzzy soft
sets, picture fuzzy sets, picture fuzzy soft sets, and similarity measures between fuzzy
soft sets, that would be important in the following talks.
Consider = {1 , 2 , 3 , . . . n } as the universe of discourse and K =
{k1 , k2 , k3 , . . . kn } is the set of parameters.
Definition 1 ([1]) A pair (, ) is said to be a soft set over the universe , in which
is a mapping defined by
: → P().
A Similarity Measure of Picture Fuzzy Soft Sets and Its Application 383
Definition 2 ([2]) Consider as the universe, be the set of parameters and P()
denote the collection of all fuzzy soft sets over . A pair (, ) is said to be a fuzzy
soft set over the universe , in which : → P() is a function defined from
→ P().
Then for ∈ , πP () = 1 − (P () + ϒP () + P ()) is called the refusal-
membership degree of in . For simpleness, we call P(P (), ϒP (), P ())
a picture fuzzy number (PFN) denoted by k = P(k , ϒk , k ), where k , ϒe , k ∈
[0, 1],
πk = 1 − (k + ϒk + k ), and k + ϒk + k ≤ 1.
Definition 4 ([17]) Let P1 and P2 be two picture fuzzy sets defined on the universe
of discourse . Then
P1 ⊆ P2 , i f P1 () ≤ P2 (), ϒP1 () ≤ ϒP2 (), P1 ()) ≥ P2 ())
where (k) () is the positive membership degree, ϒ(k) () is the neutral member-
ship degree and (k) () is the negative membership degree function respectively
with the condition, (k) () + ϒ(k) () + (e) () ≤ 1. If for any parameter k ∈
and for any ∈ , ϒ(k) () = 0, then (k) will become a Pythagorean fuzzy set
and , will turn into an intuitionistic fuzzy soft set if it is true for all k ∈ . Here,
P F S f S() represents the set of all picture fuzzy soft sets over .
384 V. Salsabeela and S. J. John
The idea of similarity measures between two PFS f Ss is discussed in this section, as
well as its equivalent weighted similarity measure, is also mentioned here.
Definition 8 Consider as an initial universal set and K as the set of all param-
eters. Suppose P F S f S() represents the set of all picture fuzzy soft sets defined
on the universe . A mapping Sm : P F S f S() × P F S f S() → [0, 1] is said to be
a similarity measure in between fuzzy soft sets , and , , represented as
Sm ( , , , ), if it satisfies the given conditions:
1. 0 ≤ Sm ( , , , ) ≤ 1;
2. Sm ( , , , ) = Sm ( , , , );
3. Sm ( , , , ) = 1 iff = ;
4. Let , ∂ be a picture fuzzy soft set, if , ⊆ , and
, ⊆ , ∂ , then Sm ( , , , ∂ ) ≤ Sm ( , , , ) and
Sm ( , , , ∂ ) ≤ Sm ( , , , ∂ ).
Now, for any PFS f S , ∈ PFS f S(), ⊆ K , we will extend the PFS f S to
the PFS f S ,ˆ in which (k) = φ,∀k ∈ / , that is, (k)
ˆ () = ϒ(k)
ˆ () =
(k)
ˆ () = 0, ∀k ∈ / . As a result, we will now treat the parameter subset of every
PFS f S over as if it were the set of parameter K , without sacrificing generality.
Definition 9 Consider = {u 1 , u 2 , . . . u n } as a universe with the parameter set K =
{k1 , k2 , . . . km }. Then, a similarity measure between two PFS f S , K and , K
can be found using the given formula:
A Similarity Measure of Picture Fuzzy Soft Sets and Its Application 385
1 1 − 21 min{|(ki ) (u j ) − (ki ) (u j )|, |(ki ) (u j ) − (ki ) (u j )|} + |ϒ(ki ) (u j ) − ϒ(ki ) (u j )|
m n
Sm (, ) =
mn
i=1 j=1
1 + 2 max{|(ki ) (u j ) − (ki ) (u j )|, |(ki ) (u j ) − (ki ) (u j )|} + |ϒ(ki ) (u j ) − ϒ(ki ) (u j )|
1
(1)
W P F S f S (, ) (2)
1
m n
1 − 21 min{|(ki ) (u j ) − (ki ) (u j )|, |(ki ) (u j ) − (ki ) (u j )|} + |ϒ(ki ) (u j ) − ϒ(ki ) (u j )|
= ∅i
n
i=1 j=1
1 + 2 max{|(ki ) (u j ) − (ki ) (u j )|, |(ki ) (u j ) − (ki ) (u j )|} + |ϒ(ki ) (u j ) − ϒ(ki ) (u j )|
1
We offer a strategy for solving a medical diagnosis problem in accordance with the
suggested similarity measure of P F S f Ss in this section. In the following application,
a similarity measure between two P F S f Ss could be used to determine if a person
has an illness or not.
A Similarity Measure of Picture Fuzzy Soft Sets and Its Application 387
Diarrhea is a common issue characterized by loose, watery, and possibly more fre-
quent bowel motions. It can occur on its own or in conjunction with other symptoms
such as nausea, vomiting, stomach discomfort, or weight loss.
Diarrhea is characterized by loose, watery feces occurring three or more times
per day. Diarrhea can be acute, chronic, or recurrent. Acute diarrhea is far more
common than chronic or persistent diarrhea. Dehydration and malabsorption are two
side effects of diarrhea. If your diarrhea lasts fewer than four days, your doctor
may not need to investigate the cause. If your diarrhea persists or you have other
symptoms, your doctor may consult your medical and family history, perform a
physical examination, or order tests to determine the cause. Here, we are attempting
to determine how likely it is that a patient with particular apparent symptoms is
suffering from diarrhea.
With the help of a medical officer, here, we first create a P F S f S for the illness
and a P F S f S for the sick person. After that, we calculate the similarity measures
between these two P F S f Ss. If they are considerably comparable, we can deduce
that the person is suffering from diarrhea. If they are not, we can rule out diarrhea.
Example 2 Suppose that there are two patients P1 , P2 admitted to a hospital with
those kinds of signs and symptoms of diarrhea. Suppose that universal set consists of
the three elements “harsh” (ϕ1 ), “soft” (ϕ2 ), and “no” (ϕ3 ). That is ϕ = {ϕ1 , ϕ2 , ϕ3 }.
Consider = {1 , 2 , 3 , 4 , 5 , 6 } as the set of parameters which indicates
definite symptoms of diarrhea, where 1 stands for watery, 2 stands for frequent
bowel movements, 3 stands for cramping or pain in the abdomen, 4 stands for
nausea, 5 stands for bloating, and 6 stands for bloody stools.
Suppose F, be a P F S f S over the universe ϕ for diarrhea, developed by
utilizing the assistance of a civil medical officer provided in Table 1.
The P F S f Ss is then constructed for two patients, as shown in Tables 2 and 3.
Therefore, we get S P F S f S F, P1 = 0.8311 ≥ 0.75 and S P F S f S F, P2 =
0.5719 ≤ 0.75. In light of these findings, we can state that the patient P1 is pos-
sibly suffering from diarrhea.
5 Conclusion
References
1. Molodtsov, D.: Soft set theory-first results. Comput. & Math. Appl. 37, 19–31 (1999)
2. Maji, P.K., Biswas, R., Roy, A.R.: Fuzzy soft sets. J. Fuzzy Math. 9, 589–602 (2001)
3. Cuong, B.C., Kreinovich, V.: Picture fuzzy sets. J. Comput. Sci. Cybern. 30, 409–420 (2014)
4. Zadeh, L.A.: Fuzzy, sets. Inf. Control. 8, 338–353 (1965)
5. Yang, Y., Liang, C., Ji, S., et al.: Adjustable soft discernibility matrix based on picture fuzzy
soft sets and its applications in decision making. J. Int. Fuzzy Syst. 29(4), 1711–1722 (2015)
6. Chen, S.M., et al.: A comparison of similarity measures of fuzzy values. Fuzzy Sets Syst. 72,
79–89 (1995)
7. Chen, S.M.: Measures of similarity between vague sets. Fuzzy Sets Syst. 74, 217–223 (1995)
8. Chen, S.M.: Similarity measures between vague sets and between elements. IEEE Trans. Syst.
Man Cybern. (Part B). 27(1), 153–168 (1997)
9. Li, F., Xu, Z.Y.: Similarity measure between vague sets. Chin. J. Softw. 12(6), 922–927 (2001)
10. Hong, D.H., Kim, C.: A note on similarity measure between vague sets and elements. Inf. Sci.
115, 83–96 (1999)
11. Pappis, C.P.: Value approximation of fuzzy systems variables. Fuzzy Sets Syst. 39, 111–115
(1991)
12. Pappis, C.P., Karacapilidis, N.I.: A comparative assessment of measures of similarity of fuzzy
values. Fuzzy Sets Syst. 56, 171–174 (1993)
A Similarity Measure of Picture Fuzzy Soft Sets and Its Application 389
13. Majumdar, P., Samanta, S.K.: Similarity measure of soft sets. New Math. Nat. Comput. 4, 1–12
(2008)
14. Atanassov, K.: Intuitionistic Fuzzy Sets. Theory and Applications, studies in Fuzziness and
Soft Computing, vol. 35, pp. 324–330. Physica-Verlag, Heidelberg (1999)
15. Szmidt, E., Kacprzyk, J.: Distances between intuitionistic fuzzy sets. Fuzzy Sets Syst. 114,
505–518 (2000)
16. Wei, G., Wei, Y.: Similarity measures of pythagorean fuzzy sets based on the cosine function
and their applications. Int. J. Intell. Syst. 33, 634–652 (2018)
17. Wei, G.: Some cosine similarity measures for picture fuzzy sets and their applications to
strategic decision making. Inform. 28, 547–564 (2017)
18. Ullah, K., Mahmood, T., Jan, N.: Similarity measures for t-spherical fuzzy sets with applications
in pattern recognition. Symmetry. 10, 193 (2018)
19. Cuong, B.: Picture fuzzy sets-first results. part 1, seminar neuro-fuzzy systems with applica-
tions, Institute of Mathematics, Hanoi (2013)
Soft Almost s-Regularity and Soft Almost
s-Normality
Abstract The present paper introduces the axioms of soft almost s-regularity and
soft almost s-normality and presents their studies in soft topological spaces.
Keywords Soft sets · Soft topology · Soft almost s-regular · Soft almost s-normal
spaces
1 Introduction
Researchers introduce many concepts to deal with uncertainty and to solve compli-
cated problems in economy, engineering, medicine, sociology and environment
because of the unsuccessful use of classical methods. The well-known theories can
be considered as a mathematical tool for dealing with uncertainty and imperfect
knowledge of the theory of fuzzy sets, theory of intuitionistic fuzzy sets, theory of
vague sets, theory of rough sets and theory of probability.
In 1999, Molodtsov [13] initiated the theory of soft sets as a new mathematical
tool to deal with uncertainty while modeling problems with incomplete information.
He [13] applied successfully soft sets in many directions such as Smoothness of
function, Game theory, Operation research, Riemann integration, Perron integration,
Probability and Theory of measurement to model the emphasis information. Maji
et al. [11] describe an application of soft set theory to decision-making problems and
gave the operation of soft sets and their properties. Later on, Ali et al. [1] improved
the work of Maji et al. [11] and Das et al. [5]. Pie and Miao [15] investigated the
relationship between soft sets to information systems. They showed that soft sets are
A. K. Prasad (B)
Department of Mathematics, Swami Vivekanand Government College, Lakhnadon, Dist-Seoni,
MP 480886, India
e-mail: [email protected]
S. S. Thakur
Department of Applied Mathematics, Jabalpur Engineering College, Jabalpur, MP 482011, India
e-mail: [email protected]
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 391
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_27
392 A. K. Prasad and S. S. Thakur
classes of special information systems. In 2011, Shabir et al. [23] initiated the study
of soft topological spaces as a generalization of topological spaces. They further
studied the concepts of closure, interior and neighborhood of soft sets. Since the
inception of soft topology, many authors such as Aygünoğlu and Aygün [2], Hussain
[9], Hussain and Ahmad [10], Min [12], Nazmul and Samanta [14], Hazra et al. [8],
Senel and Cagman [22], Zorlutuna et al. [25] and others extended and studied many
topological concepts to soft sets. The study of soft separation axioms was initiated
by Shabir and Naz [23]. In the recent past, many authors such as Georgiou et al. [6],
Guler and Kale [9], Prasad et al. [16–21], Çağman et al. [3], Varol and Aygun [24] and
others studied various soft separation axioms in soft topological spaces. The present
paper introduces two new soft separation axioms called soft almost s-regularity and
soft almost s-normality and obtains their characterizations and properties.
2 Preliminaries
Let X be a nonempty set, be the set of parameters and S(X, ) refer to the collection
of all soft sets of X relative to .
Remark 2.1 ([4]) In a STS (X, , ), the following containments are true:
(a) sCl(ξ, ) ⊆ Cl(ξ, ).
(b) Int(ξ, ) ⊆ sInt(ξ, ), for every (ξ, ) ∈ S(X, ).
Definition 2.5 ([25]) A soft set (ξ, ) ∈ S(X,) is said to be a soft point if , ∃ x ∈ X
and β ∈ with ξ(β) = {x} and ξ (β c ) = φ for each β c ∈ − {β} and denoted
by xβ . The family of all soft points over X is written as SP(X, ).
Lemma 2.3 ([8]) Let (ξ, ), (σ, ) ∈ S(X, ) and xβ ∈ SP(X, ). Then we have
Definition 2.8 ([20]) A STS(X, , ) is soft s-regular if ∀(λ, ) ∈ SC(X, ) and
xβ ∈ SP(X, ) such that xβ ∈ / (λ, ), ∃(μ, ),(ν, ) ∈ SSO(X, ) such that
xβ ∈ (μ, ), (λ, ) ⊆ (ν, ) and (μ, ) ∩ (ν, )=φ.
Remark 2.2 ([16, 20]) The axiom soft regularity implies soft s-regularity and soft
almost regularity but any soft almost regular (resp. soft s-regular space) may fail to be
soft regular. The axiom soft almost regularity and soft s-regularity are independent.
Definition 2.9 ([8]) A STS(X, , ) is soft normal if ∀(ξ, ), (σ, ) ∈
SC(X, ) such that (ξ, ) ∩ (σ, )=φ, ∃(μ, ), (λ, ) ∈ such that (ξ, ) ⊆
(μ, ),(σ, ) ⊆ (λ, ) and (μ, )∩(λ, )=φ.
Definition 2.10 ([21]) A STS(X, , ) is soft almost normal if ∀(ξ, ) ∈
SC(X, )and(σ, ) ∈ SRC(X, ) such that (ξ, )∩(σ, )=φ, ∃(μ, ), (λ, ) ∈
such that (ξ, ) ⊆ (μ, ),(σ, ) ⊆ (λ, ) and (μ, )∩(λ, )=φ.
Definition 2.11 ([19]) A STS(X, , ) is soft s-normal if ∀(ξ, ), (σ, ) ∈
SC(X, ) such that (ξ, ) ∩ (σ, )=φ, ∃(μ, ), (λ, ) ∈ SSO(X, ) such that
(ξ, ) ⊆ (μ, ),(σ, ) ⊆ (λ, ) and (μ, )∩(λ, )=φ.
Remark 2.3 ([19, 21]) The axiom soft normality implies soft s-normality and soft
almost normality but any soft almost normal (resp. soft s-normal space) may fail to be
soft normal. The axiom soft almost normality and soft s-normality are independent.
Definition 2.12 ([7]) Let (ξ, )∈ S(X, ) and xβ ∈ SP(X,). Then (ξ, ) is called
the soft neighborhood of xβ , if ∃σ, )∈ such that xβ ∈ (σ, ) ⊆ (ξ, ).
Definition 3.3 In a STS(X, , ), (ξ, ) ∈ S(X, ) is called soft regular semi-
closed if (ξ, )c ∈ SRSO(X, ).
The family of all soft regular semi-closed sets in a STS(X, , ) is written as
SRSC(X, ).
Remark 3.1 In a STS(X, , ), the next statements are true:
It can be easily verified from examples that the reverse containments may be false.
∼
Example 3.1 Let X= {x1, x2 , x3 , x4 }, ={β1 , β2 } and (λ, ), (μ, )and (ν, ) be
defined as follows:
(λ, ) = {(β 1 }), (β
1 , {x 2 , {x2 , x3 })},
(μ, ) = β1 , x2, x3 ,(β2 , {x1 }) ,
(ν, ) = { β1 , x1 , x2, x3 , (β2 , {x1 , x 2 , x3 })}.
∼
Then STS(X, , )) where = φ, X, (λ, ), (μ, ), (ν, ) , (ν, ) ∈
SSO(X, ) but (ν, ) ∈ / SRSO(X, ).
Remark 3.3 Theorem 2.1 and Remark 3.1. reveal that the axiom soft s-regularity as
well as soft almost regularity imply soft almost s-regularity. However, a soft almost
s-regular space may fail to be soft almost regular. For,
Soft Almost s-Regularity and Soft Almost s-Normality 397
Example 3.2 Let X= {x1, x2 , x3 }, ={β1 , β2 } and (λ, ), (μ, )and(ν, ) be
defined as follows:
(λ, ) = {(β1 , {x1 }), (β2 , {x2 })},
(μ, ) = {(β1 , {x2 }), (β2 , {x1 })},
(ν, ) = {(β1 , {x1 , x2 }), (β2 , {x1 , x2 })}.
∼
Then STS (X, , ), where = φ, X, (λ, ), (μ, ), (ν, ) is soft almost
s-regular but not soft almost regular.
∼
Example 3.3 Let X={x1 , x 2 , x3 }, ={β1 , β2 } and (λ, ), (μ, ), (ν, )and(δ, )
be defined as follows:
(λ, ) = {(β1 , {x1 }), (β2 , {x2 })},
(μ, ) = {(β1 , {x3 }), (β2 , {x3 })},
(ν, ) = {(β1 , {x1 , x3 }), (β2 , {x2, x 3 })},
(δ, ) = {(β1 , {x1 , x2 }), (β2 , {x1 , x 2 })}.
∼
Then STS (X, , ), where = φ, X, (λ, ), (μ, ), (ν, ), (δ, ) is
soft almost regular but not soft s-regular. The following theorem gives several
characterizations for soft almost s-regular spaces.
Theorem 3.3 The following statements are equivalent for a STS(X, , ):
Proof (a)⇒(b) Let (λ, ) ∈ SRO(X, ) such that xβ ⊆ (λ, ). Then
(λ, )c ∈ SRC (X, ) and xβ ∈ / (λ, )c . Since (X, , ) is soft almost s-regular,
∃(χ , ), (ϑ, ) ∈ SSO(X, ) such that xβ ∈ (ϑ, ), (λ, )c ⊆ (χ , ) and (χ , )∩
(ϑ, ) = φ. Since (ϑ, )⊆ (χ , )c and (χ , )c ∈ SSC(X, ), sCl(ϑ, ) ⊆
(χ , )c . This gives xβ ∈ (ϑ, ) ⊆ sCl(ϑ, ) ⊆ (λ, ). Again, (ϑ, ) ⊆
sInt(sCl(ϑ, )) ⊆ sCl(ϑ, ) ⊆ (λ, ). Let (μ, ) = sInt(sCl(ϑ, )). Then
(ϑ, ) ⊆ (μ, ) ⊆ sCl(μ, ) ⊆ sCl(ϑ, ) ⊆ (λ, ). Now (μ, ) ∈ SRSO(X, )
such that xβ ∈ (μ, ) ⊆ sCl(μ, ) ⊆ (λ, ).
(b)⇒(c) Suppose (ξ, ) ∈ SRC (X, ) and xβ ∈ / (ξ, ). Then, (λ, ) =
(ξ, )c ∈ SRO(X, ) and xβ ∈ (λ, ). By (b), ∃(ϑ, ) ∈ SRSO(X, ) such
that xβ ∈ (ϑ, ) ⊆ sCl(ϑ, ) ⊆ (λ, ). The soft set (ϑ, )c ∈ SRSC(X, ) and
398 A. K. Prasad and S. S. Thakur
(ϑ, )c ⊇ (sCl(ϑ, ))c ⊇ (ξ, ). Consequently, (ϑ, )c is a soft regular semi-closed
semi nbd of (ξ, ) with xβ ∈/ (ϑ, )c . Hence (c) holds.
(c)⇒(d) It follows from the fact that SRSC(X, ) ⊆ SSC(X, ).
(d)⇒(e) Suppose (ν, ) ∩ (ψ, ) = φ where (ψ, ) ∈ SRO(X, ). Let xβ ∈
(ν, ) ∩ (ψ, ). Since (ψ, )c ∈ SRC (X, ) such that xβ ∈ / (ψ, )c , ∃(λ, ) ∈
SSC(X, ) ∩ SSO(X, ) such that (ψ, ) ⊆ (λ, ) and xβ ∈
c
/ (λ, ). Let
(ϑ, ) ∈ SSO(X, ) for which (ψ, )c ⊆ (ϑ, ) ⊆ (λ, ). Then (ρ, ) =
(λ, )c ∈ SSO(X, ) which contains xβ and so (ρ, ) ∩ (ν, ) = φ. Also,
(ϑ, )c ∈ SSC(X, ) , sCl(ρ, ) = sCl(λ, )c ⊆ (ϑ, )c ⊆ (ψ, ).
(e)⇒(f) Let (ν, ) ∩ (ψ, ) = φ where (ν, ) = φ and (ψ, ) ∈ SRC (X, )
then (ν, ) ∩ (ψ, )c = φ and (ψ, )c ∈ SRO(X, ).Therefore by(e), ∃(ρ, ) ∈
SSO(X, ) such that (ν, ) ∩ (ρ, ) = φ and(ρ, ) ⊆ sCl(ρ, ) ⊆ (ψ, )c .
Put (δ, ) = (sCl(ρ, ))c . Then (ψ, ) ⊆ (δ, ).Consequently, (ρ, ), (δ, ) ∈
SSO(X, ) such that (ρ, )∩(δ, ) = φ, (ρ, )∩(ν, ) = φ and (ψ, ) ⊆ (δ, ).
(f) ⇒ (a)Let(λ, ) ∈ SRC(X, ) and xβ ∈ SP(X, ) such that xβ ∈ / (λ, ).
Clearly , x β ∩ (λ, )=φ. Therefore by (f), ∃(ρ, ), (δ, ) ∈ SSO(X, ) such that
(ρ, )∩(δ, ) = φ, (ρ, )∩xβ = φ and(λ, ) ⊆ (δ, ). Clearly , (ρ, )∩xβ = φ
implies xβ ∈ (ρ, ). Hence by Definition 3.1, STS(X, , ) is soft almost s-regular.
Remark 4.1 The axiom soft s-normality implies soft almost s-normality, but any
soft almost s-normal space may fail to be soft s-normal. For,
∼
Example 4.1 Let X= {x1 , x 2 , x3 , x4 , x 5 }, = {β} and
(λ, ), (μ, ), (ν, )(δ, ), (χ , ), (ψ, ), (ω, ) and (κ, ) be defined as
follows: (λ, ) = {(β, {x1 , x 2 , x3 , x4 })},
(μ, ) = {(β, {x1 , x 2 , x3 })},
(ν, ) = {(β, {x1 , x 2 , x3 , x4 , x5 })},
(δ, ) = {(β, {x2 , x3 , x4 })},
(χ , ) = {(β, {x2 , x3 })},
(ψ, ) = {(β, {x2 , x4 , x5 })},
(ω, ) = {(β, {x2 , x4 })},
(κ, ) = {(β, {x2 })}.
Soft Almost s-Regularity and Soft Almost s-Normality 399
∼
Then the STS(X, , ), where = φ, X, (λ, ), (μ, ), (ν, ), (δ, ),
(χ , ), (ψ, ), (ω, ), (κ, ) , is soft almost s-normal. For (X, , ), being the
only soft regular closed set, each non-empty soft closed set in (X, , ) intersects
with (X, , ). But the STS(X, , ) is not soft s-normal. For {β, {x 1 }} and {β, {x 5 }}
are disjoint soft closed sets but each
soft semi-open
set containing {β, {x 1 }} meets
each soft semi-open set containing β, {x 5 } .
Remark 4.2 The axiom soft almost normality implies soft almost s-normality but
any soft almost s-normal space may fail to be soft almost normal. For,
Example 4.2 Let X={x1 , x 2 , x3 , x4 }, = {β} and (λ, ), (μ, ), (ν, ), (δ, ),
(ψ, ) and (ω, ) be defined as follows: (λ, ) = {(β, {x1 , x 2 , x3 })},
(μ, ) = {(β, {x1 , x 2 })},
(ν, ) = {(β, {x2 , x 3 })},
(δ, ) = {(β, {x2 })},
(ψ, ) = {(β, {x2 , x3 , x4 })},
(ω, ) = {(β, {x3 })}.
∼
Then, STS(X, , ), where = φ, X, (λ, ), (μ, ), (ν, ), (δ, ), (ψ, ),
Remark 4.3 The axioms of almost normality and soft s-normality are independent.
For,
Example 4.3 Let X={x1 , x 2 , x3 }, = {β1 , β2 } and (λ, ), (μ, )and(ν, ) be
defined as follows:
∼
Then STS(X, , ) , where = φ, X, (λ, ), (μ, ), (ν, ) , is soft almost
normal but not soft s-normal.
∼
Then STS(X, , ), where = φ, X, (λ, ), (μ, ), (ν, ), (δ, ), (ψ, ),
The following theorem gives several characterizations for soft almost s-normal
spaces.
Theorem 4.1 The following statements are equivalent for a STS(X, , ):
(a) (X, , ) is soft almost s-normal.
(b) For every (ν, ) ∈ SC(X, ) and (ψ, ) ∈ SRO(X, ) such that (ν, ) ⊆
(ψ, ), ∃(ϑ, ) ∈ SSO (X, ) such that(ν, ) ⊆ (ϑ, ) ⊆ sCl(ϑ, ) ⊆
(ψ, ).
(c) For every (ν, ) ∈ SRC (X, ) and (ψ, ) ∈ such that (ν, ) ⊆
(ψ, ), ∃(ϑ, ) ∈ SSO (X, ) such that (ν, ) ⊆ (ϑ, ) ⊆ sCl (ϑ, ) ⊆
(ψ, ).
Proof (a)⇒(b) Let (ν, ) ∈ SC(X, )and(ψ, ) ∈ SRO(X, ) such that
(ν, ) ⊆ (ψ, ). Then (ν, ) ∩ (ψ, )c = φ, where (ν, ) ∈ SC(X, )
and (ψ, )c ∈ SRC (X, ). Therefore, ∃(ϑ, ), (λ, ) ∈ SSO(X, ) such that
(ν, ) ⊆ (ϑ, ),(ψ, )c ⊆ (λ, ) and (ϑ, ) ∩ (λ, ) = φ. Then (ν, ) ⊆
(ϑ, ) ⊆ (λ, )c ⊆ (ψ, ). Now (λ, )c ∈ SSC(X, ) so it follows that
(ν, ) ⊆ (ϑ, ) ⊆ sCl(ϑ, ) ⊆ (ψ, ).
(b)⇒(c) Let (ν, ) ∈ SRC(X, ) and (ψ, ) ∈ such that (ν, ) ⊆ (ψ, ).
Then (ψ, )c ⊆ (ν, )c . Now (ν, )c ∈ SRO(X, ) and (ψ, )c ∈ SC(X, )
such that (ψ, )c ⊆ (ν, )c , ∃(μ, ) ∈ SSO(X, ) such that (ψ, )c ⊆ (μ, ) ⊆
sCl(μ, ) ⊆ (ν, )c . Thus, (ν, ) ⊆ (sCl(μ, ))c ⊆ (μ, )c ⊆ (ψ, ). Let
(sCl(μ, ))c = (ϑ, ). Then (ϑ, ) ∈ SSO(X, ). And so, (ν, ) ⊆ (ϑ, ) ⊆
sCl(ϑ, ) ⊆ (ψ, ), since (μ, )c ∈ SSC(X, ) such that (ϑ, ) ⊆ (μ, )c .
(c)⇒(a) Let (ν, ) ∈ SC(X, ) and (ψ, ) ∈ SRC (X, ) such that (ν, ) ∩
(ψ, ) = φ. Then (ψ, ) ⊆ (ν, )c . And so ∃(ϑ, ) ∈ SSO(X, ) such that
(ψ, ) ⊆ (ϑ, ) ⊆ sCl(ϑ, ) ⊆ (ν, )c . Let (sCl(ϑ, ))c = (λ, ). Then , (λ, ) ∈
SSO(X, ) such that (ν, ) ⊆ (λ, ) with (ψ, ) ⊆ (ϑ, ) and (ϑ, )∩(λ, ) = φ.
Soft Almost s-Regularity and Soft Almost s-Normality 401
References
1. Ali, M.I., Feng, F., Liu, X., Min, W.K., Shabir, M.: On some new operations in soft set theory.
Comput. Math. Appl. 57, 1547–1553 (2009)
2. Aygünoğlu, A., Aygün, H.: Some notes on soft topological spaces. Neural Comput. Appl. 21,
113–119 (2012)
3. Çağman, N., Karataş, S., Enginoglu, S.: Soft topology. Comput. Math. Appl. 62, 351–358
(2011)
4. Chen, B.: Soft semi-open sets and related properties in soft topological spaces. Appl. Math.
Inf. Sci. 7(1), 287–294 (2013)
5. Das, S., Samanta, S.K.: Soft real sets soft real numbers and their properties. J. Fuzzy Math.
20(3), 551–576 (2012)
6. Georgiou, D.N., Megaritis, A.C., Petropoulos, V.I.: On soft topological spaces. Appl. Math.
Inf. Sci. 7(5), 1889–1901 (2013)
7. Guler, A.C., Kale, G.: Regularity and normality on soft ideal topological spaces. Ann. Fuzzy
Math. Inform. 9(3), 373–383 (2015)
8. Hazra, H., Majumdar, P. and Samanta, S.K.: Soft topology. Fuzzy Inf. Eng. 4(1), 105–115
(2012)
9. Hussain, S.: On soft regular-open (closed) sets in soft topological spaces. J. Appl. Math. Inform.
36, 59–68 (2018)
10. Hussain, S., Ahmad, B.: Some properties of soft topological spaces. Comput. Math. Appl. 62,
4058–4067 (2011)
11. Maji, P.K., Biswas, R., Roy, A.R.: Soft set theory. Comput. Math. Appl. 45, 555–562 (2003)
12. Min, W.K.: A note on soft topological spaces. Comput. Math. Appl. 62, 3524–3528 (2011)
13. Molodtsov, D.: Soft set theory-first results. Comput. Math. Appl. 37, 19–31 (1999)
14. Nazmul, S., Samanta, S.K.: Neighborhoods properties of soft topological spaces. Ann. Fuzzy
Math. Inform. 6(1), 1–15 (2013)
15. Pei, D., Miao, D.: From soft sets to information systems. Proc. Granul. IEEE 2, 617–621 (2005)
16. Prasad, A.K., Thakur, S.S.: Soft almost regular spaces. Malaya J. Mat. 7(3), 408–411 (2019)
17. Prasad, A.K., Thakur, S.S.: Soft almost I-regular spaces. J. Indian Acad. Math. 41(1), 53–60
(2019)
18. Prasad, A.K., Thakur, S.S.: Soft almost I-normal spaces. In: Presented in 23rd FAI International
Conference on Emerging Trends and Adaptation of Spirituality in Management, Artificial
Intelligence, Education, Health, Tourism, Science and Technology, 19–21 Feb 2021
19. Prasad, A.K., Thakur, S.S.: On soft s-normal, spaces. In: Presented in 4th International Confer-
ence on Mathematical Modelling, Applied Analysis and Computation (ICMMAAC-21), 5–7
Aug 2021
20. Prasad, A.K., Thakur, S.S.: On soft s-regular spaces. In: Presented in National Conference on
Mathematical Analysis and Applications NCMAA-2021, 23–24 April 2021
21. Prasad, A.K., Thakur, S.S., Thakur, M.: Soft almost normal spaces. J. Fuzzy Math. 28(3),
(2020) (In Press)
22. Senel, G., Cagman, N.: Soft topological subspaces. Ann. Fuzzy Math. Inform. 10(4), 525–535
(2015)
23. Shabir, M., Naz, M.: On soft topological spaces. Comput. Math. Appl. 61, 1786–1799 (2011)
24. Varol, B.P., Aygun, H.: On soft Hausdorff spaces. Ann. Fuzzy Math. Inform. 5(1), 15–24 (2013)
25. Zorlutuna, I., Akdag, M., Min, W.K., Atmaca, S.: Remarks on soft topological spaces. Ann.
Fuzzy Math. Inform. 3(2), 171–218 (2012)
Algebraic Properties of Spherical Fuzzy
Sets
Abstract The spherical fuzzy set (SFS) is an advanced version of fuzzy set, intu-
tionistic fuzzy set, Pythagorean fuzzy set, and picture fuzzy set. This generalized
three dimensional fuzzy set model is more realistic and accurate. In this paper, we
discuss the algebraic operations on SFSs such as union, intersection, complement,
algebraic sum, algebraic product, exponentiation operation, and scalar multiplication
operation. Also, we prove some fundamental algebraic properties of these operations.
1 Introduction
Fuzzy set theory, introduce by Zadeh [11], is one of the most powerful techniques for
tackling multi-attribute decision-making problems due to the challenges of obtaining
sufficient and accurate data for practical decision-making due to the vagueness and
ambiguity of socioeconomics. However, sometimes there are flaws and limitations
in fuzzy set theory when it comes to dealing with the task at hand in a more objective
way. Fuzzy sets are a more advanced version of classical sets that have a membership
grade for each element. To resolve some of the difficulties of fuzzy sets, Atanassov
[2] developed intuitionistic fuzzy sets. Also, many other fuzzy set generalizations
have been proposed, such as interval-valued intuitionistic fuzzy sets [3], Pythagorean
fuzzy sets [10], picture fuzzy sets [4], and so on. Recently, Ashraf et al. [1] proposed
spherical fuzzy set (SFS) as a generalization of picture fuzzy set, with each element
having three membership degrees: positive, neutral, and negative.
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 403
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_28
404 P. A. Fathima Perveen and S. J. John
Nowadays, research on these sets has become extremely fruitful, yielding numer-
ous significant theoretical and practical results. Luca and Termini [5] proposed alge-
braic properties of the class of fuzzy sets and stated that the class of generalized
characteristic function equipped with the lattice operation suggested by Zadeh is a
Brouwerian lattice. Tanaka et al. [7] studied the algebraic properties of fuzzy sets
under the operations bounded sum and bounded difference. A new idea of complex
intuitionistic fuzzy subgroups was proposed by Gulzar et al. [6]. Silambarasan et al.
[9] researched the algebraic operations on Pythagorean fuzzy matrices. Silambarasan
[8] also studied some algebraic operations of the Picture fuzzy sets.
The objective of this paper is to present certain algebraic properties for SFSs. The
spherical fuzzy model, which stands out for having a massive region of participation
of admissible triplets, is more adaptable than the previous fuzzy set models in that
it widens the field of uncertain and ambiguous information. Therefore, discussing
algebraic properties on SFSs is critical for their theoretical and practical advance-
ment. The remainder of the paper is organized in the following manner. Section 2
revisits some fundamental concepts of fuzzy sets and spherical fuzzy sets, as well as
redefines the definitions of spherical fuzzy union and spherical fuzzy intersection. In
Sect. 3, some algebraic properties of SFSs are proved based on the algebraic oper-
ations such as union, intersection, complement, algebraic sum, algebraic product,
scalar multiplication, and exponentiation. Finally, Sect. 4 concludes the work with
recommendations for future work.
2 Preliminaries
We review the fundamental notions of spherical fuzzy sets in this section and provide
new definitions for spherical fuzzy intersection and union.
Definition 1 [11] Let Σ be the universal set of discourse. A fuzzy set ℵ on Σ
is an object of the form ℵ = {(, μℵ ())| ∈ Σ}, where μℵ : Σ → [0, 1] is the
membership function of ℵ, the value μℵ () is the grade of membership of in ℵ.
Definition 2 [1] A spherical fuzzy set (SFS) ℵ over Σ can be expressed as ℵ =
{(, μℵ (), ηℵ (), ϑℵ ())| ∈ Σ}, where μℵ (), ηℵ () and ϑℵ () are the functions
defined from Σ to [0,1], are called the membership functions (positive, neutral, and
negative respectively) of ∈ Σ, with the condition 0 ≤ μ2ℵ () + ηℵ2 () + ϑℵ2 () ≤
1, ∀ ∈ Σ.
Definition 3 [1] Let ℵ and Ω be two SFSs over the universe Σ, where ℵ =
{(, μℵ (), ηℵ (), ϑℵ ())| ∈ Σ} and Ω = {(, μΩ (), ηΩ (), ϑΩ ())| ∈ Σ}.
Then ℵ is said to be a spherical fuzzy subset of Ω, denoted by ℵ ⊆ Ω, if μℵ () ≤
μΩ (), ηℵ () ≤ ηΩ (), ϑℵ () ≥ ϑΩ (), ∀ ∈ Σ.
Definition 4 [1] Let ℵ = {(, μℵ (), ηℵ (), ϑℵ ())| ∈ Σ} be a SFS over Σ. Then
the complement of ℵ is a SFS, denoted by ℵc , is defined as ℵc = {(, ϑℵ (), ηℵ (),
μℵ ())| ∈ Σ}.
Algebraic Properties of Spherical Fuzzy Sets 405
Definition 6 [1] Let ℵ and Ω be two SFSs over Σ. Then the algebraic sum and
algebraic product of ℵ and Ω, denoted by, ℵ Ω and ℵ Ω respectively, defined
as follows
1. ℵ Ω = {(, μ2ℵ () + μ2Ω () − μ2ℵ ()μ2Ω (), ηℵ ()ηΩ (), ϑℵ ()ϑΩ ())| ∈ Σ}.
2. ℵ Ω = {(, μℵ ()μΩ (), ηℵ ()ηΩ (), ϑℵ2 () + ϑΩ
2 () − ϑ 2 ()ϑ 2 ()| ∈ Σ}.
ℵ Ω
Definition 7 [1] Let ℵ be any SFS over Σ. Then the scalar multiplication operation
and the exponentiation operation of Σ, denoted by, nℵ and ℵn respectively, where n
denotes the natural number, is defined as follows
1. nℵ = {(, 1 − (1 − μ2ℵ ())n , ηℵn (), ϑℵn ())| ∈ Σ}
2. ℵn = {(, μnℵ (), ϑℵn (), 1 − (1 − ϑℵ2 ())n )| ∈ Σ}
Example 1 Let ℵ = {(1 , 0.8, 0.3, 0.2), (2 , 0.4, 0.5, 0.3), (3 , 0.9, 0.1, 0.2)} and
Ω = {(1 , 0.7, 0.4, 0.5), (2 , 0.3, 0.6, 0.4), (3 , 0.7, 0.3, 0.4)} be two SFSs over the
universe Σ = {1 , 2 , 3 }. Then
ℵc = {(1 , 0.2, 0.3, 0.8), (2 , 0.3, 0.5, 0.4), (3 , 0.2, 0.1, 0.9)}
ℵ ˜ Ω = {(1 , 0.8, 0.4, 0.2), (2 , 0.4, 0.6, 0.3), (3 , 0.9, 0.3, 0.2)}
ℵ ˜ Ω = {(1 , 0.7, 0.3, 0.5), (2 , 0.3, 0.5, 0.4), (3 , 0.7, 0.1, 0.4)}
ℵ Ω = {(1 , 0.90, 0.12, 0.10), (2 , 0.49, 0.30, 0.12), (3 , 0.95, 0.03, 0.08)}
ℵ Ω = {(1 , 0.56, 0.12, 0.53), (2 , 0.12, 0.30, 0.49), (3 , 0.63, 0.03, 0.44)}
Suppose n = 3, then
3ℵ = {(1 , 0.976, 0.027, 0.008), (2 , 0.638, 0.125, 0.027), (3 , 0.996, 0.001, 0.008)}
ℵ3 = {(1 , 0.512, 0.027, 0.339), (2 , 0.064, 0.125, 0.496), (3 , 0.729, 0.001, 0.339)}
406 P. A. Fathima Perveen and S. J. John
(3) (ℵ Ω)c ⊆ ℵc Ω c
(4) (ℵ Ω)c ⊇ ℵc Ω c
Proof We prove only (1) and (3). Other proofs can be done in similar way.
(1) ℵ Ω = ( μ2ℵ + μ2Ω − μ2ℵ μ2Ω , ηℵ ηΩ , ϑℵ ϑΩ )
(ℵ Ω)c = (ϑℵ ϑΩ ), ηℵ ηΩ , μ2ℵ + μ2Ω − μ2ℵ μ2Ω ) = ℵc Ω c
(3) (ℵ Ω)c = (ϑℵ ϑΩ , ηℵ ηΩ , μ2ℵ + μ2Ω − μ2ℵ μ2Ω ), and
ℵc Ω c = ( ϑℵ2 + ϑΩ2 − ϑℵ2 ϑΩ2 , ηℵ ηΩ , μℵ μΩ )
By Theorem 1, we have μ2ℵ + μ2Ω − μ2ℵ μ2Ω ≥ μℵ μΩ , ηℵ2 + ηΩ 2
− ηℵ2 ηΩ
2
≥
ηℵ ηΩ , and ϑℵ2 + ϑΩ2 − ϑℵ2 ϑΩ2 ≥ ϑℵ ϑΩ
⇒ μ(ℵΩ)c ≤ μℵc Ω c , η(ℵΩ)c ≤ ηℵc Ω c , and ϑ(ℵΩ)c ≥ ϑℵc Ω c
⇒ (ℵ Ω)c ⊆ ℵc Ω c
Proof We prove only (1). (2) can be prove using similar method.
Suppose that ℵ
⊆ Ω, which means μℵ ≤ μΩ , ηℵ ≤ ηΩ , and ϑℵ ≥ ϑΩ .
μℵ ≤ μΩ ⇒ 1 − (1 − μ2ℵ )n ≤ 1 − (1 − μ2Ω )n , ηℵ ≤ ηΩ ⇒ ηℵn ≤ ηΩ
n
, and ϑℵ ≥
ϑΩ ⇒ ϑℵ ≥ ϑΩ . Therefore, nℵ ⊆ nΩ.
n n
Proof The proof of (1) is given as follows. (2) can be proved in similar way.
2 n
n(ℵ Ω) = 1− 1− μ2ℵ + μ2Ω − μ2ℵ μ2Ω , (ηℵ ηΩ )n , (ϑℵ ϑΩ )n
= 1 − (1 − μ2ℵ − μ2Ω + μ2ℵ μ2Ω )n , ηℵn ηΩ n
, ϑℵn ϑΩn
= 1 − (1 − μ2ℵ )n (1 − μ2ℵ )n , ηℵn ηΩ
n
, ϑℵn ϑΩn
= nℵ nΩ
Theorem 6 Let ℵ be a SFS over the universe Σ and let n 1 and n 2 be two natural
numbers. Then
(1) n 1 ℵ n 2 ℵ = (n 1 + n 2 )ℵ
(2) ℵn1 ℵn2 = ℵ(n 1 +n 2 )
408 P. A. Fathima Perveen and S. J. John
Proof We prove
only (1). (2) can be proved in same method.
n1 ℵ n2 ℵ = 1 − (1 − μ2ℵ )n 1 + 1 − (1 − μ2ℵ )n 2 − (1 − (1 − μ2ℵ )n 1 )(1 − (1 − μ2ℵ )n 2 ),
ηℵn 1 ηℵn 2 , ϑℵn 1 ϑℵn 2
= 1 − (1 − μ2ℵ )n 1 +n 2 , ηℵn 1 +n 2 , ϑℵn 1 +n 2
= (n 1 + n 2 )ℵ
Proof The proof of (1) is given as follows. (2) can be proved in similar way. Consider
the following cases.
Case 1 : If (μℵ () or μΩ ()) = 1, for every ∈ Σ. Then
ℵ ˜ Ω = (μℵ ∧ μΩ , ηℵ ∧ ηΩ , ϑℵ ∨ ϑΩ )
(ℵ ˜ Ω) = (μℵ ∧ μΩ )2 + μ2 − (μℵ ∧ μΩ )2 μ2 , (ηℵ ∧ ηΩ )η , (ϑℵ ∨ ϑΩ )ϑ
= μ2ℵ + μ2 − μ2ℵ μ2 ∧ μ2Ω + μ2 − μ2Ω μ2 , ηℵ η ∧ ηΩ η , ϑℵ ϑ ∨ ϑΩ ϑ
= (ℵ ) ˜ (Ω )
Let S F S ∗ (Σ) be the collection of all SFSs over Σ, where for any two SFSs ℵ, Ω ∈
S F S ∗ (Σ), (μℵ () ∨ μΩ ())2 + (ηℵ () ∨ ηΩ ())2 + (ϑℵ () ∧ ϑΩ ())2 1,for every
∈ Σ.
Theorem 9 Let ℵ, Ω ∈ S F S ∗ (Σ). Then
(1) (ℵ ˜ Ω) (ℵ ˜ Ω) = ℵ Ω
(2) (ℵ ˜ Ω) (ℵ ˜ Ω) = ℵ Ω
Algebraic Properties of Spherical Fuzzy Sets 409
Proof The proof of (1) is given below. (2) can be proved using similar steps. Consider
the following two cases.
Case 1: If (μℵ () or μΩ ()) = 1, for every ∈ Σ. Then
(ℵ ˜ Ω) (ℵ ˜ Ω) = (μℵ ∨ μΩ )2 + (μℵ ∧ μΩ )2 − (μℵ ∨ μΩ )2 (μℵ ∧ μΩ )2 ,
(ηℵ ∨ ηΩ )(ηℵ ∧ ηΩ ), (ϑℵ ∧ ϑΩ )(ϑℵ ∨ ϑΩ )
= μ2ℵ + μ2Ω − μ2ℵ μ2Ω , ηℵ ηΩ , ϑℵ ϑΩ
=ℵΩ
Case 2 : If (μℵ () or μΩ ()) = 1, for some ∈ Σ, say 0 .
Without loss of generality, assume that μℵ (0 ) = 1, then ηℵ (0 ) = ϑℵ (0 ) = 0
Thus, ℵ0 ˜ Ω0 = ℵ0 and ℵ0 ˜ Ω0 = Ω0
⇒ ((ℵ ˜ Ω) (ℵ ˜ Ω))0 = (ℵ Ω)0
Therefore, in any case (ℵ ˜ Ω) (ℵ ˜ Ω) = ℵ Ω
Theorem 10 Let ℵ, Ω, ∈ S F S ∗ (Σ). Then
(1) (ℵ ˜ Ω) = (ℵ ) ˜ (Ω )
(2) (ℵ ˜ Ω) = (ℵ ) ˜ (Ω )
Proof We prove
only (1). (2) can be proved in same method.
(ℵ ˜ Ω) = (μℵ ∨ μΩ )2 + μ2 − (μℵ ∨ μΩ )2 μ , (ηℵ ∨ ηΩ )η , (ϑℵ ∧ ϑΩ )ϑ
= (μ2ℵ + μ2 ) ∨ (μ2Ω + μ2 ) − (μ2ℵ μ2 ) ∨ (μ2Ω μ2 ), ηℵ η ∨ ηΩ η , ϑℵ ϑ ∧ ϑΩ ϑ
= (μ2ℵ + μ2 ) − μ2ℵ μ2 ∨ (μ2Ω + μ2 ) − μ2Ω μ2 , ηℵ η ∨ ηΩ η , ϑℵ ϑ ∧ ϑΩ ϑ
= (ℵ ) ˜ (Ω )
Definition 8 Let ℵ be any SFS over the universe Σ. Then the concentration of ℵ is
a SFS, denoted as ℵC O N , is defined by ℵC O N = ℵ2 . That is,
ℵC O N = μ2ℵ , ηℵ2 , 1 − (1 − ϑℵ2 )2
Definition 9 Let ℵ be any SFS over the universe Σ. Then the dilation of ℵ is a SFS,
denoted as ℵ D I L , is defined by ℵ D I L = ℵ1/2 . That is,
1/2 1/2
ℵ D I L = μℵ , ηℵ , 1 − (1 − ϑℵ2 )1/2
4 Conclusion
In this paper, we have proved some algebraic properties of spherical fuzzy sets,
including commutativity, associativity, distributivity, absorption, etc. We also estab-
lished several theorems and defined new concentration and dilation of SFSs. These
results can be used in more applications of spherical fuzzy set theory. The results
of this research will be useful in the creation of the Picture fuzzy semilattice and
its algebraic property. The applicability of the suggested operators of SFSs in deci-
sion making, risk analysis, and many other uncertain and fuzzy environments can be
recommended as future work.
References
1. Ashraf, S., Abdullah, S., Mahmood, T., Ghani, F., Mahmood, T.: Spherical fuzzy sets and
their applications in multi-attribute decision making problems. J. Intell. & Fuzzy Syst. 36,
2829–2844 (2019)
2. Atanassov, K.T.: Intuitionistic fuzzy sets. Fuzzy Sets Syst. 20, 87–96 (1986)
3. Atanassov, K.T., Gargov, G.: Interval valued intuitionistic fuzzy sets. Fuzzy Sets Syst. 31,
343–349 (1989)
4. Cuong, B.C., Kreinovich, V.: Picture fuzzy sets-a new concept for computational intelligence
problems. In: Proceedings of 2013 Third World Congress on Information and Communication
Technologies (WICT 2013), pp. 1–6. IEEE (2013)
5. De Luca, A., Termini, S.: Algebraic properties of fuzzy sets. J. Math. Anal. Appl. 40(2), pp.
373–386 (1972)
6. Gulzar, M., Mateen, M.H., Alghazzawi, D., Kausar, N.: A novel applications of complex
intuitionistic fuzzy sets in group theory. IEEE Access 8, 196075–196085 (2020)
7. Mizumoto, M., Tanaka, K.: Fuzzy sets and their operations. Inf. Control. 48(1), 30–48 (1981)
8. Silambarasan, I.: Some algebraic properties of picture fuzzy sets. Bull. Int. Math. Virtual Inst.
11(3), 429–442 (2021)
9. Silambarasan, I., Sriram, S.: Algebraic operations on Pythagorean fuzzy matrices. Math. Sci.
Int. Res. J. 7(2), 406–414 (2018)
10. Yager, R.R.: Pythagorean fuzzy subsets. In: 2013 Joint IFSA World Congress and NAFIPS
Annual Meeting (IFSA/NAFIPS), pp. 57-61. IEEE (2013)
11. Zadeh, L.A.: Fuzzy sets. Inf. Control. 8(3), 338–353 (1965)
Divergence Measures of Pythagorean
Fuzzy Soft Sets
Abstract This work initiates the study of divergence measures of Pythagorean fuzzy
soft sets (PFSSs). A couple of expressions to find PFSS divergence measures are
obtained, and thus we can quantify the deviation between any two PFSSs. Also,
certain theorems based on the properties of proposed expressions are proved.
1 Introduction
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 411
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_29
412 T. M. Athira and S. J. John
Zhou et al. [15] proposed a novel divergence measure for Pythagorean fuzzy set
using the belief function from Dempster–Shafer evidence theory.
Research on the theoretical and application point of view of soft sets and their
various generalizations is growing rapidly. In [9] Peng et al. introduced Pythagorean
fuzzy soft sets (PFSSs) as a blending of soft sets (SS) and Pythagorean fuzzy
sets(PFS). Interestingly, PFSS generalizes both SS and PFS. Recently, Athira et
al. [3, 4] initiated the study of the entropy and distance measures for PFSS. To the
best of the authors’ knowledge, very little work has been done on the theory of the
PFSS. Keeping the usefulness of PFSSs to look over the given data better than IFSS
or FSS, this article focused on divergence measures of PFSS. Divergence measure
quantifies the difference between any two PFSS. PFSS with lower divergence is
“more similar”. The axiomatic definition and a couple of expressions to calculate
PFSS divergence measures are proposed. Also, certain interesting theorems based
on properties of PFSS divergence measures are proved.
This paper contains two sections besides the introduction. Section 2 contains pre-
liminary definitions and results required for our entire discussion. In Sect. 3, we
introduce divergence measures for PFSSs. Some interesting properties of divergence
measures are obtained, and we propose certain nice examples for divergence mea-
sures for PFSSs. The TOPSIS method based on PFSS divergence measure for select-
ing the best-fitting alternative is proposed.
2 Preliminaries
This section explains the fundamental definitions needed for entire discussions. Here,
U represents the universal sets and E is the set of parameters.
Definition 1 [8] The pair (S, E) is said to be soft set over U if S is a mapping from
E to P(U), where P(U) is power set of U.
Definition 3 [9] A Pythagorean fuzzy soft set (PFSS) is defined as the pair (P, E)
such that P : E → PFS(U) and PFS(U) is the collection of all Pythagorean fuzzy
subsets of U.
Proof From conditions 2 and 4, it is possible to obtain Div((P, E), (Q, E)) ≥
0 for any (P, E), (Q, E) ∈ PFSS(). Because, when (R, E) = ∅, Div((P, E)
(R, E), (Q, E) (R, E)) = Div(∅, ∅) = 0 from the condition 2. And so 0 ≤
Div((P, E), (Q, E)) from 4.
Theorem 2 Let (P, E), (Q, E) are two PFSSs over . For a PFSS divergence mea-
sure Div,
Div(P, E) (Q, E), (Q, E)) = Div(P, E), (P, E) (Q, E))
Div(P, E) (Q, E), (Q, E)) = Div(P, E), (P, E) (Q, E))
Div(P, E), (P, E) (Q, E)) = Div((P, E) (Q, E)) (P, E), (Q, E) (P, E))
≤ Div(P, E) (Q, E), (Q, E))
= Div((P, E) (Q, E)), (P, E) (Q, E) (Q, E))
≤ Div(P, E), (P, E) (Q, E))
Thus, Div(P, E) (Q, E), (Q, E)) = Div(P, E), (P, E) (Q, E))
⎧
⎪ 2 μ2P(ζ j ) (u i ) + μ2Q(ζ j ) (u i )
⎪
⎪ μ2P(ζ j ) (u i ) − μ2Q(ζ j ) (u i ) +
⎪
⎪ μ2P(ζ j ) (u i ) μ2Q(ζ j ) (u i )
⎪ if μP(ζ j ) (u i ) = 0,
1 ⎪
n m ⎨
Div2 ((P, E), (Q, E)) = 2 νP(ζ
2 (u i ) + νQ(ζ
2 (u i ) νP(ζ j ) (u i ) = 0
⎪ j) j)
j=1 ⎪
2mn νP(ζ
2
(u i ) − νQ(ζ
2
(u i )
i=1 ⎪
⎪ j) j)
⎪
⎪ νP(ζ
2
j)
(u i ) νQ(ζ
2
j)
(u i )
⎪
⎩
0 otherwise
(2)
Proof We will prove the necessary part of condition 2 and condition 3 for Divi , i =
1, 2. Condition 4 can be proved just like proof of condition 3, and the remaining parts
are trivial.
To prove necessary part of condition 2, Div1 ((P, E), (Q, E)) = 0 implies
⎛ ⎞
2μ2P(ζ j ) (u i )
μ2
⎜ P (ζ j ) i(u )
μ2P(ζ j ) (u i ) + μ2Q(ζ j ) (u i ) ⎟
log
⎜ ⎟
⎜ ⎟=0 (3)
⎜ 2μ2Q(ζ j ) (u i ) ⎟
⎝ ⎠
+μQ(ζ j ) (u i ) log 2
2
μP(ζ j ) (u i ) + μQ(ζ j ) (u i )
2
⎜ 2μ 2
u ⎟
⎝ μ2
Q(ζ j0 ) i 0
⎠
Q(ζ j0 ) u i 0 log 2
μP(ζ j ) u i0 + μQ(ζ j ) u i0
2
0 0
Thus, from Eq. 3, we get a contradiction. Hence, μP(ζ j ) (u i )) = μQ(ζ j ) (u i )), ∀i, j. and
similar way, we get νP(ζ j ) (u i )) = νQ(ζ j ) (u i )), ∀i, j. Thus Div1 ((P, E), (Q, E)) = 0
implies (P, E) = (Q, E). Now, Div2 ((P, E), (Q, E)) = 0. Since each term is posi-
tive we can conclude that
2 2
μ2P(ζ j ) (u i ) − μ2Q(ζ j ) (u i ) = 0 and νP(ζ
2
j)
(u i ) − νQ(ζ
2
j)
(u i ) ∀i, j.
Thus, Div2 ((P, E), (Q, E)) = 0 implies (P, E) = (Q, andE).
Thus, Divμ1 ((P, E) (R, E), (Q, E) (R, E)) = Divμ1 ((P, E), (Q, E)).
For the case of Div2 , the expression is symmetric. So, it is enough to consider three
cases for Divμ2 and for Divν2 .
1. μP(ζ j ) (u i ) ≤ μQ(ζ j ) (u i ) ≤ μ R (ζ j ) (u i ) for some u i and ζ j .
Then, μ2P R ζ (u i ) = μ2R ζ (u i ) and μ2Q R ζ (u i ) = μ2R(ζi ) (u i )
( j) ( j) ( j)
Thus,
μ μ
Div1 ((P, E) (R, E), (Q, E) (R, E)) = Div1 ((R, E), (R, E))
=0
μ
≤ Div1 ((P, E), (Q, E)) ∀(R, E) ∈ PFSS().
2. μP(ζ j ) (u i ) ≤ μ R (ζ j ) (u i ) ≤ μQ(ζ j ) (u i ) for some u i and ζ j .
Then, μ2P R ζ (u i ) = μ2R ζ (u i ) and μ2Q R ζ (u i ) = μ2Q ζ (u i ).
( j) ( j) ( j) ( j)
Consider the function f (x) = a12 + x12 for a given a ∈ (0, 1] and for all x ∈ [a, 1].
Since f (x) ≤ 0, f (x) is non-increasing function.
Here, take a = μQ(ζ j ) (u i ) and we have μP(ζ j ) (u i ) ≤ μ R (ζ j ) (u i ) that implies
μ2 (u i )+μ2 (u i ) μ2P(ζ ) (u i )+μ2Q(ζ ) (u i )
R (ζ j ) Q(ζ j )
μ2 (u )μ2 (u )
≤ μ2
j
(u )μ2
j
2 2
Also, μ2R ζ (u i ) − μ2Q ζ (u i ) ≤ μ2P ζ (u i ) − μ2Q(ζ j ) (u i ) .
( j) ( j) ( j)
Thus, the required inequality holds in this case also.
3. μ R (ζ j ) (u i ) ≤ μP(ζ j ) (u i ) ≤ μQ(ζ j ) (u i ) for some u i and ζ j .
Thus μ2P R ζ (u i ) = μ2P ζ (u i ) and μ2Q R ζ (u i ) = μ2Q ζ (u i )
( j) ( j) ( j) ( j)
μ μ
Div1 ((P, E) (R, E), (Q, E) (R, E)) = Div1 ((P, E), (Q, E)). Similarly, we
μ
can prove the case of Diviν ,i = 1, 2. Since Divi = Divi + Diviν , we can conclude
the third condition of definition 5.
Divergence Measures of Pythagorean Fuzzy Soft Sets 417
Hence, Divi , i = 1, 2 satisfy all the four conditions in definition 5, and thus
Divi , i = 1, 2 are divergence measure.
u1 u2 u3 u4 u5
⎛ ⎞
(0.7, 0.3)
ζ1 (0.3, 0.8) (0.1, 0.8) (0.8, 0.6) (0.7, 0.7)
ζ2⎜ (0.0, 0.9) (0.6, 0.6) (0.2, 0.6) (0.9, 0.3) (0.2, 0.3) ⎟
(P1 , E) = ⎜ ⎟
ζ3⎝ (0.4, 0.7) (0.7, 0.3) (0.9, 0.1) (0.1, 0.3) (0.5, 0.9) ⎠
ζ4 (0.8, 0.2) (0.5, 0.7) (0.2, 0.8) (0.7, 0.3) (0.7, 0.5)
u1 u2 u3 u4 u5
⎛ ⎞
(0.4, 0.3)
ζ1 (0.4, 0.8) (0.1, 0.0) (0.9, 0.2) (0.2, 0.7)
ζ2⎜ (0.0, 0.9) (0.6, 0.4) (0.2, 0.6) (0.9, 0.3) (0.1, 0.3) ⎟
(P2 , E) = ⎝ ⎜ ⎟
ζ3 (0.2, 0.2) (0.7, 0.7) (0.6, 0.1) (0.1, 0.9) (0.2, 0.9) ⎠
ζ4 (0.8, 0.3) (0.3, 0.7) (0.2, 0.3) (0.8, 0.3) (0.8, 0.5)
u1 u2 u3 u4 u5
⎛ ⎞
(0.7, 0.1)
ζ1 (0.1, 0.9) (0.5, 0.8) (0.4, 0.6) (0.7, 0.3)
ζ2⎜ (0.2, 0.3) (0.3, 0.8) (0.8, 0.5) (0.2, 0.4) (0.4, 0.3) ⎟
(P3 , E) = ⎜ ⎟
ζ3⎝ (0.1, 0.3) (0.9, 0.3) (0.9, 0.1) (0.3, 0.3) (0.5, 0.9) ⎠
ζ4 (0.8, 0.1) (0.1, 0.7) (0.4, 0.8) (0.3, 0.7) (0.7, 0.5)
Div1 ((P1 , E), (P2 , E)) = 0.133 Div2 ((P1 , E), (P2 , E)) = 0.806
Div1 ((P1 , E), (P3 , E)) = 0.142 Div2 ((P1 , E), (P3 , E)) = 1.529
Div1 ((P2 , E), (P3 , E)) = 0.194 Div2 ((P2 , E), (P3 , E)) = 2.006
From this we can conclude that the difference between (P2 , E) and (P3 , E) is much
larger than with other PFSSs.
This section explains an algorithm for the TOPSIS method based on the proposed
divergence measure under the PFSS environment and a numerical example of knowl-
edge management.
each quality of variants individually. The experts are given their opinion of agreement
and disagreement with each alternative in terms of PFSS. By considering the opinion
of each specialist significantly, we have to identify which one is the best among the
variants under study. The TOPSIS method based on the PFSS divergence measure
for selecting the best-fitting alternative is summarized below.
Step 1: Construct the decision matrices in terms of PFSSs for each {Vu : u =
1, 2, . . . r } with experts set {S1 , S2 , . . . St } is universal set and set of qualities
{Q 1 , Q 2 , . . . Q s } as the parameter set.
Step 2: Normalise the decision matrices according to which the attributes are benefit-
type or cost-type.
vi j if j is benefit-type attribute
vi j =
vicj if j is cost-type attribute
Du+
Rcu = . (10)
Du+ + Du−
Step 6: Rank the variants Vu according to the relative closeness, the bigger is the Rcu ,
the better is the variant Vu .
Divergence Measures of Pythagorean Fuzzy Soft Sets 419
Table 1 Trade T1
DM DM CM SC WM
S1 (0.12,0.23) (0.76,0.36) (0.36,0.56) (0.36,0.56) (0.81,0.21)
S2 (0.77,0.05) (0.71,0.37) (0.82,0.55) (0.84,0.18) (0.57,0.45)
S3 (0.77,0.10) (0.40,0.33) (0.92,0.05) (0.89,0.34) (0.44,0.76)
S4 (0.56,0.14) (0.78,0.22) (0.71,0.62) (0.75,0.22) (0.27,0.65)
Table 2 Trade T2
DM DM CM SC WM
S1 (0.90,0.02) (0.23,0.46) (0.56,0.67) (0.28,0.56) (0.82,0.18)
S2 (0.89,0.13) (0.77,0.21) (0.77,0.34) (0.87,0.04) (0.11,0.69)
S3 (0.58,0.32) (0.37,0.32) (0.77,0.45) (0.89,0.13) (0.43,0.53)
S4 (0.68,0.11) (0.62,0.23) (0.78,0.46) (0.45,0.26) (0.90,0.10)
Table 3 Trade T3
DM DM CM SC WM
S1 (0.68,0.22) (0.30,0.64) (0.57,0.56) (0.29,0.69) (0.61,0.45)
S2 (0.79,0.12) (0.47,0.30) (0.70,0.55) (0.76,0.24) (0.56,0.70)
S3 (0.88,0.03) (0.73,0.22) (0.72,0.45) (0.89,0.38) (0.67,0.50)
S4 (0.86,0.10) (0.90,0.21) (0.82,0.46) (0.45,0.26) (0.87,0.21)
420 T. M. Athira and S. J. John
Table 4 Trade T4
DM DM CM SC WM
S1 (0.28,0.56) (0.33,0.36) (0.52,0.76) (0.21,0.65) (0.71,0.47)
S2 (0.90,0.20) (0.71,0.21) (0.74,0.50) (0.64,0.34) (0.91,0.26)
S3 (0.78,0.12) (0.74,0.22) (0.81,0.44) (0.90,0.23) (0.87,0.25)
S4 (0.76,0.11) (0.67,0.27) (0.88,0.56) (0.58,0.21) (0.65,0.23)
Table 5 Trade T5
DM DM CM SC WM
S1 (0.58,0.34) (0.35,0.60) (0.51,0.46) (0.25,0.68) (0.37,0.81)
S2 (0.85,0.34) (0.54,0.32) (0.67,0.25) (0.26,0.64) (0.22,0.66)
S3 (0.47,0.57) (0.54,0.69) (0.23,0.68) (0.39,0.73) (0.34,0.54)
S4 (0.24,0.52) (0.61,0.25) (0.45,0.65) (0.35,0.27) (0.71,0.15)
Table 6 Trade T6
DM DM CM SC WM
S1 (0.81,0.22) (0.32,0.66) (0.56,0.67) (0.27,0.63) (0.35,0.67)
S2 (0.79,0.12) (0.76,0.33) (0.73,0.57) (0.67,0.46) (0.32,0.67)
S3 (0.67,0.24) (0.74,0.42) (0.74,0.24) (0.89,0.23) (0.74,0.23)
S4 (0.62,0.21) (0.56,0.23) (0.83,0.61) (0.57,0.26) (0.90,0.23)
Step 4:
D + = 0.0568 0.0489 0.0374 0.0378 0.1386 0.0511
D − = 0.0989 0.1110 0.1019 0.1081 0.0241 0.0882 .
Divergence Measures of Pythagorean Fuzzy Soft Sets 421
Rc1 = 0.3648 Rc2 = 0.3058 Rc3 = 0.2685 Rc4 = 0.2591 Rc5 = 0.8519 Rc6 = 0.3668
Step 6: The highest relative closeness is for Trade 5. Hence, we can conclude that
Trade 5 is the best decision.
5 Conclusion
We introduced and studied the measure of divergence for Pythagorean fuzzy soft sets.
Some nice expressions to calculate PFSS divergence measures are obtained. Certain
theorems that state the properties of PFSS divergence measures are also proven. As
future work, we can use this PFSS divergence measurement in decision problems
such as TOPSIS algorithm, pattern recognition etc.
References
1. Bhandari, D., Pal, N.R., Majumder, D.D.: Fuzzy divergence, probability measure of fuzzy
events and image thresholding. Pattern Recognit. Lett. 13(12), 857–867 (1992)
2. Montes, I., Pal, N.R., Janiš, V., Montes, S.: Divergence measures for intuitionistic fuzzy sets.
IEEE Trans. Fuzzy Syst. 23(2), 444–456 (2014)
3. Athira, T.M., John, S.J., Garg, H.: Entropy and distance measures of pythagorean fuzzy soft
sets and their applications. J. Intell. & Fuzzy Syst. 37(3), 4071–4084 (2019)
4. Athira, T.M., John, S.J., Garg, H.: A novel entropy measure of pythagorean fuzzy soft sets.
AIMS Math. 5(2), 1050–1061 (2020)
5. Garg, H., Rani, D.: Novel exponential divergence measure of complex intuitionistic fuzzy sets
with an application to the decision-making process. Sci. Iran. 28(4), 2439–2456 (2021)
6. Hwang, C.L., Yoon, K.: Methods for multiple attribute decision making. Multiple Attribute
Decision Making, pp. 58–191. Springer, Berlin, Heidelberg
7. Maji, P.K., Biswas, R., Roy, A.R.: Intuitionistic fuzzy soft sets. J. Fuzzy Math. 9(3), 677–692
(2001)
8. Molodtsov, D.: Soft set theory-first results. Comput. & Math. Appl. 37(4–5), 19–31 (1999)
9. Peng, X.D., Yang, Y., Song, J., Jiang, Y.: Pythagorean fuzzy soft set and its application. Comput.
Eng. 41(7), 224–229 (2015)
10. Xiao, F., Ding, W.: Divergence measure of Pythagorean fuzzy sets and its application in medical
diagnosis. Appl. Soft Comput. 79, 254–267 (2019)
11. Yager, R.R.: Pythagorean membership grades in multicriteria decision making. IEEE Trans.
Fuzzy Syst. 22(4), 958–965 (2013)
12. Montes, I., Pal, N.R., Montes, S.: Entropy measures for Atanassov intuitionistic fuzzy sets
based on divergence. Soft Comput. 22(15), 5051–5071 (2018)
13. Liu, B., Zhou, Q., Ding, R.X., Ni, W., Herrera, F.: Defective alternatives detection-based
multi-attribute intuitionistic fuzzy large-scale decision making model. Knowl. Based Syst.
186, 104962 (2019)
14. Thao, N.X.: Some new entropies and divergence measures of intuitionistic fuzzy sets based on
Archimedean t-conorm and application in supplier selection. Soft Comput. 25(7), 5791–5805
(2021)
15. Zhou, Q., Mo, H., Deng, Y.: A new divergence measure of pythagorean fuzzy sets based on
belief function and its application in medical diagnosis. Math. 8(1), 142 (2020)
Fuzzy-Rough Optimization Technique
for Breast Cancer Classification
Abstract Breast cancer is one of the deadly diseases amid women. The survival
rate can be increased through early detection. The classification model with high
level of predictive performance will help the medical experts to early identification
of this disease. To develop such types of robust and optimal classification model,
computational approach will be useful in early identification. In this paper, we intro-
duce hybrid intelligent fuzzy-rough classification method based on rule induction.
At initial stage, irrelevant features are removed through weak gamma evaluator.
Performance of this classification model is examined for Wisconsin Breast Cancer
Database (WBCD) and classification accuracy evaluated through F-measure. Perfor-
mance measure of fuzzy-rough set optimization technique is taken into account by
measuring sensitivity, specificity, and accuracy of the applied technique. Verification
and validation exercise of the applied technique is carried out on the basis of results
obtained in the similar direction by various realistic breast cancer images captured
by thermography.
1 Introduction
One of the most common life-threatening diseases among women is breast cancer.
Age factor, consumption of alcohol, obesity, breast implants, hormonal imbalance,
and uses of hormone-related medicines and non-breast feeding are root causes for this
K. Anitha (B)
Department of Mathematics, SRM Institute of Science and Technology,
Ramapuram, Chennai, India
e-mail: [email protected]
D. Datta
Former Nuclear Scientist, Bhabha Atomic Research Centre, Mumbai 400085, India
e-mail: [email protected]
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 423
T. Som et al. (eds.), Applied Analysis, Optimization and Soft Computing,
Springer Proceedings in Mathematics & Statistics 419,
https://doi.org/10.1007/978-981-99-0597-3_30
424 K. Anitha and D. Datta
and techniques are discussed in [11–13]. CAD system was introduced through neu-
trosophic and optimized fuzzy c-means method in [14]. Fuzzy-rough set approach
machine learning techniques are clearly explained in [15]. Correlation-based filter
technique is used to reduce the high-dimensional cancer data. Maji [16] proposed
IT2 fuzzy-rough set attribute reduction method by increasing the relevance and sig-
nificance of features. In this paper, we have proposed fuzzy-rough feature extraction
technique by rule generation based on similarity measure and it is being implemented
on thermographic image. Preliminary concepts are discussed in Sect. 2, main work
is demonstrated in Sect. 3, and experimental results are discussed in Sect. 4 followed
by concluded remarks.
2 Preliminaries
|E(X)|
6. Rough Accuracy of Approximation: |E(X)|
.
Example 1 The following information system consists of six objects, three condi-
tional attributes, and a decision attribute. Here symptoms are conditional attributes,
flu is decision attribute (Table 1).
Finite number of deterministic rules can be generated through reducts for given
dataset. The concepts of rule induction for knowledge discovery are developed in
[17, 18].
Fuzzy-Rough Optimization Technique for Breast Cancer Classification 427
Let us take two attribute sets A1 & A2 (disjoint) from an information system and
their equivalence classes are [x]A1 , [x]A2 , respectively. The dependency of attribute
set A2 on A1 is given by
n
|A1 (X)A2i |
d = γ(A1 , A2 ) = .
i=1
U
S1 is the most important attribute to give the decision about whether the patient
may get affected with flu or not. If we remove the variables from reduct and core, it
will minimize the quality of data. The main disadvantage of rough set is that it will
bring optimal result when the information system is discrete or the attribute values
are precise. It cannot be possible for high-dimensional data, especially image data.
Classical rough set deals about attributes having precise (certain) values but, in some
data, attributes are having fuzzy values. They are called hybrid attributes. For han-
dling these types of attributes, hybridization technique is needed to get optimal result.
Hybridization is the process of merging two or more than two optimization tech-
niques into single system. When the attribute takes fuzzy values, the equivalence
relation between attribute can be constructed through fuzzy-similarity relation. In
this paper, we proposed fuzzy-rough hybridization technique to identify consistent
attributes. For this technique, fuzzy-similarity relation can be considered as a base for
rough approximation. In this process, rough set indiscernibility relation is extended
to similarity relation. Let I be the indiscernibility relation which satisfies reflexive,
transitive, and symmetric. If I satisfies the following properties, then it is said to be
fuzzy-similarity relation (SR)
1. ∀x ∈ Us , I(x) ⊆ SR(x)
2. ∀x ∈ Us , ∀y ∈ SR(x), I(y) ⊆ SR(x).
From this relation, fuzzy-rough approximations are defined as follows:
428 K. Anitha and D. Datta
SR(x) = {x ∈ X : SR(x) ⊆ X}
SR(x) = SR(x).
x∈X
where d = |θ(x, v) − θ(y, v)|, μ ∈ [0, 0.5], and v ∈ [0 1] be the fuzzy normalized
attribute value T.
The fuzzy-similarity relation F on the set X of objects, then its set approximations
are given by
Fuzzy-Rough Lower Approximation: F(X) = {xi : [xi ]F ⊆ X, xi ∈ Us }.
Fuzzy-Rough Upper Approximation: F(X) = {xi : [xi ]F ∩ X = ∅, xi ∈ Us }.
Here P ⊆ Q represents μ P (x) ≤ μ Q (x).
3 Proposed Method
In this paper, we have introduced fuzzy-rough set attribute reduction based on similar-
ity relation. Let T be the target set. Fuzzy-rough approximations have been calculated
for the target set using the above definition. Then similarity measure β is calculated
by
1
β(X, Y ) = 1 − |Pos(X ) − Pos(Y )|, ∀ X, Y ⊂ U s ,
2
where 0 ≤ β(X, Y ) ≤ 1 and if β(X, Y ) = 1 then X = Y .
The attribute with maximum similarity has to be removed. The following flowchart
illustrates the proposed method.
4 Experimental Results
We examined breast cancer dataset from Wisconsin Breast Cancer Database through
Kaggle. Features are evaluated from digitalized image (FNA) based on breast mass
which portray the nature and thumbprint of nuclei cell appeared in the data(picture).
Thermographic image is analyzed through image j software and selected portions are
characterized through different statistical measures. In pre-processing stage, irrele-
vant features are removed through weak gamma evaluator and then image is converted
Fuzzy-Rough Optimization Technique for Breast Cancer Classification 429
to data table with proper attributes. As per the proposed algorithm, for the following
thermographic image 31 valid attributes have been found. There are 30 conditional
attributes with respect to various characteristics of tumor and the decision attribute
is whether the tumor is benign or malignant. The experimental results are displayed
as follows: (Figs. 1, 2, 3, 4, 5, 6, and 7):
Attributes: 30 (Conditional Attributes) + 1 (Decision Attribute − Benign, Malig-
nant) Diagnosis (Decision Attribute)
Conditional Attributes: radius_mean, texture_mean, perimeter_mean, smoothness
of tumor, concavity, symmetry,………fractal_dim_worst
Class distribution:
Total Instances: 569
357 Benign (Non-Cancerous),
212 Malignant(Cancerous) with 569 Instances.
Total number of rules generated: 126
Total number of attributes selected: 06
Description of Conditional Attributes
Classification Rules
Rule: 1
If compact_ worst <= 0.324
Then
fractal_dim_worst = 0.0001 * diagnosis = M − 0 * radius_mean − 0.1521 * smooth-
ness_mean − 0.0528 * compact_mean − 0.001 * concavity_mean − 0.0012 * sym-
metry_mean + 0.9149 * fractal_dim_mean − 0.0007 * radius_se + 0 * perime-
ter_se + 0 * area_se − 0.7479 * smoothness_se − 0.3582 * compact_se − 0.0067
* concavity_se − 0.0695 * concave points_se − 0.1174 * symmetry_se + 4.8651 *
fractal_dim_se + 0 * radius_worst − 0 * area_worst + 0.1592 * smooth_worst +
0.0678 * compact_worst + 0.0011 * concavity_worst + 0.0156 * symmetry_worst
+ 0.0028 [252/12.719%]
Rule: 2
If compact_worst <= 0.356
Then
fractal_dim_worst = 0.0006 * radius_mean − 0 * area_mean − 0.1399 * smooth-
ness_mean − 0.1164 * compact_mean − 0.0643 * concavity_mean + 0.0674 *
concave points_mean + 1.3089 * fractal_dim_mean − 0.0109 * radius_se + 0.0011
* perimeter_se − 0.1999 * compact_se − 0.0059 * concavity_se − 0.1289 * sym-
metry_se + 1.851 * fractal_dim_se + 0.0905 * smooth_worst + 0.0803 * com-
pact_worst + 0.0147 * concavity_worst − 0.0015 * concave points_worst + 0.016
* symmetry_worst − 0.017 [192/19.663%]
…………
430 K. Anitha and D. Datta
Rule: 126
If fractal_dim_mean <= 0.062
Then
fractal_dim_worst = 0.0002 * texture_mean + 0.1409 * smoothness_mean − 0.138
* compact_mean − 0.0175 * concavity_mean + 0.9885 * fractal_dim_mean +
0.004 * radius_se − 0.0004 * perimeter_se − 0.5833 * smoothness_se − 0.4537
* compact_se + 0.0305 * concavity_se − 0.5471 * concave points_se + 5.8746 *
fractal_dim_se + 0.0541 * smooth_worst + 0.0979 * compact_worst + 0.0387 *
concave points_worst − 0.0173 [125/24.827%]
Total Number of Selected attributes: 6 (Characteristics of tumor)
Hence, these six attributes identify whether the tumor is benign or malignant.
The above data is tested with other intelligent tools and their performance is
displayed below. Fuzzy-rough similarity measure covers more number of attributes
in ROI and more number of subsets are being generated, which cannot be possible
in conventional rough set since it focuses on precise values of attributes.
Fuzzy-Rough Optimization Technique for Breast Cancer Classification 433
5 Conclusion
In this paper, we have proposed fuzzy-rough similarity algorithm for early detection
of breast cancer from thermographic image. Image j software and Rosetta software
are used for image processing and rule generation. This algorithm selects minimum
number of consistent variables with maximum classification accuracy than other soft
computing techniques. The main advantage of this algorithm is that it uses similar-
ity relation which is more flexible than equivalence relation. Hence, it occupies a
greater number of variables in its boundary region. Comparative results show that
proposed algorithm gave optimal result. Future work may be extended to hybridiza-
tion technique of rough set with soft set through neighborhood relation (Tables 2
and 3).
References
1. Mahmoudzadeh, E., Montazeri, M.A., Zekri, M., Sadri, S.: Extended hidden Markov model
for optimized segmentation of breast thermography images. Infrared Phys. Technol. 72, 19–28
(2015)
2. Singh, D., Singh, A.K.: Role of image thermography in early breast cancer detection- past,
present and future. Comput. Methods Programs Biomed. 183, 105074 (2020). https://doi.org/
10.1016/j.cmpb.2019.105074. Epub 2019 Sep 7. PMID: 31525547
3. Zadeh, L.A.: Fuzzy sets. Inf. Control 8, 338–353 (1965)
4. Anitha, K.: Rough Set Theory on Topological Spaces. Rough Sets and Knowledge Technology.
RSKT 2014. Lecture Notes in Computer Science, vol. 8818. Springer, Cham (2014). https://
doi.org/10.1007/978-3-319-11740-9_7
5. Anitha, K., Thangeswari, M.: Rough set based optimization techniques. In: 6th International
Conference on Advanced Computing and Communication Systems (ICACCS), pp. 1021–1024
(2020). https://doi.org/10.1109/ICACCS48705.2020.9074212
6. Kausar, N., Munir, M., Kousar, S., Gulistan, M., Anitha, K.: Characterization of non-associative
ordered semigroups by the properties of F-ideals. Fuzzy Inf. Eng. 12, 490–508 (2020)
7. Chiranji Lal Chowdhary, D.P., Acharjya, A.: Hybrid scheme for breast cancer detection using
intuitionistic fuzzy rough set technique. Int. J. Healthc. Inf. Syst. Inform. 11 (2016)
8. Ganesh Kumar, P., Kavitha, M.S., Ahn, B.-C.: Automated detection of cancer associated genes
using a combined fuzzy-rough-set-based F-information and water swirl algorithm of human
gene expression data. PLoS ONE 11, e0167504 (2016)
9. Jaya Kumari, D.: Prognosticate breast cancer by exerting fuzzy precognition clustering tech-
nique and collate technique. Int. J. Appl. Eng. Res. 13, 15492–15502 (2018). ISSN 0973-4562
10. Golestani, N., Etehad Tavakol, M., Ng, E.Y.K.: Level set method for segmentation of infrared
breast thermograms. EXCLI J. 13, 241–251 (2014)
11. Schaefer, G., Závišek, M., Nakashima, T.: Thermography based breast cancer analysis using
statistical features and fuzzy classification. Pattern Recognit. 42, 11331137 (2009)
12. Francis, S.V., Sasikala, M., Bhavani Bharathi, G., Jaipurkar, S.D.: Breast cancer detection in
rotational thermography images using texture features. Infrared Phys. Technol. 67, 490–496
(2014)
13. Milosevic, M., Jankovic, D., Peulic, A.: Thermography based breast cancer detection using
texture features and minimum variance quantization. EXCLI J. 13, 1204–1215 (2014)
14. Gaber, T., Ismail, G., Anter, A., et al.: Thermogram breast cancer prediction approach based
on neutrosophic sets and fuzzy C-means algorithm. In: Proceedings of the 2015 37th Annual
International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC),
pp. 4254–4257. IEEE, Milan, Italy, August (2015)
15. Arunkumar, C., Ramakrishnan, S.: Prediction of cancer using customised fuzzy rough machine
learning approaches. Healthc. Technol. Lett. 6, 13–18 (2018). https://doi.org/10.1049/htl.2018.
5055. PMID: 30881694; PMCID: PMC6407447
16. Maji, P., Garai, P.: IT2 fuzzy-rough sets and max relevance-max significance criterion for
attribute selection. IEEE Trans. Cybern. 45, 1657–1668 (2015). https://doi.org/10.1109/TCYB.
2014.2357892. Epub 2014 Sep 23. PMID: 25265623
17. Ziarko, W., Shan, N.: Discovering attribute relationships, dependencies and rules by using
rough sets. In: Proceedings of the 28th Annual Hawaii International Conference on System
Sciences (HICSS’95), pp. 293–299. Hawaii (1995)
18. Ziarko, W.: Rough sets as a methodology for data mining. In: Rough Sets in Knowledge
Discovery 1: Methodology and Applications, pp. 554–576. Physica, Heidelberg (1998)
19. Liu, Y., Zhou, Q., Rakus-Andersson, E., Bai, G.: A Fuzzy-Rough Sets Based Compact Rule
Induction Method for Classifying Hybrid Data, Lecture Notes in Artificial Intelligence, vol.
7414 (2012)